2 * QEMU System Emulator block driver
4 * Copyright (c) 2011 IBM Corp.
5 * Copyright (c) 2012 Red Hat, Inc.
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:
14 * The above copyright notice and this permission notice shall be included in
15 * all copies or substantial portions of the Software.
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
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"
40 * The block job API is composed of two categories of functions.
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.
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
54 static bool is_block_job(Job *job)
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;
62 BlockJob *block_job_next(BlockJob *bjob)
64 Job *job = bjob ? &bjob->job : NULL;
68 } while (job && !is_block_job(job));
70 return job ? container_of(job, BlockJob, job) : NULL;
73 BlockJob *block_job_get(const char *id)
75 Job *job = job_get(id);
77 if (job && is_block_job(job)) {
78 return container_of(job, BlockJob, job);
84 void block_job_free(Job *job)
86 BlockJob *bjob = container_of(job, BlockJob, job);
88 block_job_remove_all_bdrv(bjob);
90 error_free(bjob->blocker);
93 static char *child_job_get_parent_desc(BdrvChild *c)
95 BlockJob *job = c->opaque;
96 return g_strdup_printf("%s job '%s'", job_type_str(&job->job), job->job.id);
99 static void child_job_drained_begin(BdrvChild *c)
101 BlockJob *job = c->opaque;
102 job_pause(&job->job);
105 static bool child_job_drained_poll(BdrvChild *c)
107 BlockJob *bjob = c->opaque;
108 Job *job = &bjob->job;
109 const BlockJobDriver *drv = block_job_driver(bjob);
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)) {
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);
127 static void child_job_drained_end(BdrvChild *c, int *drained_end_counter)
129 BlockJob *job = c->opaque;
130 job_resume(&job->job);
133 static bool child_job_can_set_aio_ctx(BdrvChild *c, AioContext *ctx,
134 GSList **ignore, Error **errp)
136 BlockJob *job = c->opaque;
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)) {
148 static void child_job_set_aio_ctx(BdrvChild *c, AioContext *ctx,
151 BlockJob *job = c->opaque;
154 for (l = job->nodes; l; l = l->next) {
155 BdrvChild *sibling = l->data;
156 if (g_slist_find(*ignore, sibling)) {
159 *ignore = g_slist_prepend(*ignore, sibling);
160 bdrv_set_aio_context_ignore(sibling->bs, ctx, ignore);
163 job->job.aio_context = ctx;
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,
176 void block_job_remove_all_bdrv(BlockJob *job)
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.
185 GSList *l = job->nodes;
186 BdrvChild *c = l->data;
188 job->nodes = l->next;
190 bdrv_op_unblock_all(c->bs, job->blocker);
191 bdrv_root_unref_child(c);
197 bool block_job_has_bdrv(BlockJob *job, BlockDriverState *bs)
201 for (el = job->nodes; el; el = el->next) {
202 BdrvChild *c = el->data;
211 int block_job_add_bdrv(BlockJob *job, const char *name, BlockDriverState *bs,
212 uint64_t perm, uint64_t shared_perm, Error **errp)
215 bool need_context_ops;
219 need_context_ops = bdrv_get_aio_context(bs) != job->job.aio_context;
221 if (need_context_ops && job->job.aio_context != qemu_get_aio_context()) {
222 aio_context_release(job->job.aio_context);
224 c = bdrv_root_attach_child(bs, name, &child_job, 0,
225 job->job.aio_context, perm, shared_perm, job,
227 if (need_context_ops && job->job.aio_context != qemu_get_aio_context()) {
228 aio_context_acquire(job->job.aio_context);
234 job->nodes = g_slist_prepend(job->nodes, c);
235 bdrv_op_block_all(bs, job->blocker);
240 static void block_job_on_idle(Notifier *n, void *opaque)
245 bool block_job_is_internal(BlockJob *job)
247 return (job->job.id == NULL);
250 const BlockJobDriver *block_job_driver(BlockJob *job)
252 return container_of(job->job.driver, BlockJobDriver, job_driver);
255 /* Assumes the job_mutex is held */
256 static bool job_timer_pending(Job *job)
258 return timer_pending(&job->sleep_timer);
261 bool block_job_set_speed(BlockJob *job, int64_t speed, Error **errp)
263 const BlockJobDriver *drv = block_job_driver(job);
264 int64_t old_speed = job->speed;
266 if (job_apply_verb(&job->job, JOB_VERB_SET_SPEED, errp) < 0) {
270 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "speed",
271 "a non-negative value");
275 ratelimit_set_speed(&job->limit, speed, BLOCK_JOB_SLICE_TIME);
279 if (drv->set_speed) {
280 drv->set_speed(job, speed);
283 if (speed && speed <= old_speed) {
287 /* kick only if a timer is pending */
288 job_enter_cond(&job->job, job_timer_pending);
293 int64_t block_job_ratelimit_get_delay(BlockJob *job, uint64_t n)
299 return ratelimit_calculate_delay(&job->limit, n);
302 BlockJobInfo *block_job_query(BlockJob *job, Error **errp)
306 if (block_job_is_internal(job)) {
307 error_setg(errp, "Cannot query QEMU internal jobs");
310 info = g_new0(BlockJobInfo, 1);
311 info->type = g_strdup(job_type_str(&job->job));
312 info->device = g_strdup(job->job.id);
313 info->busy = qatomic_read(&job->job.busy);
314 info->paused = job->job.pause_count > 0;
315 info->offset = job->job.progress.current;
316 info->len = job->job.progress.total;
317 info->speed = job->speed;
318 info->io_status = job->iostatus;
319 info->ready = job_is_ready(&job->job),
320 info->status = job->job.status;
321 info->auto_finalize = job->job.auto_finalize;
322 info->auto_dismiss = job->job.auto_dismiss;
323 info->has_error = job->job.ret != 0;
324 info->error = job->job.ret ? g_strdup(strerror(-job->job.ret)) : NULL;
328 static void block_job_iostatus_set_err(BlockJob *job, int error)
330 if (job->iostatus == BLOCK_DEVICE_IO_STATUS_OK) {
331 job->iostatus = error == ENOSPC ? BLOCK_DEVICE_IO_STATUS_NOSPACE :
332 BLOCK_DEVICE_IO_STATUS_FAILED;
336 static void block_job_event_cancelled(Notifier *n, void *opaque)
338 BlockJob *job = opaque;
340 if (block_job_is_internal(job)) {
344 qapi_event_send_block_job_cancelled(job_type(&job->job),
346 job->job.progress.total,
347 job->job.progress.current,
351 static void block_job_event_completed(Notifier *n, void *opaque)
353 BlockJob *job = opaque;
354 const char *msg = NULL;
356 if (block_job_is_internal(job)) {
360 if (job->job.ret < 0) {
361 msg = strerror(-job->job.ret);
364 qapi_event_send_block_job_completed(job_type(&job->job),
366 job->job.progress.total,
367 job->job.progress.current,
373 static void block_job_event_pending(Notifier *n, void *opaque)
375 BlockJob *job = opaque;
377 if (block_job_is_internal(job)) {
381 qapi_event_send_block_job_pending(job_type(&job->job),
385 static void block_job_event_ready(Notifier *n, void *opaque)
387 BlockJob *job = opaque;
389 if (block_job_is_internal(job)) {
393 qapi_event_send_block_job_ready(job_type(&job->job),
395 job->job.progress.total,
396 job->job.progress.current,
402 * API for block job drivers and the block layer. These functions are
403 * declared in blockjob_int.h.
406 void *block_job_create(const char *job_id, const BlockJobDriver *driver,
407 JobTxn *txn, BlockDriverState *bs, uint64_t perm,
408 uint64_t shared_perm, int64_t speed, int flags,
409 BlockCompletionFunc *cb, void *opaque, Error **errp)
414 if (job_id == NULL && !(flags & JOB_INTERNAL)) {
415 job_id = bdrv_get_device_name(bs);
418 blk = blk_new_with_bs(bs, perm, shared_perm, errp);
423 job = job_create(job_id, &driver->job_driver, txn, blk_get_aio_context(blk),
424 flags, cb, opaque, errp);
430 assert(is_block_job(&job->job));
431 assert(job->job.driver->free == &block_job_free);
432 assert(job->job.driver->user_resume == &block_job_user_resume);
436 job->finalize_cancelled_notifier.notify = block_job_event_cancelled;
437 job->finalize_completed_notifier.notify = block_job_event_completed;
438 job->pending_notifier.notify = block_job_event_pending;
439 job->ready_notifier.notify = block_job_event_ready;
440 job->idle_notifier.notify = block_job_on_idle;
442 notifier_list_add(&job->job.on_finalize_cancelled,
443 &job->finalize_cancelled_notifier);
444 notifier_list_add(&job->job.on_finalize_completed,
445 &job->finalize_completed_notifier);
446 notifier_list_add(&job->job.on_pending, &job->pending_notifier);
447 notifier_list_add(&job->job.on_ready, &job->ready_notifier);
448 notifier_list_add(&job->job.on_idle, &job->idle_notifier);
450 error_setg(&job->blocker, "block device is in use by block job: %s",
451 job_type_str(&job->job));
452 block_job_add_bdrv(job, "main node", bs, 0, BLK_PERM_ALL, &error_abort);
454 bdrv_op_unblock(bs, BLOCK_OP_TYPE_DATAPLANE, job->blocker);
456 /* Disable request queuing in the BlockBackend to avoid deadlocks on drain:
457 * The job reports that it's busy until it reaches a pause point. */
458 blk_set_disable_request_queuing(blk, true);
459 blk_set_allow_aio_context_change(blk, true);
461 /* Only set speed when necessary to avoid NotSupported error */
463 if (!block_job_set_speed(job, speed, errp)) {
464 job_early_fail(&job->job);
472 void block_job_iostatus_reset(BlockJob *job)
474 if (job->iostatus == BLOCK_DEVICE_IO_STATUS_OK) {
477 assert(job->job.user_paused && job->job.pause_count > 0);
478 job->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
481 void block_job_user_resume(Job *job)
483 BlockJob *bjob = container_of(job, BlockJob, job);
484 block_job_iostatus_reset(bjob);
487 BlockErrorAction block_job_error_action(BlockJob *job, BlockdevOnError on_err,
488 int is_read, int error)
490 BlockErrorAction action;
493 case BLOCKDEV_ON_ERROR_ENOSPC:
494 case BLOCKDEV_ON_ERROR_AUTO:
495 action = (error == ENOSPC) ?
496 BLOCK_ERROR_ACTION_STOP : BLOCK_ERROR_ACTION_REPORT;
498 case BLOCKDEV_ON_ERROR_STOP:
499 action = BLOCK_ERROR_ACTION_STOP;
501 case BLOCKDEV_ON_ERROR_REPORT:
502 action = BLOCK_ERROR_ACTION_REPORT;
504 case BLOCKDEV_ON_ERROR_IGNORE:
505 action = BLOCK_ERROR_ACTION_IGNORE;
510 if (!block_job_is_internal(job)) {
511 qapi_event_send_block_job_error(job->job.id,
512 is_read ? IO_OPERATION_TYPE_READ :
513 IO_OPERATION_TYPE_WRITE,
516 if (action == BLOCK_ERROR_ACTION_STOP) {
517 if (!job->job.user_paused) {
518 job_pause(&job->job);
519 /* make the pause user visible, which will be resumed from QMP. */
520 job->job.user_paused = true;
522 block_job_iostatus_set_err(job, error);