Merge tag 'pull-ppc-20220706' of https://gitlab.com/danielhb/qemu into staging
[qemu.git] / job.c
1 /*
2 * Background jobs (long-running operations)
3 *
4 * Copyright (c) 2011 IBM Corp.
5 * Copyright (c) 2012, 2018 Red Hat, Inc.
6 *
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:
13 *
14 * The above copyright notice and this permission notice shall be included in
15 * all copies or substantial portions of the Software.
16 *
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
23 * THE SOFTWARE.
24 */
25
26 #include "qemu/osdep.h"
27 #include "qapi/error.h"
28 #include "qemu/job.h"
29 #include "qemu/id.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"
34
35 static QLIST_HEAD(, Job) jobs = QLIST_HEAD_INITIALIZER(jobs);
36
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},
51 };
52
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},
62 };
63
64 /* Transactional group of jobs */
65 struct JobTxn {
66
67 /* Is this txn being cancelled? */
68 bool aborting;
69
70 /* List of jobs */
71 QLIST_HEAD(, Job) jobs;
72
73 /* Reference count */
74 int refcnt;
75 };
76
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
79 * job_enter. */
80 static QemuMutex job_mutex;
81
82 static void job_lock(void)
83 {
84 qemu_mutex_lock(&job_mutex);
85 }
86
87 static void job_unlock(void)
88 {
89 qemu_mutex_unlock(&job_mutex);
90 }
91
92 static void __attribute__((__constructor__)) job_init(void)
93 {
94 qemu_mutex_init(&job_mutex);
95 }
96
97 JobTxn *job_txn_new(void)
98 {
99 JobTxn *txn = g_new0(JobTxn, 1);
100 QLIST_INIT(&txn->jobs);
101 txn->refcnt = 1;
102 return txn;
103 }
104
105 static void job_txn_ref(JobTxn *txn)
106 {
107 txn->refcnt++;
108 }
109
110 void job_txn_unref(JobTxn *txn)
111 {
112 if (txn && --txn->refcnt == 0) {
113 g_free(txn);
114 }
115 }
116
117 void job_txn_add_job(JobTxn *txn, Job *job)
118 {
119 if (!txn) {
120 return;
121 }
122
123 assert(!job->txn);
124 job->txn = txn;
125
126 QLIST_INSERT_HEAD(&txn->jobs, job, txn_list);
127 job_txn_ref(txn);
128 }
129
130 static void job_txn_del_job(Job *job)
131 {
132 if (job->txn) {
133 QLIST_REMOVE(job, txn_list);
134 job_txn_unref(job->txn);
135 job->txn = NULL;
136 }
137 }
138
139 static int job_txn_apply(Job *job, int fn(Job *))
140 {
141 AioContext *inner_ctx;
142 Job *other_job, *next;
143 JobTxn *txn = job->txn;
144 int rc = 0;
145
146 /*
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.
151 */
152 job_ref(job);
153 aio_context_release(job->aio_context);
154
155 QLIST_FOREACH_SAFE(other_job, &txn->jobs, txn_list, next) {
156 inner_ctx = other_job->aio_context;
157 aio_context_acquire(inner_ctx);
158 rc = fn(other_job);
159 aio_context_release(inner_ctx);
160 if (rc) {
161 break;
162 }
163 }
164
165 /*
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.
168 */
169 aio_context_acquire(job->aio_context);
170 job_unref(job);
171 return rc;
172 }
173
174 bool job_is_internal(Job *job)
175 {
176 return (job->id == NULL);
177 }
178
179 static void job_state_transition(Job *job, JobStatus s1)
180 {
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]);
187 job->status = s1;
188
189 if (!job_is_internal(job) && s1 != s0) {
190 qapi_event_send_job_status_change(job->id, job->status);
191 }
192 }
193
194 int job_apply_verb(Job *job, JobVerb verb, Error **errp)
195 {
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]) {
201 return 0;
202 }
203 error_setg(errp, "Job '%s' in state '%s' cannot accept command verb '%s'",
204 job->id, JobStatus_str(s0), JobVerb_str(verb));
205 return -EPERM;
206 }
207
208 JobType job_type(const Job *job)
209 {
210 return job->driver->job_type;
211 }
212
213 const char *job_type_str(const Job *job)
214 {
215 return JobType_str(job_type(job));
216 }
217
218 bool job_is_cancelled(Job *job)
219 {
220 /* force_cancel may be true only if cancelled is true, too */
221 assert(job->cancelled || !job->force_cancel);
222 return job->force_cancel;
223 }
224
225 bool job_cancel_requested(Job *job)
226 {
227 return job->cancelled;
228 }
229
230 bool job_is_ready(Job *job)
231 {
232 switch (job->status) {
233 case JOB_STATUS_UNDEFINED:
234 case JOB_STATUS_CREATED:
235 case JOB_STATUS_RUNNING:
236 case JOB_STATUS_PAUSED:
237 case JOB_STATUS_WAITING:
238 case JOB_STATUS_PENDING:
239 case JOB_STATUS_ABORTING:
240 case JOB_STATUS_CONCLUDED:
241 case JOB_STATUS_NULL:
242 return false;
243 case JOB_STATUS_READY:
244 case JOB_STATUS_STANDBY:
245 return true;
246 default:
247 g_assert_not_reached();
248 }
249 return false;
250 }
251
252 bool job_is_completed(Job *job)
253 {
254 switch (job->status) {
255 case JOB_STATUS_UNDEFINED:
256 case JOB_STATUS_CREATED:
257 case JOB_STATUS_RUNNING:
258 case JOB_STATUS_PAUSED:
259 case JOB_STATUS_READY:
260 case JOB_STATUS_STANDBY:
261 return false;
262 case JOB_STATUS_WAITING:
263 case JOB_STATUS_PENDING:
264 case JOB_STATUS_ABORTING:
265 case JOB_STATUS_CONCLUDED:
266 case JOB_STATUS_NULL:
267 return true;
268 default:
269 g_assert_not_reached();
270 }
271 return false;
272 }
273
274 static bool job_started(Job *job)
275 {
276 return job->co;
277 }
278
279 static bool job_should_pause(Job *job)
280 {
281 return job->pause_count > 0;
282 }
283
284 Job *job_next(Job *job)
285 {
286 if (!job) {
287 return QLIST_FIRST(&jobs);
288 }
289 return QLIST_NEXT(job, job_list);
290 }
291
292 Job *job_get(const char *id)
293 {
294 Job *job;
295
296 QLIST_FOREACH(job, &jobs, job_list) {
297 if (job->id && !strcmp(id, job->id)) {
298 return job;
299 }
300 }
301
302 return NULL;
303 }
304
305 static void job_sleep_timer_cb(void *opaque)
306 {
307 Job *job = opaque;
308
309 job_enter(job);
310 }
311
312 void *job_create(const char *job_id, const JobDriver *driver, JobTxn *txn,
313 AioContext *ctx, int flags, BlockCompletionFunc *cb,
314 void *opaque, Error **errp)
315 {
316 Job *job;
317
318 if (job_id) {
319 if (flags & JOB_INTERNAL) {
320 error_setg(errp, "Cannot specify job ID for internal job");
321 return NULL;
322 }
323 if (!id_wellformed(job_id)) {
324 error_setg(errp, "Invalid job ID '%s'", job_id);
325 return NULL;
326 }
327 if (job_get(job_id)) {
328 error_setg(errp, "Job ID '%s' already in use", job_id);
329 return NULL;
330 }
331 } else if (!(flags & JOB_INTERNAL)) {
332 error_setg(errp, "An explicit job ID is required");
333 return NULL;
334 }
335
336 job = g_malloc0(driver->instance_size);
337 job->driver = driver;
338 job->id = g_strdup(job_id);
339 job->refcnt = 1;
340 job->aio_context = ctx;
341 job->busy = false;
342 job->paused = true;
343 job->pause_count = 1;
344 job->auto_finalize = !(flags & JOB_MANUAL_FINALIZE);
345 job->auto_dismiss = !(flags & JOB_MANUAL_DISMISS);
346 job->cb = cb;
347 job->opaque = opaque;
348
349 progress_init(&job->progress);
350
351 notifier_list_init(&job->on_finalize_cancelled);
352 notifier_list_init(&job->on_finalize_completed);
353 notifier_list_init(&job->on_pending);
354 notifier_list_init(&job->on_ready);
355 notifier_list_init(&job->on_idle);
356
357 job_state_transition(job, JOB_STATUS_CREATED);
358 aio_timer_init(qemu_get_aio_context(), &job->sleep_timer,
359 QEMU_CLOCK_REALTIME, SCALE_NS,
360 job_sleep_timer_cb, job);
361
362 QLIST_INSERT_HEAD(&jobs, job, job_list);
363
364 /* Single jobs are modeled as single-job transactions for sake of
365 * consolidating the job management logic */
366 if (!txn) {
367 txn = job_txn_new();
368 job_txn_add_job(txn, job);
369 job_txn_unref(txn);
370 } else {
371 job_txn_add_job(txn, job);
372 }
373
374 return job;
375 }
376
377 void job_ref(Job *job)
378 {
379 ++job->refcnt;
380 }
381
382 void job_unref(Job *job)
383 {
384 if (--job->refcnt == 0) {
385 assert(job->status == JOB_STATUS_NULL);
386 assert(!timer_pending(&job->sleep_timer));
387 assert(!job->txn);
388
389 if (job->driver->free) {
390 job->driver->free(job);
391 }
392
393 QLIST_REMOVE(job, job_list);
394
395 progress_destroy(&job->progress);
396 error_free(job->err);
397 g_free(job->id);
398 g_free(job);
399 }
400 }
401
402 void job_progress_update(Job *job, uint64_t done)
403 {
404 progress_work_done(&job->progress, done);
405 }
406
407 void job_progress_set_remaining(Job *job, uint64_t remaining)
408 {
409 progress_set_remaining(&job->progress, remaining);
410 }
411
412 void job_progress_increase_remaining(Job *job, uint64_t delta)
413 {
414 progress_increase_remaining(&job->progress, delta);
415 }
416
417 void job_event_cancelled(Job *job)
418 {
419 notifier_list_notify(&job->on_finalize_cancelled, job);
420 }
421
422 void job_event_completed(Job *job)
423 {
424 notifier_list_notify(&job->on_finalize_completed, job);
425 }
426
427 static void job_event_pending(Job *job)
428 {
429 notifier_list_notify(&job->on_pending, job);
430 }
431
432 static void job_event_ready(Job *job)
433 {
434 notifier_list_notify(&job->on_ready, job);
435 }
436
437 static void job_event_idle(Job *job)
438 {
439 notifier_list_notify(&job->on_idle, job);
440 }
441
442 void job_enter_cond(Job *job, bool(*fn)(Job *job))
443 {
444 if (!job_started(job)) {
445 return;
446 }
447 if (job->deferred_to_main_loop) {
448 return;
449 }
450
451 job_lock();
452 if (job->busy) {
453 job_unlock();
454 return;
455 }
456
457 if (fn && !fn(job)) {
458 job_unlock();
459 return;
460 }
461
462 assert(!job->deferred_to_main_loop);
463 timer_del(&job->sleep_timer);
464 job->busy = true;
465 job_unlock();
466 aio_co_enter(job->aio_context, job->co);
467 }
468
469 void job_enter(Job *job)
470 {
471 job_enter_cond(job, NULL);
472 }
473
474 /* Yield, and schedule a timer to reenter the coroutine after @ns nanoseconds.
475 * Reentering the job coroutine with job_enter() before the timer has expired
476 * is allowed and cancels the timer.
477 *
478 * If @ns is (uint64_t) -1, no timer is scheduled and job_enter() must be
479 * called explicitly. */
480 static void coroutine_fn job_do_yield(Job *job, uint64_t ns)
481 {
482 job_lock();
483 if (ns != -1) {
484 timer_mod(&job->sleep_timer, ns);
485 }
486 job->busy = false;
487 job_event_idle(job);
488 job_unlock();
489 qemu_coroutine_yield();
490
491 /* Set by job_enter_cond() before re-entering the coroutine. */
492 assert(job->busy);
493 }
494
495 void coroutine_fn job_pause_point(Job *job)
496 {
497 assert(job && job_started(job));
498
499 if (!job_should_pause(job)) {
500 return;
501 }
502 if (job_is_cancelled(job)) {
503 return;
504 }
505
506 if (job->driver->pause) {
507 job->driver->pause(job);
508 }
509
510 if (job_should_pause(job) && !job_is_cancelled(job)) {
511 JobStatus status = job->status;
512 job_state_transition(job, status == JOB_STATUS_READY
513 ? JOB_STATUS_STANDBY
514 : JOB_STATUS_PAUSED);
515 job->paused = true;
516 job_do_yield(job, -1);
517 job->paused = false;
518 job_state_transition(job, status);
519 }
520
521 if (job->driver->resume) {
522 job->driver->resume(job);
523 }
524 }
525
526 void job_yield(Job *job)
527 {
528 assert(job->busy);
529
530 /* Check cancellation *before* setting busy = false, too! */
531 if (job_is_cancelled(job)) {
532 return;
533 }
534
535 if (!job_should_pause(job)) {
536 job_do_yield(job, -1);
537 }
538
539 job_pause_point(job);
540 }
541
542 void coroutine_fn job_sleep_ns(Job *job, int64_t ns)
543 {
544 assert(job->busy);
545
546 /* Check cancellation *before* setting busy = false, too! */
547 if (job_is_cancelled(job)) {
548 return;
549 }
550
551 if (!job_should_pause(job)) {
552 job_do_yield(job, qemu_clock_get_ns(QEMU_CLOCK_REALTIME) + ns);
553 }
554
555 job_pause_point(job);
556 }
557
558 /* Assumes the block_job_mutex is held */
559 static bool job_timer_not_pending(Job *job)
560 {
561 return !timer_pending(&job->sleep_timer);
562 }
563
564 void job_pause(Job *job)
565 {
566 job->pause_count++;
567 if (!job->paused) {
568 job_enter(job);
569 }
570 }
571
572 void job_resume(Job *job)
573 {
574 assert(job->pause_count > 0);
575 job->pause_count--;
576 if (job->pause_count) {
577 return;
578 }
579
580 /* kick only if no timer is pending */
581 job_enter_cond(job, job_timer_not_pending);
582 }
583
584 void job_user_pause(Job *job, Error **errp)
585 {
586 if (job_apply_verb(job, JOB_VERB_PAUSE, errp)) {
587 return;
588 }
589 if (job->user_paused) {
590 error_setg(errp, "Job is already paused");
591 return;
592 }
593 job->user_paused = true;
594 job_pause(job);
595 }
596
597 bool job_user_paused(Job *job)
598 {
599 return job->user_paused;
600 }
601
602 void job_user_resume(Job *job, Error **errp)
603 {
604 assert(job);
605 if (!job->user_paused || job->pause_count <= 0) {
606 error_setg(errp, "Can't resume a job that was not paused");
607 return;
608 }
609 if (job_apply_verb(job, JOB_VERB_RESUME, errp)) {
610 return;
611 }
612 if (job->driver->user_resume) {
613 job->driver->user_resume(job);
614 }
615 job->user_paused = false;
616 job_resume(job);
617 }
618
619 static void job_do_dismiss(Job *job)
620 {
621 assert(job);
622 job->busy = false;
623 job->paused = false;
624 job->deferred_to_main_loop = true;
625
626 job_txn_del_job(job);
627
628 job_state_transition(job, JOB_STATUS_NULL);
629 job_unref(job);
630 }
631
632 void job_dismiss(Job **jobptr, Error **errp)
633 {
634 Job *job = *jobptr;
635 /* similarly to _complete, this is QMP-interface only. */
636 assert(job->id);
637 if (job_apply_verb(job, JOB_VERB_DISMISS, errp)) {
638 return;
639 }
640
641 job_do_dismiss(job);
642 *jobptr = NULL;
643 }
644
645 void job_early_fail(Job *job)
646 {
647 assert(job->status == JOB_STATUS_CREATED);
648 job_do_dismiss(job);
649 }
650
651 static void job_conclude(Job *job)
652 {
653 job_state_transition(job, JOB_STATUS_CONCLUDED);
654 if (job->auto_dismiss || !job_started(job)) {
655 job_do_dismiss(job);
656 }
657 }
658
659 static void job_update_rc(Job *job)
660 {
661 if (!job->ret && job_is_cancelled(job)) {
662 job->ret = -ECANCELED;
663 }
664 if (job->ret) {
665 if (!job->err) {
666 error_setg(&job->err, "%s", strerror(-job->ret));
667 }
668 job_state_transition(job, JOB_STATUS_ABORTING);
669 }
670 }
671
672 static void job_commit(Job *job)
673 {
674 assert(!job->ret);
675 if (job->driver->commit) {
676 job->driver->commit(job);
677 }
678 }
679
680 static void job_abort(Job *job)
681 {
682 assert(job->ret);
683 if (job->driver->abort) {
684 job->driver->abort(job);
685 }
686 }
687
688 static void job_clean(Job *job)
689 {
690 if (job->driver->clean) {
691 job->driver->clean(job);
692 }
693 }
694
695 static int job_finalize_single(Job *job)
696 {
697 assert(job_is_completed(job));
698
699 /* Ensure abort is called for late-transactional failures */
700 job_update_rc(job);
701
702 if (!job->ret) {
703 job_commit(job);
704 } else {
705 job_abort(job);
706 }
707 job_clean(job);
708
709 if (job->cb) {
710 job->cb(job->opaque, job->ret);
711 }
712
713 /* Emit events only if we actually started */
714 if (job_started(job)) {
715 if (job_is_cancelled(job)) {
716 job_event_cancelled(job);
717 } else {
718 job_event_completed(job);
719 }
720 }
721
722 job_txn_del_job(job);
723 job_conclude(job);
724 return 0;
725 }
726
727 static void job_cancel_async(Job *job, bool force)
728 {
729 if (job->driver->cancel) {
730 force = job->driver->cancel(job, force);
731 } else {
732 /* No .cancel() means the job will behave as if force-cancelled */
733 force = true;
734 }
735
736 if (job->user_paused) {
737 /* Do not call job_enter here, the caller will handle it. */
738 if (job->driver->user_resume) {
739 job->driver->user_resume(job);
740 }
741 job->user_paused = false;
742 assert(job->pause_count > 0);
743 job->pause_count--;
744 }
745
746 /*
747 * Ignore soft cancel requests after the job is already done
748 * (We will still invoke job->driver->cancel() above, but if the
749 * job driver supports soft cancelling and the job is done, that
750 * should be a no-op, too. We still call it so it can override
751 * @force.)
752 */
753 if (force || !job->deferred_to_main_loop) {
754 job->cancelled = true;
755 /* To prevent 'force == false' overriding a previous 'force == true' */
756 job->force_cancel |= force;
757 }
758 }
759
760 static void job_completed_txn_abort(Job *job)
761 {
762 AioContext *ctx;
763 JobTxn *txn = job->txn;
764 Job *other_job;
765
766 if (txn->aborting) {
767 /*
768 * We are cancelled by another job, which will handle everything.
769 */
770 return;
771 }
772 txn->aborting = true;
773 job_txn_ref(txn);
774
775 /*
776 * We can only hold the single job's AioContext lock while calling
777 * job_finalize_single() because the finalization callbacks can involve
778 * calls of AIO_WAIT_WHILE(), which could deadlock otherwise.
779 * Note that the job's AioContext may change when it is finalized.
780 */
781 job_ref(job);
782 aio_context_release(job->aio_context);
783
784 /* Other jobs are effectively cancelled by us, set the status for
785 * them; this job, however, may or may not be cancelled, depending
786 * on the caller, so leave it. */
787 QLIST_FOREACH(other_job, &txn->jobs, txn_list) {
788 if (other_job != job) {
789 ctx = other_job->aio_context;
790 aio_context_acquire(ctx);
791 /*
792 * This is a transaction: If one job failed, no result will matter.
793 * Therefore, pass force=true to terminate all other jobs as quickly
794 * as possible.
795 */
796 job_cancel_async(other_job, true);
797 aio_context_release(ctx);
798 }
799 }
800 while (!QLIST_EMPTY(&txn->jobs)) {
801 other_job = QLIST_FIRST(&txn->jobs);
802 /*
803 * The job's AioContext may change, so store it in @ctx so we
804 * release the same context that we have acquired before.
805 */
806 ctx = other_job->aio_context;
807 aio_context_acquire(ctx);
808 if (!job_is_completed(other_job)) {
809 assert(job_cancel_requested(other_job));
810 job_finish_sync(other_job, NULL, NULL);
811 }
812 job_finalize_single(other_job);
813 aio_context_release(ctx);
814 }
815
816 /*
817 * Use job_ref()/job_unref() so we can read the AioContext here
818 * even if the job went away during job_finalize_single().
819 */
820 aio_context_acquire(job->aio_context);
821 job_unref(job);
822
823 job_txn_unref(txn);
824 }
825
826 static int job_prepare(Job *job)
827 {
828 if (job->ret == 0 && job->driver->prepare) {
829 job->ret = job->driver->prepare(job);
830 job_update_rc(job);
831 }
832 return job->ret;
833 }
834
835 static int job_needs_finalize(Job *job)
836 {
837 return !job->auto_finalize;
838 }
839
840 static void job_do_finalize(Job *job)
841 {
842 int rc;
843 assert(job && job->txn);
844
845 /* prepare the transaction to complete */
846 rc = job_txn_apply(job, job_prepare);
847 if (rc) {
848 job_completed_txn_abort(job);
849 } else {
850 job_txn_apply(job, job_finalize_single);
851 }
852 }
853
854 void job_finalize(Job *job, Error **errp)
855 {
856 assert(job && job->id);
857 if (job_apply_verb(job, JOB_VERB_FINALIZE, errp)) {
858 return;
859 }
860 job_do_finalize(job);
861 }
862
863 static int job_transition_to_pending(Job *job)
864 {
865 job_state_transition(job, JOB_STATUS_PENDING);
866 if (!job->auto_finalize) {
867 job_event_pending(job);
868 }
869 return 0;
870 }
871
872 void job_transition_to_ready(Job *job)
873 {
874 job_state_transition(job, JOB_STATUS_READY);
875 job_event_ready(job);
876 }
877
878 static void job_completed_txn_success(Job *job)
879 {
880 JobTxn *txn = job->txn;
881 Job *other_job;
882
883 job_state_transition(job, JOB_STATUS_WAITING);
884
885 /*
886 * Successful completion, see if there are other running jobs in this
887 * txn.
888 */
889 QLIST_FOREACH(other_job, &txn->jobs, txn_list) {
890 if (!job_is_completed(other_job)) {
891 return;
892 }
893 assert(other_job->ret == 0);
894 }
895
896 job_txn_apply(job, job_transition_to_pending);
897
898 /* If no jobs need manual finalization, automatically do so */
899 if (job_txn_apply(job, job_needs_finalize) == 0) {
900 job_do_finalize(job);
901 }
902 }
903
904 static void job_completed(Job *job)
905 {
906 assert(job && job->txn && !job_is_completed(job));
907
908 job_update_rc(job);
909 trace_job_completed(job, job->ret);
910 if (job->ret) {
911 job_completed_txn_abort(job);
912 } else {
913 job_completed_txn_success(job);
914 }
915 }
916
917 /** Useful only as a type shim for aio_bh_schedule_oneshot. */
918 static void job_exit(void *opaque)
919 {
920 Job *job = (Job *)opaque;
921 AioContext *ctx;
922
923 job_ref(job);
924 aio_context_acquire(job->aio_context);
925
926 /* This is a lie, we're not quiescent, but still doing the completion
927 * callbacks. However, completion callbacks tend to involve operations that
928 * drain block nodes, and if .drained_poll still returned true, we would
929 * deadlock. */
930 job->busy = false;
931 job_event_idle(job);
932
933 job_completed(job);
934
935 /*
936 * Note that calling job_completed can move the job to a different
937 * aio_context, so we cannot cache from above. job_txn_apply takes care of
938 * acquiring the new lock, and we ref/unref to avoid job_completed freeing
939 * the job underneath us.
940 */
941 ctx = job->aio_context;
942 job_unref(job);
943 aio_context_release(ctx);
944 }
945
946 /**
947 * All jobs must allow a pause point before entering their job proper. This
948 * ensures that jobs can be paused prior to being started, then resumed later.
949 */
950 static void coroutine_fn job_co_entry(void *opaque)
951 {
952 Job *job = opaque;
953
954 assert(job && job->driver && job->driver->run);
955 job_pause_point(job);
956 job->ret = job->driver->run(job, &job->err);
957 job->deferred_to_main_loop = true;
958 job->busy = true;
959 aio_bh_schedule_oneshot(qemu_get_aio_context(), job_exit, job);
960 }
961
962 void job_start(Job *job)
963 {
964 assert(job && !job_started(job) && job->paused &&
965 job->driver && job->driver->run);
966 job->co = qemu_coroutine_create(job_co_entry, job);
967 job->pause_count--;
968 job->busy = true;
969 job->paused = false;
970 job_state_transition(job, JOB_STATUS_RUNNING);
971 aio_co_enter(job->aio_context, job->co);
972 }
973
974 void job_cancel(Job *job, bool force)
975 {
976 if (job->status == JOB_STATUS_CONCLUDED) {
977 job_do_dismiss(job);
978 return;
979 }
980 job_cancel_async(job, force);
981 if (!job_started(job)) {
982 job_completed(job);
983 } else if (job->deferred_to_main_loop) {
984 /*
985 * job_cancel_async() ignores soft-cancel requests for jobs
986 * that are already done (i.e. deferred to the main loop). We
987 * have to check again whether the job is really cancelled.
988 * (job_cancel_requested() and job_is_cancelled() are equivalent
989 * here, because job_cancel_async() will make soft-cancel
990 * requests no-ops when deferred_to_main_loop is true. We
991 * choose to call job_is_cancelled() to show that we invoke
992 * job_completed_txn_abort() only for force-cancelled jobs.)
993 */
994 if (job_is_cancelled(job)) {
995 job_completed_txn_abort(job);
996 }
997 } else {
998 job_enter(job);
999 }
1000 }
1001
1002 void job_user_cancel(Job *job, bool force, Error **errp)
1003 {
1004 if (job_apply_verb(job, JOB_VERB_CANCEL, errp)) {
1005 return;
1006 }
1007 job_cancel(job, force);
1008 }
1009
1010 /* A wrapper around job_cancel() taking an Error ** parameter so it may be
1011 * used with job_finish_sync() without the need for (rather nasty) function
1012 * pointer casts there. */
1013 static void job_cancel_err(Job *job, Error **errp)
1014 {
1015 job_cancel(job, false);
1016 }
1017
1018 /**
1019 * Same as job_cancel_err(), but force-cancel.
1020 */
1021 static void job_force_cancel_err(Job *job, Error **errp)
1022 {
1023 job_cancel(job, true);
1024 }
1025
1026 int job_cancel_sync(Job *job, bool force)
1027 {
1028 if (force) {
1029 return job_finish_sync(job, &job_force_cancel_err, NULL);
1030 } else {
1031 return job_finish_sync(job, &job_cancel_err, NULL);
1032 }
1033 }
1034
1035 void job_cancel_sync_all(void)
1036 {
1037 Job *job;
1038 AioContext *aio_context;
1039
1040 while ((job = job_next(NULL))) {
1041 aio_context = job->aio_context;
1042 aio_context_acquire(aio_context);
1043 job_cancel_sync(job, true);
1044 aio_context_release(aio_context);
1045 }
1046 }
1047
1048 int job_complete_sync(Job *job, Error **errp)
1049 {
1050 return job_finish_sync(job, job_complete, errp);
1051 }
1052
1053 void job_complete(Job *job, Error **errp)
1054 {
1055 /* Should not be reachable via external interface for internal jobs */
1056 assert(job->id);
1057 if (job_apply_verb(job, JOB_VERB_COMPLETE, errp)) {
1058 return;
1059 }
1060 if (job_cancel_requested(job) || !job->driver->complete) {
1061 error_setg(errp, "The active block job '%s' cannot be completed",
1062 job->id);
1063 return;
1064 }
1065
1066 job->driver->complete(job, errp);
1067 }
1068
1069 int job_finish_sync(Job *job, void (*finish)(Job *, Error **errp), Error **errp)
1070 {
1071 Error *local_err = NULL;
1072 int ret;
1073
1074 job_ref(job);
1075
1076 if (finish) {
1077 finish(job, &local_err);
1078 }
1079 if (local_err) {
1080 error_propagate(errp, local_err);
1081 job_unref(job);
1082 return -EBUSY;
1083 }
1084
1085 AIO_WAIT_WHILE(job->aio_context,
1086 (job_enter(job), !job_is_completed(job)));
1087
1088 ret = (job_is_cancelled(job) && job->ret == 0) ? -ECANCELED : job->ret;
1089 job_unref(job);
1090 return ret;
1091 }