Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
/*-
|
|
|
|
|
* Copyright (c) 2017 Chelsio Communications, Inc.
|
|
|
|
|
* All rights reserved.
|
|
|
|
|
* Written by: John Baldwin <jhb@FreeBSD.org>
|
|
|
|
|
*
|
|
|
|
|
* Redistribution and use in source and binary forms, with or without
|
|
|
|
|
* modification, are permitted provided that the following conditions
|
|
|
|
|
* are met:
|
|
|
|
|
* 1. Redistributions of source code must retain the above copyright
|
|
|
|
|
* notice, this list of conditions and the following disclaimer.
|
|
|
|
|
* 2. Redistributions in binary form must reproduce the above copyright
|
|
|
|
|
* notice, this list of conditions and the following disclaimer in the
|
|
|
|
|
* documentation and/or other materials provided with the distribution.
|
|
|
|
|
*
|
|
|
|
|
* THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
|
|
|
|
|
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
|
|
|
|
|
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
|
|
|
|
|
* ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
|
|
|
|
|
* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
|
|
|
|
|
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
|
|
|
|
|
* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
|
|
|
|
|
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
|
|
|
|
|
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
|
|
|
|
|
* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
|
|
|
|
|
* SUCH DAMAGE.
|
|
|
|
|
*/
|
|
|
|
|
|
|
|
|
|
#include <sys/cdefs.h>
|
|
|
|
|
__FBSDID("$FreeBSD$");
|
|
|
|
|
|
|
|
|
|
#include <sys/types.h>
|
|
|
|
|
#include <sys/bus.h>
|
|
|
|
|
#include <sys/lock.h>
|
|
|
|
|
#include <sys/malloc.h>
|
|
|
|
|
#include <sys/mutex.h>
|
|
|
|
|
#include <sys/module.h>
|
|
|
|
|
#include <sys/sglist.h>
|
|
|
|
|
|
|
|
|
|
#include <opencrypto/cryptodev.h>
|
|
|
|
|
#include <opencrypto/xform.h>
|
|
|
|
|
|
|
|
|
|
#include "cryptodev_if.h"
|
|
|
|
|
|
|
|
|
|
#include "common/common.h"
|
|
|
|
|
#include "crypto/t4_crypto.h"
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
|
* Requests consist of:
|
|
|
|
|
*
|
|
|
|
|
* +-------------------------------+
|
|
|
|
|
* | struct fw_crypto_lookaside_wr |
|
|
|
|
|
* +-------------------------------+
|
|
|
|
|
* | struct ulp_txpkt |
|
|
|
|
|
* +-------------------------------+
|
|
|
|
|
* | struct ulptx_idata |
|
|
|
|
|
* +-------------------------------+
|
|
|
|
|
* | struct cpl_tx_sec_pdu |
|
|
|
|
|
* +-------------------------------+
|
|
|
|
|
* | struct cpl_tls_tx_scmd_fmt |
|
|
|
|
|
* +-------------------------------+
|
|
|
|
|
* | key context header |
|
|
|
|
|
* +-------------------------------+
|
|
|
|
|
* | AES key | ----- For requests with AES
|
2018-10-29 18:24:31 -04:00
|
|
|
* +-------------------------------+
|
|
|
|
|
* | Hash state | ----- For hash-only requests
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
* +-------------------------------+ -
|
|
|
|
|
* | IPAD (16-byte aligned) | \
|
|
|
|
|
* +-------------------------------+ +---- For requests with HMAC
|
|
|
|
|
* | OPAD (16-byte aligned) | /
|
|
|
|
|
* +-------------------------------+ -
|
|
|
|
|
* | GMAC H | ----- For AES-GCM
|
|
|
|
|
* +-------------------------------+ -
|
|
|
|
|
* | struct cpl_rx_phys_dsgl | \
|
|
|
|
|
* +-------------------------------+ +---- Destination buffer for
|
|
|
|
|
* | PHYS_DSGL entries | / non-hash-only requests
|
|
|
|
|
* +-------------------------------+ -
|
2018-10-29 18:24:31 -04:00
|
|
|
* | 16 dummy bytes | ----- Only for HMAC/hash-only requests
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
* +-------------------------------+
|
|
|
|
|
* | IV | ----- If immediate IV
|
|
|
|
|
* +-------------------------------+
|
|
|
|
|
* | Payload | ----- If immediate Payload
|
|
|
|
|
* +-------------------------------+ -
|
|
|
|
|
* | struct ulptx_sgl | \
|
|
|
|
|
* +-------------------------------+ +---- If payload via SGL
|
|
|
|
|
* | SGL entries | /
|
|
|
|
|
* +-------------------------------+ -
|
|
|
|
|
*
|
|
|
|
|
* Note that the key context must be padded to ensure 16-byte alignment.
|
|
|
|
|
* For HMAC requests, the key consists of the partial hash of the IPAD
|
|
|
|
|
* followed by the partial hash of the OPAD.
|
|
|
|
|
*
|
|
|
|
|
* Replies consist of:
|
|
|
|
|
*
|
|
|
|
|
* +-------------------------------+
|
|
|
|
|
* | struct cpl_fw6_pld |
|
|
|
|
|
* +-------------------------------+
|
|
|
|
|
* | hash digest | ----- For HMAC request with
|
|
|
|
|
* +-------------------------------+ 'hash_size' set in work request
|
|
|
|
|
*
|
|
|
|
|
* A 32-bit big-endian error status word is supplied in the last 4
|
|
|
|
|
* bytes of data[0] in the CPL_FW6_PLD message. bit 0 indicates a
|
|
|
|
|
* "MAC" error and bit 1 indicates a "PAD" error.
|
|
|
|
|
*
|
|
|
|
|
* The 64-bit 'cookie' field from the fw_crypto_lookaside_wr message
|
|
|
|
|
* in the request is returned in data[1] of the CPL_FW6_PLD message.
|
|
|
|
|
*
|
|
|
|
|
* For block cipher replies, the updated IV is supplied in data[2] and
|
|
|
|
|
* data[3] of the CPL_FW6_PLD message.
|
|
|
|
|
*
|
|
|
|
|
* For hash replies where the work request set 'hash_size' to request
|
|
|
|
|
* a copy of the hash in the reply, the hash digest is supplied
|
|
|
|
|
* immediately following the CPL_FW6_PLD message.
|
|
|
|
|
*/
|
|
|
|
|
|
2018-01-24 15:08:10 -05:00
|
|
|
/*
|
|
|
|
|
* The crypto engine supports a maximum AAD size of 511 bytes.
|
|
|
|
|
*/
|
|
|
|
|
#define MAX_AAD_LEN 511
|
|
|
|
|
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
/*
|
2018-01-24 15:13:07 -05:00
|
|
|
* The documentation for CPL_RX_PHYS_DSGL claims a maximum of 32 SG
|
|
|
|
|
* entries. While the CPL includes a 16-bit length field, the T6 can
|
|
|
|
|
* sometimes hang if an error occurs while processing a request with a
|
|
|
|
|
* single DSGL entry larger than 2k.
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
*/
|
|
|
|
|
#define MAX_RX_PHYS_DSGL_SGE 32
|
2018-01-24 15:13:07 -05:00
|
|
|
#define DSGL_SGE_MAXLEN 2048
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
|
2017-05-26 16:20:40 -04:00
|
|
|
/*
|
|
|
|
|
* The adapter only supports requests with a total input or output
|
|
|
|
|
* length of 64k-1 or smaller. Longer requests either result in hung
|
|
|
|
|
* requests or incorrect results.
|
|
|
|
|
*/
|
|
|
|
|
#define MAX_REQUEST_SIZE 65535
|
|
|
|
|
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
static MALLOC_DEFINE(M_CCR, "ccr", "Chelsio T6 crypto");
|
|
|
|
|
|
|
|
|
|
struct ccr_session_hmac {
|
|
|
|
|
struct auth_hash *auth_hash;
|
|
|
|
|
int hash_len;
|
|
|
|
|
unsigned int partial_digest_len;
|
|
|
|
|
unsigned int auth_mode;
|
|
|
|
|
unsigned int mk_size;
|
2019-11-12 19:53:45 -05:00
|
|
|
char pads[CHCR_HASH_MAX_BLOCK_SIZE_128 * 2];
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
};
|
|
|
|
|
|
|
|
|
|
struct ccr_session_gmac {
|
|
|
|
|
int hash_len;
|
|
|
|
|
char ghash_h[GMAC_BLOCK_LEN];
|
|
|
|
|
};
|
|
|
|
|
|
2019-04-24 19:31:46 -04:00
|
|
|
struct ccr_session_ccm_mac {
|
|
|
|
|
int hash_len;
|
|
|
|
|
};
|
|
|
|
|
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
struct ccr_session_blkcipher {
|
|
|
|
|
unsigned int cipher_mode;
|
|
|
|
|
unsigned int key_len;
|
|
|
|
|
unsigned int iv_len;
|
|
|
|
|
__be32 key_ctx_hdr;
|
|
|
|
|
char enckey[CHCR_AES_MAX_KEY_LEN];
|
|
|
|
|
char deckey[CHCR_AES_MAX_KEY_LEN];
|
|
|
|
|
};
|
|
|
|
|
|
2020-04-10 18:27:45 -04:00
|
|
|
struct ccr_port {
|
|
|
|
|
struct sge_wrq *txq;
|
|
|
|
|
struct sge_rxq *rxq;
|
|
|
|
|
int tx_channel_id;
|
|
|
|
|
u_int active_sessions;
|
|
|
|
|
};
|
|
|
|
|
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
struct ccr_session {
|
2020-06-25 20:01:31 -04:00
|
|
|
#ifdef INVARIANTS
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
int pending;
|
2020-06-25 20:01:31 -04:00
|
|
|
#endif
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
enum { HASH, HMAC, BLKCIPHER, ETA, GCM, CCM } mode;
|
2020-04-10 18:27:45 -04:00
|
|
|
struct ccr_port *port;
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
union {
|
|
|
|
|
struct ccr_session_hmac hmac;
|
|
|
|
|
struct ccr_session_gmac gmac;
|
2019-04-24 19:31:46 -04:00
|
|
|
struct ccr_session_ccm_mac ccm_mac;
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
};
|
|
|
|
|
struct ccr_session_blkcipher blkcipher;
|
|
|
|
|
struct mtx lock;
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
|
* Pre-allocate S/G lists used when preparing a work request.
|
2020-05-25 18:23:13 -04:00
|
|
|
* 'sg_input' contains an sglist describing the entire input
|
|
|
|
|
* buffer for a 'struct cryptop'. 'sg_output' contains an
|
|
|
|
|
* sglist describing the entire output buffer. 'sg_ulptx' is
|
|
|
|
|
* used to describe the data the engine should DMA as input
|
|
|
|
|
* via ULPTX_SGL. 'sg_dsgl' is used to describe the
|
|
|
|
|
* destination that cipher text and a tag should be written
|
|
|
|
|
* to.
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
*/
|
2020-05-25 18:23:13 -04:00
|
|
|
struct sglist *sg_input;
|
|
|
|
|
struct sglist *sg_output;
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
struct sglist *sg_ulptx;
|
|
|
|
|
struct sglist *sg_dsgl;
|
2020-06-25 20:01:31 -04:00
|
|
|
};
|
|
|
|
|
|
|
|
|
|
struct ccr_softc {
|
|
|
|
|
struct adapter *adapter;
|
|
|
|
|
device_t dev;
|
|
|
|
|
uint32_t cid;
|
|
|
|
|
struct mtx lock;
|
|
|
|
|
bool detaching;
|
|
|
|
|
struct ccr_port ports[MAX_NPORTS];
|
|
|
|
|
u_int port_mask;
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
|
2018-01-24 15:11:00 -05:00
|
|
|
/*
|
|
|
|
|
* Pre-allocate a dummy output buffer for the IV and AAD for
|
|
|
|
|
* AEAD requests.
|
|
|
|
|
*/
|
|
|
|
|
char *iv_aad_buf;
|
|
|
|
|
struct sglist *sg_iv_aad;
|
|
|
|
|
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
/* Statistics. */
|
2020-06-25 20:01:31 -04:00
|
|
|
counter_u64_t stats_blkcipher_encrypt;
|
|
|
|
|
counter_u64_t stats_blkcipher_decrypt;
|
|
|
|
|
counter_u64_t stats_hash;
|
|
|
|
|
counter_u64_t stats_hmac;
|
|
|
|
|
counter_u64_t stats_eta_encrypt;
|
|
|
|
|
counter_u64_t stats_eta_decrypt;
|
|
|
|
|
counter_u64_t stats_gcm_encrypt;
|
|
|
|
|
counter_u64_t stats_gcm_decrypt;
|
|
|
|
|
counter_u64_t stats_ccm_encrypt;
|
|
|
|
|
counter_u64_t stats_ccm_decrypt;
|
|
|
|
|
counter_u64_t stats_wr_nomem;
|
|
|
|
|
counter_u64_t stats_inflight;
|
|
|
|
|
counter_u64_t stats_mac_error;
|
|
|
|
|
counter_u64_t stats_pad_error;
|
|
|
|
|
counter_u64_t stats_sglist_error;
|
|
|
|
|
counter_u64_t stats_process_error;
|
|
|
|
|
counter_u64_t stats_sw_fallback;
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
};
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
|
* Crypto requests involve two kind of scatter/gather lists.
|
|
|
|
|
*
|
|
|
|
|
* Non-hash-only requests require a PHYS_DSGL that describes the
|
|
|
|
|
* location to store the results of the encryption or decryption
|
|
|
|
|
* operation. This SGL uses a different format (PHYS_DSGL) and should
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
* exclude the skip bytes at the start of the data as well as any AAD
|
|
|
|
|
* or IV. For authenticated encryption requests it should include the
|
|
|
|
|
* destination of the hash or tag.
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
*
|
|
|
|
|
* The input payload may either be supplied inline as immediate data,
|
|
|
|
|
* or via a standard ULP_TX SGL. This SGL should include AAD,
|
|
|
|
|
* ciphertext, and the hash or tag for authenticated decryption
|
|
|
|
|
* requests.
|
|
|
|
|
*
|
|
|
|
|
* These scatter/gather lists can describe different subsets of the
|
Add support for optional separate output buffers to in-kernel crypto.
Some crypto consumers such as GELI and KTLS for file-backed sendfile
need to store their output in a separate buffer from the input.
Currently these consumers copy the contents of the input buffer into
the output buffer and queue an in-place crypto operation on the output
buffer. Using a separate output buffer avoids this copy.
- Create a new 'struct crypto_buffer' describing a crypto buffer
containing a type and type-specific fields. crp_ilen is gone,
instead buffers that use a flat kernel buffer have a cb_buf_len
field for their length. The length of other buffer types is
inferred from the backing store (e.g. uio_resid for a uio).
Requests now have two such structures: crp_buf for the input buffer,
and crp_obuf for the output buffer.
- Consumers now use helper functions (crypto_use_*,
e.g. crypto_use_mbuf()) to configure the input buffer. If an output
buffer is not configured, the request still modifies the input
buffer in-place. A consumer uses a second set of helper functions
(crypto_use_output_*) to configure an output buffer.
- Consumers must request support for separate output buffers when
creating a crypto session via the CSP_F_SEPARATE_OUTPUT flag and are
only permitted to queue a request with a separate output buffer on
sessions with this flag set. Existing drivers already reject
sessions with unknown flags, so this permits drivers to be modified
to support this extension without requiring all drivers to change.
- Several data-related functions now have matching versions that
operate on an explicit buffer (e.g. crypto_apply_buf,
crypto_contiguous_subsegment_buf, bus_dma_load_crp_buf).
- Most of the existing data-related functions operate on the input
buffer. However crypto_copyback always writes to the output buffer
if a request uses a separate output buffer.
- For the regions in input/output buffers, the following conventions
are followed:
- AAD and IV are always present in input only and their
fields are offsets into the input buffer.
- payload is always present in both buffers. If a request uses a
separate output buffer, it must set a new crp_payload_start_output
field to the offset of the payload in the output buffer.
- digest is in the input buffer for verify operations, and in the
output buffer for compute operations. crp_digest_start is relative
to the appropriate buffer.
- Add a crypto buffer cursor abstraction. This is a more general form
of some bits in the cryptosoft driver that tried to always use uio's.
However, compared to the original code, this avoids rewalking the uio
iovec array for requests with multiple vectors. It also avoids
allocate an iovec array for mbufs and populating it by instead walking
the mbuf chain directly.
- Update the cryptosoft(4) driver to support separate output buffers
making use of the cursor abstraction.
Sponsored by: Netflix
Differential Revision: https://reviews.freebsd.org/D24545
2020-05-25 18:12:04 -04:00
|
|
|
* buffers described by the crypto operation. ccr_populate_sglist()
|
|
|
|
|
* generates a scatter/gather list that covers an entire crypto
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
* operation buffer that is then used to construct the other
|
|
|
|
|
* scatter/gather lists.
|
|
|
|
|
*/
|
|
|
|
|
static int
|
Add support for optional separate output buffers to in-kernel crypto.
Some crypto consumers such as GELI and KTLS for file-backed sendfile
need to store their output in a separate buffer from the input.
Currently these consumers copy the contents of the input buffer into
the output buffer and queue an in-place crypto operation on the output
buffer. Using a separate output buffer avoids this copy.
- Create a new 'struct crypto_buffer' describing a crypto buffer
containing a type and type-specific fields. crp_ilen is gone,
instead buffers that use a flat kernel buffer have a cb_buf_len
field for their length. The length of other buffer types is
inferred from the backing store (e.g. uio_resid for a uio).
Requests now have two such structures: crp_buf for the input buffer,
and crp_obuf for the output buffer.
- Consumers now use helper functions (crypto_use_*,
e.g. crypto_use_mbuf()) to configure the input buffer. If an output
buffer is not configured, the request still modifies the input
buffer in-place. A consumer uses a second set of helper functions
(crypto_use_output_*) to configure an output buffer.
- Consumers must request support for separate output buffers when
creating a crypto session via the CSP_F_SEPARATE_OUTPUT flag and are
only permitted to queue a request with a separate output buffer on
sessions with this flag set. Existing drivers already reject
sessions with unknown flags, so this permits drivers to be modified
to support this extension without requiring all drivers to change.
- Several data-related functions now have matching versions that
operate on an explicit buffer (e.g. crypto_apply_buf,
crypto_contiguous_subsegment_buf, bus_dma_load_crp_buf).
- Most of the existing data-related functions operate on the input
buffer. However crypto_copyback always writes to the output buffer
if a request uses a separate output buffer.
- For the regions in input/output buffers, the following conventions
are followed:
- AAD and IV are always present in input only and their
fields are offsets into the input buffer.
- payload is always present in both buffers. If a request uses a
separate output buffer, it must set a new crp_payload_start_output
field to the offset of the payload in the output buffer.
- digest is in the input buffer for verify operations, and in the
output buffer for compute operations. crp_digest_start is relative
to the appropriate buffer.
- Add a crypto buffer cursor abstraction. This is a more general form
of some bits in the cryptosoft driver that tried to always use uio's.
However, compared to the original code, this avoids rewalking the uio
iovec array for requests with multiple vectors. It also avoids
allocate an iovec array for mbufs and populating it by instead walking
the mbuf chain directly.
- Update the cryptosoft(4) driver to support separate output buffers
making use of the cursor abstraction.
Sponsored by: Netflix
Differential Revision: https://reviews.freebsd.org/D24545
2020-05-25 18:12:04 -04:00
|
|
|
ccr_populate_sglist(struct sglist *sg, struct crypto_buffer *cb)
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
{
|
|
|
|
|
int error;
|
|
|
|
|
|
|
|
|
|
sglist_reset(sg);
|
Add support for optional separate output buffers to in-kernel crypto.
Some crypto consumers such as GELI and KTLS for file-backed sendfile
need to store their output in a separate buffer from the input.
Currently these consumers copy the contents of the input buffer into
the output buffer and queue an in-place crypto operation on the output
buffer. Using a separate output buffer avoids this copy.
- Create a new 'struct crypto_buffer' describing a crypto buffer
containing a type and type-specific fields. crp_ilen is gone,
instead buffers that use a flat kernel buffer have a cb_buf_len
field for their length. The length of other buffer types is
inferred from the backing store (e.g. uio_resid for a uio).
Requests now have two such structures: crp_buf for the input buffer,
and crp_obuf for the output buffer.
- Consumers now use helper functions (crypto_use_*,
e.g. crypto_use_mbuf()) to configure the input buffer. If an output
buffer is not configured, the request still modifies the input
buffer in-place. A consumer uses a second set of helper functions
(crypto_use_output_*) to configure an output buffer.
- Consumers must request support for separate output buffers when
creating a crypto session via the CSP_F_SEPARATE_OUTPUT flag and are
only permitted to queue a request with a separate output buffer on
sessions with this flag set. Existing drivers already reject
sessions with unknown flags, so this permits drivers to be modified
to support this extension without requiring all drivers to change.
- Several data-related functions now have matching versions that
operate on an explicit buffer (e.g. crypto_apply_buf,
crypto_contiguous_subsegment_buf, bus_dma_load_crp_buf).
- Most of the existing data-related functions operate on the input
buffer. However crypto_copyback always writes to the output buffer
if a request uses a separate output buffer.
- For the regions in input/output buffers, the following conventions
are followed:
- AAD and IV are always present in input only and their
fields are offsets into the input buffer.
- payload is always present in both buffers. If a request uses a
separate output buffer, it must set a new crp_payload_start_output
field to the offset of the payload in the output buffer.
- digest is in the input buffer for verify operations, and in the
output buffer for compute operations. crp_digest_start is relative
to the appropriate buffer.
- Add a crypto buffer cursor abstraction. This is a more general form
of some bits in the cryptosoft driver that tried to always use uio's.
However, compared to the original code, this avoids rewalking the uio
iovec array for requests with multiple vectors. It also avoids
allocate an iovec array for mbufs and populating it by instead walking
the mbuf chain directly.
- Update the cryptosoft(4) driver to support separate output buffers
making use of the cursor abstraction.
Sponsored by: Netflix
Differential Revision: https://reviews.freebsd.org/D24545
2020-05-25 18:12:04 -04:00
|
|
|
switch (cb->cb_type) {
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
case CRYPTO_BUF_MBUF:
|
Add support for optional separate output buffers to in-kernel crypto.
Some crypto consumers such as GELI and KTLS for file-backed sendfile
need to store their output in a separate buffer from the input.
Currently these consumers copy the contents of the input buffer into
the output buffer and queue an in-place crypto operation on the output
buffer. Using a separate output buffer avoids this copy.
- Create a new 'struct crypto_buffer' describing a crypto buffer
containing a type and type-specific fields. crp_ilen is gone,
instead buffers that use a flat kernel buffer have a cb_buf_len
field for their length. The length of other buffer types is
inferred from the backing store (e.g. uio_resid for a uio).
Requests now have two such structures: crp_buf for the input buffer,
and crp_obuf for the output buffer.
- Consumers now use helper functions (crypto_use_*,
e.g. crypto_use_mbuf()) to configure the input buffer. If an output
buffer is not configured, the request still modifies the input
buffer in-place. A consumer uses a second set of helper functions
(crypto_use_output_*) to configure an output buffer.
- Consumers must request support for separate output buffers when
creating a crypto session via the CSP_F_SEPARATE_OUTPUT flag and are
only permitted to queue a request with a separate output buffer on
sessions with this flag set. Existing drivers already reject
sessions with unknown flags, so this permits drivers to be modified
to support this extension without requiring all drivers to change.
- Several data-related functions now have matching versions that
operate on an explicit buffer (e.g. crypto_apply_buf,
crypto_contiguous_subsegment_buf, bus_dma_load_crp_buf).
- Most of the existing data-related functions operate on the input
buffer. However crypto_copyback always writes to the output buffer
if a request uses a separate output buffer.
- For the regions in input/output buffers, the following conventions
are followed:
- AAD and IV are always present in input only and their
fields are offsets into the input buffer.
- payload is always present in both buffers. If a request uses a
separate output buffer, it must set a new crp_payload_start_output
field to the offset of the payload in the output buffer.
- digest is in the input buffer for verify operations, and in the
output buffer for compute operations. crp_digest_start is relative
to the appropriate buffer.
- Add a crypto buffer cursor abstraction. This is a more general form
of some bits in the cryptosoft driver that tried to always use uio's.
However, compared to the original code, this avoids rewalking the uio
iovec array for requests with multiple vectors. It also avoids
allocate an iovec array for mbufs and populating it by instead walking
the mbuf chain directly.
- Update the cryptosoft(4) driver to support separate output buffers
making use of the cursor abstraction.
Sponsored by: Netflix
Differential Revision: https://reviews.freebsd.org/D24545
2020-05-25 18:12:04 -04:00
|
|
|
error = sglist_append_mbuf(sg, cb->cb_mbuf);
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
break;
|
|
|
|
|
case CRYPTO_BUF_UIO:
|
Add support for optional separate output buffers to in-kernel crypto.
Some crypto consumers such as GELI and KTLS for file-backed sendfile
need to store their output in a separate buffer from the input.
Currently these consumers copy the contents of the input buffer into
the output buffer and queue an in-place crypto operation on the output
buffer. Using a separate output buffer avoids this copy.
- Create a new 'struct crypto_buffer' describing a crypto buffer
containing a type and type-specific fields. crp_ilen is gone,
instead buffers that use a flat kernel buffer have a cb_buf_len
field for their length. The length of other buffer types is
inferred from the backing store (e.g. uio_resid for a uio).
Requests now have two such structures: crp_buf for the input buffer,
and crp_obuf for the output buffer.
- Consumers now use helper functions (crypto_use_*,
e.g. crypto_use_mbuf()) to configure the input buffer. If an output
buffer is not configured, the request still modifies the input
buffer in-place. A consumer uses a second set of helper functions
(crypto_use_output_*) to configure an output buffer.
- Consumers must request support for separate output buffers when
creating a crypto session via the CSP_F_SEPARATE_OUTPUT flag and are
only permitted to queue a request with a separate output buffer on
sessions with this flag set. Existing drivers already reject
sessions with unknown flags, so this permits drivers to be modified
to support this extension without requiring all drivers to change.
- Several data-related functions now have matching versions that
operate on an explicit buffer (e.g. crypto_apply_buf,
crypto_contiguous_subsegment_buf, bus_dma_load_crp_buf).
- Most of the existing data-related functions operate on the input
buffer. However crypto_copyback always writes to the output buffer
if a request uses a separate output buffer.
- For the regions in input/output buffers, the following conventions
are followed:
- AAD and IV are always present in input only and their
fields are offsets into the input buffer.
- payload is always present in both buffers. If a request uses a
separate output buffer, it must set a new crp_payload_start_output
field to the offset of the payload in the output buffer.
- digest is in the input buffer for verify operations, and in the
output buffer for compute operations. crp_digest_start is relative
to the appropriate buffer.
- Add a crypto buffer cursor abstraction. This is a more general form
of some bits in the cryptosoft driver that tried to always use uio's.
However, compared to the original code, this avoids rewalking the uio
iovec array for requests with multiple vectors. It also avoids
allocate an iovec array for mbufs and populating it by instead walking
the mbuf chain directly.
- Update the cryptosoft(4) driver to support separate output buffers
making use of the cursor abstraction.
Sponsored by: Netflix
Differential Revision: https://reviews.freebsd.org/D24545
2020-05-25 18:12:04 -04:00
|
|
|
error = sglist_append_uio(sg, cb->cb_uio);
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
break;
|
|
|
|
|
case CRYPTO_BUF_CONTIG:
|
Add support for optional separate output buffers to in-kernel crypto.
Some crypto consumers such as GELI and KTLS for file-backed sendfile
need to store their output in a separate buffer from the input.
Currently these consumers copy the contents of the input buffer into
the output buffer and queue an in-place crypto operation on the output
buffer. Using a separate output buffer avoids this copy.
- Create a new 'struct crypto_buffer' describing a crypto buffer
containing a type and type-specific fields. crp_ilen is gone,
instead buffers that use a flat kernel buffer have a cb_buf_len
field for their length. The length of other buffer types is
inferred from the backing store (e.g. uio_resid for a uio).
Requests now have two such structures: crp_buf for the input buffer,
and crp_obuf for the output buffer.
- Consumers now use helper functions (crypto_use_*,
e.g. crypto_use_mbuf()) to configure the input buffer. If an output
buffer is not configured, the request still modifies the input
buffer in-place. A consumer uses a second set of helper functions
(crypto_use_output_*) to configure an output buffer.
- Consumers must request support for separate output buffers when
creating a crypto session via the CSP_F_SEPARATE_OUTPUT flag and are
only permitted to queue a request with a separate output buffer on
sessions with this flag set. Existing drivers already reject
sessions with unknown flags, so this permits drivers to be modified
to support this extension without requiring all drivers to change.
- Several data-related functions now have matching versions that
operate on an explicit buffer (e.g. crypto_apply_buf,
crypto_contiguous_subsegment_buf, bus_dma_load_crp_buf).
- Most of the existing data-related functions operate on the input
buffer. However crypto_copyback always writes to the output buffer
if a request uses a separate output buffer.
- For the regions in input/output buffers, the following conventions
are followed:
- AAD and IV are always present in input only and their
fields are offsets into the input buffer.
- payload is always present in both buffers. If a request uses a
separate output buffer, it must set a new crp_payload_start_output
field to the offset of the payload in the output buffer.
- digest is in the input buffer for verify operations, and in the
output buffer for compute operations. crp_digest_start is relative
to the appropriate buffer.
- Add a crypto buffer cursor abstraction. This is a more general form
of some bits in the cryptosoft driver that tried to always use uio's.
However, compared to the original code, this avoids rewalking the uio
iovec array for requests with multiple vectors. It also avoids
allocate an iovec array for mbufs and populating it by instead walking
the mbuf chain directly.
- Update the cryptosoft(4) driver to support separate output buffers
making use of the cursor abstraction.
Sponsored by: Netflix
Differential Revision: https://reviews.freebsd.org/D24545
2020-05-25 18:12:04 -04:00
|
|
|
error = sglist_append(sg, cb->cb_buf, cb->cb_buf_len);
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
break;
|
2020-08-25 22:37:42 -04:00
|
|
|
case CRYPTO_BUF_VMPAGE:
|
|
|
|
|
error = sglist_append_vmpages(sg, cb->cb_vm_page,
|
|
|
|
|
cb->cb_vm_page_len, cb->cb_vm_page_offset);
|
|
|
|
|
break;
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
default:
|
|
|
|
|
error = EINVAL;
|
|
|
|
|
}
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
return (error);
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
|
* Segments in 'sg' larger than 'maxsegsize' are counted as multiple
|
|
|
|
|
* segments.
|
|
|
|
|
*/
|
|
|
|
|
static int
|
|
|
|
|
ccr_count_sgl(struct sglist *sg, int maxsegsize)
|
|
|
|
|
{
|
|
|
|
|
int i, nsegs;
|
|
|
|
|
|
|
|
|
|
nsegs = 0;
|
|
|
|
|
for (i = 0; i < sg->sg_nseg; i++)
|
|
|
|
|
nsegs += howmany(sg->sg_segs[i].ss_len, maxsegsize);
|
|
|
|
|
return (nsegs);
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
/* These functions deal with PHYS_DSGL for the reply buffer. */
|
|
|
|
|
static inline int
|
|
|
|
|
ccr_phys_dsgl_len(int nsegs)
|
|
|
|
|
{
|
|
|
|
|
int len;
|
|
|
|
|
|
|
|
|
|
len = (nsegs / 8) * sizeof(struct phys_sge_pairs);
|
|
|
|
|
if ((nsegs % 8) != 0) {
|
|
|
|
|
len += sizeof(uint16_t) * 8;
|
|
|
|
|
len += roundup2(nsegs % 8, 2) * sizeof(uint64_t);
|
|
|
|
|
}
|
|
|
|
|
return (len);
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static void
|
2020-06-25 20:01:31 -04:00
|
|
|
ccr_write_phys_dsgl(struct ccr_session *s, void *dst, int nsegs)
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
{
|
|
|
|
|
struct sglist *sg;
|
|
|
|
|
struct cpl_rx_phys_dsgl *cpl;
|
|
|
|
|
struct phys_sge_pairs *sgl;
|
|
|
|
|
vm_paddr_t paddr;
|
|
|
|
|
size_t seglen;
|
|
|
|
|
u_int i, j;
|
|
|
|
|
|
2020-06-25 20:01:31 -04:00
|
|
|
sg = s->sg_dsgl;
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
cpl = dst;
|
|
|
|
|
cpl->op_to_tid = htobe32(V_CPL_RX_PHYS_DSGL_OPCODE(CPL_RX_PHYS_DSGL) |
|
|
|
|
|
V_CPL_RX_PHYS_DSGL_ISRDMA(0));
|
|
|
|
|
cpl->pcirlxorder_to_noofsgentr = htobe32(
|
|
|
|
|
V_CPL_RX_PHYS_DSGL_PCIRLXORDER(0) |
|
|
|
|
|
V_CPL_RX_PHYS_DSGL_PCINOSNOOP(0) |
|
|
|
|
|
V_CPL_RX_PHYS_DSGL_PCITPHNTENB(0) | V_CPL_RX_PHYS_DSGL_DCAID(0) |
|
|
|
|
|
V_CPL_RX_PHYS_DSGL_NOOFSGENTR(nsegs));
|
|
|
|
|
cpl->rss_hdr_int.opcode = CPL_RX_PHYS_ADDR;
|
2020-04-10 18:27:45 -04:00
|
|
|
cpl->rss_hdr_int.qid = htobe16(s->port->rxq->iq.abs_id);
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
cpl->rss_hdr_int.hash_val = 0;
|
|
|
|
|
sgl = (struct phys_sge_pairs *)(cpl + 1);
|
|
|
|
|
j = 0;
|
|
|
|
|
for (i = 0; i < sg->sg_nseg; i++) {
|
|
|
|
|
seglen = sg->sg_segs[i].ss_len;
|
|
|
|
|
paddr = sg->sg_segs[i].ss_paddr;
|
|
|
|
|
do {
|
|
|
|
|
sgl->addr[j] = htobe64(paddr);
|
|
|
|
|
if (seglen > DSGL_SGE_MAXLEN) {
|
|
|
|
|
sgl->len[j] = htobe16(DSGL_SGE_MAXLEN);
|
|
|
|
|
paddr += DSGL_SGE_MAXLEN;
|
|
|
|
|
seglen -= DSGL_SGE_MAXLEN;
|
|
|
|
|
} else {
|
|
|
|
|
sgl->len[j] = htobe16(seglen);
|
|
|
|
|
seglen = 0;
|
|
|
|
|
}
|
|
|
|
|
j++;
|
|
|
|
|
if (j == 8) {
|
|
|
|
|
sgl++;
|
|
|
|
|
j = 0;
|
|
|
|
|
}
|
|
|
|
|
} while (seglen != 0);
|
|
|
|
|
}
|
|
|
|
|
MPASS(j + 8 * (sgl - (struct phys_sge_pairs *)(cpl + 1)) == nsegs);
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
/* These functions deal with the ULPTX_SGL for input payload. */
|
|
|
|
|
static inline int
|
|
|
|
|
ccr_ulptx_sgl_len(int nsegs)
|
|
|
|
|
{
|
|
|
|
|
u_int n;
|
|
|
|
|
|
|
|
|
|
nsegs--; /* first segment is part of ulptx_sgl */
|
|
|
|
|
n = sizeof(struct ulptx_sgl) + 8 * ((3 * nsegs) / 2 + (nsegs & 1));
|
|
|
|
|
return (roundup2(n, 16));
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static void
|
2020-06-25 20:01:31 -04:00
|
|
|
ccr_write_ulptx_sgl(struct ccr_session *s, void *dst, int nsegs)
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
{
|
|
|
|
|
struct ulptx_sgl *usgl;
|
|
|
|
|
struct sglist *sg;
|
|
|
|
|
struct sglist_seg *ss;
|
|
|
|
|
int i;
|
|
|
|
|
|
2020-06-25 20:01:31 -04:00
|
|
|
sg = s->sg_ulptx;
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
MPASS(nsegs == sg->sg_nseg);
|
|
|
|
|
ss = &sg->sg_segs[0];
|
|
|
|
|
usgl = dst;
|
|
|
|
|
usgl->cmd_nsge = htobe32(V_ULPTX_CMD(ULP_TX_SC_DSGL) |
|
|
|
|
|
V_ULPTX_NSGE(nsegs));
|
|
|
|
|
usgl->len0 = htobe32(ss->ss_len);
|
|
|
|
|
usgl->addr0 = htobe64(ss->ss_paddr);
|
|
|
|
|
ss++;
|
|
|
|
|
for (i = 0; i < sg->sg_nseg - 1; i++) {
|
|
|
|
|
usgl->sge[i / 2].len[i & 1] = htobe32(ss->ss_len);
|
|
|
|
|
usgl->sge[i / 2].addr[i & 1] = htobe64(ss->ss_paddr);
|
|
|
|
|
ss++;
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static bool
|
|
|
|
|
ccr_use_imm_data(u_int transhdr_len, u_int input_len)
|
|
|
|
|
{
|
|
|
|
|
|
|
|
|
|
if (input_len > CRYPTO_MAX_IMM_TX_PKT_LEN)
|
|
|
|
|
return (false);
|
|
|
|
|
if (roundup2(transhdr_len, 16) + roundup2(input_len, 16) >
|
|
|
|
|
SGE_MAX_WR_LEN)
|
|
|
|
|
return (false);
|
|
|
|
|
return (true);
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static void
|
2020-04-10 18:27:45 -04:00
|
|
|
ccr_populate_wreq(struct ccr_softc *sc, struct ccr_session *s,
|
|
|
|
|
struct chcr_wr *crwr, u_int kctx_len, u_int wr_len, u_int imm_len,
|
|
|
|
|
u_int sgl_len, u_int hash_size, struct cryptop *crp)
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
{
|
2019-04-24 19:31:46 -04:00
|
|
|
u_int cctx_size, idata_len;
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
|
|
|
|
|
cctx_size = sizeof(struct _key_ctx) + kctx_len;
|
|
|
|
|
crwr->wreq.op_to_cctx_size = htobe32(
|
|
|
|
|
V_FW_CRYPTO_LOOKASIDE_WR_OPCODE(FW_CRYPTO_LOOKASIDE_WR) |
|
|
|
|
|
V_FW_CRYPTO_LOOKASIDE_WR_COMPL(0) |
|
|
|
|
|
V_FW_CRYPTO_LOOKASIDE_WR_IMM_LEN(imm_len) |
|
|
|
|
|
V_FW_CRYPTO_LOOKASIDE_WR_CCTX_LOC(1) |
|
|
|
|
|
V_FW_CRYPTO_LOOKASIDE_WR_CCTX_SIZE(cctx_size >> 4));
|
|
|
|
|
crwr->wreq.len16_pkd = htobe32(
|
|
|
|
|
V_FW_CRYPTO_LOOKASIDE_WR_LEN16(wr_len / 16));
|
2018-07-17 20:56:25 -04:00
|
|
|
crwr->wreq.session_id = 0;
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
crwr->wreq.rx_chid_to_rx_q_id = htobe32(
|
2020-04-10 18:27:45 -04:00
|
|
|
V_FW_CRYPTO_LOOKASIDE_WR_RX_CHID(s->port->tx_channel_id) |
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
V_FW_CRYPTO_LOOKASIDE_WR_LCB(0) |
|
|
|
|
|
V_FW_CRYPTO_LOOKASIDE_WR_PHASH(0) |
|
2018-01-24 15:06:02 -05:00
|
|
|
V_FW_CRYPTO_LOOKASIDE_WR_IV(IV_NOP) |
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
V_FW_CRYPTO_LOOKASIDE_WR_FQIDX(0) |
|
|
|
|
|
V_FW_CRYPTO_LOOKASIDE_WR_TX_CH(0) |
|
2020-04-10 18:27:45 -04:00
|
|
|
V_FW_CRYPTO_LOOKASIDE_WR_RX_Q_ID(s->port->rxq->iq.abs_id));
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
crwr->wreq.key_addr = 0;
|
|
|
|
|
crwr->wreq.pld_size_hash_size = htobe32(
|
|
|
|
|
V_FW_CRYPTO_LOOKASIDE_WR_PLD_SIZE(sgl_len) |
|
|
|
|
|
V_FW_CRYPTO_LOOKASIDE_WR_HASH_SIZE(hash_size));
|
|
|
|
|
crwr->wreq.cookie = htobe64((uintptr_t)crp);
|
|
|
|
|
|
|
|
|
|
crwr->ulptx.cmd_dest = htobe32(V_ULPTX_CMD(ULP_TX_PKT) |
|
|
|
|
|
V_ULP_TXPKT_DATAMODIFY(0) |
|
2020-04-10 18:27:45 -04:00
|
|
|
V_ULP_TXPKT_CHANNELID(s->port->tx_channel_id) |
|
|
|
|
|
V_ULP_TXPKT_DEST(0) |
|
|
|
|
|
V_ULP_TXPKT_FID(s->port->rxq->iq.abs_id) | V_ULP_TXPKT_RO(1));
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
crwr->ulptx.len = htobe32(
|
|
|
|
|
((wr_len - sizeof(struct fw_crypto_lookaside_wr)) / 16));
|
|
|
|
|
|
|
|
|
|
crwr->sc_imm.cmd_more = htobe32(V_ULPTX_CMD(ULP_TX_SC_IMM) |
|
2019-04-24 19:31:46 -04:00
|
|
|
V_ULP_TX_SC_MORE(sgl_len != 0 ? 1 : 0));
|
|
|
|
|
idata_len = wr_len - offsetof(struct chcr_wr, sec_cpl) - sgl_len;
|
|
|
|
|
if (imm_len % 16 != 0)
|
|
|
|
|
idata_len -= 16 - imm_len % 16;
|
|
|
|
|
crwr->sc_imm.len = htobe32(idata_len);
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static int
|
2018-10-29 18:24:31 -04:00
|
|
|
ccr_hash(struct ccr_softc *sc, struct ccr_session *s, struct cryptop *crp)
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
{
|
|
|
|
|
struct chcr_wr *crwr;
|
|
|
|
|
struct wrqe *wr;
|
|
|
|
|
struct auth_hash *axf;
|
|
|
|
|
char *dst;
|
|
|
|
|
u_int hash_size_in_response, kctx_flits, kctx_len, transhdr_len, wr_len;
|
2018-10-29 18:24:31 -04:00
|
|
|
u_int hmac_ctrl, imm_len, iopad_size;
|
|
|
|
|
int error, sgl_nsegs, sgl_len, use_opad;
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
|
2017-05-26 16:20:40 -04:00
|
|
|
/* Reject requests with too large of an input buffer. */
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
if (crp->crp_payload_length > MAX_REQUEST_SIZE)
|
2017-05-26 16:20:40 -04:00
|
|
|
return (EFBIG);
|
|
|
|
|
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
axf = s->hmac.auth_hash;
|
|
|
|
|
|
2018-10-29 18:24:31 -04:00
|
|
|
if (s->mode == HMAC) {
|
|
|
|
|
use_opad = 1;
|
2018-11-16 14:08:52 -05:00
|
|
|
hmac_ctrl = SCMD_HMAC_CTRL_NO_TRUNC;
|
2018-10-29 18:24:31 -04:00
|
|
|
} else {
|
|
|
|
|
use_opad = 0;
|
2018-11-16 14:08:52 -05:00
|
|
|
hmac_ctrl = SCMD_HMAC_CTRL_NOP;
|
2018-10-29 18:24:31 -04:00
|
|
|
}
|
|
|
|
|
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
/* PADs must be 128-bit aligned. */
|
|
|
|
|
iopad_size = roundup2(s->hmac.partial_digest_len, 16);
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
|
* The 'key' part of the context includes the aligned IPAD and
|
|
|
|
|
* OPAD.
|
|
|
|
|
*/
|
2018-10-29 18:24:31 -04:00
|
|
|
kctx_len = iopad_size;
|
|
|
|
|
if (use_opad)
|
|
|
|
|
kctx_len += iopad_size;
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
hash_size_in_response = axf->hashsize;
|
|
|
|
|
transhdr_len = HASH_TRANSHDR_SIZE(kctx_len);
|
|
|
|
|
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
if (crp->crp_payload_length == 0) {
|
2017-06-08 17:06:18 -04:00
|
|
|
imm_len = axf->blocksize;
|
|
|
|
|
sgl_nsegs = 0;
|
|
|
|
|
sgl_len = 0;
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
} else if (ccr_use_imm_data(transhdr_len, crp->crp_payload_length)) {
|
|
|
|
|
imm_len = crp->crp_payload_length;
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
sgl_nsegs = 0;
|
|
|
|
|
sgl_len = 0;
|
|
|
|
|
} else {
|
|
|
|
|
imm_len = 0;
|
2020-06-25 20:01:31 -04:00
|
|
|
sglist_reset(s->sg_ulptx);
|
|
|
|
|
error = sglist_append_sglist(s->sg_ulptx, s->sg_input,
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
crp->crp_payload_start, crp->crp_payload_length);
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
if (error)
|
|
|
|
|
return (error);
|
2020-06-25 20:01:31 -04:00
|
|
|
sgl_nsegs = s->sg_ulptx->sg_nseg;
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
sgl_len = ccr_ulptx_sgl_len(sgl_nsegs);
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
wr_len = roundup2(transhdr_len, 16) + roundup2(imm_len, 16) + sgl_len;
|
2018-01-24 15:12:00 -05:00
|
|
|
if (wr_len > SGE_MAX_WR_LEN)
|
|
|
|
|
return (EFBIG);
|
2020-04-10 18:27:45 -04:00
|
|
|
wr = alloc_wrqe(wr_len, s->port->txq);
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
if (wr == NULL) {
|
2020-06-25 20:01:31 -04:00
|
|
|
counter_u64_add(sc->stats_wr_nomem, 1);
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
return (ENOMEM);
|
|
|
|
|
}
|
|
|
|
|
crwr = wrtod(wr);
|
|
|
|
|
memset(crwr, 0, wr_len);
|
|
|
|
|
|
2020-04-10 18:27:45 -04:00
|
|
|
ccr_populate_wreq(sc, s, crwr, kctx_len, wr_len, imm_len, sgl_len,
|
2018-01-24 15:06:02 -05:00
|
|
|
hash_size_in_response, crp);
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
|
|
|
|
|
crwr->sec_cpl.op_ivinsrtofst = htobe32(
|
|
|
|
|
V_CPL_TX_SEC_PDU_OPCODE(CPL_TX_SEC_PDU) |
|
2020-04-10 18:27:45 -04:00
|
|
|
V_CPL_TX_SEC_PDU_RXCHID(s->port->tx_channel_id) |
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
V_CPL_TX_SEC_PDU_ACKFOLLOWS(0) | V_CPL_TX_SEC_PDU_ULPTXLPBK(1) |
|
|
|
|
|
V_CPL_TX_SEC_PDU_CPLLEN(2) | V_CPL_TX_SEC_PDU_PLACEHOLDER(0) |
|
|
|
|
|
V_CPL_TX_SEC_PDU_IVINSRTOFST(0));
|
|
|
|
|
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
crwr->sec_cpl.pldlen = htobe32(crp->crp_payload_length == 0 ?
|
|
|
|
|
axf->blocksize : crp->crp_payload_length);
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
|
|
|
|
|
crwr->sec_cpl.cipherstop_lo_authinsert = htobe32(
|
|
|
|
|
V_CPL_TX_SEC_PDU_AUTHSTART(1) | V_CPL_TX_SEC_PDU_AUTHSTOP(0));
|
|
|
|
|
|
|
|
|
|
/* These two flits are actually a CPL_TLS_TX_SCMD_FMT. */
|
|
|
|
|
crwr->sec_cpl.seqno_numivs = htobe32(
|
|
|
|
|
V_SCMD_SEQ_NO_CTRL(0) |
|
2018-11-16 14:08:52 -05:00
|
|
|
V_SCMD_PROTO_VERSION(SCMD_PROTO_VERSION_GENERIC) |
|
|
|
|
|
V_SCMD_CIPH_MODE(SCMD_CIPH_MODE_NOP) |
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
V_SCMD_AUTH_MODE(s->hmac.auth_mode) |
|
2018-10-29 18:24:31 -04:00
|
|
|
V_SCMD_HMAC_CTRL(hmac_ctrl));
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
crwr->sec_cpl.ivgen_hdrlen = htobe32(
|
2017-06-08 17:06:18 -04:00
|
|
|
V_SCMD_LAST_FRAG(0) |
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
V_SCMD_MORE_FRAGS(crp->crp_payload_length == 0 ? 1 : 0) |
|
|
|
|
|
V_SCMD_MAC_ONLY(1));
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
|
2019-11-12 19:53:45 -05:00
|
|
|
memcpy(crwr->key_ctx.key, s->hmac.pads, kctx_len);
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
|
|
|
|
|
/* XXX: F_KEY_CONTEXT_SALT_PRESENT set, but 'salt' not set. */
|
|
|
|
|
kctx_flits = (sizeof(struct _key_ctx) + kctx_len) / 16;
|
|
|
|
|
crwr->key_ctx.ctx_hdr = htobe32(V_KEY_CONTEXT_CTX_LEN(kctx_flits) |
|
2018-10-29 18:24:31 -04:00
|
|
|
V_KEY_CONTEXT_OPAD_PRESENT(use_opad) |
|
|
|
|
|
V_KEY_CONTEXT_SALT_PRESENT(1) |
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
V_KEY_CONTEXT_CK_SIZE(CHCR_KEYCTX_NO_KEY) |
|
|
|
|
|
V_KEY_CONTEXT_MK_SIZE(s->hmac.mk_size) | V_KEY_CONTEXT_VALID(1));
|
|
|
|
|
|
|
|
|
|
dst = (char *)(crwr + 1) + kctx_len + DUMMY_BYTES;
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
if (crp->crp_payload_length == 0) {
|
2017-06-08 17:06:18 -04:00
|
|
|
dst[0] = 0x80;
|
2019-04-24 19:18:10 -04:00
|
|
|
if (s->mode == HMAC)
|
|
|
|
|
*(uint64_t *)(dst + axf->blocksize - sizeof(uint64_t)) =
|
|
|
|
|
htobe64(axf->blocksize << 3);
|
2017-06-08 17:06:18 -04:00
|
|
|
} else if (imm_len != 0)
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
crypto_copydata(crp, crp->crp_payload_start,
|
|
|
|
|
crp->crp_payload_length, dst);
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
else
|
2020-06-25 20:01:31 -04:00
|
|
|
ccr_write_ulptx_sgl(s, dst, sgl_nsegs);
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
|
|
|
|
|
/* XXX: TODO backpressure */
|
|
|
|
|
t4_wrq_tx(sc->adapter, wr);
|
|
|
|
|
|
|
|
|
|
return (0);
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static int
|
2018-10-29 18:24:31 -04:00
|
|
|
ccr_hash_done(struct ccr_softc *sc, struct ccr_session *s, struct cryptop *crp,
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
const struct cpl_fw6_pld *cpl, int error)
|
|
|
|
|
{
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
uint8_t hash[HASH_MAX_LEN];
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
if (error)
|
|
|
|
|
return (error);
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
if (crp->crp_op & CRYPTO_OP_VERIFY_DIGEST) {
|
|
|
|
|
crypto_copydata(crp, crp->crp_digest_start, s->hmac.hash_len,
|
|
|
|
|
hash);
|
|
|
|
|
if (timingsafe_bcmp((cpl + 1), hash, s->hmac.hash_len) != 0)
|
|
|
|
|
return (EBADMSG);
|
|
|
|
|
} else
|
|
|
|
|
crypto_copyback(crp, crp->crp_digest_start, s->hmac.hash_len,
|
|
|
|
|
(cpl + 1));
|
|
|
|
|
return (0);
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static int
|
2018-07-17 20:56:25 -04:00
|
|
|
ccr_blkcipher(struct ccr_softc *sc, struct ccr_session *s, struct cryptop *crp)
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
{
|
|
|
|
|
char iv[CHCR_MAX_CRYPTO_IV_LEN];
|
|
|
|
|
struct chcr_wr *crwr;
|
|
|
|
|
struct wrqe *wr;
|
|
|
|
|
char *dst;
|
2018-01-24 15:04:08 -05:00
|
|
|
u_int kctx_len, key_half, op_type, transhdr_len, wr_len;
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
u_int imm_len, iv_len;
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
int dsgl_nsegs, dsgl_len;
|
|
|
|
|
int sgl_nsegs, sgl_len;
|
|
|
|
|
int error;
|
|
|
|
|
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
if (s->blkcipher.key_len == 0 || crp->crp_payload_length == 0)
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
return (EINVAL);
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
if (s->blkcipher.cipher_mode == SCMD_CIPH_MODE_AES_CBC &&
|
|
|
|
|
(crp->crp_payload_length % AES_BLOCK_LEN) != 0)
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
return (EINVAL);
|
|
|
|
|
|
2017-05-26 16:20:40 -04:00
|
|
|
/* Reject requests with too large of an input buffer. */
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
if (crp->crp_payload_length > MAX_REQUEST_SIZE)
|
2017-05-26 16:20:40 -04:00
|
|
|
return (EFBIG);
|
|
|
|
|
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
if (CRYPTO_OP_IS_ENCRYPT(crp->crp_op))
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
op_type = CHCR_ENCRYPT_OP;
|
2018-01-24 15:15:49 -05:00
|
|
|
else
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
op_type = CHCR_DECRYPT_OP;
|
2020-05-25 18:23:13 -04:00
|
|
|
|
2020-06-25 20:01:31 -04:00
|
|
|
sglist_reset(s->sg_dsgl);
|
2020-05-25 18:23:13 -04:00
|
|
|
if (CRYPTO_HAS_OUTPUT_BUFFER(crp))
|
2020-06-25 20:01:31 -04:00
|
|
|
error = sglist_append_sglist(s->sg_dsgl, s->sg_output,
|
2020-05-25 18:23:13 -04:00
|
|
|
crp->crp_payload_output_start, crp->crp_payload_length);
|
|
|
|
|
else
|
2020-06-25 20:01:31 -04:00
|
|
|
error = sglist_append_sglist(s->sg_dsgl, s->sg_input,
|
2020-05-25 18:23:13 -04:00
|
|
|
crp->crp_payload_start, crp->crp_payload_length);
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
if (error)
|
|
|
|
|
return (error);
|
2020-06-25 20:01:31 -04:00
|
|
|
dsgl_nsegs = ccr_count_sgl(s->sg_dsgl, DSGL_SGE_MAXLEN);
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
if (dsgl_nsegs > MAX_RX_PHYS_DSGL_SGE)
|
|
|
|
|
return (EFBIG);
|
|
|
|
|
dsgl_len = ccr_phys_dsgl_len(dsgl_nsegs);
|
|
|
|
|
|
|
|
|
|
/* The 'key' must be 128-bit aligned. */
|
|
|
|
|
kctx_len = roundup2(s->blkcipher.key_len, 16);
|
|
|
|
|
transhdr_len = CIPHER_TRANSHDR_SIZE(kctx_len, dsgl_len);
|
|
|
|
|
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
/* For AES-XTS we send a 16-byte IV in the work request. */
|
|
|
|
|
if (s->blkcipher.cipher_mode == SCMD_CIPH_MODE_AES_XTS)
|
|
|
|
|
iv_len = AES_BLOCK_LEN;
|
|
|
|
|
else
|
|
|
|
|
iv_len = s->blkcipher.iv_len;
|
|
|
|
|
|
|
|
|
|
if (ccr_use_imm_data(transhdr_len, crp->crp_payload_length + iv_len)) {
|
|
|
|
|
imm_len = crp->crp_payload_length;
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
sgl_nsegs = 0;
|
|
|
|
|
sgl_len = 0;
|
|
|
|
|
} else {
|
|
|
|
|
imm_len = 0;
|
2020-06-25 20:01:31 -04:00
|
|
|
sglist_reset(s->sg_ulptx);
|
|
|
|
|
error = sglist_append_sglist(s->sg_ulptx, s->sg_input,
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
crp->crp_payload_start, crp->crp_payload_length);
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
if (error)
|
|
|
|
|
return (error);
|
2020-06-25 20:01:31 -04:00
|
|
|
sgl_nsegs = s->sg_ulptx->sg_nseg;
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
sgl_len = ccr_ulptx_sgl_len(sgl_nsegs);
|
|
|
|
|
}
|
|
|
|
|
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
wr_len = roundup2(transhdr_len, 16) + iv_len +
|
2018-01-24 15:04:08 -05:00
|
|
|
roundup2(imm_len, 16) + sgl_len;
|
2018-01-24 15:12:00 -05:00
|
|
|
if (wr_len > SGE_MAX_WR_LEN)
|
|
|
|
|
return (EFBIG);
|
2020-04-10 18:27:45 -04:00
|
|
|
wr = alloc_wrqe(wr_len, s->port->txq);
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
if (wr == NULL) {
|
2020-06-25 20:01:31 -04:00
|
|
|
counter_u64_add(sc->stats_wr_nomem, 1);
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
return (ENOMEM);
|
|
|
|
|
}
|
|
|
|
|
crwr = wrtod(wr);
|
|
|
|
|
memset(crwr, 0, wr_len);
|
|
|
|
|
|
2020-04-20 18:24:49 -04:00
|
|
|
crypto_read_iv(crp, iv);
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
|
|
|
|
|
/* Zero the remainder of the IV for AES-XTS. */
|
|
|
|
|
memset(iv + s->blkcipher.iv_len, 0, iv_len - s->blkcipher.iv_len);
|
2018-01-24 15:15:49 -05:00
|
|
|
|
2020-04-10 18:27:45 -04:00
|
|
|
ccr_populate_wreq(sc, s, crwr, kctx_len, wr_len, imm_len, sgl_len, 0,
|
2018-01-24 15:06:02 -05:00
|
|
|
crp);
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
|
|
|
|
|
crwr->sec_cpl.op_ivinsrtofst = htobe32(
|
|
|
|
|
V_CPL_TX_SEC_PDU_OPCODE(CPL_TX_SEC_PDU) |
|
2020-04-10 18:27:45 -04:00
|
|
|
V_CPL_TX_SEC_PDU_RXCHID(s->port->tx_channel_id) |
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
V_CPL_TX_SEC_PDU_ACKFOLLOWS(0) | V_CPL_TX_SEC_PDU_ULPTXLPBK(1) |
|
|
|
|
|
V_CPL_TX_SEC_PDU_CPLLEN(2) | V_CPL_TX_SEC_PDU_PLACEHOLDER(0) |
|
|
|
|
|
V_CPL_TX_SEC_PDU_IVINSRTOFST(1));
|
|
|
|
|
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
crwr->sec_cpl.pldlen = htobe32(iv_len + crp->crp_payload_length);
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
|
|
|
|
|
crwr->sec_cpl.aadstart_cipherstop_hi = htobe32(
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
V_CPL_TX_SEC_PDU_CIPHERSTART(iv_len + 1) |
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
V_CPL_TX_SEC_PDU_CIPHERSTOP_HI(0));
|
|
|
|
|
crwr->sec_cpl.cipherstop_lo_authinsert = htobe32(
|
|
|
|
|
V_CPL_TX_SEC_PDU_CIPHERSTOP_LO(0));
|
|
|
|
|
|
|
|
|
|
/* These two flits are actually a CPL_TLS_TX_SCMD_FMT. */
|
|
|
|
|
crwr->sec_cpl.seqno_numivs = htobe32(
|
|
|
|
|
V_SCMD_SEQ_NO_CTRL(0) |
|
2018-11-16 14:08:52 -05:00
|
|
|
V_SCMD_PROTO_VERSION(SCMD_PROTO_VERSION_GENERIC) |
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
V_SCMD_ENC_DEC_CTRL(op_type) |
|
|
|
|
|
V_SCMD_CIPH_MODE(s->blkcipher.cipher_mode) |
|
2018-11-16 14:08:52 -05:00
|
|
|
V_SCMD_AUTH_MODE(SCMD_AUTH_MODE_NOP) |
|
|
|
|
|
V_SCMD_HMAC_CTRL(SCMD_HMAC_CTRL_NOP) |
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
V_SCMD_IV_SIZE(iv_len / 2) |
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
V_SCMD_NUM_IVS(0));
|
|
|
|
|
crwr->sec_cpl.ivgen_hdrlen = htobe32(
|
|
|
|
|
V_SCMD_IV_GEN_CTRL(0) |
|
|
|
|
|
V_SCMD_MORE_FRAGS(0) | V_SCMD_LAST_FRAG(0) | V_SCMD_MAC_ONLY(0) |
|
|
|
|
|
V_SCMD_AADIVDROP(1) | V_SCMD_HDR_LEN(dsgl_len));
|
|
|
|
|
|
|
|
|
|
crwr->key_ctx.ctx_hdr = s->blkcipher.key_ctx_hdr;
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
switch (s->blkcipher.cipher_mode) {
|
|
|
|
|
case SCMD_CIPH_MODE_AES_CBC:
|
|
|
|
|
if (CRYPTO_OP_IS_ENCRYPT(crp->crp_op))
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
memcpy(crwr->key_ctx.key, s->blkcipher.enckey,
|
|
|
|
|
s->blkcipher.key_len);
|
|
|
|
|
else
|
|
|
|
|
memcpy(crwr->key_ctx.key, s->blkcipher.deckey,
|
|
|
|
|
s->blkcipher.key_len);
|
|
|
|
|
break;
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
case SCMD_CIPH_MODE_AES_CTR:
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
memcpy(crwr->key_ctx.key, s->blkcipher.enckey,
|
|
|
|
|
s->blkcipher.key_len);
|
|
|
|
|
break;
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
case SCMD_CIPH_MODE_AES_XTS:
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
key_half = s->blkcipher.key_len / 2;
|
|
|
|
|
memcpy(crwr->key_ctx.key, s->blkcipher.enckey + key_half,
|
|
|
|
|
key_half);
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
if (CRYPTO_OP_IS_ENCRYPT(crp->crp_op))
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
memcpy(crwr->key_ctx.key + key_half,
|
|
|
|
|
s->blkcipher.enckey, key_half);
|
|
|
|
|
else
|
|
|
|
|
memcpy(crwr->key_ctx.key + key_half,
|
|
|
|
|
s->blkcipher.deckey, key_half);
|
|
|
|
|
break;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
dst = (char *)(crwr + 1) + kctx_len;
|
2020-06-25 20:01:31 -04:00
|
|
|
ccr_write_phys_dsgl(s, dst, dsgl_nsegs);
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
dst += sizeof(struct cpl_rx_phys_dsgl) + dsgl_len;
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
memcpy(dst, iv, iv_len);
|
|
|
|
|
dst += iv_len;
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
if (imm_len != 0)
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
crypto_copydata(crp, crp->crp_payload_start,
|
|
|
|
|
crp->crp_payload_length, dst);
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
else
|
2020-06-25 20:01:31 -04:00
|
|
|
ccr_write_ulptx_sgl(s, dst, sgl_nsegs);
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
|
|
|
|
|
/* XXX: TODO backpressure */
|
|
|
|
|
t4_wrq_tx(sc->adapter, wr);
|
|
|
|
|
|
2020-06-03 18:19:52 -04:00
|
|
|
explicit_bzero(iv, sizeof(iv));
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
return (0);
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static int
|
|
|
|
|
ccr_blkcipher_done(struct ccr_softc *sc, struct ccr_session *s,
|
|
|
|
|
struct cryptop *crp, const struct cpl_fw6_pld *cpl, int error)
|
|
|
|
|
{
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
|
* The updated IV to permit chained requests is at
|
|
|
|
|
* cpl->data[2], but OCF doesn't permit chained requests.
|
|
|
|
|
*/
|
|
|
|
|
return (error);
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
|
* 'hashsize' is the length of a full digest. 'authsize' is the
|
|
|
|
|
* requested digest length for this operation which may be less
|
|
|
|
|
* than 'hashsize'.
|
|
|
|
|
*/
|
|
|
|
|
static int
|
|
|
|
|
ccr_hmac_ctrl(unsigned int hashsize, unsigned int authsize)
|
|
|
|
|
{
|
|
|
|
|
|
|
|
|
|
if (authsize == 10)
|
2018-11-16 14:08:52 -05:00
|
|
|
return (SCMD_HMAC_CTRL_TRUNC_RFC4366);
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
if (authsize == 12)
|
2018-11-16 14:08:52 -05:00
|
|
|
return (SCMD_HMAC_CTRL_IPSEC_96BIT);
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
if (authsize == hashsize / 2)
|
2018-11-16 14:08:52 -05:00
|
|
|
return (SCMD_HMAC_CTRL_DIV2);
|
|
|
|
|
return (SCMD_HMAC_CTRL_NO_TRUNC);
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static int
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
ccr_eta(struct ccr_softc *sc, struct ccr_session *s, struct cryptop *crp)
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
{
|
|
|
|
|
char iv[CHCR_MAX_CRYPTO_IV_LEN];
|
|
|
|
|
struct chcr_wr *crwr;
|
|
|
|
|
struct wrqe *wr;
|
|
|
|
|
struct auth_hash *axf;
|
|
|
|
|
char *dst;
|
2018-01-24 15:04:08 -05:00
|
|
|
u_int kctx_len, key_half, op_type, transhdr_len, wr_len;
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
u_int hash_size_in_response, imm_len, iopad_size, iv_len;
|
|
|
|
|
u_int aad_start, aad_stop;
|
|
|
|
|
u_int auth_insert;
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
u_int cipher_start, cipher_stop;
|
|
|
|
|
u_int hmac_ctrl, input_len;
|
|
|
|
|
int dsgl_nsegs, dsgl_len;
|
|
|
|
|
int sgl_nsegs, sgl_len;
|
|
|
|
|
int error;
|
|
|
|
|
|
2017-06-08 17:06:18 -04:00
|
|
|
/*
|
|
|
|
|
* If there is a need in the future, requests with an empty
|
|
|
|
|
* payload could be supported as HMAC-only requests.
|
|
|
|
|
*/
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
if (s->blkcipher.key_len == 0 || crp->crp_payload_length == 0)
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
return (EINVAL);
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
if (s->blkcipher.cipher_mode == SCMD_CIPH_MODE_AES_CBC &&
|
|
|
|
|
(crp->crp_payload_length % AES_BLOCK_LEN) != 0)
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
return (EINVAL);
|
|
|
|
|
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
/* For AES-XTS we send a 16-byte IV in the work request. */
|
|
|
|
|
if (s->blkcipher.cipher_mode == SCMD_CIPH_MODE_AES_XTS)
|
|
|
|
|
iv_len = AES_BLOCK_LEN;
|
|
|
|
|
else
|
|
|
|
|
iv_len = s->blkcipher.iv_len;
|
|
|
|
|
|
|
|
|
|
if (crp->crp_aad_length + iv_len > MAX_AAD_LEN)
|
2018-01-24 15:08:10 -05:00
|
|
|
return (EINVAL);
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
|
|
|
|
|
axf = s->hmac.auth_hash;
|
|
|
|
|
hash_size_in_response = s->hmac.hash_len;
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
if (CRYPTO_OP_IS_ENCRYPT(crp->crp_op))
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
op_type = CHCR_ENCRYPT_OP;
|
2018-01-24 15:15:49 -05:00
|
|
|
else
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
op_type = CHCR_DECRYPT_OP;
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
|
* The output buffer consists of the cipher text followed by
|
|
|
|
|
* the hash when encrypting. For decryption it only contains
|
|
|
|
|
* the plain text.
|
2018-01-24 15:11:00 -05:00
|
|
|
*
|
|
|
|
|
* Due to a firmware bug, the output buffer must include a
|
|
|
|
|
* dummy output buffer for the IV and AAD prior to the real
|
|
|
|
|
* output buffer.
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
*/
|
2017-05-26 16:20:40 -04:00
|
|
|
if (op_type == CHCR_ENCRYPT_OP) {
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
if (iv_len + crp->crp_aad_length + crp->crp_payload_length +
|
2018-01-24 15:11:00 -05:00
|
|
|
hash_size_in_response > MAX_REQUEST_SIZE)
|
2017-05-26 16:20:40 -04:00
|
|
|
return (EFBIG);
|
|
|
|
|
} else {
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
if (iv_len + crp->crp_aad_length + crp->crp_payload_length >
|
2018-01-24 15:11:00 -05:00
|
|
|
MAX_REQUEST_SIZE)
|
2017-05-26 16:20:40 -04:00
|
|
|
return (EFBIG);
|
|
|
|
|
}
|
2020-06-25 20:01:31 -04:00
|
|
|
sglist_reset(s->sg_dsgl);
|
|
|
|
|
error = sglist_append_sglist(s->sg_dsgl, sc->sg_iv_aad, 0,
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
iv_len + crp->crp_aad_length);
|
2018-01-24 15:11:00 -05:00
|
|
|
if (error)
|
|
|
|
|
return (error);
|
2020-05-25 18:23:13 -04:00
|
|
|
if (CRYPTO_HAS_OUTPUT_BUFFER(crp))
|
2020-06-25 20:01:31 -04:00
|
|
|
error = sglist_append_sglist(s->sg_dsgl, s->sg_output,
|
2020-05-25 18:23:13 -04:00
|
|
|
crp->crp_payload_output_start, crp->crp_payload_length);
|
|
|
|
|
else
|
2020-06-25 20:01:31 -04:00
|
|
|
error = sglist_append_sglist(s->sg_dsgl, s->sg_input,
|
2020-05-25 18:23:13 -04:00
|
|
|
crp->crp_payload_start, crp->crp_payload_length);
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
if (error)
|
|
|
|
|
return (error);
|
|
|
|
|
if (op_type == CHCR_ENCRYPT_OP) {
|
2020-05-25 18:23:13 -04:00
|
|
|
if (CRYPTO_HAS_OUTPUT_BUFFER(crp))
|
2020-06-25 20:01:31 -04:00
|
|
|
error = sglist_append_sglist(s->sg_dsgl, s->sg_output,
|
2020-05-25 18:23:13 -04:00
|
|
|
crp->crp_digest_start, hash_size_in_response);
|
|
|
|
|
else
|
2020-06-25 20:01:31 -04:00
|
|
|
error = sglist_append_sglist(s->sg_dsgl, s->sg_input,
|
2020-05-25 18:23:13 -04:00
|
|
|
crp->crp_digest_start, hash_size_in_response);
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
if (error)
|
|
|
|
|
return (error);
|
|
|
|
|
}
|
2020-06-25 20:01:31 -04:00
|
|
|
dsgl_nsegs = ccr_count_sgl(s->sg_dsgl, DSGL_SGE_MAXLEN);
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
if (dsgl_nsegs > MAX_RX_PHYS_DSGL_SGE)
|
|
|
|
|
return (EFBIG);
|
|
|
|
|
dsgl_len = ccr_phys_dsgl_len(dsgl_nsegs);
|
|
|
|
|
|
|
|
|
|
/* PADs must be 128-bit aligned. */
|
|
|
|
|
iopad_size = roundup2(s->hmac.partial_digest_len, 16);
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
|
* The 'key' part of the key context consists of the key followed
|
|
|
|
|
* by the IPAD and OPAD.
|
|
|
|
|
*/
|
|
|
|
|
kctx_len = roundup2(s->blkcipher.key_len, 16) + iopad_size * 2;
|
|
|
|
|
transhdr_len = CIPHER_TRANSHDR_SIZE(kctx_len, dsgl_len);
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
|
* The input buffer consists of the IV, any AAD, and then the
|
|
|
|
|
* cipher/plain text. For decryption requests the hash is
|
|
|
|
|
* appended after the cipher text.
|
2018-01-24 15:15:49 -05:00
|
|
|
*
|
|
|
|
|
* The IV is always stored at the start of the input buffer
|
|
|
|
|
* even though it may be duplicated in the payload. The
|
|
|
|
|
* crypto engine doesn't work properly if the IV offset points
|
|
|
|
|
* inside of the AAD region, so a second copy is always
|
|
|
|
|
* required.
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
*/
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
input_len = crp->crp_aad_length + crp->crp_payload_length;
|
2017-05-26 16:20:40 -04:00
|
|
|
|
|
|
|
|
/*
|
|
|
|
|
* The firmware hangs if sent a request which is a
|
|
|
|
|
* bit smaller than MAX_REQUEST_SIZE. In particular, the
|
|
|
|
|
* firmware appears to require 512 - 16 bytes of spare room
|
|
|
|
|
* along with the size of the hash even if the hash isn't
|
|
|
|
|
* included in the input buffer.
|
|
|
|
|
*/
|
|
|
|
|
if (input_len + roundup2(axf->hashsize, 16) + (512 - 16) >
|
|
|
|
|
MAX_REQUEST_SIZE)
|
|
|
|
|
return (EFBIG);
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
if (op_type == CHCR_DECRYPT_OP)
|
|
|
|
|
input_len += hash_size_in_response;
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
|
|
|
|
|
if (ccr_use_imm_data(transhdr_len, iv_len + input_len)) {
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
imm_len = input_len;
|
|
|
|
|
sgl_nsegs = 0;
|
|
|
|
|
sgl_len = 0;
|
|
|
|
|
} else {
|
|
|
|
|
imm_len = 0;
|
2020-06-25 20:01:31 -04:00
|
|
|
sglist_reset(s->sg_ulptx);
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
if (crp->crp_aad_length != 0) {
|
2020-06-22 19:41:33 -04:00
|
|
|
if (crp->crp_aad != NULL)
|
2020-06-25 20:01:31 -04:00
|
|
|
error = sglist_append(s->sg_ulptx,
|
2020-06-22 19:41:33 -04:00
|
|
|
crp->crp_aad, crp->crp_aad_length);
|
|
|
|
|
else
|
2020-06-25 20:01:31 -04:00
|
|
|
error = sglist_append_sglist(s->sg_ulptx,
|
|
|
|
|
s->sg_input, crp->crp_aad_start,
|
2020-06-22 19:41:33 -04:00
|
|
|
crp->crp_aad_length);
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
if (error)
|
|
|
|
|
return (error);
|
|
|
|
|
}
|
2020-06-25 20:01:31 -04:00
|
|
|
error = sglist_append_sglist(s->sg_ulptx, s->sg_input,
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
crp->crp_payload_start, crp->crp_payload_length);
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
if (error)
|
|
|
|
|
return (error);
|
|
|
|
|
if (op_type == CHCR_DECRYPT_OP) {
|
2020-06-25 20:01:31 -04:00
|
|
|
error = sglist_append_sglist(s->sg_ulptx, s->sg_input,
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
crp->crp_digest_start, hash_size_in_response);
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
if (error)
|
|
|
|
|
return (error);
|
|
|
|
|
}
|
2020-06-25 20:01:31 -04:00
|
|
|
sgl_nsegs = s->sg_ulptx->sg_nseg;
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
sgl_len = ccr_ulptx_sgl_len(sgl_nsegs);
|
|
|
|
|
}
|
|
|
|
|
|
2020-06-22 19:41:33 -04:00
|
|
|
/* Any AAD comes after the IV. */
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
if (crp->crp_aad_length != 0) {
|
|
|
|
|
aad_start = iv_len + 1;
|
|
|
|
|
aad_stop = aad_start + crp->crp_aad_length - 1;
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
} else {
|
|
|
|
|
aad_start = 0;
|
|
|
|
|
aad_stop = 0;
|
|
|
|
|
}
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
cipher_start = iv_len + crp->crp_aad_length + 1;
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
if (op_type == CHCR_DECRYPT_OP)
|
|
|
|
|
cipher_stop = hash_size_in_response;
|
|
|
|
|
else
|
|
|
|
|
cipher_stop = 0;
|
|
|
|
|
if (op_type == CHCR_DECRYPT_OP)
|
|
|
|
|
auth_insert = hash_size_in_response;
|
|
|
|
|
else
|
|
|
|
|
auth_insert = 0;
|
|
|
|
|
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
wr_len = roundup2(transhdr_len, 16) + iv_len + roundup2(imm_len, 16) +
|
|
|
|
|
sgl_len;
|
2018-01-24 15:12:00 -05:00
|
|
|
if (wr_len > SGE_MAX_WR_LEN)
|
|
|
|
|
return (EFBIG);
|
2020-04-10 18:27:45 -04:00
|
|
|
wr = alloc_wrqe(wr_len, s->port->txq);
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
if (wr == NULL) {
|
2020-06-25 20:01:31 -04:00
|
|
|
counter_u64_add(sc->stats_wr_nomem, 1);
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
return (ENOMEM);
|
|
|
|
|
}
|
|
|
|
|
crwr = wrtod(wr);
|
|
|
|
|
memset(crwr, 0, wr_len);
|
|
|
|
|
|
2020-04-20 18:24:49 -04:00
|
|
|
crypto_read_iv(crp, iv);
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
|
|
|
|
|
/* Zero the remainder of the IV for AES-XTS. */
|
|
|
|
|
memset(iv + s->blkcipher.iv_len, 0, iv_len - s->blkcipher.iv_len);
|
2018-01-24 15:15:49 -05:00
|
|
|
|
2020-04-10 18:27:45 -04:00
|
|
|
ccr_populate_wreq(sc, s, crwr, kctx_len, wr_len, imm_len, sgl_len,
|
2018-01-24 15:06:02 -05:00
|
|
|
op_type == CHCR_DECRYPT_OP ? hash_size_in_response : 0, crp);
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
|
|
|
|
|
crwr->sec_cpl.op_ivinsrtofst = htobe32(
|
|
|
|
|
V_CPL_TX_SEC_PDU_OPCODE(CPL_TX_SEC_PDU) |
|
2020-04-10 18:27:45 -04:00
|
|
|
V_CPL_TX_SEC_PDU_RXCHID(s->port->tx_channel_id) |
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
V_CPL_TX_SEC_PDU_ACKFOLLOWS(0) | V_CPL_TX_SEC_PDU_ULPTXLPBK(1) |
|
|
|
|
|
V_CPL_TX_SEC_PDU_CPLLEN(2) | V_CPL_TX_SEC_PDU_PLACEHOLDER(0) |
|
|
|
|
|
V_CPL_TX_SEC_PDU_IVINSRTOFST(1));
|
|
|
|
|
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
crwr->sec_cpl.pldlen = htobe32(iv_len + input_len);
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
|
|
|
|
|
crwr->sec_cpl.aadstart_cipherstop_hi = htobe32(
|
|
|
|
|
V_CPL_TX_SEC_PDU_AADSTART(aad_start) |
|
|
|
|
|
V_CPL_TX_SEC_PDU_AADSTOP(aad_stop) |
|
|
|
|
|
V_CPL_TX_SEC_PDU_CIPHERSTART(cipher_start) |
|
|
|
|
|
V_CPL_TX_SEC_PDU_CIPHERSTOP_HI(cipher_stop >> 4));
|
|
|
|
|
crwr->sec_cpl.cipherstop_lo_authinsert = htobe32(
|
|
|
|
|
V_CPL_TX_SEC_PDU_CIPHERSTOP_LO(cipher_stop & 0xf) |
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
V_CPL_TX_SEC_PDU_AUTHSTART(cipher_start) |
|
|
|
|
|
V_CPL_TX_SEC_PDU_AUTHSTOP(cipher_stop) |
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
V_CPL_TX_SEC_PDU_AUTHINSERT(auth_insert));
|
|
|
|
|
|
|
|
|
|
/* These two flits are actually a CPL_TLS_TX_SCMD_FMT. */
|
|
|
|
|
hmac_ctrl = ccr_hmac_ctrl(axf->hashsize, hash_size_in_response);
|
|
|
|
|
crwr->sec_cpl.seqno_numivs = htobe32(
|
|
|
|
|
V_SCMD_SEQ_NO_CTRL(0) |
|
2018-11-16 14:08:52 -05:00
|
|
|
V_SCMD_PROTO_VERSION(SCMD_PROTO_VERSION_GENERIC) |
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
V_SCMD_ENC_DEC_CTRL(op_type) |
|
|
|
|
|
V_SCMD_CIPH_AUTH_SEQ_CTRL(op_type == CHCR_ENCRYPT_OP ? 1 : 0) |
|
|
|
|
|
V_SCMD_CIPH_MODE(s->blkcipher.cipher_mode) |
|
|
|
|
|
V_SCMD_AUTH_MODE(s->hmac.auth_mode) |
|
|
|
|
|
V_SCMD_HMAC_CTRL(hmac_ctrl) |
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
V_SCMD_IV_SIZE(iv_len / 2) |
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
V_SCMD_NUM_IVS(0));
|
|
|
|
|
crwr->sec_cpl.ivgen_hdrlen = htobe32(
|
|
|
|
|
V_SCMD_IV_GEN_CTRL(0) |
|
|
|
|
|
V_SCMD_MORE_FRAGS(0) | V_SCMD_LAST_FRAG(0) | V_SCMD_MAC_ONLY(0) |
|
2018-01-24 15:11:00 -05:00
|
|
|
V_SCMD_AADIVDROP(0) | V_SCMD_HDR_LEN(dsgl_len));
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
|
|
|
|
|
crwr->key_ctx.ctx_hdr = s->blkcipher.key_ctx_hdr;
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
switch (s->blkcipher.cipher_mode) {
|
|
|
|
|
case SCMD_CIPH_MODE_AES_CBC:
|
|
|
|
|
if (CRYPTO_OP_IS_ENCRYPT(crp->crp_op))
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
memcpy(crwr->key_ctx.key, s->blkcipher.enckey,
|
|
|
|
|
s->blkcipher.key_len);
|
|
|
|
|
else
|
|
|
|
|
memcpy(crwr->key_ctx.key, s->blkcipher.deckey,
|
|
|
|
|
s->blkcipher.key_len);
|
|
|
|
|
break;
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
case SCMD_CIPH_MODE_AES_CTR:
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
memcpy(crwr->key_ctx.key, s->blkcipher.enckey,
|
|
|
|
|
s->blkcipher.key_len);
|
|
|
|
|
break;
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
case SCMD_CIPH_MODE_AES_XTS:
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
key_half = s->blkcipher.key_len / 2;
|
|
|
|
|
memcpy(crwr->key_ctx.key, s->blkcipher.enckey + key_half,
|
|
|
|
|
key_half);
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
if (CRYPTO_OP_IS_ENCRYPT(crp->crp_op))
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
memcpy(crwr->key_ctx.key + key_half,
|
|
|
|
|
s->blkcipher.enckey, key_half);
|
|
|
|
|
else
|
|
|
|
|
memcpy(crwr->key_ctx.key + key_half,
|
|
|
|
|
s->blkcipher.deckey, key_half);
|
|
|
|
|
break;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
dst = crwr->key_ctx.key + roundup2(s->blkcipher.key_len, 16);
|
2019-11-12 19:53:45 -05:00
|
|
|
memcpy(dst, s->hmac.pads, iopad_size * 2);
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
|
|
|
|
|
dst = (char *)(crwr + 1) + kctx_len;
|
2020-06-25 20:01:31 -04:00
|
|
|
ccr_write_phys_dsgl(s, dst, dsgl_nsegs);
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
dst += sizeof(struct cpl_rx_phys_dsgl) + dsgl_len;
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
memcpy(dst, iv, iv_len);
|
|
|
|
|
dst += iv_len;
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
if (imm_len != 0) {
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
if (crp->crp_aad_length != 0) {
|
2020-06-22 19:41:33 -04:00
|
|
|
if (crp->crp_aad != NULL)
|
|
|
|
|
memcpy(dst, crp->crp_aad, crp->crp_aad_length);
|
|
|
|
|
else
|
|
|
|
|
crypto_copydata(crp, crp->crp_aad_start,
|
|
|
|
|
crp->crp_aad_length, dst);
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
dst += crp->crp_aad_length;
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
}
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
crypto_copydata(crp, crp->crp_payload_start,
|
|
|
|
|
crp->crp_payload_length, dst);
|
|
|
|
|
dst += crp->crp_payload_length;
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
if (op_type == CHCR_DECRYPT_OP)
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
crypto_copydata(crp, crp->crp_digest_start,
|
|
|
|
|
hash_size_in_response, dst);
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
} else
|
2020-06-25 20:01:31 -04:00
|
|
|
ccr_write_ulptx_sgl(s, dst, sgl_nsegs);
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
|
|
|
|
|
/* XXX: TODO backpressure */
|
|
|
|
|
t4_wrq_tx(sc->adapter, wr);
|
|
|
|
|
|
2020-06-03 18:19:52 -04:00
|
|
|
explicit_bzero(iv, sizeof(iv));
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
return (0);
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static int
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
ccr_eta_done(struct ccr_softc *sc, struct ccr_session *s,
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
struct cryptop *crp, const struct cpl_fw6_pld *cpl, int error)
|
|
|
|
|
{
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
|
* The updated IV to permit chained requests is at
|
|
|
|
|
* cpl->data[2], but OCF doesn't permit chained requests.
|
|
|
|
|
*/
|
|
|
|
|
return (error);
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static int
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
ccr_gcm(struct ccr_softc *sc, struct ccr_session *s, struct cryptop *crp)
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
{
|
|
|
|
|
char iv[CHCR_MAX_CRYPTO_IV_LEN];
|
|
|
|
|
struct chcr_wr *crwr;
|
|
|
|
|
struct wrqe *wr;
|
|
|
|
|
char *dst;
|
2018-01-24 15:04:08 -05:00
|
|
|
u_int iv_len, kctx_len, op_type, transhdr_len, wr_len;
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
u_int hash_size_in_response, imm_len;
|
|
|
|
|
u_int aad_start, aad_stop, cipher_start, cipher_stop, auth_insert;
|
|
|
|
|
u_int hmac_ctrl, input_len;
|
|
|
|
|
int dsgl_nsegs, dsgl_len;
|
|
|
|
|
int sgl_nsegs, sgl_len;
|
|
|
|
|
int error;
|
|
|
|
|
|
|
|
|
|
if (s->blkcipher.key_len == 0)
|
|
|
|
|
return (EINVAL);
|
|
|
|
|
|
2018-01-24 15:14:57 -05:00
|
|
|
/*
|
|
|
|
|
* The crypto engine doesn't handle GCM requests with an empty
|
|
|
|
|
* payload, so handle those in software instead.
|
|
|
|
|
*/
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
if (crp->crp_payload_length == 0)
|
2018-01-24 15:14:57 -05:00
|
|
|
return (EMSGSIZE);
|
|
|
|
|
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
if (crp->crp_aad_length + AES_BLOCK_LEN > MAX_AAD_LEN)
|
2018-01-24 15:14:57 -05:00
|
|
|
return (EMSGSIZE);
|
2018-01-24 15:08:10 -05:00
|
|
|
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
hash_size_in_response = s->gmac.hash_len;
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
if (CRYPTO_OP_IS_ENCRYPT(crp->crp_op))
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
op_type = CHCR_ENCRYPT_OP;
|
2018-01-24 15:15:49 -05:00
|
|
|
else
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
op_type = CHCR_DECRYPT_OP;
|
|
|
|
|
|
|
|
|
|
/*
|
2018-01-24 15:15:49 -05:00
|
|
|
* The IV handling for GCM in OCF is a bit more complicated in
|
|
|
|
|
* that IPSec provides a full 16-byte IV (including the
|
|
|
|
|
* counter), whereas the /dev/crypto interface sometimes
|
|
|
|
|
* provides a full 16-byte IV (if no IV is provided in the
|
|
|
|
|
* ioctl) and sometimes a 12-byte IV (if the IV was explicit).
|
|
|
|
|
*
|
|
|
|
|
* When provided a 12-byte IV, assume the IV is really 16 bytes
|
|
|
|
|
* with a counter in the last 4 bytes initialized to 1.
|
|
|
|
|
*
|
|
|
|
|
* While iv_len is checked below, the value is currently
|
|
|
|
|
* always set to 12 when creating a GCM session in this driver
|
|
|
|
|
* due to limitations in OCF (there is no way to know what the
|
|
|
|
|
* IV length of a given request will be). This means that the
|
|
|
|
|
* driver always assumes as 12-byte IV for now.
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
*/
|
2018-01-24 15:15:49 -05:00
|
|
|
if (s->blkcipher.iv_len == 12)
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
iv_len = AES_BLOCK_LEN;
|
2018-01-24 15:15:49 -05:00
|
|
|
else
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
iv_len = s->blkcipher.iv_len;
|
|
|
|
|
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
/*
|
|
|
|
|
* GCM requests should always provide an explicit IV.
|
|
|
|
|
*/
|
|
|
|
|
if ((crp->crp_flags & CRYPTO_F_IV_SEPARATE) == 0)
|
|
|
|
|
return (EINVAL);
|
|
|
|
|
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
/*
|
|
|
|
|
* The output buffer consists of the cipher text followed by
|
|
|
|
|
* the tag when encrypting. For decryption it only contains
|
|
|
|
|
* the plain text.
|
2018-01-24 15:11:00 -05:00
|
|
|
*
|
|
|
|
|
* Due to a firmware bug, the output buffer must include a
|
|
|
|
|
* dummy output buffer for the IV and AAD prior to the real
|
|
|
|
|
* output buffer.
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
*/
|
2017-05-26 16:20:40 -04:00
|
|
|
if (op_type == CHCR_ENCRYPT_OP) {
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
if (iv_len + crp->crp_aad_length + crp->crp_payload_length +
|
2018-01-24 15:11:00 -05:00
|
|
|
hash_size_in_response > MAX_REQUEST_SIZE)
|
2017-05-26 16:20:40 -04:00
|
|
|
return (EFBIG);
|
|
|
|
|
} else {
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
if (iv_len + crp->crp_aad_length + crp->crp_payload_length >
|
|
|
|
|
MAX_REQUEST_SIZE)
|
2017-05-26 16:20:40 -04:00
|
|
|
return (EFBIG);
|
|
|
|
|
}
|
2020-06-25 20:01:31 -04:00
|
|
|
sglist_reset(s->sg_dsgl);
|
|
|
|
|
error = sglist_append_sglist(s->sg_dsgl, sc->sg_iv_aad, 0, iv_len +
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
crp->crp_aad_length);
|
2018-01-24 15:11:00 -05:00
|
|
|
if (error)
|
|
|
|
|
return (error);
|
2020-05-25 18:23:13 -04:00
|
|
|
if (CRYPTO_HAS_OUTPUT_BUFFER(crp))
|
2020-06-25 20:01:31 -04:00
|
|
|
error = sglist_append_sglist(s->sg_dsgl, s->sg_output,
|
2020-05-25 18:23:13 -04:00
|
|
|
crp->crp_payload_output_start, crp->crp_payload_length);
|
|
|
|
|
else
|
2020-06-25 20:01:31 -04:00
|
|
|
error = sglist_append_sglist(s->sg_dsgl, s->sg_input,
|
2020-05-25 18:23:13 -04:00
|
|
|
crp->crp_payload_start, crp->crp_payload_length);
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
if (error)
|
|
|
|
|
return (error);
|
|
|
|
|
if (op_type == CHCR_ENCRYPT_OP) {
|
2020-05-25 18:23:13 -04:00
|
|
|
if (CRYPTO_HAS_OUTPUT_BUFFER(crp))
|
2020-06-25 20:01:31 -04:00
|
|
|
error = sglist_append_sglist(s->sg_dsgl, s->sg_output,
|
2020-05-25 18:23:13 -04:00
|
|
|
crp->crp_digest_start, hash_size_in_response);
|
|
|
|
|
else
|
2020-06-25 20:01:31 -04:00
|
|
|
error = sglist_append_sglist(s->sg_dsgl, s->sg_input,
|
2020-05-25 18:23:13 -04:00
|
|
|
crp->crp_digest_start, hash_size_in_response);
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
if (error)
|
|
|
|
|
return (error);
|
|
|
|
|
}
|
2020-06-25 20:01:31 -04:00
|
|
|
dsgl_nsegs = ccr_count_sgl(s->sg_dsgl, DSGL_SGE_MAXLEN);
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
if (dsgl_nsegs > MAX_RX_PHYS_DSGL_SGE)
|
|
|
|
|
return (EFBIG);
|
|
|
|
|
dsgl_len = ccr_phys_dsgl_len(dsgl_nsegs);
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
|
* The 'key' part of the key context consists of the key followed
|
|
|
|
|
* by the Galois hash key.
|
|
|
|
|
*/
|
|
|
|
|
kctx_len = roundup2(s->blkcipher.key_len, 16) + GMAC_BLOCK_LEN;
|
|
|
|
|
transhdr_len = CIPHER_TRANSHDR_SIZE(kctx_len, dsgl_len);
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
|
* The input buffer consists of the IV, any AAD, and then the
|
|
|
|
|
* cipher/plain text. For decryption requests the hash is
|
|
|
|
|
* appended after the cipher text.
|
2018-01-24 15:15:49 -05:00
|
|
|
*
|
|
|
|
|
* The IV is always stored at the start of the input buffer
|
|
|
|
|
* even though it may be duplicated in the payload. The
|
|
|
|
|
* crypto engine doesn't work properly if the IV offset points
|
|
|
|
|
* inside of the AAD region, so a second copy is always
|
|
|
|
|
* required.
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
*/
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
input_len = crp->crp_aad_length + crp->crp_payload_length;
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
if (op_type == CHCR_DECRYPT_OP)
|
|
|
|
|
input_len += hash_size_in_response;
|
2017-05-26 16:20:40 -04:00
|
|
|
if (input_len > MAX_REQUEST_SIZE)
|
|
|
|
|
return (EFBIG);
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
if (ccr_use_imm_data(transhdr_len, iv_len + input_len)) {
|
|
|
|
|
imm_len = input_len;
|
|
|
|
|
sgl_nsegs = 0;
|
|
|
|
|
sgl_len = 0;
|
|
|
|
|
} else {
|
|
|
|
|
imm_len = 0;
|
2020-06-25 20:01:31 -04:00
|
|
|
sglist_reset(s->sg_ulptx);
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
if (crp->crp_aad_length != 0) {
|
2020-06-22 19:41:33 -04:00
|
|
|
if (crp->crp_aad != NULL)
|
2020-06-25 20:01:31 -04:00
|
|
|
error = sglist_append(s->sg_ulptx,
|
2020-06-22 19:41:33 -04:00
|
|
|
crp->crp_aad, crp->crp_aad_length);
|
|
|
|
|
else
|
2020-06-25 20:01:31 -04:00
|
|
|
error = sglist_append_sglist(s->sg_ulptx,
|
|
|
|
|
s->sg_input, crp->crp_aad_start,
|
2020-06-22 19:41:33 -04:00
|
|
|
crp->crp_aad_length);
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
if (error)
|
|
|
|
|
return (error);
|
|
|
|
|
}
|
2020-06-25 20:01:31 -04:00
|
|
|
error = sglist_append_sglist(s->sg_ulptx, s->sg_input,
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
crp->crp_payload_start, crp->crp_payload_length);
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
if (error)
|
|
|
|
|
return (error);
|
|
|
|
|
if (op_type == CHCR_DECRYPT_OP) {
|
2020-06-25 20:01:31 -04:00
|
|
|
error = sglist_append_sglist(s->sg_ulptx, s->sg_input,
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
crp->crp_digest_start, hash_size_in_response);
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
if (error)
|
|
|
|
|
return (error);
|
|
|
|
|
}
|
2020-06-25 20:01:31 -04:00
|
|
|
sgl_nsegs = s->sg_ulptx->sg_nseg;
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
sgl_len = ccr_ulptx_sgl_len(sgl_nsegs);
|
|
|
|
|
}
|
|
|
|
|
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
if (crp->crp_aad_length != 0) {
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
aad_start = iv_len + 1;
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
aad_stop = aad_start + crp->crp_aad_length - 1;
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
} else {
|
|
|
|
|
aad_start = 0;
|
|
|
|
|
aad_stop = 0;
|
|
|
|
|
}
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
cipher_start = iv_len + crp->crp_aad_length + 1;
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
if (op_type == CHCR_DECRYPT_OP)
|
|
|
|
|
cipher_stop = hash_size_in_response;
|
|
|
|
|
else
|
|
|
|
|
cipher_stop = 0;
|
|
|
|
|
if (op_type == CHCR_DECRYPT_OP)
|
|
|
|
|
auth_insert = hash_size_in_response;
|
|
|
|
|
else
|
|
|
|
|
auth_insert = 0;
|
|
|
|
|
|
2018-01-24 15:04:08 -05:00
|
|
|
wr_len = roundup2(transhdr_len, 16) + iv_len + roundup2(imm_len, 16) +
|
|
|
|
|
sgl_len;
|
2018-01-24 15:12:00 -05:00
|
|
|
if (wr_len > SGE_MAX_WR_LEN)
|
|
|
|
|
return (EFBIG);
|
2020-04-10 18:27:45 -04:00
|
|
|
wr = alloc_wrqe(wr_len, s->port->txq);
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
if (wr == NULL) {
|
2020-06-25 20:01:31 -04:00
|
|
|
counter_u64_add(sc->stats_wr_nomem, 1);
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
return (ENOMEM);
|
|
|
|
|
}
|
|
|
|
|
crwr = wrtod(wr);
|
|
|
|
|
memset(crwr, 0, wr_len);
|
|
|
|
|
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
memcpy(iv, crp->crp_iv, s->blkcipher.iv_len);
|
2018-01-24 15:15:49 -05:00
|
|
|
if (s->blkcipher.iv_len == 12)
|
|
|
|
|
*(uint32_t *)&iv[12] = htobe32(1);
|
|
|
|
|
|
2020-04-10 18:27:45 -04:00
|
|
|
ccr_populate_wreq(sc, s, crwr, kctx_len, wr_len, imm_len, sgl_len, 0,
|
2018-07-17 20:56:25 -04:00
|
|
|
crp);
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
|
|
|
|
|
crwr->sec_cpl.op_ivinsrtofst = htobe32(
|
|
|
|
|
V_CPL_TX_SEC_PDU_OPCODE(CPL_TX_SEC_PDU) |
|
2020-04-10 18:27:45 -04:00
|
|
|
V_CPL_TX_SEC_PDU_RXCHID(s->port->tx_channel_id) |
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
V_CPL_TX_SEC_PDU_ACKFOLLOWS(0) | V_CPL_TX_SEC_PDU_ULPTXLPBK(1) |
|
|
|
|
|
V_CPL_TX_SEC_PDU_CPLLEN(2) | V_CPL_TX_SEC_PDU_PLACEHOLDER(0) |
|
|
|
|
|
V_CPL_TX_SEC_PDU_IVINSRTOFST(1));
|
|
|
|
|
|
|
|
|
|
crwr->sec_cpl.pldlen = htobe32(iv_len + input_len);
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
|
* NB: cipherstop is explicitly set to 0. On encrypt it
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
* should normally be set to 0 anyway. However, for decrypt
|
|
|
|
|
* the cipher ends before the tag in the ETA case (and
|
|
|
|
|
* authstop is set to stop before the tag), but for GCM the
|
|
|
|
|
* cipher still runs to the end of the buffer. Not sure if
|
|
|
|
|
* this is intentional or a firmware quirk, but it is required
|
|
|
|
|
* for working tag validation with GCM decryption.
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
*/
|
|
|
|
|
crwr->sec_cpl.aadstart_cipherstop_hi = htobe32(
|
|
|
|
|
V_CPL_TX_SEC_PDU_AADSTART(aad_start) |
|
|
|
|
|
V_CPL_TX_SEC_PDU_AADSTOP(aad_stop) |
|
|
|
|
|
V_CPL_TX_SEC_PDU_CIPHERSTART(cipher_start) |
|
|
|
|
|
V_CPL_TX_SEC_PDU_CIPHERSTOP_HI(0));
|
|
|
|
|
crwr->sec_cpl.cipherstop_lo_authinsert = htobe32(
|
|
|
|
|
V_CPL_TX_SEC_PDU_CIPHERSTOP_LO(0) |
|
|
|
|
|
V_CPL_TX_SEC_PDU_AUTHSTART(cipher_start) |
|
|
|
|
|
V_CPL_TX_SEC_PDU_AUTHSTOP(cipher_stop) |
|
|
|
|
|
V_CPL_TX_SEC_PDU_AUTHINSERT(auth_insert));
|
|
|
|
|
|
|
|
|
|
/* These two flits are actually a CPL_TLS_TX_SCMD_FMT. */
|
|
|
|
|
hmac_ctrl = ccr_hmac_ctrl(AES_GMAC_HASH_LEN, hash_size_in_response);
|
|
|
|
|
crwr->sec_cpl.seqno_numivs = htobe32(
|
|
|
|
|
V_SCMD_SEQ_NO_CTRL(0) |
|
2018-11-16 14:08:52 -05:00
|
|
|
V_SCMD_PROTO_VERSION(SCMD_PROTO_VERSION_GENERIC) |
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
V_SCMD_ENC_DEC_CTRL(op_type) |
|
|
|
|
|
V_SCMD_CIPH_AUTH_SEQ_CTRL(op_type == CHCR_ENCRYPT_OP ? 1 : 0) |
|
2018-11-16 14:08:52 -05:00
|
|
|
V_SCMD_CIPH_MODE(SCMD_CIPH_MODE_AES_GCM) |
|
|
|
|
|
V_SCMD_AUTH_MODE(SCMD_AUTH_MODE_GHASH) |
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
V_SCMD_HMAC_CTRL(hmac_ctrl) |
|
|
|
|
|
V_SCMD_IV_SIZE(iv_len / 2) |
|
|
|
|
|
V_SCMD_NUM_IVS(0));
|
|
|
|
|
crwr->sec_cpl.ivgen_hdrlen = htobe32(
|
|
|
|
|
V_SCMD_IV_GEN_CTRL(0) |
|
|
|
|
|
V_SCMD_MORE_FRAGS(0) | V_SCMD_LAST_FRAG(0) | V_SCMD_MAC_ONLY(0) |
|
2018-01-24 15:11:00 -05:00
|
|
|
V_SCMD_AADIVDROP(0) | V_SCMD_HDR_LEN(dsgl_len));
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
|
|
|
|
|
crwr->key_ctx.ctx_hdr = s->blkcipher.key_ctx_hdr;
|
|
|
|
|
memcpy(crwr->key_ctx.key, s->blkcipher.enckey, s->blkcipher.key_len);
|
|
|
|
|
dst = crwr->key_ctx.key + roundup2(s->blkcipher.key_len, 16);
|
|
|
|
|
memcpy(dst, s->gmac.ghash_h, GMAC_BLOCK_LEN);
|
|
|
|
|
|
|
|
|
|
dst = (char *)(crwr + 1) + kctx_len;
|
2020-06-25 20:01:31 -04:00
|
|
|
ccr_write_phys_dsgl(s, dst, dsgl_nsegs);
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
dst += sizeof(struct cpl_rx_phys_dsgl) + dsgl_len;
|
2018-01-24 15:04:08 -05:00
|
|
|
memcpy(dst, iv, iv_len);
|
|
|
|
|
dst += iv_len;
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
if (imm_len != 0) {
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
if (crp->crp_aad_length != 0) {
|
2020-06-22 19:41:33 -04:00
|
|
|
if (crp->crp_aad != NULL)
|
|
|
|
|
memcpy(dst, crp->crp_aad, crp->crp_aad_length);
|
|
|
|
|
else
|
|
|
|
|
crypto_copydata(crp, crp->crp_aad_start,
|
|
|
|
|
crp->crp_aad_length, dst);
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
dst += crp->crp_aad_length;
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
}
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
crypto_copydata(crp, crp->crp_payload_start,
|
|
|
|
|
crp->crp_payload_length, dst);
|
|
|
|
|
dst += crp->crp_payload_length;
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
if (op_type == CHCR_DECRYPT_OP)
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
crypto_copydata(crp, crp->crp_digest_start,
|
|
|
|
|
hash_size_in_response, dst);
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
} else
|
2020-06-25 20:01:31 -04:00
|
|
|
ccr_write_ulptx_sgl(s, dst, sgl_nsegs);
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
|
|
|
|
|
/* XXX: TODO backpressure */
|
|
|
|
|
t4_wrq_tx(sc->adapter, wr);
|
|
|
|
|
|
2020-06-03 18:19:52 -04:00
|
|
|
explicit_bzero(iv, sizeof(iv));
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
return (0);
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static int
|
|
|
|
|
ccr_gcm_done(struct ccr_softc *sc, struct ccr_session *s,
|
|
|
|
|
struct cryptop *crp, const struct cpl_fw6_pld *cpl, int error)
|
|
|
|
|
{
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
|
* The updated IV to permit chained requests is at
|
|
|
|
|
* cpl->data[2], but OCF doesn't permit chained requests.
|
|
|
|
|
*
|
|
|
|
|
* Note that the hardware should always verify the GMAC hash.
|
|
|
|
|
*/
|
|
|
|
|
return (error);
|
|
|
|
|
}
|
|
|
|
|
|
2017-06-08 17:06:18 -04:00
|
|
|
/*
|
2018-01-24 15:14:57 -05:00
|
|
|
* Handle a GCM request that is not supported by the crypto engine by
|
|
|
|
|
* performing the operation in software. Derived from swcr_authenc().
|
2017-06-08 17:06:18 -04:00
|
|
|
*/
|
|
|
|
|
static void
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
ccr_gcm_soft(struct ccr_session *s, struct cryptop *crp)
|
2017-06-08 17:06:18 -04:00
|
|
|
{
|
2018-01-24 15:14:57 -05:00
|
|
|
struct auth_hash *axf;
|
|
|
|
|
struct enc_xform *exf;
|
2020-05-20 17:21:01 -04:00
|
|
|
void *auth_ctx, *kschedule;
|
2017-06-08 17:06:18 -04:00
|
|
|
char block[GMAC_BLOCK_LEN];
|
|
|
|
|
char digest[GMAC_DIGEST_LEN];
|
|
|
|
|
char iv[AES_BLOCK_LEN];
|
2018-01-24 15:14:57 -05:00
|
|
|
int error, i, len;
|
|
|
|
|
|
|
|
|
|
auth_ctx = NULL;
|
|
|
|
|
kschedule = NULL;
|
|
|
|
|
|
|
|
|
|
/* Initialize the MAC. */
|
|
|
|
|
switch (s->blkcipher.key_len) {
|
|
|
|
|
case 16:
|
|
|
|
|
axf = &auth_hash_nist_gmac_aes_128;
|
|
|
|
|
break;
|
|
|
|
|
case 24:
|
|
|
|
|
axf = &auth_hash_nist_gmac_aes_192;
|
|
|
|
|
break;
|
|
|
|
|
case 32:
|
|
|
|
|
axf = &auth_hash_nist_gmac_aes_256;
|
|
|
|
|
break;
|
|
|
|
|
default:
|
|
|
|
|
error = EINVAL;
|
|
|
|
|
goto out;
|
|
|
|
|
}
|
|
|
|
|
auth_ctx = malloc(axf->ctxsize, M_CCR, M_NOWAIT);
|
|
|
|
|
if (auth_ctx == NULL) {
|
|
|
|
|
error = ENOMEM;
|
|
|
|
|
goto out;
|
|
|
|
|
}
|
|
|
|
|
axf->Init(auth_ctx);
|
|
|
|
|
axf->Setkey(auth_ctx, s->blkcipher.enckey, s->blkcipher.key_len);
|
|
|
|
|
|
|
|
|
|
/* Initialize the cipher. */
|
|
|
|
|
exf = &enc_xform_aes_nist_gcm;
|
2020-05-20 17:21:01 -04:00
|
|
|
kschedule = malloc(exf->ctxsize, M_CCR, M_NOWAIT);
|
|
|
|
|
if (kschedule == NULL) {
|
|
|
|
|
error = ENOMEM;
|
|
|
|
|
goto out;
|
|
|
|
|
}
|
|
|
|
|
error = exf->setkey(kschedule, s->blkcipher.enckey,
|
2018-01-24 15:14:57 -05:00
|
|
|
s->blkcipher.key_len);
|
|
|
|
|
if (error)
|
|
|
|
|
goto out;
|
2017-06-08 17:06:18 -04:00
|
|
|
|
|
|
|
|
/*
|
|
|
|
|
* This assumes a 12-byte IV from the crp. See longer comment
|
|
|
|
|
* above in ccr_gcm() for more details.
|
|
|
|
|
*/
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
if ((crp->crp_flags & CRYPTO_F_IV_SEPARATE) == 0) {
|
|
|
|
|
error = EINVAL;
|
|
|
|
|
goto out;
|
2017-06-08 17:06:18 -04:00
|
|
|
}
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
memcpy(iv, crp->crp_iv, 12);
|
2017-06-08 17:06:18 -04:00
|
|
|
*(uint32_t *)&iv[12] = htobe32(1);
|
|
|
|
|
|
2018-01-24 15:14:57 -05:00
|
|
|
axf->Reinit(auth_ctx, iv, sizeof(iv));
|
2017-06-08 17:06:18 -04:00
|
|
|
|
|
|
|
|
/* MAC the AAD. */
|
2020-06-22 19:41:33 -04:00
|
|
|
if (crp->crp_aad != NULL) {
|
|
|
|
|
len = rounddown(crp->crp_aad_length, sizeof(block));
|
|
|
|
|
if (len != 0)
|
|
|
|
|
axf->Update(auth_ctx, crp->crp_aad, len);
|
|
|
|
|
if (crp->crp_aad_length != len) {
|
|
|
|
|
memset(block, 0, sizeof(block));
|
|
|
|
|
memcpy(block, (char *)crp->crp_aad + len,
|
|
|
|
|
crp->crp_aad_length - len);
|
|
|
|
|
axf->Update(auth_ctx, block, sizeof(block));
|
|
|
|
|
}
|
|
|
|
|
} else {
|
|
|
|
|
for (i = 0; i < crp->crp_aad_length; i += sizeof(block)) {
|
|
|
|
|
len = imin(crp->crp_aad_length - i, sizeof(block));
|
|
|
|
|
crypto_copydata(crp, crp->crp_aad_start + i, len,
|
|
|
|
|
block);
|
|
|
|
|
bzero(block + len, sizeof(block) - len);
|
|
|
|
|
axf->Update(auth_ctx, block, sizeof(block));
|
|
|
|
|
}
|
2018-01-24 15:14:57 -05:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
exf->reinit(kschedule, iv);
|
|
|
|
|
|
|
|
|
|
/* Do encryption with MAC */
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
for (i = 0; i < crp->crp_payload_length; i += sizeof(block)) {
|
|
|
|
|
len = imin(crp->crp_payload_length - i, sizeof(block));
|
|
|
|
|
crypto_copydata(crp, crp->crp_payload_start + i, len, block);
|
2018-01-24 15:14:57 -05:00
|
|
|
bzero(block + len, sizeof(block) - len);
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
if (CRYPTO_OP_IS_ENCRYPT(crp->crp_op)) {
|
2020-05-20 17:21:01 -04:00
|
|
|
exf->encrypt(kschedule, block, block);
|
2018-01-24 15:14:57 -05:00
|
|
|
axf->Update(auth_ctx, block, len);
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
crypto_copyback(crp, crp->crp_payload_start + i, len,
|
|
|
|
|
block);
|
2018-01-24 15:14:57 -05:00
|
|
|
} else {
|
|
|
|
|
axf->Update(auth_ctx, block, len);
|
|
|
|
|
}
|
2017-06-08 17:06:18 -04:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
/* Length block. */
|
|
|
|
|
bzero(block, sizeof(block));
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
((uint32_t *)block)[1] = htobe32(crp->crp_aad_length * 8);
|
|
|
|
|
((uint32_t *)block)[3] = htobe32(crp->crp_payload_length * 8);
|
2018-01-24 15:14:57 -05:00
|
|
|
axf->Update(auth_ctx, block, sizeof(block));
|
|
|
|
|
|
|
|
|
|
/* Finalize MAC. */
|
|
|
|
|
axf->Final(digest, auth_ctx);
|
2017-06-08 17:06:18 -04:00
|
|
|
|
2018-01-24 15:14:57 -05:00
|
|
|
/* Inject or validate tag. */
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
if (CRYPTO_OP_IS_ENCRYPT(crp->crp_op)) {
|
|
|
|
|
crypto_copyback(crp, crp->crp_digest_start, sizeof(digest),
|
|
|
|
|
digest);
|
2018-01-24 15:14:57 -05:00
|
|
|
error = 0;
|
2017-06-08 17:33:10 -04:00
|
|
|
} else {
|
|
|
|
|
char digest2[GMAC_DIGEST_LEN];
|
|
|
|
|
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
crypto_copydata(crp, crp->crp_digest_start, sizeof(digest2),
|
|
|
|
|
digest2);
|
2018-01-24 15:14:57 -05:00
|
|
|
if (timingsafe_bcmp(digest, digest2, sizeof(digest)) == 0) {
|
|
|
|
|
error = 0;
|
|
|
|
|
|
|
|
|
|
/* Tag matches, decrypt data. */
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
for (i = 0; i < crp->crp_payload_length;
|
|
|
|
|
i += sizeof(block)) {
|
|
|
|
|
len = imin(crp->crp_payload_length - i,
|
|
|
|
|
sizeof(block));
|
|
|
|
|
crypto_copydata(crp, crp->crp_payload_start + i,
|
|
|
|
|
len, block);
|
2018-01-24 15:14:57 -05:00
|
|
|
bzero(block + len, sizeof(block) - len);
|
2020-05-20 17:21:01 -04:00
|
|
|
exf->decrypt(kschedule, block, block);
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
crypto_copyback(crp, crp->crp_payload_start + i,
|
|
|
|
|
len, block);
|
2018-01-24 15:14:57 -05:00
|
|
|
}
|
|
|
|
|
} else
|
|
|
|
|
error = EBADMSG;
|
2020-06-03 18:11:05 -04:00
|
|
|
explicit_bzero(digest2, sizeof(digest2));
|
2018-01-24 15:14:57 -05:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
out:
|
2020-05-20 17:21:01 -04:00
|
|
|
zfree(kschedule, M_CCR);
|
|
|
|
|
zfree(auth_ctx, M_CCR);
|
2020-06-03 18:11:05 -04:00
|
|
|
explicit_bzero(block, sizeof(block));
|
|
|
|
|
explicit_bzero(iv, sizeof(iv));
|
|
|
|
|
explicit_bzero(digest, sizeof(digest));
|
2018-01-24 15:14:57 -05:00
|
|
|
crp->crp_etype = error;
|
2017-06-08 17:06:18 -04:00
|
|
|
crypto_done(crp);
|
|
|
|
|
}
|
|
|
|
|
|
2019-04-24 19:31:46 -04:00
|
|
|
static void
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
generate_ccm_b0(struct cryptop *crp, u_int hash_size_in_response,
|
|
|
|
|
const char *iv, char *b0)
|
2019-04-24 19:31:46 -04:00
|
|
|
{
|
|
|
|
|
u_int i, payload_len;
|
|
|
|
|
|
|
|
|
|
/* NB: L is already set in the first byte of the IV. */
|
|
|
|
|
memcpy(b0, iv, CCM_B0_SIZE);
|
|
|
|
|
|
|
|
|
|
/* Set length of hash in bits 3 - 5. */
|
|
|
|
|
b0[0] |= (((hash_size_in_response - 2) / 2) << 3);
|
|
|
|
|
|
|
|
|
|
/* Store the payload length as a big-endian value. */
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
payload_len = crp->crp_payload_length;
|
2019-04-24 19:31:46 -04:00
|
|
|
for (i = 0; i < iv[0]; i++) {
|
|
|
|
|
b0[CCM_CBC_BLOCK_LEN - 1 - i] = payload_len;
|
|
|
|
|
payload_len >>= 8;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
|
* If there is AAD in the request, set bit 6 in the flags
|
|
|
|
|
* field and store the AAD length as a big-endian value at the
|
|
|
|
|
* start of block 1. This only assumes a 16-bit AAD length
|
|
|
|
|
* since T6 doesn't support large AAD sizes.
|
|
|
|
|
*/
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
if (crp->crp_aad_length != 0) {
|
2019-04-24 19:31:46 -04:00
|
|
|
b0[0] |= (1 << 6);
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
*(uint16_t *)(b0 + CCM_B0_SIZE) = htobe16(crp->crp_aad_length);
|
2019-04-24 19:31:46 -04:00
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static int
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
ccr_ccm(struct ccr_softc *sc, struct ccr_session *s, struct cryptop *crp)
|
2019-04-24 19:31:46 -04:00
|
|
|
{
|
|
|
|
|
char iv[CHCR_MAX_CRYPTO_IV_LEN];
|
|
|
|
|
struct ulptx_idata *idata;
|
|
|
|
|
struct chcr_wr *crwr;
|
|
|
|
|
struct wrqe *wr;
|
|
|
|
|
char *dst;
|
|
|
|
|
u_int iv_len, kctx_len, op_type, transhdr_len, wr_len;
|
|
|
|
|
u_int aad_len, b0_len, hash_size_in_response, imm_len;
|
|
|
|
|
u_int aad_start, aad_stop, cipher_start, cipher_stop, auth_insert;
|
|
|
|
|
u_int hmac_ctrl, input_len;
|
|
|
|
|
int dsgl_nsegs, dsgl_len;
|
|
|
|
|
int sgl_nsegs, sgl_len;
|
|
|
|
|
int error;
|
|
|
|
|
|
|
|
|
|
if (s->blkcipher.key_len == 0)
|
|
|
|
|
return (EINVAL);
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
|
* The crypto engine doesn't handle CCM requests with an empty
|
|
|
|
|
* payload, so handle those in software instead.
|
|
|
|
|
*/
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
if (crp->crp_payload_length == 0)
|
2019-04-24 19:31:46 -04:00
|
|
|
return (EMSGSIZE);
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
|
* CCM always includes block 0 in the AAD before AAD from the
|
|
|
|
|
* request.
|
|
|
|
|
*/
|
|
|
|
|
b0_len = CCM_B0_SIZE;
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
if (crp->crp_aad_length != 0)
|
2019-04-24 19:31:46 -04:00
|
|
|
b0_len += CCM_AAD_FIELD_SIZE;
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
aad_len = b0_len + crp->crp_aad_length;
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
|
* CCM requests should always provide an explicit IV (really
|
|
|
|
|
* the nonce).
|
|
|
|
|
*/
|
|
|
|
|
if ((crp->crp_flags & CRYPTO_F_IV_SEPARATE) == 0)
|
|
|
|
|
return (EINVAL);
|
2019-04-24 19:31:46 -04:00
|
|
|
|
|
|
|
|
/*
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
* Always assume a 12 byte input nonce for now since that is
|
|
|
|
|
* what OCF always generates. The full IV in the work request
|
|
|
|
|
* is 16 bytes.
|
2019-04-24 19:31:46 -04:00
|
|
|
*/
|
|
|
|
|
iv_len = AES_BLOCK_LEN;
|
|
|
|
|
|
|
|
|
|
if (iv_len + aad_len > MAX_AAD_LEN)
|
|
|
|
|
return (EMSGSIZE);
|
|
|
|
|
|
|
|
|
|
hash_size_in_response = s->ccm_mac.hash_len;
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
if (CRYPTO_OP_IS_ENCRYPT(crp->crp_op))
|
2019-04-24 19:31:46 -04:00
|
|
|
op_type = CHCR_ENCRYPT_OP;
|
|
|
|
|
else
|
|
|
|
|
op_type = CHCR_DECRYPT_OP;
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
|
* The output buffer consists of the cipher text followed by
|
|
|
|
|
* the tag when encrypting. For decryption it only contains
|
|
|
|
|
* the plain text.
|
|
|
|
|
*
|
|
|
|
|
* Due to a firmware bug, the output buffer must include a
|
|
|
|
|
* dummy output buffer for the IV and AAD prior to the real
|
|
|
|
|
* output buffer.
|
|
|
|
|
*/
|
|
|
|
|
if (op_type == CHCR_ENCRYPT_OP) {
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
if (iv_len + aad_len + crp->crp_payload_length +
|
|
|
|
|
hash_size_in_response > MAX_REQUEST_SIZE)
|
2019-04-24 19:31:46 -04:00
|
|
|
return (EFBIG);
|
|
|
|
|
} else {
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
if (iv_len + aad_len + crp->crp_payload_length >
|
|
|
|
|
MAX_REQUEST_SIZE)
|
2019-04-24 19:31:46 -04:00
|
|
|
return (EFBIG);
|
|
|
|
|
}
|
2020-06-25 20:01:31 -04:00
|
|
|
sglist_reset(s->sg_dsgl);
|
|
|
|
|
error = sglist_append_sglist(s->sg_dsgl, sc->sg_iv_aad, 0, iv_len +
|
2019-04-24 19:31:46 -04:00
|
|
|
aad_len);
|
|
|
|
|
if (error)
|
|
|
|
|
return (error);
|
2020-05-25 18:23:13 -04:00
|
|
|
if (CRYPTO_HAS_OUTPUT_BUFFER(crp))
|
2020-06-25 20:01:31 -04:00
|
|
|
error = sglist_append_sglist(s->sg_dsgl, s->sg_output,
|
2020-05-25 18:23:13 -04:00
|
|
|
crp->crp_payload_output_start, crp->crp_payload_length);
|
|
|
|
|
else
|
2020-06-25 20:01:31 -04:00
|
|
|
error = sglist_append_sglist(s->sg_dsgl, s->sg_input,
|
2020-05-25 18:23:13 -04:00
|
|
|
crp->crp_payload_start, crp->crp_payload_length);
|
2019-04-24 19:31:46 -04:00
|
|
|
if (error)
|
|
|
|
|
return (error);
|
|
|
|
|
if (op_type == CHCR_ENCRYPT_OP) {
|
2020-05-25 18:23:13 -04:00
|
|
|
if (CRYPTO_HAS_OUTPUT_BUFFER(crp))
|
2020-06-25 20:01:31 -04:00
|
|
|
error = sglist_append_sglist(s->sg_dsgl, s->sg_output,
|
2020-05-25 18:23:13 -04:00
|
|
|
crp->crp_digest_start, hash_size_in_response);
|
|
|
|
|
else
|
2020-06-25 20:01:31 -04:00
|
|
|
error = sglist_append_sglist(s->sg_dsgl, s->sg_input,
|
2020-05-25 18:23:13 -04:00
|
|
|
crp->crp_digest_start, hash_size_in_response);
|
2019-04-24 19:31:46 -04:00
|
|
|
if (error)
|
|
|
|
|
return (error);
|
|
|
|
|
}
|
2020-06-25 20:01:31 -04:00
|
|
|
dsgl_nsegs = ccr_count_sgl(s->sg_dsgl, DSGL_SGE_MAXLEN);
|
2019-04-24 19:31:46 -04:00
|
|
|
if (dsgl_nsegs > MAX_RX_PHYS_DSGL_SGE)
|
|
|
|
|
return (EFBIG);
|
|
|
|
|
dsgl_len = ccr_phys_dsgl_len(dsgl_nsegs);
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
|
* The 'key' part of the key context consists of two copies of
|
|
|
|
|
* the AES key.
|
|
|
|
|
*/
|
|
|
|
|
kctx_len = roundup2(s->blkcipher.key_len, 16) * 2;
|
|
|
|
|
transhdr_len = CIPHER_TRANSHDR_SIZE(kctx_len, dsgl_len);
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
|
* The input buffer consists of the IV, AAD (including block
|
|
|
|
|
* 0), and then the cipher/plain text. For decryption
|
|
|
|
|
* requests the hash is appended after the cipher text.
|
|
|
|
|
*
|
|
|
|
|
* The IV is always stored at the start of the input buffer
|
|
|
|
|
* even though it may be duplicated in the payload. The
|
|
|
|
|
* crypto engine doesn't work properly if the IV offset points
|
|
|
|
|
* inside of the AAD region, so a second copy is always
|
|
|
|
|
* required.
|
|
|
|
|
*/
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
input_len = aad_len + crp->crp_payload_length;
|
2019-04-24 19:31:46 -04:00
|
|
|
if (op_type == CHCR_DECRYPT_OP)
|
|
|
|
|
input_len += hash_size_in_response;
|
|
|
|
|
if (input_len > MAX_REQUEST_SIZE)
|
|
|
|
|
return (EFBIG);
|
|
|
|
|
if (ccr_use_imm_data(transhdr_len, iv_len + input_len)) {
|
|
|
|
|
imm_len = input_len;
|
|
|
|
|
sgl_nsegs = 0;
|
|
|
|
|
sgl_len = 0;
|
|
|
|
|
} else {
|
|
|
|
|
/* Block 0 is passed as immediate data. */
|
|
|
|
|
imm_len = b0_len;
|
|
|
|
|
|
2020-06-25 20:01:31 -04:00
|
|
|
sglist_reset(s->sg_ulptx);
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
if (crp->crp_aad_length != 0) {
|
2020-06-22 19:41:33 -04:00
|
|
|
if (crp->crp_aad != NULL)
|
2020-06-25 20:01:31 -04:00
|
|
|
error = sglist_append(s->sg_ulptx,
|
2020-06-22 19:41:33 -04:00
|
|
|
crp->crp_aad, crp->crp_aad_length);
|
|
|
|
|
else
|
2020-06-25 20:01:31 -04:00
|
|
|
error = sglist_append_sglist(s->sg_ulptx,
|
|
|
|
|
s->sg_input, crp->crp_aad_start,
|
2020-06-22 19:41:33 -04:00
|
|
|
crp->crp_aad_length);
|
2019-04-24 19:31:46 -04:00
|
|
|
if (error)
|
|
|
|
|
return (error);
|
|
|
|
|
}
|
2020-06-25 20:01:31 -04:00
|
|
|
error = sglist_append_sglist(s->sg_ulptx, s->sg_input,
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
crp->crp_payload_start, crp->crp_payload_length);
|
2019-04-24 19:31:46 -04:00
|
|
|
if (error)
|
|
|
|
|
return (error);
|
|
|
|
|
if (op_type == CHCR_DECRYPT_OP) {
|
2020-06-25 20:01:31 -04:00
|
|
|
error = sglist_append_sglist(s->sg_ulptx, s->sg_input,
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
crp->crp_digest_start, hash_size_in_response);
|
2019-04-24 19:31:46 -04:00
|
|
|
if (error)
|
|
|
|
|
return (error);
|
|
|
|
|
}
|
2020-06-25 20:01:31 -04:00
|
|
|
sgl_nsegs = s->sg_ulptx->sg_nseg;
|
2019-04-24 19:31:46 -04:00
|
|
|
sgl_len = ccr_ulptx_sgl_len(sgl_nsegs);
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
aad_start = iv_len + 1;
|
|
|
|
|
aad_stop = aad_start + aad_len - 1;
|
|
|
|
|
cipher_start = aad_stop + 1;
|
|
|
|
|
if (op_type == CHCR_DECRYPT_OP)
|
|
|
|
|
cipher_stop = hash_size_in_response;
|
|
|
|
|
else
|
|
|
|
|
cipher_stop = 0;
|
|
|
|
|
if (op_type == CHCR_DECRYPT_OP)
|
|
|
|
|
auth_insert = hash_size_in_response;
|
|
|
|
|
else
|
|
|
|
|
auth_insert = 0;
|
|
|
|
|
|
|
|
|
|
wr_len = roundup2(transhdr_len, 16) + iv_len + roundup2(imm_len, 16) +
|
|
|
|
|
sgl_len;
|
|
|
|
|
if (wr_len > SGE_MAX_WR_LEN)
|
|
|
|
|
return (EFBIG);
|
2020-04-10 18:27:45 -04:00
|
|
|
wr = alloc_wrqe(wr_len, s->port->txq);
|
2019-04-24 19:31:46 -04:00
|
|
|
if (wr == NULL) {
|
2020-06-25 20:01:31 -04:00
|
|
|
counter_u64_add(sc->stats_wr_nomem, 1);
|
2019-04-24 19:31:46 -04:00
|
|
|
return (ENOMEM);
|
|
|
|
|
}
|
|
|
|
|
crwr = wrtod(wr);
|
|
|
|
|
memset(crwr, 0, wr_len);
|
|
|
|
|
|
|
|
|
|
/*
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
* Read the nonce from the request. Use the nonce to generate
|
|
|
|
|
* the full IV with the counter set to 0.
|
2019-04-24 19:31:46 -04:00
|
|
|
*/
|
|
|
|
|
memset(iv, 0, iv_len);
|
|
|
|
|
iv[0] = (15 - AES_CCM_IV_LEN) - 1;
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
memcpy(iv + 1, crp->crp_iv, AES_CCM_IV_LEN);
|
2019-04-24 19:31:46 -04:00
|
|
|
|
2020-04-10 18:27:45 -04:00
|
|
|
ccr_populate_wreq(sc, s, crwr, kctx_len, wr_len, imm_len, sgl_len, 0,
|
2019-04-24 19:31:46 -04:00
|
|
|
crp);
|
|
|
|
|
|
|
|
|
|
crwr->sec_cpl.op_ivinsrtofst = htobe32(
|
|
|
|
|
V_CPL_TX_SEC_PDU_OPCODE(CPL_TX_SEC_PDU) |
|
2020-04-10 18:27:45 -04:00
|
|
|
V_CPL_TX_SEC_PDU_RXCHID(s->port->tx_channel_id) |
|
2019-04-24 19:31:46 -04:00
|
|
|
V_CPL_TX_SEC_PDU_ACKFOLLOWS(0) | V_CPL_TX_SEC_PDU_ULPTXLPBK(1) |
|
|
|
|
|
V_CPL_TX_SEC_PDU_CPLLEN(2) | V_CPL_TX_SEC_PDU_PLACEHOLDER(0) |
|
|
|
|
|
V_CPL_TX_SEC_PDU_IVINSRTOFST(1));
|
|
|
|
|
|
|
|
|
|
crwr->sec_cpl.pldlen = htobe32(iv_len + input_len);
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
|
* NB: cipherstop is explicitly set to 0. See comments above
|
|
|
|
|
* in ccr_gcm().
|
|
|
|
|
*/
|
|
|
|
|
crwr->sec_cpl.aadstart_cipherstop_hi = htobe32(
|
|
|
|
|
V_CPL_TX_SEC_PDU_AADSTART(aad_start) |
|
|
|
|
|
V_CPL_TX_SEC_PDU_AADSTOP(aad_stop) |
|
|
|
|
|
V_CPL_TX_SEC_PDU_CIPHERSTART(cipher_start) |
|
|
|
|
|
V_CPL_TX_SEC_PDU_CIPHERSTOP_HI(0));
|
|
|
|
|
crwr->sec_cpl.cipherstop_lo_authinsert = htobe32(
|
|
|
|
|
V_CPL_TX_SEC_PDU_CIPHERSTOP_LO(0) |
|
|
|
|
|
V_CPL_TX_SEC_PDU_AUTHSTART(cipher_start) |
|
|
|
|
|
V_CPL_TX_SEC_PDU_AUTHSTOP(cipher_stop) |
|
|
|
|
|
V_CPL_TX_SEC_PDU_AUTHINSERT(auth_insert));
|
|
|
|
|
|
|
|
|
|
/* These two flits are actually a CPL_TLS_TX_SCMD_FMT. */
|
|
|
|
|
hmac_ctrl = ccr_hmac_ctrl(AES_CBC_MAC_HASH_LEN, hash_size_in_response);
|
|
|
|
|
crwr->sec_cpl.seqno_numivs = htobe32(
|
|
|
|
|
V_SCMD_SEQ_NO_CTRL(0) |
|
|
|
|
|
V_SCMD_PROTO_VERSION(SCMD_PROTO_VERSION_GENERIC) |
|
|
|
|
|
V_SCMD_ENC_DEC_CTRL(op_type) |
|
|
|
|
|
V_SCMD_CIPH_AUTH_SEQ_CTRL(op_type == CHCR_ENCRYPT_OP ? 0 : 1) |
|
|
|
|
|
V_SCMD_CIPH_MODE(SCMD_CIPH_MODE_AES_CCM) |
|
|
|
|
|
V_SCMD_AUTH_MODE(SCMD_AUTH_MODE_CBCMAC) |
|
|
|
|
|
V_SCMD_HMAC_CTRL(hmac_ctrl) |
|
|
|
|
|
V_SCMD_IV_SIZE(iv_len / 2) |
|
|
|
|
|
V_SCMD_NUM_IVS(0));
|
|
|
|
|
crwr->sec_cpl.ivgen_hdrlen = htobe32(
|
|
|
|
|
V_SCMD_IV_GEN_CTRL(0) |
|
|
|
|
|
V_SCMD_MORE_FRAGS(0) | V_SCMD_LAST_FRAG(0) | V_SCMD_MAC_ONLY(0) |
|
|
|
|
|
V_SCMD_AADIVDROP(0) | V_SCMD_HDR_LEN(dsgl_len));
|
|
|
|
|
|
|
|
|
|
crwr->key_ctx.ctx_hdr = s->blkcipher.key_ctx_hdr;
|
|
|
|
|
memcpy(crwr->key_ctx.key, s->blkcipher.enckey, s->blkcipher.key_len);
|
|
|
|
|
memcpy(crwr->key_ctx.key + roundup(s->blkcipher.key_len, 16),
|
|
|
|
|
s->blkcipher.enckey, s->blkcipher.key_len);
|
|
|
|
|
|
|
|
|
|
dst = (char *)(crwr + 1) + kctx_len;
|
2020-06-25 20:01:31 -04:00
|
|
|
ccr_write_phys_dsgl(s, dst, dsgl_nsegs);
|
2019-04-24 19:31:46 -04:00
|
|
|
dst += sizeof(struct cpl_rx_phys_dsgl) + dsgl_len;
|
|
|
|
|
memcpy(dst, iv, iv_len);
|
|
|
|
|
dst += iv_len;
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
generate_ccm_b0(crp, hash_size_in_response, iv, dst);
|
2019-04-24 19:31:46 -04:00
|
|
|
if (sgl_nsegs == 0) {
|
|
|
|
|
dst += b0_len;
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
if (crp->crp_aad_length != 0) {
|
2020-06-22 19:41:33 -04:00
|
|
|
if (crp->crp_aad != NULL)
|
|
|
|
|
memcpy(dst, crp->crp_aad, crp->crp_aad_length);
|
|
|
|
|
else
|
|
|
|
|
crypto_copydata(crp, crp->crp_aad_start,
|
|
|
|
|
crp->crp_aad_length, dst);
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
dst += crp->crp_aad_length;
|
2019-04-24 19:31:46 -04:00
|
|
|
}
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
crypto_copydata(crp, crp->crp_payload_start,
|
|
|
|
|
crp->crp_payload_length, dst);
|
|
|
|
|
dst += crp->crp_payload_length;
|
2019-04-24 19:31:46 -04:00
|
|
|
if (op_type == CHCR_DECRYPT_OP)
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
crypto_copydata(crp, crp->crp_digest_start,
|
|
|
|
|
hash_size_in_response, dst);
|
2019-04-24 19:31:46 -04:00
|
|
|
} else {
|
|
|
|
|
dst += CCM_B0_SIZE;
|
|
|
|
|
if (b0_len > CCM_B0_SIZE) {
|
|
|
|
|
/*
|
|
|
|
|
* If there is AAD, insert padding including a
|
|
|
|
|
* ULP_TX_SC_NOOP so that the ULP_TX_SC_DSGL
|
|
|
|
|
* is 16-byte aligned.
|
|
|
|
|
*/
|
|
|
|
|
KASSERT(b0_len - CCM_B0_SIZE == CCM_AAD_FIELD_SIZE,
|
|
|
|
|
("b0_len mismatch"));
|
|
|
|
|
memset(dst + CCM_AAD_FIELD_SIZE, 0,
|
|
|
|
|
8 - CCM_AAD_FIELD_SIZE);
|
|
|
|
|
idata = (void *)(dst + 8);
|
|
|
|
|
idata->cmd_more = htobe32(V_ULPTX_CMD(ULP_TX_SC_NOOP));
|
|
|
|
|
idata->len = htobe32(0);
|
|
|
|
|
dst = (void *)(idata + 1);
|
|
|
|
|
}
|
2020-06-25 20:01:31 -04:00
|
|
|
ccr_write_ulptx_sgl(s, dst, sgl_nsegs);
|
2019-04-24 19:31:46 -04:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
/* XXX: TODO backpressure */
|
|
|
|
|
t4_wrq_tx(sc->adapter, wr);
|
|
|
|
|
|
2020-06-03 18:19:52 -04:00
|
|
|
explicit_bzero(iv, sizeof(iv));
|
2019-04-24 19:31:46 -04:00
|
|
|
return (0);
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static int
|
|
|
|
|
ccr_ccm_done(struct ccr_softc *sc, struct ccr_session *s,
|
|
|
|
|
struct cryptop *crp, const struct cpl_fw6_pld *cpl, int error)
|
|
|
|
|
{
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
|
* The updated IV to permit chained requests is at
|
|
|
|
|
* cpl->data[2], but OCF doesn't permit chained requests.
|
|
|
|
|
*
|
|
|
|
|
* Note that the hardware should always verify the CBC MAC
|
|
|
|
|
* hash.
|
|
|
|
|
*/
|
|
|
|
|
return (error);
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
|
* Handle a CCM request that is not supported by the crypto engine by
|
|
|
|
|
* performing the operation in software. Derived from swcr_authenc().
|
|
|
|
|
*/
|
|
|
|
|
static void
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
ccr_ccm_soft(struct ccr_session *s, struct cryptop *crp)
|
2019-04-24 19:31:46 -04:00
|
|
|
{
|
|
|
|
|
struct auth_hash *axf;
|
|
|
|
|
struct enc_xform *exf;
|
|
|
|
|
union authctx *auth_ctx;
|
2020-05-20 17:21:01 -04:00
|
|
|
void *kschedule;
|
2019-04-24 19:31:46 -04:00
|
|
|
char block[CCM_CBC_BLOCK_LEN];
|
|
|
|
|
char digest[AES_CBC_MAC_HASH_LEN];
|
|
|
|
|
char iv[AES_CCM_IV_LEN];
|
|
|
|
|
int error, i, len;
|
|
|
|
|
|
|
|
|
|
auth_ctx = NULL;
|
|
|
|
|
kschedule = NULL;
|
|
|
|
|
|
|
|
|
|
/* Initialize the MAC. */
|
|
|
|
|
switch (s->blkcipher.key_len) {
|
|
|
|
|
case 16:
|
|
|
|
|
axf = &auth_hash_ccm_cbc_mac_128;
|
|
|
|
|
break;
|
|
|
|
|
case 24:
|
|
|
|
|
axf = &auth_hash_ccm_cbc_mac_192;
|
|
|
|
|
break;
|
|
|
|
|
case 32:
|
|
|
|
|
axf = &auth_hash_ccm_cbc_mac_256;
|
|
|
|
|
break;
|
|
|
|
|
default:
|
|
|
|
|
error = EINVAL;
|
|
|
|
|
goto out;
|
|
|
|
|
}
|
|
|
|
|
auth_ctx = malloc(axf->ctxsize, M_CCR, M_NOWAIT);
|
|
|
|
|
if (auth_ctx == NULL) {
|
|
|
|
|
error = ENOMEM;
|
|
|
|
|
goto out;
|
|
|
|
|
}
|
|
|
|
|
axf->Init(auth_ctx);
|
|
|
|
|
axf->Setkey(auth_ctx, s->blkcipher.enckey, s->blkcipher.key_len);
|
|
|
|
|
|
|
|
|
|
/* Initialize the cipher. */
|
|
|
|
|
exf = &enc_xform_ccm;
|
2020-05-20 17:21:01 -04:00
|
|
|
kschedule = malloc(exf->ctxsize, M_CCR, M_NOWAIT);
|
|
|
|
|
if (kschedule == NULL) {
|
|
|
|
|
error = ENOMEM;
|
|
|
|
|
goto out;
|
|
|
|
|
}
|
|
|
|
|
error = exf->setkey(kschedule, s->blkcipher.enckey,
|
2019-04-24 19:31:46 -04:00
|
|
|
s->blkcipher.key_len);
|
|
|
|
|
if (error)
|
|
|
|
|
goto out;
|
|
|
|
|
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
if ((crp->crp_flags & CRYPTO_F_IV_SEPARATE) == 0) {
|
|
|
|
|
error = EINVAL;
|
|
|
|
|
goto out;
|
2019-04-24 19:31:46 -04:00
|
|
|
}
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
memcpy(iv, crp->crp_iv, AES_CCM_IV_LEN);
|
2019-04-24 19:31:46 -04:00
|
|
|
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
auth_ctx->aes_cbc_mac_ctx.authDataLength = crp->crp_aad_length;
|
|
|
|
|
auth_ctx->aes_cbc_mac_ctx.cryptDataLength = crp->crp_payload_length;
|
2019-04-24 19:31:46 -04:00
|
|
|
axf->Reinit(auth_ctx, iv, sizeof(iv));
|
|
|
|
|
|
|
|
|
|
/* MAC the AAD. */
|
2020-06-22 19:41:33 -04:00
|
|
|
if (crp->crp_aad != NULL)
|
|
|
|
|
error = axf->Update(auth_ctx, crp->crp_aad,
|
|
|
|
|
crp->crp_aad_length);
|
|
|
|
|
else
|
|
|
|
|
error = crypto_apply(crp, crp->crp_aad_start,
|
|
|
|
|
crp->crp_aad_length, axf->Update, auth_ctx);
|
|
|
|
|
if (error)
|
|
|
|
|
goto out;
|
2019-04-24 19:31:46 -04:00
|
|
|
|
|
|
|
|
exf->reinit(kschedule, iv);
|
|
|
|
|
|
|
|
|
|
/* Do encryption/decryption with MAC */
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
for (i = 0; i < crp->crp_payload_length; i += sizeof(block)) {
|
|
|
|
|
len = imin(crp->crp_payload_length - i, sizeof(block));
|
|
|
|
|
crypto_copydata(crp, crp->crp_payload_start + i, len, block);
|
2019-04-24 19:31:46 -04:00
|
|
|
bzero(block + len, sizeof(block) - len);
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
if (CRYPTO_OP_IS_ENCRYPT(crp->crp_op)) {
|
2019-04-24 19:31:46 -04:00
|
|
|
axf->Update(auth_ctx, block, len);
|
2020-05-20 17:21:01 -04:00
|
|
|
exf->encrypt(kschedule, block, block);
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
crypto_copyback(crp, crp->crp_payload_start + i, len,
|
|
|
|
|
block);
|
2019-04-24 19:31:46 -04:00
|
|
|
} else {
|
2020-05-20 17:21:01 -04:00
|
|
|
exf->decrypt(kschedule, block, block);
|
2019-04-24 19:31:46 -04:00
|
|
|
axf->Update(auth_ctx, block, len);
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
/* Finalize MAC. */
|
|
|
|
|
axf->Final(digest, auth_ctx);
|
|
|
|
|
|
|
|
|
|
/* Inject or validate tag. */
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
if (CRYPTO_OP_IS_ENCRYPT(crp->crp_op)) {
|
|
|
|
|
crypto_copyback(crp, crp->crp_digest_start, sizeof(digest),
|
|
|
|
|
digest);
|
2019-04-24 19:31:46 -04:00
|
|
|
error = 0;
|
|
|
|
|
} else {
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
char digest2[AES_CBC_MAC_HASH_LEN];
|
2019-04-24 19:31:46 -04:00
|
|
|
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
crypto_copydata(crp, crp->crp_digest_start, sizeof(digest2),
|
|
|
|
|
digest2);
|
2019-04-24 19:31:46 -04:00
|
|
|
if (timingsafe_bcmp(digest, digest2, sizeof(digest)) == 0) {
|
|
|
|
|
error = 0;
|
|
|
|
|
|
|
|
|
|
/* Tag matches, decrypt data. */
|
|
|
|
|
exf->reinit(kschedule, iv);
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
for (i = 0; i < crp->crp_payload_length;
|
|
|
|
|
i += sizeof(block)) {
|
|
|
|
|
len = imin(crp->crp_payload_length - i,
|
|
|
|
|
sizeof(block));
|
|
|
|
|
crypto_copydata(crp, crp->crp_payload_start + i,
|
|
|
|
|
len, block);
|
2019-04-24 19:31:46 -04:00
|
|
|
bzero(block + len, sizeof(block) - len);
|
2020-05-20 17:21:01 -04:00
|
|
|
exf->decrypt(kschedule, block, block);
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
crypto_copyback(crp, crp->crp_payload_start + i,
|
|
|
|
|
len, block);
|
2019-04-24 19:31:46 -04:00
|
|
|
}
|
|
|
|
|
} else
|
|
|
|
|
error = EBADMSG;
|
2020-06-03 18:11:05 -04:00
|
|
|
explicit_bzero(digest2, sizeof(digest2));
|
2019-04-24 19:31:46 -04:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
out:
|
2020-05-20 17:21:01 -04:00
|
|
|
zfree(kschedule, M_CCR);
|
|
|
|
|
zfree(auth_ctx, M_CCR);
|
2020-06-03 18:11:05 -04:00
|
|
|
explicit_bzero(block, sizeof(block));
|
|
|
|
|
explicit_bzero(iv, sizeof(iv));
|
|
|
|
|
explicit_bzero(digest, sizeof(digest));
|
2019-04-24 19:31:46 -04:00
|
|
|
crp->crp_etype = error;
|
|
|
|
|
crypto_done(crp);
|
|
|
|
|
}
|
|
|
|
|
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
static void
|
|
|
|
|
ccr_identify(driver_t *driver, device_t parent)
|
|
|
|
|
{
|
|
|
|
|
struct adapter *sc;
|
|
|
|
|
|
|
|
|
|
sc = device_get_softc(parent);
|
|
|
|
|
if (sc->cryptocaps & FW_CAPS_CONFIG_CRYPTO_LOOKASIDE &&
|
|
|
|
|
device_find_child(parent, "ccr", -1) == NULL)
|
|
|
|
|
device_add_child(parent, "ccr", -1);
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static int
|
|
|
|
|
ccr_probe(device_t dev)
|
|
|
|
|
{
|
|
|
|
|
|
|
|
|
|
device_set_desc(dev, "Chelsio Crypto Accelerator");
|
|
|
|
|
return (BUS_PROBE_DEFAULT);
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static void
|
|
|
|
|
ccr_sysctls(struct ccr_softc *sc)
|
|
|
|
|
{
|
|
|
|
|
struct sysctl_ctx_list *ctx;
|
2020-04-10 18:27:45 -04:00
|
|
|
struct sysctl_oid *oid, *port_oid;
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
struct sysctl_oid_list *children;
|
2020-04-10 18:27:45 -04:00
|
|
|
char buf[16];
|
|
|
|
|
int i;
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
|
|
|
|
|
ctx = device_get_sysctl_ctx(sc->dev);
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
|
* dev.ccr.X.
|
|
|
|
|
*/
|
|
|
|
|
oid = device_get_sysctl_tree(sc->dev);
|
|
|
|
|
children = SYSCTL_CHILDREN(oid);
|
|
|
|
|
|
2020-04-10 18:27:45 -04:00
|
|
|
SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "port_mask", CTLFLAG_RW,
|
|
|
|
|
&sc->port_mask, 0, "Mask of enabled ports");
|
|
|
|
|
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
/*
|
|
|
|
|
* dev.ccr.X.stats.
|
|
|
|
|
*/
|
2020-02-26 09:26:36 -05:00
|
|
|
oid = SYSCTL_ADD_NODE(ctx, children, OID_AUTO, "stats",
|
|
|
|
|
CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "statistics");
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
children = SYSCTL_CHILDREN(oid);
|
|
|
|
|
|
2020-06-25 20:01:31 -04:00
|
|
|
SYSCTL_ADD_COUNTER_U64(ctx, children, OID_AUTO, "hash", CTLFLAG_RD,
|
|
|
|
|
&sc->stats_hash, "Hash requests submitted");
|
|
|
|
|
SYSCTL_ADD_COUNTER_U64(ctx, children, OID_AUTO, "hmac", CTLFLAG_RD,
|
|
|
|
|
&sc->stats_hmac, "HMAC requests submitted");
|
|
|
|
|
SYSCTL_ADD_COUNTER_U64(ctx, children, OID_AUTO, "cipher_encrypt",
|
|
|
|
|
CTLFLAG_RD, &sc->stats_blkcipher_encrypt,
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
"Cipher encryption requests submitted");
|
2020-06-25 20:01:31 -04:00
|
|
|
SYSCTL_ADD_COUNTER_U64(ctx, children, OID_AUTO, "cipher_decrypt",
|
|
|
|
|
CTLFLAG_RD, &sc->stats_blkcipher_decrypt,
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
"Cipher decryption requests submitted");
|
2020-06-25 20:01:31 -04:00
|
|
|
SYSCTL_ADD_COUNTER_U64(ctx, children, OID_AUTO, "eta_encrypt",
|
|
|
|
|
CTLFLAG_RD, &sc->stats_eta_encrypt,
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
"Combined AES+HMAC encryption requests submitted");
|
2020-06-25 20:01:31 -04:00
|
|
|
SYSCTL_ADD_COUNTER_U64(ctx, children, OID_AUTO, "eta_decrypt",
|
|
|
|
|
CTLFLAG_RD, &sc->stats_eta_decrypt,
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
"Combined AES+HMAC decryption requests submitted");
|
2020-06-25 20:01:31 -04:00
|
|
|
SYSCTL_ADD_COUNTER_U64(ctx, children, OID_AUTO, "gcm_encrypt",
|
|
|
|
|
CTLFLAG_RD, &sc->stats_gcm_encrypt,
|
|
|
|
|
"AES-GCM encryption requests submitted");
|
|
|
|
|
SYSCTL_ADD_COUNTER_U64(ctx, children, OID_AUTO, "gcm_decrypt",
|
|
|
|
|
CTLFLAG_RD, &sc->stats_gcm_decrypt,
|
|
|
|
|
"AES-GCM decryption requests submitted");
|
|
|
|
|
SYSCTL_ADD_COUNTER_U64(ctx, children, OID_AUTO, "ccm_encrypt",
|
|
|
|
|
CTLFLAG_RD, &sc->stats_ccm_encrypt,
|
|
|
|
|
"AES-CCM encryption requests submitted");
|
|
|
|
|
SYSCTL_ADD_COUNTER_U64(ctx, children, OID_AUTO, "ccm_decrypt",
|
|
|
|
|
CTLFLAG_RD, &sc->stats_ccm_decrypt,
|
|
|
|
|
"AES-CCM decryption requests submitted");
|
|
|
|
|
SYSCTL_ADD_COUNTER_U64(ctx, children, OID_AUTO, "wr_nomem", CTLFLAG_RD,
|
|
|
|
|
&sc->stats_wr_nomem, "Work request memory allocation failures");
|
|
|
|
|
SYSCTL_ADD_COUNTER_U64(ctx, children, OID_AUTO, "inflight", CTLFLAG_RD,
|
|
|
|
|
&sc->stats_inflight, "Requests currently pending");
|
|
|
|
|
SYSCTL_ADD_COUNTER_U64(ctx, children, OID_AUTO, "mac_error", CTLFLAG_RD,
|
|
|
|
|
&sc->stats_mac_error, "MAC errors");
|
|
|
|
|
SYSCTL_ADD_COUNTER_U64(ctx, children, OID_AUTO, "pad_error", CTLFLAG_RD,
|
|
|
|
|
&sc->stats_pad_error, "Padding errors");
|
|
|
|
|
SYSCTL_ADD_COUNTER_U64(ctx, children, OID_AUTO, "sglist_error",
|
|
|
|
|
CTLFLAG_RD, &sc->stats_sglist_error,
|
2017-09-14 17:06:08 -04:00
|
|
|
"Requests for which DMA mapping failed");
|
2020-06-25 20:01:31 -04:00
|
|
|
SYSCTL_ADD_COUNTER_U64(ctx, children, OID_AUTO, "process_error",
|
|
|
|
|
CTLFLAG_RD, &sc->stats_process_error,
|
|
|
|
|
"Requests failed during queueing");
|
|
|
|
|
SYSCTL_ADD_COUNTER_U64(ctx, children, OID_AUTO, "sw_fallback",
|
|
|
|
|
CTLFLAG_RD, &sc->stats_sw_fallback,
|
2018-01-24 15:14:57 -05:00
|
|
|
"Requests processed by falling back to software");
|
2020-04-10 18:27:45 -04:00
|
|
|
|
|
|
|
|
/*
|
|
|
|
|
* dev.ccr.X.stats.port
|
|
|
|
|
*/
|
|
|
|
|
port_oid = SYSCTL_ADD_NODE(ctx, children, OID_AUTO, "port",
|
|
|
|
|
CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "Per-port statistics");
|
|
|
|
|
|
|
|
|
|
for (i = 0; i < nitems(sc->ports); i++) {
|
|
|
|
|
if (sc->ports[i].rxq == NULL)
|
|
|
|
|
continue;
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
|
* dev.ccr.X.stats.port.Y
|
|
|
|
|
*/
|
|
|
|
|
snprintf(buf, sizeof(buf), "%d", i);
|
|
|
|
|
oid = SYSCTL_ADD_NODE(ctx, SYSCTL_CHILDREN(port_oid), OID_AUTO,
|
|
|
|
|
buf, CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, buf);
|
|
|
|
|
children = SYSCTL_CHILDREN(oid);
|
|
|
|
|
|
|
|
|
|
SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "active_sessions",
|
|
|
|
|
CTLFLAG_RD, &sc->ports[i].active_sessions, 0,
|
|
|
|
|
"Count of active sessions");
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static void
|
|
|
|
|
ccr_init_port(struct ccr_softc *sc, int port)
|
|
|
|
|
{
|
|
|
|
|
|
|
|
|
|
sc->ports[port].txq = &sc->adapter->sge.ctrlq[port];
|
|
|
|
|
sc->ports[port].rxq =
|
|
|
|
|
&sc->adapter->sge.rxq[sc->adapter->port[port]->vi->first_rxq];
|
|
|
|
|
sc->ports[port].tx_channel_id = port;
|
|
|
|
|
_Static_assert(sizeof(sc->port_mask) * NBBY >= MAX_NPORTS - 1,
|
|
|
|
|
"Too many ports to fit in port_mask");
|
|
|
|
|
sc->port_mask |= 1u << port;
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static int
|
|
|
|
|
ccr_attach(device_t dev)
|
|
|
|
|
{
|
|
|
|
|
struct ccr_softc *sc;
|
|
|
|
|
int32_t cid;
|
2020-04-10 18:27:45 -04:00
|
|
|
int i;
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
|
|
|
|
|
sc = device_get_softc(dev);
|
|
|
|
|
sc->dev = dev;
|
|
|
|
|
sc->adapter = device_get_softc(device_get_parent(dev));
|
2020-04-10 18:27:45 -04:00
|
|
|
for_each_port(sc->adapter, i) {
|
|
|
|
|
ccr_init_port(sc, i);
|
|
|
|
|
}
|
2018-07-17 20:56:25 -04:00
|
|
|
cid = crypto_get_driverid(dev, sizeof(struct ccr_session),
|
|
|
|
|
CRYPTOCAP_F_HARDWARE);
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
if (cid < 0) {
|
|
|
|
|
device_printf(dev, "could not get crypto driver id\n");
|
|
|
|
|
return (ENXIO);
|
|
|
|
|
}
|
|
|
|
|
sc->cid = cid;
|
|
|
|
|
sc->adapter->ccr_softc = sc;
|
|
|
|
|
|
|
|
|
|
mtx_init(&sc->lock, "ccr", NULL, MTX_DEF);
|
2018-01-24 15:11:00 -05:00
|
|
|
sc->iv_aad_buf = malloc(MAX_AAD_LEN, M_CCR, M_WAITOK);
|
|
|
|
|
sc->sg_iv_aad = sglist_build(sc->iv_aad_buf, MAX_AAD_LEN, M_WAITOK);
|
2020-06-25 20:01:31 -04:00
|
|
|
sc->stats_blkcipher_encrypt = counter_u64_alloc(M_WAITOK);
|
|
|
|
|
sc->stats_blkcipher_decrypt = counter_u64_alloc(M_WAITOK);
|
|
|
|
|
sc->stats_hash = counter_u64_alloc(M_WAITOK);
|
|
|
|
|
sc->stats_hmac = counter_u64_alloc(M_WAITOK);
|
|
|
|
|
sc->stats_eta_encrypt = counter_u64_alloc(M_WAITOK);
|
|
|
|
|
sc->stats_eta_decrypt = counter_u64_alloc(M_WAITOK);
|
|
|
|
|
sc->stats_gcm_encrypt = counter_u64_alloc(M_WAITOK);
|
|
|
|
|
sc->stats_gcm_decrypt = counter_u64_alloc(M_WAITOK);
|
|
|
|
|
sc->stats_ccm_encrypt = counter_u64_alloc(M_WAITOK);
|
|
|
|
|
sc->stats_ccm_decrypt = counter_u64_alloc(M_WAITOK);
|
|
|
|
|
sc->stats_wr_nomem = counter_u64_alloc(M_WAITOK);
|
|
|
|
|
sc->stats_inflight = counter_u64_alloc(M_WAITOK);
|
|
|
|
|
sc->stats_mac_error = counter_u64_alloc(M_WAITOK);
|
|
|
|
|
sc->stats_pad_error = counter_u64_alloc(M_WAITOK);
|
|
|
|
|
sc->stats_sglist_error = counter_u64_alloc(M_WAITOK);
|
|
|
|
|
sc->stats_process_error = counter_u64_alloc(M_WAITOK);
|
|
|
|
|
sc->stats_sw_fallback = counter_u64_alloc(M_WAITOK);
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
ccr_sysctls(sc);
|
|
|
|
|
|
|
|
|
|
return (0);
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static int
|
|
|
|
|
ccr_detach(device_t dev)
|
|
|
|
|
{
|
|
|
|
|
struct ccr_softc *sc;
|
|
|
|
|
|
|
|
|
|
sc = device_get_softc(dev);
|
|
|
|
|
|
|
|
|
|
mtx_lock(&sc->lock);
|
|
|
|
|
sc->detaching = true;
|
|
|
|
|
mtx_unlock(&sc->lock);
|
|
|
|
|
|
|
|
|
|
crypto_unregister_all(sc->cid);
|
2018-07-17 20:56:25 -04:00
|
|
|
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
mtx_destroy(&sc->lock);
|
2020-06-25 20:01:31 -04:00
|
|
|
counter_u64_free(sc->stats_blkcipher_encrypt);
|
|
|
|
|
counter_u64_free(sc->stats_blkcipher_decrypt);
|
|
|
|
|
counter_u64_free(sc->stats_hash);
|
|
|
|
|
counter_u64_free(sc->stats_hmac);
|
|
|
|
|
counter_u64_free(sc->stats_eta_encrypt);
|
|
|
|
|
counter_u64_free(sc->stats_eta_decrypt);
|
|
|
|
|
counter_u64_free(sc->stats_gcm_encrypt);
|
|
|
|
|
counter_u64_free(sc->stats_gcm_decrypt);
|
|
|
|
|
counter_u64_free(sc->stats_ccm_encrypt);
|
|
|
|
|
counter_u64_free(sc->stats_ccm_decrypt);
|
|
|
|
|
counter_u64_free(sc->stats_wr_nomem);
|
|
|
|
|
counter_u64_free(sc->stats_inflight);
|
|
|
|
|
counter_u64_free(sc->stats_mac_error);
|
|
|
|
|
counter_u64_free(sc->stats_pad_error);
|
|
|
|
|
counter_u64_free(sc->stats_sglist_error);
|
|
|
|
|
counter_u64_free(sc->stats_process_error);
|
|
|
|
|
counter_u64_free(sc->stats_sw_fallback);
|
2018-01-24 15:11:00 -05:00
|
|
|
sglist_free(sc->sg_iv_aad);
|
|
|
|
|
free(sc->iv_aad_buf, M_CCR);
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
sc->adapter->ccr_softc = NULL;
|
|
|
|
|
return (0);
|
|
|
|
|
}
|
|
|
|
|
|
2018-10-29 18:24:31 -04:00
|
|
|
static void
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
ccr_init_hash_digest(struct ccr_session *s)
|
2018-10-29 18:24:31 -04:00
|
|
|
{
|
|
|
|
|
union authctx auth_ctx;
|
|
|
|
|
struct auth_hash *axf;
|
|
|
|
|
|
|
|
|
|
axf = s->hmac.auth_hash;
|
|
|
|
|
axf->Init(&auth_ctx);
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
t4_copy_partial_hash(axf->type, &auth_ctx, s->hmac.pads);
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
}
|
|
|
|
|
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
static bool
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
ccr_aes_check_keylen(int alg, int klen)
|
|
|
|
|
{
|
|
|
|
|
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
switch (klen * 8) {
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
case 128:
|
|
|
|
|
case 192:
|
|
|
|
|
if (alg == CRYPTO_AES_XTS)
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
return (false);
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
break;
|
|
|
|
|
case 256:
|
|
|
|
|
break;
|
|
|
|
|
case 512:
|
|
|
|
|
if (alg != CRYPTO_AES_XTS)
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
return (false);
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
break;
|
|
|
|
|
default:
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
return (false);
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
}
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
return (true);
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static void
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
ccr_aes_setkey(struct ccr_session *s, const void *key, int klen)
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
{
|
|
|
|
|
unsigned int ck_size, iopad_size, kctx_flits, kctx_len, kbits, mk_size;
|
|
|
|
|
unsigned int opad_present;
|
|
|
|
|
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
if (s->blkcipher.cipher_mode == SCMD_CIPH_MODE_AES_XTS)
|
|
|
|
|
kbits = (klen / 2) * 8;
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
else
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
kbits = klen * 8;
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
switch (kbits) {
|
|
|
|
|
case 128:
|
|
|
|
|
ck_size = CHCR_KEYCTX_CIPHER_KEY_SIZE_128;
|
|
|
|
|
break;
|
|
|
|
|
case 192:
|
|
|
|
|
ck_size = CHCR_KEYCTX_CIPHER_KEY_SIZE_192;
|
|
|
|
|
break;
|
|
|
|
|
case 256:
|
|
|
|
|
ck_size = CHCR_KEYCTX_CIPHER_KEY_SIZE_256;
|
|
|
|
|
break;
|
|
|
|
|
default:
|
|
|
|
|
panic("should not get here");
|
|
|
|
|
}
|
|
|
|
|
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
s->blkcipher.key_len = klen;
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
memcpy(s->blkcipher.enckey, key, s->blkcipher.key_len);
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
switch (s->blkcipher.cipher_mode) {
|
|
|
|
|
case SCMD_CIPH_MODE_AES_CBC:
|
|
|
|
|
case SCMD_CIPH_MODE_AES_XTS:
|
2018-02-26 17:12:31 -05:00
|
|
|
t4_aes_getdeckey(s->blkcipher.deckey, key, kbits);
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
break;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
kctx_len = roundup2(s->blkcipher.key_len, 16);
|
|
|
|
|
switch (s->mode) {
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
case ETA:
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
mk_size = s->hmac.mk_size;
|
|
|
|
|
opad_present = 1;
|
|
|
|
|
iopad_size = roundup2(s->hmac.partial_digest_len, 16);
|
|
|
|
|
kctx_len += iopad_size * 2;
|
|
|
|
|
break;
|
|
|
|
|
case GCM:
|
|
|
|
|
mk_size = CHCR_KEYCTX_MAC_KEY_SIZE_128;
|
|
|
|
|
opad_present = 0;
|
|
|
|
|
kctx_len += GMAC_BLOCK_LEN;
|
|
|
|
|
break;
|
2019-04-24 19:31:46 -04:00
|
|
|
case CCM:
|
|
|
|
|
switch (kbits) {
|
|
|
|
|
case 128:
|
|
|
|
|
mk_size = CHCR_KEYCTX_MAC_KEY_SIZE_128;
|
|
|
|
|
break;
|
|
|
|
|
case 192:
|
|
|
|
|
mk_size = CHCR_KEYCTX_MAC_KEY_SIZE_192;
|
|
|
|
|
break;
|
|
|
|
|
case 256:
|
|
|
|
|
mk_size = CHCR_KEYCTX_MAC_KEY_SIZE_256;
|
|
|
|
|
break;
|
|
|
|
|
default:
|
|
|
|
|
panic("should not get here");
|
|
|
|
|
}
|
|
|
|
|
opad_present = 0;
|
|
|
|
|
kctx_len *= 2;
|
|
|
|
|
break;
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
default:
|
|
|
|
|
mk_size = CHCR_KEYCTX_NO_KEY;
|
|
|
|
|
opad_present = 0;
|
|
|
|
|
break;
|
|
|
|
|
}
|
|
|
|
|
kctx_flits = (sizeof(struct _key_ctx) + kctx_len) / 16;
|
|
|
|
|
s->blkcipher.key_ctx_hdr = htobe32(V_KEY_CONTEXT_CTX_LEN(kctx_flits) |
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
V_KEY_CONTEXT_DUAL_CK(s->blkcipher.cipher_mode ==
|
|
|
|
|
SCMD_CIPH_MODE_AES_XTS) |
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
V_KEY_CONTEXT_OPAD_PRESENT(opad_present) |
|
|
|
|
|
V_KEY_CONTEXT_SALT_PRESENT(1) | V_KEY_CONTEXT_CK_SIZE(ck_size) |
|
|
|
|
|
V_KEY_CONTEXT_MK_SIZE(mk_size) | V_KEY_CONTEXT_VALID(1));
|
|
|
|
|
}
|
|
|
|
|
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
static bool
|
|
|
|
|
ccr_auth_supported(const struct crypto_session_params *csp)
|
|
|
|
|
{
|
|
|
|
|
|
|
|
|
|
switch (csp->csp_auth_alg) {
|
|
|
|
|
case CRYPTO_SHA1:
|
|
|
|
|
case CRYPTO_SHA2_224:
|
|
|
|
|
case CRYPTO_SHA2_256:
|
|
|
|
|
case CRYPTO_SHA2_384:
|
|
|
|
|
case CRYPTO_SHA2_512:
|
|
|
|
|
case CRYPTO_SHA1_HMAC:
|
|
|
|
|
case CRYPTO_SHA2_224_HMAC:
|
|
|
|
|
case CRYPTO_SHA2_256_HMAC:
|
|
|
|
|
case CRYPTO_SHA2_384_HMAC:
|
|
|
|
|
case CRYPTO_SHA2_512_HMAC:
|
|
|
|
|
break;
|
|
|
|
|
default:
|
|
|
|
|
return (false);
|
|
|
|
|
}
|
|
|
|
|
return (true);
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static bool
|
|
|
|
|
ccr_cipher_supported(const struct crypto_session_params *csp)
|
|
|
|
|
{
|
|
|
|
|
|
|
|
|
|
switch (csp->csp_cipher_alg) {
|
|
|
|
|
case CRYPTO_AES_CBC:
|
|
|
|
|
if (csp->csp_ivlen != AES_BLOCK_LEN)
|
|
|
|
|
return (false);
|
|
|
|
|
break;
|
|
|
|
|
case CRYPTO_AES_ICM:
|
|
|
|
|
if (csp->csp_ivlen != AES_BLOCK_LEN)
|
|
|
|
|
return (false);
|
|
|
|
|
break;
|
|
|
|
|
case CRYPTO_AES_XTS:
|
|
|
|
|
if (csp->csp_ivlen != AES_XTS_IV_LEN)
|
|
|
|
|
return (false);
|
|
|
|
|
break;
|
|
|
|
|
default:
|
|
|
|
|
return (false);
|
|
|
|
|
}
|
|
|
|
|
return (ccr_aes_check_keylen(csp->csp_cipher_alg,
|
|
|
|
|
csp->csp_cipher_klen));
|
|
|
|
|
}
|
|
|
|
|
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
static int
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
ccr_cipher_mode(const struct crypto_session_params *csp)
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
{
|
|
|
|
|
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
switch (csp->csp_cipher_alg) {
|
|
|
|
|
case CRYPTO_AES_CBC:
|
|
|
|
|
return (SCMD_CIPH_MODE_AES_CBC);
|
|
|
|
|
case CRYPTO_AES_ICM:
|
|
|
|
|
return (SCMD_CIPH_MODE_AES_CTR);
|
|
|
|
|
case CRYPTO_AES_NIST_GCM_16:
|
|
|
|
|
return (SCMD_CIPH_MODE_AES_GCM);
|
|
|
|
|
case CRYPTO_AES_XTS:
|
|
|
|
|
return (SCMD_CIPH_MODE_AES_XTS);
|
|
|
|
|
case CRYPTO_AES_CCM_16:
|
|
|
|
|
return (SCMD_CIPH_MODE_AES_CCM);
|
|
|
|
|
default:
|
|
|
|
|
return (SCMD_CIPH_MODE_NOP);
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static int
|
|
|
|
|
ccr_probesession(device_t dev, const struct crypto_session_params *csp)
|
|
|
|
|
{
|
|
|
|
|
unsigned int cipher_mode;
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
|
2020-06-22 19:41:33 -04:00
|
|
|
if ((csp->csp_flags & ~(CSP_F_SEPARATE_OUTPUT | CSP_F_SEPARATE_AAD)) !=
|
|
|
|
|
0)
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
return (EINVAL);
|
|
|
|
|
switch (csp->csp_mode) {
|
|
|
|
|
case CSP_MODE_DIGEST:
|
|
|
|
|
if (!ccr_auth_supported(csp))
|
|
|
|
|
return (EINVAL);
|
|
|
|
|
break;
|
|
|
|
|
case CSP_MODE_CIPHER:
|
|
|
|
|
if (!ccr_cipher_supported(csp))
|
|
|
|
|
return (EINVAL);
|
|
|
|
|
break;
|
|
|
|
|
case CSP_MODE_AEAD:
|
|
|
|
|
switch (csp->csp_cipher_alg) {
|
|
|
|
|
case CRYPTO_AES_NIST_GCM_16:
|
|
|
|
|
if (csp->csp_ivlen != AES_GCM_IV_LEN)
|
|
|
|
|
return (EINVAL);
|
|
|
|
|
if (csp->csp_auth_mlen < 0 ||
|
|
|
|
|
csp->csp_auth_mlen > AES_GMAC_HASH_LEN)
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
return (EINVAL);
|
|
|
|
|
break;
|
2019-04-24 19:31:46 -04:00
|
|
|
case CRYPTO_AES_CCM_16:
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
if (csp->csp_ivlen != AES_CCM_IV_LEN)
|
|
|
|
|
return (EINVAL);
|
|
|
|
|
if (csp->csp_auth_mlen < 0 ||
|
|
|
|
|
csp->csp_auth_mlen > AES_CBC_MAC_HASH_LEN)
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
return (EINVAL);
|
|
|
|
|
break;
|
|
|
|
|
default:
|
|
|
|
|
return (EINVAL);
|
|
|
|
|
}
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
break;
|
|
|
|
|
case CSP_MODE_ETA:
|
|
|
|
|
if (!ccr_auth_supported(csp) || !ccr_cipher_supported(csp))
|
|
|
|
|
return (EINVAL);
|
|
|
|
|
break;
|
|
|
|
|
default:
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
return (EINVAL);
|
2018-10-29 18:24:31 -04:00
|
|
|
}
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
if (csp->csp_cipher_klen != 0) {
|
|
|
|
|
cipher_mode = ccr_cipher_mode(csp);
|
|
|
|
|
if (cipher_mode == SCMD_CIPH_MODE_NOP)
|
|
|
|
|
return (EINVAL);
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
return (CRYPTODEV_PROBE_HARDWARE);
|
|
|
|
|
}
|
|
|
|
|
|
2020-04-10 18:27:45 -04:00
|
|
|
/*
|
|
|
|
|
* Select an available port with the lowest number of active sessions.
|
|
|
|
|
*/
|
|
|
|
|
static struct ccr_port *
|
|
|
|
|
ccr_choose_port(struct ccr_softc *sc)
|
|
|
|
|
{
|
|
|
|
|
struct ccr_port *best, *p;
|
|
|
|
|
int i;
|
|
|
|
|
|
|
|
|
|
mtx_assert(&sc->lock, MA_OWNED);
|
|
|
|
|
best = NULL;
|
|
|
|
|
for (i = 0; i < nitems(sc->ports); i++) {
|
|
|
|
|
p = &sc->ports[i];
|
|
|
|
|
|
|
|
|
|
/* Ignore non-existent ports. */
|
|
|
|
|
if (p->rxq == NULL)
|
|
|
|
|
continue;
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
|
* XXX: Ignore ports whose queues aren't initialized.
|
|
|
|
|
* This is racy as the rxq can be destroyed by the
|
|
|
|
|
* associated VI detaching. Eventually ccr should use
|
|
|
|
|
* dedicated queues.
|
|
|
|
|
*/
|
|
|
|
|
if (p->rxq->iq.adapter == NULL || p->txq->adapter == NULL)
|
|
|
|
|
continue;
|
|
|
|
|
|
|
|
|
|
if ((sc->port_mask & (1u << i)) == 0)
|
|
|
|
|
continue;
|
|
|
|
|
|
|
|
|
|
if (best == NULL ||
|
|
|
|
|
p->active_sessions < best->active_sessions)
|
|
|
|
|
best = p;
|
|
|
|
|
}
|
|
|
|
|
return (best);
|
|
|
|
|
}
|
|
|
|
|
|
2020-06-25 20:01:31 -04:00
|
|
|
static void
|
|
|
|
|
ccr_delete_session(struct ccr_session *s)
|
|
|
|
|
{
|
|
|
|
|
sglist_free(s->sg_input);
|
|
|
|
|
sglist_free(s->sg_output);
|
|
|
|
|
sglist_free(s->sg_ulptx);
|
|
|
|
|
sglist_free(s->sg_dsgl);
|
|
|
|
|
mtx_destroy(&s->lock);
|
|
|
|
|
}
|
|
|
|
|
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
static int
|
|
|
|
|
ccr_newsession(device_t dev, crypto_session_t cses,
|
|
|
|
|
const struct crypto_session_params *csp)
|
|
|
|
|
{
|
|
|
|
|
struct ccr_softc *sc;
|
|
|
|
|
struct ccr_session *s;
|
|
|
|
|
struct auth_hash *auth_hash;
|
|
|
|
|
unsigned int auth_mode, cipher_mode, mk_size;
|
|
|
|
|
unsigned int partial_digest_len;
|
|
|
|
|
|
|
|
|
|
switch (csp->csp_auth_alg) {
|
|
|
|
|
case CRYPTO_SHA1:
|
|
|
|
|
case CRYPTO_SHA1_HMAC:
|
|
|
|
|
auth_hash = &auth_hash_hmac_sha1;
|
|
|
|
|
auth_mode = SCMD_AUTH_MODE_SHA1;
|
|
|
|
|
mk_size = CHCR_KEYCTX_MAC_KEY_SIZE_160;
|
|
|
|
|
partial_digest_len = SHA1_HASH_LEN;
|
|
|
|
|
break;
|
|
|
|
|
case CRYPTO_SHA2_224:
|
|
|
|
|
case CRYPTO_SHA2_224_HMAC:
|
|
|
|
|
auth_hash = &auth_hash_hmac_sha2_224;
|
|
|
|
|
auth_mode = SCMD_AUTH_MODE_SHA224;
|
|
|
|
|
mk_size = CHCR_KEYCTX_MAC_KEY_SIZE_256;
|
|
|
|
|
partial_digest_len = SHA2_256_HASH_LEN;
|
|
|
|
|
break;
|
|
|
|
|
case CRYPTO_SHA2_256:
|
|
|
|
|
case CRYPTO_SHA2_256_HMAC:
|
|
|
|
|
auth_hash = &auth_hash_hmac_sha2_256;
|
|
|
|
|
auth_mode = SCMD_AUTH_MODE_SHA256;
|
|
|
|
|
mk_size = CHCR_KEYCTX_MAC_KEY_SIZE_256;
|
|
|
|
|
partial_digest_len = SHA2_256_HASH_LEN;
|
|
|
|
|
break;
|
|
|
|
|
case CRYPTO_SHA2_384:
|
|
|
|
|
case CRYPTO_SHA2_384_HMAC:
|
|
|
|
|
auth_hash = &auth_hash_hmac_sha2_384;
|
|
|
|
|
auth_mode = SCMD_AUTH_MODE_SHA512_384;
|
|
|
|
|
mk_size = CHCR_KEYCTX_MAC_KEY_SIZE_512;
|
|
|
|
|
partial_digest_len = SHA2_512_HASH_LEN;
|
|
|
|
|
break;
|
|
|
|
|
case CRYPTO_SHA2_512:
|
|
|
|
|
case CRYPTO_SHA2_512_HMAC:
|
|
|
|
|
auth_hash = &auth_hash_hmac_sha2_512;
|
|
|
|
|
auth_mode = SCMD_AUTH_MODE_SHA512_512;
|
|
|
|
|
mk_size = CHCR_KEYCTX_MAC_KEY_SIZE_512;
|
|
|
|
|
partial_digest_len = SHA2_512_HASH_LEN;
|
|
|
|
|
break;
|
|
|
|
|
default:
|
|
|
|
|
auth_hash = NULL;
|
|
|
|
|
auth_mode = SCMD_AUTH_MODE_NOP;
|
|
|
|
|
mk_size = 0;
|
|
|
|
|
partial_digest_len = 0;
|
|
|
|
|
break;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
cipher_mode = ccr_cipher_mode(csp);
|
|
|
|
|
|
|
|
|
|
#ifdef INVARIANTS
|
|
|
|
|
switch (csp->csp_mode) {
|
|
|
|
|
case CSP_MODE_CIPHER:
|
|
|
|
|
if (cipher_mode == SCMD_CIPH_MODE_NOP ||
|
|
|
|
|
cipher_mode == SCMD_CIPH_MODE_AES_GCM ||
|
|
|
|
|
cipher_mode == SCMD_CIPH_MODE_AES_CCM)
|
|
|
|
|
panic("invalid cipher algo");
|
|
|
|
|
break;
|
|
|
|
|
case CSP_MODE_DIGEST:
|
|
|
|
|
if (auth_mode == SCMD_AUTH_MODE_NOP)
|
|
|
|
|
panic("invalid auth algo");
|
|
|
|
|
break;
|
|
|
|
|
case CSP_MODE_AEAD:
|
|
|
|
|
if (cipher_mode != SCMD_CIPH_MODE_AES_GCM &&
|
|
|
|
|
cipher_mode != SCMD_CIPH_MODE_AES_CCM)
|
|
|
|
|
panic("invalid aead cipher algo");
|
|
|
|
|
if (auth_mode != SCMD_AUTH_MODE_NOP)
|
|
|
|
|
panic("invalid aead auth aglo");
|
|
|
|
|
break;
|
|
|
|
|
case CSP_MODE_ETA:
|
|
|
|
|
if (cipher_mode == SCMD_CIPH_MODE_NOP ||
|
|
|
|
|
cipher_mode == SCMD_CIPH_MODE_AES_GCM ||
|
|
|
|
|
cipher_mode == SCMD_CIPH_MODE_AES_CCM)
|
|
|
|
|
panic("invalid cipher algo");
|
|
|
|
|
if (auth_mode == SCMD_AUTH_MODE_NOP)
|
|
|
|
|
panic("invalid auth algo");
|
|
|
|
|
break;
|
|
|
|
|
default:
|
|
|
|
|
panic("invalid csp mode");
|
|
|
|
|
}
|
|
|
|
|
#endif
|
|
|
|
|
|
2020-06-25 20:01:31 -04:00
|
|
|
s = crypto_get_driver_session(cses);
|
|
|
|
|
mtx_init(&s->lock, "ccr session", NULL, MTX_DEF);
|
|
|
|
|
s->sg_input = sglist_alloc(TX_SGL_SEGS, M_NOWAIT);
|
|
|
|
|
s->sg_output = sglist_alloc(TX_SGL_SEGS, M_NOWAIT);
|
|
|
|
|
s->sg_ulptx = sglist_alloc(TX_SGL_SEGS, M_NOWAIT);
|
|
|
|
|
s->sg_dsgl = sglist_alloc(MAX_RX_PHYS_DSGL_SGE, M_NOWAIT);
|
|
|
|
|
if (s->sg_input == NULL || s->sg_output == NULL ||
|
|
|
|
|
s->sg_ulptx == NULL || s->sg_dsgl == NULL) {
|
|
|
|
|
ccr_delete_session(s);
|
|
|
|
|
return (ENOMEM);
|
|
|
|
|
}
|
|
|
|
|
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
sc = device_get_softc(dev);
|
2019-01-15 13:53:45 -05:00
|
|
|
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
mtx_lock(&sc->lock);
|
|
|
|
|
if (sc->detaching) {
|
|
|
|
|
mtx_unlock(&sc->lock);
|
2020-06-25 20:01:31 -04:00
|
|
|
ccr_delete_session(s);
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
return (ENXIO);
|
|
|
|
|
}
|
|
|
|
|
|
2020-04-10 18:27:45 -04:00
|
|
|
s->port = ccr_choose_port(sc);
|
|
|
|
|
if (s->port == NULL) {
|
|
|
|
|
mtx_unlock(&sc->lock);
|
2020-06-25 20:01:31 -04:00
|
|
|
ccr_delete_session(s);
|
2020-04-10 18:27:45 -04:00
|
|
|
return (ENXIO);
|
|
|
|
|
}
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
switch (csp->csp_mode) {
|
|
|
|
|
case CSP_MODE_AEAD:
|
|
|
|
|
if (cipher_mode == SCMD_CIPH_MODE_AES_CCM)
|
|
|
|
|
s->mode = CCM;
|
|
|
|
|
else
|
|
|
|
|
s->mode = GCM;
|
|
|
|
|
break;
|
|
|
|
|
case CSP_MODE_ETA:
|
|
|
|
|
s->mode = ETA;
|
|
|
|
|
break;
|
|
|
|
|
case CSP_MODE_DIGEST:
|
|
|
|
|
if (csp->csp_auth_klen != 0)
|
2018-10-29 18:24:31 -04:00
|
|
|
s->mode = HMAC;
|
|
|
|
|
else
|
|
|
|
|
s->mode = HASH;
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
break;
|
|
|
|
|
case CSP_MODE_CIPHER:
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
s->mode = BLKCIPHER;
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
break;
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
}
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
|
|
|
|
|
if (s->mode == GCM) {
|
|
|
|
|
if (csp->csp_auth_mlen == 0)
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
s->gmac.hash_len = AES_GMAC_HASH_LEN;
|
|
|
|
|
else
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
s->gmac.hash_len = csp->csp_auth_mlen;
|
|
|
|
|
t4_init_gmac_hash(csp->csp_cipher_key, csp->csp_cipher_klen,
|
2019-11-12 19:53:45 -05:00
|
|
|
s->gmac.ghash_h);
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
} else if (s->mode == CCM) {
|
|
|
|
|
if (csp->csp_auth_mlen == 0)
|
2019-04-24 19:31:46 -04:00
|
|
|
s->ccm_mac.hash_len = AES_CBC_MAC_HASH_LEN;
|
|
|
|
|
else
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
s->ccm_mac.hash_len = csp->csp_auth_mlen;
|
|
|
|
|
} else if (auth_mode != SCMD_AUTH_MODE_NOP) {
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
s->hmac.auth_hash = auth_hash;
|
|
|
|
|
s->hmac.auth_mode = auth_mode;
|
|
|
|
|
s->hmac.mk_size = mk_size;
|
|
|
|
|
s->hmac.partial_digest_len = partial_digest_len;
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
if (csp->csp_auth_mlen == 0)
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
s->hmac.hash_len = auth_hash->hashsize;
|
|
|
|
|
else
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
s->hmac.hash_len = csp->csp_auth_mlen;
|
|
|
|
|
if (csp->csp_auth_key != NULL)
|
2019-11-12 19:53:45 -05:00
|
|
|
t4_init_hmac_digest(auth_hash, partial_digest_len,
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
csp->csp_auth_key, csp->csp_auth_klen,
|
|
|
|
|
s->hmac.pads);
|
2018-10-29 18:24:31 -04:00
|
|
|
else
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
ccr_init_hash_digest(s);
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
}
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
if (cipher_mode != SCMD_CIPH_MODE_NOP) {
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
s->blkcipher.cipher_mode = cipher_mode;
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
s->blkcipher.iv_len = csp->csp_ivlen;
|
|
|
|
|
if (csp->csp_cipher_key != NULL)
|
|
|
|
|
ccr_aes_setkey(s, csp->csp_cipher_key,
|
|
|
|
|
csp->csp_cipher_klen);
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
}
|
|
|
|
|
|
2020-04-10 18:27:45 -04:00
|
|
|
s->port->active_sessions++;
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
mtx_unlock(&sc->lock);
|
|
|
|
|
return (0);
|
|
|
|
|
}
|
|
|
|
|
|
2018-07-17 20:56:25 -04:00
|
|
|
static void
|
|
|
|
|
ccr_freesession(device_t dev, crypto_session_t cses)
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
{
|
|
|
|
|
struct ccr_softc *sc;
|
2018-07-17 20:56:25 -04:00
|
|
|
struct ccr_session *s;
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
|
|
|
|
|
sc = device_get_softc(dev);
|
2018-07-17 20:56:25 -04:00
|
|
|
s = crypto_get_driver_session(cses);
|
2020-06-25 20:01:31 -04:00
|
|
|
#ifdef INVARIANTS
|
2018-07-17 20:56:25 -04:00
|
|
|
if (s->pending != 0)
|
|
|
|
|
device_printf(dev,
|
|
|
|
|
"session %p freed with %d pending requests\n", s,
|
|
|
|
|
s->pending);
|
2020-06-25 20:01:31 -04:00
|
|
|
#endif
|
|
|
|
|
mtx_lock(&sc->lock);
|
2020-04-10 18:27:45 -04:00
|
|
|
s->port->active_sessions--;
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
mtx_unlock(&sc->lock);
|
2020-06-25 20:01:31 -04:00
|
|
|
ccr_delete_session(s);
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static int
|
|
|
|
|
ccr_process(device_t dev, struct cryptop *crp, int hint)
|
|
|
|
|
{
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
const struct crypto_session_params *csp;
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
struct ccr_softc *sc;
|
|
|
|
|
struct ccr_session *s;
|
|
|
|
|
int error;
|
|
|
|
|
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
csp = crypto_get_params(crp->crp_session);
|
2018-07-17 20:56:25 -04:00
|
|
|
s = crypto_get_driver_session(crp->crp_session);
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
sc = device_get_softc(dev);
|
|
|
|
|
|
2020-06-25 20:01:31 -04:00
|
|
|
mtx_lock(&s->lock);
|
|
|
|
|
error = ccr_populate_sglist(s->sg_input, &crp->crp_buf);
|
2020-05-25 18:23:13 -04:00
|
|
|
if (error == 0 && CRYPTO_HAS_OUTPUT_BUFFER(crp))
|
2020-06-25 20:01:31 -04:00
|
|
|
error = ccr_populate_sglist(s->sg_output, &crp->crp_obuf);
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
if (error) {
|
2020-06-25 20:01:31 -04:00
|
|
|
counter_u64_add(sc->stats_sglist_error, 1);
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
goto out;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
switch (s->mode) {
|
2018-10-29 18:24:31 -04:00
|
|
|
case HASH:
|
|
|
|
|
error = ccr_hash(sc, s, crp);
|
|
|
|
|
if (error == 0)
|
2020-06-25 20:01:31 -04:00
|
|
|
counter_u64_add(sc->stats_hash, 1);
|
2018-10-29 18:24:31 -04:00
|
|
|
break;
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
case HMAC:
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
if (crp->crp_auth_key != NULL)
|
2019-11-12 19:53:45 -05:00
|
|
|
t4_init_hmac_digest(s->hmac.auth_hash,
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
s->hmac.partial_digest_len, crp->crp_auth_key,
|
|
|
|
|
csp->csp_auth_klen, s->hmac.pads);
|
2018-10-29 18:24:31 -04:00
|
|
|
error = ccr_hash(sc, s, crp);
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
if (error == 0)
|
2020-06-25 20:01:31 -04:00
|
|
|
counter_u64_add(sc->stats_hmac, 1);
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
break;
|
|
|
|
|
case BLKCIPHER:
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
if (crp->crp_cipher_key != NULL)
|
|
|
|
|
ccr_aes_setkey(s, crp->crp_cipher_key,
|
|
|
|
|
csp->csp_cipher_klen);
|
2018-07-17 20:56:25 -04:00
|
|
|
error = ccr_blkcipher(sc, s, crp);
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
if (error == 0) {
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
if (CRYPTO_OP_IS_ENCRYPT(crp->crp_op))
|
2020-06-25 20:01:31 -04:00
|
|
|
counter_u64_add(sc->stats_blkcipher_encrypt, 1);
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
else
|
2020-06-25 20:01:31 -04:00
|
|
|
counter_u64_add(sc->stats_blkcipher_decrypt, 1);
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
}
|
|
|
|
|
break;
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
case ETA:
|
|
|
|
|
if (crp->crp_auth_key != NULL)
|
2019-11-12 19:53:45 -05:00
|
|
|
t4_init_hmac_digest(s->hmac.auth_hash,
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
s->hmac.partial_digest_len, crp->crp_auth_key,
|
|
|
|
|
csp->csp_auth_klen, s->hmac.pads);
|
|
|
|
|
if (crp->crp_cipher_key != NULL)
|
|
|
|
|
ccr_aes_setkey(s, crp->crp_cipher_key,
|
|
|
|
|
csp->csp_cipher_klen);
|
|
|
|
|
error = ccr_eta(sc, s, crp);
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
if (error == 0) {
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
if (CRYPTO_OP_IS_ENCRYPT(crp->crp_op))
|
2020-06-25 20:01:31 -04:00
|
|
|
counter_u64_add(sc->stats_eta_encrypt, 1);
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
else
|
2020-06-25 20:01:31 -04:00
|
|
|
counter_u64_add(sc->stats_eta_decrypt, 1);
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
}
|
|
|
|
|
break;
|
|
|
|
|
case GCM:
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
if (crp->crp_cipher_key != NULL) {
|
|
|
|
|
t4_init_gmac_hash(crp->crp_cipher_key,
|
|
|
|
|
csp->csp_cipher_klen, s->gmac.ghash_h);
|
|
|
|
|
ccr_aes_setkey(s, crp->crp_cipher_key,
|
|
|
|
|
csp->csp_cipher_klen);
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
}
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
if (crp->crp_payload_length == 0) {
|
2020-06-25 20:01:31 -04:00
|
|
|
mtx_unlock(&s->lock);
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
ccr_gcm_soft(s, crp);
|
2017-06-08 17:06:18 -04:00
|
|
|
return (0);
|
|
|
|
|
}
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
error = ccr_gcm(sc, s, crp);
|
2018-01-24 15:14:57 -05:00
|
|
|
if (error == EMSGSIZE) {
|
2020-06-25 20:01:31 -04:00
|
|
|
counter_u64_add(sc->stats_sw_fallback, 1);
|
|
|
|
|
mtx_unlock(&s->lock);
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
ccr_gcm_soft(s, crp);
|
2018-01-24 15:14:57 -05:00
|
|
|
return (0);
|
|
|
|
|
}
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
if (error == 0) {
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
if (CRYPTO_OP_IS_ENCRYPT(crp->crp_op))
|
2020-06-25 20:01:31 -04:00
|
|
|
counter_u64_add(sc->stats_gcm_encrypt, 1);
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
else
|
2020-06-25 20:01:31 -04:00
|
|
|
counter_u64_add(sc->stats_gcm_decrypt, 1);
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
}
|
|
|
|
|
break;
|
2019-04-24 19:31:46 -04:00
|
|
|
case CCM:
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
if (crp->crp_cipher_key != NULL) {
|
|
|
|
|
ccr_aes_setkey(s, crp->crp_cipher_key,
|
|
|
|
|
csp->csp_cipher_klen);
|
2019-04-24 19:31:46 -04:00
|
|
|
}
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
error = ccr_ccm(sc, s, crp);
|
2019-04-24 19:31:46 -04:00
|
|
|
if (error == EMSGSIZE) {
|
2020-06-25 20:01:31 -04:00
|
|
|
counter_u64_add(sc->stats_sw_fallback, 1);
|
|
|
|
|
mtx_unlock(&s->lock);
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
ccr_ccm_soft(s, crp);
|
2019-04-24 19:31:46 -04:00
|
|
|
return (0);
|
|
|
|
|
}
|
|
|
|
|
if (error == 0) {
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
if (CRYPTO_OP_IS_ENCRYPT(crp->crp_op))
|
2020-06-25 20:01:31 -04:00
|
|
|
counter_u64_add(sc->stats_ccm_encrypt, 1);
|
2019-04-24 19:31:46 -04:00
|
|
|
else
|
2020-06-25 20:01:31 -04:00
|
|
|
counter_u64_add(sc->stats_ccm_decrypt, 1);
|
2019-04-24 19:31:46 -04:00
|
|
|
}
|
|
|
|
|
break;
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
if (error == 0) {
|
2020-06-25 20:01:31 -04:00
|
|
|
#ifdef INVARIANTS
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
s->pending++;
|
2020-06-25 20:01:31 -04:00
|
|
|
#endif
|
|
|
|
|
counter_u64_add(sc->stats_inflight, 1);
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
} else
|
2020-06-25 20:01:31 -04:00
|
|
|
counter_u64_add(sc->stats_process_error, 1);
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
|
|
|
|
|
out:
|
2020-06-25 20:01:31 -04:00
|
|
|
mtx_unlock(&s->lock);
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
|
|
|
|
|
if (error) {
|
|
|
|
|
crp->crp_etype = error;
|
|
|
|
|
crypto_done(crp);
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
return (0);
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static int
|
|
|
|
|
do_cpl6_fw_pld(struct sge_iq *iq, const struct rss_header *rss,
|
|
|
|
|
struct mbuf *m)
|
|
|
|
|
{
|
|
|
|
|
struct ccr_softc *sc = iq->adapter->ccr_softc;
|
|
|
|
|
struct ccr_session *s;
|
|
|
|
|
const struct cpl_fw6_pld *cpl;
|
|
|
|
|
struct cryptop *crp;
|
2018-07-17 20:56:25 -04:00
|
|
|
uint32_t status;
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
int error;
|
|
|
|
|
|
|
|
|
|
if (m != NULL)
|
|
|
|
|
cpl = mtod(m, const void *);
|
|
|
|
|
else
|
|
|
|
|
cpl = (const void *)(rss + 1);
|
|
|
|
|
|
|
|
|
|
crp = (struct cryptop *)(uintptr_t)be64toh(cpl->data[1]);
|
2018-07-17 20:56:25 -04:00
|
|
|
s = crypto_get_driver_session(crp->crp_session);
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
status = be64toh(cpl->data[0]);
|
|
|
|
|
if (CHK_MAC_ERR_BIT(status) || CHK_PAD_ERR_BIT(status))
|
|
|
|
|
error = EBADMSG;
|
|
|
|
|
else
|
|
|
|
|
error = 0;
|
|
|
|
|
|
2020-06-25 20:01:31 -04:00
|
|
|
#ifdef INVARIANTS
|
|
|
|
|
mtx_lock(&s->lock);
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
s->pending--;
|
2020-06-25 20:01:31 -04:00
|
|
|
mtx_unlock(&s->lock);
|
|
|
|
|
#endif
|
|
|
|
|
counter_u64_add(sc->stats_inflight, -1);
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
|
|
|
|
|
switch (s->mode) {
|
2018-10-29 18:24:31 -04:00
|
|
|
case HASH:
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
case HMAC:
|
2018-10-29 18:24:31 -04:00
|
|
|
error = ccr_hash_done(sc, s, crp, cpl, error);
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
break;
|
|
|
|
|
case BLKCIPHER:
|
|
|
|
|
error = ccr_blkcipher_done(sc, s, crp, cpl, error);
|
|
|
|
|
break;
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
case ETA:
|
|
|
|
|
error = ccr_eta_done(sc, s, crp, cpl, error);
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
break;
|
|
|
|
|
case GCM:
|
|
|
|
|
error = ccr_gcm_done(sc, s, crp, cpl, error);
|
|
|
|
|
break;
|
2019-04-24 19:31:46 -04:00
|
|
|
case CCM:
|
|
|
|
|
error = ccr_ccm_done(sc, s, crp, cpl, error);
|
|
|
|
|
break;
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
if (error == EBADMSG) {
|
|
|
|
|
if (CHK_MAC_ERR_BIT(status))
|
2020-06-25 20:01:31 -04:00
|
|
|
counter_u64_add(sc->stats_mac_error, 1);
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
if (CHK_PAD_ERR_BIT(status))
|
2020-06-25 20:01:31 -04:00
|
|
|
counter_u64_add(sc->stats_pad_error, 1);
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
}
|
|
|
|
|
crp->crp_etype = error;
|
|
|
|
|
crypto_done(crp);
|
|
|
|
|
m_freem(m);
|
|
|
|
|
return (0);
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static int
|
|
|
|
|
ccr_modevent(module_t mod, int cmd, void *arg)
|
|
|
|
|
{
|
|
|
|
|
|
|
|
|
|
switch (cmd) {
|
|
|
|
|
case MOD_LOAD:
|
|
|
|
|
t4_register_cpl_handler(CPL_FW6_PLD, do_cpl6_fw_pld);
|
|
|
|
|
return (0);
|
|
|
|
|
case MOD_UNLOAD:
|
|
|
|
|
t4_register_cpl_handler(CPL_FW6_PLD, NULL);
|
|
|
|
|
return (0);
|
|
|
|
|
default:
|
|
|
|
|
return (EOPNOTSUPP);
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static device_method_t ccr_methods[] = {
|
|
|
|
|
DEVMETHOD(device_identify, ccr_identify),
|
|
|
|
|
DEVMETHOD(device_probe, ccr_probe),
|
|
|
|
|
DEVMETHOD(device_attach, ccr_attach),
|
|
|
|
|
DEVMETHOD(device_detach, ccr_detach),
|
|
|
|
|
|
Refactor driver and consumer interfaces for OCF (in-kernel crypto).
- The linked list of cryptoini structures used in session
initialization is replaced with a new flat structure: struct
crypto_session_params. This session includes a new mode to define
how the other fields should be interpreted. Available modes
include:
- COMPRESS (for compression/decompression)
- CIPHER (for simply encryption/decryption)
- DIGEST (computing and verifying digests)
- AEAD (combined auth and encryption such as AES-GCM and AES-CCM)
- ETA (combined auth and encryption using encrypt-then-authenticate)
Additional modes could be added in the future (e.g. if we wanted to
support TLS MtE for AES-CBC in the kernel we could add a new mode
for that. TLS modes might also affect how AAD is interpreted, etc.)
The flat structure also includes the key lengths and algorithms as
before. However, code doesn't have to walk the linked list and
switch on the algorithm to determine which key is the auth key vs
encryption key. The 'csp_auth_*' fields are always used for auth
keys and settings and 'csp_cipher_*' for cipher. (Compression
algorithms are stored in csp_cipher_alg.)
- Drivers no longer register a list of supported algorithms. This
doesn't quite work when you factor in modes (e.g. a driver might
support both AES-CBC and SHA2-256-HMAC separately but not combined
for ETA). Instead, a new 'crypto_probesession' method has been
added to the kobj interface for symmteric crypto drivers. This
method returns a negative value on success (similar to how
device_probe works) and the crypto framework uses this value to pick
the "best" driver. There are three constants for hardware
(e.g. ccr), accelerated software (e.g. aesni), and plain software
(cryptosoft) that give preference in that order. One effect of this
is that if you request only hardware when creating a new session,
you will no longer get a session using accelerated software.
Another effect is that the default setting to disallow software
crypto via /dev/crypto now disables accelerated software.
Once a driver is chosen, 'crypto_newsession' is invoked as before.
- Crypto operations are now solely described by the flat 'cryptop'
structure. The linked list of descriptors has been removed.
A separate enum has been added to describe the type of data buffer
in use instead of using CRYPTO_F_* flags to make it easier to add
more types in the future if needed (e.g. wired userspace buffers for
zero-copy). It will also make it easier to re-introduce separate
input and output buffers (in-kernel TLS would benefit from this).
Try to make the flags related to IV handling less insane:
- CRYPTO_F_IV_SEPARATE means that the IV is stored in the 'crp_iv'
member of the operation structure. If this flag is not set, the
IV is stored in the data buffer at the 'crp_iv_start' offset.
- CRYPTO_F_IV_GENERATE means that a random IV should be generated
and stored into the data buffer. This cannot be used with
CRYPTO_F_IV_SEPARATE.
If a consumer wants to deal with explicit vs implicit IVs, etc. it
can always generate the IV however it needs and store partial IVs in
the buffer and the full IV/nonce in crp_iv and set
CRYPTO_F_IV_SEPARATE.
The layout of the buffer is now described via fields in cryptop.
crp_aad_start and crp_aad_length define the boundaries of any AAD.
Previously with GCM and CCM you defined an auth crd with this range,
but for ETA your auth crd had to span both the AAD and plaintext
(and they had to be adjacent).
crp_payload_start and crp_payload_length define the boundaries of
the plaintext/ciphertext. Modes that only do a single operation
(COMPRESS, CIPHER, DIGEST) should only use this region and leave the
AAD region empty.
If a digest is present (or should be generated), it's starting
location is marked by crp_digest_start.
Instead of using the CRD_F_ENCRYPT flag to determine the direction
of the operation, cryptop now includes an 'op' field defining the
operation to perform. For digests I've added a new VERIFY digest
mode which assumes a digest is present in the input and fails the
request with EBADMSG if it doesn't match the internally-computed
digest. GCM and CCM already assumed this, and the new AEAD mode
requires this for decryption. The new ETA mode now also requires
this for decryption, so IPsec and GELI no longer do their own
authentication verification. Simple DIGEST operations can also do
this, though there are no in-tree consumers.
To eventually support some refcounting to close races, the session
cookie is now passed to crypto_getop() and clients should no longer
set crp_sesssion directly.
- Assymteric crypto operation structures should be allocated via
crypto_getkreq() and freed via crypto_freekreq(). This permits the
crypto layer to track open asym requests and close races with a
driver trying to unregister while asym requests are in flight.
- crypto_copyback, crypto_copydata, crypto_apply, and
crypto_contiguous_subsegment now accept the 'crp' object as the
first parameter instead of individual members. This makes it easier
to deal with different buffer types in the future as well as
separate input and output buffers. It's also simpler for driver
writers to use.
- bus_dmamap_load_crp() loads a DMA mapping for a crypto buffer.
This understands the various types of buffers so that drivers that
use DMA do not have to be aware of different buffer types.
- Helper routines now exist to build an auth context for HMAC IPAD
and OPAD. This reduces some duplicated work among drivers.
- Key buffers are now treated as const throughout the framework and in
device drivers. However, session key buffers provided when a session
is created are expected to remain alive for the duration of the
session.
- GCM and CCM sessions now only specify a cipher algorithm and a cipher
key. The redundant auth information is not needed or used.
- For cryptosoft, split up the code a bit such that the 'process'
callback now invokes a function pointer in the session. This
function pointer is set based on the mode (in effect) though it
simplifies a few edge cases that would otherwise be in the switch in
'process'.
It does split up GCM vs CCM which I think is more readable even if there
is some duplication.
- I changed /dev/crypto to support GMAC requests using CRYPTO_AES_NIST_GMAC
as an auth algorithm and updated cryptocheck to work with it.
- Combined cipher and auth sessions via /dev/crypto now always use ETA
mode. The COP_F_CIPHER_FIRST flag is now a no-op that is ignored.
This was actually documented as being true in crypto(4) before, but
the code had not implemented this before I added the CIPHER_FIRST
flag.
- I have not yet updated /dev/crypto to be aware of explicit modes for
sessions. I will probably do that at some point in the future as well
as teach it about IV/nonce and tag lengths for AEAD so we can support
all of the NIST KAT tests for GCM and CCM.
- I've split up the exising crypto.9 manpage into several pages
of which many are written from scratch.
- I have converted all drivers and consumers in the tree and verified
that they compile, but I have not tested all of them. I have tested
the following drivers:
- cryptosoft
- aesni (AES only)
- blake2
- ccr
and the following consumers:
- cryptodev
- IPsec
- ktls_ocf
- GELI (lightly)
I have not tested the following:
- ccp
- aesni with sha
- hifn
- kgssapi_krb5
- ubsec
- padlock
- safe
- armv8_crypto (aarch64)
- glxsb (i386)
- sec (ppc)
- cesa (armv7)
- cryptocteon (mips64)
- nlmsec (mips64)
Discussed with: cem
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D23677
2020-03-27 14:25:23 -04:00
|
|
|
DEVMETHOD(cryptodev_probesession, ccr_probesession),
|
Add a driver for the Chelsio T6 crypto accelerator engine.
The ccr(4) driver supports use of the crypto accelerator engine on
Chelsio T6 NICs in "lookaside" mode via the opencrypto framework.
Currently, the driver supports AES-CBC, AES-CTR, AES-GCM, and AES-XTS
cipher algorithms as well as the SHA1-HMAC, SHA2-256-HMAC, SHA2-384-HMAC,
and SHA2-512-HMAC authentication algorithms. The driver also supports
chaining one of AES-CBC, AES-CTR, or AES-XTS with an authentication
algorithm for encrypt-then-authenticate operations.
Note that this driver is still under active development and testing and
may not yet be ready for production use. It does pass the tests in
tests/sys/opencrypto with the exception that the AES-GCM implementation
in the driver does not yet support requests with a zero byte payload.
To use this driver currently, the "uwire" configuration must be used
along with explicitly enabling support for lookaside crypto capabilities
in the cxgbe(4) driver. These can be done by setting the following
tunables before loading the cxgbe(4) driver:
hw.cxgbe.config_file=uwire
hw.cxgbe.cryptocaps_allowed=-1
MFC after: 1 month
Relnotes: yes
Sponsored by: Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D10763
2017-05-17 18:13:07 -04:00
|
|
|
DEVMETHOD(cryptodev_newsession, ccr_newsession),
|
|
|
|
|
DEVMETHOD(cryptodev_freesession, ccr_freesession),
|
|
|
|
|
DEVMETHOD(cryptodev_process, ccr_process),
|
|
|
|
|
|
|
|
|
|
DEVMETHOD_END
|
|
|
|
|
};
|
|
|
|
|
|
|
|
|
|
static driver_t ccr_driver = {
|
|
|
|
|
"ccr",
|
|
|
|
|
ccr_methods,
|
|
|
|
|
sizeof(struct ccr_softc)
|
|
|
|
|
};
|
|
|
|
|
|
|
|
|
|
static devclass_t ccr_devclass;
|
|
|
|
|
|
|
|
|
|
DRIVER_MODULE(ccr, t6nex, ccr_driver, ccr_devclass, ccr_modevent, NULL);
|
|
|
|
|
MODULE_VERSION(ccr, 1);
|
|
|
|
|
MODULE_DEPEND(ccr, crypto, 1, 1, 1);
|
|
|
|
|
MODULE_DEPEND(ccr, t6nex, 1, 1, 1);
|