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
26 #include "block/blockjob.h"
27 #include "block/block_int.h"
28 #include "block/throttle-groups.h"
29 #include "qemu/error-report.h"
31 #define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */
33 static BlockAIOCB
*bdrv_aio_readv_em(BlockDriverState
*bs
,
34 int64_t sector_num
, QEMUIOVector
*qiov
, int nb_sectors
,
35 BlockCompletionFunc
*cb
, void *opaque
);
36 static BlockAIOCB
*bdrv_aio_writev_em(BlockDriverState
*bs
,
37 int64_t sector_num
, QEMUIOVector
*qiov
, int nb_sectors
,
38 BlockCompletionFunc
*cb
, void *opaque
);
39 static int coroutine_fn
bdrv_co_readv_em(BlockDriverState
*bs
,
40 int64_t sector_num
, int nb_sectors
,
42 static int coroutine_fn
bdrv_co_writev_em(BlockDriverState
*bs
,
43 int64_t sector_num
, int nb_sectors
,
45 static int coroutine_fn
bdrv_co_do_preadv(BlockDriverState
*bs
,
46 int64_t offset
, unsigned int bytes
, QEMUIOVector
*qiov
,
47 BdrvRequestFlags flags
);
48 static int coroutine_fn
bdrv_co_do_pwritev(BlockDriverState
*bs
,
49 int64_t offset
, unsigned int bytes
, QEMUIOVector
*qiov
,
50 BdrvRequestFlags flags
);
51 static BlockAIOCB
*bdrv_co_aio_rw_vector(BlockDriverState
*bs
,
55 BdrvRequestFlags flags
,
56 BlockCompletionFunc
*cb
,
59 static void coroutine_fn
bdrv_co_do_rw(void *opaque
);
60 static int coroutine_fn
bdrv_co_do_write_zeroes(BlockDriverState
*bs
,
61 int64_t sector_num
, int nb_sectors
, BdrvRequestFlags flags
);
63 /* throttling disk I/O limits */
64 void bdrv_set_io_limits(BlockDriverState
*bs
,
69 throttle_group_config(bs
, cfg
);
71 for (i
= 0; i
< 2; i
++) {
72 qemu_co_enter_next(&bs
->throttled_reqs
[i
]);
76 /* this function drain all the throttled IOs */
77 static bool bdrv_start_throttled_reqs(BlockDriverState
*bs
)
80 bool enabled
= bs
->io_limits_enabled
;
83 bs
->io_limits_enabled
= false;
85 for (i
= 0; i
< 2; i
++) {
86 while (qemu_co_enter_next(&bs
->throttled_reqs
[i
])) {
91 bs
->io_limits_enabled
= enabled
;
96 void bdrv_io_limits_disable(BlockDriverState
*bs
)
98 bs
->io_limits_enabled
= false;
99 bdrv_start_throttled_reqs(bs
);
100 throttle_group_unregister_bs(bs
);
103 /* should be called before bdrv_set_io_limits if a limit is set */
104 void bdrv_io_limits_enable(BlockDriverState
*bs
, const char *group
)
106 assert(!bs
->io_limits_enabled
);
107 throttle_group_register_bs(bs
, group
);
108 bs
->io_limits_enabled
= true;
111 void bdrv_io_limits_update_group(BlockDriverState
*bs
, const char *group
)
113 /* this bs is not part of any group */
114 if (!bs
->throttle_state
) {
118 /* this bs is a part of the same group than the one we want */
119 if (!g_strcmp0(throttle_group_get_name(bs
), group
)) {
123 /* need to change the group this bs belong to */
124 bdrv_io_limits_disable(bs
);
125 bdrv_io_limits_enable(bs
, group
);
128 void bdrv_setup_io_funcs(BlockDriver
*bdrv
)
130 /* Block drivers without coroutine functions need emulation */
131 if (!bdrv
->bdrv_co_readv
) {
132 bdrv
->bdrv_co_readv
= bdrv_co_readv_em
;
133 bdrv
->bdrv_co_writev
= bdrv_co_writev_em
;
135 /* bdrv_co_readv_em()/brdv_co_writev_em() work in terms of aio, so if
136 * the block driver lacks aio we need to emulate that too.
138 if (!bdrv
->bdrv_aio_readv
) {
139 /* add AIO emulation layer */
140 bdrv
->bdrv_aio_readv
= bdrv_aio_readv_em
;
141 bdrv
->bdrv_aio_writev
= bdrv_aio_writev_em
;
146 void bdrv_refresh_limits(BlockDriverState
*bs
, Error
**errp
)
148 BlockDriver
*drv
= bs
->drv
;
149 Error
*local_err
= NULL
;
151 memset(&bs
->bl
, 0, sizeof(bs
->bl
));
157 /* Take some limits from the children as a default */
159 bdrv_refresh_limits(bs
->file
, &local_err
);
161 error_propagate(errp
, local_err
);
164 bs
->bl
.opt_transfer_length
= bs
->file
->bl
.opt_transfer_length
;
165 bs
->bl
.max_transfer_length
= bs
->file
->bl
.max_transfer_length
;
166 bs
->bl
.min_mem_alignment
= bs
->file
->bl
.min_mem_alignment
;
167 bs
->bl
.opt_mem_alignment
= bs
->file
->bl
.opt_mem_alignment
;
169 bs
->bl
.min_mem_alignment
= 512;
170 bs
->bl
.opt_mem_alignment
= getpagesize();
173 if (bs
->backing_hd
) {
174 bdrv_refresh_limits(bs
->backing_hd
, &local_err
);
176 error_propagate(errp
, local_err
);
179 bs
->bl
.opt_transfer_length
=
180 MAX(bs
->bl
.opt_transfer_length
,
181 bs
->backing_hd
->bl
.opt_transfer_length
);
182 bs
->bl
.max_transfer_length
=
183 MIN_NON_ZERO(bs
->bl
.max_transfer_length
,
184 bs
->backing_hd
->bl
.max_transfer_length
);
185 bs
->bl
.opt_mem_alignment
=
186 MAX(bs
->bl
.opt_mem_alignment
,
187 bs
->backing_hd
->bl
.opt_mem_alignment
);
188 bs
->bl
.min_mem_alignment
=
189 MAX(bs
->bl
.min_mem_alignment
,
190 bs
->backing_hd
->bl
.min_mem_alignment
);
193 /* Then let the driver override it */
194 if (drv
->bdrv_refresh_limits
) {
195 drv
->bdrv_refresh_limits(bs
, errp
);
200 * The copy-on-read flag is actually a reference count so multiple users may
201 * use the feature without worrying about clobbering its previous state.
202 * Copy-on-read stays enabled until all users have called to disable it.
204 void bdrv_enable_copy_on_read(BlockDriverState
*bs
)
209 void bdrv_disable_copy_on_read(BlockDriverState
*bs
)
211 assert(bs
->copy_on_read
> 0);
215 /* Check if any requests are in-flight (including throttled requests) */
216 static bool bdrv_requests_pending(BlockDriverState
*bs
)
218 if (!QLIST_EMPTY(&bs
->tracked_requests
)) {
221 if (!qemu_co_queue_empty(&bs
->throttled_reqs
[0])) {
224 if (!qemu_co_queue_empty(&bs
->throttled_reqs
[1])) {
227 if (bs
->file
&& bdrv_requests_pending(bs
->file
)) {
230 if (bs
->backing_hd
&& bdrv_requests_pending(bs
->backing_hd
)) {
237 * Wait for pending requests to complete on a single BlockDriverState subtree
239 * See the warning in bdrv_drain_all(). This function can only be called if
240 * you are sure nothing can generate I/O because you have op blockers
243 * Note that unlike bdrv_drain_all(), the caller must hold the BlockDriverState
246 void bdrv_drain(BlockDriverState
*bs
)
252 bdrv_flush_io_queue(bs
);
253 busy
= bdrv_requests_pending(bs
);
254 busy
|= aio_poll(bdrv_get_aio_context(bs
), busy
);
259 * Wait for pending requests to complete across all BlockDriverStates
261 * This function does not flush data to disk, use bdrv_flush_all() for that
262 * after calling this function.
264 * Note that completion of an asynchronous I/O operation can trigger any
265 * number of other I/O operations on other devices---for example a coroutine
266 * can be arbitrarily complex and a constant flow of I/O can come until the
267 * coroutine is complete. Because of this, it is not possible to have a
268 * function to drain a single device's I/O queue.
270 void bdrv_drain_all(void)
272 /* Always run first iteration so any pending completion BHs run */
274 BlockDriverState
*bs
= NULL
;
275 GSList
*aio_ctxs
= NULL
, *ctx
;
277 while ((bs
= bdrv_next(bs
))) {
278 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
280 aio_context_acquire(aio_context
);
282 block_job_pause(bs
->job
);
284 aio_context_release(aio_context
);
286 if (!aio_ctxs
|| !g_slist_find(aio_ctxs
, aio_context
)) {
287 aio_ctxs
= g_slist_prepend(aio_ctxs
, aio_context
);
294 for (ctx
= aio_ctxs
; ctx
!= NULL
; ctx
= ctx
->next
) {
295 AioContext
*aio_context
= ctx
->data
;
298 aio_context_acquire(aio_context
);
299 while ((bs
= bdrv_next(bs
))) {
300 if (aio_context
== bdrv_get_aio_context(bs
)) {
301 bdrv_flush_io_queue(bs
);
302 if (bdrv_requests_pending(bs
)) {
304 aio_poll(aio_context
, busy
);
308 busy
|= aio_poll(aio_context
, false);
309 aio_context_release(aio_context
);
314 while ((bs
= bdrv_next(bs
))) {
315 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
317 aio_context_acquire(aio_context
);
319 block_job_resume(bs
->job
);
321 aio_context_release(aio_context
);
323 g_slist_free(aio_ctxs
);
327 * Remove an active request from the tracked requests list
329 * This function should be called when a tracked request is completing.
331 static void tracked_request_end(BdrvTrackedRequest
*req
)
333 if (req
->serialising
) {
334 req
->bs
->serialising_in_flight
--;
337 QLIST_REMOVE(req
, list
);
338 qemu_co_queue_restart_all(&req
->wait_queue
);
342 * Add an active request to the tracked requests list
344 static void tracked_request_begin(BdrvTrackedRequest
*req
,
345 BlockDriverState
*bs
,
347 unsigned int bytes
, bool is_write
)
349 *req
= (BdrvTrackedRequest
){
353 .is_write
= is_write
,
354 .co
= qemu_coroutine_self(),
355 .serialising
= false,
356 .overlap_offset
= offset
,
357 .overlap_bytes
= bytes
,
360 qemu_co_queue_init(&req
->wait_queue
);
362 QLIST_INSERT_HEAD(&bs
->tracked_requests
, req
, list
);
365 static void mark_request_serialising(BdrvTrackedRequest
*req
, uint64_t align
)
367 int64_t overlap_offset
= req
->offset
& ~(align
- 1);
368 unsigned int overlap_bytes
= ROUND_UP(req
->offset
+ req
->bytes
, align
)
371 if (!req
->serialising
) {
372 req
->bs
->serialising_in_flight
++;
373 req
->serialising
= true;
376 req
->overlap_offset
= MIN(req
->overlap_offset
, overlap_offset
);
377 req
->overlap_bytes
= MAX(req
->overlap_bytes
, overlap_bytes
);
381 * Round a region to cluster boundaries
383 void bdrv_round_to_clusters(BlockDriverState
*bs
,
384 int64_t sector_num
, int nb_sectors
,
385 int64_t *cluster_sector_num
,
386 int *cluster_nb_sectors
)
390 if (bdrv_get_info(bs
, &bdi
) < 0 || bdi
.cluster_size
== 0) {
391 *cluster_sector_num
= sector_num
;
392 *cluster_nb_sectors
= nb_sectors
;
394 int64_t c
= bdi
.cluster_size
/ BDRV_SECTOR_SIZE
;
395 *cluster_sector_num
= QEMU_ALIGN_DOWN(sector_num
, c
);
396 *cluster_nb_sectors
= QEMU_ALIGN_UP(sector_num
- *cluster_sector_num
+
401 static int bdrv_get_cluster_size(BlockDriverState
*bs
)
406 ret
= bdrv_get_info(bs
, &bdi
);
407 if (ret
< 0 || bdi
.cluster_size
== 0) {
408 return bs
->request_alignment
;
410 return bdi
.cluster_size
;
414 static bool tracked_request_overlaps(BdrvTrackedRequest
*req
,
415 int64_t offset
, unsigned int bytes
)
418 if (offset
>= req
->overlap_offset
+ req
->overlap_bytes
) {
422 if (req
->overlap_offset
>= offset
+ bytes
) {
428 static bool coroutine_fn
wait_serialising_requests(BdrvTrackedRequest
*self
)
430 BlockDriverState
*bs
= self
->bs
;
431 BdrvTrackedRequest
*req
;
435 if (!bs
->serialising_in_flight
) {
441 QLIST_FOREACH(req
, &bs
->tracked_requests
, list
) {
442 if (req
== self
|| (!req
->serialising
&& !self
->serialising
)) {
445 if (tracked_request_overlaps(req
, self
->overlap_offset
,
446 self
->overlap_bytes
))
448 /* Hitting this means there was a reentrant request, for
449 * example, a block driver issuing nested requests. This must
450 * never happen since it means deadlock.
452 assert(qemu_coroutine_self() != req
->co
);
454 /* If the request is already (indirectly) waiting for us, or
455 * will wait for us as soon as it wakes up, then just go on
456 * (instead of producing a deadlock in the former case). */
457 if (!req
->waiting_for
) {
458 self
->waiting_for
= req
;
459 qemu_co_queue_wait(&req
->wait_queue
);
460 self
->waiting_for
= NULL
;
472 static int bdrv_check_byte_request(BlockDriverState
*bs
, int64_t offset
,
475 if (size
> BDRV_REQUEST_MAX_SECTORS
<< BDRV_SECTOR_BITS
) {
479 if (!bdrv_is_inserted(bs
)) {
490 static int bdrv_check_request(BlockDriverState
*bs
, int64_t sector_num
,
493 if (nb_sectors
< 0 || nb_sectors
> BDRV_REQUEST_MAX_SECTORS
) {
497 return bdrv_check_byte_request(bs
, sector_num
* BDRV_SECTOR_SIZE
,
498 nb_sectors
* BDRV_SECTOR_SIZE
);
501 typedef struct RwCo
{
502 BlockDriverState
*bs
;
507 BdrvRequestFlags flags
;
510 static void coroutine_fn
bdrv_rw_co_entry(void *opaque
)
514 if (!rwco
->is_write
) {
515 rwco
->ret
= bdrv_co_do_preadv(rwco
->bs
, rwco
->offset
,
516 rwco
->qiov
->size
, rwco
->qiov
,
519 rwco
->ret
= bdrv_co_do_pwritev(rwco
->bs
, rwco
->offset
,
520 rwco
->qiov
->size
, rwco
->qiov
,
526 * Process a vectored synchronous request using coroutines
528 static int bdrv_prwv_co(BlockDriverState
*bs
, int64_t offset
,
529 QEMUIOVector
*qiov
, bool is_write
,
530 BdrvRequestFlags flags
)
537 .is_write
= is_write
,
543 * In sync call context, when the vcpu is blocked, this throttling timer
544 * will not fire; so the I/O throttling function has to be disabled here
545 * if it has been enabled.
547 if (bs
->io_limits_enabled
) {
548 fprintf(stderr
, "Disabling I/O throttling on '%s' due "
549 "to synchronous I/O.\n", bdrv_get_device_name(bs
));
550 bdrv_io_limits_disable(bs
);
553 if (qemu_in_coroutine()) {
554 /* Fast-path if already in coroutine context */
555 bdrv_rw_co_entry(&rwco
);
557 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
559 co
= qemu_coroutine_create(bdrv_rw_co_entry
);
560 qemu_coroutine_enter(co
, &rwco
);
561 while (rwco
.ret
== NOT_DONE
) {
562 aio_poll(aio_context
, true);
569 * Process a synchronous request using coroutines
571 static int bdrv_rw_co(BlockDriverState
*bs
, int64_t sector_num
, uint8_t *buf
,
572 int nb_sectors
, bool is_write
, BdrvRequestFlags flags
)
576 .iov_base
= (void *)buf
,
577 .iov_len
= nb_sectors
* BDRV_SECTOR_SIZE
,
580 if (nb_sectors
< 0 || nb_sectors
> BDRV_REQUEST_MAX_SECTORS
) {
584 qemu_iovec_init_external(&qiov
, &iov
, 1);
585 return bdrv_prwv_co(bs
, sector_num
<< BDRV_SECTOR_BITS
,
586 &qiov
, is_write
, flags
);
589 /* return < 0 if error. See bdrv_write() for the return codes */
590 int bdrv_read(BlockDriverState
*bs
, int64_t sector_num
,
591 uint8_t *buf
, int nb_sectors
)
593 return bdrv_rw_co(bs
, sector_num
, buf
, nb_sectors
, false, 0);
596 /* Just like bdrv_read(), but with I/O throttling temporarily disabled */
597 int bdrv_read_unthrottled(BlockDriverState
*bs
, int64_t sector_num
,
598 uint8_t *buf
, int nb_sectors
)
603 enabled
= bs
->io_limits_enabled
;
604 bs
->io_limits_enabled
= false;
605 ret
= bdrv_read(bs
, sector_num
, buf
, nb_sectors
);
606 bs
->io_limits_enabled
= enabled
;
610 /* Return < 0 if error. Important errors are:
611 -EIO generic I/O error (may happen for all errors)
612 -ENOMEDIUM No media inserted.
613 -EINVAL Invalid sector number or nb_sectors
614 -EACCES Trying to write a read-only device
616 int bdrv_write(BlockDriverState
*bs
, int64_t sector_num
,
617 const uint8_t *buf
, int nb_sectors
)
619 return bdrv_rw_co(bs
, sector_num
, (uint8_t *)buf
, nb_sectors
, true, 0);
622 int bdrv_write_zeroes(BlockDriverState
*bs
, int64_t sector_num
,
623 int nb_sectors
, BdrvRequestFlags flags
)
625 return bdrv_rw_co(bs
, sector_num
, NULL
, nb_sectors
, true,
626 BDRV_REQ_ZERO_WRITE
| flags
);
630 * Completely zero out a block device with the help of bdrv_write_zeroes.
631 * The operation is sped up by checking the block status and only writing
632 * zeroes to the device if they currently do not return zeroes. Optional
633 * flags are passed through to bdrv_write_zeroes (e.g. BDRV_REQ_MAY_UNMAP).
635 * Returns < 0 on error, 0 on success. For error codes see bdrv_write().
637 int bdrv_make_zero(BlockDriverState
*bs
, BdrvRequestFlags flags
)
639 int64_t target_sectors
, ret
, nb_sectors
, sector_num
= 0;
642 target_sectors
= bdrv_nb_sectors(bs
);
643 if (target_sectors
< 0) {
644 return target_sectors
;
648 nb_sectors
= MIN(target_sectors
- sector_num
, BDRV_REQUEST_MAX_SECTORS
);
649 if (nb_sectors
<= 0) {
652 ret
= bdrv_get_block_status(bs
, sector_num
, nb_sectors
, &n
);
654 error_report("error getting block status at sector %" PRId64
": %s",
655 sector_num
, strerror(-ret
));
658 if (ret
& BDRV_BLOCK_ZERO
) {
662 ret
= bdrv_write_zeroes(bs
, sector_num
, n
, flags
);
664 error_report("error writing zeroes at sector %" PRId64
": %s",
665 sector_num
, strerror(-ret
));
672 int bdrv_pread(BlockDriverState
*bs
, int64_t offset
, void *buf
, int bytes
)
676 .iov_base
= (void *)buf
,
685 qemu_iovec_init_external(&qiov
, &iov
, 1);
686 ret
= bdrv_prwv_co(bs
, offset
, &qiov
, false, 0);
694 int bdrv_pwritev(BlockDriverState
*bs
, int64_t offset
, QEMUIOVector
*qiov
)
698 ret
= bdrv_prwv_co(bs
, offset
, qiov
, true, 0);
706 int bdrv_pwrite(BlockDriverState
*bs
, int64_t offset
,
707 const void *buf
, int bytes
)
711 .iov_base
= (void *) buf
,
719 qemu_iovec_init_external(&qiov
, &iov
, 1);
720 return bdrv_pwritev(bs
, offset
, &qiov
);
724 * Writes to the file and ensures that no writes are reordered across this
725 * request (acts as a barrier)
727 * Returns 0 on success, -errno in error cases.
729 int bdrv_pwrite_sync(BlockDriverState
*bs
, int64_t offset
,
730 const void *buf
, int count
)
734 ret
= bdrv_pwrite(bs
, offset
, buf
, count
);
739 /* No flush needed for cache modes that already do it */
740 if (bs
->enable_write_cache
) {
747 static int coroutine_fn
bdrv_co_do_copy_on_readv(BlockDriverState
*bs
,
748 int64_t sector_num
, int nb_sectors
, QEMUIOVector
*qiov
)
750 /* Perform I/O through a temporary buffer so that users who scribble over
751 * their read buffer while the operation is in progress do not end up
752 * modifying the image file. This is critical for zero-copy guest I/O
753 * where anything might happen inside guest memory.
757 BlockDriver
*drv
= bs
->drv
;
759 QEMUIOVector bounce_qiov
;
760 int64_t cluster_sector_num
;
761 int cluster_nb_sectors
;
765 /* Cover entire cluster so no additional backing file I/O is required when
766 * allocating cluster in the image file.
768 bdrv_round_to_clusters(bs
, sector_num
, nb_sectors
,
769 &cluster_sector_num
, &cluster_nb_sectors
);
771 trace_bdrv_co_do_copy_on_readv(bs
, sector_num
, nb_sectors
,
772 cluster_sector_num
, cluster_nb_sectors
);
774 iov
.iov_len
= cluster_nb_sectors
* BDRV_SECTOR_SIZE
;
775 iov
.iov_base
= bounce_buffer
= qemu_try_blockalign(bs
, iov
.iov_len
);
776 if (bounce_buffer
== NULL
) {
781 qemu_iovec_init_external(&bounce_qiov
, &iov
, 1);
783 ret
= drv
->bdrv_co_readv(bs
, cluster_sector_num
, cluster_nb_sectors
,
789 if (drv
->bdrv_co_write_zeroes
&&
790 buffer_is_zero(bounce_buffer
, iov
.iov_len
)) {
791 ret
= bdrv_co_do_write_zeroes(bs
, cluster_sector_num
,
792 cluster_nb_sectors
, 0);
794 /* This does not change the data on the disk, it is not necessary
795 * to flush even in cache=writethrough mode.
797 ret
= drv
->bdrv_co_writev(bs
, cluster_sector_num
, cluster_nb_sectors
,
802 /* It might be okay to ignore write errors for guest requests. If this
803 * is a deliberate copy-on-read then we don't want to ignore the error.
804 * Simply report it in all cases.
809 skip_bytes
= (sector_num
- cluster_sector_num
) * BDRV_SECTOR_SIZE
;
810 qemu_iovec_from_buf(qiov
, 0, bounce_buffer
+ skip_bytes
,
811 nb_sectors
* BDRV_SECTOR_SIZE
);
814 qemu_vfree(bounce_buffer
);
819 * Forwards an already correctly aligned request to the BlockDriver. This
820 * handles copy on read and zeroing after EOF; any other features must be
821 * implemented by the caller.
823 static int coroutine_fn
bdrv_aligned_preadv(BlockDriverState
*bs
,
824 BdrvTrackedRequest
*req
, int64_t offset
, unsigned int bytes
,
825 int64_t align
, QEMUIOVector
*qiov
, int flags
)
827 BlockDriver
*drv
= bs
->drv
;
830 int64_t sector_num
= offset
>> BDRV_SECTOR_BITS
;
831 unsigned int nb_sectors
= bytes
>> BDRV_SECTOR_BITS
;
833 assert((offset
& (BDRV_SECTOR_SIZE
- 1)) == 0);
834 assert((bytes
& (BDRV_SECTOR_SIZE
- 1)) == 0);
835 assert(!qiov
|| bytes
== qiov
->size
);
837 /* Handle Copy on Read and associated serialisation */
838 if (flags
& BDRV_REQ_COPY_ON_READ
) {
839 /* If we touch the same cluster it counts as an overlap. This
840 * guarantees that allocating writes will be serialized and not race
841 * with each other for the same cluster. For example, in copy-on-read
842 * it ensures that the CoR read and write operations are atomic and
843 * guest writes cannot interleave between them. */
844 mark_request_serialising(req
, bdrv_get_cluster_size(bs
));
847 wait_serialising_requests(req
);
849 if (flags
& BDRV_REQ_COPY_ON_READ
) {
852 ret
= bdrv_is_allocated(bs
, sector_num
, nb_sectors
, &pnum
);
857 if (!ret
|| pnum
!= nb_sectors
) {
858 ret
= bdrv_co_do_copy_on_readv(bs
, sector_num
, nb_sectors
, qiov
);
863 /* Forward the request to the BlockDriver */
864 if (!bs
->zero_beyond_eof
) {
865 ret
= drv
->bdrv_co_readv(bs
, sector_num
, nb_sectors
, qiov
);
867 /* Read zeros after EOF */
868 int64_t total_sectors
, max_nb_sectors
;
870 total_sectors
= bdrv_nb_sectors(bs
);
871 if (total_sectors
< 0) {
876 max_nb_sectors
= ROUND_UP(MAX(0, total_sectors
- sector_num
),
877 align
>> BDRV_SECTOR_BITS
);
878 if (nb_sectors
< max_nb_sectors
) {
879 ret
= drv
->bdrv_co_readv(bs
, sector_num
, nb_sectors
, qiov
);
880 } else if (max_nb_sectors
> 0) {
881 QEMUIOVector local_qiov
;
883 qemu_iovec_init(&local_qiov
, qiov
->niov
);
884 qemu_iovec_concat(&local_qiov
, qiov
, 0,
885 max_nb_sectors
* BDRV_SECTOR_SIZE
);
887 ret
= drv
->bdrv_co_readv(bs
, sector_num
, max_nb_sectors
,
890 qemu_iovec_destroy(&local_qiov
);
895 /* Reading beyond end of file is supposed to produce zeroes */
896 if (ret
== 0 && total_sectors
< sector_num
+ nb_sectors
) {
897 uint64_t offset
= MAX(0, total_sectors
- sector_num
);
898 uint64_t bytes
= (sector_num
+ nb_sectors
- offset
) *
900 qemu_iovec_memset(qiov
, offset
* BDRV_SECTOR_SIZE
, 0, bytes
);
909 * Handle a read request in coroutine context
911 static int coroutine_fn
bdrv_co_do_preadv(BlockDriverState
*bs
,
912 int64_t offset
, unsigned int bytes
, QEMUIOVector
*qiov
,
913 BdrvRequestFlags flags
)
915 BlockDriver
*drv
= bs
->drv
;
916 BdrvTrackedRequest req
;
918 /* TODO Lift BDRV_SECTOR_SIZE restriction in BlockDriver interface */
919 uint64_t align
= MAX(BDRV_SECTOR_SIZE
, bs
->request_alignment
);
920 uint8_t *head_buf
= NULL
;
921 uint8_t *tail_buf
= NULL
;
922 QEMUIOVector local_qiov
;
923 bool use_local_qiov
= false;
930 ret
= bdrv_check_byte_request(bs
, offset
, bytes
);
935 if (bs
->copy_on_read
) {
936 flags
|= BDRV_REQ_COPY_ON_READ
;
939 /* throttling disk I/O */
940 if (bs
->io_limits_enabled
) {
941 throttle_group_co_io_limits_intercept(bs
, bytes
, false);
944 /* Align read if necessary by padding qiov */
945 if (offset
& (align
- 1)) {
946 head_buf
= qemu_blockalign(bs
, align
);
947 qemu_iovec_init(&local_qiov
, qiov
->niov
+ 2);
948 qemu_iovec_add(&local_qiov
, head_buf
, offset
& (align
- 1));
949 qemu_iovec_concat(&local_qiov
, qiov
, 0, qiov
->size
);
950 use_local_qiov
= true;
952 bytes
+= offset
& (align
- 1);
953 offset
= offset
& ~(align
- 1);
956 if ((offset
+ bytes
) & (align
- 1)) {
957 if (!use_local_qiov
) {
958 qemu_iovec_init(&local_qiov
, qiov
->niov
+ 1);
959 qemu_iovec_concat(&local_qiov
, qiov
, 0, qiov
->size
);
960 use_local_qiov
= true;
962 tail_buf
= qemu_blockalign(bs
, align
);
963 qemu_iovec_add(&local_qiov
, tail_buf
,
964 align
- ((offset
+ bytes
) & (align
- 1)));
966 bytes
= ROUND_UP(bytes
, align
);
969 tracked_request_begin(&req
, bs
, offset
, bytes
, false);
970 ret
= bdrv_aligned_preadv(bs
, &req
, offset
, bytes
, align
,
971 use_local_qiov
? &local_qiov
: qiov
,
973 tracked_request_end(&req
);
975 if (use_local_qiov
) {
976 qemu_iovec_destroy(&local_qiov
);
977 qemu_vfree(head_buf
);
978 qemu_vfree(tail_buf
);
984 static int coroutine_fn
bdrv_co_do_readv(BlockDriverState
*bs
,
985 int64_t sector_num
, int nb_sectors
, QEMUIOVector
*qiov
,
986 BdrvRequestFlags flags
)
988 if (nb_sectors
< 0 || nb_sectors
> BDRV_REQUEST_MAX_SECTORS
) {
992 return bdrv_co_do_preadv(bs
, sector_num
<< BDRV_SECTOR_BITS
,
993 nb_sectors
<< BDRV_SECTOR_BITS
, qiov
, flags
);
996 int coroutine_fn
bdrv_co_readv(BlockDriverState
*bs
, int64_t sector_num
,
997 int nb_sectors
, QEMUIOVector
*qiov
)
999 trace_bdrv_co_readv(bs
, sector_num
, nb_sectors
);
1001 return bdrv_co_do_readv(bs
, sector_num
, nb_sectors
, qiov
, 0);
1004 int coroutine_fn
bdrv_co_copy_on_readv(BlockDriverState
*bs
,
1005 int64_t sector_num
, int nb_sectors
, QEMUIOVector
*qiov
)
1007 trace_bdrv_co_copy_on_readv(bs
, sector_num
, nb_sectors
);
1009 return bdrv_co_do_readv(bs
, sector_num
, nb_sectors
, qiov
,
1010 BDRV_REQ_COPY_ON_READ
);
1013 #define MAX_WRITE_ZEROES_BOUNCE_BUFFER 32768
1015 static int coroutine_fn
bdrv_co_do_write_zeroes(BlockDriverState
*bs
,
1016 int64_t sector_num
, int nb_sectors
, BdrvRequestFlags flags
)
1018 BlockDriver
*drv
= bs
->drv
;
1020 struct iovec iov
= {0};
1023 int max_write_zeroes
= MIN_NON_ZERO(bs
->bl
.max_write_zeroes
,
1024 BDRV_REQUEST_MAX_SECTORS
);
1026 while (nb_sectors
> 0 && !ret
) {
1027 int num
= nb_sectors
;
1029 /* Align request. Block drivers can expect the "bulk" of the request
1032 if (bs
->bl
.write_zeroes_alignment
1033 && num
> bs
->bl
.write_zeroes_alignment
) {
1034 if (sector_num
% bs
->bl
.write_zeroes_alignment
!= 0) {
1035 /* Make a small request up to the first aligned sector. */
1036 num
= bs
->bl
.write_zeroes_alignment
;
1037 num
-= sector_num
% bs
->bl
.write_zeroes_alignment
;
1038 } else if ((sector_num
+ num
) % bs
->bl
.write_zeroes_alignment
!= 0) {
1039 /* Shorten the request to the last aligned sector. num cannot
1040 * underflow because num > bs->bl.write_zeroes_alignment.
1042 num
-= (sector_num
+ num
) % bs
->bl
.write_zeroes_alignment
;
1046 /* limit request size */
1047 if (num
> max_write_zeroes
) {
1048 num
= max_write_zeroes
;
1052 /* First try the efficient write zeroes operation */
1053 if (drv
->bdrv_co_write_zeroes
) {
1054 ret
= drv
->bdrv_co_write_zeroes(bs
, sector_num
, num
, flags
);
1057 if (ret
== -ENOTSUP
) {
1058 /* Fall back to bounce buffer if write zeroes is unsupported */
1059 int max_xfer_len
= MIN_NON_ZERO(bs
->bl
.max_transfer_length
,
1060 MAX_WRITE_ZEROES_BOUNCE_BUFFER
);
1061 num
= MIN(num
, max_xfer_len
);
1062 iov
.iov_len
= num
* BDRV_SECTOR_SIZE
;
1063 if (iov
.iov_base
== NULL
) {
1064 iov
.iov_base
= qemu_try_blockalign(bs
, num
* BDRV_SECTOR_SIZE
);
1065 if (iov
.iov_base
== NULL
) {
1069 memset(iov
.iov_base
, 0, num
* BDRV_SECTOR_SIZE
);
1071 qemu_iovec_init_external(&qiov
, &iov
, 1);
1073 ret
= drv
->bdrv_co_writev(bs
, sector_num
, num
, &qiov
);
1075 /* Keep bounce buffer around if it is big enough for all
1076 * all future requests.
1078 if (num
< max_xfer_len
) {
1079 qemu_vfree(iov
.iov_base
);
1080 iov
.iov_base
= NULL
;
1089 qemu_vfree(iov
.iov_base
);
1094 * Forwards an already correctly aligned write request to the BlockDriver.
1096 static int coroutine_fn
bdrv_aligned_pwritev(BlockDriverState
*bs
,
1097 BdrvTrackedRequest
*req
, int64_t offset
, unsigned int bytes
,
1098 QEMUIOVector
*qiov
, int flags
)
1100 BlockDriver
*drv
= bs
->drv
;
1104 int64_t sector_num
= offset
>> BDRV_SECTOR_BITS
;
1105 unsigned int nb_sectors
= bytes
>> BDRV_SECTOR_BITS
;
1107 assert((offset
& (BDRV_SECTOR_SIZE
- 1)) == 0);
1108 assert((bytes
& (BDRV_SECTOR_SIZE
- 1)) == 0);
1109 assert(!qiov
|| bytes
== qiov
->size
);
1111 waited
= wait_serialising_requests(req
);
1112 assert(!waited
|| !req
->serialising
);
1113 assert(req
->overlap_offset
<= offset
);
1114 assert(offset
+ bytes
<= req
->overlap_offset
+ req
->overlap_bytes
);
1116 ret
= notifier_with_return_list_notify(&bs
->before_write_notifiers
, req
);
1118 if (!ret
&& bs
->detect_zeroes
!= BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF
&&
1119 !(flags
& BDRV_REQ_ZERO_WRITE
) && drv
->bdrv_co_write_zeroes
&&
1120 qemu_iovec_is_zero(qiov
)) {
1121 flags
|= BDRV_REQ_ZERO_WRITE
;
1122 if (bs
->detect_zeroes
== BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP
) {
1123 flags
|= BDRV_REQ_MAY_UNMAP
;
1128 /* Do nothing, write notifier decided to fail this request */
1129 } else if (flags
& BDRV_REQ_ZERO_WRITE
) {
1130 BLKDBG_EVENT(bs
, BLKDBG_PWRITEV_ZERO
);
1131 ret
= bdrv_co_do_write_zeroes(bs
, sector_num
, nb_sectors
, flags
);
1133 BLKDBG_EVENT(bs
, BLKDBG_PWRITEV
);
1134 ret
= drv
->bdrv_co_writev(bs
, sector_num
, nb_sectors
, qiov
);
1136 BLKDBG_EVENT(bs
, BLKDBG_PWRITEV_DONE
);
1138 if (ret
== 0 && !bs
->enable_write_cache
) {
1139 ret
= bdrv_co_flush(bs
);
1142 bdrv_set_dirty(bs
, sector_num
, nb_sectors
);
1144 block_acct_highest_sector(&bs
->stats
, sector_num
, nb_sectors
);
1147 bs
->total_sectors
= MAX(bs
->total_sectors
, sector_num
+ nb_sectors
);
1153 static int coroutine_fn
bdrv_co_do_zero_pwritev(BlockDriverState
*bs
,
1156 BdrvRequestFlags flags
,
1157 BdrvTrackedRequest
*req
)
1159 uint8_t *buf
= NULL
;
1160 QEMUIOVector local_qiov
;
1162 uint64_t align
= MAX(BDRV_SECTOR_SIZE
, bs
->request_alignment
);
1163 unsigned int head_padding_bytes
, tail_padding_bytes
;
1166 head_padding_bytes
= offset
& (align
- 1);
1167 tail_padding_bytes
= align
- ((offset
+ bytes
) & (align
- 1));
1170 assert(flags
& BDRV_REQ_ZERO_WRITE
);
1171 if (head_padding_bytes
|| tail_padding_bytes
) {
1172 buf
= qemu_blockalign(bs
, align
);
1173 iov
= (struct iovec
) {
1177 qemu_iovec_init_external(&local_qiov
, &iov
, 1);
1179 if (head_padding_bytes
) {
1180 uint64_t zero_bytes
= MIN(bytes
, align
- head_padding_bytes
);
1182 /* RMW the unaligned part before head. */
1183 mark_request_serialising(req
, align
);
1184 wait_serialising_requests(req
);
1185 BLKDBG_EVENT(bs
, BLKDBG_PWRITEV_RMW_HEAD
);
1186 ret
= bdrv_aligned_preadv(bs
, req
, offset
& ~(align
- 1), align
,
1187 align
, &local_qiov
, 0);
1191 BLKDBG_EVENT(bs
, BLKDBG_PWRITEV_RMW_AFTER_HEAD
);
1193 memset(buf
+ head_padding_bytes
, 0, zero_bytes
);
1194 ret
= bdrv_aligned_pwritev(bs
, req
, offset
& ~(align
- 1), align
,
1196 flags
& ~BDRV_REQ_ZERO_WRITE
);
1200 offset
+= zero_bytes
;
1201 bytes
-= zero_bytes
;
1204 assert(!bytes
|| (offset
& (align
- 1)) == 0);
1205 if (bytes
>= align
) {
1206 /* Write the aligned part in the middle. */
1207 uint64_t aligned_bytes
= bytes
& ~(align
- 1);
1208 ret
= bdrv_aligned_pwritev(bs
, req
, offset
, aligned_bytes
,
1213 bytes
-= aligned_bytes
;
1214 offset
+= aligned_bytes
;
1217 assert(!bytes
|| (offset
& (align
- 1)) == 0);
1219 assert(align
== tail_padding_bytes
+ bytes
);
1220 /* RMW the unaligned part after tail. */
1221 mark_request_serialising(req
, align
);
1222 wait_serialising_requests(req
);
1223 BLKDBG_EVENT(bs
, BLKDBG_PWRITEV_RMW_TAIL
);
1224 ret
= bdrv_aligned_preadv(bs
, req
, offset
, align
,
1225 align
, &local_qiov
, 0);
1229 BLKDBG_EVENT(bs
, BLKDBG_PWRITEV_RMW_AFTER_TAIL
);
1231 memset(buf
, 0, bytes
);
1232 ret
= bdrv_aligned_pwritev(bs
, req
, offset
, align
,
1233 &local_qiov
, flags
& ~BDRV_REQ_ZERO_WRITE
);
1242 * Handle a write request in coroutine context
1244 static int coroutine_fn
bdrv_co_do_pwritev(BlockDriverState
*bs
,
1245 int64_t offset
, unsigned int bytes
, QEMUIOVector
*qiov
,
1246 BdrvRequestFlags flags
)
1248 BdrvTrackedRequest req
;
1249 /* TODO Lift BDRV_SECTOR_SIZE restriction in BlockDriver interface */
1250 uint64_t align
= MAX(BDRV_SECTOR_SIZE
, bs
->request_alignment
);
1251 uint8_t *head_buf
= NULL
;
1252 uint8_t *tail_buf
= NULL
;
1253 QEMUIOVector local_qiov
;
1254 bool use_local_qiov
= false;
1260 if (bs
->read_only
) {
1264 ret
= bdrv_check_byte_request(bs
, offset
, bytes
);
1269 /* throttling disk I/O */
1270 if (bs
->io_limits_enabled
) {
1271 throttle_group_co_io_limits_intercept(bs
, bytes
, true);
1275 * Align write if necessary by performing a read-modify-write cycle.
1276 * Pad qiov with the read parts and be sure to have a tracked request not
1277 * only for bdrv_aligned_pwritev, but also for the reads of the RMW cycle.
1279 tracked_request_begin(&req
, bs
, offset
, bytes
, true);
1282 ret
= bdrv_co_do_zero_pwritev(bs
, offset
, bytes
, flags
, &req
);
1286 if (offset
& (align
- 1)) {
1287 QEMUIOVector head_qiov
;
1288 struct iovec head_iov
;
1290 mark_request_serialising(&req
, align
);
1291 wait_serialising_requests(&req
);
1293 head_buf
= qemu_blockalign(bs
, align
);
1294 head_iov
= (struct iovec
) {
1295 .iov_base
= head_buf
,
1298 qemu_iovec_init_external(&head_qiov
, &head_iov
, 1);
1300 BLKDBG_EVENT(bs
, BLKDBG_PWRITEV_RMW_HEAD
);
1301 ret
= bdrv_aligned_preadv(bs
, &req
, offset
& ~(align
- 1), align
,
1302 align
, &head_qiov
, 0);
1306 BLKDBG_EVENT(bs
, BLKDBG_PWRITEV_RMW_AFTER_HEAD
);
1308 qemu_iovec_init(&local_qiov
, qiov
->niov
+ 2);
1309 qemu_iovec_add(&local_qiov
, head_buf
, offset
& (align
- 1));
1310 qemu_iovec_concat(&local_qiov
, qiov
, 0, qiov
->size
);
1311 use_local_qiov
= true;
1313 bytes
+= offset
& (align
- 1);
1314 offset
= offset
& ~(align
- 1);
1317 if ((offset
+ bytes
) & (align
- 1)) {
1318 QEMUIOVector tail_qiov
;
1319 struct iovec tail_iov
;
1323 mark_request_serialising(&req
, align
);
1324 waited
= wait_serialising_requests(&req
);
1325 assert(!waited
|| !use_local_qiov
);
1327 tail_buf
= qemu_blockalign(bs
, align
);
1328 tail_iov
= (struct iovec
) {
1329 .iov_base
= tail_buf
,
1332 qemu_iovec_init_external(&tail_qiov
, &tail_iov
, 1);
1334 BLKDBG_EVENT(bs
, BLKDBG_PWRITEV_RMW_TAIL
);
1335 ret
= bdrv_aligned_preadv(bs
, &req
, (offset
+ bytes
) & ~(align
- 1), align
,
1336 align
, &tail_qiov
, 0);
1340 BLKDBG_EVENT(bs
, BLKDBG_PWRITEV_RMW_AFTER_TAIL
);
1342 if (!use_local_qiov
) {
1343 qemu_iovec_init(&local_qiov
, qiov
->niov
+ 1);
1344 qemu_iovec_concat(&local_qiov
, qiov
, 0, qiov
->size
);
1345 use_local_qiov
= true;
1348 tail_bytes
= (offset
+ bytes
) & (align
- 1);
1349 qemu_iovec_add(&local_qiov
, tail_buf
+ tail_bytes
, align
- tail_bytes
);
1351 bytes
= ROUND_UP(bytes
, align
);
1354 ret
= bdrv_aligned_pwritev(bs
, &req
, offset
, bytes
,
1355 use_local_qiov
? &local_qiov
: qiov
,
1360 if (use_local_qiov
) {
1361 qemu_iovec_destroy(&local_qiov
);
1363 qemu_vfree(head_buf
);
1364 qemu_vfree(tail_buf
);
1366 tracked_request_end(&req
);
1370 static int coroutine_fn
bdrv_co_do_writev(BlockDriverState
*bs
,
1371 int64_t sector_num
, int nb_sectors
, QEMUIOVector
*qiov
,
1372 BdrvRequestFlags flags
)
1374 if (nb_sectors
< 0 || nb_sectors
> BDRV_REQUEST_MAX_SECTORS
) {
1378 return bdrv_co_do_pwritev(bs
, sector_num
<< BDRV_SECTOR_BITS
,
1379 nb_sectors
<< BDRV_SECTOR_BITS
, qiov
, flags
);
1382 int coroutine_fn
bdrv_co_writev(BlockDriverState
*bs
, int64_t sector_num
,
1383 int nb_sectors
, QEMUIOVector
*qiov
)
1385 trace_bdrv_co_writev(bs
, sector_num
, nb_sectors
);
1387 return bdrv_co_do_writev(bs
, sector_num
, nb_sectors
, qiov
, 0);
1390 int coroutine_fn
bdrv_co_write_zeroes(BlockDriverState
*bs
,
1391 int64_t sector_num
, int nb_sectors
,
1392 BdrvRequestFlags flags
)
1394 trace_bdrv_co_write_zeroes(bs
, sector_num
, nb_sectors
, flags
);
1396 if (!(bs
->open_flags
& BDRV_O_UNMAP
)) {
1397 flags
&= ~BDRV_REQ_MAY_UNMAP
;
1400 return bdrv_co_do_writev(bs
, sector_num
, nb_sectors
, NULL
,
1401 BDRV_REQ_ZERO_WRITE
| flags
);
1404 int bdrv_flush_all(void)
1406 BlockDriverState
*bs
= NULL
;
1409 while ((bs
= bdrv_next(bs
))) {
1410 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
1413 aio_context_acquire(aio_context
);
1414 ret
= bdrv_flush(bs
);
1415 if (ret
< 0 && !result
) {
1418 aio_context_release(aio_context
);
1424 typedef struct BdrvCoGetBlockStatusData
{
1425 BlockDriverState
*bs
;
1426 BlockDriverState
*base
;
1432 } BdrvCoGetBlockStatusData
;
1435 * Returns the allocation status of the specified sectors.
1436 * Drivers not implementing the functionality are assumed to not support
1437 * backing files, hence all their sectors are reported as allocated.
1439 * If 'sector_num' is beyond the end of the disk image the return value is 0
1440 * and 'pnum' is set to 0.
1442 * 'pnum' is set to the number of sectors (including and immediately following
1443 * the specified sector) that are known to be in the same
1444 * allocated/unallocated state.
1446 * 'nb_sectors' is the max value 'pnum' should be set to. If nb_sectors goes
1447 * beyond the end of the disk image it will be clamped.
1449 static int64_t coroutine_fn
bdrv_co_get_block_status(BlockDriverState
*bs
,
1451 int nb_sectors
, int *pnum
)
1453 int64_t total_sectors
;
1457 total_sectors
= bdrv_nb_sectors(bs
);
1458 if (total_sectors
< 0) {
1459 return total_sectors
;
1462 if (sector_num
>= total_sectors
) {
1467 n
= total_sectors
- sector_num
;
1468 if (n
< nb_sectors
) {
1472 if (!bs
->drv
->bdrv_co_get_block_status
) {
1474 ret
= BDRV_BLOCK_DATA
| BDRV_BLOCK_ALLOCATED
;
1475 if (bs
->drv
->protocol_name
) {
1476 ret
|= BDRV_BLOCK_OFFSET_VALID
| (sector_num
* BDRV_SECTOR_SIZE
);
1481 ret
= bs
->drv
->bdrv_co_get_block_status(bs
, sector_num
, nb_sectors
, pnum
);
1487 if (ret
& BDRV_BLOCK_RAW
) {
1488 assert(ret
& BDRV_BLOCK_OFFSET_VALID
);
1489 return bdrv_get_block_status(bs
->file
, ret
>> BDRV_SECTOR_BITS
,
1493 if (ret
& (BDRV_BLOCK_DATA
| BDRV_BLOCK_ZERO
)) {
1494 ret
|= BDRV_BLOCK_ALLOCATED
;
1496 if (bdrv_unallocated_blocks_are_zero(bs
)) {
1497 ret
|= BDRV_BLOCK_ZERO
;
1498 } else if (bs
->backing_hd
) {
1499 BlockDriverState
*bs2
= bs
->backing_hd
;
1500 int64_t nb_sectors2
= bdrv_nb_sectors(bs2
);
1501 if (nb_sectors2
>= 0 && sector_num
>= nb_sectors2
) {
1502 ret
|= BDRV_BLOCK_ZERO
;
1508 (ret
& BDRV_BLOCK_DATA
) && !(ret
& BDRV_BLOCK_ZERO
) &&
1509 (ret
& BDRV_BLOCK_OFFSET_VALID
)) {
1512 ret2
= bdrv_co_get_block_status(bs
->file
, ret
>> BDRV_SECTOR_BITS
,
1515 /* Ignore errors. This is just providing extra information, it
1516 * is useful but not necessary.
1519 /* !file_pnum indicates an offset at or beyond the EOF; it is
1520 * perfectly valid for the format block driver to point to such
1521 * offsets, so catch it and mark everything as zero */
1522 ret
|= BDRV_BLOCK_ZERO
;
1524 /* Limit request to the range reported by the protocol driver */
1526 ret
|= (ret2
& BDRV_BLOCK_ZERO
);
1534 /* Coroutine wrapper for bdrv_get_block_status() */
1535 static void coroutine_fn
bdrv_get_block_status_co_entry(void *opaque
)
1537 BdrvCoGetBlockStatusData
*data
= opaque
;
1538 BlockDriverState
*bs
= data
->bs
;
1540 data
->ret
= bdrv_co_get_block_status(bs
, data
->sector_num
, data
->nb_sectors
,
1546 * Synchronous wrapper around bdrv_co_get_block_status().
1548 * See bdrv_co_get_block_status() for details.
1550 int64_t bdrv_get_block_status(BlockDriverState
*bs
, int64_t sector_num
,
1551 int nb_sectors
, int *pnum
)
1554 BdrvCoGetBlockStatusData data
= {
1556 .sector_num
= sector_num
,
1557 .nb_sectors
= nb_sectors
,
1562 if (qemu_in_coroutine()) {
1563 /* Fast-path if already in coroutine context */
1564 bdrv_get_block_status_co_entry(&data
);
1566 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
1568 co
= qemu_coroutine_create(bdrv_get_block_status_co_entry
);
1569 qemu_coroutine_enter(co
, &data
);
1570 while (!data
.done
) {
1571 aio_poll(aio_context
, true);
1577 int coroutine_fn
bdrv_is_allocated(BlockDriverState
*bs
, int64_t sector_num
,
1578 int nb_sectors
, int *pnum
)
1580 int64_t ret
= bdrv_get_block_status(bs
, sector_num
, nb_sectors
, pnum
);
1584 return !!(ret
& BDRV_BLOCK_ALLOCATED
);
1588 * Given an image chain: ... -> [BASE] -> [INTER1] -> [INTER2] -> [TOP]
1590 * Return true if the given sector is allocated in any image between
1591 * BASE and TOP (inclusive). BASE can be NULL to check if the given
1592 * sector is allocated in any image of the chain. Return false otherwise.
1594 * 'pnum' is set to the number of sectors (including and immediately following
1595 * the specified sector) that are known to be in the same
1596 * allocated/unallocated state.
1599 int bdrv_is_allocated_above(BlockDriverState
*top
,
1600 BlockDriverState
*base
,
1602 int nb_sectors
, int *pnum
)
1604 BlockDriverState
*intermediate
;
1605 int ret
, n
= nb_sectors
;
1608 while (intermediate
&& intermediate
!= base
) {
1610 ret
= bdrv_is_allocated(intermediate
, sector_num
, nb_sectors
,
1620 * [sector_num, nb_sectors] is unallocated on top but intermediate
1623 * [sector_num+x, nr_sectors] allocated.
1625 if (n
> pnum_inter
&&
1626 (intermediate
== top
||
1627 sector_num
+ pnum_inter
< intermediate
->total_sectors
)) {
1631 intermediate
= intermediate
->backing_hd
;
1638 int bdrv_write_compressed(BlockDriverState
*bs
, int64_t sector_num
,
1639 const uint8_t *buf
, int nb_sectors
)
1641 BlockDriver
*drv
= bs
->drv
;
1647 if (!drv
->bdrv_write_compressed
) {
1650 ret
= bdrv_check_request(bs
, sector_num
, nb_sectors
);
1655 assert(QLIST_EMPTY(&bs
->dirty_bitmaps
));
1657 return drv
->bdrv_write_compressed(bs
, sector_num
, buf
, nb_sectors
);
1660 int bdrv_save_vmstate(BlockDriverState
*bs
, const uint8_t *buf
,
1661 int64_t pos
, int size
)
1664 struct iovec iov
= {
1665 .iov_base
= (void *) buf
,
1669 qemu_iovec_init_external(&qiov
, &iov
, 1);
1670 return bdrv_writev_vmstate(bs
, &qiov
, pos
);
1673 int bdrv_writev_vmstate(BlockDriverState
*bs
, QEMUIOVector
*qiov
, int64_t pos
)
1675 BlockDriver
*drv
= bs
->drv
;
1679 } else if (drv
->bdrv_save_vmstate
) {
1680 return drv
->bdrv_save_vmstate(bs
, qiov
, pos
);
1681 } else if (bs
->file
) {
1682 return bdrv_writev_vmstate(bs
->file
, qiov
, pos
);
1688 int bdrv_load_vmstate(BlockDriverState
*bs
, uint8_t *buf
,
1689 int64_t pos
, int size
)
1691 BlockDriver
*drv
= bs
->drv
;
1694 if (drv
->bdrv_load_vmstate
)
1695 return drv
->bdrv_load_vmstate(bs
, buf
, pos
, size
);
1697 return bdrv_load_vmstate(bs
->file
, buf
, pos
, size
);
1701 /**************************************************************/
1704 BlockAIOCB
*bdrv_aio_readv(BlockDriverState
*bs
, int64_t sector_num
,
1705 QEMUIOVector
*qiov
, int nb_sectors
,
1706 BlockCompletionFunc
*cb
, void *opaque
)
1708 trace_bdrv_aio_readv(bs
, sector_num
, nb_sectors
, opaque
);
1710 return bdrv_co_aio_rw_vector(bs
, sector_num
, qiov
, nb_sectors
, 0,
1714 BlockAIOCB
*bdrv_aio_writev(BlockDriverState
*bs
, int64_t sector_num
,
1715 QEMUIOVector
*qiov
, int nb_sectors
,
1716 BlockCompletionFunc
*cb
, void *opaque
)
1718 trace_bdrv_aio_writev(bs
, sector_num
, nb_sectors
, opaque
);
1720 return bdrv_co_aio_rw_vector(bs
, sector_num
, qiov
, nb_sectors
, 0,
1724 BlockAIOCB
*bdrv_aio_write_zeroes(BlockDriverState
*bs
,
1725 int64_t sector_num
, int nb_sectors
, BdrvRequestFlags flags
,
1726 BlockCompletionFunc
*cb
, void *opaque
)
1728 trace_bdrv_aio_write_zeroes(bs
, sector_num
, nb_sectors
, flags
, opaque
);
1730 return bdrv_co_aio_rw_vector(bs
, sector_num
, NULL
, nb_sectors
,
1731 BDRV_REQ_ZERO_WRITE
| flags
,
1736 typedef struct MultiwriteCB
{
1741 BlockCompletionFunc
*cb
;
1743 QEMUIOVector
*free_qiov
;
1747 static void multiwrite_user_cb(MultiwriteCB
*mcb
)
1751 for (i
= 0; i
< mcb
->num_callbacks
; i
++) {
1752 mcb
->callbacks
[i
].cb(mcb
->callbacks
[i
].opaque
, mcb
->error
);
1753 if (mcb
->callbacks
[i
].free_qiov
) {
1754 qemu_iovec_destroy(mcb
->callbacks
[i
].free_qiov
);
1756 g_free(mcb
->callbacks
[i
].free_qiov
);
1760 static void multiwrite_cb(void *opaque
, int ret
)
1762 MultiwriteCB
*mcb
= opaque
;
1764 trace_multiwrite_cb(mcb
, ret
);
1766 if (ret
< 0 && !mcb
->error
) {
1770 mcb
->num_requests
--;
1771 if (mcb
->num_requests
== 0) {
1772 multiwrite_user_cb(mcb
);
1777 static int multiwrite_req_compare(const void *a
, const void *b
)
1779 const BlockRequest
*req1
= a
, *req2
= b
;
1782 * Note that we can't simply subtract req2->sector from req1->sector
1783 * here as that could overflow the return value.
1785 if (req1
->sector
> req2
->sector
) {
1787 } else if (req1
->sector
< req2
->sector
) {
1795 * Takes a bunch of requests and tries to merge them. Returns the number of
1796 * requests that remain after merging.
1798 static int multiwrite_merge(BlockDriverState
*bs
, BlockRequest
*reqs
,
1799 int num_reqs
, MultiwriteCB
*mcb
)
1803 // Sort requests by start sector
1804 qsort(reqs
, num_reqs
, sizeof(*reqs
), &multiwrite_req_compare
);
1806 // Check if adjacent requests touch the same clusters. If so, combine them,
1807 // filling up gaps with zero sectors.
1809 for (i
= 1; i
< num_reqs
; i
++) {
1811 int64_t oldreq_last
= reqs
[outidx
].sector
+ reqs
[outidx
].nb_sectors
;
1813 // Handle exactly sequential writes and overlapping writes.
1814 if (reqs
[i
].sector
<= oldreq_last
) {
1818 if (reqs
[outidx
].qiov
->niov
+ reqs
[i
].qiov
->niov
+ 1 > IOV_MAX
) {
1822 if (bs
->bl
.max_transfer_length
&& reqs
[outidx
].nb_sectors
+
1823 reqs
[i
].nb_sectors
> bs
->bl
.max_transfer_length
) {
1829 QEMUIOVector
*qiov
= g_malloc0(sizeof(*qiov
));
1830 qemu_iovec_init(qiov
,
1831 reqs
[outidx
].qiov
->niov
+ reqs
[i
].qiov
->niov
+ 1);
1833 // Add the first request to the merged one. If the requests are
1834 // overlapping, drop the last sectors of the first request.
1835 size
= (reqs
[i
].sector
- reqs
[outidx
].sector
) << 9;
1836 qemu_iovec_concat(qiov
, reqs
[outidx
].qiov
, 0, size
);
1838 // We should need to add any zeros between the two requests
1839 assert (reqs
[i
].sector
<= oldreq_last
);
1841 // Add the second request
1842 qemu_iovec_concat(qiov
, reqs
[i
].qiov
, 0, reqs
[i
].qiov
->size
);
1844 // Add tail of first request, if necessary
1845 if (qiov
->size
< reqs
[outidx
].qiov
->size
) {
1846 qemu_iovec_concat(qiov
, reqs
[outidx
].qiov
, qiov
->size
,
1847 reqs
[outidx
].qiov
->size
- qiov
->size
);
1850 reqs
[outidx
].nb_sectors
= qiov
->size
>> 9;
1851 reqs
[outidx
].qiov
= qiov
;
1853 mcb
->callbacks
[i
].free_qiov
= reqs
[outidx
].qiov
;
1856 reqs
[outidx
].sector
= reqs
[i
].sector
;
1857 reqs
[outidx
].nb_sectors
= reqs
[i
].nb_sectors
;
1858 reqs
[outidx
].qiov
= reqs
[i
].qiov
;
1862 block_acct_merge_done(&bs
->stats
, BLOCK_ACCT_WRITE
, num_reqs
- outidx
- 1);
1868 * Submit multiple AIO write requests at once.
1870 * On success, the function returns 0 and all requests in the reqs array have
1871 * been submitted. In error case this function returns -1, and any of the
1872 * requests may or may not be submitted yet. In particular, this means that the
1873 * callback will be called for some of the requests, for others it won't. The
1874 * caller must check the error field of the BlockRequest to wait for the right
1875 * callbacks (if error != 0, no callback will be called).
1877 * The implementation may modify the contents of the reqs array, e.g. to merge
1878 * requests. However, the fields opaque and error are left unmodified as they
1879 * are used to signal failure for a single request to the caller.
1881 int bdrv_aio_multiwrite(BlockDriverState
*bs
, BlockRequest
*reqs
, int num_reqs
)
1886 /* don't submit writes if we don't have a medium */
1887 if (bs
->drv
== NULL
) {
1888 for (i
= 0; i
< num_reqs
; i
++) {
1889 reqs
[i
].error
= -ENOMEDIUM
;
1894 if (num_reqs
== 0) {
1898 // Create MultiwriteCB structure
1899 mcb
= g_malloc0(sizeof(*mcb
) + num_reqs
* sizeof(*mcb
->callbacks
));
1900 mcb
->num_requests
= 0;
1901 mcb
->num_callbacks
= num_reqs
;
1903 for (i
= 0; i
< num_reqs
; i
++) {
1904 mcb
->callbacks
[i
].cb
= reqs
[i
].cb
;
1905 mcb
->callbacks
[i
].opaque
= reqs
[i
].opaque
;
1908 // Check for mergable requests
1909 num_reqs
= multiwrite_merge(bs
, reqs
, num_reqs
, mcb
);
1911 trace_bdrv_aio_multiwrite(mcb
, mcb
->num_callbacks
, num_reqs
);
1913 /* Run the aio requests. */
1914 mcb
->num_requests
= num_reqs
;
1915 for (i
= 0; i
< num_reqs
; i
++) {
1916 bdrv_co_aio_rw_vector(bs
, reqs
[i
].sector
, reqs
[i
].qiov
,
1917 reqs
[i
].nb_sectors
, reqs
[i
].flags
,
1925 void bdrv_aio_cancel(BlockAIOCB
*acb
)
1928 bdrv_aio_cancel_async(acb
);
1929 while (acb
->refcnt
> 1) {
1930 if (acb
->aiocb_info
->get_aio_context
) {
1931 aio_poll(acb
->aiocb_info
->get_aio_context(acb
), true);
1932 } else if (acb
->bs
) {
1933 aio_poll(bdrv_get_aio_context(acb
->bs
), true);
1938 qemu_aio_unref(acb
);
1941 /* Async version of aio cancel. The caller is not blocked if the acb implements
1942 * cancel_async, otherwise we do nothing and let the request normally complete.
1943 * In either case the completion callback must be called. */
1944 void bdrv_aio_cancel_async(BlockAIOCB
*acb
)
1946 if (acb
->aiocb_info
->cancel_async
) {
1947 acb
->aiocb_info
->cancel_async(acb
);
1951 /**************************************************************/
1952 /* async block device emulation */
1954 typedef struct BlockAIOCBSync
{
1958 /* vector translation state */
1964 static const AIOCBInfo bdrv_em_aiocb_info
= {
1965 .aiocb_size
= sizeof(BlockAIOCBSync
),
1968 static void bdrv_aio_bh_cb(void *opaque
)
1970 BlockAIOCBSync
*acb
= opaque
;
1972 if (!acb
->is_write
&& acb
->ret
>= 0) {
1973 qemu_iovec_from_buf(acb
->qiov
, 0, acb
->bounce
, acb
->qiov
->size
);
1975 qemu_vfree(acb
->bounce
);
1976 acb
->common
.cb(acb
->common
.opaque
, acb
->ret
);
1977 qemu_bh_delete(acb
->bh
);
1979 qemu_aio_unref(acb
);
1982 static BlockAIOCB
*bdrv_aio_rw_vector(BlockDriverState
*bs
,
1986 BlockCompletionFunc
*cb
,
1991 BlockAIOCBSync
*acb
;
1993 acb
= qemu_aio_get(&bdrv_em_aiocb_info
, bs
, cb
, opaque
);
1994 acb
->is_write
= is_write
;
1996 acb
->bounce
= qemu_try_blockalign(bs
, qiov
->size
);
1997 acb
->bh
= aio_bh_new(bdrv_get_aio_context(bs
), bdrv_aio_bh_cb
, acb
);
1999 if (acb
->bounce
== NULL
) {
2001 } else if (is_write
) {
2002 qemu_iovec_to_buf(acb
->qiov
, 0, acb
->bounce
, qiov
->size
);
2003 acb
->ret
= bs
->drv
->bdrv_write(bs
, sector_num
, acb
->bounce
, nb_sectors
);
2005 acb
->ret
= bs
->drv
->bdrv_read(bs
, sector_num
, acb
->bounce
, nb_sectors
);
2008 qemu_bh_schedule(acb
->bh
);
2010 return &acb
->common
;
2013 static BlockAIOCB
*bdrv_aio_readv_em(BlockDriverState
*bs
,
2014 int64_t sector_num
, QEMUIOVector
*qiov
, int nb_sectors
,
2015 BlockCompletionFunc
*cb
, void *opaque
)
2017 return bdrv_aio_rw_vector(bs
, sector_num
, qiov
, nb_sectors
, cb
, opaque
, 0);
2020 static BlockAIOCB
*bdrv_aio_writev_em(BlockDriverState
*bs
,
2021 int64_t sector_num
, QEMUIOVector
*qiov
, int nb_sectors
,
2022 BlockCompletionFunc
*cb
, void *opaque
)
2024 return bdrv_aio_rw_vector(bs
, sector_num
, qiov
, nb_sectors
, cb
, opaque
, 1);
2028 typedef struct BlockAIOCBCoroutine
{
2035 } BlockAIOCBCoroutine
;
2037 static const AIOCBInfo bdrv_em_co_aiocb_info
= {
2038 .aiocb_size
= sizeof(BlockAIOCBCoroutine
),
2041 static void bdrv_co_complete(BlockAIOCBCoroutine
*acb
)
2043 if (!acb
->need_bh
) {
2044 acb
->common
.cb(acb
->common
.opaque
, acb
->req
.error
);
2045 qemu_aio_unref(acb
);
2049 static void bdrv_co_em_bh(void *opaque
)
2051 BlockAIOCBCoroutine
*acb
= opaque
;
2053 assert(!acb
->need_bh
);
2054 qemu_bh_delete(acb
->bh
);
2055 bdrv_co_complete(acb
);
2058 static void bdrv_co_maybe_schedule_bh(BlockAIOCBCoroutine
*acb
)
2060 acb
->need_bh
= false;
2061 if (acb
->req
.error
!= -EINPROGRESS
) {
2062 BlockDriverState
*bs
= acb
->common
.bs
;
2064 acb
->bh
= aio_bh_new(bdrv_get_aio_context(bs
), bdrv_co_em_bh
, acb
);
2065 qemu_bh_schedule(acb
->bh
);
2069 /* Invoke bdrv_co_do_readv/bdrv_co_do_writev */
2070 static void coroutine_fn
bdrv_co_do_rw(void *opaque
)
2072 BlockAIOCBCoroutine
*acb
= opaque
;
2073 BlockDriverState
*bs
= acb
->common
.bs
;
2075 if (!acb
->is_write
) {
2076 acb
->req
.error
= bdrv_co_do_readv(bs
, acb
->req
.sector
,
2077 acb
->req
.nb_sectors
, acb
->req
.qiov
, acb
->req
.flags
);
2079 acb
->req
.error
= bdrv_co_do_writev(bs
, acb
->req
.sector
,
2080 acb
->req
.nb_sectors
, acb
->req
.qiov
, acb
->req
.flags
);
2083 bdrv_co_complete(acb
);
2086 static BlockAIOCB
*bdrv_co_aio_rw_vector(BlockDriverState
*bs
,
2090 BdrvRequestFlags flags
,
2091 BlockCompletionFunc
*cb
,
2096 BlockAIOCBCoroutine
*acb
;
2098 acb
= qemu_aio_get(&bdrv_em_co_aiocb_info
, bs
, cb
, opaque
);
2099 acb
->need_bh
= true;
2100 acb
->req
.error
= -EINPROGRESS
;
2101 acb
->req
.sector
= sector_num
;
2102 acb
->req
.nb_sectors
= nb_sectors
;
2103 acb
->req
.qiov
= qiov
;
2104 acb
->req
.flags
= flags
;
2105 acb
->is_write
= is_write
;
2107 co
= qemu_coroutine_create(bdrv_co_do_rw
);
2108 qemu_coroutine_enter(co
, acb
);
2110 bdrv_co_maybe_schedule_bh(acb
);
2111 return &acb
->common
;
2114 static void coroutine_fn
bdrv_aio_flush_co_entry(void *opaque
)
2116 BlockAIOCBCoroutine
*acb
= opaque
;
2117 BlockDriverState
*bs
= acb
->common
.bs
;
2119 acb
->req
.error
= bdrv_co_flush(bs
);
2120 bdrv_co_complete(acb
);
2123 BlockAIOCB
*bdrv_aio_flush(BlockDriverState
*bs
,
2124 BlockCompletionFunc
*cb
, void *opaque
)
2126 trace_bdrv_aio_flush(bs
, opaque
);
2129 BlockAIOCBCoroutine
*acb
;
2131 acb
= qemu_aio_get(&bdrv_em_co_aiocb_info
, bs
, cb
, opaque
);
2132 acb
->need_bh
= true;
2133 acb
->req
.error
= -EINPROGRESS
;
2135 co
= qemu_coroutine_create(bdrv_aio_flush_co_entry
);
2136 qemu_coroutine_enter(co
, acb
);
2138 bdrv_co_maybe_schedule_bh(acb
);
2139 return &acb
->common
;
2142 static void coroutine_fn
bdrv_aio_discard_co_entry(void *opaque
)
2144 BlockAIOCBCoroutine
*acb
= opaque
;
2145 BlockDriverState
*bs
= acb
->common
.bs
;
2147 acb
->req
.error
= bdrv_co_discard(bs
, acb
->req
.sector
, acb
->req
.nb_sectors
);
2148 bdrv_co_complete(acb
);
2151 BlockAIOCB
*bdrv_aio_discard(BlockDriverState
*bs
,
2152 int64_t sector_num
, int nb_sectors
,
2153 BlockCompletionFunc
*cb
, void *opaque
)
2156 BlockAIOCBCoroutine
*acb
;
2158 trace_bdrv_aio_discard(bs
, sector_num
, nb_sectors
, opaque
);
2160 acb
= qemu_aio_get(&bdrv_em_co_aiocb_info
, bs
, cb
, opaque
);
2161 acb
->need_bh
= true;
2162 acb
->req
.error
= -EINPROGRESS
;
2163 acb
->req
.sector
= sector_num
;
2164 acb
->req
.nb_sectors
= nb_sectors
;
2165 co
= qemu_coroutine_create(bdrv_aio_discard_co_entry
);
2166 qemu_coroutine_enter(co
, acb
);
2168 bdrv_co_maybe_schedule_bh(acb
);
2169 return &acb
->common
;
2172 void *qemu_aio_get(const AIOCBInfo
*aiocb_info
, BlockDriverState
*bs
,
2173 BlockCompletionFunc
*cb
, void *opaque
)
2177 acb
= g_slice_alloc(aiocb_info
->aiocb_size
);
2178 acb
->aiocb_info
= aiocb_info
;
2181 acb
->opaque
= opaque
;
2186 void qemu_aio_ref(void *p
)
2188 BlockAIOCB
*acb
= p
;
2192 void qemu_aio_unref(void *p
)
2194 BlockAIOCB
*acb
= p
;
2195 assert(acb
->refcnt
> 0);
2196 if (--acb
->refcnt
== 0) {
2197 g_slice_free1(acb
->aiocb_info
->aiocb_size
, acb
);
2201 /**************************************************************/
2202 /* Coroutine block device emulation */
2204 typedef struct CoroutineIOCompletion
{
2205 Coroutine
*coroutine
;
2207 } CoroutineIOCompletion
;
2209 static void bdrv_co_io_em_complete(void *opaque
, int ret
)
2211 CoroutineIOCompletion
*co
= opaque
;
2214 qemu_coroutine_enter(co
->coroutine
, NULL
);
2217 static int coroutine_fn
bdrv_co_io_em(BlockDriverState
*bs
, int64_t sector_num
,
2218 int nb_sectors
, QEMUIOVector
*iov
,
2221 CoroutineIOCompletion co
= {
2222 .coroutine
= qemu_coroutine_self(),
2227 acb
= bs
->drv
->bdrv_aio_writev(bs
, sector_num
, iov
, nb_sectors
,
2228 bdrv_co_io_em_complete
, &co
);
2230 acb
= bs
->drv
->bdrv_aio_readv(bs
, sector_num
, iov
, nb_sectors
,
2231 bdrv_co_io_em_complete
, &co
);
2234 trace_bdrv_co_io_em(bs
, sector_num
, nb_sectors
, is_write
, acb
);
2238 qemu_coroutine_yield();
2243 static int coroutine_fn
bdrv_co_readv_em(BlockDriverState
*bs
,
2244 int64_t sector_num
, int nb_sectors
,
2247 return bdrv_co_io_em(bs
, sector_num
, nb_sectors
, iov
, false);
2250 static int coroutine_fn
bdrv_co_writev_em(BlockDriverState
*bs
,
2251 int64_t sector_num
, int nb_sectors
,
2254 return bdrv_co_io_em(bs
, sector_num
, nb_sectors
, iov
, true);
2257 static void coroutine_fn
bdrv_flush_co_entry(void *opaque
)
2259 RwCo
*rwco
= opaque
;
2261 rwco
->ret
= bdrv_co_flush(rwco
->bs
);
2264 int coroutine_fn
bdrv_co_flush(BlockDriverState
*bs
)
2268 if (!bs
|| !bdrv_is_inserted(bs
) || bdrv_is_read_only(bs
) ||
2273 /* Write back cached data to the OS even with cache=unsafe */
2274 BLKDBG_EVENT(bs
->file
, BLKDBG_FLUSH_TO_OS
);
2275 if (bs
->drv
->bdrv_co_flush_to_os
) {
2276 ret
= bs
->drv
->bdrv_co_flush_to_os(bs
);
2282 /* But don't actually force it to the disk with cache=unsafe */
2283 if (bs
->open_flags
& BDRV_O_NO_FLUSH
) {
2287 BLKDBG_EVENT(bs
->file
, BLKDBG_FLUSH_TO_DISK
);
2288 if (bs
->drv
->bdrv_co_flush_to_disk
) {
2289 ret
= bs
->drv
->bdrv_co_flush_to_disk(bs
);
2290 } else if (bs
->drv
->bdrv_aio_flush
) {
2292 CoroutineIOCompletion co
= {
2293 .coroutine
= qemu_coroutine_self(),
2296 acb
= bs
->drv
->bdrv_aio_flush(bs
, bdrv_co_io_em_complete
, &co
);
2300 qemu_coroutine_yield();
2305 * Some block drivers always operate in either writethrough or unsafe
2306 * mode and don't support bdrv_flush therefore. Usually qemu doesn't
2307 * know how the server works (because the behaviour is hardcoded or
2308 * depends on server-side configuration), so we can't ensure that
2309 * everything is safe on disk. Returning an error doesn't work because
2310 * that would break guests even if the server operates in writethrough
2313 * Let's hope the user knows what he's doing.
2321 /* Now flush the underlying protocol. It will also have BDRV_O_NO_FLUSH
2322 * in the case of cache=unsafe, so there are no useless flushes.
2325 return bdrv_co_flush(bs
->file
);
2328 int bdrv_flush(BlockDriverState
*bs
)
2336 if (qemu_in_coroutine()) {
2337 /* Fast-path if already in coroutine context */
2338 bdrv_flush_co_entry(&rwco
);
2340 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
2342 co
= qemu_coroutine_create(bdrv_flush_co_entry
);
2343 qemu_coroutine_enter(co
, &rwco
);
2344 while (rwco
.ret
== NOT_DONE
) {
2345 aio_poll(aio_context
, true);
2352 typedef struct DiscardCo
{
2353 BlockDriverState
*bs
;
2358 static void coroutine_fn
bdrv_discard_co_entry(void *opaque
)
2360 DiscardCo
*rwco
= opaque
;
2362 rwco
->ret
= bdrv_co_discard(rwco
->bs
, rwco
->sector_num
, rwco
->nb_sectors
);
2365 int coroutine_fn
bdrv_co_discard(BlockDriverState
*bs
, int64_t sector_num
,
2368 int max_discard
, ret
;
2374 ret
= bdrv_check_request(bs
, sector_num
, nb_sectors
);
2377 } else if (bs
->read_only
) {
2381 bdrv_reset_dirty(bs
, sector_num
, nb_sectors
);
2383 /* Do nothing if disabled. */
2384 if (!(bs
->open_flags
& BDRV_O_UNMAP
)) {
2388 if (!bs
->drv
->bdrv_co_discard
&& !bs
->drv
->bdrv_aio_discard
) {
2392 max_discard
= MIN_NON_ZERO(bs
->bl
.max_discard
, BDRV_REQUEST_MAX_SECTORS
);
2393 while (nb_sectors
> 0) {
2395 int num
= nb_sectors
;
2398 if (bs
->bl
.discard_alignment
&&
2399 num
>= bs
->bl
.discard_alignment
&&
2400 sector_num
% bs
->bl
.discard_alignment
) {
2401 if (num
> bs
->bl
.discard_alignment
) {
2402 num
= bs
->bl
.discard_alignment
;
2404 num
-= sector_num
% bs
->bl
.discard_alignment
;
2407 /* limit request size */
2408 if (num
> max_discard
) {
2412 if (bs
->drv
->bdrv_co_discard
) {
2413 ret
= bs
->drv
->bdrv_co_discard(bs
, sector_num
, num
);
2416 CoroutineIOCompletion co
= {
2417 .coroutine
= qemu_coroutine_self(),
2420 acb
= bs
->drv
->bdrv_aio_discard(bs
, sector_num
, nb_sectors
,
2421 bdrv_co_io_em_complete
, &co
);
2425 qemu_coroutine_yield();
2429 if (ret
&& ret
!= -ENOTSUP
) {
2439 int bdrv_discard(BlockDriverState
*bs
, int64_t sector_num
, int nb_sectors
)
2444 .sector_num
= sector_num
,
2445 .nb_sectors
= nb_sectors
,
2449 if (qemu_in_coroutine()) {
2450 /* Fast-path if already in coroutine context */
2451 bdrv_discard_co_entry(&rwco
);
2453 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
2455 co
= qemu_coroutine_create(bdrv_discard_co_entry
);
2456 qemu_coroutine_enter(co
, &rwco
);
2457 while (rwco
.ret
== NOT_DONE
) {
2458 aio_poll(aio_context
, true);
2465 /* needed for generic scsi interface */
2467 int bdrv_ioctl(BlockDriverState
*bs
, unsigned long int req
, void *buf
)
2469 BlockDriver
*drv
= bs
->drv
;
2471 if (drv
&& drv
->bdrv_ioctl
)
2472 return drv
->bdrv_ioctl(bs
, req
, buf
);
2476 BlockAIOCB
*bdrv_aio_ioctl(BlockDriverState
*bs
,
2477 unsigned long int req
, void *buf
,
2478 BlockCompletionFunc
*cb
, void *opaque
)
2480 BlockDriver
*drv
= bs
->drv
;
2482 if (drv
&& drv
->bdrv_aio_ioctl
)
2483 return drv
->bdrv_aio_ioctl(bs
, req
, buf
, cb
, opaque
);
2487 void *qemu_blockalign(BlockDriverState
*bs
, size_t size
)
2489 return qemu_memalign(bdrv_opt_mem_align(bs
), size
);
2492 void *qemu_blockalign0(BlockDriverState
*bs
, size_t size
)
2494 return memset(qemu_blockalign(bs
, size
), 0, size
);
2497 void *qemu_try_blockalign(BlockDriverState
*bs
, size_t size
)
2499 size_t align
= bdrv_opt_mem_align(bs
);
2501 /* Ensure that NULL is never returned on success */
2507 return qemu_try_memalign(align
, size
);
2510 void *qemu_try_blockalign0(BlockDriverState
*bs
, size_t size
)
2512 void *mem
= qemu_try_blockalign(bs
, size
);
2515 memset(mem
, 0, size
);
2522 * Check if all memory in this vector is sector aligned.
2524 bool bdrv_qiov_is_aligned(BlockDriverState
*bs
, QEMUIOVector
*qiov
)
2527 size_t alignment
= bdrv_min_mem_align(bs
);
2529 for (i
= 0; i
< qiov
->niov
; i
++) {
2530 if ((uintptr_t) qiov
->iov
[i
].iov_base
% alignment
) {
2533 if (qiov
->iov
[i
].iov_len
% alignment
) {
2541 void bdrv_add_before_write_notifier(BlockDriverState
*bs
,
2542 NotifierWithReturn
*notifier
)
2544 notifier_with_return_list_add(&bs
->before_write_notifiers
, notifier
);
2547 void bdrv_io_plug(BlockDriverState
*bs
)
2549 BlockDriver
*drv
= bs
->drv
;
2550 if (drv
&& drv
->bdrv_io_plug
) {
2551 drv
->bdrv_io_plug(bs
);
2552 } else if (bs
->file
) {
2553 bdrv_io_plug(bs
->file
);
2557 void bdrv_io_unplug(BlockDriverState
*bs
)
2559 BlockDriver
*drv
= bs
->drv
;
2560 if (drv
&& drv
->bdrv_io_unplug
) {
2561 drv
->bdrv_io_unplug(bs
);
2562 } else if (bs
->file
) {
2563 bdrv_io_unplug(bs
->file
);
2567 void bdrv_flush_io_queue(BlockDriverState
*bs
)
2569 BlockDriver
*drv
= bs
->drv
;
2570 if (drv
&& drv
->bdrv_flush_io_queue
) {
2571 drv
->bdrv_flush_io_queue(bs
);
2572 } else if (bs
->file
) {
2573 bdrv_flush_io_queue(bs
->file
);
2575 bdrv_start_throttled_reqs(bs
);