2 * QEMU System Emulator block driver
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
24 #include "config-host.h"
25 #include "qemu-common.h"
27 #include "block/block_int.h"
28 #include "block/blockjob.h"
29 #include "qemu/module.h"
30 #include "qapi/qmp/qjson.h"
31 #include "sysemu/block-backend.h"
32 #include "sysemu/sysemu.h"
33 #include "sysemu/qtest.h"
34 #include "qemu/notify.h"
35 #include "block/coroutine.h"
36 #include "block/qapi.h"
37 #include "qmp-commands.h"
38 #include "qemu/timer.h"
39 #include "qapi-event.h"
42 #include <sys/types.h>
44 #include <sys/ioctl.h>
45 #include <sys/queue.h>
55 struct BdrvDirtyBitmap
{
57 QLIST_ENTRY(BdrvDirtyBitmap
) list
;
60 #define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */
62 static BlockAIOCB
*bdrv_aio_readv_em(BlockDriverState
*bs
,
63 int64_t sector_num
, QEMUIOVector
*qiov
, int nb_sectors
,
64 BlockCompletionFunc
*cb
, void *opaque
);
65 static BlockAIOCB
*bdrv_aio_writev_em(BlockDriverState
*bs
,
66 int64_t sector_num
, QEMUIOVector
*qiov
, int nb_sectors
,
67 BlockCompletionFunc
*cb
, void *opaque
);
68 static int coroutine_fn
bdrv_co_readv_em(BlockDriverState
*bs
,
69 int64_t sector_num
, int nb_sectors
,
71 static int coroutine_fn
bdrv_co_writev_em(BlockDriverState
*bs
,
72 int64_t sector_num
, int nb_sectors
,
74 static int coroutine_fn
bdrv_co_do_preadv(BlockDriverState
*bs
,
75 int64_t offset
, unsigned int bytes
, QEMUIOVector
*qiov
,
76 BdrvRequestFlags flags
);
77 static int coroutine_fn
bdrv_co_do_pwritev(BlockDriverState
*bs
,
78 int64_t offset
, unsigned int bytes
, QEMUIOVector
*qiov
,
79 BdrvRequestFlags flags
);
80 static BlockAIOCB
*bdrv_co_aio_rw_vector(BlockDriverState
*bs
,
84 BdrvRequestFlags flags
,
85 BlockCompletionFunc
*cb
,
88 static void coroutine_fn
bdrv_co_do_rw(void *opaque
);
89 static int coroutine_fn
bdrv_co_do_write_zeroes(BlockDriverState
*bs
,
90 int64_t sector_num
, int nb_sectors
, BdrvRequestFlags flags
);
92 static QTAILQ_HEAD(, BlockDriverState
) bdrv_states
=
93 QTAILQ_HEAD_INITIALIZER(bdrv_states
);
95 static QTAILQ_HEAD(, BlockDriverState
) graph_bdrv_states
=
96 QTAILQ_HEAD_INITIALIZER(graph_bdrv_states
);
98 static QLIST_HEAD(, BlockDriver
) bdrv_drivers
=
99 QLIST_HEAD_INITIALIZER(bdrv_drivers
);
101 static void bdrv_set_dirty(BlockDriverState
*bs
, int64_t cur_sector
,
103 static void bdrv_reset_dirty(BlockDriverState
*bs
, int64_t cur_sector
,
105 /* If non-zero, use only whitelisted block drivers */
106 static int use_bdrv_whitelist
;
109 static int is_windows_drive_prefix(const char *filename
)
111 return (((filename
[0] >= 'a' && filename
[0] <= 'z') ||
112 (filename
[0] >= 'A' && filename
[0] <= 'Z')) &&
116 int is_windows_drive(const char *filename
)
118 if (is_windows_drive_prefix(filename
) &&
121 if (strstart(filename
, "\\\\.\\", NULL
) ||
122 strstart(filename
, "//./", NULL
))
128 /* throttling disk I/O limits */
129 void bdrv_set_io_limits(BlockDriverState
*bs
,
134 throttle_config(&bs
->throttle_state
, cfg
);
136 for (i
= 0; i
< 2; i
++) {
137 qemu_co_enter_next(&bs
->throttled_reqs
[i
]);
141 /* this function drain all the throttled IOs */
142 static bool bdrv_start_throttled_reqs(BlockDriverState
*bs
)
144 bool drained
= false;
145 bool enabled
= bs
->io_limits_enabled
;
148 bs
->io_limits_enabled
= false;
150 for (i
= 0; i
< 2; i
++) {
151 while (qemu_co_enter_next(&bs
->throttled_reqs
[i
])) {
156 bs
->io_limits_enabled
= enabled
;
161 void bdrv_io_limits_disable(BlockDriverState
*bs
)
163 bs
->io_limits_enabled
= false;
165 bdrv_start_throttled_reqs(bs
);
167 throttle_destroy(&bs
->throttle_state
);
170 static void bdrv_throttle_read_timer_cb(void *opaque
)
172 BlockDriverState
*bs
= opaque
;
173 qemu_co_enter_next(&bs
->throttled_reqs
[0]);
176 static void bdrv_throttle_write_timer_cb(void *opaque
)
178 BlockDriverState
*bs
= opaque
;
179 qemu_co_enter_next(&bs
->throttled_reqs
[1]);
182 /* should be called before bdrv_set_io_limits if a limit is set */
183 void bdrv_io_limits_enable(BlockDriverState
*bs
)
185 int clock_type
= QEMU_CLOCK_REALTIME
;
187 if (qtest_enabled()) {
188 /* For testing block IO throttling only */
189 clock_type
= QEMU_CLOCK_VIRTUAL
;
191 assert(!bs
->io_limits_enabled
);
192 throttle_init(&bs
->throttle_state
,
193 bdrv_get_aio_context(bs
),
195 bdrv_throttle_read_timer_cb
,
196 bdrv_throttle_write_timer_cb
,
198 bs
->io_limits_enabled
= true;
201 /* This function makes an IO wait if needed
203 * @nb_sectors: the number of sectors of the IO
204 * @is_write: is the IO a write
206 static void bdrv_io_limits_intercept(BlockDriverState
*bs
,
210 /* does this io must wait */
211 bool must_wait
= throttle_schedule_timer(&bs
->throttle_state
, is_write
);
213 /* if must wait or any request of this type throttled queue the IO */
215 !qemu_co_queue_empty(&bs
->throttled_reqs
[is_write
])) {
216 qemu_co_queue_wait(&bs
->throttled_reqs
[is_write
]);
219 /* the IO will be executed, do the accounting */
220 throttle_account(&bs
->throttle_state
, is_write
, bytes
);
223 /* if the next request must wait -> do nothing */
224 if (throttle_schedule_timer(&bs
->throttle_state
, is_write
)) {
228 /* else queue next request for execution */
229 qemu_co_queue_next(&bs
->throttled_reqs
[is_write
]);
232 size_t bdrv_opt_mem_align(BlockDriverState
*bs
)
234 if (!bs
|| !bs
->drv
) {
235 /* 4k should be on the safe side */
239 return bs
->bl
.opt_mem_alignment
;
242 /* check if the path starts with "<protocol>:" */
243 int path_has_protocol(const char *path
)
248 if (is_windows_drive(path
) ||
249 is_windows_drive_prefix(path
)) {
252 p
= path
+ strcspn(path
, ":/\\");
254 p
= path
+ strcspn(path
, ":/");
260 int path_is_absolute(const char *path
)
263 /* specific case for names like: "\\.\d:" */
264 if (is_windows_drive(path
) || is_windows_drive_prefix(path
)) {
267 return (*path
== '/' || *path
== '\\');
269 return (*path
== '/');
273 /* if filename is absolute, just copy it to dest. Otherwise, build a
274 path to it by considering it is relative to base_path. URL are
276 void path_combine(char *dest
, int dest_size
,
277 const char *base_path
,
278 const char *filename
)
285 if (path_is_absolute(filename
)) {
286 pstrcpy(dest
, dest_size
, filename
);
288 p
= strchr(base_path
, ':');
293 p1
= strrchr(base_path
, '/');
297 p2
= strrchr(base_path
, '\\');
309 if (len
> dest_size
- 1)
311 memcpy(dest
, base_path
, len
);
313 pstrcat(dest
, dest_size
, filename
);
317 void bdrv_get_full_backing_filename_from_filename(const char *backed
,
319 char *dest
, size_t sz
,
322 if (backing
[0] == '\0' || path_has_protocol(backing
) ||
323 path_is_absolute(backing
))
325 pstrcpy(dest
, sz
, backing
);
326 } else if (backed
[0] == '\0' || strstart(backed
, "json:", NULL
)) {
327 error_setg(errp
, "Cannot use relative backing file names for '%s'",
330 path_combine(dest
, sz
, backed
, backing
);
334 void bdrv_get_full_backing_filename(BlockDriverState
*bs
, char *dest
, size_t sz
,
337 char *backed
= bs
->exact_filename
[0] ? bs
->exact_filename
: bs
->filename
;
339 bdrv_get_full_backing_filename_from_filename(backed
, bs
->backing_file
,
343 void bdrv_register(BlockDriver
*bdrv
)
345 /* Block drivers without coroutine functions need emulation */
346 if (!bdrv
->bdrv_co_readv
) {
347 bdrv
->bdrv_co_readv
= bdrv_co_readv_em
;
348 bdrv
->bdrv_co_writev
= bdrv_co_writev_em
;
350 /* bdrv_co_readv_em()/brdv_co_writev_em() work in terms of aio, so if
351 * the block driver lacks aio we need to emulate that too.
353 if (!bdrv
->bdrv_aio_readv
) {
354 /* add AIO emulation layer */
355 bdrv
->bdrv_aio_readv
= bdrv_aio_readv_em
;
356 bdrv
->bdrv_aio_writev
= bdrv_aio_writev_em
;
360 QLIST_INSERT_HEAD(&bdrv_drivers
, bdrv
, list
);
363 BlockDriverState
*bdrv_new_root(void)
365 BlockDriverState
*bs
= bdrv_new();
367 QTAILQ_INSERT_TAIL(&bdrv_states
, bs
, device_list
);
371 BlockDriverState
*bdrv_new(void)
373 BlockDriverState
*bs
;
376 bs
= g_new0(BlockDriverState
, 1);
377 QLIST_INIT(&bs
->dirty_bitmaps
);
378 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
379 QLIST_INIT(&bs
->op_blockers
[i
]);
381 bdrv_iostatus_disable(bs
);
382 notifier_list_init(&bs
->close_notifiers
);
383 notifier_with_return_list_init(&bs
->before_write_notifiers
);
384 qemu_co_queue_init(&bs
->throttled_reqs
[0]);
385 qemu_co_queue_init(&bs
->throttled_reqs
[1]);
387 bs
->aio_context
= qemu_get_aio_context();
392 void bdrv_add_close_notifier(BlockDriverState
*bs
, Notifier
*notify
)
394 notifier_list_add(&bs
->close_notifiers
, notify
);
397 BlockDriver
*bdrv_find_format(const char *format_name
)
400 QLIST_FOREACH(drv1
, &bdrv_drivers
, list
) {
401 if (!strcmp(drv1
->format_name
, format_name
)) {
408 static int bdrv_is_whitelisted(BlockDriver
*drv
, bool read_only
)
410 static const char *whitelist_rw
[] = {
411 CONFIG_BDRV_RW_WHITELIST
413 static const char *whitelist_ro
[] = {
414 CONFIG_BDRV_RO_WHITELIST
418 if (!whitelist_rw
[0] && !whitelist_ro
[0]) {
419 return 1; /* no whitelist, anything goes */
422 for (p
= whitelist_rw
; *p
; p
++) {
423 if (!strcmp(drv
->format_name
, *p
)) {
428 for (p
= whitelist_ro
; *p
; p
++) {
429 if (!strcmp(drv
->format_name
, *p
)) {
437 BlockDriver
*bdrv_find_whitelisted_format(const char *format_name
,
440 BlockDriver
*drv
= bdrv_find_format(format_name
);
441 return drv
&& bdrv_is_whitelisted(drv
, read_only
) ? drv
: NULL
;
444 typedef struct CreateCo
{
452 static void coroutine_fn
bdrv_create_co_entry(void *opaque
)
454 Error
*local_err
= NULL
;
457 CreateCo
*cco
= opaque
;
460 ret
= cco
->drv
->bdrv_create(cco
->filename
, cco
->opts
, &local_err
);
462 error_propagate(&cco
->err
, local_err
);
467 int bdrv_create(BlockDriver
*drv
, const char* filename
,
468 QemuOpts
*opts
, Error
**errp
)
475 .filename
= g_strdup(filename
),
481 if (!drv
->bdrv_create
) {
482 error_setg(errp
, "Driver '%s' does not support image creation", drv
->format_name
);
487 if (qemu_in_coroutine()) {
488 /* Fast-path if already in coroutine context */
489 bdrv_create_co_entry(&cco
);
491 co
= qemu_coroutine_create(bdrv_create_co_entry
);
492 qemu_coroutine_enter(co
, &cco
);
493 while (cco
.ret
== NOT_DONE
) {
494 aio_poll(qemu_get_aio_context(), true);
501 error_propagate(errp
, cco
.err
);
503 error_setg_errno(errp
, -ret
, "Could not create image");
508 g_free(cco
.filename
);
512 int bdrv_create_file(const char *filename
, QemuOpts
*opts
, Error
**errp
)
515 Error
*local_err
= NULL
;
518 drv
= bdrv_find_protocol(filename
, true, errp
);
523 ret
= bdrv_create(drv
, filename
, opts
, &local_err
);
525 error_propagate(errp
, local_err
);
530 void bdrv_refresh_limits(BlockDriverState
*bs
, Error
**errp
)
532 BlockDriver
*drv
= bs
->drv
;
533 Error
*local_err
= NULL
;
535 memset(&bs
->bl
, 0, sizeof(bs
->bl
));
541 /* Take some limits from the children as a default */
543 bdrv_refresh_limits(bs
->file
, &local_err
);
545 error_propagate(errp
, local_err
);
548 bs
->bl
.opt_transfer_length
= bs
->file
->bl
.opt_transfer_length
;
549 bs
->bl
.max_transfer_length
= bs
->file
->bl
.max_transfer_length
;
550 bs
->bl
.opt_mem_alignment
= bs
->file
->bl
.opt_mem_alignment
;
552 bs
->bl
.opt_mem_alignment
= 512;
555 if (bs
->backing_hd
) {
556 bdrv_refresh_limits(bs
->backing_hd
, &local_err
);
558 error_propagate(errp
, local_err
);
561 bs
->bl
.opt_transfer_length
=
562 MAX(bs
->bl
.opt_transfer_length
,
563 bs
->backing_hd
->bl
.opt_transfer_length
);
564 bs
->bl
.max_transfer_length
=
565 MIN_NON_ZERO(bs
->bl
.max_transfer_length
,
566 bs
->backing_hd
->bl
.max_transfer_length
);
567 bs
->bl
.opt_mem_alignment
=
568 MAX(bs
->bl
.opt_mem_alignment
,
569 bs
->backing_hd
->bl
.opt_mem_alignment
);
572 /* Then let the driver override it */
573 if (drv
->bdrv_refresh_limits
) {
574 drv
->bdrv_refresh_limits(bs
, errp
);
579 * Try to get @bs's logical and physical block size.
580 * On success, store them in @bsz struct and return 0.
581 * On failure return -errno.
582 * @bs must not be empty.
584 int bdrv_probe_blocksizes(BlockDriverState
*bs
, BlockSizes
*bsz
)
586 BlockDriver
*drv
= bs
->drv
;
588 if (drv
&& drv
->bdrv_probe_blocksizes
) {
589 return drv
->bdrv_probe_blocksizes(bs
, bsz
);
596 * Try to get @bs's geometry (cyls, heads, sectors).
597 * On success, store them in @geo struct and return 0.
598 * On failure return -errno.
599 * @bs must not be empty.
601 int bdrv_probe_geometry(BlockDriverState
*bs
, HDGeometry
*geo
)
603 BlockDriver
*drv
= bs
->drv
;
605 if (drv
&& drv
->bdrv_probe_geometry
) {
606 return drv
->bdrv_probe_geometry(bs
, geo
);
613 * Create a uniquely-named empty temporary file.
614 * Return 0 upon success, otherwise a negative errno value.
616 int get_tmp_filename(char *filename
, int size
)
619 char temp_dir
[MAX_PATH
];
620 /* GetTempFileName requires that its output buffer (4th param)
621 have length MAX_PATH or greater. */
622 assert(size
>= MAX_PATH
);
623 return (GetTempPath(MAX_PATH
, temp_dir
)
624 && GetTempFileName(temp_dir
, "qem", 0, filename
)
625 ? 0 : -GetLastError());
629 tmpdir
= getenv("TMPDIR");
633 if (snprintf(filename
, size
, "%s/vl.XXXXXX", tmpdir
) >= size
) {
636 fd
= mkstemp(filename
);
640 if (close(fd
) != 0) {
649 * Detect host devices. By convention, /dev/cdrom[N] is always
650 * recognized as a host CDROM.
652 static BlockDriver
*find_hdev_driver(const char *filename
)
654 int score_max
= 0, score
;
655 BlockDriver
*drv
= NULL
, *d
;
657 QLIST_FOREACH(d
, &bdrv_drivers
, list
) {
658 if (d
->bdrv_probe_device
) {
659 score
= d
->bdrv_probe_device(filename
);
660 if (score
> score_max
) {
670 BlockDriver
*bdrv_find_protocol(const char *filename
,
671 bool allow_protocol_prefix
,
679 /* TODO Drivers without bdrv_file_open must be specified explicitly */
682 * XXX(hch): we really should not let host device detection
683 * override an explicit protocol specification, but moving this
684 * later breaks access to device names with colons in them.
685 * Thanks to the brain-dead persistent naming schemes on udev-
686 * based Linux systems those actually are quite common.
688 drv1
= find_hdev_driver(filename
);
693 if (!path_has_protocol(filename
) || !allow_protocol_prefix
) {
697 p
= strchr(filename
, ':');
700 if (len
> sizeof(protocol
) - 1)
701 len
= sizeof(protocol
) - 1;
702 memcpy(protocol
, filename
, len
);
703 protocol
[len
] = '\0';
704 QLIST_FOREACH(drv1
, &bdrv_drivers
, list
) {
705 if (drv1
->protocol_name
&&
706 !strcmp(drv1
->protocol_name
, protocol
)) {
711 error_setg(errp
, "Unknown protocol '%s'", protocol
);
716 * Guess image format by probing its contents.
717 * This is not a good idea when your image is raw (CVE-2008-2004), but
718 * we do it anyway for backward compatibility.
720 * @buf contains the image's first @buf_size bytes.
721 * @buf_size is the buffer size in bytes (generally BLOCK_PROBE_BUF_SIZE,
722 * but can be smaller if the image file is smaller)
723 * @filename is its filename.
725 * For all block drivers, call the bdrv_probe() method to get its
727 * Return the first block driver with the highest probing score.
729 BlockDriver
*bdrv_probe_all(const uint8_t *buf
, int buf_size
,
730 const char *filename
)
732 int score_max
= 0, score
;
733 BlockDriver
*drv
= NULL
, *d
;
735 QLIST_FOREACH(d
, &bdrv_drivers
, list
) {
737 score
= d
->bdrv_probe(buf
, buf_size
, filename
);
738 if (score
> score_max
) {
748 static int find_image_format(BlockDriverState
*bs
, const char *filename
,
749 BlockDriver
**pdrv
, Error
**errp
)
752 uint8_t buf
[BLOCK_PROBE_BUF_SIZE
];
755 /* Return the raw BlockDriver * to scsi-generic devices or empty drives */
756 if (bs
->sg
|| !bdrv_is_inserted(bs
) || bdrv_getlength(bs
) == 0) {
761 ret
= bdrv_pread(bs
, 0, buf
, sizeof(buf
));
763 error_setg_errno(errp
, -ret
, "Could not read image for determining its "
769 drv
= bdrv_probe_all(buf
, ret
, filename
);
771 error_setg(errp
, "Could not determine image format: No compatible "
780 * Set the current 'total_sectors' value
781 * Return 0 on success, -errno on error.
783 static int refresh_total_sectors(BlockDriverState
*bs
, int64_t hint
)
785 BlockDriver
*drv
= bs
->drv
;
787 /* Do not attempt drv->bdrv_getlength() on scsi-generic devices */
791 /* query actual device if possible, otherwise just trust the hint */
792 if (drv
->bdrv_getlength
) {
793 int64_t length
= drv
->bdrv_getlength(bs
);
797 hint
= DIV_ROUND_UP(length
, BDRV_SECTOR_SIZE
);
800 bs
->total_sectors
= hint
;
805 * Set open flags for a given discard mode
807 * Return 0 on success, -1 if the discard mode was invalid.
809 int bdrv_parse_discard_flags(const char *mode
, int *flags
)
811 *flags
&= ~BDRV_O_UNMAP
;
813 if (!strcmp(mode
, "off") || !strcmp(mode
, "ignore")) {
815 } else if (!strcmp(mode
, "on") || !strcmp(mode
, "unmap")) {
816 *flags
|= BDRV_O_UNMAP
;
825 * Set open flags for a given cache mode
827 * Return 0 on success, -1 if the cache mode was invalid.
829 int bdrv_parse_cache_flags(const char *mode
, int *flags
)
831 *flags
&= ~BDRV_O_CACHE_MASK
;
833 if (!strcmp(mode
, "off") || !strcmp(mode
, "none")) {
834 *flags
|= BDRV_O_NOCACHE
| BDRV_O_CACHE_WB
;
835 } else if (!strcmp(mode
, "directsync")) {
836 *flags
|= BDRV_O_NOCACHE
;
837 } else if (!strcmp(mode
, "writeback")) {
838 *flags
|= BDRV_O_CACHE_WB
;
839 } else if (!strcmp(mode
, "unsafe")) {
840 *flags
|= BDRV_O_CACHE_WB
;
841 *flags
|= BDRV_O_NO_FLUSH
;
842 } else if (!strcmp(mode
, "writethrough")) {
843 /* this is the default */
852 * The copy-on-read flag is actually a reference count so multiple users may
853 * use the feature without worrying about clobbering its previous state.
854 * Copy-on-read stays enabled until all users have called to disable it.
856 void bdrv_enable_copy_on_read(BlockDriverState
*bs
)
861 void bdrv_disable_copy_on_read(BlockDriverState
*bs
)
863 assert(bs
->copy_on_read
> 0);
868 * Returns the flags that a temporary snapshot should get, based on the
869 * originally requested flags (the originally requested image will have flags
870 * like a backing file)
872 static int bdrv_temp_snapshot_flags(int flags
)
874 return (flags
& ~BDRV_O_SNAPSHOT
) | BDRV_O_TEMPORARY
;
878 * Returns the flags that bs->file should get, based on the given flags for
881 static int bdrv_inherited_flags(int flags
)
883 /* Enable protocol handling, disable format probing for bs->file */
884 flags
|= BDRV_O_PROTOCOL
;
886 /* Our block drivers take care to send flushes and respect unmap policy,
887 * so we can enable both unconditionally on lower layers. */
888 flags
|= BDRV_O_CACHE_WB
| BDRV_O_UNMAP
;
890 /* Clear flags that only apply to the top layer */
891 flags
&= ~(BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
| BDRV_O_COPY_ON_READ
);
897 * Returns the flags that bs->backing_hd should get, based on the given flags
900 static int bdrv_backing_flags(int flags
)
902 /* backing files always opened read-only */
903 flags
&= ~(BDRV_O_RDWR
| BDRV_O_COPY_ON_READ
);
905 /* snapshot=on is handled on the top layer */
906 flags
&= ~(BDRV_O_SNAPSHOT
| BDRV_O_TEMPORARY
);
911 static int bdrv_open_flags(BlockDriverState
*bs
, int flags
)
913 int open_flags
= flags
| BDRV_O_CACHE_WB
;
916 * Clear flags that are internal to the block layer before opening the
919 open_flags
&= ~(BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
| BDRV_O_PROTOCOL
);
922 * Snapshots should be writable.
924 if (flags
& BDRV_O_TEMPORARY
) {
925 open_flags
|= BDRV_O_RDWR
;
931 static void bdrv_assign_node_name(BlockDriverState
*bs
,
932 const char *node_name
,
939 /* Check for empty string or invalid characters */
940 if (!id_wellformed(node_name
)) {
941 error_setg(errp
, "Invalid node name");
945 /* takes care of avoiding namespaces collisions */
946 if (blk_by_name(node_name
)) {
947 error_setg(errp
, "node-name=%s is conflicting with a device id",
952 /* takes care of avoiding duplicates node names */
953 if (bdrv_find_node(node_name
)) {
954 error_setg(errp
, "Duplicate node name");
958 /* copy node name into the bs and insert it into the graph list */
959 pstrcpy(bs
->node_name
, sizeof(bs
->node_name
), node_name
);
960 QTAILQ_INSERT_TAIL(&graph_bdrv_states
, bs
, node_list
);
964 * Common part for opening disk images and files
966 * Removes all processed options from *options.
968 static int bdrv_open_common(BlockDriverState
*bs
, BlockDriverState
*file
,
969 QDict
*options
, int flags
, BlockDriver
*drv
, Error
**errp
)
972 const char *filename
;
973 const char *node_name
= NULL
;
974 Error
*local_err
= NULL
;
977 assert(bs
->file
== NULL
);
978 assert(options
!= NULL
&& bs
->options
!= options
);
981 filename
= file
->filename
;
983 filename
= qdict_get_try_str(options
, "filename");
986 if (drv
->bdrv_needs_filename
&& !filename
) {
987 error_setg(errp
, "The '%s' block driver requires a file name",
992 trace_bdrv_open_common(bs
, filename
?: "", flags
, drv
->format_name
);
994 node_name
= qdict_get_try_str(options
, "node-name");
995 bdrv_assign_node_name(bs
, node_name
, &local_err
);
997 error_propagate(errp
, local_err
);
1000 qdict_del(options
, "node-name");
1002 /* bdrv_open() with directly using a protocol as drv. This layer is already
1003 * opened, so assign it to bs (while file becomes a closed BlockDriverState)
1004 * and return immediately. */
1005 if (file
!= NULL
&& drv
->bdrv_file_open
) {
1006 bdrv_swap(file
, bs
);
1010 bs
->open_flags
= flags
;
1011 bs
->guest_block_size
= 512;
1012 bs
->request_alignment
= 512;
1013 bs
->zero_beyond_eof
= true;
1014 open_flags
= bdrv_open_flags(bs
, flags
);
1015 bs
->read_only
= !(open_flags
& BDRV_O_RDWR
);
1017 if (use_bdrv_whitelist
&& !bdrv_is_whitelisted(drv
, bs
->read_only
)) {
1019 !bs
->read_only
&& bdrv_is_whitelisted(drv
, true)
1020 ? "Driver '%s' can only be used for read-only devices"
1021 : "Driver '%s' is not whitelisted",
1026 assert(bs
->copy_on_read
== 0); /* bdrv_new() and bdrv_close() make it so */
1027 if (flags
& BDRV_O_COPY_ON_READ
) {
1028 if (!bs
->read_only
) {
1029 bdrv_enable_copy_on_read(bs
);
1031 error_setg(errp
, "Can't use copy-on-read on read-only device");
1036 if (filename
!= NULL
) {
1037 pstrcpy(bs
->filename
, sizeof(bs
->filename
), filename
);
1039 bs
->filename
[0] = '\0';
1041 pstrcpy(bs
->exact_filename
, sizeof(bs
->exact_filename
), bs
->filename
);
1044 bs
->opaque
= g_malloc0(drv
->instance_size
);
1046 bs
->enable_write_cache
= !!(flags
& BDRV_O_CACHE_WB
);
1048 /* Open the image, either directly or using a protocol */
1049 if (drv
->bdrv_file_open
) {
1050 assert(file
== NULL
);
1051 assert(!drv
->bdrv_needs_filename
|| filename
!= NULL
);
1052 ret
= drv
->bdrv_file_open(bs
, options
, open_flags
, &local_err
);
1055 error_setg(errp
, "Can't use '%s' as a block driver for the "
1056 "protocol level", drv
->format_name
);
1061 ret
= drv
->bdrv_open(bs
, options
, open_flags
, &local_err
);
1066 error_propagate(errp
, local_err
);
1067 } else if (bs
->filename
[0]) {
1068 error_setg_errno(errp
, -ret
, "Could not open '%s'", bs
->filename
);
1070 error_setg_errno(errp
, -ret
, "Could not open image");
1075 if (bs
->encrypted
) {
1076 error_report("Encrypted images are deprecated");
1077 error_printf("Support for them will be removed in a future release.\n"
1078 "You can use 'qemu-img convert' to convert your image"
1079 " to an unencrypted one.\n");
1082 ret
= refresh_total_sectors(bs
, bs
->total_sectors
);
1084 error_setg_errno(errp
, -ret
, "Could not refresh total sector count");
1088 bdrv_refresh_limits(bs
, &local_err
);
1090 error_propagate(errp
, local_err
);
1095 assert(bdrv_opt_mem_align(bs
) != 0);
1096 assert((bs
->request_alignment
!= 0) || bs
->sg
);
1107 static QDict
*parse_json_filename(const char *filename
, Error
**errp
)
1109 QObject
*options_obj
;
1113 ret
= strstart(filename
, "json:", &filename
);
1116 options_obj
= qobject_from_json(filename
);
1118 error_setg(errp
, "Could not parse the JSON options");
1122 if (qobject_type(options_obj
) != QTYPE_QDICT
) {
1123 qobject_decref(options_obj
);
1124 error_setg(errp
, "Invalid JSON object given");
1128 options
= qobject_to_qdict(options_obj
);
1129 qdict_flatten(options
);
1135 * Fills in default options for opening images and converts the legacy
1136 * filename/flags pair to option QDict entries.
1138 static int bdrv_fill_options(QDict
**options
, const char **pfilename
, int flags
,
1139 BlockDriver
*drv
, Error
**errp
)
1141 const char *filename
= *pfilename
;
1142 const char *drvname
;
1143 bool protocol
= flags
& BDRV_O_PROTOCOL
;
1144 bool parse_filename
= false;
1145 Error
*local_err
= NULL
;
1147 /* Parse json: pseudo-protocol */
1148 if (filename
&& g_str_has_prefix(filename
, "json:")) {
1149 QDict
*json_options
= parse_json_filename(filename
, &local_err
);
1151 error_propagate(errp
, local_err
);
1155 /* Options given in the filename have lower priority than options
1156 * specified directly */
1157 qdict_join(*options
, json_options
, false);
1158 QDECREF(json_options
);
1159 *pfilename
= filename
= NULL
;
1162 /* Fetch the file name from the options QDict if necessary */
1163 if (protocol
&& filename
) {
1164 if (!qdict_haskey(*options
, "filename")) {
1165 qdict_put(*options
, "filename", qstring_from_str(filename
));
1166 parse_filename
= true;
1168 error_setg(errp
, "Can't specify 'file' and 'filename' options at "
1174 /* Find the right block driver */
1175 filename
= qdict_get_try_str(*options
, "filename");
1176 drvname
= qdict_get_try_str(*options
, "driver");
1180 error_setg(errp
, "Driver specified twice");
1183 drvname
= drv
->format_name
;
1184 qdict_put(*options
, "driver", qstring_from_str(drvname
));
1186 if (!drvname
&& protocol
) {
1188 drv
= bdrv_find_protocol(filename
, parse_filename
, errp
);
1193 drvname
= drv
->format_name
;
1194 qdict_put(*options
, "driver", qstring_from_str(drvname
));
1196 error_setg(errp
, "Must specify either driver or file");
1199 } else if (drvname
) {
1200 drv
= bdrv_find_format(drvname
);
1202 error_setg(errp
, "Unknown driver '%s'", drvname
);
1208 assert(drv
|| !protocol
);
1210 /* Driver-specific filename parsing */
1211 if (drv
&& drv
->bdrv_parse_filename
&& parse_filename
) {
1212 drv
->bdrv_parse_filename(filename
, *options
, &local_err
);
1214 error_propagate(errp
, local_err
);
1218 if (!drv
->bdrv_needs_filename
) {
1219 qdict_del(*options
, "filename");
1226 void bdrv_set_backing_hd(BlockDriverState
*bs
, BlockDriverState
*backing_hd
)
1229 if (bs
->backing_hd
) {
1230 assert(bs
->backing_blocker
);
1231 bdrv_op_unblock_all(bs
->backing_hd
, bs
->backing_blocker
);
1232 } else if (backing_hd
) {
1233 error_setg(&bs
->backing_blocker
,
1234 "device is used as backing hd of '%s'",
1235 bdrv_get_device_name(bs
));
1238 bs
->backing_hd
= backing_hd
;
1240 error_free(bs
->backing_blocker
);
1241 bs
->backing_blocker
= NULL
;
1244 bs
->open_flags
&= ~BDRV_O_NO_BACKING
;
1245 pstrcpy(bs
->backing_file
, sizeof(bs
->backing_file
), backing_hd
->filename
);
1246 pstrcpy(bs
->backing_format
, sizeof(bs
->backing_format
),
1247 backing_hd
->drv
? backing_hd
->drv
->format_name
: "");
1249 bdrv_op_block_all(bs
->backing_hd
, bs
->backing_blocker
);
1250 /* Otherwise we won't be able to commit due to check in bdrv_commit */
1251 bdrv_op_unblock(bs
->backing_hd
, BLOCK_OP_TYPE_COMMIT_TARGET
,
1252 bs
->backing_blocker
);
1254 bdrv_refresh_limits(bs
, NULL
);
1258 * Opens the backing file for a BlockDriverState if not yet open
1260 * options is a QDict of options to pass to the block drivers, or NULL for an
1261 * empty set of options. The reference to the QDict is transferred to this
1262 * function (even on failure), so if the caller intends to reuse the dictionary,
1263 * it needs to use QINCREF() before calling bdrv_file_open.
1265 int bdrv_open_backing_file(BlockDriverState
*bs
, QDict
*options
, Error
**errp
)
1267 char *backing_filename
= g_malloc0(PATH_MAX
);
1269 BlockDriverState
*backing_hd
;
1270 Error
*local_err
= NULL
;
1272 if (bs
->backing_hd
!= NULL
) {
1277 /* NULL means an empty set of options */
1278 if (options
== NULL
) {
1279 options
= qdict_new();
1282 bs
->open_flags
&= ~BDRV_O_NO_BACKING
;
1283 if (qdict_haskey(options
, "file.filename")) {
1284 backing_filename
[0] = '\0';
1285 } else if (bs
->backing_file
[0] == '\0' && qdict_size(options
) == 0) {
1289 bdrv_get_full_backing_filename(bs
, backing_filename
, PATH_MAX
,
1293 error_propagate(errp
, local_err
);
1299 if (!bs
->drv
|| !bs
->drv
->supports_backing
) {
1301 error_setg(errp
, "Driver doesn't support backing files");
1306 backing_hd
= bdrv_new();
1308 if (bs
->backing_format
[0] != '\0' && !qdict_haskey(options
, "driver")) {
1309 qdict_put(options
, "driver", qstring_from_str(bs
->backing_format
));
1312 assert(bs
->backing_hd
== NULL
);
1313 ret
= bdrv_open(&backing_hd
,
1314 *backing_filename
? backing_filename
: NULL
, NULL
, options
,
1315 bdrv_backing_flags(bs
->open_flags
), NULL
, &local_err
);
1317 bdrv_unref(backing_hd
);
1319 bs
->open_flags
|= BDRV_O_NO_BACKING
;
1320 error_setg(errp
, "Could not open backing file: %s",
1321 error_get_pretty(local_err
));
1322 error_free(local_err
);
1325 bdrv_set_backing_hd(bs
, backing_hd
);
1328 g_free(backing_filename
);
1333 * Opens a disk image whose options are given as BlockdevRef in another block
1336 * If allow_none is true, no image will be opened if filename is false and no
1337 * BlockdevRef is given. *pbs will remain unchanged and 0 will be returned.
1339 * bdrev_key specifies the key for the image's BlockdevRef in the options QDict.
1340 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
1341 * itself, all options starting with "${bdref_key}." are considered part of the
1344 * The BlockdevRef will be removed from the options QDict.
1346 * To conform with the behavior of bdrv_open(), *pbs has to be NULL.
1348 int bdrv_open_image(BlockDriverState
**pbs
, const char *filename
,
1349 QDict
*options
, const char *bdref_key
, int flags
,
1350 bool allow_none
, Error
**errp
)
1352 QDict
*image_options
;
1354 char *bdref_key_dot
;
1355 const char *reference
;
1358 assert(*pbs
== NULL
);
1360 bdref_key_dot
= g_strdup_printf("%s.", bdref_key
);
1361 qdict_extract_subqdict(options
, &image_options
, bdref_key_dot
);
1362 g_free(bdref_key_dot
);
1364 reference
= qdict_get_try_str(options
, bdref_key
);
1365 if (!filename
&& !reference
&& !qdict_size(image_options
)) {
1369 error_setg(errp
, "A block device must be specified for \"%s\"",
1373 QDECREF(image_options
);
1377 ret
= bdrv_open(pbs
, filename
, reference
, image_options
, flags
, NULL
, errp
);
1380 qdict_del(options
, bdref_key
);
1384 int bdrv_append_temp_snapshot(BlockDriverState
*bs
, int flags
, Error
**errp
)
1386 /* TODO: extra byte is a hack to ensure MAX_PATH space on Windows. */
1387 char *tmp_filename
= g_malloc0(PATH_MAX
+ 1);
1389 QemuOpts
*opts
= NULL
;
1390 QDict
*snapshot_options
;
1391 BlockDriverState
*bs_snapshot
;
1395 /* if snapshot, we create a temporary backing file and open it
1396 instead of opening 'filename' directly */
1398 /* Get the required size from the image */
1399 total_size
= bdrv_getlength(bs
);
1400 if (total_size
< 0) {
1402 error_setg_errno(errp
, -total_size
, "Could not get image size");
1406 /* Create the temporary image */
1407 ret
= get_tmp_filename(tmp_filename
, PATH_MAX
+ 1);
1409 error_setg_errno(errp
, -ret
, "Could not get temporary filename");
1413 opts
= qemu_opts_create(bdrv_qcow2
.create_opts
, NULL
, 0,
1415 qemu_opt_set_number(opts
, BLOCK_OPT_SIZE
, total_size
, &error_abort
);
1416 ret
= bdrv_create(&bdrv_qcow2
, tmp_filename
, opts
, &local_err
);
1417 qemu_opts_del(opts
);
1419 error_setg_errno(errp
, -ret
, "Could not create temporary overlay "
1420 "'%s': %s", tmp_filename
,
1421 error_get_pretty(local_err
));
1422 error_free(local_err
);
1426 /* Prepare a new options QDict for the temporary file */
1427 snapshot_options
= qdict_new();
1428 qdict_put(snapshot_options
, "file.driver",
1429 qstring_from_str("file"));
1430 qdict_put(snapshot_options
, "file.filename",
1431 qstring_from_str(tmp_filename
));
1433 bs_snapshot
= bdrv_new();
1435 ret
= bdrv_open(&bs_snapshot
, NULL
, NULL
, snapshot_options
,
1436 flags
, &bdrv_qcow2
, &local_err
);
1438 error_propagate(errp
, local_err
);
1442 bdrv_append(bs_snapshot
, bs
);
1445 g_free(tmp_filename
);
1450 * Opens a disk image (raw, qcow2, vmdk, ...)
1452 * options is a QDict of options to pass to the block drivers, or NULL for an
1453 * empty set of options. The reference to the QDict belongs to the block layer
1454 * after the call (even on failure), so if the caller intends to reuse the
1455 * dictionary, it needs to use QINCREF() before calling bdrv_open.
1457 * If *pbs is NULL, a new BDS will be created with a pointer to it stored there.
1458 * If it is not NULL, the referenced BDS will be reused.
1460 * The reference parameter may be used to specify an existing block device which
1461 * should be opened. If specified, neither options nor a filename may be given,
1462 * nor can an existing BDS be reused (that is, *pbs has to be NULL).
1464 int bdrv_open(BlockDriverState
**pbs
, const char *filename
,
1465 const char *reference
, QDict
*options
, int flags
,
1466 BlockDriver
*drv
, Error
**errp
)
1469 BlockDriverState
*file
= NULL
, *bs
;
1470 const char *drvname
;
1471 Error
*local_err
= NULL
;
1472 int snapshot_flags
= 0;
1477 bool options_non_empty
= options
? qdict_size(options
) : false;
1481 error_setg(errp
, "Cannot reuse an existing BDS when referencing "
1482 "another block device");
1486 if (filename
|| options_non_empty
) {
1487 error_setg(errp
, "Cannot reference an existing block device with "
1488 "additional options or a new filename");
1492 bs
= bdrv_lookup_bs(reference
, reference
, errp
);
1507 /* NULL means an empty set of options */
1508 if (options
== NULL
) {
1509 options
= qdict_new();
1512 ret
= bdrv_fill_options(&options
, &filename
, flags
, drv
, &local_err
);
1517 /* Find the right image format driver */
1519 drvname
= qdict_get_try_str(options
, "driver");
1521 drv
= bdrv_find_format(drvname
);
1522 qdict_del(options
, "driver");
1524 error_setg(errp
, "Unknown driver: '%s'", drvname
);
1530 assert(drvname
|| !(flags
& BDRV_O_PROTOCOL
));
1531 if (drv
&& !drv
->bdrv_file_open
) {
1532 /* If the user explicitly wants a format driver here, we'll need to add
1533 * another layer for the protocol in bs->file */
1534 flags
&= ~BDRV_O_PROTOCOL
;
1537 bs
->options
= options
;
1538 options
= qdict_clone_shallow(options
);
1540 /* Open image file without format layer */
1541 if ((flags
& BDRV_O_PROTOCOL
) == 0) {
1542 if (flags
& BDRV_O_RDWR
) {
1543 flags
|= BDRV_O_ALLOW_RDWR
;
1545 if (flags
& BDRV_O_SNAPSHOT
) {
1546 snapshot_flags
= bdrv_temp_snapshot_flags(flags
);
1547 flags
= bdrv_backing_flags(flags
);
1550 assert(file
== NULL
);
1551 ret
= bdrv_open_image(&file
, filename
, options
, "file",
1552 bdrv_inherited_flags(flags
),
1559 /* Image format probing */
1562 ret
= find_image_format(file
, filename
, &drv
, &local_err
);
1567 error_setg(errp
, "Must specify either driver or file");
1572 /* Open the image */
1573 ret
= bdrv_open_common(bs
, file
, options
, flags
, drv
, &local_err
);
1578 if (file
&& (bs
->file
!= file
)) {
1583 /* If there is a backing file, use it */
1584 if ((flags
& BDRV_O_NO_BACKING
) == 0) {
1585 QDict
*backing_options
;
1587 qdict_extract_subqdict(options
, &backing_options
, "backing.");
1588 ret
= bdrv_open_backing_file(bs
, backing_options
, &local_err
);
1590 goto close_and_fail
;
1594 bdrv_refresh_filename(bs
);
1596 /* For snapshot=on, create a temporary qcow2 overlay. bs points to the
1597 * temporary snapshot afterwards. */
1598 if (snapshot_flags
) {
1599 ret
= bdrv_append_temp_snapshot(bs
, snapshot_flags
, &local_err
);
1601 goto close_and_fail
;
1605 /* Check if any unknown options were used */
1606 if (options
&& (qdict_size(options
) != 0)) {
1607 const QDictEntry
*entry
= qdict_first(options
);
1608 if (flags
& BDRV_O_PROTOCOL
) {
1609 error_setg(errp
, "Block protocol '%s' doesn't support the option "
1610 "'%s'", drv
->format_name
, entry
->key
);
1612 error_setg(errp
, "Block format '%s' used by device '%s' doesn't "
1613 "support the option '%s'", drv
->format_name
,
1614 bdrv_get_device_name(bs
), entry
->key
);
1618 goto close_and_fail
;
1621 if (!bdrv_key_required(bs
)) {
1623 blk_dev_change_media_cb(bs
->blk
, true);
1625 } else if (!runstate_check(RUN_STATE_PRELAUNCH
)
1626 && !runstate_check(RUN_STATE_INMIGRATE
)
1627 && !runstate_check(RUN_STATE_PAUSED
)) { /* HACK */
1629 "Guest must be stopped for opening of encrypted image");
1631 goto close_and_fail
;
1642 QDECREF(bs
->options
);
1646 /* If *pbs is NULL, a new BDS has been created in this function and
1647 needs to be freed now. Otherwise, it does not need to be closed,
1648 since it has not really been opened yet. */
1652 error_propagate(errp
, local_err
);
1657 /* See fail path, but now the BDS has to be always closed */
1665 error_propagate(errp
, local_err
);
1670 typedef struct BlockReopenQueueEntry
{
1672 BDRVReopenState state
;
1673 QSIMPLEQ_ENTRY(BlockReopenQueueEntry
) entry
;
1674 } BlockReopenQueueEntry
;
1677 * Adds a BlockDriverState to a simple queue for an atomic, transactional
1678 * reopen of multiple devices.
1680 * bs_queue can either be an existing BlockReopenQueue that has had QSIMPLE_INIT
1681 * already performed, or alternatively may be NULL a new BlockReopenQueue will
1682 * be created and initialized. This newly created BlockReopenQueue should be
1683 * passed back in for subsequent calls that are intended to be of the same
1686 * bs is the BlockDriverState to add to the reopen queue.
1688 * flags contains the open flags for the associated bs
1690 * returns a pointer to bs_queue, which is either the newly allocated
1691 * bs_queue, or the existing bs_queue being used.
1694 BlockReopenQueue
*bdrv_reopen_queue(BlockReopenQueue
*bs_queue
,
1695 BlockDriverState
*bs
, int flags
)
1699 BlockReopenQueueEntry
*bs_entry
;
1700 if (bs_queue
== NULL
) {
1701 bs_queue
= g_new0(BlockReopenQueue
, 1);
1702 QSIMPLEQ_INIT(bs_queue
);
1705 /* bdrv_open() masks this flag out */
1706 flags
&= ~BDRV_O_PROTOCOL
;
1709 bdrv_reopen_queue(bs_queue
, bs
->file
, bdrv_inherited_flags(flags
));
1712 bs_entry
= g_new0(BlockReopenQueueEntry
, 1);
1713 QSIMPLEQ_INSERT_TAIL(bs_queue
, bs_entry
, entry
);
1715 bs_entry
->state
.bs
= bs
;
1716 bs_entry
->state
.flags
= flags
;
1722 * Reopen multiple BlockDriverStates atomically & transactionally.
1724 * The queue passed in (bs_queue) must have been built up previous
1725 * via bdrv_reopen_queue().
1727 * Reopens all BDS specified in the queue, with the appropriate
1728 * flags. All devices are prepared for reopen, and failure of any
1729 * device will cause all device changes to be abandonded, and intermediate
1732 * If all devices prepare successfully, then the changes are committed
1736 int bdrv_reopen_multiple(BlockReopenQueue
*bs_queue
, Error
**errp
)
1739 BlockReopenQueueEntry
*bs_entry
, *next
;
1740 Error
*local_err
= NULL
;
1742 assert(bs_queue
!= NULL
);
1746 QSIMPLEQ_FOREACH(bs_entry
, bs_queue
, entry
) {
1747 if (bdrv_reopen_prepare(&bs_entry
->state
, bs_queue
, &local_err
)) {
1748 error_propagate(errp
, local_err
);
1751 bs_entry
->prepared
= true;
1754 /* If we reach this point, we have success and just need to apply the
1757 QSIMPLEQ_FOREACH(bs_entry
, bs_queue
, entry
) {
1758 bdrv_reopen_commit(&bs_entry
->state
);
1764 QSIMPLEQ_FOREACH_SAFE(bs_entry
, bs_queue
, entry
, next
) {
1765 if (ret
&& bs_entry
->prepared
) {
1766 bdrv_reopen_abort(&bs_entry
->state
);
1775 /* Reopen a single BlockDriverState with the specified flags. */
1776 int bdrv_reopen(BlockDriverState
*bs
, int bdrv_flags
, Error
**errp
)
1779 Error
*local_err
= NULL
;
1780 BlockReopenQueue
*queue
= bdrv_reopen_queue(NULL
, bs
, bdrv_flags
);
1782 ret
= bdrv_reopen_multiple(queue
, &local_err
);
1783 if (local_err
!= NULL
) {
1784 error_propagate(errp
, local_err
);
1791 * Prepares a BlockDriverState for reopen. All changes are staged in the
1792 * 'opaque' field of the BDRVReopenState, which is used and allocated by
1793 * the block driver layer .bdrv_reopen_prepare()
1795 * bs is the BlockDriverState to reopen
1796 * flags are the new open flags
1797 * queue is the reopen queue
1799 * Returns 0 on success, non-zero on error. On error errp will be set
1802 * On failure, bdrv_reopen_abort() will be called to clean up any data.
1803 * It is the responsibility of the caller to then call the abort() or
1804 * commit() for any other BDS that have been left in a prepare() state
1807 int bdrv_reopen_prepare(BDRVReopenState
*reopen_state
, BlockReopenQueue
*queue
,
1811 Error
*local_err
= NULL
;
1814 assert(reopen_state
!= NULL
);
1815 assert(reopen_state
->bs
->drv
!= NULL
);
1816 drv
= reopen_state
->bs
->drv
;
1818 /* if we are to stay read-only, do not allow permission change
1820 if (!(reopen_state
->bs
->open_flags
& BDRV_O_ALLOW_RDWR
) &&
1821 reopen_state
->flags
& BDRV_O_RDWR
) {
1822 error_set(errp
, QERR_DEVICE_IS_READ_ONLY
,
1823 bdrv_get_device_name(reopen_state
->bs
));
1828 ret
= bdrv_flush(reopen_state
->bs
);
1830 error_set(errp
, ERROR_CLASS_GENERIC_ERROR
, "Error (%s) flushing drive",
1835 if (drv
->bdrv_reopen_prepare
) {
1836 ret
= drv
->bdrv_reopen_prepare(reopen_state
, queue
, &local_err
);
1838 if (local_err
!= NULL
) {
1839 error_propagate(errp
, local_err
);
1841 error_setg(errp
, "failed while preparing to reopen image '%s'",
1842 reopen_state
->bs
->filename
);
1847 /* It is currently mandatory to have a bdrv_reopen_prepare()
1848 * handler for each supported drv. */
1849 error_set(errp
, QERR_BLOCK_FORMAT_FEATURE_NOT_SUPPORTED
,
1850 drv
->format_name
, bdrv_get_device_name(reopen_state
->bs
),
1851 "reopening of file");
1863 * Takes the staged changes for the reopen from bdrv_reopen_prepare(), and
1864 * makes them final by swapping the staging BlockDriverState contents into
1865 * the active BlockDriverState contents.
1867 void bdrv_reopen_commit(BDRVReopenState
*reopen_state
)
1871 assert(reopen_state
!= NULL
);
1872 drv
= reopen_state
->bs
->drv
;
1873 assert(drv
!= NULL
);
1875 /* If there are any driver level actions to take */
1876 if (drv
->bdrv_reopen_commit
) {
1877 drv
->bdrv_reopen_commit(reopen_state
);
1880 /* set BDS specific flags now */
1881 reopen_state
->bs
->open_flags
= reopen_state
->flags
;
1882 reopen_state
->bs
->enable_write_cache
= !!(reopen_state
->flags
&
1884 reopen_state
->bs
->read_only
= !(reopen_state
->flags
& BDRV_O_RDWR
);
1886 bdrv_refresh_limits(reopen_state
->bs
, NULL
);
1890 * Abort the reopen, and delete and free the staged changes in
1893 void bdrv_reopen_abort(BDRVReopenState
*reopen_state
)
1897 assert(reopen_state
!= NULL
);
1898 drv
= reopen_state
->bs
->drv
;
1899 assert(drv
!= NULL
);
1901 if (drv
->bdrv_reopen_abort
) {
1902 drv
->bdrv_reopen_abort(reopen_state
);
1907 void bdrv_close(BlockDriverState
*bs
)
1909 BdrvAioNotifier
*ban
, *ban_next
;
1912 block_job_cancel_sync(bs
->job
);
1914 bdrv_drain_all(); /* complete I/O */
1916 bdrv_drain_all(); /* in case flush left pending I/O */
1917 notifier_list_notify(&bs
->close_notifiers
, bs
);
1920 if (bs
->backing_hd
) {
1921 BlockDriverState
*backing_hd
= bs
->backing_hd
;
1922 bdrv_set_backing_hd(bs
, NULL
);
1923 bdrv_unref(backing_hd
);
1925 bs
->drv
->bdrv_close(bs
);
1929 bs
->copy_on_read
= 0;
1930 bs
->backing_file
[0] = '\0';
1931 bs
->backing_format
[0] = '\0';
1932 bs
->total_sectors
= 0;
1936 bs
->zero_beyond_eof
= false;
1937 QDECREF(bs
->options
);
1939 QDECREF(bs
->full_open_options
);
1940 bs
->full_open_options
= NULL
;
1942 if (bs
->file
!= NULL
) {
1943 bdrv_unref(bs
->file
);
1949 blk_dev_change_media_cb(bs
->blk
, false);
1952 /*throttling disk I/O limits*/
1953 if (bs
->io_limits_enabled
) {
1954 bdrv_io_limits_disable(bs
);
1957 QLIST_FOREACH_SAFE(ban
, &bs
->aio_notifiers
, list
, ban_next
) {
1960 QLIST_INIT(&bs
->aio_notifiers
);
1963 void bdrv_close_all(void)
1965 BlockDriverState
*bs
;
1967 QTAILQ_FOREACH(bs
, &bdrv_states
, device_list
) {
1968 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
1970 aio_context_acquire(aio_context
);
1972 aio_context_release(aio_context
);
1976 /* Check if any requests are in-flight (including throttled requests) */
1977 static bool bdrv_requests_pending(BlockDriverState
*bs
)
1979 if (!QLIST_EMPTY(&bs
->tracked_requests
)) {
1982 if (!qemu_co_queue_empty(&bs
->throttled_reqs
[0])) {
1985 if (!qemu_co_queue_empty(&bs
->throttled_reqs
[1])) {
1988 if (bs
->file
&& bdrv_requests_pending(bs
->file
)) {
1991 if (bs
->backing_hd
&& bdrv_requests_pending(bs
->backing_hd
)) {
1997 static bool bdrv_drain_one(BlockDriverState
*bs
)
2001 bdrv_flush_io_queue(bs
);
2002 bdrv_start_throttled_reqs(bs
);
2003 bs_busy
= bdrv_requests_pending(bs
);
2004 bs_busy
|= aio_poll(bdrv_get_aio_context(bs
), bs_busy
);
2009 * Wait for pending requests to complete on a single BlockDriverState subtree
2011 * See the warning in bdrv_drain_all(). This function can only be called if
2012 * you are sure nothing can generate I/O because you have op blockers
2015 * Note that unlike bdrv_drain_all(), the caller must hold the BlockDriverState
2018 void bdrv_drain(BlockDriverState
*bs
)
2020 while (bdrv_drain_one(bs
)) {
2021 /* Keep iterating */
2026 * Wait for pending requests to complete across all BlockDriverStates
2028 * This function does not flush data to disk, use bdrv_flush_all() for that
2029 * after calling this function.
2031 * Note that completion of an asynchronous I/O operation can trigger any
2032 * number of other I/O operations on other devices---for example a coroutine
2033 * can be arbitrarily complex and a constant flow of I/O can come until the
2034 * coroutine is complete. Because of this, it is not possible to have a
2035 * function to drain a single device's I/O queue.
2037 void bdrv_drain_all(void)
2039 /* Always run first iteration so any pending completion BHs run */
2041 BlockDriverState
*bs
;
2046 QTAILQ_FOREACH(bs
, &bdrv_states
, device_list
) {
2047 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
2049 aio_context_acquire(aio_context
);
2050 busy
|= bdrv_drain_one(bs
);
2051 aio_context_release(aio_context
);
2056 /* make a BlockDriverState anonymous by removing from bdrv_state and
2057 * graph_bdrv_state list.
2058 Also, NULL terminate the device_name to prevent double remove */
2059 void bdrv_make_anon(BlockDriverState
*bs
)
2062 * Take care to remove bs from bdrv_states only when it's actually
2063 * in it. Note that bs->device_list.tqe_prev is initially null,
2064 * and gets set to non-null by QTAILQ_INSERT_TAIL(). Establish
2065 * the useful invariant "bs in bdrv_states iff bs->tqe_prev" by
2066 * resetting it to null on remove.
2068 if (bs
->device_list
.tqe_prev
) {
2069 QTAILQ_REMOVE(&bdrv_states
, bs
, device_list
);
2070 bs
->device_list
.tqe_prev
= NULL
;
2072 if (bs
->node_name
[0] != '\0') {
2073 QTAILQ_REMOVE(&graph_bdrv_states
, bs
, node_list
);
2075 bs
->node_name
[0] = '\0';
2078 static void bdrv_rebind(BlockDriverState
*bs
)
2080 if (bs
->drv
&& bs
->drv
->bdrv_rebind
) {
2081 bs
->drv
->bdrv_rebind(bs
);
2085 static void bdrv_move_feature_fields(BlockDriverState
*bs_dest
,
2086 BlockDriverState
*bs_src
)
2088 /* move some fields that need to stay attached to the device */
2091 bs_dest
->guest_block_size
= bs_src
->guest_block_size
;
2092 bs_dest
->copy_on_read
= bs_src
->copy_on_read
;
2094 bs_dest
->enable_write_cache
= bs_src
->enable_write_cache
;
2096 /* i/o throttled req */
2097 memcpy(&bs_dest
->throttle_state
,
2098 &bs_src
->throttle_state
,
2099 sizeof(ThrottleState
));
2100 bs_dest
->throttled_reqs
[0] = bs_src
->throttled_reqs
[0];
2101 bs_dest
->throttled_reqs
[1] = bs_src
->throttled_reqs
[1];
2102 bs_dest
->io_limits_enabled
= bs_src
->io_limits_enabled
;
2105 bs_dest
->on_read_error
= bs_src
->on_read_error
;
2106 bs_dest
->on_write_error
= bs_src
->on_write_error
;
2109 bs_dest
->iostatus_enabled
= bs_src
->iostatus_enabled
;
2110 bs_dest
->iostatus
= bs_src
->iostatus
;
2113 bs_dest
->dirty_bitmaps
= bs_src
->dirty_bitmaps
;
2115 /* reference count */
2116 bs_dest
->refcnt
= bs_src
->refcnt
;
2119 bs_dest
->job
= bs_src
->job
;
2121 /* keep the same entry in bdrv_states */
2122 bs_dest
->device_list
= bs_src
->device_list
;
2123 bs_dest
->blk
= bs_src
->blk
;
2125 memcpy(bs_dest
->op_blockers
, bs_src
->op_blockers
,
2126 sizeof(bs_dest
->op_blockers
));
2130 * Swap bs contents for two image chains while they are live,
2131 * while keeping required fields on the BlockDriverState that is
2132 * actually attached to a device.
2134 * This will modify the BlockDriverState fields, and swap contents
2135 * between bs_new and bs_old. Both bs_new and bs_old are modified.
2137 * bs_new must not be attached to a BlockBackend.
2139 * This function does not create any image files.
2141 void bdrv_swap(BlockDriverState
*bs_new
, BlockDriverState
*bs_old
)
2143 BlockDriverState tmp
;
2145 /* The code needs to swap the node_name but simply swapping node_list won't
2146 * work so first remove the nodes from the graph list, do the swap then
2147 * insert them back if needed.
2149 if (bs_new
->node_name
[0] != '\0') {
2150 QTAILQ_REMOVE(&graph_bdrv_states
, bs_new
, node_list
);
2152 if (bs_old
->node_name
[0] != '\0') {
2153 QTAILQ_REMOVE(&graph_bdrv_states
, bs_old
, node_list
);
2156 /* bs_new must be unattached and shouldn't have anything fancy enabled */
2157 assert(!bs_new
->blk
);
2158 assert(QLIST_EMPTY(&bs_new
->dirty_bitmaps
));
2159 assert(bs_new
->job
== NULL
);
2160 assert(bs_new
->io_limits_enabled
== false);
2161 assert(!throttle_have_timer(&bs_new
->throttle_state
));
2167 /* there are some fields that should not be swapped, move them back */
2168 bdrv_move_feature_fields(&tmp
, bs_old
);
2169 bdrv_move_feature_fields(bs_old
, bs_new
);
2170 bdrv_move_feature_fields(bs_new
, &tmp
);
2172 /* bs_new must remain unattached */
2173 assert(!bs_new
->blk
);
2175 /* Check a few fields that should remain attached to the device */
2176 assert(bs_new
->job
== NULL
);
2177 assert(bs_new
->io_limits_enabled
== false);
2178 assert(!throttle_have_timer(&bs_new
->throttle_state
));
2180 /* insert the nodes back into the graph node list if needed */
2181 if (bs_new
->node_name
[0] != '\0') {
2182 QTAILQ_INSERT_TAIL(&graph_bdrv_states
, bs_new
, node_list
);
2184 if (bs_old
->node_name
[0] != '\0') {
2185 QTAILQ_INSERT_TAIL(&graph_bdrv_states
, bs_old
, node_list
);
2188 bdrv_rebind(bs_new
);
2189 bdrv_rebind(bs_old
);
2193 * Add new bs contents at the top of an image chain while the chain is
2194 * live, while keeping required fields on the top layer.
2196 * This will modify the BlockDriverState fields, and swap contents
2197 * between bs_new and bs_top. Both bs_new and bs_top are modified.
2199 * bs_new must not be attached to a BlockBackend.
2201 * This function does not create any image files.
2203 void bdrv_append(BlockDriverState
*bs_new
, BlockDriverState
*bs_top
)
2205 bdrv_swap(bs_new
, bs_top
);
2207 /* The contents of 'tmp' will become bs_top, as we are
2208 * swapping bs_new and bs_top contents. */
2209 bdrv_set_backing_hd(bs_top
, bs_new
);
2212 static void bdrv_delete(BlockDriverState
*bs
)
2215 assert(bdrv_op_blocker_is_empty(bs
));
2216 assert(!bs
->refcnt
);
2217 assert(QLIST_EMPTY(&bs
->dirty_bitmaps
));
2221 /* remove from list, if necessary */
2228 * Run consistency checks on an image
2230 * Returns 0 if the check could be completed (it doesn't mean that the image is
2231 * free of errors) or -errno when an internal error occurred. The results of the
2232 * check are stored in res.
2234 int bdrv_check(BlockDriverState
*bs
, BdrvCheckResult
*res
, BdrvCheckMode fix
)
2236 if (bs
->drv
== NULL
) {
2239 if (bs
->drv
->bdrv_check
== NULL
) {
2243 memset(res
, 0, sizeof(*res
));
2244 return bs
->drv
->bdrv_check(bs
, res
, fix
);
2247 #define COMMIT_BUF_SECTORS 2048
2249 /* commit COW file into the raw image */
2250 int bdrv_commit(BlockDriverState
*bs
)
2252 BlockDriver
*drv
= bs
->drv
;
2253 int64_t sector
, total_sectors
, length
, backing_length
;
2254 int n
, ro
, open_flags
;
2256 uint8_t *buf
= NULL
;
2261 if (!bs
->backing_hd
) {
2265 if (bdrv_op_is_blocked(bs
, BLOCK_OP_TYPE_COMMIT_SOURCE
, NULL
) ||
2266 bdrv_op_is_blocked(bs
->backing_hd
, BLOCK_OP_TYPE_COMMIT_TARGET
, NULL
)) {
2270 ro
= bs
->backing_hd
->read_only
;
2271 open_flags
= bs
->backing_hd
->open_flags
;
2274 if (bdrv_reopen(bs
->backing_hd
, open_flags
| BDRV_O_RDWR
, NULL
)) {
2279 length
= bdrv_getlength(bs
);
2285 backing_length
= bdrv_getlength(bs
->backing_hd
);
2286 if (backing_length
< 0) {
2287 ret
= backing_length
;
2291 /* If our top snapshot is larger than the backing file image,
2292 * grow the backing file image if possible. If not possible,
2293 * we must return an error */
2294 if (length
> backing_length
) {
2295 ret
= bdrv_truncate(bs
->backing_hd
, length
);
2301 total_sectors
= length
>> BDRV_SECTOR_BITS
;
2303 /* qemu_try_blockalign() for bs will choose an alignment that works for
2304 * bs->backing_hd as well, so no need to compare the alignment manually. */
2305 buf
= qemu_try_blockalign(bs
, COMMIT_BUF_SECTORS
* BDRV_SECTOR_SIZE
);
2311 for (sector
= 0; sector
< total_sectors
; sector
+= n
) {
2312 ret
= bdrv_is_allocated(bs
, sector
, COMMIT_BUF_SECTORS
, &n
);
2317 ret
= bdrv_read(bs
, sector
, buf
, n
);
2322 ret
= bdrv_write(bs
->backing_hd
, sector
, buf
, n
);
2329 if (drv
->bdrv_make_empty
) {
2330 ret
= drv
->bdrv_make_empty(bs
);
2338 * Make sure all data we wrote to the backing device is actually
2341 if (bs
->backing_hd
) {
2342 bdrv_flush(bs
->backing_hd
);
2350 /* ignoring error return here */
2351 bdrv_reopen(bs
->backing_hd
, open_flags
& ~BDRV_O_RDWR
, NULL
);
2357 int bdrv_commit_all(void)
2359 BlockDriverState
*bs
;
2361 QTAILQ_FOREACH(bs
, &bdrv_states
, device_list
) {
2362 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
2364 aio_context_acquire(aio_context
);
2365 if (bs
->drv
&& bs
->backing_hd
) {
2366 int ret
= bdrv_commit(bs
);
2368 aio_context_release(aio_context
);
2372 aio_context_release(aio_context
);
2378 * Remove an active request from the tracked requests list
2380 * This function should be called when a tracked request is completing.
2382 static void tracked_request_end(BdrvTrackedRequest
*req
)
2384 if (req
->serialising
) {
2385 req
->bs
->serialising_in_flight
--;
2388 QLIST_REMOVE(req
, list
);
2389 qemu_co_queue_restart_all(&req
->wait_queue
);
2393 * Add an active request to the tracked requests list
2395 static void tracked_request_begin(BdrvTrackedRequest
*req
,
2396 BlockDriverState
*bs
,
2398 unsigned int bytes
, bool is_write
)
2400 *req
= (BdrvTrackedRequest
){
2404 .is_write
= is_write
,
2405 .co
= qemu_coroutine_self(),
2406 .serialising
= false,
2407 .overlap_offset
= offset
,
2408 .overlap_bytes
= bytes
,
2411 qemu_co_queue_init(&req
->wait_queue
);
2413 QLIST_INSERT_HEAD(&bs
->tracked_requests
, req
, list
);
2416 static void mark_request_serialising(BdrvTrackedRequest
*req
, uint64_t align
)
2418 int64_t overlap_offset
= req
->offset
& ~(align
- 1);
2419 unsigned int overlap_bytes
= ROUND_UP(req
->offset
+ req
->bytes
, align
)
2422 if (!req
->serialising
) {
2423 req
->bs
->serialising_in_flight
++;
2424 req
->serialising
= true;
2427 req
->overlap_offset
= MIN(req
->overlap_offset
, overlap_offset
);
2428 req
->overlap_bytes
= MAX(req
->overlap_bytes
, overlap_bytes
);
2432 * Round a region to cluster boundaries
2434 void bdrv_round_to_clusters(BlockDriverState
*bs
,
2435 int64_t sector_num
, int nb_sectors
,
2436 int64_t *cluster_sector_num
,
2437 int *cluster_nb_sectors
)
2439 BlockDriverInfo bdi
;
2441 if (bdrv_get_info(bs
, &bdi
) < 0 || bdi
.cluster_size
== 0) {
2442 *cluster_sector_num
= sector_num
;
2443 *cluster_nb_sectors
= nb_sectors
;
2445 int64_t c
= bdi
.cluster_size
/ BDRV_SECTOR_SIZE
;
2446 *cluster_sector_num
= QEMU_ALIGN_DOWN(sector_num
, c
);
2447 *cluster_nb_sectors
= QEMU_ALIGN_UP(sector_num
- *cluster_sector_num
+
2452 static int bdrv_get_cluster_size(BlockDriverState
*bs
)
2454 BlockDriverInfo bdi
;
2457 ret
= bdrv_get_info(bs
, &bdi
);
2458 if (ret
< 0 || bdi
.cluster_size
== 0) {
2459 return bs
->request_alignment
;
2461 return bdi
.cluster_size
;
2465 static bool tracked_request_overlaps(BdrvTrackedRequest
*req
,
2466 int64_t offset
, unsigned int bytes
)
2469 if (offset
>= req
->overlap_offset
+ req
->overlap_bytes
) {
2473 if (req
->overlap_offset
>= offset
+ bytes
) {
2479 static bool coroutine_fn
wait_serialising_requests(BdrvTrackedRequest
*self
)
2481 BlockDriverState
*bs
= self
->bs
;
2482 BdrvTrackedRequest
*req
;
2484 bool waited
= false;
2486 if (!bs
->serialising_in_flight
) {
2492 QLIST_FOREACH(req
, &bs
->tracked_requests
, list
) {
2493 if (req
== self
|| (!req
->serialising
&& !self
->serialising
)) {
2496 if (tracked_request_overlaps(req
, self
->overlap_offset
,
2497 self
->overlap_bytes
))
2499 /* Hitting this means there was a reentrant request, for
2500 * example, a block driver issuing nested requests. This must
2501 * never happen since it means deadlock.
2503 assert(qemu_coroutine_self() != req
->co
);
2505 /* If the request is already (indirectly) waiting for us, or
2506 * will wait for us as soon as it wakes up, then just go on
2507 * (instead of producing a deadlock in the former case). */
2508 if (!req
->waiting_for
) {
2509 self
->waiting_for
= req
;
2510 qemu_co_queue_wait(&req
->wait_queue
);
2511 self
->waiting_for
= NULL
;
2526 * -EINVAL - backing format specified, but no file
2527 * -ENOSPC - can't update the backing file because no space is left in the
2529 * -ENOTSUP - format driver doesn't support changing the backing file
2531 int bdrv_change_backing_file(BlockDriverState
*bs
,
2532 const char *backing_file
, const char *backing_fmt
)
2534 BlockDriver
*drv
= bs
->drv
;
2537 /* Backing file format doesn't make sense without a backing file */
2538 if (backing_fmt
&& !backing_file
) {
2542 if (drv
->bdrv_change_backing_file
!= NULL
) {
2543 ret
= drv
->bdrv_change_backing_file(bs
, backing_file
, backing_fmt
);
2549 pstrcpy(bs
->backing_file
, sizeof(bs
->backing_file
), backing_file
?: "");
2550 pstrcpy(bs
->backing_format
, sizeof(bs
->backing_format
), backing_fmt
?: "");
2556 * Finds the image layer in the chain that has 'bs' as its backing file.
2558 * active is the current topmost image.
2560 * Returns NULL if bs is not found in active's image chain,
2561 * or if active == bs.
2563 * Returns the bottommost base image if bs == NULL.
2565 BlockDriverState
*bdrv_find_overlay(BlockDriverState
*active
,
2566 BlockDriverState
*bs
)
2568 while (active
&& bs
!= active
->backing_hd
) {
2569 active
= active
->backing_hd
;
2575 /* Given a BDS, searches for the base layer. */
2576 BlockDriverState
*bdrv_find_base(BlockDriverState
*bs
)
2578 return bdrv_find_overlay(bs
, NULL
);
2581 typedef struct BlkIntermediateStates
{
2582 BlockDriverState
*bs
;
2583 QSIMPLEQ_ENTRY(BlkIntermediateStates
) entry
;
2584 } BlkIntermediateStates
;
2588 * Drops images above 'base' up to and including 'top', and sets the image
2589 * above 'top' to have base as its backing file.
2591 * Requires that the overlay to 'top' is opened r/w, so that the backing file
2592 * information in 'bs' can be properly updated.
2594 * E.g., this will convert the following chain:
2595 * bottom <- base <- intermediate <- top <- active
2599 * bottom <- base <- active
2601 * It is allowed for bottom==base, in which case it converts:
2603 * base <- intermediate <- top <- active
2609 * If backing_file_str is non-NULL, it will be used when modifying top's
2610 * overlay image metadata.
2613 * if active == top, that is considered an error
2616 int bdrv_drop_intermediate(BlockDriverState
*active
, BlockDriverState
*top
,
2617 BlockDriverState
*base
, const char *backing_file_str
)
2619 BlockDriverState
*intermediate
;
2620 BlockDriverState
*base_bs
= NULL
;
2621 BlockDriverState
*new_top_bs
= NULL
;
2622 BlkIntermediateStates
*intermediate_state
, *next
;
2625 QSIMPLEQ_HEAD(states_to_delete
, BlkIntermediateStates
) states_to_delete
;
2626 QSIMPLEQ_INIT(&states_to_delete
);
2628 if (!top
->drv
|| !base
->drv
) {
2632 new_top_bs
= bdrv_find_overlay(active
, top
);
2634 if (new_top_bs
== NULL
) {
2635 /* we could not find the image above 'top', this is an error */
2639 /* special case of new_top_bs->backing_hd already pointing to base - nothing
2640 * to do, no intermediate images */
2641 if (new_top_bs
->backing_hd
== base
) {
2648 /* now we will go down through the list, and add each BDS we find
2649 * into our deletion queue, until we hit the 'base'
2651 while (intermediate
) {
2652 intermediate_state
= g_new0(BlkIntermediateStates
, 1);
2653 intermediate_state
->bs
= intermediate
;
2654 QSIMPLEQ_INSERT_TAIL(&states_to_delete
, intermediate_state
, entry
);
2656 if (intermediate
->backing_hd
== base
) {
2657 base_bs
= intermediate
->backing_hd
;
2660 intermediate
= intermediate
->backing_hd
;
2662 if (base_bs
== NULL
) {
2663 /* something went wrong, we did not end at the base. safely
2664 * unravel everything, and exit with error */
2668 /* success - we can delete the intermediate states, and link top->base */
2669 backing_file_str
= backing_file_str
? backing_file_str
: base_bs
->filename
;
2670 ret
= bdrv_change_backing_file(new_top_bs
, backing_file_str
,
2671 base_bs
->drv
? base_bs
->drv
->format_name
: "");
2675 bdrv_set_backing_hd(new_top_bs
, base_bs
);
2677 QSIMPLEQ_FOREACH_SAFE(intermediate_state
, &states_to_delete
, entry
, next
) {
2678 /* so that bdrv_close() does not recursively close the chain */
2679 bdrv_set_backing_hd(intermediate_state
->bs
, NULL
);
2680 bdrv_unref(intermediate_state
->bs
);
2685 QSIMPLEQ_FOREACH_SAFE(intermediate_state
, &states_to_delete
, entry
, next
) {
2686 g_free(intermediate_state
);
2692 static int bdrv_check_byte_request(BlockDriverState
*bs
, int64_t offset
,
2695 if (size
> BDRV_REQUEST_MAX_SECTORS
<< BDRV_SECTOR_BITS
) {
2699 if (!bdrv_is_inserted(bs
)) {
2710 static int bdrv_check_request(BlockDriverState
*bs
, int64_t sector_num
,
2713 if (nb_sectors
< 0 || nb_sectors
> BDRV_REQUEST_MAX_SECTORS
) {
2717 return bdrv_check_byte_request(bs
, sector_num
* BDRV_SECTOR_SIZE
,
2718 nb_sectors
* BDRV_SECTOR_SIZE
);
2721 typedef struct RwCo
{
2722 BlockDriverState
*bs
;
2727 BdrvRequestFlags flags
;
2730 static void coroutine_fn
bdrv_rw_co_entry(void *opaque
)
2732 RwCo
*rwco
= opaque
;
2734 if (!rwco
->is_write
) {
2735 rwco
->ret
= bdrv_co_do_preadv(rwco
->bs
, rwco
->offset
,
2736 rwco
->qiov
->size
, rwco
->qiov
,
2739 rwco
->ret
= bdrv_co_do_pwritev(rwco
->bs
, rwco
->offset
,
2740 rwco
->qiov
->size
, rwco
->qiov
,
2746 * Process a vectored synchronous request using coroutines
2748 static int bdrv_prwv_co(BlockDriverState
*bs
, int64_t offset
,
2749 QEMUIOVector
*qiov
, bool is_write
,
2750 BdrvRequestFlags flags
)
2757 .is_write
= is_write
,
2763 * In sync call context, when the vcpu is blocked, this throttling timer
2764 * will not fire; so the I/O throttling function has to be disabled here
2765 * if it has been enabled.
2767 if (bs
->io_limits_enabled
) {
2768 fprintf(stderr
, "Disabling I/O throttling on '%s' due "
2769 "to synchronous I/O.\n", bdrv_get_device_name(bs
));
2770 bdrv_io_limits_disable(bs
);
2773 if (qemu_in_coroutine()) {
2774 /* Fast-path if already in coroutine context */
2775 bdrv_rw_co_entry(&rwco
);
2777 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
2779 co
= qemu_coroutine_create(bdrv_rw_co_entry
);
2780 qemu_coroutine_enter(co
, &rwco
);
2781 while (rwco
.ret
== NOT_DONE
) {
2782 aio_poll(aio_context
, true);
2789 * Process a synchronous request using coroutines
2791 static int bdrv_rw_co(BlockDriverState
*bs
, int64_t sector_num
, uint8_t *buf
,
2792 int nb_sectors
, bool is_write
, BdrvRequestFlags flags
)
2795 struct iovec iov
= {
2796 .iov_base
= (void *)buf
,
2797 .iov_len
= nb_sectors
* BDRV_SECTOR_SIZE
,
2800 if (nb_sectors
< 0 || nb_sectors
> BDRV_REQUEST_MAX_SECTORS
) {
2804 qemu_iovec_init_external(&qiov
, &iov
, 1);
2805 return bdrv_prwv_co(bs
, sector_num
<< BDRV_SECTOR_BITS
,
2806 &qiov
, is_write
, flags
);
2809 /* return < 0 if error. See bdrv_write() for the return codes */
2810 int bdrv_read(BlockDriverState
*bs
, int64_t sector_num
,
2811 uint8_t *buf
, int nb_sectors
)
2813 return bdrv_rw_co(bs
, sector_num
, buf
, nb_sectors
, false, 0);
2816 /* Just like bdrv_read(), but with I/O throttling temporarily disabled */
2817 int bdrv_read_unthrottled(BlockDriverState
*bs
, int64_t sector_num
,
2818 uint8_t *buf
, int nb_sectors
)
2823 enabled
= bs
->io_limits_enabled
;
2824 bs
->io_limits_enabled
= false;
2825 ret
= bdrv_read(bs
, sector_num
, buf
, nb_sectors
);
2826 bs
->io_limits_enabled
= enabled
;
2830 /* Return < 0 if error. Important errors are:
2831 -EIO generic I/O error (may happen for all errors)
2832 -ENOMEDIUM No media inserted.
2833 -EINVAL Invalid sector number or nb_sectors
2834 -EACCES Trying to write a read-only device
2836 int bdrv_write(BlockDriverState
*bs
, int64_t sector_num
,
2837 const uint8_t *buf
, int nb_sectors
)
2839 return bdrv_rw_co(bs
, sector_num
, (uint8_t *)buf
, nb_sectors
, true, 0);
2842 int bdrv_write_zeroes(BlockDriverState
*bs
, int64_t sector_num
,
2843 int nb_sectors
, BdrvRequestFlags flags
)
2845 return bdrv_rw_co(bs
, sector_num
, NULL
, nb_sectors
, true,
2846 BDRV_REQ_ZERO_WRITE
| flags
);
2850 * Completely zero out a block device with the help of bdrv_write_zeroes.
2851 * The operation is sped up by checking the block status and only writing
2852 * zeroes to the device if they currently do not return zeroes. Optional
2853 * flags are passed through to bdrv_write_zeroes (e.g. BDRV_REQ_MAY_UNMAP).
2855 * Returns < 0 on error, 0 on success. For error codes see bdrv_write().
2857 int bdrv_make_zero(BlockDriverState
*bs
, BdrvRequestFlags flags
)
2859 int64_t target_sectors
, ret
, nb_sectors
, sector_num
= 0;
2862 target_sectors
= bdrv_nb_sectors(bs
);
2863 if (target_sectors
< 0) {
2864 return target_sectors
;
2868 nb_sectors
= MIN(target_sectors
- sector_num
, BDRV_REQUEST_MAX_SECTORS
);
2869 if (nb_sectors
<= 0) {
2872 ret
= bdrv_get_block_status(bs
, sector_num
, nb_sectors
, &n
);
2874 error_report("error getting block status at sector %" PRId64
": %s",
2875 sector_num
, strerror(-ret
));
2878 if (ret
& BDRV_BLOCK_ZERO
) {
2882 ret
= bdrv_write_zeroes(bs
, sector_num
, n
, flags
);
2884 error_report("error writing zeroes at sector %" PRId64
": %s",
2885 sector_num
, strerror(-ret
));
2892 int bdrv_pread(BlockDriverState
*bs
, int64_t offset
, void *buf
, int bytes
)
2895 struct iovec iov
= {
2896 .iov_base
= (void *)buf
,
2905 qemu_iovec_init_external(&qiov
, &iov
, 1);
2906 ret
= bdrv_prwv_co(bs
, offset
, &qiov
, false, 0);
2914 int bdrv_pwritev(BlockDriverState
*bs
, int64_t offset
, QEMUIOVector
*qiov
)
2918 ret
= bdrv_prwv_co(bs
, offset
, qiov
, true, 0);
2926 int bdrv_pwrite(BlockDriverState
*bs
, int64_t offset
,
2927 const void *buf
, int bytes
)
2930 struct iovec iov
= {
2931 .iov_base
= (void *) buf
,
2939 qemu_iovec_init_external(&qiov
, &iov
, 1);
2940 return bdrv_pwritev(bs
, offset
, &qiov
);
2944 * Writes to the file and ensures that no writes are reordered across this
2945 * request (acts as a barrier)
2947 * Returns 0 on success, -errno in error cases.
2949 int bdrv_pwrite_sync(BlockDriverState
*bs
, int64_t offset
,
2950 const void *buf
, int count
)
2954 ret
= bdrv_pwrite(bs
, offset
, buf
, count
);
2959 /* No flush needed for cache modes that already do it */
2960 if (bs
->enable_write_cache
) {
2967 static int coroutine_fn
bdrv_co_do_copy_on_readv(BlockDriverState
*bs
,
2968 int64_t sector_num
, int nb_sectors
, QEMUIOVector
*qiov
)
2970 /* Perform I/O through a temporary buffer so that users who scribble over
2971 * their read buffer while the operation is in progress do not end up
2972 * modifying the image file. This is critical for zero-copy guest I/O
2973 * where anything might happen inside guest memory.
2975 void *bounce_buffer
;
2977 BlockDriver
*drv
= bs
->drv
;
2979 QEMUIOVector bounce_qiov
;
2980 int64_t cluster_sector_num
;
2981 int cluster_nb_sectors
;
2985 /* Cover entire cluster so no additional backing file I/O is required when
2986 * allocating cluster in the image file.
2988 bdrv_round_to_clusters(bs
, sector_num
, nb_sectors
,
2989 &cluster_sector_num
, &cluster_nb_sectors
);
2991 trace_bdrv_co_do_copy_on_readv(bs
, sector_num
, nb_sectors
,
2992 cluster_sector_num
, cluster_nb_sectors
);
2994 iov
.iov_len
= cluster_nb_sectors
* BDRV_SECTOR_SIZE
;
2995 iov
.iov_base
= bounce_buffer
= qemu_try_blockalign(bs
, iov
.iov_len
);
2996 if (bounce_buffer
== NULL
) {
3001 qemu_iovec_init_external(&bounce_qiov
, &iov
, 1);
3003 ret
= drv
->bdrv_co_readv(bs
, cluster_sector_num
, cluster_nb_sectors
,
3009 if (drv
->bdrv_co_write_zeroes
&&
3010 buffer_is_zero(bounce_buffer
, iov
.iov_len
)) {
3011 ret
= bdrv_co_do_write_zeroes(bs
, cluster_sector_num
,
3012 cluster_nb_sectors
, 0);
3014 /* This does not change the data on the disk, it is not necessary
3015 * to flush even in cache=writethrough mode.
3017 ret
= drv
->bdrv_co_writev(bs
, cluster_sector_num
, cluster_nb_sectors
,
3022 /* It might be okay to ignore write errors for guest requests. If this
3023 * is a deliberate copy-on-read then we don't want to ignore the error.
3024 * Simply report it in all cases.
3029 skip_bytes
= (sector_num
- cluster_sector_num
) * BDRV_SECTOR_SIZE
;
3030 qemu_iovec_from_buf(qiov
, 0, bounce_buffer
+ skip_bytes
,
3031 nb_sectors
* BDRV_SECTOR_SIZE
);
3034 qemu_vfree(bounce_buffer
);
3039 * Forwards an already correctly aligned request to the BlockDriver. This
3040 * handles copy on read and zeroing after EOF; any other features must be
3041 * implemented by the caller.
3043 static int coroutine_fn
bdrv_aligned_preadv(BlockDriverState
*bs
,
3044 BdrvTrackedRequest
*req
, int64_t offset
, unsigned int bytes
,
3045 int64_t align
, QEMUIOVector
*qiov
, int flags
)
3047 BlockDriver
*drv
= bs
->drv
;
3050 int64_t sector_num
= offset
>> BDRV_SECTOR_BITS
;
3051 unsigned int nb_sectors
= bytes
>> BDRV_SECTOR_BITS
;
3053 assert((offset
& (BDRV_SECTOR_SIZE
- 1)) == 0);
3054 assert((bytes
& (BDRV_SECTOR_SIZE
- 1)) == 0);
3055 assert(!qiov
|| bytes
== qiov
->size
);
3057 /* Handle Copy on Read and associated serialisation */
3058 if (flags
& BDRV_REQ_COPY_ON_READ
) {
3059 /* If we touch the same cluster it counts as an overlap. This
3060 * guarantees that allocating writes will be serialized and not race
3061 * with each other for the same cluster. For example, in copy-on-read
3062 * it ensures that the CoR read and write operations are atomic and
3063 * guest writes cannot interleave between them. */
3064 mark_request_serialising(req
, bdrv_get_cluster_size(bs
));
3067 wait_serialising_requests(req
);
3069 if (flags
& BDRV_REQ_COPY_ON_READ
) {
3072 ret
= bdrv_is_allocated(bs
, sector_num
, nb_sectors
, &pnum
);
3077 if (!ret
|| pnum
!= nb_sectors
) {
3078 ret
= bdrv_co_do_copy_on_readv(bs
, sector_num
, nb_sectors
, qiov
);
3083 /* Forward the request to the BlockDriver */
3084 if (!bs
->zero_beyond_eof
) {
3085 ret
= drv
->bdrv_co_readv(bs
, sector_num
, nb_sectors
, qiov
);
3087 /* Read zeros after EOF */
3088 int64_t total_sectors
, max_nb_sectors
;
3090 total_sectors
= bdrv_nb_sectors(bs
);
3091 if (total_sectors
< 0) {
3092 ret
= total_sectors
;
3096 max_nb_sectors
= ROUND_UP(MAX(0, total_sectors
- sector_num
),
3097 align
>> BDRV_SECTOR_BITS
);
3098 if (nb_sectors
< max_nb_sectors
) {
3099 ret
= drv
->bdrv_co_readv(bs
, sector_num
, nb_sectors
, qiov
);
3100 } else if (max_nb_sectors
> 0) {
3101 QEMUIOVector local_qiov
;
3103 qemu_iovec_init(&local_qiov
, qiov
->niov
);
3104 qemu_iovec_concat(&local_qiov
, qiov
, 0,
3105 max_nb_sectors
* BDRV_SECTOR_SIZE
);
3107 ret
= drv
->bdrv_co_readv(bs
, sector_num
, max_nb_sectors
,
3110 qemu_iovec_destroy(&local_qiov
);
3115 /* Reading beyond end of file is supposed to produce zeroes */
3116 if (ret
== 0 && total_sectors
< sector_num
+ nb_sectors
) {
3117 uint64_t offset
= MAX(0, total_sectors
- sector_num
);
3118 uint64_t bytes
= (sector_num
+ nb_sectors
- offset
) *
3120 qemu_iovec_memset(qiov
, offset
* BDRV_SECTOR_SIZE
, 0, bytes
);
3128 static inline uint64_t bdrv_get_align(BlockDriverState
*bs
)
3130 /* TODO Lift BDRV_SECTOR_SIZE restriction in BlockDriver interface */
3131 return MAX(BDRV_SECTOR_SIZE
, bs
->request_alignment
);
3134 static inline bool bdrv_req_is_aligned(BlockDriverState
*bs
,
3135 int64_t offset
, size_t bytes
)
3137 int64_t align
= bdrv_get_align(bs
);
3138 return !(offset
& (align
- 1) || (bytes
& (align
- 1)));
3142 * Handle a read request in coroutine context
3144 static int coroutine_fn
bdrv_co_do_preadv(BlockDriverState
*bs
,
3145 int64_t offset
, unsigned int bytes
, QEMUIOVector
*qiov
,
3146 BdrvRequestFlags flags
)
3148 BlockDriver
*drv
= bs
->drv
;
3149 BdrvTrackedRequest req
;
3151 uint64_t align
= bdrv_get_align(bs
);
3152 uint8_t *head_buf
= NULL
;
3153 uint8_t *tail_buf
= NULL
;
3154 QEMUIOVector local_qiov
;
3155 bool use_local_qiov
= false;
3162 ret
= bdrv_check_byte_request(bs
, offset
, bytes
);
3167 if (bs
->copy_on_read
) {
3168 flags
|= BDRV_REQ_COPY_ON_READ
;
3171 /* throttling disk I/O */
3172 if (bs
->io_limits_enabled
) {
3173 bdrv_io_limits_intercept(bs
, bytes
, false);
3176 /* Align read if necessary by padding qiov */
3177 if (offset
& (align
- 1)) {
3178 head_buf
= qemu_blockalign(bs
, align
);
3179 qemu_iovec_init(&local_qiov
, qiov
->niov
+ 2);
3180 qemu_iovec_add(&local_qiov
, head_buf
, offset
& (align
- 1));
3181 qemu_iovec_concat(&local_qiov
, qiov
, 0, qiov
->size
);
3182 use_local_qiov
= true;
3184 bytes
+= offset
& (align
- 1);
3185 offset
= offset
& ~(align
- 1);
3188 if ((offset
+ bytes
) & (align
- 1)) {
3189 if (!use_local_qiov
) {
3190 qemu_iovec_init(&local_qiov
, qiov
->niov
+ 1);
3191 qemu_iovec_concat(&local_qiov
, qiov
, 0, qiov
->size
);
3192 use_local_qiov
= true;
3194 tail_buf
= qemu_blockalign(bs
, align
);
3195 qemu_iovec_add(&local_qiov
, tail_buf
,
3196 align
- ((offset
+ bytes
) & (align
- 1)));
3198 bytes
= ROUND_UP(bytes
, align
);
3201 tracked_request_begin(&req
, bs
, offset
, bytes
, false);
3202 ret
= bdrv_aligned_preadv(bs
, &req
, offset
, bytes
, align
,
3203 use_local_qiov
? &local_qiov
: qiov
,
3205 tracked_request_end(&req
);
3207 if (use_local_qiov
) {
3208 qemu_iovec_destroy(&local_qiov
);
3209 qemu_vfree(head_buf
);
3210 qemu_vfree(tail_buf
);
3216 static int coroutine_fn
bdrv_co_do_readv(BlockDriverState
*bs
,
3217 int64_t sector_num
, int nb_sectors
, QEMUIOVector
*qiov
,
3218 BdrvRequestFlags flags
)
3220 if (nb_sectors
< 0 || nb_sectors
> BDRV_REQUEST_MAX_SECTORS
) {
3224 return bdrv_co_do_preadv(bs
, sector_num
<< BDRV_SECTOR_BITS
,
3225 nb_sectors
<< BDRV_SECTOR_BITS
, qiov
, flags
);
3228 int coroutine_fn
bdrv_co_readv(BlockDriverState
*bs
, int64_t sector_num
,
3229 int nb_sectors
, QEMUIOVector
*qiov
)
3231 trace_bdrv_co_readv(bs
, sector_num
, nb_sectors
);
3233 return bdrv_co_do_readv(bs
, sector_num
, nb_sectors
, qiov
, 0);
3236 int coroutine_fn
bdrv_co_copy_on_readv(BlockDriverState
*bs
,
3237 int64_t sector_num
, int nb_sectors
, QEMUIOVector
*qiov
)
3239 trace_bdrv_co_copy_on_readv(bs
, sector_num
, nb_sectors
);
3241 return bdrv_co_do_readv(bs
, sector_num
, nb_sectors
, qiov
,
3242 BDRV_REQ_COPY_ON_READ
);
3245 #define MAX_WRITE_ZEROES_BOUNCE_BUFFER 32768
3247 static int coroutine_fn
bdrv_co_do_write_zeroes(BlockDriverState
*bs
,
3248 int64_t sector_num
, int nb_sectors
, BdrvRequestFlags flags
)
3250 BlockDriver
*drv
= bs
->drv
;
3252 struct iovec iov
= {0};
3255 int max_write_zeroes
= MIN_NON_ZERO(bs
->bl
.max_write_zeroes
,
3256 BDRV_REQUEST_MAX_SECTORS
);
3258 while (nb_sectors
> 0 && !ret
) {
3259 int num
= nb_sectors
;
3261 /* Align request. Block drivers can expect the "bulk" of the request
3264 if (bs
->bl
.write_zeroes_alignment
3265 && num
> bs
->bl
.write_zeroes_alignment
) {
3266 if (sector_num
% bs
->bl
.write_zeroes_alignment
!= 0) {
3267 /* Make a small request up to the first aligned sector. */
3268 num
= bs
->bl
.write_zeroes_alignment
;
3269 num
-= sector_num
% bs
->bl
.write_zeroes_alignment
;
3270 } else if ((sector_num
+ num
) % bs
->bl
.write_zeroes_alignment
!= 0) {
3271 /* Shorten the request to the last aligned sector. num cannot
3272 * underflow because num > bs->bl.write_zeroes_alignment.
3274 num
-= (sector_num
+ num
) % bs
->bl
.write_zeroes_alignment
;
3278 /* limit request size */
3279 if (num
> max_write_zeroes
) {
3280 num
= max_write_zeroes
;
3284 /* First try the efficient write zeroes operation */
3285 if (drv
->bdrv_co_write_zeroes
) {
3286 ret
= drv
->bdrv_co_write_zeroes(bs
, sector_num
, num
, flags
);
3289 if (ret
== -ENOTSUP
) {
3290 /* Fall back to bounce buffer if write zeroes is unsupported */
3291 int max_xfer_len
= MIN_NON_ZERO(bs
->bl
.max_transfer_length
,
3292 MAX_WRITE_ZEROES_BOUNCE_BUFFER
);
3293 num
= MIN(num
, max_xfer_len
);
3294 iov
.iov_len
= num
* BDRV_SECTOR_SIZE
;
3295 if (iov
.iov_base
== NULL
) {
3296 iov
.iov_base
= qemu_try_blockalign(bs
, num
* BDRV_SECTOR_SIZE
);
3297 if (iov
.iov_base
== NULL
) {
3301 memset(iov
.iov_base
, 0, num
* BDRV_SECTOR_SIZE
);
3303 qemu_iovec_init_external(&qiov
, &iov
, 1);
3305 ret
= drv
->bdrv_co_writev(bs
, sector_num
, num
, &qiov
);
3307 /* Keep bounce buffer around if it is big enough for all
3308 * all future requests.
3310 if (num
< max_xfer_len
) {
3311 qemu_vfree(iov
.iov_base
);
3312 iov
.iov_base
= NULL
;
3321 qemu_vfree(iov
.iov_base
);
3326 * Forwards an already correctly aligned write request to the BlockDriver.
3328 static int coroutine_fn
bdrv_aligned_pwritev(BlockDriverState
*bs
,
3329 BdrvTrackedRequest
*req
, int64_t offset
, unsigned int bytes
,
3330 QEMUIOVector
*qiov
, int flags
)
3332 BlockDriver
*drv
= bs
->drv
;
3336 int64_t sector_num
= offset
>> BDRV_SECTOR_BITS
;
3337 unsigned int nb_sectors
= bytes
>> BDRV_SECTOR_BITS
;
3339 assert((offset
& (BDRV_SECTOR_SIZE
- 1)) == 0);
3340 assert((bytes
& (BDRV_SECTOR_SIZE
- 1)) == 0);
3341 assert(!qiov
|| bytes
== qiov
->size
);
3343 waited
= wait_serialising_requests(req
);
3344 assert(!waited
|| !req
->serialising
);
3345 assert(req
->overlap_offset
<= offset
);
3346 assert(offset
+ bytes
<= req
->overlap_offset
+ req
->overlap_bytes
);
3348 ret
= notifier_with_return_list_notify(&bs
->before_write_notifiers
, req
);
3350 if (!ret
&& bs
->detect_zeroes
!= BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF
&&
3351 !(flags
& BDRV_REQ_ZERO_WRITE
) && drv
->bdrv_co_write_zeroes
&&
3352 qemu_iovec_is_zero(qiov
)) {
3353 flags
|= BDRV_REQ_ZERO_WRITE
;
3354 if (bs
->detect_zeroes
== BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP
) {
3355 flags
|= BDRV_REQ_MAY_UNMAP
;
3360 /* Do nothing, write notifier decided to fail this request */
3361 } else if (flags
& BDRV_REQ_ZERO_WRITE
) {
3362 BLKDBG_EVENT(bs
, BLKDBG_PWRITEV_ZERO
);
3363 ret
= bdrv_co_do_write_zeroes(bs
, sector_num
, nb_sectors
, flags
);
3365 BLKDBG_EVENT(bs
, BLKDBG_PWRITEV
);
3366 ret
= drv
->bdrv_co_writev(bs
, sector_num
, nb_sectors
, qiov
);
3368 BLKDBG_EVENT(bs
, BLKDBG_PWRITEV_DONE
);
3370 if (ret
== 0 && !bs
->enable_write_cache
) {
3371 ret
= bdrv_co_flush(bs
);
3374 bdrv_set_dirty(bs
, sector_num
, nb_sectors
);
3376 block_acct_highest_sector(&bs
->stats
, sector_num
, nb_sectors
);
3379 bs
->total_sectors
= MAX(bs
->total_sectors
, sector_num
+ nb_sectors
);
3386 * Handle a write request in coroutine context
3388 static int coroutine_fn
bdrv_co_do_pwritev(BlockDriverState
*bs
,
3389 int64_t offset
, unsigned int bytes
, QEMUIOVector
*qiov
,
3390 BdrvRequestFlags flags
)
3392 BdrvTrackedRequest req
;
3393 uint64_t align
= bdrv_get_align(bs
);
3394 uint8_t *head_buf
= NULL
;
3395 uint8_t *tail_buf
= NULL
;
3396 QEMUIOVector local_qiov
;
3397 bool use_local_qiov
= false;
3403 if (bs
->read_only
) {
3407 ret
= bdrv_check_byte_request(bs
, offset
, bytes
);
3412 /* throttling disk I/O */
3413 if (bs
->io_limits_enabled
) {
3414 bdrv_io_limits_intercept(bs
, bytes
, true);
3418 * Align write if necessary by performing a read-modify-write cycle.
3419 * Pad qiov with the read parts and be sure to have a tracked request not
3420 * only for bdrv_aligned_pwritev, but also for the reads of the RMW cycle.
3422 tracked_request_begin(&req
, bs
, offset
, bytes
, true);
3424 if (offset
& (align
- 1)) {
3425 QEMUIOVector head_qiov
;
3426 struct iovec head_iov
;
3428 mark_request_serialising(&req
, align
);
3429 wait_serialising_requests(&req
);
3431 head_buf
= qemu_blockalign(bs
, align
);
3432 head_iov
= (struct iovec
) {
3433 .iov_base
= head_buf
,
3436 qemu_iovec_init_external(&head_qiov
, &head_iov
, 1);
3438 BLKDBG_EVENT(bs
, BLKDBG_PWRITEV_RMW_HEAD
);
3439 ret
= bdrv_aligned_preadv(bs
, &req
, offset
& ~(align
- 1), align
,
3440 align
, &head_qiov
, 0);
3444 BLKDBG_EVENT(bs
, BLKDBG_PWRITEV_RMW_AFTER_HEAD
);
3446 qemu_iovec_init(&local_qiov
, qiov
->niov
+ 2);
3447 qemu_iovec_add(&local_qiov
, head_buf
, offset
& (align
- 1));
3448 qemu_iovec_concat(&local_qiov
, qiov
, 0, qiov
->size
);
3449 use_local_qiov
= true;
3451 bytes
+= offset
& (align
- 1);
3452 offset
= offset
& ~(align
- 1);
3455 if ((offset
+ bytes
) & (align
- 1)) {
3456 QEMUIOVector tail_qiov
;
3457 struct iovec tail_iov
;
3461 mark_request_serialising(&req
, align
);
3462 waited
= wait_serialising_requests(&req
);
3463 assert(!waited
|| !use_local_qiov
);
3465 tail_buf
= qemu_blockalign(bs
, align
);
3466 tail_iov
= (struct iovec
) {
3467 .iov_base
= tail_buf
,
3470 qemu_iovec_init_external(&tail_qiov
, &tail_iov
, 1);
3472 BLKDBG_EVENT(bs
, BLKDBG_PWRITEV_RMW_TAIL
);
3473 ret
= bdrv_aligned_preadv(bs
, &req
, (offset
+ bytes
) & ~(align
- 1), align
,
3474 align
, &tail_qiov
, 0);
3478 BLKDBG_EVENT(bs
, BLKDBG_PWRITEV_RMW_AFTER_TAIL
);
3480 if (!use_local_qiov
) {
3481 qemu_iovec_init(&local_qiov
, qiov
->niov
+ 1);
3482 qemu_iovec_concat(&local_qiov
, qiov
, 0, qiov
->size
);
3483 use_local_qiov
= true;
3486 tail_bytes
= (offset
+ bytes
) & (align
- 1);
3487 qemu_iovec_add(&local_qiov
, tail_buf
+ tail_bytes
, align
- tail_bytes
);
3489 bytes
= ROUND_UP(bytes
, align
);
3492 if (use_local_qiov
) {
3493 /* Local buffer may have non-zero data. */
3494 flags
&= ~BDRV_REQ_ZERO_WRITE
;
3496 ret
= bdrv_aligned_pwritev(bs
, &req
, offset
, bytes
,
3497 use_local_qiov
? &local_qiov
: qiov
,
3501 tracked_request_end(&req
);
3503 if (use_local_qiov
) {
3504 qemu_iovec_destroy(&local_qiov
);
3506 qemu_vfree(head_buf
);
3507 qemu_vfree(tail_buf
);
3512 static int coroutine_fn
bdrv_co_do_writev(BlockDriverState
*bs
,
3513 int64_t sector_num
, int nb_sectors
, QEMUIOVector
*qiov
,
3514 BdrvRequestFlags flags
)
3516 if (nb_sectors
< 0 || nb_sectors
> BDRV_REQUEST_MAX_SECTORS
) {
3520 return bdrv_co_do_pwritev(bs
, sector_num
<< BDRV_SECTOR_BITS
,
3521 nb_sectors
<< BDRV_SECTOR_BITS
, qiov
, flags
);
3524 int coroutine_fn
bdrv_co_writev(BlockDriverState
*bs
, int64_t sector_num
,
3525 int nb_sectors
, QEMUIOVector
*qiov
)
3527 trace_bdrv_co_writev(bs
, sector_num
, nb_sectors
);
3529 return bdrv_co_do_writev(bs
, sector_num
, nb_sectors
, qiov
, 0);
3532 int coroutine_fn
bdrv_co_write_zeroes(BlockDriverState
*bs
,
3533 int64_t sector_num
, int nb_sectors
,
3534 BdrvRequestFlags flags
)
3538 trace_bdrv_co_write_zeroes(bs
, sector_num
, nb_sectors
, flags
);
3540 if (!(bs
->open_flags
& BDRV_O_UNMAP
)) {
3541 flags
&= ~BDRV_REQ_MAY_UNMAP
;
3543 if (bdrv_req_is_aligned(bs
, sector_num
<< BDRV_SECTOR_BITS
,
3544 nb_sectors
<< BDRV_SECTOR_BITS
)) {
3545 ret
= bdrv_co_do_writev(bs
, sector_num
, nb_sectors
, NULL
,
3546 BDRV_REQ_ZERO_WRITE
| flags
);
3549 QEMUIOVector local_qiov
;
3550 size_t bytes
= nb_sectors
<< BDRV_SECTOR_BITS
;
3552 buf
= qemu_memalign(bdrv_opt_mem_align(bs
), bytes
);
3553 memset(buf
, 0, bytes
);
3554 qemu_iovec_init(&local_qiov
, 1);
3555 qemu_iovec_add(&local_qiov
, buf
, bytes
);
3557 ret
= bdrv_co_do_writev(bs
, sector_num
, nb_sectors
, &local_qiov
,
3558 BDRV_REQ_ZERO_WRITE
| flags
);
3565 * Truncate file to 'offset' bytes (needed only for file protocols)
3567 int bdrv_truncate(BlockDriverState
*bs
, int64_t offset
)
3569 BlockDriver
*drv
= bs
->drv
;
3573 if (!drv
->bdrv_truncate
)
3578 ret
= drv
->bdrv_truncate(bs
, offset
);
3580 ret
= refresh_total_sectors(bs
, offset
>> BDRV_SECTOR_BITS
);
3582 blk_dev_resize_cb(bs
->blk
);
3589 * Length of a allocated file in bytes. Sparse files are counted by actual
3590 * allocated space. Return < 0 if error or unknown.
3592 int64_t bdrv_get_allocated_file_size(BlockDriverState
*bs
)
3594 BlockDriver
*drv
= bs
->drv
;
3598 if (drv
->bdrv_get_allocated_file_size
) {
3599 return drv
->bdrv_get_allocated_file_size(bs
);
3602 return bdrv_get_allocated_file_size(bs
->file
);
3608 * Return number of sectors on success, -errno on error.
3610 int64_t bdrv_nb_sectors(BlockDriverState
*bs
)
3612 BlockDriver
*drv
= bs
->drv
;
3617 if (drv
->has_variable_length
) {
3618 int ret
= refresh_total_sectors(bs
, bs
->total_sectors
);
3623 return bs
->total_sectors
;
3627 * Return length in bytes on success, -errno on error.
3628 * The length is always a multiple of BDRV_SECTOR_SIZE.
3630 int64_t bdrv_getlength(BlockDriverState
*bs
)
3632 int64_t ret
= bdrv_nb_sectors(bs
);
3634 return ret
< 0 ? ret
: ret
* BDRV_SECTOR_SIZE
;
3637 /* return 0 as number of sectors if no device present or error */
3638 void bdrv_get_geometry(BlockDriverState
*bs
, uint64_t *nb_sectors_ptr
)
3640 int64_t nb_sectors
= bdrv_nb_sectors(bs
);
3642 *nb_sectors_ptr
= nb_sectors
< 0 ? 0 : nb_sectors
;
3645 void bdrv_set_on_error(BlockDriverState
*bs
, BlockdevOnError on_read_error
,
3646 BlockdevOnError on_write_error
)
3648 bs
->on_read_error
= on_read_error
;
3649 bs
->on_write_error
= on_write_error
;
3652 BlockdevOnError
bdrv_get_on_error(BlockDriverState
*bs
, bool is_read
)
3654 return is_read
? bs
->on_read_error
: bs
->on_write_error
;
3657 BlockErrorAction
bdrv_get_error_action(BlockDriverState
*bs
, bool is_read
, int error
)
3659 BlockdevOnError on_err
= is_read
? bs
->on_read_error
: bs
->on_write_error
;
3662 case BLOCKDEV_ON_ERROR_ENOSPC
:
3663 return (error
== ENOSPC
) ?
3664 BLOCK_ERROR_ACTION_STOP
: BLOCK_ERROR_ACTION_REPORT
;
3665 case BLOCKDEV_ON_ERROR_STOP
:
3666 return BLOCK_ERROR_ACTION_STOP
;
3667 case BLOCKDEV_ON_ERROR_REPORT
:
3668 return BLOCK_ERROR_ACTION_REPORT
;
3669 case BLOCKDEV_ON_ERROR_IGNORE
:
3670 return BLOCK_ERROR_ACTION_IGNORE
;
3676 static void send_qmp_error_event(BlockDriverState
*bs
,
3677 BlockErrorAction action
,
3678 bool is_read
, int error
)
3680 IoOperationType optype
;
3682 optype
= is_read
? IO_OPERATION_TYPE_READ
: IO_OPERATION_TYPE_WRITE
;
3683 qapi_event_send_block_io_error(bdrv_get_device_name(bs
), optype
, action
,
3684 bdrv_iostatus_is_enabled(bs
),
3685 error
== ENOSPC
, strerror(error
),
3689 /* This is done by device models because, while the block layer knows
3690 * about the error, it does not know whether an operation comes from
3691 * the device or the block layer (from a job, for example).
3693 void bdrv_error_action(BlockDriverState
*bs
, BlockErrorAction action
,
3694 bool is_read
, int error
)
3698 if (action
== BLOCK_ERROR_ACTION_STOP
) {
3699 /* First set the iostatus, so that "info block" returns an iostatus
3700 * that matches the events raised so far (an additional error iostatus
3701 * is fine, but not a lost one).
3703 bdrv_iostatus_set_err(bs
, error
);
3705 /* Then raise the request to stop the VM and the event.
3706 * qemu_system_vmstop_request_prepare has two effects. First,
3707 * it ensures that the STOP event always comes after the
3708 * BLOCK_IO_ERROR event. Second, it ensures that even if management
3709 * can observe the STOP event and do a "cont" before the STOP
3710 * event is issued, the VM will not stop. In this case, vm_start()
3711 * also ensures that the STOP/RESUME pair of events is emitted.
3713 qemu_system_vmstop_request_prepare();
3714 send_qmp_error_event(bs
, action
, is_read
, error
);
3715 qemu_system_vmstop_request(RUN_STATE_IO_ERROR
);
3717 send_qmp_error_event(bs
, action
, is_read
, error
);
3721 int bdrv_is_read_only(BlockDriverState
*bs
)
3723 return bs
->read_only
;
3726 int bdrv_is_sg(BlockDriverState
*bs
)
3731 int bdrv_enable_write_cache(BlockDriverState
*bs
)
3733 return bs
->enable_write_cache
;
3736 void bdrv_set_enable_write_cache(BlockDriverState
*bs
, bool wce
)
3738 bs
->enable_write_cache
= wce
;
3740 /* so a reopen() will preserve wce */
3742 bs
->open_flags
|= BDRV_O_CACHE_WB
;
3744 bs
->open_flags
&= ~BDRV_O_CACHE_WB
;
3748 int bdrv_is_encrypted(BlockDriverState
*bs
)
3750 if (bs
->backing_hd
&& bs
->backing_hd
->encrypted
)
3752 return bs
->encrypted
;
3755 int bdrv_key_required(BlockDriverState
*bs
)
3757 BlockDriverState
*backing_hd
= bs
->backing_hd
;
3759 if (backing_hd
&& backing_hd
->encrypted
&& !backing_hd
->valid_key
)
3761 return (bs
->encrypted
&& !bs
->valid_key
);
3764 int bdrv_set_key(BlockDriverState
*bs
, const char *key
)
3767 if (bs
->backing_hd
&& bs
->backing_hd
->encrypted
) {
3768 ret
= bdrv_set_key(bs
->backing_hd
, key
);
3774 if (!bs
->encrypted
) {
3776 } else if (!bs
->drv
|| !bs
->drv
->bdrv_set_key
) {
3779 ret
= bs
->drv
->bdrv_set_key(bs
, key
);
3782 } else if (!bs
->valid_key
) {
3785 /* call the change callback now, we skipped it on open */
3786 blk_dev_change_media_cb(bs
->blk
, true);
3793 * Provide an encryption key for @bs.
3794 * If @key is non-null:
3795 * If @bs is not encrypted, fail.
3796 * Else if the key is invalid, fail.
3797 * Else set @bs's key to @key, replacing the existing key, if any.
3799 * If @bs is encrypted and still lacks a key, fail.
3801 * On failure, store an error object through @errp if non-null.
3803 void bdrv_add_key(BlockDriverState
*bs
, const char *key
, Error
**errp
)
3806 if (!bdrv_is_encrypted(bs
)) {
3807 error_setg(errp
, "Device '%s' is not encrypted",
3808 bdrv_get_device_name(bs
));
3809 } else if (bdrv_set_key(bs
, key
) < 0) {
3810 error_set(errp
, QERR_INVALID_PASSWORD
);
3813 if (bdrv_key_required(bs
)) {
3814 error_set(errp
, ERROR_CLASS_DEVICE_ENCRYPTED
,
3815 "'%s' (%s) is encrypted",
3816 bdrv_get_device_name(bs
),
3817 bdrv_get_encrypted_filename(bs
));
3822 const char *bdrv_get_format_name(BlockDriverState
*bs
)
3824 return bs
->drv
? bs
->drv
->format_name
: NULL
;
3827 static int qsort_strcmp(const void *a
, const void *b
)
3829 return strcmp(a
, b
);
3832 void bdrv_iterate_format(void (*it
)(void *opaque
, const char *name
),
3838 const char **formats
= NULL
;
3840 QLIST_FOREACH(drv
, &bdrv_drivers
, list
) {
3841 if (drv
->format_name
) {
3844 while (formats
&& i
&& !found
) {
3845 found
= !strcmp(formats
[--i
], drv
->format_name
);
3849 formats
= g_renew(const char *, formats
, count
+ 1);
3850 formats
[count
++] = drv
->format_name
;
3855 qsort(formats
, count
, sizeof(formats
[0]), qsort_strcmp
);
3857 for (i
= 0; i
< count
; i
++) {
3858 it(opaque
, formats
[i
]);
3864 /* This function is to find a node in the bs graph */
3865 BlockDriverState
*bdrv_find_node(const char *node_name
)
3867 BlockDriverState
*bs
;
3871 QTAILQ_FOREACH(bs
, &graph_bdrv_states
, node_list
) {
3872 if (!strcmp(node_name
, bs
->node_name
)) {
3879 /* Put this QMP function here so it can access the static graph_bdrv_states. */
3880 BlockDeviceInfoList
*bdrv_named_nodes_list(void)
3882 BlockDeviceInfoList
*list
, *entry
;
3883 BlockDriverState
*bs
;
3886 QTAILQ_FOREACH(bs
, &graph_bdrv_states
, node_list
) {
3887 entry
= g_malloc0(sizeof(*entry
));
3888 entry
->value
= bdrv_block_device_info(bs
);
3896 BlockDriverState
*bdrv_lookup_bs(const char *device
,
3897 const char *node_name
,
3901 BlockDriverState
*bs
;
3904 blk
= blk_by_name(device
);
3912 bs
= bdrv_find_node(node_name
);
3919 error_setg(errp
, "Cannot find device=%s nor node_name=%s",
3920 device
? device
: "",
3921 node_name
? node_name
: "");
3925 /* If 'base' is in the same chain as 'top', return true. Otherwise,
3926 * return false. If either argument is NULL, return false. */
3927 bool bdrv_chain_contains(BlockDriverState
*top
, BlockDriverState
*base
)
3929 while (top
&& top
!= base
) {
3930 top
= top
->backing_hd
;
3936 BlockDriverState
*bdrv_next_node(BlockDriverState
*bs
)
3939 return QTAILQ_FIRST(&graph_bdrv_states
);
3941 return QTAILQ_NEXT(bs
, node_list
);
3944 BlockDriverState
*bdrv_next(BlockDriverState
*bs
)
3947 return QTAILQ_FIRST(&bdrv_states
);
3949 return QTAILQ_NEXT(bs
, device_list
);
3952 const char *bdrv_get_node_name(const BlockDriverState
*bs
)
3954 return bs
->node_name
;
3957 /* TODO check what callers really want: bs->node_name or blk_name() */
3958 const char *bdrv_get_device_name(const BlockDriverState
*bs
)
3960 return bs
->blk
? blk_name(bs
->blk
) : "";
3963 int bdrv_get_flags(BlockDriverState
*bs
)
3965 return bs
->open_flags
;
3968 int bdrv_flush_all(void)
3970 BlockDriverState
*bs
;
3973 QTAILQ_FOREACH(bs
, &bdrv_states
, device_list
) {
3974 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
3977 aio_context_acquire(aio_context
);
3978 ret
= bdrv_flush(bs
);
3979 if (ret
< 0 && !result
) {
3982 aio_context_release(aio_context
);
3988 int bdrv_has_zero_init_1(BlockDriverState
*bs
)
3993 int bdrv_has_zero_init(BlockDriverState
*bs
)
3997 /* If BS is a copy on write image, it is initialized to
3998 the contents of the base image, which may not be zeroes. */
3999 if (bs
->backing_hd
) {
4002 if (bs
->drv
->bdrv_has_zero_init
) {
4003 return bs
->drv
->bdrv_has_zero_init(bs
);
4010 bool bdrv_unallocated_blocks_are_zero(BlockDriverState
*bs
)
4012 BlockDriverInfo bdi
;
4014 if (bs
->backing_hd
) {
4018 if (bdrv_get_info(bs
, &bdi
) == 0) {
4019 return bdi
.unallocated_blocks_are_zero
;
4025 bool bdrv_can_write_zeroes_with_unmap(BlockDriverState
*bs
)
4027 BlockDriverInfo bdi
;
4029 if (bs
->backing_hd
|| !(bs
->open_flags
& BDRV_O_UNMAP
)) {
4033 if (bdrv_get_info(bs
, &bdi
) == 0) {
4034 return bdi
.can_write_zeroes_with_unmap
;
4040 typedef struct BdrvCoGetBlockStatusData
{
4041 BlockDriverState
*bs
;
4042 BlockDriverState
*base
;
4048 } BdrvCoGetBlockStatusData
;
4051 * Returns the allocation status of the specified sectors.
4052 * Drivers not implementing the functionality are assumed to not support
4053 * backing files, hence all their sectors are reported as allocated.
4055 * If 'sector_num' is beyond the end of the disk image the return value is 0
4056 * and 'pnum' is set to 0.
4058 * 'pnum' is set to the number of sectors (including and immediately following
4059 * the specified sector) that are known to be in the same
4060 * allocated/unallocated state.
4062 * 'nb_sectors' is the max value 'pnum' should be set to. If nb_sectors goes
4063 * beyond the end of the disk image it will be clamped.
4065 static int64_t coroutine_fn
bdrv_co_get_block_status(BlockDriverState
*bs
,
4067 int nb_sectors
, int *pnum
)
4069 int64_t total_sectors
;
4073 total_sectors
= bdrv_nb_sectors(bs
);
4074 if (total_sectors
< 0) {
4075 return total_sectors
;
4078 if (sector_num
>= total_sectors
) {
4083 n
= total_sectors
- sector_num
;
4084 if (n
< nb_sectors
) {
4088 if (!bs
->drv
->bdrv_co_get_block_status
) {
4090 ret
= BDRV_BLOCK_DATA
| BDRV_BLOCK_ALLOCATED
;
4091 if (bs
->drv
->protocol_name
) {
4092 ret
|= BDRV_BLOCK_OFFSET_VALID
| (sector_num
* BDRV_SECTOR_SIZE
);
4097 ret
= bs
->drv
->bdrv_co_get_block_status(bs
, sector_num
, nb_sectors
, pnum
);
4103 if (ret
& BDRV_BLOCK_RAW
) {
4104 assert(ret
& BDRV_BLOCK_OFFSET_VALID
);
4105 return bdrv_get_block_status(bs
->file
, ret
>> BDRV_SECTOR_BITS
,
4109 if (ret
& (BDRV_BLOCK_DATA
| BDRV_BLOCK_ZERO
)) {
4110 ret
|= BDRV_BLOCK_ALLOCATED
;
4113 if (!(ret
& BDRV_BLOCK_DATA
) && !(ret
& BDRV_BLOCK_ZERO
)) {
4114 if (bdrv_unallocated_blocks_are_zero(bs
)) {
4115 ret
|= BDRV_BLOCK_ZERO
;
4116 } else if (bs
->backing_hd
) {
4117 BlockDriverState
*bs2
= bs
->backing_hd
;
4118 int64_t nb_sectors2
= bdrv_nb_sectors(bs2
);
4119 if (nb_sectors2
>= 0 && sector_num
>= nb_sectors2
) {
4120 ret
|= BDRV_BLOCK_ZERO
;
4126 (ret
& BDRV_BLOCK_DATA
) && !(ret
& BDRV_BLOCK_ZERO
) &&
4127 (ret
& BDRV_BLOCK_OFFSET_VALID
)) {
4130 ret2
= bdrv_co_get_block_status(bs
->file
, ret
>> BDRV_SECTOR_BITS
,
4133 /* Ignore errors. This is just providing extra information, it
4134 * is useful but not necessary.
4137 /* !file_pnum indicates an offset at or beyond the EOF; it is
4138 * perfectly valid for the format block driver to point to such
4139 * offsets, so catch it and mark everything as zero */
4140 ret
|= BDRV_BLOCK_ZERO
;
4142 /* Limit request to the range reported by the protocol driver */
4144 ret
|= (ret2
& BDRV_BLOCK_ZERO
);
4152 /* Coroutine wrapper for bdrv_get_block_status() */
4153 static void coroutine_fn
bdrv_get_block_status_co_entry(void *opaque
)
4155 BdrvCoGetBlockStatusData
*data
= opaque
;
4156 BlockDriverState
*bs
= data
->bs
;
4158 data
->ret
= bdrv_co_get_block_status(bs
, data
->sector_num
, data
->nb_sectors
,
4164 * Synchronous wrapper around bdrv_co_get_block_status().
4166 * See bdrv_co_get_block_status() for details.
4168 int64_t bdrv_get_block_status(BlockDriverState
*bs
, int64_t sector_num
,
4169 int nb_sectors
, int *pnum
)
4172 BdrvCoGetBlockStatusData data
= {
4174 .sector_num
= sector_num
,
4175 .nb_sectors
= nb_sectors
,
4180 if (qemu_in_coroutine()) {
4181 /* Fast-path if already in coroutine context */
4182 bdrv_get_block_status_co_entry(&data
);
4184 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
4186 co
= qemu_coroutine_create(bdrv_get_block_status_co_entry
);
4187 qemu_coroutine_enter(co
, &data
);
4188 while (!data
.done
) {
4189 aio_poll(aio_context
, true);
4195 int coroutine_fn
bdrv_is_allocated(BlockDriverState
*bs
, int64_t sector_num
,
4196 int nb_sectors
, int *pnum
)
4198 int64_t ret
= bdrv_get_block_status(bs
, sector_num
, nb_sectors
, pnum
);
4202 return !!(ret
& BDRV_BLOCK_ALLOCATED
);
4206 * Given an image chain: ... -> [BASE] -> [INTER1] -> [INTER2] -> [TOP]
4208 * Return true if the given sector is allocated in any image between
4209 * BASE and TOP (inclusive). BASE can be NULL to check if the given
4210 * sector is allocated in any image of the chain. Return false otherwise.
4212 * 'pnum' is set to the number of sectors (including and immediately following
4213 * the specified sector) that are known to be in the same
4214 * allocated/unallocated state.
4217 int bdrv_is_allocated_above(BlockDriverState
*top
,
4218 BlockDriverState
*base
,
4220 int nb_sectors
, int *pnum
)
4222 BlockDriverState
*intermediate
;
4223 int ret
, n
= nb_sectors
;
4226 while (intermediate
&& intermediate
!= base
) {
4228 ret
= bdrv_is_allocated(intermediate
, sector_num
, nb_sectors
,
4238 * [sector_num, nb_sectors] is unallocated on top but intermediate
4241 * [sector_num+x, nr_sectors] allocated.
4243 if (n
> pnum_inter
&&
4244 (intermediate
== top
||
4245 sector_num
+ pnum_inter
< intermediate
->total_sectors
)) {
4249 intermediate
= intermediate
->backing_hd
;
4256 const char *bdrv_get_encrypted_filename(BlockDriverState
*bs
)
4258 if (bs
->backing_hd
&& bs
->backing_hd
->encrypted
)
4259 return bs
->backing_file
;
4260 else if (bs
->encrypted
)
4261 return bs
->filename
;
4266 void bdrv_get_backing_filename(BlockDriverState
*bs
,
4267 char *filename
, int filename_size
)
4269 pstrcpy(filename
, filename_size
, bs
->backing_file
);
4272 int bdrv_write_compressed(BlockDriverState
*bs
, int64_t sector_num
,
4273 const uint8_t *buf
, int nb_sectors
)
4275 BlockDriver
*drv
= bs
->drv
;
4281 if (!drv
->bdrv_write_compressed
) {
4284 ret
= bdrv_check_request(bs
, sector_num
, nb_sectors
);
4289 assert(QLIST_EMPTY(&bs
->dirty_bitmaps
));
4291 return drv
->bdrv_write_compressed(bs
, sector_num
, buf
, nb_sectors
);
4294 int bdrv_get_info(BlockDriverState
*bs
, BlockDriverInfo
*bdi
)
4296 BlockDriver
*drv
= bs
->drv
;
4299 if (!drv
->bdrv_get_info
)
4301 memset(bdi
, 0, sizeof(*bdi
));
4302 return drv
->bdrv_get_info(bs
, bdi
);
4305 ImageInfoSpecific
*bdrv_get_specific_info(BlockDriverState
*bs
)
4307 BlockDriver
*drv
= bs
->drv
;
4308 if (drv
&& drv
->bdrv_get_specific_info
) {
4309 return drv
->bdrv_get_specific_info(bs
);
4314 int bdrv_save_vmstate(BlockDriverState
*bs
, const uint8_t *buf
,
4315 int64_t pos
, int size
)
4318 struct iovec iov
= {
4319 .iov_base
= (void *) buf
,
4323 qemu_iovec_init_external(&qiov
, &iov
, 1);
4324 return bdrv_writev_vmstate(bs
, &qiov
, pos
);
4327 int bdrv_writev_vmstate(BlockDriverState
*bs
, QEMUIOVector
*qiov
, int64_t pos
)
4329 BlockDriver
*drv
= bs
->drv
;
4333 } else if (drv
->bdrv_save_vmstate
) {
4334 return drv
->bdrv_save_vmstate(bs
, qiov
, pos
);
4335 } else if (bs
->file
) {
4336 return bdrv_writev_vmstate(bs
->file
, qiov
, pos
);
4342 int bdrv_load_vmstate(BlockDriverState
*bs
, uint8_t *buf
,
4343 int64_t pos
, int size
)
4345 BlockDriver
*drv
= bs
->drv
;
4348 if (drv
->bdrv_load_vmstate
)
4349 return drv
->bdrv_load_vmstate(bs
, buf
, pos
, size
);
4351 return bdrv_load_vmstate(bs
->file
, buf
, pos
, size
);
4355 void bdrv_debug_event(BlockDriverState
*bs
, BlkDebugEvent event
)
4357 if (!bs
|| !bs
->drv
|| !bs
->drv
->bdrv_debug_event
) {
4361 bs
->drv
->bdrv_debug_event(bs
, event
);
4364 int bdrv_debug_breakpoint(BlockDriverState
*bs
, const char *event
,
4367 while (bs
&& bs
->drv
&& !bs
->drv
->bdrv_debug_breakpoint
) {
4371 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_breakpoint
) {
4372 return bs
->drv
->bdrv_debug_breakpoint(bs
, event
, tag
);
4378 int bdrv_debug_remove_breakpoint(BlockDriverState
*bs
, const char *tag
)
4380 while (bs
&& bs
->drv
&& !bs
->drv
->bdrv_debug_remove_breakpoint
) {
4384 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_remove_breakpoint
) {
4385 return bs
->drv
->bdrv_debug_remove_breakpoint(bs
, tag
);
4391 int bdrv_debug_resume(BlockDriverState
*bs
, const char *tag
)
4393 while (bs
&& (!bs
->drv
|| !bs
->drv
->bdrv_debug_resume
)) {
4397 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_resume
) {
4398 return bs
->drv
->bdrv_debug_resume(bs
, tag
);
4404 bool bdrv_debug_is_suspended(BlockDriverState
*bs
, const char *tag
)
4406 while (bs
&& bs
->drv
&& !bs
->drv
->bdrv_debug_is_suspended
) {
4410 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_is_suspended
) {
4411 return bs
->drv
->bdrv_debug_is_suspended(bs
, tag
);
4417 int bdrv_is_snapshot(BlockDriverState
*bs
)
4419 return !!(bs
->open_flags
& BDRV_O_SNAPSHOT
);
4422 /* backing_file can either be relative, or absolute, or a protocol. If it is
4423 * relative, it must be relative to the chain. So, passing in bs->filename
4424 * from a BDS as backing_file should not be done, as that may be relative to
4425 * the CWD rather than the chain. */
4426 BlockDriverState
*bdrv_find_backing_image(BlockDriverState
*bs
,
4427 const char *backing_file
)
4429 char *filename_full
= NULL
;
4430 char *backing_file_full
= NULL
;
4431 char *filename_tmp
= NULL
;
4432 int is_protocol
= 0;
4433 BlockDriverState
*curr_bs
= NULL
;
4434 BlockDriverState
*retval
= NULL
;
4436 if (!bs
|| !bs
->drv
|| !backing_file
) {
4440 filename_full
= g_malloc(PATH_MAX
);
4441 backing_file_full
= g_malloc(PATH_MAX
);
4442 filename_tmp
= g_malloc(PATH_MAX
);
4444 is_protocol
= path_has_protocol(backing_file
);
4446 for (curr_bs
= bs
; curr_bs
->backing_hd
; curr_bs
= curr_bs
->backing_hd
) {
4448 /* If either of the filename paths is actually a protocol, then
4449 * compare unmodified paths; otherwise make paths relative */
4450 if (is_protocol
|| path_has_protocol(curr_bs
->backing_file
)) {
4451 if (strcmp(backing_file
, curr_bs
->backing_file
) == 0) {
4452 retval
= curr_bs
->backing_hd
;
4456 /* If not an absolute filename path, make it relative to the current
4457 * image's filename path */
4458 path_combine(filename_tmp
, PATH_MAX
, curr_bs
->filename
,
4461 /* We are going to compare absolute pathnames */
4462 if (!realpath(filename_tmp
, filename_full
)) {
4466 /* We need to make sure the backing filename we are comparing against
4467 * is relative to the current image filename (or absolute) */
4468 path_combine(filename_tmp
, PATH_MAX
, curr_bs
->filename
,
4469 curr_bs
->backing_file
);
4471 if (!realpath(filename_tmp
, backing_file_full
)) {
4475 if (strcmp(backing_file_full
, filename_full
) == 0) {
4476 retval
= curr_bs
->backing_hd
;
4482 g_free(filename_full
);
4483 g_free(backing_file_full
);
4484 g_free(filename_tmp
);
4488 int bdrv_get_backing_file_depth(BlockDriverState
*bs
)
4494 if (!bs
->backing_hd
) {
4498 return 1 + bdrv_get_backing_file_depth(bs
->backing_hd
);
4501 /**************************************************************/
4504 BlockAIOCB
*bdrv_aio_readv(BlockDriverState
*bs
, int64_t sector_num
,
4505 QEMUIOVector
*qiov
, int nb_sectors
,
4506 BlockCompletionFunc
*cb
, void *opaque
)
4508 trace_bdrv_aio_readv(bs
, sector_num
, nb_sectors
, opaque
);
4510 return bdrv_co_aio_rw_vector(bs
, sector_num
, qiov
, nb_sectors
, 0,
4514 BlockAIOCB
*bdrv_aio_writev(BlockDriverState
*bs
, int64_t sector_num
,
4515 QEMUIOVector
*qiov
, int nb_sectors
,
4516 BlockCompletionFunc
*cb
, void *opaque
)
4518 trace_bdrv_aio_writev(bs
, sector_num
, nb_sectors
, opaque
);
4520 return bdrv_co_aio_rw_vector(bs
, sector_num
, qiov
, nb_sectors
, 0,
4524 BlockAIOCB
*bdrv_aio_write_zeroes(BlockDriverState
*bs
,
4525 int64_t sector_num
, int nb_sectors
, BdrvRequestFlags flags
,
4526 BlockCompletionFunc
*cb
, void *opaque
)
4528 trace_bdrv_aio_write_zeroes(bs
, sector_num
, nb_sectors
, flags
, opaque
);
4530 return bdrv_co_aio_rw_vector(bs
, sector_num
, NULL
, nb_sectors
,
4531 BDRV_REQ_ZERO_WRITE
| flags
,
4536 typedef struct MultiwriteCB
{
4541 BlockCompletionFunc
*cb
;
4543 QEMUIOVector
*free_qiov
;
4547 static void multiwrite_user_cb(MultiwriteCB
*mcb
)
4551 for (i
= 0; i
< mcb
->num_callbacks
; i
++) {
4552 mcb
->callbacks
[i
].cb(mcb
->callbacks
[i
].opaque
, mcb
->error
);
4553 if (mcb
->callbacks
[i
].free_qiov
) {
4554 qemu_iovec_destroy(mcb
->callbacks
[i
].free_qiov
);
4556 g_free(mcb
->callbacks
[i
].free_qiov
);
4560 static void multiwrite_cb(void *opaque
, int ret
)
4562 MultiwriteCB
*mcb
= opaque
;
4564 trace_multiwrite_cb(mcb
, ret
);
4566 if (ret
< 0 && !mcb
->error
) {
4570 mcb
->num_requests
--;
4571 if (mcb
->num_requests
== 0) {
4572 multiwrite_user_cb(mcb
);
4577 static int multiwrite_req_compare(const void *a
, const void *b
)
4579 const BlockRequest
*req1
= a
, *req2
= b
;
4582 * Note that we can't simply subtract req2->sector from req1->sector
4583 * here as that could overflow the return value.
4585 if (req1
->sector
> req2
->sector
) {
4587 } else if (req1
->sector
< req2
->sector
) {
4595 * Takes a bunch of requests and tries to merge them. Returns the number of
4596 * requests that remain after merging.
4598 static int multiwrite_merge(BlockDriverState
*bs
, BlockRequest
*reqs
,
4599 int num_reqs
, MultiwriteCB
*mcb
)
4603 // Sort requests by start sector
4604 qsort(reqs
, num_reqs
, sizeof(*reqs
), &multiwrite_req_compare
);
4606 // Check if adjacent requests touch the same clusters. If so, combine them,
4607 // filling up gaps with zero sectors.
4609 for (i
= 1; i
< num_reqs
; i
++) {
4611 int64_t oldreq_last
= reqs
[outidx
].sector
+ reqs
[outidx
].nb_sectors
;
4613 // Handle exactly sequential writes and overlapping writes.
4614 if (reqs
[i
].sector
<= oldreq_last
) {
4618 if (reqs
[outidx
].qiov
->niov
+ reqs
[i
].qiov
->niov
+ 1 > IOV_MAX
) {
4622 if (bs
->bl
.max_transfer_length
&& reqs
[outidx
].nb_sectors
+
4623 reqs
[i
].nb_sectors
> bs
->bl
.max_transfer_length
) {
4629 QEMUIOVector
*qiov
= g_malloc0(sizeof(*qiov
));
4630 qemu_iovec_init(qiov
,
4631 reqs
[outidx
].qiov
->niov
+ reqs
[i
].qiov
->niov
+ 1);
4633 // Add the first request to the merged one. If the requests are
4634 // overlapping, drop the last sectors of the first request.
4635 size
= (reqs
[i
].sector
- reqs
[outidx
].sector
) << 9;
4636 qemu_iovec_concat(qiov
, reqs
[outidx
].qiov
, 0, size
);
4638 // We should need to add any zeros between the two requests
4639 assert (reqs
[i
].sector
<= oldreq_last
);
4641 // Add the second request
4642 qemu_iovec_concat(qiov
, reqs
[i
].qiov
, 0, reqs
[i
].qiov
->size
);
4644 // Add tail of first request, if necessary
4645 if (qiov
->size
< reqs
[outidx
].qiov
->size
) {
4646 qemu_iovec_concat(qiov
, reqs
[outidx
].qiov
, qiov
->size
,
4647 reqs
[outidx
].qiov
->size
- qiov
->size
);
4650 reqs
[outidx
].nb_sectors
= qiov
->size
>> 9;
4651 reqs
[outidx
].qiov
= qiov
;
4653 mcb
->callbacks
[i
].free_qiov
= reqs
[outidx
].qiov
;
4656 reqs
[outidx
].sector
= reqs
[i
].sector
;
4657 reqs
[outidx
].nb_sectors
= reqs
[i
].nb_sectors
;
4658 reqs
[outidx
].qiov
= reqs
[i
].qiov
;
4662 block_acct_merge_done(&bs
->stats
, BLOCK_ACCT_WRITE
, num_reqs
- outidx
- 1);
4668 * Submit multiple AIO write requests at once.
4670 * On success, the function returns 0 and all requests in the reqs array have
4671 * been submitted. In error case this function returns -1, and any of the
4672 * requests may or may not be submitted yet. In particular, this means that the
4673 * callback will be called for some of the requests, for others it won't. The
4674 * caller must check the error field of the BlockRequest to wait for the right
4675 * callbacks (if error != 0, no callback will be called).
4677 * The implementation may modify the contents of the reqs array, e.g. to merge
4678 * requests. However, the fields opaque and error are left unmodified as they
4679 * are used to signal failure for a single request to the caller.
4681 int bdrv_aio_multiwrite(BlockDriverState
*bs
, BlockRequest
*reqs
, int num_reqs
)
4686 /* don't submit writes if we don't have a medium */
4687 if (bs
->drv
== NULL
) {
4688 for (i
= 0; i
< num_reqs
; i
++) {
4689 reqs
[i
].error
= -ENOMEDIUM
;
4694 if (num_reqs
== 0) {
4698 // Create MultiwriteCB structure
4699 mcb
= g_malloc0(sizeof(*mcb
) + num_reqs
* sizeof(*mcb
->callbacks
));
4700 mcb
->num_requests
= 0;
4701 mcb
->num_callbacks
= num_reqs
;
4703 for (i
= 0; i
< num_reqs
; i
++) {
4704 mcb
->callbacks
[i
].cb
= reqs
[i
].cb
;
4705 mcb
->callbacks
[i
].opaque
= reqs
[i
].opaque
;
4708 // Check for mergable requests
4709 num_reqs
= multiwrite_merge(bs
, reqs
, num_reqs
, mcb
);
4711 trace_bdrv_aio_multiwrite(mcb
, mcb
->num_callbacks
, num_reqs
);
4713 /* Run the aio requests. */
4714 mcb
->num_requests
= num_reqs
;
4715 for (i
= 0; i
< num_reqs
; i
++) {
4716 bdrv_co_aio_rw_vector(bs
, reqs
[i
].sector
, reqs
[i
].qiov
,
4717 reqs
[i
].nb_sectors
, reqs
[i
].flags
,
4725 void bdrv_aio_cancel(BlockAIOCB
*acb
)
4728 bdrv_aio_cancel_async(acb
);
4729 while (acb
->refcnt
> 1) {
4730 if (acb
->aiocb_info
->get_aio_context
) {
4731 aio_poll(acb
->aiocb_info
->get_aio_context(acb
), true);
4732 } else if (acb
->bs
) {
4733 aio_poll(bdrv_get_aio_context(acb
->bs
), true);
4738 qemu_aio_unref(acb
);
4741 /* Async version of aio cancel. The caller is not blocked if the acb implements
4742 * cancel_async, otherwise we do nothing and let the request normally complete.
4743 * In either case the completion callback must be called. */
4744 void bdrv_aio_cancel_async(BlockAIOCB
*acb
)
4746 if (acb
->aiocb_info
->cancel_async
) {
4747 acb
->aiocb_info
->cancel_async(acb
);
4751 /**************************************************************/
4752 /* async block device emulation */
4754 typedef struct BlockAIOCBSync
{
4758 /* vector translation state */
4764 static const AIOCBInfo bdrv_em_aiocb_info
= {
4765 .aiocb_size
= sizeof(BlockAIOCBSync
),
4768 static void bdrv_aio_bh_cb(void *opaque
)
4770 BlockAIOCBSync
*acb
= opaque
;
4772 if (!acb
->is_write
&& acb
->ret
>= 0) {
4773 qemu_iovec_from_buf(acb
->qiov
, 0, acb
->bounce
, acb
->qiov
->size
);
4775 qemu_vfree(acb
->bounce
);
4776 acb
->common
.cb(acb
->common
.opaque
, acb
->ret
);
4777 qemu_bh_delete(acb
->bh
);
4779 qemu_aio_unref(acb
);
4782 static BlockAIOCB
*bdrv_aio_rw_vector(BlockDriverState
*bs
,
4786 BlockCompletionFunc
*cb
,
4791 BlockAIOCBSync
*acb
;
4793 acb
= qemu_aio_get(&bdrv_em_aiocb_info
, bs
, cb
, opaque
);
4794 acb
->is_write
= is_write
;
4796 acb
->bounce
= qemu_try_blockalign(bs
, qiov
->size
);
4797 acb
->bh
= aio_bh_new(bdrv_get_aio_context(bs
), bdrv_aio_bh_cb
, acb
);
4799 if (acb
->bounce
== NULL
) {
4801 } else if (is_write
) {
4802 qemu_iovec_to_buf(acb
->qiov
, 0, acb
->bounce
, qiov
->size
);
4803 acb
->ret
= bs
->drv
->bdrv_write(bs
, sector_num
, acb
->bounce
, nb_sectors
);
4805 acb
->ret
= bs
->drv
->bdrv_read(bs
, sector_num
, acb
->bounce
, nb_sectors
);
4808 qemu_bh_schedule(acb
->bh
);
4810 return &acb
->common
;
4813 static BlockAIOCB
*bdrv_aio_readv_em(BlockDriverState
*bs
,
4814 int64_t sector_num
, QEMUIOVector
*qiov
, int nb_sectors
,
4815 BlockCompletionFunc
*cb
, void *opaque
)
4817 return bdrv_aio_rw_vector(bs
, sector_num
, qiov
, nb_sectors
, cb
, opaque
, 0);
4820 static BlockAIOCB
*bdrv_aio_writev_em(BlockDriverState
*bs
,
4821 int64_t sector_num
, QEMUIOVector
*qiov
, int nb_sectors
,
4822 BlockCompletionFunc
*cb
, void *opaque
)
4824 return bdrv_aio_rw_vector(bs
, sector_num
, qiov
, nb_sectors
, cb
, opaque
, 1);
4828 typedef struct BlockAIOCBCoroutine
{
4834 } BlockAIOCBCoroutine
;
4836 static const AIOCBInfo bdrv_em_co_aiocb_info
= {
4837 .aiocb_size
= sizeof(BlockAIOCBCoroutine
),
4840 static void bdrv_co_em_bh(void *opaque
)
4842 BlockAIOCBCoroutine
*acb
= opaque
;
4844 acb
->common
.cb(acb
->common
.opaque
, acb
->req
.error
);
4846 qemu_bh_delete(acb
->bh
);
4847 qemu_aio_unref(acb
);
4850 /* Invoke bdrv_co_do_readv/bdrv_co_do_writev */
4851 static void coroutine_fn
bdrv_co_do_rw(void *opaque
)
4853 BlockAIOCBCoroutine
*acb
= opaque
;
4854 BlockDriverState
*bs
= acb
->common
.bs
;
4856 if (!acb
->is_write
) {
4857 acb
->req
.error
= bdrv_co_do_readv(bs
, acb
->req
.sector
,
4858 acb
->req
.nb_sectors
, acb
->req
.qiov
, acb
->req
.flags
);
4860 acb
->req
.error
= bdrv_co_do_writev(bs
, acb
->req
.sector
,
4861 acb
->req
.nb_sectors
, acb
->req
.qiov
, acb
->req
.flags
);
4864 acb
->bh
= aio_bh_new(bdrv_get_aio_context(bs
), bdrv_co_em_bh
, acb
);
4865 qemu_bh_schedule(acb
->bh
);
4868 static BlockAIOCB
*bdrv_co_aio_rw_vector(BlockDriverState
*bs
,
4872 BdrvRequestFlags flags
,
4873 BlockCompletionFunc
*cb
,
4878 BlockAIOCBCoroutine
*acb
;
4880 acb
= qemu_aio_get(&bdrv_em_co_aiocb_info
, bs
, cb
, opaque
);
4881 acb
->req
.sector
= sector_num
;
4882 acb
->req
.nb_sectors
= nb_sectors
;
4883 acb
->req
.qiov
= qiov
;
4884 acb
->req
.flags
= flags
;
4885 acb
->is_write
= is_write
;
4887 co
= qemu_coroutine_create(bdrv_co_do_rw
);
4888 qemu_coroutine_enter(co
, acb
);
4890 return &acb
->common
;
4893 static void coroutine_fn
bdrv_aio_flush_co_entry(void *opaque
)
4895 BlockAIOCBCoroutine
*acb
= opaque
;
4896 BlockDriverState
*bs
= acb
->common
.bs
;
4898 acb
->req
.error
= bdrv_co_flush(bs
);
4899 acb
->bh
= aio_bh_new(bdrv_get_aio_context(bs
), bdrv_co_em_bh
, acb
);
4900 qemu_bh_schedule(acb
->bh
);
4903 BlockAIOCB
*bdrv_aio_flush(BlockDriverState
*bs
,
4904 BlockCompletionFunc
*cb
, void *opaque
)
4906 trace_bdrv_aio_flush(bs
, opaque
);
4909 BlockAIOCBCoroutine
*acb
;
4911 acb
= qemu_aio_get(&bdrv_em_co_aiocb_info
, bs
, cb
, opaque
);
4913 co
= qemu_coroutine_create(bdrv_aio_flush_co_entry
);
4914 qemu_coroutine_enter(co
, acb
);
4916 return &acb
->common
;
4919 static void coroutine_fn
bdrv_aio_discard_co_entry(void *opaque
)
4921 BlockAIOCBCoroutine
*acb
= opaque
;
4922 BlockDriverState
*bs
= acb
->common
.bs
;
4924 acb
->req
.error
= bdrv_co_discard(bs
, acb
->req
.sector
, acb
->req
.nb_sectors
);
4925 acb
->bh
= aio_bh_new(bdrv_get_aio_context(bs
), bdrv_co_em_bh
, acb
);
4926 qemu_bh_schedule(acb
->bh
);
4929 BlockAIOCB
*bdrv_aio_discard(BlockDriverState
*bs
,
4930 int64_t sector_num
, int nb_sectors
,
4931 BlockCompletionFunc
*cb
, void *opaque
)
4934 BlockAIOCBCoroutine
*acb
;
4936 trace_bdrv_aio_discard(bs
, sector_num
, nb_sectors
, opaque
);
4938 acb
= qemu_aio_get(&bdrv_em_co_aiocb_info
, bs
, cb
, opaque
);
4939 acb
->req
.sector
= sector_num
;
4940 acb
->req
.nb_sectors
= nb_sectors
;
4941 co
= qemu_coroutine_create(bdrv_aio_discard_co_entry
);
4942 qemu_coroutine_enter(co
, acb
);
4944 return &acb
->common
;
4947 void bdrv_init(void)
4949 module_call_init(MODULE_INIT_BLOCK
);
4952 void bdrv_init_with_whitelist(void)
4954 use_bdrv_whitelist
= 1;
4958 void *qemu_aio_get(const AIOCBInfo
*aiocb_info
, BlockDriverState
*bs
,
4959 BlockCompletionFunc
*cb
, void *opaque
)
4963 acb
= g_slice_alloc(aiocb_info
->aiocb_size
);
4964 acb
->aiocb_info
= aiocb_info
;
4967 acb
->opaque
= opaque
;
4972 void qemu_aio_ref(void *p
)
4974 BlockAIOCB
*acb
= p
;
4978 void qemu_aio_unref(void *p
)
4980 BlockAIOCB
*acb
= p
;
4981 assert(acb
->refcnt
> 0);
4982 if (--acb
->refcnt
== 0) {
4983 g_slice_free1(acb
->aiocb_info
->aiocb_size
, acb
);
4987 /**************************************************************/
4988 /* Coroutine block device emulation */
4990 typedef struct CoroutineIOCompletion
{
4991 Coroutine
*coroutine
;
4993 } CoroutineIOCompletion
;
4995 static void bdrv_co_io_em_complete(void *opaque
, int ret
)
4997 CoroutineIOCompletion
*co
= opaque
;
5000 qemu_coroutine_enter(co
->coroutine
, NULL
);
5003 static int coroutine_fn
bdrv_co_io_em(BlockDriverState
*bs
, int64_t sector_num
,
5004 int nb_sectors
, QEMUIOVector
*iov
,
5007 CoroutineIOCompletion co
= {
5008 .coroutine
= qemu_coroutine_self(),
5013 acb
= bs
->drv
->bdrv_aio_writev(bs
, sector_num
, iov
, nb_sectors
,
5014 bdrv_co_io_em_complete
, &co
);
5016 acb
= bs
->drv
->bdrv_aio_readv(bs
, sector_num
, iov
, nb_sectors
,
5017 bdrv_co_io_em_complete
, &co
);
5020 trace_bdrv_co_io_em(bs
, sector_num
, nb_sectors
, is_write
, acb
);
5024 qemu_coroutine_yield();
5029 static int coroutine_fn
bdrv_co_readv_em(BlockDriverState
*bs
,
5030 int64_t sector_num
, int nb_sectors
,
5033 return bdrv_co_io_em(bs
, sector_num
, nb_sectors
, iov
, false);
5036 static int coroutine_fn
bdrv_co_writev_em(BlockDriverState
*bs
,
5037 int64_t sector_num
, int nb_sectors
,
5040 return bdrv_co_io_em(bs
, sector_num
, nb_sectors
, iov
, true);
5043 static void coroutine_fn
bdrv_flush_co_entry(void *opaque
)
5045 RwCo
*rwco
= opaque
;
5047 rwco
->ret
= bdrv_co_flush(rwco
->bs
);
5050 int coroutine_fn
bdrv_co_flush(BlockDriverState
*bs
)
5054 if (!bs
|| !bdrv_is_inserted(bs
) || bdrv_is_read_only(bs
)) {
5058 /* Write back cached data to the OS even with cache=unsafe */
5059 BLKDBG_EVENT(bs
->file
, BLKDBG_FLUSH_TO_OS
);
5060 if (bs
->drv
->bdrv_co_flush_to_os
) {
5061 ret
= bs
->drv
->bdrv_co_flush_to_os(bs
);
5067 /* But don't actually force it to the disk with cache=unsafe */
5068 if (bs
->open_flags
& BDRV_O_NO_FLUSH
) {
5072 BLKDBG_EVENT(bs
->file
, BLKDBG_FLUSH_TO_DISK
);
5073 if (bs
->drv
->bdrv_co_flush_to_disk
) {
5074 ret
= bs
->drv
->bdrv_co_flush_to_disk(bs
);
5075 } else if (bs
->drv
->bdrv_aio_flush
) {
5077 CoroutineIOCompletion co
= {
5078 .coroutine
= qemu_coroutine_self(),
5081 acb
= bs
->drv
->bdrv_aio_flush(bs
, bdrv_co_io_em_complete
, &co
);
5085 qemu_coroutine_yield();
5090 * Some block drivers always operate in either writethrough or unsafe
5091 * mode and don't support bdrv_flush therefore. Usually qemu doesn't
5092 * know how the server works (because the behaviour is hardcoded or
5093 * depends on server-side configuration), so we can't ensure that
5094 * everything is safe on disk. Returning an error doesn't work because
5095 * that would break guests even if the server operates in writethrough
5098 * Let's hope the user knows what he's doing.
5106 /* Now flush the underlying protocol. It will also have BDRV_O_NO_FLUSH
5107 * in the case of cache=unsafe, so there are no useless flushes.
5110 return bdrv_co_flush(bs
->file
);
5113 void bdrv_invalidate_cache(BlockDriverState
*bs
, Error
**errp
)
5115 Error
*local_err
= NULL
;
5122 if (!(bs
->open_flags
& BDRV_O_INCOMING
)) {
5125 bs
->open_flags
&= ~BDRV_O_INCOMING
;
5127 if (bs
->drv
->bdrv_invalidate_cache
) {
5128 bs
->drv
->bdrv_invalidate_cache(bs
, &local_err
);
5129 } else if (bs
->file
) {
5130 bdrv_invalidate_cache(bs
->file
, &local_err
);
5133 error_propagate(errp
, local_err
);
5137 ret
= refresh_total_sectors(bs
, bs
->total_sectors
);
5139 error_setg_errno(errp
, -ret
, "Could not refresh total sector count");
5144 void bdrv_invalidate_cache_all(Error
**errp
)
5146 BlockDriverState
*bs
;
5147 Error
*local_err
= NULL
;
5149 QTAILQ_FOREACH(bs
, &bdrv_states
, device_list
) {
5150 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
5152 aio_context_acquire(aio_context
);
5153 bdrv_invalidate_cache(bs
, &local_err
);
5154 aio_context_release(aio_context
);
5156 error_propagate(errp
, local_err
);
5162 int bdrv_flush(BlockDriverState
*bs
)
5170 if (qemu_in_coroutine()) {
5171 /* Fast-path if already in coroutine context */
5172 bdrv_flush_co_entry(&rwco
);
5174 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
5176 co
= qemu_coroutine_create(bdrv_flush_co_entry
);
5177 qemu_coroutine_enter(co
, &rwco
);
5178 while (rwco
.ret
== NOT_DONE
) {
5179 aio_poll(aio_context
, true);
5186 typedef struct DiscardCo
{
5187 BlockDriverState
*bs
;
5192 static void coroutine_fn
bdrv_discard_co_entry(void *opaque
)
5194 DiscardCo
*rwco
= opaque
;
5196 rwco
->ret
= bdrv_co_discard(rwco
->bs
, rwco
->sector_num
, rwco
->nb_sectors
);
5199 int coroutine_fn
bdrv_co_discard(BlockDriverState
*bs
, int64_t sector_num
,
5202 int max_discard
, ret
;
5208 ret
= bdrv_check_request(bs
, sector_num
, nb_sectors
);
5211 } else if (bs
->read_only
) {
5215 bdrv_reset_dirty(bs
, sector_num
, nb_sectors
);
5217 /* Do nothing if disabled. */
5218 if (!(bs
->open_flags
& BDRV_O_UNMAP
)) {
5222 if (!bs
->drv
->bdrv_co_discard
&& !bs
->drv
->bdrv_aio_discard
) {
5226 max_discard
= MIN_NON_ZERO(bs
->bl
.max_discard
, BDRV_REQUEST_MAX_SECTORS
);
5227 while (nb_sectors
> 0) {
5229 int num
= nb_sectors
;
5232 if (bs
->bl
.discard_alignment
&&
5233 num
>= bs
->bl
.discard_alignment
&&
5234 sector_num
% bs
->bl
.discard_alignment
) {
5235 if (num
> bs
->bl
.discard_alignment
) {
5236 num
= bs
->bl
.discard_alignment
;
5238 num
-= sector_num
% bs
->bl
.discard_alignment
;
5241 /* limit request size */
5242 if (num
> max_discard
) {
5246 if (bs
->drv
->bdrv_co_discard
) {
5247 ret
= bs
->drv
->bdrv_co_discard(bs
, sector_num
, num
);
5250 CoroutineIOCompletion co
= {
5251 .coroutine
= qemu_coroutine_self(),
5254 acb
= bs
->drv
->bdrv_aio_discard(bs
, sector_num
, nb_sectors
,
5255 bdrv_co_io_em_complete
, &co
);
5259 qemu_coroutine_yield();
5263 if (ret
&& ret
!= -ENOTSUP
) {
5273 int bdrv_discard(BlockDriverState
*bs
, int64_t sector_num
, int nb_sectors
)
5278 .sector_num
= sector_num
,
5279 .nb_sectors
= nb_sectors
,
5283 if (qemu_in_coroutine()) {
5284 /* Fast-path if already in coroutine context */
5285 bdrv_discard_co_entry(&rwco
);
5287 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
5289 co
= qemu_coroutine_create(bdrv_discard_co_entry
);
5290 qemu_coroutine_enter(co
, &rwco
);
5291 while (rwco
.ret
== NOT_DONE
) {
5292 aio_poll(aio_context
, true);
5299 /**************************************************************/
5300 /* removable device support */
5303 * Return TRUE if the media is present
5305 int bdrv_is_inserted(BlockDriverState
*bs
)
5307 BlockDriver
*drv
= bs
->drv
;
5311 if (!drv
->bdrv_is_inserted
)
5313 return drv
->bdrv_is_inserted(bs
);
5317 * Return whether the media changed since the last call to this
5318 * function, or -ENOTSUP if we don't know. Most drivers don't know.
5320 int bdrv_media_changed(BlockDriverState
*bs
)
5322 BlockDriver
*drv
= bs
->drv
;
5324 if (drv
&& drv
->bdrv_media_changed
) {
5325 return drv
->bdrv_media_changed(bs
);
5331 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
5333 void bdrv_eject(BlockDriverState
*bs
, bool eject_flag
)
5335 BlockDriver
*drv
= bs
->drv
;
5336 const char *device_name
;
5338 if (drv
&& drv
->bdrv_eject
) {
5339 drv
->bdrv_eject(bs
, eject_flag
);
5342 device_name
= bdrv_get_device_name(bs
);
5343 if (device_name
[0] != '\0') {
5344 qapi_event_send_device_tray_moved(device_name
,
5345 eject_flag
, &error_abort
);
5350 * Lock or unlock the media (if it is locked, the user won't be able
5351 * to eject it manually).
5353 void bdrv_lock_medium(BlockDriverState
*bs
, bool locked
)
5355 BlockDriver
*drv
= bs
->drv
;
5357 trace_bdrv_lock_medium(bs
, locked
);
5359 if (drv
&& drv
->bdrv_lock_medium
) {
5360 drv
->bdrv_lock_medium(bs
, locked
);
5364 /* needed for generic scsi interface */
5366 int bdrv_ioctl(BlockDriverState
*bs
, unsigned long int req
, void *buf
)
5368 BlockDriver
*drv
= bs
->drv
;
5370 if (drv
&& drv
->bdrv_ioctl
)
5371 return drv
->bdrv_ioctl(bs
, req
, buf
);
5375 BlockAIOCB
*bdrv_aio_ioctl(BlockDriverState
*bs
,
5376 unsigned long int req
, void *buf
,
5377 BlockCompletionFunc
*cb
, void *opaque
)
5379 BlockDriver
*drv
= bs
->drv
;
5381 if (drv
&& drv
->bdrv_aio_ioctl
)
5382 return drv
->bdrv_aio_ioctl(bs
, req
, buf
, cb
, opaque
);
5386 void bdrv_set_guest_block_size(BlockDriverState
*bs
, int align
)
5388 bs
->guest_block_size
= align
;
5391 void *qemu_blockalign(BlockDriverState
*bs
, size_t size
)
5393 return qemu_memalign(bdrv_opt_mem_align(bs
), size
);
5396 void *qemu_blockalign0(BlockDriverState
*bs
, size_t size
)
5398 return memset(qemu_blockalign(bs
, size
), 0, size
);
5401 void *qemu_try_blockalign(BlockDriverState
*bs
, size_t size
)
5403 size_t align
= bdrv_opt_mem_align(bs
);
5405 /* Ensure that NULL is never returned on success */
5411 return qemu_try_memalign(align
, size
);
5414 void *qemu_try_blockalign0(BlockDriverState
*bs
, size_t size
)
5416 void *mem
= qemu_try_blockalign(bs
, size
);
5419 memset(mem
, 0, size
);
5426 * Check if all memory in this vector is sector aligned.
5428 bool bdrv_qiov_is_aligned(BlockDriverState
*bs
, QEMUIOVector
*qiov
)
5431 size_t alignment
= bdrv_opt_mem_align(bs
);
5433 for (i
= 0; i
< qiov
->niov
; i
++) {
5434 if ((uintptr_t) qiov
->iov
[i
].iov_base
% alignment
) {
5437 if (qiov
->iov
[i
].iov_len
% alignment
) {
5445 BdrvDirtyBitmap
*bdrv_create_dirty_bitmap(BlockDriverState
*bs
, int granularity
,
5448 int64_t bitmap_size
;
5449 BdrvDirtyBitmap
*bitmap
;
5451 assert((granularity
& (granularity
- 1)) == 0);
5453 granularity
>>= BDRV_SECTOR_BITS
;
5454 assert(granularity
);
5455 bitmap_size
= bdrv_nb_sectors(bs
);
5456 if (bitmap_size
< 0) {
5457 error_setg_errno(errp
, -bitmap_size
, "could not get length of device");
5458 errno
= -bitmap_size
;
5461 bitmap
= g_new0(BdrvDirtyBitmap
, 1);
5462 bitmap
->bitmap
= hbitmap_alloc(bitmap_size
, ctz32(granularity
));
5463 QLIST_INSERT_HEAD(&bs
->dirty_bitmaps
, bitmap
, list
);
5467 void bdrv_release_dirty_bitmap(BlockDriverState
*bs
, BdrvDirtyBitmap
*bitmap
)
5469 BdrvDirtyBitmap
*bm
, *next
;
5470 QLIST_FOREACH_SAFE(bm
, &bs
->dirty_bitmaps
, list
, next
) {
5472 QLIST_REMOVE(bitmap
, list
);
5473 hbitmap_free(bitmap
->bitmap
);
5480 BlockDirtyInfoList
*bdrv_query_dirty_bitmaps(BlockDriverState
*bs
)
5482 BdrvDirtyBitmap
*bm
;
5483 BlockDirtyInfoList
*list
= NULL
;
5484 BlockDirtyInfoList
**plist
= &list
;
5486 QLIST_FOREACH(bm
, &bs
->dirty_bitmaps
, list
) {
5487 BlockDirtyInfo
*info
= g_new0(BlockDirtyInfo
, 1);
5488 BlockDirtyInfoList
*entry
= g_new0(BlockDirtyInfoList
, 1);
5489 info
->count
= bdrv_get_dirty_count(bs
, bm
);
5491 ((int64_t) BDRV_SECTOR_SIZE
<< hbitmap_granularity(bm
->bitmap
));
5492 entry
->value
= info
;
5494 plist
= &entry
->next
;
5500 int bdrv_get_dirty(BlockDriverState
*bs
, BdrvDirtyBitmap
*bitmap
, int64_t sector
)
5503 return hbitmap_get(bitmap
->bitmap
, sector
);
5509 void bdrv_dirty_iter_init(BlockDriverState
*bs
,
5510 BdrvDirtyBitmap
*bitmap
, HBitmapIter
*hbi
)
5512 hbitmap_iter_init(hbi
, bitmap
->bitmap
, 0);
5515 void bdrv_set_dirty_bitmap(BlockDriverState
*bs
, BdrvDirtyBitmap
*bitmap
,
5516 int64_t cur_sector
, int nr_sectors
)
5518 hbitmap_set(bitmap
->bitmap
, cur_sector
, nr_sectors
);
5521 void bdrv_reset_dirty_bitmap(BlockDriverState
*bs
, BdrvDirtyBitmap
*bitmap
,
5522 int64_t cur_sector
, int nr_sectors
)
5524 hbitmap_reset(bitmap
->bitmap
, cur_sector
, nr_sectors
);
5527 static void bdrv_set_dirty(BlockDriverState
*bs
, int64_t cur_sector
,
5530 BdrvDirtyBitmap
*bitmap
;
5531 QLIST_FOREACH(bitmap
, &bs
->dirty_bitmaps
, list
) {
5532 hbitmap_set(bitmap
->bitmap
, cur_sector
, nr_sectors
);
5536 static void bdrv_reset_dirty(BlockDriverState
*bs
, int64_t cur_sector
,
5539 BdrvDirtyBitmap
*bitmap
;
5540 QLIST_FOREACH(bitmap
, &bs
->dirty_bitmaps
, list
) {
5541 hbitmap_reset(bitmap
->bitmap
, cur_sector
, nr_sectors
);
5545 int64_t bdrv_get_dirty_count(BlockDriverState
*bs
, BdrvDirtyBitmap
*bitmap
)
5547 return hbitmap_count(bitmap
->bitmap
);
5550 /* Get a reference to bs */
5551 void bdrv_ref(BlockDriverState
*bs
)
5556 /* Release a previously grabbed reference to bs.
5557 * If after releasing, reference count is zero, the BlockDriverState is
5559 void bdrv_unref(BlockDriverState
*bs
)
5564 assert(bs
->refcnt
> 0);
5565 if (--bs
->refcnt
== 0) {
5570 struct BdrvOpBlocker
{
5572 QLIST_ENTRY(BdrvOpBlocker
) list
;
5575 bool bdrv_op_is_blocked(BlockDriverState
*bs
, BlockOpType op
, Error
**errp
)
5577 BdrvOpBlocker
*blocker
;
5578 assert((int) op
>= 0 && op
< BLOCK_OP_TYPE_MAX
);
5579 if (!QLIST_EMPTY(&bs
->op_blockers
[op
])) {
5580 blocker
= QLIST_FIRST(&bs
->op_blockers
[op
]);
5582 error_setg(errp
, "Device '%s' is busy: %s",
5583 bdrv_get_device_name(bs
),
5584 error_get_pretty(blocker
->reason
));
5591 void bdrv_op_block(BlockDriverState
*bs
, BlockOpType op
, Error
*reason
)
5593 BdrvOpBlocker
*blocker
;
5594 assert((int) op
>= 0 && op
< BLOCK_OP_TYPE_MAX
);
5596 blocker
= g_new0(BdrvOpBlocker
, 1);
5597 blocker
->reason
= reason
;
5598 QLIST_INSERT_HEAD(&bs
->op_blockers
[op
], blocker
, list
);
5601 void bdrv_op_unblock(BlockDriverState
*bs
, BlockOpType op
, Error
*reason
)
5603 BdrvOpBlocker
*blocker
, *next
;
5604 assert((int) op
>= 0 && op
< BLOCK_OP_TYPE_MAX
);
5605 QLIST_FOREACH_SAFE(blocker
, &bs
->op_blockers
[op
], list
, next
) {
5606 if (blocker
->reason
== reason
) {
5607 QLIST_REMOVE(blocker
, list
);
5613 void bdrv_op_block_all(BlockDriverState
*bs
, Error
*reason
)
5616 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
5617 bdrv_op_block(bs
, i
, reason
);
5621 void bdrv_op_unblock_all(BlockDriverState
*bs
, Error
*reason
)
5624 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
5625 bdrv_op_unblock(bs
, i
, reason
);
5629 bool bdrv_op_blocker_is_empty(BlockDriverState
*bs
)
5633 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
5634 if (!QLIST_EMPTY(&bs
->op_blockers
[i
])) {
5641 void bdrv_iostatus_enable(BlockDriverState
*bs
)
5643 bs
->iostatus_enabled
= true;
5644 bs
->iostatus
= BLOCK_DEVICE_IO_STATUS_OK
;
5647 /* The I/O status is only enabled if the drive explicitly
5648 * enables it _and_ the VM is configured to stop on errors */
5649 bool bdrv_iostatus_is_enabled(const BlockDriverState
*bs
)
5651 return (bs
->iostatus_enabled
&&
5652 (bs
->on_write_error
== BLOCKDEV_ON_ERROR_ENOSPC
||
5653 bs
->on_write_error
== BLOCKDEV_ON_ERROR_STOP
||
5654 bs
->on_read_error
== BLOCKDEV_ON_ERROR_STOP
));
5657 void bdrv_iostatus_disable(BlockDriverState
*bs
)
5659 bs
->iostatus_enabled
= false;
5662 void bdrv_iostatus_reset(BlockDriverState
*bs
)
5664 if (bdrv_iostatus_is_enabled(bs
)) {
5665 bs
->iostatus
= BLOCK_DEVICE_IO_STATUS_OK
;
5667 block_job_iostatus_reset(bs
->job
);
5672 void bdrv_iostatus_set_err(BlockDriverState
*bs
, int error
)
5674 assert(bdrv_iostatus_is_enabled(bs
));
5675 if (bs
->iostatus
== BLOCK_DEVICE_IO_STATUS_OK
) {
5676 bs
->iostatus
= error
== ENOSPC
? BLOCK_DEVICE_IO_STATUS_NOSPACE
:
5677 BLOCK_DEVICE_IO_STATUS_FAILED
;
5681 void bdrv_img_create(const char *filename
, const char *fmt
,
5682 const char *base_filename
, const char *base_fmt
,
5683 char *options
, uint64_t img_size
, int flags
,
5684 Error
**errp
, bool quiet
)
5686 QemuOptsList
*create_opts
= NULL
;
5687 QemuOpts
*opts
= NULL
;
5688 const char *backing_fmt
, *backing_file
;
5690 BlockDriver
*drv
, *proto_drv
;
5691 BlockDriver
*backing_drv
= NULL
;
5692 Error
*local_err
= NULL
;
5695 /* Find driver and parse its options */
5696 drv
= bdrv_find_format(fmt
);
5698 error_setg(errp
, "Unknown file format '%s'", fmt
);
5702 proto_drv
= bdrv_find_protocol(filename
, true, errp
);
5707 if (!drv
->create_opts
) {
5708 error_setg(errp
, "Format driver '%s' does not support image creation",
5713 if (!proto_drv
->create_opts
) {
5714 error_setg(errp
, "Protocol driver '%s' does not support image creation",
5715 proto_drv
->format_name
);
5719 create_opts
= qemu_opts_append(create_opts
, drv
->create_opts
);
5720 create_opts
= qemu_opts_append(create_opts
, proto_drv
->create_opts
);
5722 /* Create parameter list with default values */
5723 opts
= qemu_opts_create(create_opts
, NULL
, 0, &error_abort
);
5724 qemu_opt_set_number(opts
, BLOCK_OPT_SIZE
, img_size
, &error_abort
);
5726 /* Parse -o options */
5728 qemu_opts_do_parse(opts
, options
, NULL
, &local_err
);
5730 error_report_err(local_err
);
5732 error_setg(errp
, "Invalid options for file format '%s'", fmt
);
5737 if (base_filename
) {
5738 qemu_opt_set(opts
, BLOCK_OPT_BACKING_FILE
, base_filename
, &local_err
);
5740 error_setg(errp
, "Backing file not supported for file format '%s'",
5747 qemu_opt_set(opts
, BLOCK_OPT_BACKING_FMT
, base_fmt
, &local_err
);
5749 error_setg(errp
, "Backing file format not supported for file "
5750 "format '%s'", fmt
);
5755 backing_file
= qemu_opt_get(opts
, BLOCK_OPT_BACKING_FILE
);
5757 if (!strcmp(filename
, backing_file
)) {
5758 error_setg(errp
, "Error: Trying to create an image with the "
5759 "same filename as the backing file");
5764 backing_fmt
= qemu_opt_get(opts
, BLOCK_OPT_BACKING_FMT
);
5766 backing_drv
= bdrv_find_format(backing_fmt
);
5768 error_setg(errp
, "Unknown backing file format '%s'",
5774 // The size for the image must always be specified, with one exception:
5775 // If we are using a backing file, we can obtain the size from there
5776 size
= qemu_opt_get_size(opts
, BLOCK_OPT_SIZE
, 0);
5779 BlockDriverState
*bs
;
5780 char *full_backing
= g_new0(char, PATH_MAX
);
5784 bdrv_get_full_backing_filename_from_filename(filename
, backing_file
,
5785 full_backing
, PATH_MAX
,
5788 g_free(full_backing
);
5792 /* backing files always opened read-only */
5794 flags
& ~(BDRV_O_RDWR
| BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
);
5797 ret
= bdrv_open(&bs
, full_backing
, NULL
, NULL
, back_flags
,
5798 backing_drv
, &local_err
);
5799 g_free(full_backing
);
5803 size
= bdrv_getlength(bs
);
5805 error_setg_errno(errp
, -size
, "Could not get size of '%s'",
5811 qemu_opt_set_number(opts
, BLOCK_OPT_SIZE
, size
, &error_abort
);
5815 error_setg(errp
, "Image creation needs a size parameter");
5821 printf("Formatting '%s', fmt=%s", filename
, fmt
);
5822 qemu_opts_print(opts
, " ");
5826 ret
= bdrv_create(drv
, filename
, opts
, &local_err
);
5828 if (ret
== -EFBIG
) {
5829 /* This is generally a better message than whatever the driver would
5830 * deliver (especially because of the cluster_size_hint), since that
5831 * is most probably not much different from "image too large". */
5832 const char *cluster_size_hint
= "";
5833 if (qemu_opt_get_size(opts
, BLOCK_OPT_CLUSTER_SIZE
, 0)) {
5834 cluster_size_hint
= " (try using a larger cluster size)";
5836 error_setg(errp
, "The image size is too large for file format '%s'"
5837 "%s", fmt
, cluster_size_hint
);
5838 error_free(local_err
);
5843 qemu_opts_del(opts
);
5844 qemu_opts_free(create_opts
);
5846 error_propagate(errp
, local_err
);
5850 AioContext
*bdrv_get_aio_context(BlockDriverState
*bs
)
5852 return bs
->aio_context
;
5855 void bdrv_detach_aio_context(BlockDriverState
*bs
)
5857 BdrvAioNotifier
*baf
;
5863 QLIST_FOREACH(baf
, &bs
->aio_notifiers
, list
) {
5864 baf
->detach_aio_context(baf
->opaque
);
5867 if (bs
->io_limits_enabled
) {
5868 throttle_detach_aio_context(&bs
->throttle_state
);
5870 if (bs
->drv
->bdrv_detach_aio_context
) {
5871 bs
->drv
->bdrv_detach_aio_context(bs
);
5874 bdrv_detach_aio_context(bs
->file
);
5876 if (bs
->backing_hd
) {
5877 bdrv_detach_aio_context(bs
->backing_hd
);
5880 bs
->aio_context
= NULL
;
5883 void bdrv_attach_aio_context(BlockDriverState
*bs
,
5884 AioContext
*new_context
)
5886 BdrvAioNotifier
*ban
;
5892 bs
->aio_context
= new_context
;
5894 if (bs
->backing_hd
) {
5895 bdrv_attach_aio_context(bs
->backing_hd
, new_context
);
5898 bdrv_attach_aio_context(bs
->file
, new_context
);
5900 if (bs
->drv
->bdrv_attach_aio_context
) {
5901 bs
->drv
->bdrv_attach_aio_context(bs
, new_context
);
5903 if (bs
->io_limits_enabled
) {
5904 throttle_attach_aio_context(&bs
->throttle_state
, new_context
);
5907 QLIST_FOREACH(ban
, &bs
->aio_notifiers
, list
) {
5908 ban
->attached_aio_context(new_context
, ban
->opaque
);
5912 void bdrv_set_aio_context(BlockDriverState
*bs
, AioContext
*new_context
)
5914 bdrv_drain_all(); /* ensure there are no in-flight requests */
5916 bdrv_detach_aio_context(bs
);
5918 /* This function executes in the old AioContext so acquire the new one in
5919 * case it runs in a different thread.
5921 aio_context_acquire(new_context
);
5922 bdrv_attach_aio_context(bs
, new_context
);
5923 aio_context_release(new_context
);
5926 void bdrv_add_aio_context_notifier(BlockDriverState
*bs
,
5927 void (*attached_aio_context
)(AioContext
*new_context
, void *opaque
),
5928 void (*detach_aio_context
)(void *opaque
), void *opaque
)
5930 BdrvAioNotifier
*ban
= g_new(BdrvAioNotifier
, 1);
5931 *ban
= (BdrvAioNotifier
){
5932 .attached_aio_context
= attached_aio_context
,
5933 .detach_aio_context
= detach_aio_context
,
5937 QLIST_INSERT_HEAD(&bs
->aio_notifiers
, ban
, list
);
5940 void bdrv_remove_aio_context_notifier(BlockDriverState
*bs
,
5941 void (*attached_aio_context
)(AioContext
*,
5943 void (*detach_aio_context
)(void *),
5946 BdrvAioNotifier
*ban
, *ban_next
;
5948 QLIST_FOREACH_SAFE(ban
, &bs
->aio_notifiers
, list
, ban_next
) {
5949 if (ban
->attached_aio_context
== attached_aio_context
&&
5950 ban
->detach_aio_context
== detach_aio_context
&&
5951 ban
->opaque
== opaque
)
5953 QLIST_REMOVE(ban
, list
);
5963 void bdrv_add_before_write_notifier(BlockDriverState
*bs
,
5964 NotifierWithReturn
*notifier
)
5966 notifier_with_return_list_add(&bs
->before_write_notifiers
, notifier
);
5969 int bdrv_amend_options(BlockDriverState
*bs
, QemuOpts
*opts
,
5970 BlockDriverAmendStatusCB
*status_cb
)
5972 if (!bs
->drv
->bdrv_amend_options
) {
5975 return bs
->drv
->bdrv_amend_options(bs
, opts
, status_cb
);
5978 /* This function will be called by the bdrv_recurse_is_first_non_filter method
5979 * of block filter and by bdrv_is_first_non_filter.
5980 * It is used to test if the given bs is the candidate or recurse more in the
5983 bool bdrv_recurse_is_first_non_filter(BlockDriverState
*bs
,
5984 BlockDriverState
*candidate
)
5986 /* return false if basic checks fails */
5987 if (!bs
|| !bs
->drv
) {
5991 /* the code reached a non block filter driver -> check if the bs is
5992 * the same as the candidate. It's the recursion termination condition.
5994 if (!bs
->drv
->is_filter
) {
5995 return bs
== candidate
;
5997 /* Down this path the driver is a block filter driver */
5999 /* If the block filter recursion method is defined use it to recurse down
6002 if (bs
->drv
->bdrv_recurse_is_first_non_filter
) {
6003 return bs
->drv
->bdrv_recurse_is_first_non_filter(bs
, candidate
);
6006 /* the driver is a block filter but don't allow to recurse -> return false
6011 /* This function checks if the candidate is the first non filter bs down it's
6012 * bs chain. Since we don't have pointers to parents it explore all bs chains
6013 * from the top. Some filters can choose not to pass down the recursion.
6015 bool bdrv_is_first_non_filter(BlockDriverState
*candidate
)
6017 BlockDriverState
*bs
;
6019 /* walk down the bs forest recursively */
6020 QTAILQ_FOREACH(bs
, &bdrv_states
, device_list
) {
6023 /* try to recurse in this top level bs */
6024 perm
= bdrv_recurse_is_first_non_filter(bs
, candidate
);
6026 /* candidate is the first non filter */
6035 BlockDriverState
*check_to_replace_node(const char *node_name
, Error
**errp
)
6037 BlockDriverState
*to_replace_bs
= bdrv_find_node(node_name
);
6038 AioContext
*aio_context
;
6040 if (!to_replace_bs
) {
6041 error_setg(errp
, "Node name '%s' not found", node_name
);
6045 aio_context
= bdrv_get_aio_context(to_replace_bs
);
6046 aio_context_acquire(aio_context
);
6048 if (bdrv_op_is_blocked(to_replace_bs
, BLOCK_OP_TYPE_REPLACE
, errp
)) {
6049 to_replace_bs
= NULL
;
6053 /* We don't want arbitrary node of the BDS chain to be replaced only the top
6054 * most non filter in order to prevent data corruption.
6055 * Another benefit is that this tests exclude backing files which are
6056 * blocked by the backing blockers.
6058 if (!bdrv_is_first_non_filter(to_replace_bs
)) {
6059 error_setg(errp
, "Only top most non filter can be replaced");
6060 to_replace_bs
= NULL
;
6065 aio_context_release(aio_context
);
6066 return to_replace_bs
;
6069 void bdrv_io_plug(BlockDriverState
*bs
)
6071 BlockDriver
*drv
= bs
->drv
;
6072 if (drv
&& drv
->bdrv_io_plug
) {
6073 drv
->bdrv_io_plug(bs
);
6074 } else if (bs
->file
) {
6075 bdrv_io_plug(bs
->file
);
6079 void bdrv_io_unplug(BlockDriverState
*bs
)
6081 BlockDriver
*drv
= bs
->drv
;
6082 if (drv
&& drv
->bdrv_io_unplug
) {
6083 drv
->bdrv_io_unplug(bs
);
6084 } else if (bs
->file
) {
6085 bdrv_io_unplug(bs
->file
);
6089 void bdrv_flush_io_queue(BlockDriverState
*bs
)
6091 BlockDriver
*drv
= bs
->drv
;
6092 if (drv
&& drv
->bdrv_flush_io_queue
) {
6093 drv
->bdrv_flush_io_queue(bs
);
6094 } else if (bs
->file
) {
6095 bdrv_flush_io_queue(bs
->file
);
6099 static bool append_open_options(QDict
*d
, BlockDriverState
*bs
)
6101 const QDictEntry
*entry
;
6102 bool found_any
= false;
6104 for (entry
= qdict_first(bs
->options
); entry
;
6105 entry
= qdict_next(bs
->options
, entry
))
6107 /* Only take options for this level and exclude all non-driver-specific
6109 if (!strchr(qdict_entry_key(entry
), '.') &&
6110 strcmp(qdict_entry_key(entry
), "node-name"))
6112 qobject_incref(qdict_entry_value(entry
));
6113 qdict_put_obj(d
, qdict_entry_key(entry
), qdict_entry_value(entry
));
6121 /* Updates the following BDS fields:
6122 * - exact_filename: A filename which may be used for opening a block device
6123 * which (mostly) equals the given BDS (even without any
6124 * other options; so reading and writing must return the same
6125 * results, but caching etc. may be different)
6126 * - full_open_options: Options which, when given when opening a block device
6127 * (without a filename), result in a BDS (mostly)
6128 * equalling the given one
6129 * - filename: If exact_filename is set, it is copied here. Otherwise,
6130 * full_open_options is converted to a JSON object, prefixed with
6131 * "json:" (for use through the JSON pseudo protocol) and put here.
6133 void bdrv_refresh_filename(BlockDriverState
*bs
)
6135 BlockDriver
*drv
= bs
->drv
;
6142 /* This BDS's file name will most probably depend on its file's name, so
6143 * refresh that first */
6145 bdrv_refresh_filename(bs
->file
);
6148 if (drv
->bdrv_refresh_filename
) {
6149 /* Obsolete information is of no use here, so drop the old file name
6150 * information before refreshing it */
6151 bs
->exact_filename
[0] = '\0';
6152 if (bs
->full_open_options
) {
6153 QDECREF(bs
->full_open_options
);
6154 bs
->full_open_options
= NULL
;
6157 drv
->bdrv_refresh_filename(bs
);
6158 } else if (bs
->file
) {
6159 /* Try to reconstruct valid information from the underlying file */
6160 bool has_open_options
;
6162 bs
->exact_filename
[0] = '\0';
6163 if (bs
->full_open_options
) {
6164 QDECREF(bs
->full_open_options
);
6165 bs
->full_open_options
= NULL
;
6169 has_open_options
= append_open_options(opts
, bs
);
6171 /* If no specific options have been given for this BDS, the filename of
6172 * the underlying file should suffice for this one as well */
6173 if (bs
->file
->exact_filename
[0] && !has_open_options
) {
6174 strcpy(bs
->exact_filename
, bs
->file
->exact_filename
);
6176 /* Reconstructing the full options QDict is simple for most format block
6177 * drivers, as long as the full options are known for the underlying
6178 * file BDS. The full options QDict of that file BDS should somehow
6179 * contain a representation of the filename, therefore the following
6180 * suffices without querying the (exact_)filename of this BDS. */
6181 if (bs
->file
->full_open_options
) {
6182 qdict_put_obj(opts
, "driver",
6183 QOBJECT(qstring_from_str(drv
->format_name
)));
6184 QINCREF(bs
->file
->full_open_options
);
6185 qdict_put_obj(opts
, "file", QOBJECT(bs
->file
->full_open_options
));
6187 bs
->full_open_options
= opts
;
6191 } else if (!bs
->full_open_options
&& qdict_size(bs
->options
)) {
6192 /* There is no underlying file BDS (at least referenced by BDS.file),
6193 * so the full options QDict should be equal to the options given
6194 * specifically for this block device when it was opened (plus the
6195 * driver specification).
6196 * Because those options don't change, there is no need to update
6197 * full_open_options when it's already set. */
6200 append_open_options(opts
, bs
);
6201 qdict_put_obj(opts
, "driver",
6202 QOBJECT(qstring_from_str(drv
->format_name
)));
6204 if (bs
->exact_filename
[0]) {
6205 /* This may not work for all block protocol drivers (some may
6206 * require this filename to be parsed), but we have to find some
6207 * default solution here, so just include it. If some block driver
6208 * does not support pure options without any filename at all or
6209 * needs some special format of the options QDict, it needs to
6210 * implement the driver-specific bdrv_refresh_filename() function.
6212 qdict_put_obj(opts
, "filename",
6213 QOBJECT(qstring_from_str(bs
->exact_filename
)));
6216 bs
->full_open_options
= opts
;
6219 if (bs
->exact_filename
[0]) {
6220 pstrcpy(bs
->filename
, sizeof(bs
->filename
), bs
->exact_filename
);
6221 } else if (bs
->full_open_options
) {
6222 QString
*json
= qobject_to_json(QOBJECT(bs
->full_open_options
));
6223 snprintf(bs
->filename
, sizeof(bs
->filename
), "json:%s",
6224 qstring_get_str(json
));
6229 /* This accessor function purpose is to allow the device models to access the
6230 * BlockAcctStats structure embedded inside a BlockDriverState without being
6231 * aware of the BlockDriverState structure layout.
6232 * It will go away when the BlockAcctStats structure will be moved inside
6233 * the device models.
6235 BlockAcctStats
*bdrv_get_stats(BlockDriverState
*bs
)