2 * Block layer I/O functions
4 * Copyright (c) 2003 Fabrice Bellard
6 * Permission is hereby granted, free of charge, to any person obtaining a copy
7 * of this software and associated documentation files (the "Software"), to deal
8 * in the Software without restriction, including without limitation the rights
9 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10 * copies of the Software, and to permit persons to whom the Software is
11 * furnished to do so, subject to the following conditions:
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
25 #include "qemu/osdep.h"
27 #include "sysemu/block-backend.h"
28 #include "block/blockjob.h"
29 #include "block/block_int.h"
30 #include "block/throttle-groups.h"
31 #include "qemu/error-report.h"
33 #define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */
35 static BlockAIOCB
*bdrv_aio_readv_em(BlockDriverState
*bs
,
36 int64_t sector_num
, QEMUIOVector
*qiov
, int nb_sectors
,
37 BlockCompletionFunc
*cb
, void *opaque
);
38 static BlockAIOCB
*bdrv_aio_writev_em(BlockDriverState
*bs
,
39 int64_t sector_num
, QEMUIOVector
*qiov
, int nb_sectors
,
40 BlockCompletionFunc
*cb
, void *opaque
);
41 static int coroutine_fn
bdrv_co_readv_em(BlockDriverState
*bs
,
42 int64_t sector_num
, int nb_sectors
,
44 static int coroutine_fn
bdrv_co_writev_em(BlockDriverState
*bs
,
45 int64_t sector_num
, int nb_sectors
,
47 static int coroutine_fn
bdrv_co_do_pwritev(BlockDriverState
*bs
,
48 int64_t offset
, unsigned int bytes
, QEMUIOVector
*qiov
,
49 BdrvRequestFlags flags
);
50 static BlockAIOCB
*bdrv_co_aio_rw_vector(BlockDriverState
*bs
,
54 BdrvRequestFlags flags
,
55 BlockCompletionFunc
*cb
,
58 static void coroutine_fn
bdrv_co_do_rw(void *opaque
);
59 static int coroutine_fn
bdrv_co_do_write_zeroes(BlockDriverState
*bs
,
60 int64_t sector_num
, int nb_sectors
, BdrvRequestFlags flags
);
62 /* throttling disk I/O limits */
63 void bdrv_set_io_limits(BlockDriverState
*bs
,
68 throttle_group_config(bs
, cfg
);
70 for (i
= 0; i
< 2; i
++) {
71 qemu_co_enter_next(&bs
->throttled_reqs
[i
]);
75 /* this function drain all the throttled IOs */
76 static bool bdrv_start_throttled_reqs(BlockDriverState
*bs
)
79 bool enabled
= bs
->io_limits_enabled
;
82 bs
->io_limits_enabled
= false;
84 for (i
= 0; i
< 2; i
++) {
85 while (qemu_co_enter_next(&bs
->throttled_reqs
[i
])) {
90 bs
->io_limits_enabled
= enabled
;
95 void bdrv_io_limits_disable(BlockDriverState
*bs
)
97 bs
->io_limits_enabled
= false;
98 bdrv_start_throttled_reqs(bs
);
99 throttle_group_unregister_bs(bs
);
102 /* should be called before bdrv_set_io_limits if a limit is set */
103 void bdrv_io_limits_enable(BlockDriverState
*bs
, const char *group
)
105 assert(!bs
->io_limits_enabled
);
106 throttle_group_register_bs(bs
, group
);
107 bs
->io_limits_enabled
= true;
110 void bdrv_io_limits_update_group(BlockDriverState
*bs
, const char *group
)
112 /* this bs is not part of any group */
113 if (!bs
->throttle_state
) {
117 /* this bs is a part of the same group than the one we want */
118 if (!g_strcmp0(throttle_group_get_name(bs
), group
)) {
122 /* need to change the group this bs belong to */
123 bdrv_io_limits_disable(bs
);
124 bdrv_io_limits_enable(bs
, group
);
127 void bdrv_setup_io_funcs(BlockDriver
*bdrv
)
129 /* Block drivers without coroutine functions need emulation */
130 if (!bdrv
->bdrv_co_readv
) {
131 bdrv
->bdrv_co_readv
= bdrv_co_readv_em
;
132 bdrv
->bdrv_co_writev
= bdrv_co_writev_em
;
134 /* bdrv_co_readv_em()/brdv_co_writev_em() work in terms of aio, so if
135 * the block driver lacks aio we need to emulate that too.
137 if (!bdrv
->bdrv_aio_readv
) {
138 /* add AIO emulation layer */
139 bdrv
->bdrv_aio_readv
= bdrv_aio_readv_em
;
140 bdrv
->bdrv_aio_writev
= bdrv_aio_writev_em
;
145 void bdrv_refresh_limits(BlockDriverState
*bs
, Error
**errp
)
147 BlockDriver
*drv
= bs
->drv
;
148 Error
*local_err
= NULL
;
150 memset(&bs
->bl
, 0, sizeof(bs
->bl
));
156 /* Take some limits from the children as a default */
158 bdrv_refresh_limits(bs
->file
->bs
, &local_err
);
160 error_propagate(errp
, local_err
);
163 bs
->bl
.opt_transfer_length
= bs
->file
->bs
->bl
.opt_transfer_length
;
164 bs
->bl
.max_transfer_length
= bs
->file
->bs
->bl
.max_transfer_length
;
165 bs
->bl
.min_mem_alignment
= bs
->file
->bs
->bl
.min_mem_alignment
;
166 bs
->bl
.opt_mem_alignment
= bs
->file
->bs
->bl
.opt_mem_alignment
;
167 bs
->bl
.max_iov
= bs
->file
->bs
->bl
.max_iov
;
169 bs
->bl
.min_mem_alignment
= 512;
170 bs
->bl
.opt_mem_alignment
= getpagesize();
172 /* Safe default since most protocols use readv()/writev()/etc */
173 bs
->bl
.max_iov
= IOV_MAX
;
177 bdrv_refresh_limits(bs
->backing
->bs
, &local_err
);
179 error_propagate(errp
, local_err
);
182 bs
->bl
.opt_transfer_length
=
183 MAX(bs
->bl
.opt_transfer_length
,
184 bs
->backing
->bs
->bl
.opt_transfer_length
);
185 bs
->bl
.max_transfer_length
=
186 MIN_NON_ZERO(bs
->bl
.max_transfer_length
,
187 bs
->backing
->bs
->bl
.max_transfer_length
);
188 bs
->bl
.opt_mem_alignment
=
189 MAX(bs
->bl
.opt_mem_alignment
,
190 bs
->backing
->bs
->bl
.opt_mem_alignment
);
191 bs
->bl
.min_mem_alignment
=
192 MAX(bs
->bl
.min_mem_alignment
,
193 bs
->backing
->bs
->bl
.min_mem_alignment
);
196 bs
->backing
->bs
->bl
.max_iov
);
199 /* Then let the driver override it */
200 if (drv
->bdrv_refresh_limits
) {
201 drv
->bdrv_refresh_limits(bs
, errp
);
206 * The copy-on-read flag is actually a reference count so multiple users may
207 * use the feature without worrying about clobbering its previous state.
208 * Copy-on-read stays enabled until all users have called to disable it.
210 void bdrv_enable_copy_on_read(BlockDriverState
*bs
)
215 void bdrv_disable_copy_on_read(BlockDriverState
*bs
)
217 assert(bs
->copy_on_read
> 0);
221 /* Check if any requests are in-flight (including throttled requests) */
222 bool bdrv_requests_pending(BlockDriverState
*bs
)
226 if (!QLIST_EMPTY(&bs
->tracked_requests
)) {
229 if (!qemu_co_queue_empty(&bs
->throttled_reqs
[0])) {
232 if (!qemu_co_queue_empty(&bs
->throttled_reqs
[1])) {
236 QLIST_FOREACH(child
, &bs
->children
, next
) {
237 if (bdrv_requests_pending(child
->bs
)) {
245 static void bdrv_drain_recurse(BlockDriverState
*bs
)
249 if (bs
->drv
&& bs
->drv
->bdrv_drain
) {
250 bs
->drv
->bdrv_drain(bs
);
252 QLIST_FOREACH(child
, &bs
->children
, next
) {
253 bdrv_drain_recurse(child
->bs
);
258 * Wait for pending requests to complete on a single BlockDriverState subtree,
259 * and suspend block driver's internal I/O until next request arrives.
261 * Note that unlike bdrv_drain_all(), the caller must hold the BlockDriverState
264 * Only this BlockDriverState's AioContext is run, so in-flight requests must
265 * not depend on events in other AioContexts. In that case, use
266 * bdrv_drain_all() instead.
268 void bdrv_drain(BlockDriverState
*bs
)
272 bdrv_drain_recurse(bs
);
275 bdrv_flush_io_queue(bs
);
276 busy
= bdrv_requests_pending(bs
);
277 busy
|= aio_poll(bdrv_get_aio_context(bs
), busy
);
282 * Wait for pending requests to complete across all BlockDriverStates
284 * This function does not flush data to disk, use bdrv_flush_all() for that
285 * after calling this function.
287 void bdrv_drain_all(void)
289 /* Always run first iteration so any pending completion BHs run */
291 BlockDriverState
*bs
= NULL
;
292 GSList
*aio_ctxs
= NULL
, *ctx
;
294 while ((bs
= bdrv_next(bs
))) {
295 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
297 aio_context_acquire(aio_context
);
299 block_job_pause(bs
->job
);
301 bdrv_drain_recurse(bs
);
302 aio_context_release(aio_context
);
304 if (!g_slist_find(aio_ctxs
, aio_context
)) {
305 aio_ctxs
= g_slist_prepend(aio_ctxs
, aio_context
);
309 /* Note that completion of an asynchronous I/O operation can trigger any
310 * number of other I/O operations on other devices---for example a
311 * coroutine can submit an I/O request to another device in response to
312 * request completion. Therefore we must keep looping until there was no
313 * more activity rather than simply draining each device independently.
318 for (ctx
= aio_ctxs
; ctx
!= NULL
; ctx
= ctx
->next
) {
319 AioContext
*aio_context
= ctx
->data
;
322 aio_context_acquire(aio_context
);
323 while ((bs
= bdrv_next(bs
))) {
324 if (aio_context
== bdrv_get_aio_context(bs
)) {
325 bdrv_flush_io_queue(bs
);
326 if (bdrv_requests_pending(bs
)) {
328 aio_poll(aio_context
, busy
);
332 busy
|= aio_poll(aio_context
, false);
333 aio_context_release(aio_context
);
338 while ((bs
= bdrv_next(bs
))) {
339 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
341 aio_context_acquire(aio_context
);
343 block_job_resume(bs
->job
);
345 aio_context_release(aio_context
);
347 g_slist_free(aio_ctxs
);
351 * Remove an active request from the tracked requests list
353 * This function should be called when a tracked request is completing.
355 static void tracked_request_end(BdrvTrackedRequest
*req
)
357 if (req
->serialising
) {
358 req
->bs
->serialising_in_flight
--;
361 QLIST_REMOVE(req
, list
);
362 qemu_co_queue_restart_all(&req
->wait_queue
);
366 * Add an active request to the tracked requests list
368 static void tracked_request_begin(BdrvTrackedRequest
*req
,
369 BlockDriverState
*bs
,
372 enum BdrvTrackedRequestType type
)
374 *req
= (BdrvTrackedRequest
){
379 .co
= qemu_coroutine_self(),
380 .serialising
= false,
381 .overlap_offset
= offset
,
382 .overlap_bytes
= bytes
,
385 qemu_co_queue_init(&req
->wait_queue
);
387 QLIST_INSERT_HEAD(&bs
->tracked_requests
, req
, list
);
390 static void mark_request_serialising(BdrvTrackedRequest
*req
, uint64_t align
)
392 int64_t overlap_offset
= req
->offset
& ~(align
- 1);
393 unsigned int overlap_bytes
= ROUND_UP(req
->offset
+ req
->bytes
, align
)
396 if (!req
->serialising
) {
397 req
->bs
->serialising_in_flight
++;
398 req
->serialising
= true;
401 req
->overlap_offset
= MIN(req
->overlap_offset
, overlap_offset
);
402 req
->overlap_bytes
= MAX(req
->overlap_bytes
, overlap_bytes
);
406 * Round a region to cluster boundaries
408 void bdrv_round_to_clusters(BlockDriverState
*bs
,
409 int64_t sector_num
, int nb_sectors
,
410 int64_t *cluster_sector_num
,
411 int *cluster_nb_sectors
)
415 if (bdrv_get_info(bs
, &bdi
) < 0 || bdi
.cluster_size
== 0) {
416 *cluster_sector_num
= sector_num
;
417 *cluster_nb_sectors
= nb_sectors
;
419 int64_t c
= bdi
.cluster_size
/ BDRV_SECTOR_SIZE
;
420 *cluster_sector_num
= QEMU_ALIGN_DOWN(sector_num
, c
);
421 *cluster_nb_sectors
= QEMU_ALIGN_UP(sector_num
- *cluster_sector_num
+
426 static int bdrv_get_cluster_size(BlockDriverState
*bs
)
431 ret
= bdrv_get_info(bs
, &bdi
);
432 if (ret
< 0 || bdi
.cluster_size
== 0) {
433 return bs
->request_alignment
;
435 return bdi
.cluster_size
;
439 static bool tracked_request_overlaps(BdrvTrackedRequest
*req
,
440 int64_t offset
, unsigned int bytes
)
443 if (offset
>= req
->overlap_offset
+ req
->overlap_bytes
) {
447 if (req
->overlap_offset
>= offset
+ bytes
) {
453 static bool coroutine_fn
wait_serialising_requests(BdrvTrackedRequest
*self
)
455 BlockDriverState
*bs
= self
->bs
;
456 BdrvTrackedRequest
*req
;
460 if (!bs
->serialising_in_flight
) {
466 QLIST_FOREACH(req
, &bs
->tracked_requests
, list
) {
467 if (req
== self
|| (!req
->serialising
&& !self
->serialising
)) {
470 if (tracked_request_overlaps(req
, self
->overlap_offset
,
471 self
->overlap_bytes
))
473 /* Hitting this means there was a reentrant request, for
474 * example, a block driver issuing nested requests. This must
475 * never happen since it means deadlock.
477 assert(qemu_coroutine_self() != req
->co
);
479 /* If the request is already (indirectly) waiting for us, or
480 * will wait for us as soon as it wakes up, then just go on
481 * (instead of producing a deadlock in the former case). */
482 if (!req
->waiting_for
) {
483 self
->waiting_for
= req
;
484 qemu_co_queue_wait(&req
->wait_queue
);
485 self
->waiting_for
= NULL
;
497 static int bdrv_check_byte_request(BlockDriverState
*bs
, int64_t offset
,
500 if (size
> BDRV_REQUEST_MAX_SECTORS
<< BDRV_SECTOR_BITS
) {
504 if (!bdrv_is_inserted(bs
)) {
515 static int bdrv_check_request(BlockDriverState
*bs
, int64_t sector_num
,
518 if (nb_sectors
< 0 || nb_sectors
> BDRV_REQUEST_MAX_SECTORS
) {
522 return bdrv_check_byte_request(bs
, sector_num
* BDRV_SECTOR_SIZE
,
523 nb_sectors
* BDRV_SECTOR_SIZE
);
526 typedef struct RwCo
{
527 BlockDriverState
*bs
;
532 BdrvRequestFlags flags
;
535 static void coroutine_fn
bdrv_rw_co_entry(void *opaque
)
539 if (!rwco
->is_write
) {
540 rwco
->ret
= bdrv_co_do_preadv(rwco
->bs
, rwco
->offset
,
541 rwco
->qiov
->size
, rwco
->qiov
,
544 rwco
->ret
= bdrv_co_do_pwritev(rwco
->bs
, rwco
->offset
,
545 rwco
->qiov
->size
, rwco
->qiov
,
551 * Process a vectored synchronous request using coroutines
553 static int bdrv_prwv_co(BlockDriverState
*bs
, int64_t offset
,
554 QEMUIOVector
*qiov
, bool is_write
,
555 BdrvRequestFlags flags
)
562 .is_write
= is_write
,
568 * In sync call context, when the vcpu is blocked, this throttling timer
569 * will not fire; so the I/O throttling function has to be disabled here
570 * if it has been enabled.
572 if (bs
->io_limits_enabled
) {
573 fprintf(stderr
, "Disabling I/O throttling on '%s' due "
574 "to synchronous I/O.\n", bdrv_get_device_name(bs
));
575 bdrv_io_limits_disable(bs
);
578 if (qemu_in_coroutine()) {
579 /* Fast-path if already in coroutine context */
580 bdrv_rw_co_entry(&rwco
);
582 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
584 co
= qemu_coroutine_create(bdrv_rw_co_entry
);
585 qemu_coroutine_enter(co
, &rwco
);
586 while (rwco
.ret
== NOT_DONE
) {
587 aio_poll(aio_context
, true);
594 * Process a synchronous request using coroutines
596 static int bdrv_rw_co(BlockDriverState
*bs
, int64_t sector_num
, uint8_t *buf
,
597 int nb_sectors
, bool is_write
, BdrvRequestFlags flags
)
601 .iov_base
= (void *)buf
,
602 .iov_len
= nb_sectors
* BDRV_SECTOR_SIZE
,
605 if (nb_sectors
< 0 || nb_sectors
> BDRV_REQUEST_MAX_SECTORS
) {
609 qemu_iovec_init_external(&qiov
, &iov
, 1);
610 return bdrv_prwv_co(bs
, sector_num
<< BDRV_SECTOR_BITS
,
611 &qiov
, is_write
, flags
);
614 /* return < 0 if error. See bdrv_write() for the return codes */
615 int bdrv_read(BlockDriverState
*bs
, int64_t sector_num
,
616 uint8_t *buf
, int nb_sectors
)
618 return bdrv_rw_co(bs
, sector_num
, buf
, nb_sectors
, false, 0);
621 /* Just like bdrv_read(), but with I/O throttling temporarily disabled */
622 int bdrv_read_unthrottled(BlockDriverState
*bs
, int64_t sector_num
,
623 uint8_t *buf
, int nb_sectors
)
628 enabled
= bs
->io_limits_enabled
;
629 bs
->io_limits_enabled
= false;
630 ret
= bdrv_read(bs
, sector_num
, buf
, nb_sectors
);
631 bs
->io_limits_enabled
= enabled
;
635 /* Return < 0 if error. Important errors are:
636 -EIO generic I/O error (may happen for all errors)
637 -ENOMEDIUM No media inserted.
638 -EINVAL Invalid sector number or nb_sectors
639 -EACCES Trying to write a read-only device
641 int bdrv_write(BlockDriverState
*bs
, int64_t sector_num
,
642 const uint8_t *buf
, int nb_sectors
)
644 return bdrv_rw_co(bs
, sector_num
, (uint8_t *)buf
, nb_sectors
, true, 0);
647 int bdrv_write_zeroes(BlockDriverState
*bs
, int64_t sector_num
,
648 int nb_sectors
, BdrvRequestFlags flags
)
650 return bdrv_rw_co(bs
, sector_num
, NULL
, nb_sectors
, true,
651 BDRV_REQ_ZERO_WRITE
| flags
);
655 * Completely zero out a block device with the help of bdrv_write_zeroes.
656 * The operation is sped up by checking the block status and only writing
657 * zeroes to the device if they currently do not return zeroes. Optional
658 * flags are passed through to bdrv_write_zeroes (e.g. BDRV_REQ_MAY_UNMAP).
660 * Returns < 0 on error, 0 on success. For error codes see bdrv_write().
662 int bdrv_make_zero(BlockDriverState
*bs
, BdrvRequestFlags flags
)
664 int64_t target_sectors
, ret
, nb_sectors
, sector_num
= 0;
665 BlockDriverState
*file
;
668 target_sectors
= bdrv_nb_sectors(bs
);
669 if (target_sectors
< 0) {
670 return target_sectors
;
674 nb_sectors
= MIN(target_sectors
- sector_num
, BDRV_REQUEST_MAX_SECTORS
);
675 if (nb_sectors
<= 0) {
678 ret
= bdrv_get_block_status(bs
, sector_num
, nb_sectors
, &n
, &file
);
680 error_report("error getting block status at sector %" PRId64
": %s",
681 sector_num
, strerror(-ret
));
684 if (ret
& BDRV_BLOCK_ZERO
) {
688 ret
= bdrv_write_zeroes(bs
, sector_num
, n
, flags
);
690 error_report("error writing zeroes at sector %" PRId64
": %s",
691 sector_num
, strerror(-ret
));
698 int bdrv_pread(BlockDriverState
*bs
, int64_t offset
, void *buf
, int bytes
)
702 .iov_base
= (void *)buf
,
711 qemu_iovec_init_external(&qiov
, &iov
, 1);
712 ret
= bdrv_prwv_co(bs
, offset
, &qiov
, false, 0);
720 int bdrv_pwritev(BlockDriverState
*bs
, int64_t offset
, QEMUIOVector
*qiov
)
724 ret
= bdrv_prwv_co(bs
, offset
, qiov
, true, 0);
732 int bdrv_pwrite(BlockDriverState
*bs
, int64_t offset
,
733 const void *buf
, int bytes
)
737 .iov_base
= (void *) buf
,
745 qemu_iovec_init_external(&qiov
, &iov
, 1);
746 return bdrv_pwritev(bs
, offset
, &qiov
);
750 * Writes to the file and ensures that no writes are reordered across this
751 * request (acts as a barrier)
753 * Returns 0 on success, -errno in error cases.
755 int bdrv_pwrite_sync(BlockDriverState
*bs
, int64_t offset
,
756 const void *buf
, int count
)
760 ret
= bdrv_pwrite(bs
, offset
, buf
, count
);
765 /* No flush needed for cache modes that already do it */
766 if (bs
->enable_write_cache
) {
773 static int coroutine_fn
bdrv_co_do_copy_on_readv(BlockDriverState
*bs
,
774 int64_t sector_num
, int nb_sectors
, QEMUIOVector
*qiov
)
776 /* Perform I/O through a temporary buffer so that users who scribble over
777 * their read buffer while the operation is in progress do not end up
778 * modifying the image file. This is critical for zero-copy guest I/O
779 * where anything might happen inside guest memory.
783 BlockDriver
*drv
= bs
->drv
;
785 QEMUIOVector bounce_qiov
;
786 int64_t cluster_sector_num
;
787 int cluster_nb_sectors
;
791 /* Cover entire cluster so no additional backing file I/O is required when
792 * allocating cluster in the image file.
794 bdrv_round_to_clusters(bs
, sector_num
, nb_sectors
,
795 &cluster_sector_num
, &cluster_nb_sectors
);
797 trace_bdrv_co_do_copy_on_readv(bs
, sector_num
, nb_sectors
,
798 cluster_sector_num
, cluster_nb_sectors
);
800 iov
.iov_len
= cluster_nb_sectors
* BDRV_SECTOR_SIZE
;
801 iov
.iov_base
= bounce_buffer
= qemu_try_blockalign(bs
, iov
.iov_len
);
802 if (bounce_buffer
== NULL
) {
807 qemu_iovec_init_external(&bounce_qiov
, &iov
, 1);
809 ret
= drv
->bdrv_co_readv(bs
, cluster_sector_num
, cluster_nb_sectors
,
815 if (drv
->bdrv_co_write_zeroes
&&
816 buffer_is_zero(bounce_buffer
, iov
.iov_len
)) {
817 ret
= bdrv_co_do_write_zeroes(bs
, cluster_sector_num
,
818 cluster_nb_sectors
, 0);
820 /* This does not change the data on the disk, it is not necessary
821 * to flush even in cache=writethrough mode.
823 ret
= drv
->bdrv_co_writev(bs
, cluster_sector_num
, cluster_nb_sectors
,
828 /* It might be okay to ignore write errors for guest requests. If this
829 * is a deliberate copy-on-read then we don't want to ignore the error.
830 * Simply report it in all cases.
835 skip_bytes
= (sector_num
- cluster_sector_num
) * BDRV_SECTOR_SIZE
;
836 qemu_iovec_from_buf(qiov
, 0, bounce_buffer
+ skip_bytes
,
837 nb_sectors
* BDRV_SECTOR_SIZE
);
840 qemu_vfree(bounce_buffer
);
845 * Forwards an already correctly aligned request to the BlockDriver. This
846 * handles copy on read and zeroing after EOF; any other features must be
847 * implemented by the caller.
849 static int coroutine_fn
bdrv_aligned_preadv(BlockDriverState
*bs
,
850 BdrvTrackedRequest
*req
, int64_t offset
, unsigned int bytes
,
851 int64_t align
, QEMUIOVector
*qiov
, int flags
)
853 BlockDriver
*drv
= bs
->drv
;
856 int64_t sector_num
= offset
>> BDRV_SECTOR_BITS
;
857 unsigned int nb_sectors
= bytes
>> BDRV_SECTOR_BITS
;
859 assert((offset
& (BDRV_SECTOR_SIZE
- 1)) == 0);
860 assert((bytes
& (BDRV_SECTOR_SIZE
- 1)) == 0);
861 assert(!qiov
|| bytes
== qiov
->size
);
863 /* Handle Copy on Read and associated serialisation */
864 if (flags
& BDRV_REQ_COPY_ON_READ
) {
865 /* If we touch the same cluster it counts as an overlap. This
866 * guarantees that allocating writes will be serialized and not race
867 * with each other for the same cluster. For example, in copy-on-read
868 * it ensures that the CoR read and write operations are atomic and
869 * guest writes cannot interleave between them. */
870 mark_request_serialising(req
, bdrv_get_cluster_size(bs
));
873 if (!(flags
& BDRV_REQ_NO_SERIALISING
)) {
874 wait_serialising_requests(req
);
877 if (flags
& BDRV_REQ_COPY_ON_READ
) {
880 ret
= bdrv_is_allocated(bs
, sector_num
, nb_sectors
, &pnum
);
885 if (!ret
|| pnum
!= nb_sectors
) {
886 ret
= bdrv_co_do_copy_on_readv(bs
, sector_num
, nb_sectors
, qiov
);
891 /* Forward the request to the BlockDriver */
892 if (!bs
->zero_beyond_eof
) {
893 ret
= drv
->bdrv_co_readv(bs
, sector_num
, nb_sectors
, qiov
);
895 /* Read zeros after EOF */
896 int64_t total_sectors
, max_nb_sectors
;
898 total_sectors
= bdrv_nb_sectors(bs
);
899 if (total_sectors
< 0) {
904 max_nb_sectors
= ROUND_UP(MAX(0, total_sectors
- sector_num
),
905 align
>> BDRV_SECTOR_BITS
);
906 if (nb_sectors
< max_nb_sectors
) {
907 ret
= drv
->bdrv_co_readv(bs
, sector_num
, nb_sectors
, qiov
);
908 } else if (max_nb_sectors
> 0) {
909 QEMUIOVector local_qiov
;
911 qemu_iovec_init(&local_qiov
, qiov
->niov
);
912 qemu_iovec_concat(&local_qiov
, qiov
, 0,
913 max_nb_sectors
* BDRV_SECTOR_SIZE
);
915 ret
= drv
->bdrv_co_readv(bs
, sector_num
, max_nb_sectors
,
918 qemu_iovec_destroy(&local_qiov
);
923 /* Reading beyond end of file is supposed to produce zeroes */
924 if (ret
== 0 && total_sectors
< sector_num
+ nb_sectors
) {
925 uint64_t offset
= MAX(0, total_sectors
- sector_num
);
926 uint64_t bytes
= (sector_num
+ nb_sectors
- offset
) *
928 qemu_iovec_memset(qiov
, offset
* BDRV_SECTOR_SIZE
, 0, bytes
);
937 * Handle a read request in coroutine context
939 int coroutine_fn
bdrv_co_do_preadv(BlockDriverState
*bs
,
940 int64_t offset
, unsigned int bytes
, QEMUIOVector
*qiov
,
941 BdrvRequestFlags flags
)
943 BlockDriver
*drv
= bs
->drv
;
944 BdrvTrackedRequest req
;
946 /* TODO Lift BDRV_SECTOR_SIZE restriction in BlockDriver interface */
947 uint64_t align
= MAX(BDRV_SECTOR_SIZE
, bs
->request_alignment
);
948 uint8_t *head_buf
= NULL
;
949 uint8_t *tail_buf
= NULL
;
950 QEMUIOVector local_qiov
;
951 bool use_local_qiov
= false;
958 ret
= bdrv_check_byte_request(bs
, offset
, bytes
);
963 /* Don't do copy-on-read if we read data before write operation */
964 if (bs
->copy_on_read
&& !(flags
& BDRV_REQ_NO_SERIALISING
)) {
965 flags
|= BDRV_REQ_COPY_ON_READ
;
968 /* throttling disk I/O */
969 if (bs
->io_limits_enabled
) {
970 throttle_group_co_io_limits_intercept(bs
, bytes
, false);
973 /* Align read if necessary by padding qiov */
974 if (offset
& (align
- 1)) {
975 head_buf
= qemu_blockalign(bs
, align
);
976 qemu_iovec_init(&local_qiov
, qiov
->niov
+ 2);
977 qemu_iovec_add(&local_qiov
, head_buf
, offset
& (align
- 1));
978 qemu_iovec_concat(&local_qiov
, qiov
, 0, qiov
->size
);
979 use_local_qiov
= true;
981 bytes
+= offset
& (align
- 1);
982 offset
= offset
& ~(align
- 1);
985 if ((offset
+ bytes
) & (align
- 1)) {
986 if (!use_local_qiov
) {
987 qemu_iovec_init(&local_qiov
, qiov
->niov
+ 1);
988 qemu_iovec_concat(&local_qiov
, qiov
, 0, qiov
->size
);
989 use_local_qiov
= true;
991 tail_buf
= qemu_blockalign(bs
, align
);
992 qemu_iovec_add(&local_qiov
, tail_buf
,
993 align
- ((offset
+ bytes
) & (align
- 1)));
995 bytes
= ROUND_UP(bytes
, align
);
998 tracked_request_begin(&req
, bs
, offset
, bytes
, BDRV_TRACKED_READ
);
999 ret
= bdrv_aligned_preadv(bs
, &req
, offset
, bytes
, align
,
1000 use_local_qiov
? &local_qiov
: qiov
,
1002 tracked_request_end(&req
);
1004 if (use_local_qiov
) {
1005 qemu_iovec_destroy(&local_qiov
);
1006 qemu_vfree(head_buf
);
1007 qemu_vfree(tail_buf
);
1013 static int coroutine_fn
bdrv_co_do_readv(BlockDriverState
*bs
,
1014 int64_t sector_num
, int nb_sectors
, QEMUIOVector
*qiov
,
1015 BdrvRequestFlags flags
)
1017 if (nb_sectors
< 0 || nb_sectors
> BDRV_REQUEST_MAX_SECTORS
) {
1021 return bdrv_co_do_preadv(bs
, sector_num
<< BDRV_SECTOR_BITS
,
1022 nb_sectors
<< BDRV_SECTOR_BITS
, qiov
, flags
);
1025 int coroutine_fn
bdrv_co_readv(BlockDriverState
*bs
, int64_t sector_num
,
1026 int nb_sectors
, QEMUIOVector
*qiov
)
1028 trace_bdrv_co_readv(bs
, sector_num
, nb_sectors
);
1030 return bdrv_co_do_readv(bs
, sector_num
, nb_sectors
, qiov
, 0);
1033 int coroutine_fn
bdrv_co_readv_no_serialising(BlockDriverState
*bs
,
1034 int64_t sector_num
, int nb_sectors
, QEMUIOVector
*qiov
)
1036 trace_bdrv_co_readv_no_serialising(bs
, sector_num
, nb_sectors
);
1038 return bdrv_co_do_readv(bs
, sector_num
, nb_sectors
, qiov
,
1039 BDRV_REQ_NO_SERIALISING
);
1042 int coroutine_fn
bdrv_co_copy_on_readv(BlockDriverState
*bs
,
1043 int64_t sector_num
, int nb_sectors
, QEMUIOVector
*qiov
)
1045 trace_bdrv_co_copy_on_readv(bs
, sector_num
, nb_sectors
);
1047 return bdrv_co_do_readv(bs
, sector_num
, nb_sectors
, qiov
,
1048 BDRV_REQ_COPY_ON_READ
);
1051 #define MAX_WRITE_ZEROES_BOUNCE_BUFFER 32768
1053 static int coroutine_fn
bdrv_co_do_write_zeroes(BlockDriverState
*bs
,
1054 int64_t sector_num
, int nb_sectors
, BdrvRequestFlags flags
)
1056 BlockDriver
*drv
= bs
->drv
;
1058 struct iovec iov
= {0};
1061 int max_write_zeroes
= MIN_NON_ZERO(bs
->bl
.max_write_zeroes
,
1062 BDRV_REQUEST_MAX_SECTORS
);
1064 while (nb_sectors
> 0 && !ret
) {
1065 int num
= nb_sectors
;
1067 /* Align request. Block drivers can expect the "bulk" of the request
1070 if (bs
->bl
.write_zeroes_alignment
1071 && num
> bs
->bl
.write_zeroes_alignment
) {
1072 if (sector_num
% bs
->bl
.write_zeroes_alignment
!= 0) {
1073 /* Make a small request up to the first aligned sector. */
1074 num
= bs
->bl
.write_zeroes_alignment
;
1075 num
-= sector_num
% bs
->bl
.write_zeroes_alignment
;
1076 } else if ((sector_num
+ num
) % bs
->bl
.write_zeroes_alignment
!= 0) {
1077 /* Shorten the request to the last aligned sector. num cannot
1078 * underflow because num > bs->bl.write_zeroes_alignment.
1080 num
-= (sector_num
+ num
) % bs
->bl
.write_zeroes_alignment
;
1084 /* limit request size */
1085 if (num
> max_write_zeroes
) {
1086 num
= max_write_zeroes
;
1090 /* First try the efficient write zeroes operation */
1091 if (drv
->bdrv_co_write_zeroes
) {
1092 ret
= drv
->bdrv_co_write_zeroes(bs
, sector_num
, num
, flags
);
1095 if (ret
== -ENOTSUP
) {
1096 /* Fall back to bounce buffer if write zeroes is unsupported */
1097 int max_xfer_len
= MIN_NON_ZERO(bs
->bl
.max_transfer_length
,
1098 MAX_WRITE_ZEROES_BOUNCE_BUFFER
);
1099 num
= MIN(num
, max_xfer_len
);
1100 iov
.iov_len
= num
* BDRV_SECTOR_SIZE
;
1101 if (iov
.iov_base
== NULL
) {
1102 iov
.iov_base
= qemu_try_blockalign(bs
, num
* BDRV_SECTOR_SIZE
);
1103 if (iov
.iov_base
== NULL
) {
1107 memset(iov
.iov_base
, 0, num
* BDRV_SECTOR_SIZE
);
1109 qemu_iovec_init_external(&qiov
, &iov
, 1);
1111 ret
= drv
->bdrv_co_writev(bs
, sector_num
, num
, &qiov
);
1113 /* Keep bounce buffer around if it is big enough for all
1114 * all future requests.
1116 if (num
< max_xfer_len
) {
1117 qemu_vfree(iov
.iov_base
);
1118 iov
.iov_base
= NULL
;
1127 qemu_vfree(iov
.iov_base
);
1132 * Forwards an already correctly aligned write request to the BlockDriver.
1134 static int coroutine_fn
bdrv_aligned_pwritev(BlockDriverState
*bs
,
1135 BdrvTrackedRequest
*req
, int64_t offset
, unsigned int bytes
,
1136 QEMUIOVector
*qiov
, int flags
)
1138 BlockDriver
*drv
= bs
->drv
;
1142 int64_t sector_num
= offset
>> BDRV_SECTOR_BITS
;
1143 unsigned int nb_sectors
= bytes
>> BDRV_SECTOR_BITS
;
1145 assert((offset
& (BDRV_SECTOR_SIZE
- 1)) == 0);
1146 assert((bytes
& (BDRV_SECTOR_SIZE
- 1)) == 0);
1147 assert(!qiov
|| bytes
== qiov
->size
);
1149 waited
= wait_serialising_requests(req
);
1150 assert(!waited
|| !req
->serialising
);
1151 assert(req
->overlap_offset
<= offset
);
1152 assert(offset
+ bytes
<= req
->overlap_offset
+ req
->overlap_bytes
);
1154 ret
= notifier_with_return_list_notify(&bs
->before_write_notifiers
, req
);
1156 if (!ret
&& bs
->detect_zeroes
!= BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF
&&
1157 !(flags
& BDRV_REQ_ZERO_WRITE
) && drv
->bdrv_co_write_zeroes
&&
1158 qemu_iovec_is_zero(qiov
)) {
1159 flags
|= BDRV_REQ_ZERO_WRITE
;
1160 if (bs
->detect_zeroes
== BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP
) {
1161 flags
|= BDRV_REQ_MAY_UNMAP
;
1166 /* Do nothing, write notifier decided to fail this request */
1167 } else if (flags
& BDRV_REQ_ZERO_WRITE
) {
1168 bdrv_debug_event(bs
, BLKDBG_PWRITEV_ZERO
);
1169 ret
= bdrv_co_do_write_zeroes(bs
, sector_num
, nb_sectors
, flags
);
1171 bdrv_debug_event(bs
, BLKDBG_PWRITEV
);
1172 ret
= drv
->bdrv_co_writev(bs
, sector_num
, nb_sectors
, qiov
);
1174 bdrv_debug_event(bs
, BLKDBG_PWRITEV_DONE
);
1176 if (ret
== 0 && !bs
->enable_write_cache
) {
1177 ret
= bdrv_co_flush(bs
);
1180 bdrv_set_dirty(bs
, sector_num
, nb_sectors
);
1182 if (bs
->wr_highest_offset
< offset
+ bytes
) {
1183 bs
->wr_highest_offset
= offset
+ bytes
;
1187 bs
->total_sectors
= MAX(bs
->total_sectors
, sector_num
+ nb_sectors
);
1193 static int coroutine_fn
bdrv_co_do_zero_pwritev(BlockDriverState
*bs
,
1196 BdrvRequestFlags flags
,
1197 BdrvTrackedRequest
*req
)
1199 uint8_t *buf
= NULL
;
1200 QEMUIOVector local_qiov
;
1202 uint64_t align
= MAX(BDRV_SECTOR_SIZE
, bs
->request_alignment
);
1203 unsigned int head_padding_bytes
, tail_padding_bytes
;
1206 head_padding_bytes
= offset
& (align
- 1);
1207 tail_padding_bytes
= align
- ((offset
+ bytes
) & (align
- 1));
1210 assert(flags
& BDRV_REQ_ZERO_WRITE
);
1211 if (head_padding_bytes
|| tail_padding_bytes
) {
1212 buf
= qemu_blockalign(bs
, align
);
1213 iov
= (struct iovec
) {
1217 qemu_iovec_init_external(&local_qiov
, &iov
, 1);
1219 if (head_padding_bytes
) {
1220 uint64_t zero_bytes
= MIN(bytes
, align
- head_padding_bytes
);
1222 /* RMW the unaligned part before head. */
1223 mark_request_serialising(req
, align
);
1224 wait_serialising_requests(req
);
1225 bdrv_debug_event(bs
, BLKDBG_PWRITEV_RMW_HEAD
);
1226 ret
= bdrv_aligned_preadv(bs
, req
, offset
& ~(align
- 1), align
,
1227 align
, &local_qiov
, 0);
1231 bdrv_debug_event(bs
, BLKDBG_PWRITEV_RMW_AFTER_HEAD
);
1233 memset(buf
+ head_padding_bytes
, 0, zero_bytes
);
1234 ret
= bdrv_aligned_pwritev(bs
, req
, offset
& ~(align
- 1), align
,
1236 flags
& ~BDRV_REQ_ZERO_WRITE
);
1240 offset
+= zero_bytes
;
1241 bytes
-= zero_bytes
;
1244 assert(!bytes
|| (offset
& (align
- 1)) == 0);
1245 if (bytes
>= align
) {
1246 /* Write the aligned part in the middle. */
1247 uint64_t aligned_bytes
= bytes
& ~(align
- 1);
1248 ret
= bdrv_aligned_pwritev(bs
, req
, offset
, aligned_bytes
,
1253 bytes
-= aligned_bytes
;
1254 offset
+= aligned_bytes
;
1257 assert(!bytes
|| (offset
& (align
- 1)) == 0);
1259 assert(align
== tail_padding_bytes
+ bytes
);
1260 /* RMW the unaligned part after tail. */
1261 mark_request_serialising(req
, align
);
1262 wait_serialising_requests(req
);
1263 bdrv_debug_event(bs
, BLKDBG_PWRITEV_RMW_TAIL
);
1264 ret
= bdrv_aligned_preadv(bs
, req
, offset
, align
,
1265 align
, &local_qiov
, 0);
1269 bdrv_debug_event(bs
, BLKDBG_PWRITEV_RMW_AFTER_TAIL
);
1271 memset(buf
, 0, bytes
);
1272 ret
= bdrv_aligned_pwritev(bs
, req
, offset
, align
,
1273 &local_qiov
, flags
& ~BDRV_REQ_ZERO_WRITE
);
1282 * Handle a write request in coroutine context
1284 static int coroutine_fn
bdrv_co_do_pwritev(BlockDriverState
*bs
,
1285 int64_t offset
, unsigned int bytes
, QEMUIOVector
*qiov
,
1286 BdrvRequestFlags flags
)
1288 BdrvTrackedRequest req
;
1289 /* TODO Lift BDRV_SECTOR_SIZE restriction in BlockDriver interface */
1290 uint64_t align
= MAX(BDRV_SECTOR_SIZE
, bs
->request_alignment
);
1291 uint8_t *head_buf
= NULL
;
1292 uint8_t *tail_buf
= NULL
;
1293 QEMUIOVector local_qiov
;
1294 bool use_local_qiov
= false;
1300 if (bs
->read_only
) {
1303 assert(!(bs
->open_flags
& BDRV_O_INACTIVE
));
1305 ret
= bdrv_check_byte_request(bs
, offset
, bytes
);
1310 /* throttling disk I/O */
1311 if (bs
->io_limits_enabled
) {
1312 throttle_group_co_io_limits_intercept(bs
, bytes
, true);
1316 * Align write if necessary by performing a read-modify-write cycle.
1317 * Pad qiov with the read parts and be sure to have a tracked request not
1318 * only for bdrv_aligned_pwritev, but also for the reads of the RMW cycle.
1320 tracked_request_begin(&req
, bs
, offset
, bytes
, BDRV_TRACKED_WRITE
);
1323 ret
= bdrv_co_do_zero_pwritev(bs
, offset
, bytes
, flags
, &req
);
1327 if (offset
& (align
- 1)) {
1328 QEMUIOVector head_qiov
;
1329 struct iovec head_iov
;
1331 mark_request_serialising(&req
, align
);
1332 wait_serialising_requests(&req
);
1334 head_buf
= qemu_blockalign(bs
, align
);
1335 head_iov
= (struct iovec
) {
1336 .iov_base
= head_buf
,
1339 qemu_iovec_init_external(&head_qiov
, &head_iov
, 1);
1341 bdrv_debug_event(bs
, BLKDBG_PWRITEV_RMW_HEAD
);
1342 ret
= bdrv_aligned_preadv(bs
, &req
, offset
& ~(align
- 1), align
,
1343 align
, &head_qiov
, 0);
1347 bdrv_debug_event(bs
, BLKDBG_PWRITEV_RMW_AFTER_HEAD
);
1349 qemu_iovec_init(&local_qiov
, qiov
->niov
+ 2);
1350 qemu_iovec_add(&local_qiov
, head_buf
, offset
& (align
- 1));
1351 qemu_iovec_concat(&local_qiov
, qiov
, 0, qiov
->size
);
1352 use_local_qiov
= true;
1354 bytes
+= offset
& (align
- 1);
1355 offset
= offset
& ~(align
- 1);
1358 if ((offset
+ bytes
) & (align
- 1)) {
1359 QEMUIOVector tail_qiov
;
1360 struct iovec tail_iov
;
1364 mark_request_serialising(&req
, align
);
1365 waited
= wait_serialising_requests(&req
);
1366 assert(!waited
|| !use_local_qiov
);
1368 tail_buf
= qemu_blockalign(bs
, align
);
1369 tail_iov
= (struct iovec
) {
1370 .iov_base
= tail_buf
,
1373 qemu_iovec_init_external(&tail_qiov
, &tail_iov
, 1);
1375 bdrv_debug_event(bs
, BLKDBG_PWRITEV_RMW_TAIL
);
1376 ret
= bdrv_aligned_preadv(bs
, &req
, (offset
+ bytes
) & ~(align
- 1), align
,
1377 align
, &tail_qiov
, 0);
1381 bdrv_debug_event(bs
, BLKDBG_PWRITEV_RMW_AFTER_TAIL
);
1383 if (!use_local_qiov
) {
1384 qemu_iovec_init(&local_qiov
, qiov
->niov
+ 1);
1385 qemu_iovec_concat(&local_qiov
, qiov
, 0, qiov
->size
);
1386 use_local_qiov
= true;
1389 tail_bytes
= (offset
+ bytes
) & (align
- 1);
1390 qemu_iovec_add(&local_qiov
, tail_buf
+ tail_bytes
, align
- tail_bytes
);
1392 bytes
= ROUND_UP(bytes
, align
);
1395 ret
= bdrv_aligned_pwritev(bs
, &req
, offset
, bytes
,
1396 use_local_qiov
? &local_qiov
: qiov
,
1401 if (use_local_qiov
) {
1402 qemu_iovec_destroy(&local_qiov
);
1404 qemu_vfree(head_buf
);
1405 qemu_vfree(tail_buf
);
1407 tracked_request_end(&req
);
1411 static int coroutine_fn
bdrv_co_do_writev(BlockDriverState
*bs
,
1412 int64_t sector_num
, int nb_sectors
, QEMUIOVector
*qiov
,
1413 BdrvRequestFlags flags
)
1415 if (nb_sectors
< 0 || nb_sectors
> BDRV_REQUEST_MAX_SECTORS
) {
1419 return bdrv_co_do_pwritev(bs
, sector_num
<< BDRV_SECTOR_BITS
,
1420 nb_sectors
<< BDRV_SECTOR_BITS
, qiov
, flags
);
1423 int coroutine_fn
bdrv_co_writev(BlockDriverState
*bs
, int64_t sector_num
,
1424 int nb_sectors
, QEMUIOVector
*qiov
)
1426 trace_bdrv_co_writev(bs
, sector_num
, nb_sectors
);
1428 return bdrv_co_do_writev(bs
, sector_num
, nb_sectors
, qiov
, 0);
1431 int coroutine_fn
bdrv_co_write_zeroes(BlockDriverState
*bs
,
1432 int64_t sector_num
, int nb_sectors
,
1433 BdrvRequestFlags flags
)
1435 trace_bdrv_co_write_zeroes(bs
, sector_num
, nb_sectors
, flags
);
1437 if (!(bs
->open_flags
& BDRV_O_UNMAP
)) {
1438 flags
&= ~BDRV_REQ_MAY_UNMAP
;
1441 return bdrv_co_do_writev(bs
, sector_num
, nb_sectors
, NULL
,
1442 BDRV_REQ_ZERO_WRITE
| flags
);
1445 typedef struct BdrvCoGetBlockStatusData
{
1446 BlockDriverState
*bs
;
1447 BlockDriverState
*base
;
1448 BlockDriverState
**file
;
1454 } BdrvCoGetBlockStatusData
;
1457 * Returns the allocation status of the specified sectors.
1458 * Drivers not implementing the functionality are assumed to not support
1459 * backing files, hence all their sectors are reported as allocated.
1461 * If 'sector_num' is beyond the end of the disk image the return value is 0
1462 * and 'pnum' is set to 0.
1464 * 'pnum' is set to the number of sectors (including and immediately following
1465 * the specified sector) that are known to be in the same
1466 * allocated/unallocated state.
1468 * 'nb_sectors' is the max value 'pnum' should be set to. If nb_sectors goes
1469 * beyond the end of the disk image it will be clamped.
1471 * If returned value is positive and BDRV_BLOCK_OFFSET_VALID bit is set, 'file'
1472 * points to the BDS which the sector range is allocated in.
1474 static int64_t coroutine_fn
bdrv_co_get_block_status(BlockDriverState
*bs
,
1476 int nb_sectors
, int *pnum
,
1477 BlockDriverState
**file
)
1479 int64_t total_sectors
;
1483 total_sectors
= bdrv_nb_sectors(bs
);
1484 if (total_sectors
< 0) {
1485 return total_sectors
;
1488 if (sector_num
>= total_sectors
) {
1493 n
= total_sectors
- sector_num
;
1494 if (n
< nb_sectors
) {
1498 if (!bs
->drv
->bdrv_co_get_block_status
) {
1500 ret
= BDRV_BLOCK_DATA
| BDRV_BLOCK_ALLOCATED
;
1501 if (bs
->drv
->protocol_name
) {
1502 ret
|= BDRV_BLOCK_OFFSET_VALID
| (sector_num
* BDRV_SECTOR_SIZE
);
1508 ret
= bs
->drv
->bdrv_co_get_block_status(bs
, sector_num
, nb_sectors
, pnum
,
1515 if (ret
& BDRV_BLOCK_RAW
) {
1516 assert(ret
& BDRV_BLOCK_OFFSET_VALID
);
1517 return bdrv_get_block_status(bs
->file
->bs
, ret
>> BDRV_SECTOR_BITS
,
1521 if (ret
& (BDRV_BLOCK_DATA
| BDRV_BLOCK_ZERO
)) {
1522 ret
|= BDRV_BLOCK_ALLOCATED
;
1524 if (bdrv_unallocated_blocks_are_zero(bs
)) {
1525 ret
|= BDRV_BLOCK_ZERO
;
1526 } else if (bs
->backing
) {
1527 BlockDriverState
*bs2
= bs
->backing
->bs
;
1528 int64_t nb_sectors2
= bdrv_nb_sectors(bs2
);
1529 if (nb_sectors2
>= 0 && sector_num
>= nb_sectors2
) {
1530 ret
|= BDRV_BLOCK_ZERO
;
1535 if (*file
&& *file
!= bs
&&
1536 (ret
& BDRV_BLOCK_DATA
) && !(ret
& BDRV_BLOCK_ZERO
) &&
1537 (ret
& BDRV_BLOCK_OFFSET_VALID
)) {
1538 BlockDriverState
*file2
;
1541 ret2
= bdrv_co_get_block_status(*file
, ret
>> BDRV_SECTOR_BITS
,
1542 *pnum
, &file_pnum
, &file2
);
1544 /* Ignore errors. This is just providing extra information, it
1545 * is useful but not necessary.
1548 /* !file_pnum indicates an offset at or beyond the EOF; it is
1549 * perfectly valid for the format block driver to point to such
1550 * offsets, so catch it and mark everything as zero */
1551 ret
|= BDRV_BLOCK_ZERO
;
1553 /* Limit request to the range reported by the protocol driver */
1555 ret
|= (ret2
& BDRV_BLOCK_ZERO
);
1563 static int64_t coroutine_fn
bdrv_co_get_block_status_above(BlockDriverState
*bs
,
1564 BlockDriverState
*base
,
1568 BlockDriverState
**file
)
1570 BlockDriverState
*p
;
1574 for (p
= bs
; p
!= base
; p
= backing_bs(p
)) {
1575 ret
= bdrv_co_get_block_status(p
, sector_num
, nb_sectors
, pnum
, file
);
1576 if (ret
< 0 || ret
& BDRV_BLOCK_ALLOCATED
) {
1579 /* [sector_num, pnum] unallocated on this layer, which could be only
1580 * the first part of [sector_num, nb_sectors]. */
1581 nb_sectors
= MIN(nb_sectors
, *pnum
);
1586 /* Coroutine wrapper for bdrv_get_block_status_above() */
1587 static void coroutine_fn
bdrv_get_block_status_above_co_entry(void *opaque
)
1589 BdrvCoGetBlockStatusData
*data
= opaque
;
1591 data
->ret
= bdrv_co_get_block_status_above(data
->bs
, data
->base
,
1600 * Synchronous wrapper around bdrv_co_get_block_status_above().
1602 * See bdrv_co_get_block_status_above() for details.
1604 int64_t bdrv_get_block_status_above(BlockDriverState
*bs
,
1605 BlockDriverState
*base
,
1607 int nb_sectors
, int *pnum
,
1608 BlockDriverState
**file
)
1611 BdrvCoGetBlockStatusData data
= {
1615 .sector_num
= sector_num
,
1616 .nb_sectors
= nb_sectors
,
1621 if (qemu_in_coroutine()) {
1622 /* Fast-path if already in coroutine context */
1623 bdrv_get_block_status_above_co_entry(&data
);
1625 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
1627 co
= qemu_coroutine_create(bdrv_get_block_status_above_co_entry
);
1628 qemu_coroutine_enter(co
, &data
);
1629 while (!data
.done
) {
1630 aio_poll(aio_context
, true);
1636 int64_t bdrv_get_block_status(BlockDriverState
*bs
,
1638 int nb_sectors
, int *pnum
,
1639 BlockDriverState
**file
)
1641 return bdrv_get_block_status_above(bs
, backing_bs(bs
),
1642 sector_num
, nb_sectors
, pnum
, file
);
1645 int coroutine_fn
bdrv_is_allocated(BlockDriverState
*bs
, int64_t sector_num
,
1646 int nb_sectors
, int *pnum
)
1648 BlockDriverState
*file
;
1649 int64_t ret
= bdrv_get_block_status(bs
, sector_num
, nb_sectors
, pnum
,
1654 return !!(ret
& BDRV_BLOCK_ALLOCATED
);
1658 * Given an image chain: ... -> [BASE] -> [INTER1] -> [INTER2] -> [TOP]
1660 * Return true if the given sector is allocated in any image between
1661 * BASE and TOP (inclusive). BASE can be NULL to check if the given
1662 * sector is allocated in any image of the chain. Return false otherwise.
1664 * 'pnum' is set to the number of sectors (including and immediately following
1665 * the specified sector) that are known to be in the same
1666 * allocated/unallocated state.
1669 int bdrv_is_allocated_above(BlockDriverState
*top
,
1670 BlockDriverState
*base
,
1672 int nb_sectors
, int *pnum
)
1674 BlockDriverState
*intermediate
;
1675 int ret
, n
= nb_sectors
;
1678 while (intermediate
&& intermediate
!= base
) {
1680 ret
= bdrv_is_allocated(intermediate
, sector_num
, nb_sectors
,
1690 * [sector_num, nb_sectors] is unallocated on top but intermediate
1693 * [sector_num+x, nr_sectors] allocated.
1695 if (n
> pnum_inter
&&
1696 (intermediate
== top
||
1697 sector_num
+ pnum_inter
< intermediate
->total_sectors
)) {
1701 intermediate
= backing_bs(intermediate
);
1708 int bdrv_write_compressed(BlockDriverState
*bs
, int64_t sector_num
,
1709 const uint8_t *buf
, int nb_sectors
)
1711 BlockDriver
*drv
= bs
->drv
;
1717 if (!drv
->bdrv_write_compressed
) {
1720 ret
= bdrv_check_request(bs
, sector_num
, nb_sectors
);
1725 assert(QLIST_EMPTY(&bs
->dirty_bitmaps
));
1727 return drv
->bdrv_write_compressed(bs
, sector_num
, buf
, nb_sectors
);
1730 int bdrv_save_vmstate(BlockDriverState
*bs
, const uint8_t *buf
,
1731 int64_t pos
, int size
)
1734 struct iovec iov
= {
1735 .iov_base
= (void *) buf
,
1739 qemu_iovec_init_external(&qiov
, &iov
, 1);
1740 return bdrv_writev_vmstate(bs
, &qiov
, pos
);
1743 int bdrv_writev_vmstate(BlockDriverState
*bs
, QEMUIOVector
*qiov
, int64_t pos
)
1745 BlockDriver
*drv
= bs
->drv
;
1749 } else if (drv
->bdrv_save_vmstate
) {
1750 return drv
->bdrv_save_vmstate(bs
, qiov
, pos
);
1751 } else if (bs
->file
) {
1752 return bdrv_writev_vmstate(bs
->file
->bs
, qiov
, pos
);
1758 int bdrv_load_vmstate(BlockDriverState
*bs
, uint8_t *buf
,
1759 int64_t pos
, int size
)
1761 BlockDriver
*drv
= bs
->drv
;
1764 if (drv
->bdrv_load_vmstate
)
1765 return drv
->bdrv_load_vmstate(bs
, buf
, pos
, size
);
1767 return bdrv_load_vmstate(bs
->file
->bs
, buf
, pos
, size
);
1771 /**************************************************************/
1774 BlockAIOCB
*bdrv_aio_readv(BlockDriverState
*bs
, int64_t sector_num
,
1775 QEMUIOVector
*qiov
, int nb_sectors
,
1776 BlockCompletionFunc
*cb
, void *opaque
)
1778 trace_bdrv_aio_readv(bs
, sector_num
, nb_sectors
, opaque
);
1780 return bdrv_co_aio_rw_vector(bs
, sector_num
, qiov
, nb_sectors
, 0,
1784 BlockAIOCB
*bdrv_aio_writev(BlockDriverState
*bs
, int64_t sector_num
,
1785 QEMUIOVector
*qiov
, int nb_sectors
,
1786 BlockCompletionFunc
*cb
, void *opaque
)
1788 trace_bdrv_aio_writev(bs
, sector_num
, nb_sectors
, opaque
);
1790 return bdrv_co_aio_rw_vector(bs
, sector_num
, qiov
, nb_sectors
, 0,
1794 BlockAIOCB
*bdrv_aio_write_zeroes(BlockDriverState
*bs
,
1795 int64_t sector_num
, int nb_sectors
, BdrvRequestFlags flags
,
1796 BlockCompletionFunc
*cb
, void *opaque
)
1798 trace_bdrv_aio_write_zeroes(bs
, sector_num
, nb_sectors
, flags
, opaque
);
1800 return bdrv_co_aio_rw_vector(bs
, sector_num
, NULL
, nb_sectors
,
1801 BDRV_REQ_ZERO_WRITE
| flags
,
1806 typedef struct MultiwriteCB
{
1811 BlockCompletionFunc
*cb
;
1813 QEMUIOVector
*free_qiov
;
1817 static void multiwrite_user_cb(MultiwriteCB
*mcb
)
1821 for (i
= 0; i
< mcb
->num_callbacks
; i
++) {
1822 mcb
->callbacks
[i
].cb(mcb
->callbacks
[i
].opaque
, mcb
->error
);
1823 if (mcb
->callbacks
[i
].free_qiov
) {
1824 qemu_iovec_destroy(mcb
->callbacks
[i
].free_qiov
);
1826 g_free(mcb
->callbacks
[i
].free_qiov
);
1830 static void multiwrite_cb(void *opaque
, int ret
)
1832 MultiwriteCB
*mcb
= opaque
;
1834 trace_multiwrite_cb(mcb
, ret
);
1836 if (ret
< 0 && !mcb
->error
) {
1840 mcb
->num_requests
--;
1841 if (mcb
->num_requests
== 0) {
1842 multiwrite_user_cb(mcb
);
1847 static int multiwrite_req_compare(const void *a
, const void *b
)
1849 const BlockRequest
*req1
= a
, *req2
= b
;
1852 * Note that we can't simply subtract req2->sector from req1->sector
1853 * here as that could overflow the return value.
1855 if (req1
->sector
> req2
->sector
) {
1857 } else if (req1
->sector
< req2
->sector
) {
1865 * Takes a bunch of requests and tries to merge them. Returns the number of
1866 * requests that remain after merging.
1868 static int multiwrite_merge(BlockDriverState
*bs
, BlockRequest
*reqs
,
1869 int num_reqs
, MultiwriteCB
*mcb
)
1873 // Sort requests by start sector
1874 qsort(reqs
, num_reqs
, sizeof(*reqs
), &multiwrite_req_compare
);
1876 // Check if adjacent requests touch the same clusters. If so, combine them,
1877 // filling up gaps with zero sectors.
1879 for (i
= 1; i
< num_reqs
; i
++) {
1881 int64_t oldreq_last
= reqs
[outidx
].sector
+ reqs
[outidx
].nb_sectors
;
1883 // Handle exactly sequential writes and overlapping writes.
1884 if (reqs
[i
].sector
<= oldreq_last
) {
1888 if (reqs
[outidx
].qiov
->niov
+ reqs
[i
].qiov
->niov
+ 1 >
1893 if (bs
->bl
.max_transfer_length
&& reqs
[outidx
].nb_sectors
+
1894 reqs
[i
].nb_sectors
> bs
->bl
.max_transfer_length
) {
1900 QEMUIOVector
*qiov
= g_malloc0(sizeof(*qiov
));
1901 qemu_iovec_init(qiov
,
1902 reqs
[outidx
].qiov
->niov
+ reqs
[i
].qiov
->niov
+ 1);
1904 // Add the first request to the merged one. If the requests are
1905 // overlapping, drop the last sectors of the first request.
1906 size
= (reqs
[i
].sector
- reqs
[outidx
].sector
) << 9;
1907 qemu_iovec_concat(qiov
, reqs
[outidx
].qiov
, 0, size
);
1909 // We should need to add any zeros between the two requests
1910 assert (reqs
[i
].sector
<= oldreq_last
);
1912 // Add the second request
1913 qemu_iovec_concat(qiov
, reqs
[i
].qiov
, 0, reqs
[i
].qiov
->size
);
1915 // Add tail of first request, if necessary
1916 if (qiov
->size
< reqs
[outidx
].qiov
->size
) {
1917 qemu_iovec_concat(qiov
, reqs
[outidx
].qiov
, qiov
->size
,
1918 reqs
[outidx
].qiov
->size
- qiov
->size
);
1921 reqs
[outidx
].nb_sectors
= qiov
->size
>> 9;
1922 reqs
[outidx
].qiov
= qiov
;
1924 mcb
->callbacks
[i
].free_qiov
= reqs
[outidx
].qiov
;
1927 reqs
[outidx
].sector
= reqs
[i
].sector
;
1928 reqs
[outidx
].nb_sectors
= reqs
[i
].nb_sectors
;
1929 reqs
[outidx
].qiov
= reqs
[i
].qiov
;
1934 block_acct_merge_done(blk_get_stats(bs
->blk
), BLOCK_ACCT_WRITE
,
1935 num_reqs
- outidx
- 1);
1942 * Submit multiple AIO write requests at once.
1944 * On success, the function returns 0 and all requests in the reqs array have
1945 * been submitted. In error case this function returns -1, and any of the
1946 * requests may or may not be submitted yet. In particular, this means that the
1947 * callback will be called for some of the requests, for others it won't. The
1948 * caller must check the error field of the BlockRequest to wait for the right
1949 * callbacks (if error != 0, no callback will be called).
1951 * The implementation may modify the contents of the reqs array, e.g. to merge
1952 * requests. However, the fields opaque and error are left unmodified as they
1953 * are used to signal failure for a single request to the caller.
1955 int bdrv_aio_multiwrite(BlockDriverState
*bs
, BlockRequest
*reqs
, int num_reqs
)
1960 /* don't submit writes if we don't have a medium */
1961 if (bs
->drv
== NULL
) {
1962 for (i
= 0; i
< num_reqs
; i
++) {
1963 reqs
[i
].error
= -ENOMEDIUM
;
1968 if (num_reqs
== 0) {
1972 // Create MultiwriteCB structure
1973 mcb
= g_malloc0(sizeof(*mcb
) + num_reqs
* sizeof(*mcb
->callbacks
));
1974 mcb
->num_requests
= 0;
1975 mcb
->num_callbacks
= num_reqs
;
1977 for (i
= 0; i
< num_reqs
; i
++) {
1978 mcb
->callbacks
[i
].cb
= reqs
[i
].cb
;
1979 mcb
->callbacks
[i
].opaque
= reqs
[i
].opaque
;
1982 // Check for mergable requests
1983 num_reqs
= multiwrite_merge(bs
, reqs
, num_reqs
, mcb
);
1985 trace_bdrv_aio_multiwrite(mcb
, mcb
->num_callbacks
, num_reqs
);
1987 /* Run the aio requests. */
1988 mcb
->num_requests
= num_reqs
;
1989 for (i
= 0; i
< num_reqs
; i
++) {
1990 bdrv_co_aio_rw_vector(bs
, reqs
[i
].sector
, reqs
[i
].qiov
,
1991 reqs
[i
].nb_sectors
, reqs
[i
].flags
,
1999 void bdrv_aio_cancel(BlockAIOCB
*acb
)
2002 bdrv_aio_cancel_async(acb
);
2003 while (acb
->refcnt
> 1) {
2004 if (acb
->aiocb_info
->get_aio_context
) {
2005 aio_poll(acb
->aiocb_info
->get_aio_context(acb
), true);
2006 } else if (acb
->bs
) {
2007 aio_poll(bdrv_get_aio_context(acb
->bs
), true);
2012 qemu_aio_unref(acb
);
2015 /* Async version of aio cancel. The caller is not blocked if the acb implements
2016 * cancel_async, otherwise we do nothing and let the request normally complete.
2017 * In either case the completion callback must be called. */
2018 void bdrv_aio_cancel_async(BlockAIOCB
*acb
)
2020 if (acb
->aiocb_info
->cancel_async
) {
2021 acb
->aiocb_info
->cancel_async(acb
);
2025 /**************************************************************/
2026 /* async block device emulation */
2028 typedef struct BlockAIOCBSync
{
2032 /* vector translation state */
2038 static const AIOCBInfo bdrv_em_aiocb_info
= {
2039 .aiocb_size
= sizeof(BlockAIOCBSync
),
2042 static void bdrv_aio_bh_cb(void *opaque
)
2044 BlockAIOCBSync
*acb
= opaque
;
2046 if (!acb
->is_write
&& acb
->ret
>= 0) {
2047 qemu_iovec_from_buf(acb
->qiov
, 0, acb
->bounce
, acb
->qiov
->size
);
2049 qemu_vfree(acb
->bounce
);
2050 acb
->common
.cb(acb
->common
.opaque
, acb
->ret
);
2051 qemu_bh_delete(acb
->bh
);
2053 qemu_aio_unref(acb
);
2056 static BlockAIOCB
*bdrv_aio_rw_vector(BlockDriverState
*bs
,
2060 BlockCompletionFunc
*cb
,
2065 BlockAIOCBSync
*acb
;
2067 acb
= qemu_aio_get(&bdrv_em_aiocb_info
, bs
, cb
, opaque
);
2068 acb
->is_write
= is_write
;
2070 acb
->bounce
= qemu_try_blockalign(bs
, qiov
->size
);
2071 acb
->bh
= aio_bh_new(bdrv_get_aio_context(bs
), bdrv_aio_bh_cb
, acb
);
2073 if (acb
->bounce
== NULL
) {
2075 } else if (is_write
) {
2076 qemu_iovec_to_buf(acb
->qiov
, 0, acb
->bounce
, qiov
->size
);
2077 acb
->ret
= bs
->drv
->bdrv_write(bs
, sector_num
, acb
->bounce
, nb_sectors
);
2079 acb
->ret
= bs
->drv
->bdrv_read(bs
, sector_num
, acb
->bounce
, nb_sectors
);
2082 qemu_bh_schedule(acb
->bh
);
2084 return &acb
->common
;
2087 static BlockAIOCB
*bdrv_aio_readv_em(BlockDriverState
*bs
,
2088 int64_t sector_num
, QEMUIOVector
*qiov
, int nb_sectors
,
2089 BlockCompletionFunc
*cb
, void *opaque
)
2091 return bdrv_aio_rw_vector(bs
, sector_num
, qiov
, nb_sectors
, cb
, opaque
, 0);
2094 static BlockAIOCB
*bdrv_aio_writev_em(BlockDriverState
*bs
,
2095 int64_t sector_num
, QEMUIOVector
*qiov
, int nb_sectors
,
2096 BlockCompletionFunc
*cb
, void *opaque
)
2098 return bdrv_aio_rw_vector(bs
, sector_num
, qiov
, nb_sectors
, cb
, opaque
, 1);
2102 typedef struct BlockAIOCBCoroutine
{
2109 } BlockAIOCBCoroutine
;
2111 static const AIOCBInfo bdrv_em_co_aiocb_info
= {
2112 .aiocb_size
= sizeof(BlockAIOCBCoroutine
),
2115 static void bdrv_co_complete(BlockAIOCBCoroutine
*acb
)
2117 if (!acb
->need_bh
) {
2118 acb
->common
.cb(acb
->common
.opaque
, acb
->req
.error
);
2119 qemu_aio_unref(acb
);
2123 static void bdrv_co_em_bh(void *opaque
)
2125 BlockAIOCBCoroutine
*acb
= opaque
;
2127 assert(!acb
->need_bh
);
2128 qemu_bh_delete(acb
->bh
);
2129 bdrv_co_complete(acb
);
2132 static void bdrv_co_maybe_schedule_bh(BlockAIOCBCoroutine
*acb
)
2134 acb
->need_bh
= false;
2135 if (acb
->req
.error
!= -EINPROGRESS
) {
2136 BlockDriverState
*bs
= acb
->common
.bs
;
2138 acb
->bh
= aio_bh_new(bdrv_get_aio_context(bs
), bdrv_co_em_bh
, acb
);
2139 qemu_bh_schedule(acb
->bh
);
2143 /* Invoke bdrv_co_do_readv/bdrv_co_do_writev */
2144 static void coroutine_fn
bdrv_co_do_rw(void *opaque
)
2146 BlockAIOCBCoroutine
*acb
= opaque
;
2147 BlockDriverState
*bs
= acb
->common
.bs
;
2149 if (!acb
->is_write
) {
2150 acb
->req
.error
= bdrv_co_do_readv(bs
, acb
->req
.sector
,
2151 acb
->req
.nb_sectors
, acb
->req
.qiov
, acb
->req
.flags
);
2153 acb
->req
.error
= bdrv_co_do_writev(bs
, acb
->req
.sector
,
2154 acb
->req
.nb_sectors
, acb
->req
.qiov
, acb
->req
.flags
);
2157 bdrv_co_complete(acb
);
2160 static BlockAIOCB
*bdrv_co_aio_rw_vector(BlockDriverState
*bs
,
2164 BdrvRequestFlags flags
,
2165 BlockCompletionFunc
*cb
,
2170 BlockAIOCBCoroutine
*acb
;
2172 acb
= qemu_aio_get(&bdrv_em_co_aiocb_info
, bs
, cb
, opaque
);
2173 acb
->need_bh
= true;
2174 acb
->req
.error
= -EINPROGRESS
;
2175 acb
->req
.sector
= sector_num
;
2176 acb
->req
.nb_sectors
= nb_sectors
;
2177 acb
->req
.qiov
= qiov
;
2178 acb
->req
.flags
= flags
;
2179 acb
->is_write
= is_write
;
2181 co
= qemu_coroutine_create(bdrv_co_do_rw
);
2182 qemu_coroutine_enter(co
, acb
);
2184 bdrv_co_maybe_schedule_bh(acb
);
2185 return &acb
->common
;
2188 static void coroutine_fn
bdrv_aio_flush_co_entry(void *opaque
)
2190 BlockAIOCBCoroutine
*acb
= opaque
;
2191 BlockDriverState
*bs
= acb
->common
.bs
;
2193 acb
->req
.error
= bdrv_co_flush(bs
);
2194 bdrv_co_complete(acb
);
2197 BlockAIOCB
*bdrv_aio_flush(BlockDriverState
*bs
,
2198 BlockCompletionFunc
*cb
, void *opaque
)
2200 trace_bdrv_aio_flush(bs
, opaque
);
2203 BlockAIOCBCoroutine
*acb
;
2205 acb
= qemu_aio_get(&bdrv_em_co_aiocb_info
, bs
, cb
, opaque
);
2206 acb
->need_bh
= true;
2207 acb
->req
.error
= -EINPROGRESS
;
2209 co
= qemu_coroutine_create(bdrv_aio_flush_co_entry
);
2210 qemu_coroutine_enter(co
, acb
);
2212 bdrv_co_maybe_schedule_bh(acb
);
2213 return &acb
->common
;
2216 static void coroutine_fn
bdrv_aio_discard_co_entry(void *opaque
)
2218 BlockAIOCBCoroutine
*acb
= opaque
;
2219 BlockDriverState
*bs
= acb
->common
.bs
;
2221 acb
->req
.error
= bdrv_co_discard(bs
, acb
->req
.sector
, acb
->req
.nb_sectors
);
2222 bdrv_co_complete(acb
);
2225 BlockAIOCB
*bdrv_aio_discard(BlockDriverState
*bs
,
2226 int64_t sector_num
, int nb_sectors
,
2227 BlockCompletionFunc
*cb
, void *opaque
)
2230 BlockAIOCBCoroutine
*acb
;
2232 trace_bdrv_aio_discard(bs
, sector_num
, nb_sectors
, opaque
);
2234 acb
= qemu_aio_get(&bdrv_em_co_aiocb_info
, bs
, cb
, opaque
);
2235 acb
->need_bh
= true;
2236 acb
->req
.error
= -EINPROGRESS
;
2237 acb
->req
.sector
= sector_num
;
2238 acb
->req
.nb_sectors
= nb_sectors
;
2239 co
= qemu_coroutine_create(bdrv_aio_discard_co_entry
);
2240 qemu_coroutine_enter(co
, acb
);
2242 bdrv_co_maybe_schedule_bh(acb
);
2243 return &acb
->common
;
2246 void *qemu_aio_get(const AIOCBInfo
*aiocb_info
, BlockDriverState
*bs
,
2247 BlockCompletionFunc
*cb
, void *opaque
)
2251 acb
= g_malloc(aiocb_info
->aiocb_size
);
2252 acb
->aiocb_info
= aiocb_info
;
2255 acb
->opaque
= opaque
;
2260 void qemu_aio_ref(void *p
)
2262 BlockAIOCB
*acb
= p
;
2266 void qemu_aio_unref(void *p
)
2268 BlockAIOCB
*acb
= p
;
2269 assert(acb
->refcnt
> 0);
2270 if (--acb
->refcnt
== 0) {
2275 /**************************************************************/
2276 /* Coroutine block device emulation */
2278 typedef struct CoroutineIOCompletion
{
2279 Coroutine
*coroutine
;
2281 } CoroutineIOCompletion
;
2283 static void bdrv_co_io_em_complete(void *opaque
, int ret
)
2285 CoroutineIOCompletion
*co
= opaque
;
2288 qemu_coroutine_enter(co
->coroutine
, NULL
);
2291 static int coroutine_fn
bdrv_co_io_em(BlockDriverState
*bs
, int64_t sector_num
,
2292 int nb_sectors
, QEMUIOVector
*iov
,
2295 CoroutineIOCompletion co
= {
2296 .coroutine
= qemu_coroutine_self(),
2301 acb
= bs
->drv
->bdrv_aio_writev(bs
, sector_num
, iov
, nb_sectors
,
2302 bdrv_co_io_em_complete
, &co
);
2304 acb
= bs
->drv
->bdrv_aio_readv(bs
, sector_num
, iov
, nb_sectors
,
2305 bdrv_co_io_em_complete
, &co
);
2308 trace_bdrv_co_io_em(bs
, sector_num
, nb_sectors
, is_write
, acb
);
2312 qemu_coroutine_yield();
2317 static int coroutine_fn
bdrv_co_readv_em(BlockDriverState
*bs
,
2318 int64_t sector_num
, int nb_sectors
,
2321 return bdrv_co_io_em(bs
, sector_num
, nb_sectors
, iov
, false);
2324 static int coroutine_fn
bdrv_co_writev_em(BlockDriverState
*bs
,
2325 int64_t sector_num
, int nb_sectors
,
2328 return bdrv_co_io_em(bs
, sector_num
, nb_sectors
, iov
, true);
2331 static void coroutine_fn
bdrv_flush_co_entry(void *opaque
)
2333 RwCo
*rwco
= opaque
;
2335 rwco
->ret
= bdrv_co_flush(rwco
->bs
);
2338 int coroutine_fn
bdrv_co_flush(BlockDriverState
*bs
)
2341 BdrvTrackedRequest req
;
2343 if (!bs
|| !bdrv_is_inserted(bs
) || bdrv_is_read_only(bs
) ||
2348 tracked_request_begin(&req
, bs
, 0, 0, BDRV_TRACKED_FLUSH
);
2349 /* Write back cached data to the OS even with cache=unsafe */
2350 BLKDBG_EVENT(bs
->file
, BLKDBG_FLUSH_TO_OS
);
2351 if (bs
->drv
->bdrv_co_flush_to_os
) {
2352 ret
= bs
->drv
->bdrv_co_flush_to_os(bs
);
2358 /* But don't actually force it to the disk with cache=unsafe */
2359 if (bs
->open_flags
& BDRV_O_NO_FLUSH
) {
2363 BLKDBG_EVENT(bs
->file
, BLKDBG_FLUSH_TO_DISK
);
2364 if (bs
->drv
->bdrv_co_flush_to_disk
) {
2365 ret
= bs
->drv
->bdrv_co_flush_to_disk(bs
);
2366 } else if (bs
->drv
->bdrv_aio_flush
) {
2368 CoroutineIOCompletion co
= {
2369 .coroutine
= qemu_coroutine_self(),
2372 acb
= bs
->drv
->bdrv_aio_flush(bs
, bdrv_co_io_em_complete
, &co
);
2376 qemu_coroutine_yield();
2381 * Some block drivers always operate in either writethrough or unsafe
2382 * mode and don't support bdrv_flush therefore. Usually qemu doesn't
2383 * know how the server works (because the behaviour is hardcoded or
2384 * depends on server-side configuration), so we can't ensure that
2385 * everything is safe on disk. Returning an error doesn't work because
2386 * that would break guests even if the server operates in writethrough
2389 * Let's hope the user knows what he's doing.
2397 /* Now flush the underlying protocol. It will also have BDRV_O_NO_FLUSH
2398 * in the case of cache=unsafe, so there are no useless flushes.
2401 ret
= bs
->file
? bdrv_co_flush(bs
->file
->bs
) : 0;
2403 tracked_request_end(&req
);
2407 int bdrv_flush(BlockDriverState
*bs
)
2415 if (qemu_in_coroutine()) {
2416 /* Fast-path if already in coroutine context */
2417 bdrv_flush_co_entry(&rwco
);
2419 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
2421 co
= qemu_coroutine_create(bdrv_flush_co_entry
);
2422 qemu_coroutine_enter(co
, &rwco
);
2423 while (rwco
.ret
== NOT_DONE
) {
2424 aio_poll(aio_context
, true);
2431 typedef struct DiscardCo
{
2432 BlockDriverState
*bs
;
2437 static void coroutine_fn
bdrv_discard_co_entry(void *opaque
)
2439 DiscardCo
*rwco
= opaque
;
2441 rwco
->ret
= bdrv_co_discard(rwco
->bs
, rwco
->sector_num
, rwco
->nb_sectors
);
2444 int coroutine_fn
bdrv_co_discard(BlockDriverState
*bs
, int64_t sector_num
,
2447 BdrvTrackedRequest req
;
2448 int max_discard
, ret
;
2454 ret
= bdrv_check_request(bs
, sector_num
, nb_sectors
);
2457 } else if (bs
->read_only
) {
2460 assert(!(bs
->open_flags
& BDRV_O_INACTIVE
));
2462 /* Do nothing if disabled. */
2463 if (!(bs
->open_flags
& BDRV_O_UNMAP
)) {
2467 if (!bs
->drv
->bdrv_co_discard
&& !bs
->drv
->bdrv_aio_discard
) {
2471 tracked_request_begin(&req
, bs
, sector_num
, nb_sectors
,
2472 BDRV_TRACKED_DISCARD
);
2473 bdrv_set_dirty(bs
, sector_num
, nb_sectors
);
2475 max_discard
= MIN_NON_ZERO(bs
->bl
.max_discard
, BDRV_REQUEST_MAX_SECTORS
);
2476 while (nb_sectors
> 0) {
2478 int num
= nb_sectors
;
2481 if (bs
->bl
.discard_alignment
&&
2482 num
>= bs
->bl
.discard_alignment
&&
2483 sector_num
% bs
->bl
.discard_alignment
) {
2484 if (num
> bs
->bl
.discard_alignment
) {
2485 num
= bs
->bl
.discard_alignment
;
2487 num
-= sector_num
% bs
->bl
.discard_alignment
;
2490 /* limit request size */
2491 if (num
> max_discard
) {
2495 if (bs
->drv
->bdrv_co_discard
) {
2496 ret
= bs
->drv
->bdrv_co_discard(bs
, sector_num
, num
);
2499 CoroutineIOCompletion co
= {
2500 .coroutine
= qemu_coroutine_self(),
2503 acb
= bs
->drv
->bdrv_aio_discard(bs
, sector_num
, nb_sectors
,
2504 bdrv_co_io_em_complete
, &co
);
2509 qemu_coroutine_yield();
2513 if (ret
&& ret
!= -ENOTSUP
) {
2522 tracked_request_end(&req
);
2526 int bdrv_discard(BlockDriverState
*bs
, int64_t sector_num
, int nb_sectors
)
2531 .sector_num
= sector_num
,
2532 .nb_sectors
= nb_sectors
,
2536 if (qemu_in_coroutine()) {
2537 /* Fast-path if already in coroutine context */
2538 bdrv_discard_co_entry(&rwco
);
2540 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
2542 co
= qemu_coroutine_create(bdrv_discard_co_entry
);
2543 qemu_coroutine_enter(co
, &rwco
);
2544 while (rwco
.ret
== NOT_DONE
) {
2545 aio_poll(aio_context
, true);
2553 CoroutineIOCompletion
*co
;
2555 } BdrvIoctlCompletionData
;
2557 static void bdrv_ioctl_bh_cb(void *opaque
)
2559 BdrvIoctlCompletionData
*data
= opaque
;
2561 bdrv_co_io_em_complete(data
->co
, -ENOTSUP
);
2562 qemu_bh_delete(data
->bh
);
2565 static int bdrv_co_do_ioctl(BlockDriverState
*bs
, int req
, void *buf
)
2567 BlockDriver
*drv
= bs
->drv
;
2568 BdrvTrackedRequest tracked_req
;
2569 CoroutineIOCompletion co
= {
2570 .coroutine
= qemu_coroutine_self(),
2574 tracked_request_begin(&tracked_req
, bs
, 0, 0, BDRV_TRACKED_IOCTL
);
2575 if (!drv
|| !drv
->bdrv_aio_ioctl
) {
2580 acb
= drv
->bdrv_aio_ioctl(bs
, req
, buf
, bdrv_co_io_em_complete
, &co
);
2582 BdrvIoctlCompletionData
*data
= g_new(BdrvIoctlCompletionData
, 1);
2583 data
->bh
= aio_bh_new(bdrv_get_aio_context(bs
),
2584 bdrv_ioctl_bh_cb
, data
);
2586 qemu_bh_schedule(data
->bh
);
2588 qemu_coroutine_yield();
2590 tracked_request_end(&tracked_req
);
2595 BlockDriverState
*bs
;
2601 static void coroutine_fn
bdrv_co_ioctl_entry(void *opaque
)
2603 BdrvIoctlCoData
*data
= opaque
;
2604 data
->ret
= bdrv_co_do_ioctl(data
->bs
, data
->req
, data
->buf
);
2607 /* needed for generic scsi interface */
2608 int bdrv_ioctl(BlockDriverState
*bs
, unsigned long int req
, void *buf
)
2610 BdrvIoctlCoData data
= {
2614 .ret
= -EINPROGRESS
,
2617 if (qemu_in_coroutine()) {
2618 /* Fast-path if already in coroutine context */
2619 bdrv_co_ioctl_entry(&data
);
2621 Coroutine
*co
= qemu_coroutine_create(bdrv_co_ioctl_entry
);
2623 qemu_coroutine_enter(co
, &data
);
2624 while (data
.ret
== -EINPROGRESS
) {
2625 aio_poll(bdrv_get_aio_context(bs
), true);
2631 static void coroutine_fn
bdrv_co_aio_ioctl_entry(void *opaque
)
2633 BlockAIOCBCoroutine
*acb
= opaque
;
2634 acb
->req
.error
= bdrv_co_do_ioctl(acb
->common
.bs
,
2635 acb
->req
.req
, acb
->req
.buf
);
2636 bdrv_co_complete(acb
);
2639 BlockAIOCB
*bdrv_aio_ioctl(BlockDriverState
*bs
,
2640 unsigned long int req
, void *buf
,
2641 BlockCompletionFunc
*cb
, void *opaque
)
2643 BlockAIOCBCoroutine
*acb
= qemu_aio_get(&bdrv_em_co_aiocb_info
,
2647 acb
->need_bh
= true;
2648 acb
->req
.error
= -EINPROGRESS
;
2651 co
= qemu_coroutine_create(bdrv_co_aio_ioctl_entry
);
2652 qemu_coroutine_enter(co
, acb
);
2654 bdrv_co_maybe_schedule_bh(acb
);
2655 return &acb
->common
;
2658 void *qemu_blockalign(BlockDriverState
*bs
, size_t size
)
2660 return qemu_memalign(bdrv_opt_mem_align(bs
), size
);
2663 void *qemu_blockalign0(BlockDriverState
*bs
, size_t size
)
2665 return memset(qemu_blockalign(bs
, size
), 0, size
);
2668 void *qemu_try_blockalign(BlockDriverState
*bs
, size_t size
)
2670 size_t align
= bdrv_opt_mem_align(bs
);
2672 /* Ensure that NULL is never returned on success */
2678 return qemu_try_memalign(align
, size
);
2681 void *qemu_try_blockalign0(BlockDriverState
*bs
, size_t size
)
2683 void *mem
= qemu_try_blockalign(bs
, size
);
2686 memset(mem
, 0, size
);
2693 * Check if all memory in this vector is sector aligned.
2695 bool bdrv_qiov_is_aligned(BlockDriverState
*bs
, QEMUIOVector
*qiov
)
2698 size_t alignment
= bdrv_min_mem_align(bs
);
2700 for (i
= 0; i
< qiov
->niov
; i
++) {
2701 if ((uintptr_t) qiov
->iov
[i
].iov_base
% alignment
) {
2704 if (qiov
->iov
[i
].iov_len
% alignment
) {
2712 void bdrv_add_before_write_notifier(BlockDriverState
*bs
,
2713 NotifierWithReturn
*notifier
)
2715 notifier_with_return_list_add(&bs
->before_write_notifiers
, notifier
);
2718 void bdrv_io_plug(BlockDriverState
*bs
)
2720 BlockDriver
*drv
= bs
->drv
;
2721 if (drv
&& drv
->bdrv_io_plug
) {
2722 drv
->bdrv_io_plug(bs
);
2723 } else if (bs
->file
) {
2724 bdrv_io_plug(bs
->file
->bs
);
2728 void bdrv_io_unplug(BlockDriverState
*bs
)
2730 BlockDriver
*drv
= bs
->drv
;
2731 if (drv
&& drv
->bdrv_io_unplug
) {
2732 drv
->bdrv_io_unplug(bs
);
2733 } else if (bs
->file
) {
2734 bdrv_io_unplug(bs
->file
->bs
);
2738 void bdrv_flush_io_queue(BlockDriverState
*bs
)
2740 BlockDriver
*drv
= bs
->drv
;
2741 if (drv
&& drv
->bdrv_flush_io_queue
) {
2742 drv
->bdrv_flush_io_queue(bs
);
2743 } else if (bs
->file
) {
2744 bdrv_flush_io_queue(bs
->file
->bs
);
2746 bdrv_start_throttled_reqs(bs
);
2749 void bdrv_drained_begin(BlockDriverState
*bs
)
2751 if (!bs
->quiesce_counter
++) {
2752 aio_disable_external(bdrv_get_aio_context(bs
));
2757 void bdrv_drained_end(BlockDriverState
*bs
)
2759 assert(bs
->quiesce_counter
> 0);
2760 if (--bs
->quiesce_counter
> 0) {
2763 aio_enable_external(bdrv_get_aio_context(bs
));