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 "qemu/cutils.h"
34 #include "qapi/error.h"
35 #include "qemu/error-report.h"
36 #include "qemu/main-loop.h"
37 #include "sysemu/replay.h"
39 /* Maximum bounce buffer for copy-on-read and write zeroes, in bytes */
40 #define MAX_BOUNCE_BUFFER (32768 << BDRV_SECTOR_BITS)
42 static void bdrv_parent_cb_resize(BlockDriverState
*bs
);
43 static int coroutine_fn
bdrv_co_do_pwrite_zeroes(BlockDriverState
*bs
,
44 int64_t offset
, int bytes
, BdrvRequestFlags flags
);
46 static void bdrv_parent_drained_begin(BlockDriverState
*bs
, BdrvChild
*ignore
,
47 bool ignore_bds_parents
)
51 QLIST_FOREACH_SAFE(c
, &bs
->parents
, next_parent
, next
) {
52 if (c
== ignore
|| (ignore_bds_parents
&& c
->klass
->parent_is_bds
)) {
55 bdrv_parent_drained_begin_single(c
, false);
59 static void bdrv_parent_drained_end_single_no_poll(BdrvChild
*c
,
60 int *drained_end_counter
)
62 assert(c
->parent_quiesce_counter
> 0);
63 c
->parent_quiesce_counter
--;
64 if (c
->klass
->drained_end
) {
65 c
->klass
->drained_end(c
, drained_end_counter
);
69 void bdrv_parent_drained_end_single(BdrvChild
*c
)
71 int drained_end_counter
= 0;
72 bdrv_parent_drained_end_single_no_poll(c
, &drained_end_counter
);
73 BDRV_POLL_WHILE(c
->bs
, qatomic_read(&drained_end_counter
) > 0);
76 static void bdrv_parent_drained_end(BlockDriverState
*bs
, BdrvChild
*ignore
,
77 bool ignore_bds_parents
,
78 int *drained_end_counter
)
82 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
83 if (c
== ignore
|| (ignore_bds_parents
&& c
->klass
->parent_is_bds
)) {
86 bdrv_parent_drained_end_single_no_poll(c
, drained_end_counter
);
90 static bool bdrv_parent_drained_poll_single(BdrvChild
*c
)
92 if (c
->klass
->drained_poll
) {
93 return c
->klass
->drained_poll(c
);
98 static bool bdrv_parent_drained_poll(BlockDriverState
*bs
, BdrvChild
*ignore
,
99 bool ignore_bds_parents
)
104 QLIST_FOREACH_SAFE(c
, &bs
->parents
, next_parent
, next
) {
105 if (c
== ignore
|| (ignore_bds_parents
&& c
->klass
->parent_is_bds
)) {
108 busy
|= bdrv_parent_drained_poll_single(c
);
114 void bdrv_parent_drained_begin_single(BdrvChild
*c
, bool poll
)
116 c
->parent_quiesce_counter
++;
117 if (c
->klass
->drained_begin
) {
118 c
->klass
->drained_begin(c
);
121 BDRV_POLL_WHILE(c
->bs
, bdrv_parent_drained_poll_single(c
));
125 static void bdrv_merge_limits(BlockLimits
*dst
, const BlockLimits
*src
)
127 dst
->opt_transfer
= MAX(dst
->opt_transfer
, src
->opt_transfer
);
128 dst
->max_transfer
= MIN_NON_ZERO(dst
->max_transfer
, src
->max_transfer
);
129 dst
->opt_mem_alignment
= MAX(dst
->opt_mem_alignment
,
130 src
->opt_mem_alignment
);
131 dst
->min_mem_alignment
= MAX(dst
->min_mem_alignment
,
132 src
->min_mem_alignment
);
133 dst
->max_iov
= MIN_NON_ZERO(dst
->max_iov
, src
->max_iov
);
136 void bdrv_refresh_limits(BlockDriverState
*bs
, Error
**errp
)
139 BlockDriver
*drv
= bs
->drv
;
143 memset(&bs
->bl
, 0, sizeof(bs
->bl
));
149 /* Default alignment based on whether driver has byte interface */
150 bs
->bl
.request_alignment
= (drv
->bdrv_co_preadv
||
151 drv
->bdrv_aio_preadv
||
152 drv
->bdrv_co_preadv_part
) ? 1 : 512;
154 /* Take some limits from the children as a default */
156 QLIST_FOREACH(c
, &bs
->children
, next
) {
157 if (c
->role
& (BDRV_CHILD_DATA
| BDRV_CHILD_FILTERED
| BDRV_CHILD_COW
))
159 bdrv_refresh_limits(c
->bs
, errp
);
163 bdrv_merge_limits(&bs
->bl
, &c
->bs
->bl
);
169 bs
->bl
.min_mem_alignment
= 512;
170 bs
->bl
.opt_mem_alignment
= qemu_real_host_page_size
;
172 /* Safe default since most protocols use readv()/writev()/etc */
173 bs
->bl
.max_iov
= IOV_MAX
;
176 /* Then let the driver override it */
177 if (drv
->bdrv_refresh_limits
) {
178 drv
->bdrv_refresh_limits(bs
, errp
);
184 if (bs
->bl
.request_alignment
> BDRV_MAX_ALIGNMENT
) {
185 error_setg(errp
, "Driver requires too large request alignment");
190 * The copy-on-read flag is actually a reference count so multiple users may
191 * use the feature without worrying about clobbering its previous state.
192 * Copy-on-read stays enabled until all users have called to disable it.
194 void bdrv_enable_copy_on_read(BlockDriverState
*bs
)
196 qatomic_inc(&bs
->copy_on_read
);
199 void bdrv_disable_copy_on_read(BlockDriverState
*bs
)
201 int old
= qatomic_fetch_dec(&bs
->copy_on_read
);
207 BlockDriverState
*bs
;
213 bool ignore_bds_parents
;
214 int *drained_end_counter
;
217 static void coroutine_fn
bdrv_drain_invoke_entry(void *opaque
)
219 BdrvCoDrainData
*data
= opaque
;
220 BlockDriverState
*bs
= data
->bs
;
223 bs
->drv
->bdrv_co_drain_begin(bs
);
225 bs
->drv
->bdrv_co_drain_end(bs
);
228 /* Set data->done and decrement drained_end_counter before bdrv_wakeup() */
229 qatomic_mb_set(&data
->done
, true);
231 qatomic_dec(data
->drained_end_counter
);
233 bdrv_dec_in_flight(bs
);
238 /* Recursively call BlockDriver.bdrv_co_drain_begin/end callbacks */
239 static void bdrv_drain_invoke(BlockDriverState
*bs
, bool begin
,
240 int *drained_end_counter
)
242 BdrvCoDrainData
*data
;
244 if (!bs
->drv
|| (begin
&& !bs
->drv
->bdrv_co_drain_begin
) ||
245 (!begin
&& !bs
->drv
->bdrv_co_drain_end
)) {
249 data
= g_new(BdrvCoDrainData
, 1);
250 *data
= (BdrvCoDrainData
) {
254 .drained_end_counter
= drained_end_counter
,
258 qatomic_inc(drained_end_counter
);
261 /* Make sure the driver callback completes during the polling phase for
263 bdrv_inc_in_flight(bs
);
264 data
->co
= qemu_coroutine_create(bdrv_drain_invoke_entry
, data
);
265 aio_co_schedule(bdrv_get_aio_context(bs
), data
->co
);
268 /* Returns true if BDRV_POLL_WHILE() should go into a blocking aio_poll() */
269 bool bdrv_drain_poll(BlockDriverState
*bs
, bool recursive
,
270 BdrvChild
*ignore_parent
, bool ignore_bds_parents
)
272 BdrvChild
*child
, *next
;
274 if (bdrv_parent_drained_poll(bs
, ignore_parent
, ignore_bds_parents
)) {
278 if (qatomic_read(&bs
->in_flight
)) {
283 assert(!ignore_bds_parents
);
284 QLIST_FOREACH_SAFE(child
, &bs
->children
, next
, next
) {
285 if (bdrv_drain_poll(child
->bs
, recursive
, child
, false)) {
294 static bool bdrv_drain_poll_top_level(BlockDriverState
*bs
, bool recursive
,
295 BdrvChild
*ignore_parent
)
297 return bdrv_drain_poll(bs
, recursive
, ignore_parent
, false);
300 static void bdrv_do_drained_begin(BlockDriverState
*bs
, bool recursive
,
301 BdrvChild
*parent
, bool ignore_bds_parents
,
303 static void bdrv_do_drained_end(BlockDriverState
*bs
, bool recursive
,
304 BdrvChild
*parent
, bool ignore_bds_parents
,
305 int *drained_end_counter
);
307 static void bdrv_co_drain_bh_cb(void *opaque
)
309 BdrvCoDrainData
*data
= opaque
;
310 Coroutine
*co
= data
->co
;
311 BlockDriverState
*bs
= data
->bs
;
314 AioContext
*ctx
= bdrv_get_aio_context(bs
);
315 aio_context_acquire(ctx
);
316 bdrv_dec_in_flight(bs
);
318 assert(!data
->drained_end_counter
);
319 bdrv_do_drained_begin(bs
, data
->recursive
, data
->parent
,
320 data
->ignore_bds_parents
, data
->poll
);
323 bdrv_do_drained_end(bs
, data
->recursive
, data
->parent
,
324 data
->ignore_bds_parents
,
325 data
->drained_end_counter
);
327 aio_context_release(ctx
);
330 bdrv_drain_all_begin();
337 static void coroutine_fn
bdrv_co_yield_to_drain(BlockDriverState
*bs
,
338 bool begin
, bool recursive
,
340 bool ignore_bds_parents
,
342 int *drained_end_counter
)
344 BdrvCoDrainData data
;
345 Coroutine
*self
= qemu_coroutine_self();
346 AioContext
*ctx
= bdrv_get_aio_context(bs
);
347 AioContext
*co_ctx
= qemu_coroutine_get_aio_context(self
);
349 /* Calling bdrv_drain() from a BH ensures the current coroutine yields and
350 * other coroutines run if they were queued by aio_co_enter(). */
352 assert(qemu_in_coroutine());
353 data
= (BdrvCoDrainData
) {
358 .recursive
= recursive
,
360 .ignore_bds_parents
= ignore_bds_parents
,
362 .drained_end_counter
= drained_end_counter
,
366 bdrv_inc_in_flight(bs
);
370 * Temporarily drop the lock across yield or we would get deadlocks.
371 * bdrv_co_drain_bh_cb() reaquires the lock as needed.
373 * When we yield below, the lock for the current context will be
374 * released, so if this is actually the lock that protects bs, don't drop
378 aio_context_release(ctx
);
380 replay_bh_schedule_oneshot_event(ctx
, bdrv_co_drain_bh_cb
, &data
);
382 qemu_coroutine_yield();
383 /* If we are resumed from some other event (such as an aio completion or a
384 * timer callback), it is a bug in the caller that should be fixed. */
387 /* Reaquire the AioContext of bs if we dropped it */
389 aio_context_acquire(ctx
);
393 void bdrv_do_drained_begin_quiesce(BlockDriverState
*bs
,
394 BdrvChild
*parent
, bool ignore_bds_parents
)
396 assert(!qemu_in_coroutine());
398 /* Stop things in parent-to-child order */
399 if (qatomic_fetch_inc(&bs
->quiesce_counter
) == 0) {
400 aio_disable_external(bdrv_get_aio_context(bs
));
403 bdrv_parent_drained_begin(bs
, parent
, ignore_bds_parents
);
404 bdrv_drain_invoke(bs
, true, NULL
);
407 static void bdrv_do_drained_begin(BlockDriverState
*bs
, bool recursive
,
408 BdrvChild
*parent
, bool ignore_bds_parents
,
411 BdrvChild
*child
, *next
;
413 if (qemu_in_coroutine()) {
414 bdrv_co_yield_to_drain(bs
, true, recursive
, parent
, ignore_bds_parents
,
419 bdrv_do_drained_begin_quiesce(bs
, parent
, ignore_bds_parents
);
422 assert(!ignore_bds_parents
);
423 bs
->recursive_quiesce_counter
++;
424 QLIST_FOREACH_SAFE(child
, &bs
->children
, next
, next
) {
425 bdrv_do_drained_begin(child
->bs
, true, child
, ignore_bds_parents
,
431 * Wait for drained requests to finish.
433 * Calling BDRV_POLL_WHILE() only once for the top-level node is okay: The
434 * call is needed so things in this AioContext can make progress even
435 * though we don't return to the main AioContext loop - this automatically
436 * includes other nodes in the same AioContext and therefore all child
440 assert(!ignore_bds_parents
);
441 BDRV_POLL_WHILE(bs
, bdrv_drain_poll_top_level(bs
, recursive
, parent
));
445 void bdrv_drained_begin(BlockDriverState
*bs
)
447 bdrv_do_drained_begin(bs
, false, NULL
, false, true);
450 void bdrv_subtree_drained_begin(BlockDriverState
*bs
)
452 bdrv_do_drained_begin(bs
, true, NULL
, false, true);
456 * This function does not poll, nor must any of its recursively called
457 * functions. The *drained_end_counter pointee will be incremented
458 * once for every background operation scheduled, and decremented once
459 * the operation settles. Therefore, the pointer must remain valid
460 * until the pointee reaches 0. That implies that whoever sets up the
461 * pointee has to poll until it is 0.
463 * We use atomic operations to access *drained_end_counter, because
464 * (1) when called from bdrv_set_aio_context_ignore(), the subgraph of
465 * @bs may contain nodes in different AioContexts,
466 * (2) bdrv_drain_all_end() uses the same counter for all nodes,
467 * regardless of which AioContext they are in.
469 static void bdrv_do_drained_end(BlockDriverState
*bs
, bool recursive
,
470 BdrvChild
*parent
, bool ignore_bds_parents
,
471 int *drained_end_counter
)
474 int old_quiesce_counter
;
476 assert(drained_end_counter
!= NULL
);
478 if (qemu_in_coroutine()) {
479 bdrv_co_yield_to_drain(bs
, false, recursive
, parent
, ignore_bds_parents
,
480 false, drained_end_counter
);
483 assert(bs
->quiesce_counter
> 0);
485 /* Re-enable things in child-to-parent order */
486 bdrv_drain_invoke(bs
, false, drained_end_counter
);
487 bdrv_parent_drained_end(bs
, parent
, ignore_bds_parents
,
488 drained_end_counter
);
490 old_quiesce_counter
= qatomic_fetch_dec(&bs
->quiesce_counter
);
491 if (old_quiesce_counter
== 1) {
492 aio_enable_external(bdrv_get_aio_context(bs
));
496 assert(!ignore_bds_parents
);
497 bs
->recursive_quiesce_counter
--;
498 QLIST_FOREACH(child
, &bs
->children
, next
) {
499 bdrv_do_drained_end(child
->bs
, true, child
, ignore_bds_parents
,
500 drained_end_counter
);
505 void bdrv_drained_end(BlockDriverState
*bs
)
507 int drained_end_counter
= 0;
508 bdrv_do_drained_end(bs
, false, NULL
, false, &drained_end_counter
);
509 BDRV_POLL_WHILE(bs
, qatomic_read(&drained_end_counter
) > 0);
512 void bdrv_drained_end_no_poll(BlockDriverState
*bs
, int *drained_end_counter
)
514 bdrv_do_drained_end(bs
, false, NULL
, false, drained_end_counter
);
517 void bdrv_subtree_drained_end(BlockDriverState
*bs
)
519 int drained_end_counter
= 0;
520 bdrv_do_drained_end(bs
, true, NULL
, false, &drained_end_counter
);
521 BDRV_POLL_WHILE(bs
, qatomic_read(&drained_end_counter
) > 0);
524 void bdrv_apply_subtree_drain(BdrvChild
*child
, BlockDriverState
*new_parent
)
528 for (i
= 0; i
< new_parent
->recursive_quiesce_counter
; i
++) {
529 bdrv_do_drained_begin(child
->bs
, true, child
, false, true);
533 void bdrv_unapply_subtree_drain(BdrvChild
*child
, BlockDriverState
*old_parent
)
535 int drained_end_counter
= 0;
538 for (i
= 0; i
< old_parent
->recursive_quiesce_counter
; i
++) {
539 bdrv_do_drained_end(child
->bs
, true, child
, false,
540 &drained_end_counter
);
543 BDRV_POLL_WHILE(child
->bs
, qatomic_read(&drained_end_counter
) > 0);
547 * Wait for pending requests to complete on a single BlockDriverState subtree,
548 * and suspend block driver's internal I/O until next request arrives.
550 * Note that unlike bdrv_drain_all(), the caller must hold the BlockDriverState
553 void coroutine_fn
bdrv_co_drain(BlockDriverState
*bs
)
555 assert(qemu_in_coroutine());
556 bdrv_drained_begin(bs
);
557 bdrv_drained_end(bs
);
560 void bdrv_drain(BlockDriverState
*bs
)
562 bdrv_drained_begin(bs
);
563 bdrv_drained_end(bs
);
566 static void bdrv_drain_assert_idle(BlockDriverState
*bs
)
568 BdrvChild
*child
, *next
;
570 assert(qatomic_read(&bs
->in_flight
) == 0);
571 QLIST_FOREACH_SAFE(child
, &bs
->children
, next
, next
) {
572 bdrv_drain_assert_idle(child
->bs
);
576 unsigned int bdrv_drain_all_count
= 0;
578 static bool bdrv_drain_all_poll(void)
580 BlockDriverState
*bs
= NULL
;
583 /* bdrv_drain_poll() can't make changes to the graph and we are holding the
584 * main AioContext lock, so iterating bdrv_next_all_states() is safe. */
585 while ((bs
= bdrv_next_all_states(bs
))) {
586 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
587 aio_context_acquire(aio_context
);
588 result
|= bdrv_drain_poll(bs
, false, NULL
, true);
589 aio_context_release(aio_context
);
596 * Wait for pending requests to complete across all BlockDriverStates
598 * This function does not flush data to disk, use bdrv_flush_all() for that
599 * after calling this function.
601 * This pauses all block jobs and disables external clients. It must
602 * be paired with bdrv_drain_all_end().
604 * NOTE: no new block jobs or BlockDriverStates can be created between
605 * the bdrv_drain_all_begin() and bdrv_drain_all_end() calls.
607 void bdrv_drain_all_begin(void)
609 BlockDriverState
*bs
= NULL
;
611 if (qemu_in_coroutine()) {
612 bdrv_co_yield_to_drain(NULL
, true, false, NULL
, true, true, NULL
);
617 * bdrv queue is managed by record/replay,
618 * waiting for finishing the I/O requests may
621 if (replay_events_enabled()) {
625 /* AIO_WAIT_WHILE() with a NULL context can only be called from the main
626 * loop AioContext, so make sure we're in the main context. */
627 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
628 assert(bdrv_drain_all_count
< INT_MAX
);
629 bdrv_drain_all_count
++;
631 /* Quiesce all nodes, without polling in-flight requests yet. The graph
632 * cannot change during this loop. */
633 while ((bs
= bdrv_next_all_states(bs
))) {
634 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
636 aio_context_acquire(aio_context
);
637 bdrv_do_drained_begin(bs
, false, NULL
, true, false);
638 aio_context_release(aio_context
);
641 /* Now poll the in-flight requests */
642 AIO_WAIT_WHILE(NULL
, bdrv_drain_all_poll());
644 while ((bs
= bdrv_next_all_states(bs
))) {
645 bdrv_drain_assert_idle(bs
);
649 void bdrv_drain_all_end_quiesce(BlockDriverState
*bs
)
651 int drained_end_counter
= 0;
653 g_assert(bs
->quiesce_counter
> 0);
654 g_assert(!bs
->refcnt
);
656 while (bs
->quiesce_counter
) {
657 bdrv_do_drained_end(bs
, false, NULL
, true, &drained_end_counter
);
659 BDRV_POLL_WHILE(bs
, qatomic_read(&drained_end_counter
) > 0);
662 void bdrv_drain_all_end(void)
664 BlockDriverState
*bs
= NULL
;
665 int drained_end_counter
= 0;
668 * bdrv queue is managed by record/replay,
669 * waiting for finishing the I/O requests may
672 if (replay_events_enabled()) {
676 while ((bs
= bdrv_next_all_states(bs
))) {
677 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
679 aio_context_acquire(aio_context
);
680 bdrv_do_drained_end(bs
, false, NULL
, true, &drained_end_counter
);
681 aio_context_release(aio_context
);
684 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
685 AIO_WAIT_WHILE(NULL
, qatomic_read(&drained_end_counter
) > 0);
687 assert(bdrv_drain_all_count
> 0);
688 bdrv_drain_all_count
--;
691 void bdrv_drain_all(void)
693 bdrv_drain_all_begin();
694 bdrv_drain_all_end();
698 * Remove an active request from the tracked requests list
700 * This function should be called when a tracked request is completing.
702 static void tracked_request_end(BdrvTrackedRequest
*req
)
704 if (req
->serialising
) {
705 qatomic_dec(&req
->bs
->serialising_in_flight
);
708 qemu_co_mutex_lock(&req
->bs
->reqs_lock
);
709 QLIST_REMOVE(req
, list
);
710 qemu_co_queue_restart_all(&req
->wait_queue
);
711 qemu_co_mutex_unlock(&req
->bs
->reqs_lock
);
715 * Add an active request to the tracked requests list
717 static void tracked_request_begin(BdrvTrackedRequest
*req
,
718 BlockDriverState
*bs
,
721 enum BdrvTrackedRequestType type
)
723 assert(bytes
<= INT64_MAX
&& offset
<= INT64_MAX
- bytes
);
725 *req
= (BdrvTrackedRequest
){
730 .co
= qemu_coroutine_self(),
731 .serialising
= false,
732 .overlap_offset
= offset
,
733 .overlap_bytes
= bytes
,
736 qemu_co_queue_init(&req
->wait_queue
);
738 qemu_co_mutex_lock(&bs
->reqs_lock
);
739 QLIST_INSERT_HEAD(&bs
->tracked_requests
, req
, list
);
740 qemu_co_mutex_unlock(&bs
->reqs_lock
);
743 static bool tracked_request_overlaps(BdrvTrackedRequest
*req
,
744 int64_t offset
, uint64_t bytes
)
747 if (offset
>= req
->overlap_offset
+ req
->overlap_bytes
) {
751 if (req
->overlap_offset
>= offset
+ bytes
) {
757 /* Called with self->bs->reqs_lock held */
758 static BdrvTrackedRequest
*
759 bdrv_find_conflicting_request(BdrvTrackedRequest
*self
)
761 BdrvTrackedRequest
*req
;
763 QLIST_FOREACH(req
, &self
->bs
->tracked_requests
, list
) {
764 if (req
== self
|| (!req
->serialising
&& !self
->serialising
)) {
767 if (tracked_request_overlaps(req
, self
->overlap_offset
,
768 self
->overlap_bytes
))
771 * Hitting this means there was a reentrant request, for
772 * example, a block driver issuing nested requests. This must
773 * never happen since it means deadlock.
775 assert(qemu_coroutine_self() != req
->co
);
778 * If the request is already (indirectly) waiting for us, or
779 * will wait for us as soon as it wakes up, then just go on
780 * (instead of producing a deadlock in the former case).
782 if (!req
->waiting_for
) {
791 /* Called with self->bs->reqs_lock held */
792 static bool coroutine_fn
793 bdrv_wait_serialising_requests_locked(BdrvTrackedRequest
*self
)
795 BdrvTrackedRequest
*req
;
798 while ((req
= bdrv_find_conflicting_request(self
))) {
799 self
->waiting_for
= req
;
800 qemu_co_queue_wait(&req
->wait_queue
, &self
->bs
->reqs_lock
);
801 self
->waiting_for
= NULL
;
808 /* Called with req->bs->reqs_lock held */
809 static void tracked_request_set_serialising(BdrvTrackedRequest
*req
,
812 int64_t overlap_offset
= req
->offset
& ~(align
- 1);
813 uint64_t overlap_bytes
= ROUND_UP(req
->offset
+ req
->bytes
, align
)
816 if (!req
->serialising
) {
817 qatomic_inc(&req
->bs
->serialising_in_flight
);
818 req
->serialising
= true;
821 req
->overlap_offset
= MIN(req
->overlap_offset
, overlap_offset
);
822 req
->overlap_bytes
= MAX(req
->overlap_bytes
, overlap_bytes
);
826 * Return the tracked request on @bs for the current coroutine, or
827 * NULL if there is none.
829 BdrvTrackedRequest
*coroutine_fn
bdrv_co_get_self_request(BlockDriverState
*bs
)
831 BdrvTrackedRequest
*req
;
832 Coroutine
*self
= qemu_coroutine_self();
834 QLIST_FOREACH(req
, &bs
->tracked_requests
, list
) {
835 if (req
->co
== self
) {
844 * Round a region to cluster boundaries
846 void bdrv_round_to_clusters(BlockDriverState
*bs
,
847 int64_t offset
, int64_t bytes
,
848 int64_t *cluster_offset
,
849 int64_t *cluster_bytes
)
853 if (bdrv_get_info(bs
, &bdi
) < 0 || bdi
.cluster_size
== 0) {
854 *cluster_offset
= offset
;
855 *cluster_bytes
= bytes
;
857 int64_t c
= bdi
.cluster_size
;
858 *cluster_offset
= QEMU_ALIGN_DOWN(offset
, c
);
859 *cluster_bytes
= QEMU_ALIGN_UP(offset
- *cluster_offset
+ bytes
, c
);
863 static int bdrv_get_cluster_size(BlockDriverState
*bs
)
868 ret
= bdrv_get_info(bs
, &bdi
);
869 if (ret
< 0 || bdi
.cluster_size
== 0) {
870 return bs
->bl
.request_alignment
;
872 return bdi
.cluster_size
;
876 void bdrv_inc_in_flight(BlockDriverState
*bs
)
878 qatomic_inc(&bs
->in_flight
);
881 void bdrv_wakeup(BlockDriverState
*bs
)
886 void bdrv_dec_in_flight(BlockDriverState
*bs
)
888 qatomic_dec(&bs
->in_flight
);
892 static bool coroutine_fn
bdrv_wait_serialising_requests(BdrvTrackedRequest
*self
)
894 BlockDriverState
*bs
= self
->bs
;
897 if (!qatomic_read(&bs
->serialising_in_flight
)) {
901 qemu_co_mutex_lock(&bs
->reqs_lock
);
902 waited
= bdrv_wait_serialising_requests_locked(self
);
903 qemu_co_mutex_unlock(&bs
->reqs_lock
);
908 bool coroutine_fn
bdrv_make_request_serialising(BdrvTrackedRequest
*req
,
913 qemu_co_mutex_lock(&req
->bs
->reqs_lock
);
915 tracked_request_set_serialising(req
, align
);
916 waited
= bdrv_wait_serialising_requests_locked(req
);
918 qemu_co_mutex_unlock(&req
->bs
->reqs_lock
);
923 int bdrv_check_request(int64_t offset
, int64_t bytes
)
925 if (offset
< 0 || bytes
< 0) {
929 if (bytes
> BDRV_MAX_LENGTH
) {
933 if (offset
> BDRV_MAX_LENGTH
- bytes
) {
940 static int bdrv_check_request32(int64_t offset
, int64_t bytes
)
942 int ret
= bdrv_check_request(offset
, bytes
);
947 if (bytes
> BDRV_REQUEST_MAX_BYTES
) {
954 int bdrv_pwrite_zeroes(BdrvChild
*child
, int64_t offset
,
955 int bytes
, BdrvRequestFlags flags
)
957 return bdrv_pwritev(child
, offset
, bytes
, NULL
,
958 BDRV_REQ_ZERO_WRITE
| flags
);
962 * Completely zero out a block device with the help of bdrv_pwrite_zeroes.
963 * The operation is sped up by checking the block status and only writing
964 * zeroes to the device if they currently do not return zeroes. Optional
965 * flags are passed through to bdrv_pwrite_zeroes (e.g. BDRV_REQ_MAY_UNMAP,
968 * Returns < 0 on error, 0 on success. For error codes see bdrv_pwrite().
970 int bdrv_make_zero(BdrvChild
*child
, BdrvRequestFlags flags
)
973 int64_t target_size
, bytes
, offset
= 0;
974 BlockDriverState
*bs
= child
->bs
;
976 target_size
= bdrv_getlength(bs
);
977 if (target_size
< 0) {
982 bytes
= MIN(target_size
- offset
, BDRV_REQUEST_MAX_BYTES
);
986 ret
= bdrv_block_status(bs
, offset
, bytes
, &bytes
, NULL
, NULL
);
990 if (ret
& BDRV_BLOCK_ZERO
) {
994 ret
= bdrv_pwrite_zeroes(child
, offset
, bytes
, flags
);
1002 /* See bdrv_pwrite() for the return codes */
1003 int bdrv_pread(BdrvChild
*child
, int64_t offset
, void *buf
, int bytes
)
1006 QEMUIOVector qiov
= QEMU_IOVEC_INIT_BUF(qiov
, buf
, bytes
);
1012 ret
= bdrv_preadv(child
, offset
, bytes
, &qiov
, 0);
1014 return ret
< 0 ? ret
: bytes
;
1017 /* Return no. of bytes on success or < 0 on error. Important errors are:
1018 -EIO generic I/O error (may happen for all errors)
1019 -ENOMEDIUM No media inserted.
1020 -EINVAL Invalid offset or number of bytes
1021 -EACCES Trying to write a read-only device
1023 int bdrv_pwrite(BdrvChild
*child
, int64_t offset
, const void *buf
, int bytes
)
1026 QEMUIOVector qiov
= QEMU_IOVEC_INIT_BUF(qiov
, buf
, bytes
);
1032 ret
= bdrv_pwritev(child
, offset
, bytes
, &qiov
, 0);
1034 return ret
< 0 ? ret
: bytes
;
1038 * Writes to the file and ensures that no writes are reordered across this
1039 * request (acts as a barrier)
1041 * Returns 0 on success, -errno in error cases.
1043 int bdrv_pwrite_sync(BdrvChild
*child
, int64_t offset
,
1044 const void *buf
, int count
)
1048 ret
= bdrv_pwrite(child
, offset
, buf
, count
);
1053 ret
= bdrv_flush(child
->bs
);
1061 typedef struct CoroutineIOCompletion
{
1062 Coroutine
*coroutine
;
1064 } CoroutineIOCompletion
;
1066 static void bdrv_co_io_em_complete(void *opaque
, int ret
)
1068 CoroutineIOCompletion
*co
= opaque
;
1071 aio_co_wake(co
->coroutine
);
1074 static int coroutine_fn
bdrv_driver_preadv(BlockDriverState
*bs
,
1075 uint64_t offset
, uint64_t bytes
,
1077 size_t qiov_offset
, int flags
)
1079 BlockDriver
*drv
= bs
->drv
;
1081 unsigned int nb_sectors
;
1082 QEMUIOVector local_qiov
;
1085 assert(!(flags
& ~BDRV_REQ_MASK
));
1086 assert(!(flags
& BDRV_REQ_NO_FALLBACK
));
1092 if (drv
->bdrv_co_preadv_part
) {
1093 return drv
->bdrv_co_preadv_part(bs
, offset
, bytes
, qiov
, qiov_offset
,
1097 if (qiov_offset
> 0 || bytes
!= qiov
->size
) {
1098 qemu_iovec_init_slice(&local_qiov
, qiov
, qiov_offset
, bytes
);
1102 if (drv
->bdrv_co_preadv
) {
1103 ret
= drv
->bdrv_co_preadv(bs
, offset
, bytes
, qiov
, flags
);
1107 if (drv
->bdrv_aio_preadv
) {
1109 CoroutineIOCompletion co
= {
1110 .coroutine
= qemu_coroutine_self(),
1113 acb
= drv
->bdrv_aio_preadv(bs
, offset
, bytes
, qiov
, flags
,
1114 bdrv_co_io_em_complete
, &co
);
1119 qemu_coroutine_yield();
1125 sector_num
= offset
>> BDRV_SECTOR_BITS
;
1126 nb_sectors
= bytes
>> BDRV_SECTOR_BITS
;
1128 assert(QEMU_IS_ALIGNED(offset
, BDRV_SECTOR_SIZE
));
1129 assert(QEMU_IS_ALIGNED(bytes
, BDRV_SECTOR_SIZE
));
1130 assert(bytes
<= BDRV_REQUEST_MAX_BYTES
);
1131 assert(drv
->bdrv_co_readv
);
1133 ret
= drv
->bdrv_co_readv(bs
, sector_num
, nb_sectors
, qiov
);
1136 if (qiov
== &local_qiov
) {
1137 qemu_iovec_destroy(&local_qiov
);
1143 static int coroutine_fn
bdrv_driver_pwritev(BlockDriverState
*bs
,
1144 uint64_t offset
, uint64_t bytes
,
1146 size_t qiov_offset
, int flags
)
1148 BlockDriver
*drv
= bs
->drv
;
1150 unsigned int nb_sectors
;
1151 QEMUIOVector local_qiov
;
1154 assert(!(flags
& ~BDRV_REQ_MASK
));
1155 assert(!(flags
& BDRV_REQ_NO_FALLBACK
));
1161 if (drv
->bdrv_co_pwritev_part
) {
1162 ret
= drv
->bdrv_co_pwritev_part(bs
, offset
, bytes
, qiov
, qiov_offset
,
1163 flags
& bs
->supported_write_flags
);
1164 flags
&= ~bs
->supported_write_flags
;
1168 if (qiov_offset
> 0 || bytes
!= qiov
->size
) {
1169 qemu_iovec_init_slice(&local_qiov
, qiov
, qiov_offset
, bytes
);
1173 if (drv
->bdrv_co_pwritev
) {
1174 ret
= drv
->bdrv_co_pwritev(bs
, offset
, bytes
, qiov
,
1175 flags
& bs
->supported_write_flags
);
1176 flags
&= ~bs
->supported_write_flags
;
1180 if (drv
->bdrv_aio_pwritev
) {
1182 CoroutineIOCompletion co
= {
1183 .coroutine
= qemu_coroutine_self(),
1186 acb
= drv
->bdrv_aio_pwritev(bs
, offset
, bytes
, qiov
,
1187 flags
& bs
->supported_write_flags
,
1188 bdrv_co_io_em_complete
, &co
);
1189 flags
&= ~bs
->supported_write_flags
;
1193 qemu_coroutine_yield();
1199 sector_num
= offset
>> BDRV_SECTOR_BITS
;
1200 nb_sectors
= bytes
>> BDRV_SECTOR_BITS
;
1202 assert(QEMU_IS_ALIGNED(offset
, BDRV_SECTOR_SIZE
));
1203 assert(QEMU_IS_ALIGNED(bytes
, BDRV_SECTOR_SIZE
));
1204 assert(bytes
<= BDRV_REQUEST_MAX_BYTES
);
1206 assert(drv
->bdrv_co_writev
);
1207 ret
= drv
->bdrv_co_writev(bs
, sector_num
, nb_sectors
, qiov
,
1208 flags
& bs
->supported_write_flags
);
1209 flags
&= ~bs
->supported_write_flags
;
1212 if (ret
== 0 && (flags
& BDRV_REQ_FUA
)) {
1213 ret
= bdrv_co_flush(bs
);
1216 if (qiov
== &local_qiov
) {
1217 qemu_iovec_destroy(&local_qiov
);
1223 static int coroutine_fn
1224 bdrv_driver_pwritev_compressed(BlockDriverState
*bs
, uint64_t offset
,
1225 uint64_t bytes
, QEMUIOVector
*qiov
,
1228 BlockDriver
*drv
= bs
->drv
;
1229 QEMUIOVector local_qiov
;
1236 if (!block_driver_can_compress(drv
)) {
1240 if (drv
->bdrv_co_pwritev_compressed_part
) {
1241 return drv
->bdrv_co_pwritev_compressed_part(bs
, offset
, bytes
,
1245 if (qiov_offset
== 0) {
1246 return drv
->bdrv_co_pwritev_compressed(bs
, offset
, bytes
, qiov
);
1249 qemu_iovec_init_slice(&local_qiov
, qiov
, qiov_offset
, bytes
);
1250 ret
= drv
->bdrv_co_pwritev_compressed(bs
, offset
, bytes
, &local_qiov
);
1251 qemu_iovec_destroy(&local_qiov
);
1256 static int coroutine_fn
bdrv_co_do_copy_on_readv(BdrvChild
*child
,
1257 int64_t offset
, unsigned int bytes
, QEMUIOVector
*qiov
,
1258 size_t qiov_offset
, int flags
)
1260 BlockDriverState
*bs
= child
->bs
;
1262 /* Perform I/O through a temporary buffer so that users who scribble over
1263 * their read buffer while the operation is in progress do not end up
1264 * modifying the image file. This is critical for zero-copy guest I/O
1265 * where anything might happen inside guest memory.
1267 void *bounce_buffer
= NULL
;
1269 BlockDriver
*drv
= bs
->drv
;
1270 int64_t cluster_offset
;
1271 int64_t cluster_bytes
;
1274 int max_transfer
= MIN_NON_ZERO(bs
->bl
.max_transfer
,
1275 BDRV_REQUEST_MAX_BYTES
);
1276 unsigned int progress
= 0;
1284 * Do not write anything when the BDS is inactive. That is not
1285 * allowed, and it would not help.
1287 skip_write
= (bs
->open_flags
& BDRV_O_INACTIVE
);
1289 /* FIXME We cannot require callers to have write permissions when all they
1290 * are doing is a read request. If we did things right, write permissions
1291 * would be obtained anyway, but internally by the copy-on-read code. As
1292 * long as it is implemented here rather than in a separate filter driver,
1293 * the copy-on-read code doesn't have its own BdrvChild, however, for which
1294 * it could request permissions. Therefore we have to bypass the permission
1295 * system for the moment. */
1296 // assert(child->perm & (BLK_PERM_WRITE_UNCHANGED | BLK_PERM_WRITE));
1298 /* Cover entire cluster so no additional backing file I/O is required when
1299 * allocating cluster in the image file. Note that this value may exceed
1300 * BDRV_REQUEST_MAX_BYTES (even when the original read did not), which
1301 * is one reason we loop rather than doing it all at once.
1303 bdrv_round_to_clusters(bs
, offset
, bytes
, &cluster_offset
, &cluster_bytes
);
1304 skip_bytes
= offset
- cluster_offset
;
1306 trace_bdrv_co_do_copy_on_readv(bs
, offset
, bytes
,
1307 cluster_offset
, cluster_bytes
);
1309 while (cluster_bytes
) {
1313 ret
= 1; /* "already allocated", so nothing will be copied */
1314 pnum
= MIN(cluster_bytes
, max_transfer
);
1316 ret
= bdrv_is_allocated(bs
, cluster_offset
,
1317 MIN(cluster_bytes
, max_transfer
), &pnum
);
1320 * Safe to treat errors in querying allocation as if
1321 * unallocated; we'll probably fail again soon on the
1322 * read, but at least that will set a decent errno.
1324 pnum
= MIN(cluster_bytes
, max_transfer
);
1327 /* Stop at EOF if the image ends in the middle of the cluster */
1328 if (ret
== 0 && pnum
== 0) {
1329 assert(progress
>= bytes
);
1333 assert(skip_bytes
< pnum
);
1337 QEMUIOVector local_qiov
;
1339 /* Must copy-on-read; use the bounce buffer */
1340 pnum
= MIN(pnum
, MAX_BOUNCE_BUFFER
);
1341 if (!bounce_buffer
) {
1342 int64_t max_we_need
= MAX(pnum
, cluster_bytes
- pnum
);
1343 int64_t max_allowed
= MIN(max_transfer
, MAX_BOUNCE_BUFFER
);
1344 int64_t bounce_buffer_len
= MIN(max_we_need
, max_allowed
);
1346 bounce_buffer
= qemu_try_blockalign(bs
, bounce_buffer_len
);
1347 if (!bounce_buffer
) {
1352 qemu_iovec_init_buf(&local_qiov
, bounce_buffer
, pnum
);
1354 ret
= bdrv_driver_preadv(bs
, cluster_offset
, pnum
,
1360 bdrv_debug_event(bs
, BLKDBG_COR_WRITE
);
1361 if (drv
->bdrv_co_pwrite_zeroes
&&
1362 buffer_is_zero(bounce_buffer
, pnum
)) {
1363 /* FIXME: Should we (perhaps conditionally) be setting
1364 * BDRV_REQ_MAY_UNMAP, if it will allow for a sparser copy
1365 * that still correctly reads as zero? */
1366 ret
= bdrv_co_do_pwrite_zeroes(bs
, cluster_offset
, pnum
,
1367 BDRV_REQ_WRITE_UNCHANGED
);
1369 /* This does not change the data on the disk, it is not
1370 * necessary to flush even in cache=writethrough mode.
1372 ret
= bdrv_driver_pwritev(bs
, cluster_offset
, pnum
,
1374 BDRV_REQ_WRITE_UNCHANGED
);
1378 /* It might be okay to ignore write errors for guest
1379 * requests. If this is a deliberate copy-on-read
1380 * then we don't want to ignore the error. Simply
1381 * report it in all cases.
1386 if (!(flags
& BDRV_REQ_PREFETCH
)) {
1387 qemu_iovec_from_buf(qiov
, qiov_offset
+ progress
,
1388 bounce_buffer
+ skip_bytes
,
1389 MIN(pnum
- skip_bytes
, bytes
- progress
));
1391 } else if (!(flags
& BDRV_REQ_PREFETCH
)) {
1392 /* Read directly into the destination */
1393 ret
= bdrv_driver_preadv(bs
, offset
+ progress
,
1394 MIN(pnum
- skip_bytes
, bytes
- progress
),
1395 qiov
, qiov_offset
+ progress
, 0);
1401 cluster_offset
+= pnum
;
1402 cluster_bytes
-= pnum
;
1403 progress
+= pnum
- skip_bytes
;
1409 qemu_vfree(bounce_buffer
);
1414 * Forwards an already correctly aligned request to the BlockDriver. This
1415 * handles copy on read, zeroing after EOF, and fragmentation of large
1416 * reads; any other features must be implemented by the caller.
1418 static int coroutine_fn
bdrv_aligned_preadv(BdrvChild
*child
,
1419 BdrvTrackedRequest
*req
, int64_t offset
, unsigned int bytes
,
1420 int64_t align
, QEMUIOVector
*qiov
, size_t qiov_offset
, int flags
)
1422 BlockDriverState
*bs
= child
->bs
;
1423 int64_t total_bytes
, max_bytes
;
1425 uint64_t bytes_remaining
= bytes
;
1428 assert(is_power_of_2(align
));
1429 assert((offset
& (align
- 1)) == 0);
1430 assert((bytes
& (align
- 1)) == 0);
1431 assert((bs
->open_flags
& BDRV_O_NO_IO
) == 0);
1432 max_transfer
= QEMU_ALIGN_DOWN(MIN_NON_ZERO(bs
->bl
.max_transfer
, INT_MAX
),
1435 /* TODO: We would need a per-BDS .supported_read_flags and
1436 * potential fallback support, if we ever implement any read flags
1437 * to pass through to drivers. For now, there aren't any
1438 * passthrough flags. */
1439 assert(!(flags
& ~(BDRV_REQ_COPY_ON_READ
| BDRV_REQ_PREFETCH
)));
1441 /* Handle Copy on Read and associated serialisation */
1442 if (flags
& BDRV_REQ_COPY_ON_READ
) {
1443 /* If we touch the same cluster it counts as an overlap. This
1444 * guarantees that allocating writes will be serialized and not race
1445 * with each other for the same cluster. For example, in copy-on-read
1446 * it ensures that the CoR read and write operations are atomic and
1447 * guest writes cannot interleave between them. */
1448 bdrv_make_request_serialising(req
, bdrv_get_cluster_size(bs
));
1450 bdrv_wait_serialising_requests(req
);
1453 if (flags
& BDRV_REQ_COPY_ON_READ
) {
1456 ret
= bdrv_is_allocated(bs
, offset
, bytes
, &pnum
);
1461 if (!ret
|| pnum
!= bytes
) {
1462 ret
= bdrv_co_do_copy_on_readv(child
, offset
, bytes
,
1463 qiov
, qiov_offset
, flags
);
1465 } else if (flags
& BDRV_REQ_PREFETCH
) {
1470 /* Forward the request to the BlockDriver, possibly fragmenting it */
1471 total_bytes
= bdrv_getlength(bs
);
1472 if (total_bytes
< 0) {
1477 max_bytes
= ROUND_UP(MAX(0, total_bytes
- offset
), align
);
1478 if (bytes
<= max_bytes
&& bytes
<= max_transfer
) {
1479 ret
= bdrv_driver_preadv(bs
, offset
, bytes
, qiov
, qiov_offset
, 0);
1483 while (bytes_remaining
) {
1487 num
= MIN(bytes_remaining
, MIN(max_bytes
, max_transfer
));
1490 ret
= bdrv_driver_preadv(bs
, offset
+ bytes
- bytes_remaining
,
1492 qiov_offset
+ bytes
- bytes_remaining
, 0);
1495 num
= bytes_remaining
;
1496 ret
= qemu_iovec_memset(qiov
, qiov_offset
+ bytes
- bytes_remaining
,
1497 0, bytes_remaining
);
1502 bytes_remaining
-= num
;
1506 return ret
< 0 ? ret
: 0;
1512 * |<---- align ----->| |<----- align ---->|
1513 * |<- head ->|<------------- bytes ------------->|<-- tail -->|
1515 * -*----------$-------*-------- ... --------*-----$------------*---
1517 * | offset | | end |
1518 * ALIGN_DOWN(offset) ALIGN_UP(offset) ALIGN_DOWN(end) ALIGN_UP(end)
1519 * [buf ... ) [tail_buf )
1521 * @buf is an aligned allocation needed to store @head and @tail paddings. @head
1522 * is placed at the beginning of @buf and @tail at the @end.
1524 * @tail_buf is a pointer to sub-buffer, corresponding to align-sized chunk
1525 * around tail, if tail exists.
1527 * @merge_reads is true for small requests,
1528 * if @buf_len == @head + bytes + @tail. In this case it is possible that both
1529 * head and tail exist but @buf_len == align and @tail_buf == @buf.
1531 typedef struct BdrvRequestPadding
{
1538 QEMUIOVector local_qiov
;
1539 } BdrvRequestPadding
;
1541 static bool bdrv_init_padding(BlockDriverState
*bs
,
1542 int64_t offset
, int64_t bytes
,
1543 BdrvRequestPadding
*pad
)
1545 uint64_t align
= bs
->bl
.request_alignment
;
1548 memset(pad
, 0, sizeof(*pad
));
1550 pad
->head
= offset
& (align
- 1);
1551 pad
->tail
= ((offset
+ bytes
) & (align
- 1));
1553 pad
->tail
= align
- pad
->tail
;
1556 if (!pad
->head
&& !pad
->tail
) {
1560 assert(bytes
); /* Nothing good in aligning zero-length requests */
1562 sum
= pad
->head
+ bytes
+ pad
->tail
;
1563 pad
->buf_len
= (sum
> align
&& pad
->head
&& pad
->tail
) ? 2 * align
: align
;
1564 pad
->buf
= qemu_blockalign(bs
, pad
->buf_len
);
1565 pad
->merge_reads
= sum
== pad
->buf_len
;
1567 pad
->tail_buf
= pad
->buf
+ pad
->buf_len
- align
;
1573 static int bdrv_padding_rmw_read(BdrvChild
*child
,
1574 BdrvTrackedRequest
*req
,
1575 BdrvRequestPadding
*pad
,
1578 QEMUIOVector local_qiov
;
1579 BlockDriverState
*bs
= child
->bs
;
1580 uint64_t align
= bs
->bl
.request_alignment
;
1583 assert(req
->serialising
&& pad
->buf
);
1585 if (pad
->head
|| pad
->merge_reads
) {
1586 uint64_t bytes
= pad
->merge_reads
? pad
->buf_len
: align
;
1588 qemu_iovec_init_buf(&local_qiov
, pad
->buf
, bytes
);
1591 bdrv_debug_event(bs
, BLKDBG_PWRITEV_RMW_HEAD
);
1593 if (pad
->merge_reads
&& pad
->tail
) {
1594 bdrv_debug_event(bs
, BLKDBG_PWRITEV_RMW_TAIL
);
1596 ret
= bdrv_aligned_preadv(child
, req
, req
->overlap_offset
, bytes
,
1597 align
, &local_qiov
, 0, 0);
1602 bdrv_debug_event(bs
, BLKDBG_PWRITEV_RMW_AFTER_HEAD
);
1604 if (pad
->merge_reads
&& pad
->tail
) {
1605 bdrv_debug_event(bs
, BLKDBG_PWRITEV_RMW_AFTER_TAIL
);
1608 if (pad
->merge_reads
) {
1614 qemu_iovec_init_buf(&local_qiov
, pad
->tail_buf
, align
);
1616 bdrv_debug_event(bs
, BLKDBG_PWRITEV_RMW_TAIL
);
1617 ret
= bdrv_aligned_preadv(
1619 req
->overlap_offset
+ req
->overlap_bytes
- align
,
1620 align
, align
, &local_qiov
, 0, 0);
1624 bdrv_debug_event(bs
, BLKDBG_PWRITEV_RMW_AFTER_TAIL
);
1629 memset(pad
->buf
+ pad
->head
, 0, pad
->buf_len
- pad
->head
- pad
->tail
);
1635 static void bdrv_padding_destroy(BdrvRequestPadding
*pad
)
1638 qemu_vfree(pad
->buf
);
1639 qemu_iovec_destroy(&pad
->local_qiov
);
1646 * Exchange request parameters with padded request if needed. Don't include RMW
1647 * read of padding, bdrv_padding_rmw_read() should be called separately if
1650 * All parameters except @bs are in-out: they represent original request at
1651 * function call and padded (if padding needed) at function finish.
1653 * Function always succeeds.
1655 static bool bdrv_pad_request(BlockDriverState
*bs
,
1656 QEMUIOVector
**qiov
, size_t *qiov_offset
,
1657 int64_t *offset
, unsigned int *bytes
,
1658 BdrvRequestPadding
*pad
)
1660 if (!bdrv_init_padding(bs
, *offset
, *bytes
, pad
)) {
1664 qemu_iovec_init_extended(&pad
->local_qiov
, pad
->buf
, pad
->head
,
1665 *qiov
, *qiov_offset
, *bytes
,
1666 pad
->buf
+ pad
->buf_len
- pad
->tail
, pad
->tail
);
1667 *bytes
+= pad
->head
+ pad
->tail
;
1668 *offset
-= pad
->head
;
1669 *qiov
= &pad
->local_qiov
;
1675 int coroutine_fn
bdrv_co_preadv(BdrvChild
*child
,
1676 int64_t offset
, unsigned int bytes
, QEMUIOVector
*qiov
,
1677 BdrvRequestFlags flags
)
1679 return bdrv_co_preadv_part(child
, offset
, bytes
, qiov
, 0, flags
);
1682 int coroutine_fn
bdrv_co_preadv_part(BdrvChild
*child
,
1683 int64_t offset
, unsigned int bytes
,
1684 QEMUIOVector
*qiov
, size_t qiov_offset
,
1685 BdrvRequestFlags flags
)
1687 BlockDriverState
*bs
= child
->bs
;
1688 BdrvTrackedRequest req
;
1689 BdrvRequestPadding pad
;
1692 trace_bdrv_co_preadv(bs
, offset
, bytes
, flags
);
1694 if (!bdrv_is_inserted(bs
)) {
1698 ret
= bdrv_check_request32(offset
, bytes
);
1703 if (bytes
== 0 && !QEMU_IS_ALIGNED(offset
, bs
->bl
.request_alignment
)) {
1705 * Aligning zero request is nonsense. Even if driver has special meaning
1706 * of zero-length (like qcow2_co_pwritev_compressed_part), we can't pass
1707 * it to driver due to request_alignment.
1709 * Still, no reason to return an error if someone do unaligned
1710 * zero-length read occasionally.
1715 bdrv_inc_in_flight(bs
);
1717 /* Don't do copy-on-read if we read data before write operation */
1718 if (qatomic_read(&bs
->copy_on_read
)) {
1719 flags
|= BDRV_REQ_COPY_ON_READ
;
1722 bdrv_pad_request(bs
, &qiov
, &qiov_offset
, &offset
, &bytes
, &pad
);
1724 tracked_request_begin(&req
, bs
, offset
, bytes
, BDRV_TRACKED_READ
);
1725 ret
= bdrv_aligned_preadv(child
, &req
, offset
, bytes
,
1726 bs
->bl
.request_alignment
,
1727 qiov
, qiov_offset
, flags
);
1728 tracked_request_end(&req
);
1729 bdrv_dec_in_flight(bs
);
1731 bdrv_padding_destroy(&pad
);
1736 static int coroutine_fn
bdrv_co_do_pwrite_zeroes(BlockDriverState
*bs
,
1737 int64_t offset
, int bytes
, BdrvRequestFlags flags
)
1739 BlockDriver
*drv
= bs
->drv
;
1743 bool need_flush
= false;
1747 int max_write_zeroes
= MIN_NON_ZERO(bs
->bl
.max_pwrite_zeroes
, INT_MAX
);
1748 int alignment
= MAX(bs
->bl
.pwrite_zeroes_alignment
,
1749 bs
->bl
.request_alignment
);
1750 int max_transfer
= MIN_NON_ZERO(bs
->bl
.max_transfer
, MAX_BOUNCE_BUFFER
);
1756 if ((flags
& ~bs
->supported_zero_flags
) & BDRV_REQ_NO_FALLBACK
) {
1760 assert(alignment
% bs
->bl
.request_alignment
== 0);
1761 head
= offset
% alignment
;
1762 tail
= (offset
+ bytes
) % alignment
;
1763 max_write_zeroes
= QEMU_ALIGN_DOWN(max_write_zeroes
, alignment
);
1764 assert(max_write_zeroes
>= bs
->bl
.request_alignment
);
1766 while (bytes
> 0 && !ret
) {
1769 /* Align request. Block drivers can expect the "bulk" of the request
1770 * to be aligned, and that unaligned requests do not cross cluster
1774 /* Make a small request up to the first aligned sector. For
1775 * convenience, limit this request to max_transfer even if
1776 * we don't need to fall back to writes. */
1777 num
= MIN(MIN(bytes
, max_transfer
), alignment
- head
);
1778 head
= (head
+ num
) % alignment
;
1779 assert(num
< max_write_zeroes
);
1780 } else if (tail
&& num
> alignment
) {
1781 /* Shorten the request to the last aligned sector. */
1785 /* limit request size */
1786 if (num
> max_write_zeroes
) {
1787 num
= max_write_zeroes
;
1791 /* First try the efficient write zeroes operation */
1792 if (drv
->bdrv_co_pwrite_zeroes
) {
1793 ret
= drv
->bdrv_co_pwrite_zeroes(bs
, offset
, num
,
1794 flags
& bs
->supported_zero_flags
);
1795 if (ret
!= -ENOTSUP
&& (flags
& BDRV_REQ_FUA
) &&
1796 !(bs
->supported_zero_flags
& BDRV_REQ_FUA
)) {
1800 assert(!bs
->supported_zero_flags
);
1803 if (ret
== -ENOTSUP
&& !(flags
& BDRV_REQ_NO_FALLBACK
)) {
1804 /* Fall back to bounce buffer if write zeroes is unsupported */
1805 BdrvRequestFlags write_flags
= flags
& ~BDRV_REQ_ZERO_WRITE
;
1807 if ((flags
& BDRV_REQ_FUA
) &&
1808 !(bs
->supported_write_flags
& BDRV_REQ_FUA
)) {
1809 /* No need for bdrv_driver_pwrite() to do a fallback
1810 * flush on each chunk; use just one at the end */
1811 write_flags
&= ~BDRV_REQ_FUA
;
1814 num
= MIN(num
, max_transfer
);
1816 buf
= qemu_try_blockalign0(bs
, num
);
1822 qemu_iovec_init_buf(&qiov
, buf
, num
);
1824 ret
= bdrv_driver_pwritev(bs
, offset
, num
, &qiov
, 0, write_flags
);
1826 /* Keep bounce buffer around if it is big enough for all
1827 * all future requests.
1829 if (num
< max_transfer
) {
1840 if (ret
== 0 && need_flush
) {
1841 ret
= bdrv_co_flush(bs
);
1847 static inline int coroutine_fn
1848 bdrv_co_write_req_prepare(BdrvChild
*child
, int64_t offset
, uint64_t bytes
,
1849 BdrvTrackedRequest
*req
, int flags
)
1851 BlockDriverState
*bs
= child
->bs
;
1852 int64_t end_sector
= DIV_ROUND_UP(offset
+ bytes
, BDRV_SECTOR_SIZE
);
1854 if (bs
->read_only
) {
1858 assert(!(bs
->open_flags
& BDRV_O_INACTIVE
));
1859 assert((bs
->open_flags
& BDRV_O_NO_IO
) == 0);
1860 assert(!(flags
& ~BDRV_REQ_MASK
));
1861 assert(!((flags
& BDRV_REQ_NO_WAIT
) && !(flags
& BDRV_REQ_SERIALISING
)));
1863 if (flags
& BDRV_REQ_SERIALISING
) {
1864 QEMU_LOCK_GUARD(&bs
->reqs_lock
);
1866 tracked_request_set_serialising(req
, bdrv_get_cluster_size(bs
));
1868 if ((flags
& BDRV_REQ_NO_WAIT
) && bdrv_find_conflicting_request(req
)) {
1872 bdrv_wait_serialising_requests_locked(req
);
1874 bdrv_wait_serialising_requests(req
);
1877 assert(req
->overlap_offset
<= offset
);
1878 assert(offset
+ bytes
<= req
->overlap_offset
+ req
->overlap_bytes
);
1879 assert(end_sector
<= bs
->total_sectors
|| child
->perm
& BLK_PERM_RESIZE
);
1881 switch (req
->type
) {
1882 case BDRV_TRACKED_WRITE
:
1883 case BDRV_TRACKED_DISCARD
:
1884 if (flags
& BDRV_REQ_WRITE_UNCHANGED
) {
1885 assert(child
->perm
& (BLK_PERM_WRITE_UNCHANGED
| BLK_PERM_WRITE
));
1887 assert(child
->perm
& BLK_PERM_WRITE
);
1889 return notifier_with_return_list_notify(&bs
->before_write_notifiers
,
1891 case BDRV_TRACKED_TRUNCATE
:
1892 assert(child
->perm
& BLK_PERM_RESIZE
);
1899 static inline void coroutine_fn
1900 bdrv_co_write_req_finish(BdrvChild
*child
, int64_t offset
, uint64_t bytes
,
1901 BdrvTrackedRequest
*req
, int ret
)
1903 int64_t end_sector
= DIV_ROUND_UP(offset
+ bytes
, BDRV_SECTOR_SIZE
);
1904 BlockDriverState
*bs
= child
->bs
;
1906 qatomic_inc(&bs
->write_gen
);
1909 * Discard cannot extend the image, but in error handling cases, such as
1910 * when reverting a qcow2 cluster allocation, the discarded range can pass
1911 * the end of image file, so we cannot assert about BDRV_TRACKED_DISCARD
1912 * here. Instead, just skip it, since semantically a discard request
1913 * beyond EOF cannot expand the image anyway.
1916 (req
->type
== BDRV_TRACKED_TRUNCATE
||
1917 end_sector
> bs
->total_sectors
) &&
1918 req
->type
!= BDRV_TRACKED_DISCARD
) {
1919 bs
->total_sectors
= end_sector
;
1920 bdrv_parent_cb_resize(bs
);
1921 bdrv_dirty_bitmap_truncate(bs
, end_sector
<< BDRV_SECTOR_BITS
);
1924 switch (req
->type
) {
1925 case BDRV_TRACKED_WRITE
:
1926 stat64_max(&bs
->wr_highest_offset
, offset
+ bytes
);
1927 /* fall through, to set dirty bits */
1928 case BDRV_TRACKED_DISCARD
:
1929 bdrv_set_dirty(bs
, offset
, bytes
);
1938 * Forwards an already correctly aligned write request to the BlockDriver,
1939 * after possibly fragmenting it.
1941 static int coroutine_fn
bdrv_aligned_pwritev(BdrvChild
*child
,
1942 BdrvTrackedRequest
*req
, int64_t offset
, unsigned int bytes
,
1943 int64_t align
, QEMUIOVector
*qiov
, size_t qiov_offset
, int flags
)
1945 BlockDriverState
*bs
= child
->bs
;
1946 BlockDriver
*drv
= bs
->drv
;
1949 uint64_t bytes_remaining
= bytes
;
1956 if (bdrv_has_readonly_bitmaps(bs
)) {
1960 assert(is_power_of_2(align
));
1961 assert((offset
& (align
- 1)) == 0);
1962 assert((bytes
& (align
- 1)) == 0);
1963 assert(!qiov
|| qiov_offset
+ bytes
<= qiov
->size
);
1964 max_transfer
= QEMU_ALIGN_DOWN(MIN_NON_ZERO(bs
->bl
.max_transfer
, INT_MAX
),
1967 ret
= bdrv_co_write_req_prepare(child
, offset
, bytes
, req
, flags
);
1969 if (!ret
&& bs
->detect_zeroes
!= BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF
&&
1970 !(flags
& BDRV_REQ_ZERO_WRITE
) && drv
->bdrv_co_pwrite_zeroes
&&
1971 qemu_iovec_is_zero(qiov
, qiov_offset
, bytes
)) {
1972 flags
|= BDRV_REQ_ZERO_WRITE
;
1973 if (bs
->detect_zeroes
== BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP
) {
1974 flags
|= BDRV_REQ_MAY_UNMAP
;
1979 /* Do nothing, write notifier decided to fail this request */
1980 } else if (flags
& BDRV_REQ_ZERO_WRITE
) {
1981 bdrv_debug_event(bs
, BLKDBG_PWRITEV_ZERO
);
1982 ret
= bdrv_co_do_pwrite_zeroes(bs
, offset
, bytes
, flags
);
1983 } else if (flags
& BDRV_REQ_WRITE_COMPRESSED
) {
1984 ret
= bdrv_driver_pwritev_compressed(bs
, offset
, bytes
,
1986 } else if (bytes
<= max_transfer
) {
1987 bdrv_debug_event(bs
, BLKDBG_PWRITEV
);
1988 ret
= bdrv_driver_pwritev(bs
, offset
, bytes
, qiov
, qiov_offset
, flags
);
1990 bdrv_debug_event(bs
, BLKDBG_PWRITEV
);
1991 while (bytes_remaining
) {
1992 int num
= MIN(bytes_remaining
, max_transfer
);
1993 int local_flags
= flags
;
1996 if (num
< bytes_remaining
&& (flags
& BDRV_REQ_FUA
) &&
1997 !(bs
->supported_write_flags
& BDRV_REQ_FUA
)) {
1998 /* If FUA is going to be emulated by flush, we only
1999 * need to flush on the last iteration */
2000 local_flags
&= ~BDRV_REQ_FUA
;
2003 ret
= bdrv_driver_pwritev(bs
, offset
+ bytes
- bytes_remaining
,
2005 qiov_offset
+ bytes
- bytes_remaining
,
2010 bytes_remaining
-= num
;
2013 bdrv_debug_event(bs
, BLKDBG_PWRITEV_DONE
);
2018 bdrv_co_write_req_finish(child
, offset
, bytes
, req
, ret
);
2023 static int coroutine_fn
bdrv_co_do_zero_pwritev(BdrvChild
*child
,
2026 BdrvRequestFlags flags
,
2027 BdrvTrackedRequest
*req
)
2029 BlockDriverState
*bs
= child
->bs
;
2030 QEMUIOVector local_qiov
;
2031 uint64_t align
= bs
->bl
.request_alignment
;
2034 BdrvRequestPadding pad
;
2036 padding
= bdrv_init_padding(bs
, offset
, bytes
, &pad
);
2038 bdrv_make_request_serialising(req
, align
);
2040 bdrv_padding_rmw_read(child
, req
, &pad
, true);
2042 if (pad
.head
|| pad
.merge_reads
) {
2043 int64_t aligned_offset
= offset
& ~(align
- 1);
2044 int64_t write_bytes
= pad
.merge_reads
? pad
.buf_len
: align
;
2046 qemu_iovec_init_buf(&local_qiov
, pad
.buf
, write_bytes
);
2047 ret
= bdrv_aligned_pwritev(child
, req
, aligned_offset
, write_bytes
,
2048 align
, &local_qiov
, 0,
2049 flags
& ~BDRV_REQ_ZERO_WRITE
);
2050 if (ret
< 0 || pad
.merge_reads
) {
2051 /* Error or all work is done */
2054 offset
+= write_bytes
- pad
.head
;
2055 bytes
-= write_bytes
- pad
.head
;
2059 assert(!bytes
|| (offset
& (align
- 1)) == 0);
2060 if (bytes
>= align
) {
2061 /* Write the aligned part in the middle. */
2062 uint64_t aligned_bytes
= bytes
& ~(align
- 1);
2063 ret
= bdrv_aligned_pwritev(child
, req
, offset
, aligned_bytes
, align
,
2068 bytes
-= aligned_bytes
;
2069 offset
+= aligned_bytes
;
2072 assert(!bytes
|| (offset
& (align
- 1)) == 0);
2074 assert(align
== pad
.tail
+ bytes
);
2076 qemu_iovec_init_buf(&local_qiov
, pad
.tail_buf
, align
);
2077 ret
= bdrv_aligned_pwritev(child
, req
, offset
, align
, align
,
2079 flags
& ~BDRV_REQ_ZERO_WRITE
);
2083 bdrv_padding_destroy(&pad
);
2089 * Handle a write request in coroutine context
2091 int coroutine_fn
bdrv_co_pwritev(BdrvChild
*child
,
2092 int64_t offset
, unsigned int bytes
, QEMUIOVector
*qiov
,
2093 BdrvRequestFlags flags
)
2095 return bdrv_co_pwritev_part(child
, offset
, bytes
, qiov
, 0, flags
);
2098 int coroutine_fn
bdrv_co_pwritev_part(BdrvChild
*child
,
2099 int64_t offset
, unsigned int bytes
, QEMUIOVector
*qiov
, size_t qiov_offset
,
2100 BdrvRequestFlags flags
)
2102 BlockDriverState
*bs
= child
->bs
;
2103 BdrvTrackedRequest req
;
2104 uint64_t align
= bs
->bl
.request_alignment
;
2105 BdrvRequestPadding pad
;
2108 trace_bdrv_co_pwritev(child
->bs
, offset
, bytes
, flags
);
2110 if (!bdrv_is_inserted(bs
)) {
2114 ret
= bdrv_check_request32(offset
, bytes
);
2119 /* If the request is misaligned then we can't make it efficient */
2120 if ((flags
& BDRV_REQ_NO_FALLBACK
) &&
2121 !QEMU_IS_ALIGNED(offset
| bytes
, align
))
2126 if (bytes
== 0 && !QEMU_IS_ALIGNED(offset
, bs
->bl
.request_alignment
)) {
2128 * Aligning zero request is nonsense. Even if driver has special meaning
2129 * of zero-length (like qcow2_co_pwritev_compressed_part), we can't pass
2130 * it to driver due to request_alignment.
2132 * Still, no reason to return an error if someone do unaligned
2133 * zero-length write occasionally.
2138 bdrv_inc_in_flight(bs
);
2140 * Align write if necessary by performing a read-modify-write cycle.
2141 * Pad qiov with the read parts and be sure to have a tracked request not
2142 * only for bdrv_aligned_pwritev, but also for the reads of the RMW cycle.
2144 tracked_request_begin(&req
, bs
, offset
, bytes
, BDRV_TRACKED_WRITE
);
2146 if (flags
& BDRV_REQ_ZERO_WRITE
) {
2147 ret
= bdrv_co_do_zero_pwritev(child
, offset
, bytes
, flags
, &req
);
2151 if (bdrv_pad_request(bs
, &qiov
, &qiov_offset
, &offset
, &bytes
, &pad
)) {
2152 bdrv_make_request_serialising(&req
, align
);
2153 bdrv_padding_rmw_read(child
, &req
, &pad
, false);
2156 ret
= bdrv_aligned_pwritev(child
, &req
, offset
, bytes
, align
,
2157 qiov
, qiov_offset
, flags
);
2159 bdrv_padding_destroy(&pad
);
2162 tracked_request_end(&req
);
2163 bdrv_dec_in_flight(bs
);
2168 int coroutine_fn
bdrv_co_pwrite_zeroes(BdrvChild
*child
, int64_t offset
,
2169 int bytes
, BdrvRequestFlags flags
)
2171 trace_bdrv_co_pwrite_zeroes(child
->bs
, offset
, bytes
, flags
);
2173 if (!(child
->bs
->open_flags
& BDRV_O_UNMAP
)) {
2174 flags
&= ~BDRV_REQ_MAY_UNMAP
;
2177 return bdrv_co_pwritev(child
, offset
, bytes
, NULL
,
2178 BDRV_REQ_ZERO_WRITE
| flags
);
2182 * Flush ALL BDSes regardless of if they are reachable via a BlkBackend or not.
2184 int bdrv_flush_all(void)
2186 BdrvNextIterator it
;
2187 BlockDriverState
*bs
= NULL
;
2191 * bdrv queue is managed by record/replay,
2192 * creating new flush request for stopping
2193 * the VM may break the determinism
2195 if (replay_events_enabled()) {
2199 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
2200 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
2203 aio_context_acquire(aio_context
);
2204 ret
= bdrv_flush(bs
);
2205 if (ret
< 0 && !result
) {
2208 aio_context_release(aio_context
);
2215 * Returns the allocation status of the specified sectors.
2216 * Drivers not implementing the functionality are assumed to not support
2217 * backing files, hence all their sectors are reported as allocated.
2219 * If 'want_zero' is true, the caller is querying for mapping
2220 * purposes, with a focus on valid BDRV_BLOCK_OFFSET_VALID, _DATA, and
2221 * _ZERO where possible; otherwise, the result favors larger 'pnum',
2222 * with a focus on accurate BDRV_BLOCK_ALLOCATED.
2224 * If 'offset' is beyond the end of the disk image the return value is
2225 * BDRV_BLOCK_EOF and 'pnum' is set to 0.
2227 * 'bytes' is the max value 'pnum' should be set to. If bytes goes
2228 * beyond the end of the disk image it will be clamped; if 'pnum' is set to
2229 * the end of the image, then the returned value will include BDRV_BLOCK_EOF.
2231 * 'pnum' is set to the number of bytes (including and immediately
2232 * following the specified offset) that are easily known to be in the
2233 * same allocated/unallocated state. Note that a second call starting
2234 * at the original offset plus returned pnum may have the same status.
2235 * The returned value is non-zero on success except at end-of-file.
2237 * Returns negative errno on failure. Otherwise, if the
2238 * BDRV_BLOCK_OFFSET_VALID bit is set, 'map' and 'file' (if non-NULL) are
2239 * set to the host mapping and BDS corresponding to the guest offset.
2241 static int coroutine_fn
bdrv_co_block_status(BlockDriverState
*bs
,
2243 int64_t offset
, int64_t bytes
,
2244 int64_t *pnum
, int64_t *map
,
2245 BlockDriverState
**file
)
2248 int64_t n
; /* bytes */
2250 int64_t local_map
= 0;
2251 BlockDriverState
*local_file
= NULL
;
2252 int64_t aligned_offset
, aligned_bytes
;
2254 bool has_filtered_child
;
2258 total_size
= bdrv_getlength(bs
);
2259 if (total_size
< 0) {
2264 if (offset
>= total_size
) {
2265 ret
= BDRV_BLOCK_EOF
;
2273 n
= total_size
- offset
;
2278 /* Must be non-NULL or bdrv_getlength() would have failed */
2280 has_filtered_child
= bdrv_filter_child(bs
);
2281 if (!bs
->drv
->bdrv_co_block_status
&& !has_filtered_child
) {
2283 ret
= BDRV_BLOCK_DATA
| BDRV_BLOCK_ALLOCATED
;
2284 if (offset
+ bytes
== total_size
) {
2285 ret
|= BDRV_BLOCK_EOF
;
2287 if (bs
->drv
->protocol_name
) {
2288 ret
|= BDRV_BLOCK_OFFSET_VALID
;
2295 bdrv_inc_in_flight(bs
);
2297 /* Round out to request_alignment boundaries */
2298 align
= bs
->bl
.request_alignment
;
2299 aligned_offset
= QEMU_ALIGN_DOWN(offset
, align
);
2300 aligned_bytes
= ROUND_UP(offset
+ bytes
, align
) - aligned_offset
;
2302 if (bs
->drv
->bdrv_co_block_status
) {
2303 ret
= bs
->drv
->bdrv_co_block_status(bs
, want_zero
, aligned_offset
,
2304 aligned_bytes
, pnum
, &local_map
,
2307 /* Default code for filters */
2309 local_file
= bdrv_filter_bs(bs
);
2312 *pnum
= aligned_bytes
;
2313 local_map
= aligned_offset
;
2314 ret
= BDRV_BLOCK_RAW
| BDRV_BLOCK_OFFSET_VALID
;
2322 * The driver's result must be a non-zero multiple of request_alignment.
2323 * Clamp pnum and adjust map to original request.
2325 assert(*pnum
&& QEMU_IS_ALIGNED(*pnum
, align
) &&
2326 align
> offset
- aligned_offset
);
2327 if (ret
& BDRV_BLOCK_RECURSE
) {
2328 assert(ret
& BDRV_BLOCK_DATA
);
2329 assert(ret
& BDRV_BLOCK_OFFSET_VALID
);
2330 assert(!(ret
& BDRV_BLOCK_ZERO
));
2333 *pnum
-= offset
- aligned_offset
;
2334 if (*pnum
> bytes
) {
2337 if (ret
& BDRV_BLOCK_OFFSET_VALID
) {
2338 local_map
+= offset
- aligned_offset
;
2341 if (ret
& BDRV_BLOCK_RAW
) {
2342 assert(ret
& BDRV_BLOCK_OFFSET_VALID
&& local_file
);
2343 ret
= bdrv_co_block_status(local_file
, want_zero
, local_map
,
2344 *pnum
, pnum
, &local_map
, &local_file
);
2348 if (ret
& (BDRV_BLOCK_DATA
| BDRV_BLOCK_ZERO
)) {
2349 ret
|= BDRV_BLOCK_ALLOCATED
;
2350 } else if (bs
->drv
->supports_backing
) {
2351 BlockDriverState
*cow_bs
= bdrv_cow_bs(bs
);
2354 ret
|= BDRV_BLOCK_ZERO
;
2355 } else if (want_zero
) {
2356 int64_t size2
= bdrv_getlength(cow_bs
);
2358 if (size2
>= 0 && offset
>= size2
) {
2359 ret
|= BDRV_BLOCK_ZERO
;
2364 if (want_zero
&& ret
& BDRV_BLOCK_RECURSE
&&
2365 local_file
&& local_file
!= bs
&&
2366 (ret
& BDRV_BLOCK_DATA
) && !(ret
& BDRV_BLOCK_ZERO
) &&
2367 (ret
& BDRV_BLOCK_OFFSET_VALID
)) {
2371 ret2
= bdrv_co_block_status(local_file
, want_zero
, local_map
,
2372 *pnum
, &file_pnum
, NULL
, NULL
);
2374 /* Ignore errors. This is just providing extra information, it
2375 * is useful but not necessary.
2377 if (ret2
& BDRV_BLOCK_EOF
&&
2378 (!file_pnum
|| ret2
& BDRV_BLOCK_ZERO
)) {
2380 * It is valid for the format block driver to read
2381 * beyond the end of the underlying file's current
2382 * size; such areas read as zero.
2384 ret
|= BDRV_BLOCK_ZERO
;
2386 /* Limit request to the range reported by the protocol driver */
2388 ret
|= (ret2
& BDRV_BLOCK_ZERO
);
2394 bdrv_dec_in_flight(bs
);
2395 if (ret
>= 0 && offset
+ *pnum
== total_size
) {
2396 ret
|= BDRV_BLOCK_EOF
;
2409 bdrv_co_common_block_status_above(BlockDriverState
*bs
,
2410 BlockDriverState
*base
,
2417 BlockDriverState
**file
,
2421 BlockDriverState
*p
;
2425 assert(!include_base
|| base
); /* Can't include NULL base */
2432 if (!include_base
&& bs
== base
) {
2437 ret
= bdrv_co_block_status(bs
, want_zero
, offset
, bytes
, pnum
, map
, file
);
2439 if (ret
< 0 || *pnum
== 0 || ret
& BDRV_BLOCK_ALLOCATED
|| bs
== base
) {
2443 if (ret
& BDRV_BLOCK_EOF
) {
2444 eof
= offset
+ *pnum
;
2447 assert(*pnum
<= bytes
);
2450 for (p
= bdrv_filter_or_cow_bs(bs
); include_base
|| p
!= base
;
2451 p
= bdrv_filter_or_cow_bs(p
))
2453 ret
= bdrv_co_block_status(p
, want_zero
, offset
, bytes
, pnum
, map
,
2461 * The top layer deferred to this layer, and because this layer is
2462 * short, any zeroes that we synthesize beyond EOF behave as if they
2463 * were allocated at this layer.
2465 * We don't include BDRV_BLOCK_EOF into ret, as upper layer may be
2466 * larger. We'll add BDRV_BLOCK_EOF if needed at function end, see
2469 assert(ret
& BDRV_BLOCK_EOF
);
2474 ret
= BDRV_BLOCK_ZERO
| BDRV_BLOCK_ALLOCATED
;
2477 if (ret
& BDRV_BLOCK_ALLOCATED
) {
2479 * We've found the node and the status, we must break.
2481 * Drop BDRV_BLOCK_EOF, as it's not for upper layer, which may be
2482 * larger. We'll add BDRV_BLOCK_EOF if needed at function end, see
2485 ret
&= ~BDRV_BLOCK_EOF
;
2490 assert(include_base
);
2495 * OK, [offset, offset + *pnum) region is unallocated on this layer,
2496 * let's continue the diving.
2498 assert(*pnum
<= bytes
);
2502 if (offset
+ *pnum
== eof
) {
2503 ret
|= BDRV_BLOCK_EOF
;
2509 int bdrv_block_status_above(BlockDriverState
*bs
, BlockDriverState
*base
,
2510 int64_t offset
, int64_t bytes
, int64_t *pnum
,
2511 int64_t *map
, BlockDriverState
**file
)
2513 return bdrv_common_block_status_above(bs
, base
, false, true, offset
, bytes
,
2514 pnum
, map
, file
, NULL
);
2517 int bdrv_block_status(BlockDriverState
*bs
, int64_t offset
, int64_t bytes
,
2518 int64_t *pnum
, int64_t *map
, BlockDriverState
**file
)
2520 return bdrv_block_status_above(bs
, bdrv_filter_or_cow_bs(bs
),
2521 offset
, bytes
, pnum
, map
, file
);
2525 * Check @bs (and its backing chain) to see if the range defined
2526 * by @offset and @bytes is known to read as zeroes.
2527 * Return 1 if that is the case, 0 otherwise and -errno on error.
2528 * This test is meant to be fast rather than accurate so returning 0
2529 * does not guarantee non-zero data.
2531 int coroutine_fn
bdrv_co_is_zero_fast(BlockDriverState
*bs
, int64_t offset
,
2535 int64_t pnum
= bytes
;
2541 ret
= bdrv_common_block_status_above(bs
, NULL
, false, false, offset
,
2542 bytes
, &pnum
, NULL
, NULL
, NULL
);
2548 return (pnum
== bytes
) && (ret
& BDRV_BLOCK_ZERO
);
2551 int coroutine_fn
bdrv_is_allocated(BlockDriverState
*bs
, int64_t offset
,
2552 int64_t bytes
, int64_t *pnum
)
2557 ret
= bdrv_common_block_status_above(bs
, bs
, true, false, offset
,
2558 bytes
, pnum
? pnum
: &dummy
, NULL
,
2563 return !!(ret
& BDRV_BLOCK_ALLOCATED
);
2567 * Given an image chain: ... -> [BASE] -> [INTER1] -> [INTER2] -> [TOP]
2569 * Return a positive depth if (a prefix of) the given range is allocated
2570 * in any image between BASE and TOP (BASE is only included if include_base
2571 * is set). Depth 1 is TOP, 2 is the first backing layer, and so forth.
2572 * BASE can be NULL to check if the given offset is allocated in any
2573 * image of the chain. Return 0 otherwise, or negative errno on
2576 * 'pnum' is set to the number of bytes (including and immediately
2577 * following the specified offset) that are known to be in the same
2578 * allocated/unallocated state. Note that a subsequent call starting
2579 * at 'offset + *pnum' may return the same allocation status (in other
2580 * words, the result is not necessarily the maximum possible range);
2581 * but 'pnum' will only be 0 when end of file is reached.
2583 int bdrv_is_allocated_above(BlockDriverState
*top
,
2584 BlockDriverState
*base
,
2585 bool include_base
, int64_t offset
,
2586 int64_t bytes
, int64_t *pnum
)
2589 int ret
= bdrv_common_block_status_above(top
, base
, include_base
, false,
2590 offset
, bytes
, pnum
, NULL
, NULL
,
2596 if (ret
& BDRV_BLOCK_ALLOCATED
) {
2603 bdrv_co_readv_vmstate(BlockDriverState
*bs
, QEMUIOVector
*qiov
, int64_t pos
)
2605 BlockDriver
*drv
= bs
->drv
;
2606 BlockDriverState
*child_bs
= bdrv_primary_bs(bs
);
2613 bdrv_inc_in_flight(bs
);
2615 if (drv
->bdrv_load_vmstate
) {
2616 ret
= drv
->bdrv_load_vmstate(bs
, qiov
, pos
);
2617 } else if (child_bs
) {
2618 ret
= bdrv_co_readv_vmstate(child_bs
, qiov
, pos
);
2621 bdrv_dec_in_flight(bs
);
2627 bdrv_co_writev_vmstate(BlockDriverState
*bs
, QEMUIOVector
*qiov
, int64_t pos
)
2629 BlockDriver
*drv
= bs
->drv
;
2630 BlockDriverState
*child_bs
= bdrv_primary_bs(bs
);
2637 bdrv_inc_in_flight(bs
);
2639 if (drv
->bdrv_save_vmstate
) {
2640 ret
= drv
->bdrv_save_vmstate(bs
, qiov
, pos
);
2641 } else if (child_bs
) {
2642 ret
= bdrv_co_writev_vmstate(child_bs
, qiov
, pos
);
2645 bdrv_dec_in_flight(bs
);
2650 int bdrv_save_vmstate(BlockDriverState
*bs
, const uint8_t *buf
,
2651 int64_t pos
, int size
)
2653 QEMUIOVector qiov
= QEMU_IOVEC_INIT_BUF(qiov
, buf
, size
);
2654 int ret
= bdrv_writev_vmstate(bs
, &qiov
, pos
);
2656 return ret
< 0 ? ret
: size
;
2659 int bdrv_load_vmstate(BlockDriverState
*bs
, uint8_t *buf
,
2660 int64_t pos
, int size
)
2662 QEMUIOVector qiov
= QEMU_IOVEC_INIT_BUF(qiov
, buf
, size
);
2663 int ret
= bdrv_readv_vmstate(bs
, &qiov
, pos
);
2665 return ret
< 0 ? ret
: size
;
2668 /**************************************************************/
2671 void bdrv_aio_cancel(BlockAIOCB
*acb
)
2674 bdrv_aio_cancel_async(acb
);
2675 while (acb
->refcnt
> 1) {
2676 if (acb
->aiocb_info
->get_aio_context
) {
2677 aio_poll(acb
->aiocb_info
->get_aio_context(acb
), true);
2678 } else if (acb
->bs
) {
2679 /* qemu_aio_ref and qemu_aio_unref are not thread-safe, so
2680 * assert that we're not using an I/O thread. Thread-safe
2681 * code should use bdrv_aio_cancel_async exclusively.
2683 assert(bdrv_get_aio_context(acb
->bs
) == qemu_get_aio_context());
2684 aio_poll(bdrv_get_aio_context(acb
->bs
), true);
2689 qemu_aio_unref(acb
);
2692 /* Async version of aio cancel. The caller is not blocked if the acb implements
2693 * cancel_async, otherwise we do nothing and let the request normally complete.
2694 * In either case the completion callback must be called. */
2695 void bdrv_aio_cancel_async(BlockAIOCB
*acb
)
2697 if (acb
->aiocb_info
->cancel_async
) {
2698 acb
->aiocb_info
->cancel_async(acb
);
2702 /**************************************************************/
2703 /* Coroutine block device emulation */
2705 int coroutine_fn
bdrv_co_flush(BlockDriverState
*bs
)
2707 BdrvChild
*primary_child
= bdrv_primary_child(bs
);
2712 bdrv_inc_in_flight(bs
);
2714 if (!bdrv_is_inserted(bs
) || bdrv_is_read_only(bs
) ||
2719 qemu_co_mutex_lock(&bs
->reqs_lock
);
2720 current_gen
= qatomic_read(&bs
->write_gen
);
2722 /* Wait until any previous flushes are completed */
2723 while (bs
->active_flush_req
) {
2724 qemu_co_queue_wait(&bs
->flush_queue
, &bs
->reqs_lock
);
2727 /* Flushes reach this point in nondecreasing current_gen order. */
2728 bs
->active_flush_req
= true;
2729 qemu_co_mutex_unlock(&bs
->reqs_lock
);
2731 /* Write back all layers by calling one driver function */
2732 if (bs
->drv
->bdrv_co_flush
) {
2733 ret
= bs
->drv
->bdrv_co_flush(bs
);
2737 /* Write back cached data to the OS even with cache=unsafe */
2738 BLKDBG_EVENT(primary_child
, BLKDBG_FLUSH_TO_OS
);
2739 if (bs
->drv
->bdrv_co_flush_to_os
) {
2740 ret
= bs
->drv
->bdrv_co_flush_to_os(bs
);
2746 /* But don't actually force it to the disk with cache=unsafe */
2747 if (bs
->open_flags
& BDRV_O_NO_FLUSH
) {
2748 goto flush_children
;
2751 /* Check if we really need to flush anything */
2752 if (bs
->flushed_gen
== current_gen
) {
2753 goto flush_children
;
2756 BLKDBG_EVENT(primary_child
, BLKDBG_FLUSH_TO_DISK
);
2758 /* bs->drv->bdrv_co_flush() might have ejected the BDS
2759 * (even in case of apparent success) */
2763 if (bs
->drv
->bdrv_co_flush_to_disk
) {
2764 ret
= bs
->drv
->bdrv_co_flush_to_disk(bs
);
2765 } else if (bs
->drv
->bdrv_aio_flush
) {
2767 CoroutineIOCompletion co
= {
2768 .coroutine
= qemu_coroutine_self(),
2771 acb
= bs
->drv
->bdrv_aio_flush(bs
, bdrv_co_io_em_complete
, &co
);
2775 qemu_coroutine_yield();
2780 * Some block drivers always operate in either writethrough or unsafe
2781 * mode and don't support bdrv_flush therefore. Usually qemu doesn't
2782 * know how the server works (because the behaviour is hardcoded or
2783 * depends on server-side configuration), so we can't ensure that
2784 * everything is safe on disk. Returning an error doesn't work because
2785 * that would break guests even if the server operates in writethrough
2788 * Let's hope the user knows what he's doing.
2797 /* Now flush the underlying protocol. It will also have BDRV_O_NO_FLUSH
2798 * in the case of cache=unsafe, so there are no useless flushes.
2802 QLIST_FOREACH(child
, &bs
->children
, next
) {
2803 if (child
->perm
& (BLK_PERM_WRITE
| BLK_PERM_WRITE_UNCHANGED
)) {
2804 int this_child_ret
= bdrv_co_flush(child
->bs
);
2806 ret
= this_child_ret
;
2812 /* Notify any pending flushes that we have completed */
2814 bs
->flushed_gen
= current_gen
;
2817 qemu_co_mutex_lock(&bs
->reqs_lock
);
2818 bs
->active_flush_req
= false;
2819 /* Return value is ignored - it's ok if wait queue is empty */
2820 qemu_co_queue_next(&bs
->flush_queue
);
2821 qemu_co_mutex_unlock(&bs
->reqs_lock
);
2824 bdrv_dec_in_flight(bs
);
2828 int coroutine_fn
bdrv_co_pdiscard(BdrvChild
*child
, int64_t offset
,
2831 BdrvTrackedRequest req
;
2832 int max_pdiscard
, ret
;
2833 int head
, tail
, align
;
2834 BlockDriverState
*bs
= child
->bs
;
2836 if (!bs
|| !bs
->drv
|| !bdrv_is_inserted(bs
)) {
2840 if (bdrv_has_readonly_bitmaps(bs
)) {
2844 ret
= bdrv_check_request(offset
, bytes
);
2849 /* Do nothing if disabled. */
2850 if (!(bs
->open_flags
& BDRV_O_UNMAP
)) {
2854 if (!bs
->drv
->bdrv_co_pdiscard
&& !bs
->drv
->bdrv_aio_pdiscard
) {
2858 /* Discard is advisory, but some devices track and coalesce
2859 * unaligned requests, so we must pass everything down rather than
2860 * round here. Still, most devices will just silently ignore
2861 * unaligned requests (by returning -ENOTSUP), so we must fragment
2862 * the request accordingly. */
2863 align
= MAX(bs
->bl
.pdiscard_alignment
, bs
->bl
.request_alignment
);
2864 assert(align
% bs
->bl
.request_alignment
== 0);
2865 head
= offset
% align
;
2866 tail
= (offset
+ bytes
) % align
;
2868 bdrv_inc_in_flight(bs
);
2869 tracked_request_begin(&req
, bs
, offset
, bytes
, BDRV_TRACKED_DISCARD
);
2871 ret
= bdrv_co_write_req_prepare(child
, offset
, bytes
, &req
, 0);
2876 max_pdiscard
= QEMU_ALIGN_DOWN(MIN_NON_ZERO(bs
->bl
.max_pdiscard
, INT_MAX
),
2878 assert(max_pdiscard
>= bs
->bl
.request_alignment
);
2881 int64_t num
= bytes
;
2884 /* Make small requests to get to alignment boundaries. */
2885 num
= MIN(bytes
, align
- head
);
2886 if (!QEMU_IS_ALIGNED(num
, bs
->bl
.request_alignment
)) {
2887 num
%= bs
->bl
.request_alignment
;
2889 head
= (head
+ num
) % align
;
2890 assert(num
< max_pdiscard
);
2893 /* Shorten the request to the last aligned cluster. */
2895 } else if (!QEMU_IS_ALIGNED(tail
, bs
->bl
.request_alignment
) &&
2896 tail
> bs
->bl
.request_alignment
) {
2897 tail
%= bs
->bl
.request_alignment
;
2901 /* limit request size */
2902 if (num
> max_pdiscard
) {
2910 if (bs
->drv
->bdrv_co_pdiscard
) {
2911 ret
= bs
->drv
->bdrv_co_pdiscard(bs
, offset
, num
);
2914 CoroutineIOCompletion co
= {
2915 .coroutine
= qemu_coroutine_self(),
2918 acb
= bs
->drv
->bdrv_aio_pdiscard(bs
, offset
, num
,
2919 bdrv_co_io_em_complete
, &co
);
2924 qemu_coroutine_yield();
2928 if (ret
&& ret
!= -ENOTSUP
) {
2937 bdrv_co_write_req_finish(child
, req
.offset
, req
.bytes
, &req
, ret
);
2938 tracked_request_end(&req
);
2939 bdrv_dec_in_flight(bs
);
2943 int bdrv_co_ioctl(BlockDriverState
*bs
, int req
, void *buf
)
2945 BlockDriver
*drv
= bs
->drv
;
2946 CoroutineIOCompletion co
= {
2947 .coroutine
= qemu_coroutine_self(),
2951 bdrv_inc_in_flight(bs
);
2952 if (!drv
|| (!drv
->bdrv_aio_ioctl
&& !drv
->bdrv_co_ioctl
)) {
2957 if (drv
->bdrv_co_ioctl
) {
2958 co
.ret
= drv
->bdrv_co_ioctl(bs
, req
, buf
);
2960 acb
= drv
->bdrv_aio_ioctl(bs
, req
, buf
, bdrv_co_io_em_complete
, &co
);
2965 qemu_coroutine_yield();
2968 bdrv_dec_in_flight(bs
);
2972 void *qemu_blockalign(BlockDriverState
*bs
, size_t size
)
2974 return qemu_memalign(bdrv_opt_mem_align(bs
), size
);
2977 void *qemu_blockalign0(BlockDriverState
*bs
, size_t size
)
2979 return memset(qemu_blockalign(bs
, size
), 0, size
);
2982 void *qemu_try_blockalign(BlockDriverState
*bs
, size_t size
)
2984 size_t align
= bdrv_opt_mem_align(bs
);
2986 /* Ensure that NULL is never returned on success */
2992 return qemu_try_memalign(align
, size
);
2995 void *qemu_try_blockalign0(BlockDriverState
*bs
, size_t size
)
2997 void *mem
= qemu_try_blockalign(bs
, size
);
3000 memset(mem
, 0, size
);
3007 * Check if all memory in this vector is sector aligned.
3009 bool bdrv_qiov_is_aligned(BlockDriverState
*bs
, QEMUIOVector
*qiov
)
3012 size_t alignment
= bdrv_min_mem_align(bs
);
3014 for (i
= 0; i
< qiov
->niov
; i
++) {
3015 if ((uintptr_t) qiov
->iov
[i
].iov_base
% alignment
) {
3018 if (qiov
->iov
[i
].iov_len
% alignment
) {
3026 void bdrv_add_before_write_notifier(BlockDriverState
*bs
,
3027 NotifierWithReturn
*notifier
)
3029 notifier_with_return_list_add(&bs
->before_write_notifiers
, notifier
);
3032 void bdrv_io_plug(BlockDriverState
*bs
)
3036 QLIST_FOREACH(child
, &bs
->children
, next
) {
3037 bdrv_io_plug(child
->bs
);
3040 if (qatomic_fetch_inc(&bs
->io_plugged
) == 0) {
3041 BlockDriver
*drv
= bs
->drv
;
3042 if (drv
&& drv
->bdrv_io_plug
) {
3043 drv
->bdrv_io_plug(bs
);
3048 void bdrv_io_unplug(BlockDriverState
*bs
)
3052 assert(bs
->io_plugged
);
3053 if (qatomic_fetch_dec(&bs
->io_plugged
) == 1) {
3054 BlockDriver
*drv
= bs
->drv
;
3055 if (drv
&& drv
->bdrv_io_unplug
) {
3056 drv
->bdrv_io_unplug(bs
);
3060 QLIST_FOREACH(child
, &bs
->children
, next
) {
3061 bdrv_io_unplug(child
->bs
);
3065 void bdrv_register_buf(BlockDriverState
*bs
, void *host
, size_t size
)
3069 if (bs
->drv
&& bs
->drv
->bdrv_register_buf
) {
3070 bs
->drv
->bdrv_register_buf(bs
, host
, size
);
3072 QLIST_FOREACH(child
, &bs
->children
, next
) {
3073 bdrv_register_buf(child
->bs
, host
, size
);
3077 void bdrv_unregister_buf(BlockDriverState
*bs
, void *host
)
3081 if (bs
->drv
&& bs
->drv
->bdrv_unregister_buf
) {
3082 bs
->drv
->bdrv_unregister_buf(bs
, host
);
3084 QLIST_FOREACH(child
, &bs
->children
, next
) {
3085 bdrv_unregister_buf(child
->bs
, host
);
3089 static int coroutine_fn
bdrv_co_copy_range_internal(
3090 BdrvChild
*src
, uint64_t src_offset
, BdrvChild
*dst
,
3091 uint64_t dst_offset
, uint64_t bytes
,
3092 BdrvRequestFlags read_flags
, BdrvRequestFlags write_flags
,
3095 BdrvTrackedRequest req
;
3098 /* TODO We can support BDRV_REQ_NO_FALLBACK here */
3099 assert(!(read_flags
& BDRV_REQ_NO_FALLBACK
));
3100 assert(!(write_flags
& BDRV_REQ_NO_FALLBACK
));
3102 if (!dst
|| !dst
->bs
|| !bdrv_is_inserted(dst
->bs
)) {
3105 ret
= bdrv_check_request32(dst_offset
, bytes
);
3109 if (write_flags
& BDRV_REQ_ZERO_WRITE
) {
3110 return bdrv_co_pwrite_zeroes(dst
, dst_offset
, bytes
, write_flags
);
3113 if (!src
|| !src
->bs
|| !bdrv_is_inserted(src
->bs
)) {
3116 ret
= bdrv_check_request32(src_offset
, bytes
);
3121 if (!src
->bs
->drv
->bdrv_co_copy_range_from
3122 || !dst
->bs
->drv
->bdrv_co_copy_range_to
3123 || src
->bs
->encrypted
|| dst
->bs
->encrypted
) {
3128 bdrv_inc_in_flight(src
->bs
);
3129 tracked_request_begin(&req
, src
->bs
, src_offset
, bytes
,
3132 /* BDRV_REQ_SERIALISING is only for write operation */
3133 assert(!(read_flags
& BDRV_REQ_SERIALISING
));
3134 bdrv_wait_serialising_requests(&req
);
3136 ret
= src
->bs
->drv
->bdrv_co_copy_range_from(src
->bs
,
3140 read_flags
, write_flags
);
3142 tracked_request_end(&req
);
3143 bdrv_dec_in_flight(src
->bs
);
3145 bdrv_inc_in_flight(dst
->bs
);
3146 tracked_request_begin(&req
, dst
->bs
, dst_offset
, bytes
,
3147 BDRV_TRACKED_WRITE
);
3148 ret
= bdrv_co_write_req_prepare(dst
, dst_offset
, bytes
, &req
,
3151 ret
= dst
->bs
->drv
->bdrv_co_copy_range_to(dst
->bs
,
3155 read_flags
, write_flags
);
3157 bdrv_co_write_req_finish(dst
, dst_offset
, bytes
, &req
, ret
);
3158 tracked_request_end(&req
);
3159 bdrv_dec_in_flight(dst
->bs
);
3165 /* Copy range from @src to @dst.
3167 * See the comment of bdrv_co_copy_range for the parameter and return value
3169 int coroutine_fn
bdrv_co_copy_range_from(BdrvChild
*src
, uint64_t src_offset
,
3170 BdrvChild
*dst
, uint64_t dst_offset
,
3172 BdrvRequestFlags read_flags
,
3173 BdrvRequestFlags write_flags
)
3175 trace_bdrv_co_copy_range_from(src
, src_offset
, dst
, dst_offset
, bytes
,
3176 read_flags
, write_flags
);
3177 return bdrv_co_copy_range_internal(src
, src_offset
, dst
, dst_offset
,
3178 bytes
, read_flags
, write_flags
, true);
3181 /* Copy range from @src to @dst.
3183 * See the comment of bdrv_co_copy_range for the parameter and return value
3185 int coroutine_fn
bdrv_co_copy_range_to(BdrvChild
*src
, uint64_t src_offset
,
3186 BdrvChild
*dst
, uint64_t dst_offset
,
3188 BdrvRequestFlags read_flags
,
3189 BdrvRequestFlags write_flags
)
3191 trace_bdrv_co_copy_range_to(src
, src_offset
, dst
, dst_offset
, bytes
,
3192 read_flags
, write_flags
);
3193 return bdrv_co_copy_range_internal(src
, src_offset
, dst
, dst_offset
,
3194 bytes
, read_flags
, write_flags
, false);
3197 int coroutine_fn
bdrv_co_copy_range(BdrvChild
*src
, uint64_t src_offset
,
3198 BdrvChild
*dst
, uint64_t dst_offset
,
3199 uint64_t bytes
, BdrvRequestFlags read_flags
,
3200 BdrvRequestFlags write_flags
)
3202 return bdrv_co_copy_range_from(src
, src_offset
,
3204 bytes
, read_flags
, write_flags
);
3207 static void bdrv_parent_cb_resize(BlockDriverState
*bs
)
3210 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
3211 if (c
->klass
->resize
) {
3212 c
->klass
->resize(c
);
3218 * Truncate file to 'offset' bytes (needed only for file protocols)
3220 * If 'exact' is true, the file must be resized to exactly the given
3221 * 'offset'. Otherwise, it is sufficient for the node to be at least
3222 * 'offset' bytes in length.
3224 int coroutine_fn
bdrv_co_truncate(BdrvChild
*child
, int64_t offset
, bool exact
,
3225 PreallocMode prealloc
, BdrvRequestFlags flags
,
3228 BlockDriverState
*bs
= child
->bs
;
3229 BdrvChild
*filtered
, *backing
;
3230 BlockDriver
*drv
= bs
->drv
;
3231 BdrvTrackedRequest req
;
3232 int64_t old_size
, new_bytes
;
3236 /* if bs->drv == NULL, bs is closed, so there's nothing to do here */
3238 error_setg(errp
, "No medium inserted");
3242 error_setg(errp
, "Image size cannot be negative");
3246 ret
= bdrv_check_request(offset
, 0);
3248 error_setg(errp
, "Required too big image size, it must be not greater "
3249 "than %" PRId64
, BDRV_MAX_LENGTH
);
3253 old_size
= bdrv_getlength(bs
);
3255 error_setg_errno(errp
, -old_size
, "Failed to get old image size");
3259 if (offset
> old_size
) {
3260 new_bytes
= offset
- old_size
;
3265 bdrv_inc_in_flight(bs
);
3266 tracked_request_begin(&req
, bs
, offset
- new_bytes
, new_bytes
,
3267 BDRV_TRACKED_TRUNCATE
);
3269 /* If we are growing the image and potentially using preallocation for the
3270 * new area, we need to make sure that no write requests are made to it
3271 * concurrently or they might be overwritten by preallocation. */
3273 bdrv_make_request_serialising(&req
, 1);
3275 if (bs
->read_only
) {
3276 error_setg(errp
, "Image is read-only");
3280 ret
= bdrv_co_write_req_prepare(child
, offset
- new_bytes
, new_bytes
, &req
,
3283 error_setg_errno(errp
, -ret
,
3284 "Failed to prepare request for truncation");
3288 filtered
= bdrv_filter_child(bs
);
3289 backing
= bdrv_cow_child(bs
);
3292 * If the image has a backing file that is large enough that it would
3293 * provide data for the new area, we cannot leave it unallocated because
3294 * then the backing file content would become visible. Instead, zero-fill
3297 * Note that if the image has a backing file, but was opened without the
3298 * backing file, taking care of keeping things consistent with that backing
3299 * file is the user's responsibility.
3301 if (new_bytes
&& backing
) {
3302 int64_t backing_len
;
3304 backing_len
= bdrv_getlength(backing
->bs
);
3305 if (backing_len
< 0) {
3307 error_setg_errno(errp
, -ret
, "Could not get backing file size");
3311 if (backing_len
> old_size
) {
3312 flags
|= BDRV_REQ_ZERO_WRITE
;
3316 if (drv
->bdrv_co_truncate
) {
3317 if (flags
& ~bs
->supported_truncate_flags
) {
3318 error_setg(errp
, "Block driver does not support requested flags");
3322 ret
= drv
->bdrv_co_truncate(bs
, offset
, exact
, prealloc
, flags
, errp
);
3323 } else if (filtered
) {
3324 ret
= bdrv_co_truncate(filtered
, offset
, exact
, prealloc
, flags
, errp
);
3326 error_setg(errp
, "Image format driver does not support resize");
3334 ret
= refresh_total_sectors(bs
, offset
>> BDRV_SECTOR_BITS
);
3336 error_setg_errno(errp
, -ret
, "Could not refresh total sector count");
3338 offset
= bs
->total_sectors
* BDRV_SECTOR_SIZE
;
3340 /* It's possible that truncation succeeded but refresh_total_sectors
3341 * failed, but the latter doesn't affect how we should finish the request.
3342 * Pass 0 as the last parameter so that dirty bitmaps etc. are handled. */
3343 bdrv_co_write_req_finish(child
, offset
- new_bytes
, new_bytes
, &req
, 0);
3346 tracked_request_end(&req
);
3347 bdrv_dec_in_flight(bs
);