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 (!QLIST_EMPTY(&bs->tracked_requests)) {
150 QLIST_FOREACH(child, &bs->children, next) {
151 if (bdrv_requests_pending(child->bs)) {
159 static void bdrv_drain_recurse(BlockDriverState *bs)
163 if (bs->drv && bs->drv->bdrv_drain) {
164 bs->drv->bdrv_drain(bs);
166 QLIST_FOREACH(child, &bs->children, next) {
167 bdrv_drain_recurse(child->bs);
173 BlockDriverState *bs;
178 static void bdrv_drain_poll(BlockDriverState *bs)
184 busy = bdrv_requests_pending(bs);
185 busy |= aio_poll(bdrv_get_aio_context(bs), busy);
189 static void bdrv_co_drain_bh_cb(void *opaque)
191 BdrvCoDrainData *data = opaque;
192 Coroutine *co = data->co;
194 qemu_bh_delete(data->bh);
195 bdrv_drain_poll(data->bs);
197 qemu_coroutine_enter(co);
200 static void coroutine_fn bdrv_co_yield_to_drain(BlockDriverState *bs)
202 BdrvCoDrainData data;
204 /* Calling bdrv_drain() from a BH ensures the current coroutine yields and
205 * other coroutines run if they were queued from
206 * qemu_co_queue_run_restart(). */
208 assert(qemu_in_coroutine());
209 data = (BdrvCoDrainData) {
210 .co = qemu_coroutine_self(),
213 .bh = aio_bh_new(bdrv_get_aio_context(bs), bdrv_co_drain_bh_cb, &data),
215 qemu_bh_schedule(data.bh);
217 qemu_coroutine_yield();
218 /* If we are resumed from some other event (such as an aio completion or a
219 * timer callback), it is a bug in the caller that should be fixed. */
223 void bdrv_drained_begin(BlockDriverState *bs)
225 if (!bs->quiesce_counter++) {
226 aio_disable_external(bdrv_get_aio_context(bs));
227 bdrv_parent_drained_begin(bs);
230 bdrv_io_unplugged_begin(bs);
231 bdrv_drain_recurse(bs);
232 if (qemu_in_coroutine()) {
233 bdrv_co_yield_to_drain(bs);
237 bdrv_io_unplugged_end(bs);
240 void bdrv_drained_end(BlockDriverState *bs)
242 assert(bs->quiesce_counter > 0);
243 if (--bs->quiesce_counter > 0) {
247 bdrv_parent_drained_end(bs);
248 aio_enable_external(bdrv_get_aio_context(bs));
252 * Wait for pending requests to complete on a single BlockDriverState subtree,
253 * and suspend block driver's internal I/O until next request arrives.
255 * Note that unlike bdrv_drain_all(), the caller must hold the BlockDriverState
258 * Only this BlockDriverState's AioContext is run, so in-flight requests must
259 * not depend on events in other AioContexts. In that case, use
260 * bdrv_drain_all() instead.
262 void coroutine_fn bdrv_co_drain(BlockDriverState *bs)
264 assert(qemu_in_coroutine());
265 bdrv_drained_begin(bs);
266 bdrv_drained_end(bs);
269 void bdrv_drain(BlockDriverState *bs)
271 bdrv_drained_begin(bs);
272 bdrv_drained_end(bs);
276 * Wait for pending requests to complete across all BlockDriverStates
278 * This function does not flush data to disk, use bdrv_flush_all() for that
279 * after calling this function.
281 void bdrv_drain_all(void)
283 /* Always run first iteration so any pending completion BHs run */
285 BlockDriverState *bs;
287 BlockJob *job = NULL;
288 GSList *aio_ctxs = NULL, *ctx;
290 while ((job = block_job_next(job))) {
291 AioContext *aio_context = blk_get_aio_context(job->blk);
293 aio_context_acquire(aio_context);
294 block_job_pause(job);
295 aio_context_release(aio_context);
298 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
299 AioContext *aio_context = bdrv_get_aio_context(bs);
301 aio_context_acquire(aio_context);
302 bdrv_parent_drained_begin(bs);
303 bdrv_io_unplugged_begin(bs);
304 bdrv_drain_recurse(bs);
305 aio_context_release(aio_context);
307 if (!g_slist_find(aio_ctxs, aio_context)) {
308 aio_ctxs = g_slist_prepend(aio_ctxs, aio_context);
312 /* Note that completion of an asynchronous I/O operation can trigger any
313 * number of other I/O operations on other devices---for example a
314 * coroutine can submit an I/O request to another device in response to
315 * request completion. Therefore we must keep looping until there was no
316 * more activity rather than simply draining each device independently.
321 for (ctx = aio_ctxs; ctx != NULL; ctx = ctx->next) {
322 AioContext *aio_context = ctx->data;
324 aio_context_acquire(aio_context);
325 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
326 if (aio_context == bdrv_get_aio_context(bs)) {
327 if (bdrv_requests_pending(bs)) {
329 aio_poll(aio_context, busy);
333 busy |= aio_poll(aio_context, false);
334 aio_context_release(aio_context);
338 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
339 AioContext *aio_context = bdrv_get_aio_context(bs);
341 aio_context_acquire(aio_context);
342 bdrv_io_unplugged_end(bs);
343 bdrv_parent_drained_end(bs);
344 aio_context_release(aio_context);
346 g_slist_free(aio_ctxs);
349 while ((job = block_job_next(job))) {
350 AioContext *aio_context = blk_get_aio_context(job->blk);
352 aio_context_acquire(aio_context);
353 block_job_resume(job);
354 aio_context_release(aio_context);
359 * Remove an active request from the tracked requests list
361 * This function should be called when a tracked request is completing.
363 static void tracked_request_end(BdrvTrackedRequest *req)
365 if (req->serialising) {
366 req->bs->serialising_in_flight--;
369 QLIST_REMOVE(req, list);
370 qemu_co_queue_restart_all(&req->wait_queue);
374 * Add an active request to the tracked requests list
376 static void tracked_request_begin(BdrvTrackedRequest *req,
377 BlockDriverState *bs,
380 enum BdrvTrackedRequestType type)
382 *req = (BdrvTrackedRequest){
387 .co = qemu_coroutine_self(),
388 .serialising = false,
389 .overlap_offset = offset,
390 .overlap_bytes = bytes,
393 qemu_co_queue_init(&req->wait_queue);
395 QLIST_INSERT_HEAD(&bs->tracked_requests, req, list);
398 static void mark_request_serialising(BdrvTrackedRequest *req, uint64_t align)
400 int64_t overlap_offset = req->offset & ~(align - 1);
401 unsigned int overlap_bytes = ROUND_UP(req->offset + req->bytes, align)
404 if (!req->serialising) {
405 req->bs->serialising_in_flight++;
406 req->serialising = true;
409 req->overlap_offset = MIN(req->overlap_offset, overlap_offset);
410 req->overlap_bytes = MAX(req->overlap_bytes, overlap_bytes);
414 * Round a region to cluster boundaries (sector-based)
416 void bdrv_round_sectors_to_clusters(BlockDriverState *bs,
417 int64_t sector_num, int nb_sectors,
418 int64_t *cluster_sector_num,
419 int *cluster_nb_sectors)
423 if (bdrv_get_info(bs, &bdi) < 0 || bdi.cluster_size == 0) {
424 *cluster_sector_num = sector_num;
425 *cluster_nb_sectors = nb_sectors;
427 int64_t c = bdi.cluster_size / BDRV_SECTOR_SIZE;
428 *cluster_sector_num = QEMU_ALIGN_DOWN(sector_num, c);
429 *cluster_nb_sectors = QEMU_ALIGN_UP(sector_num - *cluster_sector_num +
435 * Round a region to cluster boundaries
437 void bdrv_round_to_clusters(BlockDriverState *bs,
438 int64_t offset, unsigned int bytes,
439 int64_t *cluster_offset,
440 unsigned int *cluster_bytes)
444 if (bdrv_get_info(bs, &bdi) < 0 || bdi.cluster_size == 0) {
445 *cluster_offset = offset;
446 *cluster_bytes = bytes;
448 int64_t c = bdi.cluster_size;
449 *cluster_offset = QEMU_ALIGN_DOWN(offset, c);
450 *cluster_bytes = QEMU_ALIGN_UP(offset - *cluster_offset + bytes, c);
454 static int bdrv_get_cluster_size(BlockDriverState *bs)
459 ret = bdrv_get_info(bs, &bdi);
460 if (ret < 0 || bdi.cluster_size == 0) {
461 return bs->bl.request_alignment;
463 return bdi.cluster_size;
467 static bool tracked_request_overlaps(BdrvTrackedRequest *req,
468 int64_t offset, unsigned int bytes)
471 if (offset >= req->overlap_offset + req->overlap_bytes) {
475 if (req->overlap_offset >= offset + bytes) {
481 static bool coroutine_fn wait_serialising_requests(BdrvTrackedRequest *self)
483 BlockDriverState *bs = self->bs;
484 BdrvTrackedRequest *req;
488 if (!bs->serialising_in_flight) {
494 QLIST_FOREACH(req, &bs->tracked_requests, list) {
495 if (req == self || (!req->serialising && !self->serialising)) {
498 if (tracked_request_overlaps(req, self->overlap_offset,
499 self->overlap_bytes))
501 /* Hitting this means there was a reentrant request, for
502 * example, a block driver issuing nested requests. This must
503 * never happen since it means deadlock.
505 assert(qemu_coroutine_self() != req->co);
507 /* If the request is already (indirectly) waiting for us, or
508 * will wait for us as soon as it wakes up, then just go on
509 * (instead of producing a deadlock in the former case). */
510 if (!req->waiting_for) {
511 self->waiting_for = req;
512 qemu_co_queue_wait(&req->wait_queue);
513 self->waiting_for = NULL;
525 static int bdrv_check_byte_request(BlockDriverState *bs, int64_t offset,
528 if (size > BDRV_REQUEST_MAX_SECTORS << BDRV_SECTOR_BITS) {
532 if (!bdrv_is_inserted(bs)) {
543 typedef struct RwCo {
549 BdrvRequestFlags flags;
552 static void coroutine_fn bdrv_rw_co_entry(void *opaque)
556 if (!rwco->is_write) {
557 rwco->ret = bdrv_co_preadv(rwco->child, rwco->offset,
558 rwco->qiov->size, rwco->qiov,
561 rwco->ret = bdrv_co_pwritev(rwco->child, rwco->offset,
562 rwco->qiov->size, rwco->qiov,
568 * Process a vectored synchronous request using coroutines
570 static int bdrv_prwv_co(BdrvChild *child, int64_t offset,
571 QEMUIOVector *qiov, bool is_write,
572 BdrvRequestFlags flags)
579 .is_write = is_write,
584 if (qemu_in_coroutine()) {
585 /* Fast-path if already in coroutine context */
586 bdrv_rw_co_entry(&rwco);
588 AioContext *aio_context = bdrv_get_aio_context(child->bs);
590 co = qemu_coroutine_create(bdrv_rw_co_entry, &rwco);
591 qemu_coroutine_enter(co);
592 while (rwco.ret == NOT_DONE) {
593 aio_poll(aio_context, true);
600 * Process a synchronous request using coroutines
602 static int bdrv_rw_co(BdrvChild *child, int64_t sector_num, uint8_t *buf,
603 int nb_sectors, bool is_write, BdrvRequestFlags flags)
607 .iov_base = (void *)buf,
608 .iov_len = nb_sectors * BDRV_SECTOR_SIZE,
611 if (nb_sectors < 0 || nb_sectors > BDRV_REQUEST_MAX_SECTORS) {
615 qemu_iovec_init_external(&qiov, &iov, 1);
616 return bdrv_prwv_co(child, sector_num << BDRV_SECTOR_BITS,
617 &qiov, is_write, flags);
620 /* return < 0 if error. See bdrv_write() for the return codes */
621 int bdrv_read(BdrvChild *child, int64_t sector_num,
622 uint8_t *buf, int nb_sectors)
624 return bdrv_rw_co(child, sector_num, buf, nb_sectors, false, 0);
627 /* Return < 0 if error. Important errors are:
628 -EIO generic I/O error (may happen for all errors)
629 -ENOMEDIUM No media inserted.
630 -EINVAL Invalid sector number or nb_sectors
631 -EACCES Trying to write a read-only device
633 int bdrv_write(BdrvChild *child, int64_t sector_num,
634 const uint8_t *buf, int nb_sectors)
636 return bdrv_rw_co(child, sector_num, (uint8_t *)buf, nb_sectors, true, 0);
639 int bdrv_pwrite_zeroes(BdrvChild *child, int64_t offset,
640 int count, BdrvRequestFlags flags)
648 qemu_iovec_init_external(&qiov, &iov, 1);
649 return bdrv_prwv_co(child, offset, &qiov, true,
650 BDRV_REQ_ZERO_WRITE | flags);
654 * Completely zero out a block device with the help of bdrv_pwrite_zeroes.
655 * The operation is sped up by checking the block status and only writing
656 * zeroes to the device if they currently do not return zeroes. Optional
657 * flags are passed through to bdrv_pwrite_zeroes (e.g. BDRV_REQ_MAY_UNMAP,
660 * Returns < 0 on error, 0 on success. For error codes see bdrv_write().
662 int bdrv_make_zero(BdrvChild *child, BdrvRequestFlags flags)
664 int64_t target_sectors, ret, nb_sectors, sector_num = 0;
665 BlockDriverState *bs = child->bs;
666 BlockDriverState *file;
669 target_sectors = bdrv_nb_sectors(bs);
670 if (target_sectors < 0) {
671 return target_sectors;
675 nb_sectors = MIN(target_sectors - sector_num, BDRV_REQUEST_MAX_SECTORS);
676 if (nb_sectors <= 0) {
679 ret = bdrv_get_block_status(bs, sector_num, nb_sectors, &n, &file);
681 error_report("error getting block status at sector %" PRId64 ": %s",
682 sector_num, strerror(-ret));
685 if (ret & BDRV_BLOCK_ZERO) {
689 ret = bdrv_pwrite_zeroes(child, sector_num << BDRV_SECTOR_BITS,
690 n << BDRV_SECTOR_BITS, flags);
692 error_report("error writing zeroes at sector %" PRId64 ": %s",
693 sector_num, strerror(-ret));
700 int bdrv_preadv(BdrvChild *child, int64_t offset, QEMUIOVector *qiov)
704 ret = bdrv_prwv_co(child, offset, qiov, false, 0);
712 int bdrv_pread(BdrvChild *child, int64_t offset, void *buf, int bytes)
716 .iov_base = (void *)buf,
724 qemu_iovec_init_external(&qiov, &iov, 1);
725 return bdrv_preadv(child, offset, &qiov);
728 int bdrv_pwritev(BdrvChild *child, int64_t offset, QEMUIOVector *qiov)
732 ret = bdrv_prwv_co(child, offset, qiov, true, 0);
740 int bdrv_pwrite(BdrvChild *child, int64_t offset, const void *buf, int bytes)
744 .iov_base = (void *) buf,
752 qemu_iovec_init_external(&qiov, &iov, 1);
753 return bdrv_pwritev(child, offset, &qiov);
757 * Writes to the file and ensures that no writes are reordered across this
758 * request (acts as a barrier)
760 * Returns 0 on success, -errno in error cases.
762 int bdrv_pwrite_sync(BdrvChild *child, int64_t offset,
763 const void *buf, int count)
767 ret = bdrv_pwrite(child, offset, buf, count);
772 ret = bdrv_flush(child->bs);
780 typedef struct CoroutineIOCompletion {
781 Coroutine *coroutine;
783 } CoroutineIOCompletion;
785 static void bdrv_co_io_em_complete(void *opaque, int ret)
787 CoroutineIOCompletion *co = opaque;
790 qemu_coroutine_enter(co->coroutine);
793 static int coroutine_fn bdrv_driver_preadv(BlockDriverState *bs,
794 uint64_t offset, uint64_t bytes,
795 QEMUIOVector *qiov, int flags)
797 BlockDriver *drv = bs->drv;
799 unsigned int nb_sectors;
801 assert(!(flags & ~BDRV_REQ_MASK));
803 if (drv->bdrv_co_preadv) {
804 return drv->bdrv_co_preadv(bs, offset, bytes, qiov, flags);
807 sector_num = offset >> BDRV_SECTOR_BITS;
808 nb_sectors = bytes >> BDRV_SECTOR_BITS;
810 assert((offset & (BDRV_SECTOR_SIZE - 1)) == 0);
811 assert((bytes & (BDRV_SECTOR_SIZE - 1)) == 0);
812 assert((bytes >> BDRV_SECTOR_BITS) <= BDRV_REQUEST_MAX_SECTORS);
814 if (drv->bdrv_co_readv) {
815 return drv->bdrv_co_readv(bs, sector_num, nb_sectors, qiov);
818 CoroutineIOCompletion co = {
819 .coroutine = qemu_coroutine_self(),
822 acb = bs->drv->bdrv_aio_readv(bs, sector_num, qiov, nb_sectors,
823 bdrv_co_io_em_complete, &co);
827 qemu_coroutine_yield();
833 static int coroutine_fn bdrv_driver_pwritev(BlockDriverState *bs,
834 uint64_t offset, uint64_t bytes,
835 QEMUIOVector *qiov, int flags)
837 BlockDriver *drv = bs->drv;
839 unsigned int nb_sectors;
842 assert(!(flags & ~BDRV_REQ_MASK));
844 if (drv->bdrv_co_pwritev) {
845 ret = drv->bdrv_co_pwritev(bs, offset, bytes, qiov,
846 flags & bs->supported_write_flags);
847 flags &= ~bs->supported_write_flags;
851 sector_num = offset >> BDRV_SECTOR_BITS;
852 nb_sectors = bytes >> BDRV_SECTOR_BITS;
854 assert((offset & (BDRV_SECTOR_SIZE - 1)) == 0);
855 assert((bytes & (BDRV_SECTOR_SIZE - 1)) == 0);
856 assert((bytes >> BDRV_SECTOR_BITS) <= BDRV_REQUEST_MAX_SECTORS);
858 if (drv->bdrv_co_writev_flags) {
859 ret = drv->bdrv_co_writev_flags(bs, sector_num, nb_sectors, qiov,
860 flags & bs->supported_write_flags);
861 flags &= ~bs->supported_write_flags;
862 } else if (drv->bdrv_co_writev) {
863 assert(!bs->supported_write_flags);
864 ret = drv->bdrv_co_writev(bs, sector_num, nb_sectors, qiov);
867 CoroutineIOCompletion co = {
868 .coroutine = qemu_coroutine_self(),
871 acb = bs->drv->bdrv_aio_writev(bs, sector_num, qiov, nb_sectors,
872 bdrv_co_io_em_complete, &co);
876 qemu_coroutine_yield();
882 if (ret == 0 && (flags & BDRV_REQ_FUA)) {
883 ret = bdrv_co_flush(bs);
889 static int coroutine_fn
890 bdrv_driver_pwritev_compressed(BlockDriverState *bs, uint64_t offset,
891 uint64_t bytes, QEMUIOVector *qiov)
893 BlockDriver *drv = bs->drv;
895 if (!drv->bdrv_co_pwritev_compressed) {
899 assert(QLIST_EMPTY(&bs->dirty_bitmaps));
900 return drv->bdrv_co_pwritev_compressed(bs, offset, bytes, qiov);
903 static int coroutine_fn bdrv_co_do_copy_on_readv(BlockDriverState *bs,
904 int64_t offset, unsigned int bytes, QEMUIOVector *qiov)
906 /* Perform I/O through a temporary buffer so that users who scribble over
907 * their read buffer while the operation is in progress do not end up
908 * modifying the image file. This is critical for zero-copy guest I/O
909 * where anything might happen inside guest memory.
913 BlockDriver *drv = bs->drv;
915 QEMUIOVector bounce_qiov;
916 int64_t cluster_offset;
917 unsigned int cluster_bytes;
921 /* Cover entire cluster so no additional backing file I/O is required when
922 * allocating cluster in the image file.
924 bdrv_round_to_clusters(bs, offset, bytes, &cluster_offset, &cluster_bytes);
926 trace_bdrv_co_do_copy_on_readv(bs, offset, bytes,
927 cluster_offset, cluster_bytes);
929 iov.iov_len = cluster_bytes;
930 iov.iov_base = bounce_buffer = qemu_try_blockalign(bs, iov.iov_len);
931 if (bounce_buffer == NULL) {
936 qemu_iovec_init_external(&bounce_qiov, &iov, 1);
938 ret = bdrv_driver_preadv(bs, cluster_offset, cluster_bytes,
944 if (drv->bdrv_co_pwrite_zeroes &&
945 buffer_is_zero(bounce_buffer, iov.iov_len)) {
946 /* FIXME: Should we (perhaps conditionally) be setting
947 * BDRV_REQ_MAY_UNMAP, if it will allow for a sparser copy
948 * that still correctly reads as zero? */
949 ret = bdrv_co_do_pwrite_zeroes(bs, cluster_offset, cluster_bytes, 0);
951 /* This does not change the data on the disk, it is not necessary
952 * to flush even in cache=writethrough mode.
954 ret = bdrv_driver_pwritev(bs, cluster_offset, cluster_bytes,
959 /* It might be okay to ignore write errors for guest requests. If this
960 * is a deliberate copy-on-read then we don't want to ignore the error.
961 * Simply report it in all cases.
966 skip_bytes = offset - cluster_offset;
967 qemu_iovec_from_buf(qiov, 0, bounce_buffer + skip_bytes, bytes);
970 qemu_vfree(bounce_buffer);
975 * Forwards an already correctly aligned request to the BlockDriver. This
976 * handles copy on read, zeroing after EOF, and fragmentation of large
977 * reads; any other features must be implemented by the caller.
979 static int coroutine_fn bdrv_aligned_preadv(BlockDriverState *bs,
980 BdrvTrackedRequest *req, int64_t offset, unsigned int bytes,
981 int64_t align, QEMUIOVector *qiov, int flags)
983 int64_t total_bytes, max_bytes;
985 uint64_t bytes_remaining = bytes;
988 assert(is_power_of_2(align));
989 assert((offset & (align - 1)) == 0);
990 assert((bytes & (align - 1)) == 0);
991 assert(!qiov || bytes == qiov->size);
992 assert((bs->open_flags & BDRV_O_NO_IO) == 0);
993 max_transfer = QEMU_ALIGN_DOWN(MIN_NON_ZERO(bs->bl.max_transfer, INT_MAX),
996 /* TODO: We would need a per-BDS .supported_read_flags and
997 * potential fallback support, if we ever implement any read flags
998 * to pass through to drivers. For now, there aren't any
999 * passthrough flags. */
1000 assert(!(flags & ~(BDRV_REQ_NO_SERIALISING | BDRV_REQ_COPY_ON_READ)));
1002 /* Handle Copy on Read and associated serialisation */
1003 if (flags & BDRV_REQ_COPY_ON_READ) {
1004 /* If we touch the same cluster it counts as an overlap. This
1005 * guarantees that allocating writes will be serialized and not race
1006 * with each other for the same cluster. For example, in copy-on-read
1007 * it ensures that the CoR read and write operations are atomic and
1008 * guest writes cannot interleave between them. */
1009 mark_request_serialising(req, bdrv_get_cluster_size(bs));
1012 if (!(flags & BDRV_REQ_NO_SERIALISING)) {
1013 wait_serialising_requests(req);
1016 if (flags & BDRV_REQ_COPY_ON_READ) {
1017 int64_t start_sector = offset >> BDRV_SECTOR_BITS;
1018 int64_t end_sector = DIV_ROUND_UP(offset + bytes, BDRV_SECTOR_SIZE);
1019 unsigned int nb_sectors = end_sector - start_sector;
1022 ret = bdrv_is_allocated(bs, start_sector, nb_sectors, &pnum);
1027 if (!ret || pnum != nb_sectors) {
1028 ret = bdrv_co_do_copy_on_readv(bs, offset, bytes, qiov);
1033 /* Forward the request to the BlockDriver, possibly fragmenting it */
1034 total_bytes = bdrv_getlength(bs);
1035 if (total_bytes < 0) {
1040 max_bytes = ROUND_UP(MAX(0, total_bytes - offset), align);
1041 if (bytes <= max_bytes && bytes <= max_transfer) {
1042 ret = bdrv_driver_preadv(bs, offset, bytes, qiov, 0);
1046 while (bytes_remaining) {
1050 QEMUIOVector local_qiov;
1052 num = MIN(bytes_remaining, MIN(max_bytes, max_transfer));
1054 qemu_iovec_init(&local_qiov, qiov->niov);
1055 qemu_iovec_concat(&local_qiov, qiov, bytes - bytes_remaining, num);
1057 ret = bdrv_driver_preadv(bs, offset + bytes - bytes_remaining,
1058 num, &local_qiov, 0);
1060 qemu_iovec_destroy(&local_qiov);
1062 num = bytes_remaining;
1063 ret = qemu_iovec_memset(qiov, bytes - bytes_remaining, 0,
1069 bytes_remaining -= num;
1073 return ret < 0 ? ret : 0;
1077 * Handle a read request in coroutine context
1079 int coroutine_fn bdrv_co_preadv(BdrvChild *child,
1080 int64_t offset, unsigned int bytes, QEMUIOVector *qiov,
1081 BdrvRequestFlags flags)
1083 BlockDriverState *bs = child->bs;
1084 BlockDriver *drv = bs->drv;
1085 BdrvTrackedRequest req;
1087 uint64_t align = bs->bl.request_alignment;
1088 uint8_t *head_buf = NULL;
1089 uint8_t *tail_buf = NULL;
1090 QEMUIOVector local_qiov;
1091 bool use_local_qiov = false;
1098 ret = bdrv_check_byte_request(bs, offset, bytes);
1103 /* Don't do copy-on-read if we read data before write operation */
1104 if (bs->copy_on_read && !(flags & BDRV_REQ_NO_SERIALISING)) {
1105 flags |= BDRV_REQ_COPY_ON_READ;
1108 /* Align read if necessary by padding qiov */
1109 if (offset & (align - 1)) {
1110 head_buf = qemu_blockalign(bs, align);
1111 qemu_iovec_init(&local_qiov, qiov->niov + 2);
1112 qemu_iovec_add(&local_qiov, head_buf, offset & (align - 1));
1113 qemu_iovec_concat(&local_qiov, qiov, 0, qiov->size);
1114 use_local_qiov = true;
1116 bytes += offset & (align - 1);
1117 offset = offset & ~(align - 1);
1120 if ((offset + bytes) & (align - 1)) {
1121 if (!use_local_qiov) {
1122 qemu_iovec_init(&local_qiov, qiov->niov + 1);
1123 qemu_iovec_concat(&local_qiov, qiov, 0, qiov->size);
1124 use_local_qiov = true;
1126 tail_buf = qemu_blockalign(bs, align);
1127 qemu_iovec_add(&local_qiov, tail_buf,
1128 align - ((offset + bytes) & (align - 1)));
1130 bytes = ROUND_UP(bytes, align);
1133 tracked_request_begin(&req, bs, offset, bytes, BDRV_TRACKED_READ);
1134 ret = bdrv_aligned_preadv(bs, &req, offset, bytes, align,
1135 use_local_qiov ? &local_qiov : qiov,
1137 tracked_request_end(&req);
1139 if (use_local_qiov) {
1140 qemu_iovec_destroy(&local_qiov);
1141 qemu_vfree(head_buf);
1142 qemu_vfree(tail_buf);
1148 static int coroutine_fn bdrv_co_do_readv(BdrvChild *child,
1149 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov,
1150 BdrvRequestFlags flags)
1152 if (nb_sectors < 0 || nb_sectors > BDRV_REQUEST_MAX_SECTORS) {
1156 return bdrv_co_preadv(child, sector_num << BDRV_SECTOR_BITS,
1157 nb_sectors << BDRV_SECTOR_BITS, qiov, flags);
1160 int coroutine_fn bdrv_co_readv(BdrvChild *child, int64_t sector_num,
1161 int nb_sectors, QEMUIOVector *qiov)
1163 trace_bdrv_co_readv(child->bs, sector_num, nb_sectors);
1165 return bdrv_co_do_readv(child, sector_num, nb_sectors, qiov, 0);
1168 /* Maximum buffer for write zeroes fallback, in bytes */
1169 #define MAX_WRITE_ZEROES_BOUNCE_BUFFER (32768 << BDRV_SECTOR_BITS)
1171 static int coroutine_fn bdrv_co_do_pwrite_zeroes(BlockDriverState *bs,
1172 int64_t offset, int count, BdrvRequestFlags flags)
1174 BlockDriver *drv = bs->drv;
1176 struct iovec iov = {0};
1178 bool need_flush = false;
1182 int max_write_zeroes = MIN_NON_ZERO(bs->bl.max_pwrite_zeroes, INT_MAX);
1183 int alignment = MAX(bs->bl.pwrite_zeroes_alignment,
1184 bs->bl.request_alignment);
1186 assert(alignment % bs->bl.request_alignment == 0);
1187 head = offset % alignment;
1188 tail = (offset + count) % alignment;
1189 max_write_zeroes = QEMU_ALIGN_DOWN(max_write_zeroes, alignment);
1190 assert(max_write_zeroes >= bs->bl.request_alignment);
1192 while (count > 0 && !ret) {
1195 /* Align request. Block drivers can expect the "bulk" of the request
1196 * to be aligned, and that unaligned requests do not cross cluster
1200 /* Make a small request up to the first aligned sector. */
1201 num = MIN(count, alignment - head);
1203 } else if (tail && num > alignment) {
1204 /* Shorten the request to the last aligned sector. */
1208 /* limit request size */
1209 if (num > max_write_zeroes) {
1210 num = max_write_zeroes;
1214 /* First try the efficient write zeroes operation */
1215 if (drv->bdrv_co_pwrite_zeroes) {
1216 ret = drv->bdrv_co_pwrite_zeroes(bs, offset, num,
1217 flags & bs->supported_zero_flags);
1218 if (ret != -ENOTSUP && (flags & BDRV_REQ_FUA) &&
1219 !(bs->supported_zero_flags & BDRV_REQ_FUA)) {
1223 assert(!bs->supported_zero_flags);
1226 if (ret == -ENOTSUP) {
1227 /* Fall back to bounce buffer if write zeroes is unsupported */
1228 int max_transfer = MIN_NON_ZERO(bs->bl.max_transfer,
1229 MAX_WRITE_ZEROES_BOUNCE_BUFFER);
1230 BdrvRequestFlags write_flags = flags & ~BDRV_REQ_ZERO_WRITE;
1232 if ((flags & BDRV_REQ_FUA) &&
1233 !(bs->supported_write_flags & BDRV_REQ_FUA)) {
1234 /* No need for bdrv_driver_pwrite() to do a fallback
1235 * flush on each chunk; use just one at the end */
1236 write_flags &= ~BDRV_REQ_FUA;
1239 num = MIN(num, max_transfer);
1241 if (iov.iov_base == NULL) {
1242 iov.iov_base = qemu_try_blockalign(bs, num);
1243 if (iov.iov_base == NULL) {
1247 memset(iov.iov_base, 0, num);
1249 qemu_iovec_init_external(&qiov, &iov, 1);
1251 ret = bdrv_driver_pwritev(bs, offset, num, &qiov, write_flags);
1253 /* Keep bounce buffer around if it is big enough for all
1254 * all future requests.
1256 if (num < max_transfer) {
1257 qemu_vfree(iov.iov_base);
1258 iov.iov_base = NULL;
1267 if (ret == 0 && need_flush) {
1268 ret = bdrv_co_flush(bs);
1270 qemu_vfree(iov.iov_base);
1275 * Forwards an already correctly aligned write request to the BlockDriver,
1276 * after possibly fragmenting it.
1278 static int coroutine_fn bdrv_aligned_pwritev(BlockDriverState *bs,
1279 BdrvTrackedRequest *req, int64_t offset, unsigned int bytes,
1280 int64_t align, QEMUIOVector *qiov, int flags)
1282 BlockDriver *drv = bs->drv;
1286 int64_t start_sector = offset >> BDRV_SECTOR_BITS;
1287 int64_t end_sector = DIV_ROUND_UP(offset + bytes, BDRV_SECTOR_SIZE);
1288 uint64_t bytes_remaining = bytes;
1291 assert(is_power_of_2(align));
1292 assert((offset & (align - 1)) == 0);
1293 assert((bytes & (align - 1)) == 0);
1294 assert(!qiov || bytes == qiov->size);
1295 assert((bs->open_flags & BDRV_O_NO_IO) == 0);
1296 assert(!(flags & ~BDRV_REQ_MASK));
1297 max_transfer = QEMU_ALIGN_DOWN(MIN_NON_ZERO(bs->bl.max_transfer, INT_MAX),
1300 waited = wait_serialising_requests(req);
1301 assert(!waited || !req->serialising);
1302 assert(req->overlap_offset <= offset);
1303 assert(offset + bytes <= req->overlap_offset + req->overlap_bytes);
1305 ret = notifier_with_return_list_notify(&bs->before_write_notifiers, req);
1307 if (!ret && bs->detect_zeroes != BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF &&
1308 !(flags & BDRV_REQ_ZERO_WRITE) && drv->bdrv_co_pwrite_zeroes &&
1309 qemu_iovec_is_zero(qiov)) {
1310 flags |= BDRV_REQ_ZERO_WRITE;
1311 if (bs->detect_zeroes == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP) {
1312 flags |= BDRV_REQ_MAY_UNMAP;
1317 /* Do nothing, write notifier decided to fail this request */
1318 } else if (flags & BDRV_REQ_ZERO_WRITE) {
1319 bdrv_debug_event(bs, BLKDBG_PWRITEV_ZERO);
1320 ret = bdrv_co_do_pwrite_zeroes(bs, offset, bytes, flags);
1321 } else if (bytes <= max_transfer) {
1322 bdrv_debug_event(bs, BLKDBG_PWRITEV);
1323 ret = bdrv_driver_pwritev(bs, offset, bytes, qiov, flags);
1325 bdrv_debug_event(bs, BLKDBG_PWRITEV);
1326 while (bytes_remaining) {
1327 int num = MIN(bytes_remaining, max_transfer);
1328 QEMUIOVector local_qiov;
1329 int local_flags = flags;
1332 if (num < bytes_remaining && (flags & BDRV_REQ_FUA) &&
1333 !(bs->supported_write_flags & BDRV_REQ_FUA)) {
1334 /* If FUA is going to be emulated by flush, we only
1335 * need to flush on the last iteration */
1336 local_flags &= ~BDRV_REQ_FUA;
1338 qemu_iovec_init(&local_qiov, qiov->niov);
1339 qemu_iovec_concat(&local_qiov, qiov, bytes - bytes_remaining, num);
1341 ret = bdrv_driver_pwritev(bs, offset + bytes - bytes_remaining,
1342 num, &local_qiov, local_flags);
1343 qemu_iovec_destroy(&local_qiov);
1347 bytes_remaining -= num;
1350 bdrv_debug_event(bs, BLKDBG_PWRITEV_DONE);
1353 bdrv_set_dirty(bs, start_sector, end_sector - start_sector);
1355 if (bs->wr_highest_offset < offset + bytes) {
1356 bs->wr_highest_offset = offset + bytes;
1360 bs->total_sectors = MAX(bs->total_sectors, end_sector);
1367 static int coroutine_fn bdrv_co_do_zero_pwritev(BlockDriverState *bs,
1370 BdrvRequestFlags flags,
1371 BdrvTrackedRequest *req)
1373 uint8_t *buf = NULL;
1374 QEMUIOVector local_qiov;
1376 uint64_t align = bs->bl.request_alignment;
1377 unsigned int head_padding_bytes, tail_padding_bytes;
1380 head_padding_bytes = offset & (align - 1);
1381 tail_padding_bytes = align - ((offset + bytes) & (align - 1));
1384 assert(flags & BDRV_REQ_ZERO_WRITE);
1385 if (head_padding_bytes || tail_padding_bytes) {
1386 buf = qemu_blockalign(bs, align);
1387 iov = (struct iovec) {
1391 qemu_iovec_init_external(&local_qiov, &iov, 1);
1393 if (head_padding_bytes) {
1394 uint64_t zero_bytes = MIN(bytes, align - head_padding_bytes);
1396 /* RMW the unaligned part before head. */
1397 mark_request_serialising(req, align);
1398 wait_serialising_requests(req);
1399 bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_HEAD);
1400 ret = bdrv_aligned_preadv(bs, req, offset & ~(align - 1), align,
1401 align, &local_qiov, 0);
1405 bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_AFTER_HEAD);
1407 memset(buf + head_padding_bytes, 0, zero_bytes);
1408 ret = bdrv_aligned_pwritev(bs, req, offset & ~(align - 1), align,
1410 flags & ~BDRV_REQ_ZERO_WRITE);
1414 offset += zero_bytes;
1415 bytes -= zero_bytes;
1418 assert(!bytes || (offset & (align - 1)) == 0);
1419 if (bytes >= align) {
1420 /* Write the aligned part in the middle. */
1421 uint64_t aligned_bytes = bytes & ~(align - 1);
1422 ret = bdrv_aligned_pwritev(bs, req, offset, aligned_bytes, align,
1427 bytes -= aligned_bytes;
1428 offset += aligned_bytes;
1431 assert(!bytes || (offset & (align - 1)) == 0);
1433 assert(align == tail_padding_bytes + bytes);
1434 /* RMW the unaligned part after tail. */
1435 mark_request_serialising(req, align);
1436 wait_serialising_requests(req);
1437 bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_TAIL);
1438 ret = bdrv_aligned_preadv(bs, req, offset, align,
1439 align, &local_qiov, 0);
1443 bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_AFTER_TAIL);
1445 memset(buf, 0, bytes);
1446 ret = bdrv_aligned_pwritev(bs, req, offset, align, align,
1447 &local_qiov, flags & ~BDRV_REQ_ZERO_WRITE);
1456 * Handle a write request in coroutine context
1458 int coroutine_fn bdrv_co_pwritev(BdrvChild *child,
1459 int64_t offset, unsigned int bytes, QEMUIOVector *qiov,
1460 BdrvRequestFlags flags)
1462 BlockDriverState *bs = child->bs;
1463 BdrvTrackedRequest req;
1464 uint64_t align = bs->bl.request_alignment;
1465 uint8_t *head_buf = NULL;
1466 uint8_t *tail_buf = NULL;
1467 QEMUIOVector local_qiov;
1468 bool use_local_qiov = false;
1474 if (bs->read_only) {
1477 assert(!(bs->open_flags & BDRV_O_INACTIVE));
1479 ret = bdrv_check_byte_request(bs, offset, bytes);
1485 * Align write if necessary by performing a read-modify-write cycle.
1486 * Pad qiov with the read parts and be sure to have a tracked request not
1487 * only for bdrv_aligned_pwritev, but also for the reads of the RMW cycle.
1489 tracked_request_begin(&req, bs, offset, bytes, BDRV_TRACKED_WRITE);
1492 ret = bdrv_co_do_zero_pwritev(bs, offset, bytes, flags, &req);
1496 if (offset & (align - 1)) {
1497 QEMUIOVector head_qiov;
1498 struct iovec head_iov;
1500 mark_request_serialising(&req, align);
1501 wait_serialising_requests(&req);
1503 head_buf = qemu_blockalign(bs, align);
1504 head_iov = (struct iovec) {
1505 .iov_base = head_buf,
1508 qemu_iovec_init_external(&head_qiov, &head_iov, 1);
1510 bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_HEAD);
1511 ret = bdrv_aligned_preadv(bs, &req, offset & ~(align - 1), align,
1512 align, &head_qiov, 0);
1516 bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_AFTER_HEAD);
1518 qemu_iovec_init(&local_qiov, qiov->niov + 2);
1519 qemu_iovec_add(&local_qiov, head_buf, offset & (align - 1));
1520 qemu_iovec_concat(&local_qiov, qiov, 0, qiov->size);
1521 use_local_qiov = true;
1523 bytes += offset & (align - 1);
1524 offset = offset & ~(align - 1);
1526 /* We have read the tail already if the request is smaller
1527 * than one aligned block.
1529 if (bytes < align) {
1530 qemu_iovec_add(&local_qiov, head_buf + bytes, align - bytes);
1535 if ((offset + bytes) & (align - 1)) {
1536 QEMUIOVector tail_qiov;
1537 struct iovec tail_iov;
1541 mark_request_serialising(&req, align);
1542 waited = wait_serialising_requests(&req);
1543 assert(!waited || !use_local_qiov);
1545 tail_buf = qemu_blockalign(bs, align);
1546 tail_iov = (struct iovec) {
1547 .iov_base = tail_buf,
1550 qemu_iovec_init_external(&tail_qiov, &tail_iov, 1);
1552 bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_TAIL);
1553 ret = bdrv_aligned_preadv(bs, &req, (offset + bytes) & ~(align - 1), align,
1554 align, &tail_qiov, 0);
1558 bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_AFTER_TAIL);
1560 if (!use_local_qiov) {
1561 qemu_iovec_init(&local_qiov, qiov->niov + 1);
1562 qemu_iovec_concat(&local_qiov, qiov, 0, qiov->size);
1563 use_local_qiov = true;
1566 tail_bytes = (offset + bytes) & (align - 1);
1567 qemu_iovec_add(&local_qiov, tail_buf + tail_bytes, align - tail_bytes);
1569 bytes = ROUND_UP(bytes, align);
1572 if (flags & BDRV_REQ_WRITE_COMPRESSED) {
1573 ret = bdrv_driver_pwritev_compressed(
1574 bs, offset, bytes, use_local_qiov ? &local_qiov : qiov);
1576 ret = bdrv_aligned_pwritev(bs, &req, offset, bytes, align,
1577 use_local_qiov ? &local_qiov : qiov,
1583 if (use_local_qiov) {
1584 qemu_iovec_destroy(&local_qiov);
1586 qemu_vfree(head_buf);
1587 qemu_vfree(tail_buf);
1589 tracked_request_end(&req);
1593 static int coroutine_fn bdrv_co_do_writev(BdrvChild *child,
1594 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov,
1595 BdrvRequestFlags flags)
1597 if (nb_sectors < 0 || nb_sectors > BDRV_REQUEST_MAX_SECTORS) {
1601 return bdrv_co_pwritev(child, sector_num << BDRV_SECTOR_BITS,
1602 nb_sectors << BDRV_SECTOR_BITS, qiov, flags);
1605 int coroutine_fn bdrv_co_writev(BdrvChild *child, int64_t sector_num,
1606 int nb_sectors, QEMUIOVector *qiov)
1608 trace_bdrv_co_writev(child->bs, sector_num, nb_sectors);
1610 return bdrv_co_do_writev(child, sector_num, nb_sectors, qiov, 0);
1613 int coroutine_fn bdrv_co_pwrite_zeroes(BdrvChild *child, int64_t offset,
1614 int count, BdrvRequestFlags flags)
1616 trace_bdrv_co_pwrite_zeroes(child->bs, offset, count, flags);
1618 if (!(child->bs->open_flags & BDRV_O_UNMAP)) {
1619 flags &= ~BDRV_REQ_MAY_UNMAP;
1622 return bdrv_co_pwritev(child, offset, count, NULL,
1623 BDRV_REQ_ZERO_WRITE | flags);
1626 typedef struct BdrvCoGetBlockStatusData {
1627 BlockDriverState *bs;
1628 BlockDriverState *base;
1629 BlockDriverState **file;
1635 } BdrvCoGetBlockStatusData;
1638 * Returns the allocation status of the specified sectors.
1639 * Drivers not implementing the functionality are assumed to not support
1640 * backing files, hence all their sectors are reported as allocated.
1642 * If 'sector_num' is beyond the end of the disk image the return value is 0
1643 * and 'pnum' is set to 0.
1645 * 'pnum' is set to the number of sectors (including and immediately following
1646 * the specified sector) that are known to be in the same
1647 * allocated/unallocated state.
1649 * 'nb_sectors' is the max value 'pnum' should be set to. If nb_sectors goes
1650 * beyond the end of the disk image it will be clamped.
1652 * If returned value is positive and BDRV_BLOCK_OFFSET_VALID bit is set, 'file'
1653 * points to the BDS which the sector range is allocated in.
1655 static int64_t coroutine_fn bdrv_co_get_block_status(BlockDriverState *bs,
1657 int nb_sectors, int *pnum,
1658 BlockDriverState **file)
1660 int64_t total_sectors;
1664 total_sectors = bdrv_nb_sectors(bs);
1665 if (total_sectors < 0) {
1666 return total_sectors;
1669 if (sector_num >= total_sectors) {
1674 n = total_sectors - sector_num;
1675 if (n < nb_sectors) {
1679 if (!bs->drv->bdrv_co_get_block_status) {
1681 ret = BDRV_BLOCK_DATA | BDRV_BLOCK_ALLOCATED;
1682 if (bs->drv->protocol_name) {
1683 ret |= BDRV_BLOCK_OFFSET_VALID | (sector_num * BDRV_SECTOR_SIZE);
1689 ret = bs->drv->bdrv_co_get_block_status(bs, sector_num, nb_sectors, pnum,
1696 if (ret & BDRV_BLOCK_RAW) {
1697 assert(ret & BDRV_BLOCK_OFFSET_VALID);
1698 return bdrv_get_block_status(bs->file->bs, ret >> BDRV_SECTOR_BITS,
1702 if (ret & (BDRV_BLOCK_DATA | BDRV_BLOCK_ZERO)) {
1703 ret |= BDRV_BLOCK_ALLOCATED;
1705 if (bdrv_unallocated_blocks_are_zero(bs)) {
1706 ret |= BDRV_BLOCK_ZERO;
1707 } else if (bs->backing) {
1708 BlockDriverState *bs2 = bs->backing->bs;
1709 int64_t nb_sectors2 = bdrv_nb_sectors(bs2);
1710 if (nb_sectors2 >= 0 && sector_num >= nb_sectors2) {
1711 ret |= BDRV_BLOCK_ZERO;
1716 if (*file && *file != bs &&
1717 (ret & BDRV_BLOCK_DATA) && !(ret & BDRV_BLOCK_ZERO) &&
1718 (ret & BDRV_BLOCK_OFFSET_VALID)) {
1719 BlockDriverState *file2;
1722 ret2 = bdrv_co_get_block_status(*file, ret >> BDRV_SECTOR_BITS,
1723 *pnum, &file_pnum, &file2);
1725 /* Ignore errors. This is just providing extra information, it
1726 * is useful but not necessary.
1729 /* !file_pnum indicates an offset at or beyond the EOF; it is
1730 * perfectly valid for the format block driver to point to such
1731 * offsets, so catch it and mark everything as zero */
1732 ret |= BDRV_BLOCK_ZERO;
1734 /* Limit request to the range reported by the protocol driver */
1736 ret |= (ret2 & BDRV_BLOCK_ZERO);
1744 static int64_t coroutine_fn bdrv_co_get_block_status_above(BlockDriverState *bs,
1745 BlockDriverState *base,
1749 BlockDriverState **file)
1751 BlockDriverState *p;
1755 for (p = bs; p != base; p = backing_bs(p)) {
1756 ret = bdrv_co_get_block_status(p, sector_num, nb_sectors, pnum, file);
1757 if (ret < 0 || ret & BDRV_BLOCK_ALLOCATED) {
1760 /* [sector_num, pnum] unallocated on this layer, which could be only
1761 * the first part of [sector_num, nb_sectors]. */
1762 nb_sectors = MIN(nb_sectors, *pnum);
1767 /* Coroutine wrapper for bdrv_get_block_status_above() */
1768 static void coroutine_fn bdrv_get_block_status_above_co_entry(void *opaque)
1770 BdrvCoGetBlockStatusData *data = opaque;
1772 data->ret = bdrv_co_get_block_status_above(data->bs, data->base,
1781 * Synchronous wrapper around bdrv_co_get_block_status_above().
1783 * See bdrv_co_get_block_status_above() for details.
1785 int64_t bdrv_get_block_status_above(BlockDriverState *bs,
1786 BlockDriverState *base,
1788 int nb_sectors, int *pnum,
1789 BlockDriverState **file)
1792 BdrvCoGetBlockStatusData data = {
1796 .sector_num = sector_num,
1797 .nb_sectors = nb_sectors,
1802 if (qemu_in_coroutine()) {
1803 /* Fast-path if already in coroutine context */
1804 bdrv_get_block_status_above_co_entry(&data);
1806 AioContext *aio_context = bdrv_get_aio_context(bs);
1808 co = qemu_coroutine_create(bdrv_get_block_status_above_co_entry,
1810 qemu_coroutine_enter(co);
1811 while (!data.done) {
1812 aio_poll(aio_context, true);
1818 int64_t bdrv_get_block_status(BlockDriverState *bs,
1820 int nb_sectors, int *pnum,
1821 BlockDriverState **file)
1823 return bdrv_get_block_status_above(bs, backing_bs(bs),
1824 sector_num, nb_sectors, pnum, file);
1827 int coroutine_fn bdrv_is_allocated(BlockDriverState *bs, int64_t sector_num,
1828 int nb_sectors, int *pnum)
1830 BlockDriverState *file;
1831 int64_t ret = bdrv_get_block_status(bs, sector_num, nb_sectors, pnum,
1836 return !!(ret & BDRV_BLOCK_ALLOCATED);
1840 * Given an image chain: ... -> [BASE] -> [INTER1] -> [INTER2] -> [TOP]
1842 * Return true if the given sector is allocated in any image between
1843 * BASE and TOP (inclusive). BASE can be NULL to check if the given
1844 * sector is allocated in any image of the chain. Return false otherwise.
1846 * 'pnum' is set to the number of sectors (including and immediately following
1847 * the specified sector) that are known to be in the same
1848 * allocated/unallocated state.
1851 int bdrv_is_allocated_above(BlockDriverState *top,
1852 BlockDriverState *base,
1854 int nb_sectors, int *pnum)
1856 BlockDriverState *intermediate;
1857 int ret, n = nb_sectors;
1860 while (intermediate && intermediate != base) {
1862 ret = bdrv_is_allocated(intermediate, sector_num, nb_sectors,
1872 * [sector_num, nb_sectors] is unallocated on top but intermediate
1875 * [sector_num+x, nr_sectors] allocated.
1877 if (n > pnum_inter &&
1878 (intermediate == top ||
1879 sector_num + pnum_inter < intermediate->total_sectors)) {
1883 intermediate = backing_bs(intermediate);
1890 typedef struct BdrvVmstateCo {
1891 BlockDriverState *bs;
1898 static int coroutine_fn
1899 bdrv_co_rw_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, int64_t pos,
1902 BlockDriver *drv = bs->drv;
1906 } else if (drv->bdrv_load_vmstate) {
1907 return is_read ? drv->bdrv_load_vmstate(bs, qiov, pos)
1908 : drv->bdrv_save_vmstate(bs, qiov, pos);
1909 } else if (bs->file) {
1910 return bdrv_co_rw_vmstate(bs->file->bs, qiov, pos, is_read);
1916 static void coroutine_fn bdrv_co_rw_vmstate_entry(void *opaque)
1918 BdrvVmstateCo *co = opaque;
1919 co->ret = bdrv_co_rw_vmstate(co->bs, co->qiov, co->pos, co->is_read);
1923 bdrv_rw_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, int64_t pos,
1926 if (qemu_in_coroutine()) {
1927 return bdrv_co_rw_vmstate(bs, qiov, pos, is_read);
1929 BdrvVmstateCo data = {
1934 .ret = -EINPROGRESS,
1936 Coroutine *co = qemu_coroutine_create(bdrv_co_rw_vmstate_entry, &data);
1938 qemu_coroutine_enter(co);
1939 while (data.ret == -EINPROGRESS) {
1940 aio_poll(bdrv_get_aio_context(bs), true);
1946 int bdrv_save_vmstate(BlockDriverState *bs, const uint8_t *buf,
1947 int64_t pos, int size)
1950 struct iovec iov = {
1951 .iov_base = (void *) buf,
1956 qemu_iovec_init_external(&qiov, &iov, 1);
1958 ret = bdrv_writev_vmstate(bs, &qiov, pos);
1966 int bdrv_writev_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, int64_t pos)
1968 return bdrv_rw_vmstate(bs, qiov, pos, false);
1971 int bdrv_load_vmstate(BlockDriverState *bs, uint8_t *buf,
1972 int64_t pos, int size)
1975 struct iovec iov = {
1981 qemu_iovec_init_external(&qiov, &iov, 1);
1982 ret = bdrv_readv_vmstate(bs, &qiov, pos);
1990 int bdrv_readv_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, int64_t pos)
1992 return bdrv_rw_vmstate(bs, qiov, pos, true);
1995 /**************************************************************/
1998 BlockAIOCB *bdrv_aio_readv(BdrvChild *child, int64_t sector_num,
1999 QEMUIOVector *qiov, int nb_sectors,
2000 BlockCompletionFunc *cb, void *opaque)
2002 trace_bdrv_aio_readv(child->bs, sector_num, nb_sectors, opaque);
2004 assert(nb_sectors << BDRV_SECTOR_BITS == qiov->size);
2005 return bdrv_co_aio_prw_vector(child, sector_num << BDRV_SECTOR_BITS, qiov,
2006 0, cb, opaque, false);
2009 BlockAIOCB *bdrv_aio_writev(BdrvChild *child, int64_t sector_num,
2010 QEMUIOVector *qiov, int nb_sectors,
2011 BlockCompletionFunc *cb, void *opaque)
2013 trace_bdrv_aio_writev(child->bs, sector_num, nb_sectors, opaque);
2015 assert(nb_sectors << BDRV_SECTOR_BITS == qiov->size);
2016 return bdrv_co_aio_prw_vector(child, sector_num << BDRV_SECTOR_BITS, qiov,
2017 0, cb, opaque, true);
2020 void bdrv_aio_cancel(BlockAIOCB *acb)
2023 bdrv_aio_cancel_async(acb);
2024 while (acb->refcnt > 1) {
2025 if (acb->aiocb_info->get_aio_context) {
2026 aio_poll(acb->aiocb_info->get_aio_context(acb), true);
2027 } else if (acb->bs) {
2028 aio_poll(bdrv_get_aio_context(acb->bs), true);
2033 qemu_aio_unref(acb);
2036 /* Async version of aio cancel. The caller is not blocked if the acb implements
2037 * cancel_async, otherwise we do nothing and let the request normally complete.
2038 * In either case the completion callback must be called. */
2039 void bdrv_aio_cancel_async(BlockAIOCB *acb)
2041 if (acb->aiocb_info->cancel_async) {
2042 acb->aiocb_info->cancel_async(acb);
2046 /**************************************************************/
2047 /* async block device emulation */
2049 typedef struct BlockRequest {
2051 /* Used during read, write, trim */
2058 /* Used during ioctl */
2064 BlockCompletionFunc *cb;
2070 typedef struct BlockAIOCBCoroutine {
2078 } BlockAIOCBCoroutine;
2080 static const AIOCBInfo bdrv_em_co_aiocb_info = {
2081 .aiocb_size = sizeof(BlockAIOCBCoroutine),
2084 static void bdrv_co_complete(BlockAIOCBCoroutine *acb)
2086 if (!acb->need_bh) {
2087 acb->common.cb(acb->common.opaque, acb->req.error);
2088 qemu_aio_unref(acb);
2092 static void bdrv_co_em_bh(void *opaque)
2094 BlockAIOCBCoroutine *acb = opaque;
2096 assert(!acb->need_bh);
2097 qemu_bh_delete(acb->bh);
2098 bdrv_co_complete(acb);
2101 static void bdrv_co_maybe_schedule_bh(BlockAIOCBCoroutine *acb)
2103 acb->need_bh = false;
2104 if (acb->req.error != -EINPROGRESS) {
2105 BlockDriverState *bs = acb->common.bs;
2107 acb->bh = aio_bh_new(bdrv_get_aio_context(bs), bdrv_co_em_bh, acb);
2108 qemu_bh_schedule(acb->bh);
2112 /* Invoke bdrv_co_do_readv/bdrv_co_do_writev */
2113 static void coroutine_fn bdrv_co_do_rw(void *opaque)
2115 BlockAIOCBCoroutine *acb = opaque;
2117 if (!acb->is_write) {
2118 acb->req.error = bdrv_co_preadv(acb->child, acb->req.offset,
2119 acb->req.qiov->size, acb->req.qiov, acb->req.flags);
2121 acb->req.error = bdrv_co_pwritev(acb->child, acb->req.offset,
2122 acb->req.qiov->size, acb->req.qiov, acb->req.flags);
2125 bdrv_co_complete(acb);
2128 static BlockAIOCB *bdrv_co_aio_prw_vector(BdrvChild *child,
2131 BdrvRequestFlags flags,
2132 BlockCompletionFunc *cb,
2137 BlockAIOCBCoroutine *acb;
2139 acb = qemu_aio_get(&bdrv_em_co_aiocb_info, child->bs, cb, opaque);
2141 acb->need_bh = true;
2142 acb->req.error = -EINPROGRESS;
2143 acb->req.offset = offset;
2144 acb->req.qiov = qiov;
2145 acb->req.flags = flags;
2146 acb->is_write = is_write;
2148 co = qemu_coroutine_create(bdrv_co_do_rw, acb);
2149 qemu_coroutine_enter(co);
2151 bdrv_co_maybe_schedule_bh(acb);
2152 return &acb->common;
2155 static void coroutine_fn bdrv_aio_flush_co_entry(void *opaque)
2157 BlockAIOCBCoroutine *acb = opaque;
2158 BlockDriverState *bs = acb->common.bs;
2160 acb->req.error = bdrv_co_flush(bs);
2161 bdrv_co_complete(acb);
2164 BlockAIOCB *bdrv_aio_flush(BlockDriverState *bs,
2165 BlockCompletionFunc *cb, void *opaque)
2167 trace_bdrv_aio_flush(bs, opaque);
2170 BlockAIOCBCoroutine *acb;
2172 acb = qemu_aio_get(&bdrv_em_co_aiocb_info, bs, cb, opaque);
2173 acb->need_bh = true;
2174 acb->req.error = -EINPROGRESS;
2176 co = qemu_coroutine_create(bdrv_aio_flush_co_entry, acb);
2177 qemu_coroutine_enter(co);
2179 bdrv_co_maybe_schedule_bh(acb);
2180 return &acb->common;
2183 static void coroutine_fn bdrv_aio_pdiscard_co_entry(void *opaque)
2185 BlockAIOCBCoroutine *acb = opaque;
2186 BlockDriverState *bs = acb->common.bs;
2188 acb->req.error = bdrv_co_pdiscard(bs, acb->req.offset, acb->req.bytes);
2189 bdrv_co_complete(acb);
2192 BlockAIOCB *bdrv_aio_pdiscard(BlockDriverState *bs, int64_t offset, int count,
2193 BlockCompletionFunc *cb, void *opaque)
2196 BlockAIOCBCoroutine *acb;
2198 trace_bdrv_aio_pdiscard(bs, offset, count, opaque);
2200 acb = qemu_aio_get(&bdrv_em_co_aiocb_info, bs, cb, opaque);
2201 acb->need_bh = true;
2202 acb->req.error = -EINPROGRESS;
2203 acb->req.offset = offset;
2204 acb->req.bytes = count;
2205 co = qemu_coroutine_create(bdrv_aio_pdiscard_co_entry, acb);
2206 qemu_coroutine_enter(co);
2208 bdrv_co_maybe_schedule_bh(acb);
2209 return &acb->common;
2212 void *qemu_aio_get(const AIOCBInfo *aiocb_info, BlockDriverState *bs,
2213 BlockCompletionFunc *cb, void *opaque)
2217 acb = g_malloc(aiocb_info->aiocb_size);
2218 acb->aiocb_info = aiocb_info;
2221 acb->opaque = opaque;
2226 void qemu_aio_ref(void *p)
2228 BlockAIOCB *acb = p;
2232 void qemu_aio_unref(void *p)
2234 BlockAIOCB *acb = p;
2235 assert(acb->refcnt > 0);
2236 if (--acb->refcnt == 0) {
2241 /**************************************************************/
2242 /* Coroutine block device emulation */
2244 typedef struct FlushCo {
2245 BlockDriverState *bs;
2250 static void coroutine_fn bdrv_flush_co_entry(void *opaque)
2252 FlushCo *rwco = opaque;
2254 rwco->ret = bdrv_co_flush(rwco->bs);
2257 int coroutine_fn bdrv_co_flush(BlockDriverState *bs)
2260 BdrvTrackedRequest req;
2262 if (!bs || !bdrv_is_inserted(bs) || bdrv_is_read_only(bs) ||
2267 tracked_request_begin(&req, bs, 0, 0, BDRV_TRACKED_FLUSH);
2269 int current_gen = bs->write_gen;
2271 /* Wait until any previous flushes are completed */
2272 while (bs->active_flush_req != NULL) {
2273 qemu_co_queue_wait(&bs->flush_queue);
2276 bs->active_flush_req = &req;
2278 /* Write back all layers by calling one driver function */
2279 if (bs->drv->bdrv_co_flush) {
2280 ret = bs->drv->bdrv_co_flush(bs);
2284 /* Write back cached data to the OS even with cache=unsafe */
2285 BLKDBG_EVENT(bs->file, BLKDBG_FLUSH_TO_OS);
2286 if (bs->drv->bdrv_co_flush_to_os) {
2287 ret = bs->drv->bdrv_co_flush_to_os(bs);
2293 /* But don't actually force it to the disk with cache=unsafe */
2294 if (bs->open_flags & BDRV_O_NO_FLUSH) {
2298 /* Check if we really need to flush anything */
2299 if (bs->flushed_gen == current_gen) {
2303 BLKDBG_EVENT(bs->file, BLKDBG_FLUSH_TO_DISK);
2304 if (bs->drv->bdrv_co_flush_to_disk) {
2305 ret = bs->drv->bdrv_co_flush_to_disk(bs);
2306 } else if (bs->drv->bdrv_aio_flush) {
2308 CoroutineIOCompletion co = {
2309 .coroutine = qemu_coroutine_self(),
2312 acb = bs->drv->bdrv_aio_flush(bs, bdrv_co_io_em_complete, &co);
2316 qemu_coroutine_yield();
2321 * Some block drivers always operate in either writethrough or unsafe
2322 * mode and don't support bdrv_flush therefore. Usually qemu doesn't
2323 * know how the server works (because the behaviour is hardcoded or
2324 * depends on server-side configuration), so we can't ensure that
2325 * everything is safe on disk. Returning an error doesn't work because
2326 * that would break guests even if the server operates in writethrough
2329 * Let's hope the user knows what he's doing.
2338 /* Now flush the underlying protocol. It will also have BDRV_O_NO_FLUSH
2339 * in the case of cache=unsafe, so there are no useless flushes.
2342 ret = bs->file ? bdrv_co_flush(bs->file->bs) : 0;
2344 /* Notify any pending flushes that we have completed */
2345 bs->flushed_gen = current_gen;
2346 bs->active_flush_req = NULL;
2347 /* Return value is ignored - it's ok if wait queue is empty */
2348 qemu_co_queue_next(&bs->flush_queue);
2350 tracked_request_end(&req);
2354 int bdrv_flush(BlockDriverState *bs)
2357 FlushCo flush_co = {
2362 if (qemu_in_coroutine()) {
2363 /* Fast-path if already in coroutine context */
2364 bdrv_flush_co_entry(&flush_co);
2366 AioContext *aio_context = bdrv_get_aio_context(bs);
2368 co = qemu_coroutine_create(bdrv_flush_co_entry, &flush_co);
2369 qemu_coroutine_enter(co);
2370 while (flush_co.ret == NOT_DONE) {
2371 aio_poll(aio_context, true);
2375 return flush_co.ret;
2378 typedef struct DiscardCo {
2379 BlockDriverState *bs;
2384 static void coroutine_fn bdrv_pdiscard_co_entry(void *opaque)
2386 DiscardCo *rwco = opaque;
2388 rwco->ret = bdrv_co_pdiscard(rwco->bs, rwco->offset, rwco->count);
2391 int coroutine_fn bdrv_co_pdiscard(BlockDriverState *bs, int64_t offset,
2394 BdrvTrackedRequest req;
2395 int max_pdiscard, ret;
2402 ret = bdrv_check_byte_request(bs, offset, count);
2405 } else if (bs->read_only) {
2408 assert(!(bs->open_flags & BDRV_O_INACTIVE));
2410 /* Do nothing if disabled. */
2411 if (!(bs->open_flags & BDRV_O_UNMAP)) {
2415 if (!bs->drv->bdrv_co_pdiscard && !bs->drv->bdrv_aio_pdiscard) {
2419 /* Discard is advisory, so ignore any unaligned head or tail */
2420 align = MAX(bs->bl.pdiscard_alignment, bs->bl.request_alignment);
2421 assert(align % bs->bl.request_alignment == 0);
2422 head = offset % align;
2424 head = MIN(count, align - head);
2428 count = QEMU_ALIGN_DOWN(count, align);
2433 tracked_request_begin(&req, bs, offset, count, BDRV_TRACKED_DISCARD);
2435 ret = notifier_with_return_list_notify(&bs->before_write_notifiers, &req);
2440 max_pdiscard = QEMU_ALIGN_DOWN(MIN_NON_ZERO(bs->bl.max_pdiscard, INT_MAX),
2442 assert(max_pdiscard);
2446 int num = MIN(count, max_pdiscard);
2448 if (bs->drv->bdrv_co_pdiscard) {
2449 ret = bs->drv->bdrv_co_pdiscard(bs, offset, num);
2452 CoroutineIOCompletion co = {
2453 .coroutine = qemu_coroutine_self(),
2456 acb = bs->drv->bdrv_aio_pdiscard(bs, offset, num,
2457 bdrv_co_io_em_complete, &co);
2462 qemu_coroutine_yield();
2466 if (ret && ret != -ENOTSUP) {
2476 bdrv_set_dirty(bs, req.offset >> BDRV_SECTOR_BITS,
2477 req.bytes >> BDRV_SECTOR_BITS);
2478 tracked_request_end(&req);
2482 int bdrv_pdiscard(BlockDriverState *bs, int64_t offset, int count)
2492 if (qemu_in_coroutine()) {
2493 /* Fast-path if already in coroutine context */
2494 bdrv_pdiscard_co_entry(&rwco);
2496 AioContext *aio_context = bdrv_get_aio_context(bs);
2498 co = qemu_coroutine_create(bdrv_pdiscard_co_entry, &rwco);
2499 qemu_coroutine_enter(co);
2500 while (rwco.ret == NOT_DONE) {
2501 aio_poll(aio_context, true);
2508 static int bdrv_co_do_ioctl(BlockDriverState *bs, int req, void *buf)
2510 BlockDriver *drv = bs->drv;
2511 BdrvTrackedRequest tracked_req;
2512 CoroutineIOCompletion co = {
2513 .coroutine = qemu_coroutine_self(),
2517 tracked_request_begin(&tracked_req, bs, 0, 0, BDRV_TRACKED_IOCTL);
2518 if (!drv || !drv->bdrv_aio_ioctl) {
2523 acb = drv->bdrv_aio_ioctl(bs, req, buf, bdrv_co_io_em_complete, &co);
2528 qemu_coroutine_yield();
2530 tracked_request_end(&tracked_req);
2535 BlockDriverState *bs;
2541 static void coroutine_fn bdrv_co_ioctl_entry(void *opaque)
2543 BdrvIoctlCoData *data = opaque;
2544 data->ret = bdrv_co_do_ioctl(data->bs, data->req, data->buf);
2547 /* needed for generic scsi interface */
2548 int bdrv_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
2550 BdrvIoctlCoData data = {
2554 .ret = -EINPROGRESS,
2557 if (qemu_in_coroutine()) {
2558 /* Fast-path if already in coroutine context */
2559 bdrv_co_ioctl_entry(&data);
2561 Coroutine *co = qemu_coroutine_create(bdrv_co_ioctl_entry, &data);
2563 qemu_coroutine_enter(co);
2564 while (data.ret == -EINPROGRESS) {
2565 aio_poll(bdrv_get_aio_context(bs), true);
2571 static void coroutine_fn bdrv_co_aio_ioctl_entry(void *opaque)
2573 BlockAIOCBCoroutine *acb = opaque;
2574 acb->req.error = bdrv_co_do_ioctl(acb->common.bs,
2575 acb->req.req, acb->req.buf);
2576 bdrv_co_complete(acb);
2579 BlockAIOCB *bdrv_aio_ioctl(BlockDriverState *bs,
2580 unsigned long int req, void *buf,
2581 BlockCompletionFunc *cb, void *opaque)
2583 BlockAIOCBCoroutine *acb = qemu_aio_get(&bdrv_em_co_aiocb_info,
2587 acb->need_bh = true;
2588 acb->req.error = -EINPROGRESS;
2591 co = qemu_coroutine_create(bdrv_co_aio_ioctl_entry, acb);
2592 qemu_coroutine_enter(co);
2594 bdrv_co_maybe_schedule_bh(acb);
2595 return &acb->common;
2598 void *qemu_blockalign(BlockDriverState *bs, size_t size)
2600 return qemu_memalign(bdrv_opt_mem_align(bs), size);
2603 void *qemu_blockalign0(BlockDriverState *bs, size_t size)
2605 return memset(qemu_blockalign(bs, size), 0, size);
2608 void *qemu_try_blockalign(BlockDriverState *bs, size_t size)
2610 size_t align = bdrv_opt_mem_align(bs);
2612 /* Ensure that NULL is never returned on success */
2618 return qemu_try_memalign(align, size);
2621 void *qemu_try_blockalign0(BlockDriverState *bs, size_t size)
2623 void *mem = qemu_try_blockalign(bs, size);
2626 memset(mem, 0, size);
2633 * Check if all memory in this vector is sector aligned.
2635 bool bdrv_qiov_is_aligned(BlockDriverState *bs, QEMUIOVector *qiov)
2638 size_t alignment = bdrv_min_mem_align(bs);
2640 for (i = 0; i < qiov->niov; i++) {
2641 if ((uintptr_t) qiov->iov[i].iov_base % alignment) {
2644 if (qiov->iov[i].iov_len % alignment) {
2652 void bdrv_add_before_write_notifier(BlockDriverState *bs,
2653 NotifierWithReturn *notifier)
2655 notifier_with_return_list_add(&bs->before_write_notifiers, notifier);
2658 void bdrv_io_plug(BlockDriverState *bs)
2662 QLIST_FOREACH(child, &bs->children, next) {
2663 bdrv_io_plug(child->bs);
2666 if (bs->io_plugged++ == 0 && bs->io_plug_disabled == 0) {
2667 BlockDriver *drv = bs->drv;
2668 if (drv && drv->bdrv_io_plug) {
2669 drv->bdrv_io_plug(bs);
2674 void bdrv_io_unplug(BlockDriverState *bs)
2678 assert(bs->io_plugged);
2679 if (--bs->io_plugged == 0 && bs->io_plug_disabled == 0) {
2680 BlockDriver *drv = bs->drv;
2681 if (drv && drv->bdrv_io_unplug) {
2682 drv->bdrv_io_unplug(bs);
2686 QLIST_FOREACH(child, &bs->children, next) {
2687 bdrv_io_unplug(child->bs);
2691 void bdrv_io_unplugged_begin(BlockDriverState *bs)
2695 if (bs->io_plug_disabled++ == 0 && bs->io_plugged > 0) {
2696 BlockDriver *drv = bs->drv;
2697 if (drv && drv->bdrv_io_unplug) {
2698 drv->bdrv_io_unplug(bs);
2702 QLIST_FOREACH(child, &bs->children, next) {
2703 bdrv_io_unplugged_begin(child->bs);
2707 void bdrv_io_unplugged_end(BlockDriverState *bs)
2711 assert(bs->io_plug_disabled);
2712 QLIST_FOREACH(child, &bs->children, next) {
2713 bdrv_io_unplugged_end(child->bs);
2716 if (--bs->io_plug_disabled == 0 && bs->io_plugged > 0) {
2717 BlockDriver *drv = bs->drv;
2718 if (drv && drv->bdrv_io_plug) {
2719 drv->bdrv_io_plug(bs);