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
child_job_change_aio_ctx(BdrvChild * c,AioContext * ctx,GHashTable * visited,Transaction * tran,Error ** errp)147 static bool child_job_change_aio_ctx(BdrvChild *c, AioContext *ctx,
148 GHashTable *visited, Transaction *tran,
149 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_graph_wrlock();
202 while (job->nodes) {
203 GSList *l = job->nodes;
204 BdrvChild *c = l->data;
205
206 job->nodes = l->next;
207
208 bdrv_op_unblock_all(c->bs, job->blocker);
209 bdrv_root_unref_child(c);
210
211 g_slist_free_1(l);
212 }
213 bdrv_graph_wrunlock();
214 }
215
block_job_has_bdrv(BlockJob * job,BlockDriverState * bs)216 bool block_job_has_bdrv(BlockJob *job, BlockDriverState *bs)
217 {
218 GSList *el;
219 GLOBAL_STATE_CODE();
220
221 for (el = job->nodes; el; el = el->next) {
222 BdrvChild *c = el->data;
223 if (c->bs == bs) {
224 return true;
225 }
226 }
227
228 return false;
229 }
230
block_job_add_bdrv(BlockJob * job,const char * name,BlockDriverState * bs,uint64_t perm,uint64_t shared_perm,Error ** errp)231 int block_job_add_bdrv(BlockJob *job, const char *name, BlockDriverState *bs,
232 uint64_t perm, uint64_t shared_perm, Error **errp)
233 {
234 BdrvChild *c;
235 GLOBAL_STATE_CODE();
236
237 bdrv_ref(bs);
238
239 c = bdrv_root_attach_child(bs, name, &child_job, 0, perm, shared_perm, job,
240 errp);
241 if (c == NULL) {
242 return -EPERM;
243 }
244
245 job->nodes = g_slist_prepend(job->nodes, c);
246 bdrv_op_block_all(bs, job->blocker);
247
248 return 0;
249 }
250
251 /* Called with job_mutex lock held. */
block_job_on_idle_locked(Notifier * n,void * opaque)252 static void block_job_on_idle_locked(Notifier *n, void *opaque)
253 {
254 aio_wait_kick();
255 }
256
block_job_is_internal(BlockJob * job)257 bool block_job_is_internal(BlockJob *job)
258 {
259 return (job->job.id == NULL);
260 }
261
block_job_driver(BlockJob * job)262 const BlockJobDriver *block_job_driver(BlockJob *job)
263 {
264 return container_of(job->job.driver, BlockJobDriver, job_driver);
265 }
266
267 /* Assumes the job_mutex is held */
job_timer_pending(Job * job)268 static bool job_timer_pending(Job *job)
269 {
270 return timer_pending(&job->sleep_timer);
271 }
272
block_job_set_speed_locked(BlockJob * job,int64_t speed,Error ** errp)273 bool block_job_set_speed_locked(BlockJob *job, int64_t speed, Error **errp)
274 {
275 const BlockJobDriver *drv = block_job_driver(job);
276 int64_t old_speed = job->speed;
277
278 GLOBAL_STATE_CODE();
279
280 if (job_apply_verb_locked(&job->job, JOB_VERB_SET_SPEED, errp) < 0) {
281 return false;
282 }
283 if (speed < 0) {
284 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "speed",
285 "a non-negative value");
286 return false;
287 }
288
289 ratelimit_set_speed(&job->limit, speed, BLOCK_JOB_SLICE_TIME);
290
291 job->speed = speed;
292
293 if (drv->set_speed) {
294 job_unlock();
295 drv->set_speed(job, speed);
296 job_lock();
297 }
298
299 if (speed && speed <= old_speed) {
300 return true;
301 }
302
303 /* kick only if a timer is pending */
304 job_enter_cond_locked(&job->job, job_timer_pending);
305
306 return true;
307 }
308
block_job_set_speed(BlockJob * job,int64_t speed,Error ** errp)309 static bool block_job_set_speed(BlockJob *job, int64_t speed, Error **errp)
310 {
311 JOB_LOCK_GUARD();
312 return block_job_set_speed_locked(job, speed, errp);
313 }
314
block_job_change_locked(BlockJob * job,BlockJobChangeOptions * opts,Error ** errp)315 void block_job_change_locked(BlockJob *job, BlockJobChangeOptions *opts,
316 Error **errp)
317 {
318 const BlockJobDriver *drv = block_job_driver(job);
319
320 GLOBAL_STATE_CODE();
321
322 if (job_apply_verb_locked(&job->job, JOB_VERB_CHANGE, errp)) {
323 return;
324 }
325
326 if (drv->change) {
327 job_unlock();
328 drv->change(job, opts, errp);
329 job_lock();
330 } else {
331 error_setg(errp, "Job type does not support change");
332 }
333 }
334
block_job_ratelimit_processed_bytes(BlockJob * job,uint64_t n)335 void block_job_ratelimit_processed_bytes(BlockJob *job, uint64_t n)
336 {
337 IO_CODE();
338 ratelimit_calculate_delay(&job->limit, n);
339 }
340
block_job_ratelimit_sleep(BlockJob * job)341 void block_job_ratelimit_sleep(BlockJob *job)
342 {
343 uint64_t delay_ns;
344
345 /*
346 * Sleep at least once. If the job is reentered early, keep waiting until
347 * we've waited for the full time that is necessary to keep the job at the
348 * right speed.
349 *
350 * Make sure to recalculate the delay after each (possibly interrupted)
351 * sleep because the speed can change while the job has yielded.
352 */
353 do {
354 delay_ns = ratelimit_calculate_delay(&job->limit, 0);
355 job_sleep_ns(&job->job, delay_ns);
356 } while (delay_ns && !job_is_cancelled(&job->job));
357 }
358
block_job_query_locked(BlockJob * job,Error ** errp)359 BlockJobInfo *block_job_query_locked(BlockJob *job, Error **errp)
360 {
361 BlockJobInfo *info;
362 uint64_t progress_current, progress_total;
363 const BlockJobDriver *drv = block_job_driver(job);
364
365 GLOBAL_STATE_CODE();
366
367 if (block_job_is_internal(job)) {
368 error_setg(errp, "Cannot query QEMU internal jobs");
369 return NULL;
370 }
371
372 progress_get_snapshot(&job->job.progress, &progress_current,
373 &progress_total);
374
375 info = g_new0(BlockJobInfo, 1);
376 info->type = job_type(&job->job);
377 info->device = g_strdup(job->job.id);
378 info->busy = job->job.busy;
379 info->paused = job->job.pause_count > 0;
380 info->offset = progress_current;
381 info->len = progress_total;
382 info->speed = job->speed;
383 info->io_status = job->iostatus;
384 info->ready = job_is_ready_locked(&job->job),
385 info->status = job->job.status;
386 info->auto_finalize = job->job.auto_finalize;
387 info->auto_dismiss = job->job.auto_dismiss;
388 if (job->job.ret) {
389 info->error = job->job.err ?
390 g_strdup(error_get_pretty(job->job.err)) :
391 g_strdup(strerror(-job->job.ret));
392 }
393 if (drv->query) {
394 job_unlock();
395 drv->query(job, info);
396 job_lock();
397 }
398 return info;
399 }
400
401 /* Called with job lock held */
block_job_iostatus_set_err_locked(BlockJob * job,int error)402 static void block_job_iostatus_set_err_locked(BlockJob *job, int error)
403 {
404 if (job->iostatus == BLOCK_DEVICE_IO_STATUS_OK) {
405 job->iostatus = error == ENOSPC ? BLOCK_DEVICE_IO_STATUS_NOSPACE :
406 BLOCK_DEVICE_IO_STATUS_FAILED;
407 }
408 }
409
410 /* Called with job_mutex lock held. */
block_job_event_cancelled_locked(Notifier * n,void * opaque)411 static void block_job_event_cancelled_locked(Notifier *n, void *opaque)
412 {
413 BlockJob *job = opaque;
414 uint64_t progress_current, progress_total;
415
416 if (block_job_is_internal(job)) {
417 return;
418 }
419
420 progress_get_snapshot(&job->job.progress, &progress_current,
421 &progress_total);
422
423 qapi_event_send_block_job_cancelled(job_type(&job->job),
424 job->job.id,
425 progress_total,
426 progress_current,
427 job->speed);
428 }
429
430 /* Called with job_mutex lock held. */
block_job_event_completed_locked(Notifier * n,void * opaque)431 static void block_job_event_completed_locked(Notifier *n, void *opaque)
432 {
433 BlockJob *job = opaque;
434 const char *msg = NULL;
435 uint64_t progress_current, progress_total;
436
437 if (block_job_is_internal(job)) {
438 return;
439 }
440
441 if (job->job.ret < 0) {
442 msg = error_get_pretty(job->job.err);
443 }
444
445 progress_get_snapshot(&job->job.progress, &progress_current,
446 &progress_total);
447
448 qapi_event_send_block_job_completed(job_type(&job->job),
449 job->job.id,
450 progress_total,
451 progress_current,
452 job->speed,
453 msg);
454 }
455
456 /* Called with job_mutex lock held. */
block_job_event_pending_locked(Notifier * n,void * opaque)457 static void block_job_event_pending_locked(Notifier *n, void *opaque)
458 {
459 BlockJob *job = opaque;
460
461 if (block_job_is_internal(job)) {
462 return;
463 }
464
465 qapi_event_send_block_job_pending(job_type(&job->job),
466 job->job.id);
467 }
468
469 /* Called with job_mutex lock held. */
block_job_event_ready_locked(Notifier * n,void * opaque)470 static void block_job_event_ready_locked(Notifier *n, void *opaque)
471 {
472 BlockJob *job = opaque;
473 uint64_t progress_current, progress_total;
474
475 if (block_job_is_internal(job)) {
476 return;
477 }
478
479 progress_get_snapshot(&job->job.progress, &progress_current,
480 &progress_total);
481
482 qapi_event_send_block_job_ready(job_type(&job->job),
483 job->job.id,
484 progress_total,
485 progress_current,
486 job->speed);
487 }
488
489
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)490 void *block_job_create(const char *job_id, const BlockJobDriver *driver,
491 JobTxn *txn, BlockDriverState *bs, uint64_t perm,
492 uint64_t shared_perm, int64_t speed, int flags,
493 BlockCompletionFunc *cb, void *opaque, Error **errp)
494 {
495 BlockJob *job;
496 int ret;
497 GLOBAL_STATE_CODE();
498
499 bdrv_graph_wrlock();
500
501 if (job_id == NULL && !(flags & JOB_INTERNAL)) {
502 job_id = bdrv_get_device_name(bs);
503 }
504
505 job = job_create(job_id, &driver->job_driver, txn, bdrv_get_aio_context(bs),
506 flags, cb, opaque, errp);
507 if (job == NULL) {
508 bdrv_graph_wrunlock();
509 return NULL;
510 }
511
512 assert(is_block_job(&job->job));
513 assert(job->job.driver->free == &block_job_free);
514 assert(job->job.driver->user_resume == &block_job_user_resume);
515
516 ratelimit_init(&job->limit);
517
518 job->finalize_cancelled_notifier.notify = block_job_event_cancelled_locked;
519 job->finalize_completed_notifier.notify = block_job_event_completed_locked;
520 job->pending_notifier.notify = block_job_event_pending_locked;
521 job->ready_notifier.notify = block_job_event_ready_locked;
522 job->idle_notifier.notify = block_job_on_idle_locked;
523
524 WITH_JOB_LOCK_GUARD() {
525 notifier_list_add(&job->job.on_finalize_cancelled,
526 &job->finalize_cancelled_notifier);
527 notifier_list_add(&job->job.on_finalize_completed,
528 &job->finalize_completed_notifier);
529 notifier_list_add(&job->job.on_pending, &job->pending_notifier);
530 notifier_list_add(&job->job.on_ready, &job->ready_notifier);
531 notifier_list_add(&job->job.on_idle, &job->idle_notifier);
532 }
533
534 error_setg(&job->blocker, "block device is in use by block job: %s",
535 job_type_str(&job->job));
536
537 ret = block_job_add_bdrv(job, "main node", bs, perm, shared_perm, errp);
538 if (ret < 0) {
539 goto fail;
540 }
541
542 if (!block_job_set_speed(job, speed, errp)) {
543 goto fail;
544 }
545
546 bdrv_graph_wrunlock();
547 return job;
548
549 fail:
550 bdrv_graph_wrunlock();
551 job_early_fail(&job->job);
552 return NULL;
553 }
554
block_job_iostatus_reset_locked(BlockJob * job)555 void block_job_iostatus_reset_locked(BlockJob *job)
556 {
557 GLOBAL_STATE_CODE();
558 if (job->iostatus == BLOCK_DEVICE_IO_STATUS_OK) {
559 return;
560 }
561 assert(job->job.user_paused && job->job.pause_count > 0);
562 job->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
563 }
564
block_job_iostatus_reset(BlockJob * job)565 static void block_job_iostatus_reset(BlockJob *job)
566 {
567 JOB_LOCK_GUARD();
568 block_job_iostatus_reset_locked(job);
569 }
570
block_job_user_resume(Job * job)571 void block_job_user_resume(Job *job)
572 {
573 BlockJob *bjob = container_of(job, BlockJob, job);
574 GLOBAL_STATE_CODE();
575 block_job_iostatus_reset(bjob);
576 }
577
block_job_error_action(BlockJob * job,BlockdevOnError on_err,int is_read,int error)578 BlockErrorAction block_job_error_action(BlockJob *job, BlockdevOnError on_err,
579 int is_read, int error)
580 {
581 BlockErrorAction action;
582 IO_CODE();
583
584 switch (on_err) {
585 case BLOCKDEV_ON_ERROR_ENOSPC:
586 case BLOCKDEV_ON_ERROR_AUTO:
587 action = (error == ENOSPC) ?
588 BLOCK_ERROR_ACTION_STOP : BLOCK_ERROR_ACTION_REPORT;
589 break;
590 case BLOCKDEV_ON_ERROR_STOP:
591 action = BLOCK_ERROR_ACTION_STOP;
592 break;
593 case BLOCKDEV_ON_ERROR_REPORT:
594 action = BLOCK_ERROR_ACTION_REPORT;
595 break;
596 case BLOCKDEV_ON_ERROR_IGNORE:
597 action = BLOCK_ERROR_ACTION_IGNORE;
598 break;
599 default:
600 abort();
601 }
602 if (!block_job_is_internal(job)) {
603 qapi_event_send_block_job_error(job->job.id,
604 is_read ? IO_OPERATION_TYPE_READ :
605 IO_OPERATION_TYPE_WRITE,
606 action);
607 }
608 if (action == BLOCK_ERROR_ACTION_STOP) {
609 WITH_JOB_LOCK_GUARD() {
610 if (!job->job.user_paused) {
611 job_pause_locked(&job->job);
612 /*
613 * make the pause user visible, which will be
614 * resumed from QMP.
615 */
616 job->job.user_paused = true;
617 }
618 block_job_iostatus_set_err_locked(job, error);
619 }
620 }
621 return action;
622 }
623
block_job_get_aio_context(BlockJob * job)624 AioContext *block_job_get_aio_context(BlockJob *job)
625 {
626 GLOBAL_STATE_CODE();
627 return job->job.aio_context;
628 }
629