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 "qapi/error.h"
30 #include "qemu/main-loop.h"
31 #include "block/aio-wait.h"
32 #include "trace/trace-root.h"
33 #include "qapi/qapi-events-job.h"
35 static QLIST_HEAD(, Job) jobs = QLIST_HEAD_INITIALIZER(jobs);
37 /* Job State Transition Table */
38 bool JobSTT[JOB_STATUS__MAX][JOB_STATUS__MAX] = {
39 /* U, C, R, P, Y, S, W, D, X, E, N */
40 /* U: */ [JOB_STATUS_UNDEFINED] = {0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0},
41 /* C: */ [JOB_STATUS_CREATED] = {0, 0, 1, 0, 0, 0, 0, 0, 1, 0, 1},
42 /* R: */ [JOB_STATUS_RUNNING] = {0, 0, 0, 1, 1, 0, 1, 0, 1, 0, 0},
43 /* P: */ [JOB_STATUS_PAUSED] = {0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0},
44 /* Y: */ [JOB_STATUS_READY] = {0, 0, 0, 0, 0, 1, 1, 0, 1, 0, 0},
45 /* S: */ [JOB_STATUS_STANDBY] = {0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0},
46 /* W: */ [JOB_STATUS_WAITING] = {0, 0, 0, 0, 0, 0, 0, 1, 1, 0, 0},
47 /* D: */ [JOB_STATUS_PENDING] = {0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 0},
48 /* X: */ [JOB_STATUS_ABORTING] = {0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 0},
49 /* E: */ [JOB_STATUS_CONCLUDED] = {0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1},
50 /* N: */ [JOB_STATUS_NULL] = {0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0},
53 bool JobVerbTable[JOB_VERB__MAX][JOB_STATUS__MAX] = {
54 /* U, C, R, P, Y, S, W, D, X, E, N */
55 [JOB_VERB_CANCEL] = {0, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0},
56 [JOB_VERB_PAUSE] = {0, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0},
57 [JOB_VERB_RESUME] = {0, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0},
58 [JOB_VERB_SET_SPEED] = {0, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0},
59 [JOB_VERB_COMPLETE] = {0, 0, 0, 0, 1, 1, 0, 0, 0, 0, 0},
60 [JOB_VERB_FINALIZE] = {0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0},
61 [JOB_VERB_DISMISS] = {0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0},
64 /* Transactional group of jobs */
67 /* Is this txn being cancelled? */
71 QLIST_HEAD(, Job) jobs;
77 /* Right now, this mutex is only needed to synchronize accesses to job->busy
78 * and job->sleep_timer, such as concurrent calls to job_do_yield and
80 static QemuMutex job_mutex;
82 static void job_lock(void)
84 qemu_mutex_lock(&job_mutex);
87 static void job_unlock(void)
89 qemu_mutex_unlock(&job_mutex);
92 static void __attribute__((__constructor__)) job_init(void)
94 qemu_mutex_init(&job_mutex);
97 JobTxn *job_txn_new(void)
99 JobTxn *txn = g_new0(JobTxn, 1);
100 QLIST_INIT(&txn->jobs);
105 static void job_txn_ref(JobTxn *txn)
110 void job_txn_unref(JobTxn *txn)
112 if (txn && --txn->refcnt == 0) {
117 void job_txn_add_job(JobTxn *txn, Job *job)
126 QLIST_INSERT_HEAD(&txn->jobs, job, txn_list);
130 static void job_txn_del_job(Job *job)
133 QLIST_REMOVE(job, txn_list);
134 job_txn_unref(job->txn);
139 static int job_txn_apply(Job *job, int fn(Job *))
141 AioContext *inner_ctx;
142 Job *other_job, *next;
143 JobTxn *txn = job->txn;
147 * Similar to job_completed_txn_abort, we take each job's lock before
148 * applying fn, but since we assume that outer_ctx is held by the caller,
149 * we need to release it here to avoid holding the lock twice - which would
150 * break AIO_WAIT_WHILE from within fn.
153 aio_context_release(job->aio_context);
155 QLIST_FOREACH_SAFE(other_job, &txn->jobs, txn_list, next) {
156 inner_ctx = other_job->aio_context;
157 aio_context_acquire(inner_ctx);
159 aio_context_release(inner_ctx);
166 * Note that job->aio_context might have been changed by calling fn, so we
167 * can't use a local variable to cache it.
169 aio_context_acquire(job->aio_context);
174 bool job_is_internal(Job *job)
176 return (job->id == NULL);
179 static void job_state_transition(Job *job, JobStatus s1)
181 JobStatus s0 = job->status;
182 assert(s1 >= 0 && s1 < JOB_STATUS__MAX);
183 trace_job_state_transition(job, job->ret,
184 JobSTT[s0][s1] ? "allowed" : "disallowed",
185 JobStatus_str(s0), JobStatus_str(s1));
186 assert(JobSTT[s0][s1]);
189 if (!job_is_internal(job) && s1 != s0) {
190 qapi_event_send_job_status_change(job->id, job->status);
194 int job_apply_verb(Job *job, JobVerb verb, Error **errp)
196 JobStatus s0 = job->status;
197 assert(verb >= 0 && verb < JOB_VERB__MAX);
198 trace_job_apply_verb(job, JobStatus_str(s0), JobVerb_str(verb),
199 JobVerbTable[verb][s0] ? "allowed" : "prohibited");
200 if (JobVerbTable[verb][s0]) {
203 error_setg(errp, "Job '%s' in state '%s' cannot accept command verb '%s'",
204 job->id, JobStatus_str(s0), JobVerb_str(verb));
208 JobType job_type(const Job *job)
210 return job->driver->job_type;
213 const char *job_type_str(const Job *job)
215 return JobType_str(job_type(job));
218 bool job_is_cancelled(Job *job)
220 return job->cancelled;
223 bool job_is_ready(Job *job)
225 switch (job->status) {
226 case JOB_STATUS_UNDEFINED:
227 case JOB_STATUS_CREATED:
228 case JOB_STATUS_RUNNING:
229 case JOB_STATUS_PAUSED:
230 case JOB_STATUS_WAITING:
231 case JOB_STATUS_PENDING:
232 case JOB_STATUS_ABORTING:
233 case JOB_STATUS_CONCLUDED:
234 case JOB_STATUS_NULL:
236 case JOB_STATUS_READY:
237 case JOB_STATUS_STANDBY:
240 g_assert_not_reached();
245 bool job_is_completed(Job *job)
247 switch (job->status) {
248 case JOB_STATUS_UNDEFINED:
249 case JOB_STATUS_CREATED:
250 case JOB_STATUS_RUNNING:
251 case JOB_STATUS_PAUSED:
252 case JOB_STATUS_READY:
253 case JOB_STATUS_STANDBY:
255 case JOB_STATUS_WAITING:
256 case JOB_STATUS_PENDING:
257 case JOB_STATUS_ABORTING:
258 case JOB_STATUS_CONCLUDED:
259 case JOB_STATUS_NULL:
262 g_assert_not_reached();
267 static bool job_started(Job *job)
272 static bool job_should_pause(Job *job)
274 return job->pause_count > 0;
277 Job *job_next(Job *job)
280 return QLIST_FIRST(&jobs);
282 return QLIST_NEXT(job, job_list);
285 Job *job_get(const char *id)
289 QLIST_FOREACH(job, &jobs, job_list) {
290 if (job->id && !strcmp(id, job->id)) {
298 static void job_sleep_timer_cb(void *opaque)
305 void *job_create(const char *job_id, const JobDriver *driver, JobTxn *txn,
306 AioContext *ctx, int flags, BlockCompletionFunc *cb,
307 void *opaque, Error **errp)
312 if (flags & JOB_INTERNAL) {
313 error_setg(errp, "Cannot specify job ID for internal job");
316 if (!id_wellformed(job_id)) {
317 error_setg(errp, "Invalid job ID '%s'", job_id);
320 if (job_get(job_id)) {
321 error_setg(errp, "Job ID '%s' already in use", job_id);
324 } else if (!(flags & JOB_INTERNAL)) {
325 error_setg(errp, "An explicit job ID is required");
329 job = g_malloc0(driver->instance_size);
330 job->driver = driver;
331 job->id = g_strdup(job_id);
333 job->aio_context = ctx;
336 job->pause_count = 1;
337 job->auto_finalize = !(flags & JOB_MANUAL_FINALIZE);
338 job->auto_dismiss = !(flags & JOB_MANUAL_DISMISS);
340 job->opaque = opaque;
342 progress_init(&job->progress);
344 notifier_list_init(&job->on_finalize_cancelled);
345 notifier_list_init(&job->on_finalize_completed);
346 notifier_list_init(&job->on_pending);
347 notifier_list_init(&job->on_ready);
349 job_state_transition(job, JOB_STATUS_CREATED);
350 aio_timer_init(qemu_get_aio_context(), &job->sleep_timer,
351 QEMU_CLOCK_REALTIME, SCALE_NS,
352 job_sleep_timer_cb, job);
354 QLIST_INSERT_HEAD(&jobs, job, job_list);
356 /* Single jobs are modeled as single-job transactions for sake of
357 * consolidating the job management logic */
360 job_txn_add_job(txn, job);
363 job_txn_add_job(txn, job);
369 void job_ref(Job *job)
374 void job_unref(Job *job)
376 if (--job->refcnt == 0) {
377 assert(job->status == JOB_STATUS_NULL);
378 assert(!timer_pending(&job->sleep_timer));
381 if (job->driver->free) {
382 job->driver->free(job);
385 QLIST_REMOVE(job, job_list);
387 progress_destroy(&job->progress);
388 error_free(job->err);
394 void job_progress_update(Job *job, uint64_t done)
396 progress_work_done(&job->progress, done);
399 void job_progress_set_remaining(Job *job, uint64_t remaining)
401 progress_set_remaining(&job->progress, remaining);
404 void job_progress_increase_remaining(Job *job, uint64_t delta)
406 progress_increase_remaining(&job->progress, delta);
409 void job_event_cancelled(Job *job)
411 notifier_list_notify(&job->on_finalize_cancelled, job);
414 void job_event_completed(Job *job)
416 notifier_list_notify(&job->on_finalize_completed, job);
419 static void job_event_pending(Job *job)
421 notifier_list_notify(&job->on_pending, job);
424 static void job_event_ready(Job *job)
426 notifier_list_notify(&job->on_ready, job);
429 static void job_event_idle(Job *job)
431 notifier_list_notify(&job->on_idle, job);
434 void job_enter_cond(Job *job, bool(*fn)(Job *job))
436 if (!job_started(job)) {
439 if (job->deferred_to_main_loop) {
449 if (fn && !fn(job)) {
454 assert(!job->deferred_to_main_loop);
455 timer_del(&job->sleep_timer);
458 aio_co_enter(job->aio_context, job->co);
461 void job_enter(Job *job)
463 job_enter_cond(job, NULL);
466 /* Yield, and schedule a timer to reenter the coroutine after @ns nanoseconds.
467 * Reentering the job coroutine with job_enter() before the timer has expired
468 * is allowed and cancels the timer.
470 * If @ns is (uint64_t) -1, no timer is scheduled and job_enter() must be
471 * called explicitly. */
472 static void coroutine_fn job_do_yield(Job *job, uint64_t ns)
476 timer_mod(&job->sleep_timer, ns);
481 qemu_coroutine_yield();
483 /* Set by job_enter_cond() before re-entering the coroutine. */
487 void coroutine_fn job_pause_point(Job *job)
489 assert(job && job_started(job));
491 if (!job_should_pause(job)) {
494 if (job_is_cancelled(job)) {
498 if (job->driver->pause) {
499 job->driver->pause(job);
502 if (job_should_pause(job) && !job_is_cancelled(job)) {
503 JobStatus status = job->status;
504 job_state_transition(job, status == JOB_STATUS_READY
506 : JOB_STATUS_PAUSED);
508 job_do_yield(job, -1);
510 job_state_transition(job, status);
513 if (job->driver->resume) {
514 job->driver->resume(job);
518 void job_yield(Job *job)
522 /* Check cancellation *before* setting busy = false, too! */
523 if (job_is_cancelled(job)) {
527 if (!job_should_pause(job)) {
528 job_do_yield(job, -1);
531 job_pause_point(job);
534 void coroutine_fn job_sleep_ns(Job *job, int64_t ns)
538 /* Check cancellation *before* setting busy = false, too! */
539 if (job_is_cancelled(job)) {
543 if (!job_should_pause(job)) {
544 job_do_yield(job, qemu_clock_get_ns(QEMU_CLOCK_REALTIME) + ns);
547 job_pause_point(job);
550 /* Assumes the block_job_mutex is held */
551 static bool job_timer_not_pending(Job *job)
553 return !timer_pending(&job->sleep_timer);
556 void job_pause(Job *job)
564 void job_resume(Job *job)
566 assert(job->pause_count > 0);
568 if (job->pause_count) {
572 /* kick only if no timer is pending */
573 job_enter_cond(job, job_timer_not_pending);
576 void job_user_pause(Job *job, Error **errp)
578 if (job_apply_verb(job, JOB_VERB_PAUSE, errp)) {
581 if (job->user_paused) {
582 error_setg(errp, "Job is already paused");
585 job->user_paused = true;
589 bool job_user_paused(Job *job)
591 return job->user_paused;
594 void job_user_resume(Job *job, Error **errp)
597 if (!job->user_paused || job->pause_count <= 0) {
598 error_setg(errp, "Can't resume a job that was not paused");
601 if (job_apply_verb(job, JOB_VERB_RESUME, errp)) {
604 if (job->driver->user_resume) {
605 job->driver->user_resume(job);
607 job->user_paused = false;
611 static void job_do_dismiss(Job *job)
616 job->deferred_to_main_loop = true;
618 job_txn_del_job(job);
620 job_state_transition(job, JOB_STATUS_NULL);
624 void job_dismiss(Job **jobptr, Error **errp)
627 /* similarly to _complete, this is QMP-interface only. */
629 if (job_apply_verb(job, JOB_VERB_DISMISS, errp)) {
637 void job_early_fail(Job *job)
639 assert(job->status == JOB_STATUS_CREATED);
643 static void job_conclude(Job *job)
645 job_state_transition(job, JOB_STATUS_CONCLUDED);
646 if (job->auto_dismiss || !job_started(job)) {
651 static void job_update_rc(Job *job)
653 if (!job->ret && job_is_cancelled(job)) {
654 job->ret = -ECANCELED;
658 error_setg(&job->err, "%s", strerror(-job->ret));
660 job_state_transition(job, JOB_STATUS_ABORTING);
664 static void job_commit(Job *job)
667 if (job->driver->commit) {
668 job->driver->commit(job);
672 static void job_abort(Job *job)
675 if (job->driver->abort) {
676 job->driver->abort(job);
680 static void job_clean(Job *job)
682 if (job->driver->clean) {
683 job->driver->clean(job);
687 static int job_finalize_single(Job *job)
689 assert(job_is_completed(job));
691 /* Ensure abort is called for late-transactional failures */
702 job->cb(job->opaque, job->ret);
705 /* Emit events only if we actually started */
706 if (job_started(job)) {
707 if (job_is_cancelled(job)) {
708 job_event_cancelled(job);
710 job_event_completed(job);
714 job_txn_del_job(job);
719 static void job_cancel_async(Job *job, bool force)
721 if (job->driver->cancel) {
722 job->driver->cancel(job, force);
724 if (job->user_paused) {
725 /* Do not call job_enter here, the caller will handle it. */
726 if (job->driver->user_resume) {
727 job->driver->user_resume(job);
729 job->user_paused = false;
730 assert(job->pause_count > 0);
733 job->cancelled = true;
734 /* To prevent 'force == false' overriding a previous 'force == true' */
735 job->force_cancel |= force;
738 static void job_completed_txn_abort(Job *job)
741 JobTxn *txn = job->txn;
746 * We are cancelled by another job, which will handle everything.
750 txn->aborting = true;
754 * We can only hold the single job's AioContext lock while calling
755 * job_finalize_single() because the finalization callbacks can involve
756 * calls of AIO_WAIT_WHILE(), which could deadlock otherwise.
757 * Note that the job's AioContext may change when it is finalized.
760 aio_context_release(job->aio_context);
762 /* Other jobs are effectively cancelled by us, set the status for
763 * them; this job, however, may or may not be cancelled, depending
764 * on the caller, so leave it. */
765 QLIST_FOREACH(other_job, &txn->jobs, txn_list) {
766 if (other_job != job) {
767 ctx = other_job->aio_context;
768 aio_context_acquire(ctx);
770 * This is a transaction: If one job failed, no result will matter.
771 * Therefore, pass force=true to terminate all other jobs as quickly
774 job_cancel_async(other_job, true);
775 aio_context_release(ctx);
778 while (!QLIST_EMPTY(&txn->jobs)) {
779 other_job = QLIST_FIRST(&txn->jobs);
781 * The job's AioContext may change, so store it in @ctx so we
782 * release the same context that we have acquired before.
784 ctx = other_job->aio_context;
785 aio_context_acquire(ctx);
786 if (!job_is_completed(other_job)) {
787 assert(job_is_cancelled(other_job));
788 job_finish_sync(other_job, NULL, NULL);
790 job_finalize_single(other_job);
791 aio_context_release(ctx);
795 * Use job_ref()/job_unref() so we can read the AioContext here
796 * even if the job went away during job_finalize_single().
798 aio_context_acquire(job->aio_context);
804 static int job_prepare(Job *job)
806 if (job->ret == 0 && job->driver->prepare) {
807 job->ret = job->driver->prepare(job);
813 static int job_needs_finalize(Job *job)
815 return !job->auto_finalize;
818 static void job_do_finalize(Job *job)
821 assert(job && job->txn);
823 /* prepare the transaction to complete */
824 rc = job_txn_apply(job, job_prepare);
826 job_completed_txn_abort(job);
828 job_txn_apply(job, job_finalize_single);
832 void job_finalize(Job *job, Error **errp)
834 assert(job && job->id);
835 if (job_apply_verb(job, JOB_VERB_FINALIZE, errp)) {
838 job_do_finalize(job);
841 static int job_transition_to_pending(Job *job)
843 job_state_transition(job, JOB_STATUS_PENDING);
844 if (!job->auto_finalize) {
845 job_event_pending(job);
850 void job_transition_to_ready(Job *job)
852 job_state_transition(job, JOB_STATUS_READY);
853 job_event_ready(job);
856 static void job_completed_txn_success(Job *job)
858 JobTxn *txn = job->txn;
861 job_state_transition(job, JOB_STATUS_WAITING);
864 * Successful completion, see if there are other running jobs in this
867 QLIST_FOREACH(other_job, &txn->jobs, txn_list) {
868 if (!job_is_completed(other_job)) {
871 assert(other_job->ret == 0);
874 job_txn_apply(job, job_transition_to_pending);
876 /* If no jobs need manual finalization, automatically do so */
877 if (job_txn_apply(job, job_needs_finalize) == 0) {
878 job_do_finalize(job);
882 static void job_completed(Job *job)
884 assert(job && job->txn && !job_is_completed(job));
887 trace_job_completed(job, job->ret);
889 job_completed_txn_abort(job);
891 job_completed_txn_success(job);
895 /** Useful only as a type shim for aio_bh_schedule_oneshot. */
896 static void job_exit(void *opaque)
898 Job *job = (Job *)opaque;
902 aio_context_acquire(job->aio_context);
904 /* This is a lie, we're not quiescent, but still doing the completion
905 * callbacks. However, completion callbacks tend to involve operations that
906 * drain block nodes, and if .drained_poll still returned true, we would
914 * Note that calling job_completed can move the job to a different
915 * aio_context, so we cannot cache from above. job_txn_apply takes care of
916 * acquiring the new lock, and we ref/unref to avoid job_completed freeing
917 * the job underneath us.
919 ctx = job->aio_context;
921 aio_context_release(ctx);
925 * All jobs must allow a pause point before entering their job proper. This
926 * ensures that jobs can be paused prior to being started, then resumed later.
928 static void coroutine_fn job_co_entry(void *opaque)
932 assert(job && job->driver && job->driver->run);
933 job_pause_point(job);
934 job->ret = job->driver->run(job, &job->err);
935 job->deferred_to_main_loop = true;
937 aio_bh_schedule_oneshot(qemu_get_aio_context(), job_exit, job);
940 void job_start(Job *job)
942 assert(job && !job_started(job) && job->paused &&
943 job->driver && job->driver->run);
944 job->co = qemu_coroutine_create(job_co_entry, job);
948 job_state_transition(job, JOB_STATUS_RUNNING);
949 aio_co_enter(job->aio_context, job->co);
952 void job_cancel(Job *job, bool force)
954 if (job->status == JOB_STATUS_CONCLUDED) {
958 job_cancel_async(job, force);
959 if (!job_started(job)) {
961 } else if (job->deferred_to_main_loop) {
962 job_completed_txn_abort(job);
968 void job_user_cancel(Job *job, bool force, Error **errp)
970 if (job_apply_verb(job, JOB_VERB_CANCEL, errp)) {
973 job_cancel(job, force);
976 /* A wrapper around job_cancel() taking an Error ** parameter so it may be
977 * used with job_finish_sync() without the need for (rather nasty) function
978 * pointer casts there. */
979 static void job_cancel_err(Job *job, Error **errp)
981 job_cancel(job, false);
984 int job_cancel_sync(Job *job)
986 return job_finish_sync(job, &job_cancel_err, NULL);
989 void job_cancel_sync_all(void)
992 AioContext *aio_context;
994 while ((job = job_next(NULL))) {
995 aio_context = job->aio_context;
996 aio_context_acquire(aio_context);
997 job_cancel_sync(job);
998 aio_context_release(aio_context);
1002 int job_complete_sync(Job *job, Error **errp)
1004 return job_finish_sync(job, job_complete, errp);
1007 void job_complete(Job *job, Error **errp)
1009 /* Should not be reachable via external interface for internal jobs */
1011 if (job_apply_verb(job, JOB_VERB_COMPLETE, errp)) {
1014 if (job_is_cancelled(job) || !job->driver->complete) {
1015 error_setg(errp, "The active block job '%s' cannot be completed",
1020 job->driver->complete(job, errp);
1023 int job_finish_sync(Job *job, void (*finish)(Job *, Error **errp), Error **errp)
1025 Error *local_err = NULL;
1031 finish(job, &local_err);
1034 error_propagate(errp, local_err);
1039 AIO_WAIT_WHILE(job->aio_context,
1040 (job_enter(job), !job_is_completed(job)));
1042 ret = (job_is_cancelled(job) && job->ret == 0) ? -ECANCELED : job->ret;