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
);
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
{
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
->child
, rwco
->offset
,
570 rwco
->qiov
->size
, rwco
->qiov
,
573 rwco
->ret
= bdrv_co_pwritev(rwco
->child
, rwco
->offset
,
574 rwco
->qiov
->size
, rwco
->qiov
,
580 * Process a vectored synchronous request using coroutines
582 static int bdrv_prwv_co(BdrvChild
*child
, 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(child
->bs
);
602 co
= qemu_coroutine_create(bdrv_rw_co_entry
, &rwco
);
603 qemu_coroutine_enter(co
);
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(BdrvChild
*child
, 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(child
, 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
, 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
, sector_num
, (uint8_t *)buf
, nb_sectors
, true, 0);
651 int bdrv_pwrite_zeroes(BdrvChild
*child
, int64_t offset
,
652 int count
, BdrvRequestFlags flags
)
660 qemu_iovec_init_external(&qiov
, &iov
, 1);
661 return bdrv_prwv_co(child
, offset
, &qiov
, true,
662 BDRV_REQ_ZERO_WRITE
| flags
);
666 * Completely zero out a block device with the help of bdrv_pwrite_zeroes.
667 * The operation is sped up by checking the block status and only writing
668 * zeroes to the device if they currently do not return zeroes. Optional
669 * flags are passed through to bdrv_pwrite_zeroes (e.g. BDRV_REQ_MAY_UNMAP,
672 * Returns < 0 on error, 0 on success. For error codes see bdrv_write().
674 int bdrv_make_zero(BdrvChild
*child
, BdrvRequestFlags flags
)
676 int64_t target_sectors
, ret
, nb_sectors
, sector_num
= 0;
677 BlockDriverState
*bs
= child
->bs
;
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(child
, 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
, 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
, 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
);
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(BdrvChild
*child
,
1065 int64_t offset
, unsigned int bytes
, QEMUIOVector
*qiov
,
1066 BdrvRequestFlags flags
)
1068 BlockDriverState
*bs
= child
->bs
;
1069 BlockDriver
*drv
= bs
->drv
;
1070 BdrvTrackedRequest req
;
1072 uint64_t align
= bs
->bl
.request_alignment
;
1073 uint8_t *head_buf
= NULL
;
1074 uint8_t *tail_buf
= NULL
;
1075 QEMUIOVector local_qiov
;
1076 bool use_local_qiov
= false;
1083 ret
= bdrv_check_byte_request(bs
, offset
, bytes
);
1088 /* Don't do copy-on-read if we read data before write operation */
1089 if (bs
->copy_on_read
&& !(flags
& BDRV_REQ_NO_SERIALISING
)) {
1090 flags
|= BDRV_REQ_COPY_ON_READ
;
1093 /* Align read if necessary by padding qiov */
1094 if (offset
& (align
- 1)) {
1095 head_buf
= qemu_blockalign(bs
, align
);
1096 qemu_iovec_init(&local_qiov
, qiov
->niov
+ 2);
1097 qemu_iovec_add(&local_qiov
, head_buf
, offset
& (align
- 1));
1098 qemu_iovec_concat(&local_qiov
, qiov
, 0, qiov
->size
);
1099 use_local_qiov
= true;
1101 bytes
+= offset
& (align
- 1);
1102 offset
= offset
& ~(align
- 1);
1105 if ((offset
+ bytes
) & (align
- 1)) {
1106 if (!use_local_qiov
) {
1107 qemu_iovec_init(&local_qiov
, qiov
->niov
+ 1);
1108 qemu_iovec_concat(&local_qiov
, qiov
, 0, qiov
->size
);
1109 use_local_qiov
= true;
1111 tail_buf
= qemu_blockalign(bs
, align
);
1112 qemu_iovec_add(&local_qiov
, tail_buf
,
1113 align
- ((offset
+ bytes
) & (align
- 1)));
1115 bytes
= ROUND_UP(bytes
, align
);
1118 tracked_request_begin(&req
, bs
, offset
, bytes
, BDRV_TRACKED_READ
);
1119 ret
= bdrv_aligned_preadv(bs
, &req
, offset
, bytes
, align
,
1120 use_local_qiov
? &local_qiov
: qiov
,
1122 tracked_request_end(&req
);
1124 if (use_local_qiov
) {
1125 qemu_iovec_destroy(&local_qiov
);
1126 qemu_vfree(head_buf
);
1127 qemu_vfree(tail_buf
);
1133 static int coroutine_fn
bdrv_co_do_readv(BdrvChild
*child
,
1134 int64_t sector_num
, int nb_sectors
, QEMUIOVector
*qiov
,
1135 BdrvRequestFlags flags
)
1137 if (nb_sectors
< 0 || nb_sectors
> BDRV_REQUEST_MAX_SECTORS
) {
1141 return bdrv_co_preadv(child
, sector_num
<< BDRV_SECTOR_BITS
,
1142 nb_sectors
<< BDRV_SECTOR_BITS
, qiov
, flags
);
1145 int coroutine_fn
bdrv_co_readv(BdrvChild
*child
, int64_t sector_num
,
1146 int nb_sectors
, QEMUIOVector
*qiov
)
1148 trace_bdrv_co_readv(child
->bs
, sector_num
, nb_sectors
);
1150 return bdrv_co_do_readv(child
, sector_num
, nb_sectors
, qiov
, 0);
1153 /* Maximum buffer for write zeroes fallback, in bytes */
1154 #define MAX_WRITE_ZEROES_BOUNCE_BUFFER (32768 << BDRV_SECTOR_BITS)
1156 static int coroutine_fn
bdrv_co_do_pwrite_zeroes(BlockDriverState
*bs
,
1157 int64_t offset
, int count
, BdrvRequestFlags flags
)
1159 BlockDriver
*drv
= bs
->drv
;
1161 struct iovec iov
= {0};
1163 bool need_flush
= false;
1167 int max_write_zeroes
= MIN_NON_ZERO(bs
->bl
.max_pwrite_zeroes
, INT_MAX
);
1168 int alignment
= MAX(bs
->bl
.pwrite_zeroes_alignment
,
1169 bs
->bl
.request_alignment
);
1171 assert(is_power_of_2(alignment
));
1172 head
= offset
& (alignment
- 1);
1173 tail
= (offset
+ count
) & (alignment
- 1);
1174 max_write_zeroes
&= ~(alignment
- 1);
1176 while (count
> 0 && !ret
) {
1179 /* Align request. Block drivers can expect the "bulk" of the request
1180 * to be aligned, and that unaligned requests do not cross cluster
1184 /* Make a small request up to the first aligned sector. */
1185 num
= MIN(count
, alignment
- head
);
1187 } else if (tail
&& num
> alignment
) {
1188 /* Shorten the request to the last aligned sector. */
1192 /* limit request size */
1193 if (num
> max_write_zeroes
) {
1194 num
= max_write_zeroes
;
1198 /* First try the efficient write zeroes operation */
1199 if (drv
->bdrv_co_pwrite_zeroes
) {
1200 ret
= drv
->bdrv_co_pwrite_zeroes(bs
, offset
, num
,
1201 flags
& bs
->supported_zero_flags
);
1202 if (ret
!= -ENOTSUP
&& (flags
& BDRV_REQ_FUA
) &&
1203 !(bs
->supported_zero_flags
& BDRV_REQ_FUA
)) {
1207 assert(!bs
->supported_zero_flags
);
1210 if (ret
== -ENOTSUP
) {
1211 /* Fall back to bounce buffer if write zeroes is unsupported */
1212 int max_transfer
= MIN_NON_ZERO(bs
->bl
.max_transfer
,
1213 MAX_WRITE_ZEROES_BOUNCE_BUFFER
);
1214 BdrvRequestFlags write_flags
= flags
& ~BDRV_REQ_ZERO_WRITE
;
1216 if ((flags
& BDRV_REQ_FUA
) &&
1217 !(bs
->supported_write_flags
& BDRV_REQ_FUA
)) {
1218 /* No need for bdrv_driver_pwrite() to do a fallback
1219 * flush on each chunk; use just one at the end */
1220 write_flags
&= ~BDRV_REQ_FUA
;
1223 num
= MIN(num
, max_transfer
);
1225 if (iov
.iov_base
== NULL
) {
1226 iov
.iov_base
= qemu_try_blockalign(bs
, num
);
1227 if (iov
.iov_base
== NULL
) {
1231 memset(iov
.iov_base
, 0, num
);
1233 qemu_iovec_init_external(&qiov
, &iov
, 1);
1235 ret
= bdrv_driver_pwritev(bs
, offset
, num
, &qiov
, write_flags
);
1237 /* Keep bounce buffer around if it is big enough for all
1238 * all future requests.
1240 if (num
< max_transfer
) {
1241 qemu_vfree(iov
.iov_base
);
1242 iov
.iov_base
= NULL
;
1251 if (ret
== 0 && need_flush
) {
1252 ret
= bdrv_co_flush(bs
);
1254 qemu_vfree(iov
.iov_base
);
1259 * Forwards an already correctly aligned write request to the BlockDriver.
1261 static int coroutine_fn
bdrv_aligned_pwritev(BlockDriverState
*bs
,
1262 BdrvTrackedRequest
*req
, int64_t offset
, unsigned int bytes
,
1263 int64_t align
, QEMUIOVector
*qiov
, int flags
)
1265 BlockDriver
*drv
= bs
->drv
;
1269 int64_t start_sector
= offset
>> BDRV_SECTOR_BITS
;
1270 int64_t end_sector
= DIV_ROUND_UP(offset
+ bytes
, BDRV_SECTOR_SIZE
);
1272 assert(is_power_of_2(align
));
1273 assert((offset
& (align
- 1)) == 0);
1274 assert((bytes
& (align
- 1)) == 0);
1275 assert(!qiov
|| bytes
== qiov
->size
);
1276 assert((bs
->open_flags
& BDRV_O_NO_IO
) == 0);
1277 assert(!(flags
& ~BDRV_REQ_MASK
));
1279 waited
= wait_serialising_requests(req
);
1280 assert(!waited
|| !req
->serialising
);
1281 assert(req
->overlap_offset
<= offset
);
1282 assert(offset
+ bytes
<= req
->overlap_offset
+ req
->overlap_bytes
);
1284 ret
= notifier_with_return_list_notify(&bs
->before_write_notifiers
, req
);
1286 if (!ret
&& bs
->detect_zeroes
!= BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF
&&
1287 !(flags
& BDRV_REQ_ZERO_WRITE
) && drv
->bdrv_co_pwrite_zeroes
&&
1288 qemu_iovec_is_zero(qiov
)) {
1289 flags
|= BDRV_REQ_ZERO_WRITE
;
1290 if (bs
->detect_zeroes
== BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP
) {
1291 flags
|= BDRV_REQ_MAY_UNMAP
;
1296 /* Do nothing, write notifier decided to fail this request */
1297 } else if (flags
& BDRV_REQ_ZERO_WRITE
) {
1298 bdrv_debug_event(bs
, BLKDBG_PWRITEV_ZERO
);
1299 ret
= bdrv_co_do_pwrite_zeroes(bs
, offset
, bytes
, flags
);
1301 bdrv_debug_event(bs
, BLKDBG_PWRITEV
);
1302 ret
= bdrv_driver_pwritev(bs
, offset
, bytes
, qiov
, flags
);
1304 bdrv_debug_event(bs
, BLKDBG_PWRITEV_DONE
);
1307 bdrv_set_dirty(bs
, start_sector
, end_sector
- start_sector
);
1309 if (bs
->wr_highest_offset
< offset
+ bytes
) {
1310 bs
->wr_highest_offset
= offset
+ bytes
;
1314 bs
->total_sectors
= MAX(bs
->total_sectors
, end_sector
);
1320 static int coroutine_fn
bdrv_co_do_zero_pwritev(BlockDriverState
*bs
,
1323 BdrvRequestFlags flags
,
1324 BdrvTrackedRequest
*req
)
1326 uint8_t *buf
= NULL
;
1327 QEMUIOVector local_qiov
;
1329 uint64_t align
= bs
->bl
.request_alignment
;
1330 unsigned int head_padding_bytes
, tail_padding_bytes
;
1333 head_padding_bytes
= offset
& (align
- 1);
1334 tail_padding_bytes
= align
- ((offset
+ bytes
) & (align
- 1));
1337 assert(flags
& BDRV_REQ_ZERO_WRITE
);
1338 if (head_padding_bytes
|| tail_padding_bytes
) {
1339 buf
= qemu_blockalign(bs
, align
);
1340 iov
= (struct iovec
) {
1344 qemu_iovec_init_external(&local_qiov
, &iov
, 1);
1346 if (head_padding_bytes
) {
1347 uint64_t zero_bytes
= MIN(bytes
, align
- head_padding_bytes
);
1349 /* RMW the unaligned part before head. */
1350 mark_request_serialising(req
, align
);
1351 wait_serialising_requests(req
);
1352 bdrv_debug_event(bs
, BLKDBG_PWRITEV_RMW_HEAD
);
1353 ret
= bdrv_aligned_preadv(bs
, req
, offset
& ~(align
- 1), align
,
1354 align
, &local_qiov
, 0);
1358 bdrv_debug_event(bs
, BLKDBG_PWRITEV_RMW_AFTER_HEAD
);
1360 memset(buf
+ head_padding_bytes
, 0, zero_bytes
);
1361 ret
= bdrv_aligned_pwritev(bs
, req
, offset
& ~(align
- 1), align
,
1363 flags
& ~BDRV_REQ_ZERO_WRITE
);
1367 offset
+= zero_bytes
;
1368 bytes
-= zero_bytes
;
1371 assert(!bytes
|| (offset
& (align
- 1)) == 0);
1372 if (bytes
>= align
) {
1373 /* Write the aligned part in the middle. */
1374 uint64_t aligned_bytes
= bytes
& ~(align
- 1);
1375 ret
= bdrv_aligned_pwritev(bs
, req
, offset
, aligned_bytes
, align
,
1380 bytes
-= aligned_bytes
;
1381 offset
+= aligned_bytes
;
1384 assert(!bytes
|| (offset
& (align
- 1)) == 0);
1386 assert(align
== tail_padding_bytes
+ bytes
);
1387 /* RMW the unaligned part after tail. */
1388 mark_request_serialising(req
, align
);
1389 wait_serialising_requests(req
);
1390 bdrv_debug_event(bs
, BLKDBG_PWRITEV_RMW_TAIL
);
1391 ret
= bdrv_aligned_preadv(bs
, req
, offset
, align
,
1392 align
, &local_qiov
, 0);
1396 bdrv_debug_event(bs
, BLKDBG_PWRITEV_RMW_AFTER_TAIL
);
1398 memset(buf
, 0, bytes
);
1399 ret
= bdrv_aligned_pwritev(bs
, req
, offset
, align
, align
,
1400 &local_qiov
, flags
& ~BDRV_REQ_ZERO_WRITE
);
1409 * Handle a write request in coroutine context
1411 int coroutine_fn
bdrv_co_pwritev(BdrvChild
*child
,
1412 int64_t offset
, unsigned int bytes
, QEMUIOVector
*qiov
,
1413 BdrvRequestFlags flags
)
1415 BlockDriverState
*bs
= child
->bs
;
1416 BdrvTrackedRequest req
;
1417 uint64_t align
= bs
->bl
.request_alignment
;
1418 uint8_t *head_buf
= NULL
;
1419 uint8_t *tail_buf
= NULL
;
1420 QEMUIOVector local_qiov
;
1421 bool use_local_qiov
= false;
1427 if (bs
->read_only
) {
1430 assert(!(bs
->open_flags
& BDRV_O_INACTIVE
));
1432 ret
= bdrv_check_byte_request(bs
, offset
, bytes
);
1438 * Align write if necessary by performing a read-modify-write cycle.
1439 * Pad qiov with the read parts and be sure to have a tracked request not
1440 * only for bdrv_aligned_pwritev, but also for the reads of the RMW cycle.
1442 tracked_request_begin(&req
, bs
, offset
, bytes
, BDRV_TRACKED_WRITE
);
1445 ret
= bdrv_co_do_zero_pwritev(bs
, offset
, bytes
, flags
, &req
);
1449 if (offset
& (align
- 1)) {
1450 QEMUIOVector head_qiov
;
1451 struct iovec head_iov
;
1453 mark_request_serialising(&req
, align
);
1454 wait_serialising_requests(&req
);
1456 head_buf
= qemu_blockalign(bs
, align
);
1457 head_iov
= (struct iovec
) {
1458 .iov_base
= head_buf
,
1461 qemu_iovec_init_external(&head_qiov
, &head_iov
, 1);
1463 bdrv_debug_event(bs
, BLKDBG_PWRITEV_RMW_HEAD
);
1464 ret
= bdrv_aligned_preadv(bs
, &req
, offset
& ~(align
- 1), align
,
1465 align
, &head_qiov
, 0);
1469 bdrv_debug_event(bs
, BLKDBG_PWRITEV_RMW_AFTER_HEAD
);
1471 qemu_iovec_init(&local_qiov
, qiov
->niov
+ 2);
1472 qemu_iovec_add(&local_qiov
, head_buf
, offset
& (align
- 1));
1473 qemu_iovec_concat(&local_qiov
, qiov
, 0, qiov
->size
);
1474 use_local_qiov
= true;
1476 bytes
+= offset
& (align
- 1);
1477 offset
= offset
& ~(align
- 1);
1479 /* We have read the tail already if the request is smaller
1480 * than one aligned block.
1482 if (bytes
< align
) {
1483 qemu_iovec_add(&local_qiov
, head_buf
+ bytes
, align
- bytes
);
1488 if ((offset
+ bytes
) & (align
- 1)) {
1489 QEMUIOVector tail_qiov
;
1490 struct iovec tail_iov
;
1494 mark_request_serialising(&req
, align
);
1495 waited
= wait_serialising_requests(&req
);
1496 assert(!waited
|| !use_local_qiov
);
1498 tail_buf
= qemu_blockalign(bs
, align
);
1499 tail_iov
= (struct iovec
) {
1500 .iov_base
= tail_buf
,
1503 qemu_iovec_init_external(&tail_qiov
, &tail_iov
, 1);
1505 bdrv_debug_event(bs
, BLKDBG_PWRITEV_RMW_TAIL
);
1506 ret
= bdrv_aligned_preadv(bs
, &req
, (offset
+ bytes
) & ~(align
- 1), align
,
1507 align
, &tail_qiov
, 0);
1511 bdrv_debug_event(bs
, BLKDBG_PWRITEV_RMW_AFTER_TAIL
);
1513 if (!use_local_qiov
) {
1514 qemu_iovec_init(&local_qiov
, qiov
->niov
+ 1);
1515 qemu_iovec_concat(&local_qiov
, qiov
, 0, qiov
->size
);
1516 use_local_qiov
= true;
1519 tail_bytes
= (offset
+ bytes
) & (align
- 1);
1520 qemu_iovec_add(&local_qiov
, tail_buf
+ tail_bytes
, align
- tail_bytes
);
1522 bytes
= ROUND_UP(bytes
, align
);
1525 ret
= bdrv_aligned_pwritev(bs
, &req
, offset
, bytes
, align
,
1526 use_local_qiov
? &local_qiov
: qiov
,
1531 if (use_local_qiov
) {
1532 qemu_iovec_destroy(&local_qiov
);
1534 qemu_vfree(head_buf
);
1535 qemu_vfree(tail_buf
);
1537 tracked_request_end(&req
);
1541 static int coroutine_fn
bdrv_co_do_writev(BdrvChild
*child
,
1542 int64_t sector_num
, int nb_sectors
, QEMUIOVector
*qiov
,
1543 BdrvRequestFlags flags
)
1545 if (nb_sectors
< 0 || nb_sectors
> BDRV_REQUEST_MAX_SECTORS
) {
1549 return bdrv_co_pwritev(child
, sector_num
<< BDRV_SECTOR_BITS
,
1550 nb_sectors
<< BDRV_SECTOR_BITS
, qiov
, flags
);
1553 int coroutine_fn
bdrv_co_writev(BdrvChild
*child
, int64_t sector_num
,
1554 int nb_sectors
, QEMUIOVector
*qiov
)
1556 trace_bdrv_co_writev(child
->bs
, sector_num
, nb_sectors
);
1558 return bdrv_co_do_writev(child
, sector_num
, nb_sectors
, qiov
, 0);
1561 int coroutine_fn
bdrv_co_pwrite_zeroes(BdrvChild
*child
, int64_t offset
,
1562 int count
, BdrvRequestFlags flags
)
1564 trace_bdrv_co_pwrite_zeroes(child
->bs
, offset
, count
, flags
);
1566 if (!(child
->bs
->open_flags
& BDRV_O_UNMAP
)) {
1567 flags
&= ~BDRV_REQ_MAY_UNMAP
;
1570 return bdrv_co_pwritev(child
, offset
, count
, NULL
,
1571 BDRV_REQ_ZERO_WRITE
| flags
);
1574 typedef struct BdrvCoGetBlockStatusData
{
1575 BlockDriverState
*bs
;
1576 BlockDriverState
*base
;
1577 BlockDriverState
**file
;
1583 } BdrvCoGetBlockStatusData
;
1586 * Returns the allocation status of the specified sectors.
1587 * Drivers not implementing the functionality are assumed to not support
1588 * backing files, hence all their sectors are reported as allocated.
1590 * If 'sector_num' is beyond the end of the disk image the return value is 0
1591 * and 'pnum' is set to 0.
1593 * 'pnum' is set to the number of sectors (including and immediately following
1594 * the specified sector) that are known to be in the same
1595 * allocated/unallocated state.
1597 * 'nb_sectors' is the max value 'pnum' should be set to. If nb_sectors goes
1598 * beyond the end of the disk image it will be clamped.
1600 * If returned value is positive and BDRV_BLOCK_OFFSET_VALID bit is set, 'file'
1601 * points to the BDS which the sector range is allocated in.
1603 static int64_t coroutine_fn
bdrv_co_get_block_status(BlockDriverState
*bs
,
1605 int nb_sectors
, int *pnum
,
1606 BlockDriverState
**file
)
1608 int64_t total_sectors
;
1612 total_sectors
= bdrv_nb_sectors(bs
);
1613 if (total_sectors
< 0) {
1614 return total_sectors
;
1617 if (sector_num
>= total_sectors
) {
1622 n
= total_sectors
- sector_num
;
1623 if (n
< nb_sectors
) {
1627 if (!bs
->drv
->bdrv_co_get_block_status
) {
1629 ret
= BDRV_BLOCK_DATA
| BDRV_BLOCK_ALLOCATED
;
1630 if (bs
->drv
->protocol_name
) {
1631 ret
|= BDRV_BLOCK_OFFSET_VALID
| (sector_num
* BDRV_SECTOR_SIZE
);
1637 ret
= bs
->drv
->bdrv_co_get_block_status(bs
, sector_num
, nb_sectors
, pnum
,
1644 if (ret
& BDRV_BLOCK_RAW
) {
1645 assert(ret
& BDRV_BLOCK_OFFSET_VALID
);
1646 return bdrv_get_block_status(bs
->file
->bs
, ret
>> BDRV_SECTOR_BITS
,
1650 if (ret
& (BDRV_BLOCK_DATA
| BDRV_BLOCK_ZERO
)) {
1651 ret
|= BDRV_BLOCK_ALLOCATED
;
1653 if (bdrv_unallocated_blocks_are_zero(bs
)) {
1654 ret
|= BDRV_BLOCK_ZERO
;
1655 } else if (bs
->backing
) {
1656 BlockDriverState
*bs2
= bs
->backing
->bs
;
1657 int64_t nb_sectors2
= bdrv_nb_sectors(bs2
);
1658 if (nb_sectors2
>= 0 && sector_num
>= nb_sectors2
) {
1659 ret
|= BDRV_BLOCK_ZERO
;
1664 if (*file
&& *file
!= bs
&&
1665 (ret
& BDRV_BLOCK_DATA
) && !(ret
& BDRV_BLOCK_ZERO
) &&
1666 (ret
& BDRV_BLOCK_OFFSET_VALID
)) {
1667 BlockDriverState
*file2
;
1670 ret2
= bdrv_co_get_block_status(*file
, ret
>> BDRV_SECTOR_BITS
,
1671 *pnum
, &file_pnum
, &file2
);
1673 /* Ignore errors. This is just providing extra information, it
1674 * is useful but not necessary.
1677 /* !file_pnum indicates an offset at or beyond the EOF; it is
1678 * perfectly valid for the format block driver to point to such
1679 * offsets, so catch it and mark everything as zero */
1680 ret
|= BDRV_BLOCK_ZERO
;
1682 /* Limit request to the range reported by the protocol driver */
1684 ret
|= (ret2
& BDRV_BLOCK_ZERO
);
1692 static int64_t coroutine_fn
bdrv_co_get_block_status_above(BlockDriverState
*bs
,
1693 BlockDriverState
*base
,
1697 BlockDriverState
**file
)
1699 BlockDriverState
*p
;
1703 for (p
= bs
; p
!= base
; p
= backing_bs(p
)) {
1704 ret
= bdrv_co_get_block_status(p
, sector_num
, nb_sectors
, pnum
, file
);
1705 if (ret
< 0 || ret
& BDRV_BLOCK_ALLOCATED
) {
1708 /* [sector_num, pnum] unallocated on this layer, which could be only
1709 * the first part of [sector_num, nb_sectors]. */
1710 nb_sectors
= MIN(nb_sectors
, *pnum
);
1715 /* Coroutine wrapper for bdrv_get_block_status_above() */
1716 static void coroutine_fn
bdrv_get_block_status_above_co_entry(void *opaque
)
1718 BdrvCoGetBlockStatusData
*data
= opaque
;
1720 data
->ret
= bdrv_co_get_block_status_above(data
->bs
, data
->base
,
1729 * Synchronous wrapper around bdrv_co_get_block_status_above().
1731 * See bdrv_co_get_block_status_above() for details.
1733 int64_t bdrv_get_block_status_above(BlockDriverState
*bs
,
1734 BlockDriverState
*base
,
1736 int nb_sectors
, int *pnum
,
1737 BlockDriverState
**file
)
1740 BdrvCoGetBlockStatusData data
= {
1744 .sector_num
= sector_num
,
1745 .nb_sectors
= nb_sectors
,
1750 if (qemu_in_coroutine()) {
1751 /* Fast-path if already in coroutine context */
1752 bdrv_get_block_status_above_co_entry(&data
);
1754 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
1756 co
= qemu_coroutine_create(bdrv_get_block_status_above_co_entry
,
1758 qemu_coroutine_enter(co
);
1759 while (!data
.done
) {
1760 aio_poll(aio_context
, true);
1766 int64_t bdrv_get_block_status(BlockDriverState
*bs
,
1768 int nb_sectors
, int *pnum
,
1769 BlockDriverState
**file
)
1771 return bdrv_get_block_status_above(bs
, backing_bs(bs
),
1772 sector_num
, nb_sectors
, pnum
, file
);
1775 int coroutine_fn
bdrv_is_allocated(BlockDriverState
*bs
, int64_t sector_num
,
1776 int nb_sectors
, int *pnum
)
1778 BlockDriverState
*file
;
1779 int64_t ret
= bdrv_get_block_status(bs
, sector_num
, nb_sectors
, pnum
,
1784 return !!(ret
& BDRV_BLOCK_ALLOCATED
);
1788 * Given an image chain: ... -> [BASE] -> [INTER1] -> [INTER2] -> [TOP]
1790 * Return true if the given sector is allocated in any image between
1791 * BASE and TOP (inclusive). BASE can be NULL to check if the given
1792 * sector is allocated in any image of the chain. Return false otherwise.
1794 * 'pnum' is set to the number of sectors (including and immediately following
1795 * the specified sector) that are known to be in the same
1796 * allocated/unallocated state.
1799 int bdrv_is_allocated_above(BlockDriverState
*top
,
1800 BlockDriverState
*base
,
1802 int nb_sectors
, int *pnum
)
1804 BlockDriverState
*intermediate
;
1805 int ret
, n
= nb_sectors
;
1808 while (intermediate
&& intermediate
!= base
) {
1810 ret
= bdrv_is_allocated(intermediate
, sector_num
, nb_sectors
,
1820 * [sector_num, nb_sectors] is unallocated on top but intermediate
1823 * [sector_num+x, nr_sectors] allocated.
1825 if (n
> pnum_inter
&&
1826 (intermediate
== top
||
1827 sector_num
+ pnum_inter
< intermediate
->total_sectors
)) {
1831 intermediate
= backing_bs(intermediate
);
1838 int bdrv_write_compressed(BlockDriverState
*bs
, int64_t sector_num
,
1839 const uint8_t *buf
, int nb_sectors
)
1841 BlockDriver
*drv
= bs
->drv
;
1847 if (!drv
->bdrv_write_compressed
) {
1850 ret
= bdrv_check_request(bs
, sector_num
, nb_sectors
);
1855 assert(QLIST_EMPTY(&bs
->dirty_bitmaps
));
1857 return drv
->bdrv_write_compressed(bs
, sector_num
, buf
, nb_sectors
);
1860 typedef struct BdrvVmstateCo
{
1861 BlockDriverState
*bs
;
1868 static int coroutine_fn
1869 bdrv_co_rw_vmstate(BlockDriverState
*bs
, QEMUIOVector
*qiov
, int64_t pos
,
1872 BlockDriver
*drv
= bs
->drv
;
1876 } else if (drv
->bdrv_load_vmstate
) {
1877 return is_read
? drv
->bdrv_load_vmstate(bs
, qiov
, pos
)
1878 : drv
->bdrv_save_vmstate(bs
, qiov
, pos
);
1879 } else if (bs
->file
) {
1880 return bdrv_co_rw_vmstate(bs
->file
->bs
, qiov
, pos
, is_read
);
1886 static void coroutine_fn
bdrv_co_rw_vmstate_entry(void *opaque
)
1888 BdrvVmstateCo
*co
= opaque
;
1889 co
->ret
= bdrv_co_rw_vmstate(co
->bs
, co
->qiov
, co
->pos
, co
->is_read
);
1893 bdrv_rw_vmstate(BlockDriverState
*bs
, QEMUIOVector
*qiov
, int64_t pos
,
1896 if (qemu_in_coroutine()) {
1897 return bdrv_co_rw_vmstate(bs
, qiov
, pos
, is_read
);
1899 BdrvVmstateCo data
= {
1904 .ret
= -EINPROGRESS
,
1906 Coroutine
*co
= qemu_coroutine_create(bdrv_co_rw_vmstate_entry
, &data
);
1908 qemu_coroutine_enter(co
);
1909 while (data
.ret
== -EINPROGRESS
) {
1910 aio_poll(bdrv_get_aio_context(bs
), true);
1916 int bdrv_save_vmstate(BlockDriverState
*bs
, const uint8_t *buf
,
1917 int64_t pos
, int size
)
1920 struct iovec iov
= {
1921 .iov_base
= (void *) buf
,
1926 qemu_iovec_init_external(&qiov
, &iov
, 1);
1928 ret
= bdrv_writev_vmstate(bs
, &qiov
, pos
);
1936 int bdrv_writev_vmstate(BlockDriverState
*bs
, QEMUIOVector
*qiov
, int64_t pos
)
1938 return bdrv_rw_vmstate(bs
, qiov
, pos
, false);
1941 int bdrv_load_vmstate(BlockDriverState
*bs
, uint8_t *buf
,
1942 int64_t pos
, int size
)
1945 struct iovec iov
= {
1951 qemu_iovec_init_external(&qiov
, &iov
, 1);
1952 ret
= bdrv_readv_vmstate(bs
, &qiov
, pos
);
1960 int bdrv_readv_vmstate(BlockDriverState
*bs
, QEMUIOVector
*qiov
, int64_t pos
)
1962 return bdrv_rw_vmstate(bs
, qiov
, pos
, true);
1965 /**************************************************************/
1968 BlockAIOCB
*bdrv_aio_readv(BdrvChild
*child
, int64_t sector_num
,
1969 QEMUIOVector
*qiov
, int nb_sectors
,
1970 BlockCompletionFunc
*cb
, void *opaque
)
1972 trace_bdrv_aio_readv(child
->bs
, sector_num
, nb_sectors
, opaque
);
1974 return bdrv_co_aio_rw_vector(child
, sector_num
, qiov
, nb_sectors
, 0,
1978 BlockAIOCB
*bdrv_aio_writev(BdrvChild
*child
, int64_t sector_num
,
1979 QEMUIOVector
*qiov
, int nb_sectors
,
1980 BlockCompletionFunc
*cb
, void *opaque
)
1982 trace_bdrv_aio_writev(child
->bs
, sector_num
, nb_sectors
, opaque
);
1984 return bdrv_co_aio_rw_vector(child
, sector_num
, qiov
, nb_sectors
, 0,
1988 void bdrv_aio_cancel(BlockAIOCB
*acb
)
1991 bdrv_aio_cancel_async(acb
);
1992 while (acb
->refcnt
> 1) {
1993 if (acb
->aiocb_info
->get_aio_context
) {
1994 aio_poll(acb
->aiocb_info
->get_aio_context(acb
), true);
1995 } else if (acb
->bs
) {
1996 aio_poll(bdrv_get_aio_context(acb
->bs
), true);
2001 qemu_aio_unref(acb
);
2004 /* Async version of aio cancel. The caller is not blocked if the acb implements
2005 * cancel_async, otherwise we do nothing and let the request normally complete.
2006 * In either case the completion callback must be called. */
2007 void bdrv_aio_cancel_async(BlockAIOCB
*acb
)
2009 if (acb
->aiocb_info
->cancel_async
) {
2010 acb
->aiocb_info
->cancel_async(acb
);
2014 /**************************************************************/
2015 /* async block device emulation */
2017 typedef struct BlockRequest
{
2019 /* Used during read, write, trim */
2026 /* Used during ioctl */
2032 BlockCompletionFunc
*cb
;
2038 typedef struct BlockAIOCBCoroutine
{
2046 } BlockAIOCBCoroutine
;
2048 static const AIOCBInfo bdrv_em_co_aiocb_info
= {
2049 .aiocb_size
= sizeof(BlockAIOCBCoroutine
),
2052 static void bdrv_co_complete(BlockAIOCBCoroutine
*acb
)
2054 if (!acb
->need_bh
) {
2055 acb
->common
.cb(acb
->common
.opaque
, acb
->req
.error
);
2056 qemu_aio_unref(acb
);
2060 static void bdrv_co_em_bh(void *opaque
)
2062 BlockAIOCBCoroutine
*acb
= opaque
;
2064 assert(!acb
->need_bh
);
2065 qemu_bh_delete(acb
->bh
);
2066 bdrv_co_complete(acb
);
2069 static void bdrv_co_maybe_schedule_bh(BlockAIOCBCoroutine
*acb
)
2071 acb
->need_bh
= false;
2072 if (acb
->req
.error
!= -EINPROGRESS
) {
2073 BlockDriverState
*bs
= acb
->common
.bs
;
2075 acb
->bh
= aio_bh_new(bdrv_get_aio_context(bs
), bdrv_co_em_bh
, acb
);
2076 qemu_bh_schedule(acb
->bh
);
2080 /* Invoke bdrv_co_do_readv/bdrv_co_do_writev */
2081 static void coroutine_fn
bdrv_co_do_rw(void *opaque
)
2083 BlockAIOCBCoroutine
*acb
= opaque
;
2085 if (!acb
->is_write
) {
2086 acb
->req
.error
= bdrv_co_do_readv(acb
->child
, acb
->req
.sector
,
2087 acb
->req
.nb_sectors
, acb
->req
.qiov
, acb
->req
.flags
);
2089 acb
->req
.error
= bdrv_co_do_writev(acb
->child
, acb
->req
.sector
,
2090 acb
->req
.nb_sectors
, acb
->req
.qiov
, acb
->req
.flags
);
2093 bdrv_co_complete(acb
);
2096 static BlockAIOCB
*bdrv_co_aio_rw_vector(BdrvChild
*child
,
2100 BdrvRequestFlags flags
,
2101 BlockCompletionFunc
*cb
,
2106 BlockAIOCBCoroutine
*acb
;
2108 acb
= qemu_aio_get(&bdrv_em_co_aiocb_info
, child
->bs
, cb
, opaque
);
2110 acb
->need_bh
= true;
2111 acb
->req
.error
= -EINPROGRESS
;
2112 acb
->req
.sector
= sector_num
;
2113 acb
->req
.nb_sectors
= nb_sectors
;
2114 acb
->req
.qiov
= qiov
;
2115 acb
->req
.flags
= flags
;
2116 acb
->is_write
= is_write
;
2118 co
= qemu_coroutine_create(bdrv_co_do_rw
, acb
);
2119 qemu_coroutine_enter(co
);
2121 bdrv_co_maybe_schedule_bh(acb
);
2122 return &acb
->common
;
2125 static void coroutine_fn
bdrv_aio_flush_co_entry(void *opaque
)
2127 BlockAIOCBCoroutine
*acb
= opaque
;
2128 BlockDriverState
*bs
= acb
->common
.bs
;
2130 acb
->req
.error
= bdrv_co_flush(bs
);
2131 bdrv_co_complete(acb
);
2134 BlockAIOCB
*bdrv_aio_flush(BlockDriverState
*bs
,
2135 BlockCompletionFunc
*cb
, void *opaque
)
2137 trace_bdrv_aio_flush(bs
, opaque
);
2140 BlockAIOCBCoroutine
*acb
;
2142 acb
= qemu_aio_get(&bdrv_em_co_aiocb_info
, bs
, cb
, opaque
);
2143 acb
->need_bh
= true;
2144 acb
->req
.error
= -EINPROGRESS
;
2146 co
= qemu_coroutine_create(bdrv_aio_flush_co_entry
, acb
);
2147 qemu_coroutine_enter(co
);
2149 bdrv_co_maybe_schedule_bh(acb
);
2150 return &acb
->common
;
2153 static void coroutine_fn
bdrv_aio_discard_co_entry(void *opaque
)
2155 BlockAIOCBCoroutine
*acb
= opaque
;
2156 BlockDriverState
*bs
= acb
->common
.bs
;
2158 acb
->req
.error
= bdrv_co_discard(bs
, acb
->req
.sector
, acb
->req
.nb_sectors
);
2159 bdrv_co_complete(acb
);
2162 BlockAIOCB
*bdrv_aio_discard(BlockDriverState
*bs
,
2163 int64_t sector_num
, int nb_sectors
,
2164 BlockCompletionFunc
*cb
, void *opaque
)
2167 BlockAIOCBCoroutine
*acb
;
2169 trace_bdrv_aio_discard(bs
, sector_num
, nb_sectors
, opaque
);
2171 acb
= qemu_aio_get(&bdrv_em_co_aiocb_info
, bs
, cb
, opaque
);
2172 acb
->need_bh
= true;
2173 acb
->req
.error
= -EINPROGRESS
;
2174 acb
->req
.sector
= sector_num
;
2175 acb
->req
.nb_sectors
= nb_sectors
;
2176 co
= qemu_coroutine_create(bdrv_aio_discard_co_entry
, acb
);
2177 qemu_coroutine_enter(co
);
2179 bdrv_co_maybe_schedule_bh(acb
);
2180 return &acb
->common
;
2183 void *qemu_aio_get(const AIOCBInfo
*aiocb_info
, BlockDriverState
*bs
,
2184 BlockCompletionFunc
*cb
, void *opaque
)
2188 acb
= g_malloc(aiocb_info
->aiocb_size
);
2189 acb
->aiocb_info
= aiocb_info
;
2192 acb
->opaque
= opaque
;
2197 void qemu_aio_ref(void *p
)
2199 BlockAIOCB
*acb
= p
;
2203 void qemu_aio_unref(void *p
)
2205 BlockAIOCB
*acb
= p
;
2206 assert(acb
->refcnt
> 0);
2207 if (--acb
->refcnt
== 0) {
2212 /**************************************************************/
2213 /* Coroutine block device emulation */
2215 typedef struct FlushCo
{
2216 BlockDriverState
*bs
;
2221 static void coroutine_fn
bdrv_flush_co_entry(void *opaque
)
2223 FlushCo
*rwco
= opaque
;
2225 rwco
->ret
= bdrv_co_flush(rwco
->bs
);
2228 int coroutine_fn
bdrv_co_flush(BlockDriverState
*bs
)
2231 BdrvTrackedRequest req
;
2233 if (!bs
|| !bdrv_is_inserted(bs
) || bdrv_is_read_only(bs
) ||
2238 tracked_request_begin(&req
, bs
, 0, 0, BDRV_TRACKED_FLUSH
);
2240 int current_gen
= bs
->write_gen
;
2242 /* Wait until any previous flushes are completed */
2243 while (bs
->flush_started_gen
!= bs
->flushed_gen
) {
2244 qemu_co_queue_wait(&bs
->flush_queue
);
2247 bs
->flush_started_gen
= current_gen
;
2249 /* Write back all layers by calling one driver function */
2250 if (bs
->drv
->bdrv_co_flush
) {
2251 ret
= bs
->drv
->bdrv_co_flush(bs
);
2255 /* Write back cached data to the OS even with cache=unsafe */
2256 BLKDBG_EVENT(bs
->file
, BLKDBG_FLUSH_TO_OS
);
2257 if (bs
->drv
->bdrv_co_flush_to_os
) {
2258 ret
= bs
->drv
->bdrv_co_flush_to_os(bs
);
2264 /* But don't actually force it to the disk with cache=unsafe */
2265 if (bs
->open_flags
& BDRV_O_NO_FLUSH
) {
2269 /* Check if we really need to flush anything */
2270 if (bs
->flushed_gen
== current_gen
) {
2274 BLKDBG_EVENT(bs
->file
, BLKDBG_FLUSH_TO_DISK
);
2275 if (bs
->drv
->bdrv_co_flush_to_disk
) {
2276 ret
= bs
->drv
->bdrv_co_flush_to_disk(bs
);
2277 } else if (bs
->drv
->bdrv_aio_flush
) {
2279 CoroutineIOCompletion co
= {
2280 .coroutine
= qemu_coroutine_self(),
2283 acb
= bs
->drv
->bdrv_aio_flush(bs
, bdrv_co_io_em_complete
, &co
);
2287 qemu_coroutine_yield();
2292 * Some block drivers always operate in either writethrough or unsafe
2293 * mode and don't support bdrv_flush therefore. Usually qemu doesn't
2294 * know how the server works (because the behaviour is hardcoded or
2295 * depends on server-side configuration), so we can't ensure that
2296 * everything is safe on disk. Returning an error doesn't work because
2297 * that would break guests even if the server operates in writethrough
2300 * Let's hope the user knows what he's doing.
2309 /* Now flush the underlying protocol. It will also have BDRV_O_NO_FLUSH
2310 * in the case of cache=unsafe, so there are no useless flushes.
2313 ret
= bs
->file
? bdrv_co_flush(bs
->file
->bs
) : 0;
2315 /* Notify any pending flushes that we have completed */
2316 bs
->flushed_gen
= current_gen
;
2317 qemu_co_queue_restart_all(&bs
->flush_queue
);
2319 tracked_request_end(&req
);
2323 int bdrv_flush(BlockDriverState
*bs
)
2326 FlushCo flush_co
= {
2331 if (qemu_in_coroutine()) {
2332 /* Fast-path if already in coroutine context */
2333 bdrv_flush_co_entry(&flush_co
);
2335 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
2337 co
= qemu_coroutine_create(bdrv_flush_co_entry
, &flush_co
);
2338 qemu_coroutine_enter(co
);
2339 while (flush_co
.ret
== NOT_DONE
) {
2340 aio_poll(aio_context
, true);
2344 return flush_co
.ret
;
2347 typedef struct DiscardCo
{
2348 BlockDriverState
*bs
;
2353 static void coroutine_fn
bdrv_discard_co_entry(void *opaque
)
2355 DiscardCo
*rwco
= opaque
;
2357 rwco
->ret
= bdrv_co_discard(rwco
->bs
, rwco
->sector_num
, rwco
->nb_sectors
);
2360 int coroutine_fn
bdrv_co_discard(BlockDriverState
*bs
, int64_t sector_num
,
2363 BdrvTrackedRequest req
;
2364 int max_discard
, ret
;
2370 ret
= bdrv_check_request(bs
, sector_num
, nb_sectors
);
2373 } else if (bs
->read_only
) {
2376 assert(!(bs
->open_flags
& BDRV_O_INACTIVE
));
2378 /* Do nothing if disabled. */
2379 if (!(bs
->open_flags
& BDRV_O_UNMAP
)) {
2383 if (!bs
->drv
->bdrv_co_discard
&& !bs
->drv
->bdrv_aio_discard
) {
2387 tracked_request_begin(&req
, bs
, sector_num
<< BDRV_SECTOR_BITS
,
2388 nb_sectors
<< BDRV_SECTOR_BITS
, BDRV_TRACKED_DISCARD
);
2390 ret
= notifier_with_return_list_notify(&bs
->before_write_notifiers
, &req
);
2395 max_discard
= MIN_NON_ZERO(bs
->bl
.max_pdiscard
>> BDRV_SECTOR_BITS
,
2396 BDRV_REQUEST_MAX_SECTORS
);
2397 while (nb_sectors
> 0) {
2399 int num
= nb_sectors
;
2400 int discard_alignment
= bs
->bl
.pdiscard_alignment
>> BDRV_SECTOR_BITS
;
2403 if (discard_alignment
&&
2404 num
>= discard_alignment
&&
2405 sector_num
% discard_alignment
) {
2406 if (num
> discard_alignment
) {
2407 num
= discard_alignment
;
2409 num
-= sector_num
% discard_alignment
;
2412 /* limit request size */
2413 if (num
> max_discard
) {
2417 if (bs
->drv
->bdrv_co_discard
) {
2418 ret
= bs
->drv
->bdrv_co_discard(bs
, sector_num
, num
);
2421 CoroutineIOCompletion co
= {
2422 .coroutine
= qemu_coroutine_self(),
2425 acb
= bs
->drv
->bdrv_aio_discard(bs
, sector_num
, nb_sectors
,
2426 bdrv_co_io_em_complete
, &co
);
2431 qemu_coroutine_yield();
2435 if (ret
&& ret
!= -ENOTSUP
) {
2445 bdrv_set_dirty(bs
, req
.offset
>> BDRV_SECTOR_BITS
,
2446 req
.bytes
>> BDRV_SECTOR_BITS
);
2447 tracked_request_end(&req
);
2451 int bdrv_discard(BlockDriverState
*bs
, int64_t sector_num
, int nb_sectors
)
2456 .sector_num
= sector_num
,
2457 .nb_sectors
= nb_sectors
,
2461 if (qemu_in_coroutine()) {
2462 /* Fast-path if already in coroutine context */
2463 bdrv_discard_co_entry(&rwco
);
2465 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
2467 co
= qemu_coroutine_create(bdrv_discard_co_entry
, &rwco
);
2468 qemu_coroutine_enter(co
);
2469 while (rwco
.ret
== NOT_DONE
) {
2470 aio_poll(aio_context
, true);
2477 static int bdrv_co_do_ioctl(BlockDriverState
*bs
, int req
, void *buf
)
2479 BlockDriver
*drv
= bs
->drv
;
2480 BdrvTrackedRequest tracked_req
;
2481 CoroutineIOCompletion co
= {
2482 .coroutine
= qemu_coroutine_self(),
2486 tracked_request_begin(&tracked_req
, bs
, 0, 0, BDRV_TRACKED_IOCTL
);
2487 if (!drv
|| !drv
->bdrv_aio_ioctl
) {
2492 acb
= drv
->bdrv_aio_ioctl(bs
, req
, buf
, bdrv_co_io_em_complete
, &co
);
2497 qemu_coroutine_yield();
2499 tracked_request_end(&tracked_req
);
2504 BlockDriverState
*bs
;
2510 static void coroutine_fn
bdrv_co_ioctl_entry(void *opaque
)
2512 BdrvIoctlCoData
*data
= opaque
;
2513 data
->ret
= bdrv_co_do_ioctl(data
->bs
, data
->req
, data
->buf
);
2516 /* needed for generic scsi interface */
2517 int bdrv_ioctl(BlockDriverState
*bs
, unsigned long int req
, void *buf
)
2519 BdrvIoctlCoData data
= {
2523 .ret
= -EINPROGRESS
,
2526 if (qemu_in_coroutine()) {
2527 /* Fast-path if already in coroutine context */
2528 bdrv_co_ioctl_entry(&data
);
2530 Coroutine
*co
= qemu_coroutine_create(bdrv_co_ioctl_entry
, &data
);
2532 qemu_coroutine_enter(co
);
2533 while (data
.ret
== -EINPROGRESS
) {
2534 aio_poll(bdrv_get_aio_context(bs
), true);
2540 static void coroutine_fn
bdrv_co_aio_ioctl_entry(void *opaque
)
2542 BlockAIOCBCoroutine
*acb
= opaque
;
2543 acb
->req
.error
= bdrv_co_do_ioctl(acb
->common
.bs
,
2544 acb
->req
.req
, acb
->req
.buf
);
2545 bdrv_co_complete(acb
);
2548 BlockAIOCB
*bdrv_aio_ioctl(BlockDriverState
*bs
,
2549 unsigned long int req
, void *buf
,
2550 BlockCompletionFunc
*cb
, void *opaque
)
2552 BlockAIOCBCoroutine
*acb
= qemu_aio_get(&bdrv_em_co_aiocb_info
,
2556 acb
->need_bh
= true;
2557 acb
->req
.error
= -EINPROGRESS
;
2560 co
= qemu_coroutine_create(bdrv_co_aio_ioctl_entry
, acb
);
2561 qemu_coroutine_enter(co
);
2563 bdrv_co_maybe_schedule_bh(acb
);
2564 return &acb
->common
;
2567 void *qemu_blockalign(BlockDriverState
*bs
, size_t size
)
2569 return qemu_memalign(bdrv_opt_mem_align(bs
), size
);
2572 void *qemu_blockalign0(BlockDriverState
*bs
, size_t size
)
2574 return memset(qemu_blockalign(bs
, size
), 0, size
);
2577 void *qemu_try_blockalign(BlockDriverState
*bs
, size_t size
)
2579 size_t align
= bdrv_opt_mem_align(bs
);
2581 /* Ensure that NULL is never returned on success */
2587 return qemu_try_memalign(align
, size
);
2590 void *qemu_try_blockalign0(BlockDriverState
*bs
, size_t size
)
2592 void *mem
= qemu_try_blockalign(bs
, size
);
2595 memset(mem
, 0, size
);
2602 * Check if all memory in this vector is sector aligned.
2604 bool bdrv_qiov_is_aligned(BlockDriverState
*bs
, QEMUIOVector
*qiov
)
2607 size_t alignment
= bdrv_min_mem_align(bs
);
2609 for (i
= 0; i
< qiov
->niov
; i
++) {
2610 if ((uintptr_t) qiov
->iov
[i
].iov_base
% alignment
) {
2613 if (qiov
->iov
[i
].iov_len
% alignment
) {
2621 void bdrv_add_before_write_notifier(BlockDriverState
*bs
,
2622 NotifierWithReturn
*notifier
)
2624 notifier_with_return_list_add(&bs
->before_write_notifiers
, notifier
);
2627 void bdrv_io_plug(BlockDriverState
*bs
)
2631 QLIST_FOREACH(child
, &bs
->children
, next
) {
2632 bdrv_io_plug(child
->bs
);
2635 if (bs
->io_plugged
++ == 0 && bs
->io_plug_disabled
== 0) {
2636 BlockDriver
*drv
= bs
->drv
;
2637 if (drv
&& drv
->bdrv_io_plug
) {
2638 drv
->bdrv_io_plug(bs
);
2643 void bdrv_io_unplug(BlockDriverState
*bs
)
2647 assert(bs
->io_plugged
);
2648 if (--bs
->io_plugged
== 0 && bs
->io_plug_disabled
== 0) {
2649 BlockDriver
*drv
= bs
->drv
;
2650 if (drv
&& drv
->bdrv_io_unplug
) {
2651 drv
->bdrv_io_unplug(bs
);
2655 QLIST_FOREACH(child
, &bs
->children
, next
) {
2656 bdrv_io_unplug(child
->bs
);
2660 void bdrv_io_unplugged_begin(BlockDriverState
*bs
)
2664 if (bs
->io_plug_disabled
++ == 0 && bs
->io_plugged
> 0) {
2665 BlockDriver
*drv
= bs
->drv
;
2666 if (drv
&& drv
->bdrv_io_unplug
) {
2667 drv
->bdrv_io_unplug(bs
);
2671 QLIST_FOREACH(child
, &bs
->children
, next
) {
2672 bdrv_io_unplugged_begin(child
->bs
);
2676 void bdrv_io_unplugged_end(BlockDriverState
*bs
)
2680 assert(bs
->io_plug_disabled
);
2681 QLIST_FOREACH(child
, &bs
->children
, next
) {
2682 bdrv_io_unplugged_end(child
->bs
);
2685 if (--bs
->io_plug_disabled
== 0 && bs
->io_plugged
> 0) {
2686 BlockDriver
*drv
= bs
->drv
;
2687 if (drv
&& drv
->bdrv_io_plug
) {
2688 drv
->bdrv_io_plug(bs
);