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
);
59 * The block job API is composed of two categories of functions.
61 * The first includes functions used by the monitor. The monitor is
62 * peculiar in that it accesses the block job list with block_job_get, and
63 * therefore needs consistency across block_job_get and the actual operation
64 * (e.g. block_job_set_speed). The consistency is achieved with
65 * aio_context_acquire/release. These functions are declared in blockjob.h.
67 * The second includes functions used by the block job drivers and sometimes
68 * by the core block layer. These do not care about locking, because the
69 * whole coroutine runs under the AioContext lock, and are declared in
73 BlockJob
*block_job_next(BlockJob
*job
)
76 return QLIST_FIRST(&block_jobs
);
78 return QLIST_NEXT(job
, job_list
);
81 BlockJob
*block_job_get(const char *id
)
85 QLIST_FOREACH(job
, &block_jobs
, job_list
) {
86 if (job
->id
&& !strcmp(id
, job
->id
)) {
94 static void block_job_pause(BlockJob
*job
)
99 static void block_job_resume(BlockJob
*job
)
101 assert(job
->pause_count
> 0);
103 if (job
->pause_count
) {
106 block_job_enter(job
);
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 static char *child_job_get_parent_desc(BdrvChild
*c
)
176 BlockJob
*job
= c
->opaque
;
177 return g_strdup_printf("%s job '%s'",
178 BlockJobType_lookup
[job
->driver
->job_type
],
182 static const BdrvChildRole child_job
= {
183 .get_parent_desc
= child_job_get_parent_desc
,
184 .stay_at_node
= true,
187 static void block_job_drained_begin(void *opaque
)
189 BlockJob
*job
= opaque
;
190 block_job_pause(job
);
193 static void block_job_drained_end(void *opaque
)
195 BlockJob
*job
= opaque
;
196 block_job_resume(job
);
199 static const BlockDevOps block_job_dev_ops
= {
200 .drained_begin
= block_job_drained_begin
,
201 .drained_end
= block_job_drained_end
,
204 void block_job_remove_all_bdrv(BlockJob
*job
)
207 for (l
= job
->nodes
; l
; l
= l
->next
) {
208 BdrvChild
*c
= l
->data
;
209 bdrv_op_unblock_all(c
->bs
, job
->blocker
);
210 bdrv_root_unref_child(c
);
212 g_slist_free(job
->nodes
);
216 int block_job_add_bdrv(BlockJob
*job
, const char *name
, BlockDriverState
*bs
,
217 uint64_t perm
, uint64_t shared_perm
, Error
**errp
)
221 c
= bdrv_root_attach_child(bs
, name
, &child_job
, perm
, shared_perm
,
227 job
->nodes
= g_slist_prepend(job
->nodes
, c
);
229 bdrv_op_block_all(bs
, job
->blocker
);
234 bool block_job_is_internal(BlockJob
*job
)
236 return (job
->id
== NULL
);
239 static bool block_job_started(BlockJob
*job
)
245 * All jobs must allow a pause point before entering their job proper. This
246 * ensures that jobs can be paused prior to being started, then resumed later.
248 static void coroutine_fn
block_job_co_entry(void *opaque
)
250 BlockJob
*job
= opaque
;
252 assert(job
&& job
->driver
&& job
->driver
->start
);
253 block_job_pause_point(job
);
254 job
->driver
->start(job
);
257 void block_job_start(BlockJob
*job
)
259 assert(job
&& !block_job_started(job
) && job
->paused
&&
260 job
->driver
&& job
->driver
->start
);
261 job
->co
= qemu_coroutine_create(block_job_co_entry
, job
);
265 bdrv_coroutine_enter(blk_bs(job
->blk
), job
->co
);
268 static void block_job_completed_single(BlockJob
*job
)
271 if (job
->driver
->commit
) {
272 job
->driver
->commit(job
);
275 if (job
->driver
->abort
) {
276 job
->driver
->abort(job
);
279 if (job
->driver
->clean
) {
280 job
->driver
->clean(job
);
284 job
->cb(job
->opaque
, job
->ret
);
287 /* Emit events only if we actually started */
288 if (block_job_started(job
)) {
289 if (block_job_is_cancelled(job
)) {
290 block_job_event_cancelled(job
);
292 const char *msg
= NULL
;
294 msg
= strerror(-job
->ret
);
296 block_job_event_completed(job
, msg
);
301 QLIST_REMOVE(job
, txn_list
);
302 block_job_txn_unref(job
->txn
);
304 block_job_unref(job
);
307 static void block_job_completed_txn_abort(BlockJob
*job
)
310 BlockJobTxn
*txn
= job
->txn
;
311 BlockJob
*other_job
, *next
;
315 * We are cancelled by another job, which will handle everything.
319 txn
->aborting
= true;
320 /* We are the first failed job. Cancel other jobs. */
321 QLIST_FOREACH(other_job
, &txn
->jobs
, txn_list
) {
322 ctx
= blk_get_aio_context(other_job
->blk
);
323 aio_context_acquire(ctx
);
325 QLIST_FOREACH(other_job
, &txn
->jobs
, txn_list
) {
326 if (other_job
== job
|| other_job
->completed
) {
327 /* Other jobs are "effectively" cancelled by us, set the status for
328 * them; this job, however, may or may not be cancelled, depending
329 * on the caller, so leave it. */
330 if (other_job
!= job
) {
331 other_job
->cancelled
= true;
335 block_job_cancel_sync(other_job
);
336 assert(other_job
->completed
);
338 QLIST_FOREACH_SAFE(other_job
, &txn
->jobs
, txn_list
, next
) {
339 ctx
= blk_get_aio_context(other_job
->blk
);
340 block_job_completed_single(other_job
);
341 aio_context_release(ctx
);
345 static void block_job_completed_txn_success(BlockJob
*job
)
348 BlockJobTxn
*txn
= job
->txn
;
349 BlockJob
*other_job
, *next
;
351 * Successful completion, see if there are other running jobs in this
354 QLIST_FOREACH(other_job
, &txn
->jobs
, txn_list
) {
355 if (!other_job
->completed
) {
359 /* We are the last completed job, commit the transaction. */
360 QLIST_FOREACH_SAFE(other_job
, &txn
->jobs
, txn_list
, next
) {
361 ctx
= blk_get_aio_context(other_job
->blk
);
362 aio_context_acquire(ctx
);
363 assert(other_job
->ret
== 0);
364 block_job_completed_single(other_job
);
365 aio_context_release(ctx
);
369 void block_job_set_speed(BlockJob
*job
, int64_t speed
, Error
**errp
)
371 Error
*local_err
= NULL
;
373 if (!job
->driver
->set_speed
) {
374 error_setg(errp
, QERR_UNSUPPORTED
);
377 job
->driver
->set_speed(job
, speed
, &local_err
);
379 error_propagate(errp
, local_err
);
386 void block_job_complete(BlockJob
*job
, Error
**errp
)
388 /* Should not be reachable via external interface for internal jobs */
390 if (job
->pause_count
|| job
->cancelled
||
391 !block_job_started(job
) || !job
->driver
->complete
) {
392 error_setg(errp
, "The active block job '%s' cannot be completed",
397 job
->driver
->complete(job
, errp
);
400 void block_job_user_pause(BlockJob
*job
)
402 job
->user_paused
= true;
403 block_job_pause(job
);
406 bool block_job_user_paused(BlockJob
*job
)
408 return job
->user_paused
;
411 void block_job_user_resume(BlockJob
*job
)
413 if (job
&& job
->user_paused
&& job
->pause_count
> 0) {
414 job
->user_paused
= false;
415 block_job_resume(job
);
419 void block_job_cancel(BlockJob
*job
)
421 if (block_job_started(job
)) {
422 job
->cancelled
= true;
423 block_job_iostatus_reset(job
);
424 block_job_enter(job
);
426 block_job_completed(job
, -ECANCELED
);
430 void block_job_iostatus_reset(BlockJob
*job
)
432 job
->iostatus
= BLOCK_DEVICE_IO_STATUS_OK
;
435 static int block_job_finish_sync(BlockJob
*job
,
436 void (*finish
)(BlockJob
*, Error
**errp
),
439 Error
*local_err
= NULL
;
442 assert(blk_bs(job
->blk
)->job
== job
);
446 finish(job
, &local_err
);
448 error_propagate(errp
, local_err
);
449 block_job_unref(job
);
452 /* block_job_drain calls block_job_enter, and it should be enough to
453 * induce progress until the job completes or moves to the main thread.
455 while (!job
->deferred_to_main_loop
&& !job
->completed
) {
456 block_job_drain(job
);
458 while (!job
->completed
) {
459 aio_poll(qemu_get_aio_context(), true);
461 ret
= (job
->cancelled
&& job
->ret
== 0) ? -ECANCELED
: job
->ret
;
462 block_job_unref(job
);
466 /* A wrapper around block_job_cancel() taking an Error ** parameter so it may be
467 * used with block_job_finish_sync() without the need for (rather nasty)
468 * function pointer casts there. */
469 static void block_job_cancel_err(BlockJob
*job
, Error
**errp
)
471 block_job_cancel(job
);
474 int block_job_cancel_sync(BlockJob
*job
)
476 return block_job_finish_sync(job
, &block_job_cancel_err
, NULL
);
479 void block_job_cancel_sync_all(void)
482 AioContext
*aio_context
;
484 while ((job
= QLIST_FIRST(&block_jobs
))) {
485 aio_context
= blk_get_aio_context(job
->blk
);
486 aio_context_acquire(aio_context
);
487 block_job_cancel_sync(job
);
488 aio_context_release(aio_context
);
492 int block_job_complete_sync(BlockJob
*job
, Error
**errp
)
494 return block_job_finish_sync(job
, &block_job_complete
, errp
);
497 BlockJobInfo
*block_job_query(BlockJob
*job
, Error
**errp
)
501 if (block_job_is_internal(job
)) {
502 error_setg(errp
, "Cannot query QEMU internal jobs");
505 info
= g_new0(BlockJobInfo
, 1);
506 info
->type
= g_strdup(BlockJobType_lookup
[job
->driver
->job_type
]);
507 info
->device
= g_strdup(job
->id
);
508 info
->len
= job
->len
;
509 info
->busy
= job
->busy
;
510 info
->paused
= job
->pause_count
> 0;
511 info
->offset
= job
->offset
;
512 info
->speed
= job
->speed
;
513 info
->io_status
= job
->iostatus
;
514 info
->ready
= job
->ready
;
518 static void block_job_iostatus_set_err(BlockJob
*job
, int error
)
520 if (job
->iostatus
== BLOCK_DEVICE_IO_STATUS_OK
) {
521 job
->iostatus
= error
== ENOSPC
? BLOCK_DEVICE_IO_STATUS_NOSPACE
:
522 BLOCK_DEVICE_IO_STATUS_FAILED
;
526 static void block_job_event_cancelled(BlockJob
*job
)
528 if (block_job_is_internal(job
)) {
532 qapi_event_send_block_job_cancelled(job
->driver
->job_type
,
540 static void block_job_event_completed(BlockJob
*job
, const char *msg
)
542 if (block_job_is_internal(job
)) {
546 qapi_event_send_block_job_completed(job
->driver
->job_type
,
557 * API for block job drivers and the block layer. These functions are
558 * declared in blockjob_int.h.
561 void *block_job_create(const char *job_id
, const BlockJobDriver
*driver
,
562 BlockDriverState
*bs
, uint64_t perm
,
563 uint64_t shared_perm
, int64_t speed
, int flags
,
564 BlockCompletionFunc
*cb
, void *opaque
, Error
**errp
)
571 error_setg(errp
, QERR_DEVICE_IN_USE
, bdrv_get_device_name(bs
));
575 if (job_id
== NULL
&& !(flags
& BLOCK_JOB_INTERNAL
)) {
576 job_id
= bdrv_get_device_name(bs
);
578 error_setg(errp
, "An explicit job ID is required for this node");
584 if (flags
& BLOCK_JOB_INTERNAL
) {
585 error_setg(errp
, "Cannot specify job ID for internal block job");
589 if (!id_wellformed(job_id
)) {
590 error_setg(errp
, "Invalid job ID '%s'", job_id
);
594 if (block_job_get(job_id
)) {
595 error_setg(errp
, "Job ID '%s' already in use", job_id
);
600 blk
= blk_new(perm
, shared_perm
);
601 ret
= blk_insert_bs(blk
, bs
, errp
);
607 job
= g_malloc0(driver
->instance_size
);
608 job
->driver
= driver
;
609 job
->id
= g_strdup(job_id
);
612 job
->opaque
= opaque
;
615 job
->pause_count
= 1;
618 error_setg(&job
->blocker
, "block device is in use by block job: %s",
619 BlockJobType_lookup
[driver
->job_type
]);
620 block_job_add_bdrv(job
, "main node", bs
, 0, BLK_PERM_ALL
, &error_abort
);
623 blk_set_dev_ops(blk
, &block_job_dev_ops
, job
);
624 bdrv_op_unblock(bs
, BLOCK_OP_TYPE_DATAPLANE
, job
->blocker
);
626 QLIST_INSERT_HEAD(&block_jobs
, job
, job_list
);
628 blk_add_aio_context_notifier(blk
, block_job_attached_aio_context
,
629 block_job_detach_aio_context
, job
);
631 /* Only set speed when necessary to avoid NotSupported error */
633 Error
*local_err
= NULL
;
635 block_job_set_speed(job
, speed
, &local_err
);
637 block_job_unref(job
);
638 error_propagate(errp
, local_err
);
645 void block_job_pause_all(void)
647 BlockJob
*job
= NULL
;
648 while ((job
= block_job_next(job
))) {
649 AioContext
*aio_context
= blk_get_aio_context(job
->blk
);
651 aio_context_acquire(aio_context
);
652 block_job_pause(job
);
653 aio_context_release(aio_context
);
657 void block_job_early_fail(BlockJob
*job
)
659 block_job_unref(job
);
662 void block_job_completed(BlockJob
*job
, int ret
)
664 assert(blk_bs(job
->blk
)->job
== job
);
665 assert(!job
->completed
);
666 job
->completed
= true;
669 block_job_completed_single(job
);
670 } else if (ret
< 0 || block_job_is_cancelled(job
)) {
671 block_job_completed_txn_abort(job
);
673 block_job_completed_txn_success(job
);
677 static bool block_job_should_pause(BlockJob
*job
)
679 return job
->pause_count
> 0;
682 void coroutine_fn
block_job_pause_point(BlockJob
*job
)
684 assert(job
&& block_job_started(job
));
686 if (!block_job_should_pause(job
)) {
689 if (block_job_is_cancelled(job
)) {
693 if (job
->driver
->pause
) {
694 job
->driver
->pause(job
);
697 if (block_job_should_pause(job
) && !block_job_is_cancelled(job
)) {
700 qemu_coroutine_yield(); /* wait for block_job_resume() */
705 if (job
->driver
->resume
) {
706 job
->driver
->resume(job
);
710 void block_job_resume_all(void)
712 BlockJob
*job
= NULL
;
713 while ((job
= block_job_next(job
))) {
714 AioContext
*aio_context
= blk_get_aio_context(job
->blk
);
716 aio_context_acquire(aio_context
);
717 block_job_resume(job
);
718 aio_context_release(aio_context
);
722 void block_job_enter(BlockJob
*job
)
724 if (job
->co
&& !job
->busy
) {
725 bdrv_coroutine_enter(blk_bs(job
->blk
), job
->co
);
729 bool block_job_is_cancelled(BlockJob
*job
)
731 return job
->cancelled
;
734 void block_job_sleep_ns(BlockJob
*job
, QEMUClockType type
, int64_t ns
)
738 /* Check cancellation *before* setting busy = false, too! */
739 if (block_job_is_cancelled(job
)) {
744 if (!block_job_should_pause(job
)) {
745 co_aio_sleep_ns(blk_get_aio_context(job
->blk
), type
, ns
);
749 block_job_pause_point(job
);
752 void block_job_yield(BlockJob
*job
)
756 /* Check cancellation *before* setting busy = false, too! */
757 if (block_job_is_cancelled(job
)) {
762 if (!block_job_should_pause(job
)) {
763 qemu_coroutine_yield();
767 block_job_pause_point(job
);
770 void block_job_event_ready(BlockJob
*job
)
774 if (block_job_is_internal(job
)) {
778 qapi_event_send_block_job_ready(job
->driver
->job_type
,
782 job
->speed
, &error_abort
);
785 BlockErrorAction
block_job_error_action(BlockJob
*job
, BlockdevOnError on_err
,
786 int is_read
, int error
)
788 BlockErrorAction action
;
791 case BLOCKDEV_ON_ERROR_ENOSPC
:
792 case BLOCKDEV_ON_ERROR_AUTO
:
793 action
= (error
== ENOSPC
) ?
794 BLOCK_ERROR_ACTION_STOP
: BLOCK_ERROR_ACTION_REPORT
;
796 case BLOCKDEV_ON_ERROR_STOP
:
797 action
= BLOCK_ERROR_ACTION_STOP
;
799 case BLOCKDEV_ON_ERROR_REPORT
:
800 action
= BLOCK_ERROR_ACTION_REPORT
;
802 case BLOCKDEV_ON_ERROR_IGNORE
:
803 action
= BLOCK_ERROR_ACTION_IGNORE
;
808 if (!block_job_is_internal(job
)) {
809 qapi_event_send_block_job_error(job
->id
,
810 is_read
? IO_OPERATION_TYPE_READ
:
811 IO_OPERATION_TYPE_WRITE
,
812 action
, &error_abort
);
814 if (action
== BLOCK_ERROR_ACTION_STOP
) {
815 /* make the pause user visible, which will be resumed from QMP. */
816 block_job_user_pause(job
);
817 block_job_iostatus_set_err(job
, error
);
824 AioContext
*aio_context
;
825 BlockJobDeferToMainLoopFn
*fn
;
827 } BlockJobDeferToMainLoopData
;
829 static void block_job_defer_to_main_loop_bh(void *opaque
)
831 BlockJobDeferToMainLoopData
*data
= opaque
;
832 AioContext
*aio_context
;
834 /* Prevent race with block_job_defer_to_main_loop() */
835 aio_context_acquire(data
->aio_context
);
837 /* Fetch BDS AioContext again, in case it has changed */
838 aio_context
= blk_get_aio_context(data
->job
->blk
);
839 if (aio_context
!= data
->aio_context
) {
840 aio_context_acquire(aio_context
);
843 data
->job
->deferred_to_main_loop
= false;
844 data
->fn(data
->job
, data
->opaque
);
846 if (aio_context
!= data
->aio_context
) {
847 aio_context_release(aio_context
);
850 aio_context_release(data
->aio_context
);
855 void block_job_defer_to_main_loop(BlockJob
*job
,
856 BlockJobDeferToMainLoopFn
*fn
,
859 BlockJobDeferToMainLoopData
*data
= g_malloc(sizeof(*data
));
861 data
->aio_context
= blk_get_aio_context(job
->blk
);
863 data
->opaque
= opaque
;
864 job
->deferred_to_main_loop
= true;
866 aio_bh_schedule_oneshot(qemu_get_aio_context(),
867 block_job_defer_to_main_loop_bh
, data
);
870 BlockJobTxn
*block_job_txn_new(void)
872 BlockJobTxn
*txn
= g_new0(BlockJobTxn
, 1);
873 QLIST_INIT(&txn
->jobs
);
878 static void block_job_txn_ref(BlockJobTxn
*txn
)
883 void block_job_txn_unref(BlockJobTxn
*txn
)
885 if (txn
&& --txn
->refcnt
== 0) {
890 void block_job_txn_add_job(BlockJobTxn
*txn
, BlockJob
*job
)
899 QLIST_INSERT_HEAD(&txn
->jobs
, job
, txn_list
);
900 block_job_txn_ref(txn
);