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 "qemu-common.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/coroutine.h"
37 #include "qemu/timer.h"
39 /* Transactional group of block jobs */
42 /* Is this txn being cancelled? */
46 QLIST_HEAD(, BlockJob) jobs;
53 * The block job API is composed of two categories of functions.
55 * The first includes functions used by the monitor. The monitor is
56 * peculiar in that it accesses the block job list with block_job_get, and
57 * therefore needs consistency across block_job_get and the actual operation
58 * (e.g. block_job_set_speed). The consistency is achieved with
59 * aio_context_acquire/release. These functions are declared in blockjob.h.
61 * The second includes functions used by the block job drivers and sometimes
62 * by the core block layer. These do not care about locking, because the
63 * whole coroutine runs under the AioContext lock, and are declared in
67 static bool is_block_job(Job *job)
69 return job_type(job) == JOB_TYPE_BACKUP ||
70 job_type(job) == JOB_TYPE_COMMIT ||
71 job_type(job) == JOB_TYPE_MIRROR ||
72 job_type(job) == JOB_TYPE_STREAM;
75 BlockJob *block_job_next(BlockJob *bjob)
77 Job *job = bjob ? &bjob->job : NULL;
81 } while (job && !is_block_job(job));
83 return job ? container_of(job, BlockJob, job) : NULL;
86 BlockJob *block_job_get(const char *id)
88 Job *job = job_get(id);
90 if (job && is_block_job(job)) {
91 return container_of(job, BlockJob, job);
97 BlockJobTxn *block_job_txn_new(void)
99 BlockJobTxn *txn = g_new0(BlockJobTxn, 1);
100 QLIST_INIT(&txn->jobs);
105 static void block_job_txn_ref(BlockJobTxn *txn)
110 void block_job_txn_unref(BlockJobTxn *txn)
112 if (txn && --txn->refcnt == 0) {
117 void block_job_txn_add_job(BlockJobTxn *txn, BlockJob *job)
126 QLIST_INSERT_HEAD(&txn->jobs, job, txn_list);
127 block_job_txn_ref(txn);
130 void block_job_txn_del_job(BlockJob *job)
133 QLIST_REMOVE(job, txn_list);
134 block_job_txn_unref(job->txn);
139 static void block_job_attached_aio_context(AioContext *new_context,
141 static void block_job_detach_aio_context(void *opaque);
143 void block_job_free(Job *job)
145 BlockJob *bjob = container_of(job, BlockJob, job);
146 BlockDriverState *bs = blk_bs(bjob->blk);
151 block_job_remove_all_bdrv(bjob);
152 blk_remove_aio_context_notifier(bjob->blk,
153 block_job_attached_aio_context,
154 block_job_detach_aio_context, bjob);
155 blk_unref(bjob->blk);
156 error_free(bjob->blocker);
159 static void block_job_attached_aio_context(AioContext *new_context,
162 BlockJob *job = opaque;
164 job->job.aio_context = new_context;
165 if (job->driver->attached_aio_context) {
166 job->driver->attached_aio_context(job, new_context);
169 job_resume(&job->job);
172 static void block_job_drain(BlockJob *job)
174 /* If job is !job->job.busy this kicks it into the next pause point. */
175 block_job_enter(job);
178 if (job->driver->drain) {
179 job->driver->drain(job);
183 static void block_job_detach_aio_context(void *opaque)
185 BlockJob *job = opaque;
187 /* In case the job terminates during aio_poll()... */
190 job_pause(&job->job);
192 while (!job->job.paused && !job_is_completed(&job->job)) {
193 block_job_drain(job);
196 job->job.aio_context = NULL;
197 job_unref(&job->job);
200 static char *child_job_get_parent_desc(BdrvChild *c)
202 BlockJob *job = c->opaque;
203 return g_strdup_printf("%s job '%s'", job_type_str(&job->job), job->job.id);
206 static void child_job_drained_begin(BdrvChild *c)
208 BlockJob *job = c->opaque;
209 job_pause(&job->job);
212 static void child_job_drained_end(BdrvChild *c)
214 BlockJob *job = c->opaque;
215 job_resume(&job->job);
218 static const BdrvChildRole child_job = {
219 .get_parent_desc = child_job_get_parent_desc,
220 .drained_begin = child_job_drained_begin,
221 .drained_end = child_job_drained_end,
222 .stay_at_node = true,
225 void block_job_remove_all_bdrv(BlockJob *job)
228 for (l = job->nodes; l; l = l->next) {
229 BdrvChild *c = l->data;
230 bdrv_op_unblock_all(c->bs, job->blocker);
231 bdrv_root_unref_child(c);
233 g_slist_free(job->nodes);
237 int block_job_add_bdrv(BlockJob *job, const char *name, BlockDriverState *bs,
238 uint64_t perm, uint64_t shared_perm, Error **errp)
242 c = bdrv_root_attach_child(bs, name, &child_job, perm, shared_perm,
248 job->nodes = g_slist_prepend(job->nodes, c);
250 bdrv_op_block_all(bs, job->blocker);
255 bool block_job_is_internal(BlockJob *job)
257 return (job->job.id == NULL);
260 const BlockJobDriver *block_job_driver(BlockJob *job)
265 static int block_job_prepare(BlockJob *job)
267 if (job->job.ret == 0 && job->driver->prepare) {
268 job->job.ret = job->driver->prepare(job);
273 static void job_cancel_async(Job *job, bool force)
275 if (job->user_paused) {
276 /* Do not call job_enter here, the caller will handle it. */
277 job->user_paused = false;
278 if (job->driver->user_resume) {
279 job->driver->user_resume(job);
281 assert(job->pause_count > 0);
284 job->cancelled = true;
285 /* To prevent 'force == false' overriding a previous 'force == true' */
286 job->force_cancel |= force;
289 static int block_job_txn_apply(BlockJobTxn *txn, int fn(BlockJob *), bool lock)
292 BlockJob *job, *next;
295 QLIST_FOREACH_SAFE(job, &txn->jobs, txn_list, next) {
297 ctx = blk_get_aio_context(job->blk);
298 aio_context_acquire(ctx);
302 aio_context_release(ctx);
311 static int block_job_finish_sync(BlockJob *job,
312 void (*finish)(BlockJob *, Error **errp),
315 Error *local_err = NULL;
318 assert(blk_bs(job->blk)->job == job);
323 finish(job, &local_err);
326 error_propagate(errp, local_err);
327 job_unref(&job->job);
330 /* block_job_drain calls block_job_enter, and it should be enough to
331 * induce progress until the job completes or moves to the main thread.
333 while (!job->job.deferred_to_main_loop && !job_is_completed(&job->job)) {
334 block_job_drain(job);
336 while (!job_is_completed(&job->job)) {
337 aio_poll(qemu_get_aio_context(), true);
339 ret = (job_is_cancelled(&job->job) && job->job.ret == 0)
340 ? -ECANCELED : job->job.ret;
341 job_unref(&job->job);
345 static void block_job_completed_txn_abort(BlockJob *job)
348 BlockJobTxn *txn = job->txn;
353 * We are cancelled by another job, which will handle everything.
357 txn->aborting = true;
358 block_job_txn_ref(txn);
360 /* We are the first failed job. Cancel other jobs. */
361 QLIST_FOREACH(other_job, &txn->jobs, txn_list) {
362 ctx = blk_get_aio_context(other_job->blk);
363 aio_context_acquire(ctx);
366 /* Other jobs are effectively cancelled by us, set the status for
367 * them; this job, however, may or may not be cancelled, depending
368 * on the caller, so leave it. */
369 QLIST_FOREACH(other_job, &txn->jobs, txn_list) {
370 if (other_job != job) {
371 job_cancel_async(&other_job->job, false);
374 while (!QLIST_EMPTY(&txn->jobs)) {
375 other_job = QLIST_FIRST(&txn->jobs);
376 ctx = blk_get_aio_context(other_job->blk);
377 if (!job_is_completed(&other_job->job)) {
378 assert(job_is_cancelled(&other_job->job));
379 block_job_finish_sync(other_job, NULL, NULL);
381 job_finalize_single(&other_job->job);
382 aio_context_release(ctx);
385 block_job_txn_unref(txn);
388 static int block_job_needs_finalize(BlockJob *job)
390 return !job->job.auto_finalize;
393 static int block_job_finalize_single(BlockJob *job)
395 return job_finalize_single(&job->job);
398 static void block_job_do_finalize(BlockJob *job)
401 assert(job && job->txn);
403 /* prepare the transaction to complete */
404 rc = block_job_txn_apply(job->txn, block_job_prepare, true);
406 block_job_completed_txn_abort(job);
408 block_job_txn_apply(job->txn, block_job_finalize_single, true);
412 static int block_job_transition_to_pending(BlockJob *job)
414 job_state_transition(&job->job, JOB_STATUS_PENDING);
415 if (!job->job.auto_finalize) {
416 job_event_pending(&job->job);
421 static void block_job_completed_txn_success(BlockJob *job)
423 BlockJobTxn *txn = job->txn;
426 job_state_transition(&job->job, JOB_STATUS_WAITING);
429 * Successful completion, see if there are other running jobs in this
432 QLIST_FOREACH(other_job, &txn->jobs, txn_list) {
433 if (!job_is_completed(&other_job->job)) {
436 assert(other_job->job.ret == 0);
439 block_job_txn_apply(txn, block_job_transition_to_pending, false);
441 /* If no jobs need manual finalization, automatically do so */
442 if (block_job_txn_apply(txn, block_job_needs_finalize, false) == 0) {
443 block_job_do_finalize(job);
447 /* Assumes the job_mutex is held */
448 static bool job_timer_pending(Job *job)
450 return timer_pending(&job->sleep_timer);
453 void block_job_set_speed(BlockJob *job, int64_t speed, Error **errp)
455 int64_t old_speed = job->speed;
457 if (job_apply_verb(&job->job, JOB_VERB_SET_SPEED, errp)) {
461 error_setg(errp, QERR_INVALID_PARAMETER, "speed");
465 ratelimit_set_speed(&job->limit, speed, BLOCK_JOB_SLICE_TIME);
468 if (speed && speed <= old_speed) {
472 /* kick only if a timer is pending */
473 job_enter_cond(&job->job, job_timer_pending);
476 int64_t block_job_ratelimit_get_delay(BlockJob *job, uint64_t n)
482 return ratelimit_calculate_delay(&job->limit, n);
485 void block_job_complete(BlockJob *job, Error **errp)
487 /* Should not be reachable via external interface for internal jobs */
489 if (job_apply_verb(&job->job, JOB_VERB_COMPLETE, errp)) {
492 if (job->job.pause_count || job_is_cancelled(&job->job) ||
493 !job->driver->complete)
495 error_setg(errp, "The active block job '%s' cannot be completed",
500 job->driver->complete(job, errp);
503 void block_job_finalize(BlockJob *job, Error **errp)
505 assert(job && job->job.id);
506 if (job_apply_verb(&job->job, JOB_VERB_FINALIZE, errp)) {
509 block_job_do_finalize(job);
512 void block_job_dismiss(BlockJob **jobptr, Error **errp)
514 BlockJob *job = *jobptr;
515 /* similarly to _complete, this is QMP-interface only. */
517 if (job_apply_verb(&job->job, JOB_VERB_DISMISS, errp)) {
521 job_do_dismiss(&job->job);
525 void block_job_cancel(BlockJob *job, bool force)
527 if (job->job.status == JOB_STATUS_CONCLUDED) {
528 job_do_dismiss(&job->job);
531 job_cancel_async(&job->job, force);
532 if (!job_started(&job->job)) {
533 block_job_completed(job, -ECANCELED);
534 } else if (job->job.deferred_to_main_loop) {
535 block_job_completed_txn_abort(job);
537 block_job_enter(job);
541 void block_job_user_cancel(BlockJob *job, bool force, Error **errp)
543 if (job_apply_verb(&job->job, JOB_VERB_CANCEL, errp)) {
546 block_job_cancel(job, force);
549 /* A wrapper around block_job_cancel() taking an Error ** parameter so it may be
550 * used with block_job_finish_sync() without the need for (rather nasty)
551 * function pointer casts there. */
552 static void block_job_cancel_err(BlockJob *job, Error **errp)
554 block_job_cancel(job, false);
557 int block_job_cancel_sync(BlockJob *job)
559 return block_job_finish_sync(job, &block_job_cancel_err, NULL);
562 void block_job_cancel_sync_all(void)
565 AioContext *aio_context;
567 while ((job = block_job_next(NULL))) {
568 aio_context = blk_get_aio_context(job->blk);
569 aio_context_acquire(aio_context);
570 block_job_cancel_sync(job);
571 aio_context_release(aio_context);
575 int block_job_complete_sync(BlockJob *job, Error **errp)
577 return block_job_finish_sync(job, &block_job_complete, errp);
580 void block_job_progress_update(BlockJob *job, uint64_t done)
585 void block_job_progress_set_remaining(BlockJob *job, uint64_t remaining)
587 job->len = job->offset + remaining;
590 BlockJobInfo *block_job_query(BlockJob *job, Error **errp)
594 if (block_job_is_internal(job)) {
595 error_setg(errp, "Cannot query QEMU internal jobs");
598 info = g_new0(BlockJobInfo, 1);
599 info->type = g_strdup(job_type_str(&job->job));
600 info->device = g_strdup(job->job.id);
601 info->len = job->len;
602 info->busy = atomic_read(&job->job.busy);
603 info->paused = job->job.pause_count > 0;
604 info->offset = job->offset;
605 info->speed = job->speed;
606 info->io_status = job->iostatus;
607 info->ready = job->ready;
608 info->status = job->job.status;
609 info->auto_finalize = job->job.auto_finalize;
610 info->auto_dismiss = job->job.auto_dismiss;
611 info->has_error = job->job.ret != 0;
612 info->error = job->job.ret ? g_strdup(strerror(-job->job.ret)) : NULL;
616 static void block_job_iostatus_set_err(BlockJob *job, int error)
618 if (job->iostatus == BLOCK_DEVICE_IO_STATUS_OK) {
619 job->iostatus = error == ENOSPC ? BLOCK_DEVICE_IO_STATUS_NOSPACE :
620 BLOCK_DEVICE_IO_STATUS_FAILED;
624 static void block_job_event_cancelled(Notifier *n, void *opaque)
626 BlockJob *job = opaque;
628 if (block_job_is_internal(job)) {
632 qapi_event_send_block_job_cancelled(job_type(&job->job),
640 static void block_job_event_completed(Notifier *n, void *opaque)
642 BlockJob *job = opaque;
643 const char *msg = NULL;
645 if (block_job_is_internal(job)) {
649 if (job->job.ret < 0) {
650 msg = strerror(-job->job.ret);
653 qapi_event_send_block_job_completed(job_type(&job->job),
663 static void block_job_event_pending(Notifier *n, void *opaque)
665 BlockJob *job = opaque;
667 if (block_job_is_internal(job)) {
671 qapi_event_send_block_job_pending(job_type(&job->job),
677 * API for block job drivers and the block layer. These functions are
678 * declared in blockjob_int.h.
681 void *block_job_create(const char *job_id, const BlockJobDriver *driver,
682 BlockJobTxn *txn, BlockDriverState *bs, uint64_t perm,
683 uint64_t shared_perm, int64_t speed, int flags,
684 BlockCompletionFunc *cb, void *opaque, Error **errp)
691 error_setg(errp, QERR_DEVICE_IN_USE, bdrv_get_device_name(bs));
695 if (job_id == NULL && !(flags & JOB_INTERNAL)) {
696 job_id = bdrv_get_device_name(bs);
699 blk = blk_new(perm, shared_perm);
700 ret = blk_insert_bs(blk, bs, errp);
706 job = job_create(job_id, &driver->job_driver, blk_get_aio_context(blk),
707 flags, cb, opaque, errp);
713 assert(is_block_job(&job->job));
714 assert(job->job.driver->free == &block_job_free);
715 assert(job->job.driver->user_resume == &block_job_user_resume);
717 job->driver = driver;
720 job->finalize_cancelled_notifier.notify = block_job_event_cancelled;
721 job->finalize_completed_notifier.notify = block_job_event_completed;
722 job->pending_notifier.notify = block_job_event_pending;
724 notifier_list_add(&job->job.on_finalize_cancelled,
725 &job->finalize_cancelled_notifier);
726 notifier_list_add(&job->job.on_finalize_completed,
727 &job->finalize_completed_notifier);
728 notifier_list_add(&job->job.on_pending, &job->pending_notifier);
730 error_setg(&job->blocker, "block device is in use by block job: %s",
731 job_type_str(&job->job));
732 block_job_add_bdrv(job, "main node", bs, 0, BLK_PERM_ALL, &error_abort);
735 bdrv_op_unblock(bs, BLOCK_OP_TYPE_DATAPLANE, job->blocker);
737 blk_add_aio_context_notifier(blk, block_job_attached_aio_context,
738 block_job_detach_aio_context, job);
740 /* Only set speed when necessary to avoid NotSupported error */
742 Error *local_err = NULL;
744 block_job_set_speed(job, speed, &local_err);
746 job_early_fail(&job->job);
747 error_propagate(errp, local_err);
752 /* Single jobs are modeled as single-job transactions for sake of
753 * consolidating the job management logic */
755 txn = block_job_txn_new();
756 block_job_txn_add_job(txn, job);
757 block_job_txn_unref(txn);
759 block_job_txn_add_job(txn, job);
765 void block_job_completed(BlockJob *job, int ret)
767 assert(job && job->txn && !job_is_completed(&job->job));
768 assert(blk_bs(job->blk)->job == job);
770 job_update_rc(&job->job);
771 trace_block_job_completed(job, ret, job->job.ret);
773 block_job_completed_txn_abort(job);
775 block_job_completed_txn_success(job);
779 void block_job_enter(BlockJob *job)
781 job_enter_cond(&job->job, NULL);
784 void block_job_yield(BlockJob *job)
786 assert(job->job.busy);
788 /* Check cancellation *before* setting busy = false, too! */
789 if (job_is_cancelled(&job->job)) {
793 if (!job_should_pause(&job->job)) {
794 job_do_yield(&job->job, -1);
797 job_pause_point(&job->job);
800 void block_job_iostatus_reset(BlockJob *job)
802 if (job->iostatus == BLOCK_DEVICE_IO_STATUS_OK) {
805 assert(job->job.user_paused && job->job.pause_count > 0);
806 job->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
809 void block_job_user_resume(Job *job)
811 BlockJob *bjob = container_of(job, BlockJob, job);
812 block_job_iostatus_reset(bjob);
815 void block_job_event_ready(BlockJob *job)
817 job_state_transition(&job->job, JOB_STATUS_READY);
820 if (block_job_is_internal(job)) {
824 qapi_event_send_block_job_ready(job_type(&job->job),
828 job->speed, &error_abort);
831 BlockErrorAction block_job_error_action(BlockJob *job, BlockdevOnError on_err,
832 int is_read, int error)
834 BlockErrorAction action;
837 case BLOCKDEV_ON_ERROR_ENOSPC:
838 case BLOCKDEV_ON_ERROR_AUTO:
839 action = (error == ENOSPC) ?
840 BLOCK_ERROR_ACTION_STOP : BLOCK_ERROR_ACTION_REPORT;
842 case BLOCKDEV_ON_ERROR_STOP:
843 action = BLOCK_ERROR_ACTION_STOP;
845 case BLOCKDEV_ON_ERROR_REPORT:
846 action = BLOCK_ERROR_ACTION_REPORT;
848 case BLOCKDEV_ON_ERROR_IGNORE:
849 action = BLOCK_ERROR_ACTION_IGNORE;
854 if (!block_job_is_internal(job)) {
855 qapi_event_send_block_job_error(job->job.id,
856 is_read ? IO_OPERATION_TYPE_READ :
857 IO_OPERATION_TYPE_WRITE,
858 action, &error_abort);
860 if (action == BLOCK_ERROR_ACTION_STOP) {
861 job_pause(&job->job);
862 /* make the pause user visible, which will be resumed from QMP. */
863 job->job.user_paused = true;
864 block_job_iostatus_set_err(job, error);