Merge tag 'block-pull-request' of https://gitlab.com/stefanha/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 GLOBAL_STATE_CODE();
385
386 if (--job->refcnt == 0) {
387 assert(job->status == JOB_STATUS_NULL);
388 assert(!timer_pending(&job->sleep_timer));
389 assert(!job->txn);
390
391 if (job->driver->free) {
392 job->driver->free(job);
393 }
394
395 QLIST_REMOVE(job, job_list);
396
397 progress_destroy(&job->progress);
398 error_free(job->err);
399 g_free(job->id);
400 g_free(job);
401 }
402 }
403
404 void job_progress_update(Job *job, uint64_t done)
405 {
406 progress_work_done(&job->progress, done);
407 }
408
409 void job_progress_set_remaining(Job *job, uint64_t remaining)
410 {
411 progress_set_remaining(&job->progress, remaining);
412 }
413
414 void job_progress_increase_remaining(Job *job, uint64_t delta)
415 {
416 progress_increase_remaining(&job->progress, delta);
417 }
418
419 void job_event_cancelled(Job *job)
420 {
421 notifier_list_notify(&job->on_finalize_cancelled, job);
422 }
423
424 void job_event_completed(Job *job)
425 {
426 notifier_list_notify(&job->on_finalize_completed, job);
427 }
428
429 static void job_event_pending(Job *job)
430 {
431 notifier_list_notify(&job->on_pending, job);
432 }
433
434 static void job_event_ready(Job *job)
435 {
436 notifier_list_notify(&job->on_ready, job);
437 }
438
439 static void job_event_idle(Job *job)
440 {
441 notifier_list_notify(&job->on_idle, job);
442 }
443
444 void job_enter_cond(Job *job, bool(*fn)(Job *job))
445 {
446 if (!job_started(job)) {
447 return;
448 }
449 if (job->deferred_to_main_loop) {
450 return;
451 }
452
453 job_lock();
454 if (job->busy) {
455 job_unlock();
456 return;
457 }
458
459 if (fn && !fn(job)) {
460 job_unlock();
461 return;
462 }
463
464 assert(!job->deferred_to_main_loop);
465 timer_del(&job->sleep_timer);
466 job->busy = true;
467 job_unlock();
468 aio_co_enter(job->aio_context, job->co);
469 }
470
471 void job_enter(Job *job)
472 {
473 job_enter_cond(job, NULL);
474 }
475
476 /* Yield, and schedule a timer to reenter the coroutine after @ns nanoseconds.
477 * Reentering the job coroutine with job_enter() before the timer has expired
478 * is allowed and cancels the timer.
479 *
480 * If @ns is (uint64_t) -1, no timer is scheduled and job_enter() must be
481 * called explicitly. */
482 static void coroutine_fn job_do_yield(Job *job, uint64_t ns)
483 {
484 job_lock();
485 if (ns != -1) {
486 timer_mod(&job->sleep_timer, ns);
487 }
488 job->busy = false;
489 job_event_idle(job);
490 job_unlock();
491 qemu_coroutine_yield();
492
493 /* Set by job_enter_cond() before re-entering the coroutine. */
494 assert(job->busy);
495 }
496
497 void coroutine_fn job_pause_point(Job *job)
498 {
499 assert(job && job_started(job));
500
501 if (!job_should_pause(job)) {
502 return;
503 }
504 if (job_is_cancelled(job)) {
505 return;
506 }
507
508 if (job->driver->pause) {
509 job->driver->pause(job);
510 }
511
512 if (job_should_pause(job) && !job_is_cancelled(job)) {
513 JobStatus status = job->status;
514 job_state_transition(job, status == JOB_STATUS_READY
515 ? JOB_STATUS_STANDBY
516 : JOB_STATUS_PAUSED);
517 job->paused = true;
518 job_do_yield(job, -1);
519 job->paused = false;
520 job_state_transition(job, status);
521 }
522
523 if (job->driver->resume) {
524 job->driver->resume(job);
525 }
526 }
527
528 void job_yield(Job *job)
529 {
530 assert(job->busy);
531
532 /* Check cancellation *before* setting busy = false, too! */
533 if (job_is_cancelled(job)) {
534 return;
535 }
536
537 if (!job_should_pause(job)) {
538 job_do_yield(job, -1);
539 }
540
541 job_pause_point(job);
542 }
543
544 void coroutine_fn job_sleep_ns(Job *job, int64_t ns)
545 {
546 assert(job->busy);
547
548 /* Check cancellation *before* setting busy = false, too! */
549 if (job_is_cancelled(job)) {
550 return;
551 }
552
553 if (!job_should_pause(job)) {
554 job_do_yield(job, qemu_clock_get_ns(QEMU_CLOCK_REALTIME) + ns);
555 }
556
557 job_pause_point(job);
558 }
559
560 /* Assumes the block_job_mutex is held */
561 static bool job_timer_not_pending(Job *job)
562 {
563 return !timer_pending(&job->sleep_timer);
564 }
565
566 void job_pause(Job *job)
567 {
568 job->pause_count++;
569 if (!job->paused) {
570 job_enter(job);
571 }
572 }
573
574 void job_resume(Job *job)
575 {
576 assert(job->pause_count > 0);
577 job->pause_count--;
578 if (job->pause_count) {
579 return;
580 }
581
582 /* kick only if no timer is pending */
583 job_enter_cond(job, job_timer_not_pending);
584 }
585
586 void job_user_pause(Job *job, Error **errp)
587 {
588 if (job_apply_verb(job, JOB_VERB_PAUSE, errp)) {
589 return;
590 }
591 if (job->user_paused) {
592 error_setg(errp, "Job is already paused");
593 return;
594 }
595 job->user_paused = true;
596 job_pause(job);
597 }
598
599 bool job_user_paused(Job *job)
600 {
601 return job->user_paused;
602 }
603
604 void job_user_resume(Job *job, Error **errp)
605 {
606 assert(job);
607 GLOBAL_STATE_CODE();
608 if (!job->user_paused || job->pause_count <= 0) {
609 error_setg(errp, "Can't resume a job that was not paused");
610 return;
611 }
612 if (job_apply_verb(job, JOB_VERB_RESUME, errp)) {
613 return;
614 }
615 if (job->driver->user_resume) {
616 job->driver->user_resume(job);
617 }
618 job->user_paused = false;
619 job_resume(job);
620 }
621
622 static void job_do_dismiss(Job *job)
623 {
624 assert(job);
625 job->busy = false;
626 job->paused = false;
627 job->deferred_to_main_loop = true;
628
629 job_txn_del_job(job);
630
631 job_state_transition(job, JOB_STATUS_NULL);
632 job_unref(job);
633 }
634
635 void job_dismiss(Job **jobptr, Error **errp)
636 {
637 Job *job = *jobptr;
638 /* similarly to _complete, this is QMP-interface only. */
639 assert(job->id);
640 if (job_apply_verb(job, JOB_VERB_DISMISS, errp)) {
641 return;
642 }
643
644 job_do_dismiss(job);
645 *jobptr = NULL;
646 }
647
648 void job_early_fail(Job *job)
649 {
650 assert(job->status == JOB_STATUS_CREATED);
651 job_do_dismiss(job);
652 }
653
654 static void job_conclude(Job *job)
655 {
656 job_state_transition(job, JOB_STATUS_CONCLUDED);
657 if (job->auto_dismiss || !job_started(job)) {
658 job_do_dismiss(job);
659 }
660 }
661
662 static void job_update_rc(Job *job)
663 {
664 if (!job->ret && job_is_cancelled(job)) {
665 job->ret = -ECANCELED;
666 }
667 if (job->ret) {
668 if (!job->err) {
669 error_setg(&job->err, "%s", strerror(-job->ret));
670 }
671 job_state_transition(job, JOB_STATUS_ABORTING);
672 }
673 }
674
675 static void job_commit(Job *job)
676 {
677 assert(!job->ret);
678 GLOBAL_STATE_CODE();
679 if (job->driver->commit) {
680 job->driver->commit(job);
681 }
682 }
683
684 static void job_abort(Job *job)
685 {
686 assert(job->ret);
687 GLOBAL_STATE_CODE();
688 if (job->driver->abort) {
689 job->driver->abort(job);
690 }
691 }
692
693 static void job_clean(Job *job)
694 {
695 GLOBAL_STATE_CODE();
696 if (job->driver->clean) {
697 job->driver->clean(job);
698 }
699 }
700
701 static int job_finalize_single(Job *job)
702 {
703 assert(job_is_completed(job));
704
705 /* Ensure abort is called for late-transactional failures */
706 job_update_rc(job);
707
708 if (!job->ret) {
709 job_commit(job);
710 } else {
711 job_abort(job);
712 }
713 job_clean(job);
714
715 if (job->cb) {
716 job->cb(job->opaque, job->ret);
717 }
718
719 /* Emit events only if we actually started */
720 if (job_started(job)) {
721 if (job_is_cancelled(job)) {
722 job_event_cancelled(job);
723 } else {
724 job_event_completed(job);
725 }
726 }
727
728 job_txn_del_job(job);
729 job_conclude(job);
730 return 0;
731 }
732
733 static void job_cancel_async(Job *job, bool force)
734 {
735 GLOBAL_STATE_CODE();
736 if (job->driver->cancel) {
737 force = job->driver->cancel(job, force);
738 } else {
739 /* No .cancel() means the job will behave as if force-cancelled */
740 force = true;
741 }
742
743 if (job->user_paused) {
744 /* Do not call job_enter here, the caller will handle it. */
745 if (job->driver->user_resume) {
746 job->driver->user_resume(job);
747 }
748 job->user_paused = false;
749 assert(job->pause_count > 0);
750 job->pause_count--;
751 }
752
753 /*
754 * Ignore soft cancel requests after the job is already done
755 * (We will still invoke job->driver->cancel() above, but if the
756 * job driver supports soft cancelling and the job is done, that
757 * should be a no-op, too. We still call it so it can override
758 * @force.)
759 */
760 if (force || !job->deferred_to_main_loop) {
761 job->cancelled = true;
762 /* To prevent 'force == false' overriding a previous 'force == true' */
763 job->force_cancel |= force;
764 }
765 }
766
767 static void job_completed_txn_abort(Job *job)
768 {
769 AioContext *ctx;
770 JobTxn *txn = job->txn;
771 Job *other_job;
772
773 if (txn->aborting) {
774 /*
775 * We are cancelled by another job, which will handle everything.
776 */
777 return;
778 }
779 txn->aborting = true;
780 job_txn_ref(txn);
781
782 /*
783 * We can only hold the single job's AioContext lock while calling
784 * job_finalize_single() because the finalization callbacks can involve
785 * calls of AIO_WAIT_WHILE(), which could deadlock otherwise.
786 * Note that the job's AioContext may change when it is finalized.
787 */
788 job_ref(job);
789 aio_context_release(job->aio_context);
790
791 /* Other jobs are effectively cancelled by us, set the status for
792 * them; this job, however, may or may not be cancelled, depending
793 * on the caller, so leave it. */
794 QLIST_FOREACH(other_job, &txn->jobs, txn_list) {
795 if (other_job != job) {
796 ctx = other_job->aio_context;
797 aio_context_acquire(ctx);
798 /*
799 * This is a transaction: If one job failed, no result will matter.
800 * Therefore, pass force=true to terminate all other jobs as quickly
801 * as possible.
802 */
803 job_cancel_async(other_job, true);
804 aio_context_release(ctx);
805 }
806 }
807 while (!QLIST_EMPTY(&txn->jobs)) {
808 other_job = QLIST_FIRST(&txn->jobs);
809 /*
810 * The job's AioContext may change, so store it in @ctx so we
811 * release the same context that we have acquired before.
812 */
813 ctx = other_job->aio_context;
814 aio_context_acquire(ctx);
815 if (!job_is_completed(other_job)) {
816 assert(job_cancel_requested(other_job));
817 job_finish_sync(other_job, NULL, NULL);
818 }
819 job_finalize_single(other_job);
820 aio_context_release(ctx);
821 }
822
823 /*
824 * Use job_ref()/job_unref() so we can read the AioContext here
825 * even if the job went away during job_finalize_single().
826 */
827 aio_context_acquire(job->aio_context);
828 job_unref(job);
829
830 job_txn_unref(txn);
831 }
832
833 static int job_prepare(Job *job)
834 {
835 GLOBAL_STATE_CODE();
836 if (job->ret == 0 && job->driver->prepare) {
837 job->ret = job->driver->prepare(job);
838 job_update_rc(job);
839 }
840 return job->ret;
841 }
842
843 static int job_needs_finalize(Job *job)
844 {
845 return !job->auto_finalize;
846 }
847
848 static void job_do_finalize(Job *job)
849 {
850 int rc;
851 assert(job && job->txn);
852
853 /* prepare the transaction to complete */
854 rc = job_txn_apply(job, job_prepare);
855 if (rc) {
856 job_completed_txn_abort(job);
857 } else {
858 job_txn_apply(job, job_finalize_single);
859 }
860 }
861
862 void job_finalize(Job *job, Error **errp)
863 {
864 assert(job && job->id);
865 if (job_apply_verb(job, JOB_VERB_FINALIZE, errp)) {
866 return;
867 }
868 job_do_finalize(job);
869 }
870
871 static int job_transition_to_pending(Job *job)
872 {
873 job_state_transition(job, JOB_STATUS_PENDING);
874 if (!job->auto_finalize) {
875 job_event_pending(job);
876 }
877 return 0;
878 }
879
880 void job_transition_to_ready(Job *job)
881 {
882 job_state_transition(job, JOB_STATUS_READY);
883 job_event_ready(job);
884 }
885
886 static void job_completed_txn_success(Job *job)
887 {
888 JobTxn *txn = job->txn;
889 Job *other_job;
890
891 job_state_transition(job, JOB_STATUS_WAITING);
892
893 /*
894 * Successful completion, see if there are other running jobs in this
895 * txn.
896 */
897 QLIST_FOREACH(other_job, &txn->jobs, txn_list) {
898 if (!job_is_completed(other_job)) {
899 return;
900 }
901 assert(other_job->ret == 0);
902 }
903
904 job_txn_apply(job, job_transition_to_pending);
905
906 /* If no jobs need manual finalization, automatically do so */
907 if (job_txn_apply(job, job_needs_finalize) == 0) {
908 job_do_finalize(job);
909 }
910 }
911
912 static void job_completed(Job *job)
913 {
914 assert(job && job->txn && !job_is_completed(job));
915
916 job_update_rc(job);
917 trace_job_completed(job, job->ret);
918 if (job->ret) {
919 job_completed_txn_abort(job);
920 } else {
921 job_completed_txn_success(job);
922 }
923 }
924
925 /** Useful only as a type shim for aio_bh_schedule_oneshot. */
926 static void job_exit(void *opaque)
927 {
928 Job *job = (Job *)opaque;
929 AioContext *ctx;
930
931 job_ref(job);
932 aio_context_acquire(job->aio_context);
933
934 /* This is a lie, we're not quiescent, but still doing the completion
935 * callbacks. However, completion callbacks tend to involve operations that
936 * drain block nodes, and if .drained_poll still returned true, we would
937 * deadlock. */
938 job->busy = false;
939 job_event_idle(job);
940
941 job_completed(job);
942
943 /*
944 * Note that calling job_completed can move the job to a different
945 * aio_context, so we cannot cache from above. job_txn_apply takes care of
946 * acquiring the new lock, and we ref/unref to avoid job_completed freeing
947 * the job underneath us.
948 */
949 ctx = job->aio_context;
950 job_unref(job);
951 aio_context_release(ctx);
952 }
953
954 /**
955 * All jobs must allow a pause point before entering their job proper. This
956 * ensures that jobs can be paused prior to being started, then resumed later.
957 */
958 static void coroutine_fn job_co_entry(void *opaque)
959 {
960 Job *job = opaque;
961
962 assert(job && job->driver && job->driver->run);
963 assert(job->aio_context == qemu_get_current_aio_context());
964 job_pause_point(job);
965 job->ret = job->driver->run(job, &job->err);
966 job->deferred_to_main_loop = true;
967 job->busy = true;
968 aio_bh_schedule_oneshot(qemu_get_aio_context(), job_exit, job);
969 }
970
971 void job_start(Job *job)
972 {
973 assert(job && !job_started(job) && job->paused &&
974 job->driver && job->driver->run);
975 job->co = qemu_coroutine_create(job_co_entry, job);
976 job->pause_count--;
977 job->busy = true;
978 job->paused = false;
979 job_state_transition(job, JOB_STATUS_RUNNING);
980 aio_co_enter(job->aio_context, job->co);
981 }
982
983 void job_cancel(Job *job, bool force)
984 {
985 if (job->status == JOB_STATUS_CONCLUDED) {
986 job_do_dismiss(job);
987 return;
988 }
989 job_cancel_async(job, force);
990 if (!job_started(job)) {
991 job_completed(job);
992 } else if (job->deferred_to_main_loop) {
993 /*
994 * job_cancel_async() ignores soft-cancel requests for jobs
995 * that are already done (i.e. deferred to the main loop). We
996 * have to check again whether the job is really cancelled.
997 * (job_cancel_requested() and job_is_cancelled() are equivalent
998 * here, because job_cancel_async() will make soft-cancel
999 * requests no-ops when deferred_to_main_loop is true. We
1000 * choose to call job_is_cancelled() to show that we invoke
1001 * job_completed_txn_abort() only for force-cancelled jobs.)
1002 */
1003 if (job_is_cancelled(job)) {
1004 job_completed_txn_abort(job);
1005 }
1006 } else {
1007 job_enter(job);
1008 }
1009 }
1010
1011 void job_user_cancel(Job *job, bool force, Error **errp)
1012 {
1013 if (job_apply_verb(job, JOB_VERB_CANCEL, errp)) {
1014 return;
1015 }
1016 job_cancel(job, force);
1017 }
1018
1019 /* A wrapper around job_cancel() taking an Error ** parameter so it may be
1020 * used with job_finish_sync() without the need for (rather nasty) function
1021 * pointer casts there. */
1022 static void job_cancel_err(Job *job, Error **errp)
1023 {
1024 job_cancel(job, false);
1025 }
1026
1027 /**
1028 * Same as job_cancel_err(), but force-cancel.
1029 */
1030 static void job_force_cancel_err(Job *job, Error **errp)
1031 {
1032 job_cancel(job, true);
1033 }
1034
1035 int job_cancel_sync(Job *job, bool force)
1036 {
1037 if (force) {
1038 return job_finish_sync(job, &job_force_cancel_err, NULL);
1039 } else {
1040 return job_finish_sync(job, &job_cancel_err, NULL);
1041 }
1042 }
1043
1044 void job_cancel_sync_all(void)
1045 {
1046 Job *job;
1047 AioContext *aio_context;
1048
1049 while ((job = job_next(NULL))) {
1050 aio_context = job->aio_context;
1051 aio_context_acquire(aio_context);
1052 job_cancel_sync(job, true);
1053 aio_context_release(aio_context);
1054 }
1055 }
1056
1057 int job_complete_sync(Job *job, Error **errp)
1058 {
1059 return job_finish_sync(job, job_complete, errp);
1060 }
1061
1062 void job_complete(Job *job, Error **errp)
1063 {
1064 /* Should not be reachable via external interface for internal jobs */
1065 assert(job->id);
1066 GLOBAL_STATE_CODE();
1067 if (job_apply_verb(job, JOB_VERB_COMPLETE, errp)) {
1068 return;
1069 }
1070 if (job_cancel_requested(job) || !job->driver->complete) {
1071 error_setg(errp, "The active block job '%s' cannot be completed",
1072 job->id);
1073 return;
1074 }
1075
1076 job->driver->complete(job, errp);
1077 }
1078
1079 int job_finish_sync(Job *job, void (*finish)(Job *, Error **errp), Error **errp)
1080 {
1081 Error *local_err = NULL;
1082 int ret;
1083
1084 job_ref(job);
1085
1086 if (finish) {
1087 finish(job, &local_err);
1088 }
1089 if (local_err) {
1090 error_propagate(errp, local_err);
1091 job_unref(job);
1092 return -EBUSY;
1093 }
1094
1095 AIO_WAIT_WHILE(job->aio_context,
1096 (job_enter(job), !job_is_completed(job)));
1097
1098 ret = (job_is_cancelled(job) && job->ret == 0) ? -ECANCELED : job->ret;
1099 job_unref(job);
1100 return ret;
1101 }