xref: /openbmc/qemu/blockjob.c (revision 3d7b8974)
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 "sysemu/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 
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 
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 
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 
71 BlockJob *block_job_get(const char *id)
72 {
73     JOB_LOCK_GUARD();
74     return block_job_get_locked(id);
75 }
76 
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 
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 
93 static void child_job_drained_begin(BdrvChild *c)
94 {
95     BlockJob *job = c->opaque;
96     job_pause(&job->job);
97 }
98 
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 
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 
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 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 
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 
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     while (job->nodes) {
202         GSList *l = job->nodes;
203         BdrvChild *c = l->data;
204 
205         job->nodes = l->next;
206 
207         bdrv_op_unblock_all(c->bs, job->blocker);
208         bdrv_root_unref_child(c);
209 
210         g_slist_free_1(l);
211     }
212 }
213 
214 bool block_job_has_bdrv(BlockJob *job, BlockDriverState *bs)
215 {
216     GSList *el;
217     GLOBAL_STATE_CODE();
218 
219     for (el = job->nodes; el; el = el->next) {
220         BdrvChild *c = el->data;
221         if (c->bs == bs) {
222             return true;
223         }
224     }
225 
226     return false;
227 }
228 
229 int block_job_add_bdrv(BlockJob *job, const char *name, BlockDriverState *bs,
230                        uint64_t perm, uint64_t shared_perm, Error **errp)
231 {
232     BdrvChild *c;
233     AioContext *ctx = bdrv_get_aio_context(bs);
234     bool need_context_ops;
235     GLOBAL_STATE_CODE();
236 
237     bdrv_ref(bs);
238 
239     need_context_ops = ctx != job->job.aio_context;
240 
241     if (need_context_ops) {
242         if (job->job.aio_context != qemu_get_aio_context()) {
243             aio_context_release(job->job.aio_context);
244         }
245         aio_context_acquire(ctx);
246     }
247     c = bdrv_root_attach_child(bs, name, &child_job, 0, perm, shared_perm, job,
248                                errp);
249     if (need_context_ops) {
250         aio_context_release(ctx);
251         if (job->job.aio_context != qemu_get_aio_context()) {
252             aio_context_acquire(job->job.aio_context);
253         }
254     }
255     if (c == NULL) {
256         return -EPERM;
257     }
258 
259     job->nodes = g_slist_prepend(job->nodes, c);
260     bdrv_op_block_all(bs, job->blocker);
261 
262     return 0;
263 }
264 
265 /* Called with job_mutex lock held. */
266 static void block_job_on_idle_locked(Notifier *n, void *opaque)
267 {
268     aio_wait_kick();
269 }
270 
271 bool block_job_is_internal(BlockJob *job)
272 {
273     return (job->job.id == NULL);
274 }
275 
276 const BlockJobDriver *block_job_driver(BlockJob *job)
277 {
278     return container_of(job->job.driver, BlockJobDriver, job_driver);
279 }
280 
281 /* Assumes the job_mutex is held */
282 static bool job_timer_pending(Job *job)
283 {
284     return timer_pending(&job->sleep_timer);
285 }
286 
287 bool block_job_set_speed_locked(BlockJob *job, int64_t speed, Error **errp)
288 {
289     const BlockJobDriver *drv = block_job_driver(job);
290     int64_t old_speed = job->speed;
291 
292     GLOBAL_STATE_CODE();
293 
294     if (job_apply_verb_locked(&job->job, JOB_VERB_SET_SPEED, errp) < 0) {
295         return false;
296     }
297     if (speed < 0) {
298         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "speed",
299                    "a non-negative value");
300         return false;
301     }
302 
303     ratelimit_set_speed(&job->limit, speed, BLOCK_JOB_SLICE_TIME);
304 
305     job->speed = speed;
306 
307     if (drv->set_speed) {
308         job_unlock();
309         drv->set_speed(job, speed);
310         job_lock();
311     }
312 
313     if (speed && speed <= old_speed) {
314         return true;
315     }
316 
317     /* kick only if a timer is pending */
318     job_enter_cond_locked(&job->job, job_timer_pending);
319 
320     return true;
321 }
322 
323 static bool block_job_set_speed(BlockJob *job, int64_t speed, Error **errp)
324 {
325     JOB_LOCK_GUARD();
326     return block_job_set_speed_locked(job, speed, errp);
327 }
328 
329 void block_job_ratelimit_processed_bytes(BlockJob *job, uint64_t n)
330 {
331     IO_CODE();
332     ratelimit_calculate_delay(&job->limit, n);
333 }
334 
335 void block_job_ratelimit_sleep(BlockJob *job)
336 {
337     uint64_t delay_ns;
338 
339     /*
340      * Sleep at least once. If the job is reentered early, keep waiting until
341      * we've waited for the full time that is necessary to keep the job at the
342      * right speed.
343      *
344      * Make sure to recalculate the delay after each (possibly interrupted)
345      * sleep because the speed can change while the job has yielded.
346      */
347     do {
348         delay_ns = ratelimit_calculate_delay(&job->limit, 0);
349         job_sleep_ns(&job->job, delay_ns);
350     } while (delay_ns && !job_is_cancelled(&job->job));
351 }
352 
353 BlockJobInfo *block_job_query_locked(BlockJob *job, Error **errp)
354 {
355     BlockJobInfo *info;
356     uint64_t progress_current, progress_total;
357 
358     GLOBAL_STATE_CODE();
359 
360     if (block_job_is_internal(job)) {
361         error_setg(errp, "Cannot query QEMU internal jobs");
362         return NULL;
363     }
364 
365     progress_get_snapshot(&job->job.progress, &progress_current,
366                           &progress_total);
367 
368     info = g_new0(BlockJobInfo, 1);
369     info->type      = g_strdup(job_type_str(&job->job));
370     info->device    = g_strdup(job->job.id);
371     info->busy      = job->job.busy;
372     info->paused    = job->job.pause_count > 0;
373     info->offset    = progress_current;
374     info->len       = progress_total;
375     info->speed     = job->speed;
376     info->io_status = job->iostatus;
377     info->ready     = job_is_ready_locked(&job->job),
378     info->status    = job->job.status;
379     info->auto_finalize = job->job.auto_finalize;
380     info->auto_dismiss  = job->job.auto_dismiss;
381     if (job->job.ret) {
382         info->error = job->job.err ?
383                         g_strdup(error_get_pretty(job->job.err)) :
384                         g_strdup(strerror(-job->job.ret));
385     }
386     return info;
387 }
388 
389 /* Called with job lock held */
390 static void block_job_iostatus_set_err_locked(BlockJob *job, int error)
391 {
392     if (job->iostatus == BLOCK_DEVICE_IO_STATUS_OK) {
393         job->iostatus = error == ENOSPC ? BLOCK_DEVICE_IO_STATUS_NOSPACE :
394                                           BLOCK_DEVICE_IO_STATUS_FAILED;
395     }
396 }
397 
398 /* Called with job_mutex lock held. */
399 static void block_job_event_cancelled_locked(Notifier *n, void *opaque)
400 {
401     BlockJob *job = opaque;
402     uint64_t progress_current, progress_total;
403 
404     if (block_job_is_internal(job)) {
405         return;
406     }
407 
408     progress_get_snapshot(&job->job.progress, &progress_current,
409                           &progress_total);
410 
411     qapi_event_send_block_job_cancelled(job_type(&job->job),
412                                         job->job.id,
413                                         progress_total,
414                                         progress_current,
415                                         job->speed);
416 }
417 
418 /* Called with job_mutex lock held. */
419 static void block_job_event_completed_locked(Notifier *n, void *opaque)
420 {
421     BlockJob *job = opaque;
422     const char *msg = NULL;
423     uint64_t progress_current, progress_total;
424 
425     if (block_job_is_internal(job)) {
426         return;
427     }
428 
429     if (job->job.ret < 0) {
430         msg = error_get_pretty(job->job.err);
431     }
432 
433     progress_get_snapshot(&job->job.progress, &progress_current,
434                           &progress_total);
435 
436     qapi_event_send_block_job_completed(job_type(&job->job),
437                                         job->job.id,
438                                         progress_total,
439                                         progress_current,
440                                         job->speed,
441                                         msg);
442 }
443 
444 /* Called with job_mutex lock held. */
445 static void block_job_event_pending_locked(Notifier *n, void *opaque)
446 {
447     BlockJob *job = opaque;
448 
449     if (block_job_is_internal(job)) {
450         return;
451     }
452 
453     qapi_event_send_block_job_pending(job_type(&job->job),
454                                       job->job.id);
455 }
456 
457 /* Called with job_mutex lock held. */
458 static void block_job_event_ready_locked(Notifier *n, void *opaque)
459 {
460     BlockJob *job = opaque;
461     uint64_t progress_current, progress_total;
462 
463     if (block_job_is_internal(job)) {
464         return;
465     }
466 
467     progress_get_snapshot(&job->job.progress, &progress_current,
468                           &progress_total);
469 
470     qapi_event_send_block_job_ready(job_type(&job->job),
471                                     job->job.id,
472                                     progress_total,
473                                     progress_current,
474                                     job->speed);
475 }
476 
477 
478 void *block_job_create(const char *job_id, const BlockJobDriver *driver,
479                        JobTxn *txn, BlockDriverState *bs, uint64_t perm,
480                        uint64_t shared_perm, int64_t speed, int flags,
481                        BlockCompletionFunc *cb, void *opaque, Error **errp)
482 {
483     BlockJob *job;
484     int ret;
485     GLOBAL_STATE_CODE();
486 
487     if (job_id == NULL && !(flags & JOB_INTERNAL)) {
488         job_id = bdrv_get_device_name(bs);
489     }
490 
491     job = job_create(job_id, &driver->job_driver, txn, bdrv_get_aio_context(bs),
492                      flags, cb, opaque, errp);
493     if (job == NULL) {
494         return NULL;
495     }
496 
497     assert(is_block_job(&job->job));
498     assert(job->job.driver->free == &block_job_free);
499     assert(job->job.driver->user_resume == &block_job_user_resume);
500 
501     ratelimit_init(&job->limit);
502 
503     job->finalize_cancelled_notifier.notify = block_job_event_cancelled_locked;
504     job->finalize_completed_notifier.notify = block_job_event_completed_locked;
505     job->pending_notifier.notify = block_job_event_pending_locked;
506     job->ready_notifier.notify = block_job_event_ready_locked;
507     job->idle_notifier.notify = block_job_on_idle_locked;
508 
509     WITH_JOB_LOCK_GUARD() {
510         notifier_list_add(&job->job.on_finalize_cancelled,
511                           &job->finalize_cancelled_notifier);
512         notifier_list_add(&job->job.on_finalize_completed,
513                           &job->finalize_completed_notifier);
514         notifier_list_add(&job->job.on_pending, &job->pending_notifier);
515         notifier_list_add(&job->job.on_ready, &job->ready_notifier);
516         notifier_list_add(&job->job.on_idle, &job->idle_notifier);
517     }
518 
519     error_setg(&job->blocker, "block device is in use by block job: %s",
520                job_type_str(&job->job));
521 
522     ret = block_job_add_bdrv(job, "main node", bs, perm, shared_perm, errp);
523     if (ret < 0) {
524         goto fail;
525     }
526 
527     bdrv_op_unblock(bs, BLOCK_OP_TYPE_DATAPLANE, job->blocker);
528 
529     if (!block_job_set_speed(job, speed, errp)) {
530         goto fail;
531     }
532 
533     return job;
534 
535 fail:
536     job_early_fail(&job->job);
537     return NULL;
538 }
539 
540 void block_job_iostatus_reset_locked(BlockJob *job)
541 {
542     GLOBAL_STATE_CODE();
543     if (job->iostatus == BLOCK_DEVICE_IO_STATUS_OK) {
544         return;
545     }
546     assert(job->job.user_paused && job->job.pause_count > 0);
547     job->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
548 }
549 
550 static void block_job_iostatus_reset(BlockJob *job)
551 {
552     JOB_LOCK_GUARD();
553     block_job_iostatus_reset_locked(job);
554 }
555 
556 void block_job_user_resume(Job *job)
557 {
558     BlockJob *bjob = container_of(job, BlockJob, job);
559     GLOBAL_STATE_CODE();
560     block_job_iostatus_reset(bjob);
561 }
562 
563 BlockErrorAction block_job_error_action(BlockJob *job, BlockdevOnError on_err,
564                                         int is_read, int error)
565 {
566     BlockErrorAction action;
567     IO_CODE();
568 
569     switch (on_err) {
570     case BLOCKDEV_ON_ERROR_ENOSPC:
571     case BLOCKDEV_ON_ERROR_AUTO:
572         action = (error == ENOSPC) ?
573                  BLOCK_ERROR_ACTION_STOP : BLOCK_ERROR_ACTION_REPORT;
574         break;
575     case BLOCKDEV_ON_ERROR_STOP:
576         action = BLOCK_ERROR_ACTION_STOP;
577         break;
578     case BLOCKDEV_ON_ERROR_REPORT:
579         action = BLOCK_ERROR_ACTION_REPORT;
580         break;
581     case BLOCKDEV_ON_ERROR_IGNORE:
582         action = BLOCK_ERROR_ACTION_IGNORE;
583         break;
584     default:
585         abort();
586     }
587     if (!block_job_is_internal(job)) {
588         qapi_event_send_block_job_error(job->job.id,
589                                         is_read ? IO_OPERATION_TYPE_READ :
590                                         IO_OPERATION_TYPE_WRITE,
591                                         action);
592     }
593     if (action == BLOCK_ERROR_ACTION_STOP) {
594         WITH_JOB_LOCK_GUARD() {
595             if (!job->job.user_paused) {
596                 job_pause_locked(&job->job);
597                 /*
598                  * make the pause user visible, which will be
599                  * resumed from QMP.
600                  */
601                 job->job.user_paused = true;
602             }
603             block_job_iostatus_set_err_locked(job, error);
604         }
605     }
606     return action;
607 }
608 
609 AioContext *block_job_get_aio_context(BlockJob *job)
610 {
611     GLOBAL_STATE_CODE();
612     return job->job.aio_context;
613 }
614