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 "qemu/osdep.h"
25 #include "block/trace.h"
26 #include "block/block_int.h"
27 #include "block/blockjob.h"
28 #include "block/nbd.h"
29 #include "qemu/error-report.h"
30 #include "module_block.h"
31 #include "qemu/module.h"
32 #include "qapi/qmp/qerror.h"
33 #include "qapi/qmp/qbool.h"
34 #include "qapi/qmp/qjson.h"
35 #include "sysemu/block-backend.h"
36 #include "sysemu/sysemu.h"
37 #include "qemu/notify.h"
38 #include "qemu/coroutine.h"
39 #include "block/qapi.h"
40 #include "qmp-commands.h"
41 #include "qemu/timer.h"
42 #include "qapi-event.h"
43 #include "qemu/cutils.h"
45 #include "qapi/util.h"
48 #include <sys/ioctl.h>
49 #include <sys/queue.h>
59 #define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */
61 static QTAILQ_HEAD(, BlockDriverState
) graph_bdrv_states
=
62 QTAILQ_HEAD_INITIALIZER(graph_bdrv_states
);
64 static QTAILQ_HEAD(, BlockDriverState
) all_bdrv_states
=
65 QTAILQ_HEAD_INITIALIZER(all_bdrv_states
);
67 static QLIST_HEAD(, BlockDriver
) bdrv_drivers
=
68 QLIST_HEAD_INITIALIZER(bdrv_drivers
);
70 static BlockDriverState
*bdrv_open_inherit(const char *filename
,
71 const char *reference
,
72 QDict
*options
, int flags
,
73 BlockDriverState
*parent
,
74 const BdrvChildRole
*child_role
,
77 /* If non-zero, use only whitelisted block drivers */
78 static int use_bdrv_whitelist
;
81 static int is_windows_drive_prefix(const char *filename
)
83 return (((filename
[0] >= 'a' && filename
[0] <= 'z') ||
84 (filename
[0] >= 'A' && filename
[0] <= 'Z')) &&
88 int is_windows_drive(const char *filename
)
90 if (is_windows_drive_prefix(filename
) &&
93 if (strstart(filename
, "\\\\.\\", NULL
) ||
94 strstart(filename
, "//./", NULL
))
100 size_t bdrv_opt_mem_align(BlockDriverState
*bs
)
102 if (!bs
|| !bs
->drv
) {
103 /* page size or 4k (hdd sector size) should be on the safe side */
104 return MAX(4096, getpagesize());
107 return bs
->bl
.opt_mem_alignment
;
110 size_t bdrv_min_mem_align(BlockDriverState
*bs
)
112 if (!bs
|| !bs
->drv
) {
113 /* page size or 4k (hdd sector size) should be on the safe side */
114 return MAX(4096, getpagesize());
117 return bs
->bl
.min_mem_alignment
;
120 /* check if the path starts with "<protocol>:" */
121 int path_has_protocol(const char *path
)
126 if (is_windows_drive(path
) ||
127 is_windows_drive_prefix(path
)) {
130 p
= path
+ strcspn(path
, ":/\\");
132 p
= path
+ strcspn(path
, ":/");
138 int path_is_absolute(const char *path
)
141 /* specific case for names like: "\\.\d:" */
142 if (is_windows_drive(path
) || is_windows_drive_prefix(path
)) {
145 return (*path
== '/' || *path
== '\\');
147 return (*path
== '/');
151 /* if filename is absolute, just copy it to dest. Otherwise, build a
152 path to it by considering it is relative to base_path. URL are
154 void path_combine(char *dest
, int dest_size
,
155 const char *base_path
,
156 const char *filename
)
163 if (path_is_absolute(filename
)) {
164 pstrcpy(dest
, dest_size
, filename
);
166 const char *protocol_stripped
= NULL
;
168 if (path_has_protocol(base_path
)) {
169 protocol_stripped
= strchr(base_path
, ':');
170 if (protocol_stripped
) {
174 p
= protocol_stripped
?: base_path
;
176 p1
= strrchr(base_path
, '/');
180 p2
= strrchr(base_path
, '\\');
192 if (len
> dest_size
- 1)
194 memcpy(dest
, base_path
, len
);
196 pstrcat(dest
, dest_size
, filename
);
200 /* Returns whether the image file is opened as read-only. Note that this can
201 * return false and writing to the image file is still not possible because the
202 * image is inactivated. */
203 bool bdrv_is_read_only(BlockDriverState
*bs
)
205 return bs
->read_only
;
208 /* Returns whether the image file can be written to right now */
209 bool bdrv_is_writable(BlockDriverState
*bs
)
211 return !bdrv_is_read_only(bs
) && !(bs
->open_flags
& BDRV_O_INACTIVE
);
214 int bdrv_can_set_read_only(BlockDriverState
*bs
, bool read_only
, Error
**errp
)
216 /* Do not set read_only if copy_on_read is enabled */
217 if (bs
->copy_on_read
&& read_only
) {
218 error_setg(errp
, "Can't set node '%s' to r/o with copy-on-read enabled",
219 bdrv_get_device_or_node_name(bs
));
223 /* Do not clear read_only if it is prohibited */
224 if (!read_only
&& !(bs
->open_flags
& BDRV_O_ALLOW_RDWR
)) {
225 error_setg(errp
, "Node '%s' is read only",
226 bdrv_get_device_or_node_name(bs
));
233 int bdrv_set_read_only(BlockDriverState
*bs
, bool read_only
, Error
**errp
)
237 ret
= bdrv_can_set_read_only(bs
, read_only
, errp
);
242 bs
->read_only
= read_only
;
246 void bdrv_get_full_backing_filename_from_filename(const char *backed
,
248 char *dest
, size_t sz
,
251 if (backing
[0] == '\0' || path_has_protocol(backing
) ||
252 path_is_absolute(backing
))
254 pstrcpy(dest
, sz
, backing
);
255 } else if (backed
[0] == '\0' || strstart(backed
, "json:", NULL
)) {
256 error_setg(errp
, "Cannot use relative backing file names for '%s'",
259 path_combine(dest
, sz
, backed
, backing
);
263 void bdrv_get_full_backing_filename(BlockDriverState
*bs
, char *dest
, size_t sz
,
266 char *backed
= bs
->exact_filename
[0] ? bs
->exact_filename
: bs
->filename
;
268 bdrv_get_full_backing_filename_from_filename(backed
, bs
->backing_file
,
272 void bdrv_register(BlockDriver
*bdrv
)
274 QLIST_INSERT_HEAD(&bdrv_drivers
, bdrv
, list
);
277 BlockDriverState
*bdrv_new(void)
279 BlockDriverState
*bs
;
282 bs
= g_new0(BlockDriverState
, 1);
283 QLIST_INIT(&bs
->dirty_bitmaps
);
284 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
285 QLIST_INIT(&bs
->op_blockers
[i
]);
287 notifier_with_return_list_init(&bs
->before_write_notifiers
);
289 bs
->aio_context
= qemu_get_aio_context();
291 qemu_co_queue_init(&bs
->flush_queue
);
293 QTAILQ_INSERT_TAIL(&all_bdrv_states
, bs
, bs_list
);
298 static BlockDriver
*bdrv_do_find_format(const char *format_name
)
302 QLIST_FOREACH(drv1
, &bdrv_drivers
, list
) {
303 if (!strcmp(drv1
->format_name
, format_name
)) {
311 BlockDriver
*bdrv_find_format(const char *format_name
)
316 drv1
= bdrv_do_find_format(format_name
);
321 /* The driver isn't registered, maybe we need to load a module */
322 for (i
= 0; i
< (int)ARRAY_SIZE(block_driver_modules
); ++i
) {
323 if (!strcmp(block_driver_modules
[i
].format_name
, format_name
)) {
324 block_module_load_one(block_driver_modules
[i
].library_name
);
329 return bdrv_do_find_format(format_name
);
332 static int bdrv_is_whitelisted(BlockDriver
*drv
, bool read_only
)
334 static const char *whitelist_rw
[] = {
335 CONFIG_BDRV_RW_WHITELIST
337 static const char *whitelist_ro
[] = {
338 CONFIG_BDRV_RO_WHITELIST
342 if (!whitelist_rw
[0] && !whitelist_ro
[0]) {
343 return 1; /* no whitelist, anything goes */
346 for (p
= whitelist_rw
; *p
; p
++) {
347 if (!strcmp(drv
->format_name
, *p
)) {
352 for (p
= whitelist_ro
; *p
; p
++) {
353 if (!strcmp(drv
->format_name
, *p
)) {
361 bool bdrv_uses_whitelist(void)
363 return use_bdrv_whitelist
;
366 typedef struct CreateCo
{
374 static void coroutine_fn
bdrv_create_co_entry(void *opaque
)
376 Error
*local_err
= NULL
;
379 CreateCo
*cco
= opaque
;
382 ret
= cco
->drv
->bdrv_create(cco
->filename
, cco
->opts
, &local_err
);
383 error_propagate(&cco
->err
, local_err
);
387 int bdrv_create(BlockDriver
*drv
, const char* filename
,
388 QemuOpts
*opts
, Error
**errp
)
395 .filename
= g_strdup(filename
),
401 if (!drv
->bdrv_create
) {
402 error_setg(errp
, "Driver '%s' does not support image creation", drv
->format_name
);
407 if (qemu_in_coroutine()) {
408 /* Fast-path if already in coroutine context */
409 bdrv_create_co_entry(&cco
);
411 co
= qemu_coroutine_create(bdrv_create_co_entry
, &cco
);
412 qemu_coroutine_enter(co
);
413 while (cco
.ret
== NOT_DONE
) {
414 aio_poll(qemu_get_aio_context(), true);
421 error_propagate(errp
, cco
.err
);
423 error_setg_errno(errp
, -ret
, "Could not create image");
428 g_free(cco
.filename
);
432 int bdrv_create_file(const char *filename
, QemuOpts
*opts
, Error
**errp
)
435 Error
*local_err
= NULL
;
438 drv
= bdrv_find_protocol(filename
, true, errp
);
443 ret
= bdrv_create(drv
, filename
, opts
, &local_err
);
444 error_propagate(errp
, local_err
);
449 * Try to get @bs's logical and physical block size.
450 * On success, store them in @bsz struct and return 0.
451 * On failure return -errno.
452 * @bs must not be empty.
454 int bdrv_probe_blocksizes(BlockDriverState
*bs
, BlockSizes
*bsz
)
456 BlockDriver
*drv
= bs
->drv
;
458 if (drv
&& drv
->bdrv_probe_blocksizes
) {
459 return drv
->bdrv_probe_blocksizes(bs
, bsz
);
466 * Try to get @bs's geometry (cyls, heads, sectors).
467 * On success, store them in @geo struct and return 0.
468 * On failure return -errno.
469 * @bs must not be empty.
471 int bdrv_probe_geometry(BlockDriverState
*bs
, HDGeometry
*geo
)
473 BlockDriver
*drv
= bs
->drv
;
475 if (drv
&& drv
->bdrv_probe_geometry
) {
476 return drv
->bdrv_probe_geometry(bs
, geo
);
483 * Create a uniquely-named empty temporary file.
484 * Return 0 upon success, otherwise a negative errno value.
486 int get_tmp_filename(char *filename
, int size
)
489 char temp_dir
[MAX_PATH
];
490 /* GetTempFileName requires that its output buffer (4th param)
491 have length MAX_PATH or greater. */
492 assert(size
>= MAX_PATH
);
493 return (GetTempPath(MAX_PATH
, temp_dir
)
494 && GetTempFileName(temp_dir
, "qem", 0, filename
)
495 ? 0 : -GetLastError());
499 tmpdir
= getenv("TMPDIR");
503 if (snprintf(filename
, size
, "%s/vl.XXXXXX", tmpdir
) >= size
) {
506 fd
= mkstemp(filename
);
510 if (close(fd
) != 0) {
519 * Detect host devices. By convention, /dev/cdrom[N] is always
520 * recognized as a host CDROM.
522 static BlockDriver
*find_hdev_driver(const char *filename
)
524 int score_max
= 0, score
;
525 BlockDriver
*drv
= NULL
, *d
;
527 QLIST_FOREACH(d
, &bdrv_drivers
, list
) {
528 if (d
->bdrv_probe_device
) {
529 score
= d
->bdrv_probe_device(filename
);
530 if (score
> score_max
) {
540 static BlockDriver
*bdrv_do_find_protocol(const char *protocol
)
544 QLIST_FOREACH(drv1
, &bdrv_drivers
, list
) {
545 if (drv1
->protocol_name
&& !strcmp(drv1
->protocol_name
, protocol
)) {
553 BlockDriver
*bdrv_find_protocol(const char *filename
,
554 bool allow_protocol_prefix
,
563 /* TODO Drivers without bdrv_file_open must be specified explicitly */
566 * XXX(hch): we really should not let host device detection
567 * override an explicit protocol specification, but moving this
568 * later breaks access to device names with colons in them.
569 * Thanks to the brain-dead persistent naming schemes on udev-
570 * based Linux systems those actually are quite common.
572 drv1
= find_hdev_driver(filename
);
577 if (!path_has_protocol(filename
) || !allow_protocol_prefix
) {
581 p
= strchr(filename
, ':');
584 if (len
> sizeof(protocol
) - 1)
585 len
= sizeof(protocol
) - 1;
586 memcpy(protocol
, filename
, len
);
587 protocol
[len
] = '\0';
589 drv1
= bdrv_do_find_protocol(protocol
);
594 for (i
= 0; i
< (int)ARRAY_SIZE(block_driver_modules
); ++i
) {
595 if (block_driver_modules
[i
].protocol_name
&&
596 !strcmp(block_driver_modules
[i
].protocol_name
, protocol
)) {
597 block_module_load_one(block_driver_modules
[i
].library_name
);
602 drv1
= bdrv_do_find_protocol(protocol
);
604 error_setg(errp
, "Unknown protocol '%s'", protocol
);
610 * Guess image format by probing its contents.
611 * This is not a good idea when your image is raw (CVE-2008-2004), but
612 * we do it anyway for backward compatibility.
614 * @buf contains the image's first @buf_size bytes.
615 * @buf_size is the buffer size in bytes (generally BLOCK_PROBE_BUF_SIZE,
616 * but can be smaller if the image file is smaller)
617 * @filename is its filename.
619 * For all block drivers, call the bdrv_probe() method to get its
621 * Return the first block driver with the highest probing score.
623 BlockDriver
*bdrv_probe_all(const uint8_t *buf
, int buf_size
,
624 const char *filename
)
626 int score_max
= 0, score
;
627 BlockDriver
*drv
= NULL
, *d
;
629 QLIST_FOREACH(d
, &bdrv_drivers
, list
) {
631 score
= d
->bdrv_probe(buf
, buf_size
, filename
);
632 if (score
> score_max
) {
642 static int find_image_format(BlockBackend
*file
, const char *filename
,
643 BlockDriver
**pdrv
, Error
**errp
)
646 uint8_t buf
[BLOCK_PROBE_BUF_SIZE
];
649 /* Return the raw BlockDriver * to scsi-generic devices or empty drives */
650 if (blk_is_sg(file
) || !blk_is_inserted(file
) || blk_getlength(file
) == 0) {
655 ret
= blk_pread(file
, 0, buf
, sizeof(buf
));
657 error_setg_errno(errp
, -ret
, "Could not read image for determining its "
663 drv
= bdrv_probe_all(buf
, ret
, filename
);
665 error_setg(errp
, "Could not determine image format: No compatible "
674 * Set the current 'total_sectors' value
675 * Return 0 on success, -errno on error.
677 static int refresh_total_sectors(BlockDriverState
*bs
, int64_t hint
)
679 BlockDriver
*drv
= bs
->drv
;
681 /* Do not attempt drv->bdrv_getlength() on scsi-generic devices */
685 /* query actual device if possible, otherwise just trust the hint */
686 if (drv
->bdrv_getlength
) {
687 int64_t length
= drv
->bdrv_getlength(bs
);
691 hint
= DIV_ROUND_UP(length
, BDRV_SECTOR_SIZE
);
694 bs
->total_sectors
= hint
;
699 * Combines a QDict of new block driver @options with any missing options taken
700 * from @old_options, so that leaving out an option defaults to its old value.
702 static void bdrv_join_options(BlockDriverState
*bs
, QDict
*options
,
705 if (bs
->drv
&& bs
->drv
->bdrv_join_options
) {
706 bs
->drv
->bdrv_join_options(options
, old_options
);
708 qdict_join(options
, old_options
, false);
713 * Set open flags for a given discard mode
715 * Return 0 on success, -1 if the discard mode was invalid.
717 int bdrv_parse_discard_flags(const char *mode
, int *flags
)
719 *flags
&= ~BDRV_O_UNMAP
;
721 if (!strcmp(mode
, "off") || !strcmp(mode
, "ignore")) {
723 } else if (!strcmp(mode
, "on") || !strcmp(mode
, "unmap")) {
724 *flags
|= BDRV_O_UNMAP
;
733 * Set open flags for a given cache mode
735 * Return 0 on success, -1 if the cache mode was invalid.
737 int bdrv_parse_cache_mode(const char *mode
, int *flags
, bool *writethrough
)
739 *flags
&= ~BDRV_O_CACHE_MASK
;
741 if (!strcmp(mode
, "off") || !strcmp(mode
, "none")) {
742 *writethrough
= false;
743 *flags
|= BDRV_O_NOCACHE
;
744 } else if (!strcmp(mode
, "directsync")) {
745 *writethrough
= true;
746 *flags
|= BDRV_O_NOCACHE
;
747 } else if (!strcmp(mode
, "writeback")) {
748 *writethrough
= false;
749 } else if (!strcmp(mode
, "unsafe")) {
750 *writethrough
= false;
751 *flags
|= BDRV_O_NO_FLUSH
;
752 } else if (!strcmp(mode
, "writethrough")) {
753 *writethrough
= true;
761 static char *bdrv_child_get_parent_desc(BdrvChild
*c
)
763 BlockDriverState
*parent
= c
->opaque
;
764 return g_strdup(bdrv_get_device_or_node_name(parent
));
767 static void bdrv_child_cb_drained_begin(BdrvChild
*child
)
769 BlockDriverState
*bs
= child
->opaque
;
770 bdrv_drained_begin(bs
);
773 static void bdrv_child_cb_drained_end(BdrvChild
*child
)
775 BlockDriverState
*bs
= child
->opaque
;
776 bdrv_drained_end(bs
);
779 static int bdrv_child_cb_inactivate(BdrvChild
*child
)
781 BlockDriverState
*bs
= child
->opaque
;
782 assert(bs
->open_flags
& BDRV_O_INACTIVE
);
787 * Returns the options and flags that a temporary snapshot should get, based on
788 * the originally requested flags (the originally requested image will have
789 * flags like a backing file)
791 static void bdrv_temp_snapshot_options(int *child_flags
, QDict
*child_options
,
792 int parent_flags
, QDict
*parent_options
)
794 *child_flags
= (parent_flags
& ~BDRV_O_SNAPSHOT
) | BDRV_O_TEMPORARY
;
796 /* For temporary files, unconditional cache=unsafe is fine */
797 qdict_set_default_str(child_options
, BDRV_OPT_CACHE_DIRECT
, "off");
798 qdict_set_default_str(child_options
, BDRV_OPT_CACHE_NO_FLUSH
, "on");
800 /* Copy the read-only option from the parent */
801 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_READ_ONLY
);
803 /* aio=native doesn't work for cache.direct=off, so disable it for the
804 * temporary snapshot */
805 *child_flags
&= ~BDRV_O_NATIVE_AIO
;
809 * Returns the options and flags that bs->file should get if a protocol driver
810 * is expected, based on the given options and flags for the parent BDS
812 static void bdrv_inherited_options(int *child_flags
, QDict
*child_options
,
813 int parent_flags
, QDict
*parent_options
)
815 int flags
= parent_flags
;
817 /* Enable protocol handling, disable format probing for bs->file */
818 flags
|= BDRV_O_PROTOCOL
;
820 /* If the cache mode isn't explicitly set, inherit direct and no-flush from
822 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_CACHE_DIRECT
);
823 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_CACHE_NO_FLUSH
);
824 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_FORCE_SHARE
);
826 /* Inherit the read-only option from the parent if it's not set */
827 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_READ_ONLY
);
829 /* Our block drivers take care to send flushes and respect unmap policy,
830 * so we can default to enable both on lower layers regardless of the
831 * corresponding parent options. */
832 qdict_set_default_str(child_options
, BDRV_OPT_DISCARD
, "unmap");
834 /* Clear flags that only apply to the top layer */
835 flags
&= ~(BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
| BDRV_O_COPY_ON_READ
|
838 *child_flags
= flags
;
841 const BdrvChildRole child_file
= {
842 .get_parent_desc
= bdrv_child_get_parent_desc
,
843 .inherit_options
= bdrv_inherited_options
,
844 .drained_begin
= bdrv_child_cb_drained_begin
,
845 .drained_end
= bdrv_child_cb_drained_end
,
846 .inactivate
= bdrv_child_cb_inactivate
,
850 * Returns the options and flags that bs->file should get if the use of formats
851 * (and not only protocols) is permitted for it, based on the given options and
852 * flags for the parent BDS
854 static void bdrv_inherited_fmt_options(int *child_flags
, QDict
*child_options
,
855 int parent_flags
, QDict
*parent_options
)
857 child_file
.inherit_options(child_flags
, child_options
,
858 parent_flags
, parent_options
);
860 *child_flags
&= ~(BDRV_O_PROTOCOL
| BDRV_O_NO_IO
);
863 const BdrvChildRole child_format
= {
864 .get_parent_desc
= bdrv_child_get_parent_desc
,
865 .inherit_options
= bdrv_inherited_fmt_options
,
866 .drained_begin
= bdrv_child_cb_drained_begin
,
867 .drained_end
= bdrv_child_cb_drained_end
,
868 .inactivate
= bdrv_child_cb_inactivate
,
871 static void bdrv_backing_attach(BdrvChild
*c
)
873 BlockDriverState
*parent
= c
->opaque
;
874 BlockDriverState
*backing_hd
= c
->bs
;
876 assert(!parent
->backing_blocker
);
877 error_setg(&parent
->backing_blocker
,
878 "node is used as backing hd of '%s'",
879 bdrv_get_device_or_node_name(parent
));
881 parent
->open_flags
&= ~BDRV_O_NO_BACKING
;
882 pstrcpy(parent
->backing_file
, sizeof(parent
->backing_file
),
883 backing_hd
->filename
);
884 pstrcpy(parent
->backing_format
, sizeof(parent
->backing_format
),
885 backing_hd
->drv
? backing_hd
->drv
->format_name
: "");
887 bdrv_op_block_all(backing_hd
, parent
->backing_blocker
);
888 /* Otherwise we won't be able to commit or stream */
889 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_COMMIT_TARGET
,
890 parent
->backing_blocker
);
891 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_STREAM
,
892 parent
->backing_blocker
);
894 * We do backup in 3 ways:
896 * The target bs is new opened, and the source is top BDS
898 * Both the source and the target are top BDSes.
899 * 3. internal backup(used for block replication)
900 * Both the source and the target are backing file
902 * In case 1 and 2, neither the source nor the target is the backing file.
903 * In case 3, we will block the top BDS, so there is only one block job
904 * for the top BDS and its backing chain.
906 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_BACKUP_SOURCE
,
907 parent
->backing_blocker
);
908 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_BACKUP_TARGET
,
909 parent
->backing_blocker
);
912 static void bdrv_backing_detach(BdrvChild
*c
)
914 BlockDriverState
*parent
= c
->opaque
;
916 assert(parent
->backing_blocker
);
917 bdrv_op_unblock_all(c
->bs
, parent
->backing_blocker
);
918 error_free(parent
->backing_blocker
);
919 parent
->backing_blocker
= NULL
;
923 * Returns the options and flags that bs->backing should get, based on the
924 * given options and flags for the parent BDS
926 static void bdrv_backing_options(int *child_flags
, QDict
*child_options
,
927 int parent_flags
, QDict
*parent_options
)
929 int flags
= parent_flags
;
931 /* The cache mode is inherited unmodified for backing files; except WCE,
932 * which is only applied on the top level (BlockBackend) */
933 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_CACHE_DIRECT
);
934 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_CACHE_NO_FLUSH
);
935 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_FORCE_SHARE
);
937 /* backing files always opened read-only */
938 qdict_set_default_str(child_options
, BDRV_OPT_READ_ONLY
, "on");
939 flags
&= ~BDRV_O_COPY_ON_READ
;
941 /* snapshot=on is handled on the top layer */
942 flags
&= ~(BDRV_O_SNAPSHOT
| BDRV_O_TEMPORARY
);
944 *child_flags
= flags
;
947 const BdrvChildRole child_backing
= {
948 .get_parent_desc
= bdrv_child_get_parent_desc
,
949 .attach
= bdrv_backing_attach
,
950 .detach
= bdrv_backing_detach
,
951 .inherit_options
= bdrv_backing_options
,
952 .drained_begin
= bdrv_child_cb_drained_begin
,
953 .drained_end
= bdrv_child_cb_drained_end
,
954 .inactivate
= bdrv_child_cb_inactivate
,
957 static int bdrv_open_flags(BlockDriverState
*bs
, int flags
)
959 int open_flags
= flags
;
962 * Clear flags that are internal to the block layer before opening the
965 open_flags
&= ~(BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
| BDRV_O_PROTOCOL
);
968 * Snapshots should be writable.
970 if (flags
& BDRV_O_TEMPORARY
) {
971 open_flags
|= BDRV_O_RDWR
;
977 static void update_flags_from_options(int *flags
, QemuOpts
*opts
)
979 *flags
&= ~BDRV_O_CACHE_MASK
;
981 assert(qemu_opt_find(opts
, BDRV_OPT_CACHE_NO_FLUSH
));
982 if (qemu_opt_get_bool(opts
, BDRV_OPT_CACHE_NO_FLUSH
, false)) {
983 *flags
|= BDRV_O_NO_FLUSH
;
986 assert(qemu_opt_find(opts
, BDRV_OPT_CACHE_DIRECT
));
987 if (qemu_opt_get_bool(opts
, BDRV_OPT_CACHE_DIRECT
, false)) {
988 *flags
|= BDRV_O_NOCACHE
;
991 *flags
&= ~BDRV_O_RDWR
;
993 assert(qemu_opt_find(opts
, BDRV_OPT_READ_ONLY
));
994 if (!qemu_opt_get_bool(opts
, BDRV_OPT_READ_ONLY
, false)) {
995 *flags
|= BDRV_O_RDWR
;
1000 static void update_options_from_flags(QDict
*options
, int flags
)
1002 if (!qdict_haskey(options
, BDRV_OPT_CACHE_DIRECT
)) {
1003 qdict_put_bool(options
, BDRV_OPT_CACHE_DIRECT
, flags
& BDRV_O_NOCACHE
);
1005 if (!qdict_haskey(options
, BDRV_OPT_CACHE_NO_FLUSH
)) {
1006 qdict_put_bool(options
, BDRV_OPT_CACHE_NO_FLUSH
,
1007 flags
& BDRV_O_NO_FLUSH
);
1009 if (!qdict_haskey(options
, BDRV_OPT_READ_ONLY
)) {
1010 qdict_put_bool(options
, BDRV_OPT_READ_ONLY
, !(flags
& BDRV_O_RDWR
));
1014 static void bdrv_assign_node_name(BlockDriverState
*bs
,
1015 const char *node_name
,
1018 char *gen_node_name
= NULL
;
1021 node_name
= gen_node_name
= id_generate(ID_BLOCK
);
1022 } else if (!id_wellformed(node_name
)) {
1024 * Check for empty string or invalid characters, but not if it is
1025 * generated (generated names use characters not available to the user)
1027 error_setg(errp
, "Invalid node name");
1031 /* takes care of avoiding namespaces collisions */
1032 if (blk_by_name(node_name
)) {
1033 error_setg(errp
, "node-name=%s is conflicting with a device id",
1038 /* takes care of avoiding duplicates node names */
1039 if (bdrv_find_node(node_name
)) {
1040 error_setg(errp
, "Duplicate node name");
1044 /* copy node name into the bs and insert it into the graph list */
1045 pstrcpy(bs
->node_name
, sizeof(bs
->node_name
), node_name
);
1046 QTAILQ_INSERT_TAIL(&graph_bdrv_states
, bs
, node_list
);
1048 g_free(gen_node_name
);
1051 static int bdrv_open_driver(BlockDriverState
*bs
, BlockDriver
*drv
,
1052 const char *node_name
, QDict
*options
,
1053 int open_flags
, Error
**errp
)
1055 Error
*local_err
= NULL
;
1058 bdrv_assign_node_name(bs
, node_name
, &local_err
);
1060 error_propagate(errp
, local_err
);
1065 bs
->read_only
= !(bs
->open_flags
& BDRV_O_RDWR
);
1066 bs
->opaque
= g_malloc0(drv
->instance_size
);
1068 if (drv
->bdrv_file_open
) {
1069 assert(!drv
->bdrv_needs_filename
|| bs
->filename
[0]);
1070 ret
= drv
->bdrv_file_open(bs
, options
, open_flags
, &local_err
);
1071 } else if (drv
->bdrv_open
) {
1072 ret
= drv
->bdrv_open(bs
, options
, open_flags
, &local_err
);
1079 error_propagate(errp
, local_err
);
1080 } else if (bs
->filename
[0]) {
1081 error_setg_errno(errp
, -ret
, "Could not open '%s'", bs
->filename
);
1083 error_setg_errno(errp
, -ret
, "Could not open image");
1088 ret
= refresh_total_sectors(bs
, bs
->total_sectors
);
1090 error_setg_errno(errp
, -ret
, "Could not refresh total sector count");
1094 bdrv_refresh_limits(bs
, &local_err
);
1096 error_propagate(errp
, local_err
);
1101 assert(bdrv_opt_mem_align(bs
) != 0);
1102 assert(bdrv_min_mem_align(bs
) != 0);
1103 assert(is_power_of_2(bs
->bl
.request_alignment
));
1108 /* FIXME Close bs first if already opened*/
1115 BlockDriverState
*bdrv_new_open_driver(BlockDriver
*drv
, const char *node_name
,
1116 int flags
, Error
**errp
)
1118 BlockDriverState
*bs
;
1122 bs
->open_flags
= flags
;
1123 bs
->explicit_options
= qdict_new();
1124 bs
->options
= qdict_new();
1127 update_options_from_flags(bs
->options
, flags
);
1129 ret
= bdrv_open_driver(bs
, drv
, node_name
, bs
->options
, flags
, errp
);
1131 QDECREF(bs
->explicit_options
);
1132 QDECREF(bs
->options
);
1140 QemuOptsList bdrv_runtime_opts
= {
1141 .name
= "bdrv_common",
1142 .head
= QTAILQ_HEAD_INITIALIZER(bdrv_runtime_opts
.head
),
1145 .name
= "node-name",
1146 .type
= QEMU_OPT_STRING
,
1147 .help
= "Node name of the block device node",
1151 .type
= QEMU_OPT_STRING
,
1152 .help
= "Block driver to use for the node",
1155 .name
= BDRV_OPT_CACHE_DIRECT
,
1156 .type
= QEMU_OPT_BOOL
,
1157 .help
= "Bypass software writeback cache on the host",
1160 .name
= BDRV_OPT_CACHE_NO_FLUSH
,
1161 .type
= QEMU_OPT_BOOL
,
1162 .help
= "Ignore flush requests",
1165 .name
= BDRV_OPT_READ_ONLY
,
1166 .type
= QEMU_OPT_BOOL
,
1167 .help
= "Node is opened in read-only mode",
1170 .name
= "detect-zeroes",
1171 .type
= QEMU_OPT_STRING
,
1172 .help
= "try to optimize zero writes (off, on, unmap)",
1176 .type
= QEMU_OPT_STRING
,
1177 .help
= "discard operation (ignore/off, unmap/on)",
1180 .name
= BDRV_OPT_FORCE_SHARE
,
1181 .type
= QEMU_OPT_BOOL
,
1182 .help
= "always accept other writers (default: off)",
1184 { /* end of list */ }
1189 * Common part for opening disk images and files
1191 * Removes all processed options from *options.
1193 static int bdrv_open_common(BlockDriverState
*bs
, BlockBackend
*file
,
1194 QDict
*options
, Error
**errp
)
1196 int ret
, open_flags
;
1197 const char *filename
;
1198 const char *driver_name
= NULL
;
1199 const char *node_name
= NULL
;
1200 const char *discard
;
1201 const char *detect_zeroes
;
1204 Error
*local_err
= NULL
;
1206 assert(bs
->file
== NULL
);
1207 assert(options
!= NULL
&& bs
->options
!= options
);
1209 opts
= qemu_opts_create(&bdrv_runtime_opts
, NULL
, 0, &error_abort
);
1210 qemu_opts_absorb_qdict(opts
, options
, &local_err
);
1212 error_propagate(errp
, local_err
);
1217 update_flags_from_options(&bs
->open_flags
, opts
);
1219 driver_name
= qemu_opt_get(opts
, "driver");
1220 drv
= bdrv_find_format(driver_name
);
1221 assert(drv
!= NULL
);
1223 bs
->force_share
= qemu_opt_get_bool(opts
, BDRV_OPT_FORCE_SHARE
, false);
1225 if (bs
->force_share
&& (bs
->open_flags
& BDRV_O_RDWR
)) {
1227 BDRV_OPT_FORCE_SHARE
1228 "=on can only be used with read-only images");
1234 filename
= blk_bs(file
)->filename
;
1237 * Caution: while qdict_get_try_str() is fine, getting
1238 * non-string types would require more care. When @options
1239 * come from -blockdev or blockdev_add, its members are typed
1240 * according to the QAPI schema, but when they come from
1241 * -drive, they're all QString.
1243 filename
= qdict_get_try_str(options
, "filename");
1246 if (drv
->bdrv_needs_filename
&& (!filename
|| !filename
[0])) {
1247 error_setg(errp
, "The '%s' block driver requires a file name",
1253 trace_bdrv_open_common(bs
, filename
?: "", bs
->open_flags
,
1256 bs
->read_only
= !(bs
->open_flags
& BDRV_O_RDWR
);
1258 if (use_bdrv_whitelist
&& !bdrv_is_whitelisted(drv
, bs
->read_only
)) {
1260 !bs
->read_only
&& bdrv_is_whitelisted(drv
, true)
1261 ? "Driver '%s' can only be used for read-only devices"
1262 : "Driver '%s' is not whitelisted",
1268 assert(bs
->copy_on_read
== 0); /* bdrv_new() and bdrv_close() make it so */
1269 if (bs
->open_flags
& BDRV_O_COPY_ON_READ
) {
1270 if (!bs
->read_only
) {
1271 bdrv_enable_copy_on_read(bs
);
1273 error_setg(errp
, "Can't use copy-on-read on read-only device");
1279 discard
= qemu_opt_get(opts
, "discard");
1280 if (discard
!= NULL
) {
1281 if (bdrv_parse_discard_flags(discard
, &bs
->open_flags
) != 0) {
1282 error_setg(errp
, "Invalid discard option");
1288 detect_zeroes
= qemu_opt_get(opts
, "detect-zeroes");
1289 if (detect_zeroes
) {
1290 BlockdevDetectZeroesOptions value
=
1291 qapi_enum_parse(BlockdevDetectZeroesOptions_lookup
,
1293 BLOCKDEV_DETECT_ZEROES_OPTIONS__MAX
,
1294 BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF
,
1297 error_propagate(errp
, local_err
);
1302 if (value
== BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP
&&
1303 !(bs
->open_flags
& BDRV_O_UNMAP
))
1305 error_setg(errp
, "setting detect-zeroes to unmap is not allowed "
1306 "without setting discard operation to unmap");
1311 bs
->detect_zeroes
= value
;
1314 if (filename
!= NULL
) {
1315 pstrcpy(bs
->filename
, sizeof(bs
->filename
), filename
);
1317 bs
->filename
[0] = '\0';
1319 pstrcpy(bs
->exact_filename
, sizeof(bs
->exact_filename
), bs
->filename
);
1321 /* Open the image, either directly or using a protocol */
1322 open_flags
= bdrv_open_flags(bs
, bs
->open_flags
);
1323 node_name
= qemu_opt_get(opts
, "node-name");
1325 assert(!drv
->bdrv_file_open
|| file
== NULL
);
1326 ret
= bdrv_open_driver(bs
, drv
, node_name
, options
, open_flags
, errp
);
1331 qemu_opts_del(opts
);
1335 qemu_opts_del(opts
);
1339 static QDict
*parse_json_filename(const char *filename
, Error
**errp
)
1341 QObject
*options_obj
;
1345 ret
= strstart(filename
, "json:", &filename
);
1348 options_obj
= qobject_from_json(filename
, errp
);
1350 /* Work around qobject_from_json() lossage TODO fix that */
1351 if (errp
&& !*errp
) {
1352 error_setg(errp
, "Could not parse the JSON options");
1355 error_prepend(errp
, "Could not parse the JSON options: ");
1359 options
= qobject_to_qdict(options_obj
);
1361 qobject_decref(options_obj
);
1362 error_setg(errp
, "Invalid JSON object given");
1366 qdict_flatten(options
);
1371 static void parse_json_protocol(QDict
*options
, const char **pfilename
,
1374 QDict
*json_options
;
1375 Error
*local_err
= NULL
;
1377 /* Parse json: pseudo-protocol */
1378 if (!*pfilename
|| !g_str_has_prefix(*pfilename
, "json:")) {
1382 json_options
= parse_json_filename(*pfilename
, &local_err
);
1384 error_propagate(errp
, local_err
);
1388 /* Options given in the filename have lower priority than options
1389 * specified directly */
1390 qdict_join(options
, json_options
, false);
1391 QDECREF(json_options
);
1396 * Fills in default options for opening images and converts the legacy
1397 * filename/flags pair to option QDict entries.
1398 * The BDRV_O_PROTOCOL flag in *flags will be set or cleared accordingly if a
1399 * block driver has been specified explicitly.
1401 static int bdrv_fill_options(QDict
**options
, const char *filename
,
1402 int *flags
, Error
**errp
)
1404 const char *drvname
;
1405 bool protocol
= *flags
& BDRV_O_PROTOCOL
;
1406 bool parse_filename
= false;
1407 BlockDriver
*drv
= NULL
;
1408 Error
*local_err
= NULL
;
1411 * Caution: while qdict_get_try_str() is fine, getting non-string
1412 * types would require more care. When @options come from
1413 * -blockdev or blockdev_add, its members are typed according to
1414 * the QAPI schema, but when they come from -drive, they're all
1417 drvname
= qdict_get_try_str(*options
, "driver");
1419 drv
= bdrv_find_format(drvname
);
1421 error_setg(errp
, "Unknown driver '%s'", drvname
);
1424 /* If the user has explicitly specified the driver, this choice should
1425 * override the BDRV_O_PROTOCOL flag */
1426 protocol
= drv
->bdrv_file_open
;
1430 *flags
|= BDRV_O_PROTOCOL
;
1432 *flags
&= ~BDRV_O_PROTOCOL
;
1435 /* Translate cache options from flags into options */
1436 update_options_from_flags(*options
, *flags
);
1438 /* Fetch the file name from the options QDict if necessary */
1439 if (protocol
&& filename
) {
1440 if (!qdict_haskey(*options
, "filename")) {
1441 qdict_put_str(*options
, "filename", filename
);
1442 parse_filename
= true;
1444 error_setg(errp
, "Can't specify 'file' and 'filename' options at "
1450 /* Find the right block driver */
1451 /* See cautionary note on accessing @options above */
1452 filename
= qdict_get_try_str(*options
, "filename");
1454 if (!drvname
&& protocol
) {
1456 drv
= bdrv_find_protocol(filename
, parse_filename
, errp
);
1461 drvname
= drv
->format_name
;
1462 qdict_put_str(*options
, "driver", drvname
);
1464 error_setg(errp
, "Must specify either driver or file");
1469 assert(drv
|| !protocol
);
1471 /* Driver-specific filename parsing */
1472 if (drv
&& drv
->bdrv_parse_filename
&& parse_filename
) {
1473 drv
->bdrv_parse_filename(filename
, *options
, &local_err
);
1475 error_propagate(errp
, local_err
);
1479 if (!drv
->bdrv_needs_filename
) {
1480 qdict_del(*options
, "filename");
1487 static int bdrv_child_check_perm(BdrvChild
*c
, uint64_t perm
, uint64_t shared
,
1488 GSList
*ignore_children
, Error
**errp
);
1489 static void bdrv_child_abort_perm_update(BdrvChild
*c
);
1490 static void bdrv_child_set_perm(BdrvChild
*c
, uint64_t perm
, uint64_t shared
);
1492 static void bdrv_child_perm(BlockDriverState
*bs
, BlockDriverState
*child_bs
,
1494 const BdrvChildRole
*role
,
1495 uint64_t parent_perm
, uint64_t parent_shared
,
1496 uint64_t *nperm
, uint64_t *nshared
)
1498 if (bs
->drv
&& bs
->drv
->bdrv_child_perm
) {
1499 bs
->drv
->bdrv_child_perm(bs
, c
, role
,
1500 parent_perm
, parent_shared
,
1503 if (child_bs
&& child_bs
->force_share
) {
1504 *nshared
= BLK_PERM_ALL
;
1509 * Check whether permissions on this node can be changed in a way that
1510 * @cumulative_perms and @cumulative_shared_perms are the new cumulative
1511 * permissions of all its parents. This involves checking whether all necessary
1512 * permission changes to child nodes can be performed.
1514 * A call to this function must always be followed by a call to bdrv_set_perm()
1515 * or bdrv_abort_perm_update().
1517 static int bdrv_check_perm(BlockDriverState
*bs
, uint64_t cumulative_perms
,
1518 uint64_t cumulative_shared_perms
,
1519 GSList
*ignore_children
, Error
**errp
)
1521 BlockDriver
*drv
= bs
->drv
;
1525 /* Write permissions never work with read-only images */
1526 if ((cumulative_perms
& (BLK_PERM_WRITE
| BLK_PERM_WRITE_UNCHANGED
)) &&
1527 !bdrv_is_writable(bs
))
1529 error_setg(errp
, "Block node is read-only");
1533 /* Check this node */
1538 if (drv
->bdrv_check_perm
) {
1539 return drv
->bdrv_check_perm(bs
, cumulative_perms
,
1540 cumulative_shared_perms
, errp
);
1543 /* Drivers that never have children can omit .bdrv_child_perm() */
1544 if (!drv
->bdrv_child_perm
) {
1545 assert(QLIST_EMPTY(&bs
->children
));
1549 /* Check all children */
1550 QLIST_FOREACH(c
, &bs
->children
, next
) {
1551 uint64_t cur_perm
, cur_shared
;
1552 bdrv_child_perm(bs
, c
->bs
, c
, c
->role
,
1553 cumulative_perms
, cumulative_shared_perms
,
1554 &cur_perm
, &cur_shared
);
1555 ret
= bdrv_child_check_perm(c
, cur_perm
, cur_shared
, ignore_children
,
1566 * Notifies drivers that after a previous bdrv_check_perm() call, the
1567 * permission update is not performed and any preparations made for it (e.g.
1568 * taken file locks) need to be undone.
1570 * This function recursively notifies all child nodes.
1572 static void bdrv_abort_perm_update(BlockDriverState
*bs
)
1574 BlockDriver
*drv
= bs
->drv
;
1581 if (drv
->bdrv_abort_perm_update
) {
1582 drv
->bdrv_abort_perm_update(bs
);
1585 QLIST_FOREACH(c
, &bs
->children
, next
) {
1586 bdrv_child_abort_perm_update(c
);
1590 static void bdrv_set_perm(BlockDriverState
*bs
, uint64_t cumulative_perms
,
1591 uint64_t cumulative_shared_perms
)
1593 BlockDriver
*drv
= bs
->drv
;
1600 /* Update this node */
1601 if (drv
->bdrv_set_perm
) {
1602 drv
->bdrv_set_perm(bs
, cumulative_perms
, cumulative_shared_perms
);
1605 /* Drivers that never have children can omit .bdrv_child_perm() */
1606 if (!drv
->bdrv_child_perm
) {
1607 assert(QLIST_EMPTY(&bs
->children
));
1611 /* Update all children */
1612 QLIST_FOREACH(c
, &bs
->children
, next
) {
1613 uint64_t cur_perm
, cur_shared
;
1614 bdrv_child_perm(bs
, c
->bs
, c
, c
->role
,
1615 cumulative_perms
, cumulative_shared_perms
,
1616 &cur_perm
, &cur_shared
);
1617 bdrv_child_set_perm(c
, cur_perm
, cur_shared
);
1621 static void bdrv_get_cumulative_perm(BlockDriverState
*bs
, uint64_t *perm
,
1622 uint64_t *shared_perm
)
1625 uint64_t cumulative_perms
= 0;
1626 uint64_t cumulative_shared_perms
= BLK_PERM_ALL
;
1628 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
1629 cumulative_perms
|= c
->perm
;
1630 cumulative_shared_perms
&= c
->shared_perm
;
1633 *perm
= cumulative_perms
;
1634 *shared_perm
= cumulative_shared_perms
;
1637 static char *bdrv_child_user_desc(BdrvChild
*c
)
1639 if (c
->role
->get_parent_desc
) {
1640 return c
->role
->get_parent_desc(c
);
1643 return g_strdup("another user");
1646 char *bdrv_perm_names(uint64_t perm
)
1652 { BLK_PERM_CONSISTENT_READ
, "consistent read" },
1653 { BLK_PERM_WRITE
, "write" },
1654 { BLK_PERM_WRITE_UNCHANGED
, "write unchanged" },
1655 { BLK_PERM_RESIZE
, "resize" },
1656 { BLK_PERM_GRAPH_MOD
, "change children" },
1660 char *result
= g_strdup("");
1661 struct perm_name
*p
;
1663 for (p
= permissions
; p
->name
; p
++) {
1664 if (perm
& p
->perm
) {
1666 result
= g_strdup_printf("%s%s%s", old
, *old
? ", " : "", p
->name
);
1675 * Checks whether a new reference to @bs can be added if the new user requires
1676 * @new_used_perm/@new_shared_perm as its permissions. If @ignore_children is
1677 * set, the BdrvChild objects in this list are ignored in the calculations;
1678 * this allows checking permission updates for an existing reference.
1680 * Needs to be followed by a call to either bdrv_set_perm() or
1681 * bdrv_abort_perm_update(). */
1682 static int bdrv_check_update_perm(BlockDriverState
*bs
, uint64_t new_used_perm
,
1683 uint64_t new_shared_perm
,
1684 GSList
*ignore_children
, Error
**errp
)
1687 uint64_t cumulative_perms
= new_used_perm
;
1688 uint64_t cumulative_shared_perms
= new_shared_perm
;
1690 /* There is no reason why anyone couldn't tolerate write_unchanged */
1691 assert(new_shared_perm
& BLK_PERM_WRITE_UNCHANGED
);
1693 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
1694 if (g_slist_find(ignore_children
, c
)) {
1698 if ((new_used_perm
& c
->shared_perm
) != new_used_perm
) {
1699 char *user
= bdrv_child_user_desc(c
);
1700 char *perm_names
= bdrv_perm_names(new_used_perm
& ~c
->shared_perm
);
1701 error_setg(errp
, "Conflicts with use by %s as '%s', which does not "
1703 user
, c
->name
, perm_names
, bdrv_get_node_name(c
->bs
));
1709 if ((c
->perm
& new_shared_perm
) != c
->perm
) {
1710 char *user
= bdrv_child_user_desc(c
);
1711 char *perm_names
= bdrv_perm_names(c
->perm
& ~new_shared_perm
);
1712 error_setg(errp
, "Conflicts with use by %s as '%s', which uses "
1714 user
, c
->name
, perm_names
, bdrv_get_node_name(c
->bs
));
1720 cumulative_perms
|= c
->perm
;
1721 cumulative_shared_perms
&= c
->shared_perm
;
1724 return bdrv_check_perm(bs
, cumulative_perms
, cumulative_shared_perms
,
1725 ignore_children
, errp
);
1728 /* Needs to be followed by a call to either bdrv_child_set_perm() or
1729 * bdrv_child_abort_perm_update(). */
1730 static int bdrv_child_check_perm(BdrvChild
*c
, uint64_t perm
, uint64_t shared
,
1731 GSList
*ignore_children
, Error
**errp
)
1735 ignore_children
= g_slist_prepend(g_slist_copy(ignore_children
), c
);
1736 ret
= bdrv_check_update_perm(c
->bs
, perm
, shared
, ignore_children
, errp
);
1737 g_slist_free(ignore_children
);
1742 static void bdrv_child_set_perm(BdrvChild
*c
, uint64_t perm
, uint64_t shared
)
1744 uint64_t cumulative_perms
, cumulative_shared_perms
;
1747 c
->shared_perm
= shared
;
1749 bdrv_get_cumulative_perm(c
->bs
, &cumulative_perms
,
1750 &cumulative_shared_perms
);
1751 bdrv_set_perm(c
->bs
, cumulative_perms
, cumulative_shared_perms
);
1754 static void bdrv_child_abort_perm_update(BdrvChild
*c
)
1756 bdrv_abort_perm_update(c
->bs
);
1759 int bdrv_child_try_set_perm(BdrvChild
*c
, uint64_t perm
, uint64_t shared
,
1764 ret
= bdrv_child_check_perm(c
, perm
, shared
, NULL
, errp
);
1766 bdrv_child_abort_perm_update(c
);
1770 bdrv_child_set_perm(c
, perm
, shared
);
1775 #define DEFAULT_PERM_PASSTHROUGH (BLK_PERM_CONSISTENT_READ \
1777 | BLK_PERM_WRITE_UNCHANGED \
1779 #define DEFAULT_PERM_UNCHANGED (BLK_PERM_ALL & ~DEFAULT_PERM_PASSTHROUGH)
1781 void bdrv_filter_default_perms(BlockDriverState
*bs
, BdrvChild
*c
,
1782 const BdrvChildRole
*role
,
1783 uint64_t perm
, uint64_t shared
,
1784 uint64_t *nperm
, uint64_t *nshared
)
1787 *nperm
= perm
& DEFAULT_PERM_PASSTHROUGH
;
1788 *nshared
= (shared
& DEFAULT_PERM_PASSTHROUGH
) | DEFAULT_PERM_UNCHANGED
;
1792 *nperm
= (perm
& DEFAULT_PERM_PASSTHROUGH
) |
1793 (c
->perm
& DEFAULT_PERM_UNCHANGED
);
1794 *nshared
= (shared
& DEFAULT_PERM_PASSTHROUGH
) |
1795 (c
->shared_perm
& DEFAULT_PERM_UNCHANGED
);
1798 void bdrv_format_default_perms(BlockDriverState
*bs
, BdrvChild
*c
,
1799 const BdrvChildRole
*role
,
1800 uint64_t perm
, uint64_t shared
,
1801 uint64_t *nperm
, uint64_t *nshared
)
1803 bool backing
= (role
== &child_backing
);
1804 assert(role
== &child_backing
|| role
== &child_file
);
1807 /* Apart from the modifications below, the same permissions are
1808 * forwarded and left alone as for filters */
1809 bdrv_filter_default_perms(bs
, c
, role
, perm
, shared
, &perm
, &shared
);
1811 /* Format drivers may touch metadata even if the guest doesn't write */
1812 if (bdrv_is_writable(bs
)) {
1813 perm
|= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
1816 /* bs->file always needs to be consistent because of the metadata. We
1817 * can never allow other users to resize or write to it. */
1818 perm
|= BLK_PERM_CONSISTENT_READ
;
1819 shared
&= ~(BLK_PERM_WRITE
| BLK_PERM_RESIZE
);
1821 /* We want consistent read from backing files if the parent needs it.
1822 * No other operations are performed on backing files. */
1823 perm
&= BLK_PERM_CONSISTENT_READ
;
1825 /* If the parent can deal with changing data, we're okay with a
1826 * writable and resizable backing file. */
1827 /* TODO Require !(perm & BLK_PERM_CONSISTENT_READ), too? */
1828 if (shared
& BLK_PERM_WRITE
) {
1829 shared
= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
1834 shared
|= BLK_PERM_CONSISTENT_READ
| BLK_PERM_GRAPH_MOD
|
1835 BLK_PERM_WRITE_UNCHANGED
;
1838 if (bs
->open_flags
& BDRV_O_INACTIVE
) {
1839 shared
|= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
1846 static void bdrv_replace_child_noperm(BdrvChild
*child
,
1847 BlockDriverState
*new_bs
)
1849 BlockDriverState
*old_bs
= child
->bs
;
1851 if (old_bs
&& new_bs
) {
1852 assert(bdrv_get_aio_context(old_bs
) == bdrv_get_aio_context(new_bs
));
1855 if (old_bs
->quiesce_counter
&& child
->role
->drained_end
) {
1856 child
->role
->drained_end(child
);
1858 if (child
->role
->detach
) {
1859 child
->role
->detach(child
);
1861 QLIST_REMOVE(child
, next_parent
);
1867 QLIST_INSERT_HEAD(&new_bs
->parents
, child
, next_parent
);
1868 if (new_bs
->quiesce_counter
&& child
->role
->drained_begin
) {
1869 child
->role
->drained_begin(child
);
1872 if (child
->role
->attach
) {
1873 child
->role
->attach(child
);
1879 * Updates @child to change its reference to point to @new_bs, including
1880 * checking and applying the necessary permisson updates both to the old node
1883 * NULL is passed as @new_bs for removing the reference before freeing @child.
1885 * If @new_bs is not NULL, bdrv_check_perm() must be called beforehand, as this
1886 * function uses bdrv_set_perm() to update the permissions according to the new
1887 * reference that @new_bs gets.
1889 static void bdrv_replace_child(BdrvChild
*child
, BlockDriverState
*new_bs
)
1891 BlockDriverState
*old_bs
= child
->bs
;
1892 uint64_t perm
, shared_perm
;
1895 /* Update permissions for old node. This is guaranteed to succeed
1896 * because we're just taking a parent away, so we're loosening
1898 bdrv_get_cumulative_perm(old_bs
, &perm
, &shared_perm
);
1899 bdrv_check_perm(old_bs
, perm
, shared_perm
, NULL
, &error_abort
);
1900 bdrv_set_perm(old_bs
, perm
, shared_perm
);
1903 bdrv_replace_child_noperm(child
, new_bs
);
1906 bdrv_get_cumulative_perm(new_bs
, &perm
, &shared_perm
);
1907 bdrv_set_perm(new_bs
, perm
, shared_perm
);
1911 BdrvChild
*bdrv_root_attach_child(BlockDriverState
*child_bs
,
1912 const char *child_name
,
1913 const BdrvChildRole
*child_role
,
1914 uint64_t perm
, uint64_t shared_perm
,
1915 void *opaque
, Error
**errp
)
1920 ret
= bdrv_check_update_perm(child_bs
, perm
, shared_perm
, NULL
, errp
);
1922 bdrv_abort_perm_update(child_bs
);
1926 child
= g_new(BdrvChild
, 1);
1927 *child
= (BdrvChild
) {
1929 .name
= g_strdup(child_name
),
1932 .shared_perm
= shared_perm
,
1936 /* This performs the matching bdrv_set_perm() for the above check. */
1937 bdrv_replace_child(child
, child_bs
);
1942 BdrvChild
*bdrv_attach_child(BlockDriverState
*parent_bs
,
1943 BlockDriverState
*child_bs
,
1944 const char *child_name
,
1945 const BdrvChildRole
*child_role
,
1949 uint64_t perm
, shared_perm
;
1951 bdrv_get_cumulative_perm(parent_bs
, &perm
, &shared_perm
);
1953 assert(parent_bs
->drv
);
1954 assert(bdrv_get_aio_context(parent_bs
) == bdrv_get_aio_context(child_bs
));
1955 bdrv_child_perm(parent_bs
, child_bs
, NULL
, child_role
,
1956 perm
, shared_perm
, &perm
, &shared_perm
);
1958 child
= bdrv_root_attach_child(child_bs
, child_name
, child_role
,
1959 perm
, shared_perm
, parent_bs
, errp
);
1960 if (child
== NULL
) {
1964 QLIST_INSERT_HEAD(&parent_bs
->children
, child
, next
);
1968 static void bdrv_detach_child(BdrvChild
*child
)
1970 if (child
->next
.le_prev
) {
1971 QLIST_REMOVE(child
, next
);
1972 child
->next
.le_prev
= NULL
;
1975 bdrv_replace_child(child
, NULL
);
1977 g_free(child
->name
);
1981 void bdrv_root_unref_child(BdrvChild
*child
)
1983 BlockDriverState
*child_bs
;
1985 child_bs
= child
->bs
;
1986 bdrv_detach_child(child
);
1987 bdrv_unref(child_bs
);
1990 void bdrv_unref_child(BlockDriverState
*parent
, BdrvChild
*child
)
1992 if (child
== NULL
) {
1996 if (child
->bs
->inherits_from
== parent
) {
1999 /* Remove inherits_from only when the last reference between parent and
2000 * child->bs goes away. */
2001 QLIST_FOREACH(c
, &parent
->children
, next
) {
2002 if (c
!= child
&& c
->bs
== child
->bs
) {
2007 child
->bs
->inherits_from
= NULL
;
2011 bdrv_root_unref_child(child
);
2015 static void bdrv_parent_cb_change_media(BlockDriverState
*bs
, bool load
)
2018 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
2019 if (c
->role
->change_media
) {
2020 c
->role
->change_media(c
, load
);
2025 static void bdrv_parent_cb_resize(BlockDriverState
*bs
)
2028 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
2029 if (c
->role
->resize
) {
2036 * Sets the backing file link of a BDS. A new reference is created; callers
2037 * which don't need their own reference any more must call bdrv_unref().
2039 void bdrv_set_backing_hd(BlockDriverState
*bs
, BlockDriverState
*backing_hd
,
2043 bdrv_ref(backing_hd
);
2047 bdrv_unref_child(bs
, bs
->backing
);
2055 bs
->backing
= bdrv_attach_child(bs
, backing_hd
, "backing", &child_backing
,
2058 bdrv_unref(backing_hd
);
2061 bdrv_refresh_filename(bs
);
2064 bdrv_refresh_limits(bs
, NULL
);
2068 * Opens the backing file for a BlockDriverState if not yet open
2070 * bdref_key specifies the key for the image's BlockdevRef in the options QDict.
2071 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
2072 * itself, all options starting with "${bdref_key}." are considered part of the
2075 * TODO Can this be unified with bdrv_open_image()?
2077 int bdrv_open_backing_file(BlockDriverState
*bs
, QDict
*parent_options
,
2078 const char *bdref_key
, Error
**errp
)
2080 char *backing_filename
= g_malloc0(PATH_MAX
);
2081 char *bdref_key_dot
;
2082 const char *reference
= NULL
;
2084 BlockDriverState
*backing_hd
;
2086 QDict
*tmp_parent_options
= NULL
;
2087 Error
*local_err
= NULL
;
2089 if (bs
->backing
!= NULL
) {
2093 /* NULL means an empty set of options */
2094 if (parent_options
== NULL
) {
2095 tmp_parent_options
= qdict_new();
2096 parent_options
= tmp_parent_options
;
2099 bs
->open_flags
&= ~BDRV_O_NO_BACKING
;
2101 bdref_key_dot
= g_strdup_printf("%s.", bdref_key
);
2102 qdict_extract_subqdict(parent_options
, &options
, bdref_key_dot
);
2103 g_free(bdref_key_dot
);
2106 * Caution: while qdict_get_try_str() is fine, getting non-string
2107 * types would require more care. When @parent_options come from
2108 * -blockdev or blockdev_add, its members are typed according to
2109 * the QAPI schema, but when they come from -drive, they're all
2112 reference
= qdict_get_try_str(parent_options
, bdref_key
);
2113 if (reference
|| qdict_haskey(options
, "file.filename")) {
2114 backing_filename
[0] = '\0';
2115 } else if (bs
->backing_file
[0] == '\0' && qdict_size(options
) == 0) {
2119 bdrv_get_full_backing_filename(bs
, backing_filename
, PATH_MAX
,
2123 error_propagate(errp
, local_err
);
2129 if (!bs
->drv
|| !bs
->drv
->supports_backing
) {
2131 error_setg(errp
, "Driver doesn't support backing files");
2136 if (bs
->backing_format
[0] != '\0' && !qdict_haskey(options
, "driver")) {
2137 qdict_put_str(options
, "driver", bs
->backing_format
);
2140 backing_hd
= bdrv_open_inherit(*backing_filename
? backing_filename
: NULL
,
2141 reference
, options
, 0, bs
, &child_backing
,
2144 bs
->open_flags
|= BDRV_O_NO_BACKING
;
2145 error_prepend(errp
, "Could not open backing file: ");
2150 /* Hook up the backing file link; drop our reference, bs owns the
2151 * backing_hd reference now */
2152 bdrv_set_backing_hd(bs
, backing_hd
, &local_err
);
2153 bdrv_unref(backing_hd
);
2155 error_propagate(errp
, local_err
);
2160 qdict_del(parent_options
, bdref_key
);
2163 g_free(backing_filename
);
2164 QDECREF(tmp_parent_options
);
2168 static BlockDriverState
*
2169 bdrv_open_child_bs(const char *filename
, QDict
*options
, const char *bdref_key
,
2170 BlockDriverState
*parent
, const BdrvChildRole
*child_role
,
2171 bool allow_none
, Error
**errp
)
2173 BlockDriverState
*bs
= NULL
;
2174 QDict
*image_options
;
2175 char *bdref_key_dot
;
2176 const char *reference
;
2178 assert(child_role
!= NULL
);
2180 bdref_key_dot
= g_strdup_printf("%s.", bdref_key
);
2181 qdict_extract_subqdict(options
, &image_options
, bdref_key_dot
);
2182 g_free(bdref_key_dot
);
2185 * Caution: while qdict_get_try_str() is fine, getting non-string
2186 * types would require more care. When @options come from
2187 * -blockdev or blockdev_add, its members are typed according to
2188 * the QAPI schema, but when they come from -drive, they're all
2191 reference
= qdict_get_try_str(options
, bdref_key
);
2192 if (!filename
&& !reference
&& !qdict_size(image_options
)) {
2194 error_setg(errp
, "A block device must be specified for \"%s\"",
2197 QDECREF(image_options
);
2201 bs
= bdrv_open_inherit(filename
, reference
, image_options
, 0,
2202 parent
, child_role
, errp
);
2208 qdict_del(options
, bdref_key
);
2213 * Opens a disk image whose options are given as BlockdevRef in another block
2216 * If allow_none is true, no image will be opened if filename is false and no
2217 * BlockdevRef is given. NULL will be returned, but errp remains unset.
2219 * bdrev_key specifies the key for the image's BlockdevRef in the options QDict.
2220 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
2221 * itself, all options starting with "${bdref_key}." are considered part of the
2224 * The BlockdevRef will be removed from the options QDict.
2226 BdrvChild
*bdrv_open_child(const char *filename
,
2227 QDict
*options
, const char *bdref_key
,
2228 BlockDriverState
*parent
,
2229 const BdrvChildRole
*child_role
,
2230 bool allow_none
, Error
**errp
)
2233 BlockDriverState
*bs
;
2235 bs
= bdrv_open_child_bs(filename
, options
, bdref_key
, parent
, child_role
,
2241 c
= bdrv_attach_child(parent
, bs
, bdref_key
, child_role
, errp
);
2250 static BlockDriverState
*bdrv_append_temp_snapshot(BlockDriverState
*bs
,
2252 QDict
*snapshot_options
,
2255 /* TODO: extra byte is a hack to ensure MAX_PATH space on Windows. */
2256 char *tmp_filename
= g_malloc0(PATH_MAX
+ 1);
2258 QemuOpts
*opts
= NULL
;
2259 BlockDriverState
*bs_snapshot
= NULL
;
2260 Error
*local_err
= NULL
;
2263 /* if snapshot, we create a temporary backing file and open it
2264 instead of opening 'filename' directly */
2266 /* Get the required size from the image */
2267 total_size
= bdrv_getlength(bs
);
2268 if (total_size
< 0) {
2269 error_setg_errno(errp
, -total_size
, "Could not get image size");
2273 /* Create the temporary image */
2274 ret
= get_tmp_filename(tmp_filename
, PATH_MAX
+ 1);
2276 error_setg_errno(errp
, -ret
, "Could not get temporary filename");
2280 opts
= qemu_opts_create(bdrv_qcow2
.create_opts
, NULL
, 0,
2282 qemu_opt_set_number(opts
, BLOCK_OPT_SIZE
, total_size
, &error_abort
);
2283 ret
= bdrv_create(&bdrv_qcow2
, tmp_filename
, opts
, errp
);
2284 qemu_opts_del(opts
);
2286 error_prepend(errp
, "Could not create temporary overlay '%s': ",
2291 /* Prepare options QDict for the temporary file */
2292 qdict_put_str(snapshot_options
, "file.driver", "file");
2293 qdict_put_str(snapshot_options
, "file.filename", tmp_filename
);
2294 qdict_put_str(snapshot_options
, "driver", "qcow2");
2296 bs_snapshot
= bdrv_open(NULL
, NULL
, snapshot_options
, flags
, errp
);
2297 snapshot_options
= NULL
;
2302 /* bdrv_append() consumes a strong reference to bs_snapshot
2303 * (i.e. it will call bdrv_unref() on it) even on error, so in
2304 * order to be able to return one, we have to increase
2305 * bs_snapshot's refcount here */
2306 bdrv_ref(bs_snapshot
);
2307 bdrv_append(bs_snapshot
, bs
, &local_err
);
2309 error_propagate(errp
, local_err
);
2315 QDECREF(snapshot_options
);
2316 g_free(tmp_filename
);
2321 * Opens a disk image (raw, qcow2, vmdk, ...)
2323 * options is a QDict of options to pass to the block drivers, or NULL for an
2324 * empty set of options. The reference to the QDict belongs to the block layer
2325 * after the call (even on failure), so if the caller intends to reuse the
2326 * dictionary, it needs to use QINCREF() before calling bdrv_open.
2328 * If *pbs is NULL, a new BDS will be created with a pointer to it stored there.
2329 * If it is not NULL, the referenced BDS will be reused.
2331 * The reference parameter may be used to specify an existing block device which
2332 * should be opened. If specified, neither options nor a filename may be given,
2333 * nor can an existing BDS be reused (that is, *pbs has to be NULL).
2335 static BlockDriverState
*bdrv_open_inherit(const char *filename
,
2336 const char *reference
,
2337 QDict
*options
, int flags
,
2338 BlockDriverState
*parent
,
2339 const BdrvChildRole
*child_role
,
2343 BlockBackend
*file
= NULL
;
2344 BlockDriverState
*bs
;
2345 BlockDriver
*drv
= NULL
;
2346 const char *drvname
;
2347 const char *backing
;
2348 Error
*local_err
= NULL
;
2349 QDict
*snapshot_options
= NULL
;
2350 int snapshot_flags
= 0;
2352 assert(!child_role
|| !flags
);
2353 assert(!child_role
== !parent
);
2356 bool options_non_empty
= options
? qdict_size(options
) : false;
2359 if (filename
|| options_non_empty
) {
2360 error_setg(errp
, "Cannot reference an existing block device with "
2361 "additional options or a new filename");
2365 bs
= bdrv_lookup_bs(reference
, reference
, errp
);
2376 /* NULL means an empty set of options */
2377 if (options
== NULL
) {
2378 options
= qdict_new();
2381 /* json: syntax counts as explicit options, as if in the QDict */
2382 parse_json_protocol(options
, &filename
, &local_err
);
2387 bs
->explicit_options
= qdict_clone_shallow(options
);
2390 bs
->inherits_from
= parent
;
2391 child_role
->inherit_options(&flags
, options
,
2392 parent
->open_flags
, parent
->options
);
2395 ret
= bdrv_fill_options(&options
, filename
, &flags
, &local_err
);
2401 * Set the BDRV_O_RDWR and BDRV_O_ALLOW_RDWR flags.
2402 * Caution: getting a boolean member of @options requires care.
2403 * When @options come from -blockdev or blockdev_add, members are
2404 * typed according to the QAPI schema, but when they come from
2405 * -drive, they're all QString.
2407 if (g_strcmp0(qdict_get_try_str(options
, BDRV_OPT_READ_ONLY
), "on") &&
2408 !qdict_get_try_bool(options
, BDRV_OPT_READ_ONLY
, false)) {
2409 flags
|= (BDRV_O_RDWR
| BDRV_O_ALLOW_RDWR
);
2411 flags
&= ~BDRV_O_RDWR
;
2414 if (flags
& BDRV_O_SNAPSHOT
) {
2415 snapshot_options
= qdict_new();
2416 bdrv_temp_snapshot_options(&snapshot_flags
, snapshot_options
,
2418 /* Let bdrv_backing_options() override "read-only" */
2419 qdict_del(options
, BDRV_OPT_READ_ONLY
);
2420 bdrv_backing_options(&flags
, options
, flags
, options
);
2423 bs
->open_flags
= flags
;
2424 bs
->options
= options
;
2425 options
= qdict_clone_shallow(options
);
2427 /* Find the right image format driver */
2428 /* See cautionary note on accessing @options above */
2429 drvname
= qdict_get_try_str(options
, "driver");
2431 drv
= bdrv_find_format(drvname
);
2433 error_setg(errp
, "Unknown driver: '%s'", drvname
);
2438 assert(drvname
|| !(flags
& BDRV_O_PROTOCOL
));
2440 /* See cautionary note on accessing @options above */
2441 backing
= qdict_get_try_str(options
, "backing");
2442 if (backing
&& *backing
== '\0') {
2443 flags
|= BDRV_O_NO_BACKING
;
2444 qdict_del(options
, "backing");
2447 /* Open image file without format layer. This BlockBackend is only used for
2448 * probing, the block drivers will do their own bdrv_open_child() for the
2449 * same BDS, which is why we put the node name back into options. */
2450 if ((flags
& BDRV_O_PROTOCOL
) == 0) {
2451 BlockDriverState
*file_bs
;
2453 file_bs
= bdrv_open_child_bs(filename
, options
, "file", bs
,
2454 &child_file
, true, &local_err
);
2458 if (file_bs
!= NULL
) {
2459 file
= blk_new(BLK_PERM_CONSISTENT_READ
, BLK_PERM_ALL
);
2460 blk_insert_bs(file
, file_bs
, &local_err
);
2461 bdrv_unref(file_bs
);
2466 qdict_put_str(options
, "file", bdrv_get_node_name(file_bs
));
2470 /* Image format probing */
2473 ret
= find_image_format(file
, filename
, &drv
, &local_err
);
2478 * This option update would logically belong in bdrv_fill_options(),
2479 * but we first need to open bs->file for the probing to work, while
2480 * opening bs->file already requires the (mostly) final set of options
2481 * so that cache mode etc. can be inherited.
2483 * Adding the driver later is somewhat ugly, but it's not an option
2484 * that would ever be inherited, so it's correct. We just need to make
2485 * sure to update both bs->options (which has the full effective
2486 * options for bs) and options (which has file.* already removed).
2488 qdict_put_str(bs
->options
, "driver", drv
->format_name
);
2489 qdict_put_str(options
, "driver", drv
->format_name
);
2491 error_setg(errp
, "Must specify either driver or file");
2495 /* BDRV_O_PROTOCOL must be set iff a protocol BDS is about to be created */
2496 assert(!!(flags
& BDRV_O_PROTOCOL
) == !!drv
->bdrv_file_open
);
2497 /* file must be NULL if a protocol BDS is about to be created
2498 * (the inverse results in an error message from bdrv_open_common()) */
2499 assert(!(flags
& BDRV_O_PROTOCOL
) || !file
);
2501 /* Open the image */
2502 ret
= bdrv_open_common(bs
, file
, options
, &local_err
);
2512 /* If there is a backing file, use it */
2513 if ((flags
& BDRV_O_NO_BACKING
) == 0) {
2514 ret
= bdrv_open_backing_file(bs
, options
, "backing", &local_err
);
2516 goto close_and_fail
;
2520 bdrv_refresh_filename(bs
);
2522 /* Check if any unknown options were used */
2523 if (qdict_size(options
) != 0) {
2524 const QDictEntry
*entry
= qdict_first(options
);
2525 if (flags
& BDRV_O_PROTOCOL
) {
2526 error_setg(errp
, "Block protocol '%s' doesn't support the option "
2527 "'%s'", drv
->format_name
, entry
->key
);
2530 "Block format '%s' does not support the option '%s'",
2531 drv
->format_name
, entry
->key
);
2534 goto close_and_fail
;
2537 if (!bdrv_key_required(bs
)) {
2538 bdrv_parent_cb_change_media(bs
, true);
2539 } else if (!runstate_check(RUN_STATE_PRELAUNCH
)
2540 && !runstate_check(RUN_STATE_INMIGRATE
)
2541 && !runstate_check(RUN_STATE_PAUSED
)) { /* HACK */
2543 "Guest must be stopped for opening of encrypted image");
2544 goto close_and_fail
;
2549 /* For snapshot=on, create a temporary qcow2 overlay. bs points to the
2550 * temporary snapshot afterwards. */
2551 if (snapshot_flags
) {
2552 BlockDriverState
*snapshot_bs
;
2553 snapshot_bs
= bdrv_append_temp_snapshot(bs
, snapshot_flags
,
2554 snapshot_options
, &local_err
);
2555 snapshot_options
= NULL
;
2557 goto close_and_fail
;
2559 /* We are not going to return bs but the overlay on top of it
2560 * (snapshot_bs); thus, we have to drop the strong reference to bs
2561 * (which we obtained by calling bdrv_new()). bs will not be deleted,
2562 * though, because the overlay still has a reference to it. */
2571 if (bs
->file
!= NULL
) {
2572 bdrv_unref_child(bs
, bs
->file
);
2574 QDECREF(snapshot_options
);
2575 QDECREF(bs
->explicit_options
);
2576 QDECREF(bs
->options
);
2580 error_propagate(errp
, local_err
);
2585 QDECREF(snapshot_options
);
2587 error_propagate(errp
, local_err
);
2591 BlockDriverState
*bdrv_open(const char *filename
, const char *reference
,
2592 QDict
*options
, int flags
, Error
**errp
)
2594 return bdrv_open_inherit(filename
, reference
, options
, flags
, NULL
,
2598 typedef struct BlockReopenQueueEntry
{
2600 BDRVReopenState state
;
2601 QSIMPLEQ_ENTRY(BlockReopenQueueEntry
) entry
;
2602 } BlockReopenQueueEntry
;
2605 * Adds a BlockDriverState to a simple queue for an atomic, transactional
2606 * reopen of multiple devices.
2608 * bs_queue can either be an existing BlockReopenQueue that has had QSIMPLE_INIT
2609 * already performed, or alternatively may be NULL a new BlockReopenQueue will
2610 * be created and initialized. This newly created BlockReopenQueue should be
2611 * passed back in for subsequent calls that are intended to be of the same
2614 * bs is the BlockDriverState to add to the reopen queue.
2616 * options contains the changed options for the associated bs
2617 * (the BlockReopenQueue takes ownership)
2619 * flags contains the open flags for the associated bs
2621 * returns a pointer to bs_queue, which is either the newly allocated
2622 * bs_queue, or the existing bs_queue being used.
2625 static BlockReopenQueue
*bdrv_reopen_queue_child(BlockReopenQueue
*bs_queue
,
2626 BlockDriverState
*bs
,
2629 const BdrvChildRole
*role
,
2630 QDict
*parent_options
,
2635 BlockReopenQueueEntry
*bs_entry
;
2637 QDict
*old_options
, *explicit_options
;
2639 if (bs_queue
== NULL
) {
2640 bs_queue
= g_new0(BlockReopenQueue
, 1);
2641 QSIMPLEQ_INIT(bs_queue
);
2645 options
= qdict_new();
2648 /* Check if this BlockDriverState is already in the queue */
2649 QSIMPLEQ_FOREACH(bs_entry
, bs_queue
, entry
) {
2650 if (bs
== bs_entry
->state
.bs
) {
2656 * Precedence of options:
2657 * 1. Explicitly passed in options (highest)
2658 * 2. Set in flags (only for top level)
2659 * 3. Retained from explicitly set options of bs
2660 * 4. Inherited from parent node
2661 * 5. Retained from effective options of bs
2664 if (!parent_options
) {
2666 * Any setting represented by flags is always updated. If the
2667 * corresponding QDict option is set, it takes precedence. Otherwise
2668 * the flag is translated into a QDict option. The old setting of bs is
2671 update_options_from_flags(options
, flags
);
2674 /* Old explicitly set values (don't overwrite by inherited value) */
2676 old_options
= qdict_clone_shallow(bs_entry
->state
.explicit_options
);
2678 old_options
= qdict_clone_shallow(bs
->explicit_options
);
2680 bdrv_join_options(bs
, options
, old_options
);
2681 QDECREF(old_options
);
2683 explicit_options
= qdict_clone_shallow(options
);
2685 /* Inherit from parent node */
2686 if (parent_options
) {
2688 role
->inherit_options(&flags
, options
, parent_flags
, parent_options
);
2691 /* Old values are used for options that aren't set yet */
2692 old_options
= qdict_clone_shallow(bs
->options
);
2693 bdrv_join_options(bs
, options
, old_options
);
2694 QDECREF(old_options
);
2696 /* bdrv_open() masks this flag out */
2697 flags
&= ~BDRV_O_PROTOCOL
;
2699 QLIST_FOREACH(child
, &bs
->children
, next
) {
2700 QDict
*new_child_options
;
2701 char *child_key_dot
;
2703 /* reopen can only change the options of block devices that were
2704 * implicitly created and inherited options. For other (referenced)
2705 * block devices, a syntax like "backing.foo" results in an error. */
2706 if (child
->bs
->inherits_from
!= bs
) {
2710 child_key_dot
= g_strdup_printf("%s.", child
->name
);
2711 qdict_extract_subqdict(options
, &new_child_options
, child_key_dot
);
2712 g_free(child_key_dot
);
2714 bdrv_reopen_queue_child(bs_queue
, child
->bs
, new_child_options
, 0,
2715 child
->role
, options
, flags
);
2719 bs_entry
= g_new0(BlockReopenQueueEntry
, 1);
2720 QSIMPLEQ_INSERT_TAIL(bs_queue
, bs_entry
, entry
);
2722 QDECREF(bs_entry
->state
.options
);
2723 QDECREF(bs_entry
->state
.explicit_options
);
2726 bs_entry
->state
.bs
= bs
;
2727 bs_entry
->state
.options
= options
;
2728 bs_entry
->state
.explicit_options
= explicit_options
;
2729 bs_entry
->state
.flags
= flags
;
2734 BlockReopenQueue
*bdrv_reopen_queue(BlockReopenQueue
*bs_queue
,
2735 BlockDriverState
*bs
,
2736 QDict
*options
, int flags
)
2738 return bdrv_reopen_queue_child(bs_queue
, bs
, options
, flags
,
2743 * Reopen multiple BlockDriverStates atomically & transactionally.
2745 * The queue passed in (bs_queue) must have been built up previous
2746 * via bdrv_reopen_queue().
2748 * Reopens all BDS specified in the queue, with the appropriate
2749 * flags. All devices are prepared for reopen, and failure of any
2750 * device will cause all device changes to be abandonded, and intermediate
2753 * If all devices prepare successfully, then the changes are committed
2757 int bdrv_reopen_multiple(AioContext
*ctx
, BlockReopenQueue
*bs_queue
, Error
**errp
)
2760 BlockReopenQueueEntry
*bs_entry
, *next
;
2761 Error
*local_err
= NULL
;
2763 assert(bs_queue
!= NULL
);
2765 aio_context_release(ctx
);
2766 bdrv_drain_all_begin();
2767 aio_context_acquire(ctx
);
2769 QSIMPLEQ_FOREACH(bs_entry
, bs_queue
, entry
) {
2770 if (bdrv_reopen_prepare(&bs_entry
->state
, bs_queue
, &local_err
)) {
2771 error_propagate(errp
, local_err
);
2774 bs_entry
->prepared
= true;
2777 /* If we reach this point, we have success and just need to apply the
2780 QSIMPLEQ_FOREACH(bs_entry
, bs_queue
, entry
) {
2781 bdrv_reopen_commit(&bs_entry
->state
);
2787 QSIMPLEQ_FOREACH_SAFE(bs_entry
, bs_queue
, entry
, next
) {
2788 if (ret
&& bs_entry
->prepared
) {
2789 bdrv_reopen_abort(&bs_entry
->state
);
2791 QDECREF(bs_entry
->state
.explicit_options
);
2793 QDECREF(bs_entry
->state
.options
);
2798 bdrv_drain_all_end();
2804 /* Reopen a single BlockDriverState with the specified flags. */
2805 int bdrv_reopen(BlockDriverState
*bs
, int bdrv_flags
, Error
**errp
)
2808 Error
*local_err
= NULL
;
2809 BlockReopenQueue
*queue
= bdrv_reopen_queue(NULL
, bs
, NULL
, bdrv_flags
);
2811 ret
= bdrv_reopen_multiple(bdrv_get_aio_context(bs
), queue
, &local_err
);
2812 if (local_err
!= NULL
) {
2813 error_propagate(errp
, local_err
);
2820 * Prepares a BlockDriverState for reopen. All changes are staged in the
2821 * 'opaque' field of the BDRVReopenState, which is used and allocated by
2822 * the block driver layer .bdrv_reopen_prepare()
2824 * bs is the BlockDriverState to reopen
2825 * flags are the new open flags
2826 * queue is the reopen queue
2828 * Returns 0 on success, non-zero on error. On error errp will be set
2831 * On failure, bdrv_reopen_abort() will be called to clean up any data.
2832 * It is the responsibility of the caller to then call the abort() or
2833 * commit() for any other BDS that have been left in a prepare() state
2836 int bdrv_reopen_prepare(BDRVReopenState
*reopen_state
, BlockReopenQueue
*queue
,
2840 Error
*local_err
= NULL
;
2846 assert(reopen_state
!= NULL
);
2847 assert(reopen_state
->bs
->drv
!= NULL
);
2848 drv
= reopen_state
->bs
->drv
;
2850 /* Process generic block layer options */
2851 opts
= qemu_opts_create(&bdrv_runtime_opts
, NULL
, 0, &error_abort
);
2852 qemu_opts_absorb_qdict(opts
, reopen_state
->options
, &local_err
);
2854 error_propagate(errp
, local_err
);
2859 update_flags_from_options(&reopen_state
->flags
, opts
);
2861 /* node-name and driver must be unchanged. Put them back into the QDict, so
2862 * that they are checked at the end of this function. */
2863 value
= qemu_opt_get(opts
, "node-name");
2865 qdict_put_str(reopen_state
->options
, "node-name", value
);
2868 value
= qemu_opt_get(opts
, "driver");
2870 qdict_put_str(reopen_state
->options
, "driver", value
);
2873 /* If we are to stay read-only, do not allow permission change
2874 * to r/w. Attempting to set to r/w may fail if either BDRV_O_ALLOW_RDWR is
2875 * not set, or if the BDS still has copy_on_read enabled */
2876 read_only
= !(reopen_state
->flags
& BDRV_O_RDWR
);
2877 ret
= bdrv_can_set_read_only(reopen_state
->bs
, read_only
, &local_err
);
2879 error_propagate(errp
, local_err
);
2884 ret
= bdrv_flush(reopen_state
->bs
);
2886 error_setg_errno(errp
, -ret
, "Error flushing drive");
2890 if (drv
->bdrv_reopen_prepare
) {
2891 ret
= drv
->bdrv_reopen_prepare(reopen_state
, queue
, &local_err
);
2893 if (local_err
!= NULL
) {
2894 error_propagate(errp
, local_err
);
2896 error_setg(errp
, "failed while preparing to reopen image '%s'",
2897 reopen_state
->bs
->filename
);
2902 /* It is currently mandatory to have a bdrv_reopen_prepare()
2903 * handler for each supported drv. */
2904 error_setg(errp
, "Block format '%s' used by node '%s' "
2905 "does not support reopening files", drv
->format_name
,
2906 bdrv_get_device_or_node_name(reopen_state
->bs
));
2911 /* Options that are not handled are only okay if they are unchanged
2912 * compared to the old state. It is expected that some options are only
2913 * used for the initial open, but not reopen (e.g. filename) */
2914 if (qdict_size(reopen_state
->options
)) {
2915 const QDictEntry
*entry
= qdict_first(reopen_state
->options
);
2918 QString
*new_obj
= qobject_to_qstring(entry
->value
);
2919 const char *new = qstring_get_str(new_obj
);
2921 * Caution: while qdict_get_try_str() is fine, getting
2922 * non-string types would require more care. When
2923 * bs->options come from -blockdev or blockdev_add, its
2924 * members are typed according to the QAPI schema, but
2925 * when they come from -drive, they're all QString.
2927 const char *old
= qdict_get_try_str(reopen_state
->bs
->options
,
2930 if (!old
|| strcmp(new, old
)) {
2931 error_setg(errp
, "Cannot change the option '%s'", entry
->key
);
2935 } while ((entry
= qdict_next(reopen_state
->options
, entry
)));
2941 qemu_opts_del(opts
);
2946 * Takes the staged changes for the reopen from bdrv_reopen_prepare(), and
2947 * makes them final by swapping the staging BlockDriverState contents into
2948 * the active BlockDriverState contents.
2950 void bdrv_reopen_commit(BDRVReopenState
*reopen_state
)
2954 assert(reopen_state
!= NULL
);
2955 drv
= reopen_state
->bs
->drv
;
2956 assert(drv
!= NULL
);
2958 /* If there are any driver level actions to take */
2959 if (drv
->bdrv_reopen_commit
) {
2960 drv
->bdrv_reopen_commit(reopen_state
);
2963 /* set BDS specific flags now */
2964 QDECREF(reopen_state
->bs
->explicit_options
);
2966 reopen_state
->bs
->explicit_options
= reopen_state
->explicit_options
;
2967 reopen_state
->bs
->open_flags
= reopen_state
->flags
;
2968 reopen_state
->bs
->read_only
= !(reopen_state
->flags
& BDRV_O_RDWR
);
2970 bdrv_refresh_limits(reopen_state
->bs
, NULL
);
2974 * Abort the reopen, and delete and free the staged changes in
2977 void bdrv_reopen_abort(BDRVReopenState
*reopen_state
)
2981 assert(reopen_state
!= NULL
);
2982 drv
= reopen_state
->bs
->drv
;
2983 assert(drv
!= NULL
);
2985 if (drv
->bdrv_reopen_abort
) {
2986 drv
->bdrv_reopen_abort(reopen_state
);
2989 QDECREF(reopen_state
->explicit_options
);
2993 static void bdrv_close(BlockDriverState
*bs
)
2995 BdrvAioNotifier
*ban
, *ban_next
;
2998 assert(!bs
->refcnt
);
3000 bdrv_drained_begin(bs
); /* complete I/O */
3002 bdrv_drain(bs
); /* in case flush left pending I/O */
3004 bdrv_release_named_dirty_bitmaps(bs
);
3005 assert(QLIST_EMPTY(&bs
->dirty_bitmaps
));
3008 BdrvChild
*child
, *next
;
3010 bs
->drv
->bdrv_close(bs
);
3013 bdrv_set_backing_hd(bs
, NULL
, &error_abort
);
3015 if (bs
->file
!= NULL
) {
3016 bdrv_unref_child(bs
, bs
->file
);
3020 QLIST_FOREACH_SAFE(child
, &bs
->children
, next
, next
) {
3021 /* TODO Remove bdrv_unref() from drivers' close function and use
3022 * bdrv_unref_child() here */
3023 if (child
->bs
->inherits_from
== bs
) {
3024 child
->bs
->inherits_from
= NULL
;
3026 bdrv_detach_child(child
);
3031 bs
->copy_on_read
= 0;
3032 bs
->backing_file
[0] = '\0';
3033 bs
->backing_format
[0] = '\0';
3034 bs
->total_sectors
= 0;
3035 bs
->encrypted
= false;
3036 bs
->valid_key
= false;
3038 QDECREF(bs
->options
);
3039 QDECREF(bs
->explicit_options
);
3041 QDECREF(bs
->full_open_options
);
3042 bs
->full_open_options
= NULL
;
3045 QLIST_FOREACH_SAFE(ban
, &bs
->aio_notifiers
, list
, ban_next
) {
3048 QLIST_INIT(&bs
->aio_notifiers
);
3049 bdrv_drained_end(bs
);
3052 void bdrv_close_all(void)
3054 block_job_cancel_sync_all();
3055 nbd_export_close_all();
3057 /* Drop references from requests still in flight, such as canceled block
3058 * jobs whose AIO context has not been polled yet */
3061 blk_remove_all_bs();
3062 blockdev_close_all_bdrv_states();
3064 assert(QTAILQ_EMPTY(&all_bdrv_states
));
3067 static bool should_update_child(BdrvChild
*c
, BlockDriverState
*to
)
3071 if (c
->role
->stay_at_node
) {
3075 if (c
->role
== &child_backing
) {
3076 /* If @from is a backing file of @to, ignore the child to avoid
3077 * creating a loop. We only want to change the pointer of other
3079 QLIST_FOREACH(to_c
, &to
->children
, next
) {
3092 void bdrv_replace_node(BlockDriverState
*from
, BlockDriverState
*to
,
3095 BdrvChild
*c
, *next
;
3096 GSList
*list
= NULL
, *p
;
3097 uint64_t old_perm
, old_shared
;
3098 uint64_t perm
= 0, shared
= BLK_PERM_ALL
;
3101 assert(!atomic_read(&from
->in_flight
));
3102 assert(!atomic_read(&to
->in_flight
));
3104 /* Make sure that @from doesn't go away until we have successfully attached
3105 * all of its parents to @to. */
3108 /* Put all parents into @list and calculate their cumulative permissions */
3109 QLIST_FOREACH_SAFE(c
, &from
->parents
, next_parent
, next
) {
3110 if (!should_update_child(c
, to
)) {
3113 list
= g_slist_prepend(list
, c
);
3115 shared
&= c
->shared_perm
;
3118 /* Check whether the required permissions can be granted on @to, ignoring
3119 * all BdrvChild in @list so that they can't block themselves. */
3120 ret
= bdrv_check_update_perm(to
, perm
, shared
, list
, errp
);
3122 bdrv_abort_perm_update(to
);
3126 /* Now actually perform the change. We performed the permission check for
3127 * all elements of @list at once, so set the permissions all at once at the
3129 for (p
= list
; p
!= NULL
; p
= p
->next
) {
3133 bdrv_replace_child_noperm(c
, to
);
3137 bdrv_get_cumulative_perm(to
, &old_perm
, &old_shared
);
3138 bdrv_set_perm(to
, old_perm
| perm
, old_shared
| shared
);
3146 * Add new bs contents at the top of an image chain while the chain is
3147 * live, while keeping required fields on the top layer.
3149 * This will modify the BlockDriverState fields, and swap contents
3150 * between bs_new and bs_top. Both bs_new and bs_top are modified.
3152 * bs_new must not be attached to a BlockBackend.
3154 * This function does not create any image files.
3156 * bdrv_append() takes ownership of a bs_new reference and unrefs it because
3157 * that's what the callers commonly need. bs_new will be referenced by the old
3158 * parents of bs_top after bdrv_append() returns. If the caller needs to keep a
3159 * reference of its own, it must call bdrv_ref().
3161 void bdrv_append(BlockDriverState
*bs_new
, BlockDriverState
*bs_top
,
3164 Error
*local_err
= NULL
;
3166 bdrv_set_backing_hd(bs_new
, bs_top
, &local_err
);
3168 error_propagate(errp
, local_err
);
3172 bdrv_replace_node(bs_top
, bs_new
, &local_err
);
3174 error_propagate(errp
, local_err
);
3175 bdrv_set_backing_hd(bs_new
, NULL
, &error_abort
);
3179 /* bs_new is now referenced by its new parents, we don't need the
3180 * additional reference any more. */
3185 static void bdrv_delete(BlockDriverState
*bs
)
3188 assert(bdrv_op_blocker_is_empty(bs
));
3189 assert(!bs
->refcnt
);
3193 /* remove from list, if necessary */
3194 if (bs
->node_name
[0] != '\0') {
3195 QTAILQ_REMOVE(&graph_bdrv_states
, bs
, node_list
);
3197 QTAILQ_REMOVE(&all_bdrv_states
, bs
, bs_list
);
3203 * Run consistency checks on an image
3205 * Returns 0 if the check could be completed (it doesn't mean that the image is
3206 * free of errors) or -errno when an internal error occurred. The results of the
3207 * check are stored in res.
3209 int bdrv_check(BlockDriverState
*bs
, BdrvCheckResult
*res
, BdrvCheckMode fix
)
3211 if (bs
->drv
== NULL
) {
3214 if (bs
->drv
->bdrv_check
== NULL
) {
3218 memset(res
, 0, sizeof(*res
));
3219 return bs
->drv
->bdrv_check(bs
, res
, fix
);
3225 * -EINVAL - backing format specified, but no file
3226 * -ENOSPC - can't update the backing file because no space is left in the
3228 * -ENOTSUP - format driver doesn't support changing the backing file
3230 int bdrv_change_backing_file(BlockDriverState
*bs
,
3231 const char *backing_file
, const char *backing_fmt
)
3233 BlockDriver
*drv
= bs
->drv
;
3236 /* Backing file format doesn't make sense without a backing file */
3237 if (backing_fmt
&& !backing_file
) {
3241 if (drv
->bdrv_change_backing_file
!= NULL
) {
3242 ret
= drv
->bdrv_change_backing_file(bs
, backing_file
, backing_fmt
);
3248 pstrcpy(bs
->backing_file
, sizeof(bs
->backing_file
), backing_file
?: "");
3249 pstrcpy(bs
->backing_format
, sizeof(bs
->backing_format
), backing_fmt
?: "");
3255 * Finds the image layer in the chain that has 'bs' as its backing file.
3257 * active is the current topmost image.
3259 * Returns NULL if bs is not found in active's image chain,
3260 * or if active == bs.
3262 * Returns the bottommost base image if bs == NULL.
3264 BlockDriverState
*bdrv_find_overlay(BlockDriverState
*active
,
3265 BlockDriverState
*bs
)
3267 while (active
&& bs
!= backing_bs(active
)) {
3268 active
= backing_bs(active
);
3274 /* Given a BDS, searches for the base layer. */
3275 BlockDriverState
*bdrv_find_base(BlockDriverState
*bs
)
3277 return bdrv_find_overlay(bs
, NULL
);
3281 * Drops images above 'base' up to and including 'top', and sets the image
3282 * above 'top' to have base as its backing file.
3284 * Requires that the overlay to 'top' is opened r/w, so that the backing file
3285 * information in 'bs' can be properly updated.
3287 * E.g., this will convert the following chain:
3288 * bottom <- base <- intermediate <- top <- active
3292 * bottom <- base <- active
3294 * It is allowed for bottom==base, in which case it converts:
3296 * base <- intermediate <- top <- active
3302 * If backing_file_str is non-NULL, it will be used when modifying top's
3303 * overlay image metadata.
3306 * if active == top, that is considered an error
3309 int bdrv_drop_intermediate(BlockDriverState
*active
, BlockDriverState
*top
,
3310 BlockDriverState
*base
, const char *backing_file_str
)
3312 BlockDriverState
*new_top_bs
= NULL
;
3313 Error
*local_err
= NULL
;
3316 if (!top
->drv
|| !base
->drv
) {
3320 new_top_bs
= bdrv_find_overlay(active
, top
);
3322 if (new_top_bs
== NULL
) {
3323 /* we could not find the image above 'top', this is an error */
3327 /* special case of new_top_bs->backing->bs already pointing to base - nothing
3328 * to do, no intermediate images */
3329 if (backing_bs(new_top_bs
) == base
) {
3334 /* Make sure that base is in the backing chain of top */
3335 if (!bdrv_chain_contains(top
, base
)) {
3339 /* success - we can delete the intermediate states, and link top->base */
3340 backing_file_str
= backing_file_str
? backing_file_str
: base
->filename
;
3341 ret
= bdrv_change_backing_file(new_top_bs
, backing_file_str
,
3342 base
->drv
? base
->drv
->format_name
: "");
3347 bdrv_set_backing_hd(new_top_bs
, base
, &local_err
);
3350 error_report_err(local_err
);
3360 * Truncate file to 'offset' bytes (needed only for file protocols)
3362 int bdrv_truncate(BdrvChild
*child
, int64_t offset
, Error
**errp
)
3364 BlockDriverState
*bs
= child
->bs
;
3365 BlockDriver
*drv
= bs
->drv
;
3368 assert(child
->perm
& BLK_PERM_RESIZE
);
3371 error_setg(errp
, "No medium inserted");
3374 if (!drv
->bdrv_truncate
) {
3375 error_setg(errp
, "Image format driver does not support resize");
3378 if (bs
->read_only
) {
3379 error_setg(errp
, "Image is read-only");
3383 assert(!(bs
->open_flags
& BDRV_O_INACTIVE
));
3385 ret
= drv
->bdrv_truncate(bs
, offset
, errp
);
3387 ret
= refresh_total_sectors(bs
, offset
>> BDRV_SECTOR_BITS
);
3388 bdrv_dirty_bitmap_truncate(bs
);
3389 bdrv_parent_cb_resize(bs
);
3396 * Length of a allocated file in bytes. Sparse files are counted by actual
3397 * allocated space. Return < 0 if error or unknown.
3399 int64_t bdrv_get_allocated_file_size(BlockDriverState
*bs
)
3401 BlockDriver
*drv
= bs
->drv
;
3405 if (drv
->bdrv_get_allocated_file_size
) {
3406 return drv
->bdrv_get_allocated_file_size(bs
);
3409 return bdrv_get_allocated_file_size(bs
->file
->bs
);
3415 * Return number of sectors on success, -errno on error.
3417 int64_t bdrv_nb_sectors(BlockDriverState
*bs
)
3419 BlockDriver
*drv
= bs
->drv
;
3424 if (drv
->has_variable_length
) {
3425 int ret
= refresh_total_sectors(bs
, bs
->total_sectors
);
3430 return bs
->total_sectors
;
3434 * Return length in bytes on success, -errno on error.
3435 * The length is always a multiple of BDRV_SECTOR_SIZE.
3437 int64_t bdrv_getlength(BlockDriverState
*bs
)
3439 int64_t ret
= bdrv_nb_sectors(bs
);
3441 ret
= ret
> INT64_MAX
/ BDRV_SECTOR_SIZE
? -EFBIG
: ret
;
3442 return ret
< 0 ? ret
: ret
* BDRV_SECTOR_SIZE
;
3445 /* return 0 as number of sectors if no device present or error */
3446 void bdrv_get_geometry(BlockDriverState
*bs
, uint64_t *nb_sectors_ptr
)
3448 int64_t nb_sectors
= bdrv_nb_sectors(bs
);
3450 *nb_sectors_ptr
= nb_sectors
< 0 ? 0 : nb_sectors
;
3453 bool bdrv_is_sg(BlockDriverState
*bs
)
3458 bool bdrv_is_encrypted(BlockDriverState
*bs
)
3460 if (bs
->backing
&& bs
->backing
->bs
->encrypted
) {
3463 return bs
->encrypted
;
3466 bool bdrv_key_required(BlockDriverState
*bs
)
3468 BdrvChild
*backing
= bs
->backing
;
3470 if (backing
&& backing
->bs
->encrypted
&& !backing
->bs
->valid_key
) {
3473 return (bs
->encrypted
&& !bs
->valid_key
);
3476 int bdrv_set_key(BlockDriverState
*bs
, const char *key
)
3479 if (bs
->backing
&& bs
->backing
->bs
->encrypted
) {
3480 ret
= bdrv_set_key(bs
->backing
->bs
, key
);
3486 if (!bs
->encrypted
) {
3488 } else if (!bs
->drv
|| !bs
->drv
->bdrv_set_key
) {
3491 ret
= bs
->drv
->bdrv_set_key(bs
, key
);
3493 bs
->valid_key
= false;
3494 } else if (!bs
->valid_key
) {
3495 /* call the change callback now, we skipped it on open */
3496 bs
->valid_key
= true;
3497 bdrv_parent_cb_change_media(bs
, true);
3503 * Provide an encryption key for @bs.
3504 * If @key is non-null:
3505 * If @bs is not encrypted, fail.
3506 * Else if the key is invalid, fail.
3507 * Else set @bs's key to @key, replacing the existing key, if any.
3509 * If @bs is encrypted and still lacks a key, fail.
3511 * On failure, store an error object through @errp if non-null.
3513 void bdrv_add_key(BlockDriverState
*bs
, const char *key
, Error
**errp
)
3516 if (!bdrv_is_encrypted(bs
)) {
3517 error_setg(errp
, "Node '%s' is not encrypted",
3518 bdrv_get_device_or_node_name(bs
));
3519 } else if (bdrv_set_key(bs
, key
) < 0) {
3520 error_setg(errp
, QERR_INVALID_PASSWORD
);
3523 if (bdrv_key_required(bs
)) {
3524 error_set(errp
, ERROR_CLASS_DEVICE_ENCRYPTED
,
3525 "'%s' (%s) is encrypted",
3526 bdrv_get_device_or_node_name(bs
),
3527 bdrv_get_encrypted_filename(bs
));
3532 const char *bdrv_get_format_name(BlockDriverState
*bs
)
3534 return bs
->drv
? bs
->drv
->format_name
: NULL
;
3537 static int qsort_strcmp(const void *a
, const void *b
)
3539 return strcmp(*(char *const *)a
, *(char *const *)b
);
3542 void bdrv_iterate_format(void (*it
)(void *opaque
, const char *name
),
3548 const char **formats
= NULL
;
3550 QLIST_FOREACH(drv
, &bdrv_drivers
, list
) {
3551 if (drv
->format_name
) {
3554 while (formats
&& i
&& !found
) {
3555 found
= !strcmp(formats
[--i
], drv
->format_name
);
3559 formats
= g_renew(const char *, formats
, count
+ 1);
3560 formats
[count
++] = drv
->format_name
;
3565 for (i
= 0; i
< (int)ARRAY_SIZE(block_driver_modules
); i
++) {
3566 const char *format_name
= block_driver_modules
[i
].format_name
;
3572 while (formats
&& j
&& !found
) {
3573 found
= !strcmp(formats
[--j
], format_name
);
3577 formats
= g_renew(const char *, formats
, count
+ 1);
3578 formats
[count
++] = format_name
;
3583 qsort(formats
, count
, sizeof(formats
[0]), qsort_strcmp
);
3585 for (i
= 0; i
< count
; i
++) {
3586 it(opaque
, formats
[i
]);
3592 /* This function is to find a node in the bs graph */
3593 BlockDriverState
*bdrv_find_node(const char *node_name
)
3595 BlockDriverState
*bs
;
3599 QTAILQ_FOREACH(bs
, &graph_bdrv_states
, node_list
) {
3600 if (!strcmp(node_name
, bs
->node_name
)) {
3607 /* Put this QMP function here so it can access the static graph_bdrv_states. */
3608 BlockDeviceInfoList
*bdrv_named_nodes_list(Error
**errp
)
3610 BlockDeviceInfoList
*list
, *entry
;
3611 BlockDriverState
*bs
;
3614 QTAILQ_FOREACH(bs
, &graph_bdrv_states
, node_list
) {
3615 BlockDeviceInfo
*info
= bdrv_block_device_info(NULL
, bs
, errp
);
3617 qapi_free_BlockDeviceInfoList(list
);
3620 entry
= g_malloc0(sizeof(*entry
));
3621 entry
->value
= info
;
3629 BlockDriverState
*bdrv_lookup_bs(const char *device
,
3630 const char *node_name
,
3634 BlockDriverState
*bs
;
3637 blk
= blk_by_name(device
);
3642 error_setg(errp
, "Device '%s' has no medium", device
);
3650 bs
= bdrv_find_node(node_name
);
3657 error_setg(errp
, "Cannot find device=%s nor node_name=%s",
3658 device
? device
: "",
3659 node_name
? node_name
: "");
3663 /* If 'base' is in the same chain as 'top', return true. Otherwise,
3664 * return false. If either argument is NULL, return false. */
3665 bool bdrv_chain_contains(BlockDriverState
*top
, BlockDriverState
*base
)
3667 while (top
&& top
!= base
) {
3668 top
= backing_bs(top
);
3674 BlockDriverState
*bdrv_next_node(BlockDriverState
*bs
)
3677 return QTAILQ_FIRST(&graph_bdrv_states
);
3679 return QTAILQ_NEXT(bs
, node_list
);
3682 const char *bdrv_get_node_name(const BlockDriverState
*bs
)
3684 return bs
->node_name
;
3687 const char *bdrv_get_parent_name(const BlockDriverState
*bs
)
3692 /* If multiple parents have a name, just pick the first one. */
3693 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
3694 if (c
->role
->get_name
) {
3695 name
= c
->role
->get_name(c
);
3696 if (name
&& *name
) {
3705 /* TODO check what callers really want: bs->node_name or blk_name() */
3706 const char *bdrv_get_device_name(const BlockDriverState
*bs
)
3708 return bdrv_get_parent_name(bs
) ?: "";
3711 /* This can be used to identify nodes that might not have a device
3712 * name associated. Since node and device names live in the same
3713 * namespace, the result is unambiguous. The exception is if both are
3714 * absent, then this returns an empty (non-null) string. */
3715 const char *bdrv_get_device_or_node_name(const BlockDriverState
*bs
)
3717 return bdrv_get_parent_name(bs
) ?: bs
->node_name
;
3720 int bdrv_get_flags(BlockDriverState
*bs
)
3722 return bs
->open_flags
;
3725 int bdrv_has_zero_init_1(BlockDriverState
*bs
)
3730 int bdrv_has_zero_init(BlockDriverState
*bs
)
3734 /* If BS is a copy on write image, it is initialized to
3735 the contents of the base image, which may not be zeroes. */
3739 if (bs
->drv
->bdrv_has_zero_init
) {
3740 return bs
->drv
->bdrv_has_zero_init(bs
);
3747 bool bdrv_unallocated_blocks_are_zero(BlockDriverState
*bs
)
3749 BlockDriverInfo bdi
;
3755 if (bdrv_get_info(bs
, &bdi
) == 0) {
3756 return bdi
.unallocated_blocks_are_zero
;
3762 bool bdrv_can_write_zeroes_with_unmap(BlockDriverState
*bs
)
3764 BlockDriverInfo bdi
;
3766 if (!(bs
->open_flags
& BDRV_O_UNMAP
)) {
3770 if (bdrv_get_info(bs
, &bdi
) == 0) {
3771 return bdi
.can_write_zeroes_with_unmap
;
3777 const char *bdrv_get_encrypted_filename(BlockDriverState
*bs
)
3779 if (bs
->backing
&& bs
->backing
->bs
->encrypted
)
3780 return bs
->backing_file
;
3781 else if (bs
->encrypted
)
3782 return bs
->filename
;
3787 void bdrv_get_backing_filename(BlockDriverState
*bs
,
3788 char *filename
, int filename_size
)
3790 pstrcpy(filename
, filename_size
, bs
->backing_file
);
3793 int bdrv_get_info(BlockDriverState
*bs
, BlockDriverInfo
*bdi
)
3795 BlockDriver
*drv
= bs
->drv
;
3798 if (!drv
->bdrv_get_info
)
3800 memset(bdi
, 0, sizeof(*bdi
));
3801 return drv
->bdrv_get_info(bs
, bdi
);
3804 ImageInfoSpecific
*bdrv_get_specific_info(BlockDriverState
*bs
)
3806 BlockDriver
*drv
= bs
->drv
;
3807 if (drv
&& drv
->bdrv_get_specific_info
) {
3808 return drv
->bdrv_get_specific_info(bs
);
3813 void bdrv_debug_event(BlockDriverState
*bs
, BlkdebugEvent event
)
3815 if (!bs
|| !bs
->drv
|| !bs
->drv
->bdrv_debug_event
) {
3819 bs
->drv
->bdrv_debug_event(bs
, event
);
3822 int bdrv_debug_breakpoint(BlockDriverState
*bs
, const char *event
,
3825 while (bs
&& bs
->drv
&& !bs
->drv
->bdrv_debug_breakpoint
) {
3826 bs
= bs
->file
? bs
->file
->bs
: NULL
;
3829 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_breakpoint
) {
3830 return bs
->drv
->bdrv_debug_breakpoint(bs
, event
, tag
);
3836 int bdrv_debug_remove_breakpoint(BlockDriverState
*bs
, const char *tag
)
3838 while (bs
&& bs
->drv
&& !bs
->drv
->bdrv_debug_remove_breakpoint
) {
3839 bs
= bs
->file
? bs
->file
->bs
: NULL
;
3842 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_remove_breakpoint
) {
3843 return bs
->drv
->bdrv_debug_remove_breakpoint(bs
, tag
);
3849 int bdrv_debug_resume(BlockDriverState
*bs
, const char *tag
)
3851 while (bs
&& (!bs
->drv
|| !bs
->drv
->bdrv_debug_resume
)) {
3852 bs
= bs
->file
? bs
->file
->bs
: NULL
;
3855 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_resume
) {
3856 return bs
->drv
->bdrv_debug_resume(bs
, tag
);
3862 bool bdrv_debug_is_suspended(BlockDriverState
*bs
, const char *tag
)
3864 while (bs
&& bs
->drv
&& !bs
->drv
->bdrv_debug_is_suspended
) {
3865 bs
= bs
->file
? bs
->file
->bs
: NULL
;
3868 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_is_suspended
) {
3869 return bs
->drv
->bdrv_debug_is_suspended(bs
, tag
);
3875 /* backing_file can either be relative, or absolute, or a protocol. If it is
3876 * relative, it must be relative to the chain. So, passing in bs->filename
3877 * from a BDS as backing_file should not be done, as that may be relative to
3878 * the CWD rather than the chain. */
3879 BlockDriverState
*bdrv_find_backing_image(BlockDriverState
*bs
,
3880 const char *backing_file
)
3882 char *filename_full
= NULL
;
3883 char *backing_file_full
= NULL
;
3884 char *filename_tmp
= NULL
;
3885 int is_protocol
= 0;
3886 BlockDriverState
*curr_bs
= NULL
;
3887 BlockDriverState
*retval
= NULL
;
3888 Error
*local_error
= NULL
;
3890 if (!bs
|| !bs
->drv
|| !backing_file
) {
3894 filename_full
= g_malloc(PATH_MAX
);
3895 backing_file_full
= g_malloc(PATH_MAX
);
3896 filename_tmp
= g_malloc(PATH_MAX
);
3898 is_protocol
= path_has_protocol(backing_file
);
3900 for (curr_bs
= bs
; curr_bs
->backing
; curr_bs
= curr_bs
->backing
->bs
) {
3902 /* If either of the filename paths is actually a protocol, then
3903 * compare unmodified paths; otherwise make paths relative */
3904 if (is_protocol
|| path_has_protocol(curr_bs
->backing_file
)) {
3905 if (strcmp(backing_file
, curr_bs
->backing_file
) == 0) {
3906 retval
= curr_bs
->backing
->bs
;
3909 /* Also check against the full backing filename for the image */
3910 bdrv_get_full_backing_filename(curr_bs
, backing_file_full
, PATH_MAX
,
3912 if (local_error
== NULL
) {
3913 if (strcmp(backing_file
, backing_file_full
) == 0) {
3914 retval
= curr_bs
->backing
->bs
;
3918 error_free(local_error
);
3922 /* If not an absolute filename path, make it relative to the current
3923 * image's filename path */
3924 path_combine(filename_tmp
, PATH_MAX
, curr_bs
->filename
,
3927 /* We are going to compare absolute pathnames */
3928 if (!realpath(filename_tmp
, filename_full
)) {
3932 /* We need to make sure the backing filename we are comparing against
3933 * is relative to the current image filename (or absolute) */
3934 path_combine(filename_tmp
, PATH_MAX
, curr_bs
->filename
,
3935 curr_bs
->backing_file
);
3937 if (!realpath(filename_tmp
, backing_file_full
)) {
3941 if (strcmp(backing_file_full
, filename_full
) == 0) {
3942 retval
= curr_bs
->backing
->bs
;
3948 g_free(filename_full
);
3949 g_free(backing_file_full
);
3950 g_free(filename_tmp
);
3954 int bdrv_get_backing_file_depth(BlockDriverState
*bs
)
3964 return 1 + bdrv_get_backing_file_depth(bs
->backing
->bs
);
3967 void bdrv_init(void)
3969 module_call_init(MODULE_INIT_BLOCK
);
3972 void bdrv_init_with_whitelist(void)
3974 use_bdrv_whitelist
= 1;
3978 void bdrv_invalidate_cache(BlockDriverState
*bs
, Error
**errp
)
3980 BdrvChild
*child
, *parent
;
3981 uint64_t perm
, shared_perm
;
3982 Error
*local_err
= NULL
;
3989 if (!(bs
->open_flags
& BDRV_O_INACTIVE
)) {
3993 QLIST_FOREACH(child
, &bs
->children
, next
) {
3994 bdrv_invalidate_cache(child
->bs
, &local_err
);
3996 error_propagate(errp
, local_err
);
4001 bs
->open_flags
&= ~BDRV_O_INACTIVE
;
4002 if (bs
->drv
->bdrv_invalidate_cache
) {
4003 bs
->drv
->bdrv_invalidate_cache(bs
, &local_err
);
4005 bs
->open_flags
|= BDRV_O_INACTIVE
;
4006 error_propagate(errp
, local_err
);
4011 ret
= refresh_total_sectors(bs
, bs
->total_sectors
);
4013 bs
->open_flags
|= BDRV_O_INACTIVE
;
4014 error_setg_errno(errp
, -ret
, "Could not refresh total sector count");
4018 /* Update permissions, they may differ for inactive nodes */
4019 bdrv_get_cumulative_perm(bs
, &perm
, &shared_perm
);
4020 ret
= bdrv_check_perm(bs
, perm
, shared_perm
, NULL
, &local_err
);
4022 bs
->open_flags
|= BDRV_O_INACTIVE
;
4023 error_propagate(errp
, local_err
);
4026 bdrv_set_perm(bs
, perm
, shared_perm
);
4028 QLIST_FOREACH(parent
, &bs
->parents
, next_parent
) {
4029 if (parent
->role
->activate
) {
4030 parent
->role
->activate(parent
, &local_err
);
4032 error_propagate(errp
, local_err
);
4039 void bdrv_invalidate_cache_all(Error
**errp
)
4041 BlockDriverState
*bs
;
4042 Error
*local_err
= NULL
;
4043 BdrvNextIterator it
;
4045 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
4046 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
4048 aio_context_acquire(aio_context
);
4049 bdrv_invalidate_cache(bs
, &local_err
);
4050 aio_context_release(aio_context
);
4052 error_propagate(errp
, local_err
);
4058 static int bdrv_inactivate_recurse(BlockDriverState
*bs
,
4061 BdrvChild
*child
, *parent
;
4064 if (!setting_flag
&& bs
->drv
->bdrv_inactivate
) {
4065 ret
= bs
->drv
->bdrv_inactivate(bs
);
4072 uint64_t perm
, shared_perm
;
4074 bs
->open_flags
|= BDRV_O_INACTIVE
;
4076 QLIST_FOREACH(parent
, &bs
->parents
, next_parent
) {
4077 if (parent
->role
->inactivate
) {
4078 ret
= parent
->role
->inactivate(parent
);
4080 bs
->open_flags
&= ~BDRV_O_INACTIVE
;
4086 /* Update permissions, they may differ for inactive nodes */
4087 bdrv_get_cumulative_perm(bs
, &perm
, &shared_perm
);
4088 bdrv_check_perm(bs
, perm
, shared_perm
, NULL
, &error_abort
);
4089 bdrv_set_perm(bs
, perm
, shared_perm
);
4092 QLIST_FOREACH(child
, &bs
->children
, next
) {
4093 ret
= bdrv_inactivate_recurse(child
->bs
, setting_flag
);
4102 int bdrv_inactivate_all(void)
4104 BlockDriverState
*bs
= NULL
;
4105 BdrvNextIterator it
;
4109 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
4110 aio_context_acquire(bdrv_get_aio_context(bs
));
4113 /* We do two passes of inactivation. The first pass calls to drivers'
4114 * .bdrv_inactivate callbacks recursively so all cache is flushed to disk;
4115 * the second pass sets the BDRV_O_INACTIVE flag so that no further write
4117 for (pass
= 0; pass
< 2; pass
++) {
4118 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
4119 ret
= bdrv_inactivate_recurse(bs
, pass
);
4127 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
4128 aio_context_release(bdrv_get_aio_context(bs
));
4134 /**************************************************************/
4135 /* removable device support */
4138 * Return TRUE if the media is present
4140 bool bdrv_is_inserted(BlockDriverState
*bs
)
4142 BlockDriver
*drv
= bs
->drv
;
4148 if (drv
->bdrv_is_inserted
) {
4149 return drv
->bdrv_is_inserted(bs
);
4151 QLIST_FOREACH(child
, &bs
->children
, next
) {
4152 if (!bdrv_is_inserted(child
->bs
)) {
4160 * Return whether the media changed since the last call to this
4161 * function, or -ENOTSUP if we don't know. Most drivers don't know.
4163 int bdrv_media_changed(BlockDriverState
*bs
)
4165 BlockDriver
*drv
= bs
->drv
;
4167 if (drv
&& drv
->bdrv_media_changed
) {
4168 return drv
->bdrv_media_changed(bs
);
4174 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
4176 void bdrv_eject(BlockDriverState
*bs
, bool eject_flag
)
4178 BlockDriver
*drv
= bs
->drv
;
4180 if (drv
&& drv
->bdrv_eject
) {
4181 drv
->bdrv_eject(bs
, eject_flag
);
4186 * Lock or unlock the media (if it is locked, the user won't be able
4187 * to eject it manually).
4189 void bdrv_lock_medium(BlockDriverState
*bs
, bool locked
)
4191 BlockDriver
*drv
= bs
->drv
;
4193 trace_bdrv_lock_medium(bs
, locked
);
4195 if (drv
&& drv
->bdrv_lock_medium
) {
4196 drv
->bdrv_lock_medium(bs
, locked
);
4200 /* Get a reference to bs */
4201 void bdrv_ref(BlockDriverState
*bs
)
4206 /* Release a previously grabbed reference to bs.
4207 * If after releasing, reference count is zero, the BlockDriverState is
4209 void bdrv_unref(BlockDriverState
*bs
)
4214 assert(bs
->refcnt
> 0);
4215 if (--bs
->refcnt
== 0) {
4220 struct BdrvOpBlocker
{
4222 QLIST_ENTRY(BdrvOpBlocker
) list
;
4225 bool bdrv_op_is_blocked(BlockDriverState
*bs
, BlockOpType op
, Error
**errp
)
4227 BdrvOpBlocker
*blocker
;
4228 assert((int) op
>= 0 && op
< BLOCK_OP_TYPE_MAX
);
4229 if (!QLIST_EMPTY(&bs
->op_blockers
[op
])) {
4230 blocker
= QLIST_FIRST(&bs
->op_blockers
[op
]);
4232 *errp
= error_copy(blocker
->reason
);
4233 error_prepend(errp
, "Node '%s' is busy: ",
4234 bdrv_get_device_or_node_name(bs
));
4241 void bdrv_op_block(BlockDriverState
*bs
, BlockOpType op
, Error
*reason
)
4243 BdrvOpBlocker
*blocker
;
4244 assert((int) op
>= 0 && op
< BLOCK_OP_TYPE_MAX
);
4246 blocker
= g_new0(BdrvOpBlocker
, 1);
4247 blocker
->reason
= reason
;
4248 QLIST_INSERT_HEAD(&bs
->op_blockers
[op
], blocker
, list
);
4251 void bdrv_op_unblock(BlockDriverState
*bs
, BlockOpType op
, Error
*reason
)
4253 BdrvOpBlocker
*blocker
, *next
;
4254 assert((int) op
>= 0 && op
< BLOCK_OP_TYPE_MAX
);
4255 QLIST_FOREACH_SAFE(blocker
, &bs
->op_blockers
[op
], list
, next
) {
4256 if (blocker
->reason
== reason
) {
4257 QLIST_REMOVE(blocker
, list
);
4263 void bdrv_op_block_all(BlockDriverState
*bs
, Error
*reason
)
4266 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
4267 bdrv_op_block(bs
, i
, reason
);
4271 void bdrv_op_unblock_all(BlockDriverState
*bs
, Error
*reason
)
4274 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
4275 bdrv_op_unblock(bs
, i
, reason
);
4279 bool bdrv_op_blocker_is_empty(BlockDriverState
*bs
)
4283 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
4284 if (!QLIST_EMPTY(&bs
->op_blockers
[i
])) {
4291 void bdrv_img_create(const char *filename
, const char *fmt
,
4292 const char *base_filename
, const char *base_fmt
,
4293 char *options
, uint64_t img_size
, int flags
, bool quiet
,
4296 QemuOptsList
*create_opts
= NULL
;
4297 QemuOpts
*opts
= NULL
;
4298 const char *backing_fmt
, *backing_file
;
4300 BlockDriver
*drv
, *proto_drv
;
4301 Error
*local_err
= NULL
;
4304 /* Find driver and parse its options */
4305 drv
= bdrv_find_format(fmt
);
4307 error_setg(errp
, "Unknown file format '%s'", fmt
);
4311 proto_drv
= bdrv_find_protocol(filename
, true, errp
);
4316 if (!drv
->create_opts
) {
4317 error_setg(errp
, "Format driver '%s' does not support image creation",
4322 if (!proto_drv
->create_opts
) {
4323 error_setg(errp
, "Protocol driver '%s' does not support image creation",
4324 proto_drv
->format_name
);
4328 create_opts
= qemu_opts_append(create_opts
, drv
->create_opts
);
4329 create_opts
= qemu_opts_append(create_opts
, proto_drv
->create_opts
);
4331 /* Create parameter list with default values */
4332 opts
= qemu_opts_create(create_opts
, NULL
, 0, &error_abort
);
4333 qemu_opt_set_number(opts
, BLOCK_OPT_SIZE
, img_size
, &error_abort
);
4335 /* Parse -o options */
4337 qemu_opts_do_parse(opts
, options
, NULL
, &local_err
);
4339 error_report_err(local_err
);
4341 error_setg(errp
, "Invalid options for file format '%s'", fmt
);
4346 if (base_filename
) {
4347 qemu_opt_set(opts
, BLOCK_OPT_BACKING_FILE
, base_filename
, &local_err
);
4349 error_setg(errp
, "Backing file not supported for file format '%s'",
4356 qemu_opt_set(opts
, BLOCK_OPT_BACKING_FMT
, base_fmt
, &local_err
);
4358 error_setg(errp
, "Backing file format not supported for file "
4359 "format '%s'", fmt
);
4364 backing_file
= qemu_opt_get(opts
, BLOCK_OPT_BACKING_FILE
);
4366 if (!strcmp(filename
, backing_file
)) {
4367 error_setg(errp
, "Error: Trying to create an image with the "
4368 "same filename as the backing file");
4373 backing_fmt
= qemu_opt_get(opts
, BLOCK_OPT_BACKING_FMT
);
4375 // The size for the image must always be specified, with one exception:
4376 // If we are using a backing file, we can obtain the size from there
4377 size
= qemu_opt_get_size(opts
, BLOCK_OPT_SIZE
, 0);
4380 BlockDriverState
*bs
;
4381 char *full_backing
= g_new0(char, PATH_MAX
);
4384 QDict
*backing_options
= NULL
;
4386 bdrv_get_full_backing_filename_from_filename(filename
, backing_file
,
4387 full_backing
, PATH_MAX
,
4390 g_free(full_backing
);
4394 /* backing files always opened read-only */
4396 back_flags
&= ~(BDRV_O_RDWR
| BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
);
4399 backing_options
= qdict_new();
4400 qdict_put_str(backing_options
, "driver", backing_fmt
);
4403 bs
= bdrv_open(full_backing
, NULL
, backing_options
, back_flags
,
4405 g_free(full_backing
);
4409 size
= bdrv_getlength(bs
);
4411 error_setg_errno(errp
, -size
, "Could not get size of '%s'",
4417 qemu_opt_set_number(opts
, BLOCK_OPT_SIZE
, size
, &error_abort
);
4421 error_setg(errp
, "Image creation needs a size parameter");
4427 printf("Formatting '%s', fmt=%s ", filename
, fmt
);
4428 qemu_opts_print(opts
, " ");
4432 ret
= bdrv_create(drv
, filename
, opts
, &local_err
);
4434 if (ret
== -EFBIG
) {
4435 /* This is generally a better message than whatever the driver would
4436 * deliver (especially because of the cluster_size_hint), since that
4437 * is most probably not much different from "image too large". */
4438 const char *cluster_size_hint
= "";
4439 if (qemu_opt_get_size(opts
, BLOCK_OPT_CLUSTER_SIZE
, 0)) {
4440 cluster_size_hint
= " (try using a larger cluster size)";
4442 error_setg(errp
, "The image size is too large for file format '%s'"
4443 "%s", fmt
, cluster_size_hint
);
4444 error_free(local_err
);
4449 qemu_opts_del(opts
);
4450 qemu_opts_free(create_opts
);
4451 error_propagate(errp
, local_err
);
4454 AioContext
*bdrv_get_aio_context(BlockDriverState
*bs
)
4456 return bs
->aio_context
;
4459 void bdrv_coroutine_enter(BlockDriverState
*bs
, Coroutine
*co
)
4461 aio_co_enter(bdrv_get_aio_context(bs
), co
);
4464 static void bdrv_do_remove_aio_context_notifier(BdrvAioNotifier
*ban
)
4466 QLIST_REMOVE(ban
, list
);
4470 void bdrv_detach_aio_context(BlockDriverState
*bs
)
4472 BdrvAioNotifier
*baf
, *baf_tmp
;
4479 assert(!bs
->walking_aio_notifiers
);
4480 bs
->walking_aio_notifiers
= true;
4481 QLIST_FOREACH_SAFE(baf
, &bs
->aio_notifiers
, list
, baf_tmp
) {
4483 bdrv_do_remove_aio_context_notifier(baf
);
4485 baf
->detach_aio_context(baf
->opaque
);
4488 /* Never mind iterating again to check for ->deleted. bdrv_close() will
4489 * remove remaining aio notifiers if we aren't called again.
4491 bs
->walking_aio_notifiers
= false;
4493 if (bs
->drv
->bdrv_detach_aio_context
) {
4494 bs
->drv
->bdrv_detach_aio_context(bs
);
4496 QLIST_FOREACH(child
, &bs
->children
, next
) {
4497 bdrv_detach_aio_context(child
->bs
);
4500 bs
->aio_context
= NULL
;
4503 void bdrv_attach_aio_context(BlockDriverState
*bs
,
4504 AioContext
*new_context
)
4506 BdrvAioNotifier
*ban
, *ban_tmp
;
4513 bs
->aio_context
= new_context
;
4515 QLIST_FOREACH(child
, &bs
->children
, next
) {
4516 bdrv_attach_aio_context(child
->bs
, new_context
);
4518 if (bs
->drv
->bdrv_attach_aio_context
) {
4519 bs
->drv
->bdrv_attach_aio_context(bs
, new_context
);
4522 assert(!bs
->walking_aio_notifiers
);
4523 bs
->walking_aio_notifiers
= true;
4524 QLIST_FOREACH_SAFE(ban
, &bs
->aio_notifiers
, list
, ban_tmp
) {
4526 bdrv_do_remove_aio_context_notifier(ban
);
4528 ban
->attached_aio_context(new_context
, ban
->opaque
);
4531 bs
->walking_aio_notifiers
= false;
4534 void bdrv_set_aio_context(BlockDriverState
*bs
, AioContext
*new_context
)
4536 AioContext
*ctx
= bdrv_get_aio_context(bs
);
4538 aio_disable_external(ctx
);
4539 bdrv_parent_drained_begin(bs
);
4540 bdrv_drain(bs
); /* ensure there are no in-flight requests */
4542 while (aio_poll(ctx
, false)) {
4543 /* wait for all bottom halves to execute */
4546 bdrv_detach_aio_context(bs
);
4548 /* This function executes in the old AioContext so acquire the new one in
4549 * case it runs in a different thread.
4551 aio_context_acquire(new_context
);
4552 bdrv_attach_aio_context(bs
, new_context
);
4553 bdrv_parent_drained_end(bs
);
4554 aio_enable_external(ctx
);
4555 aio_context_release(new_context
);
4558 void bdrv_add_aio_context_notifier(BlockDriverState
*bs
,
4559 void (*attached_aio_context
)(AioContext
*new_context
, void *opaque
),
4560 void (*detach_aio_context
)(void *opaque
), void *opaque
)
4562 BdrvAioNotifier
*ban
= g_new(BdrvAioNotifier
, 1);
4563 *ban
= (BdrvAioNotifier
){
4564 .attached_aio_context
= attached_aio_context
,
4565 .detach_aio_context
= detach_aio_context
,
4569 QLIST_INSERT_HEAD(&bs
->aio_notifiers
, ban
, list
);
4572 void bdrv_remove_aio_context_notifier(BlockDriverState
*bs
,
4573 void (*attached_aio_context
)(AioContext
*,
4575 void (*detach_aio_context
)(void *),
4578 BdrvAioNotifier
*ban
, *ban_next
;
4580 QLIST_FOREACH_SAFE(ban
, &bs
->aio_notifiers
, list
, ban_next
) {
4581 if (ban
->attached_aio_context
== attached_aio_context
&&
4582 ban
->detach_aio_context
== detach_aio_context
&&
4583 ban
->opaque
== opaque
&&
4584 ban
->deleted
== false)
4586 if (bs
->walking_aio_notifiers
) {
4587 ban
->deleted
= true;
4589 bdrv_do_remove_aio_context_notifier(ban
);
4598 int bdrv_amend_options(BlockDriverState
*bs
, QemuOpts
*opts
,
4599 BlockDriverAmendStatusCB
*status_cb
, void *cb_opaque
)
4601 if (!bs
->drv
->bdrv_amend_options
) {
4604 return bs
->drv
->bdrv_amend_options(bs
, opts
, status_cb
, cb_opaque
);
4607 /* This function will be called by the bdrv_recurse_is_first_non_filter method
4608 * of block filter and by bdrv_is_first_non_filter.
4609 * It is used to test if the given bs is the candidate or recurse more in the
4612 bool bdrv_recurse_is_first_non_filter(BlockDriverState
*bs
,
4613 BlockDriverState
*candidate
)
4615 /* return false if basic checks fails */
4616 if (!bs
|| !bs
->drv
) {
4620 /* the code reached a non block filter driver -> check if the bs is
4621 * the same as the candidate. It's the recursion termination condition.
4623 if (!bs
->drv
->is_filter
) {
4624 return bs
== candidate
;
4626 /* Down this path the driver is a block filter driver */
4628 /* If the block filter recursion method is defined use it to recurse down
4631 if (bs
->drv
->bdrv_recurse_is_first_non_filter
) {
4632 return bs
->drv
->bdrv_recurse_is_first_non_filter(bs
, candidate
);
4635 /* the driver is a block filter but don't allow to recurse -> return false
4640 /* This function checks if the candidate is the first non filter bs down it's
4641 * bs chain. Since we don't have pointers to parents it explore all bs chains
4642 * from the top. Some filters can choose not to pass down the recursion.
4644 bool bdrv_is_first_non_filter(BlockDriverState
*candidate
)
4646 BlockDriverState
*bs
;
4647 BdrvNextIterator it
;
4649 /* walk down the bs forest recursively */
4650 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
4653 /* try to recurse in this top level bs */
4654 perm
= bdrv_recurse_is_first_non_filter(bs
, candidate
);
4656 /* candidate is the first non filter */
4665 BlockDriverState
*check_to_replace_node(BlockDriverState
*parent_bs
,
4666 const char *node_name
, Error
**errp
)
4668 BlockDriverState
*to_replace_bs
= bdrv_find_node(node_name
);
4669 AioContext
*aio_context
;
4671 if (!to_replace_bs
) {
4672 error_setg(errp
, "Node name '%s' not found", node_name
);
4676 aio_context
= bdrv_get_aio_context(to_replace_bs
);
4677 aio_context_acquire(aio_context
);
4679 if (bdrv_op_is_blocked(to_replace_bs
, BLOCK_OP_TYPE_REPLACE
, errp
)) {
4680 to_replace_bs
= NULL
;
4684 /* We don't want arbitrary node of the BDS chain to be replaced only the top
4685 * most non filter in order to prevent data corruption.
4686 * Another benefit is that this tests exclude backing files which are
4687 * blocked by the backing blockers.
4689 if (!bdrv_recurse_is_first_non_filter(parent_bs
, to_replace_bs
)) {
4690 error_setg(errp
, "Only top most non filter can be replaced");
4691 to_replace_bs
= NULL
;
4696 aio_context_release(aio_context
);
4697 return to_replace_bs
;
4700 static bool append_open_options(QDict
*d
, BlockDriverState
*bs
)
4702 const QDictEntry
*entry
;
4705 bool found_any
= false;
4708 for (entry
= qdict_first(bs
->options
); entry
;
4709 entry
= qdict_next(bs
->options
, entry
))
4711 /* Exclude options for children */
4712 QLIST_FOREACH(child
, &bs
->children
, next
) {
4713 if (strstart(qdict_entry_key(entry
), child
->name
, &p
)
4714 && (!*p
|| *p
== '.'))
4723 /* And exclude all non-driver-specific options */
4724 for (desc
= bdrv_runtime_opts
.desc
; desc
->name
; desc
++) {
4725 if (!strcmp(qdict_entry_key(entry
), desc
->name
)) {
4733 qobject_incref(qdict_entry_value(entry
));
4734 qdict_put_obj(d
, qdict_entry_key(entry
), qdict_entry_value(entry
));
4741 /* Updates the following BDS fields:
4742 * - exact_filename: A filename which may be used for opening a block device
4743 * which (mostly) equals the given BDS (even without any
4744 * other options; so reading and writing must return the same
4745 * results, but caching etc. may be different)
4746 * - full_open_options: Options which, when given when opening a block device
4747 * (without a filename), result in a BDS (mostly)
4748 * equalling the given one
4749 * - filename: If exact_filename is set, it is copied here. Otherwise,
4750 * full_open_options is converted to a JSON object, prefixed with
4751 * "json:" (for use through the JSON pseudo protocol) and put here.
4753 void bdrv_refresh_filename(BlockDriverState
*bs
)
4755 BlockDriver
*drv
= bs
->drv
;
4762 /* This BDS's file name will most probably depend on its file's name, so
4763 * refresh that first */
4765 bdrv_refresh_filename(bs
->file
->bs
);
4768 if (drv
->bdrv_refresh_filename
) {
4769 /* Obsolete information is of no use here, so drop the old file name
4770 * information before refreshing it */
4771 bs
->exact_filename
[0] = '\0';
4772 if (bs
->full_open_options
) {
4773 QDECREF(bs
->full_open_options
);
4774 bs
->full_open_options
= NULL
;
4778 append_open_options(opts
, bs
);
4779 drv
->bdrv_refresh_filename(bs
, opts
);
4781 } else if (bs
->file
) {
4782 /* Try to reconstruct valid information from the underlying file */
4783 bool has_open_options
;
4785 bs
->exact_filename
[0] = '\0';
4786 if (bs
->full_open_options
) {
4787 QDECREF(bs
->full_open_options
);
4788 bs
->full_open_options
= NULL
;
4792 has_open_options
= append_open_options(opts
, bs
);
4794 /* If no specific options have been given for this BDS, the filename of
4795 * the underlying file should suffice for this one as well */
4796 if (bs
->file
->bs
->exact_filename
[0] && !has_open_options
) {
4797 strcpy(bs
->exact_filename
, bs
->file
->bs
->exact_filename
);
4799 /* Reconstructing the full options QDict is simple for most format block
4800 * drivers, as long as the full options are known for the underlying
4801 * file BDS. The full options QDict of that file BDS should somehow
4802 * contain a representation of the filename, therefore the following
4803 * suffices without querying the (exact_)filename of this BDS. */
4804 if (bs
->file
->bs
->full_open_options
) {
4805 qdict_put_str(opts
, "driver", drv
->format_name
);
4806 QINCREF(bs
->file
->bs
->full_open_options
);
4807 qdict_put(opts
, "file", bs
->file
->bs
->full_open_options
);
4809 bs
->full_open_options
= opts
;
4813 } else if (!bs
->full_open_options
&& qdict_size(bs
->options
)) {
4814 /* There is no underlying file BDS (at least referenced by BDS.file),
4815 * so the full options QDict should be equal to the options given
4816 * specifically for this block device when it was opened (plus the
4817 * driver specification).
4818 * Because those options don't change, there is no need to update
4819 * full_open_options when it's already set. */
4822 append_open_options(opts
, bs
);
4823 qdict_put_str(opts
, "driver", drv
->format_name
);
4825 if (bs
->exact_filename
[0]) {
4826 /* This may not work for all block protocol drivers (some may
4827 * require this filename to be parsed), but we have to find some
4828 * default solution here, so just include it. If some block driver
4829 * does not support pure options without any filename at all or
4830 * needs some special format of the options QDict, it needs to
4831 * implement the driver-specific bdrv_refresh_filename() function.
4833 qdict_put_str(opts
, "filename", bs
->exact_filename
);
4836 bs
->full_open_options
= opts
;
4839 if (bs
->exact_filename
[0]) {
4840 pstrcpy(bs
->filename
, sizeof(bs
->filename
), bs
->exact_filename
);
4841 } else if (bs
->full_open_options
) {
4842 QString
*json
= qobject_to_json(QOBJECT(bs
->full_open_options
));
4843 snprintf(bs
->filename
, sizeof(bs
->filename
), "json:%s",
4844 qstring_get_str(json
));
4850 * Hot add/remove a BDS's child. So the user can take a child offline when
4851 * it is broken and take a new child online
4853 void bdrv_add_child(BlockDriverState
*parent_bs
, BlockDriverState
*child_bs
,
4857 if (!parent_bs
->drv
|| !parent_bs
->drv
->bdrv_add_child
) {
4858 error_setg(errp
, "The node %s does not support adding a child",
4859 bdrv_get_device_or_node_name(parent_bs
));
4863 if (!QLIST_EMPTY(&child_bs
->parents
)) {
4864 error_setg(errp
, "The node %s already has a parent",
4865 child_bs
->node_name
);
4869 parent_bs
->drv
->bdrv_add_child(parent_bs
, child_bs
, errp
);
4872 void bdrv_del_child(BlockDriverState
*parent_bs
, BdrvChild
*child
, Error
**errp
)
4876 if (!parent_bs
->drv
|| !parent_bs
->drv
->bdrv_del_child
) {
4877 error_setg(errp
, "The node %s does not support removing a child",
4878 bdrv_get_device_or_node_name(parent_bs
));
4882 QLIST_FOREACH(tmp
, &parent_bs
->children
, next
) {
4889 error_setg(errp
, "The node %s does not have a child named %s",
4890 bdrv_get_device_or_node_name(parent_bs
),
4891 bdrv_get_device_or_node_name(child
->bs
));
4895 parent_bs
->drv
->bdrv_del_child(parent_bs
, child
, errp
);