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_preadv(BlockDriverState
*bs
,
48 int64_t offset
, unsigned int bytes
, QEMUIOVector
*qiov
,
49 BdrvRequestFlags flags
);
50 static int coroutine_fn
bdrv_co_do_pwritev(BlockDriverState
*bs
,
51 int64_t offset
, unsigned int bytes
, QEMUIOVector
*qiov
,
52 BdrvRequestFlags flags
);
53 static BlockAIOCB
*bdrv_co_aio_rw_vector(BlockDriverState
*bs
,
57 BdrvRequestFlags flags
,
58 BlockCompletionFunc
*cb
,
61 static void coroutine_fn
bdrv_co_do_rw(void *opaque
);
62 static int coroutine_fn
bdrv_co_do_write_zeroes(BlockDriverState
*bs
,
63 int64_t sector_num
, int nb_sectors
, BdrvRequestFlags flags
);
65 /* throttling disk I/O limits */
66 void bdrv_set_io_limits(BlockDriverState
*bs
,
71 throttle_group_config(bs
, cfg
);
73 for (i
= 0; i
< 2; i
++) {
74 qemu_co_enter_next(&bs
->throttled_reqs
[i
]);
78 /* this function drain all the throttled IOs */
79 static bool bdrv_start_throttled_reqs(BlockDriverState
*bs
)
82 bool enabled
= bs
->io_limits_enabled
;
85 bs
->io_limits_enabled
= false;
87 for (i
= 0; i
< 2; i
++) {
88 while (qemu_co_enter_next(&bs
->throttled_reqs
[i
])) {
93 bs
->io_limits_enabled
= enabled
;
98 void bdrv_io_limits_disable(BlockDriverState
*bs
)
100 bs
->io_limits_enabled
= false;
101 bdrv_start_throttled_reqs(bs
);
102 throttle_group_unregister_bs(bs
);
105 /* should be called before bdrv_set_io_limits if a limit is set */
106 void bdrv_io_limits_enable(BlockDriverState
*bs
, const char *group
)
108 assert(!bs
->io_limits_enabled
);
109 throttle_group_register_bs(bs
, group
);
110 bs
->io_limits_enabled
= true;
113 void bdrv_io_limits_update_group(BlockDriverState
*bs
, const char *group
)
115 /* this bs is not part of any group */
116 if (!bs
->throttle_state
) {
120 /* this bs is a part of the same group than the one we want */
121 if (!g_strcmp0(throttle_group_get_name(bs
), group
)) {
125 /* need to change the group this bs belong to */
126 bdrv_io_limits_disable(bs
);
127 bdrv_io_limits_enable(bs
, group
);
130 void bdrv_setup_io_funcs(BlockDriver
*bdrv
)
132 /* Block drivers without coroutine functions need emulation */
133 if (!bdrv
->bdrv_co_readv
) {
134 bdrv
->bdrv_co_readv
= bdrv_co_readv_em
;
135 bdrv
->bdrv_co_writev
= bdrv_co_writev_em
;
137 /* bdrv_co_readv_em()/brdv_co_writev_em() work in terms of aio, so if
138 * the block driver lacks aio we need to emulate that too.
140 if (!bdrv
->bdrv_aio_readv
) {
141 /* add AIO emulation layer */
142 bdrv
->bdrv_aio_readv
= bdrv_aio_readv_em
;
143 bdrv
->bdrv_aio_writev
= bdrv_aio_writev_em
;
148 void bdrv_refresh_limits(BlockDriverState
*bs
, Error
**errp
)
150 BlockDriver
*drv
= bs
->drv
;
151 Error
*local_err
= NULL
;
153 memset(&bs
->bl
, 0, sizeof(bs
->bl
));
159 /* Take some limits from the children as a default */
161 bdrv_refresh_limits(bs
->file
->bs
, &local_err
);
163 error_propagate(errp
, local_err
);
166 bs
->bl
.opt_transfer_length
= bs
->file
->bs
->bl
.opt_transfer_length
;
167 bs
->bl
.max_transfer_length
= bs
->file
->bs
->bl
.max_transfer_length
;
168 bs
->bl
.min_mem_alignment
= bs
->file
->bs
->bl
.min_mem_alignment
;
169 bs
->bl
.opt_mem_alignment
= bs
->file
->bs
->bl
.opt_mem_alignment
;
170 bs
->bl
.max_iov
= bs
->file
->bs
->bl
.max_iov
;
172 bs
->bl
.min_mem_alignment
= 512;
173 bs
->bl
.opt_mem_alignment
= getpagesize();
175 /* Safe default since most protocols use readv()/writev()/etc */
176 bs
->bl
.max_iov
= IOV_MAX
;
180 bdrv_refresh_limits(bs
->backing
->bs
, &local_err
);
182 error_propagate(errp
, local_err
);
185 bs
->bl
.opt_transfer_length
=
186 MAX(bs
->bl
.opt_transfer_length
,
187 bs
->backing
->bs
->bl
.opt_transfer_length
);
188 bs
->bl
.max_transfer_length
=
189 MIN_NON_ZERO(bs
->bl
.max_transfer_length
,
190 bs
->backing
->bs
->bl
.max_transfer_length
);
191 bs
->bl
.opt_mem_alignment
=
192 MAX(bs
->bl
.opt_mem_alignment
,
193 bs
->backing
->bs
->bl
.opt_mem_alignment
);
194 bs
->bl
.min_mem_alignment
=
195 MAX(bs
->bl
.min_mem_alignment
,
196 bs
->backing
->bs
->bl
.min_mem_alignment
);
199 bs
->backing
->bs
->bl
.max_iov
);
202 /* Then let the driver override it */
203 if (drv
->bdrv_refresh_limits
) {
204 drv
->bdrv_refresh_limits(bs
, errp
);
209 * The copy-on-read flag is actually a reference count so multiple users may
210 * use the feature without worrying about clobbering its previous state.
211 * Copy-on-read stays enabled until all users have called to disable it.
213 void bdrv_enable_copy_on_read(BlockDriverState
*bs
)
218 void bdrv_disable_copy_on_read(BlockDriverState
*bs
)
220 assert(bs
->copy_on_read
> 0);
224 /* Check if any requests are in-flight (including throttled requests) */
225 bool bdrv_requests_pending(BlockDriverState
*bs
)
229 if (!QLIST_EMPTY(&bs
->tracked_requests
)) {
232 if (!qemu_co_queue_empty(&bs
->throttled_reqs
[0])) {
235 if (!qemu_co_queue_empty(&bs
->throttled_reqs
[1])) {
239 QLIST_FOREACH(child
, &bs
->children
, next
) {
240 if (bdrv_requests_pending(child
->bs
)) {
248 static void bdrv_drain_recurse(BlockDriverState
*bs
)
252 if (bs
->drv
&& bs
->drv
->bdrv_drain
) {
253 bs
->drv
->bdrv_drain(bs
);
255 QLIST_FOREACH(child
, &bs
->children
, next
) {
256 bdrv_drain_recurse(child
->bs
);
261 * Wait for pending requests to complete on a single BlockDriverState subtree,
262 * and suspend block driver's internal I/O until next request arrives.
264 * Note that unlike bdrv_drain_all(), the caller must hold the BlockDriverState
267 * Only this BlockDriverState's AioContext is run, so in-flight requests must
268 * not depend on events in other AioContexts. In that case, use
269 * bdrv_drain_all() instead.
271 void bdrv_drain(BlockDriverState
*bs
)
275 bdrv_drain_recurse(bs
);
278 bdrv_flush_io_queue(bs
);
279 busy
= bdrv_requests_pending(bs
);
280 busy
|= aio_poll(bdrv_get_aio_context(bs
), busy
);
285 * Wait for pending requests to complete across all BlockDriverStates
287 * This function does not flush data to disk, use bdrv_flush_all() for that
288 * after calling this function.
290 void bdrv_drain_all(void)
292 /* Always run first iteration so any pending completion BHs run */
294 BlockDriverState
*bs
= NULL
;
295 GSList
*aio_ctxs
= NULL
, *ctx
;
297 while ((bs
= bdrv_next(bs
))) {
298 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
300 aio_context_acquire(aio_context
);
302 block_job_pause(bs
->job
);
304 aio_context_release(aio_context
);
306 if (!g_slist_find(aio_ctxs
, aio_context
)) {
307 aio_ctxs
= g_slist_prepend(aio_ctxs
, aio_context
);
311 /* Note that completion of an asynchronous I/O operation can trigger any
312 * number of other I/O operations on other devices---for example a
313 * coroutine can submit an I/O request to another device in response to
314 * request completion. Therefore we must keep looping until there was no
315 * more activity rather than simply draining each device independently.
320 for (ctx
= aio_ctxs
; ctx
!= NULL
; ctx
= ctx
->next
) {
321 AioContext
*aio_context
= ctx
->data
;
324 aio_context_acquire(aio_context
);
325 while ((bs
= bdrv_next(bs
))) {
326 if (aio_context
== bdrv_get_aio_context(bs
)) {
327 bdrv_flush_io_queue(bs
);
328 if (bdrv_requests_pending(bs
)) {
330 aio_poll(aio_context
, busy
);
334 busy
|= aio_poll(aio_context
, false);
335 aio_context_release(aio_context
);
340 while ((bs
= bdrv_next(bs
))) {
341 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
343 aio_context_acquire(aio_context
);
345 block_job_resume(bs
->job
);
347 aio_context_release(aio_context
);
349 g_slist_free(aio_ctxs
);
353 * Remove an active request from the tracked requests list
355 * This function should be called when a tracked request is completing.
357 static void tracked_request_end(BdrvTrackedRequest
*req
)
359 if (req
->serialising
) {
360 req
->bs
->serialising_in_flight
--;
363 QLIST_REMOVE(req
, list
);
364 qemu_co_queue_restart_all(&req
->wait_queue
);
368 * Add an active request to the tracked requests list
370 static void tracked_request_begin(BdrvTrackedRequest
*req
,
371 BlockDriverState
*bs
,
374 enum BdrvTrackedRequestType type
)
376 *req
= (BdrvTrackedRequest
){
381 .co
= qemu_coroutine_self(),
382 .serialising
= false,
383 .overlap_offset
= offset
,
384 .overlap_bytes
= bytes
,
387 qemu_co_queue_init(&req
->wait_queue
);
389 QLIST_INSERT_HEAD(&bs
->tracked_requests
, req
, list
);
392 static void mark_request_serialising(BdrvTrackedRequest
*req
, uint64_t align
)
394 int64_t overlap_offset
= req
->offset
& ~(align
- 1);
395 unsigned int overlap_bytes
= ROUND_UP(req
->offset
+ req
->bytes
, align
)
398 if (!req
->serialising
) {
399 req
->bs
->serialising_in_flight
++;
400 req
->serialising
= true;
403 req
->overlap_offset
= MIN(req
->overlap_offset
, overlap_offset
);
404 req
->overlap_bytes
= MAX(req
->overlap_bytes
, overlap_bytes
);
408 * Round a region to cluster boundaries
410 void bdrv_round_to_clusters(BlockDriverState
*bs
,
411 int64_t sector_num
, int nb_sectors
,
412 int64_t *cluster_sector_num
,
413 int *cluster_nb_sectors
)
417 if (bdrv_get_info(bs
, &bdi
) < 0 || bdi
.cluster_size
== 0) {
418 *cluster_sector_num
= sector_num
;
419 *cluster_nb_sectors
= nb_sectors
;
421 int64_t c
= bdi
.cluster_size
/ BDRV_SECTOR_SIZE
;
422 *cluster_sector_num
= QEMU_ALIGN_DOWN(sector_num
, c
);
423 *cluster_nb_sectors
= QEMU_ALIGN_UP(sector_num
- *cluster_sector_num
+
428 static int bdrv_get_cluster_size(BlockDriverState
*bs
)
433 ret
= bdrv_get_info(bs
, &bdi
);
434 if (ret
< 0 || bdi
.cluster_size
== 0) {
435 return bs
->request_alignment
;
437 return bdi
.cluster_size
;
441 static bool tracked_request_overlaps(BdrvTrackedRequest
*req
,
442 int64_t offset
, unsigned int bytes
)
445 if (offset
>= req
->overlap_offset
+ req
->overlap_bytes
) {
449 if (req
->overlap_offset
>= offset
+ bytes
) {
455 static bool coroutine_fn
wait_serialising_requests(BdrvTrackedRequest
*self
)
457 BlockDriverState
*bs
= self
->bs
;
458 BdrvTrackedRequest
*req
;
462 if (!bs
->serialising_in_flight
) {
468 QLIST_FOREACH(req
, &bs
->tracked_requests
, list
) {
469 if (req
== self
|| (!req
->serialising
&& !self
->serialising
)) {
472 if (tracked_request_overlaps(req
, self
->overlap_offset
,
473 self
->overlap_bytes
))
475 /* Hitting this means there was a reentrant request, for
476 * example, a block driver issuing nested requests. This must
477 * never happen since it means deadlock.
479 assert(qemu_coroutine_self() != req
->co
);
481 /* If the request is already (indirectly) waiting for us, or
482 * will wait for us as soon as it wakes up, then just go on
483 * (instead of producing a deadlock in the former case). */
484 if (!req
->waiting_for
) {
485 self
->waiting_for
= req
;
486 qemu_co_queue_wait(&req
->wait_queue
);
487 self
->waiting_for
= NULL
;
499 static int bdrv_check_byte_request(BlockDriverState
*bs
, int64_t offset
,
502 if (size
> BDRV_REQUEST_MAX_SECTORS
<< BDRV_SECTOR_BITS
) {
506 if (!bdrv_is_inserted(bs
)) {
517 static int bdrv_check_request(BlockDriverState
*bs
, int64_t sector_num
,
520 if (nb_sectors
< 0 || nb_sectors
> BDRV_REQUEST_MAX_SECTORS
) {
524 return bdrv_check_byte_request(bs
, sector_num
* BDRV_SECTOR_SIZE
,
525 nb_sectors
* BDRV_SECTOR_SIZE
);
528 typedef struct RwCo
{
529 BlockDriverState
*bs
;
534 BdrvRequestFlags flags
;
537 static void coroutine_fn
bdrv_rw_co_entry(void *opaque
)
541 if (!rwco
->is_write
) {
542 rwco
->ret
= bdrv_co_do_preadv(rwco
->bs
, rwco
->offset
,
543 rwco
->qiov
->size
, rwco
->qiov
,
546 rwco
->ret
= bdrv_co_do_pwritev(rwco
->bs
, rwco
->offset
,
547 rwco
->qiov
->size
, rwco
->qiov
,
553 * Process a vectored synchronous request using coroutines
555 static int bdrv_prwv_co(BlockDriverState
*bs
, int64_t offset
,
556 QEMUIOVector
*qiov
, bool is_write
,
557 BdrvRequestFlags flags
)
564 .is_write
= is_write
,
570 * In sync call context, when the vcpu is blocked, this throttling timer
571 * will not fire; so the I/O throttling function has to be disabled here
572 * if it has been enabled.
574 if (bs
->io_limits_enabled
) {
575 fprintf(stderr
, "Disabling I/O throttling on '%s' due "
576 "to synchronous I/O.\n", bdrv_get_device_name(bs
));
577 bdrv_io_limits_disable(bs
);
580 if (qemu_in_coroutine()) {
581 /* Fast-path if already in coroutine context */
582 bdrv_rw_co_entry(&rwco
);
584 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
586 co
= qemu_coroutine_create(bdrv_rw_co_entry
);
587 qemu_coroutine_enter(co
, &rwco
);
588 while (rwco
.ret
== NOT_DONE
) {
589 aio_poll(aio_context
, true);
596 * Process a synchronous request using coroutines
598 static int bdrv_rw_co(BlockDriverState
*bs
, int64_t sector_num
, uint8_t *buf
,
599 int nb_sectors
, bool is_write
, BdrvRequestFlags flags
)
603 .iov_base
= (void *)buf
,
604 .iov_len
= nb_sectors
* BDRV_SECTOR_SIZE
,
607 if (nb_sectors
< 0 || nb_sectors
> BDRV_REQUEST_MAX_SECTORS
) {
611 qemu_iovec_init_external(&qiov
, &iov
, 1);
612 return bdrv_prwv_co(bs
, sector_num
<< BDRV_SECTOR_BITS
,
613 &qiov
, is_write
, flags
);
616 /* return < 0 if error. See bdrv_write() for the return codes */
617 int bdrv_read(BlockDriverState
*bs
, int64_t sector_num
,
618 uint8_t *buf
, int nb_sectors
)
620 return bdrv_rw_co(bs
, sector_num
, buf
, nb_sectors
, false, 0);
623 /* Just like bdrv_read(), but with I/O throttling temporarily disabled */
624 int bdrv_read_unthrottled(BlockDriverState
*bs
, int64_t sector_num
,
625 uint8_t *buf
, int nb_sectors
)
630 enabled
= bs
->io_limits_enabled
;
631 bs
->io_limits_enabled
= false;
632 ret
= bdrv_read(bs
, sector_num
, buf
, nb_sectors
);
633 bs
->io_limits_enabled
= enabled
;
637 /* Return < 0 if error. Important errors are:
638 -EIO generic I/O error (may happen for all errors)
639 -ENOMEDIUM No media inserted.
640 -EINVAL Invalid sector number or nb_sectors
641 -EACCES Trying to write a read-only device
643 int bdrv_write(BlockDriverState
*bs
, int64_t sector_num
,
644 const uint8_t *buf
, int nb_sectors
)
646 return bdrv_rw_co(bs
, sector_num
, (uint8_t *)buf
, nb_sectors
, true, 0);
649 int bdrv_write_zeroes(BlockDriverState
*bs
, int64_t sector_num
,
650 int nb_sectors
, BdrvRequestFlags flags
)
652 return bdrv_rw_co(bs
, sector_num
, NULL
, nb_sectors
, true,
653 BDRV_REQ_ZERO_WRITE
| flags
);
657 * Completely zero out a block device with the help of bdrv_write_zeroes.
658 * The operation is sped up by checking the block status and only writing
659 * zeroes to the device if they currently do not return zeroes. Optional
660 * flags are passed through to bdrv_write_zeroes (e.g. BDRV_REQ_MAY_UNMAP).
662 * Returns < 0 on error, 0 on success. For error codes see bdrv_write().
664 int bdrv_make_zero(BlockDriverState
*bs
, BdrvRequestFlags flags
)
666 int64_t target_sectors
, ret
, nb_sectors
, sector_num
= 0;
667 BlockDriverState
*file
;
670 target_sectors
= bdrv_nb_sectors(bs
);
671 if (target_sectors
< 0) {
672 return target_sectors
;
676 nb_sectors
= MIN(target_sectors
- sector_num
, BDRV_REQUEST_MAX_SECTORS
);
677 if (nb_sectors
<= 0) {
680 ret
= bdrv_get_block_status(bs
, sector_num
, nb_sectors
, &n
, &file
);
682 error_report("error getting block status at sector %" PRId64
": %s",
683 sector_num
, strerror(-ret
));
686 if (ret
& BDRV_BLOCK_ZERO
) {
690 ret
= bdrv_write_zeroes(bs
, sector_num
, n
, flags
);
692 error_report("error writing zeroes at sector %" PRId64
": %s",
693 sector_num
, strerror(-ret
));
700 int bdrv_pread(BlockDriverState
*bs
, int64_t offset
, void *buf
, int bytes
)
704 .iov_base
= (void *)buf
,
713 qemu_iovec_init_external(&qiov
, &iov
, 1);
714 ret
= bdrv_prwv_co(bs
, offset
, &qiov
, false, 0);
722 int bdrv_pwritev(BlockDriverState
*bs
, int64_t offset
, QEMUIOVector
*qiov
)
726 ret
= bdrv_prwv_co(bs
, offset
, qiov
, true, 0);
734 int bdrv_pwrite(BlockDriverState
*bs
, int64_t offset
,
735 const void *buf
, int bytes
)
739 .iov_base
= (void *) buf
,
747 qemu_iovec_init_external(&qiov
, &iov
, 1);
748 return bdrv_pwritev(bs
, offset
, &qiov
);
752 * Writes to the file and ensures that no writes are reordered across this
753 * request (acts as a barrier)
755 * Returns 0 on success, -errno in error cases.
757 int bdrv_pwrite_sync(BlockDriverState
*bs
, int64_t offset
,
758 const void *buf
, int count
)
762 ret
= bdrv_pwrite(bs
, offset
, buf
, count
);
767 /* No flush needed for cache modes that already do it */
768 if (bs
->enable_write_cache
) {
775 static int coroutine_fn
bdrv_co_do_copy_on_readv(BlockDriverState
*bs
,
776 int64_t sector_num
, int nb_sectors
, QEMUIOVector
*qiov
)
778 /* Perform I/O through a temporary buffer so that users who scribble over
779 * their read buffer while the operation is in progress do not end up
780 * modifying the image file. This is critical for zero-copy guest I/O
781 * where anything might happen inside guest memory.
785 BlockDriver
*drv
= bs
->drv
;
787 QEMUIOVector bounce_qiov
;
788 int64_t cluster_sector_num
;
789 int cluster_nb_sectors
;
793 /* Cover entire cluster so no additional backing file I/O is required when
794 * allocating cluster in the image file.
796 bdrv_round_to_clusters(bs
, sector_num
, nb_sectors
,
797 &cluster_sector_num
, &cluster_nb_sectors
);
799 trace_bdrv_co_do_copy_on_readv(bs
, sector_num
, nb_sectors
,
800 cluster_sector_num
, cluster_nb_sectors
);
802 iov
.iov_len
= cluster_nb_sectors
* BDRV_SECTOR_SIZE
;
803 iov
.iov_base
= bounce_buffer
= qemu_try_blockalign(bs
, iov
.iov_len
);
804 if (bounce_buffer
== NULL
) {
809 qemu_iovec_init_external(&bounce_qiov
, &iov
, 1);
811 ret
= drv
->bdrv_co_readv(bs
, cluster_sector_num
, cluster_nb_sectors
,
817 if (drv
->bdrv_co_write_zeroes
&&
818 buffer_is_zero(bounce_buffer
, iov
.iov_len
)) {
819 ret
= bdrv_co_do_write_zeroes(bs
, cluster_sector_num
,
820 cluster_nb_sectors
, 0);
822 /* This does not change the data on the disk, it is not necessary
823 * to flush even in cache=writethrough mode.
825 ret
= drv
->bdrv_co_writev(bs
, cluster_sector_num
, cluster_nb_sectors
,
830 /* It might be okay to ignore write errors for guest requests. If this
831 * is a deliberate copy-on-read then we don't want to ignore the error.
832 * Simply report it in all cases.
837 skip_bytes
= (sector_num
- cluster_sector_num
) * BDRV_SECTOR_SIZE
;
838 qemu_iovec_from_buf(qiov
, 0, bounce_buffer
+ skip_bytes
,
839 nb_sectors
* BDRV_SECTOR_SIZE
);
842 qemu_vfree(bounce_buffer
);
847 * Forwards an already correctly aligned request to the BlockDriver. This
848 * handles copy on read and zeroing after EOF; any other features must be
849 * implemented by the caller.
851 static int coroutine_fn
bdrv_aligned_preadv(BlockDriverState
*bs
,
852 BdrvTrackedRequest
*req
, int64_t offset
, unsigned int bytes
,
853 int64_t align
, QEMUIOVector
*qiov
, int flags
)
855 BlockDriver
*drv
= bs
->drv
;
858 int64_t sector_num
= offset
>> BDRV_SECTOR_BITS
;
859 unsigned int nb_sectors
= bytes
>> BDRV_SECTOR_BITS
;
861 assert((offset
& (BDRV_SECTOR_SIZE
- 1)) == 0);
862 assert((bytes
& (BDRV_SECTOR_SIZE
- 1)) == 0);
863 assert(!qiov
|| bytes
== qiov
->size
);
865 /* Handle Copy on Read and associated serialisation */
866 if (flags
& BDRV_REQ_COPY_ON_READ
) {
867 /* If we touch the same cluster it counts as an overlap. This
868 * guarantees that allocating writes will be serialized and not race
869 * with each other for the same cluster. For example, in copy-on-read
870 * it ensures that the CoR read and write operations are atomic and
871 * guest writes cannot interleave between them. */
872 mark_request_serialising(req
, bdrv_get_cluster_size(bs
));
875 if (!(flags
& BDRV_REQ_NO_SERIALISING
)) {
876 wait_serialising_requests(req
);
879 if (flags
& BDRV_REQ_COPY_ON_READ
) {
882 ret
= bdrv_is_allocated(bs
, sector_num
, nb_sectors
, &pnum
);
887 if (!ret
|| pnum
!= nb_sectors
) {
888 ret
= bdrv_co_do_copy_on_readv(bs
, sector_num
, nb_sectors
, qiov
);
893 /* Forward the request to the BlockDriver */
894 if (!bs
->zero_beyond_eof
) {
895 ret
= drv
->bdrv_co_readv(bs
, sector_num
, nb_sectors
, qiov
);
897 /* Read zeros after EOF */
898 int64_t total_sectors
, max_nb_sectors
;
900 total_sectors
= bdrv_nb_sectors(bs
);
901 if (total_sectors
< 0) {
906 max_nb_sectors
= ROUND_UP(MAX(0, total_sectors
- sector_num
),
907 align
>> BDRV_SECTOR_BITS
);
908 if (nb_sectors
< max_nb_sectors
) {
909 ret
= drv
->bdrv_co_readv(bs
, sector_num
, nb_sectors
, qiov
);
910 } else if (max_nb_sectors
> 0) {
911 QEMUIOVector local_qiov
;
913 qemu_iovec_init(&local_qiov
, qiov
->niov
);
914 qemu_iovec_concat(&local_qiov
, qiov
, 0,
915 max_nb_sectors
* BDRV_SECTOR_SIZE
);
917 ret
= drv
->bdrv_co_readv(bs
, sector_num
, max_nb_sectors
,
920 qemu_iovec_destroy(&local_qiov
);
925 /* Reading beyond end of file is supposed to produce zeroes */
926 if (ret
== 0 && total_sectors
< sector_num
+ nb_sectors
) {
927 uint64_t offset
= MAX(0, total_sectors
- sector_num
);
928 uint64_t bytes
= (sector_num
+ nb_sectors
- offset
) *
930 qemu_iovec_memset(qiov
, offset
* BDRV_SECTOR_SIZE
, 0, bytes
);
939 * Handle a read request in coroutine context
941 static int coroutine_fn
bdrv_co_do_preadv(BlockDriverState
*bs
,
942 int64_t offset
, unsigned int bytes
, QEMUIOVector
*qiov
,
943 BdrvRequestFlags flags
)
945 BlockDriver
*drv
= bs
->drv
;
946 BdrvTrackedRequest req
;
948 /* TODO Lift BDRV_SECTOR_SIZE restriction in BlockDriver interface */
949 uint64_t align
= MAX(BDRV_SECTOR_SIZE
, bs
->request_alignment
);
950 uint8_t *head_buf
= NULL
;
951 uint8_t *tail_buf
= NULL
;
952 QEMUIOVector local_qiov
;
953 bool use_local_qiov
= false;
960 ret
= bdrv_check_byte_request(bs
, offset
, bytes
);
965 /* Don't do copy-on-read if we read data before write operation */
966 if (bs
->copy_on_read
&& !(flags
& BDRV_REQ_NO_SERIALISING
)) {
967 flags
|= BDRV_REQ_COPY_ON_READ
;
970 /* throttling disk I/O */
971 if (bs
->io_limits_enabled
) {
972 throttle_group_co_io_limits_intercept(bs
, bytes
, false);
975 /* Align read if necessary by padding qiov */
976 if (offset
& (align
- 1)) {
977 head_buf
= qemu_blockalign(bs
, align
);
978 qemu_iovec_init(&local_qiov
, qiov
->niov
+ 2);
979 qemu_iovec_add(&local_qiov
, head_buf
, offset
& (align
- 1));
980 qemu_iovec_concat(&local_qiov
, qiov
, 0, qiov
->size
);
981 use_local_qiov
= true;
983 bytes
+= offset
& (align
- 1);
984 offset
= offset
& ~(align
- 1);
987 if ((offset
+ bytes
) & (align
- 1)) {
988 if (!use_local_qiov
) {
989 qemu_iovec_init(&local_qiov
, qiov
->niov
+ 1);
990 qemu_iovec_concat(&local_qiov
, qiov
, 0, qiov
->size
);
991 use_local_qiov
= true;
993 tail_buf
= qemu_blockalign(bs
, align
);
994 qemu_iovec_add(&local_qiov
, tail_buf
,
995 align
- ((offset
+ bytes
) & (align
- 1)));
997 bytes
= ROUND_UP(bytes
, align
);
1000 tracked_request_begin(&req
, bs
, offset
, bytes
, BDRV_TRACKED_READ
);
1001 ret
= bdrv_aligned_preadv(bs
, &req
, offset
, bytes
, align
,
1002 use_local_qiov
? &local_qiov
: qiov
,
1004 tracked_request_end(&req
);
1006 if (use_local_qiov
) {
1007 qemu_iovec_destroy(&local_qiov
);
1008 qemu_vfree(head_buf
);
1009 qemu_vfree(tail_buf
);
1015 static int coroutine_fn
bdrv_co_do_readv(BlockDriverState
*bs
,
1016 int64_t sector_num
, int nb_sectors
, QEMUIOVector
*qiov
,
1017 BdrvRequestFlags flags
)
1019 if (nb_sectors
< 0 || nb_sectors
> BDRV_REQUEST_MAX_SECTORS
) {
1023 return bdrv_co_do_preadv(bs
, sector_num
<< BDRV_SECTOR_BITS
,
1024 nb_sectors
<< BDRV_SECTOR_BITS
, qiov
, flags
);
1027 int coroutine_fn
bdrv_co_readv(BlockDriverState
*bs
, int64_t sector_num
,
1028 int nb_sectors
, QEMUIOVector
*qiov
)
1030 trace_bdrv_co_readv(bs
, sector_num
, nb_sectors
);
1032 return bdrv_co_do_readv(bs
, sector_num
, nb_sectors
, qiov
, 0);
1035 int coroutine_fn
bdrv_co_readv_no_serialising(BlockDriverState
*bs
,
1036 int64_t sector_num
, int nb_sectors
, QEMUIOVector
*qiov
)
1038 trace_bdrv_co_readv_no_serialising(bs
, sector_num
, nb_sectors
);
1040 return bdrv_co_do_readv(bs
, sector_num
, nb_sectors
, qiov
,
1041 BDRV_REQ_NO_SERIALISING
);
1044 int coroutine_fn
bdrv_co_copy_on_readv(BlockDriverState
*bs
,
1045 int64_t sector_num
, int nb_sectors
, QEMUIOVector
*qiov
)
1047 trace_bdrv_co_copy_on_readv(bs
, sector_num
, nb_sectors
);
1049 return bdrv_co_do_readv(bs
, sector_num
, nb_sectors
, qiov
,
1050 BDRV_REQ_COPY_ON_READ
);
1053 #define MAX_WRITE_ZEROES_BOUNCE_BUFFER 32768
1055 static int coroutine_fn
bdrv_co_do_write_zeroes(BlockDriverState
*bs
,
1056 int64_t sector_num
, int nb_sectors
, BdrvRequestFlags flags
)
1058 BlockDriver
*drv
= bs
->drv
;
1060 struct iovec iov
= {0};
1063 int max_write_zeroes
= MIN_NON_ZERO(bs
->bl
.max_write_zeroes
,
1064 BDRV_REQUEST_MAX_SECTORS
);
1066 while (nb_sectors
> 0 && !ret
) {
1067 int num
= nb_sectors
;
1069 /* Align request. Block drivers can expect the "bulk" of the request
1072 if (bs
->bl
.write_zeroes_alignment
1073 && num
> bs
->bl
.write_zeroes_alignment
) {
1074 if (sector_num
% bs
->bl
.write_zeroes_alignment
!= 0) {
1075 /* Make a small request up to the first aligned sector. */
1076 num
= bs
->bl
.write_zeroes_alignment
;
1077 num
-= sector_num
% bs
->bl
.write_zeroes_alignment
;
1078 } else if ((sector_num
+ num
) % bs
->bl
.write_zeroes_alignment
!= 0) {
1079 /* Shorten the request to the last aligned sector. num cannot
1080 * underflow because num > bs->bl.write_zeroes_alignment.
1082 num
-= (sector_num
+ num
) % bs
->bl
.write_zeroes_alignment
;
1086 /* limit request size */
1087 if (num
> max_write_zeroes
) {
1088 num
= max_write_zeroes
;
1092 /* First try the efficient write zeroes operation */
1093 if (drv
->bdrv_co_write_zeroes
) {
1094 ret
= drv
->bdrv_co_write_zeroes(bs
, sector_num
, num
, flags
);
1097 if (ret
== -ENOTSUP
) {
1098 /* Fall back to bounce buffer if write zeroes is unsupported */
1099 int max_xfer_len
= MIN_NON_ZERO(bs
->bl
.max_transfer_length
,
1100 MAX_WRITE_ZEROES_BOUNCE_BUFFER
);
1101 num
= MIN(num
, max_xfer_len
);
1102 iov
.iov_len
= num
* BDRV_SECTOR_SIZE
;
1103 if (iov
.iov_base
== NULL
) {
1104 iov
.iov_base
= qemu_try_blockalign(bs
, num
* BDRV_SECTOR_SIZE
);
1105 if (iov
.iov_base
== NULL
) {
1109 memset(iov
.iov_base
, 0, num
* BDRV_SECTOR_SIZE
);
1111 qemu_iovec_init_external(&qiov
, &iov
, 1);
1113 ret
= drv
->bdrv_co_writev(bs
, sector_num
, num
, &qiov
);
1115 /* Keep bounce buffer around if it is big enough for all
1116 * all future requests.
1118 if (num
< max_xfer_len
) {
1119 qemu_vfree(iov
.iov_base
);
1120 iov
.iov_base
= NULL
;
1129 qemu_vfree(iov
.iov_base
);
1134 * Forwards an already correctly aligned write request to the BlockDriver.
1136 static int coroutine_fn
bdrv_aligned_pwritev(BlockDriverState
*bs
,
1137 BdrvTrackedRequest
*req
, int64_t offset
, unsigned int bytes
,
1138 QEMUIOVector
*qiov
, int flags
)
1140 BlockDriver
*drv
= bs
->drv
;
1144 int64_t sector_num
= offset
>> BDRV_SECTOR_BITS
;
1145 unsigned int nb_sectors
= bytes
>> BDRV_SECTOR_BITS
;
1147 assert((offset
& (BDRV_SECTOR_SIZE
- 1)) == 0);
1148 assert((bytes
& (BDRV_SECTOR_SIZE
- 1)) == 0);
1149 assert(!qiov
|| bytes
== qiov
->size
);
1151 waited
= wait_serialising_requests(req
);
1152 assert(!waited
|| !req
->serialising
);
1153 assert(req
->overlap_offset
<= offset
);
1154 assert(offset
+ bytes
<= req
->overlap_offset
+ req
->overlap_bytes
);
1156 ret
= notifier_with_return_list_notify(&bs
->before_write_notifiers
, req
);
1158 if (!ret
&& bs
->detect_zeroes
!= BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF
&&
1159 !(flags
& BDRV_REQ_ZERO_WRITE
) && drv
->bdrv_co_write_zeroes
&&
1160 qemu_iovec_is_zero(qiov
)) {
1161 flags
|= BDRV_REQ_ZERO_WRITE
;
1162 if (bs
->detect_zeroes
== BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP
) {
1163 flags
|= BDRV_REQ_MAY_UNMAP
;
1168 /* Do nothing, write notifier decided to fail this request */
1169 } else if (flags
& BDRV_REQ_ZERO_WRITE
) {
1170 bdrv_debug_event(bs
, BLKDBG_PWRITEV_ZERO
);
1171 ret
= bdrv_co_do_write_zeroes(bs
, sector_num
, nb_sectors
, flags
);
1173 bdrv_debug_event(bs
, BLKDBG_PWRITEV
);
1174 ret
= drv
->bdrv_co_writev(bs
, sector_num
, nb_sectors
, qiov
);
1176 bdrv_debug_event(bs
, BLKDBG_PWRITEV_DONE
);
1178 if (ret
== 0 && !bs
->enable_write_cache
) {
1179 ret
= bdrv_co_flush(bs
);
1182 bdrv_set_dirty(bs
, sector_num
, nb_sectors
);
1184 if (bs
->wr_highest_offset
< offset
+ bytes
) {
1185 bs
->wr_highest_offset
= offset
+ bytes
;
1189 bs
->total_sectors
= MAX(bs
->total_sectors
, sector_num
+ nb_sectors
);
1195 static int coroutine_fn
bdrv_co_do_zero_pwritev(BlockDriverState
*bs
,
1198 BdrvRequestFlags flags
,
1199 BdrvTrackedRequest
*req
)
1201 uint8_t *buf
= NULL
;
1202 QEMUIOVector local_qiov
;
1204 uint64_t align
= MAX(BDRV_SECTOR_SIZE
, bs
->request_alignment
);
1205 unsigned int head_padding_bytes
, tail_padding_bytes
;
1208 head_padding_bytes
= offset
& (align
- 1);
1209 tail_padding_bytes
= align
- ((offset
+ bytes
) & (align
- 1));
1212 assert(flags
& BDRV_REQ_ZERO_WRITE
);
1213 if (head_padding_bytes
|| tail_padding_bytes
) {
1214 buf
= qemu_blockalign(bs
, align
);
1215 iov
= (struct iovec
) {
1219 qemu_iovec_init_external(&local_qiov
, &iov
, 1);
1221 if (head_padding_bytes
) {
1222 uint64_t zero_bytes
= MIN(bytes
, align
- head_padding_bytes
);
1224 /* RMW the unaligned part before head. */
1225 mark_request_serialising(req
, align
);
1226 wait_serialising_requests(req
);
1227 bdrv_debug_event(bs
, BLKDBG_PWRITEV_RMW_HEAD
);
1228 ret
= bdrv_aligned_preadv(bs
, req
, offset
& ~(align
- 1), align
,
1229 align
, &local_qiov
, 0);
1233 bdrv_debug_event(bs
, BLKDBG_PWRITEV_RMW_AFTER_HEAD
);
1235 memset(buf
+ head_padding_bytes
, 0, zero_bytes
);
1236 ret
= bdrv_aligned_pwritev(bs
, req
, offset
& ~(align
- 1), align
,
1238 flags
& ~BDRV_REQ_ZERO_WRITE
);
1242 offset
+= zero_bytes
;
1243 bytes
-= zero_bytes
;
1246 assert(!bytes
|| (offset
& (align
- 1)) == 0);
1247 if (bytes
>= align
) {
1248 /* Write the aligned part in the middle. */
1249 uint64_t aligned_bytes
= bytes
& ~(align
- 1);
1250 ret
= bdrv_aligned_pwritev(bs
, req
, offset
, aligned_bytes
,
1255 bytes
-= aligned_bytes
;
1256 offset
+= aligned_bytes
;
1259 assert(!bytes
|| (offset
& (align
- 1)) == 0);
1261 assert(align
== tail_padding_bytes
+ bytes
);
1262 /* RMW the unaligned part after tail. */
1263 mark_request_serialising(req
, align
);
1264 wait_serialising_requests(req
);
1265 bdrv_debug_event(bs
, BLKDBG_PWRITEV_RMW_TAIL
);
1266 ret
= bdrv_aligned_preadv(bs
, req
, offset
, align
,
1267 align
, &local_qiov
, 0);
1271 bdrv_debug_event(bs
, BLKDBG_PWRITEV_RMW_AFTER_TAIL
);
1273 memset(buf
, 0, bytes
);
1274 ret
= bdrv_aligned_pwritev(bs
, req
, offset
, align
,
1275 &local_qiov
, flags
& ~BDRV_REQ_ZERO_WRITE
);
1284 * Handle a write request in coroutine context
1286 static int coroutine_fn
bdrv_co_do_pwritev(BlockDriverState
*bs
,
1287 int64_t offset
, unsigned int bytes
, QEMUIOVector
*qiov
,
1288 BdrvRequestFlags flags
)
1290 BdrvTrackedRequest req
;
1291 /* TODO Lift BDRV_SECTOR_SIZE restriction in BlockDriver interface */
1292 uint64_t align
= MAX(BDRV_SECTOR_SIZE
, bs
->request_alignment
);
1293 uint8_t *head_buf
= NULL
;
1294 uint8_t *tail_buf
= NULL
;
1295 QEMUIOVector local_qiov
;
1296 bool use_local_qiov
= false;
1302 if (bs
->read_only
) {
1305 assert(!(bs
->open_flags
& BDRV_O_INACTIVE
));
1307 ret
= bdrv_check_byte_request(bs
, offset
, bytes
);
1312 /* throttling disk I/O */
1313 if (bs
->io_limits_enabled
) {
1314 throttle_group_co_io_limits_intercept(bs
, bytes
, true);
1318 * Align write if necessary by performing a read-modify-write cycle.
1319 * Pad qiov with the read parts and be sure to have a tracked request not
1320 * only for bdrv_aligned_pwritev, but also for the reads of the RMW cycle.
1322 tracked_request_begin(&req
, bs
, offset
, bytes
, BDRV_TRACKED_WRITE
);
1325 ret
= bdrv_co_do_zero_pwritev(bs
, offset
, bytes
, flags
, &req
);
1329 if (offset
& (align
- 1)) {
1330 QEMUIOVector head_qiov
;
1331 struct iovec head_iov
;
1333 mark_request_serialising(&req
, align
);
1334 wait_serialising_requests(&req
);
1336 head_buf
= qemu_blockalign(bs
, align
);
1337 head_iov
= (struct iovec
) {
1338 .iov_base
= head_buf
,
1341 qemu_iovec_init_external(&head_qiov
, &head_iov
, 1);
1343 bdrv_debug_event(bs
, BLKDBG_PWRITEV_RMW_HEAD
);
1344 ret
= bdrv_aligned_preadv(bs
, &req
, offset
& ~(align
- 1), align
,
1345 align
, &head_qiov
, 0);
1349 bdrv_debug_event(bs
, BLKDBG_PWRITEV_RMW_AFTER_HEAD
);
1351 qemu_iovec_init(&local_qiov
, qiov
->niov
+ 2);
1352 qemu_iovec_add(&local_qiov
, head_buf
, offset
& (align
- 1));
1353 qemu_iovec_concat(&local_qiov
, qiov
, 0, qiov
->size
);
1354 use_local_qiov
= true;
1356 bytes
+= offset
& (align
- 1);
1357 offset
= offset
& ~(align
- 1);
1360 if ((offset
+ bytes
) & (align
- 1)) {
1361 QEMUIOVector tail_qiov
;
1362 struct iovec tail_iov
;
1366 mark_request_serialising(&req
, align
);
1367 waited
= wait_serialising_requests(&req
);
1368 assert(!waited
|| !use_local_qiov
);
1370 tail_buf
= qemu_blockalign(bs
, align
);
1371 tail_iov
= (struct iovec
) {
1372 .iov_base
= tail_buf
,
1375 qemu_iovec_init_external(&tail_qiov
, &tail_iov
, 1);
1377 bdrv_debug_event(bs
, BLKDBG_PWRITEV_RMW_TAIL
);
1378 ret
= bdrv_aligned_preadv(bs
, &req
, (offset
+ bytes
) & ~(align
- 1), align
,
1379 align
, &tail_qiov
, 0);
1383 bdrv_debug_event(bs
, BLKDBG_PWRITEV_RMW_AFTER_TAIL
);
1385 if (!use_local_qiov
) {
1386 qemu_iovec_init(&local_qiov
, qiov
->niov
+ 1);
1387 qemu_iovec_concat(&local_qiov
, qiov
, 0, qiov
->size
);
1388 use_local_qiov
= true;
1391 tail_bytes
= (offset
+ bytes
) & (align
- 1);
1392 qemu_iovec_add(&local_qiov
, tail_buf
+ tail_bytes
, align
- tail_bytes
);
1394 bytes
= ROUND_UP(bytes
, align
);
1397 ret
= bdrv_aligned_pwritev(bs
, &req
, offset
, bytes
,
1398 use_local_qiov
? &local_qiov
: qiov
,
1403 if (use_local_qiov
) {
1404 qemu_iovec_destroy(&local_qiov
);
1406 qemu_vfree(head_buf
);
1407 qemu_vfree(tail_buf
);
1409 tracked_request_end(&req
);
1413 static int coroutine_fn
bdrv_co_do_writev(BlockDriverState
*bs
,
1414 int64_t sector_num
, int nb_sectors
, QEMUIOVector
*qiov
,
1415 BdrvRequestFlags flags
)
1417 if (nb_sectors
< 0 || nb_sectors
> BDRV_REQUEST_MAX_SECTORS
) {
1421 return bdrv_co_do_pwritev(bs
, sector_num
<< BDRV_SECTOR_BITS
,
1422 nb_sectors
<< BDRV_SECTOR_BITS
, qiov
, flags
);
1425 int coroutine_fn
bdrv_co_writev(BlockDriverState
*bs
, int64_t sector_num
,
1426 int nb_sectors
, QEMUIOVector
*qiov
)
1428 trace_bdrv_co_writev(bs
, sector_num
, nb_sectors
);
1430 return bdrv_co_do_writev(bs
, sector_num
, nb_sectors
, qiov
, 0);
1433 int coroutine_fn
bdrv_co_write_zeroes(BlockDriverState
*bs
,
1434 int64_t sector_num
, int nb_sectors
,
1435 BdrvRequestFlags flags
)
1437 trace_bdrv_co_write_zeroes(bs
, sector_num
, nb_sectors
, flags
);
1439 if (!(bs
->open_flags
& BDRV_O_UNMAP
)) {
1440 flags
&= ~BDRV_REQ_MAY_UNMAP
;
1443 return bdrv_co_do_writev(bs
, sector_num
, nb_sectors
, NULL
,
1444 BDRV_REQ_ZERO_WRITE
| flags
);
1447 int bdrv_flush_all(void)
1449 BlockDriverState
*bs
= NULL
;
1452 while ((bs
= bdrv_next(bs
))) {
1453 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
1456 aio_context_acquire(aio_context
);
1457 ret
= bdrv_flush(bs
);
1458 if (ret
< 0 && !result
) {
1461 aio_context_release(aio_context
);
1467 typedef struct BdrvCoGetBlockStatusData
{
1468 BlockDriverState
*bs
;
1469 BlockDriverState
*base
;
1470 BlockDriverState
**file
;
1476 } BdrvCoGetBlockStatusData
;
1479 * Returns the allocation status of the specified sectors.
1480 * Drivers not implementing the functionality are assumed to not support
1481 * backing files, hence all their sectors are reported as allocated.
1483 * If 'sector_num' is beyond the end of the disk image the return value is 0
1484 * and 'pnum' is set to 0.
1486 * 'pnum' is set to the number of sectors (including and immediately following
1487 * the specified sector) that are known to be in the same
1488 * allocated/unallocated state.
1490 * 'nb_sectors' is the max value 'pnum' should be set to. If nb_sectors goes
1491 * beyond the end of the disk image it will be clamped.
1493 * If returned value is positive and BDRV_BLOCK_OFFSET_VALID bit is set, 'file'
1494 * points to the BDS which the sector range is allocated in.
1496 static int64_t coroutine_fn
bdrv_co_get_block_status(BlockDriverState
*bs
,
1498 int nb_sectors
, int *pnum
,
1499 BlockDriverState
**file
)
1501 int64_t total_sectors
;
1505 total_sectors
= bdrv_nb_sectors(bs
);
1506 if (total_sectors
< 0) {
1507 return total_sectors
;
1510 if (sector_num
>= total_sectors
) {
1515 n
= total_sectors
- sector_num
;
1516 if (n
< nb_sectors
) {
1520 if (!bs
->drv
->bdrv_co_get_block_status
) {
1522 ret
= BDRV_BLOCK_DATA
| BDRV_BLOCK_ALLOCATED
;
1523 if (bs
->drv
->protocol_name
) {
1524 ret
|= BDRV_BLOCK_OFFSET_VALID
| (sector_num
* BDRV_SECTOR_SIZE
);
1530 ret
= bs
->drv
->bdrv_co_get_block_status(bs
, sector_num
, nb_sectors
, pnum
,
1537 if (ret
& BDRV_BLOCK_RAW
) {
1538 assert(ret
& BDRV_BLOCK_OFFSET_VALID
);
1539 return bdrv_get_block_status(bs
->file
->bs
, ret
>> BDRV_SECTOR_BITS
,
1543 if (ret
& (BDRV_BLOCK_DATA
| BDRV_BLOCK_ZERO
)) {
1544 ret
|= BDRV_BLOCK_ALLOCATED
;
1546 if (bdrv_unallocated_blocks_are_zero(bs
)) {
1547 ret
|= BDRV_BLOCK_ZERO
;
1548 } else if (bs
->backing
) {
1549 BlockDriverState
*bs2
= bs
->backing
->bs
;
1550 int64_t nb_sectors2
= bdrv_nb_sectors(bs2
);
1551 if (nb_sectors2
>= 0 && sector_num
>= nb_sectors2
) {
1552 ret
|= BDRV_BLOCK_ZERO
;
1558 (ret
& BDRV_BLOCK_DATA
) && !(ret
& BDRV_BLOCK_ZERO
) &&
1559 (ret
& BDRV_BLOCK_OFFSET_VALID
)) {
1560 BlockDriverState
*file2
;
1563 ret2
= bdrv_co_get_block_status(bs
->file
->bs
, ret
>> BDRV_SECTOR_BITS
,
1564 *pnum
, &file_pnum
, &file2
);
1566 /* Ignore errors. This is just providing extra information, it
1567 * is useful but not necessary.
1570 /* !file_pnum indicates an offset at or beyond the EOF; it is
1571 * perfectly valid for the format block driver to point to such
1572 * offsets, so catch it and mark everything as zero */
1573 ret
|= BDRV_BLOCK_ZERO
;
1575 /* Limit request to the range reported by the protocol driver */
1577 ret
|= (ret2
& BDRV_BLOCK_ZERO
);
1585 static int64_t coroutine_fn
bdrv_co_get_block_status_above(BlockDriverState
*bs
,
1586 BlockDriverState
*base
,
1590 BlockDriverState
**file
)
1592 BlockDriverState
*p
;
1596 for (p
= bs
; p
!= base
; p
= backing_bs(p
)) {
1597 ret
= bdrv_co_get_block_status(p
, sector_num
, nb_sectors
, pnum
, file
);
1598 if (ret
< 0 || ret
& BDRV_BLOCK_ALLOCATED
) {
1601 /* [sector_num, pnum] unallocated on this layer, which could be only
1602 * the first part of [sector_num, nb_sectors]. */
1603 nb_sectors
= MIN(nb_sectors
, *pnum
);
1608 /* Coroutine wrapper for bdrv_get_block_status_above() */
1609 static void coroutine_fn
bdrv_get_block_status_above_co_entry(void *opaque
)
1611 BdrvCoGetBlockStatusData
*data
= opaque
;
1613 data
->ret
= bdrv_co_get_block_status_above(data
->bs
, data
->base
,
1622 * Synchronous wrapper around bdrv_co_get_block_status_above().
1624 * See bdrv_co_get_block_status_above() for details.
1626 int64_t bdrv_get_block_status_above(BlockDriverState
*bs
,
1627 BlockDriverState
*base
,
1629 int nb_sectors
, int *pnum
,
1630 BlockDriverState
**file
)
1633 BdrvCoGetBlockStatusData data
= {
1637 .sector_num
= sector_num
,
1638 .nb_sectors
= nb_sectors
,
1643 if (qemu_in_coroutine()) {
1644 /* Fast-path if already in coroutine context */
1645 bdrv_get_block_status_above_co_entry(&data
);
1647 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
1649 co
= qemu_coroutine_create(bdrv_get_block_status_above_co_entry
);
1650 qemu_coroutine_enter(co
, &data
);
1651 while (!data
.done
) {
1652 aio_poll(aio_context
, true);
1658 int64_t bdrv_get_block_status(BlockDriverState
*bs
,
1660 int nb_sectors
, int *pnum
,
1661 BlockDriverState
**file
)
1663 return bdrv_get_block_status_above(bs
, backing_bs(bs
),
1664 sector_num
, nb_sectors
, pnum
, file
);
1667 int coroutine_fn
bdrv_is_allocated(BlockDriverState
*bs
, int64_t sector_num
,
1668 int nb_sectors
, int *pnum
)
1670 BlockDriverState
*file
;
1671 int64_t ret
= bdrv_get_block_status(bs
, sector_num
, nb_sectors
, pnum
,
1676 return !!(ret
& BDRV_BLOCK_ALLOCATED
);
1680 * Given an image chain: ... -> [BASE] -> [INTER1] -> [INTER2] -> [TOP]
1682 * Return true if the given sector is allocated in any image between
1683 * BASE and TOP (inclusive). BASE can be NULL to check if the given
1684 * sector is allocated in any image of the chain. Return false otherwise.
1686 * 'pnum' is set to the number of sectors (including and immediately following
1687 * the specified sector) that are known to be in the same
1688 * allocated/unallocated state.
1691 int bdrv_is_allocated_above(BlockDriverState
*top
,
1692 BlockDriverState
*base
,
1694 int nb_sectors
, int *pnum
)
1696 BlockDriverState
*intermediate
;
1697 int ret
, n
= nb_sectors
;
1700 while (intermediate
&& intermediate
!= base
) {
1702 ret
= bdrv_is_allocated(intermediate
, sector_num
, nb_sectors
,
1712 * [sector_num, nb_sectors] is unallocated on top but intermediate
1715 * [sector_num+x, nr_sectors] allocated.
1717 if (n
> pnum_inter
&&
1718 (intermediate
== top
||
1719 sector_num
+ pnum_inter
< intermediate
->total_sectors
)) {
1723 intermediate
= backing_bs(intermediate
);
1730 int bdrv_write_compressed(BlockDriverState
*bs
, int64_t sector_num
,
1731 const uint8_t *buf
, int nb_sectors
)
1733 BlockDriver
*drv
= bs
->drv
;
1739 if (!drv
->bdrv_write_compressed
) {
1742 ret
= bdrv_check_request(bs
, sector_num
, nb_sectors
);
1747 assert(QLIST_EMPTY(&bs
->dirty_bitmaps
));
1749 return drv
->bdrv_write_compressed(bs
, sector_num
, buf
, nb_sectors
);
1752 int bdrv_save_vmstate(BlockDriverState
*bs
, const uint8_t *buf
,
1753 int64_t pos
, int size
)
1756 struct iovec iov
= {
1757 .iov_base
= (void *) buf
,
1761 qemu_iovec_init_external(&qiov
, &iov
, 1);
1762 return bdrv_writev_vmstate(bs
, &qiov
, pos
);
1765 int bdrv_writev_vmstate(BlockDriverState
*bs
, QEMUIOVector
*qiov
, int64_t pos
)
1767 BlockDriver
*drv
= bs
->drv
;
1771 } else if (drv
->bdrv_save_vmstate
) {
1772 return drv
->bdrv_save_vmstate(bs
, qiov
, pos
);
1773 } else if (bs
->file
) {
1774 return bdrv_writev_vmstate(bs
->file
->bs
, qiov
, pos
);
1780 int bdrv_load_vmstate(BlockDriverState
*bs
, uint8_t *buf
,
1781 int64_t pos
, int size
)
1783 BlockDriver
*drv
= bs
->drv
;
1786 if (drv
->bdrv_load_vmstate
)
1787 return drv
->bdrv_load_vmstate(bs
, buf
, pos
, size
);
1789 return bdrv_load_vmstate(bs
->file
->bs
, buf
, pos
, size
);
1793 /**************************************************************/
1796 BlockAIOCB
*bdrv_aio_readv(BlockDriverState
*bs
, int64_t sector_num
,
1797 QEMUIOVector
*qiov
, int nb_sectors
,
1798 BlockCompletionFunc
*cb
, void *opaque
)
1800 trace_bdrv_aio_readv(bs
, sector_num
, nb_sectors
, opaque
);
1802 return bdrv_co_aio_rw_vector(bs
, sector_num
, qiov
, nb_sectors
, 0,
1806 BlockAIOCB
*bdrv_aio_writev(BlockDriverState
*bs
, int64_t sector_num
,
1807 QEMUIOVector
*qiov
, int nb_sectors
,
1808 BlockCompletionFunc
*cb
, void *opaque
)
1810 trace_bdrv_aio_writev(bs
, sector_num
, nb_sectors
, opaque
);
1812 return bdrv_co_aio_rw_vector(bs
, sector_num
, qiov
, nb_sectors
, 0,
1816 BlockAIOCB
*bdrv_aio_write_zeroes(BlockDriverState
*bs
,
1817 int64_t sector_num
, int nb_sectors
, BdrvRequestFlags flags
,
1818 BlockCompletionFunc
*cb
, void *opaque
)
1820 trace_bdrv_aio_write_zeroes(bs
, sector_num
, nb_sectors
, flags
, opaque
);
1822 return bdrv_co_aio_rw_vector(bs
, sector_num
, NULL
, nb_sectors
,
1823 BDRV_REQ_ZERO_WRITE
| flags
,
1828 typedef struct MultiwriteCB
{
1833 BlockCompletionFunc
*cb
;
1835 QEMUIOVector
*free_qiov
;
1839 static void multiwrite_user_cb(MultiwriteCB
*mcb
)
1843 for (i
= 0; i
< mcb
->num_callbacks
; i
++) {
1844 mcb
->callbacks
[i
].cb(mcb
->callbacks
[i
].opaque
, mcb
->error
);
1845 if (mcb
->callbacks
[i
].free_qiov
) {
1846 qemu_iovec_destroy(mcb
->callbacks
[i
].free_qiov
);
1848 g_free(mcb
->callbacks
[i
].free_qiov
);
1852 static void multiwrite_cb(void *opaque
, int ret
)
1854 MultiwriteCB
*mcb
= opaque
;
1856 trace_multiwrite_cb(mcb
, ret
);
1858 if (ret
< 0 && !mcb
->error
) {
1862 mcb
->num_requests
--;
1863 if (mcb
->num_requests
== 0) {
1864 multiwrite_user_cb(mcb
);
1869 static int multiwrite_req_compare(const void *a
, const void *b
)
1871 const BlockRequest
*req1
= a
, *req2
= b
;
1874 * Note that we can't simply subtract req2->sector from req1->sector
1875 * here as that could overflow the return value.
1877 if (req1
->sector
> req2
->sector
) {
1879 } else if (req1
->sector
< req2
->sector
) {
1887 * Takes a bunch of requests and tries to merge them. Returns the number of
1888 * requests that remain after merging.
1890 static int multiwrite_merge(BlockDriverState
*bs
, BlockRequest
*reqs
,
1891 int num_reqs
, MultiwriteCB
*mcb
)
1895 // Sort requests by start sector
1896 qsort(reqs
, num_reqs
, sizeof(*reqs
), &multiwrite_req_compare
);
1898 // Check if adjacent requests touch the same clusters. If so, combine them,
1899 // filling up gaps with zero sectors.
1901 for (i
= 1; i
< num_reqs
; i
++) {
1903 int64_t oldreq_last
= reqs
[outidx
].sector
+ reqs
[outidx
].nb_sectors
;
1905 // Handle exactly sequential writes and overlapping writes.
1906 if (reqs
[i
].sector
<= oldreq_last
) {
1910 if (reqs
[outidx
].qiov
->niov
+ reqs
[i
].qiov
->niov
+ 1 >
1915 if (bs
->bl
.max_transfer_length
&& reqs
[outidx
].nb_sectors
+
1916 reqs
[i
].nb_sectors
> bs
->bl
.max_transfer_length
) {
1922 QEMUIOVector
*qiov
= g_malloc0(sizeof(*qiov
));
1923 qemu_iovec_init(qiov
,
1924 reqs
[outidx
].qiov
->niov
+ reqs
[i
].qiov
->niov
+ 1);
1926 // Add the first request to the merged one. If the requests are
1927 // overlapping, drop the last sectors of the first request.
1928 size
= (reqs
[i
].sector
- reqs
[outidx
].sector
) << 9;
1929 qemu_iovec_concat(qiov
, reqs
[outidx
].qiov
, 0, size
);
1931 // We should need to add any zeros between the two requests
1932 assert (reqs
[i
].sector
<= oldreq_last
);
1934 // Add the second request
1935 qemu_iovec_concat(qiov
, reqs
[i
].qiov
, 0, reqs
[i
].qiov
->size
);
1937 // Add tail of first request, if necessary
1938 if (qiov
->size
< reqs
[outidx
].qiov
->size
) {
1939 qemu_iovec_concat(qiov
, reqs
[outidx
].qiov
, qiov
->size
,
1940 reqs
[outidx
].qiov
->size
- qiov
->size
);
1943 reqs
[outidx
].nb_sectors
= qiov
->size
>> 9;
1944 reqs
[outidx
].qiov
= qiov
;
1946 mcb
->callbacks
[i
].free_qiov
= reqs
[outidx
].qiov
;
1949 reqs
[outidx
].sector
= reqs
[i
].sector
;
1950 reqs
[outidx
].nb_sectors
= reqs
[i
].nb_sectors
;
1951 reqs
[outidx
].qiov
= reqs
[i
].qiov
;
1956 block_acct_merge_done(blk_get_stats(bs
->blk
), BLOCK_ACCT_WRITE
,
1957 num_reqs
- outidx
- 1);
1964 * Submit multiple AIO write requests at once.
1966 * On success, the function returns 0 and all requests in the reqs array have
1967 * been submitted. In error case this function returns -1, and any of the
1968 * requests may or may not be submitted yet. In particular, this means that the
1969 * callback will be called for some of the requests, for others it won't. The
1970 * caller must check the error field of the BlockRequest to wait for the right
1971 * callbacks (if error != 0, no callback will be called).
1973 * The implementation may modify the contents of the reqs array, e.g. to merge
1974 * requests. However, the fields opaque and error are left unmodified as they
1975 * are used to signal failure for a single request to the caller.
1977 int bdrv_aio_multiwrite(BlockDriverState
*bs
, BlockRequest
*reqs
, int num_reqs
)
1982 /* don't submit writes if we don't have a medium */
1983 if (bs
->drv
== NULL
) {
1984 for (i
= 0; i
< num_reqs
; i
++) {
1985 reqs
[i
].error
= -ENOMEDIUM
;
1990 if (num_reqs
== 0) {
1994 // Create MultiwriteCB structure
1995 mcb
= g_malloc0(sizeof(*mcb
) + num_reqs
* sizeof(*mcb
->callbacks
));
1996 mcb
->num_requests
= 0;
1997 mcb
->num_callbacks
= num_reqs
;
1999 for (i
= 0; i
< num_reqs
; i
++) {
2000 mcb
->callbacks
[i
].cb
= reqs
[i
].cb
;
2001 mcb
->callbacks
[i
].opaque
= reqs
[i
].opaque
;
2004 // Check for mergable requests
2005 num_reqs
= multiwrite_merge(bs
, reqs
, num_reqs
, mcb
);
2007 trace_bdrv_aio_multiwrite(mcb
, mcb
->num_callbacks
, num_reqs
);
2009 /* Run the aio requests. */
2010 mcb
->num_requests
= num_reqs
;
2011 for (i
= 0; i
< num_reqs
; i
++) {
2012 bdrv_co_aio_rw_vector(bs
, reqs
[i
].sector
, reqs
[i
].qiov
,
2013 reqs
[i
].nb_sectors
, reqs
[i
].flags
,
2021 void bdrv_aio_cancel(BlockAIOCB
*acb
)
2024 bdrv_aio_cancel_async(acb
);
2025 while (acb
->refcnt
> 1) {
2026 if (acb
->aiocb_info
->get_aio_context
) {
2027 aio_poll(acb
->aiocb_info
->get_aio_context(acb
), true);
2028 } else if (acb
->bs
) {
2029 aio_poll(bdrv_get_aio_context(acb
->bs
), true);
2034 qemu_aio_unref(acb
);
2037 /* Async version of aio cancel. The caller is not blocked if the acb implements
2038 * cancel_async, otherwise we do nothing and let the request normally complete.
2039 * In either case the completion callback must be called. */
2040 void bdrv_aio_cancel_async(BlockAIOCB
*acb
)
2042 if (acb
->aiocb_info
->cancel_async
) {
2043 acb
->aiocb_info
->cancel_async(acb
);
2047 /**************************************************************/
2048 /* async block device emulation */
2050 typedef struct BlockAIOCBSync
{
2054 /* vector translation state */
2060 static const AIOCBInfo bdrv_em_aiocb_info
= {
2061 .aiocb_size
= sizeof(BlockAIOCBSync
),
2064 static void bdrv_aio_bh_cb(void *opaque
)
2066 BlockAIOCBSync
*acb
= opaque
;
2068 if (!acb
->is_write
&& acb
->ret
>= 0) {
2069 qemu_iovec_from_buf(acb
->qiov
, 0, acb
->bounce
, acb
->qiov
->size
);
2071 qemu_vfree(acb
->bounce
);
2072 acb
->common
.cb(acb
->common
.opaque
, acb
->ret
);
2073 qemu_bh_delete(acb
->bh
);
2075 qemu_aio_unref(acb
);
2078 static BlockAIOCB
*bdrv_aio_rw_vector(BlockDriverState
*bs
,
2082 BlockCompletionFunc
*cb
,
2087 BlockAIOCBSync
*acb
;
2089 acb
= qemu_aio_get(&bdrv_em_aiocb_info
, bs
, cb
, opaque
);
2090 acb
->is_write
= is_write
;
2092 acb
->bounce
= qemu_try_blockalign(bs
, qiov
->size
);
2093 acb
->bh
= aio_bh_new(bdrv_get_aio_context(bs
), bdrv_aio_bh_cb
, acb
);
2095 if (acb
->bounce
== NULL
) {
2097 } else if (is_write
) {
2098 qemu_iovec_to_buf(acb
->qiov
, 0, acb
->bounce
, qiov
->size
);
2099 acb
->ret
= bs
->drv
->bdrv_write(bs
, sector_num
, acb
->bounce
, nb_sectors
);
2101 acb
->ret
= bs
->drv
->bdrv_read(bs
, sector_num
, acb
->bounce
, nb_sectors
);
2104 qemu_bh_schedule(acb
->bh
);
2106 return &acb
->common
;
2109 static BlockAIOCB
*bdrv_aio_readv_em(BlockDriverState
*bs
,
2110 int64_t sector_num
, QEMUIOVector
*qiov
, int nb_sectors
,
2111 BlockCompletionFunc
*cb
, void *opaque
)
2113 return bdrv_aio_rw_vector(bs
, sector_num
, qiov
, nb_sectors
, cb
, opaque
, 0);
2116 static BlockAIOCB
*bdrv_aio_writev_em(BlockDriverState
*bs
,
2117 int64_t sector_num
, QEMUIOVector
*qiov
, int nb_sectors
,
2118 BlockCompletionFunc
*cb
, void *opaque
)
2120 return bdrv_aio_rw_vector(bs
, sector_num
, qiov
, nb_sectors
, cb
, opaque
, 1);
2124 typedef struct BlockAIOCBCoroutine
{
2131 } BlockAIOCBCoroutine
;
2133 static const AIOCBInfo bdrv_em_co_aiocb_info
= {
2134 .aiocb_size
= sizeof(BlockAIOCBCoroutine
),
2137 static void bdrv_co_complete(BlockAIOCBCoroutine
*acb
)
2139 if (!acb
->need_bh
) {
2140 acb
->common
.cb(acb
->common
.opaque
, acb
->req
.error
);
2141 qemu_aio_unref(acb
);
2145 static void bdrv_co_em_bh(void *opaque
)
2147 BlockAIOCBCoroutine
*acb
= opaque
;
2149 assert(!acb
->need_bh
);
2150 qemu_bh_delete(acb
->bh
);
2151 bdrv_co_complete(acb
);
2154 static void bdrv_co_maybe_schedule_bh(BlockAIOCBCoroutine
*acb
)
2156 acb
->need_bh
= false;
2157 if (acb
->req
.error
!= -EINPROGRESS
) {
2158 BlockDriverState
*bs
= acb
->common
.bs
;
2160 acb
->bh
= aio_bh_new(bdrv_get_aio_context(bs
), bdrv_co_em_bh
, acb
);
2161 qemu_bh_schedule(acb
->bh
);
2165 /* Invoke bdrv_co_do_readv/bdrv_co_do_writev */
2166 static void coroutine_fn
bdrv_co_do_rw(void *opaque
)
2168 BlockAIOCBCoroutine
*acb
= opaque
;
2169 BlockDriverState
*bs
= acb
->common
.bs
;
2171 if (!acb
->is_write
) {
2172 acb
->req
.error
= bdrv_co_do_readv(bs
, acb
->req
.sector
,
2173 acb
->req
.nb_sectors
, acb
->req
.qiov
, acb
->req
.flags
);
2175 acb
->req
.error
= bdrv_co_do_writev(bs
, acb
->req
.sector
,
2176 acb
->req
.nb_sectors
, acb
->req
.qiov
, acb
->req
.flags
);
2179 bdrv_co_complete(acb
);
2182 static BlockAIOCB
*bdrv_co_aio_rw_vector(BlockDriverState
*bs
,
2186 BdrvRequestFlags flags
,
2187 BlockCompletionFunc
*cb
,
2192 BlockAIOCBCoroutine
*acb
;
2194 acb
= qemu_aio_get(&bdrv_em_co_aiocb_info
, bs
, cb
, opaque
);
2195 acb
->need_bh
= true;
2196 acb
->req
.error
= -EINPROGRESS
;
2197 acb
->req
.sector
= sector_num
;
2198 acb
->req
.nb_sectors
= nb_sectors
;
2199 acb
->req
.qiov
= qiov
;
2200 acb
->req
.flags
= flags
;
2201 acb
->is_write
= is_write
;
2203 co
= qemu_coroutine_create(bdrv_co_do_rw
);
2204 qemu_coroutine_enter(co
, acb
);
2206 bdrv_co_maybe_schedule_bh(acb
);
2207 return &acb
->common
;
2210 static void coroutine_fn
bdrv_aio_flush_co_entry(void *opaque
)
2212 BlockAIOCBCoroutine
*acb
= opaque
;
2213 BlockDriverState
*bs
= acb
->common
.bs
;
2215 acb
->req
.error
= bdrv_co_flush(bs
);
2216 bdrv_co_complete(acb
);
2219 BlockAIOCB
*bdrv_aio_flush(BlockDriverState
*bs
,
2220 BlockCompletionFunc
*cb
, void *opaque
)
2222 trace_bdrv_aio_flush(bs
, opaque
);
2225 BlockAIOCBCoroutine
*acb
;
2227 acb
= qemu_aio_get(&bdrv_em_co_aiocb_info
, bs
, cb
, opaque
);
2228 acb
->need_bh
= true;
2229 acb
->req
.error
= -EINPROGRESS
;
2231 co
= qemu_coroutine_create(bdrv_aio_flush_co_entry
);
2232 qemu_coroutine_enter(co
, acb
);
2234 bdrv_co_maybe_schedule_bh(acb
);
2235 return &acb
->common
;
2238 static void coroutine_fn
bdrv_aio_discard_co_entry(void *opaque
)
2240 BlockAIOCBCoroutine
*acb
= opaque
;
2241 BlockDriverState
*bs
= acb
->common
.bs
;
2243 acb
->req
.error
= bdrv_co_discard(bs
, acb
->req
.sector
, acb
->req
.nb_sectors
);
2244 bdrv_co_complete(acb
);
2247 BlockAIOCB
*bdrv_aio_discard(BlockDriverState
*bs
,
2248 int64_t sector_num
, int nb_sectors
,
2249 BlockCompletionFunc
*cb
, void *opaque
)
2252 BlockAIOCBCoroutine
*acb
;
2254 trace_bdrv_aio_discard(bs
, sector_num
, nb_sectors
, opaque
);
2256 acb
= qemu_aio_get(&bdrv_em_co_aiocb_info
, bs
, cb
, opaque
);
2257 acb
->need_bh
= true;
2258 acb
->req
.error
= -EINPROGRESS
;
2259 acb
->req
.sector
= sector_num
;
2260 acb
->req
.nb_sectors
= nb_sectors
;
2261 co
= qemu_coroutine_create(bdrv_aio_discard_co_entry
);
2262 qemu_coroutine_enter(co
, acb
);
2264 bdrv_co_maybe_schedule_bh(acb
);
2265 return &acb
->common
;
2268 void *qemu_aio_get(const AIOCBInfo
*aiocb_info
, BlockDriverState
*bs
,
2269 BlockCompletionFunc
*cb
, void *opaque
)
2273 acb
= g_malloc(aiocb_info
->aiocb_size
);
2274 acb
->aiocb_info
= aiocb_info
;
2277 acb
->opaque
= opaque
;
2282 void qemu_aio_ref(void *p
)
2284 BlockAIOCB
*acb
= p
;
2288 void qemu_aio_unref(void *p
)
2290 BlockAIOCB
*acb
= p
;
2291 assert(acb
->refcnt
> 0);
2292 if (--acb
->refcnt
== 0) {
2297 /**************************************************************/
2298 /* Coroutine block device emulation */
2300 typedef struct CoroutineIOCompletion
{
2301 Coroutine
*coroutine
;
2303 } CoroutineIOCompletion
;
2305 static void bdrv_co_io_em_complete(void *opaque
, int ret
)
2307 CoroutineIOCompletion
*co
= opaque
;
2310 qemu_coroutine_enter(co
->coroutine
, NULL
);
2313 static int coroutine_fn
bdrv_co_io_em(BlockDriverState
*bs
, int64_t sector_num
,
2314 int nb_sectors
, QEMUIOVector
*iov
,
2317 CoroutineIOCompletion co
= {
2318 .coroutine
= qemu_coroutine_self(),
2323 acb
= bs
->drv
->bdrv_aio_writev(bs
, sector_num
, iov
, nb_sectors
,
2324 bdrv_co_io_em_complete
, &co
);
2326 acb
= bs
->drv
->bdrv_aio_readv(bs
, sector_num
, iov
, nb_sectors
,
2327 bdrv_co_io_em_complete
, &co
);
2330 trace_bdrv_co_io_em(bs
, sector_num
, nb_sectors
, is_write
, acb
);
2334 qemu_coroutine_yield();
2339 static int coroutine_fn
bdrv_co_readv_em(BlockDriverState
*bs
,
2340 int64_t sector_num
, int nb_sectors
,
2343 return bdrv_co_io_em(bs
, sector_num
, nb_sectors
, iov
, false);
2346 static int coroutine_fn
bdrv_co_writev_em(BlockDriverState
*bs
,
2347 int64_t sector_num
, int nb_sectors
,
2350 return bdrv_co_io_em(bs
, sector_num
, nb_sectors
, iov
, true);
2353 static void coroutine_fn
bdrv_flush_co_entry(void *opaque
)
2355 RwCo
*rwco
= opaque
;
2357 rwco
->ret
= bdrv_co_flush(rwco
->bs
);
2360 int coroutine_fn
bdrv_co_flush(BlockDriverState
*bs
)
2363 BdrvTrackedRequest req
;
2365 if (!bs
|| !bdrv_is_inserted(bs
) || bdrv_is_read_only(bs
) ||
2370 tracked_request_begin(&req
, bs
, 0, 0, BDRV_TRACKED_FLUSH
);
2371 /* Write back cached data to the OS even with cache=unsafe */
2372 BLKDBG_EVENT(bs
->file
, BLKDBG_FLUSH_TO_OS
);
2373 if (bs
->drv
->bdrv_co_flush_to_os
) {
2374 ret
= bs
->drv
->bdrv_co_flush_to_os(bs
);
2380 /* But don't actually force it to the disk with cache=unsafe */
2381 if (bs
->open_flags
& BDRV_O_NO_FLUSH
) {
2385 BLKDBG_EVENT(bs
->file
, BLKDBG_FLUSH_TO_DISK
);
2386 if (bs
->drv
->bdrv_co_flush_to_disk
) {
2387 ret
= bs
->drv
->bdrv_co_flush_to_disk(bs
);
2388 } else if (bs
->drv
->bdrv_aio_flush
) {
2390 CoroutineIOCompletion co
= {
2391 .coroutine
= qemu_coroutine_self(),
2394 acb
= bs
->drv
->bdrv_aio_flush(bs
, bdrv_co_io_em_complete
, &co
);
2398 qemu_coroutine_yield();
2403 * Some block drivers always operate in either writethrough or unsafe
2404 * mode and don't support bdrv_flush therefore. Usually qemu doesn't
2405 * know how the server works (because the behaviour is hardcoded or
2406 * depends on server-side configuration), so we can't ensure that
2407 * everything is safe on disk. Returning an error doesn't work because
2408 * that would break guests even if the server operates in writethrough
2411 * Let's hope the user knows what he's doing.
2419 /* Now flush the underlying protocol. It will also have BDRV_O_NO_FLUSH
2420 * in the case of cache=unsafe, so there are no useless flushes.
2423 ret
= bs
->file
? bdrv_co_flush(bs
->file
->bs
) : 0;
2425 tracked_request_end(&req
);
2429 int bdrv_flush(BlockDriverState
*bs
)
2437 if (qemu_in_coroutine()) {
2438 /* Fast-path if already in coroutine context */
2439 bdrv_flush_co_entry(&rwco
);
2441 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
2443 co
= qemu_coroutine_create(bdrv_flush_co_entry
);
2444 qemu_coroutine_enter(co
, &rwco
);
2445 while (rwco
.ret
== NOT_DONE
) {
2446 aio_poll(aio_context
, true);
2453 typedef struct DiscardCo
{
2454 BlockDriverState
*bs
;
2459 static void coroutine_fn
bdrv_discard_co_entry(void *opaque
)
2461 DiscardCo
*rwco
= opaque
;
2463 rwco
->ret
= bdrv_co_discard(rwco
->bs
, rwco
->sector_num
, rwco
->nb_sectors
);
2466 int coroutine_fn
bdrv_co_discard(BlockDriverState
*bs
, int64_t sector_num
,
2469 BdrvTrackedRequest req
;
2470 int max_discard
, ret
;
2476 ret
= bdrv_check_request(bs
, sector_num
, nb_sectors
);
2479 } else if (bs
->read_only
) {
2482 assert(!(bs
->open_flags
& BDRV_O_INACTIVE
));
2484 /* Do nothing if disabled. */
2485 if (!(bs
->open_flags
& BDRV_O_UNMAP
)) {
2489 if (!bs
->drv
->bdrv_co_discard
&& !bs
->drv
->bdrv_aio_discard
) {
2493 tracked_request_begin(&req
, bs
, sector_num
, nb_sectors
,
2494 BDRV_TRACKED_DISCARD
);
2495 bdrv_set_dirty(bs
, sector_num
, nb_sectors
);
2497 max_discard
= MIN_NON_ZERO(bs
->bl
.max_discard
, BDRV_REQUEST_MAX_SECTORS
);
2498 while (nb_sectors
> 0) {
2500 int num
= nb_sectors
;
2503 if (bs
->bl
.discard_alignment
&&
2504 num
>= bs
->bl
.discard_alignment
&&
2505 sector_num
% bs
->bl
.discard_alignment
) {
2506 if (num
> bs
->bl
.discard_alignment
) {
2507 num
= bs
->bl
.discard_alignment
;
2509 num
-= sector_num
% bs
->bl
.discard_alignment
;
2512 /* limit request size */
2513 if (num
> max_discard
) {
2517 if (bs
->drv
->bdrv_co_discard
) {
2518 ret
= bs
->drv
->bdrv_co_discard(bs
, sector_num
, num
);
2521 CoroutineIOCompletion co
= {
2522 .coroutine
= qemu_coroutine_self(),
2525 acb
= bs
->drv
->bdrv_aio_discard(bs
, sector_num
, nb_sectors
,
2526 bdrv_co_io_em_complete
, &co
);
2531 qemu_coroutine_yield();
2535 if (ret
&& ret
!= -ENOTSUP
) {
2544 tracked_request_end(&req
);
2548 int bdrv_discard(BlockDriverState
*bs
, int64_t sector_num
, int nb_sectors
)
2553 .sector_num
= sector_num
,
2554 .nb_sectors
= nb_sectors
,
2558 if (qemu_in_coroutine()) {
2559 /* Fast-path if already in coroutine context */
2560 bdrv_discard_co_entry(&rwco
);
2562 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
2564 co
= qemu_coroutine_create(bdrv_discard_co_entry
);
2565 qemu_coroutine_enter(co
, &rwco
);
2566 while (rwco
.ret
== NOT_DONE
) {
2567 aio_poll(aio_context
, true);
2575 CoroutineIOCompletion
*co
;
2577 } BdrvIoctlCompletionData
;
2579 static void bdrv_ioctl_bh_cb(void *opaque
)
2581 BdrvIoctlCompletionData
*data
= opaque
;
2583 bdrv_co_io_em_complete(data
->co
, -ENOTSUP
);
2584 qemu_bh_delete(data
->bh
);
2587 static int bdrv_co_do_ioctl(BlockDriverState
*bs
, int req
, void *buf
)
2589 BlockDriver
*drv
= bs
->drv
;
2590 BdrvTrackedRequest tracked_req
;
2591 CoroutineIOCompletion co
= {
2592 .coroutine
= qemu_coroutine_self(),
2596 tracked_request_begin(&tracked_req
, bs
, 0, 0, BDRV_TRACKED_IOCTL
);
2597 if (!drv
|| !drv
->bdrv_aio_ioctl
) {
2602 acb
= drv
->bdrv_aio_ioctl(bs
, req
, buf
, bdrv_co_io_em_complete
, &co
);
2604 BdrvIoctlCompletionData
*data
= g_new(BdrvIoctlCompletionData
, 1);
2605 data
->bh
= aio_bh_new(bdrv_get_aio_context(bs
),
2606 bdrv_ioctl_bh_cb
, data
);
2608 qemu_bh_schedule(data
->bh
);
2610 qemu_coroutine_yield();
2612 tracked_request_end(&tracked_req
);
2617 BlockDriverState
*bs
;
2623 static void coroutine_fn
bdrv_co_ioctl_entry(void *opaque
)
2625 BdrvIoctlCoData
*data
= opaque
;
2626 data
->ret
= bdrv_co_do_ioctl(data
->bs
, data
->req
, data
->buf
);
2629 /* needed for generic scsi interface */
2630 int bdrv_ioctl(BlockDriverState
*bs
, unsigned long int req
, void *buf
)
2632 BdrvIoctlCoData data
= {
2636 .ret
= -EINPROGRESS
,
2639 if (qemu_in_coroutine()) {
2640 /* Fast-path if already in coroutine context */
2641 bdrv_co_ioctl_entry(&data
);
2643 Coroutine
*co
= qemu_coroutine_create(bdrv_co_ioctl_entry
);
2645 qemu_coroutine_enter(co
, &data
);
2646 while (data
.ret
== -EINPROGRESS
) {
2647 aio_poll(bdrv_get_aio_context(bs
), true);
2653 static void coroutine_fn
bdrv_co_aio_ioctl_entry(void *opaque
)
2655 BlockAIOCBCoroutine
*acb
= opaque
;
2656 acb
->req
.error
= bdrv_co_do_ioctl(acb
->common
.bs
,
2657 acb
->req
.req
, acb
->req
.buf
);
2658 bdrv_co_complete(acb
);
2661 BlockAIOCB
*bdrv_aio_ioctl(BlockDriverState
*bs
,
2662 unsigned long int req
, void *buf
,
2663 BlockCompletionFunc
*cb
, void *opaque
)
2665 BlockAIOCBCoroutine
*acb
= qemu_aio_get(&bdrv_em_co_aiocb_info
,
2669 acb
->need_bh
= true;
2670 acb
->req
.error
= -EINPROGRESS
;
2673 co
= qemu_coroutine_create(bdrv_co_aio_ioctl_entry
);
2674 qemu_coroutine_enter(co
, acb
);
2676 bdrv_co_maybe_schedule_bh(acb
);
2677 return &acb
->common
;
2680 void *qemu_blockalign(BlockDriverState
*bs
, size_t size
)
2682 return qemu_memalign(bdrv_opt_mem_align(bs
), size
);
2685 void *qemu_blockalign0(BlockDriverState
*bs
, size_t size
)
2687 return memset(qemu_blockalign(bs
, size
), 0, size
);
2690 void *qemu_try_blockalign(BlockDriverState
*bs
, size_t size
)
2692 size_t align
= bdrv_opt_mem_align(bs
);
2694 /* Ensure that NULL is never returned on success */
2700 return qemu_try_memalign(align
, size
);
2703 void *qemu_try_blockalign0(BlockDriverState
*bs
, size_t size
)
2705 void *mem
= qemu_try_blockalign(bs
, size
);
2708 memset(mem
, 0, size
);
2715 * Check if all memory in this vector is sector aligned.
2717 bool bdrv_qiov_is_aligned(BlockDriverState
*bs
, QEMUIOVector
*qiov
)
2720 size_t alignment
= bdrv_min_mem_align(bs
);
2722 for (i
= 0; i
< qiov
->niov
; i
++) {
2723 if ((uintptr_t) qiov
->iov
[i
].iov_base
% alignment
) {
2726 if (qiov
->iov
[i
].iov_len
% alignment
) {
2734 void bdrv_add_before_write_notifier(BlockDriverState
*bs
,
2735 NotifierWithReturn
*notifier
)
2737 notifier_with_return_list_add(&bs
->before_write_notifiers
, notifier
);
2740 void bdrv_io_plug(BlockDriverState
*bs
)
2742 BlockDriver
*drv
= bs
->drv
;
2743 if (drv
&& drv
->bdrv_io_plug
) {
2744 drv
->bdrv_io_plug(bs
);
2745 } else if (bs
->file
) {
2746 bdrv_io_plug(bs
->file
->bs
);
2750 void bdrv_io_unplug(BlockDriverState
*bs
)
2752 BlockDriver
*drv
= bs
->drv
;
2753 if (drv
&& drv
->bdrv_io_unplug
) {
2754 drv
->bdrv_io_unplug(bs
);
2755 } else if (bs
->file
) {
2756 bdrv_io_unplug(bs
->file
->bs
);
2760 void bdrv_flush_io_queue(BlockDriverState
*bs
)
2762 BlockDriver
*drv
= bs
->drv
;
2763 if (drv
&& drv
->bdrv_flush_io_queue
) {
2764 drv
->bdrv_flush_io_queue(bs
);
2765 } else if (bs
->file
) {
2766 bdrv_flush_io_queue(bs
->file
->bs
);
2768 bdrv_start_throttled_reqs(bs
);
2771 void bdrv_drained_begin(BlockDriverState
*bs
)
2773 if (!bs
->quiesce_counter
++) {
2774 aio_disable_external(bdrv_get_aio_context(bs
));
2779 void bdrv_drained_end(BlockDriverState
*bs
)
2781 assert(bs
->quiesce_counter
> 0);
2782 if (--bs
->quiesce_counter
> 0) {
2785 aio_enable_external(bdrv_get_aio_context(bs
));