Merge tag 'for-upstream' of https://gitlab.com/bonzini/qemu into staging
[qemu.git] / blockjob.c
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/block.h"
28 #include "block/blockjob_int.h"
29 #include "block/block_int.h"
30 #include "block/trace.h"
31 #include "sysemu/block-backend.h"
32 #include "qapi/error.h"
33 #include "qapi/qapi-events-block-core.h"
34 #include "qapi/qmp/qerror.h"
35 #include "qemu/coroutine.h"
36 #include "qemu/main-loop.h"
37 #include "qemu/timer.h"
38
39 /*
40 * The block job API is composed of two categories of functions.
41 *
42 * The first includes functions used by the monitor. The monitor is
43 * peculiar in that it accesses the block job list with block_job_get, and
44 * therefore needs consistency across block_job_get and the actual operation
45 * (e.g. block_job_set_speed). The consistency is achieved with
46 * aio_context_acquire/release. These functions are declared in blockjob.h.
47 *
48 * The second includes functions used by the block job drivers and sometimes
49 * by the core block layer. These do not care about locking, because the
50 * whole coroutine runs under the AioContext lock, and are declared in
51 * blockjob_int.h.
52 */
53
54 static bool is_block_job(Job *job)
55 {
56 return job_type(job) == JOB_TYPE_BACKUP ||
57 job_type(job) == JOB_TYPE_COMMIT ||
58 job_type(job) == JOB_TYPE_MIRROR ||
59 job_type(job) == JOB_TYPE_STREAM;
60 }
61
62 BlockJob *block_job_next(BlockJob *bjob)
63 {
64 Job *job = bjob ? &bjob->job : NULL;
65 GLOBAL_STATE_CODE();
66
67 do {
68 job = job_next(job);
69 } while (job && !is_block_job(job));
70
71 return job ? container_of(job, BlockJob, job) : NULL;
72 }
73
74 BlockJob *block_job_get(const char *id)
75 {
76 Job *job = job_get(id);
77 GLOBAL_STATE_CODE();
78
79 if (job && is_block_job(job)) {
80 return container_of(job, BlockJob, job);
81 } else {
82 return NULL;
83 }
84 }
85
86 void block_job_free(Job *job)
87 {
88 BlockJob *bjob = container_of(job, BlockJob, job);
89 GLOBAL_STATE_CODE();
90
91 block_job_remove_all_bdrv(bjob);
92 ratelimit_destroy(&bjob->limit);
93 error_free(bjob->blocker);
94 }
95
96 static char *child_job_get_parent_desc(BdrvChild *c)
97 {
98 BlockJob *job = c->opaque;
99 return g_strdup_printf("%s job '%s'", job_type_str(&job->job), job->job.id);
100 }
101
102 static void child_job_drained_begin(BdrvChild *c)
103 {
104 BlockJob *job = c->opaque;
105 job_pause(&job->job);
106 }
107
108 static bool child_job_drained_poll(BdrvChild *c)
109 {
110 BlockJob *bjob = c->opaque;
111 Job *job = &bjob->job;
112 const BlockJobDriver *drv = block_job_driver(bjob);
113
114 /* An inactive or completed job doesn't have any pending requests. Jobs
115 * with !job->busy are either already paused or have a pause point after
116 * being reentered, so no job driver code will run before they pause. */
117 if (!job->busy || job_is_completed(job)) {
118 return false;
119 }
120
121 /* Otherwise, assume that it isn't fully stopped yet, but allow the job to
122 * override this assumption. */
123 if (drv->drained_poll) {
124 return drv->drained_poll(bjob);
125 } else {
126 return true;
127 }
128 }
129
130 static void child_job_drained_end(BdrvChild *c, int *drained_end_counter)
131 {
132 BlockJob *job = c->opaque;
133 job_resume(&job->job);
134 }
135
136 static bool child_job_can_set_aio_ctx(BdrvChild *c, AioContext *ctx,
137 GSList **ignore, Error **errp)
138 {
139 BlockJob *job = c->opaque;
140 GSList *l;
141
142 for (l = job->nodes; l; l = l->next) {
143 BdrvChild *sibling = l->data;
144 if (!bdrv_child_can_set_aio_context(sibling, ctx, ignore, errp)) {
145 return false;
146 }
147 }
148 return true;
149 }
150
151 static void child_job_set_aio_ctx(BdrvChild *c, AioContext *ctx,
152 GSList **ignore)
153 {
154 BlockJob *job = c->opaque;
155 GSList *l;
156
157 for (l = job->nodes; l; l = l->next) {
158 BdrvChild *sibling = l->data;
159 if (g_slist_find(*ignore, sibling)) {
160 continue;
161 }
162 *ignore = g_slist_prepend(*ignore, sibling);
163 bdrv_set_aio_context_ignore(sibling->bs, ctx, ignore);
164 }
165
166 job->job.aio_context = ctx;
167 }
168
169 static AioContext *child_job_get_parent_aio_context(BdrvChild *c)
170 {
171 BlockJob *job = c->opaque;
172
173 return job->job.aio_context;
174 }
175
176 static const BdrvChildClass child_job = {
177 .get_parent_desc = child_job_get_parent_desc,
178 .drained_begin = child_job_drained_begin,
179 .drained_poll = child_job_drained_poll,
180 .drained_end = child_job_drained_end,
181 .can_set_aio_ctx = child_job_can_set_aio_ctx,
182 .set_aio_ctx = child_job_set_aio_ctx,
183 .stay_at_node = true,
184 .get_parent_aio_context = child_job_get_parent_aio_context,
185 };
186
187 void block_job_remove_all_bdrv(BlockJob *job)
188 {
189 GLOBAL_STATE_CODE();
190 /*
191 * bdrv_root_unref_child() may reach child_job_[can_]set_aio_ctx(),
192 * which will also traverse job->nodes, so consume the list one by
193 * one to make sure that such a concurrent access does not attempt
194 * to process an already freed BdrvChild.
195 */
196 while (job->nodes) {
197 GSList *l = job->nodes;
198 BdrvChild *c = l->data;
199
200 job->nodes = l->next;
201
202 bdrv_op_unblock_all(c->bs, job->blocker);
203 bdrv_root_unref_child(c);
204
205 g_slist_free_1(l);
206 }
207 }
208
209 bool block_job_has_bdrv(BlockJob *job, BlockDriverState *bs)
210 {
211 GSList *el;
212 GLOBAL_STATE_CODE();
213
214 for (el = job->nodes; el; el = el->next) {
215 BdrvChild *c = el->data;
216 if (c->bs == bs) {
217 return true;
218 }
219 }
220
221 return false;
222 }
223
224 int block_job_add_bdrv(BlockJob *job, const char *name, BlockDriverState *bs,
225 uint64_t perm, uint64_t shared_perm, Error **errp)
226 {
227 BdrvChild *c;
228 bool need_context_ops;
229 GLOBAL_STATE_CODE();
230
231 bdrv_ref(bs);
232
233 need_context_ops = bdrv_get_aio_context(bs) != job->job.aio_context;
234
235 if (need_context_ops && job->job.aio_context != qemu_get_aio_context()) {
236 aio_context_release(job->job.aio_context);
237 }
238 c = bdrv_root_attach_child(bs, name, &child_job, 0, perm, shared_perm, job,
239 errp);
240 if (need_context_ops && job->job.aio_context != qemu_get_aio_context()) {
241 aio_context_acquire(job->job.aio_context);
242 }
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 static void block_job_on_idle(Notifier *n, void *opaque)
254 {
255 aio_wait_kick();
256 }
257
258 bool block_job_is_internal(BlockJob *job)
259 {
260 return (job->job.id == NULL);
261 }
262
263 const BlockJobDriver *block_job_driver(BlockJob *job)
264 {
265 return container_of(job->job.driver, BlockJobDriver, job_driver);
266 }
267
268 /* Assumes the job_mutex is held */
269 static bool job_timer_pending(Job *job)
270 {
271 return timer_pending(&job->sleep_timer);
272 }
273
274 bool block_job_set_speed(BlockJob *job, int64_t speed, Error **errp)
275 {
276 const BlockJobDriver *drv = block_job_driver(job);
277 int64_t old_speed = job->speed;
278
279 GLOBAL_STATE_CODE();
280
281 if (job_apply_verb(&job->job, JOB_VERB_SET_SPEED, errp) < 0) {
282 return false;
283 }
284 if (speed < 0) {
285 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "speed",
286 "a non-negative value");
287 return false;
288 }
289
290 ratelimit_set_speed(&job->limit, speed, BLOCK_JOB_SLICE_TIME);
291
292 job->speed = speed;
293
294 if (drv->set_speed) {
295 drv->set_speed(job, speed);
296 }
297
298 if (speed && speed <= old_speed) {
299 return true;
300 }
301
302 /* kick only if a timer is pending */
303 job_enter_cond(&job->job, job_timer_pending);
304
305 return true;
306 }
307
308 int64_t block_job_ratelimit_get_delay(BlockJob *job, uint64_t n)
309 {
310 IO_CODE();
311 return ratelimit_calculate_delay(&job->limit, n);
312 }
313
314 BlockJobInfo *block_job_query(BlockJob *job, Error **errp)
315 {
316 BlockJobInfo *info;
317 uint64_t progress_current, progress_total;
318
319 GLOBAL_STATE_CODE();
320
321 if (block_job_is_internal(job)) {
322 error_setg(errp, "Cannot query QEMU internal jobs");
323 return NULL;
324 }
325
326 progress_get_snapshot(&job->job.progress, &progress_current,
327 &progress_total);
328
329 info = g_new0(BlockJobInfo, 1);
330 info->type = g_strdup(job_type_str(&job->job));
331 info->device = g_strdup(job->job.id);
332 info->busy = qatomic_read(&job->job.busy);
333 info->paused = job->job.pause_count > 0;
334 info->offset = progress_current;
335 info->len = progress_total;
336 info->speed = job->speed;
337 info->io_status = job->iostatus;
338 info->ready = job_is_ready(&job->job),
339 info->status = job->job.status;
340 info->auto_finalize = job->job.auto_finalize;
341 info->auto_dismiss = job->job.auto_dismiss;
342 if (job->job.ret) {
343 info->has_error = true;
344 info->error = job->job.err ?
345 g_strdup(error_get_pretty(job->job.err)) :
346 g_strdup(strerror(-job->job.ret));
347 }
348 return info;
349 }
350
351 static void block_job_iostatus_set_err(BlockJob *job, int error)
352 {
353 if (job->iostatus == BLOCK_DEVICE_IO_STATUS_OK) {
354 job->iostatus = error == ENOSPC ? BLOCK_DEVICE_IO_STATUS_NOSPACE :
355 BLOCK_DEVICE_IO_STATUS_FAILED;
356 }
357 }
358
359 static void block_job_event_cancelled(Notifier *n, void *opaque)
360 {
361 BlockJob *job = opaque;
362 uint64_t progress_current, progress_total;
363
364 if (block_job_is_internal(job)) {
365 return;
366 }
367
368 progress_get_snapshot(&job->job.progress, &progress_current,
369 &progress_total);
370
371 qapi_event_send_block_job_cancelled(job_type(&job->job),
372 job->job.id,
373 progress_total,
374 progress_current,
375 job->speed);
376 }
377
378 static void block_job_event_completed(Notifier *n, void *opaque)
379 {
380 BlockJob *job = opaque;
381 const char *msg = NULL;
382 uint64_t progress_current, progress_total;
383
384 if (block_job_is_internal(job)) {
385 return;
386 }
387
388 if (job->job.ret < 0) {
389 msg = error_get_pretty(job->job.err);
390 }
391
392 progress_get_snapshot(&job->job.progress, &progress_current,
393 &progress_total);
394
395 qapi_event_send_block_job_completed(job_type(&job->job),
396 job->job.id,
397 progress_total,
398 progress_current,
399 job->speed,
400 !!msg,
401 msg);
402 }
403
404 static void block_job_event_pending(Notifier *n, void *opaque)
405 {
406 BlockJob *job = opaque;
407
408 if (block_job_is_internal(job)) {
409 return;
410 }
411
412 qapi_event_send_block_job_pending(job_type(&job->job),
413 job->job.id);
414 }
415
416 static void block_job_event_ready(Notifier *n, void *opaque)
417 {
418 BlockJob *job = opaque;
419 uint64_t progress_current, progress_total;
420
421 if (block_job_is_internal(job)) {
422 return;
423 }
424
425 progress_get_snapshot(&job->job.progress, &progress_current,
426 &progress_total);
427
428 qapi_event_send_block_job_ready(job_type(&job->job),
429 job->job.id,
430 progress_total,
431 progress_current,
432 job->speed);
433 }
434
435
436 /*
437 * API for block job drivers and the block layer. These functions are
438 * declared in blockjob_int.h.
439 */
440
441 void *block_job_create(const char *job_id, const BlockJobDriver *driver,
442 JobTxn *txn, BlockDriverState *bs, uint64_t perm,
443 uint64_t shared_perm, int64_t speed, int flags,
444 BlockCompletionFunc *cb, void *opaque, Error **errp)
445 {
446 BlockJob *job;
447 int ret;
448 GLOBAL_STATE_CODE();
449
450 if (job_id == NULL && !(flags & JOB_INTERNAL)) {
451 job_id = bdrv_get_device_name(bs);
452 }
453
454 job = job_create(job_id, &driver->job_driver, txn, bdrv_get_aio_context(bs),
455 flags, cb, opaque, errp);
456 if (job == NULL) {
457 return NULL;
458 }
459
460 assert(is_block_job(&job->job));
461 assert(job->job.driver->free == &block_job_free);
462 assert(job->job.driver->user_resume == &block_job_user_resume);
463
464 ratelimit_init(&job->limit);
465
466 job->finalize_cancelled_notifier.notify = block_job_event_cancelled;
467 job->finalize_completed_notifier.notify = block_job_event_completed;
468 job->pending_notifier.notify = block_job_event_pending;
469 job->ready_notifier.notify = block_job_event_ready;
470 job->idle_notifier.notify = block_job_on_idle;
471
472 notifier_list_add(&job->job.on_finalize_cancelled,
473 &job->finalize_cancelled_notifier);
474 notifier_list_add(&job->job.on_finalize_completed,
475 &job->finalize_completed_notifier);
476 notifier_list_add(&job->job.on_pending, &job->pending_notifier);
477 notifier_list_add(&job->job.on_ready, &job->ready_notifier);
478 notifier_list_add(&job->job.on_idle, &job->idle_notifier);
479
480 error_setg(&job->blocker, "block device is in use by block job: %s",
481 job_type_str(&job->job));
482
483 ret = block_job_add_bdrv(job, "main node", bs, perm, shared_perm, errp);
484 if (ret < 0) {
485 goto fail;
486 }
487
488 bdrv_op_unblock(bs, BLOCK_OP_TYPE_DATAPLANE, job->blocker);
489
490 if (!block_job_set_speed(job, speed, errp)) {
491 goto fail;
492 }
493
494 return job;
495
496 fail:
497 job_early_fail(&job->job);
498 return NULL;
499 }
500
501 void block_job_iostatus_reset(BlockJob *job)
502 {
503 GLOBAL_STATE_CODE();
504 if (job->iostatus == BLOCK_DEVICE_IO_STATUS_OK) {
505 return;
506 }
507 assert(job->job.user_paused && job->job.pause_count > 0);
508 job->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
509 }
510
511 void block_job_user_resume(Job *job)
512 {
513 BlockJob *bjob = container_of(job, BlockJob, job);
514 GLOBAL_STATE_CODE();
515 block_job_iostatus_reset(bjob);
516 }
517
518 BlockErrorAction block_job_error_action(BlockJob *job, BlockdevOnError on_err,
519 int is_read, int error)
520 {
521 BlockErrorAction action;
522 IO_CODE();
523
524 switch (on_err) {
525 case BLOCKDEV_ON_ERROR_ENOSPC:
526 case BLOCKDEV_ON_ERROR_AUTO:
527 action = (error == ENOSPC) ?
528 BLOCK_ERROR_ACTION_STOP : BLOCK_ERROR_ACTION_REPORT;
529 break;
530 case BLOCKDEV_ON_ERROR_STOP:
531 action = BLOCK_ERROR_ACTION_STOP;
532 break;
533 case BLOCKDEV_ON_ERROR_REPORT:
534 action = BLOCK_ERROR_ACTION_REPORT;
535 break;
536 case BLOCKDEV_ON_ERROR_IGNORE:
537 action = BLOCK_ERROR_ACTION_IGNORE;
538 break;
539 default:
540 abort();
541 }
542 if (!block_job_is_internal(job)) {
543 qapi_event_send_block_job_error(job->job.id,
544 is_read ? IO_OPERATION_TYPE_READ :
545 IO_OPERATION_TYPE_WRITE,
546 action);
547 }
548 if (action == BLOCK_ERROR_ACTION_STOP) {
549 if (!job->job.user_paused) {
550 job_pause(&job->job);
551 /* make the pause user visible, which will be resumed from QMP. */
552 job->job.user_paused = true;
553 }
554 block_job_iostatus_set_err(job, error);
555 }
556 return action;
557 }
558
559 AioContext *block_job_get_aio_context(BlockJob *job)
560 {
561 GLOBAL_STATE_CODE();
562 return job->job.aio_context;
563 }