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"
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 *), bool lock)
145 QLIST_FOREACH_SAFE(job, &txn->jobs, txn_list, next) {
147 ctx = job->aio_context;
148 aio_context_acquire(ctx);
152 aio_context_release(ctx);
161 bool job_is_internal(Job *job)
163 return (job->id == NULL);
166 static void job_state_transition(Job *job, JobStatus s1)
168 JobStatus s0 = job->status;
169 assert(s1 >= 0 && s1 <= JOB_STATUS__MAX);
170 trace_job_state_transition(job, job->ret,
171 JobSTT[s0][s1] ? "allowed" : "disallowed",
172 JobStatus_str(s0), JobStatus_str(s1));
173 assert(JobSTT[s0][s1]);
176 if (!job_is_internal(job) && s1 != s0) {
177 qapi_event_send_job_status_change(job->id, job->status, &error_abort);
181 int job_apply_verb(Job *job, JobVerb verb, Error **errp)
183 JobStatus s0 = job->status;
184 assert(verb >= 0 && verb <= JOB_VERB__MAX);
185 trace_job_apply_verb(job, JobStatus_str(s0), JobVerb_str(verb),
186 JobVerbTable[verb][s0] ? "allowed" : "prohibited");
187 if (JobVerbTable[verb][s0]) {
190 error_setg(errp, "Job '%s' in state '%s' cannot accept command verb '%s'",
191 job->id, JobStatus_str(s0), JobVerb_str(verb));
195 JobType job_type(const Job *job)
197 return job->driver->job_type;
200 const char *job_type_str(const Job *job)
202 return JobType_str(job_type(job));
205 bool job_is_cancelled(Job *job)
207 return job->cancelled;
210 bool job_is_ready(Job *job)
212 switch (job->status) {
213 case JOB_STATUS_UNDEFINED:
214 case JOB_STATUS_CREATED:
215 case JOB_STATUS_RUNNING:
216 case JOB_STATUS_PAUSED:
217 case JOB_STATUS_WAITING:
218 case JOB_STATUS_PENDING:
219 case JOB_STATUS_ABORTING:
220 case JOB_STATUS_CONCLUDED:
221 case JOB_STATUS_NULL:
223 case JOB_STATUS_READY:
224 case JOB_STATUS_STANDBY:
227 g_assert_not_reached();
232 bool job_is_completed(Job *job)
234 switch (job->status) {
235 case JOB_STATUS_UNDEFINED:
236 case JOB_STATUS_CREATED:
237 case JOB_STATUS_RUNNING:
238 case JOB_STATUS_PAUSED:
239 case JOB_STATUS_READY:
240 case JOB_STATUS_STANDBY:
242 case JOB_STATUS_WAITING:
243 case JOB_STATUS_PENDING:
244 case JOB_STATUS_ABORTING:
245 case JOB_STATUS_CONCLUDED:
246 case JOB_STATUS_NULL:
249 g_assert_not_reached();
254 static bool job_started(Job *job)
259 static bool job_should_pause(Job *job)
261 return job->pause_count > 0;
264 Job *job_next(Job *job)
267 return QLIST_FIRST(&jobs);
269 return QLIST_NEXT(job, job_list);
272 Job *job_get(const char *id)
276 QLIST_FOREACH(job, &jobs, job_list) {
277 if (job->id && !strcmp(id, job->id)) {
285 static void job_sleep_timer_cb(void *opaque)
292 void *job_create(const char *job_id, const JobDriver *driver, JobTxn *txn,
293 AioContext *ctx, int flags, BlockCompletionFunc *cb,
294 void *opaque, Error **errp)
299 if (flags & JOB_INTERNAL) {
300 error_setg(errp, "Cannot specify job ID for internal job");
303 if (!id_wellformed(job_id)) {
304 error_setg(errp, "Invalid job ID '%s'", job_id);
307 if (job_get(job_id)) {
308 error_setg(errp, "Job ID '%s' already in use", job_id);
311 } else if (!(flags & JOB_INTERNAL)) {
312 error_setg(errp, "An explicit job ID is required");
316 job = g_malloc0(driver->instance_size);
317 job->driver = driver;
318 job->id = g_strdup(job_id);
320 job->aio_context = ctx;
323 job->pause_count = 1;
324 job->auto_finalize = !(flags & JOB_MANUAL_FINALIZE);
325 job->auto_dismiss = !(flags & JOB_MANUAL_DISMISS);
327 job->opaque = opaque;
329 notifier_list_init(&job->on_finalize_cancelled);
330 notifier_list_init(&job->on_finalize_completed);
331 notifier_list_init(&job->on_pending);
332 notifier_list_init(&job->on_ready);
334 job_state_transition(job, JOB_STATUS_CREATED);
335 aio_timer_init(qemu_get_aio_context(), &job->sleep_timer,
336 QEMU_CLOCK_REALTIME, SCALE_NS,
337 job_sleep_timer_cb, job);
339 QLIST_INSERT_HEAD(&jobs, job, job_list);
341 /* Single jobs are modeled as single-job transactions for sake of
342 * consolidating the job management logic */
345 job_txn_add_job(txn, job);
348 job_txn_add_job(txn, job);
354 void job_ref(Job *job)
359 void job_unref(Job *job)
361 if (--job->refcnt == 0) {
362 assert(job->status == JOB_STATUS_NULL);
363 assert(!timer_pending(&job->sleep_timer));
366 if (job->driver->free) {
367 job->driver->free(job);
370 QLIST_REMOVE(job, job_list);
378 void job_progress_update(Job *job, uint64_t done)
380 job->progress_current += done;
383 void job_progress_set_remaining(Job *job, uint64_t remaining)
385 job->progress_total = job->progress_current + remaining;
388 void job_event_cancelled(Job *job)
390 notifier_list_notify(&job->on_finalize_cancelled, job);
393 void job_event_completed(Job *job)
395 notifier_list_notify(&job->on_finalize_completed, job);
398 static void job_event_pending(Job *job)
400 notifier_list_notify(&job->on_pending, job);
403 static void job_event_ready(Job *job)
405 notifier_list_notify(&job->on_ready, job);
408 void job_enter_cond(Job *job, bool(*fn)(Job *job))
410 if (!job_started(job)) {
413 if (job->deferred_to_main_loop) {
423 if (fn && !fn(job)) {
428 assert(!job->deferred_to_main_loop);
429 timer_del(&job->sleep_timer);
432 aio_co_wake(job->co);
435 void job_enter(Job *job)
437 job_enter_cond(job, NULL);
440 /* Yield, and schedule a timer to reenter the coroutine after @ns nanoseconds.
441 * Reentering the job coroutine with job_enter() before the timer has expired
442 * is allowed and cancels the timer.
444 * If @ns is (uint64_t) -1, no timer is scheduled and job_enter() must be
445 * called explicitly. */
446 static void coroutine_fn job_do_yield(Job *job, uint64_t ns)
450 timer_mod(&job->sleep_timer, ns);
454 qemu_coroutine_yield();
456 /* Set by job_enter_cond() before re-entering the coroutine. */
460 void coroutine_fn job_pause_point(Job *job)
462 assert(job && job_started(job));
464 if (!job_should_pause(job)) {
467 if (job_is_cancelled(job)) {
471 if (job->driver->pause) {
472 job->driver->pause(job);
475 if (job_should_pause(job) && !job_is_cancelled(job)) {
476 JobStatus status = job->status;
477 job_state_transition(job, status == JOB_STATUS_READY
479 : JOB_STATUS_PAUSED);
481 job_do_yield(job, -1);
483 job_state_transition(job, status);
486 if (job->driver->resume) {
487 job->driver->resume(job);
491 void job_yield(Job *job)
495 /* Check cancellation *before* setting busy = false, too! */
496 if (job_is_cancelled(job)) {
500 if (!job_should_pause(job)) {
501 job_do_yield(job, -1);
504 job_pause_point(job);
507 void coroutine_fn job_sleep_ns(Job *job, int64_t ns)
511 /* Check cancellation *before* setting busy = false, too! */
512 if (job_is_cancelled(job)) {
516 if (!job_should_pause(job)) {
517 job_do_yield(job, qemu_clock_get_ns(QEMU_CLOCK_REALTIME) + ns);
520 job_pause_point(job);
523 void job_drain(Job *job)
525 /* If job is !busy this kicks it into the next pause point. */
528 if (job->driver->drain) {
529 job->driver->drain(job);
535 * All jobs must allow a pause point before entering their job proper. This
536 * ensures that jobs can be paused prior to being started, then resumed later.
538 static void coroutine_fn job_co_entry(void *opaque)
542 assert(job && job->driver && job->driver->start);
543 job_pause_point(job);
544 job->driver->start(job);
548 void job_start(Job *job)
550 assert(job && !job_started(job) && job->paused &&
551 job->driver && job->driver->start);
552 job->co = qemu_coroutine_create(job_co_entry, job);
556 job_state_transition(job, JOB_STATUS_RUNNING);
557 aio_co_enter(job->aio_context, job->co);
560 /* Assumes the block_job_mutex is held */
561 static bool job_timer_not_pending(Job *job)
563 return !timer_pending(&job->sleep_timer);
566 void job_pause(Job *job)
571 void job_resume(Job *job)
573 assert(job->pause_count > 0);
575 if (job->pause_count) {
579 /* kick only if no timer is pending */
580 job_enter_cond(job, job_timer_not_pending);
583 void job_user_pause(Job *job, Error **errp)
585 if (job_apply_verb(job, JOB_VERB_PAUSE, errp)) {
588 if (job->user_paused) {
589 error_setg(errp, "Job is already paused");
592 job->user_paused = true;
596 bool job_user_paused(Job *job)
598 return job->user_paused;
601 void job_user_resume(Job *job, Error **errp)
604 if (!job->user_paused || job->pause_count <= 0) {
605 error_setg(errp, "Can't resume a job that was not paused");
608 if (job_apply_verb(job, JOB_VERB_RESUME, errp)) {
611 if (job->driver->user_resume) {
612 job->driver->user_resume(job);
614 job->user_paused = false;
618 static void job_do_dismiss(Job *job)
623 job->deferred_to_main_loop = true;
625 job_txn_del_job(job);
627 job_state_transition(job, JOB_STATUS_NULL);
631 void job_dismiss(Job **jobptr, Error **errp)
634 /* similarly to _complete, this is QMP-interface only. */
636 if (job_apply_verb(job, JOB_VERB_DISMISS, errp)) {
644 void job_early_fail(Job *job)
646 assert(job->status == JOB_STATUS_CREATED);
650 static void job_conclude(Job *job)
652 job_state_transition(job, JOB_STATUS_CONCLUDED);
653 if (job->auto_dismiss || !job_started(job)) {
658 static void job_update_rc(Job *job)
660 if (!job->ret && job_is_cancelled(job)) {
661 job->ret = -ECANCELED;
665 job->error = g_strdup(strerror(-job->ret));
667 job_state_transition(job, JOB_STATUS_ABORTING);
671 static void job_commit(Job *job)
674 if (job->driver->commit) {
675 job->driver->commit(job);
679 static void job_abort(Job *job)
682 if (job->driver->abort) {
683 job->driver->abort(job);
687 static void job_clean(Job *job)
689 if (job->driver->clean) {
690 job->driver->clean(job);
694 static int job_finalize_single(Job *job)
696 assert(job_is_completed(job));
698 /* Ensure abort is called for late-transactional failures */
709 job->cb(job->opaque, job->ret);
712 /* Emit events only if we actually started */
713 if (job_started(job)) {
714 if (job_is_cancelled(job)) {
715 job_event_cancelled(job);
717 job_event_completed(job);
721 job_txn_del_job(job);
726 static void job_cancel_async(Job *job, bool force)
728 if (job->user_paused) {
729 /* Do not call job_enter here, the caller will handle it. */
730 job->user_paused = false;
731 if (job->driver->user_resume) {
732 job->driver->user_resume(job);
734 assert(job->pause_count > 0);
737 job->cancelled = true;
738 /* To prevent 'force == false' overriding a previous 'force == true' */
739 job->force_cancel |= force;
742 static void job_completed_txn_abort(Job *job)
745 JobTxn *txn = job->txn;
750 * We are cancelled by another job, which will handle everything.
754 txn->aborting = true;
757 /* We are the first failed job. Cancel other jobs. */
758 QLIST_FOREACH(other_job, &txn->jobs, txn_list) {
759 ctx = other_job->aio_context;
760 aio_context_acquire(ctx);
763 /* Other jobs are effectively cancelled by us, set the status for
764 * them; this job, however, may or may not be cancelled, depending
765 * on the caller, so leave it. */
766 QLIST_FOREACH(other_job, &txn->jobs, txn_list) {
767 if (other_job != job) {
768 job_cancel_async(other_job, false);
771 while (!QLIST_EMPTY(&txn->jobs)) {
772 other_job = QLIST_FIRST(&txn->jobs);
773 ctx = other_job->aio_context;
774 if (!job_is_completed(other_job)) {
775 assert(job_is_cancelled(other_job));
776 job_finish_sync(other_job, NULL, NULL);
778 job_finalize_single(other_job);
779 aio_context_release(ctx);
785 static int job_prepare(Job *job)
787 if (job->ret == 0 && job->driver->prepare) {
788 job->ret = job->driver->prepare(job);
794 static int job_needs_finalize(Job *job)
796 return !job->auto_finalize;
799 static void job_do_finalize(Job *job)
802 assert(job && job->txn);
804 /* prepare the transaction to complete */
805 rc = job_txn_apply(job->txn, job_prepare, true);
807 job_completed_txn_abort(job);
809 job_txn_apply(job->txn, job_finalize_single, true);
813 void job_finalize(Job *job, Error **errp)
815 assert(job && job->id);
816 if (job_apply_verb(job, JOB_VERB_FINALIZE, errp)) {
819 job_do_finalize(job);
822 static int job_transition_to_pending(Job *job)
824 job_state_transition(job, JOB_STATUS_PENDING);
825 if (!job->auto_finalize) {
826 job_event_pending(job);
831 void job_transition_to_ready(Job *job)
833 job_state_transition(job, JOB_STATUS_READY);
834 job_event_ready(job);
837 static void job_completed_txn_success(Job *job)
839 JobTxn *txn = job->txn;
842 job_state_transition(job, JOB_STATUS_WAITING);
845 * Successful completion, see if there are other running jobs in this
848 QLIST_FOREACH(other_job, &txn->jobs, txn_list) {
849 if (!job_is_completed(other_job)) {
852 assert(other_job->ret == 0);
855 job_txn_apply(txn, job_transition_to_pending, false);
857 /* If no jobs need manual finalization, automatically do so */
858 if (job_txn_apply(txn, job_needs_finalize, false) == 0) {
859 job_do_finalize(job);
863 void job_completed(Job *job, int ret, Error *error)
865 assert(job && job->txn && !job_is_completed(job));
869 assert(job->ret < 0);
870 job->error = g_strdup(error_get_pretty(error));
875 trace_job_completed(job, ret, job->ret);
877 job_completed_txn_abort(job);
879 job_completed_txn_success(job);
883 void job_cancel(Job *job, bool force)
885 if (job->status == JOB_STATUS_CONCLUDED) {
889 job_cancel_async(job, force);
890 if (!job_started(job)) {
891 job_completed(job, -ECANCELED, NULL);
892 } else if (job->deferred_to_main_loop) {
893 job_completed_txn_abort(job);
899 void job_user_cancel(Job *job, bool force, Error **errp)
901 if (job_apply_verb(job, JOB_VERB_CANCEL, errp)) {
904 job_cancel(job, force);
907 /* A wrapper around job_cancel() taking an Error ** parameter so it may be
908 * used with job_finish_sync() without the need for (rather nasty) function
909 * pointer casts there. */
910 static void job_cancel_err(Job *job, Error **errp)
912 job_cancel(job, false);
915 int job_cancel_sync(Job *job)
917 return job_finish_sync(job, &job_cancel_err, NULL);
920 void job_cancel_sync_all(void)
923 AioContext *aio_context;
925 while ((job = job_next(NULL))) {
926 aio_context = job->aio_context;
927 aio_context_acquire(aio_context);
928 job_cancel_sync(job);
929 aio_context_release(aio_context);
933 int job_complete_sync(Job *job, Error **errp)
935 return job_finish_sync(job, job_complete, errp);
938 void job_complete(Job *job, Error **errp)
940 /* Should not be reachable via external interface for internal jobs */
942 if (job_apply_verb(job, JOB_VERB_COMPLETE, errp)) {
945 if (job->pause_count || job_is_cancelled(job) || !job->driver->complete) {
946 error_setg(errp, "The active block job '%s' cannot be completed",
951 job->driver->complete(job, errp);
957 JobDeferToMainLoopFn *fn;
959 } JobDeferToMainLoopData;
961 static void job_defer_to_main_loop_bh(void *opaque)
963 JobDeferToMainLoopData *data = opaque;
964 Job *job = data->job;
965 AioContext *aio_context = job->aio_context;
967 aio_context_acquire(aio_context);
968 data->fn(data->job, data->opaque);
969 aio_context_release(aio_context);
974 void job_defer_to_main_loop(Job *job, JobDeferToMainLoopFn *fn, void *opaque)
976 JobDeferToMainLoopData *data = g_malloc(sizeof(*data));
979 data->opaque = opaque;
980 job->deferred_to_main_loop = true;
982 aio_bh_schedule_oneshot(qemu_get_aio_context(),
983 job_defer_to_main_loop_bh, data);
986 int job_finish_sync(Job *job, void (*finish)(Job *, Error **errp), Error **errp)
988 Error *local_err = NULL;
994 finish(job, &local_err);
997 error_propagate(errp, local_err);
1001 /* job_drain calls job_enter, and it should be enough to induce progress
1002 * until the job completes or moves to the main thread. */
1003 while (!job->deferred_to_main_loop && !job_is_completed(job)) {
1006 while (!job_is_completed(job)) {
1007 aio_poll(qemu_get_aio_context(), true);
1009 ret = (job_is_cancelled(job) && job->ret == 0) ? -ECANCELED : job->ret;