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"
40 * The block job API is composed of two categories of functions.
42 * The first includes functions used by the monitor. The monitor is
43 * peculiar in that it accesses the block job list with block_job_get, and
44 * therefore needs consistency across block_job_get and the actual operation
45 * (e.g. block_job_set_speed). The consistency is achieved with
46 * aio_context_acquire/release. These functions are declared in blockjob.h.
48 * The second includes functions used by the block job drivers and sometimes
49 * by the core block layer. These do not care about locking, because the
50 * whole coroutine runs under the AioContext lock, and are declared in
54 static bool is_block_job(Job
*job
)
56 return job_type(job
) == JOB_TYPE_BACKUP
||
57 job_type(job
) == JOB_TYPE_COMMIT
||
58 job_type(job
) == JOB_TYPE_MIRROR
||
59 job_type(job
) == JOB_TYPE_STREAM
;
62 BlockJob
*block_job_next(BlockJob
*bjob
)
64 Job
*job
= bjob
? &bjob
->job
: NULL
;
68 } while (job
&& !is_block_job(job
));
70 return job
? container_of(job
, BlockJob
, job
) : NULL
;
73 BlockJob
*block_job_get(const char *id
)
75 Job
*job
= job_get(id
);
77 if (job
&& is_block_job(job
)) {
78 return container_of(job
, BlockJob
, job
);
84 static void block_job_attached_aio_context(AioContext
*new_context
,
86 static void block_job_detach_aio_context(void *opaque
);
88 void block_job_free(Job
*job
)
90 BlockJob
*bjob
= container_of(job
, BlockJob
, job
);
91 BlockDriverState
*bs
= blk_bs(bjob
->blk
);
94 block_job_remove_all_bdrv(bjob
);
95 blk_remove_aio_context_notifier(bjob
->blk
,
96 block_job_attached_aio_context
,
97 block_job_detach_aio_context
, bjob
);
99 error_free(bjob
->blocker
);
102 static void block_job_attached_aio_context(AioContext
*new_context
,
105 BlockJob
*job
= opaque
;
106 const JobDriver
*drv
= job
->job
.driver
;
107 BlockJobDriver
*bjdrv
= container_of(drv
, BlockJobDriver
, job_driver
);
109 job
->job
.aio_context
= new_context
;
110 if (bjdrv
->attached_aio_context
) {
111 bjdrv
->attached_aio_context(job
, new_context
);
114 job_resume(&job
->job
);
117 void block_job_drain(Job
*job
)
119 BlockJob
*bjob
= container_of(job
, BlockJob
, job
);
120 const JobDriver
*drv
= job
->driver
;
121 BlockJobDriver
*bjdrv
= container_of(drv
, BlockJobDriver
, job_driver
);
123 blk_drain(bjob
->blk
);
129 static void block_job_detach_aio_context(void *opaque
)
131 BlockJob
*job
= opaque
;
133 /* In case the job terminates during aio_poll()... */
136 job_pause(&job
->job
);
138 while (!job
->job
.paused
&& !job_is_completed(&job
->job
)) {
139 job_drain(&job
->job
);
142 job
->job
.aio_context
= NULL
;
143 job_unref(&job
->job
);
146 static char *child_job_get_parent_desc(BdrvChild
*c
)
148 BlockJob
*job
= c
->opaque
;
149 return g_strdup_printf("%s job '%s'", job_type_str(&job
->job
), job
->job
.id
);
152 static void child_job_drained_begin(BdrvChild
*c
)
154 BlockJob
*job
= c
->opaque
;
155 job_pause(&job
->job
);
158 static bool child_job_drained_poll(BdrvChild
*c
)
160 BlockJob
*bjob
= c
->opaque
;
161 Job
*job
= &bjob
->job
;
162 const BlockJobDriver
*drv
= block_job_driver(bjob
);
164 /* An inactive or completed job doesn't have any pending requests. Jobs
165 * with !job->busy are either already paused or have a pause point after
166 * being reentered, so no job driver code will run before they pause. */
167 if (!job
->busy
|| job_is_completed(job
) || job
->deferred_to_main_loop
) {
171 /* Otherwise, assume that it isn't fully stopped yet, but allow the job to
172 * override this assumption. */
173 if (drv
->drained_poll
) {
174 return drv
->drained_poll(bjob
);
180 static void child_job_drained_end(BdrvChild
*c
)
182 BlockJob
*job
= c
->opaque
;
183 job_resume(&job
->job
);
186 static const BdrvChildRole child_job
= {
187 .get_parent_desc
= child_job_get_parent_desc
,
188 .drained_begin
= child_job_drained_begin
,
189 .drained_poll
= child_job_drained_poll
,
190 .drained_end
= child_job_drained_end
,
191 .stay_at_node
= true,
194 void block_job_remove_all_bdrv(BlockJob
*job
)
197 for (l
= job
->nodes
; l
; l
= l
->next
) {
198 BdrvChild
*c
= l
->data
;
199 bdrv_op_unblock_all(c
->bs
, job
->blocker
);
200 bdrv_root_unref_child(c
);
202 g_slist_free(job
->nodes
);
206 int block_job_add_bdrv(BlockJob
*job
, const char *name
, BlockDriverState
*bs
,
207 uint64_t perm
, uint64_t shared_perm
, Error
**errp
)
211 c
= bdrv_root_attach_child(bs
, name
, &child_job
, perm
, shared_perm
,
217 job
->nodes
= g_slist_prepend(job
->nodes
, c
);
219 bdrv_op_block_all(bs
, job
->blocker
);
224 void block_job_wakeup_all_bdrv(BlockJob
*job
)
228 for (l
= job
->nodes
; l
; l
= l
->next
) {
229 BdrvChild
*c
= l
->data
;
234 static void block_job_on_idle(Notifier
*n
, void *opaque
)
236 BlockJob
*job
= opaque
;
237 block_job_wakeup_all_bdrv(job
);
240 bool block_job_is_internal(BlockJob
*job
)
242 return (job
->job
.id
== NULL
);
245 const BlockJobDriver
*block_job_driver(BlockJob
*job
)
247 return container_of(job
->job
.driver
, BlockJobDriver
, job_driver
);
250 /* Assumes the job_mutex is held */
251 static bool job_timer_pending(Job
*job
)
253 return timer_pending(&job
->sleep_timer
);
256 void block_job_set_speed(BlockJob
*job
, int64_t speed
, Error
**errp
)
258 int64_t old_speed
= job
->speed
;
260 if (job_apply_verb(&job
->job
, JOB_VERB_SET_SPEED
, errp
)) {
264 error_setg(errp
, QERR_INVALID_PARAMETER
, "speed");
268 ratelimit_set_speed(&job
->limit
, speed
, BLOCK_JOB_SLICE_TIME
);
271 if (speed
&& speed
<= old_speed
) {
275 /* kick only if a timer is pending */
276 job_enter_cond(&job
->job
, job_timer_pending
);
279 int64_t block_job_ratelimit_get_delay(BlockJob
*job
, uint64_t n
)
285 return ratelimit_calculate_delay(&job
->limit
, n
);
288 BlockJobInfo
*block_job_query(BlockJob
*job
, Error
**errp
)
292 if (block_job_is_internal(job
)) {
293 error_setg(errp
, "Cannot query QEMU internal jobs");
296 info
= g_new0(BlockJobInfo
, 1);
297 info
->type
= g_strdup(job_type_str(&job
->job
));
298 info
->device
= g_strdup(job
->job
.id
);
299 info
->busy
= atomic_read(&job
->job
.busy
);
300 info
->paused
= job
->job
.pause_count
> 0;
301 info
->offset
= job
->job
.progress_current
;
302 info
->len
= job
->job
.progress_total
;
303 info
->speed
= job
->speed
;
304 info
->io_status
= job
->iostatus
;
305 info
->ready
= job_is_ready(&job
->job
),
306 info
->status
= job
->job
.status
;
307 info
->auto_finalize
= job
->job
.auto_finalize
;
308 info
->auto_dismiss
= job
->job
.auto_dismiss
;
309 info
->has_error
= job
->job
.ret
!= 0;
310 info
->error
= job
->job
.ret
? g_strdup(strerror(-job
->job
.ret
)) : NULL
;
314 static void block_job_iostatus_set_err(BlockJob
*job
, int error
)
316 if (job
->iostatus
== BLOCK_DEVICE_IO_STATUS_OK
) {
317 job
->iostatus
= error
== ENOSPC
? BLOCK_DEVICE_IO_STATUS_NOSPACE
:
318 BLOCK_DEVICE_IO_STATUS_FAILED
;
322 static void block_job_event_cancelled(Notifier
*n
, void *opaque
)
324 BlockJob
*job
= opaque
;
326 if (block_job_is_internal(job
)) {
330 qapi_event_send_block_job_cancelled(job_type(&job
->job
),
332 job
->job
.progress_total
,
333 job
->job
.progress_current
,
337 static void block_job_event_completed(Notifier
*n
, void *opaque
)
339 BlockJob
*job
= opaque
;
340 const char *msg
= NULL
;
342 if (block_job_is_internal(job
)) {
346 if (job
->job
.ret
< 0) {
347 msg
= strerror(-job
->job
.ret
);
350 qapi_event_send_block_job_completed(job_type(&job
->job
),
352 job
->job
.progress_total
,
353 job
->job
.progress_current
,
359 static void block_job_event_pending(Notifier
*n
, void *opaque
)
361 BlockJob
*job
= opaque
;
363 if (block_job_is_internal(job
)) {
367 qapi_event_send_block_job_pending(job_type(&job
->job
),
371 static void block_job_event_ready(Notifier
*n
, void *opaque
)
373 BlockJob
*job
= opaque
;
375 if (block_job_is_internal(job
)) {
379 qapi_event_send_block_job_ready(job_type(&job
->job
),
381 job
->job
.progress_total
,
382 job
->job
.progress_current
,
388 * API for block job drivers and the block layer. These functions are
389 * declared in blockjob_int.h.
392 void *block_job_create(const char *job_id
, const BlockJobDriver
*driver
,
393 JobTxn
*txn
, BlockDriverState
*bs
, uint64_t perm
,
394 uint64_t shared_perm
, int64_t speed
, int flags
,
395 BlockCompletionFunc
*cb
, void *opaque
, Error
**errp
)
402 error_setg(errp
, QERR_DEVICE_IN_USE
, bdrv_get_device_name(bs
));
406 if (job_id
== NULL
&& !(flags
& JOB_INTERNAL
)) {
407 job_id
= bdrv_get_device_name(bs
);
410 blk
= blk_new(perm
, shared_perm
);
411 ret
= blk_insert_bs(blk
, bs
, errp
);
417 job
= job_create(job_id
, &driver
->job_driver
, txn
, blk_get_aio_context(blk
),
418 flags
, cb
, opaque
, errp
);
424 assert(is_block_job(&job
->job
));
425 assert(job
->job
.driver
->free
== &block_job_free
);
426 assert(job
->job
.driver
->user_resume
== &block_job_user_resume
);
427 assert(job
->job
.driver
->drain
== &block_job_drain
);
431 job
->finalize_cancelled_notifier
.notify
= block_job_event_cancelled
;
432 job
->finalize_completed_notifier
.notify
= block_job_event_completed
;
433 job
->pending_notifier
.notify
= block_job_event_pending
;
434 job
->ready_notifier
.notify
= block_job_event_ready
;
435 job
->idle_notifier
.notify
= block_job_on_idle
;
437 notifier_list_add(&job
->job
.on_finalize_cancelled
,
438 &job
->finalize_cancelled_notifier
);
439 notifier_list_add(&job
->job
.on_finalize_completed
,
440 &job
->finalize_completed_notifier
);
441 notifier_list_add(&job
->job
.on_pending
, &job
->pending_notifier
);
442 notifier_list_add(&job
->job
.on_ready
, &job
->ready_notifier
);
443 notifier_list_add(&job
->job
.on_idle
, &job
->idle_notifier
);
445 error_setg(&job
->blocker
, "block device is in use by block job: %s",
446 job_type_str(&job
->job
));
447 block_job_add_bdrv(job
, "main node", bs
, 0, BLK_PERM_ALL
, &error_abort
);
450 bdrv_op_unblock(bs
, BLOCK_OP_TYPE_DATAPLANE
, job
->blocker
);
452 blk_add_aio_context_notifier(blk
, block_job_attached_aio_context
,
453 block_job_detach_aio_context
, job
);
455 /* Only set speed when necessary to avoid NotSupported error */
457 Error
*local_err
= NULL
;
459 block_job_set_speed(job
, speed
, &local_err
);
461 job_early_fail(&job
->job
);
462 error_propagate(errp
, local_err
);
470 void block_job_iostatus_reset(BlockJob
*job
)
472 if (job
->iostatus
== BLOCK_DEVICE_IO_STATUS_OK
) {
475 assert(job
->job
.user_paused
&& job
->job
.pause_count
> 0);
476 job
->iostatus
= BLOCK_DEVICE_IO_STATUS_OK
;
479 void block_job_user_resume(Job
*job
)
481 BlockJob
*bjob
= container_of(job
, BlockJob
, job
);
482 block_job_iostatus_reset(bjob
);
485 BlockErrorAction
block_job_error_action(BlockJob
*job
, BlockdevOnError on_err
,
486 int is_read
, int error
)
488 BlockErrorAction action
;
491 case BLOCKDEV_ON_ERROR_ENOSPC
:
492 case BLOCKDEV_ON_ERROR_AUTO
:
493 action
= (error
== ENOSPC
) ?
494 BLOCK_ERROR_ACTION_STOP
: BLOCK_ERROR_ACTION_REPORT
;
496 case BLOCKDEV_ON_ERROR_STOP
:
497 action
= BLOCK_ERROR_ACTION_STOP
;
499 case BLOCKDEV_ON_ERROR_REPORT
:
500 action
= BLOCK_ERROR_ACTION_REPORT
;
502 case BLOCKDEV_ON_ERROR_IGNORE
:
503 action
= BLOCK_ERROR_ACTION_IGNORE
;
508 if (!block_job_is_internal(job
)) {
509 qapi_event_send_block_job_error(job
->job
.id
,
510 is_read
? IO_OPERATION_TYPE_READ
:
511 IO_OPERATION_TYPE_WRITE
,
514 if (action
== BLOCK_ERROR_ACTION_STOP
) {
515 job_pause(&job
->job
);
516 /* make the pause user visible, which will be resumed from QMP. */
517 job
->job
.user_paused
= true;
518 block_job_iostatus_set_err(job
, error
);