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 /* Right now, this mutex is only needed to synchronize accesses to job->busy
40 * and job->sleep_timer, such as concurrent calls to block_job_do_yield and
42 static QemuMutex block_job_mutex
;
44 /* BlockJob State Transition Table */
45 bool BlockJobSTT
[BLOCK_JOB_STATUS__MAX
][BLOCK_JOB_STATUS__MAX
] = {
46 /* U, C, R, P, Y, S, W, D, X, E, N */
47 /* U: */ [BLOCK_JOB_STATUS_UNDEFINED
] = {0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0},
48 /* C: */ [BLOCK_JOB_STATUS_CREATED
] = {0, 0, 1, 0, 0, 0, 0, 0, 1, 0, 1},
49 /* R: */ [BLOCK_JOB_STATUS_RUNNING
] = {0, 0, 0, 1, 1, 0, 1, 0, 1, 0, 0},
50 /* P: */ [BLOCK_JOB_STATUS_PAUSED
] = {0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0},
51 /* Y: */ [BLOCK_JOB_STATUS_READY
] = {0, 0, 0, 0, 0, 1, 1, 0, 1, 0, 0},
52 /* S: */ [BLOCK_JOB_STATUS_STANDBY
] = {0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0},
53 /* W: */ [BLOCK_JOB_STATUS_WAITING
] = {0, 0, 0, 0, 0, 0, 0, 1, 1, 0, 0},
54 /* D: */ [BLOCK_JOB_STATUS_PENDING
] = {0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 0},
55 /* X: */ [BLOCK_JOB_STATUS_ABORTING
] = {0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 0},
56 /* E: */ [BLOCK_JOB_STATUS_CONCLUDED
] = {0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1},
57 /* N: */ [BLOCK_JOB_STATUS_NULL
] = {0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0},
60 bool BlockJobVerbTable
[BLOCK_JOB_VERB__MAX
][BLOCK_JOB_STATUS__MAX
] = {
61 /* U, C, R, P, Y, S, W, D, X, E, N */
62 [BLOCK_JOB_VERB_CANCEL
] = {0, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0},
63 [BLOCK_JOB_VERB_PAUSE
] = {0, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0},
64 [BLOCK_JOB_VERB_RESUME
] = {0, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0},
65 [BLOCK_JOB_VERB_SET_SPEED
] = {0, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0},
66 [BLOCK_JOB_VERB_COMPLETE
] = {0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0},
67 [BLOCK_JOB_VERB_FINALIZE
] = {0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0},
68 [BLOCK_JOB_VERB_DISMISS
] = {0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0},
71 static void block_job_state_transition(BlockJob
*job
, BlockJobStatus s1
)
73 BlockJobStatus s0
= job
->status
;
74 assert(s1
>= 0 && s1
<= BLOCK_JOB_STATUS__MAX
);
75 trace_block_job_state_transition(job
, job
->ret
, BlockJobSTT
[s0
][s1
] ?
76 "allowed" : "disallowed",
77 BlockJobStatus_str(s0
),
78 BlockJobStatus_str(s1
));
79 assert(BlockJobSTT
[s0
][s1
]);
83 static int block_job_apply_verb(BlockJob
*job
, BlockJobVerb bv
, Error
**errp
)
85 assert(bv
>= 0 && bv
<= BLOCK_JOB_VERB__MAX
);
86 trace_block_job_apply_verb(job
, BlockJobStatus_str(job
->status
),
88 BlockJobVerbTable
[bv
][job
->status
] ?
89 "allowed" : "prohibited");
90 if (BlockJobVerbTable
[bv
][job
->status
]) {
93 error_setg(errp
, "Job '%s' in state '%s' cannot accept command verb '%s'",
94 job
->job
.id
, BlockJobStatus_str(job
->status
),
95 BlockJobVerb_str(bv
));
99 static void block_job_lock(void)
101 qemu_mutex_lock(&block_job_mutex
);
104 static void block_job_unlock(void)
106 qemu_mutex_unlock(&block_job_mutex
);
109 static void __attribute__((__constructor__
)) block_job_init(void)
111 qemu_mutex_init(&block_job_mutex
);
114 static void block_job_event_cancelled(BlockJob
*job
);
115 static void block_job_event_completed(BlockJob
*job
, const char *msg
);
116 static int block_job_event_pending(BlockJob
*job
);
117 static void block_job_enter_cond(BlockJob
*job
, bool(*fn
)(BlockJob
*job
));
119 /* Transactional group of block jobs */
122 /* Is this txn being cancelled? */
126 QLIST_HEAD(, BlockJob
) jobs
;
128 /* Reference count */
133 * The block job API is composed of two categories of functions.
135 * The first includes functions used by the monitor. The monitor is
136 * peculiar in that it accesses the block job list with block_job_get, and
137 * therefore needs consistency across block_job_get and the actual operation
138 * (e.g. block_job_set_speed). The consistency is achieved with
139 * aio_context_acquire/release. These functions are declared in blockjob.h.
141 * The second includes functions used by the block job drivers and sometimes
142 * by the core block layer. These do not care about locking, because the
143 * whole coroutine runs under the AioContext lock, and are declared in
147 static bool is_block_job(Job
*job
)
149 return job_type(job
) == JOB_TYPE_BACKUP
||
150 job_type(job
) == JOB_TYPE_COMMIT
||
151 job_type(job
) == JOB_TYPE_MIRROR
||
152 job_type(job
) == JOB_TYPE_STREAM
;
155 BlockJob
*block_job_next(BlockJob
*bjob
)
157 Job
*job
= bjob
? &bjob
->job
: NULL
;
161 } while (job
&& !is_block_job(job
));
163 return job
? container_of(job
, BlockJob
, job
) : NULL
;
166 BlockJob
*block_job_get(const char *id
)
168 Job
*job
= job_get(id
);
170 if (job
&& is_block_job(job
)) {
171 return container_of(job
, BlockJob
, job
);
177 BlockJobTxn
*block_job_txn_new(void)
179 BlockJobTxn
*txn
= g_new0(BlockJobTxn
, 1);
180 QLIST_INIT(&txn
->jobs
);
185 static void block_job_txn_ref(BlockJobTxn
*txn
)
190 void block_job_txn_unref(BlockJobTxn
*txn
)
192 if (txn
&& --txn
->refcnt
== 0) {
197 void block_job_txn_add_job(BlockJobTxn
*txn
, BlockJob
*job
)
206 QLIST_INSERT_HEAD(&txn
->jobs
, job
, txn_list
);
207 block_job_txn_ref(txn
);
210 static void block_job_txn_del_job(BlockJob
*job
)
213 QLIST_REMOVE(job
, txn_list
);
214 block_job_txn_unref(job
->txn
);
219 /* Assumes the block_job_mutex is held */
220 static bool block_job_timer_pending(BlockJob
*job
)
222 return timer_pending(&job
->sleep_timer
);
225 /* Assumes the block_job_mutex is held */
226 static bool block_job_timer_not_pending(BlockJob
*job
)
228 return !block_job_timer_pending(job
);
231 static void block_job_pause(BlockJob
*job
)
236 static void block_job_resume(BlockJob
*job
)
238 assert(job
->pause_count
> 0);
240 if (job
->pause_count
) {
244 /* kick only if no timer is pending */
245 block_job_enter_cond(job
, block_job_timer_not_pending
);
248 void block_job_ref(BlockJob
*job
)
253 static void block_job_attached_aio_context(AioContext
*new_context
,
255 static void block_job_detach_aio_context(void *opaque
);
257 void block_job_unref(BlockJob
*job
)
259 if (--job
->refcnt
== 0) {
260 assert(job
->status
== BLOCK_JOB_STATUS_NULL
);
262 BlockDriverState
*bs
= blk_bs(job
->blk
);
264 block_job_remove_all_bdrv(job
);
265 blk_remove_aio_context_notifier(job
->blk
,
266 block_job_attached_aio_context
,
267 block_job_detach_aio_context
, job
);
269 error_free(job
->blocker
);
270 assert(!timer_pending(&job
->sleep_timer
));
271 job_delete(&job
->job
);
275 static void block_job_attached_aio_context(AioContext
*new_context
,
278 BlockJob
*job
= opaque
;
280 if (job
->driver
->attached_aio_context
) {
281 job
->driver
->attached_aio_context(job
, new_context
);
284 block_job_resume(job
);
287 static void block_job_drain(BlockJob
*job
)
289 /* If job is !job->busy this kicks it into the next pause point. */
290 block_job_enter(job
);
293 if (job
->driver
->drain
) {
294 job
->driver
->drain(job
);
298 static void block_job_detach_aio_context(void *opaque
)
300 BlockJob
*job
= opaque
;
302 /* In case the job terminates during aio_poll()... */
305 block_job_pause(job
);
307 while (!job
->paused
&& !job
->completed
) {
308 block_job_drain(job
);
311 block_job_unref(job
);
314 static char *child_job_get_parent_desc(BdrvChild
*c
)
316 BlockJob
*job
= c
->opaque
;
317 return g_strdup_printf("%s job '%s'", job_type_str(&job
->job
), job
->job
.id
);
320 static void child_job_drained_begin(BdrvChild
*c
)
322 BlockJob
*job
= c
->opaque
;
323 block_job_pause(job
);
326 static void child_job_drained_end(BdrvChild
*c
)
328 BlockJob
*job
= c
->opaque
;
329 block_job_resume(job
);
332 static const BdrvChildRole child_job
= {
333 .get_parent_desc
= child_job_get_parent_desc
,
334 .drained_begin
= child_job_drained_begin
,
335 .drained_end
= child_job_drained_end
,
336 .stay_at_node
= true,
339 void block_job_remove_all_bdrv(BlockJob
*job
)
342 for (l
= job
->nodes
; l
; l
= l
->next
) {
343 BdrvChild
*c
= l
->data
;
344 bdrv_op_unblock_all(c
->bs
, job
->blocker
);
345 bdrv_root_unref_child(c
);
347 g_slist_free(job
->nodes
);
351 int block_job_add_bdrv(BlockJob
*job
, const char *name
, BlockDriverState
*bs
,
352 uint64_t perm
, uint64_t shared_perm
, Error
**errp
)
356 c
= bdrv_root_attach_child(bs
, name
, &child_job
, perm
, shared_perm
,
362 job
->nodes
= g_slist_prepend(job
->nodes
, c
);
364 bdrv_op_block_all(bs
, job
->blocker
);
369 bool block_job_is_internal(BlockJob
*job
)
371 return (job
->job
.id
== NULL
);
374 static bool block_job_started(BlockJob
*job
)
379 const BlockJobDriver
*block_job_driver(BlockJob
*job
)
385 * All jobs must allow a pause point before entering their job proper. This
386 * ensures that jobs can be paused prior to being started, then resumed later.
388 static void coroutine_fn
block_job_co_entry(void *opaque
)
390 BlockJob
*job
= opaque
;
392 assert(job
&& job
->driver
&& job
->driver
->start
);
393 block_job_pause_point(job
);
394 job
->driver
->start(job
);
397 static void block_job_sleep_timer_cb(void *opaque
)
399 BlockJob
*job
= opaque
;
401 block_job_enter(job
);
404 void block_job_start(BlockJob
*job
)
406 assert(job
&& !block_job_started(job
) && job
->paused
&&
407 job
->driver
&& job
->driver
->start
);
408 job
->co
= qemu_coroutine_create(block_job_co_entry
, job
);
412 block_job_state_transition(job
, BLOCK_JOB_STATUS_RUNNING
);
413 bdrv_coroutine_enter(blk_bs(job
->blk
), job
->co
);
416 static void block_job_decommission(BlockJob
*job
)
419 job
->completed
= true;
422 job
->deferred_to_main_loop
= true;
423 block_job_txn_del_job(job
);
424 block_job_state_transition(job
, BLOCK_JOB_STATUS_NULL
);
425 block_job_unref(job
);
428 static void block_job_do_dismiss(BlockJob
*job
)
430 block_job_decommission(job
);
433 static void block_job_conclude(BlockJob
*job
)
435 block_job_state_transition(job
, BLOCK_JOB_STATUS_CONCLUDED
);
436 if (job
->auto_dismiss
|| !block_job_started(job
)) {
437 block_job_do_dismiss(job
);
441 static void block_job_update_rc(BlockJob
*job
)
443 if (!job
->ret
&& block_job_is_cancelled(job
)) {
444 job
->ret
= -ECANCELED
;
447 block_job_state_transition(job
, BLOCK_JOB_STATUS_ABORTING
);
451 static int block_job_prepare(BlockJob
*job
)
453 if (job
->ret
== 0 && job
->driver
->prepare
) {
454 job
->ret
= job
->driver
->prepare(job
);
459 static void block_job_commit(BlockJob
*job
)
462 if (job
->driver
->commit
) {
463 job
->driver
->commit(job
);
467 static void block_job_abort(BlockJob
*job
)
470 if (job
->driver
->abort
) {
471 job
->driver
->abort(job
);
475 static void block_job_clean(BlockJob
*job
)
477 if (job
->driver
->clean
) {
478 job
->driver
->clean(job
);
482 static int block_job_finalize_single(BlockJob
*job
)
484 assert(job
->completed
);
486 /* Ensure abort is called for late-transactional failures */
487 block_job_update_rc(job
);
490 block_job_commit(job
);
492 block_job_abort(job
);
494 block_job_clean(job
);
497 job
->cb(job
->opaque
, job
->ret
);
500 /* Emit events only if we actually started */
501 if (block_job_started(job
)) {
502 if (block_job_is_cancelled(job
)) {
503 block_job_event_cancelled(job
);
505 const char *msg
= NULL
;
507 msg
= strerror(-job
->ret
);
509 block_job_event_completed(job
, msg
);
513 block_job_txn_del_job(job
);
514 block_job_conclude(job
);
518 static void block_job_cancel_async(BlockJob
*job
, bool force
)
520 if (job
->iostatus
!= BLOCK_DEVICE_IO_STATUS_OK
) {
521 block_job_iostatus_reset(job
);
523 if (job
->user_paused
) {
524 /* Do not call block_job_enter here, the caller will handle it. */
525 job
->user_paused
= false;
528 job
->cancelled
= true;
529 /* To prevent 'force == false' overriding a previous 'force == true' */
533 static int block_job_txn_apply(BlockJobTxn
*txn
, int fn(BlockJob
*), bool lock
)
536 BlockJob
*job
, *next
;
539 QLIST_FOREACH_SAFE(job
, &txn
->jobs
, txn_list
, next
) {
541 ctx
= blk_get_aio_context(job
->blk
);
542 aio_context_acquire(ctx
);
546 aio_context_release(ctx
);
555 static int block_job_finish_sync(BlockJob
*job
,
556 void (*finish
)(BlockJob
*, Error
**errp
),
559 Error
*local_err
= NULL
;
562 assert(blk_bs(job
->blk
)->job
== job
);
567 finish(job
, &local_err
);
570 error_propagate(errp
, local_err
);
571 block_job_unref(job
);
574 /* block_job_drain calls block_job_enter, and it should be enough to
575 * induce progress until the job completes or moves to the main thread.
577 while (!job
->deferred_to_main_loop
&& !job
->completed
) {
578 block_job_drain(job
);
580 while (!job
->completed
) {
581 aio_poll(qemu_get_aio_context(), true);
583 ret
= (job
->cancelled
&& job
->ret
== 0) ? -ECANCELED
: job
->ret
;
584 block_job_unref(job
);
588 static void block_job_completed_txn_abort(BlockJob
*job
)
591 BlockJobTxn
*txn
= job
->txn
;
596 * We are cancelled by another job, which will handle everything.
600 txn
->aborting
= true;
601 block_job_txn_ref(txn
);
603 /* We are the first failed job. Cancel other jobs. */
604 QLIST_FOREACH(other_job
, &txn
->jobs
, txn_list
) {
605 ctx
= blk_get_aio_context(other_job
->blk
);
606 aio_context_acquire(ctx
);
609 /* Other jobs are effectively cancelled by us, set the status for
610 * them; this job, however, may or may not be cancelled, depending
611 * on the caller, so leave it. */
612 QLIST_FOREACH(other_job
, &txn
->jobs
, txn_list
) {
613 if (other_job
!= job
) {
614 block_job_cancel_async(other_job
, false);
617 while (!QLIST_EMPTY(&txn
->jobs
)) {
618 other_job
= QLIST_FIRST(&txn
->jobs
);
619 ctx
= blk_get_aio_context(other_job
->blk
);
620 if (!other_job
->completed
) {
621 assert(other_job
->cancelled
);
622 block_job_finish_sync(other_job
, NULL
, NULL
);
624 block_job_finalize_single(other_job
);
625 aio_context_release(ctx
);
628 block_job_txn_unref(txn
);
631 static int block_job_needs_finalize(BlockJob
*job
)
633 return !job
->auto_finalize
;
636 static void block_job_do_finalize(BlockJob
*job
)
639 assert(job
&& job
->txn
);
641 /* prepare the transaction to complete */
642 rc
= block_job_txn_apply(job
->txn
, block_job_prepare
, true);
644 block_job_completed_txn_abort(job
);
646 block_job_txn_apply(job
->txn
, block_job_finalize_single
, true);
650 static void block_job_completed_txn_success(BlockJob
*job
)
652 BlockJobTxn
*txn
= job
->txn
;
655 block_job_state_transition(job
, BLOCK_JOB_STATUS_WAITING
);
658 * Successful completion, see if there are other running jobs in this
661 QLIST_FOREACH(other_job
, &txn
->jobs
, txn_list
) {
662 if (!other_job
->completed
) {
665 assert(other_job
->ret
== 0);
668 block_job_txn_apply(txn
, block_job_event_pending
, false);
670 /* If no jobs need manual finalization, automatically do so */
671 if (block_job_txn_apply(txn
, block_job_needs_finalize
, false) == 0) {
672 block_job_do_finalize(job
);
676 void block_job_set_speed(BlockJob
*job
, int64_t speed
, Error
**errp
)
678 int64_t old_speed
= job
->speed
;
680 if (block_job_apply_verb(job
, BLOCK_JOB_VERB_SET_SPEED
, errp
)) {
684 error_setg(errp
, QERR_INVALID_PARAMETER
, "speed");
688 ratelimit_set_speed(&job
->limit
, speed
, BLOCK_JOB_SLICE_TIME
);
691 if (speed
&& speed
<= old_speed
) {
695 /* kick only if a timer is pending */
696 block_job_enter_cond(job
, block_job_timer_pending
);
699 int64_t block_job_ratelimit_get_delay(BlockJob
*job
, uint64_t n
)
705 return ratelimit_calculate_delay(&job
->limit
, n
);
708 void block_job_complete(BlockJob
*job
, Error
**errp
)
710 /* Should not be reachable via external interface for internal jobs */
712 if (block_job_apply_verb(job
, BLOCK_JOB_VERB_COMPLETE
, errp
)) {
715 if (job
->pause_count
|| job
->cancelled
|| !job
->driver
->complete
) {
716 error_setg(errp
, "The active block job '%s' cannot be completed",
721 job
->driver
->complete(job
, errp
);
724 void block_job_finalize(BlockJob
*job
, Error
**errp
)
726 assert(job
&& job
->job
.id
);
727 if (block_job_apply_verb(job
, BLOCK_JOB_VERB_FINALIZE
, errp
)) {
730 block_job_do_finalize(job
);
733 void block_job_dismiss(BlockJob
**jobptr
, Error
**errp
)
735 BlockJob
*job
= *jobptr
;
736 /* similarly to _complete, this is QMP-interface only. */
738 if (block_job_apply_verb(job
, BLOCK_JOB_VERB_DISMISS
, errp
)) {
742 block_job_do_dismiss(job
);
746 void block_job_user_pause(BlockJob
*job
, Error
**errp
)
748 if (block_job_apply_verb(job
, BLOCK_JOB_VERB_PAUSE
, errp
)) {
751 if (job
->user_paused
) {
752 error_setg(errp
, "Job is already paused");
755 job
->user_paused
= true;
756 block_job_pause(job
);
759 bool block_job_user_paused(BlockJob
*job
)
761 return job
->user_paused
;
764 void block_job_user_resume(BlockJob
*job
, Error
**errp
)
767 if (!job
->user_paused
|| job
->pause_count
<= 0) {
768 error_setg(errp
, "Can't resume a job that was not paused");
771 if (block_job_apply_verb(job
, BLOCK_JOB_VERB_RESUME
, errp
)) {
774 block_job_iostatus_reset(job
);
775 job
->user_paused
= false;
776 block_job_resume(job
);
779 void block_job_cancel(BlockJob
*job
, bool force
)
781 if (job
->status
== BLOCK_JOB_STATUS_CONCLUDED
) {
782 block_job_do_dismiss(job
);
785 block_job_cancel_async(job
, force
);
786 if (!block_job_started(job
)) {
787 block_job_completed(job
, -ECANCELED
);
788 } else if (job
->deferred_to_main_loop
) {
789 block_job_completed_txn_abort(job
);
791 block_job_enter(job
);
795 void block_job_user_cancel(BlockJob
*job
, bool force
, Error
**errp
)
797 if (block_job_apply_verb(job
, BLOCK_JOB_VERB_CANCEL
, errp
)) {
800 block_job_cancel(job
, force
);
803 /* A wrapper around block_job_cancel() taking an Error ** parameter so it may be
804 * used with block_job_finish_sync() without the need for (rather nasty)
805 * function pointer casts there. */
806 static void block_job_cancel_err(BlockJob
*job
, Error
**errp
)
808 block_job_cancel(job
, false);
811 int block_job_cancel_sync(BlockJob
*job
)
813 return block_job_finish_sync(job
, &block_job_cancel_err
, NULL
);
816 void block_job_cancel_sync_all(void)
819 AioContext
*aio_context
;
821 while ((job
= block_job_next(NULL
))) {
822 aio_context
= blk_get_aio_context(job
->blk
);
823 aio_context_acquire(aio_context
);
824 block_job_cancel_sync(job
);
825 aio_context_release(aio_context
);
829 int block_job_complete_sync(BlockJob
*job
, Error
**errp
)
831 return block_job_finish_sync(job
, &block_job_complete
, errp
);
834 void block_job_progress_update(BlockJob
*job
, uint64_t done
)
839 void block_job_progress_set_remaining(BlockJob
*job
, uint64_t remaining
)
841 job
->len
= job
->offset
+ remaining
;
844 BlockJobInfo
*block_job_query(BlockJob
*job
, Error
**errp
)
848 if (block_job_is_internal(job
)) {
849 error_setg(errp
, "Cannot query QEMU internal jobs");
852 info
= g_new0(BlockJobInfo
, 1);
853 info
->type
= g_strdup(job_type_str(&job
->job
));
854 info
->device
= g_strdup(job
->job
.id
);
855 info
->len
= job
->len
;
856 info
->busy
= atomic_read(&job
->busy
);
857 info
->paused
= job
->pause_count
> 0;
858 info
->offset
= job
->offset
;
859 info
->speed
= job
->speed
;
860 info
->io_status
= job
->iostatus
;
861 info
->ready
= job
->ready
;
862 info
->status
= job
->status
;
863 info
->auto_finalize
= job
->auto_finalize
;
864 info
->auto_dismiss
= job
->auto_dismiss
;
865 info
->has_error
= job
->ret
!= 0;
866 info
->error
= job
->ret
? g_strdup(strerror(-job
->ret
)) : NULL
;
870 static void block_job_iostatus_set_err(BlockJob
*job
, int error
)
872 if (job
->iostatus
== BLOCK_DEVICE_IO_STATUS_OK
) {
873 job
->iostatus
= error
== ENOSPC
? BLOCK_DEVICE_IO_STATUS_NOSPACE
:
874 BLOCK_DEVICE_IO_STATUS_FAILED
;
878 static void block_job_event_cancelled(BlockJob
*job
)
880 if (block_job_is_internal(job
)) {
884 qapi_event_send_block_job_cancelled(job_type(&job
->job
),
892 static void block_job_event_completed(BlockJob
*job
, const char *msg
)
894 if (block_job_is_internal(job
)) {
898 qapi_event_send_block_job_completed(job_type(&job
->job
),
908 static int block_job_event_pending(BlockJob
*job
)
910 block_job_state_transition(job
, BLOCK_JOB_STATUS_PENDING
);
911 if (!job
->auto_finalize
&& !block_job_is_internal(job
)) {
912 qapi_event_send_block_job_pending(job_type(&job
->job
),
920 * API for block job drivers and the block layer. These functions are
921 * declared in blockjob_int.h.
924 void *block_job_create(const char *job_id
, const BlockJobDriver
*driver
,
925 BlockJobTxn
*txn
, BlockDriverState
*bs
, uint64_t perm
,
926 uint64_t shared_perm
, int64_t speed
, int flags
,
927 BlockCompletionFunc
*cb
, void *opaque
, Error
**errp
)
934 error_setg(errp
, QERR_DEVICE_IN_USE
, bdrv_get_device_name(bs
));
938 if (job_id
== NULL
&& !(flags
& BLOCK_JOB_INTERNAL
)) {
939 job_id
= bdrv_get_device_name(bs
);
941 error_setg(errp
, "An explicit job ID is required for this node");
947 if (flags
& BLOCK_JOB_INTERNAL
) {
948 error_setg(errp
, "Cannot specify job ID for internal block job");
953 blk
= blk_new(perm
, shared_perm
);
954 ret
= blk_insert_bs(blk
, bs
, errp
);
960 job
= job_create(job_id
, &driver
->job_driver
, errp
);
966 assert(is_block_job(&job
->job
));
968 job
->driver
= driver
;
971 job
->opaque
= opaque
;
974 job
->pause_count
= 1;
976 job
->auto_finalize
= !(flags
& BLOCK_JOB_MANUAL_FINALIZE
);
977 job
->auto_dismiss
= !(flags
& BLOCK_JOB_MANUAL_DISMISS
);
978 block_job_state_transition(job
, BLOCK_JOB_STATUS_CREATED
);
979 aio_timer_init(qemu_get_aio_context(), &job
->sleep_timer
,
980 QEMU_CLOCK_REALTIME
, SCALE_NS
,
981 block_job_sleep_timer_cb
, job
);
983 error_setg(&job
->blocker
, "block device is in use by block job: %s",
984 job_type_str(&job
->job
));
985 block_job_add_bdrv(job
, "main node", bs
, 0, BLK_PERM_ALL
, &error_abort
);
988 bdrv_op_unblock(bs
, BLOCK_OP_TYPE_DATAPLANE
, job
->blocker
);
990 blk_add_aio_context_notifier(blk
, block_job_attached_aio_context
,
991 block_job_detach_aio_context
, job
);
993 /* Only set speed when necessary to avoid NotSupported error */
995 Error
*local_err
= NULL
;
997 block_job_set_speed(job
, speed
, &local_err
);
999 block_job_early_fail(job
);
1000 error_propagate(errp
, local_err
);
1005 /* Single jobs are modeled as single-job transactions for sake of
1006 * consolidating the job management logic */
1008 txn
= block_job_txn_new();
1009 block_job_txn_add_job(txn
, job
);
1010 block_job_txn_unref(txn
);
1012 block_job_txn_add_job(txn
, job
);
1018 void block_job_early_fail(BlockJob
*job
)
1020 assert(job
->status
== BLOCK_JOB_STATUS_CREATED
);
1021 block_job_decommission(job
);
1024 void block_job_completed(BlockJob
*job
, int ret
)
1026 assert(job
&& job
->txn
&& !job
->completed
);
1027 assert(blk_bs(job
->blk
)->job
== job
);
1028 job
->completed
= true;
1030 block_job_update_rc(job
);
1031 trace_block_job_completed(job
, ret
, job
->ret
);
1033 block_job_completed_txn_abort(job
);
1035 block_job_completed_txn_success(job
);
1039 static bool block_job_should_pause(BlockJob
*job
)
1041 return job
->pause_count
> 0;
1044 /* Yield, and schedule a timer to reenter the coroutine after @ns nanoseconds.
1045 * Reentering the job coroutine with block_job_enter() before the timer has
1046 * expired is allowed and cancels the timer.
1048 * If @ns is (uint64_t) -1, no timer is scheduled and block_job_enter() must be
1049 * called explicitly. */
1050 static void block_job_do_yield(BlockJob
*job
, uint64_t ns
)
1054 timer_mod(&job
->sleep_timer
, ns
);
1058 qemu_coroutine_yield();
1060 /* Set by block_job_enter before re-entering the coroutine. */
1064 void coroutine_fn
block_job_pause_point(BlockJob
*job
)
1066 assert(job
&& block_job_started(job
));
1068 if (!block_job_should_pause(job
)) {
1071 if (block_job_is_cancelled(job
)) {
1075 if (job
->driver
->pause
) {
1076 job
->driver
->pause(job
);
1079 if (block_job_should_pause(job
) && !block_job_is_cancelled(job
)) {
1080 BlockJobStatus status
= job
->status
;
1081 block_job_state_transition(job
, status
== BLOCK_JOB_STATUS_READY
? \
1082 BLOCK_JOB_STATUS_STANDBY
: \
1083 BLOCK_JOB_STATUS_PAUSED
);
1085 block_job_do_yield(job
, -1);
1086 job
->paused
= false;
1087 block_job_state_transition(job
, status
);
1090 if (job
->driver
->resume
) {
1091 job
->driver
->resume(job
);
1096 * Conditionally enter a block_job pending a call to fn() while
1097 * under the block_job_lock critical section.
1099 static void block_job_enter_cond(BlockJob
*job
, bool(*fn
)(BlockJob
*job
))
1101 if (!block_job_started(job
)) {
1104 if (job
->deferred_to_main_loop
) {
1114 if (fn
&& !fn(job
)) {
1119 assert(!job
->deferred_to_main_loop
);
1120 timer_del(&job
->sleep_timer
);
1123 aio_co_wake(job
->co
);
1126 void block_job_enter(BlockJob
*job
)
1128 block_job_enter_cond(job
, NULL
);
1131 bool block_job_is_cancelled(BlockJob
*job
)
1133 return job
->cancelled
;
1136 void block_job_sleep_ns(BlockJob
*job
, int64_t ns
)
1140 /* Check cancellation *before* setting busy = false, too! */
1141 if (block_job_is_cancelled(job
)) {
1145 if (!block_job_should_pause(job
)) {
1146 block_job_do_yield(job
, qemu_clock_get_ns(QEMU_CLOCK_REALTIME
) + ns
);
1149 block_job_pause_point(job
);
1152 void block_job_yield(BlockJob
*job
)
1156 /* Check cancellation *before* setting busy = false, too! */
1157 if (block_job_is_cancelled(job
)) {
1161 if (!block_job_should_pause(job
)) {
1162 block_job_do_yield(job
, -1);
1165 block_job_pause_point(job
);
1168 void block_job_iostatus_reset(BlockJob
*job
)
1170 if (job
->iostatus
== BLOCK_DEVICE_IO_STATUS_OK
) {
1173 assert(job
->user_paused
&& job
->pause_count
> 0);
1174 job
->iostatus
= BLOCK_DEVICE_IO_STATUS_OK
;
1177 void block_job_event_ready(BlockJob
*job
)
1179 block_job_state_transition(job
, BLOCK_JOB_STATUS_READY
);
1182 if (block_job_is_internal(job
)) {
1186 qapi_event_send_block_job_ready(job_type(&job
->job
),
1190 job
->speed
, &error_abort
);
1193 BlockErrorAction
block_job_error_action(BlockJob
*job
, BlockdevOnError on_err
,
1194 int is_read
, int error
)
1196 BlockErrorAction action
;
1199 case BLOCKDEV_ON_ERROR_ENOSPC
:
1200 case BLOCKDEV_ON_ERROR_AUTO
:
1201 action
= (error
== ENOSPC
) ?
1202 BLOCK_ERROR_ACTION_STOP
: BLOCK_ERROR_ACTION_REPORT
;
1204 case BLOCKDEV_ON_ERROR_STOP
:
1205 action
= BLOCK_ERROR_ACTION_STOP
;
1207 case BLOCKDEV_ON_ERROR_REPORT
:
1208 action
= BLOCK_ERROR_ACTION_REPORT
;
1210 case BLOCKDEV_ON_ERROR_IGNORE
:
1211 action
= BLOCK_ERROR_ACTION_IGNORE
;
1216 if (!block_job_is_internal(job
)) {
1217 qapi_event_send_block_job_error(job
->job
.id
,
1218 is_read
? IO_OPERATION_TYPE_READ
:
1219 IO_OPERATION_TYPE_WRITE
,
1220 action
, &error_abort
);
1222 if (action
== BLOCK_ERROR_ACTION_STOP
) {
1223 block_job_pause(job
);
1224 /* make the pause user visible, which will be resumed from QMP. */
1225 job
->user_paused
= true;
1226 block_job_iostatus_set_err(job
, error
);
1233 AioContext
*aio_context
;
1234 BlockJobDeferToMainLoopFn
*fn
;
1236 } BlockJobDeferToMainLoopData
;
1238 static void block_job_defer_to_main_loop_bh(void *opaque
)
1240 BlockJobDeferToMainLoopData
*data
= opaque
;
1241 AioContext
*aio_context
;
1243 /* Prevent race with block_job_defer_to_main_loop() */
1244 aio_context_acquire(data
->aio_context
);
1246 /* Fetch BDS AioContext again, in case it has changed */
1247 aio_context
= blk_get_aio_context(data
->job
->blk
);
1248 if (aio_context
!= data
->aio_context
) {
1249 aio_context_acquire(aio_context
);
1252 data
->fn(data
->job
, data
->opaque
);
1254 if (aio_context
!= data
->aio_context
) {
1255 aio_context_release(aio_context
);
1258 aio_context_release(data
->aio_context
);
1263 void block_job_defer_to_main_loop(BlockJob
*job
,
1264 BlockJobDeferToMainLoopFn
*fn
,
1267 BlockJobDeferToMainLoopData
*data
= g_malloc(sizeof(*data
));
1269 data
->aio_context
= blk_get_aio_context(job
->blk
);
1271 data
->opaque
= opaque
;
1272 job
->deferred_to_main_loop
= true;
1274 aio_bh_schedule_oneshot(qemu_get_aio_context(),
1275 block_job_defer_to_main_loop_bh
, data
);