bind9/lib/isc/quota.c
Tony Finch c377e0a9e3
Help thread sanitizer to cope with liburcu
All the places the qp-trie code was using `call_rcu()` needed
`__tsan_release()` and `__tsan_acquire()` annotations, so
add a couple of wrappers to encapsulate this pattern.

With these wrappers, the tests run almost clean under thread
sanitizer. The remaining problems are due to `rcu_barrier()`
which can be suppressed using `.tsan-suppress`. It does not
suppress the whole of `liburcu`, because we would like thread
sanitizer to detect problems in `call_rcu()` callbacks, which
are called from `liburcu`.

The CI jobs have been updated to use `.tsan-suppress` by
default, except for a special-case job that needs the
additional suppressions in `.tsan-suppress-extra`.

We might be able to get rid of some of this after liburcu gains
support for thread sanitizer.

Note: the `rcu_barrier()` suppression is not entirely effective:
tsan sometimes reports races that originate inside `rcu_barrier()`
but tsan has discarded the stack so it does not have the
information required to suppress the report. These "races" can
be made much easier to reproduce by adding `atexit_sleep_ms=1000`
to `TSAN_OPTIONS`. The problem with tsan's short memory can be
addressed by increasing `history_size`: when it is large enough
(6 or 7) the `rcu_barrier()` stack usually survives long enough
for suppression to work.
2023-05-12 20:48:31 +01:00

129 lines
3.3 KiB
C

/*
* Copyright (C) Internet Systems Consortium, Inc. ("ISC")
*
* SPDX-License-Identifier: MPL-2.0
*
* This Source Code Form is subject to the terms of the Mozilla Public
* License, v. 2.0. If a copy of the MPL was not distributed with this
* file, you can obtain one at https://mozilla.org/MPL/2.0/.
*
* See the COPYRIGHT file distributed with this work for additional
* information regarding copyright ownership.
*/
/*! \file */
#include <stddef.h>
#include <isc/atomic.h>
#include <isc/quota.h>
#include <isc/urcu.h>
#include <isc/util.h>
#define QUOTA_MAGIC ISC_MAGIC('Q', 'U', 'O', 'T')
#define VALID_QUOTA(p) ISC_MAGIC_VALID(p, QUOTA_MAGIC)
void
isc_quota_init(isc_quota_t *quota, unsigned int max) {
atomic_init(&quota->max, max);
atomic_init(&quota->used, 0);
atomic_init(&quota->soft, 0);
cds_wfcq_init(&quota->jobs.head, &quota->jobs.tail);
ISC_LINK_INIT(quota, link);
quota->magic = QUOTA_MAGIC;
}
void
isc_quota_soft(isc_quota_t *quota, unsigned int soft) {
REQUIRE(VALID_QUOTA(quota));
REQUIRE(atomic_load_relaxed(&quota->max) > soft);
atomic_store_relaxed(&quota->soft, soft);
}
void
isc_quota_max(isc_quota_t *quota, unsigned int max) {
REQUIRE(VALID_QUOTA(quota));
atomic_store_relaxed(&quota->max, max);
}
unsigned int
isc_quota_getmax(isc_quota_t *quota) {
REQUIRE(VALID_QUOTA(quota));
return (atomic_load_relaxed(&quota->max));
}
unsigned int
isc_quota_getsoft(isc_quota_t *quota) {
REQUIRE(VALID_QUOTA(quota));
return (atomic_load_relaxed(&quota->soft));
}
unsigned int
isc_quota_getused(isc_quota_t *quota) {
REQUIRE(VALID_QUOTA(quota));
return (atomic_load_relaxed(&quota->used));
}
void
isc_quota_release(isc_quota_t *quota) {
/*
* We are using the cds_wfcq_dequeue_blocking() variant here that
* has an internal mutex because we need synchronization on
* multiple dequeues running from different threads.
*/
struct cds_wfcq_node *node =
cds_wfcq_dequeue_blocking(&quota->jobs.head, &quota->jobs.tail);
if (node == NULL) {
uint_fast32_t used = atomic_fetch_sub_relaxed(&quota->used, 1);
INSIST(used > 0);
return;
}
isc_job_t *job = isc_urcu_container(node, isc_job_t, wfcq_node);
job->cb(job->cbarg);
}
isc_result_t
isc_quota_acquire_cb(isc_quota_t *quota, isc_job_t *job, isc_job_cb cb,
void *cbarg) {
REQUIRE(VALID_QUOTA(quota));
REQUIRE(job == NULL || cb != NULL);
uint_fast32_t used = atomic_fetch_add_relaxed(&quota->used, 1);
uint_fast32_t max = atomic_load_relaxed(&quota->max);
if (max != 0 && used >= max) {
(void)atomic_fetch_sub_relaxed(&quota->used, 1);
if (job != NULL) {
job->cb = cb;
job->cbarg = cbarg;
cds_wfcq_node_init(&job->wfcq_node);
/*
* The cds_wfcq_enqueue() is non-blocking (no internal
* mutex involved), so it offers a slight advantage.
*/
__tsan_release(job);
cds_wfcq_enqueue(&quota->jobs.head, &quota->jobs.tail,
&job->wfcq_node);
}
return (ISC_R_QUOTA);
}
uint_fast32_t soft = atomic_load_relaxed(&quota->soft);
if (soft != 0 && used >= soft) {
return (ISC_R_SOFTQUOTA);
}
return (ISC_R_SUCCESS);
}
void
isc_quota_destroy(isc_quota_t *quota) {
REQUIRE(VALID_QUOTA(quota));
quota->magic = 0;
INSIST(atomic_load(&quota->used) == 0);
INSIST(cds_wfcq_empty(&quota->jobs.head, &quota->jobs.tail));
cds_wfcq_destroy(&quota->jobs.head, &quota->jobs.tail);
}