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-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, 0, 0, 0, 0, 0, 0}, |
60 | [JOB_VERB_FINALIZE] = {0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0}, |
61 | [JOB_VERB_DISMISS] = {0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0}, |
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(JobTxn *txn, int fn(Job *)) |
140 | { |
141 | Job *job, *next; |
142 | int rc = 0; |
143 | |
144 | QLIST_FOREACH_SAFE(job, &txn->jobs, txn_list, next) { |
145 | rc = fn(job); |
146 | if (rc) { |
147 | break; |
148 | } |
149 | } |
150 | return rc; |
151 | } |
152 | |
153 | bool job_is_internal(Job *job) |
154 | { |
155 | return (job->id == NULL); |
156 | } |
157 | |
158 | static void job_state_transition(Job *job, JobStatus s1) |
159 | { |
160 | JobStatus s0 = job->status; |
161 | assert(s1 >= 0 && s1 < JOB_STATUS__MAX); |
162 | trace_job_state_transition(job, job->ret, |
163 | JobSTT[s0][s1] ? "allowed" : "disallowed" , |
164 | JobStatus_str(s0), JobStatus_str(s1)); |
165 | assert(JobSTT[s0][s1]); |
166 | job->status = s1; |
167 | |
168 | if (!job_is_internal(job) && s1 != s0) { |
169 | qapi_event_send_job_status_change(job->id, job->status); |
170 | } |
171 | } |
172 | |
173 | int job_apply_verb(Job *job, JobVerb verb, Error **errp) |
174 | { |
175 | JobStatus s0 = job->status; |
176 | assert(verb >= 0 && verb < JOB_VERB__MAX); |
177 | trace_job_apply_verb(job, JobStatus_str(s0), JobVerb_str(verb), |
178 | JobVerbTable[verb][s0] ? "allowed" : "prohibited" ); |
179 | if (JobVerbTable[verb][s0]) { |
180 | return 0; |
181 | } |
182 | error_setg(errp, "Job '%s' in state '%s' cannot accept command verb '%s'" , |
183 | job->id, JobStatus_str(s0), JobVerb_str(verb)); |
184 | return -EPERM; |
185 | } |
186 | |
187 | JobType job_type(const Job *job) |
188 | { |
189 | return job->driver->job_type; |
190 | } |
191 | |
192 | const char *job_type_str(const Job *job) |
193 | { |
194 | return JobType_str(job_type(job)); |
195 | } |
196 | |
197 | bool job_is_cancelled(Job *job) |
198 | { |
199 | return job->cancelled; |
200 | } |
201 | |
202 | bool job_is_ready(Job *job) |
203 | { |
204 | switch (job->status) { |
205 | case JOB_STATUS_UNDEFINED: |
206 | case JOB_STATUS_CREATED: |
207 | case JOB_STATUS_RUNNING: |
208 | case JOB_STATUS_PAUSED: |
209 | case JOB_STATUS_WAITING: |
210 | case JOB_STATUS_PENDING: |
211 | case JOB_STATUS_ABORTING: |
212 | case JOB_STATUS_CONCLUDED: |
213 | case JOB_STATUS_NULL: |
214 | return false; |
215 | case JOB_STATUS_READY: |
216 | case JOB_STATUS_STANDBY: |
217 | return true; |
218 | default: |
219 | g_assert_not_reached(); |
220 | } |
221 | return false; |
222 | } |
223 | |
224 | bool job_is_completed(Job *job) |
225 | { |
226 | switch (job->status) { |
227 | case JOB_STATUS_UNDEFINED: |
228 | case JOB_STATUS_CREATED: |
229 | case JOB_STATUS_RUNNING: |
230 | case JOB_STATUS_PAUSED: |
231 | case JOB_STATUS_READY: |
232 | case JOB_STATUS_STANDBY: |
233 | return false; |
234 | case JOB_STATUS_WAITING: |
235 | case JOB_STATUS_PENDING: |
236 | case JOB_STATUS_ABORTING: |
237 | case JOB_STATUS_CONCLUDED: |
238 | case JOB_STATUS_NULL: |
239 | return true; |
240 | default: |
241 | g_assert_not_reached(); |
242 | } |
243 | return false; |
244 | } |
245 | |
246 | static bool job_started(Job *job) |
247 | { |
248 | return job->co; |
249 | } |
250 | |
251 | static bool job_should_pause(Job *job) |
252 | { |
253 | return job->pause_count > 0; |
254 | } |
255 | |
256 | Job *job_next(Job *job) |
257 | { |
258 | if (!job) { |
259 | return QLIST_FIRST(&jobs); |
260 | } |
261 | return QLIST_NEXT(job, job_list); |
262 | } |
263 | |
264 | Job *job_get(const char *id) |
265 | { |
266 | Job *job; |
267 | |
268 | QLIST_FOREACH(job, &jobs, job_list) { |
269 | if (job->id && !strcmp(id, job->id)) { |
270 | return job; |
271 | } |
272 | } |
273 | |
274 | return NULL; |
275 | } |
276 | |
277 | static void job_sleep_timer_cb(void *opaque) |
278 | { |
279 | Job *job = opaque; |
280 | |
281 | job_enter(job); |
282 | } |
283 | |
284 | void *job_create(const char *job_id, const JobDriver *driver, JobTxn *txn, |
285 | AioContext *ctx, int flags, BlockCompletionFunc *cb, |
286 | void *opaque, Error **errp) |
287 | { |
288 | Job *job; |
289 | |
290 | if (job_id) { |
291 | if (flags & JOB_INTERNAL) { |
292 | error_setg(errp, "Cannot specify job ID for internal job" ); |
293 | return NULL; |
294 | } |
295 | if (!id_wellformed(job_id)) { |
296 | error_setg(errp, "Invalid job ID '%s'" , job_id); |
297 | return NULL; |
298 | } |
299 | if (job_get(job_id)) { |
300 | error_setg(errp, "Job ID '%s' already in use" , job_id); |
301 | return NULL; |
302 | } |
303 | } else if (!(flags & JOB_INTERNAL)) { |
304 | error_setg(errp, "An explicit job ID is required" ); |
305 | return NULL; |
306 | } |
307 | |
308 | job = g_malloc0(driver->instance_size); |
309 | job->driver = driver; |
310 | job->id = g_strdup(job_id); |
311 | job->refcnt = 1; |
312 | job->aio_context = ctx; |
313 | job->busy = false; |
314 | job->paused = true; |
315 | job->pause_count = 1; |
316 | job->auto_finalize = !(flags & JOB_MANUAL_FINALIZE); |
317 | job->auto_dismiss = !(flags & JOB_MANUAL_DISMISS); |
318 | job->cb = cb; |
319 | job->opaque = opaque; |
320 | |
321 | notifier_list_init(&job->on_finalize_cancelled); |
322 | notifier_list_init(&job->on_finalize_completed); |
323 | notifier_list_init(&job->on_pending); |
324 | notifier_list_init(&job->on_ready); |
325 | |
326 | job_state_transition(job, JOB_STATUS_CREATED); |
327 | aio_timer_init(qemu_get_aio_context(), &job->sleep_timer, |
328 | QEMU_CLOCK_REALTIME, SCALE_NS, |
329 | job_sleep_timer_cb, job); |
330 | |
331 | QLIST_INSERT_HEAD(&jobs, job, job_list); |
332 | |
333 | /* Single jobs are modeled as single-job transactions for sake of |
334 | * consolidating the job management logic */ |
335 | if (!txn) { |
336 | txn = job_txn_new(); |
337 | job_txn_add_job(txn, job); |
338 | job_txn_unref(txn); |
339 | } else { |
340 | job_txn_add_job(txn, job); |
341 | } |
342 | |
343 | return job; |
344 | } |
345 | |
346 | void job_ref(Job *job) |
347 | { |
348 | ++job->refcnt; |
349 | } |
350 | |
351 | void job_unref(Job *job) |
352 | { |
353 | if (--job->refcnt == 0) { |
354 | assert(job->status == JOB_STATUS_NULL); |
355 | assert(!timer_pending(&job->sleep_timer)); |
356 | assert(!job->txn); |
357 | |
358 | if (job->driver->free) { |
359 | job->driver->free(job); |
360 | } |
361 | |
362 | QLIST_REMOVE(job, job_list); |
363 | |
364 | error_free(job->err); |
365 | g_free(job->id); |
366 | g_free(job); |
367 | } |
368 | } |
369 | |
370 | void job_progress_update(Job *job, uint64_t done) |
371 | { |
372 | job->progress_current += done; |
373 | } |
374 | |
375 | void job_progress_set_remaining(Job *job, uint64_t remaining) |
376 | { |
377 | job->progress_total = job->progress_current + remaining; |
378 | } |
379 | |
380 | void job_progress_increase_remaining(Job *job, uint64_t delta) |
381 | { |
382 | job->progress_total += delta; |
383 | } |
384 | |
385 | void job_event_cancelled(Job *job) |
386 | { |
387 | notifier_list_notify(&job->on_finalize_cancelled, job); |
388 | } |
389 | |
390 | void job_event_completed(Job *job) |
391 | { |
392 | notifier_list_notify(&job->on_finalize_completed, job); |
393 | } |
394 | |
395 | static void job_event_pending(Job *job) |
396 | { |
397 | notifier_list_notify(&job->on_pending, job); |
398 | } |
399 | |
400 | static void job_event_ready(Job *job) |
401 | { |
402 | notifier_list_notify(&job->on_ready, job); |
403 | } |
404 | |
405 | static void job_event_idle(Job *job) |
406 | { |
407 | notifier_list_notify(&job->on_idle, job); |
408 | } |
409 | |
410 | void job_enter_cond(Job *job, bool(*fn)(Job *job)) |
411 | { |
412 | if (!job_started(job)) { |
413 | return; |
414 | } |
415 | if (job->deferred_to_main_loop) { |
416 | return; |
417 | } |
418 | |
419 | job_lock(); |
420 | if (job->busy) { |
421 | job_unlock(); |
422 | return; |
423 | } |
424 | |
425 | if (fn && !fn(job)) { |
426 | job_unlock(); |
427 | return; |
428 | } |
429 | |
430 | assert(!job->deferred_to_main_loop); |
431 | timer_del(&job->sleep_timer); |
432 | job->busy = true; |
433 | job_unlock(); |
434 | aio_co_enter(job->aio_context, job->co); |
435 | } |
436 | |
437 | void job_enter(Job *job) |
438 | { |
439 | job_enter_cond(job, NULL); |
440 | } |
441 | |
442 | /* Yield, and schedule a timer to reenter the coroutine after @ns nanoseconds. |
443 | * Reentering the job coroutine with job_enter() before the timer has expired |
444 | * is allowed and cancels the timer. |
445 | * |
446 | * If @ns is (uint64_t) -1, no timer is scheduled and job_enter() must be |
447 | * called explicitly. */ |
448 | static void coroutine_fn job_do_yield(Job *job, uint64_t ns) |
449 | { |
450 | job_lock(); |
451 | if (ns != -1) { |
452 | timer_mod(&job->sleep_timer, ns); |
453 | } |
454 | job->busy = false; |
455 | job_event_idle(job); |
456 | job_unlock(); |
457 | qemu_coroutine_yield(); |
458 | |
459 | /* Set by job_enter_cond() before re-entering the coroutine. */ |
460 | assert(job->busy); |
461 | } |
462 | |
463 | void coroutine_fn job_pause_point(Job *job) |
464 | { |
465 | assert(job && job_started(job)); |
466 | |
467 | if (!job_should_pause(job)) { |
468 | return; |
469 | } |
470 | if (job_is_cancelled(job)) { |
471 | return; |
472 | } |
473 | |
474 | if (job->driver->pause) { |
475 | job->driver->pause(job); |
476 | } |
477 | |
478 | if (job_should_pause(job) && !job_is_cancelled(job)) { |
479 | JobStatus status = job->status; |
480 | job_state_transition(job, status == JOB_STATUS_READY |
481 | ? JOB_STATUS_STANDBY |
482 | : JOB_STATUS_PAUSED); |
483 | job->paused = true; |
484 | job_do_yield(job, -1); |
485 | job->paused = false; |
486 | job_state_transition(job, status); |
487 | } |
488 | |
489 | if (job->driver->resume) { |
490 | job->driver->resume(job); |
491 | } |
492 | } |
493 | |
494 | void job_yield(Job *job) |
495 | { |
496 | assert(job->busy); |
497 | |
498 | /* Check cancellation *before* setting busy = false, too! */ |
499 | if (job_is_cancelled(job)) { |
500 | return; |
501 | } |
502 | |
503 | if (!job_should_pause(job)) { |
504 | job_do_yield(job, -1); |
505 | } |
506 | |
507 | job_pause_point(job); |
508 | } |
509 | |
510 | void coroutine_fn job_sleep_ns(Job *job, int64_t ns) |
511 | { |
512 | assert(job->busy); |
513 | |
514 | /* Check cancellation *before* setting busy = false, too! */ |
515 | if (job_is_cancelled(job)) { |
516 | return; |
517 | } |
518 | |
519 | if (!job_should_pause(job)) { |
520 | job_do_yield(job, qemu_clock_get_ns(QEMU_CLOCK_REALTIME) + ns); |
521 | } |
522 | |
523 | job_pause_point(job); |
524 | } |
525 | |
526 | void job_drain(Job *job) |
527 | { |
528 | /* If job is !busy this kicks it into the next pause point. */ |
529 | job_enter(job); |
530 | |
531 | if (job->driver->drain) { |
532 | job->driver->drain(job); |
533 | } |
534 | } |
535 | |
536 | /* Assumes the block_job_mutex is held */ |
537 | static bool job_timer_not_pending(Job *job) |
538 | { |
539 | return !timer_pending(&job->sleep_timer); |
540 | } |
541 | |
542 | void job_pause(Job *job) |
543 | { |
544 | job->pause_count++; |
545 | } |
546 | |
547 | void job_resume(Job *job) |
548 | { |
549 | assert(job->pause_count > 0); |
550 | job->pause_count--; |
551 | if (job->pause_count) { |
552 | return; |
553 | } |
554 | |
555 | /* kick only if no timer is pending */ |
556 | job_enter_cond(job, job_timer_not_pending); |
557 | } |
558 | |
559 | void job_user_pause(Job *job, Error **errp) |
560 | { |
561 | if (job_apply_verb(job, JOB_VERB_PAUSE, errp)) { |
562 | return; |
563 | } |
564 | if (job->user_paused) { |
565 | error_setg(errp, "Job is already paused" ); |
566 | return; |
567 | } |
568 | job->user_paused = true; |
569 | job_pause(job); |
570 | } |
571 | |
572 | bool job_user_paused(Job *job) |
573 | { |
574 | return job->user_paused; |
575 | } |
576 | |
577 | void job_user_resume(Job *job, Error **errp) |
578 | { |
579 | assert(job); |
580 | if (!job->user_paused || job->pause_count <= 0) { |
581 | error_setg(errp, "Can't resume a job that was not paused" ); |
582 | return; |
583 | } |
584 | if (job_apply_verb(job, JOB_VERB_RESUME, errp)) { |
585 | return; |
586 | } |
587 | if (job->driver->user_resume) { |
588 | job->driver->user_resume(job); |
589 | } |
590 | job->user_paused = false; |
591 | job_resume(job); |
592 | } |
593 | |
594 | static void job_do_dismiss(Job *job) |
595 | { |
596 | assert(job); |
597 | job->busy = false; |
598 | job->paused = false; |
599 | job->deferred_to_main_loop = true; |
600 | |
601 | job_txn_del_job(job); |
602 | |
603 | job_state_transition(job, JOB_STATUS_NULL); |
604 | job_unref(job); |
605 | } |
606 | |
607 | void job_dismiss(Job **jobptr, Error **errp) |
608 | { |
609 | Job *job = *jobptr; |
610 | /* similarly to _complete, this is QMP-interface only. */ |
611 | assert(job->id); |
612 | if (job_apply_verb(job, JOB_VERB_DISMISS, errp)) { |
613 | return; |
614 | } |
615 | |
616 | job_do_dismiss(job); |
617 | *jobptr = NULL; |
618 | } |
619 | |
620 | void job_early_fail(Job *job) |
621 | { |
622 | assert(job->status == JOB_STATUS_CREATED); |
623 | job_do_dismiss(job); |
624 | } |
625 | |
626 | static void job_conclude(Job *job) |
627 | { |
628 | job_state_transition(job, JOB_STATUS_CONCLUDED); |
629 | if (job->auto_dismiss || !job_started(job)) { |
630 | job_do_dismiss(job); |
631 | } |
632 | } |
633 | |
634 | static void job_update_rc(Job *job) |
635 | { |
636 | if (!job->ret && job_is_cancelled(job)) { |
637 | job->ret = -ECANCELED; |
638 | } |
639 | if (job->ret) { |
640 | if (!job->err) { |
641 | error_setg(&job->err, "%s" , strerror(-job->ret)); |
642 | } |
643 | job_state_transition(job, JOB_STATUS_ABORTING); |
644 | } |
645 | } |
646 | |
647 | static void job_commit(Job *job) |
648 | { |
649 | assert(!job->ret); |
650 | if (job->driver->commit) { |
651 | job->driver->commit(job); |
652 | } |
653 | } |
654 | |
655 | static void job_abort(Job *job) |
656 | { |
657 | assert(job->ret); |
658 | if (job->driver->abort) { |
659 | job->driver->abort(job); |
660 | } |
661 | } |
662 | |
663 | static void job_clean(Job *job) |
664 | { |
665 | if (job->driver->clean) { |
666 | job->driver->clean(job); |
667 | } |
668 | } |
669 | |
670 | static int job_finalize_single(Job *job) |
671 | { |
672 | assert(job_is_completed(job)); |
673 | |
674 | /* Ensure abort is called for late-transactional failures */ |
675 | job_update_rc(job); |
676 | |
677 | if (!job->ret) { |
678 | job_commit(job); |
679 | } else { |
680 | job_abort(job); |
681 | } |
682 | job_clean(job); |
683 | |
684 | if (job->cb) { |
685 | job->cb(job->opaque, job->ret); |
686 | } |
687 | |
688 | /* Emit events only if we actually started */ |
689 | if (job_started(job)) { |
690 | if (job_is_cancelled(job)) { |
691 | job_event_cancelled(job); |
692 | } else { |
693 | job_event_completed(job); |
694 | } |
695 | } |
696 | |
697 | job_txn_del_job(job); |
698 | job_conclude(job); |
699 | return 0; |
700 | } |
701 | |
702 | static void job_cancel_async(Job *job, bool force) |
703 | { |
704 | if (job->user_paused) { |
705 | /* Do not call job_enter here, the caller will handle it. */ |
706 | if (job->driver->user_resume) { |
707 | job->driver->user_resume(job); |
708 | } |
709 | job->user_paused = false; |
710 | assert(job->pause_count > 0); |
711 | job->pause_count--; |
712 | } |
713 | job->cancelled = true; |
714 | /* To prevent 'force == false' overriding a previous 'force == true' */ |
715 | job->force_cancel |= force; |
716 | } |
717 | |
718 | static void job_completed_txn_abort(Job *job) |
719 | { |
720 | AioContext *outer_ctx = job->aio_context; |
721 | AioContext *ctx; |
722 | JobTxn *txn = job->txn; |
723 | Job *other_job; |
724 | |
725 | if (txn->aborting) { |
726 | /* |
727 | * We are cancelled by another job, which will handle everything. |
728 | */ |
729 | return; |
730 | } |
731 | txn->aborting = true; |
732 | job_txn_ref(txn); |
733 | |
734 | /* We can only hold the single job's AioContext lock while calling |
735 | * job_finalize_single() because the finalization callbacks can involve |
736 | * calls of AIO_WAIT_WHILE(), which could deadlock otherwise. */ |
737 | aio_context_release(outer_ctx); |
738 | |
739 | /* Other jobs are effectively cancelled by us, set the status for |
740 | * them; this job, however, may or may not be cancelled, depending |
741 | * on the caller, so leave it. */ |
742 | QLIST_FOREACH(other_job, &txn->jobs, txn_list) { |
743 | if (other_job != job) { |
744 | ctx = other_job->aio_context; |
745 | aio_context_acquire(ctx); |
746 | job_cancel_async(other_job, false); |
747 | aio_context_release(ctx); |
748 | } |
749 | } |
750 | while (!QLIST_EMPTY(&txn->jobs)) { |
751 | other_job = QLIST_FIRST(&txn->jobs); |
752 | ctx = other_job->aio_context; |
753 | aio_context_acquire(ctx); |
754 | if (!job_is_completed(other_job)) { |
755 | assert(job_is_cancelled(other_job)); |
756 | job_finish_sync(other_job, NULL, NULL); |
757 | } |
758 | job_finalize_single(other_job); |
759 | aio_context_release(ctx); |
760 | } |
761 | |
762 | aio_context_acquire(outer_ctx); |
763 | |
764 | job_txn_unref(txn); |
765 | } |
766 | |
767 | static int job_prepare(Job *job) |
768 | { |
769 | if (job->ret == 0 && job->driver->prepare) { |
770 | job->ret = job->driver->prepare(job); |
771 | job_update_rc(job); |
772 | } |
773 | return job->ret; |
774 | } |
775 | |
776 | static int job_needs_finalize(Job *job) |
777 | { |
778 | return !job->auto_finalize; |
779 | } |
780 | |
781 | static void job_do_finalize(Job *job) |
782 | { |
783 | int rc; |
784 | assert(job && job->txn); |
785 | |
786 | /* prepare the transaction to complete */ |
787 | rc = job_txn_apply(job->txn, job_prepare); |
788 | if (rc) { |
789 | job_completed_txn_abort(job); |
790 | } else { |
791 | job_txn_apply(job->txn, job_finalize_single); |
792 | } |
793 | } |
794 | |
795 | void job_finalize(Job *job, Error **errp) |
796 | { |
797 | assert(job && job->id); |
798 | if (job_apply_verb(job, JOB_VERB_FINALIZE, errp)) { |
799 | return; |
800 | } |
801 | job_do_finalize(job); |
802 | } |
803 | |
804 | static int job_transition_to_pending(Job *job) |
805 | { |
806 | job_state_transition(job, JOB_STATUS_PENDING); |
807 | if (!job->auto_finalize) { |
808 | job_event_pending(job); |
809 | } |
810 | return 0; |
811 | } |
812 | |
813 | void job_transition_to_ready(Job *job) |
814 | { |
815 | job_state_transition(job, JOB_STATUS_READY); |
816 | job_event_ready(job); |
817 | } |
818 | |
819 | static void job_completed_txn_success(Job *job) |
820 | { |
821 | JobTxn *txn = job->txn; |
822 | Job *other_job; |
823 | |
824 | job_state_transition(job, JOB_STATUS_WAITING); |
825 | |
826 | /* |
827 | * Successful completion, see if there are other running jobs in this |
828 | * txn. |
829 | */ |
830 | QLIST_FOREACH(other_job, &txn->jobs, txn_list) { |
831 | if (!job_is_completed(other_job)) { |
832 | return; |
833 | } |
834 | assert(other_job->ret == 0); |
835 | } |
836 | |
837 | job_txn_apply(txn, job_transition_to_pending); |
838 | |
839 | /* If no jobs need manual finalization, automatically do so */ |
840 | if (job_txn_apply(txn, job_needs_finalize) == 0) { |
841 | job_do_finalize(job); |
842 | } |
843 | } |
844 | |
845 | static void job_completed(Job *job) |
846 | { |
847 | assert(job && job->txn && !job_is_completed(job)); |
848 | |
849 | job_update_rc(job); |
850 | trace_job_completed(job, job->ret); |
851 | if (job->ret) { |
852 | job_completed_txn_abort(job); |
853 | } else { |
854 | job_completed_txn_success(job); |
855 | } |
856 | } |
857 | |
858 | /** Useful only as a type shim for aio_bh_schedule_oneshot. */ |
859 | static void job_exit(void *opaque) |
860 | { |
861 | Job *job = (Job *)opaque; |
862 | AioContext *ctx = job->aio_context; |
863 | |
864 | aio_context_acquire(ctx); |
865 | |
866 | /* This is a lie, we're not quiescent, but still doing the completion |
867 | * callbacks. However, completion callbacks tend to involve operations that |
868 | * drain block nodes, and if .drained_poll still returned true, we would |
869 | * deadlock. */ |
870 | job->busy = false; |
871 | job_event_idle(job); |
872 | |
873 | job_completed(job); |
874 | |
875 | aio_context_release(ctx); |
876 | } |
877 | |
878 | /** |
879 | * All jobs must allow a pause point before entering their job proper. This |
880 | * ensures that jobs can be paused prior to being started, then resumed later. |
881 | */ |
882 | static void coroutine_fn job_co_entry(void *opaque) |
883 | { |
884 | Job *job = opaque; |
885 | |
886 | assert(job && job->driver && job->driver->run); |
887 | job_pause_point(job); |
888 | job->ret = job->driver->run(job, &job->err); |
889 | job->deferred_to_main_loop = true; |
890 | job->busy = true; |
891 | aio_bh_schedule_oneshot(qemu_get_aio_context(), job_exit, job); |
892 | } |
893 | |
894 | void job_start(Job *job) |
895 | { |
896 | assert(job && !job_started(job) && job->paused && |
897 | job->driver && job->driver->run); |
898 | job->co = qemu_coroutine_create(job_co_entry, job); |
899 | job->pause_count--; |
900 | job->busy = true; |
901 | job->paused = false; |
902 | job_state_transition(job, JOB_STATUS_RUNNING); |
903 | aio_co_enter(job->aio_context, job->co); |
904 | } |
905 | |
906 | void job_cancel(Job *job, bool force) |
907 | { |
908 | if (job->status == JOB_STATUS_CONCLUDED) { |
909 | job_do_dismiss(job); |
910 | return; |
911 | } |
912 | job_cancel_async(job, force); |
913 | if (!job_started(job)) { |
914 | job_completed(job); |
915 | } else if (job->deferred_to_main_loop) { |
916 | job_completed_txn_abort(job); |
917 | } else { |
918 | job_enter(job); |
919 | } |
920 | } |
921 | |
922 | void job_user_cancel(Job *job, bool force, Error **errp) |
923 | { |
924 | if (job_apply_verb(job, JOB_VERB_CANCEL, errp)) { |
925 | return; |
926 | } |
927 | job_cancel(job, force); |
928 | } |
929 | |
930 | /* A wrapper around job_cancel() taking an Error ** parameter so it may be |
931 | * used with job_finish_sync() without the need for (rather nasty) function |
932 | * pointer casts there. */ |
933 | static void job_cancel_err(Job *job, Error **errp) |
934 | { |
935 | job_cancel(job, false); |
936 | } |
937 | |
938 | int job_cancel_sync(Job *job) |
939 | { |
940 | return job_finish_sync(job, &job_cancel_err, NULL); |
941 | } |
942 | |
943 | void job_cancel_sync_all(void) |
944 | { |
945 | Job *job; |
946 | AioContext *aio_context; |
947 | |
948 | while ((job = job_next(NULL))) { |
949 | aio_context = job->aio_context; |
950 | aio_context_acquire(aio_context); |
951 | job_cancel_sync(job); |
952 | aio_context_release(aio_context); |
953 | } |
954 | } |
955 | |
956 | int job_complete_sync(Job *job, Error **errp) |
957 | { |
958 | return job_finish_sync(job, job_complete, errp); |
959 | } |
960 | |
961 | void job_complete(Job *job, Error **errp) |
962 | { |
963 | /* Should not be reachable via external interface for internal jobs */ |
964 | assert(job->id); |
965 | if (job_apply_verb(job, JOB_VERB_COMPLETE, errp)) { |
966 | return; |
967 | } |
968 | if (job->pause_count || job_is_cancelled(job) || !job->driver->complete) { |
969 | error_setg(errp, "The active block job '%s' cannot be completed" , |
970 | job->id); |
971 | return; |
972 | } |
973 | |
974 | job->driver->complete(job, errp); |
975 | } |
976 | |
977 | int job_finish_sync(Job *job, void (*finish)(Job *, Error **errp), Error **errp) |
978 | { |
979 | Error *local_err = NULL; |
980 | int ret; |
981 | |
982 | job_ref(job); |
983 | |
984 | if (finish) { |
985 | finish(job, &local_err); |
986 | } |
987 | if (local_err) { |
988 | error_propagate(errp, local_err); |
989 | job_unref(job); |
990 | return -EBUSY; |
991 | } |
992 | |
993 | AIO_WAIT_WHILE(job->aio_context, |
994 | (job_drain(job), !job_is_completed(job))); |
995 | |
996 | ret = (job_is_cancelled(job) && job->ret == 0) ? -ECANCELED : job->ret; |
997 | job_unref(job); |
998 | return ret; |
999 | } |
1000 | |