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);
160 static void job_state_transition(Job *job, JobStatus s1)
162 JobStatus s0 = job->status;
163 assert(s1 >= 0 && s1 <= JOB_STATUS__MAX);
164 trace_job_state_transition(job, job->ret,
165 JobSTT[s0][s1] ? "allowed" : "disallowed",
166 JobStatus_str(s0), JobStatus_str(s1));
167 assert(JobSTT[s0][s1]);
171 int job_apply_verb(Job *job, JobVerb verb, Error **errp)
173 JobStatus s0 = job->status;
174 assert(verb >= 0 && verb <= JOB_VERB__MAX);
175 trace_job_apply_verb(job, JobStatus_str(s0), JobVerb_str(verb),
176 JobVerbTable[verb][s0] ? "allowed" : "prohibited");
177 if (JobVerbTable[verb][s0]) {
180 error_setg(errp, "Job '%s' in state '%s' cannot accept command verb '%s'",
181 job->id, JobStatus_str(s0), JobVerb_str(verb));
185 JobType job_type(const Job *job)
187 return job->driver->job_type;
190 const char *job_type_str(const Job *job)
192 return JobType_str(job_type(job));
195 bool job_is_cancelled(Job *job)
197 return job->cancelled;
200 bool job_is_ready(Job *job)
202 switch (job->status) {
203 case JOB_STATUS_UNDEFINED:
204 case JOB_STATUS_CREATED:
205 case JOB_STATUS_RUNNING:
206 case JOB_STATUS_PAUSED:
207 case JOB_STATUS_WAITING:
208 case JOB_STATUS_PENDING:
209 case JOB_STATUS_ABORTING:
210 case JOB_STATUS_CONCLUDED:
211 case JOB_STATUS_NULL:
213 case JOB_STATUS_READY:
214 case JOB_STATUS_STANDBY:
217 g_assert_not_reached();
222 bool job_is_completed(Job *job)
224 switch (job->status) {
225 case JOB_STATUS_UNDEFINED:
226 case JOB_STATUS_CREATED:
227 case JOB_STATUS_RUNNING:
228 case JOB_STATUS_PAUSED:
229 case JOB_STATUS_READY:
230 case JOB_STATUS_STANDBY:
232 case JOB_STATUS_WAITING:
233 case JOB_STATUS_PENDING:
234 case JOB_STATUS_ABORTING:
235 case JOB_STATUS_CONCLUDED:
236 case JOB_STATUS_NULL:
239 g_assert_not_reached();
244 static bool job_started(Job *job)
249 static bool job_should_pause(Job *job)
251 return job->pause_count > 0;
254 Job *job_next(Job *job)
257 return QLIST_FIRST(&jobs);
259 return QLIST_NEXT(job, job_list);
262 Job *job_get(const char *id)
266 QLIST_FOREACH(job, &jobs, job_list) {
267 if (job->id && !strcmp(id, job->id)) {
275 static void job_sleep_timer_cb(void *opaque)
282 void *job_create(const char *job_id, const JobDriver *driver, JobTxn *txn,
283 AioContext *ctx, int flags, BlockCompletionFunc *cb,
284 void *opaque, Error **errp)
289 if (flags & JOB_INTERNAL) {
290 error_setg(errp, "Cannot specify job ID for internal job");
293 if (!id_wellformed(job_id)) {
294 error_setg(errp, "Invalid job ID '%s'", job_id);
297 if (job_get(job_id)) {
298 error_setg(errp, "Job ID '%s' already in use", job_id);
301 } else if (!(flags & JOB_INTERNAL)) {
302 error_setg(errp, "An explicit job ID is required");
306 job = g_malloc0(driver->instance_size);
307 job->driver = driver;
308 job->id = g_strdup(job_id);
310 job->aio_context = ctx;
313 job->pause_count = 1;
314 job->auto_finalize = !(flags & JOB_MANUAL_FINALIZE);
315 job->auto_dismiss = !(flags & JOB_MANUAL_DISMISS);
317 job->opaque = opaque;
319 notifier_list_init(&job->on_finalize_cancelled);
320 notifier_list_init(&job->on_finalize_completed);
321 notifier_list_init(&job->on_pending);
322 notifier_list_init(&job->on_ready);
324 job_state_transition(job, JOB_STATUS_CREATED);
325 aio_timer_init(qemu_get_aio_context(), &job->sleep_timer,
326 QEMU_CLOCK_REALTIME, SCALE_NS,
327 job_sleep_timer_cb, job);
329 QLIST_INSERT_HEAD(&jobs, job, job_list);
331 /* Single jobs are modeled as single-job transactions for sake of
332 * consolidating the job management logic */
335 job_txn_add_job(txn, job);
338 job_txn_add_job(txn, job);
344 void job_ref(Job *job)
349 void job_unref(Job *job)
351 if (--job->refcnt == 0) {
352 assert(job->status == JOB_STATUS_NULL);
353 assert(!timer_pending(&job->sleep_timer));
356 if (job->driver->free) {
357 job->driver->free(job);
360 QLIST_REMOVE(job, job_list);
367 void job_progress_update(Job *job, uint64_t done)
369 job->progress_current += done;
372 void job_progress_set_remaining(Job *job, uint64_t remaining)
374 job->progress_total = job->progress_current + remaining;
377 void job_event_cancelled(Job *job)
379 notifier_list_notify(&job->on_finalize_cancelled, job);
382 void job_event_completed(Job *job)
384 notifier_list_notify(&job->on_finalize_completed, job);
387 static void job_event_pending(Job *job)
389 notifier_list_notify(&job->on_pending, job);
392 static void job_event_ready(Job *job)
394 notifier_list_notify(&job->on_ready, job);
397 void job_enter_cond(Job *job, bool(*fn)(Job *job))
399 if (!job_started(job)) {
402 if (job->deferred_to_main_loop) {
412 if (fn && !fn(job)) {
417 assert(!job->deferred_to_main_loop);
418 timer_del(&job->sleep_timer);
421 aio_co_wake(job->co);
424 void job_enter(Job *job)
426 job_enter_cond(job, NULL);
429 /* Yield, and schedule a timer to reenter the coroutine after @ns nanoseconds.
430 * Reentering the job coroutine with job_enter() before the timer has expired
431 * is allowed and cancels the timer.
433 * If @ns is (uint64_t) -1, no timer is scheduled and job_enter() must be
434 * called explicitly. */
435 static void coroutine_fn job_do_yield(Job *job, uint64_t ns)
439 timer_mod(&job->sleep_timer, ns);
443 qemu_coroutine_yield();
445 /* Set by job_enter_cond() before re-entering the coroutine. */
449 void coroutine_fn job_pause_point(Job *job)
451 assert(job && job_started(job));
453 if (!job_should_pause(job)) {
456 if (job_is_cancelled(job)) {
460 if (job->driver->pause) {
461 job->driver->pause(job);
464 if (job_should_pause(job) && !job_is_cancelled(job)) {
465 JobStatus status = job->status;
466 job_state_transition(job, status == JOB_STATUS_READY
468 : JOB_STATUS_PAUSED);
470 job_do_yield(job, -1);
472 job_state_transition(job, status);
475 if (job->driver->resume) {
476 job->driver->resume(job);
480 void job_yield(Job *job)
484 /* Check cancellation *before* setting busy = false, too! */
485 if (job_is_cancelled(job)) {
489 if (!job_should_pause(job)) {
490 job_do_yield(job, -1);
493 job_pause_point(job);
496 void coroutine_fn job_sleep_ns(Job *job, int64_t ns)
500 /* Check cancellation *before* setting busy = false, too! */
501 if (job_is_cancelled(job)) {
505 if (!job_should_pause(job)) {
506 job_do_yield(job, qemu_clock_get_ns(QEMU_CLOCK_REALTIME) + ns);
509 job_pause_point(job);
512 void job_drain(Job *job)
514 /* If job is !busy this kicks it into the next pause point. */
517 if (job->driver->drain) {
518 job->driver->drain(job);
524 * All jobs must allow a pause point before entering their job proper. This
525 * ensures that jobs can be paused prior to being started, then resumed later.
527 static void coroutine_fn job_co_entry(void *opaque)
531 assert(job && job->driver && job->driver->start);
532 job_pause_point(job);
533 job->driver->start(job);
537 void job_start(Job *job)
539 assert(job && !job_started(job) && job->paused &&
540 job->driver && job->driver->start);
541 job->co = qemu_coroutine_create(job_co_entry, job);
545 job_state_transition(job, JOB_STATUS_RUNNING);
546 aio_co_enter(job->aio_context, job->co);
549 /* Assumes the block_job_mutex is held */
550 static bool job_timer_not_pending(Job *job)
552 return !timer_pending(&job->sleep_timer);
555 void job_pause(Job *job)
560 void job_resume(Job *job)
562 assert(job->pause_count > 0);
564 if (job->pause_count) {
568 /* kick only if no timer is pending */
569 job_enter_cond(job, job_timer_not_pending);
572 void job_user_pause(Job *job, Error **errp)
574 if (job_apply_verb(job, JOB_VERB_PAUSE, errp)) {
577 if (job->user_paused) {
578 error_setg(errp, "Job is already paused");
581 job->user_paused = true;
585 bool job_user_paused(Job *job)
587 return job->user_paused;
590 void job_user_resume(Job *job, Error **errp)
593 if (!job->user_paused || job->pause_count <= 0) {
594 error_setg(errp, "Can't resume a job that was not paused");
597 if (job_apply_verb(job, JOB_VERB_RESUME, errp)) {
600 if (job->driver->user_resume) {
601 job->driver->user_resume(job);
603 job->user_paused = false;
607 static void job_do_dismiss(Job *job)
612 job->deferred_to_main_loop = true;
614 job_txn_del_job(job);
616 job_state_transition(job, JOB_STATUS_NULL);
620 void job_dismiss(Job **jobptr, Error **errp)
623 /* similarly to _complete, this is QMP-interface only. */
625 if (job_apply_verb(job, JOB_VERB_DISMISS, errp)) {
633 void job_early_fail(Job *job)
635 assert(job->status == JOB_STATUS_CREATED);
639 static void job_conclude(Job *job)
641 job_state_transition(job, JOB_STATUS_CONCLUDED);
642 if (job->auto_dismiss || !job_started(job)) {
647 static void job_update_rc(Job *job)
649 if (!job->ret && job_is_cancelled(job)) {
650 job->ret = -ECANCELED;
653 job_state_transition(job, JOB_STATUS_ABORTING);
657 static void job_commit(Job *job)
660 if (job->driver->commit) {
661 job->driver->commit(job);
665 static void job_abort(Job *job)
668 if (job->driver->abort) {
669 job->driver->abort(job);
673 static void job_clean(Job *job)
675 if (job->driver->clean) {
676 job->driver->clean(job);
680 static int job_finalize_single(Job *job)
682 assert(job_is_completed(job));
684 /* Ensure abort is called for late-transactional failures */
695 job->cb(job->opaque, job->ret);
698 /* Emit events only if we actually started */
699 if (job_started(job)) {
700 if (job_is_cancelled(job)) {
701 job_event_cancelled(job);
703 job_event_completed(job);
707 job_txn_del_job(job);
712 static void job_cancel_async(Job *job, bool force)
714 if (job->user_paused) {
715 /* Do not call job_enter here, the caller will handle it. */
716 job->user_paused = false;
717 if (job->driver->user_resume) {
718 job->driver->user_resume(job);
720 assert(job->pause_count > 0);
723 job->cancelled = true;
724 /* To prevent 'force == false' overriding a previous 'force == true' */
725 job->force_cancel |= force;
728 static void job_completed_txn_abort(Job *job)
731 JobTxn *txn = job->txn;
736 * We are cancelled by another job, which will handle everything.
740 txn->aborting = true;
743 /* We are the first failed job. Cancel other jobs. */
744 QLIST_FOREACH(other_job, &txn->jobs, txn_list) {
745 ctx = other_job->aio_context;
746 aio_context_acquire(ctx);
749 /* Other jobs are effectively cancelled by us, set the status for
750 * them; this job, however, may or may not be cancelled, depending
751 * on the caller, so leave it. */
752 QLIST_FOREACH(other_job, &txn->jobs, txn_list) {
753 if (other_job != job) {
754 job_cancel_async(other_job, false);
757 while (!QLIST_EMPTY(&txn->jobs)) {
758 other_job = QLIST_FIRST(&txn->jobs);
759 ctx = other_job->aio_context;
760 if (!job_is_completed(other_job)) {
761 assert(job_is_cancelled(other_job));
762 job_finish_sync(other_job, NULL, NULL);
764 job_finalize_single(other_job);
765 aio_context_release(ctx);
771 static int job_prepare(Job *job)
773 if (job->ret == 0 && job->driver->prepare) {
774 job->ret = job->driver->prepare(job);
779 static int job_needs_finalize(Job *job)
781 return !job->auto_finalize;
784 static void job_do_finalize(Job *job)
787 assert(job && job->txn);
789 /* prepare the transaction to complete */
790 rc = job_txn_apply(job->txn, job_prepare, true);
792 job_completed_txn_abort(job);
794 job_txn_apply(job->txn, job_finalize_single, true);
798 void job_finalize(Job *job, Error **errp)
800 assert(job && job->id);
801 if (job_apply_verb(job, JOB_VERB_FINALIZE, errp)) {
804 job_do_finalize(job);
807 static int job_transition_to_pending(Job *job)
809 job_state_transition(job, JOB_STATUS_PENDING);
810 if (!job->auto_finalize) {
811 job_event_pending(job);
816 void job_transition_to_ready(Job *job)
818 job_state_transition(job, JOB_STATUS_READY);
819 job_event_ready(job);
822 static void job_completed_txn_success(Job *job)
824 JobTxn *txn = job->txn;
827 job_state_transition(job, JOB_STATUS_WAITING);
830 * Successful completion, see if there are other running jobs in this
833 QLIST_FOREACH(other_job, &txn->jobs, txn_list) {
834 if (!job_is_completed(other_job)) {
837 assert(other_job->ret == 0);
840 job_txn_apply(txn, job_transition_to_pending, false);
842 /* If no jobs need manual finalization, automatically do so */
843 if (job_txn_apply(txn, job_needs_finalize, false) == 0) {
844 job_do_finalize(job);
848 void job_completed(Job *job, int ret)
850 assert(job && job->txn && !job_is_completed(job));
853 trace_job_completed(job, ret, job->ret);
855 job_completed_txn_abort(job);
857 job_completed_txn_success(job);
861 void job_cancel(Job *job, bool force)
863 if (job->status == JOB_STATUS_CONCLUDED) {
867 job_cancel_async(job, force);
868 if (!job_started(job)) {
869 job_completed(job, -ECANCELED);
870 } else if (job->deferred_to_main_loop) {
871 job_completed_txn_abort(job);
877 void job_user_cancel(Job *job, bool force, Error **errp)
879 if (job_apply_verb(job, JOB_VERB_CANCEL, errp)) {
882 job_cancel(job, force);
885 /* A wrapper around job_cancel() taking an Error ** parameter so it may be
886 * used with job_finish_sync() without the need for (rather nasty) function
887 * pointer casts there. */
888 static void job_cancel_err(Job *job, Error **errp)
890 job_cancel(job, false);
893 int job_cancel_sync(Job *job)
895 return job_finish_sync(job, &job_cancel_err, NULL);
898 void job_cancel_sync_all(void)
901 AioContext *aio_context;
903 while ((job = job_next(NULL))) {
904 aio_context = job->aio_context;
905 aio_context_acquire(aio_context);
906 job_cancel_sync(job);
907 aio_context_release(aio_context);
911 int job_complete_sync(Job *job, Error **errp)
913 return job_finish_sync(job, job_complete, errp);
916 void job_complete(Job *job, Error **errp)
918 /* Should not be reachable via external interface for internal jobs */
920 if (job_apply_verb(job, JOB_VERB_COMPLETE, errp)) {
923 if (job->pause_count || job_is_cancelled(job) || !job->driver->complete) {
924 error_setg(errp, "The active block job '%s' cannot be completed",
929 job->driver->complete(job, errp);
935 JobDeferToMainLoopFn *fn;
937 } JobDeferToMainLoopData;
939 static void job_defer_to_main_loop_bh(void *opaque)
941 JobDeferToMainLoopData *data = opaque;
942 Job *job = data->job;
943 AioContext *aio_context = job->aio_context;
945 aio_context_acquire(aio_context);
946 data->fn(data->job, data->opaque);
947 aio_context_release(aio_context);
952 void job_defer_to_main_loop(Job *job, JobDeferToMainLoopFn *fn, void *opaque)
954 JobDeferToMainLoopData *data = g_malloc(sizeof(*data));
957 data->opaque = opaque;
958 job->deferred_to_main_loop = true;
960 aio_bh_schedule_oneshot(qemu_get_aio_context(),
961 job_defer_to_main_loop_bh, data);
964 int job_finish_sync(Job *job, void (*finish)(Job *, Error **errp), Error **errp)
966 Error *local_err = NULL;
972 finish(job, &local_err);
975 error_propagate(errp, local_err);
979 /* job_drain calls job_enter, and it should be enough to induce progress
980 * until the job completes or moves to the main thread. */
981 while (!job->deferred_to_main_loop && !job_is_completed(job)) {
984 while (!job_is_completed(job)) {
985 aio_poll(qemu_get_aio_context(), true);
987 ret = (job_is_cancelled(job) && job->ret == 0) ? -ECANCELED : job->ret;