2 * Block layer I/O functions
4 * Copyright (c) 2003 Fabrice Bellard
6 * Permission is hereby granted, free of charge, to any person obtaining a copy
7 * of this software and associated documentation files (the "Software"), to deal
8 * in the Software without restriction, including without limitation the rights
9 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10 * copies of the Software, and to permit persons to whom the Software is
11 * furnished to do so, subject to the following conditions:
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
26 #include "sysemu/block-backend.h"
27 #include "block/blockjob.h"
28 #include "block/block_int.h"
29 #include "block/throttle-groups.h"
30 #include "qemu/error-report.h"
32 #define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */
34 static BlockAIOCB *bdrv_aio_readv_em(BlockDriverState *bs,
35 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
36 BlockCompletionFunc *cb, void *opaque);
37 static BlockAIOCB *bdrv_aio_writev_em(BlockDriverState *bs,
38 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
39 BlockCompletionFunc *cb, void *opaque);
40 static int coroutine_fn bdrv_co_readv_em(BlockDriverState *bs,
41 int64_t sector_num, int nb_sectors,
43 static int coroutine_fn bdrv_co_writev_em(BlockDriverState *bs,
44 int64_t sector_num, int nb_sectors,
46 static int coroutine_fn bdrv_co_do_preadv(BlockDriverState *bs,
47 int64_t offset, unsigned int bytes, QEMUIOVector *qiov,
48 BdrvRequestFlags flags);
49 static int coroutine_fn bdrv_co_do_pwritev(BlockDriverState *bs,
50 int64_t offset, unsigned int bytes, QEMUIOVector *qiov,
51 BdrvRequestFlags flags);
52 static BlockAIOCB *bdrv_co_aio_rw_vector(BlockDriverState *bs,
56 BdrvRequestFlags flags,
57 BlockCompletionFunc *cb,
60 static void coroutine_fn bdrv_co_do_rw(void *opaque);
61 static int coroutine_fn bdrv_co_do_write_zeroes(BlockDriverState *bs,
62 int64_t sector_num, int nb_sectors, BdrvRequestFlags flags);
64 /* throttling disk I/O limits */
65 void bdrv_set_io_limits(BlockDriverState *bs,
70 throttle_group_config(bs, cfg);
72 for (i = 0; i < 2; i++) {
73 qemu_co_enter_next(&bs->throttled_reqs[i]);
77 /* this function drain all the throttled IOs */
78 static bool bdrv_start_throttled_reqs(BlockDriverState *bs)
81 bool enabled = bs->io_limits_enabled;
84 bs->io_limits_enabled = false;
86 for (i = 0; i < 2; i++) {
87 while (qemu_co_enter_next(&bs->throttled_reqs[i])) {
92 bs->io_limits_enabled = enabled;
97 void bdrv_io_limits_disable(BlockDriverState *bs)
99 bs->io_limits_enabled = false;
100 bdrv_start_throttled_reqs(bs);
101 throttle_group_unregister_bs(bs);
104 /* should be called before bdrv_set_io_limits if a limit is set */
105 void bdrv_io_limits_enable(BlockDriverState *bs, const char *group)
107 assert(!bs->io_limits_enabled);
108 throttle_group_register_bs(bs, group);
109 bs->io_limits_enabled = true;
112 void bdrv_io_limits_update_group(BlockDriverState *bs, const char *group)
114 /* this bs is not part of any group */
115 if (!bs->throttle_state) {
119 /* this bs is a part of the same group than the one we want */
120 if (!g_strcmp0(throttle_group_get_name(bs), group)) {
124 /* need to change the group this bs belong to */
125 bdrv_io_limits_disable(bs);
126 bdrv_io_limits_enable(bs, group);
129 void bdrv_setup_io_funcs(BlockDriver *bdrv)
131 /* Block drivers without coroutine functions need emulation */
132 if (!bdrv->bdrv_co_readv) {
133 bdrv->bdrv_co_readv = bdrv_co_readv_em;
134 bdrv->bdrv_co_writev = bdrv_co_writev_em;
136 /* bdrv_co_readv_em()/brdv_co_writev_em() work in terms of aio, so if
137 * the block driver lacks aio we need to emulate that too.
139 if (!bdrv->bdrv_aio_readv) {
140 /* add AIO emulation layer */
141 bdrv->bdrv_aio_readv = bdrv_aio_readv_em;
142 bdrv->bdrv_aio_writev = bdrv_aio_writev_em;
147 void bdrv_refresh_limits(BlockDriverState *bs, Error **errp)
149 BlockDriver *drv = bs->drv;
150 Error *local_err = NULL;
152 memset(&bs->bl, 0, sizeof(bs->bl));
158 /* Take some limits from the children as a default */
160 bdrv_refresh_limits(bs->file->bs, &local_err);
162 error_propagate(errp, local_err);
165 bs->bl.opt_transfer_length = bs->file->bs->bl.opt_transfer_length;
166 bs->bl.max_transfer_length = bs->file->bs->bl.max_transfer_length;
167 bs->bl.min_mem_alignment = bs->file->bs->bl.min_mem_alignment;
168 bs->bl.opt_mem_alignment = bs->file->bs->bl.opt_mem_alignment;
169 bs->bl.max_iov = bs->file->bs->bl.max_iov;
171 bs->bl.min_mem_alignment = 512;
172 bs->bl.opt_mem_alignment = getpagesize();
174 /* Safe default since most protocols use readv()/writev()/etc */
175 bs->bl.max_iov = IOV_MAX;
179 bdrv_refresh_limits(bs->backing->bs, &local_err);
181 error_propagate(errp, local_err);
184 bs->bl.opt_transfer_length =
185 MAX(bs->bl.opt_transfer_length,
186 bs->backing->bs->bl.opt_transfer_length);
187 bs->bl.max_transfer_length =
188 MIN_NON_ZERO(bs->bl.max_transfer_length,
189 bs->backing->bs->bl.max_transfer_length);
190 bs->bl.opt_mem_alignment =
191 MAX(bs->bl.opt_mem_alignment,
192 bs->backing->bs->bl.opt_mem_alignment);
193 bs->bl.min_mem_alignment =
194 MAX(bs->bl.min_mem_alignment,
195 bs->backing->bs->bl.min_mem_alignment);
198 bs->backing->bs->bl.max_iov);
201 /* Then let the driver override it */
202 if (drv->bdrv_refresh_limits) {
203 drv->bdrv_refresh_limits(bs, errp);
208 * The copy-on-read flag is actually a reference count so multiple users may
209 * use the feature without worrying about clobbering its previous state.
210 * Copy-on-read stays enabled until all users have called to disable it.
212 void bdrv_enable_copy_on_read(BlockDriverState *bs)
217 void bdrv_disable_copy_on_read(BlockDriverState *bs)
219 assert(bs->copy_on_read > 0);
223 /* Check if any requests are in-flight (including throttled requests) */
224 bool bdrv_requests_pending(BlockDriverState *bs)
228 if (!QLIST_EMPTY(&bs->tracked_requests)) {
231 if (!qemu_co_queue_empty(&bs->throttled_reqs[0])) {
234 if (!qemu_co_queue_empty(&bs->throttled_reqs[1])) {
238 QLIST_FOREACH(child, &bs->children, next) {
239 if (bdrv_requests_pending(child->bs)) {
247 static void bdrv_drain_recurse(BlockDriverState *bs)
251 if (bs->drv && bs->drv->bdrv_drain) {
252 bs->drv->bdrv_drain(bs);
254 QLIST_FOREACH(child, &bs->children, next) {
255 bdrv_drain_recurse(child->bs);
260 * Wait for pending requests to complete on a single BlockDriverState subtree,
261 * and suspend block driver's internal I/O until next request arrives.
263 * Note that unlike bdrv_drain_all(), the caller must hold the BlockDriverState
266 * Only this BlockDriverState's AioContext is run, so in-flight requests must
267 * not depend on events in other AioContexts. In that case, use
268 * bdrv_drain_all() instead.
270 void bdrv_drain(BlockDriverState *bs)
274 bdrv_drain_recurse(bs);
277 bdrv_flush_io_queue(bs);
278 busy = bdrv_requests_pending(bs);
279 busy |= aio_poll(bdrv_get_aio_context(bs), busy);
284 * Wait for pending requests to complete across all BlockDriverStates
286 * This function does not flush data to disk, use bdrv_flush_all() for that
287 * after calling this function.
289 void bdrv_drain_all(void)
291 /* Always run first iteration so any pending completion BHs run */
293 BlockDriverState *bs = NULL;
294 GSList *aio_ctxs = NULL, *ctx;
296 while ((bs = bdrv_next(bs))) {
297 AioContext *aio_context = bdrv_get_aio_context(bs);
299 aio_context_acquire(aio_context);
301 block_job_pause(bs->job);
303 aio_context_release(aio_context);
305 if (!g_slist_find(aio_ctxs, aio_context)) {
306 aio_ctxs = g_slist_prepend(aio_ctxs, aio_context);
310 /* Note that completion of an asynchronous I/O operation can trigger any
311 * number of other I/O operations on other devices---for example a
312 * coroutine can submit an I/O request to another device in response to
313 * request completion. Therefore we must keep looping until there was no
314 * more activity rather than simply draining each device independently.
319 for (ctx = aio_ctxs; ctx != NULL; ctx = ctx->next) {
320 AioContext *aio_context = ctx->data;
323 aio_context_acquire(aio_context);
324 while ((bs = bdrv_next(bs))) {
325 if (aio_context == bdrv_get_aio_context(bs)) {
326 bdrv_flush_io_queue(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);
339 while ((bs = bdrv_next(bs))) {
340 AioContext *aio_context = bdrv_get_aio_context(bs);
342 aio_context_acquire(aio_context);
344 block_job_resume(bs->job);
346 aio_context_release(aio_context);
348 g_slist_free(aio_ctxs);
352 * Remove an active request from the tracked requests list
354 * This function should be called when a tracked request is completing.
356 static void tracked_request_end(BdrvTrackedRequest *req)
358 if (req->serialising) {
359 req->bs->serialising_in_flight--;
362 QLIST_REMOVE(req, list);
363 qemu_co_queue_restart_all(&req->wait_queue);
367 * Add an active request to the tracked requests list
369 static void tracked_request_begin(BdrvTrackedRequest *req,
370 BlockDriverState *bs,
373 enum BdrvTrackedRequestType type)
375 *req = (BdrvTrackedRequest){
380 .co = qemu_coroutine_self(),
381 .serialising = false,
382 .overlap_offset = offset,
383 .overlap_bytes = bytes,
386 qemu_co_queue_init(&req->wait_queue);
388 QLIST_INSERT_HEAD(&bs->tracked_requests, req, list);
391 static void mark_request_serialising(BdrvTrackedRequest *req, uint64_t align)
393 int64_t overlap_offset = req->offset & ~(align - 1);
394 unsigned int overlap_bytes = ROUND_UP(req->offset + req->bytes, align)
397 if (!req->serialising) {
398 req->bs->serialising_in_flight++;
399 req->serialising = true;
402 req->overlap_offset = MIN(req->overlap_offset, overlap_offset);
403 req->overlap_bytes = MAX(req->overlap_bytes, overlap_bytes);
407 * Round a region to cluster boundaries
409 void bdrv_round_to_clusters(BlockDriverState *bs,
410 int64_t sector_num, int nb_sectors,
411 int64_t *cluster_sector_num,
412 int *cluster_nb_sectors)
416 if (bdrv_get_info(bs, &bdi) < 0 || bdi.cluster_size == 0) {
417 *cluster_sector_num = sector_num;
418 *cluster_nb_sectors = nb_sectors;
420 int64_t c = bdi.cluster_size / BDRV_SECTOR_SIZE;
421 *cluster_sector_num = QEMU_ALIGN_DOWN(sector_num, c);
422 *cluster_nb_sectors = QEMU_ALIGN_UP(sector_num - *cluster_sector_num +
427 static int bdrv_get_cluster_size(BlockDriverState *bs)
432 ret = bdrv_get_info(bs, &bdi);
433 if (ret < 0 || bdi.cluster_size == 0) {
434 return bs->request_alignment;
436 return bdi.cluster_size;
440 static bool tracked_request_overlaps(BdrvTrackedRequest *req,
441 int64_t offset, unsigned int bytes)
444 if (offset >= req->overlap_offset + req->overlap_bytes) {
448 if (req->overlap_offset >= offset + bytes) {
454 static bool coroutine_fn wait_serialising_requests(BdrvTrackedRequest *self)
456 BlockDriverState *bs = self->bs;
457 BdrvTrackedRequest *req;
461 if (!bs->serialising_in_flight) {
467 QLIST_FOREACH(req, &bs->tracked_requests, list) {
468 if (req == self || (!req->serialising && !self->serialising)) {
471 if (tracked_request_overlaps(req, self->overlap_offset,
472 self->overlap_bytes))
474 /* Hitting this means there was a reentrant request, for
475 * example, a block driver issuing nested requests. This must
476 * never happen since it means deadlock.
478 assert(qemu_coroutine_self() != req->co);
480 /* If the request is already (indirectly) waiting for us, or
481 * will wait for us as soon as it wakes up, then just go on
482 * (instead of producing a deadlock in the former case). */
483 if (!req->waiting_for) {
484 self->waiting_for = req;
485 qemu_co_queue_wait(&req->wait_queue);
486 self->waiting_for = NULL;
498 static int bdrv_check_byte_request(BlockDriverState *bs, int64_t offset,
501 if (size > BDRV_REQUEST_MAX_SECTORS << BDRV_SECTOR_BITS) {
505 if (!bdrv_is_inserted(bs)) {
516 static int bdrv_check_request(BlockDriverState *bs, int64_t sector_num,
519 if (nb_sectors < 0 || nb_sectors > BDRV_REQUEST_MAX_SECTORS) {
523 return bdrv_check_byte_request(bs, sector_num * BDRV_SECTOR_SIZE,
524 nb_sectors * BDRV_SECTOR_SIZE);
527 typedef struct RwCo {
528 BlockDriverState *bs;
533 BdrvRequestFlags flags;
536 static void coroutine_fn bdrv_rw_co_entry(void *opaque)
540 if (!rwco->is_write) {
541 rwco->ret = bdrv_co_do_preadv(rwco->bs, rwco->offset,
542 rwco->qiov->size, rwco->qiov,
545 rwco->ret = bdrv_co_do_pwritev(rwco->bs, rwco->offset,
546 rwco->qiov->size, rwco->qiov,
552 * Process a vectored synchronous request using coroutines
554 static int bdrv_prwv_co(BlockDriverState *bs, int64_t offset,
555 QEMUIOVector *qiov, bool is_write,
556 BdrvRequestFlags flags)
563 .is_write = is_write,
569 * In sync call context, when the vcpu is blocked, this throttling timer
570 * will not fire; so the I/O throttling function has to be disabled here
571 * if it has been enabled.
573 if (bs->io_limits_enabled) {
574 fprintf(stderr, "Disabling I/O throttling on '%s' due "
575 "to synchronous I/O.\n", bdrv_get_device_name(bs));
576 bdrv_io_limits_disable(bs);
579 if (qemu_in_coroutine()) {
580 /* Fast-path if already in coroutine context */
581 bdrv_rw_co_entry(&rwco);
583 AioContext *aio_context = bdrv_get_aio_context(bs);
585 co = qemu_coroutine_create(bdrv_rw_co_entry);
586 qemu_coroutine_enter(co, &rwco);
587 while (rwco.ret == NOT_DONE) {
588 aio_poll(aio_context, true);
595 * Process a synchronous request using coroutines
597 static int bdrv_rw_co(BlockDriverState *bs, int64_t sector_num, uint8_t *buf,
598 int nb_sectors, bool is_write, BdrvRequestFlags flags)
602 .iov_base = (void *)buf,
603 .iov_len = nb_sectors * BDRV_SECTOR_SIZE,
606 if (nb_sectors < 0 || nb_sectors > BDRV_REQUEST_MAX_SECTORS) {
610 qemu_iovec_init_external(&qiov, &iov, 1);
611 return bdrv_prwv_co(bs, sector_num << BDRV_SECTOR_BITS,
612 &qiov, is_write, flags);
615 /* return < 0 if error. See bdrv_write() for the return codes */
616 int bdrv_read(BlockDriverState *bs, int64_t sector_num,
617 uint8_t *buf, int nb_sectors)
619 return bdrv_rw_co(bs, sector_num, buf, nb_sectors, false, 0);
622 /* Just like bdrv_read(), but with I/O throttling temporarily disabled */
623 int bdrv_read_unthrottled(BlockDriverState *bs, int64_t sector_num,
624 uint8_t *buf, int nb_sectors)
629 enabled = bs->io_limits_enabled;
630 bs->io_limits_enabled = false;
631 ret = bdrv_read(bs, sector_num, buf, nb_sectors);
632 bs->io_limits_enabled = enabled;
636 /* Return < 0 if error. Important errors are:
637 -EIO generic I/O error (may happen for all errors)
638 -ENOMEDIUM No media inserted.
639 -EINVAL Invalid sector number or nb_sectors
640 -EACCES Trying to write a read-only device
642 int bdrv_write(BlockDriverState *bs, int64_t sector_num,
643 const uint8_t *buf, int nb_sectors)
645 return bdrv_rw_co(bs, sector_num, (uint8_t *)buf, nb_sectors, true, 0);
648 int bdrv_write_zeroes(BlockDriverState *bs, int64_t sector_num,
649 int nb_sectors, BdrvRequestFlags flags)
651 return bdrv_rw_co(bs, sector_num, NULL, nb_sectors, true,
652 BDRV_REQ_ZERO_WRITE | flags);
656 * Completely zero out a block device with the help of bdrv_write_zeroes.
657 * The operation is sped up by checking the block status and only writing
658 * zeroes to the device if they currently do not return zeroes. Optional
659 * flags are passed through to bdrv_write_zeroes (e.g. BDRV_REQ_MAY_UNMAP).
661 * Returns < 0 on error, 0 on success. For error codes see bdrv_write().
663 int bdrv_make_zero(BlockDriverState *bs, BdrvRequestFlags flags)
665 int64_t target_sectors, ret, nb_sectors, sector_num = 0;
668 target_sectors = bdrv_nb_sectors(bs);
669 if (target_sectors < 0) {
670 return target_sectors;
674 nb_sectors = MIN(target_sectors - sector_num, BDRV_REQUEST_MAX_SECTORS);
675 if (nb_sectors <= 0) {
678 ret = bdrv_get_block_status(bs, sector_num, nb_sectors, &n);
680 error_report("error getting block status at sector %" PRId64 ": %s",
681 sector_num, strerror(-ret));
684 if (ret & BDRV_BLOCK_ZERO) {
688 ret = bdrv_write_zeroes(bs, sector_num, n, flags);
690 error_report("error writing zeroes at sector %" PRId64 ": %s",
691 sector_num, strerror(-ret));
698 int bdrv_pread(BlockDriverState *bs, int64_t offset, void *buf, int bytes)
702 .iov_base = (void *)buf,
711 qemu_iovec_init_external(&qiov, &iov, 1);
712 ret = bdrv_prwv_co(bs, offset, &qiov, false, 0);
720 int bdrv_pwritev(BlockDriverState *bs, int64_t offset, QEMUIOVector *qiov)
724 ret = bdrv_prwv_co(bs, offset, qiov, true, 0);
732 int bdrv_pwrite(BlockDriverState *bs, int64_t offset,
733 const void *buf, int bytes)
737 .iov_base = (void *) buf,
745 qemu_iovec_init_external(&qiov, &iov, 1);
746 return bdrv_pwritev(bs, offset, &qiov);
750 * Writes to the file and ensures that no writes are reordered across this
751 * request (acts as a barrier)
753 * Returns 0 on success, -errno in error cases.
755 int bdrv_pwrite_sync(BlockDriverState *bs, int64_t offset,
756 const void *buf, int count)
760 ret = bdrv_pwrite(bs, offset, buf, count);
765 /* No flush needed for cache modes that already do it */
766 if (bs->enable_write_cache) {
773 static int coroutine_fn bdrv_co_do_copy_on_readv(BlockDriverState *bs,
774 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov)
776 /* Perform I/O through a temporary buffer so that users who scribble over
777 * their read buffer while the operation is in progress do not end up
778 * modifying the image file. This is critical for zero-copy guest I/O
779 * where anything might happen inside guest memory.
783 BlockDriver *drv = bs->drv;
785 QEMUIOVector bounce_qiov;
786 int64_t cluster_sector_num;
787 int cluster_nb_sectors;
791 /* Cover entire cluster so no additional backing file I/O is required when
792 * allocating cluster in the image file.
794 bdrv_round_to_clusters(bs, sector_num, nb_sectors,
795 &cluster_sector_num, &cluster_nb_sectors);
797 trace_bdrv_co_do_copy_on_readv(bs, sector_num, nb_sectors,
798 cluster_sector_num, cluster_nb_sectors);
800 iov.iov_len = cluster_nb_sectors * BDRV_SECTOR_SIZE;
801 iov.iov_base = bounce_buffer = qemu_try_blockalign(bs, iov.iov_len);
802 if (bounce_buffer == NULL) {
807 qemu_iovec_init_external(&bounce_qiov, &iov, 1);
809 ret = drv->bdrv_co_readv(bs, cluster_sector_num, cluster_nb_sectors,
815 if (drv->bdrv_co_write_zeroes &&
816 buffer_is_zero(bounce_buffer, iov.iov_len)) {
817 ret = bdrv_co_do_write_zeroes(bs, cluster_sector_num,
818 cluster_nb_sectors, 0);
820 /* This does not change the data on the disk, it is not necessary
821 * to flush even in cache=writethrough mode.
823 ret = drv->bdrv_co_writev(bs, cluster_sector_num, cluster_nb_sectors,
828 /* It might be okay to ignore write errors for guest requests. If this
829 * is a deliberate copy-on-read then we don't want to ignore the error.
830 * Simply report it in all cases.
835 skip_bytes = (sector_num - cluster_sector_num) * BDRV_SECTOR_SIZE;
836 qemu_iovec_from_buf(qiov, 0, bounce_buffer + skip_bytes,
837 nb_sectors * BDRV_SECTOR_SIZE);
840 qemu_vfree(bounce_buffer);
845 * Forwards an already correctly aligned request to the BlockDriver. This
846 * handles copy on read and zeroing after EOF; any other features must be
847 * implemented by the caller.
849 static int coroutine_fn bdrv_aligned_preadv(BlockDriverState *bs,
850 BdrvTrackedRequest *req, int64_t offset, unsigned int bytes,
851 int64_t align, QEMUIOVector *qiov, int flags)
853 BlockDriver *drv = bs->drv;
856 int64_t sector_num = offset >> BDRV_SECTOR_BITS;
857 unsigned int nb_sectors = bytes >> BDRV_SECTOR_BITS;
859 assert((offset & (BDRV_SECTOR_SIZE - 1)) == 0);
860 assert((bytes & (BDRV_SECTOR_SIZE - 1)) == 0);
861 assert(!qiov || bytes == qiov->size);
863 /* Handle Copy on Read and associated serialisation */
864 if (flags & BDRV_REQ_COPY_ON_READ) {
865 /* If we touch the same cluster it counts as an overlap. This
866 * guarantees that allocating writes will be serialized and not race
867 * with each other for the same cluster. For example, in copy-on-read
868 * it ensures that the CoR read and write operations are atomic and
869 * guest writes cannot interleave between them. */
870 mark_request_serialising(req, bdrv_get_cluster_size(bs));
873 if (!(flags & BDRV_REQ_NO_SERIALISING)) {
874 wait_serialising_requests(req);
877 if (flags & BDRV_REQ_COPY_ON_READ) {
880 ret = bdrv_is_allocated(bs, sector_num, nb_sectors, &pnum);
885 if (!ret || pnum != nb_sectors) {
886 ret = bdrv_co_do_copy_on_readv(bs, sector_num, nb_sectors, qiov);
891 /* Forward the request to the BlockDriver */
892 if (!bs->zero_beyond_eof) {
893 ret = drv->bdrv_co_readv(bs, sector_num, nb_sectors, qiov);
895 /* Read zeros after EOF */
896 int64_t total_sectors, max_nb_sectors;
898 total_sectors = bdrv_nb_sectors(bs);
899 if (total_sectors < 0) {
904 max_nb_sectors = ROUND_UP(MAX(0, total_sectors - sector_num),
905 align >> BDRV_SECTOR_BITS);
906 if (nb_sectors < max_nb_sectors) {
907 ret = drv->bdrv_co_readv(bs, sector_num, nb_sectors, qiov);
908 } else if (max_nb_sectors > 0) {
909 QEMUIOVector local_qiov;
911 qemu_iovec_init(&local_qiov, qiov->niov);
912 qemu_iovec_concat(&local_qiov, qiov, 0,
913 max_nb_sectors * BDRV_SECTOR_SIZE);
915 ret = drv->bdrv_co_readv(bs, sector_num, max_nb_sectors,
918 qemu_iovec_destroy(&local_qiov);
923 /* Reading beyond end of file is supposed to produce zeroes */
924 if (ret == 0 && total_sectors < sector_num + nb_sectors) {
925 uint64_t offset = MAX(0, total_sectors - sector_num);
926 uint64_t bytes = (sector_num + nb_sectors - offset) *
928 qemu_iovec_memset(qiov, offset * BDRV_SECTOR_SIZE, 0, bytes);
937 * Handle a read request in coroutine context
939 static int coroutine_fn bdrv_co_do_preadv(BlockDriverState *bs,
940 int64_t offset, unsigned int bytes, QEMUIOVector *qiov,
941 BdrvRequestFlags flags)
943 BlockDriver *drv = bs->drv;
944 BdrvTrackedRequest req;
946 /* TODO Lift BDRV_SECTOR_SIZE restriction in BlockDriver interface */
947 uint64_t align = MAX(BDRV_SECTOR_SIZE, bs->request_alignment);
948 uint8_t *head_buf = NULL;
949 uint8_t *tail_buf = NULL;
950 QEMUIOVector local_qiov;
951 bool use_local_qiov = false;
958 ret = bdrv_check_byte_request(bs, offset, bytes);
963 /* Don't do copy-on-read if we read data before write operation */
964 if (bs->copy_on_read && !(flags & BDRV_REQ_NO_SERIALISING)) {
965 flags |= BDRV_REQ_COPY_ON_READ;
968 /* throttling disk I/O */
969 if (bs->io_limits_enabled) {
970 throttle_group_co_io_limits_intercept(bs, bytes, false);
973 /* Align read if necessary by padding qiov */
974 if (offset & (align - 1)) {
975 head_buf = qemu_blockalign(bs, align);
976 qemu_iovec_init(&local_qiov, qiov->niov + 2);
977 qemu_iovec_add(&local_qiov, head_buf, offset & (align - 1));
978 qemu_iovec_concat(&local_qiov, qiov, 0, qiov->size);
979 use_local_qiov = true;
981 bytes += offset & (align - 1);
982 offset = offset & ~(align - 1);
985 if ((offset + bytes) & (align - 1)) {
986 if (!use_local_qiov) {
987 qemu_iovec_init(&local_qiov, qiov->niov + 1);
988 qemu_iovec_concat(&local_qiov, qiov, 0, qiov->size);
989 use_local_qiov = true;
991 tail_buf = qemu_blockalign(bs, align);
992 qemu_iovec_add(&local_qiov, tail_buf,
993 align - ((offset + bytes) & (align - 1)));
995 bytes = ROUND_UP(bytes, align);
998 tracked_request_begin(&req, bs, offset, bytes, BDRV_TRACKED_READ);
999 ret = bdrv_aligned_preadv(bs, &req, offset, bytes, align,
1000 use_local_qiov ? &local_qiov : qiov,
1002 tracked_request_end(&req);
1004 if (use_local_qiov) {
1005 qemu_iovec_destroy(&local_qiov);
1006 qemu_vfree(head_buf);
1007 qemu_vfree(tail_buf);
1013 static int coroutine_fn bdrv_co_do_readv(BlockDriverState *bs,
1014 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov,
1015 BdrvRequestFlags flags)
1017 if (nb_sectors < 0 || nb_sectors > BDRV_REQUEST_MAX_SECTORS) {
1021 return bdrv_co_do_preadv(bs, sector_num << BDRV_SECTOR_BITS,
1022 nb_sectors << BDRV_SECTOR_BITS, qiov, flags);
1025 int coroutine_fn bdrv_co_readv(BlockDriverState *bs, int64_t sector_num,
1026 int nb_sectors, QEMUIOVector *qiov)
1028 trace_bdrv_co_readv(bs, sector_num, nb_sectors);
1030 return bdrv_co_do_readv(bs, sector_num, nb_sectors, qiov, 0);
1033 int coroutine_fn bdrv_co_readv_no_serialising(BlockDriverState *bs,
1034 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov)
1036 trace_bdrv_co_readv_no_serialising(bs, sector_num, nb_sectors);
1038 return bdrv_co_do_readv(bs, sector_num, nb_sectors, qiov,
1039 BDRV_REQ_NO_SERIALISING);
1042 int coroutine_fn bdrv_co_copy_on_readv(BlockDriverState *bs,
1043 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov)
1045 trace_bdrv_co_copy_on_readv(bs, sector_num, nb_sectors);
1047 return bdrv_co_do_readv(bs, sector_num, nb_sectors, qiov,
1048 BDRV_REQ_COPY_ON_READ);
1051 #define MAX_WRITE_ZEROES_BOUNCE_BUFFER 32768
1053 static int coroutine_fn bdrv_co_do_write_zeroes(BlockDriverState *bs,
1054 int64_t sector_num, int nb_sectors, BdrvRequestFlags flags)
1056 BlockDriver *drv = bs->drv;
1058 struct iovec iov = {0};
1061 int max_write_zeroes = MIN_NON_ZERO(bs->bl.max_write_zeroes,
1062 BDRV_REQUEST_MAX_SECTORS);
1064 while (nb_sectors > 0 && !ret) {
1065 int num = nb_sectors;
1067 /* Align request. Block drivers can expect the "bulk" of the request
1070 if (bs->bl.write_zeroes_alignment
1071 && num > bs->bl.write_zeroes_alignment) {
1072 if (sector_num % bs->bl.write_zeroes_alignment != 0) {
1073 /* Make a small request up to the first aligned sector. */
1074 num = bs->bl.write_zeroes_alignment;
1075 num -= sector_num % bs->bl.write_zeroes_alignment;
1076 } else if ((sector_num + num) % bs->bl.write_zeroes_alignment != 0) {
1077 /* Shorten the request to the last aligned sector. num cannot
1078 * underflow because num > bs->bl.write_zeroes_alignment.
1080 num -= (sector_num + num) % bs->bl.write_zeroes_alignment;
1084 /* limit request size */
1085 if (num > max_write_zeroes) {
1086 num = max_write_zeroes;
1090 /* First try the efficient write zeroes operation */
1091 if (drv->bdrv_co_write_zeroes) {
1092 ret = drv->bdrv_co_write_zeroes(bs, sector_num, num, flags);
1095 if (ret == -ENOTSUP) {
1096 /* Fall back to bounce buffer if write zeroes is unsupported */
1097 int max_xfer_len = MIN_NON_ZERO(bs->bl.max_transfer_length,
1098 MAX_WRITE_ZEROES_BOUNCE_BUFFER);
1099 num = MIN(num, max_xfer_len);
1100 iov.iov_len = num * BDRV_SECTOR_SIZE;
1101 if (iov.iov_base == NULL) {
1102 iov.iov_base = qemu_try_blockalign(bs, num * BDRV_SECTOR_SIZE);
1103 if (iov.iov_base == NULL) {
1107 memset(iov.iov_base, 0, num * BDRV_SECTOR_SIZE);
1109 qemu_iovec_init_external(&qiov, &iov, 1);
1111 ret = drv->bdrv_co_writev(bs, sector_num, num, &qiov);
1113 /* Keep bounce buffer around if it is big enough for all
1114 * all future requests.
1116 if (num < max_xfer_len) {
1117 qemu_vfree(iov.iov_base);
1118 iov.iov_base = NULL;
1127 qemu_vfree(iov.iov_base);
1132 * Forwards an already correctly aligned write request to the BlockDriver.
1134 static int coroutine_fn bdrv_aligned_pwritev(BlockDriverState *bs,
1135 BdrvTrackedRequest *req, int64_t offset, unsigned int bytes,
1136 QEMUIOVector *qiov, int flags)
1138 BlockDriver *drv = bs->drv;
1142 int64_t sector_num = offset >> BDRV_SECTOR_BITS;
1143 unsigned int nb_sectors = bytes >> BDRV_SECTOR_BITS;
1145 assert((offset & (BDRV_SECTOR_SIZE - 1)) == 0);
1146 assert((bytes & (BDRV_SECTOR_SIZE - 1)) == 0);
1147 assert(!qiov || bytes == qiov->size);
1149 waited = wait_serialising_requests(req);
1150 assert(!waited || !req->serialising);
1151 assert(req->overlap_offset <= offset);
1152 assert(offset + bytes <= req->overlap_offset + req->overlap_bytes);
1154 ret = notifier_with_return_list_notify(&bs->before_write_notifiers, req);
1156 if (!ret && bs->detect_zeroes != BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF &&
1157 !(flags & BDRV_REQ_ZERO_WRITE) && drv->bdrv_co_write_zeroes &&
1158 qemu_iovec_is_zero(qiov)) {
1159 flags |= BDRV_REQ_ZERO_WRITE;
1160 if (bs->detect_zeroes == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP) {
1161 flags |= BDRV_REQ_MAY_UNMAP;
1166 /* Do nothing, write notifier decided to fail this request */
1167 } else if (flags & BDRV_REQ_ZERO_WRITE) {
1168 bdrv_debug_event(bs, BLKDBG_PWRITEV_ZERO);
1169 ret = bdrv_co_do_write_zeroes(bs, sector_num, nb_sectors, flags);
1171 bdrv_debug_event(bs, BLKDBG_PWRITEV);
1172 ret = drv->bdrv_co_writev(bs, sector_num, nb_sectors, qiov);
1174 bdrv_debug_event(bs, BLKDBG_PWRITEV_DONE);
1176 if (ret == 0 && !bs->enable_write_cache) {
1177 ret = bdrv_co_flush(bs);
1180 bdrv_set_dirty(bs, sector_num, nb_sectors);
1182 if (bs->wr_highest_offset < offset + bytes) {
1183 bs->wr_highest_offset = offset + bytes;
1187 bs->total_sectors = MAX(bs->total_sectors, sector_num + nb_sectors);
1193 static int coroutine_fn bdrv_co_do_zero_pwritev(BlockDriverState *bs,
1196 BdrvRequestFlags flags,
1197 BdrvTrackedRequest *req)
1199 uint8_t *buf = NULL;
1200 QEMUIOVector local_qiov;
1202 uint64_t align = MAX(BDRV_SECTOR_SIZE, bs->request_alignment);
1203 unsigned int head_padding_bytes, tail_padding_bytes;
1206 head_padding_bytes = offset & (align - 1);
1207 tail_padding_bytes = align - ((offset + bytes) & (align - 1));
1210 assert(flags & BDRV_REQ_ZERO_WRITE);
1211 if (head_padding_bytes || tail_padding_bytes) {
1212 buf = qemu_blockalign(bs, align);
1213 iov = (struct iovec) {
1217 qemu_iovec_init_external(&local_qiov, &iov, 1);
1219 if (head_padding_bytes) {
1220 uint64_t zero_bytes = MIN(bytes, align - head_padding_bytes);
1222 /* RMW the unaligned part before head. */
1223 mark_request_serialising(req, align);
1224 wait_serialising_requests(req);
1225 bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_HEAD);
1226 ret = bdrv_aligned_preadv(bs, req, offset & ~(align - 1), align,
1227 align, &local_qiov, 0);
1231 bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_AFTER_HEAD);
1233 memset(buf + head_padding_bytes, 0, zero_bytes);
1234 ret = bdrv_aligned_pwritev(bs, req, offset & ~(align - 1), align,
1236 flags & ~BDRV_REQ_ZERO_WRITE);
1240 offset += zero_bytes;
1241 bytes -= zero_bytes;
1244 assert(!bytes || (offset & (align - 1)) == 0);
1245 if (bytes >= align) {
1246 /* Write the aligned part in the middle. */
1247 uint64_t aligned_bytes = bytes & ~(align - 1);
1248 ret = bdrv_aligned_pwritev(bs, req, offset, aligned_bytes,
1253 bytes -= aligned_bytes;
1254 offset += aligned_bytes;
1257 assert(!bytes || (offset & (align - 1)) == 0);
1259 assert(align == tail_padding_bytes + bytes);
1260 /* RMW the unaligned part after tail. */
1261 mark_request_serialising(req, align);
1262 wait_serialising_requests(req);
1263 bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_TAIL);
1264 ret = bdrv_aligned_preadv(bs, req, offset, align,
1265 align, &local_qiov, 0);
1269 bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_AFTER_TAIL);
1271 memset(buf, 0, bytes);
1272 ret = bdrv_aligned_pwritev(bs, req, offset, align,
1273 &local_qiov, flags & ~BDRV_REQ_ZERO_WRITE);
1282 * Handle a write request in coroutine context
1284 static int coroutine_fn bdrv_co_do_pwritev(BlockDriverState *bs,
1285 int64_t offset, unsigned int bytes, QEMUIOVector *qiov,
1286 BdrvRequestFlags flags)
1288 BdrvTrackedRequest req;
1289 /* TODO Lift BDRV_SECTOR_SIZE restriction in BlockDriver interface */
1290 uint64_t align = MAX(BDRV_SECTOR_SIZE, bs->request_alignment);
1291 uint8_t *head_buf = NULL;
1292 uint8_t *tail_buf = NULL;
1293 QEMUIOVector local_qiov;
1294 bool use_local_qiov = false;
1300 if (bs->read_only) {
1304 ret = bdrv_check_byte_request(bs, offset, bytes);
1309 /* throttling disk I/O */
1310 if (bs->io_limits_enabled) {
1311 throttle_group_co_io_limits_intercept(bs, bytes, true);
1315 * Align write if necessary by performing a read-modify-write cycle.
1316 * Pad qiov with the read parts and be sure to have a tracked request not
1317 * only for bdrv_aligned_pwritev, but also for the reads of the RMW cycle.
1319 tracked_request_begin(&req, bs, offset, bytes, BDRV_TRACKED_WRITE);
1322 ret = bdrv_co_do_zero_pwritev(bs, offset, bytes, flags, &req);
1326 if (offset & (align - 1)) {
1327 QEMUIOVector head_qiov;
1328 struct iovec head_iov;
1330 mark_request_serialising(&req, align);
1331 wait_serialising_requests(&req);
1333 head_buf = qemu_blockalign(bs, align);
1334 head_iov = (struct iovec) {
1335 .iov_base = head_buf,
1338 qemu_iovec_init_external(&head_qiov, &head_iov, 1);
1340 bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_HEAD);
1341 ret = bdrv_aligned_preadv(bs, &req, offset & ~(align - 1), align,
1342 align, &head_qiov, 0);
1346 bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_AFTER_HEAD);
1348 qemu_iovec_init(&local_qiov, qiov->niov + 2);
1349 qemu_iovec_add(&local_qiov, head_buf, offset & (align - 1));
1350 qemu_iovec_concat(&local_qiov, qiov, 0, qiov->size);
1351 use_local_qiov = true;
1353 bytes += offset & (align - 1);
1354 offset = offset & ~(align - 1);
1357 if ((offset + bytes) & (align - 1)) {
1358 QEMUIOVector tail_qiov;
1359 struct iovec tail_iov;
1363 mark_request_serialising(&req, align);
1364 waited = wait_serialising_requests(&req);
1365 assert(!waited || !use_local_qiov);
1367 tail_buf = qemu_blockalign(bs, align);
1368 tail_iov = (struct iovec) {
1369 .iov_base = tail_buf,
1372 qemu_iovec_init_external(&tail_qiov, &tail_iov, 1);
1374 bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_TAIL);
1375 ret = bdrv_aligned_preadv(bs, &req, (offset + bytes) & ~(align - 1), align,
1376 align, &tail_qiov, 0);
1380 bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_AFTER_TAIL);
1382 if (!use_local_qiov) {
1383 qemu_iovec_init(&local_qiov, qiov->niov + 1);
1384 qemu_iovec_concat(&local_qiov, qiov, 0, qiov->size);
1385 use_local_qiov = true;
1388 tail_bytes = (offset + bytes) & (align - 1);
1389 qemu_iovec_add(&local_qiov, tail_buf + tail_bytes, align - tail_bytes);
1391 bytes = ROUND_UP(bytes, align);
1394 ret = bdrv_aligned_pwritev(bs, &req, offset, bytes,
1395 use_local_qiov ? &local_qiov : qiov,
1400 if (use_local_qiov) {
1401 qemu_iovec_destroy(&local_qiov);
1403 qemu_vfree(head_buf);
1404 qemu_vfree(tail_buf);
1406 tracked_request_end(&req);
1410 static int coroutine_fn bdrv_co_do_writev(BlockDriverState *bs,
1411 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov,
1412 BdrvRequestFlags flags)
1414 if (nb_sectors < 0 || nb_sectors > BDRV_REQUEST_MAX_SECTORS) {
1418 return bdrv_co_do_pwritev(bs, sector_num << BDRV_SECTOR_BITS,
1419 nb_sectors << BDRV_SECTOR_BITS, qiov, flags);
1422 int coroutine_fn bdrv_co_writev(BlockDriverState *bs, int64_t sector_num,
1423 int nb_sectors, QEMUIOVector *qiov)
1425 trace_bdrv_co_writev(bs, sector_num, nb_sectors);
1427 return bdrv_co_do_writev(bs, sector_num, nb_sectors, qiov, 0);
1430 int coroutine_fn bdrv_co_write_zeroes(BlockDriverState *bs,
1431 int64_t sector_num, int nb_sectors,
1432 BdrvRequestFlags flags)
1434 trace_bdrv_co_write_zeroes(bs, sector_num, nb_sectors, flags);
1436 if (!(bs->open_flags & BDRV_O_UNMAP)) {
1437 flags &= ~BDRV_REQ_MAY_UNMAP;
1440 return bdrv_co_do_writev(bs, sector_num, nb_sectors, NULL,
1441 BDRV_REQ_ZERO_WRITE | flags);
1444 int bdrv_flush_all(void)
1446 BlockDriverState *bs = NULL;
1449 while ((bs = bdrv_next(bs))) {
1450 AioContext *aio_context = bdrv_get_aio_context(bs);
1453 aio_context_acquire(aio_context);
1454 ret = bdrv_flush(bs);
1455 if (ret < 0 && !result) {
1458 aio_context_release(aio_context);
1464 typedef struct BdrvCoGetBlockStatusData {
1465 BlockDriverState *bs;
1466 BlockDriverState *base;
1472 } BdrvCoGetBlockStatusData;
1475 * Returns the allocation status of the specified sectors.
1476 * Drivers not implementing the functionality are assumed to not support
1477 * backing files, hence all their sectors are reported as allocated.
1479 * If 'sector_num' is beyond the end of the disk image the return value is 0
1480 * and 'pnum' is set to 0.
1482 * 'pnum' is set to the number of sectors (including and immediately following
1483 * the specified sector) that are known to be in the same
1484 * allocated/unallocated state.
1486 * 'nb_sectors' is the max value 'pnum' should be set to. If nb_sectors goes
1487 * beyond the end of the disk image it will be clamped.
1489 static int64_t coroutine_fn bdrv_co_get_block_status(BlockDriverState *bs,
1491 int nb_sectors, int *pnum)
1493 int64_t total_sectors;
1497 total_sectors = bdrv_nb_sectors(bs);
1498 if (total_sectors < 0) {
1499 return total_sectors;
1502 if (sector_num >= total_sectors) {
1507 n = total_sectors - sector_num;
1508 if (n < nb_sectors) {
1512 if (!bs->drv->bdrv_co_get_block_status) {
1514 ret = BDRV_BLOCK_DATA | BDRV_BLOCK_ALLOCATED;
1515 if (bs->drv->protocol_name) {
1516 ret |= BDRV_BLOCK_OFFSET_VALID | (sector_num * BDRV_SECTOR_SIZE);
1521 ret = bs->drv->bdrv_co_get_block_status(bs, sector_num, nb_sectors, pnum);
1527 if (ret & BDRV_BLOCK_RAW) {
1528 assert(ret & BDRV_BLOCK_OFFSET_VALID);
1529 return bdrv_get_block_status(bs->file->bs, ret >> BDRV_SECTOR_BITS,
1533 if (ret & (BDRV_BLOCK_DATA | BDRV_BLOCK_ZERO)) {
1534 ret |= BDRV_BLOCK_ALLOCATED;
1536 if (bdrv_unallocated_blocks_are_zero(bs)) {
1537 ret |= BDRV_BLOCK_ZERO;
1538 } else if (bs->backing) {
1539 BlockDriverState *bs2 = bs->backing->bs;
1540 int64_t nb_sectors2 = bdrv_nb_sectors(bs2);
1541 if (nb_sectors2 >= 0 && sector_num >= nb_sectors2) {
1542 ret |= BDRV_BLOCK_ZERO;
1548 (ret & BDRV_BLOCK_DATA) && !(ret & BDRV_BLOCK_ZERO) &&
1549 (ret & BDRV_BLOCK_OFFSET_VALID)) {
1552 ret2 = bdrv_co_get_block_status(bs->file->bs, ret >> BDRV_SECTOR_BITS,
1555 /* Ignore errors. This is just providing extra information, it
1556 * is useful but not necessary.
1559 /* !file_pnum indicates an offset at or beyond the EOF; it is
1560 * perfectly valid for the format block driver to point to such
1561 * offsets, so catch it and mark everything as zero */
1562 ret |= BDRV_BLOCK_ZERO;
1564 /* Limit request to the range reported by the protocol driver */
1566 ret |= (ret2 & BDRV_BLOCK_ZERO);
1574 static int64_t coroutine_fn bdrv_co_get_block_status_above(BlockDriverState *bs,
1575 BlockDriverState *base,
1580 BlockDriverState *p;
1584 for (p = bs; p != base; p = backing_bs(p)) {
1585 ret = bdrv_co_get_block_status(p, sector_num, nb_sectors, pnum);
1586 if (ret < 0 || ret & BDRV_BLOCK_ALLOCATED) {
1589 /* [sector_num, pnum] unallocated on this layer, which could be only
1590 * the first part of [sector_num, nb_sectors]. */
1591 nb_sectors = MIN(nb_sectors, *pnum);
1596 /* Coroutine wrapper for bdrv_get_block_status_above() */
1597 static void coroutine_fn bdrv_get_block_status_above_co_entry(void *opaque)
1599 BdrvCoGetBlockStatusData *data = opaque;
1601 data->ret = bdrv_co_get_block_status_above(data->bs, data->base,
1609 * Synchronous wrapper around bdrv_co_get_block_status_above().
1611 * See bdrv_co_get_block_status_above() for details.
1613 int64_t bdrv_get_block_status_above(BlockDriverState *bs,
1614 BlockDriverState *base,
1616 int nb_sectors, int *pnum)
1619 BdrvCoGetBlockStatusData data = {
1622 .sector_num = sector_num,
1623 .nb_sectors = nb_sectors,
1628 if (qemu_in_coroutine()) {
1629 /* Fast-path if already in coroutine context */
1630 bdrv_get_block_status_above_co_entry(&data);
1632 AioContext *aio_context = bdrv_get_aio_context(bs);
1634 co = qemu_coroutine_create(bdrv_get_block_status_above_co_entry);
1635 qemu_coroutine_enter(co, &data);
1636 while (!data.done) {
1637 aio_poll(aio_context, true);
1643 int64_t bdrv_get_block_status(BlockDriverState *bs,
1645 int nb_sectors, int *pnum)
1647 return bdrv_get_block_status_above(bs, backing_bs(bs),
1648 sector_num, nb_sectors, pnum);
1651 int coroutine_fn bdrv_is_allocated(BlockDriverState *bs, int64_t sector_num,
1652 int nb_sectors, int *pnum)
1654 int64_t ret = bdrv_get_block_status(bs, sector_num, nb_sectors, pnum);
1658 return !!(ret & BDRV_BLOCK_ALLOCATED);
1662 * Given an image chain: ... -> [BASE] -> [INTER1] -> [INTER2] -> [TOP]
1664 * Return true if the given sector is allocated in any image between
1665 * BASE and TOP (inclusive). BASE can be NULL to check if the given
1666 * sector is allocated in any image of the chain. Return false otherwise.
1668 * 'pnum' is set to the number of sectors (including and immediately following
1669 * the specified sector) that are known to be in the same
1670 * allocated/unallocated state.
1673 int bdrv_is_allocated_above(BlockDriverState *top,
1674 BlockDriverState *base,
1676 int nb_sectors, int *pnum)
1678 BlockDriverState *intermediate;
1679 int ret, n = nb_sectors;
1682 while (intermediate && intermediate != base) {
1684 ret = bdrv_is_allocated(intermediate, sector_num, nb_sectors,
1694 * [sector_num, nb_sectors] is unallocated on top but intermediate
1697 * [sector_num+x, nr_sectors] allocated.
1699 if (n > pnum_inter &&
1700 (intermediate == top ||
1701 sector_num + pnum_inter < intermediate->total_sectors)) {
1705 intermediate = backing_bs(intermediate);
1712 int bdrv_write_compressed(BlockDriverState *bs, int64_t sector_num,
1713 const uint8_t *buf, int nb_sectors)
1715 BlockDriver *drv = bs->drv;
1721 if (!drv->bdrv_write_compressed) {
1724 ret = bdrv_check_request(bs, sector_num, nb_sectors);
1729 assert(QLIST_EMPTY(&bs->dirty_bitmaps));
1731 return drv->bdrv_write_compressed(bs, sector_num, buf, nb_sectors);
1734 int bdrv_save_vmstate(BlockDriverState *bs, const uint8_t *buf,
1735 int64_t pos, int size)
1738 struct iovec iov = {
1739 .iov_base = (void *) buf,
1743 qemu_iovec_init_external(&qiov, &iov, 1);
1744 return bdrv_writev_vmstate(bs, &qiov, pos);
1747 int bdrv_writev_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, int64_t pos)
1749 BlockDriver *drv = bs->drv;
1753 } else if (drv->bdrv_save_vmstate) {
1754 return drv->bdrv_save_vmstate(bs, qiov, pos);
1755 } else if (bs->file) {
1756 return bdrv_writev_vmstate(bs->file->bs, qiov, pos);
1762 int bdrv_load_vmstate(BlockDriverState *bs, uint8_t *buf,
1763 int64_t pos, int size)
1765 BlockDriver *drv = bs->drv;
1768 if (drv->bdrv_load_vmstate)
1769 return drv->bdrv_load_vmstate(bs, buf, pos, size);
1771 return bdrv_load_vmstate(bs->file->bs, buf, pos, size);
1775 /**************************************************************/
1778 BlockAIOCB *bdrv_aio_readv(BlockDriverState *bs, int64_t sector_num,
1779 QEMUIOVector *qiov, int nb_sectors,
1780 BlockCompletionFunc *cb, void *opaque)
1782 trace_bdrv_aio_readv(bs, sector_num, nb_sectors, opaque);
1784 return bdrv_co_aio_rw_vector(bs, sector_num, qiov, nb_sectors, 0,
1788 BlockAIOCB *bdrv_aio_writev(BlockDriverState *bs, int64_t sector_num,
1789 QEMUIOVector *qiov, int nb_sectors,
1790 BlockCompletionFunc *cb, void *opaque)
1792 trace_bdrv_aio_writev(bs, sector_num, nb_sectors, opaque);
1794 return bdrv_co_aio_rw_vector(bs, sector_num, qiov, nb_sectors, 0,
1798 BlockAIOCB *bdrv_aio_write_zeroes(BlockDriverState *bs,
1799 int64_t sector_num, int nb_sectors, BdrvRequestFlags flags,
1800 BlockCompletionFunc *cb, void *opaque)
1802 trace_bdrv_aio_write_zeroes(bs, sector_num, nb_sectors, flags, opaque);
1804 return bdrv_co_aio_rw_vector(bs, sector_num, NULL, nb_sectors,
1805 BDRV_REQ_ZERO_WRITE | flags,
1810 typedef struct MultiwriteCB {
1815 BlockCompletionFunc *cb;
1817 QEMUIOVector *free_qiov;
1821 static void multiwrite_user_cb(MultiwriteCB *mcb)
1825 for (i = 0; i < mcb->num_callbacks; i++) {
1826 mcb->callbacks[i].cb(mcb->callbacks[i].opaque, mcb->error);
1827 if (mcb->callbacks[i].free_qiov) {
1828 qemu_iovec_destroy(mcb->callbacks[i].free_qiov);
1830 g_free(mcb->callbacks[i].free_qiov);
1834 static void multiwrite_cb(void *opaque, int ret)
1836 MultiwriteCB *mcb = opaque;
1838 trace_multiwrite_cb(mcb, ret);
1840 if (ret < 0 && !mcb->error) {
1844 mcb->num_requests--;
1845 if (mcb->num_requests == 0) {
1846 multiwrite_user_cb(mcb);
1851 static int multiwrite_req_compare(const void *a, const void *b)
1853 const BlockRequest *req1 = a, *req2 = b;
1856 * Note that we can't simply subtract req2->sector from req1->sector
1857 * here as that could overflow the return value.
1859 if (req1->sector > req2->sector) {
1861 } else if (req1->sector < req2->sector) {
1869 * Takes a bunch of requests and tries to merge them. Returns the number of
1870 * requests that remain after merging.
1872 static int multiwrite_merge(BlockDriverState *bs, BlockRequest *reqs,
1873 int num_reqs, MultiwriteCB *mcb)
1877 // Sort requests by start sector
1878 qsort(reqs, num_reqs, sizeof(*reqs), &multiwrite_req_compare);
1880 // Check if adjacent requests touch the same clusters. If so, combine them,
1881 // filling up gaps with zero sectors.
1883 for (i = 1; i < num_reqs; i++) {
1885 int64_t oldreq_last = reqs[outidx].sector + reqs[outidx].nb_sectors;
1887 // Handle exactly sequential writes and overlapping writes.
1888 if (reqs[i].sector <= oldreq_last) {
1892 if (reqs[outidx].qiov->niov + reqs[i].qiov->niov + 1 >
1897 if (bs->bl.max_transfer_length && reqs[outidx].nb_sectors +
1898 reqs[i].nb_sectors > bs->bl.max_transfer_length) {
1904 QEMUIOVector *qiov = g_malloc0(sizeof(*qiov));
1905 qemu_iovec_init(qiov,
1906 reqs[outidx].qiov->niov + reqs[i].qiov->niov + 1);
1908 // Add the first request to the merged one. If the requests are
1909 // overlapping, drop the last sectors of the first request.
1910 size = (reqs[i].sector - reqs[outidx].sector) << 9;
1911 qemu_iovec_concat(qiov, reqs[outidx].qiov, 0, size);
1913 // We should need to add any zeros between the two requests
1914 assert (reqs[i].sector <= oldreq_last);
1916 // Add the second request
1917 qemu_iovec_concat(qiov, reqs[i].qiov, 0, reqs[i].qiov->size);
1919 // Add tail of first request, if necessary
1920 if (qiov->size < reqs[outidx].qiov->size) {
1921 qemu_iovec_concat(qiov, reqs[outidx].qiov, qiov->size,
1922 reqs[outidx].qiov->size - qiov->size);
1925 reqs[outidx].nb_sectors = qiov->size >> 9;
1926 reqs[outidx].qiov = qiov;
1928 mcb->callbacks[i].free_qiov = reqs[outidx].qiov;
1931 reqs[outidx].sector = reqs[i].sector;
1932 reqs[outidx].nb_sectors = reqs[i].nb_sectors;
1933 reqs[outidx].qiov = reqs[i].qiov;
1938 block_acct_merge_done(blk_get_stats(bs->blk), BLOCK_ACCT_WRITE,
1939 num_reqs - outidx - 1);
1946 * Submit multiple AIO write requests at once.
1948 * On success, the function returns 0 and all requests in the reqs array have
1949 * been submitted. In error case this function returns -1, and any of the
1950 * requests may or may not be submitted yet. In particular, this means that the
1951 * callback will be called for some of the requests, for others it won't. The
1952 * caller must check the error field of the BlockRequest to wait for the right
1953 * callbacks (if error != 0, no callback will be called).
1955 * The implementation may modify the contents of the reqs array, e.g. to merge
1956 * requests. However, the fields opaque and error are left unmodified as they
1957 * are used to signal failure for a single request to the caller.
1959 int bdrv_aio_multiwrite(BlockDriverState *bs, BlockRequest *reqs, int num_reqs)
1964 /* don't submit writes if we don't have a medium */
1965 if (bs->drv == NULL) {
1966 for (i = 0; i < num_reqs; i++) {
1967 reqs[i].error = -ENOMEDIUM;
1972 if (num_reqs == 0) {
1976 // Create MultiwriteCB structure
1977 mcb = g_malloc0(sizeof(*mcb) + num_reqs * sizeof(*mcb->callbacks));
1978 mcb->num_requests = 0;
1979 mcb->num_callbacks = num_reqs;
1981 for (i = 0; i < num_reqs; i++) {
1982 mcb->callbacks[i].cb = reqs[i].cb;
1983 mcb->callbacks[i].opaque = reqs[i].opaque;
1986 // Check for mergable requests
1987 num_reqs = multiwrite_merge(bs, reqs, num_reqs, mcb);
1989 trace_bdrv_aio_multiwrite(mcb, mcb->num_callbacks, num_reqs);
1991 /* Run the aio requests. */
1992 mcb->num_requests = num_reqs;
1993 for (i = 0; i < num_reqs; i++) {
1994 bdrv_co_aio_rw_vector(bs, reqs[i].sector, reqs[i].qiov,
1995 reqs[i].nb_sectors, reqs[i].flags,
2003 void bdrv_aio_cancel(BlockAIOCB *acb)
2006 bdrv_aio_cancel_async(acb);
2007 while (acb->refcnt > 1) {
2008 if (acb->aiocb_info->get_aio_context) {
2009 aio_poll(acb->aiocb_info->get_aio_context(acb), true);
2010 } else if (acb->bs) {
2011 aio_poll(bdrv_get_aio_context(acb->bs), true);
2016 qemu_aio_unref(acb);
2019 /* Async version of aio cancel. The caller is not blocked if the acb implements
2020 * cancel_async, otherwise we do nothing and let the request normally complete.
2021 * In either case the completion callback must be called. */
2022 void bdrv_aio_cancel_async(BlockAIOCB *acb)
2024 if (acb->aiocb_info->cancel_async) {
2025 acb->aiocb_info->cancel_async(acb);
2029 /**************************************************************/
2030 /* async block device emulation */
2032 typedef struct BlockAIOCBSync {
2036 /* vector translation state */
2042 static const AIOCBInfo bdrv_em_aiocb_info = {
2043 .aiocb_size = sizeof(BlockAIOCBSync),
2046 static void bdrv_aio_bh_cb(void *opaque)
2048 BlockAIOCBSync *acb = opaque;
2050 if (!acb->is_write && acb->ret >= 0) {
2051 qemu_iovec_from_buf(acb->qiov, 0, acb->bounce, acb->qiov->size);
2053 qemu_vfree(acb->bounce);
2054 acb->common.cb(acb->common.opaque, acb->ret);
2055 qemu_bh_delete(acb->bh);
2057 qemu_aio_unref(acb);
2060 static BlockAIOCB *bdrv_aio_rw_vector(BlockDriverState *bs,
2064 BlockCompletionFunc *cb,
2069 BlockAIOCBSync *acb;
2071 acb = qemu_aio_get(&bdrv_em_aiocb_info, bs, cb, opaque);
2072 acb->is_write = is_write;
2074 acb->bounce = qemu_try_blockalign(bs, qiov->size);
2075 acb->bh = aio_bh_new(bdrv_get_aio_context(bs), bdrv_aio_bh_cb, acb);
2077 if (acb->bounce == NULL) {
2079 } else if (is_write) {
2080 qemu_iovec_to_buf(acb->qiov, 0, acb->bounce, qiov->size);
2081 acb->ret = bs->drv->bdrv_write(bs, sector_num, acb->bounce, nb_sectors);
2083 acb->ret = bs->drv->bdrv_read(bs, sector_num, acb->bounce, nb_sectors);
2086 qemu_bh_schedule(acb->bh);
2088 return &acb->common;
2091 static BlockAIOCB *bdrv_aio_readv_em(BlockDriverState *bs,
2092 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
2093 BlockCompletionFunc *cb, void *opaque)
2095 return bdrv_aio_rw_vector(bs, sector_num, qiov, nb_sectors, cb, opaque, 0);
2098 static BlockAIOCB *bdrv_aio_writev_em(BlockDriverState *bs,
2099 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
2100 BlockCompletionFunc *cb, void *opaque)
2102 return bdrv_aio_rw_vector(bs, sector_num, qiov, nb_sectors, cb, opaque, 1);
2106 typedef struct BlockAIOCBCoroutine {
2113 } BlockAIOCBCoroutine;
2115 static const AIOCBInfo bdrv_em_co_aiocb_info = {
2116 .aiocb_size = sizeof(BlockAIOCBCoroutine),
2119 static void bdrv_co_complete(BlockAIOCBCoroutine *acb)
2121 if (!acb->need_bh) {
2122 acb->common.cb(acb->common.opaque, acb->req.error);
2123 qemu_aio_unref(acb);
2127 static void bdrv_co_em_bh(void *opaque)
2129 BlockAIOCBCoroutine *acb = opaque;
2131 assert(!acb->need_bh);
2132 qemu_bh_delete(acb->bh);
2133 bdrv_co_complete(acb);
2136 static void bdrv_co_maybe_schedule_bh(BlockAIOCBCoroutine *acb)
2138 acb->need_bh = false;
2139 if (acb->req.error != -EINPROGRESS) {
2140 BlockDriverState *bs = acb->common.bs;
2142 acb->bh = aio_bh_new(bdrv_get_aio_context(bs), bdrv_co_em_bh, acb);
2143 qemu_bh_schedule(acb->bh);
2147 /* Invoke bdrv_co_do_readv/bdrv_co_do_writev */
2148 static void coroutine_fn bdrv_co_do_rw(void *opaque)
2150 BlockAIOCBCoroutine *acb = opaque;
2151 BlockDriverState *bs = acb->common.bs;
2153 if (!acb->is_write) {
2154 acb->req.error = bdrv_co_do_readv(bs, acb->req.sector,
2155 acb->req.nb_sectors, acb->req.qiov, acb->req.flags);
2157 acb->req.error = bdrv_co_do_writev(bs, acb->req.sector,
2158 acb->req.nb_sectors, acb->req.qiov, acb->req.flags);
2161 bdrv_co_complete(acb);
2164 static BlockAIOCB *bdrv_co_aio_rw_vector(BlockDriverState *bs,
2168 BdrvRequestFlags flags,
2169 BlockCompletionFunc *cb,
2174 BlockAIOCBCoroutine *acb;
2176 acb = qemu_aio_get(&bdrv_em_co_aiocb_info, bs, cb, opaque);
2177 acb->need_bh = true;
2178 acb->req.error = -EINPROGRESS;
2179 acb->req.sector = sector_num;
2180 acb->req.nb_sectors = nb_sectors;
2181 acb->req.qiov = qiov;
2182 acb->req.flags = flags;
2183 acb->is_write = is_write;
2185 co = qemu_coroutine_create(bdrv_co_do_rw);
2186 qemu_coroutine_enter(co, acb);
2188 bdrv_co_maybe_schedule_bh(acb);
2189 return &acb->common;
2192 static void coroutine_fn bdrv_aio_flush_co_entry(void *opaque)
2194 BlockAIOCBCoroutine *acb = opaque;
2195 BlockDriverState *bs = acb->common.bs;
2197 acb->req.error = bdrv_co_flush(bs);
2198 bdrv_co_complete(acb);
2201 BlockAIOCB *bdrv_aio_flush(BlockDriverState *bs,
2202 BlockCompletionFunc *cb, void *opaque)
2204 trace_bdrv_aio_flush(bs, opaque);
2207 BlockAIOCBCoroutine *acb;
2209 acb = qemu_aio_get(&bdrv_em_co_aiocb_info, bs, cb, opaque);
2210 acb->need_bh = true;
2211 acb->req.error = -EINPROGRESS;
2213 co = qemu_coroutine_create(bdrv_aio_flush_co_entry);
2214 qemu_coroutine_enter(co, acb);
2216 bdrv_co_maybe_schedule_bh(acb);
2217 return &acb->common;
2220 static void coroutine_fn bdrv_aio_discard_co_entry(void *opaque)
2222 BlockAIOCBCoroutine *acb = opaque;
2223 BlockDriverState *bs = acb->common.bs;
2225 acb->req.error = bdrv_co_discard(bs, acb->req.sector, acb->req.nb_sectors);
2226 bdrv_co_complete(acb);
2229 BlockAIOCB *bdrv_aio_discard(BlockDriverState *bs,
2230 int64_t sector_num, int nb_sectors,
2231 BlockCompletionFunc *cb, void *opaque)
2234 BlockAIOCBCoroutine *acb;
2236 trace_bdrv_aio_discard(bs, sector_num, nb_sectors, opaque);
2238 acb = qemu_aio_get(&bdrv_em_co_aiocb_info, bs, cb, opaque);
2239 acb->need_bh = true;
2240 acb->req.error = -EINPROGRESS;
2241 acb->req.sector = sector_num;
2242 acb->req.nb_sectors = nb_sectors;
2243 co = qemu_coroutine_create(bdrv_aio_discard_co_entry);
2244 qemu_coroutine_enter(co, acb);
2246 bdrv_co_maybe_schedule_bh(acb);
2247 return &acb->common;
2250 void *qemu_aio_get(const AIOCBInfo *aiocb_info, BlockDriverState *bs,
2251 BlockCompletionFunc *cb, void *opaque)
2255 acb = g_malloc(aiocb_info->aiocb_size);
2256 acb->aiocb_info = aiocb_info;
2259 acb->opaque = opaque;
2264 void qemu_aio_ref(void *p)
2266 BlockAIOCB *acb = p;
2270 void qemu_aio_unref(void *p)
2272 BlockAIOCB *acb = p;
2273 assert(acb->refcnt > 0);
2274 if (--acb->refcnt == 0) {
2279 /**************************************************************/
2280 /* Coroutine block device emulation */
2282 typedef struct CoroutineIOCompletion {
2283 Coroutine *coroutine;
2285 } CoroutineIOCompletion;
2287 static void bdrv_co_io_em_complete(void *opaque, int ret)
2289 CoroutineIOCompletion *co = opaque;
2292 qemu_coroutine_enter(co->coroutine, NULL);
2295 static int coroutine_fn bdrv_co_io_em(BlockDriverState *bs, int64_t sector_num,
2296 int nb_sectors, QEMUIOVector *iov,
2299 CoroutineIOCompletion co = {
2300 .coroutine = qemu_coroutine_self(),
2305 acb = bs->drv->bdrv_aio_writev(bs, sector_num, iov, nb_sectors,
2306 bdrv_co_io_em_complete, &co);
2308 acb = bs->drv->bdrv_aio_readv(bs, sector_num, iov, nb_sectors,
2309 bdrv_co_io_em_complete, &co);
2312 trace_bdrv_co_io_em(bs, sector_num, nb_sectors, is_write, acb);
2316 qemu_coroutine_yield();
2321 static int coroutine_fn bdrv_co_readv_em(BlockDriverState *bs,
2322 int64_t sector_num, int nb_sectors,
2325 return bdrv_co_io_em(bs, sector_num, nb_sectors, iov, false);
2328 static int coroutine_fn bdrv_co_writev_em(BlockDriverState *bs,
2329 int64_t sector_num, int nb_sectors,
2332 return bdrv_co_io_em(bs, sector_num, nb_sectors, iov, true);
2335 static void coroutine_fn bdrv_flush_co_entry(void *opaque)
2337 RwCo *rwco = opaque;
2339 rwco->ret = bdrv_co_flush(rwco->bs);
2342 int coroutine_fn bdrv_co_flush(BlockDriverState *bs)
2345 BdrvTrackedRequest req;
2347 if (!bs || !bdrv_is_inserted(bs) || bdrv_is_read_only(bs) ||
2352 tracked_request_begin(&req, bs, 0, 0, BDRV_TRACKED_FLUSH);
2353 /* Write back cached data to the OS even with cache=unsafe */
2354 BLKDBG_EVENT(bs->file, BLKDBG_FLUSH_TO_OS);
2355 if (bs->drv->bdrv_co_flush_to_os) {
2356 ret = bs->drv->bdrv_co_flush_to_os(bs);
2362 /* But don't actually force it to the disk with cache=unsafe */
2363 if (bs->open_flags & BDRV_O_NO_FLUSH) {
2367 BLKDBG_EVENT(bs->file, BLKDBG_FLUSH_TO_DISK);
2368 if (bs->drv->bdrv_co_flush_to_disk) {
2369 ret = bs->drv->bdrv_co_flush_to_disk(bs);
2370 } else if (bs->drv->bdrv_aio_flush) {
2372 CoroutineIOCompletion co = {
2373 .coroutine = qemu_coroutine_self(),
2376 acb = bs->drv->bdrv_aio_flush(bs, bdrv_co_io_em_complete, &co);
2380 qemu_coroutine_yield();
2385 * Some block drivers always operate in either writethrough or unsafe
2386 * mode and don't support bdrv_flush therefore. Usually qemu doesn't
2387 * know how the server works (because the behaviour is hardcoded or
2388 * depends on server-side configuration), so we can't ensure that
2389 * everything is safe on disk. Returning an error doesn't work because
2390 * that would break guests even if the server operates in writethrough
2393 * Let's hope the user knows what he's doing.
2401 /* Now flush the underlying protocol. It will also have BDRV_O_NO_FLUSH
2402 * in the case of cache=unsafe, so there are no useless flushes.
2405 ret = bs->file ? bdrv_co_flush(bs->file->bs) : 0;
2407 tracked_request_end(&req);
2411 int bdrv_flush(BlockDriverState *bs)
2419 if (qemu_in_coroutine()) {
2420 /* Fast-path if already in coroutine context */
2421 bdrv_flush_co_entry(&rwco);
2423 AioContext *aio_context = bdrv_get_aio_context(bs);
2425 co = qemu_coroutine_create(bdrv_flush_co_entry);
2426 qemu_coroutine_enter(co, &rwco);
2427 while (rwco.ret == NOT_DONE) {
2428 aio_poll(aio_context, true);
2435 typedef struct DiscardCo {
2436 BlockDriverState *bs;
2441 static void coroutine_fn bdrv_discard_co_entry(void *opaque)
2443 DiscardCo *rwco = opaque;
2445 rwco->ret = bdrv_co_discard(rwco->bs, rwco->sector_num, rwco->nb_sectors);
2448 int coroutine_fn bdrv_co_discard(BlockDriverState *bs, int64_t sector_num,
2451 BdrvTrackedRequest req;
2452 int max_discard, ret;
2458 ret = bdrv_check_request(bs, sector_num, nb_sectors);
2461 } else if (bs->read_only) {
2465 /* Do nothing if disabled. */
2466 if (!(bs->open_flags & BDRV_O_UNMAP)) {
2470 if (!bs->drv->bdrv_co_discard && !bs->drv->bdrv_aio_discard) {
2474 tracked_request_begin(&req, bs, sector_num, nb_sectors,
2475 BDRV_TRACKED_DISCARD);
2476 bdrv_set_dirty(bs, sector_num, nb_sectors);
2478 max_discard = MIN_NON_ZERO(bs->bl.max_discard, BDRV_REQUEST_MAX_SECTORS);
2479 while (nb_sectors > 0) {
2481 int num = nb_sectors;
2484 if (bs->bl.discard_alignment &&
2485 num >= bs->bl.discard_alignment &&
2486 sector_num % bs->bl.discard_alignment) {
2487 if (num > bs->bl.discard_alignment) {
2488 num = bs->bl.discard_alignment;
2490 num -= sector_num % bs->bl.discard_alignment;
2493 /* limit request size */
2494 if (num > max_discard) {
2498 if (bs->drv->bdrv_co_discard) {
2499 ret = bs->drv->bdrv_co_discard(bs, sector_num, num);
2502 CoroutineIOCompletion co = {
2503 .coroutine = qemu_coroutine_self(),
2506 acb = bs->drv->bdrv_aio_discard(bs, sector_num, nb_sectors,
2507 bdrv_co_io_em_complete, &co);
2512 qemu_coroutine_yield();
2516 if (ret && ret != -ENOTSUP) {
2525 tracked_request_end(&req);
2529 int bdrv_discard(BlockDriverState *bs, int64_t sector_num, int nb_sectors)
2534 .sector_num = sector_num,
2535 .nb_sectors = nb_sectors,
2539 if (qemu_in_coroutine()) {
2540 /* Fast-path if already in coroutine context */
2541 bdrv_discard_co_entry(&rwco);
2543 AioContext *aio_context = bdrv_get_aio_context(bs);
2545 co = qemu_coroutine_create(bdrv_discard_co_entry);
2546 qemu_coroutine_enter(co, &rwco);
2547 while (rwco.ret == NOT_DONE) {
2548 aio_poll(aio_context, true);
2556 CoroutineIOCompletion *co;
2558 } BdrvIoctlCompletionData;
2560 static void bdrv_ioctl_bh_cb(void *opaque)
2562 BdrvIoctlCompletionData *data = opaque;
2564 bdrv_co_io_em_complete(data->co, -ENOTSUP);
2565 qemu_bh_delete(data->bh);
2568 static int bdrv_co_do_ioctl(BlockDriverState *bs, int req, void *buf)
2570 BlockDriver *drv = bs->drv;
2571 BdrvTrackedRequest tracked_req;
2572 CoroutineIOCompletion co = {
2573 .coroutine = qemu_coroutine_self(),
2577 tracked_request_begin(&tracked_req, bs, 0, 0, BDRV_TRACKED_IOCTL);
2578 if (!drv || !drv->bdrv_aio_ioctl) {
2583 acb = drv->bdrv_aio_ioctl(bs, req, buf, bdrv_co_io_em_complete, &co);
2585 BdrvIoctlCompletionData *data = g_new(BdrvIoctlCompletionData, 1);
2586 data->bh = aio_bh_new(bdrv_get_aio_context(bs),
2587 bdrv_ioctl_bh_cb, data);
2589 qemu_bh_schedule(data->bh);
2591 qemu_coroutine_yield();
2593 tracked_request_end(&tracked_req);
2598 BlockDriverState *bs;
2604 static void coroutine_fn bdrv_co_ioctl_entry(void *opaque)
2606 BdrvIoctlCoData *data = opaque;
2607 data->ret = bdrv_co_do_ioctl(data->bs, data->req, data->buf);
2610 /* needed for generic scsi interface */
2611 int bdrv_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
2613 BdrvIoctlCoData data = {
2617 .ret = -EINPROGRESS,
2620 if (qemu_in_coroutine()) {
2621 /* Fast-path if already in coroutine context */
2622 bdrv_co_ioctl_entry(&data);
2624 Coroutine *co = qemu_coroutine_create(bdrv_co_ioctl_entry);
2626 qemu_coroutine_enter(co, &data);
2627 while (data.ret == -EINPROGRESS) {
2628 aio_poll(bdrv_get_aio_context(bs), true);
2634 static void coroutine_fn bdrv_co_aio_ioctl_entry(void *opaque)
2636 BlockAIOCBCoroutine *acb = opaque;
2637 acb->req.error = bdrv_co_do_ioctl(acb->common.bs,
2638 acb->req.req, acb->req.buf);
2639 bdrv_co_complete(acb);
2642 BlockAIOCB *bdrv_aio_ioctl(BlockDriverState *bs,
2643 unsigned long int req, void *buf,
2644 BlockCompletionFunc *cb, void *opaque)
2646 BlockAIOCBCoroutine *acb = qemu_aio_get(&bdrv_em_co_aiocb_info,
2650 acb->need_bh = true;
2651 acb->req.error = -EINPROGRESS;
2654 co = qemu_coroutine_create(bdrv_co_aio_ioctl_entry);
2655 qemu_coroutine_enter(co, acb);
2657 bdrv_co_maybe_schedule_bh(acb);
2658 return &acb->common;
2661 void *qemu_blockalign(BlockDriverState *bs, size_t size)
2663 return qemu_memalign(bdrv_opt_mem_align(bs), size);
2666 void *qemu_blockalign0(BlockDriverState *bs, size_t size)
2668 return memset(qemu_blockalign(bs, size), 0, size);
2671 void *qemu_try_blockalign(BlockDriverState *bs, size_t size)
2673 size_t align = bdrv_opt_mem_align(bs);
2675 /* Ensure that NULL is never returned on success */
2681 return qemu_try_memalign(align, size);
2684 void *qemu_try_blockalign0(BlockDriverState *bs, size_t size)
2686 void *mem = qemu_try_blockalign(bs, size);
2689 memset(mem, 0, size);
2696 * Check if all memory in this vector is sector aligned.
2698 bool bdrv_qiov_is_aligned(BlockDriverState *bs, QEMUIOVector *qiov)
2701 size_t alignment = bdrv_min_mem_align(bs);
2703 for (i = 0; i < qiov->niov; i++) {
2704 if ((uintptr_t) qiov->iov[i].iov_base % alignment) {
2707 if (qiov->iov[i].iov_len % alignment) {
2715 void bdrv_add_before_write_notifier(BlockDriverState *bs,
2716 NotifierWithReturn *notifier)
2718 notifier_with_return_list_add(&bs->before_write_notifiers, notifier);
2721 void bdrv_io_plug(BlockDriverState *bs)
2723 BlockDriver *drv = bs->drv;
2724 if (drv && drv->bdrv_io_plug) {
2725 drv->bdrv_io_plug(bs);
2726 } else if (bs->file) {
2727 bdrv_io_plug(bs->file->bs);
2731 void bdrv_io_unplug(BlockDriverState *bs)
2733 BlockDriver *drv = bs->drv;
2734 if (drv && drv->bdrv_io_unplug) {
2735 drv->bdrv_io_unplug(bs);
2736 } else if (bs->file) {
2737 bdrv_io_unplug(bs->file->bs);
2741 void bdrv_flush_io_queue(BlockDriverState *bs)
2743 BlockDriver *drv = bs->drv;
2744 if (drv && drv->bdrv_flush_io_queue) {
2745 drv->bdrv_flush_io_queue(bs);
2746 } else if (bs->file) {
2747 bdrv_flush_io_queue(bs->file->bs);
2749 bdrv_start_throttled_reqs(bs);
2752 void bdrv_drained_begin(BlockDriverState *bs)
2754 if (!bs->quiesce_counter++) {
2755 aio_disable_external(bdrv_get_aio_context(bs));
2760 void bdrv_drained_end(BlockDriverState *bs)
2762 assert(bs->quiesce_counter > 0);
2763 if (--bs->quiesce_counter > 0) {
2766 aio_enable_external(bdrv_get_aio_context(bs));