Merge remote-tracking branch 'remotes/dgilbert-gitlab/tags/pull-virtiofs-20210304...
[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
66 do {
67 job = job_next(job);
68 } while (job && !is_block_job(job));
69
70 return job ? container_of(job, BlockJob, job) : NULL;
71 }
72
73 BlockJob *block_job_get(const char *id)
74 {
75 Job *job = job_get(id);
76
77 if (job && is_block_job(job)) {
78 return container_of(job, BlockJob, job);
79 } else {
80 return NULL;
81 }
82 }
83
84 void block_job_free(Job *job)
85 {
86 BlockJob *bjob = container_of(job, BlockJob, job);
87
88 block_job_remove_all_bdrv(bjob);
89 blk_unref(bjob->blk);
90 error_free(bjob->blocker);
91 }
92
93 static char *child_job_get_parent_desc(BdrvChild *c)
94 {
95 BlockJob *job = c->opaque;
96 return g_strdup_printf("%s job '%s'", job_type_str(&job->job), job->job.id);
97 }
98
99 static void child_job_drained_begin(BdrvChild *c)
100 {
101 BlockJob *job = c->opaque;
102 job_pause(&job->job);
103 }
104
105 static bool child_job_drained_poll(BdrvChild *c)
106 {
107 BlockJob *bjob = c->opaque;
108 Job *job = &bjob->job;
109 const BlockJobDriver *drv = block_job_driver(bjob);
110
111 /* An inactive or completed job doesn't have any pending requests. Jobs
112 * with !job->busy are either already paused or have a pause point after
113 * being reentered, so no job driver code will run before they pause. */
114 if (!job->busy || job_is_completed(job)) {
115 return false;
116 }
117
118 /* Otherwise, assume that it isn't fully stopped yet, but allow the job to
119 * override this assumption. */
120 if (drv->drained_poll) {
121 return drv->drained_poll(bjob);
122 } else {
123 return true;
124 }
125 }
126
127 static void child_job_drained_end(BdrvChild *c, int *drained_end_counter)
128 {
129 BlockJob *job = c->opaque;
130 job_resume(&job->job);
131 }
132
133 static bool child_job_can_set_aio_ctx(BdrvChild *c, AioContext *ctx,
134 GSList **ignore, Error **errp)
135 {
136 BlockJob *job = c->opaque;
137 GSList *l;
138
139 for (l = job->nodes; l; l = l->next) {
140 BdrvChild *sibling = l->data;
141 if (!bdrv_child_can_set_aio_context(sibling, ctx, ignore, errp)) {
142 return false;
143 }
144 }
145 return true;
146 }
147
148 static void child_job_set_aio_ctx(BdrvChild *c, AioContext *ctx,
149 GSList **ignore)
150 {
151 BlockJob *job = c->opaque;
152 GSList *l;
153
154 for (l = job->nodes; l; l = l->next) {
155 BdrvChild *sibling = l->data;
156 if (g_slist_find(*ignore, sibling)) {
157 continue;
158 }
159 *ignore = g_slist_prepend(*ignore, sibling);
160 bdrv_set_aio_context_ignore(sibling->bs, ctx, ignore);
161 }
162
163 job->job.aio_context = ctx;
164 }
165
166 static const BdrvChildClass child_job = {
167 .get_parent_desc = child_job_get_parent_desc,
168 .drained_begin = child_job_drained_begin,
169 .drained_poll = child_job_drained_poll,
170 .drained_end = child_job_drained_end,
171 .can_set_aio_ctx = child_job_can_set_aio_ctx,
172 .set_aio_ctx = child_job_set_aio_ctx,
173 .stay_at_node = true,
174 };
175
176 void block_job_remove_all_bdrv(BlockJob *job)
177 {
178 /*
179 * bdrv_root_unref_child() may reach child_job_[can_]set_aio_ctx(),
180 * which will also traverse job->nodes, so consume the list one by
181 * one to make sure that such a concurrent access does not attempt
182 * to process an already freed BdrvChild.
183 */
184 while (job->nodes) {
185 GSList *l = job->nodes;
186 BdrvChild *c = l->data;
187
188 job->nodes = l->next;
189
190 bdrv_op_unblock_all(c->bs, job->blocker);
191 bdrv_root_unref_child(c);
192
193 g_slist_free_1(l);
194 }
195 }
196
197 bool block_job_has_bdrv(BlockJob *job, BlockDriverState *bs)
198 {
199 GSList *el;
200
201 for (el = job->nodes; el; el = el->next) {
202 BdrvChild *c = el->data;
203 if (c->bs == bs) {
204 return true;
205 }
206 }
207
208 return false;
209 }
210
211 int block_job_add_bdrv(BlockJob *job, const char *name, BlockDriverState *bs,
212 uint64_t perm, uint64_t shared_perm, Error **errp)
213 {
214 BdrvChild *c;
215 bool need_context_ops;
216
217 bdrv_ref(bs);
218
219 need_context_ops = bdrv_get_aio_context(bs) != job->job.aio_context;
220
221 if (need_context_ops && job->job.aio_context != qemu_get_aio_context()) {
222 aio_context_release(job->job.aio_context);
223 }
224 c = bdrv_root_attach_child(bs, name, &child_job, 0,
225 job->job.aio_context, perm, shared_perm, job,
226 errp);
227 if (need_context_ops && job->job.aio_context != qemu_get_aio_context()) {
228 aio_context_acquire(job->job.aio_context);
229 }
230 if (c == NULL) {
231 return -EPERM;
232 }
233
234 job->nodes = g_slist_prepend(job->nodes, c);
235 bdrv_op_block_all(bs, job->blocker);
236
237 return 0;
238 }
239
240 static void block_job_on_idle(Notifier *n, void *opaque)
241 {
242 aio_wait_kick();
243 }
244
245 bool block_job_is_internal(BlockJob *job)
246 {
247 return (job->job.id == NULL);
248 }
249
250 const BlockJobDriver *block_job_driver(BlockJob *job)
251 {
252 return container_of(job->job.driver, BlockJobDriver, job_driver);
253 }
254
255 /* Assumes the job_mutex is held */
256 static bool job_timer_pending(Job *job)
257 {
258 return timer_pending(&job->sleep_timer);
259 }
260
261 void block_job_set_speed(BlockJob *job, int64_t speed, Error **errp)
262 {
263 const BlockJobDriver *drv = block_job_driver(job);
264 int64_t old_speed = job->speed;
265
266 if (job_apply_verb(&job->job, JOB_VERB_SET_SPEED, errp)) {
267 return;
268 }
269 if (speed < 0) {
270 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "speed",
271 "a non-negative value");
272 return;
273 }
274
275 ratelimit_set_speed(&job->limit, speed, BLOCK_JOB_SLICE_TIME);
276
277 job->speed = speed;
278
279 if (drv->set_speed) {
280 drv->set_speed(job, speed);
281 }
282
283 if (speed && speed <= old_speed) {
284 return;
285 }
286
287 /* kick only if a timer is pending */
288 job_enter_cond(&job->job, job_timer_pending);
289 }
290
291 int64_t block_job_ratelimit_get_delay(BlockJob *job, uint64_t n)
292 {
293 if (!job->speed) {
294 return 0;
295 }
296
297 return ratelimit_calculate_delay(&job->limit, n);
298 }
299
300 BlockJobInfo *block_job_query(BlockJob *job, Error **errp)
301 {
302 BlockJobInfo *info;
303
304 if (block_job_is_internal(job)) {
305 error_setg(errp, "Cannot query QEMU internal jobs");
306 return NULL;
307 }
308 info = g_new0(BlockJobInfo, 1);
309 info->type = g_strdup(job_type_str(&job->job));
310 info->device = g_strdup(job->job.id);
311 info->busy = qatomic_read(&job->job.busy);
312 info->paused = job->job.pause_count > 0;
313 info->offset = job->job.progress.current;
314 info->len = job->job.progress.total;
315 info->speed = job->speed;
316 info->io_status = job->iostatus;
317 info->ready = job_is_ready(&job->job),
318 info->status = job->job.status;
319 info->auto_finalize = job->job.auto_finalize;
320 info->auto_dismiss = job->job.auto_dismiss;
321 info->has_error = job->job.ret != 0;
322 info->error = job->job.ret ? g_strdup(strerror(-job->job.ret)) : NULL;
323 return info;
324 }
325
326 static void block_job_iostatus_set_err(BlockJob *job, int error)
327 {
328 if (job->iostatus == BLOCK_DEVICE_IO_STATUS_OK) {
329 job->iostatus = error == ENOSPC ? BLOCK_DEVICE_IO_STATUS_NOSPACE :
330 BLOCK_DEVICE_IO_STATUS_FAILED;
331 }
332 }
333
334 static void block_job_event_cancelled(Notifier *n, void *opaque)
335 {
336 BlockJob *job = opaque;
337
338 if (block_job_is_internal(job)) {
339 return;
340 }
341
342 qapi_event_send_block_job_cancelled(job_type(&job->job),
343 job->job.id,
344 job->job.progress.total,
345 job->job.progress.current,
346 job->speed);
347 }
348
349 static void block_job_event_completed(Notifier *n, void *opaque)
350 {
351 BlockJob *job = opaque;
352 const char *msg = NULL;
353
354 if (block_job_is_internal(job)) {
355 return;
356 }
357
358 if (job->job.ret < 0) {
359 msg = strerror(-job->job.ret);
360 }
361
362 qapi_event_send_block_job_completed(job_type(&job->job),
363 job->job.id,
364 job->job.progress.total,
365 job->job.progress.current,
366 job->speed,
367 !!msg,
368 msg);
369 }
370
371 static void block_job_event_pending(Notifier *n, void *opaque)
372 {
373 BlockJob *job = opaque;
374
375 if (block_job_is_internal(job)) {
376 return;
377 }
378
379 qapi_event_send_block_job_pending(job_type(&job->job),
380 job->job.id);
381 }
382
383 static void block_job_event_ready(Notifier *n, void *opaque)
384 {
385 BlockJob *job = opaque;
386
387 if (block_job_is_internal(job)) {
388 return;
389 }
390
391 qapi_event_send_block_job_ready(job_type(&job->job),
392 job->job.id,
393 job->job.progress.total,
394 job->job.progress.current,
395 job->speed);
396 }
397
398
399 /*
400 * API for block job drivers and the block layer. These functions are
401 * declared in blockjob_int.h.
402 */
403
404 void *block_job_create(const char *job_id, const BlockJobDriver *driver,
405 JobTxn *txn, BlockDriverState *bs, uint64_t perm,
406 uint64_t shared_perm, int64_t speed, int flags,
407 BlockCompletionFunc *cb, void *opaque, Error **errp)
408 {
409 BlockBackend *blk;
410 BlockJob *job;
411
412 if (job_id == NULL && !(flags & JOB_INTERNAL)) {
413 job_id = bdrv_get_device_name(bs);
414 }
415
416 blk = blk_new_with_bs(bs, perm, shared_perm, errp);
417 if (!blk) {
418 return NULL;
419 }
420
421 job = job_create(job_id, &driver->job_driver, txn, blk_get_aio_context(blk),
422 flags, cb, opaque, errp);
423 if (job == NULL) {
424 blk_unref(blk);
425 return NULL;
426 }
427
428 assert(is_block_job(&job->job));
429 assert(job->job.driver->free == &block_job_free);
430 assert(job->job.driver->user_resume == &block_job_user_resume);
431
432 job->blk = blk;
433
434 job->finalize_cancelled_notifier.notify = block_job_event_cancelled;
435 job->finalize_completed_notifier.notify = block_job_event_completed;
436 job->pending_notifier.notify = block_job_event_pending;
437 job->ready_notifier.notify = block_job_event_ready;
438 job->idle_notifier.notify = block_job_on_idle;
439
440 notifier_list_add(&job->job.on_finalize_cancelled,
441 &job->finalize_cancelled_notifier);
442 notifier_list_add(&job->job.on_finalize_completed,
443 &job->finalize_completed_notifier);
444 notifier_list_add(&job->job.on_pending, &job->pending_notifier);
445 notifier_list_add(&job->job.on_ready, &job->ready_notifier);
446 notifier_list_add(&job->job.on_idle, &job->idle_notifier);
447
448 error_setg(&job->blocker, "block device is in use by block job: %s",
449 job_type_str(&job->job));
450 block_job_add_bdrv(job, "main node", bs, 0, BLK_PERM_ALL, &error_abort);
451
452 bdrv_op_unblock(bs, BLOCK_OP_TYPE_DATAPLANE, job->blocker);
453
454 /* Disable request queuing in the BlockBackend to avoid deadlocks on drain:
455 * The job reports that it's busy until it reaches a pause point. */
456 blk_set_disable_request_queuing(blk, true);
457 blk_set_allow_aio_context_change(blk, true);
458
459 /* Only set speed when necessary to avoid NotSupported error */
460 if (speed != 0) {
461 Error *local_err = NULL;
462
463 block_job_set_speed(job, speed, &local_err);
464 if (local_err) {
465 job_early_fail(&job->job);
466 error_propagate(errp, local_err);
467 return NULL;
468 }
469 }
470
471 return job;
472 }
473
474 void block_job_iostatus_reset(BlockJob *job)
475 {
476 if (job->iostatus == BLOCK_DEVICE_IO_STATUS_OK) {
477 return;
478 }
479 assert(job->job.user_paused && job->job.pause_count > 0);
480 job->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
481 }
482
483 void block_job_user_resume(Job *job)
484 {
485 BlockJob *bjob = container_of(job, BlockJob, job);
486 block_job_iostatus_reset(bjob);
487 }
488
489 BlockErrorAction block_job_error_action(BlockJob *job, BlockdevOnError on_err,
490 int is_read, int error)
491 {
492 BlockErrorAction action;
493
494 switch (on_err) {
495 case BLOCKDEV_ON_ERROR_ENOSPC:
496 case BLOCKDEV_ON_ERROR_AUTO:
497 action = (error == ENOSPC) ?
498 BLOCK_ERROR_ACTION_STOP : BLOCK_ERROR_ACTION_REPORT;
499 break;
500 case BLOCKDEV_ON_ERROR_STOP:
501 action = BLOCK_ERROR_ACTION_STOP;
502 break;
503 case BLOCKDEV_ON_ERROR_REPORT:
504 action = BLOCK_ERROR_ACTION_REPORT;
505 break;
506 case BLOCKDEV_ON_ERROR_IGNORE:
507 action = BLOCK_ERROR_ACTION_IGNORE;
508 break;
509 default:
510 abort();
511 }
512 if (!block_job_is_internal(job)) {
513 qapi_event_send_block_job_error(job->job.id,
514 is_read ? IO_OPERATION_TYPE_READ :
515 IO_OPERATION_TYPE_WRITE,
516 action);
517 }
518 if (action == BLOCK_ERROR_ACTION_STOP) {
519 if (!job->job.user_paused) {
520 job_pause(&job->job);
521 /* make the pause user visible, which will be resumed from QMP. */
522 job->job.user_paused = true;
523 }
524 block_job_iostatus_set_err(job, error);
525 }
526 return action;
527 }