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 return its duplicate. Otherwise, build a
156 path to it by considering it is relative to base_path. URL are
158 char *path_combine(const char *base_path
, const char *filename
)
160 const char *protocol_stripped
= NULL
;
165 if (path_is_absolute(filename
)) {
166 return g_strdup(filename
);
169 if (path_has_protocol(base_path
)) {
170 protocol_stripped
= strchr(base_path
, ':');
171 if (protocol_stripped
) {
175 p
= protocol_stripped
?: base_path
;
177 p1
= strrchr(base_path
, '/');
181 p2
= strrchr(base_path
, '\\');
182 if (!p1
|| p2
> p1
) {
197 result
= g_malloc(len
+ strlen(filename
) + 1);
198 memcpy(result
, base_path
, len
);
199 strcpy(result
+ len
, filename
);
204 static void path_combine_deprecated(char *dest
, int dest_size
,
205 const char *base_path
,
206 const char *filename
)
208 char *combined
= path_combine(base_path
, filename
);
209 pstrcpy(dest
, dest_size
, combined
);
214 * Helper function for bdrv_parse_filename() implementations to remove optional
215 * protocol prefixes (especially "file:") from a filename and for putting the
216 * stripped filename into the options QDict if there is such a prefix.
218 void bdrv_parse_filename_strip_prefix(const char *filename
, const char *prefix
,
221 if (strstart(filename
, prefix
, &filename
)) {
222 /* Stripping the explicit protocol prefix may result in a protocol
223 * prefix being (wrongly) detected (if the filename contains a colon) */
224 if (path_has_protocol(filename
)) {
225 QString
*fat_filename
;
227 /* This means there is some colon before the first slash; therefore,
228 * this cannot be an absolute path */
229 assert(!path_is_absolute(filename
));
231 /* And we can thus fix the protocol detection issue by prefixing it
233 fat_filename
= qstring_from_str("./");
234 qstring_append(fat_filename
, filename
);
236 assert(!path_has_protocol(qstring_get_str(fat_filename
)));
238 qdict_put(options
, "filename", fat_filename
);
240 /* If no protocol prefix was detected, we can use the shortened
242 qdict_put_str(options
, "filename", filename
);
248 /* Returns whether the image file is opened as read-only. Note that this can
249 * return false and writing to the image file is still not possible because the
250 * image is inactivated. */
251 bool bdrv_is_read_only(BlockDriverState
*bs
)
253 return bs
->read_only
;
256 int bdrv_can_set_read_only(BlockDriverState
*bs
, bool read_only
,
257 bool ignore_allow_rdw
, Error
**errp
)
259 /* Do not set read_only if copy_on_read is enabled */
260 if (bs
->copy_on_read
&& read_only
) {
261 error_setg(errp
, "Can't set node '%s' to r/o with copy-on-read enabled",
262 bdrv_get_device_or_node_name(bs
));
266 /* Do not clear read_only if it is prohibited */
267 if (!read_only
&& !(bs
->open_flags
& BDRV_O_ALLOW_RDWR
) &&
270 error_setg(errp
, "Node '%s' is read only",
271 bdrv_get_device_or_node_name(bs
));
279 * Called by a driver that can only provide a read-only image.
281 * Returns 0 if the node is already read-only or it could switch the node to
282 * read-only because BDRV_O_AUTO_RDONLY is set.
284 * Returns -EACCES if the node is read-write and BDRV_O_AUTO_RDONLY is not set
285 * or bdrv_can_set_read_only() forbids making the node read-only. If @errmsg
286 * is not NULL, it is used as the error message for the Error object.
288 int bdrv_apply_auto_read_only(BlockDriverState
*bs
, const char *errmsg
,
293 if (!(bs
->open_flags
& BDRV_O_RDWR
)) {
296 if (!(bs
->open_flags
& BDRV_O_AUTO_RDONLY
)) {
300 ret
= bdrv_can_set_read_only(bs
, true, false, NULL
);
305 bs
->read_only
= true;
306 bs
->open_flags
&= ~BDRV_O_RDWR
;
311 error_setg(errp
, "%s", errmsg
?: "Image is read-only");
315 void bdrv_get_full_backing_filename_from_filename(const char *backed
,
317 char *dest
, size_t sz
,
320 if (backing
[0] == '\0' || path_has_protocol(backing
) ||
321 path_is_absolute(backing
))
323 pstrcpy(dest
, sz
, backing
);
324 } else if (backed
[0] == '\0' || strstart(backed
, "json:", NULL
)) {
325 error_setg(errp
, "Cannot use relative backing file names for '%s'",
328 path_combine_deprecated(dest
, sz
, backed
, backing
);
332 void bdrv_get_full_backing_filename(BlockDriverState
*bs
, char *dest
, size_t sz
,
337 bdrv_refresh_filename(bs
);
339 backed
= bs
->exact_filename
[0] ? bs
->exact_filename
: bs
->filename
;
340 bdrv_get_full_backing_filename_from_filename(backed
, bs
->backing_file
,
344 void bdrv_register(BlockDriver
*bdrv
)
346 QLIST_INSERT_HEAD(&bdrv_drivers
, bdrv
, list
);
349 BlockDriverState
*bdrv_new(void)
351 BlockDriverState
*bs
;
354 bs
= g_new0(BlockDriverState
, 1);
355 QLIST_INIT(&bs
->dirty_bitmaps
);
356 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
357 QLIST_INIT(&bs
->op_blockers
[i
]);
359 notifier_with_return_list_init(&bs
->before_write_notifiers
);
360 qemu_co_mutex_init(&bs
->reqs_lock
);
361 qemu_mutex_init(&bs
->dirty_bitmap_mutex
);
363 bs
->aio_context
= qemu_get_aio_context();
365 qemu_co_queue_init(&bs
->flush_queue
);
367 for (i
= 0; i
< bdrv_drain_all_count
; i
++) {
368 bdrv_drained_begin(bs
);
371 QTAILQ_INSERT_TAIL(&all_bdrv_states
, bs
, bs_list
);
376 static BlockDriver
*bdrv_do_find_format(const char *format_name
)
380 QLIST_FOREACH(drv1
, &bdrv_drivers
, list
) {
381 if (!strcmp(drv1
->format_name
, format_name
)) {
389 BlockDriver
*bdrv_find_format(const char *format_name
)
394 drv1
= bdrv_do_find_format(format_name
);
399 /* The driver isn't registered, maybe we need to load a module */
400 for (i
= 0; i
< (int)ARRAY_SIZE(block_driver_modules
); ++i
) {
401 if (!strcmp(block_driver_modules
[i
].format_name
, format_name
)) {
402 block_module_load_one(block_driver_modules
[i
].library_name
);
407 return bdrv_do_find_format(format_name
);
410 int bdrv_is_whitelisted(BlockDriver
*drv
, bool read_only
)
412 static const char *whitelist_rw
[] = {
413 CONFIG_BDRV_RW_WHITELIST
415 static const char *whitelist_ro
[] = {
416 CONFIG_BDRV_RO_WHITELIST
420 if (!whitelist_rw
[0] && !whitelist_ro
[0]) {
421 return 1; /* no whitelist, anything goes */
424 for (p
= whitelist_rw
; *p
; p
++) {
425 if (!strcmp(drv
->format_name
, *p
)) {
430 for (p
= whitelist_ro
; *p
; p
++) {
431 if (!strcmp(drv
->format_name
, *p
)) {
439 bool bdrv_uses_whitelist(void)
441 return use_bdrv_whitelist
;
444 typedef struct CreateCo
{
452 static void coroutine_fn
bdrv_create_co_entry(void *opaque
)
454 Error
*local_err
= NULL
;
457 CreateCo
*cco
= opaque
;
460 ret
= cco
->drv
->bdrv_co_create_opts(cco
->filename
, cco
->opts
, &local_err
);
461 error_propagate(&cco
->err
, local_err
);
465 int bdrv_create(BlockDriver
*drv
, const char* filename
,
466 QemuOpts
*opts
, Error
**errp
)
473 .filename
= g_strdup(filename
),
479 if (!drv
->bdrv_co_create_opts
) {
480 error_setg(errp
, "Driver '%s' does not support image creation", drv
->format_name
);
485 if (qemu_in_coroutine()) {
486 /* Fast-path if already in coroutine context */
487 bdrv_create_co_entry(&cco
);
489 co
= qemu_coroutine_create(bdrv_create_co_entry
, &cco
);
490 qemu_coroutine_enter(co
);
491 while (cco
.ret
== NOT_DONE
) {
492 aio_poll(qemu_get_aio_context(), true);
499 error_propagate(errp
, cco
.err
);
501 error_setg_errno(errp
, -ret
, "Could not create image");
506 g_free(cco
.filename
);
510 int bdrv_create_file(const char *filename
, QemuOpts
*opts
, Error
**errp
)
513 Error
*local_err
= NULL
;
516 drv
= bdrv_find_protocol(filename
, true, errp
);
521 ret
= bdrv_create(drv
, filename
, opts
, &local_err
);
522 error_propagate(errp
, local_err
);
527 * Try to get @bs's logical and physical block size.
528 * On success, store them in @bsz struct and return 0.
529 * On failure return -errno.
530 * @bs must not be empty.
532 int bdrv_probe_blocksizes(BlockDriverState
*bs
, BlockSizes
*bsz
)
534 BlockDriver
*drv
= bs
->drv
;
536 if (drv
&& drv
->bdrv_probe_blocksizes
) {
537 return drv
->bdrv_probe_blocksizes(bs
, bsz
);
538 } else if (drv
&& drv
->is_filter
&& bs
->file
) {
539 return bdrv_probe_blocksizes(bs
->file
->bs
, bsz
);
546 * Try to get @bs's geometry (cyls, heads, sectors).
547 * On success, store them in @geo struct and return 0.
548 * On failure return -errno.
549 * @bs must not be empty.
551 int bdrv_probe_geometry(BlockDriverState
*bs
, HDGeometry
*geo
)
553 BlockDriver
*drv
= bs
->drv
;
555 if (drv
&& drv
->bdrv_probe_geometry
) {
556 return drv
->bdrv_probe_geometry(bs
, geo
);
557 } else if (drv
&& drv
->is_filter
&& bs
->file
) {
558 return bdrv_probe_geometry(bs
->file
->bs
, geo
);
565 * Create a uniquely-named empty temporary file.
566 * Return 0 upon success, otherwise a negative errno value.
568 int get_tmp_filename(char *filename
, int size
)
571 char temp_dir
[MAX_PATH
];
572 /* GetTempFileName requires that its output buffer (4th param)
573 have length MAX_PATH or greater. */
574 assert(size
>= MAX_PATH
);
575 return (GetTempPath(MAX_PATH
, temp_dir
)
576 && GetTempFileName(temp_dir
, "qem", 0, filename
)
577 ? 0 : -GetLastError());
581 tmpdir
= getenv("TMPDIR");
585 if (snprintf(filename
, size
, "%s/vl.XXXXXX", tmpdir
) >= size
) {
588 fd
= mkstemp(filename
);
592 if (close(fd
) != 0) {
601 * Detect host devices. By convention, /dev/cdrom[N] is always
602 * recognized as a host CDROM.
604 static BlockDriver
*find_hdev_driver(const char *filename
)
606 int score_max
= 0, score
;
607 BlockDriver
*drv
= NULL
, *d
;
609 QLIST_FOREACH(d
, &bdrv_drivers
, list
) {
610 if (d
->bdrv_probe_device
) {
611 score
= d
->bdrv_probe_device(filename
);
612 if (score
> score_max
) {
622 static BlockDriver
*bdrv_do_find_protocol(const char *protocol
)
626 QLIST_FOREACH(drv1
, &bdrv_drivers
, list
) {
627 if (drv1
->protocol_name
&& !strcmp(drv1
->protocol_name
, protocol
)) {
635 BlockDriver
*bdrv_find_protocol(const char *filename
,
636 bool allow_protocol_prefix
,
645 /* TODO Drivers without bdrv_file_open must be specified explicitly */
648 * XXX(hch): we really should not let host device detection
649 * override an explicit protocol specification, but moving this
650 * later breaks access to device names with colons in them.
651 * Thanks to the brain-dead persistent naming schemes on udev-
652 * based Linux systems those actually are quite common.
654 drv1
= find_hdev_driver(filename
);
659 if (!path_has_protocol(filename
) || !allow_protocol_prefix
) {
663 p
= strchr(filename
, ':');
666 if (len
> sizeof(protocol
) - 1)
667 len
= sizeof(protocol
) - 1;
668 memcpy(protocol
, filename
, len
);
669 protocol
[len
] = '\0';
671 drv1
= bdrv_do_find_protocol(protocol
);
676 for (i
= 0; i
< (int)ARRAY_SIZE(block_driver_modules
); ++i
) {
677 if (block_driver_modules
[i
].protocol_name
&&
678 !strcmp(block_driver_modules
[i
].protocol_name
, protocol
)) {
679 block_module_load_one(block_driver_modules
[i
].library_name
);
684 drv1
= bdrv_do_find_protocol(protocol
);
686 error_setg(errp
, "Unknown protocol '%s'", protocol
);
692 * Guess image format by probing its contents.
693 * This is not a good idea when your image is raw (CVE-2008-2004), but
694 * we do it anyway for backward compatibility.
696 * @buf contains the image's first @buf_size bytes.
697 * @buf_size is the buffer size in bytes (generally BLOCK_PROBE_BUF_SIZE,
698 * but can be smaller if the image file is smaller)
699 * @filename is its filename.
701 * For all block drivers, call the bdrv_probe() method to get its
703 * Return the first block driver with the highest probing score.
705 BlockDriver
*bdrv_probe_all(const uint8_t *buf
, int buf_size
,
706 const char *filename
)
708 int score_max
= 0, score
;
709 BlockDriver
*drv
= NULL
, *d
;
711 QLIST_FOREACH(d
, &bdrv_drivers
, list
) {
713 score
= d
->bdrv_probe(buf
, buf_size
, filename
);
714 if (score
> score_max
) {
724 static int find_image_format(BlockBackend
*file
, const char *filename
,
725 BlockDriver
**pdrv
, Error
**errp
)
728 uint8_t buf
[BLOCK_PROBE_BUF_SIZE
];
731 /* Return the raw BlockDriver * to scsi-generic devices or empty drives */
732 if (blk_is_sg(file
) || !blk_is_inserted(file
) || blk_getlength(file
) == 0) {
737 ret
= blk_pread(file
, 0, buf
, sizeof(buf
));
739 error_setg_errno(errp
, -ret
, "Could not read image for determining its "
745 drv
= bdrv_probe_all(buf
, ret
, filename
);
747 error_setg(errp
, "Could not determine image format: No compatible "
756 * Set the current 'total_sectors' value
757 * Return 0 on success, -errno on error.
759 int refresh_total_sectors(BlockDriverState
*bs
, int64_t hint
)
761 BlockDriver
*drv
= bs
->drv
;
767 /* Do not attempt drv->bdrv_getlength() on scsi-generic devices */
771 /* query actual device if possible, otherwise just trust the hint */
772 if (drv
->bdrv_getlength
) {
773 int64_t length
= drv
->bdrv_getlength(bs
);
777 hint
= DIV_ROUND_UP(length
, BDRV_SECTOR_SIZE
);
780 bs
->total_sectors
= hint
;
785 * Combines a QDict of new block driver @options with any missing options taken
786 * from @old_options, so that leaving out an option defaults to its old value.
788 static void bdrv_join_options(BlockDriverState
*bs
, QDict
*options
,
791 if (bs
->drv
&& bs
->drv
->bdrv_join_options
) {
792 bs
->drv
->bdrv_join_options(options
, old_options
);
794 qdict_join(options
, old_options
, false);
798 static BlockdevDetectZeroesOptions
bdrv_parse_detect_zeroes(QemuOpts
*opts
,
802 Error
*local_err
= NULL
;
803 char *value
= qemu_opt_get_del(opts
, "detect-zeroes");
804 BlockdevDetectZeroesOptions detect_zeroes
=
805 qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup
, value
,
806 BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF
, &local_err
);
809 error_propagate(errp
, local_err
);
810 return detect_zeroes
;
813 if (detect_zeroes
== BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP
&&
814 !(open_flags
& BDRV_O_UNMAP
))
816 error_setg(errp
, "setting detect-zeroes to unmap is not allowed "
817 "without setting discard operation to unmap");
820 return detect_zeroes
;
824 * Set open flags for a given discard mode
826 * Return 0 on success, -1 if the discard mode was invalid.
828 int bdrv_parse_discard_flags(const char *mode
, int *flags
)
830 *flags
&= ~BDRV_O_UNMAP
;
832 if (!strcmp(mode
, "off") || !strcmp(mode
, "ignore")) {
834 } else if (!strcmp(mode
, "on") || !strcmp(mode
, "unmap")) {
835 *flags
|= BDRV_O_UNMAP
;
844 * Set open flags for a given cache mode
846 * Return 0 on success, -1 if the cache mode was invalid.
848 int bdrv_parse_cache_mode(const char *mode
, int *flags
, bool *writethrough
)
850 *flags
&= ~BDRV_O_CACHE_MASK
;
852 if (!strcmp(mode
, "off") || !strcmp(mode
, "none")) {
853 *writethrough
= false;
854 *flags
|= BDRV_O_NOCACHE
;
855 } else if (!strcmp(mode
, "directsync")) {
856 *writethrough
= true;
857 *flags
|= BDRV_O_NOCACHE
;
858 } else if (!strcmp(mode
, "writeback")) {
859 *writethrough
= false;
860 } else if (!strcmp(mode
, "unsafe")) {
861 *writethrough
= false;
862 *flags
|= BDRV_O_NO_FLUSH
;
863 } else if (!strcmp(mode
, "writethrough")) {
864 *writethrough
= true;
872 static char *bdrv_child_get_parent_desc(BdrvChild
*c
)
874 BlockDriverState
*parent
= c
->opaque
;
875 return g_strdup(bdrv_get_device_or_node_name(parent
));
878 static void bdrv_child_cb_drained_begin(BdrvChild
*child
)
880 BlockDriverState
*bs
= child
->opaque
;
881 bdrv_do_drained_begin_quiesce(bs
, NULL
, false);
884 static bool bdrv_child_cb_drained_poll(BdrvChild
*child
)
886 BlockDriverState
*bs
= child
->opaque
;
887 return bdrv_drain_poll(bs
, false, NULL
, false);
890 static void bdrv_child_cb_drained_end(BdrvChild
*child
)
892 BlockDriverState
*bs
= child
->opaque
;
893 bdrv_drained_end(bs
);
896 static void bdrv_child_cb_attach(BdrvChild
*child
)
898 BlockDriverState
*bs
= child
->opaque
;
899 bdrv_apply_subtree_drain(child
, bs
);
902 static void bdrv_child_cb_detach(BdrvChild
*child
)
904 BlockDriverState
*bs
= child
->opaque
;
905 bdrv_unapply_subtree_drain(child
, bs
);
908 static int bdrv_child_cb_inactivate(BdrvChild
*child
)
910 BlockDriverState
*bs
= child
->opaque
;
911 assert(bs
->open_flags
& BDRV_O_INACTIVE
);
916 * Returns the options and flags that a temporary snapshot should get, based on
917 * the originally requested flags (the originally requested image will have
918 * flags like a backing file)
920 static void bdrv_temp_snapshot_options(int *child_flags
, QDict
*child_options
,
921 int parent_flags
, QDict
*parent_options
)
923 *child_flags
= (parent_flags
& ~BDRV_O_SNAPSHOT
) | BDRV_O_TEMPORARY
;
925 /* For temporary files, unconditional cache=unsafe is fine */
926 qdict_set_default_str(child_options
, BDRV_OPT_CACHE_DIRECT
, "off");
927 qdict_set_default_str(child_options
, BDRV_OPT_CACHE_NO_FLUSH
, "on");
929 /* Copy the read-only option from the parent */
930 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_READ_ONLY
);
932 /* aio=native doesn't work for cache.direct=off, so disable it for the
933 * temporary snapshot */
934 *child_flags
&= ~BDRV_O_NATIVE_AIO
;
938 * Returns the options and flags that bs->file should get if a protocol driver
939 * is expected, based on the given options and flags for the parent BDS
941 static void bdrv_inherited_options(int *child_flags
, QDict
*child_options
,
942 int parent_flags
, QDict
*parent_options
)
944 int flags
= parent_flags
;
946 /* Enable protocol handling, disable format probing for bs->file */
947 flags
|= BDRV_O_PROTOCOL
;
949 /* If the cache mode isn't explicitly set, inherit direct and no-flush from
951 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_CACHE_DIRECT
);
952 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_CACHE_NO_FLUSH
);
953 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_FORCE_SHARE
);
955 /* Inherit the read-only option from the parent if it's not set */
956 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_READ_ONLY
);
957 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_AUTO_READ_ONLY
);
959 /* Our block drivers take care to send flushes and respect unmap policy,
960 * so we can default to enable both on lower layers regardless of the
961 * corresponding parent options. */
962 qdict_set_default_str(child_options
, BDRV_OPT_DISCARD
, "unmap");
964 /* Clear flags that only apply to the top layer */
965 flags
&= ~(BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
| BDRV_O_COPY_ON_READ
|
968 *child_flags
= flags
;
971 const BdrvChildRole child_file
= {
972 .parent_is_bds
= true,
973 .get_parent_desc
= bdrv_child_get_parent_desc
,
974 .inherit_options
= bdrv_inherited_options
,
975 .drained_begin
= bdrv_child_cb_drained_begin
,
976 .drained_poll
= bdrv_child_cb_drained_poll
,
977 .drained_end
= bdrv_child_cb_drained_end
,
978 .attach
= bdrv_child_cb_attach
,
979 .detach
= bdrv_child_cb_detach
,
980 .inactivate
= bdrv_child_cb_inactivate
,
984 * Returns the options and flags that bs->file should get if the use of formats
985 * (and not only protocols) is permitted for it, based on the given options and
986 * flags for the parent BDS
988 static void bdrv_inherited_fmt_options(int *child_flags
, QDict
*child_options
,
989 int parent_flags
, QDict
*parent_options
)
991 child_file
.inherit_options(child_flags
, child_options
,
992 parent_flags
, parent_options
);
994 *child_flags
&= ~(BDRV_O_PROTOCOL
| BDRV_O_NO_IO
);
997 const BdrvChildRole child_format
= {
998 .parent_is_bds
= true,
999 .get_parent_desc
= bdrv_child_get_parent_desc
,
1000 .inherit_options
= bdrv_inherited_fmt_options
,
1001 .drained_begin
= bdrv_child_cb_drained_begin
,
1002 .drained_poll
= bdrv_child_cb_drained_poll
,
1003 .drained_end
= bdrv_child_cb_drained_end
,
1004 .attach
= bdrv_child_cb_attach
,
1005 .detach
= bdrv_child_cb_detach
,
1006 .inactivate
= bdrv_child_cb_inactivate
,
1009 static void bdrv_backing_attach(BdrvChild
*c
)
1011 BlockDriverState
*parent
= c
->opaque
;
1012 BlockDriverState
*backing_hd
= c
->bs
;
1014 assert(!parent
->backing_blocker
);
1015 error_setg(&parent
->backing_blocker
,
1016 "node is used as backing hd of '%s'",
1017 bdrv_get_device_or_node_name(parent
));
1019 bdrv_refresh_filename(backing_hd
);
1021 parent
->open_flags
&= ~BDRV_O_NO_BACKING
;
1022 pstrcpy(parent
->backing_file
, sizeof(parent
->backing_file
),
1023 backing_hd
->filename
);
1024 pstrcpy(parent
->backing_format
, sizeof(parent
->backing_format
),
1025 backing_hd
->drv
? backing_hd
->drv
->format_name
: "");
1027 bdrv_op_block_all(backing_hd
, parent
->backing_blocker
);
1028 /* Otherwise we won't be able to commit or stream */
1029 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_COMMIT_TARGET
,
1030 parent
->backing_blocker
);
1031 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_STREAM
,
1032 parent
->backing_blocker
);
1034 * We do backup in 3 ways:
1036 * The target bs is new opened, and the source is top BDS
1037 * 2. blockdev backup
1038 * Both the source and the target are top BDSes.
1039 * 3. internal backup(used for block replication)
1040 * Both the source and the target are backing file
1042 * In case 1 and 2, neither the source nor the target is the backing file.
1043 * In case 3, we will block the top BDS, so there is only one block job
1044 * for the top BDS and its backing chain.
1046 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_BACKUP_SOURCE
,
1047 parent
->backing_blocker
);
1048 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_BACKUP_TARGET
,
1049 parent
->backing_blocker
);
1051 bdrv_child_cb_attach(c
);
1054 static void bdrv_backing_detach(BdrvChild
*c
)
1056 BlockDriverState
*parent
= c
->opaque
;
1058 assert(parent
->backing_blocker
);
1059 bdrv_op_unblock_all(c
->bs
, parent
->backing_blocker
);
1060 error_free(parent
->backing_blocker
);
1061 parent
->backing_blocker
= NULL
;
1063 bdrv_child_cb_detach(c
);
1067 * Returns the options and flags that bs->backing should get, based on the
1068 * given options and flags for the parent BDS
1070 static void bdrv_backing_options(int *child_flags
, QDict
*child_options
,
1071 int parent_flags
, QDict
*parent_options
)
1073 int flags
= parent_flags
;
1075 /* The cache mode is inherited unmodified for backing files; except WCE,
1076 * which is only applied on the top level (BlockBackend) */
1077 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_CACHE_DIRECT
);
1078 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_CACHE_NO_FLUSH
);
1079 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_FORCE_SHARE
);
1081 /* backing files always opened read-only */
1082 qdict_set_default_str(child_options
, BDRV_OPT_READ_ONLY
, "on");
1083 qdict_set_default_str(child_options
, BDRV_OPT_AUTO_READ_ONLY
, "off");
1084 flags
&= ~BDRV_O_COPY_ON_READ
;
1086 /* snapshot=on is handled on the top layer */
1087 flags
&= ~(BDRV_O_SNAPSHOT
| BDRV_O_TEMPORARY
);
1089 *child_flags
= flags
;
1092 static int bdrv_backing_update_filename(BdrvChild
*c
, BlockDriverState
*base
,
1093 const char *filename
, Error
**errp
)
1095 BlockDriverState
*parent
= c
->opaque
;
1096 bool read_only
= bdrv_is_read_only(parent
);
1100 ret
= bdrv_reopen_set_read_only(parent
, false, errp
);
1106 ret
= bdrv_change_backing_file(parent
, filename
,
1107 base
->drv
? base
->drv
->format_name
: "");
1109 error_setg_errno(errp
, -ret
, "Could not update backing file link");
1113 bdrv_reopen_set_read_only(parent
, true, NULL
);
1119 const BdrvChildRole child_backing
= {
1120 .parent_is_bds
= true,
1121 .get_parent_desc
= bdrv_child_get_parent_desc
,
1122 .attach
= bdrv_backing_attach
,
1123 .detach
= bdrv_backing_detach
,
1124 .inherit_options
= bdrv_backing_options
,
1125 .drained_begin
= bdrv_child_cb_drained_begin
,
1126 .drained_poll
= bdrv_child_cb_drained_poll
,
1127 .drained_end
= bdrv_child_cb_drained_end
,
1128 .inactivate
= bdrv_child_cb_inactivate
,
1129 .update_filename
= bdrv_backing_update_filename
,
1132 static int bdrv_open_flags(BlockDriverState
*bs
, int flags
)
1134 int open_flags
= flags
;
1137 * Clear flags that are internal to the block layer before opening the
1140 open_flags
&= ~(BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
| BDRV_O_PROTOCOL
);
1143 * Snapshots should be writable.
1145 if (flags
& BDRV_O_TEMPORARY
) {
1146 open_flags
|= BDRV_O_RDWR
;
1152 static void update_flags_from_options(int *flags
, QemuOpts
*opts
)
1154 *flags
&= ~(BDRV_O_CACHE_MASK
| BDRV_O_RDWR
| BDRV_O_AUTO_RDONLY
);
1156 if (qemu_opt_get_bool_del(opts
, BDRV_OPT_CACHE_NO_FLUSH
, false)) {
1157 *flags
|= BDRV_O_NO_FLUSH
;
1160 if (qemu_opt_get_bool_del(opts
, BDRV_OPT_CACHE_DIRECT
, false)) {
1161 *flags
|= BDRV_O_NOCACHE
;
1164 if (!qemu_opt_get_bool_del(opts
, BDRV_OPT_READ_ONLY
, false)) {
1165 *flags
|= BDRV_O_RDWR
;
1168 if (qemu_opt_get_bool_del(opts
, BDRV_OPT_AUTO_READ_ONLY
, false)) {
1169 *flags
|= BDRV_O_AUTO_RDONLY
;
1173 static void update_options_from_flags(QDict
*options
, int flags
)
1175 if (!qdict_haskey(options
, BDRV_OPT_CACHE_DIRECT
)) {
1176 qdict_put_bool(options
, BDRV_OPT_CACHE_DIRECT
, flags
& BDRV_O_NOCACHE
);
1178 if (!qdict_haskey(options
, BDRV_OPT_CACHE_NO_FLUSH
)) {
1179 qdict_put_bool(options
, BDRV_OPT_CACHE_NO_FLUSH
,
1180 flags
& BDRV_O_NO_FLUSH
);
1182 if (!qdict_haskey(options
, BDRV_OPT_READ_ONLY
)) {
1183 qdict_put_bool(options
, BDRV_OPT_READ_ONLY
, !(flags
& BDRV_O_RDWR
));
1185 if (!qdict_haskey(options
, BDRV_OPT_AUTO_READ_ONLY
)) {
1186 qdict_put_bool(options
, BDRV_OPT_AUTO_READ_ONLY
,
1187 flags
& BDRV_O_AUTO_RDONLY
);
1191 static void bdrv_assign_node_name(BlockDriverState
*bs
,
1192 const char *node_name
,
1195 char *gen_node_name
= NULL
;
1198 node_name
= gen_node_name
= id_generate(ID_BLOCK
);
1199 } else if (!id_wellformed(node_name
)) {
1201 * Check for empty string or invalid characters, but not if it is
1202 * generated (generated names use characters not available to the user)
1204 error_setg(errp
, "Invalid node name");
1208 /* takes care of avoiding namespaces collisions */
1209 if (blk_by_name(node_name
)) {
1210 error_setg(errp
, "node-name=%s is conflicting with a device id",
1215 /* takes care of avoiding duplicates node names */
1216 if (bdrv_find_node(node_name
)) {
1217 error_setg(errp
, "Duplicate node name");
1221 /* Make sure that the node name isn't truncated */
1222 if (strlen(node_name
) >= sizeof(bs
->node_name
)) {
1223 error_setg(errp
, "Node name too long");
1227 /* copy node name into the bs and insert it into the graph list */
1228 pstrcpy(bs
->node_name
, sizeof(bs
->node_name
), node_name
);
1229 QTAILQ_INSERT_TAIL(&graph_bdrv_states
, bs
, node_list
);
1231 g_free(gen_node_name
);
1234 static int bdrv_open_driver(BlockDriverState
*bs
, BlockDriver
*drv
,
1235 const char *node_name
, QDict
*options
,
1236 int open_flags
, Error
**errp
)
1238 Error
*local_err
= NULL
;
1241 bdrv_assign_node_name(bs
, node_name
, &local_err
);
1243 error_propagate(errp
, local_err
);
1248 bs
->read_only
= !(bs
->open_flags
& BDRV_O_RDWR
);
1249 bs
->opaque
= g_malloc0(drv
->instance_size
);
1251 if (drv
->bdrv_file_open
) {
1252 assert(!drv
->bdrv_needs_filename
|| bs
->filename
[0]);
1253 ret
= drv
->bdrv_file_open(bs
, options
, open_flags
, &local_err
);
1254 } else if (drv
->bdrv_open
) {
1255 ret
= drv
->bdrv_open(bs
, options
, open_flags
, &local_err
);
1262 error_propagate(errp
, local_err
);
1263 } else if (bs
->filename
[0]) {
1264 error_setg_errno(errp
, -ret
, "Could not open '%s'", bs
->filename
);
1266 error_setg_errno(errp
, -ret
, "Could not open image");
1271 ret
= refresh_total_sectors(bs
, bs
->total_sectors
);
1273 error_setg_errno(errp
, -ret
, "Could not refresh total sector count");
1277 bdrv_refresh_limits(bs
, &local_err
);
1279 error_propagate(errp
, local_err
);
1283 assert(bdrv_opt_mem_align(bs
) != 0);
1284 assert(bdrv_min_mem_align(bs
) != 0);
1285 assert(is_power_of_2(bs
->bl
.request_alignment
));
1287 for (i
= 0; i
< bs
->quiesce_counter
; i
++) {
1288 if (drv
->bdrv_co_drain_begin
) {
1289 drv
->bdrv_co_drain_begin(bs
);
1296 if (bs
->file
!= NULL
) {
1297 bdrv_unref_child(bs
, bs
->file
);
1305 BlockDriverState
*bdrv_new_open_driver(BlockDriver
*drv
, const char *node_name
,
1306 int flags
, Error
**errp
)
1308 BlockDriverState
*bs
;
1312 bs
->open_flags
= flags
;
1313 bs
->explicit_options
= qdict_new();
1314 bs
->options
= qdict_new();
1317 update_options_from_flags(bs
->options
, flags
);
1319 ret
= bdrv_open_driver(bs
, drv
, node_name
, bs
->options
, flags
, errp
);
1321 qobject_unref(bs
->explicit_options
);
1322 bs
->explicit_options
= NULL
;
1323 qobject_unref(bs
->options
);
1332 QemuOptsList bdrv_runtime_opts
= {
1333 .name
= "bdrv_common",
1334 .head
= QTAILQ_HEAD_INITIALIZER(bdrv_runtime_opts
.head
),
1337 .name
= "node-name",
1338 .type
= QEMU_OPT_STRING
,
1339 .help
= "Node name of the block device node",
1343 .type
= QEMU_OPT_STRING
,
1344 .help
= "Block driver to use for the node",
1347 .name
= BDRV_OPT_CACHE_DIRECT
,
1348 .type
= QEMU_OPT_BOOL
,
1349 .help
= "Bypass software writeback cache on the host",
1352 .name
= BDRV_OPT_CACHE_NO_FLUSH
,
1353 .type
= QEMU_OPT_BOOL
,
1354 .help
= "Ignore flush requests",
1357 .name
= BDRV_OPT_READ_ONLY
,
1358 .type
= QEMU_OPT_BOOL
,
1359 .help
= "Node is opened in read-only mode",
1362 .name
= BDRV_OPT_AUTO_READ_ONLY
,
1363 .type
= QEMU_OPT_BOOL
,
1364 .help
= "Node can become read-only if opening read-write fails",
1367 .name
= "detect-zeroes",
1368 .type
= QEMU_OPT_STRING
,
1369 .help
= "try to optimize zero writes (off, on, unmap)",
1372 .name
= BDRV_OPT_DISCARD
,
1373 .type
= QEMU_OPT_STRING
,
1374 .help
= "discard operation (ignore/off, unmap/on)",
1377 .name
= BDRV_OPT_FORCE_SHARE
,
1378 .type
= QEMU_OPT_BOOL
,
1379 .help
= "always accept other writers (default: off)",
1381 { /* end of list */ }
1386 * Common part for opening disk images and files
1388 * Removes all processed options from *options.
1390 static int bdrv_open_common(BlockDriverState
*bs
, BlockBackend
*file
,
1391 QDict
*options
, Error
**errp
)
1393 int ret
, open_flags
;
1394 const char *filename
;
1395 const char *driver_name
= NULL
;
1396 const char *node_name
= NULL
;
1397 const char *discard
;
1400 Error
*local_err
= NULL
;
1402 assert(bs
->file
== NULL
);
1403 assert(options
!= NULL
&& bs
->options
!= options
);
1405 opts
= qemu_opts_create(&bdrv_runtime_opts
, NULL
, 0, &error_abort
);
1406 qemu_opts_absorb_qdict(opts
, options
, &local_err
);
1408 error_propagate(errp
, local_err
);
1413 update_flags_from_options(&bs
->open_flags
, opts
);
1415 driver_name
= qemu_opt_get(opts
, "driver");
1416 drv
= bdrv_find_format(driver_name
);
1417 assert(drv
!= NULL
);
1419 bs
->force_share
= qemu_opt_get_bool(opts
, BDRV_OPT_FORCE_SHARE
, false);
1421 if (bs
->force_share
&& (bs
->open_flags
& BDRV_O_RDWR
)) {
1423 BDRV_OPT_FORCE_SHARE
1424 "=on can only be used with read-only images");
1430 bdrv_refresh_filename(blk_bs(file
));
1431 filename
= blk_bs(file
)->filename
;
1434 * Caution: while qdict_get_try_str() is fine, getting
1435 * non-string types would require more care. When @options
1436 * come from -blockdev or blockdev_add, its members are typed
1437 * according to the QAPI schema, but when they come from
1438 * -drive, they're all QString.
1440 filename
= qdict_get_try_str(options
, "filename");
1443 if (drv
->bdrv_needs_filename
&& (!filename
|| !filename
[0])) {
1444 error_setg(errp
, "The '%s' block driver requires a file name",
1450 trace_bdrv_open_common(bs
, filename
?: "", bs
->open_flags
,
1453 bs
->read_only
= !(bs
->open_flags
& BDRV_O_RDWR
);
1455 if (use_bdrv_whitelist
&& !bdrv_is_whitelisted(drv
, bs
->read_only
)) {
1456 if (!bs
->read_only
&& bdrv_is_whitelisted(drv
, true)) {
1457 ret
= bdrv_apply_auto_read_only(bs
, NULL
, NULL
);
1463 !bs
->read_only
&& bdrv_is_whitelisted(drv
, true)
1464 ? "Driver '%s' can only be used for read-only devices"
1465 : "Driver '%s' is not whitelisted",
1471 /* bdrv_new() and bdrv_close() make it so */
1472 assert(atomic_read(&bs
->copy_on_read
) == 0);
1474 if (bs
->open_flags
& BDRV_O_COPY_ON_READ
) {
1475 if (!bs
->read_only
) {
1476 bdrv_enable_copy_on_read(bs
);
1478 error_setg(errp
, "Can't use copy-on-read on read-only device");
1484 discard
= qemu_opt_get(opts
, BDRV_OPT_DISCARD
);
1485 if (discard
!= NULL
) {
1486 if (bdrv_parse_discard_flags(discard
, &bs
->open_flags
) != 0) {
1487 error_setg(errp
, "Invalid discard option");
1494 bdrv_parse_detect_zeroes(opts
, bs
->open_flags
, &local_err
);
1496 error_propagate(errp
, local_err
);
1501 if (filename
!= NULL
) {
1502 pstrcpy(bs
->filename
, sizeof(bs
->filename
), filename
);
1504 bs
->filename
[0] = '\0';
1506 pstrcpy(bs
->exact_filename
, sizeof(bs
->exact_filename
), bs
->filename
);
1508 /* Open the image, either directly or using a protocol */
1509 open_flags
= bdrv_open_flags(bs
, bs
->open_flags
);
1510 node_name
= qemu_opt_get(opts
, "node-name");
1512 assert(!drv
->bdrv_file_open
|| file
== NULL
);
1513 ret
= bdrv_open_driver(bs
, drv
, node_name
, options
, open_flags
, errp
);
1518 qemu_opts_del(opts
);
1522 qemu_opts_del(opts
);
1526 static QDict
*parse_json_filename(const char *filename
, Error
**errp
)
1528 QObject
*options_obj
;
1532 ret
= strstart(filename
, "json:", &filename
);
1535 options_obj
= qobject_from_json(filename
, errp
);
1537 error_prepend(errp
, "Could not parse the JSON options: ");
1541 options
= qobject_to(QDict
, options_obj
);
1543 qobject_unref(options_obj
);
1544 error_setg(errp
, "Invalid JSON object given");
1548 qdict_flatten(options
);
1553 static void parse_json_protocol(QDict
*options
, const char **pfilename
,
1556 QDict
*json_options
;
1557 Error
*local_err
= NULL
;
1559 /* Parse json: pseudo-protocol */
1560 if (!*pfilename
|| !g_str_has_prefix(*pfilename
, "json:")) {
1564 json_options
= parse_json_filename(*pfilename
, &local_err
);
1566 error_propagate(errp
, local_err
);
1570 /* Options given in the filename have lower priority than options
1571 * specified directly */
1572 qdict_join(options
, json_options
, false);
1573 qobject_unref(json_options
);
1578 * Fills in default options for opening images and converts the legacy
1579 * filename/flags pair to option QDict entries.
1580 * The BDRV_O_PROTOCOL flag in *flags will be set or cleared accordingly if a
1581 * block driver has been specified explicitly.
1583 static int bdrv_fill_options(QDict
**options
, const char *filename
,
1584 int *flags
, Error
**errp
)
1586 const char *drvname
;
1587 bool protocol
= *flags
& BDRV_O_PROTOCOL
;
1588 bool parse_filename
= false;
1589 BlockDriver
*drv
= NULL
;
1590 Error
*local_err
= NULL
;
1593 * Caution: while qdict_get_try_str() is fine, getting non-string
1594 * types would require more care. When @options come from
1595 * -blockdev or blockdev_add, its members are typed according to
1596 * the QAPI schema, but when they come from -drive, they're all
1599 drvname
= qdict_get_try_str(*options
, "driver");
1601 drv
= bdrv_find_format(drvname
);
1603 error_setg(errp
, "Unknown driver '%s'", drvname
);
1606 /* If the user has explicitly specified the driver, this choice should
1607 * override the BDRV_O_PROTOCOL flag */
1608 protocol
= drv
->bdrv_file_open
;
1612 *flags
|= BDRV_O_PROTOCOL
;
1614 *flags
&= ~BDRV_O_PROTOCOL
;
1617 /* Translate cache options from flags into options */
1618 update_options_from_flags(*options
, *flags
);
1620 /* Fetch the file name from the options QDict if necessary */
1621 if (protocol
&& filename
) {
1622 if (!qdict_haskey(*options
, "filename")) {
1623 qdict_put_str(*options
, "filename", filename
);
1624 parse_filename
= true;
1626 error_setg(errp
, "Can't specify 'file' and 'filename' options at "
1632 /* Find the right block driver */
1633 /* See cautionary note on accessing @options above */
1634 filename
= qdict_get_try_str(*options
, "filename");
1636 if (!drvname
&& protocol
) {
1638 drv
= bdrv_find_protocol(filename
, parse_filename
, errp
);
1643 drvname
= drv
->format_name
;
1644 qdict_put_str(*options
, "driver", drvname
);
1646 error_setg(errp
, "Must specify either driver or file");
1651 assert(drv
|| !protocol
);
1653 /* Driver-specific filename parsing */
1654 if (drv
&& drv
->bdrv_parse_filename
&& parse_filename
) {
1655 drv
->bdrv_parse_filename(filename
, *options
, &local_err
);
1657 error_propagate(errp
, local_err
);
1661 if (!drv
->bdrv_needs_filename
) {
1662 qdict_del(*options
, "filename");
1669 static int bdrv_child_check_perm(BdrvChild
*c
, BlockReopenQueue
*q
,
1670 uint64_t perm
, uint64_t shared
,
1671 GSList
*ignore_children
, Error
**errp
);
1672 static void bdrv_child_abort_perm_update(BdrvChild
*c
);
1673 static void bdrv_child_set_perm(BdrvChild
*c
, uint64_t perm
, uint64_t shared
);
1675 typedef struct BlockReopenQueueEntry
{
1677 BDRVReopenState state
;
1678 QSIMPLEQ_ENTRY(BlockReopenQueueEntry
) entry
;
1679 } BlockReopenQueueEntry
;
1682 * Return the flags that @bs will have after the reopens in @q have
1683 * successfully completed. If @q is NULL (or @bs is not contained in @q),
1684 * return the current flags.
1686 static int bdrv_reopen_get_flags(BlockReopenQueue
*q
, BlockDriverState
*bs
)
1688 BlockReopenQueueEntry
*entry
;
1691 QSIMPLEQ_FOREACH(entry
, q
, entry
) {
1692 if (entry
->state
.bs
== bs
) {
1693 return entry
->state
.flags
;
1698 return bs
->open_flags
;
1701 /* Returns whether the image file can be written to after the reopen queue @q
1702 * has been successfully applied, or right now if @q is NULL. */
1703 static bool bdrv_is_writable_after_reopen(BlockDriverState
*bs
,
1704 BlockReopenQueue
*q
)
1706 int flags
= bdrv_reopen_get_flags(q
, bs
);
1708 return (flags
& (BDRV_O_RDWR
| BDRV_O_INACTIVE
)) == BDRV_O_RDWR
;
1712 * Return whether the BDS can be written to. This is not necessarily
1713 * the same as !bdrv_is_read_only(bs), as inactivated images may not
1714 * be written to but do not count as read-only images.
1716 bool bdrv_is_writable(BlockDriverState
*bs
)
1718 return bdrv_is_writable_after_reopen(bs
, NULL
);
1721 static void bdrv_child_perm(BlockDriverState
*bs
, BlockDriverState
*child_bs
,
1722 BdrvChild
*c
, const BdrvChildRole
*role
,
1723 BlockReopenQueue
*reopen_queue
,
1724 uint64_t parent_perm
, uint64_t parent_shared
,
1725 uint64_t *nperm
, uint64_t *nshared
)
1727 if (bs
->drv
&& bs
->drv
->bdrv_child_perm
) {
1728 bs
->drv
->bdrv_child_perm(bs
, c
, role
, reopen_queue
,
1729 parent_perm
, parent_shared
,
1732 /* TODO Take force_share from reopen_queue */
1733 if (child_bs
&& child_bs
->force_share
) {
1734 *nshared
= BLK_PERM_ALL
;
1739 * Check whether permissions on this node can be changed in a way that
1740 * @cumulative_perms and @cumulative_shared_perms are the new cumulative
1741 * permissions of all its parents. This involves checking whether all necessary
1742 * permission changes to child nodes can be performed.
1744 * A call to this function must always be followed by a call to bdrv_set_perm()
1745 * or bdrv_abort_perm_update().
1747 static int bdrv_check_perm(BlockDriverState
*bs
, BlockReopenQueue
*q
,
1748 uint64_t cumulative_perms
,
1749 uint64_t cumulative_shared_perms
,
1750 GSList
*ignore_children
, Error
**errp
)
1752 BlockDriver
*drv
= bs
->drv
;
1756 /* Write permissions never work with read-only images */
1757 if ((cumulative_perms
& (BLK_PERM_WRITE
| BLK_PERM_WRITE_UNCHANGED
)) &&
1758 !bdrv_is_writable_after_reopen(bs
, q
))
1760 error_setg(errp
, "Block node is read-only");
1764 /* Check this node */
1769 if (drv
->bdrv_check_perm
) {
1770 return drv
->bdrv_check_perm(bs
, cumulative_perms
,
1771 cumulative_shared_perms
, errp
);
1774 /* Drivers that never have children can omit .bdrv_child_perm() */
1775 if (!drv
->bdrv_child_perm
) {
1776 assert(QLIST_EMPTY(&bs
->children
));
1780 /* Check all children */
1781 QLIST_FOREACH(c
, &bs
->children
, next
) {
1782 uint64_t cur_perm
, cur_shared
;
1783 bdrv_child_perm(bs
, c
->bs
, c
, c
->role
, q
,
1784 cumulative_perms
, cumulative_shared_perms
,
1785 &cur_perm
, &cur_shared
);
1786 ret
= bdrv_child_check_perm(c
, q
, cur_perm
, cur_shared
,
1787 ignore_children
, errp
);
1797 * Notifies drivers that after a previous bdrv_check_perm() call, the
1798 * permission update is not performed and any preparations made for it (e.g.
1799 * taken file locks) need to be undone.
1801 * This function recursively notifies all child nodes.
1803 static void bdrv_abort_perm_update(BlockDriverState
*bs
)
1805 BlockDriver
*drv
= bs
->drv
;
1812 if (drv
->bdrv_abort_perm_update
) {
1813 drv
->bdrv_abort_perm_update(bs
);
1816 QLIST_FOREACH(c
, &bs
->children
, next
) {
1817 bdrv_child_abort_perm_update(c
);
1821 static void bdrv_set_perm(BlockDriverState
*bs
, uint64_t cumulative_perms
,
1822 uint64_t cumulative_shared_perms
)
1824 BlockDriver
*drv
= bs
->drv
;
1831 /* Update this node */
1832 if (drv
->bdrv_set_perm
) {
1833 drv
->bdrv_set_perm(bs
, cumulative_perms
, cumulative_shared_perms
);
1836 /* Drivers that never have children can omit .bdrv_child_perm() */
1837 if (!drv
->bdrv_child_perm
) {
1838 assert(QLIST_EMPTY(&bs
->children
));
1842 /* Update all children */
1843 QLIST_FOREACH(c
, &bs
->children
, next
) {
1844 uint64_t cur_perm
, cur_shared
;
1845 bdrv_child_perm(bs
, c
->bs
, c
, c
->role
, NULL
,
1846 cumulative_perms
, cumulative_shared_perms
,
1847 &cur_perm
, &cur_shared
);
1848 bdrv_child_set_perm(c
, cur_perm
, cur_shared
);
1852 static void bdrv_get_cumulative_perm(BlockDriverState
*bs
, uint64_t *perm
,
1853 uint64_t *shared_perm
)
1856 uint64_t cumulative_perms
= 0;
1857 uint64_t cumulative_shared_perms
= BLK_PERM_ALL
;
1859 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
1860 cumulative_perms
|= c
->perm
;
1861 cumulative_shared_perms
&= c
->shared_perm
;
1864 *perm
= cumulative_perms
;
1865 *shared_perm
= cumulative_shared_perms
;
1868 static char *bdrv_child_user_desc(BdrvChild
*c
)
1870 if (c
->role
->get_parent_desc
) {
1871 return c
->role
->get_parent_desc(c
);
1874 return g_strdup("another user");
1877 char *bdrv_perm_names(uint64_t perm
)
1883 { BLK_PERM_CONSISTENT_READ
, "consistent read" },
1884 { BLK_PERM_WRITE
, "write" },
1885 { BLK_PERM_WRITE_UNCHANGED
, "write unchanged" },
1886 { BLK_PERM_RESIZE
, "resize" },
1887 { BLK_PERM_GRAPH_MOD
, "change children" },
1891 char *result
= g_strdup("");
1892 struct perm_name
*p
;
1894 for (p
= permissions
; p
->name
; p
++) {
1895 if (perm
& p
->perm
) {
1897 result
= g_strdup_printf("%s%s%s", old
, *old
? ", " : "", p
->name
);
1906 * Checks whether a new reference to @bs can be added if the new user requires
1907 * @new_used_perm/@new_shared_perm as its permissions. If @ignore_children is
1908 * set, the BdrvChild objects in this list are ignored in the calculations;
1909 * this allows checking permission updates for an existing reference.
1911 * Needs to be followed by a call to either bdrv_set_perm() or
1912 * bdrv_abort_perm_update(). */
1913 static int bdrv_check_update_perm(BlockDriverState
*bs
, BlockReopenQueue
*q
,
1914 uint64_t new_used_perm
,
1915 uint64_t new_shared_perm
,
1916 GSList
*ignore_children
, Error
**errp
)
1919 uint64_t cumulative_perms
= new_used_perm
;
1920 uint64_t cumulative_shared_perms
= new_shared_perm
;
1922 /* There is no reason why anyone couldn't tolerate write_unchanged */
1923 assert(new_shared_perm
& BLK_PERM_WRITE_UNCHANGED
);
1925 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
1926 if (g_slist_find(ignore_children
, c
)) {
1930 if ((new_used_perm
& c
->shared_perm
) != new_used_perm
) {
1931 char *user
= bdrv_child_user_desc(c
);
1932 char *perm_names
= bdrv_perm_names(new_used_perm
& ~c
->shared_perm
);
1933 error_setg(errp
, "Conflicts with use by %s as '%s', which does not "
1935 user
, c
->name
, perm_names
, bdrv_get_node_name(c
->bs
));
1941 if ((c
->perm
& new_shared_perm
) != c
->perm
) {
1942 char *user
= bdrv_child_user_desc(c
);
1943 char *perm_names
= bdrv_perm_names(c
->perm
& ~new_shared_perm
);
1944 error_setg(errp
, "Conflicts with use by %s as '%s', which uses "
1946 user
, c
->name
, perm_names
, bdrv_get_node_name(c
->bs
));
1952 cumulative_perms
|= c
->perm
;
1953 cumulative_shared_perms
&= c
->shared_perm
;
1956 return bdrv_check_perm(bs
, q
, cumulative_perms
, cumulative_shared_perms
,
1957 ignore_children
, errp
);
1960 /* Needs to be followed by a call to either bdrv_child_set_perm() or
1961 * bdrv_child_abort_perm_update(). */
1962 static int bdrv_child_check_perm(BdrvChild
*c
, BlockReopenQueue
*q
,
1963 uint64_t perm
, uint64_t shared
,
1964 GSList
*ignore_children
, Error
**errp
)
1968 ignore_children
= g_slist_prepend(g_slist_copy(ignore_children
), c
);
1969 ret
= bdrv_check_update_perm(c
->bs
, q
, perm
, shared
, ignore_children
, errp
);
1970 g_slist_free(ignore_children
);
1976 if (!c
->has_backup_perm
) {
1977 c
->has_backup_perm
= true;
1978 c
->backup_perm
= c
->perm
;
1979 c
->backup_shared_perm
= c
->shared_perm
;
1982 * Note: it's OK if c->has_backup_perm was already set, as we can find the
1983 * same child twice during check_perm procedure
1987 c
->shared_perm
= shared
;
1992 static void bdrv_child_set_perm(BdrvChild
*c
, uint64_t perm
, uint64_t shared
)
1994 uint64_t cumulative_perms
, cumulative_shared_perms
;
1996 c
->has_backup_perm
= false;
1999 c
->shared_perm
= shared
;
2001 bdrv_get_cumulative_perm(c
->bs
, &cumulative_perms
,
2002 &cumulative_shared_perms
);
2003 bdrv_set_perm(c
->bs
, cumulative_perms
, cumulative_shared_perms
);
2006 static void bdrv_child_abort_perm_update(BdrvChild
*c
)
2008 if (c
->has_backup_perm
) {
2009 c
->perm
= c
->backup_perm
;
2010 c
->shared_perm
= c
->backup_shared_perm
;
2011 c
->has_backup_perm
= false;
2014 bdrv_abort_perm_update(c
->bs
);
2017 int bdrv_child_try_set_perm(BdrvChild
*c
, uint64_t perm
, uint64_t shared
,
2022 ret
= bdrv_child_check_perm(c
, NULL
, perm
, shared
, NULL
, errp
);
2024 bdrv_child_abort_perm_update(c
);
2028 bdrv_child_set_perm(c
, perm
, shared
);
2033 void bdrv_filter_default_perms(BlockDriverState
*bs
, BdrvChild
*c
,
2034 const BdrvChildRole
*role
,
2035 BlockReopenQueue
*reopen_queue
,
2036 uint64_t perm
, uint64_t shared
,
2037 uint64_t *nperm
, uint64_t *nshared
)
2040 *nperm
= perm
& DEFAULT_PERM_PASSTHROUGH
;
2041 *nshared
= (shared
& DEFAULT_PERM_PASSTHROUGH
) | DEFAULT_PERM_UNCHANGED
;
2045 *nperm
= (perm
& DEFAULT_PERM_PASSTHROUGH
) |
2046 (c
->perm
& DEFAULT_PERM_UNCHANGED
);
2047 *nshared
= (shared
& DEFAULT_PERM_PASSTHROUGH
) |
2048 (c
->shared_perm
& DEFAULT_PERM_UNCHANGED
);
2051 void bdrv_format_default_perms(BlockDriverState
*bs
, BdrvChild
*c
,
2052 const BdrvChildRole
*role
,
2053 BlockReopenQueue
*reopen_queue
,
2054 uint64_t perm
, uint64_t shared
,
2055 uint64_t *nperm
, uint64_t *nshared
)
2057 bool backing
= (role
== &child_backing
);
2058 assert(role
== &child_backing
|| role
== &child_file
);
2061 int flags
= bdrv_reopen_get_flags(reopen_queue
, bs
);
2063 /* Apart from the modifications below, the same permissions are
2064 * forwarded and left alone as for filters */
2065 bdrv_filter_default_perms(bs
, c
, role
, reopen_queue
, perm
, shared
,
2068 /* Format drivers may touch metadata even if the guest doesn't write */
2069 if (bdrv_is_writable_after_reopen(bs
, reopen_queue
)) {
2070 perm
|= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2073 /* bs->file always needs to be consistent because of the metadata. We
2074 * can never allow other users to resize or write to it. */
2075 if (!(flags
& BDRV_O_NO_IO
)) {
2076 perm
|= BLK_PERM_CONSISTENT_READ
;
2078 shared
&= ~(BLK_PERM_WRITE
| BLK_PERM_RESIZE
);
2080 /* We want consistent read from backing files if the parent needs it.
2081 * No other operations are performed on backing files. */
2082 perm
&= BLK_PERM_CONSISTENT_READ
;
2084 /* If the parent can deal with changing data, we're okay with a
2085 * writable and resizable backing file. */
2086 /* TODO Require !(perm & BLK_PERM_CONSISTENT_READ), too? */
2087 if (shared
& BLK_PERM_WRITE
) {
2088 shared
= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2093 shared
|= BLK_PERM_CONSISTENT_READ
| BLK_PERM_GRAPH_MOD
|
2094 BLK_PERM_WRITE_UNCHANGED
;
2097 if (bs
->open_flags
& BDRV_O_INACTIVE
) {
2098 shared
|= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2105 static void bdrv_replace_child_noperm(BdrvChild
*child
,
2106 BlockDriverState
*new_bs
)
2108 BlockDriverState
*old_bs
= child
->bs
;
2111 if (old_bs
&& new_bs
) {
2112 assert(bdrv_get_aio_context(old_bs
) == bdrv_get_aio_context(new_bs
));
2115 /* Detach first so that the recursive drain sections coming from @child
2116 * are already gone and we only end the drain sections that came from
2118 if (child
->role
->detach
) {
2119 child
->role
->detach(child
);
2121 if (old_bs
->quiesce_counter
&& child
->role
->drained_end
) {
2122 int num
= old_bs
->quiesce_counter
;
2123 if (child
->role
->parent_is_bds
) {
2124 num
-= bdrv_drain_all_count
;
2127 for (i
= 0; i
< num
; i
++) {
2128 child
->role
->drained_end(child
);
2131 QLIST_REMOVE(child
, next_parent
);
2137 QLIST_INSERT_HEAD(&new_bs
->parents
, child
, next_parent
);
2138 if (new_bs
->quiesce_counter
&& child
->role
->drained_begin
) {
2139 int num
= new_bs
->quiesce_counter
;
2140 if (child
->role
->parent_is_bds
) {
2141 num
-= bdrv_drain_all_count
;
2144 for (i
= 0; i
< num
; i
++) {
2145 bdrv_parent_drained_begin_single(child
, true);
2149 /* Attach only after starting new drained sections, so that recursive
2150 * drain sections coming from @child don't get an extra .drained_begin
2152 if (child
->role
->attach
) {
2153 child
->role
->attach(child
);
2159 * Updates @child to change its reference to point to @new_bs, including
2160 * checking and applying the necessary permisson updates both to the old node
2163 * NULL is passed as @new_bs for removing the reference before freeing @child.
2165 * If @new_bs is not NULL, bdrv_check_perm() must be called beforehand, as this
2166 * function uses bdrv_set_perm() to update the permissions according to the new
2167 * reference that @new_bs gets.
2169 static void bdrv_replace_child(BdrvChild
*child
, BlockDriverState
*new_bs
)
2171 BlockDriverState
*old_bs
= child
->bs
;
2172 uint64_t perm
, shared_perm
;
2174 bdrv_replace_child_noperm(child
, new_bs
);
2177 /* Update permissions for old node. This is guaranteed to succeed
2178 * because we're just taking a parent away, so we're loosening
2180 bdrv_get_cumulative_perm(old_bs
, &perm
, &shared_perm
);
2181 bdrv_check_perm(old_bs
, NULL
, perm
, shared_perm
, NULL
, &error_abort
);
2182 bdrv_set_perm(old_bs
, perm
, shared_perm
);
2186 bdrv_get_cumulative_perm(new_bs
, &perm
, &shared_perm
);
2187 bdrv_set_perm(new_bs
, perm
, shared_perm
);
2191 BdrvChild
*bdrv_root_attach_child(BlockDriverState
*child_bs
,
2192 const char *child_name
,
2193 const BdrvChildRole
*child_role
,
2194 uint64_t perm
, uint64_t shared_perm
,
2195 void *opaque
, Error
**errp
)
2200 ret
= bdrv_check_update_perm(child_bs
, NULL
, perm
, shared_perm
, NULL
, errp
);
2202 bdrv_abort_perm_update(child_bs
);
2206 child
= g_new(BdrvChild
, 1);
2207 *child
= (BdrvChild
) {
2209 .name
= g_strdup(child_name
),
2212 .shared_perm
= shared_perm
,
2216 /* This performs the matching bdrv_set_perm() for the above check. */
2217 bdrv_replace_child(child
, child_bs
);
2222 BdrvChild
*bdrv_attach_child(BlockDriverState
*parent_bs
,
2223 BlockDriverState
*child_bs
,
2224 const char *child_name
,
2225 const BdrvChildRole
*child_role
,
2229 uint64_t perm
, shared_perm
;
2231 bdrv_get_cumulative_perm(parent_bs
, &perm
, &shared_perm
);
2233 assert(parent_bs
->drv
);
2234 assert(bdrv_get_aio_context(parent_bs
) == bdrv_get_aio_context(child_bs
));
2235 bdrv_child_perm(parent_bs
, child_bs
, NULL
, child_role
, NULL
,
2236 perm
, shared_perm
, &perm
, &shared_perm
);
2238 child
= bdrv_root_attach_child(child_bs
, child_name
, child_role
,
2239 perm
, shared_perm
, parent_bs
, errp
);
2240 if (child
== NULL
) {
2244 QLIST_INSERT_HEAD(&parent_bs
->children
, child
, next
);
2248 static void bdrv_detach_child(BdrvChild
*child
)
2250 if (child
->next
.le_prev
) {
2251 QLIST_REMOVE(child
, next
);
2252 child
->next
.le_prev
= NULL
;
2255 bdrv_replace_child(child
, NULL
);
2257 g_free(child
->name
);
2261 void bdrv_root_unref_child(BdrvChild
*child
)
2263 BlockDriverState
*child_bs
;
2265 child_bs
= child
->bs
;
2266 bdrv_detach_child(child
);
2267 bdrv_unref(child_bs
);
2270 void bdrv_unref_child(BlockDriverState
*parent
, BdrvChild
*child
)
2272 if (child
== NULL
) {
2276 if (child
->bs
->inherits_from
== parent
) {
2279 /* Remove inherits_from only when the last reference between parent and
2280 * child->bs goes away. */
2281 QLIST_FOREACH(c
, &parent
->children
, next
) {
2282 if (c
!= child
&& c
->bs
== child
->bs
) {
2287 child
->bs
->inherits_from
= NULL
;
2291 bdrv_root_unref_child(child
);
2295 static void bdrv_parent_cb_change_media(BlockDriverState
*bs
, bool load
)
2298 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
2299 if (c
->role
->change_media
) {
2300 c
->role
->change_media(c
, load
);
2305 /* Return true if you can reach parent going through child->inherits_from
2306 * recursively. If parent or child are NULL, return false */
2307 static bool bdrv_inherits_from_recursive(BlockDriverState
*child
,
2308 BlockDriverState
*parent
)
2310 while (child
&& child
!= parent
) {
2311 child
= child
->inherits_from
;
2314 return child
!= NULL
;
2318 * Sets the backing file link of a BDS. A new reference is created; callers
2319 * which don't need their own reference any more must call bdrv_unref().
2321 void bdrv_set_backing_hd(BlockDriverState
*bs
, BlockDriverState
*backing_hd
,
2324 bool update_inherits_from
= bdrv_chain_contains(bs
, backing_hd
) &&
2325 bdrv_inherits_from_recursive(backing_hd
, bs
);
2328 bdrv_ref(backing_hd
);
2332 bdrv_unref_child(bs
, bs
->backing
);
2340 bs
->backing
= bdrv_attach_child(bs
, backing_hd
, "backing", &child_backing
,
2342 /* If backing_hd was already part of bs's backing chain, and
2343 * inherits_from pointed recursively to bs then let's update it to
2344 * point directly to bs (else it will become NULL). */
2345 if (update_inherits_from
) {
2346 backing_hd
->inherits_from
= bs
;
2349 bdrv_unref(backing_hd
);
2353 bdrv_refresh_limits(bs
, NULL
);
2357 * Opens the backing file for a BlockDriverState if not yet open
2359 * bdref_key specifies the key for the image's BlockdevRef in the options QDict.
2360 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
2361 * itself, all options starting with "${bdref_key}." are considered part of the
2364 * TODO Can this be unified with bdrv_open_image()?
2366 int bdrv_open_backing_file(BlockDriverState
*bs
, QDict
*parent_options
,
2367 const char *bdref_key
, Error
**errp
)
2369 char *backing_filename
= g_malloc0(PATH_MAX
);
2370 char *bdref_key_dot
;
2371 const char *reference
= NULL
;
2373 bool implicit_backing
= false;
2374 BlockDriverState
*backing_hd
;
2376 QDict
*tmp_parent_options
= NULL
;
2377 Error
*local_err
= NULL
;
2379 if (bs
->backing
!= NULL
) {
2383 /* NULL means an empty set of options */
2384 if (parent_options
== NULL
) {
2385 tmp_parent_options
= qdict_new();
2386 parent_options
= tmp_parent_options
;
2389 bs
->open_flags
&= ~BDRV_O_NO_BACKING
;
2391 bdref_key_dot
= g_strdup_printf("%s.", bdref_key
);
2392 qdict_extract_subqdict(parent_options
, &options
, bdref_key_dot
);
2393 g_free(bdref_key_dot
);
2396 * Caution: while qdict_get_try_str() is fine, getting non-string
2397 * types would require more care. When @parent_options come from
2398 * -blockdev or blockdev_add, its members are typed according to
2399 * the QAPI schema, but when they come from -drive, they're all
2402 reference
= qdict_get_try_str(parent_options
, bdref_key
);
2403 if (reference
|| qdict_haskey(options
, "file.filename")) {
2404 backing_filename
[0] = '\0';
2405 } else if (bs
->backing_file
[0] == '\0' && qdict_size(options
) == 0) {
2406 qobject_unref(options
);
2409 if (qdict_size(options
) == 0) {
2410 /* If the user specifies options that do not modify the
2411 * backing file's behavior, we might still consider it the
2412 * implicit backing file. But it's easier this way, and
2413 * just specifying some of the backing BDS's options is
2414 * only possible with -drive anyway (otherwise the QAPI
2415 * schema forces the user to specify everything). */
2416 implicit_backing
= !strcmp(bs
->auto_backing_file
, bs
->backing_file
);
2419 bdrv_get_full_backing_filename(bs
, backing_filename
, PATH_MAX
,
2423 error_propagate(errp
, local_err
);
2424 qobject_unref(options
);
2429 if (!bs
->drv
|| !bs
->drv
->supports_backing
) {
2431 error_setg(errp
, "Driver doesn't support backing files");
2432 qobject_unref(options
);
2437 bs
->backing_format
[0] != '\0' && !qdict_haskey(options
, "driver")) {
2438 qdict_put_str(options
, "driver", bs
->backing_format
);
2441 backing_hd
= bdrv_open_inherit(*backing_filename
? backing_filename
: NULL
,
2442 reference
, options
, 0, bs
, &child_backing
,
2445 bs
->open_flags
|= BDRV_O_NO_BACKING
;
2446 error_prepend(errp
, "Could not open backing file: ");
2450 bdrv_set_aio_context(backing_hd
, bdrv_get_aio_context(bs
));
2452 if (implicit_backing
) {
2453 bdrv_refresh_filename(backing_hd
);
2454 pstrcpy(bs
->auto_backing_file
, sizeof(bs
->auto_backing_file
),
2455 backing_hd
->filename
);
2458 /* Hook up the backing file link; drop our reference, bs owns the
2459 * backing_hd reference now */
2460 bdrv_set_backing_hd(bs
, backing_hd
, &local_err
);
2461 bdrv_unref(backing_hd
);
2463 error_propagate(errp
, local_err
);
2468 qdict_del(parent_options
, bdref_key
);
2471 g_free(backing_filename
);
2472 qobject_unref(tmp_parent_options
);
2476 static BlockDriverState
*
2477 bdrv_open_child_bs(const char *filename
, QDict
*options
, const char *bdref_key
,
2478 BlockDriverState
*parent
, const BdrvChildRole
*child_role
,
2479 bool allow_none
, Error
**errp
)
2481 BlockDriverState
*bs
= NULL
;
2482 QDict
*image_options
;
2483 char *bdref_key_dot
;
2484 const char *reference
;
2486 assert(child_role
!= NULL
);
2488 bdref_key_dot
= g_strdup_printf("%s.", bdref_key
);
2489 qdict_extract_subqdict(options
, &image_options
, bdref_key_dot
);
2490 g_free(bdref_key_dot
);
2493 * Caution: while qdict_get_try_str() is fine, getting non-string
2494 * types would require more care. When @options come from
2495 * -blockdev or blockdev_add, its members are typed according to
2496 * the QAPI schema, but when they come from -drive, they're all
2499 reference
= qdict_get_try_str(options
, bdref_key
);
2500 if (!filename
&& !reference
&& !qdict_size(image_options
)) {
2502 error_setg(errp
, "A block device must be specified for \"%s\"",
2505 qobject_unref(image_options
);
2509 bs
= bdrv_open_inherit(filename
, reference
, image_options
, 0,
2510 parent
, child_role
, errp
);
2516 qdict_del(options
, bdref_key
);
2521 * Opens a disk image whose options are given as BlockdevRef in another block
2524 * If allow_none is true, no image will be opened if filename is false and no
2525 * BlockdevRef is given. NULL will be returned, but errp remains unset.
2527 * bdrev_key specifies the key for the image's BlockdevRef in the options QDict.
2528 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
2529 * itself, all options starting with "${bdref_key}." are considered part of the
2532 * The BlockdevRef will be removed from the options QDict.
2534 BdrvChild
*bdrv_open_child(const char *filename
,
2535 QDict
*options
, const char *bdref_key
,
2536 BlockDriverState
*parent
,
2537 const BdrvChildRole
*child_role
,
2538 bool allow_none
, Error
**errp
)
2541 BlockDriverState
*bs
;
2543 bs
= bdrv_open_child_bs(filename
, options
, bdref_key
, parent
, child_role
,
2549 c
= bdrv_attach_child(parent
, bs
, bdref_key
, child_role
, errp
);
2558 /* TODO Future callers may need to specify parent/child_role in order for
2559 * option inheritance to work. Existing callers use it for the root node. */
2560 BlockDriverState
*bdrv_open_blockdev_ref(BlockdevRef
*ref
, Error
**errp
)
2562 BlockDriverState
*bs
= NULL
;
2563 Error
*local_err
= NULL
;
2564 QObject
*obj
= NULL
;
2565 QDict
*qdict
= NULL
;
2566 const char *reference
= NULL
;
2569 if (ref
->type
== QTYPE_QSTRING
) {
2570 reference
= ref
->u
.reference
;
2572 BlockdevOptions
*options
= &ref
->u
.definition
;
2573 assert(ref
->type
== QTYPE_QDICT
);
2575 v
= qobject_output_visitor_new(&obj
);
2576 visit_type_BlockdevOptions(v
, NULL
, &options
, &local_err
);
2578 error_propagate(errp
, local_err
);
2581 visit_complete(v
, &obj
);
2583 qdict
= qobject_to(QDict
, obj
);
2584 qdict_flatten(qdict
);
2586 /* bdrv_open_inherit() defaults to the values in bdrv_flags (for
2587 * compatibility with other callers) rather than what we want as the
2588 * real defaults. Apply the defaults here instead. */
2589 qdict_set_default_str(qdict
, BDRV_OPT_CACHE_DIRECT
, "off");
2590 qdict_set_default_str(qdict
, BDRV_OPT_CACHE_NO_FLUSH
, "off");
2591 qdict_set_default_str(qdict
, BDRV_OPT_READ_ONLY
, "off");
2592 qdict_set_default_str(qdict
, BDRV_OPT_AUTO_READ_ONLY
, "off");
2596 bs
= bdrv_open_inherit(NULL
, reference
, qdict
, 0, NULL
, NULL
, errp
);
2605 static BlockDriverState
*bdrv_append_temp_snapshot(BlockDriverState
*bs
,
2607 QDict
*snapshot_options
,
2610 /* TODO: extra byte is a hack to ensure MAX_PATH space on Windows. */
2611 char *tmp_filename
= g_malloc0(PATH_MAX
+ 1);
2613 QemuOpts
*opts
= NULL
;
2614 BlockDriverState
*bs_snapshot
= NULL
;
2615 Error
*local_err
= NULL
;
2618 /* if snapshot, we create a temporary backing file and open it
2619 instead of opening 'filename' directly */
2621 /* Get the required size from the image */
2622 total_size
= bdrv_getlength(bs
);
2623 if (total_size
< 0) {
2624 error_setg_errno(errp
, -total_size
, "Could not get image size");
2628 /* Create the temporary image */
2629 ret
= get_tmp_filename(tmp_filename
, PATH_MAX
+ 1);
2631 error_setg_errno(errp
, -ret
, "Could not get temporary filename");
2635 opts
= qemu_opts_create(bdrv_qcow2
.create_opts
, NULL
, 0,
2637 qemu_opt_set_number(opts
, BLOCK_OPT_SIZE
, total_size
, &error_abort
);
2638 ret
= bdrv_create(&bdrv_qcow2
, tmp_filename
, opts
, errp
);
2639 qemu_opts_del(opts
);
2641 error_prepend(errp
, "Could not create temporary overlay '%s': ",
2646 /* Prepare options QDict for the temporary file */
2647 qdict_put_str(snapshot_options
, "file.driver", "file");
2648 qdict_put_str(snapshot_options
, "file.filename", tmp_filename
);
2649 qdict_put_str(snapshot_options
, "driver", "qcow2");
2651 bs_snapshot
= bdrv_open(NULL
, NULL
, snapshot_options
, flags
, errp
);
2652 snapshot_options
= NULL
;
2657 /* bdrv_append() consumes a strong reference to bs_snapshot
2658 * (i.e. it will call bdrv_unref() on it) even on error, so in
2659 * order to be able to return one, we have to increase
2660 * bs_snapshot's refcount here */
2661 bdrv_ref(bs_snapshot
);
2662 bdrv_append(bs_snapshot
, bs
, &local_err
);
2664 error_propagate(errp
, local_err
);
2670 qobject_unref(snapshot_options
);
2671 g_free(tmp_filename
);
2676 * Opens a disk image (raw, qcow2, vmdk, ...)
2678 * options is a QDict of options to pass to the block drivers, or NULL for an
2679 * empty set of options. The reference to the QDict belongs to the block layer
2680 * after the call (even on failure), so if the caller intends to reuse the
2681 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
2683 * If *pbs is NULL, a new BDS will be created with a pointer to it stored there.
2684 * If it is not NULL, the referenced BDS will be reused.
2686 * The reference parameter may be used to specify an existing block device which
2687 * should be opened. If specified, neither options nor a filename may be given,
2688 * nor can an existing BDS be reused (that is, *pbs has to be NULL).
2690 static BlockDriverState
*bdrv_open_inherit(const char *filename
,
2691 const char *reference
,
2692 QDict
*options
, int flags
,
2693 BlockDriverState
*parent
,
2694 const BdrvChildRole
*child_role
,
2698 BlockBackend
*file
= NULL
;
2699 BlockDriverState
*bs
;
2700 BlockDriver
*drv
= NULL
;
2702 const char *drvname
;
2703 const char *backing
;
2704 Error
*local_err
= NULL
;
2705 QDict
*snapshot_options
= NULL
;
2706 int snapshot_flags
= 0;
2708 assert(!child_role
|| !flags
);
2709 assert(!child_role
== !parent
);
2712 bool options_non_empty
= options
? qdict_size(options
) : false;
2713 qobject_unref(options
);
2715 if (filename
|| options_non_empty
) {
2716 error_setg(errp
, "Cannot reference an existing block device with "
2717 "additional options or a new filename");
2721 bs
= bdrv_lookup_bs(reference
, reference
, errp
);
2732 /* NULL means an empty set of options */
2733 if (options
== NULL
) {
2734 options
= qdict_new();
2737 /* json: syntax counts as explicit options, as if in the QDict */
2738 parse_json_protocol(options
, &filename
, &local_err
);
2743 bs
->explicit_options
= qdict_clone_shallow(options
);
2746 bs
->inherits_from
= parent
;
2747 child_role
->inherit_options(&flags
, options
,
2748 parent
->open_flags
, parent
->options
);
2751 ret
= bdrv_fill_options(&options
, filename
, &flags
, &local_err
);
2757 * Set the BDRV_O_RDWR and BDRV_O_ALLOW_RDWR flags.
2758 * Caution: getting a boolean member of @options requires care.
2759 * When @options come from -blockdev or blockdev_add, members are
2760 * typed according to the QAPI schema, but when they come from
2761 * -drive, they're all QString.
2763 if (g_strcmp0(qdict_get_try_str(options
, BDRV_OPT_READ_ONLY
), "on") &&
2764 !qdict_get_try_bool(options
, BDRV_OPT_READ_ONLY
, false)) {
2765 flags
|= (BDRV_O_RDWR
| BDRV_O_ALLOW_RDWR
);
2767 flags
&= ~BDRV_O_RDWR
;
2770 if (flags
& BDRV_O_SNAPSHOT
) {
2771 snapshot_options
= qdict_new();
2772 bdrv_temp_snapshot_options(&snapshot_flags
, snapshot_options
,
2774 /* Let bdrv_backing_options() override "read-only" */
2775 qdict_del(options
, BDRV_OPT_READ_ONLY
);
2776 bdrv_backing_options(&flags
, options
, flags
, options
);
2779 bs
->open_flags
= flags
;
2780 bs
->options
= options
;
2781 options
= qdict_clone_shallow(options
);
2783 /* Find the right image format driver */
2784 /* See cautionary note on accessing @options above */
2785 drvname
= qdict_get_try_str(options
, "driver");
2787 drv
= bdrv_find_format(drvname
);
2789 error_setg(errp
, "Unknown driver: '%s'", drvname
);
2794 assert(drvname
|| !(flags
& BDRV_O_PROTOCOL
));
2796 /* See cautionary note on accessing @options above */
2797 backing
= qdict_get_try_str(options
, "backing");
2798 if (qobject_to(QNull
, qdict_get(options
, "backing")) != NULL
||
2799 (backing
&& *backing
== '\0'))
2802 warn_report("Use of \"backing\": \"\" is deprecated; "
2803 "use \"backing\": null instead");
2805 flags
|= BDRV_O_NO_BACKING
;
2806 qdict_del(options
, "backing");
2809 /* Open image file without format layer. This BlockBackend is only used for
2810 * probing, the block drivers will do their own bdrv_open_child() for the
2811 * same BDS, which is why we put the node name back into options. */
2812 if ((flags
& BDRV_O_PROTOCOL
) == 0) {
2813 BlockDriverState
*file_bs
;
2815 file_bs
= bdrv_open_child_bs(filename
, options
, "file", bs
,
2816 &child_file
, true, &local_err
);
2820 if (file_bs
!= NULL
) {
2821 /* Not requesting BLK_PERM_CONSISTENT_READ because we're only
2822 * looking at the header to guess the image format. This works even
2823 * in cases where a guest would not see a consistent state. */
2824 file
= blk_new(0, BLK_PERM_ALL
);
2825 blk_insert_bs(file
, file_bs
, &local_err
);
2826 bdrv_unref(file_bs
);
2831 qdict_put_str(options
, "file", bdrv_get_node_name(file_bs
));
2835 /* Image format probing */
2838 ret
= find_image_format(file
, filename
, &drv
, &local_err
);
2843 * This option update would logically belong in bdrv_fill_options(),
2844 * but we first need to open bs->file for the probing to work, while
2845 * opening bs->file already requires the (mostly) final set of options
2846 * so that cache mode etc. can be inherited.
2848 * Adding the driver later is somewhat ugly, but it's not an option
2849 * that would ever be inherited, so it's correct. We just need to make
2850 * sure to update both bs->options (which has the full effective
2851 * options for bs) and options (which has file.* already removed).
2853 qdict_put_str(bs
->options
, "driver", drv
->format_name
);
2854 qdict_put_str(options
, "driver", drv
->format_name
);
2856 error_setg(errp
, "Must specify either driver or file");
2860 /* BDRV_O_PROTOCOL must be set iff a protocol BDS is about to be created */
2861 assert(!!(flags
& BDRV_O_PROTOCOL
) == !!drv
->bdrv_file_open
);
2862 /* file must be NULL if a protocol BDS is about to be created
2863 * (the inverse results in an error message from bdrv_open_common()) */
2864 assert(!(flags
& BDRV_O_PROTOCOL
) || !file
);
2866 /* Open the image */
2867 ret
= bdrv_open_common(bs
, file
, options
, &local_err
);
2877 /* If there is a backing file, use it */
2878 if ((flags
& BDRV_O_NO_BACKING
) == 0) {
2879 ret
= bdrv_open_backing_file(bs
, options
, "backing", &local_err
);
2881 goto close_and_fail
;
2885 /* Remove all children options and references
2886 * from bs->options and bs->explicit_options */
2887 QLIST_FOREACH(child
, &bs
->children
, next
) {
2888 char *child_key_dot
;
2889 child_key_dot
= g_strdup_printf("%s.", child
->name
);
2890 qdict_extract_subqdict(bs
->explicit_options
, NULL
, child_key_dot
);
2891 qdict_extract_subqdict(bs
->options
, NULL
, child_key_dot
);
2892 qdict_del(bs
->explicit_options
, child
->name
);
2893 qdict_del(bs
->options
, child
->name
);
2894 g_free(child_key_dot
);
2897 /* Check if any unknown options were used */
2898 if (qdict_size(options
) != 0) {
2899 const QDictEntry
*entry
= qdict_first(options
);
2900 if (flags
& BDRV_O_PROTOCOL
) {
2901 error_setg(errp
, "Block protocol '%s' doesn't support the option "
2902 "'%s'", drv
->format_name
, entry
->key
);
2905 "Block format '%s' does not support the option '%s'",
2906 drv
->format_name
, entry
->key
);
2909 goto close_and_fail
;
2912 bdrv_parent_cb_change_media(bs
, true);
2914 qobject_unref(options
);
2917 /* For snapshot=on, create a temporary qcow2 overlay. bs points to the
2918 * temporary snapshot afterwards. */
2919 if (snapshot_flags
) {
2920 BlockDriverState
*snapshot_bs
;
2921 snapshot_bs
= bdrv_append_temp_snapshot(bs
, snapshot_flags
,
2922 snapshot_options
, &local_err
);
2923 snapshot_options
= NULL
;
2925 goto close_and_fail
;
2927 /* We are not going to return bs but the overlay on top of it
2928 * (snapshot_bs); thus, we have to drop the strong reference to bs
2929 * (which we obtained by calling bdrv_new()). bs will not be deleted,
2930 * though, because the overlay still has a reference to it. */
2939 qobject_unref(snapshot_options
);
2940 qobject_unref(bs
->explicit_options
);
2941 qobject_unref(bs
->options
);
2942 qobject_unref(options
);
2944 bs
->explicit_options
= NULL
;
2946 error_propagate(errp
, local_err
);
2951 qobject_unref(snapshot_options
);
2952 qobject_unref(options
);
2953 error_propagate(errp
, local_err
);
2957 BlockDriverState
*bdrv_open(const char *filename
, const char *reference
,
2958 QDict
*options
, int flags
, Error
**errp
)
2960 return bdrv_open_inherit(filename
, reference
, options
, flags
, NULL
,
2965 * Adds a BlockDriverState to a simple queue for an atomic, transactional
2966 * reopen of multiple devices.
2968 * bs_queue can either be an existing BlockReopenQueue that has had QSIMPLE_INIT
2969 * already performed, or alternatively may be NULL a new BlockReopenQueue will
2970 * be created and initialized. This newly created BlockReopenQueue should be
2971 * passed back in for subsequent calls that are intended to be of the same
2974 * bs is the BlockDriverState to add to the reopen queue.
2976 * options contains the changed options for the associated bs
2977 * (the BlockReopenQueue takes ownership)
2979 * flags contains the open flags for the associated bs
2981 * returns a pointer to bs_queue, which is either the newly allocated
2982 * bs_queue, or the existing bs_queue being used.
2984 * bs must be drained between bdrv_reopen_queue() and bdrv_reopen_multiple().
2986 static BlockReopenQueue
*bdrv_reopen_queue_child(BlockReopenQueue
*bs_queue
,
2987 BlockDriverState
*bs
,
2989 const BdrvChildRole
*role
,
2990 QDict
*parent_options
,
2995 BlockReopenQueueEntry
*bs_entry
;
2997 QDict
*old_options
, *explicit_options
, *options_copy
;
3001 /* Make sure that the caller remembered to use a drained section. This is
3002 * important to avoid graph changes between the recursive queuing here and
3003 * bdrv_reopen_multiple(). */
3004 assert(bs
->quiesce_counter
> 0);
3006 if (bs_queue
== NULL
) {
3007 bs_queue
= g_new0(BlockReopenQueue
, 1);
3008 QSIMPLEQ_INIT(bs_queue
);
3012 options
= qdict_new();
3015 /* Check if this BlockDriverState is already in the queue */
3016 QSIMPLEQ_FOREACH(bs_entry
, bs_queue
, entry
) {
3017 if (bs
== bs_entry
->state
.bs
) {
3023 * Precedence of options:
3024 * 1. Explicitly passed in options (highest)
3025 * 2. Retained from explicitly set options of bs
3026 * 3. Inherited from parent node
3027 * 4. Retained from effective options of bs
3030 /* Old explicitly set values (don't overwrite by inherited value) */
3032 old_options
= qdict_clone_shallow(bs_entry
->state
.explicit_options
);
3034 old_options
= qdict_clone_shallow(bs
->explicit_options
);
3036 bdrv_join_options(bs
, options
, old_options
);
3037 qobject_unref(old_options
);
3039 explicit_options
= qdict_clone_shallow(options
);
3041 /* Inherit from parent node */
3042 if (parent_options
) {
3044 role
->inherit_options(&flags
, options
, parent_flags
, parent_options
);
3046 flags
= bdrv_get_flags(bs
);
3049 /* Old values are used for options that aren't set yet */
3050 old_options
= qdict_clone_shallow(bs
->options
);
3051 bdrv_join_options(bs
, options
, old_options
);
3052 qobject_unref(old_options
);
3054 /* We have the final set of options so let's update the flags */
3055 options_copy
= qdict_clone_shallow(options
);
3056 opts
= qemu_opts_create(&bdrv_runtime_opts
, NULL
, 0, &error_abort
);
3057 qemu_opts_absorb_qdict(opts
, options_copy
, NULL
);
3058 update_flags_from_options(&flags
, opts
);
3059 qemu_opts_del(opts
);
3060 qobject_unref(options_copy
);
3062 /* bdrv_open_inherit() sets and clears some additional flags internally */
3063 flags
&= ~BDRV_O_PROTOCOL
;
3064 if (flags
& BDRV_O_RDWR
) {
3065 flags
|= BDRV_O_ALLOW_RDWR
;
3069 bs_entry
= g_new0(BlockReopenQueueEntry
, 1);
3070 QSIMPLEQ_INSERT_TAIL(bs_queue
, bs_entry
, entry
);
3072 qobject_unref(bs_entry
->state
.options
);
3073 qobject_unref(bs_entry
->state
.explicit_options
);
3076 bs_entry
->state
.bs
= bs
;
3077 bs_entry
->state
.options
= options
;
3078 bs_entry
->state
.explicit_options
= explicit_options
;
3079 bs_entry
->state
.flags
= flags
;
3081 /* This needs to be overwritten in bdrv_reopen_prepare() */
3082 bs_entry
->state
.perm
= UINT64_MAX
;
3083 bs_entry
->state
.shared_perm
= 0;
3085 QLIST_FOREACH(child
, &bs
->children
, next
) {
3086 QDict
*new_child_options
;
3087 char *child_key_dot
;
3089 /* reopen can only change the options of block devices that were
3090 * implicitly created and inherited options. For other (referenced)
3091 * block devices, a syntax like "backing.foo" results in an error. */
3092 if (child
->bs
->inherits_from
!= bs
) {
3096 child_key_dot
= g_strdup_printf("%s.", child
->name
);
3097 qdict_extract_subqdict(explicit_options
, NULL
, child_key_dot
);
3098 qdict_extract_subqdict(options
, &new_child_options
, child_key_dot
);
3099 g_free(child_key_dot
);
3101 bdrv_reopen_queue_child(bs_queue
, child
->bs
, new_child_options
,
3102 child
->role
, options
, flags
);
3108 BlockReopenQueue
*bdrv_reopen_queue(BlockReopenQueue
*bs_queue
,
3109 BlockDriverState
*bs
,
3112 return bdrv_reopen_queue_child(bs_queue
, bs
, options
, NULL
, NULL
, 0);
3116 * Reopen multiple BlockDriverStates atomically & transactionally.
3118 * The queue passed in (bs_queue) must have been built up previous
3119 * via bdrv_reopen_queue().
3121 * Reopens all BDS specified in the queue, with the appropriate
3122 * flags. All devices are prepared for reopen, and failure of any
3123 * device will cause all device changes to be abandoned, and intermediate
3126 * If all devices prepare successfully, then the changes are committed
3129 * All affected nodes must be drained between bdrv_reopen_queue() and
3130 * bdrv_reopen_multiple().
3132 int bdrv_reopen_multiple(AioContext
*ctx
, BlockReopenQueue
*bs_queue
, Error
**errp
)
3135 BlockReopenQueueEntry
*bs_entry
, *next
;
3136 Error
*local_err
= NULL
;
3138 assert(bs_queue
!= NULL
);
3140 QSIMPLEQ_FOREACH(bs_entry
, bs_queue
, entry
) {
3141 assert(bs_entry
->state
.bs
->quiesce_counter
> 0);
3142 if (bdrv_reopen_prepare(&bs_entry
->state
, bs_queue
, &local_err
)) {
3143 error_propagate(errp
, local_err
);
3146 bs_entry
->prepared
= true;
3149 /* If we reach this point, we have success and just need to apply the
3152 QSIMPLEQ_FOREACH(bs_entry
, bs_queue
, entry
) {
3153 bdrv_reopen_commit(&bs_entry
->state
);
3159 QSIMPLEQ_FOREACH_SAFE(bs_entry
, bs_queue
, entry
, next
) {
3161 if (bs_entry
->prepared
) {
3162 bdrv_reopen_abort(&bs_entry
->state
);
3164 qobject_unref(bs_entry
->state
.explicit_options
);
3165 qobject_unref(bs_entry
->state
.options
);
3174 int bdrv_reopen_set_read_only(BlockDriverState
*bs
, bool read_only
,
3178 BlockReopenQueue
*queue
;
3179 QDict
*opts
= qdict_new();
3181 qdict_put_bool(opts
, BDRV_OPT_READ_ONLY
, read_only
);
3183 bdrv_subtree_drained_begin(bs
);
3184 queue
= bdrv_reopen_queue(NULL
, bs
, opts
);
3185 ret
= bdrv_reopen_multiple(bdrv_get_aio_context(bs
), queue
, errp
);
3186 bdrv_subtree_drained_end(bs
);
3191 static BlockReopenQueueEntry
*find_parent_in_reopen_queue(BlockReopenQueue
*q
,
3194 BlockReopenQueueEntry
*entry
;
3196 QSIMPLEQ_FOREACH(entry
, q
, entry
) {
3197 BlockDriverState
*bs
= entry
->state
.bs
;
3200 QLIST_FOREACH(child
, &bs
->children
, next
) {
3210 static void bdrv_reopen_perm(BlockReopenQueue
*q
, BlockDriverState
*bs
,
3211 uint64_t *perm
, uint64_t *shared
)
3214 BlockReopenQueueEntry
*parent
;
3215 uint64_t cumulative_perms
= 0;
3216 uint64_t cumulative_shared_perms
= BLK_PERM_ALL
;
3218 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
3219 parent
= find_parent_in_reopen_queue(q
, c
);
3221 cumulative_perms
|= c
->perm
;
3222 cumulative_shared_perms
&= c
->shared_perm
;
3224 uint64_t nperm
, nshared
;
3226 bdrv_child_perm(parent
->state
.bs
, bs
, c
, c
->role
, q
,
3227 parent
->state
.perm
, parent
->state
.shared_perm
,
3230 cumulative_perms
|= nperm
;
3231 cumulative_shared_perms
&= nshared
;
3234 *perm
= cumulative_perms
;
3235 *shared
= cumulative_shared_perms
;
3239 * Prepares a BlockDriverState for reopen. All changes are staged in the
3240 * 'opaque' field of the BDRVReopenState, which is used and allocated by
3241 * the block driver layer .bdrv_reopen_prepare()
3243 * bs is the BlockDriverState to reopen
3244 * flags are the new open flags
3245 * queue is the reopen queue
3247 * Returns 0 on success, non-zero on error. On error errp will be set
3250 * On failure, bdrv_reopen_abort() will be called to clean up any data.
3251 * It is the responsibility of the caller to then call the abort() or
3252 * commit() for any other BDS that have been left in a prepare() state
3255 int bdrv_reopen_prepare(BDRVReopenState
*reopen_state
, BlockReopenQueue
*queue
,
3260 Error
*local_err
= NULL
;
3263 QDict
*orig_reopen_opts
;
3264 char *discard
= NULL
;
3266 bool drv_prepared
= false;
3268 assert(reopen_state
!= NULL
);
3269 assert(reopen_state
->bs
->drv
!= NULL
);
3270 drv
= reopen_state
->bs
->drv
;
3272 /* This function and each driver's bdrv_reopen_prepare() remove
3273 * entries from reopen_state->options as they are processed, so
3274 * we need to make a copy of the original QDict. */
3275 orig_reopen_opts
= qdict_clone_shallow(reopen_state
->options
);
3277 /* Process generic block layer options */
3278 opts
= qemu_opts_create(&bdrv_runtime_opts
, NULL
, 0, &error_abort
);
3279 qemu_opts_absorb_qdict(opts
, reopen_state
->options
, &local_err
);
3281 error_propagate(errp
, local_err
);
3286 /* This was already called in bdrv_reopen_queue_child() so the flags
3287 * are up-to-date. This time we simply want to remove the options from
3288 * QemuOpts in order to indicate that they have been processed. */
3289 old_flags
= reopen_state
->flags
;
3290 update_flags_from_options(&reopen_state
->flags
, opts
);
3291 assert(old_flags
== reopen_state
->flags
);
3293 discard
= qemu_opt_get_del(opts
, BDRV_OPT_DISCARD
);
3294 if (discard
!= NULL
) {
3295 if (bdrv_parse_discard_flags(discard
, &reopen_state
->flags
) != 0) {
3296 error_setg(errp
, "Invalid discard option");
3302 reopen_state
->detect_zeroes
=
3303 bdrv_parse_detect_zeroes(opts
, reopen_state
->flags
, &local_err
);
3305 error_propagate(errp
, local_err
);
3310 /* All other options (including node-name and driver) must be unchanged.
3311 * Put them back into the QDict, so that they are checked at the end
3312 * of this function. */
3313 qemu_opts_to_qdict(opts
, reopen_state
->options
);
3315 /* If we are to stay read-only, do not allow permission change
3316 * to r/w. Attempting to set to r/w may fail if either BDRV_O_ALLOW_RDWR is
3317 * not set, or if the BDS still has copy_on_read enabled */
3318 read_only
= !(reopen_state
->flags
& BDRV_O_RDWR
);
3319 ret
= bdrv_can_set_read_only(reopen_state
->bs
, read_only
, true, &local_err
);
3321 error_propagate(errp
, local_err
);
3325 /* Calculate required permissions after reopening */
3326 bdrv_reopen_perm(queue
, reopen_state
->bs
,
3327 &reopen_state
->perm
, &reopen_state
->shared_perm
);
3329 ret
= bdrv_flush(reopen_state
->bs
);
3331 error_setg_errno(errp
, -ret
, "Error flushing drive");
3335 if (drv
->bdrv_reopen_prepare
) {
3336 ret
= drv
->bdrv_reopen_prepare(reopen_state
, queue
, &local_err
);
3338 if (local_err
!= NULL
) {
3339 error_propagate(errp
, local_err
);
3341 bdrv_refresh_filename(reopen_state
->bs
);
3342 error_setg(errp
, "failed while preparing to reopen image '%s'",
3343 reopen_state
->bs
->filename
);
3348 /* It is currently mandatory to have a bdrv_reopen_prepare()
3349 * handler for each supported drv. */
3350 error_setg(errp
, "Block format '%s' used by node '%s' "
3351 "does not support reopening files", drv
->format_name
,
3352 bdrv_get_device_or_node_name(reopen_state
->bs
));
3357 drv_prepared
= true;
3359 /* Options that are not handled are only okay if they are unchanged
3360 * compared to the old state. It is expected that some options are only
3361 * used for the initial open, but not reopen (e.g. filename) */
3362 if (qdict_size(reopen_state
->options
)) {
3363 const QDictEntry
*entry
= qdict_first(reopen_state
->options
);
3366 QObject
*new = entry
->value
;
3367 QObject
*old
= qdict_get(reopen_state
->bs
->options
, entry
->key
);
3369 /* Allow child references (child_name=node_name) as long as they
3370 * point to the current child (i.e. everything stays the same). */
3371 if (qobject_type(new) == QTYPE_QSTRING
) {
3373 QLIST_FOREACH(child
, &reopen_state
->bs
->children
, next
) {
3374 if (!strcmp(child
->name
, entry
->key
)) {
3380 const char *str
= qobject_get_try_str(new);
3381 if (!strcmp(child
->bs
->node_name
, str
)) {
3382 continue; /* Found child with this name, skip option */
3388 * TODO: When using -drive to specify blockdev options, all values
3389 * will be strings; however, when using -blockdev, blockdev-add or
3390 * filenames using the json:{} pseudo-protocol, they will be
3392 * In contrast, reopening options are (currently) always strings
3393 * (because you can only specify them through qemu-io; all other
3394 * callers do not specify any options).
3395 * Therefore, when using anything other than -drive to create a BDS,
3396 * this cannot detect non-string options as unchanged, because
3397 * qobject_is_equal() always returns false for objects of different
3398 * type. In the future, this should be remedied by correctly typing
3399 * all options. For now, this is not too big of an issue because
3400 * the user can simply omit options which cannot be changed anyway,
3401 * so they will stay unchanged.
3403 if (!qobject_is_equal(new, old
)) {
3404 error_setg(errp
, "Cannot change the option '%s'", entry
->key
);
3408 } while ((entry
= qdict_next(reopen_state
->options
, entry
)));
3411 ret
= bdrv_check_perm(reopen_state
->bs
, queue
, reopen_state
->perm
,
3412 reopen_state
->shared_perm
, NULL
, errp
);
3419 /* Restore the original reopen_state->options QDict */
3420 qobject_unref(reopen_state
->options
);
3421 reopen_state
->options
= qobject_ref(orig_reopen_opts
);
3424 if (ret
< 0 && drv_prepared
) {
3425 /* drv->bdrv_reopen_prepare() has succeeded, so we need to
3426 * call drv->bdrv_reopen_abort() before signaling an error
3427 * (bdrv_reopen_multiple() will not call bdrv_reopen_abort()
3428 * when the respective bdrv_reopen_prepare() has failed) */
3429 if (drv
->bdrv_reopen_abort
) {
3430 drv
->bdrv_reopen_abort(reopen_state
);
3433 qemu_opts_del(opts
);
3434 qobject_unref(orig_reopen_opts
);
3440 * Takes the staged changes for the reopen from bdrv_reopen_prepare(), and
3441 * makes them final by swapping the staging BlockDriverState contents into
3442 * the active BlockDriverState contents.
3444 void bdrv_reopen_commit(BDRVReopenState
*reopen_state
)
3447 BlockDriverState
*bs
;
3449 bool old_can_write
, new_can_write
;
3451 assert(reopen_state
!= NULL
);
3452 bs
= reopen_state
->bs
;
3454 assert(drv
!= NULL
);
3457 !bdrv_is_read_only(bs
) && !(bdrv_get_flags(bs
) & BDRV_O_INACTIVE
);
3459 /* If there are any driver level actions to take */
3460 if (drv
->bdrv_reopen_commit
) {
3461 drv
->bdrv_reopen_commit(reopen_state
);
3464 /* set BDS specific flags now */
3465 qobject_unref(bs
->explicit_options
);
3466 qobject_unref(bs
->options
);
3468 bs
->explicit_options
= reopen_state
->explicit_options
;
3469 bs
->options
= reopen_state
->options
;
3470 bs
->open_flags
= reopen_state
->flags
;
3471 bs
->read_only
= !(reopen_state
->flags
& BDRV_O_RDWR
);
3472 bs
->detect_zeroes
= reopen_state
->detect_zeroes
;
3474 /* Remove child references from bs->options and bs->explicit_options.
3475 * Child options were already removed in bdrv_reopen_queue_child() */
3476 QLIST_FOREACH(child
, &bs
->children
, next
) {
3477 qdict_del(bs
->explicit_options
, child
->name
);
3478 qdict_del(bs
->options
, child
->name
);
3481 bdrv_refresh_limits(bs
, NULL
);
3483 bdrv_set_perm(reopen_state
->bs
, reopen_state
->perm
,
3484 reopen_state
->shared_perm
);
3487 !bdrv_is_read_only(bs
) && !(bdrv_get_flags(bs
) & BDRV_O_INACTIVE
);
3488 if (!old_can_write
&& new_can_write
&& drv
->bdrv_reopen_bitmaps_rw
) {
3489 Error
*local_err
= NULL
;
3490 if (drv
->bdrv_reopen_bitmaps_rw(bs
, &local_err
) < 0) {
3491 /* This is not fatal, bitmaps just left read-only, so all following
3492 * writes will fail. User can remove read-only bitmaps to unblock
3495 error_reportf_err(local_err
,
3496 "%s: Failed to make dirty bitmaps writable: ",
3497 bdrv_get_node_name(bs
));
3503 * Abort the reopen, and delete and free the staged changes in
3506 void bdrv_reopen_abort(BDRVReopenState
*reopen_state
)
3510 assert(reopen_state
!= NULL
);
3511 drv
= reopen_state
->bs
->drv
;
3512 assert(drv
!= NULL
);
3514 if (drv
->bdrv_reopen_abort
) {
3515 drv
->bdrv_reopen_abort(reopen_state
);
3518 bdrv_abort_perm_update(reopen_state
->bs
);
3522 static void bdrv_close(BlockDriverState
*bs
)
3524 BdrvAioNotifier
*ban
, *ban_next
;
3525 BdrvChild
*child
, *next
;
3528 assert(!bs
->refcnt
);
3530 bdrv_drained_begin(bs
); /* complete I/O */
3532 bdrv_drain(bs
); /* in case flush left pending I/O */
3535 if (bs
->drv
->bdrv_close
) {
3536 bs
->drv
->bdrv_close(bs
);
3541 bdrv_set_backing_hd(bs
, NULL
, &error_abort
);
3543 if (bs
->file
!= NULL
) {
3544 bdrv_unref_child(bs
, bs
->file
);
3548 QLIST_FOREACH_SAFE(child
, &bs
->children
, next
, next
) {
3549 /* TODO Remove bdrv_unref() from drivers' close function and use
3550 * bdrv_unref_child() here */
3551 if (child
->bs
->inherits_from
== bs
) {
3552 child
->bs
->inherits_from
= NULL
;
3554 bdrv_detach_child(child
);
3559 atomic_set(&bs
->copy_on_read
, 0);
3560 bs
->backing_file
[0] = '\0';
3561 bs
->backing_format
[0] = '\0';
3562 bs
->total_sectors
= 0;
3563 bs
->encrypted
= false;
3565 qobject_unref(bs
->options
);
3566 qobject_unref(bs
->explicit_options
);
3568 bs
->explicit_options
= NULL
;
3569 qobject_unref(bs
->full_open_options
);
3570 bs
->full_open_options
= NULL
;
3572 bdrv_release_named_dirty_bitmaps(bs
);
3573 assert(QLIST_EMPTY(&bs
->dirty_bitmaps
));
3575 QLIST_FOREACH_SAFE(ban
, &bs
->aio_notifiers
, list
, ban_next
) {
3578 QLIST_INIT(&bs
->aio_notifiers
);
3579 bdrv_drained_end(bs
);
3582 void bdrv_close_all(void)
3584 assert(job_next(NULL
) == NULL
);
3585 nbd_export_close_all();
3587 /* Drop references from requests still in flight, such as canceled block
3588 * jobs whose AIO context has not been polled yet */
3591 blk_remove_all_bs();
3592 blockdev_close_all_bdrv_states();
3594 assert(QTAILQ_EMPTY(&all_bdrv_states
));
3597 static bool should_update_child(BdrvChild
*c
, BlockDriverState
*to
)
3603 if (c
->role
->stay_at_node
) {
3607 /* If the child @c belongs to the BDS @to, replacing the current
3608 * c->bs by @to would mean to create a loop.
3610 * Such a case occurs when appending a BDS to a backing chain.
3611 * For instance, imagine the following chain:
3613 * guest device -> node A -> further backing chain...
3615 * Now we create a new BDS B which we want to put on top of this
3616 * chain, so we first attach A as its backing node:
3621 * guest device -> node A -> further backing chain...
3623 * Finally we want to replace A by B. When doing that, we want to
3624 * replace all pointers to A by pointers to B -- except for the
3625 * pointer from B because (1) that would create a loop, and (2)
3626 * that pointer should simply stay intact:
3628 * guest device -> node B
3631 * node A -> further backing chain...
3633 * In general, when replacing a node A (c->bs) by a node B (@to),
3634 * if A is a child of B, that means we cannot replace A by B there
3635 * because that would create a loop. Silently detaching A from B
3636 * is also not really an option. So overall just leaving A in
3637 * place there is the most sensible choice.
3639 * We would also create a loop in any cases where @c is only
3640 * indirectly referenced by @to. Prevent this by returning false
3641 * if @c is found (by breadth-first search) anywhere in the whole
3646 found
= g_hash_table_new(NULL
, NULL
);
3647 g_hash_table_add(found
, to
);
3648 queue
= g_queue_new();
3649 g_queue_push_tail(queue
, to
);
3651 while (!g_queue_is_empty(queue
)) {
3652 BlockDriverState
*v
= g_queue_pop_head(queue
);
3655 QLIST_FOREACH(c2
, &v
->children
, next
) {
3661 if (g_hash_table_contains(found
, c2
->bs
)) {
3665 g_queue_push_tail(queue
, c2
->bs
);
3666 g_hash_table_add(found
, c2
->bs
);
3670 g_queue_free(queue
);
3671 g_hash_table_destroy(found
);
3676 void bdrv_replace_node(BlockDriverState
*from
, BlockDriverState
*to
,
3679 BdrvChild
*c
, *next
;
3680 GSList
*list
= NULL
, *p
;
3681 uint64_t old_perm
, old_shared
;
3682 uint64_t perm
= 0, shared
= BLK_PERM_ALL
;
3685 assert(!atomic_read(&from
->in_flight
));
3686 assert(!atomic_read(&to
->in_flight
));
3688 /* Make sure that @from doesn't go away until we have successfully attached
3689 * all of its parents to @to. */
3692 /* Put all parents into @list and calculate their cumulative permissions */
3693 QLIST_FOREACH_SAFE(c
, &from
->parents
, next_parent
, next
) {
3694 assert(c
->bs
== from
);
3695 if (!should_update_child(c
, to
)) {
3698 list
= g_slist_prepend(list
, c
);
3700 shared
&= c
->shared_perm
;
3703 /* Check whether the required permissions can be granted on @to, ignoring
3704 * all BdrvChild in @list so that they can't block themselves. */
3705 ret
= bdrv_check_update_perm(to
, NULL
, perm
, shared
, list
, errp
);
3707 bdrv_abort_perm_update(to
);
3711 /* Now actually perform the change. We performed the permission check for
3712 * all elements of @list at once, so set the permissions all at once at the
3714 for (p
= list
; p
!= NULL
; p
= p
->next
) {
3718 bdrv_replace_child_noperm(c
, to
);
3722 bdrv_get_cumulative_perm(to
, &old_perm
, &old_shared
);
3723 bdrv_set_perm(to
, old_perm
| perm
, old_shared
| shared
);
3731 * Add new bs contents at the top of an image chain while the chain is
3732 * live, while keeping required fields on the top layer.
3734 * This will modify the BlockDriverState fields, and swap contents
3735 * between bs_new and bs_top. Both bs_new and bs_top are modified.
3737 * bs_new must not be attached to a BlockBackend.
3739 * This function does not create any image files.
3741 * bdrv_append() takes ownership of a bs_new reference and unrefs it because
3742 * that's what the callers commonly need. bs_new will be referenced by the old
3743 * parents of bs_top after bdrv_append() returns. If the caller needs to keep a
3744 * reference of its own, it must call bdrv_ref().
3746 void bdrv_append(BlockDriverState
*bs_new
, BlockDriverState
*bs_top
,
3749 Error
*local_err
= NULL
;
3751 bdrv_set_backing_hd(bs_new
, bs_top
, &local_err
);
3753 error_propagate(errp
, local_err
);
3757 bdrv_replace_node(bs_top
, bs_new
, &local_err
);
3759 error_propagate(errp
, local_err
);
3760 bdrv_set_backing_hd(bs_new
, NULL
, &error_abort
);
3764 /* bs_new is now referenced by its new parents, we don't need the
3765 * additional reference any more. */
3770 static void bdrv_delete(BlockDriverState
*bs
)
3773 assert(bdrv_op_blocker_is_empty(bs
));
3774 assert(!bs
->refcnt
);
3778 /* remove from list, if necessary */
3779 if (bs
->node_name
[0] != '\0') {
3780 QTAILQ_REMOVE(&graph_bdrv_states
, bs
, node_list
);
3782 QTAILQ_REMOVE(&all_bdrv_states
, bs
, bs_list
);
3788 * Run consistency checks on an image
3790 * Returns 0 if the check could be completed (it doesn't mean that the image is
3791 * free of errors) or -errno when an internal error occurred. The results of the
3792 * check are stored in res.
3794 static int coroutine_fn
bdrv_co_check(BlockDriverState
*bs
,
3795 BdrvCheckResult
*res
, BdrvCheckMode fix
)
3797 if (bs
->drv
== NULL
) {
3800 if (bs
->drv
->bdrv_co_check
== NULL
) {
3804 memset(res
, 0, sizeof(*res
));
3805 return bs
->drv
->bdrv_co_check(bs
, res
, fix
);
3808 typedef struct CheckCo
{
3809 BlockDriverState
*bs
;
3810 BdrvCheckResult
*res
;
3815 static void bdrv_check_co_entry(void *opaque
)
3817 CheckCo
*cco
= opaque
;
3818 cco
->ret
= bdrv_co_check(cco
->bs
, cco
->res
, cco
->fix
);
3822 int bdrv_check(BlockDriverState
*bs
,
3823 BdrvCheckResult
*res
, BdrvCheckMode fix
)
3829 .ret
= -EINPROGRESS
,
3833 if (qemu_in_coroutine()) {
3834 /* Fast-path if already in coroutine context */
3835 bdrv_check_co_entry(&cco
);
3837 co
= qemu_coroutine_create(bdrv_check_co_entry
, &cco
);
3838 bdrv_coroutine_enter(bs
, co
);
3839 BDRV_POLL_WHILE(bs
, cco
.ret
== -EINPROGRESS
);
3848 * -EINVAL - backing format specified, but no file
3849 * -ENOSPC - can't update the backing file because no space is left in the
3851 * -ENOTSUP - format driver doesn't support changing the backing file
3853 int bdrv_change_backing_file(BlockDriverState
*bs
,
3854 const char *backing_file
, const char *backing_fmt
)
3856 BlockDriver
*drv
= bs
->drv
;
3863 /* Backing file format doesn't make sense without a backing file */
3864 if (backing_fmt
&& !backing_file
) {
3868 if (drv
->bdrv_change_backing_file
!= NULL
) {
3869 ret
= drv
->bdrv_change_backing_file(bs
, backing_file
, backing_fmt
);
3875 pstrcpy(bs
->backing_file
, sizeof(bs
->backing_file
), backing_file
?: "");
3876 pstrcpy(bs
->backing_format
, sizeof(bs
->backing_format
), backing_fmt
?: "");
3877 pstrcpy(bs
->auto_backing_file
, sizeof(bs
->auto_backing_file
),
3878 backing_file
?: "");
3884 * Finds the image layer in the chain that has 'bs' as its backing file.
3886 * active is the current topmost image.
3888 * Returns NULL if bs is not found in active's image chain,
3889 * or if active == bs.
3891 * Returns the bottommost base image if bs == NULL.
3893 BlockDriverState
*bdrv_find_overlay(BlockDriverState
*active
,
3894 BlockDriverState
*bs
)
3896 while (active
&& bs
!= backing_bs(active
)) {
3897 active
= backing_bs(active
);
3903 /* Given a BDS, searches for the base layer. */
3904 BlockDriverState
*bdrv_find_base(BlockDriverState
*bs
)
3906 return bdrv_find_overlay(bs
, NULL
);
3910 * Drops images above 'base' up to and including 'top', and sets the image
3911 * above 'top' to have base as its backing file.
3913 * Requires that the overlay to 'top' is opened r/w, so that the backing file
3914 * information in 'bs' can be properly updated.
3916 * E.g., this will convert the following chain:
3917 * bottom <- base <- intermediate <- top <- active
3921 * bottom <- base <- active
3923 * It is allowed for bottom==base, in which case it converts:
3925 * base <- intermediate <- top <- active
3931 * If backing_file_str is non-NULL, it will be used when modifying top's
3932 * overlay image metadata.
3935 * if active == top, that is considered an error
3938 int bdrv_drop_intermediate(BlockDriverState
*top
, BlockDriverState
*base
,
3939 const char *backing_file_str
)
3941 BlockDriverState
*explicit_top
= top
;
3942 bool update_inherits_from
;
3943 BdrvChild
*c
, *next
;
3944 Error
*local_err
= NULL
;
3949 if (!top
->drv
|| !base
->drv
) {
3953 /* Make sure that base is in the backing chain of top */
3954 if (!bdrv_chain_contains(top
, base
)) {
3958 /* If 'base' recursively inherits from 'top' then we should set
3959 * base->inherits_from to top->inherits_from after 'top' and all
3960 * other intermediate nodes have been dropped.
3961 * If 'top' is an implicit node (e.g. "commit_top") we should skip
3962 * it because no one inherits from it. We use explicit_top for that. */
3963 while (explicit_top
&& explicit_top
->implicit
) {
3964 explicit_top
= backing_bs(explicit_top
);
3966 update_inherits_from
= bdrv_inherits_from_recursive(base
, explicit_top
);
3968 /* success - we can delete the intermediate states, and link top->base */
3969 /* TODO Check graph modification op blockers (BLK_PERM_GRAPH_MOD) once
3970 * we've figured out how they should work. */
3971 if (!backing_file_str
) {
3972 bdrv_refresh_filename(base
);
3973 backing_file_str
= base
->filename
;
3976 QLIST_FOREACH_SAFE(c
, &top
->parents
, next_parent
, next
) {
3977 /* Check whether we are allowed to switch c from top to base */
3978 GSList
*ignore_children
= g_slist_prepend(NULL
, c
);
3979 bdrv_check_update_perm(base
, NULL
, c
->perm
, c
->shared_perm
,
3980 ignore_children
, &local_err
);
3981 g_slist_free(ignore_children
);
3984 error_report_err(local_err
);
3988 /* If so, update the backing file path in the image file */
3989 if (c
->role
->update_filename
) {
3990 ret
= c
->role
->update_filename(c
, base
, backing_file_str
,
3993 bdrv_abort_perm_update(base
);
3994 error_report_err(local_err
);
3999 /* Do the actual switch in the in-memory graph.
4000 * Completes bdrv_check_update_perm() transaction internally. */
4002 bdrv_replace_child(c
, base
);
4006 if (update_inherits_from
) {
4007 base
->inherits_from
= explicit_top
->inherits_from
;
4017 * Length of a allocated file in bytes. Sparse files are counted by actual
4018 * allocated space. Return < 0 if error or unknown.
4020 int64_t bdrv_get_allocated_file_size(BlockDriverState
*bs
)
4022 BlockDriver
*drv
= bs
->drv
;
4026 if (drv
->bdrv_get_allocated_file_size
) {
4027 return drv
->bdrv_get_allocated_file_size(bs
);
4030 return bdrv_get_allocated_file_size(bs
->file
->bs
);
4037 * @drv: Format driver
4038 * @opts: Creation options for new image
4039 * @in_bs: Existing image containing data for new image (may be NULL)
4040 * @errp: Error object
4041 * Returns: A #BlockMeasureInfo (free using qapi_free_BlockMeasureInfo())
4044 * Calculate file size required to create a new image.
4046 * If @in_bs is given then space for allocated clusters and zero clusters
4047 * from that image are included in the calculation. If @opts contains a
4048 * backing file that is shared by @in_bs then backing clusters may be omitted
4049 * from the calculation.
4051 * If @in_bs is NULL then the calculation includes no allocated clusters
4052 * unless a preallocation option is given in @opts.
4054 * Note that @in_bs may use a different BlockDriver from @drv.
4056 * If an error occurs the @errp pointer is set.
4058 BlockMeasureInfo
*bdrv_measure(BlockDriver
*drv
, QemuOpts
*opts
,
4059 BlockDriverState
*in_bs
, Error
**errp
)
4061 if (!drv
->bdrv_measure
) {
4062 error_setg(errp
, "Block driver '%s' does not support size measurement",
4067 return drv
->bdrv_measure(opts
, in_bs
, errp
);
4071 * Return number of sectors on success, -errno on error.
4073 int64_t bdrv_nb_sectors(BlockDriverState
*bs
)
4075 BlockDriver
*drv
= bs
->drv
;
4080 if (drv
->has_variable_length
) {
4081 int ret
= refresh_total_sectors(bs
, bs
->total_sectors
);
4086 return bs
->total_sectors
;
4090 * Return length in bytes on success, -errno on error.
4091 * The length is always a multiple of BDRV_SECTOR_SIZE.
4093 int64_t bdrv_getlength(BlockDriverState
*bs
)
4095 int64_t ret
= bdrv_nb_sectors(bs
);
4097 ret
= ret
> INT64_MAX
/ BDRV_SECTOR_SIZE
? -EFBIG
: ret
;
4098 return ret
< 0 ? ret
: ret
* BDRV_SECTOR_SIZE
;
4101 /* return 0 as number of sectors if no device present or error */
4102 void bdrv_get_geometry(BlockDriverState
*bs
, uint64_t *nb_sectors_ptr
)
4104 int64_t nb_sectors
= bdrv_nb_sectors(bs
);
4106 *nb_sectors_ptr
= nb_sectors
< 0 ? 0 : nb_sectors
;
4109 bool bdrv_is_sg(BlockDriverState
*bs
)
4114 bool bdrv_is_encrypted(BlockDriverState
*bs
)
4116 if (bs
->backing
&& bs
->backing
->bs
->encrypted
) {
4119 return bs
->encrypted
;
4122 const char *bdrv_get_format_name(BlockDriverState
*bs
)
4124 return bs
->drv
? bs
->drv
->format_name
: NULL
;
4127 static int qsort_strcmp(const void *a
, const void *b
)
4129 return strcmp(*(char *const *)a
, *(char *const *)b
);
4132 void bdrv_iterate_format(void (*it
)(void *opaque
, const char *name
),
4138 const char **formats
= NULL
;
4140 QLIST_FOREACH(drv
, &bdrv_drivers
, list
) {
4141 if (drv
->format_name
) {
4144 while (formats
&& i
&& !found
) {
4145 found
= !strcmp(formats
[--i
], drv
->format_name
);
4149 formats
= g_renew(const char *, formats
, count
+ 1);
4150 formats
[count
++] = drv
->format_name
;
4155 for (i
= 0; i
< (int)ARRAY_SIZE(block_driver_modules
); i
++) {
4156 const char *format_name
= block_driver_modules
[i
].format_name
;
4162 while (formats
&& j
&& !found
) {
4163 found
= !strcmp(formats
[--j
], format_name
);
4167 formats
= g_renew(const char *, formats
, count
+ 1);
4168 formats
[count
++] = format_name
;
4173 qsort(formats
, count
, sizeof(formats
[0]), qsort_strcmp
);
4175 for (i
= 0; i
< count
; i
++) {
4176 it(opaque
, formats
[i
]);
4182 /* This function is to find a node in the bs graph */
4183 BlockDriverState
*bdrv_find_node(const char *node_name
)
4185 BlockDriverState
*bs
;
4189 QTAILQ_FOREACH(bs
, &graph_bdrv_states
, node_list
) {
4190 if (!strcmp(node_name
, bs
->node_name
)) {
4197 /* Put this QMP function here so it can access the static graph_bdrv_states. */
4198 BlockDeviceInfoList
*bdrv_named_nodes_list(Error
**errp
)
4200 BlockDeviceInfoList
*list
, *entry
;
4201 BlockDriverState
*bs
;
4204 QTAILQ_FOREACH(bs
, &graph_bdrv_states
, node_list
) {
4205 BlockDeviceInfo
*info
= bdrv_block_device_info(NULL
, bs
, errp
);
4207 qapi_free_BlockDeviceInfoList(list
);
4210 entry
= g_malloc0(sizeof(*entry
));
4211 entry
->value
= info
;
4219 #define QAPI_LIST_ADD(list, element) do { \
4220 typeof(list) _tmp = g_new(typeof(*(list)), 1); \
4221 _tmp->value = (element); \
4222 _tmp->next = (list); \
4226 typedef struct XDbgBlockGraphConstructor
{
4227 XDbgBlockGraph
*graph
;
4228 GHashTable
*graph_nodes
;
4229 } XDbgBlockGraphConstructor
;
4231 static XDbgBlockGraphConstructor
*xdbg_graph_new(void)
4233 XDbgBlockGraphConstructor
*gr
= g_new(XDbgBlockGraphConstructor
, 1);
4235 gr
->graph
= g_new0(XDbgBlockGraph
, 1);
4236 gr
->graph_nodes
= g_hash_table_new(NULL
, NULL
);
4241 static XDbgBlockGraph
*xdbg_graph_finalize(XDbgBlockGraphConstructor
*gr
)
4243 XDbgBlockGraph
*graph
= gr
->graph
;
4245 g_hash_table_destroy(gr
->graph_nodes
);
4251 static uintptr_t xdbg_graph_node_num(XDbgBlockGraphConstructor
*gr
, void *node
)
4253 uintptr_t ret
= (uintptr_t)g_hash_table_lookup(gr
->graph_nodes
, node
);
4260 * Start counting from 1, not 0, because 0 interferes with not-found (NULL)
4261 * answer of g_hash_table_lookup.
4263 ret
= g_hash_table_size(gr
->graph_nodes
) + 1;
4264 g_hash_table_insert(gr
->graph_nodes
, node
, (void *)ret
);
4269 static void xdbg_graph_add_node(XDbgBlockGraphConstructor
*gr
, void *node
,
4270 XDbgBlockGraphNodeType type
, const char *name
)
4272 XDbgBlockGraphNode
*n
;
4274 n
= g_new0(XDbgBlockGraphNode
, 1);
4276 n
->id
= xdbg_graph_node_num(gr
, node
);
4278 n
->name
= g_strdup(name
);
4280 QAPI_LIST_ADD(gr
->graph
->nodes
, n
);
4283 static void xdbg_graph_add_edge(XDbgBlockGraphConstructor
*gr
, void *parent
,
4284 const BdrvChild
*child
)
4288 BlockPermission num
;
4291 static const PermissionMap permissions
[] = {
4292 { BLK_PERM_CONSISTENT_READ
, BLOCK_PERMISSION_CONSISTENT_READ
},
4293 { BLK_PERM_WRITE
, BLOCK_PERMISSION_WRITE
},
4294 { BLK_PERM_WRITE_UNCHANGED
, BLOCK_PERMISSION_WRITE_UNCHANGED
},
4295 { BLK_PERM_RESIZE
, BLOCK_PERMISSION_RESIZE
},
4296 { BLK_PERM_GRAPH_MOD
, BLOCK_PERMISSION_GRAPH_MOD
},
4299 const PermissionMap
*p
;
4300 XDbgBlockGraphEdge
*edge
;
4302 QEMU_BUILD_BUG_ON(1UL << (ARRAY_SIZE(permissions
) - 1) != BLK_PERM_ALL
+ 1);
4304 edge
= g_new0(XDbgBlockGraphEdge
, 1);
4306 edge
->parent
= xdbg_graph_node_num(gr
, parent
);
4307 edge
->child
= xdbg_graph_node_num(gr
, child
->bs
);
4308 edge
->name
= g_strdup(child
->name
);
4310 for (p
= permissions
; p
->flag
; p
++) {
4311 if (p
->flag
& child
->perm
) {
4312 QAPI_LIST_ADD(edge
->perm
, p
->num
);
4314 if (p
->flag
& child
->shared_perm
) {
4315 QAPI_LIST_ADD(edge
->shared_perm
, p
->num
);
4319 QAPI_LIST_ADD(gr
->graph
->edges
, edge
);
4323 XDbgBlockGraph
*bdrv_get_xdbg_block_graph(Error
**errp
)
4327 BlockDriverState
*bs
;
4329 XDbgBlockGraphConstructor
*gr
= xdbg_graph_new();
4331 for (blk
= blk_all_next(NULL
); blk
; blk
= blk_all_next(blk
)) {
4332 char *allocated_name
= NULL
;
4333 const char *name
= blk_name(blk
);
4336 name
= allocated_name
= blk_get_attached_dev_id(blk
);
4338 xdbg_graph_add_node(gr
, blk
, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_BACKEND
,
4340 g_free(allocated_name
);
4341 if (blk_root(blk
)) {
4342 xdbg_graph_add_edge(gr
, blk
, blk_root(blk
));
4346 for (job
= block_job_next(NULL
); job
; job
= block_job_next(job
)) {
4349 xdbg_graph_add_node(gr
, job
, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_JOB
,
4351 for (el
= job
->nodes
; el
; el
= el
->next
) {
4352 xdbg_graph_add_edge(gr
, job
, (BdrvChild
*)el
->data
);
4356 QTAILQ_FOREACH(bs
, &graph_bdrv_states
, node_list
) {
4357 xdbg_graph_add_node(gr
, bs
, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_DRIVER
,
4359 QLIST_FOREACH(child
, &bs
->children
, next
) {
4360 xdbg_graph_add_edge(gr
, bs
, child
);
4364 return xdbg_graph_finalize(gr
);
4367 BlockDriverState
*bdrv_lookup_bs(const char *device
,
4368 const char *node_name
,
4372 BlockDriverState
*bs
;
4375 blk
= blk_by_name(device
);
4380 error_setg(errp
, "Device '%s' has no medium", device
);
4388 bs
= bdrv_find_node(node_name
);
4395 error_setg(errp
, "Cannot find device=%s nor node_name=%s",
4396 device
? device
: "",
4397 node_name
? node_name
: "");
4401 /* If 'base' is in the same chain as 'top', return true. Otherwise,
4402 * return false. If either argument is NULL, return false. */
4403 bool bdrv_chain_contains(BlockDriverState
*top
, BlockDriverState
*base
)
4405 while (top
&& top
!= base
) {
4406 top
= backing_bs(top
);
4412 BlockDriverState
*bdrv_next_node(BlockDriverState
*bs
)
4415 return QTAILQ_FIRST(&graph_bdrv_states
);
4417 return QTAILQ_NEXT(bs
, node_list
);
4420 BlockDriverState
*bdrv_next_all_states(BlockDriverState
*bs
)
4423 return QTAILQ_FIRST(&all_bdrv_states
);
4425 return QTAILQ_NEXT(bs
, bs_list
);
4428 const char *bdrv_get_node_name(const BlockDriverState
*bs
)
4430 return bs
->node_name
;
4433 const char *bdrv_get_parent_name(const BlockDriverState
*bs
)
4438 /* If multiple parents have a name, just pick the first one. */
4439 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
4440 if (c
->role
->get_name
) {
4441 name
= c
->role
->get_name(c
);
4442 if (name
&& *name
) {
4451 /* TODO check what callers really want: bs->node_name or blk_name() */
4452 const char *bdrv_get_device_name(const BlockDriverState
*bs
)
4454 return bdrv_get_parent_name(bs
) ?: "";
4457 /* This can be used to identify nodes that might not have a device
4458 * name associated. Since node and device names live in the same
4459 * namespace, the result is unambiguous. The exception is if both are
4460 * absent, then this returns an empty (non-null) string. */
4461 const char *bdrv_get_device_or_node_name(const BlockDriverState
*bs
)
4463 return bdrv_get_parent_name(bs
) ?: bs
->node_name
;
4466 int bdrv_get_flags(BlockDriverState
*bs
)
4468 return bs
->open_flags
;
4471 int bdrv_has_zero_init_1(BlockDriverState
*bs
)
4476 int bdrv_has_zero_init(BlockDriverState
*bs
)
4482 /* If BS is a copy on write image, it is initialized to
4483 the contents of the base image, which may not be zeroes. */
4487 if (bs
->drv
->bdrv_has_zero_init
) {
4488 return bs
->drv
->bdrv_has_zero_init(bs
);
4490 if (bs
->file
&& bs
->drv
->is_filter
) {
4491 return bdrv_has_zero_init(bs
->file
->bs
);
4498 bool bdrv_unallocated_blocks_are_zero(BlockDriverState
*bs
)
4500 BlockDriverInfo bdi
;
4506 if (bdrv_get_info(bs
, &bdi
) == 0) {
4507 return bdi
.unallocated_blocks_are_zero
;
4513 bool bdrv_can_write_zeroes_with_unmap(BlockDriverState
*bs
)
4515 if (!(bs
->open_flags
& BDRV_O_UNMAP
)) {
4519 return bs
->supported_zero_flags
& BDRV_REQ_MAY_UNMAP
;
4522 void bdrv_get_backing_filename(BlockDriverState
*bs
,
4523 char *filename
, int filename_size
)
4525 pstrcpy(filename
, filename_size
, bs
->backing_file
);
4528 int bdrv_get_info(BlockDriverState
*bs
, BlockDriverInfo
*bdi
)
4530 BlockDriver
*drv
= bs
->drv
;
4531 /* if bs->drv == NULL, bs is closed, so there's nothing to do here */
4535 if (!drv
->bdrv_get_info
) {
4536 if (bs
->file
&& drv
->is_filter
) {
4537 return bdrv_get_info(bs
->file
->bs
, bdi
);
4541 memset(bdi
, 0, sizeof(*bdi
));
4542 return drv
->bdrv_get_info(bs
, bdi
);
4545 ImageInfoSpecific
*bdrv_get_specific_info(BlockDriverState
*bs
,
4548 BlockDriver
*drv
= bs
->drv
;
4549 if (drv
&& drv
->bdrv_get_specific_info
) {
4550 return drv
->bdrv_get_specific_info(bs
, errp
);
4555 void bdrv_debug_event(BlockDriverState
*bs
, BlkdebugEvent event
)
4557 if (!bs
|| !bs
->drv
|| !bs
->drv
->bdrv_debug_event
) {
4561 bs
->drv
->bdrv_debug_event(bs
, event
);
4564 int bdrv_debug_breakpoint(BlockDriverState
*bs
, const char *event
,
4567 while (bs
&& bs
->drv
&& !bs
->drv
->bdrv_debug_breakpoint
) {
4568 bs
= bs
->file
? bs
->file
->bs
: NULL
;
4571 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_breakpoint
) {
4572 return bs
->drv
->bdrv_debug_breakpoint(bs
, event
, tag
);
4578 int bdrv_debug_remove_breakpoint(BlockDriverState
*bs
, const char *tag
)
4580 while (bs
&& bs
->drv
&& !bs
->drv
->bdrv_debug_remove_breakpoint
) {
4581 bs
= bs
->file
? bs
->file
->bs
: NULL
;
4584 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_remove_breakpoint
) {
4585 return bs
->drv
->bdrv_debug_remove_breakpoint(bs
, tag
);
4591 int bdrv_debug_resume(BlockDriverState
*bs
, const char *tag
)
4593 while (bs
&& (!bs
->drv
|| !bs
->drv
->bdrv_debug_resume
)) {
4594 bs
= bs
->file
? bs
->file
->bs
: NULL
;
4597 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_resume
) {
4598 return bs
->drv
->bdrv_debug_resume(bs
, tag
);
4604 bool bdrv_debug_is_suspended(BlockDriverState
*bs
, const char *tag
)
4606 while (bs
&& bs
->drv
&& !bs
->drv
->bdrv_debug_is_suspended
) {
4607 bs
= bs
->file
? bs
->file
->bs
: NULL
;
4610 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_is_suspended
) {
4611 return bs
->drv
->bdrv_debug_is_suspended(bs
, tag
);
4617 /* backing_file can either be relative, or absolute, or a protocol. If it is
4618 * relative, it must be relative to the chain. So, passing in bs->filename
4619 * from a BDS as backing_file should not be done, as that may be relative to
4620 * the CWD rather than the chain. */
4621 BlockDriverState
*bdrv_find_backing_image(BlockDriverState
*bs
,
4622 const char *backing_file
)
4624 char *filename_full
= NULL
;
4625 char *backing_file_full
= NULL
;
4626 char *filename_tmp
= NULL
;
4627 int is_protocol
= 0;
4628 BlockDriverState
*curr_bs
= NULL
;
4629 BlockDriverState
*retval
= NULL
;
4630 Error
*local_error
= NULL
;
4632 if (!bs
|| !bs
->drv
|| !backing_file
) {
4636 filename_full
= g_malloc(PATH_MAX
);
4637 backing_file_full
= g_malloc(PATH_MAX
);
4638 filename_tmp
= g_malloc(PATH_MAX
);
4640 is_protocol
= path_has_protocol(backing_file
);
4642 /* This will recursively refresh everything in the backing chain */
4643 bdrv_refresh_filename(bs
);
4645 for (curr_bs
= bs
; curr_bs
->backing
; curr_bs
= curr_bs
->backing
->bs
) {
4647 /* If either of the filename paths is actually a protocol, then
4648 * compare unmodified paths; otherwise make paths relative */
4649 if (is_protocol
|| path_has_protocol(curr_bs
->backing_file
)) {
4650 if (strcmp(backing_file
, curr_bs
->backing_file
) == 0) {
4651 retval
= curr_bs
->backing
->bs
;
4654 /* Also check against the full backing filename for the image */
4655 bdrv_get_full_backing_filename(curr_bs
, backing_file_full
, PATH_MAX
,
4657 if (local_error
== NULL
) {
4658 if (strcmp(backing_file
, backing_file_full
) == 0) {
4659 retval
= curr_bs
->backing
->bs
;
4663 error_free(local_error
);
4667 /* If not an absolute filename path, make it relative to the current
4668 * image's filename path */
4669 path_combine_deprecated(filename_tmp
, PATH_MAX
, curr_bs
->filename
,
4672 /* We are going to compare absolute pathnames */
4673 if (!realpath(filename_tmp
, filename_full
)) {
4677 /* We need to make sure the backing filename we are comparing against
4678 * is relative to the current image filename (or absolute) */
4679 path_combine_deprecated(filename_tmp
, PATH_MAX
, curr_bs
->filename
,
4680 curr_bs
->backing_file
);
4682 if (!realpath(filename_tmp
, backing_file_full
)) {
4686 if (strcmp(backing_file_full
, filename_full
) == 0) {
4687 retval
= curr_bs
->backing
->bs
;
4693 g_free(filename_full
);
4694 g_free(backing_file_full
);
4695 g_free(filename_tmp
);
4699 void bdrv_init(void)
4701 module_call_init(MODULE_INIT_BLOCK
);
4704 void bdrv_init_with_whitelist(void)
4706 use_bdrv_whitelist
= 1;
4710 static void coroutine_fn
bdrv_co_invalidate_cache(BlockDriverState
*bs
,
4713 BdrvChild
*child
, *parent
;
4714 uint64_t perm
, shared_perm
;
4715 Error
*local_err
= NULL
;
4717 BdrvDirtyBitmap
*bm
;
4723 if (!(bs
->open_flags
& BDRV_O_INACTIVE
)) {
4727 QLIST_FOREACH(child
, &bs
->children
, next
) {
4728 bdrv_co_invalidate_cache(child
->bs
, &local_err
);
4730 error_propagate(errp
, local_err
);
4736 * Update permissions, they may differ for inactive nodes.
4738 * Note that the required permissions of inactive images are always a
4739 * subset of the permissions required after activating the image. This
4740 * allows us to just get the permissions upfront without restricting
4741 * drv->bdrv_invalidate_cache().
4743 * It also means that in error cases, we don't have to try and revert to
4744 * the old permissions (which is an operation that could fail, too). We can
4745 * just keep the extended permissions for the next time that an activation
4746 * of the image is tried.
4748 bs
->open_flags
&= ~BDRV_O_INACTIVE
;
4749 bdrv_get_cumulative_perm(bs
, &perm
, &shared_perm
);
4750 ret
= bdrv_check_perm(bs
, NULL
, perm
, shared_perm
, NULL
, &local_err
);
4752 bs
->open_flags
|= BDRV_O_INACTIVE
;
4753 error_propagate(errp
, local_err
);
4756 bdrv_set_perm(bs
, perm
, shared_perm
);
4758 if (bs
->drv
->bdrv_co_invalidate_cache
) {
4759 bs
->drv
->bdrv_co_invalidate_cache(bs
, &local_err
);
4761 bs
->open_flags
|= BDRV_O_INACTIVE
;
4762 error_propagate(errp
, local_err
);
4767 for (bm
= bdrv_dirty_bitmap_next(bs
, NULL
); bm
;
4768 bm
= bdrv_dirty_bitmap_next(bs
, bm
))
4770 bdrv_dirty_bitmap_set_migration(bm
, false);
4773 ret
= refresh_total_sectors(bs
, bs
->total_sectors
);
4775 bs
->open_flags
|= BDRV_O_INACTIVE
;
4776 error_setg_errno(errp
, -ret
, "Could not refresh total sector count");
4780 QLIST_FOREACH(parent
, &bs
->parents
, next_parent
) {
4781 if (parent
->role
->activate
) {
4782 parent
->role
->activate(parent
, &local_err
);
4784 bs
->open_flags
|= BDRV_O_INACTIVE
;
4785 error_propagate(errp
, local_err
);
4792 typedef struct InvalidateCacheCo
{
4793 BlockDriverState
*bs
;
4796 } InvalidateCacheCo
;
4798 static void coroutine_fn
bdrv_invalidate_cache_co_entry(void *opaque
)
4800 InvalidateCacheCo
*ico
= opaque
;
4801 bdrv_co_invalidate_cache(ico
->bs
, ico
->errp
);
4806 void bdrv_invalidate_cache(BlockDriverState
*bs
, Error
**errp
)
4809 InvalidateCacheCo ico
= {
4815 if (qemu_in_coroutine()) {
4816 /* Fast-path if already in coroutine context */
4817 bdrv_invalidate_cache_co_entry(&ico
);
4819 co
= qemu_coroutine_create(bdrv_invalidate_cache_co_entry
, &ico
);
4820 bdrv_coroutine_enter(bs
, co
);
4821 BDRV_POLL_WHILE(bs
, !ico
.done
);
4825 void bdrv_invalidate_cache_all(Error
**errp
)
4827 BlockDriverState
*bs
;
4828 Error
*local_err
= NULL
;
4829 BdrvNextIterator it
;
4831 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
4832 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
4834 aio_context_acquire(aio_context
);
4835 bdrv_invalidate_cache(bs
, &local_err
);
4836 aio_context_release(aio_context
);
4838 error_propagate(errp
, local_err
);
4839 bdrv_next_cleanup(&it
);
4845 static bool bdrv_has_bds_parent(BlockDriverState
*bs
, bool only_active
)
4849 QLIST_FOREACH(parent
, &bs
->parents
, next_parent
) {
4850 if (parent
->role
->parent_is_bds
) {
4851 BlockDriverState
*parent_bs
= parent
->opaque
;
4852 if (!only_active
|| !(parent_bs
->open_flags
& BDRV_O_INACTIVE
)) {
4861 static int bdrv_inactivate_recurse(BlockDriverState
*bs
)
4863 BdrvChild
*child
, *parent
;
4864 uint64_t perm
, shared_perm
;
4871 /* Make sure that we don't inactivate a child before its parent.
4872 * It will be covered by recursion from the yet active parent. */
4873 if (bdrv_has_bds_parent(bs
, true)) {
4877 assert(!(bs
->open_flags
& BDRV_O_INACTIVE
));
4879 /* Inactivate this node */
4880 if (bs
->drv
->bdrv_inactivate
) {
4881 ret
= bs
->drv
->bdrv_inactivate(bs
);
4887 QLIST_FOREACH(parent
, &bs
->parents
, next_parent
) {
4888 if (parent
->role
->inactivate
) {
4889 ret
= parent
->role
->inactivate(parent
);
4896 bs
->open_flags
|= BDRV_O_INACTIVE
;
4898 /* Update permissions, they may differ for inactive nodes */
4899 bdrv_get_cumulative_perm(bs
, &perm
, &shared_perm
);
4900 bdrv_check_perm(bs
, NULL
, perm
, shared_perm
, NULL
, &error_abort
);
4901 bdrv_set_perm(bs
, perm
, shared_perm
);
4904 /* Recursively inactivate children */
4905 QLIST_FOREACH(child
, &bs
->children
, next
) {
4906 ret
= bdrv_inactivate_recurse(child
->bs
);
4915 int bdrv_inactivate_all(void)
4917 BlockDriverState
*bs
= NULL
;
4918 BdrvNextIterator it
;
4920 GSList
*aio_ctxs
= NULL
, *ctx
;
4922 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
4923 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
4925 if (!g_slist_find(aio_ctxs
, aio_context
)) {
4926 aio_ctxs
= g_slist_prepend(aio_ctxs
, aio_context
);
4927 aio_context_acquire(aio_context
);
4931 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
4932 /* Nodes with BDS parents are covered by recursion from the last
4933 * parent that gets inactivated. Don't inactivate them a second
4934 * time if that has already happened. */
4935 if (bdrv_has_bds_parent(bs
, false)) {
4938 ret
= bdrv_inactivate_recurse(bs
);
4940 bdrv_next_cleanup(&it
);
4946 for (ctx
= aio_ctxs
; ctx
!= NULL
; ctx
= ctx
->next
) {
4947 AioContext
*aio_context
= ctx
->data
;
4948 aio_context_release(aio_context
);
4950 g_slist_free(aio_ctxs
);
4955 /**************************************************************/
4956 /* removable device support */
4959 * Return TRUE if the media is present
4961 bool bdrv_is_inserted(BlockDriverState
*bs
)
4963 BlockDriver
*drv
= bs
->drv
;
4969 if (drv
->bdrv_is_inserted
) {
4970 return drv
->bdrv_is_inserted(bs
);
4972 QLIST_FOREACH(child
, &bs
->children
, next
) {
4973 if (!bdrv_is_inserted(child
->bs
)) {
4981 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
4983 void bdrv_eject(BlockDriverState
*bs
, bool eject_flag
)
4985 BlockDriver
*drv
= bs
->drv
;
4987 if (drv
&& drv
->bdrv_eject
) {
4988 drv
->bdrv_eject(bs
, eject_flag
);
4993 * Lock or unlock the media (if it is locked, the user won't be able
4994 * to eject it manually).
4996 void bdrv_lock_medium(BlockDriverState
*bs
, bool locked
)
4998 BlockDriver
*drv
= bs
->drv
;
5000 trace_bdrv_lock_medium(bs
, locked
);
5002 if (drv
&& drv
->bdrv_lock_medium
) {
5003 drv
->bdrv_lock_medium(bs
, locked
);
5007 /* Get a reference to bs */
5008 void bdrv_ref(BlockDriverState
*bs
)
5013 /* Release a previously grabbed reference to bs.
5014 * If after releasing, reference count is zero, the BlockDriverState is
5016 void bdrv_unref(BlockDriverState
*bs
)
5021 assert(bs
->refcnt
> 0);
5022 if (--bs
->refcnt
== 0) {
5027 struct BdrvOpBlocker
{
5029 QLIST_ENTRY(BdrvOpBlocker
) list
;
5032 bool bdrv_op_is_blocked(BlockDriverState
*bs
, BlockOpType op
, Error
**errp
)
5034 BdrvOpBlocker
*blocker
;
5035 assert((int) op
>= 0 && op
< BLOCK_OP_TYPE_MAX
);
5036 if (!QLIST_EMPTY(&bs
->op_blockers
[op
])) {
5037 blocker
= QLIST_FIRST(&bs
->op_blockers
[op
]);
5038 error_propagate_prepend(errp
, error_copy(blocker
->reason
),
5039 "Node '%s' is busy: ",
5040 bdrv_get_device_or_node_name(bs
));
5046 void bdrv_op_block(BlockDriverState
*bs
, BlockOpType op
, Error
*reason
)
5048 BdrvOpBlocker
*blocker
;
5049 assert((int) op
>= 0 && op
< BLOCK_OP_TYPE_MAX
);
5051 blocker
= g_new0(BdrvOpBlocker
, 1);
5052 blocker
->reason
= reason
;
5053 QLIST_INSERT_HEAD(&bs
->op_blockers
[op
], blocker
, list
);
5056 void bdrv_op_unblock(BlockDriverState
*bs
, BlockOpType op
, Error
*reason
)
5058 BdrvOpBlocker
*blocker
, *next
;
5059 assert((int) op
>= 0 && op
< BLOCK_OP_TYPE_MAX
);
5060 QLIST_FOREACH_SAFE(blocker
, &bs
->op_blockers
[op
], list
, next
) {
5061 if (blocker
->reason
== reason
) {
5062 QLIST_REMOVE(blocker
, list
);
5068 void bdrv_op_block_all(BlockDriverState
*bs
, Error
*reason
)
5071 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
5072 bdrv_op_block(bs
, i
, reason
);
5076 void bdrv_op_unblock_all(BlockDriverState
*bs
, Error
*reason
)
5079 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
5080 bdrv_op_unblock(bs
, i
, reason
);
5084 bool bdrv_op_blocker_is_empty(BlockDriverState
*bs
)
5088 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
5089 if (!QLIST_EMPTY(&bs
->op_blockers
[i
])) {
5096 void bdrv_img_create(const char *filename
, const char *fmt
,
5097 const char *base_filename
, const char *base_fmt
,
5098 char *options
, uint64_t img_size
, int flags
, bool quiet
,
5101 QemuOptsList
*create_opts
= NULL
;
5102 QemuOpts
*opts
= NULL
;
5103 const char *backing_fmt
, *backing_file
;
5105 BlockDriver
*drv
, *proto_drv
;
5106 Error
*local_err
= NULL
;
5109 /* Find driver and parse its options */
5110 drv
= bdrv_find_format(fmt
);
5112 error_setg(errp
, "Unknown file format '%s'", fmt
);
5116 proto_drv
= bdrv_find_protocol(filename
, true, errp
);
5121 if (!drv
->create_opts
) {
5122 error_setg(errp
, "Format driver '%s' does not support image creation",
5127 if (!proto_drv
->create_opts
) {
5128 error_setg(errp
, "Protocol driver '%s' does not support image creation",
5129 proto_drv
->format_name
);
5133 create_opts
= qemu_opts_append(create_opts
, drv
->create_opts
);
5134 create_opts
= qemu_opts_append(create_opts
, proto_drv
->create_opts
);
5136 /* Create parameter list with default values */
5137 opts
= qemu_opts_create(create_opts
, NULL
, 0, &error_abort
);
5138 qemu_opt_set_number(opts
, BLOCK_OPT_SIZE
, img_size
, &error_abort
);
5140 /* Parse -o options */
5142 qemu_opts_do_parse(opts
, options
, NULL
, &local_err
);
5148 if (base_filename
) {
5149 qemu_opt_set(opts
, BLOCK_OPT_BACKING_FILE
, base_filename
, &local_err
);
5151 error_setg(errp
, "Backing file not supported for file format '%s'",
5158 qemu_opt_set(opts
, BLOCK_OPT_BACKING_FMT
, base_fmt
, &local_err
);
5160 error_setg(errp
, "Backing file format not supported for file "
5161 "format '%s'", fmt
);
5166 backing_file
= qemu_opt_get(opts
, BLOCK_OPT_BACKING_FILE
);
5168 if (!strcmp(filename
, backing_file
)) {
5169 error_setg(errp
, "Error: Trying to create an image with the "
5170 "same filename as the backing file");
5175 backing_fmt
= qemu_opt_get(opts
, BLOCK_OPT_BACKING_FMT
);
5177 /* The size for the image must always be specified, unless we have a backing
5178 * file and we have not been forbidden from opening it. */
5179 size
= qemu_opt_get_size(opts
, BLOCK_OPT_SIZE
, img_size
);
5180 if (backing_file
&& !(flags
& BDRV_O_NO_BACKING
)) {
5181 BlockDriverState
*bs
;
5182 char *full_backing
= g_new0(char, PATH_MAX
);
5184 QDict
*backing_options
= NULL
;
5186 bdrv_get_full_backing_filename_from_filename(filename
, backing_file
,
5187 full_backing
, PATH_MAX
,
5190 g_free(full_backing
);
5194 /* backing files always opened read-only */
5196 back_flags
&= ~(BDRV_O_RDWR
| BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
);
5198 backing_options
= qdict_new();
5200 qdict_put_str(backing_options
, "driver", backing_fmt
);
5202 qdict_put_bool(backing_options
, BDRV_OPT_FORCE_SHARE
, true);
5204 bs
= bdrv_open(full_backing
, NULL
, backing_options
, back_flags
,
5206 g_free(full_backing
);
5207 if (!bs
&& size
!= -1) {
5208 /* Couldn't open BS, but we have a size, so it's nonfatal */
5209 warn_reportf_err(local_err
,
5210 "Could not verify backing image. "
5211 "This may become an error in future versions.\n");
5214 /* Couldn't open bs, do not have size */
5215 error_append_hint(&local_err
,
5216 "Could not open backing image to determine size.\n");
5220 /* Opened BS, have no size */
5221 size
= bdrv_getlength(bs
);
5223 error_setg_errno(errp
, -size
, "Could not get size of '%s'",
5228 qemu_opt_set_number(opts
, BLOCK_OPT_SIZE
, size
, &error_abort
);
5232 } /* (backing_file && !(flags & BDRV_O_NO_BACKING)) */
5235 error_setg(errp
, "Image creation needs a size parameter");
5240 printf("Formatting '%s', fmt=%s ", filename
, fmt
);
5241 qemu_opts_print(opts
, " ");
5245 ret
= bdrv_create(drv
, filename
, opts
, &local_err
);
5247 if (ret
== -EFBIG
) {
5248 /* This is generally a better message than whatever the driver would
5249 * deliver (especially because of the cluster_size_hint), since that
5250 * is most probably not much different from "image too large". */
5251 const char *cluster_size_hint
= "";
5252 if (qemu_opt_get_size(opts
, BLOCK_OPT_CLUSTER_SIZE
, 0)) {
5253 cluster_size_hint
= " (try using a larger cluster size)";
5255 error_setg(errp
, "The image size is too large for file format '%s'"
5256 "%s", fmt
, cluster_size_hint
);
5257 error_free(local_err
);
5262 qemu_opts_del(opts
);
5263 qemu_opts_free(create_opts
);
5264 error_propagate(errp
, local_err
);
5267 AioContext
*bdrv_get_aio_context(BlockDriverState
*bs
)
5269 return bs
? bs
->aio_context
: qemu_get_aio_context();
5272 void bdrv_coroutine_enter(BlockDriverState
*bs
, Coroutine
*co
)
5274 aio_co_enter(bdrv_get_aio_context(bs
), co
);
5277 static void bdrv_do_remove_aio_context_notifier(BdrvAioNotifier
*ban
)
5279 QLIST_REMOVE(ban
, list
);
5283 void bdrv_detach_aio_context(BlockDriverState
*bs
)
5285 BdrvAioNotifier
*baf
, *baf_tmp
;
5292 assert(!bs
->walking_aio_notifiers
);
5293 bs
->walking_aio_notifiers
= true;
5294 QLIST_FOREACH_SAFE(baf
, &bs
->aio_notifiers
, list
, baf_tmp
) {
5296 bdrv_do_remove_aio_context_notifier(baf
);
5298 baf
->detach_aio_context(baf
->opaque
);
5301 /* Never mind iterating again to check for ->deleted. bdrv_close() will
5302 * remove remaining aio notifiers if we aren't called again.
5304 bs
->walking_aio_notifiers
= false;
5306 if (bs
->drv
->bdrv_detach_aio_context
) {
5307 bs
->drv
->bdrv_detach_aio_context(bs
);
5309 QLIST_FOREACH(child
, &bs
->children
, next
) {
5310 bdrv_detach_aio_context(child
->bs
);
5313 if (bs
->quiesce_counter
) {
5314 aio_enable_external(bs
->aio_context
);
5316 bs
->aio_context
= NULL
;
5319 void bdrv_attach_aio_context(BlockDriverState
*bs
,
5320 AioContext
*new_context
)
5322 BdrvAioNotifier
*ban
, *ban_tmp
;
5329 if (bs
->quiesce_counter
) {
5330 aio_disable_external(new_context
);
5333 bs
->aio_context
= new_context
;
5335 QLIST_FOREACH(child
, &bs
->children
, next
) {
5336 bdrv_attach_aio_context(child
->bs
, new_context
);
5338 if (bs
->drv
->bdrv_attach_aio_context
) {
5339 bs
->drv
->bdrv_attach_aio_context(bs
, new_context
);
5342 assert(!bs
->walking_aio_notifiers
);
5343 bs
->walking_aio_notifiers
= true;
5344 QLIST_FOREACH_SAFE(ban
, &bs
->aio_notifiers
, list
, ban_tmp
) {
5346 bdrv_do_remove_aio_context_notifier(ban
);
5348 ban
->attached_aio_context(new_context
, ban
->opaque
);
5351 bs
->walking_aio_notifiers
= false;
5354 /* The caller must own the AioContext lock for the old AioContext of bs, but it
5355 * must not own the AioContext lock for new_context (unless new_context is
5356 * the same as the current context of bs). */
5357 void bdrv_set_aio_context(BlockDriverState
*bs
, AioContext
*new_context
)
5359 if (bdrv_get_aio_context(bs
) == new_context
) {
5363 bdrv_drained_begin(bs
);
5364 bdrv_detach_aio_context(bs
);
5366 /* This function executes in the old AioContext so acquire the new one in
5367 * case it runs in a different thread.
5369 aio_context_acquire(new_context
);
5370 bdrv_attach_aio_context(bs
, new_context
);
5371 bdrv_drained_end(bs
);
5372 aio_context_release(new_context
);
5375 void bdrv_add_aio_context_notifier(BlockDriverState
*bs
,
5376 void (*attached_aio_context
)(AioContext
*new_context
, void *opaque
),
5377 void (*detach_aio_context
)(void *opaque
), void *opaque
)
5379 BdrvAioNotifier
*ban
= g_new(BdrvAioNotifier
, 1);
5380 *ban
= (BdrvAioNotifier
){
5381 .attached_aio_context
= attached_aio_context
,
5382 .detach_aio_context
= detach_aio_context
,
5386 QLIST_INSERT_HEAD(&bs
->aio_notifiers
, ban
, list
);
5389 void bdrv_remove_aio_context_notifier(BlockDriverState
*bs
,
5390 void (*attached_aio_context
)(AioContext
*,
5392 void (*detach_aio_context
)(void *),
5395 BdrvAioNotifier
*ban
, *ban_next
;
5397 QLIST_FOREACH_SAFE(ban
, &bs
->aio_notifiers
, list
, ban_next
) {
5398 if (ban
->attached_aio_context
== attached_aio_context
&&
5399 ban
->detach_aio_context
== detach_aio_context
&&
5400 ban
->opaque
== opaque
&&
5401 ban
->deleted
== false)
5403 if (bs
->walking_aio_notifiers
) {
5404 ban
->deleted
= true;
5406 bdrv_do_remove_aio_context_notifier(ban
);
5415 int bdrv_amend_options(BlockDriverState
*bs
, QemuOpts
*opts
,
5416 BlockDriverAmendStatusCB
*status_cb
, void *cb_opaque
,
5420 error_setg(errp
, "Node is ejected");
5423 if (!bs
->drv
->bdrv_amend_options
) {
5424 error_setg(errp
, "Block driver '%s' does not support option amendment",
5425 bs
->drv
->format_name
);
5428 return bs
->drv
->bdrv_amend_options(bs
, opts
, status_cb
, cb_opaque
, errp
);
5431 /* This function will be called by the bdrv_recurse_is_first_non_filter method
5432 * of block filter and by bdrv_is_first_non_filter.
5433 * It is used to test if the given bs is the candidate or recurse more in the
5436 bool bdrv_recurse_is_first_non_filter(BlockDriverState
*bs
,
5437 BlockDriverState
*candidate
)
5439 /* return false if basic checks fails */
5440 if (!bs
|| !bs
->drv
) {
5444 /* the code reached a non block filter driver -> check if the bs is
5445 * the same as the candidate. It's the recursion termination condition.
5447 if (!bs
->drv
->is_filter
) {
5448 return bs
== candidate
;
5450 /* Down this path the driver is a block filter driver */
5452 /* If the block filter recursion method is defined use it to recurse down
5455 if (bs
->drv
->bdrv_recurse_is_first_non_filter
) {
5456 return bs
->drv
->bdrv_recurse_is_first_non_filter(bs
, candidate
);
5459 /* the driver is a block filter but don't allow to recurse -> return false
5464 /* This function checks if the candidate is the first non filter bs down it's
5465 * bs chain. Since we don't have pointers to parents it explore all bs chains
5466 * from the top. Some filters can choose not to pass down the recursion.
5468 bool bdrv_is_first_non_filter(BlockDriverState
*candidate
)
5470 BlockDriverState
*bs
;
5471 BdrvNextIterator it
;
5473 /* walk down the bs forest recursively */
5474 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
5477 /* try to recurse in this top level bs */
5478 perm
= bdrv_recurse_is_first_non_filter(bs
, candidate
);
5480 /* candidate is the first non filter */
5482 bdrv_next_cleanup(&it
);
5490 BlockDriverState
*check_to_replace_node(BlockDriverState
*parent_bs
,
5491 const char *node_name
, Error
**errp
)
5493 BlockDriverState
*to_replace_bs
= bdrv_find_node(node_name
);
5494 AioContext
*aio_context
;
5496 if (!to_replace_bs
) {
5497 error_setg(errp
, "Node name '%s' not found", node_name
);
5501 aio_context
= bdrv_get_aio_context(to_replace_bs
);
5502 aio_context_acquire(aio_context
);
5504 if (bdrv_op_is_blocked(to_replace_bs
, BLOCK_OP_TYPE_REPLACE
, errp
)) {
5505 to_replace_bs
= NULL
;
5509 /* We don't want arbitrary node of the BDS chain to be replaced only the top
5510 * most non filter in order to prevent data corruption.
5511 * Another benefit is that this tests exclude backing files which are
5512 * blocked by the backing blockers.
5514 if (!bdrv_recurse_is_first_non_filter(parent_bs
, to_replace_bs
)) {
5515 error_setg(errp
, "Only top most non filter can be replaced");
5516 to_replace_bs
= NULL
;
5521 aio_context_release(aio_context
);
5522 return to_replace_bs
;
5525 static bool append_open_options(QDict
*d
, BlockDriverState
*bs
)
5527 const QDictEntry
*entry
;
5529 bool found_any
= false;
5531 for (entry
= qdict_first(bs
->options
); entry
;
5532 entry
= qdict_next(bs
->options
, entry
))
5534 /* Exclude all non-driver-specific options */
5535 for (desc
= bdrv_runtime_opts
.desc
; desc
->name
; desc
++) {
5536 if (!strcmp(qdict_entry_key(entry
), desc
->name
)) {
5544 qdict_put_obj(d
, qdict_entry_key(entry
),
5545 qobject_ref(qdict_entry_value(entry
)));
5552 /* Note: This function may return false positives; it may return true
5553 * even if opening the backing file specified by bs's image header
5554 * would result in exactly bs->backing. */
5555 static bool bdrv_backing_overridden(BlockDriverState
*bs
)
5558 return strcmp(bs
->auto_backing_file
,
5559 bs
->backing
->bs
->filename
);
5561 /* No backing BDS, so if the image header reports any backing
5562 * file, it must have been suppressed */
5563 return bs
->auto_backing_file
[0] != '\0';
5567 /* Updates the following BDS fields:
5568 * - exact_filename: A filename which may be used for opening a block device
5569 * which (mostly) equals the given BDS (even without any
5570 * other options; so reading and writing must return the same
5571 * results, but caching etc. may be different)
5572 * - full_open_options: Options which, when given when opening a block device
5573 * (without a filename), result in a BDS (mostly)
5574 * equalling the given one
5575 * - filename: If exact_filename is set, it is copied here. Otherwise,
5576 * full_open_options is converted to a JSON object, prefixed with
5577 * "json:" (for use through the JSON pseudo protocol) and put here.
5579 void bdrv_refresh_filename(BlockDriverState
*bs
)
5581 BlockDriver
*drv
= bs
->drv
;
5584 bool backing_overridden
;
5590 /* This BDS's file name may depend on any of its children's file names, so
5591 * refresh those first */
5592 QLIST_FOREACH(child
, &bs
->children
, next
) {
5593 bdrv_refresh_filename(child
->bs
);
5597 /* For implicit nodes, just copy everything from the single child */
5598 child
= QLIST_FIRST(&bs
->children
);
5599 assert(QLIST_NEXT(child
, next
) == NULL
);
5601 pstrcpy(bs
->exact_filename
, sizeof(bs
->exact_filename
),
5602 child
->bs
->exact_filename
);
5603 pstrcpy(bs
->filename
, sizeof(bs
->filename
), child
->bs
->filename
);
5605 bs
->full_open_options
= qobject_ref(child
->bs
->full_open_options
);
5610 backing_overridden
= bdrv_backing_overridden(bs
);
5612 if (bs
->open_flags
& BDRV_O_NO_IO
) {
5613 /* Without I/O, the backing file does not change anything.
5614 * Therefore, in such a case (primarily qemu-img), we can
5615 * pretend the backing file has not been overridden even if
5616 * it technically has been. */
5617 backing_overridden
= false;
5620 if (drv
->bdrv_refresh_filename
) {
5621 /* Obsolete information is of no use here, so drop the old file name
5622 * information before refreshing it */
5623 bs
->exact_filename
[0] = '\0';
5624 if (bs
->full_open_options
) {
5625 qobject_unref(bs
->full_open_options
);
5626 bs
->full_open_options
= NULL
;
5630 append_open_options(opts
, bs
);
5631 drv
->bdrv_refresh_filename(bs
, opts
);
5632 qobject_unref(opts
);
5633 } else if (bs
->file
) {
5634 /* Try to reconstruct valid information from the underlying file */
5635 bool has_open_options
;
5637 bs
->exact_filename
[0] = '\0';
5638 if (bs
->full_open_options
) {
5639 qobject_unref(bs
->full_open_options
);
5640 bs
->full_open_options
= NULL
;
5644 has_open_options
= append_open_options(opts
, bs
);
5645 has_open_options
|= backing_overridden
;
5647 /* If no specific options have been given for this BDS, the filename of
5648 * the underlying file should suffice for this one as well */
5649 if (bs
->file
->bs
->exact_filename
[0] && !has_open_options
) {
5650 strcpy(bs
->exact_filename
, bs
->file
->bs
->exact_filename
);
5652 /* Reconstructing the full options QDict is simple for most format block
5653 * drivers, as long as the full options are known for the underlying
5654 * file BDS. The full options QDict of that file BDS should somehow
5655 * contain a representation of the filename, therefore the following
5656 * suffices without querying the (exact_)filename of this BDS. */
5657 if (bs
->file
->bs
->full_open_options
&&
5658 (!bs
->backing
|| bs
->backing
->bs
->full_open_options
))
5660 qdict_put_str(opts
, "driver", drv
->format_name
);
5661 qdict_put(opts
, "file",
5662 qobject_ref(bs
->file
->bs
->full_open_options
));
5665 qdict_put(opts
, "backing",
5666 qobject_ref(bs
->backing
->bs
->full_open_options
));
5667 } else if (backing_overridden
) {
5668 qdict_put_null(opts
, "backing");
5671 bs
->full_open_options
= opts
;
5673 qobject_unref(opts
);
5675 } else if (!bs
->full_open_options
&& qdict_size(bs
->options
)) {
5676 /* There is no underlying file BDS (at least referenced by BDS.file),
5677 * so the full options QDict should be equal to the options given
5678 * specifically for this block device when it was opened (plus the
5679 * driver specification).
5680 * Because those options don't change, there is no need to update
5681 * full_open_options when it's already set. */
5684 append_open_options(opts
, bs
);
5685 qdict_put_str(opts
, "driver", drv
->format_name
);
5687 if (bs
->exact_filename
[0]) {
5688 /* This may not work for all block protocol drivers (some may
5689 * require this filename to be parsed), but we have to find some
5690 * default solution here, so just include it. If some block driver
5691 * does not support pure options without any filename at all or
5692 * needs some special format of the options QDict, it needs to
5693 * implement the driver-specific bdrv_refresh_filename() function.
5695 qdict_put_str(opts
, "filename", bs
->exact_filename
);
5698 bs
->full_open_options
= opts
;
5701 if (bs
->exact_filename
[0]) {
5702 pstrcpy(bs
->filename
, sizeof(bs
->filename
), bs
->exact_filename
);
5703 } else if (bs
->full_open_options
) {
5704 QString
*json
= qobject_to_json(QOBJECT(bs
->full_open_options
));
5705 snprintf(bs
->filename
, sizeof(bs
->filename
), "json:%s",
5706 qstring_get_str(json
));
5707 qobject_unref(json
);
5712 * Hot add/remove a BDS's child. So the user can take a child offline when
5713 * it is broken and take a new child online
5715 void bdrv_add_child(BlockDriverState
*parent_bs
, BlockDriverState
*child_bs
,
5719 if (!parent_bs
->drv
|| !parent_bs
->drv
->bdrv_add_child
) {
5720 error_setg(errp
, "The node %s does not support adding a child",
5721 bdrv_get_device_or_node_name(parent_bs
));
5725 if (!QLIST_EMPTY(&child_bs
->parents
)) {
5726 error_setg(errp
, "The node %s already has a parent",
5727 child_bs
->node_name
);
5731 parent_bs
->drv
->bdrv_add_child(parent_bs
, child_bs
, errp
);
5734 void bdrv_del_child(BlockDriverState
*parent_bs
, BdrvChild
*child
, Error
**errp
)
5738 if (!parent_bs
->drv
|| !parent_bs
->drv
->bdrv_del_child
) {
5739 error_setg(errp
, "The node %s does not support removing a child",
5740 bdrv_get_device_or_node_name(parent_bs
));
5744 QLIST_FOREACH(tmp
, &parent_bs
->children
, next
) {
5751 error_setg(errp
, "The node %s does not have a child named %s",
5752 bdrv_get_device_or_node_name(parent_bs
),
5753 bdrv_get_device_or_node_name(child
->bs
));
5757 parent_bs
->drv
->bdrv_del_child(parent_bs
, child
, errp
);
5760 bool bdrv_can_store_new_dirty_bitmap(BlockDriverState
*bs
, const char *name
,
5761 uint32_t granularity
, Error
**errp
)
5763 BlockDriver
*drv
= bs
->drv
;
5766 error_setg_errno(errp
, ENOMEDIUM
,
5767 "Can't store persistent bitmaps to %s",
5768 bdrv_get_device_or_node_name(bs
));
5772 if (!drv
->bdrv_can_store_new_dirty_bitmap
) {
5773 error_setg_errno(errp
, ENOTSUP
,
5774 "Can't store persistent bitmaps to %s",
5775 bdrv_get_device_or_node_name(bs
));
5779 return drv
->bdrv_can_store_new_dirty_bitmap(bs
, name
, granularity
, errp
);