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(BlockDriverState
*bs
,
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 void bdrv_refresh_limits(BlockDriverState
*bs
, Error
**errp
)
72 BlockDriver
*drv
= bs
->drv
;
73 Error
*local_err
= NULL
;
75 memset(&bs
->bl
, 0, sizeof(bs
->bl
));
81 /* Take some limits from the children as a default */
83 bdrv_refresh_limits(bs
->file
->bs
, &local_err
);
85 error_propagate(errp
, local_err
);
88 bs
->bl
.opt_transfer_length
= bs
->file
->bs
->bl
.opt_transfer_length
;
89 bs
->bl
.max_transfer_length
= bs
->file
->bs
->bl
.max_transfer_length
;
90 bs
->bl
.min_mem_alignment
= bs
->file
->bs
->bl
.min_mem_alignment
;
91 bs
->bl
.opt_mem_alignment
= bs
->file
->bs
->bl
.opt_mem_alignment
;
92 bs
->bl
.max_iov
= bs
->file
->bs
->bl
.max_iov
;
94 bs
->bl
.min_mem_alignment
= 512;
95 bs
->bl
.opt_mem_alignment
= getpagesize();
97 /* Safe default since most protocols use readv()/writev()/etc */
98 bs
->bl
.max_iov
= IOV_MAX
;
102 bdrv_refresh_limits(bs
->backing
->bs
, &local_err
);
104 error_propagate(errp
, local_err
);
107 bs
->bl
.opt_transfer_length
=
108 MAX(bs
->bl
.opt_transfer_length
,
109 bs
->backing
->bs
->bl
.opt_transfer_length
);
110 bs
->bl
.max_transfer_length
=
111 MIN_NON_ZERO(bs
->bl
.max_transfer_length
,
112 bs
->backing
->bs
->bl
.max_transfer_length
);
113 bs
->bl
.opt_mem_alignment
=
114 MAX(bs
->bl
.opt_mem_alignment
,
115 bs
->backing
->bs
->bl
.opt_mem_alignment
);
116 bs
->bl
.min_mem_alignment
=
117 MAX(bs
->bl
.min_mem_alignment
,
118 bs
->backing
->bs
->bl
.min_mem_alignment
);
121 bs
->backing
->bs
->bl
.max_iov
);
124 /* Then let the driver override it */
125 if (drv
->bdrv_refresh_limits
) {
126 drv
->bdrv_refresh_limits(bs
, errp
);
131 * The copy-on-read flag is actually a reference count so multiple users may
132 * use the feature without worrying about clobbering its previous state.
133 * Copy-on-read stays enabled until all users have called to disable it.
135 void bdrv_enable_copy_on_read(BlockDriverState
*bs
)
140 void bdrv_disable_copy_on_read(BlockDriverState
*bs
)
142 assert(bs
->copy_on_read
> 0);
146 /* Check if any requests are in-flight (including throttled requests) */
147 bool bdrv_requests_pending(BlockDriverState
*bs
)
151 if (!QLIST_EMPTY(&bs
->tracked_requests
)) {
155 QLIST_FOREACH(child
, &bs
->children
, next
) {
156 if (bdrv_requests_pending(child
->bs
)) {
164 static void bdrv_drain_recurse(BlockDriverState
*bs
)
168 if (bs
->drv
&& bs
->drv
->bdrv_drain
) {
169 bs
->drv
->bdrv_drain(bs
);
171 QLIST_FOREACH(child
, &bs
->children
, next
) {
172 bdrv_drain_recurse(child
->bs
);
178 BlockDriverState
*bs
;
183 static void bdrv_drain_poll(BlockDriverState
*bs
)
189 busy
= bdrv_requests_pending(bs
);
190 busy
|= aio_poll(bdrv_get_aio_context(bs
), busy
);
194 static void bdrv_co_drain_bh_cb(void *opaque
)
196 BdrvCoDrainData
*data
= opaque
;
197 Coroutine
*co
= data
->co
;
199 qemu_bh_delete(data
->bh
);
200 bdrv_drain_poll(data
->bs
);
202 qemu_coroutine_enter(co
, NULL
);
205 static void coroutine_fn
bdrv_co_yield_to_drain(BlockDriverState
*bs
)
207 BdrvCoDrainData data
;
209 /* Calling bdrv_drain() from a BH ensures the current coroutine yields and
210 * other coroutines run if they were queued from
211 * qemu_co_queue_run_restart(). */
213 assert(qemu_in_coroutine());
214 data
= (BdrvCoDrainData
) {
215 .co
= qemu_coroutine_self(),
218 .bh
= aio_bh_new(bdrv_get_aio_context(bs
), bdrv_co_drain_bh_cb
, &data
),
220 qemu_bh_schedule(data
.bh
);
222 qemu_coroutine_yield();
223 /* If we are resumed from some other event (such as an aio completion or a
224 * timer callback), it is a bug in the caller that should be fixed. */
228 void bdrv_drained_begin(BlockDriverState
*bs
)
230 if (!bs
->quiesce_counter
++) {
231 aio_disable_external(bdrv_get_aio_context(bs
));
232 bdrv_parent_drained_begin(bs
);
235 bdrv_io_unplugged_begin(bs
);
236 bdrv_drain_recurse(bs
);
237 if (qemu_in_coroutine()) {
238 bdrv_co_yield_to_drain(bs
);
242 bdrv_io_unplugged_end(bs
);
245 void bdrv_drained_end(BlockDriverState
*bs
)
247 assert(bs
->quiesce_counter
> 0);
248 if (--bs
->quiesce_counter
> 0) {
252 bdrv_parent_drained_end(bs
);
253 aio_enable_external(bdrv_get_aio_context(bs
));
257 * Wait for pending requests to complete on a single BlockDriverState subtree,
258 * and suspend block driver's internal I/O until next request arrives.
260 * Note that unlike bdrv_drain_all(), the caller must hold the BlockDriverState
263 * Only this BlockDriverState's AioContext is run, so in-flight requests must
264 * not depend on events in other AioContexts. In that case, use
265 * bdrv_drain_all() instead.
267 void coroutine_fn
bdrv_co_drain(BlockDriverState
*bs
)
269 assert(qemu_in_coroutine());
270 bdrv_drained_begin(bs
);
271 bdrv_drained_end(bs
);
274 void bdrv_drain(BlockDriverState
*bs
)
276 bdrv_drained_begin(bs
);
277 bdrv_drained_end(bs
);
281 * Wait for pending requests to complete across all BlockDriverStates
283 * This function does not flush data to disk, use bdrv_flush_all() for that
284 * after calling this function.
286 void bdrv_drain_all(void)
288 /* Always run first iteration so any pending completion BHs run */
290 BlockDriverState
*bs
;
292 BlockJob
*job
= NULL
;
293 GSList
*aio_ctxs
= NULL
, *ctx
;
295 while ((job
= block_job_next(job
))) {
296 AioContext
*aio_context
= blk_get_aio_context(job
->blk
);
298 aio_context_acquire(aio_context
);
299 block_job_pause(job
);
300 aio_context_release(aio_context
);
303 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
304 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
306 aio_context_acquire(aio_context
);
307 bdrv_parent_drained_begin(bs
);
308 bdrv_io_unplugged_begin(bs
);
309 bdrv_drain_recurse(bs
);
310 aio_context_release(aio_context
);
312 if (!g_slist_find(aio_ctxs
, aio_context
)) {
313 aio_ctxs
= g_slist_prepend(aio_ctxs
, aio_context
);
317 /* Note that completion of an asynchronous I/O operation can trigger any
318 * number of other I/O operations on other devices---for example a
319 * coroutine can submit an I/O request to another device in response to
320 * request completion. Therefore we must keep looping until there was no
321 * more activity rather than simply draining each device independently.
326 for (ctx
= aio_ctxs
; ctx
!= NULL
; ctx
= ctx
->next
) {
327 AioContext
*aio_context
= ctx
->data
;
329 aio_context_acquire(aio_context
);
330 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
331 if (aio_context
== bdrv_get_aio_context(bs
)) {
332 if (bdrv_requests_pending(bs
)) {
334 aio_poll(aio_context
, busy
);
338 busy
|= aio_poll(aio_context
, false);
339 aio_context_release(aio_context
);
343 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
344 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
346 aio_context_acquire(aio_context
);
347 bdrv_io_unplugged_end(bs
);
348 bdrv_parent_drained_end(bs
);
349 aio_context_release(aio_context
);
351 g_slist_free(aio_ctxs
);
354 while ((job
= block_job_next(job
))) {
355 AioContext
*aio_context
= blk_get_aio_context(job
->blk
);
357 aio_context_acquire(aio_context
);
358 block_job_resume(job
);
359 aio_context_release(aio_context
);
364 * Remove an active request from the tracked requests list
366 * This function should be called when a tracked request is completing.
368 static void tracked_request_end(BdrvTrackedRequest
*req
)
370 if (req
->serialising
) {
371 req
->bs
->serialising_in_flight
--;
374 QLIST_REMOVE(req
, list
);
375 qemu_co_queue_restart_all(&req
->wait_queue
);
379 * Add an active request to the tracked requests list
381 static void tracked_request_begin(BdrvTrackedRequest
*req
,
382 BlockDriverState
*bs
,
385 enum BdrvTrackedRequestType type
)
387 *req
= (BdrvTrackedRequest
){
392 .co
= qemu_coroutine_self(),
393 .serialising
= false,
394 .overlap_offset
= offset
,
395 .overlap_bytes
= bytes
,
398 qemu_co_queue_init(&req
->wait_queue
);
400 QLIST_INSERT_HEAD(&bs
->tracked_requests
, req
, list
);
403 static void mark_request_serialising(BdrvTrackedRequest
*req
, uint64_t align
)
405 int64_t overlap_offset
= req
->offset
& ~(align
- 1);
406 unsigned int overlap_bytes
= ROUND_UP(req
->offset
+ req
->bytes
, align
)
409 if (!req
->serialising
) {
410 req
->bs
->serialising_in_flight
++;
411 req
->serialising
= true;
414 req
->overlap_offset
= MIN(req
->overlap_offset
, overlap_offset
);
415 req
->overlap_bytes
= MAX(req
->overlap_bytes
, overlap_bytes
);
419 * Round a region to cluster boundaries (sector-based)
421 void bdrv_round_sectors_to_clusters(BlockDriverState
*bs
,
422 int64_t sector_num
, int nb_sectors
,
423 int64_t *cluster_sector_num
,
424 int *cluster_nb_sectors
)
428 if (bdrv_get_info(bs
, &bdi
) < 0 || bdi
.cluster_size
== 0) {
429 *cluster_sector_num
= sector_num
;
430 *cluster_nb_sectors
= nb_sectors
;
432 int64_t c
= bdi
.cluster_size
/ BDRV_SECTOR_SIZE
;
433 *cluster_sector_num
= QEMU_ALIGN_DOWN(sector_num
, c
);
434 *cluster_nb_sectors
= QEMU_ALIGN_UP(sector_num
- *cluster_sector_num
+
440 * Round a region to cluster boundaries
442 void bdrv_round_to_clusters(BlockDriverState
*bs
,
443 int64_t offset
, unsigned int bytes
,
444 int64_t *cluster_offset
,
445 unsigned int *cluster_bytes
)
449 if (bdrv_get_info(bs
, &bdi
) < 0 || bdi
.cluster_size
== 0) {
450 *cluster_offset
= offset
;
451 *cluster_bytes
= bytes
;
453 int64_t c
= bdi
.cluster_size
;
454 *cluster_offset
= QEMU_ALIGN_DOWN(offset
, c
);
455 *cluster_bytes
= QEMU_ALIGN_UP(offset
- *cluster_offset
+ bytes
, c
);
459 static int bdrv_get_cluster_size(BlockDriverState
*bs
)
464 ret
= bdrv_get_info(bs
, &bdi
);
465 if (ret
< 0 || bdi
.cluster_size
== 0) {
466 return bs
->request_alignment
;
468 return bdi
.cluster_size
;
472 static bool tracked_request_overlaps(BdrvTrackedRequest
*req
,
473 int64_t offset
, unsigned int bytes
)
476 if (offset
>= req
->overlap_offset
+ req
->overlap_bytes
) {
480 if (req
->overlap_offset
>= offset
+ bytes
) {
486 static bool coroutine_fn
wait_serialising_requests(BdrvTrackedRequest
*self
)
488 BlockDriverState
*bs
= self
->bs
;
489 BdrvTrackedRequest
*req
;
493 if (!bs
->serialising_in_flight
) {
499 QLIST_FOREACH(req
, &bs
->tracked_requests
, list
) {
500 if (req
== self
|| (!req
->serialising
&& !self
->serialising
)) {
503 if (tracked_request_overlaps(req
, self
->overlap_offset
,
504 self
->overlap_bytes
))
506 /* Hitting this means there was a reentrant request, for
507 * example, a block driver issuing nested requests. This must
508 * never happen since it means deadlock.
510 assert(qemu_coroutine_self() != req
->co
);
512 /* If the request is already (indirectly) waiting for us, or
513 * will wait for us as soon as it wakes up, then just go on
514 * (instead of producing a deadlock in the former case). */
515 if (!req
->waiting_for
) {
516 self
->waiting_for
= req
;
517 qemu_co_queue_wait(&req
->wait_queue
);
518 self
->waiting_for
= NULL
;
530 static int bdrv_check_byte_request(BlockDriverState
*bs
, int64_t offset
,
533 if (size
> BDRV_REQUEST_MAX_SECTORS
<< BDRV_SECTOR_BITS
) {
537 if (!bdrv_is_inserted(bs
)) {
548 static int bdrv_check_request(BlockDriverState
*bs
, int64_t sector_num
,
551 if (nb_sectors
< 0 || nb_sectors
> BDRV_REQUEST_MAX_SECTORS
) {
555 return bdrv_check_byte_request(bs
, sector_num
* BDRV_SECTOR_SIZE
,
556 nb_sectors
* BDRV_SECTOR_SIZE
);
559 typedef struct RwCo
{
560 BlockDriverState
*bs
;
565 BdrvRequestFlags flags
;
568 static void coroutine_fn
bdrv_rw_co_entry(void *opaque
)
572 if (!rwco
->is_write
) {
573 rwco
->ret
= bdrv_co_preadv(rwco
->bs
, rwco
->offset
,
574 rwco
->qiov
->size
, rwco
->qiov
,
577 rwco
->ret
= bdrv_co_pwritev(rwco
->bs
, rwco
->offset
,
578 rwco
->qiov
->size
, rwco
->qiov
,
584 * Process a vectored synchronous request using coroutines
586 static int bdrv_prwv_co(BlockDriverState
*bs
, int64_t offset
,
587 QEMUIOVector
*qiov
, bool is_write
,
588 BdrvRequestFlags flags
)
595 .is_write
= is_write
,
600 if (qemu_in_coroutine()) {
601 /* Fast-path if already in coroutine context */
602 bdrv_rw_co_entry(&rwco
);
604 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
606 co
= qemu_coroutine_create(bdrv_rw_co_entry
);
607 qemu_coroutine_enter(co
, &rwco
);
608 while (rwco
.ret
== NOT_DONE
) {
609 aio_poll(aio_context
, true);
616 * Process a synchronous request using coroutines
618 static int bdrv_rw_co(BlockDriverState
*bs
, int64_t sector_num
, uint8_t *buf
,
619 int nb_sectors
, bool is_write
, BdrvRequestFlags flags
)
623 .iov_base
= (void *)buf
,
624 .iov_len
= nb_sectors
* BDRV_SECTOR_SIZE
,
627 if (nb_sectors
< 0 || nb_sectors
> BDRV_REQUEST_MAX_SECTORS
) {
631 qemu_iovec_init_external(&qiov
, &iov
, 1);
632 return bdrv_prwv_co(bs
, sector_num
<< BDRV_SECTOR_BITS
,
633 &qiov
, is_write
, flags
);
636 /* return < 0 if error. See bdrv_write() for the return codes */
637 int bdrv_read(BlockDriverState
*bs
, int64_t sector_num
,
638 uint8_t *buf
, int nb_sectors
)
640 return bdrv_rw_co(bs
, sector_num
, buf
, nb_sectors
, false, 0);
643 /* Return < 0 if error. Important errors are:
644 -EIO generic I/O error (may happen for all errors)
645 -ENOMEDIUM No media inserted.
646 -EINVAL Invalid sector number or nb_sectors
647 -EACCES Trying to write a read-only device
649 int bdrv_write(BlockDriverState
*bs
, int64_t sector_num
,
650 const uint8_t *buf
, int nb_sectors
)
652 return bdrv_rw_co(bs
, sector_num
, (uint8_t *)buf
, nb_sectors
, true, 0);
655 int bdrv_pwrite_zeroes(BlockDriverState
*bs
, int64_t offset
,
656 int count
, BdrvRequestFlags flags
)
664 qemu_iovec_init_external(&qiov
, &iov
, 1);
665 return bdrv_prwv_co(bs
, offset
, &qiov
, true,
666 BDRV_REQ_ZERO_WRITE
| flags
);
670 * Completely zero out a block device with the help of bdrv_pwrite_zeroes.
671 * The operation is sped up by checking the block status and only writing
672 * zeroes to the device if they currently do not return zeroes. Optional
673 * flags are passed through to bdrv_pwrite_zeroes (e.g. BDRV_REQ_MAY_UNMAP,
676 * Returns < 0 on error, 0 on success. For error codes see bdrv_write().
678 int bdrv_make_zero(BlockDriverState
*bs
, BdrvRequestFlags flags
)
680 int64_t target_sectors
, ret
, nb_sectors
, sector_num
= 0;
681 BlockDriverState
*file
;
684 target_sectors
= bdrv_nb_sectors(bs
);
685 if (target_sectors
< 0) {
686 return target_sectors
;
690 nb_sectors
= MIN(target_sectors
- sector_num
, BDRV_REQUEST_MAX_SECTORS
);
691 if (nb_sectors
<= 0) {
694 ret
= bdrv_get_block_status(bs
, sector_num
, nb_sectors
, &n
, &file
);
696 error_report("error getting block status at sector %" PRId64
": %s",
697 sector_num
, strerror(-ret
));
700 if (ret
& BDRV_BLOCK_ZERO
) {
704 ret
= bdrv_pwrite_zeroes(bs
, sector_num
<< BDRV_SECTOR_BITS
,
705 n
<< BDRV_SECTOR_BITS
, flags
);
707 error_report("error writing zeroes at sector %" PRId64
": %s",
708 sector_num
, strerror(-ret
));
715 int bdrv_preadv(BlockDriverState
*bs
, int64_t offset
, QEMUIOVector
*qiov
)
719 ret
= bdrv_prwv_co(bs
, offset
, qiov
, false, 0);
727 int bdrv_pread(BlockDriverState
*bs
, int64_t offset
, void *buf
, int bytes
)
731 .iov_base
= (void *)buf
,
739 qemu_iovec_init_external(&qiov
, &iov
, 1);
740 return bdrv_preadv(bs
, offset
, &qiov
);
743 int bdrv_pwritev(BlockDriverState
*bs
, int64_t offset
, QEMUIOVector
*qiov
)
747 ret
= bdrv_prwv_co(bs
, offset
, qiov
, true, 0);
755 int bdrv_pwrite(BlockDriverState
*bs
, int64_t offset
,
756 const void *buf
, int bytes
)
760 .iov_base
= (void *) buf
,
768 qemu_iovec_init_external(&qiov
, &iov
, 1);
769 return bdrv_pwritev(bs
, offset
, &qiov
);
773 * Writes to the file and ensures that no writes are reordered across this
774 * request (acts as a barrier)
776 * Returns 0 on success, -errno in error cases.
778 int bdrv_pwrite_sync(BlockDriverState
*bs
, int64_t offset
,
779 const void *buf
, int count
)
783 ret
= bdrv_pwrite(bs
, offset
, buf
, count
);
788 ret
= bdrv_flush(bs
);
796 typedef struct CoroutineIOCompletion
{
797 Coroutine
*coroutine
;
799 } CoroutineIOCompletion
;
801 static void bdrv_co_io_em_complete(void *opaque
, int ret
)
803 CoroutineIOCompletion
*co
= opaque
;
806 qemu_coroutine_enter(co
->coroutine
, NULL
);
809 static int coroutine_fn
bdrv_driver_preadv(BlockDriverState
*bs
,
810 uint64_t offset
, uint64_t bytes
,
811 QEMUIOVector
*qiov
, int flags
)
813 BlockDriver
*drv
= bs
->drv
;
815 unsigned int nb_sectors
;
817 assert(!(flags
& ~BDRV_REQ_MASK
));
819 if (drv
->bdrv_co_preadv
) {
820 return drv
->bdrv_co_preadv(bs
, offset
, bytes
, qiov
, flags
);
823 sector_num
= offset
>> BDRV_SECTOR_BITS
;
824 nb_sectors
= bytes
>> BDRV_SECTOR_BITS
;
826 assert((offset
& (BDRV_SECTOR_SIZE
- 1)) == 0);
827 assert((bytes
& (BDRV_SECTOR_SIZE
- 1)) == 0);
828 assert((bytes
>> BDRV_SECTOR_BITS
) <= BDRV_REQUEST_MAX_SECTORS
);
830 if (drv
->bdrv_co_readv
) {
831 return drv
->bdrv_co_readv(bs
, sector_num
, nb_sectors
, qiov
);
834 CoroutineIOCompletion co
= {
835 .coroutine
= qemu_coroutine_self(),
838 acb
= bs
->drv
->bdrv_aio_readv(bs
, sector_num
, qiov
, nb_sectors
,
839 bdrv_co_io_em_complete
, &co
);
843 qemu_coroutine_yield();
849 static int coroutine_fn
bdrv_driver_pwritev(BlockDriverState
*bs
,
850 uint64_t offset
, uint64_t bytes
,
851 QEMUIOVector
*qiov
, int flags
)
853 BlockDriver
*drv
= bs
->drv
;
855 unsigned int nb_sectors
;
858 assert(!(flags
& ~BDRV_REQ_MASK
));
860 if (drv
->bdrv_co_pwritev
) {
861 ret
= drv
->bdrv_co_pwritev(bs
, offset
, bytes
, qiov
,
862 flags
& bs
->supported_write_flags
);
863 flags
&= ~bs
->supported_write_flags
;
867 sector_num
= offset
>> BDRV_SECTOR_BITS
;
868 nb_sectors
= bytes
>> BDRV_SECTOR_BITS
;
870 assert((offset
& (BDRV_SECTOR_SIZE
- 1)) == 0);
871 assert((bytes
& (BDRV_SECTOR_SIZE
- 1)) == 0);
872 assert((bytes
>> BDRV_SECTOR_BITS
) <= BDRV_REQUEST_MAX_SECTORS
);
874 if (drv
->bdrv_co_writev_flags
) {
875 ret
= drv
->bdrv_co_writev_flags(bs
, sector_num
, nb_sectors
, qiov
,
876 flags
& bs
->supported_write_flags
);
877 flags
&= ~bs
->supported_write_flags
;
878 } else if (drv
->bdrv_co_writev
) {
879 assert(!bs
->supported_write_flags
);
880 ret
= drv
->bdrv_co_writev(bs
, sector_num
, nb_sectors
, qiov
);
883 CoroutineIOCompletion co
= {
884 .coroutine
= qemu_coroutine_self(),
887 acb
= bs
->drv
->bdrv_aio_writev(bs
, sector_num
, qiov
, nb_sectors
,
888 bdrv_co_io_em_complete
, &co
);
892 qemu_coroutine_yield();
898 if (ret
== 0 && (flags
& BDRV_REQ_FUA
)) {
899 ret
= bdrv_co_flush(bs
);
905 static int coroutine_fn
bdrv_co_do_copy_on_readv(BlockDriverState
*bs
,
906 int64_t offset
, unsigned int bytes
, QEMUIOVector
*qiov
)
908 /* Perform I/O through a temporary buffer so that users who scribble over
909 * their read buffer while the operation is in progress do not end up
910 * modifying the image file. This is critical for zero-copy guest I/O
911 * where anything might happen inside guest memory.
915 BlockDriver
*drv
= bs
->drv
;
917 QEMUIOVector bounce_qiov
;
918 int64_t cluster_offset
;
919 unsigned int cluster_bytes
;
923 /* Cover entire cluster so no additional backing file I/O is required when
924 * allocating cluster in the image file.
926 bdrv_round_to_clusters(bs
, offset
, bytes
, &cluster_offset
, &cluster_bytes
);
928 trace_bdrv_co_do_copy_on_readv(bs
, offset
, bytes
,
929 cluster_offset
, cluster_bytes
);
931 iov
.iov_len
= cluster_bytes
;
932 iov
.iov_base
= bounce_buffer
= qemu_try_blockalign(bs
, iov
.iov_len
);
933 if (bounce_buffer
== NULL
) {
938 qemu_iovec_init_external(&bounce_qiov
, &iov
, 1);
940 ret
= bdrv_driver_preadv(bs
, cluster_offset
, cluster_bytes
,
946 if (drv
->bdrv_co_pwrite_zeroes
&&
947 buffer_is_zero(bounce_buffer
, iov
.iov_len
)) {
948 ret
= bdrv_co_do_pwrite_zeroes(bs
, cluster_offset
, cluster_bytes
, 0);
950 /* This does not change the data on the disk, it is not necessary
951 * to flush even in cache=writethrough mode.
953 ret
= bdrv_driver_pwritev(bs
, cluster_offset
, cluster_bytes
,
958 /* It might be okay to ignore write errors for guest requests. If this
959 * is a deliberate copy-on-read then we don't want to ignore the error.
960 * Simply report it in all cases.
965 skip_bytes
= offset
- cluster_offset
;
966 qemu_iovec_from_buf(qiov
, 0, bounce_buffer
+ skip_bytes
, bytes
);
969 qemu_vfree(bounce_buffer
);
974 * Forwards an already correctly aligned request to the BlockDriver. This
975 * handles copy on read and zeroing after EOF; any other features must be
976 * implemented by the caller.
978 static int coroutine_fn
bdrv_aligned_preadv(BlockDriverState
*bs
,
979 BdrvTrackedRequest
*req
, int64_t offset
, unsigned int bytes
,
980 int64_t align
, QEMUIOVector
*qiov
, int flags
)
982 int64_t total_bytes
, max_bytes
;
985 assert(is_power_of_2(align
));
986 assert((offset
& (align
- 1)) == 0);
987 assert((bytes
& (align
- 1)) == 0);
988 assert(!qiov
|| bytes
== qiov
->size
);
989 assert((bs
->open_flags
& BDRV_O_NO_IO
) == 0);
990 assert(!(flags
& ~BDRV_REQ_MASK
));
992 /* Handle Copy on Read and associated serialisation */
993 if (flags
& BDRV_REQ_COPY_ON_READ
) {
994 /* If we touch the same cluster it counts as an overlap. This
995 * guarantees that allocating writes will be serialized and not race
996 * with each other for the same cluster. For example, in copy-on-read
997 * it ensures that the CoR read and write operations are atomic and
998 * guest writes cannot interleave between them. */
999 mark_request_serialising(req
, bdrv_get_cluster_size(bs
));
1002 if (!(flags
& BDRV_REQ_NO_SERIALISING
)) {
1003 wait_serialising_requests(req
);
1006 if (flags
& BDRV_REQ_COPY_ON_READ
) {
1007 int64_t start_sector
= offset
>> BDRV_SECTOR_BITS
;
1008 int64_t end_sector
= DIV_ROUND_UP(offset
+ bytes
, BDRV_SECTOR_SIZE
);
1009 unsigned int nb_sectors
= end_sector
- start_sector
;
1012 ret
= bdrv_is_allocated(bs
, start_sector
, nb_sectors
, &pnum
);
1017 if (!ret
|| pnum
!= nb_sectors
) {
1018 ret
= bdrv_co_do_copy_on_readv(bs
, offset
, bytes
, qiov
);
1023 /* Forward the request to the BlockDriver */
1024 total_bytes
= bdrv_getlength(bs
);
1025 if (total_bytes
< 0) {
1030 max_bytes
= ROUND_UP(MAX(0, total_bytes
- offset
), align
);
1031 if (bytes
< max_bytes
) {
1032 ret
= bdrv_driver_preadv(bs
, offset
, bytes
, qiov
, 0);
1033 } else if (max_bytes
> 0) {
1034 QEMUIOVector local_qiov
;
1036 qemu_iovec_init(&local_qiov
, qiov
->niov
);
1037 qemu_iovec_concat(&local_qiov
, qiov
, 0, max_bytes
);
1039 ret
= bdrv_driver_preadv(bs
, offset
, max_bytes
, &local_qiov
, 0);
1041 qemu_iovec_destroy(&local_qiov
);
1046 /* Reading beyond end of file is supposed to produce zeroes */
1047 if (ret
== 0 && total_bytes
< offset
+ bytes
) {
1048 uint64_t zero_offset
= MAX(0, total_bytes
- offset
);
1049 uint64_t zero_bytes
= offset
+ bytes
- zero_offset
;
1050 qemu_iovec_memset(qiov
, zero_offset
, 0, zero_bytes
);
1058 * Handle a read request in coroutine context
1060 int coroutine_fn
bdrv_co_preadv(BlockDriverState
*bs
,
1061 int64_t offset
, unsigned int bytes
, QEMUIOVector
*qiov
,
1062 BdrvRequestFlags flags
)
1064 BlockDriver
*drv
= bs
->drv
;
1065 BdrvTrackedRequest req
;
1067 uint64_t align
= bs
->request_alignment
;
1068 uint8_t *head_buf
= NULL
;
1069 uint8_t *tail_buf
= NULL
;
1070 QEMUIOVector local_qiov
;
1071 bool use_local_qiov
= false;
1078 ret
= bdrv_check_byte_request(bs
, offset
, bytes
);
1083 /* Don't do copy-on-read if we read data before write operation */
1084 if (bs
->copy_on_read
&& !(flags
& BDRV_REQ_NO_SERIALISING
)) {
1085 flags
|= BDRV_REQ_COPY_ON_READ
;
1088 /* Align read if necessary by padding qiov */
1089 if (offset
& (align
- 1)) {
1090 head_buf
= qemu_blockalign(bs
, align
);
1091 qemu_iovec_init(&local_qiov
, qiov
->niov
+ 2);
1092 qemu_iovec_add(&local_qiov
, head_buf
, offset
& (align
- 1));
1093 qemu_iovec_concat(&local_qiov
, qiov
, 0, qiov
->size
);
1094 use_local_qiov
= true;
1096 bytes
+= offset
& (align
- 1);
1097 offset
= offset
& ~(align
- 1);
1100 if ((offset
+ bytes
) & (align
- 1)) {
1101 if (!use_local_qiov
) {
1102 qemu_iovec_init(&local_qiov
, qiov
->niov
+ 1);
1103 qemu_iovec_concat(&local_qiov
, qiov
, 0, qiov
->size
);
1104 use_local_qiov
= true;
1106 tail_buf
= qemu_blockalign(bs
, align
);
1107 qemu_iovec_add(&local_qiov
, tail_buf
,
1108 align
- ((offset
+ bytes
) & (align
- 1)));
1110 bytes
= ROUND_UP(bytes
, align
);
1113 tracked_request_begin(&req
, bs
, offset
, bytes
, BDRV_TRACKED_READ
);
1114 ret
= bdrv_aligned_preadv(bs
, &req
, offset
, bytes
, align
,
1115 use_local_qiov
? &local_qiov
: qiov
,
1117 tracked_request_end(&req
);
1119 if (use_local_qiov
) {
1120 qemu_iovec_destroy(&local_qiov
);
1121 qemu_vfree(head_buf
);
1122 qemu_vfree(tail_buf
);
1128 static int coroutine_fn
bdrv_co_do_readv(BlockDriverState
*bs
,
1129 int64_t sector_num
, int nb_sectors
, QEMUIOVector
*qiov
,
1130 BdrvRequestFlags flags
)
1132 if (nb_sectors
< 0 || nb_sectors
> BDRV_REQUEST_MAX_SECTORS
) {
1136 return bdrv_co_preadv(bs
, sector_num
<< BDRV_SECTOR_BITS
,
1137 nb_sectors
<< BDRV_SECTOR_BITS
, qiov
, flags
);
1140 int coroutine_fn
bdrv_co_readv(BlockDriverState
*bs
, int64_t sector_num
,
1141 int nb_sectors
, QEMUIOVector
*qiov
)
1143 trace_bdrv_co_readv(bs
, sector_num
, nb_sectors
);
1145 return bdrv_co_do_readv(bs
, sector_num
, nb_sectors
, qiov
, 0);
1148 #define MAX_WRITE_ZEROES_BOUNCE_BUFFER 32768
1150 static int coroutine_fn
bdrv_co_do_pwrite_zeroes(BlockDriverState
*bs
,
1151 int64_t offset
, int count
, BdrvRequestFlags flags
)
1153 BlockDriver
*drv
= bs
->drv
;
1155 struct iovec iov
= {0};
1157 bool need_flush
= false;
1161 int max_write_zeroes
= MIN_NON_ZERO(bs
->bl
.max_pwrite_zeroes
, INT_MAX
);
1162 int alignment
= MAX(bs
->bl
.pwrite_zeroes_alignment
?: 1,
1163 bs
->request_alignment
);
1165 assert(is_power_of_2(alignment
));
1166 head
= offset
& (alignment
- 1);
1167 tail
= (offset
+ count
) & (alignment
- 1);
1168 max_write_zeroes
&= ~(alignment
- 1);
1170 while (count
> 0 && !ret
) {
1173 /* Align request. Block drivers can expect the "bulk" of the request
1174 * to be aligned, and that unaligned requests do not cross cluster
1178 /* Make a small request up to the first aligned sector. */
1179 num
= MIN(count
, alignment
- head
);
1181 } else if (tail
&& num
> alignment
) {
1182 /* Shorten the request to the last aligned sector. */
1186 /* limit request size */
1187 if (num
> max_write_zeroes
) {
1188 num
= max_write_zeroes
;
1192 /* First try the efficient write zeroes operation */
1193 if (drv
->bdrv_co_pwrite_zeroes
) {
1194 ret
= drv
->bdrv_co_pwrite_zeroes(bs
, offset
, num
,
1195 flags
& bs
->supported_zero_flags
);
1196 if (ret
!= -ENOTSUP
&& (flags
& BDRV_REQ_FUA
) &&
1197 !(bs
->supported_zero_flags
& BDRV_REQ_FUA
)) {
1201 assert(!bs
->supported_zero_flags
);
1204 if (ret
== -ENOTSUP
) {
1205 /* Fall back to bounce buffer if write zeroes is unsupported */
1206 int max_xfer_len
= MIN_NON_ZERO(bs
->bl
.max_transfer_length
,
1207 MAX_WRITE_ZEROES_BOUNCE_BUFFER
);
1208 BdrvRequestFlags write_flags
= flags
& ~BDRV_REQ_ZERO_WRITE
;
1210 if ((flags
& BDRV_REQ_FUA
) &&
1211 !(bs
->supported_write_flags
& BDRV_REQ_FUA
)) {
1212 /* No need for bdrv_driver_pwrite() to do a fallback
1213 * flush on each chunk; use just one at the end */
1214 write_flags
&= ~BDRV_REQ_FUA
;
1217 num
= MIN(num
, max_xfer_len
<< BDRV_SECTOR_BITS
);
1219 if (iov
.iov_base
== NULL
) {
1220 iov
.iov_base
= qemu_try_blockalign(bs
, num
);
1221 if (iov
.iov_base
== NULL
) {
1225 memset(iov
.iov_base
, 0, num
);
1227 qemu_iovec_init_external(&qiov
, &iov
, 1);
1229 ret
= bdrv_driver_pwritev(bs
, offset
, num
, &qiov
, write_flags
);
1231 /* Keep bounce buffer around if it is big enough for all
1232 * all future requests.
1234 if (num
< max_xfer_len
<< BDRV_SECTOR_BITS
) {
1235 qemu_vfree(iov
.iov_base
);
1236 iov
.iov_base
= NULL
;
1245 if (ret
== 0 && need_flush
) {
1246 ret
= bdrv_co_flush(bs
);
1248 qemu_vfree(iov
.iov_base
);
1253 * Forwards an already correctly aligned write request to the BlockDriver.
1255 static int coroutine_fn
bdrv_aligned_pwritev(BlockDriverState
*bs
,
1256 BdrvTrackedRequest
*req
, int64_t offset
, unsigned int bytes
,
1257 int64_t align
, QEMUIOVector
*qiov
, int flags
)
1259 BlockDriver
*drv
= bs
->drv
;
1263 int64_t start_sector
= offset
>> BDRV_SECTOR_BITS
;
1264 int64_t end_sector
= DIV_ROUND_UP(offset
+ bytes
, BDRV_SECTOR_SIZE
);
1266 assert(is_power_of_2(align
));
1267 assert((offset
& (align
- 1)) == 0);
1268 assert((bytes
& (align
- 1)) == 0);
1269 assert(!qiov
|| bytes
== qiov
->size
);
1270 assert((bs
->open_flags
& BDRV_O_NO_IO
) == 0);
1271 assert(!(flags
& ~BDRV_REQ_MASK
));
1273 waited
= wait_serialising_requests(req
);
1274 assert(!waited
|| !req
->serialising
);
1275 assert(req
->overlap_offset
<= offset
);
1276 assert(offset
+ bytes
<= req
->overlap_offset
+ req
->overlap_bytes
);
1278 ret
= notifier_with_return_list_notify(&bs
->before_write_notifiers
, req
);
1280 if (!ret
&& bs
->detect_zeroes
!= BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF
&&
1281 !(flags
& BDRV_REQ_ZERO_WRITE
) && drv
->bdrv_co_pwrite_zeroes
&&
1282 qemu_iovec_is_zero(qiov
)) {
1283 flags
|= BDRV_REQ_ZERO_WRITE
;
1284 if (bs
->detect_zeroes
== BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP
) {
1285 flags
|= BDRV_REQ_MAY_UNMAP
;
1290 /* Do nothing, write notifier decided to fail this request */
1291 } else if (flags
& BDRV_REQ_ZERO_WRITE
) {
1292 bdrv_debug_event(bs
, BLKDBG_PWRITEV_ZERO
);
1293 ret
= bdrv_co_do_pwrite_zeroes(bs
, offset
, bytes
, flags
);
1295 bdrv_debug_event(bs
, BLKDBG_PWRITEV
);
1296 ret
= bdrv_driver_pwritev(bs
, offset
, bytes
, qiov
, flags
);
1298 bdrv_debug_event(bs
, BLKDBG_PWRITEV_DONE
);
1300 bdrv_set_dirty(bs
, start_sector
, end_sector
- start_sector
);
1302 if (bs
->wr_highest_offset
< offset
+ bytes
) {
1303 bs
->wr_highest_offset
= offset
+ bytes
;
1307 bs
->total_sectors
= MAX(bs
->total_sectors
, end_sector
);
1313 static int coroutine_fn
bdrv_co_do_zero_pwritev(BlockDriverState
*bs
,
1316 BdrvRequestFlags flags
,
1317 BdrvTrackedRequest
*req
)
1319 uint8_t *buf
= NULL
;
1320 QEMUIOVector local_qiov
;
1322 uint64_t align
= bs
->request_alignment
;
1323 unsigned int head_padding_bytes
, tail_padding_bytes
;
1326 head_padding_bytes
= offset
& (align
- 1);
1327 tail_padding_bytes
= align
- ((offset
+ bytes
) & (align
- 1));
1330 assert(flags
& BDRV_REQ_ZERO_WRITE
);
1331 if (head_padding_bytes
|| tail_padding_bytes
) {
1332 buf
= qemu_blockalign(bs
, align
);
1333 iov
= (struct iovec
) {
1337 qemu_iovec_init_external(&local_qiov
, &iov
, 1);
1339 if (head_padding_bytes
) {
1340 uint64_t zero_bytes
= MIN(bytes
, align
- head_padding_bytes
);
1342 /* RMW the unaligned part before head. */
1343 mark_request_serialising(req
, align
);
1344 wait_serialising_requests(req
);
1345 bdrv_debug_event(bs
, BLKDBG_PWRITEV_RMW_HEAD
);
1346 ret
= bdrv_aligned_preadv(bs
, req
, offset
& ~(align
- 1), align
,
1347 align
, &local_qiov
, 0);
1351 bdrv_debug_event(bs
, BLKDBG_PWRITEV_RMW_AFTER_HEAD
);
1353 memset(buf
+ head_padding_bytes
, 0, zero_bytes
);
1354 ret
= bdrv_aligned_pwritev(bs
, req
, offset
& ~(align
- 1), align
,
1356 flags
& ~BDRV_REQ_ZERO_WRITE
);
1360 offset
+= zero_bytes
;
1361 bytes
-= zero_bytes
;
1364 assert(!bytes
|| (offset
& (align
- 1)) == 0);
1365 if (bytes
>= align
) {
1366 /* Write the aligned part in the middle. */
1367 uint64_t aligned_bytes
= bytes
& ~(align
- 1);
1368 ret
= bdrv_aligned_pwritev(bs
, req
, offset
, aligned_bytes
, align
,
1373 bytes
-= aligned_bytes
;
1374 offset
+= aligned_bytes
;
1377 assert(!bytes
|| (offset
& (align
- 1)) == 0);
1379 assert(align
== tail_padding_bytes
+ bytes
);
1380 /* RMW the unaligned part after tail. */
1381 mark_request_serialising(req
, align
);
1382 wait_serialising_requests(req
);
1383 bdrv_debug_event(bs
, BLKDBG_PWRITEV_RMW_TAIL
);
1384 ret
= bdrv_aligned_preadv(bs
, req
, offset
, align
,
1385 align
, &local_qiov
, 0);
1389 bdrv_debug_event(bs
, BLKDBG_PWRITEV_RMW_AFTER_TAIL
);
1391 memset(buf
, 0, bytes
);
1392 ret
= bdrv_aligned_pwritev(bs
, req
, offset
, align
, align
,
1393 &local_qiov
, flags
& ~BDRV_REQ_ZERO_WRITE
);
1402 * Handle a write request in coroutine context
1404 int coroutine_fn
bdrv_co_pwritev(BlockDriverState
*bs
,
1405 int64_t offset
, unsigned int bytes
, QEMUIOVector
*qiov
,
1406 BdrvRequestFlags flags
)
1408 BdrvTrackedRequest req
;
1409 uint64_t align
= bs
->request_alignment
;
1410 uint8_t *head_buf
= NULL
;
1411 uint8_t *tail_buf
= NULL
;
1412 QEMUIOVector local_qiov
;
1413 bool use_local_qiov
= false;
1419 if (bs
->read_only
) {
1422 assert(!(bs
->open_flags
& BDRV_O_INACTIVE
));
1424 ret
= bdrv_check_byte_request(bs
, offset
, bytes
);
1430 * Align write if necessary by performing a read-modify-write cycle.
1431 * Pad qiov with the read parts and be sure to have a tracked request not
1432 * only for bdrv_aligned_pwritev, but also for the reads of the RMW cycle.
1434 tracked_request_begin(&req
, bs
, offset
, bytes
, BDRV_TRACKED_WRITE
);
1437 ret
= bdrv_co_do_zero_pwritev(bs
, offset
, bytes
, flags
, &req
);
1441 if (offset
& (align
- 1)) {
1442 QEMUIOVector head_qiov
;
1443 struct iovec head_iov
;
1445 mark_request_serialising(&req
, align
);
1446 wait_serialising_requests(&req
);
1448 head_buf
= qemu_blockalign(bs
, align
);
1449 head_iov
= (struct iovec
) {
1450 .iov_base
= head_buf
,
1453 qemu_iovec_init_external(&head_qiov
, &head_iov
, 1);
1455 bdrv_debug_event(bs
, BLKDBG_PWRITEV_RMW_HEAD
);
1456 ret
= bdrv_aligned_preadv(bs
, &req
, offset
& ~(align
- 1), align
,
1457 align
, &head_qiov
, 0);
1461 bdrv_debug_event(bs
, BLKDBG_PWRITEV_RMW_AFTER_HEAD
);
1463 qemu_iovec_init(&local_qiov
, qiov
->niov
+ 2);
1464 qemu_iovec_add(&local_qiov
, head_buf
, offset
& (align
- 1));
1465 qemu_iovec_concat(&local_qiov
, qiov
, 0, qiov
->size
);
1466 use_local_qiov
= true;
1468 bytes
+= offset
& (align
- 1);
1469 offset
= offset
& ~(align
- 1);
1471 /* We have read the tail already if the request is smaller
1472 * than one aligned block.
1474 if (bytes
< align
) {
1475 qemu_iovec_add(&local_qiov
, head_buf
+ bytes
, align
- bytes
);
1480 if ((offset
+ bytes
) & (align
- 1)) {
1481 QEMUIOVector tail_qiov
;
1482 struct iovec tail_iov
;
1486 mark_request_serialising(&req
, align
);
1487 waited
= wait_serialising_requests(&req
);
1488 assert(!waited
|| !use_local_qiov
);
1490 tail_buf
= qemu_blockalign(bs
, align
);
1491 tail_iov
= (struct iovec
) {
1492 .iov_base
= tail_buf
,
1495 qemu_iovec_init_external(&tail_qiov
, &tail_iov
, 1);
1497 bdrv_debug_event(bs
, BLKDBG_PWRITEV_RMW_TAIL
);
1498 ret
= bdrv_aligned_preadv(bs
, &req
, (offset
+ bytes
) & ~(align
- 1), align
,
1499 align
, &tail_qiov
, 0);
1503 bdrv_debug_event(bs
, BLKDBG_PWRITEV_RMW_AFTER_TAIL
);
1505 if (!use_local_qiov
) {
1506 qemu_iovec_init(&local_qiov
, qiov
->niov
+ 1);
1507 qemu_iovec_concat(&local_qiov
, qiov
, 0, qiov
->size
);
1508 use_local_qiov
= true;
1511 tail_bytes
= (offset
+ bytes
) & (align
- 1);
1512 qemu_iovec_add(&local_qiov
, tail_buf
+ tail_bytes
, align
- tail_bytes
);
1514 bytes
= ROUND_UP(bytes
, align
);
1517 ret
= bdrv_aligned_pwritev(bs
, &req
, offset
, bytes
, align
,
1518 use_local_qiov
? &local_qiov
: qiov
,
1523 if (use_local_qiov
) {
1524 qemu_iovec_destroy(&local_qiov
);
1526 qemu_vfree(head_buf
);
1527 qemu_vfree(tail_buf
);
1529 tracked_request_end(&req
);
1533 static int coroutine_fn
bdrv_co_do_writev(BlockDriverState
*bs
,
1534 int64_t sector_num
, int nb_sectors
, QEMUIOVector
*qiov
,
1535 BdrvRequestFlags flags
)
1537 if (nb_sectors
< 0 || nb_sectors
> BDRV_REQUEST_MAX_SECTORS
) {
1541 return bdrv_co_pwritev(bs
, sector_num
<< BDRV_SECTOR_BITS
,
1542 nb_sectors
<< BDRV_SECTOR_BITS
, qiov
, flags
);
1545 int coroutine_fn
bdrv_co_writev(BlockDriverState
*bs
, int64_t sector_num
,
1546 int nb_sectors
, QEMUIOVector
*qiov
)
1548 trace_bdrv_co_writev(bs
, sector_num
, nb_sectors
);
1550 return bdrv_co_do_writev(bs
, sector_num
, nb_sectors
, qiov
, 0);
1553 int coroutine_fn
bdrv_co_pwrite_zeroes(BlockDriverState
*bs
,
1554 int64_t offset
, int count
,
1555 BdrvRequestFlags flags
)
1557 trace_bdrv_co_pwrite_zeroes(bs
, offset
, count
, flags
);
1559 if (!(bs
->open_flags
& BDRV_O_UNMAP
)) {
1560 flags
&= ~BDRV_REQ_MAY_UNMAP
;
1563 return bdrv_co_pwritev(bs
, offset
, count
, NULL
,
1564 BDRV_REQ_ZERO_WRITE
| flags
);
1567 typedef struct BdrvCoGetBlockStatusData
{
1568 BlockDriverState
*bs
;
1569 BlockDriverState
*base
;
1570 BlockDriverState
**file
;
1576 } BdrvCoGetBlockStatusData
;
1579 * Returns the allocation status of the specified sectors.
1580 * Drivers not implementing the functionality are assumed to not support
1581 * backing files, hence all their sectors are reported as allocated.
1583 * If 'sector_num' is beyond the end of the disk image the return value is 0
1584 * and 'pnum' is set to 0.
1586 * 'pnum' is set to the number of sectors (including and immediately following
1587 * the specified sector) that are known to be in the same
1588 * allocated/unallocated state.
1590 * 'nb_sectors' is the max value 'pnum' should be set to. If nb_sectors goes
1591 * beyond the end of the disk image it will be clamped.
1593 * If returned value is positive and BDRV_BLOCK_OFFSET_VALID bit is set, 'file'
1594 * points to the BDS which the sector range is allocated in.
1596 static int64_t coroutine_fn
bdrv_co_get_block_status(BlockDriverState
*bs
,
1598 int nb_sectors
, int *pnum
,
1599 BlockDriverState
**file
)
1601 int64_t total_sectors
;
1605 total_sectors
= bdrv_nb_sectors(bs
);
1606 if (total_sectors
< 0) {
1607 return total_sectors
;
1610 if (sector_num
>= total_sectors
) {
1615 n
= total_sectors
- sector_num
;
1616 if (n
< nb_sectors
) {
1620 if (!bs
->drv
->bdrv_co_get_block_status
) {
1622 ret
= BDRV_BLOCK_DATA
| BDRV_BLOCK_ALLOCATED
;
1623 if (bs
->drv
->protocol_name
) {
1624 ret
|= BDRV_BLOCK_OFFSET_VALID
| (sector_num
* BDRV_SECTOR_SIZE
);
1630 ret
= bs
->drv
->bdrv_co_get_block_status(bs
, sector_num
, nb_sectors
, pnum
,
1637 if (ret
& BDRV_BLOCK_RAW
) {
1638 assert(ret
& BDRV_BLOCK_OFFSET_VALID
);
1639 return bdrv_get_block_status(bs
->file
->bs
, ret
>> BDRV_SECTOR_BITS
,
1643 if (ret
& (BDRV_BLOCK_DATA
| BDRV_BLOCK_ZERO
)) {
1644 ret
|= BDRV_BLOCK_ALLOCATED
;
1646 if (bdrv_unallocated_blocks_are_zero(bs
)) {
1647 ret
|= BDRV_BLOCK_ZERO
;
1648 } else if (bs
->backing
) {
1649 BlockDriverState
*bs2
= bs
->backing
->bs
;
1650 int64_t nb_sectors2
= bdrv_nb_sectors(bs2
);
1651 if (nb_sectors2
>= 0 && sector_num
>= nb_sectors2
) {
1652 ret
|= BDRV_BLOCK_ZERO
;
1657 if (*file
&& *file
!= bs
&&
1658 (ret
& BDRV_BLOCK_DATA
) && !(ret
& BDRV_BLOCK_ZERO
) &&
1659 (ret
& BDRV_BLOCK_OFFSET_VALID
)) {
1660 BlockDriverState
*file2
;
1663 ret2
= bdrv_co_get_block_status(*file
, ret
>> BDRV_SECTOR_BITS
,
1664 *pnum
, &file_pnum
, &file2
);
1666 /* Ignore errors. This is just providing extra information, it
1667 * is useful but not necessary.
1670 /* !file_pnum indicates an offset at or beyond the EOF; it is
1671 * perfectly valid for the format block driver to point to such
1672 * offsets, so catch it and mark everything as zero */
1673 ret
|= BDRV_BLOCK_ZERO
;
1675 /* Limit request to the range reported by the protocol driver */
1677 ret
|= (ret2
& BDRV_BLOCK_ZERO
);
1685 static int64_t coroutine_fn
bdrv_co_get_block_status_above(BlockDriverState
*bs
,
1686 BlockDriverState
*base
,
1690 BlockDriverState
**file
)
1692 BlockDriverState
*p
;
1696 for (p
= bs
; p
!= base
; p
= backing_bs(p
)) {
1697 ret
= bdrv_co_get_block_status(p
, sector_num
, nb_sectors
, pnum
, file
);
1698 if (ret
< 0 || ret
& BDRV_BLOCK_ALLOCATED
) {
1701 /* [sector_num, pnum] unallocated on this layer, which could be only
1702 * the first part of [sector_num, nb_sectors]. */
1703 nb_sectors
= MIN(nb_sectors
, *pnum
);
1708 /* Coroutine wrapper for bdrv_get_block_status_above() */
1709 static void coroutine_fn
bdrv_get_block_status_above_co_entry(void *opaque
)
1711 BdrvCoGetBlockStatusData
*data
= opaque
;
1713 data
->ret
= bdrv_co_get_block_status_above(data
->bs
, data
->base
,
1722 * Synchronous wrapper around bdrv_co_get_block_status_above().
1724 * See bdrv_co_get_block_status_above() for details.
1726 int64_t bdrv_get_block_status_above(BlockDriverState
*bs
,
1727 BlockDriverState
*base
,
1729 int nb_sectors
, int *pnum
,
1730 BlockDriverState
**file
)
1733 BdrvCoGetBlockStatusData data
= {
1737 .sector_num
= sector_num
,
1738 .nb_sectors
= nb_sectors
,
1743 if (qemu_in_coroutine()) {
1744 /* Fast-path if already in coroutine context */
1745 bdrv_get_block_status_above_co_entry(&data
);
1747 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
1749 co
= qemu_coroutine_create(bdrv_get_block_status_above_co_entry
);
1750 qemu_coroutine_enter(co
, &data
);
1751 while (!data
.done
) {
1752 aio_poll(aio_context
, true);
1758 int64_t bdrv_get_block_status(BlockDriverState
*bs
,
1760 int nb_sectors
, int *pnum
,
1761 BlockDriverState
**file
)
1763 return bdrv_get_block_status_above(bs
, backing_bs(bs
),
1764 sector_num
, nb_sectors
, pnum
, file
);
1767 int coroutine_fn
bdrv_is_allocated(BlockDriverState
*bs
, int64_t sector_num
,
1768 int nb_sectors
, int *pnum
)
1770 BlockDriverState
*file
;
1771 int64_t ret
= bdrv_get_block_status(bs
, sector_num
, nb_sectors
, pnum
,
1776 return !!(ret
& BDRV_BLOCK_ALLOCATED
);
1780 * Given an image chain: ... -> [BASE] -> [INTER1] -> [INTER2] -> [TOP]
1782 * Return true if the given sector is allocated in any image between
1783 * BASE and TOP (inclusive). BASE can be NULL to check if the given
1784 * sector is allocated in any image of the chain. Return false otherwise.
1786 * 'pnum' is set to the number of sectors (including and immediately following
1787 * the specified sector) that are known to be in the same
1788 * allocated/unallocated state.
1791 int bdrv_is_allocated_above(BlockDriverState
*top
,
1792 BlockDriverState
*base
,
1794 int nb_sectors
, int *pnum
)
1796 BlockDriverState
*intermediate
;
1797 int ret
, n
= nb_sectors
;
1800 while (intermediate
&& intermediate
!= base
) {
1802 ret
= bdrv_is_allocated(intermediate
, sector_num
, nb_sectors
,
1812 * [sector_num, nb_sectors] is unallocated on top but intermediate
1815 * [sector_num+x, nr_sectors] allocated.
1817 if (n
> pnum_inter
&&
1818 (intermediate
== top
||
1819 sector_num
+ pnum_inter
< intermediate
->total_sectors
)) {
1823 intermediate
= backing_bs(intermediate
);
1830 int bdrv_write_compressed(BlockDriverState
*bs
, int64_t sector_num
,
1831 const uint8_t *buf
, int nb_sectors
)
1833 BlockDriver
*drv
= bs
->drv
;
1839 if (!drv
->bdrv_write_compressed
) {
1842 ret
= bdrv_check_request(bs
, sector_num
, nb_sectors
);
1847 assert(QLIST_EMPTY(&bs
->dirty_bitmaps
));
1849 return drv
->bdrv_write_compressed(bs
, sector_num
, buf
, nb_sectors
);
1852 typedef struct BdrvVmstateCo
{
1853 BlockDriverState
*bs
;
1860 static int coroutine_fn
1861 bdrv_co_rw_vmstate(BlockDriverState
*bs
, QEMUIOVector
*qiov
, int64_t pos
,
1864 BlockDriver
*drv
= bs
->drv
;
1868 } else if (drv
->bdrv_load_vmstate
) {
1869 return is_read
? drv
->bdrv_load_vmstate(bs
, qiov
, pos
)
1870 : drv
->bdrv_save_vmstate(bs
, qiov
, pos
);
1871 } else if (bs
->file
) {
1872 return bdrv_co_rw_vmstate(bs
->file
->bs
, qiov
, pos
, is_read
);
1878 static void coroutine_fn
bdrv_co_rw_vmstate_entry(void *opaque
)
1880 BdrvVmstateCo
*co
= opaque
;
1881 co
->ret
= bdrv_co_rw_vmstate(co
->bs
, co
->qiov
, co
->pos
, co
->is_read
);
1885 bdrv_rw_vmstate(BlockDriverState
*bs
, QEMUIOVector
*qiov
, int64_t pos
,
1888 if (qemu_in_coroutine()) {
1889 return bdrv_co_rw_vmstate(bs
, qiov
, pos
, is_read
);
1891 BdrvVmstateCo data
= {
1896 .ret
= -EINPROGRESS
,
1898 Coroutine
*co
= qemu_coroutine_create(bdrv_co_rw_vmstate_entry
);
1900 qemu_coroutine_enter(co
, &data
);
1901 while (data
.ret
== -EINPROGRESS
) {
1902 aio_poll(bdrv_get_aio_context(bs
), true);
1908 int bdrv_save_vmstate(BlockDriverState
*bs
, const uint8_t *buf
,
1909 int64_t pos
, int size
)
1912 struct iovec iov
= {
1913 .iov_base
= (void *) buf
,
1918 qemu_iovec_init_external(&qiov
, &iov
, 1);
1920 ret
= bdrv_writev_vmstate(bs
, &qiov
, pos
);
1928 int bdrv_writev_vmstate(BlockDriverState
*bs
, QEMUIOVector
*qiov
, int64_t pos
)
1930 return bdrv_rw_vmstate(bs
, qiov
, pos
, false);
1933 int bdrv_load_vmstate(BlockDriverState
*bs
, uint8_t *buf
,
1934 int64_t pos
, int size
)
1937 struct iovec iov
= {
1943 qemu_iovec_init_external(&qiov
, &iov
, 1);
1944 ret
= bdrv_readv_vmstate(bs
, &qiov
, pos
);
1952 int bdrv_readv_vmstate(BlockDriverState
*bs
, QEMUIOVector
*qiov
, int64_t pos
)
1954 return bdrv_rw_vmstate(bs
, qiov
, pos
, true);
1957 /**************************************************************/
1960 BlockAIOCB
*bdrv_aio_readv(BlockDriverState
*bs
, int64_t sector_num
,
1961 QEMUIOVector
*qiov
, int nb_sectors
,
1962 BlockCompletionFunc
*cb
, void *opaque
)
1964 trace_bdrv_aio_readv(bs
, sector_num
, nb_sectors
, opaque
);
1966 return bdrv_co_aio_rw_vector(bs
, sector_num
, qiov
, nb_sectors
, 0,
1970 BlockAIOCB
*bdrv_aio_writev(BlockDriverState
*bs
, int64_t sector_num
,
1971 QEMUIOVector
*qiov
, int nb_sectors
,
1972 BlockCompletionFunc
*cb
, void *opaque
)
1974 trace_bdrv_aio_writev(bs
, sector_num
, nb_sectors
, opaque
);
1976 return bdrv_co_aio_rw_vector(bs
, sector_num
, qiov
, nb_sectors
, 0,
1980 void bdrv_aio_cancel(BlockAIOCB
*acb
)
1983 bdrv_aio_cancel_async(acb
);
1984 while (acb
->refcnt
> 1) {
1985 if (acb
->aiocb_info
->get_aio_context
) {
1986 aio_poll(acb
->aiocb_info
->get_aio_context(acb
), true);
1987 } else if (acb
->bs
) {
1988 aio_poll(bdrv_get_aio_context(acb
->bs
), true);
1993 qemu_aio_unref(acb
);
1996 /* Async version of aio cancel. The caller is not blocked if the acb implements
1997 * cancel_async, otherwise we do nothing and let the request normally complete.
1998 * In either case the completion callback must be called. */
1999 void bdrv_aio_cancel_async(BlockAIOCB
*acb
)
2001 if (acb
->aiocb_info
->cancel_async
) {
2002 acb
->aiocb_info
->cancel_async(acb
);
2006 /**************************************************************/
2007 /* async block device emulation */
2009 typedef struct BlockRequest
{
2011 /* Used during read, write, trim */
2018 /* Used during ioctl */
2024 BlockCompletionFunc
*cb
;
2030 typedef struct BlockAIOCBCoroutine
{
2037 } BlockAIOCBCoroutine
;
2039 static const AIOCBInfo bdrv_em_co_aiocb_info
= {
2040 .aiocb_size
= sizeof(BlockAIOCBCoroutine
),
2043 static void bdrv_co_complete(BlockAIOCBCoroutine
*acb
)
2045 if (!acb
->need_bh
) {
2046 acb
->common
.cb(acb
->common
.opaque
, acb
->req
.error
);
2047 qemu_aio_unref(acb
);
2051 static void bdrv_co_em_bh(void *opaque
)
2053 BlockAIOCBCoroutine
*acb
= opaque
;
2055 assert(!acb
->need_bh
);
2056 qemu_bh_delete(acb
->bh
);
2057 bdrv_co_complete(acb
);
2060 static void bdrv_co_maybe_schedule_bh(BlockAIOCBCoroutine
*acb
)
2062 acb
->need_bh
= false;
2063 if (acb
->req
.error
!= -EINPROGRESS
) {
2064 BlockDriverState
*bs
= acb
->common
.bs
;
2066 acb
->bh
= aio_bh_new(bdrv_get_aio_context(bs
), bdrv_co_em_bh
, acb
);
2067 qemu_bh_schedule(acb
->bh
);
2071 /* Invoke bdrv_co_do_readv/bdrv_co_do_writev */
2072 static void coroutine_fn
bdrv_co_do_rw(void *opaque
)
2074 BlockAIOCBCoroutine
*acb
= opaque
;
2075 BlockDriverState
*bs
= acb
->common
.bs
;
2077 if (!acb
->is_write
) {
2078 acb
->req
.error
= bdrv_co_do_readv(bs
, acb
->req
.sector
,
2079 acb
->req
.nb_sectors
, acb
->req
.qiov
, acb
->req
.flags
);
2081 acb
->req
.error
= bdrv_co_do_writev(bs
, acb
->req
.sector
,
2082 acb
->req
.nb_sectors
, acb
->req
.qiov
, acb
->req
.flags
);
2085 bdrv_co_complete(acb
);
2088 static BlockAIOCB
*bdrv_co_aio_rw_vector(BlockDriverState
*bs
,
2092 BdrvRequestFlags flags
,
2093 BlockCompletionFunc
*cb
,
2098 BlockAIOCBCoroutine
*acb
;
2100 acb
= qemu_aio_get(&bdrv_em_co_aiocb_info
, bs
, cb
, opaque
);
2101 acb
->need_bh
= true;
2102 acb
->req
.error
= -EINPROGRESS
;
2103 acb
->req
.sector
= sector_num
;
2104 acb
->req
.nb_sectors
= nb_sectors
;
2105 acb
->req
.qiov
= qiov
;
2106 acb
->req
.flags
= flags
;
2107 acb
->is_write
= is_write
;
2109 co
= qemu_coroutine_create(bdrv_co_do_rw
);
2110 qemu_coroutine_enter(co
, acb
);
2112 bdrv_co_maybe_schedule_bh(acb
);
2113 return &acb
->common
;
2116 static void coroutine_fn
bdrv_aio_flush_co_entry(void *opaque
)
2118 BlockAIOCBCoroutine
*acb
= opaque
;
2119 BlockDriverState
*bs
= acb
->common
.bs
;
2121 acb
->req
.error
= bdrv_co_flush(bs
);
2122 bdrv_co_complete(acb
);
2125 BlockAIOCB
*bdrv_aio_flush(BlockDriverState
*bs
,
2126 BlockCompletionFunc
*cb
, void *opaque
)
2128 trace_bdrv_aio_flush(bs
, opaque
);
2131 BlockAIOCBCoroutine
*acb
;
2133 acb
= qemu_aio_get(&bdrv_em_co_aiocb_info
, bs
, cb
, opaque
);
2134 acb
->need_bh
= true;
2135 acb
->req
.error
= -EINPROGRESS
;
2137 co
= qemu_coroutine_create(bdrv_aio_flush_co_entry
);
2138 qemu_coroutine_enter(co
, acb
);
2140 bdrv_co_maybe_schedule_bh(acb
);
2141 return &acb
->common
;
2144 static void coroutine_fn
bdrv_aio_discard_co_entry(void *opaque
)
2146 BlockAIOCBCoroutine
*acb
= opaque
;
2147 BlockDriverState
*bs
= acb
->common
.bs
;
2149 acb
->req
.error
= bdrv_co_discard(bs
, acb
->req
.sector
, acb
->req
.nb_sectors
);
2150 bdrv_co_complete(acb
);
2153 BlockAIOCB
*bdrv_aio_discard(BlockDriverState
*bs
,
2154 int64_t sector_num
, int nb_sectors
,
2155 BlockCompletionFunc
*cb
, void *opaque
)
2158 BlockAIOCBCoroutine
*acb
;
2160 trace_bdrv_aio_discard(bs
, sector_num
, nb_sectors
, opaque
);
2162 acb
= qemu_aio_get(&bdrv_em_co_aiocb_info
, bs
, cb
, opaque
);
2163 acb
->need_bh
= true;
2164 acb
->req
.error
= -EINPROGRESS
;
2165 acb
->req
.sector
= sector_num
;
2166 acb
->req
.nb_sectors
= nb_sectors
;
2167 co
= qemu_coroutine_create(bdrv_aio_discard_co_entry
);
2168 qemu_coroutine_enter(co
, acb
);
2170 bdrv_co_maybe_schedule_bh(acb
);
2171 return &acb
->common
;
2174 void *qemu_aio_get(const AIOCBInfo
*aiocb_info
, BlockDriverState
*bs
,
2175 BlockCompletionFunc
*cb
, void *opaque
)
2179 acb
= g_malloc(aiocb_info
->aiocb_size
);
2180 acb
->aiocb_info
= aiocb_info
;
2183 acb
->opaque
= opaque
;
2188 void qemu_aio_ref(void *p
)
2190 BlockAIOCB
*acb
= p
;
2194 void qemu_aio_unref(void *p
)
2196 BlockAIOCB
*acb
= p
;
2197 assert(acb
->refcnt
> 0);
2198 if (--acb
->refcnt
== 0) {
2203 /**************************************************************/
2204 /* Coroutine block device emulation */
2206 static void coroutine_fn
bdrv_flush_co_entry(void *opaque
)
2208 RwCo
*rwco
= opaque
;
2210 rwco
->ret
= bdrv_co_flush(rwco
->bs
);
2213 int coroutine_fn
bdrv_co_flush(BlockDriverState
*bs
)
2216 BdrvTrackedRequest req
;
2218 if (!bs
|| !bdrv_is_inserted(bs
) || bdrv_is_read_only(bs
) ||
2223 tracked_request_begin(&req
, bs
, 0, 0, BDRV_TRACKED_FLUSH
);
2225 /* Write back all layers by calling one driver function */
2226 if (bs
->drv
->bdrv_co_flush
) {
2227 ret
= bs
->drv
->bdrv_co_flush(bs
);
2231 /* Write back cached data to the OS even with cache=unsafe */
2232 BLKDBG_EVENT(bs
->file
, BLKDBG_FLUSH_TO_OS
);
2233 if (bs
->drv
->bdrv_co_flush_to_os
) {
2234 ret
= bs
->drv
->bdrv_co_flush_to_os(bs
);
2240 /* But don't actually force it to the disk with cache=unsafe */
2241 if (bs
->open_flags
& BDRV_O_NO_FLUSH
) {
2245 BLKDBG_EVENT(bs
->file
, BLKDBG_FLUSH_TO_DISK
);
2246 if (bs
->drv
->bdrv_co_flush_to_disk
) {
2247 ret
= bs
->drv
->bdrv_co_flush_to_disk(bs
);
2248 } else if (bs
->drv
->bdrv_aio_flush
) {
2250 CoroutineIOCompletion co
= {
2251 .coroutine
= qemu_coroutine_self(),
2254 acb
= bs
->drv
->bdrv_aio_flush(bs
, bdrv_co_io_em_complete
, &co
);
2258 qemu_coroutine_yield();
2263 * Some block drivers always operate in either writethrough or unsafe
2264 * mode and don't support bdrv_flush therefore. Usually qemu doesn't
2265 * know how the server works (because the behaviour is hardcoded or
2266 * depends on server-side configuration), so we can't ensure that
2267 * everything is safe on disk. Returning an error doesn't work because
2268 * that would break guests even if the server operates in writethrough
2271 * Let's hope the user knows what he's doing.
2279 /* Now flush the underlying protocol. It will also have BDRV_O_NO_FLUSH
2280 * in the case of cache=unsafe, so there are no useless flushes.
2283 ret
= bs
->file
? bdrv_co_flush(bs
->file
->bs
) : 0;
2285 tracked_request_end(&req
);
2289 int bdrv_flush(BlockDriverState
*bs
)
2297 if (qemu_in_coroutine()) {
2298 /* Fast-path if already in coroutine context */
2299 bdrv_flush_co_entry(&rwco
);
2301 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
2303 co
= qemu_coroutine_create(bdrv_flush_co_entry
);
2304 qemu_coroutine_enter(co
, &rwco
);
2305 while (rwco
.ret
== NOT_DONE
) {
2306 aio_poll(aio_context
, true);
2313 typedef struct DiscardCo
{
2314 BlockDriverState
*bs
;
2319 static void coroutine_fn
bdrv_discard_co_entry(void *opaque
)
2321 DiscardCo
*rwco
= opaque
;
2323 rwco
->ret
= bdrv_co_discard(rwco
->bs
, rwco
->sector_num
, rwco
->nb_sectors
);
2326 int coroutine_fn
bdrv_co_discard(BlockDriverState
*bs
, int64_t sector_num
,
2329 BdrvTrackedRequest req
;
2330 int max_discard
, ret
;
2336 ret
= bdrv_check_request(bs
, sector_num
, nb_sectors
);
2339 } else if (bs
->read_only
) {
2342 assert(!(bs
->open_flags
& BDRV_O_INACTIVE
));
2344 /* Do nothing if disabled. */
2345 if (!(bs
->open_flags
& BDRV_O_UNMAP
)) {
2349 if (!bs
->drv
->bdrv_co_discard
&& !bs
->drv
->bdrv_aio_discard
) {
2353 tracked_request_begin(&req
, bs
, sector_num
<< BDRV_SECTOR_BITS
,
2354 nb_sectors
<< BDRV_SECTOR_BITS
, BDRV_TRACKED_DISCARD
);
2356 ret
= notifier_with_return_list_notify(&bs
->before_write_notifiers
, &req
);
2361 max_discard
= MIN_NON_ZERO(bs
->bl
.max_discard
, BDRV_REQUEST_MAX_SECTORS
);
2362 while (nb_sectors
> 0) {
2364 int num
= nb_sectors
;
2367 if (bs
->bl
.discard_alignment
&&
2368 num
>= bs
->bl
.discard_alignment
&&
2369 sector_num
% bs
->bl
.discard_alignment
) {
2370 if (num
> bs
->bl
.discard_alignment
) {
2371 num
= bs
->bl
.discard_alignment
;
2373 num
-= sector_num
% bs
->bl
.discard_alignment
;
2376 /* limit request size */
2377 if (num
> max_discard
) {
2381 if (bs
->drv
->bdrv_co_discard
) {
2382 ret
= bs
->drv
->bdrv_co_discard(bs
, sector_num
, num
);
2385 CoroutineIOCompletion co
= {
2386 .coroutine
= qemu_coroutine_self(),
2389 acb
= bs
->drv
->bdrv_aio_discard(bs
, sector_num
, nb_sectors
,
2390 bdrv_co_io_em_complete
, &co
);
2395 qemu_coroutine_yield();
2399 if (ret
&& ret
!= -ENOTSUP
) {
2408 bdrv_set_dirty(bs
, req
.offset
>> BDRV_SECTOR_BITS
,
2409 req
.bytes
>> BDRV_SECTOR_BITS
);
2410 tracked_request_end(&req
);
2414 int bdrv_discard(BlockDriverState
*bs
, int64_t sector_num
, int nb_sectors
)
2419 .sector_num
= sector_num
,
2420 .nb_sectors
= nb_sectors
,
2424 if (qemu_in_coroutine()) {
2425 /* Fast-path if already in coroutine context */
2426 bdrv_discard_co_entry(&rwco
);
2428 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
2430 co
= qemu_coroutine_create(bdrv_discard_co_entry
);
2431 qemu_coroutine_enter(co
, &rwco
);
2432 while (rwco
.ret
== NOT_DONE
) {
2433 aio_poll(aio_context
, true);
2440 static int bdrv_co_do_ioctl(BlockDriverState
*bs
, int req
, void *buf
)
2442 BlockDriver
*drv
= bs
->drv
;
2443 BdrvTrackedRequest tracked_req
;
2444 CoroutineIOCompletion co
= {
2445 .coroutine
= qemu_coroutine_self(),
2449 tracked_request_begin(&tracked_req
, bs
, 0, 0, BDRV_TRACKED_IOCTL
);
2450 if (!drv
|| !drv
->bdrv_aio_ioctl
) {
2455 acb
= drv
->bdrv_aio_ioctl(bs
, req
, buf
, bdrv_co_io_em_complete
, &co
);
2460 qemu_coroutine_yield();
2462 tracked_request_end(&tracked_req
);
2467 BlockDriverState
*bs
;
2473 static void coroutine_fn
bdrv_co_ioctl_entry(void *opaque
)
2475 BdrvIoctlCoData
*data
= opaque
;
2476 data
->ret
= bdrv_co_do_ioctl(data
->bs
, data
->req
, data
->buf
);
2479 /* needed for generic scsi interface */
2480 int bdrv_ioctl(BlockDriverState
*bs
, unsigned long int req
, void *buf
)
2482 BdrvIoctlCoData data
= {
2486 .ret
= -EINPROGRESS
,
2489 if (qemu_in_coroutine()) {
2490 /* Fast-path if already in coroutine context */
2491 bdrv_co_ioctl_entry(&data
);
2493 Coroutine
*co
= qemu_coroutine_create(bdrv_co_ioctl_entry
);
2495 qemu_coroutine_enter(co
, &data
);
2496 while (data
.ret
== -EINPROGRESS
) {
2497 aio_poll(bdrv_get_aio_context(bs
), true);
2503 static void coroutine_fn
bdrv_co_aio_ioctl_entry(void *opaque
)
2505 BlockAIOCBCoroutine
*acb
= opaque
;
2506 acb
->req
.error
= bdrv_co_do_ioctl(acb
->common
.bs
,
2507 acb
->req
.req
, acb
->req
.buf
);
2508 bdrv_co_complete(acb
);
2511 BlockAIOCB
*bdrv_aio_ioctl(BlockDriverState
*bs
,
2512 unsigned long int req
, void *buf
,
2513 BlockCompletionFunc
*cb
, void *opaque
)
2515 BlockAIOCBCoroutine
*acb
= qemu_aio_get(&bdrv_em_co_aiocb_info
,
2519 acb
->need_bh
= true;
2520 acb
->req
.error
= -EINPROGRESS
;
2523 co
= qemu_coroutine_create(bdrv_co_aio_ioctl_entry
);
2524 qemu_coroutine_enter(co
, acb
);
2526 bdrv_co_maybe_schedule_bh(acb
);
2527 return &acb
->common
;
2530 void *qemu_blockalign(BlockDriverState
*bs
, size_t size
)
2532 return qemu_memalign(bdrv_opt_mem_align(bs
), size
);
2535 void *qemu_blockalign0(BlockDriverState
*bs
, size_t size
)
2537 return memset(qemu_blockalign(bs
, size
), 0, size
);
2540 void *qemu_try_blockalign(BlockDriverState
*bs
, size_t size
)
2542 size_t align
= bdrv_opt_mem_align(bs
);
2544 /* Ensure that NULL is never returned on success */
2550 return qemu_try_memalign(align
, size
);
2553 void *qemu_try_blockalign0(BlockDriverState
*bs
, size_t size
)
2555 void *mem
= qemu_try_blockalign(bs
, size
);
2558 memset(mem
, 0, size
);
2565 * Check if all memory in this vector is sector aligned.
2567 bool bdrv_qiov_is_aligned(BlockDriverState
*bs
, QEMUIOVector
*qiov
)
2570 size_t alignment
= bdrv_min_mem_align(bs
);
2572 for (i
= 0; i
< qiov
->niov
; i
++) {
2573 if ((uintptr_t) qiov
->iov
[i
].iov_base
% alignment
) {
2576 if (qiov
->iov
[i
].iov_len
% alignment
) {
2584 void bdrv_add_before_write_notifier(BlockDriverState
*bs
,
2585 NotifierWithReturn
*notifier
)
2587 notifier_with_return_list_add(&bs
->before_write_notifiers
, notifier
);
2590 void bdrv_io_plug(BlockDriverState
*bs
)
2594 QLIST_FOREACH(child
, &bs
->children
, next
) {
2595 bdrv_io_plug(child
->bs
);
2598 if (bs
->io_plugged
++ == 0 && bs
->io_plug_disabled
== 0) {
2599 BlockDriver
*drv
= bs
->drv
;
2600 if (drv
&& drv
->bdrv_io_plug
) {
2601 drv
->bdrv_io_plug(bs
);
2606 void bdrv_io_unplug(BlockDriverState
*bs
)
2610 assert(bs
->io_plugged
);
2611 if (--bs
->io_plugged
== 0 && bs
->io_plug_disabled
== 0) {
2612 BlockDriver
*drv
= bs
->drv
;
2613 if (drv
&& drv
->bdrv_io_unplug
) {
2614 drv
->bdrv_io_unplug(bs
);
2618 QLIST_FOREACH(child
, &bs
->children
, next
) {
2619 bdrv_io_unplug(child
->bs
);
2623 void bdrv_io_unplugged_begin(BlockDriverState
*bs
)
2627 if (bs
->io_plug_disabled
++ == 0 && bs
->io_plugged
> 0) {
2628 BlockDriver
*drv
= bs
->drv
;
2629 if (drv
&& drv
->bdrv_io_unplug
) {
2630 drv
->bdrv_io_unplug(bs
);
2634 QLIST_FOREACH(child
, &bs
->children
, next
) {
2635 bdrv_io_unplugged_begin(child
->bs
);
2639 void bdrv_io_unplugged_end(BlockDriverState
*bs
)
2643 assert(bs
->io_plug_disabled
);
2644 QLIST_FOREACH(child
, &bs
->children
, next
) {
2645 bdrv_io_unplugged_end(child
->bs
);
2648 if (--bs
->io_plug_disabled
== 0 && bs
->io_plugged
> 0) {
2649 BlockDriver
*drv
= bs
->drv
;
2650 if (drv
&& drv
->bdrv_io_plug
) {
2651 drv
->bdrv_io_plug(bs
);