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
25 #include "qemu/osdep.h"
26 #include "block/trace.h"
27 #include "block/block_int.h"
28 #include "block/blockjob.h"
29 #include "block/nbd.h"
30 #include "block/qdict.h"
31 #include "qemu/error-report.h"
32 #include "module_block.h"
33 #include "qemu/module.h"
34 #include "qapi/error.h"
35 #include "qapi/qmp/qdict.h"
36 #include "qapi/qmp/qjson.h"
37 #include "qapi/qmp/qnull.h"
38 #include "qapi/qmp/qstring.h"
39 #include "qapi/qobject-output-visitor.h"
40 #include "qapi/qapi-visit-block-core.h"
41 #include "sysemu/block-backend.h"
42 #include "sysemu/sysemu.h"
43 #include "qemu/notify.h"
44 #include "qemu/option.h"
45 #include "qemu/coroutine.h"
46 #include "block/qapi.h"
47 #include "qemu/timer.h"
48 #include "qemu/cutils.h"
52 #include <sys/ioctl.h>
53 #include <sys/queue.h>
63 #define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */
65 static QTAILQ_HEAD(, BlockDriverState
) graph_bdrv_states
=
66 QTAILQ_HEAD_INITIALIZER(graph_bdrv_states
);
68 static QTAILQ_HEAD(, BlockDriverState
) all_bdrv_states
=
69 QTAILQ_HEAD_INITIALIZER(all_bdrv_states
);
71 static QLIST_HEAD(, BlockDriver
) bdrv_drivers
=
72 QLIST_HEAD_INITIALIZER(bdrv_drivers
);
74 static BlockDriverState
*bdrv_open_inherit(const char *filename
,
75 const char *reference
,
76 QDict
*options
, int flags
,
77 BlockDriverState
*parent
,
78 const BdrvChildRole
*child_role
,
81 /* If non-zero, use only whitelisted block drivers */
82 static int use_bdrv_whitelist
;
85 static int is_windows_drive_prefix(const char *filename
)
87 return (((filename
[0] >= 'a' && filename
[0] <= 'z') ||
88 (filename
[0] >= 'A' && filename
[0] <= 'Z')) &&
92 int is_windows_drive(const char *filename
)
94 if (is_windows_drive_prefix(filename
) &&
97 if (strstart(filename
, "\\\\.\\", NULL
) ||
98 strstart(filename
, "//./", NULL
))
104 size_t bdrv_opt_mem_align(BlockDriverState
*bs
)
106 if (!bs
|| !bs
->drv
) {
107 /* page size or 4k (hdd sector size) should be on the safe side */
108 return MAX(4096, getpagesize());
111 return bs
->bl
.opt_mem_alignment
;
114 size_t bdrv_min_mem_align(BlockDriverState
*bs
)
116 if (!bs
|| !bs
->drv
) {
117 /* page size or 4k (hdd sector size) should be on the safe side */
118 return MAX(4096, getpagesize());
121 return bs
->bl
.min_mem_alignment
;
124 /* check if the path starts with "<protocol>:" */
125 int path_has_protocol(const char *path
)
130 if (is_windows_drive(path
) ||
131 is_windows_drive_prefix(path
)) {
134 p
= path
+ strcspn(path
, ":/\\");
136 p
= path
+ strcspn(path
, ":/");
142 int path_is_absolute(const char *path
)
145 /* specific case for names like: "\\.\d:" */
146 if (is_windows_drive(path
) || is_windows_drive_prefix(path
)) {
149 return (*path
== '/' || *path
== '\\');
151 return (*path
== '/');
155 /* if filename is absolute, just copy it to dest. Otherwise, build a
156 path to it by considering it is relative to base_path. URL are
158 void path_combine(char *dest
, int dest_size
,
159 const char *base_path
,
160 const char *filename
)
167 if (path_is_absolute(filename
)) {
168 pstrcpy(dest
, dest_size
, filename
);
170 const char *protocol_stripped
= NULL
;
172 if (path_has_protocol(base_path
)) {
173 protocol_stripped
= strchr(base_path
, ':');
174 if (protocol_stripped
) {
178 p
= protocol_stripped
?: base_path
;
180 p1
= strrchr(base_path
, '/');
184 p2
= strrchr(base_path
, '\\');
196 if (len
> dest_size
- 1)
198 memcpy(dest
, base_path
, len
);
200 pstrcat(dest
, dest_size
, filename
);
205 * Helper function for bdrv_parse_filename() implementations to remove optional
206 * protocol prefixes (especially "file:") from a filename and for putting the
207 * stripped filename into the options QDict if there is such a prefix.
209 void bdrv_parse_filename_strip_prefix(const char *filename
, const char *prefix
,
212 if (strstart(filename
, prefix
, &filename
)) {
213 /* Stripping the explicit protocol prefix may result in a protocol
214 * prefix being (wrongly) detected (if the filename contains a colon) */
215 if (path_has_protocol(filename
)) {
216 QString
*fat_filename
;
218 /* This means there is some colon before the first slash; therefore,
219 * this cannot be an absolute path */
220 assert(!path_is_absolute(filename
));
222 /* And we can thus fix the protocol detection issue by prefixing it
224 fat_filename
= qstring_from_str("./");
225 qstring_append(fat_filename
, filename
);
227 assert(!path_has_protocol(qstring_get_str(fat_filename
)));
229 qdict_put(options
, "filename", fat_filename
);
231 /* If no protocol prefix was detected, we can use the shortened
233 qdict_put_str(options
, "filename", filename
);
239 /* Returns whether the image file is opened as read-only. Note that this can
240 * return false and writing to the image file is still not possible because the
241 * image is inactivated. */
242 bool bdrv_is_read_only(BlockDriverState
*bs
)
244 return bs
->read_only
;
247 int bdrv_can_set_read_only(BlockDriverState
*bs
, bool read_only
,
248 bool ignore_allow_rdw
, Error
**errp
)
250 /* Do not set read_only if copy_on_read is enabled */
251 if (bs
->copy_on_read
&& read_only
) {
252 error_setg(errp
, "Can't set node '%s' to r/o with copy-on-read enabled",
253 bdrv_get_device_or_node_name(bs
));
257 /* Do not clear read_only if it is prohibited */
258 if (!read_only
&& !(bs
->open_flags
& BDRV_O_ALLOW_RDWR
) &&
261 error_setg(errp
, "Node '%s' is read only",
262 bdrv_get_device_or_node_name(bs
));
270 * Called by a driver that can only provide a read-only image.
272 * Returns 0 if the node is already read-only or it could switch the node to
273 * read-only because BDRV_O_AUTO_RDONLY is set.
275 * Returns -EACCES if the node is read-write and BDRV_O_AUTO_RDONLY is not set
276 * or bdrv_can_set_read_only() forbids making the node read-only. If @errmsg
277 * is not NULL, it is used as the error message for the Error object.
279 int bdrv_apply_auto_read_only(BlockDriverState
*bs
, const char *errmsg
,
284 if (!(bs
->open_flags
& BDRV_O_RDWR
)) {
287 if (!(bs
->open_flags
& BDRV_O_AUTO_RDONLY
)) {
291 ret
= bdrv_can_set_read_only(bs
, true, false, NULL
);
296 bs
->read_only
= true;
297 bs
->open_flags
&= ~BDRV_O_RDWR
;
302 error_setg(errp
, "%s", errmsg
?: "Image is read-only");
306 void bdrv_get_full_backing_filename_from_filename(const char *backed
,
308 char *dest
, size_t sz
,
311 if (backing
[0] == '\0' || path_has_protocol(backing
) ||
312 path_is_absolute(backing
))
314 pstrcpy(dest
, sz
, backing
);
315 } else if (backed
[0] == '\0' || strstart(backed
, "json:", NULL
)) {
316 error_setg(errp
, "Cannot use relative backing file names for '%s'",
319 path_combine(dest
, sz
, backed
, backing
);
323 void bdrv_get_full_backing_filename(BlockDriverState
*bs
, char *dest
, size_t sz
,
328 bdrv_refresh_filename(bs
);
330 backed
= bs
->exact_filename
[0] ? bs
->exact_filename
: bs
->filename
;
331 bdrv_get_full_backing_filename_from_filename(backed
, bs
->backing_file
,
335 void bdrv_register(BlockDriver
*bdrv
)
337 QLIST_INSERT_HEAD(&bdrv_drivers
, bdrv
, list
);
340 BlockDriverState
*bdrv_new(void)
342 BlockDriverState
*bs
;
345 bs
= g_new0(BlockDriverState
, 1);
346 QLIST_INIT(&bs
->dirty_bitmaps
);
347 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
348 QLIST_INIT(&bs
->op_blockers
[i
]);
350 notifier_with_return_list_init(&bs
->before_write_notifiers
);
351 qemu_co_mutex_init(&bs
->reqs_lock
);
352 qemu_mutex_init(&bs
->dirty_bitmap_mutex
);
354 bs
->aio_context
= qemu_get_aio_context();
356 qemu_co_queue_init(&bs
->flush_queue
);
358 for (i
= 0; i
< bdrv_drain_all_count
; i
++) {
359 bdrv_drained_begin(bs
);
362 QTAILQ_INSERT_TAIL(&all_bdrv_states
, bs
, bs_list
);
367 static BlockDriver
*bdrv_do_find_format(const char *format_name
)
371 QLIST_FOREACH(drv1
, &bdrv_drivers
, list
) {
372 if (!strcmp(drv1
->format_name
, format_name
)) {
380 BlockDriver
*bdrv_find_format(const char *format_name
)
385 drv1
= bdrv_do_find_format(format_name
);
390 /* The driver isn't registered, maybe we need to load a module */
391 for (i
= 0; i
< (int)ARRAY_SIZE(block_driver_modules
); ++i
) {
392 if (!strcmp(block_driver_modules
[i
].format_name
, format_name
)) {
393 block_module_load_one(block_driver_modules
[i
].library_name
);
398 return bdrv_do_find_format(format_name
);
401 int bdrv_is_whitelisted(BlockDriver
*drv
, bool read_only
)
403 static const char *whitelist_rw
[] = {
404 CONFIG_BDRV_RW_WHITELIST
406 static const char *whitelist_ro
[] = {
407 CONFIG_BDRV_RO_WHITELIST
411 if (!whitelist_rw
[0] && !whitelist_ro
[0]) {
412 return 1; /* no whitelist, anything goes */
415 for (p
= whitelist_rw
; *p
; p
++) {
416 if (!strcmp(drv
->format_name
, *p
)) {
421 for (p
= whitelist_ro
; *p
; p
++) {
422 if (!strcmp(drv
->format_name
, *p
)) {
430 bool bdrv_uses_whitelist(void)
432 return use_bdrv_whitelist
;
435 typedef struct CreateCo
{
443 static void coroutine_fn
bdrv_create_co_entry(void *opaque
)
445 Error
*local_err
= NULL
;
448 CreateCo
*cco
= opaque
;
451 ret
= cco
->drv
->bdrv_co_create_opts(cco
->filename
, cco
->opts
, &local_err
);
452 error_propagate(&cco
->err
, local_err
);
456 int bdrv_create(BlockDriver
*drv
, const char* filename
,
457 QemuOpts
*opts
, Error
**errp
)
464 .filename
= g_strdup(filename
),
470 if (!drv
->bdrv_co_create_opts
) {
471 error_setg(errp
, "Driver '%s' does not support image creation", drv
->format_name
);
476 if (qemu_in_coroutine()) {
477 /* Fast-path if already in coroutine context */
478 bdrv_create_co_entry(&cco
);
480 co
= qemu_coroutine_create(bdrv_create_co_entry
, &cco
);
481 qemu_coroutine_enter(co
);
482 while (cco
.ret
== NOT_DONE
) {
483 aio_poll(qemu_get_aio_context(), true);
490 error_propagate(errp
, cco
.err
);
492 error_setg_errno(errp
, -ret
, "Could not create image");
497 g_free(cco
.filename
);
501 int bdrv_create_file(const char *filename
, QemuOpts
*opts
, Error
**errp
)
504 Error
*local_err
= NULL
;
507 drv
= bdrv_find_protocol(filename
, true, errp
);
512 ret
= bdrv_create(drv
, filename
, opts
, &local_err
);
513 error_propagate(errp
, local_err
);
518 * Try to get @bs's logical and physical block size.
519 * On success, store them in @bsz struct and return 0.
520 * On failure return -errno.
521 * @bs must not be empty.
523 int bdrv_probe_blocksizes(BlockDriverState
*bs
, BlockSizes
*bsz
)
525 BlockDriver
*drv
= bs
->drv
;
527 if (drv
&& drv
->bdrv_probe_blocksizes
) {
528 return drv
->bdrv_probe_blocksizes(bs
, bsz
);
529 } else if (drv
&& drv
->is_filter
&& bs
->file
) {
530 return bdrv_probe_blocksizes(bs
->file
->bs
, bsz
);
537 * Try to get @bs's geometry (cyls, heads, sectors).
538 * On success, store them in @geo struct and return 0.
539 * On failure return -errno.
540 * @bs must not be empty.
542 int bdrv_probe_geometry(BlockDriverState
*bs
, HDGeometry
*geo
)
544 BlockDriver
*drv
= bs
->drv
;
546 if (drv
&& drv
->bdrv_probe_geometry
) {
547 return drv
->bdrv_probe_geometry(bs
, geo
);
548 } else if (drv
&& drv
->is_filter
&& bs
->file
) {
549 return bdrv_probe_geometry(bs
->file
->bs
, geo
);
556 * Create a uniquely-named empty temporary file.
557 * Return 0 upon success, otherwise a negative errno value.
559 int get_tmp_filename(char *filename
, int size
)
562 char temp_dir
[MAX_PATH
];
563 /* GetTempFileName requires that its output buffer (4th param)
564 have length MAX_PATH or greater. */
565 assert(size
>= MAX_PATH
);
566 return (GetTempPath(MAX_PATH
, temp_dir
)
567 && GetTempFileName(temp_dir
, "qem", 0, filename
)
568 ? 0 : -GetLastError());
572 tmpdir
= getenv("TMPDIR");
576 if (snprintf(filename
, size
, "%s/vl.XXXXXX", tmpdir
) >= size
) {
579 fd
= mkstemp(filename
);
583 if (close(fd
) != 0) {
592 * Detect host devices. By convention, /dev/cdrom[N] is always
593 * recognized as a host CDROM.
595 static BlockDriver
*find_hdev_driver(const char *filename
)
597 int score_max
= 0, score
;
598 BlockDriver
*drv
= NULL
, *d
;
600 QLIST_FOREACH(d
, &bdrv_drivers
, list
) {
601 if (d
->bdrv_probe_device
) {
602 score
= d
->bdrv_probe_device(filename
);
603 if (score
> score_max
) {
613 static BlockDriver
*bdrv_do_find_protocol(const char *protocol
)
617 QLIST_FOREACH(drv1
, &bdrv_drivers
, list
) {
618 if (drv1
->protocol_name
&& !strcmp(drv1
->protocol_name
, protocol
)) {
626 BlockDriver
*bdrv_find_protocol(const char *filename
,
627 bool allow_protocol_prefix
,
636 /* TODO Drivers without bdrv_file_open must be specified explicitly */
639 * XXX(hch): we really should not let host device detection
640 * override an explicit protocol specification, but moving this
641 * later breaks access to device names with colons in them.
642 * Thanks to the brain-dead persistent naming schemes on udev-
643 * based Linux systems those actually are quite common.
645 drv1
= find_hdev_driver(filename
);
650 if (!path_has_protocol(filename
) || !allow_protocol_prefix
) {
654 p
= strchr(filename
, ':');
657 if (len
> sizeof(protocol
) - 1)
658 len
= sizeof(protocol
) - 1;
659 memcpy(protocol
, filename
, len
);
660 protocol
[len
] = '\0';
662 drv1
= bdrv_do_find_protocol(protocol
);
667 for (i
= 0; i
< (int)ARRAY_SIZE(block_driver_modules
); ++i
) {
668 if (block_driver_modules
[i
].protocol_name
&&
669 !strcmp(block_driver_modules
[i
].protocol_name
, protocol
)) {
670 block_module_load_one(block_driver_modules
[i
].library_name
);
675 drv1
= bdrv_do_find_protocol(protocol
);
677 error_setg(errp
, "Unknown protocol '%s'", protocol
);
683 * Guess image format by probing its contents.
684 * This is not a good idea when your image is raw (CVE-2008-2004), but
685 * we do it anyway for backward compatibility.
687 * @buf contains the image's first @buf_size bytes.
688 * @buf_size is the buffer size in bytes (generally BLOCK_PROBE_BUF_SIZE,
689 * but can be smaller if the image file is smaller)
690 * @filename is its filename.
692 * For all block drivers, call the bdrv_probe() method to get its
694 * Return the first block driver with the highest probing score.
696 BlockDriver
*bdrv_probe_all(const uint8_t *buf
, int buf_size
,
697 const char *filename
)
699 int score_max
= 0, score
;
700 BlockDriver
*drv
= NULL
, *d
;
702 QLIST_FOREACH(d
, &bdrv_drivers
, list
) {
704 score
= d
->bdrv_probe(buf
, buf_size
, filename
);
705 if (score
> score_max
) {
715 static int find_image_format(BlockBackend
*file
, const char *filename
,
716 BlockDriver
**pdrv
, Error
**errp
)
719 uint8_t buf
[BLOCK_PROBE_BUF_SIZE
];
722 /* Return the raw BlockDriver * to scsi-generic devices or empty drives */
723 if (blk_is_sg(file
) || !blk_is_inserted(file
) || blk_getlength(file
) == 0) {
728 ret
= blk_pread(file
, 0, buf
, sizeof(buf
));
730 error_setg_errno(errp
, -ret
, "Could not read image for determining its "
736 drv
= bdrv_probe_all(buf
, ret
, filename
);
738 error_setg(errp
, "Could not determine image format: No compatible "
747 * Set the current 'total_sectors' value
748 * Return 0 on success, -errno on error.
750 int refresh_total_sectors(BlockDriverState
*bs
, int64_t hint
)
752 BlockDriver
*drv
= bs
->drv
;
758 /* Do not attempt drv->bdrv_getlength() on scsi-generic devices */
762 /* query actual device if possible, otherwise just trust the hint */
763 if (drv
->bdrv_getlength
) {
764 int64_t length
= drv
->bdrv_getlength(bs
);
768 hint
= DIV_ROUND_UP(length
, BDRV_SECTOR_SIZE
);
771 bs
->total_sectors
= hint
;
776 * Combines a QDict of new block driver @options with any missing options taken
777 * from @old_options, so that leaving out an option defaults to its old value.
779 static void bdrv_join_options(BlockDriverState
*bs
, QDict
*options
,
782 if (bs
->drv
&& bs
->drv
->bdrv_join_options
) {
783 bs
->drv
->bdrv_join_options(options
, old_options
);
785 qdict_join(options
, old_options
, false);
789 static BlockdevDetectZeroesOptions
bdrv_parse_detect_zeroes(QemuOpts
*opts
,
793 Error
*local_err
= NULL
;
794 char *value
= qemu_opt_get_del(opts
, "detect-zeroes");
795 BlockdevDetectZeroesOptions detect_zeroes
=
796 qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup
, value
,
797 BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF
, &local_err
);
800 error_propagate(errp
, local_err
);
801 return detect_zeroes
;
804 if (detect_zeroes
== BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP
&&
805 !(open_flags
& BDRV_O_UNMAP
))
807 error_setg(errp
, "setting detect-zeroes to unmap is not allowed "
808 "without setting discard operation to unmap");
811 return detect_zeroes
;
815 * Set open flags for a given discard mode
817 * Return 0 on success, -1 if the discard mode was invalid.
819 int bdrv_parse_discard_flags(const char *mode
, int *flags
)
821 *flags
&= ~BDRV_O_UNMAP
;
823 if (!strcmp(mode
, "off") || !strcmp(mode
, "ignore")) {
825 } else if (!strcmp(mode
, "on") || !strcmp(mode
, "unmap")) {
826 *flags
|= BDRV_O_UNMAP
;
835 * Set open flags for a given cache mode
837 * Return 0 on success, -1 if the cache mode was invalid.
839 int bdrv_parse_cache_mode(const char *mode
, int *flags
, bool *writethrough
)
841 *flags
&= ~BDRV_O_CACHE_MASK
;
843 if (!strcmp(mode
, "off") || !strcmp(mode
, "none")) {
844 *writethrough
= false;
845 *flags
|= BDRV_O_NOCACHE
;
846 } else if (!strcmp(mode
, "directsync")) {
847 *writethrough
= true;
848 *flags
|= BDRV_O_NOCACHE
;
849 } else if (!strcmp(mode
, "writeback")) {
850 *writethrough
= false;
851 } else if (!strcmp(mode
, "unsafe")) {
852 *writethrough
= false;
853 *flags
|= BDRV_O_NO_FLUSH
;
854 } else if (!strcmp(mode
, "writethrough")) {
855 *writethrough
= true;
863 static char *bdrv_child_get_parent_desc(BdrvChild
*c
)
865 BlockDriverState
*parent
= c
->opaque
;
866 return g_strdup(bdrv_get_device_or_node_name(parent
));
869 static void bdrv_child_cb_drained_begin(BdrvChild
*child
)
871 BlockDriverState
*bs
= child
->opaque
;
872 bdrv_do_drained_begin_quiesce(bs
, NULL
, false);
875 static bool bdrv_child_cb_drained_poll(BdrvChild
*child
)
877 BlockDriverState
*bs
= child
->opaque
;
878 return bdrv_drain_poll(bs
, false, NULL
, false);
881 static void bdrv_child_cb_drained_end(BdrvChild
*child
)
883 BlockDriverState
*bs
= child
->opaque
;
884 bdrv_drained_end(bs
);
887 static void bdrv_child_cb_attach(BdrvChild
*child
)
889 BlockDriverState
*bs
= child
->opaque
;
890 bdrv_apply_subtree_drain(child
, bs
);
893 static void bdrv_child_cb_detach(BdrvChild
*child
)
895 BlockDriverState
*bs
= child
->opaque
;
896 bdrv_unapply_subtree_drain(child
, bs
);
899 static int bdrv_child_cb_inactivate(BdrvChild
*child
)
901 BlockDriverState
*bs
= child
->opaque
;
902 assert(bs
->open_flags
& BDRV_O_INACTIVE
);
907 * Returns the options and flags that a temporary snapshot should get, based on
908 * the originally requested flags (the originally requested image will have
909 * flags like a backing file)
911 static void bdrv_temp_snapshot_options(int *child_flags
, QDict
*child_options
,
912 int parent_flags
, QDict
*parent_options
)
914 *child_flags
= (parent_flags
& ~BDRV_O_SNAPSHOT
) | BDRV_O_TEMPORARY
;
916 /* For temporary files, unconditional cache=unsafe is fine */
917 qdict_set_default_str(child_options
, BDRV_OPT_CACHE_DIRECT
, "off");
918 qdict_set_default_str(child_options
, BDRV_OPT_CACHE_NO_FLUSH
, "on");
920 /* Copy the read-only option from the parent */
921 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_READ_ONLY
);
923 /* aio=native doesn't work for cache.direct=off, so disable it for the
924 * temporary snapshot */
925 *child_flags
&= ~BDRV_O_NATIVE_AIO
;
929 * Returns the options and flags that bs->file should get if a protocol driver
930 * is expected, based on the given options and flags for the parent BDS
932 static void bdrv_inherited_options(int *child_flags
, QDict
*child_options
,
933 int parent_flags
, QDict
*parent_options
)
935 int flags
= parent_flags
;
937 /* Enable protocol handling, disable format probing for bs->file */
938 flags
|= BDRV_O_PROTOCOL
;
940 /* If the cache mode isn't explicitly set, inherit direct and no-flush from
942 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_CACHE_DIRECT
);
943 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_CACHE_NO_FLUSH
);
944 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_FORCE_SHARE
);
946 /* Inherit the read-only option from the parent if it's not set */
947 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_READ_ONLY
);
948 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_AUTO_READ_ONLY
);
950 /* Our block drivers take care to send flushes and respect unmap policy,
951 * so we can default to enable both on lower layers regardless of the
952 * corresponding parent options. */
953 qdict_set_default_str(child_options
, BDRV_OPT_DISCARD
, "unmap");
955 /* Clear flags that only apply to the top layer */
956 flags
&= ~(BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
| BDRV_O_COPY_ON_READ
|
959 *child_flags
= flags
;
962 const BdrvChildRole child_file
= {
963 .parent_is_bds
= true,
964 .get_parent_desc
= bdrv_child_get_parent_desc
,
965 .inherit_options
= bdrv_inherited_options
,
966 .drained_begin
= bdrv_child_cb_drained_begin
,
967 .drained_poll
= bdrv_child_cb_drained_poll
,
968 .drained_end
= bdrv_child_cb_drained_end
,
969 .attach
= bdrv_child_cb_attach
,
970 .detach
= bdrv_child_cb_detach
,
971 .inactivate
= bdrv_child_cb_inactivate
,
975 * Returns the options and flags that bs->file should get if the use of formats
976 * (and not only protocols) is permitted for it, based on the given options and
977 * flags for the parent BDS
979 static void bdrv_inherited_fmt_options(int *child_flags
, QDict
*child_options
,
980 int parent_flags
, QDict
*parent_options
)
982 child_file
.inherit_options(child_flags
, child_options
,
983 parent_flags
, parent_options
);
985 *child_flags
&= ~(BDRV_O_PROTOCOL
| BDRV_O_NO_IO
);
988 const BdrvChildRole child_format
= {
989 .parent_is_bds
= true,
990 .get_parent_desc
= bdrv_child_get_parent_desc
,
991 .inherit_options
= bdrv_inherited_fmt_options
,
992 .drained_begin
= bdrv_child_cb_drained_begin
,
993 .drained_poll
= bdrv_child_cb_drained_poll
,
994 .drained_end
= bdrv_child_cb_drained_end
,
995 .attach
= bdrv_child_cb_attach
,
996 .detach
= bdrv_child_cb_detach
,
997 .inactivate
= bdrv_child_cb_inactivate
,
1000 static void bdrv_backing_attach(BdrvChild
*c
)
1002 BlockDriverState
*parent
= c
->opaque
;
1003 BlockDriverState
*backing_hd
= c
->bs
;
1005 assert(!parent
->backing_blocker
);
1006 error_setg(&parent
->backing_blocker
,
1007 "node is used as backing hd of '%s'",
1008 bdrv_get_device_or_node_name(parent
));
1010 bdrv_refresh_filename(backing_hd
);
1012 parent
->open_flags
&= ~BDRV_O_NO_BACKING
;
1013 pstrcpy(parent
->backing_file
, sizeof(parent
->backing_file
),
1014 backing_hd
->filename
);
1015 pstrcpy(parent
->backing_format
, sizeof(parent
->backing_format
),
1016 backing_hd
->drv
? backing_hd
->drv
->format_name
: "");
1018 bdrv_op_block_all(backing_hd
, parent
->backing_blocker
);
1019 /* Otherwise we won't be able to commit or stream */
1020 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_COMMIT_TARGET
,
1021 parent
->backing_blocker
);
1022 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_STREAM
,
1023 parent
->backing_blocker
);
1025 * We do backup in 3 ways:
1027 * The target bs is new opened, and the source is top BDS
1028 * 2. blockdev backup
1029 * Both the source and the target are top BDSes.
1030 * 3. internal backup(used for block replication)
1031 * Both the source and the target are backing file
1033 * In case 1 and 2, neither the source nor the target is the backing file.
1034 * In case 3, we will block the top BDS, so there is only one block job
1035 * for the top BDS and its backing chain.
1037 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_BACKUP_SOURCE
,
1038 parent
->backing_blocker
);
1039 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_BACKUP_TARGET
,
1040 parent
->backing_blocker
);
1042 bdrv_child_cb_attach(c
);
1045 static void bdrv_backing_detach(BdrvChild
*c
)
1047 BlockDriverState
*parent
= c
->opaque
;
1049 assert(parent
->backing_blocker
);
1050 bdrv_op_unblock_all(c
->bs
, parent
->backing_blocker
);
1051 error_free(parent
->backing_blocker
);
1052 parent
->backing_blocker
= NULL
;
1054 bdrv_child_cb_detach(c
);
1058 * Returns the options and flags that bs->backing should get, based on the
1059 * given options and flags for the parent BDS
1061 static void bdrv_backing_options(int *child_flags
, QDict
*child_options
,
1062 int parent_flags
, QDict
*parent_options
)
1064 int flags
= parent_flags
;
1066 /* The cache mode is inherited unmodified for backing files; except WCE,
1067 * which is only applied on the top level (BlockBackend) */
1068 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_CACHE_DIRECT
);
1069 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_CACHE_NO_FLUSH
);
1070 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_FORCE_SHARE
);
1072 /* backing files always opened read-only */
1073 qdict_set_default_str(child_options
, BDRV_OPT_READ_ONLY
, "on");
1074 qdict_set_default_str(child_options
, BDRV_OPT_AUTO_READ_ONLY
, "off");
1075 flags
&= ~BDRV_O_COPY_ON_READ
;
1077 /* snapshot=on is handled on the top layer */
1078 flags
&= ~(BDRV_O_SNAPSHOT
| BDRV_O_TEMPORARY
);
1080 *child_flags
= flags
;
1083 static int bdrv_backing_update_filename(BdrvChild
*c
, BlockDriverState
*base
,
1084 const char *filename
, Error
**errp
)
1086 BlockDriverState
*parent
= c
->opaque
;
1087 bool read_only
= bdrv_is_read_only(parent
);
1091 ret
= bdrv_reopen_set_read_only(parent
, false, errp
);
1097 ret
= bdrv_change_backing_file(parent
, filename
,
1098 base
->drv
? base
->drv
->format_name
: "");
1100 error_setg_errno(errp
, -ret
, "Could not update backing file link");
1104 bdrv_reopen_set_read_only(parent
, true, NULL
);
1110 const BdrvChildRole child_backing
= {
1111 .parent_is_bds
= true,
1112 .get_parent_desc
= bdrv_child_get_parent_desc
,
1113 .attach
= bdrv_backing_attach
,
1114 .detach
= bdrv_backing_detach
,
1115 .inherit_options
= bdrv_backing_options
,
1116 .drained_begin
= bdrv_child_cb_drained_begin
,
1117 .drained_poll
= bdrv_child_cb_drained_poll
,
1118 .drained_end
= bdrv_child_cb_drained_end
,
1119 .inactivate
= bdrv_child_cb_inactivate
,
1120 .update_filename
= bdrv_backing_update_filename
,
1123 static int bdrv_open_flags(BlockDriverState
*bs
, int flags
)
1125 int open_flags
= flags
;
1128 * Clear flags that are internal to the block layer before opening the
1131 open_flags
&= ~(BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
| BDRV_O_PROTOCOL
);
1134 * Snapshots should be writable.
1136 if (flags
& BDRV_O_TEMPORARY
) {
1137 open_flags
|= BDRV_O_RDWR
;
1143 static void update_flags_from_options(int *flags
, QemuOpts
*opts
)
1145 *flags
&= ~(BDRV_O_CACHE_MASK
| BDRV_O_RDWR
| BDRV_O_AUTO_RDONLY
);
1147 if (qemu_opt_get_bool_del(opts
, BDRV_OPT_CACHE_NO_FLUSH
, false)) {
1148 *flags
|= BDRV_O_NO_FLUSH
;
1151 if (qemu_opt_get_bool_del(opts
, BDRV_OPT_CACHE_DIRECT
, false)) {
1152 *flags
|= BDRV_O_NOCACHE
;
1155 if (!qemu_opt_get_bool_del(opts
, BDRV_OPT_READ_ONLY
, false)) {
1156 *flags
|= BDRV_O_RDWR
;
1159 if (qemu_opt_get_bool_del(opts
, BDRV_OPT_AUTO_READ_ONLY
, false)) {
1160 *flags
|= BDRV_O_AUTO_RDONLY
;
1164 static void update_options_from_flags(QDict
*options
, int flags
)
1166 if (!qdict_haskey(options
, BDRV_OPT_CACHE_DIRECT
)) {
1167 qdict_put_bool(options
, BDRV_OPT_CACHE_DIRECT
, flags
& BDRV_O_NOCACHE
);
1169 if (!qdict_haskey(options
, BDRV_OPT_CACHE_NO_FLUSH
)) {
1170 qdict_put_bool(options
, BDRV_OPT_CACHE_NO_FLUSH
,
1171 flags
& BDRV_O_NO_FLUSH
);
1173 if (!qdict_haskey(options
, BDRV_OPT_READ_ONLY
)) {
1174 qdict_put_bool(options
, BDRV_OPT_READ_ONLY
, !(flags
& BDRV_O_RDWR
));
1176 if (!qdict_haskey(options
, BDRV_OPT_AUTO_READ_ONLY
)) {
1177 qdict_put_bool(options
, BDRV_OPT_AUTO_READ_ONLY
,
1178 flags
& BDRV_O_AUTO_RDONLY
);
1182 static void bdrv_assign_node_name(BlockDriverState
*bs
,
1183 const char *node_name
,
1186 char *gen_node_name
= NULL
;
1189 node_name
= gen_node_name
= id_generate(ID_BLOCK
);
1190 } else if (!id_wellformed(node_name
)) {
1192 * Check for empty string or invalid characters, but not if it is
1193 * generated (generated names use characters not available to the user)
1195 error_setg(errp
, "Invalid node name");
1199 /* takes care of avoiding namespaces collisions */
1200 if (blk_by_name(node_name
)) {
1201 error_setg(errp
, "node-name=%s is conflicting with a device id",
1206 /* takes care of avoiding duplicates node names */
1207 if (bdrv_find_node(node_name
)) {
1208 error_setg(errp
, "Duplicate node name");
1212 /* Make sure that the node name isn't truncated */
1213 if (strlen(node_name
) >= sizeof(bs
->node_name
)) {
1214 error_setg(errp
, "Node name too long");
1218 /* copy node name into the bs and insert it into the graph list */
1219 pstrcpy(bs
->node_name
, sizeof(bs
->node_name
), node_name
);
1220 QTAILQ_INSERT_TAIL(&graph_bdrv_states
, bs
, node_list
);
1222 g_free(gen_node_name
);
1225 static int bdrv_open_driver(BlockDriverState
*bs
, BlockDriver
*drv
,
1226 const char *node_name
, QDict
*options
,
1227 int open_flags
, Error
**errp
)
1229 Error
*local_err
= NULL
;
1232 bdrv_assign_node_name(bs
, node_name
, &local_err
);
1234 error_propagate(errp
, local_err
);
1239 bs
->read_only
= !(bs
->open_flags
& BDRV_O_RDWR
);
1240 bs
->opaque
= g_malloc0(drv
->instance_size
);
1242 if (drv
->bdrv_file_open
) {
1243 assert(!drv
->bdrv_needs_filename
|| bs
->filename
[0]);
1244 ret
= drv
->bdrv_file_open(bs
, options
, open_flags
, &local_err
);
1245 } else if (drv
->bdrv_open
) {
1246 ret
= drv
->bdrv_open(bs
, options
, open_flags
, &local_err
);
1253 error_propagate(errp
, local_err
);
1254 } else if (bs
->filename
[0]) {
1255 error_setg_errno(errp
, -ret
, "Could not open '%s'", bs
->filename
);
1257 error_setg_errno(errp
, -ret
, "Could not open image");
1262 ret
= refresh_total_sectors(bs
, bs
->total_sectors
);
1264 error_setg_errno(errp
, -ret
, "Could not refresh total sector count");
1268 bdrv_refresh_limits(bs
, &local_err
);
1270 error_propagate(errp
, local_err
);
1274 assert(bdrv_opt_mem_align(bs
) != 0);
1275 assert(bdrv_min_mem_align(bs
) != 0);
1276 assert(is_power_of_2(bs
->bl
.request_alignment
));
1278 for (i
= 0; i
< bs
->quiesce_counter
; i
++) {
1279 if (drv
->bdrv_co_drain_begin
) {
1280 drv
->bdrv_co_drain_begin(bs
);
1287 if (bs
->file
!= NULL
) {
1288 bdrv_unref_child(bs
, bs
->file
);
1296 BlockDriverState
*bdrv_new_open_driver(BlockDriver
*drv
, const char *node_name
,
1297 int flags
, Error
**errp
)
1299 BlockDriverState
*bs
;
1303 bs
->open_flags
= flags
;
1304 bs
->explicit_options
= qdict_new();
1305 bs
->options
= qdict_new();
1308 update_options_from_flags(bs
->options
, flags
);
1310 ret
= bdrv_open_driver(bs
, drv
, node_name
, bs
->options
, flags
, errp
);
1312 qobject_unref(bs
->explicit_options
);
1313 bs
->explicit_options
= NULL
;
1314 qobject_unref(bs
->options
);
1323 QemuOptsList bdrv_runtime_opts
= {
1324 .name
= "bdrv_common",
1325 .head
= QTAILQ_HEAD_INITIALIZER(bdrv_runtime_opts
.head
),
1328 .name
= "node-name",
1329 .type
= QEMU_OPT_STRING
,
1330 .help
= "Node name of the block device node",
1334 .type
= QEMU_OPT_STRING
,
1335 .help
= "Block driver to use for the node",
1338 .name
= BDRV_OPT_CACHE_DIRECT
,
1339 .type
= QEMU_OPT_BOOL
,
1340 .help
= "Bypass software writeback cache on the host",
1343 .name
= BDRV_OPT_CACHE_NO_FLUSH
,
1344 .type
= QEMU_OPT_BOOL
,
1345 .help
= "Ignore flush requests",
1348 .name
= BDRV_OPT_READ_ONLY
,
1349 .type
= QEMU_OPT_BOOL
,
1350 .help
= "Node is opened in read-only mode",
1353 .name
= BDRV_OPT_AUTO_READ_ONLY
,
1354 .type
= QEMU_OPT_BOOL
,
1355 .help
= "Node can become read-only if opening read-write fails",
1358 .name
= "detect-zeroes",
1359 .type
= QEMU_OPT_STRING
,
1360 .help
= "try to optimize zero writes (off, on, unmap)",
1363 .name
= BDRV_OPT_DISCARD
,
1364 .type
= QEMU_OPT_STRING
,
1365 .help
= "discard operation (ignore/off, unmap/on)",
1368 .name
= BDRV_OPT_FORCE_SHARE
,
1369 .type
= QEMU_OPT_BOOL
,
1370 .help
= "always accept other writers (default: off)",
1372 { /* end of list */ }
1377 * Common part for opening disk images and files
1379 * Removes all processed options from *options.
1381 static int bdrv_open_common(BlockDriverState
*bs
, BlockBackend
*file
,
1382 QDict
*options
, Error
**errp
)
1384 int ret
, open_flags
;
1385 const char *filename
;
1386 const char *driver_name
= NULL
;
1387 const char *node_name
= NULL
;
1388 const char *discard
;
1391 Error
*local_err
= NULL
;
1393 assert(bs
->file
== NULL
);
1394 assert(options
!= NULL
&& bs
->options
!= options
);
1396 opts
= qemu_opts_create(&bdrv_runtime_opts
, NULL
, 0, &error_abort
);
1397 qemu_opts_absorb_qdict(opts
, options
, &local_err
);
1399 error_propagate(errp
, local_err
);
1404 update_flags_from_options(&bs
->open_flags
, opts
);
1406 driver_name
= qemu_opt_get(opts
, "driver");
1407 drv
= bdrv_find_format(driver_name
);
1408 assert(drv
!= NULL
);
1410 bs
->force_share
= qemu_opt_get_bool(opts
, BDRV_OPT_FORCE_SHARE
, false);
1412 if (bs
->force_share
&& (bs
->open_flags
& BDRV_O_RDWR
)) {
1414 BDRV_OPT_FORCE_SHARE
1415 "=on can only be used with read-only images");
1421 bdrv_refresh_filename(blk_bs(file
));
1422 filename
= blk_bs(file
)->filename
;
1425 * Caution: while qdict_get_try_str() is fine, getting
1426 * non-string types would require more care. When @options
1427 * come from -blockdev or blockdev_add, its members are typed
1428 * according to the QAPI schema, but when they come from
1429 * -drive, they're all QString.
1431 filename
= qdict_get_try_str(options
, "filename");
1434 if (drv
->bdrv_needs_filename
&& (!filename
|| !filename
[0])) {
1435 error_setg(errp
, "The '%s' block driver requires a file name",
1441 trace_bdrv_open_common(bs
, filename
?: "", bs
->open_flags
,
1444 bs
->read_only
= !(bs
->open_flags
& BDRV_O_RDWR
);
1446 if (use_bdrv_whitelist
&& !bdrv_is_whitelisted(drv
, bs
->read_only
)) {
1447 if (!bs
->read_only
&& bdrv_is_whitelisted(drv
, true)) {
1448 ret
= bdrv_apply_auto_read_only(bs
, NULL
, NULL
);
1454 !bs
->read_only
&& bdrv_is_whitelisted(drv
, true)
1455 ? "Driver '%s' can only be used for read-only devices"
1456 : "Driver '%s' is not whitelisted",
1462 /* bdrv_new() and bdrv_close() make it so */
1463 assert(atomic_read(&bs
->copy_on_read
) == 0);
1465 if (bs
->open_flags
& BDRV_O_COPY_ON_READ
) {
1466 if (!bs
->read_only
) {
1467 bdrv_enable_copy_on_read(bs
);
1469 error_setg(errp
, "Can't use copy-on-read on read-only device");
1475 discard
= qemu_opt_get(opts
, BDRV_OPT_DISCARD
);
1476 if (discard
!= NULL
) {
1477 if (bdrv_parse_discard_flags(discard
, &bs
->open_flags
) != 0) {
1478 error_setg(errp
, "Invalid discard option");
1485 bdrv_parse_detect_zeroes(opts
, bs
->open_flags
, &local_err
);
1487 error_propagate(errp
, local_err
);
1492 if (filename
!= NULL
) {
1493 pstrcpy(bs
->filename
, sizeof(bs
->filename
), filename
);
1495 bs
->filename
[0] = '\0';
1497 pstrcpy(bs
->exact_filename
, sizeof(bs
->exact_filename
), bs
->filename
);
1499 /* Open the image, either directly or using a protocol */
1500 open_flags
= bdrv_open_flags(bs
, bs
->open_flags
);
1501 node_name
= qemu_opt_get(opts
, "node-name");
1503 assert(!drv
->bdrv_file_open
|| file
== NULL
);
1504 ret
= bdrv_open_driver(bs
, drv
, node_name
, options
, open_flags
, errp
);
1509 qemu_opts_del(opts
);
1513 qemu_opts_del(opts
);
1517 static QDict
*parse_json_filename(const char *filename
, Error
**errp
)
1519 QObject
*options_obj
;
1523 ret
= strstart(filename
, "json:", &filename
);
1526 options_obj
= qobject_from_json(filename
, errp
);
1528 error_prepend(errp
, "Could not parse the JSON options: ");
1532 options
= qobject_to(QDict
, options_obj
);
1534 qobject_unref(options_obj
);
1535 error_setg(errp
, "Invalid JSON object given");
1539 qdict_flatten(options
);
1544 static void parse_json_protocol(QDict
*options
, const char **pfilename
,
1547 QDict
*json_options
;
1548 Error
*local_err
= NULL
;
1550 /* Parse json: pseudo-protocol */
1551 if (!*pfilename
|| !g_str_has_prefix(*pfilename
, "json:")) {
1555 json_options
= parse_json_filename(*pfilename
, &local_err
);
1557 error_propagate(errp
, local_err
);
1561 /* Options given in the filename have lower priority than options
1562 * specified directly */
1563 qdict_join(options
, json_options
, false);
1564 qobject_unref(json_options
);
1569 * Fills in default options for opening images and converts the legacy
1570 * filename/flags pair to option QDict entries.
1571 * The BDRV_O_PROTOCOL flag in *flags will be set or cleared accordingly if a
1572 * block driver has been specified explicitly.
1574 static int bdrv_fill_options(QDict
**options
, const char *filename
,
1575 int *flags
, Error
**errp
)
1577 const char *drvname
;
1578 bool protocol
= *flags
& BDRV_O_PROTOCOL
;
1579 bool parse_filename
= false;
1580 BlockDriver
*drv
= NULL
;
1581 Error
*local_err
= NULL
;
1584 * Caution: while qdict_get_try_str() is fine, getting non-string
1585 * types would require more care. When @options come from
1586 * -blockdev or blockdev_add, its members are typed according to
1587 * the QAPI schema, but when they come from -drive, they're all
1590 drvname
= qdict_get_try_str(*options
, "driver");
1592 drv
= bdrv_find_format(drvname
);
1594 error_setg(errp
, "Unknown driver '%s'", drvname
);
1597 /* If the user has explicitly specified the driver, this choice should
1598 * override the BDRV_O_PROTOCOL flag */
1599 protocol
= drv
->bdrv_file_open
;
1603 *flags
|= BDRV_O_PROTOCOL
;
1605 *flags
&= ~BDRV_O_PROTOCOL
;
1608 /* Translate cache options from flags into options */
1609 update_options_from_flags(*options
, *flags
);
1611 /* Fetch the file name from the options QDict if necessary */
1612 if (protocol
&& filename
) {
1613 if (!qdict_haskey(*options
, "filename")) {
1614 qdict_put_str(*options
, "filename", filename
);
1615 parse_filename
= true;
1617 error_setg(errp
, "Can't specify 'file' and 'filename' options at "
1623 /* Find the right block driver */
1624 /* See cautionary note on accessing @options above */
1625 filename
= qdict_get_try_str(*options
, "filename");
1627 if (!drvname
&& protocol
) {
1629 drv
= bdrv_find_protocol(filename
, parse_filename
, errp
);
1634 drvname
= drv
->format_name
;
1635 qdict_put_str(*options
, "driver", drvname
);
1637 error_setg(errp
, "Must specify either driver or file");
1642 assert(drv
|| !protocol
);
1644 /* Driver-specific filename parsing */
1645 if (drv
&& drv
->bdrv_parse_filename
&& parse_filename
) {
1646 drv
->bdrv_parse_filename(filename
, *options
, &local_err
);
1648 error_propagate(errp
, local_err
);
1652 if (!drv
->bdrv_needs_filename
) {
1653 qdict_del(*options
, "filename");
1660 static int bdrv_child_check_perm(BdrvChild
*c
, BlockReopenQueue
*q
,
1661 uint64_t perm
, uint64_t shared
,
1662 GSList
*ignore_children
, Error
**errp
);
1663 static void bdrv_child_abort_perm_update(BdrvChild
*c
);
1664 static void bdrv_child_set_perm(BdrvChild
*c
, uint64_t perm
, uint64_t shared
);
1666 typedef struct BlockReopenQueueEntry
{
1668 BDRVReopenState state
;
1669 QSIMPLEQ_ENTRY(BlockReopenQueueEntry
) entry
;
1670 } BlockReopenQueueEntry
;
1673 * Return the flags that @bs will have after the reopens in @q have
1674 * successfully completed. If @q is NULL (or @bs is not contained in @q),
1675 * return the current flags.
1677 static int bdrv_reopen_get_flags(BlockReopenQueue
*q
, BlockDriverState
*bs
)
1679 BlockReopenQueueEntry
*entry
;
1682 QSIMPLEQ_FOREACH(entry
, q
, entry
) {
1683 if (entry
->state
.bs
== bs
) {
1684 return entry
->state
.flags
;
1689 return bs
->open_flags
;
1692 /* Returns whether the image file can be written to after the reopen queue @q
1693 * has been successfully applied, or right now if @q is NULL. */
1694 static bool bdrv_is_writable_after_reopen(BlockDriverState
*bs
,
1695 BlockReopenQueue
*q
)
1697 int flags
= bdrv_reopen_get_flags(q
, bs
);
1699 return (flags
& (BDRV_O_RDWR
| BDRV_O_INACTIVE
)) == BDRV_O_RDWR
;
1703 * Return whether the BDS can be written to. This is not necessarily
1704 * the same as !bdrv_is_read_only(bs), as inactivated images may not
1705 * be written to but do not count as read-only images.
1707 bool bdrv_is_writable(BlockDriverState
*bs
)
1709 return bdrv_is_writable_after_reopen(bs
, NULL
);
1712 static void bdrv_child_perm(BlockDriverState
*bs
, BlockDriverState
*child_bs
,
1713 BdrvChild
*c
, const BdrvChildRole
*role
,
1714 BlockReopenQueue
*reopen_queue
,
1715 uint64_t parent_perm
, uint64_t parent_shared
,
1716 uint64_t *nperm
, uint64_t *nshared
)
1718 if (bs
->drv
&& bs
->drv
->bdrv_child_perm
) {
1719 bs
->drv
->bdrv_child_perm(bs
, c
, role
, reopen_queue
,
1720 parent_perm
, parent_shared
,
1723 /* TODO Take force_share from reopen_queue */
1724 if (child_bs
&& child_bs
->force_share
) {
1725 *nshared
= BLK_PERM_ALL
;
1730 * Check whether permissions on this node can be changed in a way that
1731 * @cumulative_perms and @cumulative_shared_perms are the new cumulative
1732 * permissions of all its parents. This involves checking whether all necessary
1733 * permission changes to child nodes can be performed.
1735 * A call to this function must always be followed by a call to bdrv_set_perm()
1736 * or bdrv_abort_perm_update().
1738 static int bdrv_check_perm(BlockDriverState
*bs
, BlockReopenQueue
*q
,
1739 uint64_t cumulative_perms
,
1740 uint64_t cumulative_shared_perms
,
1741 GSList
*ignore_children
, Error
**errp
)
1743 BlockDriver
*drv
= bs
->drv
;
1747 /* Write permissions never work with read-only images */
1748 if ((cumulative_perms
& (BLK_PERM_WRITE
| BLK_PERM_WRITE_UNCHANGED
)) &&
1749 !bdrv_is_writable_after_reopen(bs
, q
))
1751 error_setg(errp
, "Block node is read-only");
1755 /* Check this node */
1760 if (drv
->bdrv_check_perm
) {
1761 return drv
->bdrv_check_perm(bs
, cumulative_perms
,
1762 cumulative_shared_perms
, errp
);
1765 /* Drivers that never have children can omit .bdrv_child_perm() */
1766 if (!drv
->bdrv_child_perm
) {
1767 assert(QLIST_EMPTY(&bs
->children
));
1771 /* Check all children */
1772 QLIST_FOREACH(c
, &bs
->children
, next
) {
1773 uint64_t cur_perm
, cur_shared
;
1774 bdrv_child_perm(bs
, c
->bs
, c
, c
->role
, q
,
1775 cumulative_perms
, cumulative_shared_perms
,
1776 &cur_perm
, &cur_shared
);
1777 ret
= bdrv_child_check_perm(c
, q
, cur_perm
, cur_shared
,
1778 ignore_children
, errp
);
1788 * Notifies drivers that after a previous bdrv_check_perm() call, the
1789 * permission update is not performed and any preparations made for it (e.g.
1790 * taken file locks) need to be undone.
1792 * This function recursively notifies all child nodes.
1794 static void bdrv_abort_perm_update(BlockDriverState
*bs
)
1796 BlockDriver
*drv
= bs
->drv
;
1803 if (drv
->bdrv_abort_perm_update
) {
1804 drv
->bdrv_abort_perm_update(bs
);
1807 QLIST_FOREACH(c
, &bs
->children
, next
) {
1808 bdrv_child_abort_perm_update(c
);
1812 static void bdrv_set_perm(BlockDriverState
*bs
, uint64_t cumulative_perms
,
1813 uint64_t cumulative_shared_perms
)
1815 BlockDriver
*drv
= bs
->drv
;
1822 /* Update this node */
1823 if (drv
->bdrv_set_perm
) {
1824 drv
->bdrv_set_perm(bs
, cumulative_perms
, cumulative_shared_perms
);
1827 /* Drivers that never have children can omit .bdrv_child_perm() */
1828 if (!drv
->bdrv_child_perm
) {
1829 assert(QLIST_EMPTY(&bs
->children
));
1833 /* Update all children */
1834 QLIST_FOREACH(c
, &bs
->children
, next
) {
1835 uint64_t cur_perm
, cur_shared
;
1836 bdrv_child_perm(bs
, c
->bs
, c
, c
->role
, NULL
,
1837 cumulative_perms
, cumulative_shared_perms
,
1838 &cur_perm
, &cur_shared
);
1839 bdrv_child_set_perm(c
, cur_perm
, cur_shared
);
1843 static void bdrv_get_cumulative_perm(BlockDriverState
*bs
, uint64_t *perm
,
1844 uint64_t *shared_perm
)
1847 uint64_t cumulative_perms
= 0;
1848 uint64_t cumulative_shared_perms
= BLK_PERM_ALL
;
1850 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
1851 cumulative_perms
|= c
->perm
;
1852 cumulative_shared_perms
&= c
->shared_perm
;
1855 *perm
= cumulative_perms
;
1856 *shared_perm
= cumulative_shared_perms
;
1859 static char *bdrv_child_user_desc(BdrvChild
*c
)
1861 if (c
->role
->get_parent_desc
) {
1862 return c
->role
->get_parent_desc(c
);
1865 return g_strdup("another user");
1868 char *bdrv_perm_names(uint64_t perm
)
1874 { BLK_PERM_CONSISTENT_READ
, "consistent read" },
1875 { BLK_PERM_WRITE
, "write" },
1876 { BLK_PERM_WRITE_UNCHANGED
, "write unchanged" },
1877 { BLK_PERM_RESIZE
, "resize" },
1878 { BLK_PERM_GRAPH_MOD
, "change children" },
1882 char *result
= g_strdup("");
1883 struct perm_name
*p
;
1885 for (p
= permissions
; p
->name
; p
++) {
1886 if (perm
& p
->perm
) {
1888 result
= g_strdup_printf("%s%s%s", old
, *old
? ", " : "", p
->name
);
1897 * Checks whether a new reference to @bs can be added if the new user requires
1898 * @new_used_perm/@new_shared_perm as its permissions. If @ignore_children is
1899 * set, the BdrvChild objects in this list are ignored in the calculations;
1900 * this allows checking permission updates for an existing reference.
1902 * Needs to be followed by a call to either bdrv_set_perm() or
1903 * bdrv_abort_perm_update(). */
1904 static int bdrv_check_update_perm(BlockDriverState
*bs
, BlockReopenQueue
*q
,
1905 uint64_t new_used_perm
,
1906 uint64_t new_shared_perm
,
1907 GSList
*ignore_children
, Error
**errp
)
1910 uint64_t cumulative_perms
= new_used_perm
;
1911 uint64_t cumulative_shared_perms
= new_shared_perm
;
1913 /* There is no reason why anyone couldn't tolerate write_unchanged */
1914 assert(new_shared_perm
& BLK_PERM_WRITE_UNCHANGED
);
1916 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
1917 if (g_slist_find(ignore_children
, c
)) {
1921 if ((new_used_perm
& c
->shared_perm
) != new_used_perm
) {
1922 char *user
= bdrv_child_user_desc(c
);
1923 char *perm_names
= bdrv_perm_names(new_used_perm
& ~c
->shared_perm
);
1924 error_setg(errp
, "Conflicts with use by %s as '%s', which does not "
1926 user
, c
->name
, perm_names
, bdrv_get_node_name(c
->bs
));
1932 if ((c
->perm
& new_shared_perm
) != c
->perm
) {
1933 char *user
= bdrv_child_user_desc(c
);
1934 char *perm_names
= bdrv_perm_names(c
->perm
& ~new_shared_perm
);
1935 error_setg(errp
, "Conflicts with use by %s as '%s', which uses "
1937 user
, c
->name
, perm_names
, bdrv_get_node_name(c
->bs
));
1943 cumulative_perms
|= c
->perm
;
1944 cumulative_shared_perms
&= c
->shared_perm
;
1947 return bdrv_check_perm(bs
, q
, cumulative_perms
, cumulative_shared_perms
,
1948 ignore_children
, errp
);
1951 /* Needs to be followed by a call to either bdrv_child_set_perm() or
1952 * bdrv_child_abort_perm_update(). */
1953 static int bdrv_child_check_perm(BdrvChild
*c
, BlockReopenQueue
*q
,
1954 uint64_t perm
, uint64_t shared
,
1955 GSList
*ignore_children
, Error
**errp
)
1959 ignore_children
= g_slist_prepend(g_slist_copy(ignore_children
), c
);
1960 ret
= bdrv_check_update_perm(c
->bs
, q
, perm
, shared
, ignore_children
, errp
);
1961 g_slist_free(ignore_children
);
1967 if (!c
->has_backup_perm
) {
1968 c
->has_backup_perm
= true;
1969 c
->backup_perm
= c
->perm
;
1970 c
->backup_shared_perm
= c
->shared_perm
;
1973 * Note: it's OK if c->has_backup_perm was already set, as we can find the
1974 * same child twice during check_perm procedure
1978 c
->shared_perm
= shared
;
1983 static void bdrv_child_set_perm(BdrvChild
*c
, uint64_t perm
, uint64_t shared
)
1985 uint64_t cumulative_perms
, cumulative_shared_perms
;
1987 c
->has_backup_perm
= false;
1990 c
->shared_perm
= shared
;
1992 bdrv_get_cumulative_perm(c
->bs
, &cumulative_perms
,
1993 &cumulative_shared_perms
);
1994 bdrv_set_perm(c
->bs
, cumulative_perms
, cumulative_shared_perms
);
1997 static void bdrv_child_abort_perm_update(BdrvChild
*c
)
1999 if (c
->has_backup_perm
) {
2000 c
->perm
= c
->backup_perm
;
2001 c
->shared_perm
= c
->backup_shared_perm
;
2002 c
->has_backup_perm
= false;
2005 bdrv_abort_perm_update(c
->bs
);
2008 int bdrv_child_try_set_perm(BdrvChild
*c
, uint64_t perm
, uint64_t shared
,
2013 ret
= bdrv_child_check_perm(c
, NULL
, perm
, shared
, NULL
, errp
);
2015 bdrv_child_abort_perm_update(c
);
2019 bdrv_child_set_perm(c
, perm
, shared
);
2024 void bdrv_filter_default_perms(BlockDriverState
*bs
, BdrvChild
*c
,
2025 const BdrvChildRole
*role
,
2026 BlockReopenQueue
*reopen_queue
,
2027 uint64_t perm
, uint64_t shared
,
2028 uint64_t *nperm
, uint64_t *nshared
)
2031 *nperm
= perm
& DEFAULT_PERM_PASSTHROUGH
;
2032 *nshared
= (shared
& DEFAULT_PERM_PASSTHROUGH
) | DEFAULT_PERM_UNCHANGED
;
2036 *nperm
= (perm
& DEFAULT_PERM_PASSTHROUGH
) |
2037 (c
->perm
& DEFAULT_PERM_UNCHANGED
);
2038 *nshared
= (shared
& DEFAULT_PERM_PASSTHROUGH
) |
2039 (c
->shared_perm
& DEFAULT_PERM_UNCHANGED
);
2042 void bdrv_format_default_perms(BlockDriverState
*bs
, BdrvChild
*c
,
2043 const BdrvChildRole
*role
,
2044 BlockReopenQueue
*reopen_queue
,
2045 uint64_t perm
, uint64_t shared
,
2046 uint64_t *nperm
, uint64_t *nshared
)
2048 bool backing
= (role
== &child_backing
);
2049 assert(role
== &child_backing
|| role
== &child_file
);
2052 int flags
= bdrv_reopen_get_flags(reopen_queue
, bs
);
2054 /* Apart from the modifications below, the same permissions are
2055 * forwarded and left alone as for filters */
2056 bdrv_filter_default_perms(bs
, c
, role
, reopen_queue
, perm
, shared
,
2059 /* Format drivers may touch metadata even if the guest doesn't write */
2060 if (bdrv_is_writable_after_reopen(bs
, reopen_queue
)) {
2061 perm
|= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2064 /* bs->file always needs to be consistent because of the metadata. We
2065 * can never allow other users to resize or write to it. */
2066 if (!(flags
& BDRV_O_NO_IO
)) {
2067 perm
|= BLK_PERM_CONSISTENT_READ
;
2069 shared
&= ~(BLK_PERM_WRITE
| BLK_PERM_RESIZE
);
2071 /* We want consistent read from backing files if the parent needs it.
2072 * No other operations are performed on backing files. */
2073 perm
&= BLK_PERM_CONSISTENT_READ
;
2075 /* If the parent can deal with changing data, we're okay with a
2076 * writable and resizable backing file. */
2077 /* TODO Require !(perm & BLK_PERM_CONSISTENT_READ), too? */
2078 if (shared
& BLK_PERM_WRITE
) {
2079 shared
= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2084 shared
|= BLK_PERM_CONSISTENT_READ
| BLK_PERM_GRAPH_MOD
|
2085 BLK_PERM_WRITE_UNCHANGED
;
2088 if (bs
->open_flags
& BDRV_O_INACTIVE
) {
2089 shared
|= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2096 static void bdrv_replace_child_noperm(BdrvChild
*child
,
2097 BlockDriverState
*new_bs
)
2099 BlockDriverState
*old_bs
= child
->bs
;
2102 if (old_bs
&& new_bs
) {
2103 assert(bdrv_get_aio_context(old_bs
) == bdrv_get_aio_context(new_bs
));
2106 /* Detach first so that the recursive drain sections coming from @child
2107 * are already gone and we only end the drain sections that came from
2109 if (child
->role
->detach
) {
2110 child
->role
->detach(child
);
2112 if (old_bs
->quiesce_counter
&& child
->role
->drained_end
) {
2113 int num
= old_bs
->quiesce_counter
;
2114 if (child
->role
->parent_is_bds
) {
2115 num
-= bdrv_drain_all_count
;
2118 for (i
= 0; i
< num
; i
++) {
2119 child
->role
->drained_end(child
);
2122 QLIST_REMOVE(child
, next_parent
);
2128 QLIST_INSERT_HEAD(&new_bs
->parents
, child
, next_parent
);
2129 if (new_bs
->quiesce_counter
&& child
->role
->drained_begin
) {
2130 int num
= new_bs
->quiesce_counter
;
2131 if (child
->role
->parent_is_bds
) {
2132 num
-= bdrv_drain_all_count
;
2135 for (i
= 0; i
< num
; i
++) {
2136 bdrv_parent_drained_begin_single(child
, true);
2140 /* Attach only after starting new drained sections, so that recursive
2141 * drain sections coming from @child don't get an extra .drained_begin
2143 if (child
->role
->attach
) {
2144 child
->role
->attach(child
);
2150 * Updates @child to change its reference to point to @new_bs, including
2151 * checking and applying the necessary permisson updates both to the old node
2154 * NULL is passed as @new_bs for removing the reference before freeing @child.
2156 * If @new_bs is not NULL, bdrv_check_perm() must be called beforehand, as this
2157 * function uses bdrv_set_perm() to update the permissions according to the new
2158 * reference that @new_bs gets.
2160 static void bdrv_replace_child(BdrvChild
*child
, BlockDriverState
*new_bs
)
2162 BlockDriverState
*old_bs
= child
->bs
;
2163 uint64_t perm
, shared_perm
;
2165 bdrv_replace_child_noperm(child
, new_bs
);
2168 /* Update permissions for old node. This is guaranteed to succeed
2169 * because we're just taking a parent away, so we're loosening
2171 bdrv_get_cumulative_perm(old_bs
, &perm
, &shared_perm
);
2172 bdrv_check_perm(old_bs
, NULL
, perm
, shared_perm
, NULL
, &error_abort
);
2173 bdrv_set_perm(old_bs
, perm
, shared_perm
);
2177 bdrv_get_cumulative_perm(new_bs
, &perm
, &shared_perm
);
2178 bdrv_set_perm(new_bs
, perm
, shared_perm
);
2182 BdrvChild
*bdrv_root_attach_child(BlockDriverState
*child_bs
,
2183 const char *child_name
,
2184 const BdrvChildRole
*child_role
,
2185 uint64_t perm
, uint64_t shared_perm
,
2186 void *opaque
, Error
**errp
)
2191 ret
= bdrv_check_update_perm(child_bs
, NULL
, perm
, shared_perm
, NULL
, errp
);
2193 bdrv_abort_perm_update(child_bs
);
2197 child
= g_new(BdrvChild
, 1);
2198 *child
= (BdrvChild
) {
2200 .name
= g_strdup(child_name
),
2203 .shared_perm
= shared_perm
,
2207 /* This performs the matching bdrv_set_perm() for the above check. */
2208 bdrv_replace_child(child
, child_bs
);
2213 BdrvChild
*bdrv_attach_child(BlockDriverState
*parent_bs
,
2214 BlockDriverState
*child_bs
,
2215 const char *child_name
,
2216 const BdrvChildRole
*child_role
,
2220 uint64_t perm
, shared_perm
;
2222 bdrv_get_cumulative_perm(parent_bs
, &perm
, &shared_perm
);
2224 assert(parent_bs
->drv
);
2225 assert(bdrv_get_aio_context(parent_bs
) == bdrv_get_aio_context(child_bs
));
2226 bdrv_child_perm(parent_bs
, child_bs
, NULL
, child_role
, NULL
,
2227 perm
, shared_perm
, &perm
, &shared_perm
);
2229 child
= bdrv_root_attach_child(child_bs
, child_name
, child_role
,
2230 perm
, shared_perm
, parent_bs
, errp
);
2231 if (child
== NULL
) {
2235 QLIST_INSERT_HEAD(&parent_bs
->children
, child
, next
);
2239 static void bdrv_detach_child(BdrvChild
*child
)
2241 if (child
->next
.le_prev
) {
2242 QLIST_REMOVE(child
, next
);
2243 child
->next
.le_prev
= NULL
;
2246 bdrv_replace_child(child
, NULL
);
2248 g_free(child
->name
);
2252 void bdrv_root_unref_child(BdrvChild
*child
)
2254 BlockDriverState
*child_bs
;
2256 child_bs
= child
->bs
;
2257 bdrv_detach_child(child
);
2258 bdrv_unref(child_bs
);
2261 void bdrv_unref_child(BlockDriverState
*parent
, BdrvChild
*child
)
2263 if (child
== NULL
) {
2267 if (child
->bs
->inherits_from
== parent
) {
2270 /* Remove inherits_from only when the last reference between parent and
2271 * child->bs goes away. */
2272 QLIST_FOREACH(c
, &parent
->children
, next
) {
2273 if (c
!= child
&& c
->bs
== child
->bs
) {
2278 child
->bs
->inherits_from
= NULL
;
2282 bdrv_root_unref_child(child
);
2286 static void bdrv_parent_cb_change_media(BlockDriverState
*bs
, bool load
)
2289 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
2290 if (c
->role
->change_media
) {
2291 c
->role
->change_media(c
, load
);
2296 /* Return true if you can reach parent going through child->inherits_from
2297 * recursively. If parent or child are NULL, return false */
2298 static bool bdrv_inherits_from_recursive(BlockDriverState
*child
,
2299 BlockDriverState
*parent
)
2301 while (child
&& child
!= parent
) {
2302 child
= child
->inherits_from
;
2305 return child
!= NULL
;
2309 * Sets the backing file link of a BDS. A new reference is created; callers
2310 * which don't need their own reference any more must call bdrv_unref().
2312 void bdrv_set_backing_hd(BlockDriverState
*bs
, BlockDriverState
*backing_hd
,
2315 bool update_inherits_from
= bdrv_chain_contains(bs
, backing_hd
) &&
2316 bdrv_inherits_from_recursive(backing_hd
, bs
);
2319 bdrv_ref(backing_hd
);
2323 bdrv_unref_child(bs
, bs
->backing
);
2331 bs
->backing
= bdrv_attach_child(bs
, backing_hd
, "backing", &child_backing
,
2333 /* If backing_hd was already part of bs's backing chain, and
2334 * inherits_from pointed recursively to bs then let's update it to
2335 * point directly to bs (else it will become NULL). */
2336 if (update_inherits_from
) {
2337 backing_hd
->inherits_from
= bs
;
2340 bdrv_unref(backing_hd
);
2344 bdrv_refresh_limits(bs
, NULL
);
2348 * Opens the backing file for a BlockDriverState if not yet open
2350 * bdref_key specifies the key for the image's BlockdevRef in the options QDict.
2351 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
2352 * itself, all options starting with "${bdref_key}." are considered part of the
2355 * TODO Can this be unified with bdrv_open_image()?
2357 int bdrv_open_backing_file(BlockDriverState
*bs
, QDict
*parent_options
,
2358 const char *bdref_key
, Error
**errp
)
2360 char *backing_filename
= g_malloc0(PATH_MAX
);
2361 char *bdref_key_dot
;
2362 const char *reference
= NULL
;
2364 bool implicit_backing
= false;
2365 BlockDriverState
*backing_hd
;
2367 QDict
*tmp_parent_options
= NULL
;
2368 Error
*local_err
= NULL
;
2370 if (bs
->backing
!= NULL
) {
2374 /* NULL means an empty set of options */
2375 if (parent_options
== NULL
) {
2376 tmp_parent_options
= qdict_new();
2377 parent_options
= tmp_parent_options
;
2380 bs
->open_flags
&= ~BDRV_O_NO_BACKING
;
2382 bdref_key_dot
= g_strdup_printf("%s.", bdref_key
);
2383 qdict_extract_subqdict(parent_options
, &options
, bdref_key_dot
);
2384 g_free(bdref_key_dot
);
2387 * Caution: while qdict_get_try_str() is fine, getting non-string
2388 * types would require more care. When @parent_options come from
2389 * -blockdev or blockdev_add, its members are typed according to
2390 * the QAPI schema, but when they come from -drive, they're all
2393 reference
= qdict_get_try_str(parent_options
, bdref_key
);
2394 if (reference
|| qdict_haskey(options
, "file.filename")) {
2395 backing_filename
[0] = '\0';
2396 } else if (bs
->backing_file
[0] == '\0' && qdict_size(options
) == 0) {
2397 qobject_unref(options
);
2400 if (qdict_size(options
) == 0) {
2401 /* If the user specifies options that do not modify the
2402 * backing file's behavior, we might still consider it the
2403 * implicit backing file. But it's easier this way, and
2404 * just specifying some of the backing BDS's options is
2405 * only possible with -drive anyway (otherwise the QAPI
2406 * schema forces the user to specify everything). */
2407 implicit_backing
= !strcmp(bs
->auto_backing_file
, bs
->backing_file
);
2410 bdrv_get_full_backing_filename(bs
, backing_filename
, PATH_MAX
,
2414 error_propagate(errp
, local_err
);
2415 qobject_unref(options
);
2420 if (!bs
->drv
|| !bs
->drv
->supports_backing
) {
2422 error_setg(errp
, "Driver doesn't support backing files");
2423 qobject_unref(options
);
2428 bs
->backing_format
[0] != '\0' && !qdict_haskey(options
, "driver")) {
2429 qdict_put_str(options
, "driver", bs
->backing_format
);
2432 backing_hd
= bdrv_open_inherit(*backing_filename
? backing_filename
: NULL
,
2433 reference
, options
, 0, bs
, &child_backing
,
2436 bs
->open_flags
|= BDRV_O_NO_BACKING
;
2437 error_prepend(errp
, "Could not open backing file: ");
2441 bdrv_set_aio_context(backing_hd
, bdrv_get_aio_context(bs
));
2443 if (implicit_backing
) {
2444 bdrv_refresh_filename(backing_hd
);
2445 pstrcpy(bs
->auto_backing_file
, sizeof(bs
->auto_backing_file
),
2446 backing_hd
->filename
);
2449 /* Hook up the backing file link; drop our reference, bs owns the
2450 * backing_hd reference now */
2451 bdrv_set_backing_hd(bs
, backing_hd
, &local_err
);
2452 bdrv_unref(backing_hd
);
2454 error_propagate(errp
, local_err
);
2459 qdict_del(parent_options
, bdref_key
);
2462 g_free(backing_filename
);
2463 qobject_unref(tmp_parent_options
);
2467 static BlockDriverState
*
2468 bdrv_open_child_bs(const char *filename
, QDict
*options
, const char *bdref_key
,
2469 BlockDriverState
*parent
, const BdrvChildRole
*child_role
,
2470 bool allow_none
, Error
**errp
)
2472 BlockDriverState
*bs
= NULL
;
2473 QDict
*image_options
;
2474 char *bdref_key_dot
;
2475 const char *reference
;
2477 assert(child_role
!= NULL
);
2479 bdref_key_dot
= g_strdup_printf("%s.", bdref_key
);
2480 qdict_extract_subqdict(options
, &image_options
, bdref_key_dot
);
2481 g_free(bdref_key_dot
);
2484 * Caution: while qdict_get_try_str() is fine, getting non-string
2485 * types would require more care. When @options come from
2486 * -blockdev or blockdev_add, its members are typed according to
2487 * the QAPI schema, but when they come from -drive, they're all
2490 reference
= qdict_get_try_str(options
, bdref_key
);
2491 if (!filename
&& !reference
&& !qdict_size(image_options
)) {
2493 error_setg(errp
, "A block device must be specified for \"%s\"",
2496 qobject_unref(image_options
);
2500 bs
= bdrv_open_inherit(filename
, reference
, image_options
, 0,
2501 parent
, child_role
, errp
);
2507 qdict_del(options
, bdref_key
);
2512 * Opens a disk image whose options are given as BlockdevRef in another block
2515 * If allow_none is true, no image will be opened if filename is false and no
2516 * BlockdevRef is given. NULL will be returned, but errp remains unset.
2518 * bdrev_key specifies the key for the image's BlockdevRef in the options QDict.
2519 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
2520 * itself, all options starting with "${bdref_key}." are considered part of the
2523 * The BlockdevRef will be removed from the options QDict.
2525 BdrvChild
*bdrv_open_child(const char *filename
,
2526 QDict
*options
, const char *bdref_key
,
2527 BlockDriverState
*parent
,
2528 const BdrvChildRole
*child_role
,
2529 bool allow_none
, Error
**errp
)
2532 BlockDriverState
*bs
;
2534 bs
= bdrv_open_child_bs(filename
, options
, bdref_key
, parent
, child_role
,
2540 c
= bdrv_attach_child(parent
, bs
, bdref_key
, child_role
, errp
);
2549 /* TODO Future callers may need to specify parent/child_role in order for
2550 * option inheritance to work. Existing callers use it for the root node. */
2551 BlockDriverState
*bdrv_open_blockdev_ref(BlockdevRef
*ref
, Error
**errp
)
2553 BlockDriverState
*bs
= NULL
;
2554 Error
*local_err
= NULL
;
2555 QObject
*obj
= NULL
;
2556 QDict
*qdict
= NULL
;
2557 const char *reference
= NULL
;
2560 if (ref
->type
== QTYPE_QSTRING
) {
2561 reference
= ref
->u
.reference
;
2563 BlockdevOptions
*options
= &ref
->u
.definition
;
2564 assert(ref
->type
== QTYPE_QDICT
);
2566 v
= qobject_output_visitor_new(&obj
);
2567 visit_type_BlockdevOptions(v
, NULL
, &options
, &local_err
);
2569 error_propagate(errp
, local_err
);
2572 visit_complete(v
, &obj
);
2574 qdict
= qobject_to(QDict
, obj
);
2575 qdict_flatten(qdict
);
2577 /* bdrv_open_inherit() defaults to the values in bdrv_flags (for
2578 * compatibility with other callers) rather than what we want as the
2579 * real defaults. Apply the defaults here instead. */
2580 qdict_set_default_str(qdict
, BDRV_OPT_CACHE_DIRECT
, "off");
2581 qdict_set_default_str(qdict
, BDRV_OPT_CACHE_NO_FLUSH
, "off");
2582 qdict_set_default_str(qdict
, BDRV_OPT_READ_ONLY
, "off");
2583 qdict_set_default_str(qdict
, BDRV_OPT_AUTO_READ_ONLY
, "off");
2587 bs
= bdrv_open_inherit(NULL
, reference
, qdict
, 0, NULL
, NULL
, errp
);
2596 static BlockDriverState
*bdrv_append_temp_snapshot(BlockDriverState
*bs
,
2598 QDict
*snapshot_options
,
2601 /* TODO: extra byte is a hack to ensure MAX_PATH space on Windows. */
2602 char *tmp_filename
= g_malloc0(PATH_MAX
+ 1);
2604 QemuOpts
*opts
= NULL
;
2605 BlockDriverState
*bs_snapshot
= NULL
;
2606 Error
*local_err
= NULL
;
2609 /* if snapshot, we create a temporary backing file and open it
2610 instead of opening 'filename' directly */
2612 /* Get the required size from the image */
2613 total_size
= bdrv_getlength(bs
);
2614 if (total_size
< 0) {
2615 error_setg_errno(errp
, -total_size
, "Could not get image size");
2619 /* Create the temporary image */
2620 ret
= get_tmp_filename(tmp_filename
, PATH_MAX
+ 1);
2622 error_setg_errno(errp
, -ret
, "Could not get temporary filename");
2626 opts
= qemu_opts_create(bdrv_qcow2
.create_opts
, NULL
, 0,
2628 qemu_opt_set_number(opts
, BLOCK_OPT_SIZE
, total_size
, &error_abort
);
2629 ret
= bdrv_create(&bdrv_qcow2
, tmp_filename
, opts
, errp
);
2630 qemu_opts_del(opts
);
2632 error_prepend(errp
, "Could not create temporary overlay '%s': ",
2637 /* Prepare options QDict for the temporary file */
2638 qdict_put_str(snapshot_options
, "file.driver", "file");
2639 qdict_put_str(snapshot_options
, "file.filename", tmp_filename
);
2640 qdict_put_str(snapshot_options
, "driver", "qcow2");
2642 bs_snapshot
= bdrv_open(NULL
, NULL
, snapshot_options
, flags
, errp
);
2643 snapshot_options
= NULL
;
2648 /* bdrv_append() consumes a strong reference to bs_snapshot
2649 * (i.e. it will call bdrv_unref() on it) even on error, so in
2650 * order to be able to return one, we have to increase
2651 * bs_snapshot's refcount here */
2652 bdrv_ref(bs_snapshot
);
2653 bdrv_append(bs_snapshot
, bs
, &local_err
);
2655 error_propagate(errp
, local_err
);
2661 qobject_unref(snapshot_options
);
2662 g_free(tmp_filename
);
2667 * Opens a disk image (raw, qcow2, vmdk, ...)
2669 * options is a QDict of options to pass to the block drivers, or NULL for an
2670 * empty set of options. The reference to the QDict belongs to the block layer
2671 * after the call (even on failure), so if the caller intends to reuse the
2672 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
2674 * If *pbs is NULL, a new BDS will be created with a pointer to it stored there.
2675 * If it is not NULL, the referenced BDS will be reused.
2677 * The reference parameter may be used to specify an existing block device which
2678 * should be opened. If specified, neither options nor a filename may be given,
2679 * nor can an existing BDS be reused (that is, *pbs has to be NULL).
2681 static BlockDriverState
*bdrv_open_inherit(const char *filename
,
2682 const char *reference
,
2683 QDict
*options
, int flags
,
2684 BlockDriverState
*parent
,
2685 const BdrvChildRole
*child_role
,
2689 BlockBackend
*file
= NULL
;
2690 BlockDriverState
*bs
;
2691 BlockDriver
*drv
= NULL
;
2693 const char *drvname
;
2694 const char *backing
;
2695 Error
*local_err
= NULL
;
2696 QDict
*snapshot_options
= NULL
;
2697 int snapshot_flags
= 0;
2699 assert(!child_role
|| !flags
);
2700 assert(!child_role
== !parent
);
2703 bool options_non_empty
= options
? qdict_size(options
) : false;
2704 qobject_unref(options
);
2706 if (filename
|| options_non_empty
) {
2707 error_setg(errp
, "Cannot reference an existing block device with "
2708 "additional options or a new filename");
2712 bs
= bdrv_lookup_bs(reference
, reference
, errp
);
2723 /* NULL means an empty set of options */
2724 if (options
== NULL
) {
2725 options
= qdict_new();
2728 /* json: syntax counts as explicit options, as if in the QDict */
2729 parse_json_protocol(options
, &filename
, &local_err
);
2734 bs
->explicit_options
= qdict_clone_shallow(options
);
2737 bs
->inherits_from
= parent
;
2738 child_role
->inherit_options(&flags
, options
,
2739 parent
->open_flags
, parent
->options
);
2742 ret
= bdrv_fill_options(&options
, filename
, &flags
, &local_err
);
2748 * Set the BDRV_O_RDWR and BDRV_O_ALLOW_RDWR flags.
2749 * Caution: getting a boolean member of @options requires care.
2750 * When @options come from -blockdev or blockdev_add, members are
2751 * typed according to the QAPI schema, but when they come from
2752 * -drive, they're all QString.
2754 if (g_strcmp0(qdict_get_try_str(options
, BDRV_OPT_READ_ONLY
), "on") &&
2755 !qdict_get_try_bool(options
, BDRV_OPT_READ_ONLY
, false)) {
2756 flags
|= (BDRV_O_RDWR
| BDRV_O_ALLOW_RDWR
);
2758 flags
&= ~BDRV_O_RDWR
;
2761 if (flags
& BDRV_O_SNAPSHOT
) {
2762 snapshot_options
= qdict_new();
2763 bdrv_temp_snapshot_options(&snapshot_flags
, snapshot_options
,
2765 /* Let bdrv_backing_options() override "read-only" */
2766 qdict_del(options
, BDRV_OPT_READ_ONLY
);
2767 bdrv_backing_options(&flags
, options
, flags
, options
);
2770 bs
->open_flags
= flags
;
2771 bs
->options
= options
;
2772 options
= qdict_clone_shallow(options
);
2774 /* Find the right image format driver */
2775 /* See cautionary note on accessing @options above */
2776 drvname
= qdict_get_try_str(options
, "driver");
2778 drv
= bdrv_find_format(drvname
);
2780 error_setg(errp
, "Unknown driver: '%s'", drvname
);
2785 assert(drvname
|| !(flags
& BDRV_O_PROTOCOL
));
2787 /* See cautionary note on accessing @options above */
2788 backing
= qdict_get_try_str(options
, "backing");
2789 if (qobject_to(QNull
, qdict_get(options
, "backing")) != NULL
||
2790 (backing
&& *backing
== '\0'))
2793 warn_report("Use of \"backing\": \"\" is deprecated; "
2794 "use \"backing\": null instead");
2796 flags
|= BDRV_O_NO_BACKING
;
2797 qdict_del(options
, "backing");
2800 /* Open image file without format layer. This BlockBackend is only used for
2801 * probing, the block drivers will do their own bdrv_open_child() for the
2802 * same BDS, which is why we put the node name back into options. */
2803 if ((flags
& BDRV_O_PROTOCOL
) == 0) {
2804 BlockDriverState
*file_bs
;
2806 file_bs
= bdrv_open_child_bs(filename
, options
, "file", bs
,
2807 &child_file
, true, &local_err
);
2811 if (file_bs
!= NULL
) {
2812 /* Not requesting BLK_PERM_CONSISTENT_READ because we're only
2813 * looking at the header to guess the image format. This works even
2814 * in cases where a guest would not see a consistent state. */
2815 file
= blk_new(0, BLK_PERM_ALL
);
2816 blk_insert_bs(file
, file_bs
, &local_err
);
2817 bdrv_unref(file_bs
);
2822 qdict_put_str(options
, "file", bdrv_get_node_name(file_bs
));
2826 /* Image format probing */
2829 ret
= find_image_format(file
, filename
, &drv
, &local_err
);
2834 * This option update would logically belong in bdrv_fill_options(),
2835 * but we first need to open bs->file for the probing to work, while
2836 * opening bs->file already requires the (mostly) final set of options
2837 * so that cache mode etc. can be inherited.
2839 * Adding the driver later is somewhat ugly, but it's not an option
2840 * that would ever be inherited, so it's correct. We just need to make
2841 * sure to update both bs->options (which has the full effective
2842 * options for bs) and options (which has file.* already removed).
2844 qdict_put_str(bs
->options
, "driver", drv
->format_name
);
2845 qdict_put_str(options
, "driver", drv
->format_name
);
2847 error_setg(errp
, "Must specify either driver or file");
2851 /* BDRV_O_PROTOCOL must be set iff a protocol BDS is about to be created */
2852 assert(!!(flags
& BDRV_O_PROTOCOL
) == !!drv
->bdrv_file_open
);
2853 /* file must be NULL if a protocol BDS is about to be created
2854 * (the inverse results in an error message from bdrv_open_common()) */
2855 assert(!(flags
& BDRV_O_PROTOCOL
) || !file
);
2857 /* Open the image */
2858 ret
= bdrv_open_common(bs
, file
, options
, &local_err
);
2868 /* If there is a backing file, use it */
2869 if ((flags
& BDRV_O_NO_BACKING
) == 0) {
2870 ret
= bdrv_open_backing_file(bs
, options
, "backing", &local_err
);
2872 goto close_and_fail
;
2876 /* Remove all children options and references
2877 * from bs->options and bs->explicit_options */
2878 QLIST_FOREACH(child
, &bs
->children
, next
) {
2879 char *child_key_dot
;
2880 child_key_dot
= g_strdup_printf("%s.", child
->name
);
2881 qdict_extract_subqdict(bs
->explicit_options
, NULL
, child_key_dot
);
2882 qdict_extract_subqdict(bs
->options
, NULL
, child_key_dot
);
2883 qdict_del(bs
->explicit_options
, child
->name
);
2884 qdict_del(bs
->options
, child
->name
);
2885 g_free(child_key_dot
);
2888 /* Check if any unknown options were used */
2889 if (qdict_size(options
) != 0) {
2890 const QDictEntry
*entry
= qdict_first(options
);
2891 if (flags
& BDRV_O_PROTOCOL
) {
2892 error_setg(errp
, "Block protocol '%s' doesn't support the option "
2893 "'%s'", drv
->format_name
, entry
->key
);
2896 "Block format '%s' does not support the option '%s'",
2897 drv
->format_name
, entry
->key
);
2900 goto close_and_fail
;
2903 bdrv_parent_cb_change_media(bs
, true);
2905 qobject_unref(options
);
2908 /* For snapshot=on, create a temporary qcow2 overlay. bs points to the
2909 * temporary snapshot afterwards. */
2910 if (snapshot_flags
) {
2911 BlockDriverState
*snapshot_bs
;
2912 snapshot_bs
= bdrv_append_temp_snapshot(bs
, snapshot_flags
,
2913 snapshot_options
, &local_err
);
2914 snapshot_options
= NULL
;
2916 goto close_and_fail
;
2918 /* We are not going to return bs but the overlay on top of it
2919 * (snapshot_bs); thus, we have to drop the strong reference to bs
2920 * (which we obtained by calling bdrv_new()). bs will not be deleted,
2921 * though, because the overlay still has a reference to it. */
2930 qobject_unref(snapshot_options
);
2931 qobject_unref(bs
->explicit_options
);
2932 qobject_unref(bs
->options
);
2933 qobject_unref(options
);
2935 bs
->explicit_options
= NULL
;
2937 error_propagate(errp
, local_err
);
2942 qobject_unref(snapshot_options
);
2943 qobject_unref(options
);
2944 error_propagate(errp
, local_err
);
2948 BlockDriverState
*bdrv_open(const char *filename
, const char *reference
,
2949 QDict
*options
, int flags
, Error
**errp
)
2951 return bdrv_open_inherit(filename
, reference
, options
, flags
, NULL
,
2956 * Adds a BlockDriverState to a simple queue for an atomic, transactional
2957 * reopen of multiple devices.
2959 * bs_queue can either be an existing BlockReopenQueue that has had QSIMPLE_INIT
2960 * already performed, or alternatively may be NULL a new BlockReopenQueue will
2961 * be created and initialized. This newly created BlockReopenQueue should be
2962 * passed back in for subsequent calls that are intended to be of the same
2965 * bs is the BlockDriverState to add to the reopen queue.
2967 * options contains the changed options for the associated bs
2968 * (the BlockReopenQueue takes ownership)
2970 * flags contains the open flags for the associated bs
2972 * returns a pointer to bs_queue, which is either the newly allocated
2973 * bs_queue, or the existing bs_queue being used.
2975 * bs must be drained between bdrv_reopen_queue() and bdrv_reopen_multiple().
2977 static BlockReopenQueue
*bdrv_reopen_queue_child(BlockReopenQueue
*bs_queue
,
2978 BlockDriverState
*bs
,
2980 const BdrvChildRole
*role
,
2981 QDict
*parent_options
,
2986 BlockReopenQueueEntry
*bs_entry
;
2988 QDict
*old_options
, *explicit_options
, *options_copy
;
2992 /* Make sure that the caller remembered to use a drained section. This is
2993 * important to avoid graph changes between the recursive queuing here and
2994 * bdrv_reopen_multiple(). */
2995 assert(bs
->quiesce_counter
> 0);
2997 if (bs_queue
== NULL
) {
2998 bs_queue
= g_new0(BlockReopenQueue
, 1);
2999 QSIMPLEQ_INIT(bs_queue
);
3003 options
= qdict_new();
3006 /* Check if this BlockDriverState is already in the queue */
3007 QSIMPLEQ_FOREACH(bs_entry
, bs_queue
, entry
) {
3008 if (bs
== bs_entry
->state
.bs
) {
3014 * Precedence of options:
3015 * 1. Explicitly passed in options (highest)
3016 * 2. Retained from explicitly set options of bs
3017 * 3. Inherited from parent node
3018 * 4. Retained from effective options of bs
3021 /* Old explicitly set values (don't overwrite by inherited value) */
3023 old_options
= qdict_clone_shallow(bs_entry
->state
.explicit_options
);
3025 old_options
= qdict_clone_shallow(bs
->explicit_options
);
3027 bdrv_join_options(bs
, options
, old_options
);
3028 qobject_unref(old_options
);
3030 explicit_options
= qdict_clone_shallow(options
);
3032 /* Inherit from parent node */
3033 if (parent_options
) {
3035 role
->inherit_options(&flags
, options
, parent_flags
, parent_options
);
3037 flags
= bdrv_get_flags(bs
);
3040 /* Old values are used for options that aren't set yet */
3041 old_options
= qdict_clone_shallow(bs
->options
);
3042 bdrv_join_options(bs
, options
, old_options
);
3043 qobject_unref(old_options
);
3045 /* We have the final set of options so let's update the flags */
3046 options_copy
= qdict_clone_shallow(options
);
3047 opts
= qemu_opts_create(&bdrv_runtime_opts
, NULL
, 0, &error_abort
);
3048 qemu_opts_absorb_qdict(opts
, options_copy
, NULL
);
3049 update_flags_from_options(&flags
, opts
);
3050 qemu_opts_del(opts
);
3051 qobject_unref(options_copy
);
3053 /* bdrv_open_inherit() sets and clears some additional flags internally */
3054 flags
&= ~BDRV_O_PROTOCOL
;
3055 if (flags
& BDRV_O_RDWR
) {
3056 flags
|= BDRV_O_ALLOW_RDWR
;
3060 bs_entry
= g_new0(BlockReopenQueueEntry
, 1);
3061 QSIMPLEQ_INSERT_TAIL(bs_queue
, bs_entry
, entry
);
3063 qobject_unref(bs_entry
->state
.options
);
3064 qobject_unref(bs_entry
->state
.explicit_options
);
3067 bs_entry
->state
.bs
= bs
;
3068 bs_entry
->state
.options
= options
;
3069 bs_entry
->state
.explicit_options
= explicit_options
;
3070 bs_entry
->state
.flags
= flags
;
3072 /* This needs to be overwritten in bdrv_reopen_prepare() */
3073 bs_entry
->state
.perm
= UINT64_MAX
;
3074 bs_entry
->state
.shared_perm
= 0;
3076 QLIST_FOREACH(child
, &bs
->children
, next
) {
3077 QDict
*new_child_options
;
3078 char *child_key_dot
;
3080 /* reopen can only change the options of block devices that were
3081 * implicitly created and inherited options. For other (referenced)
3082 * block devices, a syntax like "backing.foo" results in an error. */
3083 if (child
->bs
->inherits_from
!= bs
) {
3087 child_key_dot
= g_strdup_printf("%s.", child
->name
);
3088 qdict_extract_subqdict(explicit_options
, NULL
, child_key_dot
);
3089 qdict_extract_subqdict(options
, &new_child_options
, child_key_dot
);
3090 g_free(child_key_dot
);
3092 bdrv_reopen_queue_child(bs_queue
, child
->bs
, new_child_options
,
3093 child
->role
, options
, flags
);
3099 BlockReopenQueue
*bdrv_reopen_queue(BlockReopenQueue
*bs_queue
,
3100 BlockDriverState
*bs
,
3103 return bdrv_reopen_queue_child(bs_queue
, bs
, options
, NULL
, NULL
, 0);
3107 * Reopen multiple BlockDriverStates atomically & transactionally.
3109 * The queue passed in (bs_queue) must have been built up previous
3110 * via bdrv_reopen_queue().
3112 * Reopens all BDS specified in the queue, with the appropriate
3113 * flags. All devices are prepared for reopen, and failure of any
3114 * device will cause all device changes to be abandoned, and intermediate
3117 * If all devices prepare successfully, then the changes are committed
3120 * All affected nodes must be drained between bdrv_reopen_queue() and
3121 * bdrv_reopen_multiple().
3123 int bdrv_reopen_multiple(AioContext
*ctx
, BlockReopenQueue
*bs_queue
, Error
**errp
)
3126 BlockReopenQueueEntry
*bs_entry
, *next
;
3127 Error
*local_err
= NULL
;
3129 assert(bs_queue
!= NULL
);
3131 QSIMPLEQ_FOREACH(bs_entry
, bs_queue
, entry
) {
3132 assert(bs_entry
->state
.bs
->quiesce_counter
> 0);
3133 if (bdrv_reopen_prepare(&bs_entry
->state
, bs_queue
, &local_err
)) {
3134 error_propagate(errp
, local_err
);
3137 bs_entry
->prepared
= true;
3140 /* If we reach this point, we have success and just need to apply the
3143 QSIMPLEQ_FOREACH(bs_entry
, bs_queue
, entry
) {
3144 bdrv_reopen_commit(&bs_entry
->state
);
3150 QSIMPLEQ_FOREACH_SAFE(bs_entry
, bs_queue
, entry
, next
) {
3152 if (bs_entry
->prepared
) {
3153 bdrv_reopen_abort(&bs_entry
->state
);
3155 qobject_unref(bs_entry
->state
.explicit_options
);
3156 qobject_unref(bs_entry
->state
.options
);
3165 int bdrv_reopen_set_read_only(BlockDriverState
*bs
, bool read_only
,
3169 BlockReopenQueue
*queue
;
3170 QDict
*opts
= qdict_new();
3172 qdict_put_bool(opts
, BDRV_OPT_READ_ONLY
, read_only
);
3174 bdrv_subtree_drained_begin(bs
);
3175 queue
= bdrv_reopen_queue(NULL
, bs
, opts
);
3176 ret
= bdrv_reopen_multiple(bdrv_get_aio_context(bs
), queue
, errp
);
3177 bdrv_subtree_drained_end(bs
);
3182 static BlockReopenQueueEntry
*find_parent_in_reopen_queue(BlockReopenQueue
*q
,
3185 BlockReopenQueueEntry
*entry
;
3187 QSIMPLEQ_FOREACH(entry
, q
, entry
) {
3188 BlockDriverState
*bs
= entry
->state
.bs
;
3191 QLIST_FOREACH(child
, &bs
->children
, next
) {
3201 static void bdrv_reopen_perm(BlockReopenQueue
*q
, BlockDriverState
*bs
,
3202 uint64_t *perm
, uint64_t *shared
)
3205 BlockReopenQueueEntry
*parent
;
3206 uint64_t cumulative_perms
= 0;
3207 uint64_t cumulative_shared_perms
= BLK_PERM_ALL
;
3209 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
3210 parent
= find_parent_in_reopen_queue(q
, c
);
3212 cumulative_perms
|= c
->perm
;
3213 cumulative_shared_perms
&= c
->shared_perm
;
3215 uint64_t nperm
, nshared
;
3217 bdrv_child_perm(parent
->state
.bs
, bs
, c
, c
->role
, q
,
3218 parent
->state
.perm
, parent
->state
.shared_perm
,
3221 cumulative_perms
|= nperm
;
3222 cumulative_shared_perms
&= nshared
;
3225 *perm
= cumulative_perms
;
3226 *shared
= cumulative_shared_perms
;
3230 * Prepares a BlockDriverState for reopen. All changes are staged in the
3231 * 'opaque' field of the BDRVReopenState, which is used and allocated by
3232 * the block driver layer .bdrv_reopen_prepare()
3234 * bs is the BlockDriverState to reopen
3235 * flags are the new open flags
3236 * queue is the reopen queue
3238 * Returns 0 on success, non-zero on error. On error errp will be set
3241 * On failure, bdrv_reopen_abort() will be called to clean up any data.
3242 * It is the responsibility of the caller to then call the abort() or
3243 * commit() for any other BDS that have been left in a prepare() state
3246 int bdrv_reopen_prepare(BDRVReopenState
*reopen_state
, BlockReopenQueue
*queue
,
3251 Error
*local_err
= NULL
;
3254 QDict
*orig_reopen_opts
;
3255 char *discard
= NULL
;
3257 bool drv_prepared
= false;
3259 assert(reopen_state
!= NULL
);
3260 assert(reopen_state
->bs
->drv
!= NULL
);
3261 drv
= reopen_state
->bs
->drv
;
3263 /* This function and each driver's bdrv_reopen_prepare() remove
3264 * entries from reopen_state->options as they are processed, so
3265 * we need to make a copy of the original QDict. */
3266 orig_reopen_opts
= qdict_clone_shallow(reopen_state
->options
);
3268 /* Process generic block layer options */
3269 opts
= qemu_opts_create(&bdrv_runtime_opts
, NULL
, 0, &error_abort
);
3270 qemu_opts_absorb_qdict(opts
, reopen_state
->options
, &local_err
);
3272 error_propagate(errp
, local_err
);
3277 /* This was already called in bdrv_reopen_queue_child() so the flags
3278 * are up-to-date. This time we simply want to remove the options from
3279 * QemuOpts in order to indicate that they have been processed. */
3280 old_flags
= reopen_state
->flags
;
3281 update_flags_from_options(&reopen_state
->flags
, opts
);
3282 assert(old_flags
== reopen_state
->flags
);
3284 discard
= qemu_opt_get_del(opts
, BDRV_OPT_DISCARD
);
3285 if (discard
!= NULL
) {
3286 if (bdrv_parse_discard_flags(discard
, &reopen_state
->flags
) != 0) {
3287 error_setg(errp
, "Invalid discard option");
3293 reopen_state
->detect_zeroes
=
3294 bdrv_parse_detect_zeroes(opts
, reopen_state
->flags
, &local_err
);
3296 error_propagate(errp
, local_err
);
3301 /* All other options (including node-name and driver) must be unchanged.
3302 * Put them back into the QDict, so that they are checked at the end
3303 * of this function. */
3304 qemu_opts_to_qdict(opts
, reopen_state
->options
);
3306 /* If we are to stay read-only, do not allow permission change
3307 * to r/w. Attempting to set to r/w may fail if either BDRV_O_ALLOW_RDWR is
3308 * not set, or if the BDS still has copy_on_read enabled */
3309 read_only
= !(reopen_state
->flags
& BDRV_O_RDWR
);
3310 ret
= bdrv_can_set_read_only(reopen_state
->bs
, read_only
, true, &local_err
);
3312 error_propagate(errp
, local_err
);
3316 /* Calculate required permissions after reopening */
3317 bdrv_reopen_perm(queue
, reopen_state
->bs
,
3318 &reopen_state
->perm
, &reopen_state
->shared_perm
);
3320 ret
= bdrv_flush(reopen_state
->bs
);
3322 error_setg_errno(errp
, -ret
, "Error flushing drive");
3326 if (drv
->bdrv_reopen_prepare
) {
3327 ret
= drv
->bdrv_reopen_prepare(reopen_state
, queue
, &local_err
);
3329 if (local_err
!= NULL
) {
3330 error_propagate(errp
, local_err
);
3332 bdrv_refresh_filename(reopen_state
->bs
);
3333 error_setg(errp
, "failed while preparing to reopen image '%s'",
3334 reopen_state
->bs
->filename
);
3339 /* It is currently mandatory to have a bdrv_reopen_prepare()
3340 * handler for each supported drv. */
3341 error_setg(errp
, "Block format '%s' used by node '%s' "
3342 "does not support reopening files", drv
->format_name
,
3343 bdrv_get_device_or_node_name(reopen_state
->bs
));
3348 drv_prepared
= true;
3350 /* Options that are not handled are only okay if they are unchanged
3351 * compared to the old state. It is expected that some options are only
3352 * used for the initial open, but not reopen (e.g. filename) */
3353 if (qdict_size(reopen_state
->options
)) {
3354 const QDictEntry
*entry
= qdict_first(reopen_state
->options
);
3357 QObject
*new = entry
->value
;
3358 QObject
*old
= qdict_get(reopen_state
->bs
->options
, entry
->key
);
3360 /* Allow child references (child_name=node_name) as long as they
3361 * point to the current child (i.e. everything stays the same). */
3362 if (qobject_type(new) == QTYPE_QSTRING
) {
3364 QLIST_FOREACH(child
, &reopen_state
->bs
->children
, next
) {
3365 if (!strcmp(child
->name
, entry
->key
)) {
3371 const char *str
= qobject_get_try_str(new);
3372 if (!strcmp(child
->bs
->node_name
, str
)) {
3373 continue; /* Found child with this name, skip option */
3379 * TODO: When using -drive to specify blockdev options, all values
3380 * will be strings; however, when using -blockdev, blockdev-add or
3381 * filenames using the json:{} pseudo-protocol, they will be
3383 * In contrast, reopening options are (currently) always strings
3384 * (because you can only specify them through qemu-io; all other
3385 * callers do not specify any options).
3386 * Therefore, when using anything other than -drive to create a BDS,
3387 * this cannot detect non-string options as unchanged, because
3388 * qobject_is_equal() always returns false for objects of different
3389 * type. In the future, this should be remedied by correctly typing
3390 * all options. For now, this is not too big of an issue because
3391 * the user can simply omit options which cannot be changed anyway,
3392 * so they will stay unchanged.
3394 if (!qobject_is_equal(new, old
)) {
3395 error_setg(errp
, "Cannot change the option '%s'", entry
->key
);
3399 } while ((entry
= qdict_next(reopen_state
->options
, entry
)));
3402 ret
= bdrv_check_perm(reopen_state
->bs
, queue
, reopen_state
->perm
,
3403 reopen_state
->shared_perm
, NULL
, errp
);
3410 /* Restore the original reopen_state->options QDict */
3411 qobject_unref(reopen_state
->options
);
3412 reopen_state
->options
= qobject_ref(orig_reopen_opts
);
3415 if (ret
< 0 && drv_prepared
) {
3416 /* drv->bdrv_reopen_prepare() has succeeded, so we need to
3417 * call drv->bdrv_reopen_abort() before signaling an error
3418 * (bdrv_reopen_multiple() will not call bdrv_reopen_abort()
3419 * when the respective bdrv_reopen_prepare() has failed) */
3420 if (drv
->bdrv_reopen_abort
) {
3421 drv
->bdrv_reopen_abort(reopen_state
);
3424 qemu_opts_del(opts
);
3425 qobject_unref(orig_reopen_opts
);
3431 * Takes the staged changes for the reopen from bdrv_reopen_prepare(), and
3432 * makes them final by swapping the staging BlockDriverState contents into
3433 * the active BlockDriverState contents.
3435 void bdrv_reopen_commit(BDRVReopenState
*reopen_state
)
3438 BlockDriverState
*bs
;
3440 bool old_can_write
, new_can_write
;
3442 assert(reopen_state
!= NULL
);
3443 bs
= reopen_state
->bs
;
3445 assert(drv
!= NULL
);
3448 !bdrv_is_read_only(bs
) && !(bdrv_get_flags(bs
) & BDRV_O_INACTIVE
);
3450 /* If there are any driver level actions to take */
3451 if (drv
->bdrv_reopen_commit
) {
3452 drv
->bdrv_reopen_commit(reopen_state
);
3455 /* set BDS specific flags now */
3456 qobject_unref(bs
->explicit_options
);
3457 qobject_unref(bs
->options
);
3459 bs
->explicit_options
= reopen_state
->explicit_options
;
3460 bs
->options
= reopen_state
->options
;
3461 bs
->open_flags
= reopen_state
->flags
;
3462 bs
->read_only
= !(reopen_state
->flags
& BDRV_O_RDWR
);
3463 bs
->detect_zeroes
= reopen_state
->detect_zeroes
;
3465 /* Remove child references from bs->options and bs->explicit_options.
3466 * Child options were already removed in bdrv_reopen_queue_child() */
3467 QLIST_FOREACH(child
, &bs
->children
, next
) {
3468 qdict_del(bs
->explicit_options
, child
->name
);
3469 qdict_del(bs
->options
, child
->name
);
3472 bdrv_refresh_limits(bs
, NULL
);
3474 bdrv_set_perm(reopen_state
->bs
, reopen_state
->perm
,
3475 reopen_state
->shared_perm
);
3478 !bdrv_is_read_only(bs
) && !(bdrv_get_flags(bs
) & BDRV_O_INACTIVE
);
3479 if (!old_can_write
&& new_can_write
&& drv
->bdrv_reopen_bitmaps_rw
) {
3480 Error
*local_err
= NULL
;
3481 if (drv
->bdrv_reopen_bitmaps_rw(bs
, &local_err
) < 0) {
3482 /* This is not fatal, bitmaps just left read-only, so all following
3483 * writes will fail. User can remove read-only bitmaps to unblock
3486 error_reportf_err(local_err
,
3487 "%s: Failed to make dirty bitmaps writable: ",
3488 bdrv_get_node_name(bs
));
3494 * Abort the reopen, and delete and free the staged changes in
3497 void bdrv_reopen_abort(BDRVReopenState
*reopen_state
)
3501 assert(reopen_state
!= NULL
);
3502 drv
= reopen_state
->bs
->drv
;
3503 assert(drv
!= NULL
);
3505 if (drv
->bdrv_reopen_abort
) {
3506 drv
->bdrv_reopen_abort(reopen_state
);
3509 bdrv_abort_perm_update(reopen_state
->bs
);
3513 static void bdrv_close(BlockDriverState
*bs
)
3515 BdrvAioNotifier
*ban
, *ban_next
;
3516 BdrvChild
*child
, *next
;
3519 assert(!bs
->refcnt
);
3521 bdrv_drained_begin(bs
); /* complete I/O */
3523 bdrv_drain(bs
); /* in case flush left pending I/O */
3526 if (bs
->drv
->bdrv_close
) {
3527 bs
->drv
->bdrv_close(bs
);
3532 bdrv_set_backing_hd(bs
, NULL
, &error_abort
);
3534 if (bs
->file
!= NULL
) {
3535 bdrv_unref_child(bs
, bs
->file
);
3539 QLIST_FOREACH_SAFE(child
, &bs
->children
, next
, next
) {
3540 /* TODO Remove bdrv_unref() from drivers' close function and use
3541 * bdrv_unref_child() here */
3542 if (child
->bs
->inherits_from
== bs
) {
3543 child
->bs
->inherits_from
= NULL
;
3545 bdrv_detach_child(child
);
3550 atomic_set(&bs
->copy_on_read
, 0);
3551 bs
->backing_file
[0] = '\0';
3552 bs
->backing_format
[0] = '\0';
3553 bs
->total_sectors
= 0;
3554 bs
->encrypted
= false;
3556 qobject_unref(bs
->options
);
3557 qobject_unref(bs
->explicit_options
);
3559 bs
->explicit_options
= NULL
;
3560 qobject_unref(bs
->full_open_options
);
3561 bs
->full_open_options
= NULL
;
3563 bdrv_release_named_dirty_bitmaps(bs
);
3564 assert(QLIST_EMPTY(&bs
->dirty_bitmaps
));
3566 QLIST_FOREACH_SAFE(ban
, &bs
->aio_notifiers
, list
, ban_next
) {
3569 QLIST_INIT(&bs
->aio_notifiers
);
3570 bdrv_drained_end(bs
);
3573 void bdrv_close_all(void)
3575 assert(job_next(NULL
) == NULL
);
3576 nbd_export_close_all();
3578 /* Drop references from requests still in flight, such as canceled block
3579 * jobs whose AIO context has not been polled yet */
3582 blk_remove_all_bs();
3583 blockdev_close_all_bdrv_states();
3585 assert(QTAILQ_EMPTY(&all_bdrv_states
));
3588 static bool should_update_child(BdrvChild
*c
, BlockDriverState
*to
)
3594 if (c
->role
->stay_at_node
) {
3598 /* If the child @c belongs to the BDS @to, replacing the current
3599 * c->bs by @to would mean to create a loop.
3601 * Such a case occurs when appending a BDS to a backing chain.
3602 * For instance, imagine the following chain:
3604 * guest device -> node A -> further backing chain...
3606 * Now we create a new BDS B which we want to put on top of this
3607 * chain, so we first attach A as its backing node:
3612 * guest device -> node A -> further backing chain...
3614 * Finally we want to replace A by B. When doing that, we want to
3615 * replace all pointers to A by pointers to B -- except for the
3616 * pointer from B because (1) that would create a loop, and (2)
3617 * that pointer should simply stay intact:
3619 * guest device -> node B
3622 * node A -> further backing chain...
3624 * In general, when replacing a node A (c->bs) by a node B (@to),
3625 * if A is a child of B, that means we cannot replace A by B there
3626 * because that would create a loop. Silently detaching A from B
3627 * is also not really an option. So overall just leaving A in
3628 * place there is the most sensible choice.
3630 * We would also create a loop in any cases where @c is only
3631 * indirectly referenced by @to. Prevent this by returning false
3632 * if @c is found (by breadth-first search) anywhere in the whole
3637 found
= g_hash_table_new(NULL
, NULL
);
3638 g_hash_table_add(found
, to
);
3639 queue
= g_queue_new();
3640 g_queue_push_tail(queue
, to
);
3642 while (!g_queue_is_empty(queue
)) {
3643 BlockDriverState
*v
= g_queue_pop_head(queue
);
3646 QLIST_FOREACH(c2
, &v
->children
, next
) {
3652 if (g_hash_table_contains(found
, c2
->bs
)) {
3656 g_queue_push_tail(queue
, c2
->bs
);
3657 g_hash_table_add(found
, c2
->bs
);
3661 g_queue_free(queue
);
3662 g_hash_table_destroy(found
);
3667 void bdrv_replace_node(BlockDriverState
*from
, BlockDriverState
*to
,
3670 BdrvChild
*c
, *next
;
3671 GSList
*list
= NULL
, *p
;
3672 uint64_t old_perm
, old_shared
;
3673 uint64_t perm
= 0, shared
= BLK_PERM_ALL
;
3676 assert(!atomic_read(&from
->in_flight
));
3677 assert(!atomic_read(&to
->in_flight
));
3679 /* Make sure that @from doesn't go away until we have successfully attached
3680 * all of its parents to @to. */
3683 /* Put all parents into @list and calculate their cumulative permissions */
3684 QLIST_FOREACH_SAFE(c
, &from
->parents
, next_parent
, next
) {
3685 assert(c
->bs
== from
);
3686 if (!should_update_child(c
, to
)) {
3689 list
= g_slist_prepend(list
, c
);
3691 shared
&= c
->shared_perm
;
3694 /* Check whether the required permissions can be granted on @to, ignoring
3695 * all BdrvChild in @list so that they can't block themselves. */
3696 ret
= bdrv_check_update_perm(to
, NULL
, perm
, shared
, list
, errp
);
3698 bdrv_abort_perm_update(to
);
3702 /* Now actually perform the change. We performed the permission check for
3703 * all elements of @list at once, so set the permissions all at once at the
3705 for (p
= list
; p
!= NULL
; p
= p
->next
) {
3709 bdrv_replace_child_noperm(c
, to
);
3713 bdrv_get_cumulative_perm(to
, &old_perm
, &old_shared
);
3714 bdrv_set_perm(to
, old_perm
| perm
, old_shared
| shared
);
3722 * Add new bs contents at the top of an image chain while the chain is
3723 * live, while keeping required fields on the top layer.
3725 * This will modify the BlockDriverState fields, and swap contents
3726 * between bs_new and bs_top. Both bs_new and bs_top are modified.
3728 * bs_new must not be attached to a BlockBackend.
3730 * This function does not create any image files.
3732 * bdrv_append() takes ownership of a bs_new reference and unrefs it because
3733 * that's what the callers commonly need. bs_new will be referenced by the old
3734 * parents of bs_top after bdrv_append() returns. If the caller needs to keep a
3735 * reference of its own, it must call bdrv_ref().
3737 void bdrv_append(BlockDriverState
*bs_new
, BlockDriverState
*bs_top
,
3740 Error
*local_err
= NULL
;
3742 bdrv_set_backing_hd(bs_new
, bs_top
, &local_err
);
3744 error_propagate(errp
, local_err
);
3748 bdrv_replace_node(bs_top
, bs_new
, &local_err
);
3750 error_propagate(errp
, local_err
);
3751 bdrv_set_backing_hd(bs_new
, NULL
, &error_abort
);
3755 /* bs_new is now referenced by its new parents, we don't need the
3756 * additional reference any more. */
3761 static void bdrv_delete(BlockDriverState
*bs
)
3764 assert(bdrv_op_blocker_is_empty(bs
));
3765 assert(!bs
->refcnt
);
3769 /* remove from list, if necessary */
3770 if (bs
->node_name
[0] != '\0') {
3771 QTAILQ_REMOVE(&graph_bdrv_states
, bs
, node_list
);
3773 QTAILQ_REMOVE(&all_bdrv_states
, bs
, bs_list
);
3779 * Run consistency checks on an image
3781 * Returns 0 if the check could be completed (it doesn't mean that the image is
3782 * free of errors) or -errno when an internal error occurred. The results of the
3783 * check are stored in res.
3785 static int coroutine_fn
bdrv_co_check(BlockDriverState
*bs
,
3786 BdrvCheckResult
*res
, BdrvCheckMode fix
)
3788 if (bs
->drv
== NULL
) {
3791 if (bs
->drv
->bdrv_co_check
== NULL
) {
3795 memset(res
, 0, sizeof(*res
));
3796 return bs
->drv
->bdrv_co_check(bs
, res
, fix
);
3799 typedef struct CheckCo
{
3800 BlockDriverState
*bs
;
3801 BdrvCheckResult
*res
;
3806 static void bdrv_check_co_entry(void *opaque
)
3808 CheckCo
*cco
= opaque
;
3809 cco
->ret
= bdrv_co_check(cco
->bs
, cco
->res
, cco
->fix
);
3813 int bdrv_check(BlockDriverState
*bs
,
3814 BdrvCheckResult
*res
, BdrvCheckMode fix
)
3820 .ret
= -EINPROGRESS
,
3824 if (qemu_in_coroutine()) {
3825 /* Fast-path if already in coroutine context */
3826 bdrv_check_co_entry(&cco
);
3828 co
= qemu_coroutine_create(bdrv_check_co_entry
, &cco
);
3829 bdrv_coroutine_enter(bs
, co
);
3830 BDRV_POLL_WHILE(bs
, cco
.ret
== -EINPROGRESS
);
3839 * -EINVAL - backing format specified, but no file
3840 * -ENOSPC - can't update the backing file because no space is left in the
3842 * -ENOTSUP - format driver doesn't support changing the backing file
3844 int bdrv_change_backing_file(BlockDriverState
*bs
,
3845 const char *backing_file
, const char *backing_fmt
)
3847 BlockDriver
*drv
= bs
->drv
;
3854 /* Backing file format doesn't make sense without a backing file */
3855 if (backing_fmt
&& !backing_file
) {
3859 if (drv
->bdrv_change_backing_file
!= NULL
) {
3860 ret
= drv
->bdrv_change_backing_file(bs
, backing_file
, backing_fmt
);
3866 pstrcpy(bs
->backing_file
, sizeof(bs
->backing_file
), backing_file
?: "");
3867 pstrcpy(bs
->backing_format
, sizeof(bs
->backing_format
), backing_fmt
?: "");
3868 pstrcpy(bs
->auto_backing_file
, sizeof(bs
->auto_backing_file
),
3869 backing_file
?: "");
3875 * Finds the image layer in the chain that has 'bs' as its backing file.
3877 * active is the current topmost image.
3879 * Returns NULL if bs is not found in active's image chain,
3880 * or if active == bs.
3882 * Returns the bottommost base image if bs == NULL.
3884 BlockDriverState
*bdrv_find_overlay(BlockDriverState
*active
,
3885 BlockDriverState
*bs
)
3887 while (active
&& bs
!= backing_bs(active
)) {
3888 active
= backing_bs(active
);
3894 /* Given a BDS, searches for the base layer. */
3895 BlockDriverState
*bdrv_find_base(BlockDriverState
*bs
)
3897 return bdrv_find_overlay(bs
, NULL
);
3901 * Drops images above 'base' up to and including 'top', and sets the image
3902 * above 'top' to have base as its backing file.
3904 * Requires that the overlay to 'top' is opened r/w, so that the backing file
3905 * information in 'bs' can be properly updated.
3907 * E.g., this will convert the following chain:
3908 * bottom <- base <- intermediate <- top <- active
3912 * bottom <- base <- active
3914 * It is allowed for bottom==base, in which case it converts:
3916 * base <- intermediate <- top <- active
3922 * If backing_file_str is non-NULL, it will be used when modifying top's
3923 * overlay image metadata.
3926 * if active == top, that is considered an error
3929 int bdrv_drop_intermediate(BlockDriverState
*top
, BlockDriverState
*base
,
3930 const char *backing_file_str
)
3932 BlockDriverState
*explicit_top
= top
;
3933 bool update_inherits_from
;
3934 BdrvChild
*c
, *next
;
3935 Error
*local_err
= NULL
;
3940 if (!top
->drv
|| !base
->drv
) {
3944 /* Make sure that base is in the backing chain of top */
3945 if (!bdrv_chain_contains(top
, base
)) {
3949 /* If 'base' recursively inherits from 'top' then we should set
3950 * base->inherits_from to top->inherits_from after 'top' and all
3951 * other intermediate nodes have been dropped.
3952 * If 'top' is an implicit node (e.g. "commit_top") we should skip
3953 * it because no one inherits from it. We use explicit_top for that. */
3954 while (explicit_top
&& explicit_top
->implicit
) {
3955 explicit_top
= backing_bs(explicit_top
);
3957 update_inherits_from
= bdrv_inherits_from_recursive(base
, explicit_top
);
3959 /* success - we can delete the intermediate states, and link top->base */
3960 /* TODO Check graph modification op blockers (BLK_PERM_GRAPH_MOD) once
3961 * we've figured out how they should work. */
3962 if (!backing_file_str
) {
3963 bdrv_refresh_filename(base
);
3964 backing_file_str
= base
->filename
;
3967 QLIST_FOREACH_SAFE(c
, &top
->parents
, next_parent
, next
) {
3968 /* Check whether we are allowed to switch c from top to base */
3969 GSList
*ignore_children
= g_slist_prepend(NULL
, c
);
3970 bdrv_check_update_perm(base
, NULL
, c
->perm
, c
->shared_perm
,
3971 ignore_children
, &local_err
);
3972 g_slist_free(ignore_children
);
3975 error_report_err(local_err
);
3979 /* If so, update the backing file path in the image file */
3980 if (c
->role
->update_filename
) {
3981 ret
= c
->role
->update_filename(c
, base
, backing_file_str
,
3984 bdrv_abort_perm_update(base
);
3985 error_report_err(local_err
);
3990 /* Do the actual switch in the in-memory graph.
3991 * Completes bdrv_check_update_perm() transaction internally. */
3993 bdrv_replace_child(c
, base
);
3997 if (update_inherits_from
) {
3998 base
->inherits_from
= explicit_top
->inherits_from
;
4008 * Length of a allocated file in bytes. Sparse files are counted by actual
4009 * allocated space. Return < 0 if error or unknown.
4011 int64_t bdrv_get_allocated_file_size(BlockDriverState
*bs
)
4013 BlockDriver
*drv
= bs
->drv
;
4017 if (drv
->bdrv_get_allocated_file_size
) {
4018 return drv
->bdrv_get_allocated_file_size(bs
);
4021 return bdrv_get_allocated_file_size(bs
->file
->bs
);
4028 * @drv: Format driver
4029 * @opts: Creation options for new image
4030 * @in_bs: Existing image containing data for new image (may be NULL)
4031 * @errp: Error object
4032 * Returns: A #BlockMeasureInfo (free using qapi_free_BlockMeasureInfo())
4035 * Calculate file size required to create a new image.
4037 * If @in_bs is given then space for allocated clusters and zero clusters
4038 * from that image are included in the calculation. If @opts contains a
4039 * backing file that is shared by @in_bs then backing clusters may be omitted
4040 * from the calculation.
4042 * If @in_bs is NULL then the calculation includes no allocated clusters
4043 * unless a preallocation option is given in @opts.
4045 * Note that @in_bs may use a different BlockDriver from @drv.
4047 * If an error occurs the @errp pointer is set.
4049 BlockMeasureInfo
*bdrv_measure(BlockDriver
*drv
, QemuOpts
*opts
,
4050 BlockDriverState
*in_bs
, Error
**errp
)
4052 if (!drv
->bdrv_measure
) {
4053 error_setg(errp
, "Block driver '%s' does not support size measurement",
4058 return drv
->bdrv_measure(opts
, in_bs
, errp
);
4062 * Return number of sectors on success, -errno on error.
4064 int64_t bdrv_nb_sectors(BlockDriverState
*bs
)
4066 BlockDriver
*drv
= bs
->drv
;
4071 if (drv
->has_variable_length
) {
4072 int ret
= refresh_total_sectors(bs
, bs
->total_sectors
);
4077 return bs
->total_sectors
;
4081 * Return length in bytes on success, -errno on error.
4082 * The length is always a multiple of BDRV_SECTOR_SIZE.
4084 int64_t bdrv_getlength(BlockDriverState
*bs
)
4086 int64_t ret
= bdrv_nb_sectors(bs
);
4088 ret
= ret
> INT64_MAX
/ BDRV_SECTOR_SIZE
? -EFBIG
: ret
;
4089 return ret
< 0 ? ret
: ret
* BDRV_SECTOR_SIZE
;
4092 /* return 0 as number of sectors if no device present or error */
4093 void bdrv_get_geometry(BlockDriverState
*bs
, uint64_t *nb_sectors_ptr
)
4095 int64_t nb_sectors
= bdrv_nb_sectors(bs
);
4097 *nb_sectors_ptr
= nb_sectors
< 0 ? 0 : nb_sectors
;
4100 bool bdrv_is_sg(BlockDriverState
*bs
)
4105 bool bdrv_is_encrypted(BlockDriverState
*bs
)
4107 if (bs
->backing
&& bs
->backing
->bs
->encrypted
) {
4110 return bs
->encrypted
;
4113 const char *bdrv_get_format_name(BlockDriverState
*bs
)
4115 return bs
->drv
? bs
->drv
->format_name
: NULL
;
4118 static int qsort_strcmp(const void *a
, const void *b
)
4120 return strcmp(*(char *const *)a
, *(char *const *)b
);
4123 void bdrv_iterate_format(void (*it
)(void *opaque
, const char *name
),
4129 const char **formats
= NULL
;
4131 QLIST_FOREACH(drv
, &bdrv_drivers
, list
) {
4132 if (drv
->format_name
) {
4135 while (formats
&& i
&& !found
) {
4136 found
= !strcmp(formats
[--i
], drv
->format_name
);
4140 formats
= g_renew(const char *, formats
, count
+ 1);
4141 formats
[count
++] = drv
->format_name
;
4146 for (i
= 0; i
< (int)ARRAY_SIZE(block_driver_modules
); i
++) {
4147 const char *format_name
= block_driver_modules
[i
].format_name
;
4153 while (formats
&& j
&& !found
) {
4154 found
= !strcmp(formats
[--j
], format_name
);
4158 formats
= g_renew(const char *, formats
, count
+ 1);
4159 formats
[count
++] = format_name
;
4164 qsort(formats
, count
, sizeof(formats
[0]), qsort_strcmp
);
4166 for (i
= 0; i
< count
; i
++) {
4167 it(opaque
, formats
[i
]);
4173 /* This function is to find a node in the bs graph */
4174 BlockDriverState
*bdrv_find_node(const char *node_name
)
4176 BlockDriverState
*bs
;
4180 QTAILQ_FOREACH(bs
, &graph_bdrv_states
, node_list
) {
4181 if (!strcmp(node_name
, bs
->node_name
)) {
4188 /* Put this QMP function here so it can access the static graph_bdrv_states. */
4189 BlockDeviceInfoList
*bdrv_named_nodes_list(Error
**errp
)
4191 BlockDeviceInfoList
*list
, *entry
;
4192 BlockDriverState
*bs
;
4195 QTAILQ_FOREACH(bs
, &graph_bdrv_states
, node_list
) {
4196 BlockDeviceInfo
*info
= bdrv_block_device_info(NULL
, bs
, errp
);
4198 qapi_free_BlockDeviceInfoList(list
);
4201 entry
= g_malloc0(sizeof(*entry
));
4202 entry
->value
= info
;
4210 #define QAPI_LIST_ADD(list, element) do { \
4211 typeof(list) _tmp = g_new(typeof(*(list)), 1); \
4212 _tmp->value = (element); \
4213 _tmp->next = (list); \
4217 typedef struct XDbgBlockGraphConstructor
{
4218 XDbgBlockGraph
*graph
;
4219 GHashTable
*graph_nodes
;
4220 } XDbgBlockGraphConstructor
;
4222 static XDbgBlockGraphConstructor
*xdbg_graph_new(void)
4224 XDbgBlockGraphConstructor
*gr
= g_new(XDbgBlockGraphConstructor
, 1);
4226 gr
->graph
= g_new0(XDbgBlockGraph
, 1);
4227 gr
->graph_nodes
= g_hash_table_new(NULL
, NULL
);
4232 static XDbgBlockGraph
*xdbg_graph_finalize(XDbgBlockGraphConstructor
*gr
)
4234 XDbgBlockGraph
*graph
= gr
->graph
;
4236 g_hash_table_destroy(gr
->graph_nodes
);
4242 static uintptr_t xdbg_graph_node_num(XDbgBlockGraphConstructor
*gr
, void *node
)
4244 uintptr_t ret
= (uintptr_t)g_hash_table_lookup(gr
->graph_nodes
, node
);
4251 * Start counting from 1, not 0, because 0 interferes with not-found (NULL)
4252 * answer of g_hash_table_lookup.
4254 ret
= g_hash_table_size(gr
->graph_nodes
) + 1;
4255 g_hash_table_insert(gr
->graph_nodes
, node
, (void *)ret
);
4260 static void xdbg_graph_add_node(XDbgBlockGraphConstructor
*gr
, void *node
,
4261 XDbgBlockGraphNodeType type
, const char *name
)
4263 XDbgBlockGraphNode
*n
;
4265 n
= g_new0(XDbgBlockGraphNode
, 1);
4267 n
->id
= xdbg_graph_node_num(gr
, node
);
4269 n
->name
= g_strdup(name
);
4271 QAPI_LIST_ADD(gr
->graph
->nodes
, n
);
4274 static void xdbg_graph_add_edge(XDbgBlockGraphConstructor
*gr
, void *parent
,
4275 const BdrvChild
*child
)
4279 BlockPermission num
;
4282 static const PermissionMap permissions
[] = {
4283 { BLK_PERM_CONSISTENT_READ
, BLOCK_PERMISSION_CONSISTENT_READ
},
4284 { BLK_PERM_WRITE
, BLOCK_PERMISSION_WRITE
},
4285 { BLK_PERM_WRITE_UNCHANGED
, BLOCK_PERMISSION_WRITE_UNCHANGED
},
4286 { BLK_PERM_RESIZE
, BLOCK_PERMISSION_RESIZE
},
4287 { BLK_PERM_GRAPH_MOD
, BLOCK_PERMISSION_GRAPH_MOD
},
4290 const PermissionMap
*p
;
4291 XDbgBlockGraphEdge
*edge
;
4293 QEMU_BUILD_BUG_ON(1UL << (ARRAY_SIZE(permissions
) - 1) != BLK_PERM_ALL
+ 1);
4295 edge
= g_new0(XDbgBlockGraphEdge
, 1);
4297 edge
->parent
= xdbg_graph_node_num(gr
, parent
);
4298 edge
->child
= xdbg_graph_node_num(gr
, child
->bs
);
4299 edge
->name
= g_strdup(child
->name
);
4301 for (p
= permissions
; p
->flag
; p
++) {
4302 if (p
->flag
& child
->perm
) {
4303 QAPI_LIST_ADD(edge
->perm
, p
->num
);
4305 if (p
->flag
& child
->shared_perm
) {
4306 QAPI_LIST_ADD(edge
->shared_perm
, p
->num
);
4310 QAPI_LIST_ADD(gr
->graph
->edges
, edge
);
4314 XDbgBlockGraph
*bdrv_get_xdbg_block_graph(Error
**errp
)
4318 BlockDriverState
*bs
;
4320 XDbgBlockGraphConstructor
*gr
= xdbg_graph_new();
4322 for (blk
= blk_all_next(NULL
); blk
; blk
= blk_all_next(blk
)) {
4323 char *allocated_name
= NULL
;
4324 const char *name
= blk_name(blk
);
4327 name
= allocated_name
= blk_get_attached_dev_id(blk
);
4329 xdbg_graph_add_node(gr
, blk
, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_BACKEND
,
4331 g_free(allocated_name
);
4332 if (blk_root(blk
)) {
4333 xdbg_graph_add_edge(gr
, blk
, blk_root(blk
));
4337 for (job
= block_job_next(NULL
); job
; job
= block_job_next(job
)) {
4340 xdbg_graph_add_node(gr
, job
, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_JOB
,
4342 for (el
= job
->nodes
; el
; el
= el
->next
) {
4343 xdbg_graph_add_edge(gr
, job
, (BdrvChild
*)el
->data
);
4347 QTAILQ_FOREACH(bs
, &graph_bdrv_states
, node_list
) {
4348 xdbg_graph_add_node(gr
, bs
, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_DRIVER
,
4350 QLIST_FOREACH(child
, &bs
->children
, next
) {
4351 xdbg_graph_add_edge(gr
, bs
, child
);
4355 return xdbg_graph_finalize(gr
);
4358 BlockDriverState
*bdrv_lookup_bs(const char *device
,
4359 const char *node_name
,
4363 BlockDriverState
*bs
;
4366 blk
= blk_by_name(device
);
4371 error_setg(errp
, "Device '%s' has no medium", device
);
4379 bs
= bdrv_find_node(node_name
);
4386 error_setg(errp
, "Cannot find device=%s nor node_name=%s",
4387 device
? device
: "",
4388 node_name
? node_name
: "");
4392 /* If 'base' is in the same chain as 'top', return true. Otherwise,
4393 * return false. If either argument is NULL, return false. */
4394 bool bdrv_chain_contains(BlockDriverState
*top
, BlockDriverState
*base
)
4396 while (top
&& top
!= base
) {
4397 top
= backing_bs(top
);
4403 BlockDriverState
*bdrv_next_node(BlockDriverState
*bs
)
4406 return QTAILQ_FIRST(&graph_bdrv_states
);
4408 return QTAILQ_NEXT(bs
, node_list
);
4411 BlockDriverState
*bdrv_next_all_states(BlockDriverState
*bs
)
4414 return QTAILQ_FIRST(&all_bdrv_states
);
4416 return QTAILQ_NEXT(bs
, bs_list
);
4419 const char *bdrv_get_node_name(const BlockDriverState
*bs
)
4421 return bs
->node_name
;
4424 const char *bdrv_get_parent_name(const BlockDriverState
*bs
)
4429 /* If multiple parents have a name, just pick the first one. */
4430 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
4431 if (c
->role
->get_name
) {
4432 name
= c
->role
->get_name(c
);
4433 if (name
&& *name
) {
4442 /* TODO check what callers really want: bs->node_name or blk_name() */
4443 const char *bdrv_get_device_name(const BlockDriverState
*bs
)
4445 return bdrv_get_parent_name(bs
) ?: "";
4448 /* This can be used to identify nodes that might not have a device
4449 * name associated. Since node and device names live in the same
4450 * namespace, the result is unambiguous. The exception is if both are
4451 * absent, then this returns an empty (non-null) string. */
4452 const char *bdrv_get_device_or_node_name(const BlockDriverState
*bs
)
4454 return bdrv_get_parent_name(bs
) ?: bs
->node_name
;
4457 int bdrv_get_flags(BlockDriverState
*bs
)
4459 return bs
->open_flags
;
4462 int bdrv_has_zero_init_1(BlockDriverState
*bs
)
4467 int bdrv_has_zero_init(BlockDriverState
*bs
)
4473 /* If BS is a copy on write image, it is initialized to
4474 the contents of the base image, which may not be zeroes. */
4478 if (bs
->drv
->bdrv_has_zero_init
) {
4479 return bs
->drv
->bdrv_has_zero_init(bs
);
4481 if (bs
->file
&& bs
->drv
->is_filter
) {
4482 return bdrv_has_zero_init(bs
->file
->bs
);
4489 bool bdrv_unallocated_blocks_are_zero(BlockDriverState
*bs
)
4491 BlockDriverInfo bdi
;
4497 if (bdrv_get_info(bs
, &bdi
) == 0) {
4498 return bdi
.unallocated_blocks_are_zero
;
4504 bool bdrv_can_write_zeroes_with_unmap(BlockDriverState
*bs
)
4506 if (!(bs
->open_flags
& BDRV_O_UNMAP
)) {
4510 return bs
->supported_zero_flags
& BDRV_REQ_MAY_UNMAP
;
4513 void bdrv_get_backing_filename(BlockDriverState
*bs
,
4514 char *filename
, int filename_size
)
4516 pstrcpy(filename
, filename_size
, bs
->backing_file
);
4519 int bdrv_get_info(BlockDriverState
*bs
, BlockDriverInfo
*bdi
)
4521 BlockDriver
*drv
= bs
->drv
;
4522 /* if bs->drv == NULL, bs is closed, so there's nothing to do here */
4526 if (!drv
->bdrv_get_info
) {
4527 if (bs
->file
&& drv
->is_filter
) {
4528 return bdrv_get_info(bs
->file
->bs
, bdi
);
4532 memset(bdi
, 0, sizeof(*bdi
));
4533 return drv
->bdrv_get_info(bs
, bdi
);
4536 ImageInfoSpecific
*bdrv_get_specific_info(BlockDriverState
*bs
,
4539 BlockDriver
*drv
= bs
->drv
;
4540 if (drv
&& drv
->bdrv_get_specific_info
) {
4541 return drv
->bdrv_get_specific_info(bs
, errp
);
4546 void bdrv_debug_event(BlockDriverState
*bs
, BlkdebugEvent event
)
4548 if (!bs
|| !bs
->drv
|| !bs
->drv
->bdrv_debug_event
) {
4552 bs
->drv
->bdrv_debug_event(bs
, event
);
4555 int bdrv_debug_breakpoint(BlockDriverState
*bs
, const char *event
,
4558 while (bs
&& bs
->drv
&& !bs
->drv
->bdrv_debug_breakpoint
) {
4559 bs
= bs
->file
? bs
->file
->bs
: NULL
;
4562 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_breakpoint
) {
4563 return bs
->drv
->bdrv_debug_breakpoint(bs
, event
, tag
);
4569 int bdrv_debug_remove_breakpoint(BlockDriverState
*bs
, const char *tag
)
4571 while (bs
&& bs
->drv
&& !bs
->drv
->bdrv_debug_remove_breakpoint
) {
4572 bs
= bs
->file
? bs
->file
->bs
: NULL
;
4575 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_remove_breakpoint
) {
4576 return bs
->drv
->bdrv_debug_remove_breakpoint(bs
, tag
);
4582 int bdrv_debug_resume(BlockDriverState
*bs
, const char *tag
)
4584 while (bs
&& (!bs
->drv
|| !bs
->drv
->bdrv_debug_resume
)) {
4585 bs
= bs
->file
? bs
->file
->bs
: NULL
;
4588 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_resume
) {
4589 return bs
->drv
->bdrv_debug_resume(bs
, tag
);
4595 bool bdrv_debug_is_suspended(BlockDriverState
*bs
, const char *tag
)
4597 while (bs
&& bs
->drv
&& !bs
->drv
->bdrv_debug_is_suspended
) {
4598 bs
= bs
->file
? bs
->file
->bs
: NULL
;
4601 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_is_suspended
) {
4602 return bs
->drv
->bdrv_debug_is_suspended(bs
, tag
);
4608 /* backing_file can either be relative, or absolute, or a protocol. If it is
4609 * relative, it must be relative to the chain. So, passing in bs->filename
4610 * from a BDS as backing_file should not be done, as that may be relative to
4611 * the CWD rather than the chain. */
4612 BlockDriverState
*bdrv_find_backing_image(BlockDriverState
*bs
,
4613 const char *backing_file
)
4615 char *filename_full
= NULL
;
4616 char *backing_file_full
= NULL
;
4617 char *filename_tmp
= NULL
;
4618 int is_protocol
= 0;
4619 BlockDriverState
*curr_bs
= NULL
;
4620 BlockDriverState
*retval
= NULL
;
4621 Error
*local_error
= NULL
;
4623 if (!bs
|| !bs
->drv
|| !backing_file
) {
4627 filename_full
= g_malloc(PATH_MAX
);
4628 backing_file_full
= g_malloc(PATH_MAX
);
4629 filename_tmp
= g_malloc(PATH_MAX
);
4631 is_protocol
= path_has_protocol(backing_file
);
4633 /* This will recursively refresh everything in the backing chain */
4634 bdrv_refresh_filename(bs
);
4636 for (curr_bs
= bs
; curr_bs
->backing
; curr_bs
= curr_bs
->backing
->bs
) {
4638 /* If either of the filename paths is actually a protocol, then
4639 * compare unmodified paths; otherwise make paths relative */
4640 if (is_protocol
|| path_has_protocol(curr_bs
->backing_file
)) {
4641 if (strcmp(backing_file
, curr_bs
->backing_file
) == 0) {
4642 retval
= curr_bs
->backing
->bs
;
4645 /* Also check against the full backing filename for the image */
4646 bdrv_get_full_backing_filename(curr_bs
, backing_file_full
, PATH_MAX
,
4648 if (local_error
== NULL
) {
4649 if (strcmp(backing_file
, backing_file_full
) == 0) {
4650 retval
= curr_bs
->backing
->bs
;
4654 error_free(local_error
);
4658 /* If not an absolute filename path, make it relative to the current
4659 * image's filename path */
4660 path_combine(filename_tmp
, PATH_MAX
, curr_bs
->filename
,
4663 /* We are going to compare absolute pathnames */
4664 if (!realpath(filename_tmp
, filename_full
)) {
4668 /* We need to make sure the backing filename we are comparing against
4669 * is relative to the current image filename (or absolute) */
4670 path_combine(filename_tmp
, PATH_MAX
, curr_bs
->filename
,
4671 curr_bs
->backing_file
);
4673 if (!realpath(filename_tmp
, backing_file_full
)) {
4677 if (strcmp(backing_file_full
, filename_full
) == 0) {
4678 retval
= curr_bs
->backing
->bs
;
4684 g_free(filename_full
);
4685 g_free(backing_file_full
);
4686 g_free(filename_tmp
);
4690 void bdrv_init(void)
4692 module_call_init(MODULE_INIT_BLOCK
);
4695 void bdrv_init_with_whitelist(void)
4697 use_bdrv_whitelist
= 1;
4701 static void coroutine_fn
bdrv_co_invalidate_cache(BlockDriverState
*bs
,
4704 BdrvChild
*child
, *parent
;
4705 uint64_t perm
, shared_perm
;
4706 Error
*local_err
= NULL
;
4708 BdrvDirtyBitmap
*bm
;
4714 if (!(bs
->open_flags
& BDRV_O_INACTIVE
)) {
4718 QLIST_FOREACH(child
, &bs
->children
, next
) {
4719 bdrv_co_invalidate_cache(child
->bs
, &local_err
);
4721 error_propagate(errp
, local_err
);
4727 * Update permissions, they may differ for inactive nodes.
4729 * Note that the required permissions of inactive images are always a
4730 * subset of the permissions required after activating the image. This
4731 * allows us to just get the permissions upfront without restricting
4732 * drv->bdrv_invalidate_cache().
4734 * It also means that in error cases, we don't have to try and revert to
4735 * the old permissions (which is an operation that could fail, too). We can
4736 * just keep the extended permissions for the next time that an activation
4737 * of the image is tried.
4739 bs
->open_flags
&= ~BDRV_O_INACTIVE
;
4740 bdrv_get_cumulative_perm(bs
, &perm
, &shared_perm
);
4741 ret
= bdrv_check_perm(bs
, NULL
, perm
, shared_perm
, NULL
, &local_err
);
4743 bs
->open_flags
|= BDRV_O_INACTIVE
;
4744 error_propagate(errp
, local_err
);
4747 bdrv_set_perm(bs
, perm
, shared_perm
);
4749 if (bs
->drv
->bdrv_co_invalidate_cache
) {
4750 bs
->drv
->bdrv_co_invalidate_cache(bs
, &local_err
);
4752 bs
->open_flags
|= BDRV_O_INACTIVE
;
4753 error_propagate(errp
, local_err
);
4758 for (bm
= bdrv_dirty_bitmap_next(bs
, NULL
); bm
;
4759 bm
= bdrv_dirty_bitmap_next(bs
, bm
))
4761 bdrv_dirty_bitmap_set_migration(bm
, false);
4764 ret
= refresh_total_sectors(bs
, bs
->total_sectors
);
4766 bs
->open_flags
|= BDRV_O_INACTIVE
;
4767 error_setg_errno(errp
, -ret
, "Could not refresh total sector count");
4771 QLIST_FOREACH(parent
, &bs
->parents
, next_parent
) {
4772 if (parent
->role
->activate
) {
4773 parent
->role
->activate(parent
, &local_err
);
4775 bs
->open_flags
|= BDRV_O_INACTIVE
;
4776 error_propagate(errp
, local_err
);
4783 typedef struct InvalidateCacheCo
{
4784 BlockDriverState
*bs
;
4787 } InvalidateCacheCo
;
4789 static void coroutine_fn
bdrv_invalidate_cache_co_entry(void *opaque
)
4791 InvalidateCacheCo
*ico
= opaque
;
4792 bdrv_co_invalidate_cache(ico
->bs
, ico
->errp
);
4797 void bdrv_invalidate_cache(BlockDriverState
*bs
, Error
**errp
)
4800 InvalidateCacheCo ico
= {
4806 if (qemu_in_coroutine()) {
4807 /* Fast-path if already in coroutine context */
4808 bdrv_invalidate_cache_co_entry(&ico
);
4810 co
= qemu_coroutine_create(bdrv_invalidate_cache_co_entry
, &ico
);
4811 bdrv_coroutine_enter(bs
, co
);
4812 BDRV_POLL_WHILE(bs
, !ico
.done
);
4816 void bdrv_invalidate_cache_all(Error
**errp
)
4818 BlockDriverState
*bs
;
4819 Error
*local_err
= NULL
;
4820 BdrvNextIterator it
;
4822 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
4823 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
4825 aio_context_acquire(aio_context
);
4826 bdrv_invalidate_cache(bs
, &local_err
);
4827 aio_context_release(aio_context
);
4829 error_propagate(errp
, local_err
);
4830 bdrv_next_cleanup(&it
);
4836 static bool bdrv_has_bds_parent(BlockDriverState
*bs
, bool only_active
)
4840 QLIST_FOREACH(parent
, &bs
->parents
, next_parent
) {
4841 if (parent
->role
->parent_is_bds
) {
4842 BlockDriverState
*parent_bs
= parent
->opaque
;
4843 if (!only_active
|| !(parent_bs
->open_flags
& BDRV_O_INACTIVE
)) {
4852 static int bdrv_inactivate_recurse(BlockDriverState
*bs
)
4854 BdrvChild
*child
, *parent
;
4855 uint64_t perm
, shared_perm
;
4862 /* Make sure that we don't inactivate a child before its parent.
4863 * It will be covered by recursion from the yet active parent. */
4864 if (bdrv_has_bds_parent(bs
, true)) {
4868 assert(!(bs
->open_flags
& BDRV_O_INACTIVE
));
4870 /* Inactivate this node */
4871 if (bs
->drv
->bdrv_inactivate
) {
4872 ret
= bs
->drv
->bdrv_inactivate(bs
);
4878 QLIST_FOREACH(parent
, &bs
->parents
, next_parent
) {
4879 if (parent
->role
->inactivate
) {
4880 ret
= parent
->role
->inactivate(parent
);
4887 bs
->open_flags
|= BDRV_O_INACTIVE
;
4889 /* Update permissions, they may differ for inactive nodes */
4890 bdrv_get_cumulative_perm(bs
, &perm
, &shared_perm
);
4891 bdrv_check_perm(bs
, NULL
, perm
, shared_perm
, NULL
, &error_abort
);
4892 bdrv_set_perm(bs
, perm
, shared_perm
);
4895 /* Recursively inactivate children */
4896 QLIST_FOREACH(child
, &bs
->children
, next
) {
4897 ret
= bdrv_inactivate_recurse(child
->bs
);
4906 int bdrv_inactivate_all(void)
4908 BlockDriverState
*bs
= NULL
;
4909 BdrvNextIterator it
;
4911 GSList
*aio_ctxs
= NULL
, *ctx
;
4913 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
4914 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
4916 if (!g_slist_find(aio_ctxs
, aio_context
)) {
4917 aio_ctxs
= g_slist_prepend(aio_ctxs
, aio_context
);
4918 aio_context_acquire(aio_context
);
4922 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
4923 /* Nodes with BDS parents are covered by recursion from the last
4924 * parent that gets inactivated. Don't inactivate them a second
4925 * time if that has already happened. */
4926 if (bdrv_has_bds_parent(bs
, false)) {
4929 ret
= bdrv_inactivate_recurse(bs
);
4931 bdrv_next_cleanup(&it
);
4937 for (ctx
= aio_ctxs
; ctx
!= NULL
; ctx
= ctx
->next
) {
4938 AioContext
*aio_context
= ctx
->data
;
4939 aio_context_release(aio_context
);
4941 g_slist_free(aio_ctxs
);
4946 /**************************************************************/
4947 /* removable device support */
4950 * Return TRUE if the media is present
4952 bool bdrv_is_inserted(BlockDriverState
*bs
)
4954 BlockDriver
*drv
= bs
->drv
;
4960 if (drv
->bdrv_is_inserted
) {
4961 return drv
->bdrv_is_inserted(bs
);
4963 QLIST_FOREACH(child
, &bs
->children
, next
) {
4964 if (!bdrv_is_inserted(child
->bs
)) {
4972 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
4974 void bdrv_eject(BlockDriverState
*bs
, bool eject_flag
)
4976 BlockDriver
*drv
= bs
->drv
;
4978 if (drv
&& drv
->bdrv_eject
) {
4979 drv
->bdrv_eject(bs
, eject_flag
);
4984 * Lock or unlock the media (if it is locked, the user won't be able
4985 * to eject it manually).
4987 void bdrv_lock_medium(BlockDriverState
*bs
, bool locked
)
4989 BlockDriver
*drv
= bs
->drv
;
4991 trace_bdrv_lock_medium(bs
, locked
);
4993 if (drv
&& drv
->bdrv_lock_medium
) {
4994 drv
->bdrv_lock_medium(bs
, locked
);
4998 /* Get a reference to bs */
4999 void bdrv_ref(BlockDriverState
*bs
)
5004 /* Release a previously grabbed reference to bs.
5005 * If after releasing, reference count is zero, the BlockDriverState is
5007 void bdrv_unref(BlockDriverState
*bs
)
5012 assert(bs
->refcnt
> 0);
5013 if (--bs
->refcnt
== 0) {
5018 struct BdrvOpBlocker
{
5020 QLIST_ENTRY(BdrvOpBlocker
) list
;
5023 bool bdrv_op_is_blocked(BlockDriverState
*bs
, BlockOpType op
, Error
**errp
)
5025 BdrvOpBlocker
*blocker
;
5026 assert((int) op
>= 0 && op
< BLOCK_OP_TYPE_MAX
);
5027 if (!QLIST_EMPTY(&bs
->op_blockers
[op
])) {
5028 blocker
= QLIST_FIRST(&bs
->op_blockers
[op
]);
5029 error_propagate_prepend(errp
, error_copy(blocker
->reason
),
5030 "Node '%s' is busy: ",
5031 bdrv_get_device_or_node_name(bs
));
5037 void bdrv_op_block(BlockDriverState
*bs
, BlockOpType op
, Error
*reason
)
5039 BdrvOpBlocker
*blocker
;
5040 assert((int) op
>= 0 && op
< BLOCK_OP_TYPE_MAX
);
5042 blocker
= g_new0(BdrvOpBlocker
, 1);
5043 blocker
->reason
= reason
;
5044 QLIST_INSERT_HEAD(&bs
->op_blockers
[op
], blocker
, list
);
5047 void bdrv_op_unblock(BlockDriverState
*bs
, BlockOpType op
, Error
*reason
)
5049 BdrvOpBlocker
*blocker
, *next
;
5050 assert((int) op
>= 0 && op
< BLOCK_OP_TYPE_MAX
);
5051 QLIST_FOREACH_SAFE(blocker
, &bs
->op_blockers
[op
], list
, next
) {
5052 if (blocker
->reason
== reason
) {
5053 QLIST_REMOVE(blocker
, list
);
5059 void bdrv_op_block_all(BlockDriverState
*bs
, Error
*reason
)
5062 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
5063 bdrv_op_block(bs
, i
, reason
);
5067 void bdrv_op_unblock_all(BlockDriverState
*bs
, Error
*reason
)
5070 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
5071 bdrv_op_unblock(bs
, i
, reason
);
5075 bool bdrv_op_blocker_is_empty(BlockDriverState
*bs
)
5079 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
5080 if (!QLIST_EMPTY(&bs
->op_blockers
[i
])) {
5087 void bdrv_img_create(const char *filename
, const char *fmt
,
5088 const char *base_filename
, const char *base_fmt
,
5089 char *options
, uint64_t img_size
, int flags
, bool quiet
,
5092 QemuOptsList
*create_opts
= NULL
;
5093 QemuOpts
*opts
= NULL
;
5094 const char *backing_fmt
, *backing_file
;
5096 BlockDriver
*drv
, *proto_drv
;
5097 Error
*local_err
= NULL
;
5100 /* Find driver and parse its options */
5101 drv
= bdrv_find_format(fmt
);
5103 error_setg(errp
, "Unknown file format '%s'", fmt
);
5107 proto_drv
= bdrv_find_protocol(filename
, true, errp
);
5112 if (!drv
->create_opts
) {
5113 error_setg(errp
, "Format driver '%s' does not support image creation",
5118 if (!proto_drv
->create_opts
) {
5119 error_setg(errp
, "Protocol driver '%s' does not support image creation",
5120 proto_drv
->format_name
);
5124 create_opts
= qemu_opts_append(create_opts
, drv
->create_opts
);
5125 create_opts
= qemu_opts_append(create_opts
, proto_drv
->create_opts
);
5127 /* Create parameter list with default values */
5128 opts
= qemu_opts_create(create_opts
, NULL
, 0, &error_abort
);
5129 qemu_opt_set_number(opts
, BLOCK_OPT_SIZE
, img_size
, &error_abort
);
5131 /* Parse -o options */
5133 qemu_opts_do_parse(opts
, options
, NULL
, &local_err
);
5139 if (base_filename
) {
5140 qemu_opt_set(opts
, BLOCK_OPT_BACKING_FILE
, base_filename
, &local_err
);
5142 error_setg(errp
, "Backing file not supported for file format '%s'",
5149 qemu_opt_set(opts
, BLOCK_OPT_BACKING_FMT
, base_fmt
, &local_err
);
5151 error_setg(errp
, "Backing file format not supported for file "
5152 "format '%s'", fmt
);
5157 backing_file
= qemu_opt_get(opts
, BLOCK_OPT_BACKING_FILE
);
5159 if (!strcmp(filename
, backing_file
)) {
5160 error_setg(errp
, "Error: Trying to create an image with the "
5161 "same filename as the backing file");
5166 backing_fmt
= qemu_opt_get(opts
, BLOCK_OPT_BACKING_FMT
);
5168 /* The size for the image must always be specified, unless we have a backing
5169 * file and we have not been forbidden from opening it. */
5170 size
= qemu_opt_get_size(opts
, BLOCK_OPT_SIZE
, img_size
);
5171 if (backing_file
&& !(flags
& BDRV_O_NO_BACKING
)) {
5172 BlockDriverState
*bs
;
5173 char *full_backing
= g_new0(char, PATH_MAX
);
5175 QDict
*backing_options
= NULL
;
5177 bdrv_get_full_backing_filename_from_filename(filename
, backing_file
,
5178 full_backing
, PATH_MAX
,
5181 g_free(full_backing
);
5185 /* backing files always opened read-only */
5187 back_flags
&= ~(BDRV_O_RDWR
| BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
);
5189 backing_options
= qdict_new();
5191 qdict_put_str(backing_options
, "driver", backing_fmt
);
5193 qdict_put_bool(backing_options
, BDRV_OPT_FORCE_SHARE
, true);
5195 bs
= bdrv_open(full_backing
, NULL
, backing_options
, back_flags
,
5197 g_free(full_backing
);
5198 if (!bs
&& size
!= -1) {
5199 /* Couldn't open BS, but we have a size, so it's nonfatal */
5200 warn_reportf_err(local_err
,
5201 "Could not verify backing image. "
5202 "This may become an error in future versions.\n");
5205 /* Couldn't open bs, do not have size */
5206 error_append_hint(&local_err
,
5207 "Could not open backing image to determine size.\n");
5211 /* Opened BS, have no size */
5212 size
= bdrv_getlength(bs
);
5214 error_setg_errno(errp
, -size
, "Could not get size of '%s'",
5219 qemu_opt_set_number(opts
, BLOCK_OPT_SIZE
, size
, &error_abort
);
5223 } /* (backing_file && !(flags & BDRV_O_NO_BACKING)) */
5226 error_setg(errp
, "Image creation needs a size parameter");
5231 printf("Formatting '%s', fmt=%s ", filename
, fmt
);
5232 qemu_opts_print(opts
, " ");
5236 ret
= bdrv_create(drv
, filename
, opts
, &local_err
);
5238 if (ret
== -EFBIG
) {
5239 /* This is generally a better message than whatever the driver would
5240 * deliver (especially because of the cluster_size_hint), since that
5241 * is most probably not much different from "image too large". */
5242 const char *cluster_size_hint
= "";
5243 if (qemu_opt_get_size(opts
, BLOCK_OPT_CLUSTER_SIZE
, 0)) {
5244 cluster_size_hint
= " (try using a larger cluster size)";
5246 error_setg(errp
, "The image size is too large for file format '%s'"
5247 "%s", fmt
, cluster_size_hint
);
5248 error_free(local_err
);
5253 qemu_opts_del(opts
);
5254 qemu_opts_free(create_opts
);
5255 error_propagate(errp
, local_err
);
5258 AioContext
*bdrv_get_aio_context(BlockDriverState
*bs
)
5260 return bs
? bs
->aio_context
: qemu_get_aio_context();
5263 void bdrv_coroutine_enter(BlockDriverState
*bs
, Coroutine
*co
)
5265 aio_co_enter(bdrv_get_aio_context(bs
), co
);
5268 static void bdrv_do_remove_aio_context_notifier(BdrvAioNotifier
*ban
)
5270 QLIST_REMOVE(ban
, list
);
5274 void bdrv_detach_aio_context(BlockDriverState
*bs
)
5276 BdrvAioNotifier
*baf
, *baf_tmp
;
5283 assert(!bs
->walking_aio_notifiers
);
5284 bs
->walking_aio_notifiers
= true;
5285 QLIST_FOREACH_SAFE(baf
, &bs
->aio_notifiers
, list
, baf_tmp
) {
5287 bdrv_do_remove_aio_context_notifier(baf
);
5289 baf
->detach_aio_context(baf
->opaque
);
5292 /* Never mind iterating again to check for ->deleted. bdrv_close() will
5293 * remove remaining aio notifiers if we aren't called again.
5295 bs
->walking_aio_notifiers
= false;
5297 if (bs
->drv
->bdrv_detach_aio_context
) {
5298 bs
->drv
->bdrv_detach_aio_context(bs
);
5300 QLIST_FOREACH(child
, &bs
->children
, next
) {
5301 bdrv_detach_aio_context(child
->bs
);
5304 if (bs
->quiesce_counter
) {
5305 aio_enable_external(bs
->aio_context
);
5307 bs
->aio_context
= NULL
;
5310 void bdrv_attach_aio_context(BlockDriverState
*bs
,
5311 AioContext
*new_context
)
5313 BdrvAioNotifier
*ban
, *ban_tmp
;
5320 if (bs
->quiesce_counter
) {
5321 aio_disable_external(new_context
);
5324 bs
->aio_context
= new_context
;
5326 QLIST_FOREACH(child
, &bs
->children
, next
) {
5327 bdrv_attach_aio_context(child
->bs
, new_context
);
5329 if (bs
->drv
->bdrv_attach_aio_context
) {
5330 bs
->drv
->bdrv_attach_aio_context(bs
, new_context
);
5333 assert(!bs
->walking_aio_notifiers
);
5334 bs
->walking_aio_notifiers
= true;
5335 QLIST_FOREACH_SAFE(ban
, &bs
->aio_notifiers
, list
, ban_tmp
) {
5337 bdrv_do_remove_aio_context_notifier(ban
);
5339 ban
->attached_aio_context(new_context
, ban
->opaque
);
5342 bs
->walking_aio_notifiers
= false;
5345 /* The caller must own the AioContext lock for the old AioContext of bs, but it
5346 * must not own the AioContext lock for new_context (unless new_context is
5347 * the same as the current context of bs). */
5348 void bdrv_set_aio_context(BlockDriverState
*bs
, AioContext
*new_context
)
5350 if (bdrv_get_aio_context(bs
) == new_context
) {
5354 bdrv_drained_begin(bs
);
5355 bdrv_detach_aio_context(bs
);
5357 /* This function executes in the old AioContext so acquire the new one in
5358 * case it runs in a different thread.
5360 aio_context_acquire(new_context
);
5361 bdrv_attach_aio_context(bs
, new_context
);
5362 bdrv_drained_end(bs
);
5363 aio_context_release(new_context
);
5366 void bdrv_add_aio_context_notifier(BlockDriverState
*bs
,
5367 void (*attached_aio_context
)(AioContext
*new_context
, void *opaque
),
5368 void (*detach_aio_context
)(void *opaque
), void *opaque
)
5370 BdrvAioNotifier
*ban
= g_new(BdrvAioNotifier
, 1);
5371 *ban
= (BdrvAioNotifier
){
5372 .attached_aio_context
= attached_aio_context
,
5373 .detach_aio_context
= detach_aio_context
,
5377 QLIST_INSERT_HEAD(&bs
->aio_notifiers
, ban
, list
);
5380 void bdrv_remove_aio_context_notifier(BlockDriverState
*bs
,
5381 void (*attached_aio_context
)(AioContext
*,
5383 void (*detach_aio_context
)(void *),
5386 BdrvAioNotifier
*ban
, *ban_next
;
5388 QLIST_FOREACH_SAFE(ban
, &bs
->aio_notifiers
, list
, ban_next
) {
5389 if (ban
->attached_aio_context
== attached_aio_context
&&
5390 ban
->detach_aio_context
== detach_aio_context
&&
5391 ban
->opaque
== opaque
&&
5392 ban
->deleted
== false)
5394 if (bs
->walking_aio_notifiers
) {
5395 ban
->deleted
= true;
5397 bdrv_do_remove_aio_context_notifier(ban
);
5406 int bdrv_amend_options(BlockDriverState
*bs
, QemuOpts
*opts
,
5407 BlockDriverAmendStatusCB
*status_cb
, void *cb_opaque
,
5411 error_setg(errp
, "Node is ejected");
5414 if (!bs
->drv
->bdrv_amend_options
) {
5415 error_setg(errp
, "Block driver '%s' does not support option amendment",
5416 bs
->drv
->format_name
);
5419 return bs
->drv
->bdrv_amend_options(bs
, opts
, status_cb
, cb_opaque
, errp
);
5422 /* This function will be called by the bdrv_recurse_is_first_non_filter method
5423 * of block filter and by bdrv_is_first_non_filter.
5424 * It is used to test if the given bs is the candidate or recurse more in the
5427 bool bdrv_recurse_is_first_non_filter(BlockDriverState
*bs
,
5428 BlockDriverState
*candidate
)
5430 /* return false if basic checks fails */
5431 if (!bs
|| !bs
->drv
) {
5435 /* the code reached a non block filter driver -> check if the bs is
5436 * the same as the candidate. It's the recursion termination condition.
5438 if (!bs
->drv
->is_filter
) {
5439 return bs
== candidate
;
5441 /* Down this path the driver is a block filter driver */
5443 /* If the block filter recursion method is defined use it to recurse down
5446 if (bs
->drv
->bdrv_recurse_is_first_non_filter
) {
5447 return bs
->drv
->bdrv_recurse_is_first_non_filter(bs
, candidate
);
5450 /* the driver is a block filter but don't allow to recurse -> return false
5455 /* This function checks if the candidate is the first non filter bs down it's
5456 * bs chain. Since we don't have pointers to parents it explore all bs chains
5457 * from the top. Some filters can choose not to pass down the recursion.
5459 bool bdrv_is_first_non_filter(BlockDriverState
*candidate
)
5461 BlockDriverState
*bs
;
5462 BdrvNextIterator it
;
5464 /* walk down the bs forest recursively */
5465 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
5468 /* try to recurse in this top level bs */
5469 perm
= bdrv_recurse_is_first_non_filter(bs
, candidate
);
5471 /* candidate is the first non filter */
5473 bdrv_next_cleanup(&it
);
5481 BlockDriverState
*check_to_replace_node(BlockDriverState
*parent_bs
,
5482 const char *node_name
, Error
**errp
)
5484 BlockDriverState
*to_replace_bs
= bdrv_find_node(node_name
);
5485 AioContext
*aio_context
;
5487 if (!to_replace_bs
) {
5488 error_setg(errp
, "Node name '%s' not found", node_name
);
5492 aio_context
= bdrv_get_aio_context(to_replace_bs
);
5493 aio_context_acquire(aio_context
);
5495 if (bdrv_op_is_blocked(to_replace_bs
, BLOCK_OP_TYPE_REPLACE
, errp
)) {
5496 to_replace_bs
= NULL
;
5500 /* We don't want arbitrary node of the BDS chain to be replaced only the top
5501 * most non filter in order to prevent data corruption.
5502 * Another benefit is that this tests exclude backing files which are
5503 * blocked by the backing blockers.
5505 if (!bdrv_recurse_is_first_non_filter(parent_bs
, to_replace_bs
)) {
5506 error_setg(errp
, "Only top most non filter can be replaced");
5507 to_replace_bs
= NULL
;
5512 aio_context_release(aio_context
);
5513 return to_replace_bs
;
5516 static bool append_open_options(QDict
*d
, BlockDriverState
*bs
)
5518 const QDictEntry
*entry
;
5520 bool found_any
= false;
5522 for (entry
= qdict_first(bs
->options
); entry
;
5523 entry
= qdict_next(bs
->options
, entry
))
5525 /* Exclude all non-driver-specific options */
5526 for (desc
= bdrv_runtime_opts
.desc
; desc
->name
; desc
++) {
5527 if (!strcmp(qdict_entry_key(entry
), desc
->name
)) {
5535 qdict_put_obj(d
, qdict_entry_key(entry
),
5536 qobject_ref(qdict_entry_value(entry
)));
5543 /* Note: This function may return false positives; it may return true
5544 * even if opening the backing file specified by bs's image header
5545 * would result in exactly bs->backing. */
5546 static bool bdrv_backing_overridden(BlockDriverState
*bs
)
5549 return strcmp(bs
->auto_backing_file
,
5550 bs
->backing
->bs
->filename
);
5552 /* No backing BDS, so if the image header reports any backing
5553 * file, it must have been suppressed */
5554 return bs
->auto_backing_file
[0] != '\0';
5558 /* Updates the following BDS fields:
5559 * - exact_filename: A filename which may be used for opening a block device
5560 * which (mostly) equals the given BDS (even without any
5561 * other options; so reading and writing must return the same
5562 * results, but caching etc. may be different)
5563 * - full_open_options: Options which, when given when opening a block device
5564 * (without a filename), result in a BDS (mostly)
5565 * equalling the given one
5566 * - filename: If exact_filename is set, it is copied here. Otherwise,
5567 * full_open_options is converted to a JSON object, prefixed with
5568 * "json:" (for use through the JSON pseudo protocol) and put here.
5570 void bdrv_refresh_filename(BlockDriverState
*bs
)
5572 BlockDriver
*drv
= bs
->drv
;
5575 bool backing_overridden
;
5581 /* This BDS's file name may depend on any of its children's file names, so
5582 * refresh those first */
5583 QLIST_FOREACH(child
, &bs
->children
, next
) {
5584 bdrv_refresh_filename(child
->bs
);
5588 /* For implicit nodes, just copy everything from the single child */
5589 child
= QLIST_FIRST(&bs
->children
);
5590 assert(QLIST_NEXT(child
, next
) == NULL
);
5592 pstrcpy(bs
->exact_filename
, sizeof(bs
->exact_filename
),
5593 child
->bs
->exact_filename
);
5594 pstrcpy(bs
->filename
, sizeof(bs
->filename
), child
->bs
->filename
);
5596 bs
->full_open_options
= qobject_ref(child
->bs
->full_open_options
);
5601 backing_overridden
= bdrv_backing_overridden(bs
);
5603 if (bs
->open_flags
& BDRV_O_NO_IO
) {
5604 /* Without I/O, the backing file does not change anything.
5605 * Therefore, in such a case (primarily qemu-img), we can
5606 * pretend the backing file has not been overridden even if
5607 * it technically has been. */
5608 backing_overridden
= false;
5611 if (drv
->bdrv_refresh_filename
) {
5612 /* Obsolete information is of no use here, so drop the old file name
5613 * information before refreshing it */
5614 bs
->exact_filename
[0] = '\0';
5615 if (bs
->full_open_options
) {
5616 qobject_unref(bs
->full_open_options
);
5617 bs
->full_open_options
= NULL
;
5621 append_open_options(opts
, bs
);
5622 drv
->bdrv_refresh_filename(bs
, opts
);
5623 qobject_unref(opts
);
5624 } else if (bs
->file
) {
5625 /* Try to reconstruct valid information from the underlying file */
5626 bool has_open_options
;
5628 bs
->exact_filename
[0] = '\0';
5629 if (bs
->full_open_options
) {
5630 qobject_unref(bs
->full_open_options
);
5631 bs
->full_open_options
= NULL
;
5635 has_open_options
= append_open_options(opts
, bs
);
5636 has_open_options
|= backing_overridden
;
5638 /* If no specific options have been given for this BDS, the filename of
5639 * the underlying file should suffice for this one as well */
5640 if (bs
->file
->bs
->exact_filename
[0] && !has_open_options
) {
5641 strcpy(bs
->exact_filename
, bs
->file
->bs
->exact_filename
);
5643 /* Reconstructing the full options QDict is simple for most format block
5644 * drivers, as long as the full options are known for the underlying
5645 * file BDS. The full options QDict of that file BDS should somehow
5646 * contain a representation of the filename, therefore the following
5647 * suffices without querying the (exact_)filename of this BDS. */
5648 if (bs
->file
->bs
->full_open_options
&&
5649 (!bs
->backing
|| bs
->backing
->bs
->full_open_options
))
5651 qdict_put_str(opts
, "driver", drv
->format_name
);
5652 qdict_put(opts
, "file",
5653 qobject_ref(bs
->file
->bs
->full_open_options
));
5656 qdict_put(opts
, "backing",
5657 qobject_ref(bs
->backing
->bs
->full_open_options
));
5658 } else if (backing_overridden
) {
5659 qdict_put_null(opts
, "backing");
5662 bs
->full_open_options
= opts
;
5664 qobject_unref(opts
);
5666 } else if (!bs
->full_open_options
&& qdict_size(bs
->options
)) {
5667 /* There is no underlying file BDS (at least referenced by BDS.file),
5668 * so the full options QDict should be equal to the options given
5669 * specifically for this block device when it was opened (plus the
5670 * driver specification).
5671 * Because those options don't change, there is no need to update
5672 * full_open_options when it's already set. */
5675 append_open_options(opts
, bs
);
5676 qdict_put_str(opts
, "driver", drv
->format_name
);
5678 if (bs
->exact_filename
[0]) {
5679 /* This may not work for all block protocol drivers (some may
5680 * require this filename to be parsed), but we have to find some
5681 * default solution here, so just include it. If some block driver
5682 * does not support pure options without any filename at all or
5683 * needs some special format of the options QDict, it needs to
5684 * implement the driver-specific bdrv_refresh_filename() function.
5686 qdict_put_str(opts
, "filename", bs
->exact_filename
);
5689 bs
->full_open_options
= opts
;
5692 if (bs
->exact_filename
[0]) {
5693 pstrcpy(bs
->filename
, sizeof(bs
->filename
), bs
->exact_filename
);
5694 } else if (bs
->full_open_options
) {
5695 QString
*json
= qobject_to_json(QOBJECT(bs
->full_open_options
));
5696 snprintf(bs
->filename
, sizeof(bs
->filename
), "json:%s",
5697 qstring_get_str(json
));
5698 qobject_unref(json
);
5703 * Hot add/remove a BDS's child. So the user can take a child offline when
5704 * it is broken and take a new child online
5706 void bdrv_add_child(BlockDriverState
*parent_bs
, BlockDriverState
*child_bs
,
5710 if (!parent_bs
->drv
|| !parent_bs
->drv
->bdrv_add_child
) {
5711 error_setg(errp
, "The node %s does not support adding a child",
5712 bdrv_get_device_or_node_name(parent_bs
));
5716 if (!QLIST_EMPTY(&child_bs
->parents
)) {
5717 error_setg(errp
, "The node %s already has a parent",
5718 child_bs
->node_name
);
5722 parent_bs
->drv
->bdrv_add_child(parent_bs
, child_bs
, errp
);
5725 void bdrv_del_child(BlockDriverState
*parent_bs
, BdrvChild
*child
, Error
**errp
)
5729 if (!parent_bs
->drv
|| !parent_bs
->drv
->bdrv_del_child
) {
5730 error_setg(errp
, "The node %s does not support removing a child",
5731 bdrv_get_device_or_node_name(parent_bs
));
5735 QLIST_FOREACH(tmp
, &parent_bs
->children
, next
) {
5742 error_setg(errp
, "The node %s does not have a child named %s",
5743 bdrv_get_device_or_node_name(parent_bs
),
5744 bdrv_get_device_or_node_name(child
->bs
));
5748 parent_bs
->drv
->bdrv_del_child(parent_bs
, child
, errp
);
5751 bool bdrv_can_store_new_dirty_bitmap(BlockDriverState
*bs
, const char *name
,
5752 uint32_t granularity
, Error
**errp
)
5754 BlockDriver
*drv
= bs
->drv
;
5757 error_setg_errno(errp
, ENOMEDIUM
,
5758 "Can't store persistent bitmaps to %s",
5759 bdrv_get_device_or_node_name(bs
));
5763 if (!drv
->bdrv_can_store_new_dirty_bitmap
) {
5764 error_setg_errno(errp
, ENOTSUP
,
5765 "Can't store persistent bitmaps to %s",
5766 bdrv_get_device_or_node_name(bs
));
5770 return drv
->bdrv_can_store_new_dirty_bitmap(bs
, name
, granularity
, errp
);