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 "monitor/monitor.h"
28 #include "block/block_int.h"
29 #include "block/blockjob.h"
30 #include "qemu/module.h"
31 #include "qapi/qmp/qjson.h"
32 #include "sysemu/sysemu.h"
33 #include "qemu/notify.h"
34 #include "block/coroutine.h"
35 #include "block/qapi.h"
36 #include "qmp-commands.h"
37 #include "qemu/timer.h"
40 #include <sys/types.h>
42 #include <sys/ioctl.h>
43 #include <sys/queue.h>
53 struct BdrvDirtyBitmap
{
55 QLIST_ENTRY(BdrvDirtyBitmap
) list
;
58 #define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */
60 static void bdrv_dev_change_media_cb(BlockDriverState
*bs
, bool load
);
61 static BlockDriverAIOCB
*bdrv_aio_readv_em(BlockDriverState
*bs
,
62 int64_t sector_num
, QEMUIOVector
*qiov
, int nb_sectors
,
63 BlockDriverCompletionFunc
*cb
, void *opaque
);
64 static BlockDriverAIOCB
*bdrv_aio_writev_em(BlockDriverState
*bs
,
65 int64_t sector_num
, QEMUIOVector
*qiov
, int nb_sectors
,
66 BlockDriverCompletionFunc
*cb
, void *opaque
);
67 static int coroutine_fn
bdrv_co_readv_em(BlockDriverState
*bs
,
68 int64_t sector_num
, int nb_sectors
,
70 static int coroutine_fn
bdrv_co_writev_em(BlockDriverState
*bs
,
71 int64_t sector_num
, int nb_sectors
,
73 static int coroutine_fn
bdrv_co_do_preadv(BlockDriverState
*bs
,
74 int64_t offset
, unsigned int bytes
, QEMUIOVector
*qiov
,
75 BdrvRequestFlags flags
);
76 static int coroutine_fn
bdrv_co_do_pwritev(BlockDriverState
*bs
,
77 int64_t offset
, unsigned int bytes
, QEMUIOVector
*qiov
,
78 BdrvRequestFlags flags
);
79 static BlockDriverAIOCB
*bdrv_co_aio_rw_vector(BlockDriverState
*bs
,
83 BdrvRequestFlags flags
,
84 BlockDriverCompletionFunc
*cb
,
87 static void coroutine_fn
bdrv_co_do_rw(void *opaque
);
88 static int coroutine_fn
bdrv_co_do_write_zeroes(BlockDriverState
*bs
,
89 int64_t sector_num
, int nb_sectors
, BdrvRequestFlags flags
);
91 static QTAILQ_HEAD(, BlockDriverState
) bdrv_states
=
92 QTAILQ_HEAD_INITIALIZER(bdrv_states
);
94 static QTAILQ_HEAD(, BlockDriverState
) graph_bdrv_states
=
95 QTAILQ_HEAD_INITIALIZER(graph_bdrv_states
);
97 static QLIST_HEAD(, BlockDriver
) bdrv_drivers
=
98 QLIST_HEAD_INITIALIZER(bdrv_drivers
);
100 /* If non-zero, use only whitelisted block drivers */
101 static int use_bdrv_whitelist
;
104 static int is_windows_drive_prefix(const char *filename
)
106 return (((filename
[0] >= 'a' && filename
[0] <= 'z') ||
107 (filename
[0] >= 'A' && filename
[0] <= 'Z')) &&
111 int is_windows_drive(const char *filename
)
113 if (is_windows_drive_prefix(filename
) &&
116 if (strstart(filename
, "\\\\.\\", NULL
) ||
117 strstart(filename
, "//./", NULL
))
123 /* throttling disk I/O limits */
124 void bdrv_set_io_limits(BlockDriverState
*bs
,
129 throttle_config(&bs
->throttle_state
, cfg
);
131 for (i
= 0; i
< 2; i
++) {
132 qemu_co_enter_next(&bs
->throttled_reqs
[i
]);
136 /* this function drain all the throttled IOs */
137 static bool bdrv_start_throttled_reqs(BlockDriverState
*bs
)
139 bool drained
= false;
140 bool enabled
= bs
->io_limits_enabled
;
143 bs
->io_limits_enabled
= false;
145 for (i
= 0; i
< 2; i
++) {
146 while (qemu_co_enter_next(&bs
->throttled_reqs
[i
])) {
151 bs
->io_limits_enabled
= enabled
;
156 void bdrv_io_limits_disable(BlockDriverState
*bs
)
158 bs
->io_limits_enabled
= false;
160 bdrv_start_throttled_reqs(bs
);
162 throttle_destroy(&bs
->throttle_state
);
165 static void bdrv_throttle_read_timer_cb(void *opaque
)
167 BlockDriverState
*bs
= opaque
;
168 qemu_co_enter_next(&bs
->throttled_reqs
[0]);
171 static void bdrv_throttle_write_timer_cb(void *opaque
)
173 BlockDriverState
*bs
= opaque
;
174 qemu_co_enter_next(&bs
->throttled_reqs
[1]);
177 /* should be called before bdrv_set_io_limits if a limit is set */
178 void bdrv_io_limits_enable(BlockDriverState
*bs
)
180 assert(!bs
->io_limits_enabled
);
181 throttle_init(&bs
->throttle_state
,
183 bdrv_throttle_read_timer_cb
,
184 bdrv_throttle_write_timer_cb
,
186 bs
->io_limits_enabled
= true;
189 /* This function makes an IO wait if needed
191 * @nb_sectors: the number of sectors of the IO
192 * @is_write: is the IO a write
194 static void bdrv_io_limits_intercept(BlockDriverState
*bs
,
198 /* does this io must wait */
199 bool must_wait
= throttle_schedule_timer(&bs
->throttle_state
, is_write
);
201 /* if must wait or any request of this type throttled queue the IO */
203 !qemu_co_queue_empty(&bs
->throttled_reqs
[is_write
])) {
204 qemu_co_queue_wait(&bs
->throttled_reqs
[is_write
]);
207 /* the IO will be executed, do the accounting */
208 throttle_account(&bs
->throttle_state
, is_write
, bytes
);
211 /* if the next request must wait -> do nothing */
212 if (throttle_schedule_timer(&bs
->throttle_state
, is_write
)) {
216 /* else queue next request for execution */
217 qemu_co_queue_next(&bs
->throttled_reqs
[is_write
]);
220 size_t bdrv_opt_mem_align(BlockDriverState
*bs
)
222 if (!bs
|| !bs
->drv
) {
223 /* 4k should be on the safe side */
227 return bs
->bl
.opt_mem_alignment
;
230 /* check if the path starts with "<protocol>:" */
231 static int path_has_protocol(const char *path
)
236 if (is_windows_drive(path
) ||
237 is_windows_drive_prefix(path
)) {
240 p
= path
+ strcspn(path
, ":/\\");
242 p
= path
+ strcspn(path
, ":/");
248 int path_is_absolute(const char *path
)
251 /* specific case for names like: "\\.\d:" */
252 if (is_windows_drive(path
) || is_windows_drive_prefix(path
)) {
255 return (*path
== '/' || *path
== '\\');
257 return (*path
== '/');
261 /* if filename is absolute, just copy it to dest. Otherwise, build a
262 path to it by considering it is relative to base_path. URL are
264 void path_combine(char *dest
, int dest_size
,
265 const char *base_path
,
266 const char *filename
)
273 if (path_is_absolute(filename
)) {
274 pstrcpy(dest
, dest_size
, filename
);
276 p
= strchr(base_path
, ':');
281 p1
= strrchr(base_path
, '/');
285 p2
= strrchr(base_path
, '\\');
297 if (len
> dest_size
- 1)
299 memcpy(dest
, base_path
, len
);
301 pstrcat(dest
, dest_size
, filename
);
305 void bdrv_get_full_backing_filename(BlockDriverState
*bs
, char *dest
, size_t sz
)
307 if (bs
->backing_file
[0] == '\0' || path_has_protocol(bs
->backing_file
)) {
308 pstrcpy(dest
, sz
, bs
->backing_file
);
310 path_combine(dest
, sz
, bs
->filename
, bs
->backing_file
);
314 void bdrv_register(BlockDriver
*bdrv
)
316 /* Block drivers without coroutine functions need emulation */
317 if (!bdrv
->bdrv_co_readv
) {
318 bdrv
->bdrv_co_readv
= bdrv_co_readv_em
;
319 bdrv
->bdrv_co_writev
= bdrv_co_writev_em
;
321 /* bdrv_co_readv_em()/brdv_co_writev_em() work in terms of aio, so if
322 * the block driver lacks aio we need to emulate that too.
324 if (!bdrv
->bdrv_aio_readv
) {
325 /* add AIO emulation layer */
326 bdrv
->bdrv_aio_readv
= bdrv_aio_readv_em
;
327 bdrv
->bdrv_aio_writev
= bdrv_aio_writev_em
;
331 QLIST_INSERT_HEAD(&bdrv_drivers
, bdrv
, list
);
334 /* create a new block device (by default it is empty) */
335 BlockDriverState
*bdrv_new(const char *device_name
, Error
**errp
)
337 BlockDriverState
*bs
;
340 if (bdrv_find(device_name
)) {
341 error_setg(errp
, "Device with id '%s' already exists",
345 if (bdrv_find_node(device_name
)) {
346 error_setg(errp
, "Device with node-name '%s' already exists",
351 bs
= g_malloc0(sizeof(BlockDriverState
));
352 QLIST_INIT(&bs
->dirty_bitmaps
);
353 pstrcpy(bs
->device_name
, sizeof(bs
->device_name
), device_name
);
354 if (device_name
[0] != '\0') {
355 QTAILQ_INSERT_TAIL(&bdrv_states
, bs
, device_list
);
357 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
358 QLIST_INIT(&bs
->op_blockers
[i
]);
360 bdrv_iostatus_disable(bs
);
361 notifier_list_init(&bs
->close_notifiers
);
362 notifier_with_return_list_init(&bs
->before_write_notifiers
);
363 qemu_co_queue_init(&bs
->throttled_reqs
[0]);
364 qemu_co_queue_init(&bs
->throttled_reqs
[1]);
370 void bdrv_add_close_notifier(BlockDriverState
*bs
, Notifier
*notify
)
372 notifier_list_add(&bs
->close_notifiers
, notify
);
375 BlockDriver
*bdrv_find_format(const char *format_name
)
378 QLIST_FOREACH(drv1
, &bdrv_drivers
, list
) {
379 if (!strcmp(drv1
->format_name
, format_name
)) {
386 static int bdrv_is_whitelisted(BlockDriver
*drv
, bool read_only
)
388 static const char *whitelist_rw
[] = {
389 CONFIG_BDRV_RW_WHITELIST
391 static const char *whitelist_ro
[] = {
392 CONFIG_BDRV_RO_WHITELIST
396 if (!whitelist_rw
[0] && !whitelist_ro
[0]) {
397 return 1; /* no whitelist, anything goes */
400 for (p
= whitelist_rw
; *p
; p
++) {
401 if (!strcmp(drv
->format_name
, *p
)) {
406 for (p
= whitelist_ro
; *p
; p
++) {
407 if (!strcmp(drv
->format_name
, *p
)) {
415 BlockDriver
*bdrv_find_whitelisted_format(const char *format_name
,
418 BlockDriver
*drv
= bdrv_find_format(format_name
);
419 return drv
&& bdrv_is_whitelisted(drv
, read_only
) ? drv
: NULL
;
422 typedef struct CreateCo
{
425 QEMUOptionParameter
*options
;
430 static void coroutine_fn
bdrv_create_co_entry(void *opaque
)
432 Error
*local_err
= NULL
;
435 CreateCo
*cco
= opaque
;
438 ret
= cco
->drv
->bdrv_create(cco
->filename
, cco
->options
, &local_err
);
440 error_propagate(&cco
->err
, local_err
);
445 int bdrv_create(BlockDriver
*drv
, const char* filename
,
446 QEMUOptionParameter
*options
, Error
**errp
)
453 .filename
= g_strdup(filename
),
459 if (!drv
->bdrv_create
) {
460 error_setg(errp
, "Driver '%s' does not support image creation", drv
->format_name
);
465 if (qemu_in_coroutine()) {
466 /* Fast-path if already in coroutine context */
467 bdrv_create_co_entry(&cco
);
469 co
= qemu_coroutine_create(bdrv_create_co_entry
);
470 qemu_coroutine_enter(co
, &cco
);
471 while (cco
.ret
== NOT_DONE
) {
479 error_propagate(errp
, cco
.err
);
481 error_setg_errno(errp
, -ret
, "Could not create image");
486 g_free(cco
.filename
);
490 int bdrv_create_file(const char* filename
, QEMUOptionParameter
*options
,
494 Error
*local_err
= NULL
;
497 drv
= bdrv_find_protocol(filename
, true);
499 error_setg(errp
, "Could not find protocol for file '%s'", filename
);
503 ret
= bdrv_create(drv
, filename
, options
, &local_err
);
505 error_propagate(errp
, local_err
);
510 int bdrv_refresh_limits(BlockDriverState
*bs
)
512 BlockDriver
*drv
= bs
->drv
;
514 memset(&bs
->bl
, 0, sizeof(bs
->bl
));
520 /* Take some limits from the children as a default */
522 bdrv_refresh_limits(bs
->file
);
523 bs
->bl
.opt_transfer_length
= bs
->file
->bl
.opt_transfer_length
;
524 bs
->bl
.opt_mem_alignment
= bs
->file
->bl
.opt_mem_alignment
;
526 bs
->bl
.opt_mem_alignment
= 512;
529 if (bs
->backing_hd
) {
530 bdrv_refresh_limits(bs
->backing_hd
);
531 bs
->bl
.opt_transfer_length
=
532 MAX(bs
->bl
.opt_transfer_length
,
533 bs
->backing_hd
->bl
.opt_transfer_length
);
534 bs
->bl
.opt_mem_alignment
=
535 MAX(bs
->bl
.opt_mem_alignment
,
536 bs
->backing_hd
->bl
.opt_mem_alignment
);
539 /* Then let the driver override it */
540 if (drv
->bdrv_refresh_limits
) {
541 return drv
->bdrv_refresh_limits(bs
);
548 * Create a uniquely-named empty temporary file.
549 * Return 0 upon success, otherwise a negative errno value.
551 int get_tmp_filename(char *filename
, int size
)
554 char temp_dir
[MAX_PATH
];
555 /* GetTempFileName requires that its output buffer (4th param)
556 have length MAX_PATH or greater. */
557 assert(size
>= MAX_PATH
);
558 return (GetTempPath(MAX_PATH
, temp_dir
)
559 && GetTempFileName(temp_dir
, "qem", 0, filename
)
560 ? 0 : -GetLastError());
564 tmpdir
= getenv("TMPDIR");
568 if (snprintf(filename
, size
, "%s/vl.XXXXXX", tmpdir
) >= size
) {
571 fd
= mkstemp(filename
);
575 if (close(fd
) != 0) {
584 * Detect host devices. By convention, /dev/cdrom[N] is always
585 * recognized as a host CDROM.
587 static BlockDriver
*find_hdev_driver(const char *filename
)
589 int score_max
= 0, score
;
590 BlockDriver
*drv
= NULL
, *d
;
592 QLIST_FOREACH(d
, &bdrv_drivers
, list
) {
593 if (d
->bdrv_probe_device
) {
594 score
= d
->bdrv_probe_device(filename
);
595 if (score
> score_max
) {
605 BlockDriver
*bdrv_find_protocol(const char *filename
,
606 bool allow_protocol_prefix
)
613 /* TODO Drivers without bdrv_file_open must be specified explicitly */
616 * XXX(hch): we really should not let host device detection
617 * override an explicit protocol specification, but moving this
618 * later breaks access to device names with colons in them.
619 * Thanks to the brain-dead persistent naming schemes on udev-
620 * based Linux systems those actually are quite common.
622 drv1
= find_hdev_driver(filename
);
627 if (!path_has_protocol(filename
) || !allow_protocol_prefix
) {
628 return bdrv_find_format("file");
631 p
= strchr(filename
, ':');
634 if (len
> sizeof(protocol
) - 1)
635 len
= sizeof(protocol
) - 1;
636 memcpy(protocol
, filename
, len
);
637 protocol
[len
] = '\0';
638 QLIST_FOREACH(drv1
, &bdrv_drivers
, list
) {
639 if (drv1
->protocol_name
&&
640 !strcmp(drv1
->protocol_name
, protocol
)) {
647 static int find_image_format(BlockDriverState
*bs
, const char *filename
,
648 BlockDriver
**pdrv
, Error
**errp
)
650 int score
, score_max
;
651 BlockDriver
*drv1
, *drv
;
655 /* Return the raw BlockDriver * to scsi-generic devices or empty drives */
656 if (bs
->sg
|| !bdrv_is_inserted(bs
) || bdrv_getlength(bs
) == 0) {
657 drv
= bdrv_find_format("raw");
659 error_setg(errp
, "Could not find raw image format");
666 ret
= bdrv_pread(bs
, 0, buf
, sizeof(buf
));
668 error_setg_errno(errp
, -ret
, "Could not read image for determining its "
676 QLIST_FOREACH(drv1
, &bdrv_drivers
, list
) {
677 if (drv1
->bdrv_probe
) {
678 score
= drv1
->bdrv_probe(buf
, ret
, filename
);
679 if (score
> score_max
) {
686 error_setg(errp
, "Could not determine image format: No compatible "
695 * Set the current 'total_sectors' value
697 static int refresh_total_sectors(BlockDriverState
*bs
, int64_t hint
)
699 BlockDriver
*drv
= bs
->drv
;
701 /* Do not attempt drv->bdrv_getlength() on scsi-generic devices */
705 /* query actual device if possible, otherwise just trust the hint */
706 if (drv
->bdrv_getlength
) {
707 int64_t length
= drv
->bdrv_getlength(bs
);
711 hint
= DIV_ROUND_UP(length
, BDRV_SECTOR_SIZE
);
714 bs
->total_sectors
= hint
;
719 * Set open flags for a given discard mode
721 * Return 0 on success, -1 if the discard mode was invalid.
723 int bdrv_parse_discard_flags(const char *mode
, int *flags
)
725 *flags
&= ~BDRV_O_UNMAP
;
727 if (!strcmp(mode
, "off") || !strcmp(mode
, "ignore")) {
729 } else if (!strcmp(mode
, "on") || !strcmp(mode
, "unmap")) {
730 *flags
|= BDRV_O_UNMAP
;
739 * Set open flags for a given cache mode
741 * Return 0 on success, -1 if the cache mode was invalid.
743 int bdrv_parse_cache_flags(const char *mode
, int *flags
)
745 *flags
&= ~BDRV_O_CACHE_MASK
;
747 if (!strcmp(mode
, "off") || !strcmp(mode
, "none")) {
748 *flags
|= BDRV_O_NOCACHE
| BDRV_O_CACHE_WB
;
749 } else if (!strcmp(mode
, "directsync")) {
750 *flags
|= BDRV_O_NOCACHE
;
751 } else if (!strcmp(mode
, "writeback")) {
752 *flags
|= BDRV_O_CACHE_WB
;
753 } else if (!strcmp(mode
, "unsafe")) {
754 *flags
|= BDRV_O_CACHE_WB
;
755 *flags
|= BDRV_O_NO_FLUSH
;
756 } else if (!strcmp(mode
, "writethrough")) {
757 /* this is the default */
766 * The copy-on-read flag is actually a reference count so multiple users may
767 * use the feature without worrying about clobbering its previous state.
768 * Copy-on-read stays enabled until all users have called to disable it.
770 void bdrv_enable_copy_on_read(BlockDriverState
*bs
)
775 void bdrv_disable_copy_on_read(BlockDriverState
*bs
)
777 assert(bs
->copy_on_read
> 0);
782 * Returns the flags that a temporary snapshot should get, based on the
783 * originally requested flags (the originally requested image will have flags
784 * like a backing file)
786 static int bdrv_temp_snapshot_flags(int flags
)
788 return (flags
& ~BDRV_O_SNAPSHOT
) | BDRV_O_TEMPORARY
;
792 * Returns the flags that bs->file should get, based on the given flags for
795 static int bdrv_inherited_flags(int flags
)
797 /* Enable protocol handling, disable format probing for bs->file */
798 flags
|= BDRV_O_PROTOCOL
;
800 /* Our block drivers take care to send flushes and respect unmap policy,
801 * so we can enable both unconditionally on lower layers. */
802 flags
|= BDRV_O_CACHE_WB
| BDRV_O_UNMAP
;
804 /* Clear flags that only apply to the top layer */
805 flags
&= ~(BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
| BDRV_O_COPY_ON_READ
);
811 * Returns the flags that bs->backing_hd should get, based on the given flags
814 static int bdrv_backing_flags(int flags
)
816 /* backing files always opened read-only */
817 flags
&= ~(BDRV_O_RDWR
| BDRV_O_COPY_ON_READ
);
819 /* snapshot=on is handled on the top layer */
820 flags
&= ~(BDRV_O_SNAPSHOT
| BDRV_O_TEMPORARY
);
825 static int bdrv_open_flags(BlockDriverState
*bs
, int flags
)
827 int open_flags
= flags
| BDRV_O_CACHE_WB
;
830 * Clear flags that are internal to the block layer before opening the
833 open_flags
&= ~(BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
);
836 * Snapshots should be writable.
838 if (flags
& BDRV_O_TEMPORARY
) {
839 open_flags
|= BDRV_O_RDWR
;
845 static void bdrv_assign_node_name(BlockDriverState
*bs
,
846 const char *node_name
,
853 /* empty string node name is invalid */
854 if (node_name
[0] == '\0') {
855 error_setg(errp
, "Empty node name");
859 /* takes care of avoiding namespaces collisions */
860 if (bdrv_find(node_name
)) {
861 error_setg(errp
, "node-name=%s is conflicting with a device id",
866 /* takes care of avoiding duplicates node names */
867 if (bdrv_find_node(node_name
)) {
868 error_setg(errp
, "Duplicate node name");
872 /* copy node name into the bs and insert it into the graph list */
873 pstrcpy(bs
->node_name
, sizeof(bs
->node_name
), node_name
);
874 QTAILQ_INSERT_TAIL(&graph_bdrv_states
, bs
, node_list
);
878 * Common part for opening disk images and files
880 * Removes all processed options from *options.
882 static int bdrv_open_common(BlockDriverState
*bs
, BlockDriverState
*file
,
883 QDict
*options
, int flags
, BlockDriver
*drv
, Error
**errp
)
886 const char *filename
;
887 const char *node_name
= NULL
;
888 Error
*local_err
= NULL
;
891 assert(bs
->file
== NULL
);
892 assert(options
!= NULL
&& bs
->options
!= options
);
895 filename
= file
->filename
;
897 filename
= qdict_get_try_str(options
, "filename");
900 if (drv
->bdrv_needs_filename
&& !filename
) {
901 error_setg(errp
, "The '%s' block driver requires a file name",
906 trace_bdrv_open_common(bs
, filename
?: "", flags
, drv
->format_name
);
908 node_name
= qdict_get_try_str(options
, "node-name");
909 bdrv_assign_node_name(bs
, node_name
, &local_err
);
911 error_propagate(errp
, local_err
);
914 qdict_del(options
, "node-name");
916 /* bdrv_open() with directly using a protocol as drv. This layer is already
917 * opened, so assign it to bs (while file becomes a closed BlockDriverState)
918 * and return immediately. */
919 if (file
!= NULL
&& drv
->bdrv_file_open
) {
924 bs
->open_flags
= flags
;
925 bs
->guest_block_size
= 512;
926 bs
->request_alignment
= 512;
927 bs
->zero_beyond_eof
= true;
928 open_flags
= bdrv_open_flags(bs
, flags
);
929 bs
->read_only
= !(open_flags
& BDRV_O_RDWR
);
931 if (use_bdrv_whitelist
&& !bdrv_is_whitelisted(drv
, bs
->read_only
)) {
933 !bs
->read_only
&& bdrv_is_whitelisted(drv
, true)
934 ? "Driver '%s' can only be used for read-only devices"
935 : "Driver '%s' is not whitelisted",
940 assert(bs
->copy_on_read
== 0); /* bdrv_new() and bdrv_close() make it so */
941 if (flags
& BDRV_O_COPY_ON_READ
) {
942 if (!bs
->read_only
) {
943 bdrv_enable_copy_on_read(bs
);
945 error_setg(errp
, "Can't use copy-on-read on read-only device");
950 if (filename
!= NULL
) {
951 pstrcpy(bs
->filename
, sizeof(bs
->filename
), filename
);
953 bs
->filename
[0] = '\0';
957 bs
->opaque
= g_malloc0(drv
->instance_size
);
959 bs
->enable_write_cache
= !!(flags
& BDRV_O_CACHE_WB
);
961 /* Open the image, either directly or using a protocol */
962 if (drv
->bdrv_file_open
) {
963 assert(file
== NULL
);
964 assert(!drv
->bdrv_needs_filename
|| filename
!= NULL
);
965 ret
= drv
->bdrv_file_open(bs
, options
, open_flags
, &local_err
);
968 error_setg(errp
, "Can't use '%s' as a block driver for the "
969 "protocol level", drv
->format_name
);
974 ret
= drv
->bdrv_open(bs
, options
, open_flags
, &local_err
);
979 error_propagate(errp
, local_err
);
980 } else if (bs
->filename
[0]) {
981 error_setg_errno(errp
, -ret
, "Could not open '%s'", bs
->filename
);
983 error_setg_errno(errp
, -ret
, "Could not open image");
988 ret
= refresh_total_sectors(bs
, bs
->total_sectors
);
990 error_setg_errno(errp
, -ret
, "Could not refresh total sector count");
994 bdrv_refresh_limits(bs
);
995 assert(bdrv_opt_mem_align(bs
) != 0);
996 assert((bs
->request_alignment
!= 0) || bs
->sg
);
1008 * Opens a file using a protocol (file, host_device, nbd, ...)
1010 * options is an indirect pointer to a QDict of options to pass to the block
1011 * drivers, or pointer to NULL for an empty set of options. If this function
1012 * takes ownership of the QDict reference, it will set *options to NULL;
1013 * otherwise, it will contain unused/unrecognized options after this function
1014 * returns. Then, the caller is responsible for freeing it. If it intends to
1015 * reuse the QDict, QINCREF() should be called beforehand.
1017 static int bdrv_file_open(BlockDriverState
*bs
, const char *filename
,
1018 QDict
**options
, int flags
, Error
**errp
)
1021 const char *drvname
;
1022 bool parse_filename
= false;
1023 Error
*local_err
= NULL
;
1026 /* Fetch the file name from the options QDict if necessary */
1028 filename
= qdict_get_try_str(*options
, "filename");
1029 } else if (filename
&& !qdict_haskey(*options
, "filename")) {
1030 qdict_put(*options
, "filename", qstring_from_str(filename
));
1031 parse_filename
= true;
1033 error_setg(errp
, "Can't specify 'file' and 'filename' options at the "
1039 /* Find the right block driver */
1040 drvname
= qdict_get_try_str(*options
, "driver");
1042 drv
= bdrv_find_format(drvname
);
1044 error_setg(errp
, "Unknown driver '%s'", drvname
);
1046 qdict_del(*options
, "driver");
1047 } else if (filename
) {
1048 drv
= bdrv_find_protocol(filename
, parse_filename
);
1050 error_setg(errp
, "Unknown protocol");
1053 error_setg(errp
, "Must specify either driver or file");
1058 /* errp has been set already */
1063 /* Parse the filename and open it */
1064 if (drv
->bdrv_parse_filename
&& parse_filename
) {
1065 drv
->bdrv_parse_filename(filename
, *options
, &local_err
);
1067 error_propagate(errp
, local_err
);
1072 if (!drv
->bdrv_needs_filename
) {
1073 qdict_del(*options
, "filename");
1075 filename
= qdict_get_str(*options
, "filename");
1079 if (!drv
->bdrv_file_open
) {
1080 ret
= bdrv_open(&bs
, filename
, NULL
, *options
, flags
, drv
, &local_err
);
1083 ret
= bdrv_open_common(bs
, NULL
, *options
, flags
, drv
, &local_err
);
1086 error_propagate(errp
, local_err
);
1097 void bdrv_set_backing_hd(BlockDriverState
*bs
, BlockDriverState
*backing_hd
)
1100 if (bs
->backing_hd
) {
1101 assert(bs
->backing_blocker
);
1102 bdrv_op_unblock_all(bs
->backing_hd
, bs
->backing_blocker
);
1103 } else if (backing_hd
) {
1104 error_setg(&bs
->backing_blocker
,
1105 "device is used as backing hd of '%s'",
1109 bs
->backing_hd
= backing_hd
;
1111 error_free(bs
->backing_blocker
);
1112 bs
->backing_blocker
= NULL
;
1115 bs
->open_flags
&= ~BDRV_O_NO_BACKING
;
1116 pstrcpy(bs
->backing_file
, sizeof(bs
->backing_file
), backing_hd
->filename
);
1117 pstrcpy(bs
->backing_format
, sizeof(bs
->backing_format
),
1118 backing_hd
->drv
? backing_hd
->drv
->format_name
: "");
1120 bdrv_op_block_all(bs
->backing_hd
, bs
->backing_blocker
);
1121 /* Otherwise we won't be able to commit due to check in bdrv_commit */
1122 bdrv_op_unblock(bs
->backing_hd
, BLOCK_OP_TYPE_COMMIT
,
1123 bs
->backing_blocker
);
1125 bdrv_refresh_limits(bs
);
1129 * Opens the backing file for a BlockDriverState if not yet open
1131 * options is a QDict of options to pass to the block drivers, or NULL for an
1132 * empty set of options. The reference to the QDict is transferred to this
1133 * function (even on failure), so if the caller intends to reuse the dictionary,
1134 * it needs to use QINCREF() before calling bdrv_file_open.
1136 int bdrv_open_backing_file(BlockDriverState
*bs
, QDict
*options
, Error
**errp
)
1138 char *backing_filename
= g_malloc0(PATH_MAX
);
1140 BlockDriver
*back_drv
= NULL
;
1141 BlockDriverState
*backing_hd
;
1142 Error
*local_err
= NULL
;
1144 if (bs
->backing_hd
!= NULL
) {
1149 /* NULL means an empty set of options */
1150 if (options
== NULL
) {
1151 options
= qdict_new();
1154 bs
->open_flags
&= ~BDRV_O_NO_BACKING
;
1155 if (qdict_haskey(options
, "file.filename")) {
1156 backing_filename
[0] = '\0';
1157 } else if (bs
->backing_file
[0] == '\0' && qdict_size(options
) == 0) {
1161 bdrv_get_full_backing_filename(bs
, backing_filename
, PATH_MAX
);
1164 backing_hd
= bdrv_new("", errp
);
1166 if (bs
->backing_format
[0] != '\0') {
1167 back_drv
= bdrv_find_format(bs
->backing_format
);
1170 assert(bs
->backing_hd
== NULL
);
1171 ret
= bdrv_open(&backing_hd
,
1172 *backing_filename
? backing_filename
: NULL
, NULL
, options
,
1173 bdrv_backing_flags(bs
->open_flags
), back_drv
, &local_err
);
1175 bdrv_unref(backing_hd
);
1177 bs
->open_flags
|= BDRV_O_NO_BACKING
;
1178 error_setg(errp
, "Could not open backing file: %s",
1179 error_get_pretty(local_err
));
1180 error_free(local_err
);
1183 bdrv_set_backing_hd(bs
, backing_hd
);
1186 g_free(backing_filename
);
1191 * Opens a disk image whose options are given as BlockdevRef in another block
1194 * If allow_none is true, no image will be opened if filename is false and no
1195 * BlockdevRef is given. *pbs will remain unchanged and 0 will be returned.
1197 * bdrev_key specifies the key for the image's BlockdevRef in the options QDict.
1198 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
1199 * itself, all options starting with "${bdref_key}." are considered part of the
1202 * The BlockdevRef will be removed from the options QDict.
1204 * To conform with the behavior of bdrv_open(), *pbs has to be NULL.
1206 int bdrv_open_image(BlockDriverState
**pbs
, const char *filename
,
1207 QDict
*options
, const char *bdref_key
, int flags
,
1208 bool allow_none
, Error
**errp
)
1210 QDict
*image_options
;
1212 char *bdref_key_dot
;
1213 const char *reference
;
1216 assert(*pbs
== NULL
);
1218 bdref_key_dot
= g_strdup_printf("%s.", bdref_key
);
1219 qdict_extract_subqdict(options
, &image_options
, bdref_key_dot
);
1220 g_free(bdref_key_dot
);
1222 reference
= qdict_get_try_str(options
, bdref_key
);
1223 if (!filename
&& !reference
&& !qdict_size(image_options
)) {
1227 error_setg(errp
, "A block device must be specified for \"%s\"",
1231 QDECREF(image_options
);
1235 ret
= bdrv_open(pbs
, filename
, reference
, image_options
, flags
, NULL
, errp
);
1238 qdict_del(options
, bdref_key
);
1242 void bdrv_append_temp_snapshot(BlockDriverState
*bs
, int flags
, Error
**errp
)
1244 /* TODO: extra byte is a hack to ensure MAX_PATH space on Windows. */
1245 char *tmp_filename
= g_malloc0(PATH_MAX
+ 1);
1247 BlockDriver
*bdrv_qcow2
;
1248 QEMUOptionParameter
*create_options
;
1249 QDict
*snapshot_options
;
1250 BlockDriverState
*bs_snapshot
;
1254 /* if snapshot, we create a temporary backing file and open it
1255 instead of opening 'filename' directly */
1257 /* Get the required size from the image */
1258 total_size
= bdrv_getlength(bs
);
1259 if (total_size
< 0) {
1260 error_setg_errno(errp
, -total_size
, "Could not get image size");
1263 total_size
&= BDRV_SECTOR_MASK
;
1265 /* Create the temporary image */
1266 ret
= get_tmp_filename(tmp_filename
, PATH_MAX
+ 1);
1268 error_setg_errno(errp
, -ret
, "Could not get temporary filename");
1272 bdrv_qcow2
= bdrv_find_format("qcow2");
1273 create_options
= parse_option_parameters("", bdrv_qcow2
->create_options
,
1276 set_option_parameter_int(create_options
, BLOCK_OPT_SIZE
, total_size
);
1278 ret
= bdrv_create(bdrv_qcow2
, tmp_filename
, create_options
, &local_err
);
1279 free_option_parameters(create_options
);
1281 error_setg_errno(errp
, -ret
, "Could not create temporary overlay "
1282 "'%s': %s", tmp_filename
,
1283 error_get_pretty(local_err
));
1284 error_free(local_err
);
1288 /* Prepare a new options QDict for the temporary file */
1289 snapshot_options
= qdict_new();
1290 qdict_put(snapshot_options
, "file.driver",
1291 qstring_from_str("file"));
1292 qdict_put(snapshot_options
, "file.filename",
1293 qstring_from_str(tmp_filename
));
1295 bs_snapshot
= bdrv_new("", &error_abort
);
1297 ret
= bdrv_open(&bs_snapshot
, NULL
, NULL
, snapshot_options
,
1298 flags
, bdrv_qcow2
, &local_err
);
1300 error_propagate(errp
, local_err
);
1304 bdrv_append(bs_snapshot
, bs
);
1307 g_free(tmp_filename
);
1310 static QDict
*parse_json_filename(const char *filename
, Error
**errp
)
1312 QObject
*options_obj
;
1316 ret
= strstart(filename
, "json:", &filename
);
1319 options_obj
= qobject_from_json(filename
);
1321 error_setg(errp
, "Could not parse the JSON options");
1325 if (qobject_type(options_obj
) != QTYPE_QDICT
) {
1326 qobject_decref(options_obj
);
1327 error_setg(errp
, "Invalid JSON object given");
1331 options
= qobject_to_qdict(options_obj
);
1332 qdict_flatten(options
);
1338 * Opens a disk image (raw, qcow2, vmdk, ...)
1340 * options is a QDict of options to pass to the block drivers, or NULL for an
1341 * empty set of options. The reference to the QDict belongs to the block layer
1342 * after the call (even on failure), so if the caller intends to reuse the
1343 * dictionary, it needs to use QINCREF() before calling bdrv_open.
1345 * If *pbs is NULL, a new BDS will be created with a pointer to it stored there.
1346 * If it is not NULL, the referenced BDS will be reused.
1348 * The reference parameter may be used to specify an existing block device which
1349 * should be opened. If specified, neither options nor a filename may be given,
1350 * nor can an existing BDS be reused (that is, *pbs has to be NULL).
1352 int bdrv_open(BlockDriverState
**pbs
, const char *filename
,
1353 const char *reference
, QDict
*options
, int flags
,
1354 BlockDriver
*drv
, Error
**errp
)
1357 BlockDriverState
*file
= NULL
, *bs
;
1358 const char *drvname
;
1359 Error
*local_err
= NULL
;
1360 int snapshot_flags
= 0;
1365 bool options_non_empty
= options
? qdict_size(options
) : false;
1369 error_setg(errp
, "Cannot reuse an existing BDS when referencing "
1370 "another block device");
1374 if (filename
|| options_non_empty
) {
1375 error_setg(errp
, "Cannot reference an existing block device with "
1376 "additional options or a new filename");
1380 bs
= bdrv_lookup_bs(reference
, reference
, errp
);
1392 bs
= bdrv_new("", &error_abort
);
1395 /* NULL means an empty set of options */
1396 if (options
== NULL
) {
1397 options
= qdict_new();
1400 if (filename
&& g_str_has_prefix(filename
, "json:")) {
1401 QDict
*json_options
= parse_json_filename(filename
, &local_err
);
1407 /* Options given in the filename have lower priority than options
1408 * specified directly */
1409 qdict_join(options
, json_options
, false);
1410 QDECREF(json_options
);
1414 bs
->options
= options
;
1415 options
= qdict_clone_shallow(options
);
1417 if (flags
& BDRV_O_PROTOCOL
) {
1419 ret
= bdrv_file_open(bs
, filename
, &options
, flags
& ~BDRV_O_PROTOCOL
,
1424 } else if (bs
->drv
) {
1425 goto close_and_fail
;
1431 /* Open image file without format layer */
1432 if (flags
& BDRV_O_RDWR
) {
1433 flags
|= BDRV_O_ALLOW_RDWR
;
1435 if (flags
& BDRV_O_SNAPSHOT
) {
1436 snapshot_flags
= bdrv_temp_snapshot_flags(flags
);
1437 flags
= bdrv_backing_flags(flags
);
1440 assert(file
== NULL
);
1441 ret
= bdrv_open_image(&file
, filename
, options
, "file",
1442 bdrv_inherited_flags(flags
),
1448 /* Find the right image format driver */
1449 drvname
= qdict_get_try_str(options
, "driver");
1451 drv
= bdrv_find_format(drvname
);
1452 qdict_del(options
, "driver");
1454 error_setg(errp
, "Invalid driver: '%s'", drvname
);
1462 ret
= find_image_format(file
, filename
, &drv
, &local_err
);
1464 error_setg(errp
, "Must specify either driver or file");
1474 /* Open the image */
1475 ret
= bdrv_open_common(bs
, file
, options
, flags
, drv
, &local_err
);
1480 if (file
&& (bs
->file
!= file
)) {
1485 /* If there is a backing file, use it */
1486 if ((flags
& BDRV_O_NO_BACKING
) == 0) {
1487 QDict
*backing_options
;
1489 qdict_extract_subqdict(options
, &backing_options
, "backing.");
1490 ret
= bdrv_open_backing_file(bs
, backing_options
, &local_err
);
1492 goto close_and_fail
;
1496 /* For snapshot=on, create a temporary qcow2 overlay. bs points to the
1497 * temporary snapshot afterwards. */
1498 if (snapshot_flags
) {
1499 bdrv_append_temp_snapshot(bs
, snapshot_flags
, &local_err
);
1501 error_propagate(errp
, local_err
);
1502 goto close_and_fail
;
1508 /* Check if any unknown options were used */
1509 if (options
&& (qdict_size(options
) != 0)) {
1510 const QDictEntry
*entry
= qdict_first(options
);
1511 if (flags
& BDRV_O_PROTOCOL
) {
1512 error_setg(errp
, "Block protocol '%s' doesn't support the option "
1513 "'%s'", drv
->format_name
, entry
->key
);
1515 error_setg(errp
, "Block format '%s' used by device '%s' doesn't "
1516 "support the option '%s'", drv
->format_name
,
1517 bs
->device_name
, entry
->key
);
1521 goto close_and_fail
;
1524 if (!bdrv_key_required(bs
)) {
1525 bdrv_dev_change_media_cb(bs
, true);
1526 } else if (!runstate_check(RUN_STATE_PRELAUNCH
)
1527 && !runstate_check(RUN_STATE_INMIGRATE
)
1528 && !runstate_check(RUN_STATE_PAUSED
)) { /* HACK */
1530 "Guest must be stopped for opening of encrypted image");
1532 goto close_and_fail
;
1543 QDECREF(bs
->options
);
1547 /* If *pbs is NULL, a new BDS has been created in this function and
1548 needs to be freed now. Otherwise, it does not need to be closed,
1549 since it has not really been opened yet. */
1553 error_propagate(errp
, local_err
);
1558 /* See fail path, but now the BDS has to be always closed */
1566 error_propagate(errp
, local_err
);
1571 typedef struct BlockReopenQueueEntry
{
1573 BDRVReopenState state
;
1574 QSIMPLEQ_ENTRY(BlockReopenQueueEntry
) entry
;
1575 } BlockReopenQueueEntry
;
1578 * Adds a BlockDriverState to a simple queue for an atomic, transactional
1579 * reopen of multiple devices.
1581 * bs_queue can either be an existing BlockReopenQueue that has had QSIMPLE_INIT
1582 * already performed, or alternatively may be NULL a new BlockReopenQueue will
1583 * be created and initialized. This newly created BlockReopenQueue should be
1584 * passed back in for subsequent calls that are intended to be of the same
1587 * bs is the BlockDriverState to add to the reopen queue.
1589 * flags contains the open flags for the associated bs
1591 * returns a pointer to bs_queue, which is either the newly allocated
1592 * bs_queue, or the existing bs_queue being used.
1595 BlockReopenQueue
*bdrv_reopen_queue(BlockReopenQueue
*bs_queue
,
1596 BlockDriverState
*bs
, int flags
)
1600 BlockReopenQueueEntry
*bs_entry
;
1601 if (bs_queue
== NULL
) {
1602 bs_queue
= g_new0(BlockReopenQueue
, 1);
1603 QSIMPLEQ_INIT(bs_queue
);
1606 /* bdrv_open() masks this flag out */
1607 flags
&= ~BDRV_O_PROTOCOL
;
1610 bdrv_reopen_queue(bs_queue
, bs
->file
, bdrv_inherited_flags(flags
));
1613 bs_entry
= g_new0(BlockReopenQueueEntry
, 1);
1614 QSIMPLEQ_INSERT_TAIL(bs_queue
, bs_entry
, entry
);
1616 bs_entry
->state
.bs
= bs
;
1617 bs_entry
->state
.flags
= flags
;
1623 * Reopen multiple BlockDriverStates atomically & transactionally.
1625 * The queue passed in (bs_queue) must have been built up previous
1626 * via bdrv_reopen_queue().
1628 * Reopens all BDS specified in the queue, with the appropriate
1629 * flags. All devices are prepared for reopen, and failure of any
1630 * device will cause all device changes to be abandonded, and intermediate
1633 * If all devices prepare successfully, then the changes are committed
1637 int bdrv_reopen_multiple(BlockReopenQueue
*bs_queue
, Error
**errp
)
1640 BlockReopenQueueEntry
*bs_entry
, *next
;
1641 Error
*local_err
= NULL
;
1643 assert(bs_queue
!= NULL
);
1647 QSIMPLEQ_FOREACH(bs_entry
, bs_queue
, entry
) {
1648 if (bdrv_reopen_prepare(&bs_entry
->state
, bs_queue
, &local_err
)) {
1649 error_propagate(errp
, local_err
);
1652 bs_entry
->prepared
= true;
1655 /* If we reach this point, we have success and just need to apply the
1658 QSIMPLEQ_FOREACH(bs_entry
, bs_queue
, entry
) {
1659 bdrv_reopen_commit(&bs_entry
->state
);
1665 QSIMPLEQ_FOREACH_SAFE(bs_entry
, bs_queue
, entry
, next
) {
1666 if (ret
&& bs_entry
->prepared
) {
1667 bdrv_reopen_abort(&bs_entry
->state
);
1676 /* Reopen a single BlockDriverState with the specified flags. */
1677 int bdrv_reopen(BlockDriverState
*bs
, int bdrv_flags
, Error
**errp
)
1680 Error
*local_err
= NULL
;
1681 BlockReopenQueue
*queue
= bdrv_reopen_queue(NULL
, bs
, bdrv_flags
);
1683 ret
= bdrv_reopen_multiple(queue
, &local_err
);
1684 if (local_err
!= NULL
) {
1685 error_propagate(errp
, local_err
);
1692 * Prepares a BlockDriverState for reopen. All changes are staged in the
1693 * 'opaque' field of the BDRVReopenState, which is used and allocated by
1694 * the block driver layer .bdrv_reopen_prepare()
1696 * bs is the BlockDriverState to reopen
1697 * flags are the new open flags
1698 * queue is the reopen queue
1700 * Returns 0 on success, non-zero on error. On error errp will be set
1703 * On failure, bdrv_reopen_abort() will be called to clean up any data.
1704 * It is the responsibility of the caller to then call the abort() or
1705 * commit() for any other BDS that have been left in a prepare() state
1708 int bdrv_reopen_prepare(BDRVReopenState
*reopen_state
, BlockReopenQueue
*queue
,
1712 Error
*local_err
= NULL
;
1715 assert(reopen_state
!= NULL
);
1716 assert(reopen_state
->bs
->drv
!= NULL
);
1717 drv
= reopen_state
->bs
->drv
;
1719 /* if we are to stay read-only, do not allow permission change
1721 if (!(reopen_state
->bs
->open_flags
& BDRV_O_ALLOW_RDWR
) &&
1722 reopen_state
->flags
& BDRV_O_RDWR
) {
1723 error_set(errp
, QERR_DEVICE_IS_READ_ONLY
,
1724 reopen_state
->bs
->device_name
);
1729 ret
= bdrv_flush(reopen_state
->bs
);
1731 error_set(errp
, ERROR_CLASS_GENERIC_ERROR
, "Error (%s) flushing drive",
1736 if (drv
->bdrv_reopen_prepare
) {
1737 ret
= drv
->bdrv_reopen_prepare(reopen_state
, queue
, &local_err
);
1739 if (local_err
!= NULL
) {
1740 error_propagate(errp
, local_err
);
1742 error_setg(errp
, "failed while preparing to reopen image '%s'",
1743 reopen_state
->bs
->filename
);
1748 /* It is currently mandatory to have a bdrv_reopen_prepare()
1749 * handler for each supported drv. */
1750 error_set(errp
, QERR_BLOCK_FORMAT_FEATURE_NOT_SUPPORTED
,
1751 drv
->format_name
, reopen_state
->bs
->device_name
,
1752 "reopening of file");
1764 * Takes the staged changes for the reopen from bdrv_reopen_prepare(), and
1765 * makes them final by swapping the staging BlockDriverState contents into
1766 * the active BlockDriverState contents.
1768 void bdrv_reopen_commit(BDRVReopenState
*reopen_state
)
1772 assert(reopen_state
!= NULL
);
1773 drv
= reopen_state
->bs
->drv
;
1774 assert(drv
!= NULL
);
1776 /* If there are any driver level actions to take */
1777 if (drv
->bdrv_reopen_commit
) {
1778 drv
->bdrv_reopen_commit(reopen_state
);
1781 /* set BDS specific flags now */
1782 reopen_state
->bs
->open_flags
= reopen_state
->flags
;
1783 reopen_state
->bs
->enable_write_cache
= !!(reopen_state
->flags
&
1785 reopen_state
->bs
->read_only
= !(reopen_state
->flags
& BDRV_O_RDWR
);
1787 bdrv_refresh_limits(reopen_state
->bs
);
1791 * Abort the reopen, and delete and free the staged changes in
1794 void bdrv_reopen_abort(BDRVReopenState
*reopen_state
)
1798 assert(reopen_state
!= NULL
);
1799 drv
= reopen_state
->bs
->drv
;
1800 assert(drv
!= NULL
);
1802 if (drv
->bdrv_reopen_abort
) {
1803 drv
->bdrv_reopen_abort(reopen_state
);
1808 void bdrv_close(BlockDriverState
*bs
)
1811 block_job_cancel_sync(bs
->job
);
1813 bdrv_drain_all(); /* complete I/O */
1815 bdrv_drain_all(); /* in case flush left pending I/O */
1816 notifier_list_notify(&bs
->close_notifiers
, bs
);
1819 if (bs
->backing_hd
) {
1820 BlockDriverState
*backing_hd
= bs
->backing_hd
;
1821 bdrv_set_backing_hd(bs
, NULL
);
1822 bdrv_unref(backing_hd
);
1824 bs
->drv
->bdrv_close(bs
);
1828 bs
->copy_on_read
= 0;
1829 bs
->backing_file
[0] = '\0';
1830 bs
->backing_format
[0] = '\0';
1831 bs
->total_sectors
= 0;
1836 bs
->zero_beyond_eof
= false;
1837 QDECREF(bs
->options
);
1840 if (bs
->file
!= NULL
) {
1841 bdrv_unref(bs
->file
);
1846 bdrv_dev_change_media_cb(bs
, false);
1848 /*throttling disk I/O limits*/
1849 if (bs
->io_limits_enabled
) {
1850 bdrv_io_limits_disable(bs
);
1854 void bdrv_close_all(void)
1856 BlockDriverState
*bs
;
1858 QTAILQ_FOREACH(bs
, &bdrv_states
, device_list
) {
1859 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
1861 aio_context_acquire(aio_context
);
1863 aio_context_release(aio_context
);
1867 /* Check if any requests are in-flight (including throttled requests) */
1868 static bool bdrv_requests_pending(BlockDriverState
*bs
)
1870 if (!QLIST_EMPTY(&bs
->tracked_requests
)) {
1873 if (!qemu_co_queue_empty(&bs
->throttled_reqs
[0])) {
1876 if (!qemu_co_queue_empty(&bs
->throttled_reqs
[1])) {
1879 if (bs
->file
&& bdrv_requests_pending(bs
->file
)) {
1882 if (bs
->backing_hd
&& bdrv_requests_pending(bs
->backing_hd
)) {
1889 * Wait for pending requests to complete across all BlockDriverStates
1891 * This function does not flush data to disk, use bdrv_flush_all() for that
1892 * after calling this function.
1894 * Note that completion of an asynchronous I/O operation can trigger any
1895 * number of other I/O operations on other devices---for example a coroutine
1896 * can be arbitrarily complex and a constant flow of I/O can come until the
1897 * coroutine is complete. Because of this, it is not possible to have a
1898 * function to drain a single device's I/O queue.
1900 void bdrv_drain_all(void)
1902 /* Always run first iteration so any pending completion BHs run */
1904 BlockDriverState
*bs
;
1909 QTAILQ_FOREACH(bs
, &bdrv_states
, device_list
) {
1910 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
1913 aio_context_acquire(aio_context
);
1914 bdrv_start_throttled_reqs(bs
);
1915 bs_busy
= bdrv_requests_pending(bs
);
1916 bs_busy
|= aio_poll(aio_context
, bs_busy
);
1917 aio_context_release(aio_context
);
1924 /* make a BlockDriverState anonymous by removing from bdrv_state and
1925 * graph_bdrv_state list.
1926 Also, NULL terminate the device_name to prevent double remove */
1927 void bdrv_make_anon(BlockDriverState
*bs
)
1929 if (bs
->device_name
[0] != '\0') {
1930 QTAILQ_REMOVE(&bdrv_states
, bs
, device_list
);
1932 bs
->device_name
[0] = '\0';
1933 if (bs
->node_name
[0] != '\0') {
1934 QTAILQ_REMOVE(&graph_bdrv_states
, bs
, node_list
);
1936 bs
->node_name
[0] = '\0';
1939 static void bdrv_rebind(BlockDriverState
*bs
)
1941 if (bs
->drv
&& bs
->drv
->bdrv_rebind
) {
1942 bs
->drv
->bdrv_rebind(bs
);
1946 static void bdrv_move_feature_fields(BlockDriverState
*bs_dest
,
1947 BlockDriverState
*bs_src
)
1949 /* move some fields that need to stay attached to the device */
1952 bs_dest
->dev_ops
= bs_src
->dev_ops
;
1953 bs_dest
->dev_opaque
= bs_src
->dev_opaque
;
1954 bs_dest
->dev
= bs_src
->dev
;
1955 bs_dest
->guest_block_size
= bs_src
->guest_block_size
;
1956 bs_dest
->copy_on_read
= bs_src
->copy_on_read
;
1958 bs_dest
->enable_write_cache
= bs_src
->enable_write_cache
;
1960 /* i/o throttled req */
1961 memcpy(&bs_dest
->throttle_state
,
1962 &bs_src
->throttle_state
,
1963 sizeof(ThrottleState
));
1964 bs_dest
->throttled_reqs
[0] = bs_src
->throttled_reqs
[0];
1965 bs_dest
->throttled_reqs
[1] = bs_src
->throttled_reqs
[1];
1966 bs_dest
->io_limits_enabled
= bs_src
->io_limits_enabled
;
1969 bs_dest
->on_read_error
= bs_src
->on_read_error
;
1970 bs_dest
->on_write_error
= bs_src
->on_write_error
;
1973 bs_dest
->iostatus_enabled
= bs_src
->iostatus_enabled
;
1974 bs_dest
->iostatus
= bs_src
->iostatus
;
1977 bs_dest
->dirty_bitmaps
= bs_src
->dirty_bitmaps
;
1979 /* reference count */
1980 bs_dest
->refcnt
= bs_src
->refcnt
;
1983 bs_dest
->job
= bs_src
->job
;
1985 /* keep the same entry in bdrv_states */
1986 pstrcpy(bs_dest
->device_name
, sizeof(bs_dest
->device_name
),
1987 bs_src
->device_name
);
1988 bs_dest
->device_list
= bs_src
->device_list
;
1989 memcpy(bs_dest
->op_blockers
, bs_src
->op_blockers
,
1990 sizeof(bs_dest
->op_blockers
));
1994 * Swap bs contents for two image chains while they are live,
1995 * while keeping required fields on the BlockDriverState that is
1996 * actually attached to a device.
1998 * This will modify the BlockDriverState fields, and swap contents
1999 * between bs_new and bs_old. Both bs_new and bs_old are modified.
2001 * bs_new is required to be anonymous.
2003 * This function does not create any image files.
2005 void bdrv_swap(BlockDriverState
*bs_new
, BlockDriverState
*bs_old
)
2007 BlockDriverState tmp
;
2009 /* The code needs to swap the node_name but simply swapping node_list won't
2010 * work so first remove the nodes from the graph list, do the swap then
2011 * insert them back if needed.
2013 if (bs_new
->node_name
[0] != '\0') {
2014 QTAILQ_REMOVE(&graph_bdrv_states
, bs_new
, node_list
);
2016 if (bs_old
->node_name
[0] != '\0') {
2017 QTAILQ_REMOVE(&graph_bdrv_states
, bs_old
, node_list
);
2020 /* bs_new must be anonymous and shouldn't have anything fancy enabled */
2021 assert(bs_new
->device_name
[0] == '\0');
2022 assert(QLIST_EMPTY(&bs_new
->dirty_bitmaps
));
2023 assert(bs_new
->job
== NULL
);
2024 assert(bs_new
->dev
== NULL
);
2025 assert(bs_new
->io_limits_enabled
== false);
2026 assert(!throttle_have_timer(&bs_new
->throttle_state
));
2032 /* there are some fields that should not be swapped, move them back */
2033 bdrv_move_feature_fields(&tmp
, bs_old
);
2034 bdrv_move_feature_fields(bs_old
, bs_new
);
2035 bdrv_move_feature_fields(bs_new
, &tmp
);
2037 /* bs_new shouldn't be in bdrv_states even after the swap! */
2038 assert(bs_new
->device_name
[0] == '\0');
2040 /* Check a few fields that should remain attached to the device */
2041 assert(bs_new
->dev
== NULL
);
2042 assert(bs_new
->job
== NULL
);
2043 assert(bs_new
->io_limits_enabled
== false);
2044 assert(!throttle_have_timer(&bs_new
->throttle_state
));
2046 /* insert the nodes back into the graph node list if needed */
2047 if (bs_new
->node_name
[0] != '\0') {
2048 QTAILQ_INSERT_TAIL(&graph_bdrv_states
, bs_new
, node_list
);
2050 if (bs_old
->node_name
[0] != '\0') {
2051 QTAILQ_INSERT_TAIL(&graph_bdrv_states
, bs_old
, node_list
);
2054 bdrv_rebind(bs_new
);
2055 bdrv_rebind(bs_old
);
2059 * Add new bs contents at the top of an image chain while the chain is
2060 * live, while keeping required fields on the top layer.
2062 * This will modify the BlockDriverState fields, and swap contents
2063 * between bs_new and bs_top. Both bs_new and bs_top are modified.
2065 * bs_new is required to be anonymous.
2067 * This function does not create any image files.
2069 void bdrv_append(BlockDriverState
*bs_new
, BlockDriverState
*bs_top
)
2071 bdrv_swap(bs_new
, bs_top
);
2073 /* The contents of 'tmp' will become bs_top, as we are
2074 * swapping bs_new and bs_top contents. */
2075 bdrv_set_backing_hd(bs_top
, bs_new
);
2078 static void bdrv_delete(BlockDriverState
*bs
)
2082 assert(bdrv_op_blocker_is_empty(bs
));
2083 assert(!bs
->refcnt
);
2084 assert(QLIST_EMPTY(&bs
->dirty_bitmaps
));
2088 /* remove from list, if necessary */
2094 int bdrv_attach_dev(BlockDriverState
*bs
, void *dev
)
2095 /* TODO change to DeviceState *dev when all users are qdevified */
2101 bdrv_iostatus_reset(bs
);
2105 /* TODO qdevified devices don't use this, remove when devices are qdevified */
2106 void bdrv_attach_dev_nofail(BlockDriverState
*bs
, void *dev
)
2108 if (bdrv_attach_dev(bs
, dev
) < 0) {
2113 void bdrv_detach_dev(BlockDriverState
*bs
, void *dev
)
2114 /* TODO change to DeviceState *dev when all users are qdevified */
2116 assert(bs
->dev
== dev
);
2119 bs
->dev_opaque
= NULL
;
2120 bs
->guest_block_size
= 512;
2123 /* TODO change to return DeviceState * when all users are qdevified */
2124 void *bdrv_get_attached_dev(BlockDriverState
*bs
)
2129 void bdrv_set_dev_ops(BlockDriverState
*bs
, const BlockDevOps
*ops
,
2133 bs
->dev_opaque
= opaque
;
2136 void bdrv_emit_qmp_error_event(const BlockDriverState
*bdrv
,
2137 enum MonitorEvent ev
,
2138 BlockErrorAction action
, bool is_read
)
2141 const char *action_str
;
2144 case BDRV_ACTION_REPORT
:
2145 action_str
= "report";
2147 case BDRV_ACTION_IGNORE
:
2148 action_str
= "ignore";
2150 case BDRV_ACTION_STOP
:
2151 action_str
= "stop";
2157 data
= qobject_from_jsonf("{ 'device': %s, 'action': %s, 'operation': %s }",
2160 is_read
? "read" : "write");
2161 monitor_protocol_event(ev
, data
);
2163 qobject_decref(data
);
2166 static void bdrv_emit_qmp_eject_event(BlockDriverState
*bs
, bool ejected
)
2170 data
= qobject_from_jsonf("{ 'device': %s, 'tray-open': %i }",
2171 bdrv_get_device_name(bs
), ejected
);
2172 monitor_protocol_event(QEVENT_DEVICE_TRAY_MOVED
, data
);
2174 qobject_decref(data
);
2177 static void bdrv_dev_change_media_cb(BlockDriverState
*bs
, bool load
)
2179 if (bs
->dev_ops
&& bs
->dev_ops
->change_media_cb
) {
2180 bool tray_was_closed
= !bdrv_dev_is_tray_open(bs
);
2181 bs
->dev_ops
->change_media_cb(bs
->dev_opaque
, load
);
2182 if (tray_was_closed
) {
2184 bdrv_emit_qmp_eject_event(bs
, true);
2188 bdrv_emit_qmp_eject_event(bs
, false);
2193 bool bdrv_dev_has_removable_media(BlockDriverState
*bs
)
2195 return !bs
->dev
|| (bs
->dev_ops
&& bs
->dev_ops
->change_media_cb
);
2198 void bdrv_dev_eject_request(BlockDriverState
*bs
, bool force
)
2200 if (bs
->dev_ops
&& bs
->dev_ops
->eject_request_cb
) {
2201 bs
->dev_ops
->eject_request_cb(bs
->dev_opaque
, force
);
2205 bool bdrv_dev_is_tray_open(BlockDriverState
*bs
)
2207 if (bs
->dev_ops
&& bs
->dev_ops
->is_tray_open
) {
2208 return bs
->dev_ops
->is_tray_open(bs
->dev_opaque
);
2213 static void bdrv_dev_resize_cb(BlockDriverState
*bs
)
2215 if (bs
->dev_ops
&& bs
->dev_ops
->resize_cb
) {
2216 bs
->dev_ops
->resize_cb(bs
->dev_opaque
);
2220 bool bdrv_dev_is_medium_locked(BlockDriverState
*bs
)
2222 if (bs
->dev_ops
&& bs
->dev_ops
->is_medium_locked
) {
2223 return bs
->dev_ops
->is_medium_locked(bs
->dev_opaque
);
2229 * Run consistency checks on an image
2231 * Returns 0 if the check could be completed (it doesn't mean that the image is
2232 * free of errors) or -errno when an internal error occurred. The results of the
2233 * check are stored in res.
2235 int bdrv_check(BlockDriverState
*bs
, BdrvCheckResult
*res
, BdrvCheckMode fix
)
2237 if (bs
->drv
->bdrv_check
== NULL
) {
2241 memset(res
, 0, sizeof(*res
));
2242 return bs
->drv
->bdrv_check(bs
, res
, fix
);
2245 #define COMMIT_BUF_SECTORS 2048
2247 /* commit COW file into the raw image */
2248 int bdrv_commit(BlockDriverState
*bs
)
2250 BlockDriver
*drv
= bs
->drv
;
2251 int64_t sector
, total_sectors
, length
, backing_length
;
2252 int n
, ro
, open_flags
;
2254 uint8_t *buf
= NULL
;
2255 char filename
[PATH_MAX
];
2260 if (!bs
->backing_hd
) {
2264 if (bdrv_op_is_blocked(bs
, BLOCK_OP_TYPE_COMMIT
, NULL
) ||
2265 bdrv_op_is_blocked(bs
->backing_hd
, BLOCK_OP_TYPE_COMMIT
, NULL
)) {
2269 ro
= bs
->backing_hd
->read_only
;
2270 /* Use pstrcpy (not strncpy): filename must be NUL-terminated. */
2271 pstrcpy(filename
, sizeof(filename
), bs
->backing_hd
->filename
);
2272 open_flags
= bs
->backing_hd
->open_flags
;
2275 if (bdrv_reopen(bs
->backing_hd
, open_flags
| BDRV_O_RDWR
, NULL
)) {
2280 length
= bdrv_getlength(bs
);
2286 backing_length
= bdrv_getlength(bs
->backing_hd
);
2287 if (backing_length
< 0) {
2288 ret
= backing_length
;
2292 /* If our top snapshot is larger than the backing file image,
2293 * grow the backing file image if possible. If not possible,
2294 * we must return an error */
2295 if (length
> backing_length
) {
2296 ret
= bdrv_truncate(bs
->backing_hd
, length
);
2302 total_sectors
= length
>> BDRV_SECTOR_BITS
;
2303 buf
= g_malloc(COMMIT_BUF_SECTORS
* BDRV_SECTOR_SIZE
);
2305 for (sector
= 0; sector
< total_sectors
; sector
+= n
) {
2306 ret
= bdrv_is_allocated(bs
, sector
, COMMIT_BUF_SECTORS
, &n
);
2311 ret
= bdrv_read(bs
, sector
, buf
, n
);
2316 ret
= bdrv_write(bs
->backing_hd
, sector
, buf
, n
);
2323 if (drv
->bdrv_make_empty
) {
2324 ret
= drv
->bdrv_make_empty(bs
);
2332 * Make sure all data we wrote to the backing device is actually
2335 if (bs
->backing_hd
) {
2336 bdrv_flush(bs
->backing_hd
);
2344 /* ignoring error return here */
2345 bdrv_reopen(bs
->backing_hd
, open_flags
& ~BDRV_O_RDWR
, NULL
);
2351 int bdrv_commit_all(void)
2353 BlockDriverState
*bs
;
2355 QTAILQ_FOREACH(bs
, &bdrv_states
, device_list
) {
2356 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
2358 aio_context_acquire(aio_context
);
2359 if (bs
->drv
&& bs
->backing_hd
) {
2360 int ret
= bdrv_commit(bs
);
2362 aio_context_release(aio_context
);
2366 aio_context_release(aio_context
);
2372 * Remove an active request from the tracked requests list
2374 * This function should be called when a tracked request is completing.
2376 static void tracked_request_end(BdrvTrackedRequest
*req
)
2378 if (req
->serialising
) {
2379 req
->bs
->serialising_in_flight
--;
2382 QLIST_REMOVE(req
, list
);
2383 qemu_co_queue_restart_all(&req
->wait_queue
);
2387 * Add an active request to the tracked requests list
2389 static void tracked_request_begin(BdrvTrackedRequest
*req
,
2390 BlockDriverState
*bs
,
2392 unsigned int bytes
, bool is_write
)
2394 *req
= (BdrvTrackedRequest
){
2398 .is_write
= is_write
,
2399 .co
= qemu_coroutine_self(),
2400 .serialising
= false,
2401 .overlap_offset
= offset
,
2402 .overlap_bytes
= bytes
,
2405 qemu_co_queue_init(&req
->wait_queue
);
2407 QLIST_INSERT_HEAD(&bs
->tracked_requests
, req
, list
);
2410 static void mark_request_serialising(BdrvTrackedRequest
*req
, uint64_t align
)
2412 int64_t overlap_offset
= req
->offset
& ~(align
- 1);
2413 unsigned int overlap_bytes
= ROUND_UP(req
->offset
+ req
->bytes
, align
)
2416 if (!req
->serialising
) {
2417 req
->bs
->serialising_in_flight
++;
2418 req
->serialising
= true;
2421 req
->overlap_offset
= MIN(req
->overlap_offset
, overlap_offset
);
2422 req
->overlap_bytes
= MAX(req
->overlap_bytes
, overlap_bytes
);
2426 * Round a region to cluster boundaries
2428 void bdrv_round_to_clusters(BlockDriverState
*bs
,
2429 int64_t sector_num
, int nb_sectors
,
2430 int64_t *cluster_sector_num
,
2431 int *cluster_nb_sectors
)
2433 BlockDriverInfo bdi
;
2435 if (bdrv_get_info(bs
, &bdi
) < 0 || bdi
.cluster_size
== 0) {
2436 *cluster_sector_num
= sector_num
;
2437 *cluster_nb_sectors
= nb_sectors
;
2439 int64_t c
= bdi
.cluster_size
/ BDRV_SECTOR_SIZE
;
2440 *cluster_sector_num
= QEMU_ALIGN_DOWN(sector_num
, c
);
2441 *cluster_nb_sectors
= QEMU_ALIGN_UP(sector_num
- *cluster_sector_num
+
2446 static int bdrv_get_cluster_size(BlockDriverState
*bs
)
2448 BlockDriverInfo bdi
;
2451 ret
= bdrv_get_info(bs
, &bdi
);
2452 if (ret
< 0 || bdi
.cluster_size
== 0) {
2453 return bs
->request_alignment
;
2455 return bdi
.cluster_size
;
2459 static bool tracked_request_overlaps(BdrvTrackedRequest
*req
,
2460 int64_t offset
, unsigned int bytes
)
2463 if (offset
>= req
->overlap_offset
+ req
->overlap_bytes
) {
2467 if (req
->overlap_offset
>= offset
+ bytes
) {
2473 static bool coroutine_fn
wait_serialising_requests(BdrvTrackedRequest
*self
)
2475 BlockDriverState
*bs
= self
->bs
;
2476 BdrvTrackedRequest
*req
;
2478 bool waited
= false;
2480 if (!bs
->serialising_in_flight
) {
2486 QLIST_FOREACH(req
, &bs
->tracked_requests
, list
) {
2487 if (req
== self
|| (!req
->serialising
&& !self
->serialising
)) {
2490 if (tracked_request_overlaps(req
, self
->overlap_offset
,
2491 self
->overlap_bytes
))
2493 /* Hitting this means there was a reentrant request, for
2494 * example, a block driver issuing nested requests. This must
2495 * never happen since it means deadlock.
2497 assert(qemu_coroutine_self() != req
->co
);
2499 /* If the request is already (indirectly) waiting for us, or
2500 * will wait for us as soon as it wakes up, then just go on
2501 * (instead of producing a deadlock in the former case). */
2502 if (!req
->waiting_for
) {
2503 self
->waiting_for
= req
;
2504 qemu_co_queue_wait(&req
->wait_queue
);
2505 self
->waiting_for
= NULL
;
2520 * -EINVAL - backing format specified, but no file
2521 * -ENOSPC - can't update the backing file because no space is left in the
2523 * -ENOTSUP - format driver doesn't support changing the backing file
2525 int bdrv_change_backing_file(BlockDriverState
*bs
,
2526 const char *backing_file
, const char *backing_fmt
)
2528 BlockDriver
*drv
= bs
->drv
;
2531 /* Backing file format doesn't make sense without a backing file */
2532 if (backing_fmt
&& !backing_file
) {
2536 if (drv
->bdrv_change_backing_file
!= NULL
) {
2537 ret
= drv
->bdrv_change_backing_file(bs
, backing_file
, backing_fmt
);
2543 pstrcpy(bs
->backing_file
, sizeof(bs
->backing_file
), backing_file
?: "");
2544 pstrcpy(bs
->backing_format
, sizeof(bs
->backing_format
), backing_fmt
?: "");
2550 * Finds the image layer in the chain that has 'bs' as its backing file.
2552 * active is the current topmost image.
2554 * Returns NULL if bs is not found in active's image chain,
2555 * or if active == bs.
2557 BlockDriverState
*bdrv_find_overlay(BlockDriverState
*active
,
2558 BlockDriverState
*bs
)
2560 BlockDriverState
*overlay
= NULL
;
2561 BlockDriverState
*intermediate
;
2563 assert(active
!= NULL
);
2566 /* if bs is the same as active, then by definition it has no overlay
2572 intermediate
= active
;
2573 while (intermediate
->backing_hd
) {
2574 if (intermediate
->backing_hd
== bs
) {
2575 overlay
= intermediate
;
2578 intermediate
= intermediate
->backing_hd
;
2584 typedef struct BlkIntermediateStates
{
2585 BlockDriverState
*bs
;
2586 QSIMPLEQ_ENTRY(BlkIntermediateStates
) entry
;
2587 } BlkIntermediateStates
;
2591 * Drops images above 'base' up to and including 'top', and sets the image
2592 * above 'top' to have base as its backing file.
2594 * Requires that the overlay to 'top' is opened r/w, so that the backing file
2595 * information in 'bs' can be properly updated.
2597 * E.g., this will convert the following chain:
2598 * bottom <- base <- intermediate <- top <- active
2602 * bottom <- base <- active
2604 * It is allowed for bottom==base, in which case it converts:
2606 * base <- intermediate <- top <- active
2613 * if active == top, that is considered an error
2616 int bdrv_drop_intermediate(BlockDriverState
*active
, BlockDriverState
*top
,
2617 BlockDriverState
*base
)
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_malloc0(sizeof(BlkIntermediateStates
));
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 ret
= bdrv_change_backing_file(new_top_bs
, base_bs
->filename
,
2670 base_bs
->drv
? base_bs
->drv
->format_name
: "");
2674 bdrv_set_backing_hd(new_top_bs
, base_bs
);
2676 QSIMPLEQ_FOREACH_SAFE(intermediate_state
, &states_to_delete
, entry
, next
) {
2677 /* so that bdrv_close() does not recursively close the chain */
2678 bdrv_set_backing_hd(intermediate_state
->bs
, NULL
);
2679 bdrv_unref(intermediate_state
->bs
);
2684 QSIMPLEQ_FOREACH_SAFE(intermediate_state
, &states_to_delete
, entry
, next
) {
2685 g_free(intermediate_state
);
2691 static int bdrv_check_byte_request(BlockDriverState
*bs
, int64_t offset
,
2696 if (size
> INT_MAX
) {
2700 if (!bdrv_is_inserted(bs
))
2706 len
= bdrv_getlength(bs
);
2711 if ((offset
> len
) || (len
- offset
< size
))
2717 static int bdrv_check_request(BlockDriverState
*bs
, int64_t sector_num
,
2720 if (nb_sectors
< 0 || nb_sectors
> INT_MAX
/ BDRV_SECTOR_SIZE
) {
2724 return bdrv_check_byte_request(bs
, sector_num
* BDRV_SECTOR_SIZE
,
2725 nb_sectors
* BDRV_SECTOR_SIZE
);
2728 typedef struct RwCo
{
2729 BlockDriverState
*bs
;
2734 BdrvRequestFlags flags
;
2737 static void coroutine_fn
bdrv_rw_co_entry(void *opaque
)
2739 RwCo
*rwco
= opaque
;
2741 if (!rwco
->is_write
) {
2742 rwco
->ret
= bdrv_co_do_preadv(rwco
->bs
, rwco
->offset
,
2743 rwco
->qiov
->size
, rwco
->qiov
,
2746 rwco
->ret
= bdrv_co_do_pwritev(rwco
->bs
, rwco
->offset
,
2747 rwco
->qiov
->size
, rwco
->qiov
,
2753 * Process a vectored synchronous request using coroutines
2755 static int bdrv_prwv_co(BlockDriverState
*bs
, int64_t offset
,
2756 QEMUIOVector
*qiov
, bool is_write
,
2757 BdrvRequestFlags flags
)
2764 .is_write
= is_write
,
2770 * In sync call context, when the vcpu is blocked, this throttling timer
2771 * will not fire; so the I/O throttling function has to be disabled here
2772 * if it has been enabled.
2774 if (bs
->io_limits_enabled
) {
2775 fprintf(stderr
, "Disabling I/O throttling on '%s' due "
2776 "to synchronous I/O.\n", bdrv_get_device_name(bs
));
2777 bdrv_io_limits_disable(bs
);
2780 if (qemu_in_coroutine()) {
2781 /* Fast-path if already in coroutine context */
2782 bdrv_rw_co_entry(&rwco
);
2784 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
2786 co
= qemu_coroutine_create(bdrv_rw_co_entry
);
2787 qemu_coroutine_enter(co
, &rwco
);
2788 while (rwco
.ret
== NOT_DONE
) {
2789 aio_poll(aio_context
, true);
2796 * Process a synchronous request using coroutines
2798 static int bdrv_rw_co(BlockDriverState
*bs
, int64_t sector_num
, uint8_t *buf
,
2799 int nb_sectors
, bool is_write
, BdrvRequestFlags flags
)
2802 struct iovec iov
= {
2803 .iov_base
= (void *)buf
,
2804 .iov_len
= nb_sectors
* BDRV_SECTOR_SIZE
,
2807 if (nb_sectors
< 0 || nb_sectors
> INT_MAX
/ BDRV_SECTOR_SIZE
) {
2811 qemu_iovec_init_external(&qiov
, &iov
, 1);
2812 return bdrv_prwv_co(bs
, sector_num
<< BDRV_SECTOR_BITS
,
2813 &qiov
, is_write
, flags
);
2816 /* return < 0 if error. See bdrv_write() for the return codes */
2817 int bdrv_read(BlockDriverState
*bs
, int64_t sector_num
,
2818 uint8_t *buf
, int nb_sectors
)
2820 return bdrv_rw_co(bs
, sector_num
, buf
, nb_sectors
, false, 0);
2823 /* Just like bdrv_read(), but with I/O throttling temporarily disabled */
2824 int bdrv_read_unthrottled(BlockDriverState
*bs
, int64_t sector_num
,
2825 uint8_t *buf
, int nb_sectors
)
2830 enabled
= bs
->io_limits_enabled
;
2831 bs
->io_limits_enabled
= false;
2832 ret
= bdrv_read(bs
, sector_num
, buf
, nb_sectors
);
2833 bs
->io_limits_enabled
= enabled
;
2837 /* Return < 0 if error. Important errors are:
2838 -EIO generic I/O error (may happen for all errors)
2839 -ENOMEDIUM No media inserted.
2840 -EINVAL Invalid sector number or nb_sectors
2841 -EACCES Trying to write a read-only device
2843 int bdrv_write(BlockDriverState
*bs
, int64_t sector_num
,
2844 const uint8_t *buf
, int nb_sectors
)
2846 return bdrv_rw_co(bs
, sector_num
, (uint8_t *)buf
, nb_sectors
, true, 0);
2849 int bdrv_write_zeroes(BlockDriverState
*bs
, int64_t sector_num
,
2850 int nb_sectors
, BdrvRequestFlags flags
)
2852 return bdrv_rw_co(bs
, sector_num
, NULL
, nb_sectors
, true,
2853 BDRV_REQ_ZERO_WRITE
| flags
);
2857 * Completely zero out a block device with the help of bdrv_write_zeroes.
2858 * The operation is sped up by checking the block status and only writing
2859 * zeroes to the device if they currently do not return zeroes. Optional
2860 * flags are passed through to bdrv_write_zeroes (e.g. BDRV_REQ_MAY_UNMAP).
2862 * Returns < 0 on error, 0 on success. For error codes see bdrv_write().
2864 int bdrv_make_zero(BlockDriverState
*bs
, BdrvRequestFlags flags
)
2866 int64_t target_size
;
2867 int64_t ret
, nb_sectors
, sector_num
= 0;
2870 target_size
= bdrv_getlength(bs
);
2871 if (target_size
< 0) {
2874 target_size
/= BDRV_SECTOR_SIZE
;
2877 nb_sectors
= target_size
- sector_num
;
2878 if (nb_sectors
<= 0) {
2881 if (nb_sectors
> INT_MAX
) {
2882 nb_sectors
= INT_MAX
;
2884 ret
= bdrv_get_block_status(bs
, sector_num
, nb_sectors
, &n
);
2886 error_report("error getting block status at sector %" PRId64
": %s",
2887 sector_num
, strerror(-ret
));
2890 if (ret
& BDRV_BLOCK_ZERO
) {
2894 ret
= bdrv_write_zeroes(bs
, sector_num
, n
, flags
);
2896 error_report("error writing zeroes at sector %" PRId64
": %s",
2897 sector_num
, strerror(-ret
));
2904 int bdrv_pread(BlockDriverState
*bs
, int64_t offset
, void *buf
, int bytes
)
2907 struct iovec iov
= {
2908 .iov_base
= (void *)buf
,
2917 qemu_iovec_init_external(&qiov
, &iov
, 1);
2918 ret
= bdrv_prwv_co(bs
, offset
, &qiov
, false, 0);
2926 int bdrv_pwritev(BlockDriverState
*bs
, int64_t offset
, QEMUIOVector
*qiov
)
2930 ret
= bdrv_prwv_co(bs
, offset
, qiov
, true, 0);
2938 int bdrv_pwrite(BlockDriverState
*bs
, int64_t offset
,
2939 const void *buf
, int bytes
)
2942 struct iovec iov
= {
2943 .iov_base
= (void *) buf
,
2951 qemu_iovec_init_external(&qiov
, &iov
, 1);
2952 return bdrv_pwritev(bs
, offset
, &qiov
);
2956 * Writes to the file and ensures that no writes are reordered across this
2957 * request (acts as a barrier)
2959 * Returns 0 on success, -errno in error cases.
2961 int bdrv_pwrite_sync(BlockDriverState
*bs
, int64_t offset
,
2962 const void *buf
, int count
)
2966 ret
= bdrv_pwrite(bs
, offset
, buf
, count
);
2971 /* No flush needed for cache modes that already do it */
2972 if (bs
->enable_write_cache
) {
2979 static int coroutine_fn
bdrv_co_do_copy_on_readv(BlockDriverState
*bs
,
2980 int64_t sector_num
, int nb_sectors
, QEMUIOVector
*qiov
)
2982 /* Perform I/O through a temporary buffer so that users who scribble over
2983 * their read buffer while the operation is in progress do not end up
2984 * modifying the image file. This is critical for zero-copy guest I/O
2985 * where anything might happen inside guest memory.
2987 void *bounce_buffer
;
2989 BlockDriver
*drv
= bs
->drv
;
2991 QEMUIOVector bounce_qiov
;
2992 int64_t cluster_sector_num
;
2993 int cluster_nb_sectors
;
2997 /* Cover entire cluster so no additional backing file I/O is required when
2998 * allocating cluster in the image file.
3000 bdrv_round_to_clusters(bs
, sector_num
, nb_sectors
,
3001 &cluster_sector_num
, &cluster_nb_sectors
);
3003 trace_bdrv_co_do_copy_on_readv(bs
, sector_num
, nb_sectors
,
3004 cluster_sector_num
, cluster_nb_sectors
);
3006 iov
.iov_len
= cluster_nb_sectors
* BDRV_SECTOR_SIZE
;
3007 iov
.iov_base
= bounce_buffer
= qemu_blockalign(bs
, iov
.iov_len
);
3008 qemu_iovec_init_external(&bounce_qiov
, &iov
, 1);
3010 ret
= drv
->bdrv_co_readv(bs
, cluster_sector_num
, cluster_nb_sectors
,
3016 if (drv
->bdrv_co_write_zeroes
&&
3017 buffer_is_zero(bounce_buffer
, iov
.iov_len
)) {
3018 ret
= bdrv_co_do_write_zeroes(bs
, cluster_sector_num
,
3019 cluster_nb_sectors
, 0);
3021 /* This does not change the data on the disk, it is not necessary
3022 * to flush even in cache=writethrough mode.
3024 ret
= drv
->bdrv_co_writev(bs
, cluster_sector_num
, cluster_nb_sectors
,
3029 /* It might be okay to ignore write errors for guest requests. If this
3030 * is a deliberate copy-on-read then we don't want to ignore the error.
3031 * Simply report it in all cases.
3036 skip_bytes
= (sector_num
- cluster_sector_num
) * BDRV_SECTOR_SIZE
;
3037 qemu_iovec_from_buf(qiov
, 0, bounce_buffer
+ skip_bytes
,
3038 nb_sectors
* BDRV_SECTOR_SIZE
);
3041 qemu_vfree(bounce_buffer
);
3046 * Forwards an already correctly aligned request to the BlockDriver. This
3047 * handles copy on read and zeroing after EOF; any other features must be
3048 * implemented by the caller.
3050 static int coroutine_fn
bdrv_aligned_preadv(BlockDriverState
*bs
,
3051 BdrvTrackedRequest
*req
, int64_t offset
, unsigned int bytes
,
3052 int64_t align
, QEMUIOVector
*qiov
, int flags
)
3054 BlockDriver
*drv
= bs
->drv
;
3057 int64_t sector_num
= offset
>> BDRV_SECTOR_BITS
;
3058 unsigned int nb_sectors
= bytes
>> BDRV_SECTOR_BITS
;
3060 assert((offset
& (BDRV_SECTOR_SIZE
- 1)) == 0);
3061 assert((bytes
& (BDRV_SECTOR_SIZE
- 1)) == 0);
3063 /* Handle Copy on Read and associated serialisation */
3064 if (flags
& BDRV_REQ_COPY_ON_READ
) {
3065 /* If we touch the same cluster it counts as an overlap. This
3066 * guarantees that allocating writes will be serialized and not race
3067 * with each other for the same cluster. For example, in copy-on-read
3068 * it ensures that the CoR read and write operations are atomic and
3069 * guest writes cannot interleave between them. */
3070 mark_request_serialising(req
, bdrv_get_cluster_size(bs
));
3073 wait_serialising_requests(req
);
3075 if (flags
& BDRV_REQ_COPY_ON_READ
) {
3078 ret
= bdrv_is_allocated(bs
, sector_num
, nb_sectors
, &pnum
);
3083 if (!ret
|| pnum
!= nb_sectors
) {
3084 ret
= bdrv_co_do_copy_on_readv(bs
, sector_num
, nb_sectors
, qiov
);
3089 /* Forward the request to the BlockDriver */
3090 if (!(bs
->zero_beyond_eof
&& bs
->growable
)) {
3091 ret
= drv
->bdrv_co_readv(bs
, sector_num
, nb_sectors
, qiov
);
3093 /* Read zeros after EOF of growable BDSes */
3094 int64_t len
, total_sectors
, max_nb_sectors
;
3096 len
= bdrv_getlength(bs
);
3102 total_sectors
= DIV_ROUND_UP(len
, BDRV_SECTOR_SIZE
);
3103 max_nb_sectors
= ROUND_UP(MAX(0, total_sectors
- sector_num
),
3104 align
>> BDRV_SECTOR_BITS
);
3105 if (max_nb_sectors
> 0) {
3106 ret
= drv
->bdrv_co_readv(bs
, sector_num
,
3107 MIN(nb_sectors
, max_nb_sectors
), qiov
);
3112 /* Reading beyond end of file is supposed to produce zeroes */
3113 if (ret
== 0 && total_sectors
< sector_num
+ nb_sectors
) {
3114 uint64_t offset
= MAX(0, total_sectors
- sector_num
);
3115 uint64_t bytes
= (sector_num
+ nb_sectors
- offset
) *
3117 qemu_iovec_memset(qiov
, offset
* BDRV_SECTOR_SIZE
, 0, bytes
);
3126 * Handle a read request in coroutine context
3128 static int coroutine_fn
bdrv_co_do_preadv(BlockDriverState
*bs
,
3129 int64_t offset
, unsigned int bytes
, QEMUIOVector
*qiov
,
3130 BdrvRequestFlags flags
)
3132 BlockDriver
*drv
= bs
->drv
;
3133 BdrvTrackedRequest req
;
3135 /* TODO Lift BDRV_SECTOR_SIZE restriction in BlockDriver interface */
3136 uint64_t align
= MAX(BDRV_SECTOR_SIZE
, bs
->request_alignment
);
3137 uint8_t *head_buf
= NULL
;
3138 uint8_t *tail_buf
= NULL
;
3139 QEMUIOVector local_qiov
;
3140 bool use_local_qiov
= false;
3146 if (bdrv_check_byte_request(bs
, offset
, bytes
)) {
3150 if (bs
->copy_on_read
) {
3151 flags
|= BDRV_REQ_COPY_ON_READ
;
3154 /* throttling disk I/O */
3155 if (bs
->io_limits_enabled
) {
3156 bdrv_io_limits_intercept(bs
, bytes
, false);
3159 /* Align read if necessary by padding qiov */
3160 if (offset
& (align
- 1)) {
3161 head_buf
= qemu_blockalign(bs
, align
);
3162 qemu_iovec_init(&local_qiov
, qiov
->niov
+ 2);
3163 qemu_iovec_add(&local_qiov
, head_buf
, offset
& (align
- 1));
3164 qemu_iovec_concat(&local_qiov
, qiov
, 0, qiov
->size
);
3165 use_local_qiov
= true;
3167 bytes
+= offset
& (align
- 1);
3168 offset
= offset
& ~(align
- 1);
3171 if ((offset
+ bytes
) & (align
- 1)) {
3172 if (!use_local_qiov
) {
3173 qemu_iovec_init(&local_qiov
, qiov
->niov
+ 1);
3174 qemu_iovec_concat(&local_qiov
, qiov
, 0, qiov
->size
);
3175 use_local_qiov
= true;
3177 tail_buf
= qemu_blockalign(bs
, align
);
3178 qemu_iovec_add(&local_qiov
, tail_buf
,
3179 align
- ((offset
+ bytes
) & (align
- 1)));
3181 bytes
= ROUND_UP(bytes
, align
);
3184 tracked_request_begin(&req
, bs
, offset
, bytes
, false);
3185 ret
= bdrv_aligned_preadv(bs
, &req
, offset
, bytes
, align
,
3186 use_local_qiov
? &local_qiov
: qiov
,
3188 tracked_request_end(&req
);
3190 if (use_local_qiov
) {
3191 qemu_iovec_destroy(&local_qiov
);
3192 qemu_vfree(head_buf
);
3193 qemu_vfree(tail_buf
);
3199 static int coroutine_fn
bdrv_co_do_readv(BlockDriverState
*bs
,
3200 int64_t sector_num
, int nb_sectors
, QEMUIOVector
*qiov
,
3201 BdrvRequestFlags flags
)
3203 if (nb_sectors
< 0 || nb_sectors
> (UINT_MAX
>> BDRV_SECTOR_BITS
)) {
3207 return bdrv_co_do_preadv(bs
, sector_num
<< BDRV_SECTOR_BITS
,
3208 nb_sectors
<< BDRV_SECTOR_BITS
, qiov
, flags
);
3211 int coroutine_fn
bdrv_co_readv(BlockDriverState
*bs
, int64_t sector_num
,
3212 int nb_sectors
, QEMUIOVector
*qiov
)
3214 trace_bdrv_co_readv(bs
, sector_num
, nb_sectors
);
3216 return bdrv_co_do_readv(bs
, sector_num
, nb_sectors
, qiov
, 0);
3219 int coroutine_fn
bdrv_co_copy_on_readv(BlockDriverState
*bs
,
3220 int64_t sector_num
, int nb_sectors
, QEMUIOVector
*qiov
)
3222 trace_bdrv_co_copy_on_readv(bs
, sector_num
, nb_sectors
);
3224 return bdrv_co_do_readv(bs
, sector_num
, nb_sectors
, qiov
,
3225 BDRV_REQ_COPY_ON_READ
);
3228 /* if no limit is specified in the BlockLimits use a default
3229 * of 32768 512-byte sectors (16 MiB) per request.
3231 #define MAX_WRITE_ZEROES_DEFAULT 32768
3233 static int coroutine_fn
bdrv_co_do_write_zeroes(BlockDriverState
*bs
,
3234 int64_t sector_num
, int nb_sectors
, BdrvRequestFlags flags
)
3236 BlockDriver
*drv
= bs
->drv
;
3238 struct iovec iov
= {0};
3241 int max_write_zeroes
= bs
->bl
.max_write_zeroes
?
3242 bs
->bl
.max_write_zeroes
: MAX_WRITE_ZEROES_DEFAULT
;
3244 while (nb_sectors
> 0 && !ret
) {
3245 int num
= nb_sectors
;
3247 /* Align request. Block drivers can expect the "bulk" of the request
3250 if (bs
->bl
.write_zeroes_alignment
3251 && num
> bs
->bl
.write_zeroes_alignment
) {
3252 if (sector_num
% bs
->bl
.write_zeroes_alignment
!= 0) {
3253 /* Make a small request up to the first aligned sector. */
3254 num
= bs
->bl
.write_zeroes_alignment
;
3255 num
-= sector_num
% bs
->bl
.write_zeroes_alignment
;
3256 } else if ((sector_num
+ num
) % bs
->bl
.write_zeroes_alignment
!= 0) {
3257 /* Shorten the request to the last aligned sector. num cannot
3258 * underflow because num > bs->bl.write_zeroes_alignment.
3260 num
-= (sector_num
+ num
) % bs
->bl
.write_zeroes_alignment
;
3264 /* limit request size */
3265 if (num
> max_write_zeroes
) {
3266 num
= max_write_zeroes
;
3270 /* First try the efficient write zeroes operation */
3271 if (drv
->bdrv_co_write_zeroes
) {
3272 ret
= drv
->bdrv_co_write_zeroes(bs
, sector_num
, num
, flags
);
3275 if (ret
== -ENOTSUP
) {
3276 /* Fall back to bounce buffer if write zeroes is unsupported */
3277 iov
.iov_len
= num
* BDRV_SECTOR_SIZE
;
3278 if (iov
.iov_base
== NULL
) {
3279 iov
.iov_base
= qemu_blockalign(bs
, num
* BDRV_SECTOR_SIZE
);
3280 memset(iov
.iov_base
, 0, num
* BDRV_SECTOR_SIZE
);
3282 qemu_iovec_init_external(&qiov
, &iov
, 1);
3284 ret
= drv
->bdrv_co_writev(bs
, sector_num
, num
, &qiov
);
3286 /* Keep bounce buffer around if it is big enough for all
3287 * all future requests.
3289 if (num
< max_write_zeroes
) {
3290 qemu_vfree(iov
.iov_base
);
3291 iov
.iov_base
= NULL
;
3299 qemu_vfree(iov
.iov_base
);
3304 * Forwards an already correctly aligned write request to the BlockDriver.
3306 static int coroutine_fn
bdrv_aligned_pwritev(BlockDriverState
*bs
,
3307 BdrvTrackedRequest
*req
, int64_t offset
, unsigned int bytes
,
3308 QEMUIOVector
*qiov
, int flags
)
3310 BlockDriver
*drv
= bs
->drv
;
3314 int64_t sector_num
= offset
>> BDRV_SECTOR_BITS
;
3315 unsigned int nb_sectors
= bytes
>> BDRV_SECTOR_BITS
;
3317 assert((offset
& (BDRV_SECTOR_SIZE
- 1)) == 0);
3318 assert((bytes
& (BDRV_SECTOR_SIZE
- 1)) == 0);
3320 waited
= wait_serialising_requests(req
);
3321 assert(!waited
|| !req
->serialising
);
3322 assert(req
->overlap_offset
<= offset
);
3323 assert(offset
+ bytes
<= req
->overlap_offset
+ req
->overlap_bytes
);
3325 ret
= notifier_with_return_list_notify(&bs
->before_write_notifiers
, req
);
3327 if (!ret
&& bs
->detect_zeroes
!= BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF
&&
3328 !(flags
& BDRV_REQ_ZERO_WRITE
) && drv
->bdrv_co_write_zeroes
&&
3329 qemu_iovec_is_zero(qiov
)) {
3330 flags
|= BDRV_REQ_ZERO_WRITE
;
3331 if (bs
->detect_zeroes
== BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP
) {
3332 flags
|= BDRV_REQ_MAY_UNMAP
;
3337 /* Do nothing, write notifier decided to fail this request */
3338 } else if (flags
& BDRV_REQ_ZERO_WRITE
) {
3339 BLKDBG_EVENT(bs
, BLKDBG_PWRITEV_ZERO
);
3340 ret
= bdrv_co_do_write_zeroes(bs
, sector_num
, nb_sectors
, flags
);
3342 BLKDBG_EVENT(bs
, BLKDBG_PWRITEV
);
3343 ret
= drv
->bdrv_co_writev(bs
, sector_num
, nb_sectors
, qiov
);
3345 BLKDBG_EVENT(bs
, BLKDBG_PWRITEV_DONE
);
3347 if (ret
== 0 && !bs
->enable_write_cache
) {
3348 ret
= bdrv_co_flush(bs
);
3351 bdrv_set_dirty(bs
, sector_num
, nb_sectors
);
3353 if (bs
->wr_highest_sector
< sector_num
+ nb_sectors
- 1) {
3354 bs
->wr_highest_sector
= sector_num
+ nb_sectors
- 1;
3356 if (bs
->growable
&& ret
>= 0) {
3357 bs
->total_sectors
= MAX(bs
->total_sectors
, sector_num
+ nb_sectors
);
3364 * Handle a write request in coroutine context
3366 static int coroutine_fn
bdrv_co_do_pwritev(BlockDriverState
*bs
,
3367 int64_t offset
, unsigned int bytes
, QEMUIOVector
*qiov
,
3368 BdrvRequestFlags flags
)
3370 BdrvTrackedRequest req
;
3371 /* TODO Lift BDRV_SECTOR_SIZE restriction in BlockDriver interface */
3372 uint64_t align
= MAX(BDRV_SECTOR_SIZE
, bs
->request_alignment
);
3373 uint8_t *head_buf
= NULL
;
3374 uint8_t *tail_buf
= NULL
;
3375 QEMUIOVector local_qiov
;
3376 bool use_local_qiov
= false;
3382 if (bs
->read_only
) {
3385 if (bdrv_check_byte_request(bs
, offset
, bytes
)) {
3389 /* throttling disk I/O */
3390 if (bs
->io_limits_enabled
) {
3391 bdrv_io_limits_intercept(bs
, bytes
, true);
3395 * Align write if necessary by performing a read-modify-write cycle.
3396 * Pad qiov with the read parts and be sure to have a tracked request not
3397 * only for bdrv_aligned_pwritev, but also for the reads of the RMW cycle.
3399 tracked_request_begin(&req
, bs
, offset
, bytes
, true);
3401 if (offset
& (align
- 1)) {
3402 QEMUIOVector head_qiov
;
3403 struct iovec head_iov
;
3405 mark_request_serialising(&req
, align
);
3406 wait_serialising_requests(&req
);
3408 head_buf
= qemu_blockalign(bs
, align
);
3409 head_iov
= (struct iovec
) {
3410 .iov_base
= head_buf
,
3413 qemu_iovec_init_external(&head_qiov
, &head_iov
, 1);
3415 BLKDBG_EVENT(bs
, BLKDBG_PWRITEV_RMW_HEAD
);
3416 ret
= bdrv_aligned_preadv(bs
, &req
, offset
& ~(align
- 1), align
,
3417 align
, &head_qiov
, 0);
3421 BLKDBG_EVENT(bs
, BLKDBG_PWRITEV_RMW_AFTER_HEAD
);
3423 qemu_iovec_init(&local_qiov
, qiov
->niov
+ 2);
3424 qemu_iovec_add(&local_qiov
, head_buf
, offset
& (align
- 1));
3425 qemu_iovec_concat(&local_qiov
, qiov
, 0, qiov
->size
);
3426 use_local_qiov
= true;
3428 bytes
+= offset
& (align
- 1);
3429 offset
= offset
& ~(align
- 1);
3432 if ((offset
+ bytes
) & (align
- 1)) {
3433 QEMUIOVector tail_qiov
;
3434 struct iovec tail_iov
;
3438 mark_request_serialising(&req
, align
);
3439 waited
= wait_serialising_requests(&req
);
3440 assert(!waited
|| !use_local_qiov
);
3442 tail_buf
= qemu_blockalign(bs
, align
);
3443 tail_iov
= (struct iovec
) {
3444 .iov_base
= tail_buf
,
3447 qemu_iovec_init_external(&tail_qiov
, &tail_iov
, 1);
3449 BLKDBG_EVENT(bs
, BLKDBG_PWRITEV_RMW_TAIL
);
3450 ret
= bdrv_aligned_preadv(bs
, &req
, (offset
+ bytes
) & ~(align
- 1), align
,
3451 align
, &tail_qiov
, 0);
3455 BLKDBG_EVENT(bs
, BLKDBG_PWRITEV_RMW_AFTER_TAIL
);
3457 if (!use_local_qiov
) {
3458 qemu_iovec_init(&local_qiov
, qiov
->niov
+ 1);
3459 qemu_iovec_concat(&local_qiov
, qiov
, 0, qiov
->size
);
3460 use_local_qiov
= true;
3463 tail_bytes
= (offset
+ bytes
) & (align
- 1);
3464 qemu_iovec_add(&local_qiov
, tail_buf
+ tail_bytes
, align
- tail_bytes
);
3466 bytes
= ROUND_UP(bytes
, align
);
3469 ret
= bdrv_aligned_pwritev(bs
, &req
, offset
, bytes
,
3470 use_local_qiov
? &local_qiov
: qiov
,
3474 tracked_request_end(&req
);
3476 if (use_local_qiov
) {
3477 qemu_iovec_destroy(&local_qiov
);
3479 qemu_vfree(head_buf
);
3480 qemu_vfree(tail_buf
);
3485 static int coroutine_fn
bdrv_co_do_writev(BlockDriverState
*bs
,
3486 int64_t sector_num
, int nb_sectors
, QEMUIOVector
*qiov
,
3487 BdrvRequestFlags flags
)
3489 if (nb_sectors
< 0 || nb_sectors
> (INT_MAX
>> BDRV_SECTOR_BITS
)) {
3493 return bdrv_co_do_pwritev(bs
, sector_num
<< BDRV_SECTOR_BITS
,
3494 nb_sectors
<< BDRV_SECTOR_BITS
, qiov
, flags
);
3497 int coroutine_fn
bdrv_co_writev(BlockDriverState
*bs
, int64_t sector_num
,
3498 int nb_sectors
, QEMUIOVector
*qiov
)
3500 trace_bdrv_co_writev(bs
, sector_num
, nb_sectors
);
3502 return bdrv_co_do_writev(bs
, sector_num
, nb_sectors
, qiov
, 0);
3505 int coroutine_fn
bdrv_co_write_zeroes(BlockDriverState
*bs
,
3506 int64_t sector_num
, int nb_sectors
,
3507 BdrvRequestFlags flags
)
3509 trace_bdrv_co_write_zeroes(bs
, sector_num
, nb_sectors
, flags
);
3511 if (!(bs
->open_flags
& BDRV_O_UNMAP
)) {
3512 flags
&= ~BDRV_REQ_MAY_UNMAP
;
3515 return bdrv_co_do_writev(bs
, sector_num
, nb_sectors
, NULL
,
3516 BDRV_REQ_ZERO_WRITE
| flags
);
3520 * Truncate file to 'offset' bytes (needed only for file protocols)
3522 int bdrv_truncate(BlockDriverState
*bs
, int64_t offset
)
3524 BlockDriver
*drv
= bs
->drv
;
3528 if (!drv
->bdrv_truncate
)
3532 if (bdrv_op_is_blocked(bs
, BLOCK_OP_TYPE_RESIZE
, NULL
)) {
3535 ret
= drv
->bdrv_truncate(bs
, offset
);
3537 ret
= refresh_total_sectors(bs
, offset
>> BDRV_SECTOR_BITS
);
3538 bdrv_dev_resize_cb(bs
);
3544 * Length of a allocated file in bytes. Sparse files are counted by actual
3545 * allocated space. Return < 0 if error or unknown.
3547 int64_t bdrv_get_allocated_file_size(BlockDriverState
*bs
)
3549 BlockDriver
*drv
= bs
->drv
;
3553 if (drv
->bdrv_get_allocated_file_size
) {
3554 return drv
->bdrv_get_allocated_file_size(bs
);
3557 return bdrv_get_allocated_file_size(bs
->file
);
3563 * Length of a file in bytes. Return < 0 if error or unknown.
3565 int64_t bdrv_getlength(BlockDriverState
*bs
)
3567 BlockDriver
*drv
= bs
->drv
;
3571 if (drv
->has_variable_length
) {
3572 int ret
= refresh_total_sectors(bs
, bs
->total_sectors
);
3577 return bs
->total_sectors
* BDRV_SECTOR_SIZE
;
3580 /* return 0 as number of sectors if no device present or error */
3581 void bdrv_get_geometry(BlockDriverState
*bs
, uint64_t *nb_sectors_ptr
)
3584 length
= bdrv_getlength(bs
);
3588 length
= length
>> BDRV_SECTOR_BITS
;
3589 *nb_sectors_ptr
= length
;
3592 void bdrv_set_on_error(BlockDriverState
*bs
, BlockdevOnError on_read_error
,
3593 BlockdevOnError on_write_error
)
3595 bs
->on_read_error
= on_read_error
;
3596 bs
->on_write_error
= on_write_error
;
3599 BlockdevOnError
bdrv_get_on_error(BlockDriverState
*bs
, bool is_read
)
3601 return is_read
? bs
->on_read_error
: bs
->on_write_error
;
3604 BlockErrorAction
bdrv_get_error_action(BlockDriverState
*bs
, bool is_read
, int error
)
3606 BlockdevOnError on_err
= is_read
? bs
->on_read_error
: bs
->on_write_error
;
3609 case BLOCKDEV_ON_ERROR_ENOSPC
:
3610 return (error
== ENOSPC
) ? BDRV_ACTION_STOP
: BDRV_ACTION_REPORT
;
3611 case BLOCKDEV_ON_ERROR_STOP
:
3612 return BDRV_ACTION_STOP
;
3613 case BLOCKDEV_ON_ERROR_REPORT
:
3614 return BDRV_ACTION_REPORT
;
3615 case BLOCKDEV_ON_ERROR_IGNORE
:
3616 return BDRV_ACTION_IGNORE
;
3622 /* This is done by device models because, while the block layer knows
3623 * about the error, it does not know whether an operation comes from
3624 * the device or the block layer (from a job, for example).
3626 void bdrv_error_action(BlockDriverState
*bs
, BlockErrorAction action
,
3627 bool is_read
, int error
)
3630 bdrv_emit_qmp_error_event(bs
, QEVENT_BLOCK_IO_ERROR
, action
, is_read
);
3631 if (action
== BDRV_ACTION_STOP
) {
3632 vm_stop(RUN_STATE_IO_ERROR
);
3633 bdrv_iostatus_set_err(bs
, error
);
3637 int bdrv_is_read_only(BlockDriverState
*bs
)
3639 return bs
->read_only
;
3642 int bdrv_is_sg(BlockDriverState
*bs
)
3647 int bdrv_enable_write_cache(BlockDriverState
*bs
)
3649 return bs
->enable_write_cache
;
3652 void bdrv_set_enable_write_cache(BlockDriverState
*bs
, bool wce
)
3654 bs
->enable_write_cache
= wce
;
3656 /* so a reopen() will preserve wce */
3658 bs
->open_flags
|= BDRV_O_CACHE_WB
;
3660 bs
->open_flags
&= ~BDRV_O_CACHE_WB
;
3664 int bdrv_is_encrypted(BlockDriverState
*bs
)
3666 if (bs
->backing_hd
&& bs
->backing_hd
->encrypted
)
3668 return bs
->encrypted
;
3671 int bdrv_key_required(BlockDriverState
*bs
)
3673 BlockDriverState
*backing_hd
= bs
->backing_hd
;
3675 if (backing_hd
&& backing_hd
->encrypted
&& !backing_hd
->valid_key
)
3677 return (bs
->encrypted
&& !bs
->valid_key
);
3680 int bdrv_set_key(BlockDriverState
*bs
, const char *key
)
3683 if (bs
->backing_hd
&& bs
->backing_hd
->encrypted
) {
3684 ret
= bdrv_set_key(bs
->backing_hd
, key
);
3690 if (!bs
->encrypted
) {
3692 } else if (!bs
->drv
|| !bs
->drv
->bdrv_set_key
) {
3695 ret
= bs
->drv
->bdrv_set_key(bs
, key
);
3698 } else if (!bs
->valid_key
) {
3700 /* call the change callback now, we skipped it on open */
3701 bdrv_dev_change_media_cb(bs
, true);
3706 const char *bdrv_get_format_name(BlockDriverState
*bs
)
3708 return bs
->drv
? bs
->drv
->format_name
: NULL
;
3711 void bdrv_iterate_format(void (*it
)(void *opaque
, const char *name
),
3716 const char **formats
= NULL
;
3718 QLIST_FOREACH(drv
, &bdrv_drivers
, list
) {
3719 if (drv
->format_name
) {
3722 while (formats
&& i
&& !found
) {
3723 found
= !strcmp(formats
[--i
], drv
->format_name
);
3727 formats
= g_realloc(formats
, (count
+ 1) * sizeof(char *));
3728 formats
[count
++] = drv
->format_name
;
3729 it(opaque
, drv
->format_name
);
3736 /* This function is to find block backend bs */
3737 BlockDriverState
*bdrv_find(const char *name
)
3739 BlockDriverState
*bs
;
3741 QTAILQ_FOREACH(bs
, &bdrv_states
, device_list
) {
3742 if (!strcmp(name
, bs
->device_name
)) {
3749 /* This function is to find a node in the bs graph */
3750 BlockDriverState
*bdrv_find_node(const char *node_name
)
3752 BlockDriverState
*bs
;
3756 QTAILQ_FOREACH(bs
, &graph_bdrv_states
, node_list
) {
3757 if (!strcmp(node_name
, bs
->node_name
)) {
3764 /* Put this QMP function here so it can access the static graph_bdrv_states. */
3765 BlockDeviceInfoList
*bdrv_named_nodes_list(void)
3767 BlockDeviceInfoList
*list
, *entry
;
3768 BlockDriverState
*bs
;
3771 QTAILQ_FOREACH(bs
, &graph_bdrv_states
, node_list
) {
3772 entry
= g_malloc0(sizeof(*entry
));
3773 entry
->value
= bdrv_block_device_info(bs
);
3781 BlockDriverState
*bdrv_lookup_bs(const char *device
,
3782 const char *node_name
,
3785 BlockDriverState
*bs
= NULL
;
3788 bs
= bdrv_find(device
);
3796 bs
= bdrv_find_node(node_name
);
3803 error_setg(errp
, "Cannot find device=%s nor node_name=%s",
3804 device
? device
: "",
3805 node_name
? node_name
: "");
3809 BlockDriverState
*bdrv_next(BlockDriverState
*bs
)
3812 return QTAILQ_FIRST(&bdrv_states
);
3814 return QTAILQ_NEXT(bs
, device_list
);
3817 void bdrv_iterate(void (*it
)(void *opaque
, BlockDriverState
*bs
), void *opaque
)
3819 BlockDriverState
*bs
;
3821 QTAILQ_FOREACH(bs
, &bdrv_states
, device_list
) {
3826 const char *bdrv_get_device_name(BlockDriverState
*bs
)
3828 return bs
->device_name
;
3831 int bdrv_get_flags(BlockDriverState
*bs
)
3833 return bs
->open_flags
;
3836 int bdrv_flush_all(void)
3838 BlockDriverState
*bs
;
3841 QTAILQ_FOREACH(bs
, &bdrv_states
, device_list
) {
3842 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
3845 aio_context_acquire(aio_context
);
3846 ret
= bdrv_flush(bs
);
3847 if (ret
< 0 && !result
) {
3850 aio_context_release(aio_context
);
3856 int bdrv_has_zero_init_1(BlockDriverState
*bs
)
3861 int bdrv_has_zero_init(BlockDriverState
*bs
)
3865 /* If BS is a copy on write image, it is initialized to
3866 the contents of the base image, which may not be zeroes. */
3867 if (bs
->backing_hd
) {
3870 if (bs
->drv
->bdrv_has_zero_init
) {
3871 return bs
->drv
->bdrv_has_zero_init(bs
);
3878 bool bdrv_unallocated_blocks_are_zero(BlockDriverState
*bs
)
3880 BlockDriverInfo bdi
;
3882 if (bs
->backing_hd
) {
3886 if (bdrv_get_info(bs
, &bdi
) == 0) {
3887 return bdi
.unallocated_blocks_are_zero
;
3893 bool bdrv_can_write_zeroes_with_unmap(BlockDriverState
*bs
)
3895 BlockDriverInfo bdi
;
3897 if (bs
->backing_hd
|| !(bs
->open_flags
& BDRV_O_UNMAP
)) {
3901 if (bdrv_get_info(bs
, &bdi
) == 0) {
3902 return bdi
.can_write_zeroes_with_unmap
;
3908 typedef struct BdrvCoGetBlockStatusData
{
3909 BlockDriverState
*bs
;
3910 BlockDriverState
*base
;
3916 } BdrvCoGetBlockStatusData
;
3919 * Returns true iff the specified sector is present in the disk image. Drivers
3920 * not implementing the functionality are assumed to not support backing files,
3921 * hence all their sectors are reported as allocated.
3923 * If 'sector_num' is beyond the end of the disk image the return value is 0
3924 * and 'pnum' is set to 0.
3926 * 'pnum' is set to the number of sectors (including and immediately following
3927 * the specified sector) that are known to be in the same
3928 * allocated/unallocated state.
3930 * 'nb_sectors' is the max value 'pnum' should be set to. If nb_sectors goes
3931 * beyond the end of the disk image it will be clamped.
3933 static int64_t coroutine_fn
bdrv_co_get_block_status(BlockDriverState
*bs
,
3935 int nb_sectors
, int *pnum
)
3941 length
= bdrv_getlength(bs
);
3946 if (sector_num
>= (length
>> BDRV_SECTOR_BITS
)) {
3951 n
= bs
->total_sectors
- sector_num
;
3952 if (n
< nb_sectors
) {
3956 if (!bs
->drv
->bdrv_co_get_block_status
) {
3958 ret
= BDRV_BLOCK_DATA
| BDRV_BLOCK_ALLOCATED
;
3959 if (bs
->drv
->protocol_name
) {
3960 ret
|= BDRV_BLOCK_OFFSET_VALID
| (sector_num
* BDRV_SECTOR_SIZE
);
3965 ret
= bs
->drv
->bdrv_co_get_block_status(bs
, sector_num
, nb_sectors
, pnum
);
3971 if (ret
& BDRV_BLOCK_RAW
) {
3972 assert(ret
& BDRV_BLOCK_OFFSET_VALID
);
3973 return bdrv_get_block_status(bs
->file
, ret
>> BDRV_SECTOR_BITS
,
3977 if (ret
& (BDRV_BLOCK_DATA
| BDRV_BLOCK_ZERO
)) {
3978 ret
|= BDRV_BLOCK_ALLOCATED
;
3981 if (!(ret
& BDRV_BLOCK_DATA
) && !(ret
& BDRV_BLOCK_ZERO
)) {
3982 if (bdrv_unallocated_blocks_are_zero(bs
)) {
3983 ret
|= BDRV_BLOCK_ZERO
;
3984 } else if (bs
->backing_hd
) {
3985 BlockDriverState
*bs2
= bs
->backing_hd
;
3986 int64_t length2
= bdrv_getlength(bs2
);
3987 if (length2
>= 0 && sector_num
>= (length2
>> BDRV_SECTOR_BITS
)) {
3988 ret
|= BDRV_BLOCK_ZERO
;
3994 (ret
& BDRV_BLOCK_DATA
) && !(ret
& BDRV_BLOCK_ZERO
) &&
3995 (ret
& BDRV_BLOCK_OFFSET_VALID
)) {
3996 ret2
= bdrv_co_get_block_status(bs
->file
, ret
>> BDRV_SECTOR_BITS
,
3999 /* Ignore errors. This is just providing extra information, it
4000 * is useful but not necessary.
4002 ret
|= (ret2
& BDRV_BLOCK_ZERO
);
4009 /* Coroutine wrapper for bdrv_get_block_status() */
4010 static void coroutine_fn
bdrv_get_block_status_co_entry(void *opaque
)
4012 BdrvCoGetBlockStatusData
*data
= opaque
;
4013 BlockDriverState
*bs
= data
->bs
;
4015 data
->ret
= bdrv_co_get_block_status(bs
, data
->sector_num
, data
->nb_sectors
,
4021 * Synchronous wrapper around bdrv_co_get_block_status().
4023 * See bdrv_co_get_block_status() for details.
4025 int64_t bdrv_get_block_status(BlockDriverState
*bs
, int64_t sector_num
,
4026 int nb_sectors
, int *pnum
)
4029 BdrvCoGetBlockStatusData data
= {
4031 .sector_num
= sector_num
,
4032 .nb_sectors
= nb_sectors
,
4037 if (qemu_in_coroutine()) {
4038 /* Fast-path if already in coroutine context */
4039 bdrv_get_block_status_co_entry(&data
);
4041 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
4043 co
= qemu_coroutine_create(bdrv_get_block_status_co_entry
);
4044 qemu_coroutine_enter(co
, &data
);
4045 while (!data
.done
) {
4046 aio_poll(aio_context
, true);
4052 int coroutine_fn
bdrv_is_allocated(BlockDriverState
*bs
, int64_t sector_num
,
4053 int nb_sectors
, int *pnum
)
4055 int64_t ret
= bdrv_get_block_status(bs
, sector_num
, nb_sectors
, pnum
);
4059 return (ret
& BDRV_BLOCK_ALLOCATED
);
4063 * Given an image chain: ... -> [BASE] -> [INTER1] -> [INTER2] -> [TOP]
4065 * Return true if the given sector is allocated in any image between
4066 * BASE and TOP (inclusive). BASE can be NULL to check if the given
4067 * sector is allocated in any image of the chain. Return false otherwise.
4069 * 'pnum' is set to the number of sectors (including and immediately following
4070 * the specified sector) that are known to be in the same
4071 * allocated/unallocated state.
4074 int bdrv_is_allocated_above(BlockDriverState
*top
,
4075 BlockDriverState
*base
,
4077 int nb_sectors
, int *pnum
)
4079 BlockDriverState
*intermediate
;
4080 int ret
, n
= nb_sectors
;
4083 while (intermediate
&& intermediate
!= base
) {
4085 ret
= bdrv_is_allocated(intermediate
, sector_num
, nb_sectors
,
4095 * [sector_num, nb_sectors] is unallocated on top but intermediate
4098 * [sector_num+x, nr_sectors] allocated.
4100 if (n
> pnum_inter
&&
4101 (intermediate
== top
||
4102 sector_num
+ pnum_inter
< intermediate
->total_sectors
)) {
4106 intermediate
= intermediate
->backing_hd
;
4113 const char *bdrv_get_encrypted_filename(BlockDriverState
*bs
)
4115 if (bs
->backing_hd
&& bs
->backing_hd
->encrypted
)
4116 return bs
->backing_file
;
4117 else if (bs
->encrypted
)
4118 return bs
->filename
;
4123 void bdrv_get_backing_filename(BlockDriverState
*bs
,
4124 char *filename
, int filename_size
)
4126 pstrcpy(filename
, filename_size
, bs
->backing_file
);
4129 int bdrv_write_compressed(BlockDriverState
*bs
, int64_t sector_num
,
4130 const uint8_t *buf
, int nb_sectors
)
4132 BlockDriver
*drv
= bs
->drv
;
4135 if (!drv
->bdrv_write_compressed
)
4137 if (bdrv_check_request(bs
, sector_num
, nb_sectors
))
4140 assert(QLIST_EMPTY(&bs
->dirty_bitmaps
));
4142 return drv
->bdrv_write_compressed(bs
, sector_num
, buf
, nb_sectors
);
4145 int bdrv_get_info(BlockDriverState
*bs
, BlockDriverInfo
*bdi
)
4147 BlockDriver
*drv
= bs
->drv
;
4150 if (!drv
->bdrv_get_info
)
4152 memset(bdi
, 0, sizeof(*bdi
));
4153 return drv
->bdrv_get_info(bs
, bdi
);
4156 ImageInfoSpecific
*bdrv_get_specific_info(BlockDriverState
*bs
)
4158 BlockDriver
*drv
= bs
->drv
;
4159 if (drv
&& drv
->bdrv_get_specific_info
) {
4160 return drv
->bdrv_get_specific_info(bs
);
4165 int bdrv_save_vmstate(BlockDriverState
*bs
, const uint8_t *buf
,
4166 int64_t pos
, int size
)
4169 struct iovec iov
= {
4170 .iov_base
= (void *) buf
,
4174 qemu_iovec_init_external(&qiov
, &iov
, 1);
4175 return bdrv_writev_vmstate(bs
, &qiov
, pos
);
4178 int bdrv_writev_vmstate(BlockDriverState
*bs
, QEMUIOVector
*qiov
, int64_t pos
)
4180 BlockDriver
*drv
= bs
->drv
;
4184 } else if (drv
->bdrv_save_vmstate
) {
4185 return drv
->bdrv_save_vmstate(bs
, qiov
, pos
);
4186 } else if (bs
->file
) {
4187 return bdrv_writev_vmstate(bs
->file
, qiov
, pos
);
4193 int bdrv_load_vmstate(BlockDriverState
*bs
, uint8_t *buf
,
4194 int64_t pos
, int size
)
4196 BlockDriver
*drv
= bs
->drv
;
4199 if (drv
->bdrv_load_vmstate
)
4200 return drv
->bdrv_load_vmstate(bs
, buf
, pos
, size
);
4202 return bdrv_load_vmstate(bs
->file
, buf
, pos
, size
);
4206 void bdrv_debug_event(BlockDriverState
*bs
, BlkDebugEvent event
)
4208 if (!bs
|| !bs
->drv
|| !bs
->drv
->bdrv_debug_event
) {
4212 bs
->drv
->bdrv_debug_event(bs
, event
);
4215 int bdrv_debug_breakpoint(BlockDriverState
*bs
, const char *event
,
4218 while (bs
&& bs
->drv
&& !bs
->drv
->bdrv_debug_breakpoint
) {
4222 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_breakpoint
) {
4223 return bs
->drv
->bdrv_debug_breakpoint(bs
, event
, tag
);
4229 int bdrv_debug_remove_breakpoint(BlockDriverState
*bs
, const char *tag
)
4231 while (bs
&& bs
->drv
&& !bs
->drv
->bdrv_debug_remove_breakpoint
) {
4235 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_remove_breakpoint
) {
4236 return bs
->drv
->bdrv_debug_remove_breakpoint(bs
, tag
);
4242 int bdrv_debug_resume(BlockDriverState
*bs
, const char *tag
)
4244 while (bs
&& (!bs
->drv
|| !bs
->drv
->bdrv_debug_resume
)) {
4248 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_resume
) {
4249 return bs
->drv
->bdrv_debug_resume(bs
, tag
);
4255 bool bdrv_debug_is_suspended(BlockDriverState
*bs
, const char *tag
)
4257 while (bs
&& bs
->drv
&& !bs
->drv
->bdrv_debug_is_suspended
) {
4261 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_is_suspended
) {
4262 return bs
->drv
->bdrv_debug_is_suspended(bs
, tag
);
4268 int bdrv_is_snapshot(BlockDriverState
*bs
)
4270 return !!(bs
->open_flags
& BDRV_O_SNAPSHOT
);
4273 /* backing_file can either be relative, or absolute, or a protocol. If it is
4274 * relative, it must be relative to the chain. So, passing in bs->filename
4275 * from a BDS as backing_file should not be done, as that may be relative to
4276 * the CWD rather than the chain. */
4277 BlockDriverState
*bdrv_find_backing_image(BlockDriverState
*bs
,
4278 const char *backing_file
)
4280 char *filename_full
= NULL
;
4281 char *backing_file_full
= NULL
;
4282 char *filename_tmp
= NULL
;
4283 int is_protocol
= 0;
4284 BlockDriverState
*curr_bs
= NULL
;
4285 BlockDriverState
*retval
= NULL
;
4287 if (!bs
|| !bs
->drv
|| !backing_file
) {
4291 filename_full
= g_malloc(PATH_MAX
);
4292 backing_file_full
= g_malloc(PATH_MAX
);
4293 filename_tmp
= g_malloc(PATH_MAX
);
4295 is_protocol
= path_has_protocol(backing_file
);
4297 for (curr_bs
= bs
; curr_bs
->backing_hd
; curr_bs
= curr_bs
->backing_hd
) {
4299 /* If either of the filename paths is actually a protocol, then
4300 * compare unmodified paths; otherwise make paths relative */
4301 if (is_protocol
|| path_has_protocol(curr_bs
->backing_file
)) {
4302 if (strcmp(backing_file
, curr_bs
->backing_file
) == 0) {
4303 retval
= curr_bs
->backing_hd
;
4307 /* If not an absolute filename path, make it relative to the current
4308 * image's filename path */
4309 path_combine(filename_tmp
, PATH_MAX
, curr_bs
->filename
,
4312 /* We are going to compare absolute pathnames */
4313 if (!realpath(filename_tmp
, filename_full
)) {
4317 /* We need to make sure the backing filename we are comparing against
4318 * is relative to the current image filename (or absolute) */
4319 path_combine(filename_tmp
, PATH_MAX
, curr_bs
->filename
,
4320 curr_bs
->backing_file
);
4322 if (!realpath(filename_tmp
, backing_file_full
)) {
4326 if (strcmp(backing_file_full
, filename_full
) == 0) {
4327 retval
= curr_bs
->backing_hd
;
4333 g_free(filename_full
);
4334 g_free(backing_file_full
);
4335 g_free(filename_tmp
);
4339 int bdrv_get_backing_file_depth(BlockDriverState
*bs
)
4345 if (!bs
->backing_hd
) {
4349 return 1 + bdrv_get_backing_file_depth(bs
->backing_hd
);
4352 BlockDriverState
*bdrv_find_base(BlockDriverState
*bs
)
4354 BlockDriverState
*curr_bs
= NULL
;
4362 while (curr_bs
->backing_hd
) {
4363 curr_bs
= curr_bs
->backing_hd
;
4368 /**************************************************************/
4371 BlockDriverAIOCB
*bdrv_aio_readv(BlockDriverState
*bs
, int64_t sector_num
,
4372 QEMUIOVector
*qiov
, int nb_sectors
,
4373 BlockDriverCompletionFunc
*cb
, void *opaque
)
4375 trace_bdrv_aio_readv(bs
, sector_num
, nb_sectors
, opaque
);
4377 return bdrv_co_aio_rw_vector(bs
, sector_num
, qiov
, nb_sectors
, 0,
4381 BlockDriverAIOCB
*bdrv_aio_writev(BlockDriverState
*bs
, int64_t sector_num
,
4382 QEMUIOVector
*qiov
, int nb_sectors
,
4383 BlockDriverCompletionFunc
*cb
, void *opaque
)
4385 trace_bdrv_aio_writev(bs
, sector_num
, nb_sectors
, opaque
);
4387 return bdrv_co_aio_rw_vector(bs
, sector_num
, qiov
, nb_sectors
, 0,
4391 BlockDriverAIOCB
*bdrv_aio_write_zeroes(BlockDriverState
*bs
,
4392 int64_t sector_num
, int nb_sectors
, BdrvRequestFlags flags
,
4393 BlockDriverCompletionFunc
*cb
, void *opaque
)
4395 trace_bdrv_aio_write_zeroes(bs
, sector_num
, nb_sectors
, flags
, opaque
);
4397 return bdrv_co_aio_rw_vector(bs
, sector_num
, NULL
, nb_sectors
,
4398 BDRV_REQ_ZERO_WRITE
| flags
,
4403 typedef struct MultiwriteCB
{
4408 BlockDriverCompletionFunc
*cb
;
4410 QEMUIOVector
*free_qiov
;
4414 static void multiwrite_user_cb(MultiwriteCB
*mcb
)
4418 for (i
= 0; i
< mcb
->num_callbacks
; i
++) {
4419 mcb
->callbacks
[i
].cb(mcb
->callbacks
[i
].opaque
, mcb
->error
);
4420 if (mcb
->callbacks
[i
].free_qiov
) {
4421 qemu_iovec_destroy(mcb
->callbacks
[i
].free_qiov
);
4423 g_free(mcb
->callbacks
[i
].free_qiov
);
4427 static void multiwrite_cb(void *opaque
, int ret
)
4429 MultiwriteCB
*mcb
= opaque
;
4431 trace_multiwrite_cb(mcb
, ret
);
4433 if (ret
< 0 && !mcb
->error
) {
4437 mcb
->num_requests
--;
4438 if (mcb
->num_requests
== 0) {
4439 multiwrite_user_cb(mcb
);
4444 static int multiwrite_req_compare(const void *a
, const void *b
)
4446 const BlockRequest
*req1
= a
, *req2
= b
;
4449 * Note that we can't simply subtract req2->sector from req1->sector
4450 * here as that could overflow the return value.
4452 if (req1
->sector
> req2
->sector
) {
4454 } else if (req1
->sector
< req2
->sector
) {
4462 * Takes a bunch of requests and tries to merge them. Returns the number of
4463 * requests that remain after merging.
4465 static int multiwrite_merge(BlockDriverState
*bs
, BlockRequest
*reqs
,
4466 int num_reqs
, MultiwriteCB
*mcb
)
4470 // Sort requests by start sector
4471 qsort(reqs
, num_reqs
, sizeof(*reqs
), &multiwrite_req_compare
);
4473 // Check if adjacent requests touch the same clusters. If so, combine them,
4474 // filling up gaps with zero sectors.
4476 for (i
= 1; i
< num_reqs
; i
++) {
4478 int64_t oldreq_last
= reqs
[outidx
].sector
+ reqs
[outidx
].nb_sectors
;
4480 // Handle exactly sequential writes and overlapping writes.
4481 if (reqs
[i
].sector
<= oldreq_last
) {
4485 if (reqs
[outidx
].qiov
->niov
+ reqs
[i
].qiov
->niov
+ 1 > IOV_MAX
) {
4491 QEMUIOVector
*qiov
= g_malloc0(sizeof(*qiov
));
4492 qemu_iovec_init(qiov
,
4493 reqs
[outidx
].qiov
->niov
+ reqs
[i
].qiov
->niov
+ 1);
4495 // Add the first request to the merged one. If the requests are
4496 // overlapping, drop the last sectors of the first request.
4497 size
= (reqs
[i
].sector
- reqs
[outidx
].sector
) << 9;
4498 qemu_iovec_concat(qiov
, reqs
[outidx
].qiov
, 0, size
);
4500 // We should need to add any zeros between the two requests
4501 assert (reqs
[i
].sector
<= oldreq_last
);
4503 // Add the second request
4504 qemu_iovec_concat(qiov
, reqs
[i
].qiov
, 0, reqs
[i
].qiov
->size
);
4506 reqs
[outidx
].nb_sectors
= qiov
->size
>> 9;
4507 reqs
[outidx
].qiov
= qiov
;
4509 mcb
->callbacks
[i
].free_qiov
= reqs
[outidx
].qiov
;
4512 reqs
[outidx
].sector
= reqs
[i
].sector
;
4513 reqs
[outidx
].nb_sectors
= reqs
[i
].nb_sectors
;
4514 reqs
[outidx
].qiov
= reqs
[i
].qiov
;
4522 * Submit multiple AIO write requests at once.
4524 * On success, the function returns 0 and all requests in the reqs array have
4525 * been submitted. In error case this function returns -1, and any of the
4526 * requests may or may not be submitted yet. In particular, this means that the
4527 * callback will be called for some of the requests, for others it won't. The
4528 * caller must check the error field of the BlockRequest to wait for the right
4529 * callbacks (if error != 0, no callback will be called).
4531 * The implementation may modify the contents of the reqs array, e.g. to merge
4532 * requests. However, the fields opaque and error are left unmodified as they
4533 * are used to signal failure for a single request to the caller.
4535 int bdrv_aio_multiwrite(BlockDriverState
*bs
, BlockRequest
*reqs
, int num_reqs
)
4540 /* don't submit writes if we don't have a medium */
4541 if (bs
->drv
== NULL
) {
4542 for (i
= 0; i
< num_reqs
; i
++) {
4543 reqs
[i
].error
= -ENOMEDIUM
;
4548 if (num_reqs
== 0) {
4552 // Create MultiwriteCB structure
4553 mcb
= g_malloc0(sizeof(*mcb
) + num_reqs
* sizeof(*mcb
->callbacks
));
4554 mcb
->num_requests
= 0;
4555 mcb
->num_callbacks
= num_reqs
;
4557 for (i
= 0; i
< num_reqs
; i
++) {
4558 mcb
->callbacks
[i
].cb
= reqs
[i
].cb
;
4559 mcb
->callbacks
[i
].opaque
= reqs
[i
].opaque
;
4562 // Check for mergable requests
4563 num_reqs
= multiwrite_merge(bs
, reqs
, num_reqs
, mcb
);
4565 trace_bdrv_aio_multiwrite(mcb
, mcb
->num_callbacks
, num_reqs
);
4567 /* Run the aio requests. */
4568 mcb
->num_requests
= num_reqs
;
4569 for (i
= 0; i
< num_reqs
; i
++) {
4570 bdrv_co_aio_rw_vector(bs
, reqs
[i
].sector
, reqs
[i
].qiov
,
4571 reqs
[i
].nb_sectors
, reqs
[i
].flags
,
4579 void bdrv_aio_cancel(BlockDriverAIOCB
*acb
)
4581 acb
->aiocb_info
->cancel(acb
);
4584 /**************************************************************/
4585 /* async block device emulation */
4587 typedef struct BlockDriverAIOCBSync
{
4588 BlockDriverAIOCB common
;
4591 /* vector translation state */
4595 } BlockDriverAIOCBSync
;
4597 static void bdrv_aio_cancel_em(BlockDriverAIOCB
*blockacb
)
4599 BlockDriverAIOCBSync
*acb
=
4600 container_of(blockacb
, BlockDriverAIOCBSync
, common
);
4601 qemu_bh_delete(acb
->bh
);
4603 qemu_aio_release(acb
);
4606 static const AIOCBInfo bdrv_em_aiocb_info
= {
4607 .aiocb_size
= sizeof(BlockDriverAIOCBSync
),
4608 .cancel
= bdrv_aio_cancel_em
,
4611 static void bdrv_aio_bh_cb(void *opaque
)
4613 BlockDriverAIOCBSync
*acb
= opaque
;
4616 qemu_iovec_from_buf(acb
->qiov
, 0, acb
->bounce
, acb
->qiov
->size
);
4617 qemu_vfree(acb
->bounce
);
4618 acb
->common
.cb(acb
->common
.opaque
, acb
->ret
);
4619 qemu_bh_delete(acb
->bh
);
4621 qemu_aio_release(acb
);
4624 static BlockDriverAIOCB
*bdrv_aio_rw_vector(BlockDriverState
*bs
,
4628 BlockDriverCompletionFunc
*cb
,
4633 BlockDriverAIOCBSync
*acb
;
4635 acb
= qemu_aio_get(&bdrv_em_aiocb_info
, bs
, cb
, opaque
);
4636 acb
->is_write
= is_write
;
4638 acb
->bounce
= qemu_blockalign(bs
, qiov
->size
);
4639 acb
->bh
= aio_bh_new(bdrv_get_aio_context(bs
), bdrv_aio_bh_cb
, acb
);
4642 qemu_iovec_to_buf(acb
->qiov
, 0, acb
->bounce
, qiov
->size
);
4643 acb
->ret
= bs
->drv
->bdrv_write(bs
, sector_num
, acb
->bounce
, nb_sectors
);
4645 acb
->ret
= bs
->drv
->bdrv_read(bs
, sector_num
, acb
->bounce
, nb_sectors
);
4648 qemu_bh_schedule(acb
->bh
);
4650 return &acb
->common
;
4653 static BlockDriverAIOCB
*bdrv_aio_readv_em(BlockDriverState
*bs
,
4654 int64_t sector_num
, QEMUIOVector
*qiov
, int nb_sectors
,
4655 BlockDriverCompletionFunc
*cb
, void *opaque
)
4657 return bdrv_aio_rw_vector(bs
, sector_num
, qiov
, nb_sectors
, cb
, opaque
, 0);
4660 static BlockDriverAIOCB
*bdrv_aio_writev_em(BlockDriverState
*bs
,
4661 int64_t sector_num
, QEMUIOVector
*qiov
, int nb_sectors
,
4662 BlockDriverCompletionFunc
*cb
, void *opaque
)
4664 return bdrv_aio_rw_vector(bs
, sector_num
, qiov
, nb_sectors
, cb
, opaque
, 1);
4668 typedef struct BlockDriverAIOCBCoroutine
{
4669 BlockDriverAIOCB common
;
4674 } BlockDriverAIOCBCoroutine
;
4676 static void bdrv_aio_co_cancel_em(BlockDriverAIOCB
*blockacb
)
4678 AioContext
*aio_context
= bdrv_get_aio_context(blockacb
->bs
);
4679 BlockDriverAIOCBCoroutine
*acb
=
4680 container_of(blockacb
, BlockDriverAIOCBCoroutine
, common
);
4685 aio_poll(aio_context
, true);
4689 static const AIOCBInfo bdrv_em_co_aiocb_info
= {
4690 .aiocb_size
= sizeof(BlockDriverAIOCBCoroutine
),
4691 .cancel
= bdrv_aio_co_cancel_em
,
4694 static void bdrv_co_em_bh(void *opaque
)
4696 BlockDriverAIOCBCoroutine
*acb
= opaque
;
4698 acb
->common
.cb(acb
->common
.opaque
, acb
->req
.error
);
4704 qemu_bh_delete(acb
->bh
);
4705 qemu_aio_release(acb
);
4708 /* Invoke bdrv_co_do_readv/bdrv_co_do_writev */
4709 static void coroutine_fn
bdrv_co_do_rw(void *opaque
)
4711 BlockDriverAIOCBCoroutine
*acb
= opaque
;
4712 BlockDriverState
*bs
= acb
->common
.bs
;
4714 if (!acb
->is_write
) {
4715 acb
->req
.error
= bdrv_co_do_readv(bs
, acb
->req
.sector
,
4716 acb
->req
.nb_sectors
, acb
->req
.qiov
, acb
->req
.flags
);
4718 acb
->req
.error
= bdrv_co_do_writev(bs
, acb
->req
.sector
,
4719 acb
->req
.nb_sectors
, acb
->req
.qiov
, acb
->req
.flags
);
4722 acb
->bh
= aio_bh_new(bdrv_get_aio_context(bs
), bdrv_co_em_bh
, acb
);
4723 qemu_bh_schedule(acb
->bh
);
4726 static BlockDriverAIOCB
*bdrv_co_aio_rw_vector(BlockDriverState
*bs
,
4730 BdrvRequestFlags flags
,
4731 BlockDriverCompletionFunc
*cb
,
4736 BlockDriverAIOCBCoroutine
*acb
;
4738 acb
= qemu_aio_get(&bdrv_em_co_aiocb_info
, bs
, cb
, opaque
);
4739 acb
->req
.sector
= sector_num
;
4740 acb
->req
.nb_sectors
= nb_sectors
;
4741 acb
->req
.qiov
= qiov
;
4742 acb
->req
.flags
= flags
;
4743 acb
->is_write
= is_write
;
4746 co
= qemu_coroutine_create(bdrv_co_do_rw
);
4747 qemu_coroutine_enter(co
, acb
);
4749 return &acb
->common
;
4752 static void coroutine_fn
bdrv_aio_flush_co_entry(void *opaque
)
4754 BlockDriverAIOCBCoroutine
*acb
= opaque
;
4755 BlockDriverState
*bs
= acb
->common
.bs
;
4757 acb
->req
.error
= bdrv_co_flush(bs
);
4758 acb
->bh
= aio_bh_new(bdrv_get_aio_context(bs
), bdrv_co_em_bh
, acb
);
4759 qemu_bh_schedule(acb
->bh
);
4762 BlockDriverAIOCB
*bdrv_aio_flush(BlockDriverState
*bs
,
4763 BlockDriverCompletionFunc
*cb
, void *opaque
)
4765 trace_bdrv_aio_flush(bs
, opaque
);
4768 BlockDriverAIOCBCoroutine
*acb
;
4770 acb
= qemu_aio_get(&bdrv_em_co_aiocb_info
, bs
, cb
, opaque
);
4773 co
= qemu_coroutine_create(bdrv_aio_flush_co_entry
);
4774 qemu_coroutine_enter(co
, acb
);
4776 return &acb
->common
;
4779 static void coroutine_fn
bdrv_aio_discard_co_entry(void *opaque
)
4781 BlockDriverAIOCBCoroutine
*acb
= opaque
;
4782 BlockDriverState
*bs
= acb
->common
.bs
;
4784 acb
->req
.error
= bdrv_co_discard(bs
, acb
->req
.sector
, acb
->req
.nb_sectors
);
4785 acb
->bh
= aio_bh_new(bdrv_get_aio_context(bs
), bdrv_co_em_bh
, acb
);
4786 qemu_bh_schedule(acb
->bh
);
4789 BlockDriverAIOCB
*bdrv_aio_discard(BlockDriverState
*bs
,
4790 int64_t sector_num
, int nb_sectors
,
4791 BlockDriverCompletionFunc
*cb
, void *opaque
)
4794 BlockDriverAIOCBCoroutine
*acb
;
4796 trace_bdrv_aio_discard(bs
, sector_num
, nb_sectors
, opaque
);
4798 acb
= qemu_aio_get(&bdrv_em_co_aiocb_info
, bs
, cb
, opaque
);
4799 acb
->req
.sector
= sector_num
;
4800 acb
->req
.nb_sectors
= nb_sectors
;
4802 co
= qemu_coroutine_create(bdrv_aio_discard_co_entry
);
4803 qemu_coroutine_enter(co
, acb
);
4805 return &acb
->common
;
4808 void bdrv_init(void)
4810 module_call_init(MODULE_INIT_BLOCK
);
4813 void bdrv_init_with_whitelist(void)
4815 use_bdrv_whitelist
= 1;
4819 void *qemu_aio_get(const AIOCBInfo
*aiocb_info
, BlockDriverState
*bs
,
4820 BlockDriverCompletionFunc
*cb
, void *opaque
)
4822 BlockDriverAIOCB
*acb
;
4824 acb
= g_slice_alloc(aiocb_info
->aiocb_size
);
4825 acb
->aiocb_info
= aiocb_info
;
4828 acb
->opaque
= opaque
;
4832 void qemu_aio_release(void *p
)
4834 BlockDriverAIOCB
*acb
= p
;
4835 g_slice_free1(acb
->aiocb_info
->aiocb_size
, acb
);
4838 /**************************************************************/
4839 /* Coroutine block device emulation */
4841 typedef struct CoroutineIOCompletion
{
4842 Coroutine
*coroutine
;
4844 } CoroutineIOCompletion
;
4846 static void bdrv_co_io_em_complete(void *opaque
, int ret
)
4848 CoroutineIOCompletion
*co
= opaque
;
4851 qemu_coroutine_enter(co
->coroutine
, NULL
);
4854 static int coroutine_fn
bdrv_co_io_em(BlockDriverState
*bs
, int64_t sector_num
,
4855 int nb_sectors
, QEMUIOVector
*iov
,
4858 CoroutineIOCompletion co
= {
4859 .coroutine
= qemu_coroutine_self(),
4861 BlockDriverAIOCB
*acb
;
4864 acb
= bs
->drv
->bdrv_aio_writev(bs
, sector_num
, iov
, nb_sectors
,
4865 bdrv_co_io_em_complete
, &co
);
4867 acb
= bs
->drv
->bdrv_aio_readv(bs
, sector_num
, iov
, nb_sectors
,
4868 bdrv_co_io_em_complete
, &co
);
4871 trace_bdrv_co_io_em(bs
, sector_num
, nb_sectors
, is_write
, acb
);
4875 qemu_coroutine_yield();
4880 static int coroutine_fn
bdrv_co_readv_em(BlockDriverState
*bs
,
4881 int64_t sector_num
, int nb_sectors
,
4884 return bdrv_co_io_em(bs
, sector_num
, nb_sectors
, iov
, false);
4887 static int coroutine_fn
bdrv_co_writev_em(BlockDriverState
*bs
,
4888 int64_t sector_num
, int nb_sectors
,
4891 return bdrv_co_io_em(bs
, sector_num
, nb_sectors
, iov
, true);
4894 static void coroutine_fn
bdrv_flush_co_entry(void *opaque
)
4896 RwCo
*rwco
= opaque
;
4898 rwco
->ret
= bdrv_co_flush(rwco
->bs
);
4901 int coroutine_fn
bdrv_co_flush(BlockDriverState
*bs
)
4905 if (!bs
|| !bdrv_is_inserted(bs
) || bdrv_is_read_only(bs
)) {
4909 /* Write back cached data to the OS even with cache=unsafe */
4910 BLKDBG_EVENT(bs
->file
, BLKDBG_FLUSH_TO_OS
);
4911 if (bs
->drv
->bdrv_co_flush_to_os
) {
4912 ret
= bs
->drv
->bdrv_co_flush_to_os(bs
);
4918 /* But don't actually force it to the disk with cache=unsafe */
4919 if (bs
->open_flags
& BDRV_O_NO_FLUSH
) {
4923 BLKDBG_EVENT(bs
->file
, BLKDBG_FLUSH_TO_DISK
);
4924 if (bs
->drv
->bdrv_co_flush_to_disk
) {
4925 ret
= bs
->drv
->bdrv_co_flush_to_disk(bs
);
4926 } else if (bs
->drv
->bdrv_aio_flush
) {
4927 BlockDriverAIOCB
*acb
;
4928 CoroutineIOCompletion co
= {
4929 .coroutine
= qemu_coroutine_self(),
4932 acb
= bs
->drv
->bdrv_aio_flush(bs
, bdrv_co_io_em_complete
, &co
);
4936 qemu_coroutine_yield();
4941 * Some block drivers always operate in either writethrough or unsafe
4942 * mode and don't support bdrv_flush therefore. Usually qemu doesn't
4943 * know how the server works (because the behaviour is hardcoded or
4944 * depends on server-side configuration), so we can't ensure that
4945 * everything is safe on disk. Returning an error doesn't work because
4946 * that would break guests even if the server operates in writethrough
4949 * Let's hope the user knows what he's doing.
4957 /* Now flush the underlying protocol. It will also have BDRV_O_NO_FLUSH
4958 * in the case of cache=unsafe, so there are no useless flushes.
4961 return bdrv_co_flush(bs
->file
);
4964 void bdrv_invalidate_cache(BlockDriverState
*bs
, Error
**errp
)
4966 Error
*local_err
= NULL
;
4973 if (bs
->drv
->bdrv_invalidate_cache
) {
4974 bs
->drv
->bdrv_invalidate_cache(bs
, &local_err
);
4975 } else if (bs
->file
) {
4976 bdrv_invalidate_cache(bs
->file
, &local_err
);
4979 error_propagate(errp
, local_err
);
4983 ret
= refresh_total_sectors(bs
, bs
->total_sectors
);
4985 error_setg_errno(errp
, -ret
, "Could not refresh total sector count");
4990 void bdrv_invalidate_cache_all(Error
**errp
)
4992 BlockDriverState
*bs
;
4993 Error
*local_err
= NULL
;
4995 QTAILQ_FOREACH(bs
, &bdrv_states
, device_list
) {
4996 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
4998 aio_context_acquire(aio_context
);
4999 bdrv_invalidate_cache(bs
, &local_err
);
5000 aio_context_release(aio_context
);
5002 error_propagate(errp
, local_err
);
5008 void bdrv_clear_incoming_migration_all(void)
5010 BlockDriverState
*bs
;
5012 QTAILQ_FOREACH(bs
, &bdrv_states
, device_list
) {
5013 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
5015 aio_context_acquire(aio_context
);
5016 bs
->open_flags
= bs
->open_flags
& ~(BDRV_O_INCOMING
);
5017 aio_context_release(aio_context
);
5021 int bdrv_flush(BlockDriverState
*bs
)
5029 if (qemu_in_coroutine()) {
5030 /* Fast-path if already in coroutine context */
5031 bdrv_flush_co_entry(&rwco
);
5033 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
5035 co
= qemu_coroutine_create(bdrv_flush_co_entry
);
5036 qemu_coroutine_enter(co
, &rwco
);
5037 while (rwco
.ret
== NOT_DONE
) {
5038 aio_poll(aio_context
, true);
5045 typedef struct DiscardCo
{
5046 BlockDriverState
*bs
;
5051 static void coroutine_fn
bdrv_discard_co_entry(void *opaque
)
5053 DiscardCo
*rwco
= opaque
;
5055 rwco
->ret
= bdrv_co_discard(rwco
->bs
, rwco
->sector_num
, rwco
->nb_sectors
);
5058 /* if no limit is specified in the BlockLimits use a default
5059 * of 32768 512-byte sectors (16 MiB) per request.
5061 #define MAX_DISCARD_DEFAULT 32768
5063 int coroutine_fn
bdrv_co_discard(BlockDriverState
*bs
, int64_t sector_num
,
5070 } else if (bdrv_check_request(bs
, sector_num
, nb_sectors
)) {
5072 } else if (bs
->read_only
) {
5076 bdrv_reset_dirty(bs
, sector_num
, nb_sectors
);
5078 /* Do nothing if disabled. */
5079 if (!(bs
->open_flags
& BDRV_O_UNMAP
)) {
5083 if (!bs
->drv
->bdrv_co_discard
&& !bs
->drv
->bdrv_aio_discard
) {
5087 max_discard
= bs
->bl
.max_discard
? bs
->bl
.max_discard
: MAX_DISCARD_DEFAULT
;
5088 while (nb_sectors
> 0) {
5090 int num
= nb_sectors
;
5093 if (bs
->bl
.discard_alignment
&&
5094 num
>= bs
->bl
.discard_alignment
&&
5095 sector_num
% bs
->bl
.discard_alignment
) {
5096 if (num
> bs
->bl
.discard_alignment
) {
5097 num
= bs
->bl
.discard_alignment
;
5099 num
-= sector_num
% bs
->bl
.discard_alignment
;
5102 /* limit request size */
5103 if (num
> max_discard
) {
5107 if (bs
->drv
->bdrv_co_discard
) {
5108 ret
= bs
->drv
->bdrv_co_discard(bs
, sector_num
, num
);
5110 BlockDriverAIOCB
*acb
;
5111 CoroutineIOCompletion co
= {
5112 .coroutine
= qemu_coroutine_self(),
5115 acb
= bs
->drv
->bdrv_aio_discard(bs
, sector_num
, nb_sectors
,
5116 bdrv_co_io_em_complete
, &co
);
5120 qemu_coroutine_yield();
5124 if (ret
&& ret
!= -ENOTSUP
) {
5134 int bdrv_discard(BlockDriverState
*bs
, int64_t sector_num
, int nb_sectors
)
5139 .sector_num
= sector_num
,
5140 .nb_sectors
= nb_sectors
,
5144 if (qemu_in_coroutine()) {
5145 /* Fast-path if already in coroutine context */
5146 bdrv_discard_co_entry(&rwco
);
5148 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
5150 co
= qemu_coroutine_create(bdrv_discard_co_entry
);
5151 qemu_coroutine_enter(co
, &rwco
);
5152 while (rwco
.ret
== NOT_DONE
) {
5153 aio_poll(aio_context
, true);
5160 /**************************************************************/
5161 /* removable device support */
5164 * Return TRUE if the media is present
5166 int bdrv_is_inserted(BlockDriverState
*bs
)
5168 BlockDriver
*drv
= bs
->drv
;
5172 if (!drv
->bdrv_is_inserted
)
5174 return drv
->bdrv_is_inserted(bs
);
5178 * Return whether the media changed since the last call to this
5179 * function, or -ENOTSUP if we don't know. Most drivers don't know.
5181 int bdrv_media_changed(BlockDriverState
*bs
)
5183 BlockDriver
*drv
= bs
->drv
;
5185 if (drv
&& drv
->bdrv_media_changed
) {
5186 return drv
->bdrv_media_changed(bs
);
5192 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
5194 void bdrv_eject(BlockDriverState
*bs
, bool eject_flag
)
5196 BlockDriver
*drv
= bs
->drv
;
5198 if (drv
&& drv
->bdrv_eject
) {
5199 drv
->bdrv_eject(bs
, eject_flag
);
5202 if (bs
->device_name
[0] != '\0') {
5203 bdrv_emit_qmp_eject_event(bs
, eject_flag
);
5208 * Lock or unlock the media (if it is locked, the user won't be able
5209 * to eject it manually).
5211 void bdrv_lock_medium(BlockDriverState
*bs
, bool locked
)
5213 BlockDriver
*drv
= bs
->drv
;
5215 trace_bdrv_lock_medium(bs
, locked
);
5217 if (drv
&& drv
->bdrv_lock_medium
) {
5218 drv
->bdrv_lock_medium(bs
, locked
);
5222 /* needed for generic scsi interface */
5224 int bdrv_ioctl(BlockDriverState
*bs
, unsigned long int req
, void *buf
)
5226 BlockDriver
*drv
= bs
->drv
;
5228 if (drv
&& drv
->bdrv_ioctl
)
5229 return drv
->bdrv_ioctl(bs
, req
, buf
);
5233 BlockDriverAIOCB
*bdrv_aio_ioctl(BlockDriverState
*bs
,
5234 unsigned long int req
, void *buf
,
5235 BlockDriverCompletionFunc
*cb
, void *opaque
)
5237 BlockDriver
*drv
= bs
->drv
;
5239 if (drv
&& drv
->bdrv_aio_ioctl
)
5240 return drv
->bdrv_aio_ioctl(bs
, req
, buf
, cb
, opaque
);
5244 void bdrv_set_guest_block_size(BlockDriverState
*bs
, int align
)
5246 bs
->guest_block_size
= align
;
5249 void *qemu_blockalign(BlockDriverState
*bs
, size_t size
)
5251 return qemu_memalign(bdrv_opt_mem_align(bs
), size
);
5255 * Check if all memory in this vector is sector aligned.
5257 bool bdrv_qiov_is_aligned(BlockDriverState
*bs
, QEMUIOVector
*qiov
)
5260 size_t alignment
= bdrv_opt_mem_align(bs
);
5262 for (i
= 0; i
< qiov
->niov
; i
++) {
5263 if ((uintptr_t) qiov
->iov
[i
].iov_base
% alignment
) {
5266 if (qiov
->iov
[i
].iov_len
% alignment
) {
5274 BdrvDirtyBitmap
*bdrv_create_dirty_bitmap(BlockDriverState
*bs
, int granularity
,
5277 int64_t bitmap_size
;
5278 BdrvDirtyBitmap
*bitmap
;
5280 assert((granularity
& (granularity
- 1)) == 0);
5282 granularity
>>= BDRV_SECTOR_BITS
;
5283 assert(granularity
);
5284 bitmap_size
= bdrv_getlength(bs
);
5285 if (bitmap_size
< 0) {
5286 error_setg_errno(errp
, -bitmap_size
, "could not get length of device");
5287 errno
= -bitmap_size
;
5290 bitmap_size
>>= BDRV_SECTOR_BITS
;
5291 bitmap
= g_malloc0(sizeof(BdrvDirtyBitmap
));
5292 bitmap
->bitmap
= hbitmap_alloc(bitmap_size
, ffs(granularity
) - 1);
5293 QLIST_INSERT_HEAD(&bs
->dirty_bitmaps
, bitmap
, list
);
5297 void bdrv_release_dirty_bitmap(BlockDriverState
*bs
, BdrvDirtyBitmap
*bitmap
)
5299 BdrvDirtyBitmap
*bm
, *next
;
5300 QLIST_FOREACH_SAFE(bm
, &bs
->dirty_bitmaps
, list
, next
) {
5302 QLIST_REMOVE(bitmap
, list
);
5303 hbitmap_free(bitmap
->bitmap
);
5310 BlockDirtyInfoList
*bdrv_query_dirty_bitmaps(BlockDriverState
*bs
)
5312 BdrvDirtyBitmap
*bm
;
5313 BlockDirtyInfoList
*list
= NULL
;
5314 BlockDirtyInfoList
**plist
= &list
;
5316 QLIST_FOREACH(bm
, &bs
->dirty_bitmaps
, list
) {
5317 BlockDirtyInfo
*info
= g_malloc0(sizeof(BlockDirtyInfo
));
5318 BlockDirtyInfoList
*entry
= g_malloc0(sizeof(BlockDirtyInfoList
));
5319 info
->count
= bdrv_get_dirty_count(bs
, bm
);
5321 ((int64_t) BDRV_SECTOR_SIZE
<< hbitmap_granularity(bm
->bitmap
));
5322 entry
->value
= info
;
5324 plist
= &entry
->next
;
5330 int bdrv_get_dirty(BlockDriverState
*bs
, BdrvDirtyBitmap
*bitmap
, int64_t sector
)
5333 return hbitmap_get(bitmap
->bitmap
, sector
);
5339 void bdrv_dirty_iter_init(BlockDriverState
*bs
,
5340 BdrvDirtyBitmap
*bitmap
, HBitmapIter
*hbi
)
5342 hbitmap_iter_init(hbi
, bitmap
->bitmap
, 0);
5345 void bdrv_set_dirty(BlockDriverState
*bs
, int64_t cur_sector
,
5348 BdrvDirtyBitmap
*bitmap
;
5349 QLIST_FOREACH(bitmap
, &bs
->dirty_bitmaps
, list
) {
5350 hbitmap_set(bitmap
->bitmap
, cur_sector
, nr_sectors
);
5354 void bdrv_reset_dirty(BlockDriverState
*bs
, int64_t cur_sector
, int nr_sectors
)
5356 BdrvDirtyBitmap
*bitmap
;
5357 QLIST_FOREACH(bitmap
, &bs
->dirty_bitmaps
, list
) {
5358 hbitmap_reset(bitmap
->bitmap
, cur_sector
, nr_sectors
);
5362 int64_t bdrv_get_dirty_count(BlockDriverState
*bs
, BdrvDirtyBitmap
*bitmap
)
5364 return hbitmap_count(bitmap
->bitmap
);
5367 /* Get a reference to bs */
5368 void bdrv_ref(BlockDriverState
*bs
)
5373 /* Release a previously grabbed reference to bs.
5374 * If after releasing, reference count is zero, the BlockDriverState is
5376 void bdrv_unref(BlockDriverState
*bs
)
5378 assert(bs
->refcnt
> 0);
5379 if (--bs
->refcnt
== 0) {
5384 struct BdrvOpBlocker
{
5386 QLIST_ENTRY(BdrvOpBlocker
) list
;
5389 bool bdrv_op_is_blocked(BlockDriverState
*bs
, BlockOpType op
, Error
**errp
)
5391 BdrvOpBlocker
*blocker
;
5392 assert((int) op
>= 0 && op
< BLOCK_OP_TYPE_MAX
);
5393 if (!QLIST_EMPTY(&bs
->op_blockers
[op
])) {
5394 blocker
= QLIST_FIRST(&bs
->op_blockers
[op
]);
5396 error_setg(errp
, "Device '%s' is busy: %s",
5397 bs
->device_name
, error_get_pretty(blocker
->reason
));
5404 void bdrv_op_block(BlockDriverState
*bs
, BlockOpType op
, Error
*reason
)
5406 BdrvOpBlocker
*blocker
;
5407 assert((int) op
>= 0 && op
< BLOCK_OP_TYPE_MAX
);
5409 blocker
= g_malloc0(sizeof(BdrvOpBlocker
));
5410 blocker
->reason
= reason
;
5411 QLIST_INSERT_HEAD(&bs
->op_blockers
[op
], blocker
, list
);
5414 void bdrv_op_unblock(BlockDriverState
*bs
, BlockOpType op
, Error
*reason
)
5416 BdrvOpBlocker
*blocker
, *next
;
5417 assert((int) op
>= 0 && op
< BLOCK_OP_TYPE_MAX
);
5418 QLIST_FOREACH_SAFE(blocker
, &bs
->op_blockers
[op
], list
, next
) {
5419 if (blocker
->reason
== reason
) {
5420 QLIST_REMOVE(blocker
, list
);
5426 void bdrv_op_block_all(BlockDriverState
*bs
, Error
*reason
)
5429 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
5430 bdrv_op_block(bs
, i
, reason
);
5434 void bdrv_op_unblock_all(BlockDriverState
*bs
, Error
*reason
)
5437 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
5438 bdrv_op_unblock(bs
, i
, reason
);
5442 bool bdrv_op_blocker_is_empty(BlockDriverState
*bs
)
5446 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
5447 if (!QLIST_EMPTY(&bs
->op_blockers
[i
])) {
5454 void bdrv_iostatus_enable(BlockDriverState
*bs
)
5456 bs
->iostatus_enabled
= true;
5457 bs
->iostatus
= BLOCK_DEVICE_IO_STATUS_OK
;
5460 /* The I/O status is only enabled if the drive explicitly
5461 * enables it _and_ the VM is configured to stop on errors */
5462 bool bdrv_iostatus_is_enabled(const BlockDriverState
*bs
)
5464 return (bs
->iostatus_enabled
&&
5465 (bs
->on_write_error
== BLOCKDEV_ON_ERROR_ENOSPC
||
5466 bs
->on_write_error
== BLOCKDEV_ON_ERROR_STOP
||
5467 bs
->on_read_error
== BLOCKDEV_ON_ERROR_STOP
));
5470 void bdrv_iostatus_disable(BlockDriverState
*bs
)
5472 bs
->iostatus_enabled
= false;
5475 void bdrv_iostatus_reset(BlockDriverState
*bs
)
5477 if (bdrv_iostatus_is_enabled(bs
)) {
5478 bs
->iostatus
= BLOCK_DEVICE_IO_STATUS_OK
;
5480 block_job_iostatus_reset(bs
->job
);
5485 void bdrv_iostatus_set_err(BlockDriverState
*bs
, int error
)
5487 assert(bdrv_iostatus_is_enabled(bs
));
5488 if (bs
->iostatus
== BLOCK_DEVICE_IO_STATUS_OK
) {
5489 bs
->iostatus
= error
== ENOSPC
? BLOCK_DEVICE_IO_STATUS_NOSPACE
:
5490 BLOCK_DEVICE_IO_STATUS_FAILED
;
5495 bdrv_acct_start(BlockDriverState
*bs
, BlockAcctCookie
*cookie
, int64_t bytes
,
5496 enum BlockAcctType type
)
5498 assert(type
< BDRV_MAX_IOTYPE
);
5500 cookie
->bytes
= bytes
;
5501 cookie
->start_time_ns
= get_clock();
5502 cookie
->type
= type
;
5506 bdrv_acct_done(BlockDriverState
*bs
, BlockAcctCookie
*cookie
)
5508 assert(cookie
->type
< BDRV_MAX_IOTYPE
);
5510 bs
->nr_bytes
[cookie
->type
] += cookie
->bytes
;
5511 bs
->nr_ops
[cookie
->type
]++;
5512 bs
->total_time_ns
[cookie
->type
] += get_clock() - cookie
->start_time_ns
;
5515 void bdrv_img_create(const char *filename
, const char *fmt
,
5516 const char *base_filename
, const char *base_fmt
,
5517 char *options
, uint64_t img_size
, int flags
,
5518 Error
**errp
, bool quiet
)
5520 QEMUOptionParameter
*param
= NULL
, *create_options
= NULL
;
5521 QEMUOptionParameter
*backing_fmt
, *backing_file
, *size
;
5522 BlockDriver
*drv
, *proto_drv
;
5523 BlockDriver
*backing_drv
= NULL
;
5524 Error
*local_err
= NULL
;
5527 /* Find driver and parse its options */
5528 drv
= bdrv_find_format(fmt
);
5530 error_setg(errp
, "Unknown file format '%s'", fmt
);
5534 proto_drv
= bdrv_find_protocol(filename
, true);
5536 error_setg(errp
, "Unknown protocol '%s'", filename
);
5540 create_options
= append_option_parameters(create_options
,
5541 drv
->create_options
);
5542 create_options
= append_option_parameters(create_options
,
5543 proto_drv
->create_options
);
5545 /* Create parameter list with default values */
5546 param
= parse_option_parameters("", create_options
, param
);
5548 set_option_parameter_int(param
, BLOCK_OPT_SIZE
, img_size
);
5550 /* Parse -o options */
5552 param
= parse_option_parameters(options
, create_options
, param
);
5553 if (param
== NULL
) {
5554 error_setg(errp
, "Invalid options for file format '%s'.", fmt
);
5559 if (base_filename
) {
5560 if (set_option_parameter(param
, BLOCK_OPT_BACKING_FILE
,
5562 error_setg(errp
, "Backing file not supported for file format '%s'",
5569 if (set_option_parameter(param
, BLOCK_OPT_BACKING_FMT
, base_fmt
)) {
5570 error_setg(errp
, "Backing file format not supported for file "
5571 "format '%s'", fmt
);
5576 backing_file
= get_option_parameter(param
, BLOCK_OPT_BACKING_FILE
);
5577 if (backing_file
&& backing_file
->value
.s
) {
5578 if (!strcmp(filename
, backing_file
->value
.s
)) {
5579 error_setg(errp
, "Error: Trying to create an image with the "
5580 "same filename as the backing file");
5585 backing_fmt
= get_option_parameter(param
, BLOCK_OPT_BACKING_FMT
);
5586 if (backing_fmt
&& backing_fmt
->value
.s
) {
5587 backing_drv
= bdrv_find_format(backing_fmt
->value
.s
);
5589 error_setg(errp
, "Unknown backing file format '%s'",
5590 backing_fmt
->value
.s
);
5595 // The size for the image must always be specified, with one exception:
5596 // If we are using a backing file, we can obtain the size from there
5597 size
= get_option_parameter(param
, BLOCK_OPT_SIZE
);
5598 if (size
&& size
->value
.n
== -1) {
5599 if (backing_file
&& backing_file
->value
.s
) {
5600 BlockDriverState
*bs
;
5605 /* backing files always opened read-only */
5607 flags
& ~(BDRV_O_RDWR
| BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
);
5610 ret
= bdrv_open(&bs
, backing_file
->value
.s
, NULL
, NULL
, back_flags
,
5611 backing_drv
, &local_err
);
5613 error_setg_errno(errp
, -ret
, "Could not open '%s': %s",
5614 backing_file
->value
.s
,
5615 error_get_pretty(local_err
));
5616 error_free(local_err
);
5620 bdrv_get_geometry(bs
, &size
);
5623 snprintf(buf
, sizeof(buf
), "%" PRId64
, size
);
5624 set_option_parameter(param
, BLOCK_OPT_SIZE
, buf
);
5628 error_setg(errp
, "Image creation needs a size parameter");
5634 printf("Formatting '%s', fmt=%s ", filename
, fmt
);
5635 print_option_parameters(param
);
5638 ret
= bdrv_create(drv
, filename
, param
, &local_err
);
5639 if (ret
== -EFBIG
) {
5640 /* This is generally a better message than whatever the driver would
5641 * deliver (especially because of the cluster_size_hint), since that
5642 * is most probably not much different from "image too large". */
5643 const char *cluster_size_hint
= "";
5644 if (get_option_parameter(create_options
, BLOCK_OPT_CLUSTER_SIZE
)) {
5645 cluster_size_hint
= " (try using a larger cluster size)";
5647 error_setg(errp
, "The image size is too large for file format '%s'"
5648 "%s", fmt
, cluster_size_hint
);
5649 error_free(local_err
);
5654 free_option_parameters(create_options
);
5655 free_option_parameters(param
);
5658 error_propagate(errp
, local_err
);
5662 AioContext
*bdrv_get_aio_context(BlockDriverState
*bs
)
5664 /* Currently BlockDriverState always uses the main loop AioContext */
5665 return qemu_get_aio_context();
5668 void bdrv_add_before_write_notifier(BlockDriverState
*bs
,
5669 NotifierWithReturn
*notifier
)
5671 notifier_with_return_list_add(&bs
->before_write_notifiers
, notifier
);
5674 int bdrv_amend_options(BlockDriverState
*bs
, QEMUOptionParameter
*options
)
5676 if (bs
->drv
->bdrv_amend_options
== NULL
) {
5679 return bs
->drv
->bdrv_amend_options(bs
, options
);
5682 /* This function will be called by the bdrv_recurse_is_first_non_filter method
5683 * of block filter and by bdrv_is_first_non_filter.
5684 * It is used to test if the given bs is the candidate or recurse more in the
5687 bool bdrv_recurse_is_first_non_filter(BlockDriverState
*bs
,
5688 BlockDriverState
*candidate
)
5690 /* return false if basic checks fails */
5691 if (!bs
|| !bs
->drv
) {
5695 /* the code reached a non block filter driver -> check if the bs is
5696 * the same as the candidate. It's the recursion termination condition.
5698 if (!bs
->drv
->is_filter
) {
5699 return bs
== candidate
;
5701 /* Down this path the driver is a block filter driver */
5703 /* If the block filter recursion method is defined use it to recurse down
5706 if (bs
->drv
->bdrv_recurse_is_first_non_filter
) {
5707 return bs
->drv
->bdrv_recurse_is_first_non_filter(bs
, candidate
);
5710 /* the driver is a block filter but don't allow to recurse -> return false
5715 /* This function checks if the candidate is the first non filter bs down it's
5716 * bs chain. Since we don't have pointers to parents it explore all bs chains
5717 * from the top. Some filters can choose not to pass down the recursion.
5719 bool bdrv_is_first_non_filter(BlockDriverState
*candidate
)
5721 BlockDriverState
*bs
;
5723 /* walk down the bs forest recursively */
5724 QTAILQ_FOREACH(bs
, &bdrv_states
, device_list
) {
5727 /* try to recurse in this top level bs */
5728 perm
= bdrv_recurse_is_first_non_filter(bs
, candidate
);
5730 /* candidate is the first non filter */