2 * Block layer I/O functions
4 * Copyright (c) 2003 Fabrice Bellard
6 * Permission is hereby granted, free of charge, to any person obtaining a copy
7 * of this software and associated documentation files (the "Software"), to deal
8 * in the Software without restriction, including without limitation the rights
9 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10 * copies of the Software, and to permit persons to whom the Software is
11 * furnished to do so, subject to the following conditions:
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
25 #include "qemu/osdep.h"
27 #include "sysemu/block-backend.h"
28 #include "block/blockjob.h"
29 #include "block/block_int.h"
30 #include "qemu/cutils.h"
31 #include "qapi/error.h"
32 #include "qemu/error-report.h"
34 #define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */
36 static BlockAIOCB *bdrv_co_aio_rw_vector(BlockDriverState *bs,
40 BdrvRequestFlags flags,
41 BlockCompletionFunc *cb,
44 static void coroutine_fn bdrv_co_do_rw(void *opaque);
45 static int coroutine_fn bdrv_co_do_pwrite_zeroes(BlockDriverState *bs,
46 int64_t offset, int count, BdrvRequestFlags flags);
48 static void bdrv_parent_drained_begin(BlockDriverState *bs)
52 QLIST_FOREACH(c, &bs->parents, next_parent) {
53 if (c->role->drained_begin) {
54 c->role->drained_begin(c);
59 static void bdrv_parent_drained_end(BlockDriverState *bs)
63 QLIST_FOREACH(c, &bs->parents, next_parent) {
64 if (c->role->drained_end) {
65 c->role->drained_end(c);
70 void bdrv_refresh_limits(BlockDriverState *bs, Error **errp)
72 BlockDriver *drv = bs->drv;
73 Error *local_err = NULL;
75 memset(&bs->bl, 0, sizeof(bs->bl));
81 /* Take some limits from the children as a default */
83 bdrv_refresh_limits(bs->file->bs, &local_err);
85 error_propagate(errp, local_err);
88 bs->bl.opt_transfer_length = bs->file->bs->bl.opt_transfer_length;
89 bs->bl.max_transfer_length = bs->file->bs->bl.max_transfer_length;
90 bs->bl.min_mem_alignment = bs->file->bs->bl.min_mem_alignment;
91 bs->bl.opt_mem_alignment = bs->file->bs->bl.opt_mem_alignment;
92 bs->bl.max_iov = bs->file->bs->bl.max_iov;
94 bs->bl.min_mem_alignment = 512;
95 bs->bl.opt_mem_alignment = getpagesize();
97 /* Safe default since most protocols use readv()/writev()/etc */
98 bs->bl.max_iov = IOV_MAX;
102 bdrv_refresh_limits(bs->backing->bs, &local_err);
104 error_propagate(errp, local_err);
107 bs->bl.opt_transfer_length =
108 MAX(bs->bl.opt_transfer_length,
109 bs->backing->bs->bl.opt_transfer_length);
110 bs->bl.max_transfer_length =
111 MIN_NON_ZERO(bs->bl.max_transfer_length,
112 bs->backing->bs->bl.max_transfer_length);
113 bs->bl.opt_mem_alignment =
114 MAX(bs->bl.opt_mem_alignment,
115 bs->backing->bs->bl.opt_mem_alignment);
116 bs->bl.min_mem_alignment =
117 MAX(bs->bl.min_mem_alignment,
118 bs->backing->bs->bl.min_mem_alignment);
121 bs->backing->bs->bl.max_iov);
124 /* Then let the driver override it */
125 if (drv->bdrv_refresh_limits) {
126 drv->bdrv_refresh_limits(bs, errp);
131 * The copy-on-read flag is actually a reference count so multiple users may
132 * use the feature without worrying about clobbering its previous state.
133 * Copy-on-read stays enabled until all users have called to disable it.
135 void bdrv_enable_copy_on_read(BlockDriverState *bs)
140 void bdrv_disable_copy_on_read(BlockDriverState *bs)
142 assert(bs->copy_on_read > 0);
146 /* Check if any requests are in-flight (including throttled requests) */
147 bool bdrv_requests_pending(BlockDriverState *bs)
151 if (!QLIST_EMPTY(&bs->tracked_requests)) {
155 QLIST_FOREACH(child, &bs->children, next) {
156 if (bdrv_requests_pending(child->bs)) {
164 static void bdrv_drain_recurse(BlockDriverState *bs)
168 if (bs->drv && bs->drv->bdrv_drain) {
169 bs->drv->bdrv_drain(bs);
171 QLIST_FOREACH(child, &bs->children, next) {
172 bdrv_drain_recurse(child->bs);
178 BlockDriverState *bs;
183 static void bdrv_drain_poll(BlockDriverState *bs)
189 busy = bdrv_requests_pending(bs);
190 busy |= aio_poll(bdrv_get_aio_context(bs), busy);
194 static void bdrv_co_drain_bh_cb(void *opaque)
196 BdrvCoDrainData *data = opaque;
197 Coroutine *co = data->co;
199 qemu_bh_delete(data->bh);
200 bdrv_drain_poll(data->bs);
202 qemu_coroutine_enter(co, NULL);
205 static void coroutine_fn bdrv_co_yield_to_drain(BlockDriverState *bs)
207 BdrvCoDrainData data;
209 /* Calling bdrv_drain() from a BH ensures the current coroutine yields and
210 * other coroutines run if they were queued from
211 * qemu_co_queue_run_restart(). */
213 assert(qemu_in_coroutine());
214 data = (BdrvCoDrainData) {
215 .co = qemu_coroutine_self(),
218 .bh = aio_bh_new(bdrv_get_aio_context(bs), bdrv_co_drain_bh_cb, &data),
220 qemu_bh_schedule(data.bh);
222 qemu_coroutine_yield();
223 /* If we are resumed from some other event (such as an aio completion or a
224 * timer callback), it is a bug in the caller that should be fixed. */
228 void bdrv_drained_begin(BlockDriverState *bs)
230 if (!bs->quiesce_counter++) {
231 aio_disable_external(bdrv_get_aio_context(bs));
232 bdrv_parent_drained_begin(bs);
235 bdrv_io_unplugged_begin(bs);
236 bdrv_drain_recurse(bs);
237 if (qemu_in_coroutine()) {
238 bdrv_co_yield_to_drain(bs);
242 bdrv_io_unplugged_end(bs);
245 void bdrv_drained_end(BlockDriverState *bs)
247 assert(bs->quiesce_counter > 0);
248 if (--bs->quiesce_counter > 0) {
252 bdrv_parent_drained_end(bs);
253 aio_enable_external(bdrv_get_aio_context(bs));
257 * Wait for pending requests to complete on a single BlockDriverState subtree,
258 * and suspend block driver's internal I/O until next request arrives.
260 * Note that unlike bdrv_drain_all(), the caller must hold the BlockDriverState
263 * Only this BlockDriverState's AioContext is run, so in-flight requests must
264 * not depend on events in other AioContexts. In that case, use
265 * bdrv_drain_all() instead.
267 void coroutine_fn bdrv_co_drain(BlockDriverState *bs)
269 assert(qemu_in_coroutine());
270 bdrv_drained_begin(bs);
271 bdrv_drained_end(bs);
274 void bdrv_drain(BlockDriverState *bs)
276 bdrv_drained_begin(bs);
277 bdrv_drained_end(bs);
281 * Wait for pending requests to complete across all BlockDriverStates
283 * This function does not flush data to disk, use bdrv_flush_all() for that
284 * after calling this function.
286 void bdrv_drain_all(void)
288 /* Always run first iteration so any pending completion BHs run */
290 BlockDriverState *bs;
292 GSList *aio_ctxs = NULL, *ctx;
294 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
295 AioContext *aio_context = bdrv_get_aio_context(bs);
297 aio_context_acquire(aio_context);
299 block_job_pause(bs->job);
301 bdrv_parent_drained_begin(bs);
302 bdrv_io_unplugged_begin(bs);
303 bdrv_drain_recurse(bs);
304 aio_context_release(aio_context);
306 if (!g_slist_find(aio_ctxs, aio_context)) {
307 aio_ctxs = g_slist_prepend(aio_ctxs, aio_context);
311 /* Note that completion of an asynchronous I/O operation can trigger any
312 * number of other I/O operations on other devices---for example a
313 * coroutine can submit an I/O request to another device in response to
314 * request completion. Therefore we must keep looping until there was no
315 * more activity rather than simply draining each device independently.
320 for (ctx = aio_ctxs; ctx != NULL; ctx = ctx->next) {
321 AioContext *aio_context = ctx->data;
323 aio_context_acquire(aio_context);
324 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
325 if (aio_context == bdrv_get_aio_context(bs)) {
326 if (bdrv_requests_pending(bs)) {
328 aio_poll(aio_context, busy);
332 busy |= aio_poll(aio_context, false);
333 aio_context_release(aio_context);
337 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
338 AioContext *aio_context = bdrv_get_aio_context(bs);
340 aio_context_acquire(aio_context);
341 bdrv_io_unplugged_end(bs);
342 bdrv_parent_drained_end(bs);
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 (sector-based)
409 void bdrv_round_sectors_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 +
428 * Round a region to cluster boundaries
430 void bdrv_round_to_clusters(BlockDriverState *bs,
431 int64_t offset, unsigned int bytes,
432 int64_t *cluster_offset,
433 unsigned int *cluster_bytes)
437 if (bdrv_get_info(bs, &bdi) < 0 || bdi.cluster_size == 0) {
438 *cluster_offset = offset;
439 *cluster_bytes = bytes;
441 int64_t c = bdi.cluster_size;
442 *cluster_offset = QEMU_ALIGN_DOWN(offset, c);
443 *cluster_bytes = QEMU_ALIGN_UP(offset - *cluster_offset + bytes, c);
447 static int bdrv_get_cluster_size(BlockDriverState *bs)
452 ret = bdrv_get_info(bs, &bdi);
453 if (ret < 0 || bdi.cluster_size == 0) {
454 return bs->request_alignment;
456 return bdi.cluster_size;
460 static bool tracked_request_overlaps(BdrvTrackedRequest *req,
461 int64_t offset, unsigned int bytes)
464 if (offset >= req->overlap_offset + req->overlap_bytes) {
468 if (req->overlap_offset >= offset + bytes) {
474 static bool coroutine_fn wait_serialising_requests(BdrvTrackedRequest *self)
476 BlockDriverState *bs = self->bs;
477 BdrvTrackedRequest *req;
481 if (!bs->serialising_in_flight) {
487 QLIST_FOREACH(req, &bs->tracked_requests, list) {
488 if (req == self || (!req->serialising && !self->serialising)) {
491 if (tracked_request_overlaps(req, self->overlap_offset,
492 self->overlap_bytes))
494 /* Hitting this means there was a reentrant request, for
495 * example, a block driver issuing nested requests. This must
496 * never happen since it means deadlock.
498 assert(qemu_coroutine_self() != req->co);
500 /* If the request is already (indirectly) waiting for us, or
501 * will wait for us as soon as it wakes up, then just go on
502 * (instead of producing a deadlock in the former case). */
503 if (!req->waiting_for) {
504 self->waiting_for = req;
505 qemu_co_queue_wait(&req->wait_queue);
506 self->waiting_for = NULL;
518 static int bdrv_check_byte_request(BlockDriverState *bs, int64_t offset,
521 if (size > BDRV_REQUEST_MAX_SECTORS << BDRV_SECTOR_BITS) {
525 if (!bdrv_is_inserted(bs)) {
536 static int bdrv_check_request(BlockDriverState *bs, int64_t sector_num,
539 if (nb_sectors < 0 || nb_sectors > BDRV_REQUEST_MAX_SECTORS) {
543 return bdrv_check_byte_request(bs, sector_num * BDRV_SECTOR_SIZE,
544 nb_sectors * BDRV_SECTOR_SIZE);
547 typedef struct RwCo {
548 BlockDriverState *bs;
553 BdrvRequestFlags flags;
556 static void coroutine_fn bdrv_rw_co_entry(void *opaque)
560 if (!rwco->is_write) {
561 rwco->ret = bdrv_co_preadv(rwco->bs, rwco->offset,
562 rwco->qiov->size, rwco->qiov,
565 rwco->ret = bdrv_co_pwritev(rwco->bs, rwco->offset,
566 rwco->qiov->size, rwco->qiov,
572 * Process a vectored synchronous request using coroutines
574 static int bdrv_prwv_co(BlockDriverState *bs, int64_t offset,
575 QEMUIOVector *qiov, bool is_write,
576 BdrvRequestFlags flags)
583 .is_write = is_write,
588 if (qemu_in_coroutine()) {
589 /* Fast-path if already in coroutine context */
590 bdrv_rw_co_entry(&rwco);
592 AioContext *aio_context = bdrv_get_aio_context(bs);
594 co = qemu_coroutine_create(bdrv_rw_co_entry);
595 qemu_coroutine_enter(co, &rwco);
596 while (rwco.ret == NOT_DONE) {
597 aio_poll(aio_context, true);
604 * Process a synchronous request using coroutines
606 static int bdrv_rw_co(BlockDriverState *bs, int64_t sector_num, uint8_t *buf,
607 int nb_sectors, bool is_write, BdrvRequestFlags flags)
611 .iov_base = (void *)buf,
612 .iov_len = nb_sectors * BDRV_SECTOR_SIZE,
615 if (nb_sectors < 0 || nb_sectors > BDRV_REQUEST_MAX_SECTORS) {
619 qemu_iovec_init_external(&qiov, &iov, 1);
620 return bdrv_prwv_co(bs, sector_num << BDRV_SECTOR_BITS,
621 &qiov, is_write, flags);
624 /* return < 0 if error. See bdrv_write() for the return codes */
625 int bdrv_read(BlockDriverState *bs, int64_t sector_num,
626 uint8_t *buf, int nb_sectors)
628 return bdrv_rw_co(bs, sector_num, buf, nb_sectors, false, 0);
631 /* Return < 0 if error. Important errors are:
632 -EIO generic I/O error (may happen for all errors)
633 -ENOMEDIUM No media inserted.
634 -EINVAL Invalid sector number or nb_sectors
635 -EACCES Trying to write a read-only device
637 int bdrv_write(BlockDriverState *bs, int64_t sector_num,
638 const uint8_t *buf, int nb_sectors)
640 return bdrv_rw_co(bs, sector_num, (uint8_t *)buf, nb_sectors, true, 0);
643 int bdrv_pwrite_zeroes(BlockDriverState *bs, int64_t offset,
644 int count, BdrvRequestFlags flags)
652 qemu_iovec_init_external(&qiov, &iov, 1);
653 return bdrv_prwv_co(bs, offset, &qiov, true,
654 BDRV_REQ_ZERO_WRITE | flags);
658 * Completely zero out a block device with the help of bdrv_pwrite_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_pwrite_zeroes (e.g. BDRV_REQ_MAY_UNMAP,
664 * Returns < 0 on error, 0 on success. For error codes see bdrv_write().
666 int bdrv_make_zero(BlockDriverState *bs, BdrvRequestFlags flags)
668 int64_t target_sectors, ret, nb_sectors, sector_num = 0;
669 BlockDriverState *file;
672 target_sectors = bdrv_nb_sectors(bs);
673 if (target_sectors < 0) {
674 return target_sectors;
678 nb_sectors = MIN(target_sectors - sector_num, BDRV_REQUEST_MAX_SECTORS);
679 if (nb_sectors <= 0) {
682 ret = bdrv_get_block_status(bs, sector_num, nb_sectors, &n, &file);
684 error_report("error getting block status at sector %" PRId64 ": %s",
685 sector_num, strerror(-ret));
688 if (ret & BDRV_BLOCK_ZERO) {
692 ret = bdrv_pwrite_zeroes(bs, sector_num << BDRV_SECTOR_BITS,
693 n << BDRV_SECTOR_BITS, flags);
695 error_report("error writing zeroes at sector %" PRId64 ": %s",
696 sector_num, strerror(-ret));
703 int bdrv_preadv(BlockDriverState *bs, int64_t offset, QEMUIOVector *qiov)
707 ret = bdrv_prwv_co(bs, offset, qiov, false, 0);
715 int bdrv_pread(BlockDriverState *bs, int64_t offset, void *buf, int bytes)
719 .iov_base = (void *)buf,
727 qemu_iovec_init_external(&qiov, &iov, 1);
728 return bdrv_preadv(bs, offset, &qiov);
731 int bdrv_pwritev(BlockDriverState *bs, int64_t offset, QEMUIOVector *qiov)
735 ret = bdrv_prwv_co(bs, offset, qiov, true, 0);
743 int bdrv_pwrite(BlockDriverState *bs, int64_t offset,
744 const void *buf, int bytes)
748 .iov_base = (void *) buf,
756 qemu_iovec_init_external(&qiov, &iov, 1);
757 return bdrv_pwritev(bs, offset, &qiov);
761 * Writes to the file and ensures that no writes are reordered across this
762 * request (acts as a barrier)
764 * Returns 0 on success, -errno in error cases.
766 int bdrv_pwrite_sync(BlockDriverState *bs, int64_t offset,
767 const void *buf, int count)
771 ret = bdrv_pwrite(bs, offset, buf, count);
776 ret = bdrv_flush(bs);
784 typedef struct CoroutineIOCompletion {
785 Coroutine *coroutine;
787 } CoroutineIOCompletion;
789 static void bdrv_co_io_em_complete(void *opaque, int ret)
791 CoroutineIOCompletion *co = opaque;
794 qemu_coroutine_enter(co->coroutine, NULL);
797 static int coroutine_fn bdrv_driver_preadv(BlockDriverState *bs,
798 uint64_t offset, uint64_t bytes,
799 QEMUIOVector *qiov, int flags)
801 BlockDriver *drv = bs->drv;
803 unsigned int nb_sectors;
805 assert(!(flags & ~BDRV_REQ_MASK));
807 if (drv->bdrv_co_preadv) {
808 return drv->bdrv_co_preadv(bs, offset, bytes, qiov, flags);
811 sector_num = offset >> BDRV_SECTOR_BITS;
812 nb_sectors = bytes >> BDRV_SECTOR_BITS;
814 assert((offset & (BDRV_SECTOR_SIZE - 1)) == 0);
815 assert((bytes & (BDRV_SECTOR_SIZE - 1)) == 0);
816 assert((bytes >> BDRV_SECTOR_BITS) <= BDRV_REQUEST_MAX_SECTORS);
818 if (drv->bdrv_co_readv) {
819 return drv->bdrv_co_readv(bs, sector_num, nb_sectors, qiov);
822 CoroutineIOCompletion co = {
823 .coroutine = qemu_coroutine_self(),
826 acb = bs->drv->bdrv_aio_readv(bs, sector_num, qiov, nb_sectors,
827 bdrv_co_io_em_complete, &co);
831 qemu_coroutine_yield();
837 static int coroutine_fn bdrv_driver_pwritev(BlockDriverState *bs,
838 uint64_t offset, uint64_t bytes,
839 QEMUIOVector *qiov, int flags)
841 BlockDriver *drv = bs->drv;
843 unsigned int nb_sectors;
846 assert(!(flags & ~BDRV_REQ_MASK));
848 if (drv->bdrv_co_pwritev) {
849 ret = drv->bdrv_co_pwritev(bs, offset, bytes, qiov,
850 flags & bs->supported_write_flags);
851 flags &= ~bs->supported_write_flags;
855 sector_num = offset >> BDRV_SECTOR_BITS;
856 nb_sectors = bytes >> BDRV_SECTOR_BITS;
858 assert((offset & (BDRV_SECTOR_SIZE - 1)) == 0);
859 assert((bytes & (BDRV_SECTOR_SIZE - 1)) == 0);
860 assert((bytes >> BDRV_SECTOR_BITS) <= BDRV_REQUEST_MAX_SECTORS);
862 if (drv->bdrv_co_writev_flags) {
863 ret = drv->bdrv_co_writev_flags(bs, sector_num, nb_sectors, qiov,
864 flags & bs->supported_write_flags);
865 flags &= ~bs->supported_write_flags;
866 } else if (drv->bdrv_co_writev) {
867 assert(!bs->supported_write_flags);
868 ret = drv->bdrv_co_writev(bs, sector_num, nb_sectors, qiov);
871 CoroutineIOCompletion co = {
872 .coroutine = qemu_coroutine_self(),
875 acb = bs->drv->bdrv_aio_writev(bs, sector_num, qiov, nb_sectors,
876 bdrv_co_io_em_complete, &co);
880 qemu_coroutine_yield();
886 if (ret == 0 && (flags & BDRV_REQ_FUA)) {
887 ret = bdrv_co_flush(bs);
893 static int coroutine_fn bdrv_co_do_copy_on_readv(BlockDriverState *bs,
894 int64_t offset, unsigned int bytes, QEMUIOVector *qiov)
896 /* Perform I/O through a temporary buffer so that users who scribble over
897 * their read buffer while the operation is in progress do not end up
898 * modifying the image file. This is critical for zero-copy guest I/O
899 * where anything might happen inside guest memory.
903 BlockDriver *drv = bs->drv;
905 QEMUIOVector bounce_qiov;
906 int64_t cluster_offset;
907 unsigned int cluster_bytes;
911 /* Cover entire cluster so no additional backing file I/O is required when
912 * allocating cluster in the image file.
914 bdrv_round_to_clusters(bs, offset, bytes, &cluster_offset, &cluster_bytes);
916 trace_bdrv_co_do_copy_on_readv(bs, offset, bytes,
917 cluster_offset, cluster_bytes);
919 iov.iov_len = cluster_bytes;
920 iov.iov_base = bounce_buffer = qemu_try_blockalign(bs, iov.iov_len);
921 if (bounce_buffer == NULL) {
926 qemu_iovec_init_external(&bounce_qiov, &iov, 1);
928 ret = bdrv_driver_preadv(bs, cluster_offset, cluster_bytes,
934 if (drv->bdrv_co_pwrite_zeroes &&
935 buffer_is_zero(bounce_buffer, iov.iov_len)) {
936 ret = bdrv_co_do_pwrite_zeroes(bs, cluster_offset, cluster_bytes, 0);
938 /* This does not change the data on the disk, it is not necessary
939 * to flush even in cache=writethrough mode.
941 ret = bdrv_driver_pwritev(bs, cluster_offset, cluster_bytes,
946 /* It might be okay to ignore write errors for guest requests. If this
947 * is a deliberate copy-on-read then we don't want to ignore the error.
948 * Simply report it in all cases.
953 skip_bytes = offset - cluster_offset;
954 qemu_iovec_from_buf(qiov, 0, bounce_buffer + skip_bytes, bytes);
957 qemu_vfree(bounce_buffer);
962 * Forwards an already correctly aligned request to the BlockDriver. This
963 * handles copy on read and zeroing after EOF; any other features must be
964 * implemented by the caller.
966 static int coroutine_fn bdrv_aligned_preadv(BlockDriverState *bs,
967 BdrvTrackedRequest *req, int64_t offset, unsigned int bytes,
968 int64_t align, QEMUIOVector *qiov, int flags)
972 assert(is_power_of_2(align));
973 assert((offset & (align - 1)) == 0);
974 assert((bytes & (align - 1)) == 0);
975 assert(!qiov || bytes == qiov->size);
976 assert((bs->open_flags & BDRV_O_NO_IO) == 0);
977 assert(!(flags & ~BDRV_REQ_MASK));
979 /* Handle Copy on Read and associated serialisation */
980 if (flags & BDRV_REQ_COPY_ON_READ) {
981 /* If we touch the same cluster it counts as an overlap. This
982 * guarantees that allocating writes will be serialized and not race
983 * with each other for the same cluster. For example, in copy-on-read
984 * it ensures that the CoR read and write operations are atomic and
985 * guest writes cannot interleave between them. */
986 mark_request_serialising(req, bdrv_get_cluster_size(bs));
989 if (!(flags & BDRV_REQ_NO_SERIALISING)) {
990 wait_serialising_requests(req);
993 if (flags & BDRV_REQ_COPY_ON_READ) {
994 int64_t start_sector = offset >> BDRV_SECTOR_BITS;
995 int64_t end_sector = DIV_ROUND_UP(offset + bytes, BDRV_SECTOR_SIZE);
996 unsigned int nb_sectors = end_sector - start_sector;
999 ret = bdrv_is_allocated(bs, start_sector, nb_sectors, &pnum);
1004 if (!ret || pnum != nb_sectors) {
1005 ret = bdrv_co_do_copy_on_readv(bs, offset, bytes, qiov);
1010 /* Forward the request to the BlockDriver */
1011 if (!bs->zero_beyond_eof) {
1012 ret = bdrv_driver_preadv(bs, offset, bytes, qiov, 0);
1014 /* Read zeros after EOF */
1015 int64_t total_bytes, max_bytes;
1017 total_bytes = bdrv_getlength(bs);
1018 if (total_bytes < 0) {
1023 max_bytes = ROUND_UP(MAX(0, total_bytes - offset), align);
1024 if (bytes < max_bytes) {
1025 ret = bdrv_driver_preadv(bs, offset, bytes, qiov, 0);
1026 } else if (max_bytes > 0) {
1027 QEMUIOVector local_qiov;
1029 qemu_iovec_init(&local_qiov, qiov->niov);
1030 qemu_iovec_concat(&local_qiov, qiov, 0, max_bytes);
1032 ret = bdrv_driver_preadv(bs, offset, max_bytes, &local_qiov, 0);
1034 qemu_iovec_destroy(&local_qiov);
1039 /* Reading beyond end of file is supposed to produce zeroes */
1040 if (ret == 0 && total_bytes < offset + bytes) {
1041 uint64_t zero_offset = MAX(0, total_bytes - offset);
1042 uint64_t zero_bytes = offset + bytes - zero_offset;
1043 qemu_iovec_memset(qiov, zero_offset, 0, zero_bytes);
1052 * Handle a read request in coroutine context
1054 int coroutine_fn bdrv_co_preadv(BlockDriverState *bs,
1055 int64_t offset, unsigned int bytes, QEMUIOVector *qiov,
1056 BdrvRequestFlags flags)
1058 BlockDriver *drv = bs->drv;
1059 BdrvTrackedRequest req;
1061 uint64_t align = bs->request_alignment;
1062 uint8_t *head_buf = NULL;
1063 uint8_t *tail_buf = NULL;
1064 QEMUIOVector local_qiov;
1065 bool use_local_qiov = false;
1072 ret = bdrv_check_byte_request(bs, offset, bytes);
1077 /* Don't do copy-on-read if we read data before write operation */
1078 if (bs->copy_on_read && !(flags & BDRV_REQ_NO_SERIALISING)) {
1079 flags |= BDRV_REQ_COPY_ON_READ;
1082 /* Align read if necessary by padding qiov */
1083 if (offset & (align - 1)) {
1084 head_buf = qemu_blockalign(bs, align);
1085 qemu_iovec_init(&local_qiov, qiov->niov + 2);
1086 qemu_iovec_add(&local_qiov, head_buf, offset & (align - 1));
1087 qemu_iovec_concat(&local_qiov, qiov, 0, qiov->size);
1088 use_local_qiov = true;
1090 bytes += offset & (align - 1);
1091 offset = offset & ~(align - 1);
1094 if ((offset + bytes) & (align - 1)) {
1095 if (!use_local_qiov) {
1096 qemu_iovec_init(&local_qiov, qiov->niov + 1);
1097 qemu_iovec_concat(&local_qiov, qiov, 0, qiov->size);
1098 use_local_qiov = true;
1100 tail_buf = qemu_blockalign(bs, align);
1101 qemu_iovec_add(&local_qiov, tail_buf,
1102 align - ((offset + bytes) & (align - 1)));
1104 bytes = ROUND_UP(bytes, align);
1107 tracked_request_begin(&req, bs, offset, bytes, BDRV_TRACKED_READ);
1108 ret = bdrv_aligned_preadv(bs, &req, offset, bytes, align,
1109 use_local_qiov ? &local_qiov : qiov,
1111 tracked_request_end(&req);
1113 if (use_local_qiov) {
1114 qemu_iovec_destroy(&local_qiov);
1115 qemu_vfree(head_buf);
1116 qemu_vfree(tail_buf);
1122 static int coroutine_fn bdrv_co_do_readv(BlockDriverState *bs,
1123 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov,
1124 BdrvRequestFlags flags)
1126 if (nb_sectors < 0 || nb_sectors > BDRV_REQUEST_MAX_SECTORS) {
1130 return bdrv_co_preadv(bs, sector_num << BDRV_SECTOR_BITS,
1131 nb_sectors << BDRV_SECTOR_BITS, qiov, flags);
1134 int coroutine_fn bdrv_co_readv(BlockDriverState *bs, int64_t sector_num,
1135 int nb_sectors, QEMUIOVector *qiov)
1137 trace_bdrv_co_readv(bs, sector_num, nb_sectors);
1139 return bdrv_co_do_readv(bs, sector_num, nb_sectors, qiov, 0);
1142 #define MAX_WRITE_ZEROES_BOUNCE_BUFFER 32768
1144 static int coroutine_fn bdrv_co_do_pwrite_zeroes(BlockDriverState *bs,
1145 int64_t offset, int count, BdrvRequestFlags flags)
1147 BlockDriver *drv = bs->drv;
1149 struct iovec iov = {0};
1151 bool need_flush = false;
1155 int max_write_zeroes = MIN_NON_ZERO(bs->bl.max_pwrite_zeroes, INT_MAX);
1156 int alignment = MAX(bs->bl.pwrite_zeroes_alignment ?: 1,
1157 bs->request_alignment);
1159 assert(is_power_of_2(alignment));
1160 head = offset & (alignment - 1);
1161 tail = (offset + count) & (alignment - 1);
1162 max_write_zeroes &= ~(alignment - 1);
1164 while (count > 0 && !ret) {
1167 /* Align request. Block drivers can expect the "bulk" of the request
1168 * to be aligned, and that unaligned requests do not cross cluster
1172 /* Make a small request up to the first aligned sector. */
1173 num = MIN(count, alignment - head);
1175 } else if (tail && num > alignment) {
1176 /* Shorten the request to the last aligned sector. */
1180 /* limit request size */
1181 if (num > max_write_zeroes) {
1182 num = max_write_zeroes;
1186 /* First try the efficient write zeroes operation */
1187 if (drv->bdrv_co_pwrite_zeroes) {
1188 ret = drv->bdrv_co_pwrite_zeroes(bs, offset, num,
1189 flags & bs->supported_zero_flags);
1190 if (ret != -ENOTSUP && (flags & BDRV_REQ_FUA) &&
1191 !(bs->supported_zero_flags & BDRV_REQ_FUA)) {
1195 assert(!bs->supported_zero_flags);
1198 if (ret == -ENOTSUP) {
1199 /* Fall back to bounce buffer if write zeroes is unsupported */
1200 int max_xfer_len = MIN_NON_ZERO(bs->bl.max_transfer_length,
1201 MAX_WRITE_ZEROES_BOUNCE_BUFFER);
1202 BdrvRequestFlags write_flags = flags & ~BDRV_REQ_ZERO_WRITE;
1204 if ((flags & BDRV_REQ_FUA) &&
1205 !(bs->supported_write_flags & BDRV_REQ_FUA)) {
1206 /* No need for bdrv_driver_pwrite() to do a fallback
1207 * flush on each chunk; use just one at the end */
1208 write_flags &= ~BDRV_REQ_FUA;
1211 num = MIN(num, max_xfer_len << BDRV_SECTOR_BITS);
1213 if (iov.iov_base == NULL) {
1214 iov.iov_base = qemu_try_blockalign(bs, num);
1215 if (iov.iov_base == NULL) {
1219 memset(iov.iov_base, 0, num);
1221 qemu_iovec_init_external(&qiov, &iov, 1);
1223 ret = bdrv_driver_pwritev(bs, offset, num, &qiov, write_flags);
1225 /* Keep bounce buffer around if it is big enough for all
1226 * all future requests.
1228 if (num < max_xfer_len << BDRV_SECTOR_BITS) {
1229 qemu_vfree(iov.iov_base);
1230 iov.iov_base = NULL;
1239 if (ret == 0 && need_flush) {
1240 ret = bdrv_co_flush(bs);
1242 qemu_vfree(iov.iov_base);
1247 * Forwards an already correctly aligned write request to the BlockDriver.
1249 static int coroutine_fn bdrv_aligned_pwritev(BlockDriverState *bs,
1250 BdrvTrackedRequest *req, int64_t offset, unsigned int bytes,
1251 QEMUIOVector *qiov, int flags)
1253 BlockDriver *drv = bs->drv;
1257 int64_t start_sector = offset >> BDRV_SECTOR_BITS;
1258 int64_t end_sector = DIV_ROUND_UP(offset + bytes, BDRV_SECTOR_SIZE);
1260 assert(!qiov || bytes == qiov->size);
1261 assert((bs->open_flags & BDRV_O_NO_IO) == 0);
1262 assert(!(flags & ~BDRV_REQ_MASK));
1264 waited = wait_serialising_requests(req);
1265 assert(!waited || !req->serialising);
1266 assert(req->overlap_offset <= offset);
1267 assert(offset + bytes <= req->overlap_offset + req->overlap_bytes);
1269 ret = notifier_with_return_list_notify(&bs->before_write_notifiers, req);
1271 if (!ret && bs->detect_zeroes != BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF &&
1272 !(flags & BDRV_REQ_ZERO_WRITE) && drv->bdrv_co_pwrite_zeroes &&
1273 qemu_iovec_is_zero(qiov)) {
1274 flags |= BDRV_REQ_ZERO_WRITE;
1275 if (bs->detect_zeroes == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP) {
1276 flags |= BDRV_REQ_MAY_UNMAP;
1281 /* Do nothing, write notifier decided to fail this request */
1282 } else if (flags & BDRV_REQ_ZERO_WRITE) {
1283 bdrv_debug_event(bs, BLKDBG_PWRITEV_ZERO);
1284 ret = bdrv_co_do_pwrite_zeroes(bs, offset, bytes, flags);
1286 bdrv_debug_event(bs, BLKDBG_PWRITEV);
1287 ret = bdrv_driver_pwritev(bs, offset, bytes, qiov, flags);
1289 bdrv_debug_event(bs, BLKDBG_PWRITEV_DONE);
1291 bdrv_set_dirty(bs, start_sector, end_sector - start_sector);
1293 if (bs->wr_highest_offset < offset + bytes) {
1294 bs->wr_highest_offset = offset + bytes;
1298 bs->total_sectors = MAX(bs->total_sectors, end_sector);
1304 static int coroutine_fn bdrv_co_do_zero_pwritev(BlockDriverState *bs,
1307 BdrvRequestFlags flags,
1308 BdrvTrackedRequest *req)
1310 uint8_t *buf = NULL;
1311 QEMUIOVector local_qiov;
1313 uint64_t align = bs->request_alignment;
1314 unsigned int head_padding_bytes, tail_padding_bytes;
1317 head_padding_bytes = offset & (align - 1);
1318 tail_padding_bytes = align - ((offset + bytes) & (align - 1));
1321 assert(flags & BDRV_REQ_ZERO_WRITE);
1322 if (head_padding_bytes || tail_padding_bytes) {
1323 buf = qemu_blockalign(bs, align);
1324 iov = (struct iovec) {
1328 qemu_iovec_init_external(&local_qiov, &iov, 1);
1330 if (head_padding_bytes) {
1331 uint64_t zero_bytes = MIN(bytes, align - head_padding_bytes);
1333 /* RMW the unaligned part before head. */
1334 mark_request_serialising(req, align);
1335 wait_serialising_requests(req);
1336 bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_HEAD);
1337 ret = bdrv_aligned_preadv(bs, req, offset & ~(align - 1), align,
1338 align, &local_qiov, 0);
1342 bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_AFTER_HEAD);
1344 memset(buf + head_padding_bytes, 0, zero_bytes);
1345 ret = bdrv_aligned_pwritev(bs, req, offset & ~(align - 1), align,
1347 flags & ~BDRV_REQ_ZERO_WRITE);
1351 offset += zero_bytes;
1352 bytes -= zero_bytes;
1355 assert(!bytes || (offset & (align - 1)) == 0);
1356 if (bytes >= align) {
1357 /* Write the aligned part in the middle. */
1358 uint64_t aligned_bytes = bytes & ~(align - 1);
1359 ret = bdrv_aligned_pwritev(bs, req, offset, aligned_bytes,
1364 bytes -= aligned_bytes;
1365 offset += aligned_bytes;
1368 assert(!bytes || (offset & (align - 1)) == 0);
1370 assert(align == tail_padding_bytes + bytes);
1371 /* RMW the unaligned part after tail. */
1372 mark_request_serialising(req, align);
1373 wait_serialising_requests(req);
1374 bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_TAIL);
1375 ret = bdrv_aligned_preadv(bs, req, offset, align,
1376 align, &local_qiov, 0);
1380 bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_AFTER_TAIL);
1382 memset(buf, 0, bytes);
1383 ret = bdrv_aligned_pwritev(bs, req, offset, align,
1384 &local_qiov, flags & ~BDRV_REQ_ZERO_WRITE);
1393 * Handle a write request in coroutine context
1395 int coroutine_fn bdrv_co_pwritev(BlockDriverState *bs,
1396 int64_t offset, unsigned int bytes, QEMUIOVector *qiov,
1397 BdrvRequestFlags flags)
1399 BdrvTrackedRequest req;
1400 uint64_t align = bs->request_alignment;
1401 uint8_t *head_buf = NULL;
1402 uint8_t *tail_buf = NULL;
1403 QEMUIOVector local_qiov;
1404 bool use_local_qiov = false;
1410 if (bs->read_only) {
1413 assert(!(bs->open_flags & BDRV_O_INACTIVE));
1415 ret = bdrv_check_byte_request(bs, offset, bytes);
1421 * Align write if necessary by performing a read-modify-write cycle.
1422 * Pad qiov with the read parts and be sure to have a tracked request not
1423 * only for bdrv_aligned_pwritev, but also for the reads of the RMW cycle.
1425 tracked_request_begin(&req, bs, offset, bytes, BDRV_TRACKED_WRITE);
1428 ret = bdrv_co_do_zero_pwritev(bs, offset, bytes, flags, &req);
1432 if (offset & (align - 1)) {
1433 QEMUIOVector head_qiov;
1434 struct iovec head_iov;
1436 mark_request_serialising(&req, align);
1437 wait_serialising_requests(&req);
1439 head_buf = qemu_blockalign(bs, align);
1440 head_iov = (struct iovec) {
1441 .iov_base = head_buf,
1444 qemu_iovec_init_external(&head_qiov, &head_iov, 1);
1446 bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_HEAD);
1447 ret = bdrv_aligned_preadv(bs, &req, offset & ~(align - 1), align,
1448 align, &head_qiov, 0);
1452 bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_AFTER_HEAD);
1454 qemu_iovec_init(&local_qiov, qiov->niov + 2);
1455 qemu_iovec_add(&local_qiov, head_buf, offset & (align - 1));
1456 qemu_iovec_concat(&local_qiov, qiov, 0, qiov->size);
1457 use_local_qiov = true;
1459 bytes += offset & (align - 1);
1460 offset = offset & ~(align - 1);
1462 /* We have read the tail already if the request is smaller
1463 * than one aligned block.
1465 if (bytes < align) {
1466 qemu_iovec_add(&local_qiov, head_buf + bytes, align - bytes);
1471 if ((offset + bytes) & (align - 1)) {
1472 QEMUIOVector tail_qiov;
1473 struct iovec tail_iov;
1477 mark_request_serialising(&req, align);
1478 waited = wait_serialising_requests(&req);
1479 assert(!waited || !use_local_qiov);
1481 tail_buf = qemu_blockalign(bs, align);
1482 tail_iov = (struct iovec) {
1483 .iov_base = tail_buf,
1486 qemu_iovec_init_external(&tail_qiov, &tail_iov, 1);
1488 bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_TAIL);
1489 ret = bdrv_aligned_preadv(bs, &req, (offset + bytes) & ~(align - 1), align,
1490 align, &tail_qiov, 0);
1494 bdrv_debug_event(bs, BLKDBG_PWRITEV_RMW_AFTER_TAIL);
1496 if (!use_local_qiov) {
1497 qemu_iovec_init(&local_qiov, qiov->niov + 1);
1498 qemu_iovec_concat(&local_qiov, qiov, 0, qiov->size);
1499 use_local_qiov = true;
1502 tail_bytes = (offset + bytes) & (align - 1);
1503 qemu_iovec_add(&local_qiov, tail_buf + tail_bytes, align - tail_bytes);
1505 bytes = ROUND_UP(bytes, align);
1508 ret = bdrv_aligned_pwritev(bs, &req, offset, bytes,
1509 use_local_qiov ? &local_qiov : qiov,
1514 if (use_local_qiov) {
1515 qemu_iovec_destroy(&local_qiov);
1517 qemu_vfree(head_buf);
1518 qemu_vfree(tail_buf);
1520 tracked_request_end(&req);
1524 static int coroutine_fn bdrv_co_do_writev(BlockDriverState *bs,
1525 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov,
1526 BdrvRequestFlags flags)
1528 if (nb_sectors < 0 || nb_sectors > BDRV_REQUEST_MAX_SECTORS) {
1532 return bdrv_co_pwritev(bs, sector_num << BDRV_SECTOR_BITS,
1533 nb_sectors << BDRV_SECTOR_BITS, qiov, flags);
1536 int coroutine_fn bdrv_co_writev(BlockDriverState *bs, int64_t sector_num,
1537 int nb_sectors, QEMUIOVector *qiov)
1539 trace_bdrv_co_writev(bs, sector_num, nb_sectors);
1541 return bdrv_co_do_writev(bs, sector_num, nb_sectors, qiov, 0);
1544 int coroutine_fn bdrv_co_pwrite_zeroes(BlockDriverState *bs,
1545 int64_t offset, int count,
1546 BdrvRequestFlags flags)
1548 trace_bdrv_co_pwrite_zeroes(bs, offset, count, flags);
1550 if (!(bs->open_flags & BDRV_O_UNMAP)) {
1551 flags &= ~BDRV_REQ_MAY_UNMAP;
1554 return bdrv_co_pwritev(bs, offset, count, NULL,
1555 BDRV_REQ_ZERO_WRITE | flags);
1558 typedef struct BdrvCoGetBlockStatusData {
1559 BlockDriverState *bs;
1560 BlockDriverState *base;
1561 BlockDriverState **file;
1567 } BdrvCoGetBlockStatusData;
1570 * Returns the allocation status of the specified sectors.
1571 * Drivers not implementing the functionality are assumed to not support
1572 * backing files, hence all their sectors are reported as allocated.
1574 * If 'sector_num' is beyond the end of the disk image the return value is 0
1575 * and 'pnum' is set to 0.
1577 * 'pnum' is set to the number of sectors (including and immediately following
1578 * the specified sector) that are known to be in the same
1579 * allocated/unallocated state.
1581 * 'nb_sectors' is the max value 'pnum' should be set to. If nb_sectors goes
1582 * beyond the end of the disk image it will be clamped.
1584 * If returned value is positive and BDRV_BLOCK_OFFSET_VALID bit is set, 'file'
1585 * points to the BDS which the sector range is allocated in.
1587 static int64_t coroutine_fn bdrv_co_get_block_status(BlockDriverState *bs,
1589 int nb_sectors, int *pnum,
1590 BlockDriverState **file)
1592 int64_t total_sectors;
1596 total_sectors = bdrv_nb_sectors(bs);
1597 if (total_sectors < 0) {
1598 return total_sectors;
1601 if (sector_num >= total_sectors) {
1606 n = total_sectors - sector_num;
1607 if (n < nb_sectors) {
1611 if (!bs->drv->bdrv_co_get_block_status) {
1613 ret = BDRV_BLOCK_DATA | BDRV_BLOCK_ALLOCATED;
1614 if (bs->drv->protocol_name) {
1615 ret |= BDRV_BLOCK_OFFSET_VALID | (sector_num * BDRV_SECTOR_SIZE);
1621 ret = bs->drv->bdrv_co_get_block_status(bs, sector_num, nb_sectors, pnum,
1628 if (ret & BDRV_BLOCK_RAW) {
1629 assert(ret & BDRV_BLOCK_OFFSET_VALID);
1630 return bdrv_get_block_status(bs->file->bs, ret >> BDRV_SECTOR_BITS,
1634 if (ret & (BDRV_BLOCK_DATA | BDRV_BLOCK_ZERO)) {
1635 ret |= BDRV_BLOCK_ALLOCATED;
1637 if (bdrv_unallocated_blocks_are_zero(bs)) {
1638 ret |= BDRV_BLOCK_ZERO;
1639 } else if (bs->backing) {
1640 BlockDriverState *bs2 = bs->backing->bs;
1641 int64_t nb_sectors2 = bdrv_nb_sectors(bs2);
1642 if (nb_sectors2 >= 0 && sector_num >= nb_sectors2) {
1643 ret |= BDRV_BLOCK_ZERO;
1648 if (*file && *file != bs &&
1649 (ret & BDRV_BLOCK_DATA) && !(ret & BDRV_BLOCK_ZERO) &&
1650 (ret & BDRV_BLOCK_OFFSET_VALID)) {
1651 BlockDriverState *file2;
1654 ret2 = bdrv_co_get_block_status(*file, ret >> BDRV_SECTOR_BITS,
1655 *pnum, &file_pnum, &file2);
1657 /* Ignore errors. This is just providing extra information, it
1658 * is useful but not necessary.
1661 /* !file_pnum indicates an offset at or beyond the EOF; it is
1662 * perfectly valid for the format block driver to point to such
1663 * offsets, so catch it and mark everything as zero */
1664 ret |= BDRV_BLOCK_ZERO;
1666 /* Limit request to the range reported by the protocol driver */
1668 ret |= (ret2 & BDRV_BLOCK_ZERO);
1676 static int64_t coroutine_fn bdrv_co_get_block_status_above(BlockDriverState *bs,
1677 BlockDriverState *base,
1681 BlockDriverState **file)
1683 BlockDriverState *p;
1687 for (p = bs; p != base; p = backing_bs(p)) {
1688 ret = bdrv_co_get_block_status(p, sector_num, nb_sectors, pnum, file);
1689 if (ret < 0 || ret & BDRV_BLOCK_ALLOCATED) {
1692 /* [sector_num, pnum] unallocated on this layer, which could be only
1693 * the first part of [sector_num, nb_sectors]. */
1694 nb_sectors = MIN(nb_sectors, *pnum);
1699 /* Coroutine wrapper for bdrv_get_block_status_above() */
1700 static void coroutine_fn bdrv_get_block_status_above_co_entry(void *opaque)
1702 BdrvCoGetBlockStatusData *data = opaque;
1704 data->ret = bdrv_co_get_block_status_above(data->bs, data->base,
1713 * Synchronous wrapper around bdrv_co_get_block_status_above().
1715 * See bdrv_co_get_block_status_above() for details.
1717 int64_t bdrv_get_block_status_above(BlockDriverState *bs,
1718 BlockDriverState *base,
1720 int nb_sectors, int *pnum,
1721 BlockDriverState **file)
1724 BdrvCoGetBlockStatusData data = {
1728 .sector_num = sector_num,
1729 .nb_sectors = nb_sectors,
1734 if (qemu_in_coroutine()) {
1735 /* Fast-path if already in coroutine context */
1736 bdrv_get_block_status_above_co_entry(&data);
1738 AioContext *aio_context = bdrv_get_aio_context(bs);
1740 co = qemu_coroutine_create(bdrv_get_block_status_above_co_entry);
1741 qemu_coroutine_enter(co, &data);
1742 while (!data.done) {
1743 aio_poll(aio_context, true);
1749 int64_t bdrv_get_block_status(BlockDriverState *bs,
1751 int nb_sectors, int *pnum,
1752 BlockDriverState **file)
1754 return bdrv_get_block_status_above(bs, backing_bs(bs),
1755 sector_num, nb_sectors, pnum, file);
1758 int coroutine_fn bdrv_is_allocated(BlockDriverState *bs, int64_t sector_num,
1759 int nb_sectors, int *pnum)
1761 BlockDriverState *file;
1762 int64_t ret = bdrv_get_block_status(bs, sector_num, nb_sectors, pnum,
1767 return !!(ret & BDRV_BLOCK_ALLOCATED);
1771 * Given an image chain: ... -> [BASE] -> [INTER1] -> [INTER2] -> [TOP]
1773 * Return true if the given sector is allocated in any image between
1774 * BASE and TOP (inclusive). BASE can be NULL to check if the given
1775 * sector is allocated in any image of the chain. Return false otherwise.
1777 * 'pnum' is set to the number of sectors (including and immediately following
1778 * the specified sector) that are known to be in the same
1779 * allocated/unallocated state.
1782 int bdrv_is_allocated_above(BlockDriverState *top,
1783 BlockDriverState *base,
1785 int nb_sectors, int *pnum)
1787 BlockDriverState *intermediate;
1788 int ret, n = nb_sectors;
1791 while (intermediate && intermediate != base) {
1793 ret = bdrv_is_allocated(intermediate, sector_num, nb_sectors,
1803 * [sector_num, nb_sectors] is unallocated on top but intermediate
1806 * [sector_num+x, nr_sectors] allocated.
1808 if (n > pnum_inter &&
1809 (intermediate == top ||
1810 sector_num + pnum_inter < intermediate->total_sectors)) {
1814 intermediate = backing_bs(intermediate);
1821 int bdrv_write_compressed(BlockDriverState *bs, int64_t sector_num,
1822 const uint8_t *buf, int nb_sectors)
1824 BlockDriver *drv = bs->drv;
1830 if (!drv->bdrv_write_compressed) {
1833 ret = bdrv_check_request(bs, sector_num, nb_sectors);
1838 assert(QLIST_EMPTY(&bs->dirty_bitmaps));
1840 return drv->bdrv_write_compressed(bs, sector_num, buf, nb_sectors);
1843 int bdrv_save_vmstate(BlockDriverState *bs, const uint8_t *buf,
1844 int64_t pos, int size)
1847 struct iovec iov = {
1848 .iov_base = (void *) buf,
1853 qemu_iovec_init_external(&qiov, &iov, 1);
1855 ret = bdrv_writev_vmstate(bs, &qiov, pos);
1863 int bdrv_writev_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, int64_t pos)
1865 BlockDriver *drv = bs->drv;
1869 } else if (drv->bdrv_save_vmstate) {
1870 return drv->bdrv_save_vmstate(bs, qiov, pos);
1871 } else if (bs->file) {
1872 return bdrv_writev_vmstate(bs->file->bs, qiov, pos);
1878 int bdrv_load_vmstate(BlockDriverState *bs, uint8_t *buf,
1879 int64_t pos, int size)
1882 struct iovec iov = {
1888 qemu_iovec_init_external(&qiov, &iov, 1);
1889 ret = bdrv_readv_vmstate(bs, &qiov, pos);
1897 int bdrv_readv_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, int64_t pos)
1899 BlockDriver *drv = bs->drv;
1903 } else if (drv->bdrv_load_vmstate) {
1904 return drv->bdrv_load_vmstate(bs, qiov, pos);
1905 } else if (bs->file) {
1906 return bdrv_readv_vmstate(bs->file->bs, qiov, pos);
1912 /**************************************************************/
1915 BlockAIOCB *bdrv_aio_readv(BlockDriverState *bs, int64_t sector_num,
1916 QEMUIOVector *qiov, int nb_sectors,
1917 BlockCompletionFunc *cb, void *opaque)
1919 trace_bdrv_aio_readv(bs, sector_num, nb_sectors, opaque);
1921 return bdrv_co_aio_rw_vector(bs, sector_num, qiov, nb_sectors, 0,
1925 BlockAIOCB *bdrv_aio_writev(BlockDriverState *bs, int64_t sector_num,
1926 QEMUIOVector *qiov, int nb_sectors,
1927 BlockCompletionFunc *cb, void *opaque)
1929 trace_bdrv_aio_writev(bs, sector_num, nb_sectors, opaque);
1931 return bdrv_co_aio_rw_vector(bs, sector_num, qiov, nb_sectors, 0,
1935 void bdrv_aio_cancel(BlockAIOCB *acb)
1938 bdrv_aio_cancel_async(acb);
1939 while (acb->refcnt > 1) {
1940 if (acb->aiocb_info->get_aio_context) {
1941 aio_poll(acb->aiocb_info->get_aio_context(acb), true);
1942 } else if (acb->bs) {
1943 aio_poll(bdrv_get_aio_context(acb->bs), true);
1948 qemu_aio_unref(acb);
1951 /* Async version of aio cancel. The caller is not blocked if the acb implements
1952 * cancel_async, otherwise we do nothing and let the request normally complete.
1953 * In either case the completion callback must be called. */
1954 void bdrv_aio_cancel_async(BlockAIOCB *acb)
1956 if (acb->aiocb_info->cancel_async) {
1957 acb->aiocb_info->cancel_async(acb);
1961 /**************************************************************/
1962 /* async block device emulation */
1964 typedef struct BlockRequest {
1966 /* Used during read, write, trim */
1973 /* Used during ioctl */
1979 BlockCompletionFunc *cb;
1985 typedef struct BlockAIOCBCoroutine {
1992 } BlockAIOCBCoroutine;
1994 static const AIOCBInfo bdrv_em_co_aiocb_info = {
1995 .aiocb_size = sizeof(BlockAIOCBCoroutine),
1998 static void bdrv_co_complete(BlockAIOCBCoroutine *acb)
2000 if (!acb->need_bh) {
2001 acb->common.cb(acb->common.opaque, acb->req.error);
2002 qemu_aio_unref(acb);
2006 static void bdrv_co_em_bh(void *opaque)
2008 BlockAIOCBCoroutine *acb = opaque;
2010 assert(!acb->need_bh);
2011 qemu_bh_delete(acb->bh);
2012 bdrv_co_complete(acb);
2015 static void bdrv_co_maybe_schedule_bh(BlockAIOCBCoroutine *acb)
2017 acb->need_bh = false;
2018 if (acb->req.error != -EINPROGRESS) {
2019 BlockDriverState *bs = acb->common.bs;
2021 acb->bh = aio_bh_new(bdrv_get_aio_context(bs), bdrv_co_em_bh, acb);
2022 qemu_bh_schedule(acb->bh);
2026 /* Invoke bdrv_co_do_readv/bdrv_co_do_writev */
2027 static void coroutine_fn bdrv_co_do_rw(void *opaque)
2029 BlockAIOCBCoroutine *acb = opaque;
2030 BlockDriverState *bs = acb->common.bs;
2032 if (!acb->is_write) {
2033 acb->req.error = bdrv_co_do_readv(bs, acb->req.sector,
2034 acb->req.nb_sectors, acb->req.qiov, acb->req.flags);
2036 acb->req.error = bdrv_co_do_writev(bs, acb->req.sector,
2037 acb->req.nb_sectors, acb->req.qiov, acb->req.flags);
2040 bdrv_co_complete(acb);
2043 static BlockAIOCB *bdrv_co_aio_rw_vector(BlockDriverState *bs,
2047 BdrvRequestFlags flags,
2048 BlockCompletionFunc *cb,
2053 BlockAIOCBCoroutine *acb;
2055 acb = qemu_aio_get(&bdrv_em_co_aiocb_info, bs, cb, opaque);
2056 acb->need_bh = true;
2057 acb->req.error = -EINPROGRESS;
2058 acb->req.sector = sector_num;
2059 acb->req.nb_sectors = nb_sectors;
2060 acb->req.qiov = qiov;
2061 acb->req.flags = flags;
2062 acb->is_write = is_write;
2064 co = qemu_coroutine_create(bdrv_co_do_rw);
2065 qemu_coroutine_enter(co, acb);
2067 bdrv_co_maybe_schedule_bh(acb);
2068 return &acb->common;
2071 static void coroutine_fn bdrv_aio_flush_co_entry(void *opaque)
2073 BlockAIOCBCoroutine *acb = opaque;
2074 BlockDriverState *bs = acb->common.bs;
2076 acb->req.error = bdrv_co_flush(bs);
2077 bdrv_co_complete(acb);
2080 BlockAIOCB *bdrv_aio_flush(BlockDriverState *bs,
2081 BlockCompletionFunc *cb, void *opaque)
2083 trace_bdrv_aio_flush(bs, opaque);
2086 BlockAIOCBCoroutine *acb;
2088 acb = qemu_aio_get(&bdrv_em_co_aiocb_info, bs, cb, opaque);
2089 acb->need_bh = true;
2090 acb->req.error = -EINPROGRESS;
2092 co = qemu_coroutine_create(bdrv_aio_flush_co_entry);
2093 qemu_coroutine_enter(co, acb);
2095 bdrv_co_maybe_schedule_bh(acb);
2096 return &acb->common;
2099 static void coroutine_fn bdrv_aio_discard_co_entry(void *opaque)
2101 BlockAIOCBCoroutine *acb = opaque;
2102 BlockDriverState *bs = acb->common.bs;
2104 acb->req.error = bdrv_co_discard(bs, acb->req.sector, acb->req.nb_sectors);
2105 bdrv_co_complete(acb);
2108 BlockAIOCB *bdrv_aio_discard(BlockDriverState *bs,
2109 int64_t sector_num, int nb_sectors,
2110 BlockCompletionFunc *cb, void *opaque)
2113 BlockAIOCBCoroutine *acb;
2115 trace_bdrv_aio_discard(bs, sector_num, nb_sectors, opaque);
2117 acb = qemu_aio_get(&bdrv_em_co_aiocb_info, bs, cb, opaque);
2118 acb->need_bh = true;
2119 acb->req.error = -EINPROGRESS;
2120 acb->req.sector = sector_num;
2121 acb->req.nb_sectors = nb_sectors;
2122 co = qemu_coroutine_create(bdrv_aio_discard_co_entry);
2123 qemu_coroutine_enter(co, acb);
2125 bdrv_co_maybe_schedule_bh(acb);
2126 return &acb->common;
2129 void *qemu_aio_get(const AIOCBInfo *aiocb_info, BlockDriverState *bs,
2130 BlockCompletionFunc *cb, void *opaque)
2134 acb = g_malloc(aiocb_info->aiocb_size);
2135 acb->aiocb_info = aiocb_info;
2138 acb->opaque = opaque;
2143 void qemu_aio_ref(void *p)
2145 BlockAIOCB *acb = p;
2149 void qemu_aio_unref(void *p)
2151 BlockAIOCB *acb = p;
2152 assert(acb->refcnt > 0);
2153 if (--acb->refcnt == 0) {
2158 /**************************************************************/
2159 /* Coroutine block device emulation */
2161 static void coroutine_fn bdrv_flush_co_entry(void *opaque)
2163 RwCo *rwco = opaque;
2165 rwco->ret = bdrv_co_flush(rwco->bs);
2168 int coroutine_fn bdrv_co_flush(BlockDriverState *bs)
2171 BdrvTrackedRequest req;
2173 if (!bs || !bdrv_is_inserted(bs) || bdrv_is_read_only(bs) ||
2178 tracked_request_begin(&req, bs, 0, 0, BDRV_TRACKED_FLUSH);
2180 /* Write back all layers by calling one driver function */
2181 if (bs->drv->bdrv_co_flush) {
2182 ret = bs->drv->bdrv_co_flush(bs);
2186 /* Write back cached data to the OS even with cache=unsafe */
2187 BLKDBG_EVENT(bs->file, BLKDBG_FLUSH_TO_OS);
2188 if (bs->drv->bdrv_co_flush_to_os) {
2189 ret = bs->drv->bdrv_co_flush_to_os(bs);
2195 /* But don't actually force it to the disk with cache=unsafe */
2196 if (bs->open_flags & BDRV_O_NO_FLUSH) {
2200 BLKDBG_EVENT(bs->file, BLKDBG_FLUSH_TO_DISK);
2201 if (bs->drv->bdrv_co_flush_to_disk) {
2202 ret = bs->drv->bdrv_co_flush_to_disk(bs);
2203 } else if (bs->drv->bdrv_aio_flush) {
2205 CoroutineIOCompletion co = {
2206 .coroutine = qemu_coroutine_self(),
2209 acb = bs->drv->bdrv_aio_flush(bs, bdrv_co_io_em_complete, &co);
2213 qemu_coroutine_yield();
2218 * Some block drivers always operate in either writethrough or unsafe
2219 * mode and don't support bdrv_flush therefore. Usually qemu doesn't
2220 * know how the server works (because the behaviour is hardcoded or
2221 * depends on server-side configuration), so we can't ensure that
2222 * everything is safe on disk. Returning an error doesn't work because
2223 * that would break guests even if the server operates in writethrough
2226 * Let's hope the user knows what he's doing.
2234 /* Now flush the underlying protocol. It will also have BDRV_O_NO_FLUSH
2235 * in the case of cache=unsafe, so there are no useless flushes.
2238 ret = bs->file ? bdrv_co_flush(bs->file->bs) : 0;
2240 tracked_request_end(&req);
2244 int bdrv_flush(BlockDriverState *bs)
2252 if (qemu_in_coroutine()) {
2253 /* Fast-path if already in coroutine context */
2254 bdrv_flush_co_entry(&rwco);
2256 AioContext *aio_context = bdrv_get_aio_context(bs);
2258 co = qemu_coroutine_create(bdrv_flush_co_entry);
2259 qemu_coroutine_enter(co, &rwco);
2260 while (rwco.ret == NOT_DONE) {
2261 aio_poll(aio_context, true);
2268 typedef struct DiscardCo {
2269 BlockDriverState *bs;
2274 static void coroutine_fn bdrv_discard_co_entry(void *opaque)
2276 DiscardCo *rwco = opaque;
2278 rwco->ret = bdrv_co_discard(rwco->bs, rwco->sector_num, rwco->nb_sectors);
2281 int coroutine_fn bdrv_co_discard(BlockDriverState *bs, int64_t sector_num,
2284 BdrvTrackedRequest req;
2285 int max_discard, ret;
2291 ret = bdrv_check_request(bs, sector_num, nb_sectors);
2294 } else if (bs->read_only) {
2297 assert(!(bs->open_flags & BDRV_O_INACTIVE));
2299 /* Do nothing if disabled. */
2300 if (!(bs->open_flags & BDRV_O_UNMAP)) {
2304 if (!bs->drv->bdrv_co_discard && !bs->drv->bdrv_aio_discard) {
2308 tracked_request_begin(&req, bs, sector_num, nb_sectors,
2309 BDRV_TRACKED_DISCARD);
2310 bdrv_set_dirty(bs, sector_num, nb_sectors);
2312 max_discard = MIN_NON_ZERO(bs->bl.max_discard, BDRV_REQUEST_MAX_SECTORS);
2313 while (nb_sectors > 0) {
2315 int num = nb_sectors;
2318 if (bs->bl.discard_alignment &&
2319 num >= bs->bl.discard_alignment &&
2320 sector_num % bs->bl.discard_alignment) {
2321 if (num > bs->bl.discard_alignment) {
2322 num = bs->bl.discard_alignment;
2324 num -= sector_num % bs->bl.discard_alignment;
2327 /* limit request size */
2328 if (num > max_discard) {
2332 if (bs->drv->bdrv_co_discard) {
2333 ret = bs->drv->bdrv_co_discard(bs, sector_num, num);
2336 CoroutineIOCompletion co = {
2337 .coroutine = qemu_coroutine_self(),
2340 acb = bs->drv->bdrv_aio_discard(bs, sector_num, nb_sectors,
2341 bdrv_co_io_em_complete, &co);
2346 qemu_coroutine_yield();
2350 if (ret && ret != -ENOTSUP) {
2359 tracked_request_end(&req);
2363 int bdrv_discard(BlockDriverState *bs, int64_t sector_num, int nb_sectors)
2368 .sector_num = sector_num,
2369 .nb_sectors = nb_sectors,
2373 if (qemu_in_coroutine()) {
2374 /* Fast-path if already in coroutine context */
2375 bdrv_discard_co_entry(&rwco);
2377 AioContext *aio_context = bdrv_get_aio_context(bs);
2379 co = qemu_coroutine_create(bdrv_discard_co_entry);
2380 qemu_coroutine_enter(co, &rwco);
2381 while (rwco.ret == NOT_DONE) {
2382 aio_poll(aio_context, true);
2389 static int bdrv_co_do_ioctl(BlockDriverState *bs, int req, void *buf)
2391 BlockDriver *drv = bs->drv;
2392 BdrvTrackedRequest tracked_req;
2393 CoroutineIOCompletion co = {
2394 .coroutine = qemu_coroutine_self(),
2398 tracked_request_begin(&tracked_req, bs, 0, 0, BDRV_TRACKED_IOCTL);
2399 if (!drv || !drv->bdrv_aio_ioctl) {
2404 acb = drv->bdrv_aio_ioctl(bs, req, buf, bdrv_co_io_em_complete, &co);
2409 qemu_coroutine_yield();
2411 tracked_request_end(&tracked_req);
2416 BlockDriverState *bs;
2422 static void coroutine_fn bdrv_co_ioctl_entry(void *opaque)
2424 BdrvIoctlCoData *data = opaque;
2425 data->ret = bdrv_co_do_ioctl(data->bs, data->req, data->buf);
2428 /* needed for generic scsi interface */
2429 int bdrv_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
2431 BdrvIoctlCoData data = {
2435 .ret = -EINPROGRESS,
2438 if (qemu_in_coroutine()) {
2439 /* Fast-path if already in coroutine context */
2440 bdrv_co_ioctl_entry(&data);
2442 Coroutine *co = qemu_coroutine_create(bdrv_co_ioctl_entry);
2444 qemu_coroutine_enter(co, &data);
2445 while (data.ret == -EINPROGRESS) {
2446 aio_poll(bdrv_get_aio_context(bs), true);
2452 static void coroutine_fn bdrv_co_aio_ioctl_entry(void *opaque)
2454 BlockAIOCBCoroutine *acb = opaque;
2455 acb->req.error = bdrv_co_do_ioctl(acb->common.bs,
2456 acb->req.req, acb->req.buf);
2457 bdrv_co_complete(acb);
2460 BlockAIOCB *bdrv_aio_ioctl(BlockDriverState *bs,
2461 unsigned long int req, void *buf,
2462 BlockCompletionFunc *cb, void *opaque)
2464 BlockAIOCBCoroutine *acb = qemu_aio_get(&bdrv_em_co_aiocb_info,
2468 acb->need_bh = true;
2469 acb->req.error = -EINPROGRESS;
2472 co = qemu_coroutine_create(bdrv_co_aio_ioctl_entry);
2473 qemu_coroutine_enter(co, acb);
2475 bdrv_co_maybe_schedule_bh(acb);
2476 return &acb->common;
2479 void *qemu_blockalign(BlockDriverState *bs, size_t size)
2481 return qemu_memalign(bdrv_opt_mem_align(bs), size);
2484 void *qemu_blockalign0(BlockDriverState *bs, size_t size)
2486 return memset(qemu_blockalign(bs, size), 0, size);
2489 void *qemu_try_blockalign(BlockDriverState *bs, size_t size)
2491 size_t align = bdrv_opt_mem_align(bs);
2493 /* Ensure that NULL is never returned on success */
2499 return qemu_try_memalign(align, size);
2502 void *qemu_try_blockalign0(BlockDriverState *bs, size_t size)
2504 void *mem = qemu_try_blockalign(bs, size);
2507 memset(mem, 0, size);
2514 * Check if all memory in this vector is sector aligned.
2516 bool bdrv_qiov_is_aligned(BlockDriverState *bs, QEMUIOVector *qiov)
2519 size_t alignment = bdrv_min_mem_align(bs);
2521 for (i = 0; i < qiov->niov; i++) {
2522 if ((uintptr_t) qiov->iov[i].iov_base % alignment) {
2525 if (qiov->iov[i].iov_len % alignment) {
2533 void bdrv_add_before_write_notifier(BlockDriverState *bs,
2534 NotifierWithReturn *notifier)
2536 notifier_with_return_list_add(&bs->before_write_notifiers, notifier);
2539 void bdrv_io_plug(BlockDriverState *bs)
2543 QLIST_FOREACH(child, &bs->children, next) {
2544 bdrv_io_plug(child->bs);
2547 if (bs->io_plugged++ == 0 && bs->io_plug_disabled == 0) {
2548 BlockDriver *drv = bs->drv;
2549 if (drv && drv->bdrv_io_plug) {
2550 drv->bdrv_io_plug(bs);
2555 void bdrv_io_unplug(BlockDriverState *bs)
2559 assert(bs->io_plugged);
2560 if (--bs->io_plugged == 0 && bs->io_plug_disabled == 0) {
2561 BlockDriver *drv = bs->drv;
2562 if (drv && drv->bdrv_io_unplug) {
2563 drv->bdrv_io_unplug(bs);
2567 QLIST_FOREACH(child, &bs->children, next) {
2568 bdrv_io_unplug(child->bs);
2572 void bdrv_io_unplugged_begin(BlockDriverState *bs)
2576 if (bs->io_plug_disabled++ == 0 && bs->io_plugged > 0) {
2577 BlockDriver *drv = bs->drv;
2578 if (drv && drv->bdrv_io_unplug) {
2579 drv->bdrv_io_unplug(bs);
2583 QLIST_FOREACH(child, &bs->children, next) {
2584 bdrv_io_unplugged_begin(child->bs);
2588 void bdrv_io_unplugged_end(BlockDriverState *bs)
2592 assert(bs->io_plug_disabled);
2593 QLIST_FOREACH(child, &bs->children, next) {
2594 bdrv_io_unplugged_end(child->bs);
2597 if (--bs->io_plug_disabled == 0 && bs->io_plugged > 0) {
2598 BlockDriver *drv = bs->drv;
2599 if (drv && drv->bdrv_io_plug) {
2600 drv->bdrv_io_plug(bs);