2 * Background jobs (long-running operations)
4 * Copyright (c) 2011 IBM Corp.
5 * Copyright (c) 2012, 2018 Red Hat, Inc.
7 * Permission is hereby granted, free of charge, to any person obtaining a copy
8 * of this software and associated documentation files (the "Software"), to deal
9 * in the Software without restriction, including without limitation the rights
10 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
11 * copies of the Software, and to permit persons to whom the Software is
12 * furnished to do so, subject to the following conditions:
14 * The above copyright notice and this permission notice shall be included in
15 * all copies or substantial portions of the Software.
17 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
18 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
19 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
20 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
21 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
22 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
26 #include "qemu/osdep.h"
27 #include "qemu-common.h"
28 #include "qapi/error.h"
31 #include "qemu/main-loop.h"
32 #include "trace-root.h"
34 static QLIST_HEAD(, Job
) jobs
= QLIST_HEAD_INITIALIZER(jobs
);
36 /* Job State Transition Table */
37 bool JobSTT
[JOB_STATUS__MAX
][JOB_STATUS__MAX
] = {
38 /* U, C, R, P, Y, S, W, D, X, E, N */
39 /* U: */ [JOB_STATUS_UNDEFINED
] = {0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0},
40 /* C: */ [JOB_STATUS_CREATED
] = {0, 0, 1, 0, 0, 0, 0, 0, 1, 0, 1},
41 /* R: */ [JOB_STATUS_RUNNING
] = {0, 0, 0, 1, 1, 0, 1, 0, 1, 0, 0},
42 /* P: */ [JOB_STATUS_PAUSED
] = {0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0},
43 /* Y: */ [JOB_STATUS_READY
] = {0, 0, 0, 0, 0, 1, 1, 0, 1, 0, 0},
44 /* S: */ [JOB_STATUS_STANDBY
] = {0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0},
45 /* W: */ [JOB_STATUS_WAITING
] = {0, 0, 0, 0, 0, 0, 0, 1, 1, 0, 0},
46 /* D: */ [JOB_STATUS_PENDING
] = {0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 0},
47 /* X: */ [JOB_STATUS_ABORTING
] = {0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 0},
48 /* E: */ [JOB_STATUS_CONCLUDED
] = {0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1},
49 /* N: */ [JOB_STATUS_NULL
] = {0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0},
52 bool JobVerbTable
[JOB_VERB__MAX
][JOB_STATUS__MAX
] = {
53 /* U, C, R, P, Y, S, W, D, X, E, N */
54 [JOB_VERB_CANCEL
] = {0, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0},
55 [JOB_VERB_PAUSE
] = {0, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0},
56 [JOB_VERB_RESUME
] = {0, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0},
57 [JOB_VERB_SET_SPEED
] = {0, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0},
58 [JOB_VERB_COMPLETE
] = {0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0},
59 [JOB_VERB_FINALIZE
] = {0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0},
60 [JOB_VERB_DISMISS
] = {0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0},
63 /* Transactional group of jobs */
66 /* Is this txn being cancelled? */
70 QLIST_HEAD(, Job
) jobs
;
76 /* Right now, this mutex is only needed to synchronize accesses to job->busy
77 * and job->sleep_timer, such as concurrent calls to job_do_yield and
79 static QemuMutex job_mutex
;
81 static void job_lock(void)
83 qemu_mutex_lock(&job_mutex
);
86 static void job_unlock(void)
88 qemu_mutex_unlock(&job_mutex
);
91 static void __attribute__((__constructor__
)) job_init(void)
93 qemu_mutex_init(&job_mutex
);
96 JobTxn
*job_txn_new(void)
98 JobTxn
*txn
= g_new0(JobTxn
, 1);
99 QLIST_INIT(&txn
->jobs
);
104 static void job_txn_ref(JobTxn
*txn
)
109 void job_txn_unref(JobTxn
*txn
)
111 if (txn
&& --txn
->refcnt
== 0) {
116 void job_txn_add_job(JobTxn
*txn
, Job
*job
)
125 QLIST_INSERT_HEAD(&txn
->jobs
, job
, txn_list
);
129 static void job_txn_del_job(Job
*job
)
132 QLIST_REMOVE(job
, txn_list
);
133 job_txn_unref(job
->txn
);
138 static int job_txn_apply(JobTxn
*txn
, int fn(Job
*), bool lock
)
144 QLIST_FOREACH_SAFE(job
, &txn
->jobs
, txn_list
, next
) {
146 ctx
= job
->aio_context
;
147 aio_context_acquire(ctx
);
151 aio_context_release(ctx
);
161 /* TODO Make static once the whole state machine is in job.c */
162 void job_state_transition(Job
*job
, JobStatus s1
)
164 JobStatus s0
= job
->status
;
165 assert(s1
>= 0 && s1
<= JOB_STATUS__MAX
);
166 trace_job_state_transition(job
, job
->ret
,
167 JobSTT
[s0
][s1
] ? "allowed" : "disallowed",
168 JobStatus_str(s0
), JobStatus_str(s1
));
169 assert(JobSTT
[s0
][s1
]);
173 int job_apply_verb(Job
*job
, JobVerb verb
, Error
**errp
)
175 JobStatus s0
= job
->status
;
176 assert(verb
>= 0 && verb
<= JOB_VERB__MAX
);
177 trace_job_apply_verb(job
, JobStatus_str(s0
), JobVerb_str(verb
),
178 JobVerbTable
[verb
][s0
] ? "allowed" : "prohibited");
179 if (JobVerbTable
[verb
][s0
]) {
182 error_setg(errp
, "Job '%s' in state '%s' cannot accept command verb '%s'",
183 job
->id
, JobStatus_str(s0
), JobVerb_str(verb
));
187 JobType
job_type(const Job
*job
)
189 return job
->driver
->job_type
;
192 const char *job_type_str(const Job
*job
)
194 return JobType_str(job_type(job
));
197 bool job_is_cancelled(Job
*job
)
199 return job
->cancelled
;
202 bool job_is_completed(Job
*job
)
204 switch (job
->status
) {
205 case JOB_STATUS_UNDEFINED
:
206 case JOB_STATUS_CREATED
:
207 case JOB_STATUS_RUNNING
:
208 case JOB_STATUS_PAUSED
:
209 case JOB_STATUS_READY
:
210 case JOB_STATUS_STANDBY
:
212 case JOB_STATUS_WAITING
:
213 case JOB_STATUS_PENDING
:
214 case JOB_STATUS_ABORTING
:
215 case JOB_STATUS_CONCLUDED
:
216 case JOB_STATUS_NULL
:
219 g_assert_not_reached();
224 static bool job_started(Job
*job
)
229 static bool job_should_pause(Job
*job
)
231 return job
->pause_count
> 0;
234 Job
*job_next(Job
*job
)
237 return QLIST_FIRST(&jobs
);
239 return QLIST_NEXT(job
, job_list
);
242 Job
*job_get(const char *id
)
246 QLIST_FOREACH(job
, &jobs
, job_list
) {
247 if (job
->id
&& !strcmp(id
, job
->id
)) {
255 static void job_sleep_timer_cb(void *opaque
)
262 void *job_create(const char *job_id
, const JobDriver
*driver
, JobTxn
*txn
,
263 AioContext
*ctx
, int flags
, BlockCompletionFunc
*cb
,
264 void *opaque
, Error
**errp
)
269 if (flags
& JOB_INTERNAL
) {
270 error_setg(errp
, "Cannot specify job ID for internal job");
273 if (!id_wellformed(job_id
)) {
274 error_setg(errp
, "Invalid job ID '%s'", job_id
);
277 if (job_get(job_id
)) {
278 error_setg(errp
, "Job ID '%s' already in use", job_id
);
281 } else if (!(flags
& JOB_INTERNAL
)) {
282 error_setg(errp
, "An explicit job ID is required");
286 job
= g_malloc0(driver
->instance_size
);
287 job
->driver
= driver
;
288 job
->id
= g_strdup(job_id
);
290 job
->aio_context
= ctx
;
293 job
->pause_count
= 1;
294 job
->auto_finalize
= !(flags
& JOB_MANUAL_FINALIZE
);
295 job
->auto_dismiss
= !(flags
& JOB_MANUAL_DISMISS
);
297 job
->opaque
= opaque
;
299 notifier_list_init(&job
->on_finalize_cancelled
);
300 notifier_list_init(&job
->on_finalize_completed
);
301 notifier_list_init(&job
->on_pending
);
303 job_state_transition(job
, JOB_STATUS_CREATED
);
304 aio_timer_init(qemu_get_aio_context(), &job
->sleep_timer
,
305 QEMU_CLOCK_REALTIME
, SCALE_NS
,
306 job_sleep_timer_cb
, job
);
308 QLIST_INSERT_HEAD(&jobs
, job
, job_list
);
310 /* Single jobs are modeled as single-job transactions for sake of
311 * consolidating the job management logic */
314 job_txn_add_job(txn
, job
);
317 job_txn_add_job(txn
, job
);
323 void job_ref(Job
*job
)
328 void job_unref(Job
*job
)
330 if (--job
->refcnt
== 0) {
331 assert(job
->status
== JOB_STATUS_NULL
);
332 assert(!timer_pending(&job
->sleep_timer
));
335 if (job
->driver
->free
) {
336 job
->driver
->free(job
);
339 QLIST_REMOVE(job
, job_list
);
346 void job_event_cancelled(Job
*job
)
348 notifier_list_notify(&job
->on_finalize_cancelled
, job
);
351 void job_event_completed(Job
*job
)
353 notifier_list_notify(&job
->on_finalize_completed
, job
);
356 static void job_event_pending(Job
*job
)
358 notifier_list_notify(&job
->on_pending
, job
);
361 void job_enter_cond(Job
*job
, bool(*fn
)(Job
*job
))
363 if (!job_started(job
)) {
366 if (job
->deferred_to_main_loop
) {
376 if (fn
&& !fn(job
)) {
381 assert(!job
->deferred_to_main_loop
);
382 timer_del(&job
->sleep_timer
);
385 aio_co_wake(job
->co
);
388 void job_enter(Job
*job
)
390 job_enter_cond(job
, NULL
);
393 /* Yield, and schedule a timer to reenter the coroutine after @ns nanoseconds.
394 * Reentering the job coroutine with job_enter() before the timer has expired
395 * is allowed and cancels the timer.
397 * If @ns is (uint64_t) -1, no timer is scheduled and job_enter() must be
398 * called explicitly. */
399 static void coroutine_fn
job_do_yield(Job
*job
, uint64_t ns
)
403 timer_mod(&job
->sleep_timer
, ns
);
407 qemu_coroutine_yield();
409 /* Set by job_enter_cond() before re-entering the coroutine. */
413 void coroutine_fn
job_pause_point(Job
*job
)
415 assert(job
&& job_started(job
));
417 if (!job_should_pause(job
)) {
420 if (job_is_cancelled(job
)) {
424 if (job
->driver
->pause
) {
425 job
->driver
->pause(job
);
428 if (job_should_pause(job
) && !job_is_cancelled(job
)) {
429 JobStatus status
= job
->status
;
430 job_state_transition(job
, status
== JOB_STATUS_READY
432 : JOB_STATUS_PAUSED
);
434 job_do_yield(job
, -1);
436 job_state_transition(job
, status
);
439 if (job
->driver
->resume
) {
440 job
->driver
->resume(job
);
444 void job_yield(Job
*job
)
448 /* Check cancellation *before* setting busy = false, too! */
449 if (job_is_cancelled(job
)) {
453 if (!job_should_pause(job
)) {
454 job_do_yield(job
, -1);
457 job_pause_point(job
);
460 void coroutine_fn
job_sleep_ns(Job
*job
, int64_t ns
)
464 /* Check cancellation *before* setting busy = false, too! */
465 if (job_is_cancelled(job
)) {
469 if (!job_should_pause(job
)) {
470 job_do_yield(job
, qemu_clock_get_ns(QEMU_CLOCK_REALTIME
) + ns
);
473 job_pause_point(job
);
476 void job_drain(Job
*job
)
478 /* If job is !busy this kicks it into the next pause point. */
481 if (job
->driver
->drain
) {
482 job
->driver
->drain(job
);
488 * All jobs must allow a pause point before entering their job proper. This
489 * ensures that jobs can be paused prior to being started, then resumed later.
491 static void coroutine_fn
job_co_entry(void *opaque
)
495 assert(job
&& job
->driver
&& job
->driver
->start
);
496 job_pause_point(job
);
497 job
->driver
->start(job
);
501 void job_start(Job
*job
)
503 assert(job
&& !job_started(job
) && job
->paused
&&
504 job
->driver
&& job
->driver
->start
);
505 job
->co
= qemu_coroutine_create(job_co_entry
, job
);
509 job_state_transition(job
, JOB_STATUS_RUNNING
);
510 aio_co_enter(job
->aio_context
, job
->co
);
513 /* Assumes the block_job_mutex is held */
514 static bool job_timer_not_pending(Job
*job
)
516 return !timer_pending(&job
->sleep_timer
);
519 void job_pause(Job
*job
)
524 void job_resume(Job
*job
)
526 assert(job
->pause_count
> 0);
528 if (job
->pause_count
) {
532 /* kick only if no timer is pending */
533 job_enter_cond(job
, job_timer_not_pending
);
536 void job_user_pause(Job
*job
, Error
**errp
)
538 if (job_apply_verb(job
, JOB_VERB_PAUSE
, errp
)) {
541 if (job
->user_paused
) {
542 error_setg(errp
, "Job is already paused");
545 job
->user_paused
= true;
549 bool job_user_paused(Job
*job
)
551 return job
->user_paused
;
554 void job_user_resume(Job
*job
, Error
**errp
)
557 if (!job
->user_paused
|| job
->pause_count
<= 0) {
558 error_setg(errp
, "Can't resume a job that was not paused");
561 if (job_apply_verb(job
, JOB_VERB_RESUME
, errp
)) {
564 if (job
->driver
->user_resume
) {
565 job
->driver
->user_resume(job
);
567 job
->user_paused
= false;
571 void job_do_dismiss(Job
*job
)
576 job
->deferred_to_main_loop
= true;
578 job_txn_del_job(job
);
580 job_state_transition(job
, JOB_STATUS_NULL
);
584 void job_early_fail(Job
*job
)
586 assert(job
->status
== JOB_STATUS_CREATED
);
590 static void job_conclude(Job
*job
)
592 job_state_transition(job
, JOB_STATUS_CONCLUDED
);
593 if (job
->auto_dismiss
|| !job_started(job
)) {
598 static void job_update_rc(Job
*job
)
600 if (!job
->ret
&& job_is_cancelled(job
)) {
601 job
->ret
= -ECANCELED
;
604 job_state_transition(job
, JOB_STATUS_ABORTING
);
608 static void job_commit(Job
*job
)
611 if (job
->driver
->commit
) {
612 job
->driver
->commit(job
);
616 static void job_abort(Job
*job
)
619 if (job
->driver
->abort
) {
620 job
->driver
->abort(job
);
624 static void job_clean(Job
*job
)
626 if (job
->driver
->clean
) {
627 job
->driver
->clean(job
);
631 static int job_finalize_single(Job
*job
)
633 assert(job_is_completed(job
));
635 /* Ensure abort is called for late-transactional failures */
646 job
->cb(job
->opaque
, job
->ret
);
649 /* Emit events only if we actually started */
650 if (job_started(job
)) {
651 if (job_is_cancelled(job
)) {
652 job_event_cancelled(job
);
654 job_event_completed(job
);
658 job_txn_del_job(job
);
663 static void job_cancel_async(Job
*job
, bool force
)
665 if (job
->user_paused
) {
666 /* Do not call job_enter here, the caller will handle it. */
667 job
->user_paused
= false;
668 if (job
->driver
->user_resume
) {
669 job
->driver
->user_resume(job
);
671 assert(job
->pause_count
> 0);
674 job
->cancelled
= true;
675 /* To prevent 'force == false' overriding a previous 'force == true' */
676 job
->force_cancel
|= force
;
679 static void job_completed_txn_abort(Job
*job
)
682 JobTxn
*txn
= job
->txn
;
687 * We are cancelled by another job, which will handle everything.
691 txn
->aborting
= true;
694 /* We are the first failed job. Cancel other jobs. */
695 QLIST_FOREACH(other_job
, &txn
->jobs
, txn_list
) {
696 ctx
= other_job
->aio_context
;
697 aio_context_acquire(ctx
);
700 /* Other jobs are effectively cancelled by us, set the status for
701 * them; this job, however, may or may not be cancelled, depending
702 * on the caller, so leave it. */
703 QLIST_FOREACH(other_job
, &txn
->jobs
, txn_list
) {
704 if (other_job
!= job
) {
705 job_cancel_async(other_job
, false);
708 while (!QLIST_EMPTY(&txn
->jobs
)) {
709 other_job
= QLIST_FIRST(&txn
->jobs
);
710 ctx
= other_job
->aio_context
;
711 if (!job_is_completed(other_job
)) {
712 assert(job_is_cancelled(other_job
));
713 job_finish_sync(other_job
, NULL
, NULL
);
715 job_finalize_single(other_job
);
716 aio_context_release(ctx
);
722 static int job_prepare(Job
*job
)
724 if (job
->ret
== 0 && job
->driver
->prepare
) {
725 job
->ret
= job
->driver
->prepare(job
);
730 static int job_needs_finalize(Job
*job
)
732 return !job
->auto_finalize
;
735 static void job_do_finalize(Job
*job
)
738 assert(job
&& job
->txn
);
740 /* prepare the transaction to complete */
741 rc
= job_txn_apply(job
->txn
, job_prepare
, true);
743 job_completed_txn_abort(job
);
745 job_txn_apply(job
->txn
, job_finalize_single
, true);
749 void job_finalize(Job
*job
, Error
**errp
)
751 assert(job
&& job
->id
);
752 if (job_apply_verb(job
, JOB_VERB_FINALIZE
, errp
)) {
755 job_do_finalize(job
);
758 static int job_transition_to_pending(Job
*job
)
760 job_state_transition(job
, JOB_STATUS_PENDING
);
761 if (!job
->auto_finalize
) {
762 job_event_pending(job
);
767 static void job_completed_txn_success(Job
*job
)
769 JobTxn
*txn
= job
->txn
;
772 job_state_transition(job
, JOB_STATUS_WAITING
);
775 * Successful completion, see if there are other running jobs in this
778 QLIST_FOREACH(other_job
, &txn
->jobs
, txn_list
) {
779 if (!job_is_completed(other_job
)) {
782 assert(other_job
->ret
== 0);
785 job_txn_apply(txn
, job_transition_to_pending
, false);
787 /* If no jobs need manual finalization, automatically do so */
788 if (job_txn_apply(txn
, job_needs_finalize
, false) == 0) {
789 job_do_finalize(job
);
793 void job_completed(Job
*job
, int ret
)
795 assert(job
&& job
->txn
&& !job_is_completed(job
));
798 trace_job_completed(job
, ret
, job
->ret
);
800 job_completed_txn_abort(job
);
802 job_completed_txn_success(job
);
806 void job_cancel(Job
*job
, bool force
)
808 if (job
->status
== JOB_STATUS_CONCLUDED
) {
812 job_cancel_async(job
, force
);
813 if (!job_started(job
)) {
814 job_completed(job
, -ECANCELED
);
815 } else if (job
->deferred_to_main_loop
) {
816 job_completed_txn_abort(job
);
822 void job_user_cancel(Job
*job
, bool force
, Error
**errp
)
824 if (job_apply_verb(job
, JOB_VERB_CANCEL
, errp
)) {
827 job_cancel(job
, force
);
830 /* A wrapper around job_cancel() taking an Error ** parameter so it may be
831 * used with job_finish_sync() without the need for (rather nasty) function
832 * pointer casts there. */
833 static void job_cancel_err(Job
*job
, Error
**errp
)
835 job_cancel(job
, false);
838 int job_cancel_sync(Job
*job
)
840 return job_finish_sync(job
, &job_cancel_err
, NULL
);
843 void job_cancel_sync_all(void)
846 AioContext
*aio_context
;
848 while ((job
= job_next(NULL
))) {
849 aio_context
= job
->aio_context
;
850 aio_context_acquire(aio_context
);
851 job_cancel_sync(job
);
852 aio_context_release(aio_context
);
856 int job_complete_sync(Job
*job
, Error
**errp
)
858 return job_finish_sync(job
, job_complete
, errp
);
861 void job_complete(Job
*job
, Error
**errp
)
863 /* Should not be reachable via external interface for internal jobs */
865 if (job_apply_verb(job
, JOB_VERB_COMPLETE
, errp
)) {
868 if (job
->pause_count
|| job_is_cancelled(job
) || !job
->driver
->complete
) {
869 error_setg(errp
, "The active block job '%s' cannot be completed",
874 job
->driver
->complete(job
, errp
);
880 JobDeferToMainLoopFn
*fn
;
882 } JobDeferToMainLoopData
;
884 static void job_defer_to_main_loop_bh(void *opaque
)
886 JobDeferToMainLoopData
*data
= opaque
;
887 Job
*job
= data
->job
;
888 AioContext
*aio_context
= job
->aio_context
;
890 aio_context_acquire(aio_context
);
891 data
->fn(data
->job
, data
->opaque
);
892 aio_context_release(aio_context
);
897 void job_defer_to_main_loop(Job
*job
, JobDeferToMainLoopFn
*fn
, void *opaque
)
899 JobDeferToMainLoopData
*data
= g_malloc(sizeof(*data
));
902 data
->opaque
= opaque
;
903 job
->deferred_to_main_loop
= true;
905 aio_bh_schedule_oneshot(qemu_get_aio_context(),
906 job_defer_to_main_loop_bh
, data
);
909 int job_finish_sync(Job
*job
, void (*finish
)(Job
*, Error
**errp
), Error
**errp
)
911 Error
*local_err
= NULL
;
917 finish(job
, &local_err
);
920 error_propagate(errp
, local_err
);
924 /* job_drain calls job_enter, and it should be enough to induce progress
925 * until the job completes or moves to the main thread. */
926 while (!job
->deferred_to_main_loop
&& !job_is_completed(job
)) {
929 while (!job_is_completed(job
)) {
930 aio_poll(qemu_get_aio_context(), true);
932 ret
= (job_is_cancelled(job
) && job
->ret
== 0) ? -ECANCELED
: job
->ret
;