mirror_zfs/lib/libzpool/taskq.c
George.Wilson cc92e9d0c3 3246 ZFS I/O deadman thread
Reviewed by: Matt Ahrens <matthew.ahrens@delphix.com>
Reviewed by: Eric Schrock <eric.schrock@delphix.com>
Reviewed by: Christopher Siden <chris.siden@delphix.com>
Approved by: Garrett D'Amore <garrett@damore.org>

NOTES: This patch has been reworked from the original in the
following ways to accomidate Linux ZFS implementation

*) Usage of the cyclic interface was replaced by the delayed taskq
   interface.  This avoids the need to implement new compatibility
   code and allows us to rely on the existing taskq implementation.

*) An extern for zfs_txg_synctime_ms was added to sys/dsl_pool.h
   because declaring externs in source files as was done in the
   original patch is just plain wrong.

*) Instead of panicing the system when the deadman triggers a
   zevent describing the blocked vdev and the first pending I/O
   is posted.  If the panic behavior is desired Linux provides
   other generic methods to panic the system when threads are
   observed to hang.

*) For reference, to delay zios by 30 seconds for testing you can
   use zinject as follows: 'zinject -d <vdev> -D30 <pool>'

References:
  illumos/illumos-gate@283b84606b
  https://www.illumos.org/issues/3246

Ported-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes #1396
2013-05-01 17:05:52 -07:00

368 lines
8.4 KiB
C

/*
* CDDL HEADER START
*
* The contents of this file are subject to the terms of the
* Common Development and Distribution License (the "License").
* You may not use this file except in compliance with the License.
*
* You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
* or http://www.opensolaris.org/os/licensing.
* See the License for the specific language governing permissions
* and limitations under the License.
*
* When distributing Covered Code, include this CDDL HEADER in each
* file and include the License file at usr/src/OPENSOLARIS.LICENSE.
* If applicable, add the following below this CDDL HEADER, with the
* fields enclosed by brackets "[]" replaced with your own identifying
* information: Portions Copyright [yyyy] [name of copyright owner]
*
* CDDL HEADER END
*/
/*
* Copyright 2010 Sun Microsystems, Inc. All rights reserved.
* Use is subject to license terms.
*/
/*
* Copyright 2011 Nexenta Systems, Inc. All rights reserved.
* Copyright 2012 Garrett D'Amore <garrett@damore.org>. All rights reserved.
*/
#include <sys/zfs_context.h>
int taskq_now;
taskq_t *system_taskq;
#define TASKQ_ACTIVE 0x00010000
struct taskq {
kmutex_t tq_lock;
krwlock_t tq_threadlock;
kcondvar_t tq_dispatch_cv;
kcondvar_t tq_wait_cv;
kthread_t **tq_threadlist;
int tq_flags;
int tq_active;
int tq_nthreads;
int tq_nalloc;
int tq_minalloc;
int tq_maxalloc;
kcondvar_t tq_maxalloc_cv;
int tq_maxalloc_wait;
taskq_ent_t *tq_freelist;
taskq_ent_t tq_task;
};
static taskq_ent_t *
task_alloc(taskq_t *tq, int tqflags)
{
taskq_ent_t *t;
int rv;
again: if ((t = tq->tq_freelist) != NULL && tq->tq_nalloc >= tq->tq_minalloc) {
ASSERT(!(t->tqent_flags & TQENT_FLAG_PREALLOC));
tq->tq_freelist = t->tqent_next;
} else {
if (tq->tq_nalloc >= tq->tq_maxalloc) {
if (!(tqflags & KM_SLEEP))
return (NULL);
/*
* We don't want to exceed tq_maxalloc, but we can't
* wait for other tasks to complete (and thus free up
* task structures) without risking deadlock with
* the caller. So, we just delay for one second
* to throttle the allocation rate. If we have tasks
* complete before one second timeout expires then
* taskq_ent_free will signal us and we will
* immediately retry the allocation.
*/
tq->tq_maxalloc_wait++;
rv = cv_timedwait(&tq->tq_maxalloc_cv,
&tq->tq_lock, ddi_get_lbolt() + hz);
tq->tq_maxalloc_wait--;
if (rv > 0)
goto again; /* signaled */
}
mutex_exit(&tq->tq_lock);
t = kmem_alloc(sizeof (taskq_ent_t), tqflags);
mutex_enter(&tq->tq_lock);
if (t != NULL) {
/* Make sure we start without any flags */
t->tqent_flags = 0;
tq->tq_nalloc++;
}
}
return (t);
}
static void
task_free(taskq_t *tq, taskq_ent_t *t)
{
if (tq->tq_nalloc <= tq->tq_minalloc) {
t->tqent_next = tq->tq_freelist;
tq->tq_freelist = t;
} else {
tq->tq_nalloc--;
mutex_exit(&tq->tq_lock);
kmem_free(t, sizeof (taskq_ent_t));
mutex_enter(&tq->tq_lock);
}
if (tq->tq_maxalloc_wait)
cv_signal(&tq->tq_maxalloc_cv);
}
taskqid_t
taskq_dispatch(taskq_t *tq, task_func_t func, void *arg, uint_t tqflags)
{
taskq_ent_t *t;
if (taskq_now) {
func(arg);
return (1);
}
mutex_enter(&tq->tq_lock);
ASSERT(tq->tq_flags & TASKQ_ACTIVE);
if ((t = task_alloc(tq, tqflags)) == NULL) {
mutex_exit(&tq->tq_lock);
return (0);
}
if (tqflags & TQ_FRONT) {
t->tqent_next = tq->tq_task.tqent_next;
t->tqent_prev = &tq->tq_task;
} else {
t->tqent_next = &tq->tq_task;
t->tqent_prev = tq->tq_task.tqent_prev;
}
t->tqent_next->tqent_prev = t;
t->tqent_prev->tqent_next = t;
t->tqent_func = func;
t->tqent_arg = arg;
t->tqent_flags = 0;
cv_signal(&tq->tq_dispatch_cv);
mutex_exit(&tq->tq_lock);
return (1);
}
taskqid_t
taskq_dispatch_delay(taskq_t *tq, task_func_t func, void *arg, uint_t tqflags,
clock_t expire_time)
{
return (0);
}
int
taskq_empty_ent(taskq_ent_t *t)
{
return t->tqent_next == NULL;
}
void
taskq_init_ent(taskq_ent_t *t)
{
t->tqent_next = NULL;
t->tqent_prev = NULL;
t->tqent_func = NULL;
t->tqent_arg = NULL;
t->tqent_flags = 0;
}
void
taskq_dispatch_ent(taskq_t *tq, task_func_t func, void *arg, uint_t flags,
taskq_ent_t *t)
{
ASSERT(func != NULL);
ASSERT(!(tq->tq_flags & TASKQ_DYNAMIC));
/*
* Mark it as a prealloc'd task. This is important
* to ensure that we don't free it later.
*/
t->tqent_flags |= TQENT_FLAG_PREALLOC;
/*
* Enqueue the task to the underlying queue.
*/
mutex_enter(&tq->tq_lock);
if (flags & TQ_FRONT) {
t->tqent_next = tq->tq_task.tqent_next;
t->tqent_prev = &tq->tq_task;
} else {
t->tqent_next = &tq->tq_task;
t->tqent_prev = tq->tq_task.tqent_prev;
}
t->tqent_next->tqent_prev = t;
t->tqent_prev->tqent_next = t;
t->tqent_func = func;
t->tqent_arg = arg;
cv_signal(&tq->tq_dispatch_cv);
mutex_exit(&tq->tq_lock);
}
void
taskq_wait(taskq_t *tq)
{
mutex_enter(&tq->tq_lock);
while (tq->tq_task.tqent_next != &tq->tq_task || tq->tq_active != 0)
cv_wait(&tq->tq_wait_cv, &tq->tq_lock);
mutex_exit(&tq->tq_lock);
}
static void
taskq_thread(void *arg)
{
taskq_t *tq = arg;
taskq_ent_t *t;
boolean_t prealloc;
mutex_enter(&tq->tq_lock);
while (tq->tq_flags & TASKQ_ACTIVE) {
if ((t = tq->tq_task.tqent_next) == &tq->tq_task) {
if (--tq->tq_active == 0)
cv_broadcast(&tq->tq_wait_cv);
cv_wait(&tq->tq_dispatch_cv, &tq->tq_lock);
tq->tq_active++;
continue;
}
t->tqent_prev->tqent_next = t->tqent_next;
t->tqent_next->tqent_prev = t->tqent_prev;
t->tqent_next = NULL;
t->tqent_prev = NULL;
prealloc = t->tqent_flags & TQENT_FLAG_PREALLOC;
mutex_exit(&tq->tq_lock);
rw_enter(&tq->tq_threadlock, RW_READER);
t->tqent_func(t->tqent_arg);
rw_exit(&tq->tq_threadlock);
mutex_enter(&tq->tq_lock);
if (!prealloc)
task_free(tq, t);
}
tq->tq_nthreads--;
cv_broadcast(&tq->tq_wait_cv);
mutex_exit(&tq->tq_lock);
thread_exit();
}
/*ARGSUSED*/
taskq_t *
taskq_create(const char *name, int nthreads, pri_t pri,
int minalloc, int maxalloc, uint_t flags)
{
taskq_t *tq = kmem_zalloc(sizeof (taskq_t), KM_SLEEP);
int t;
if (flags & TASKQ_THREADS_CPU_PCT) {
int pct;
ASSERT3S(nthreads, >=, 0);
ASSERT3S(nthreads, <=, 100);
pct = MIN(nthreads, 100);
pct = MAX(pct, 0);
nthreads = (sysconf(_SC_NPROCESSORS_ONLN) * pct) / 100;
nthreads = MAX(nthreads, 1); /* need at least 1 thread */
} else {
ASSERT3S(nthreads, >=, 1);
}
rw_init(&tq->tq_threadlock, NULL, RW_DEFAULT, NULL);
mutex_init(&tq->tq_lock, NULL, MUTEX_DEFAULT, NULL);
cv_init(&tq->tq_dispatch_cv, NULL, CV_DEFAULT, NULL);
cv_init(&tq->tq_wait_cv, NULL, CV_DEFAULT, NULL);
cv_init(&tq->tq_maxalloc_cv, NULL, CV_DEFAULT, NULL);
tq->tq_flags = flags | TASKQ_ACTIVE;
tq->tq_active = nthreads;
tq->tq_nthreads = nthreads;
tq->tq_minalloc = minalloc;
tq->tq_maxalloc = maxalloc;
tq->tq_task.tqent_next = &tq->tq_task;
tq->tq_task.tqent_prev = &tq->tq_task;
tq->tq_threadlist = kmem_alloc(nthreads*sizeof(kthread_t *), KM_SLEEP);
if (flags & TASKQ_PREPOPULATE) {
mutex_enter(&tq->tq_lock);
while (minalloc-- > 0)
task_free(tq, task_alloc(tq, KM_SLEEP));
mutex_exit(&tq->tq_lock);
}
for (t = 0; t < nthreads; t++)
VERIFY((tq->tq_threadlist[t] = thread_create(NULL, 0,
taskq_thread, tq, TS_RUN, NULL, 0, 0)) != NULL);
return (tq);
}
void
taskq_destroy(taskq_t *tq)
{
int nthreads = tq->tq_nthreads;
taskq_wait(tq);
mutex_enter(&tq->tq_lock);
tq->tq_flags &= ~TASKQ_ACTIVE;
cv_broadcast(&tq->tq_dispatch_cv);
while (tq->tq_nthreads != 0)
cv_wait(&tq->tq_wait_cv, &tq->tq_lock);
tq->tq_minalloc = 0;
while (tq->tq_nalloc != 0) {
ASSERT(tq->tq_freelist != NULL);
task_free(tq, task_alloc(tq, KM_SLEEP));
}
mutex_exit(&tq->tq_lock);
kmem_free(tq->tq_threadlist, nthreads * sizeof (kthread_t *));
rw_destroy(&tq->tq_threadlock);
mutex_destroy(&tq->tq_lock);
cv_destroy(&tq->tq_dispatch_cv);
cv_destroy(&tq->tq_wait_cv);
cv_destroy(&tq->tq_maxalloc_cv);
kmem_free(tq, sizeof (taskq_t));
}
int
taskq_member(taskq_t *tq, kthread_t *t)
{
int i;
if (taskq_now)
return (1);
for (i = 0; i < tq->tq_nthreads; i++)
if (tq->tq_threadlist[i] == t)
return (1);
return (0);
}
int
taskq_cancel_id(taskq_t *tq, taskqid_t id)
{
return (ENOENT);
}
void
system_taskq_init(void)
{
system_taskq = taskq_create("system_taskq", 64, minclsyspri, 4, 512,
TASKQ_DYNAMIC | TASKQ_PREPOPULATE);
}
void
system_taskq_fini(void)
{
taskq_destroy(system_taskq);
system_taskq = NULL; /* defensive */
}