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 BlockJob
*block_job_next(BlockJob
*job
)
61 return QLIST_FIRST(&block_jobs
);
63 return QLIST_NEXT(job
, job_list
);
66 BlockJob
*block_job_get(const char *id
)
70 QLIST_FOREACH(job
, &block_jobs
, job_list
) {
71 if (job
->id
&& !strcmp(id
, job
->id
)) {
79 static void block_job_attached_aio_context(AioContext
*new_context
,
82 BlockJob
*job
= opaque
;
84 if (job
->driver
->attached_aio_context
) {
85 job
->driver
->attached_aio_context(job
, new_context
);
88 block_job_resume(job
);
91 static void block_job_drain(BlockJob
*job
)
93 /* If job is !job->busy this kicks it into the next pause point. */
97 if (job
->driver
->drain
) {
98 job
->driver
->drain(job
);
102 static void block_job_detach_aio_context(void *opaque
)
104 BlockJob
*job
= opaque
;
106 /* In case the job terminates during aio_poll()... */
109 block_job_pause(job
);
111 while (!job
->paused
&& !job
->completed
) {
112 block_job_drain(job
);
115 block_job_unref(job
);
118 void block_job_add_bdrv(BlockJob
*job
, BlockDriverState
*bs
)
120 job
->nodes
= g_slist_prepend(job
->nodes
, bs
);
122 bdrv_op_block_all(bs
, job
->blocker
);
125 void *block_job_create(const char *job_id
, const BlockJobDriver
*driver
,
126 BlockDriverState
*bs
, int64_t speed
, int flags
,
127 BlockCompletionFunc
*cb
, void *opaque
, Error
**errp
)
134 error_setg(errp
, QERR_DEVICE_IN_USE
, bdrv_get_device_name(bs
));
138 if (job_id
== NULL
&& !(flags
& BLOCK_JOB_INTERNAL
)) {
139 job_id
= bdrv_get_device_name(bs
);
141 error_setg(errp
, "An explicit job ID is required for this node");
147 if (flags
& BLOCK_JOB_INTERNAL
) {
148 error_setg(errp
, "Cannot specify job ID for internal block job");
152 if (!id_wellformed(job_id
)) {
153 error_setg(errp
, "Invalid job ID '%s'", job_id
);
157 if (block_job_get(job_id
)) {
158 error_setg(errp
, "Job ID '%s' already in use", job_id
);
163 /* FIXME Use real permissions */
164 blk
= blk_new(0, BLK_PERM_ALL
);
165 ret
= blk_insert_bs(blk
, bs
, errp
);
171 job
= g_malloc0(driver
->instance_size
);
172 error_setg(&job
->blocker
, "block device is in use by block job: %s",
173 BlockJobType_lookup
[driver
->job_type
]);
174 block_job_add_bdrv(job
, bs
);
175 bdrv_op_unblock(bs
, BLOCK_OP_TYPE_DATAPLANE
, job
->blocker
);
177 job
->driver
= driver
;
178 job
->id
= g_strdup(job_id
);
181 job
->opaque
= opaque
;
184 job
->pause_count
= 1;
188 QLIST_INSERT_HEAD(&block_jobs
, job
, job_list
);
190 blk_add_aio_context_notifier(blk
, block_job_attached_aio_context
,
191 block_job_detach_aio_context
, job
);
193 /* Only set speed when necessary to avoid NotSupported error */
195 Error
*local_err
= NULL
;
197 block_job_set_speed(job
, speed
, &local_err
);
199 block_job_unref(job
);
200 error_propagate(errp
, local_err
);
207 bool block_job_is_internal(BlockJob
*job
)
209 return (job
->id
== NULL
);
212 static bool block_job_started(BlockJob
*job
)
217 void block_job_start(BlockJob
*job
)
219 assert(job
&& !block_job_started(job
) && job
->paused
&&
220 !job
->busy
&& job
->driver
->start
);
221 job
->co
= qemu_coroutine_create(job
->driver
->start
, job
);
222 if (--job
->pause_count
== 0) {
225 qemu_coroutine_enter(job
->co
);
229 void block_job_ref(BlockJob
*job
)
234 void block_job_unref(BlockJob
*job
)
236 if (--job
->refcnt
== 0) {
238 BlockDriverState
*bs
= blk_bs(job
->blk
);
240 for (l
= job
->nodes
; l
; l
= l
->next
) {
242 bdrv_op_unblock_all(bs
, job
->blocker
);
245 g_slist_free(job
->nodes
);
246 blk_remove_aio_context_notifier(job
->blk
,
247 block_job_attached_aio_context
,
248 block_job_detach_aio_context
, job
);
250 error_free(job
->blocker
);
252 QLIST_REMOVE(job
, job_list
);
257 static void block_job_completed_single(BlockJob
*job
)
260 if (job
->driver
->commit
) {
261 job
->driver
->commit(job
);
264 if (job
->driver
->abort
) {
265 job
->driver
->abort(job
);
268 if (job
->driver
->clean
) {
269 job
->driver
->clean(job
);
273 job
->cb(job
->opaque
, job
->ret
);
276 /* Emit events only if we actually started */
277 if (block_job_started(job
)) {
278 if (block_job_is_cancelled(job
)) {
279 block_job_event_cancelled(job
);
281 const char *msg
= NULL
;
283 msg
= strerror(-job
->ret
);
285 block_job_event_completed(job
, msg
);
290 QLIST_REMOVE(job
, txn_list
);
291 block_job_txn_unref(job
->txn
);
293 block_job_unref(job
);
296 static void block_job_completed_txn_abort(BlockJob
*job
)
299 BlockJobTxn
*txn
= job
->txn
;
300 BlockJob
*other_job
, *next
;
304 * We are cancelled by another job, which will handle everything.
308 txn
->aborting
= true;
309 /* We are the first failed job. Cancel other jobs. */
310 QLIST_FOREACH(other_job
, &txn
->jobs
, txn_list
) {
311 ctx
= blk_get_aio_context(other_job
->blk
);
312 aio_context_acquire(ctx
);
314 QLIST_FOREACH(other_job
, &txn
->jobs
, txn_list
) {
315 if (other_job
== job
|| other_job
->completed
) {
316 /* Other jobs are "effectively" cancelled by us, set the status for
317 * them; this job, however, may or may not be cancelled, depending
318 * on the caller, so leave it. */
319 if (other_job
!= job
) {
320 other_job
->cancelled
= true;
324 block_job_cancel_sync(other_job
);
325 assert(other_job
->completed
);
327 QLIST_FOREACH_SAFE(other_job
, &txn
->jobs
, txn_list
, next
) {
328 ctx
= blk_get_aio_context(other_job
->blk
);
329 block_job_completed_single(other_job
);
330 aio_context_release(ctx
);
334 static void block_job_completed_txn_success(BlockJob
*job
)
337 BlockJobTxn
*txn
= job
->txn
;
338 BlockJob
*other_job
, *next
;
340 * Successful completion, see if there are other running jobs in this
343 QLIST_FOREACH(other_job
, &txn
->jobs
, txn_list
) {
344 if (!other_job
->completed
) {
348 /* We are the last completed job, commit the transaction. */
349 QLIST_FOREACH_SAFE(other_job
, &txn
->jobs
, txn_list
, next
) {
350 ctx
= blk_get_aio_context(other_job
->blk
);
351 aio_context_acquire(ctx
);
352 assert(other_job
->ret
== 0);
353 block_job_completed_single(other_job
);
354 aio_context_release(ctx
);
358 void block_job_completed(BlockJob
*job
, int ret
)
360 assert(blk_bs(job
->blk
)->job
== job
);
361 assert(!job
->completed
);
362 job
->completed
= true;
365 block_job_completed_single(job
);
366 } else if (ret
< 0 || block_job_is_cancelled(job
)) {
367 block_job_completed_txn_abort(job
);
369 block_job_completed_txn_success(job
);
373 void block_job_set_speed(BlockJob
*job
, int64_t speed
, Error
**errp
)
375 Error
*local_err
= NULL
;
377 if (!job
->driver
->set_speed
) {
378 error_setg(errp
, QERR_UNSUPPORTED
);
381 job
->driver
->set_speed(job
, speed
, &local_err
);
383 error_propagate(errp
, local_err
);
390 void block_job_complete(BlockJob
*job
, Error
**errp
)
392 /* Should not be reachable via external interface for internal jobs */
394 if (job
->pause_count
|| job
->cancelled
||
395 !block_job_started(job
) || !job
->driver
->complete
) {
396 error_setg(errp
, "The active block job '%s' cannot be completed",
401 job
->driver
->complete(job
, errp
);
404 void block_job_pause(BlockJob
*job
)
409 void block_job_user_pause(BlockJob
*job
)
411 job
->user_paused
= true;
412 block_job_pause(job
);
415 static bool block_job_should_pause(BlockJob
*job
)
417 return job
->pause_count
> 0;
420 bool block_job_user_paused(BlockJob
*job
)
422 return job
? job
->user_paused
: 0;
425 void coroutine_fn
block_job_pause_point(BlockJob
*job
)
427 assert(job
&& block_job_started(job
));
429 if (!block_job_should_pause(job
)) {
432 if (block_job_is_cancelled(job
)) {
436 if (job
->driver
->pause
) {
437 job
->driver
->pause(job
);
440 if (block_job_should_pause(job
) && !block_job_is_cancelled(job
)) {
443 qemu_coroutine_yield(); /* wait for block_job_resume() */
448 if (job
->driver
->resume
) {
449 job
->driver
->resume(job
);
453 void block_job_resume(BlockJob
*job
)
455 assert(job
->pause_count
> 0);
457 if (job
->pause_count
) {
460 block_job_enter(job
);
463 void block_job_user_resume(BlockJob
*job
)
465 if (job
&& job
->user_paused
&& job
->pause_count
> 0) {
466 job
->user_paused
= false;
467 block_job_resume(job
);
471 void block_job_enter(BlockJob
*job
)
473 if (job
->co
&& !job
->busy
) {
474 qemu_coroutine_enter(job
->co
);
478 void block_job_cancel(BlockJob
*job
)
480 if (block_job_started(job
)) {
481 job
->cancelled
= true;
482 block_job_iostatus_reset(job
);
483 block_job_enter(job
);
485 block_job_completed(job
, -ECANCELED
);
489 bool block_job_is_cancelled(BlockJob
*job
)
491 return job
->cancelled
;
494 void block_job_iostatus_reset(BlockJob
*job
)
496 job
->iostatus
= BLOCK_DEVICE_IO_STATUS_OK
;
497 if (job
->driver
->iostatus_reset
) {
498 job
->driver
->iostatus_reset(job
);
502 static int block_job_finish_sync(BlockJob
*job
,
503 void (*finish
)(BlockJob
*, Error
**errp
),
506 Error
*local_err
= NULL
;
509 assert(blk_bs(job
->blk
)->job
== job
);
513 finish(job
, &local_err
);
515 error_propagate(errp
, local_err
);
516 block_job_unref(job
);
519 /* block_job_drain calls block_job_enter, and it should be enough to
520 * induce progress until the job completes or moves to the main thread.
522 while (!job
->deferred_to_main_loop
&& !job
->completed
) {
523 block_job_drain(job
);
525 while (!job
->completed
) {
526 aio_poll(qemu_get_aio_context(), true);
528 ret
= (job
->cancelled
&& job
->ret
== 0) ? -ECANCELED
: job
->ret
;
529 block_job_unref(job
);
533 /* A wrapper around block_job_cancel() taking an Error ** parameter so it may be
534 * used with block_job_finish_sync() without the need for (rather nasty)
535 * function pointer casts there. */
536 static void block_job_cancel_err(BlockJob
*job
, Error
**errp
)
538 block_job_cancel(job
);
541 int block_job_cancel_sync(BlockJob
*job
)
543 return block_job_finish_sync(job
, &block_job_cancel_err
, NULL
);
546 void block_job_cancel_sync_all(void)
549 AioContext
*aio_context
;
551 while ((job
= QLIST_FIRST(&block_jobs
))) {
552 aio_context
= blk_get_aio_context(job
->blk
);
553 aio_context_acquire(aio_context
);
554 block_job_cancel_sync(job
);
555 aio_context_release(aio_context
);
559 int block_job_complete_sync(BlockJob
*job
, Error
**errp
)
561 return block_job_finish_sync(job
, &block_job_complete
, errp
);
564 void block_job_sleep_ns(BlockJob
*job
, QEMUClockType type
, int64_t ns
)
568 /* Check cancellation *before* setting busy = false, too! */
569 if (block_job_is_cancelled(job
)) {
574 if (!block_job_should_pause(job
)) {
575 co_aio_sleep_ns(blk_get_aio_context(job
->blk
), type
, ns
);
579 block_job_pause_point(job
);
582 void block_job_yield(BlockJob
*job
)
586 /* Check cancellation *before* setting busy = false, too! */
587 if (block_job_is_cancelled(job
)) {
592 if (!block_job_should_pause(job
)) {
593 qemu_coroutine_yield();
597 block_job_pause_point(job
);
600 BlockJobInfo
*block_job_query(BlockJob
*job
, Error
**errp
)
604 if (block_job_is_internal(job
)) {
605 error_setg(errp
, "Cannot query QEMU internal jobs");
608 info
= g_new0(BlockJobInfo
, 1);
609 info
->type
= g_strdup(BlockJobType_lookup
[job
->driver
->job_type
]);
610 info
->device
= g_strdup(job
->id
);
611 info
->len
= job
->len
;
612 info
->busy
= job
->busy
;
613 info
->paused
= job
->pause_count
> 0;
614 info
->offset
= job
->offset
;
615 info
->speed
= job
->speed
;
616 info
->io_status
= job
->iostatus
;
617 info
->ready
= job
->ready
;
621 static void block_job_iostatus_set_err(BlockJob
*job
, int error
)
623 if (job
->iostatus
== BLOCK_DEVICE_IO_STATUS_OK
) {
624 job
->iostatus
= error
== ENOSPC
? BLOCK_DEVICE_IO_STATUS_NOSPACE
:
625 BLOCK_DEVICE_IO_STATUS_FAILED
;
629 static void block_job_event_cancelled(BlockJob
*job
)
631 if (block_job_is_internal(job
)) {
635 qapi_event_send_block_job_cancelled(job
->driver
->job_type
,
643 static void block_job_event_completed(BlockJob
*job
, const char *msg
)
645 if (block_job_is_internal(job
)) {
649 qapi_event_send_block_job_completed(job
->driver
->job_type
,
659 void block_job_event_ready(BlockJob
*job
)
663 if (block_job_is_internal(job
)) {
667 qapi_event_send_block_job_ready(job
->driver
->job_type
,
671 job
->speed
, &error_abort
);
674 BlockErrorAction
block_job_error_action(BlockJob
*job
, BlockdevOnError on_err
,
675 int is_read
, int error
)
677 BlockErrorAction action
;
680 case BLOCKDEV_ON_ERROR_ENOSPC
:
681 case BLOCKDEV_ON_ERROR_AUTO
:
682 action
= (error
== ENOSPC
) ?
683 BLOCK_ERROR_ACTION_STOP
: BLOCK_ERROR_ACTION_REPORT
;
685 case BLOCKDEV_ON_ERROR_STOP
:
686 action
= BLOCK_ERROR_ACTION_STOP
;
688 case BLOCKDEV_ON_ERROR_REPORT
:
689 action
= BLOCK_ERROR_ACTION_REPORT
;
691 case BLOCKDEV_ON_ERROR_IGNORE
:
692 action
= BLOCK_ERROR_ACTION_IGNORE
;
697 if (!block_job_is_internal(job
)) {
698 qapi_event_send_block_job_error(job
->id
,
699 is_read
? IO_OPERATION_TYPE_READ
:
700 IO_OPERATION_TYPE_WRITE
,
701 action
, &error_abort
);
703 if (action
== BLOCK_ERROR_ACTION_STOP
) {
704 /* make the pause user visible, which will be resumed from QMP. */
705 block_job_user_pause(job
);
706 block_job_iostatus_set_err(job
, error
);
713 AioContext
*aio_context
;
714 BlockJobDeferToMainLoopFn
*fn
;
716 } BlockJobDeferToMainLoopData
;
718 static void block_job_defer_to_main_loop_bh(void *opaque
)
720 BlockJobDeferToMainLoopData
*data
= opaque
;
721 AioContext
*aio_context
;
723 /* Prevent race with block_job_defer_to_main_loop() */
724 aio_context_acquire(data
->aio_context
);
726 /* Fetch BDS AioContext again, in case it has changed */
727 aio_context
= blk_get_aio_context(data
->job
->blk
);
728 aio_context_acquire(aio_context
);
730 data
->job
->deferred_to_main_loop
= false;
731 data
->fn(data
->job
, data
->opaque
);
733 aio_context_release(aio_context
);
735 aio_context_release(data
->aio_context
);
740 void block_job_defer_to_main_loop(BlockJob
*job
,
741 BlockJobDeferToMainLoopFn
*fn
,
744 BlockJobDeferToMainLoopData
*data
= g_malloc(sizeof(*data
));
746 data
->aio_context
= blk_get_aio_context(job
->blk
);
748 data
->opaque
= opaque
;
749 job
->deferred_to_main_loop
= true;
751 aio_bh_schedule_oneshot(qemu_get_aio_context(),
752 block_job_defer_to_main_loop_bh
, data
);
755 BlockJobTxn
*block_job_txn_new(void)
757 BlockJobTxn
*txn
= g_new0(BlockJobTxn
, 1);
758 QLIST_INIT(&txn
->jobs
);
763 static void block_job_txn_ref(BlockJobTxn
*txn
)
768 void block_job_txn_unref(BlockJobTxn
*txn
)
770 if (txn
&& --txn
->refcnt
== 0) {
775 void block_job_txn_add_job(BlockJobTxn
*txn
, BlockJob
*job
)
784 QLIST_INSERT_HEAD(&txn
->jobs
, job
, txn_list
);
785 block_job_txn_ref(txn
);