2 * Block layer I/O functions
4 * Copyright (c) 2003 Fabrice Bellard
6 * Permission is hereby granted, free of charge, to any person obtaining a copy
7 * of this software and associated documentation files (the "Software"), to deal
8 * in the Software without restriction, including without limitation the rights
9 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10 * copies of the Software, and to permit persons to whom the Software is
11 * furnished to do so, subject to the following conditions:
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
25 #include "qemu/osdep.h"
27 #include "sysemu/block-backend.h"
28 #include "block/blockjob.h"
29 #include "block/block_int.h"
30 #include "qemu/cutils.h"
31 #include "qapi/error.h"
32 #include "qemu/error-report.h"
34 #define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */
36 static BlockAIOCB *bdrv_co_aio_prw_vector(BdrvChild *child,
39 BdrvRequestFlags flags,
40 BlockCompletionFunc *cb,
43 static void coroutine_fn bdrv_co_do_rw(void *opaque);
44 static int coroutine_fn bdrv_co_do_pwrite_zeroes(BlockDriverState *bs,
45 int64_t offset, int count, BdrvRequestFlags flags);
47 static void bdrv_parent_drained_begin(BlockDriverState *bs)
51 QLIST_FOREACH(c, &bs->parents, next_parent) {
52 if (c->role->drained_begin) {
53 c->role->drained_begin(c);
58 static void bdrv_parent_drained_end(BlockDriverState *bs)
62 QLIST_FOREACH(c, &bs->parents, next_parent) {
63 if (c->role->drained_end) {
64 c->role->drained_end(c);
69 static void bdrv_merge_limits(BlockLimits *dst, const BlockLimits *src)
71 dst->opt_transfer = MAX(dst->opt_transfer, src->opt_transfer);
72 dst->max_transfer = MIN_NON_ZERO(dst->max_transfer, src->max_transfer);
73 dst->opt_mem_alignment = MAX(dst->opt_mem_alignment,
74 src->opt_mem_alignment);
75 dst->min_mem_alignment = MAX(dst->min_mem_alignment,
76 src->min_mem_alignment);
77 dst->max_iov = MIN_NON_ZERO(dst->max_iov, src->max_iov);
80 void bdrv_refresh_limits(BlockDriverState *bs, Error **errp)
82 BlockDriver *drv = bs->drv;
83 Error *local_err = NULL;
85 memset(&bs->bl, 0, sizeof(bs->bl));
91 /* Default alignment based on whether driver has byte interface */
92 bs->bl.request_alignment = drv->bdrv_co_preadv ? 1 : 512;
94 /* Take some limits from the children as a default */
96 bdrv_refresh_limits(bs->file->bs, &local_err);
98 error_propagate(errp, local_err);
101 bdrv_merge_limits(&bs->bl, &bs->file->bs->bl);
103 bs->bl.min_mem_alignment = 512;
104 bs->bl.opt_mem_alignment = getpagesize();
106 /* Safe default since most protocols use readv()/writev()/etc */
107 bs->bl.max_iov = IOV_MAX;
111 bdrv_refresh_limits(bs->backing->bs, &local_err);
113 error_propagate(errp, local_err);
116 bdrv_merge_limits(&bs->bl, &bs->backing->bs->bl);
119 /* Then let the driver override it */
120 if (drv->bdrv_refresh_limits) {
121 drv->bdrv_refresh_limits(bs, errp);
126 * The copy-on-read flag is actually a reference count so multiple users may
127 * use the feature without worrying about clobbering its previous state.
128 * Copy-on-read stays enabled until all users have called to disable it.
130 void bdrv_enable_copy_on_read(BlockDriverState *bs)
135 void bdrv_disable_copy_on_read(BlockDriverState *bs)
137 assert(bs->copy_on_read > 0);
141 /* Check if any requests are in-flight (including throttled requests) */
142 bool bdrv_requests_pending(BlockDriverState *bs)
146 if (atomic_read(&bs->in_flight)) {
150 QLIST_FOREACH(child, &bs->children, next) {
151 if (bdrv_requests_pending(child->bs)) {
159 static bool bdrv_drain_recurse(BlockDriverState *bs)
164 waited = BDRV_POLL_WHILE(bs, atomic_read(&bs->in_flight) > 0);
166 if (bs->drv && bs->drv->bdrv_drain) {
167 bs->drv->bdrv_drain(bs);
170 QLIST_FOREACH(child, &bs->children, next) {
171 waited |= bdrv_drain_recurse(child->bs);
179 BlockDriverState *bs;
183 static void bdrv_co_drain_bh_cb(void *opaque)
185 BdrvCoDrainData *data = opaque;
186 Coroutine *co = data->co;
187 BlockDriverState *bs = data->bs;
189 bdrv_dec_in_flight(bs);
190 bdrv_drained_begin(bs);
192 qemu_coroutine_enter(co);
195 static void coroutine_fn bdrv_co_yield_to_drain(BlockDriverState *bs)
197 BdrvCoDrainData data;
199 /* Calling bdrv_drain() from a BH ensures the current coroutine yields and
200 * other coroutines run if they were queued from
201 * qemu_co_queue_run_restart(). */
203 assert(qemu_in_coroutine());
204 data = (BdrvCoDrainData) {
205 .co = qemu_coroutine_self(),
209 bdrv_inc_in_flight(bs);
210 aio_bh_schedule_oneshot(bdrv_get_aio_context(bs),
211 bdrv_co_drain_bh_cb, &data);
213 qemu_coroutine_yield();
214 /* If we are resumed from some other event (such as an aio completion or a
215 * timer callback), it is a bug in the caller that should be fixed. */
219 void bdrv_drained_begin(BlockDriverState *bs)
221 if (qemu_in_coroutine()) {
222 bdrv_co_yield_to_drain(bs);
226 if (!bs->quiesce_counter++) {
227 aio_disable_external(bdrv_get_aio_context(bs));
228 bdrv_parent_drained_begin(bs);
231 bdrv_io_unplugged_begin(bs);
232 bdrv_drain_recurse(bs);
233 bdrv_io_unplugged_end(bs);
236 void bdrv_drained_end(BlockDriverState *bs)
238 assert(bs->quiesce_counter > 0);
239 if (--bs->quiesce_counter > 0) {
243 bdrv_parent_drained_end(bs);
244 aio_enable_external(bdrv_get_aio_context(bs));
248 * Wait for pending requests to complete on a single BlockDriverState subtree,
249 * and suspend block driver's internal I/O until next request arrives.
251 * Note that unlike bdrv_drain_all(), the caller must hold the BlockDriverState
254 * Only this BlockDriverState's AioContext is run, so in-flight requests must
255 * not depend on events in other AioContexts. In that case, use
256 * bdrv_drain_all() instead.
258 void coroutine_fn bdrv_co_drain(BlockDriverState *bs)
260 assert(qemu_in_coroutine());
261 bdrv_drained_begin(bs);
262 bdrv_drained_end(bs);
265 void bdrv_drain(BlockDriverState *bs)
267 bdrv_drained_begin(bs);
268 bdrv_drained_end(bs);
272 * Wait for pending requests to complete across all BlockDriverStates
274 * This function does not flush data to disk, use bdrv_flush_all() for that
275 * after calling this function.
277 * This pauses all block jobs and disables external clients. It must
278 * be paired with bdrv_drain_all_end().
280 * NOTE: no new block jobs or BlockDriverStates can be created between
281 * the bdrv_drain_all_begin() and bdrv_drain_all_end() calls.
283 void bdrv_drain_all_begin(void)
285 /* Always run first iteration so any pending completion BHs run */
287 BlockDriverState *bs;
289 BlockJob *job = NULL;
290 GSList *aio_ctxs = NULL, *ctx;
292 while ((job = block_job_next(job))) {
293 AioContext *aio_context = blk_get_aio_context(job->blk);
295 aio_context_acquire(aio_context);
296 block_job_pause(job);
297 aio_context_release(aio_context);
300 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
301 AioContext *aio_context = bdrv_get_aio_context(bs);
303 aio_context_acquire(aio_context);
304 bdrv_parent_drained_begin(bs);
305 bdrv_io_unplugged_begin(bs);
306 aio_disable_external(aio_context);
307 aio_context_release(aio_context);
309 if (!g_slist_find(aio_ctxs, aio_context)) {
310 aio_ctxs = g_slist_prepend(aio_ctxs, aio_context);
314 /* Note that completion of an asynchronous I/O operation can trigger any
315 * number of other I/O operations on other devices---for example a
316 * coroutine can submit an I/O request to another device in response to
317 * request completion. Therefore we must keep looping until there was no
318 * more activity rather than simply draining each device independently.
323 for (ctx = aio_ctxs; ctx != NULL; ctx = ctx->next) {
324 AioContext *aio_context = ctx->data;
326 aio_context_acquire(aio_context);
327 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
328 if (aio_context == bdrv_get_aio_context(bs)) {
329 waited |= bdrv_drain_recurse(bs);
332 aio_context_release(aio_context);
336 g_slist_free(aio_ctxs);
339 void bdrv_drain_all_end(void)
341 BlockDriverState *bs;
343 BlockJob *job = NULL;
345 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
346 AioContext *aio_context = bdrv_get_aio_context(bs);
348 aio_context_acquire(aio_context);
349 aio_enable_external(aio_context);
350 bdrv_io_unplugged_end(bs);
351 bdrv_parent_drained_end(bs);
352 aio_context_release(aio_context);
355 while ((job = block_job_next(job))) {
356 AioContext *aio_context = blk_get_aio_context(job->blk);
358 aio_context_acquire(aio_context);
359 block_job_resume(job);
360 aio_context_release(aio_context);
364 void bdrv_drain_all(void)
366 bdrv_drain_all_begin();
367 bdrv_drain_all_end();
371 * Remove an active request from the tracked requests list
373 * This function should be called when a tracked request is completing.
375 static void tracked_request_end(BdrvTrackedRequest *req)
377 if (req->serialising) {
378 req->bs->serialising_in_flight--;
381 QLIST_REMOVE(req, list);
382 qemu_co_queue_restart_all(&req->wait_queue);
386 * Add an active request to the tracked requests list
388 static void tracked_request_begin(BdrvTrackedRequest *req,
389 BlockDriverState *bs,
392 enum BdrvTrackedRequestType type)
394 *req = (BdrvTrackedRequest){
399 .co = qemu_coroutine_self(),
400 .serialising = false,
401 .overlap_offset = offset,
402 .overlap_bytes = bytes,
405 qemu_co_queue_init(&req->wait_queue);
407 QLIST_INSERT_HEAD(&bs->tracked_requests, req, list);
410 static void mark_request_serialising(BdrvTrackedRequest *req, uint64_t align)
412 int64_t overlap_offset = req->offset & ~(align - 1);
413 unsigned int overlap_bytes = ROUND_UP(req->offset + req->bytes, align)
416 if (!req->serialising) {
417 req->bs->serialising_in_flight++;
418 req->serialising = true;
421 req->overlap_offset = MIN(req->overlap_offset, overlap_offset);
422 req->overlap_bytes = MAX(req->overlap_bytes, overlap_bytes);
426 * Round a region to cluster boundaries (sector-based)
428 void bdrv_round_sectors_to_clusters(BlockDriverState *bs,
429 int64_t sector_num, int nb_sectors,
430 int64_t *cluster_sector_num,
431 int *cluster_nb_sectors)
435 if (bdrv_get_info(bs, &bdi) < 0 || bdi.cluster_size == 0) {
436 *cluster_sector_num = sector_num;
437 *cluster_nb_sectors = nb_sectors;
439 int64_t c = bdi.cluster_size / BDRV_SECTOR_SIZE;
440 *cluster_sector_num = QEMU_ALIGN_DOWN(sector_num, c);
441 *cluster_nb_sectors = QEMU_ALIGN_UP(sector_num - *cluster_sector_num +
447 * Round a region to cluster boundaries
449 void bdrv_round_to_clusters(BlockDriverState *bs,
450 int64_t offset, unsigned int bytes,
451 int64_t *cluster_offset,
452 unsigned int *cluster_bytes)
456 if (bdrv_get_info(bs, &bdi) < 0 || bdi.cluster_size == 0) {
457 *cluster_offset = offset;
458 *cluster_bytes = bytes;
460 int64_t c = bdi.cluster_size;
461 *cluster_offset = QEMU_ALIGN_DOWN(offset, c);
462 *cluster_bytes = QEMU_ALIGN_UP(offset - *cluster_offset + bytes, c);
466 static int bdrv_get_cluster_size(BlockDriverState *bs)
471 ret = bdrv_get_info(bs, &bdi);
472 if (ret < 0 || bdi.cluster_size == 0) {
473 return bs->bl.request_alignment;
475 return bdi.cluster_size;
479 static bool tracked_request_overlaps(BdrvTrackedRequest *req,
480 int64_t offset, unsigned int bytes)
483 if (offset >= req->overlap_offset + req->overlap_bytes) {
487 if (req->overlap_offset >= offset + bytes) {
493 void bdrv_inc_in_flight(BlockDriverState *bs)
495 atomic_inc(&bs->in_flight);
498 static void dummy_bh_cb(void *opaque)
502 void bdrv_wakeup(BlockDriverState *bs)
505 aio_bh_schedule_oneshot(qemu_get_aio_context(), dummy_bh_cb, NULL);
509 void bdrv_dec_in_flight(BlockDriverState *bs)
511 atomic_dec(&bs->in_flight);
515 static bool coroutine_fn wait_serialising_requests(BdrvTrackedRequest *self)
517 BlockDriverState *bs = self->bs;
518 BdrvTrackedRequest *req;
522 if (!bs->serialising_in_flight) {
528 QLIST_FOREACH(req, &bs->tracked_requests, list) {
529 if (req == self || (!req->serialising && !self->serialising)) {
532 if (tracked_request_overlaps(req, self->overlap_offset,
533 self->overlap_bytes))
535 /* Hitting this means there was a reentrant request, for
536 * example, a block driver issuing nested requests. This must
537 * never happen since it means deadlock.
539 assert(qemu_coroutine_self() != req->co);
541 /* If the request is already (indirectly) waiting for us, or
542 * will wait for us as soon as it wakes up, then just go on
543 * (instead of producing a deadlock in the former case). */
544 if (!req->waiting_for) {
545 self->waiting_for = req;
546 qemu_co_queue_wait(&req->wait_queue);
547 self->waiting_for = NULL;
559 static int bdrv_check_byte_request(BlockDriverState *bs, int64_t offset,
562 if (size > BDRV_REQUEST_MAX_SECTORS << BDRV_SECTOR_BITS) {
566 if (!bdrv_is_inserted(bs)) {
577 typedef struct RwCo {
583 BdrvRequestFlags flags;
586 static void coroutine_fn bdrv_rw_co_entry(void *opaque)
590 if (!rwco->is_write) {
591 rwco->ret = bdrv_co_preadv(rwco->child, rwco->offset,
592 rwco->qiov->size, rwco->qiov,
595 rwco->ret = bdrv_co_pwritev(rwco->child, rwco->offset,
596 rwco->qiov->size, rwco->qiov,
602 * Process a vectored synchronous request using coroutines
604 static int bdrv_prwv_co(BdrvChild *child, int64_t offset,
605 QEMUIOVector *qiov, bool is_write,
606 BdrvRequestFlags flags)
613 .is_write = is_write,
618 if (qemu_in_coroutine()) {
619 /* Fast-path if already in coroutine context */
620 bdrv_rw_co_entry(&rwco);
622 co = qemu_coroutine_create(bdrv_rw_co_entry, &rwco);
623 qemu_coroutine_enter(co);
624 BDRV_POLL_WHILE(child->bs, rwco.ret == NOT_DONE);
630 * Process a synchronous request using coroutines
632 static int bdrv_rw_co(BdrvChild *child, int64_t sector_num, uint8_t *buf,
633 int nb_sectors, bool is_write, BdrvRequestFlags flags)
637 .iov_base = (void *)buf,
638 .iov_len = nb_sectors * BDRV_SECTOR_SIZE,
641 if (nb_sectors < 0 || nb_sectors > BDRV_REQUEST_MAX_SECTORS) {
645 qemu_iovec_init_external(&qiov, &iov, 1);
646 return bdrv_prwv_co(child, sector_num << BDRV_SECTOR_BITS,
647 &qiov, is_write, flags);
650 /* return < 0 if error. See bdrv_write() for the return codes */
651 int bdrv_read(BdrvChild *child, int64_t sector_num,
652 uint8_t *buf, int nb_sectors)
654 return bdrv_rw_co(child, sector_num, buf, nb_sectors, false, 0);
657 /* Return < 0 if error. Important errors are:
658 -EIO generic I/O error (may happen for all errors)
659 -ENOMEDIUM No media inserted.
660 -EINVAL Invalid sector number or nb_sectors
661 -EACCES Trying to write a read-only device
663 int bdrv_write(BdrvChild *child, int64_t sector_num,
664 const uint8_t *buf, int nb_sectors)
666 return bdrv_rw_co(child, sector_num, (uint8_t *)buf, nb_sectors, true, 0);
669 int bdrv_pwrite_zeroes(BdrvChild *child, int64_t offset,
670 int count, BdrvRequestFlags flags)
678 qemu_iovec_init_external(&qiov, &iov, 1);
679 return bdrv_prwv_co(child, offset, &qiov, true,
680 BDRV_REQ_ZERO_WRITE | flags);
684 * Completely zero out a block device with the help of bdrv_pwrite_zeroes.
685 * The operation is sped up by checking the block status and only writing
686 * zeroes to the device if they currently do not return zeroes. Optional
687 * flags are passed through to bdrv_pwrite_zeroes (e.g. BDRV_REQ_MAY_UNMAP,
690 * Returns < 0 on error, 0 on success. For error codes see bdrv_write().
692 int bdrv_make_zero(BdrvChild *child, BdrvRequestFlags flags)
694 int64_t target_sectors, ret, nb_sectors, sector_num = 0;
695 BlockDriverState *bs = child->bs;
696 BlockDriverState *file;
699 target_sectors = bdrv_nb_sectors(bs);
700 if (target_sectors < 0) {
701 return target_sectors;
705 nb_sectors = MIN(target_sectors - sector_num, BDRV_REQUEST_MAX_SECTORS);
706 if (nb_sectors <= 0) {
709 ret = bdrv_get_block_status(bs, sector_num, nb_sectors, &n, &file);
711 error_report("error getting block status at sector %" PRId64 ": %s",
712 sector_num, strerror(-ret));
715 if (ret & BDRV_BLOCK_ZERO) {
719 ret = bdrv_pwrite_zeroes(child, sector_num << BDRV_SECTOR_BITS,
720 n << BDRV_SECTOR_BITS, flags);
722 error_report("error writing zeroes at sector %" PRId64 ": %s",
723 sector_num, strerror(-ret));
730 int bdrv_preadv(BdrvChild *child, int64_t offset, QEMUIOVector *qiov)
734 ret = bdrv_prwv_co(child, offset, qiov, false, 0);
742 int bdrv_pread(BdrvChild *child, int64_t offset, void *buf, int bytes)
746 .iov_base = (void *)buf,
754 qemu_iovec_init_external(&qiov, &iov, 1);
755 return bdrv_preadv(child, offset, &qiov);
758 int bdrv_pwritev(BdrvChild *child, int64_t offset, QEMUIOVector *qiov)
762 ret = bdrv_prwv_co(child, offset, qiov, true, 0);
770 int bdrv_pwrite(BdrvChild *child, int64_t offset, const void *buf, int bytes)
774 .iov_base = (void *) buf,
782 qemu_iovec_init_external(&qiov, &iov, 1);
783 return bdrv_pwritev(child, offset, &qiov);
787 * Writes to the file and ensures that no writes are reordered across this
788 * request (acts as a barrier)
790 * Returns 0 on success, -errno in error cases.
792 int bdrv_pwrite_sync(BdrvChild *child, int64_t offset,
793 const void *buf, int count)
797 ret = bdrv_pwrite(child, offset, buf, count);
802 ret = bdrv_flush(child->bs);
810 typedef struct CoroutineIOCompletion {
811 Coroutine *coroutine;
813 } CoroutineIOCompletion;
815 static void bdrv_co_io_em_complete(void *opaque, int ret)
817 CoroutineIOCompletion *co = opaque;
820 qemu_coroutine_enter(co->coroutine);
823 static int coroutine_fn bdrv_driver_preadv(BlockDriverState *bs,
824 uint64_t offset, uint64_t bytes,
825 QEMUIOVector *qiov, int flags)
827 BlockDriver *drv = bs->drv;
829 unsigned int nb_sectors;
831 assert(!(flags & ~BDRV_REQ_MASK));
833 if (drv->bdrv_co_preadv) {
834 return drv->bdrv_co_preadv(bs, offset, bytes, qiov, flags);
837 sector_num = offset >> BDRV_SECTOR_BITS;
838 nb_sectors = bytes >> BDRV_SECTOR_BITS;
840 assert((offset & (BDRV_SECTOR_SIZE - 1)) == 0);
841 assert((bytes & (BDRV_SECTOR_SIZE - 1)) == 0);
842 assert((bytes >> BDRV_SECTOR_BITS) <= BDRV_REQUEST_MAX_SECTORS);
844 if (drv->bdrv_co_readv) {
845 return drv->bdrv_co_readv(bs, sector_num, nb_sectors, qiov);
848 CoroutineIOCompletion co = {
849 .coroutine = qemu_coroutine_self(),
852 acb = bs->drv->bdrv_aio_readv(bs, sector_num, qiov, nb_sectors,
853 bdrv_co_io_em_complete, &co);
857 qemu_coroutine_yield();
863 static int coroutine_fn bdrv_driver_pwritev(BlockDriverState *bs,
864 uint64_t offset, uint64_t bytes,
865 QEMUIOVector *qiov, int flags)
867 BlockDriver *drv = bs->drv;
869 unsigned int nb_sectors;
872 assert(!(flags & ~BDRV_REQ_MASK));
874 if (drv->bdrv_co_pwritev) {
875 ret = drv->bdrv_co_pwritev(bs, offset, bytes, qiov,
876 flags & bs->supported_write_flags);
877 flags &= ~bs->supported_write_flags;
881 sector_num = offset >> BDRV_SECTOR_BITS;
882 nb_sectors = bytes >> BDRV_SECTOR_BITS;
884 assert((offset & (BDRV_SECTOR_SIZE - 1)) == 0);
885 assert((bytes & (BDRV_SECTOR_SIZE - 1)) == 0);
886 assert((bytes >> BDRV_SECTOR_BITS) <= BDRV_REQUEST_MAX_SECTORS);
888 if (drv->bdrv_co_writev_flags) {
889 ret = drv->bdrv_co_writev_flags(bs, sector_num, nb_sectors, qiov,
890 flags & bs->supported_write_flags);
891 flags &= ~bs->supported_write_flags;
892 } else if (drv->bdrv_co_writev) {
893 assert(!bs->supported_write_flags);
894 ret = drv->bdrv_co_writev(bs, sector_num, nb_sectors, qiov);
897 CoroutineIOCompletion co = {
898 .coroutine = qemu_coroutine_self(),
901 acb = bs->drv->bdrv_aio_writev(bs, sector_num, qiov, nb_sectors,
902 bdrv_co_io_em_complete, &co);
906 qemu_coroutine_yield();
912 if (ret == 0 && (flags & BDRV_REQ_FUA)) {
913 ret = bdrv_co_flush(bs);
919 static int coroutine_fn
920 bdrv_driver_pwritev_compressed(BlockDriverState *bs, uint64_t offset,
921 uint64_t bytes, QEMUIOVector *qiov)
923 BlockDriver *drv = bs->drv;
925 if (!drv->bdrv_co_pwritev_compressed) {
929 return drv->bdrv_co_pwritev_compressed(bs, offset, bytes, qiov);
932 static int coroutine_fn bdrv_co_do_copy_on_readv(BlockDriverState *bs,
933 int64_t offset, unsigned int bytes, QEMUIOVector *qiov)
935 /* Perform I/O through a temporary buffer so that users who scribble over
936 * their read buffer while the operation is in progress do not end up
937 * modifying the image file. This is critical for zero-copy guest I/O
938 * where anything might happen inside guest memory.
942 BlockDriver *drv = bs->drv;
944 QEMUIOVector bounce_qiov;
945 int64_t cluster_offset;
946 unsigned int cluster_bytes;
950 /* Cover entire cluster so no additional backing file I/O is required when
951 * allocating cluster in the image file.
953 bdrv_round_to_clusters(bs, offset, bytes, &cluster_offset, &cluster_bytes);
955 trace_bdrv_co_do_copy_on_readv(bs, offset, bytes,
956 cluster_offset, cluster_bytes);
958 iov.iov_len = cluster_bytes;
959 iov.iov_base = bounce_buffer = qemu_try_blockalign(bs, iov.iov_len);
960 if (bounce_buffer == NULL) {
965 qemu_iovec_init_external(&bounce_qiov, &iov, 1);
967 ret = bdrv_driver_preadv(bs, cluster_offset, cluster_bytes,
973 if (drv->bdrv_co_pwrite_zeroes &&
974 buffer_is_zero(bounce_buffer, iov.iov_len)) {
975 /* FIXME: Should we (perhaps conditionally) be setting
976 * BDRV_REQ_MAY_UNMAP, if it will allow for a sparser copy
977 * that still correctly reads as zero? */
978 ret = bdrv_co_do_pwrite_zeroes(bs, cluster_offset, cluster_bytes, 0);
980 /* This does not change the data on the disk, it is not necessary
981 * to flush even in cache=writethrough mode.
983 ret = bdrv_driver_pwritev(bs, cluster_offset, cluster_bytes,
988 /* It might be okay to ignore write errors for guest requests. If this
989 * is a deliberate copy-on-read then we don't want to ignore the error.
990 * Simply report it in all cases.
995 skip_bytes = offset - cluster_offset;
996 qemu_iovec_from_buf(qiov, 0, bounce_buffer + skip_bytes, bytes);
999 qemu_vfree(bounce_buffer);
1004 * Forwards an already correctly aligned request to the BlockDriver. This
1005 * handles copy on read, zeroing after EOF, and fragmentation of large
1006 * reads; any other features must be implemented by the caller.
1008 static int coroutine_fn bdrv_aligned_preadv(BlockDriverState *bs,
1009 BdrvTrackedRequest *req, int64_t offset, unsigned int bytes,
1010 int64_t align, QEMUIOVector *qiov, int flags)
1012 int64_t total_bytes, max_bytes;
1014 uint64_t bytes_remaining = bytes;
1017 assert(is_power_of_2(align));
1018 assert((offset & (align - 1)) == 0);
1019 assert((bytes & (align - 1)) == 0);
1020 assert(!qiov || bytes == qiov->size);
1021 assert((bs->open_flags & BDRV_O_NO_IO) == 0);
1022 max_transfer = QEMU_ALIGN_DOWN(MIN_NON_ZERO(bs->bl.max_transfer, INT_MAX),
1025 /* TODO: We would need a per-BDS .supported_read_flags and
1026 * potential fallback support, if we ever implement any read flags
1027 * to pass through to drivers. For now, there aren't any
1028 * passthrough flags. */
1029 assert(!(flags & ~(BDRV_REQ_NO_SERIALISING | BDRV_REQ_COPY_ON_READ)));
1031 /* Handle Copy on Read and associated serialisation */
1032 if (flags & BDRV_REQ_COPY_ON_READ) {
1033 /* If we touch the same cluster it counts as an overlap. This
1034 * guarantees that allocating writes will be serialized and not race
1035 * with each other for the same cluster. For example, in copy-on-read
1036 * it ensures that the CoR read and write operations are atomic and
1037 * guest writes cannot interleave between them. */
1038 mark_request_serialising(req, bdrv_get_cluster_size(bs));
1041 if (!(flags & BDRV_REQ_NO_SERIALISING)) {
1042 wait_serialising_requests(req);
1045 if (flags & BDRV_REQ_COPY_ON_READ) {
1046 int64_t start_sector = offset >> BDRV_SECTOR_BITS;
1047 int64_t end_sector = DIV_ROUND_UP(offset + bytes, BDRV_SECTOR_SIZE);
1048 unsigned int nb_sectors = end_sector - start_sector;
1051 ret = bdrv_is_allocated(bs, start_sector, nb_sectors, &pnum);
1056 if (!ret || pnum != nb_sectors) {
1057 ret = bdrv_co_do_copy_on_readv(bs, offset, bytes, qiov);
1062 /* Forward the request to the BlockDriver, possibly fragmenting it */
1063 total_bytes = bdrv_getlength(bs);
1064 if (total_bytes < 0) {
1069 max_bytes = ROUND_UP(MAX(0, total_bytes - offset), align);
1070 if (bytes <= max_bytes && bytes <= max_transfer) {
1071 ret = bdrv_driver_preadv(bs, offset, bytes, qiov, 0);
1075 while (bytes_remaining) {
1079 QEMUIOVector local_qiov;
1081 num = MIN(bytes_remaining, MIN(max_bytes, max_transfer));
1083 qemu_iovec_init(&local_qiov, qiov->niov);
1084 qemu_iovec_concat(&local_qiov, qiov, bytes - bytes_remaining, num);
1086 ret = bdrv_driver_preadv(bs, offset + bytes - bytes_remaining,
1087 num, &local_qiov, 0);
1089 qemu_iovec_destroy(&local_qiov);
1091 num = bytes_remaining;
1092 ret = qemu_iovec_memset(qiov, bytes - bytes_remaining, 0,
1098 bytes_remaining -= num;
1102 return ret < 0 ? ret : 0;
1106 * Handle a read request in coroutine context
1108 int coroutine_fn bdrv_co_preadv(BdrvChild *child,
1109 int64_t offset, unsigned int bytes, QEMUIOVector *qiov,
1110 BdrvRequestFlags flags)
1112 BlockDriverState *bs = child->bs;
1113 BlockDriver *drv = bs->drv;
1114 BdrvTrackedRequest req;
1116 uint64_t align = bs->bl.request_alignment;
1117 uint8_t *head_buf = NULL;
1118 uint8_t *tail_buf = NULL;
1119 QEMUIOVector local_qiov;
1120 bool use_local_qiov = false;
1127 ret = bdrv_check_byte_request(bs, offset, bytes);
1132 bdrv_inc_in_flight(bs);
1134 /* Don't do copy-on-read if we read data before write operation */
1135 if (bs->copy_on_read && !(flags & BDRV_REQ_NO_SERIALISING)) {
1136 flags |= BDRV_REQ_COPY_ON_READ;
1139 /* Align read if necessary by padding qiov */
1140 if (offset & (align - 1)) {
1141 head_buf = qemu_blockalign(bs, align);
1142 qemu_iovec_init(&local_qiov, qiov->niov + 2);
1143 qemu_iovec_add(&local_qiov, head_buf, offset & (align - 1));
1144 qemu_iovec_concat(&local_qiov, qiov, 0, qiov->size);
1145 use_local_qiov = true;
1147 bytes += offset & (align - 1);
1148 offset = offset & ~(align - 1);
1151 if ((offset + bytes) & (align - 1)) {
1152 if (!use_local_qiov) {
1153 qemu_iovec_init(&local_qiov, qiov->niov + 1);
1154 qemu_iovec_concat(&local_qiov, qiov, 0, qiov->size);
1155 use_local_qiov = true;
1157 tail_buf = qemu_blockalign(bs, align);
1158 qemu_iovec_add(&local_qiov, tail_buf,
1159 align - ((offset + bytes) & (align - 1)));
1161 bytes = ROUND_UP(bytes, align);
1164 tracked_request_begin(&req, bs, offset, bytes, BDRV_TRACKED_READ);
1165 ret = bdrv_aligned_preadv(bs, &req, offset, bytes, align,
1166 use_local_qiov ? &local_qiov : qiov,
1168 tracked_request_end(&req);
1169 bdrv_dec_in_flight(bs);
1171 if (use_local_qiov) {
1172 qemu_iovec_destroy(&local_qiov);
1173 qemu_vfree(head_buf);
1174 qemu_vfree(tail_buf);
1180 static int coroutine_fn bdrv_co_do_readv(BdrvChild *child,
1181 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov,
1182 BdrvRequestFlags flags)
1184 if (nb_sectors < 0 || nb_sectors > BDRV_REQUEST_MAX_SECTORS) {
1188 return bdrv_co_preadv(child, sector_num << BDRV_SECTOR_BITS,
1189 nb_sectors << BDRV_SECTOR_BITS, qiov, flags);
1192 int coroutine_fn bdrv_co_readv(BdrvChild *child, int64_t sector_num,
1193 int nb_sectors, QEMUIOVector *qiov)
1195 trace_bdrv_co_readv(child->bs, sector_num, nb_sectors);
1197 return bdrv_co_do_readv(child, sector_num, nb_sectors, qiov, 0);
1200 /* Maximum buffer for write zeroes fallback, in bytes */
1201 #define MAX_WRITE_ZEROES_BOUNCE_BUFFER (32768 << BDRV_SECTOR_BITS)
1203 static int coroutine_fn bdrv_co_do_pwrite_zeroes(BlockDriverState *bs,
1204 int64_t offset, int count, BdrvRequestFlags flags)
1206 BlockDriver *drv = bs->drv;
1208 struct iovec iov = {0};
1210 bool need_flush = false;
1214 int max_write_zeroes = MIN_NON_ZERO(bs->bl.max_pwrite_zeroes, INT_MAX);
1215 int alignment = MAX(bs->bl.pwrite_zeroes_alignment,
1216 bs->bl.request_alignment);
1218 assert(alignment % bs->bl.request_alignment == 0);
1219 head = offset % alignment;
1220 tail = (offset + count) % alignment;
1221 max_write_zeroes = QEMU_ALIGN_DOWN(max_write_zeroes, alignment);
1222 assert(max_write_zeroes >= bs->bl.request_alignment);
1224 while (count > 0 && !ret) {
1227 /* Align request. Block drivers can expect the "bulk" of the request
1228 * to be aligned, and that unaligned requests do not cross cluster
1232 /* Make a small request up to the first aligned sector. */
1233 num = MIN(count, alignment - head);
1235 } else if (tail && num > alignment) {
1236 /* Shorten the request to the last aligned sector. */
1240 /* limit request size */
1241 if (num > max_write_zeroes) {
1242 num = max_write_zeroes;
1246 /* First try the efficient write zeroes operation */
1247 if (drv->bdrv_co_pwrite_zeroes) {
1248 ret = drv->bdrv_co_pwrite_zeroes(bs, offset, num,
1249 flags & bs->supported_zero_flags);
1250 if (ret != -ENOTSUP && (flags & BDRV_REQ_FUA) &&
1251 !(bs->supported_zero_flags & BDRV_REQ_FUA)) {
1255 assert(!bs->supported_zero_flags);
1258 if (ret == -ENOTSUP) {
1259 /* Fall back to bounce buffer if write zeroes is unsupported */
1260 int max_transfer = MIN_NON_ZERO(bs->bl.max_transfer,
1261 MAX_WRITE_ZEROES_BOUNCE_BUFFER);
1262 BdrvRequestFlags write_flags = flags & ~BDRV_REQ_ZERO_WRITE;
1264 if ((flags & BDRV_REQ_FUA) &&
1265 !(bs->supported_write_flags & BDRV_REQ_FUA)) {
1266 /* No need for bdrv_driver_pwrite() to do a fallback
1267 * flush on each chunk; use just one at the end */
1268 write_flags &= ~BDRV_REQ_FUA;
1271 num = MIN(num, max_transfer);
1273 if (iov.iov_base == NULL) {
1274 iov.iov_base = qemu_try_blockalign(bs, num);
1275 if (iov.iov_base == NULL) {
1279 memset(iov.iov_base, 0, num);
1281 qemu_iovec_init_external(&qiov, &iov, 1);
1283 ret = bdrv_driver_pwritev(bs, offset, num, &qiov, write_flags);
1285 /* Keep bounce buffer around if it is big enough for all
1286 * all future requests.
1288 if (num < max_transfer) {
1289 qemu_vfree(iov.iov_base);
1290 iov.iov_base = NULL;
1299 if (ret == 0 && need_flush) {
1300 ret = bdrv_co_flush(bs);
1302 qemu_vfree(iov.iov_base);
1307 * Forwards an already correctly aligned write request to the BlockDriver,
1308 * after possibly fragmenting it.
1310 static int coroutine_fn bdrv_aligned_pwritev(BlockDriverState *bs,
1311 BdrvTrackedRequest *req, int64_t offset, unsigned int bytes,
1312 int64_t align, QEMUIOVector *qiov, int flags)
1314 BlockDriver *drv = bs->drv;
1318 int64_t start_sector = offset >> BDRV_SECTOR_BITS;
1319 int64_t end_sector = DIV_ROUND_UP(offset + bytes, BDRV_SECTOR_SIZE);
1320 uint64_t bytes_remaining = bytes;
1323 assert(is_power_of_2(align));
1324 assert((offset & (align - 1)) == 0);
1325 assert((bytes & (align - 1)) == 0);
1326 assert(!qiov || bytes == qiov->size);
1327 assert((bs->open_flags & BDRV_O_NO_IO) == 0);
1328 assert(!(flags & ~BDRV_REQ_MASK));
1329 max_transfer = QEMU_ALIGN_DOWN(MIN_NON_ZERO(bs->bl.max_transfer, INT_MAX),
1332 waited = wait_serialising_requests(req);
1333 assert(!waited || !req->serialising);
1334 assert(req->overlap_offset <= offset);
1335 assert(offset + bytes <= req->overlap_offset + req->overlap_bytes);
1337 ret = notifier_with_return_list_notify(&bs->before_write_notifiers, req);
1339 if (!ret && bs->detect_zeroes != BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF &&
1340 !(flags & BDRV_REQ_ZERO_WRITE) && drv->bdrv_co_pwrite_zeroes &&
1341 qemu_iovec_is_zero(qiov)) {
1342 flags |= BDRV_REQ_ZERO_WRITE;
1343 if (bs->detect_zeroes == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP) {
1344 flags |= BDRV_REQ_MAY_UNMAP;
1349 /* Do nothing, write notifier decided to fail this request */
1350 } else if (flags & BDRV_REQ_ZERO_WRITE) {
1351 bdrv_debug_event(bs, BLKDBG_PWRITEV_ZERO);
1352 ret = bdrv_co_do_pwrite_zeroes(bs, offset, bytes, flags);
1353 } else if (flags & BDRV_REQ_WRITE_COMPRESSED) {
1354 ret = bdrv_driver_pwritev_compressed(bs, offset, bytes, qiov);
1355 } else if (bytes <= max_transfer) {
1356 bdrv_debug_event(bs, BLKDBG_PWRITEV);
1357 ret = bdrv_driver_pwritev(bs, offset, bytes, qiov, flags);
1359 bdrv_debug_event(bs, BLKDBG_PWRITEV);
1360 while (bytes_remaining) {
1361 int num = MIN(bytes_remaining, max_transfer);
1362 QEMUIOVector local_qiov;
1363 int local_flags = flags;
1366 if (num < bytes_remaining && (flags & BDRV_REQ_FUA) &&
1367 !(bs->supported_write_flags & BDRV_REQ_FUA)) {
1368 /* If FUA is going to be emulated by flush, we only
1369 * need to flush on the last iteration */
1370 local_flags &= ~BDRV_REQ_FUA;
1372 qemu_iovec_init(&local_qiov, qiov->niov);
1373 qemu_iovec_concat(&local_qiov, qiov, bytes - bytes_remaining, num);
1375 ret = bdrv_driver_pwritev(bs, offset + bytes - bytes_remaining,
1376 num, &local_qiov, local_flags);
1377 qemu_iovec_destroy(&local_qiov);
1381 bytes_remaining -= num;
1384 bdrv_debug_event(bs, BLKDBG_PWRITEV_DONE);
1387 bdrv_set_dirty(bs, start_sector, end_sector - start_sector);
1389 if (bs->wr_highest_offset < offset + bytes) {
1390 bs->wr_highest_offset = offset + bytes;
1394 bs->total_sectors = MAX(bs->total_sectors, end_sector);
1401 static int coroutine_fn bdrv_co_do_zero_pwritev(BlockDriverState *bs,
1404 BdrvRequestFlags flags,
1405 BdrvTrackedRequest *req)
1407 uint8_t *buf = NULL;
1408 QEMUIOVector local_qiov;
1410 uint64_t align = bs->bl.request_alignment;
1411 unsigned int head_padding_bytes, tail_padding_bytes;
1414 head_padding_bytes = offset & (align - 1);
1415 tail_padding_bytes = align - ((offset + bytes) & (align - 1));
1418 assert(flags & BDRV_REQ_ZERO_WRITE);
1419 if (head_padding_bytes || tail_padding_bytes) {
1420 buf = qemu_blockalign(bs, align);
1421 iov = (struct iovec) {
1425 qemu_iovec_init_external(&local_qiov, &iov, 1);
1427 if (head_padding_bytes) {
1428 uint64_t zero_bytes = MIN(bytes, align - head_padding_bytes);
1430 /* RMW the unaligned part before head. */
1431 mark_request_serialising(req, align);
1432 wait_serialising_requests(req);
1433 bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_HEAD);
1434 ret = bdrv_aligned_preadv(bs, req, offset & ~(align - 1), align,
1435 align, &local_qiov, 0);
1439 bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_AFTER_HEAD);
1441 memset(buf + head_padding_bytes, 0, zero_bytes);
1442 ret = bdrv_aligned_pwritev(bs, req, offset & ~(align - 1), align,
1444 flags & ~BDRV_REQ_ZERO_WRITE);
1448 offset += zero_bytes;
1449 bytes -= zero_bytes;
1452 assert(!bytes || (offset & (align - 1)) == 0);
1453 if (bytes >= align) {
1454 /* Write the aligned part in the middle. */
1455 uint64_t aligned_bytes = bytes & ~(align - 1);
1456 ret = bdrv_aligned_pwritev(bs, req, offset, aligned_bytes, align,
1461 bytes -= aligned_bytes;
1462 offset += aligned_bytes;
1465 assert(!bytes || (offset & (align - 1)) == 0);
1467 assert(align == tail_padding_bytes + bytes);
1468 /* RMW the unaligned part after tail. */
1469 mark_request_serialising(req, align);
1470 wait_serialising_requests(req);
1471 bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_TAIL);
1472 ret = bdrv_aligned_preadv(bs, req, offset, align,
1473 align, &local_qiov, 0);
1477 bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_AFTER_TAIL);
1479 memset(buf, 0, bytes);
1480 ret = bdrv_aligned_pwritev(bs, req, offset, align, align,
1481 &local_qiov, flags & ~BDRV_REQ_ZERO_WRITE);
1490 * Handle a write request in coroutine context
1492 int coroutine_fn bdrv_co_pwritev(BdrvChild *child,
1493 int64_t offset, unsigned int bytes, QEMUIOVector *qiov,
1494 BdrvRequestFlags flags)
1496 BlockDriverState *bs = child->bs;
1497 BdrvTrackedRequest req;
1498 uint64_t align = bs->bl.request_alignment;
1499 uint8_t *head_buf = NULL;
1500 uint8_t *tail_buf = NULL;
1501 QEMUIOVector local_qiov;
1502 bool use_local_qiov = false;
1508 if (bs->read_only) {
1511 assert(!(bs->open_flags & BDRV_O_INACTIVE));
1513 ret = bdrv_check_byte_request(bs, offset, bytes);
1518 bdrv_inc_in_flight(bs);
1520 * Align write if necessary by performing a read-modify-write cycle.
1521 * Pad qiov with the read parts and be sure to have a tracked request not
1522 * only for bdrv_aligned_pwritev, but also for the reads of the RMW cycle.
1524 tracked_request_begin(&req, bs, offset, bytes, BDRV_TRACKED_WRITE);
1527 ret = bdrv_co_do_zero_pwritev(bs, offset, bytes, flags, &req);
1531 if (offset & (align - 1)) {
1532 QEMUIOVector head_qiov;
1533 struct iovec head_iov;
1535 mark_request_serialising(&req, align);
1536 wait_serialising_requests(&req);
1538 head_buf = qemu_blockalign(bs, align);
1539 head_iov = (struct iovec) {
1540 .iov_base = head_buf,
1543 qemu_iovec_init_external(&head_qiov, &head_iov, 1);
1545 bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_HEAD);
1546 ret = bdrv_aligned_preadv(bs, &req, offset & ~(align - 1), align,
1547 align, &head_qiov, 0);
1551 bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_AFTER_HEAD);
1553 qemu_iovec_init(&local_qiov, qiov->niov + 2);
1554 qemu_iovec_add(&local_qiov, head_buf, offset & (align - 1));
1555 qemu_iovec_concat(&local_qiov, qiov, 0, qiov->size);
1556 use_local_qiov = true;
1558 bytes += offset & (align - 1);
1559 offset = offset & ~(align - 1);
1561 /* We have read the tail already if the request is smaller
1562 * than one aligned block.
1564 if (bytes < align) {
1565 qemu_iovec_add(&local_qiov, head_buf + bytes, align - bytes);
1570 if ((offset + bytes) & (align - 1)) {
1571 QEMUIOVector tail_qiov;
1572 struct iovec tail_iov;
1576 mark_request_serialising(&req, align);
1577 waited = wait_serialising_requests(&req);
1578 assert(!waited || !use_local_qiov);
1580 tail_buf = qemu_blockalign(bs, align);
1581 tail_iov = (struct iovec) {
1582 .iov_base = tail_buf,
1585 qemu_iovec_init_external(&tail_qiov, &tail_iov, 1);
1587 bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_TAIL);
1588 ret = bdrv_aligned_preadv(bs, &req, (offset + bytes) & ~(align - 1), align,
1589 align, &tail_qiov, 0);
1593 bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_AFTER_TAIL);
1595 if (!use_local_qiov) {
1596 qemu_iovec_init(&local_qiov, qiov->niov + 1);
1597 qemu_iovec_concat(&local_qiov, qiov, 0, qiov->size);
1598 use_local_qiov = true;
1601 tail_bytes = (offset + bytes) & (align - 1);
1602 qemu_iovec_add(&local_qiov, tail_buf + tail_bytes, align - tail_bytes);
1604 bytes = ROUND_UP(bytes, align);
1607 ret = bdrv_aligned_pwritev(bs, &req, offset, bytes, align,
1608 use_local_qiov ? &local_qiov : qiov,
1613 if (use_local_qiov) {
1614 qemu_iovec_destroy(&local_qiov);
1616 qemu_vfree(head_buf);
1617 qemu_vfree(tail_buf);
1619 tracked_request_end(&req);
1620 bdrv_dec_in_flight(bs);
1624 static int coroutine_fn bdrv_co_do_writev(BdrvChild *child,
1625 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov,
1626 BdrvRequestFlags flags)
1628 if (nb_sectors < 0 || nb_sectors > BDRV_REQUEST_MAX_SECTORS) {
1632 return bdrv_co_pwritev(child, sector_num << BDRV_SECTOR_BITS,
1633 nb_sectors << BDRV_SECTOR_BITS, qiov, flags);
1636 int coroutine_fn bdrv_co_writev(BdrvChild *child, int64_t sector_num,
1637 int nb_sectors, QEMUIOVector *qiov)
1639 trace_bdrv_co_writev(child->bs, sector_num, nb_sectors);
1641 return bdrv_co_do_writev(child, sector_num, nb_sectors, qiov, 0);
1644 int coroutine_fn bdrv_co_pwrite_zeroes(BdrvChild *child, int64_t offset,
1645 int count, BdrvRequestFlags flags)
1647 trace_bdrv_co_pwrite_zeroes(child->bs, offset, count, flags);
1649 if (!(child->bs->open_flags & BDRV_O_UNMAP)) {
1650 flags &= ~BDRV_REQ_MAY_UNMAP;
1653 return bdrv_co_pwritev(child, offset, count, NULL,
1654 BDRV_REQ_ZERO_WRITE | flags);
1658 * Flush ALL BDSes regardless of if they are reachable via a BlkBackend or not.
1660 int bdrv_flush_all(void)
1662 BdrvNextIterator it;
1663 BlockDriverState *bs = NULL;
1666 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
1667 AioContext *aio_context = bdrv_get_aio_context(bs);
1670 aio_context_acquire(aio_context);
1671 ret = bdrv_flush(bs);
1672 if (ret < 0 && !result) {
1675 aio_context_release(aio_context);
1682 typedef struct BdrvCoGetBlockStatusData {
1683 BlockDriverState *bs;
1684 BlockDriverState *base;
1685 BlockDriverState **file;
1691 } BdrvCoGetBlockStatusData;
1694 * Returns the allocation status of the specified sectors.
1695 * Drivers not implementing the functionality are assumed to not support
1696 * backing files, hence all their sectors are reported as allocated.
1698 * If 'sector_num' is beyond the end of the disk image the return value is 0
1699 * and 'pnum' is set to 0.
1701 * 'pnum' is set to the number of sectors (including and immediately following
1702 * the specified sector) that are known to be in the same
1703 * allocated/unallocated state.
1705 * 'nb_sectors' is the max value 'pnum' should be set to. If nb_sectors goes
1706 * beyond the end of the disk image it will be clamped.
1708 * If returned value is positive and BDRV_BLOCK_OFFSET_VALID bit is set, 'file'
1709 * points to the BDS which the sector range is allocated in.
1711 static int64_t coroutine_fn bdrv_co_get_block_status(BlockDriverState *bs,
1713 int nb_sectors, int *pnum,
1714 BlockDriverState **file)
1716 int64_t total_sectors;
1720 total_sectors = bdrv_nb_sectors(bs);
1721 if (total_sectors < 0) {
1722 return total_sectors;
1725 if (sector_num >= total_sectors) {
1730 n = total_sectors - sector_num;
1731 if (n < nb_sectors) {
1735 if (!bs->drv->bdrv_co_get_block_status) {
1737 ret = BDRV_BLOCK_DATA | BDRV_BLOCK_ALLOCATED;
1738 if (bs->drv->protocol_name) {
1739 ret |= BDRV_BLOCK_OFFSET_VALID | (sector_num * BDRV_SECTOR_SIZE);
1745 bdrv_inc_in_flight(bs);
1746 ret = bs->drv->bdrv_co_get_block_status(bs, sector_num, nb_sectors, pnum,
1753 if (ret & BDRV_BLOCK_RAW) {
1754 assert(ret & BDRV_BLOCK_OFFSET_VALID);
1755 ret = bdrv_get_block_status(bs->file->bs, ret >> BDRV_SECTOR_BITS,
1760 if (ret & (BDRV_BLOCK_DATA | BDRV_BLOCK_ZERO)) {
1761 ret |= BDRV_BLOCK_ALLOCATED;
1763 if (bdrv_unallocated_blocks_are_zero(bs)) {
1764 ret |= BDRV_BLOCK_ZERO;
1765 } else if (bs->backing) {
1766 BlockDriverState *bs2 = bs->backing->bs;
1767 int64_t nb_sectors2 = bdrv_nb_sectors(bs2);
1768 if (nb_sectors2 >= 0 && sector_num >= nb_sectors2) {
1769 ret |= BDRV_BLOCK_ZERO;
1774 if (*file && *file != bs &&
1775 (ret & BDRV_BLOCK_DATA) && !(ret & BDRV_BLOCK_ZERO) &&
1776 (ret & BDRV_BLOCK_OFFSET_VALID)) {
1777 BlockDriverState *file2;
1780 ret2 = bdrv_co_get_block_status(*file, ret >> BDRV_SECTOR_BITS,
1781 *pnum, &file_pnum, &file2);
1783 /* Ignore errors. This is just providing extra information, it
1784 * is useful but not necessary.
1787 /* !file_pnum indicates an offset at or beyond the EOF; it is
1788 * perfectly valid for the format block driver to point to such
1789 * offsets, so catch it and mark everything as zero */
1790 ret |= BDRV_BLOCK_ZERO;
1792 /* Limit request to the range reported by the protocol driver */
1794 ret |= (ret2 & BDRV_BLOCK_ZERO);
1800 bdrv_dec_in_flight(bs);
1804 static int64_t coroutine_fn bdrv_co_get_block_status_above(BlockDriverState *bs,
1805 BlockDriverState *base,
1809 BlockDriverState **file)
1811 BlockDriverState *p;
1815 for (p = bs; p != base; p = backing_bs(p)) {
1816 ret = bdrv_co_get_block_status(p, sector_num, nb_sectors, pnum, file);
1817 if (ret < 0 || ret & BDRV_BLOCK_ALLOCATED) {
1820 /* [sector_num, pnum] unallocated on this layer, which could be only
1821 * the first part of [sector_num, nb_sectors]. */
1822 nb_sectors = MIN(nb_sectors, *pnum);
1827 /* Coroutine wrapper for bdrv_get_block_status_above() */
1828 static void coroutine_fn bdrv_get_block_status_above_co_entry(void *opaque)
1830 BdrvCoGetBlockStatusData *data = opaque;
1832 data->ret = bdrv_co_get_block_status_above(data->bs, data->base,
1841 * Synchronous wrapper around bdrv_co_get_block_status_above().
1843 * See bdrv_co_get_block_status_above() for details.
1845 int64_t bdrv_get_block_status_above(BlockDriverState *bs,
1846 BlockDriverState *base,
1848 int nb_sectors, int *pnum,
1849 BlockDriverState **file)
1852 BdrvCoGetBlockStatusData data = {
1856 .sector_num = sector_num,
1857 .nb_sectors = nb_sectors,
1862 if (qemu_in_coroutine()) {
1863 /* Fast-path if already in coroutine context */
1864 bdrv_get_block_status_above_co_entry(&data);
1866 co = qemu_coroutine_create(bdrv_get_block_status_above_co_entry,
1868 qemu_coroutine_enter(co);
1869 BDRV_POLL_WHILE(bs, !data.done);
1874 int64_t bdrv_get_block_status(BlockDriverState *bs,
1876 int nb_sectors, int *pnum,
1877 BlockDriverState **file)
1879 return bdrv_get_block_status_above(bs, backing_bs(bs),
1880 sector_num, nb_sectors, pnum, file);
1883 int coroutine_fn bdrv_is_allocated(BlockDriverState *bs, int64_t sector_num,
1884 int nb_sectors, int *pnum)
1886 BlockDriverState *file;
1887 int64_t ret = bdrv_get_block_status(bs, sector_num, nb_sectors, pnum,
1892 return !!(ret & BDRV_BLOCK_ALLOCATED);
1896 * Given an image chain: ... -> [BASE] -> [INTER1] -> [INTER2] -> [TOP]
1898 * Return true if the given sector is allocated in any image between
1899 * BASE and TOP (inclusive). BASE can be NULL to check if the given
1900 * sector is allocated in any image of the chain. Return false otherwise.
1902 * 'pnum' is set to the number of sectors (including and immediately following
1903 * the specified sector) that are known to be in the same
1904 * allocated/unallocated state.
1907 int bdrv_is_allocated_above(BlockDriverState *top,
1908 BlockDriverState *base,
1910 int nb_sectors, int *pnum)
1912 BlockDriverState *intermediate;
1913 int ret, n = nb_sectors;
1916 while (intermediate && intermediate != base) {
1918 ret = bdrv_is_allocated(intermediate, sector_num, nb_sectors,
1928 * [sector_num, nb_sectors] is unallocated on top but intermediate
1931 * [sector_num+x, nr_sectors] allocated.
1933 if (n > pnum_inter &&
1934 (intermediate == top ||
1935 sector_num + pnum_inter < intermediate->total_sectors)) {
1939 intermediate = backing_bs(intermediate);
1946 typedef struct BdrvVmstateCo {
1947 BlockDriverState *bs;
1954 static int coroutine_fn
1955 bdrv_co_rw_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, int64_t pos,
1958 BlockDriver *drv = bs->drv;
1962 } else if (drv->bdrv_load_vmstate) {
1963 return is_read ? drv->bdrv_load_vmstate(bs, qiov, pos)
1964 : drv->bdrv_save_vmstate(bs, qiov, pos);
1965 } else if (bs->file) {
1966 return bdrv_co_rw_vmstate(bs->file->bs, qiov, pos, is_read);
1972 static void coroutine_fn bdrv_co_rw_vmstate_entry(void *opaque)
1974 BdrvVmstateCo *co = opaque;
1975 co->ret = bdrv_co_rw_vmstate(co->bs, co->qiov, co->pos, co->is_read);
1979 bdrv_rw_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, int64_t pos,
1982 if (qemu_in_coroutine()) {
1983 return bdrv_co_rw_vmstate(bs, qiov, pos, is_read);
1985 BdrvVmstateCo data = {
1990 .ret = -EINPROGRESS,
1992 Coroutine *co = qemu_coroutine_create(bdrv_co_rw_vmstate_entry, &data);
1994 qemu_coroutine_enter(co);
1995 while (data.ret == -EINPROGRESS) {
1996 aio_poll(bdrv_get_aio_context(bs), true);
2002 int bdrv_save_vmstate(BlockDriverState *bs, const uint8_t *buf,
2003 int64_t pos, int size)
2006 struct iovec iov = {
2007 .iov_base = (void *) buf,
2012 qemu_iovec_init_external(&qiov, &iov, 1);
2014 ret = bdrv_writev_vmstate(bs, &qiov, pos);
2022 int bdrv_writev_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, int64_t pos)
2024 return bdrv_rw_vmstate(bs, qiov, pos, false);
2027 int bdrv_load_vmstate(BlockDriverState *bs, uint8_t *buf,
2028 int64_t pos, int size)
2031 struct iovec iov = {
2037 qemu_iovec_init_external(&qiov, &iov, 1);
2038 ret = bdrv_readv_vmstate(bs, &qiov, pos);
2046 int bdrv_readv_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, int64_t pos)
2048 return bdrv_rw_vmstate(bs, qiov, pos, true);
2051 /**************************************************************/
2054 BlockAIOCB *bdrv_aio_readv(BdrvChild *child, int64_t sector_num,
2055 QEMUIOVector *qiov, int nb_sectors,
2056 BlockCompletionFunc *cb, void *opaque)
2058 trace_bdrv_aio_readv(child->bs, sector_num, nb_sectors, opaque);
2060 assert(nb_sectors << BDRV_SECTOR_BITS == qiov->size);
2061 return bdrv_co_aio_prw_vector(child, sector_num << BDRV_SECTOR_BITS, qiov,
2062 0, cb, opaque, false);
2065 BlockAIOCB *bdrv_aio_writev(BdrvChild *child, int64_t sector_num,
2066 QEMUIOVector *qiov, int nb_sectors,
2067 BlockCompletionFunc *cb, void *opaque)
2069 trace_bdrv_aio_writev(child->bs, sector_num, nb_sectors, opaque);
2071 assert(nb_sectors << BDRV_SECTOR_BITS == qiov->size);
2072 return bdrv_co_aio_prw_vector(child, sector_num << BDRV_SECTOR_BITS, qiov,
2073 0, cb, opaque, true);
2076 void bdrv_aio_cancel(BlockAIOCB *acb)
2079 bdrv_aio_cancel_async(acb);
2080 while (acb->refcnt > 1) {
2081 if (acb->aiocb_info->get_aio_context) {
2082 aio_poll(acb->aiocb_info->get_aio_context(acb), true);
2083 } else if (acb->bs) {
2084 aio_poll(bdrv_get_aio_context(acb->bs), true);
2089 qemu_aio_unref(acb);
2092 /* Async version of aio cancel. The caller is not blocked if the acb implements
2093 * cancel_async, otherwise we do nothing and let the request normally complete.
2094 * In either case the completion callback must be called. */
2095 void bdrv_aio_cancel_async(BlockAIOCB *acb)
2097 if (acb->aiocb_info->cancel_async) {
2098 acb->aiocb_info->cancel_async(acb);
2102 /**************************************************************/
2103 /* async block device emulation */
2105 typedef struct BlockRequest {
2107 /* Used during read, write, trim */
2114 /* Used during ioctl */
2120 BlockCompletionFunc *cb;
2126 typedef struct BlockAIOCBCoroutine {
2133 } BlockAIOCBCoroutine;
2135 static const AIOCBInfo bdrv_em_co_aiocb_info = {
2136 .aiocb_size = sizeof(BlockAIOCBCoroutine),
2139 static void bdrv_co_complete(BlockAIOCBCoroutine *acb)
2141 if (!acb->need_bh) {
2142 bdrv_dec_in_flight(acb->common.bs);
2143 acb->common.cb(acb->common.opaque, acb->req.error);
2144 qemu_aio_unref(acb);
2148 static void bdrv_co_em_bh(void *opaque)
2150 BlockAIOCBCoroutine *acb = opaque;
2152 assert(!acb->need_bh);
2153 bdrv_co_complete(acb);
2156 static void bdrv_co_maybe_schedule_bh(BlockAIOCBCoroutine *acb)
2158 acb->need_bh = false;
2159 if (acb->req.error != -EINPROGRESS) {
2160 BlockDriverState *bs = acb->common.bs;
2162 aio_bh_schedule_oneshot(bdrv_get_aio_context(bs), bdrv_co_em_bh, acb);
2166 /* Invoke bdrv_co_do_readv/bdrv_co_do_writev */
2167 static void coroutine_fn bdrv_co_do_rw(void *opaque)
2169 BlockAIOCBCoroutine *acb = opaque;
2171 if (!acb->is_write) {
2172 acb->req.error = bdrv_co_preadv(acb->child, acb->req.offset,
2173 acb->req.qiov->size, acb->req.qiov, acb->req.flags);
2175 acb->req.error = bdrv_co_pwritev(acb->child, acb->req.offset,
2176 acb->req.qiov->size, acb->req.qiov, acb->req.flags);
2179 bdrv_co_complete(acb);
2182 static BlockAIOCB *bdrv_co_aio_prw_vector(BdrvChild *child,
2185 BdrvRequestFlags flags,
2186 BlockCompletionFunc *cb,
2191 BlockAIOCBCoroutine *acb;
2193 /* Matched by bdrv_co_complete's bdrv_dec_in_flight. */
2194 bdrv_inc_in_flight(child->bs);
2196 acb = qemu_aio_get(&bdrv_em_co_aiocb_info, child->bs, cb, opaque);
2198 acb->need_bh = true;
2199 acb->req.error = -EINPROGRESS;
2200 acb->req.offset = offset;
2201 acb->req.qiov = qiov;
2202 acb->req.flags = flags;
2203 acb->is_write = is_write;
2205 co = qemu_coroutine_create(bdrv_co_do_rw, acb);
2206 qemu_coroutine_enter(co);
2208 bdrv_co_maybe_schedule_bh(acb);
2209 return &acb->common;
2212 static void coroutine_fn bdrv_aio_flush_co_entry(void *opaque)
2214 BlockAIOCBCoroutine *acb = opaque;
2215 BlockDriverState *bs = acb->common.bs;
2217 acb->req.error = bdrv_co_flush(bs);
2218 bdrv_co_complete(acb);
2221 BlockAIOCB *bdrv_aio_flush(BlockDriverState *bs,
2222 BlockCompletionFunc *cb, void *opaque)
2224 trace_bdrv_aio_flush(bs, opaque);
2227 BlockAIOCBCoroutine *acb;
2229 /* Matched by bdrv_co_complete's bdrv_dec_in_flight. */
2230 bdrv_inc_in_flight(bs);
2232 acb = qemu_aio_get(&bdrv_em_co_aiocb_info, bs, cb, opaque);
2233 acb->need_bh = true;
2234 acb->req.error = -EINPROGRESS;
2236 co = qemu_coroutine_create(bdrv_aio_flush_co_entry, acb);
2237 qemu_coroutine_enter(co);
2239 bdrv_co_maybe_schedule_bh(acb);
2240 return &acb->common;
2243 void *qemu_aio_get(const AIOCBInfo *aiocb_info, BlockDriverState *bs,
2244 BlockCompletionFunc *cb, void *opaque)
2248 acb = g_malloc(aiocb_info->aiocb_size);
2249 acb->aiocb_info = aiocb_info;
2252 acb->opaque = opaque;
2257 void qemu_aio_ref(void *p)
2259 BlockAIOCB *acb = p;
2263 void qemu_aio_unref(void *p)
2265 BlockAIOCB *acb = p;
2266 assert(acb->refcnt > 0);
2267 if (--acb->refcnt == 0) {
2272 /**************************************************************/
2273 /* Coroutine block device emulation */
2275 typedef struct FlushCo {
2276 BlockDriverState *bs;
2281 static void coroutine_fn bdrv_flush_co_entry(void *opaque)
2283 FlushCo *rwco = opaque;
2285 rwco->ret = bdrv_co_flush(rwco->bs);
2288 int coroutine_fn bdrv_co_flush(BlockDriverState *bs)
2292 if (!bs || !bdrv_is_inserted(bs) || bdrv_is_read_only(bs) ||
2297 bdrv_inc_in_flight(bs);
2299 int current_gen = bs->write_gen;
2301 /* Wait until any previous flushes are completed */
2302 while (bs->active_flush_req) {
2303 qemu_co_queue_wait(&bs->flush_queue);
2306 bs->active_flush_req = true;
2308 /* Write back all layers by calling one driver function */
2309 if (bs->drv->bdrv_co_flush) {
2310 ret = bs->drv->bdrv_co_flush(bs);
2314 /* Write back cached data to the OS even with cache=unsafe */
2315 BLKDBG_EVENT(bs->file, BLKDBG_FLUSH_TO_OS);
2316 if (bs->drv->bdrv_co_flush_to_os) {
2317 ret = bs->drv->bdrv_co_flush_to_os(bs);
2323 /* But don't actually force it to the disk with cache=unsafe */
2324 if (bs->open_flags & BDRV_O_NO_FLUSH) {
2328 /* Check if we really need to flush anything */
2329 if (bs->flushed_gen == current_gen) {
2333 BLKDBG_EVENT(bs->file, BLKDBG_FLUSH_TO_DISK);
2334 if (bs->drv->bdrv_co_flush_to_disk) {
2335 ret = bs->drv->bdrv_co_flush_to_disk(bs);
2336 } else if (bs->drv->bdrv_aio_flush) {
2338 CoroutineIOCompletion co = {
2339 .coroutine = qemu_coroutine_self(),
2342 acb = bs->drv->bdrv_aio_flush(bs, bdrv_co_io_em_complete, &co);
2346 qemu_coroutine_yield();
2351 * Some block drivers always operate in either writethrough or unsafe
2352 * mode and don't support bdrv_flush therefore. Usually qemu doesn't
2353 * know how the server works (because the behaviour is hardcoded or
2354 * depends on server-side configuration), so we can't ensure that
2355 * everything is safe on disk. Returning an error doesn't work because
2356 * that would break guests even if the server operates in writethrough
2359 * Let's hope the user knows what he's doing.
2368 /* Now flush the underlying protocol. It will also have BDRV_O_NO_FLUSH
2369 * in the case of cache=unsafe, so there are no useless flushes.
2372 ret = bs->file ? bdrv_co_flush(bs->file->bs) : 0;
2374 /* Notify any pending flushes that we have completed */
2375 bs->flushed_gen = current_gen;
2376 bs->active_flush_req = false;
2377 /* Return value is ignored - it's ok if wait queue is empty */
2378 qemu_co_queue_next(&bs->flush_queue);
2380 bdrv_dec_in_flight(bs);
2384 int bdrv_flush(BlockDriverState *bs)
2387 FlushCo flush_co = {
2392 if (qemu_in_coroutine()) {
2393 /* Fast-path if already in coroutine context */
2394 bdrv_flush_co_entry(&flush_co);
2396 co = qemu_coroutine_create(bdrv_flush_co_entry, &flush_co);
2397 qemu_coroutine_enter(co);
2398 BDRV_POLL_WHILE(bs, flush_co.ret == NOT_DONE);
2401 return flush_co.ret;
2404 typedef struct DiscardCo {
2405 BlockDriverState *bs;
2410 static void coroutine_fn bdrv_pdiscard_co_entry(void *opaque)
2412 DiscardCo *rwco = opaque;
2414 rwco->ret = bdrv_co_pdiscard(rwco->bs, rwco->offset, rwco->count);
2417 int coroutine_fn bdrv_co_pdiscard(BlockDriverState *bs, int64_t offset,
2420 BdrvTrackedRequest req;
2421 int max_pdiscard, ret;
2428 ret = bdrv_check_byte_request(bs, offset, count);
2431 } else if (bs->read_only) {
2434 assert(!(bs->open_flags & BDRV_O_INACTIVE));
2436 /* Do nothing if disabled. */
2437 if (!(bs->open_flags & BDRV_O_UNMAP)) {
2441 if (!bs->drv->bdrv_co_pdiscard && !bs->drv->bdrv_aio_pdiscard) {
2445 /* Discard is advisory, so ignore any unaligned head or tail */
2446 align = MAX(bs->bl.pdiscard_alignment, bs->bl.request_alignment);
2447 assert(align % bs->bl.request_alignment == 0);
2448 head = offset % align;
2450 head = MIN(count, align - head);
2454 count = QEMU_ALIGN_DOWN(count, align);
2459 bdrv_inc_in_flight(bs);
2460 tracked_request_begin(&req, bs, offset, count, BDRV_TRACKED_DISCARD);
2462 ret = notifier_with_return_list_notify(&bs->before_write_notifiers, &req);
2467 max_pdiscard = QEMU_ALIGN_DOWN(MIN_NON_ZERO(bs->bl.max_pdiscard, INT_MAX),
2469 assert(max_pdiscard);
2473 int num = MIN(count, max_pdiscard);
2475 if (bs->drv->bdrv_co_pdiscard) {
2476 ret = bs->drv->bdrv_co_pdiscard(bs, offset, num);
2479 CoroutineIOCompletion co = {
2480 .coroutine = qemu_coroutine_self(),
2483 acb = bs->drv->bdrv_aio_pdiscard(bs, offset, num,
2484 bdrv_co_io_em_complete, &co);
2489 qemu_coroutine_yield();
2493 if (ret && ret != -ENOTSUP) {
2503 bdrv_set_dirty(bs, req.offset >> BDRV_SECTOR_BITS,
2504 req.bytes >> BDRV_SECTOR_BITS);
2505 tracked_request_end(&req);
2506 bdrv_dec_in_flight(bs);
2510 int bdrv_pdiscard(BlockDriverState *bs, int64_t offset, int count)
2520 if (qemu_in_coroutine()) {
2521 /* Fast-path if already in coroutine context */
2522 bdrv_pdiscard_co_entry(&rwco);
2524 co = qemu_coroutine_create(bdrv_pdiscard_co_entry, &rwco);
2525 qemu_coroutine_enter(co);
2526 BDRV_POLL_WHILE(bs, rwco.ret == NOT_DONE);
2532 int bdrv_co_ioctl(BlockDriverState *bs, int req, void *buf)
2534 BlockDriver *drv = bs->drv;
2535 CoroutineIOCompletion co = {
2536 .coroutine = qemu_coroutine_self(),
2540 bdrv_inc_in_flight(bs);
2541 if (!drv || (!drv->bdrv_aio_ioctl && !drv->bdrv_co_ioctl)) {
2546 if (drv->bdrv_co_ioctl) {
2547 co.ret = drv->bdrv_co_ioctl(bs, req, buf);
2549 acb = drv->bdrv_aio_ioctl(bs, req, buf, bdrv_co_io_em_complete, &co);
2554 qemu_coroutine_yield();
2557 bdrv_dec_in_flight(bs);
2561 void *qemu_blockalign(BlockDriverState *bs, size_t size)
2563 return qemu_memalign(bdrv_opt_mem_align(bs), size);
2566 void *qemu_blockalign0(BlockDriverState *bs, size_t size)
2568 return memset(qemu_blockalign(bs, size), 0, size);
2571 void *qemu_try_blockalign(BlockDriverState *bs, size_t size)
2573 size_t align = bdrv_opt_mem_align(bs);
2575 /* Ensure that NULL is never returned on success */
2581 return qemu_try_memalign(align, size);
2584 void *qemu_try_blockalign0(BlockDriverState *bs, size_t size)
2586 void *mem = qemu_try_blockalign(bs, size);
2589 memset(mem, 0, size);
2596 * Check if all memory in this vector is sector aligned.
2598 bool bdrv_qiov_is_aligned(BlockDriverState *bs, QEMUIOVector *qiov)
2601 size_t alignment = bdrv_min_mem_align(bs);
2603 for (i = 0; i < qiov->niov; i++) {
2604 if ((uintptr_t) qiov->iov[i].iov_base % alignment) {
2607 if (qiov->iov[i].iov_len % alignment) {
2615 void bdrv_add_before_write_notifier(BlockDriverState *bs,
2616 NotifierWithReturn *notifier)
2618 notifier_with_return_list_add(&bs->before_write_notifiers, notifier);
2621 void bdrv_io_plug(BlockDriverState *bs)
2625 QLIST_FOREACH(child, &bs->children, next) {
2626 bdrv_io_plug(child->bs);
2629 if (bs->io_plugged++ == 0 && bs->io_plug_disabled == 0) {
2630 BlockDriver *drv = bs->drv;
2631 if (drv && drv->bdrv_io_plug) {
2632 drv->bdrv_io_plug(bs);
2637 void bdrv_io_unplug(BlockDriverState *bs)
2641 assert(bs->io_plugged);
2642 if (--bs->io_plugged == 0 && bs->io_plug_disabled == 0) {
2643 BlockDriver *drv = bs->drv;
2644 if (drv && drv->bdrv_io_unplug) {
2645 drv->bdrv_io_unplug(bs);
2649 QLIST_FOREACH(child, &bs->children, next) {
2650 bdrv_io_unplug(child->bs);
2654 void bdrv_io_unplugged_begin(BlockDriverState *bs)
2658 if (bs->io_plug_disabled++ == 0 && bs->io_plugged > 0) {
2659 BlockDriver *drv = bs->drv;
2660 if (drv && drv->bdrv_io_unplug) {
2661 drv->bdrv_io_unplug(bs);
2665 QLIST_FOREACH(child, &bs->children, next) {
2666 bdrv_io_unplugged_begin(child->bs);
2670 void bdrv_io_unplugged_end(BlockDriverState *bs)
2674 assert(bs->io_plug_disabled);
2675 QLIST_FOREACH(child, &bs->children, next) {
2676 bdrv_io_unplugged_end(child->bs);
2679 if (--bs->io_plug_disabled == 0 && bs->io_plugged > 0) {
2680 BlockDriver *drv = bs->drv;
2681 if (drv && drv->bdrv_io_plug) {
2682 drv->bdrv_io_plug(bs);