opnsense-src/sys/dev/cxgbe/crypto/t4_crypto.c

2937 lines
83 KiB
C
Raw Normal View History

/*-
* 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
* +-------------------------------+
* | Hash state | ----- For hash-only requests
* +-------------------------------+ -
* | 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
* +-------------------------------+ -
* | 16 dummy bytes | ----- Only for HMAC/hash-only requests
* +-------------------------------+
* | 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.
*/
/*
* The crypto engine supports a maximum AAD size of 511 bytes.
*/
#define MAX_AAD_LEN 511
/*
* 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.
*/
#define MAX_RX_PHYS_DSGL_SGE 32
#define DSGL_SGE_MAXLEN 2048
/*
* 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
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;
char pads[CHCR_HASH_MAX_BLOCK_SIZE_128 * 2];
};
struct ccr_session_gmac {
int hash_len;
char ghash_h[GMAC_BLOCK_LEN];
};
struct ccr_session_ccm_mac {
int hash_len;
};
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];
};
struct ccr_port {
struct sge_wrq *txq;
struct sge_rxq *rxq;
int tx_channel_id;
u_int active_sessions;
};
struct ccr_session {
#ifdef INVARIANTS
int pending;
#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;
struct ccr_port *port;
union {
struct ccr_session_hmac hmac;
struct ccr_session_gmac gmac;
struct ccr_session_ccm_mac ccm_mac;
};
struct ccr_session_blkcipher blkcipher;
struct mtx lock;
/*
* Pre-allocate S/G lists used when preparing a work request.
* '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.
*/
struct sglist *sg_input;
struct sglist *sg_output;
struct sglist *sg_ulptx;
struct sglist *sg_dsgl;
};
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;
/*
* Pre-allocate a dummy output buffer for the IV and AAD for
* AEAD requests.
*/
char *iv_aad_buf;
struct sglist *sg_iv_aad;
/* Statistics. */
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;
};
/*
* 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.
*
* 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
* 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)
{
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;
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;
}
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
ccr_write_phys_dsgl(struct ccr_session *s, void *dst, int nsegs)
{
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;
sg = s->sg_dsgl;
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;
cpl->rss_hdr_int.qid = htobe16(s->port->rxq->iq.abs_id);
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
ccr_write_ulptx_sgl(struct ccr_session *s, void *dst, int nsegs)
{
struct ulptx_sgl *usgl;
struct sglist *sg;
struct sglist_seg *ss;
int i;
sg = s->sg_ulptx;
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
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)
{
u_int cctx_size, idata_len;
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));
crwr->wreq.session_id = 0;
crwr->wreq.rx_chid_to_rx_q_id = htobe32(
V_FW_CRYPTO_LOOKASIDE_WR_RX_CHID(s->port->tx_channel_id) |
V_FW_CRYPTO_LOOKASIDE_WR_LCB(0) |
V_FW_CRYPTO_LOOKASIDE_WR_PHASH(0) |
V_FW_CRYPTO_LOOKASIDE_WR_IV(IV_NOP) |
V_FW_CRYPTO_LOOKASIDE_WR_FQIDX(0) |
V_FW_CRYPTO_LOOKASIDE_WR_TX_CH(0) |
V_FW_CRYPTO_LOOKASIDE_WR_RX_Q_ID(s->port->rxq->iq.abs_id));
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) |
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));
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) |
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);
}
static int
ccr_hash(struct ccr_softc *sc, struct ccr_session *s, struct cryptop *crp)
{
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;
u_int hmac_ctrl, imm_len, iopad_size;
int error, sgl_nsegs, sgl_len, use_opad;
/* 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)
return (EFBIG);
axf = s->hmac.auth_hash;
if (s->mode == HMAC) {
use_opad = 1;
hmac_ctrl = SCMD_HMAC_CTRL_NO_TRUNC;
} else {
use_opad = 0;
hmac_ctrl = SCMD_HMAC_CTRL_NOP;
}
/* 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.
*/
kctx_len = iopad_size;
if (use_opad)
kctx_len += iopad_size;
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) {
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;
sgl_nsegs = 0;
sgl_len = 0;
} else {
imm_len = 0;
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);
if (error)
return (error);
sgl_nsegs = s->sg_ulptx->sg_nseg;
sgl_len = ccr_ulptx_sgl_len(sgl_nsegs);
}
wr_len = roundup2(transhdr_len, 16) + roundup2(imm_len, 16) + sgl_len;
if (wr_len > SGE_MAX_WR_LEN)
return (EFBIG);
wr = alloc_wrqe(wr_len, s->port->txq);
if (wr == NULL) {
counter_u64_add(sc->stats_wr_nomem, 1);
return (ENOMEM);
}
crwr = wrtod(wr);
memset(crwr, 0, wr_len);
ccr_populate_wreq(sc, s, crwr, kctx_len, wr_len, imm_len, sgl_len,
hash_size_in_response, crp);
crwr->sec_cpl.op_ivinsrtofst = htobe32(
V_CPL_TX_SEC_PDU_OPCODE(CPL_TX_SEC_PDU) |
V_CPL_TX_SEC_PDU_RXCHID(s->port->tx_channel_id) |
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);
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) |
V_SCMD_PROTO_VERSION(SCMD_PROTO_VERSION_GENERIC) |
V_SCMD_CIPH_MODE(SCMD_CIPH_MODE_NOP) |
V_SCMD_AUTH_MODE(s->hmac.auth_mode) |
V_SCMD_HMAC_CTRL(hmac_ctrl));
crwr->sec_cpl.ivgen_hdrlen = htobe32(
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));
memcpy(crwr->key_ctx.key, s->hmac.pads, kctx_len);
/* 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) |
V_KEY_CONTEXT_OPAD_PRESENT(use_opad) |
V_KEY_CONTEXT_SALT_PRESENT(1) |
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) {
dst[0] = 0x80;
if (s->mode == HMAC)
*(uint64_t *)(dst + axf->blocksize - sizeof(uint64_t)) =
htobe64(axf->blocksize << 3);
} 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);
else
ccr_write_ulptx_sgl(s, dst, sgl_nsegs);
/* XXX: TODO backpressure */
t4_wrq_tx(sc->adapter, wr);
return (0);
}
static int
ccr_hash_done(struct ccr_softc *sc, struct ccr_session *s, struct cryptop *crp,
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];
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);
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);
}
static int
ccr_blkcipher(struct ccr_softc *sc, struct ccr_session *s, struct cryptop *crp)
{
char iv[CHCR_MAX_CRYPTO_IV_LEN];
struct chcr_wr *crwr;
struct wrqe *wr;
char *dst;
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;
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)
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)
return (EINVAL);
/* 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)
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))
op_type = CHCR_ENCRYPT_OP;
else
op_type = CHCR_DECRYPT_OP;
sglist_reset(s->sg_dsgl);
if (CRYPTO_HAS_OUTPUT_BUFFER(crp))
error = sglist_append_sglist(s->sg_dsgl, s->sg_output,
crp->crp_payload_output_start, crp->crp_payload_length);
else
error = sglist_append_sglist(s->sg_dsgl, s->sg_input,
crp->crp_payload_start, crp->crp_payload_length);
if (error)
return (error);
dsgl_nsegs = ccr_count_sgl(s->sg_dsgl, DSGL_SGE_MAXLEN);
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;
sgl_nsegs = 0;
sgl_len = 0;
} else {
imm_len = 0;
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);
if (error)
return (error);
sgl_nsegs = s->sg_ulptx->sg_nseg;
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 +
roundup2(imm_len, 16) + sgl_len;
if (wr_len > SGE_MAX_WR_LEN)
return (EFBIG);
wr = alloc_wrqe(wr_len, s->port->txq);
if (wr == NULL) {
counter_u64_add(sc->stats_wr_nomem, 1);
return (ENOMEM);
}
crwr = wrtod(wr);
memset(crwr, 0, wr_len);
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);
ccr_populate_wreq(sc, s, crwr, kctx_len, wr_len, imm_len, sgl_len, 0,
crp);
crwr->sec_cpl.op_ivinsrtofst = htobe32(
V_CPL_TX_SEC_PDU_OPCODE(CPL_TX_SEC_PDU) |
V_CPL_TX_SEC_PDU_RXCHID(s->port->tx_channel_id) |
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);
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) |
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) |
V_SCMD_PROTO_VERSION(SCMD_PROTO_VERSION_GENERIC) |
V_SCMD_ENC_DEC_CTRL(op_type) |
V_SCMD_CIPH_MODE(s->blkcipher.cipher_mode) |
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) |
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))
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:
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:
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))
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;
ccr_write_phys_dsgl(s, dst, dsgl_nsegs);
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;
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);
else
ccr_write_ulptx_sgl(s, dst, sgl_nsegs);
/* XXX: TODO backpressure */
t4_wrq_tx(sc->adapter, wr);
explicit_bzero(iv, sizeof(iv));
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)
return (SCMD_HMAC_CTRL_TRUNC_RFC4366);
if (authsize == 12)
return (SCMD_HMAC_CTRL_IPSEC_96BIT);
if (authsize == hashsize / 2)
return (SCMD_HMAC_CTRL_DIV2);
return (SCMD_HMAC_CTRL_NO_TRUNC);
}
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)
{
char iv[CHCR_MAX_CRYPTO_IV_LEN];
struct chcr_wr *crwr;
struct wrqe *wr;
struct auth_hash *axf;
char *dst;
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;
u_int cipher_start, cipher_stop;
u_int hmac_ctrl, input_len;
int dsgl_nsegs, dsgl_len;
int sgl_nsegs, sgl_len;
int error;
/*
* 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)
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)
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)
return (EINVAL);
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))
op_type = CHCR_ENCRYPT_OP;
else
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.
*
* 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 + crp->crp_aad_length + crp->crp_payload_length +
hash_size_in_response > MAX_REQUEST_SIZE)
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)
return (EFBIG);
}
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);
if (error)
return (error);
if (CRYPTO_HAS_OUTPUT_BUFFER(crp))
error = sglist_append_sglist(s->sg_dsgl, s->sg_output,
crp->crp_payload_output_start, crp->crp_payload_length);
else
error = sglist_append_sglist(s->sg_dsgl, s->sg_input,
crp->crp_payload_start, crp->crp_payload_length);
if (error)
return (error);
if (op_type == CHCR_ENCRYPT_OP) {
if (CRYPTO_HAS_OUTPUT_BUFFER(crp))
error = sglist_append_sglist(s->sg_dsgl, s->sg_output,
crp->crp_digest_start, hash_size_in_response);
else
error = sglist_append_sglist(s->sg_dsgl, s->sg_input,
crp->crp_digest_start, hash_size_in_response);
if (error)
return (error);
}
dsgl_nsegs = ccr_count_sgl(s->sg_dsgl, DSGL_SGE_MAXLEN);
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.
*
* 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 = crp->crp_aad_length + crp->crp_payload_length;
/*
* 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);
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)) {
imm_len = input_len;
sgl_nsegs = 0;
sgl_len = 0;
} else {
imm_len = 0;
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) {
if (crp->crp_aad != NULL)
error = sglist_append(s->sg_ulptx,
crp->crp_aad, crp->crp_aad_length);
else
error = sglist_append_sglist(s->sg_ulptx,
s->sg_input, crp->crp_aad_start,
crp->crp_aad_length);
if (error)
return (error);
}
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);
if (error)
return (error);
if (op_type == CHCR_DECRYPT_OP) {
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);
if (error)
return (error);
}
sgl_nsegs = s->sg_ulptx->sg_nseg;
sgl_len = ccr_ulptx_sgl_len(sgl_nsegs);
}
/* 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;
} 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;
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;
if (wr_len > SGE_MAX_WR_LEN)
return (EFBIG);
wr = alloc_wrqe(wr_len, s->port->txq);
if (wr == NULL) {
counter_u64_add(sc->stats_wr_nomem, 1);
return (ENOMEM);
}
crwr = wrtod(wr);
memset(crwr, 0, wr_len);
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);
ccr_populate_wreq(sc, s, crwr, kctx_len, wr_len, imm_len, sgl_len,
op_type == CHCR_DECRYPT_OP ? hash_size_in_response : 0, crp);
crwr->sec_cpl.op_ivinsrtofst = htobe32(
V_CPL_TX_SEC_PDU_OPCODE(CPL_TX_SEC_PDU) |
V_CPL_TX_SEC_PDU_RXCHID(s->port->tx_channel_id) |
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);
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) |
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) |
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 ? 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) |
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;
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))
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:
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:
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))
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);
memcpy(dst, s->hmac.pads, iopad_size * 2);
dst = (char *)(crwr + 1) + kctx_len;
ccr_write_phys_dsgl(s, dst, dsgl_nsegs);
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;
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) {
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;
}
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;
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);
} else
ccr_write_ulptx_sgl(s, dst, sgl_nsegs);
/* XXX: TODO backpressure */
t4_wrq_tx(sc->adapter, wr);
explicit_bzero(iv, sizeof(iv));
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,
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)
{
char iv[CHCR_MAX_CRYPTO_IV_LEN];
struct chcr_wr *crwr;
struct wrqe *wr;
char *dst;
u_int iv_len, kctx_len, op_type, transhdr_len, wr_len;
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);
/*
* 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)
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)
return (EMSGSIZE);
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))
op_type = CHCR_ENCRYPT_OP;
else
op_type = CHCR_DECRYPT_OP;
/*
* 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.
*/
if (s->blkcipher.iv_len == 12)
iv_len = AES_BLOCK_LEN;
else
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);
/*
* 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 + crp->crp_aad_length + crp->crp_payload_length +
hash_size_in_response > MAX_REQUEST_SIZE)
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)
return (EFBIG);
}
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);
if (error)
return (error);
if (CRYPTO_HAS_OUTPUT_BUFFER(crp))
error = sglist_append_sglist(s->sg_dsgl, s->sg_output,
crp->crp_payload_output_start, crp->crp_payload_length);
else
error = sglist_append_sglist(s->sg_dsgl, s->sg_input,
crp->crp_payload_start, crp->crp_payload_length);
if (error)
return (error);
if (op_type == CHCR_ENCRYPT_OP) {
if (CRYPTO_HAS_OUTPUT_BUFFER(crp))
error = sglist_append_sglist(s->sg_dsgl, s->sg_output,
crp->crp_digest_start, hash_size_in_response);
else
error = sglist_append_sglist(s->sg_dsgl, s->sg_input,
crp->crp_digest_start, hash_size_in_response);
if (error)
return (error);
}
dsgl_nsegs = ccr_count_sgl(s->sg_dsgl, DSGL_SGE_MAXLEN);
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.
*
* 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 = crp->crp_aad_length + crp->crp_payload_length;
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 {
imm_len = 0;
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) {
if (crp->crp_aad != NULL)
error = sglist_append(s->sg_ulptx,
crp->crp_aad, crp->crp_aad_length);
else
error = sglist_append_sglist(s->sg_ulptx,
s->sg_input, crp->crp_aad_start,
crp->crp_aad_length);
if (error)
return (error);
}
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);
if (error)
return (error);
if (op_type == CHCR_DECRYPT_OP) {
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);
if (error)
return (error);
}
sgl_nsegs = s->sg_ulptx->sg_nseg;
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) {
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;
} 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;
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);
wr = alloc_wrqe(wr_len, s->port->txq);
if (wr == NULL) {
counter_u64_add(sc->stats_wr_nomem, 1);
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);
if (s->blkcipher.iv_len == 12)
*(uint32_t *)&iv[12] = htobe32(1);
ccr_populate_wreq(sc, s, crwr, kctx_len, wr_len, imm_len, sgl_len, 0,
crp);
crwr->sec_cpl.op_ivinsrtofst = htobe32(
V_CPL_TX_SEC_PDU_OPCODE(CPL_TX_SEC_PDU) |
V_CPL_TX_SEC_PDU_RXCHID(s->port->tx_channel_id) |
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.
*/
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) |
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 ? 1 : 0) |
V_SCMD_CIPH_MODE(SCMD_CIPH_MODE_AES_GCM) |
V_SCMD_AUTH_MODE(SCMD_AUTH_MODE_GHASH) |
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);
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;
ccr_write_phys_dsgl(s, dst, dsgl_nsegs);
dst += sizeof(struct cpl_rx_phys_dsgl) + dsgl_len;
memcpy(dst, iv, iv_len);
dst += iv_len;
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) {
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;
}
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;
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);
} else
ccr_write_ulptx_sgl(s, dst, sgl_nsegs);
/* XXX: TODO backpressure */
t4_wrq_tx(sc->adapter, wr);
explicit_bzero(iv, sizeof(iv));
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);
}
/*
* Handle a GCM 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_gcm_soft(struct ccr_session *s, struct cryptop *crp)
{
struct auth_hash *axf;
struct enc_xform *exf;
void *auth_ctx, *kschedule;
char block[GMAC_BLOCK_LEN];
char digest[GMAC_DIGEST_LEN];
char iv[AES_BLOCK_LEN];
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;
kschedule = malloc(exf->ctxsize, M_CCR, M_NOWAIT);
if (kschedule == NULL) {
error = ENOMEM;
goto out;
}
error = exf->setkey(kschedule, s->blkcipher.enckey,
s->blkcipher.key_len);
if (error)
goto out;
/*
* 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;
}
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);
*(uint32_t *)&iv[12] = htobe32(1);
axf->Reinit(auth_ctx, iv, sizeof(iv));
/* MAC the AAD. */
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));
}
}
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);
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)) {
exf->encrypt(kschedule, block, block);
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);
} else {
axf->Update(auth_ctx, block, len);
}
}
/* 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);
axf->Update(auth_ctx, block, sizeof(block));
/* 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);
error = 0;
} 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);
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);
bzero(block + len, sizeof(block) - len);
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);
}
} else
error = EBADMSG;
explicit_bzero(digest2, sizeof(digest2));
}
out:
zfree(kschedule, M_CCR);
zfree(auth_ctx, M_CCR);
explicit_bzero(block, sizeof(block));
explicit_bzero(iv, sizeof(iv));
explicit_bzero(digest, sizeof(digest));
crp->crp_etype = error;
crypto_done(crp);
}
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)
{
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;
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) {
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);
}
}
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)
{
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)
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)
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);
/*
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.
*/
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))
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)
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)
return (EFBIG);
}
sglist_reset(s->sg_dsgl);
error = sglist_append_sglist(s->sg_dsgl, sc->sg_iv_aad, 0, iv_len +
aad_len);
if (error)
return (error);
if (CRYPTO_HAS_OUTPUT_BUFFER(crp))
error = sglist_append_sglist(s->sg_dsgl, s->sg_output,
crp->crp_payload_output_start, crp->crp_payload_length);
else
error = sglist_append_sglist(s->sg_dsgl, s->sg_input,
crp->crp_payload_start, crp->crp_payload_length);
if (error)
return (error);
if (op_type == CHCR_ENCRYPT_OP) {
if (CRYPTO_HAS_OUTPUT_BUFFER(crp))
error = sglist_append_sglist(s->sg_dsgl, s->sg_output,
crp->crp_digest_start, hash_size_in_response);
else
error = sglist_append_sglist(s->sg_dsgl, s->sg_input,
crp->crp_digest_start, hash_size_in_response);
if (error)
return (error);
}
dsgl_nsegs = ccr_count_sgl(s->sg_dsgl, DSGL_SGE_MAXLEN);
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;
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;
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) {
if (crp->crp_aad != NULL)
error = sglist_append(s->sg_ulptx,
crp->crp_aad, crp->crp_aad_length);
else
error = sglist_append_sglist(s->sg_ulptx,
s->sg_input, crp->crp_aad_start,
crp->crp_aad_length);
if (error)
return (error);
}
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);
if (error)
return (error);
if (op_type == CHCR_DECRYPT_OP) {
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);
if (error)
return (error);
}
sgl_nsegs = s->sg_ulptx->sg_nseg;
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);
wr = alloc_wrqe(wr_len, s->port->txq);
if (wr == NULL) {
counter_u64_add(sc->stats_wr_nomem, 1);
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.
*/
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);
ccr_populate_wreq(sc, s, crwr, kctx_len, wr_len, imm_len, sgl_len, 0,
crp);
crwr->sec_cpl.op_ivinsrtofst = htobe32(
V_CPL_TX_SEC_PDU_OPCODE(CPL_TX_SEC_PDU) |
V_CPL_TX_SEC_PDU_RXCHID(s->port->tx_channel_id) |
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;
ccr_write_phys_dsgl(s, dst, dsgl_nsegs);
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);
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) {
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;
}
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;
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);
} 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);
}
ccr_write_ulptx_sgl(s, dst, sgl_nsegs);
}
/* XXX: TODO backpressure */
t4_wrq_tx(sc->adapter, wr);
explicit_bzero(iv, sizeof(iv));
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)
{
struct auth_hash *axf;
struct enc_xform *exf;
union authctx *auth_ctx;
void *kschedule;
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;
kschedule = malloc(exf->ctxsize, M_CCR, M_NOWAIT);
if (kschedule == NULL) {
error = ENOMEM;
goto out;
}
error = exf->setkey(kschedule, s->blkcipher.enckey,
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;
}
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);
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;
axf->Reinit(auth_ctx, iv, sizeof(iv));
/* MAC the AAD. */
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;
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);
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)) {
axf->Update(auth_ctx, block, len);
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);
} else {
exf->decrypt(kschedule, block, block);
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);
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];
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);
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);
bzero(block + len, sizeof(block) - len);
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);
}
} else
error = EBADMSG;
explicit_bzero(digest2, sizeof(digest2));
}
out:
zfree(kschedule, M_CCR);
zfree(auth_ctx, M_CCR);
explicit_bzero(block, sizeof(block));
explicit_bzero(iv, sizeof(iv));
explicit_bzero(digest, sizeof(digest));
crp->crp_etype = error;
crypto_done(crp);
}
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;
struct sysctl_oid *oid, *port_oid;
struct sysctl_oid_list *children;
char buf[16];
int i;
ctx = device_get_sysctl_ctx(sc->dev);
/*
* dev.ccr.X.
*/
oid = device_get_sysctl_tree(sc->dev);
children = SYSCTL_CHILDREN(oid);
SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "port_mask", CTLFLAG_RW,
&sc->port_mask, 0, "Mask of enabled ports");
/*
* dev.ccr.X.stats.
*/
oid = SYSCTL_ADD_NODE(ctx, children, OID_AUTO, "stats",
CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "statistics");
children = SYSCTL_CHILDREN(oid);
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,
"Cipher encryption requests submitted");
SYSCTL_ADD_COUNTER_U64(ctx, children, OID_AUTO, "cipher_decrypt",
CTLFLAG_RD, &sc->stats_blkcipher_decrypt,
"Cipher decryption requests submitted");
SYSCTL_ADD_COUNTER_U64(ctx, children, OID_AUTO, "eta_encrypt",
CTLFLAG_RD, &sc->stats_eta_encrypt,
"Combined AES+HMAC encryption requests submitted");
SYSCTL_ADD_COUNTER_U64(ctx, children, OID_AUTO, "eta_decrypt",
CTLFLAG_RD, &sc->stats_eta_decrypt,
"Combined AES+HMAC decryption requests submitted");
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,
"Requests for which DMA mapping failed");
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,
"Requests processed by falling back to software");
/*
* 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;
}
static int
ccr_attach(device_t dev)
{
struct ccr_softc *sc;
int32_t cid;
int i;
sc = device_get_softc(dev);
sc->dev = dev;
sc->adapter = device_get_softc(device_get_parent(dev));
for_each_port(sc->adapter, i) {
ccr_init_port(sc, i);
}
cid = crypto_get_driverid(dev, sizeof(struct ccr_session),
CRYPTOCAP_F_HARDWARE);
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);
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);
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);
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);
mtx_destroy(&sc->lock);
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);
sglist_free(sc->sg_iv_aad);
free(sc->iv_aad_buf, M_CCR);
sc->adapter->ccr_softc = NULL;
return (0);
}
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)
{
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);
}
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_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) {
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);
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);
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);
}
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);
}
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)
{
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;
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;
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;
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:
t4_aes_getdeckey(s->blkcipher.deckey, key, kbits);
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:
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;
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;
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) |
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));
}
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)
{
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;
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)
return (EINVAL);
break;
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)
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:
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 (csp->csp_cipher_klen != 0) {
cipher_mode = ccr_cipher_mode(csp);
if (cipher_mode == SCMD_CIPH_MODE_NOP)
return (EINVAL);
}
return (CRYPTODEV_PROBE_HARDWARE);
}
/*
* 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);
}
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
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);
}
sc = device_get_softc(dev);
mtx_lock(&sc->lock);
if (sc->detaching) {
mtx_unlock(&sc->lock);
ccr_delete_session(s);
return (ENXIO);
}
s->port = ccr_choose_port(sc);
if (s->port == NULL) {
mtx_unlock(&sc->lock);
ccr_delete_session(s);
return (ENXIO);
}
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)
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:
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;
}
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)
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,
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)
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) {
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)
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)
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);
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);
}
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) {
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);
}
s->port->active_sessions++;
mtx_unlock(&sc->lock);
return (0);
}
static void
ccr_freesession(device_t dev, crypto_session_t cses)
{
struct ccr_softc *sc;
struct ccr_session *s;
sc = device_get_softc(dev);
s = crypto_get_driver_session(cses);
#ifdef INVARIANTS
if (s->pending != 0)
device_printf(dev,
"session %p freed with %d pending requests\n", s,
s->pending);
#endif
mtx_lock(&sc->lock);
s->port->active_sessions--;
mtx_unlock(&sc->lock);
ccr_delete_session(s);
}
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;
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);
s = crypto_get_driver_session(crp->crp_session);
sc = device_get_softc(dev);
mtx_lock(&s->lock);
error = ccr_populate_sglist(s->sg_input, &crp->crp_buf);
if (error == 0 && CRYPTO_HAS_OUTPUT_BUFFER(crp))
error = ccr_populate_sglist(s->sg_output, &crp->crp_obuf);
if (error) {
counter_u64_add(sc->stats_sglist_error, 1);
goto out;
}
switch (s->mode) {
case HASH:
error = ccr_hash(sc, s, crp);
if (error == 0)
counter_u64_add(sc->stats_hash, 1);
break;
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)
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);
error = ccr_hash(sc, s, crp);
if (error == 0)
counter_u64_add(sc->stats_hmac, 1);
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);
error = ccr_blkcipher(sc, s, crp);
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))
counter_u64_add(sc->stats_blkcipher_encrypt, 1);
else
counter_u64_add(sc->stats_blkcipher_decrypt, 1);
}
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)
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);
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))
counter_u64_add(sc->stats_eta_encrypt, 1);
else
counter_u64_add(sc->stats_eta_decrypt, 1);
}
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);
}
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) {
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);
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);
if (error == EMSGSIZE) {
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);
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))
counter_u64_add(sc->stats_gcm_encrypt, 1);
else
counter_u64_add(sc->stats_gcm_decrypt, 1);
}
break;
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);
}
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);
if (error == EMSGSIZE) {
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);
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))
counter_u64_add(sc->stats_ccm_encrypt, 1);
else
counter_u64_add(sc->stats_ccm_decrypt, 1);
}
break;
}
if (error == 0) {
#ifdef INVARIANTS
s->pending++;
#endif
counter_u64_add(sc->stats_inflight, 1);
} else
counter_u64_add(sc->stats_process_error, 1);
out:
mtx_unlock(&s->lock);
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;
uint32_t status;
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]);
s = crypto_get_driver_session(crp->crp_session);
status = be64toh(cpl->data[0]);
if (CHK_MAC_ERR_BIT(status) || CHK_PAD_ERR_BIT(status))
error = EBADMSG;
else
error = 0;
#ifdef INVARIANTS
mtx_lock(&s->lock);
s->pending--;
mtx_unlock(&s->lock);
#endif
counter_u64_add(sc->stats_inflight, -1);
switch (s->mode) {
case HASH:
case HMAC:
error = ccr_hash_done(sc, s, crp, cpl, error);
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);
break;
case GCM:
error = ccr_gcm_done(sc, s, crp, cpl, error);
break;
case CCM:
error = ccr_ccm_done(sc, s, crp, cpl, error);
break;
}
if (error == EBADMSG) {
if (CHK_MAC_ERR_BIT(status))
counter_u64_add(sc->stats_mac_error, 1);
if (CHK_PAD_ERR_BIT(status))
counter_u64_add(sc->stats_pad_error, 1);
}
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),
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);