xref: /openbmc/qemu/job.c (revision 30b6852c)
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 
356     job_state_transition(job, JOB_STATUS_CREATED);
357     aio_timer_init(qemu_get_aio_context(), &job->sleep_timer,
358                    QEMU_CLOCK_REALTIME, SCALE_NS,
359                    job_sleep_timer_cb, job);
360 
361     QLIST_INSERT_HEAD(&jobs, job, job_list);
362 
363     /* Single jobs are modeled as single-job transactions for sake of
364      * consolidating the job management logic */
365     if (!txn) {
366         txn = job_txn_new();
367         job_txn_add_job(txn, job);
368         job_txn_unref(txn);
369     } else {
370         job_txn_add_job(txn, job);
371     }
372 
373     return job;
374 }
375 
376 void job_ref(Job *job)
377 {
378     ++job->refcnt;
379 }
380 
381 void job_unref(Job *job)
382 {
383     if (--job->refcnt == 0) {
384         assert(job->status == JOB_STATUS_NULL);
385         assert(!timer_pending(&job->sleep_timer));
386         assert(!job->txn);
387 
388         if (job->driver->free) {
389             job->driver->free(job);
390         }
391 
392         QLIST_REMOVE(job, job_list);
393 
394         progress_destroy(&job->progress);
395         error_free(job->err);
396         g_free(job->id);
397         g_free(job);
398     }
399 }
400 
401 void job_progress_update(Job *job, uint64_t done)
402 {
403     progress_work_done(&job->progress, done);
404 }
405 
406 void job_progress_set_remaining(Job *job, uint64_t remaining)
407 {
408     progress_set_remaining(&job->progress, remaining);
409 }
410 
411 void job_progress_increase_remaining(Job *job, uint64_t delta)
412 {
413     progress_increase_remaining(&job->progress, delta);
414 }
415 
416 void job_event_cancelled(Job *job)
417 {
418     notifier_list_notify(&job->on_finalize_cancelled, job);
419 }
420 
421 void job_event_completed(Job *job)
422 {
423     notifier_list_notify(&job->on_finalize_completed, job);
424 }
425 
426 static void job_event_pending(Job *job)
427 {
428     notifier_list_notify(&job->on_pending, job);
429 }
430 
431 static void job_event_ready(Job *job)
432 {
433     notifier_list_notify(&job->on_ready, job);
434 }
435 
436 static void job_event_idle(Job *job)
437 {
438     notifier_list_notify(&job->on_idle, job);
439 }
440 
441 void job_enter_cond(Job *job, bool(*fn)(Job *job))
442 {
443     if (!job_started(job)) {
444         return;
445     }
446     if (job->deferred_to_main_loop) {
447         return;
448     }
449 
450     job_lock();
451     if (job->busy) {
452         job_unlock();
453         return;
454     }
455 
456     if (fn && !fn(job)) {
457         job_unlock();
458         return;
459     }
460 
461     assert(!job->deferred_to_main_loop);
462     timer_del(&job->sleep_timer);
463     job->busy = true;
464     job_unlock();
465     aio_co_enter(job->aio_context, job->co);
466 }
467 
468 void job_enter(Job *job)
469 {
470     job_enter_cond(job, NULL);
471 }
472 
473 /* Yield, and schedule a timer to reenter the coroutine after @ns nanoseconds.
474  * Reentering the job coroutine with job_enter() before the timer has expired
475  * is allowed and cancels the timer.
476  *
477  * If @ns is (uint64_t) -1, no timer is scheduled and job_enter() must be
478  * called explicitly. */
479 static void coroutine_fn job_do_yield(Job *job, uint64_t ns)
480 {
481     job_lock();
482     if (ns != -1) {
483         timer_mod(&job->sleep_timer, ns);
484     }
485     job->busy = false;
486     job_event_idle(job);
487     job_unlock();
488     qemu_coroutine_yield();
489 
490     /* Set by job_enter_cond() before re-entering the coroutine.  */
491     assert(job->busy);
492 }
493 
494 void coroutine_fn job_pause_point(Job *job)
495 {
496     assert(job && job_started(job));
497 
498     if (!job_should_pause(job)) {
499         return;
500     }
501     if (job_is_cancelled(job)) {
502         return;
503     }
504 
505     if (job->driver->pause) {
506         job->driver->pause(job);
507     }
508 
509     if (job_should_pause(job) && !job_is_cancelled(job)) {
510         JobStatus status = job->status;
511         job_state_transition(job, status == JOB_STATUS_READY
512                                   ? JOB_STATUS_STANDBY
513                                   : JOB_STATUS_PAUSED);
514         job->paused = true;
515         job_do_yield(job, -1);
516         job->paused = false;
517         job_state_transition(job, status);
518     }
519 
520     if (job->driver->resume) {
521         job->driver->resume(job);
522     }
523 }
524 
525 void job_yield(Job *job)
526 {
527     assert(job->busy);
528 
529     /* Check cancellation *before* setting busy = false, too!  */
530     if (job_is_cancelled(job)) {
531         return;
532     }
533 
534     if (!job_should_pause(job)) {
535         job_do_yield(job, -1);
536     }
537 
538     job_pause_point(job);
539 }
540 
541 void coroutine_fn job_sleep_ns(Job *job, int64_t ns)
542 {
543     assert(job->busy);
544 
545     /* Check cancellation *before* setting busy = false, too!  */
546     if (job_is_cancelled(job)) {
547         return;
548     }
549 
550     if (!job_should_pause(job)) {
551         job_do_yield(job, qemu_clock_get_ns(QEMU_CLOCK_REALTIME) + ns);
552     }
553 
554     job_pause_point(job);
555 }
556 
557 /* Assumes the block_job_mutex is held */
558 static bool job_timer_not_pending(Job *job)
559 {
560     return !timer_pending(&job->sleep_timer);
561 }
562 
563 void job_pause(Job *job)
564 {
565     job->pause_count++;
566     if (!job->paused) {
567         job_enter(job);
568     }
569 }
570 
571 void job_resume(Job *job)
572 {
573     assert(job->pause_count > 0);
574     job->pause_count--;
575     if (job->pause_count) {
576         return;
577     }
578 
579     /* kick only if no timer is pending */
580     job_enter_cond(job, job_timer_not_pending);
581 }
582 
583 void job_user_pause(Job *job, Error **errp)
584 {
585     if (job_apply_verb(job, JOB_VERB_PAUSE, errp)) {
586         return;
587     }
588     if (job->user_paused) {
589         error_setg(errp, "Job is already paused");
590         return;
591     }
592     job->user_paused = true;
593     job_pause(job);
594 }
595 
596 bool job_user_paused(Job *job)
597 {
598     return job->user_paused;
599 }
600 
601 void job_user_resume(Job *job, Error **errp)
602 {
603     assert(job);
604     if (!job->user_paused || job->pause_count <= 0) {
605         error_setg(errp, "Can't resume a job that was not paused");
606         return;
607     }
608     if (job_apply_verb(job, JOB_VERB_RESUME, errp)) {
609         return;
610     }
611     if (job->driver->user_resume) {
612         job->driver->user_resume(job);
613     }
614     job->user_paused = false;
615     job_resume(job);
616 }
617 
618 static void job_do_dismiss(Job *job)
619 {
620     assert(job);
621     job->busy = false;
622     job->paused = false;
623     job->deferred_to_main_loop = true;
624 
625     job_txn_del_job(job);
626 
627     job_state_transition(job, JOB_STATUS_NULL);
628     job_unref(job);
629 }
630 
631 void job_dismiss(Job **jobptr, Error **errp)
632 {
633     Job *job = *jobptr;
634     /* similarly to _complete, this is QMP-interface only. */
635     assert(job->id);
636     if (job_apply_verb(job, JOB_VERB_DISMISS, errp)) {
637         return;
638     }
639 
640     job_do_dismiss(job);
641     *jobptr = NULL;
642 }
643 
644 void job_early_fail(Job *job)
645 {
646     assert(job->status == JOB_STATUS_CREATED);
647     job_do_dismiss(job);
648 }
649 
650 static void job_conclude(Job *job)
651 {
652     job_state_transition(job, JOB_STATUS_CONCLUDED);
653     if (job->auto_dismiss || !job_started(job)) {
654         job_do_dismiss(job);
655     }
656 }
657 
658 static void job_update_rc(Job *job)
659 {
660     if (!job->ret && job_is_cancelled(job)) {
661         job->ret = -ECANCELED;
662     }
663     if (job->ret) {
664         if (!job->err) {
665             error_setg(&job->err, "%s", strerror(-job->ret));
666         }
667         job_state_transition(job, JOB_STATUS_ABORTING);
668     }
669 }
670 
671 static void job_commit(Job *job)
672 {
673     assert(!job->ret);
674     if (job->driver->commit) {
675         job->driver->commit(job);
676     }
677 }
678 
679 static void job_abort(Job *job)
680 {
681     assert(job->ret);
682     if (job->driver->abort) {
683         job->driver->abort(job);
684     }
685 }
686 
687 static void job_clean(Job *job)
688 {
689     if (job->driver->clean) {
690         job->driver->clean(job);
691     }
692 }
693 
694 static int job_finalize_single(Job *job)
695 {
696     assert(job_is_completed(job));
697 
698     /* Ensure abort is called for late-transactional failures */
699     job_update_rc(job);
700 
701     if (!job->ret) {
702         job_commit(job);
703     } else {
704         job_abort(job);
705     }
706     job_clean(job);
707 
708     if (job->cb) {
709         job->cb(job->opaque, job->ret);
710     }
711 
712     /* Emit events only if we actually started */
713     if (job_started(job)) {
714         if (job_is_cancelled(job)) {
715             job_event_cancelled(job);
716         } else {
717             job_event_completed(job);
718         }
719     }
720 
721     job_txn_del_job(job);
722     job_conclude(job);
723     return 0;
724 }
725 
726 static void job_cancel_async(Job *job, bool force)
727 {
728     if (job->driver->cancel) {
729         force = job->driver->cancel(job, force);
730     } else {
731         /* No .cancel() means the job will behave as if force-cancelled */
732         force = true;
733     }
734 
735     if (job->user_paused) {
736         /* Do not call job_enter here, the caller will handle it.  */
737         if (job->driver->user_resume) {
738             job->driver->user_resume(job);
739         }
740         job->user_paused = false;
741         assert(job->pause_count > 0);
742         job->pause_count--;
743     }
744 
745     /*
746      * Ignore soft cancel requests after the job is already done
747      * (We will still invoke job->driver->cancel() above, but if the
748      * job driver supports soft cancelling and the job is done, that
749      * should be a no-op, too.  We still call it so it can override
750      * @force.)
751      */
752     if (force || !job->deferred_to_main_loop) {
753         job->cancelled = true;
754         /* To prevent 'force == false' overriding a previous 'force == true' */
755         job->force_cancel |= force;
756     }
757 }
758 
759 static void job_completed_txn_abort(Job *job)
760 {
761     AioContext *ctx;
762     JobTxn *txn = job->txn;
763     Job *other_job;
764 
765     if (txn->aborting) {
766         /*
767          * We are cancelled by another job, which will handle everything.
768          */
769         return;
770     }
771     txn->aborting = true;
772     job_txn_ref(txn);
773 
774     /*
775      * We can only hold the single job's AioContext lock while calling
776      * job_finalize_single() because the finalization callbacks can involve
777      * calls of AIO_WAIT_WHILE(), which could deadlock otherwise.
778      * Note that the job's AioContext may change when it is finalized.
779      */
780     job_ref(job);
781     aio_context_release(job->aio_context);
782 
783     /* Other jobs are effectively cancelled by us, set the status for
784      * them; this job, however, may or may not be cancelled, depending
785      * on the caller, so leave it. */
786     QLIST_FOREACH(other_job, &txn->jobs, txn_list) {
787         if (other_job != job) {
788             ctx = other_job->aio_context;
789             aio_context_acquire(ctx);
790             /*
791              * This is a transaction: If one job failed, no result will matter.
792              * Therefore, pass force=true to terminate all other jobs as quickly
793              * as possible.
794              */
795             job_cancel_async(other_job, true);
796             aio_context_release(ctx);
797         }
798     }
799     while (!QLIST_EMPTY(&txn->jobs)) {
800         other_job = QLIST_FIRST(&txn->jobs);
801         /*
802          * The job's AioContext may change, so store it in @ctx so we
803          * release the same context that we have acquired before.
804          */
805         ctx = other_job->aio_context;
806         aio_context_acquire(ctx);
807         if (!job_is_completed(other_job)) {
808             assert(job_cancel_requested(other_job));
809             job_finish_sync(other_job, NULL, NULL);
810         }
811         job_finalize_single(other_job);
812         aio_context_release(ctx);
813     }
814 
815     /*
816      * Use job_ref()/job_unref() so we can read the AioContext here
817      * even if the job went away during job_finalize_single().
818      */
819     aio_context_acquire(job->aio_context);
820     job_unref(job);
821 
822     job_txn_unref(txn);
823 }
824 
825 static int job_prepare(Job *job)
826 {
827     if (job->ret == 0 && job->driver->prepare) {
828         job->ret = job->driver->prepare(job);
829         job_update_rc(job);
830     }
831     return job->ret;
832 }
833 
834 static int job_needs_finalize(Job *job)
835 {
836     return !job->auto_finalize;
837 }
838 
839 static void job_do_finalize(Job *job)
840 {
841     int rc;
842     assert(job && job->txn);
843 
844     /* prepare the transaction to complete */
845     rc = job_txn_apply(job, job_prepare);
846     if (rc) {
847         job_completed_txn_abort(job);
848     } else {
849         job_txn_apply(job, job_finalize_single);
850     }
851 }
852 
853 void job_finalize(Job *job, Error **errp)
854 {
855     assert(job && job->id);
856     if (job_apply_verb(job, JOB_VERB_FINALIZE, errp)) {
857         return;
858     }
859     job_do_finalize(job);
860 }
861 
862 static int job_transition_to_pending(Job *job)
863 {
864     job_state_transition(job, JOB_STATUS_PENDING);
865     if (!job->auto_finalize) {
866         job_event_pending(job);
867     }
868     return 0;
869 }
870 
871 void job_transition_to_ready(Job *job)
872 {
873     job_state_transition(job, JOB_STATUS_READY);
874     job_event_ready(job);
875 }
876 
877 static void job_completed_txn_success(Job *job)
878 {
879     JobTxn *txn = job->txn;
880     Job *other_job;
881 
882     job_state_transition(job, JOB_STATUS_WAITING);
883 
884     /*
885      * Successful completion, see if there are other running jobs in this
886      * txn.
887      */
888     QLIST_FOREACH(other_job, &txn->jobs, txn_list) {
889         if (!job_is_completed(other_job)) {
890             return;
891         }
892         assert(other_job->ret == 0);
893     }
894 
895     job_txn_apply(job, job_transition_to_pending);
896 
897     /* If no jobs need manual finalization, automatically do so */
898     if (job_txn_apply(job, job_needs_finalize) == 0) {
899         job_do_finalize(job);
900     }
901 }
902 
903 static void job_completed(Job *job)
904 {
905     assert(job && job->txn && !job_is_completed(job));
906 
907     job_update_rc(job);
908     trace_job_completed(job, job->ret);
909     if (job->ret) {
910         job_completed_txn_abort(job);
911     } else {
912         job_completed_txn_success(job);
913     }
914 }
915 
916 /** Useful only as a type shim for aio_bh_schedule_oneshot. */
917 static void job_exit(void *opaque)
918 {
919     Job *job = (Job *)opaque;
920     AioContext *ctx;
921 
922     job_ref(job);
923     aio_context_acquire(job->aio_context);
924 
925     /* This is a lie, we're not quiescent, but still doing the completion
926      * callbacks. However, completion callbacks tend to involve operations that
927      * drain block nodes, and if .drained_poll still returned true, we would
928      * deadlock. */
929     job->busy = false;
930     job_event_idle(job);
931 
932     job_completed(job);
933 
934     /*
935      * Note that calling job_completed can move the job to a different
936      * aio_context, so we cannot cache from above. job_txn_apply takes care of
937      * acquiring the new lock, and we ref/unref to avoid job_completed freeing
938      * the job underneath us.
939      */
940     ctx = job->aio_context;
941     job_unref(job);
942     aio_context_release(ctx);
943 }
944 
945 /**
946  * All jobs must allow a pause point before entering their job proper. This
947  * ensures that jobs can be paused prior to being started, then resumed later.
948  */
949 static void coroutine_fn job_co_entry(void *opaque)
950 {
951     Job *job = opaque;
952 
953     assert(job && job->driver && job->driver->run);
954     job_pause_point(job);
955     job->ret = job->driver->run(job, &job->err);
956     job->deferred_to_main_loop = true;
957     job->busy = true;
958     aio_bh_schedule_oneshot(qemu_get_aio_context(), job_exit, job);
959 }
960 
961 void job_start(Job *job)
962 {
963     assert(job && !job_started(job) && job->paused &&
964            job->driver && job->driver->run);
965     job->co = qemu_coroutine_create(job_co_entry, job);
966     job->pause_count--;
967     job->busy = true;
968     job->paused = false;
969     job_state_transition(job, JOB_STATUS_RUNNING);
970     aio_co_enter(job->aio_context, job->co);
971 }
972 
973 void job_cancel(Job *job, bool force)
974 {
975     if (job->status == JOB_STATUS_CONCLUDED) {
976         job_do_dismiss(job);
977         return;
978     }
979     job_cancel_async(job, force);
980     if (!job_started(job)) {
981         job_completed(job);
982     } else if (job->deferred_to_main_loop) {
983         /*
984          * job_cancel_async() ignores soft-cancel requests for jobs
985          * that are already done (i.e. deferred to the main loop).  We
986          * have to check again whether the job is really cancelled.
987          * (job_cancel_requested() and job_is_cancelled() are equivalent
988          * here, because job_cancel_async() will make soft-cancel
989          * requests no-ops when deferred_to_main_loop is true.  We
990          * choose to call job_is_cancelled() to show that we invoke
991          * job_completed_txn_abort() only for force-cancelled jobs.)
992          */
993         if (job_is_cancelled(job)) {
994             job_completed_txn_abort(job);
995         }
996     } else {
997         job_enter(job);
998     }
999 }
1000 
1001 void job_user_cancel(Job *job, bool force, Error **errp)
1002 {
1003     if (job_apply_verb(job, JOB_VERB_CANCEL, errp)) {
1004         return;
1005     }
1006     job_cancel(job, force);
1007 }
1008 
1009 /* A wrapper around job_cancel() taking an Error ** parameter so it may be
1010  * used with job_finish_sync() without the need for (rather nasty) function
1011  * pointer casts there. */
1012 static void job_cancel_err(Job *job, Error **errp)
1013 {
1014     job_cancel(job, false);
1015 }
1016 
1017 /**
1018  * Same as job_cancel_err(), but force-cancel.
1019  */
1020 static void job_force_cancel_err(Job *job, Error **errp)
1021 {
1022     job_cancel(job, true);
1023 }
1024 
1025 int job_cancel_sync(Job *job, bool force)
1026 {
1027     if (force) {
1028         return job_finish_sync(job, &job_force_cancel_err, NULL);
1029     } else {
1030         return job_finish_sync(job, &job_cancel_err, NULL);
1031     }
1032 }
1033 
1034 void job_cancel_sync_all(void)
1035 {
1036     Job *job;
1037     AioContext *aio_context;
1038 
1039     while ((job = job_next(NULL))) {
1040         aio_context = job->aio_context;
1041         aio_context_acquire(aio_context);
1042         job_cancel_sync(job, true);
1043         aio_context_release(aio_context);
1044     }
1045 }
1046 
1047 int job_complete_sync(Job *job, Error **errp)
1048 {
1049     return job_finish_sync(job, job_complete, errp);
1050 }
1051 
1052 void job_complete(Job *job, Error **errp)
1053 {
1054     /* Should not be reachable via external interface for internal jobs */
1055     assert(job->id);
1056     if (job_apply_verb(job, JOB_VERB_COMPLETE, errp)) {
1057         return;
1058     }
1059     if (job_cancel_requested(job) || !job->driver->complete) {
1060         error_setg(errp, "The active block job '%s' cannot be completed",
1061                    job->id);
1062         return;
1063     }
1064 
1065     job->driver->complete(job, errp);
1066 }
1067 
1068 int job_finish_sync(Job *job, void (*finish)(Job *, Error **errp), Error **errp)
1069 {
1070     Error *local_err = NULL;
1071     int ret;
1072 
1073     job_ref(job);
1074 
1075     if (finish) {
1076         finish(job, &local_err);
1077     }
1078     if (local_err) {
1079         error_propagate(errp, local_err);
1080         job_unref(job);
1081         return -EBUSY;
1082     }
1083 
1084     AIO_WAIT_WHILE(job->aio_context,
1085                    (job_enter(job), !job_is_completed(job)));
1086 
1087     ret = (job_is_cancelled(job) && job->ret == 0) ? -ECANCELED : job->ret;
1088     job_unref(job);
1089     return ret;
1090 }
1091