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 "qemu/cutils.h"
33 #include "qapi/error.h"
34 #include "qemu/error-report.h"
35 #include "qemu/main-loop.h"
37 #define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */
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
->role
->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
->role
->drained_end
) {
65 c
->role
->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
, atomic_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
->role
->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
->role
->drained_poll
) {
93 return c
->role
->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
->role
->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
->role
->drained_begin
) {
118 c
->role
->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
)
138 BlockDriver
*drv
= bs
->drv
;
139 Error
*local_err
= NULL
;
141 memset(&bs
->bl
, 0, sizeof(bs
->bl
));
147 /* Default alignment based on whether driver has byte interface */
148 bs
->bl
.request_alignment
= (drv
->bdrv_co_preadv
||
149 drv
->bdrv_aio_preadv
) ? 1 : 512;
151 /* Take some limits from the children as a default */
153 bdrv_refresh_limits(bs
->file
->bs
, &local_err
);
155 error_propagate(errp
, local_err
);
158 bdrv_merge_limits(&bs
->bl
, &bs
->file
->bs
->bl
);
160 bs
->bl
.min_mem_alignment
= 512;
161 bs
->bl
.opt_mem_alignment
= getpagesize();
163 /* Safe default since most protocols use readv()/writev()/etc */
164 bs
->bl
.max_iov
= IOV_MAX
;
168 bdrv_refresh_limits(bs
->backing
->bs
, &local_err
);
170 error_propagate(errp
, local_err
);
173 bdrv_merge_limits(&bs
->bl
, &bs
->backing
->bs
->bl
);
176 /* Then let the driver override it */
177 if (drv
->bdrv_refresh_limits
) {
178 drv
->bdrv_refresh_limits(bs
, errp
);
183 * The copy-on-read flag is actually a reference count so multiple users may
184 * use the feature without worrying about clobbering its previous state.
185 * Copy-on-read stays enabled until all users have called to disable it.
187 void bdrv_enable_copy_on_read(BlockDriverState
*bs
)
189 atomic_inc(&bs
->copy_on_read
);
192 void bdrv_disable_copy_on_read(BlockDriverState
*bs
)
194 int old
= atomic_fetch_dec(&bs
->copy_on_read
);
200 BlockDriverState
*bs
;
206 bool ignore_bds_parents
;
207 int *drained_end_counter
;
210 static void coroutine_fn
bdrv_drain_invoke_entry(void *opaque
)
212 BdrvCoDrainData
*data
= opaque
;
213 BlockDriverState
*bs
= data
->bs
;
216 bs
->drv
->bdrv_co_drain_begin(bs
);
218 bs
->drv
->bdrv_co_drain_end(bs
);
221 /* Set data->done and decrement drained_end_counter before bdrv_wakeup() */
222 atomic_mb_set(&data
->done
, true);
224 atomic_dec(data
->drained_end_counter
);
226 bdrv_dec_in_flight(bs
);
231 /* Recursively call BlockDriver.bdrv_co_drain_begin/end callbacks */
232 static void bdrv_drain_invoke(BlockDriverState
*bs
, bool begin
,
233 int *drained_end_counter
)
235 BdrvCoDrainData
*data
;
237 if (!bs
->drv
|| (begin
&& !bs
->drv
->bdrv_co_drain_begin
) ||
238 (!begin
&& !bs
->drv
->bdrv_co_drain_end
)) {
242 data
= g_new(BdrvCoDrainData
, 1);
243 *data
= (BdrvCoDrainData
) {
247 .drained_end_counter
= drained_end_counter
,
251 atomic_inc(drained_end_counter
);
254 /* Make sure the driver callback completes during the polling phase for
256 bdrv_inc_in_flight(bs
);
257 data
->co
= qemu_coroutine_create(bdrv_drain_invoke_entry
, data
);
258 aio_co_schedule(bdrv_get_aio_context(bs
), data
->co
);
261 /* Returns true if BDRV_POLL_WHILE() should go into a blocking aio_poll() */
262 bool bdrv_drain_poll(BlockDriverState
*bs
, bool recursive
,
263 BdrvChild
*ignore_parent
, bool ignore_bds_parents
)
265 BdrvChild
*child
, *next
;
267 if (bdrv_parent_drained_poll(bs
, ignore_parent
, ignore_bds_parents
)) {
271 if (atomic_read(&bs
->in_flight
)) {
276 assert(!ignore_bds_parents
);
277 QLIST_FOREACH_SAFE(child
, &bs
->children
, next
, next
) {
278 if (bdrv_drain_poll(child
->bs
, recursive
, child
, false)) {
287 static bool bdrv_drain_poll_top_level(BlockDriverState
*bs
, bool recursive
,
288 BdrvChild
*ignore_parent
)
290 return bdrv_drain_poll(bs
, recursive
, ignore_parent
, false);
293 static void bdrv_do_drained_begin(BlockDriverState
*bs
, bool recursive
,
294 BdrvChild
*parent
, bool ignore_bds_parents
,
296 static void bdrv_do_drained_end(BlockDriverState
*bs
, bool recursive
,
297 BdrvChild
*parent
, bool ignore_bds_parents
,
298 int *drained_end_counter
);
300 static void bdrv_co_drain_bh_cb(void *opaque
)
302 BdrvCoDrainData
*data
= opaque
;
303 Coroutine
*co
= data
->co
;
304 BlockDriverState
*bs
= data
->bs
;
307 AioContext
*ctx
= bdrv_get_aio_context(bs
);
308 AioContext
*co_ctx
= qemu_coroutine_get_aio_context(co
);
311 * When the coroutine yielded, the lock for its home context was
312 * released, so we need to re-acquire it here. If it explicitly
313 * acquired a different context, the lock is still held and we don't
314 * want to lock it a second time (or AIO_WAIT_WHILE() would hang).
317 aio_context_acquire(ctx
);
319 bdrv_dec_in_flight(bs
);
321 assert(!data
->drained_end_counter
);
322 bdrv_do_drained_begin(bs
, data
->recursive
, data
->parent
,
323 data
->ignore_bds_parents
, data
->poll
);
326 bdrv_do_drained_end(bs
, data
->recursive
, data
->parent
,
327 data
->ignore_bds_parents
,
328 data
->drained_end_counter
);
331 aio_context_release(ctx
);
335 bdrv_drain_all_begin();
342 static void coroutine_fn
bdrv_co_yield_to_drain(BlockDriverState
*bs
,
343 bool begin
, bool recursive
,
345 bool ignore_bds_parents
,
347 int *drained_end_counter
)
349 BdrvCoDrainData data
;
351 /* Calling bdrv_drain() from a BH ensures the current coroutine yields and
352 * other coroutines run if they were queued by aio_co_enter(). */
354 assert(qemu_in_coroutine());
355 data
= (BdrvCoDrainData
) {
356 .co
= qemu_coroutine_self(),
360 .recursive
= recursive
,
362 .ignore_bds_parents
= ignore_bds_parents
,
364 .drained_end_counter
= drained_end_counter
,
368 bdrv_inc_in_flight(bs
);
370 aio_bh_schedule_oneshot(bdrv_get_aio_context(bs
),
371 bdrv_co_drain_bh_cb
, &data
);
373 qemu_coroutine_yield();
374 /* If we are resumed from some other event (such as an aio completion or a
375 * timer callback), it is a bug in the caller that should be fixed. */
379 void bdrv_do_drained_begin_quiesce(BlockDriverState
*bs
,
380 BdrvChild
*parent
, bool ignore_bds_parents
)
382 assert(!qemu_in_coroutine());
384 /* Stop things in parent-to-child order */
385 if (atomic_fetch_inc(&bs
->quiesce_counter
) == 0) {
386 aio_disable_external(bdrv_get_aio_context(bs
));
389 bdrv_parent_drained_begin(bs
, parent
, ignore_bds_parents
);
390 bdrv_drain_invoke(bs
, true, NULL
);
393 static void bdrv_do_drained_begin(BlockDriverState
*bs
, bool recursive
,
394 BdrvChild
*parent
, bool ignore_bds_parents
,
397 BdrvChild
*child
, *next
;
399 if (qemu_in_coroutine()) {
400 bdrv_co_yield_to_drain(bs
, true, recursive
, parent
, ignore_bds_parents
,
405 bdrv_do_drained_begin_quiesce(bs
, parent
, ignore_bds_parents
);
408 assert(!ignore_bds_parents
);
409 bs
->recursive_quiesce_counter
++;
410 QLIST_FOREACH_SAFE(child
, &bs
->children
, next
, next
) {
411 bdrv_do_drained_begin(child
->bs
, true, child
, ignore_bds_parents
,
417 * Wait for drained requests to finish.
419 * Calling BDRV_POLL_WHILE() only once for the top-level node is okay: The
420 * call is needed so things in this AioContext can make progress even
421 * though we don't return to the main AioContext loop - this automatically
422 * includes other nodes in the same AioContext and therefore all child
426 assert(!ignore_bds_parents
);
427 BDRV_POLL_WHILE(bs
, bdrv_drain_poll_top_level(bs
, recursive
, parent
));
431 void bdrv_drained_begin(BlockDriverState
*bs
)
433 bdrv_do_drained_begin(bs
, false, NULL
, false, true);
436 void bdrv_subtree_drained_begin(BlockDriverState
*bs
)
438 bdrv_do_drained_begin(bs
, true, NULL
, false, true);
442 * This function does not poll, nor must any of its recursively called
443 * functions. The *drained_end_counter pointee will be incremented
444 * once for every background operation scheduled, and decremented once
445 * the operation settles. Therefore, the pointer must remain valid
446 * until the pointee reaches 0. That implies that whoever sets up the
447 * pointee has to poll until it is 0.
449 * We use atomic operations to access *drained_end_counter, because
450 * (1) when called from bdrv_set_aio_context_ignore(), the subgraph of
451 * @bs may contain nodes in different AioContexts,
452 * (2) bdrv_drain_all_end() uses the same counter for all nodes,
453 * regardless of which AioContext they are in.
455 static void bdrv_do_drained_end(BlockDriverState
*bs
, bool recursive
,
456 BdrvChild
*parent
, bool ignore_bds_parents
,
457 int *drained_end_counter
)
460 int old_quiesce_counter
;
462 assert(drained_end_counter
!= NULL
);
464 if (qemu_in_coroutine()) {
465 bdrv_co_yield_to_drain(bs
, false, recursive
, parent
, ignore_bds_parents
,
466 false, drained_end_counter
);
469 assert(bs
->quiesce_counter
> 0);
471 /* Re-enable things in child-to-parent order */
472 bdrv_drain_invoke(bs
, false, drained_end_counter
);
473 bdrv_parent_drained_end(bs
, parent
, ignore_bds_parents
,
474 drained_end_counter
);
476 old_quiesce_counter
= atomic_fetch_dec(&bs
->quiesce_counter
);
477 if (old_quiesce_counter
== 1) {
478 aio_enable_external(bdrv_get_aio_context(bs
));
482 assert(!ignore_bds_parents
);
483 bs
->recursive_quiesce_counter
--;
484 QLIST_FOREACH(child
, &bs
->children
, next
) {
485 bdrv_do_drained_end(child
->bs
, true, child
, ignore_bds_parents
,
486 drained_end_counter
);
491 void bdrv_drained_end(BlockDriverState
*bs
)
493 int drained_end_counter
= 0;
494 bdrv_do_drained_end(bs
, false, NULL
, false, &drained_end_counter
);
495 BDRV_POLL_WHILE(bs
, atomic_read(&drained_end_counter
) > 0);
498 void bdrv_drained_end_no_poll(BlockDriverState
*bs
, int *drained_end_counter
)
500 bdrv_do_drained_end(bs
, false, NULL
, false, drained_end_counter
);
503 void bdrv_subtree_drained_end(BlockDriverState
*bs
)
505 int drained_end_counter
= 0;
506 bdrv_do_drained_end(bs
, true, NULL
, false, &drained_end_counter
);
507 BDRV_POLL_WHILE(bs
, atomic_read(&drained_end_counter
) > 0);
510 void bdrv_apply_subtree_drain(BdrvChild
*child
, BlockDriverState
*new_parent
)
514 for (i
= 0; i
< new_parent
->recursive_quiesce_counter
; i
++) {
515 bdrv_do_drained_begin(child
->bs
, true, child
, false, true);
519 void bdrv_unapply_subtree_drain(BdrvChild
*child
, BlockDriverState
*old_parent
)
521 int drained_end_counter
= 0;
524 for (i
= 0; i
< old_parent
->recursive_quiesce_counter
; i
++) {
525 bdrv_do_drained_end(child
->bs
, true, child
, false,
526 &drained_end_counter
);
529 BDRV_POLL_WHILE(child
->bs
, atomic_read(&drained_end_counter
) > 0);
533 * Wait for pending requests to complete on a single BlockDriverState subtree,
534 * and suspend block driver's internal I/O until next request arrives.
536 * Note that unlike bdrv_drain_all(), the caller must hold the BlockDriverState
539 void coroutine_fn
bdrv_co_drain(BlockDriverState
*bs
)
541 assert(qemu_in_coroutine());
542 bdrv_drained_begin(bs
);
543 bdrv_drained_end(bs
);
546 void bdrv_drain(BlockDriverState
*bs
)
548 bdrv_drained_begin(bs
);
549 bdrv_drained_end(bs
);
552 static void bdrv_drain_assert_idle(BlockDriverState
*bs
)
554 BdrvChild
*child
, *next
;
556 assert(atomic_read(&bs
->in_flight
) == 0);
557 QLIST_FOREACH_SAFE(child
, &bs
->children
, next
, next
) {
558 bdrv_drain_assert_idle(child
->bs
);
562 unsigned int bdrv_drain_all_count
= 0;
564 static bool bdrv_drain_all_poll(void)
566 BlockDriverState
*bs
= NULL
;
569 /* bdrv_drain_poll() can't make changes to the graph and we are holding the
570 * main AioContext lock, so iterating bdrv_next_all_states() is safe. */
571 while ((bs
= bdrv_next_all_states(bs
))) {
572 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
573 aio_context_acquire(aio_context
);
574 result
|= bdrv_drain_poll(bs
, false, NULL
, true);
575 aio_context_release(aio_context
);
582 * Wait for pending requests to complete across all BlockDriverStates
584 * This function does not flush data to disk, use bdrv_flush_all() for that
585 * after calling this function.
587 * This pauses all block jobs and disables external clients. It must
588 * be paired with bdrv_drain_all_end().
590 * NOTE: no new block jobs or BlockDriverStates can be created between
591 * the bdrv_drain_all_begin() and bdrv_drain_all_end() calls.
593 void bdrv_drain_all_begin(void)
595 BlockDriverState
*bs
= NULL
;
597 if (qemu_in_coroutine()) {
598 bdrv_co_yield_to_drain(NULL
, true, false, NULL
, true, true, NULL
);
602 /* AIO_WAIT_WHILE() with a NULL context can only be called from the main
603 * loop AioContext, so make sure we're in the main context. */
604 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
605 assert(bdrv_drain_all_count
< INT_MAX
);
606 bdrv_drain_all_count
++;
608 /* Quiesce all nodes, without polling in-flight requests yet. The graph
609 * cannot change during this loop. */
610 while ((bs
= bdrv_next_all_states(bs
))) {
611 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
613 aio_context_acquire(aio_context
);
614 bdrv_do_drained_begin(bs
, false, NULL
, true, false);
615 aio_context_release(aio_context
);
618 /* Now poll the in-flight requests */
619 AIO_WAIT_WHILE(NULL
, bdrv_drain_all_poll());
621 while ((bs
= bdrv_next_all_states(bs
))) {
622 bdrv_drain_assert_idle(bs
);
626 void bdrv_drain_all_end(void)
628 BlockDriverState
*bs
= NULL
;
629 int drained_end_counter
= 0;
631 while ((bs
= bdrv_next_all_states(bs
))) {
632 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
634 aio_context_acquire(aio_context
);
635 bdrv_do_drained_end(bs
, false, NULL
, true, &drained_end_counter
);
636 aio_context_release(aio_context
);
639 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
640 AIO_WAIT_WHILE(NULL
, atomic_read(&drained_end_counter
) > 0);
642 assert(bdrv_drain_all_count
> 0);
643 bdrv_drain_all_count
--;
646 void bdrv_drain_all(void)
648 bdrv_drain_all_begin();
649 bdrv_drain_all_end();
653 * Remove an active request from the tracked requests list
655 * This function should be called when a tracked request is completing.
657 static void tracked_request_end(BdrvTrackedRequest
*req
)
659 if (req
->serialising
) {
660 atomic_dec(&req
->bs
->serialising_in_flight
);
663 qemu_co_mutex_lock(&req
->bs
->reqs_lock
);
664 QLIST_REMOVE(req
, list
);
665 qemu_co_queue_restart_all(&req
->wait_queue
);
666 qemu_co_mutex_unlock(&req
->bs
->reqs_lock
);
670 * Add an active request to the tracked requests list
672 static void tracked_request_begin(BdrvTrackedRequest
*req
,
673 BlockDriverState
*bs
,
676 enum BdrvTrackedRequestType type
)
678 assert(bytes
<= INT64_MAX
&& offset
<= INT64_MAX
- bytes
);
680 *req
= (BdrvTrackedRequest
){
685 .co
= qemu_coroutine_self(),
686 .serialising
= false,
687 .overlap_offset
= offset
,
688 .overlap_bytes
= bytes
,
691 qemu_co_queue_init(&req
->wait_queue
);
693 qemu_co_mutex_lock(&bs
->reqs_lock
);
694 QLIST_INSERT_HEAD(&bs
->tracked_requests
, req
, list
);
695 qemu_co_mutex_unlock(&bs
->reqs_lock
);
698 static void mark_request_serialising(BdrvTrackedRequest
*req
, uint64_t align
)
700 int64_t overlap_offset
= req
->offset
& ~(align
- 1);
701 uint64_t overlap_bytes
= ROUND_UP(req
->offset
+ req
->bytes
, align
)
704 if (!req
->serialising
) {
705 atomic_inc(&req
->bs
->serialising_in_flight
);
706 req
->serialising
= true;
709 req
->overlap_offset
= MIN(req
->overlap_offset
, overlap_offset
);
710 req
->overlap_bytes
= MAX(req
->overlap_bytes
, overlap_bytes
);
713 static bool is_request_serialising_and_aligned(BdrvTrackedRequest
*req
)
716 * If the request is serialising, overlap_offset and overlap_bytes are set,
717 * so we can check if the request is aligned. Otherwise, don't care and
721 return req
->serialising
&& (req
->offset
== req
->overlap_offset
) &&
722 (req
->bytes
== req
->overlap_bytes
);
726 * Round a region to cluster boundaries
728 void bdrv_round_to_clusters(BlockDriverState
*bs
,
729 int64_t offset
, int64_t bytes
,
730 int64_t *cluster_offset
,
731 int64_t *cluster_bytes
)
735 if (bdrv_get_info(bs
, &bdi
) < 0 || bdi
.cluster_size
== 0) {
736 *cluster_offset
= offset
;
737 *cluster_bytes
= bytes
;
739 int64_t c
= bdi
.cluster_size
;
740 *cluster_offset
= QEMU_ALIGN_DOWN(offset
, c
);
741 *cluster_bytes
= QEMU_ALIGN_UP(offset
- *cluster_offset
+ bytes
, c
);
745 static int bdrv_get_cluster_size(BlockDriverState
*bs
)
750 ret
= bdrv_get_info(bs
, &bdi
);
751 if (ret
< 0 || bdi
.cluster_size
== 0) {
752 return bs
->bl
.request_alignment
;
754 return bdi
.cluster_size
;
758 static bool tracked_request_overlaps(BdrvTrackedRequest
*req
,
759 int64_t offset
, uint64_t bytes
)
762 if (offset
>= req
->overlap_offset
+ req
->overlap_bytes
) {
766 if (req
->overlap_offset
>= offset
+ bytes
) {
772 void bdrv_inc_in_flight(BlockDriverState
*bs
)
774 atomic_inc(&bs
->in_flight
);
777 void bdrv_wakeup(BlockDriverState
*bs
)
782 void bdrv_dec_in_flight(BlockDriverState
*bs
)
784 atomic_dec(&bs
->in_flight
);
788 static bool coroutine_fn
wait_serialising_requests(BdrvTrackedRequest
*self
)
790 BlockDriverState
*bs
= self
->bs
;
791 BdrvTrackedRequest
*req
;
795 if (!atomic_read(&bs
->serialising_in_flight
)) {
801 qemu_co_mutex_lock(&bs
->reqs_lock
);
802 QLIST_FOREACH(req
, &bs
->tracked_requests
, list
) {
803 if (req
== self
|| (!req
->serialising
&& !self
->serialising
)) {
806 if (tracked_request_overlaps(req
, self
->overlap_offset
,
807 self
->overlap_bytes
))
809 /* Hitting this means there was a reentrant request, for
810 * example, a block driver issuing nested requests. This must
811 * never happen since it means deadlock.
813 assert(qemu_coroutine_self() != req
->co
);
815 /* If the request is already (indirectly) waiting for us, or
816 * will wait for us as soon as it wakes up, then just go on
817 * (instead of producing a deadlock in the former case). */
818 if (!req
->waiting_for
) {
819 self
->waiting_for
= req
;
820 qemu_co_queue_wait(&req
->wait_queue
, &bs
->reqs_lock
);
821 self
->waiting_for
= NULL
;
828 qemu_co_mutex_unlock(&bs
->reqs_lock
);
834 static int bdrv_check_byte_request(BlockDriverState
*bs
, int64_t offset
,
837 if (size
> BDRV_REQUEST_MAX_BYTES
) {
841 if (!bdrv_is_inserted(bs
)) {
852 typedef struct RwCo
{
858 BdrvRequestFlags flags
;
861 static void coroutine_fn
bdrv_rw_co_entry(void *opaque
)
865 if (!rwco
->is_write
) {
866 rwco
->ret
= bdrv_co_preadv(rwco
->child
, rwco
->offset
,
867 rwco
->qiov
->size
, rwco
->qiov
,
870 rwco
->ret
= bdrv_co_pwritev(rwco
->child
, rwco
->offset
,
871 rwco
->qiov
->size
, rwco
->qiov
,
878 * Process a vectored synchronous request using coroutines
880 static int bdrv_prwv_co(BdrvChild
*child
, int64_t offset
,
881 QEMUIOVector
*qiov
, bool is_write
,
882 BdrvRequestFlags flags
)
889 .is_write
= is_write
,
894 if (qemu_in_coroutine()) {
895 /* Fast-path if already in coroutine context */
896 bdrv_rw_co_entry(&rwco
);
898 co
= qemu_coroutine_create(bdrv_rw_co_entry
, &rwco
);
899 bdrv_coroutine_enter(child
->bs
, co
);
900 BDRV_POLL_WHILE(child
->bs
, rwco
.ret
== NOT_DONE
);
905 int bdrv_pwrite_zeroes(BdrvChild
*child
, int64_t offset
,
906 int bytes
, BdrvRequestFlags flags
)
908 QEMUIOVector qiov
= QEMU_IOVEC_INIT_BUF(qiov
, NULL
, bytes
);
910 return bdrv_prwv_co(child
, offset
, &qiov
, true,
911 BDRV_REQ_ZERO_WRITE
| flags
);
915 * Completely zero out a block device with the help of bdrv_pwrite_zeroes.
916 * The operation is sped up by checking the block status and only writing
917 * zeroes to the device if they currently do not return zeroes. Optional
918 * flags are passed through to bdrv_pwrite_zeroes (e.g. BDRV_REQ_MAY_UNMAP,
921 * Returns < 0 on error, 0 on success. For error codes see bdrv_write().
923 int bdrv_make_zero(BdrvChild
*child
, BdrvRequestFlags flags
)
926 int64_t target_size
, bytes
, offset
= 0;
927 BlockDriverState
*bs
= child
->bs
;
929 target_size
= bdrv_getlength(bs
);
930 if (target_size
< 0) {
935 bytes
= MIN(target_size
- offset
, BDRV_REQUEST_MAX_BYTES
);
939 ret
= bdrv_block_status(bs
, offset
, bytes
, &bytes
, NULL
, NULL
);
943 if (ret
& BDRV_BLOCK_ZERO
) {
947 ret
= bdrv_pwrite_zeroes(child
, offset
, bytes
, flags
);
955 int bdrv_preadv(BdrvChild
*child
, int64_t offset
, QEMUIOVector
*qiov
)
959 ret
= bdrv_prwv_co(child
, offset
, qiov
, false, 0);
967 /* See bdrv_pwrite() for the return codes */
968 int bdrv_pread(BdrvChild
*child
, int64_t offset
, void *buf
, int bytes
)
970 QEMUIOVector qiov
= QEMU_IOVEC_INIT_BUF(qiov
, buf
, bytes
);
976 return bdrv_preadv(child
, offset
, &qiov
);
979 int bdrv_pwritev(BdrvChild
*child
, int64_t offset
, QEMUIOVector
*qiov
)
983 ret
= bdrv_prwv_co(child
, offset
, qiov
, true, 0);
991 /* Return no. of bytes on success or < 0 on error. Important errors are:
992 -EIO generic I/O error (may happen for all errors)
993 -ENOMEDIUM No media inserted.
994 -EINVAL Invalid offset or number of bytes
995 -EACCES Trying to write a read-only device
997 int bdrv_pwrite(BdrvChild
*child
, int64_t offset
, const void *buf
, int bytes
)
999 QEMUIOVector qiov
= QEMU_IOVEC_INIT_BUF(qiov
, buf
, bytes
);
1005 return bdrv_pwritev(child
, offset
, &qiov
);
1009 * Writes to the file and ensures that no writes are reordered across this
1010 * request (acts as a barrier)
1012 * Returns 0 on success, -errno in error cases.
1014 int bdrv_pwrite_sync(BdrvChild
*child
, int64_t offset
,
1015 const void *buf
, int count
)
1019 ret
= bdrv_pwrite(child
, offset
, buf
, count
);
1024 ret
= bdrv_flush(child
->bs
);
1032 typedef struct CoroutineIOCompletion
{
1033 Coroutine
*coroutine
;
1035 } CoroutineIOCompletion
;
1037 static void bdrv_co_io_em_complete(void *opaque
, int ret
)
1039 CoroutineIOCompletion
*co
= opaque
;
1042 aio_co_wake(co
->coroutine
);
1045 static int coroutine_fn
bdrv_driver_preadv(BlockDriverState
*bs
,
1046 uint64_t offset
, uint64_t bytes
,
1047 QEMUIOVector
*qiov
, int flags
)
1049 BlockDriver
*drv
= bs
->drv
;
1051 unsigned int nb_sectors
;
1053 assert(!(flags
& ~BDRV_REQ_MASK
));
1054 assert(!(flags
& BDRV_REQ_NO_FALLBACK
));
1060 if (drv
->bdrv_co_preadv
) {
1061 return drv
->bdrv_co_preadv(bs
, offset
, bytes
, qiov
, flags
);
1064 if (drv
->bdrv_aio_preadv
) {
1066 CoroutineIOCompletion co
= {
1067 .coroutine
= qemu_coroutine_self(),
1070 acb
= drv
->bdrv_aio_preadv(bs
, offset
, bytes
, qiov
, flags
,
1071 bdrv_co_io_em_complete
, &co
);
1075 qemu_coroutine_yield();
1080 sector_num
= offset
>> BDRV_SECTOR_BITS
;
1081 nb_sectors
= bytes
>> BDRV_SECTOR_BITS
;
1083 assert((offset
& (BDRV_SECTOR_SIZE
- 1)) == 0);
1084 assert((bytes
& (BDRV_SECTOR_SIZE
- 1)) == 0);
1085 assert(bytes
<= BDRV_REQUEST_MAX_BYTES
);
1086 assert(drv
->bdrv_co_readv
);
1088 return drv
->bdrv_co_readv(bs
, sector_num
, nb_sectors
, qiov
);
1091 static int coroutine_fn
bdrv_driver_pwritev(BlockDriverState
*bs
,
1092 uint64_t offset
, uint64_t bytes
,
1093 QEMUIOVector
*qiov
, int flags
)
1095 BlockDriver
*drv
= bs
->drv
;
1097 unsigned int nb_sectors
;
1100 assert(!(flags
& ~BDRV_REQ_MASK
));
1101 assert(!(flags
& BDRV_REQ_NO_FALLBACK
));
1107 if (drv
->bdrv_co_pwritev
) {
1108 ret
= drv
->bdrv_co_pwritev(bs
, offset
, bytes
, qiov
,
1109 flags
& bs
->supported_write_flags
);
1110 flags
&= ~bs
->supported_write_flags
;
1114 if (drv
->bdrv_aio_pwritev
) {
1116 CoroutineIOCompletion co
= {
1117 .coroutine
= qemu_coroutine_self(),
1120 acb
= drv
->bdrv_aio_pwritev(bs
, offset
, bytes
, qiov
,
1121 flags
& bs
->supported_write_flags
,
1122 bdrv_co_io_em_complete
, &co
);
1123 flags
&= ~bs
->supported_write_flags
;
1127 qemu_coroutine_yield();
1133 sector_num
= offset
>> BDRV_SECTOR_BITS
;
1134 nb_sectors
= bytes
>> BDRV_SECTOR_BITS
;
1136 assert((offset
& (BDRV_SECTOR_SIZE
- 1)) == 0);
1137 assert((bytes
& (BDRV_SECTOR_SIZE
- 1)) == 0);
1138 assert(bytes
<= BDRV_REQUEST_MAX_BYTES
);
1140 assert(drv
->bdrv_co_writev
);
1141 ret
= drv
->bdrv_co_writev(bs
, sector_num
, nb_sectors
, qiov
,
1142 flags
& bs
->supported_write_flags
);
1143 flags
&= ~bs
->supported_write_flags
;
1146 if (ret
== 0 && (flags
& BDRV_REQ_FUA
)) {
1147 ret
= bdrv_co_flush(bs
);
1153 static int coroutine_fn
1154 bdrv_driver_pwritev_compressed(BlockDriverState
*bs
, uint64_t offset
,
1155 uint64_t bytes
, QEMUIOVector
*qiov
)
1157 BlockDriver
*drv
= bs
->drv
;
1163 if (!drv
->bdrv_co_pwritev_compressed
) {
1167 return drv
->bdrv_co_pwritev_compressed(bs
, offset
, bytes
, qiov
);
1170 static int coroutine_fn
bdrv_co_do_copy_on_readv(BdrvChild
*child
,
1171 int64_t offset
, unsigned int bytes
, QEMUIOVector
*qiov
)
1173 BlockDriverState
*bs
= child
->bs
;
1175 /* Perform I/O through a temporary buffer so that users who scribble over
1176 * their read buffer while the operation is in progress do not end up
1177 * modifying the image file. This is critical for zero-copy guest I/O
1178 * where anything might happen inside guest memory.
1180 void *bounce_buffer
;
1182 BlockDriver
*drv
= bs
->drv
;
1183 QEMUIOVector local_qiov
;
1184 int64_t cluster_offset
;
1185 int64_t cluster_bytes
;
1188 int max_transfer
= MIN_NON_ZERO(bs
->bl
.max_transfer
,
1189 BDRV_REQUEST_MAX_BYTES
);
1190 unsigned int progress
= 0;
1196 /* FIXME We cannot require callers to have write permissions when all they
1197 * are doing is a read request. If we did things right, write permissions
1198 * would be obtained anyway, but internally by the copy-on-read code. As
1199 * long as it is implemented here rather than in a separate filter driver,
1200 * the copy-on-read code doesn't have its own BdrvChild, however, for which
1201 * it could request permissions. Therefore we have to bypass the permission
1202 * system for the moment. */
1203 // assert(child->perm & (BLK_PERM_WRITE_UNCHANGED | BLK_PERM_WRITE));
1205 /* Cover entire cluster so no additional backing file I/O is required when
1206 * allocating cluster in the image file. Note that this value may exceed
1207 * BDRV_REQUEST_MAX_BYTES (even when the original read did not), which
1208 * is one reason we loop rather than doing it all at once.
1210 bdrv_round_to_clusters(bs
, offset
, bytes
, &cluster_offset
, &cluster_bytes
);
1211 skip_bytes
= offset
- cluster_offset
;
1213 trace_bdrv_co_do_copy_on_readv(bs
, offset
, bytes
,
1214 cluster_offset
, cluster_bytes
);
1216 bounce_buffer
= qemu_try_blockalign(bs
,
1217 MIN(MIN(max_transfer
, cluster_bytes
),
1218 MAX_BOUNCE_BUFFER
));
1219 if (bounce_buffer
== NULL
) {
1224 while (cluster_bytes
) {
1227 ret
= bdrv_is_allocated(bs
, cluster_offset
,
1228 MIN(cluster_bytes
, max_transfer
), &pnum
);
1230 /* Safe to treat errors in querying allocation as if
1231 * unallocated; we'll probably fail again soon on the
1232 * read, but at least that will set a decent errno.
1234 pnum
= MIN(cluster_bytes
, max_transfer
);
1237 /* Stop at EOF if the image ends in the middle of the cluster */
1238 if (ret
== 0 && pnum
== 0) {
1239 assert(progress
>= bytes
);
1243 assert(skip_bytes
< pnum
);
1246 /* Must copy-on-read; use the bounce buffer */
1247 pnum
= MIN(pnum
, MAX_BOUNCE_BUFFER
);
1248 qemu_iovec_init_buf(&local_qiov
, bounce_buffer
, pnum
);
1250 ret
= bdrv_driver_preadv(bs
, cluster_offset
, pnum
,
1256 bdrv_debug_event(bs
, BLKDBG_COR_WRITE
);
1257 if (drv
->bdrv_co_pwrite_zeroes
&&
1258 buffer_is_zero(bounce_buffer
, pnum
)) {
1259 /* FIXME: Should we (perhaps conditionally) be setting
1260 * BDRV_REQ_MAY_UNMAP, if it will allow for a sparser copy
1261 * that still correctly reads as zero? */
1262 ret
= bdrv_co_do_pwrite_zeroes(bs
, cluster_offset
, pnum
,
1263 BDRV_REQ_WRITE_UNCHANGED
);
1265 /* This does not change the data on the disk, it is not
1266 * necessary to flush even in cache=writethrough mode.
1268 ret
= bdrv_driver_pwritev(bs
, cluster_offset
, pnum
,
1270 BDRV_REQ_WRITE_UNCHANGED
);
1274 /* It might be okay to ignore write errors for guest
1275 * requests. If this is a deliberate copy-on-read
1276 * then we don't want to ignore the error. Simply
1277 * report it in all cases.
1282 qemu_iovec_from_buf(qiov
, progress
, bounce_buffer
+ skip_bytes
,
1285 /* Read directly into the destination */
1286 qemu_iovec_init(&local_qiov
, qiov
->niov
);
1287 qemu_iovec_concat(&local_qiov
, qiov
, progress
, pnum
- skip_bytes
);
1288 ret
= bdrv_driver_preadv(bs
, offset
+ progress
, local_qiov
.size
,
1290 qemu_iovec_destroy(&local_qiov
);
1296 cluster_offset
+= pnum
;
1297 cluster_bytes
-= pnum
;
1298 progress
+= pnum
- skip_bytes
;
1304 qemu_vfree(bounce_buffer
);
1309 * Forwards an already correctly aligned request to the BlockDriver. This
1310 * handles copy on read, zeroing after EOF, and fragmentation of large
1311 * reads; any other features must be implemented by the caller.
1313 static int coroutine_fn
bdrv_aligned_preadv(BdrvChild
*child
,
1314 BdrvTrackedRequest
*req
, int64_t offset
, unsigned int bytes
,
1315 int64_t align
, QEMUIOVector
*qiov
, int flags
)
1317 BlockDriverState
*bs
= child
->bs
;
1318 int64_t total_bytes
, max_bytes
;
1320 uint64_t bytes_remaining
= bytes
;
1323 assert(is_power_of_2(align
));
1324 assert((offset
& (align
- 1)) == 0);
1325 assert((bytes
& (align
- 1)) == 0);
1326 assert(!qiov
|| bytes
== qiov
->size
);
1327 assert((bs
->open_flags
& BDRV_O_NO_IO
) == 0);
1328 max_transfer
= QEMU_ALIGN_DOWN(MIN_NON_ZERO(bs
->bl
.max_transfer
, INT_MAX
),
1331 /* TODO: We would need a per-BDS .supported_read_flags and
1332 * potential fallback support, if we ever implement any read flags
1333 * to pass through to drivers. For now, there aren't any
1334 * passthrough flags. */
1335 assert(!(flags
& ~(BDRV_REQ_NO_SERIALISING
| BDRV_REQ_COPY_ON_READ
)));
1337 /* Handle Copy on Read and associated serialisation */
1338 if (flags
& BDRV_REQ_COPY_ON_READ
) {
1339 /* If we touch the same cluster it counts as an overlap. This
1340 * guarantees that allocating writes will be serialized and not race
1341 * with each other for the same cluster. For example, in copy-on-read
1342 * it ensures that the CoR read and write operations are atomic and
1343 * guest writes cannot interleave between them. */
1344 mark_request_serialising(req
, bdrv_get_cluster_size(bs
));
1347 /* BDRV_REQ_SERIALISING is only for write operation */
1348 assert(!(flags
& BDRV_REQ_SERIALISING
));
1350 if (!(flags
& BDRV_REQ_NO_SERIALISING
)) {
1351 wait_serialising_requests(req
);
1354 if (flags
& BDRV_REQ_COPY_ON_READ
) {
1357 ret
= bdrv_is_allocated(bs
, offset
, bytes
, &pnum
);
1362 if (!ret
|| pnum
!= bytes
) {
1363 ret
= bdrv_co_do_copy_on_readv(child
, offset
, bytes
, qiov
);
1368 /* Forward the request to the BlockDriver, possibly fragmenting it */
1369 total_bytes
= bdrv_getlength(bs
);
1370 if (total_bytes
< 0) {
1375 max_bytes
= ROUND_UP(MAX(0, total_bytes
- offset
), align
);
1376 if (bytes
<= max_bytes
&& bytes
<= max_transfer
) {
1377 ret
= bdrv_driver_preadv(bs
, offset
, bytes
, qiov
, 0);
1381 while (bytes_remaining
) {
1385 QEMUIOVector local_qiov
;
1387 num
= MIN(bytes_remaining
, MIN(max_bytes
, max_transfer
));
1389 qemu_iovec_init(&local_qiov
, qiov
->niov
);
1390 qemu_iovec_concat(&local_qiov
, qiov
, bytes
- bytes_remaining
, num
);
1392 ret
= bdrv_driver_preadv(bs
, offset
+ bytes
- bytes_remaining
,
1393 num
, &local_qiov
, 0);
1395 qemu_iovec_destroy(&local_qiov
);
1397 num
= bytes_remaining
;
1398 ret
= qemu_iovec_memset(qiov
, bytes
- bytes_remaining
, 0,
1404 bytes_remaining
-= num
;
1408 return ret
< 0 ? ret
: 0;
1412 * Handle a read request in coroutine context
1414 int coroutine_fn
bdrv_co_preadv(BdrvChild
*child
,
1415 int64_t offset
, unsigned int bytes
, QEMUIOVector
*qiov
,
1416 BdrvRequestFlags flags
)
1418 BlockDriverState
*bs
= child
->bs
;
1419 BlockDriver
*drv
= bs
->drv
;
1420 BdrvTrackedRequest req
;
1422 uint64_t align
= bs
->bl
.request_alignment
;
1423 uint8_t *head_buf
= NULL
;
1424 uint8_t *tail_buf
= NULL
;
1425 QEMUIOVector local_qiov
;
1426 bool use_local_qiov
= false;
1429 trace_bdrv_co_preadv(child
->bs
, offset
, bytes
, flags
);
1435 ret
= bdrv_check_byte_request(bs
, offset
, bytes
);
1440 bdrv_inc_in_flight(bs
);
1442 /* Don't do copy-on-read if we read data before write operation */
1443 if (atomic_read(&bs
->copy_on_read
) && !(flags
& BDRV_REQ_NO_SERIALISING
)) {
1444 flags
|= BDRV_REQ_COPY_ON_READ
;
1447 /* Align read if necessary by padding qiov */
1448 if (offset
& (align
- 1)) {
1449 head_buf
= qemu_blockalign(bs
, align
);
1450 qemu_iovec_init(&local_qiov
, qiov
->niov
+ 2);
1451 qemu_iovec_add(&local_qiov
, head_buf
, offset
& (align
- 1));
1452 qemu_iovec_concat(&local_qiov
, qiov
, 0, qiov
->size
);
1453 use_local_qiov
= true;
1455 bytes
+= offset
& (align
- 1);
1456 offset
= offset
& ~(align
- 1);
1459 if ((offset
+ bytes
) & (align
- 1)) {
1460 if (!use_local_qiov
) {
1461 qemu_iovec_init(&local_qiov
, qiov
->niov
+ 1);
1462 qemu_iovec_concat(&local_qiov
, qiov
, 0, qiov
->size
);
1463 use_local_qiov
= true;
1465 tail_buf
= qemu_blockalign(bs
, align
);
1466 qemu_iovec_add(&local_qiov
, tail_buf
,
1467 align
- ((offset
+ bytes
) & (align
- 1)));
1469 bytes
= ROUND_UP(bytes
, align
);
1472 tracked_request_begin(&req
, bs
, offset
, bytes
, BDRV_TRACKED_READ
);
1473 ret
= bdrv_aligned_preadv(child
, &req
, offset
, bytes
, align
,
1474 use_local_qiov
? &local_qiov
: qiov
,
1476 tracked_request_end(&req
);
1477 bdrv_dec_in_flight(bs
);
1479 if (use_local_qiov
) {
1480 qemu_iovec_destroy(&local_qiov
);
1481 qemu_vfree(head_buf
);
1482 qemu_vfree(tail_buf
);
1488 static int coroutine_fn
bdrv_co_do_pwrite_zeroes(BlockDriverState
*bs
,
1489 int64_t offset
, int bytes
, BdrvRequestFlags flags
)
1491 BlockDriver
*drv
= bs
->drv
;
1495 bool need_flush
= false;
1499 int max_write_zeroes
= MIN_NON_ZERO(bs
->bl
.max_pwrite_zeroes
, INT_MAX
);
1500 int alignment
= MAX(bs
->bl
.pwrite_zeroes_alignment
,
1501 bs
->bl
.request_alignment
);
1502 int max_transfer
= MIN_NON_ZERO(bs
->bl
.max_transfer
, MAX_BOUNCE_BUFFER
);
1508 if ((flags
& ~bs
->supported_zero_flags
) & BDRV_REQ_NO_FALLBACK
) {
1512 assert(alignment
% bs
->bl
.request_alignment
== 0);
1513 head
= offset
% alignment
;
1514 tail
= (offset
+ bytes
) % alignment
;
1515 max_write_zeroes
= QEMU_ALIGN_DOWN(max_write_zeroes
, alignment
);
1516 assert(max_write_zeroes
>= bs
->bl
.request_alignment
);
1518 while (bytes
> 0 && !ret
) {
1521 /* Align request. Block drivers can expect the "bulk" of the request
1522 * to be aligned, and that unaligned requests do not cross cluster
1526 /* Make a small request up to the first aligned sector. For
1527 * convenience, limit this request to max_transfer even if
1528 * we don't need to fall back to writes. */
1529 num
= MIN(MIN(bytes
, max_transfer
), alignment
- head
);
1530 head
= (head
+ num
) % alignment
;
1531 assert(num
< max_write_zeroes
);
1532 } else if (tail
&& num
> alignment
) {
1533 /* Shorten the request to the last aligned sector. */
1537 /* limit request size */
1538 if (num
> max_write_zeroes
) {
1539 num
= max_write_zeroes
;
1543 /* First try the efficient write zeroes operation */
1544 if (drv
->bdrv_co_pwrite_zeroes
) {
1545 ret
= drv
->bdrv_co_pwrite_zeroes(bs
, offset
, num
,
1546 flags
& bs
->supported_zero_flags
);
1547 if (ret
!= -ENOTSUP
&& (flags
& BDRV_REQ_FUA
) &&
1548 !(bs
->supported_zero_flags
& BDRV_REQ_FUA
)) {
1552 assert(!bs
->supported_zero_flags
);
1555 if (ret
< 0 && !(flags
& BDRV_REQ_NO_FALLBACK
)) {
1556 /* Fall back to bounce buffer if write zeroes is unsupported */
1557 BdrvRequestFlags write_flags
= flags
& ~BDRV_REQ_ZERO_WRITE
;
1559 if ((flags
& BDRV_REQ_FUA
) &&
1560 !(bs
->supported_write_flags
& BDRV_REQ_FUA
)) {
1561 /* No need for bdrv_driver_pwrite() to do a fallback
1562 * flush on each chunk; use just one at the end */
1563 write_flags
&= ~BDRV_REQ_FUA
;
1566 num
= MIN(num
, max_transfer
);
1568 buf
= qemu_try_blockalign0(bs
, num
);
1574 qemu_iovec_init_buf(&qiov
, buf
, num
);
1576 ret
= bdrv_driver_pwritev(bs
, offset
, num
, &qiov
, write_flags
);
1578 /* Keep bounce buffer around if it is big enough for all
1579 * all future requests.
1581 if (num
< max_transfer
) {
1592 if (ret
== 0 && need_flush
) {
1593 ret
= bdrv_co_flush(bs
);
1599 static inline int coroutine_fn
1600 bdrv_co_write_req_prepare(BdrvChild
*child
, int64_t offset
, uint64_t bytes
,
1601 BdrvTrackedRequest
*req
, int flags
)
1603 BlockDriverState
*bs
= child
->bs
;
1605 int64_t end_sector
= DIV_ROUND_UP(offset
+ bytes
, BDRV_SECTOR_SIZE
);
1607 if (bs
->read_only
) {
1611 /* BDRV_REQ_NO_SERIALISING is only for read operation */
1612 assert(!(flags
& BDRV_REQ_NO_SERIALISING
));
1613 assert(!(bs
->open_flags
& BDRV_O_INACTIVE
));
1614 assert((bs
->open_flags
& BDRV_O_NO_IO
) == 0);
1615 assert(!(flags
& ~BDRV_REQ_MASK
));
1617 if (flags
& BDRV_REQ_SERIALISING
) {
1618 mark_request_serialising(req
, bdrv_get_cluster_size(bs
));
1621 waited
= wait_serialising_requests(req
);
1623 assert(!waited
|| !req
->serialising
||
1624 is_request_serialising_and_aligned(req
));
1625 assert(req
->overlap_offset
<= offset
);
1626 assert(offset
+ bytes
<= req
->overlap_offset
+ req
->overlap_bytes
);
1627 assert(end_sector
<= bs
->total_sectors
|| child
->perm
& BLK_PERM_RESIZE
);
1629 switch (req
->type
) {
1630 case BDRV_TRACKED_WRITE
:
1631 case BDRV_TRACKED_DISCARD
:
1632 if (flags
& BDRV_REQ_WRITE_UNCHANGED
) {
1633 assert(child
->perm
& (BLK_PERM_WRITE_UNCHANGED
| BLK_PERM_WRITE
));
1635 assert(child
->perm
& BLK_PERM_WRITE
);
1637 return notifier_with_return_list_notify(&bs
->before_write_notifiers
,
1639 case BDRV_TRACKED_TRUNCATE
:
1640 assert(child
->perm
& BLK_PERM_RESIZE
);
1647 static inline void coroutine_fn
1648 bdrv_co_write_req_finish(BdrvChild
*child
, int64_t offset
, uint64_t bytes
,
1649 BdrvTrackedRequest
*req
, int ret
)
1651 int64_t end_sector
= DIV_ROUND_UP(offset
+ bytes
, BDRV_SECTOR_SIZE
);
1652 BlockDriverState
*bs
= child
->bs
;
1654 atomic_inc(&bs
->write_gen
);
1657 * Discard cannot extend the image, but in error handling cases, such as
1658 * when reverting a qcow2 cluster allocation, the discarded range can pass
1659 * the end of image file, so we cannot assert about BDRV_TRACKED_DISCARD
1660 * here. Instead, just skip it, since semantically a discard request
1661 * beyond EOF cannot expand the image anyway.
1664 (req
->type
== BDRV_TRACKED_TRUNCATE
||
1665 end_sector
> bs
->total_sectors
) &&
1666 req
->type
!= BDRV_TRACKED_DISCARD
) {
1667 bs
->total_sectors
= end_sector
;
1668 bdrv_parent_cb_resize(bs
);
1669 bdrv_dirty_bitmap_truncate(bs
, end_sector
<< BDRV_SECTOR_BITS
);
1672 switch (req
->type
) {
1673 case BDRV_TRACKED_WRITE
:
1674 stat64_max(&bs
->wr_highest_offset
, offset
+ bytes
);
1675 /* fall through, to set dirty bits */
1676 case BDRV_TRACKED_DISCARD
:
1677 bdrv_set_dirty(bs
, offset
, bytes
);
1686 * Forwards an already correctly aligned write request to the BlockDriver,
1687 * after possibly fragmenting it.
1689 static int coroutine_fn
bdrv_aligned_pwritev(BdrvChild
*child
,
1690 BdrvTrackedRequest
*req
, int64_t offset
, unsigned int bytes
,
1691 int64_t align
, QEMUIOVector
*qiov
, int flags
)
1693 BlockDriverState
*bs
= child
->bs
;
1694 BlockDriver
*drv
= bs
->drv
;
1697 uint64_t bytes_remaining
= bytes
;
1704 if (bdrv_has_readonly_bitmaps(bs
)) {
1708 assert(is_power_of_2(align
));
1709 assert((offset
& (align
- 1)) == 0);
1710 assert((bytes
& (align
- 1)) == 0);
1711 assert(!qiov
|| bytes
== qiov
->size
);
1712 max_transfer
= QEMU_ALIGN_DOWN(MIN_NON_ZERO(bs
->bl
.max_transfer
, INT_MAX
),
1715 ret
= bdrv_co_write_req_prepare(child
, offset
, bytes
, req
, flags
);
1717 if (!ret
&& bs
->detect_zeroes
!= BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF
&&
1718 !(flags
& BDRV_REQ_ZERO_WRITE
) && drv
->bdrv_co_pwrite_zeroes
&&
1719 qemu_iovec_is_zero(qiov
)) {
1720 flags
|= BDRV_REQ_ZERO_WRITE
;
1721 if (bs
->detect_zeroes
== BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP
) {
1722 flags
|= BDRV_REQ_MAY_UNMAP
;
1727 /* Do nothing, write notifier decided to fail this request */
1728 } else if (flags
& BDRV_REQ_ZERO_WRITE
) {
1729 bdrv_debug_event(bs
, BLKDBG_PWRITEV_ZERO
);
1730 ret
= bdrv_co_do_pwrite_zeroes(bs
, offset
, bytes
, flags
);
1731 } else if (flags
& BDRV_REQ_WRITE_COMPRESSED
) {
1732 ret
= bdrv_driver_pwritev_compressed(bs
, offset
, bytes
, qiov
);
1733 } else if (bytes
<= max_transfer
) {
1734 bdrv_debug_event(bs
, BLKDBG_PWRITEV
);
1735 ret
= bdrv_driver_pwritev(bs
, offset
, bytes
, qiov
, flags
);
1737 bdrv_debug_event(bs
, BLKDBG_PWRITEV
);
1738 while (bytes_remaining
) {
1739 int num
= MIN(bytes_remaining
, max_transfer
);
1740 QEMUIOVector local_qiov
;
1741 int local_flags
= flags
;
1744 if (num
< bytes_remaining
&& (flags
& BDRV_REQ_FUA
) &&
1745 !(bs
->supported_write_flags
& BDRV_REQ_FUA
)) {
1746 /* If FUA is going to be emulated by flush, we only
1747 * need to flush on the last iteration */
1748 local_flags
&= ~BDRV_REQ_FUA
;
1750 qemu_iovec_init(&local_qiov
, qiov
->niov
);
1751 qemu_iovec_concat(&local_qiov
, qiov
, bytes
- bytes_remaining
, num
);
1753 ret
= bdrv_driver_pwritev(bs
, offset
+ bytes
- bytes_remaining
,
1754 num
, &local_qiov
, local_flags
);
1755 qemu_iovec_destroy(&local_qiov
);
1759 bytes_remaining
-= num
;
1762 bdrv_debug_event(bs
, BLKDBG_PWRITEV_DONE
);
1767 bdrv_co_write_req_finish(child
, offset
, bytes
, req
, ret
);
1772 static int coroutine_fn
bdrv_co_do_zero_pwritev(BdrvChild
*child
,
1775 BdrvRequestFlags flags
,
1776 BdrvTrackedRequest
*req
)
1778 BlockDriverState
*bs
= child
->bs
;
1779 uint8_t *buf
= NULL
;
1780 QEMUIOVector local_qiov
;
1781 uint64_t align
= bs
->bl
.request_alignment
;
1782 unsigned int head_padding_bytes
, tail_padding_bytes
;
1785 head_padding_bytes
= offset
& (align
- 1);
1786 tail_padding_bytes
= (align
- (offset
+ bytes
)) & (align
- 1);
1789 assert(flags
& BDRV_REQ_ZERO_WRITE
);
1790 if (head_padding_bytes
|| tail_padding_bytes
) {
1791 buf
= qemu_blockalign(bs
, align
);
1792 qemu_iovec_init_buf(&local_qiov
, buf
, align
);
1794 if (head_padding_bytes
) {
1795 uint64_t zero_bytes
= MIN(bytes
, align
- head_padding_bytes
);
1797 /* RMW the unaligned part before head. */
1798 mark_request_serialising(req
, align
);
1799 wait_serialising_requests(req
);
1800 bdrv_debug_event(bs
, BLKDBG_PWRITEV_RMW_HEAD
);
1801 ret
= bdrv_aligned_preadv(child
, req
, offset
& ~(align
- 1), align
,
1802 align
, &local_qiov
, 0);
1806 bdrv_debug_event(bs
, BLKDBG_PWRITEV_RMW_AFTER_HEAD
);
1808 memset(buf
+ head_padding_bytes
, 0, zero_bytes
);
1809 ret
= bdrv_aligned_pwritev(child
, req
, offset
& ~(align
- 1), align
,
1811 flags
& ~BDRV_REQ_ZERO_WRITE
);
1815 offset
+= zero_bytes
;
1816 bytes
-= zero_bytes
;
1819 assert(!bytes
|| (offset
& (align
- 1)) == 0);
1820 if (bytes
>= align
) {
1821 /* Write the aligned part in the middle. */
1822 uint64_t aligned_bytes
= bytes
& ~(align
- 1);
1823 ret
= bdrv_aligned_pwritev(child
, req
, offset
, aligned_bytes
, align
,
1828 bytes
-= aligned_bytes
;
1829 offset
+= aligned_bytes
;
1832 assert(!bytes
|| (offset
& (align
- 1)) == 0);
1834 assert(align
== tail_padding_bytes
+ bytes
);
1835 /* RMW the unaligned part after tail. */
1836 mark_request_serialising(req
, align
);
1837 wait_serialising_requests(req
);
1838 bdrv_debug_event(bs
, BLKDBG_PWRITEV_RMW_TAIL
);
1839 ret
= bdrv_aligned_preadv(child
, req
, offset
, align
,
1840 align
, &local_qiov
, 0);
1844 bdrv_debug_event(bs
, BLKDBG_PWRITEV_RMW_AFTER_TAIL
);
1846 memset(buf
, 0, bytes
);
1847 ret
= bdrv_aligned_pwritev(child
, req
, offset
, align
, align
,
1848 &local_qiov
, flags
& ~BDRV_REQ_ZERO_WRITE
);
1857 * Handle a write request in coroutine context
1859 int coroutine_fn
bdrv_co_pwritev(BdrvChild
*child
,
1860 int64_t offset
, unsigned int bytes
, QEMUIOVector
*qiov
,
1861 BdrvRequestFlags flags
)
1863 BlockDriverState
*bs
= child
->bs
;
1864 BdrvTrackedRequest req
;
1865 uint64_t align
= bs
->bl
.request_alignment
;
1866 uint8_t *head_buf
= NULL
;
1867 uint8_t *tail_buf
= NULL
;
1868 QEMUIOVector local_qiov
;
1869 bool use_local_qiov
= false;
1872 trace_bdrv_co_pwritev(child
->bs
, offset
, bytes
, flags
);
1878 ret
= bdrv_check_byte_request(bs
, offset
, bytes
);
1883 bdrv_inc_in_flight(bs
);
1885 * Align write if necessary by performing a read-modify-write cycle.
1886 * Pad qiov with the read parts and be sure to have a tracked request not
1887 * only for bdrv_aligned_pwritev, but also for the reads of the RMW cycle.
1889 tracked_request_begin(&req
, bs
, offset
, bytes
, BDRV_TRACKED_WRITE
);
1891 if (flags
& BDRV_REQ_ZERO_WRITE
) {
1892 ret
= bdrv_co_do_zero_pwritev(child
, offset
, bytes
, flags
, &req
);
1896 if (offset
& (align
- 1)) {
1897 QEMUIOVector head_qiov
;
1899 mark_request_serialising(&req
, align
);
1900 wait_serialising_requests(&req
);
1902 head_buf
= qemu_blockalign(bs
, align
);
1903 qemu_iovec_init_buf(&head_qiov
, head_buf
, align
);
1905 bdrv_debug_event(bs
, BLKDBG_PWRITEV_RMW_HEAD
);
1906 ret
= bdrv_aligned_preadv(child
, &req
, offset
& ~(align
- 1), align
,
1907 align
, &head_qiov
, 0);
1911 bdrv_debug_event(bs
, BLKDBG_PWRITEV_RMW_AFTER_HEAD
);
1913 qemu_iovec_init(&local_qiov
, qiov
->niov
+ 2);
1914 qemu_iovec_add(&local_qiov
, head_buf
, offset
& (align
- 1));
1915 qemu_iovec_concat(&local_qiov
, qiov
, 0, qiov
->size
);
1916 use_local_qiov
= true;
1918 bytes
+= offset
& (align
- 1);
1919 offset
= offset
& ~(align
- 1);
1921 /* We have read the tail already if the request is smaller
1922 * than one aligned block.
1924 if (bytes
< align
) {
1925 qemu_iovec_add(&local_qiov
, head_buf
+ bytes
, align
- bytes
);
1930 if ((offset
+ bytes
) & (align
- 1)) {
1931 QEMUIOVector tail_qiov
;
1935 mark_request_serialising(&req
, align
);
1936 waited
= wait_serialising_requests(&req
);
1937 assert(!waited
|| !use_local_qiov
);
1939 tail_buf
= qemu_blockalign(bs
, align
);
1940 qemu_iovec_init_buf(&tail_qiov
, tail_buf
, align
);
1942 bdrv_debug_event(bs
, BLKDBG_PWRITEV_RMW_TAIL
);
1943 ret
= bdrv_aligned_preadv(child
, &req
, (offset
+ bytes
) & ~(align
- 1),
1944 align
, align
, &tail_qiov
, 0);
1948 bdrv_debug_event(bs
, BLKDBG_PWRITEV_RMW_AFTER_TAIL
);
1950 if (!use_local_qiov
) {
1951 qemu_iovec_init(&local_qiov
, qiov
->niov
+ 1);
1952 qemu_iovec_concat(&local_qiov
, qiov
, 0, qiov
->size
);
1953 use_local_qiov
= true;
1956 tail_bytes
= (offset
+ bytes
) & (align
- 1);
1957 qemu_iovec_add(&local_qiov
, tail_buf
+ tail_bytes
, align
- tail_bytes
);
1959 bytes
= ROUND_UP(bytes
, align
);
1962 ret
= bdrv_aligned_pwritev(child
, &req
, offset
, bytes
, align
,
1963 use_local_qiov
? &local_qiov
: qiov
,
1968 if (use_local_qiov
) {
1969 qemu_iovec_destroy(&local_qiov
);
1971 qemu_vfree(head_buf
);
1972 qemu_vfree(tail_buf
);
1974 tracked_request_end(&req
);
1975 bdrv_dec_in_flight(bs
);
1979 int coroutine_fn
bdrv_co_pwrite_zeroes(BdrvChild
*child
, int64_t offset
,
1980 int bytes
, BdrvRequestFlags flags
)
1982 trace_bdrv_co_pwrite_zeroes(child
->bs
, offset
, bytes
, flags
);
1984 if (!(child
->bs
->open_flags
& BDRV_O_UNMAP
)) {
1985 flags
&= ~BDRV_REQ_MAY_UNMAP
;
1988 return bdrv_co_pwritev(child
, offset
, bytes
, NULL
,
1989 BDRV_REQ_ZERO_WRITE
| flags
);
1993 * Flush ALL BDSes regardless of if they are reachable via a BlkBackend or not.
1995 int bdrv_flush_all(void)
1997 BdrvNextIterator it
;
1998 BlockDriverState
*bs
= NULL
;
2001 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
2002 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
2005 aio_context_acquire(aio_context
);
2006 ret
= bdrv_flush(bs
);
2007 if (ret
< 0 && !result
) {
2010 aio_context_release(aio_context
);
2017 typedef struct BdrvCoBlockStatusData
{
2018 BlockDriverState
*bs
;
2019 BlockDriverState
*base
;
2025 BlockDriverState
**file
;
2028 } BdrvCoBlockStatusData
;
2030 int coroutine_fn
bdrv_co_block_status_from_file(BlockDriverState
*bs
,
2036 BlockDriverState
**file
)
2038 assert(bs
->file
&& bs
->file
->bs
);
2041 *file
= bs
->file
->bs
;
2042 return BDRV_BLOCK_RAW
| BDRV_BLOCK_OFFSET_VALID
;
2045 int coroutine_fn
bdrv_co_block_status_from_backing(BlockDriverState
*bs
,
2051 BlockDriverState
**file
)
2053 assert(bs
->backing
&& bs
->backing
->bs
);
2056 *file
= bs
->backing
->bs
;
2057 return BDRV_BLOCK_RAW
| BDRV_BLOCK_OFFSET_VALID
;
2061 * Returns the allocation status of the specified sectors.
2062 * Drivers not implementing the functionality are assumed to not support
2063 * backing files, hence all their sectors are reported as allocated.
2065 * If 'want_zero' is true, the caller is querying for mapping
2066 * purposes, with a focus on valid BDRV_BLOCK_OFFSET_VALID, _DATA, and
2067 * _ZERO where possible; otherwise, the result favors larger 'pnum',
2068 * with a focus on accurate BDRV_BLOCK_ALLOCATED.
2070 * If 'offset' is beyond the end of the disk image the return value is
2071 * BDRV_BLOCK_EOF and 'pnum' is set to 0.
2073 * 'bytes' is the max value 'pnum' should be set to. If bytes goes
2074 * beyond the end of the disk image it will be clamped; if 'pnum' is set to
2075 * the end of the image, then the returned value will include BDRV_BLOCK_EOF.
2077 * 'pnum' is set to the number of bytes (including and immediately
2078 * following the specified offset) that are easily known to be in the
2079 * same allocated/unallocated state. Note that a second call starting
2080 * at the original offset plus returned pnum may have the same status.
2081 * The returned value is non-zero on success except at end-of-file.
2083 * Returns negative errno on failure. Otherwise, if the
2084 * BDRV_BLOCK_OFFSET_VALID bit is set, 'map' and 'file' (if non-NULL) are
2085 * set to the host mapping and BDS corresponding to the guest offset.
2087 static int coroutine_fn
bdrv_co_block_status(BlockDriverState
*bs
,
2089 int64_t offset
, int64_t bytes
,
2090 int64_t *pnum
, int64_t *map
,
2091 BlockDriverState
**file
)
2094 int64_t n
; /* bytes */
2096 int64_t local_map
= 0;
2097 BlockDriverState
*local_file
= NULL
;
2098 int64_t aligned_offset
, aligned_bytes
;
2103 total_size
= bdrv_getlength(bs
);
2104 if (total_size
< 0) {
2109 if (offset
>= total_size
) {
2110 ret
= BDRV_BLOCK_EOF
;
2118 n
= total_size
- offset
;
2123 /* Must be non-NULL or bdrv_getlength() would have failed */
2125 if (!bs
->drv
->bdrv_co_block_status
) {
2127 ret
= BDRV_BLOCK_DATA
| BDRV_BLOCK_ALLOCATED
;
2128 if (offset
+ bytes
== total_size
) {
2129 ret
|= BDRV_BLOCK_EOF
;
2131 if (bs
->drv
->protocol_name
) {
2132 ret
|= BDRV_BLOCK_OFFSET_VALID
;
2139 bdrv_inc_in_flight(bs
);
2141 /* Round out to request_alignment boundaries */
2142 align
= bs
->bl
.request_alignment
;
2143 aligned_offset
= QEMU_ALIGN_DOWN(offset
, align
);
2144 aligned_bytes
= ROUND_UP(offset
+ bytes
, align
) - aligned_offset
;
2146 ret
= bs
->drv
->bdrv_co_block_status(bs
, want_zero
, aligned_offset
,
2147 aligned_bytes
, pnum
, &local_map
,
2155 * The driver's result must be a non-zero multiple of request_alignment.
2156 * Clamp pnum and adjust map to original request.
2158 assert(*pnum
&& QEMU_IS_ALIGNED(*pnum
, align
) &&
2159 align
> offset
- aligned_offset
);
2160 if (ret
& BDRV_BLOCK_RECURSE
) {
2161 assert(ret
& BDRV_BLOCK_DATA
);
2162 assert(ret
& BDRV_BLOCK_OFFSET_VALID
);
2163 assert(!(ret
& BDRV_BLOCK_ZERO
));
2166 *pnum
-= offset
- aligned_offset
;
2167 if (*pnum
> bytes
) {
2170 if (ret
& BDRV_BLOCK_OFFSET_VALID
) {
2171 local_map
+= offset
- aligned_offset
;
2174 if (ret
& BDRV_BLOCK_RAW
) {
2175 assert(ret
& BDRV_BLOCK_OFFSET_VALID
&& local_file
);
2176 ret
= bdrv_co_block_status(local_file
, want_zero
, local_map
,
2177 *pnum
, pnum
, &local_map
, &local_file
);
2181 if (ret
& (BDRV_BLOCK_DATA
| BDRV_BLOCK_ZERO
)) {
2182 ret
|= BDRV_BLOCK_ALLOCATED
;
2183 } else if (want_zero
) {
2184 if (bdrv_unallocated_blocks_are_zero(bs
)) {
2185 ret
|= BDRV_BLOCK_ZERO
;
2186 } else if (bs
->backing
) {
2187 BlockDriverState
*bs2
= bs
->backing
->bs
;
2188 int64_t size2
= bdrv_getlength(bs2
);
2190 if (size2
>= 0 && offset
>= size2
) {
2191 ret
|= BDRV_BLOCK_ZERO
;
2196 if (want_zero
&& ret
& BDRV_BLOCK_RECURSE
&&
2197 local_file
&& local_file
!= bs
&&
2198 (ret
& BDRV_BLOCK_DATA
) && !(ret
& BDRV_BLOCK_ZERO
) &&
2199 (ret
& BDRV_BLOCK_OFFSET_VALID
)) {
2203 ret2
= bdrv_co_block_status(local_file
, want_zero
, local_map
,
2204 *pnum
, &file_pnum
, NULL
, NULL
);
2206 /* Ignore errors. This is just providing extra information, it
2207 * is useful but not necessary.
2209 if (ret2
& BDRV_BLOCK_EOF
&&
2210 (!file_pnum
|| ret2
& BDRV_BLOCK_ZERO
)) {
2212 * It is valid for the format block driver to read
2213 * beyond the end of the underlying file's current
2214 * size; such areas read as zero.
2216 ret
|= BDRV_BLOCK_ZERO
;
2218 /* Limit request to the range reported by the protocol driver */
2220 ret
|= (ret2
& BDRV_BLOCK_ZERO
);
2226 bdrv_dec_in_flight(bs
);
2227 if (ret
>= 0 && offset
+ *pnum
== total_size
) {
2228 ret
|= BDRV_BLOCK_EOF
;
2240 static int coroutine_fn
bdrv_co_block_status_above(BlockDriverState
*bs
,
2241 BlockDriverState
*base
,
2247 BlockDriverState
**file
)
2249 BlockDriverState
*p
;
2254 for (p
= bs
; p
!= base
; p
= backing_bs(p
)) {
2255 ret
= bdrv_co_block_status(p
, want_zero
, offset
, bytes
, pnum
, map
,
2260 if (ret
& BDRV_BLOCK_ZERO
&& ret
& BDRV_BLOCK_EOF
&& !first
) {
2262 * Reading beyond the end of the file continues to read
2263 * zeroes, but we can only widen the result to the
2264 * unallocated length we learned from an earlier
2269 if (ret
& (BDRV_BLOCK_ZERO
| BDRV_BLOCK_DATA
)) {
2272 /* [offset, pnum] unallocated on this layer, which could be only
2273 * the first part of [offset, bytes]. */
2274 bytes
= MIN(bytes
, *pnum
);
2280 /* Coroutine wrapper for bdrv_block_status_above() */
2281 static void coroutine_fn
bdrv_block_status_above_co_entry(void *opaque
)
2283 BdrvCoBlockStatusData
*data
= opaque
;
2285 data
->ret
= bdrv_co_block_status_above(data
->bs
, data
->base
,
2287 data
->offset
, data
->bytes
,
2288 data
->pnum
, data
->map
, data
->file
);
2294 * Synchronous wrapper around bdrv_co_block_status_above().
2296 * See bdrv_co_block_status_above() for details.
2298 static int bdrv_common_block_status_above(BlockDriverState
*bs
,
2299 BlockDriverState
*base
,
2300 bool want_zero
, int64_t offset
,
2301 int64_t bytes
, int64_t *pnum
,
2303 BlockDriverState
**file
)
2306 BdrvCoBlockStatusData data
= {
2309 .want_zero
= want_zero
,
2318 if (qemu_in_coroutine()) {
2319 /* Fast-path if already in coroutine context */
2320 bdrv_block_status_above_co_entry(&data
);
2322 co
= qemu_coroutine_create(bdrv_block_status_above_co_entry
, &data
);
2323 bdrv_coroutine_enter(bs
, co
);
2324 BDRV_POLL_WHILE(bs
, !data
.done
);
2329 int bdrv_block_status_above(BlockDriverState
*bs
, BlockDriverState
*base
,
2330 int64_t offset
, int64_t bytes
, int64_t *pnum
,
2331 int64_t *map
, BlockDriverState
**file
)
2333 return bdrv_common_block_status_above(bs
, base
, true, offset
, bytes
,
2337 int bdrv_block_status(BlockDriverState
*bs
, int64_t offset
, int64_t bytes
,
2338 int64_t *pnum
, int64_t *map
, BlockDriverState
**file
)
2340 return bdrv_block_status_above(bs
, backing_bs(bs
),
2341 offset
, bytes
, pnum
, map
, file
);
2344 int coroutine_fn
bdrv_is_allocated(BlockDriverState
*bs
, int64_t offset
,
2345 int64_t bytes
, int64_t *pnum
)
2350 ret
= bdrv_common_block_status_above(bs
, backing_bs(bs
), false, offset
,
2351 bytes
, pnum
? pnum
: &dummy
, NULL
,
2356 return !!(ret
& BDRV_BLOCK_ALLOCATED
);
2360 * Given an image chain: ... -> [BASE] -> [INTER1] -> [INTER2] -> [TOP]
2362 * Return 1 if (a prefix of) the given range is allocated in any image
2363 * between BASE and TOP (BASE is only included if include_base is set).
2364 * BASE can be NULL to check if the given offset is allocated in any
2365 * image of the chain. Return 0 otherwise, or negative errno on
2368 * 'pnum' is set to the number of bytes (including and immediately
2369 * following the specified offset) that are known to be in the same
2370 * allocated/unallocated state. Note that a subsequent call starting
2371 * at 'offset + *pnum' may return the same allocation status (in other
2372 * words, the result is not necessarily the maximum possible range);
2373 * but 'pnum' will only be 0 when end of file is reached.
2376 int bdrv_is_allocated_above(BlockDriverState
*top
,
2377 BlockDriverState
*base
,
2378 bool include_base
, int64_t offset
,
2379 int64_t bytes
, int64_t *pnum
)
2381 BlockDriverState
*intermediate
;
2385 assert(base
|| !include_base
);
2388 while (include_base
|| intermediate
!= base
) {
2392 assert(intermediate
);
2393 ret
= bdrv_is_allocated(intermediate
, offset
, bytes
, &pnum_inter
);
2402 size_inter
= bdrv_getlength(intermediate
);
2403 if (size_inter
< 0) {
2406 if (n
> pnum_inter
&&
2407 (intermediate
== top
|| offset
+ pnum_inter
< size_inter
)) {
2411 if (intermediate
== base
) {
2415 intermediate
= backing_bs(intermediate
);
2422 typedef struct BdrvVmstateCo
{
2423 BlockDriverState
*bs
;
2430 static int coroutine_fn
2431 bdrv_co_rw_vmstate(BlockDriverState
*bs
, QEMUIOVector
*qiov
, int64_t pos
,
2434 BlockDriver
*drv
= bs
->drv
;
2437 bdrv_inc_in_flight(bs
);
2441 } else if (drv
->bdrv_load_vmstate
) {
2443 ret
= drv
->bdrv_load_vmstate(bs
, qiov
, pos
);
2445 ret
= drv
->bdrv_save_vmstate(bs
, qiov
, pos
);
2447 } else if (bs
->file
) {
2448 ret
= bdrv_co_rw_vmstate(bs
->file
->bs
, qiov
, pos
, is_read
);
2451 bdrv_dec_in_flight(bs
);
2455 static void coroutine_fn
bdrv_co_rw_vmstate_entry(void *opaque
)
2457 BdrvVmstateCo
*co
= opaque
;
2458 co
->ret
= bdrv_co_rw_vmstate(co
->bs
, co
->qiov
, co
->pos
, co
->is_read
);
2463 bdrv_rw_vmstate(BlockDriverState
*bs
, QEMUIOVector
*qiov
, int64_t pos
,
2466 if (qemu_in_coroutine()) {
2467 return bdrv_co_rw_vmstate(bs
, qiov
, pos
, is_read
);
2469 BdrvVmstateCo data
= {
2474 .ret
= -EINPROGRESS
,
2476 Coroutine
*co
= qemu_coroutine_create(bdrv_co_rw_vmstate_entry
, &data
);
2478 bdrv_coroutine_enter(bs
, co
);
2479 BDRV_POLL_WHILE(bs
, data
.ret
== -EINPROGRESS
);
2484 int bdrv_save_vmstate(BlockDriverState
*bs
, const uint8_t *buf
,
2485 int64_t pos
, int size
)
2487 QEMUIOVector qiov
= QEMU_IOVEC_INIT_BUF(qiov
, buf
, size
);
2490 ret
= bdrv_writev_vmstate(bs
, &qiov
, pos
);
2498 int bdrv_writev_vmstate(BlockDriverState
*bs
, QEMUIOVector
*qiov
, int64_t pos
)
2500 return bdrv_rw_vmstate(bs
, qiov
, pos
, false);
2503 int bdrv_load_vmstate(BlockDriverState
*bs
, uint8_t *buf
,
2504 int64_t pos
, int size
)
2506 QEMUIOVector qiov
= QEMU_IOVEC_INIT_BUF(qiov
, buf
, size
);
2509 ret
= bdrv_readv_vmstate(bs
, &qiov
, pos
);
2517 int bdrv_readv_vmstate(BlockDriverState
*bs
, QEMUIOVector
*qiov
, int64_t pos
)
2519 return bdrv_rw_vmstate(bs
, qiov
, pos
, true);
2522 /**************************************************************/
2525 void bdrv_aio_cancel(BlockAIOCB
*acb
)
2528 bdrv_aio_cancel_async(acb
);
2529 while (acb
->refcnt
> 1) {
2530 if (acb
->aiocb_info
->get_aio_context
) {
2531 aio_poll(acb
->aiocb_info
->get_aio_context(acb
), true);
2532 } else if (acb
->bs
) {
2533 /* qemu_aio_ref and qemu_aio_unref are not thread-safe, so
2534 * assert that we're not using an I/O thread. Thread-safe
2535 * code should use bdrv_aio_cancel_async exclusively.
2537 assert(bdrv_get_aio_context(acb
->bs
) == qemu_get_aio_context());
2538 aio_poll(bdrv_get_aio_context(acb
->bs
), true);
2543 qemu_aio_unref(acb
);
2546 /* Async version of aio cancel. The caller is not blocked if the acb implements
2547 * cancel_async, otherwise we do nothing and let the request normally complete.
2548 * In either case the completion callback must be called. */
2549 void bdrv_aio_cancel_async(BlockAIOCB
*acb
)
2551 if (acb
->aiocb_info
->cancel_async
) {
2552 acb
->aiocb_info
->cancel_async(acb
);
2556 /**************************************************************/
2557 /* Coroutine block device emulation */
2559 typedef struct FlushCo
{
2560 BlockDriverState
*bs
;
2565 static void coroutine_fn
bdrv_flush_co_entry(void *opaque
)
2567 FlushCo
*rwco
= opaque
;
2569 rwco
->ret
= bdrv_co_flush(rwco
->bs
);
2573 int coroutine_fn
bdrv_co_flush(BlockDriverState
*bs
)
2578 bdrv_inc_in_flight(bs
);
2580 if (!bdrv_is_inserted(bs
) || bdrv_is_read_only(bs
) ||
2585 qemu_co_mutex_lock(&bs
->reqs_lock
);
2586 current_gen
= atomic_read(&bs
->write_gen
);
2588 /* Wait until any previous flushes are completed */
2589 while (bs
->active_flush_req
) {
2590 qemu_co_queue_wait(&bs
->flush_queue
, &bs
->reqs_lock
);
2593 /* Flushes reach this point in nondecreasing current_gen order. */
2594 bs
->active_flush_req
= true;
2595 qemu_co_mutex_unlock(&bs
->reqs_lock
);
2597 /* Write back all layers by calling one driver function */
2598 if (bs
->drv
->bdrv_co_flush
) {
2599 ret
= bs
->drv
->bdrv_co_flush(bs
);
2603 /* Write back cached data to the OS even with cache=unsafe */
2604 BLKDBG_EVENT(bs
->file
, BLKDBG_FLUSH_TO_OS
);
2605 if (bs
->drv
->bdrv_co_flush_to_os
) {
2606 ret
= bs
->drv
->bdrv_co_flush_to_os(bs
);
2612 /* But don't actually force it to the disk with cache=unsafe */
2613 if (bs
->open_flags
& BDRV_O_NO_FLUSH
) {
2617 /* Check if we really need to flush anything */
2618 if (bs
->flushed_gen
== current_gen
) {
2622 BLKDBG_EVENT(bs
->file
, BLKDBG_FLUSH_TO_DISK
);
2624 /* bs->drv->bdrv_co_flush() might have ejected the BDS
2625 * (even in case of apparent success) */
2629 if (bs
->drv
->bdrv_co_flush_to_disk
) {
2630 ret
= bs
->drv
->bdrv_co_flush_to_disk(bs
);
2631 } else if (bs
->drv
->bdrv_aio_flush
) {
2633 CoroutineIOCompletion co
= {
2634 .coroutine
= qemu_coroutine_self(),
2637 acb
= bs
->drv
->bdrv_aio_flush(bs
, bdrv_co_io_em_complete
, &co
);
2641 qemu_coroutine_yield();
2646 * Some block drivers always operate in either writethrough or unsafe
2647 * mode and don't support bdrv_flush therefore. Usually qemu doesn't
2648 * know how the server works (because the behaviour is hardcoded or
2649 * depends on server-side configuration), so we can't ensure that
2650 * everything is safe on disk. Returning an error doesn't work because
2651 * that would break guests even if the server operates in writethrough
2654 * Let's hope the user knows what he's doing.
2663 /* Now flush the underlying protocol. It will also have BDRV_O_NO_FLUSH
2664 * in the case of cache=unsafe, so there are no useless flushes.
2667 ret
= bs
->file
? bdrv_co_flush(bs
->file
->bs
) : 0;
2669 /* Notify any pending flushes that we have completed */
2671 bs
->flushed_gen
= current_gen
;
2674 qemu_co_mutex_lock(&bs
->reqs_lock
);
2675 bs
->active_flush_req
= false;
2676 /* Return value is ignored - it's ok if wait queue is empty */
2677 qemu_co_queue_next(&bs
->flush_queue
);
2678 qemu_co_mutex_unlock(&bs
->reqs_lock
);
2681 bdrv_dec_in_flight(bs
);
2685 int bdrv_flush(BlockDriverState
*bs
)
2688 FlushCo flush_co
= {
2693 if (qemu_in_coroutine()) {
2694 /* Fast-path if already in coroutine context */
2695 bdrv_flush_co_entry(&flush_co
);
2697 co
= qemu_coroutine_create(bdrv_flush_co_entry
, &flush_co
);
2698 bdrv_coroutine_enter(bs
, co
);
2699 BDRV_POLL_WHILE(bs
, flush_co
.ret
== NOT_DONE
);
2702 return flush_co
.ret
;
2705 typedef struct DiscardCo
{
2711 static void coroutine_fn
bdrv_pdiscard_co_entry(void *opaque
)
2713 DiscardCo
*rwco
= opaque
;
2715 rwco
->ret
= bdrv_co_pdiscard(rwco
->child
, rwco
->offset
, rwco
->bytes
);
2719 int coroutine_fn
bdrv_co_pdiscard(BdrvChild
*child
, int64_t offset
,
2722 BdrvTrackedRequest req
;
2723 int max_pdiscard
, ret
;
2724 int head
, tail
, align
;
2725 BlockDriverState
*bs
= child
->bs
;
2727 if (!bs
|| !bs
->drv
|| !bdrv_is_inserted(bs
)) {
2731 if (bdrv_has_readonly_bitmaps(bs
)) {
2735 if (offset
< 0 || bytes
< 0 || bytes
> INT64_MAX
- offset
) {
2739 /* Do nothing if disabled. */
2740 if (!(bs
->open_flags
& BDRV_O_UNMAP
)) {
2744 if (!bs
->drv
->bdrv_co_pdiscard
&& !bs
->drv
->bdrv_aio_pdiscard
) {
2748 /* Discard is advisory, but some devices track and coalesce
2749 * unaligned requests, so we must pass everything down rather than
2750 * round here. Still, most devices will just silently ignore
2751 * unaligned requests (by returning -ENOTSUP), so we must fragment
2752 * the request accordingly. */
2753 align
= MAX(bs
->bl
.pdiscard_alignment
, bs
->bl
.request_alignment
);
2754 assert(align
% bs
->bl
.request_alignment
== 0);
2755 head
= offset
% align
;
2756 tail
= (offset
+ bytes
) % align
;
2758 bdrv_inc_in_flight(bs
);
2759 tracked_request_begin(&req
, bs
, offset
, bytes
, BDRV_TRACKED_DISCARD
);
2761 ret
= bdrv_co_write_req_prepare(child
, offset
, bytes
, &req
, 0);
2766 max_pdiscard
= QEMU_ALIGN_DOWN(MIN_NON_ZERO(bs
->bl
.max_pdiscard
, INT_MAX
),
2768 assert(max_pdiscard
>= bs
->bl
.request_alignment
);
2771 int64_t num
= bytes
;
2774 /* Make small requests to get to alignment boundaries. */
2775 num
= MIN(bytes
, align
- head
);
2776 if (!QEMU_IS_ALIGNED(num
, bs
->bl
.request_alignment
)) {
2777 num
%= bs
->bl
.request_alignment
;
2779 head
= (head
+ num
) % align
;
2780 assert(num
< max_pdiscard
);
2783 /* Shorten the request to the last aligned cluster. */
2785 } else if (!QEMU_IS_ALIGNED(tail
, bs
->bl
.request_alignment
) &&
2786 tail
> bs
->bl
.request_alignment
) {
2787 tail
%= bs
->bl
.request_alignment
;
2791 /* limit request size */
2792 if (num
> max_pdiscard
) {
2800 if (bs
->drv
->bdrv_co_pdiscard
) {
2801 ret
= bs
->drv
->bdrv_co_pdiscard(bs
, offset
, num
);
2804 CoroutineIOCompletion co
= {
2805 .coroutine
= qemu_coroutine_self(),
2808 acb
= bs
->drv
->bdrv_aio_pdiscard(bs
, offset
, num
,
2809 bdrv_co_io_em_complete
, &co
);
2814 qemu_coroutine_yield();
2818 if (ret
&& ret
!= -ENOTSUP
) {
2827 bdrv_co_write_req_finish(child
, req
.offset
, req
.bytes
, &req
, ret
);
2828 tracked_request_end(&req
);
2829 bdrv_dec_in_flight(bs
);
2833 int bdrv_pdiscard(BdrvChild
*child
, int64_t offset
, int64_t bytes
)
2843 if (qemu_in_coroutine()) {
2844 /* Fast-path if already in coroutine context */
2845 bdrv_pdiscard_co_entry(&rwco
);
2847 co
= qemu_coroutine_create(bdrv_pdiscard_co_entry
, &rwco
);
2848 bdrv_coroutine_enter(child
->bs
, co
);
2849 BDRV_POLL_WHILE(child
->bs
, rwco
.ret
== NOT_DONE
);
2855 int bdrv_co_ioctl(BlockDriverState
*bs
, int req
, void *buf
)
2857 BlockDriver
*drv
= bs
->drv
;
2858 CoroutineIOCompletion co
= {
2859 .coroutine
= qemu_coroutine_self(),
2863 bdrv_inc_in_flight(bs
);
2864 if (!drv
|| (!drv
->bdrv_aio_ioctl
&& !drv
->bdrv_co_ioctl
)) {
2869 if (drv
->bdrv_co_ioctl
) {
2870 co
.ret
= drv
->bdrv_co_ioctl(bs
, req
, buf
);
2872 acb
= drv
->bdrv_aio_ioctl(bs
, req
, buf
, bdrv_co_io_em_complete
, &co
);
2877 qemu_coroutine_yield();
2880 bdrv_dec_in_flight(bs
);
2884 void *qemu_blockalign(BlockDriverState
*bs
, size_t size
)
2886 return qemu_memalign(bdrv_opt_mem_align(bs
), size
);
2889 void *qemu_blockalign0(BlockDriverState
*bs
, size_t size
)
2891 return memset(qemu_blockalign(bs
, size
), 0, size
);
2894 void *qemu_try_blockalign(BlockDriverState
*bs
, size_t size
)
2896 size_t align
= bdrv_opt_mem_align(bs
);
2898 /* Ensure that NULL is never returned on success */
2904 return qemu_try_memalign(align
, size
);
2907 void *qemu_try_blockalign0(BlockDriverState
*bs
, size_t size
)
2909 void *mem
= qemu_try_blockalign(bs
, size
);
2912 memset(mem
, 0, size
);
2919 * Check if all memory in this vector is sector aligned.
2921 bool bdrv_qiov_is_aligned(BlockDriverState
*bs
, QEMUIOVector
*qiov
)
2924 size_t alignment
= bdrv_min_mem_align(bs
);
2926 for (i
= 0; i
< qiov
->niov
; i
++) {
2927 if ((uintptr_t) qiov
->iov
[i
].iov_base
% alignment
) {
2930 if (qiov
->iov
[i
].iov_len
% alignment
) {
2938 void bdrv_add_before_write_notifier(BlockDriverState
*bs
,
2939 NotifierWithReturn
*notifier
)
2941 notifier_with_return_list_add(&bs
->before_write_notifiers
, notifier
);
2944 void bdrv_io_plug(BlockDriverState
*bs
)
2948 QLIST_FOREACH(child
, &bs
->children
, next
) {
2949 bdrv_io_plug(child
->bs
);
2952 if (atomic_fetch_inc(&bs
->io_plugged
) == 0) {
2953 BlockDriver
*drv
= bs
->drv
;
2954 if (drv
&& drv
->bdrv_io_plug
) {
2955 drv
->bdrv_io_plug(bs
);
2960 void bdrv_io_unplug(BlockDriverState
*bs
)
2964 assert(bs
->io_plugged
);
2965 if (atomic_fetch_dec(&bs
->io_plugged
) == 1) {
2966 BlockDriver
*drv
= bs
->drv
;
2967 if (drv
&& drv
->bdrv_io_unplug
) {
2968 drv
->bdrv_io_unplug(bs
);
2972 QLIST_FOREACH(child
, &bs
->children
, next
) {
2973 bdrv_io_unplug(child
->bs
);
2977 void bdrv_register_buf(BlockDriverState
*bs
, void *host
, size_t size
)
2981 if (bs
->drv
&& bs
->drv
->bdrv_register_buf
) {
2982 bs
->drv
->bdrv_register_buf(bs
, host
, size
);
2984 QLIST_FOREACH(child
, &bs
->children
, next
) {
2985 bdrv_register_buf(child
->bs
, host
, size
);
2989 void bdrv_unregister_buf(BlockDriverState
*bs
, void *host
)
2993 if (bs
->drv
&& bs
->drv
->bdrv_unregister_buf
) {
2994 bs
->drv
->bdrv_unregister_buf(bs
, host
);
2996 QLIST_FOREACH(child
, &bs
->children
, next
) {
2997 bdrv_unregister_buf(child
->bs
, host
);
3001 static int coroutine_fn
bdrv_co_copy_range_internal(
3002 BdrvChild
*src
, uint64_t src_offset
, BdrvChild
*dst
,
3003 uint64_t dst_offset
, uint64_t bytes
,
3004 BdrvRequestFlags read_flags
, BdrvRequestFlags write_flags
,
3007 BdrvTrackedRequest req
;
3010 /* TODO We can support BDRV_REQ_NO_FALLBACK here */
3011 assert(!(read_flags
& BDRV_REQ_NO_FALLBACK
));
3012 assert(!(write_flags
& BDRV_REQ_NO_FALLBACK
));
3014 if (!dst
|| !dst
->bs
) {
3017 ret
= bdrv_check_byte_request(dst
->bs
, dst_offset
, bytes
);
3021 if (write_flags
& BDRV_REQ_ZERO_WRITE
) {
3022 return bdrv_co_pwrite_zeroes(dst
, dst_offset
, bytes
, write_flags
);
3025 if (!src
|| !src
->bs
) {
3028 ret
= bdrv_check_byte_request(src
->bs
, src_offset
, bytes
);
3033 if (!src
->bs
->drv
->bdrv_co_copy_range_from
3034 || !dst
->bs
->drv
->bdrv_co_copy_range_to
3035 || src
->bs
->encrypted
|| dst
->bs
->encrypted
) {
3040 bdrv_inc_in_flight(src
->bs
);
3041 tracked_request_begin(&req
, src
->bs
, src_offset
, bytes
,
3044 /* BDRV_REQ_SERIALISING is only for write operation */
3045 assert(!(read_flags
& BDRV_REQ_SERIALISING
));
3046 if (!(read_flags
& BDRV_REQ_NO_SERIALISING
)) {
3047 wait_serialising_requests(&req
);
3050 ret
= src
->bs
->drv
->bdrv_co_copy_range_from(src
->bs
,
3054 read_flags
, write_flags
);
3056 tracked_request_end(&req
);
3057 bdrv_dec_in_flight(src
->bs
);
3059 bdrv_inc_in_flight(dst
->bs
);
3060 tracked_request_begin(&req
, dst
->bs
, dst_offset
, bytes
,
3061 BDRV_TRACKED_WRITE
);
3062 ret
= bdrv_co_write_req_prepare(dst
, dst_offset
, bytes
, &req
,
3065 ret
= dst
->bs
->drv
->bdrv_co_copy_range_to(dst
->bs
,
3069 read_flags
, write_flags
);
3071 bdrv_co_write_req_finish(dst
, dst_offset
, bytes
, &req
, ret
);
3072 tracked_request_end(&req
);
3073 bdrv_dec_in_flight(dst
->bs
);
3079 /* Copy range from @src to @dst.
3081 * See the comment of bdrv_co_copy_range for the parameter and return value
3083 int coroutine_fn
bdrv_co_copy_range_from(BdrvChild
*src
, uint64_t src_offset
,
3084 BdrvChild
*dst
, uint64_t dst_offset
,
3086 BdrvRequestFlags read_flags
,
3087 BdrvRequestFlags write_flags
)
3089 trace_bdrv_co_copy_range_from(src
, src_offset
, dst
, dst_offset
, bytes
,
3090 read_flags
, write_flags
);
3091 return bdrv_co_copy_range_internal(src
, src_offset
, dst
, dst_offset
,
3092 bytes
, read_flags
, write_flags
, true);
3095 /* Copy range from @src to @dst.
3097 * See the comment of bdrv_co_copy_range for the parameter and return value
3099 int coroutine_fn
bdrv_co_copy_range_to(BdrvChild
*src
, uint64_t src_offset
,
3100 BdrvChild
*dst
, uint64_t dst_offset
,
3102 BdrvRequestFlags read_flags
,
3103 BdrvRequestFlags write_flags
)
3105 trace_bdrv_co_copy_range_to(src
, src_offset
, dst
, dst_offset
, bytes
,
3106 read_flags
, write_flags
);
3107 return bdrv_co_copy_range_internal(src
, src_offset
, dst
, dst_offset
,
3108 bytes
, read_flags
, write_flags
, false);
3111 int coroutine_fn
bdrv_co_copy_range(BdrvChild
*src
, uint64_t src_offset
,
3112 BdrvChild
*dst
, uint64_t dst_offset
,
3113 uint64_t bytes
, BdrvRequestFlags read_flags
,
3114 BdrvRequestFlags write_flags
)
3116 return bdrv_co_copy_range_from(src
, src_offset
,
3118 bytes
, read_flags
, write_flags
);
3121 static void bdrv_parent_cb_resize(BlockDriverState
*bs
)
3124 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
3125 if (c
->role
->resize
) {
3132 * Truncate file to 'offset' bytes (needed only for file protocols)
3134 int coroutine_fn
bdrv_co_truncate(BdrvChild
*child
, int64_t offset
,
3135 PreallocMode prealloc
, Error
**errp
)
3137 BlockDriverState
*bs
= child
->bs
;
3138 BlockDriver
*drv
= bs
->drv
;
3139 BdrvTrackedRequest req
;
3140 int64_t old_size
, new_bytes
;
3144 /* if bs->drv == NULL, bs is closed, so there's nothing to do here */
3146 error_setg(errp
, "No medium inserted");
3150 error_setg(errp
, "Image size cannot be negative");
3154 old_size
= bdrv_getlength(bs
);
3156 error_setg_errno(errp
, -old_size
, "Failed to get old image size");
3160 if (offset
> old_size
) {
3161 new_bytes
= offset
- old_size
;
3166 bdrv_inc_in_flight(bs
);
3167 tracked_request_begin(&req
, bs
, offset
- new_bytes
, new_bytes
,
3168 BDRV_TRACKED_TRUNCATE
);
3170 /* If we are growing the image and potentially using preallocation for the
3171 * new area, we need to make sure that no write requests are made to it
3172 * concurrently or they might be overwritten by preallocation. */
3174 mark_request_serialising(&req
, 1);
3176 if (bs
->read_only
) {
3177 error_setg(errp
, "Image is read-only");
3181 ret
= bdrv_co_write_req_prepare(child
, offset
- new_bytes
, new_bytes
, &req
,
3184 error_setg_errno(errp
, -ret
,
3185 "Failed to prepare request for truncation");
3189 if (!drv
->bdrv_co_truncate
) {
3190 if (bs
->file
&& drv
->is_filter
) {
3191 ret
= bdrv_co_truncate(bs
->file
, offset
, prealloc
, errp
);
3194 error_setg(errp
, "Image format driver does not support resize");
3199 ret
= drv
->bdrv_co_truncate(bs
, offset
, prealloc
, errp
);
3203 ret
= refresh_total_sectors(bs
, offset
>> BDRV_SECTOR_BITS
);
3205 error_setg_errno(errp
, -ret
, "Could not refresh total sector count");
3207 offset
= bs
->total_sectors
* BDRV_SECTOR_SIZE
;
3209 /* It's possible that truncation succeeded but refresh_total_sectors
3210 * failed, but the latter doesn't affect how we should finish the request.
3211 * Pass 0 as the last parameter so that dirty bitmaps etc. are handled. */
3212 bdrv_co_write_req_finish(child
, offset
- new_bytes
, new_bytes
, &req
, 0);
3215 tracked_request_end(&req
);
3216 bdrv_dec_in_flight(bs
);
3221 typedef struct TruncateCo
{
3224 PreallocMode prealloc
;
3229 static void coroutine_fn
bdrv_truncate_co_entry(void *opaque
)
3231 TruncateCo
*tco
= opaque
;
3232 tco
->ret
= bdrv_co_truncate(tco
->child
, tco
->offset
, tco
->prealloc
,
3237 int bdrv_truncate(BdrvChild
*child
, int64_t offset
, PreallocMode prealloc
,
3244 .prealloc
= prealloc
,
3249 if (qemu_in_coroutine()) {
3250 /* Fast-path if already in coroutine context */
3251 bdrv_truncate_co_entry(&tco
);
3253 co
= qemu_coroutine_create(bdrv_truncate_co_entry
, &tco
);
3254 bdrv_coroutine_enter(child
->bs
, co
);
3255 BDRV_POLL_WHILE(child
->bs
, tco
.ret
== NOT_DONE
);