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_ready(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_WAITING
:
210 case JOB_STATUS_PENDING
:
211 case JOB_STATUS_ABORTING
:
212 case JOB_STATUS_CONCLUDED
:
213 case JOB_STATUS_NULL
:
215 case JOB_STATUS_READY
:
216 case JOB_STATUS_STANDBY
:
219 g_assert_not_reached();
224 bool job_is_completed(Job
*job
)
226 switch (job
->status
) {
227 case JOB_STATUS_UNDEFINED
:
228 case JOB_STATUS_CREATED
:
229 case JOB_STATUS_RUNNING
:
230 case JOB_STATUS_PAUSED
:
231 case JOB_STATUS_READY
:
232 case JOB_STATUS_STANDBY
:
234 case JOB_STATUS_WAITING
:
235 case JOB_STATUS_PENDING
:
236 case JOB_STATUS_ABORTING
:
237 case JOB_STATUS_CONCLUDED
:
238 case JOB_STATUS_NULL
:
241 g_assert_not_reached();
246 static bool job_started(Job
*job
)
251 static bool job_should_pause(Job
*job
)
253 return job
->pause_count
> 0;
256 Job
*job_next(Job
*job
)
259 return QLIST_FIRST(&jobs
);
261 return QLIST_NEXT(job
, job_list
);
264 Job
*job_get(const char *id
)
268 QLIST_FOREACH(job
, &jobs
, job_list
) {
269 if (job
->id
&& !strcmp(id
, job
->id
)) {
277 static void job_sleep_timer_cb(void *opaque
)
284 void *job_create(const char *job_id
, const JobDriver
*driver
, JobTxn
*txn
,
285 AioContext
*ctx
, int flags
, BlockCompletionFunc
*cb
,
286 void *opaque
, Error
**errp
)
291 if (flags
& JOB_INTERNAL
) {
292 error_setg(errp
, "Cannot specify job ID for internal job");
295 if (!id_wellformed(job_id
)) {
296 error_setg(errp
, "Invalid job ID '%s'", job_id
);
299 if (job_get(job_id
)) {
300 error_setg(errp
, "Job ID '%s' already in use", job_id
);
303 } else if (!(flags
& JOB_INTERNAL
)) {
304 error_setg(errp
, "An explicit job ID is required");
308 job
= g_malloc0(driver
->instance_size
);
309 job
->driver
= driver
;
310 job
->id
= g_strdup(job_id
);
312 job
->aio_context
= ctx
;
315 job
->pause_count
= 1;
316 job
->auto_finalize
= !(flags
& JOB_MANUAL_FINALIZE
);
317 job
->auto_dismiss
= !(flags
& JOB_MANUAL_DISMISS
);
319 job
->opaque
= opaque
;
321 notifier_list_init(&job
->on_finalize_cancelled
);
322 notifier_list_init(&job
->on_finalize_completed
);
323 notifier_list_init(&job
->on_pending
);
325 job_state_transition(job
, JOB_STATUS_CREATED
);
326 aio_timer_init(qemu_get_aio_context(), &job
->sleep_timer
,
327 QEMU_CLOCK_REALTIME
, SCALE_NS
,
328 job_sleep_timer_cb
, job
);
330 QLIST_INSERT_HEAD(&jobs
, job
, job_list
);
332 /* Single jobs are modeled as single-job transactions for sake of
333 * consolidating the job management logic */
336 job_txn_add_job(txn
, job
);
339 job_txn_add_job(txn
, job
);
345 void job_ref(Job
*job
)
350 void job_unref(Job
*job
)
352 if (--job
->refcnt
== 0) {
353 assert(job
->status
== JOB_STATUS_NULL
);
354 assert(!timer_pending(&job
->sleep_timer
));
357 if (job
->driver
->free
) {
358 job
->driver
->free(job
);
361 QLIST_REMOVE(job
, job_list
);
368 void job_event_cancelled(Job
*job
)
370 notifier_list_notify(&job
->on_finalize_cancelled
, job
);
373 void job_event_completed(Job
*job
)
375 notifier_list_notify(&job
->on_finalize_completed
, job
);
378 static void job_event_pending(Job
*job
)
380 notifier_list_notify(&job
->on_pending
, job
);
383 void job_enter_cond(Job
*job
, bool(*fn
)(Job
*job
))
385 if (!job_started(job
)) {
388 if (job
->deferred_to_main_loop
) {
398 if (fn
&& !fn(job
)) {
403 assert(!job
->deferred_to_main_loop
);
404 timer_del(&job
->sleep_timer
);
407 aio_co_wake(job
->co
);
410 void job_enter(Job
*job
)
412 job_enter_cond(job
, NULL
);
415 /* Yield, and schedule a timer to reenter the coroutine after @ns nanoseconds.
416 * Reentering the job coroutine with job_enter() before the timer has expired
417 * is allowed and cancels the timer.
419 * If @ns is (uint64_t) -1, no timer is scheduled and job_enter() must be
420 * called explicitly. */
421 static void coroutine_fn
job_do_yield(Job
*job
, uint64_t ns
)
425 timer_mod(&job
->sleep_timer
, ns
);
429 qemu_coroutine_yield();
431 /* Set by job_enter_cond() before re-entering the coroutine. */
435 void coroutine_fn
job_pause_point(Job
*job
)
437 assert(job
&& job_started(job
));
439 if (!job_should_pause(job
)) {
442 if (job_is_cancelled(job
)) {
446 if (job
->driver
->pause
) {
447 job
->driver
->pause(job
);
450 if (job_should_pause(job
) && !job_is_cancelled(job
)) {
451 JobStatus status
= job
->status
;
452 job_state_transition(job
, status
== JOB_STATUS_READY
454 : JOB_STATUS_PAUSED
);
456 job_do_yield(job
, -1);
458 job_state_transition(job
, status
);
461 if (job
->driver
->resume
) {
462 job
->driver
->resume(job
);
466 void job_yield(Job
*job
)
470 /* Check cancellation *before* setting busy = false, too! */
471 if (job_is_cancelled(job
)) {
475 if (!job_should_pause(job
)) {
476 job_do_yield(job
, -1);
479 job_pause_point(job
);
482 void coroutine_fn
job_sleep_ns(Job
*job
, int64_t ns
)
486 /* Check cancellation *before* setting busy = false, too! */
487 if (job_is_cancelled(job
)) {
491 if (!job_should_pause(job
)) {
492 job_do_yield(job
, qemu_clock_get_ns(QEMU_CLOCK_REALTIME
) + ns
);
495 job_pause_point(job
);
498 void job_drain(Job
*job
)
500 /* If job is !busy this kicks it into the next pause point. */
503 if (job
->driver
->drain
) {
504 job
->driver
->drain(job
);
510 * All jobs must allow a pause point before entering their job proper. This
511 * ensures that jobs can be paused prior to being started, then resumed later.
513 static void coroutine_fn
job_co_entry(void *opaque
)
517 assert(job
&& job
->driver
&& job
->driver
->start
);
518 job_pause_point(job
);
519 job
->driver
->start(job
);
523 void job_start(Job
*job
)
525 assert(job
&& !job_started(job
) && job
->paused
&&
526 job
->driver
&& job
->driver
->start
);
527 job
->co
= qemu_coroutine_create(job_co_entry
, job
);
531 job_state_transition(job
, JOB_STATUS_RUNNING
);
532 aio_co_enter(job
->aio_context
, job
->co
);
535 /* Assumes the block_job_mutex is held */
536 static bool job_timer_not_pending(Job
*job
)
538 return !timer_pending(&job
->sleep_timer
);
541 void job_pause(Job
*job
)
546 void job_resume(Job
*job
)
548 assert(job
->pause_count
> 0);
550 if (job
->pause_count
) {
554 /* kick only if no timer is pending */
555 job_enter_cond(job
, job_timer_not_pending
);
558 void job_user_pause(Job
*job
, Error
**errp
)
560 if (job_apply_verb(job
, JOB_VERB_PAUSE
, errp
)) {
563 if (job
->user_paused
) {
564 error_setg(errp
, "Job is already paused");
567 job
->user_paused
= true;
571 bool job_user_paused(Job
*job
)
573 return job
->user_paused
;
576 void job_user_resume(Job
*job
, Error
**errp
)
579 if (!job
->user_paused
|| job
->pause_count
<= 0) {
580 error_setg(errp
, "Can't resume a job that was not paused");
583 if (job_apply_verb(job
, JOB_VERB_RESUME
, errp
)) {
586 if (job
->driver
->user_resume
) {
587 job
->driver
->user_resume(job
);
589 job
->user_paused
= false;
593 static void job_do_dismiss(Job
*job
)
598 job
->deferred_to_main_loop
= true;
600 job_txn_del_job(job
);
602 job_state_transition(job
, JOB_STATUS_NULL
);
606 void job_dismiss(Job
**jobptr
, Error
**errp
)
609 /* similarly to _complete, this is QMP-interface only. */
611 if (job_apply_verb(job
, JOB_VERB_DISMISS
, errp
)) {
619 void job_early_fail(Job
*job
)
621 assert(job
->status
== JOB_STATUS_CREATED
);
625 static void job_conclude(Job
*job
)
627 job_state_transition(job
, JOB_STATUS_CONCLUDED
);
628 if (job
->auto_dismiss
|| !job_started(job
)) {
633 static void job_update_rc(Job
*job
)
635 if (!job
->ret
&& job_is_cancelled(job
)) {
636 job
->ret
= -ECANCELED
;
639 job_state_transition(job
, JOB_STATUS_ABORTING
);
643 static void job_commit(Job
*job
)
646 if (job
->driver
->commit
) {
647 job
->driver
->commit(job
);
651 static void job_abort(Job
*job
)
654 if (job
->driver
->abort
) {
655 job
->driver
->abort(job
);
659 static void job_clean(Job
*job
)
661 if (job
->driver
->clean
) {
662 job
->driver
->clean(job
);
666 static int job_finalize_single(Job
*job
)
668 assert(job_is_completed(job
));
670 /* Ensure abort is called for late-transactional failures */
681 job
->cb(job
->opaque
, job
->ret
);
684 /* Emit events only if we actually started */
685 if (job_started(job
)) {
686 if (job_is_cancelled(job
)) {
687 job_event_cancelled(job
);
689 job_event_completed(job
);
693 job_txn_del_job(job
);
698 static void job_cancel_async(Job
*job
, bool force
)
700 if (job
->user_paused
) {
701 /* Do not call job_enter here, the caller will handle it. */
702 job
->user_paused
= false;
703 if (job
->driver
->user_resume
) {
704 job
->driver
->user_resume(job
);
706 assert(job
->pause_count
> 0);
709 job
->cancelled
= true;
710 /* To prevent 'force == false' overriding a previous 'force == true' */
711 job
->force_cancel
|= force
;
714 static void job_completed_txn_abort(Job
*job
)
717 JobTxn
*txn
= job
->txn
;
722 * We are cancelled by another job, which will handle everything.
726 txn
->aborting
= true;
729 /* We are the first failed job. Cancel other jobs. */
730 QLIST_FOREACH(other_job
, &txn
->jobs
, txn_list
) {
731 ctx
= other_job
->aio_context
;
732 aio_context_acquire(ctx
);
735 /* Other jobs are effectively cancelled by us, set the status for
736 * them; this job, however, may or may not be cancelled, depending
737 * on the caller, so leave it. */
738 QLIST_FOREACH(other_job
, &txn
->jobs
, txn_list
) {
739 if (other_job
!= job
) {
740 job_cancel_async(other_job
, false);
743 while (!QLIST_EMPTY(&txn
->jobs
)) {
744 other_job
= QLIST_FIRST(&txn
->jobs
);
745 ctx
= other_job
->aio_context
;
746 if (!job_is_completed(other_job
)) {
747 assert(job_is_cancelled(other_job
));
748 job_finish_sync(other_job
, NULL
, NULL
);
750 job_finalize_single(other_job
);
751 aio_context_release(ctx
);
757 static int job_prepare(Job
*job
)
759 if (job
->ret
== 0 && job
->driver
->prepare
) {
760 job
->ret
= job
->driver
->prepare(job
);
765 static int job_needs_finalize(Job
*job
)
767 return !job
->auto_finalize
;
770 static void job_do_finalize(Job
*job
)
773 assert(job
&& job
->txn
);
775 /* prepare the transaction to complete */
776 rc
= job_txn_apply(job
->txn
, job_prepare
, true);
778 job_completed_txn_abort(job
);
780 job_txn_apply(job
->txn
, job_finalize_single
, true);
784 void job_finalize(Job
*job
, Error
**errp
)
786 assert(job
&& job
->id
);
787 if (job_apply_verb(job
, JOB_VERB_FINALIZE
, errp
)) {
790 job_do_finalize(job
);
793 static int job_transition_to_pending(Job
*job
)
795 job_state_transition(job
, JOB_STATUS_PENDING
);
796 if (!job
->auto_finalize
) {
797 job_event_pending(job
);
802 static void job_completed_txn_success(Job
*job
)
804 JobTxn
*txn
= job
->txn
;
807 job_state_transition(job
, JOB_STATUS_WAITING
);
810 * Successful completion, see if there are other running jobs in this
813 QLIST_FOREACH(other_job
, &txn
->jobs
, txn_list
) {
814 if (!job_is_completed(other_job
)) {
817 assert(other_job
->ret
== 0);
820 job_txn_apply(txn
, job_transition_to_pending
, false);
822 /* If no jobs need manual finalization, automatically do so */
823 if (job_txn_apply(txn
, job_needs_finalize
, false) == 0) {
824 job_do_finalize(job
);
828 void job_completed(Job
*job
, int ret
)
830 assert(job
&& job
->txn
&& !job_is_completed(job
));
833 trace_job_completed(job
, ret
, job
->ret
);
835 job_completed_txn_abort(job
);
837 job_completed_txn_success(job
);
841 void job_cancel(Job
*job
, bool force
)
843 if (job
->status
== JOB_STATUS_CONCLUDED
) {
847 job_cancel_async(job
, force
);
848 if (!job_started(job
)) {
849 job_completed(job
, -ECANCELED
);
850 } else if (job
->deferred_to_main_loop
) {
851 job_completed_txn_abort(job
);
857 void job_user_cancel(Job
*job
, bool force
, Error
**errp
)
859 if (job_apply_verb(job
, JOB_VERB_CANCEL
, errp
)) {
862 job_cancel(job
, force
);
865 /* A wrapper around job_cancel() taking an Error ** parameter so it may be
866 * used with job_finish_sync() without the need for (rather nasty) function
867 * pointer casts there. */
868 static void job_cancel_err(Job
*job
, Error
**errp
)
870 job_cancel(job
, false);
873 int job_cancel_sync(Job
*job
)
875 return job_finish_sync(job
, &job_cancel_err
, NULL
);
878 void job_cancel_sync_all(void)
881 AioContext
*aio_context
;
883 while ((job
= job_next(NULL
))) {
884 aio_context
= job
->aio_context
;
885 aio_context_acquire(aio_context
);
886 job_cancel_sync(job
);
887 aio_context_release(aio_context
);
891 int job_complete_sync(Job
*job
, Error
**errp
)
893 return job_finish_sync(job
, job_complete
, errp
);
896 void job_complete(Job
*job
, Error
**errp
)
898 /* Should not be reachable via external interface for internal jobs */
900 if (job_apply_verb(job
, JOB_VERB_COMPLETE
, errp
)) {
903 if (job
->pause_count
|| job_is_cancelled(job
) || !job
->driver
->complete
) {
904 error_setg(errp
, "The active block job '%s' cannot be completed",
909 job
->driver
->complete(job
, errp
);
915 JobDeferToMainLoopFn
*fn
;
917 } JobDeferToMainLoopData
;
919 static void job_defer_to_main_loop_bh(void *opaque
)
921 JobDeferToMainLoopData
*data
= opaque
;
922 Job
*job
= data
->job
;
923 AioContext
*aio_context
= job
->aio_context
;
925 aio_context_acquire(aio_context
);
926 data
->fn(data
->job
, data
->opaque
);
927 aio_context_release(aio_context
);
932 void job_defer_to_main_loop(Job
*job
, JobDeferToMainLoopFn
*fn
, void *opaque
)
934 JobDeferToMainLoopData
*data
= g_malloc(sizeof(*data
));
937 data
->opaque
= opaque
;
938 job
->deferred_to_main_loop
= true;
940 aio_bh_schedule_oneshot(qemu_get_aio_context(),
941 job_defer_to_main_loop_bh
, data
);
944 int job_finish_sync(Job
*job
, void (*finish
)(Job
*, Error
**errp
), Error
**errp
)
946 Error
*local_err
= NULL
;
952 finish(job
, &local_err
);
955 error_propagate(errp
, local_err
);
959 /* job_drain calls job_enter, and it should be enough to induce progress
960 * until the job completes or moves to the main thread. */
961 while (!job
->deferred_to_main_loop
&& !job_is_completed(job
)) {
964 while (!job_is_completed(job
)) {
965 aio_poll(qemu_get_aio_context(), true);
967 ret
= (job_is_cancelled(job
) && job
->ret
== 0) ? -ECANCELED
: job
->ret
;