1 /*
2 * QEMU System Emulator block driver
3 *
4 * Copyright (c) 2011 IBM Corp.
5 * Copyright (c) 2012 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 "block/aio-wait.h"
28 #include "block/block.h"
29 #include "block/blockjob_int.h"
30 #include "block/block_int.h"
31 #include "block/trace.h"
32 #include "system/block-backend.h"
33 #include "qapi/error.h"
34 #include "qapi/qapi-events-block-core.h"
35 #include "qapi/qmp/qerror.h"
36 #include "qemu/main-loop.h"
37 #include "qemu/timer.h"
38
is_block_job(Job * job)39 static bool is_block_job(Job *job)
40 {
41 return job_type(job) == JOB_TYPE_BACKUP ||
42 job_type(job) == JOB_TYPE_COMMIT ||
43 job_type(job) == JOB_TYPE_MIRROR ||
44 job_type(job) == JOB_TYPE_STREAM;
45 }
46
block_job_next_locked(BlockJob * bjob)47 BlockJob *block_job_next_locked(BlockJob *bjob)
48 {
49 Job *job = bjob ? &bjob->job : NULL;
50 GLOBAL_STATE_CODE();
51
52 do {
53 job = job_next_locked(job);
54 } while (job && !is_block_job(job));
55
56 return job ? container_of(job, BlockJob, job) : NULL;
57 }
58
block_job_get_locked(const char * id)59 BlockJob *block_job_get_locked(const char *id)
60 {
61 Job *job = job_get_locked(id);
62 GLOBAL_STATE_CODE();
63
64 if (job && is_block_job(job)) {
65 return container_of(job, BlockJob, job);
66 } else {
67 return NULL;
68 }
69 }
70
block_job_get(const char * id)71 BlockJob *block_job_get(const char *id)
72 {
73 JOB_LOCK_GUARD();
74 return block_job_get_locked(id);
75 }
76
block_job_free(Job * job)77 void block_job_free(Job *job)
78 {
79 BlockJob *bjob = container_of(job, BlockJob, job);
80 GLOBAL_STATE_CODE();
81
82 block_job_remove_all_bdrv(bjob);
83 ratelimit_destroy(&bjob->limit);
84 error_free(bjob->blocker);
85 }
86
child_job_get_parent_desc(BdrvChild * c)87 static char *child_job_get_parent_desc(BdrvChild *c)
88 {
89 BlockJob *job = c->opaque;
90 return g_strdup_printf("%s job '%s'", job_type_str(&job->job), job->job.id);
91 }
92
child_job_drained_begin(BdrvChild * c)93 static void child_job_drained_begin(BdrvChild *c)
94 {
95 BlockJob *job = c->opaque;
96 job_pause(&job->job);
97 }
98
child_job_drained_poll(BdrvChild * c)99 static bool child_job_drained_poll(BdrvChild *c)
100 {
101 BlockJob *bjob = c->opaque;
102 Job *job = &bjob->job;
103 const BlockJobDriver *drv = block_job_driver(bjob);
104
105 /* An inactive or completed job doesn't have any pending requests. Jobs
106 * with !job->busy are either already paused or have a pause point after
107 * being reentered, so no job driver code will run before they pause. */
108 WITH_JOB_LOCK_GUARD() {
109 if (!job->busy || job_is_completed_locked(job)) {
110 return false;
111 }
112 }
113
114 /* Otherwise, assume that it isn't fully stopped yet, but allow the job to
115 * override this assumption. */
116 if (drv->drained_poll) {
117 return drv->drained_poll(bjob);
118 } else {
119 return true;
120 }
121 }
122
child_job_drained_end(BdrvChild * c)123 static void child_job_drained_end(BdrvChild *c)
124 {
125 BlockJob *job = c->opaque;
126 job_resume(&job->job);
127 }
128
129 typedef struct BdrvStateChildJobContext {
130 AioContext *new_ctx;
131 BlockJob *job;
132 } BdrvStateChildJobContext;
133
child_job_set_aio_ctx_commit(void * opaque)134 static void child_job_set_aio_ctx_commit(void *opaque)
135 {
136 BdrvStateChildJobContext *s = opaque;
137 BlockJob *job = s->job;
138
139 job_set_aio_context(&job->job, s->new_ctx);
140 }
141
142 static TransactionActionDrv change_child_job_context = {
143 .commit = child_job_set_aio_ctx_commit,
144 .clean = g_free,
145 };
146
147 static bool GRAPH_RDLOCK
child_job_change_aio_ctx(BdrvChild * c,AioContext * ctx,GHashTable * visited,Transaction * tran,Error ** errp)148 child_job_change_aio_ctx(BdrvChild *c, AioContext *ctx, GHashTable *visited,
149 Transaction *tran, Error **errp)
150 {
151 BlockJob *job = c->opaque;
152 BdrvStateChildJobContext *s;
153 GSList *l;
154
155 for (l = job->nodes; l; l = l->next) {
156 BdrvChild *sibling = l->data;
157 if (!bdrv_child_change_aio_context(sibling, ctx, visited,
158 tran, errp)) {
159 return false;
160 }
161 }
162
163 s = g_new(BdrvStateChildJobContext, 1);
164 *s = (BdrvStateChildJobContext) {
165 .new_ctx = ctx,
166 .job = job,
167 };
168
169 tran_add(tran, &change_child_job_context, s);
170 return true;
171 }
172
child_job_get_parent_aio_context(BdrvChild * c)173 static AioContext *child_job_get_parent_aio_context(BdrvChild *c)
174 {
175 BlockJob *job = c->opaque;
176 IO_CODE();
177 JOB_LOCK_GUARD();
178
179 return job->job.aio_context;
180 }
181
182 static const BdrvChildClass child_job = {
183 .get_parent_desc = child_job_get_parent_desc,
184 .drained_begin = child_job_drained_begin,
185 .drained_poll = child_job_drained_poll,
186 .drained_end = child_job_drained_end,
187 .change_aio_ctx = child_job_change_aio_ctx,
188 .stay_at_node = true,
189 .get_parent_aio_context = child_job_get_parent_aio_context,
190 };
191
block_job_remove_all_bdrv(BlockJob * job)192 void block_job_remove_all_bdrv(BlockJob *job)
193 {
194 GLOBAL_STATE_CODE();
195 /*
196 * bdrv_root_unref_child() may reach child_job_[can_]set_aio_ctx(),
197 * which will also traverse job->nodes, so consume the list one by
198 * one to make sure that such a concurrent access does not attempt
199 * to process an already freed BdrvChild.
200 */
201 bdrv_drain_all_begin();
202 bdrv_graph_wrlock();
203 while (job->nodes) {
204 GSList *l = job->nodes;
205 BdrvChild *c = l->data;
206
207 job->nodes = l->next;
208
209 bdrv_op_unblock_all(c->bs, job->blocker);
210 bdrv_root_unref_child(c);
211
212 g_slist_free_1(l);
213 }
214 bdrv_graph_wrunlock();
215 bdrv_drain_all_end();
216 }
217
block_job_has_bdrv(BlockJob * job,BlockDriverState * bs)218 bool block_job_has_bdrv(BlockJob *job, BlockDriverState *bs)
219 {
220 GSList *el;
221 GLOBAL_STATE_CODE();
222
223 for (el = job->nodes; el; el = el->next) {
224 BdrvChild *c = el->data;
225 if (c->bs == bs) {
226 return true;
227 }
228 }
229
230 return false;
231 }
232
block_job_add_bdrv(BlockJob * job,const char * name,BlockDriverState * bs,uint64_t perm,uint64_t shared_perm,Error ** errp)233 int block_job_add_bdrv(BlockJob *job, const char *name, BlockDriverState *bs,
234 uint64_t perm, uint64_t shared_perm, Error **errp)
235 {
236 BdrvChild *c;
237 GLOBAL_STATE_CODE();
238
239 bdrv_ref(bs);
240
241 c = bdrv_root_attach_child(bs, name, &child_job, 0, perm, shared_perm, job,
242 errp);
243 if (c == NULL) {
244 return -EPERM;
245 }
246
247 job->nodes = g_slist_prepend(job->nodes, c);
248 bdrv_op_block_all(bs, job->blocker);
249
250 return 0;
251 }
252
253 /* Called with job_mutex lock held. */
block_job_on_idle_locked(Notifier * n,void * opaque)254 static void block_job_on_idle_locked(Notifier *n, void *opaque)
255 {
256 aio_wait_kick();
257 }
258
block_job_is_internal(BlockJob * job)259 bool block_job_is_internal(BlockJob *job)
260 {
261 return (job->job.id == NULL);
262 }
263
block_job_driver(BlockJob * job)264 const BlockJobDriver *block_job_driver(BlockJob *job)
265 {
266 return container_of(job->job.driver, BlockJobDriver, job_driver);
267 }
268
269 /* Assumes the job_mutex is held */
job_timer_pending(Job * job)270 static bool job_timer_pending(Job *job)
271 {
272 return timer_pending(&job->sleep_timer);
273 }
274
block_job_set_speed_locked(BlockJob * job,int64_t speed,Error ** errp)275 bool block_job_set_speed_locked(BlockJob *job, int64_t speed, Error **errp)
276 {
277 const BlockJobDriver *drv = block_job_driver(job);
278 int64_t old_speed = job->speed;
279
280 GLOBAL_STATE_CODE();
281
282 if (job_apply_verb_locked(&job->job, JOB_VERB_SET_SPEED, errp) < 0) {
283 return false;
284 }
285 if (speed < 0) {
286 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "speed",
287 "a non-negative value");
288 return false;
289 }
290
291 ratelimit_set_speed(&job->limit, speed, BLOCK_JOB_SLICE_TIME);
292
293 job->speed = speed;
294
295 if (drv->set_speed) {
296 job_unlock();
297 drv->set_speed(job, speed);
298 job_lock();
299 }
300
301 if (speed && speed <= old_speed) {
302 return true;
303 }
304
305 /* kick only if a timer is pending */
306 job_enter_cond_locked(&job->job, job_timer_pending);
307
308 return true;
309 }
310
block_job_set_speed(BlockJob * job,int64_t speed,Error ** errp)311 static bool block_job_set_speed(BlockJob *job, int64_t speed, Error **errp)
312 {
313 JOB_LOCK_GUARD();
314 return block_job_set_speed_locked(job, speed, errp);
315 }
316
block_job_change_locked(BlockJob * job,BlockJobChangeOptions * opts,Error ** errp)317 void block_job_change_locked(BlockJob *job, BlockJobChangeOptions *opts,
318 Error **errp)
319 {
320 const BlockJobDriver *drv = block_job_driver(job);
321
322 GLOBAL_STATE_CODE();
323
324 if (job_apply_verb_locked(&job->job, JOB_VERB_CHANGE, errp)) {
325 return;
326 }
327
328 if (drv->change) {
329 job_unlock();
330 drv->change(job, opts, errp);
331 job_lock();
332 } else {
333 error_setg(errp, "Job type does not support change");
334 }
335 }
336
block_job_ratelimit_processed_bytes(BlockJob * job,uint64_t n)337 void block_job_ratelimit_processed_bytes(BlockJob *job, uint64_t n)
338 {
339 IO_CODE();
340 ratelimit_calculate_delay(&job->limit, n);
341 }
342
block_job_ratelimit_sleep(BlockJob * job)343 void block_job_ratelimit_sleep(BlockJob *job)
344 {
345 uint64_t delay_ns;
346
347 /*
348 * Sleep at least once. If the job is reentered early, keep waiting until
349 * we've waited for the full time that is necessary to keep the job at the
350 * right speed.
351 *
352 * Make sure to recalculate the delay after each (possibly interrupted)
353 * sleep because the speed can change while the job has yielded.
354 */
355 do {
356 delay_ns = ratelimit_calculate_delay(&job->limit, 0);
357 job_sleep_ns(&job->job, delay_ns);
358 } while (delay_ns && !job_is_cancelled(&job->job));
359 }
360
block_job_query_locked(BlockJob * job,Error ** errp)361 BlockJobInfo *block_job_query_locked(BlockJob *job, Error **errp)
362 {
363 BlockJobInfo *info;
364 uint64_t progress_current, progress_total;
365 const BlockJobDriver *drv = block_job_driver(job);
366
367 GLOBAL_STATE_CODE();
368
369 if (block_job_is_internal(job)) {
370 error_setg(errp, "Cannot query QEMU internal jobs");
371 return NULL;
372 }
373
374 progress_get_snapshot(&job->job.progress, &progress_current,
375 &progress_total);
376
377 info = g_new0(BlockJobInfo, 1);
378 info->type = job_type(&job->job);
379 info->device = g_strdup(job->job.id);
380 info->busy = job->job.busy;
381 info->paused = job->job.pause_count > 0;
382 info->offset = progress_current;
383 info->len = progress_total;
384 info->speed = job->speed;
385 info->io_status = job->iostatus;
386 info->ready = job_is_ready_locked(&job->job),
387 info->status = job->job.status;
388 info->auto_finalize = job->job.auto_finalize;
389 info->auto_dismiss = job->job.auto_dismiss;
390 if (job->job.ret) {
391 info->error = job->job.err ?
392 g_strdup(error_get_pretty(job->job.err)) :
393 g_strdup(strerror(-job->job.ret));
394 }
395 if (drv->query) {
396 job_unlock();
397 drv->query(job, info);
398 job_lock();
399 }
400 return info;
401 }
402
403 /* Called with job lock held */
block_job_iostatus_set_err_locked(BlockJob * job,int error)404 static void block_job_iostatus_set_err_locked(BlockJob *job, int error)
405 {
406 if (job->iostatus == BLOCK_DEVICE_IO_STATUS_OK) {
407 job->iostatus = error == ENOSPC ? BLOCK_DEVICE_IO_STATUS_NOSPACE :
408 BLOCK_DEVICE_IO_STATUS_FAILED;
409 }
410 }
411
412 /* Called with job_mutex lock held. */
block_job_event_cancelled_locked(Notifier * n,void * opaque)413 static void block_job_event_cancelled_locked(Notifier *n, void *opaque)
414 {
415 BlockJob *job = opaque;
416 uint64_t progress_current, progress_total;
417
418 if (block_job_is_internal(job)) {
419 return;
420 }
421
422 progress_get_snapshot(&job->job.progress, &progress_current,
423 &progress_total);
424
425 qapi_event_send_block_job_cancelled(job_type(&job->job),
426 job->job.id,
427 progress_total,
428 progress_current,
429 job->speed);
430 }
431
432 /* Called with job_mutex lock held. */
block_job_event_completed_locked(Notifier * n,void * opaque)433 static void block_job_event_completed_locked(Notifier *n, void *opaque)
434 {
435 BlockJob *job = opaque;
436 const char *msg = NULL;
437 uint64_t progress_current, progress_total;
438
439 if (block_job_is_internal(job)) {
440 return;
441 }
442
443 if (job->job.ret < 0) {
444 msg = error_get_pretty(job->job.err);
445 }
446
447 progress_get_snapshot(&job->job.progress, &progress_current,
448 &progress_total);
449
450 qapi_event_send_block_job_completed(job_type(&job->job),
451 job->job.id,
452 progress_total,
453 progress_current,
454 job->speed,
455 msg);
456 }
457
458 /* Called with job_mutex lock held. */
block_job_event_pending_locked(Notifier * n,void * opaque)459 static void block_job_event_pending_locked(Notifier *n, void *opaque)
460 {
461 BlockJob *job = opaque;
462
463 if (block_job_is_internal(job)) {
464 return;
465 }
466
467 qapi_event_send_block_job_pending(job_type(&job->job),
468 job->job.id);
469 }
470
471 /* Called with job_mutex lock held. */
block_job_event_ready_locked(Notifier * n,void * opaque)472 static void block_job_event_ready_locked(Notifier *n, void *opaque)
473 {
474 BlockJob *job = opaque;
475 uint64_t progress_current, progress_total;
476
477 if (block_job_is_internal(job)) {
478 return;
479 }
480
481 progress_get_snapshot(&job->job.progress, &progress_current,
482 &progress_total);
483
484 qapi_event_send_block_job_ready(job_type(&job->job),
485 job->job.id,
486 progress_total,
487 progress_current,
488 job->speed);
489 }
490
491
block_job_create(const char * job_id,const BlockJobDriver * driver,JobTxn * txn,BlockDriverState * bs,uint64_t perm,uint64_t shared_perm,int64_t speed,int flags,BlockCompletionFunc * cb,void * opaque,Error ** errp)492 void *block_job_create(const char *job_id, const BlockJobDriver *driver,
493 JobTxn *txn, BlockDriverState *bs, uint64_t perm,
494 uint64_t shared_perm, int64_t speed, int flags,
495 BlockCompletionFunc *cb, void *opaque, Error **errp)
496 {
497 BlockJob *job;
498 int ret;
499 GLOBAL_STATE_CODE();
500
501 bdrv_drain_all_begin();
502 bdrv_graph_wrlock();
503
504 if (job_id == NULL && !(flags & JOB_INTERNAL)) {
505 job_id = bdrv_get_device_name(bs);
506 }
507
508 job = job_create(job_id, &driver->job_driver, txn, bdrv_get_aio_context(bs),
509 flags, cb, opaque, errp);
510 if (job == NULL) {
511 bdrv_graph_wrunlock();
512 bdrv_drain_all_end();
513 return NULL;
514 }
515
516 assert(is_block_job(&job->job));
517 assert(job->job.driver->free == &block_job_free);
518 assert(job->job.driver->user_resume == &block_job_user_resume);
519
520 ratelimit_init(&job->limit);
521
522 job->finalize_cancelled_notifier.notify = block_job_event_cancelled_locked;
523 job->finalize_completed_notifier.notify = block_job_event_completed_locked;
524 job->pending_notifier.notify = block_job_event_pending_locked;
525 job->ready_notifier.notify = block_job_event_ready_locked;
526 job->idle_notifier.notify = block_job_on_idle_locked;
527
528 WITH_JOB_LOCK_GUARD() {
529 notifier_list_add(&job->job.on_finalize_cancelled,
530 &job->finalize_cancelled_notifier);
531 notifier_list_add(&job->job.on_finalize_completed,
532 &job->finalize_completed_notifier);
533 notifier_list_add(&job->job.on_pending, &job->pending_notifier);
534 notifier_list_add(&job->job.on_ready, &job->ready_notifier);
535 notifier_list_add(&job->job.on_idle, &job->idle_notifier);
536 }
537
538 error_setg(&job->blocker, "block device is in use by block job: %s",
539 job_type_str(&job->job));
540
541 ret = block_job_add_bdrv(job, "main node", bs, perm, shared_perm, errp);
542 if (ret < 0) {
543 goto fail;
544 }
545
546 if (!block_job_set_speed(job, speed, errp)) {
547 goto fail;
548 }
549
550 bdrv_graph_wrunlock();
551 bdrv_drain_all_end();
552 return job;
553
554 fail:
555 bdrv_graph_wrunlock();
556 bdrv_drain_all_end();
557 job_early_fail(&job->job);
558 return NULL;
559 }
560
block_job_iostatus_reset_locked(BlockJob * job)561 void block_job_iostatus_reset_locked(BlockJob *job)
562 {
563 GLOBAL_STATE_CODE();
564 if (job->iostatus == BLOCK_DEVICE_IO_STATUS_OK) {
565 return;
566 }
567 assert(job->job.user_paused && job->job.pause_count > 0);
568 job->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
569 }
570
block_job_iostatus_reset(BlockJob * job)571 static void block_job_iostatus_reset(BlockJob *job)
572 {
573 JOB_LOCK_GUARD();
574 block_job_iostatus_reset_locked(job);
575 }
576
block_job_user_resume(Job * job)577 void block_job_user_resume(Job *job)
578 {
579 BlockJob *bjob = container_of(job, BlockJob, job);
580 GLOBAL_STATE_CODE();
581 block_job_iostatus_reset(bjob);
582 }
583
block_job_error_action(BlockJob * job,BlockdevOnError on_err,int is_read,int error)584 BlockErrorAction block_job_error_action(BlockJob *job, BlockdevOnError on_err,
585 int is_read, int error)
586 {
587 BlockErrorAction action;
588 IO_CODE();
589
590 switch (on_err) {
591 case BLOCKDEV_ON_ERROR_ENOSPC:
592 case BLOCKDEV_ON_ERROR_AUTO:
593 action = (error == ENOSPC) ?
594 BLOCK_ERROR_ACTION_STOP : BLOCK_ERROR_ACTION_REPORT;
595 break;
596 case BLOCKDEV_ON_ERROR_STOP:
597 action = BLOCK_ERROR_ACTION_STOP;
598 break;
599 case BLOCKDEV_ON_ERROR_REPORT:
600 action = BLOCK_ERROR_ACTION_REPORT;
601 break;
602 case BLOCKDEV_ON_ERROR_IGNORE:
603 action = BLOCK_ERROR_ACTION_IGNORE;
604 break;
605 default:
606 abort();
607 }
608 if (!block_job_is_internal(job)) {
609 qapi_event_send_block_job_error(job->job.id,
610 is_read ? IO_OPERATION_TYPE_READ :
611 IO_OPERATION_TYPE_WRITE,
612 action);
613 }
614 if (action == BLOCK_ERROR_ACTION_STOP) {
615 WITH_JOB_LOCK_GUARD() {
616 if (!job->job.user_paused) {
617 job_pause_locked(&job->job);
618 /*
619 * make the pause user visible, which will be
620 * resumed from QMP.
621 */
622 job->job.user_paused = true;
623 }
624 block_job_iostatus_set_err_locked(job, error);
625 }
626 }
627 return action;
628 }
629
block_job_get_aio_context(BlockJob * job)630 AioContext *block_job_get_aio_context(BlockJob *job)
631 {
632 GLOBAL_STATE_CODE();
633 return job->job.aio_context;
634 }
635