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/aio-wait.h"
29 #include "block/blockjob.h"
30 #include "block/blockjob_int.h"
31 #include "block/block_int.h"
32 #include "block/coroutines.h"
33 #include "block/write-threshold.h"
34 #include "qemu/cutils.h"
35 #include "qapi/error.h"
36 #include "qemu/error-report.h"
37 #include "qemu/main-loop.h"
38 #include "sysemu/replay.h"
40 /* Maximum bounce buffer for copy-on-read and write zeroes, in bytes */
41 #define MAX_BOUNCE_BUFFER (32768 << BDRV_SECTOR_BITS)
43 static void bdrv_parent_cb_resize(BlockDriverState
*bs
);
44 static int coroutine_fn
bdrv_co_do_pwrite_zeroes(BlockDriverState
*bs
,
45 int64_t offset
, int64_t bytes
, BdrvRequestFlags flags
);
47 static void bdrv_parent_drained_begin(BlockDriverState
*bs
, BdrvChild
*ignore
,
48 bool ignore_bds_parents
)
52 QLIST_FOREACH_SAFE(c
, &bs
->parents
, next_parent
, next
) {
53 if (c
== ignore
|| (ignore_bds_parents
&& c
->klass
->parent_is_bds
)) {
56 bdrv_parent_drained_begin_single(c
, false);
60 static void bdrv_parent_drained_end_single_no_poll(BdrvChild
*c
,
61 int *drained_end_counter
)
63 assert(c
->parent_quiesce_counter
> 0);
64 c
->parent_quiesce_counter
--;
65 if (c
->klass
->drained_end
) {
66 c
->klass
->drained_end(c
, drained_end_counter
);
70 void bdrv_parent_drained_end_single(BdrvChild
*c
)
72 int drained_end_counter
= 0;
73 bdrv_parent_drained_end_single_no_poll(c
, &drained_end_counter
);
74 BDRV_POLL_WHILE(c
->bs
, qatomic_read(&drained_end_counter
) > 0);
77 static void bdrv_parent_drained_end(BlockDriverState
*bs
, BdrvChild
*ignore
,
78 bool ignore_bds_parents
,
79 int *drained_end_counter
)
83 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
84 if (c
== ignore
|| (ignore_bds_parents
&& c
->klass
->parent_is_bds
)) {
87 bdrv_parent_drained_end_single_no_poll(c
, drained_end_counter
);
91 static bool bdrv_parent_drained_poll_single(BdrvChild
*c
)
93 if (c
->klass
->drained_poll
) {
94 return c
->klass
->drained_poll(c
);
99 static bool bdrv_parent_drained_poll(BlockDriverState
*bs
, BdrvChild
*ignore
,
100 bool ignore_bds_parents
)
105 QLIST_FOREACH_SAFE(c
, &bs
->parents
, next_parent
, next
) {
106 if (c
== ignore
|| (ignore_bds_parents
&& c
->klass
->parent_is_bds
)) {
109 busy
|= bdrv_parent_drained_poll_single(c
);
115 void bdrv_parent_drained_begin_single(BdrvChild
*c
, bool poll
)
117 c
->parent_quiesce_counter
++;
118 if (c
->klass
->drained_begin
) {
119 c
->klass
->drained_begin(c
);
122 BDRV_POLL_WHILE(c
->bs
, bdrv_parent_drained_poll_single(c
));
126 static void bdrv_merge_limits(BlockLimits
*dst
, const BlockLimits
*src
)
128 dst
->opt_transfer
= MAX(dst
->opt_transfer
, src
->opt_transfer
);
129 dst
->max_transfer
= MIN_NON_ZERO(dst
->max_transfer
, src
->max_transfer
);
130 dst
->max_hw_transfer
= MIN_NON_ZERO(dst
->max_hw_transfer
,
131 src
->max_hw_transfer
);
132 dst
->opt_mem_alignment
= MAX(dst
->opt_mem_alignment
,
133 src
->opt_mem_alignment
);
134 dst
->min_mem_alignment
= MAX(dst
->min_mem_alignment
,
135 src
->min_mem_alignment
);
136 dst
->max_iov
= MIN_NON_ZERO(dst
->max_iov
, src
->max_iov
);
139 typedef struct BdrvRefreshLimitsState
{
140 BlockDriverState
*bs
;
142 } BdrvRefreshLimitsState
;
144 static void bdrv_refresh_limits_abort(void *opaque
)
146 BdrvRefreshLimitsState
*s
= opaque
;
148 s
->bs
->bl
= s
->old_bl
;
151 static TransactionActionDrv bdrv_refresh_limits_drv
= {
152 .abort
= bdrv_refresh_limits_abort
,
156 /* @tran is allowed to be NULL, in this case no rollback is possible. */
157 void bdrv_refresh_limits(BlockDriverState
*bs
, Transaction
*tran
, Error
**errp
)
160 BlockDriver
*drv
= bs
->drv
;
165 BdrvRefreshLimitsState
*s
= g_new(BdrvRefreshLimitsState
, 1);
166 *s
= (BdrvRefreshLimitsState
) {
170 tran_add(tran
, &bdrv_refresh_limits_drv
, s
);
173 memset(&bs
->bl
, 0, sizeof(bs
->bl
));
179 /* Default alignment based on whether driver has byte interface */
180 bs
->bl
.request_alignment
= (drv
->bdrv_co_preadv
||
181 drv
->bdrv_aio_preadv
||
182 drv
->bdrv_co_preadv_part
) ? 1 : 512;
184 /* Take some limits from the children as a default */
186 QLIST_FOREACH(c
, &bs
->children
, next
) {
187 if (c
->role
& (BDRV_CHILD_DATA
| BDRV_CHILD_FILTERED
| BDRV_CHILD_COW
))
189 bdrv_refresh_limits(c
->bs
, tran
, errp
);
193 bdrv_merge_limits(&bs
->bl
, &c
->bs
->bl
);
199 bs
->bl
.min_mem_alignment
= 512;
200 bs
->bl
.opt_mem_alignment
= qemu_real_host_page_size
;
202 /* Safe default since most protocols use readv()/writev()/etc */
203 bs
->bl
.max_iov
= IOV_MAX
;
206 /* Then let the driver override it */
207 if (drv
->bdrv_refresh_limits
) {
208 drv
->bdrv_refresh_limits(bs
, errp
);
214 if (bs
->bl
.request_alignment
> BDRV_MAX_ALIGNMENT
) {
215 error_setg(errp
, "Driver requires too large request alignment");
220 * The copy-on-read flag is actually a reference count so multiple users may
221 * use the feature without worrying about clobbering its previous state.
222 * Copy-on-read stays enabled until all users have called to disable it.
224 void bdrv_enable_copy_on_read(BlockDriverState
*bs
)
226 qatomic_inc(&bs
->copy_on_read
);
229 void bdrv_disable_copy_on_read(BlockDriverState
*bs
)
231 int old
= qatomic_fetch_dec(&bs
->copy_on_read
);
237 BlockDriverState
*bs
;
243 bool ignore_bds_parents
;
244 int *drained_end_counter
;
247 static void coroutine_fn
bdrv_drain_invoke_entry(void *opaque
)
249 BdrvCoDrainData
*data
= opaque
;
250 BlockDriverState
*bs
= data
->bs
;
253 bs
->drv
->bdrv_co_drain_begin(bs
);
255 bs
->drv
->bdrv_co_drain_end(bs
);
258 /* Set data->done and decrement drained_end_counter before bdrv_wakeup() */
259 qatomic_mb_set(&data
->done
, true);
261 qatomic_dec(data
->drained_end_counter
);
263 bdrv_dec_in_flight(bs
);
268 /* Recursively call BlockDriver.bdrv_co_drain_begin/end callbacks */
269 static void bdrv_drain_invoke(BlockDriverState
*bs
, bool begin
,
270 int *drained_end_counter
)
272 BdrvCoDrainData
*data
;
274 if (!bs
->drv
|| (begin
&& !bs
->drv
->bdrv_co_drain_begin
) ||
275 (!begin
&& !bs
->drv
->bdrv_co_drain_end
)) {
279 data
= g_new(BdrvCoDrainData
, 1);
280 *data
= (BdrvCoDrainData
) {
284 .drained_end_counter
= drained_end_counter
,
288 qatomic_inc(drained_end_counter
);
291 /* Make sure the driver callback completes during the polling phase for
293 bdrv_inc_in_flight(bs
);
294 data
->co
= qemu_coroutine_create(bdrv_drain_invoke_entry
, data
);
295 aio_co_schedule(bdrv_get_aio_context(bs
), data
->co
);
298 /* Returns true if BDRV_POLL_WHILE() should go into a blocking aio_poll() */
299 bool bdrv_drain_poll(BlockDriverState
*bs
, bool recursive
,
300 BdrvChild
*ignore_parent
, bool ignore_bds_parents
)
302 BdrvChild
*child
, *next
;
304 if (bdrv_parent_drained_poll(bs
, ignore_parent
, ignore_bds_parents
)) {
308 if (qatomic_read(&bs
->in_flight
)) {
313 assert(!ignore_bds_parents
);
314 QLIST_FOREACH_SAFE(child
, &bs
->children
, next
, next
) {
315 if (bdrv_drain_poll(child
->bs
, recursive
, child
, false)) {
324 static bool bdrv_drain_poll_top_level(BlockDriverState
*bs
, bool recursive
,
325 BdrvChild
*ignore_parent
)
327 return bdrv_drain_poll(bs
, recursive
, ignore_parent
, false);
330 static void bdrv_do_drained_begin(BlockDriverState
*bs
, bool recursive
,
331 BdrvChild
*parent
, bool ignore_bds_parents
,
333 static void bdrv_do_drained_end(BlockDriverState
*bs
, bool recursive
,
334 BdrvChild
*parent
, bool ignore_bds_parents
,
335 int *drained_end_counter
);
337 static void bdrv_co_drain_bh_cb(void *opaque
)
339 BdrvCoDrainData
*data
= opaque
;
340 Coroutine
*co
= data
->co
;
341 BlockDriverState
*bs
= data
->bs
;
344 AioContext
*ctx
= bdrv_get_aio_context(bs
);
345 aio_context_acquire(ctx
);
346 bdrv_dec_in_flight(bs
);
348 assert(!data
->drained_end_counter
);
349 bdrv_do_drained_begin(bs
, data
->recursive
, data
->parent
,
350 data
->ignore_bds_parents
, data
->poll
);
353 bdrv_do_drained_end(bs
, data
->recursive
, data
->parent
,
354 data
->ignore_bds_parents
,
355 data
->drained_end_counter
);
357 aio_context_release(ctx
);
360 bdrv_drain_all_begin();
367 static void coroutine_fn
bdrv_co_yield_to_drain(BlockDriverState
*bs
,
368 bool begin
, bool recursive
,
370 bool ignore_bds_parents
,
372 int *drained_end_counter
)
374 BdrvCoDrainData data
;
375 Coroutine
*self
= qemu_coroutine_self();
376 AioContext
*ctx
= bdrv_get_aio_context(bs
);
377 AioContext
*co_ctx
= qemu_coroutine_get_aio_context(self
);
379 /* Calling bdrv_drain() from a BH ensures the current coroutine yields and
380 * other coroutines run if they were queued by aio_co_enter(). */
382 assert(qemu_in_coroutine());
383 data
= (BdrvCoDrainData
) {
388 .recursive
= recursive
,
390 .ignore_bds_parents
= ignore_bds_parents
,
392 .drained_end_counter
= drained_end_counter
,
396 bdrv_inc_in_flight(bs
);
400 * Temporarily drop the lock across yield or we would get deadlocks.
401 * bdrv_co_drain_bh_cb() reaquires the lock as needed.
403 * When we yield below, the lock for the current context will be
404 * released, so if this is actually the lock that protects bs, don't drop
408 aio_context_release(ctx
);
410 replay_bh_schedule_oneshot_event(ctx
, bdrv_co_drain_bh_cb
, &data
);
412 qemu_coroutine_yield();
413 /* If we are resumed from some other event (such as an aio completion or a
414 * timer callback), it is a bug in the caller that should be fixed. */
417 /* Reaquire the AioContext of bs if we dropped it */
419 aio_context_acquire(ctx
);
423 void bdrv_do_drained_begin_quiesce(BlockDriverState
*bs
,
424 BdrvChild
*parent
, bool ignore_bds_parents
)
426 assert(!qemu_in_coroutine());
428 /* Stop things in parent-to-child order */
429 if (qatomic_fetch_inc(&bs
->quiesce_counter
) == 0) {
430 aio_disable_external(bdrv_get_aio_context(bs
));
433 bdrv_parent_drained_begin(bs
, parent
, ignore_bds_parents
);
434 bdrv_drain_invoke(bs
, true, NULL
);
437 static void bdrv_do_drained_begin(BlockDriverState
*bs
, bool recursive
,
438 BdrvChild
*parent
, bool ignore_bds_parents
,
441 BdrvChild
*child
, *next
;
443 if (qemu_in_coroutine()) {
444 bdrv_co_yield_to_drain(bs
, true, recursive
, parent
, ignore_bds_parents
,
449 bdrv_do_drained_begin_quiesce(bs
, parent
, ignore_bds_parents
);
452 assert(!ignore_bds_parents
);
453 bs
->recursive_quiesce_counter
++;
454 QLIST_FOREACH_SAFE(child
, &bs
->children
, next
, next
) {
455 bdrv_do_drained_begin(child
->bs
, true, child
, ignore_bds_parents
,
461 * Wait for drained requests to finish.
463 * Calling BDRV_POLL_WHILE() only once for the top-level node is okay: The
464 * call is needed so things in this AioContext can make progress even
465 * though we don't return to the main AioContext loop - this automatically
466 * includes other nodes in the same AioContext and therefore all child
470 assert(!ignore_bds_parents
);
471 BDRV_POLL_WHILE(bs
, bdrv_drain_poll_top_level(bs
, recursive
, parent
));
475 void bdrv_drained_begin(BlockDriverState
*bs
)
477 bdrv_do_drained_begin(bs
, false, NULL
, false, true);
480 void bdrv_subtree_drained_begin(BlockDriverState
*bs
)
482 bdrv_do_drained_begin(bs
, true, NULL
, false, true);
486 * This function does not poll, nor must any of its recursively called
487 * functions. The *drained_end_counter pointee will be incremented
488 * once for every background operation scheduled, and decremented once
489 * the operation settles. Therefore, the pointer must remain valid
490 * until the pointee reaches 0. That implies that whoever sets up the
491 * pointee has to poll until it is 0.
493 * We use atomic operations to access *drained_end_counter, because
494 * (1) when called from bdrv_set_aio_context_ignore(), the subgraph of
495 * @bs may contain nodes in different AioContexts,
496 * (2) bdrv_drain_all_end() uses the same counter for all nodes,
497 * regardless of which AioContext they are in.
499 static void bdrv_do_drained_end(BlockDriverState
*bs
, bool recursive
,
500 BdrvChild
*parent
, bool ignore_bds_parents
,
501 int *drained_end_counter
)
504 int old_quiesce_counter
;
506 assert(drained_end_counter
!= NULL
);
508 if (qemu_in_coroutine()) {
509 bdrv_co_yield_to_drain(bs
, false, recursive
, parent
, ignore_bds_parents
,
510 false, drained_end_counter
);
513 assert(bs
->quiesce_counter
> 0);
515 /* Re-enable things in child-to-parent order */
516 bdrv_drain_invoke(bs
, false, drained_end_counter
);
517 bdrv_parent_drained_end(bs
, parent
, ignore_bds_parents
,
518 drained_end_counter
);
520 old_quiesce_counter
= qatomic_fetch_dec(&bs
->quiesce_counter
);
521 if (old_quiesce_counter
== 1) {
522 aio_enable_external(bdrv_get_aio_context(bs
));
526 assert(!ignore_bds_parents
);
527 bs
->recursive_quiesce_counter
--;
528 QLIST_FOREACH(child
, &bs
->children
, next
) {
529 bdrv_do_drained_end(child
->bs
, true, child
, ignore_bds_parents
,
530 drained_end_counter
);
535 void bdrv_drained_end(BlockDriverState
*bs
)
537 int drained_end_counter
= 0;
538 bdrv_do_drained_end(bs
, false, NULL
, false, &drained_end_counter
);
539 BDRV_POLL_WHILE(bs
, qatomic_read(&drained_end_counter
) > 0);
542 void bdrv_drained_end_no_poll(BlockDriverState
*bs
, int *drained_end_counter
)
544 bdrv_do_drained_end(bs
, false, NULL
, false, drained_end_counter
);
547 void bdrv_subtree_drained_end(BlockDriverState
*bs
)
549 int drained_end_counter
= 0;
550 bdrv_do_drained_end(bs
, true, NULL
, false, &drained_end_counter
);
551 BDRV_POLL_WHILE(bs
, qatomic_read(&drained_end_counter
) > 0);
554 void bdrv_apply_subtree_drain(BdrvChild
*child
, BlockDriverState
*new_parent
)
558 for (i
= 0; i
< new_parent
->recursive_quiesce_counter
; i
++) {
559 bdrv_do_drained_begin(child
->bs
, true, child
, false, true);
563 void bdrv_unapply_subtree_drain(BdrvChild
*child
, BlockDriverState
*old_parent
)
565 int drained_end_counter
= 0;
568 for (i
= 0; i
< old_parent
->recursive_quiesce_counter
; i
++) {
569 bdrv_do_drained_end(child
->bs
, true, child
, false,
570 &drained_end_counter
);
573 BDRV_POLL_WHILE(child
->bs
, qatomic_read(&drained_end_counter
) > 0);
577 * Wait for pending requests to complete on a single BlockDriverState subtree,
578 * and suspend block driver's internal I/O until next request arrives.
580 * Note that unlike bdrv_drain_all(), the caller must hold the BlockDriverState
583 void coroutine_fn
bdrv_co_drain(BlockDriverState
*bs
)
585 assert(qemu_in_coroutine());
586 bdrv_drained_begin(bs
);
587 bdrv_drained_end(bs
);
590 void bdrv_drain(BlockDriverState
*bs
)
592 bdrv_drained_begin(bs
);
593 bdrv_drained_end(bs
);
596 static void bdrv_drain_assert_idle(BlockDriverState
*bs
)
598 BdrvChild
*child
, *next
;
600 assert(qatomic_read(&bs
->in_flight
) == 0);
601 QLIST_FOREACH_SAFE(child
, &bs
->children
, next
, next
) {
602 bdrv_drain_assert_idle(child
->bs
);
606 unsigned int bdrv_drain_all_count
= 0;
608 static bool bdrv_drain_all_poll(void)
610 BlockDriverState
*bs
= NULL
;
613 /* bdrv_drain_poll() can't make changes to the graph and we are holding the
614 * main AioContext lock, so iterating bdrv_next_all_states() is safe. */
615 while ((bs
= bdrv_next_all_states(bs
))) {
616 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
617 aio_context_acquire(aio_context
);
618 result
|= bdrv_drain_poll(bs
, false, NULL
, true);
619 aio_context_release(aio_context
);
626 * Wait for pending requests to complete across all BlockDriverStates
628 * This function does not flush data to disk, use bdrv_flush_all() for that
629 * after calling this function.
631 * This pauses all block jobs and disables external clients. It must
632 * be paired with bdrv_drain_all_end().
634 * NOTE: no new block jobs or BlockDriverStates can be created between
635 * the bdrv_drain_all_begin() and bdrv_drain_all_end() calls.
637 void bdrv_drain_all_begin(void)
639 BlockDriverState
*bs
= NULL
;
641 if (qemu_in_coroutine()) {
642 bdrv_co_yield_to_drain(NULL
, true, false, NULL
, true, true, NULL
);
647 * bdrv queue is managed by record/replay,
648 * waiting for finishing the I/O requests may
651 if (replay_events_enabled()) {
655 /* AIO_WAIT_WHILE() with a NULL context can only be called from the main
656 * loop AioContext, so make sure we're in the main context. */
657 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
658 assert(bdrv_drain_all_count
< INT_MAX
);
659 bdrv_drain_all_count
++;
661 /* Quiesce all nodes, without polling in-flight requests yet. The graph
662 * cannot change during this loop. */
663 while ((bs
= bdrv_next_all_states(bs
))) {
664 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
666 aio_context_acquire(aio_context
);
667 bdrv_do_drained_begin(bs
, false, NULL
, true, false);
668 aio_context_release(aio_context
);
671 /* Now poll the in-flight requests */
672 AIO_WAIT_WHILE(NULL
, bdrv_drain_all_poll());
674 while ((bs
= bdrv_next_all_states(bs
))) {
675 bdrv_drain_assert_idle(bs
);
679 void bdrv_drain_all_end_quiesce(BlockDriverState
*bs
)
681 int drained_end_counter
= 0;
683 g_assert(bs
->quiesce_counter
> 0);
684 g_assert(!bs
->refcnt
);
686 while (bs
->quiesce_counter
) {
687 bdrv_do_drained_end(bs
, false, NULL
, true, &drained_end_counter
);
689 BDRV_POLL_WHILE(bs
, qatomic_read(&drained_end_counter
) > 0);
692 void bdrv_drain_all_end(void)
694 BlockDriverState
*bs
= NULL
;
695 int drained_end_counter
= 0;
698 * bdrv queue is managed by record/replay,
699 * waiting for finishing the I/O requests may
702 if (replay_events_enabled()) {
706 while ((bs
= bdrv_next_all_states(bs
))) {
707 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
709 aio_context_acquire(aio_context
);
710 bdrv_do_drained_end(bs
, false, NULL
, true, &drained_end_counter
);
711 aio_context_release(aio_context
);
714 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
715 AIO_WAIT_WHILE(NULL
, qatomic_read(&drained_end_counter
) > 0);
717 assert(bdrv_drain_all_count
> 0);
718 bdrv_drain_all_count
--;
721 void bdrv_drain_all(void)
723 bdrv_drain_all_begin();
724 bdrv_drain_all_end();
728 * Remove an active request from the tracked requests list
730 * This function should be called when a tracked request is completing.
732 static void tracked_request_end(BdrvTrackedRequest
*req
)
734 if (req
->serialising
) {
735 qatomic_dec(&req
->bs
->serialising_in_flight
);
738 qemu_co_mutex_lock(&req
->bs
->reqs_lock
);
739 QLIST_REMOVE(req
, list
);
740 qemu_co_queue_restart_all(&req
->wait_queue
);
741 qemu_co_mutex_unlock(&req
->bs
->reqs_lock
);
745 * Add an active request to the tracked requests list
747 static void tracked_request_begin(BdrvTrackedRequest
*req
,
748 BlockDriverState
*bs
,
751 enum BdrvTrackedRequestType type
)
753 bdrv_check_request(offset
, bytes
, &error_abort
);
755 *req
= (BdrvTrackedRequest
){
760 .co
= qemu_coroutine_self(),
761 .serialising
= false,
762 .overlap_offset
= offset
,
763 .overlap_bytes
= bytes
,
766 qemu_co_queue_init(&req
->wait_queue
);
768 qemu_co_mutex_lock(&bs
->reqs_lock
);
769 QLIST_INSERT_HEAD(&bs
->tracked_requests
, req
, list
);
770 qemu_co_mutex_unlock(&bs
->reqs_lock
);
773 static bool tracked_request_overlaps(BdrvTrackedRequest
*req
,
774 int64_t offset
, int64_t bytes
)
776 bdrv_check_request(offset
, bytes
, &error_abort
);
779 if (offset
>= req
->overlap_offset
+ req
->overlap_bytes
) {
783 if (req
->overlap_offset
>= offset
+ bytes
) {
789 /* Called with self->bs->reqs_lock held */
790 static BdrvTrackedRequest
*
791 bdrv_find_conflicting_request(BdrvTrackedRequest
*self
)
793 BdrvTrackedRequest
*req
;
795 QLIST_FOREACH(req
, &self
->bs
->tracked_requests
, list
) {
796 if (req
== self
|| (!req
->serialising
&& !self
->serialising
)) {
799 if (tracked_request_overlaps(req
, self
->overlap_offset
,
800 self
->overlap_bytes
))
803 * Hitting this means there was a reentrant request, for
804 * example, a block driver issuing nested requests. This must
805 * never happen since it means deadlock.
807 assert(qemu_coroutine_self() != req
->co
);
810 * If the request is already (indirectly) waiting for us, or
811 * will wait for us as soon as it wakes up, then just go on
812 * (instead of producing a deadlock in the former case).
814 if (!req
->waiting_for
) {
823 /* Called with self->bs->reqs_lock held */
824 static bool coroutine_fn
825 bdrv_wait_serialising_requests_locked(BdrvTrackedRequest
*self
)
827 BdrvTrackedRequest
*req
;
830 while ((req
= bdrv_find_conflicting_request(self
))) {
831 self
->waiting_for
= req
;
832 qemu_co_queue_wait(&req
->wait_queue
, &self
->bs
->reqs_lock
);
833 self
->waiting_for
= NULL
;
840 /* Called with req->bs->reqs_lock held */
841 static void tracked_request_set_serialising(BdrvTrackedRequest
*req
,
844 int64_t overlap_offset
= req
->offset
& ~(align
- 1);
845 int64_t overlap_bytes
=
846 ROUND_UP(req
->offset
+ req
->bytes
, align
) - overlap_offset
;
848 bdrv_check_request(req
->offset
, req
->bytes
, &error_abort
);
850 if (!req
->serialising
) {
851 qatomic_inc(&req
->bs
->serialising_in_flight
);
852 req
->serialising
= true;
855 req
->overlap_offset
= MIN(req
->overlap_offset
, overlap_offset
);
856 req
->overlap_bytes
= MAX(req
->overlap_bytes
, overlap_bytes
);
860 * Return the tracked request on @bs for the current coroutine, or
861 * NULL if there is none.
863 BdrvTrackedRequest
*coroutine_fn
bdrv_co_get_self_request(BlockDriverState
*bs
)
865 BdrvTrackedRequest
*req
;
866 Coroutine
*self
= qemu_coroutine_self();
868 QLIST_FOREACH(req
, &bs
->tracked_requests
, list
) {
869 if (req
->co
== self
) {
878 * Round a region to cluster boundaries
880 void bdrv_round_to_clusters(BlockDriverState
*bs
,
881 int64_t offset
, int64_t bytes
,
882 int64_t *cluster_offset
,
883 int64_t *cluster_bytes
)
887 if (bdrv_get_info(bs
, &bdi
) < 0 || bdi
.cluster_size
== 0) {
888 *cluster_offset
= offset
;
889 *cluster_bytes
= bytes
;
891 int64_t c
= bdi
.cluster_size
;
892 *cluster_offset
= QEMU_ALIGN_DOWN(offset
, c
);
893 *cluster_bytes
= QEMU_ALIGN_UP(offset
- *cluster_offset
+ bytes
, c
);
897 static int bdrv_get_cluster_size(BlockDriverState
*bs
)
902 ret
= bdrv_get_info(bs
, &bdi
);
903 if (ret
< 0 || bdi
.cluster_size
== 0) {
904 return bs
->bl
.request_alignment
;
906 return bdi
.cluster_size
;
910 void bdrv_inc_in_flight(BlockDriverState
*bs
)
912 qatomic_inc(&bs
->in_flight
);
915 void bdrv_wakeup(BlockDriverState
*bs
)
920 void bdrv_dec_in_flight(BlockDriverState
*bs
)
922 qatomic_dec(&bs
->in_flight
);
926 static bool coroutine_fn
bdrv_wait_serialising_requests(BdrvTrackedRequest
*self
)
928 BlockDriverState
*bs
= self
->bs
;
931 if (!qatomic_read(&bs
->serialising_in_flight
)) {
935 qemu_co_mutex_lock(&bs
->reqs_lock
);
936 waited
= bdrv_wait_serialising_requests_locked(self
);
937 qemu_co_mutex_unlock(&bs
->reqs_lock
);
942 bool coroutine_fn
bdrv_make_request_serialising(BdrvTrackedRequest
*req
,
947 qemu_co_mutex_lock(&req
->bs
->reqs_lock
);
949 tracked_request_set_serialising(req
, align
);
950 waited
= bdrv_wait_serialising_requests_locked(req
);
952 qemu_co_mutex_unlock(&req
->bs
->reqs_lock
);
957 static int bdrv_check_qiov_request(int64_t offset
, int64_t bytes
,
958 QEMUIOVector
*qiov
, size_t qiov_offset
,
962 * Check generic offset/bytes correctness
966 error_setg(errp
, "offset is negative: %" PRIi64
, offset
);
971 error_setg(errp
, "bytes is negative: %" PRIi64
, bytes
);
975 if (bytes
> BDRV_MAX_LENGTH
) {
976 error_setg(errp
, "bytes(%" PRIi64
") exceeds maximum(%" PRIi64
")",
977 bytes
, BDRV_MAX_LENGTH
);
981 if (offset
> BDRV_MAX_LENGTH
) {
982 error_setg(errp
, "offset(%" PRIi64
") exceeds maximum(%" PRIi64
")",
983 offset
, BDRV_MAX_LENGTH
);
987 if (offset
> BDRV_MAX_LENGTH
- bytes
) {
988 error_setg(errp
, "sum of offset(%" PRIi64
") and bytes(%" PRIi64
") "
989 "exceeds maximum(%" PRIi64
")", offset
, bytes
,
999 * Check qiov and qiov_offset
1002 if (qiov_offset
> qiov
->size
) {
1003 error_setg(errp
, "qiov_offset(%zu) overflow io vector size(%zu)",
1004 qiov_offset
, qiov
->size
);
1008 if (bytes
> qiov
->size
- qiov_offset
) {
1009 error_setg(errp
, "bytes(%" PRIi64
") + qiov_offset(%zu) overflow io "
1010 "vector size(%zu)", bytes
, qiov_offset
, qiov
->size
);
1017 int bdrv_check_request(int64_t offset
, int64_t bytes
, Error
**errp
)
1019 return bdrv_check_qiov_request(offset
, bytes
, NULL
, 0, errp
);
1022 static int bdrv_check_request32(int64_t offset
, int64_t bytes
,
1023 QEMUIOVector
*qiov
, size_t qiov_offset
)
1025 int ret
= bdrv_check_qiov_request(offset
, bytes
, qiov
, qiov_offset
, NULL
);
1030 if (bytes
> BDRV_REQUEST_MAX_BYTES
) {
1037 int bdrv_pwrite_zeroes(BdrvChild
*child
, int64_t offset
,
1038 int64_t bytes
, BdrvRequestFlags flags
)
1040 return bdrv_pwritev(child
, offset
, bytes
, NULL
,
1041 BDRV_REQ_ZERO_WRITE
| flags
);
1045 * Completely zero out a block device with the help of bdrv_pwrite_zeroes.
1046 * The operation is sped up by checking the block status and only writing
1047 * zeroes to the device if they currently do not return zeroes. Optional
1048 * flags are passed through to bdrv_pwrite_zeroes (e.g. BDRV_REQ_MAY_UNMAP,
1051 * Returns < 0 on error, 0 on success. For error codes see bdrv_pwrite().
1053 int bdrv_make_zero(BdrvChild
*child
, BdrvRequestFlags flags
)
1056 int64_t target_size
, bytes
, offset
= 0;
1057 BlockDriverState
*bs
= child
->bs
;
1059 target_size
= bdrv_getlength(bs
);
1060 if (target_size
< 0) {
1065 bytes
= MIN(target_size
- offset
, BDRV_REQUEST_MAX_BYTES
);
1069 ret
= bdrv_block_status(bs
, offset
, bytes
, &bytes
, NULL
, NULL
);
1073 if (ret
& BDRV_BLOCK_ZERO
) {
1077 ret
= bdrv_pwrite_zeroes(child
, offset
, bytes
, flags
);
1085 /* See bdrv_pwrite() for the return codes */
1086 int bdrv_pread(BdrvChild
*child
, int64_t offset
, void *buf
, int64_t bytes
)
1089 QEMUIOVector qiov
= QEMU_IOVEC_INIT_BUF(qiov
, buf
, bytes
);
1095 ret
= bdrv_preadv(child
, offset
, bytes
, &qiov
, 0);
1097 return ret
< 0 ? ret
: bytes
;
1100 /* Return no. of bytes on success or < 0 on error. Important errors are:
1101 -EIO generic I/O error (may happen for all errors)
1102 -ENOMEDIUM No media inserted.
1103 -EINVAL Invalid offset or number of bytes
1104 -EACCES Trying to write a read-only device
1106 int bdrv_pwrite(BdrvChild
*child
, int64_t offset
, const void *buf
,
1110 QEMUIOVector qiov
= QEMU_IOVEC_INIT_BUF(qiov
, buf
, bytes
);
1116 ret
= bdrv_pwritev(child
, offset
, bytes
, &qiov
, 0);
1118 return ret
< 0 ? ret
: bytes
;
1122 * Writes to the file and ensures that no writes are reordered across this
1123 * request (acts as a barrier)
1125 * Returns 0 on success, -errno in error cases.
1127 int bdrv_pwrite_sync(BdrvChild
*child
, int64_t offset
,
1128 const void *buf
, int64_t count
)
1132 ret
= bdrv_pwrite(child
, offset
, buf
, count
);
1137 ret
= bdrv_flush(child
->bs
);
1145 typedef struct CoroutineIOCompletion
{
1146 Coroutine
*coroutine
;
1148 } CoroutineIOCompletion
;
1150 static void bdrv_co_io_em_complete(void *opaque
, int ret
)
1152 CoroutineIOCompletion
*co
= opaque
;
1155 aio_co_wake(co
->coroutine
);
1158 static int coroutine_fn
bdrv_driver_preadv(BlockDriverState
*bs
,
1159 int64_t offset
, int64_t bytes
,
1161 size_t qiov_offset
, int flags
)
1163 BlockDriver
*drv
= bs
->drv
;
1165 unsigned int nb_sectors
;
1166 QEMUIOVector local_qiov
;
1169 bdrv_check_qiov_request(offset
, bytes
, qiov
, qiov_offset
, &error_abort
);
1170 assert(!(flags
& ~BDRV_REQ_MASK
));
1171 assert(!(flags
& BDRV_REQ_NO_FALLBACK
));
1177 if (drv
->bdrv_co_preadv_part
) {
1178 return drv
->bdrv_co_preadv_part(bs
, offset
, bytes
, qiov
, qiov_offset
,
1182 if (qiov_offset
> 0 || bytes
!= qiov
->size
) {
1183 qemu_iovec_init_slice(&local_qiov
, qiov
, qiov_offset
, bytes
);
1187 if (drv
->bdrv_co_preadv
) {
1188 ret
= drv
->bdrv_co_preadv(bs
, offset
, bytes
, qiov
, flags
);
1192 if (drv
->bdrv_aio_preadv
) {
1194 CoroutineIOCompletion co
= {
1195 .coroutine
= qemu_coroutine_self(),
1198 acb
= drv
->bdrv_aio_preadv(bs
, offset
, bytes
, qiov
, flags
,
1199 bdrv_co_io_em_complete
, &co
);
1204 qemu_coroutine_yield();
1210 sector_num
= offset
>> BDRV_SECTOR_BITS
;
1211 nb_sectors
= bytes
>> BDRV_SECTOR_BITS
;
1213 assert(QEMU_IS_ALIGNED(offset
, BDRV_SECTOR_SIZE
));
1214 assert(QEMU_IS_ALIGNED(bytes
, BDRV_SECTOR_SIZE
));
1215 assert(bytes
<= BDRV_REQUEST_MAX_BYTES
);
1216 assert(drv
->bdrv_co_readv
);
1218 ret
= drv
->bdrv_co_readv(bs
, sector_num
, nb_sectors
, qiov
);
1221 if (qiov
== &local_qiov
) {
1222 qemu_iovec_destroy(&local_qiov
);
1228 static int coroutine_fn
bdrv_driver_pwritev(BlockDriverState
*bs
,
1229 int64_t offset
, int64_t bytes
,
1231 size_t qiov_offset
, int flags
)
1233 BlockDriver
*drv
= bs
->drv
;
1235 unsigned int nb_sectors
;
1236 QEMUIOVector local_qiov
;
1239 bdrv_check_qiov_request(offset
, bytes
, qiov
, qiov_offset
, &error_abort
);
1240 assert(!(flags
& ~BDRV_REQ_MASK
));
1241 assert(!(flags
& BDRV_REQ_NO_FALLBACK
));
1247 if (drv
->bdrv_co_pwritev_part
) {
1248 ret
= drv
->bdrv_co_pwritev_part(bs
, offset
, bytes
, qiov
, qiov_offset
,
1249 flags
& bs
->supported_write_flags
);
1250 flags
&= ~bs
->supported_write_flags
;
1254 if (qiov_offset
> 0 || bytes
!= qiov
->size
) {
1255 qemu_iovec_init_slice(&local_qiov
, qiov
, qiov_offset
, bytes
);
1259 if (drv
->bdrv_co_pwritev
) {
1260 ret
= drv
->bdrv_co_pwritev(bs
, offset
, bytes
, qiov
,
1261 flags
& bs
->supported_write_flags
);
1262 flags
&= ~bs
->supported_write_flags
;
1266 if (drv
->bdrv_aio_pwritev
) {
1268 CoroutineIOCompletion co
= {
1269 .coroutine
= qemu_coroutine_self(),
1272 acb
= drv
->bdrv_aio_pwritev(bs
, offset
, bytes
, qiov
,
1273 flags
& bs
->supported_write_flags
,
1274 bdrv_co_io_em_complete
, &co
);
1275 flags
&= ~bs
->supported_write_flags
;
1279 qemu_coroutine_yield();
1285 sector_num
= offset
>> BDRV_SECTOR_BITS
;
1286 nb_sectors
= bytes
>> BDRV_SECTOR_BITS
;
1288 assert(QEMU_IS_ALIGNED(offset
, BDRV_SECTOR_SIZE
));
1289 assert(QEMU_IS_ALIGNED(bytes
, BDRV_SECTOR_SIZE
));
1290 assert(bytes
<= BDRV_REQUEST_MAX_BYTES
);
1292 assert(drv
->bdrv_co_writev
);
1293 ret
= drv
->bdrv_co_writev(bs
, sector_num
, nb_sectors
, qiov
,
1294 flags
& bs
->supported_write_flags
);
1295 flags
&= ~bs
->supported_write_flags
;
1298 if (ret
== 0 && (flags
& BDRV_REQ_FUA
)) {
1299 ret
= bdrv_co_flush(bs
);
1302 if (qiov
== &local_qiov
) {
1303 qemu_iovec_destroy(&local_qiov
);
1309 static int coroutine_fn
1310 bdrv_driver_pwritev_compressed(BlockDriverState
*bs
, int64_t offset
,
1311 int64_t bytes
, QEMUIOVector
*qiov
,
1314 BlockDriver
*drv
= bs
->drv
;
1315 QEMUIOVector local_qiov
;
1318 bdrv_check_qiov_request(offset
, bytes
, qiov
, qiov_offset
, &error_abort
);
1324 if (!block_driver_can_compress(drv
)) {
1328 if (drv
->bdrv_co_pwritev_compressed_part
) {
1329 return drv
->bdrv_co_pwritev_compressed_part(bs
, offset
, bytes
,
1333 if (qiov_offset
== 0) {
1334 return drv
->bdrv_co_pwritev_compressed(bs
, offset
, bytes
, qiov
);
1337 qemu_iovec_init_slice(&local_qiov
, qiov
, qiov_offset
, bytes
);
1338 ret
= drv
->bdrv_co_pwritev_compressed(bs
, offset
, bytes
, &local_qiov
);
1339 qemu_iovec_destroy(&local_qiov
);
1344 static int coroutine_fn
bdrv_co_do_copy_on_readv(BdrvChild
*child
,
1345 int64_t offset
, int64_t bytes
, QEMUIOVector
*qiov
,
1346 size_t qiov_offset
, int flags
)
1348 BlockDriverState
*bs
= child
->bs
;
1350 /* Perform I/O through a temporary buffer so that users who scribble over
1351 * their read buffer while the operation is in progress do not end up
1352 * modifying the image file. This is critical for zero-copy guest I/O
1353 * where anything might happen inside guest memory.
1355 void *bounce_buffer
= NULL
;
1357 BlockDriver
*drv
= bs
->drv
;
1358 int64_t cluster_offset
;
1359 int64_t cluster_bytes
;
1362 int max_transfer
= MIN_NON_ZERO(bs
->bl
.max_transfer
,
1363 BDRV_REQUEST_MAX_BYTES
);
1364 int64_t progress
= 0;
1367 bdrv_check_qiov_request(offset
, bytes
, qiov
, qiov_offset
, &error_abort
);
1374 * Do not write anything when the BDS is inactive. That is not
1375 * allowed, and it would not help.
1377 skip_write
= (bs
->open_flags
& BDRV_O_INACTIVE
);
1379 /* FIXME We cannot require callers to have write permissions when all they
1380 * are doing is a read request. If we did things right, write permissions
1381 * would be obtained anyway, but internally by the copy-on-read code. As
1382 * long as it is implemented here rather than in a separate filter driver,
1383 * the copy-on-read code doesn't have its own BdrvChild, however, for which
1384 * it could request permissions. Therefore we have to bypass the permission
1385 * system for the moment. */
1386 // assert(child->perm & (BLK_PERM_WRITE_UNCHANGED | BLK_PERM_WRITE));
1388 /* Cover entire cluster so no additional backing file I/O is required when
1389 * allocating cluster in the image file. Note that this value may exceed
1390 * BDRV_REQUEST_MAX_BYTES (even when the original read did not), which
1391 * is one reason we loop rather than doing it all at once.
1393 bdrv_round_to_clusters(bs
, offset
, bytes
, &cluster_offset
, &cluster_bytes
);
1394 skip_bytes
= offset
- cluster_offset
;
1396 trace_bdrv_co_do_copy_on_readv(bs
, offset
, bytes
,
1397 cluster_offset
, cluster_bytes
);
1399 while (cluster_bytes
) {
1403 ret
= 1; /* "already allocated", so nothing will be copied */
1404 pnum
= MIN(cluster_bytes
, max_transfer
);
1406 ret
= bdrv_is_allocated(bs
, cluster_offset
,
1407 MIN(cluster_bytes
, max_transfer
), &pnum
);
1410 * Safe to treat errors in querying allocation as if
1411 * unallocated; we'll probably fail again soon on the
1412 * read, but at least that will set a decent errno.
1414 pnum
= MIN(cluster_bytes
, max_transfer
);
1417 /* Stop at EOF if the image ends in the middle of the cluster */
1418 if (ret
== 0 && pnum
== 0) {
1419 assert(progress
>= bytes
);
1423 assert(skip_bytes
< pnum
);
1427 QEMUIOVector local_qiov
;
1429 /* Must copy-on-read; use the bounce buffer */
1430 pnum
= MIN(pnum
, MAX_BOUNCE_BUFFER
);
1431 if (!bounce_buffer
) {
1432 int64_t max_we_need
= MAX(pnum
, cluster_bytes
- pnum
);
1433 int64_t max_allowed
= MIN(max_transfer
, MAX_BOUNCE_BUFFER
);
1434 int64_t bounce_buffer_len
= MIN(max_we_need
, max_allowed
);
1436 bounce_buffer
= qemu_try_blockalign(bs
, bounce_buffer_len
);
1437 if (!bounce_buffer
) {
1442 qemu_iovec_init_buf(&local_qiov
, bounce_buffer
, pnum
);
1444 ret
= bdrv_driver_preadv(bs
, cluster_offset
, pnum
,
1450 bdrv_debug_event(bs
, BLKDBG_COR_WRITE
);
1451 if (drv
->bdrv_co_pwrite_zeroes
&&
1452 buffer_is_zero(bounce_buffer
, pnum
)) {
1453 /* FIXME: Should we (perhaps conditionally) be setting
1454 * BDRV_REQ_MAY_UNMAP, if it will allow for a sparser copy
1455 * that still correctly reads as zero? */
1456 ret
= bdrv_co_do_pwrite_zeroes(bs
, cluster_offset
, pnum
,
1457 BDRV_REQ_WRITE_UNCHANGED
);
1459 /* This does not change the data on the disk, it is not
1460 * necessary to flush even in cache=writethrough mode.
1462 ret
= bdrv_driver_pwritev(bs
, cluster_offset
, pnum
,
1464 BDRV_REQ_WRITE_UNCHANGED
);
1468 /* It might be okay to ignore write errors for guest
1469 * requests. If this is a deliberate copy-on-read
1470 * then we don't want to ignore the error. Simply
1471 * report it in all cases.
1476 if (!(flags
& BDRV_REQ_PREFETCH
)) {
1477 qemu_iovec_from_buf(qiov
, qiov_offset
+ progress
,
1478 bounce_buffer
+ skip_bytes
,
1479 MIN(pnum
- skip_bytes
, bytes
- progress
));
1481 } else if (!(flags
& BDRV_REQ_PREFETCH
)) {
1482 /* Read directly into the destination */
1483 ret
= bdrv_driver_preadv(bs
, offset
+ progress
,
1484 MIN(pnum
- skip_bytes
, bytes
- progress
),
1485 qiov
, qiov_offset
+ progress
, 0);
1491 cluster_offset
+= pnum
;
1492 cluster_bytes
-= pnum
;
1493 progress
+= pnum
- skip_bytes
;
1499 qemu_vfree(bounce_buffer
);
1504 * Forwards an already correctly aligned request to the BlockDriver. This
1505 * handles copy on read, zeroing after EOF, and fragmentation of large
1506 * reads; any other features must be implemented by the caller.
1508 static int coroutine_fn
bdrv_aligned_preadv(BdrvChild
*child
,
1509 BdrvTrackedRequest
*req
, int64_t offset
, int64_t bytes
,
1510 int64_t align
, QEMUIOVector
*qiov
, size_t qiov_offset
, int flags
)
1512 BlockDriverState
*bs
= child
->bs
;
1513 int64_t total_bytes
, max_bytes
;
1515 int64_t bytes_remaining
= bytes
;
1518 bdrv_check_qiov_request(offset
, bytes
, qiov
, qiov_offset
, &error_abort
);
1519 assert(is_power_of_2(align
));
1520 assert((offset
& (align
- 1)) == 0);
1521 assert((bytes
& (align
- 1)) == 0);
1522 assert((bs
->open_flags
& BDRV_O_NO_IO
) == 0);
1523 max_transfer
= QEMU_ALIGN_DOWN(MIN_NON_ZERO(bs
->bl
.max_transfer
, INT_MAX
),
1526 /* TODO: We would need a per-BDS .supported_read_flags and
1527 * potential fallback support, if we ever implement any read flags
1528 * to pass through to drivers. For now, there aren't any
1529 * passthrough flags. */
1530 assert(!(flags
& ~(BDRV_REQ_COPY_ON_READ
| BDRV_REQ_PREFETCH
)));
1532 /* Handle Copy on Read and associated serialisation */
1533 if (flags
& BDRV_REQ_COPY_ON_READ
) {
1534 /* If we touch the same cluster it counts as an overlap. This
1535 * guarantees that allocating writes will be serialized and not race
1536 * with each other for the same cluster. For example, in copy-on-read
1537 * it ensures that the CoR read and write operations are atomic and
1538 * guest writes cannot interleave between them. */
1539 bdrv_make_request_serialising(req
, bdrv_get_cluster_size(bs
));
1541 bdrv_wait_serialising_requests(req
);
1544 if (flags
& BDRV_REQ_COPY_ON_READ
) {
1547 /* The flag BDRV_REQ_COPY_ON_READ has reached its addressee */
1548 flags
&= ~BDRV_REQ_COPY_ON_READ
;
1550 ret
= bdrv_is_allocated(bs
, offset
, bytes
, &pnum
);
1555 if (!ret
|| pnum
!= bytes
) {
1556 ret
= bdrv_co_do_copy_on_readv(child
, offset
, bytes
,
1557 qiov
, qiov_offset
, flags
);
1559 } else if (flags
& BDRV_REQ_PREFETCH
) {
1564 /* Forward the request to the BlockDriver, possibly fragmenting it */
1565 total_bytes
= bdrv_getlength(bs
);
1566 if (total_bytes
< 0) {
1571 assert(!(flags
& ~bs
->supported_read_flags
));
1573 max_bytes
= ROUND_UP(MAX(0, total_bytes
- offset
), align
);
1574 if (bytes
<= max_bytes
&& bytes
<= max_transfer
) {
1575 ret
= bdrv_driver_preadv(bs
, offset
, bytes
, qiov
, qiov_offset
, flags
);
1579 while (bytes_remaining
) {
1583 num
= MIN(bytes_remaining
, MIN(max_bytes
, max_transfer
));
1586 ret
= bdrv_driver_preadv(bs
, offset
+ bytes
- bytes_remaining
,
1588 qiov_offset
+ bytes
- bytes_remaining
,
1592 num
= bytes_remaining
;
1593 ret
= qemu_iovec_memset(qiov
, qiov_offset
+ bytes
- bytes_remaining
,
1594 0, bytes_remaining
);
1599 bytes_remaining
-= num
;
1603 return ret
< 0 ? ret
: 0;
1609 * |<---- align ----->| |<----- align ---->|
1610 * |<- head ->|<------------- bytes ------------->|<-- tail -->|
1612 * -*----------$-------*-------- ... --------*-----$------------*---
1614 * | offset | | end |
1615 * ALIGN_DOWN(offset) ALIGN_UP(offset) ALIGN_DOWN(end) ALIGN_UP(end)
1616 * [buf ... ) [tail_buf )
1618 * @buf is an aligned allocation needed to store @head and @tail paddings. @head
1619 * is placed at the beginning of @buf and @tail at the @end.
1621 * @tail_buf is a pointer to sub-buffer, corresponding to align-sized chunk
1622 * around tail, if tail exists.
1624 * @merge_reads is true for small requests,
1625 * if @buf_len == @head + bytes + @tail. In this case it is possible that both
1626 * head and tail exist but @buf_len == align and @tail_buf == @buf.
1628 typedef struct BdrvRequestPadding
{
1635 QEMUIOVector local_qiov
;
1636 } BdrvRequestPadding
;
1638 static bool bdrv_init_padding(BlockDriverState
*bs
,
1639 int64_t offset
, int64_t bytes
,
1640 BdrvRequestPadding
*pad
)
1642 int64_t align
= bs
->bl
.request_alignment
;
1645 bdrv_check_request(offset
, bytes
, &error_abort
);
1646 assert(align
<= INT_MAX
); /* documented in block/block_int.h */
1647 assert(align
<= SIZE_MAX
/ 2); /* so we can allocate the buffer */
1649 memset(pad
, 0, sizeof(*pad
));
1651 pad
->head
= offset
& (align
- 1);
1652 pad
->tail
= ((offset
+ bytes
) & (align
- 1));
1654 pad
->tail
= align
- pad
->tail
;
1657 if (!pad
->head
&& !pad
->tail
) {
1661 assert(bytes
); /* Nothing good in aligning zero-length requests */
1663 sum
= pad
->head
+ bytes
+ pad
->tail
;
1664 pad
->buf_len
= (sum
> align
&& pad
->head
&& pad
->tail
) ? 2 * align
: align
;
1665 pad
->buf
= qemu_blockalign(bs
, pad
->buf_len
);
1666 pad
->merge_reads
= sum
== pad
->buf_len
;
1668 pad
->tail_buf
= pad
->buf
+ pad
->buf_len
- align
;
1674 static int bdrv_padding_rmw_read(BdrvChild
*child
,
1675 BdrvTrackedRequest
*req
,
1676 BdrvRequestPadding
*pad
,
1679 QEMUIOVector local_qiov
;
1680 BlockDriverState
*bs
= child
->bs
;
1681 uint64_t align
= bs
->bl
.request_alignment
;
1684 assert(req
->serialising
&& pad
->buf
);
1686 if (pad
->head
|| pad
->merge_reads
) {
1687 int64_t bytes
= pad
->merge_reads
? pad
->buf_len
: align
;
1689 qemu_iovec_init_buf(&local_qiov
, pad
->buf
, bytes
);
1692 bdrv_debug_event(bs
, BLKDBG_PWRITEV_RMW_HEAD
);
1694 if (pad
->merge_reads
&& pad
->tail
) {
1695 bdrv_debug_event(bs
, BLKDBG_PWRITEV_RMW_TAIL
);
1697 ret
= bdrv_aligned_preadv(child
, req
, req
->overlap_offset
, bytes
,
1698 align
, &local_qiov
, 0, 0);
1703 bdrv_debug_event(bs
, BLKDBG_PWRITEV_RMW_AFTER_HEAD
);
1705 if (pad
->merge_reads
&& pad
->tail
) {
1706 bdrv_debug_event(bs
, BLKDBG_PWRITEV_RMW_AFTER_TAIL
);
1709 if (pad
->merge_reads
) {
1715 qemu_iovec_init_buf(&local_qiov
, pad
->tail_buf
, align
);
1717 bdrv_debug_event(bs
, BLKDBG_PWRITEV_RMW_TAIL
);
1718 ret
= bdrv_aligned_preadv(
1720 req
->overlap_offset
+ req
->overlap_bytes
- align
,
1721 align
, align
, &local_qiov
, 0, 0);
1725 bdrv_debug_event(bs
, BLKDBG_PWRITEV_RMW_AFTER_TAIL
);
1730 memset(pad
->buf
+ pad
->head
, 0, pad
->buf_len
- pad
->head
- pad
->tail
);
1736 static void bdrv_padding_destroy(BdrvRequestPadding
*pad
)
1739 qemu_vfree(pad
->buf
);
1740 qemu_iovec_destroy(&pad
->local_qiov
);
1742 memset(pad
, 0, sizeof(*pad
));
1748 * Exchange request parameters with padded request if needed. Don't include RMW
1749 * read of padding, bdrv_padding_rmw_read() should be called separately if
1752 * Request parameters (@qiov, &qiov_offset, &offset, &bytes) are in-out:
1753 * - on function start they represent original request
1754 * - on failure or when padding is not needed they are unchanged
1755 * - on success when padding is needed they represent padded request
1757 static int bdrv_pad_request(BlockDriverState
*bs
,
1758 QEMUIOVector
**qiov
, size_t *qiov_offset
,
1759 int64_t *offset
, int64_t *bytes
,
1760 BdrvRequestPadding
*pad
, bool *padded
)
1764 bdrv_check_qiov_request(*offset
, *bytes
, *qiov
, *qiov_offset
, &error_abort
);
1766 if (!bdrv_init_padding(bs
, *offset
, *bytes
, pad
)) {
1773 ret
= qemu_iovec_init_extended(&pad
->local_qiov
, pad
->buf
, pad
->head
,
1774 *qiov
, *qiov_offset
, *bytes
,
1775 pad
->buf
+ pad
->buf_len
- pad
->tail
,
1778 bdrv_padding_destroy(pad
);
1781 *bytes
+= pad
->head
+ pad
->tail
;
1782 *offset
-= pad
->head
;
1783 *qiov
= &pad
->local_qiov
;
1792 int coroutine_fn
bdrv_co_preadv(BdrvChild
*child
,
1793 int64_t offset
, int64_t bytes
, QEMUIOVector
*qiov
,
1794 BdrvRequestFlags flags
)
1796 return bdrv_co_preadv_part(child
, offset
, bytes
, qiov
, 0, flags
);
1799 int coroutine_fn
bdrv_co_preadv_part(BdrvChild
*child
,
1800 int64_t offset
, int64_t bytes
,
1801 QEMUIOVector
*qiov
, size_t qiov_offset
,
1802 BdrvRequestFlags flags
)
1804 BlockDriverState
*bs
= child
->bs
;
1805 BdrvTrackedRequest req
;
1806 BdrvRequestPadding pad
;
1809 trace_bdrv_co_preadv_part(bs
, offset
, bytes
, flags
);
1811 if (!bdrv_is_inserted(bs
)) {
1815 ret
= bdrv_check_request32(offset
, bytes
, qiov
, qiov_offset
);
1820 if (bytes
== 0 && !QEMU_IS_ALIGNED(offset
, bs
->bl
.request_alignment
)) {
1822 * Aligning zero request is nonsense. Even if driver has special meaning
1823 * of zero-length (like qcow2_co_pwritev_compressed_part), we can't pass
1824 * it to driver due to request_alignment.
1826 * Still, no reason to return an error if someone do unaligned
1827 * zero-length read occasionally.
1832 bdrv_inc_in_flight(bs
);
1834 /* Don't do copy-on-read if we read data before write operation */
1835 if (qatomic_read(&bs
->copy_on_read
)) {
1836 flags
|= BDRV_REQ_COPY_ON_READ
;
1839 ret
= bdrv_pad_request(bs
, &qiov
, &qiov_offset
, &offset
, &bytes
, &pad
,
1845 tracked_request_begin(&req
, bs
, offset
, bytes
, BDRV_TRACKED_READ
);
1846 ret
= bdrv_aligned_preadv(child
, &req
, offset
, bytes
,
1847 bs
->bl
.request_alignment
,
1848 qiov
, qiov_offset
, flags
);
1849 tracked_request_end(&req
);
1850 bdrv_dec_in_flight(bs
);
1852 bdrv_padding_destroy(&pad
);
1857 static int coroutine_fn
bdrv_co_do_pwrite_zeroes(BlockDriverState
*bs
,
1858 int64_t offset
, int64_t bytes
, BdrvRequestFlags flags
)
1860 BlockDriver
*drv
= bs
->drv
;
1864 bool need_flush
= false;
1868 int max_write_zeroes
= MIN_NON_ZERO(bs
->bl
.max_pwrite_zeroes
, INT_MAX
);
1869 int alignment
= MAX(bs
->bl
.pwrite_zeroes_alignment
,
1870 bs
->bl
.request_alignment
);
1871 int max_transfer
= MIN_NON_ZERO(bs
->bl
.max_transfer
, MAX_BOUNCE_BUFFER
);
1873 bdrv_check_request(offset
, bytes
, &error_abort
);
1879 if ((flags
& ~bs
->supported_zero_flags
) & BDRV_REQ_NO_FALLBACK
) {
1883 assert(alignment
% bs
->bl
.request_alignment
== 0);
1884 head
= offset
% alignment
;
1885 tail
= (offset
+ bytes
) % alignment
;
1886 max_write_zeroes
= QEMU_ALIGN_DOWN(max_write_zeroes
, alignment
);
1887 assert(max_write_zeroes
>= bs
->bl
.request_alignment
);
1889 while (bytes
> 0 && !ret
) {
1890 int64_t num
= bytes
;
1892 /* Align request. Block drivers can expect the "bulk" of the request
1893 * to be aligned, and that unaligned requests do not cross cluster
1897 /* Make a small request up to the first aligned sector. For
1898 * convenience, limit this request to max_transfer even if
1899 * we don't need to fall back to writes. */
1900 num
= MIN(MIN(bytes
, max_transfer
), alignment
- head
);
1901 head
= (head
+ num
) % alignment
;
1902 assert(num
< max_write_zeroes
);
1903 } else if (tail
&& num
> alignment
) {
1904 /* Shorten the request to the last aligned sector. */
1908 /* limit request size */
1909 if (num
> max_write_zeroes
) {
1910 num
= max_write_zeroes
;
1914 /* First try the efficient write zeroes operation */
1915 if (drv
->bdrv_co_pwrite_zeroes
) {
1916 ret
= drv
->bdrv_co_pwrite_zeroes(bs
, offset
, num
,
1917 flags
& bs
->supported_zero_flags
);
1918 if (ret
!= -ENOTSUP
&& (flags
& BDRV_REQ_FUA
) &&
1919 !(bs
->supported_zero_flags
& BDRV_REQ_FUA
)) {
1923 assert(!bs
->supported_zero_flags
);
1926 if (ret
== -ENOTSUP
&& !(flags
& BDRV_REQ_NO_FALLBACK
)) {
1927 /* Fall back to bounce buffer if write zeroes is unsupported */
1928 BdrvRequestFlags write_flags
= flags
& ~BDRV_REQ_ZERO_WRITE
;
1930 if ((flags
& BDRV_REQ_FUA
) &&
1931 !(bs
->supported_write_flags
& BDRV_REQ_FUA
)) {
1932 /* No need for bdrv_driver_pwrite() to do a fallback
1933 * flush on each chunk; use just one at the end */
1934 write_flags
&= ~BDRV_REQ_FUA
;
1937 num
= MIN(num
, max_transfer
);
1939 buf
= qemu_try_blockalign0(bs
, num
);
1945 qemu_iovec_init_buf(&qiov
, buf
, num
);
1947 ret
= bdrv_driver_pwritev(bs
, offset
, num
, &qiov
, 0, write_flags
);
1949 /* Keep bounce buffer around if it is big enough for all
1950 * all future requests.
1952 if (num
< max_transfer
) {
1963 if (ret
== 0 && need_flush
) {
1964 ret
= bdrv_co_flush(bs
);
1970 static inline int coroutine_fn
1971 bdrv_co_write_req_prepare(BdrvChild
*child
, int64_t offset
, int64_t bytes
,
1972 BdrvTrackedRequest
*req
, int flags
)
1974 BlockDriverState
*bs
= child
->bs
;
1976 bdrv_check_request(offset
, bytes
, &error_abort
);
1978 if (bdrv_is_read_only(bs
)) {
1982 assert(!(bs
->open_flags
& BDRV_O_INACTIVE
));
1983 assert((bs
->open_flags
& BDRV_O_NO_IO
) == 0);
1984 assert(!(flags
& ~BDRV_REQ_MASK
));
1985 assert(!((flags
& BDRV_REQ_NO_WAIT
) && !(flags
& BDRV_REQ_SERIALISING
)));
1987 if (flags
& BDRV_REQ_SERIALISING
) {
1988 QEMU_LOCK_GUARD(&bs
->reqs_lock
);
1990 tracked_request_set_serialising(req
, bdrv_get_cluster_size(bs
));
1992 if ((flags
& BDRV_REQ_NO_WAIT
) && bdrv_find_conflicting_request(req
)) {
1996 bdrv_wait_serialising_requests_locked(req
);
1998 bdrv_wait_serialising_requests(req
);
2001 assert(req
->overlap_offset
<= offset
);
2002 assert(offset
+ bytes
<= req
->overlap_offset
+ req
->overlap_bytes
);
2003 assert(offset
+ bytes
<= bs
->total_sectors
* BDRV_SECTOR_SIZE
||
2004 child
->perm
& BLK_PERM_RESIZE
);
2006 switch (req
->type
) {
2007 case BDRV_TRACKED_WRITE
:
2008 case BDRV_TRACKED_DISCARD
:
2009 if (flags
& BDRV_REQ_WRITE_UNCHANGED
) {
2010 assert(child
->perm
& (BLK_PERM_WRITE_UNCHANGED
| BLK_PERM_WRITE
));
2012 assert(child
->perm
& BLK_PERM_WRITE
);
2014 bdrv_write_threshold_check_write(bs
, offset
, bytes
);
2016 case BDRV_TRACKED_TRUNCATE
:
2017 assert(child
->perm
& BLK_PERM_RESIZE
);
2024 static inline void coroutine_fn
2025 bdrv_co_write_req_finish(BdrvChild
*child
, int64_t offset
, int64_t bytes
,
2026 BdrvTrackedRequest
*req
, int ret
)
2028 int64_t end_sector
= DIV_ROUND_UP(offset
+ bytes
, BDRV_SECTOR_SIZE
);
2029 BlockDriverState
*bs
= child
->bs
;
2031 bdrv_check_request(offset
, bytes
, &error_abort
);
2033 qatomic_inc(&bs
->write_gen
);
2036 * Discard cannot extend the image, but in error handling cases, such as
2037 * when reverting a qcow2 cluster allocation, the discarded range can pass
2038 * the end of image file, so we cannot assert about BDRV_TRACKED_DISCARD
2039 * here. Instead, just skip it, since semantically a discard request
2040 * beyond EOF cannot expand the image anyway.
2043 (req
->type
== BDRV_TRACKED_TRUNCATE
||
2044 end_sector
> bs
->total_sectors
) &&
2045 req
->type
!= BDRV_TRACKED_DISCARD
) {
2046 bs
->total_sectors
= end_sector
;
2047 bdrv_parent_cb_resize(bs
);
2048 bdrv_dirty_bitmap_truncate(bs
, end_sector
<< BDRV_SECTOR_BITS
);
2051 switch (req
->type
) {
2052 case BDRV_TRACKED_WRITE
:
2053 stat64_max(&bs
->wr_highest_offset
, offset
+ bytes
);
2054 /* fall through, to set dirty bits */
2055 case BDRV_TRACKED_DISCARD
:
2056 bdrv_set_dirty(bs
, offset
, bytes
);
2065 * Forwards an already correctly aligned write request to the BlockDriver,
2066 * after possibly fragmenting it.
2068 static int coroutine_fn
bdrv_aligned_pwritev(BdrvChild
*child
,
2069 BdrvTrackedRequest
*req
, int64_t offset
, int64_t bytes
,
2070 int64_t align
, QEMUIOVector
*qiov
, size_t qiov_offset
, int flags
)
2072 BlockDriverState
*bs
= child
->bs
;
2073 BlockDriver
*drv
= bs
->drv
;
2076 int64_t bytes_remaining
= bytes
;
2079 bdrv_check_qiov_request(offset
, bytes
, qiov
, qiov_offset
, &error_abort
);
2085 if (bdrv_has_readonly_bitmaps(bs
)) {
2089 assert(is_power_of_2(align
));
2090 assert((offset
& (align
- 1)) == 0);
2091 assert((bytes
& (align
- 1)) == 0);
2092 max_transfer
= QEMU_ALIGN_DOWN(MIN_NON_ZERO(bs
->bl
.max_transfer
, INT_MAX
),
2095 ret
= bdrv_co_write_req_prepare(child
, offset
, bytes
, req
, flags
);
2097 if (!ret
&& bs
->detect_zeroes
!= BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF
&&
2098 !(flags
& BDRV_REQ_ZERO_WRITE
) && drv
->bdrv_co_pwrite_zeroes
&&
2099 qemu_iovec_is_zero(qiov
, qiov_offset
, bytes
)) {
2100 flags
|= BDRV_REQ_ZERO_WRITE
;
2101 if (bs
->detect_zeroes
== BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP
) {
2102 flags
|= BDRV_REQ_MAY_UNMAP
;
2107 /* Do nothing, write notifier decided to fail this request */
2108 } else if (flags
& BDRV_REQ_ZERO_WRITE
) {
2109 bdrv_debug_event(bs
, BLKDBG_PWRITEV_ZERO
);
2110 ret
= bdrv_co_do_pwrite_zeroes(bs
, offset
, bytes
, flags
);
2111 } else if (flags
& BDRV_REQ_WRITE_COMPRESSED
) {
2112 ret
= bdrv_driver_pwritev_compressed(bs
, offset
, bytes
,
2114 } else if (bytes
<= max_transfer
) {
2115 bdrv_debug_event(bs
, BLKDBG_PWRITEV
);
2116 ret
= bdrv_driver_pwritev(bs
, offset
, bytes
, qiov
, qiov_offset
, flags
);
2118 bdrv_debug_event(bs
, BLKDBG_PWRITEV
);
2119 while (bytes_remaining
) {
2120 int num
= MIN(bytes_remaining
, max_transfer
);
2121 int local_flags
= flags
;
2124 if (num
< bytes_remaining
&& (flags
& BDRV_REQ_FUA
) &&
2125 !(bs
->supported_write_flags
& BDRV_REQ_FUA
)) {
2126 /* If FUA is going to be emulated by flush, we only
2127 * need to flush on the last iteration */
2128 local_flags
&= ~BDRV_REQ_FUA
;
2131 ret
= bdrv_driver_pwritev(bs
, offset
+ bytes
- bytes_remaining
,
2133 qiov_offset
+ bytes
- bytes_remaining
,
2138 bytes_remaining
-= num
;
2141 bdrv_debug_event(bs
, BLKDBG_PWRITEV_DONE
);
2146 bdrv_co_write_req_finish(child
, offset
, bytes
, req
, ret
);
2151 static int coroutine_fn
bdrv_co_do_zero_pwritev(BdrvChild
*child
,
2154 BdrvRequestFlags flags
,
2155 BdrvTrackedRequest
*req
)
2157 BlockDriverState
*bs
= child
->bs
;
2158 QEMUIOVector local_qiov
;
2159 uint64_t align
= bs
->bl
.request_alignment
;
2162 BdrvRequestPadding pad
;
2164 padding
= bdrv_init_padding(bs
, offset
, bytes
, &pad
);
2166 bdrv_make_request_serialising(req
, align
);
2168 bdrv_padding_rmw_read(child
, req
, &pad
, true);
2170 if (pad
.head
|| pad
.merge_reads
) {
2171 int64_t aligned_offset
= offset
& ~(align
- 1);
2172 int64_t write_bytes
= pad
.merge_reads
? pad
.buf_len
: align
;
2174 qemu_iovec_init_buf(&local_qiov
, pad
.buf
, write_bytes
);
2175 ret
= bdrv_aligned_pwritev(child
, req
, aligned_offset
, write_bytes
,
2176 align
, &local_qiov
, 0,
2177 flags
& ~BDRV_REQ_ZERO_WRITE
);
2178 if (ret
< 0 || pad
.merge_reads
) {
2179 /* Error or all work is done */
2182 offset
+= write_bytes
- pad
.head
;
2183 bytes
-= write_bytes
- pad
.head
;
2187 assert(!bytes
|| (offset
& (align
- 1)) == 0);
2188 if (bytes
>= align
) {
2189 /* Write the aligned part in the middle. */
2190 int64_t aligned_bytes
= bytes
& ~(align
- 1);
2191 ret
= bdrv_aligned_pwritev(child
, req
, offset
, aligned_bytes
, align
,
2196 bytes
-= aligned_bytes
;
2197 offset
+= aligned_bytes
;
2200 assert(!bytes
|| (offset
& (align
- 1)) == 0);
2202 assert(align
== pad
.tail
+ bytes
);
2204 qemu_iovec_init_buf(&local_qiov
, pad
.tail_buf
, align
);
2205 ret
= bdrv_aligned_pwritev(child
, req
, offset
, align
, align
,
2207 flags
& ~BDRV_REQ_ZERO_WRITE
);
2211 bdrv_padding_destroy(&pad
);
2217 * Handle a write request in coroutine context
2219 int coroutine_fn
bdrv_co_pwritev(BdrvChild
*child
,
2220 int64_t offset
, int64_t bytes
, QEMUIOVector
*qiov
,
2221 BdrvRequestFlags flags
)
2223 return bdrv_co_pwritev_part(child
, offset
, bytes
, qiov
, 0, flags
);
2226 int coroutine_fn
bdrv_co_pwritev_part(BdrvChild
*child
,
2227 int64_t offset
, int64_t bytes
, QEMUIOVector
*qiov
, size_t qiov_offset
,
2228 BdrvRequestFlags flags
)
2230 BlockDriverState
*bs
= child
->bs
;
2231 BdrvTrackedRequest req
;
2232 uint64_t align
= bs
->bl
.request_alignment
;
2233 BdrvRequestPadding pad
;
2235 bool padded
= false;
2237 trace_bdrv_co_pwritev_part(child
->bs
, offset
, bytes
, flags
);
2239 if (!bdrv_is_inserted(bs
)) {
2243 ret
= bdrv_check_request32(offset
, bytes
, qiov
, qiov_offset
);
2248 /* If the request is misaligned then we can't make it efficient */
2249 if ((flags
& BDRV_REQ_NO_FALLBACK
) &&
2250 !QEMU_IS_ALIGNED(offset
| bytes
, align
))
2255 if (bytes
== 0 && !QEMU_IS_ALIGNED(offset
, bs
->bl
.request_alignment
)) {
2257 * Aligning zero request is nonsense. Even if driver has special meaning
2258 * of zero-length (like qcow2_co_pwritev_compressed_part), we can't pass
2259 * it to driver due to request_alignment.
2261 * Still, no reason to return an error if someone do unaligned
2262 * zero-length write occasionally.
2267 if (!(flags
& BDRV_REQ_ZERO_WRITE
)) {
2269 * Pad request for following read-modify-write cycle.
2270 * bdrv_co_do_zero_pwritev() does aligning by itself, so, we do
2271 * alignment only if there is no ZERO flag.
2273 ret
= bdrv_pad_request(bs
, &qiov
, &qiov_offset
, &offset
, &bytes
, &pad
,
2280 bdrv_inc_in_flight(bs
);
2281 tracked_request_begin(&req
, bs
, offset
, bytes
, BDRV_TRACKED_WRITE
);
2283 if (flags
& BDRV_REQ_ZERO_WRITE
) {
2285 ret
= bdrv_co_do_zero_pwritev(child
, offset
, bytes
, flags
, &req
);
2291 * Request was unaligned to request_alignment and therefore
2292 * padded. We are going to do read-modify-write, and must
2293 * serialize the request to prevent interactions of the
2294 * widened region with other transactions.
2296 bdrv_make_request_serialising(&req
, align
);
2297 bdrv_padding_rmw_read(child
, &req
, &pad
, false);
2300 ret
= bdrv_aligned_pwritev(child
, &req
, offset
, bytes
, align
,
2301 qiov
, qiov_offset
, flags
);
2303 bdrv_padding_destroy(&pad
);
2306 tracked_request_end(&req
);
2307 bdrv_dec_in_flight(bs
);
2312 int coroutine_fn
bdrv_co_pwrite_zeroes(BdrvChild
*child
, int64_t offset
,
2313 int64_t bytes
, BdrvRequestFlags flags
)
2315 trace_bdrv_co_pwrite_zeroes(child
->bs
, offset
, bytes
, flags
);
2317 if (!(child
->bs
->open_flags
& BDRV_O_UNMAP
)) {
2318 flags
&= ~BDRV_REQ_MAY_UNMAP
;
2321 return bdrv_co_pwritev(child
, offset
, bytes
, NULL
,
2322 BDRV_REQ_ZERO_WRITE
| flags
);
2326 * Flush ALL BDSes regardless of if they are reachable via a BlkBackend or not.
2328 int bdrv_flush_all(void)
2330 BdrvNextIterator it
;
2331 BlockDriverState
*bs
= NULL
;
2335 * bdrv queue is managed by record/replay,
2336 * creating new flush request for stopping
2337 * the VM may break the determinism
2339 if (replay_events_enabled()) {
2343 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
2344 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
2347 aio_context_acquire(aio_context
);
2348 ret
= bdrv_flush(bs
);
2349 if (ret
< 0 && !result
) {
2352 aio_context_release(aio_context
);
2359 * Returns the allocation status of the specified sectors.
2360 * Drivers not implementing the functionality are assumed to not support
2361 * backing files, hence all their sectors are reported as allocated.
2363 * If 'want_zero' is true, the caller is querying for mapping
2364 * purposes, with a focus on valid BDRV_BLOCK_OFFSET_VALID, _DATA, and
2365 * _ZERO where possible; otherwise, the result favors larger 'pnum',
2366 * with a focus on accurate BDRV_BLOCK_ALLOCATED.
2368 * If 'offset' is beyond the end of the disk image the return value is
2369 * BDRV_BLOCK_EOF and 'pnum' is set to 0.
2371 * 'bytes' is the max value 'pnum' should be set to. If bytes goes
2372 * beyond the end of the disk image it will be clamped; if 'pnum' is set to
2373 * the end of the image, then the returned value will include BDRV_BLOCK_EOF.
2375 * 'pnum' is set to the number of bytes (including and immediately
2376 * following the specified offset) that are easily known to be in the
2377 * same allocated/unallocated state. Note that a second call starting
2378 * at the original offset plus returned pnum may have the same status.
2379 * The returned value is non-zero on success except at end-of-file.
2381 * Returns negative errno on failure. Otherwise, if the
2382 * BDRV_BLOCK_OFFSET_VALID bit is set, 'map' and 'file' (if non-NULL) are
2383 * set to the host mapping and BDS corresponding to the guest offset.
2385 static int coroutine_fn
bdrv_co_block_status(BlockDriverState
*bs
,
2387 int64_t offset
, int64_t bytes
,
2388 int64_t *pnum
, int64_t *map
,
2389 BlockDriverState
**file
)
2392 int64_t n
; /* bytes */
2394 int64_t local_map
= 0;
2395 BlockDriverState
*local_file
= NULL
;
2396 int64_t aligned_offset
, aligned_bytes
;
2398 bool has_filtered_child
;
2402 total_size
= bdrv_getlength(bs
);
2403 if (total_size
< 0) {
2408 if (offset
>= total_size
) {
2409 ret
= BDRV_BLOCK_EOF
;
2417 n
= total_size
- offset
;
2422 /* Must be non-NULL or bdrv_getlength() would have failed */
2424 has_filtered_child
= bdrv_filter_child(bs
);
2425 if (!bs
->drv
->bdrv_co_block_status
&& !has_filtered_child
) {
2427 ret
= BDRV_BLOCK_DATA
| BDRV_BLOCK_ALLOCATED
;
2428 if (offset
+ bytes
== total_size
) {
2429 ret
|= BDRV_BLOCK_EOF
;
2431 if (bs
->drv
->protocol_name
) {
2432 ret
|= BDRV_BLOCK_OFFSET_VALID
;
2439 bdrv_inc_in_flight(bs
);
2441 /* Round out to request_alignment boundaries */
2442 align
= bs
->bl
.request_alignment
;
2443 aligned_offset
= QEMU_ALIGN_DOWN(offset
, align
);
2444 aligned_bytes
= ROUND_UP(offset
+ bytes
, align
) - aligned_offset
;
2446 if (bs
->drv
->bdrv_co_block_status
) {
2447 ret
= bs
->drv
->bdrv_co_block_status(bs
, want_zero
, aligned_offset
,
2448 aligned_bytes
, pnum
, &local_map
,
2451 /* Default code for filters */
2453 local_file
= bdrv_filter_bs(bs
);
2456 *pnum
= aligned_bytes
;
2457 local_map
= aligned_offset
;
2458 ret
= BDRV_BLOCK_RAW
| BDRV_BLOCK_OFFSET_VALID
;
2466 * The driver's result must be a non-zero multiple of request_alignment.
2467 * Clamp pnum and adjust map to original request.
2469 assert(*pnum
&& QEMU_IS_ALIGNED(*pnum
, align
) &&
2470 align
> offset
- aligned_offset
);
2471 if (ret
& BDRV_BLOCK_RECURSE
) {
2472 assert(ret
& BDRV_BLOCK_DATA
);
2473 assert(ret
& BDRV_BLOCK_OFFSET_VALID
);
2474 assert(!(ret
& BDRV_BLOCK_ZERO
));
2477 *pnum
-= offset
- aligned_offset
;
2478 if (*pnum
> bytes
) {
2481 if (ret
& BDRV_BLOCK_OFFSET_VALID
) {
2482 local_map
+= offset
- aligned_offset
;
2485 if (ret
& BDRV_BLOCK_RAW
) {
2486 assert(ret
& BDRV_BLOCK_OFFSET_VALID
&& local_file
);
2487 ret
= bdrv_co_block_status(local_file
, want_zero
, local_map
,
2488 *pnum
, pnum
, &local_map
, &local_file
);
2492 if (ret
& (BDRV_BLOCK_DATA
| BDRV_BLOCK_ZERO
)) {
2493 ret
|= BDRV_BLOCK_ALLOCATED
;
2494 } else if (bs
->drv
->supports_backing
) {
2495 BlockDriverState
*cow_bs
= bdrv_cow_bs(bs
);
2498 ret
|= BDRV_BLOCK_ZERO
;
2499 } else if (want_zero
) {
2500 int64_t size2
= bdrv_getlength(cow_bs
);
2502 if (size2
>= 0 && offset
>= size2
) {
2503 ret
|= BDRV_BLOCK_ZERO
;
2508 if (want_zero
&& ret
& BDRV_BLOCK_RECURSE
&&
2509 local_file
&& local_file
!= bs
&&
2510 (ret
& BDRV_BLOCK_DATA
) && !(ret
& BDRV_BLOCK_ZERO
) &&
2511 (ret
& BDRV_BLOCK_OFFSET_VALID
)) {
2515 ret2
= bdrv_co_block_status(local_file
, want_zero
, local_map
,
2516 *pnum
, &file_pnum
, NULL
, NULL
);
2518 /* Ignore errors. This is just providing extra information, it
2519 * is useful but not necessary.
2521 if (ret2
& BDRV_BLOCK_EOF
&&
2522 (!file_pnum
|| ret2
& BDRV_BLOCK_ZERO
)) {
2524 * It is valid for the format block driver to read
2525 * beyond the end of the underlying file's current
2526 * size; such areas read as zero.
2528 ret
|= BDRV_BLOCK_ZERO
;
2530 /* Limit request to the range reported by the protocol driver */
2532 ret
|= (ret2
& BDRV_BLOCK_ZERO
);
2538 bdrv_dec_in_flight(bs
);
2539 if (ret
>= 0 && offset
+ *pnum
== total_size
) {
2540 ret
|= BDRV_BLOCK_EOF
;
2553 bdrv_co_common_block_status_above(BlockDriverState
*bs
,
2554 BlockDriverState
*base
,
2561 BlockDriverState
**file
,
2565 BlockDriverState
*p
;
2569 assert(!include_base
|| base
); /* Can't include NULL base */
2576 if (!include_base
&& bs
== base
) {
2581 ret
= bdrv_co_block_status(bs
, want_zero
, offset
, bytes
, pnum
, map
, file
);
2583 if (ret
< 0 || *pnum
== 0 || ret
& BDRV_BLOCK_ALLOCATED
|| bs
== base
) {
2587 if (ret
& BDRV_BLOCK_EOF
) {
2588 eof
= offset
+ *pnum
;
2591 assert(*pnum
<= bytes
);
2594 for (p
= bdrv_filter_or_cow_bs(bs
); include_base
|| p
!= base
;
2595 p
= bdrv_filter_or_cow_bs(p
))
2597 ret
= bdrv_co_block_status(p
, want_zero
, offset
, bytes
, pnum
, map
,
2605 * The top layer deferred to this layer, and because this layer is
2606 * short, any zeroes that we synthesize beyond EOF behave as if they
2607 * were allocated at this layer.
2609 * We don't include BDRV_BLOCK_EOF into ret, as upper layer may be
2610 * larger. We'll add BDRV_BLOCK_EOF if needed at function end, see
2613 assert(ret
& BDRV_BLOCK_EOF
);
2618 ret
= BDRV_BLOCK_ZERO
| BDRV_BLOCK_ALLOCATED
;
2621 if (ret
& BDRV_BLOCK_ALLOCATED
) {
2623 * We've found the node and the status, we must break.
2625 * Drop BDRV_BLOCK_EOF, as it's not for upper layer, which may be
2626 * larger. We'll add BDRV_BLOCK_EOF if needed at function end, see
2629 ret
&= ~BDRV_BLOCK_EOF
;
2634 assert(include_base
);
2639 * OK, [offset, offset + *pnum) region is unallocated on this layer,
2640 * let's continue the diving.
2642 assert(*pnum
<= bytes
);
2646 if (offset
+ *pnum
== eof
) {
2647 ret
|= BDRV_BLOCK_EOF
;
2653 int bdrv_block_status_above(BlockDriverState
*bs
, BlockDriverState
*base
,
2654 int64_t offset
, int64_t bytes
, int64_t *pnum
,
2655 int64_t *map
, BlockDriverState
**file
)
2657 return bdrv_common_block_status_above(bs
, base
, false, true, offset
, bytes
,
2658 pnum
, map
, file
, NULL
);
2661 int bdrv_block_status(BlockDriverState
*bs
, int64_t offset
, int64_t bytes
,
2662 int64_t *pnum
, int64_t *map
, BlockDriverState
**file
)
2664 return bdrv_block_status_above(bs
, bdrv_filter_or_cow_bs(bs
),
2665 offset
, bytes
, pnum
, map
, file
);
2669 * Check @bs (and its backing chain) to see if the range defined
2670 * by @offset and @bytes is known to read as zeroes.
2671 * Return 1 if that is the case, 0 otherwise and -errno on error.
2672 * This test is meant to be fast rather than accurate so returning 0
2673 * does not guarantee non-zero data.
2675 int coroutine_fn
bdrv_co_is_zero_fast(BlockDriverState
*bs
, int64_t offset
,
2679 int64_t pnum
= bytes
;
2685 ret
= bdrv_common_block_status_above(bs
, NULL
, false, false, offset
,
2686 bytes
, &pnum
, NULL
, NULL
, NULL
);
2692 return (pnum
== bytes
) && (ret
& BDRV_BLOCK_ZERO
);
2695 int coroutine_fn
bdrv_is_allocated(BlockDriverState
*bs
, int64_t offset
,
2696 int64_t bytes
, int64_t *pnum
)
2701 ret
= bdrv_common_block_status_above(bs
, bs
, true, false, offset
,
2702 bytes
, pnum
? pnum
: &dummy
, NULL
,
2707 return !!(ret
& BDRV_BLOCK_ALLOCATED
);
2711 * Given an image chain: ... -> [BASE] -> [INTER1] -> [INTER2] -> [TOP]
2713 * Return a positive depth if (a prefix of) the given range is allocated
2714 * in any image between BASE and TOP (BASE is only included if include_base
2715 * is set). Depth 1 is TOP, 2 is the first backing layer, and so forth.
2716 * BASE can be NULL to check if the given offset is allocated in any
2717 * image of the chain. Return 0 otherwise, or negative errno on
2720 * 'pnum' is set to the number of bytes (including and immediately
2721 * following the specified offset) that are known to be in the same
2722 * allocated/unallocated state. Note that a subsequent call starting
2723 * at 'offset + *pnum' may return the same allocation status (in other
2724 * words, the result is not necessarily the maximum possible range);
2725 * but 'pnum' will only be 0 when end of file is reached.
2727 int bdrv_is_allocated_above(BlockDriverState
*top
,
2728 BlockDriverState
*base
,
2729 bool include_base
, int64_t offset
,
2730 int64_t bytes
, int64_t *pnum
)
2733 int ret
= bdrv_common_block_status_above(top
, base
, include_base
, false,
2734 offset
, bytes
, pnum
, NULL
, NULL
,
2740 if (ret
& BDRV_BLOCK_ALLOCATED
) {
2747 bdrv_co_readv_vmstate(BlockDriverState
*bs
, QEMUIOVector
*qiov
, int64_t pos
)
2749 BlockDriver
*drv
= bs
->drv
;
2750 BlockDriverState
*child_bs
= bdrv_primary_bs(bs
);
2757 bdrv_inc_in_flight(bs
);
2759 if (drv
->bdrv_load_vmstate
) {
2760 ret
= drv
->bdrv_load_vmstate(bs
, qiov
, pos
);
2761 } else if (child_bs
) {
2762 ret
= bdrv_co_readv_vmstate(child_bs
, qiov
, pos
);
2765 bdrv_dec_in_flight(bs
);
2771 bdrv_co_writev_vmstate(BlockDriverState
*bs
, QEMUIOVector
*qiov
, int64_t pos
)
2773 BlockDriver
*drv
= bs
->drv
;
2774 BlockDriverState
*child_bs
= bdrv_primary_bs(bs
);
2781 bdrv_inc_in_flight(bs
);
2783 if (drv
->bdrv_save_vmstate
) {
2784 ret
= drv
->bdrv_save_vmstate(bs
, qiov
, pos
);
2785 } else if (child_bs
) {
2786 ret
= bdrv_co_writev_vmstate(child_bs
, qiov
, pos
);
2789 bdrv_dec_in_flight(bs
);
2794 int bdrv_save_vmstate(BlockDriverState
*bs
, const uint8_t *buf
,
2795 int64_t pos
, int size
)
2797 QEMUIOVector qiov
= QEMU_IOVEC_INIT_BUF(qiov
, buf
, size
);
2798 int ret
= bdrv_writev_vmstate(bs
, &qiov
, pos
);
2800 return ret
< 0 ? ret
: size
;
2803 int bdrv_load_vmstate(BlockDriverState
*bs
, uint8_t *buf
,
2804 int64_t pos
, int size
)
2806 QEMUIOVector qiov
= QEMU_IOVEC_INIT_BUF(qiov
, buf
, size
);
2807 int ret
= bdrv_readv_vmstate(bs
, &qiov
, pos
);
2809 return ret
< 0 ? ret
: size
;
2812 /**************************************************************/
2815 void bdrv_aio_cancel(BlockAIOCB
*acb
)
2818 bdrv_aio_cancel_async(acb
);
2819 while (acb
->refcnt
> 1) {
2820 if (acb
->aiocb_info
->get_aio_context
) {
2821 aio_poll(acb
->aiocb_info
->get_aio_context(acb
), true);
2822 } else if (acb
->bs
) {
2823 /* qemu_aio_ref and qemu_aio_unref are not thread-safe, so
2824 * assert that we're not using an I/O thread. Thread-safe
2825 * code should use bdrv_aio_cancel_async exclusively.
2827 assert(bdrv_get_aio_context(acb
->bs
) == qemu_get_aio_context());
2828 aio_poll(bdrv_get_aio_context(acb
->bs
), true);
2833 qemu_aio_unref(acb
);
2836 /* Async version of aio cancel. The caller is not blocked if the acb implements
2837 * cancel_async, otherwise we do nothing and let the request normally complete.
2838 * In either case the completion callback must be called. */
2839 void bdrv_aio_cancel_async(BlockAIOCB
*acb
)
2841 if (acb
->aiocb_info
->cancel_async
) {
2842 acb
->aiocb_info
->cancel_async(acb
);
2846 /**************************************************************/
2847 /* Coroutine block device emulation */
2849 int coroutine_fn
bdrv_co_flush(BlockDriverState
*bs
)
2851 BdrvChild
*primary_child
= bdrv_primary_child(bs
);
2856 bdrv_inc_in_flight(bs
);
2858 if (!bdrv_is_inserted(bs
) || bdrv_is_read_only(bs
) ||
2863 qemu_co_mutex_lock(&bs
->reqs_lock
);
2864 current_gen
= qatomic_read(&bs
->write_gen
);
2866 /* Wait until any previous flushes are completed */
2867 while (bs
->active_flush_req
) {
2868 qemu_co_queue_wait(&bs
->flush_queue
, &bs
->reqs_lock
);
2871 /* Flushes reach this point in nondecreasing current_gen order. */
2872 bs
->active_flush_req
= true;
2873 qemu_co_mutex_unlock(&bs
->reqs_lock
);
2875 /* Write back all layers by calling one driver function */
2876 if (bs
->drv
->bdrv_co_flush
) {
2877 ret
= bs
->drv
->bdrv_co_flush(bs
);
2881 /* Write back cached data to the OS even with cache=unsafe */
2882 BLKDBG_EVENT(primary_child
, BLKDBG_FLUSH_TO_OS
);
2883 if (bs
->drv
->bdrv_co_flush_to_os
) {
2884 ret
= bs
->drv
->bdrv_co_flush_to_os(bs
);
2890 /* But don't actually force it to the disk with cache=unsafe */
2891 if (bs
->open_flags
& BDRV_O_NO_FLUSH
) {
2892 goto flush_children
;
2895 /* Check if we really need to flush anything */
2896 if (bs
->flushed_gen
== current_gen
) {
2897 goto flush_children
;
2900 BLKDBG_EVENT(primary_child
, BLKDBG_FLUSH_TO_DISK
);
2902 /* bs->drv->bdrv_co_flush() might have ejected the BDS
2903 * (even in case of apparent success) */
2907 if (bs
->drv
->bdrv_co_flush_to_disk
) {
2908 ret
= bs
->drv
->bdrv_co_flush_to_disk(bs
);
2909 } else if (bs
->drv
->bdrv_aio_flush
) {
2911 CoroutineIOCompletion co
= {
2912 .coroutine
= qemu_coroutine_self(),
2915 acb
= bs
->drv
->bdrv_aio_flush(bs
, bdrv_co_io_em_complete
, &co
);
2919 qemu_coroutine_yield();
2924 * Some block drivers always operate in either writethrough or unsafe
2925 * mode and don't support bdrv_flush therefore. Usually qemu doesn't
2926 * know how the server works (because the behaviour is hardcoded or
2927 * depends on server-side configuration), so we can't ensure that
2928 * everything is safe on disk. Returning an error doesn't work because
2929 * that would break guests even if the server operates in writethrough
2932 * Let's hope the user knows what he's doing.
2941 /* Now flush the underlying protocol. It will also have BDRV_O_NO_FLUSH
2942 * in the case of cache=unsafe, so there are no useless flushes.
2946 QLIST_FOREACH(child
, &bs
->children
, next
) {
2947 if (child
->perm
& (BLK_PERM_WRITE
| BLK_PERM_WRITE_UNCHANGED
)) {
2948 int this_child_ret
= bdrv_co_flush(child
->bs
);
2950 ret
= this_child_ret
;
2956 /* Notify any pending flushes that we have completed */
2958 bs
->flushed_gen
= current_gen
;
2961 qemu_co_mutex_lock(&bs
->reqs_lock
);
2962 bs
->active_flush_req
= false;
2963 /* Return value is ignored - it's ok if wait queue is empty */
2964 qemu_co_queue_next(&bs
->flush_queue
);
2965 qemu_co_mutex_unlock(&bs
->reqs_lock
);
2968 bdrv_dec_in_flight(bs
);
2972 int coroutine_fn
bdrv_co_pdiscard(BdrvChild
*child
, int64_t offset
,
2975 BdrvTrackedRequest req
;
2976 int max_pdiscard
, ret
;
2977 int head
, tail
, align
;
2978 BlockDriverState
*bs
= child
->bs
;
2980 if (!bs
|| !bs
->drv
|| !bdrv_is_inserted(bs
)) {
2984 if (bdrv_has_readonly_bitmaps(bs
)) {
2988 ret
= bdrv_check_request(offset
, bytes
, NULL
);
2993 /* Do nothing if disabled. */
2994 if (!(bs
->open_flags
& BDRV_O_UNMAP
)) {
2998 if (!bs
->drv
->bdrv_co_pdiscard
&& !bs
->drv
->bdrv_aio_pdiscard
) {
3002 /* Discard is advisory, but some devices track and coalesce
3003 * unaligned requests, so we must pass everything down rather than
3004 * round here. Still, most devices will just silently ignore
3005 * unaligned requests (by returning -ENOTSUP), so we must fragment
3006 * the request accordingly. */
3007 align
= MAX(bs
->bl
.pdiscard_alignment
, bs
->bl
.request_alignment
);
3008 assert(align
% bs
->bl
.request_alignment
== 0);
3009 head
= offset
% align
;
3010 tail
= (offset
+ bytes
) % align
;
3012 bdrv_inc_in_flight(bs
);
3013 tracked_request_begin(&req
, bs
, offset
, bytes
, BDRV_TRACKED_DISCARD
);
3015 ret
= bdrv_co_write_req_prepare(child
, offset
, bytes
, &req
, 0);
3020 max_pdiscard
= QEMU_ALIGN_DOWN(MIN_NON_ZERO(bs
->bl
.max_pdiscard
, INT_MAX
),
3022 assert(max_pdiscard
>= bs
->bl
.request_alignment
);
3025 int64_t num
= bytes
;
3028 /* Make small requests to get to alignment boundaries. */
3029 num
= MIN(bytes
, align
- head
);
3030 if (!QEMU_IS_ALIGNED(num
, bs
->bl
.request_alignment
)) {
3031 num
%= bs
->bl
.request_alignment
;
3033 head
= (head
+ num
) % align
;
3034 assert(num
< max_pdiscard
);
3037 /* Shorten the request to the last aligned cluster. */
3039 } else if (!QEMU_IS_ALIGNED(tail
, bs
->bl
.request_alignment
) &&
3040 tail
> bs
->bl
.request_alignment
) {
3041 tail
%= bs
->bl
.request_alignment
;
3045 /* limit request size */
3046 if (num
> max_pdiscard
) {
3054 if (bs
->drv
->bdrv_co_pdiscard
) {
3055 ret
= bs
->drv
->bdrv_co_pdiscard(bs
, offset
, num
);
3058 CoroutineIOCompletion co
= {
3059 .coroutine
= qemu_coroutine_self(),
3062 acb
= bs
->drv
->bdrv_aio_pdiscard(bs
, offset
, num
,
3063 bdrv_co_io_em_complete
, &co
);
3068 qemu_coroutine_yield();
3072 if (ret
&& ret
!= -ENOTSUP
) {
3081 bdrv_co_write_req_finish(child
, req
.offset
, req
.bytes
, &req
, ret
);
3082 tracked_request_end(&req
);
3083 bdrv_dec_in_flight(bs
);
3087 int bdrv_co_ioctl(BlockDriverState
*bs
, int req
, void *buf
)
3089 BlockDriver
*drv
= bs
->drv
;
3090 CoroutineIOCompletion co
= {
3091 .coroutine
= qemu_coroutine_self(),
3095 bdrv_inc_in_flight(bs
);
3096 if (!drv
|| (!drv
->bdrv_aio_ioctl
&& !drv
->bdrv_co_ioctl
)) {
3101 if (drv
->bdrv_co_ioctl
) {
3102 co
.ret
= drv
->bdrv_co_ioctl(bs
, req
, buf
);
3104 acb
= drv
->bdrv_aio_ioctl(bs
, req
, buf
, bdrv_co_io_em_complete
, &co
);
3109 qemu_coroutine_yield();
3112 bdrv_dec_in_flight(bs
);
3116 void *qemu_blockalign(BlockDriverState
*bs
, size_t size
)
3118 return qemu_memalign(bdrv_opt_mem_align(bs
), size
);
3121 void *qemu_blockalign0(BlockDriverState
*bs
, size_t size
)
3123 return memset(qemu_blockalign(bs
, size
), 0, size
);
3126 void *qemu_try_blockalign(BlockDriverState
*bs
, size_t size
)
3128 size_t align
= bdrv_opt_mem_align(bs
);
3130 /* Ensure that NULL is never returned on success */
3136 return qemu_try_memalign(align
, size
);
3139 void *qemu_try_blockalign0(BlockDriverState
*bs
, size_t size
)
3141 void *mem
= qemu_try_blockalign(bs
, size
);
3144 memset(mem
, 0, size
);
3151 * Check if all memory in this vector is sector aligned.
3153 bool bdrv_qiov_is_aligned(BlockDriverState
*bs
, QEMUIOVector
*qiov
)
3156 size_t alignment
= bdrv_min_mem_align(bs
);
3158 for (i
= 0; i
< qiov
->niov
; i
++) {
3159 if ((uintptr_t) qiov
->iov
[i
].iov_base
% alignment
) {
3162 if (qiov
->iov
[i
].iov_len
% alignment
) {
3170 void bdrv_io_plug(BlockDriverState
*bs
)
3174 QLIST_FOREACH(child
, &bs
->children
, next
) {
3175 bdrv_io_plug(child
->bs
);
3178 if (qatomic_fetch_inc(&bs
->io_plugged
) == 0) {
3179 BlockDriver
*drv
= bs
->drv
;
3180 if (drv
&& drv
->bdrv_io_plug
) {
3181 drv
->bdrv_io_plug(bs
);
3186 void bdrv_io_unplug(BlockDriverState
*bs
)
3190 assert(bs
->io_plugged
);
3191 if (qatomic_fetch_dec(&bs
->io_plugged
) == 1) {
3192 BlockDriver
*drv
= bs
->drv
;
3193 if (drv
&& drv
->bdrv_io_unplug
) {
3194 drv
->bdrv_io_unplug(bs
);
3198 QLIST_FOREACH(child
, &bs
->children
, next
) {
3199 bdrv_io_unplug(child
->bs
);
3203 void bdrv_register_buf(BlockDriverState
*bs
, void *host
, size_t size
)
3207 if (bs
->drv
&& bs
->drv
->bdrv_register_buf
) {
3208 bs
->drv
->bdrv_register_buf(bs
, host
, size
);
3210 QLIST_FOREACH(child
, &bs
->children
, next
) {
3211 bdrv_register_buf(child
->bs
, host
, size
);
3215 void bdrv_unregister_buf(BlockDriverState
*bs
, void *host
)
3219 if (bs
->drv
&& bs
->drv
->bdrv_unregister_buf
) {
3220 bs
->drv
->bdrv_unregister_buf(bs
, host
);
3222 QLIST_FOREACH(child
, &bs
->children
, next
) {
3223 bdrv_unregister_buf(child
->bs
, host
);
3227 static int coroutine_fn
bdrv_co_copy_range_internal(
3228 BdrvChild
*src
, int64_t src_offset
, BdrvChild
*dst
,
3229 int64_t dst_offset
, int64_t bytes
,
3230 BdrvRequestFlags read_flags
, BdrvRequestFlags write_flags
,
3233 BdrvTrackedRequest req
;
3236 /* TODO We can support BDRV_REQ_NO_FALLBACK here */
3237 assert(!(read_flags
& BDRV_REQ_NO_FALLBACK
));
3238 assert(!(write_flags
& BDRV_REQ_NO_FALLBACK
));
3240 if (!dst
|| !dst
->bs
|| !bdrv_is_inserted(dst
->bs
)) {
3243 ret
= bdrv_check_request32(dst_offset
, bytes
, NULL
, 0);
3247 if (write_flags
& BDRV_REQ_ZERO_WRITE
) {
3248 return bdrv_co_pwrite_zeroes(dst
, dst_offset
, bytes
, write_flags
);
3251 if (!src
|| !src
->bs
|| !bdrv_is_inserted(src
->bs
)) {
3254 ret
= bdrv_check_request32(src_offset
, bytes
, NULL
, 0);
3259 if (!src
->bs
->drv
->bdrv_co_copy_range_from
3260 || !dst
->bs
->drv
->bdrv_co_copy_range_to
3261 || src
->bs
->encrypted
|| dst
->bs
->encrypted
) {
3266 bdrv_inc_in_flight(src
->bs
);
3267 tracked_request_begin(&req
, src
->bs
, src_offset
, bytes
,
3270 /* BDRV_REQ_SERIALISING is only for write operation */
3271 assert(!(read_flags
& BDRV_REQ_SERIALISING
));
3272 bdrv_wait_serialising_requests(&req
);
3274 ret
= src
->bs
->drv
->bdrv_co_copy_range_from(src
->bs
,
3278 read_flags
, write_flags
);
3280 tracked_request_end(&req
);
3281 bdrv_dec_in_flight(src
->bs
);
3283 bdrv_inc_in_flight(dst
->bs
);
3284 tracked_request_begin(&req
, dst
->bs
, dst_offset
, bytes
,
3285 BDRV_TRACKED_WRITE
);
3286 ret
= bdrv_co_write_req_prepare(dst
, dst_offset
, bytes
, &req
,
3289 ret
= dst
->bs
->drv
->bdrv_co_copy_range_to(dst
->bs
,
3293 read_flags
, write_flags
);
3295 bdrv_co_write_req_finish(dst
, dst_offset
, bytes
, &req
, ret
);
3296 tracked_request_end(&req
);
3297 bdrv_dec_in_flight(dst
->bs
);
3303 /* Copy range from @src to @dst.
3305 * See the comment of bdrv_co_copy_range for the parameter and return value
3307 int coroutine_fn
bdrv_co_copy_range_from(BdrvChild
*src
, int64_t src_offset
,
3308 BdrvChild
*dst
, int64_t dst_offset
,
3310 BdrvRequestFlags read_flags
,
3311 BdrvRequestFlags write_flags
)
3313 trace_bdrv_co_copy_range_from(src
, src_offset
, dst
, dst_offset
, bytes
,
3314 read_flags
, write_flags
);
3315 return bdrv_co_copy_range_internal(src
, src_offset
, dst
, dst_offset
,
3316 bytes
, read_flags
, write_flags
, true);
3319 /* Copy range from @src to @dst.
3321 * See the comment of bdrv_co_copy_range for the parameter and return value
3323 int coroutine_fn
bdrv_co_copy_range_to(BdrvChild
*src
, int64_t src_offset
,
3324 BdrvChild
*dst
, int64_t dst_offset
,
3326 BdrvRequestFlags read_flags
,
3327 BdrvRequestFlags write_flags
)
3329 trace_bdrv_co_copy_range_to(src
, src_offset
, dst
, dst_offset
, bytes
,
3330 read_flags
, write_flags
);
3331 return bdrv_co_copy_range_internal(src
, src_offset
, dst
, dst_offset
,
3332 bytes
, read_flags
, write_flags
, false);
3335 int coroutine_fn
bdrv_co_copy_range(BdrvChild
*src
, int64_t src_offset
,
3336 BdrvChild
*dst
, int64_t dst_offset
,
3337 int64_t bytes
, BdrvRequestFlags read_flags
,
3338 BdrvRequestFlags write_flags
)
3340 return bdrv_co_copy_range_from(src
, src_offset
,
3342 bytes
, read_flags
, write_flags
);
3345 static void bdrv_parent_cb_resize(BlockDriverState
*bs
)
3348 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
3349 if (c
->klass
->resize
) {
3350 c
->klass
->resize(c
);
3356 * Truncate file to 'offset' bytes (needed only for file protocols)
3358 * If 'exact' is true, the file must be resized to exactly the given
3359 * 'offset'. Otherwise, it is sufficient for the node to be at least
3360 * 'offset' bytes in length.
3362 int coroutine_fn
bdrv_co_truncate(BdrvChild
*child
, int64_t offset
, bool exact
,
3363 PreallocMode prealloc
, BdrvRequestFlags flags
,
3366 BlockDriverState
*bs
= child
->bs
;
3367 BdrvChild
*filtered
, *backing
;
3368 BlockDriver
*drv
= bs
->drv
;
3369 BdrvTrackedRequest req
;
3370 int64_t old_size
, new_bytes
;
3374 /* if bs->drv == NULL, bs is closed, so there's nothing to do here */
3376 error_setg(errp
, "No medium inserted");
3380 error_setg(errp
, "Image size cannot be negative");
3384 ret
= bdrv_check_request(offset
, 0, errp
);
3389 old_size
= bdrv_getlength(bs
);
3391 error_setg_errno(errp
, -old_size
, "Failed to get old image size");
3395 if (bdrv_is_read_only(bs
)) {
3396 error_setg(errp
, "Image is read-only");
3400 if (offset
> old_size
) {
3401 new_bytes
= offset
- old_size
;
3406 bdrv_inc_in_flight(bs
);
3407 tracked_request_begin(&req
, bs
, offset
- new_bytes
, new_bytes
,
3408 BDRV_TRACKED_TRUNCATE
);
3410 /* If we are growing the image and potentially using preallocation for the
3411 * new area, we need to make sure that no write requests are made to it
3412 * concurrently or they might be overwritten by preallocation. */
3414 bdrv_make_request_serialising(&req
, 1);
3416 ret
= bdrv_co_write_req_prepare(child
, offset
- new_bytes
, new_bytes
, &req
,
3419 error_setg_errno(errp
, -ret
,
3420 "Failed to prepare request for truncation");
3424 filtered
= bdrv_filter_child(bs
);
3425 backing
= bdrv_cow_child(bs
);
3428 * If the image has a backing file that is large enough that it would
3429 * provide data for the new area, we cannot leave it unallocated because
3430 * then the backing file content would become visible. Instead, zero-fill
3433 * Note that if the image has a backing file, but was opened without the
3434 * backing file, taking care of keeping things consistent with that backing
3435 * file is the user's responsibility.
3437 if (new_bytes
&& backing
) {
3438 int64_t backing_len
;
3440 backing_len
= bdrv_getlength(backing
->bs
);
3441 if (backing_len
< 0) {
3443 error_setg_errno(errp
, -ret
, "Could not get backing file size");
3447 if (backing_len
> old_size
) {
3448 flags
|= BDRV_REQ_ZERO_WRITE
;
3452 if (drv
->bdrv_co_truncate
) {
3453 if (flags
& ~bs
->supported_truncate_flags
) {
3454 error_setg(errp
, "Block driver does not support requested flags");
3458 ret
= drv
->bdrv_co_truncate(bs
, offset
, exact
, prealloc
, flags
, errp
);
3459 } else if (filtered
) {
3460 ret
= bdrv_co_truncate(filtered
, offset
, exact
, prealloc
, flags
, errp
);
3462 error_setg(errp
, "Image format driver does not support resize");
3470 ret
= refresh_total_sectors(bs
, offset
>> BDRV_SECTOR_BITS
);
3472 error_setg_errno(errp
, -ret
, "Could not refresh total sector count");
3474 offset
= bs
->total_sectors
* BDRV_SECTOR_SIZE
;
3476 /* It's possible that truncation succeeded but refresh_total_sectors
3477 * failed, but the latter doesn't affect how we should finish the request.
3478 * Pass 0 as the last parameter so that dirty bitmaps etc. are handled. */
3479 bdrv_co_write_req_finish(child
, offset
- new_bytes
, new_bytes
, &req
, 0);
3482 tracked_request_end(&req
);
3483 bdrv_dec_in_flight(bs
);
3488 void bdrv_cancel_in_flight(BlockDriverState
*bs
)
3490 if (!bs
|| !bs
->drv
) {
3494 if (bs
->drv
->bdrv_cancel_in_flight
) {
3495 bs
->drv
->bdrv_cancel_in_flight(bs
);