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
->bs
, &local_err
);
161 error_propagate(errp
, local_err
);
164 bs
->bl
.opt_transfer_length
= bs
->file
->bs
->bl
.opt_transfer_length
;
165 bs
->bl
.max_transfer_length
= bs
->file
->bs
->bl
.max_transfer_length
;
166 bs
->bl
.min_mem_alignment
= bs
->file
->bs
->bl
.min_mem_alignment
;
167 bs
->bl
.opt_mem_alignment
= bs
->file
->bs
->bl
.opt_mem_alignment
;
169 bs
->bl
.min_mem_alignment
= 512;
170 bs
->bl
.opt_mem_alignment
= getpagesize();
174 bdrv_refresh_limits(bs
->backing
->bs
, &local_err
);
176 error_propagate(errp
, local_err
);
179 bs
->bl
.opt_transfer_length
=
180 MAX(bs
->bl
.opt_transfer_length
,
181 bs
->backing
->bs
->bl
.opt_transfer_length
);
182 bs
->bl
.max_transfer_length
=
183 MIN_NON_ZERO(bs
->bl
.max_transfer_length
,
184 bs
->backing
->bs
->bl
.max_transfer_length
);
185 bs
->bl
.opt_mem_alignment
=
186 MAX(bs
->bl
.opt_mem_alignment
,
187 bs
->backing
->bs
->bl
.opt_mem_alignment
);
188 bs
->bl
.min_mem_alignment
=
189 MAX(bs
->bl
.min_mem_alignment
,
190 bs
->backing
->bs
->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
->bs
)) {
230 if (bs
->backing
&& bdrv_requests_pending(bs
->backing
->bs
)) {
237 * Wait for pending requests to complete on a single BlockDriverState subtree
239 * Note that unlike bdrv_drain_all(), the caller must hold the BlockDriverState
242 * Only this BlockDriverState's AioContext is run, so in-flight requests must
243 * not depend on events in other AioContexts. In that case, use
244 * bdrv_drain_all() instead.
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 void bdrv_drain_all(void)
266 /* Always run first iteration so any pending completion BHs run */
268 BlockDriverState
*bs
= NULL
;
269 GSList
*aio_ctxs
= NULL
, *ctx
;
271 while ((bs
= bdrv_next(bs
))) {
272 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
274 aio_context_acquire(aio_context
);
276 block_job_pause(bs
->job
);
278 aio_context_release(aio_context
);
280 if (!g_slist_find(aio_ctxs
, aio_context
)) {
281 aio_ctxs
= g_slist_prepend(aio_ctxs
, aio_context
);
285 /* Note that completion of an asynchronous I/O operation can trigger any
286 * number of other I/O operations on other devices---for example a
287 * coroutine can submit an I/O request to another device in response to
288 * request completion. Therefore we must keep looping until there was no
289 * more activity rather than simply draining each device independently.
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 /* Don't do copy-on-read if we read data before write operation */
936 if (bs
->copy_on_read
&& !(flags
& BDRV_REQ_NO_COPY_ON_READ
)) {
937 flags
|= BDRV_REQ_COPY_ON_READ
;
940 /* throttling disk I/O */
941 if (bs
->io_limits_enabled
) {
942 throttle_group_co_io_limits_intercept(bs
, bytes
, false);
945 /* Align read if necessary by padding qiov */
946 if (offset
& (align
- 1)) {
947 head_buf
= qemu_blockalign(bs
, align
);
948 qemu_iovec_init(&local_qiov
, qiov
->niov
+ 2);
949 qemu_iovec_add(&local_qiov
, head_buf
, offset
& (align
- 1));
950 qemu_iovec_concat(&local_qiov
, qiov
, 0, qiov
->size
);
951 use_local_qiov
= true;
953 bytes
+= offset
& (align
- 1);
954 offset
= offset
& ~(align
- 1);
957 if ((offset
+ bytes
) & (align
- 1)) {
958 if (!use_local_qiov
) {
959 qemu_iovec_init(&local_qiov
, qiov
->niov
+ 1);
960 qemu_iovec_concat(&local_qiov
, qiov
, 0, qiov
->size
);
961 use_local_qiov
= true;
963 tail_buf
= qemu_blockalign(bs
, align
);
964 qemu_iovec_add(&local_qiov
, tail_buf
,
965 align
- ((offset
+ bytes
) & (align
- 1)));
967 bytes
= ROUND_UP(bytes
, align
);
970 tracked_request_begin(&req
, bs
, offset
, bytes
, false);
971 ret
= bdrv_aligned_preadv(bs
, &req
, offset
, bytes
, align
,
972 use_local_qiov
? &local_qiov
: qiov
,
974 tracked_request_end(&req
);
976 if (use_local_qiov
) {
977 qemu_iovec_destroy(&local_qiov
);
978 qemu_vfree(head_buf
);
979 qemu_vfree(tail_buf
);
985 static int coroutine_fn
bdrv_co_do_readv(BlockDriverState
*bs
,
986 int64_t sector_num
, int nb_sectors
, QEMUIOVector
*qiov
,
987 BdrvRequestFlags flags
)
989 if (nb_sectors
< 0 || nb_sectors
> BDRV_REQUEST_MAX_SECTORS
) {
993 return bdrv_co_do_preadv(bs
, sector_num
<< BDRV_SECTOR_BITS
,
994 nb_sectors
<< BDRV_SECTOR_BITS
, qiov
, flags
);
997 int coroutine_fn
bdrv_co_readv(BlockDriverState
*bs
, int64_t sector_num
,
998 int nb_sectors
, QEMUIOVector
*qiov
)
1000 trace_bdrv_co_readv(bs
, sector_num
, nb_sectors
);
1002 return bdrv_co_do_readv(bs
, sector_num
, nb_sectors
, qiov
, 0);
1005 int coroutine_fn
bdrv_co_no_copy_on_readv(BlockDriverState
*bs
,
1006 int64_t sector_num
, int nb_sectors
, QEMUIOVector
*qiov
)
1008 trace_bdrv_co_no_copy_on_readv(bs
, sector_num
, nb_sectors
);
1010 return bdrv_co_do_readv(bs
, sector_num
, nb_sectors
, qiov
,
1011 BDRV_REQ_NO_COPY_ON_READ
);
1014 int coroutine_fn
bdrv_co_copy_on_readv(BlockDriverState
*bs
,
1015 int64_t sector_num
, int nb_sectors
, QEMUIOVector
*qiov
)
1017 trace_bdrv_co_copy_on_readv(bs
, sector_num
, nb_sectors
);
1019 return bdrv_co_do_readv(bs
, sector_num
, nb_sectors
, qiov
,
1020 BDRV_REQ_COPY_ON_READ
);
1023 #define MAX_WRITE_ZEROES_BOUNCE_BUFFER 32768
1025 static int coroutine_fn
bdrv_co_do_write_zeroes(BlockDriverState
*bs
,
1026 int64_t sector_num
, int nb_sectors
, BdrvRequestFlags flags
)
1028 BlockDriver
*drv
= bs
->drv
;
1030 struct iovec iov
= {0};
1033 int max_write_zeroes
= MIN_NON_ZERO(bs
->bl
.max_write_zeroes
,
1034 BDRV_REQUEST_MAX_SECTORS
);
1036 while (nb_sectors
> 0 && !ret
) {
1037 int num
= nb_sectors
;
1039 /* Align request. Block drivers can expect the "bulk" of the request
1042 if (bs
->bl
.write_zeroes_alignment
1043 && num
> bs
->bl
.write_zeroes_alignment
) {
1044 if (sector_num
% bs
->bl
.write_zeroes_alignment
!= 0) {
1045 /* Make a small request up to the first aligned sector. */
1046 num
= bs
->bl
.write_zeroes_alignment
;
1047 num
-= sector_num
% bs
->bl
.write_zeroes_alignment
;
1048 } else if ((sector_num
+ num
) % bs
->bl
.write_zeroes_alignment
!= 0) {
1049 /* Shorten the request to the last aligned sector. num cannot
1050 * underflow because num > bs->bl.write_zeroes_alignment.
1052 num
-= (sector_num
+ num
) % bs
->bl
.write_zeroes_alignment
;
1056 /* limit request size */
1057 if (num
> max_write_zeroes
) {
1058 num
= max_write_zeroes
;
1062 /* First try the efficient write zeroes operation */
1063 if (drv
->bdrv_co_write_zeroes
) {
1064 ret
= drv
->bdrv_co_write_zeroes(bs
, sector_num
, num
, flags
);
1067 if (ret
== -ENOTSUP
) {
1068 /* Fall back to bounce buffer if write zeroes is unsupported */
1069 int max_xfer_len
= MIN_NON_ZERO(bs
->bl
.max_transfer_length
,
1070 MAX_WRITE_ZEROES_BOUNCE_BUFFER
);
1071 num
= MIN(num
, max_xfer_len
);
1072 iov
.iov_len
= num
* BDRV_SECTOR_SIZE
;
1073 if (iov
.iov_base
== NULL
) {
1074 iov
.iov_base
= qemu_try_blockalign(bs
, num
* BDRV_SECTOR_SIZE
);
1075 if (iov
.iov_base
== NULL
) {
1079 memset(iov
.iov_base
, 0, num
* BDRV_SECTOR_SIZE
);
1081 qemu_iovec_init_external(&qiov
, &iov
, 1);
1083 ret
= drv
->bdrv_co_writev(bs
, sector_num
, num
, &qiov
);
1085 /* Keep bounce buffer around if it is big enough for all
1086 * all future requests.
1088 if (num
< max_xfer_len
) {
1089 qemu_vfree(iov
.iov_base
);
1090 iov
.iov_base
= NULL
;
1099 qemu_vfree(iov
.iov_base
);
1104 * Forwards an already correctly aligned write request to the BlockDriver.
1106 static int coroutine_fn
bdrv_aligned_pwritev(BlockDriverState
*bs
,
1107 BdrvTrackedRequest
*req
, int64_t offset
, unsigned int bytes
,
1108 QEMUIOVector
*qiov
, int flags
)
1110 BlockDriver
*drv
= bs
->drv
;
1114 int64_t sector_num
= offset
>> BDRV_SECTOR_BITS
;
1115 unsigned int nb_sectors
= bytes
>> BDRV_SECTOR_BITS
;
1117 assert((offset
& (BDRV_SECTOR_SIZE
- 1)) == 0);
1118 assert((bytes
& (BDRV_SECTOR_SIZE
- 1)) == 0);
1119 assert(!qiov
|| bytes
== qiov
->size
);
1121 waited
= wait_serialising_requests(req
);
1122 assert(!waited
|| !req
->serialising
);
1123 assert(req
->overlap_offset
<= offset
);
1124 assert(offset
+ bytes
<= req
->overlap_offset
+ req
->overlap_bytes
);
1126 ret
= notifier_with_return_list_notify(&bs
->before_write_notifiers
, req
);
1128 if (!ret
&& bs
->detect_zeroes
!= BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF
&&
1129 !(flags
& BDRV_REQ_ZERO_WRITE
) && drv
->bdrv_co_write_zeroes
&&
1130 qemu_iovec_is_zero(qiov
)) {
1131 flags
|= BDRV_REQ_ZERO_WRITE
;
1132 if (bs
->detect_zeroes
== BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP
) {
1133 flags
|= BDRV_REQ_MAY_UNMAP
;
1138 /* Do nothing, write notifier decided to fail this request */
1139 } else if (flags
& BDRV_REQ_ZERO_WRITE
) {
1140 bdrv_debug_event(bs
, BLKDBG_PWRITEV_ZERO
);
1141 ret
= bdrv_co_do_write_zeroes(bs
, sector_num
, nb_sectors
, flags
);
1143 bdrv_debug_event(bs
, BLKDBG_PWRITEV
);
1144 ret
= drv
->bdrv_co_writev(bs
, sector_num
, nb_sectors
, qiov
);
1146 bdrv_debug_event(bs
, BLKDBG_PWRITEV_DONE
);
1148 if (ret
== 0 && !bs
->enable_write_cache
) {
1149 ret
= bdrv_co_flush(bs
);
1152 bdrv_set_dirty(bs
, sector_num
, nb_sectors
);
1154 block_acct_highest_sector(&bs
->stats
, sector_num
, nb_sectors
);
1157 bs
->total_sectors
= MAX(bs
->total_sectors
, sector_num
+ nb_sectors
);
1163 static int coroutine_fn
bdrv_co_do_zero_pwritev(BlockDriverState
*bs
,
1166 BdrvRequestFlags flags
,
1167 BdrvTrackedRequest
*req
)
1169 uint8_t *buf
= NULL
;
1170 QEMUIOVector local_qiov
;
1172 uint64_t align
= MAX(BDRV_SECTOR_SIZE
, bs
->request_alignment
);
1173 unsigned int head_padding_bytes
, tail_padding_bytes
;
1176 head_padding_bytes
= offset
& (align
- 1);
1177 tail_padding_bytes
= align
- ((offset
+ bytes
) & (align
- 1));
1180 assert(flags
& BDRV_REQ_ZERO_WRITE
);
1181 if (head_padding_bytes
|| tail_padding_bytes
) {
1182 buf
= qemu_blockalign(bs
, align
);
1183 iov
= (struct iovec
) {
1187 qemu_iovec_init_external(&local_qiov
, &iov
, 1);
1189 if (head_padding_bytes
) {
1190 uint64_t zero_bytes
= MIN(bytes
, align
- head_padding_bytes
);
1192 /* RMW the unaligned part before head. */
1193 mark_request_serialising(req
, align
);
1194 wait_serialising_requests(req
);
1195 bdrv_debug_event(bs
, BLKDBG_PWRITEV_RMW_HEAD
);
1196 ret
= bdrv_aligned_preadv(bs
, req
, offset
& ~(align
- 1), align
,
1197 align
, &local_qiov
, 0);
1201 bdrv_debug_event(bs
, BLKDBG_PWRITEV_RMW_AFTER_HEAD
);
1203 memset(buf
+ head_padding_bytes
, 0, zero_bytes
);
1204 ret
= bdrv_aligned_pwritev(bs
, req
, offset
& ~(align
- 1), align
,
1206 flags
& ~BDRV_REQ_ZERO_WRITE
);
1210 offset
+= zero_bytes
;
1211 bytes
-= zero_bytes
;
1214 assert(!bytes
|| (offset
& (align
- 1)) == 0);
1215 if (bytes
>= align
) {
1216 /* Write the aligned part in the middle. */
1217 uint64_t aligned_bytes
= bytes
& ~(align
- 1);
1218 ret
= bdrv_aligned_pwritev(bs
, req
, offset
, aligned_bytes
,
1223 bytes
-= aligned_bytes
;
1224 offset
+= aligned_bytes
;
1227 assert(!bytes
|| (offset
& (align
- 1)) == 0);
1229 assert(align
== tail_padding_bytes
+ bytes
);
1230 /* RMW the unaligned part after tail. */
1231 mark_request_serialising(req
, align
);
1232 wait_serialising_requests(req
);
1233 bdrv_debug_event(bs
, BLKDBG_PWRITEV_RMW_TAIL
);
1234 ret
= bdrv_aligned_preadv(bs
, req
, offset
, align
,
1235 align
, &local_qiov
, 0);
1239 bdrv_debug_event(bs
, BLKDBG_PWRITEV_RMW_AFTER_TAIL
);
1241 memset(buf
, 0, bytes
);
1242 ret
= bdrv_aligned_pwritev(bs
, req
, offset
, align
,
1243 &local_qiov
, flags
& ~BDRV_REQ_ZERO_WRITE
);
1252 * Handle a write request in coroutine context
1254 static int coroutine_fn
bdrv_co_do_pwritev(BlockDriverState
*bs
,
1255 int64_t offset
, unsigned int bytes
, QEMUIOVector
*qiov
,
1256 BdrvRequestFlags flags
)
1258 BdrvTrackedRequest req
;
1259 /* TODO Lift BDRV_SECTOR_SIZE restriction in BlockDriver interface */
1260 uint64_t align
= MAX(BDRV_SECTOR_SIZE
, bs
->request_alignment
);
1261 uint8_t *head_buf
= NULL
;
1262 uint8_t *tail_buf
= NULL
;
1263 QEMUIOVector local_qiov
;
1264 bool use_local_qiov
= false;
1270 if (bs
->read_only
) {
1274 ret
= bdrv_check_byte_request(bs
, offset
, bytes
);
1279 /* throttling disk I/O */
1280 if (bs
->io_limits_enabled
) {
1281 throttle_group_co_io_limits_intercept(bs
, bytes
, true);
1285 * Align write if necessary by performing a read-modify-write cycle.
1286 * Pad qiov with the read parts and be sure to have a tracked request not
1287 * only for bdrv_aligned_pwritev, but also for the reads of the RMW cycle.
1289 tracked_request_begin(&req
, bs
, offset
, bytes
, true);
1292 ret
= bdrv_co_do_zero_pwritev(bs
, offset
, bytes
, flags
, &req
);
1296 if (offset
& (align
- 1)) {
1297 QEMUIOVector head_qiov
;
1298 struct iovec head_iov
;
1300 mark_request_serialising(&req
, align
);
1301 wait_serialising_requests(&req
);
1303 head_buf
= qemu_blockalign(bs
, align
);
1304 head_iov
= (struct iovec
) {
1305 .iov_base
= head_buf
,
1308 qemu_iovec_init_external(&head_qiov
, &head_iov
, 1);
1310 bdrv_debug_event(bs
, BLKDBG_PWRITEV_RMW_HEAD
);
1311 ret
= bdrv_aligned_preadv(bs
, &req
, offset
& ~(align
- 1), align
,
1312 align
, &head_qiov
, 0);
1316 bdrv_debug_event(bs
, BLKDBG_PWRITEV_RMW_AFTER_HEAD
);
1318 qemu_iovec_init(&local_qiov
, qiov
->niov
+ 2);
1319 qemu_iovec_add(&local_qiov
, head_buf
, offset
& (align
- 1));
1320 qemu_iovec_concat(&local_qiov
, qiov
, 0, qiov
->size
);
1321 use_local_qiov
= true;
1323 bytes
+= offset
& (align
- 1);
1324 offset
= offset
& ~(align
- 1);
1327 if ((offset
+ bytes
) & (align
- 1)) {
1328 QEMUIOVector tail_qiov
;
1329 struct iovec tail_iov
;
1333 mark_request_serialising(&req
, align
);
1334 waited
= wait_serialising_requests(&req
);
1335 assert(!waited
|| !use_local_qiov
);
1337 tail_buf
= qemu_blockalign(bs
, align
);
1338 tail_iov
= (struct iovec
) {
1339 .iov_base
= tail_buf
,
1342 qemu_iovec_init_external(&tail_qiov
, &tail_iov
, 1);
1344 bdrv_debug_event(bs
, BLKDBG_PWRITEV_RMW_TAIL
);
1345 ret
= bdrv_aligned_preadv(bs
, &req
, (offset
+ bytes
) & ~(align
- 1), align
,
1346 align
, &tail_qiov
, 0);
1350 bdrv_debug_event(bs
, BLKDBG_PWRITEV_RMW_AFTER_TAIL
);
1352 if (!use_local_qiov
) {
1353 qemu_iovec_init(&local_qiov
, qiov
->niov
+ 1);
1354 qemu_iovec_concat(&local_qiov
, qiov
, 0, qiov
->size
);
1355 use_local_qiov
= true;
1358 tail_bytes
= (offset
+ bytes
) & (align
- 1);
1359 qemu_iovec_add(&local_qiov
, tail_buf
+ tail_bytes
, align
- tail_bytes
);
1361 bytes
= ROUND_UP(bytes
, align
);
1364 ret
= bdrv_aligned_pwritev(bs
, &req
, offset
, bytes
,
1365 use_local_qiov
? &local_qiov
: qiov
,
1370 if (use_local_qiov
) {
1371 qemu_iovec_destroy(&local_qiov
);
1373 qemu_vfree(head_buf
);
1374 qemu_vfree(tail_buf
);
1376 tracked_request_end(&req
);
1380 static int coroutine_fn
bdrv_co_do_writev(BlockDriverState
*bs
,
1381 int64_t sector_num
, int nb_sectors
, QEMUIOVector
*qiov
,
1382 BdrvRequestFlags flags
)
1384 if (nb_sectors
< 0 || nb_sectors
> BDRV_REQUEST_MAX_SECTORS
) {
1388 return bdrv_co_do_pwritev(bs
, sector_num
<< BDRV_SECTOR_BITS
,
1389 nb_sectors
<< BDRV_SECTOR_BITS
, qiov
, flags
);
1392 int coroutine_fn
bdrv_co_writev(BlockDriverState
*bs
, int64_t sector_num
,
1393 int nb_sectors
, QEMUIOVector
*qiov
)
1395 trace_bdrv_co_writev(bs
, sector_num
, nb_sectors
);
1397 return bdrv_co_do_writev(bs
, sector_num
, nb_sectors
, qiov
, 0);
1400 int coroutine_fn
bdrv_co_write_zeroes(BlockDriverState
*bs
,
1401 int64_t sector_num
, int nb_sectors
,
1402 BdrvRequestFlags flags
)
1404 trace_bdrv_co_write_zeroes(bs
, sector_num
, nb_sectors
, flags
);
1406 if (!(bs
->open_flags
& BDRV_O_UNMAP
)) {
1407 flags
&= ~BDRV_REQ_MAY_UNMAP
;
1410 return bdrv_co_do_writev(bs
, sector_num
, nb_sectors
, NULL
,
1411 BDRV_REQ_ZERO_WRITE
| flags
);
1414 int bdrv_flush_all(void)
1416 BlockDriverState
*bs
= NULL
;
1419 while ((bs
= bdrv_next(bs
))) {
1420 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
1423 aio_context_acquire(aio_context
);
1424 ret
= bdrv_flush(bs
);
1425 if (ret
< 0 && !result
) {
1428 aio_context_release(aio_context
);
1434 typedef struct BdrvCoGetBlockStatusData
{
1435 BlockDriverState
*bs
;
1436 BlockDriverState
*base
;
1442 } BdrvCoGetBlockStatusData
;
1445 * Returns the allocation status of the specified sectors.
1446 * Drivers not implementing the functionality are assumed to not support
1447 * backing files, hence all their sectors are reported as allocated.
1449 * If 'sector_num' is beyond the end of the disk image the return value is 0
1450 * and 'pnum' is set to 0.
1452 * 'pnum' is set to the number of sectors (including and immediately following
1453 * the specified sector) that are known to be in the same
1454 * allocated/unallocated state.
1456 * 'nb_sectors' is the max value 'pnum' should be set to. If nb_sectors goes
1457 * beyond the end of the disk image it will be clamped.
1459 static int64_t coroutine_fn
bdrv_co_get_block_status(BlockDriverState
*bs
,
1461 int nb_sectors
, int *pnum
)
1463 int64_t total_sectors
;
1467 total_sectors
= bdrv_nb_sectors(bs
);
1468 if (total_sectors
< 0) {
1469 return total_sectors
;
1472 if (sector_num
>= total_sectors
) {
1477 n
= total_sectors
- sector_num
;
1478 if (n
< nb_sectors
) {
1482 if (!bs
->drv
->bdrv_co_get_block_status
) {
1484 ret
= BDRV_BLOCK_DATA
| BDRV_BLOCK_ALLOCATED
;
1485 if (bs
->drv
->protocol_name
) {
1486 ret
|= BDRV_BLOCK_OFFSET_VALID
| (sector_num
* BDRV_SECTOR_SIZE
);
1491 ret
= bs
->drv
->bdrv_co_get_block_status(bs
, sector_num
, nb_sectors
, pnum
);
1497 if (ret
& BDRV_BLOCK_RAW
) {
1498 assert(ret
& BDRV_BLOCK_OFFSET_VALID
);
1499 return bdrv_get_block_status(bs
->file
->bs
, ret
>> BDRV_SECTOR_BITS
,
1503 if (ret
& (BDRV_BLOCK_DATA
| BDRV_BLOCK_ZERO
)) {
1504 ret
|= BDRV_BLOCK_ALLOCATED
;
1506 if (bdrv_unallocated_blocks_are_zero(bs
)) {
1507 ret
|= BDRV_BLOCK_ZERO
;
1508 } else if (bs
->backing
) {
1509 BlockDriverState
*bs2
= bs
->backing
->bs
;
1510 int64_t nb_sectors2
= bdrv_nb_sectors(bs2
);
1511 if (nb_sectors2
>= 0 && sector_num
>= nb_sectors2
) {
1512 ret
|= BDRV_BLOCK_ZERO
;
1518 (ret
& BDRV_BLOCK_DATA
) && !(ret
& BDRV_BLOCK_ZERO
) &&
1519 (ret
& BDRV_BLOCK_OFFSET_VALID
)) {
1522 ret2
= bdrv_co_get_block_status(bs
->file
->bs
, ret
>> BDRV_SECTOR_BITS
,
1525 /* Ignore errors. This is just providing extra information, it
1526 * is useful but not necessary.
1529 /* !file_pnum indicates an offset at or beyond the EOF; it is
1530 * perfectly valid for the format block driver to point to such
1531 * offsets, so catch it and mark everything as zero */
1532 ret
|= BDRV_BLOCK_ZERO
;
1534 /* Limit request to the range reported by the protocol driver */
1536 ret
|= (ret2
& BDRV_BLOCK_ZERO
);
1544 static int64_t coroutine_fn
bdrv_co_get_block_status_above(BlockDriverState
*bs
,
1545 BlockDriverState
*base
,
1550 BlockDriverState
*p
;
1554 for (p
= bs
; p
!= base
; p
= backing_bs(p
)) {
1555 ret
= bdrv_co_get_block_status(p
, sector_num
, nb_sectors
, pnum
);
1556 if (ret
< 0 || ret
& BDRV_BLOCK_ALLOCATED
) {
1559 /* [sector_num, pnum] unallocated on this layer, which could be only
1560 * the first part of [sector_num, nb_sectors]. */
1561 nb_sectors
= MIN(nb_sectors
, *pnum
);
1566 /* Coroutine wrapper for bdrv_get_block_status_above() */
1567 static void coroutine_fn
bdrv_get_block_status_above_co_entry(void *opaque
)
1569 BdrvCoGetBlockStatusData
*data
= opaque
;
1571 data
->ret
= bdrv_co_get_block_status_above(data
->bs
, data
->base
,
1579 * Synchronous wrapper around bdrv_co_get_block_status_above().
1581 * See bdrv_co_get_block_status_above() for details.
1583 int64_t bdrv_get_block_status_above(BlockDriverState
*bs
,
1584 BlockDriverState
*base
,
1586 int nb_sectors
, int *pnum
)
1589 BdrvCoGetBlockStatusData data
= {
1592 .sector_num
= sector_num
,
1593 .nb_sectors
= nb_sectors
,
1598 if (qemu_in_coroutine()) {
1599 /* Fast-path if already in coroutine context */
1600 bdrv_get_block_status_above_co_entry(&data
);
1602 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
1604 co
= qemu_coroutine_create(bdrv_get_block_status_above_co_entry
);
1605 qemu_coroutine_enter(co
, &data
);
1606 while (!data
.done
) {
1607 aio_poll(aio_context
, true);
1613 int64_t bdrv_get_block_status(BlockDriverState
*bs
,
1615 int nb_sectors
, int *pnum
)
1617 return bdrv_get_block_status_above(bs
, backing_bs(bs
),
1618 sector_num
, nb_sectors
, pnum
);
1621 int coroutine_fn
bdrv_is_allocated(BlockDriverState
*bs
, int64_t sector_num
,
1622 int nb_sectors
, int *pnum
)
1624 int64_t ret
= bdrv_get_block_status(bs
, sector_num
, nb_sectors
, pnum
);
1628 return !!(ret
& BDRV_BLOCK_ALLOCATED
);
1632 * Given an image chain: ... -> [BASE] -> [INTER1] -> [INTER2] -> [TOP]
1634 * Return true if the given sector is allocated in any image between
1635 * BASE and TOP (inclusive). BASE can be NULL to check if the given
1636 * sector is allocated in any image of the chain. Return false otherwise.
1638 * 'pnum' is set to the number of sectors (including and immediately following
1639 * the specified sector) that are known to be in the same
1640 * allocated/unallocated state.
1643 int bdrv_is_allocated_above(BlockDriverState
*top
,
1644 BlockDriverState
*base
,
1646 int nb_sectors
, int *pnum
)
1648 BlockDriverState
*intermediate
;
1649 int ret
, n
= nb_sectors
;
1652 while (intermediate
&& intermediate
!= base
) {
1654 ret
= bdrv_is_allocated(intermediate
, sector_num
, nb_sectors
,
1664 * [sector_num, nb_sectors] is unallocated on top but intermediate
1667 * [sector_num+x, nr_sectors] allocated.
1669 if (n
> pnum_inter
&&
1670 (intermediate
== top
||
1671 sector_num
+ pnum_inter
< intermediate
->total_sectors
)) {
1675 intermediate
= backing_bs(intermediate
);
1682 int bdrv_write_compressed(BlockDriverState
*bs
, int64_t sector_num
,
1683 const uint8_t *buf
, int nb_sectors
)
1685 BlockDriver
*drv
= bs
->drv
;
1691 if (!drv
->bdrv_write_compressed
) {
1694 ret
= bdrv_check_request(bs
, sector_num
, nb_sectors
);
1699 assert(QLIST_EMPTY(&bs
->dirty_bitmaps
));
1701 return drv
->bdrv_write_compressed(bs
, sector_num
, buf
, nb_sectors
);
1704 int bdrv_save_vmstate(BlockDriverState
*bs
, const uint8_t *buf
,
1705 int64_t pos
, int size
)
1708 struct iovec iov
= {
1709 .iov_base
= (void *) buf
,
1713 qemu_iovec_init_external(&qiov
, &iov
, 1);
1714 return bdrv_writev_vmstate(bs
, &qiov
, pos
);
1717 int bdrv_writev_vmstate(BlockDriverState
*bs
, QEMUIOVector
*qiov
, int64_t pos
)
1719 BlockDriver
*drv
= bs
->drv
;
1723 } else if (drv
->bdrv_save_vmstate
) {
1724 return drv
->bdrv_save_vmstate(bs
, qiov
, pos
);
1725 } else if (bs
->file
) {
1726 return bdrv_writev_vmstate(bs
->file
->bs
, qiov
, pos
);
1732 int bdrv_load_vmstate(BlockDriverState
*bs
, uint8_t *buf
,
1733 int64_t pos
, int size
)
1735 BlockDriver
*drv
= bs
->drv
;
1738 if (drv
->bdrv_load_vmstate
)
1739 return drv
->bdrv_load_vmstate(bs
, buf
, pos
, size
);
1741 return bdrv_load_vmstate(bs
->file
->bs
, buf
, pos
, size
);
1745 /**************************************************************/
1748 BlockAIOCB
*bdrv_aio_readv(BlockDriverState
*bs
, int64_t sector_num
,
1749 QEMUIOVector
*qiov
, int nb_sectors
,
1750 BlockCompletionFunc
*cb
, void *opaque
)
1752 trace_bdrv_aio_readv(bs
, sector_num
, nb_sectors
, opaque
);
1754 return bdrv_co_aio_rw_vector(bs
, sector_num
, qiov
, nb_sectors
, 0,
1758 BlockAIOCB
*bdrv_aio_writev(BlockDriverState
*bs
, int64_t sector_num
,
1759 QEMUIOVector
*qiov
, int nb_sectors
,
1760 BlockCompletionFunc
*cb
, void *opaque
)
1762 trace_bdrv_aio_writev(bs
, sector_num
, nb_sectors
, opaque
);
1764 return bdrv_co_aio_rw_vector(bs
, sector_num
, qiov
, nb_sectors
, 0,
1768 BlockAIOCB
*bdrv_aio_write_zeroes(BlockDriverState
*bs
,
1769 int64_t sector_num
, int nb_sectors
, BdrvRequestFlags flags
,
1770 BlockCompletionFunc
*cb
, void *opaque
)
1772 trace_bdrv_aio_write_zeroes(bs
, sector_num
, nb_sectors
, flags
, opaque
);
1774 return bdrv_co_aio_rw_vector(bs
, sector_num
, NULL
, nb_sectors
,
1775 BDRV_REQ_ZERO_WRITE
| flags
,
1780 typedef struct MultiwriteCB
{
1785 BlockCompletionFunc
*cb
;
1787 QEMUIOVector
*free_qiov
;
1791 static void multiwrite_user_cb(MultiwriteCB
*mcb
)
1795 for (i
= 0; i
< mcb
->num_callbacks
; i
++) {
1796 mcb
->callbacks
[i
].cb(mcb
->callbacks
[i
].opaque
, mcb
->error
);
1797 if (mcb
->callbacks
[i
].free_qiov
) {
1798 qemu_iovec_destroy(mcb
->callbacks
[i
].free_qiov
);
1800 g_free(mcb
->callbacks
[i
].free_qiov
);
1804 static void multiwrite_cb(void *opaque
, int ret
)
1806 MultiwriteCB
*mcb
= opaque
;
1808 trace_multiwrite_cb(mcb
, ret
);
1810 if (ret
< 0 && !mcb
->error
) {
1814 mcb
->num_requests
--;
1815 if (mcb
->num_requests
== 0) {
1816 multiwrite_user_cb(mcb
);
1821 static int multiwrite_req_compare(const void *a
, const void *b
)
1823 const BlockRequest
*req1
= a
, *req2
= b
;
1826 * Note that we can't simply subtract req2->sector from req1->sector
1827 * here as that could overflow the return value.
1829 if (req1
->sector
> req2
->sector
) {
1831 } else if (req1
->sector
< req2
->sector
) {
1839 * Takes a bunch of requests and tries to merge them. Returns the number of
1840 * requests that remain after merging.
1842 static int multiwrite_merge(BlockDriverState
*bs
, BlockRequest
*reqs
,
1843 int num_reqs
, MultiwriteCB
*mcb
)
1847 // Sort requests by start sector
1848 qsort(reqs
, num_reqs
, sizeof(*reqs
), &multiwrite_req_compare
);
1850 // Check if adjacent requests touch the same clusters. If so, combine them,
1851 // filling up gaps with zero sectors.
1853 for (i
= 1; i
< num_reqs
; i
++) {
1855 int64_t oldreq_last
= reqs
[outidx
].sector
+ reqs
[outidx
].nb_sectors
;
1857 // Handle exactly sequential writes and overlapping writes.
1858 if (reqs
[i
].sector
<= oldreq_last
) {
1862 if (reqs
[outidx
].qiov
->niov
+ reqs
[i
].qiov
->niov
+ 1 > IOV_MAX
) {
1866 if (bs
->bl
.max_transfer_length
&& reqs
[outidx
].nb_sectors
+
1867 reqs
[i
].nb_sectors
> bs
->bl
.max_transfer_length
) {
1873 QEMUIOVector
*qiov
= g_malloc0(sizeof(*qiov
));
1874 qemu_iovec_init(qiov
,
1875 reqs
[outidx
].qiov
->niov
+ reqs
[i
].qiov
->niov
+ 1);
1877 // Add the first request to the merged one. If the requests are
1878 // overlapping, drop the last sectors of the first request.
1879 size
= (reqs
[i
].sector
- reqs
[outidx
].sector
) << 9;
1880 qemu_iovec_concat(qiov
, reqs
[outidx
].qiov
, 0, size
);
1882 // We should need to add any zeros between the two requests
1883 assert (reqs
[i
].sector
<= oldreq_last
);
1885 // Add the second request
1886 qemu_iovec_concat(qiov
, reqs
[i
].qiov
, 0, reqs
[i
].qiov
->size
);
1888 // Add tail of first request, if necessary
1889 if (qiov
->size
< reqs
[outidx
].qiov
->size
) {
1890 qemu_iovec_concat(qiov
, reqs
[outidx
].qiov
, qiov
->size
,
1891 reqs
[outidx
].qiov
->size
- qiov
->size
);
1894 reqs
[outidx
].nb_sectors
= qiov
->size
>> 9;
1895 reqs
[outidx
].qiov
= qiov
;
1897 mcb
->callbacks
[i
].free_qiov
= reqs
[outidx
].qiov
;
1900 reqs
[outidx
].sector
= reqs
[i
].sector
;
1901 reqs
[outidx
].nb_sectors
= reqs
[i
].nb_sectors
;
1902 reqs
[outidx
].qiov
= reqs
[i
].qiov
;
1906 block_acct_merge_done(&bs
->stats
, BLOCK_ACCT_WRITE
, num_reqs
- outidx
- 1);
1912 * Submit multiple AIO write requests at once.
1914 * On success, the function returns 0 and all requests in the reqs array have
1915 * been submitted. In error case this function returns -1, and any of the
1916 * requests may or may not be submitted yet. In particular, this means that the
1917 * callback will be called for some of the requests, for others it won't. The
1918 * caller must check the error field of the BlockRequest to wait for the right
1919 * callbacks (if error != 0, no callback will be called).
1921 * The implementation may modify the contents of the reqs array, e.g. to merge
1922 * requests. However, the fields opaque and error are left unmodified as they
1923 * are used to signal failure for a single request to the caller.
1925 int bdrv_aio_multiwrite(BlockDriverState
*bs
, BlockRequest
*reqs
, int num_reqs
)
1930 /* don't submit writes if we don't have a medium */
1931 if (bs
->drv
== NULL
) {
1932 for (i
= 0; i
< num_reqs
; i
++) {
1933 reqs
[i
].error
= -ENOMEDIUM
;
1938 if (num_reqs
== 0) {
1942 // Create MultiwriteCB structure
1943 mcb
= g_malloc0(sizeof(*mcb
) + num_reqs
* sizeof(*mcb
->callbacks
));
1944 mcb
->num_requests
= 0;
1945 mcb
->num_callbacks
= num_reqs
;
1947 for (i
= 0; i
< num_reqs
; i
++) {
1948 mcb
->callbacks
[i
].cb
= reqs
[i
].cb
;
1949 mcb
->callbacks
[i
].opaque
= reqs
[i
].opaque
;
1952 // Check for mergable requests
1953 num_reqs
= multiwrite_merge(bs
, reqs
, num_reqs
, mcb
);
1955 trace_bdrv_aio_multiwrite(mcb
, mcb
->num_callbacks
, num_reqs
);
1957 /* Run the aio requests. */
1958 mcb
->num_requests
= num_reqs
;
1959 for (i
= 0; i
< num_reqs
; i
++) {
1960 bdrv_co_aio_rw_vector(bs
, reqs
[i
].sector
, reqs
[i
].qiov
,
1961 reqs
[i
].nb_sectors
, reqs
[i
].flags
,
1969 void bdrv_aio_cancel(BlockAIOCB
*acb
)
1972 bdrv_aio_cancel_async(acb
);
1973 while (acb
->refcnt
> 1) {
1974 if (acb
->aiocb_info
->get_aio_context
) {
1975 aio_poll(acb
->aiocb_info
->get_aio_context(acb
), true);
1976 } else if (acb
->bs
) {
1977 aio_poll(bdrv_get_aio_context(acb
->bs
), true);
1982 qemu_aio_unref(acb
);
1985 /* Async version of aio cancel. The caller is not blocked if the acb implements
1986 * cancel_async, otherwise we do nothing and let the request normally complete.
1987 * In either case the completion callback must be called. */
1988 void bdrv_aio_cancel_async(BlockAIOCB
*acb
)
1990 if (acb
->aiocb_info
->cancel_async
) {
1991 acb
->aiocb_info
->cancel_async(acb
);
1995 /**************************************************************/
1996 /* async block device emulation */
1998 typedef struct BlockAIOCBSync
{
2002 /* vector translation state */
2008 static const AIOCBInfo bdrv_em_aiocb_info
= {
2009 .aiocb_size
= sizeof(BlockAIOCBSync
),
2012 static void bdrv_aio_bh_cb(void *opaque
)
2014 BlockAIOCBSync
*acb
= opaque
;
2016 if (!acb
->is_write
&& acb
->ret
>= 0) {
2017 qemu_iovec_from_buf(acb
->qiov
, 0, acb
->bounce
, acb
->qiov
->size
);
2019 qemu_vfree(acb
->bounce
);
2020 acb
->common
.cb(acb
->common
.opaque
, acb
->ret
);
2021 qemu_bh_delete(acb
->bh
);
2023 qemu_aio_unref(acb
);
2026 static BlockAIOCB
*bdrv_aio_rw_vector(BlockDriverState
*bs
,
2030 BlockCompletionFunc
*cb
,
2035 BlockAIOCBSync
*acb
;
2037 acb
= qemu_aio_get(&bdrv_em_aiocb_info
, bs
, cb
, opaque
);
2038 acb
->is_write
= is_write
;
2040 acb
->bounce
= qemu_try_blockalign(bs
, qiov
->size
);
2041 acb
->bh
= aio_bh_new(bdrv_get_aio_context(bs
), bdrv_aio_bh_cb
, acb
);
2043 if (acb
->bounce
== NULL
) {
2045 } else if (is_write
) {
2046 qemu_iovec_to_buf(acb
->qiov
, 0, acb
->bounce
, qiov
->size
);
2047 acb
->ret
= bs
->drv
->bdrv_write(bs
, sector_num
, acb
->bounce
, nb_sectors
);
2049 acb
->ret
= bs
->drv
->bdrv_read(bs
, sector_num
, acb
->bounce
, nb_sectors
);
2052 qemu_bh_schedule(acb
->bh
);
2054 return &acb
->common
;
2057 static BlockAIOCB
*bdrv_aio_readv_em(BlockDriverState
*bs
,
2058 int64_t sector_num
, QEMUIOVector
*qiov
, int nb_sectors
,
2059 BlockCompletionFunc
*cb
, void *opaque
)
2061 return bdrv_aio_rw_vector(bs
, sector_num
, qiov
, nb_sectors
, cb
, opaque
, 0);
2064 static BlockAIOCB
*bdrv_aio_writev_em(BlockDriverState
*bs
,
2065 int64_t sector_num
, QEMUIOVector
*qiov
, int nb_sectors
,
2066 BlockCompletionFunc
*cb
, void *opaque
)
2068 return bdrv_aio_rw_vector(bs
, sector_num
, qiov
, nb_sectors
, cb
, opaque
, 1);
2072 typedef struct BlockAIOCBCoroutine
{
2079 } BlockAIOCBCoroutine
;
2081 static const AIOCBInfo bdrv_em_co_aiocb_info
= {
2082 .aiocb_size
= sizeof(BlockAIOCBCoroutine
),
2085 static void bdrv_co_complete(BlockAIOCBCoroutine
*acb
)
2087 if (!acb
->need_bh
) {
2088 acb
->common
.cb(acb
->common
.opaque
, acb
->req
.error
);
2089 qemu_aio_unref(acb
);
2093 static void bdrv_co_em_bh(void *opaque
)
2095 BlockAIOCBCoroutine
*acb
= opaque
;
2097 assert(!acb
->need_bh
);
2098 qemu_bh_delete(acb
->bh
);
2099 bdrv_co_complete(acb
);
2102 static void bdrv_co_maybe_schedule_bh(BlockAIOCBCoroutine
*acb
)
2104 acb
->need_bh
= false;
2105 if (acb
->req
.error
!= -EINPROGRESS
) {
2106 BlockDriverState
*bs
= acb
->common
.bs
;
2108 acb
->bh
= aio_bh_new(bdrv_get_aio_context(bs
), bdrv_co_em_bh
, acb
);
2109 qemu_bh_schedule(acb
->bh
);
2113 /* Invoke bdrv_co_do_readv/bdrv_co_do_writev */
2114 static void coroutine_fn
bdrv_co_do_rw(void *opaque
)
2116 BlockAIOCBCoroutine
*acb
= opaque
;
2117 BlockDriverState
*bs
= acb
->common
.bs
;
2119 if (!acb
->is_write
) {
2120 acb
->req
.error
= bdrv_co_do_readv(bs
, acb
->req
.sector
,
2121 acb
->req
.nb_sectors
, acb
->req
.qiov
, acb
->req
.flags
);
2123 acb
->req
.error
= bdrv_co_do_writev(bs
, acb
->req
.sector
,
2124 acb
->req
.nb_sectors
, acb
->req
.qiov
, acb
->req
.flags
);
2127 bdrv_co_complete(acb
);
2130 static BlockAIOCB
*bdrv_co_aio_rw_vector(BlockDriverState
*bs
,
2134 BdrvRequestFlags flags
,
2135 BlockCompletionFunc
*cb
,
2140 BlockAIOCBCoroutine
*acb
;
2142 acb
= qemu_aio_get(&bdrv_em_co_aiocb_info
, bs
, cb
, opaque
);
2143 acb
->need_bh
= true;
2144 acb
->req
.error
= -EINPROGRESS
;
2145 acb
->req
.sector
= sector_num
;
2146 acb
->req
.nb_sectors
= nb_sectors
;
2147 acb
->req
.qiov
= qiov
;
2148 acb
->req
.flags
= flags
;
2149 acb
->is_write
= is_write
;
2151 co
= qemu_coroutine_create(bdrv_co_do_rw
);
2152 qemu_coroutine_enter(co
, acb
);
2154 bdrv_co_maybe_schedule_bh(acb
);
2155 return &acb
->common
;
2158 static void coroutine_fn
bdrv_aio_flush_co_entry(void *opaque
)
2160 BlockAIOCBCoroutine
*acb
= opaque
;
2161 BlockDriverState
*bs
= acb
->common
.bs
;
2163 acb
->req
.error
= bdrv_co_flush(bs
);
2164 bdrv_co_complete(acb
);
2167 BlockAIOCB
*bdrv_aio_flush(BlockDriverState
*bs
,
2168 BlockCompletionFunc
*cb
, void *opaque
)
2170 trace_bdrv_aio_flush(bs
, opaque
);
2173 BlockAIOCBCoroutine
*acb
;
2175 acb
= qemu_aio_get(&bdrv_em_co_aiocb_info
, bs
, cb
, opaque
);
2176 acb
->need_bh
= true;
2177 acb
->req
.error
= -EINPROGRESS
;
2179 co
= qemu_coroutine_create(bdrv_aio_flush_co_entry
);
2180 qemu_coroutine_enter(co
, acb
);
2182 bdrv_co_maybe_schedule_bh(acb
);
2183 return &acb
->common
;
2186 static void coroutine_fn
bdrv_aio_discard_co_entry(void *opaque
)
2188 BlockAIOCBCoroutine
*acb
= opaque
;
2189 BlockDriverState
*bs
= acb
->common
.bs
;
2191 acb
->req
.error
= bdrv_co_discard(bs
, acb
->req
.sector
, acb
->req
.nb_sectors
);
2192 bdrv_co_complete(acb
);
2195 BlockAIOCB
*bdrv_aio_discard(BlockDriverState
*bs
,
2196 int64_t sector_num
, int nb_sectors
,
2197 BlockCompletionFunc
*cb
, void *opaque
)
2200 BlockAIOCBCoroutine
*acb
;
2202 trace_bdrv_aio_discard(bs
, sector_num
, nb_sectors
, opaque
);
2204 acb
= qemu_aio_get(&bdrv_em_co_aiocb_info
, bs
, cb
, opaque
);
2205 acb
->need_bh
= true;
2206 acb
->req
.error
= -EINPROGRESS
;
2207 acb
->req
.sector
= sector_num
;
2208 acb
->req
.nb_sectors
= nb_sectors
;
2209 co
= qemu_coroutine_create(bdrv_aio_discard_co_entry
);
2210 qemu_coroutine_enter(co
, acb
);
2212 bdrv_co_maybe_schedule_bh(acb
);
2213 return &acb
->common
;
2216 void *qemu_aio_get(const AIOCBInfo
*aiocb_info
, BlockDriverState
*bs
,
2217 BlockCompletionFunc
*cb
, void *opaque
)
2221 acb
= g_malloc(aiocb_info
->aiocb_size
);
2222 acb
->aiocb_info
= aiocb_info
;
2225 acb
->opaque
= opaque
;
2230 void qemu_aio_ref(void *p
)
2232 BlockAIOCB
*acb
= p
;
2236 void qemu_aio_unref(void *p
)
2238 BlockAIOCB
*acb
= p
;
2239 assert(acb
->refcnt
> 0);
2240 if (--acb
->refcnt
== 0) {
2245 /**************************************************************/
2246 /* Coroutine block device emulation */
2248 typedef struct CoroutineIOCompletion
{
2249 Coroutine
*coroutine
;
2251 } CoroutineIOCompletion
;
2253 static void bdrv_co_io_em_complete(void *opaque
, int ret
)
2255 CoroutineIOCompletion
*co
= opaque
;
2258 qemu_coroutine_enter(co
->coroutine
, NULL
);
2261 static int coroutine_fn
bdrv_co_io_em(BlockDriverState
*bs
, int64_t sector_num
,
2262 int nb_sectors
, QEMUIOVector
*iov
,
2265 CoroutineIOCompletion co
= {
2266 .coroutine
= qemu_coroutine_self(),
2271 acb
= bs
->drv
->bdrv_aio_writev(bs
, sector_num
, iov
, nb_sectors
,
2272 bdrv_co_io_em_complete
, &co
);
2274 acb
= bs
->drv
->bdrv_aio_readv(bs
, sector_num
, iov
, nb_sectors
,
2275 bdrv_co_io_em_complete
, &co
);
2278 trace_bdrv_co_io_em(bs
, sector_num
, nb_sectors
, is_write
, acb
);
2282 qemu_coroutine_yield();
2287 static int coroutine_fn
bdrv_co_readv_em(BlockDriverState
*bs
,
2288 int64_t sector_num
, int nb_sectors
,
2291 return bdrv_co_io_em(bs
, sector_num
, nb_sectors
, iov
, false);
2294 static int coroutine_fn
bdrv_co_writev_em(BlockDriverState
*bs
,
2295 int64_t sector_num
, int nb_sectors
,
2298 return bdrv_co_io_em(bs
, sector_num
, nb_sectors
, iov
, true);
2301 static void coroutine_fn
bdrv_flush_co_entry(void *opaque
)
2303 RwCo
*rwco
= opaque
;
2305 rwco
->ret
= bdrv_co_flush(rwco
->bs
);
2308 int coroutine_fn
bdrv_co_flush(BlockDriverState
*bs
)
2312 if (!bs
|| !bdrv_is_inserted(bs
) || bdrv_is_read_only(bs
) ||
2317 /* Write back cached data to the OS even with cache=unsafe */
2318 BLKDBG_EVENT(bs
->file
, BLKDBG_FLUSH_TO_OS
);
2319 if (bs
->drv
->bdrv_co_flush_to_os
) {
2320 ret
= bs
->drv
->bdrv_co_flush_to_os(bs
);
2326 /* But don't actually force it to the disk with cache=unsafe */
2327 if (bs
->open_flags
& BDRV_O_NO_FLUSH
) {
2331 BLKDBG_EVENT(bs
->file
, BLKDBG_FLUSH_TO_DISK
);
2332 if (bs
->drv
->bdrv_co_flush_to_disk
) {
2333 ret
= bs
->drv
->bdrv_co_flush_to_disk(bs
);
2334 } else if (bs
->drv
->bdrv_aio_flush
) {
2336 CoroutineIOCompletion co
= {
2337 .coroutine
= qemu_coroutine_self(),
2340 acb
= bs
->drv
->bdrv_aio_flush(bs
, bdrv_co_io_em_complete
, &co
);
2344 qemu_coroutine_yield();
2349 * Some block drivers always operate in either writethrough or unsafe
2350 * mode and don't support bdrv_flush therefore. Usually qemu doesn't
2351 * know how the server works (because the behaviour is hardcoded or
2352 * depends on server-side configuration), so we can't ensure that
2353 * everything is safe on disk. Returning an error doesn't work because
2354 * that would break guests even if the server operates in writethrough
2357 * Let's hope the user knows what he's doing.
2365 /* Now flush the underlying protocol. It will also have BDRV_O_NO_FLUSH
2366 * in the case of cache=unsafe, so there are no useless flushes.
2369 return bs
->file
? bdrv_co_flush(bs
->file
->bs
) : 0;
2372 int bdrv_flush(BlockDriverState
*bs
)
2380 if (qemu_in_coroutine()) {
2381 /* Fast-path if already in coroutine context */
2382 bdrv_flush_co_entry(&rwco
);
2384 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
2386 co
= qemu_coroutine_create(bdrv_flush_co_entry
);
2387 qemu_coroutine_enter(co
, &rwco
);
2388 while (rwco
.ret
== NOT_DONE
) {
2389 aio_poll(aio_context
, true);
2396 typedef struct DiscardCo
{
2397 BlockDriverState
*bs
;
2402 static void coroutine_fn
bdrv_discard_co_entry(void *opaque
)
2404 DiscardCo
*rwco
= opaque
;
2406 rwco
->ret
= bdrv_co_discard(rwco
->bs
, rwco
->sector_num
, rwco
->nb_sectors
);
2409 int coroutine_fn
bdrv_co_discard(BlockDriverState
*bs
, int64_t sector_num
,
2412 int max_discard
, ret
;
2418 ret
= bdrv_check_request(bs
, sector_num
, nb_sectors
);
2421 } else if (bs
->read_only
) {
2425 /* Do nothing if disabled. */
2426 if (!(bs
->open_flags
& BDRV_O_UNMAP
)) {
2430 if (!bs
->drv
->bdrv_co_discard
&& !bs
->drv
->bdrv_aio_discard
) {
2434 bdrv_set_dirty(bs
, sector_num
, nb_sectors
);
2436 max_discard
= MIN_NON_ZERO(bs
->bl
.max_discard
, BDRV_REQUEST_MAX_SECTORS
);
2437 while (nb_sectors
> 0) {
2439 int num
= nb_sectors
;
2442 if (bs
->bl
.discard_alignment
&&
2443 num
>= bs
->bl
.discard_alignment
&&
2444 sector_num
% bs
->bl
.discard_alignment
) {
2445 if (num
> bs
->bl
.discard_alignment
) {
2446 num
= bs
->bl
.discard_alignment
;
2448 num
-= sector_num
% bs
->bl
.discard_alignment
;
2451 /* limit request size */
2452 if (num
> max_discard
) {
2456 if (bs
->drv
->bdrv_co_discard
) {
2457 ret
= bs
->drv
->bdrv_co_discard(bs
, sector_num
, num
);
2460 CoroutineIOCompletion co
= {
2461 .coroutine
= qemu_coroutine_self(),
2464 acb
= bs
->drv
->bdrv_aio_discard(bs
, sector_num
, nb_sectors
,
2465 bdrv_co_io_em_complete
, &co
);
2469 qemu_coroutine_yield();
2473 if (ret
&& ret
!= -ENOTSUP
) {
2483 int bdrv_discard(BlockDriverState
*bs
, int64_t sector_num
, int nb_sectors
)
2488 .sector_num
= sector_num
,
2489 .nb_sectors
= nb_sectors
,
2493 if (qemu_in_coroutine()) {
2494 /* Fast-path if already in coroutine context */
2495 bdrv_discard_co_entry(&rwco
);
2497 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
2499 co
= qemu_coroutine_create(bdrv_discard_co_entry
);
2500 qemu_coroutine_enter(co
, &rwco
);
2501 while (rwco
.ret
== NOT_DONE
) {
2502 aio_poll(aio_context
, true);
2509 /* needed for generic scsi interface */
2511 int bdrv_ioctl(BlockDriverState
*bs
, unsigned long int req
, void *buf
)
2513 BlockDriver
*drv
= bs
->drv
;
2515 if (drv
&& drv
->bdrv_ioctl
)
2516 return drv
->bdrv_ioctl(bs
, req
, buf
);
2520 BlockAIOCB
*bdrv_aio_ioctl(BlockDriverState
*bs
,
2521 unsigned long int req
, void *buf
,
2522 BlockCompletionFunc
*cb
, void *opaque
)
2524 BlockDriver
*drv
= bs
->drv
;
2526 if (drv
&& drv
->bdrv_aio_ioctl
)
2527 return drv
->bdrv_aio_ioctl(bs
, req
, buf
, cb
, opaque
);
2531 void *qemu_blockalign(BlockDriverState
*bs
, size_t size
)
2533 return qemu_memalign(bdrv_opt_mem_align(bs
), size
);
2536 void *qemu_blockalign0(BlockDriverState
*bs
, size_t size
)
2538 return memset(qemu_blockalign(bs
, size
), 0, size
);
2541 void *qemu_try_blockalign(BlockDriverState
*bs
, size_t size
)
2543 size_t align
= bdrv_opt_mem_align(bs
);
2545 /* Ensure that NULL is never returned on success */
2551 return qemu_try_memalign(align
, size
);
2554 void *qemu_try_blockalign0(BlockDriverState
*bs
, size_t size
)
2556 void *mem
= qemu_try_blockalign(bs
, size
);
2559 memset(mem
, 0, size
);
2566 * Check if all memory in this vector is sector aligned.
2568 bool bdrv_qiov_is_aligned(BlockDriverState
*bs
, QEMUIOVector
*qiov
)
2571 size_t alignment
= bdrv_min_mem_align(bs
);
2573 for (i
= 0; i
< qiov
->niov
; i
++) {
2574 if ((uintptr_t) qiov
->iov
[i
].iov_base
% alignment
) {
2577 if (qiov
->iov
[i
].iov_len
% alignment
) {
2585 void bdrv_add_before_write_notifier(BlockDriverState
*bs
,
2586 NotifierWithReturn
*notifier
)
2588 notifier_with_return_list_add(&bs
->before_write_notifiers
, notifier
);
2591 void bdrv_io_plug(BlockDriverState
*bs
)
2593 BlockDriver
*drv
= bs
->drv
;
2594 if (drv
&& drv
->bdrv_io_plug
) {
2595 drv
->bdrv_io_plug(bs
);
2596 } else if (bs
->file
) {
2597 bdrv_io_plug(bs
->file
->bs
);
2601 void bdrv_io_unplug(BlockDriverState
*bs
)
2603 BlockDriver
*drv
= bs
->drv
;
2604 if (drv
&& drv
->bdrv_io_unplug
) {
2605 drv
->bdrv_io_unplug(bs
);
2606 } else if (bs
->file
) {
2607 bdrv_io_unplug(bs
->file
->bs
);
2611 void bdrv_flush_io_queue(BlockDriverState
*bs
)
2613 BlockDriver
*drv
= bs
->drv
;
2614 if (drv
&& drv
->bdrv_flush_io_queue
) {
2615 drv
->bdrv_flush_io_queue(bs
);
2616 } else if (bs
->file
) {
2617 bdrv_flush_io_queue(bs
->file
->bs
);
2619 bdrv_start_throttled_reqs(bs
);