2 * Block layer I/O functions
4 * Copyright (c) 2003 Fabrice Bellard
6 * Permission is hereby granted, free of charge, to any person obtaining a copy
7 * of this software and associated documentation files (the "Software"), to deal
8 * in the Software without restriction, including without limitation the rights
9 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10 * copies of the Software, and to permit persons to whom the Software is
11 * furnished to do so, subject to the following conditions:
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
25 #include "qemu/osdep.h"
27 #include "sysemu/block-backend.h"
28 #include "block/blockjob.h"
29 #include "block/block_int.h"
30 #include "qemu/cutils.h"
31 #include "qapi/error.h"
32 #include "qemu/error-report.h"
34 #define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */
36 static BlockAIOCB
*bdrv_co_aio_prw_vector(BdrvChild
*child
,
39 BdrvRequestFlags flags
,
40 BlockCompletionFunc
*cb
,
43 static void coroutine_fn
bdrv_co_do_rw(void *opaque
);
44 static int coroutine_fn
bdrv_co_do_pwrite_zeroes(BlockDriverState
*bs
,
45 int64_t offset
, int count
, BdrvRequestFlags flags
);
47 static void bdrv_parent_drained_begin(BlockDriverState
*bs
)
51 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
52 if (c
->role
->drained_begin
) {
53 c
->role
->drained_begin(c
);
58 static void bdrv_parent_drained_end(BlockDriverState
*bs
)
62 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
63 if (c
->role
->drained_end
) {
64 c
->role
->drained_end(c
);
69 static void bdrv_merge_limits(BlockLimits
*dst
, const BlockLimits
*src
)
71 dst
->opt_transfer
= MAX(dst
->opt_transfer
, src
->opt_transfer
);
72 dst
->max_transfer
= MIN_NON_ZERO(dst
->max_transfer
, src
->max_transfer
);
73 dst
->opt_mem_alignment
= MAX(dst
->opt_mem_alignment
,
74 src
->opt_mem_alignment
);
75 dst
->min_mem_alignment
= MAX(dst
->min_mem_alignment
,
76 src
->min_mem_alignment
);
77 dst
->max_iov
= MIN_NON_ZERO(dst
->max_iov
, src
->max_iov
);
80 void bdrv_refresh_limits(BlockDriverState
*bs
, Error
**errp
)
82 BlockDriver
*drv
= bs
->drv
;
83 Error
*local_err
= NULL
;
85 memset(&bs
->bl
, 0, sizeof(bs
->bl
));
91 /* Default alignment based on whether driver has byte interface */
92 bs
->bl
.request_alignment
= drv
->bdrv_co_preadv
? 1 : 512;
94 /* Take some limits from the children as a default */
96 bdrv_refresh_limits(bs
->file
->bs
, &local_err
);
98 error_propagate(errp
, local_err
);
101 bdrv_merge_limits(&bs
->bl
, &bs
->file
->bs
->bl
);
103 bs
->bl
.min_mem_alignment
= 512;
104 bs
->bl
.opt_mem_alignment
= getpagesize();
106 /* Safe default since most protocols use readv()/writev()/etc */
107 bs
->bl
.max_iov
= IOV_MAX
;
111 bdrv_refresh_limits(bs
->backing
->bs
, &local_err
);
113 error_propagate(errp
, local_err
);
116 bdrv_merge_limits(&bs
->bl
, &bs
->backing
->bs
->bl
);
119 /* Then let the driver override it */
120 if (drv
->bdrv_refresh_limits
) {
121 drv
->bdrv_refresh_limits(bs
, errp
);
126 * The copy-on-read flag is actually a reference count so multiple users may
127 * use the feature without worrying about clobbering its previous state.
128 * Copy-on-read stays enabled until all users have called to disable it.
130 void bdrv_enable_copy_on_read(BlockDriverState
*bs
)
135 void bdrv_disable_copy_on_read(BlockDriverState
*bs
)
137 assert(bs
->copy_on_read
> 0);
141 /* Check if any requests are in-flight (including throttled requests) */
142 bool bdrv_requests_pending(BlockDriverState
*bs
)
146 if (atomic_read(&bs
->in_flight
)) {
150 QLIST_FOREACH(child
, &bs
->children
, next
) {
151 if (bdrv_requests_pending(child
->bs
)) {
159 static bool bdrv_drain_recurse(BlockDriverState
*bs
)
164 waited
= BDRV_POLL_WHILE(bs
, atomic_read(&bs
->in_flight
) > 0);
166 if (bs
->drv
&& bs
->drv
->bdrv_drain
) {
167 bs
->drv
->bdrv_drain(bs
);
170 QLIST_FOREACH(child
, &bs
->children
, next
) {
171 waited
|= bdrv_drain_recurse(child
->bs
);
179 BlockDriverState
*bs
;
183 static void bdrv_co_drain_bh_cb(void *opaque
)
185 BdrvCoDrainData
*data
= opaque
;
186 Coroutine
*co
= data
->co
;
187 BlockDriverState
*bs
= data
->bs
;
189 bdrv_dec_in_flight(bs
);
190 bdrv_drained_begin(bs
);
195 static void coroutine_fn
bdrv_co_yield_to_drain(BlockDriverState
*bs
)
197 BdrvCoDrainData data
;
199 /* Calling bdrv_drain() from a BH ensures the current coroutine yields and
200 * other coroutines run if they were queued from
201 * qemu_co_queue_run_restart(). */
203 assert(qemu_in_coroutine());
204 data
= (BdrvCoDrainData
) {
205 .co
= qemu_coroutine_self(),
209 bdrv_inc_in_flight(bs
);
210 aio_bh_schedule_oneshot(bdrv_get_aio_context(bs
),
211 bdrv_co_drain_bh_cb
, &data
);
213 qemu_coroutine_yield();
214 /* If we are resumed from some other event (such as an aio completion or a
215 * timer callback), it is a bug in the caller that should be fixed. */
219 void bdrv_drained_begin(BlockDriverState
*bs
)
221 if (qemu_in_coroutine()) {
222 bdrv_co_yield_to_drain(bs
);
226 if (!bs
->quiesce_counter
++) {
227 aio_disable_external(bdrv_get_aio_context(bs
));
228 bdrv_parent_drained_begin(bs
);
231 bdrv_drain_recurse(bs
);
234 void bdrv_drained_end(BlockDriverState
*bs
)
236 assert(bs
->quiesce_counter
> 0);
237 if (--bs
->quiesce_counter
> 0) {
241 bdrv_parent_drained_end(bs
);
242 aio_enable_external(bdrv_get_aio_context(bs
));
246 * Wait for pending requests to complete on a single BlockDriverState subtree,
247 * and suspend block driver's internal I/O until next request arrives.
249 * Note that unlike bdrv_drain_all(), the caller must hold the BlockDriverState
252 * Only this BlockDriverState's AioContext is run, so in-flight requests must
253 * not depend on events in other AioContexts. In that case, use
254 * bdrv_drain_all() instead.
256 void coroutine_fn
bdrv_co_drain(BlockDriverState
*bs
)
258 assert(qemu_in_coroutine());
259 bdrv_drained_begin(bs
);
260 bdrv_drained_end(bs
);
263 void bdrv_drain(BlockDriverState
*bs
)
265 bdrv_drained_begin(bs
);
266 bdrv_drained_end(bs
);
270 * Wait for pending requests to complete across all BlockDriverStates
272 * This function does not flush data to disk, use bdrv_flush_all() for that
273 * after calling this function.
275 * This pauses all block jobs and disables external clients. It must
276 * be paired with bdrv_drain_all_end().
278 * NOTE: no new block jobs or BlockDriverStates can be created between
279 * the bdrv_drain_all_begin() and bdrv_drain_all_end() calls.
281 void bdrv_drain_all_begin(void)
283 /* Always run first iteration so any pending completion BHs run */
285 BlockDriverState
*bs
;
287 BlockJob
*job
= NULL
;
288 GSList
*aio_ctxs
= NULL
, *ctx
;
290 while ((job
= block_job_next(job
))) {
291 AioContext
*aio_context
= blk_get_aio_context(job
->blk
);
293 aio_context_acquire(aio_context
);
294 block_job_pause(job
);
295 aio_context_release(aio_context
);
298 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
299 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
301 aio_context_acquire(aio_context
);
302 bdrv_parent_drained_begin(bs
);
303 aio_disable_external(aio_context
);
304 aio_context_release(aio_context
);
306 if (!g_slist_find(aio_ctxs
, aio_context
)) {
307 aio_ctxs
= g_slist_prepend(aio_ctxs
, aio_context
);
311 /* Note that completion of an asynchronous I/O operation can trigger any
312 * number of other I/O operations on other devices---for example a
313 * coroutine can submit an I/O request to another device in response to
314 * request completion. Therefore we must keep looping until there was no
315 * more activity rather than simply draining each device independently.
320 for (ctx
= aio_ctxs
; ctx
!= NULL
; ctx
= ctx
->next
) {
321 AioContext
*aio_context
= ctx
->data
;
323 aio_context_acquire(aio_context
);
324 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
325 if (aio_context
== bdrv_get_aio_context(bs
)) {
326 waited
|= bdrv_drain_recurse(bs
);
329 aio_context_release(aio_context
);
333 g_slist_free(aio_ctxs
);
336 void bdrv_drain_all_end(void)
338 BlockDriverState
*bs
;
340 BlockJob
*job
= NULL
;
342 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
343 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
345 aio_context_acquire(aio_context
);
346 aio_enable_external(aio_context
);
347 bdrv_parent_drained_end(bs
);
348 aio_context_release(aio_context
);
351 while ((job
= block_job_next(job
))) {
352 AioContext
*aio_context
= blk_get_aio_context(job
->blk
);
354 aio_context_acquire(aio_context
);
355 block_job_resume(job
);
356 aio_context_release(aio_context
);
360 void bdrv_drain_all(void)
362 bdrv_drain_all_begin();
363 bdrv_drain_all_end();
367 * Remove an active request from the tracked requests list
369 * This function should be called when a tracked request is completing.
371 static void tracked_request_end(BdrvTrackedRequest
*req
)
373 if (req
->serialising
) {
374 req
->bs
->serialising_in_flight
--;
377 QLIST_REMOVE(req
, list
);
378 qemu_co_queue_restart_all(&req
->wait_queue
);
382 * Add an active request to the tracked requests list
384 static void tracked_request_begin(BdrvTrackedRequest
*req
,
385 BlockDriverState
*bs
,
388 enum BdrvTrackedRequestType type
)
390 *req
= (BdrvTrackedRequest
){
395 .co
= qemu_coroutine_self(),
396 .serialising
= false,
397 .overlap_offset
= offset
,
398 .overlap_bytes
= bytes
,
401 qemu_co_queue_init(&req
->wait_queue
);
403 QLIST_INSERT_HEAD(&bs
->tracked_requests
, req
, list
);
406 static void mark_request_serialising(BdrvTrackedRequest
*req
, uint64_t align
)
408 int64_t overlap_offset
= req
->offset
& ~(align
- 1);
409 unsigned int overlap_bytes
= ROUND_UP(req
->offset
+ req
->bytes
, align
)
412 if (!req
->serialising
) {
413 req
->bs
->serialising_in_flight
++;
414 req
->serialising
= true;
417 req
->overlap_offset
= MIN(req
->overlap_offset
, overlap_offset
);
418 req
->overlap_bytes
= MAX(req
->overlap_bytes
, overlap_bytes
);
422 * Round a region to cluster boundaries (sector-based)
424 void bdrv_round_sectors_to_clusters(BlockDriverState
*bs
,
425 int64_t sector_num
, int nb_sectors
,
426 int64_t *cluster_sector_num
,
427 int *cluster_nb_sectors
)
431 if (bdrv_get_info(bs
, &bdi
) < 0 || bdi
.cluster_size
== 0) {
432 *cluster_sector_num
= sector_num
;
433 *cluster_nb_sectors
= nb_sectors
;
435 int64_t c
= bdi
.cluster_size
/ BDRV_SECTOR_SIZE
;
436 *cluster_sector_num
= QEMU_ALIGN_DOWN(sector_num
, c
);
437 *cluster_nb_sectors
= QEMU_ALIGN_UP(sector_num
- *cluster_sector_num
+
443 * Round a region to cluster boundaries
445 void bdrv_round_to_clusters(BlockDriverState
*bs
,
446 int64_t offset
, unsigned int bytes
,
447 int64_t *cluster_offset
,
448 unsigned int *cluster_bytes
)
452 if (bdrv_get_info(bs
, &bdi
) < 0 || bdi
.cluster_size
== 0) {
453 *cluster_offset
= offset
;
454 *cluster_bytes
= bytes
;
456 int64_t c
= bdi
.cluster_size
;
457 *cluster_offset
= QEMU_ALIGN_DOWN(offset
, c
);
458 *cluster_bytes
= QEMU_ALIGN_UP(offset
- *cluster_offset
+ bytes
, c
);
462 static int bdrv_get_cluster_size(BlockDriverState
*bs
)
467 ret
= bdrv_get_info(bs
, &bdi
);
468 if (ret
< 0 || bdi
.cluster_size
== 0) {
469 return bs
->bl
.request_alignment
;
471 return bdi
.cluster_size
;
475 static bool tracked_request_overlaps(BdrvTrackedRequest
*req
,
476 int64_t offset
, unsigned int bytes
)
479 if (offset
>= req
->overlap_offset
+ req
->overlap_bytes
) {
483 if (req
->overlap_offset
>= offset
+ bytes
) {
489 void bdrv_inc_in_flight(BlockDriverState
*bs
)
491 atomic_inc(&bs
->in_flight
);
494 static void dummy_bh_cb(void *opaque
)
498 void bdrv_wakeup(BlockDriverState
*bs
)
501 aio_bh_schedule_oneshot(qemu_get_aio_context(), dummy_bh_cb
, NULL
);
505 void bdrv_dec_in_flight(BlockDriverState
*bs
)
507 atomic_dec(&bs
->in_flight
);
511 static bool coroutine_fn
wait_serialising_requests(BdrvTrackedRequest
*self
)
513 BlockDriverState
*bs
= self
->bs
;
514 BdrvTrackedRequest
*req
;
518 if (!bs
->serialising_in_flight
) {
524 QLIST_FOREACH(req
, &bs
->tracked_requests
, list
) {
525 if (req
== self
|| (!req
->serialising
&& !self
->serialising
)) {
528 if (tracked_request_overlaps(req
, self
->overlap_offset
,
529 self
->overlap_bytes
))
531 /* Hitting this means there was a reentrant request, for
532 * example, a block driver issuing nested requests. This must
533 * never happen since it means deadlock.
535 assert(qemu_coroutine_self() != req
->co
);
537 /* If the request is already (indirectly) waiting for us, or
538 * will wait for us as soon as it wakes up, then just go on
539 * (instead of producing a deadlock in the former case). */
540 if (!req
->waiting_for
) {
541 self
->waiting_for
= req
;
542 qemu_co_queue_wait(&req
->wait_queue
, NULL
);
543 self
->waiting_for
= NULL
;
555 static int bdrv_check_byte_request(BlockDriverState
*bs
, int64_t offset
,
558 if (size
> BDRV_REQUEST_MAX_SECTORS
<< BDRV_SECTOR_BITS
) {
562 if (!bdrv_is_inserted(bs
)) {
573 typedef struct RwCo
{
579 BdrvRequestFlags flags
;
582 static void coroutine_fn
bdrv_rw_co_entry(void *opaque
)
586 if (!rwco
->is_write
) {
587 rwco
->ret
= bdrv_co_preadv(rwco
->child
, rwco
->offset
,
588 rwco
->qiov
->size
, rwco
->qiov
,
591 rwco
->ret
= bdrv_co_pwritev(rwco
->child
, rwco
->offset
,
592 rwco
->qiov
->size
, rwco
->qiov
,
598 * Process a vectored synchronous request using coroutines
600 static int bdrv_prwv_co(BdrvChild
*child
, int64_t offset
,
601 QEMUIOVector
*qiov
, bool is_write
,
602 BdrvRequestFlags flags
)
609 .is_write
= is_write
,
614 if (qemu_in_coroutine()) {
615 /* Fast-path if already in coroutine context */
616 bdrv_rw_co_entry(&rwco
);
618 co
= qemu_coroutine_create(bdrv_rw_co_entry
, &rwco
);
619 qemu_coroutine_enter(co
);
620 BDRV_POLL_WHILE(child
->bs
, rwco
.ret
== NOT_DONE
);
626 * Process a synchronous request using coroutines
628 static int bdrv_rw_co(BdrvChild
*child
, int64_t sector_num
, uint8_t *buf
,
629 int nb_sectors
, bool is_write
, BdrvRequestFlags flags
)
633 .iov_base
= (void *)buf
,
634 .iov_len
= nb_sectors
* BDRV_SECTOR_SIZE
,
637 if (nb_sectors
< 0 || nb_sectors
> BDRV_REQUEST_MAX_SECTORS
) {
641 qemu_iovec_init_external(&qiov
, &iov
, 1);
642 return bdrv_prwv_co(child
, sector_num
<< BDRV_SECTOR_BITS
,
643 &qiov
, is_write
, flags
);
646 /* return < 0 if error. See bdrv_write() for the return codes */
647 int bdrv_read(BdrvChild
*child
, int64_t sector_num
,
648 uint8_t *buf
, int nb_sectors
)
650 return bdrv_rw_co(child
, sector_num
, buf
, nb_sectors
, false, 0);
653 /* Return < 0 if error. Important errors are:
654 -EIO generic I/O error (may happen for all errors)
655 -ENOMEDIUM No media inserted.
656 -EINVAL Invalid sector number or nb_sectors
657 -EACCES Trying to write a read-only device
659 int bdrv_write(BdrvChild
*child
, int64_t sector_num
,
660 const uint8_t *buf
, int nb_sectors
)
662 return bdrv_rw_co(child
, sector_num
, (uint8_t *)buf
, nb_sectors
, true, 0);
665 int bdrv_pwrite_zeroes(BdrvChild
*child
, int64_t offset
,
666 int count
, BdrvRequestFlags flags
)
674 qemu_iovec_init_external(&qiov
, &iov
, 1);
675 return bdrv_prwv_co(child
, offset
, &qiov
, true,
676 BDRV_REQ_ZERO_WRITE
| flags
);
680 * Completely zero out a block device with the help of bdrv_pwrite_zeroes.
681 * The operation is sped up by checking the block status and only writing
682 * zeroes to the device if they currently do not return zeroes. Optional
683 * flags are passed through to bdrv_pwrite_zeroes (e.g. BDRV_REQ_MAY_UNMAP,
686 * Returns < 0 on error, 0 on success. For error codes see bdrv_write().
688 int bdrv_make_zero(BdrvChild
*child
, BdrvRequestFlags flags
)
690 int64_t target_sectors
, ret
, nb_sectors
, sector_num
= 0;
691 BlockDriverState
*bs
= child
->bs
;
692 BlockDriverState
*file
;
695 target_sectors
= bdrv_nb_sectors(bs
);
696 if (target_sectors
< 0) {
697 return target_sectors
;
701 nb_sectors
= MIN(target_sectors
- sector_num
, BDRV_REQUEST_MAX_SECTORS
);
702 if (nb_sectors
<= 0) {
705 ret
= bdrv_get_block_status(bs
, sector_num
, nb_sectors
, &n
, &file
);
707 error_report("error getting block status at sector %" PRId64
": %s",
708 sector_num
, strerror(-ret
));
711 if (ret
& BDRV_BLOCK_ZERO
) {
715 ret
= bdrv_pwrite_zeroes(child
, sector_num
<< BDRV_SECTOR_BITS
,
716 n
<< BDRV_SECTOR_BITS
, flags
);
718 error_report("error writing zeroes at sector %" PRId64
": %s",
719 sector_num
, strerror(-ret
));
726 int bdrv_preadv(BdrvChild
*child
, int64_t offset
, QEMUIOVector
*qiov
)
730 ret
= bdrv_prwv_co(child
, offset
, qiov
, false, 0);
738 int bdrv_pread(BdrvChild
*child
, int64_t offset
, void *buf
, int bytes
)
742 .iov_base
= (void *)buf
,
750 qemu_iovec_init_external(&qiov
, &iov
, 1);
751 return bdrv_preadv(child
, offset
, &qiov
);
754 int bdrv_pwritev(BdrvChild
*child
, int64_t offset
, QEMUIOVector
*qiov
)
758 ret
= bdrv_prwv_co(child
, offset
, qiov
, true, 0);
766 int bdrv_pwrite(BdrvChild
*child
, int64_t offset
, const void *buf
, int bytes
)
770 .iov_base
= (void *) buf
,
778 qemu_iovec_init_external(&qiov
, &iov
, 1);
779 return bdrv_pwritev(child
, offset
, &qiov
);
783 * Writes to the file and ensures that no writes are reordered across this
784 * request (acts as a barrier)
786 * Returns 0 on success, -errno in error cases.
788 int bdrv_pwrite_sync(BdrvChild
*child
, int64_t offset
,
789 const void *buf
, int count
)
793 ret
= bdrv_pwrite(child
, offset
, buf
, count
);
798 ret
= bdrv_flush(child
->bs
);
806 typedef struct CoroutineIOCompletion
{
807 Coroutine
*coroutine
;
809 } CoroutineIOCompletion
;
811 static void bdrv_co_io_em_complete(void *opaque
, int ret
)
813 CoroutineIOCompletion
*co
= opaque
;
816 aio_co_wake(co
->coroutine
);
819 static int coroutine_fn
bdrv_driver_preadv(BlockDriverState
*bs
,
820 uint64_t offset
, uint64_t bytes
,
821 QEMUIOVector
*qiov
, int flags
)
823 BlockDriver
*drv
= bs
->drv
;
825 unsigned int nb_sectors
;
827 assert(!(flags
& ~BDRV_REQ_MASK
));
829 if (drv
->bdrv_co_preadv
) {
830 return drv
->bdrv_co_preadv(bs
, offset
, bytes
, qiov
, flags
);
833 sector_num
= offset
>> BDRV_SECTOR_BITS
;
834 nb_sectors
= bytes
>> BDRV_SECTOR_BITS
;
836 assert((offset
& (BDRV_SECTOR_SIZE
- 1)) == 0);
837 assert((bytes
& (BDRV_SECTOR_SIZE
- 1)) == 0);
838 assert((bytes
>> BDRV_SECTOR_BITS
) <= BDRV_REQUEST_MAX_SECTORS
);
840 if (drv
->bdrv_co_readv
) {
841 return drv
->bdrv_co_readv(bs
, sector_num
, nb_sectors
, qiov
);
844 CoroutineIOCompletion co
= {
845 .coroutine
= qemu_coroutine_self(),
848 acb
= bs
->drv
->bdrv_aio_readv(bs
, sector_num
, qiov
, nb_sectors
,
849 bdrv_co_io_em_complete
, &co
);
853 qemu_coroutine_yield();
859 static int coroutine_fn
bdrv_driver_pwritev(BlockDriverState
*bs
,
860 uint64_t offset
, uint64_t bytes
,
861 QEMUIOVector
*qiov
, int flags
)
863 BlockDriver
*drv
= bs
->drv
;
865 unsigned int nb_sectors
;
868 assert(!(flags
& ~BDRV_REQ_MASK
));
870 if (drv
->bdrv_co_pwritev
) {
871 ret
= drv
->bdrv_co_pwritev(bs
, offset
, bytes
, qiov
,
872 flags
& bs
->supported_write_flags
);
873 flags
&= ~bs
->supported_write_flags
;
877 sector_num
= offset
>> BDRV_SECTOR_BITS
;
878 nb_sectors
= bytes
>> BDRV_SECTOR_BITS
;
880 assert((offset
& (BDRV_SECTOR_SIZE
- 1)) == 0);
881 assert((bytes
& (BDRV_SECTOR_SIZE
- 1)) == 0);
882 assert((bytes
>> BDRV_SECTOR_BITS
) <= BDRV_REQUEST_MAX_SECTORS
);
884 if (drv
->bdrv_co_writev_flags
) {
885 ret
= drv
->bdrv_co_writev_flags(bs
, sector_num
, nb_sectors
, qiov
,
886 flags
& bs
->supported_write_flags
);
887 flags
&= ~bs
->supported_write_flags
;
888 } else if (drv
->bdrv_co_writev
) {
889 assert(!bs
->supported_write_flags
);
890 ret
= drv
->bdrv_co_writev(bs
, sector_num
, nb_sectors
, qiov
);
893 CoroutineIOCompletion co
= {
894 .coroutine
= qemu_coroutine_self(),
897 acb
= bs
->drv
->bdrv_aio_writev(bs
, sector_num
, qiov
, nb_sectors
,
898 bdrv_co_io_em_complete
, &co
);
902 qemu_coroutine_yield();
908 if (ret
== 0 && (flags
& BDRV_REQ_FUA
)) {
909 ret
= bdrv_co_flush(bs
);
915 static int coroutine_fn
916 bdrv_driver_pwritev_compressed(BlockDriverState
*bs
, uint64_t offset
,
917 uint64_t bytes
, QEMUIOVector
*qiov
)
919 BlockDriver
*drv
= bs
->drv
;
921 if (!drv
->bdrv_co_pwritev_compressed
) {
925 return drv
->bdrv_co_pwritev_compressed(bs
, offset
, bytes
, qiov
);
928 static int coroutine_fn
bdrv_co_do_copy_on_readv(BdrvChild
*child
,
929 int64_t offset
, unsigned int bytes
, QEMUIOVector
*qiov
)
931 BlockDriverState
*bs
= child
->bs
;
933 /* Perform I/O through a temporary buffer so that users who scribble over
934 * their read buffer while the operation is in progress do not end up
935 * modifying the image file. This is critical for zero-copy guest I/O
936 * where anything might happen inside guest memory.
940 BlockDriver
*drv
= bs
->drv
;
942 QEMUIOVector bounce_qiov
;
943 int64_t cluster_offset
;
944 unsigned int cluster_bytes
;
948 assert(child
->perm
& (BLK_PERM_WRITE_UNCHANGED
| BLK_PERM_WRITE
));
950 /* Cover entire cluster so no additional backing file I/O is required when
951 * allocating cluster in the image file.
953 bdrv_round_to_clusters(bs
, offset
, bytes
, &cluster_offset
, &cluster_bytes
);
955 trace_bdrv_co_do_copy_on_readv(bs
, offset
, bytes
,
956 cluster_offset
, cluster_bytes
);
958 iov
.iov_len
= cluster_bytes
;
959 iov
.iov_base
= bounce_buffer
= qemu_try_blockalign(bs
, iov
.iov_len
);
960 if (bounce_buffer
== NULL
) {
965 qemu_iovec_init_external(&bounce_qiov
, &iov
, 1);
967 ret
= bdrv_driver_preadv(bs
, cluster_offset
, cluster_bytes
,
973 if (drv
->bdrv_co_pwrite_zeroes
&&
974 buffer_is_zero(bounce_buffer
, iov
.iov_len
)) {
975 /* FIXME: Should we (perhaps conditionally) be setting
976 * BDRV_REQ_MAY_UNMAP, if it will allow for a sparser copy
977 * that still correctly reads as zero? */
978 ret
= bdrv_co_do_pwrite_zeroes(bs
, cluster_offset
, cluster_bytes
, 0);
980 /* This does not change the data on the disk, it is not necessary
981 * to flush even in cache=writethrough mode.
983 ret
= bdrv_driver_pwritev(bs
, cluster_offset
, cluster_bytes
,
988 /* It might be okay to ignore write errors for guest requests. If this
989 * is a deliberate copy-on-read then we don't want to ignore the error.
990 * Simply report it in all cases.
995 skip_bytes
= offset
- cluster_offset
;
996 qemu_iovec_from_buf(qiov
, 0, bounce_buffer
+ skip_bytes
, bytes
);
999 qemu_vfree(bounce_buffer
);
1004 * Forwards an already correctly aligned request to the BlockDriver. This
1005 * handles copy on read, zeroing after EOF, and fragmentation of large
1006 * reads; any other features must be implemented by the caller.
1008 static int coroutine_fn
bdrv_aligned_preadv(BdrvChild
*child
,
1009 BdrvTrackedRequest
*req
, int64_t offset
, unsigned int bytes
,
1010 int64_t align
, QEMUIOVector
*qiov
, int flags
)
1012 BlockDriverState
*bs
= child
->bs
;
1013 int64_t total_bytes
, max_bytes
;
1015 uint64_t bytes_remaining
= bytes
;
1018 assert(is_power_of_2(align
));
1019 assert((offset
& (align
- 1)) == 0);
1020 assert((bytes
& (align
- 1)) == 0);
1021 assert(!qiov
|| bytes
== qiov
->size
);
1022 assert((bs
->open_flags
& BDRV_O_NO_IO
) == 0);
1023 max_transfer
= QEMU_ALIGN_DOWN(MIN_NON_ZERO(bs
->bl
.max_transfer
, INT_MAX
),
1026 /* TODO: We would need a per-BDS .supported_read_flags and
1027 * potential fallback support, if we ever implement any read flags
1028 * to pass through to drivers. For now, there aren't any
1029 * passthrough flags. */
1030 assert(!(flags
& ~(BDRV_REQ_NO_SERIALISING
| BDRV_REQ_COPY_ON_READ
)));
1032 /* Handle Copy on Read and associated serialisation */
1033 if (flags
& BDRV_REQ_COPY_ON_READ
) {
1034 /* If we touch the same cluster it counts as an overlap. This
1035 * guarantees that allocating writes will be serialized and not race
1036 * with each other for the same cluster. For example, in copy-on-read
1037 * it ensures that the CoR read and write operations are atomic and
1038 * guest writes cannot interleave between them. */
1039 mark_request_serialising(req
, bdrv_get_cluster_size(bs
));
1042 if (!(flags
& BDRV_REQ_NO_SERIALISING
)) {
1043 wait_serialising_requests(req
);
1046 if (flags
& BDRV_REQ_COPY_ON_READ
) {
1047 int64_t start_sector
= offset
>> BDRV_SECTOR_BITS
;
1048 int64_t end_sector
= DIV_ROUND_UP(offset
+ bytes
, BDRV_SECTOR_SIZE
);
1049 unsigned int nb_sectors
= end_sector
- start_sector
;
1052 ret
= bdrv_is_allocated(bs
, start_sector
, nb_sectors
, &pnum
);
1057 if (!ret
|| pnum
!= nb_sectors
) {
1058 ret
= bdrv_co_do_copy_on_readv(child
, offset
, bytes
, qiov
);
1063 /* Forward the request to the BlockDriver, possibly fragmenting it */
1064 total_bytes
= bdrv_getlength(bs
);
1065 if (total_bytes
< 0) {
1070 max_bytes
= ROUND_UP(MAX(0, total_bytes
- offset
), align
);
1071 if (bytes
<= max_bytes
&& bytes
<= max_transfer
) {
1072 ret
= bdrv_driver_preadv(bs
, offset
, bytes
, qiov
, 0);
1076 while (bytes_remaining
) {
1080 QEMUIOVector local_qiov
;
1082 num
= MIN(bytes_remaining
, MIN(max_bytes
, max_transfer
));
1084 qemu_iovec_init(&local_qiov
, qiov
->niov
);
1085 qemu_iovec_concat(&local_qiov
, qiov
, bytes
- bytes_remaining
, num
);
1087 ret
= bdrv_driver_preadv(bs
, offset
+ bytes
- bytes_remaining
,
1088 num
, &local_qiov
, 0);
1090 qemu_iovec_destroy(&local_qiov
);
1092 num
= bytes_remaining
;
1093 ret
= qemu_iovec_memset(qiov
, bytes
- bytes_remaining
, 0,
1099 bytes_remaining
-= num
;
1103 return ret
< 0 ? ret
: 0;
1107 * Handle a read request in coroutine context
1109 int coroutine_fn
bdrv_co_preadv(BdrvChild
*child
,
1110 int64_t offset
, unsigned int bytes
, QEMUIOVector
*qiov
,
1111 BdrvRequestFlags flags
)
1113 BlockDriverState
*bs
= child
->bs
;
1114 BlockDriver
*drv
= bs
->drv
;
1115 BdrvTrackedRequest req
;
1117 uint64_t align
= bs
->bl
.request_alignment
;
1118 uint8_t *head_buf
= NULL
;
1119 uint8_t *tail_buf
= NULL
;
1120 QEMUIOVector local_qiov
;
1121 bool use_local_qiov
= false;
1128 ret
= bdrv_check_byte_request(bs
, offset
, bytes
);
1133 bdrv_inc_in_flight(bs
);
1135 /* Don't do copy-on-read if we read data before write operation */
1136 if (bs
->copy_on_read
&& !(flags
& BDRV_REQ_NO_SERIALISING
)) {
1137 flags
|= BDRV_REQ_COPY_ON_READ
;
1140 /* Align read if necessary by padding qiov */
1141 if (offset
& (align
- 1)) {
1142 head_buf
= qemu_blockalign(bs
, align
);
1143 qemu_iovec_init(&local_qiov
, qiov
->niov
+ 2);
1144 qemu_iovec_add(&local_qiov
, head_buf
, offset
& (align
- 1));
1145 qemu_iovec_concat(&local_qiov
, qiov
, 0, qiov
->size
);
1146 use_local_qiov
= true;
1148 bytes
+= offset
& (align
- 1);
1149 offset
= offset
& ~(align
- 1);
1152 if ((offset
+ bytes
) & (align
- 1)) {
1153 if (!use_local_qiov
) {
1154 qemu_iovec_init(&local_qiov
, qiov
->niov
+ 1);
1155 qemu_iovec_concat(&local_qiov
, qiov
, 0, qiov
->size
);
1156 use_local_qiov
= true;
1158 tail_buf
= qemu_blockalign(bs
, align
);
1159 qemu_iovec_add(&local_qiov
, tail_buf
,
1160 align
- ((offset
+ bytes
) & (align
- 1)));
1162 bytes
= ROUND_UP(bytes
, align
);
1165 tracked_request_begin(&req
, bs
, offset
, bytes
, BDRV_TRACKED_READ
);
1166 ret
= bdrv_aligned_preadv(child
, &req
, offset
, bytes
, align
,
1167 use_local_qiov
? &local_qiov
: qiov
,
1169 tracked_request_end(&req
);
1170 bdrv_dec_in_flight(bs
);
1172 if (use_local_qiov
) {
1173 qemu_iovec_destroy(&local_qiov
);
1174 qemu_vfree(head_buf
);
1175 qemu_vfree(tail_buf
);
1181 static int coroutine_fn
bdrv_co_do_readv(BdrvChild
*child
,
1182 int64_t sector_num
, int nb_sectors
, QEMUIOVector
*qiov
,
1183 BdrvRequestFlags flags
)
1185 if (nb_sectors
< 0 || nb_sectors
> BDRV_REQUEST_MAX_SECTORS
) {
1189 return bdrv_co_preadv(child
, sector_num
<< BDRV_SECTOR_BITS
,
1190 nb_sectors
<< BDRV_SECTOR_BITS
, qiov
, flags
);
1193 int coroutine_fn
bdrv_co_readv(BdrvChild
*child
, int64_t sector_num
,
1194 int nb_sectors
, QEMUIOVector
*qiov
)
1196 trace_bdrv_co_readv(child
->bs
, sector_num
, nb_sectors
);
1198 return bdrv_co_do_readv(child
, sector_num
, nb_sectors
, qiov
, 0);
1201 /* Maximum buffer for write zeroes fallback, in bytes */
1202 #define MAX_WRITE_ZEROES_BOUNCE_BUFFER (32768 << BDRV_SECTOR_BITS)
1204 static int coroutine_fn
bdrv_co_do_pwrite_zeroes(BlockDriverState
*bs
,
1205 int64_t offset
, int count
, BdrvRequestFlags flags
)
1207 BlockDriver
*drv
= bs
->drv
;
1209 struct iovec iov
= {0};
1211 bool need_flush
= false;
1215 int max_write_zeroes
= MIN_NON_ZERO(bs
->bl
.max_pwrite_zeroes
, INT_MAX
);
1216 int alignment
= MAX(bs
->bl
.pwrite_zeroes_alignment
,
1217 bs
->bl
.request_alignment
);
1218 int max_transfer
= MIN_NON_ZERO(bs
->bl
.max_transfer
,
1219 MAX_WRITE_ZEROES_BOUNCE_BUFFER
);
1221 assert(alignment
% bs
->bl
.request_alignment
== 0);
1222 head
= offset
% alignment
;
1223 tail
= (offset
+ count
) % alignment
;
1224 max_write_zeroes
= QEMU_ALIGN_DOWN(max_write_zeroes
, alignment
);
1225 assert(max_write_zeroes
>= bs
->bl
.request_alignment
);
1227 while (count
> 0 && !ret
) {
1230 /* Align request. Block drivers can expect the "bulk" of the request
1231 * to be aligned, and that unaligned requests do not cross cluster
1235 /* Make a small request up to the first aligned sector. For
1236 * convenience, limit this request to max_transfer even if
1237 * we don't need to fall back to writes. */
1238 num
= MIN(MIN(count
, max_transfer
), alignment
- head
);
1239 head
= (head
+ num
) % alignment
;
1240 assert(num
< max_write_zeroes
);
1241 } else if (tail
&& num
> alignment
) {
1242 /* Shorten the request to the last aligned sector. */
1246 /* limit request size */
1247 if (num
> max_write_zeroes
) {
1248 num
= max_write_zeroes
;
1252 /* First try the efficient write zeroes operation */
1253 if (drv
->bdrv_co_pwrite_zeroes
) {
1254 ret
= drv
->bdrv_co_pwrite_zeroes(bs
, offset
, num
,
1255 flags
& bs
->supported_zero_flags
);
1256 if (ret
!= -ENOTSUP
&& (flags
& BDRV_REQ_FUA
) &&
1257 !(bs
->supported_zero_flags
& BDRV_REQ_FUA
)) {
1261 assert(!bs
->supported_zero_flags
);
1264 if (ret
== -ENOTSUP
) {
1265 /* Fall back to bounce buffer if write zeroes is unsupported */
1266 BdrvRequestFlags write_flags
= flags
& ~BDRV_REQ_ZERO_WRITE
;
1268 if ((flags
& BDRV_REQ_FUA
) &&
1269 !(bs
->supported_write_flags
& BDRV_REQ_FUA
)) {
1270 /* No need for bdrv_driver_pwrite() to do a fallback
1271 * flush on each chunk; use just one at the end */
1272 write_flags
&= ~BDRV_REQ_FUA
;
1275 num
= MIN(num
, max_transfer
);
1277 if (iov
.iov_base
== NULL
) {
1278 iov
.iov_base
= qemu_try_blockalign(bs
, num
);
1279 if (iov
.iov_base
== NULL
) {
1283 memset(iov
.iov_base
, 0, num
);
1285 qemu_iovec_init_external(&qiov
, &iov
, 1);
1287 ret
= bdrv_driver_pwritev(bs
, offset
, num
, &qiov
, write_flags
);
1289 /* Keep bounce buffer around if it is big enough for all
1290 * all future requests.
1292 if (num
< max_transfer
) {
1293 qemu_vfree(iov
.iov_base
);
1294 iov
.iov_base
= NULL
;
1303 if (ret
== 0 && need_flush
) {
1304 ret
= bdrv_co_flush(bs
);
1306 qemu_vfree(iov
.iov_base
);
1311 * Forwards an already correctly aligned write request to the BlockDriver,
1312 * after possibly fragmenting it.
1314 static int coroutine_fn
bdrv_aligned_pwritev(BdrvChild
*child
,
1315 BdrvTrackedRequest
*req
, int64_t offset
, unsigned int bytes
,
1316 int64_t align
, QEMUIOVector
*qiov
, int flags
)
1318 BlockDriverState
*bs
= child
->bs
;
1319 BlockDriver
*drv
= bs
->drv
;
1323 int64_t start_sector
= offset
>> BDRV_SECTOR_BITS
;
1324 int64_t end_sector
= DIV_ROUND_UP(offset
+ bytes
, BDRV_SECTOR_SIZE
);
1325 uint64_t bytes_remaining
= bytes
;
1328 assert(is_power_of_2(align
));
1329 assert((offset
& (align
- 1)) == 0);
1330 assert((bytes
& (align
- 1)) == 0);
1331 assert(!qiov
|| bytes
== qiov
->size
);
1332 assert((bs
->open_flags
& BDRV_O_NO_IO
) == 0);
1333 assert(!(flags
& ~BDRV_REQ_MASK
));
1334 max_transfer
= QEMU_ALIGN_DOWN(MIN_NON_ZERO(bs
->bl
.max_transfer
, INT_MAX
),
1337 waited
= wait_serialising_requests(req
);
1338 assert(!waited
|| !req
->serialising
);
1339 assert(req
->overlap_offset
<= offset
);
1340 assert(offset
+ bytes
<= req
->overlap_offset
+ req
->overlap_bytes
);
1341 assert(child
->perm
& BLK_PERM_WRITE
);
1343 ret
= notifier_with_return_list_notify(&bs
->before_write_notifiers
, req
);
1345 if (!ret
&& bs
->detect_zeroes
!= BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF
&&
1346 !(flags
& BDRV_REQ_ZERO_WRITE
) && drv
->bdrv_co_pwrite_zeroes
&&
1347 qemu_iovec_is_zero(qiov
)) {
1348 flags
|= BDRV_REQ_ZERO_WRITE
;
1349 if (bs
->detect_zeroes
== BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP
) {
1350 flags
|= BDRV_REQ_MAY_UNMAP
;
1355 /* Do nothing, write notifier decided to fail this request */
1356 } else if (flags
& BDRV_REQ_ZERO_WRITE
) {
1357 bdrv_debug_event(bs
, BLKDBG_PWRITEV_ZERO
);
1358 ret
= bdrv_co_do_pwrite_zeroes(bs
, offset
, bytes
, flags
);
1359 } else if (flags
& BDRV_REQ_WRITE_COMPRESSED
) {
1360 ret
= bdrv_driver_pwritev_compressed(bs
, offset
, bytes
, qiov
);
1361 } else if (bytes
<= max_transfer
) {
1362 bdrv_debug_event(bs
, BLKDBG_PWRITEV
);
1363 ret
= bdrv_driver_pwritev(bs
, offset
, bytes
, qiov
, flags
);
1365 bdrv_debug_event(bs
, BLKDBG_PWRITEV
);
1366 while (bytes_remaining
) {
1367 int num
= MIN(bytes_remaining
, max_transfer
);
1368 QEMUIOVector local_qiov
;
1369 int local_flags
= flags
;
1372 if (num
< bytes_remaining
&& (flags
& BDRV_REQ_FUA
) &&
1373 !(bs
->supported_write_flags
& BDRV_REQ_FUA
)) {
1374 /* If FUA is going to be emulated by flush, we only
1375 * need to flush on the last iteration */
1376 local_flags
&= ~BDRV_REQ_FUA
;
1378 qemu_iovec_init(&local_qiov
, qiov
->niov
);
1379 qemu_iovec_concat(&local_qiov
, qiov
, bytes
- bytes_remaining
, num
);
1381 ret
= bdrv_driver_pwritev(bs
, offset
+ bytes
- bytes_remaining
,
1382 num
, &local_qiov
, local_flags
);
1383 qemu_iovec_destroy(&local_qiov
);
1387 bytes_remaining
-= num
;
1390 bdrv_debug_event(bs
, BLKDBG_PWRITEV_DONE
);
1393 bdrv_set_dirty(bs
, start_sector
, end_sector
- start_sector
);
1395 if (bs
->wr_highest_offset
< offset
+ bytes
) {
1396 bs
->wr_highest_offset
= offset
+ bytes
;
1400 bs
->total_sectors
= MAX(bs
->total_sectors
, end_sector
);
1407 static int coroutine_fn
bdrv_co_do_zero_pwritev(BdrvChild
*child
,
1410 BdrvRequestFlags flags
,
1411 BdrvTrackedRequest
*req
)
1413 BlockDriverState
*bs
= child
->bs
;
1414 uint8_t *buf
= NULL
;
1415 QEMUIOVector local_qiov
;
1417 uint64_t align
= bs
->bl
.request_alignment
;
1418 unsigned int head_padding_bytes
, tail_padding_bytes
;
1421 head_padding_bytes
= offset
& (align
- 1);
1422 tail_padding_bytes
= align
- ((offset
+ bytes
) & (align
- 1));
1425 assert(flags
& BDRV_REQ_ZERO_WRITE
);
1426 if (head_padding_bytes
|| tail_padding_bytes
) {
1427 buf
= qemu_blockalign(bs
, align
);
1428 iov
= (struct iovec
) {
1432 qemu_iovec_init_external(&local_qiov
, &iov
, 1);
1434 if (head_padding_bytes
) {
1435 uint64_t zero_bytes
= MIN(bytes
, align
- head_padding_bytes
);
1437 /* RMW the unaligned part before head. */
1438 mark_request_serialising(req
, align
);
1439 wait_serialising_requests(req
);
1440 bdrv_debug_event(bs
, BLKDBG_PWRITEV_RMW_HEAD
);
1441 ret
= bdrv_aligned_preadv(child
, req
, offset
& ~(align
- 1), align
,
1442 align
, &local_qiov
, 0);
1446 bdrv_debug_event(bs
, BLKDBG_PWRITEV_RMW_AFTER_HEAD
);
1448 memset(buf
+ head_padding_bytes
, 0, zero_bytes
);
1449 ret
= bdrv_aligned_pwritev(child
, req
, offset
& ~(align
- 1), align
,
1451 flags
& ~BDRV_REQ_ZERO_WRITE
);
1455 offset
+= zero_bytes
;
1456 bytes
-= zero_bytes
;
1459 assert(!bytes
|| (offset
& (align
- 1)) == 0);
1460 if (bytes
>= align
) {
1461 /* Write the aligned part in the middle. */
1462 uint64_t aligned_bytes
= bytes
& ~(align
- 1);
1463 ret
= bdrv_aligned_pwritev(child
, req
, offset
, aligned_bytes
, align
,
1468 bytes
-= aligned_bytes
;
1469 offset
+= aligned_bytes
;
1472 assert(!bytes
|| (offset
& (align
- 1)) == 0);
1474 assert(align
== tail_padding_bytes
+ bytes
);
1475 /* RMW the unaligned part after tail. */
1476 mark_request_serialising(req
, align
);
1477 wait_serialising_requests(req
);
1478 bdrv_debug_event(bs
, BLKDBG_PWRITEV_RMW_TAIL
);
1479 ret
= bdrv_aligned_preadv(child
, req
, offset
, align
,
1480 align
, &local_qiov
, 0);
1484 bdrv_debug_event(bs
, BLKDBG_PWRITEV_RMW_AFTER_TAIL
);
1486 memset(buf
, 0, bytes
);
1487 ret
= bdrv_aligned_pwritev(child
, req
, offset
, align
, align
,
1488 &local_qiov
, flags
& ~BDRV_REQ_ZERO_WRITE
);
1497 * Handle a write request in coroutine context
1499 int coroutine_fn
bdrv_co_pwritev(BdrvChild
*child
,
1500 int64_t offset
, unsigned int bytes
, QEMUIOVector
*qiov
,
1501 BdrvRequestFlags flags
)
1503 BlockDriverState
*bs
= child
->bs
;
1504 BdrvTrackedRequest req
;
1505 uint64_t align
= bs
->bl
.request_alignment
;
1506 uint8_t *head_buf
= NULL
;
1507 uint8_t *tail_buf
= NULL
;
1508 QEMUIOVector local_qiov
;
1509 bool use_local_qiov
= false;
1515 if (bs
->read_only
) {
1518 assert(!(bs
->open_flags
& BDRV_O_INACTIVE
));
1520 ret
= bdrv_check_byte_request(bs
, offset
, bytes
);
1525 bdrv_inc_in_flight(bs
);
1527 * Align write if necessary by performing a read-modify-write cycle.
1528 * Pad qiov with the read parts and be sure to have a tracked request not
1529 * only for bdrv_aligned_pwritev, but also for the reads of the RMW cycle.
1531 tracked_request_begin(&req
, bs
, offset
, bytes
, BDRV_TRACKED_WRITE
);
1534 ret
= bdrv_co_do_zero_pwritev(child
, offset
, bytes
, flags
, &req
);
1538 if (offset
& (align
- 1)) {
1539 QEMUIOVector head_qiov
;
1540 struct iovec head_iov
;
1542 mark_request_serialising(&req
, align
);
1543 wait_serialising_requests(&req
);
1545 head_buf
= qemu_blockalign(bs
, align
);
1546 head_iov
= (struct iovec
) {
1547 .iov_base
= head_buf
,
1550 qemu_iovec_init_external(&head_qiov
, &head_iov
, 1);
1552 bdrv_debug_event(bs
, BLKDBG_PWRITEV_RMW_HEAD
);
1553 ret
= bdrv_aligned_preadv(child
, &req
, offset
& ~(align
- 1), align
,
1554 align
, &head_qiov
, 0);
1558 bdrv_debug_event(bs
, BLKDBG_PWRITEV_RMW_AFTER_HEAD
);
1560 qemu_iovec_init(&local_qiov
, qiov
->niov
+ 2);
1561 qemu_iovec_add(&local_qiov
, head_buf
, offset
& (align
- 1));
1562 qemu_iovec_concat(&local_qiov
, qiov
, 0, qiov
->size
);
1563 use_local_qiov
= true;
1565 bytes
+= offset
& (align
- 1);
1566 offset
= offset
& ~(align
- 1);
1568 /* We have read the tail already if the request is smaller
1569 * than one aligned block.
1571 if (bytes
< align
) {
1572 qemu_iovec_add(&local_qiov
, head_buf
+ bytes
, align
- bytes
);
1577 if ((offset
+ bytes
) & (align
- 1)) {
1578 QEMUIOVector tail_qiov
;
1579 struct iovec tail_iov
;
1583 mark_request_serialising(&req
, align
);
1584 waited
= wait_serialising_requests(&req
);
1585 assert(!waited
|| !use_local_qiov
);
1587 tail_buf
= qemu_blockalign(bs
, align
);
1588 tail_iov
= (struct iovec
) {
1589 .iov_base
= tail_buf
,
1592 qemu_iovec_init_external(&tail_qiov
, &tail_iov
, 1);
1594 bdrv_debug_event(bs
, BLKDBG_PWRITEV_RMW_TAIL
);
1595 ret
= bdrv_aligned_preadv(child
, &req
, (offset
+ bytes
) & ~(align
- 1),
1596 align
, align
, &tail_qiov
, 0);
1600 bdrv_debug_event(bs
, BLKDBG_PWRITEV_RMW_AFTER_TAIL
);
1602 if (!use_local_qiov
) {
1603 qemu_iovec_init(&local_qiov
, qiov
->niov
+ 1);
1604 qemu_iovec_concat(&local_qiov
, qiov
, 0, qiov
->size
);
1605 use_local_qiov
= true;
1608 tail_bytes
= (offset
+ bytes
) & (align
- 1);
1609 qemu_iovec_add(&local_qiov
, tail_buf
+ tail_bytes
, align
- tail_bytes
);
1611 bytes
= ROUND_UP(bytes
, align
);
1614 ret
= bdrv_aligned_pwritev(child
, &req
, offset
, bytes
, align
,
1615 use_local_qiov
? &local_qiov
: qiov
,
1620 if (use_local_qiov
) {
1621 qemu_iovec_destroy(&local_qiov
);
1623 qemu_vfree(head_buf
);
1624 qemu_vfree(tail_buf
);
1626 tracked_request_end(&req
);
1627 bdrv_dec_in_flight(bs
);
1631 static int coroutine_fn
bdrv_co_do_writev(BdrvChild
*child
,
1632 int64_t sector_num
, int nb_sectors
, QEMUIOVector
*qiov
,
1633 BdrvRequestFlags flags
)
1635 if (nb_sectors
< 0 || nb_sectors
> BDRV_REQUEST_MAX_SECTORS
) {
1639 return bdrv_co_pwritev(child
, sector_num
<< BDRV_SECTOR_BITS
,
1640 nb_sectors
<< BDRV_SECTOR_BITS
, qiov
, flags
);
1643 int coroutine_fn
bdrv_co_writev(BdrvChild
*child
, int64_t sector_num
,
1644 int nb_sectors
, QEMUIOVector
*qiov
)
1646 trace_bdrv_co_writev(child
->bs
, sector_num
, nb_sectors
);
1648 return bdrv_co_do_writev(child
, sector_num
, nb_sectors
, qiov
, 0);
1651 int coroutine_fn
bdrv_co_pwrite_zeroes(BdrvChild
*child
, int64_t offset
,
1652 int count
, BdrvRequestFlags flags
)
1654 trace_bdrv_co_pwrite_zeroes(child
->bs
, offset
, count
, flags
);
1656 if (!(child
->bs
->open_flags
& BDRV_O_UNMAP
)) {
1657 flags
&= ~BDRV_REQ_MAY_UNMAP
;
1660 return bdrv_co_pwritev(child
, offset
, count
, NULL
,
1661 BDRV_REQ_ZERO_WRITE
| flags
);
1665 * Flush ALL BDSes regardless of if they are reachable via a BlkBackend or not.
1667 int bdrv_flush_all(void)
1669 BdrvNextIterator it
;
1670 BlockDriverState
*bs
= NULL
;
1673 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
1674 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
1677 aio_context_acquire(aio_context
);
1678 ret
= bdrv_flush(bs
);
1679 if (ret
< 0 && !result
) {
1682 aio_context_release(aio_context
);
1689 typedef struct BdrvCoGetBlockStatusData
{
1690 BlockDriverState
*bs
;
1691 BlockDriverState
*base
;
1692 BlockDriverState
**file
;
1698 } BdrvCoGetBlockStatusData
;
1701 * Returns the allocation status of the specified sectors.
1702 * Drivers not implementing the functionality are assumed to not support
1703 * backing files, hence all their sectors are reported as allocated.
1705 * If 'sector_num' is beyond the end of the disk image the return value is 0
1706 * and 'pnum' is set to 0.
1708 * 'pnum' is set to the number of sectors (including and immediately following
1709 * the specified sector) that are known to be in the same
1710 * allocated/unallocated state.
1712 * 'nb_sectors' is the max value 'pnum' should be set to. If nb_sectors goes
1713 * beyond the end of the disk image it will be clamped.
1715 * If returned value is positive and BDRV_BLOCK_OFFSET_VALID bit is set, 'file'
1716 * points to the BDS which the sector range is allocated in.
1718 static int64_t coroutine_fn
bdrv_co_get_block_status(BlockDriverState
*bs
,
1720 int nb_sectors
, int *pnum
,
1721 BlockDriverState
**file
)
1723 int64_t total_sectors
;
1727 total_sectors
= bdrv_nb_sectors(bs
);
1728 if (total_sectors
< 0) {
1729 return total_sectors
;
1732 if (sector_num
>= total_sectors
) {
1737 n
= total_sectors
- sector_num
;
1738 if (n
< nb_sectors
) {
1742 if (!bs
->drv
->bdrv_co_get_block_status
) {
1744 ret
= BDRV_BLOCK_DATA
| BDRV_BLOCK_ALLOCATED
;
1745 if (bs
->drv
->protocol_name
) {
1746 ret
|= BDRV_BLOCK_OFFSET_VALID
| (sector_num
* BDRV_SECTOR_SIZE
);
1752 bdrv_inc_in_flight(bs
);
1753 ret
= bs
->drv
->bdrv_co_get_block_status(bs
, sector_num
, nb_sectors
, pnum
,
1760 if (ret
& BDRV_BLOCK_RAW
) {
1761 assert(ret
& BDRV_BLOCK_OFFSET_VALID
);
1762 ret
= bdrv_get_block_status(bs
->file
->bs
, ret
>> BDRV_SECTOR_BITS
,
1767 if (ret
& (BDRV_BLOCK_DATA
| BDRV_BLOCK_ZERO
)) {
1768 ret
|= BDRV_BLOCK_ALLOCATED
;
1770 if (bdrv_unallocated_blocks_are_zero(bs
)) {
1771 ret
|= BDRV_BLOCK_ZERO
;
1772 } else if (bs
->backing
) {
1773 BlockDriverState
*bs2
= bs
->backing
->bs
;
1774 int64_t nb_sectors2
= bdrv_nb_sectors(bs2
);
1775 if (nb_sectors2
>= 0 && sector_num
>= nb_sectors2
) {
1776 ret
|= BDRV_BLOCK_ZERO
;
1781 if (*file
&& *file
!= bs
&&
1782 (ret
& BDRV_BLOCK_DATA
) && !(ret
& BDRV_BLOCK_ZERO
) &&
1783 (ret
& BDRV_BLOCK_OFFSET_VALID
)) {
1784 BlockDriverState
*file2
;
1787 ret2
= bdrv_co_get_block_status(*file
, ret
>> BDRV_SECTOR_BITS
,
1788 *pnum
, &file_pnum
, &file2
);
1790 /* Ignore errors. This is just providing extra information, it
1791 * is useful but not necessary.
1794 /* !file_pnum indicates an offset at or beyond the EOF; it is
1795 * perfectly valid for the format block driver to point to such
1796 * offsets, so catch it and mark everything as zero */
1797 ret
|= BDRV_BLOCK_ZERO
;
1799 /* Limit request to the range reported by the protocol driver */
1801 ret
|= (ret2
& BDRV_BLOCK_ZERO
);
1807 bdrv_dec_in_flight(bs
);
1811 static int64_t coroutine_fn
bdrv_co_get_block_status_above(BlockDriverState
*bs
,
1812 BlockDriverState
*base
,
1816 BlockDriverState
**file
)
1818 BlockDriverState
*p
;
1822 for (p
= bs
; p
!= base
; p
= backing_bs(p
)) {
1823 ret
= bdrv_co_get_block_status(p
, sector_num
, nb_sectors
, pnum
, file
);
1824 if (ret
< 0 || ret
& BDRV_BLOCK_ALLOCATED
) {
1827 /* [sector_num, pnum] unallocated on this layer, which could be only
1828 * the first part of [sector_num, nb_sectors]. */
1829 nb_sectors
= MIN(nb_sectors
, *pnum
);
1834 /* Coroutine wrapper for bdrv_get_block_status_above() */
1835 static void coroutine_fn
bdrv_get_block_status_above_co_entry(void *opaque
)
1837 BdrvCoGetBlockStatusData
*data
= opaque
;
1839 data
->ret
= bdrv_co_get_block_status_above(data
->bs
, data
->base
,
1848 * Synchronous wrapper around bdrv_co_get_block_status_above().
1850 * See bdrv_co_get_block_status_above() for details.
1852 int64_t bdrv_get_block_status_above(BlockDriverState
*bs
,
1853 BlockDriverState
*base
,
1855 int nb_sectors
, int *pnum
,
1856 BlockDriverState
**file
)
1859 BdrvCoGetBlockStatusData data
= {
1863 .sector_num
= sector_num
,
1864 .nb_sectors
= nb_sectors
,
1869 if (qemu_in_coroutine()) {
1870 /* Fast-path if already in coroutine context */
1871 bdrv_get_block_status_above_co_entry(&data
);
1873 co
= qemu_coroutine_create(bdrv_get_block_status_above_co_entry
,
1875 qemu_coroutine_enter(co
);
1876 BDRV_POLL_WHILE(bs
, !data
.done
);
1881 int64_t bdrv_get_block_status(BlockDriverState
*bs
,
1883 int nb_sectors
, int *pnum
,
1884 BlockDriverState
**file
)
1886 return bdrv_get_block_status_above(bs
, backing_bs(bs
),
1887 sector_num
, nb_sectors
, pnum
, file
);
1890 int coroutine_fn
bdrv_is_allocated(BlockDriverState
*bs
, int64_t sector_num
,
1891 int nb_sectors
, int *pnum
)
1893 BlockDriverState
*file
;
1894 int64_t ret
= bdrv_get_block_status(bs
, sector_num
, nb_sectors
, pnum
,
1899 return !!(ret
& BDRV_BLOCK_ALLOCATED
);
1903 * Given an image chain: ... -> [BASE] -> [INTER1] -> [INTER2] -> [TOP]
1905 * Return true if the given sector is allocated in any image between
1906 * BASE and TOP (inclusive). BASE can be NULL to check if the given
1907 * sector is allocated in any image of the chain. Return false otherwise.
1909 * 'pnum' is set to the number of sectors (including and immediately following
1910 * the specified sector) that are known to be in the same
1911 * allocated/unallocated state.
1914 int bdrv_is_allocated_above(BlockDriverState
*top
,
1915 BlockDriverState
*base
,
1917 int nb_sectors
, int *pnum
)
1919 BlockDriverState
*intermediate
;
1920 int ret
, n
= nb_sectors
;
1923 while (intermediate
&& intermediate
!= base
) {
1925 ret
= bdrv_is_allocated(intermediate
, sector_num
, nb_sectors
,
1935 * [sector_num, nb_sectors] is unallocated on top but intermediate
1938 * [sector_num+x, nr_sectors] allocated.
1940 if (n
> pnum_inter
&&
1941 (intermediate
== top
||
1942 sector_num
+ pnum_inter
< intermediate
->total_sectors
)) {
1946 intermediate
= backing_bs(intermediate
);
1953 typedef struct BdrvVmstateCo
{
1954 BlockDriverState
*bs
;
1961 static int coroutine_fn
1962 bdrv_co_rw_vmstate(BlockDriverState
*bs
, QEMUIOVector
*qiov
, int64_t pos
,
1965 BlockDriver
*drv
= bs
->drv
;
1969 } else if (drv
->bdrv_load_vmstate
) {
1970 return is_read
? drv
->bdrv_load_vmstate(bs
, qiov
, pos
)
1971 : drv
->bdrv_save_vmstate(bs
, qiov
, pos
);
1972 } else if (bs
->file
) {
1973 return bdrv_co_rw_vmstate(bs
->file
->bs
, qiov
, pos
, is_read
);
1979 static void coroutine_fn
bdrv_co_rw_vmstate_entry(void *opaque
)
1981 BdrvVmstateCo
*co
= opaque
;
1982 co
->ret
= bdrv_co_rw_vmstate(co
->bs
, co
->qiov
, co
->pos
, co
->is_read
);
1986 bdrv_rw_vmstate(BlockDriverState
*bs
, QEMUIOVector
*qiov
, int64_t pos
,
1989 if (qemu_in_coroutine()) {
1990 return bdrv_co_rw_vmstate(bs
, qiov
, pos
, is_read
);
1992 BdrvVmstateCo data
= {
1997 .ret
= -EINPROGRESS
,
1999 Coroutine
*co
= qemu_coroutine_create(bdrv_co_rw_vmstate_entry
, &data
);
2001 qemu_coroutine_enter(co
);
2002 while (data
.ret
== -EINPROGRESS
) {
2003 aio_poll(bdrv_get_aio_context(bs
), true);
2009 int bdrv_save_vmstate(BlockDriverState
*bs
, const uint8_t *buf
,
2010 int64_t pos
, int size
)
2013 struct iovec iov
= {
2014 .iov_base
= (void *) buf
,
2019 qemu_iovec_init_external(&qiov
, &iov
, 1);
2021 ret
= bdrv_writev_vmstate(bs
, &qiov
, pos
);
2029 int bdrv_writev_vmstate(BlockDriverState
*bs
, QEMUIOVector
*qiov
, int64_t pos
)
2031 return bdrv_rw_vmstate(bs
, qiov
, pos
, false);
2034 int bdrv_load_vmstate(BlockDriverState
*bs
, uint8_t *buf
,
2035 int64_t pos
, int size
)
2038 struct iovec iov
= {
2044 qemu_iovec_init_external(&qiov
, &iov
, 1);
2045 ret
= bdrv_readv_vmstate(bs
, &qiov
, pos
);
2053 int bdrv_readv_vmstate(BlockDriverState
*bs
, QEMUIOVector
*qiov
, int64_t pos
)
2055 return bdrv_rw_vmstate(bs
, qiov
, pos
, true);
2058 /**************************************************************/
2061 BlockAIOCB
*bdrv_aio_readv(BdrvChild
*child
, int64_t sector_num
,
2062 QEMUIOVector
*qiov
, int nb_sectors
,
2063 BlockCompletionFunc
*cb
, void *opaque
)
2065 trace_bdrv_aio_readv(child
->bs
, sector_num
, nb_sectors
, opaque
);
2067 assert(nb_sectors
<< BDRV_SECTOR_BITS
== qiov
->size
);
2068 return bdrv_co_aio_prw_vector(child
, sector_num
<< BDRV_SECTOR_BITS
, qiov
,
2069 0, cb
, opaque
, false);
2072 BlockAIOCB
*bdrv_aio_writev(BdrvChild
*child
, int64_t sector_num
,
2073 QEMUIOVector
*qiov
, int nb_sectors
,
2074 BlockCompletionFunc
*cb
, void *opaque
)
2076 trace_bdrv_aio_writev(child
->bs
, sector_num
, nb_sectors
, opaque
);
2078 assert(nb_sectors
<< BDRV_SECTOR_BITS
== qiov
->size
);
2079 return bdrv_co_aio_prw_vector(child
, sector_num
<< BDRV_SECTOR_BITS
, qiov
,
2080 0, cb
, opaque
, true);
2083 void bdrv_aio_cancel(BlockAIOCB
*acb
)
2086 bdrv_aio_cancel_async(acb
);
2087 while (acb
->refcnt
> 1) {
2088 if (acb
->aiocb_info
->get_aio_context
) {
2089 aio_poll(acb
->aiocb_info
->get_aio_context(acb
), true);
2090 } else if (acb
->bs
) {
2091 /* qemu_aio_ref and qemu_aio_unref are not thread-safe, so
2092 * assert that we're not using an I/O thread. Thread-safe
2093 * code should use bdrv_aio_cancel_async exclusively.
2095 assert(bdrv_get_aio_context(acb
->bs
) == qemu_get_aio_context());
2096 aio_poll(bdrv_get_aio_context(acb
->bs
), true);
2101 qemu_aio_unref(acb
);
2104 /* Async version of aio cancel. The caller is not blocked if the acb implements
2105 * cancel_async, otherwise we do nothing and let the request normally complete.
2106 * In either case the completion callback must be called. */
2107 void bdrv_aio_cancel_async(BlockAIOCB
*acb
)
2109 if (acb
->aiocb_info
->cancel_async
) {
2110 acb
->aiocb_info
->cancel_async(acb
);
2114 /**************************************************************/
2115 /* async block device emulation */
2117 typedef struct BlockRequest
{
2119 /* Used during read, write, trim */
2126 /* Used during ioctl */
2132 BlockCompletionFunc
*cb
;
2138 typedef struct BlockAIOCBCoroutine
{
2145 } BlockAIOCBCoroutine
;
2147 static const AIOCBInfo bdrv_em_co_aiocb_info
= {
2148 .aiocb_size
= sizeof(BlockAIOCBCoroutine
),
2151 static void bdrv_co_complete(BlockAIOCBCoroutine
*acb
)
2153 if (!acb
->need_bh
) {
2154 bdrv_dec_in_flight(acb
->common
.bs
);
2155 acb
->common
.cb(acb
->common
.opaque
, acb
->req
.error
);
2156 qemu_aio_unref(acb
);
2160 static void bdrv_co_em_bh(void *opaque
)
2162 BlockAIOCBCoroutine
*acb
= opaque
;
2164 assert(!acb
->need_bh
);
2165 bdrv_co_complete(acb
);
2168 static void bdrv_co_maybe_schedule_bh(BlockAIOCBCoroutine
*acb
)
2170 acb
->need_bh
= false;
2171 if (acb
->req
.error
!= -EINPROGRESS
) {
2172 BlockDriverState
*bs
= acb
->common
.bs
;
2174 aio_bh_schedule_oneshot(bdrv_get_aio_context(bs
), bdrv_co_em_bh
, acb
);
2178 /* Invoke bdrv_co_do_readv/bdrv_co_do_writev */
2179 static void coroutine_fn
bdrv_co_do_rw(void *opaque
)
2181 BlockAIOCBCoroutine
*acb
= opaque
;
2183 if (!acb
->is_write
) {
2184 acb
->req
.error
= bdrv_co_preadv(acb
->child
, acb
->req
.offset
,
2185 acb
->req
.qiov
->size
, acb
->req
.qiov
, acb
->req
.flags
);
2187 acb
->req
.error
= bdrv_co_pwritev(acb
->child
, acb
->req
.offset
,
2188 acb
->req
.qiov
->size
, acb
->req
.qiov
, acb
->req
.flags
);
2191 bdrv_co_complete(acb
);
2194 static BlockAIOCB
*bdrv_co_aio_prw_vector(BdrvChild
*child
,
2197 BdrvRequestFlags flags
,
2198 BlockCompletionFunc
*cb
,
2203 BlockAIOCBCoroutine
*acb
;
2205 /* Matched by bdrv_co_complete's bdrv_dec_in_flight. */
2206 bdrv_inc_in_flight(child
->bs
);
2208 acb
= qemu_aio_get(&bdrv_em_co_aiocb_info
, child
->bs
, cb
, opaque
);
2210 acb
->need_bh
= true;
2211 acb
->req
.error
= -EINPROGRESS
;
2212 acb
->req
.offset
= offset
;
2213 acb
->req
.qiov
= qiov
;
2214 acb
->req
.flags
= flags
;
2215 acb
->is_write
= is_write
;
2217 co
= qemu_coroutine_create(bdrv_co_do_rw
, acb
);
2218 qemu_coroutine_enter(co
);
2220 bdrv_co_maybe_schedule_bh(acb
);
2221 return &acb
->common
;
2224 static void coroutine_fn
bdrv_aio_flush_co_entry(void *opaque
)
2226 BlockAIOCBCoroutine
*acb
= opaque
;
2227 BlockDriverState
*bs
= acb
->common
.bs
;
2229 acb
->req
.error
= bdrv_co_flush(bs
);
2230 bdrv_co_complete(acb
);
2233 BlockAIOCB
*bdrv_aio_flush(BlockDriverState
*bs
,
2234 BlockCompletionFunc
*cb
, void *opaque
)
2236 trace_bdrv_aio_flush(bs
, opaque
);
2239 BlockAIOCBCoroutine
*acb
;
2241 /* Matched by bdrv_co_complete's bdrv_dec_in_flight. */
2242 bdrv_inc_in_flight(bs
);
2244 acb
= qemu_aio_get(&bdrv_em_co_aiocb_info
, bs
, cb
, opaque
);
2245 acb
->need_bh
= true;
2246 acb
->req
.error
= -EINPROGRESS
;
2248 co
= qemu_coroutine_create(bdrv_aio_flush_co_entry
, acb
);
2249 qemu_coroutine_enter(co
);
2251 bdrv_co_maybe_schedule_bh(acb
);
2252 return &acb
->common
;
2255 /**************************************************************/
2256 /* Coroutine block device emulation */
2258 typedef struct FlushCo
{
2259 BlockDriverState
*bs
;
2264 static void coroutine_fn
bdrv_flush_co_entry(void *opaque
)
2266 FlushCo
*rwco
= opaque
;
2268 rwco
->ret
= bdrv_co_flush(rwco
->bs
);
2271 int coroutine_fn
bdrv_co_flush(BlockDriverState
*bs
)
2275 if (!bs
|| !bdrv_is_inserted(bs
) || bdrv_is_read_only(bs
) ||
2280 bdrv_inc_in_flight(bs
);
2282 int current_gen
= bs
->write_gen
;
2284 /* Wait until any previous flushes are completed */
2285 while (bs
->active_flush_req
) {
2286 qemu_co_queue_wait(&bs
->flush_queue
, NULL
);
2289 bs
->active_flush_req
= true;
2291 /* Write back all layers by calling one driver function */
2292 if (bs
->drv
->bdrv_co_flush
) {
2293 ret
= bs
->drv
->bdrv_co_flush(bs
);
2297 /* Write back cached data to the OS even with cache=unsafe */
2298 BLKDBG_EVENT(bs
->file
, BLKDBG_FLUSH_TO_OS
);
2299 if (bs
->drv
->bdrv_co_flush_to_os
) {
2300 ret
= bs
->drv
->bdrv_co_flush_to_os(bs
);
2306 /* But don't actually force it to the disk with cache=unsafe */
2307 if (bs
->open_flags
& BDRV_O_NO_FLUSH
) {
2311 /* Check if we really need to flush anything */
2312 if (bs
->flushed_gen
== current_gen
) {
2316 BLKDBG_EVENT(bs
->file
, BLKDBG_FLUSH_TO_DISK
);
2317 if (bs
->drv
->bdrv_co_flush_to_disk
) {
2318 ret
= bs
->drv
->bdrv_co_flush_to_disk(bs
);
2319 } else if (bs
->drv
->bdrv_aio_flush
) {
2321 CoroutineIOCompletion co
= {
2322 .coroutine
= qemu_coroutine_self(),
2325 acb
= bs
->drv
->bdrv_aio_flush(bs
, bdrv_co_io_em_complete
, &co
);
2329 qemu_coroutine_yield();
2334 * Some block drivers always operate in either writethrough or unsafe
2335 * mode and don't support bdrv_flush therefore. Usually qemu doesn't
2336 * know how the server works (because the behaviour is hardcoded or
2337 * depends on server-side configuration), so we can't ensure that
2338 * everything is safe on disk. Returning an error doesn't work because
2339 * that would break guests even if the server operates in writethrough
2342 * Let's hope the user knows what he's doing.
2351 /* Now flush the underlying protocol. It will also have BDRV_O_NO_FLUSH
2352 * in the case of cache=unsafe, so there are no useless flushes.
2355 ret
= bs
->file
? bdrv_co_flush(bs
->file
->bs
) : 0;
2357 /* Notify any pending flushes that we have completed */
2359 bs
->flushed_gen
= current_gen
;
2361 bs
->active_flush_req
= false;
2362 /* Return value is ignored - it's ok if wait queue is empty */
2363 qemu_co_queue_next(&bs
->flush_queue
);
2365 bdrv_dec_in_flight(bs
);
2369 int bdrv_flush(BlockDriverState
*bs
)
2372 FlushCo flush_co
= {
2377 if (qemu_in_coroutine()) {
2378 /* Fast-path if already in coroutine context */
2379 bdrv_flush_co_entry(&flush_co
);
2381 co
= qemu_coroutine_create(bdrv_flush_co_entry
, &flush_co
);
2382 qemu_coroutine_enter(co
);
2383 BDRV_POLL_WHILE(bs
, flush_co
.ret
== NOT_DONE
);
2386 return flush_co
.ret
;
2389 typedef struct DiscardCo
{
2390 BlockDriverState
*bs
;
2395 static void coroutine_fn
bdrv_pdiscard_co_entry(void *opaque
)
2397 DiscardCo
*rwco
= opaque
;
2399 rwco
->ret
= bdrv_co_pdiscard(rwco
->bs
, rwco
->offset
, rwco
->count
);
2402 int coroutine_fn
bdrv_co_pdiscard(BlockDriverState
*bs
, int64_t offset
,
2405 BdrvTrackedRequest req
;
2406 int max_pdiscard
, ret
;
2407 int head
, tail
, align
;
2413 ret
= bdrv_check_byte_request(bs
, offset
, count
);
2416 } else if (bs
->read_only
) {
2419 assert(!(bs
->open_flags
& BDRV_O_INACTIVE
));
2421 /* Do nothing if disabled. */
2422 if (!(bs
->open_flags
& BDRV_O_UNMAP
)) {
2426 if (!bs
->drv
->bdrv_co_pdiscard
&& !bs
->drv
->bdrv_aio_pdiscard
) {
2430 /* Discard is advisory, but some devices track and coalesce
2431 * unaligned requests, so we must pass everything down rather than
2432 * round here. Still, most devices will just silently ignore
2433 * unaligned requests (by returning -ENOTSUP), so we must fragment
2434 * the request accordingly. */
2435 align
= MAX(bs
->bl
.pdiscard_alignment
, bs
->bl
.request_alignment
);
2436 assert(align
% bs
->bl
.request_alignment
== 0);
2437 head
= offset
% align
;
2438 tail
= (offset
+ count
) % align
;
2440 bdrv_inc_in_flight(bs
);
2441 tracked_request_begin(&req
, bs
, offset
, count
, BDRV_TRACKED_DISCARD
);
2443 ret
= notifier_with_return_list_notify(&bs
->before_write_notifiers
, &req
);
2448 max_pdiscard
= QEMU_ALIGN_DOWN(MIN_NON_ZERO(bs
->bl
.max_pdiscard
, INT_MAX
),
2450 assert(max_pdiscard
>= bs
->bl
.request_alignment
);
2457 /* Make small requests to get to alignment boundaries. */
2458 num
= MIN(count
, align
- head
);
2459 if (!QEMU_IS_ALIGNED(num
, bs
->bl
.request_alignment
)) {
2460 num
%= bs
->bl
.request_alignment
;
2462 head
= (head
+ num
) % align
;
2463 assert(num
< max_pdiscard
);
2466 /* Shorten the request to the last aligned cluster. */
2468 } else if (!QEMU_IS_ALIGNED(tail
, bs
->bl
.request_alignment
) &&
2469 tail
> bs
->bl
.request_alignment
) {
2470 tail
%= bs
->bl
.request_alignment
;
2474 /* limit request size */
2475 if (num
> max_pdiscard
) {
2479 if (bs
->drv
->bdrv_co_pdiscard
) {
2480 ret
= bs
->drv
->bdrv_co_pdiscard(bs
, offset
, num
);
2483 CoroutineIOCompletion co
= {
2484 .coroutine
= qemu_coroutine_self(),
2487 acb
= bs
->drv
->bdrv_aio_pdiscard(bs
, offset
, num
,
2488 bdrv_co_io_em_complete
, &co
);
2493 qemu_coroutine_yield();
2497 if (ret
&& ret
!= -ENOTSUP
) {
2507 bdrv_set_dirty(bs
, req
.offset
>> BDRV_SECTOR_BITS
,
2508 req
.bytes
>> BDRV_SECTOR_BITS
);
2509 tracked_request_end(&req
);
2510 bdrv_dec_in_flight(bs
);
2514 int bdrv_pdiscard(BlockDriverState
*bs
, int64_t offset
, int count
)
2524 if (qemu_in_coroutine()) {
2525 /* Fast-path if already in coroutine context */
2526 bdrv_pdiscard_co_entry(&rwco
);
2528 co
= qemu_coroutine_create(bdrv_pdiscard_co_entry
, &rwco
);
2529 qemu_coroutine_enter(co
);
2530 BDRV_POLL_WHILE(bs
, rwco
.ret
== NOT_DONE
);
2536 int bdrv_co_ioctl(BlockDriverState
*bs
, int req
, void *buf
)
2538 BlockDriver
*drv
= bs
->drv
;
2539 CoroutineIOCompletion co
= {
2540 .coroutine
= qemu_coroutine_self(),
2544 bdrv_inc_in_flight(bs
);
2545 if (!drv
|| (!drv
->bdrv_aio_ioctl
&& !drv
->bdrv_co_ioctl
)) {
2550 if (drv
->bdrv_co_ioctl
) {
2551 co
.ret
= drv
->bdrv_co_ioctl(bs
, req
, buf
);
2553 acb
= drv
->bdrv_aio_ioctl(bs
, req
, buf
, bdrv_co_io_em_complete
, &co
);
2558 qemu_coroutine_yield();
2561 bdrv_dec_in_flight(bs
);
2565 void *qemu_blockalign(BlockDriverState
*bs
, size_t size
)
2567 return qemu_memalign(bdrv_opt_mem_align(bs
), size
);
2570 void *qemu_blockalign0(BlockDriverState
*bs
, size_t size
)
2572 return memset(qemu_blockalign(bs
, size
), 0, size
);
2575 void *qemu_try_blockalign(BlockDriverState
*bs
, size_t size
)
2577 size_t align
= bdrv_opt_mem_align(bs
);
2579 /* Ensure that NULL is never returned on success */
2585 return qemu_try_memalign(align
, size
);
2588 void *qemu_try_blockalign0(BlockDriverState
*bs
, size_t size
)
2590 void *mem
= qemu_try_blockalign(bs
, size
);
2593 memset(mem
, 0, size
);
2600 * Check if all memory in this vector is sector aligned.
2602 bool bdrv_qiov_is_aligned(BlockDriverState
*bs
, QEMUIOVector
*qiov
)
2605 size_t alignment
= bdrv_min_mem_align(bs
);
2607 for (i
= 0; i
< qiov
->niov
; i
++) {
2608 if ((uintptr_t) qiov
->iov
[i
].iov_base
% alignment
) {
2611 if (qiov
->iov
[i
].iov_len
% alignment
) {
2619 void bdrv_add_before_write_notifier(BlockDriverState
*bs
,
2620 NotifierWithReturn
*notifier
)
2622 notifier_with_return_list_add(&bs
->before_write_notifiers
, notifier
);
2625 void bdrv_io_plug(BlockDriverState
*bs
)
2629 QLIST_FOREACH(child
, &bs
->children
, next
) {
2630 bdrv_io_plug(child
->bs
);
2633 if (bs
->io_plugged
++ == 0) {
2634 BlockDriver
*drv
= bs
->drv
;
2635 if (drv
&& drv
->bdrv_io_plug
) {
2636 drv
->bdrv_io_plug(bs
);
2641 void bdrv_io_unplug(BlockDriverState
*bs
)
2645 assert(bs
->io_plugged
);
2646 if (--bs
->io_plugged
== 0) {
2647 BlockDriver
*drv
= bs
->drv
;
2648 if (drv
&& drv
->bdrv_io_unplug
) {
2649 drv
->bdrv_io_unplug(bs
);
2653 QLIST_FOREACH(child
, &bs
->children
, next
) {
2654 bdrv_io_unplug(child
->bs
);