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 GSList
*aio_ctxs
= NULL
, *ctx
;
294 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
295 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
297 aio_context_acquire(aio_context
);
299 block_job_pause(bs
->job
);
301 bdrv_parent_drained_begin(bs
);
302 bdrv_io_unplugged_begin(bs
);
303 bdrv_drain_recurse(bs
);
304 aio_context_release(aio_context
);
306 if (!g_slist_find(aio_ctxs
, aio_context
)) {
307 aio_ctxs
= g_slist_prepend(aio_ctxs
, aio_context
);
311 /* Note that completion of an asynchronous I/O operation can trigger any
312 * number of other I/O operations on other devices---for example a
313 * coroutine can submit an I/O request to another device in response to
314 * request completion. Therefore we must keep looping until there was no
315 * more activity rather than simply draining each device independently.
320 for (ctx
= aio_ctxs
; ctx
!= NULL
; ctx
= ctx
->next
) {
321 AioContext
*aio_context
= ctx
->data
;
323 aio_context_acquire(aio_context
);
324 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
325 if (aio_context
== bdrv_get_aio_context(bs
)) {
326 if (bdrv_requests_pending(bs
)) {
328 aio_poll(aio_context
, busy
);
332 busy
|= aio_poll(aio_context
, false);
333 aio_context_release(aio_context
);
337 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
338 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
340 aio_context_acquire(aio_context
);
341 bdrv_io_unplugged_end(bs
);
342 bdrv_parent_drained_end(bs
);
344 block_job_resume(bs
->job
);
346 aio_context_release(aio_context
);
348 g_slist_free(aio_ctxs
);
352 * Remove an active request from the tracked requests list
354 * This function should be called when a tracked request is completing.
356 static void tracked_request_end(BdrvTrackedRequest
*req
)
358 if (req
->serialising
) {
359 req
->bs
->serialising_in_flight
--;
362 QLIST_REMOVE(req
, list
);
363 qemu_co_queue_restart_all(&req
->wait_queue
);
367 * Add an active request to the tracked requests list
369 static void tracked_request_begin(BdrvTrackedRequest
*req
,
370 BlockDriverState
*bs
,
373 enum BdrvTrackedRequestType type
)
375 *req
= (BdrvTrackedRequest
){
380 .co
= qemu_coroutine_self(),
381 .serialising
= false,
382 .overlap_offset
= offset
,
383 .overlap_bytes
= bytes
,
386 qemu_co_queue_init(&req
->wait_queue
);
388 QLIST_INSERT_HEAD(&bs
->tracked_requests
, req
, list
);
391 static void mark_request_serialising(BdrvTrackedRequest
*req
, uint64_t align
)
393 int64_t overlap_offset
= req
->offset
& ~(align
- 1);
394 unsigned int overlap_bytes
= ROUND_UP(req
->offset
+ req
->bytes
, align
)
397 if (!req
->serialising
) {
398 req
->bs
->serialising_in_flight
++;
399 req
->serialising
= true;
402 req
->overlap_offset
= MIN(req
->overlap_offset
, overlap_offset
);
403 req
->overlap_bytes
= MAX(req
->overlap_bytes
, overlap_bytes
);
407 * Round a region to cluster boundaries (sector-based)
409 void bdrv_round_sectors_to_clusters(BlockDriverState
*bs
,
410 int64_t sector_num
, int nb_sectors
,
411 int64_t *cluster_sector_num
,
412 int *cluster_nb_sectors
)
416 if (bdrv_get_info(bs
, &bdi
) < 0 || bdi
.cluster_size
== 0) {
417 *cluster_sector_num
= sector_num
;
418 *cluster_nb_sectors
= nb_sectors
;
420 int64_t c
= bdi
.cluster_size
/ BDRV_SECTOR_SIZE
;
421 *cluster_sector_num
= QEMU_ALIGN_DOWN(sector_num
, c
);
422 *cluster_nb_sectors
= QEMU_ALIGN_UP(sector_num
- *cluster_sector_num
+
428 * Round a region to cluster boundaries
430 void bdrv_round_to_clusters(BlockDriverState
*bs
,
431 int64_t offset
, unsigned int bytes
,
432 int64_t *cluster_offset
,
433 unsigned int *cluster_bytes
)
437 if (bdrv_get_info(bs
, &bdi
) < 0 || bdi
.cluster_size
== 0) {
438 *cluster_offset
= offset
;
439 *cluster_bytes
= bytes
;
441 int64_t c
= bdi
.cluster_size
;
442 *cluster_offset
= QEMU_ALIGN_DOWN(offset
, c
);
443 *cluster_bytes
= QEMU_ALIGN_UP(offset
- *cluster_offset
+ bytes
, c
);
447 static int bdrv_get_cluster_size(BlockDriverState
*bs
)
452 ret
= bdrv_get_info(bs
, &bdi
);
453 if (ret
< 0 || bdi
.cluster_size
== 0) {
454 return bs
->request_alignment
;
456 return bdi
.cluster_size
;
460 static bool tracked_request_overlaps(BdrvTrackedRequest
*req
,
461 int64_t offset
, unsigned int bytes
)
464 if (offset
>= req
->overlap_offset
+ req
->overlap_bytes
) {
468 if (req
->overlap_offset
>= offset
+ bytes
) {
474 static bool coroutine_fn
wait_serialising_requests(BdrvTrackedRequest
*self
)
476 BlockDriverState
*bs
= self
->bs
;
477 BdrvTrackedRequest
*req
;
481 if (!bs
->serialising_in_flight
) {
487 QLIST_FOREACH(req
, &bs
->tracked_requests
, list
) {
488 if (req
== self
|| (!req
->serialising
&& !self
->serialising
)) {
491 if (tracked_request_overlaps(req
, self
->overlap_offset
,
492 self
->overlap_bytes
))
494 /* Hitting this means there was a reentrant request, for
495 * example, a block driver issuing nested requests. This must
496 * never happen since it means deadlock.
498 assert(qemu_coroutine_self() != req
->co
);
500 /* If the request is already (indirectly) waiting for us, or
501 * will wait for us as soon as it wakes up, then just go on
502 * (instead of producing a deadlock in the former case). */
503 if (!req
->waiting_for
) {
504 self
->waiting_for
= req
;
505 qemu_co_queue_wait(&req
->wait_queue
);
506 self
->waiting_for
= NULL
;
518 static int bdrv_check_byte_request(BlockDriverState
*bs
, int64_t offset
,
521 if (size
> BDRV_REQUEST_MAX_SECTORS
<< BDRV_SECTOR_BITS
) {
525 if (!bdrv_is_inserted(bs
)) {
536 static int bdrv_check_request(BlockDriverState
*bs
, int64_t sector_num
,
539 if (nb_sectors
< 0 || nb_sectors
> BDRV_REQUEST_MAX_SECTORS
) {
543 return bdrv_check_byte_request(bs
, sector_num
* BDRV_SECTOR_SIZE
,
544 nb_sectors
* BDRV_SECTOR_SIZE
);
547 typedef struct RwCo
{
548 BlockDriverState
*bs
;
553 BdrvRequestFlags flags
;
556 static void coroutine_fn
bdrv_rw_co_entry(void *opaque
)
560 if (!rwco
->is_write
) {
561 rwco
->ret
= bdrv_co_preadv(rwco
->bs
, rwco
->offset
,
562 rwco
->qiov
->size
, rwco
->qiov
,
565 rwco
->ret
= bdrv_co_pwritev(rwco
->bs
, rwco
->offset
,
566 rwco
->qiov
->size
, rwco
->qiov
,
572 * Process a vectored synchronous request using coroutines
574 static int bdrv_prwv_co(BlockDriverState
*bs
, int64_t offset
,
575 QEMUIOVector
*qiov
, bool is_write
,
576 BdrvRequestFlags flags
)
583 .is_write
= is_write
,
588 if (qemu_in_coroutine()) {
589 /* Fast-path if already in coroutine context */
590 bdrv_rw_co_entry(&rwco
);
592 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
594 co
= qemu_coroutine_create(bdrv_rw_co_entry
);
595 qemu_coroutine_enter(co
, &rwco
);
596 while (rwco
.ret
== NOT_DONE
) {
597 aio_poll(aio_context
, true);
604 * Process a synchronous request using coroutines
606 static int bdrv_rw_co(BlockDriverState
*bs
, int64_t sector_num
, uint8_t *buf
,
607 int nb_sectors
, bool is_write
, BdrvRequestFlags flags
)
611 .iov_base
= (void *)buf
,
612 .iov_len
= nb_sectors
* BDRV_SECTOR_SIZE
,
615 if (nb_sectors
< 0 || nb_sectors
> BDRV_REQUEST_MAX_SECTORS
) {
619 qemu_iovec_init_external(&qiov
, &iov
, 1);
620 return bdrv_prwv_co(bs
, sector_num
<< BDRV_SECTOR_BITS
,
621 &qiov
, is_write
, flags
);
624 /* return < 0 if error. See bdrv_write() for the return codes */
625 int bdrv_read(BlockDriverState
*bs
, int64_t sector_num
,
626 uint8_t *buf
, int nb_sectors
)
628 return bdrv_rw_co(bs
, sector_num
, buf
, nb_sectors
, false, 0);
631 /* Return < 0 if error. Important errors are:
632 -EIO generic I/O error (may happen for all errors)
633 -ENOMEDIUM No media inserted.
634 -EINVAL Invalid sector number or nb_sectors
635 -EACCES Trying to write a read-only device
637 int bdrv_write(BlockDriverState
*bs
, int64_t sector_num
,
638 const uint8_t *buf
, int nb_sectors
)
640 return bdrv_rw_co(bs
, sector_num
, (uint8_t *)buf
, nb_sectors
, true, 0);
643 int bdrv_pwrite_zeroes(BlockDriverState
*bs
, int64_t offset
,
644 int count
, BdrvRequestFlags flags
)
652 qemu_iovec_init_external(&qiov
, &iov
, 1);
653 return bdrv_prwv_co(bs
, offset
, &qiov
, true,
654 BDRV_REQ_ZERO_WRITE
| flags
);
658 * Completely zero out a block device with the help of bdrv_pwrite_zeroes.
659 * The operation is sped up by checking the block status and only writing
660 * zeroes to the device if they currently do not return zeroes. Optional
661 * flags are passed through to bdrv_pwrite_zeroes (e.g. BDRV_REQ_MAY_UNMAP,
664 * Returns < 0 on error, 0 on success. For error codes see bdrv_write().
666 int bdrv_make_zero(BlockDriverState
*bs
, BdrvRequestFlags flags
)
668 int64_t target_sectors
, ret
, nb_sectors
, sector_num
= 0;
669 BlockDriverState
*file
;
672 target_sectors
= bdrv_nb_sectors(bs
);
673 if (target_sectors
< 0) {
674 return target_sectors
;
678 nb_sectors
= MIN(target_sectors
- sector_num
, BDRV_REQUEST_MAX_SECTORS
);
679 if (nb_sectors
<= 0) {
682 ret
= bdrv_get_block_status(bs
, sector_num
, nb_sectors
, &n
, &file
);
684 error_report("error getting block status at sector %" PRId64
": %s",
685 sector_num
, strerror(-ret
));
688 if (ret
& BDRV_BLOCK_ZERO
) {
692 ret
= bdrv_pwrite_zeroes(bs
, sector_num
<< BDRV_SECTOR_BITS
,
693 n
<< BDRV_SECTOR_BITS
, flags
);
695 error_report("error writing zeroes at sector %" PRId64
": %s",
696 sector_num
, strerror(-ret
));
703 int bdrv_pread(BlockDriverState
*bs
, int64_t offset
, void *buf
, int bytes
)
707 .iov_base
= (void *)buf
,
716 qemu_iovec_init_external(&qiov
, &iov
, 1);
717 ret
= bdrv_prwv_co(bs
, offset
, &qiov
, false, 0);
725 int bdrv_pwritev(BlockDriverState
*bs
, int64_t offset
, QEMUIOVector
*qiov
)
729 ret
= bdrv_prwv_co(bs
, offset
, qiov
, true, 0);
737 int bdrv_pwrite(BlockDriverState
*bs
, int64_t offset
,
738 const void *buf
, int bytes
)
742 .iov_base
= (void *) buf
,
750 qemu_iovec_init_external(&qiov
, &iov
, 1);
751 return bdrv_pwritev(bs
, offset
, &qiov
);
755 * Writes to the file and ensures that no writes are reordered across this
756 * request (acts as a barrier)
758 * Returns 0 on success, -errno in error cases.
760 int bdrv_pwrite_sync(BlockDriverState
*bs
, int64_t offset
,
761 const void *buf
, int count
)
765 ret
= bdrv_pwrite(bs
, offset
, buf
, count
);
770 ret
= bdrv_flush(bs
);
778 typedef struct CoroutineIOCompletion
{
779 Coroutine
*coroutine
;
781 } CoroutineIOCompletion
;
783 static void bdrv_co_io_em_complete(void *opaque
, int ret
)
785 CoroutineIOCompletion
*co
= opaque
;
788 qemu_coroutine_enter(co
->coroutine
, NULL
);
791 static int coroutine_fn
bdrv_driver_preadv(BlockDriverState
*bs
,
792 uint64_t offset
, uint64_t bytes
,
793 QEMUIOVector
*qiov
, int flags
)
795 BlockDriver
*drv
= bs
->drv
;
797 unsigned int nb_sectors
;
799 assert(!(flags
& ~BDRV_REQ_MASK
));
801 if (drv
->bdrv_co_preadv
) {
802 return drv
->bdrv_co_preadv(bs
, offset
, bytes
, qiov
, flags
);
805 sector_num
= offset
>> BDRV_SECTOR_BITS
;
806 nb_sectors
= bytes
>> BDRV_SECTOR_BITS
;
808 assert((offset
& (BDRV_SECTOR_SIZE
- 1)) == 0);
809 assert((bytes
& (BDRV_SECTOR_SIZE
- 1)) == 0);
810 assert((bytes
>> BDRV_SECTOR_BITS
) <= BDRV_REQUEST_MAX_SECTORS
);
812 if (drv
->bdrv_co_readv
) {
813 return drv
->bdrv_co_readv(bs
, sector_num
, nb_sectors
, qiov
);
816 CoroutineIOCompletion co
= {
817 .coroutine
= qemu_coroutine_self(),
820 acb
= bs
->drv
->bdrv_aio_readv(bs
, sector_num
, qiov
, nb_sectors
,
821 bdrv_co_io_em_complete
, &co
);
825 qemu_coroutine_yield();
831 static int coroutine_fn
bdrv_driver_pwritev(BlockDriverState
*bs
,
832 uint64_t offset
, uint64_t bytes
,
833 QEMUIOVector
*qiov
, int flags
)
835 BlockDriver
*drv
= bs
->drv
;
837 unsigned int nb_sectors
;
840 assert(!(flags
& ~BDRV_REQ_MASK
));
842 if (drv
->bdrv_co_pwritev
) {
843 ret
= drv
->bdrv_co_pwritev(bs
, offset
, bytes
, qiov
,
844 flags
& bs
->supported_write_flags
);
845 flags
&= ~bs
->supported_write_flags
;
849 sector_num
= offset
>> BDRV_SECTOR_BITS
;
850 nb_sectors
= bytes
>> BDRV_SECTOR_BITS
;
852 assert((offset
& (BDRV_SECTOR_SIZE
- 1)) == 0);
853 assert((bytes
& (BDRV_SECTOR_SIZE
- 1)) == 0);
854 assert((bytes
>> BDRV_SECTOR_BITS
) <= BDRV_REQUEST_MAX_SECTORS
);
856 if (drv
->bdrv_co_writev_flags
) {
857 ret
= drv
->bdrv_co_writev_flags(bs
, sector_num
, nb_sectors
, qiov
,
858 flags
& bs
->supported_write_flags
);
859 flags
&= ~bs
->supported_write_flags
;
860 } else if (drv
->bdrv_co_writev
) {
861 assert(!bs
->supported_write_flags
);
862 ret
= drv
->bdrv_co_writev(bs
, sector_num
, nb_sectors
, qiov
);
865 CoroutineIOCompletion co
= {
866 .coroutine
= qemu_coroutine_self(),
869 acb
= bs
->drv
->bdrv_aio_writev(bs
, sector_num
, qiov
, nb_sectors
,
870 bdrv_co_io_em_complete
, &co
);
874 qemu_coroutine_yield();
880 if (ret
== 0 && (flags
& BDRV_REQ_FUA
)) {
881 ret
= bdrv_co_flush(bs
);
887 static int coroutine_fn
bdrv_co_do_copy_on_readv(BlockDriverState
*bs
,
888 int64_t offset
, unsigned int bytes
, QEMUIOVector
*qiov
)
890 /* Perform I/O through a temporary buffer so that users who scribble over
891 * their read buffer while the operation is in progress do not end up
892 * modifying the image file. This is critical for zero-copy guest I/O
893 * where anything might happen inside guest memory.
897 BlockDriver
*drv
= bs
->drv
;
899 QEMUIOVector bounce_qiov
;
900 int64_t cluster_offset
;
901 unsigned int cluster_bytes
;
905 /* Cover entire cluster so no additional backing file I/O is required when
906 * allocating cluster in the image file.
908 bdrv_round_to_clusters(bs
, offset
, bytes
, &cluster_offset
, &cluster_bytes
);
910 trace_bdrv_co_do_copy_on_readv(bs
, offset
, bytes
,
911 cluster_offset
, cluster_bytes
);
913 iov
.iov_len
= cluster_bytes
;
914 iov
.iov_base
= bounce_buffer
= qemu_try_blockalign(bs
, iov
.iov_len
);
915 if (bounce_buffer
== NULL
) {
920 qemu_iovec_init_external(&bounce_qiov
, &iov
, 1);
922 ret
= bdrv_driver_preadv(bs
, cluster_offset
, cluster_bytes
,
928 if (drv
->bdrv_co_pwrite_zeroes
&&
929 buffer_is_zero(bounce_buffer
, iov
.iov_len
)) {
930 ret
= bdrv_co_do_pwrite_zeroes(bs
, cluster_offset
, cluster_bytes
, 0);
932 /* This does not change the data on the disk, it is not necessary
933 * to flush even in cache=writethrough mode.
935 ret
= bdrv_driver_pwritev(bs
, cluster_offset
, cluster_bytes
,
940 /* It might be okay to ignore write errors for guest requests. If this
941 * is a deliberate copy-on-read then we don't want to ignore the error.
942 * Simply report it in all cases.
947 skip_bytes
= offset
- cluster_offset
;
948 qemu_iovec_from_buf(qiov
, 0, bounce_buffer
+ skip_bytes
, bytes
);
951 qemu_vfree(bounce_buffer
);
956 * Forwards an already correctly aligned request to the BlockDriver. This
957 * handles copy on read and zeroing after EOF; any other features must be
958 * implemented by the caller.
960 static int coroutine_fn
bdrv_aligned_preadv(BlockDriverState
*bs
,
961 BdrvTrackedRequest
*req
, int64_t offset
, unsigned int bytes
,
962 int64_t align
, QEMUIOVector
*qiov
, int flags
)
966 int64_t sector_num
= offset
>> BDRV_SECTOR_BITS
;
967 unsigned int nb_sectors
= bytes
>> BDRV_SECTOR_BITS
;
969 assert((offset
& (BDRV_SECTOR_SIZE
- 1)) == 0);
970 assert((bytes
& (BDRV_SECTOR_SIZE
- 1)) == 0);
971 assert(!qiov
|| bytes
== qiov
->size
);
972 assert((bs
->open_flags
& BDRV_O_NO_IO
) == 0);
973 assert(!(flags
& ~BDRV_REQ_MASK
));
975 /* Handle Copy on Read and associated serialisation */
976 if (flags
& BDRV_REQ_COPY_ON_READ
) {
977 /* If we touch the same cluster it counts as an overlap. This
978 * guarantees that allocating writes will be serialized and not race
979 * with each other for the same cluster. For example, in copy-on-read
980 * it ensures that the CoR read and write operations are atomic and
981 * guest writes cannot interleave between them. */
982 mark_request_serialising(req
, bdrv_get_cluster_size(bs
));
985 if (!(flags
& BDRV_REQ_NO_SERIALISING
)) {
986 wait_serialising_requests(req
);
989 if (flags
& BDRV_REQ_COPY_ON_READ
) {
992 ret
= bdrv_is_allocated(bs
, sector_num
, nb_sectors
, &pnum
);
997 if (!ret
|| pnum
!= nb_sectors
) {
998 ret
= bdrv_co_do_copy_on_readv(bs
, offset
, bytes
, qiov
);
1003 /* Forward the request to the BlockDriver */
1004 if (!bs
->zero_beyond_eof
) {
1005 ret
= bdrv_driver_preadv(bs
, offset
, bytes
, qiov
, 0);
1007 /* Read zeros after EOF */
1008 int64_t total_sectors
, max_nb_sectors
;
1010 total_sectors
= bdrv_nb_sectors(bs
);
1011 if (total_sectors
< 0) {
1012 ret
= total_sectors
;
1016 max_nb_sectors
= ROUND_UP(MAX(0, total_sectors
- sector_num
),
1017 align
>> BDRV_SECTOR_BITS
);
1018 if (nb_sectors
< max_nb_sectors
) {
1019 ret
= bdrv_driver_preadv(bs
, offset
, bytes
, qiov
, 0);
1020 } else if (max_nb_sectors
> 0) {
1021 QEMUIOVector local_qiov
;
1023 qemu_iovec_init(&local_qiov
, qiov
->niov
);
1024 qemu_iovec_concat(&local_qiov
, qiov
, 0,
1025 max_nb_sectors
* BDRV_SECTOR_SIZE
);
1027 ret
= bdrv_driver_preadv(bs
, offset
,
1028 max_nb_sectors
* BDRV_SECTOR_SIZE
,
1031 qemu_iovec_destroy(&local_qiov
);
1036 /* Reading beyond end of file is supposed to produce zeroes */
1037 if (ret
== 0 && total_sectors
< sector_num
+ nb_sectors
) {
1038 uint64_t offset
= MAX(0, total_sectors
- sector_num
);
1039 uint64_t bytes
= (sector_num
+ nb_sectors
- offset
) *
1041 qemu_iovec_memset(qiov
, offset
* BDRV_SECTOR_SIZE
, 0, bytes
);
1050 * Handle a read request in coroutine context
1052 int coroutine_fn
bdrv_co_preadv(BlockDriverState
*bs
,
1053 int64_t offset
, unsigned int bytes
, QEMUIOVector
*qiov
,
1054 BdrvRequestFlags flags
)
1056 BlockDriver
*drv
= bs
->drv
;
1057 BdrvTrackedRequest req
;
1059 /* TODO Lift BDRV_SECTOR_SIZE restriction in BlockDriver interface */
1060 uint64_t align
= MAX(BDRV_SECTOR_SIZE
, bs
->request_alignment
);
1061 uint8_t *head_buf
= NULL
;
1062 uint8_t *tail_buf
= NULL
;
1063 QEMUIOVector local_qiov
;
1064 bool use_local_qiov
= false;
1071 ret
= bdrv_check_byte_request(bs
, offset
, bytes
);
1076 /* Don't do copy-on-read if we read data before write operation */
1077 if (bs
->copy_on_read
&& !(flags
& BDRV_REQ_NO_SERIALISING
)) {
1078 flags
|= BDRV_REQ_COPY_ON_READ
;
1081 /* Align read if necessary by padding qiov */
1082 if (offset
& (align
- 1)) {
1083 head_buf
= qemu_blockalign(bs
, align
);
1084 qemu_iovec_init(&local_qiov
, qiov
->niov
+ 2);
1085 qemu_iovec_add(&local_qiov
, head_buf
, offset
& (align
- 1));
1086 qemu_iovec_concat(&local_qiov
, qiov
, 0, qiov
->size
);
1087 use_local_qiov
= true;
1089 bytes
+= offset
& (align
- 1);
1090 offset
= offset
& ~(align
- 1);
1093 if ((offset
+ bytes
) & (align
- 1)) {
1094 if (!use_local_qiov
) {
1095 qemu_iovec_init(&local_qiov
, qiov
->niov
+ 1);
1096 qemu_iovec_concat(&local_qiov
, qiov
, 0, qiov
->size
);
1097 use_local_qiov
= true;
1099 tail_buf
= qemu_blockalign(bs
, align
);
1100 qemu_iovec_add(&local_qiov
, tail_buf
,
1101 align
- ((offset
+ bytes
) & (align
- 1)));
1103 bytes
= ROUND_UP(bytes
, align
);
1106 tracked_request_begin(&req
, bs
, offset
, bytes
, BDRV_TRACKED_READ
);
1107 ret
= bdrv_aligned_preadv(bs
, &req
, offset
, bytes
, align
,
1108 use_local_qiov
? &local_qiov
: qiov
,
1110 tracked_request_end(&req
);
1112 if (use_local_qiov
) {
1113 qemu_iovec_destroy(&local_qiov
);
1114 qemu_vfree(head_buf
);
1115 qemu_vfree(tail_buf
);
1121 static int coroutine_fn
bdrv_co_do_readv(BlockDriverState
*bs
,
1122 int64_t sector_num
, int nb_sectors
, QEMUIOVector
*qiov
,
1123 BdrvRequestFlags flags
)
1125 if (nb_sectors
< 0 || nb_sectors
> BDRV_REQUEST_MAX_SECTORS
) {
1129 return bdrv_co_preadv(bs
, sector_num
<< BDRV_SECTOR_BITS
,
1130 nb_sectors
<< BDRV_SECTOR_BITS
, qiov
, flags
);
1133 int coroutine_fn
bdrv_co_readv(BlockDriverState
*bs
, int64_t sector_num
,
1134 int nb_sectors
, QEMUIOVector
*qiov
)
1136 trace_bdrv_co_readv(bs
, sector_num
, nb_sectors
);
1138 return bdrv_co_do_readv(bs
, sector_num
, nb_sectors
, qiov
, 0);
1141 #define MAX_WRITE_ZEROES_BOUNCE_BUFFER 32768
1143 static int coroutine_fn
bdrv_co_do_pwrite_zeroes(BlockDriverState
*bs
,
1144 int64_t offset
, int count
, BdrvRequestFlags flags
)
1146 BlockDriver
*drv
= bs
->drv
;
1148 struct iovec iov
= {0};
1150 bool need_flush
= false;
1154 int max_write_zeroes
= MIN_NON_ZERO(bs
->bl
.max_pwrite_zeroes
, INT_MAX
);
1155 int alignment
= MAX(bs
->bl
.pwrite_zeroes_alignment
?: 1,
1156 bs
->request_alignment
);
1158 assert(is_power_of_2(alignment
));
1159 head
= offset
& (alignment
- 1);
1160 tail
= (offset
+ count
) & (alignment
- 1);
1161 max_write_zeroes
&= ~(alignment
- 1);
1163 while (count
> 0 && !ret
) {
1166 /* Align request. Block drivers can expect the "bulk" of the request
1167 * to be aligned, and that unaligned requests do not cross cluster
1171 /* Make a small request up to the first aligned sector. */
1172 num
= MIN(count
, alignment
- head
);
1174 } else if (tail
&& num
> alignment
) {
1175 /* Shorten the request to the last aligned sector. */
1179 /* limit request size */
1180 if (num
> max_write_zeroes
) {
1181 num
= max_write_zeroes
;
1185 /* First try the efficient write zeroes operation */
1186 if (drv
->bdrv_co_pwrite_zeroes
) {
1187 ret
= drv
->bdrv_co_pwrite_zeroes(bs
, offset
, num
,
1188 flags
& bs
->supported_zero_flags
);
1189 if (ret
!= -ENOTSUP
&& (flags
& BDRV_REQ_FUA
) &&
1190 !(bs
->supported_zero_flags
& BDRV_REQ_FUA
)) {
1194 assert(!bs
->supported_zero_flags
);
1197 if (ret
== -ENOTSUP
) {
1198 /* Fall back to bounce buffer if write zeroes is unsupported */
1199 int max_xfer_len
= MIN_NON_ZERO(bs
->bl
.max_transfer_length
,
1200 MAX_WRITE_ZEROES_BOUNCE_BUFFER
);
1201 BdrvRequestFlags write_flags
= flags
& ~BDRV_REQ_ZERO_WRITE
;
1203 if ((flags
& BDRV_REQ_FUA
) &&
1204 !(bs
->supported_write_flags
& BDRV_REQ_FUA
)) {
1205 /* No need for bdrv_driver_pwrite() to do a fallback
1206 * flush on each chunk; use just one at the end */
1207 write_flags
&= ~BDRV_REQ_FUA
;
1210 num
= MIN(num
, max_xfer_len
<< BDRV_SECTOR_BITS
);
1212 if (iov
.iov_base
== NULL
) {
1213 iov
.iov_base
= qemu_try_blockalign(bs
, num
);
1214 if (iov
.iov_base
== NULL
) {
1218 memset(iov
.iov_base
, 0, num
);
1220 qemu_iovec_init_external(&qiov
, &iov
, 1);
1222 ret
= bdrv_driver_pwritev(bs
, offset
, num
, &qiov
, write_flags
);
1224 /* Keep bounce buffer around if it is big enough for all
1225 * all future requests.
1227 if (num
< max_xfer_len
<< BDRV_SECTOR_BITS
) {
1228 qemu_vfree(iov
.iov_base
);
1229 iov
.iov_base
= NULL
;
1238 if (ret
== 0 && need_flush
) {
1239 ret
= bdrv_co_flush(bs
);
1241 qemu_vfree(iov
.iov_base
);
1246 * Forwards an already correctly aligned write request to the BlockDriver.
1248 static int coroutine_fn
bdrv_aligned_pwritev(BlockDriverState
*bs
,
1249 BdrvTrackedRequest
*req
, int64_t offset
, unsigned int bytes
,
1250 QEMUIOVector
*qiov
, int flags
)
1252 BlockDriver
*drv
= bs
->drv
;
1256 int64_t sector_num
= offset
>> BDRV_SECTOR_BITS
;
1257 unsigned int nb_sectors
= bytes
>> BDRV_SECTOR_BITS
;
1259 assert((offset
& (BDRV_SECTOR_SIZE
- 1)) == 0);
1260 assert((bytes
& (BDRV_SECTOR_SIZE
- 1)) == 0);
1261 assert(!qiov
|| bytes
== qiov
->size
);
1262 assert((bs
->open_flags
& BDRV_O_NO_IO
) == 0);
1263 assert(!(flags
& ~BDRV_REQ_MASK
));
1265 waited
= wait_serialising_requests(req
);
1266 assert(!waited
|| !req
->serialising
);
1267 assert(req
->overlap_offset
<= offset
);
1268 assert(offset
+ bytes
<= req
->overlap_offset
+ req
->overlap_bytes
);
1270 ret
= notifier_with_return_list_notify(&bs
->before_write_notifiers
, req
);
1272 if (!ret
&& bs
->detect_zeroes
!= BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF
&&
1273 !(flags
& BDRV_REQ_ZERO_WRITE
) && drv
->bdrv_co_pwrite_zeroes
&&
1274 qemu_iovec_is_zero(qiov
)) {
1275 flags
|= BDRV_REQ_ZERO_WRITE
;
1276 if (bs
->detect_zeroes
== BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP
) {
1277 flags
|= BDRV_REQ_MAY_UNMAP
;
1282 /* Do nothing, write notifier decided to fail this request */
1283 } else if (flags
& BDRV_REQ_ZERO_WRITE
) {
1284 bdrv_debug_event(bs
, BLKDBG_PWRITEV_ZERO
);
1285 ret
= bdrv_co_do_pwrite_zeroes(bs
, sector_num
<< BDRV_SECTOR_BITS
,
1286 nb_sectors
<< BDRV_SECTOR_BITS
, flags
);
1288 bdrv_debug_event(bs
, BLKDBG_PWRITEV
);
1289 ret
= bdrv_driver_pwritev(bs
, offset
, bytes
, qiov
, flags
);
1291 bdrv_debug_event(bs
, BLKDBG_PWRITEV_DONE
);
1293 bdrv_set_dirty(bs
, sector_num
, nb_sectors
);
1295 if (bs
->wr_highest_offset
< offset
+ bytes
) {
1296 bs
->wr_highest_offset
= offset
+ bytes
;
1300 bs
->total_sectors
= MAX(bs
->total_sectors
, sector_num
+ nb_sectors
);
1306 static int coroutine_fn
bdrv_co_do_zero_pwritev(BlockDriverState
*bs
,
1309 BdrvRequestFlags flags
,
1310 BdrvTrackedRequest
*req
)
1312 uint8_t *buf
= NULL
;
1313 QEMUIOVector local_qiov
;
1315 uint64_t align
= MAX(BDRV_SECTOR_SIZE
, bs
->request_alignment
);
1316 unsigned int head_padding_bytes
, tail_padding_bytes
;
1319 head_padding_bytes
= offset
& (align
- 1);
1320 tail_padding_bytes
= align
- ((offset
+ bytes
) & (align
- 1));
1323 assert(flags
& BDRV_REQ_ZERO_WRITE
);
1324 if (head_padding_bytes
|| tail_padding_bytes
) {
1325 buf
= qemu_blockalign(bs
, align
);
1326 iov
= (struct iovec
) {
1330 qemu_iovec_init_external(&local_qiov
, &iov
, 1);
1332 if (head_padding_bytes
) {
1333 uint64_t zero_bytes
= MIN(bytes
, align
- head_padding_bytes
);
1335 /* RMW the unaligned part before head. */
1336 mark_request_serialising(req
, align
);
1337 wait_serialising_requests(req
);
1338 bdrv_debug_event(bs
, BLKDBG_PWRITEV_RMW_HEAD
);
1339 ret
= bdrv_aligned_preadv(bs
, req
, offset
& ~(align
- 1), align
,
1340 align
, &local_qiov
, 0);
1344 bdrv_debug_event(bs
, BLKDBG_PWRITEV_RMW_AFTER_HEAD
);
1346 memset(buf
+ head_padding_bytes
, 0, zero_bytes
);
1347 ret
= bdrv_aligned_pwritev(bs
, req
, offset
& ~(align
- 1), align
,
1349 flags
& ~BDRV_REQ_ZERO_WRITE
);
1353 offset
+= zero_bytes
;
1354 bytes
-= zero_bytes
;
1357 assert(!bytes
|| (offset
& (align
- 1)) == 0);
1358 if (bytes
>= align
) {
1359 /* Write the aligned part in the middle. */
1360 uint64_t aligned_bytes
= bytes
& ~(align
- 1);
1361 ret
= bdrv_aligned_pwritev(bs
, req
, offset
, aligned_bytes
,
1366 bytes
-= aligned_bytes
;
1367 offset
+= aligned_bytes
;
1370 assert(!bytes
|| (offset
& (align
- 1)) == 0);
1372 assert(align
== tail_padding_bytes
+ bytes
);
1373 /* RMW the unaligned part after tail. */
1374 mark_request_serialising(req
, align
);
1375 wait_serialising_requests(req
);
1376 bdrv_debug_event(bs
, BLKDBG_PWRITEV_RMW_TAIL
);
1377 ret
= bdrv_aligned_preadv(bs
, req
, offset
, align
,
1378 align
, &local_qiov
, 0);
1382 bdrv_debug_event(bs
, BLKDBG_PWRITEV_RMW_AFTER_TAIL
);
1384 memset(buf
, 0, bytes
);
1385 ret
= bdrv_aligned_pwritev(bs
, req
, offset
, align
,
1386 &local_qiov
, flags
& ~BDRV_REQ_ZERO_WRITE
);
1395 * Handle a write request in coroutine context
1397 int coroutine_fn
bdrv_co_pwritev(BlockDriverState
*bs
,
1398 int64_t offset
, unsigned int bytes
, QEMUIOVector
*qiov
,
1399 BdrvRequestFlags flags
)
1401 BdrvTrackedRequest req
;
1402 /* TODO Lift BDRV_SECTOR_SIZE restriction in BlockDriver interface */
1403 uint64_t align
= MAX(BDRV_SECTOR_SIZE
, bs
->request_alignment
);
1404 uint8_t *head_buf
= NULL
;
1405 uint8_t *tail_buf
= NULL
;
1406 QEMUIOVector local_qiov
;
1407 bool use_local_qiov
= false;
1413 if (bs
->read_only
) {
1416 assert(!(bs
->open_flags
& BDRV_O_INACTIVE
));
1418 ret
= bdrv_check_byte_request(bs
, offset
, bytes
);
1424 * Align write if necessary by performing a read-modify-write cycle.
1425 * Pad qiov with the read parts and be sure to have a tracked request not
1426 * only for bdrv_aligned_pwritev, but also for the reads of the RMW cycle.
1428 tracked_request_begin(&req
, bs
, offset
, bytes
, BDRV_TRACKED_WRITE
);
1431 ret
= bdrv_co_do_zero_pwritev(bs
, offset
, bytes
, flags
, &req
);
1435 if (offset
& (align
- 1)) {
1436 QEMUIOVector head_qiov
;
1437 struct iovec head_iov
;
1439 mark_request_serialising(&req
, align
);
1440 wait_serialising_requests(&req
);
1442 head_buf
= qemu_blockalign(bs
, align
);
1443 head_iov
= (struct iovec
) {
1444 .iov_base
= head_buf
,
1447 qemu_iovec_init_external(&head_qiov
, &head_iov
, 1);
1449 bdrv_debug_event(bs
, BLKDBG_PWRITEV_RMW_HEAD
);
1450 ret
= bdrv_aligned_preadv(bs
, &req
, offset
& ~(align
- 1), align
,
1451 align
, &head_qiov
, 0);
1455 bdrv_debug_event(bs
, BLKDBG_PWRITEV_RMW_AFTER_HEAD
);
1457 qemu_iovec_init(&local_qiov
, qiov
->niov
+ 2);
1458 qemu_iovec_add(&local_qiov
, head_buf
, offset
& (align
- 1));
1459 qemu_iovec_concat(&local_qiov
, qiov
, 0, qiov
->size
);
1460 use_local_qiov
= true;
1462 bytes
+= offset
& (align
- 1);
1463 offset
= offset
& ~(align
- 1);
1465 /* We have read the tail already if the request is smaller
1466 * than one aligned block.
1468 if (bytes
< align
) {
1469 qemu_iovec_add(&local_qiov
, head_buf
+ bytes
, align
- bytes
);
1474 if ((offset
+ bytes
) & (align
- 1)) {
1475 QEMUIOVector tail_qiov
;
1476 struct iovec tail_iov
;
1480 mark_request_serialising(&req
, align
);
1481 waited
= wait_serialising_requests(&req
);
1482 assert(!waited
|| !use_local_qiov
);
1484 tail_buf
= qemu_blockalign(bs
, align
);
1485 tail_iov
= (struct iovec
) {
1486 .iov_base
= tail_buf
,
1489 qemu_iovec_init_external(&tail_qiov
, &tail_iov
, 1);
1491 bdrv_debug_event(bs
, BLKDBG_PWRITEV_RMW_TAIL
);
1492 ret
= bdrv_aligned_preadv(bs
, &req
, (offset
+ bytes
) & ~(align
- 1), align
,
1493 align
, &tail_qiov
, 0);
1497 bdrv_debug_event(bs
, BLKDBG_PWRITEV_RMW_AFTER_TAIL
);
1499 if (!use_local_qiov
) {
1500 qemu_iovec_init(&local_qiov
, qiov
->niov
+ 1);
1501 qemu_iovec_concat(&local_qiov
, qiov
, 0, qiov
->size
);
1502 use_local_qiov
= true;
1505 tail_bytes
= (offset
+ bytes
) & (align
- 1);
1506 qemu_iovec_add(&local_qiov
, tail_buf
+ tail_bytes
, align
- tail_bytes
);
1508 bytes
= ROUND_UP(bytes
, align
);
1511 ret
= bdrv_aligned_pwritev(bs
, &req
, offset
, bytes
,
1512 use_local_qiov
? &local_qiov
: qiov
,
1517 if (use_local_qiov
) {
1518 qemu_iovec_destroy(&local_qiov
);
1520 qemu_vfree(head_buf
);
1521 qemu_vfree(tail_buf
);
1523 tracked_request_end(&req
);
1527 static int coroutine_fn
bdrv_co_do_writev(BlockDriverState
*bs
,
1528 int64_t sector_num
, int nb_sectors
, QEMUIOVector
*qiov
,
1529 BdrvRequestFlags flags
)
1531 if (nb_sectors
< 0 || nb_sectors
> BDRV_REQUEST_MAX_SECTORS
) {
1535 return bdrv_co_pwritev(bs
, sector_num
<< BDRV_SECTOR_BITS
,
1536 nb_sectors
<< BDRV_SECTOR_BITS
, qiov
, flags
);
1539 int coroutine_fn
bdrv_co_writev(BlockDriverState
*bs
, int64_t sector_num
,
1540 int nb_sectors
, QEMUIOVector
*qiov
)
1542 trace_bdrv_co_writev(bs
, sector_num
, nb_sectors
);
1544 return bdrv_co_do_writev(bs
, sector_num
, nb_sectors
, qiov
, 0);
1547 int coroutine_fn
bdrv_co_pwrite_zeroes(BlockDriverState
*bs
,
1548 int64_t offset
, int count
,
1549 BdrvRequestFlags flags
)
1551 trace_bdrv_co_pwrite_zeroes(bs
, offset
, count
, flags
);
1553 if (!(bs
->open_flags
& BDRV_O_UNMAP
)) {
1554 flags
&= ~BDRV_REQ_MAY_UNMAP
;
1557 return bdrv_co_pwritev(bs
, offset
, count
, NULL
,
1558 BDRV_REQ_ZERO_WRITE
| flags
);
1561 typedef struct BdrvCoGetBlockStatusData
{
1562 BlockDriverState
*bs
;
1563 BlockDriverState
*base
;
1564 BlockDriverState
**file
;
1570 } BdrvCoGetBlockStatusData
;
1573 * Returns the allocation status of the specified sectors.
1574 * Drivers not implementing the functionality are assumed to not support
1575 * backing files, hence all their sectors are reported as allocated.
1577 * If 'sector_num' is beyond the end of the disk image the return value is 0
1578 * and 'pnum' is set to 0.
1580 * 'pnum' is set to the number of sectors (including and immediately following
1581 * the specified sector) that are known to be in the same
1582 * allocated/unallocated state.
1584 * 'nb_sectors' is the max value 'pnum' should be set to. If nb_sectors goes
1585 * beyond the end of the disk image it will be clamped.
1587 * If returned value is positive and BDRV_BLOCK_OFFSET_VALID bit is set, 'file'
1588 * points to the BDS which the sector range is allocated in.
1590 static int64_t coroutine_fn
bdrv_co_get_block_status(BlockDriverState
*bs
,
1592 int nb_sectors
, int *pnum
,
1593 BlockDriverState
**file
)
1595 int64_t total_sectors
;
1599 total_sectors
= bdrv_nb_sectors(bs
);
1600 if (total_sectors
< 0) {
1601 return total_sectors
;
1604 if (sector_num
>= total_sectors
) {
1609 n
= total_sectors
- sector_num
;
1610 if (n
< nb_sectors
) {
1614 if (!bs
->drv
->bdrv_co_get_block_status
) {
1616 ret
= BDRV_BLOCK_DATA
| BDRV_BLOCK_ALLOCATED
;
1617 if (bs
->drv
->protocol_name
) {
1618 ret
|= BDRV_BLOCK_OFFSET_VALID
| (sector_num
* BDRV_SECTOR_SIZE
);
1624 ret
= bs
->drv
->bdrv_co_get_block_status(bs
, sector_num
, nb_sectors
, pnum
,
1631 if (ret
& BDRV_BLOCK_RAW
) {
1632 assert(ret
& BDRV_BLOCK_OFFSET_VALID
);
1633 return bdrv_get_block_status(bs
->file
->bs
, ret
>> BDRV_SECTOR_BITS
,
1637 if (ret
& (BDRV_BLOCK_DATA
| BDRV_BLOCK_ZERO
)) {
1638 ret
|= BDRV_BLOCK_ALLOCATED
;
1640 if (bdrv_unallocated_blocks_are_zero(bs
)) {
1641 ret
|= BDRV_BLOCK_ZERO
;
1642 } else if (bs
->backing
) {
1643 BlockDriverState
*bs2
= bs
->backing
->bs
;
1644 int64_t nb_sectors2
= bdrv_nb_sectors(bs2
);
1645 if (nb_sectors2
>= 0 && sector_num
>= nb_sectors2
) {
1646 ret
|= BDRV_BLOCK_ZERO
;
1651 if (*file
&& *file
!= bs
&&
1652 (ret
& BDRV_BLOCK_DATA
) && !(ret
& BDRV_BLOCK_ZERO
) &&
1653 (ret
& BDRV_BLOCK_OFFSET_VALID
)) {
1654 BlockDriverState
*file2
;
1657 ret2
= bdrv_co_get_block_status(*file
, ret
>> BDRV_SECTOR_BITS
,
1658 *pnum
, &file_pnum
, &file2
);
1660 /* Ignore errors. This is just providing extra information, it
1661 * is useful but not necessary.
1664 /* !file_pnum indicates an offset at or beyond the EOF; it is
1665 * perfectly valid for the format block driver to point to such
1666 * offsets, so catch it and mark everything as zero */
1667 ret
|= BDRV_BLOCK_ZERO
;
1669 /* Limit request to the range reported by the protocol driver */
1671 ret
|= (ret2
& BDRV_BLOCK_ZERO
);
1679 static int64_t coroutine_fn
bdrv_co_get_block_status_above(BlockDriverState
*bs
,
1680 BlockDriverState
*base
,
1684 BlockDriverState
**file
)
1686 BlockDriverState
*p
;
1690 for (p
= bs
; p
!= base
; p
= backing_bs(p
)) {
1691 ret
= bdrv_co_get_block_status(p
, sector_num
, nb_sectors
, pnum
, file
);
1692 if (ret
< 0 || ret
& BDRV_BLOCK_ALLOCATED
) {
1695 /* [sector_num, pnum] unallocated on this layer, which could be only
1696 * the first part of [sector_num, nb_sectors]. */
1697 nb_sectors
= MIN(nb_sectors
, *pnum
);
1702 /* Coroutine wrapper for bdrv_get_block_status_above() */
1703 static void coroutine_fn
bdrv_get_block_status_above_co_entry(void *opaque
)
1705 BdrvCoGetBlockStatusData
*data
= opaque
;
1707 data
->ret
= bdrv_co_get_block_status_above(data
->bs
, data
->base
,
1716 * Synchronous wrapper around bdrv_co_get_block_status_above().
1718 * See bdrv_co_get_block_status_above() for details.
1720 int64_t bdrv_get_block_status_above(BlockDriverState
*bs
,
1721 BlockDriverState
*base
,
1723 int nb_sectors
, int *pnum
,
1724 BlockDriverState
**file
)
1727 BdrvCoGetBlockStatusData data
= {
1731 .sector_num
= sector_num
,
1732 .nb_sectors
= nb_sectors
,
1737 if (qemu_in_coroutine()) {
1738 /* Fast-path if already in coroutine context */
1739 bdrv_get_block_status_above_co_entry(&data
);
1741 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
1743 co
= qemu_coroutine_create(bdrv_get_block_status_above_co_entry
);
1744 qemu_coroutine_enter(co
, &data
);
1745 while (!data
.done
) {
1746 aio_poll(aio_context
, true);
1752 int64_t bdrv_get_block_status(BlockDriverState
*bs
,
1754 int nb_sectors
, int *pnum
,
1755 BlockDriverState
**file
)
1757 return bdrv_get_block_status_above(bs
, backing_bs(bs
),
1758 sector_num
, nb_sectors
, pnum
, file
);
1761 int coroutine_fn
bdrv_is_allocated(BlockDriverState
*bs
, int64_t sector_num
,
1762 int nb_sectors
, int *pnum
)
1764 BlockDriverState
*file
;
1765 int64_t ret
= bdrv_get_block_status(bs
, sector_num
, nb_sectors
, pnum
,
1770 return !!(ret
& BDRV_BLOCK_ALLOCATED
);
1774 * Given an image chain: ... -> [BASE] -> [INTER1] -> [INTER2] -> [TOP]
1776 * Return true if the given sector is allocated in any image between
1777 * BASE and TOP (inclusive). BASE can be NULL to check if the given
1778 * sector is allocated in any image of the chain. Return false otherwise.
1780 * 'pnum' is set to the number of sectors (including and immediately following
1781 * the specified sector) that are known to be in the same
1782 * allocated/unallocated state.
1785 int bdrv_is_allocated_above(BlockDriverState
*top
,
1786 BlockDriverState
*base
,
1788 int nb_sectors
, int *pnum
)
1790 BlockDriverState
*intermediate
;
1791 int ret
, n
= nb_sectors
;
1794 while (intermediate
&& intermediate
!= base
) {
1796 ret
= bdrv_is_allocated(intermediate
, sector_num
, nb_sectors
,
1806 * [sector_num, nb_sectors] is unallocated on top but intermediate
1809 * [sector_num+x, nr_sectors] allocated.
1811 if (n
> pnum_inter
&&
1812 (intermediate
== top
||
1813 sector_num
+ pnum_inter
< intermediate
->total_sectors
)) {
1817 intermediate
= backing_bs(intermediate
);
1824 int bdrv_write_compressed(BlockDriverState
*bs
, int64_t sector_num
,
1825 const uint8_t *buf
, int nb_sectors
)
1827 BlockDriver
*drv
= bs
->drv
;
1833 if (!drv
->bdrv_write_compressed
) {
1836 ret
= bdrv_check_request(bs
, sector_num
, nb_sectors
);
1841 assert(QLIST_EMPTY(&bs
->dirty_bitmaps
));
1843 return drv
->bdrv_write_compressed(bs
, sector_num
, buf
, nb_sectors
);
1846 int bdrv_save_vmstate(BlockDriverState
*bs
, const uint8_t *buf
,
1847 int64_t pos
, int size
)
1850 struct iovec iov
= {
1851 .iov_base
= (void *) buf
,
1855 qemu_iovec_init_external(&qiov
, &iov
, 1);
1856 return bdrv_writev_vmstate(bs
, &qiov
, pos
);
1859 int bdrv_writev_vmstate(BlockDriverState
*bs
, QEMUIOVector
*qiov
, int64_t pos
)
1861 BlockDriver
*drv
= bs
->drv
;
1865 } else if (drv
->bdrv_save_vmstate
) {
1866 return drv
->bdrv_save_vmstate(bs
, qiov
, pos
);
1867 } else if (bs
->file
) {
1868 return bdrv_writev_vmstate(bs
->file
->bs
, qiov
, pos
);
1874 int bdrv_load_vmstate(BlockDriverState
*bs
, uint8_t *buf
,
1875 int64_t pos
, int size
)
1877 BlockDriver
*drv
= bs
->drv
;
1880 if (drv
->bdrv_load_vmstate
)
1881 return drv
->bdrv_load_vmstate(bs
, buf
, pos
, size
);
1883 return bdrv_load_vmstate(bs
->file
->bs
, buf
, pos
, size
);
1887 /**************************************************************/
1890 BlockAIOCB
*bdrv_aio_readv(BlockDriverState
*bs
, int64_t sector_num
,
1891 QEMUIOVector
*qiov
, int nb_sectors
,
1892 BlockCompletionFunc
*cb
, void *opaque
)
1894 trace_bdrv_aio_readv(bs
, sector_num
, nb_sectors
, opaque
);
1896 return bdrv_co_aio_rw_vector(bs
, sector_num
, qiov
, nb_sectors
, 0,
1900 BlockAIOCB
*bdrv_aio_writev(BlockDriverState
*bs
, int64_t sector_num
,
1901 QEMUIOVector
*qiov
, int nb_sectors
,
1902 BlockCompletionFunc
*cb
, void *opaque
)
1904 trace_bdrv_aio_writev(bs
, sector_num
, nb_sectors
, opaque
);
1906 return bdrv_co_aio_rw_vector(bs
, sector_num
, qiov
, nb_sectors
, 0,
1910 void bdrv_aio_cancel(BlockAIOCB
*acb
)
1913 bdrv_aio_cancel_async(acb
);
1914 while (acb
->refcnt
> 1) {
1915 if (acb
->aiocb_info
->get_aio_context
) {
1916 aio_poll(acb
->aiocb_info
->get_aio_context(acb
), true);
1917 } else if (acb
->bs
) {
1918 aio_poll(bdrv_get_aio_context(acb
->bs
), true);
1923 qemu_aio_unref(acb
);
1926 /* Async version of aio cancel. The caller is not blocked if the acb implements
1927 * cancel_async, otherwise we do nothing and let the request normally complete.
1928 * In either case the completion callback must be called. */
1929 void bdrv_aio_cancel_async(BlockAIOCB
*acb
)
1931 if (acb
->aiocb_info
->cancel_async
) {
1932 acb
->aiocb_info
->cancel_async(acb
);
1936 /**************************************************************/
1937 /* async block device emulation */
1939 typedef struct BlockRequest
{
1941 /* Used during read, write, trim */
1948 /* Used during ioctl */
1954 BlockCompletionFunc
*cb
;
1960 typedef struct BlockAIOCBCoroutine
{
1967 } BlockAIOCBCoroutine
;
1969 static const AIOCBInfo bdrv_em_co_aiocb_info
= {
1970 .aiocb_size
= sizeof(BlockAIOCBCoroutine
),
1973 static void bdrv_co_complete(BlockAIOCBCoroutine
*acb
)
1975 if (!acb
->need_bh
) {
1976 acb
->common
.cb(acb
->common
.opaque
, acb
->req
.error
);
1977 qemu_aio_unref(acb
);
1981 static void bdrv_co_em_bh(void *opaque
)
1983 BlockAIOCBCoroutine
*acb
= opaque
;
1985 assert(!acb
->need_bh
);
1986 qemu_bh_delete(acb
->bh
);
1987 bdrv_co_complete(acb
);
1990 static void bdrv_co_maybe_schedule_bh(BlockAIOCBCoroutine
*acb
)
1992 acb
->need_bh
= false;
1993 if (acb
->req
.error
!= -EINPROGRESS
) {
1994 BlockDriverState
*bs
= acb
->common
.bs
;
1996 acb
->bh
= aio_bh_new(bdrv_get_aio_context(bs
), bdrv_co_em_bh
, acb
);
1997 qemu_bh_schedule(acb
->bh
);
2001 /* Invoke bdrv_co_do_readv/bdrv_co_do_writev */
2002 static void coroutine_fn
bdrv_co_do_rw(void *opaque
)
2004 BlockAIOCBCoroutine
*acb
= opaque
;
2005 BlockDriverState
*bs
= acb
->common
.bs
;
2007 if (!acb
->is_write
) {
2008 acb
->req
.error
= bdrv_co_do_readv(bs
, acb
->req
.sector
,
2009 acb
->req
.nb_sectors
, acb
->req
.qiov
, acb
->req
.flags
);
2011 acb
->req
.error
= bdrv_co_do_writev(bs
, acb
->req
.sector
,
2012 acb
->req
.nb_sectors
, acb
->req
.qiov
, acb
->req
.flags
);
2015 bdrv_co_complete(acb
);
2018 static BlockAIOCB
*bdrv_co_aio_rw_vector(BlockDriverState
*bs
,
2022 BdrvRequestFlags flags
,
2023 BlockCompletionFunc
*cb
,
2028 BlockAIOCBCoroutine
*acb
;
2030 acb
= qemu_aio_get(&bdrv_em_co_aiocb_info
, bs
, cb
, opaque
);
2031 acb
->need_bh
= true;
2032 acb
->req
.error
= -EINPROGRESS
;
2033 acb
->req
.sector
= sector_num
;
2034 acb
->req
.nb_sectors
= nb_sectors
;
2035 acb
->req
.qiov
= qiov
;
2036 acb
->req
.flags
= flags
;
2037 acb
->is_write
= is_write
;
2039 co
= qemu_coroutine_create(bdrv_co_do_rw
);
2040 qemu_coroutine_enter(co
, acb
);
2042 bdrv_co_maybe_schedule_bh(acb
);
2043 return &acb
->common
;
2046 static void coroutine_fn
bdrv_aio_flush_co_entry(void *opaque
)
2048 BlockAIOCBCoroutine
*acb
= opaque
;
2049 BlockDriverState
*bs
= acb
->common
.bs
;
2051 acb
->req
.error
= bdrv_co_flush(bs
);
2052 bdrv_co_complete(acb
);
2055 BlockAIOCB
*bdrv_aio_flush(BlockDriverState
*bs
,
2056 BlockCompletionFunc
*cb
, void *opaque
)
2058 trace_bdrv_aio_flush(bs
, opaque
);
2061 BlockAIOCBCoroutine
*acb
;
2063 acb
= qemu_aio_get(&bdrv_em_co_aiocb_info
, bs
, cb
, opaque
);
2064 acb
->need_bh
= true;
2065 acb
->req
.error
= -EINPROGRESS
;
2067 co
= qemu_coroutine_create(bdrv_aio_flush_co_entry
);
2068 qemu_coroutine_enter(co
, acb
);
2070 bdrv_co_maybe_schedule_bh(acb
);
2071 return &acb
->common
;
2074 static void coroutine_fn
bdrv_aio_discard_co_entry(void *opaque
)
2076 BlockAIOCBCoroutine
*acb
= opaque
;
2077 BlockDriverState
*bs
= acb
->common
.bs
;
2079 acb
->req
.error
= bdrv_co_discard(bs
, acb
->req
.sector
, acb
->req
.nb_sectors
);
2080 bdrv_co_complete(acb
);
2083 BlockAIOCB
*bdrv_aio_discard(BlockDriverState
*bs
,
2084 int64_t sector_num
, int nb_sectors
,
2085 BlockCompletionFunc
*cb
, void *opaque
)
2088 BlockAIOCBCoroutine
*acb
;
2090 trace_bdrv_aio_discard(bs
, sector_num
, nb_sectors
, opaque
);
2092 acb
= qemu_aio_get(&bdrv_em_co_aiocb_info
, bs
, cb
, opaque
);
2093 acb
->need_bh
= true;
2094 acb
->req
.error
= -EINPROGRESS
;
2095 acb
->req
.sector
= sector_num
;
2096 acb
->req
.nb_sectors
= nb_sectors
;
2097 co
= qemu_coroutine_create(bdrv_aio_discard_co_entry
);
2098 qemu_coroutine_enter(co
, acb
);
2100 bdrv_co_maybe_schedule_bh(acb
);
2101 return &acb
->common
;
2104 void *qemu_aio_get(const AIOCBInfo
*aiocb_info
, BlockDriverState
*bs
,
2105 BlockCompletionFunc
*cb
, void *opaque
)
2109 acb
= g_malloc(aiocb_info
->aiocb_size
);
2110 acb
->aiocb_info
= aiocb_info
;
2113 acb
->opaque
= opaque
;
2118 void qemu_aio_ref(void *p
)
2120 BlockAIOCB
*acb
= p
;
2124 void qemu_aio_unref(void *p
)
2126 BlockAIOCB
*acb
= p
;
2127 assert(acb
->refcnt
> 0);
2128 if (--acb
->refcnt
== 0) {
2133 /**************************************************************/
2134 /* Coroutine block device emulation */
2136 static void coroutine_fn
bdrv_flush_co_entry(void *opaque
)
2138 RwCo
*rwco
= opaque
;
2140 rwco
->ret
= bdrv_co_flush(rwco
->bs
);
2143 int coroutine_fn
bdrv_co_flush(BlockDriverState
*bs
)
2146 BdrvTrackedRequest req
;
2148 if (!bs
|| !bdrv_is_inserted(bs
) || bdrv_is_read_only(bs
) ||
2153 tracked_request_begin(&req
, bs
, 0, 0, BDRV_TRACKED_FLUSH
);
2155 /* Write back all layers by calling one driver function */
2156 if (bs
->drv
->bdrv_co_flush
) {
2157 ret
= bs
->drv
->bdrv_co_flush(bs
);
2161 /* Write back cached data to the OS even with cache=unsafe */
2162 BLKDBG_EVENT(bs
->file
, BLKDBG_FLUSH_TO_OS
);
2163 if (bs
->drv
->bdrv_co_flush_to_os
) {
2164 ret
= bs
->drv
->bdrv_co_flush_to_os(bs
);
2170 /* But don't actually force it to the disk with cache=unsafe */
2171 if (bs
->open_flags
& BDRV_O_NO_FLUSH
) {
2175 BLKDBG_EVENT(bs
->file
, BLKDBG_FLUSH_TO_DISK
);
2176 if (bs
->drv
->bdrv_co_flush_to_disk
) {
2177 ret
= bs
->drv
->bdrv_co_flush_to_disk(bs
);
2178 } else if (bs
->drv
->bdrv_aio_flush
) {
2180 CoroutineIOCompletion co
= {
2181 .coroutine
= qemu_coroutine_self(),
2184 acb
= bs
->drv
->bdrv_aio_flush(bs
, bdrv_co_io_em_complete
, &co
);
2188 qemu_coroutine_yield();
2193 * Some block drivers always operate in either writethrough or unsafe
2194 * mode and don't support bdrv_flush therefore. Usually qemu doesn't
2195 * know how the server works (because the behaviour is hardcoded or
2196 * depends on server-side configuration), so we can't ensure that
2197 * everything is safe on disk. Returning an error doesn't work because
2198 * that would break guests even if the server operates in writethrough
2201 * Let's hope the user knows what he's doing.
2209 /* Now flush the underlying protocol. It will also have BDRV_O_NO_FLUSH
2210 * in the case of cache=unsafe, so there are no useless flushes.
2213 ret
= bs
->file
? bdrv_co_flush(bs
->file
->bs
) : 0;
2215 tracked_request_end(&req
);
2219 int bdrv_flush(BlockDriverState
*bs
)
2227 if (qemu_in_coroutine()) {
2228 /* Fast-path if already in coroutine context */
2229 bdrv_flush_co_entry(&rwco
);
2231 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
2233 co
= qemu_coroutine_create(bdrv_flush_co_entry
);
2234 qemu_coroutine_enter(co
, &rwco
);
2235 while (rwco
.ret
== NOT_DONE
) {
2236 aio_poll(aio_context
, true);
2243 typedef struct DiscardCo
{
2244 BlockDriverState
*bs
;
2249 static void coroutine_fn
bdrv_discard_co_entry(void *opaque
)
2251 DiscardCo
*rwco
= opaque
;
2253 rwco
->ret
= bdrv_co_discard(rwco
->bs
, rwco
->sector_num
, rwco
->nb_sectors
);
2256 int coroutine_fn
bdrv_co_discard(BlockDriverState
*bs
, int64_t sector_num
,
2259 BdrvTrackedRequest req
;
2260 int max_discard
, ret
;
2266 ret
= bdrv_check_request(bs
, sector_num
, nb_sectors
);
2269 } else if (bs
->read_only
) {
2272 assert(!(bs
->open_flags
& BDRV_O_INACTIVE
));
2274 /* Do nothing if disabled. */
2275 if (!(bs
->open_flags
& BDRV_O_UNMAP
)) {
2279 if (!bs
->drv
->bdrv_co_discard
&& !bs
->drv
->bdrv_aio_discard
) {
2283 tracked_request_begin(&req
, bs
, sector_num
, nb_sectors
,
2284 BDRV_TRACKED_DISCARD
);
2285 bdrv_set_dirty(bs
, sector_num
, nb_sectors
);
2287 max_discard
= MIN_NON_ZERO(bs
->bl
.max_discard
, BDRV_REQUEST_MAX_SECTORS
);
2288 while (nb_sectors
> 0) {
2290 int num
= nb_sectors
;
2293 if (bs
->bl
.discard_alignment
&&
2294 num
>= bs
->bl
.discard_alignment
&&
2295 sector_num
% bs
->bl
.discard_alignment
) {
2296 if (num
> bs
->bl
.discard_alignment
) {
2297 num
= bs
->bl
.discard_alignment
;
2299 num
-= sector_num
% bs
->bl
.discard_alignment
;
2302 /* limit request size */
2303 if (num
> max_discard
) {
2307 if (bs
->drv
->bdrv_co_discard
) {
2308 ret
= bs
->drv
->bdrv_co_discard(bs
, sector_num
, num
);
2311 CoroutineIOCompletion co
= {
2312 .coroutine
= qemu_coroutine_self(),
2315 acb
= bs
->drv
->bdrv_aio_discard(bs
, sector_num
, nb_sectors
,
2316 bdrv_co_io_em_complete
, &co
);
2321 qemu_coroutine_yield();
2325 if (ret
&& ret
!= -ENOTSUP
) {
2334 tracked_request_end(&req
);
2338 int bdrv_discard(BlockDriverState
*bs
, int64_t sector_num
, int nb_sectors
)
2343 .sector_num
= sector_num
,
2344 .nb_sectors
= nb_sectors
,
2348 if (qemu_in_coroutine()) {
2349 /* Fast-path if already in coroutine context */
2350 bdrv_discard_co_entry(&rwco
);
2352 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
2354 co
= qemu_coroutine_create(bdrv_discard_co_entry
);
2355 qemu_coroutine_enter(co
, &rwco
);
2356 while (rwco
.ret
== NOT_DONE
) {
2357 aio_poll(aio_context
, true);
2364 static int bdrv_co_do_ioctl(BlockDriverState
*bs
, int req
, void *buf
)
2366 BlockDriver
*drv
= bs
->drv
;
2367 BdrvTrackedRequest tracked_req
;
2368 CoroutineIOCompletion co
= {
2369 .coroutine
= qemu_coroutine_self(),
2373 tracked_request_begin(&tracked_req
, bs
, 0, 0, BDRV_TRACKED_IOCTL
);
2374 if (!drv
|| !drv
->bdrv_aio_ioctl
) {
2379 acb
= drv
->bdrv_aio_ioctl(bs
, req
, buf
, bdrv_co_io_em_complete
, &co
);
2384 qemu_coroutine_yield();
2386 tracked_request_end(&tracked_req
);
2391 BlockDriverState
*bs
;
2397 static void coroutine_fn
bdrv_co_ioctl_entry(void *opaque
)
2399 BdrvIoctlCoData
*data
= opaque
;
2400 data
->ret
= bdrv_co_do_ioctl(data
->bs
, data
->req
, data
->buf
);
2403 /* needed for generic scsi interface */
2404 int bdrv_ioctl(BlockDriverState
*bs
, unsigned long int req
, void *buf
)
2406 BdrvIoctlCoData data
= {
2410 .ret
= -EINPROGRESS
,
2413 if (qemu_in_coroutine()) {
2414 /* Fast-path if already in coroutine context */
2415 bdrv_co_ioctl_entry(&data
);
2417 Coroutine
*co
= qemu_coroutine_create(bdrv_co_ioctl_entry
);
2419 qemu_coroutine_enter(co
, &data
);
2420 while (data
.ret
== -EINPROGRESS
) {
2421 aio_poll(bdrv_get_aio_context(bs
), true);
2427 static void coroutine_fn
bdrv_co_aio_ioctl_entry(void *opaque
)
2429 BlockAIOCBCoroutine
*acb
= opaque
;
2430 acb
->req
.error
= bdrv_co_do_ioctl(acb
->common
.bs
,
2431 acb
->req
.req
, acb
->req
.buf
);
2432 bdrv_co_complete(acb
);
2435 BlockAIOCB
*bdrv_aio_ioctl(BlockDriverState
*bs
,
2436 unsigned long int req
, void *buf
,
2437 BlockCompletionFunc
*cb
, void *opaque
)
2439 BlockAIOCBCoroutine
*acb
= qemu_aio_get(&bdrv_em_co_aiocb_info
,
2443 acb
->need_bh
= true;
2444 acb
->req
.error
= -EINPROGRESS
;
2447 co
= qemu_coroutine_create(bdrv_co_aio_ioctl_entry
);
2448 qemu_coroutine_enter(co
, acb
);
2450 bdrv_co_maybe_schedule_bh(acb
);
2451 return &acb
->common
;
2454 void *qemu_blockalign(BlockDriverState
*bs
, size_t size
)
2456 return qemu_memalign(bdrv_opt_mem_align(bs
), size
);
2459 void *qemu_blockalign0(BlockDriverState
*bs
, size_t size
)
2461 return memset(qemu_blockalign(bs
, size
), 0, size
);
2464 void *qemu_try_blockalign(BlockDriverState
*bs
, size_t size
)
2466 size_t align
= bdrv_opt_mem_align(bs
);
2468 /* Ensure that NULL is never returned on success */
2474 return qemu_try_memalign(align
, size
);
2477 void *qemu_try_blockalign0(BlockDriverState
*bs
, size_t size
)
2479 void *mem
= qemu_try_blockalign(bs
, size
);
2482 memset(mem
, 0, size
);
2489 * Check if all memory in this vector is sector aligned.
2491 bool bdrv_qiov_is_aligned(BlockDriverState
*bs
, QEMUIOVector
*qiov
)
2494 size_t alignment
= bdrv_min_mem_align(bs
);
2496 for (i
= 0; i
< qiov
->niov
; i
++) {
2497 if ((uintptr_t) qiov
->iov
[i
].iov_base
% alignment
) {
2500 if (qiov
->iov
[i
].iov_len
% alignment
) {
2508 void bdrv_add_before_write_notifier(BlockDriverState
*bs
,
2509 NotifierWithReturn
*notifier
)
2511 notifier_with_return_list_add(&bs
->before_write_notifiers
, notifier
);
2514 void bdrv_io_plug(BlockDriverState
*bs
)
2518 QLIST_FOREACH(child
, &bs
->children
, next
) {
2519 bdrv_io_plug(child
->bs
);
2522 if (bs
->io_plugged
++ == 0 && bs
->io_plug_disabled
== 0) {
2523 BlockDriver
*drv
= bs
->drv
;
2524 if (drv
&& drv
->bdrv_io_plug
) {
2525 drv
->bdrv_io_plug(bs
);
2530 void bdrv_io_unplug(BlockDriverState
*bs
)
2534 assert(bs
->io_plugged
);
2535 if (--bs
->io_plugged
== 0 && bs
->io_plug_disabled
== 0) {
2536 BlockDriver
*drv
= bs
->drv
;
2537 if (drv
&& drv
->bdrv_io_unplug
) {
2538 drv
->bdrv_io_unplug(bs
);
2542 QLIST_FOREACH(child
, &bs
->children
, next
) {
2543 bdrv_io_unplug(child
->bs
);
2547 void bdrv_io_unplugged_begin(BlockDriverState
*bs
)
2551 if (bs
->io_plug_disabled
++ == 0 && bs
->io_plugged
> 0) {
2552 BlockDriver
*drv
= bs
->drv
;
2553 if (drv
&& drv
->bdrv_io_unplug
) {
2554 drv
->bdrv_io_unplug(bs
);
2558 QLIST_FOREACH(child
, &bs
->children
, next
) {
2559 bdrv_io_unplugged_begin(child
->bs
);
2563 void bdrv_io_unplugged_end(BlockDriverState
*bs
)
2567 assert(bs
->io_plug_disabled
);
2568 QLIST_FOREACH(child
, &bs
->children
, next
) {
2569 bdrv_io_unplugged_end(child
->bs
);
2572 if (--bs
->io_plug_disabled
== 0 && bs
->io_plugged
> 0) {
2573 BlockDriver
*drv
= bs
->drv
;
2574 if (drv
&& drv
->bdrv_io_plug
) {
2575 drv
->bdrv_io_plug(bs
);