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_prw_vector(BdrvChild
*child
,
39 BdrvRequestFlags flags
,
40 BlockCompletionFunc
*cb
,
43 static void coroutine_fn
bdrv_co_do_rw(void *opaque
);
44 static int coroutine_fn
bdrv_co_do_pwrite_zeroes(BlockDriverState
*bs
,
45 int64_t offset
, int count
, BdrvRequestFlags flags
);
47 static void bdrv_parent_drained_begin(BlockDriverState
*bs
)
51 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
52 if (c
->role
->drained_begin
) {
53 c
->role
->drained_begin(c
);
58 static void bdrv_parent_drained_end(BlockDriverState
*bs
)
62 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
63 if (c
->role
->drained_end
) {
64 c
->role
->drained_end(c
);
69 static void bdrv_merge_limits(BlockLimits
*dst
, const BlockLimits
*src
)
71 dst
->opt_transfer
= MAX(dst
->opt_transfer
, src
->opt_transfer
);
72 dst
->max_transfer
= MIN_NON_ZERO(dst
->max_transfer
, src
->max_transfer
);
73 dst
->opt_mem_alignment
= MAX(dst
->opt_mem_alignment
,
74 src
->opt_mem_alignment
);
75 dst
->min_mem_alignment
= MAX(dst
->min_mem_alignment
,
76 src
->min_mem_alignment
);
77 dst
->max_iov
= MIN_NON_ZERO(dst
->max_iov
, src
->max_iov
);
80 void bdrv_refresh_limits(BlockDriverState
*bs
, Error
**errp
)
82 BlockDriver
*drv
= bs
->drv
;
83 Error
*local_err
= NULL
;
85 memset(&bs
->bl
, 0, sizeof(bs
->bl
));
91 /* Default alignment based on whether driver has byte interface */
92 bs
->bl
.request_alignment
= drv
->bdrv_co_preadv
? 1 : 512;
94 /* Take some limits from the children as a default */
96 bdrv_refresh_limits(bs
->file
->bs
, &local_err
);
98 error_propagate(errp
, local_err
);
101 bdrv_merge_limits(&bs
->bl
, &bs
->file
->bs
->bl
);
103 bs
->bl
.min_mem_alignment
= 512;
104 bs
->bl
.opt_mem_alignment
= getpagesize();
106 /* Safe default since most protocols use readv()/writev()/etc */
107 bs
->bl
.max_iov
= IOV_MAX
;
111 bdrv_refresh_limits(bs
->backing
->bs
, &local_err
);
113 error_propagate(errp
, local_err
);
116 bdrv_merge_limits(&bs
->bl
, &bs
->backing
->bs
->bl
);
119 /* Then let the driver override it */
120 if (drv
->bdrv_refresh_limits
) {
121 drv
->bdrv_refresh_limits(bs
, errp
);
126 * The copy-on-read flag is actually a reference count so multiple users may
127 * use the feature without worrying about clobbering its previous state.
128 * Copy-on-read stays enabled until all users have called to disable it.
130 void bdrv_enable_copy_on_read(BlockDriverState
*bs
)
135 void bdrv_disable_copy_on_read(BlockDriverState
*bs
)
137 assert(bs
->copy_on_read
> 0);
141 /* Check if any requests are in-flight (including throttled requests) */
142 bool bdrv_requests_pending(BlockDriverState
*bs
)
146 if (!QLIST_EMPTY(&bs
->tracked_requests
)) {
150 QLIST_FOREACH(child
, &bs
->children
, next
) {
151 if (bdrv_requests_pending(child
->bs
)) {
159 static void bdrv_drain_recurse(BlockDriverState
*bs
)
163 if (bs
->drv
&& bs
->drv
->bdrv_drain
) {
164 bs
->drv
->bdrv_drain(bs
);
166 QLIST_FOREACH(child
, &bs
->children
, next
) {
167 bdrv_drain_recurse(child
->bs
);
173 BlockDriverState
*bs
;
178 static void bdrv_drain_poll(BlockDriverState
*bs
)
184 busy
= bdrv_requests_pending(bs
);
185 busy
|= aio_poll(bdrv_get_aio_context(bs
), busy
);
189 static void bdrv_co_drain_bh_cb(void *opaque
)
191 BdrvCoDrainData
*data
= opaque
;
192 Coroutine
*co
= data
->co
;
194 qemu_bh_delete(data
->bh
);
195 bdrv_drain_poll(data
->bs
);
197 qemu_coroutine_enter(co
);
200 static void coroutine_fn
bdrv_co_yield_to_drain(BlockDriverState
*bs
)
202 BdrvCoDrainData data
;
204 /* Calling bdrv_drain() from a BH ensures the current coroutine yields and
205 * other coroutines run if they were queued from
206 * qemu_co_queue_run_restart(). */
208 assert(qemu_in_coroutine());
209 data
= (BdrvCoDrainData
) {
210 .co
= qemu_coroutine_self(),
213 .bh
= aio_bh_new(bdrv_get_aio_context(bs
), bdrv_co_drain_bh_cb
, &data
),
215 qemu_bh_schedule(data
.bh
);
217 qemu_coroutine_yield();
218 /* If we are resumed from some other event (such as an aio completion or a
219 * timer callback), it is a bug in the caller that should be fixed. */
223 void bdrv_drained_begin(BlockDriverState
*bs
)
225 if (!bs
->quiesce_counter
++) {
226 aio_disable_external(bdrv_get_aio_context(bs
));
227 bdrv_parent_drained_begin(bs
);
230 bdrv_io_unplugged_begin(bs
);
231 bdrv_drain_recurse(bs
);
232 if (qemu_in_coroutine()) {
233 bdrv_co_yield_to_drain(bs
);
237 bdrv_io_unplugged_end(bs
);
240 void bdrv_drained_end(BlockDriverState
*bs
)
242 assert(bs
->quiesce_counter
> 0);
243 if (--bs
->quiesce_counter
> 0) {
247 bdrv_parent_drained_end(bs
);
248 aio_enable_external(bdrv_get_aio_context(bs
));
252 * Wait for pending requests to complete on a single BlockDriverState subtree,
253 * and suspend block driver's internal I/O until next request arrives.
255 * Note that unlike bdrv_drain_all(), the caller must hold the BlockDriverState
258 * Only this BlockDriverState's AioContext is run, so in-flight requests must
259 * not depend on events in other AioContexts. In that case, use
260 * bdrv_drain_all() instead.
262 void coroutine_fn
bdrv_co_drain(BlockDriverState
*bs
)
264 assert(qemu_in_coroutine());
265 bdrv_drained_begin(bs
);
266 bdrv_drained_end(bs
);
269 void bdrv_drain(BlockDriverState
*bs
)
271 bdrv_drained_begin(bs
);
272 bdrv_drained_end(bs
);
276 * Wait for pending requests to complete across all BlockDriverStates
278 * This function does not flush data to disk, use bdrv_flush_all() for that
279 * after calling this function.
281 void bdrv_drain_all(void)
283 /* Always run first iteration so any pending completion BHs run */
285 BlockDriverState
*bs
;
287 BlockJob
*job
= NULL
;
288 GSList
*aio_ctxs
= NULL
, *ctx
;
290 while ((job
= block_job_next(job
))) {
291 AioContext
*aio_context
= blk_get_aio_context(job
->blk
);
293 aio_context_acquire(aio_context
);
294 block_job_pause(job
);
295 aio_context_release(aio_context
);
298 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
299 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
301 aio_context_acquire(aio_context
);
302 bdrv_parent_drained_begin(bs
);
303 bdrv_io_unplugged_begin(bs
);
304 bdrv_drain_recurse(bs
);
305 aio_context_release(aio_context
);
307 if (!g_slist_find(aio_ctxs
, aio_context
)) {
308 aio_ctxs
= g_slist_prepend(aio_ctxs
, aio_context
);
312 /* Note that completion of an asynchronous I/O operation can trigger any
313 * number of other I/O operations on other devices---for example a
314 * coroutine can submit an I/O request to another device in response to
315 * request completion. Therefore we must keep looping until there was no
316 * more activity rather than simply draining each device independently.
321 for (ctx
= aio_ctxs
; ctx
!= NULL
; ctx
= ctx
->next
) {
322 AioContext
*aio_context
= ctx
->data
;
324 aio_context_acquire(aio_context
);
325 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
326 if (aio_context
== bdrv_get_aio_context(bs
)) {
327 if (bdrv_requests_pending(bs
)) {
329 aio_poll(aio_context
, busy
);
333 busy
|= aio_poll(aio_context
, false);
334 aio_context_release(aio_context
);
338 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
339 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
341 aio_context_acquire(aio_context
);
342 bdrv_io_unplugged_end(bs
);
343 bdrv_parent_drained_end(bs
);
344 aio_context_release(aio_context
);
346 g_slist_free(aio_ctxs
);
349 while ((job
= block_job_next(job
))) {
350 AioContext
*aio_context
= blk_get_aio_context(job
->blk
);
352 aio_context_acquire(aio_context
);
353 block_job_resume(job
);
354 aio_context_release(aio_context
);
359 * Remove an active request from the tracked requests list
361 * This function should be called when a tracked request is completing.
363 static void tracked_request_end(BdrvTrackedRequest
*req
)
365 if (req
->serialising
) {
366 req
->bs
->serialising_in_flight
--;
369 QLIST_REMOVE(req
, list
);
370 qemu_co_queue_restart_all(&req
->wait_queue
);
374 * Add an active request to the tracked requests list
376 static void tracked_request_begin(BdrvTrackedRequest
*req
,
377 BlockDriverState
*bs
,
380 enum BdrvTrackedRequestType type
)
382 *req
= (BdrvTrackedRequest
){
387 .co
= qemu_coroutine_self(),
388 .serialising
= false,
389 .overlap_offset
= offset
,
390 .overlap_bytes
= bytes
,
393 qemu_co_queue_init(&req
->wait_queue
);
395 QLIST_INSERT_HEAD(&bs
->tracked_requests
, req
, list
);
398 static void mark_request_serialising(BdrvTrackedRequest
*req
, uint64_t align
)
400 int64_t overlap_offset
= req
->offset
& ~(align
- 1);
401 unsigned int overlap_bytes
= ROUND_UP(req
->offset
+ req
->bytes
, align
)
404 if (!req
->serialising
) {
405 req
->bs
->serialising_in_flight
++;
406 req
->serialising
= true;
409 req
->overlap_offset
= MIN(req
->overlap_offset
, overlap_offset
);
410 req
->overlap_bytes
= MAX(req
->overlap_bytes
, overlap_bytes
);
414 * Round a region to cluster boundaries (sector-based)
416 void bdrv_round_sectors_to_clusters(BlockDriverState
*bs
,
417 int64_t sector_num
, int nb_sectors
,
418 int64_t *cluster_sector_num
,
419 int *cluster_nb_sectors
)
423 if (bdrv_get_info(bs
, &bdi
) < 0 || bdi
.cluster_size
== 0) {
424 *cluster_sector_num
= sector_num
;
425 *cluster_nb_sectors
= nb_sectors
;
427 int64_t c
= bdi
.cluster_size
/ BDRV_SECTOR_SIZE
;
428 *cluster_sector_num
= QEMU_ALIGN_DOWN(sector_num
, c
);
429 *cluster_nb_sectors
= QEMU_ALIGN_UP(sector_num
- *cluster_sector_num
+
435 * Round a region to cluster boundaries
437 void bdrv_round_to_clusters(BlockDriverState
*bs
,
438 int64_t offset
, unsigned int bytes
,
439 int64_t *cluster_offset
,
440 unsigned int *cluster_bytes
)
444 if (bdrv_get_info(bs
, &bdi
) < 0 || bdi
.cluster_size
== 0) {
445 *cluster_offset
= offset
;
446 *cluster_bytes
= bytes
;
448 int64_t c
= bdi
.cluster_size
;
449 *cluster_offset
= QEMU_ALIGN_DOWN(offset
, c
);
450 *cluster_bytes
= QEMU_ALIGN_UP(offset
- *cluster_offset
+ bytes
, c
);
454 static int bdrv_get_cluster_size(BlockDriverState
*bs
)
459 ret
= bdrv_get_info(bs
, &bdi
);
460 if (ret
< 0 || bdi
.cluster_size
== 0) {
461 return bs
->bl
.request_alignment
;
463 return bdi
.cluster_size
;
467 static bool tracked_request_overlaps(BdrvTrackedRequest
*req
,
468 int64_t offset
, unsigned int bytes
)
471 if (offset
>= req
->overlap_offset
+ req
->overlap_bytes
) {
475 if (req
->overlap_offset
>= offset
+ bytes
) {
481 static bool coroutine_fn
wait_serialising_requests(BdrvTrackedRequest
*self
)
483 BlockDriverState
*bs
= self
->bs
;
484 BdrvTrackedRequest
*req
;
488 if (!bs
->serialising_in_flight
) {
494 QLIST_FOREACH(req
, &bs
->tracked_requests
, list
) {
495 if (req
== self
|| (!req
->serialising
&& !self
->serialising
)) {
498 if (tracked_request_overlaps(req
, self
->overlap_offset
,
499 self
->overlap_bytes
))
501 /* Hitting this means there was a reentrant request, for
502 * example, a block driver issuing nested requests. This must
503 * never happen since it means deadlock.
505 assert(qemu_coroutine_self() != req
->co
);
507 /* If the request is already (indirectly) waiting for us, or
508 * will wait for us as soon as it wakes up, then just go on
509 * (instead of producing a deadlock in the former case). */
510 if (!req
->waiting_for
) {
511 self
->waiting_for
= req
;
512 qemu_co_queue_wait(&req
->wait_queue
);
513 self
->waiting_for
= NULL
;
525 static int bdrv_check_byte_request(BlockDriverState
*bs
, int64_t offset
,
528 if (size
> BDRV_REQUEST_MAX_SECTORS
<< BDRV_SECTOR_BITS
) {
532 if (!bdrv_is_inserted(bs
)) {
543 typedef struct RwCo
{
549 BdrvRequestFlags flags
;
552 static void coroutine_fn
bdrv_rw_co_entry(void *opaque
)
556 if (!rwco
->is_write
) {
557 rwco
->ret
= bdrv_co_preadv(rwco
->child
, rwco
->offset
,
558 rwco
->qiov
->size
, rwco
->qiov
,
561 rwco
->ret
= bdrv_co_pwritev(rwco
->child
, rwco
->offset
,
562 rwco
->qiov
->size
, rwco
->qiov
,
568 * Process a vectored synchronous request using coroutines
570 static int bdrv_prwv_co(BdrvChild
*child
, int64_t offset
,
571 QEMUIOVector
*qiov
, bool is_write
,
572 BdrvRequestFlags flags
)
579 .is_write
= is_write
,
584 if (qemu_in_coroutine()) {
585 /* Fast-path if already in coroutine context */
586 bdrv_rw_co_entry(&rwco
);
588 AioContext
*aio_context
= bdrv_get_aio_context(child
->bs
);
590 co
= qemu_coroutine_create(bdrv_rw_co_entry
, &rwco
);
591 qemu_coroutine_enter(co
);
592 while (rwco
.ret
== NOT_DONE
) {
593 aio_poll(aio_context
, true);
600 * Process a synchronous request using coroutines
602 static int bdrv_rw_co(BdrvChild
*child
, int64_t sector_num
, uint8_t *buf
,
603 int nb_sectors
, bool is_write
, BdrvRequestFlags flags
)
607 .iov_base
= (void *)buf
,
608 .iov_len
= nb_sectors
* BDRV_SECTOR_SIZE
,
611 if (nb_sectors
< 0 || nb_sectors
> BDRV_REQUEST_MAX_SECTORS
) {
615 qemu_iovec_init_external(&qiov
, &iov
, 1);
616 return bdrv_prwv_co(child
, sector_num
<< BDRV_SECTOR_BITS
,
617 &qiov
, is_write
, flags
);
620 /* return < 0 if error. See bdrv_write() for the return codes */
621 int bdrv_read(BdrvChild
*child
, int64_t sector_num
,
622 uint8_t *buf
, int nb_sectors
)
624 return bdrv_rw_co(child
, sector_num
, buf
, nb_sectors
, false, 0);
627 /* Return < 0 if error. Important errors are:
628 -EIO generic I/O error (may happen for all errors)
629 -ENOMEDIUM No media inserted.
630 -EINVAL Invalid sector number or nb_sectors
631 -EACCES Trying to write a read-only device
633 int bdrv_write(BdrvChild
*child
, int64_t sector_num
,
634 const uint8_t *buf
, int nb_sectors
)
636 return bdrv_rw_co(child
, sector_num
, (uint8_t *)buf
, nb_sectors
, true, 0);
639 int bdrv_pwrite_zeroes(BdrvChild
*child
, int64_t offset
,
640 int count
, BdrvRequestFlags flags
)
648 qemu_iovec_init_external(&qiov
, &iov
, 1);
649 return bdrv_prwv_co(child
, offset
, &qiov
, true,
650 BDRV_REQ_ZERO_WRITE
| flags
);
654 * Completely zero out a block device with the help of bdrv_pwrite_zeroes.
655 * The operation is sped up by checking the block status and only writing
656 * zeroes to the device if they currently do not return zeroes. Optional
657 * flags are passed through to bdrv_pwrite_zeroes (e.g. BDRV_REQ_MAY_UNMAP,
660 * Returns < 0 on error, 0 on success. For error codes see bdrv_write().
662 int bdrv_make_zero(BdrvChild
*child
, BdrvRequestFlags flags
)
664 int64_t target_sectors
, ret
, nb_sectors
, sector_num
= 0;
665 BlockDriverState
*bs
= child
->bs
;
666 BlockDriverState
*file
;
669 target_sectors
= bdrv_nb_sectors(bs
);
670 if (target_sectors
< 0) {
671 return target_sectors
;
675 nb_sectors
= MIN(target_sectors
- sector_num
, BDRV_REQUEST_MAX_SECTORS
);
676 if (nb_sectors
<= 0) {
679 ret
= bdrv_get_block_status(bs
, sector_num
, nb_sectors
, &n
, &file
);
681 error_report("error getting block status at sector %" PRId64
": %s",
682 sector_num
, strerror(-ret
));
685 if (ret
& BDRV_BLOCK_ZERO
) {
689 ret
= bdrv_pwrite_zeroes(child
, sector_num
<< BDRV_SECTOR_BITS
,
690 n
<< BDRV_SECTOR_BITS
, flags
);
692 error_report("error writing zeroes at sector %" PRId64
": %s",
693 sector_num
, strerror(-ret
));
700 int bdrv_preadv(BdrvChild
*child
, int64_t offset
, QEMUIOVector
*qiov
)
704 ret
= bdrv_prwv_co(child
, offset
, qiov
, false, 0);
712 int bdrv_pread(BdrvChild
*child
, int64_t offset
, void *buf
, int bytes
)
716 .iov_base
= (void *)buf
,
724 qemu_iovec_init_external(&qiov
, &iov
, 1);
725 return bdrv_preadv(child
, offset
, &qiov
);
728 int bdrv_pwritev(BdrvChild
*child
, int64_t offset
, QEMUIOVector
*qiov
)
732 ret
= bdrv_prwv_co(child
, offset
, qiov
, true, 0);
740 int bdrv_pwrite(BdrvChild
*child
, int64_t offset
, const void *buf
, int bytes
)
744 .iov_base
= (void *) buf
,
752 qemu_iovec_init_external(&qiov
, &iov
, 1);
753 return bdrv_pwritev(child
, offset
, &qiov
);
757 * Writes to the file and ensures that no writes are reordered across this
758 * request (acts as a barrier)
760 * Returns 0 on success, -errno in error cases.
762 int bdrv_pwrite_sync(BdrvChild
*child
, int64_t offset
,
763 const void *buf
, int count
)
767 ret
= bdrv_pwrite(child
, offset
, buf
, count
);
772 ret
= bdrv_flush(child
->bs
);
780 typedef struct CoroutineIOCompletion
{
781 Coroutine
*coroutine
;
783 } CoroutineIOCompletion
;
785 static void bdrv_co_io_em_complete(void *opaque
, int ret
)
787 CoroutineIOCompletion
*co
= opaque
;
790 qemu_coroutine_enter(co
->coroutine
);
793 static int coroutine_fn
bdrv_driver_preadv(BlockDriverState
*bs
,
794 uint64_t offset
, uint64_t bytes
,
795 QEMUIOVector
*qiov
, int flags
)
797 BlockDriver
*drv
= bs
->drv
;
799 unsigned int nb_sectors
;
801 assert(!(flags
& ~BDRV_REQ_MASK
));
803 if (drv
->bdrv_co_preadv
) {
804 return drv
->bdrv_co_preadv(bs
, offset
, bytes
, qiov
, flags
);
807 sector_num
= offset
>> BDRV_SECTOR_BITS
;
808 nb_sectors
= bytes
>> BDRV_SECTOR_BITS
;
810 assert((offset
& (BDRV_SECTOR_SIZE
- 1)) == 0);
811 assert((bytes
& (BDRV_SECTOR_SIZE
- 1)) == 0);
812 assert((bytes
>> BDRV_SECTOR_BITS
) <= BDRV_REQUEST_MAX_SECTORS
);
814 if (drv
->bdrv_co_readv
) {
815 return drv
->bdrv_co_readv(bs
, sector_num
, nb_sectors
, qiov
);
818 CoroutineIOCompletion co
= {
819 .coroutine
= qemu_coroutine_self(),
822 acb
= bs
->drv
->bdrv_aio_readv(bs
, sector_num
, qiov
, nb_sectors
,
823 bdrv_co_io_em_complete
, &co
);
827 qemu_coroutine_yield();
833 static int coroutine_fn
bdrv_driver_pwritev(BlockDriverState
*bs
,
834 uint64_t offset
, uint64_t bytes
,
835 QEMUIOVector
*qiov
, int flags
)
837 BlockDriver
*drv
= bs
->drv
;
839 unsigned int nb_sectors
;
842 assert(!(flags
& ~BDRV_REQ_MASK
));
844 if (drv
->bdrv_co_pwritev
) {
845 ret
= drv
->bdrv_co_pwritev(bs
, offset
, bytes
, qiov
,
846 flags
& bs
->supported_write_flags
);
847 flags
&= ~bs
->supported_write_flags
;
851 sector_num
= offset
>> BDRV_SECTOR_BITS
;
852 nb_sectors
= bytes
>> BDRV_SECTOR_BITS
;
854 assert((offset
& (BDRV_SECTOR_SIZE
- 1)) == 0);
855 assert((bytes
& (BDRV_SECTOR_SIZE
- 1)) == 0);
856 assert((bytes
>> BDRV_SECTOR_BITS
) <= BDRV_REQUEST_MAX_SECTORS
);
858 if (drv
->bdrv_co_writev_flags
) {
859 ret
= drv
->bdrv_co_writev_flags(bs
, sector_num
, nb_sectors
, qiov
,
860 flags
& bs
->supported_write_flags
);
861 flags
&= ~bs
->supported_write_flags
;
862 } else if (drv
->bdrv_co_writev
) {
863 assert(!bs
->supported_write_flags
);
864 ret
= drv
->bdrv_co_writev(bs
, sector_num
, nb_sectors
, qiov
);
867 CoroutineIOCompletion co
= {
868 .coroutine
= qemu_coroutine_self(),
871 acb
= bs
->drv
->bdrv_aio_writev(bs
, sector_num
, qiov
, nb_sectors
,
872 bdrv_co_io_em_complete
, &co
);
876 qemu_coroutine_yield();
882 if (ret
== 0 && (flags
& BDRV_REQ_FUA
)) {
883 ret
= bdrv_co_flush(bs
);
889 static int coroutine_fn
890 bdrv_driver_pwritev_compressed(BlockDriverState
*bs
, uint64_t offset
,
891 uint64_t bytes
, QEMUIOVector
*qiov
)
893 BlockDriver
*drv
= bs
->drv
;
895 if (!drv
->bdrv_co_pwritev_compressed
) {
899 return drv
->bdrv_co_pwritev_compressed(bs
, offset
, bytes
, qiov
);
902 static int coroutine_fn
bdrv_co_do_copy_on_readv(BlockDriverState
*bs
,
903 int64_t offset
, unsigned int bytes
, QEMUIOVector
*qiov
)
905 /* Perform I/O through a temporary buffer so that users who scribble over
906 * their read buffer while the operation is in progress do not end up
907 * modifying the image file. This is critical for zero-copy guest I/O
908 * where anything might happen inside guest memory.
912 BlockDriver
*drv
= bs
->drv
;
914 QEMUIOVector bounce_qiov
;
915 int64_t cluster_offset
;
916 unsigned int cluster_bytes
;
920 /* Cover entire cluster so no additional backing file I/O is required when
921 * allocating cluster in the image file.
923 bdrv_round_to_clusters(bs
, offset
, bytes
, &cluster_offset
, &cluster_bytes
);
925 trace_bdrv_co_do_copy_on_readv(bs
, offset
, bytes
,
926 cluster_offset
, cluster_bytes
);
928 iov
.iov_len
= cluster_bytes
;
929 iov
.iov_base
= bounce_buffer
= qemu_try_blockalign(bs
, iov
.iov_len
);
930 if (bounce_buffer
== NULL
) {
935 qemu_iovec_init_external(&bounce_qiov
, &iov
, 1);
937 ret
= bdrv_driver_preadv(bs
, cluster_offset
, cluster_bytes
,
943 if (drv
->bdrv_co_pwrite_zeroes
&&
944 buffer_is_zero(bounce_buffer
, iov
.iov_len
)) {
945 /* FIXME: Should we (perhaps conditionally) be setting
946 * BDRV_REQ_MAY_UNMAP, if it will allow for a sparser copy
947 * that still correctly reads as zero? */
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, zeroing after EOF, and fragmentation of large
976 * reads; any other features must be 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
;
984 uint64_t bytes_remaining
= bytes
;
987 assert(is_power_of_2(align
));
988 assert((offset
& (align
- 1)) == 0);
989 assert((bytes
& (align
- 1)) == 0);
990 assert(!qiov
|| bytes
== qiov
->size
);
991 assert((bs
->open_flags
& BDRV_O_NO_IO
) == 0);
992 max_transfer
= QEMU_ALIGN_DOWN(MIN_NON_ZERO(bs
->bl
.max_transfer
, INT_MAX
),
995 /* TODO: We would need a per-BDS .supported_read_flags and
996 * potential fallback support, if we ever implement any read flags
997 * to pass through to drivers. For now, there aren't any
998 * passthrough flags. */
999 assert(!(flags
& ~(BDRV_REQ_NO_SERIALISING
| BDRV_REQ_COPY_ON_READ
)));
1001 /* Handle Copy on Read and associated serialisation */
1002 if (flags
& BDRV_REQ_COPY_ON_READ
) {
1003 /* If we touch the same cluster it counts as an overlap. This
1004 * guarantees that allocating writes will be serialized and not race
1005 * with each other for the same cluster. For example, in copy-on-read
1006 * it ensures that the CoR read and write operations are atomic and
1007 * guest writes cannot interleave between them. */
1008 mark_request_serialising(req
, bdrv_get_cluster_size(bs
));
1011 if (!(flags
& BDRV_REQ_NO_SERIALISING
)) {
1012 wait_serialising_requests(req
);
1015 if (flags
& BDRV_REQ_COPY_ON_READ
) {
1016 int64_t start_sector
= offset
>> BDRV_SECTOR_BITS
;
1017 int64_t end_sector
= DIV_ROUND_UP(offset
+ bytes
, BDRV_SECTOR_SIZE
);
1018 unsigned int nb_sectors
= end_sector
- start_sector
;
1021 ret
= bdrv_is_allocated(bs
, start_sector
, nb_sectors
, &pnum
);
1026 if (!ret
|| pnum
!= nb_sectors
) {
1027 ret
= bdrv_co_do_copy_on_readv(bs
, offset
, bytes
, qiov
);
1032 /* Forward the request to the BlockDriver, possibly fragmenting it */
1033 total_bytes
= bdrv_getlength(bs
);
1034 if (total_bytes
< 0) {
1039 max_bytes
= ROUND_UP(MAX(0, total_bytes
- offset
), align
);
1040 if (bytes
<= max_bytes
&& bytes
<= max_transfer
) {
1041 ret
= bdrv_driver_preadv(bs
, offset
, bytes
, qiov
, 0);
1045 while (bytes_remaining
) {
1049 QEMUIOVector local_qiov
;
1051 num
= MIN(bytes_remaining
, MIN(max_bytes
, max_transfer
));
1053 qemu_iovec_init(&local_qiov
, qiov
->niov
);
1054 qemu_iovec_concat(&local_qiov
, qiov
, bytes
- bytes_remaining
, num
);
1056 ret
= bdrv_driver_preadv(bs
, offset
+ bytes
- bytes_remaining
,
1057 num
, &local_qiov
, 0);
1059 qemu_iovec_destroy(&local_qiov
);
1061 num
= bytes_remaining
;
1062 ret
= qemu_iovec_memset(qiov
, bytes
- bytes_remaining
, 0,
1068 bytes_remaining
-= num
;
1072 return ret
< 0 ? ret
: 0;
1076 * Handle a read request in coroutine context
1078 int coroutine_fn
bdrv_co_preadv(BdrvChild
*child
,
1079 int64_t offset
, unsigned int bytes
, QEMUIOVector
*qiov
,
1080 BdrvRequestFlags flags
)
1082 BlockDriverState
*bs
= child
->bs
;
1083 BlockDriver
*drv
= bs
->drv
;
1084 BdrvTrackedRequest req
;
1086 uint64_t align
= bs
->bl
.request_alignment
;
1087 uint8_t *head_buf
= NULL
;
1088 uint8_t *tail_buf
= NULL
;
1089 QEMUIOVector local_qiov
;
1090 bool use_local_qiov
= false;
1097 ret
= bdrv_check_byte_request(bs
, offset
, bytes
);
1102 /* Don't do copy-on-read if we read data before write operation */
1103 if (bs
->copy_on_read
&& !(flags
& BDRV_REQ_NO_SERIALISING
)) {
1104 flags
|= BDRV_REQ_COPY_ON_READ
;
1107 /* Align read if necessary by padding qiov */
1108 if (offset
& (align
- 1)) {
1109 head_buf
= qemu_blockalign(bs
, align
);
1110 qemu_iovec_init(&local_qiov
, qiov
->niov
+ 2);
1111 qemu_iovec_add(&local_qiov
, head_buf
, offset
& (align
- 1));
1112 qemu_iovec_concat(&local_qiov
, qiov
, 0, qiov
->size
);
1113 use_local_qiov
= true;
1115 bytes
+= offset
& (align
- 1);
1116 offset
= offset
& ~(align
- 1);
1119 if ((offset
+ bytes
) & (align
- 1)) {
1120 if (!use_local_qiov
) {
1121 qemu_iovec_init(&local_qiov
, qiov
->niov
+ 1);
1122 qemu_iovec_concat(&local_qiov
, qiov
, 0, qiov
->size
);
1123 use_local_qiov
= true;
1125 tail_buf
= qemu_blockalign(bs
, align
);
1126 qemu_iovec_add(&local_qiov
, tail_buf
,
1127 align
- ((offset
+ bytes
) & (align
- 1)));
1129 bytes
= ROUND_UP(bytes
, align
);
1132 tracked_request_begin(&req
, bs
, offset
, bytes
, BDRV_TRACKED_READ
);
1133 ret
= bdrv_aligned_preadv(bs
, &req
, offset
, bytes
, align
,
1134 use_local_qiov
? &local_qiov
: qiov
,
1136 tracked_request_end(&req
);
1138 if (use_local_qiov
) {
1139 qemu_iovec_destroy(&local_qiov
);
1140 qemu_vfree(head_buf
);
1141 qemu_vfree(tail_buf
);
1147 static int coroutine_fn
bdrv_co_do_readv(BdrvChild
*child
,
1148 int64_t sector_num
, int nb_sectors
, QEMUIOVector
*qiov
,
1149 BdrvRequestFlags flags
)
1151 if (nb_sectors
< 0 || nb_sectors
> BDRV_REQUEST_MAX_SECTORS
) {
1155 return bdrv_co_preadv(child
, sector_num
<< BDRV_SECTOR_BITS
,
1156 nb_sectors
<< BDRV_SECTOR_BITS
, qiov
, flags
);
1159 int coroutine_fn
bdrv_co_readv(BdrvChild
*child
, int64_t sector_num
,
1160 int nb_sectors
, QEMUIOVector
*qiov
)
1162 trace_bdrv_co_readv(child
->bs
, sector_num
, nb_sectors
);
1164 return bdrv_co_do_readv(child
, sector_num
, nb_sectors
, qiov
, 0);
1167 /* Maximum buffer for write zeroes fallback, in bytes */
1168 #define MAX_WRITE_ZEROES_BOUNCE_BUFFER (32768 << BDRV_SECTOR_BITS)
1170 static int coroutine_fn
bdrv_co_do_pwrite_zeroes(BlockDriverState
*bs
,
1171 int64_t offset
, int count
, BdrvRequestFlags flags
)
1173 BlockDriver
*drv
= bs
->drv
;
1175 struct iovec iov
= {0};
1177 bool need_flush
= false;
1181 int max_write_zeroes
= MIN_NON_ZERO(bs
->bl
.max_pwrite_zeroes
, INT_MAX
);
1182 int alignment
= MAX(bs
->bl
.pwrite_zeroes_alignment
,
1183 bs
->bl
.request_alignment
);
1185 assert(alignment
% bs
->bl
.request_alignment
== 0);
1186 head
= offset
% alignment
;
1187 tail
= (offset
+ count
) % alignment
;
1188 max_write_zeroes
= QEMU_ALIGN_DOWN(max_write_zeroes
, alignment
);
1189 assert(max_write_zeroes
>= bs
->bl
.request_alignment
);
1191 while (count
> 0 && !ret
) {
1194 /* Align request. Block drivers can expect the "bulk" of the request
1195 * to be aligned, and that unaligned requests do not cross cluster
1199 /* Make a small request up to the first aligned sector. */
1200 num
= MIN(count
, alignment
- head
);
1202 } else if (tail
&& num
> alignment
) {
1203 /* Shorten the request to the last aligned sector. */
1207 /* limit request size */
1208 if (num
> max_write_zeroes
) {
1209 num
= max_write_zeroes
;
1213 /* First try the efficient write zeroes operation */
1214 if (drv
->bdrv_co_pwrite_zeroes
) {
1215 ret
= drv
->bdrv_co_pwrite_zeroes(bs
, offset
, num
,
1216 flags
& bs
->supported_zero_flags
);
1217 if (ret
!= -ENOTSUP
&& (flags
& BDRV_REQ_FUA
) &&
1218 !(bs
->supported_zero_flags
& BDRV_REQ_FUA
)) {
1222 assert(!bs
->supported_zero_flags
);
1225 if (ret
== -ENOTSUP
) {
1226 /* Fall back to bounce buffer if write zeroes is unsupported */
1227 int max_transfer
= MIN_NON_ZERO(bs
->bl
.max_transfer
,
1228 MAX_WRITE_ZEROES_BOUNCE_BUFFER
);
1229 BdrvRequestFlags write_flags
= flags
& ~BDRV_REQ_ZERO_WRITE
;
1231 if ((flags
& BDRV_REQ_FUA
) &&
1232 !(bs
->supported_write_flags
& BDRV_REQ_FUA
)) {
1233 /* No need for bdrv_driver_pwrite() to do a fallback
1234 * flush on each chunk; use just one at the end */
1235 write_flags
&= ~BDRV_REQ_FUA
;
1238 num
= MIN(num
, max_transfer
);
1240 if (iov
.iov_base
== NULL
) {
1241 iov
.iov_base
= qemu_try_blockalign(bs
, num
);
1242 if (iov
.iov_base
== NULL
) {
1246 memset(iov
.iov_base
, 0, num
);
1248 qemu_iovec_init_external(&qiov
, &iov
, 1);
1250 ret
= bdrv_driver_pwritev(bs
, offset
, num
, &qiov
, write_flags
);
1252 /* Keep bounce buffer around if it is big enough for all
1253 * all future requests.
1255 if (num
< max_transfer
) {
1256 qemu_vfree(iov
.iov_base
);
1257 iov
.iov_base
= NULL
;
1266 if (ret
== 0 && need_flush
) {
1267 ret
= bdrv_co_flush(bs
);
1269 qemu_vfree(iov
.iov_base
);
1274 * Forwards an already correctly aligned write request to the BlockDriver,
1275 * after possibly fragmenting it.
1277 static int coroutine_fn
bdrv_aligned_pwritev(BlockDriverState
*bs
,
1278 BdrvTrackedRequest
*req
, int64_t offset
, unsigned int bytes
,
1279 int64_t align
, QEMUIOVector
*qiov
, int flags
)
1281 BlockDriver
*drv
= bs
->drv
;
1285 int64_t start_sector
= offset
>> BDRV_SECTOR_BITS
;
1286 int64_t end_sector
= DIV_ROUND_UP(offset
+ bytes
, BDRV_SECTOR_SIZE
);
1287 uint64_t bytes_remaining
= bytes
;
1290 assert(is_power_of_2(align
));
1291 assert((offset
& (align
- 1)) == 0);
1292 assert((bytes
& (align
- 1)) == 0);
1293 assert(!qiov
|| bytes
== qiov
->size
);
1294 assert((bs
->open_flags
& BDRV_O_NO_IO
) == 0);
1295 assert(!(flags
& ~BDRV_REQ_MASK
));
1296 max_transfer
= QEMU_ALIGN_DOWN(MIN_NON_ZERO(bs
->bl
.max_transfer
, INT_MAX
),
1299 waited
= wait_serialising_requests(req
);
1300 assert(!waited
|| !req
->serialising
);
1301 assert(req
->overlap_offset
<= offset
);
1302 assert(offset
+ bytes
<= req
->overlap_offset
+ req
->overlap_bytes
);
1304 ret
= notifier_with_return_list_notify(&bs
->before_write_notifiers
, req
);
1306 if (!ret
&& bs
->detect_zeroes
!= BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF
&&
1307 !(flags
& BDRV_REQ_ZERO_WRITE
) && drv
->bdrv_co_pwrite_zeroes
&&
1308 qemu_iovec_is_zero(qiov
)) {
1309 flags
|= BDRV_REQ_ZERO_WRITE
;
1310 if (bs
->detect_zeroes
== BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP
) {
1311 flags
|= BDRV_REQ_MAY_UNMAP
;
1316 /* Do nothing, write notifier decided to fail this request */
1317 } else if (flags
& BDRV_REQ_ZERO_WRITE
) {
1318 bdrv_debug_event(bs
, BLKDBG_PWRITEV_ZERO
);
1319 ret
= bdrv_co_do_pwrite_zeroes(bs
, offset
, bytes
, flags
);
1320 } else if (flags
& BDRV_REQ_WRITE_COMPRESSED
) {
1321 ret
= bdrv_driver_pwritev_compressed(bs
, offset
, bytes
, qiov
);
1322 } else if (bytes
<= max_transfer
) {
1323 bdrv_debug_event(bs
, BLKDBG_PWRITEV
);
1324 ret
= bdrv_driver_pwritev(bs
, offset
, bytes
, qiov
, flags
);
1326 bdrv_debug_event(bs
, BLKDBG_PWRITEV
);
1327 while (bytes_remaining
) {
1328 int num
= MIN(bytes_remaining
, max_transfer
);
1329 QEMUIOVector local_qiov
;
1330 int local_flags
= flags
;
1333 if (num
< bytes_remaining
&& (flags
& BDRV_REQ_FUA
) &&
1334 !(bs
->supported_write_flags
& BDRV_REQ_FUA
)) {
1335 /* If FUA is going to be emulated by flush, we only
1336 * need to flush on the last iteration */
1337 local_flags
&= ~BDRV_REQ_FUA
;
1339 qemu_iovec_init(&local_qiov
, qiov
->niov
);
1340 qemu_iovec_concat(&local_qiov
, qiov
, bytes
- bytes_remaining
, num
);
1342 ret
= bdrv_driver_pwritev(bs
, offset
+ bytes
- bytes_remaining
,
1343 num
, &local_qiov
, local_flags
);
1344 qemu_iovec_destroy(&local_qiov
);
1348 bytes_remaining
-= num
;
1351 bdrv_debug_event(bs
, BLKDBG_PWRITEV_DONE
);
1354 bdrv_set_dirty(bs
, start_sector
, end_sector
- start_sector
);
1356 if (bs
->wr_highest_offset
< offset
+ bytes
) {
1357 bs
->wr_highest_offset
= offset
+ bytes
;
1361 bs
->total_sectors
= MAX(bs
->total_sectors
, end_sector
);
1368 static int coroutine_fn
bdrv_co_do_zero_pwritev(BlockDriverState
*bs
,
1371 BdrvRequestFlags flags
,
1372 BdrvTrackedRequest
*req
)
1374 uint8_t *buf
= NULL
;
1375 QEMUIOVector local_qiov
;
1377 uint64_t align
= bs
->bl
.request_alignment
;
1378 unsigned int head_padding_bytes
, tail_padding_bytes
;
1381 head_padding_bytes
= offset
& (align
- 1);
1382 tail_padding_bytes
= align
- ((offset
+ bytes
) & (align
- 1));
1385 assert(flags
& BDRV_REQ_ZERO_WRITE
);
1386 if (head_padding_bytes
|| tail_padding_bytes
) {
1387 buf
= qemu_blockalign(bs
, align
);
1388 iov
= (struct iovec
) {
1392 qemu_iovec_init_external(&local_qiov
, &iov
, 1);
1394 if (head_padding_bytes
) {
1395 uint64_t zero_bytes
= MIN(bytes
, align
- head_padding_bytes
);
1397 /* RMW the unaligned part before head. */
1398 mark_request_serialising(req
, align
);
1399 wait_serialising_requests(req
);
1400 bdrv_debug_event(bs
, BLKDBG_PWRITEV_RMW_HEAD
);
1401 ret
= bdrv_aligned_preadv(bs
, req
, offset
& ~(align
- 1), align
,
1402 align
, &local_qiov
, 0);
1406 bdrv_debug_event(bs
, BLKDBG_PWRITEV_RMW_AFTER_HEAD
);
1408 memset(buf
+ head_padding_bytes
, 0, zero_bytes
);
1409 ret
= bdrv_aligned_pwritev(bs
, req
, offset
& ~(align
- 1), align
,
1411 flags
& ~BDRV_REQ_ZERO_WRITE
);
1415 offset
+= zero_bytes
;
1416 bytes
-= zero_bytes
;
1419 assert(!bytes
|| (offset
& (align
- 1)) == 0);
1420 if (bytes
>= align
) {
1421 /* Write the aligned part in the middle. */
1422 uint64_t aligned_bytes
= bytes
& ~(align
- 1);
1423 ret
= bdrv_aligned_pwritev(bs
, req
, offset
, aligned_bytes
, align
,
1428 bytes
-= aligned_bytes
;
1429 offset
+= aligned_bytes
;
1432 assert(!bytes
|| (offset
& (align
- 1)) == 0);
1434 assert(align
== tail_padding_bytes
+ bytes
);
1435 /* RMW the unaligned part after tail. */
1436 mark_request_serialising(req
, align
);
1437 wait_serialising_requests(req
);
1438 bdrv_debug_event(bs
, BLKDBG_PWRITEV_RMW_TAIL
);
1439 ret
= bdrv_aligned_preadv(bs
, req
, offset
, align
,
1440 align
, &local_qiov
, 0);
1444 bdrv_debug_event(bs
, BLKDBG_PWRITEV_RMW_AFTER_TAIL
);
1446 memset(buf
, 0, bytes
);
1447 ret
= bdrv_aligned_pwritev(bs
, req
, offset
, align
, align
,
1448 &local_qiov
, flags
& ~BDRV_REQ_ZERO_WRITE
);
1457 * Handle a write request in coroutine context
1459 int coroutine_fn
bdrv_co_pwritev(BdrvChild
*child
,
1460 int64_t offset
, unsigned int bytes
, QEMUIOVector
*qiov
,
1461 BdrvRequestFlags flags
)
1463 BlockDriverState
*bs
= child
->bs
;
1464 BdrvTrackedRequest req
;
1465 uint64_t align
= bs
->bl
.request_alignment
;
1466 uint8_t *head_buf
= NULL
;
1467 uint8_t *tail_buf
= NULL
;
1468 QEMUIOVector local_qiov
;
1469 bool use_local_qiov
= false;
1475 if (bs
->read_only
) {
1478 assert(!(bs
->open_flags
& BDRV_O_INACTIVE
));
1480 ret
= bdrv_check_byte_request(bs
, offset
, bytes
);
1486 * Align write if necessary by performing a read-modify-write cycle.
1487 * Pad qiov with the read parts and be sure to have a tracked request not
1488 * only for bdrv_aligned_pwritev, but also for the reads of the RMW cycle.
1490 tracked_request_begin(&req
, bs
, offset
, bytes
, BDRV_TRACKED_WRITE
);
1493 ret
= bdrv_co_do_zero_pwritev(bs
, offset
, bytes
, flags
, &req
);
1497 if (offset
& (align
- 1)) {
1498 QEMUIOVector head_qiov
;
1499 struct iovec head_iov
;
1501 mark_request_serialising(&req
, align
);
1502 wait_serialising_requests(&req
);
1504 head_buf
= qemu_blockalign(bs
, align
);
1505 head_iov
= (struct iovec
) {
1506 .iov_base
= head_buf
,
1509 qemu_iovec_init_external(&head_qiov
, &head_iov
, 1);
1511 bdrv_debug_event(bs
, BLKDBG_PWRITEV_RMW_HEAD
);
1512 ret
= bdrv_aligned_preadv(bs
, &req
, offset
& ~(align
- 1), align
,
1513 align
, &head_qiov
, 0);
1517 bdrv_debug_event(bs
, BLKDBG_PWRITEV_RMW_AFTER_HEAD
);
1519 qemu_iovec_init(&local_qiov
, qiov
->niov
+ 2);
1520 qemu_iovec_add(&local_qiov
, head_buf
, offset
& (align
- 1));
1521 qemu_iovec_concat(&local_qiov
, qiov
, 0, qiov
->size
);
1522 use_local_qiov
= true;
1524 bytes
+= offset
& (align
- 1);
1525 offset
= offset
& ~(align
- 1);
1527 /* We have read the tail already if the request is smaller
1528 * than one aligned block.
1530 if (bytes
< align
) {
1531 qemu_iovec_add(&local_qiov
, head_buf
+ bytes
, align
- bytes
);
1536 if ((offset
+ bytes
) & (align
- 1)) {
1537 QEMUIOVector tail_qiov
;
1538 struct iovec tail_iov
;
1542 mark_request_serialising(&req
, align
);
1543 waited
= wait_serialising_requests(&req
);
1544 assert(!waited
|| !use_local_qiov
);
1546 tail_buf
= qemu_blockalign(bs
, align
);
1547 tail_iov
= (struct iovec
) {
1548 .iov_base
= tail_buf
,
1551 qemu_iovec_init_external(&tail_qiov
, &tail_iov
, 1);
1553 bdrv_debug_event(bs
, BLKDBG_PWRITEV_RMW_TAIL
);
1554 ret
= bdrv_aligned_preadv(bs
, &req
, (offset
+ bytes
) & ~(align
- 1), align
,
1555 align
, &tail_qiov
, 0);
1559 bdrv_debug_event(bs
, BLKDBG_PWRITEV_RMW_AFTER_TAIL
);
1561 if (!use_local_qiov
) {
1562 qemu_iovec_init(&local_qiov
, qiov
->niov
+ 1);
1563 qemu_iovec_concat(&local_qiov
, qiov
, 0, qiov
->size
);
1564 use_local_qiov
= true;
1567 tail_bytes
= (offset
+ bytes
) & (align
- 1);
1568 qemu_iovec_add(&local_qiov
, tail_buf
+ tail_bytes
, align
- tail_bytes
);
1570 bytes
= ROUND_UP(bytes
, align
);
1573 ret
= bdrv_aligned_pwritev(bs
, &req
, offset
, bytes
, align
,
1574 use_local_qiov
? &local_qiov
: qiov
,
1579 if (use_local_qiov
) {
1580 qemu_iovec_destroy(&local_qiov
);
1582 qemu_vfree(head_buf
);
1583 qemu_vfree(tail_buf
);
1585 tracked_request_end(&req
);
1589 static int coroutine_fn
bdrv_co_do_writev(BdrvChild
*child
,
1590 int64_t sector_num
, int nb_sectors
, QEMUIOVector
*qiov
,
1591 BdrvRequestFlags flags
)
1593 if (nb_sectors
< 0 || nb_sectors
> BDRV_REQUEST_MAX_SECTORS
) {
1597 return bdrv_co_pwritev(child
, sector_num
<< BDRV_SECTOR_BITS
,
1598 nb_sectors
<< BDRV_SECTOR_BITS
, qiov
, flags
);
1601 int coroutine_fn
bdrv_co_writev(BdrvChild
*child
, int64_t sector_num
,
1602 int nb_sectors
, QEMUIOVector
*qiov
)
1604 trace_bdrv_co_writev(child
->bs
, sector_num
, nb_sectors
);
1606 return bdrv_co_do_writev(child
, sector_num
, nb_sectors
, qiov
, 0);
1609 int coroutine_fn
bdrv_co_pwrite_zeroes(BdrvChild
*child
, int64_t offset
,
1610 int count
, BdrvRequestFlags flags
)
1612 trace_bdrv_co_pwrite_zeroes(child
->bs
, offset
, count
, flags
);
1614 if (!(child
->bs
->open_flags
& BDRV_O_UNMAP
)) {
1615 flags
&= ~BDRV_REQ_MAY_UNMAP
;
1618 return bdrv_co_pwritev(child
, offset
, count
, NULL
,
1619 BDRV_REQ_ZERO_WRITE
| flags
);
1623 * Flush ALL BDSes regardless of if they are reachable via a BlkBackend or not.
1625 int bdrv_flush_all(void)
1627 BdrvNextIterator it
;
1628 BlockDriverState
*bs
= NULL
;
1631 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
1632 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
1635 aio_context_acquire(aio_context
);
1636 ret
= bdrv_flush(bs
);
1637 if (ret
< 0 && !result
) {
1640 aio_context_release(aio_context
);
1647 typedef struct BdrvCoGetBlockStatusData
{
1648 BlockDriverState
*bs
;
1649 BlockDriverState
*base
;
1650 BlockDriverState
**file
;
1656 } BdrvCoGetBlockStatusData
;
1659 * Returns the allocation status of the specified sectors.
1660 * Drivers not implementing the functionality are assumed to not support
1661 * backing files, hence all their sectors are reported as allocated.
1663 * If 'sector_num' is beyond the end of the disk image the return value is 0
1664 * and 'pnum' is set to 0.
1666 * 'pnum' is set to the number of sectors (including and immediately following
1667 * the specified sector) that are known to be in the same
1668 * allocated/unallocated state.
1670 * 'nb_sectors' is the max value 'pnum' should be set to. If nb_sectors goes
1671 * beyond the end of the disk image it will be clamped.
1673 * If returned value is positive and BDRV_BLOCK_OFFSET_VALID bit is set, 'file'
1674 * points to the BDS which the sector range is allocated in.
1676 static int64_t coroutine_fn
bdrv_co_get_block_status(BlockDriverState
*bs
,
1678 int nb_sectors
, int *pnum
,
1679 BlockDriverState
**file
)
1681 int64_t total_sectors
;
1685 total_sectors
= bdrv_nb_sectors(bs
);
1686 if (total_sectors
< 0) {
1687 return total_sectors
;
1690 if (sector_num
>= total_sectors
) {
1695 n
= total_sectors
- sector_num
;
1696 if (n
< nb_sectors
) {
1700 if (!bs
->drv
->bdrv_co_get_block_status
) {
1702 ret
= BDRV_BLOCK_DATA
| BDRV_BLOCK_ALLOCATED
;
1703 if (bs
->drv
->protocol_name
) {
1704 ret
|= BDRV_BLOCK_OFFSET_VALID
| (sector_num
* BDRV_SECTOR_SIZE
);
1710 ret
= bs
->drv
->bdrv_co_get_block_status(bs
, sector_num
, nb_sectors
, pnum
,
1717 if (ret
& BDRV_BLOCK_RAW
) {
1718 assert(ret
& BDRV_BLOCK_OFFSET_VALID
);
1719 return bdrv_get_block_status(bs
->file
->bs
, ret
>> BDRV_SECTOR_BITS
,
1723 if (ret
& (BDRV_BLOCK_DATA
| BDRV_BLOCK_ZERO
)) {
1724 ret
|= BDRV_BLOCK_ALLOCATED
;
1726 if (bdrv_unallocated_blocks_are_zero(bs
)) {
1727 ret
|= BDRV_BLOCK_ZERO
;
1728 } else if (bs
->backing
) {
1729 BlockDriverState
*bs2
= bs
->backing
->bs
;
1730 int64_t nb_sectors2
= bdrv_nb_sectors(bs2
);
1731 if (nb_sectors2
>= 0 && sector_num
>= nb_sectors2
) {
1732 ret
|= BDRV_BLOCK_ZERO
;
1737 if (*file
&& *file
!= bs
&&
1738 (ret
& BDRV_BLOCK_DATA
) && !(ret
& BDRV_BLOCK_ZERO
) &&
1739 (ret
& BDRV_BLOCK_OFFSET_VALID
)) {
1740 BlockDriverState
*file2
;
1743 ret2
= bdrv_co_get_block_status(*file
, ret
>> BDRV_SECTOR_BITS
,
1744 *pnum
, &file_pnum
, &file2
);
1746 /* Ignore errors. This is just providing extra information, it
1747 * is useful but not necessary.
1750 /* !file_pnum indicates an offset at or beyond the EOF; it is
1751 * perfectly valid for the format block driver to point to such
1752 * offsets, so catch it and mark everything as zero */
1753 ret
|= BDRV_BLOCK_ZERO
;
1755 /* Limit request to the range reported by the protocol driver */
1757 ret
|= (ret2
& BDRV_BLOCK_ZERO
);
1765 static int64_t coroutine_fn
bdrv_co_get_block_status_above(BlockDriverState
*bs
,
1766 BlockDriverState
*base
,
1770 BlockDriverState
**file
)
1772 BlockDriverState
*p
;
1776 for (p
= bs
; p
!= base
; p
= backing_bs(p
)) {
1777 ret
= bdrv_co_get_block_status(p
, sector_num
, nb_sectors
, pnum
, file
);
1778 if (ret
< 0 || ret
& BDRV_BLOCK_ALLOCATED
) {
1781 /* [sector_num, pnum] unallocated on this layer, which could be only
1782 * the first part of [sector_num, nb_sectors]. */
1783 nb_sectors
= MIN(nb_sectors
, *pnum
);
1788 /* Coroutine wrapper for bdrv_get_block_status_above() */
1789 static void coroutine_fn
bdrv_get_block_status_above_co_entry(void *opaque
)
1791 BdrvCoGetBlockStatusData
*data
= opaque
;
1793 data
->ret
= bdrv_co_get_block_status_above(data
->bs
, data
->base
,
1802 * Synchronous wrapper around bdrv_co_get_block_status_above().
1804 * See bdrv_co_get_block_status_above() for details.
1806 int64_t bdrv_get_block_status_above(BlockDriverState
*bs
,
1807 BlockDriverState
*base
,
1809 int nb_sectors
, int *pnum
,
1810 BlockDriverState
**file
)
1813 BdrvCoGetBlockStatusData data
= {
1817 .sector_num
= sector_num
,
1818 .nb_sectors
= nb_sectors
,
1823 if (qemu_in_coroutine()) {
1824 /* Fast-path if already in coroutine context */
1825 bdrv_get_block_status_above_co_entry(&data
);
1827 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
1829 co
= qemu_coroutine_create(bdrv_get_block_status_above_co_entry
,
1831 qemu_coroutine_enter(co
);
1832 while (!data
.done
) {
1833 aio_poll(aio_context
, true);
1839 int64_t bdrv_get_block_status(BlockDriverState
*bs
,
1841 int nb_sectors
, int *pnum
,
1842 BlockDriverState
**file
)
1844 return bdrv_get_block_status_above(bs
, backing_bs(bs
),
1845 sector_num
, nb_sectors
, pnum
, file
);
1848 int coroutine_fn
bdrv_is_allocated(BlockDriverState
*bs
, int64_t sector_num
,
1849 int nb_sectors
, int *pnum
)
1851 BlockDriverState
*file
;
1852 int64_t ret
= bdrv_get_block_status(bs
, sector_num
, nb_sectors
, pnum
,
1857 return !!(ret
& BDRV_BLOCK_ALLOCATED
);
1861 * Given an image chain: ... -> [BASE] -> [INTER1] -> [INTER2] -> [TOP]
1863 * Return true if the given sector is allocated in any image between
1864 * BASE and TOP (inclusive). BASE can be NULL to check if the given
1865 * sector is allocated in any image of the chain. Return false otherwise.
1867 * 'pnum' is set to the number of sectors (including and immediately following
1868 * the specified sector) that are known to be in the same
1869 * allocated/unallocated state.
1872 int bdrv_is_allocated_above(BlockDriverState
*top
,
1873 BlockDriverState
*base
,
1875 int nb_sectors
, int *pnum
)
1877 BlockDriverState
*intermediate
;
1878 int ret
, n
= nb_sectors
;
1881 while (intermediate
&& intermediate
!= base
) {
1883 ret
= bdrv_is_allocated(intermediate
, sector_num
, nb_sectors
,
1893 * [sector_num, nb_sectors] is unallocated on top but intermediate
1896 * [sector_num+x, nr_sectors] allocated.
1898 if (n
> pnum_inter
&&
1899 (intermediate
== top
||
1900 sector_num
+ pnum_inter
< intermediate
->total_sectors
)) {
1904 intermediate
= backing_bs(intermediate
);
1911 typedef struct BdrvVmstateCo
{
1912 BlockDriverState
*bs
;
1919 static int coroutine_fn
1920 bdrv_co_rw_vmstate(BlockDriverState
*bs
, QEMUIOVector
*qiov
, int64_t pos
,
1923 BlockDriver
*drv
= bs
->drv
;
1927 } else if (drv
->bdrv_load_vmstate
) {
1928 return is_read
? drv
->bdrv_load_vmstate(bs
, qiov
, pos
)
1929 : drv
->bdrv_save_vmstate(bs
, qiov
, pos
);
1930 } else if (bs
->file
) {
1931 return bdrv_co_rw_vmstate(bs
->file
->bs
, qiov
, pos
, is_read
);
1937 static void coroutine_fn
bdrv_co_rw_vmstate_entry(void *opaque
)
1939 BdrvVmstateCo
*co
= opaque
;
1940 co
->ret
= bdrv_co_rw_vmstate(co
->bs
, co
->qiov
, co
->pos
, co
->is_read
);
1944 bdrv_rw_vmstate(BlockDriverState
*bs
, QEMUIOVector
*qiov
, int64_t pos
,
1947 if (qemu_in_coroutine()) {
1948 return bdrv_co_rw_vmstate(bs
, qiov
, pos
, is_read
);
1950 BdrvVmstateCo data
= {
1955 .ret
= -EINPROGRESS
,
1957 Coroutine
*co
= qemu_coroutine_create(bdrv_co_rw_vmstate_entry
, &data
);
1959 qemu_coroutine_enter(co
);
1960 while (data
.ret
== -EINPROGRESS
) {
1961 aio_poll(bdrv_get_aio_context(bs
), true);
1967 int bdrv_save_vmstate(BlockDriverState
*bs
, const uint8_t *buf
,
1968 int64_t pos
, int size
)
1971 struct iovec iov
= {
1972 .iov_base
= (void *) buf
,
1977 qemu_iovec_init_external(&qiov
, &iov
, 1);
1979 ret
= bdrv_writev_vmstate(bs
, &qiov
, pos
);
1987 int bdrv_writev_vmstate(BlockDriverState
*bs
, QEMUIOVector
*qiov
, int64_t pos
)
1989 return bdrv_rw_vmstate(bs
, qiov
, pos
, false);
1992 int bdrv_load_vmstate(BlockDriverState
*bs
, uint8_t *buf
,
1993 int64_t pos
, int size
)
1996 struct iovec iov
= {
2002 qemu_iovec_init_external(&qiov
, &iov
, 1);
2003 ret
= bdrv_readv_vmstate(bs
, &qiov
, pos
);
2011 int bdrv_readv_vmstate(BlockDriverState
*bs
, QEMUIOVector
*qiov
, int64_t pos
)
2013 return bdrv_rw_vmstate(bs
, qiov
, pos
, true);
2016 /**************************************************************/
2019 BlockAIOCB
*bdrv_aio_readv(BdrvChild
*child
, int64_t sector_num
,
2020 QEMUIOVector
*qiov
, int nb_sectors
,
2021 BlockCompletionFunc
*cb
, void *opaque
)
2023 trace_bdrv_aio_readv(child
->bs
, sector_num
, nb_sectors
, opaque
);
2025 assert(nb_sectors
<< BDRV_SECTOR_BITS
== qiov
->size
);
2026 return bdrv_co_aio_prw_vector(child
, sector_num
<< BDRV_SECTOR_BITS
, qiov
,
2027 0, cb
, opaque
, false);
2030 BlockAIOCB
*bdrv_aio_writev(BdrvChild
*child
, int64_t sector_num
,
2031 QEMUIOVector
*qiov
, int nb_sectors
,
2032 BlockCompletionFunc
*cb
, void *opaque
)
2034 trace_bdrv_aio_writev(child
->bs
, sector_num
, nb_sectors
, opaque
);
2036 assert(nb_sectors
<< BDRV_SECTOR_BITS
== qiov
->size
);
2037 return bdrv_co_aio_prw_vector(child
, sector_num
<< BDRV_SECTOR_BITS
, qiov
,
2038 0, cb
, opaque
, true);
2041 void bdrv_aio_cancel(BlockAIOCB
*acb
)
2044 bdrv_aio_cancel_async(acb
);
2045 while (acb
->refcnt
> 1) {
2046 if (acb
->aiocb_info
->get_aio_context
) {
2047 aio_poll(acb
->aiocb_info
->get_aio_context(acb
), true);
2048 } else if (acb
->bs
) {
2049 aio_poll(bdrv_get_aio_context(acb
->bs
), true);
2054 qemu_aio_unref(acb
);
2057 /* Async version of aio cancel. The caller is not blocked if the acb implements
2058 * cancel_async, otherwise we do nothing and let the request normally complete.
2059 * In either case the completion callback must be called. */
2060 void bdrv_aio_cancel_async(BlockAIOCB
*acb
)
2062 if (acb
->aiocb_info
->cancel_async
) {
2063 acb
->aiocb_info
->cancel_async(acb
);
2067 /**************************************************************/
2068 /* async block device emulation */
2070 typedef struct BlockRequest
{
2072 /* Used during read, write, trim */
2079 /* Used during ioctl */
2085 BlockCompletionFunc
*cb
;
2091 typedef struct BlockAIOCBCoroutine
{
2099 } BlockAIOCBCoroutine
;
2101 static const AIOCBInfo bdrv_em_co_aiocb_info
= {
2102 .aiocb_size
= sizeof(BlockAIOCBCoroutine
),
2105 static void bdrv_co_complete(BlockAIOCBCoroutine
*acb
)
2107 if (!acb
->need_bh
) {
2108 acb
->common
.cb(acb
->common
.opaque
, acb
->req
.error
);
2109 qemu_aio_unref(acb
);
2113 static void bdrv_co_em_bh(void *opaque
)
2115 BlockAIOCBCoroutine
*acb
= opaque
;
2117 assert(!acb
->need_bh
);
2118 qemu_bh_delete(acb
->bh
);
2119 bdrv_co_complete(acb
);
2122 static void bdrv_co_maybe_schedule_bh(BlockAIOCBCoroutine
*acb
)
2124 acb
->need_bh
= false;
2125 if (acb
->req
.error
!= -EINPROGRESS
) {
2126 BlockDriverState
*bs
= acb
->common
.bs
;
2128 acb
->bh
= aio_bh_new(bdrv_get_aio_context(bs
), bdrv_co_em_bh
, acb
);
2129 qemu_bh_schedule(acb
->bh
);
2133 /* Invoke bdrv_co_do_readv/bdrv_co_do_writev */
2134 static void coroutine_fn
bdrv_co_do_rw(void *opaque
)
2136 BlockAIOCBCoroutine
*acb
= opaque
;
2138 if (!acb
->is_write
) {
2139 acb
->req
.error
= bdrv_co_preadv(acb
->child
, acb
->req
.offset
,
2140 acb
->req
.qiov
->size
, acb
->req
.qiov
, acb
->req
.flags
);
2142 acb
->req
.error
= bdrv_co_pwritev(acb
->child
, acb
->req
.offset
,
2143 acb
->req
.qiov
->size
, acb
->req
.qiov
, acb
->req
.flags
);
2146 bdrv_co_complete(acb
);
2149 static BlockAIOCB
*bdrv_co_aio_prw_vector(BdrvChild
*child
,
2152 BdrvRequestFlags flags
,
2153 BlockCompletionFunc
*cb
,
2158 BlockAIOCBCoroutine
*acb
;
2160 acb
= qemu_aio_get(&bdrv_em_co_aiocb_info
, child
->bs
, cb
, opaque
);
2162 acb
->need_bh
= true;
2163 acb
->req
.error
= -EINPROGRESS
;
2164 acb
->req
.offset
= offset
;
2165 acb
->req
.qiov
= qiov
;
2166 acb
->req
.flags
= flags
;
2167 acb
->is_write
= is_write
;
2169 co
= qemu_coroutine_create(bdrv_co_do_rw
, acb
);
2170 qemu_coroutine_enter(co
);
2172 bdrv_co_maybe_schedule_bh(acb
);
2173 return &acb
->common
;
2176 static void coroutine_fn
bdrv_aio_flush_co_entry(void *opaque
)
2178 BlockAIOCBCoroutine
*acb
= opaque
;
2179 BlockDriverState
*bs
= acb
->common
.bs
;
2181 acb
->req
.error
= bdrv_co_flush(bs
);
2182 bdrv_co_complete(acb
);
2185 BlockAIOCB
*bdrv_aio_flush(BlockDriverState
*bs
,
2186 BlockCompletionFunc
*cb
, void *opaque
)
2188 trace_bdrv_aio_flush(bs
, opaque
);
2191 BlockAIOCBCoroutine
*acb
;
2193 acb
= qemu_aio_get(&bdrv_em_co_aiocb_info
, bs
, cb
, opaque
);
2194 acb
->need_bh
= true;
2195 acb
->req
.error
= -EINPROGRESS
;
2197 co
= qemu_coroutine_create(bdrv_aio_flush_co_entry
, acb
);
2198 qemu_coroutine_enter(co
);
2200 bdrv_co_maybe_schedule_bh(acb
);
2201 return &acb
->common
;
2204 static void coroutine_fn
bdrv_aio_pdiscard_co_entry(void *opaque
)
2206 BlockAIOCBCoroutine
*acb
= opaque
;
2207 BlockDriverState
*bs
= acb
->common
.bs
;
2209 acb
->req
.error
= bdrv_co_pdiscard(bs
, acb
->req
.offset
, acb
->req
.bytes
);
2210 bdrv_co_complete(acb
);
2213 BlockAIOCB
*bdrv_aio_pdiscard(BlockDriverState
*bs
, int64_t offset
, int count
,
2214 BlockCompletionFunc
*cb
, void *opaque
)
2217 BlockAIOCBCoroutine
*acb
;
2219 trace_bdrv_aio_pdiscard(bs
, offset
, count
, opaque
);
2221 acb
= qemu_aio_get(&bdrv_em_co_aiocb_info
, bs
, cb
, opaque
);
2222 acb
->need_bh
= true;
2223 acb
->req
.error
= -EINPROGRESS
;
2224 acb
->req
.offset
= offset
;
2225 acb
->req
.bytes
= count
;
2226 co
= qemu_coroutine_create(bdrv_aio_pdiscard_co_entry
, acb
);
2227 qemu_coroutine_enter(co
);
2229 bdrv_co_maybe_schedule_bh(acb
);
2230 return &acb
->common
;
2233 void *qemu_aio_get(const AIOCBInfo
*aiocb_info
, BlockDriverState
*bs
,
2234 BlockCompletionFunc
*cb
, void *opaque
)
2238 acb
= g_malloc(aiocb_info
->aiocb_size
);
2239 acb
->aiocb_info
= aiocb_info
;
2242 acb
->opaque
= opaque
;
2247 void qemu_aio_ref(void *p
)
2249 BlockAIOCB
*acb
= p
;
2253 void qemu_aio_unref(void *p
)
2255 BlockAIOCB
*acb
= p
;
2256 assert(acb
->refcnt
> 0);
2257 if (--acb
->refcnt
== 0) {
2262 /**************************************************************/
2263 /* Coroutine block device emulation */
2265 typedef struct FlushCo
{
2266 BlockDriverState
*bs
;
2271 static void coroutine_fn
bdrv_flush_co_entry(void *opaque
)
2273 FlushCo
*rwco
= opaque
;
2275 rwco
->ret
= bdrv_co_flush(rwco
->bs
);
2278 int coroutine_fn
bdrv_co_flush(BlockDriverState
*bs
)
2281 BdrvTrackedRequest req
;
2283 if (!bs
|| !bdrv_is_inserted(bs
) || bdrv_is_read_only(bs
) ||
2288 tracked_request_begin(&req
, bs
, 0, 0, BDRV_TRACKED_FLUSH
);
2290 int current_gen
= bs
->write_gen
;
2292 /* Wait until any previous flushes are completed */
2293 while (bs
->active_flush_req
!= NULL
) {
2294 qemu_co_queue_wait(&bs
->flush_queue
);
2297 bs
->active_flush_req
= &req
;
2299 /* Write back all layers by calling one driver function */
2300 if (bs
->drv
->bdrv_co_flush
) {
2301 ret
= bs
->drv
->bdrv_co_flush(bs
);
2305 /* Write back cached data to the OS even with cache=unsafe */
2306 BLKDBG_EVENT(bs
->file
, BLKDBG_FLUSH_TO_OS
);
2307 if (bs
->drv
->bdrv_co_flush_to_os
) {
2308 ret
= bs
->drv
->bdrv_co_flush_to_os(bs
);
2314 /* But don't actually force it to the disk with cache=unsafe */
2315 if (bs
->open_flags
& BDRV_O_NO_FLUSH
) {
2319 /* Check if we really need to flush anything */
2320 if (bs
->flushed_gen
== current_gen
) {
2324 BLKDBG_EVENT(bs
->file
, BLKDBG_FLUSH_TO_DISK
);
2325 if (bs
->drv
->bdrv_co_flush_to_disk
) {
2326 ret
= bs
->drv
->bdrv_co_flush_to_disk(bs
);
2327 } else if (bs
->drv
->bdrv_aio_flush
) {
2329 CoroutineIOCompletion co
= {
2330 .coroutine
= qemu_coroutine_self(),
2333 acb
= bs
->drv
->bdrv_aio_flush(bs
, bdrv_co_io_em_complete
, &co
);
2337 qemu_coroutine_yield();
2342 * Some block drivers always operate in either writethrough or unsafe
2343 * mode and don't support bdrv_flush therefore. Usually qemu doesn't
2344 * know how the server works (because the behaviour is hardcoded or
2345 * depends on server-side configuration), so we can't ensure that
2346 * everything is safe on disk. Returning an error doesn't work because
2347 * that would break guests even if the server operates in writethrough
2350 * Let's hope the user knows what he's doing.
2359 /* Now flush the underlying protocol. It will also have BDRV_O_NO_FLUSH
2360 * in the case of cache=unsafe, so there are no useless flushes.
2363 ret
= bs
->file
? bdrv_co_flush(bs
->file
->bs
) : 0;
2365 /* Notify any pending flushes that we have completed */
2366 bs
->flushed_gen
= current_gen
;
2367 bs
->active_flush_req
= NULL
;
2368 /* Return value is ignored - it's ok if wait queue is empty */
2369 qemu_co_queue_next(&bs
->flush_queue
);
2371 tracked_request_end(&req
);
2375 int bdrv_flush(BlockDriverState
*bs
)
2378 FlushCo flush_co
= {
2383 if (qemu_in_coroutine()) {
2384 /* Fast-path if already in coroutine context */
2385 bdrv_flush_co_entry(&flush_co
);
2387 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
2389 co
= qemu_coroutine_create(bdrv_flush_co_entry
, &flush_co
);
2390 qemu_coroutine_enter(co
);
2391 while (flush_co
.ret
== NOT_DONE
) {
2392 aio_poll(aio_context
, true);
2396 return flush_co
.ret
;
2399 typedef struct DiscardCo
{
2400 BlockDriverState
*bs
;
2405 static void coroutine_fn
bdrv_pdiscard_co_entry(void *opaque
)
2407 DiscardCo
*rwco
= opaque
;
2409 rwco
->ret
= bdrv_co_pdiscard(rwco
->bs
, rwco
->offset
, rwco
->count
);
2412 int coroutine_fn
bdrv_co_pdiscard(BlockDriverState
*bs
, int64_t offset
,
2415 BdrvTrackedRequest req
;
2416 int max_pdiscard
, ret
;
2423 ret
= bdrv_check_byte_request(bs
, offset
, count
);
2426 } else if (bs
->read_only
) {
2429 assert(!(bs
->open_flags
& BDRV_O_INACTIVE
));
2431 /* Do nothing if disabled. */
2432 if (!(bs
->open_flags
& BDRV_O_UNMAP
)) {
2436 if (!bs
->drv
->bdrv_co_pdiscard
&& !bs
->drv
->bdrv_aio_pdiscard
) {
2440 /* Discard is advisory, so ignore any unaligned head or tail */
2441 align
= MAX(bs
->bl
.pdiscard_alignment
, bs
->bl
.request_alignment
);
2442 assert(align
% bs
->bl
.request_alignment
== 0);
2443 head
= offset
% align
;
2445 head
= MIN(count
, align
- head
);
2449 count
= QEMU_ALIGN_DOWN(count
, align
);
2454 tracked_request_begin(&req
, bs
, offset
, count
, BDRV_TRACKED_DISCARD
);
2456 ret
= notifier_with_return_list_notify(&bs
->before_write_notifiers
, &req
);
2461 max_pdiscard
= QEMU_ALIGN_DOWN(MIN_NON_ZERO(bs
->bl
.max_pdiscard
, INT_MAX
),
2463 assert(max_pdiscard
);
2467 int num
= MIN(count
, max_pdiscard
);
2469 if (bs
->drv
->bdrv_co_pdiscard
) {
2470 ret
= bs
->drv
->bdrv_co_pdiscard(bs
, offset
, num
);
2473 CoroutineIOCompletion co
= {
2474 .coroutine
= qemu_coroutine_self(),
2477 acb
= bs
->drv
->bdrv_aio_pdiscard(bs
, offset
, num
,
2478 bdrv_co_io_em_complete
, &co
);
2483 qemu_coroutine_yield();
2487 if (ret
&& ret
!= -ENOTSUP
) {
2497 bdrv_set_dirty(bs
, req
.offset
>> BDRV_SECTOR_BITS
,
2498 req
.bytes
>> BDRV_SECTOR_BITS
);
2499 tracked_request_end(&req
);
2503 int bdrv_pdiscard(BlockDriverState
*bs
, int64_t offset
, int count
)
2513 if (qemu_in_coroutine()) {
2514 /* Fast-path if already in coroutine context */
2515 bdrv_pdiscard_co_entry(&rwco
);
2517 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
2519 co
= qemu_coroutine_create(bdrv_pdiscard_co_entry
, &rwco
);
2520 qemu_coroutine_enter(co
);
2521 while (rwco
.ret
== NOT_DONE
) {
2522 aio_poll(aio_context
, true);
2529 static int bdrv_co_do_ioctl(BlockDriverState
*bs
, int req
, void *buf
)
2531 BlockDriver
*drv
= bs
->drv
;
2532 BdrvTrackedRequest tracked_req
;
2533 CoroutineIOCompletion co
= {
2534 .coroutine
= qemu_coroutine_self(),
2538 tracked_request_begin(&tracked_req
, bs
, 0, 0, BDRV_TRACKED_IOCTL
);
2539 if (!drv
|| !drv
->bdrv_aio_ioctl
) {
2544 acb
= drv
->bdrv_aio_ioctl(bs
, req
, buf
, bdrv_co_io_em_complete
, &co
);
2549 qemu_coroutine_yield();
2551 tracked_request_end(&tracked_req
);
2556 BlockDriverState
*bs
;
2562 static void coroutine_fn
bdrv_co_ioctl_entry(void *opaque
)
2564 BdrvIoctlCoData
*data
= opaque
;
2565 data
->ret
= bdrv_co_do_ioctl(data
->bs
, data
->req
, data
->buf
);
2568 /* needed for generic scsi interface */
2569 int bdrv_ioctl(BlockDriverState
*bs
, unsigned long int req
, void *buf
)
2571 BdrvIoctlCoData data
= {
2575 .ret
= -EINPROGRESS
,
2578 if (qemu_in_coroutine()) {
2579 /* Fast-path if already in coroutine context */
2580 bdrv_co_ioctl_entry(&data
);
2582 Coroutine
*co
= qemu_coroutine_create(bdrv_co_ioctl_entry
, &data
);
2584 qemu_coroutine_enter(co
);
2585 while (data
.ret
== -EINPROGRESS
) {
2586 aio_poll(bdrv_get_aio_context(bs
), true);
2592 static void coroutine_fn
bdrv_co_aio_ioctl_entry(void *opaque
)
2594 BlockAIOCBCoroutine
*acb
= opaque
;
2595 acb
->req
.error
= bdrv_co_do_ioctl(acb
->common
.bs
,
2596 acb
->req
.req
, acb
->req
.buf
);
2597 bdrv_co_complete(acb
);
2600 BlockAIOCB
*bdrv_aio_ioctl(BlockDriverState
*bs
,
2601 unsigned long int req
, void *buf
,
2602 BlockCompletionFunc
*cb
, void *opaque
)
2604 BlockAIOCBCoroutine
*acb
= qemu_aio_get(&bdrv_em_co_aiocb_info
,
2608 acb
->need_bh
= true;
2609 acb
->req
.error
= -EINPROGRESS
;
2612 co
= qemu_coroutine_create(bdrv_co_aio_ioctl_entry
, acb
);
2613 qemu_coroutine_enter(co
);
2615 bdrv_co_maybe_schedule_bh(acb
);
2616 return &acb
->common
;
2619 void *qemu_blockalign(BlockDriverState
*bs
, size_t size
)
2621 return qemu_memalign(bdrv_opt_mem_align(bs
), size
);
2624 void *qemu_blockalign0(BlockDriverState
*bs
, size_t size
)
2626 return memset(qemu_blockalign(bs
, size
), 0, size
);
2629 void *qemu_try_blockalign(BlockDriverState
*bs
, size_t size
)
2631 size_t align
= bdrv_opt_mem_align(bs
);
2633 /* Ensure that NULL is never returned on success */
2639 return qemu_try_memalign(align
, size
);
2642 void *qemu_try_blockalign0(BlockDriverState
*bs
, size_t size
)
2644 void *mem
= qemu_try_blockalign(bs
, size
);
2647 memset(mem
, 0, size
);
2654 * Check if all memory in this vector is sector aligned.
2656 bool bdrv_qiov_is_aligned(BlockDriverState
*bs
, QEMUIOVector
*qiov
)
2659 size_t alignment
= bdrv_min_mem_align(bs
);
2661 for (i
= 0; i
< qiov
->niov
; i
++) {
2662 if ((uintptr_t) qiov
->iov
[i
].iov_base
% alignment
) {
2665 if (qiov
->iov
[i
].iov_len
% alignment
) {
2673 void bdrv_add_before_write_notifier(BlockDriverState
*bs
,
2674 NotifierWithReturn
*notifier
)
2676 notifier_with_return_list_add(&bs
->before_write_notifiers
, notifier
);
2679 void bdrv_io_plug(BlockDriverState
*bs
)
2683 QLIST_FOREACH(child
, &bs
->children
, next
) {
2684 bdrv_io_plug(child
->bs
);
2687 if (bs
->io_plugged
++ == 0 && bs
->io_plug_disabled
== 0) {
2688 BlockDriver
*drv
= bs
->drv
;
2689 if (drv
&& drv
->bdrv_io_plug
) {
2690 drv
->bdrv_io_plug(bs
);
2695 void bdrv_io_unplug(BlockDriverState
*bs
)
2699 assert(bs
->io_plugged
);
2700 if (--bs
->io_plugged
== 0 && bs
->io_plug_disabled
== 0) {
2701 BlockDriver
*drv
= bs
->drv
;
2702 if (drv
&& drv
->bdrv_io_unplug
) {
2703 drv
->bdrv_io_unplug(bs
);
2707 QLIST_FOREACH(child
, &bs
->children
, next
) {
2708 bdrv_io_unplug(child
->bs
);
2712 void bdrv_io_unplugged_begin(BlockDriverState
*bs
)
2716 if (bs
->io_plug_disabled
++ == 0 && bs
->io_plugged
> 0) {
2717 BlockDriver
*drv
= bs
->drv
;
2718 if (drv
&& drv
->bdrv_io_unplug
) {
2719 drv
->bdrv_io_unplug(bs
);
2723 QLIST_FOREACH(child
, &bs
->children
, next
) {
2724 bdrv_io_unplugged_begin(child
->bs
);
2728 void bdrv_io_unplugged_end(BlockDriverState
*bs
)
2732 assert(bs
->io_plug_disabled
);
2733 QLIST_FOREACH(child
, &bs
->children
, next
) {
2734 bdrv_io_unplugged_end(child
->bs
);
2737 if (--bs
->io_plug_disabled
== 0 && bs
->io_plugged
> 0) {
2738 BlockDriver
*drv
= bs
->drv
;
2739 if (drv
&& drv
->bdrv_io_plug
) {
2740 drv
->bdrv_io_plug(bs
);