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, zeroing after EOF, and fragmentation of large
975 * reads; any other features must be 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
;
983 uint64_t bytes_remaining
= bytes
;
986 assert(is_power_of_2(align
));
987 assert((offset
& (align
- 1)) == 0);
988 assert((bytes
& (align
- 1)) == 0);
989 assert(!qiov
|| bytes
== qiov
->size
);
990 assert((bs
->open_flags
& BDRV_O_NO_IO
) == 0);
991 max_transfer
= QEMU_ALIGN_DOWN(MIN_NON_ZERO(bs
->bl
.max_transfer
, INT_MAX
),
994 /* TODO: We would need a per-BDS .supported_read_flags and
995 * potential fallback support, if we ever implement any read flags
996 * to pass through to drivers. For now, there aren't any
997 * passthrough flags. */
998 assert(!(flags
& ~(BDRV_REQ_NO_SERIALISING
| BDRV_REQ_COPY_ON_READ
)));
1000 /* Handle Copy on Read and associated serialisation */
1001 if (flags
& BDRV_REQ_COPY_ON_READ
) {
1002 /* If we touch the same cluster it counts as an overlap. This
1003 * guarantees that allocating writes will be serialized and not race
1004 * with each other for the same cluster. For example, in copy-on-read
1005 * it ensures that the CoR read and write operations are atomic and
1006 * guest writes cannot interleave between them. */
1007 mark_request_serialising(req
, bdrv_get_cluster_size(bs
));
1010 if (!(flags
& BDRV_REQ_NO_SERIALISING
)) {
1011 wait_serialising_requests(req
);
1014 if (flags
& BDRV_REQ_COPY_ON_READ
) {
1015 int64_t start_sector
= offset
>> BDRV_SECTOR_BITS
;
1016 int64_t end_sector
= DIV_ROUND_UP(offset
+ bytes
, BDRV_SECTOR_SIZE
);
1017 unsigned int nb_sectors
= end_sector
- start_sector
;
1020 ret
= bdrv_is_allocated(bs
, start_sector
, nb_sectors
, &pnum
);
1025 if (!ret
|| pnum
!= nb_sectors
) {
1026 ret
= bdrv_co_do_copy_on_readv(bs
, offset
, bytes
, qiov
);
1031 /* Forward the request to the BlockDriver, possibly fragmenting it */
1032 total_bytes
= bdrv_getlength(bs
);
1033 if (total_bytes
< 0) {
1038 max_bytes
= ROUND_UP(MAX(0, total_bytes
- offset
), align
);
1039 if (bytes
<= max_bytes
&& bytes
<= max_transfer
) {
1040 ret
= bdrv_driver_preadv(bs
, offset
, bytes
, qiov
, 0);
1044 while (bytes_remaining
) {
1048 QEMUIOVector local_qiov
;
1050 num
= MIN(bytes_remaining
, MIN(max_bytes
, max_transfer
));
1052 qemu_iovec_init(&local_qiov
, qiov
->niov
);
1053 qemu_iovec_concat(&local_qiov
, qiov
, bytes
- bytes_remaining
, num
);
1055 ret
= bdrv_driver_preadv(bs
, offset
+ bytes
- bytes_remaining
,
1056 num
, &local_qiov
, 0);
1058 qemu_iovec_destroy(&local_qiov
);
1060 num
= bytes_remaining
;
1061 ret
= qemu_iovec_memset(qiov
, bytes
- bytes_remaining
, 0,
1067 bytes_remaining
-= num
;
1071 return ret
< 0 ? ret
: 0;
1075 * Handle a read request in coroutine context
1077 int coroutine_fn
bdrv_co_preadv(BdrvChild
*child
,
1078 int64_t offset
, unsigned int bytes
, QEMUIOVector
*qiov
,
1079 BdrvRequestFlags flags
)
1081 BlockDriverState
*bs
= child
->bs
;
1082 BlockDriver
*drv
= bs
->drv
;
1083 BdrvTrackedRequest req
;
1085 uint64_t align
= bs
->bl
.request_alignment
;
1086 uint8_t *head_buf
= NULL
;
1087 uint8_t *tail_buf
= NULL
;
1088 QEMUIOVector local_qiov
;
1089 bool use_local_qiov
= false;
1096 ret
= bdrv_check_byte_request(bs
, offset
, bytes
);
1101 /* Don't do copy-on-read if we read data before write operation */
1102 if (bs
->copy_on_read
&& !(flags
& BDRV_REQ_NO_SERIALISING
)) {
1103 flags
|= BDRV_REQ_COPY_ON_READ
;
1106 /* Align read if necessary by padding qiov */
1107 if (offset
& (align
- 1)) {
1108 head_buf
= qemu_blockalign(bs
, align
);
1109 qemu_iovec_init(&local_qiov
, qiov
->niov
+ 2);
1110 qemu_iovec_add(&local_qiov
, head_buf
, offset
& (align
- 1));
1111 qemu_iovec_concat(&local_qiov
, qiov
, 0, qiov
->size
);
1112 use_local_qiov
= true;
1114 bytes
+= offset
& (align
- 1);
1115 offset
= offset
& ~(align
- 1);
1118 if ((offset
+ bytes
) & (align
- 1)) {
1119 if (!use_local_qiov
) {
1120 qemu_iovec_init(&local_qiov
, qiov
->niov
+ 1);
1121 qemu_iovec_concat(&local_qiov
, qiov
, 0, qiov
->size
);
1122 use_local_qiov
= true;
1124 tail_buf
= qemu_blockalign(bs
, align
);
1125 qemu_iovec_add(&local_qiov
, tail_buf
,
1126 align
- ((offset
+ bytes
) & (align
- 1)));
1128 bytes
= ROUND_UP(bytes
, align
);
1131 tracked_request_begin(&req
, bs
, offset
, bytes
, BDRV_TRACKED_READ
);
1132 ret
= bdrv_aligned_preadv(bs
, &req
, offset
, bytes
, align
,
1133 use_local_qiov
? &local_qiov
: qiov
,
1135 tracked_request_end(&req
);
1137 if (use_local_qiov
) {
1138 qemu_iovec_destroy(&local_qiov
);
1139 qemu_vfree(head_buf
);
1140 qemu_vfree(tail_buf
);
1146 static int coroutine_fn
bdrv_co_do_readv(BdrvChild
*child
,
1147 int64_t sector_num
, int nb_sectors
, QEMUIOVector
*qiov
,
1148 BdrvRequestFlags flags
)
1150 if (nb_sectors
< 0 || nb_sectors
> BDRV_REQUEST_MAX_SECTORS
) {
1154 return bdrv_co_preadv(child
, sector_num
<< BDRV_SECTOR_BITS
,
1155 nb_sectors
<< BDRV_SECTOR_BITS
, qiov
, flags
);
1158 int coroutine_fn
bdrv_co_readv(BdrvChild
*child
, int64_t sector_num
,
1159 int nb_sectors
, QEMUIOVector
*qiov
)
1161 trace_bdrv_co_readv(child
->bs
, sector_num
, nb_sectors
);
1163 return bdrv_co_do_readv(child
, sector_num
, nb_sectors
, qiov
, 0);
1166 /* Maximum buffer for write zeroes fallback, in bytes */
1167 #define MAX_WRITE_ZEROES_BOUNCE_BUFFER (32768 << BDRV_SECTOR_BITS)
1169 static int coroutine_fn
bdrv_co_do_pwrite_zeroes(BlockDriverState
*bs
,
1170 int64_t offset
, int count
, BdrvRequestFlags flags
)
1172 BlockDriver
*drv
= bs
->drv
;
1174 struct iovec iov
= {0};
1176 bool need_flush
= false;
1180 int max_write_zeroes
= MIN_NON_ZERO(bs
->bl
.max_pwrite_zeroes
, INT_MAX
);
1181 int alignment
= MAX(bs
->bl
.pwrite_zeroes_alignment
,
1182 bs
->bl
.request_alignment
);
1184 assert(is_power_of_2(alignment
));
1185 head
= offset
& (alignment
- 1);
1186 tail
= (offset
+ count
) & (alignment
- 1);
1187 max_write_zeroes
&= ~(alignment
- 1);
1189 while (count
> 0 && !ret
) {
1192 /* Align request. Block drivers can expect the "bulk" of the request
1193 * to be aligned, and that unaligned requests do not cross cluster
1197 /* Make a small request up to the first aligned sector. */
1198 num
= MIN(count
, alignment
- head
);
1200 } else if (tail
&& num
> alignment
) {
1201 /* Shorten the request to the last aligned sector. */
1205 /* limit request size */
1206 if (num
> max_write_zeroes
) {
1207 num
= max_write_zeroes
;
1211 /* First try the efficient write zeroes operation */
1212 if (drv
->bdrv_co_pwrite_zeroes
) {
1213 ret
= drv
->bdrv_co_pwrite_zeroes(bs
, offset
, num
,
1214 flags
& bs
->supported_zero_flags
);
1215 if (ret
!= -ENOTSUP
&& (flags
& BDRV_REQ_FUA
) &&
1216 !(bs
->supported_zero_flags
& BDRV_REQ_FUA
)) {
1220 assert(!bs
->supported_zero_flags
);
1223 if (ret
== -ENOTSUP
) {
1224 /* Fall back to bounce buffer if write zeroes is unsupported */
1225 int max_transfer
= MIN_NON_ZERO(bs
->bl
.max_transfer
,
1226 MAX_WRITE_ZEROES_BOUNCE_BUFFER
);
1227 BdrvRequestFlags write_flags
= flags
& ~BDRV_REQ_ZERO_WRITE
;
1229 if ((flags
& BDRV_REQ_FUA
) &&
1230 !(bs
->supported_write_flags
& BDRV_REQ_FUA
)) {
1231 /* No need for bdrv_driver_pwrite() to do a fallback
1232 * flush on each chunk; use just one at the end */
1233 write_flags
&= ~BDRV_REQ_FUA
;
1236 num
= MIN(num
, max_transfer
);
1238 if (iov
.iov_base
== NULL
) {
1239 iov
.iov_base
= qemu_try_blockalign(bs
, num
);
1240 if (iov
.iov_base
== NULL
) {
1244 memset(iov
.iov_base
, 0, num
);
1246 qemu_iovec_init_external(&qiov
, &iov
, 1);
1248 ret
= bdrv_driver_pwritev(bs
, offset
, num
, &qiov
, write_flags
);
1250 /* Keep bounce buffer around if it is big enough for all
1251 * all future requests.
1253 if (num
< max_transfer
) {
1254 qemu_vfree(iov
.iov_base
);
1255 iov
.iov_base
= NULL
;
1264 if (ret
== 0 && need_flush
) {
1265 ret
= bdrv_co_flush(bs
);
1267 qemu_vfree(iov
.iov_base
);
1272 * Forwards an already correctly aligned write request to the BlockDriver,
1273 * after possibly fragmenting it.
1275 static int coroutine_fn
bdrv_aligned_pwritev(BlockDriverState
*bs
,
1276 BdrvTrackedRequest
*req
, int64_t offset
, unsigned int bytes
,
1277 int64_t align
, QEMUIOVector
*qiov
, int flags
)
1279 BlockDriver
*drv
= bs
->drv
;
1283 int64_t start_sector
= offset
>> BDRV_SECTOR_BITS
;
1284 int64_t end_sector
= DIV_ROUND_UP(offset
+ bytes
, BDRV_SECTOR_SIZE
);
1285 uint64_t bytes_remaining
= bytes
;
1288 assert(is_power_of_2(align
));
1289 assert((offset
& (align
- 1)) == 0);
1290 assert((bytes
& (align
- 1)) == 0);
1291 assert(!qiov
|| bytes
== qiov
->size
);
1292 assert((bs
->open_flags
& BDRV_O_NO_IO
) == 0);
1293 assert(!(flags
& ~BDRV_REQ_MASK
));
1294 max_transfer
= QEMU_ALIGN_DOWN(MIN_NON_ZERO(bs
->bl
.max_transfer
, INT_MAX
),
1297 waited
= wait_serialising_requests(req
);
1298 assert(!waited
|| !req
->serialising
);
1299 assert(req
->overlap_offset
<= offset
);
1300 assert(offset
+ bytes
<= req
->overlap_offset
+ req
->overlap_bytes
);
1302 ret
= notifier_with_return_list_notify(&bs
->before_write_notifiers
, req
);
1304 if (!ret
&& bs
->detect_zeroes
!= BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF
&&
1305 !(flags
& BDRV_REQ_ZERO_WRITE
) && drv
->bdrv_co_pwrite_zeroes
&&
1306 qemu_iovec_is_zero(qiov
)) {
1307 flags
|= BDRV_REQ_ZERO_WRITE
;
1308 if (bs
->detect_zeroes
== BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP
) {
1309 flags
|= BDRV_REQ_MAY_UNMAP
;
1314 /* Do nothing, write notifier decided to fail this request */
1315 } else if (flags
& BDRV_REQ_ZERO_WRITE
) {
1316 bdrv_debug_event(bs
, BLKDBG_PWRITEV_ZERO
);
1317 ret
= bdrv_co_do_pwrite_zeroes(bs
, offset
, bytes
, flags
);
1318 } else if (bytes
<= max_transfer
) {
1319 bdrv_debug_event(bs
, BLKDBG_PWRITEV
);
1320 ret
= bdrv_driver_pwritev(bs
, offset
, bytes
, qiov
, flags
);
1322 bdrv_debug_event(bs
, BLKDBG_PWRITEV
);
1323 while (bytes_remaining
) {
1324 int num
= MIN(bytes_remaining
, max_transfer
);
1325 QEMUIOVector local_qiov
;
1326 int local_flags
= flags
;
1329 if (num
< bytes_remaining
&& (flags
& BDRV_REQ_FUA
) &&
1330 !(bs
->supported_write_flags
& BDRV_REQ_FUA
)) {
1331 /* If FUA is going to be emulated by flush, we only
1332 * need to flush on the last iteration */
1333 local_flags
&= ~BDRV_REQ_FUA
;
1335 qemu_iovec_init(&local_qiov
, qiov
->niov
);
1336 qemu_iovec_concat(&local_qiov
, qiov
, bytes
- bytes_remaining
, num
);
1338 ret
= bdrv_driver_pwritev(bs
, offset
+ bytes
- bytes_remaining
,
1339 num
, &local_qiov
, local_flags
);
1340 qemu_iovec_destroy(&local_qiov
);
1344 bytes_remaining
-= num
;
1347 bdrv_debug_event(bs
, BLKDBG_PWRITEV_DONE
);
1350 bdrv_set_dirty(bs
, start_sector
, end_sector
- start_sector
);
1352 if (bs
->wr_highest_offset
< offset
+ bytes
) {
1353 bs
->wr_highest_offset
= offset
+ bytes
;
1357 bs
->total_sectors
= MAX(bs
->total_sectors
, end_sector
);
1364 static int coroutine_fn
bdrv_co_do_zero_pwritev(BlockDriverState
*bs
,
1367 BdrvRequestFlags flags
,
1368 BdrvTrackedRequest
*req
)
1370 uint8_t *buf
= NULL
;
1371 QEMUIOVector local_qiov
;
1373 uint64_t align
= bs
->bl
.request_alignment
;
1374 unsigned int head_padding_bytes
, tail_padding_bytes
;
1377 head_padding_bytes
= offset
& (align
- 1);
1378 tail_padding_bytes
= align
- ((offset
+ bytes
) & (align
- 1));
1381 assert(flags
& BDRV_REQ_ZERO_WRITE
);
1382 if (head_padding_bytes
|| tail_padding_bytes
) {
1383 buf
= qemu_blockalign(bs
, align
);
1384 iov
= (struct iovec
) {
1388 qemu_iovec_init_external(&local_qiov
, &iov
, 1);
1390 if (head_padding_bytes
) {
1391 uint64_t zero_bytes
= MIN(bytes
, align
- head_padding_bytes
);
1393 /* RMW the unaligned part before head. */
1394 mark_request_serialising(req
, align
);
1395 wait_serialising_requests(req
);
1396 bdrv_debug_event(bs
, BLKDBG_PWRITEV_RMW_HEAD
);
1397 ret
= bdrv_aligned_preadv(bs
, req
, offset
& ~(align
- 1), align
,
1398 align
, &local_qiov
, 0);
1402 bdrv_debug_event(bs
, BLKDBG_PWRITEV_RMW_AFTER_HEAD
);
1404 memset(buf
+ head_padding_bytes
, 0, zero_bytes
);
1405 ret
= bdrv_aligned_pwritev(bs
, req
, offset
& ~(align
- 1), align
,
1407 flags
& ~BDRV_REQ_ZERO_WRITE
);
1411 offset
+= zero_bytes
;
1412 bytes
-= zero_bytes
;
1415 assert(!bytes
|| (offset
& (align
- 1)) == 0);
1416 if (bytes
>= align
) {
1417 /* Write the aligned part in the middle. */
1418 uint64_t aligned_bytes
= bytes
& ~(align
- 1);
1419 ret
= bdrv_aligned_pwritev(bs
, req
, offset
, aligned_bytes
, align
,
1424 bytes
-= aligned_bytes
;
1425 offset
+= aligned_bytes
;
1428 assert(!bytes
|| (offset
& (align
- 1)) == 0);
1430 assert(align
== tail_padding_bytes
+ bytes
);
1431 /* RMW the unaligned part after tail. */
1432 mark_request_serialising(req
, align
);
1433 wait_serialising_requests(req
);
1434 bdrv_debug_event(bs
, BLKDBG_PWRITEV_RMW_TAIL
);
1435 ret
= bdrv_aligned_preadv(bs
, req
, offset
, align
,
1436 align
, &local_qiov
, 0);
1440 bdrv_debug_event(bs
, BLKDBG_PWRITEV_RMW_AFTER_TAIL
);
1442 memset(buf
, 0, bytes
);
1443 ret
= bdrv_aligned_pwritev(bs
, req
, offset
, align
, align
,
1444 &local_qiov
, flags
& ~BDRV_REQ_ZERO_WRITE
);
1453 * Handle a write request in coroutine context
1455 int coroutine_fn
bdrv_co_pwritev(BdrvChild
*child
,
1456 int64_t offset
, unsigned int bytes
, QEMUIOVector
*qiov
,
1457 BdrvRequestFlags flags
)
1459 BlockDriverState
*bs
= child
->bs
;
1460 BdrvTrackedRequest req
;
1461 uint64_t align
= bs
->bl
.request_alignment
;
1462 uint8_t *head_buf
= NULL
;
1463 uint8_t *tail_buf
= NULL
;
1464 QEMUIOVector local_qiov
;
1465 bool use_local_qiov
= false;
1471 if (bs
->read_only
) {
1474 assert(!(bs
->open_flags
& BDRV_O_INACTIVE
));
1476 ret
= bdrv_check_byte_request(bs
, offset
, bytes
);
1482 * Align write if necessary by performing a read-modify-write cycle.
1483 * Pad qiov with the read parts and be sure to have a tracked request not
1484 * only for bdrv_aligned_pwritev, but also for the reads of the RMW cycle.
1486 tracked_request_begin(&req
, bs
, offset
, bytes
, BDRV_TRACKED_WRITE
);
1489 ret
= bdrv_co_do_zero_pwritev(bs
, offset
, bytes
, flags
, &req
);
1493 if (offset
& (align
- 1)) {
1494 QEMUIOVector head_qiov
;
1495 struct iovec head_iov
;
1497 mark_request_serialising(&req
, align
);
1498 wait_serialising_requests(&req
);
1500 head_buf
= qemu_blockalign(bs
, align
);
1501 head_iov
= (struct iovec
) {
1502 .iov_base
= head_buf
,
1505 qemu_iovec_init_external(&head_qiov
, &head_iov
, 1);
1507 bdrv_debug_event(bs
, BLKDBG_PWRITEV_RMW_HEAD
);
1508 ret
= bdrv_aligned_preadv(bs
, &req
, offset
& ~(align
- 1), align
,
1509 align
, &head_qiov
, 0);
1513 bdrv_debug_event(bs
, BLKDBG_PWRITEV_RMW_AFTER_HEAD
);
1515 qemu_iovec_init(&local_qiov
, qiov
->niov
+ 2);
1516 qemu_iovec_add(&local_qiov
, head_buf
, offset
& (align
- 1));
1517 qemu_iovec_concat(&local_qiov
, qiov
, 0, qiov
->size
);
1518 use_local_qiov
= true;
1520 bytes
+= offset
& (align
- 1);
1521 offset
= offset
& ~(align
- 1);
1523 /* We have read the tail already if the request is smaller
1524 * than one aligned block.
1526 if (bytes
< align
) {
1527 qemu_iovec_add(&local_qiov
, head_buf
+ bytes
, align
- bytes
);
1532 if ((offset
+ bytes
) & (align
- 1)) {
1533 QEMUIOVector tail_qiov
;
1534 struct iovec tail_iov
;
1538 mark_request_serialising(&req
, align
);
1539 waited
= wait_serialising_requests(&req
);
1540 assert(!waited
|| !use_local_qiov
);
1542 tail_buf
= qemu_blockalign(bs
, align
);
1543 tail_iov
= (struct iovec
) {
1544 .iov_base
= tail_buf
,
1547 qemu_iovec_init_external(&tail_qiov
, &tail_iov
, 1);
1549 bdrv_debug_event(bs
, BLKDBG_PWRITEV_RMW_TAIL
);
1550 ret
= bdrv_aligned_preadv(bs
, &req
, (offset
+ bytes
) & ~(align
- 1), align
,
1551 align
, &tail_qiov
, 0);
1555 bdrv_debug_event(bs
, BLKDBG_PWRITEV_RMW_AFTER_TAIL
);
1557 if (!use_local_qiov
) {
1558 qemu_iovec_init(&local_qiov
, qiov
->niov
+ 1);
1559 qemu_iovec_concat(&local_qiov
, qiov
, 0, qiov
->size
);
1560 use_local_qiov
= true;
1563 tail_bytes
= (offset
+ bytes
) & (align
- 1);
1564 qemu_iovec_add(&local_qiov
, tail_buf
+ tail_bytes
, align
- tail_bytes
);
1566 bytes
= ROUND_UP(bytes
, align
);
1569 ret
= bdrv_aligned_pwritev(bs
, &req
, offset
, bytes
, align
,
1570 use_local_qiov
? &local_qiov
: qiov
,
1575 if (use_local_qiov
) {
1576 qemu_iovec_destroy(&local_qiov
);
1578 qemu_vfree(head_buf
);
1579 qemu_vfree(tail_buf
);
1581 tracked_request_end(&req
);
1585 static int coroutine_fn
bdrv_co_do_writev(BdrvChild
*child
,
1586 int64_t sector_num
, int nb_sectors
, QEMUIOVector
*qiov
,
1587 BdrvRequestFlags flags
)
1589 if (nb_sectors
< 0 || nb_sectors
> BDRV_REQUEST_MAX_SECTORS
) {
1593 return bdrv_co_pwritev(child
, sector_num
<< BDRV_SECTOR_BITS
,
1594 nb_sectors
<< BDRV_SECTOR_BITS
, qiov
, flags
);
1597 int coroutine_fn
bdrv_co_writev(BdrvChild
*child
, int64_t sector_num
,
1598 int nb_sectors
, QEMUIOVector
*qiov
)
1600 trace_bdrv_co_writev(child
->bs
, sector_num
, nb_sectors
);
1602 return bdrv_co_do_writev(child
, sector_num
, nb_sectors
, qiov
, 0);
1605 int coroutine_fn
bdrv_co_pwrite_zeroes(BdrvChild
*child
, int64_t offset
,
1606 int count
, BdrvRequestFlags flags
)
1608 trace_bdrv_co_pwrite_zeroes(child
->bs
, offset
, count
, flags
);
1610 if (!(child
->bs
->open_flags
& BDRV_O_UNMAP
)) {
1611 flags
&= ~BDRV_REQ_MAY_UNMAP
;
1614 return bdrv_co_pwritev(child
, offset
, count
, NULL
,
1615 BDRV_REQ_ZERO_WRITE
| flags
);
1618 typedef struct BdrvCoGetBlockStatusData
{
1619 BlockDriverState
*bs
;
1620 BlockDriverState
*base
;
1621 BlockDriverState
**file
;
1627 } BdrvCoGetBlockStatusData
;
1630 * Returns the allocation status of the specified sectors.
1631 * Drivers not implementing the functionality are assumed to not support
1632 * backing files, hence all their sectors are reported as allocated.
1634 * If 'sector_num' is beyond the end of the disk image the return value is 0
1635 * and 'pnum' is set to 0.
1637 * 'pnum' is set to the number of sectors (including and immediately following
1638 * the specified sector) that are known to be in the same
1639 * allocated/unallocated state.
1641 * 'nb_sectors' is the max value 'pnum' should be set to. If nb_sectors goes
1642 * beyond the end of the disk image it will be clamped.
1644 * If returned value is positive and BDRV_BLOCK_OFFSET_VALID bit is set, 'file'
1645 * points to the BDS which the sector range is allocated in.
1647 static int64_t coroutine_fn
bdrv_co_get_block_status(BlockDriverState
*bs
,
1649 int nb_sectors
, int *pnum
,
1650 BlockDriverState
**file
)
1652 int64_t total_sectors
;
1656 total_sectors
= bdrv_nb_sectors(bs
);
1657 if (total_sectors
< 0) {
1658 return total_sectors
;
1661 if (sector_num
>= total_sectors
) {
1666 n
= total_sectors
- sector_num
;
1667 if (n
< nb_sectors
) {
1671 if (!bs
->drv
->bdrv_co_get_block_status
) {
1673 ret
= BDRV_BLOCK_DATA
| BDRV_BLOCK_ALLOCATED
;
1674 if (bs
->drv
->protocol_name
) {
1675 ret
|= BDRV_BLOCK_OFFSET_VALID
| (sector_num
* BDRV_SECTOR_SIZE
);
1681 ret
= bs
->drv
->bdrv_co_get_block_status(bs
, sector_num
, nb_sectors
, pnum
,
1688 if (ret
& BDRV_BLOCK_RAW
) {
1689 assert(ret
& BDRV_BLOCK_OFFSET_VALID
);
1690 return bdrv_get_block_status(bs
->file
->bs
, ret
>> BDRV_SECTOR_BITS
,
1694 if (ret
& (BDRV_BLOCK_DATA
| BDRV_BLOCK_ZERO
)) {
1695 ret
|= BDRV_BLOCK_ALLOCATED
;
1697 if (bdrv_unallocated_blocks_are_zero(bs
)) {
1698 ret
|= BDRV_BLOCK_ZERO
;
1699 } else if (bs
->backing
) {
1700 BlockDriverState
*bs2
= bs
->backing
->bs
;
1701 int64_t nb_sectors2
= bdrv_nb_sectors(bs2
);
1702 if (nb_sectors2
>= 0 && sector_num
>= nb_sectors2
) {
1703 ret
|= BDRV_BLOCK_ZERO
;
1708 if (*file
&& *file
!= bs
&&
1709 (ret
& BDRV_BLOCK_DATA
) && !(ret
& BDRV_BLOCK_ZERO
) &&
1710 (ret
& BDRV_BLOCK_OFFSET_VALID
)) {
1711 BlockDriverState
*file2
;
1714 ret2
= bdrv_co_get_block_status(*file
, ret
>> BDRV_SECTOR_BITS
,
1715 *pnum
, &file_pnum
, &file2
);
1717 /* Ignore errors. This is just providing extra information, it
1718 * is useful but not necessary.
1721 /* !file_pnum indicates an offset at or beyond the EOF; it is
1722 * perfectly valid for the format block driver to point to such
1723 * offsets, so catch it and mark everything as zero */
1724 ret
|= BDRV_BLOCK_ZERO
;
1726 /* Limit request to the range reported by the protocol driver */
1728 ret
|= (ret2
& BDRV_BLOCK_ZERO
);
1736 static int64_t coroutine_fn
bdrv_co_get_block_status_above(BlockDriverState
*bs
,
1737 BlockDriverState
*base
,
1741 BlockDriverState
**file
)
1743 BlockDriverState
*p
;
1747 for (p
= bs
; p
!= base
; p
= backing_bs(p
)) {
1748 ret
= bdrv_co_get_block_status(p
, sector_num
, nb_sectors
, pnum
, file
);
1749 if (ret
< 0 || ret
& BDRV_BLOCK_ALLOCATED
) {
1752 /* [sector_num, pnum] unallocated on this layer, which could be only
1753 * the first part of [sector_num, nb_sectors]. */
1754 nb_sectors
= MIN(nb_sectors
, *pnum
);
1759 /* Coroutine wrapper for bdrv_get_block_status_above() */
1760 static void coroutine_fn
bdrv_get_block_status_above_co_entry(void *opaque
)
1762 BdrvCoGetBlockStatusData
*data
= opaque
;
1764 data
->ret
= bdrv_co_get_block_status_above(data
->bs
, data
->base
,
1773 * Synchronous wrapper around bdrv_co_get_block_status_above().
1775 * See bdrv_co_get_block_status_above() for details.
1777 int64_t bdrv_get_block_status_above(BlockDriverState
*bs
,
1778 BlockDriverState
*base
,
1780 int nb_sectors
, int *pnum
,
1781 BlockDriverState
**file
)
1784 BdrvCoGetBlockStatusData data
= {
1788 .sector_num
= sector_num
,
1789 .nb_sectors
= nb_sectors
,
1794 if (qemu_in_coroutine()) {
1795 /* Fast-path if already in coroutine context */
1796 bdrv_get_block_status_above_co_entry(&data
);
1798 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
1800 co
= qemu_coroutine_create(bdrv_get_block_status_above_co_entry
,
1802 qemu_coroutine_enter(co
);
1803 while (!data
.done
) {
1804 aio_poll(aio_context
, true);
1810 int64_t bdrv_get_block_status(BlockDriverState
*bs
,
1812 int nb_sectors
, int *pnum
,
1813 BlockDriverState
**file
)
1815 return bdrv_get_block_status_above(bs
, backing_bs(bs
),
1816 sector_num
, nb_sectors
, pnum
, file
);
1819 int coroutine_fn
bdrv_is_allocated(BlockDriverState
*bs
, int64_t sector_num
,
1820 int nb_sectors
, int *pnum
)
1822 BlockDriverState
*file
;
1823 int64_t ret
= bdrv_get_block_status(bs
, sector_num
, nb_sectors
, pnum
,
1828 return !!(ret
& BDRV_BLOCK_ALLOCATED
);
1832 * Given an image chain: ... -> [BASE] -> [INTER1] -> [INTER2] -> [TOP]
1834 * Return true if the given sector is allocated in any image between
1835 * BASE and TOP (inclusive). BASE can be NULL to check if the given
1836 * sector is allocated in any image of the chain. Return false otherwise.
1838 * 'pnum' is set to the number of sectors (including and immediately following
1839 * the specified sector) that are known to be in the same
1840 * allocated/unallocated state.
1843 int bdrv_is_allocated_above(BlockDriverState
*top
,
1844 BlockDriverState
*base
,
1846 int nb_sectors
, int *pnum
)
1848 BlockDriverState
*intermediate
;
1849 int ret
, n
= nb_sectors
;
1852 while (intermediate
&& intermediate
!= base
) {
1854 ret
= bdrv_is_allocated(intermediate
, sector_num
, nb_sectors
,
1864 * [sector_num, nb_sectors] is unallocated on top but intermediate
1867 * [sector_num+x, nr_sectors] allocated.
1869 if (n
> pnum_inter
&&
1870 (intermediate
== top
||
1871 sector_num
+ pnum_inter
< intermediate
->total_sectors
)) {
1875 intermediate
= backing_bs(intermediate
);
1882 int bdrv_write_compressed(BlockDriverState
*bs
, int64_t sector_num
,
1883 const uint8_t *buf
, int nb_sectors
)
1885 BlockDriver
*drv
= bs
->drv
;
1891 if (!drv
->bdrv_write_compressed
) {
1894 ret
= bdrv_check_request(bs
, sector_num
, nb_sectors
);
1899 assert(QLIST_EMPTY(&bs
->dirty_bitmaps
));
1901 return drv
->bdrv_write_compressed(bs
, sector_num
, buf
, nb_sectors
);
1904 typedef struct BdrvVmstateCo
{
1905 BlockDriverState
*bs
;
1912 static int coroutine_fn
1913 bdrv_co_rw_vmstate(BlockDriverState
*bs
, QEMUIOVector
*qiov
, int64_t pos
,
1916 BlockDriver
*drv
= bs
->drv
;
1920 } else if (drv
->bdrv_load_vmstate
) {
1921 return is_read
? drv
->bdrv_load_vmstate(bs
, qiov
, pos
)
1922 : drv
->bdrv_save_vmstate(bs
, qiov
, pos
);
1923 } else if (bs
->file
) {
1924 return bdrv_co_rw_vmstate(bs
->file
->bs
, qiov
, pos
, is_read
);
1930 static void coroutine_fn
bdrv_co_rw_vmstate_entry(void *opaque
)
1932 BdrvVmstateCo
*co
= opaque
;
1933 co
->ret
= bdrv_co_rw_vmstate(co
->bs
, co
->qiov
, co
->pos
, co
->is_read
);
1937 bdrv_rw_vmstate(BlockDriverState
*bs
, QEMUIOVector
*qiov
, int64_t pos
,
1940 if (qemu_in_coroutine()) {
1941 return bdrv_co_rw_vmstate(bs
, qiov
, pos
, is_read
);
1943 BdrvVmstateCo data
= {
1948 .ret
= -EINPROGRESS
,
1950 Coroutine
*co
= qemu_coroutine_create(bdrv_co_rw_vmstate_entry
, &data
);
1952 qemu_coroutine_enter(co
);
1953 while (data
.ret
== -EINPROGRESS
) {
1954 aio_poll(bdrv_get_aio_context(bs
), true);
1960 int bdrv_save_vmstate(BlockDriverState
*bs
, const uint8_t *buf
,
1961 int64_t pos
, int size
)
1964 struct iovec iov
= {
1965 .iov_base
= (void *) buf
,
1970 qemu_iovec_init_external(&qiov
, &iov
, 1);
1972 ret
= bdrv_writev_vmstate(bs
, &qiov
, pos
);
1980 int bdrv_writev_vmstate(BlockDriverState
*bs
, QEMUIOVector
*qiov
, int64_t pos
)
1982 return bdrv_rw_vmstate(bs
, qiov
, pos
, false);
1985 int bdrv_load_vmstate(BlockDriverState
*bs
, uint8_t *buf
,
1986 int64_t pos
, int size
)
1989 struct iovec iov
= {
1995 qemu_iovec_init_external(&qiov
, &iov
, 1);
1996 ret
= bdrv_readv_vmstate(bs
, &qiov
, pos
);
2004 int bdrv_readv_vmstate(BlockDriverState
*bs
, QEMUIOVector
*qiov
, int64_t pos
)
2006 return bdrv_rw_vmstate(bs
, qiov
, pos
, true);
2009 /**************************************************************/
2012 BlockAIOCB
*bdrv_aio_readv(BdrvChild
*child
, int64_t sector_num
,
2013 QEMUIOVector
*qiov
, int nb_sectors
,
2014 BlockCompletionFunc
*cb
, void *opaque
)
2016 trace_bdrv_aio_readv(child
->bs
, sector_num
, nb_sectors
, opaque
);
2018 return bdrv_co_aio_rw_vector(child
, sector_num
, qiov
, nb_sectors
, 0,
2022 BlockAIOCB
*bdrv_aio_writev(BdrvChild
*child
, int64_t sector_num
,
2023 QEMUIOVector
*qiov
, int nb_sectors
,
2024 BlockCompletionFunc
*cb
, void *opaque
)
2026 trace_bdrv_aio_writev(child
->bs
, sector_num
, nb_sectors
, opaque
);
2028 return bdrv_co_aio_rw_vector(child
, sector_num
, qiov
, nb_sectors
, 0,
2032 void bdrv_aio_cancel(BlockAIOCB
*acb
)
2035 bdrv_aio_cancel_async(acb
);
2036 while (acb
->refcnt
> 1) {
2037 if (acb
->aiocb_info
->get_aio_context
) {
2038 aio_poll(acb
->aiocb_info
->get_aio_context(acb
), true);
2039 } else if (acb
->bs
) {
2040 aio_poll(bdrv_get_aio_context(acb
->bs
), true);
2045 qemu_aio_unref(acb
);
2048 /* Async version of aio cancel. The caller is not blocked if the acb implements
2049 * cancel_async, otherwise we do nothing and let the request normally complete.
2050 * In either case the completion callback must be called. */
2051 void bdrv_aio_cancel_async(BlockAIOCB
*acb
)
2053 if (acb
->aiocb_info
->cancel_async
) {
2054 acb
->aiocb_info
->cancel_async(acb
);
2058 /**************************************************************/
2059 /* async block device emulation */
2061 typedef struct BlockRequest
{
2063 /* Used during read, write, trim */
2070 /* Used during ioctl */
2076 BlockCompletionFunc
*cb
;
2082 typedef struct BlockAIOCBCoroutine
{
2090 } BlockAIOCBCoroutine
;
2092 static const AIOCBInfo bdrv_em_co_aiocb_info
= {
2093 .aiocb_size
= sizeof(BlockAIOCBCoroutine
),
2096 static void bdrv_co_complete(BlockAIOCBCoroutine
*acb
)
2098 if (!acb
->need_bh
) {
2099 acb
->common
.cb(acb
->common
.opaque
, acb
->req
.error
);
2100 qemu_aio_unref(acb
);
2104 static void bdrv_co_em_bh(void *opaque
)
2106 BlockAIOCBCoroutine
*acb
= opaque
;
2108 assert(!acb
->need_bh
);
2109 qemu_bh_delete(acb
->bh
);
2110 bdrv_co_complete(acb
);
2113 static void bdrv_co_maybe_schedule_bh(BlockAIOCBCoroutine
*acb
)
2115 acb
->need_bh
= false;
2116 if (acb
->req
.error
!= -EINPROGRESS
) {
2117 BlockDriverState
*bs
= acb
->common
.bs
;
2119 acb
->bh
= aio_bh_new(bdrv_get_aio_context(bs
), bdrv_co_em_bh
, acb
);
2120 qemu_bh_schedule(acb
->bh
);
2124 /* Invoke bdrv_co_do_readv/bdrv_co_do_writev */
2125 static void coroutine_fn
bdrv_co_do_rw(void *opaque
)
2127 BlockAIOCBCoroutine
*acb
= opaque
;
2129 if (!acb
->is_write
) {
2130 acb
->req
.error
= bdrv_co_do_readv(acb
->child
, acb
->req
.sector
,
2131 acb
->req
.nb_sectors
, acb
->req
.qiov
, acb
->req
.flags
);
2133 acb
->req
.error
= bdrv_co_do_writev(acb
->child
, acb
->req
.sector
,
2134 acb
->req
.nb_sectors
, acb
->req
.qiov
, acb
->req
.flags
);
2137 bdrv_co_complete(acb
);
2140 static BlockAIOCB
*bdrv_co_aio_rw_vector(BdrvChild
*child
,
2144 BdrvRequestFlags flags
,
2145 BlockCompletionFunc
*cb
,
2150 BlockAIOCBCoroutine
*acb
;
2152 acb
= qemu_aio_get(&bdrv_em_co_aiocb_info
, child
->bs
, cb
, opaque
);
2154 acb
->need_bh
= true;
2155 acb
->req
.error
= -EINPROGRESS
;
2156 acb
->req
.sector
= sector_num
;
2157 acb
->req
.nb_sectors
= nb_sectors
;
2158 acb
->req
.qiov
= qiov
;
2159 acb
->req
.flags
= flags
;
2160 acb
->is_write
= is_write
;
2162 co
= qemu_coroutine_create(bdrv_co_do_rw
, acb
);
2163 qemu_coroutine_enter(co
);
2165 bdrv_co_maybe_schedule_bh(acb
);
2166 return &acb
->common
;
2169 static void coroutine_fn
bdrv_aio_flush_co_entry(void *opaque
)
2171 BlockAIOCBCoroutine
*acb
= opaque
;
2172 BlockDriverState
*bs
= acb
->common
.bs
;
2174 acb
->req
.error
= bdrv_co_flush(bs
);
2175 bdrv_co_complete(acb
);
2178 BlockAIOCB
*bdrv_aio_flush(BlockDriverState
*bs
,
2179 BlockCompletionFunc
*cb
, void *opaque
)
2181 trace_bdrv_aio_flush(bs
, opaque
);
2184 BlockAIOCBCoroutine
*acb
;
2186 acb
= qemu_aio_get(&bdrv_em_co_aiocb_info
, bs
, cb
, opaque
);
2187 acb
->need_bh
= true;
2188 acb
->req
.error
= -EINPROGRESS
;
2190 co
= qemu_coroutine_create(bdrv_aio_flush_co_entry
, acb
);
2191 qemu_coroutine_enter(co
);
2193 bdrv_co_maybe_schedule_bh(acb
);
2194 return &acb
->common
;
2197 static void coroutine_fn
bdrv_aio_discard_co_entry(void *opaque
)
2199 BlockAIOCBCoroutine
*acb
= opaque
;
2200 BlockDriverState
*bs
= acb
->common
.bs
;
2202 acb
->req
.error
= bdrv_co_discard(bs
, acb
->req
.sector
, acb
->req
.nb_sectors
);
2203 bdrv_co_complete(acb
);
2206 BlockAIOCB
*bdrv_aio_discard(BlockDriverState
*bs
,
2207 int64_t sector_num
, int nb_sectors
,
2208 BlockCompletionFunc
*cb
, void *opaque
)
2211 BlockAIOCBCoroutine
*acb
;
2213 trace_bdrv_aio_discard(bs
, sector_num
, nb_sectors
, opaque
);
2215 acb
= qemu_aio_get(&bdrv_em_co_aiocb_info
, bs
, cb
, opaque
);
2216 acb
->need_bh
= true;
2217 acb
->req
.error
= -EINPROGRESS
;
2218 acb
->req
.sector
= sector_num
;
2219 acb
->req
.nb_sectors
= nb_sectors
;
2220 co
= qemu_coroutine_create(bdrv_aio_discard_co_entry
, acb
);
2221 qemu_coroutine_enter(co
);
2223 bdrv_co_maybe_schedule_bh(acb
);
2224 return &acb
->common
;
2227 void *qemu_aio_get(const AIOCBInfo
*aiocb_info
, BlockDriverState
*bs
,
2228 BlockCompletionFunc
*cb
, void *opaque
)
2232 acb
= g_malloc(aiocb_info
->aiocb_size
);
2233 acb
->aiocb_info
= aiocb_info
;
2236 acb
->opaque
= opaque
;
2241 void qemu_aio_ref(void *p
)
2243 BlockAIOCB
*acb
= p
;
2247 void qemu_aio_unref(void *p
)
2249 BlockAIOCB
*acb
= p
;
2250 assert(acb
->refcnt
> 0);
2251 if (--acb
->refcnt
== 0) {
2256 /**************************************************************/
2257 /* Coroutine block device emulation */
2259 typedef struct FlushCo
{
2260 BlockDriverState
*bs
;
2265 static void coroutine_fn
bdrv_flush_co_entry(void *opaque
)
2267 FlushCo
*rwco
= opaque
;
2269 rwco
->ret
= bdrv_co_flush(rwco
->bs
);
2272 int coroutine_fn
bdrv_co_flush(BlockDriverState
*bs
)
2275 BdrvTrackedRequest req
;
2277 if (!bs
|| !bdrv_is_inserted(bs
) || bdrv_is_read_only(bs
) ||
2282 tracked_request_begin(&req
, bs
, 0, 0, BDRV_TRACKED_FLUSH
);
2284 int current_gen
= bs
->write_gen
;
2286 /* Wait until any previous flushes are completed */
2287 while (bs
->flush_started_gen
!= bs
->flushed_gen
) {
2288 qemu_co_queue_wait(&bs
->flush_queue
);
2291 bs
->flush_started_gen
= current_gen
;
2293 /* Write back all layers by calling one driver function */
2294 if (bs
->drv
->bdrv_co_flush
) {
2295 ret
= bs
->drv
->bdrv_co_flush(bs
);
2299 /* Write back cached data to the OS even with cache=unsafe */
2300 BLKDBG_EVENT(bs
->file
, BLKDBG_FLUSH_TO_OS
);
2301 if (bs
->drv
->bdrv_co_flush_to_os
) {
2302 ret
= bs
->drv
->bdrv_co_flush_to_os(bs
);
2308 /* But don't actually force it to the disk with cache=unsafe */
2309 if (bs
->open_flags
& BDRV_O_NO_FLUSH
) {
2313 /* Check if we really need to flush anything */
2314 if (bs
->flushed_gen
== current_gen
) {
2318 BLKDBG_EVENT(bs
->file
, BLKDBG_FLUSH_TO_DISK
);
2319 if (bs
->drv
->bdrv_co_flush_to_disk
) {
2320 ret
= bs
->drv
->bdrv_co_flush_to_disk(bs
);
2321 } else if (bs
->drv
->bdrv_aio_flush
) {
2323 CoroutineIOCompletion co
= {
2324 .coroutine
= qemu_coroutine_self(),
2327 acb
= bs
->drv
->bdrv_aio_flush(bs
, bdrv_co_io_em_complete
, &co
);
2331 qemu_coroutine_yield();
2336 * Some block drivers always operate in either writethrough or unsafe
2337 * mode and don't support bdrv_flush therefore. Usually qemu doesn't
2338 * know how the server works (because the behaviour is hardcoded or
2339 * depends on server-side configuration), so we can't ensure that
2340 * everything is safe on disk. Returning an error doesn't work because
2341 * that would break guests even if the server operates in writethrough
2344 * Let's hope the user knows what he's doing.
2353 /* Now flush the underlying protocol. It will also have BDRV_O_NO_FLUSH
2354 * in the case of cache=unsafe, so there are no useless flushes.
2357 ret
= bs
->file
? bdrv_co_flush(bs
->file
->bs
) : 0;
2359 /* Notify any pending flushes that we have completed */
2360 bs
->flushed_gen
= current_gen
;
2361 qemu_co_queue_restart_all(&bs
->flush_queue
);
2363 tracked_request_end(&req
);
2367 int bdrv_flush(BlockDriverState
*bs
)
2370 FlushCo flush_co
= {
2375 if (qemu_in_coroutine()) {
2376 /* Fast-path if already in coroutine context */
2377 bdrv_flush_co_entry(&flush_co
);
2379 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
2381 co
= qemu_coroutine_create(bdrv_flush_co_entry
, &flush_co
);
2382 qemu_coroutine_enter(co
);
2383 while (flush_co
.ret
== NOT_DONE
) {
2384 aio_poll(aio_context
, true);
2388 return flush_co
.ret
;
2391 typedef struct DiscardCo
{
2392 BlockDriverState
*bs
;
2397 static void coroutine_fn
bdrv_discard_co_entry(void *opaque
)
2399 DiscardCo
*rwco
= opaque
;
2401 rwco
->ret
= bdrv_co_discard(rwco
->bs
, rwco
->sector_num
, rwco
->nb_sectors
);
2404 int coroutine_fn
bdrv_co_discard(BlockDriverState
*bs
, int64_t sector_num
,
2407 BdrvTrackedRequest req
;
2408 int max_discard
, ret
;
2414 ret
= bdrv_check_request(bs
, sector_num
, nb_sectors
);
2417 } else if (bs
->read_only
) {
2420 assert(!(bs
->open_flags
& BDRV_O_INACTIVE
));
2422 /* Do nothing if disabled. */
2423 if (!(bs
->open_flags
& BDRV_O_UNMAP
)) {
2427 if (!bs
->drv
->bdrv_co_discard
&& !bs
->drv
->bdrv_aio_discard
) {
2431 tracked_request_begin(&req
, bs
, sector_num
<< BDRV_SECTOR_BITS
,
2432 nb_sectors
<< BDRV_SECTOR_BITS
, BDRV_TRACKED_DISCARD
);
2434 ret
= notifier_with_return_list_notify(&bs
->before_write_notifiers
, &req
);
2439 max_discard
= MIN_NON_ZERO(bs
->bl
.max_pdiscard
>> BDRV_SECTOR_BITS
,
2440 BDRV_REQUEST_MAX_SECTORS
);
2441 while (nb_sectors
> 0) {
2443 int num
= nb_sectors
;
2444 int discard_alignment
= bs
->bl
.pdiscard_alignment
>> BDRV_SECTOR_BITS
;
2447 if (discard_alignment
&&
2448 num
>= discard_alignment
&&
2449 sector_num
% discard_alignment
) {
2450 if (num
> discard_alignment
) {
2451 num
= discard_alignment
;
2453 num
-= sector_num
% discard_alignment
;
2456 /* limit request size */
2457 if (num
> max_discard
) {
2461 if (bs
->drv
->bdrv_co_discard
) {
2462 ret
= bs
->drv
->bdrv_co_discard(bs
, sector_num
, num
);
2465 CoroutineIOCompletion co
= {
2466 .coroutine
= qemu_coroutine_self(),
2469 acb
= bs
->drv
->bdrv_aio_discard(bs
, sector_num
, nb_sectors
,
2470 bdrv_co_io_em_complete
, &co
);
2475 qemu_coroutine_yield();
2479 if (ret
&& ret
!= -ENOTSUP
) {
2489 bdrv_set_dirty(bs
, req
.offset
>> BDRV_SECTOR_BITS
,
2490 req
.bytes
>> BDRV_SECTOR_BITS
);
2491 tracked_request_end(&req
);
2495 int bdrv_discard(BlockDriverState
*bs
, int64_t sector_num
, int nb_sectors
)
2500 .sector_num
= sector_num
,
2501 .nb_sectors
= nb_sectors
,
2505 if (qemu_in_coroutine()) {
2506 /* Fast-path if already in coroutine context */
2507 bdrv_discard_co_entry(&rwco
);
2509 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
2511 co
= qemu_coroutine_create(bdrv_discard_co_entry
, &rwco
);
2512 qemu_coroutine_enter(co
);
2513 while (rwco
.ret
== NOT_DONE
) {
2514 aio_poll(aio_context
, true);
2521 static int bdrv_co_do_ioctl(BlockDriverState
*bs
, int req
, void *buf
)
2523 BlockDriver
*drv
= bs
->drv
;
2524 BdrvTrackedRequest tracked_req
;
2525 CoroutineIOCompletion co
= {
2526 .coroutine
= qemu_coroutine_self(),
2530 tracked_request_begin(&tracked_req
, bs
, 0, 0, BDRV_TRACKED_IOCTL
);
2531 if (!drv
|| !drv
->bdrv_aio_ioctl
) {
2536 acb
= drv
->bdrv_aio_ioctl(bs
, req
, buf
, bdrv_co_io_em_complete
, &co
);
2541 qemu_coroutine_yield();
2543 tracked_request_end(&tracked_req
);
2548 BlockDriverState
*bs
;
2554 static void coroutine_fn
bdrv_co_ioctl_entry(void *opaque
)
2556 BdrvIoctlCoData
*data
= opaque
;
2557 data
->ret
= bdrv_co_do_ioctl(data
->bs
, data
->req
, data
->buf
);
2560 /* needed for generic scsi interface */
2561 int bdrv_ioctl(BlockDriverState
*bs
, unsigned long int req
, void *buf
)
2563 BdrvIoctlCoData data
= {
2567 .ret
= -EINPROGRESS
,
2570 if (qemu_in_coroutine()) {
2571 /* Fast-path if already in coroutine context */
2572 bdrv_co_ioctl_entry(&data
);
2574 Coroutine
*co
= qemu_coroutine_create(bdrv_co_ioctl_entry
, &data
);
2576 qemu_coroutine_enter(co
);
2577 while (data
.ret
== -EINPROGRESS
) {
2578 aio_poll(bdrv_get_aio_context(bs
), true);
2584 static void coroutine_fn
bdrv_co_aio_ioctl_entry(void *opaque
)
2586 BlockAIOCBCoroutine
*acb
= opaque
;
2587 acb
->req
.error
= bdrv_co_do_ioctl(acb
->common
.bs
,
2588 acb
->req
.req
, acb
->req
.buf
);
2589 bdrv_co_complete(acb
);
2592 BlockAIOCB
*bdrv_aio_ioctl(BlockDriverState
*bs
,
2593 unsigned long int req
, void *buf
,
2594 BlockCompletionFunc
*cb
, void *opaque
)
2596 BlockAIOCBCoroutine
*acb
= qemu_aio_get(&bdrv_em_co_aiocb_info
,
2600 acb
->need_bh
= true;
2601 acb
->req
.error
= -EINPROGRESS
;
2604 co
= qemu_coroutine_create(bdrv_co_aio_ioctl_entry
, acb
);
2605 qemu_coroutine_enter(co
);
2607 bdrv_co_maybe_schedule_bh(acb
);
2608 return &acb
->common
;
2611 void *qemu_blockalign(BlockDriverState
*bs
, size_t size
)
2613 return qemu_memalign(bdrv_opt_mem_align(bs
), size
);
2616 void *qemu_blockalign0(BlockDriverState
*bs
, size_t size
)
2618 return memset(qemu_blockalign(bs
, size
), 0, size
);
2621 void *qemu_try_blockalign(BlockDriverState
*bs
, size_t size
)
2623 size_t align
= bdrv_opt_mem_align(bs
);
2625 /* Ensure that NULL is never returned on success */
2631 return qemu_try_memalign(align
, size
);
2634 void *qemu_try_blockalign0(BlockDriverState
*bs
, size_t size
)
2636 void *mem
= qemu_try_blockalign(bs
, size
);
2639 memset(mem
, 0, size
);
2646 * Check if all memory in this vector is sector aligned.
2648 bool bdrv_qiov_is_aligned(BlockDriverState
*bs
, QEMUIOVector
*qiov
)
2651 size_t alignment
= bdrv_min_mem_align(bs
);
2653 for (i
= 0; i
< qiov
->niov
; i
++) {
2654 if ((uintptr_t) qiov
->iov
[i
].iov_base
% alignment
) {
2657 if (qiov
->iov
[i
].iov_len
% alignment
) {
2665 void bdrv_add_before_write_notifier(BlockDriverState
*bs
,
2666 NotifierWithReturn
*notifier
)
2668 notifier_with_return_list_add(&bs
->before_write_notifiers
, notifier
);
2671 void bdrv_io_plug(BlockDriverState
*bs
)
2675 QLIST_FOREACH(child
, &bs
->children
, next
) {
2676 bdrv_io_plug(child
->bs
);
2679 if (bs
->io_plugged
++ == 0 && bs
->io_plug_disabled
== 0) {
2680 BlockDriver
*drv
= bs
->drv
;
2681 if (drv
&& drv
->bdrv_io_plug
) {
2682 drv
->bdrv_io_plug(bs
);
2687 void bdrv_io_unplug(BlockDriverState
*bs
)
2691 assert(bs
->io_plugged
);
2692 if (--bs
->io_plugged
== 0 && bs
->io_plug_disabled
== 0) {
2693 BlockDriver
*drv
= bs
->drv
;
2694 if (drv
&& drv
->bdrv_io_unplug
) {
2695 drv
->bdrv_io_unplug(bs
);
2699 QLIST_FOREACH(child
, &bs
->children
, next
) {
2700 bdrv_io_unplug(child
->bs
);
2704 void bdrv_io_unplugged_begin(BlockDriverState
*bs
)
2708 if (bs
->io_plug_disabled
++ == 0 && bs
->io_plugged
> 0) {
2709 BlockDriver
*drv
= bs
->drv
;
2710 if (drv
&& drv
->bdrv_io_unplug
) {
2711 drv
->bdrv_io_unplug(bs
);
2715 QLIST_FOREACH(child
, &bs
->children
, next
) {
2716 bdrv_io_unplugged_begin(child
->bs
);
2720 void bdrv_io_unplugged_end(BlockDriverState
*bs
)
2724 assert(bs
->io_plug_disabled
);
2725 QLIST_FOREACH(child
, &bs
->children
, next
) {
2726 bdrv_io_unplugged_end(child
->bs
);
2729 if (--bs
->io_plug_disabled
== 0 && bs
->io_plugged
> 0) {
2730 BlockDriver
*drv
= bs
->drv
;
2731 if (drv
&& drv
->bdrv_io_plug
) {
2732 drv
->bdrv_io_plug(bs
);