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_rw_vector(BdrvChild
*child
,
40 BdrvRequestFlags flags
,
41 BlockCompletionFunc
*cb
,
44 static void coroutine_fn
bdrv_co_do_rw(void *opaque
);
45 static int coroutine_fn
bdrv_co_do_pwrite_zeroes(BlockDriverState
*bs
,
46 int64_t offset
, int count
, BdrvRequestFlags flags
);
48 static void bdrv_parent_drained_begin(BlockDriverState
*bs
)
52 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
53 if (c
->role
->drained_begin
) {
54 c
->role
->drained_begin(c
);
59 static void bdrv_parent_drained_end(BlockDriverState
*bs
)
63 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
64 if (c
->role
->drained_end
) {
65 c
->role
->drained_end(c
);
70 static void bdrv_merge_limits(BlockLimits
*dst
, const BlockLimits
*src
)
72 dst
->opt_transfer
= MAX(dst
->opt_transfer
, src
->opt_transfer
);
73 dst
->max_transfer
= MIN_NON_ZERO(dst
->max_transfer
, src
->max_transfer
);
74 dst
->opt_mem_alignment
= MAX(dst
->opt_mem_alignment
,
75 src
->opt_mem_alignment
);
76 dst
->min_mem_alignment
= MAX(dst
->min_mem_alignment
,
77 src
->min_mem_alignment
);
78 dst
->max_iov
= MIN_NON_ZERO(dst
->max_iov
, src
->max_iov
);
81 void bdrv_refresh_limits(BlockDriverState
*bs
, Error
**errp
)
83 BlockDriver
*drv
= bs
->drv
;
84 Error
*local_err
= NULL
;
86 memset(&bs
->bl
, 0, sizeof(bs
->bl
));
92 /* Default alignment based on whether driver has byte interface */
93 bs
->bl
.request_alignment
= drv
->bdrv_co_preadv
? 1 : 512;
95 /* Take some limits from the children as a default */
97 bdrv_refresh_limits(bs
->file
->bs
, &local_err
);
99 error_propagate(errp
, local_err
);
102 bdrv_merge_limits(&bs
->bl
, &bs
->file
->bs
->bl
);
104 bs
->bl
.min_mem_alignment
= 512;
105 bs
->bl
.opt_mem_alignment
= getpagesize();
107 /* Safe default since most protocols use readv()/writev()/etc */
108 bs
->bl
.max_iov
= IOV_MAX
;
112 bdrv_refresh_limits(bs
->backing
->bs
, &local_err
);
114 error_propagate(errp
, local_err
);
117 bdrv_merge_limits(&bs
->bl
, &bs
->backing
->bs
->bl
);
120 /* Then let the driver override it */
121 if (drv
->bdrv_refresh_limits
) {
122 drv
->bdrv_refresh_limits(bs
, errp
);
127 * The copy-on-read flag is actually a reference count so multiple users may
128 * use the feature without worrying about clobbering its previous state.
129 * Copy-on-read stays enabled until all users have called to disable it.
131 void bdrv_enable_copy_on_read(BlockDriverState
*bs
)
136 void bdrv_disable_copy_on_read(BlockDriverState
*bs
)
138 assert(bs
->copy_on_read
> 0);
142 /* Check if any requests are in-flight (including throttled requests) */
143 bool bdrv_requests_pending(BlockDriverState
*bs
)
147 if (!QLIST_EMPTY(&bs
->tracked_requests
)) {
151 QLIST_FOREACH(child
, &bs
->children
, next
) {
152 if (bdrv_requests_pending(child
->bs
)) {
160 static void bdrv_drain_recurse(BlockDriverState
*bs
)
164 if (bs
->drv
&& bs
->drv
->bdrv_drain
) {
165 bs
->drv
->bdrv_drain(bs
);
167 QLIST_FOREACH(child
, &bs
->children
, next
) {
168 bdrv_drain_recurse(child
->bs
);
174 BlockDriverState
*bs
;
179 static void bdrv_drain_poll(BlockDriverState
*bs
)
185 busy
= bdrv_requests_pending(bs
);
186 busy
|= aio_poll(bdrv_get_aio_context(bs
), busy
);
190 static void bdrv_co_drain_bh_cb(void *opaque
)
192 BdrvCoDrainData
*data
= opaque
;
193 Coroutine
*co
= data
->co
;
195 qemu_bh_delete(data
->bh
);
196 bdrv_drain_poll(data
->bs
);
198 qemu_coroutine_enter(co
, NULL
);
201 static void coroutine_fn
bdrv_co_yield_to_drain(BlockDriverState
*bs
)
203 BdrvCoDrainData data
;
205 /* Calling bdrv_drain() from a BH ensures the current coroutine yields and
206 * other coroutines run if they were queued from
207 * qemu_co_queue_run_restart(). */
209 assert(qemu_in_coroutine());
210 data
= (BdrvCoDrainData
) {
211 .co
= qemu_coroutine_self(),
214 .bh
= aio_bh_new(bdrv_get_aio_context(bs
), bdrv_co_drain_bh_cb
, &data
),
216 qemu_bh_schedule(data
.bh
);
218 qemu_coroutine_yield();
219 /* If we are resumed from some other event (such as an aio completion or a
220 * timer callback), it is a bug in the caller that should be fixed. */
224 void bdrv_drained_begin(BlockDriverState
*bs
)
226 if (!bs
->quiesce_counter
++) {
227 aio_disable_external(bdrv_get_aio_context(bs
));
228 bdrv_parent_drained_begin(bs
);
231 bdrv_io_unplugged_begin(bs
);
232 bdrv_drain_recurse(bs
);
233 if (qemu_in_coroutine()) {
234 bdrv_co_yield_to_drain(bs
);
238 bdrv_io_unplugged_end(bs
);
241 void bdrv_drained_end(BlockDriverState
*bs
)
243 assert(bs
->quiesce_counter
> 0);
244 if (--bs
->quiesce_counter
> 0) {
248 bdrv_parent_drained_end(bs
);
249 aio_enable_external(bdrv_get_aio_context(bs
));
253 * Wait for pending requests to complete on a single BlockDriverState subtree,
254 * and suspend block driver's internal I/O until next request arrives.
256 * Note that unlike bdrv_drain_all(), the caller must hold the BlockDriverState
259 * Only this BlockDriverState's AioContext is run, so in-flight requests must
260 * not depend on events in other AioContexts. In that case, use
261 * bdrv_drain_all() instead.
263 void coroutine_fn
bdrv_co_drain(BlockDriverState
*bs
)
265 assert(qemu_in_coroutine());
266 bdrv_drained_begin(bs
);
267 bdrv_drained_end(bs
);
270 void bdrv_drain(BlockDriverState
*bs
)
272 bdrv_drained_begin(bs
);
273 bdrv_drained_end(bs
);
277 * Wait for pending requests to complete across all BlockDriverStates
279 * This function does not flush data to disk, use bdrv_flush_all() for that
280 * after calling this function.
282 void bdrv_drain_all(void)
284 /* Always run first iteration so any pending completion BHs run */
286 BlockDriverState
*bs
;
288 BlockJob
*job
= NULL
;
289 GSList
*aio_ctxs
= NULL
, *ctx
;
291 while ((job
= block_job_next(job
))) {
292 AioContext
*aio_context
= blk_get_aio_context(job
->blk
);
294 aio_context_acquire(aio_context
);
295 block_job_pause(job
);
296 aio_context_release(aio_context
);
299 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
300 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
302 aio_context_acquire(aio_context
);
303 bdrv_parent_drained_begin(bs
);
304 bdrv_io_unplugged_begin(bs
);
305 bdrv_drain_recurse(bs
);
306 aio_context_release(aio_context
);
308 if (!g_slist_find(aio_ctxs
, aio_context
)) {
309 aio_ctxs
= g_slist_prepend(aio_ctxs
, aio_context
);
313 /* Note that completion of an asynchronous I/O operation can trigger any
314 * number of other I/O operations on other devices---for example a
315 * coroutine can submit an I/O request to another device in response to
316 * request completion. Therefore we must keep looping until there was no
317 * more activity rather than simply draining each device independently.
322 for (ctx
= aio_ctxs
; ctx
!= NULL
; ctx
= ctx
->next
) {
323 AioContext
*aio_context
= ctx
->data
;
325 aio_context_acquire(aio_context
);
326 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
327 if (aio_context
== bdrv_get_aio_context(bs
)) {
328 if (bdrv_requests_pending(bs
)) {
330 aio_poll(aio_context
, busy
);
334 busy
|= aio_poll(aio_context
, false);
335 aio_context_release(aio_context
);
339 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
340 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
342 aio_context_acquire(aio_context
);
343 bdrv_io_unplugged_end(bs
);
344 bdrv_parent_drained_end(bs
);
345 aio_context_release(aio_context
);
347 g_slist_free(aio_ctxs
);
350 while ((job
= block_job_next(job
))) {
351 AioContext
*aio_context
= blk_get_aio_context(job
->blk
);
353 aio_context_acquire(aio_context
);
354 block_job_resume(job
);
355 aio_context_release(aio_context
);
360 * Remove an active request from the tracked requests list
362 * This function should be called when a tracked request is completing.
364 static void tracked_request_end(BdrvTrackedRequest
*req
)
366 if (req
->serialising
) {
367 req
->bs
->serialising_in_flight
--;
370 QLIST_REMOVE(req
, list
);
371 qemu_co_queue_restart_all(&req
->wait_queue
);
375 * Add an active request to the tracked requests list
377 static void tracked_request_begin(BdrvTrackedRequest
*req
,
378 BlockDriverState
*bs
,
381 enum BdrvTrackedRequestType type
)
383 *req
= (BdrvTrackedRequest
){
388 .co
= qemu_coroutine_self(),
389 .serialising
= false,
390 .overlap_offset
= offset
,
391 .overlap_bytes
= bytes
,
394 qemu_co_queue_init(&req
->wait_queue
);
396 QLIST_INSERT_HEAD(&bs
->tracked_requests
, req
, list
);
399 static void mark_request_serialising(BdrvTrackedRequest
*req
, uint64_t align
)
401 int64_t overlap_offset
= req
->offset
& ~(align
- 1);
402 unsigned int overlap_bytes
= ROUND_UP(req
->offset
+ req
->bytes
, align
)
405 if (!req
->serialising
) {
406 req
->bs
->serialising_in_flight
++;
407 req
->serialising
= true;
410 req
->overlap_offset
= MIN(req
->overlap_offset
, overlap_offset
);
411 req
->overlap_bytes
= MAX(req
->overlap_bytes
, overlap_bytes
);
415 * Round a region to cluster boundaries (sector-based)
417 void bdrv_round_sectors_to_clusters(BlockDriverState
*bs
,
418 int64_t sector_num
, int nb_sectors
,
419 int64_t *cluster_sector_num
,
420 int *cluster_nb_sectors
)
424 if (bdrv_get_info(bs
, &bdi
) < 0 || bdi
.cluster_size
== 0) {
425 *cluster_sector_num
= sector_num
;
426 *cluster_nb_sectors
= nb_sectors
;
428 int64_t c
= bdi
.cluster_size
/ BDRV_SECTOR_SIZE
;
429 *cluster_sector_num
= QEMU_ALIGN_DOWN(sector_num
, c
);
430 *cluster_nb_sectors
= QEMU_ALIGN_UP(sector_num
- *cluster_sector_num
+
436 * Round a region to cluster boundaries
438 void bdrv_round_to_clusters(BlockDriverState
*bs
,
439 int64_t offset
, unsigned int bytes
,
440 int64_t *cluster_offset
,
441 unsigned int *cluster_bytes
)
445 if (bdrv_get_info(bs
, &bdi
) < 0 || bdi
.cluster_size
== 0) {
446 *cluster_offset
= offset
;
447 *cluster_bytes
= bytes
;
449 int64_t c
= bdi
.cluster_size
;
450 *cluster_offset
= QEMU_ALIGN_DOWN(offset
, c
);
451 *cluster_bytes
= QEMU_ALIGN_UP(offset
- *cluster_offset
+ bytes
, c
);
455 static int bdrv_get_cluster_size(BlockDriverState
*bs
)
460 ret
= bdrv_get_info(bs
, &bdi
);
461 if (ret
< 0 || bdi
.cluster_size
== 0) {
462 return bs
->bl
.request_alignment
;
464 return bdi
.cluster_size
;
468 static bool tracked_request_overlaps(BdrvTrackedRequest
*req
,
469 int64_t offset
, unsigned int bytes
)
472 if (offset
>= req
->overlap_offset
+ req
->overlap_bytes
) {
476 if (req
->overlap_offset
>= offset
+ bytes
) {
482 static bool coroutine_fn
wait_serialising_requests(BdrvTrackedRequest
*self
)
484 BlockDriverState
*bs
= self
->bs
;
485 BdrvTrackedRequest
*req
;
489 if (!bs
->serialising_in_flight
) {
495 QLIST_FOREACH(req
, &bs
->tracked_requests
, list
) {
496 if (req
== self
|| (!req
->serialising
&& !self
->serialising
)) {
499 if (tracked_request_overlaps(req
, self
->overlap_offset
,
500 self
->overlap_bytes
))
502 /* Hitting this means there was a reentrant request, for
503 * example, a block driver issuing nested requests. This must
504 * never happen since it means deadlock.
506 assert(qemu_coroutine_self() != req
->co
);
508 /* If the request is already (indirectly) waiting for us, or
509 * will wait for us as soon as it wakes up, then just go on
510 * (instead of producing a deadlock in the former case). */
511 if (!req
->waiting_for
) {
512 self
->waiting_for
= req
;
513 qemu_co_queue_wait(&req
->wait_queue
);
514 self
->waiting_for
= NULL
;
526 static int bdrv_check_byte_request(BlockDriverState
*bs
, int64_t offset
,
529 if (size
> BDRV_REQUEST_MAX_SECTORS
<< BDRV_SECTOR_BITS
) {
533 if (!bdrv_is_inserted(bs
)) {
544 static int bdrv_check_request(BlockDriverState
*bs
, int64_t sector_num
,
547 if (nb_sectors
< 0 || nb_sectors
> BDRV_REQUEST_MAX_SECTORS
) {
551 return bdrv_check_byte_request(bs
, sector_num
* BDRV_SECTOR_SIZE
,
552 nb_sectors
* BDRV_SECTOR_SIZE
);
555 typedef struct RwCo
{
556 BlockDriverState
*bs
;
561 BdrvRequestFlags flags
;
564 static void coroutine_fn
bdrv_rw_co_entry(void *opaque
)
568 if (!rwco
->is_write
) {
569 rwco
->ret
= bdrv_co_preadv(rwco
->bs
, rwco
->offset
,
570 rwco
->qiov
->size
, rwco
->qiov
,
573 rwco
->ret
= bdrv_co_pwritev(rwco
->bs
, rwco
->offset
,
574 rwco
->qiov
->size
, rwco
->qiov
,
580 * Process a vectored synchronous request using coroutines
582 static int bdrv_prwv_co(BlockDriverState
*bs
, int64_t offset
,
583 QEMUIOVector
*qiov
, bool is_write
,
584 BdrvRequestFlags flags
)
591 .is_write
= is_write
,
596 if (qemu_in_coroutine()) {
597 /* Fast-path if already in coroutine context */
598 bdrv_rw_co_entry(&rwco
);
600 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
602 co
= qemu_coroutine_create(bdrv_rw_co_entry
);
603 qemu_coroutine_enter(co
, &rwco
);
604 while (rwco
.ret
== NOT_DONE
) {
605 aio_poll(aio_context
, true);
612 * Process a synchronous request using coroutines
614 static int bdrv_rw_co(BlockDriverState
*bs
, int64_t sector_num
, uint8_t *buf
,
615 int nb_sectors
, bool is_write
, BdrvRequestFlags flags
)
619 .iov_base
= (void *)buf
,
620 .iov_len
= nb_sectors
* BDRV_SECTOR_SIZE
,
623 if (nb_sectors
< 0 || nb_sectors
> BDRV_REQUEST_MAX_SECTORS
) {
627 qemu_iovec_init_external(&qiov
, &iov
, 1);
628 return bdrv_prwv_co(bs
, sector_num
<< BDRV_SECTOR_BITS
,
629 &qiov
, is_write
, flags
);
632 /* return < 0 if error. See bdrv_write() for the return codes */
633 int bdrv_read(BdrvChild
*child
, int64_t sector_num
,
634 uint8_t *buf
, int nb_sectors
)
636 return bdrv_rw_co(child
->bs
, sector_num
, buf
, nb_sectors
, false, 0);
639 /* Return < 0 if error. Important errors are:
640 -EIO generic I/O error (may happen for all errors)
641 -ENOMEDIUM No media inserted.
642 -EINVAL Invalid sector number or nb_sectors
643 -EACCES Trying to write a read-only device
645 int bdrv_write(BdrvChild
*child
, int64_t sector_num
,
646 const uint8_t *buf
, int nb_sectors
)
648 return bdrv_rw_co(child
->bs
, sector_num
, (uint8_t *)buf
, nb_sectors
,
652 int bdrv_pwrite_zeroes(BlockDriverState
*bs
, int64_t offset
,
653 int count
, BdrvRequestFlags flags
)
661 qemu_iovec_init_external(&qiov
, &iov
, 1);
662 return bdrv_prwv_co(bs
, offset
, &qiov
, true,
663 BDRV_REQ_ZERO_WRITE
| flags
);
667 * Completely zero out a block device with the help of bdrv_pwrite_zeroes.
668 * The operation is sped up by checking the block status and only writing
669 * zeroes to the device if they currently do not return zeroes. Optional
670 * flags are passed through to bdrv_pwrite_zeroes (e.g. BDRV_REQ_MAY_UNMAP,
673 * Returns < 0 on error, 0 on success. For error codes see bdrv_write().
675 int bdrv_make_zero(BlockDriverState
*bs
, BdrvRequestFlags flags
)
677 int64_t target_sectors
, ret
, nb_sectors
, sector_num
= 0;
678 BlockDriverState
*file
;
681 target_sectors
= bdrv_nb_sectors(bs
);
682 if (target_sectors
< 0) {
683 return target_sectors
;
687 nb_sectors
= MIN(target_sectors
- sector_num
, BDRV_REQUEST_MAX_SECTORS
);
688 if (nb_sectors
<= 0) {
691 ret
= bdrv_get_block_status(bs
, sector_num
, nb_sectors
, &n
, &file
);
693 error_report("error getting block status at sector %" PRId64
": %s",
694 sector_num
, strerror(-ret
));
697 if (ret
& BDRV_BLOCK_ZERO
) {
701 ret
= bdrv_pwrite_zeroes(bs
, sector_num
<< BDRV_SECTOR_BITS
,
702 n
<< BDRV_SECTOR_BITS
, flags
);
704 error_report("error writing zeroes at sector %" PRId64
": %s",
705 sector_num
, strerror(-ret
));
712 int bdrv_preadv(BdrvChild
*child
, int64_t offset
, QEMUIOVector
*qiov
)
716 ret
= bdrv_prwv_co(child
->bs
, offset
, qiov
, false, 0);
724 int bdrv_pread(BdrvChild
*child
, int64_t offset
, void *buf
, int bytes
)
728 .iov_base
= (void *)buf
,
736 qemu_iovec_init_external(&qiov
, &iov
, 1);
737 return bdrv_preadv(child
, offset
, &qiov
);
740 int bdrv_pwritev(BdrvChild
*child
, int64_t offset
, QEMUIOVector
*qiov
)
744 ret
= bdrv_prwv_co(child
->bs
, offset
, qiov
, true, 0);
752 int bdrv_pwrite(BdrvChild
*child
, int64_t offset
, const void *buf
, int bytes
)
756 .iov_base
= (void *) buf
,
764 qemu_iovec_init_external(&qiov
, &iov
, 1);
765 return bdrv_pwritev(child
, offset
, &qiov
);
769 * Writes to the file and ensures that no writes are reordered across this
770 * request (acts as a barrier)
772 * Returns 0 on success, -errno in error cases.
774 int bdrv_pwrite_sync(BdrvChild
*child
, int64_t offset
,
775 const void *buf
, int count
)
779 ret
= bdrv_pwrite(child
, offset
, buf
, count
);
784 ret
= bdrv_flush(child
->bs
);
792 typedef struct CoroutineIOCompletion
{
793 Coroutine
*coroutine
;
795 } CoroutineIOCompletion
;
797 static void bdrv_co_io_em_complete(void *opaque
, int ret
)
799 CoroutineIOCompletion
*co
= opaque
;
802 qemu_coroutine_enter(co
->coroutine
, NULL
);
805 static int coroutine_fn
bdrv_driver_preadv(BlockDriverState
*bs
,
806 uint64_t offset
, uint64_t bytes
,
807 QEMUIOVector
*qiov
, int flags
)
809 BlockDriver
*drv
= bs
->drv
;
811 unsigned int nb_sectors
;
813 assert(!(flags
& ~BDRV_REQ_MASK
));
815 if (drv
->bdrv_co_preadv
) {
816 return drv
->bdrv_co_preadv(bs
, offset
, bytes
, qiov
, flags
);
819 sector_num
= offset
>> BDRV_SECTOR_BITS
;
820 nb_sectors
= bytes
>> BDRV_SECTOR_BITS
;
822 assert((offset
& (BDRV_SECTOR_SIZE
- 1)) == 0);
823 assert((bytes
& (BDRV_SECTOR_SIZE
- 1)) == 0);
824 assert((bytes
>> BDRV_SECTOR_BITS
) <= BDRV_REQUEST_MAX_SECTORS
);
826 if (drv
->bdrv_co_readv
) {
827 return drv
->bdrv_co_readv(bs
, sector_num
, nb_sectors
, qiov
);
830 CoroutineIOCompletion co
= {
831 .coroutine
= qemu_coroutine_self(),
834 acb
= bs
->drv
->bdrv_aio_readv(bs
, sector_num
, qiov
, nb_sectors
,
835 bdrv_co_io_em_complete
, &co
);
839 qemu_coroutine_yield();
845 static int coroutine_fn
bdrv_driver_pwritev(BlockDriverState
*bs
,
846 uint64_t offset
, uint64_t bytes
,
847 QEMUIOVector
*qiov
, int flags
)
849 BlockDriver
*drv
= bs
->drv
;
851 unsigned int nb_sectors
;
854 assert(!(flags
& ~BDRV_REQ_MASK
));
856 if (drv
->bdrv_co_pwritev
) {
857 ret
= drv
->bdrv_co_pwritev(bs
, offset
, bytes
, qiov
,
858 flags
& bs
->supported_write_flags
);
859 flags
&= ~bs
->supported_write_flags
;
863 sector_num
= offset
>> BDRV_SECTOR_BITS
;
864 nb_sectors
= bytes
>> BDRV_SECTOR_BITS
;
866 assert((offset
& (BDRV_SECTOR_SIZE
- 1)) == 0);
867 assert((bytes
& (BDRV_SECTOR_SIZE
- 1)) == 0);
868 assert((bytes
>> BDRV_SECTOR_BITS
) <= BDRV_REQUEST_MAX_SECTORS
);
870 if (drv
->bdrv_co_writev_flags
) {
871 ret
= drv
->bdrv_co_writev_flags(bs
, sector_num
, nb_sectors
, qiov
,
872 flags
& bs
->supported_write_flags
);
873 flags
&= ~bs
->supported_write_flags
;
874 } else if (drv
->bdrv_co_writev
) {
875 assert(!bs
->supported_write_flags
);
876 ret
= drv
->bdrv_co_writev(bs
, sector_num
, nb_sectors
, qiov
);
879 CoroutineIOCompletion co
= {
880 .coroutine
= qemu_coroutine_self(),
883 acb
= bs
->drv
->bdrv_aio_writev(bs
, sector_num
, qiov
, nb_sectors
,
884 bdrv_co_io_em_complete
, &co
);
888 qemu_coroutine_yield();
894 if (ret
== 0 && (flags
& BDRV_REQ_FUA
)) {
895 ret
= bdrv_co_flush(bs
);
901 static int coroutine_fn
bdrv_co_do_copy_on_readv(BlockDriverState
*bs
,
902 int64_t offset
, unsigned int bytes
, QEMUIOVector
*qiov
)
904 /* Perform I/O through a temporary buffer so that users who scribble over
905 * their read buffer while the operation is in progress do not end up
906 * modifying the image file. This is critical for zero-copy guest I/O
907 * where anything might happen inside guest memory.
911 BlockDriver
*drv
= bs
->drv
;
913 QEMUIOVector bounce_qiov
;
914 int64_t cluster_offset
;
915 unsigned int cluster_bytes
;
919 /* Cover entire cluster so no additional backing file I/O is required when
920 * allocating cluster in the image file.
922 bdrv_round_to_clusters(bs
, offset
, bytes
, &cluster_offset
, &cluster_bytes
);
924 trace_bdrv_co_do_copy_on_readv(bs
, offset
, bytes
,
925 cluster_offset
, cluster_bytes
);
927 iov
.iov_len
= cluster_bytes
;
928 iov
.iov_base
= bounce_buffer
= qemu_try_blockalign(bs
, iov
.iov_len
);
929 if (bounce_buffer
== NULL
) {
934 qemu_iovec_init_external(&bounce_qiov
, &iov
, 1);
936 ret
= bdrv_driver_preadv(bs
, cluster_offset
, cluster_bytes
,
942 if (drv
->bdrv_co_pwrite_zeroes
&&
943 buffer_is_zero(bounce_buffer
, iov
.iov_len
)) {
944 /* FIXME: Should we (perhaps conditionally) be setting
945 * BDRV_REQ_MAY_UNMAP, if it will allow for a sparser copy
946 * that still correctly reads as zero? */
947 ret
= bdrv_co_do_pwrite_zeroes(bs
, cluster_offset
, cluster_bytes
, 0);
949 /* This does not change the data on the disk, it is not necessary
950 * to flush even in cache=writethrough mode.
952 ret
= bdrv_driver_pwritev(bs
, cluster_offset
, cluster_bytes
,
957 /* It might be okay to ignore write errors for guest requests. If this
958 * is a deliberate copy-on-read then we don't want to ignore the error.
959 * Simply report it in all cases.
964 skip_bytes
= offset
- cluster_offset
;
965 qemu_iovec_from_buf(qiov
, 0, bounce_buffer
+ skip_bytes
, bytes
);
968 qemu_vfree(bounce_buffer
);
973 * Forwards an already correctly aligned request to the BlockDriver. This
974 * handles copy on read and zeroing after EOF; any other features must be
975 * implemented by the caller.
977 static int coroutine_fn
bdrv_aligned_preadv(BlockDriverState
*bs
,
978 BdrvTrackedRequest
*req
, int64_t offset
, unsigned int bytes
,
979 int64_t align
, QEMUIOVector
*qiov
, int flags
)
981 int64_t total_bytes
, max_bytes
;
984 assert(is_power_of_2(align
));
985 assert((offset
& (align
- 1)) == 0);
986 assert((bytes
& (align
- 1)) == 0);
987 assert(!qiov
|| bytes
== qiov
->size
);
988 assert((bs
->open_flags
& BDRV_O_NO_IO
) == 0);
990 /* TODO: We would need a per-BDS .supported_read_flags and
991 * potential fallback support, if we ever implement any read flags
992 * to pass through to drivers. For now, there aren't any
993 * passthrough flags. */
994 assert(!(flags
& ~(BDRV_REQ_NO_SERIALISING
| BDRV_REQ_COPY_ON_READ
)));
996 /* Handle Copy on Read and associated serialisation */
997 if (flags
& BDRV_REQ_COPY_ON_READ
) {
998 /* If we touch the same cluster it counts as an overlap. This
999 * guarantees that allocating writes will be serialized and not race
1000 * with each other for the same cluster. For example, in copy-on-read
1001 * it ensures that the CoR read and write operations are atomic and
1002 * guest writes cannot interleave between them. */
1003 mark_request_serialising(req
, bdrv_get_cluster_size(bs
));
1006 if (!(flags
& BDRV_REQ_NO_SERIALISING
)) {
1007 wait_serialising_requests(req
);
1010 if (flags
& BDRV_REQ_COPY_ON_READ
) {
1011 int64_t start_sector
= offset
>> BDRV_SECTOR_BITS
;
1012 int64_t end_sector
= DIV_ROUND_UP(offset
+ bytes
, BDRV_SECTOR_SIZE
);
1013 unsigned int nb_sectors
= end_sector
- start_sector
;
1016 ret
= bdrv_is_allocated(bs
, start_sector
, nb_sectors
, &pnum
);
1021 if (!ret
|| pnum
!= nb_sectors
) {
1022 ret
= bdrv_co_do_copy_on_readv(bs
, offset
, bytes
, qiov
);
1027 /* Forward the request to the BlockDriver */
1028 total_bytes
= bdrv_getlength(bs
);
1029 if (total_bytes
< 0) {
1034 max_bytes
= ROUND_UP(MAX(0, total_bytes
- offset
), align
);
1035 if (bytes
<= max_bytes
) {
1036 ret
= bdrv_driver_preadv(bs
, offset
, bytes
, qiov
, 0);
1037 } else if (max_bytes
> 0) {
1038 QEMUIOVector local_qiov
;
1040 qemu_iovec_init(&local_qiov
, qiov
->niov
);
1041 qemu_iovec_concat(&local_qiov
, qiov
, 0, max_bytes
);
1043 ret
= bdrv_driver_preadv(bs
, offset
, max_bytes
, &local_qiov
, 0);
1045 qemu_iovec_destroy(&local_qiov
);
1050 /* Reading beyond end of file is supposed to produce zeroes */
1051 if (ret
== 0 && total_bytes
< offset
+ bytes
) {
1052 uint64_t zero_offset
= MAX(0, total_bytes
- offset
);
1053 uint64_t zero_bytes
= offset
+ bytes
- zero_offset
;
1054 qemu_iovec_memset(qiov
, zero_offset
, 0, zero_bytes
);
1062 * Handle a read request in coroutine context
1064 int coroutine_fn
bdrv_co_preadv(BlockDriverState
*bs
,
1065 int64_t offset
, unsigned int bytes
, QEMUIOVector
*qiov
,
1066 BdrvRequestFlags flags
)
1068 BlockDriver
*drv
= bs
->drv
;
1069 BdrvTrackedRequest req
;
1071 uint64_t align
= bs
->bl
.request_alignment
;
1072 uint8_t *head_buf
= NULL
;
1073 uint8_t *tail_buf
= NULL
;
1074 QEMUIOVector local_qiov
;
1075 bool use_local_qiov
= false;
1082 ret
= bdrv_check_byte_request(bs
, offset
, bytes
);
1087 /* Don't do copy-on-read if we read data before write operation */
1088 if (bs
->copy_on_read
&& !(flags
& BDRV_REQ_NO_SERIALISING
)) {
1089 flags
|= BDRV_REQ_COPY_ON_READ
;
1092 /* Align read if necessary by padding qiov */
1093 if (offset
& (align
- 1)) {
1094 head_buf
= qemu_blockalign(bs
, align
);
1095 qemu_iovec_init(&local_qiov
, qiov
->niov
+ 2);
1096 qemu_iovec_add(&local_qiov
, head_buf
, offset
& (align
- 1));
1097 qemu_iovec_concat(&local_qiov
, qiov
, 0, qiov
->size
);
1098 use_local_qiov
= true;
1100 bytes
+= offset
& (align
- 1);
1101 offset
= offset
& ~(align
- 1);
1104 if ((offset
+ bytes
) & (align
- 1)) {
1105 if (!use_local_qiov
) {
1106 qemu_iovec_init(&local_qiov
, qiov
->niov
+ 1);
1107 qemu_iovec_concat(&local_qiov
, qiov
, 0, qiov
->size
);
1108 use_local_qiov
= true;
1110 tail_buf
= qemu_blockalign(bs
, align
);
1111 qemu_iovec_add(&local_qiov
, tail_buf
,
1112 align
- ((offset
+ bytes
) & (align
- 1)));
1114 bytes
= ROUND_UP(bytes
, align
);
1117 tracked_request_begin(&req
, bs
, offset
, bytes
, BDRV_TRACKED_READ
);
1118 ret
= bdrv_aligned_preadv(bs
, &req
, offset
, bytes
, align
,
1119 use_local_qiov
? &local_qiov
: qiov
,
1121 tracked_request_end(&req
);
1123 if (use_local_qiov
) {
1124 qemu_iovec_destroy(&local_qiov
);
1125 qemu_vfree(head_buf
);
1126 qemu_vfree(tail_buf
);
1132 static int coroutine_fn
bdrv_co_do_readv(BdrvChild
*child
,
1133 int64_t sector_num
, int nb_sectors
, QEMUIOVector
*qiov
,
1134 BdrvRequestFlags flags
)
1136 if (nb_sectors
< 0 || nb_sectors
> BDRV_REQUEST_MAX_SECTORS
) {
1140 return bdrv_co_preadv(child
->bs
, sector_num
<< BDRV_SECTOR_BITS
,
1141 nb_sectors
<< BDRV_SECTOR_BITS
, qiov
, flags
);
1144 int coroutine_fn
bdrv_co_readv(BdrvChild
*child
, int64_t sector_num
,
1145 int nb_sectors
, QEMUIOVector
*qiov
)
1147 trace_bdrv_co_readv(child
->bs
, sector_num
, nb_sectors
);
1149 return bdrv_co_do_readv(child
, sector_num
, nb_sectors
, qiov
, 0);
1152 /* Maximum buffer for write zeroes fallback, in bytes */
1153 #define MAX_WRITE_ZEROES_BOUNCE_BUFFER (32768 << BDRV_SECTOR_BITS)
1155 static int coroutine_fn
bdrv_co_do_pwrite_zeroes(BlockDriverState
*bs
,
1156 int64_t offset
, int count
, BdrvRequestFlags flags
)
1158 BlockDriver
*drv
= bs
->drv
;
1160 struct iovec iov
= {0};
1162 bool need_flush
= false;
1166 int max_write_zeroes
= MIN_NON_ZERO(bs
->bl
.max_pwrite_zeroes
, INT_MAX
);
1167 int alignment
= MAX(bs
->bl
.pwrite_zeroes_alignment
,
1168 bs
->bl
.request_alignment
);
1170 assert(is_power_of_2(alignment
));
1171 head
= offset
& (alignment
- 1);
1172 tail
= (offset
+ count
) & (alignment
- 1);
1173 max_write_zeroes
&= ~(alignment
- 1);
1175 while (count
> 0 && !ret
) {
1178 /* Align request. Block drivers can expect the "bulk" of the request
1179 * to be aligned, and that unaligned requests do not cross cluster
1183 /* Make a small request up to the first aligned sector. */
1184 num
= MIN(count
, alignment
- head
);
1186 } else if (tail
&& num
> alignment
) {
1187 /* Shorten the request to the last aligned sector. */
1191 /* limit request size */
1192 if (num
> max_write_zeroes
) {
1193 num
= max_write_zeroes
;
1197 /* First try the efficient write zeroes operation */
1198 if (drv
->bdrv_co_pwrite_zeroes
) {
1199 ret
= drv
->bdrv_co_pwrite_zeroes(bs
, offset
, num
,
1200 flags
& bs
->supported_zero_flags
);
1201 if (ret
!= -ENOTSUP
&& (flags
& BDRV_REQ_FUA
) &&
1202 !(bs
->supported_zero_flags
& BDRV_REQ_FUA
)) {
1206 assert(!bs
->supported_zero_flags
);
1209 if (ret
== -ENOTSUP
) {
1210 /* Fall back to bounce buffer if write zeroes is unsupported */
1211 int max_transfer
= MIN_NON_ZERO(bs
->bl
.max_transfer
,
1212 MAX_WRITE_ZEROES_BOUNCE_BUFFER
);
1213 BdrvRequestFlags write_flags
= flags
& ~BDRV_REQ_ZERO_WRITE
;
1215 if ((flags
& BDRV_REQ_FUA
) &&
1216 !(bs
->supported_write_flags
& BDRV_REQ_FUA
)) {
1217 /* No need for bdrv_driver_pwrite() to do a fallback
1218 * flush on each chunk; use just one at the end */
1219 write_flags
&= ~BDRV_REQ_FUA
;
1222 num
= MIN(num
, max_transfer
);
1224 if (iov
.iov_base
== NULL
) {
1225 iov
.iov_base
= qemu_try_blockalign(bs
, num
);
1226 if (iov
.iov_base
== NULL
) {
1230 memset(iov
.iov_base
, 0, num
);
1232 qemu_iovec_init_external(&qiov
, &iov
, 1);
1234 ret
= bdrv_driver_pwritev(bs
, offset
, num
, &qiov
, write_flags
);
1236 /* Keep bounce buffer around if it is big enough for all
1237 * all future requests.
1239 if (num
< max_transfer
) {
1240 qemu_vfree(iov
.iov_base
);
1241 iov
.iov_base
= NULL
;
1250 if (ret
== 0 && need_flush
) {
1251 ret
= bdrv_co_flush(bs
);
1253 qemu_vfree(iov
.iov_base
);
1258 * Forwards an already correctly aligned write request to the BlockDriver.
1260 static int coroutine_fn
bdrv_aligned_pwritev(BlockDriverState
*bs
,
1261 BdrvTrackedRequest
*req
, int64_t offset
, unsigned int bytes
,
1262 int64_t align
, QEMUIOVector
*qiov
, int flags
)
1264 BlockDriver
*drv
= bs
->drv
;
1268 int64_t start_sector
= offset
>> BDRV_SECTOR_BITS
;
1269 int64_t end_sector
= DIV_ROUND_UP(offset
+ bytes
, BDRV_SECTOR_SIZE
);
1271 assert(is_power_of_2(align
));
1272 assert((offset
& (align
- 1)) == 0);
1273 assert((bytes
& (align
- 1)) == 0);
1274 assert(!qiov
|| bytes
== qiov
->size
);
1275 assert((bs
->open_flags
& BDRV_O_NO_IO
) == 0);
1276 assert(!(flags
& ~BDRV_REQ_MASK
));
1278 waited
= wait_serialising_requests(req
);
1279 assert(!waited
|| !req
->serialising
);
1280 assert(req
->overlap_offset
<= offset
);
1281 assert(offset
+ bytes
<= req
->overlap_offset
+ req
->overlap_bytes
);
1283 ret
= notifier_with_return_list_notify(&bs
->before_write_notifiers
, req
);
1285 if (!ret
&& bs
->detect_zeroes
!= BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF
&&
1286 !(flags
& BDRV_REQ_ZERO_WRITE
) && drv
->bdrv_co_pwrite_zeroes
&&
1287 qemu_iovec_is_zero(qiov
)) {
1288 flags
|= BDRV_REQ_ZERO_WRITE
;
1289 if (bs
->detect_zeroes
== BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP
) {
1290 flags
|= BDRV_REQ_MAY_UNMAP
;
1295 /* Do nothing, write notifier decided to fail this request */
1296 } else if (flags
& BDRV_REQ_ZERO_WRITE
) {
1297 bdrv_debug_event(bs
, BLKDBG_PWRITEV_ZERO
);
1298 ret
= bdrv_co_do_pwrite_zeroes(bs
, offset
, bytes
, flags
);
1300 bdrv_debug_event(bs
, BLKDBG_PWRITEV
);
1301 ret
= bdrv_driver_pwritev(bs
, offset
, bytes
, qiov
, flags
);
1303 bdrv_debug_event(bs
, BLKDBG_PWRITEV_DONE
);
1305 bdrv_set_dirty(bs
, start_sector
, end_sector
- start_sector
);
1307 if (bs
->wr_highest_offset
< offset
+ bytes
) {
1308 bs
->wr_highest_offset
= offset
+ bytes
;
1312 bs
->total_sectors
= MAX(bs
->total_sectors
, end_sector
);
1318 static int coroutine_fn
bdrv_co_do_zero_pwritev(BlockDriverState
*bs
,
1321 BdrvRequestFlags flags
,
1322 BdrvTrackedRequest
*req
)
1324 uint8_t *buf
= NULL
;
1325 QEMUIOVector local_qiov
;
1327 uint64_t align
= bs
->bl
.request_alignment
;
1328 unsigned int head_padding_bytes
, tail_padding_bytes
;
1331 head_padding_bytes
= offset
& (align
- 1);
1332 tail_padding_bytes
= align
- ((offset
+ bytes
) & (align
- 1));
1335 assert(flags
& BDRV_REQ_ZERO_WRITE
);
1336 if (head_padding_bytes
|| tail_padding_bytes
) {
1337 buf
= qemu_blockalign(bs
, align
);
1338 iov
= (struct iovec
) {
1342 qemu_iovec_init_external(&local_qiov
, &iov
, 1);
1344 if (head_padding_bytes
) {
1345 uint64_t zero_bytes
= MIN(bytes
, align
- head_padding_bytes
);
1347 /* RMW the unaligned part before head. */
1348 mark_request_serialising(req
, align
);
1349 wait_serialising_requests(req
);
1350 bdrv_debug_event(bs
, BLKDBG_PWRITEV_RMW_HEAD
);
1351 ret
= bdrv_aligned_preadv(bs
, req
, offset
& ~(align
- 1), align
,
1352 align
, &local_qiov
, 0);
1356 bdrv_debug_event(bs
, BLKDBG_PWRITEV_RMW_AFTER_HEAD
);
1358 memset(buf
+ head_padding_bytes
, 0, zero_bytes
);
1359 ret
= bdrv_aligned_pwritev(bs
, req
, offset
& ~(align
- 1), align
,
1361 flags
& ~BDRV_REQ_ZERO_WRITE
);
1365 offset
+= zero_bytes
;
1366 bytes
-= zero_bytes
;
1369 assert(!bytes
|| (offset
& (align
- 1)) == 0);
1370 if (bytes
>= align
) {
1371 /* Write the aligned part in the middle. */
1372 uint64_t aligned_bytes
= bytes
& ~(align
- 1);
1373 ret
= bdrv_aligned_pwritev(bs
, req
, offset
, aligned_bytes
, align
,
1378 bytes
-= aligned_bytes
;
1379 offset
+= aligned_bytes
;
1382 assert(!bytes
|| (offset
& (align
- 1)) == 0);
1384 assert(align
== tail_padding_bytes
+ bytes
);
1385 /* RMW the unaligned part after tail. */
1386 mark_request_serialising(req
, align
);
1387 wait_serialising_requests(req
);
1388 bdrv_debug_event(bs
, BLKDBG_PWRITEV_RMW_TAIL
);
1389 ret
= bdrv_aligned_preadv(bs
, req
, offset
, align
,
1390 align
, &local_qiov
, 0);
1394 bdrv_debug_event(bs
, BLKDBG_PWRITEV_RMW_AFTER_TAIL
);
1396 memset(buf
, 0, bytes
);
1397 ret
= bdrv_aligned_pwritev(bs
, req
, offset
, align
, align
,
1398 &local_qiov
, flags
& ~BDRV_REQ_ZERO_WRITE
);
1407 * Handle a write request in coroutine context
1409 int coroutine_fn
bdrv_co_pwritev(BlockDriverState
*bs
,
1410 int64_t offset
, unsigned int bytes
, QEMUIOVector
*qiov
,
1411 BdrvRequestFlags flags
)
1413 BdrvTrackedRequest req
;
1414 uint64_t align
= bs
->bl
.request_alignment
;
1415 uint8_t *head_buf
= NULL
;
1416 uint8_t *tail_buf
= NULL
;
1417 QEMUIOVector local_qiov
;
1418 bool use_local_qiov
= false;
1424 if (bs
->read_only
) {
1427 assert(!(bs
->open_flags
& BDRV_O_INACTIVE
));
1429 ret
= bdrv_check_byte_request(bs
, offset
, bytes
);
1435 * Align write if necessary by performing a read-modify-write cycle.
1436 * Pad qiov with the read parts and be sure to have a tracked request not
1437 * only for bdrv_aligned_pwritev, but also for the reads of the RMW cycle.
1439 tracked_request_begin(&req
, bs
, offset
, bytes
, BDRV_TRACKED_WRITE
);
1442 ret
= bdrv_co_do_zero_pwritev(bs
, offset
, bytes
, flags
, &req
);
1446 if (offset
& (align
- 1)) {
1447 QEMUIOVector head_qiov
;
1448 struct iovec head_iov
;
1450 mark_request_serialising(&req
, align
);
1451 wait_serialising_requests(&req
);
1453 head_buf
= qemu_blockalign(bs
, align
);
1454 head_iov
= (struct iovec
) {
1455 .iov_base
= head_buf
,
1458 qemu_iovec_init_external(&head_qiov
, &head_iov
, 1);
1460 bdrv_debug_event(bs
, BLKDBG_PWRITEV_RMW_HEAD
);
1461 ret
= bdrv_aligned_preadv(bs
, &req
, offset
& ~(align
- 1), align
,
1462 align
, &head_qiov
, 0);
1466 bdrv_debug_event(bs
, BLKDBG_PWRITEV_RMW_AFTER_HEAD
);
1468 qemu_iovec_init(&local_qiov
, qiov
->niov
+ 2);
1469 qemu_iovec_add(&local_qiov
, head_buf
, offset
& (align
- 1));
1470 qemu_iovec_concat(&local_qiov
, qiov
, 0, qiov
->size
);
1471 use_local_qiov
= true;
1473 bytes
+= offset
& (align
- 1);
1474 offset
= offset
& ~(align
- 1);
1476 /* We have read the tail already if the request is smaller
1477 * than one aligned block.
1479 if (bytes
< align
) {
1480 qemu_iovec_add(&local_qiov
, head_buf
+ bytes
, align
- bytes
);
1485 if ((offset
+ bytes
) & (align
- 1)) {
1486 QEMUIOVector tail_qiov
;
1487 struct iovec tail_iov
;
1491 mark_request_serialising(&req
, align
);
1492 waited
= wait_serialising_requests(&req
);
1493 assert(!waited
|| !use_local_qiov
);
1495 tail_buf
= qemu_blockalign(bs
, align
);
1496 tail_iov
= (struct iovec
) {
1497 .iov_base
= tail_buf
,
1500 qemu_iovec_init_external(&tail_qiov
, &tail_iov
, 1);
1502 bdrv_debug_event(bs
, BLKDBG_PWRITEV_RMW_TAIL
);
1503 ret
= bdrv_aligned_preadv(bs
, &req
, (offset
+ bytes
) & ~(align
- 1), align
,
1504 align
, &tail_qiov
, 0);
1508 bdrv_debug_event(bs
, BLKDBG_PWRITEV_RMW_AFTER_TAIL
);
1510 if (!use_local_qiov
) {
1511 qemu_iovec_init(&local_qiov
, qiov
->niov
+ 1);
1512 qemu_iovec_concat(&local_qiov
, qiov
, 0, qiov
->size
);
1513 use_local_qiov
= true;
1516 tail_bytes
= (offset
+ bytes
) & (align
- 1);
1517 qemu_iovec_add(&local_qiov
, tail_buf
+ tail_bytes
, align
- tail_bytes
);
1519 bytes
= ROUND_UP(bytes
, align
);
1522 ret
= bdrv_aligned_pwritev(bs
, &req
, offset
, bytes
, align
,
1523 use_local_qiov
? &local_qiov
: qiov
,
1528 if (use_local_qiov
) {
1529 qemu_iovec_destroy(&local_qiov
);
1531 qemu_vfree(head_buf
);
1532 qemu_vfree(tail_buf
);
1534 tracked_request_end(&req
);
1538 static int coroutine_fn
bdrv_co_do_writev(BdrvChild
*child
,
1539 int64_t sector_num
, int nb_sectors
, QEMUIOVector
*qiov
,
1540 BdrvRequestFlags flags
)
1542 if (nb_sectors
< 0 || nb_sectors
> BDRV_REQUEST_MAX_SECTORS
) {
1546 return bdrv_co_pwritev(child
->bs
, sector_num
<< BDRV_SECTOR_BITS
,
1547 nb_sectors
<< BDRV_SECTOR_BITS
, qiov
, flags
);
1550 int coroutine_fn
bdrv_co_writev(BdrvChild
*child
, int64_t sector_num
,
1551 int nb_sectors
, QEMUIOVector
*qiov
)
1553 trace_bdrv_co_writev(child
->bs
, sector_num
, nb_sectors
);
1555 return bdrv_co_do_writev(child
, sector_num
, nb_sectors
, qiov
, 0);
1558 int coroutine_fn
bdrv_co_pwrite_zeroes(BlockDriverState
*bs
,
1559 int64_t offset
, int count
,
1560 BdrvRequestFlags flags
)
1562 trace_bdrv_co_pwrite_zeroes(bs
, offset
, count
, flags
);
1564 if (!(bs
->open_flags
& BDRV_O_UNMAP
)) {
1565 flags
&= ~BDRV_REQ_MAY_UNMAP
;
1568 return bdrv_co_pwritev(bs
, offset
, count
, NULL
,
1569 BDRV_REQ_ZERO_WRITE
| flags
);
1572 typedef struct BdrvCoGetBlockStatusData
{
1573 BlockDriverState
*bs
;
1574 BlockDriverState
*base
;
1575 BlockDriverState
**file
;
1581 } BdrvCoGetBlockStatusData
;
1584 * Returns the allocation status of the specified sectors.
1585 * Drivers not implementing the functionality are assumed to not support
1586 * backing files, hence all their sectors are reported as allocated.
1588 * If 'sector_num' is beyond the end of the disk image the return value is 0
1589 * and 'pnum' is set to 0.
1591 * 'pnum' is set to the number of sectors (including and immediately following
1592 * the specified sector) that are known to be in the same
1593 * allocated/unallocated state.
1595 * 'nb_sectors' is the max value 'pnum' should be set to. If nb_sectors goes
1596 * beyond the end of the disk image it will be clamped.
1598 * If returned value is positive and BDRV_BLOCK_OFFSET_VALID bit is set, 'file'
1599 * points to the BDS which the sector range is allocated in.
1601 static int64_t coroutine_fn
bdrv_co_get_block_status(BlockDriverState
*bs
,
1603 int nb_sectors
, int *pnum
,
1604 BlockDriverState
**file
)
1606 int64_t total_sectors
;
1610 total_sectors
= bdrv_nb_sectors(bs
);
1611 if (total_sectors
< 0) {
1612 return total_sectors
;
1615 if (sector_num
>= total_sectors
) {
1620 n
= total_sectors
- sector_num
;
1621 if (n
< nb_sectors
) {
1625 if (!bs
->drv
->bdrv_co_get_block_status
) {
1627 ret
= BDRV_BLOCK_DATA
| BDRV_BLOCK_ALLOCATED
;
1628 if (bs
->drv
->protocol_name
) {
1629 ret
|= BDRV_BLOCK_OFFSET_VALID
| (sector_num
* BDRV_SECTOR_SIZE
);
1635 ret
= bs
->drv
->bdrv_co_get_block_status(bs
, sector_num
, nb_sectors
, pnum
,
1642 if (ret
& BDRV_BLOCK_RAW
) {
1643 assert(ret
& BDRV_BLOCK_OFFSET_VALID
);
1644 return bdrv_get_block_status(bs
->file
->bs
, ret
>> BDRV_SECTOR_BITS
,
1648 if (ret
& (BDRV_BLOCK_DATA
| BDRV_BLOCK_ZERO
)) {
1649 ret
|= BDRV_BLOCK_ALLOCATED
;
1651 if (bdrv_unallocated_blocks_are_zero(bs
)) {
1652 ret
|= BDRV_BLOCK_ZERO
;
1653 } else if (bs
->backing
) {
1654 BlockDriverState
*bs2
= bs
->backing
->bs
;
1655 int64_t nb_sectors2
= bdrv_nb_sectors(bs2
);
1656 if (nb_sectors2
>= 0 && sector_num
>= nb_sectors2
) {
1657 ret
|= BDRV_BLOCK_ZERO
;
1662 if (*file
&& *file
!= bs
&&
1663 (ret
& BDRV_BLOCK_DATA
) && !(ret
& BDRV_BLOCK_ZERO
) &&
1664 (ret
& BDRV_BLOCK_OFFSET_VALID
)) {
1665 BlockDriverState
*file2
;
1668 ret2
= bdrv_co_get_block_status(*file
, ret
>> BDRV_SECTOR_BITS
,
1669 *pnum
, &file_pnum
, &file2
);
1671 /* Ignore errors. This is just providing extra information, it
1672 * is useful but not necessary.
1675 /* !file_pnum indicates an offset at or beyond the EOF; it is
1676 * perfectly valid for the format block driver to point to such
1677 * offsets, so catch it and mark everything as zero */
1678 ret
|= BDRV_BLOCK_ZERO
;
1680 /* Limit request to the range reported by the protocol driver */
1682 ret
|= (ret2
& BDRV_BLOCK_ZERO
);
1690 static int64_t coroutine_fn
bdrv_co_get_block_status_above(BlockDriverState
*bs
,
1691 BlockDriverState
*base
,
1695 BlockDriverState
**file
)
1697 BlockDriverState
*p
;
1701 for (p
= bs
; p
!= base
; p
= backing_bs(p
)) {
1702 ret
= bdrv_co_get_block_status(p
, sector_num
, nb_sectors
, pnum
, file
);
1703 if (ret
< 0 || ret
& BDRV_BLOCK_ALLOCATED
) {
1706 /* [sector_num, pnum] unallocated on this layer, which could be only
1707 * the first part of [sector_num, nb_sectors]. */
1708 nb_sectors
= MIN(nb_sectors
, *pnum
);
1713 /* Coroutine wrapper for bdrv_get_block_status_above() */
1714 static void coroutine_fn
bdrv_get_block_status_above_co_entry(void *opaque
)
1716 BdrvCoGetBlockStatusData
*data
= opaque
;
1718 data
->ret
= bdrv_co_get_block_status_above(data
->bs
, data
->base
,
1727 * Synchronous wrapper around bdrv_co_get_block_status_above().
1729 * See bdrv_co_get_block_status_above() for details.
1731 int64_t bdrv_get_block_status_above(BlockDriverState
*bs
,
1732 BlockDriverState
*base
,
1734 int nb_sectors
, int *pnum
,
1735 BlockDriverState
**file
)
1738 BdrvCoGetBlockStatusData data
= {
1742 .sector_num
= sector_num
,
1743 .nb_sectors
= nb_sectors
,
1748 if (qemu_in_coroutine()) {
1749 /* Fast-path if already in coroutine context */
1750 bdrv_get_block_status_above_co_entry(&data
);
1752 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
1754 co
= qemu_coroutine_create(bdrv_get_block_status_above_co_entry
);
1755 qemu_coroutine_enter(co
, &data
);
1756 while (!data
.done
) {
1757 aio_poll(aio_context
, true);
1763 int64_t bdrv_get_block_status(BlockDriverState
*bs
,
1765 int nb_sectors
, int *pnum
,
1766 BlockDriverState
**file
)
1768 return bdrv_get_block_status_above(bs
, backing_bs(bs
),
1769 sector_num
, nb_sectors
, pnum
, file
);
1772 int coroutine_fn
bdrv_is_allocated(BlockDriverState
*bs
, int64_t sector_num
,
1773 int nb_sectors
, int *pnum
)
1775 BlockDriverState
*file
;
1776 int64_t ret
= bdrv_get_block_status(bs
, sector_num
, nb_sectors
, pnum
,
1781 return !!(ret
& BDRV_BLOCK_ALLOCATED
);
1785 * Given an image chain: ... -> [BASE] -> [INTER1] -> [INTER2] -> [TOP]
1787 * Return true if the given sector is allocated in any image between
1788 * BASE and TOP (inclusive). BASE can be NULL to check if the given
1789 * sector is allocated in any image of the chain. Return false otherwise.
1791 * 'pnum' is set to the number of sectors (including and immediately following
1792 * the specified sector) that are known to be in the same
1793 * allocated/unallocated state.
1796 int bdrv_is_allocated_above(BlockDriverState
*top
,
1797 BlockDriverState
*base
,
1799 int nb_sectors
, int *pnum
)
1801 BlockDriverState
*intermediate
;
1802 int ret
, n
= nb_sectors
;
1805 while (intermediate
&& intermediate
!= base
) {
1807 ret
= bdrv_is_allocated(intermediate
, sector_num
, nb_sectors
,
1817 * [sector_num, nb_sectors] is unallocated on top but intermediate
1820 * [sector_num+x, nr_sectors] allocated.
1822 if (n
> pnum_inter
&&
1823 (intermediate
== top
||
1824 sector_num
+ pnum_inter
< intermediate
->total_sectors
)) {
1828 intermediate
= backing_bs(intermediate
);
1835 int bdrv_write_compressed(BlockDriverState
*bs
, int64_t sector_num
,
1836 const uint8_t *buf
, int nb_sectors
)
1838 BlockDriver
*drv
= bs
->drv
;
1844 if (!drv
->bdrv_write_compressed
) {
1847 ret
= bdrv_check_request(bs
, sector_num
, nb_sectors
);
1852 assert(QLIST_EMPTY(&bs
->dirty_bitmaps
));
1854 return drv
->bdrv_write_compressed(bs
, sector_num
, buf
, nb_sectors
);
1857 typedef struct BdrvVmstateCo
{
1858 BlockDriverState
*bs
;
1865 static int coroutine_fn
1866 bdrv_co_rw_vmstate(BlockDriverState
*bs
, QEMUIOVector
*qiov
, int64_t pos
,
1869 BlockDriver
*drv
= bs
->drv
;
1873 } else if (drv
->bdrv_load_vmstate
) {
1874 return is_read
? drv
->bdrv_load_vmstate(bs
, qiov
, pos
)
1875 : drv
->bdrv_save_vmstate(bs
, qiov
, pos
);
1876 } else if (bs
->file
) {
1877 return bdrv_co_rw_vmstate(bs
->file
->bs
, qiov
, pos
, is_read
);
1883 static void coroutine_fn
bdrv_co_rw_vmstate_entry(void *opaque
)
1885 BdrvVmstateCo
*co
= opaque
;
1886 co
->ret
= bdrv_co_rw_vmstate(co
->bs
, co
->qiov
, co
->pos
, co
->is_read
);
1890 bdrv_rw_vmstate(BlockDriverState
*bs
, QEMUIOVector
*qiov
, int64_t pos
,
1893 if (qemu_in_coroutine()) {
1894 return bdrv_co_rw_vmstate(bs
, qiov
, pos
, is_read
);
1896 BdrvVmstateCo data
= {
1901 .ret
= -EINPROGRESS
,
1903 Coroutine
*co
= qemu_coroutine_create(bdrv_co_rw_vmstate_entry
);
1905 qemu_coroutine_enter(co
, &data
);
1906 while (data
.ret
== -EINPROGRESS
) {
1907 aio_poll(bdrv_get_aio_context(bs
), true);
1913 int bdrv_save_vmstate(BlockDriverState
*bs
, const uint8_t *buf
,
1914 int64_t pos
, int size
)
1917 struct iovec iov
= {
1918 .iov_base
= (void *) buf
,
1923 qemu_iovec_init_external(&qiov
, &iov
, 1);
1925 ret
= bdrv_writev_vmstate(bs
, &qiov
, pos
);
1933 int bdrv_writev_vmstate(BlockDriverState
*bs
, QEMUIOVector
*qiov
, int64_t pos
)
1935 return bdrv_rw_vmstate(bs
, qiov
, pos
, false);
1938 int bdrv_load_vmstate(BlockDriverState
*bs
, uint8_t *buf
,
1939 int64_t pos
, int size
)
1942 struct iovec iov
= {
1948 qemu_iovec_init_external(&qiov
, &iov
, 1);
1949 ret
= bdrv_readv_vmstate(bs
, &qiov
, pos
);
1957 int bdrv_readv_vmstate(BlockDriverState
*bs
, QEMUIOVector
*qiov
, int64_t pos
)
1959 return bdrv_rw_vmstate(bs
, qiov
, pos
, true);
1962 /**************************************************************/
1965 BlockAIOCB
*bdrv_aio_readv(BdrvChild
*child
, int64_t sector_num
,
1966 QEMUIOVector
*qiov
, int nb_sectors
,
1967 BlockCompletionFunc
*cb
, void *opaque
)
1969 trace_bdrv_aio_readv(child
->bs
, sector_num
, nb_sectors
, opaque
);
1971 return bdrv_co_aio_rw_vector(child
, sector_num
, qiov
, nb_sectors
, 0,
1975 BlockAIOCB
*bdrv_aio_writev(BdrvChild
*child
, int64_t sector_num
,
1976 QEMUIOVector
*qiov
, int nb_sectors
,
1977 BlockCompletionFunc
*cb
, void *opaque
)
1979 trace_bdrv_aio_writev(child
->bs
, sector_num
, nb_sectors
, opaque
);
1981 return bdrv_co_aio_rw_vector(child
, sector_num
, qiov
, nb_sectors
, 0,
1985 void bdrv_aio_cancel(BlockAIOCB
*acb
)
1988 bdrv_aio_cancel_async(acb
);
1989 while (acb
->refcnt
> 1) {
1990 if (acb
->aiocb_info
->get_aio_context
) {
1991 aio_poll(acb
->aiocb_info
->get_aio_context(acb
), true);
1992 } else if (acb
->bs
) {
1993 aio_poll(bdrv_get_aio_context(acb
->bs
), true);
1998 qemu_aio_unref(acb
);
2001 /* Async version of aio cancel. The caller is not blocked if the acb implements
2002 * cancel_async, otherwise we do nothing and let the request normally complete.
2003 * In either case the completion callback must be called. */
2004 void bdrv_aio_cancel_async(BlockAIOCB
*acb
)
2006 if (acb
->aiocb_info
->cancel_async
) {
2007 acb
->aiocb_info
->cancel_async(acb
);
2011 /**************************************************************/
2012 /* async block device emulation */
2014 typedef struct BlockRequest
{
2016 /* Used during read, write, trim */
2023 /* Used during ioctl */
2029 BlockCompletionFunc
*cb
;
2035 typedef struct BlockAIOCBCoroutine
{
2043 } BlockAIOCBCoroutine
;
2045 static const AIOCBInfo bdrv_em_co_aiocb_info
= {
2046 .aiocb_size
= sizeof(BlockAIOCBCoroutine
),
2049 static void bdrv_co_complete(BlockAIOCBCoroutine
*acb
)
2051 if (!acb
->need_bh
) {
2052 acb
->common
.cb(acb
->common
.opaque
, acb
->req
.error
);
2053 qemu_aio_unref(acb
);
2057 static void bdrv_co_em_bh(void *opaque
)
2059 BlockAIOCBCoroutine
*acb
= opaque
;
2061 assert(!acb
->need_bh
);
2062 qemu_bh_delete(acb
->bh
);
2063 bdrv_co_complete(acb
);
2066 static void bdrv_co_maybe_schedule_bh(BlockAIOCBCoroutine
*acb
)
2068 acb
->need_bh
= false;
2069 if (acb
->req
.error
!= -EINPROGRESS
) {
2070 BlockDriverState
*bs
= acb
->common
.bs
;
2072 acb
->bh
= aio_bh_new(bdrv_get_aio_context(bs
), bdrv_co_em_bh
, acb
);
2073 qemu_bh_schedule(acb
->bh
);
2077 /* Invoke bdrv_co_do_readv/bdrv_co_do_writev */
2078 static void coroutine_fn
bdrv_co_do_rw(void *opaque
)
2080 BlockAIOCBCoroutine
*acb
= opaque
;
2082 if (!acb
->is_write
) {
2083 acb
->req
.error
= bdrv_co_do_readv(acb
->child
, acb
->req
.sector
,
2084 acb
->req
.nb_sectors
, acb
->req
.qiov
, acb
->req
.flags
);
2086 acb
->req
.error
= bdrv_co_do_writev(acb
->child
, acb
->req
.sector
,
2087 acb
->req
.nb_sectors
, acb
->req
.qiov
, acb
->req
.flags
);
2090 bdrv_co_complete(acb
);
2093 static BlockAIOCB
*bdrv_co_aio_rw_vector(BdrvChild
*child
,
2097 BdrvRequestFlags flags
,
2098 BlockCompletionFunc
*cb
,
2103 BlockAIOCBCoroutine
*acb
;
2105 acb
= qemu_aio_get(&bdrv_em_co_aiocb_info
, child
->bs
, cb
, opaque
);
2107 acb
->need_bh
= true;
2108 acb
->req
.error
= -EINPROGRESS
;
2109 acb
->req
.sector
= sector_num
;
2110 acb
->req
.nb_sectors
= nb_sectors
;
2111 acb
->req
.qiov
= qiov
;
2112 acb
->req
.flags
= flags
;
2113 acb
->is_write
= is_write
;
2115 co
= qemu_coroutine_create(bdrv_co_do_rw
);
2116 qemu_coroutine_enter(co
, acb
);
2118 bdrv_co_maybe_schedule_bh(acb
);
2119 return &acb
->common
;
2122 static void coroutine_fn
bdrv_aio_flush_co_entry(void *opaque
)
2124 BlockAIOCBCoroutine
*acb
= opaque
;
2125 BlockDriverState
*bs
= acb
->common
.bs
;
2127 acb
->req
.error
= bdrv_co_flush(bs
);
2128 bdrv_co_complete(acb
);
2131 BlockAIOCB
*bdrv_aio_flush(BlockDriverState
*bs
,
2132 BlockCompletionFunc
*cb
, void *opaque
)
2134 trace_bdrv_aio_flush(bs
, opaque
);
2137 BlockAIOCBCoroutine
*acb
;
2139 acb
= qemu_aio_get(&bdrv_em_co_aiocb_info
, bs
, cb
, opaque
);
2140 acb
->need_bh
= true;
2141 acb
->req
.error
= -EINPROGRESS
;
2143 co
= qemu_coroutine_create(bdrv_aio_flush_co_entry
);
2144 qemu_coroutine_enter(co
, acb
);
2146 bdrv_co_maybe_schedule_bh(acb
);
2147 return &acb
->common
;
2150 static void coroutine_fn
bdrv_aio_discard_co_entry(void *opaque
)
2152 BlockAIOCBCoroutine
*acb
= opaque
;
2153 BlockDriverState
*bs
= acb
->common
.bs
;
2155 acb
->req
.error
= bdrv_co_discard(bs
, acb
->req
.sector
, acb
->req
.nb_sectors
);
2156 bdrv_co_complete(acb
);
2159 BlockAIOCB
*bdrv_aio_discard(BlockDriverState
*bs
,
2160 int64_t sector_num
, int nb_sectors
,
2161 BlockCompletionFunc
*cb
, void *opaque
)
2164 BlockAIOCBCoroutine
*acb
;
2166 trace_bdrv_aio_discard(bs
, sector_num
, nb_sectors
, opaque
);
2168 acb
= qemu_aio_get(&bdrv_em_co_aiocb_info
, bs
, cb
, opaque
);
2169 acb
->need_bh
= true;
2170 acb
->req
.error
= -EINPROGRESS
;
2171 acb
->req
.sector
= sector_num
;
2172 acb
->req
.nb_sectors
= nb_sectors
;
2173 co
= qemu_coroutine_create(bdrv_aio_discard_co_entry
);
2174 qemu_coroutine_enter(co
, acb
);
2176 bdrv_co_maybe_schedule_bh(acb
);
2177 return &acb
->common
;
2180 void *qemu_aio_get(const AIOCBInfo
*aiocb_info
, BlockDriverState
*bs
,
2181 BlockCompletionFunc
*cb
, void *opaque
)
2185 acb
= g_malloc(aiocb_info
->aiocb_size
);
2186 acb
->aiocb_info
= aiocb_info
;
2189 acb
->opaque
= opaque
;
2194 void qemu_aio_ref(void *p
)
2196 BlockAIOCB
*acb
= p
;
2200 void qemu_aio_unref(void *p
)
2202 BlockAIOCB
*acb
= p
;
2203 assert(acb
->refcnt
> 0);
2204 if (--acb
->refcnt
== 0) {
2209 /**************************************************************/
2210 /* Coroutine block device emulation */
2212 static void coroutine_fn
bdrv_flush_co_entry(void *opaque
)
2214 RwCo
*rwco
= opaque
;
2216 rwco
->ret
= bdrv_co_flush(rwco
->bs
);
2219 int coroutine_fn
bdrv_co_flush(BlockDriverState
*bs
)
2222 BdrvTrackedRequest req
;
2224 if (!bs
|| !bdrv_is_inserted(bs
) || bdrv_is_read_only(bs
) ||
2229 tracked_request_begin(&req
, bs
, 0, 0, BDRV_TRACKED_FLUSH
);
2231 /* Write back all layers by calling one driver function */
2232 if (bs
->drv
->bdrv_co_flush
) {
2233 ret
= bs
->drv
->bdrv_co_flush(bs
);
2237 /* Write back cached data to the OS even with cache=unsafe */
2238 BLKDBG_EVENT(bs
->file
, BLKDBG_FLUSH_TO_OS
);
2239 if (bs
->drv
->bdrv_co_flush_to_os
) {
2240 ret
= bs
->drv
->bdrv_co_flush_to_os(bs
);
2246 /* But don't actually force it to the disk with cache=unsafe */
2247 if (bs
->open_flags
& BDRV_O_NO_FLUSH
) {
2251 BLKDBG_EVENT(bs
->file
, BLKDBG_FLUSH_TO_DISK
);
2252 if (bs
->drv
->bdrv_co_flush_to_disk
) {
2253 ret
= bs
->drv
->bdrv_co_flush_to_disk(bs
);
2254 } else if (bs
->drv
->bdrv_aio_flush
) {
2256 CoroutineIOCompletion co
= {
2257 .coroutine
= qemu_coroutine_self(),
2260 acb
= bs
->drv
->bdrv_aio_flush(bs
, bdrv_co_io_em_complete
, &co
);
2264 qemu_coroutine_yield();
2269 * Some block drivers always operate in either writethrough or unsafe
2270 * mode and don't support bdrv_flush therefore. Usually qemu doesn't
2271 * know how the server works (because the behaviour is hardcoded or
2272 * depends on server-side configuration), so we can't ensure that
2273 * everything is safe on disk. Returning an error doesn't work because
2274 * that would break guests even if the server operates in writethrough
2277 * Let's hope the user knows what he's doing.
2285 /* Now flush the underlying protocol. It will also have BDRV_O_NO_FLUSH
2286 * in the case of cache=unsafe, so there are no useless flushes.
2289 ret
= bs
->file
? bdrv_co_flush(bs
->file
->bs
) : 0;
2291 tracked_request_end(&req
);
2295 int bdrv_flush(BlockDriverState
*bs
)
2303 if (qemu_in_coroutine()) {
2304 /* Fast-path if already in coroutine context */
2305 bdrv_flush_co_entry(&rwco
);
2307 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
2309 co
= qemu_coroutine_create(bdrv_flush_co_entry
);
2310 qemu_coroutine_enter(co
, &rwco
);
2311 while (rwco
.ret
== NOT_DONE
) {
2312 aio_poll(aio_context
, true);
2319 typedef struct DiscardCo
{
2320 BlockDriverState
*bs
;
2325 static void coroutine_fn
bdrv_discard_co_entry(void *opaque
)
2327 DiscardCo
*rwco
= opaque
;
2329 rwco
->ret
= bdrv_co_discard(rwco
->bs
, rwco
->sector_num
, rwco
->nb_sectors
);
2332 int coroutine_fn
bdrv_co_discard(BlockDriverState
*bs
, int64_t sector_num
,
2335 BdrvTrackedRequest req
;
2336 int max_discard
, ret
;
2342 ret
= bdrv_check_request(bs
, sector_num
, nb_sectors
);
2345 } else if (bs
->read_only
) {
2348 assert(!(bs
->open_flags
& BDRV_O_INACTIVE
));
2350 /* Do nothing if disabled. */
2351 if (!(bs
->open_flags
& BDRV_O_UNMAP
)) {
2355 if (!bs
->drv
->bdrv_co_discard
&& !bs
->drv
->bdrv_aio_discard
) {
2359 tracked_request_begin(&req
, bs
, sector_num
<< BDRV_SECTOR_BITS
,
2360 nb_sectors
<< BDRV_SECTOR_BITS
, BDRV_TRACKED_DISCARD
);
2362 ret
= notifier_with_return_list_notify(&bs
->before_write_notifiers
, &req
);
2367 max_discard
= MIN_NON_ZERO(bs
->bl
.max_pdiscard
>> BDRV_SECTOR_BITS
,
2368 BDRV_REQUEST_MAX_SECTORS
);
2369 while (nb_sectors
> 0) {
2371 int num
= nb_sectors
;
2372 int discard_alignment
= bs
->bl
.pdiscard_alignment
>> BDRV_SECTOR_BITS
;
2375 if (discard_alignment
&&
2376 num
>= discard_alignment
&&
2377 sector_num
% discard_alignment
) {
2378 if (num
> discard_alignment
) {
2379 num
= discard_alignment
;
2381 num
-= sector_num
% discard_alignment
;
2384 /* limit request size */
2385 if (num
> max_discard
) {
2389 if (bs
->drv
->bdrv_co_discard
) {
2390 ret
= bs
->drv
->bdrv_co_discard(bs
, sector_num
, num
);
2393 CoroutineIOCompletion co
= {
2394 .coroutine
= qemu_coroutine_self(),
2397 acb
= bs
->drv
->bdrv_aio_discard(bs
, sector_num
, nb_sectors
,
2398 bdrv_co_io_em_complete
, &co
);
2403 qemu_coroutine_yield();
2407 if (ret
&& ret
!= -ENOTSUP
) {
2416 bdrv_set_dirty(bs
, req
.offset
>> BDRV_SECTOR_BITS
,
2417 req
.bytes
>> BDRV_SECTOR_BITS
);
2418 tracked_request_end(&req
);
2422 int bdrv_discard(BlockDriverState
*bs
, int64_t sector_num
, int nb_sectors
)
2427 .sector_num
= sector_num
,
2428 .nb_sectors
= nb_sectors
,
2432 if (qemu_in_coroutine()) {
2433 /* Fast-path if already in coroutine context */
2434 bdrv_discard_co_entry(&rwco
);
2436 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
2438 co
= qemu_coroutine_create(bdrv_discard_co_entry
);
2439 qemu_coroutine_enter(co
, &rwco
);
2440 while (rwco
.ret
== NOT_DONE
) {
2441 aio_poll(aio_context
, true);
2448 static int bdrv_co_do_ioctl(BlockDriverState
*bs
, int req
, void *buf
)
2450 BlockDriver
*drv
= bs
->drv
;
2451 BdrvTrackedRequest tracked_req
;
2452 CoroutineIOCompletion co
= {
2453 .coroutine
= qemu_coroutine_self(),
2457 tracked_request_begin(&tracked_req
, bs
, 0, 0, BDRV_TRACKED_IOCTL
);
2458 if (!drv
|| !drv
->bdrv_aio_ioctl
) {
2463 acb
= drv
->bdrv_aio_ioctl(bs
, req
, buf
, bdrv_co_io_em_complete
, &co
);
2468 qemu_coroutine_yield();
2470 tracked_request_end(&tracked_req
);
2475 BlockDriverState
*bs
;
2481 static void coroutine_fn
bdrv_co_ioctl_entry(void *opaque
)
2483 BdrvIoctlCoData
*data
= opaque
;
2484 data
->ret
= bdrv_co_do_ioctl(data
->bs
, data
->req
, data
->buf
);
2487 /* needed for generic scsi interface */
2488 int bdrv_ioctl(BlockDriverState
*bs
, unsigned long int req
, void *buf
)
2490 BdrvIoctlCoData data
= {
2494 .ret
= -EINPROGRESS
,
2497 if (qemu_in_coroutine()) {
2498 /* Fast-path if already in coroutine context */
2499 bdrv_co_ioctl_entry(&data
);
2501 Coroutine
*co
= qemu_coroutine_create(bdrv_co_ioctl_entry
);
2503 qemu_coroutine_enter(co
, &data
);
2504 while (data
.ret
== -EINPROGRESS
) {
2505 aio_poll(bdrv_get_aio_context(bs
), true);
2511 static void coroutine_fn
bdrv_co_aio_ioctl_entry(void *opaque
)
2513 BlockAIOCBCoroutine
*acb
= opaque
;
2514 acb
->req
.error
= bdrv_co_do_ioctl(acb
->common
.bs
,
2515 acb
->req
.req
, acb
->req
.buf
);
2516 bdrv_co_complete(acb
);
2519 BlockAIOCB
*bdrv_aio_ioctl(BlockDriverState
*bs
,
2520 unsigned long int req
, void *buf
,
2521 BlockCompletionFunc
*cb
, void *opaque
)
2523 BlockAIOCBCoroutine
*acb
= qemu_aio_get(&bdrv_em_co_aiocb_info
,
2527 acb
->need_bh
= true;
2528 acb
->req
.error
= -EINPROGRESS
;
2531 co
= qemu_coroutine_create(bdrv_co_aio_ioctl_entry
);
2532 qemu_coroutine_enter(co
, acb
);
2534 bdrv_co_maybe_schedule_bh(acb
);
2535 return &acb
->common
;
2538 void *qemu_blockalign(BlockDriverState
*bs
, size_t size
)
2540 return qemu_memalign(bdrv_opt_mem_align(bs
), size
);
2543 void *qemu_blockalign0(BlockDriverState
*bs
, size_t size
)
2545 return memset(qemu_blockalign(bs
, size
), 0, size
);
2548 void *qemu_try_blockalign(BlockDriverState
*bs
, size_t size
)
2550 size_t align
= bdrv_opt_mem_align(bs
);
2552 /* Ensure that NULL is never returned on success */
2558 return qemu_try_memalign(align
, size
);
2561 void *qemu_try_blockalign0(BlockDriverState
*bs
, size_t size
)
2563 void *mem
= qemu_try_blockalign(bs
, size
);
2566 memset(mem
, 0, size
);
2573 * Check if all memory in this vector is sector aligned.
2575 bool bdrv_qiov_is_aligned(BlockDriverState
*bs
, QEMUIOVector
*qiov
)
2578 size_t alignment
= bdrv_min_mem_align(bs
);
2580 for (i
= 0; i
< qiov
->niov
; i
++) {
2581 if ((uintptr_t) qiov
->iov
[i
].iov_base
% alignment
) {
2584 if (qiov
->iov
[i
].iov_len
% alignment
) {
2592 void bdrv_add_before_write_notifier(BlockDriverState
*bs
,
2593 NotifierWithReturn
*notifier
)
2595 notifier_with_return_list_add(&bs
->before_write_notifiers
, notifier
);
2598 void bdrv_io_plug(BlockDriverState
*bs
)
2602 QLIST_FOREACH(child
, &bs
->children
, next
) {
2603 bdrv_io_plug(child
->bs
);
2606 if (bs
->io_plugged
++ == 0 && bs
->io_plug_disabled
== 0) {
2607 BlockDriver
*drv
= bs
->drv
;
2608 if (drv
&& drv
->bdrv_io_plug
) {
2609 drv
->bdrv_io_plug(bs
);
2614 void bdrv_io_unplug(BlockDriverState
*bs
)
2618 assert(bs
->io_plugged
);
2619 if (--bs
->io_plugged
== 0 && bs
->io_plug_disabled
== 0) {
2620 BlockDriver
*drv
= bs
->drv
;
2621 if (drv
&& drv
->bdrv_io_unplug
) {
2622 drv
->bdrv_io_unplug(bs
);
2626 QLIST_FOREACH(child
, &bs
->children
, next
) {
2627 bdrv_io_unplug(child
->bs
);
2631 void bdrv_io_unplugged_begin(BlockDriverState
*bs
)
2635 if (bs
->io_plug_disabled
++ == 0 && bs
->io_plugged
> 0) {
2636 BlockDriver
*drv
= bs
->drv
;
2637 if (drv
&& drv
->bdrv_io_unplug
) {
2638 drv
->bdrv_io_unplug(bs
);
2642 QLIST_FOREACH(child
, &bs
->children
, next
) {
2643 bdrv_io_unplugged_begin(child
->bs
);
2647 void bdrv_io_unplugged_end(BlockDriverState
*bs
)
2651 assert(bs
->io_plug_disabled
);
2652 QLIST_FOREACH(child
, &bs
->children
, next
) {
2653 bdrv_io_unplugged_end(child
->bs
);
2656 if (--bs
->io_plug_disabled
== 0 && bs
->io_plugged
> 0) {
2657 BlockDriver
*drv
= bs
->drv
;
2658 if (drv
&& drv
->bdrv_io_plug
) {
2659 drv
->bdrv_io_plug(bs
);