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 "sysemu/block-backend.h"
32 #include "qapi/qmp/qerror.h"
33 #include "qapi/qmp/qjson.h"
34 #include "qemu/coroutine.h"
36 #include "qmp-commands.h"
37 #include "qemu/timer.h"
38 #include "qapi-event.h"
40 static void block_job_event_cancelled(BlockJob *job);
41 static void block_job_event_completed(BlockJob *job, const char *msg);
43 /* Transactional group of block jobs */
46 /* Is this txn being cancelled? */
50 QLIST_HEAD(, BlockJob) jobs;
56 static QLIST_HEAD(, BlockJob) block_jobs = QLIST_HEAD_INITIALIZER(block_jobs);
58 static char *child_job_get_parent_desc(BdrvChild *c)
60 BlockJob *job = c->opaque;
61 return g_strdup_printf("%s job '%s'",
62 BlockJobType_lookup[job->driver->job_type],
66 static const BdrvChildRole child_job = {
67 .get_parent_desc = child_job_get_parent_desc,
71 static void block_job_drained_begin(void *opaque)
73 BlockJob *job = opaque;
77 static void block_job_drained_end(void *opaque)
79 BlockJob *job = opaque;
80 block_job_resume(job);
83 static const BlockDevOps block_job_dev_ops = {
84 .drained_begin = block_job_drained_begin,
85 .drained_end = block_job_drained_end,
88 BlockJob *block_job_next(BlockJob *job)
91 return QLIST_FIRST(&block_jobs);
93 return QLIST_NEXT(job, job_list);
96 BlockJob *block_job_get(const char *id)
100 QLIST_FOREACH(job, &block_jobs, job_list) {
101 if (job->id && !strcmp(id, job->id)) {
109 static void block_job_ref(BlockJob *job)
114 static void block_job_attached_aio_context(AioContext *new_context,
116 static void block_job_detach_aio_context(void *opaque);
118 static void block_job_unref(BlockJob *job)
120 if (--job->refcnt == 0) {
121 BlockDriverState *bs = blk_bs(job->blk);
123 block_job_remove_all_bdrv(job);
124 blk_remove_aio_context_notifier(job->blk,
125 block_job_attached_aio_context,
126 block_job_detach_aio_context, job);
128 error_free(job->blocker);
130 QLIST_REMOVE(job, job_list);
135 static void block_job_attached_aio_context(AioContext *new_context,
138 BlockJob *job = opaque;
140 if (job->driver->attached_aio_context) {
141 job->driver->attached_aio_context(job, new_context);
144 block_job_resume(job);
147 static void block_job_drain(BlockJob *job)
149 /* If job is !job->busy this kicks it into the next pause point. */
150 block_job_enter(job);
153 if (job->driver->drain) {
154 job->driver->drain(job);
158 static void block_job_detach_aio_context(void *opaque)
160 BlockJob *job = opaque;
162 /* In case the job terminates during aio_poll()... */
165 block_job_pause(job);
167 while (!job->paused && !job->completed) {
168 block_job_drain(job);
171 block_job_unref(job);
174 void block_job_remove_all_bdrv(BlockJob *job)
177 for (l = job->nodes; l; l = l->next) {
178 BdrvChild *c = l->data;
179 bdrv_op_unblock_all(c->bs, job->blocker);
180 bdrv_root_unref_child(c);
182 g_slist_free(job->nodes);
186 int block_job_add_bdrv(BlockJob *job, const char *name, BlockDriverState *bs,
187 uint64_t perm, uint64_t shared_perm, Error **errp)
191 c = bdrv_root_attach_child(bs, name, &child_job, perm, shared_perm,
197 job->nodes = g_slist_prepend(job->nodes, c);
199 bdrv_op_block_all(bs, job->blocker);
204 void *block_job_create(const char *job_id, const BlockJobDriver *driver,
205 BlockDriverState *bs, uint64_t perm,
206 uint64_t shared_perm, int64_t speed, int flags,
207 BlockCompletionFunc *cb, void *opaque, Error **errp)
214 error_setg(errp, QERR_DEVICE_IN_USE, bdrv_get_device_name(bs));
218 if (job_id == NULL && !(flags & BLOCK_JOB_INTERNAL)) {
219 job_id = bdrv_get_device_name(bs);
221 error_setg(errp, "An explicit job ID is required for this node");
227 if (flags & BLOCK_JOB_INTERNAL) {
228 error_setg(errp, "Cannot specify job ID for internal block job");
232 if (!id_wellformed(job_id)) {
233 error_setg(errp, "Invalid job ID '%s'", job_id);
237 if (block_job_get(job_id)) {
238 error_setg(errp, "Job ID '%s' already in use", job_id);
243 blk = blk_new(perm, shared_perm);
244 ret = blk_insert_bs(blk, bs, errp);
250 job = g_malloc0(driver->instance_size);
251 job->driver = driver;
252 job->id = g_strdup(job_id);
255 job->opaque = opaque;
258 job->pause_count = 1;
261 error_setg(&job->blocker, "block device is in use by block job: %s",
262 BlockJobType_lookup[driver->job_type]);
263 block_job_add_bdrv(job, "main node", bs, 0, BLK_PERM_ALL, &error_abort);
266 blk_set_dev_ops(blk, &block_job_dev_ops, job);
267 bdrv_op_unblock(bs, BLOCK_OP_TYPE_DATAPLANE, job->blocker);
269 QLIST_INSERT_HEAD(&block_jobs, job, job_list);
271 blk_add_aio_context_notifier(blk, block_job_attached_aio_context,
272 block_job_detach_aio_context, job);
274 /* Only set speed when necessary to avoid NotSupported error */
276 Error *local_err = NULL;
278 block_job_set_speed(job, speed, &local_err);
280 block_job_unref(job);
281 error_propagate(errp, local_err);
288 bool block_job_is_internal(BlockJob *job)
290 return (job->id == NULL);
293 static bool block_job_started(BlockJob *job)
299 * All jobs must allow a pause point before entering their job proper. This
300 * ensures that jobs can be paused prior to being started, then resumed later.
302 static void coroutine_fn block_job_co_entry(void *opaque)
304 BlockJob *job = opaque;
306 assert(job && job->driver && job->driver->start);
307 block_job_pause_point(job);
308 job->driver->start(job);
311 void block_job_start(BlockJob *job)
313 assert(job && !block_job_started(job) && job->paused &&
314 job->driver && job->driver->start);
315 job->co = qemu_coroutine_create(block_job_co_entry, job);
319 bdrv_coroutine_enter(blk_bs(job->blk), job->co);
322 void block_job_early_fail(BlockJob *job)
324 block_job_unref(job);
327 static void block_job_completed_single(BlockJob *job)
330 if (job->driver->commit) {
331 job->driver->commit(job);
334 if (job->driver->abort) {
335 job->driver->abort(job);
338 if (job->driver->clean) {
339 job->driver->clean(job);
343 job->cb(job->opaque, job->ret);
346 /* Emit events only if we actually started */
347 if (block_job_started(job)) {
348 if (block_job_is_cancelled(job)) {
349 block_job_event_cancelled(job);
351 const char *msg = NULL;
353 msg = strerror(-job->ret);
355 block_job_event_completed(job, msg);
360 QLIST_REMOVE(job, txn_list);
361 block_job_txn_unref(job->txn);
363 block_job_unref(job);
366 static void block_job_completed_txn_abort(BlockJob *job)
369 BlockJobTxn *txn = job->txn;
370 BlockJob *other_job, *next;
374 * We are cancelled by another job, which will handle everything.
378 txn->aborting = true;
379 /* We are the first failed job. Cancel other jobs. */
380 QLIST_FOREACH(other_job, &txn->jobs, txn_list) {
381 ctx = blk_get_aio_context(other_job->blk);
382 aio_context_acquire(ctx);
384 QLIST_FOREACH(other_job, &txn->jobs, txn_list) {
385 if (other_job == job || other_job->completed) {
386 /* Other jobs are "effectively" cancelled by us, set the status for
387 * them; this job, however, may or may not be cancelled, depending
388 * on the caller, so leave it. */
389 if (other_job != job) {
390 other_job->cancelled = true;
394 block_job_cancel_sync(other_job);
395 assert(other_job->completed);
397 QLIST_FOREACH_SAFE(other_job, &txn->jobs, txn_list, next) {
398 ctx = blk_get_aio_context(other_job->blk);
399 block_job_completed_single(other_job);
400 aio_context_release(ctx);
404 static void block_job_completed_txn_success(BlockJob *job)
407 BlockJobTxn *txn = job->txn;
408 BlockJob *other_job, *next;
410 * Successful completion, see if there are other running jobs in this
413 QLIST_FOREACH(other_job, &txn->jobs, txn_list) {
414 if (!other_job->completed) {
418 /* We are the last completed job, commit the transaction. */
419 QLIST_FOREACH_SAFE(other_job, &txn->jobs, txn_list, next) {
420 ctx = blk_get_aio_context(other_job->blk);
421 aio_context_acquire(ctx);
422 assert(other_job->ret == 0);
423 block_job_completed_single(other_job);
424 aio_context_release(ctx);
428 void block_job_completed(BlockJob *job, int ret)
430 assert(blk_bs(job->blk)->job == job);
431 assert(!job->completed);
432 job->completed = true;
435 block_job_completed_single(job);
436 } else if (ret < 0 || block_job_is_cancelled(job)) {
437 block_job_completed_txn_abort(job);
439 block_job_completed_txn_success(job);
443 void block_job_set_speed(BlockJob *job, int64_t speed, Error **errp)
445 Error *local_err = NULL;
447 if (!job->driver->set_speed) {
448 error_setg(errp, QERR_UNSUPPORTED);
451 job->driver->set_speed(job, speed, &local_err);
453 error_propagate(errp, local_err);
460 void block_job_complete(BlockJob *job, Error **errp)
462 /* Should not be reachable via external interface for internal jobs */
464 if (job->pause_count || job->cancelled ||
465 !block_job_started(job) || !job->driver->complete) {
466 error_setg(errp, "The active block job '%s' cannot be completed",
471 job->driver->complete(job, errp);
474 void block_job_pause(BlockJob *job)
479 void block_job_user_pause(BlockJob *job)
481 job->user_paused = true;
482 block_job_pause(job);
485 static bool block_job_should_pause(BlockJob *job)
487 return job->pause_count > 0;
490 bool block_job_user_paused(BlockJob *job)
492 return job->user_paused;
495 void coroutine_fn block_job_pause_point(BlockJob *job)
497 assert(job && block_job_started(job));
499 if (!block_job_should_pause(job)) {
502 if (block_job_is_cancelled(job)) {
506 if (job->driver->pause) {
507 job->driver->pause(job);
510 if (block_job_should_pause(job) && !block_job_is_cancelled(job)) {
513 qemu_coroutine_yield(); /* wait for block_job_resume() */
518 if (job->driver->resume) {
519 job->driver->resume(job);
523 void block_job_resume(BlockJob *job)
525 assert(job->pause_count > 0);
527 if (job->pause_count) {
530 block_job_enter(job);
533 void block_job_user_resume(BlockJob *job)
535 if (job && job->user_paused && job->pause_count > 0) {
536 job->user_paused = false;
537 block_job_resume(job);
541 void block_job_enter(BlockJob *job)
543 if (job->co && !job->busy) {
544 bdrv_coroutine_enter(blk_bs(job->blk), job->co);
548 void block_job_cancel(BlockJob *job)
550 if (block_job_started(job)) {
551 job->cancelled = true;
552 block_job_iostatus_reset(job);
553 block_job_enter(job);
555 block_job_completed(job, -ECANCELED);
559 bool block_job_is_cancelled(BlockJob *job)
561 return job->cancelled;
564 void block_job_iostatus_reset(BlockJob *job)
566 job->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
569 static int block_job_finish_sync(BlockJob *job,
570 void (*finish)(BlockJob *, Error **errp),
573 Error *local_err = NULL;
576 assert(blk_bs(job->blk)->job == job);
580 finish(job, &local_err);
582 error_propagate(errp, local_err);
583 block_job_unref(job);
586 /* block_job_drain calls block_job_enter, and it should be enough to
587 * induce progress until the job completes or moves to the main thread.
589 while (!job->deferred_to_main_loop && !job->completed) {
590 block_job_drain(job);
592 while (!job->completed) {
593 aio_poll(qemu_get_aio_context(), true);
595 ret = (job->cancelled && job->ret == 0) ? -ECANCELED : job->ret;
596 block_job_unref(job);
600 /* A wrapper around block_job_cancel() taking an Error ** parameter so it may be
601 * used with block_job_finish_sync() without the need for (rather nasty)
602 * function pointer casts there. */
603 static void block_job_cancel_err(BlockJob *job, Error **errp)
605 block_job_cancel(job);
608 int block_job_cancel_sync(BlockJob *job)
610 return block_job_finish_sync(job, &block_job_cancel_err, NULL);
613 void block_job_cancel_sync_all(void)
616 AioContext *aio_context;
618 while ((job = QLIST_FIRST(&block_jobs))) {
619 aio_context = blk_get_aio_context(job->blk);
620 aio_context_acquire(aio_context);
621 block_job_cancel_sync(job);
622 aio_context_release(aio_context);
626 int block_job_complete_sync(BlockJob *job, Error **errp)
628 return block_job_finish_sync(job, &block_job_complete, errp);
631 void block_job_sleep_ns(BlockJob *job, QEMUClockType type, int64_t ns)
635 /* Check cancellation *before* setting busy = false, too! */
636 if (block_job_is_cancelled(job)) {
641 if (!block_job_should_pause(job)) {
642 co_aio_sleep_ns(blk_get_aio_context(job->blk), type, ns);
646 block_job_pause_point(job);
649 void block_job_yield(BlockJob *job)
653 /* Check cancellation *before* setting busy = false, too! */
654 if (block_job_is_cancelled(job)) {
659 if (!block_job_should_pause(job)) {
660 qemu_coroutine_yield();
664 block_job_pause_point(job);
667 BlockJobInfo *block_job_query(BlockJob *job, Error **errp)
671 if (block_job_is_internal(job)) {
672 error_setg(errp, "Cannot query QEMU internal jobs");
675 info = g_new0(BlockJobInfo, 1);
676 info->type = g_strdup(BlockJobType_lookup[job->driver->job_type]);
677 info->device = g_strdup(job->id);
678 info->len = job->len;
679 info->busy = job->busy;
680 info->paused = job->pause_count > 0;
681 info->offset = job->offset;
682 info->speed = job->speed;
683 info->io_status = job->iostatus;
684 info->ready = job->ready;
688 static void block_job_iostatus_set_err(BlockJob *job, int error)
690 if (job->iostatus == BLOCK_DEVICE_IO_STATUS_OK) {
691 job->iostatus = error == ENOSPC ? BLOCK_DEVICE_IO_STATUS_NOSPACE :
692 BLOCK_DEVICE_IO_STATUS_FAILED;
696 static void block_job_event_cancelled(BlockJob *job)
698 if (block_job_is_internal(job)) {
702 qapi_event_send_block_job_cancelled(job->driver->job_type,
710 static void block_job_event_completed(BlockJob *job, const char *msg)
712 if (block_job_is_internal(job)) {
716 qapi_event_send_block_job_completed(job->driver->job_type,
726 void block_job_event_ready(BlockJob *job)
730 if (block_job_is_internal(job)) {
734 qapi_event_send_block_job_ready(job->driver->job_type,
738 job->speed, &error_abort);
741 BlockErrorAction block_job_error_action(BlockJob *job, BlockdevOnError on_err,
742 int is_read, int error)
744 BlockErrorAction action;
747 case BLOCKDEV_ON_ERROR_ENOSPC:
748 case BLOCKDEV_ON_ERROR_AUTO:
749 action = (error == ENOSPC) ?
750 BLOCK_ERROR_ACTION_STOP : BLOCK_ERROR_ACTION_REPORT;
752 case BLOCKDEV_ON_ERROR_STOP:
753 action = BLOCK_ERROR_ACTION_STOP;
755 case BLOCKDEV_ON_ERROR_REPORT:
756 action = BLOCK_ERROR_ACTION_REPORT;
758 case BLOCKDEV_ON_ERROR_IGNORE:
759 action = BLOCK_ERROR_ACTION_IGNORE;
764 if (!block_job_is_internal(job)) {
765 qapi_event_send_block_job_error(job->id,
766 is_read ? IO_OPERATION_TYPE_READ :
767 IO_OPERATION_TYPE_WRITE,
768 action, &error_abort);
770 if (action == BLOCK_ERROR_ACTION_STOP) {
771 /* make the pause user visible, which will be resumed from QMP. */
772 block_job_user_pause(job);
773 block_job_iostatus_set_err(job, error);
780 AioContext *aio_context;
781 BlockJobDeferToMainLoopFn *fn;
783 } BlockJobDeferToMainLoopData;
785 static void block_job_defer_to_main_loop_bh(void *opaque)
787 BlockJobDeferToMainLoopData *data = opaque;
788 AioContext *aio_context;
790 /* Prevent race with block_job_defer_to_main_loop() */
791 aio_context_acquire(data->aio_context);
793 /* Fetch BDS AioContext again, in case it has changed */
794 aio_context = blk_get_aio_context(data->job->blk);
795 if (aio_context != data->aio_context) {
796 aio_context_acquire(aio_context);
799 data->job->deferred_to_main_loop = false;
800 data->fn(data->job, data->opaque);
802 if (aio_context != data->aio_context) {
803 aio_context_release(aio_context);
806 aio_context_release(data->aio_context);
811 void block_job_defer_to_main_loop(BlockJob *job,
812 BlockJobDeferToMainLoopFn *fn,
815 BlockJobDeferToMainLoopData *data = g_malloc(sizeof(*data));
817 data->aio_context = blk_get_aio_context(job->blk);
819 data->opaque = opaque;
820 job->deferred_to_main_loop = true;
822 aio_bh_schedule_oneshot(qemu_get_aio_context(),
823 block_job_defer_to_main_loop_bh, data);
826 BlockJobTxn *block_job_txn_new(void)
828 BlockJobTxn *txn = g_new0(BlockJobTxn, 1);
829 QLIST_INIT(&txn->jobs);
834 static void block_job_txn_ref(BlockJobTxn *txn)
839 void block_job_txn_unref(BlockJobTxn *txn)
841 if (txn && --txn->refcnt == 0) {
846 void block_job_txn_add_job(BlockJobTxn *txn, BlockJob *job)
855 QLIST_INSERT_HEAD(&txn->jobs, job, txn_list);
856 block_job_txn_ref(txn);