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-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, 0, 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(JobTxn *txn, int fn(Job *))
144 QLIST_FOREACH_SAFE(job, &txn->jobs, txn_list, next) {
153 bool job_is_internal(Job *job)
155 return (job->id == NULL);
158 static void job_state_transition(Job *job, JobStatus s1)
160 JobStatus s0 = job->status;
161 assert(s1 >= 0 && s1 < JOB_STATUS__MAX);
162 trace_job_state_transition(job, job->ret,
163 JobSTT[s0][s1] ? "allowed" : "disallowed",
164 JobStatus_str(s0), JobStatus_str(s1));
165 assert(JobSTT[s0][s1]);
168 if (!job_is_internal(job) && s1 != s0) {
169 qapi_event_send_job_status_change(job->id, job->status);
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);
324 notifier_list_init(&job->on_ready);
326 job_state_transition(job, JOB_STATUS_CREATED);
327 aio_timer_init(qemu_get_aio_context(), &job->sleep_timer,
328 QEMU_CLOCK_REALTIME, SCALE_NS,
329 job_sleep_timer_cb, job);
331 QLIST_INSERT_HEAD(&jobs, job, job_list);
333 /* Single jobs are modeled as single-job transactions for sake of
334 * consolidating the job management logic */
337 job_txn_add_job(txn, job);
340 job_txn_add_job(txn, job);
346 void job_ref(Job *job)
351 void job_unref(Job *job)
353 if (--job->refcnt == 0) {
354 assert(job->status == JOB_STATUS_NULL);
355 assert(!timer_pending(&job->sleep_timer));
358 if (job->driver->free) {
359 job->driver->free(job);
362 QLIST_REMOVE(job, job_list);
364 error_free(job->err);
370 void job_progress_update(Job *job, uint64_t done)
372 job->progress_current += done;
375 void job_progress_set_remaining(Job *job, uint64_t remaining)
377 job->progress_total = job->progress_current + remaining;
380 void job_progress_increase_remaining(Job *job, uint64_t delta)
382 job->progress_total += delta;
385 void job_event_cancelled(Job *job)
387 notifier_list_notify(&job->on_finalize_cancelled, job);
390 void job_event_completed(Job *job)
392 notifier_list_notify(&job->on_finalize_completed, job);
395 static void job_event_pending(Job *job)
397 notifier_list_notify(&job->on_pending, job);
400 static void job_event_ready(Job *job)
402 notifier_list_notify(&job->on_ready, job);
405 static void job_event_idle(Job *job)
407 notifier_list_notify(&job->on_idle, job);
410 void job_enter_cond(Job *job, bool(*fn)(Job *job))
412 if (!job_started(job)) {
415 if (job->deferred_to_main_loop) {
425 if (fn && !fn(job)) {
430 assert(!job->deferred_to_main_loop);
431 timer_del(&job->sleep_timer);
434 aio_co_enter(job->aio_context, job->co);
437 void job_enter(Job *job)
439 job_enter_cond(job, NULL);
442 /* Yield, and schedule a timer to reenter the coroutine after @ns nanoseconds.
443 * Reentering the job coroutine with job_enter() before the timer has expired
444 * is allowed and cancels the timer.
446 * If @ns is (uint64_t) -1, no timer is scheduled and job_enter() must be
447 * called explicitly. */
448 static void coroutine_fn job_do_yield(Job *job, uint64_t ns)
452 timer_mod(&job->sleep_timer, ns);
457 qemu_coroutine_yield();
459 /* Set by job_enter_cond() before re-entering the coroutine. */
463 void coroutine_fn job_pause_point(Job *job)
465 assert(job && job_started(job));
467 if (!job_should_pause(job)) {
470 if (job_is_cancelled(job)) {
474 if (job->driver->pause) {
475 job->driver->pause(job);
478 if (job_should_pause(job) && !job_is_cancelled(job)) {
479 JobStatus status = job->status;
480 job_state_transition(job, status == JOB_STATUS_READY
482 : JOB_STATUS_PAUSED);
484 job_do_yield(job, -1);
486 job_state_transition(job, status);
489 if (job->driver->resume) {
490 job->driver->resume(job);
494 void job_yield(Job *job)
498 /* Check cancellation *before* setting busy = false, too! */
499 if (job_is_cancelled(job)) {
503 if (!job_should_pause(job)) {
504 job_do_yield(job, -1);
507 job_pause_point(job);
510 void coroutine_fn job_sleep_ns(Job *job, int64_t ns)
514 /* Check cancellation *before* setting busy = false, too! */
515 if (job_is_cancelled(job)) {
519 if (!job_should_pause(job)) {
520 job_do_yield(job, qemu_clock_get_ns(QEMU_CLOCK_REALTIME) + ns);
523 job_pause_point(job);
526 /* Assumes the block_job_mutex is held */
527 static bool job_timer_not_pending(Job *job)
529 return !timer_pending(&job->sleep_timer);
532 void job_pause(Job *job)
537 void job_resume(Job *job)
539 assert(job->pause_count > 0);
541 if (job->pause_count) {
545 /* kick only if no timer is pending */
546 job_enter_cond(job, job_timer_not_pending);
549 void job_user_pause(Job *job, Error **errp)
551 if (job_apply_verb(job, JOB_VERB_PAUSE, errp)) {
554 if (job->user_paused) {
555 error_setg(errp, "Job is already paused");
558 job->user_paused = true;
562 bool job_user_paused(Job *job)
564 return job->user_paused;
567 void job_user_resume(Job *job, Error **errp)
570 if (!job->user_paused || job->pause_count <= 0) {
571 error_setg(errp, "Can't resume a job that was not paused");
574 if (job_apply_verb(job, JOB_VERB_RESUME, errp)) {
577 if (job->driver->user_resume) {
578 job->driver->user_resume(job);
580 job->user_paused = false;
584 static void job_do_dismiss(Job *job)
589 job->deferred_to_main_loop = true;
591 job_txn_del_job(job);
593 job_state_transition(job, JOB_STATUS_NULL);
597 void job_dismiss(Job **jobptr, Error **errp)
600 /* similarly to _complete, this is QMP-interface only. */
602 if (job_apply_verb(job, JOB_VERB_DISMISS, errp)) {
610 void job_early_fail(Job *job)
612 assert(job->status == JOB_STATUS_CREATED);
616 static void job_conclude(Job *job)
618 job_state_transition(job, JOB_STATUS_CONCLUDED);
619 if (job->auto_dismiss || !job_started(job)) {
624 static void job_update_rc(Job *job)
626 if (!job->ret && job_is_cancelled(job)) {
627 job->ret = -ECANCELED;
631 error_setg(&job->err, "%s", strerror(-job->ret));
633 job_state_transition(job, JOB_STATUS_ABORTING);
637 static void job_commit(Job *job)
640 if (job->driver->commit) {
641 job->driver->commit(job);
645 static void job_abort(Job *job)
648 if (job->driver->abort) {
649 job->driver->abort(job);
653 static void job_clean(Job *job)
655 if (job->driver->clean) {
656 job->driver->clean(job);
660 static int job_finalize_single(Job *job)
662 assert(job_is_completed(job));
664 /* Ensure abort is called for late-transactional failures */
675 job->cb(job->opaque, job->ret);
678 /* Emit events only if we actually started */
679 if (job_started(job)) {
680 if (job_is_cancelled(job)) {
681 job_event_cancelled(job);
683 job_event_completed(job);
687 job_txn_del_job(job);
692 static void job_cancel_async(Job *job, bool force)
694 if (job->user_paused) {
695 /* Do not call job_enter here, the caller will handle it. */
696 if (job->driver->user_resume) {
697 job->driver->user_resume(job);
699 job->user_paused = false;
700 assert(job->pause_count > 0);
703 job->cancelled = true;
704 /* To prevent 'force == false' overriding a previous 'force == true' */
705 job->force_cancel |= force;
708 static void job_completed_txn_abort(Job *job)
710 AioContext *outer_ctx = job->aio_context;
712 JobTxn *txn = job->txn;
717 * We are cancelled by another job, which will handle everything.
721 txn->aborting = true;
724 /* We can only hold the single job's AioContext lock while calling
725 * job_finalize_single() because the finalization callbacks can involve
726 * calls of AIO_WAIT_WHILE(), which could deadlock otherwise. */
727 aio_context_release(outer_ctx);
729 /* Other jobs are effectively cancelled by us, set the status for
730 * them; this job, however, may or may not be cancelled, depending
731 * on the caller, so leave it. */
732 QLIST_FOREACH(other_job, &txn->jobs, txn_list) {
733 if (other_job != job) {
734 ctx = other_job->aio_context;
735 aio_context_acquire(ctx);
736 job_cancel_async(other_job, false);
737 aio_context_release(ctx);
740 while (!QLIST_EMPTY(&txn->jobs)) {
741 other_job = QLIST_FIRST(&txn->jobs);
742 ctx = other_job->aio_context;
743 aio_context_acquire(ctx);
744 if (!job_is_completed(other_job)) {
745 assert(job_is_cancelled(other_job));
746 job_finish_sync(other_job, NULL, NULL);
748 job_finalize_single(other_job);
749 aio_context_release(ctx);
752 aio_context_acquire(outer_ctx);
757 static int job_prepare(Job *job)
759 if (job->ret == 0 && job->driver->prepare) {
760 job->ret = job->driver->prepare(job);
766 static int job_needs_finalize(Job *job)
768 return !job->auto_finalize;
771 static void job_do_finalize(Job *job)
774 assert(job && job->txn);
776 /* prepare the transaction to complete */
777 rc = job_txn_apply(job->txn, job_prepare);
779 job_completed_txn_abort(job);
781 job_txn_apply(job->txn, job_finalize_single);
785 void job_finalize(Job *job, Error **errp)
787 assert(job && job->id);
788 if (job_apply_verb(job, JOB_VERB_FINALIZE, errp)) {
791 job_do_finalize(job);
794 static int job_transition_to_pending(Job *job)
796 job_state_transition(job, JOB_STATUS_PENDING);
797 if (!job->auto_finalize) {
798 job_event_pending(job);
803 void job_transition_to_ready(Job *job)
805 job_state_transition(job, JOB_STATUS_READY);
806 job_event_ready(job);
809 static void job_completed_txn_success(Job *job)
811 JobTxn *txn = job->txn;
814 job_state_transition(job, JOB_STATUS_WAITING);
817 * Successful completion, see if there are other running jobs in this
820 QLIST_FOREACH(other_job, &txn->jobs, txn_list) {
821 if (!job_is_completed(other_job)) {
824 assert(other_job->ret == 0);
827 job_txn_apply(txn, job_transition_to_pending);
829 /* If no jobs need manual finalization, automatically do so */
830 if (job_txn_apply(txn, job_needs_finalize) == 0) {
831 job_do_finalize(job);
835 static void job_completed(Job *job)
837 assert(job && job->txn && !job_is_completed(job));
840 trace_job_completed(job, job->ret);
842 job_completed_txn_abort(job);
844 job_completed_txn_success(job);
848 /** Useful only as a type shim for aio_bh_schedule_oneshot. */
849 static void job_exit(void *opaque)
851 Job *job = (Job *)opaque;
852 AioContext *ctx = job->aio_context;
854 aio_context_acquire(ctx);
856 /* This is a lie, we're not quiescent, but still doing the completion
857 * callbacks. However, completion callbacks tend to involve operations that
858 * drain block nodes, and if .drained_poll still returned true, we would
865 aio_context_release(ctx);
869 * All jobs must allow a pause point before entering their job proper. This
870 * ensures that jobs can be paused prior to being started, then resumed later.
872 static void coroutine_fn job_co_entry(void *opaque)
876 assert(job && job->driver && job->driver->run);
877 job_pause_point(job);
878 job->ret = job->driver->run(job, &job->err);
879 job->deferred_to_main_loop = true;
881 aio_bh_schedule_oneshot(qemu_get_aio_context(), job_exit, job);
884 void job_start(Job *job)
886 assert(job && !job_started(job) && job->paused &&
887 job->driver && job->driver->run);
888 job->co = qemu_coroutine_create(job_co_entry, job);
892 job_state_transition(job, JOB_STATUS_RUNNING);
893 aio_co_enter(job->aio_context, job->co);
896 void job_cancel(Job *job, bool force)
898 if (job->status == JOB_STATUS_CONCLUDED) {
902 job_cancel_async(job, force);
903 if (!job_started(job)) {
905 } else if (job->deferred_to_main_loop) {
906 job_completed_txn_abort(job);
912 void job_user_cancel(Job *job, bool force, Error **errp)
914 if (job_apply_verb(job, JOB_VERB_CANCEL, errp)) {
917 job_cancel(job, force);
920 /* A wrapper around job_cancel() taking an Error ** parameter so it may be
921 * used with job_finish_sync() without the need for (rather nasty) function
922 * pointer casts there. */
923 static void job_cancel_err(Job *job, Error **errp)
925 job_cancel(job, false);
928 int job_cancel_sync(Job *job)
930 return job_finish_sync(job, &job_cancel_err, NULL);
933 void job_cancel_sync_all(void)
936 AioContext *aio_context;
938 while ((job = job_next(NULL))) {
939 aio_context = job->aio_context;
940 aio_context_acquire(aio_context);
941 job_cancel_sync(job);
942 aio_context_release(aio_context);
946 int job_complete_sync(Job *job, Error **errp)
948 return job_finish_sync(job, job_complete, errp);
951 void job_complete(Job *job, Error **errp)
953 /* Should not be reachable via external interface for internal jobs */
955 if (job_apply_verb(job, JOB_VERB_COMPLETE, errp)) {
958 if (job->pause_count || job_is_cancelled(job) || !job->driver->complete) {
959 error_setg(errp, "The active block job '%s' cannot be completed",
964 job->driver->complete(job, errp);
967 int job_finish_sync(Job *job, void (*finish)(Job *, Error **errp), Error **errp)
969 Error *local_err = NULL;
975 finish(job, &local_err);
978 error_propagate(errp, local_err);
983 AIO_WAIT_WHILE(job->aio_context,
984 (job_enter(job), !job_is_completed(job)));
986 ret = (job_is_cancelled(job) && job->ret == 0) ? -ECANCELED : job->ret;