2 * Block layer I/O functions
4 * Copyright (c) 2003 Fabrice Bellard
6 * Permission is hereby granted, free of charge, to any person obtaining a copy
7 * of this software and associated documentation files (the "Software"), to deal
8 * in the Software without restriction, including without limitation the rights
9 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10 * copies of the Software, and to permit persons to whom the Software is
11 * furnished to do so, subject to the following conditions:
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
25 #include "qemu/osdep.h"
27 #include "sysemu/block-backend.h"
28 #include "block/blockjob.h"
29 #include "block/block_int.h"
30 #include "block/throttle-groups.h"
31 #include "qemu/error-report.h"
33 #define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */
35 static BlockAIOCB *bdrv_aio_readv_em(BlockDriverState *bs,
36 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
37 BlockCompletionFunc *cb, void *opaque);
38 static BlockAIOCB *bdrv_aio_writev_em(BlockDriverState *bs,
39 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
40 BlockCompletionFunc *cb, void *opaque);
41 static int coroutine_fn bdrv_co_readv_em(BlockDriverState *bs,
42 int64_t sector_num, int nb_sectors,
44 static int coroutine_fn bdrv_co_writev_em(BlockDriverState *bs,
45 int64_t sector_num, int nb_sectors,
47 static int coroutine_fn bdrv_co_do_preadv(BlockDriverState *bs,
48 int64_t offset, unsigned int bytes, QEMUIOVector *qiov,
49 BdrvRequestFlags flags);
50 static int coroutine_fn bdrv_co_do_pwritev(BlockDriverState *bs,
51 int64_t offset, unsigned int bytes, QEMUIOVector *qiov,
52 BdrvRequestFlags flags);
53 static BlockAIOCB *bdrv_co_aio_rw_vector(BlockDriverState *bs,
57 BdrvRequestFlags flags,
58 BlockCompletionFunc *cb,
61 static void coroutine_fn bdrv_co_do_rw(void *opaque);
62 static int coroutine_fn bdrv_co_do_write_zeroes(BlockDriverState *bs,
63 int64_t sector_num, int nb_sectors, BdrvRequestFlags flags);
65 /* throttling disk I/O limits */
66 void bdrv_set_io_limits(BlockDriverState *bs,
71 throttle_group_config(bs, cfg);
73 for (i = 0; i < 2; i++) {
74 qemu_co_enter_next(&bs->throttled_reqs[i]);
78 /* this function drain all the throttled IOs */
79 static bool bdrv_start_throttled_reqs(BlockDriverState *bs)
82 bool enabled = bs->io_limits_enabled;
85 bs->io_limits_enabled = false;
87 for (i = 0; i < 2; i++) {
88 while (qemu_co_enter_next(&bs->throttled_reqs[i])) {
93 bs->io_limits_enabled = enabled;
98 void bdrv_io_limits_disable(BlockDriverState *bs)
100 bs->io_limits_enabled = false;
101 bdrv_start_throttled_reqs(bs);
102 throttle_group_unregister_bs(bs);
105 /* should be called before bdrv_set_io_limits if a limit is set */
106 void bdrv_io_limits_enable(BlockDriverState *bs, const char *group)
108 assert(!bs->io_limits_enabled);
109 throttle_group_register_bs(bs, group);
110 bs->io_limits_enabled = true;
113 void bdrv_io_limits_update_group(BlockDriverState *bs, const char *group)
115 /* this bs is not part of any group */
116 if (!bs->throttle_state) {
120 /* this bs is a part of the same group than the one we want */
121 if (!g_strcmp0(throttle_group_get_name(bs), group)) {
125 /* need to change the group this bs belong to */
126 bdrv_io_limits_disable(bs);
127 bdrv_io_limits_enable(bs, group);
130 void bdrv_setup_io_funcs(BlockDriver *bdrv)
132 /* Block drivers without coroutine functions need emulation */
133 if (!bdrv->bdrv_co_readv) {
134 bdrv->bdrv_co_readv = bdrv_co_readv_em;
135 bdrv->bdrv_co_writev = bdrv_co_writev_em;
137 /* bdrv_co_readv_em()/brdv_co_writev_em() work in terms of aio, so if
138 * the block driver lacks aio we need to emulate that too.
140 if (!bdrv->bdrv_aio_readv) {
141 /* add AIO emulation layer */
142 bdrv->bdrv_aio_readv = bdrv_aio_readv_em;
143 bdrv->bdrv_aio_writev = bdrv_aio_writev_em;
148 void bdrv_refresh_limits(BlockDriverState *bs, Error **errp)
150 BlockDriver *drv = bs->drv;
151 Error *local_err = NULL;
153 memset(&bs->bl, 0, sizeof(bs->bl));
159 /* Take some limits from the children as a default */
161 bdrv_refresh_limits(bs->file->bs, &local_err);
163 error_propagate(errp, local_err);
166 bs->bl.opt_transfer_length = bs->file->bs->bl.opt_transfer_length;
167 bs->bl.max_transfer_length = bs->file->bs->bl.max_transfer_length;
168 bs->bl.min_mem_alignment = bs->file->bs->bl.min_mem_alignment;
169 bs->bl.opt_mem_alignment = bs->file->bs->bl.opt_mem_alignment;
170 bs->bl.max_iov = bs->file->bs->bl.max_iov;
172 bs->bl.min_mem_alignment = 512;
173 bs->bl.opt_mem_alignment = getpagesize();
175 /* Safe default since most protocols use readv()/writev()/etc */
176 bs->bl.max_iov = IOV_MAX;
180 bdrv_refresh_limits(bs->backing->bs, &local_err);
182 error_propagate(errp, local_err);
185 bs->bl.opt_transfer_length =
186 MAX(bs->bl.opt_transfer_length,
187 bs->backing->bs->bl.opt_transfer_length);
188 bs->bl.max_transfer_length =
189 MIN_NON_ZERO(bs->bl.max_transfer_length,
190 bs->backing->bs->bl.max_transfer_length);
191 bs->bl.opt_mem_alignment =
192 MAX(bs->bl.opt_mem_alignment,
193 bs->backing->bs->bl.opt_mem_alignment);
194 bs->bl.min_mem_alignment =
195 MAX(bs->bl.min_mem_alignment,
196 bs->backing->bs->bl.min_mem_alignment);
199 bs->backing->bs->bl.max_iov);
202 /* Then let the driver override it */
203 if (drv->bdrv_refresh_limits) {
204 drv->bdrv_refresh_limits(bs, errp);
209 * The copy-on-read flag is actually a reference count so multiple users may
210 * use the feature without worrying about clobbering its previous state.
211 * Copy-on-read stays enabled until all users have called to disable it.
213 void bdrv_enable_copy_on_read(BlockDriverState *bs)
218 void bdrv_disable_copy_on_read(BlockDriverState *bs)
220 assert(bs->copy_on_read > 0);
224 /* Check if any requests are in-flight (including throttled requests) */
225 bool bdrv_requests_pending(BlockDriverState *bs)
229 if (!QLIST_EMPTY(&bs->tracked_requests)) {
232 if (!qemu_co_queue_empty(&bs->throttled_reqs[0])) {
235 if (!qemu_co_queue_empty(&bs->throttled_reqs[1])) {
239 QLIST_FOREACH(child, &bs->children, next) {
240 if (bdrv_requests_pending(child->bs)) {
248 static void bdrv_drain_recurse(BlockDriverState *bs)
252 if (bs->drv && bs->drv->bdrv_drain) {
253 bs->drv->bdrv_drain(bs);
255 QLIST_FOREACH(child, &bs->children, next) {
256 bdrv_drain_recurse(child->bs);
261 * Wait for pending requests to complete on a single BlockDriverState subtree,
262 * and suspend block driver's internal I/O until next request arrives.
264 * Note that unlike bdrv_drain_all(), the caller must hold the BlockDriverState
267 * Only this BlockDriverState's AioContext is run, so in-flight requests must
268 * not depend on events in other AioContexts. In that case, use
269 * bdrv_drain_all() instead.
271 void bdrv_drain(BlockDriverState *bs)
275 bdrv_drain_recurse(bs);
278 bdrv_flush_io_queue(bs);
279 busy = bdrv_requests_pending(bs);
280 busy |= aio_poll(bdrv_get_aio_context(bs), busy);
285 * Wait for pending requests to complete across all BlockDriverStates
287 * This function does not flush data to disk, use bdrv_flush_all() for that
288 * after calling this function.
290 void bdrv_drain_all(void)
292 /* Always run first iteration so any pending completion BHs run */
294 BlockDriverState *bs = NULL;
295 GSList *aio_ctxs = NULL, *ctx;
297 while ((bs = bdrv_next(bs))) {
298 AioContext *aio_context = bdrv_get_aio_context(bs);
300 aio_context_acquire(aio_context);
302 block_job_pause(bs->job);
304 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;
325 aio_context_acquire(aio_context);
326 while ((bs = bdrv_next(bs))) {
327 if (aio_context == bdrv_get_aio_context(bs)) {
328 bdrv_flush_io_queue(bs);
329 if (bdrv_requests_pending(bs)) {
331 aio_poll(aio_context, busy);
335 busy |= aio_poll(aio_context, false);
336 aio_context_release(aio_context);
341 while ((bs = bdrv_next(bs))) {
342 AioContext *aio_context = bdrv_get_aio_context(bs);
344 aio_context_acquire(aio_context);
346 block_job_resume(bs->job);
348 aio_context_release(aio_context);
350 g_slist_free(aio_ctxs);
354 * Remove an active request from the tracked requests list
356 * This function should be called when a tracked request is completing.
358 static void tracked_request_end(BdrvTrackedRequest *req)
360 if (req->serialising) {
361 req->bs->serialising_in_flight--;
364 QLIST_REMOVE(req, list);
365 qemu_co_queue_restart_all(&req->wait_queue);
369 * Add an active request to the tracked requests list
371 static void tracked_request_begin(BdrvTrackedRequest *req,
372 BlockDriverState *bs,
375 enum BdrvTrackedRequestType type)
377 *req = (BdrvTrackedRequest){
382 .co = qemu_coroutine_self(),
383 .serialising = false,
384 .overlap_offset = offset,
385 .overlap_bytes = bytes,
388 qemu_co_queue_init(&req->wait_queue);
390 QLIST_INSERT_HEAD(&bs->tracked_requests, req, list);
393 static void mark_request_serialising(BdrvTrackedRequest *req, uint64_t align)
395 int64_t overlap_offset = req->offset & ~(align - 1);
396 unsigned int overlap_bytes = ROUND_UP(req->offset + req->bytes, align)
399 if (!req->serialising) {
400 req->bs->serialising_in_flight++;
401 req->serialising = true;
404 req->overlap_offset = MIN(req->overlap_offset, overlap_offset);
405 req->overlap_bytes = MAX(req->overlap_bytes, overlap_bytes);
409 * Round a region to cluster boundaries
411 void bdrv_round_to_clusters(BlockDriverState *bs,
412 int64_t sector_num, int nb_sectors,
413 int64_t *cluster_sector_num,
414 int *cluster_nb_sectors)
418 if (bdrv_get_info(bs, &bdi) < 0 || bdi.cluster_size == 0) {
419 *cluster_sector_num = sector_num;
420 *cluster_nb_sectors = nb_sectors;
422 int64_t c = bdi.cluster_size / BDRV_SECTOR_SIZE;
423 *cluster_sector_num = QEMU_ALIGN_DOWN(sector_num, c);
424 *cluster_nb_sectors = QEMU_ALIGN_UP(sector_num - *cluster_sector_num +
429 static int bdrv_get_cluster_size(BlockDriverState *bs)
434 ret = bdrv_get_info(bs, &bdi);
435 if (ret < 0 || bdi.cluster_size == 0) {
436 return bs->request_alignment;
438 return bdi.cluster_size;
442 static bool tracked_request_overlaps(BdrvTrackedRequest *req,
443 int64_t offset, unsigned int bytes)
446 if (offset >= req->overlap_offset + req->overlap_bytes) {
450 if (req->overlap_offset >= offset + bytes) {
456 static bool coroutine_fn wait_serialising_requests(BdrvTrackedRequest *self)
458 BlockDriverState *bs = self->bs;
459 BdrvTrackedRequest *req;
463 if (!bs->serialising_in_flight) {
469 QLIST_FOREACH(req, &bs->tracked_requests, list) {
470 if (req == self || (!req->serialising && !self->serialising)) {
473 if (tracked_request_overlaps(req, self->overlap_offset,
474 self->overlap_bytes))
476 /* Hitting this means there was a reentrant request, for
477 * example, a block driver issuing nested requests. This must
478 * never happen since it means deadlock.
480 assert(qemu_coroutine_self() != req->co);
482 /* If the request is already (indirectly) waiting for us, or
483 * will wait for us as soon as it wakes up, then just go on
484 * (instead of producing a deadlock in the former case). */
485 if (!req->waiting_for) {
486 self->waiting_for = req;
487 qemu_co_queue_wait(&req->wait_queue);
488 self->waiting_for = NULL;
500 static int bdrv_check_byte_request(BlockDriverState *bs, int64_t offset,
503 if (size > BDRV_REQUEST_MAX_SECTORS << BDRV_SECTOR_BITS) {
507 if (!bdrv_is_inserted(bs)) {
518 static int bdrv_check_request(BlockDriverState *bs, int64_t sector_num,
521 if (nb_sectors < 0 || nb_sectors > BDRV_REQUEST_MAX_SECTORS) {
525 return bdrv_check_byte_request(bs, sector_num * BDRV_SECTOR_SIZE,
526 nb_sectors * BDRV_SECTOR_SIZE);
529 typedef struct RwCo {
530 BlockDriverState *bs;
535 BdrvRequestFlags flags;
538 static void coroutine_fn bdrv_rw_co_entry(void *opaque)
542 if (!rwco->is_write) {
543 rwco->ret = bdrv_co_do_preadv(rwco->bs, rwco->offset,
544 rwco->qiov->size, rwco->qiov,
547 rwco->ret = bdrv_co_do_pwritev(rwco->bs, rwco->offset,
548 rwco->qiov->size, rwco->qiov,
554 * Process a vectored synchronous request using coroutines
556 static int bdrv_prwv_co(BlockDriverState *bs, int64_t offset,
557 QEMUIOVector *qiov, bool is_write,
558 BdrvRequestFlags flags)
565 .is_write = is_write,
571 * In sync call context, when the vcpu is blocked, this throttling timer
572 * will not fire; so the I/O throttling function has to be disabled here
573 * if it has been enabled.
575 if (bs->io_limits_enabled) {
576 fprintf(stderr, "Disabling I/O throttling on '%s' due "
577 "to synchronous I/O.\n", bdrv_get_device_name(bs));
578 bdrv_io_limits_disable(bs);
581 if (qemu_in_coroutine()) {
582 /* Fast-path if already in coroutine context */
583 bdrv_rw_co_entry(&rwco);
585 AioContext *aio_context = bdrv_get_aio_context(bs);
587 co = qemu_coroutine_create(bdrv_rw_co_entry);
588 qemu_coroutine_enter(co, &rwco);
589 while (rwco.ret == NOT_DONE) {
590 aio_poll(aio_context, true);
597 * Process a synchronous request using coroutines
599 static int bdrv_rw_co(BlockDriverState *bs, int64_t sector_num, uint8_t *buf,
600 int nb_sectors, bool is_write, BdrvRequestFlags flags)
604 .iov_base = (void *)buf,
605 .iov_len = nb_sectors * BDRV_SECTOR_SIZE,
608 if (nb_sectors < 0 || nb_sectors > BDRV_REQUEST_MAX_SECTORS) {
612 qemu_iovec_init_external(&qiov, &iov, 1);
613 return bdrv_prwv_co(bs, sector_num << BDRV_SECTOR_BITS,
614 &qiov, is_write, flags);
617 /* return < 0 if error. See bdrv_write() for the return codes */
618 int bdrv_read(BlockDriverState *bs, int64_t sector_num,
619 uint8_t *buf, int nb_sectors)
621 return bdrv_rw_co(bs, sector_num, buf, nb_sectors, false, 0);
624 /* Just like bdrv_read(), but with I/O throttling temporarily disabled */
625 int bdrv_read_unthrottled(BlockDriverState *bs, int64_t sector_num,
626 uint8_t *buf, int nb_sectors)
631 enabled = bs->io_limits_enabled;
632 bs->io_limits_enabled = false;
633 ret = bdrv_read(bs, sector_num, buf, nb_sectors);
634 bs->io_limits_enabled = enabled;
638 /* Return < 0 if error. Important errors are:
639 -EIO generic I/O error (may happen for all errors)
640 -ENOMEDIUM No media inserted.
641 -EINVAL Invalid sector number or nb_sectors
642 -EACCES Trying to write a read-only device
644 int bdrv_write(BlockDriverState *bs, int64_t sector_num,
645 const uint8_t *buf, int nb_sectors)
647 return bdrv_rw_co(bs, sector_num, (uint8_t *)buf, nb_sectors, true, 0);
650 int bdrv_write_zeroes(BlockDriverState *bs, int64_t sector_num,
651 int nb_sectors, BdrvRequestFlags flags)
653 return bdrv_rw_co(bs, sector_num, NULL, nb_sectors, true,
654 BDRV_REQ_ZERO_WRITE | flags);
658 * Completely zero out a block device with the help of bdrv_write_zeroes.
659 * The operation is sped up by checking the block status and only writing
660 * zeroes to the device if they currently do not return zeroes. Optional
661 * flags are passed through to bdrv_write_zeroes (e.g. BDRV_REQ_MAY_UNMAP).
663 * Returns < 0 on error, 0 on success. For error codes see bdrv_write().
665 int bdrv_make_zero(BlockDriverState *bs, BdrvRequestFlags flags)
667 int64_t target_sectors, ret, nb_sectors, sector_num = 0;
668 BlockDriverState *file;
671 target_sectors = bdrv_nb_sectors(bs);
672 if (target_sectors < 0) {
673 return target_sectors;
677 nb_sectors = MIN(target_sectors - sector_num, BDRV_REQUEST_MAX_SECTORS);
678 if (nb_sectors <= 0) {
681 ret = bdrv_get_block_status(bs, sector_num, nb_sectors, &n, &file);
683 error_report("error getting block status at sector %" PRId64 ": %s",
684 sector_num, strerror(-ret));
687 if (ret & BDRV_BLOCK_ZERO) {
691 ret = bdrv_write_zeroes(bs, sector_num, n, flags);
693 error_report("error writing zeroes at sector %" PRId64 ": %s",
694 sector_num, strerror(-ret));
701 int bdrv_pread(BlockDriverState *bs, int64_t offset, void *buf, int bytes)
705 .iov_base = (void *)buf,
714 qemu_iovec_init_external(&qiov, &iov, 1);
715 ret = bdrv_prwv_co(bs, offset, &qiov, false, 0);
723 int bdrv_pwritev(BlockDriverState *bs, int64_t offset, QEMUIOVector *qiov)
727 ret = bdrv_prwv_co(bs, offset, qiov, true, 0);
735 int bdrv_pwrite(BlockDriverState *bs, int64_t offset,
736 const void *buf, int bytes)
740 .iov_base = (void *) buf,
748 qemu_iovec_init_external(&qiov, &iov, 1);
749 return bdrv_pwritev(bs, offset, &qiov);
753 * Writes to the file and ensures that no writes are reordered across this
754 * request (acts as a barrier)
756 * Returns 0 on success, -errno in error cases.
758 int bdrv_pwrite_sync(BlockDriverState *bs, int64_t offset,
759 const void *buf, int count)
763 ret = bdrv_pwrite(bs, offset, buf, count);
768 /* No flush needed for cache modes that already do it */
769 if (bs->enable_write_cache) {
776 static int coroutine_fn bdrv_co_do_copy_on_readv(BlockDriverState *bs,
777 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov)
779 /* Perform I/O through a temporary buffer so that users who scribble over
780 * their read buffer while the operation is in progress do not end up
781 * modifying the image file. This is critical for zero-copy guest I/O
782 * where anything might happen inside guest memory.
786 BlockDriver *drv = bs->drv;
788 QEMUIOVector bounce_qiov;
789 int64_t cluster_sector_num;
790 int cluster_nb_sectors;
794 /* Cover entire cluster so no additional backing file I/O is required when
795 * allocating cluster in the image file.
797 bdrv_round_to_clusters(bs, sector_num, nb_sectors,
798 &cluster_sector_num, &cluster_nb_sectors);
800 trace_bdrv_co_do_copy_on_readv(bs, sector_num, nb_sectors,
801 cluster_sector_num, cluster_nb_sectors);
803 iov.iov_len = cluster_nb_sectors * BDRV_SECTOR_SIZE;
804 iov.iov_base = bounce_buffer = qemu_try_blockalign(bs, iov.iov_len);
805 if (bounce_buffer == NULL) {
810 qemu_iovec_init_external(&bounce_qiov, &iov, 1);
812 ret = drv->bdrv_co_readv(bs, cluster_sector_num, cluster_nb_sectors,
818 if (drv->bdrv_co_write_zeroes &&
819 buffer_is_zero(bounce_buffer, iov.iov_len)) {
820 ret = bdrv_co_do_write_zeroes(bs, cluster_sector_num,
821 cluster_nb_sectors, 0);
823 /* This does not change the data on the disk, it is not necessary
824 * to flush even in cache=writethrough mode.
826 ret = drv->bdrv_co_writev(bs, cluster_sector_num, cluster_nb_sectors,
831 /* It might be okay to ignore write errors for guest requests. If this
832 * is a deliberate copy-on-read then we don't want to ignore the error.
833 * Simply report it in all cases.
838 skip_bytes = (sector_num - cluster_sector_num) * BDRV_SECTOR_SIZE;
839 qemu_iovec_from_buf(qiov, 0, bounce_buffer + skip_bytes,
840 nb_sectors * BDRV_SECTOR_SIZE);
843 qemu_vfree(bounce_buffer);
848 * Forwards an already correctly aligned request to the BlockDriver. This
849 * handles copy on read and zeroing after EOF; any other features must be
850 * implemented by the caller.
852 static int coroutine_fn bdrv_aligned_preadv(BlockDriverState *bs,
853 BdrvTrackedRequest *req, int64_t offset, unsigned int bytes,
854 int64_t align, QEMUIOVector *qiov, int flags)
856 BlockDriver *drv = bs->drv;
859 int64_t sector_num = offset >> BDRV_SECTOR_BITS;
860 unsigned int nb_sectors = bytes >> BDRV_SECTOR_BITS;
862 assert((offset & (BDRV_SECTOR_SIZE - 1)) == 0);
863 assert((bytes & (BDRV_SECTOR_SIZE - 1)) == 0);
864 assert(!qiov || bytes == qiov->size);
866 /* Handle Copy on Read and associated serialisation */
867 if (flags & BDRV_REQ_COPY_ON_READ) {
868 /* If we touch the same cluster it counts as an overlap. This
869 * guarantees that allocating writes will be serialized and not race
870 * with each other for the same cluster. For example, in copy-on-read
871 * it ensures that the CoR read and write operations are atomic and
872 * guest writes cannot interleave between them. */
873 mark_request_serialising(req, bdrv_get_cluster_size(bs));
876 if (!(flags & BDRV_REQ_NO_SERIALISING)) {
877 wait_serialising_requests(req);
880 if (flags & BDRV_REQ_COPY_ON_READ) {
883 ret = bdrv_is_allocated(bs, sector_num, nb_sectors, &pnum);
888 if (!ret || pnum != nb_sectors) {
889 ret = bdrv_co_do_copy_on_readv(bs, sector_num, nb_sectors, qiov);
894 /* Forward the request to the BlockDriver */
895 if (!bs->zero_beyond_eof) {
896 ret = drv->bdrv_co_readv(bs, sector_num, nb_sectors, qiov);
898 /* Read zeros after EOF */
899 int64_t total_sectors, max_nb_sectors;
901 total_sectors = bdrv_nb_sectors(bs);
902 if (total_sectors < 0) {
907 max_nb_sectors = ROUND_UP(MAX(0, total_sectors - sector_num),
908 align >> BDRV_SECTOR_BITS);
909 if (nb_sectors < max_nb_sectors) {
910 ret = drv->bdrv_co_readv(bs, sector_num, nb_sectors, qiov);
911 } else if (max_nb_sectors > 0) {
912 QEMUIOVector local_qiov;
914 qemu_iovec_init(&local_qiov, qiov->niov);
915 qemu_iovec_concat(&local_qiov, qiov, 0,
916 max_nb_sectors * BDRV_SECTOR_SIZE);
918 ret = drv->bdrv_co_readv(bs, sector_num, max_nb_sectors,
921 qemu_iovec_destroy(&local_qiov);
926 /* Reading beyond end of file is supposed to produce zeroes */
927 if (ret == 0 && total_sectors < sector_num + nb_sectors) {
928 uint64_t offset = MAX(0, total_sectors - sector_num);
929 uint64_t bytes = (sector_num + nb_sectors - offset) *
931 qemu_iovec_memset(qiov, offset * BDRV_SECTOR_SIZE, 0, bytes);
940 * Handle a read request in coroutine context
942 static int coroutine_fn bdrv_co_do_preadv(BlockDriverState *bs,
943 int64_t offset, unsigned int bytes, QEMUIOVector *qiov,
944 BdrvRequestFlags flags)
946 BlockDriver *drv = bs->drv;
947 BdrvTrackedRequest req;
949 /* TODO Lift BDRV_SECTOR_SIZE restriction in BlockDriver interface */
950 uint64_t align = MAX(BDRV_SECTOR_SIZE, bs->request_alignment);
951 uint8_t *head_buf = NULL;
952 uint8_t *tail_buf = NULL;
953 QEMUIOVector local_qiov;
954 bool use_local_qiov = false;
961 ret = bdrv_check_byte_request(bs, offset, bytes);
966 /* Don't do copy-on-read if we read data before write operation */
967 if (bs->copy_on_read && !(flags & BDRV_REQ_NO_SERIALISING)) {
968 flags |= BDRV_REQ_COPY_ON_READ;
971 /* throttling disk I/O */
972 if (bs->io_limits_enabled) {
973 throttle_group_co_io_limits_intercept(bs, bytes, false);
976 /* Align read if necessary by padding qiov */
977 if (offset & (align - 1)) {
978 head_buf = qemu_blockalign(bs, align);
979 qemu_iovec_init(&local_qiov, qiov->niov + 2);
980 qemu_iovec_add(&local_qiov, head_buf, offset & (align - 1));
981 qemu_iovec_concat(&local_qiov, qiov, 0, qiov->size);
982 use_local_qiov = true;
984 bytes += offset & (align - 1);
985 offset = offset & ~(align - 1);
988 if ((offset + bytes) & (align - 1)) {
989 if (!use_local_qiov) {
990 qemu_iovec_init(&local_qiov, qiov->niov + 1);
991 qemu_iovec_concat(&local_qiov, qiov, 0, qiov->size);
992 use_local_qiov = true;
994 tail_buf = qemu_blockalign(bs, align);
995 qemu_iovec_add(&local_qiov, tail_buf,
996 align - ((offset + bytes) & (align - 1)));
998 bytes = ROUND_UP(bytes, align);
1001 tracked_request_begin(&req, bs, offset, bytes, BDRV_TRACKED_READ);
1002 ret = bdrv_aligned_preadv(bs, &req, offset, bytes, align,
1003 use_local_qiov ? &local_qiov : qiov,
1005 tracked_request_end(&req);
1007 if (use_local_qiov) {
1008 qemu_iovec_destroy(&local_qiov);
1009 qemu_vfree(head_buf);
1010 qemu_vfree(tail_buf);
1016 static int coroutine_fn bdrv_co_do_readv(BlockDriverState *bs,
1017 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov,
1018 BdrvRequestFlags flags)
1020 if (nb_sectors < 0 || nb_sectors > BDRV_REQUEST_MAX_SECTORS) {
1024 return bdrv_co_do_preadv(bs, sector_num << BDRV_SECTOR_BITS,
1025 nb_sectors << BDRV_SECTOR_BITS, qiov, flags);
1028 int coroutine_fn bdrv_co_readv(BlockDriverState *bs, int64_t sector_num,
1029 int nb_sectors, QEMUIOVector *qiov)
1031 trace_bdrv_co_readv(bs, sector_num, nb_sectors);
1033 return bdrv_co_do_readv(bs, sector_num, nb_sectors, qiov, 0);
1036 int coroutine_fn bdrv_co_readv_no_serialising(BlockDriverState *bs,
1037 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov)
1039 trace_bdrv_co_readv_no_serialising(bs, sector_num, nb_sectors);
1041 return bdrv_co_do_readv(bs, sector_num, nb_sectors, qiov,
1042 BDRV_REQ_NO_SERIALISING);
1045 int coroutine_fn bdrv_co_copy_on_readv(BlockDriverState *bs,
1046 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov)
1048 trace_bdrv_co_copy_on_readv(bs, sector_num, nb_sectors);
1050 return bdrv_co_do_readv(bs, sector_num, nb_sectors, qiov,
1051 BDRV_REQ_COPY_ON_READ);
1054 #define MAX_WRITE_ZEROES_BOUNCE_BUFFER 32768
1056 static int coroutine_fn bdrv_co_do_write_zeroes(BlockDriverState *bs,
1057 int64_t sector_num, int nb_sectors, BdrvRequestFlags flags)
1059 BlockDriver *drv = bs->drv;
1061 struct iovec iov = {0};
1064 int max_write_zeroes = MIN_NON_ZERO(bs->bl.max_write_zeroes,
1065 BDRV_REQUEST_MAX_SECTORS);
1067 while (nb_sectors > 0 && !ret) {
1068 int num = nb_sectors;
1070 /* Align request. Block drivers can expect the "bulk" of the request
1073 if (bs->bl.write_zeroes_alignment
1074 && num > bs->bl.write_zeroes_alignment) {
1075 if (sector_num % bs->bl.write_zeroes_alignment != 0) {
1076 /* Make a small request up to the first aligned sector. */
1077 num = bs->bl.write_zeroes_alignment;
1078 num -= sector_num % bs->bl.write_zeroes_alignment;
1079 } else if ((sector_num + num) % bs->bl.write_zeroes_alignment != 0) {
1080 /* Shorten the request to the last aligned sector. num cannot
1081 * underflow because num > bs->bl.write_zeroes_alignment.
1083 num -= (sector_num + num) % bs->bl.write_zeroes_alignment;
1087 /* limit request size */
1088 if (num > max_write_zeroes) {
1089 num = max_write_zeroes;
1093 /* First try the efficient write zeroes operation */
1094 if (drv->bdrv_co_write_zeroes) {
1095 ret = drv->bdrv_co_write_zeroes(bs, sector_num, num, flags);
1098 if (ret == -ENOTSUP) {
1099 /* Fall back to bounce buffer if write zeroes is unsupported */
1100 int max_xfer_len = MIN_NON_ZERO(bs->bl.max_transfer_length,
1101 MAX_WRITE_ZEROES_BOUNCE_BUFFER);
1102 num = MIN(num, max_xfer_len);
1103 iov.iov_len = num * BDRV_SECTOR_SIZE;
1104 if (iov.iov_base == NULL) {
1105 iov.iov_base = qemu_try_blockalign(bs, num * BDRV_SECTOR_SIZE);
1106 if (iov.iov_base == NULL) {
1110 memset(iov.iov_base, 0, num * BDRV_SECTOR_SIZE);
1112 qemu_iovec_init_external(&qiov, &iov, 1);
1114 ret = drv->bdrv_co_writev(bs, sector_num, num, &qiov);
1116 /* Keep bounce buffer around if it is big enough for all
1117 * all future requests.
1119 if (num < max_xfer_len) {
1120 qemu_vfree(iov.iov_base);
1121 iov.iov_base = NULL;
1130 qemu_vfree(iov.iov_base);
1135 * Forwards an already correctly aligned write request to the BlockDriver.
1137 static int coroutine_fn bdrv_aligned_pwritev(BlockDriverState *bs,
1138 BdrvTrackedRequest *req, int64_t offset, unsigned int bytes,
1139 QEMUIOVector *qiov, int flags)
1141 BlockDriver *drv = bs->drv;
1145 int64_t sector_num = offset >> BDRV_SECTOR_BITS;
1146 unsigned int nb_sectors = bytes >> BDRV_SECTOR_BITS;
1148 assert((offset & (BDRV_SECTOR_SIZE - 1)) == 0);
1149 assert((bytes & (BDRV_SECTOR_SIZE - 1)) == 0);
1150 assert(!qiov || bytes == qiov->size);
1152 waited = wait_serialising_requests(req);
1153 assert(!waited || !req->serialising);
1154 assert(req->overlap_offset <= offset);
1155 assert(offset + bytes <= req->overlap_offset + req->overlap_bytes);
1157 ret = notifier_with_return_list_notify(&bs->before_write_notifiers, req);
1159 if (!ret && bs->detect_zeroes != BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF &&
1160 !(flags & BDRV_REQ_ZERO_WRITE) && drv->bdrv_co_write_zeroes &&
1161 qemu_iovec_is_zero(qiov)) {
1162 flags |= BDRV_REQ_ZERO_WRITE;
1163 if (bs->detect_zeroes == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP) {
1164 flags |= BDRV_REQ_MAY_UNMAP;
1169 /* Do nothing, write notifier decided to fail this request */
1170 } else if (flags & BDRV_REQ_ZERO_WRITE) {
1171 bdrv_debug_event(bs, BLKDBG_PWRITEV_ZERO);
1172 ret = bdrv_co_do_write_zeroes(bs, sector_num, nb_sectors, flags);
1174 bdrv_debug_event(bs, BLKDBG_PWRITEV);
1175 ret = drv->bdrv_co_writev(bs, sector_num, nb_sectors, qiov);
1177 bdrv_debug_event(bs, BLKDBG_PWRITEV_DONE);
1179 if (ret == 0 && !bs->enable_write_cache) {
1180 ret = bdrv_co_flush(bs);
1183 bdrv_set_dirty(bs, sector_num, nb_sectors);
1185 if (bs->wr_highest_offset < offset + bytes) {
1186 bs->wr_highest_offset = offset + bytes;
1190 bs->total_sectors = MAX(bs->total_sectors, sector_num + nb_sectors);
1196 static int coroutine_fn bdrv_co_do_zero_pwritev(BlockDriverState *bs,
1199 BdrvRequestFlags flags,
1200 BdrvTrackedRequest *req)
1202 uint8_t *buf = NULL;
1203 QEMUIOVector local_qiov;
1205 uint64_t align = MAX(BDRV_SECTOR_SIZE, bs->request_alignment);
1206 unsigned int head_padding_bytes, tail_padding_bytes;
1209 head_padding_bytes = offset & (align - 1);
1210 tail_padding_bytes = align - ((offset + bytes) & (align - 1));
1213 assert(flags & BDRV_REQ_ZERO_WRITE);
1214 if (head_padding_bytes || tail_padding_bytes) {
1215 buf = qemu_blockalign(bs, align);
1216 iov = (struct iovec) {
1220 qemu_iovec_init_external(&local_qiov, &iov, 1);
1222 if (head_padding_bytes) {
1223 uint64_t zero_bytes = MIN(bytes, align - head_padding_bytes);
1225 /* RMW the unaligned part before head. */
1226 mark_request_serialising(req, align);
1227 wait_serialising_requests(req);
1228 bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_HEAD);
1229 ret = bdrv_aligned_preadv(bs, req, offset & ~(align - 1), align,
1230 align, &local_qiov, 0);
1234 bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_AFTER_HEAD);
1236 memset(buf + head_padding_bytes, 0, zero_bytes);
1237 ret = bdrv_aligned_pwritev(bs, req, offset & ~(align - 1), align,
1239 flags & ~BDRV_REQ_ZERO_WRITE);
1243 offset += zero_bytes;
1244 bytes -= zero_bytes;
1247 assert(!bytes || (offset & (align - 1)) == 0);
1248 if (bytes >= align) {
1249 /* Write the aligned part in the middle. */
1250 uint64_t aligned_bytes = bytes & ~(align - 1);
1251 ret = bdrv_aligned_pwritev(bs, req, offset, aligned_bytes,
1256 bytes -= aligned_bytes;
1257 offset += aligned_bytes;
1260 assert(!bytes || (offset & (align - 1)) == 0);
1262 assert(align == tail_padding_bytes + bytes);
1263 /* RMW the unaligned part after tail. */
1264 mark_request_serialising(req, align);
1265 wait_serialising_requests(req);
1266 bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_TAIL);
1267 ret = bdrv_aligned_preadv(bs, req, offset, align,
1268 align, &local_qiov, 0);
1272 bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_AFTER_TAIL);
1274 memset(buf, 0, bytes);
1275 ret = bdrv_aligned_pwritev(bs, req, offset, align,
1276 &local_qiov, flags & ~BDRV_REQ_ZERO_WRITE);
1285 * Handle a write request in coroutine context
1287 static int coroutine_fn bdrv_co_do_pwritev(BlockDriverState *bs,
1288 int64_t offset, unsigned int bytes, QEMUIOVector *qiov,
1289 BdrvRequestFlags flags)
1291 BdrvTrackedRequest req;
1292 /* TODO Lift BDRV_SECTOR_SIZE restriction in BlockDriver interface */
1293 uint64_t align = MAX(BDRV_SECTOR_SIZE, bs->request_alignment);
1294 uint8_t *head_buf = NULL;
1295 uint8_t *tail_buf = NULL;
1296 QEMUIOVector local_qiov;
1297 bool use_local_qiov = false;
1303 if (bs->read_only) {
1306 assert(!(bs->open_flags & BDRV_O_INACTIVE));
1308 ret = bdrv_check_byte_request(bs, offset, bytes);
1313 /* throttling disk I/O */
1314 if (bs->io_limits_enabled) {
1315 throttle_group_co_io_limits_intercept(bs, bytes, true);
1319 * Align write if necessary by performing a read-modify-write cycle.
1320 * Pad qiov with the read parts and be sure to have a tracked request not
1321 * only for bdrv_aligned_pwritev, but also for the reads of the RMW cycle.
1323 tracked_request_begin(&req, bs, offset, bytes, BDRV_TRACKED_WRITE);
1326 ret = bdrv_co_do_zero_pwritev(bs, offset, bytes, flags, &req);
1330 if (offset & (align - 1)) {
1331 QEMUIOVector head_qiov;
1332 struct iovec head_iov;
1334 mark_request_serialising(&req, align);
1335 wait_serialising_requests(&req);
1337 head_buf = qemu_blockalign(bs, align);
1338 head_iov = (struct iovec) {
1339 .iov_base = head_buf,
1342 qemu_iovec_init_external(&head_qiov, &head_iov, 1);
1344 bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_HEAD);
1345 ret = bdrv_aligned_preadv(bs, &req, offset & ~(align - 1), align,
1346 align, &head_qiov, 0);
1350 bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_AFTER_HEAD);
1352 qemu_iovec_init(&local_qiov, qiov->niov + 2);
1353 qemu_iovec_add(&local_qiov, head_buf, offset & (align - 1));
1354 qemu_iovec_concat(&local_qiov, qiov, 0, qiov->size);
1355 use_local_qiov = true;
1357 bytes += offset & (align - 1);
1358 offset = offset & ~(align - 1);
1361 if ((offset + bytes) & (align - 1)) {
1362 QEMUIOVector tail_qiov;
1363 struct iovec tail_iov;
1367 mark_request_serialising(&req, align);
1368 waited = wait_serialising_requests(&req);
1369 assert(!waited || !use_local_qiov);
1371 tail_buf = qemu_blockalign(bs, align);
1372 tail_iov = (struct iovec) {
1373 .iov_base = tail_buf,
1376 qemu_iovec_init_external(&tail_qiov, &tail_iov, 1);
1378 bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_TAIL);
1379 ret = bdrv_aligned_preadv(bs, &req, (offset + bytes) & ~(align - 1), align,
1380 align, &tail_qiov, 0);
1384 bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_AFTER_TAIL);
1386 if (!use_local_qiov) {
1387 qemu_iovec_init(&local_qiov, qiov->niov + 1);
1388 qemu_iovec_concat(&local_qiov, qiov, 0, qiov->size);
1389 use_local_qiov = true;
1392 tail_bytes = (offset + bytes) & (align - 1);
1393 qemu_iovec_add(&local_qiov, tail_buf + tail_bytes, align - tail_bytes);
1395 bytes = ROUND_UP(bytes, align);
1398 ret = bdrv_aligned_pwritev(bs, &req, offset, bytes,
1399 use_local_qiov ? &local_qiov : qiov,
1404 if (use_local_qiov) {
1405 qemu_iovec_destroy(&local_qiov);
1407 qemu_vfree(head_buf);
1408 qemu_vfree(tail_buf);
1410 tracked_request_end(&req);
1414 static int coroutine_fn bdrv_co_do_writev(BlockDriverState *bs,
1415 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov,
1416 BdrvRequestFlags flags)
1418 if (nb_sectors < 0 || nb_sectors > BDRV_REQUEST_MAX_SECTORS) {
1422 return bdrv_co_do_pwritev(bs, sector_num << BDRV_SECTOR_BITS,
1423 nb_sectors << BDRV_SECTOR_BITS, qiov, flags);
1426 int coroutine_fn bdrv_co_writev(BlockDriverState *bs, int64_t sector_num,
1427 int nb_sectors, QEMUIOVector *qiov)
1429 trace_bdrv_co_writev(bs, sector_num, nb_sectors);
1431 return bdrv_co_do_writev(bs, sector_num, nb_sectors, qiov, 0);
1434 int coroutine_fn bdrv_co_write_zeroes(BlockDriverState *bs,
1435 int64_t sector_num, int nb_sectors,
1436 BdrvRequestFlags flags)
1438 trace_bdrv_co_write_zeroes(bs, sector_num, nb_sectors, flags);
1440 if (!(bs->open_flags & BDRV_O_UNMAP)) {
1441 flags &= ~BDRV_REQ_MAY_UNMAP;
1444 return bdrv_co_do_writev(bs, sector_num, nb_sectors, NULL,
1445 BDRV_REQ_ZERO_WRITE | flags);
1448 typedef struct BdrvCoGetBlockStatusData {
1449 BlockDriverState *bs;
1450 BlockDriverState *base;
1451 BlockDriverState **file;
1457 } BdrvCoGetBlockStatusData;
1460 * Returns the allocation status of the specified sectors.
1461 * Drivers not implementing the functionality are assumed to not support
1462 * backing files, hence all their sectors are reported as allocated.
1464 * If 'sector_num' is beyond the end of the disk image the return value is 0
1465 * and 'pnum' is set to 0.
1467 * 'pnum' is set to the number of sectors (including and immediately following
1468 * the specified sector) that are known to be in the same
1469 * allocated/unallocated state.
1471 * 'nb_sectors' is the max value 'pnum' should be set to. If nb_sectors goes
1472 * beyond the end of the disk image it will be clamped.
1474 * If returned value is positive and BDRV_BLOCK_OFFSET_VALID bit is set, 'file'
1475 * points to the BDS which the sector range is allocated in.
1477 static int64_t coroutine_fn bdrv_co_get_block_status(BlockDriverState *bs,
1479 int nb_sectors, int *pnum,
1480 BlockDriverState **file)
1482 int64_t total_sectors;
1486 total_sectors = bdrv_nb_sectors(bs);
1487 if (total_sectors < 0) {
1488 return total_sectors;
1491 if (sector_num >= total_sectors) {
1496 n = total_sectors - sector_num;
1497 if (n < nb_sectors) {
1501 if (!bs->drv->bdrv_co_get_block_status) {
1503 ret = BDRV_BLOCK_DATA | BDRV_BLOCK_ALLOCATED;
1504 if (bs->drv->protocol_name) {
1505 ret |= BDRV_BLOCK_OFFSET_VALID | (sector_num * BDRV_SECTOR_SIZE);
1511 ret = bs->drv->bdrv_co_get_block_status(bs, sector_num, nb_sectors, pnum,
1518 if (ret & BDRV_BLOCK_RAW) {
1519 assert(ret & BDRV_BLOCK_OFFSET_VALID);
1520 return bdrv_get_block_status(bs->file->bs, ret >> BDRV_SECTOR_BITS,
1524 if (ret & (BDRV_BLOCK_DATA | BDRV_BLOCK_ZERO)) {
1525 ret |= BDRV_BLOCK_ALLOCATED;
1527 if (bdrv_unallocated_blocks_are_zero(bs)) {
1528 ret |= BDRV_BLOCK_ZERO;
1529 } else if (bs->backing) {
1530 BlockDriverState *bs2 = bs->backing->bs;
1531 int64_t nb_sectors2 = bdrv_nb_sectors(bs2);
1532 if (nb_sectors2 >= 0 && sector_num >= nb_sectors2) {
1533 ret |= BDRV_BLOCK_ZERO;
1538 if (*file && *file != bs &&
1539 (ret & BDRV_BLOCK_DATA) && !(ret & BDRV_BLOCK_ZERO) &&
1540 (ret & BDRV_BLOCK_OFFSET_VALID)) {
1541 BlockDriverState *file2;
1544 ret2 = bdrv_co_get_block_status(*file, ret >> BDRV_SECTOR_BITS,
1545 *pnum, &file_pnum, &file2);
1547 /* Ignore errors. This is just providing extra information, it
1548 * is useful but not necessary.
1551 /* !file_pnum indicates an offset at or beyond the EOF; it is
1552 * perfectly valid for the format block driver to point to such
1553 * offsets, so catch it and mark everything as zero */
1554 ret |= BDRV_BLOCK_ZERO;
1556 /* Limit request to the range reported by the protocol driver */
1558 ret |= (ret2 & BDRV_BLOCK_ZERO);
1566 static int64_t coroutine_fn bdrv_co_get_block_status_above(BlockDriverState *bs,
1567 BlockDriverState *base,
1571 BlockDriverState **file)
1573 BlockDriverState *p;
1577 for (p = bs; p != base; p = backing_bs(p)) {
1578 ret = bdrv_co_get_block_status(p, sector_num, nb_sectors, pnum, file);
1579 if (ret < 0 || ret & BDRV_BLOCK_ALLOCATED) {
1582 /* [sector_num, pnum] unallocated on this layer, which could be only
1583 * the first part of [sector_num, nb_sectors]. */
1584 nb_sectors = MIN(nb_sectors, *pnum);
1589 /* Coroutine wrapper for bdrv_get_block_status_above() */
1590 static void coroutine_fn bdrv_get_block_status_above_co_entry(void *opaque)
1592 BdrvCoGetBlockStatusData *data = opaque;
1594 data->ret = bdrv_co_get_block_status_above(data->bs, data->base,
1603 * Synchronous wrapper around bdrv_co_get_block_status_above().
1605 * See bdrv_co_get_block_status_above() for details.
1607 int64_t bdrv_get_block_status_above(BlockDriverState *bs,
1608 BlockDriverState *base,
1610 int nb_sectors, int *pnum,
1611 BlockDriverState **file)
1614 BdrvCoGetBlockStatusData data = {
1618 .sector_num = sector_num,
1619 .nb_sectors = nb_sectors,
1624 if (qemu_in_coroutine()) {
1625 /* Fast-path if already in coroutine context */
1626 bdrv_get_block_status_above_co_entry(&data);
1628 AioContext *aio_context = bdrv_get_aio_context(bs);
1630 co = qemu_coroutine_create(bdrv_get_block_status_above_co_entry);
1631 qemu_coroutine_enter(co, &data);
1632 while (!data.done) {
1633 aio_poll(aio_context, true);
1639 int64_t bdrv_get_block_status(BlockDriverState *bs,
1641 int nb_sectors, int *pnum,
1642 BlockDriverState **file)
1644 return bdrv_get_block_status_above(bs, backing_bs(bs),
1645 sector_num, nb_sectors, pnum, file);
1648 int coroutine_fn bdrv_is_allocated(BlockDriverState *bs, int64_t sector_num,
1649 int nb_sectors, int *pnum)
1651 BlockDriverState *file;
1652 int64_t ret = bdrv_get_block_status(bs, sector_num, nb_sectors, pnum,
1657 return !!(ret & BDRV_BLOCK_ALLOCATED);
1661 * Given an image chain: ... -> [BASE] -> [INTER1] -> [INTER2] -> [TOP]
1663 * Return true if the given sector is allocated in any image between
1664 * BASE and TOP (inclusive). BASE can be NULL to check if the given
1665 * sector is allocated in any image of the chain. Return false otherwise.
1667 * 'pnum' is set to the number of sectors (including and immediately following
1668 * the specified sector) that are known to be in the same
1669 * allocated/unallocated state.
1672 int bdrv_is_allocated_above(BlockDriverState *top,
1673 BlockDriverState *base,
1675 int nb_sectors, int *pnum)
1677 BlockDriverState *intermediate;
1678 int ret, n = nb_sectors;
1681 while (intermediate && intermediate != base) {
1683 ret = bdrv_is_allocated(intermediate, sector_num, nb_sectors,
1693 * [sector_num, nb_sectors] is unallocated on top but intermediate
1696 * [sector_num+x, nr_sectors] allocated.
1698 if (n > pnum_inter &&
1699 (intermediate == top ||
1700 sector_num + pnum_inter < intermediate->total_sectors)) {
1704 intermediate = backing_bs(intermediate);
1711 int bdrv_write_compressed(BlockDriverState *bs, int64_t sector_num,
1712 const uint8_t *buf, int nb_sectors)
1714 BlockDriver *drv = bs->drv;
1720 if (!drv->bdrv_write_compressed) {
1723 ret = bdrv_check_request(bs, sector_num, nb_sectors);
1728 assert(QLIST_EMPTY(&bs->dirty_bitmaps));
1730 return drv->bdrv_write_compressed(bs, sector_num, buf, nb_sectors);
1733 int bdrv_save_vmstate(BlockDriverState *bs, const uint8_t *buf,
1734 int64_t pos, int size)
1737 struct iovec iov = {
1738 .iov_base = (void *) buf,
1742 qemu_iovec_init_external(&qiov, &iov, 1);
1743 return bdrv_writev_vmstate(bs, &qiov, pos);
1746 int bdrv_writev_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, int64_t pos)
1748 BlockDriver *drv = bs->drv;
1752 } else if (drv->bdrv_save_vmstate) {
1753 return drv->bdrv_save_vmstate(bs, qiov, pos);
1754 } else if (bs->file) {
1755 return bdrv_writev_vmstate(bs->file->bs, qiov, pos);
1761 int bdrv_load_vmstate(BlockDriverState *bs, uint8_t *buf,
1762 int64_t pos, int size)
1764 BlockDriver *drv = bs->drv;
1767 if (drv->bdrv_load_vmstate)
1768 return drv->bdrv_load_vmstate(bs, buf, pos, size);
1770 return bdrv_load_vmstate(bs->file->bs, buf, pos, size);
1774 /**************************************************************/
1777 BlockAIOCB *bdrv_aio_readv(BlockDriverState *bs, int64_t sector_num,
1778 QEMUIOVector *qiov, int nb_sectors,
1779 BlockCompletionFunc *cb, void *opaque)
1781 trace_bdrv_aio_readv(bs, sector_num, nb_sectors, opaque);
1783 return bdrv_co_aio_rw_vector(bs, sector_num, qiov, nb_sectors, 0,
1787 BlockAIOCB *bdrv_aio_writev(BlockDriverState *bs, int64_t sector_num,
1788 QEMUIOVector *qiov, int nb_sectors,
1789 BlockCompletionFunc *cb, void *opaque)
1791 trace_bdrv_aio_writev(bs, sector_num, nb_sectors, opaque);
1793 return bdrv_co_aio_rw_vector(bs, sector_num, qiov, nb_sectors, 0,
1797 BlockAIOCB *bdrv_aio_write_zeroes(BlockDriverState *bs,
1798 int64_t sector_num, int nb_sectors, BdrvRequestFlags flags,
1799 BlockCompletionFunc *cb, void *opaque)
1801 trace_bdrv_aio_write_zeroes(bs, sector_num, nb_sectors, flags, opaque);
1803 return bdrv_co_aio_rw_vector(bs, sector_num, NULL, nb_sectors,
1804 BDRV_REQ_ZERO_WRITE | flags,
1809 typedef struct MultiwriteCB {
1814 BlockCompletionFunc *cb;
1816 QEMUIOVector *free_qiov;
1820 static void multiwrite_user_cb(MultiwriteCB *mcb)
1824 for (i = 0; i < mcb->num_callbacks; i++) {
1825 mcb->callbacks[i].cb(mcb->callbacks[i].opaque, mcb->error);
1826 if (mcb->callbacks[i].free_qiov) {
1827 qemu_iovec_destroy(mcb->callbacks[i].free_qiov);
1829 g_free(mcb->callbacks[i].free_qiov);
1833 static void multiwrite_cb(void *opaque, int ret)
1835 MultiwriteCB *mcb = opaque;
1837 trace_multiwrite_cb(mcb, ret);
1839 if (ret < 0 && !mcb->error) {
1843 mcb->num_requests--;
1844 if (mcb->num_requests == 0) {
1845 multiwrite_user_cb(mcb);
1850 static int multiwrite_req_compare(const void *a, const void *b)
1852 const BlockRequest *req1 = a, *req2 = b;
1855 * Note that we can't simply subtract req2->sector from req1->sector
1856 * here as that could overflow the return value.
1858 if (req1->sector > req2->sector) {
1860 } else if (req1->sector < req2->sector) {
1868 * Takes a bunch of requests and tries to merge them. Returns the number of
1869 * requests that remain after merging.
1871 static int multiwrite_merge(BlockDriverState *bs, BlockRequest *reqs,
1872 int num_reqs, MultiwriteCB *mcb)
1876 // Sort requests by start sector
1877 qsort(reqs, num_reqs, sizeof(*reqs), &multiwrite_req_compare);
1879 // Check if adjacent requests touch the same clusters. If so, combine them,
1880 // filling up gaps with zero sectors.
1882 for (i = 1; i < num_reqs; i++) {
1884 int64_t oldreq_last = reqs[outidx].sector + reqs[outidx].nb_sectors;
1886 // Handle exactly sequential writes and overlapping writes.
1887 if (reqs[i].sector <= oldreq_last) {
1891 if (reqs[outidx].qiov->niov + reqs[i].qiov->niov + 1 >
1896 if (bs->bl.max_transfer_length && reqs[outidx].nb_sectors +
1897 reqs[i].nb_sectors > bs->bl.max_transfer_length) {
1903 QEMUIOVector *qiov = g_malloc0(sizeof(*qiov));
1904 qemu_iovec_init(qiov,
1905 reqs[outidx].qiov->niov + reqs[i].qiov->niov + 1);
1907 // Add the first request to the merged one. If the requests are
1908 // overlapping, drop the last sectors of the first request.
1909 size = (reqs[i].sector - reqs[outidx].sector) << 9;
1910 qemu_iovec_concat(qiov, reqs[outidx].qiov, 0, size);
1912 // We should need to add any zeros between the two requests
1913 assert (reqs[i].sector <= oldreq_last);
1915 // Add the second request
1916 qemu_iovec_concat(qiov, reqs[i].qiov, 0, reqs[i].qiov->size);
1918 // Add tail of first request, if necessary
1919 if (qiov->size < reqs[outidx].qiov->size) {
1920 qemu_iovec_concat(qiov, reqs[outidx].qiov, qiov->size,
1921 reqs[outidx].qiov->size - qiov->size);
1924 reqs[outidx].nb_sectors = qiov->size >> 9;
1925 reqs[outidx].qiov = qiov;
1927 mcb->callbacks[i].free_qiov = reqs[outidx].qiov;
1930 reqs[outidx].sector = reqs[i].sector;
1931 reqs[outidx].nb_sectors = reqs[i].nb_sectors;
1932 reqs[outidx].qiov = reqs[i].qiov;
1937 block_acct_merge_done(blk_get_stats(bs->blk), BLOCK_ACCT_WRITE,
1938 num_reqs - outidx - 1);
1945 * Submit multiple AIO write requests at once.
1947 * On success, the function returns 0 and all requests in the reqs array have
1948 * been submitted. In error case this function returns -1, and any of the
1949 * requests may or may not be submitted yet. In particular, this means that the
1950 * callback will be called for some of the requests, for others it won't. The
1951 * caller must check the error field of the BlockRequest to wait for the right
1952 * callbacks (if error != 0, no callback will be called).
1954 * The implementation may modify the contents of the reqs array, e.g. to merge
1955 * requests. However, the fields opaque and error are left unmodified as they
1956 * are used to signal failure for a single request to the caller.
1958 int bdrv_aio_multiwrite(BlockDriverState *bs, BlockRequest *reqs, int num_reqs)
1963 /* don't submit writes if we don't have a medium */
1964 if (bs->drv == NULL) {
1965 for (i = 0; i < num_reqs; i++) {
1966 reqs[i].error = -ENOMEDIUM;
1971 if (num_reqs == 0) {
1975 // Create MultiwriteCB structure
1976 mcb = g_malloc0(sizeof(*mcb) + num_reqs * sizeof(*mcb->callbacks));
1977 mcb->num_requests = 0;
1978 mcb->num_callbacks = num_reqs;
1980 for (i = 0; i < num_reqs; i++) {
1981 mcb->callbacks[i].cb = reqs[i].cb;
1982 mcb->callbacks[i].opaque = reqs[i].opaque;
1985 // Check for mergable requests
1986 num_reqs = multiwrite_merge(bs, reqs, num_reqs, mcb);
1988 trace_bdrv_aio_multiwrite(mcb, mcb->num_callbacks, num_reqs);
1990 /* Run the aio requests. */
1991 mcb->num_requests = num_reqs;
1992 for (i = 0; i < num_reqs; i++) {
1993 bdrv_co_aio_rw_vector(bs, reqs[i].sector, reqs[i].qiov,
1994 reqs[i].nb_sectors, reqs[i].flags,
2002 void bdrv_aio_cancel(BlockAIOCB *acb)
2005 bdrv_aio_cancel_async(acb);
2006 while (acb->refcnt > 1) {
2007 if (acb->aiocb_info->get_aio_context) {
2008 aio_poll(acb->aiocb_info->get_aio_context(acb), true);
2009 } else if (acb->bs) {
2010 aio_poll(bdrv_get_aio_context(acb->bs), true);
2015 qemu_aio_unref(acb);
2018 /* Async version of aio cancel. The caller is not blocked if the acb implements
2019 * cancel_async, otherwise we do nothing and let the request normally complete.
2020 * In either case the completion callback must be called. */
2021 void bdrv_aio_cancel_async(BlockAIOCB *acb)
2023 if (acb->aiocb_info->cancel_async) {
2024 acb->aiocb_info->cancel_async(acb);
2028 /**************************************************************/
2029 /* async block device emulation */
2031 typedef struct BlockAIOCBSync {
2035 /* vector translation state */
2041 static const AIOCBInfo bdrv_em_aiocb_info = {
2042 .aiocb_size = sizeof(BlockAIOCBSync),
2045 static void bdrv_aio_bh_cb(void *opaque)
2047 BlockAIOCBSync *acb = opaque;
2049 if (!acb->is_write && acb->ret >= 0) {
2050 qemu_iovec_from_buf(acb->qiov, 0, acb->bounce, acb->qiov->size);
2052 qemu_vfree(acb->bounce);
2053 acb->common.cb(acb->common.opaque, acb->ret);
2054 qemu_bh_delete(acb->bh);
2056 qemu_aio_unref(acb);
2059 static BlockAIOCB *bdrv_aio_rw_vector(BlockDriverState *bs,
2063 BlockCompletionFunc *cb,
2068 BlockAIOCBSync *acb;
2070 acb = qemu_aio_get(&bdrv_em_aiocb_info, bs, cb, opaque);
2071 acb->is_write = is_write;
2073 acb->bounce = qemu_try_blockalign(bs, qiov->size);
2074 acb->bh = aio_bh_new(bdrv_get_aio_context(bs), bdrv_aio_bh_cb, acb);
2076 if (acb->bounce == NULL) {
2078 } else if (is_write) {
2079 qemu_iovec_to_buf(acb->qiov, 0, acb->bounce, qiov->size);
2080 acb->ret = bs->drv->bdrv_write(bs, sector_num, acb->bounce, nb_sectors);
2082 acb->ret = bs->drv->bdrv_read(bs, sector_num, acb->bounce, nb_sectors);
2085 qemu_bh_schedule(acb->bh);
2087 return &acb->common;
2090 static BlockAIOCB *bdrv_aio_readv_em(BlockDriverState *bs,
2091 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
2092 BlockCompletionFunc *cb, void *opaque)
2094 return bdrv_aio_rw_vector(bs, sector_num, qiov, nb_sectors, cb, opaque, 0);
2097 static BlockAIOCB *bdrv_aio_writev_em(BlockDriverState *bs,
2098 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
2099 BlockCompletionFunc *cb, void *opaque)
2101 return bdrv_aio_rw_vector(bs, sector_num, qiov, nb_sectors, cb, opaque, 1);
2105 typedef struct BlockAIOCBCoroutine {
2112 } BlockAIOCBCoroutine;
2114 static const AIOCBInfo bdrv_em_co_aiocb_info = {
2115 .aiocb_size = sizeof(BlockAIOCBCoroutine),
2118 static void bdrv_co_complete(BlockAIOCBCoroutine *acb)
2120 if (!acb->need_bh) {
2121 acb->common.cb(acb->common.opaque, acb->req.error);
2122 qemu_aio_unref(acb);
2126 static void bdrv_co_em_bh(void *opaque)
2128 BlockAIOCBCoroutine *acb = opaque;
2130 assert(!acb->need_bh);
2131 qemu_bh_delete(acb->bh);
2132 bdrv_co_complete(acb);
2135 static void bdrv_co_maybe_schedule_bh(BlockAIOCBCoroutine *acb)
2137 acb->need_bh = false;
2138 if (acb->req.error != -EINPROGRESS) {
2139 BlockDriverState *bs = acb->common.bs;
2141 acb->bh = aio_bh_new(bdrv_get_aio_context(bs), bdrv_co_em_bh, acb);
2142 qemu_bh_schedule(acb->bh);
2146 /* Invoke bdrv_co_do_readv/bdrv_co_do_writev */
2147 static void coroutine_fn bdrv_co_do_rw(void *opaque)
2149 BlockAIOCBCoroutine *acb = opaque;
2150 BlockDriverState *bs = acb->common.bs;
2152 if (!acb->is_write) {
2153 acb->req.error = bdrv_co_do_readv(bs, acb->req.sector,
2154 acb->req.nb_sectors, acb->req.qiov, acb->req.flags);
2156 acb->req.error = bdrv_co_do_writev(bs, acb->req.sector,
2157 acb->req.nb_sectors, acb->req.qiov, acb->req.flags);
2160 bdrv_co_complete(acb);
2163 static BlockAIOCB *bdrv_co_aio_rw_vector(BlockDriverState *bs,
2167 BdrvRequestFlags flags,
2168 BlockCompletionFunc *cb,
2173 BlockAIOCBCoroutine *acb;
2175 acb = qemu_aio_get(&bdrv_em_co_aiocb_info, bs, cb, opaque);
2176 acb->need_bh = true;
2177 acb->req.error = -EINPROGRESS;
2178 acb->req.sector = sector_num;
2179 acb->req.nb_sectors = nb_sectors;
2180 acb->req.qiov = qiov;
2181 acb->req.flags = flags;
2182 acb->is_write = is_write;
2184 co = qemu_coroutine_create(bdrv_co_do_rw);
2185 qemu_coroutine_enter(co, acb);
2187 bdrv_co_maybe_schedule_bh(acb);
2188 return &acb->common;
2191 static void coroutine_fn bdrv_aio_flush_co_entry(void *opaque)
2193 BlockAIOCBCoroutine *acb = opaque;
2194 BlockDriverState *bs = acb->common.bs;
2196 acb->req.error = bdrv_co_flush(bs);
2197 bdrv_co_complete(acb);
2200 BlockAIOCB *bdrv_aio_flush(BlockDriverState *bs,
2201 BlockCompletionFunc *cb, void *opaque)
2203 trace_bdrv_aio_flush(bs, opaque);
2206 BlockAIOCBCoroutine *acb;
2208 acb = qemu_aio_get(&bdrv_em_co_aiocb_info, bs, cb, opaque);
2209 acb->need_bh = true;
2210 acb->req.error = -EINPROGRESS;
2212 co = qemu_coroutine_create(bdrv_aio_flush_co_entry);
2213 qemu_coroutine_enter(co, acb);
2215 bdrv_co_maybe_schedule_bh(acb);
2216 return &acb->common;
2219 static void coroutine_fn bdrv_aio_discard_co_entry(void *opaque)
2221 BlockAIOCBCoroutine *acb = opaque;
2222 BlockDriverState *bs = acb->common.bs;
2224 acb->req.error = bdrv_co_discard(bs, acb->req.sector, acb->req.nb_sectors);
2225 bdrv_co_complete(acb);
2228 BlockAIOCB *bdrv_aio_discard(BlockDriverState *bs,
2229 int64_t sector_num, int nb_sectors,
2230 BlockCompletionFunc *cb, void *opaque)
2233 BlockAIOCBCoroutine *acb;
2235 trace_bdrv_aio_discard(bs, sector_num, nb_sectors, opaque);
2237 acb = qemu_aio_get(&bdrv_em_co_aiocb_info, bs, cb, opaque);
2238 acb->need_bh = true;
2239 acb->req.error = -EINPROGRESS;
2240 acb->req.sector = sector_num;
2241 acb->req.nb_sectors = nb_sectors;
2242 co = qemu_coroutine_create(bdrv_aio_discard_co_entry);
2243 qemu_coroutine_enter(co, acb);
2245 bdrv_co_maybe_schedule_bh(acb);
2246 return &acb->common;
2249 void *qemu_aio_get(const AIOCBInfo *aiocb_info, BlockDriverState *bs,
2250 BlockCompletionFunc *cb, void *opaque)
2254 acb = g_malloc(aiocb_info->aiocb_size);
2255 acb->aiocb_info = aiocb_info;
2258 acb->opaque = opaque;
2263 void qemu_aio_ref(void *p)
2265 BlockAIOCB *acb = p;
2269 void qemu_aio_unref(void *p)
2271 BlockAIOCB *acb = p;
2272 assert(acb->refcnt > 0);
2273 if (--acb->refcnt == 0) {
2278 /**************************************************************/
2279 /* Coroutine block device emulation */
2281 typedef struct CoroutineIOCompletion {
2282 Coroutine *coroutine;
2284 } CoroutineIOCompletion;
2286 static void bdrv_co_io_em_complete(void *opaque, int ret)
2288 CoroutineIOCompletion *co = opaque;
2291 qemu_coroutine_enter(co->coroutine, NULL);
2294 static int coroutine_fn bdrv_co_io_em(BlockDriverState *bs, int64_t sector_num,
2295 int nb_sectors, QEMUIOVector *iov,
2298 CoroutineIOCompletion co = {
2299 .coroutine = qemu_coroutine_self(),
2304 acb = bs->drv->bdrv_aio_writev(bs, sector_num, iov, nb_sectors,
2305 bdrv_co_io_em_complete, &co);
2307 acb = bs->drv->bdrv_aio_readv(bs, sector_num, iov, nb_sectors,
2308 bdrv_co_io_em_complete, &co);
2311 trace_bdrv_co_io_em(bs, sector_num, nb_sectors, is_write, acb);
2315 qemu_coroutine_yield();
2320 static int coroutine_fn bdrv_co_readv_em(BlockDriverState *bs,
2321 int64_t sector_num, int nb_sectors,
2324 return bdrv_co_io_em(bs, sector_num, nb_sectors, iov, false);
2327 static int coroutine_fn bdrv_co_writev_em(BlockDriverState *bs,
2328 int64_t sector_num, int nb_sectors,
2331 return bdrv_co_io_em(bs, sector_num, nb_sectors, iov, true);
2334 static void coroutine_fn bdrv_flush_co_entry(void *opaque)
2336 RwCo *rwco = opaque;
2338 rwco->ret = bdrv_co_flush(rwco->bs);
2341 int coroutine_fn bdrv_co_flush(BlockDriverState *bs)
2344 BdrvTrackedRequest req;
2346 if (!bs || !bdrv_is_inserted(bs) || bdrv_is_read_only(bs) ||
2351 tracked_request_begin(&req, bs, 0, 0, BDRV_TRACKED_FLUSH);
2352 /* Write back cached data to the OS even with cache=unsafe */
2353 BLKDBG_EVENT(bs->file, BLKDBG_FLUSH_TO_OS);
2354 if (bs->drv->bdrv_co_flush_to_os) {
2355 ret = bs->drv->bdrv_co_flush_to_os(bs);
2361 /* But don't actually force it to the disk with cache=unsafe */
2362 if (bs->open_flags & BDRV_O_NO_FLUSH) {
2366 BLKDBG_EVENT(bs->file, BLKDBG_FLUSH_TO_DISK);
2367 if (bs->drv->bdrv_co_flush_to_disk) {
2368 ret = bs->drv->bdrv_co_flush_to_disk(bs);
2369 } else if (bs->drv->bdrv_aio_flush) {
2371 CoroutineIOCompletion co = {
2372 .coroutine = qemu_coroutine_self(),
2375 acb = bs->drv->bdrv_aio_flush(bs, bdrv_co_io_em_complete, &co);
2379 qemu_coroutine_yield();
2384 * Some block drivers always operate in either writethrough or unsafe
2385 * mode and don't support bdrv_flush therefore. Usually qemu doesn't
2386 * know how the server works (because the behaviour is hardcoded or
2387 * depends on server-side configuration), so we can't ensure that
2388 * everything is safe on disk. Returning an error doesn't work because
2389 * that would break guests even if the server operates in writethrough
2392 * Let's hope the user knows what he's doing.
2400 /* Now flush the underlying protocol. It will also have BDRV_O_NO_FLUSH
2401 * in the case of cache=unsafe, so there are no useless flushes.
2404 ret = bs->file ? bdrv_co_flush(bs->file->bs) : 0;
2406 tracked_request_end(&req);
2410 int bdrv_flush(BlockDriverState *bs)
2418 if (qemu_in_coroutine()) {
2419 /* Fast-path if already in coroutine context */
2420 bdrv_flush_co_entry(&rwco);
2422 AioContext *aio_context = bdrv_get_aio_context(bs);
2424 co = qemu_coroutine_create(bdrv_flush_co_entry);
2425 qemu_coroutine_enter(co, &rwco);
2426 while (rwco.ret == NOT_DONE) {
2427 aio_poll(aio_context, true);
2434 typedef struct DiscardCo {
2435 BlockDriverState *bs;
2440 static void coroutine_fn bdrv_discard_co_entry(void *opaque)
2442 DiscardCo *rwco = opaque;
2444 rwco->ret = bdrv_co_discard(rwco->bs, rwco->sector_num, rwco->nb_sectors);
2447 int coroutine_fn bdrv_co_discard(BlockDriverState *bs, int64_t sector_num,
2450 BdrvTrackedRequest req;
2451 int max_discard, ret;
2457 ret = bdrv_check_request(bs, sector_num, nb_sectors);
2460 } else if (bs->read_only) {
2463 assert(!(bs->open_flags & BDRV_O_INACTIVE));
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));