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
);
205 * Helper function for bdrv_parse_filename() implementations to remove optional
206 * protocol prefixes (especially "file:") from a filename and for putting the
207 * stripped filename into the options QDict if there is such a prefix.
209 void bdrv_parse_filename_strip_prefix(const char *filename
, const char *prefix
,
212 if (strstart(filename
, prefix
, &filename
)) {
213 /* Stripping the explicit protocol prefix may result in a protocol
214 * prefix being (wrongly) detected (if the filename contains a colon) */
215 if (path_has_protocol(filename
)) {
216 QString
*fat_filename
;
218 /* This means there is some colon before the first slash; therefore,
219 * this cannot be an absolute path */
220 assert(!path_is_absolute(filename
));
222 /* And we can thus fix the protocol detection issue by prefixing it
224 fat_filename
= qstring_from_str("./");
225 qstring_append(fat_filename
, filename
);
227 assert(!path_has_protocol(qstring_get_str(fat_filename
)));
229 qdict_put(options
, "filename", fat_filename
);
231 /* If no protocol prefix was detected, we can use the shortened
233 qdict_put_str(options
, "filename", filename
);
239 /* Returns whether the image file is opened as read-only. Note that this can
240 * return false and writing to the image file is still not possible because the
241 * image is inactivated. */
242 bool bdrv_is_read_only(BlockDriverState
*bs
)
244 return bs
->read_only
;
247 int bdrv_can_set_read_only(BlockDriverState
*bs
, bool read_only
,
248 bool ignore_allow_rdw
, Error
**errp
)
250 /* Do not set read_only if copy_on_read is enabled */
251 if (bs
->copy_on_read
&& read_only
) {
252 error_setg(errp
, "Can't set node '%s' to r/o with copy-on-read enabled",
253 bdrv_get_device_or_node_name(bs
));
257 /* Do not clear read_only if it is prohibited */
258 if (!read_only
&& !(bs
->open_flags
& BDRV_O_ALLOW_RDWR
) &&
261 error_setg(errp
, "Node '%s' is read only",
262 bdrv_get_device_or_node_name(bs
));
270 * Called by a driver that can only provide a read-only image.
272 * Returns 0 if the node is already read-only or it could switch the node to
273 * read-only because BDRV_O_AUTO_RDONLY is set.
275 * Returns -EACCES if the node is read-write and BDRV_O_AUTO_RDONLY is not set
276 * or bdrv_can_set_read_only() forbids making the node read-only. If @errmsg
277 * is not NULL, it is used as the error message for the Error object.
279 int bdrv_apply_auto_read_only(BlockDriverState
*bs
, const char *errmsg
,
284 if (!(bs
->open_flags
& BDRV_O_RDWR
)) {
287 if (!(bs
->open_flags
& BDRV_O_AUTO_RDONLY
)) {
291 ret
= bdrv_can_set_read_only(bs
, true, false, NULL
);
296 bs
->read_only
= true;
297 bs
->open_flags
&= ~BDRV_O_RDWR
;
302 error_setg(errp
, "%s", errmsg
?: "Image is read-only");
307 * If @backing is empty, this function returns NULL without setting
308 * @errp. In all other cases, NULL will only be returned with @errp
311 * Therefore, a return value of NULL without @errp set means that
312 * there is no backing file; if @errp is set, there is one but its
313 * absolute filename cannot be generated.
315 char *bdrv_get_full_backing_filename_from_filename(const char *backed
,
319 if (backing
[0] == '\0') {
321 } else if (path_has_protocol(backing
) || path_is_absolute(backing
)) {
322 return g_strdup(backing
);
323 } else if (backed
[0] == '\0' || strstart(backed
, "json:", NULL
)) {
324 error_setg(errp
, "Cannot use relative backing file names for '%s'",
328 return path_combine(backed
, backing
);
333 * If @filename is empty or NULL, this function returns NULL without
334 * setting @errp. In all other cases, NULL will only be returned with
337 static char *bdrv_make_absolute_filename(BlockDriverState
*relative_to
,
338 const char *filename
, Error
**errp
)
340 char *dir
, *full_name
;
342 if (!filename
|| filename
[0] == '\0') {
344 } else if (path_has_protocol(filename
) || path_is_absolute(filename
)) {
345 return g_strdup(filename
);
348 dir
= bdrv_dirname(relative_to
, errp
);
353 full_name
= g_strconcat(dir
, filename
, NULL
);
358 char *bdrv_get_full_backing_filename(BlockDriverState
*bs
, Error
**errp
)
360 return bdrv_make_absolute_filename(bs
, bs
->backing_file
, errp
);
363 void bdrv_register(BlockDriver
*bdrv
)
365 QLIST_INSERT_HEAD(&bdrv_drivers
, bdrv
, list
);
368 BlockDriverState
*bdrv_new(void)
370 BlockDriverState
*bs
;
373 bs
= g_new0(BlockDriverState
, 1);
374 QLIST_INIT(&bs
->dirty_bitmaps
);
375 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
376 QLIST_INIT(&bs
->op_blockers
[i
]);
378 notifier_with_return_list_init(&bs
->before_write_notifiers
);
379 qemu_co_mutex_init(&bs
->reqs_lock
);
380 qemu_mutex_init(&bs
->dirty_bitmap_mutex
);
382 bs
->aio_context
= qemu_get_aio_context();
384 qemu_co_queue_init(&bs
->flush_queue
);
386 for (i
= 0; i
< bdrv_drain_all_count
; i
++) {
387 bdrv_drained_begin(bs
);
390 QTAILQ_INSERT_TAIL(&all_bdrv_states
, bs
, bs_list
);
395 static BlockDriver
*bdrv_do_find_format(const char *format_name
)
399 QLIST_FOREACH(drv1
, &bdrv_drivers
, list
) {
400 if (!strcmp(drv1
->format_name
, format_name
)) {
408 BlockDriver
*bdrv_find_format(const char *format_name
)
413 drv1
= bdrv_do_find_format(format_name
);
418 /* The driver isn't registered, maybe we need to load a module */
419 for (i
= 0; i
< (int)ARRAY_SIZE(block_driver_modules
); ++i
) {
420 if (!strcmp(block_driver_modules
[i
].format_name
, format_name
)) {
421 block_module_load_one(block_driver_modules
[i
].library_name
);
426 return bdrv_do_find_format(format_name
);
429 static int bdrv_format_is_whitelisted(const char *format_name
, bool read_only
)
431 static const char *whitelist_rw
[] = {
432 CONFIG_BDRV_RW_WHITELIST
434 static const char *whitelist_ro
[] = {
435 CONFIG_BDRV_RO_WHITELIST
439 if (!whitelist_rw
[0] && !whitelist_ro
[0]) {
440 return 1; /* no whitelist, anything goes */
443 for (p
= whitelist_rw
; *p
; p
++) {
444 if (!strcmp(format_name
, *p
)) {
449 for (p
= whitelist_ro
; *p
; p
++) {
450 if (!strcmp(format_name
, *p
)) {
458 int bdrv_is_whitelisted(BlockDriver
*drv
, bool read_only
)
460 return bdrv_format_is_whitelisted(drv
->format_name
, read_only
);
463 bool bdrv_uses_whitelist(void)
465 return use_bdrv_whitelist
;
468 typedef struct CreateCo
{
476 static void coroutine_fn
bdrv_create_co_entry(void *opaque
)
478 Error
*local_err
= NULL
;
481 CreateCo
*cco
= opaque
;
484 ret
= cco
->drv
->bdrv_co_create_opts(cco
->filename
, cco
->opts
, &local_err
);
485 error_propagate(&cco
->err
, local_err
);
489 int bdrv_create(BlockDriver
*drv
, const char* filename
,
490 QemuOpts
*opts
, Error
**errp
)
497 .filename
= g_strdup(filename
),
503 if (!drv
->bdrv_co_create_opts
) {
504 error_setg(errp
, "Driver '%s' does not support image creation", drv
->format_name
);
509 if (qemu_in_coroutine()) {
510 /* Fast-path if already in coroutine context */
511 bdrv_create_co_entry(&cco
);
513 co
= qemu_coroutine_create(bdrv_create_co_entry
, &cco
);
514 qemu_coroutine_enter(co
);
515 while (cco
.ret
== NOT_DONE
) {
516 aio_poll(qemu_get_aio_context(), true);
523 error_propagate(errp
, cco
.err
);
525 error_setg_errno(errp
, -ret
, "Could not create image");
530 g_free(cco
.filename
);
534 int bdrv_create_file(const char *filename
, QemuOpts
*opts
, Error
**errp
)
537 Error
*local_err
= NULL
;
540 drv
= bdrv_find_protocol(filename
, true, errp
);
545 ret
= bdrv_create(drv
, filename
, opts
, &local_err
);
546 error_propagate(errp
, local_err
);
551 * Try to get @bs's logical and physical block size.
552 * On success, store them in @bsz struct and return 0.
553 * On failure return -errno.
554 * @bs must not be empty.
556 int bdrv_probe_blocksizes(BlockDriverState
*bs
, BlockSizes
*bsz
)
558 BlockDriver
*drv
= bs
->drv
;
560 if (drv
&& drv
->bdrv_probe_blocksizes
) {
561 return drv
->bdrv_probe_blocksizes(bs
, bsz
);
562 } else if (drv
&& drv
->is_filter
&& bs
->file
) {
563 return bdrv_probe_blocksizes(bs
->file
->bs
, bsz
);
570 * Try to get @bs's geometry (cyls, heads, sectors).
571 * On success, store them in @geo struct and return 0.
572 * On failure return -errno.
573 * @bs must not be empty.
575 int bdrv_probe_geometry(BlockDriverState
*bs
, HDGeometry
*geo
)
577 BlockDriver
*drv
= bs
->drv
;
579 if (drv
&& drv
->bdrv_probe_geometry
) {
580 return drv
->bdrv_probe_geometry(bs
, geo
);
581 } else if (drv
&& drv
->is_filter
&& bs
->file
) {
582 return bdrv_probe_geometry(bs
->file
->bs
, geo
);
589 * Create a uniquely-named empty temporary file.
590 * Return 0 upon success, otherwise a negative errno value.
592 int get_tmp_filename(char *filename
, int size
)
595 char temp_dir
[MAX_PATH
];
596 /* GetTempFileName requires that its output buffer (4th param)
597 have length MAX_PATH or greater. */
598 assert(size
>= MAX_PATH
);
599 return (GetTempPath(MAX_PATH
, temp_dir
)
600 && GetTempFileName(temp_dir
, "qem", 0, filename
)
601 ? 0 : -GetLastError());
605 tmpdir
= getenv("TMPDIR");
609 if (snprintf(filename
, size
, "%s/vl.XXXXXX", tmpdir
) >= size
) {
612 fd
= mkstemp(filename
);
616 if (close(fd
) != 0) {
625 * Detect host devices. By convention, /dev/cdrom[N] is always
626 * recognized as a host CDROM.
628 static BlockDriver
*find_hdev_driver(const char *filename
)
630 int score_max
= 0, score
;
631 BlockDriver
*drv
= NULL
, *d
;
633 QLIST_FOREACH(d
, &bdrv_drivers
, list
) {
634 if (d
->bdrv_probe_device
) {
635 score
= d
->bdrv_probe_device(filename
);
636 if (score
> score_max
) {
646 static BlockDriver
*bdrv_do_find_protocol(const char *protocol
)
650 QLIST_FOREACH(drv1
, &bdrv_drivers
, list
) {
651 if (drv1
->protocol_name
&& !strcmp(drv1
->protocol_name
, protocol
)) {
659 BlockDriver
*bdrv_find_protocol(const char *filename
,
660 bool allow_protocol_prefix
,
669 /* TODO Drivers without bdrv_file_open must be specified explicitly */
672 * XXX(hch): we really should not let host device detection
673 * override an explicit protocol specification, but moving this
674 * later breaks access to device names with colons in them.
675 * Thanks to the brain-dead persistent naming schemes on udev-
676 * based Linux systems those actually are quite common.
678 drv1
= find_hdev_driver(filename
);
683 if (!path_has_protocol(filename
) || !allow_protocol_prefix
) {
687 p
= strchr(filename
, ':');
690 if (len
> sizeof(protocol
) - 1)
691 len
= sizeof(protocol
) - 1;
692 memcpy(protocol
, filename
, len
);
693 protocol
[len
] = '\0';
695 drv1
= bdrv_do_find_protocol(protocol
);
700 for (i
= 0; i
< (int)ARRAY_SIZE(block_driver_modules
); ++i
) {
701 if (block_driver_modules
[i
].protocol_name
&&
702 !strcmp(block_driver_modules
[i
].protocol_name
, protocol
)) {
703 block_module_load_one(block_driver_modules
[i
].library_name
);
708 drv1
= bdrv_do_find_protocol(protocol
);
710 error_setg(errp
, "Unknown protocol '%s'", protocol
);
716 * Guess image format by probing its contents.
717 * This is not a good idea when your image is raw (CVE-2008-2004), but
718 * we do it anyway for backward compatibility.
720 * @buf contains the image's first @buf_size bytes.
721 * @buf_size is the buffer size in bytes (generally BLOCK_PROBE_BUF_SIZE,
722 * but can be smaller if the image file is smaller)
723 * @filename is its filename.
725 * For all block drivers, call the bdrv_probe() method to get its
727 * Return the first block driver with the highest probing score.
729 BlockDriver
*bdrv_probe_all(const uint8_t *buf
, int buf_size
,
730 const char *filename
)
732 int score_max
= 0, score
;
733 BlockDriver
*drv
= NULL
, *d
;
735 QLIST_FOREACH(d
, &bdrv_drivers
, list
) {
737 score
= d
->bdrv_probe(buf
, buf_size
, filename
);
738 if (score
> score_max
) {
748 static int find_image_format(BlockBackend
*file
, const char *filename
,
749 BlockDriver
**pdrv
, Error
**errp
)
752 uint8_t buf
[BLOCK_PROBE_BUF_SIZE
];
755 /* Return the raw BlockDriver * to scsi-generic devices or empty drives */
756 if (blk_is_sg(file
) || !blk_is_inserted(file
) || blk_getlength(file
) == 0) {
761 ret
= blk_pread(file
, 0, buf
, sizeof(buf
));
763 error_setg_errno(errp
, -ret
, "Could not read image for determining its "
769 drv
= bdrv_probe_all(buf
, ret
, filename
);
771 error_setg(errp
, "Could not determine image format: No compatible "
780 * Set the current 'total_sectors' value
781 * Return 0 on success, -errno on error.
783 int refresh_total_sectors(BlockDriverState
*bs
, int64_t hint
)
785 BlockDriver
*drv
= bs
->drv
;
791 /* Do not attempt drv->bdrv_getlength() on scsi-generic devices */
795 /* query actual device if possible, otherwise just trust the hint */
796 if (drv
->bdrv_getlength
) {
797 int64_t length
= drv
->bdrv_getlength(bs
);
801 hint
= DIV_ROUND_UP(length
, BDRV_SECTOR_SIZE
);
804 bs
->total_sectors
= hint
;
809 * Combines a QDict of new block driver @options with any missing options taken
810 * from @old_options, so that leaving out an option defaults to its old value.
812 static void bdrv_join_options(BlockDriverState
*bs
, QDict
*options
,
815 if (bs
->drv
&& bs
->drv
->bdrv_join_options
) {
816 bs
->drv
->bdrv_join_options(options
, old_options
);
818 qdict_join(options
, old_options
, false);
822 static BlockdevDetectZeroesOptions
bdrv_parse_detect_zeroes(QemuOpts
*opts
,
826 Error
*local_err
= NULL
;
827 char *value
= qemu_opt_get_del(opts
, "detect-zeroes");
828 BlockdevDetectZeroesOptions detect_zeroes
=
829 qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup
, value
,
830 BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF
, &local_err
);
833 error_propagate(errp
, local_err
);
834 return detect_zeroes
;
837 if (detect_zeroes
== BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP
&&
838 !(open_flags
& BDRV_O_UNMAP
))
840 error_setg(errp
, "setting detect-zeroes to unmap is not allowed "
841 "without setting discard operation to unmap");
844 return detect_zeroes
;
848 * Set open flags for a given discard mode
850 * Return 0 on success, -1 if the discard mode was invalid.
852 int bdrv_parse_discard_flags(const char *mode
, int *flags
)
854 *flags
&= ~BDRV_O_UNMAP
;
856 if (!strcmp(mode
, "off") || !strcmp(mode
, "ignore")) {
858 } else if (!strcmp(mode
, "on") || !strcmp(mode
, "unmap")) {
859 *flags
|= BDRV_O_UNMAP
;
868 * Set open flags for a given cache mode
870 * Return 0 on success, -1 if the cache mode was invalid.
872 int bdrv_parse_cache_mode(const char *mode
, int *flags
, bool *writethrough
)
874 *flags
&= ~BDRV_O_CACHE_MASK
;
876 if (!strcmp(mode
, "off") || !strcmp(mode
, "none")) {
877 *writethrough
= false;
878 *flags
|= BDRV_O_NOCACHE
;
879 } else if (!strcmp(mode
, "directsync")) {
880 *writethrough
= true;
881 *flags
|= BDRV_O_NOCACHE
;
882 } else if (!strcmp(mode
, "writeback")) {
883 *writethrough
= false;
884 } else if (!strcmp(mode
, "unsafe")) {
885 *writethrough
= false;
886 *flags
|= BDRV_O_NO_FLUSH
;
887 } else if (!strcmp(mode
, "writethrough")) {
888 *writethrough
= true;
896 static char *bdrv_child_get_parent_desc(BdrvChild
*c
)
898 BlockDriverState
*parent
= c
->opaque
;
899 return g_strdup(bdrv_get_device_or_node_name(parent
));
902 static void bdrv_child_cb_drained_begin(BdrvChild
*child
)
904 BlockDriverState
*bs
= child
->opaque
;
905 bdrv_do_drained_begin_quiesce(bs
, NULL
, false);
908 static bool bdrv_child_cb_drained_poll(BdrvChild
*child
)
910 BlockDriverState
*bs
= child
->opaque
;
911 return bdrv_drain_poll(bs
, false, NULL
, false);
914 static void bdrv_child_cb_drained_end(BdrvChild
*child
)
916 BlockDriverState
*bs
= child
->opaque
;
917 bdrv_drained_end(bs
);
920 static void bdrv_child_cb_attach(BdrvChild
*child
)
922 BlockDriverState
*bs
= child
->opaque
;
923 bdrv_apply_subtree_drain(child
, bs
);
926 static void bdrv_child_cb_detach(BdrvChild
*child
)
928 BlockDriverState
*bs
= child
->opaque
;
929 bdrv_unapply_subtree_drain(child
, bs
);
932 static int bdrv_child_cb_inactivate(BdrvChild
*child
)
934 BlockDriverState
*bs
= child
->opaque
;
935 assert(bs
->open_flags
& BDRV_O_INACTIVE
);
939 static bool bdrv_child_cb_can_set_aio_ctx(BdrvChild
*child
, AioContext
*ctx
,
940 GSList
**ignore
, Error
**errp
)
942 BlockDriverState
*bs
= child
->opaque
;
943 return bdrv_can_set_aio_context(bs
, ctx
, ignore
, errp
);
946 static void bdrv_child_cb_set_aio_ctx(BdrvChild
*child
, AioContext
*ctx
,
949 BlockDriverState
*bs
= child
->opaque
;
950 return bdrv_set_aio_context_ignore(bs
, ctx
, ignore
);
954 * Returns the options and flags that a temporary snapshot should get, based on
955 * the originally requested flags (the originally requested image will have
956 * flags like a backing file)
958 static void bdrv_temp_snapshot_options(int *child_flags
, QDict
*child_options
,
959 int parent_flags
, QDict
*parent_options
)
961 *child_flags
= (parent_flags
& ~BDRV_O_SNAPSHOT
) | BDRV_O_TEMPORARY
;
963 /* For temporary files, unconditional cache=unsafe is fine */
964 qdict_set_default_str(child_options
, BDRV_OPT_CACHE_DIRECT
, "off");
965 qdict_set_default_str(child_options
, BDRV_OPT_CACHE_NO_FLUSH
, "on");
967 /* Copy the read-only and discard options from the parent */
968 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_READ_ONLY
);
969 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_DISCARD
);
971 /* aio=native doesn't work for cache.direct=off, so disable it for the
972 * temporary snapshot */
973 *child_flags
&= ~BDRV_O_NATIVE_AIO
;
977 * Returns the options and flags that bs->file should get if a protocol driver
978 * is expected, based on the given options and flags for the parent BDS
980 static void bdrv_inherited_options(int *child_flags
, QDict
*child_options
,
981 int parent_flags
, QDict
*parent_options
)
983 int flags
= parent_flags
;
985 /* Enable protocol handling, disable format probing for bs->file */
986 flags
|= BDRV_O_PROTOCOL
;
988 /* If the cache mode isn't explicitly set, inherit direct and no-flush from
990 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_CACHE_DIRECT
);
991 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_CACHE_NO_FLUSH
);
992 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_FORCE_SHARE
);
994 /* Inherit the read-only option from the parent if it's not set */
995 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_READ_ONLY
);
996 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_AUTO_READ_ONLY
);
998 /* Our block drivers take care to send flushes and respect unmap policy,
999 * so we can default to enable both on lower layers regardless of the
1000 * corresponding parent options. */
1001 qdict_set_default_str(child_options
, BDRV_OPT_DISCARD
, "unmap");
1003 /* Clear flags that only apply to the top layer */
1004 flags
&= ~(BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
| BDRV_O_COPY_ON_READ
|
1007 *child_flags
= flags
;
1010 const BdrvChildRole child_file
= {
1011 .parent_is_bds
= true,
1012 .get_parent_desc
= bdrv_child_get_parent_desc
,
1013 .inherit_options
= bdrv_inherited_options
,
1014 .drained_begin
= bdrv_child_cb_drained_begin
,
1015 .drained_poll
= bdrv_child_cb_drained_poll
,
1016 .drained_end
= bdrv_child_cb_drained_end
,
1017 .attach
= bdrv_child_cb_attach
,
1018 .detach
= bdrv_child_cb_detach
,
1019 .inactivate
= bdrv_child_cb_inactivate
,
1020 .can_set_aio_ctx
= bdrv_child_cb_can_set_aio_ctx
,
1021 .set_aio_ctx
= bdrv_child_cb_set_aio_ctx
,
1025 * Returns the options and flags that bs->file should get if the use of formats
1026 * (and not only protocols) is permitted for it, based on the given options and
1027 * flags for the parent BDS
1029 static void bdrv_inherited_fmt_options(int *child_flags
, QDict
*child_options
,
1030 int parent_flags
, QDict
*parent_options
)
1032 child_file
.inherit_options(child_flags
, child_options
,
1033 parent_flags
, parent_options
);
1035 *child_flags
&= ~(BDRV_O_PROTOCOL
| BDRV_O_NO_IO
);
1038 const BdrvChildRole child_format
= {
1039 .parent_is_bds
= true,
1040 .get_parent_desc
= bdrv_child_get_parent_desc
,
1041 .inherit_options
= bdrv_inherited_fmt_options
,
1042 .drained_begin
= bdrv_child_cb_drained_begin
,
1043 .drained_poll
= bdrv_child_cb_drained_poll
,
1044 .drained_end
= bdrv_child_cb_drained_end
,
1045 .attach
= bdrv_child_cb_attach
,
1046 .detach
= bdrv_child_cb_detach
,
1047 .inactivate
= bdrv_child_cb_inactivate
,
1048 .can_set_aio_ctx
= bdrv_child_cb_can_set_aio_ctx
,
1049 .set_aio_ctx
= bdrv_child_cb_set_aio_ctx
,
1052 static void bdrv_backing_attach(BdrvChild
*c
)
1054 BlockDriverState
*parent
= c
->opaque
;
1055 BlockDriverState
*backing_hd
= c
->bs
;
1057 assert(!parent
->backing_blocker
);
1058 error_setg(&parent
->backing_blocker
,
1059 "node is used as backing hd of '%s'",
1060 bdrv_get_device_or_node_name(parent
));
1062 bdrv_refresh_filename(backing_hd
);
1064 parent
->open_flags
&= ~BDRV_O_NO_BACKING
;
1065 pstrcpy(parent
->backing_file
, sizeof(parent
->backing_file
),
1066 backing_hd
->filename
);
1067 pstrcpy(parent
->backing_format
, sizeof(parent
->backing_format
),
1068 backing_hd
->drv
? backing_hd
->drv
->format_name
: "");
1070 bdrv_op_block_all(backing_hd
, parent
->backing_blocker
);
1071 /* Otherwise we won't be able to commit or stream */
1072 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_COMMIT_TARGET
,
1073 parent
->backing_blocker
);
1074 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_STREAM
,
1075 parent
->backing_blocker
);
1077 * We do backup in 3 ways:
1079 * The target bs is new opened, and the source is top BDS
1080 * 2. blockdev backup
1081 * Both the source and the target are top BDSes.
1082 * 3. internal backup(used for block replication)
1083 * Both the source and the target are backing file
1085 * In case 1 and 2, neither the source nor the target is the backing file.
1086 * In case 3, we will block the top BDS, so there is only one block job
1087 * for the top BDS and its backing chain.
1089 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_BACKUP_SOURCE
,
1090 parent
->backing_blocker
);
1091 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_BACKUP_TARGET
,
1092 parent
->backing_blocker
);
1094 bdrv_child_cb_attach(c
);
1097 static void bdrv_backing_detach(BdrvChild
*c
)
1099 BlockDriverState
*parent
= c
->opaque
;
1101 assert(parent
->backing_blocker
);
1102 bdrv_op_unblock_all(c
->bs
, parent
->backing_blocker
);
1103 error_free(parent
->backing_blocker
);
1104 parent
->backing_blocker
= NULL
;
1106 bdrv_child_cb_detach(c
);
1110 * Returns the options and flags that bs->backing should get, based on the
1111 * given options and flags for the parent BDS
1113 static void bdrv_backing_options(int *child_flags
, QDict
*child_options
,
1114 int parent_flags
, QDict
*parent_options
)
1116 int flags
= parent_flags
;
1118 /* The cache mode is inherited unmodified for backing files; except WCE,
1119 * which is only applied on the top level (BlockBackend) */
1120 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_CACHE_DIRECT
);
1121 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_CACHE_NO_FLUSH
);
1122 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_FORCE_SHARE
);
1124 /* backing files always opened read-only */
1125 qdict_set_default_str(child_options
, BDRV_OPT_READ_ONLY
, "on");
1126 qdict_set_default_str(child_options
, BDRV_OPT_AUTO_READ_ONLY
, "off");
1127 flags
&= ~BDRV_O_COPY_ON_READ
;
1129 /* snapshot=on is handled on the top layer */
1130 flags
&= ~(BDRV_O_SNAPSHOT
| BDRV_O_TEMPORARY
);
1132 *child_flags
= flags
;
1135 static int bdrv_backing_update_filename(BdrvChild
*c
, BlockDriverState
*base
,
1136 const char *filename
, Error
**errp
)
1138 BlockDriverState
*parent
= c
->opaque
;
1139 bool read_only
= bdrv_is_read_only(parent
);
1143 ret
= bdrv_reopen_set_read_only(parent
, false, errp
);
1149 ret
= bdrv_change_backing_file(parent
, filename
,
1150 base
->drv
? base
->drv
->format_name
: "");
1152 error_setg_errno(errp
, -ret
, "Could not update backing file link");
1156 bdrv_reopen_set_read_only(parent
, true, NULL
);
1162 const BdrvChildRole child_backing
= {
1163 .parent_is_bds
= true,
1164 .get_parent_desc
= bdrv_child_get_parent_desc
,
1165 .attach
= bdrv_backing_attach
,
1166 .detach
= bdrv_backing_detach
,
1167 .inherit_options
= bdrv_backing_options
,
1168 .drained_begin
= bdrv_child_cb_drained_begin
,
1169 .drained_poll
= bdrv_child_cb_drained_poll
,
1170 .drained_end
= bdrv_child_cb_drained_end
,
1171 .inactivate
= bdrv_child_cb_inactivate
,
1172 .update_filename
= bdrv_backing_update_filename
,
1173 .can_set_aio_ctx
= bdrv_child_cb_can_set_aio_ctx
,
1174 .set_aio_ctx
= bdrv_child_cb_set_aio_ctx
,
1177 static int bdrv_open_flags(BlockDriverState
*bs
, int flags
)
1179 int open_flags
= flags
;
1182 * Clear flags that are internal to the block layer before opening the
1185 open_flags
&= ~(BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
| BDRV_O_PROTOCOL
);
1190 static void update_flags_from_options(int *flags
, QemuOpts
*opts
)
1192 *flags
&= ~(BDRV_O_CACHE_MASK
| BDRV_O_RDWR
| BDRV_O_AUTO_RDONLY
);
1194 if (qemu_opt_get_bool_del(opts
, BDRV_OPT_CACHE_NO_FLUSH
, false)) {
1195 *flags
|= BDRV_O_NO_FLUSH
;
1198 if (qemu_opt_get_bool_del(opts
, BDRV_OPT_CACHE_DIRECT
, false)) {
1199 *flags
|= BDRV_O_NOCACHE
;
1202 if (!qemu_opt_get_bool_del(opts
, BDRV_OPT_READ_ONLY
, false)) {
1203 *flags
|= BDRV_O_RDWR
;
1206 if (qemu_opt_get_bool_del(opts
, BDRV_OPT_AUTO_READ_ONLY
, false)) {
1207 *flags
|= BDRV_O_AUTO_RDONLY
;
1211 static void update_options_from_flags(QDict
*options
, int flags
)
1213 if (!qdict_haskey(options
, BDRV_OPT_CACHE_DIRECT
)) {
1214 qdict_put_bool(options
, BDRV_OPT_CACHE_DIRECT
, flags
& BDRV_O_NOCACHE
);
1216 if (!qdict_haskey(options
, BDRV_OPT_CACHE_NO_FLUSH
)) {
1217 qdict_put_bool(options
, BDRV_OPT_CACHE_NO_FLUSH
,
1218 flags
& BDRV_O_NO_FLUSH
);
1220 if (!qdict_haskey(options
, BDRV_OPT_READ_ONLY
)) {
1221 qdict_put_bool(options
, BDRV_OPT_READ_ONLY
, !(flags
& BDRV_O_RDWR
));
1223 if (!qdict_haskey(options
, BDRV_OPT_AUTO_READ_ONLY
)) {
1224 qdict_put_bool(options
, BDRV_OPT_AUTO_READ_ONLY
,
1225 flags
& BDRV_O_AUTO_RDONLY
);
1229 static void bdrv_assign_node_name(BlockDriverState
*bs
,
1230 const char *node_name
,
1233 char *gen_node_name
= NULL
;
1236 node_name
= gen_node_name
= id_generate(ID_BLOCK
);
1237 } else if (!id_wellformed(node_name
)) {
1239 * Check for empty string or invalid characters, but not if it is
1240 * generated (generated names use characters not available to the user)
1242 error_setg(errp
, "Invalid node name");
1246 /* takes care of avoiding namespaces collisions */
1247 if (blk_by_name(node_name
)) {
1248 error_setg(errp
, "node-name=%s is conflicting with a device id",
1253 /* takes care of avoiding duplicates node names */
1254 if (bdrv_find_node(node_name
)) {
1255 error_setg(errp
, "Duplicate node name");
1259 /* Make sure that the node name isn't truncated */
1260 if (strlen(node_name
) >= sizeof(bs
->node_name
)) {
1261 error_setg(errp
, "Node name too long");
1265 /* copy node name into the bs and insert it into the graph list */
1266 pstrcpy(bs
->node_name
, sizeof(bs
->node_name
), node_name
);
1267 QTAILQ_INSERT_TAIL(&graph_bdrv_states
, bs
, node_list
);
1269 g_free(gen_node_name
);
1272 static int bdrv_open_driver(BlockDriverState
*bs
, BlockDriver
*drv
,
1273 const char *node_name
, QDict
*options
,
1274 int open_flags
, Error
**errp
)
1276 Error
*local_err
= NULL
;
1279 bdrv_assign_node_name(bs
, node_name
, &local_err
);
1281 error_propagate(errp
, local_err
);
1286 bs
->read_only
= !(bs
->open_flags
& BDRV_O_RDWR
);
1287 bs
->opaque
= g_malloc0(drv
->instance_size
);
1289 if (drv
->bdrv_file_open
) {
1290 assert(!drv
->bdrv_needs_filename
|| bs
->filename
[0]);
1291 ret
= drv
->bdrv_file_open(bs
, options
, open_flags
, &local_err
);
1292 } else if (drv
->bdrv_open
) {
1293 ret
= drv
->bdrv_open(bs
, options
, open_flags
, &local_err
);
1300 error_propagate(errp
, local_err
);
1301 } else if (bs
->filename
[0]) {
1302 error_setg_errno(errp
, -ret
, "Could not open '%s'", bs
->filename
);
1304 error_setg_errno(errp
, -ret
, "Could not open image");
1309 ret
= refresh_total_sectors(bs
, bs
->total_sectors
);
1311 error_setg_errno(errp
, -ret
, "Could not refresh total sector count");
1315 bdrv_refresh_limits(bs
, &local_err
);
1317 error_propagate(errp
, local_err
);
1321 assert(bdrv_opt_mem_align(bs
) != 0);
1322 assert(bdrv_min_mem_align(bs
) != 0);
1323 assert(is_power_of_2(bs
->bl
.request_alignment
));
1325 for (i
= 0; i
< bs
->quiesce_counter
; i
++) {
1326 if (drv
->bdrv_co_drain_begin
) {
1327 drv
->bdrv_co_drain_begin(bs
);
1334 if (bs
->file
!= NULL
) {
1335 bdrv_unref_child(bs
, bs
->file
);
1343 BlockDriverState
*bdrv_new_open_driver(BlockDriver
*drv
, const char *node_name
,
1344 int flags
, Error
**errp
)
1346 BlockDriverState
*bs
;
1350 bs
->open_flags
= flags
;
1351 bs
->explicit_options
= qdict_new();
1352 bs
->options
= qdict_new();
1355 update_options_from_flags(bs
->options
, flags
);
1357 ret
= bdrv_open_driver(bs
, drv
, node_name
, bs
->options
, flags
, errp
);
1359 qobject_unref(bs
->explicit_options
);
1360 bs
->explicit_options
= NULL
;
1361 qobject_unref(bs
->options
);
1370 QemuOptsList bdrv_runtime_opts
= {
1371 .name
= "bdrv_common",
1372 .head
= QTAILQ_HEAD_INITIALIZER(bdrv_runtime_opts
.head
),
1375 .name
= "node-name",
1376 .type
= QEMU_OPT_STRING
,
1377 .help
= "Node name of the block device node",
1381 .type
= QEMU_OPT_STRING
,
1382 .help
= "Block driver to use for the node",
1385 .name
= BDRV_OPT_CACHE_DIRECT
,
1386 .type
= QEMU_OPT_BOOL
,
1387 .help
= "Bypass software writeback cache on the host",
1390 .name
= BDRV_OPT_CACHE_NO_FLUSH
,
1391 .type
= QEMU_OPT_BOOL
,
1392 .help
= "Ignore flush requests",
1395 .name
= BDRV_OPT_READ_ONLY
,
1396 .type
= QEMU_OPT_BOOL
,
1397 .help
= "Node is opened in read-only mode",
1400 .name
= BDRV_OPT_AUTO_READ_ONLY
,
1401 .type
= QEMU_OPT_BOOL
,
1402 .help
= "Node can become read-only if opening read-write fails",
1405 .name
= "detect-zeroes",
1406 .type
= QEMU_OPT_STRING
,
1407 .help
= "try to optimize zero writes (off, on, unmap)",
1410 .name
= BDRV_OPT_DISCARD
,
1411 .type
= QEMU_OPT_STRING
,
1412 .help
= "discard operation (ignore/off, unmap/on)",
1415 .name
= BDRV_OPT_FORCE_SHARE
,
1416 .type
= QEMU_OPT_BOOL
,
1417 .help
= "always accept other writers (default: off)",
1419 { /* end of list */ }
1424 * Common part for opening disk images and files
1426 * Removes all processed options from *options.
1428 static int bdrv_open_common(BlockDriverState
*bs
, BlockBackend
*file
,
1429 QDict
*options
, Error
**errp
)
1431 int ret
, open_flags
;
1432 const char *filename
;
1433 const char *driver_name
= NULL
;
1434 const char *node_name
= NULL
;
1435 const char *discard
;
1438 Error
*local_err
= NULL
;
1440 assert(bs
->file
== NULL
);
1441 assert(options
!= NULL
&& bs
->options
!= options
);
1443 opts
= qemu_opts_create(&bdrv_runtime_opts
, NULL
, 0, &error_abort
);
1444 qemu_opts_absorb_qdict(opts
, options
, &local_err
);
1446 error_propagate(errp
, local_err
);
1451 update_flags_from_options(&bs
->open_flags
, opts
);
1453 driver_name
= qemu_opt_get(opts
, "driver");
1454 drv
= bdrv_find_format(driver_name
);
1455 assert(drv
!= NULL
);
1457 bs
->force_share
= qemu_opt_get_bool(opts
, BDRV_OPT_FORCE_SHARE
, false);
1459 if (bs
->force_share
&& (bs
->open_flags
& BDRV_O_RDWR
)) {
1461 BDRV_OPT_FORCE_SHARE
1462 "=on can only be used with read-only images");
1468 bdrv_refresh_filename(blk_bs(file
));
1469 filename
= blk_bs(file
)->filename
;
1472 * Caution: while qdict_get_try_str() is fine, getting
1473 * non-string types would require more care. When @options
1474 * come from -blockdev or blockdev_add, its members are typed
1475 * according to the QAPI schema, but when they come from
1476 * -drive, they're all QString.
1478 filename
= qdict_get_try_str(options
, "filename");
1481 if (drv
->bdrv_needs_filename
&& (!filename
|| !filename
[0])) {
1482 error_setg(errp
, "The '%s' block driver requires a file name",
1488 trace_bdrv_open_common(bs
, filename
?: "", bs
->open_flags
,
1491 bs
->read_only
= !(bs
->open_flags
& BDRV_O_RDWR
);
1493 if (use_bdrv_whitelist
&& !bdrv_is_whitelisted(drv
, bs
->read_only
)) {
1494 if (!bs
->read_only
&& bdrv_is_whitelisted(drv
, true)) {
1495 ret
= bdrv_apply_auto_read_only(bs
, NULL
, NULL
);
1501 !bs
->read_only
&& bdrv_is_whitelisted(drv
, true)
1502 ? "Driver '%s' can only be used for read-only devices"
1503 : "Driver '%s' is not whitelisted",
1509 /* bdrv_new() and bdrv_close() make it so */
1510 assert(atomic_read(&bs
->copy_on_read
) == 0);
1512 if (bs
->open_flags
& BDRV_O_COPY_ON_READ
) {
1513 if (!bs
->read_only
) {
1514 bdrv_enable_copy_on_read(bs
);
1516 error_setg(errp
, "Can't use copy-on-read on read-only device");
1522 discard
= qemu_opt_get(opts
, BDRV_OPT_DISCARD
);
1523 if (discard
!= NULL
) {
1524 if (bdrv_parse_discard_flags(discard
, &bs
->open_flags
) != 0) {
1525 error_setg(errp
, "Invalid discard option");
1532 bdrv_parse_detect_zeroes(opts
, bs
->open_flags
, &local_err
);
1534 error_propagate(errp
, local_err
);
1539 if (filename
!= NULL
) {
1540 pstrcpy(bs
->filename
, sizeof(bs
->filename
), filename
);
1542 bs
->filename
[0] = '\0';
1544 pstrcpy(bs
->exact_filename
, sizeof(bs
->exact_filename
), bs
->filename
);
1546 /* Open the image, either directly or using a protocol */
1547 open_flags
= bdrv_open_flags(bs
, bs
->open_flags
);
1548 node_name
= qemu_opt_get(opts
, "node-name");
1550 assert(!drv
->bdrv_file_open
|| file
== NULL
);
1551 ret
= bdrv_open_driver(bs
, drv
, node_name
, options
, open_flags
, errp
);
1556 qemu_opts_del(opts
);
1560 qemu_opts_del(opts
);
1564 static QDict
*parse_json_filename(const char *filename
, Error
**errp
)
1566 QObject
*options_obj
;
1570 ret
= strstart(filename
, "json:", &filename
);
1573 options_obj
= qobject_from_json(filename
, errp
);
1575 error_prepend(errp
, "Could not parse the JSON options: ");
1579 options
= qobject_to(QDict
, options_obj
);
1581 qobject_unref(options_obj
);
1582 error_setg(errp
, "Invalid JSON object given");
1586 qdict_flatten(options
);
1591 static void parse_json_protocol(QDict
*options
, const char **pfilename
,
1594 QDict
*json_options
;
1595 Error
*local_err
= NULL
;
1597 /* Parse json: pseudo-protocol */
1598 if (!*pfilename
|| !g_str_has_prefix(*pfilename
, "json:")) {
1602 json_options
= parse_json_filename(*pfilename
, &local_err
);
1604 error_propagate(errp
, local_err
);
1608 /* Options given in the filename have lower priority than options
1609 * specified directly */
1610 qdict_join(options
, json_options
, false);
1611 qobject_unref(json_options
);
1616 * Fills in default options for opening images and converts the legacy
1617 * filename/flags pair to option QDict entries.
1618 * The BDRV_O_PROTOCOL flag in *flags will be set or cleared accordingly if a
1619 * block driver has been specified explicitly.
1621 static int bdrv_fill_options(QDict
**options
, const char *filename
,
1622 int *flags
, Error
**errp
)
1624 const char *drvname
;
1625 bool protocol
= *flags
& BDRV_O_PROTOCOL
;
1626 bool parse_filename
= false;
1627 BlockDriver
*drv
= NULL
;
1628 Error
*local_err
= NULL
;
1631 * Caution: while qdict_get_try_str() is fine, getting non-string
1632 * types would require more care. When @options come from
1633 * -blockdev or blockdev_add, its members are typed according to
1634 * the QAPI schema, but when they come from -drive, they're all
1637 drvname
= qdict_get_try_str(*options
, "driver");
1639 drv
= bdrv_find_format(drvname
);
1641 error_setg(errp
, "Unknown driver '%s'", drvname
);
1644 /* If the user has explicitly specified the driver, this choice should
1645 * override the BDRV_O_PROTOCOL flag */
1646 protocol
= drv
->bdrv_file_open
;
1650 *flags
|= BDRV_O_PROTOCOL
;
1652 *flags
&= ~BDRV_O_PROTOCOL
;
1655 /* Translate cache options from flags into options */
1656 update_options_from_flags(*options
, *flags
);
1658 /* Fetch the file name from the options QDict if necessary */
1659 if (protocol
&& filename
) {
1660 if (!qdict_haskey(*options
, "filename")) {
1661 qdict_put_str(*options
, "filename", filename
);
1662 parse_filename
= true;
1664 error_setg(errp
, "Can't specify 'file' and 'filename' options at "
1670 /* Find the right block driver */
1671 /* See cautionary note on accessing @options above */
1672 filename
= qdict_get_try_str(*options
, "filename");
1674 if (!drvname
&& protocol
) {
1676 drv
= bdrv_find_protocol(filename
, parse_filename
, errp
);
1681 drvname
= drv
->format_name
;
1682 qdict_put_str(*options
, "driver", drvname
);
1684 error_setg(errp
, "Must specify either driver or file");
1689 assert(drv
|| !protocol
);
1691 /* Driver-specific filename parsing */
1692 if (drv
&& drv
->bdrv_parse_filename
&& parse_filename
) {
1693 drv
->bdrv_parse_filename(filename
, *options
, &local_err
);
1695 error_propagate(errp
, local_err
);
1699 if (!drv
->bdrv_needs_filename
) {
1700 qdict_del(*options
, "filename");
1707 static int bdrv_child_check_perm(BdrvChild
*c
, BlockReopenQueue
*q
,
1708 uint64_t perm
, uint64_t shared
,
1709 GSList
*ignore_children
,
1710 bool *tighten_restrictions
, Error
**errp
);
1711 static void bdrv_child_abort_perm_update(BdrvChild
*c
);
1712 static void bdrv_child_set_perm(BdrvChild
*c
, uint64_t perm
, uint64_t shared
);
1713 static void bdrv_get_cumulative_perm(BlockDriverState
*bs
, uint64_t *perm
,
1714 uint64_t *shared_perm
);
1716 typedef struct BlockReopenQueueEntry
{
1719 BDRVReopenState state
;
1720 QSIMPLEQ_ENTRY(BlockReopenQueueEntry
) entry
;
1721 } BlockReopenQueueEntry
;
1724 * Return the flags that @bs will have after the reopens in @q have
1725 * successfully completed. If @q is NULL (or @bs is not contained in @q),
1726 * return the current flags.
1728 static int bdrv_reopen_get_flags(BlockReopenQueue
*q
, BlockDriverState
*bs
)
1730 BlockReopenQueueEntry
*entry
;
1733 QSIMPLEQ_FOREACH(entry
, q
, entry
) {
1734 if (entry
->state
.bs
== bs
) {
1735 return entry
->state
.flags
;
1740 return bs
->open_flags
;
1743 /* Returns whether the image file can be written to after the reopen queue @q
1744 * has been successfully applied, or right now if @q is NULL. */
1745 static bool bdrv_is_writable_after_reopen(BlockDriverState
*bs
,
1746 BlockReopenQueue
*q
)
1748 int flags
= bdrv_reopen_get_flags(q
, bs
);
1750 return (flags
& (BDRV_O_RDWR
| BDRV_O_INACTIVE
)) == BDRV_O_RDWR
;
1754 * Return whether the BDS can be written to. This is not necessarily
1755 * the same as !bdrv_is_read_only(bs), as inactivated images may not
1756 * be written to but do not count as read-only images.
1758 bool bdrv_is_writable(BlockDriverState
*bs
)
1760 return bdrv_is_writable_after_reopen(bs
, NULL
);
1763 static void bdrv_child_perm(BlockDriverState
*bs
, BlockDriverState
*child_bs
,
1764 BdrvChild
*c
, const BdrvChildRole
*role
,
1765 BlockReopenQueue
*reopen_queue
,
1766 uint64_t parent_perm
, uint64_t parent_shared
,
1767 uint64_t *nperm
, uint64_t *nshared
)
1769 assert(bs
->drv
&& bs
->drv
->bdrv_child_perm
);
1770 bs
->drv
->bdrv_child_perm(bs
, c
, role
, reopen_queue
,
1771 parent_perm
, parent_shared
,
1773 /* TODO Take force_share from reopen_queue */
1774 if (child_bs
&& child_bs
->force_share
) {
1775 *nshared
= BLK_PERM_ALL
;
1780 * Check whether permissions on this node can be changed in a way that
1781 * @cumulative_perms and @cumulative_shared_perms are the new cumulative
1782 * permissions of all its parents. This involves checking whether all necessary
1783 * permission changes to child nodes can be performed.
1785 * Will set *tighten_restrictions to true if and only if new permissions have to
1786 * be taken or currently shared permissions are to be unshared. Otherwise,
1787 * errors are not fatal as long as the caller accepts that the restrictions
1788 * remain tighter than they need to be. The caller still has to abort the
1790 * @tighten_restrictions cannot be used together with @q: When reopening, we may
1791 * encounter fatal errors even though no restrictions are to be tightened. For
1792 * example, changing a node from RW to RO will fail if the WRITE permission is
1795 * A call to this function must always be followed by a call to bdrv_set_perm()
1796 * or bdrv_abort_perm_update().
1798 static int bdrv_check_perm(BlockDriverState
*bs
, BlockReopenQueue
*q
,
1799 uint64_t cumulative_perms
,
1800 uint64_t cumulative_shared_perms
,
1801 GSList
*ignore_children
,
1802 bool *tighten_restrictions
, Error
**errp
)
1804 BlockDriver
*drv
= bs
->drv
;
1808 assert(!q
|| !tighten_restrictions
);
1810 if (tighten_restrictions
) {
1811 uint64_t current_perms
, current_shared
;
1812 uint64_t added_perms
, removed_shared_perms
;
1814 bdrv_get_cumulative_perm(bs
, ¤t_perms
, ¤t_shared
);
1816 added_perms
= cumulative_perms
& ~current_perms
;
1817 removed_shared_perms
= current_shared
& ~cumulative_shared_perms
;
1819 *tighten_restrictions
= added_perms
|| removed_shared_perms
;
1822 /* Write permissions never work with read-only images */
1823 if ((cumulative_perms
& (BLK_PERM_WRITE
| BLK_PERM_WRITE_UNCHANGED
)) &&
1824 !bdrv_is_writable_after_reopen(bs
, q
))
1826 if (!bdrv_is_writable_after_reopen(bs
, NULL
)) {
1827 error_setg(errp
, "Block node is read-only");
1829 uint64_t current_perms
, current_shared
;
1830 bdrv_get_cumulative_perm(bs
, ¤t_perms
, ¤t_shared
);
1831 if (current_perms
& (BLK_PERM_WRITE
| BLK_PERM_WRITE_UNCHANGED
)) {
1832 error_setg(errp
, "Cannot make block node read-only, there is "
1835 error_setg(errp
, "Cannot make block node read-only and create "
1843 /* Check this node */
1848 if (drv
->bdrv_check_perm
) {
1849 return drv
->bdrv_check_perm(bs
, cumulative_perms
,
1850 cumulative_shared_perms
, errp
);
1853 /* Drivers that never have children can omit .bdrv_child_perm() */
1854 if (!drv
->bdrv_child_perm
) {
1855 assert(QLIST_EMPTY(&bs
->children
));
1859 /* Check all children */
1860 QLIST_FOREACH(c
, &bs
->children
, next
) {
1861 uint64_t cur_perm
, cur_shared
;
1862 bool child_tighten_restr
;
1864 bdrv_child_perm(bs
, c
->bs
, c
, c
->role
, q
,
1865 cumulative_perms
, cumulative_shared_perms
,
1866 &cur_perm
, &cur_shared
);
1867 ret
= bdrv_child_check_perm(c
, q
, cur_perm
, cur_shared
, ignore_children
,
1868 tighten_restrictions
? &child_tighten_restr
1871 if (tighten_restrictions
) {
1872 *tighten_restrictions
|= child_tighten_restr
;
1883 * Notifies drivers that after a previous bdrv_check_perm() call, the
1884 * permission update is not performed and any preparations made for it (e.g.
1885 * taken file locks) need to be undone.
1887 * This function recursively notifies all child nodes.
1889 static void bdrv_abort_perm_update(BlockDriverState
*bs
)
1891 BlockDriver
*drv
= bs
->drv
;
1898 if (drv
->bdrv_abort_perm_update
) {
1899 drv
->bdrv_abort_perm_update(bs
);
1902 QLIST_FOREACH(c
, &bs
->children
, next
) {
1903 bdrv_child_abort_perm_update(c
);
1907 static void bdrv_set_perm(BlockDriverState
*bs
, uint64_t cumulative_perms
,
1908 uint64_t cumulative_shared_perms
)
1910 BlockDriver
*drv
= bs
->drv
;
1917 /* Update this node */
1918 if (drv
->bdrv_set_perm
) {
1919 drv
->bdrv_set_perm(bs
, cumulative_perms
, cumulative_shared_perms
);
1922 /* Drivers that never have children can omit .bdrv_child_perm() */
1923 if (!drv
->bdrv_child_perm
) {
1924 assert(QLIST_EMPTY(&bs
->children
));
1928 /* Update all children */
1929 QLIST_FOREACH(c
, &bs
->children
, next
) {
1930 uint64_t cur_perm
, cur_shared
;
1931 bdrv_child_perm(bs
, c
->bs
, c
, c
->role
, NULL
,
1932 cumulative_perms
, cumulative_shared_perms
,
1933 &cur_perm
, &cur_shared
);
1934 bdrv_child_set_perm(c
, cur_perm
, cur_shared
);
1938 static void bdrv_get_cumulative_perm(BlockDriverState
*bs
, uint64_t *perm
,
1939 uint64_t *shared_perm
)
1942 uint64_t cumulative_perms
= 0;
1943 uint64_t cumulative_shared_perms
= BLK_PERM_ALL
;
1945 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
1946 cumulative_perms
|= c
->perm
;
1947 cumulative_shared_perms
&= c
->shared_perm
;
1950 *perm
= cumulative_perms
;
1951 *shared_perm
= cumulative_shared_perms
;
1954 static char *bdrv_child_user_desc(BdrvChild
*c
)
1956 if (c
->role
->get_parent_desc
) {
1957 return c
->role
->get_parent_desc(c
);
1960 return g_strdup("another user");
1963 char *bdrv_perm_names(uint64_t perm
)
1969 { BLK_PERM_CONSISTENT_READ
, "consistent read" },
1970 { BLK_PERM_WRITE
, "write" },
1971 { BLK_PERM_WRITE_UNCHANGED
, "write unchanged" },
1972 { BLK_PERM_RESIZE
, "resize" },
1973 { BLK_PERM_GRAPH_MOD
, "change children" },
1977 char *result
= g_strdup("");
1978 struct perm_name
*p
;
1980 for (p
= permissions
; p
->name
; p
++) {
1981 if (perm
& p
->perm
) {
1983 result
= g_strdup_printf("%s%s%s", old
, *old
? ", " : "", p
->name
);
1992 * Checks whether a new reference to @bs can be added if the new user requires
1993 * @new_used_perm/@new_shared_perm as its permissions. If @ignore_children is
1994 * set, the BdrvChild objects in this list are ignored in the calculations;
1995 * this allows checking permission updates for an existing reference.
1997 * See bdrv_check_perm() for the semantics of @tighten_restrictions.
1999 * Needs to be followed by a call to either bdrv_set_perm() or
2000 * bdrv_abort_perm_update(). */
2001 static int bdrv_check_update_perm(BlockDriverState
*bs
, BlockReopenQueue
*q
,
2002 uint64_t new_used_perm
,
2003 uint64_t new_shared_perm
,
2004 GSList
*ignore_children
,
2005 bool *tighten_restrictions
,
2009 uint64_t cumulative_perms
= new_used_perm
;
2010 uint64_t cumulative_shared_perms
= new_shared_perm
;
2012 assert(!q
|| !tighten_restrictions
);
2014 /* There is no reason why anyone couldn't tolerate write_unchanged */
2015 assert(new_shared_perm
& BLK_PERM_WRITE_UNCHANGED
);
2017 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
2018 if (g_slist_find(ignore_children
, c
)) {
2022 if ((new_used_perm
& c
->shared_perm
) != new_used_perm
) {
2023 char *user
= bdrv_child_user_desc(c
);
2024 char *perm_names
= bdrv_perm_names(new_used_perm
& ~c
->shared_perm
);
2026 if (tighten_restrictions
) {
2027 *tighten_restrictions
= true;
2030 error_setg(errp
, "Conflicts with use by %s as '%s', which does not "
2032 user
, c
->name
, perm_names
, bdrv_get_node_name(c
->bs
));
2038 if ((c
->perm
& new_shared_perm
) != c
->perm
) {
2039 char *user
= bdrv_child_user_desc(c
);
2040 char *perm_names
= bdrv_perm_names(c
->perm
& ~new_shared_perm
);
2042 if (tighten_restrictions
) {
2043 *tighten_restrictions
= true;
2046 error_setg(errp
, "Conflicts with use by %s as '%s', which uses "
2048 user
, c
->name
, perm_names
, bdrv_get_node_name(c
->bs
));
2054 cumulative_perms
|= c
->perm
;
2055 cumulative_shared_perms
&= c
->shared_perm
;
2058 return bdrv_check_perm(bs
, q
, cumulative_perms
, cumulative_shared_perms
,
2059 ignore_children
, tighten_restrictions
, errp
);
2062 /* Needs to be followed by a call to either bdrv_child_set_perm() or
2063 * bdrv_child_abort_perm_update(). */
2064 static int bdrv_child_check_perm(BdrvChild
*c
, BlockReopenQueue
*q
,
2065 uint64_t perm
, uint64_t shared
,
2066 GSList
*ignore_children
,
2067 bool *tighten_restrictions
, Error
**errp
)
2071 ignore_children
= g_slist_prepend(g_slist_copy(ignore_children
), c
);
2072 ret
= bdrv_check_update_perm(c
->bs
, q
, perm
, shared
, ignore_children
,
2073 tighten_restrictions
, errp
);
2074 g_slist_free(ignore_children
);
2080 if (!c
->has_backup_perm
) {
2081 c
->has_backup_perm
= true;
2082 c
->backup_perm
= c
->perm
;
2083 c
->backup_shared_perm
= c
->shared_perm
;
2086 * Note: it's OK if c->has_backup_perm was already set, as we can find the
2087 * same child twice during check_perm procedure
2091 c
->shared_perm
= shared
;
2096 static void bdrv_child_set_perm(BdrvChild
*c
, uint64_t perm
, uint64_t shared
)
2098 uint64_t cumulative_perms
, cumulative_shared_perms
;
2100 c
->has_backup_perm
= false;
2103 c
->shared_perm
= shared
;
2105 bdrv_get_cumulative_perm(c
->bs
, &cumulative_perms
,
2106 &cumulative_shared_perms
);
2107 bdrv_set_perm(c
->bs
, cumulative_perms
, cumulative_shared_perms
);
2110 static void bdrv_child_abort_perm_update(BdrvChild
*c
)
2112 if (c
->has_backup_perm
) {
2113 c
->perm
= c
->backup_perm
;
2114 c
->shared_perm
= c
->backup_shared_perm
;
2115 c
->has_backup_perm
= false;
2118 bdrv_abort_perm_update(c
->bs
);
2121 int bdrv_child_try_set_perm(BdrvChild
*c
, uint64_t perm
, uint64_t shared
,
2124 Error
*local_err
= NULL
;
2126 bool tighten_restrictions
;
2128 ret
= bdrv_child_check_perm(c
, NULL
, perm
, shared
, NULL
,
2129 &tighten_restrictions
, &local_err
);
2131 bdrv_child_abort_perm_update(c
);
2132 if (tighten_restrictions
) {
2133 error_propagate(errp
, local_err
);
2136 * Our caller may intend to only loosen restrictions and
2137 * does not expect this function to fail. Errors are not
2138 * fatal in such a case, so we can just hide them from our
2141 error_free(local_err
);
2147 bdrv_child_set_perm(c
, perm
, shared
);
2152 int bdrv_child_refresh_perms(BlockDriverState
*bs
, BdrvChild
*c
, Error
**errp
)
2154 uint64_t parent_perms
, parent_shared
;
2155 uint64_t perms
, shared
;
2157 bdrv_get_cumulative_perm(bs
, &parent_perms
, &parent_shared
);
2158 bdrv_child_perm(bs
, c
->bs
, c
, c
->role
, NULL
, parent_perms
, parent_shared
,
2161 return bdrv_child_try_set_perm(c
, perms
, shared
, errp
);
2164 void bdrv_filter_default_perms(BlockDriverState
*bs
, BdrvChild
*c
,
2165 const BdrvChildRole
*role
,
2166 BlockReopenQueue
*reopen_queue
,
2167 uint64_t perm
, uint64_t shared
,
2168 uint64_t *nperm
, uint64_t *nshared
)
2171 *nperm
= perm
& DEFAULT_PERM_PASSTHROUGH
;
2172 *nshared
= (shared
& DEFAULT_PERM_PASSTHROUGH
) | DEFAULT_PERM_UNCHANGED
;
2176 *nperm
= (perm
& DEFAULT_PERM_PASSTHROUGH
) |
2177 (c
->perm
& DEFAULT_PERM_UNCHANGED
);
2178 *nshared
= (shared
& DEFAULT_PERM_PASSTHROUGH
) |
2179 (c
->shared_perm
& DEFAULT_PERM_UNCHANGED
);
2182 void bdrv_format_default_perms(BlockDriverState
*bs
, BdrvChild
*c
,
2183 const BdrvChildRole
*role
,
2184 BlockReopenQueue
*reopen_queue
,
2185 uint64_t perm
, uint64_t shared
,
2186 uint64_t *nperm
, uint64_t *nshared
)
2188 bool backing
= (role
== &child_backing
);
2189 assert(role
== &child_backing
|| role
== &child_file
);
2192 int flags
= bdrv_reopen_get_flags(reopen_queue
, bs
);
2194 /* Apart from the modifications below, the same permissions are
2195 * forwarded and left alone as for filters */
2196 bdrv_filter_default_perms(bs
, c
, role
, reopen_queue
, perm
, shared
,
2199 /* Format drivers may touch metadata even if the guest doesn't write */
2200 if (bdrv_is_writable_after_reopen(bs
, reopen_queue
)) {
2201 perm
|= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2204 /* bs->file always needs to be consistent because of the metadata. We
2205 * can never allow other users to resize or write to it. */
2206 if (!(flags
& BDRV_O_NO_IO
)) {
2207 perm
|= BLK_PERM_CONSISTENT_READ
;
2209 shared
&= ~(BLK_PERM_WRITE
| BLK_PERM_RESIZE
);
2211 /* We want consistent read from backing files if the parent needs it.
2212 * No other operations are performed on backing files. */
2213 perm
&= BLK_PERM_CONSISTENT_READ
;
2215 /* If the parent can deal with changing data, we're okay with a
2216 * writable and resizable backing file. */
2217 /* TODO Require !(perm & BLK_PERM_CONSISTENT_READ), too? */
2218 if (shared
& BLK_PERM_WRITE
) {
2219 shared
= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2224 shared
|= BLK_PERM_CONSISTENT_READ
| BLK_PERM_GRAPH_MOD
|
2225 BLK_PERM_WRITE_UNCHANGED
;
2228 if (bs
->open_flags
& BDRV_O_INACTIVE
) {
2229 shared
|= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2236 static void bdrv_replace_child_noperm(BdrvChild
*child
,
2237 BlockDriverState
*new_bs
)
2239 BlockDriverState
*old_bs
= child
->bs
;
2242 assert(!child
->frozen
);
2244 if (old_bs
&& new_bs
) {
2245 assert(bdrv_get_aio_context(old_bs
) == bdrv_get_aio_context(new_bs
));
2248 /* Detach first so that the recursive drain sections coming from @child
2249 * are already gone and we only end the drain sections that came from
2251 if (child
->role
->detach
) {
2252 child
->role
->detach(child
);
2254 if (old_bs
->quiesce_counter
&& child
->role
->drained_end
) {
2255 int num
= old_bs
->quiesce_counter
;
2256 if (child
->role
->parent_is_bds
) {
2257 num
-= bdrv_drain_all_count
;
2260 for (i
= 0; i
< num
; i
++) {
2261 child
->role
->drained_end(child
);
2264 QLIST_REMOVE(child
, next_parent
);
2270 QLIST_INSERT_HEAD(&new_bs
->parents
, child
, next_parent
);
2271 if (new_bs
->quiesce_counter
&& child
->role
->drained_begin
) {
2272 int num
= new_bs
->quiesce_counter
;
2273 if (child
->role
->parent_is_bds
) {
2274 num
-= bdrv_drain_all_count
;
2277 for (i
= 0; i
< num
; i
++) {
2278 bdrv_parent_drained_begin_single(child
, true);
2282 /* Attach only after starting new drained sections, so that recursive
2283 * drain sections coming from @child don't get an extra .drained_begin
2285 if (child
->role
->attach
) {
2286 child
->role
->attach(child
);
2292 * Updates @child to change its reference to point to @new_bs, including
2293 * checking and applying the necessary permisson updates both to the old node
2296 * NULL is passed as @new_bs for removing the reference before freeing @child.
2298 * If @new_bs is not NULL, bdrv_check_perm() must be called beforehand, as this
2299 * function uses bdrv_set_perm() to update the permissions according to the new
2300 * reference that @new_bs gets.
2302 static void bdrv_replace_child(BdrvChild
*child
, BlockDriverState
*new_bs
)
2304 BlockDriverState
*old_bs
= child
->bs
;
2305 uint64_t perm
, shared_perm
;
2307 bdrv_replace_child_noperm(child
, new_bs
);
2310 * Start with the new node's permissions. If @new_bs is a (direct
2311 * or indirect) child of @old_bs, we must complete the permission
2312 * update on @new_bs before we loosen the restrictions on @old_bs.
2313 * Otherwise, bdrv_check_perm() on @old_bs would re-initiate
2314 * updating the permissions of @new_bs, and thus not purely loosen
2318 bdrv_get_cumulative_perm(new_bs
, &perm
, &shared_perm
);
2319 bdrv_set_perm(new_bs
, perm
, shared_perm
);
2323 /* Update permissions for old node. This is guaranteed to succeed
2324 * because we're just taking a parent away, so we're loosening
2326 bool tighten_restrictions
;
2329 bdrv_get_cumulative_perm(old_bs
, &perm
, &shared_perm
);
2330 ret
= bdrv_check_perm(old_bs
, NULL
, perm
, shared_perm
, NULL
,
2331 &tighten_restrictions
, NULL
);
2332 assert(tighten_restrictions
== false);
2334 /* We only tried to loosen restrictions, so errors are not fatal */
2335 bdrv_abort_perm_update(old_bs
);
2337 bdrv_set_perm(old_bs
, perm
, shared_perm
);
2340 /* When the parent requiring a non-default AioContext is removed, the
2341 * node moves back to the main AioContext */
2342 bdrv_try_set_aio_context(old_bs
, qemu_get_aio_context(), NULL
);
2347 * This function steals the reference to child_bs from the caller.
2348 * That reference is later dropped by bdrv_root_unref_child().
2350 * On failure NULL is returned, errp is set and the reference to
2351 * child_bs is also dropped.
2353 * The caller must hold the AioContext lock @child_bs, but not that of @ctx
2354 * (unless @child_bs is already in @ctx).
2356 BdrvChild
*bdrv_root_attach_child(BlockDriverState
*child_bs
,
2357 const char *child_name
,
2358 const BdrvChildRole
*child_role
,
2360 uint64_t perm
, uint64_t shared_perm
,
2361 void *opaque
, Error
**errp
)
2364 Error
*local_err
= NULL
;
2367 ret
= bdrv_check_update_perm(child_bs
, NULL
, perm
, shared_perm
, NULL
, NULL
,
2370 bdrv_abort_perm_update(child_bs
);
2371 bdrv_unref(child_bs
);
2375 child
= g_new(BdrvChild
, 1);
2376 *child
= (BdrvChild
) {
2378 .name
= g_strdup(child_name
),
2381 .shared_perm
= shared_perm
,
2385 /* If the AioContexts don't match, first try to move the subtree of
2386 * child_bs into the AioContext of the new parent. If this doesn't work,
2387 * try moving the parent into the AioContext of child_bs instead. */
2388 if (bdrv_get_aio_context(child_bs
) != ctx
) {
2389 ret
= bdrv_try_set_aio_context(child_bs
, ctx
, &local_err
);
2390 if (ret
< 0 && child_role
->can_set_aio_ctx
) {
2391 GSList
*ignore
= g_slist_prepend(NULL
, child
);;
2392 ctx
= bdrv_get_aio_context(child_bs
);
2393 if (child_role
->can_set_aio_ctx(child
, ctx
, &ignore
, NULL
)) {
2394 error_free(local_err
);
2396 g_slist_free(ignore
);
2397 ignore
= g_slist_prepend(NULL
, child
);;
2398 child_role
->set_aio_ctx(child
, ctx
, &ignore
);
2400 g_slist_free(ignore
);
2403 error_propagate(errp
, local_err
);
2405 bdrv_abort_perm_update(child_bs
);
2410 /* This performs the matching bdrv_set_perm() for the above check. */
2411 bdrv_replace_child(child
, child_bs
);
2417 * This function transfers the reference to child_bs from the caller
2418 * to parent_bs. That reference is later dropped by parent_bs on
2419 * bdrv_close() or if someone calls bdrv_unref_child().
2421 * On failure NULL is returned, errp is set and the reference to
2422 * child_bs is also dropped.
2424 * If @parent_bs and @child_bs are in different AioContexts, the caller must
2425 * hold the AioContext lock for @child_bs, but not for @parent_bs.
2427 BdrvChild
*bdrv_attach_child(BlockDriverState
*parent_bs
,
2428 BlockDriverState
*child_bs
,
2429 const char *child_name
,
2430 const BdrvChildRole
*child_role
,
2434 uint64_t perm
, shared_perm
;
2436 bdrv_get_cumulative_perm(parent_bs
, &perm
, &shared_perm
);
2438 assert(parent_bs
->drv
);
2439 bdrv_child_perm(parent_bs
, child_bs
, NULL
, child_role
, NULL
,
2440 perm
, shared_perm
, &perm
, &shared_perm
);
2442 child
= bdrv_root_attach_child(child_bs
, child_name
, child_role
,
2443 bdrv_get_aio_context(parent_bs
),
2444 perm
, shared_perm
, parent_bs
, errp
);
2445 if (child
== NULL
) {
2449 QLIST_INSERT_HEAD(&parent_bs
->children
, child
, next
);
2453 static void bdrv_detach_child(BdrvChild
*child
)
2455 if (child
->next
.le_prev
) {
2456 QLIST_REMOVE(child
, next
);
2457 child
->next
.le_prev
= NULL
;
2460 bdrv_replace_child(child
, NULL
);
2462 g_free(child
->name
);
2466 void bdrv_root_unref_child(BdrvChild
*child
)
2468 BlockDriverState
*child_bs
;
2470 child_bs
= child
->bs
;
2471 bdrv_detach_child(child
);
2472 bdrv_unref(child_bs
);
2476 * Clear all inherits_from pointers from children and grandchildren of
2477 * @root that point to @root, where necessary.
2479 static void bdrv_unset_inherits_from(BlockDriverState
*root
, BdrvChild
*child
)
2483 if (child
->bs
->inherits_from
== root
) {
2485 * Remove inherits_from only when the last reference between root and
2486 * child->bs goes away.
2488 QLIST_FOREACH(c
, &root
->children
, next
) {
2489 if (c
!= child
&& c
->bs
== child
->bs
) {
2494 child
->bs
->inherits_from
= NULL
;
2498 QLIST_FOREACH(c
, &child
->bs
->children
, next
) {
2499 bdrv_unset_inherits_from(root
, c
);
2503 void bdrv_unref_child(BlockDriverState
*parent
, BdrvChild
*child
)
2505 if (child
== NULL
) {
2509 bdrv_unset_inherits_from(parent
, child
);
2510 bdrv_root_unref_child(child
);
2514 static void bdrv_parent_cb_change_media(BlockDriverState
*bs
, bool load
)
2517 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
2518 if (c
->role
->change_media
) {
2519 c
->role
->change_media(c
, load
);
2524 /* Return true if you can reach parent going through child->inherits_from
2525 * recursively. If parent or child are NULL, return false */
2526 static bool bdrv_inherits_from_recursive(BlockDriverState
*child
,
2527 BlockDriverState
*parent
)
2529 while (child
&& child
!= parent
) {
2530 child
= child
->inherits_from
;
2533 return child
!= NULL
;
2537 * Sets the backing file link of a BDS. A new reference is created; callers
2538 * which don't need their own reference any more must call bdrv_unref().
2540 void bdrv_set_backing_hd(BlockDriverState
*bs
, BlockDriverState
*backing_hd
,
2543 bool update_inherits_from
= bdrv_chain_contains(bs
, backing_hd
) &&
2544 bdrv_inherits_from_recursive(backing_hd
, bs
);
2546 if (bdrv_is_backing_chain_frozen(bs
, backing_bs(bs
), errp
)) {
2551 bdrv_ref(backing_hd
);
2555 bdrv_unref_child(bs
, bs
->backing
);
2563 bs
->backing
= bdrv_attach_child(bs
, backing_hd
, "backing", &child_backing
,
2565 /* If backing_hd was already part of bs's backing chain, and
2566 * inherits_from pointed recursively to bs then let's update it to
2567 * point directly to bs (else it will become NULL). */
2568 if (bs
->backing
&& update_inherits_from
) {
2569 backing_hd
->inherits_from
= bs
;
2573 bdrv_refresh_limits(bs
, NULL
);
2577 * Opens the backing file for a BlockDriverState if not yet open
2579 * bdref_key specifies the key for the image's BlockdevRef in the options QDict.
2580 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
2581 * itself, all options starting with "${bdref_key}." are considered part of the
2584 * TODO Can this be unified with bdrv_open_image()?
2586 int bdrv_open_backing_file(BlockDriverState
*bs
, QDict
*parent_options
,
2587 const char *bdref_key
, Error
**errp
)
2589 char *backing_filename
= NULL
;
2590 char *bdref_key_dot
;
2591 const char *reference
= NULL
;
2593 bool implicit_backing
= false;
2594 BlockDriverState
*backing_hd
;
2596 QDict
*tmp_parent_options
= NULL
;
2597 Error
*local_err
= NULL
;
2599 if (bs
->backing
!= NULL
) {
2603 /* NULL means an empty set of options */
2604 if (parent_options
== NULL
) {
2605 tmp_parent_options
= qdict_new();
2606 parent_options
= tmp_parent_options
;
2609 bs
->open_flags
&= ~BDRV_O_NO_BACKING
;
2611 bdref_key_dot
= g_strdup_printf("%s.", bdref_key
);
2612 qdict_extract_subqdict(parent_options
, &options
, bdref_key_dot
);
2613 g_free(bdref_key_dot
);
2616 * Caution: while qdict_get_try_str() is fine, getting non-string
2617 * types would require more care. When @parent_options come from
2618 * -blockdev or blockdev_add, its members are typed according to
2619 * the QAPI schema, but when they come from -drive, they're all
2622 reference
= qdict_get_try_str(parent_options
, bdref_key
);
2623 if (reference
|| qdict_haskey(options
, "file.filename")) {
2624 /* keep backing_filename NULL */
2625 } else if (bs
->backing_file
[0] == '\0' && qdict_size(options
) == 0) {
2626 qobject_unref(options
);
2629 if (qdict_size(options
) == 0) {
2630 /* If the user specifies options that do not modify the
2631 * backing file's behavior, we might still consider it the
2632 * implicit backing file. But it's easier this way, and
2633 * just specifying some of the backing BDS's options is
2634 * only possible with -drive anyway (otherwise the QAPI
2635 * schema forces the user to specify everything). */
2636 implicit_backing
= !strcmp(bs
->auto_backing_file
, bs
->backing_file
);
2639 backing_filename
= bdrv_get_full_backing_filename(bs
, &local_err
);
2642 error_propagate(errp
, local_err
);
2643 qobject_unref(options
);
2648 if (!bs
->drv
|| !bs
->drv
->supports_backing
) {
2650 error_setg(errp
, "Driver doesn't support backing files");
2651 qobject_unref(options
);
2656 bs
->backing_format
[0] != '\0' && !qdict_haskey(options
, "driver")) {
2657 qdict_put_str(options
, "driver", bs
->backing_format
);
2660 backing_hd
= bdrv_open_inherit(backing_filename
, reference
, options
, 0, bs
,
2661 &child_backing
, errp
);
2663 bs
->open_flags
|= BDRV_O_NO_BACKING
;
2664 error_prepend(errp
, "Could not open backing file: ");
2669 if (implicit_backing
) {
2670 bdrv_refresh_filename(backing_hd
);
2671 pstrcpy(bs
->auto_backing_file
, sizeof(bs
->auto_backing_file
),
2672 backing_hd
->filename
);
2675 /* Hook up the backing file link; drop our reference, bs owns the
2676 * backing_hd reference now */
2677 bdrv_set_backing_hd(bs
, backing_hd
, &local_err
);
2678 bdrv_unref(backing_hd
);
2680 error_propagate(errp
, local_err
);
2685 qdict_del(parent_options
, bdref_key
);
2688 g_free(backing_filename
);
2689 qobject_unref(tmp_parent_options
);
2693 static BlockDriverState
*
2694 bdrv_open_child_bs(const char *filename
, QDict
*options
, const char *bdref_key
,
2695 BlockDriverState
*parent
, const BdrvChildRole
*child_role
,
2696 bool allow_none
, Error
**errp
)
2698 BlockDriverState
*bs
= NULL
;
2699 QDict
*image_options
;
2700 char *bdref_key_dot
;
2701 const char *reference
;
2703 assert(child_role
!= NULL
);
2705 bdref_key_dot
= g_strdup_printf("%s.", bdref_key
);
2706 qdict_extract_subqdict(options
, &image_options
, bdref_key_dot
);
2707 g_free(bdref_key_dot
);
2710 * Caution: while qdict_get_try_str() is fine, getting non-string
2711 * types would require more care. When @options come from
2712 * -blockdev or blockdev_add, its members are typed according to
2713 * the QAPI schema, but when they come from -drive, they're all
2716 reference
= qdict_get_try_str(options
, bdref_key
);
2717 if (!filename
&& !reference
&& !qdict_size(image_options
)) {
2719 error_setg(errp
, "A block device must be specified for \"%s\"",
2722 qobject_unref(image_options
);
2726 bs
= bdrv_open_inherit(filename
, reference
, image_options
, 0,
2727 parent
, child_role
, errp
);
2733 qdict_del(options
, bdref_key
);
2738 * Opens a disk image whose options are given as BlockdevRef in another block
2741 * If allow_none is true, no image will be opened if filename is false and no
2742 * BlockdevRef is given. NULL will be returned, but errp remains unset.
2744 * bdrev_key specifies the key for the image's BlockdevRef in the options QDict.
2745 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
2746 * itself, all options starting with "${bdref_key}." are considered part of the
2749 * The BlockdevRef will be removed from the options QDict.
2751 BdrvChild
*bdrv_open_child(const char *filename
,
2752 QDict
*options
, const char *bdref_key
,
2753 BlockDriverState
*parent
,
2754 const BdrvChildRole
*child_role
,
2755 bool allow_none
, Error
**errp
)
2757 BlockDriverState
*bs
;
2759 bs
= bdrv_open_child_bs(filename
, options
, bdref_key
, parent
, child_role
,
2765 return bdrv_attach_child(parent
, bs
, bdref_key
, child_role
, errp
);
2768 /* TODO Future callers may need to specify parent/child_role in order for
2769 * option inheritance to work. Existing callers use it for the root node. */
2770 BlockDriverState
*bdrv_open_blockdev_ref(BlockdevRef
*ref
, Error
**errp
)
2772 BlockDriverState
*bs
= NULL
;
2773 Error
*local_err
= NULL
;
2774 QObject
*obj
= NULL
;
2775 QDict
*qdict
= NULL
;
2776 const char *reference
= NULL
;
2779 if (ref
->type
== QTYPE_QSTRING
) {
2780 reference
= ref
->u
.reference
;
2782 BlockdevOptions
*options
= &ref
->u
.definition
;
2783 assert(ref
->type
== QTYPE_QDICT
);
2785 v
= qobject_output_visitor_new(&obj
);
2786 visit_type_BlockdevOptions(v
, NULL
, &options
, &local_err
);
2788 error_propagate(errp
, local_err
);
2791 visit_complete(v
, &obj
);
2793 qdict
= qobject_to(QDict
, obj
);
2794 qdict_flatten(qdict
);
2796 /* bdrv_open_inherit() defaults to the values in bdrv_flags (for
2797 * compatibility with other callers) rather than what we want as the
2798 * real defaults. Apply the defaults here instead. */
2799 qdict_set_default_str(qdict
, BDRV_OPT_CACHE_DIRECT
, "off");
2800 qdict_set_default_str(qdict
, BDRV_OPT_CACHE_NO_FLUSH
, "off");
2801 qdict_set_default_str(qdict
, BDRV_OPT_READ_ONLY
, "off");
2802 qdict_set_default_str(qdict
, BDRV_OPT_AUTO_READ_ONLY
, "off");
2806 bs
= bdrv_open_inherit(NULL
, reference
, qdict
, 0, NULL
, NULL
, errp
);
2815 static BlockDriverState
*bdrv_append_temp_snapshot(BlockDriverState
*bs
,
2817 QDict
*snapshot_options
,
2820 /* TODO: extra byte is a hack to ensure MAX_PATH space on Windows. */
2821 char *tmp_filename
= g_malloc0(PATH_MAX
+ 1);
2823 QemuOpts
*opts
= NULL
;
2824 BlockDriverState
*bs_snapshot
= NULL
;
2825 Error
*local_err
= NULL
;
2828 /* if snapshot, we create a temporary backing file and open it
2829 instead of opening 'filename' directly */
2831 /* Get the required size from the image */
2832 total_size
= bdrv_getlength(bs
);
2833 if (total_size
< 0) {
2834 error_setg_errno(errp
, -total_size
, "Could not get image size");
2838 /* Create the temporary image */
2839 ret
= get_tmp_filename(tmp_filename
, PATH_MAX
+ 1);
2841 error_setg_errno(errp
, -ret
, "Could not get temporary filename");
2845 opts
= qemu_opts_create(bdrv_qcow2
.create_opts
, NULL
, 0,
2847 qemu_opt_set_number(opts
, BLOCK_OPT_SIZE
, total_size
, &error_abort
);
2848 ret
= bdrv_create(&bdrv_qcow2
, tmp_filename
, opts
, errp
);
2849 qemu_opts_del(opts
);
2851 error_prepend(errp
, "Could not create temporary overlay '%s': ",
2856 /* Prepare options QDict for the temporary file */
2857 qdict_put_str(snapshot_options
, "file.driver", "file");
2858 qdict_put_str(snapshot_options
, "file.filename", tmp_filename
);
2859 qdict_put_str(snapshot_options
, "driver", "qcow2");
2861 bs_snapshot
= bdrv_open(NULL
, NULL
, snapshot_options
, flags
, errp
);
2862 snapshot_options
= NULL
;
2867 /* bdrv_append() consumes a strong reference to bs_snapshot
2868 * (i.e. it will call bdrv_unref() on it) even on error, so in
2869 * order to be able to return one, we have to increase
2870 * bs_snapshot's refcount here */
2871 bdrv_ref(bs_snapshot
);
2872 bdrv_append(bs_snapshot
, bs
, &local_err
);
2874 error_propagate(errp
, local_err
);
2880 qobject_unref(snapshot_options
);
2881 g_free(tmp_filename
);
2886 * Opens a disk image (raw, qcow2, vmdk, ...)
2888 * options is a QDict of options to pass to the block drivers, or NULL for an
2889 * empty set of options. The reference to the QDict belongs to the block layer
2890 * after the call (even on failure), so if the caller intends to reuse the
2891 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
2893 * If *pbs is NULL, a new BDS will be created with a pointer to it stored there.
2894 * If it is not NULL, the referenced BDS will be reused.
2896 * The reference parameter may be used to specify an existing block device which
2897 * should be opened. If specified, neither options nor a filename may be given,
2898 * nor can an existing BDS be reused (that is, *pbs has to be NULL).
2900 static BlockDriverState
*bdrv_open_inherit(const char *filename
,
2901 const char *reference
,
2902 QDict
*options
, int flags
,
2903 BlockDriverState
*parent
,
2904 const BdrvChildRole
*child_role
,
2908 BlockBackend
*file
= NULL
;
2909 BlockDriverState
*bs
;
2910 BlockDriver
*drv
= NULL
;
2912 const char *drvname
;
2913 const char *backing
;
2914 Error
*local_err
= NULL
;
2915 QDict
*snapshot_options
= NULL
;
2916 int snapshot_flags
= 0;
2918 assert(!child_role
|| !flags
);
2919 assert(!child_role
== !parent
);
2922 bool options_non_empty
= options
? qdict_size(options
) : false;
2923 qobject_unref(options
);
2925 if (filename
|| options_non_empty
) {
2926 error_setg(errp
, "Cannot reference an existing block device with "
2927 "additional options or a new filename");
2931 bs
= bdrv_lookup_bs(reference
, reference
, errp
);
2942 /* NULL means an empty set of options */
2943 if (options
== NULL
) {
2944 options
= qdict_new();
2947 /* json: syntax counts as explicit options, as if in the QDict */
2948 parse_json_protocol(options
, &filename
, &local_err
);
2953 bs
->explicit_options
= qdict_clone_shallow(options
);
2956 bs
->inherits_from
= parent
;
2957 child_role
->inherit_options(&flags
, options
,
2958 parent
->open_flags
, parent
->options
);
2961 ret
= bdrv_fill_options(&options
, filename
, &flags
, &local_err
);
2967 * Set the BDRV_O_RDWR and BDRV_O_ALLOW_RDWR flags.
2968 * Caution: getting a boolean member of @options requires care.
2969 * When @options come from -blockdev or blockdev_add, members are
2970 * typed according to the QAPI schema, but when they come from
2971 * -drive, they're all QString.
2973 if (g_strcmp0(qdict_get_try_str(options
, BDRV_OPT_READ_ONLY
), "on") &&
2974 !qdict_get_try_bool(options
, BDRV_OPT_READ_ONLY
, false)) {
2975 flags
|= (BDRV_O_RDWR
| BDRV_O_ALLOW_RDWR
);
2977 flags
&= ~BDRV_O_RDWR
;
2980 if (flags
& BDRV_O_SNAPSHOT
) {
2981 snapshot_options
= qdict_new();
2982 bdrv_temp_snapshot_options(&snapshot_flags
, snapshot_options
,
2984 /* Let bdrv_backing_options() override "read-only" */
2985 qdict_del(options
, BDRV_OPT_READ_ONLY
);
2986 bdrv_backing_options(&flags
, options
, flags
, options
);
2989 bs
->open_flags
= flags
;
2990 bs
->options
= options
;
2991 options
= qdict_clone_shallow(options
);
2993 /* Find the right image format driver */
2994 /* See cautionary note on accessing @options above */
2995 drvname
= qdict_get_try_str(options
, "driver");
2997 drv
= bdrv_find_format(drvname
);
2999 error_setg(errp
, "Unknown driver: '%s'", drvname
);
3004 assert(drvname
|| !(flags
& BDRV_O_PROTOCOL
));
3006 /* See cautionary note on accessing @options above */
3007 backing
= qdict_get_try_str(options
, "backing");
3008 if (qobject_to(QNull
, qdict_get(options
, "backing")) != NULL
||
3009 (backing
&& *backing
== '\0'))
3012 warn_report("Use of \"backing\": \"\" is deprecated; "
3013 "use \"backing\": null instead");
3015 flags
|= BDRV_O_NO_BACKING
;
3016 qdict_del(options
, "backing");
3019 /* Open image file without format layer. This BlockBackend is only used for
3020 * probing, the block drivers will do their own bdrv_open_child() for the
3021 * same BDS, which is why we put the node name back into options. */
3022 if ((flags
& BDRV_O_PROTOCOL
) == 0) {
3023 BlockDriverState
*file_bs
;
3025 file_bs
= bdrv_open_child_bs(filename
, options
, "file", bs
,
3026 &child_file
, true, &local_err
);
3030 if (file_bs
!= NULL
) {
3031 /* Not requesting BLK_PERM_CONSISTENT_READ because we're only
3032 * looking at the header to guess the image format. This works even
3033 * in cases where a guest would not see a consistent state. */
3034 file
= blk_new(bdrv_get_aio_context(file_bs
), 0, BLK_PERM_ALL
);
3035 blk_insert_bs(file
, file_bs
, &local_err
);
3036 bdrv_unref(file_bs
);
3041 qdict_put_str(options
, "file", bdrv_get_node_name(file_bs
));
3045 /* Image format probing */
3048 ret
= find_image_format(file
, filename
, &drv
, &local_err
);
3053 * This option update would logically belong in bdrv_fill_options(),
3054 * but we first need to open bs->file for the probing to work, while
3055 * opening bs->file already requires the (mostly) final set of options
3056 * so that cache mode etc. can be inherited.
3058 * Adding the driver later is somewhat ugly, but it's not an option
3059 * that would ever be inherited, so it's correct. We just need to make
3060 * sure to update both bs->options (which has the full effective
3061 * options for bs) and options (which has file.* already removed).
3063 qdict_put_str(bs
->options
, "driver", drv
->format_name
);
3064 qdict_put_str(options
, "driver", drv
->format_name
);
3066 error_setg(errp
, "Must specify either driver or file");
3070 /* BDRV_O_PROTOCOL must be set iff a protocol BDS is about to be created */
3071 assert(!!(flags
& BDRV_O_PROTOCOL
) == !!drv
->bdrv_file_open
);
3072 /* file must be NULL if a protocol BDS is about to be created
3073 * (the inverse results in an error message from bdrv_open_common()) */
3074 assert(!(flags
& BDRV_O_PROTOCOL
) || !file
);
3076 /* Open the image */
3077 ret
= bdrv_open_common(bs
, file
, options
, &local_err
);
3087 /* If there is a backing file, use it */
3088 if ((flags
& BDRV_O_NO_BACKING
) == 0) {
3089 ret
= bdrv_open_backing_file(bs
, options
, "backing", &local_err
);
3091 goto close_and_fail
;
3095 /* Remove all children options and references
3096 * from bs->options and bs->explicit_options */
3097 QLIST_FOREACH(child
, &bs
->children
, next
) {
3098 char *child_key_dot
;
3099 child_key_dot
= g_strdup_printf("%s.", child
->name
);
3100 qdict_extract_subqdict(bs
->explicit_options
, NULL
, child_key_dot
);
3101 qdict_extract_subqdict(bs
->options
, NULL
, child_key_dot
);
3102 qdict_del(bs
->explicit_options
, child
->name
);
3103 qdict_del(bs
->options
, child
->name
);
3104 g_free(child_key_dot
);
3107 /* Check if any unknown options were used */
3108 if (qdict_size(options
) != 0) {
3109 const QDictEntry
*entry
= qdict_first(options
);
3110 if (flags
& BDRV_O_PROTOCOL
) {
3111 error_setg(errp
, "Block protocol '%s' doesn't support the option "
3112 "'%s'", drv
->format_name
, entry
->key
);
3115 "Block format '%s' does not support the option '%s'",
3116 drv
->format_name
, entry
->key
);
3119 goto close_and_fail
;
3122 bdrv_parent_cb_change_media(bs
, true);
3124 qobject_unref(options
);
3127 /* For snapshot=on, create a temporary qcow2 overlay. bs points to the
3128 * temporary snapshot afterwards. */
3129 if (snapshot_flags
) {
3130 BlockDriverState
*snapshot_bs
;
3131 snapshot_bs
= bdrv_append_temp_snapshot(bs
, snapshot_flags
,
3132 snapshot_options
, &local_err
);
3133 snapshot_options
= NULL
;
3135 goto close_and_fail
;
3137 /* We are not going to return bs but the overlay on top of it
3138 * (snapshot_bs); thus, we have to drop the strong reference to bs
3139 * (which we obtained by calling bdrv_new()). bs will not be deleted,
3140 * though, because the overlay still has a reference to it. */
3149 qobject_unref(snapshot_options
);
3150 qobject_unref(bs
->explicit_options
);
3151 qobject_unref(bs
->options
);
3152 qobject_unref(options
);
3154 bs
->explicit_options
= NULL
;
3156 error_propagate(errp
, local_err
);
3161 qobject_unref(snapshot_options
);
3162 qobject_unref(options
);
3163 error_propagate(errp
, local_err
);
3167 BlockDriverState
*bdrv_open(const char *filename
, const char *reference
,
3168 QDict
*options
, int flags
, Error
**errp
)
3170 return bdrv_open_inherit(filename
, reference
, options
, flags
, NULL
,
3174 /* Return true if the NULL-terminated @list contains @str */
3175 static bool is_str_in_list(const char *str
, const char *const *list
)
3179 for (i
= 0; list
[i
] != NULL
; i
++) {
3180 if (!strcmp(str
, list
[i
])) {
3189 * Check that every option set in @bs->options is also set in
3192 * Options listed in the common_options list and in
3193 * @bs->drv->mutable_opts are skipped.
3195 * Return 0 on success, otherwise return -EINVAL and set @errp.
3197 static int bdrv_reset_options_allowed(BlockDriverState
*bs
,
3198 const QDict
*new_opts
, Error
**errp
)
3200 const QDictEntry
*e
;
3201 /* These options are common to all block drivers and are handled
3202 * in bdrv_reopen_prepare() so they can be left out of @new_opts */
3203 const char *const common_options
[] = {
3204 "node-name", "discard", "cache.direct", "cache.no-flush",
3205 "read-only", "auto-read-only", "detect-zeroes", NULL
3208 for (e
= qdict_first(bs
->options
); e
; e
= qdict_next(bs
->options
, e
)) {
3209 if (!qdict_haskey(new_opts
, e
->key
) &&
3210 !is_str_in_list(e
->key
, common_options
) &&
3211 !is_str_in_list(e
->key
, bs
->drv
->mutable_opts
)) {
3212 error_setg(errp
, "Option '%s' cannot be reset "
3213 "to its default value", e
->key
);
3222 * Returns true if @child can be reached recursively from @bs
3224 static bool bdrv_recurse_has_child(BlockDriverState
*bs
,
3225 BlockDriverState
*child
)
3233 QLIST_FOREACH(c
, &bs
->children
, next
) {
3234 if (bdrv_recurse_has_child(c
->bs
, child
)) {
3243 * Adds a BlockDriverState to a simple queue for an atomic, transactional
3244 * reopen of multiple devices.
3246 * bs_queue can either be an existing BlockReopenQueue that has had QSIMPLE_INIT
3247 * already performed, or alternatively may be NULL a new BlockReopenQueue will
3248 * be created and initialized. This newly created BlockReopenQueue should be
3249 * passed back in for subsequent calls that are intended to be of the same
3252 * bs is the BlockDriverState to add to the reopen queue.
3254 * options contains the changed options for the associated bs
3255 * (the BlockReopenQueue takes ownership)
3257 * flags contains the open flags for the associated bs
3259 * returns a pointer to bs_queue, which is either the newly allocated
3260 * bs_queue, or the existing bs_queue being used.
3262 * bs must be drained between bdrv_reopen_queue() and bdrv_reopen_multiple().
3264 static BlockReopenQueue
*bdrv_reopen_queue_child(BlockReopenQueue
*bs_queue
,
3265 BlockDriverState
*bs
,
3267 const BdrvChildRole
*role
,
3268 QDict
*parent_options
,
3274 BlockReopenQueueEntry
*bs_entry
;
3276 QDict
*old_options
, *explicit_options
, *options_copy
;
3280 /* Make sure that the caller remembered to use a drained section. This is
3281 * important to avoid graph changes between the recursive queuing here and
3282 * bdrv_reopen_multiple(). */
3283 assert(bs
->quiesce_counter
> 0);
3285 if (bs_queue
== NULL
) {
3286 bs_queue
= g_new0(BlockReopenQueue
, 1);
3287 QSIMPLEQ_INIT(bs_queue
);
3291 options
= qdict_new();
3294 /* Check if this BlockDriverState is already in the queue */
3295 QSIMPLEQ_FOREACH(bs_entry
, bs_queue
, entry
) {
3296 if (bs
== bs_entry
->state
.bs
) {
3302 * Precedence of options:
3303 * 1. Explicitly passed in options (highest)
3304 * 2. Retained from explicitly set options of bs
3305 * 3. Inherited from parent node
3306 * 4. Retained from effective options of bs
3309 /* Old explicitly set values (don't overwrite by inherited value) */
3310 if (bs_entry
|| keep_old_opts
) {
3311 old_options
= qdict_clone_shallow(bs_entry
?
3312 bs_entry
->state
.explicit_options
:
3313 bs
->explicit_options
);
3314 bdrv_join_options(bs
, options
, old_options
);
3315 qobject_unref(old_options
);
3318 explicit_options
= qdict_clone_shallow(options
);
3320 /* Inherit from parent node */
3321 if (parent_options
) {
3323 role
->inherit_options(&flags
, options
, parent_flags
, parent_options
);
3325 flags
= bdrv_get_flags(bs
);
3328 if (keep_old_opts
) {
3329 /* Old values are used for options that aren't set yet */
3330 old_options
= qdict_clone_shallow(bs
->options
);
3331 bdrv_join_options(bs
, options
, old_options
);
3332 qobject_unref(old_options
);
3335 /* We have the final set of options so let's update the flags */
3336 options_copy
= qdict_clone_shallow(options
);
3337 opts
= qemu_opts_create(&bdrv_runtime_opts
, NULL
, 0, &error_abort
);
3338 qemu_opts_absorb_qdict(opts
, options_copy
, NULL
);
3339 update_flags_from_options(&flags
, opts
);
3340 qemu_opts_del(opts
);
3341 qobject_unref(options_copy
);
3343 /* bdrv_open_inherit() sets and clears some additional flags internally */
3344 flags
&= ~BDRV_O_PROTOCOL
;
3345 if (flags
& BDRV_O_RDWR
) {
3346 flags
|= BDRV_O_ALLOW_RDWR
;
3350 bs_entry
= g_new0(BlockReopenQueueEntry
, 1);
3351 QSIMPLEQ_INSERT_TAIL(bs_queue
, bs_entry
, entry
);
3353 qobject_unref(bs_entry
->state
.options
);
3354 qobject_unref(bs_entry
->state
.explicit_options
);
3357 bs_entry
->state
.bs
= bs
;
3358 bs_entry
->state
.options
= options
;
3359 bs_entry
->state
.explicit_options
= explicit_options
;
3360 bs_entry
->state
.flags
= flags
;
3362 /* This needs to be overwritten in bdrv_reopen_prepare() */
3363 bs_entry
->state
.perm
= UINT64_MAX
;
3364 bs_entry
->state
.shared_perm
= 0;
3367 * If keep_old_opts is false then it means that unspecified
3368 * options must be reset to their original value. We don't allow
3369 * resetting 'backing' but we need to know if the option is
3370 * missing in order to decide if we have to return an error.
3372 if (!keep_old_opts
) {
3373 bs_entry
->state
.backing_missing
=
3374 !qdict_haskey(options
, "backing") &&
3375 !qdict_haskey(options
, "backing.driver");
3378 QLIST_FOREACH(child
, &bs
->children
, next
) {
3379 QDict
*new_child_options
= NULL
;
3380 bool child_keep_old
= keep_old_opts
;
3382 /* reopen can only change the options of block devices that were
3383 * implicitly created and inherited options. For other (referenced)
3384 * block devices, a syntax like "backing.foo" results in an error. */
3385 if (child
->bs
->inherits_from
!= bs
) {
3389 /* Check if the options contain a child reference */
3390 if (qdict_haskey(options
, child
->name
)) {
3391 const char *childref
= qdict_get_try_str(options
, child
->name
);
3393 * The current child must not be reopened if the child
3394 * reference is null or points to a different node.
3396 if (g_strcmp0(childref
, child
->bs
->node_name
)) {
3400 * If the child reference points to the current child then
3401 * reopen it with its existing set of options (note that
3402 * it can still inherit new options from the parent).
3404 child_keep_old
= true;
3406 /* Extract child options ("child-name.*") */
3407 char *child_key_dot
= g_strdup_printf("%s.", child
->name
);
3408 qdict_extract_subqdict(explicit_options
, NULL
, child_key_dot
);
3409 qdict_extract_subqdict(options
, &new_child_options
, child_key_dot
);
3410 g_free(child_key_dot
);
3413 bdrv_reopen_queue_child(bs_queue
, child
->bs
, new_child_options
,
3414 child
->role
, options
, flags
, child_keep_old
);
3420 BlockReopenQueue
*bdrv_reopen_queue(BlockReopenQueue
*bs_queue
,
3421 BlockDriverState
*bs
,
3422 QDict
*options
, bool keep_old_opts
)
3424 return bdrv_reopen_queue_child(bs_queue
, bs
, options
, NULL
, NULL
, 0,
3429 * Reopen multiple BlockDriverStates atomically & transactionally.
3431 * The queue passed in (bs_queue) must have been built up previous
3432 * via bdrv_reopen_queue().
3434 * Reopens all BDS specified in the queue, with the appropriate
3435 * flags. All devices are prepared for reopen, and failure of any
3436 * device will cause all device changes to be abandoned, and intermediate
3439 * If all devices prepare successfully, then the changes are committed
3442 * All affected nodes must be drained between bdrv_reopen_queue() and
3443 * bdrv_reopen_multiple().
3445 int bdrv_reopen_multiple(BlockReopenQueue
*bs_queue
, Error
**errp
)
3448 BlockReopenQueueEntry
*bs_entry
, *next
;
3450 assert(bs_queue
!= NULL
);
3452 QSIMPLEQ_FOREACH(bs_entry
, bs_queue
, entry
) {
3453 assert(bs_entry
->state
.bs
->quiesce_counter
> 0);
3454 if (bdrv_reopen_prepare(&bs_entry
->state
, bs_queue
, errp
)) {
3457 bs_entry
->prepared
= true;
3460 QSIMPLEQ_FOREACH(bs_entry
, bs_queue
, entry
) {
3461 BDRVReopenState
*state
= &bs_entry
->state
;
3462 ret
= bdrv_check_perm(state
->bs
, bs_queue
, state
->perm
,
3463 state
->shared_perm
, NULL
, NULL
, errp
);
3467 /* Check if new_backing_bs would accept the new permissions */
3468 if (state
->replace_backing_bs
&& state
->new_backing_bs
) {
3469 uint64_t nperm
, nshared
;
3470 bdrv_child_perm(state
->bs
, state
->new_backing_bs
,
3471 NULL
, &child_backing
, bs_queue
,
3472 state
->perm
, state
->shared_perm
,
3474 ret
= bdrv_check_update_perm(state
->new_backing_bs
, NULL
,
3475 nperm
, nshared
, NULL
, NULL
, errp
);
3480 bs_entry
->perms_checked
= true;
3483 /* If we reach this point, we have success and just need to apply the
3486 QSIMPLEQ_FOREACH(bs_entry
, bs_queue
, entry
) {
3487 bdrv_reopen_commit(&bs_entry
->state
);
3492 QSIMPLEQ_FOREACH_SAFE(bs_entry
, bs_queue
, entry
, next
) {
3493 BDRVReopenState
*state
= &bs_entry
->state
;
3495 if (!bs_entry
->perms_checked
) {
3500 bdrv_set_perm(state
->bs
, state
->perm
, state
->shared_perm
);
3502 bdrv_abort_perm_update(state
->bs
);
3503 if (state
->replace_backing_bs
&& state
->new_backing_bs
) {
3504 bdrv_abort_perm_update(state
->new_backing_bs
);
3509 QSIMPLEQ_FOREACH_SAFE(bs_entry
, bs_queue
, entry
, next
) {
3511 if (bs_entry
->prepared
) {
3512 bdrv_reopen_abort(&bs_entry
->state
);
3514 qobject_unref(bs_entry
->state
.explicit_options
);
3515 qobject_unref(bs_entry
->state
.options
);
3517 if (bs_entry
->state
.new_backing_bs
) {
3518 bdrv_unref(bs_entry
->state
.new_backing_bs
);
3527 int bdrv_reopen_set_read_only(BlockDriverState
*bs
, bool read_only
,
3531 BlockReopenQueue
*queue
;
3532 QDict
*opts
= qdict_new();
3534 qdict_put_bool(opts
, BDRV_OPT_READ_ONLY
, read_only
);
3536 bdrv_subtree_drained_begin(bs
);
3537 queue
= bdrv_reopen_queue(NULL
, bs
, opts
, true);
3538 ret
= bdrv_reopen_multiple(queue
, errp
);
3539 bdrv_subtree_drained_end(bs
);
3544 static BlockReopenQueueEntry
*find_parent_in_reopen_queue(BlockReopenQueue
*q
,
3547 BlockReopenQueueEntry
*entry
;
3549 QSIMPLEQ_FOREACH(entry
, q
, entry
) {
3550 BlockDriverState
*bs
= entry
->state
.bs
;
3553 QLIST_FOREACH(child
, &bs
->children
, next
) {
3563 static void bdrv_reopen_perm(BlockReopenQueue
*q
, BlockDriverState
*bs
,
3564 uint64_t *perm
, uint64_t *shared
)
3567 BlockReopenQueueEntry
*parent
;
3568 uint64_t cumulative_perms
= 0;
3569 uint64_t cumulative_shared_perms
= BLK_PERM_ALL
;
3571 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
3572 parent
= find_parent_in_reopen_queue(q
, c
);
3574 cumulative_perms
|= c
->perm
;
3575 cumulative_shared_perms
&= c
->shared_perm
;
3577 uint64_t nperm
, nshared
;
3579 bdrv_child_perm(parent
->state
.bs
, bs
, c
, c
->role
, q
,
3580 parent
->state
.perm
, parent
->state
.shared_perm
,
3583 cumulative_perms
|= nperm
;
3584 cumulative_shared_perms
&= nshared
;
3587 *perm
= cumulative_perms
;
3588 *shared
= cumulative_shared_perms
;
3592 * Take a BDRVReopenState and check if the value of 'backing' in the
3593 * reopen_state->options QDict is valid or not.
3595 * If 'backing' is missing from the QDict then return 0.
3597 * If 'backing' contains the node name of the backing file of
3598 * reopen_state->bs then return 0.
3600 * If 'backing' contains a different node name (or is null) then check
3601 * whether the current backing file can be replaced with the new one.
3602 * If that's the case then reopen_state->replace_backing_bs is set to
3603 * true and reopen_state->new_backing_bs contains a pointer to the new
3604 * backing BlockDriverState (or NULL).
3606 * Return 0 on success, otherwise return < 0 and set @errp.
3608 static int bdrv_reopen_parse_backing(BDRVReopenState
*reopen_state
,
3611 BlockDriverState
*bs
= reopen_state
->bs
;
3612 BlockDriverState
*overlay_bs
, *new_backing_bs
;
3616 value
= qdict_get(reopen_state
->options
, "backing");
3617 if (value
== NULL
) {
3621 switch (qobject_type(value
)) {
3623 new_backing_bs
= NULL
;
3626 str
= qobject_get_try_str(value
);
3627 new_backing_bs
= bdrv_lookup_bs(NULL
, str
, errp
);
3628 if (new_backing_bs
== NULL
) {
3630 } else if (bdrv_recurse_has_child(new_backing_bs
, bs
)) {
3631 error_setg(errp
, "Making '%s' a backing file of '%s' "
3632 "would create a cycle", str
, bs
->node_name
);
3637 /* 'backing' does not allow any other data type */
3638 g_assert_not_reached();
3642 * TODO: before removing the x- prefix from x-blockdev-reopen we
3643 * should move the new backing file into the right AioContext
3644 * instead of returning an error.
3646 if (new_backing_bs
) {
3647 if (bdrv_get_aio_context(new_backing_bs
) != bdrv_get_aio_context(bs
)) {
3648 error_setg(errp
, "Cannot use a new backing file "
3649 "with a different AioContext");
3655 * Find the "actual" backing file by skipping all links that point
3656 * to an implicit node, if any (e.g. a commit filter node).
3659 while (backing_bs(overlay_bs
) && backing_bs(overlay_bs
)->implicit
) {
3660 overlay_bs
= backing_bs(overlay_bs
);
3663 /* If we want to replace the backing file we need some extra checks */
3664 if (new_backing_bs
!= backing_bs(overlay_bs
)) {
3665 /* Check for implicit nodes between bs and its backing file */
3666 if (bs
!= overlay_bs
) {
3667 error_setg(errp
, "Cannot change backing link if '%s' has "
3668 "an implicit backing file", bs
->node_name
);
3671 /* Check if the backing link that we want to replace is frozen */
3672 if (bdrv_is_backing_chain_frozen(overlay_bs
, backing_bs(overlay_bs
),
3676 reopen_state
->replace_backing_bs
= true;
3677 if (new_backing_bs
) {
3678 bdrv_ref(new_backing_bs
);
3679 reopen_state
->new_backing_bs
= new_backing_bs
;
3687 * Prepares a BlockDriverState for reopen. All changes are staged in the
3688 * 'opaque' field of the BDRVReopenState, which is used and allocated by
3689 * the block driver layer .bdrv_reopen_prepare()
3691 * bs is the BlockDriverState to reopen
3692 * flags are the new open flags
3693 * queue is the reopen queue
3695 * Returns 0 on success, non-zero on error. On error errp will be set
3698 * On failure, bdrv_reopen_abort() will be called to clean up any data.
3699 * It is the responsibility of the caller to then call the abort() or
3700 * commit() for any other BDS that have been left in a prepare() state
3703 int bdrv_reopen_prepare(BDRVReopenState
*reopen_state
, BlockReopenQueue
*queue
,
3708 Error
*local_err
= NULL
;
3711 QDict
*orig_reopen_opts
;
3712 char *discard
= NULL
;
3714 bool drv_prepared
= false;
3716 assert(reopen_state
!= NULL
);
3717 assert(reopen_state
->bs
->drv
!= NULL
);
3718 drv
= reopen_state
->bs
->drv
;
3720 /* This function and each driver's bdrv_reopen_prepare() remove
3721 * entries from reopen_state->options as they are processed, so
3722 * we need to make a copy of the original QDict. */
3723 orig_reopen_opts
= qdict_clone_shallow(reopen_state
->options
);
3725 /* Process generic block layer options */
3726 opts
= qemu_opts_create(&bdrv_runtime_opts
, NULL
, 0, &error_abort
);
3727 qemu_opts_absorb_qdict(opts
, reopen_state
->options
, &local_err
);
3729 error_propagate(errp
, local_err
);
3734 /* This was already called in bdrv_reopen_queue_child() so the flags
3735 * are up-to-date. This time we simply want to remove the options from
3736 * QemuOpts in order to indicate that they have been processed. */
3737 old_flags
= reopen_state
->flags
;
3738 update_flags_from_options(&reopen_state
->flags
, opts
);
3739 assert(old_flags
== reopen_state
->flags
);
3741 discard
= qemu_opt_get_del(opts
, BDRV_OPT_DISCARD
);
3742 if (discard
!= NULL
) {
3743 if (bdrv_parse_discard_flags(discard
, &reopen_state
->flags
) != 0) {
3744 error_setg(errp
, "Invalid discard option");
3750 reopen_state
->detect_zeroes
=
3751 bdrv_parse_detect_zeroes(opts
, reopen_state
->flags
, &local_err
);
3753 error_propagate(errp
, local_err
);
3758 /* All other options (including node-name and driver) must be unchanged.
3759 * Put them back into the QDict, so that they are checked at the end
3760 * of this function. */
3761 qemu_opts_to_qdict(opts
, reopen_state
->options
);
3763 /* If we are to stay read-only, do not allow permission change
3764 * to r/w. Attempting to set to r/w may fail if either BDRV_O_ALLOW_RDWR is
3765 * not set, or if the BDS still has copy_on_read enabled */
3766 read_only
= !(reopen_state
->flags
& BDRV_O_RDWR
);
3767 ret
= bdrv_can_set_read_only(reopen_state
->bs
, read_only
, true, &local_err
);
3769 error_propagate(errp
, local_err
);
3773 /* Calculate required permissions after reopening */
3774 bdrv_reopen_perm(queue
, reopen_state
->bs
,
3775 &reopen_state
->perm
, &reopen_state
->shared_perm
);
3777 ret
= bdrv_flush(reopen_state
->bs
);
3779 error_setg_errno(errp
, -ret
, "Error flushing drive");
3783 if (drv
->bdrv_reopen_prepare
) {
3785 * If a driver-specific option is missing, it means that we
3786 * should reset it to its default value.
3787 * But not all options allow that, so we need to check it first.
3789 ret
= bdrv_reset_options_allowed(reopen_state
->bs
,
3790 reopen_state
->options
, errp
);
3795 ret
= drv
->bdrv_reopen_prepare(reopen_state
, queue
, &local_err
);
3797 if (local_err
!= NULL
) {
3798 error_propagate(errp
, local_err
);
3800 bdrv_refresh_filename(reopen_state
->bs
);
3801 error_setg(errp
, "failed while preparing to reopen image '%s'",
3802 reopen_state
->bs
->filename
);
3807 /* It is currently mandatory to have a bdrv_reopen_prepare()
3808 * handler for each supported drv. */
3809 error_setg(errp
, "Block format '%s' used by node '%s' "
3810 "does not support reopening files", drv
->format_name
,
3811 bdrv_get_device_or_node_name(reopen_state
->bs
));
3816 drv_prepared
= true;
3819 * We must provide the 'backing' option if the BDS has a backing
3820 * file or if the image file has a backing file name as part of
3821 * its metadata. Otherwise the 'backing' option can be omitted.
3823 if (drv
->supports_backing
&& reopen_state
->backing_missing
&&
3824 (backing_bs(reopen_state
->bs
) || reopen_state
->bs
->backing_file
[0])) {
3825 error_setg(errp
, "backing is missing for '%s'",
3826 reopen_state
->bs
->node_name
);
3832 * Allow changing the 'backing' option. The new value can be
3833 * either a reference to an existing node (using its node name)
3834 * or NULL to simply detach the current backing file.
3836 ret
= bdrv_reopen_parse_backing(reopen_state
, errp
);
3840 qdict_del(reopen_state
->options
, "backing");
3842 /* Options that are not handled are only okay if they are unchanged
3843 * compared to the old state. It is expected that some options are only
3844 * used for the initial open, but not reopen (e.g. filename) */
3845 if (qdict_size(reopen_state
->options
)) {
3846 const QDictEntry
*entry
= qdict_first(reopen_state
->options
);
3849 QObject
*new = entry
->value
;
3850 QObject
*old
= qdict_get(reopen_state
->bs
->options
, entry
->key
);
3852 /* Allow child references (child_name=node_name) as long as they
3853 * point to the current child (i.e. everything stays the same). */
3854 if (qobject_type(new) == QTYPE_QSTRING
) {
3856 QLIST_FOREACH(child
, &reopen_state
->bs
->children
, next
) {
3857 if (!strcmp(child
->name
, entry
->key
)) {
3863 const char *str
= qobject_get_try_str(new);
3864 if (!strcmp(child
->bs
->node_name
, str
)) {
3865 continue; /* Found child with this name, skip option */
3871 * TODO: When using -drive to specify blockdev options, all values
3872 * will be strings; however, when using -blockdev, blockdev-add or
3873 * filenames using the json:{} pseudo-protocol, they will be
3875 * In contrast, reopening options are (currently) always strings
3876 * (because you can only specify them through qemu-io; all other
3877 * callers do not specify any options).
3878 * Therefore, when using anything other than -drive to create a BDS,
3879 * this cannot detect non-string options as unchanged, because
3880 * qobject_is_equal() always returns false for objects of different
3881 * type. In the future, this should be remedied by correctly typing
3882 * all options. For now, this is not too big of an issue because
3883 * the user can simply omit options which cannot be changed anyway,
3884 * so they will stay unchanged.
3886 if (!qobject_is_equal(new, old
)) {
3887 error_setg(errp
, "Cannot change the option '%s'", entry
->key
);
3891 } while ((entry
= qdict_next(reopen_state
->options
, entry
)));
3896 /* Restore the original reopen_state->options QDict */
3897 qobject_unref(reopen_state
->options
);
3898 reopen_state
->options
= qobject_ref(orig_reopen_opts
);
3901 if (ret
< 0 && drv_prepared
) {
3902 /* drv->bdrv_reopen_prepare() has succeeded, so we need to
3903 * call drv->bdrv_reopen_abort() before signaling an error
3904 * (bdrv_reopen_multiple() will not call bdrv_reopen_abort()
3905 * when the respective bdrv_reopen_prepare() has failed) */
3906 if (drv
->bdrv_reopen_abort
) {
3907 drv
->bdrv_reopen_abort(reopen_state
);
3910 qemu_opts_del(opts
);
3911 qobject_unref(orig_reopen_opts
);
3917 * Takes the staged changes for the reopen from bdrv_reopen_prepare(), and
3918 * makes them final by swapping the staging BlockDriverState contents into
3919 * the active BlockDriverState contents.
3921 void bdrv_reopen_commit(BDRVReopenState
*reopen_state
)
3924 BlockDriverState
*bs
;
3926 bool old_can_write
, new_can_write
;
3928 assert(reopen_state
!= NULL
);
3929 bs
= reopen_state
->bs
;
3931 assert(drv
!= NULL
);
3934 !bdrv_is_read_only(bs
) && !(bdrv_get_flags(bs
) & BDRV_O_INACTIVE
);
3936 /* If there are any driver level actions to take */
3937 if (drv
->bdrv_reopen_commit
) {
3938 drv
->bdrv_reopen_commit(reopen_state
);
3941 /* set BDS specific flags now */
3942 qobject_unref(bs
->explicit_options
);
3943 qobject_unref(bs
->options
);
3945 bs
->explicit_options
= reopen_state
->explicit_options
;
3946 bs
->options
= reopen_state
->options
;
3947 bs
->open_flags
= reopen_state
->flags
;
3948 bs
->read_only
= !(reopen_state
->flags
& BDRV_O_RDWR
);
3949 bs
->detect_zeroes
= reopen_state
->detect_zeroes
;
3951 if (reopen_state
->replace_backing_bs
) {
3952 qdict_del(bs
->explicit_options
, "backing");
3953 qdict_del(bs
->options
, "backing");
3956 /* Remove child references from bs->options and bs->explicit_options.
3957 * Child options were already removed in bdrv_reopen_queue_child() */
3958 QLIST_FOREACH(child
, &bs
->children
, next
) {
3959 qdict_del(bs
->explicit_options
, child
->name
);
3960 qdict_del(bs
->options
, child
->name
);
3964 * Change the backing file if a new one was specified. We do this
3965 * after updating bs->options, so bdrv_refresh_filename() (called
3966 * from bdrv_set_backing_hd()) has the new values.
3968 if (reopen_state
->replace_backing_bs
) {
3969 BlockDriverState
*old_backing_bs
= backing_bs(bs
);
3970 assert(!old_backing_bs
|| !old_backing_bs
->implicit
);
3971 /* Abort the permission update on the backing bs we're detaching */
3972 if (old_backing_bs
) {
3973 bdrv_abort_perm_update(old_backing_bs
);
3975 bdrv_set_backing_hd(bs
, reopen_state
->new_backing_bs
, &error_abort
);
3978 bdrv_refresh_limits(bs
, NULL
);
3981 !bdrv_is_read_only(bs
) && !(bdrv_get_flags(bs
) & BDRV_O_INACTIVE
);
3982 if (!old_can_write
&& new_can_write
&& drv
->bdrv_reopen_bitmaps_rw
) {
3983 Error
*local_err
= NULL
;
3984 if (drv
->bdrv_reopen_bitmaps_rw(bs
, &local_err
) < 0) {
3985 /* This is not fatal, bitmaps just left read-only, so all following
3986 * writes will fail. User can remove read-only bitmaps to unblock
3989 error_reportf_err(local_err
,
3990 "%s: Failed to make dirty bitmaps writable: ",
3991 bdrv_get_node_name(bs
));
3997 * Abort the reopen, and delete and free the staged changes in
4000 void bdrv_reopen_abort(BDRVReopenState
*reopen_state
)
4004 assert(reopen_state
!= NULL
);
4005 drv
= reopen_state
->bs
->drv
;
4006 assert(drv
!= NULL
);
4008 if (drv
->bdrv_reopen_abort
) {
4009 drv
->bdrv_reopen_abort(reopen_state
);
4014 static void bdrv_close(BlockDriverState
*bs
)
4016 BdrvAioNotifier
*ban
, *ban_next
;
4017 BdrvChild
*child
, *next
;
4019 assert(!bs
->refcnt
);
4021 bdrv_drained_begin(bs
); /* complete I/O */
4023 bdrv_drain(bs
); /* in case flush left pending I/O */
4026 if (bs
->drv
->bdrv_close
) {
4027 bs
->drv
->bdrv_close(bs
);
4032 QLIST_FOREACH_SAFE(child
, &bs
->children
, next
, next
) {
4033 bdrv_unref_child(bs
, child
);
4040 atomic_set(&bs
->copy_on_read
, 0);
4041 bs
->backing_file
[0] = '\0';
4042 bs
->backing_format
[0] = '\0';
4043 bs
->total_sectors
= 0;
4044 bs
->encrypted
= false;
4046 qobject_unref(bs
->options
);
4047 qobject_unref(bs
->explicit_options
);
4049 bs
->explicit_options
= NULL
;
4050 qobject_unref(bs
->full_open_options
);
4051 bs
->full_open_options
= NULL
;
4053 bdrv_release_named_dirty_bitmaps(bs
);
4054 assert(QLIST_EMPTY(&bs
->dirty_bitmaps
));
4056 QLIST_FOREACH_SAFE(ban
, &bs
->aio_notifiers
, list
, ban_next
) {
4059 QLIST_INIT(&bs
->aio_notifiers
);
4060 bdrv_drained_end(bs
);
4063 void bdrv_close_all(void)
4065 assert(job_next(NULL
) == NULL
);
4066 nbd_export_close_all();
4068 /* Drop references from requests still in flight, such as canceled block
4069 * jobs whose AIO context has not been polled yet */
4072 blk_remove_all_bs();
4073 blockdev_close_all_bdrv_states();
4075 assert(QTAILQ_EMPTY(&all_bdrv_states
));
4078 static bool should_update_child(BdrvChild
*c
, BlockDriverState
*to
)
4084 if (c
->role
->stay_at_node
) {
4088 /* If the child @c belongs to the BDS @to, replacing the current
4089 * c->bs by @to would mean to create a loop.
4091 * Such a case occurs when appending a BDS to a backing chain.
4092 * For instance, imagine the following chain:
4094 * guest device -> node A -> further backing chain...
4096 * Now we create a new BDS B which we want to put on top of this
4097 * chain, so we first attach A as its backing node:
4102 * guest device -> node A -> further backing chain...
4104 * Finally we want to replace A by B. When doing that, we want to
4105 * replace all pointers to A by pointers to B -- except for the
4106 * pointer from B because (1) that would create a loop, and (2)
4107 * that pointer should simply stay intact:
4109 * guest device -> node B
4112 * node A -> further backing chain...
4114 * In general, when replacing a node A (c->bs) by a node B (@to),
4115 * if A is a child of B, that means we cannot replace A by B there
4116 * because that would create a loop. Silently detaching A from B
4117 * is also not really an option. So overall just leaving A in
4118 * place there is the most sensible choice.
4120 * We would also create a loop in any cases where @c is only
4121 * indirectly referenced by @to. Prevent this by returning false
4122 * if @c is found (by breadth-first search) anywhere in the whole
4127 found
= g_hash_table_new(NULL
, NULL
);
4128 g_hash_table_add(found
, to
);
4129 queue
= g_queue_new();
4130 g_queue_push_tail(queue
, to
);
4132 while (!g_queue_is_empty(queue
)) {
4133 BlockDriverState
*v
= g_queue_pop_head(queue
);
4136 QLIST_FOREACH(c2
, &v
->children
, next
) {
4142 if (g_hash_table_contains(found
, c2
->bs
)) {
4146 g_queue_push_tail(queue
, c2
->bs
);
4147 g_hash_table_add(found
, c2
->bs
);
4151 g_queue_free(queue
);
4152 g_hash_table_destroy(found
);
4157 void bdrv_replace_node(BlockDriverState
*from
, BlockDriverState
*to
,
4160 BdrvChild
*c
, *next
;
4161 GSList
*list
= NULL
, *p
;
4162 uint64_t old_perm
, old_shared
;
4163 uint64_t perm
= 0, shared
= BLK_PERM_ALL
;
4166 /* Make sure that @from doesn't go away until we have successfully attached
4167 * all of its parents to @to. */
4170 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
4171 bdrv_drained_begin(from
);
4173 /* Put all parents into @list and calculate their cumulative permissions */
4174 QLIST_FOREACH_SAFE(c
, &from
->parents
, next_parent
, next
) {
4175 assert(c
->bs
== from
);
4176 if (!should_update_child(c
, to
)) {
4180 error_setg(errp
, "Cannot change '%s' link to '%s'",
4181 c
->name
, from
->node_name
);
4184 list
= g_slist_prepend(list
, c
);
4186 shared
&= c
->shared_perm
;
4189 /* Check whether the required permissions can be granted on @to, ignoring
4190 * all BdrvChild in @list so that they can't block themselves. */
4191 ret
= bdrv_check_update_perm(to
, NULL
, perm
, shared
, list
, NULL
, errp
);
4193 bdrv_abort_perm_update(to
);
4197 /* Now actually perform the change. We performed the permission check for
4198 * all elements of @list at once, so set the permissions all at once at the
4200 for (p
= list
; p
!= NULL
; p
= p
->next
) {
4204 bdrv_replace_child_noperm(c
, to
);
4208 bdrv_get_cumulative_perm(to
, &old_perm
, &old_shared
);
4209 bdrv_set_perm(to
, old_perm
| perm
, old_shared
| shared
);
4213 bdrv_drained_end(from
);
4218 * Add new bs contents at the top of an image chain while the chain is
4219 * live, while keeping required fields on the top layer.
4221 * This will modify the BlockDriverState fields, and swap contents
4222 * between bs_new and bs_top. Both bs_new and bs_top are modified.
4224 * bs_new must not be attached to a BlockBackend.
4226 * This function does not create any image files.
4228 * bdrv_append() takes ownership of a bs_new reference and unrefs it because
4229 * that's what the callers commonly need. bs_new will be referenced by the old
4230 * parents of bs_top after bdrv_append() returns. If the caller needs to keep a
4231 * reference of its own, it must call bdrv_ref().
4233 void bdrv_append(BlockDriverState
*bs_new
, BlockDriverState
*bs_top
,
4236 Error
*local_err
= NULL
;
4238 bdrv_set_backing_hd(bs_new
, bs_top
, &local_err
);
4240 error_propagate(errp
, local_err
);
4244 bdrv_replace_node(bs_top
, bs_new
, &local_err
);
4246 error_propagate(errp
, local_err
);
4247 bdrv_set_backing_hd(bs_new
, NULL
, &error_abort
);
4251 /* bs_new is now referenced by its new parents, we don't need the
4252 * additional reference any more. */
4257 static void bdrv_delete(BlockDriverState
*bs
)
4259 assert(bdrv_op_blocker_is_empty(bs
));
4260 assert(!bs
->refcnt
);
4262 /* remove from list, if necessary */
4263 if (bs
->node_name
[0] != '\0') {
4264 QTAILQ_REMOVE(&graph_bdrv_states
, bs
, node_list
);
4266 QTAILQ_REMOVE(&all_bdrv_states
, bs
, bs_list
);
4274 * Run consistency checks on an image
4276 * Returns 0 if the check could be completed (it doesn't mean that the image is
4277 * free of errors) or -errno when an internal error occurred. The results of the
4278 * check are stored in res.
4280 static int coroutine_fn
bdrv_co_check(BlockDriverState
*bs
,
4281 BdrvCheckResult
*res
, BdrvCheckMode fix
)
4283 if (bs
->drv
== NULL
) {
4286 if (bs
->drv
->bdrv_co_check
== NULL
) {
4290 memset(res
, 0, sizeof(*res
));
4291 return bs
->drv
->bdrv_co_check(bs
, res
, fix
);
4294 typedef struct CheckCo
{
4295 BlockDriverState
*bs
;
4296 BdrvCheckResult
*res
;
4301 static void coroutine_fn
bdrv_check_co_entry(void *opaque
)
4303 CheckCo
*cco
= opaque
;
4304 cco
->ret
= bdrv_co_check(cco
->bs
, cco
->res
, cco
->fix
);
4308 int bdrv_check(BlockDriverState
*bs
,
4309 BdrvCheckResult
*res
, BdrvCheckMode fix
)
4315 .ret
= -EINPROGRESS
,
4319 if (qemu_in_coroutine()) {
4320 /* Fast-path if already in coroutine context */
4321 bdrv_check_co_entry(&cco
);
4323 co
= qemu_coroutine_create(bdrv_check_co_entry
, &cco
);
4324 bdrv_coroutine_enter(bs
, co
);
4325 BDRV_POLL_WHILE(bs
, cco
.ret
== -EINPROGRESS
);
4334 * -EINVAL - backing format specified, but no file
4335 * -ENOSPC - can't update the backing file because no space is left in the
4337 * -ENOTSUP - format driver doesn't support changing the backing file
4339 int bdrv_change_backing_file(BlockDriverState
*bs
,
4340 const char *backing_file
, const char *backing_fmt
)
4342 BlockDriver
*drv
= bs
->drv
;
4349 /* Backing file format doesn't make sense without a backing file */
4350 if (backing_fmt
&& !backing_file
) {
4354 if (drv
->bdrv_change_backing_file
!= NULL
) {
4355 ret
= drv
->bdrv_change_backing_file(bs
, backing_file
, backing_fmt
);
4361 pstrcpy(bs
->backing_file
, sizeof(bs
->backing_file
), backing_file
?: "");
4362 pstrcpy(bs
->backing_format
, sizeof(bs
->backing_format
), backing_fmt
?: "");
4363 pstrcpy(bs
->auto_backing_file
, sizeof(bs
->auto_backing_file
),
4364 backing_file
?: "");
4370 * Finds the image layer in the chain that has 'bs' as its backing file.
4372 * active is the current topmost image.
4374 * Returns NULL if bs is not found in active's image chain,
4375 * or if active == bs.
4377 * Returns the bottommost base image if bs == NULL.
4379 BlockDriverState
*bdrv_find_overlay(BlockDriverState
*active
,
4380 BlockDriverState
*bs
)
4382 while (active
&& bs
!= backing_bs(active
)) {
4383 active
= backing_bs(active
);
4389 /* Given a BDS, searches for the base layer. */
4390 BlockDriverState
*bdrv_find_base(BlockDriverState
*bs
)
4392 return bdrv_find_overlay(bs
, NULL
);
4396 * Return true if at least one of the backing links between @bs and
4397 * @base is frozen. @errp is set if that's the case.
4398 * @base must be reachable from @bs, or NULL.
4400 bool bdrv_is_backing_chain_frozen(BlockDriverState
*bs
, BlockDriverState
*base
,
4403 BlockDriverState
*i
;
4405 for (i
= bs
; i
!= base
; i
= backing_bs(i
)) {
4406 if (i
->backing
&& i
->backing
->frozen
) {
4407 error_setg(errp
, "Cannot change '%s' link from '%s' to '%s'",
4408 i
->backing
->name
, i
->node_name
,
4409 backing_bs(i
)->node_name
);
4418 * Freeze all backing links between @bs and @base.
4419 * If any of the links is already frozen the operation is aborted and
4420 * none of the links are modified.
4421 * @base must be reachable from @bs, or NULL.
4422 * Returns 0 on success. On failure returns < 0 and sets @errp.
4424 int bdrv_freeze_backing_chain(BlockDriverState
*bs
, BlockDriverState
*base
,
4427 BlockDriverState
*i
;
4429 if (bdrv_is_backing_chain_frozen(bs
, base
, errp
)) {
4433 for (i
= bs
; i
!= base
; i
= backing_bs(i
)) {
4434 if (i
->backing
&& backing_bs(i
)->never_freeze
) {
4435 error_setg(errp
, "Cannot freeze '%s' link to '%s'",
4436 i
->backing
->name
, backing_bs(i
)->node_name
);
4441 for (i
= bs
; i
!= base
; i
= backing_bs(i
)) {
4443 i
->backing
->frozen
= true;
4451 * Unfreeze all backing links between @bs and @base. The caller must
4452 * ensure that all links are frozen before using this function.
4453 * @base must be reachable from @bs, or NULL.
4455 void bdrv_unfreeze_backing_chain(BlockDriverState
*bs
, BlockDriverState
*base
)
4457 BlockDriverState
*i
;
4459 for (i
= bs
; i
!= base
; i
= backing_bs(i
)) {
4461 assert(i
->backing
->frozen
);
4462 i
->backing
->frozen
= false;
4468 * Drops images above 'base' up to and including 'top', and sets the image
4469 * above 'top' to have base as its backing file.
4471 * Requires that the overlay to 'top' is opened r/w, so that the backing file
4472 * information in 'bs' can be properly updated.
4474 * E.g., this will convert the following chain:
4475 * bottom <- base <- intermediate <- top <- active
4479 * bottom <- base <- active
4481 * It is allowed for bottom==base, in which case it converts:
4483 * base <- intermediate <- top <- active
4489 * If backing_file_str is non-NULL, it will be used when modifying top's
4490 * overlay image metadata.
4493 * if active == top, that is considered an error
4496 int bdrv_drop_intermediate(BlockDriverState
*top
, BlockDriverState
*base
,
4497 const char *backing_file_str
)
4499 BlockDriverState
*explicit_top
= top
;
4500 bool update_inherits_from
;
4501 BdrvChild
*c
, *next
;
4502 Error
*local_err
= NULL
;
4507 if (!top
->drv
|| !base
->drv
) {
4511 /* Make sure that base is in the backing chain of top */
4512 if (!bdrv_chain_contains(top
, base
)) {
4516 /* This function changes all links that point to top and makes
4517 * them point to base. Check that none of them is frozen. */
4518 QLIST_FOREACH(c
, &top
->parents
, next_parent
) {
4524 /* If 'base' recursively inherits from 'top' then we should set
4525 * base->inherits_from to top->inherits_from after 'top' and all
4526 * other intermediate nodes have been dropped.
4527 * If 'top' is an implicit node (e.g. "commit_top") we should skip
4528 * it because no one inherits from it. We use explicit_top for that. */
4529 while (explicit_top
&& explicit_top
->implicit
) {
4530 explicit_top
= backing_bs(explicit_top
);
4532 update_inherits_from
= bdrv_inherits_from_recursive(base
, explicit_top
);
4534 /* success - we can delete the intermediate states, and link top->base */
4535 /* TODO Check graph modification op blockers (BLK_PERM_GRAPH_MOD) once
4536 * we've figured out how they should work. */
4537 if (!backing_file_str
) {
4538 bdrv_refresh_filename(base
);
4539 backing_file_str
= base
->filename
;
4542 QLIST_FOREACH_SAFE(c
, &top
->parents
, next_parent
, next
) {
4543 /* Check whether we are allowed to switch c from top to base */
4544 GSList
*ignore_children
= g_slist_prepend(NULL
, c
);
4545 ret
= bdrv_check_update_perm(base
, NULL
, c
->perm
, c
->shared_perm
,
4546 ignore_children
, NULL
, &local_err
);
4547 g_slist_free(ignore_children
);
4549 error_report_err(local_err
);
4553 /* If so, update the backing file path in the image file */
4554 if (c
->role
->update_filename
) {
4555 ret
= c
->role
->update_filename(c
, base
, backing_file_str
,
4558 bdrv_abort_perm_update(base
);
4559 error_report_err(local_err
);
4564 /* Do the actual switch in the in-memory graph.
4565 * Completes bdrv_check_update_perm() transaction internally. */
4567 bdrv_replace_child(c
, base
);
4571 if (update_inherits_from
) {
4572 base
->inherits_from
= explicit_top
->inherits_from
;
4582 * Length of a allocated file in bytes. Sparse files are counted by actual
4583 * allocated space. Return < 0 if error or unknown.
4585 int64_t bdrv_get_allocated_file_size(BlockDriverState
*bs
)
4587 BlockDriver
*drv
= bs
->drv
;
4591 if (drv
->bdrv_get_allocated_file_size
) {
4592 return drv
->bdrv_get_allocated_file_size(bs
);
4595 return bdrv_get_allocated_file_size(bs
->file
->bs
);
4602 * @drv: Format driver
4603 * @opts: Creation options for new image
4604 * @in_bs: Existing image containing data for new image (may be NULL)
4605 * @errp: Error object
4606 * Returns: A #BlockMeasureInfo (free using qapi_free_BlockMeasureInfo())
4609 * Calculate file size required to create a new image.
4611 * If @in_bs is given then space for allocated clusters and zero clusters
4612 * from that image are included in the calculation. If @opts contains a
4613 * backing file that is shared by @in_bs then backing clusters may be omitted
4614 * from the calculation.
4616 * If @in_bs is NULL then the calculation includes no allocated clusters
4617 * unless a preallocation option is given in @opts.
4619 * Note that @in_bs may use a different BlockDriver from @drv.
4621 * If an error occurs the @errp pointer is set.
4623 BlockMeasureInfo
*bdrv_measure(BlockDriver
*drv
, QemuOpts
*opts
,
4624 BlockDriverState
*in_bs
, Error
**errp
)
4626 if (!drv
->bdrv_measure
) {
4627 error_setg(errp
, "Block driver '%s' does not support size measurement",
4632 return drv
->bdrv_measure(opts
, in_bs
, errp
);
4636 * Return number of sectors on success, -errno on error.
4638 int64_t bdrv_nb_sectors(BlockDriverState
*bs
)
4640 BlockDriver
*drv
= bs
->drv
;
4645 if (drv
->has_variable_length
) {
4646 int ret
= refresh_total_sectors(bs
, bs
->total_sectors
);
4651 return bs
->total_sectors
;
4655 * Return length in bytes on success, -errno on error.
4656 * The length is always a multiple of BDRV_SECTOR_SIZE.
4658 int64_t bdrv_getlength(BlockDriverState
*bs
)
4660 int64_t ret
= bdrv_nb_sectors(bs
);
4662 ret
= ret
> INT64_MAX
/ BDRV_SECTOR_SIZE
? -EFBIG
: ret
;
4663 return ret
< 0 ? ret
: ret
* BDRV_SECTOR_SIZE
;
4666 /* return 0 as number of sectors if no device present or error */
4667 void bdrv_get_geometry(BlockDriverState
*bs
, uint64_t *nb_sectors_ptr
)
4669 int64_t nb_sectors
= bdrv_nb_sectors(bs
);
4671 *nb_sectors_ptr
= nb_sectors
< 0 ? 0 : nb_sectors
;
4674 bool bdrv_is_sg(BlockDriverState
*bs
)
4679 bool bdrv_is_encrypted(BlockDriverState
*bs
)
4681 if (bs
->backing
&& bs
->backing
->bs
->encrypted
) {
4684 return bs
->encrypted
;
4687 const char *bdrv_get_format_name(BlockDriverState
*bs
)
4689 return bs
->drv
? bs
->drv
->format_name
: NULL
;
4692 static int qsort_strcmp(const void *a
, const void *b
)
4694 return strcmp(*(char *const *)a
, *(char *const *)b
);
4697 void bdrv_iterate_format(void (*it
)(void *opaque
, const char *name
),
4698 void *opaque
, bool read_only
)
4703 const char **formats
= NULL
;
4705 QLIST_FOREACH(drv
, &bdrv_drivers
, list
) {
4706 if (drv
->format_name
) {
4710 if (use_bdrv_whitelist
&& !bdrv_is_whitelisted(drv
, read_only
)) {
4714 while (formats
&& i
&& !found
) {
4715 found
= !strcmp(formats
[--i
], drv
->format_name
);
4719 formats
= g_renew(const char *, formats
, count
+ 1);
4720 formats
[count
++] = drv
->format_name
;
4725 for (i
= 0; i
< (int)ARRAY_SIZE(block_driver_modules
); i
++) {
4726 const char *format_name
= block_driver_modules
[i
].format_name
;
4732 if (use_bdrv_whitelist
&&
4733 !bdrv_format_is_whitelisted(format_name
, read_only
)) {
4737 while (formats
&& j
&& !found
) {
4738 found
= !strcmp(formats
[--j
], format_name
);
4742 formats
= g_renew(const char *, formats
, count
+ 1);
4743 formats
[count
++] = format_name
;
4748 qsort(formats
, count
, sizeof(formats
[0]), qsort_strcmp
);
4750 for (i
= 0; i
< count
; i
++) {
4751 it(opaque
, formats
[i
]);
4757 /* This function is to find a node in the bs graph */
4758 BlockDriverState
*bdrv_find_node(const char *node_name
)
4760 BlockDriverState
*bs
;
4764 QTAILQ_FOREACH(bs
, &graph_bdrv_states
, node_list
) {
4765 if (!strcmp(node_name
, bs
->node_name
)) {
4772 /* Put this QMP function here so it can access the static graph_bdrv_states. */
4773 BlockDeviceInfoList
*bdrv_named_nodes_list(Error
**errp
)
4775 BlockDeviceInfoList
*list
, *entry
;
4776 BlockDriverState
*bs
;
4779 QTAILQ_FOREACH(bs
, &graph_bdrv_states
, node_list
) {
4780 BlockDeviceInfo
*info
= bdrv_block_device_info(NULL
, bs
, errp
);
4782 qapi_free_BlockDeviceInfoList(list
);
4785 entry
= g_malloc0(sizeof(*entry
));
4786 entry
->value
= info
;
4794 #define QAPI_LIST_ADD(list, element) do { \
4795 typeof(list) _tmp = g_new(typeof(*(list)), 1); \
4796 _tmp->value = (element); \
4797 _tmp->next = (list); \
4801 typedef struct XDbgBlockGraphConstructor
{
4802 XDbgBlockGraph
*graph
;
4803 GHashTable
*graph_nodes
;
4804 } XDbgBlockGraphConstructor
;
4806 static XDbgBlockGraphConstructor
*xdbg_graph_new(void)
4808 XDbgBlockGraphConstructor
*gr
= g_new(XDbgBlockGraphConstructor
, 1);
4810 gr
->graph
= g_new0(XDbgBlockGraph
, 1);
4811 gr
->graph_nodes
= g_hash_table_new(NULL
, NULL
);
4816 static XDbgBlockGraph
*xdbg_graph_finalize(XDbgBlockGraphConstructor
*gr
)
4818 XDbgBlockGraph
*graph
= gr
->graph
;
4820 g_hash_table_destroy(gr
->graph_nodes
);
4826 static uintptr_t xdbg_graph_node_num(XDbgBlockGraphConstructor
*gr
, void *node
)
4828 uintptr_t ret
= (uintptr_t)g_hash_table_lookup(gr
->graph_nodes
, node
);
4835 * Start counting from 1, not 0, because 0 interferes with not-found (NULL)
4836 * answer of g_hash_table_lookup.
4838 ret
= g_hash_table_size(gr
->graph_nodes
) + 1;
4839 g_hash_table_insert(gr
->graph_nodes
, node
, (void *)ret
);
4844 static void xdbg_graph_add_node(XDbgBlockGraphConstructor
*gr
, void *node
,
4845 XDbgBlockGraphNodeType type
, const char *name
)
4847 XDbgBlockGraphNode
*n
;
4849 n
= g_new0(XDbgBlockGraphNode
, 1);
4851 n
->id
= xdbg_graph_node_num(gr
, node
);
4853 n
->name
= g_strdup(name
);
4855 QAPI_LIST_ADD(gr
->graph
->nodes
, n
);
4858 static void xdbg_graph_add_edge(XDbgBlockGraphConstructor
*gr
, void *parent
,
4859 const BdrvChild
*child
)
4863 BlockPermission num
;
4866 static const PermissionMap permissions
[] = {
4867 { BLK_PERM_CONSISTENT_READ
, BLOCK_PERMISSION_CONSISTENT_READ
},
4868 { BLK_PERM_WRITE
, BLOCK_PERMISSION_WRITE
},
4869 { BLK_PERM_WRITE_UNCHANGED
, BLOCK_PERMISSION_WRITE_UNCHANGED
},
4870 { BLK_PERM_RESIZE
, BLOCK_PERMISSION_RESIZE
},
4871 { BLK_PERM_GRAPH_MOD
, BLOCK_PERMISSION_GRAPH_MOD
},
4874 const PermissionMap
*p
;
4875 XDbgBlockGraphEdge
*edge
;
4877 QEMU_BUILD_BUG_ON(1UL << (ARRAY_SIZE(permissions
) - 1) != BLK_PERM_ALL
+ 1);
4879 edge
= g_new0(XDbgBlockGraphEdge
, 1);
4881 edge
->parent
= xdbg_graph_node_num(gr
, parent
);
4882 edge
->child
= xdbg_graph_node_num(gr
, child
->bs
);
4883 edge
->name
= g_strdup(child
->name
);
4885 for (p
= permissions
; p
->flag
; p
++) {
4886 if (p
->flag
& child
->perm
) {
4887 QAPI_LIST_ADD(edge
->perm
, p
->num
);
4889 if (p
->flag
& child
->shared_perm
) {
4890 QAPI_LIST_ADD(edge
->shared_perm
, p
->num
);
4894 QAPI_LIST_ADD(gr
->graph
->edges
, edge
);
4898 XDbgBlockGraph
*bdrv_get_xdbg_block_graph(Error
**errp
)
4902 BlockDriverState
*bs
;
4904 XDbgBlockGraphConstructor
*gr
= xdbg_graph_new();
4906 for (blk
= blk_all_next(NULL
); blk
; blk
= blk_all_next(blk
)) {
4907 char *allocated_name
= NULL
;
4908 const char *name
= blk_name(blk
);
4911 name
= allocated_name
= blk_get_attached_dev_id(blk
);
4913 xdbg_graph_add_node(gr
, blk
, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_BACKEND
,
4915 g_free(allocated_name
);
4916 if (blk_root(blk
)) {
4917 xdbg_graph_add_edge(gr
, blk
, blk_root(blk
));
4921 for (job
= block_job_next(NULL
); job
; job
= block_job_next(job
)) {
4924 xdbg_graph_add_node(gr
, job
, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_JOB
,
4926 for (el
= job
->nodes
; el
; el
= el
->next
) {
4927 xdbg_graph_add_edge(gr
, job
, (BdrvChild
*)el
->data
);
4931 QTAILQ_FOREACH(bs
, &graph_bdrv_states
, node_list
) {
4932 xdbg_graph_add_node(gr
, bs
, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_DRIVER
,
4934 QLIST_FOREACH(child
, &bs
->children
, next
) {
4935 xdbg_graph_add_edge(gr
, bs
, child
);
4939 return xdbg_graph_finalize(gr
);
4942 BlockDriverState
*bdrv_lookup_bs(const char *device
,
4943 const char *node_name
,
4947 BlockDriverState
*bs
;
4950 blk
= blk_by_name(device
);
4955 error_setg(errp
, "Device '%s' has no medium", device
);
4963 bs
= bdrv_find_node(node_name
);
4970 error_setg(errp
, "Cannot find device=%s nor node_name=%s",
4971 device
? device
: "",
4972 node_name
? node_name
: "");
4976 /* If 'base' is in the same chain as 'top', return true. Otherwise,
4977 * return false. If either argument is NULL, return false. */
4978 bool bdrv_chain_contains(BlockDriverState
*top
, BlockDriverState
*base
)
4980 while (top
&& top
!= base
) {
4981 top
= backing_bs(top
);
4987 BlockDriverState
*bdrv_next_node(BlockDriverState
*bs
)
4990 return QTAILQ_FIRST(&graph_bdrv_states
);
4992 return QTAILQ_NEXT(bs
, node_list
);
4995 BlockDriverState
*bdrv_next_all_states(BlockDriverState
*bs
)
4998 return QTAILQ_FIRST(&all_bdrv_states
);
5000 return QTAILQ_NEXT(bs
, bs_list
);
5003 const char *bdrv_get_node_name(const BlockDriverState
*bs
)
5005 return bs
->node_name
;
5008 const char *bdrv_get_parent_name(const BlockDriverState
*bs
)
5013 /* If multiple parents have a name, just pick the first one. */
5014 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
5015 if (c
->role
->get_name
) {
5016 name
= c
->role
->get_name(c
);
5017 if (name
&& *name
) {
5026 /* TODO check what callers really want: bs->node_name or blk_name() */
5027 const char *bdrv_get_device_name(const BlockDriverState
*bs
)
5029 return bdrv_get_parent_name(bs
) ?: "";
5032 /* This can be used to identify nodes that might not have a device
5033 * name associated. Since node and device names live in the same
5034 * namespace, the result is unambiguous. The exception is if both are
5035 * absent, then this returns an empty (non-null) string. */
5036 const char *bdrv_get_device_or_node_name(const BlockDriverState
*bs
)
5038 return bdrv_get_parent_name(bs
) ?: bs
->node_name
;
5041 int bdrv_get_flags(BlockDriverState
*bs
)
5043 return bs
->open_flags
;
5046 int bdrv_has_zero_init_1(BlockDriverState
*bs
)
5051 int bdrv_has_zero_init(BlockDriverState
*bs
)
5057 /* If BS is a copy on write image, it is initialized to
5058 the contents of the base image, which may not be zeroes. */
5062 if (bs
->drv
->bdrv_has_zero_init
) {
5063 return bs
->drv
->bdrv_has_zero_init(bs
);
5065 if (bs
->file
&& bs
->drv
->is_filter
) {
5066 return bdrv_has_zero_init(bs
->file
->bs
);
5073 bool bdrv_unallocated_blocks_are_zero(BlockDriverState
*bs
)
5075 BlockDriverInfo bdi
;
5081 if (bdrv_get_info(bs
, &bdi
) == 0) {
5082 return bdi
.unallocated_blocks_are_zero
;
5088 bool bdrv_can_write_zeroes_with_unmap(BlockDriverState
*bs
)
5090 if (!(bs
->open_flags
& BDRV_O_UNMAP
)) {
5094 return bs
->supported_zero_flags
& BDRV_REQ_MAY_UNMAP
;
5097 void bdrv_get_backing_filename(BlockDriverState
*bs
,
5098 char *filename
, int filename_size
)
5100 pstrcpy(filename
, filename_size
, bs
->backing_file
);
5103 int bdrv_get_info(BlockDriverState
*bs
, BlockDriverInfo
*bdi
)
5105 BlockDriver
*drv
= bs
->drv
;
5106 /* if bs->drv == NULL, bs is closed, so there's nothing to do here */
5110 if (!drv
->bdrv_get_info
) {
5111 if (bs
->file
&& drv
->is_filter
) {
5112 return bdrv_get_info(bs
->file
->bs
, bdi
);
5116 memset(bdi
, 0, sizeof(*bdi
));
5117 return drv
->bdrv_get_info(bs
, bdi
);
5120 ImageInfoSpecific
*bdrv_get_specific_info(BlockDriverState
*bs
,
5123 BlockDriver
*drv
= bs
->drv
;
5124 if (drv
&& drv
->bdrv_get_specific_info
) {
5125 return drv
->bdrv_get_specific_info(bs
, errp
);
5130 void bdrv_debug_event(BlockDriverState
*bs
, BlkdebugEvent event
)
5132 if (!bs
|| !bs
->drv
|| !bs
->drv
->bdrv_debug_event
) {
5136 bs
->drv
->bdrv_debug_event(bs
, event
);
5139 int bdrv_debug_breakpoint(BlockDriverState
*bs
, const char *event
,
5142 while (bs
&& bs
->drv
&& !bs
->drv
->bdrv_debug_breakpoint
) {
5143 bs
= bs
->file
? bs
->file
->bs
: NULL
;
5146 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_breakpoint
) {
5147 return bs
->drv
->bdrv_debug_breakpoint(bs
, event
, tag
);
5153 int bdrv_debug_remove_breakpoint(BlockDriverState
*bs
, const char *tag
)
5155 while (bs
&& bs
->drv
&& !bs
->drv
->bdrv_debug_remove_breakpoint
) {
5156 bs
= bs
->file
? bs
->file
->bs
: NULL
;
5159 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_remove_breakpoint
) {
5160 return bs
->drv
->bdrv_debug_remove_breakpoint(bs
, tag
);
5166 int bdrv_debug_resume(BlockDriverState
*bs
, const char *tag
)
5168 while (bs
&& (!bs
->drv
|| !bs
->drv
->bdrv_debug_resume
)) {
5169 bs
= bs
->file
? bs
->file
->bs
: NULL
;
5172 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_resume
) {
5173 return bs
->drv
->bdrv_debug_resume(bs
, tag
);
5179 bool bdrv_debug_is_suspended(BlockDriverState
*bs
, const char *tag
)
5181 while (bs
&& bs
->drv
&& !bs
->drv
->bdrv_debug_is_suspended
) {
5182 bs
= bs
->file
? bs
->file
->bs
: NULL
;
5185 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_is_suspended
) {
5186 return bs
->drv
->bdrv_debug_is_suspended(bs
, tag
);
5192 /* backing_file can either be relative, or absolute, or a protocol. If it is
5193 * relative, it must be relative to the chain. So, passing in bs->filename
5194 * from a BDS as backing_file should not be done, as that may be relative to
5195 * the CWD rather than the chain. */
5196 BlockDriverState
*bdrv_find_backing_image(BlockDriverState
*bs
,
5197 const char *backing_file
)
5199 char *filename_full
= NULL
;
5200 char *backing_file_full
= NULL
;
5201 char *filename_tmp
= NULL
;
5202 int is_protocol
= 0;
5203 BlockDriverState
*curr_bs
= NULL
;
5204 BlockDriverState
*retval
= NULL
;
5206 if (!bs
|| !bs
->drv
|| !backing_file
) {
5210 filename_full
= g_malloc(PATH_MAX
);
5211 backing_file_full
= g_malloc(PATH_MAX
);
5213 is_protocol
= path_has_protocol(backing_file
);
5215 for (curr_bs
= bs
; curr_bs
->backing
; curr_bs
= curr_bs
->backing
->bs
) {
5217 /* If either of the filename paths is actually a protocol, then
5218 * compare unmodified paths; otherwise make paths relative */
5219 if (is_protocol
|| path_has_protocol(curr_bs
->backing_file
)) {
5220 char *backing_file_full_ret
;
5222 if (strcmp(backing_file
, curr_bs
->backing_file
) == 0) {
5223 retval
= curr_bs
->backing
->bs
;
5226 /* Also check against the full backing filename for the image */
5227 backing_file_full_ret
= bdrv_get_full_backing_filename(curr_bs
,
5229 if (backing_file_full_ret
) {
5230 bool equal
= strcmp(backing_file
, backing_file_full_ret
) == 0;
5231 g_free(backing_file_full_ret
);
5233 retval
= curr_bs
->backing
->bs
;
5238 /* If not an absolute filename path, make it relative to the current
5239 * image's filename path */
5240 filename_tmp
= bdrv_make_absolute_filename(curr_bs
, backing_file
,
5242 /* We are going to compare canonicalized absolute pathnames */
5243 if (!filename_tmp
|| !realpath(filename_tmp
, filename_full
)) {
5244 g_free(filename_tmp
);
5247 g_free(filename_tmp
);
5249 /* We need to make sure the backing filename we are comparing against
5250 * is relative to the current image filename (or absolute) */
5251 filename_tmp
= bdrv_get_full_backing_filename(curr_bs
, NULL
);
5252 if (!filename_tmp
|| !realpath(filename_tmp
, backing_file_full
)) {
5253 g_free(filename_tmp
);
5256 g_free(filename_tmp
);
5258 if (strcmp(backing_file_full
, filename_full
) == 0) {
5259 retval
= curr_bs
->backing
->bs
;
5265 g_free(filename_full
);
5266 g_free(backing_file_full
);
5270 void bdrv_init(void)
5272 module_call_init(MODULE_INIT_BLOCK
);
5275 void bdrv_init_with_whitelist(void)
5277 use_bdrv_whitelist
= 1;
5281 static void coroutine_fn
bdrv_co_invalidate_cache(BlockDriverState
*bs
,
5284 BdrvChild
*child
, *parent
;
5285 uint64_t perm
, shared_perm
;
5286 Error
*local_err
= NULL
;
5288 BdrvDirtyBitmap
*bm
;
5294 if (!(bs
->open_flags
& BDRV_O_INACTIVE
)) {
5298 QLIST_FOREACH(child
, &bs
->children
, next
) {
5299 bdrv_co_invalidate_cache(child
->bs
, &local_err
);
5301 error_propagate(errp
, local_err
);
5307 * Update permissions, they may differ for inactive nodes.
5309 * Note that the required permissions of inactive images are always a
5310 * subset of the permissions required after activating the image. This
5311 * allows us to just get the permissions upfront without restricting
5312 * drv->bdrv_invalidate_cache().
5314 * It also means that in error cases, we don't have to try and revert to
5315 * the old permissions (which is an operation that could fail, too). We can
5316 * just keep the extended permissions for the next time that an activation
5317 * of the image is tried.
5319 bs
->open_flags
&= ~BDRV_O_INACTIVE
;
5320 bdrv_get_cumulative_perm(bs
, &perm
, &shared_perm
);
5321 ret
= bdrv_check_perm(bs
, NULL
, perm
, shared_perm
, NULL
, NULL
, &local_err
);
5323 bs
->open_flags
|= BDRV_O_INACTIVE
;
5324 error_propagate(errp
, local_err
);
5327 bdrv_set_perm(bs
, perm
, shared_perm
);
5329 if (bs
->drv
->bdrv_co_invalidate_cache
) {
5330 bs
->drv
->bdrv_co_invalidate_cache(bs
, &local_err
);
5332 bs
->open_flags
|= BDRV_O_INACTIVE
;
5333 error_propagate(errp
, local_err
);
5338 for (bm
= bdrv_dirty_bitmap_next(bs
, NULL
); bm
;
5339 bm
= bdrv_dirty_bitmap_next(bs
, bm
))
5341 bdrv_dirty_bitmap_set_migration(bm
, false);
5344 ret
= refresh_total_sectors(bs
, bs
->total_sectors
);
5346 bs
->open_flags
|= BDRV_O_INACTIVE
;
5347 error_setg_errno(errp
, -ret
, "Could not refresh total sector count");
5351 QLIST_FOREACH(parent
, &bs
->parents
, next_parent
) {
5352 if (parent
->role
->activate
) {
5353 parent
->role
->activate(parent
, &local_err
);
5355 bs
->open_flags
|= BDRV_O_INACTIVE
;
5356 error_propagate(errp
, local_err
);
5363 typedef struct InvalidateCacheCo
{
5364 BlockDriverState
*bs
;
5367 } InvalidateCacheCo
;
5369 static void coroutine_fn
bdrv_invalidate_cache_co_entry(void *opaque
)
5371 InvalidateCacheCo
*ico
= opaque
;
5372 bdrv_co_invalidate_cache(ico
->bs
, ico
->errp
);
5377 void bdrv_invalidate_cache(BlockDriverState
*bs
, Error
**errp
)
5380 InvalidateCacheCo ico
= {
5386 if (qemu_in_coroutine()) {
5387 /* Fast-path if already in coroutine context */
5388 bdrv_invalidate_cache_co_entry(&ico
);
5390 co
= qemu_coroutine_create(bdrv_invalidate_cache_co_entry
, &ico
);
5391 bdrv_coroutine_enter(bs
, co
);
5392 BDRV_POLL_WHILE(bs
, !ico
.done
);
5396 void bdrv_invalidate_cache_all(Error
**errp
)
5398 BlockDriverState
*bs
;
5399 Error
*local_err
= NULL
;
5400 BdrvNextIterator it
;
5402 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
5403 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
5405 aio_context_acquire(aio_context
);
5406 bdrv_invalidate_cache(bs
, &local_err
);
5407 aio_context_release(aio_context
);
5409 error_propagate(errp
, local_err
);
5410 bdrv_next_cleanup(&it
);
5416 static bool bdrv_has_bds_parent(BlockDriverState
*bs
, bool only_active
)
5420 QLIST_FOREACH(parent
, &bs
->parents
, next_parent
) {
5421 if (parent
->role
->parent_is_bds
) {
5422 BlockDriverState
*parent_bs
= parent
->opaque
;
5423 if (!only_active
|| !(parent_bs
->open_flags
& BDRV_O_INACTIVE
)) {
5432 static int bdrv_inactivate_recurse(BlockDriverState
*bs
)
5434 BdrvChild
*child
, *parent
;
5435 bool tighten_restrictions
;
5436 uint64_t perm
, shared_perm
;
5443 /* Make sure that we don't inactivate a child before its parent.
5444 * It will be covered by recursion from the yet active parent. */
5445 if (bdrv_has_bds_parent(bs
, true)) {
5449 assert(!(bs
->open_flags
& BDRV_O_INACTIVE
));
5451 /* Inactivate this node */
5452 if (bs
->drv
->bdrv_inactivate
) {
5453 ret
= bs
->drv
->bdrv_inactivate(bs
);
5459 QLIST_FOREACH(parent
, &bs
->parents
, next_parent
) {
5460 if (parent
->role
->inactivate
) {
5461 ret
= parent
->role
->inactivate(parent
);
5468 bs
->open_flags
|= BDRV_O_INACTIVE
;
5470 /* Update permissions, they may differ for inactive nodes */
5471 bdrv_get_cumulative_perm(bs
, &perm
, &shared_perm
);
5472 ret
= bdrv_check_perm(bs
, NULL
, perm
, shared_perm
, NULL
,
5473 &tighten_restrictions
, NULL
);
5474 assert(tighten_restrictions
== false);
5476 /* We only tried to loosen restrictions, so errors are not fatal */
5477 bdrv_abort_perm_update(bs
);
5479 bdrv_set_perm(bs
, perm
, shared_perm
);
5483 /* Recursively inactivate children */
5484 QLIST_FOREACH(child
, &bs
->children
, next
) {
5485 ret
= bdrv_inactivate_recurse(child
->bs
);
5494 int bdrv_inactivate_all(void)
5496 BlockDriverState
*bs
= NULL
;
5497 BdrvNextIterator it
;
5499 GSList
*aio_ctxs
= NULL
, *ctx
;
5501 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
5502 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
5504 if (!g_slist_find(aio_ctxs
, aio_context
)) {
5505 aio_ctxs
= g_slist_prepend(aio_ctxs
, aio_context
);
5506 aio_context_acquire(aio_context
);
5510 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
5511 /* Nodes with BDS parents are covered by recursion from the last
5512 * parent that gets inactivated. Don't inactivate them a second
5513 * time if that has already happened. */
5514 if (bdrv_has_bds_parent(bs
, false)) {
5517 ret
= bdrv_inactivate_recurse(bs
);
5519 bdrv_next_cleanup(&it
);
5525 for (ctx
= aio_ctxs
; ctx
!= NULL
; ctx
= ctx
->next
) {
5526 AioContext
*aio_context
= ctx
->data
;
5527 aio_context_release(aio_context
);
5529 g_slist_free(aio_ctxs
);
5534 /**************************************************************/
5535 /* removable device support */
5538 * Return TRUE if the media is present
5540 bool bdrv_is_inserted(BlockDriverState
*bs
)
5542 BlockDriver
*drv
= bs
->drv
;
5548 if (drv
->bdrv_is_inserted
) {
5549 return drv
->bdrv_is_inserted(bs
);
5551 QLIST_FOREACH(child
, &bs
->children
, next
) {
5552 if (!bdrv_is_inserted(child
->bs
)) {
5560 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
5562 void bdrv_eject(BlockDriverState
*bs
, bool eject_flag
)
5564 BlockDriver
*drv
= bs
->drv
;
5566 if (drv
&& drv
->bdrv_eject
) {
5567 drv
->bdrv_eject(bs
, eject_flag
);
5572 * Lock or unlock the media (if it is locked, the user won't be able
5573 * to eject it manually).
5575 void bdrv_lock_medium(BlockDriverState
*bs
, bool locked
)
5577 BlockDriver
*drv
= bs
->drv
;
5579 trace_bdrv_lock_medium(bs
, locked
);
5581 if (drv
&& drv
->bdrv_lock_medium
) {
5582 drv
->bdrv_lock_medium(bs
, locked
);
5586 /* Get a reference to bs */
5587 void bdrv_ref(BlockDriverState
*bs
)
5592 /* Release a previously grabbed reference to bs.
5593 * If after releasing, reference count is zero, the BlockDriverState is
5595 void bdrv_unref(BlockDriverState
*bs
)
5600 assert(bs
->refcnt
> 0);
5601 if (--bs
->refcnt
== 0) {
5606 struct BdrvOpBlocker
{
5608 QLIST_ENTRY(BdrvOpBlocker
) list
;
5611 bool bdrv_op_is_blocked(BlockDriverState
*bs
, BlockOpType op
, Error
**errp
)
5613 BdrvOpBlocker
*blocker
;
5614 assert((int) op
>= 0 && op
< BLOCK_OP_TYPE_MAX
);
5615 if (!QLIST_EMPTY(&bs
->op_blockers
[op
])) {
5616 blocker
= QLIST_FIRST(&bs
->op_blockers
[op
]);
5617 error_propagate_prepend(errp
, error_copy(blocker
->reason
),
5618 "Node '%s' is busy: ",
5619 bdrv_get_device_or_node_name(bs
));
5625 void bdrv_op_block(BlockDriverState
*bs
, BlockOpType op
, Error
*reason
)
5627 BdrvOpBlocker
*blocker
;
5628 assert((int) op
>= 0 && op
< BLOCK_OP_TYPE_MAX
);
5630 blocker
= g_new0(BdrvOpBlocker
, 1);
5631 blocker
->reason
= reason
;
5632 QLIST_INSERT_HEAD(&bs
->op_blockers
[op
], blocker
, list
);
5635 void bdrv_op_unblock(BlockDriverState
*bs
, BlockOpType op
, Error
*reason
)
5637 BdrvOpBlocker
*blocker
, *next
;
5638 assert((int) op
>= 0 && op
< BLOCK_OP_TYPE_MAX
);
5639 QLIST_FOREACH_SAFE(blocker
, &bs
->op_blockers
[op
], list
, next
) {
5640 if (blocker
->reason
== reason
) {
5641 QLIST_REMOVE(blocker
, list
);
5647 void bdrv_op_block_all(BlockDriverState
*bs
, Error
*reason
)
5650 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
5651 bdrv_op_block(bs
, i
, reason
);
5655 void bdrv_op_unblock_all(BlockDriverState
*bs
, Error
*reason
)
5658 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
5659 bdrv_op_unblock(bs
, i
, reason
);
5663 bool bdrv_op_blocker_is_empty(BlockDriverState
*bs
)
5667 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
5668 if (!QLIST_EMPTY(&bs
->op_blockers
[i
])) {
5675 void bdrv_img_create(const char *filename
, const char *fmt
,
5676 const char *base_filename
, const char *base_fmt
,
5677 char *options
, uint64_t img_size
, int flags
, bool quiet
,
5680 QemuOptsList
*create_opts
= NULL
;
5681 QemuOpts
*opts
= NULL
;
5682 const char *backing_fmt
, *backing_file
;
5684 BlockDriver
*drv
, *proto_drv
;
5685 Error
*local_err
= NULL
;
5688 /* Find driver and parse its options */
5689 drv
= bdrv_find_format(fmt
);
5691 error_setg(errp
, "Unknown file format '%s'", fmt
);
5695 proto_drv
= bdrv_find_protocol(filename
, true, errp
);
5700 if (!drv
->create_opts
) {
5701 error_setg(errp
, "Format driver '%s' does not support image creation",
5706 if (!proto_drv
->create_opts
) {
5707 error_setg(errp
, "Protocol driver '%s' does not support image creation",
5708 proto_drv
->format_name
);
5712 create_opts
= qemu_opts_append(create_opts
, drv
->create_opts
);
5713 create_opts
= qemu_opts_append(create_opts
, proto_drv
->create_opts
);
5715 /* Create parameter list with default values */
5716 opts
= qemu_opts_create(create_opts
, NULL
, 0, &error_abort
);
5717 qemu_opt_set_number(opts
, BLOCK_OPT_SIZE
, img_size
, &error_abort
);
5719 /* Parse -o options */
5721 qemu_opts_do_parse(opts
, options
, NULL
, &local_err
);
5727 if (base_filename
) {
5728 qemu_opt_set(opts
, BLOCK_OPT_BACKING_FILE
, base_filename
, &local_err
);
5730 error_setg(errp
, "Backing file not supported for file format '%s'",
5737 qemu_opt_set(opts
, BLOCK_OPT_BACKING_FMT
, base_fmt
, &local_err
);
5739 error_setg(errp
, "Backing file format not supported for file "
5740 "format '%s'", fmt
);
5745 backing_file
= qemu_opt_get(opts
, BLOCK_OPT_BACKING_FILE
);
5747 if (!strcmp(filename
, backing_file
)) {
5748 error_setg(errp
, "Error: Trying to create an image with the "
5749 "same filename as the backing file");
5754 backing_fmt
= qemu_opt_get(opts
, BLOCK_OPT_BACKING_FMT
);
5756 /* The size for the image must always be specified, unless we have a backing
5757 * file and we have not been forbidden from opening it. */
5758 size
= qemu_opt_get_size(opts
, BLOCK_OPT_SIZE
, img_size
);
5759 if (backing_file
&& !(flags
& BDRV_O_NO_BACKING
)) {
5760 BlockDriverState
*bs
;
5763 QDict
*backing_options
= NULL
;
5766 bdrv_get_full_backing_filename_from_filename(filename
, backing_file
,
5771 assert(full_backing
);
5773 /* backing files always opened read-only */
5775 back_flags
&= ~(BDRV_O_RDWR
| BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
);
5777 backing_options
= qdict_new();
5779 qdict_put_str(backing_options
, "driver", backing_fmt
);
5781 qdict_put_bool(backing_options
, BDRV_OPT_FORCE_SHARE
, true);
5783 bs
= bdrv_open(full_backing
, NULL
, backing_options
, back_flags
,
5785 g_free(full_backing
);
5786 if (!bs
&& size
!= -1) {
5787 /* Couldn't open BS, but we have a size, so it's nonfatal */
5788 warn_reportf_err(local_err
,
5789 "Could not verify backing image. "
5790 "This may become an error in future versions.\n");
5793 /* Couldn't open bs, do not have size */
5794 error_append_hint(&local_err
,
5795 "Could not open backing image to determine size.\n");
5799 /* Opened BS, have no size */
5800 size
= bdrv_getlength(bs
);
5802 error_setg_errno(errp
, -size
, "Could not get size of '%s'",
5807 qemu_opt_set_number(opts
, BLOCK_OPT_SIZE
, size
, &error_abort
);
5811 } /* (backing_file && !(flags & BDRV_O_NO_BACKING)) */
5814 error_setg(errp
, "Image creation needs a size parameter");
5819 printf("Formatting '%s', fmt=%s ", filename
, fmt
);
5820 qemu_opts_print(opts
, " ");
5824 ret
= bdrv_create(drv
, filename
, opts
, &local_err
);
5826 if (ret
== -EFBIG
) {
5827 /* This is generally a better message than whatever the driver would
5828 * deliver (especially because of the cluster_size_hint), since that
5829 * is most probably not much different from "image too large". */
5830 const char *cluster_size_hint
= "";
5831 if (qemu_opt_get_size(opts
, BLOCK_OPT_CLUSTER_SIZE
, 0)) {
5832 cluster_size_hint
= " (try using a larger cluster size)";
5834 error_setg(errp
, "The image size is too large for file format '%s'"
5835 "%s", fmt
, cluster_size_hint
);
5836 error_free(local_err
);
5841 qemu_opts_del(opts
);
5842 qemu_opts_free(create_opts
);
5843 error_propagate(errp
, local_err
);
5846 AioContext
*bdrv_get_aio_context(BlockDriverState
*bs
)
5848 return bs
? bs
->aio_context
: qemu_get_aio_context();
5851 void bdrv_coroutine_enter(BlockDriverState
*bs
, Coroutine
*co
)
5853 aio_co_enter(bdrv_get_aio_context(bs
), co
);
5856 static void bdrv_do_remove_aio_context_notifier(BdrvAioNotifier
*ban
)
5858 QLIST_REMOVE(ban
, list
);
5862 static void bdrv_detach_aio_context(BlockDriverState
*bs
)
5864 BdrvAioNotifier
*baf
, *baf_tmp
;
5866 assert(!bs
->walking_aio_notifiers
);
5867 bs
->walking_aio_notifiers
= true;
5868 QLIST_FOREACH_SAFE(baf
, &bs
->aio_notifiers
, list
, baf_tmp
) {
5870 bdrv_do_remove_aio_context_notifier(baf
);
5872 baf
->detach_aio_context(baf
->opaque
);
5875 /* Never mind iterating again to check for ->deleted. bdrv_close() will
5876 * remove remaining aio notifiers if we aren't called again.
5878 bs
->walking_aio_notifiers
= false;
5880 if (bs
->drv
&& bs
->drv
->bdrv_detach_aio_context
) {
5881 bs
->drv
->bdrv_detach_aio_context(bs
);
5884 if (bs
->quiesce_counter
) {
5885 aio_enable_external(bs
->aio_context
);
5887 bs
->aio_context
= NULL
;
5890 static void bdrv_attach_aio_context(BlockDriverState
*bs
,
5891 AioContext
*new_context
)
5893 BdrvAioNotifier
*ban
, *ban_tmp
;
5895 if (bs
->quiesce_counter
) {
5896 aio_disable_external(new_context
);
5899 bs
->aio_context
= new_context
;
5901 if (bs
->drv
&& bs
->drv
->bdrv_attach_aio_context
) {
5902 bs
->drv
->bdrv_attach_aio_context(bs
, new_context
);
5905 assert(!bs
->walking_aio_notifiers
);
5906 bs
->walking_aio_notifiers
= true;
5907 QLIST_FOREACH_SAFE(ban
, &bs
->aio_notifiers
, list
, ban_tmp
) {
5909 bdrv_do_remove_aio_context_notifier(ban
);
5911 ban
->attached_aio_context(new_context
, ban
->opaque
);
5914 bs
->walking_aio_notifiers
= false;
5918 * Changes the AioContext used for fd handlers, timers, and BHs by this
5919 * BlockDriverState and all its children and parents.
5921 * The caller must own the AioContext lock for the old AioContext of bs, but it
5922 * must not own the AioContext lock for new_context (unless new_context is the
5923 * same as the current context of bs).
5925 * @ignore will accumulate all visited BdrvChild object. The caller is
5926 * responsible for freeing the list afterwards.
5928 void bdrv_set_aio_context_ignore(BlockDriverState
*bs
,
5929 AioContext
*new_context
, GSList
**ignore
)
5933 if (bdrv_get_aio_context(bs
) == new_context
) {
5937 bdrv_drained_begin(bs
);
5939 QLIST_FOREACH(child
, &bs
->children
, next
) {
5940 if (g_slist_find(*ignore
, child
)) {
5943 *ignore
= g_slist_prepend(*ignore
, child
);
5944 bdrv_set_aio_context_ignore(child
->bs
, new_context
, ignore
);
5946 QLIST_FOREACH(child
, &bs
->parents
, next_parent
) {
5947 if (g_slist_find(*ignore
, child
)) {
5950 assert(child
->role
->set_aio_ctx
);
5951 *ignore
= g_slist_prepend(*ignore
, child
);
5952 child
->role
->set_aio_ctx(child
, new_context
, ignore
);
5955 bdrv_detach_aio_context(bs
);
5957 /* This function executes in the old AioContext so acquire the new one in
5958 * case it runs in a different thread.
5960 aio_context_acquire(new_context
);
5961 bdrv_attach_aio_context(bs
, new_context
);
5962 bdrv_drained_end(bs
);
5963 aio_context_release(new_context
);
5966 static bool bdrv_parent_can_set_aio_context(BdrvChild
*c
, AioContext
*ctx
,
5967 GSList
**ignore
, Error
**errp
)
5969 if (g_slist_find(*ignore
, c
)) {
5972 *ignore
= g_slist_prepend(*ignore
, c
);
5974 /* A BdrvChildRole that doesn't handle AioContext changes cannot
5975 * tolerate any AioContext changes */
5976 if (!c
->role
->can_set_aio_ctx
) {
5977 char *user
= bdrv_child_user_desc(c
);
5978 error_setg(errp
, "Changing iothreads is not supported by %s", user
);
5982 if (!c
->role
->can_set_aio_ctx(c
, ctx
, ignore
, errp
)) {
5983 assert(!errp
|| *errp
);
5989 bool bdrv_child_can_set_aio_context(BdrvChild
*c
, AioContext
*ctx
,
5990 GSList
**ignore
, Error
**errp
)
5992 if (g_slist_find(*ignore
, c
)) {
5995 *ignore
= g_slist_prepend(*ignore
, c
);
5996 return bdrv_can_set_aio_context(c
->bs
, ctx
, ignore
, errp
);
5999 /* @ignore will accumulate all visited BdrvChild object. The caller is
6000 * responsible for freeing the list afterwards. */
6001 bool bdrv_can_set_aio_context(BlockDriverState
*bs
, AioContext
*ctx
,
6002 GSList
**ignore
, Error
**errp
)
6006 if (bdrv_get_aio_context(bs
) == ctx
) {
6010 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
6011 if (!bdrv_parent_can_set_aio_context(c
, ctx
, ignore
, errp
)) {
6015 QLIST_FOREACH(c
, &bs
->children
, next
) {
6016 if (!bdrv_child_can_set_aio_context(c
, ctx
, ignore
, errp
)) {
6024 int bdrv_child_try_set_aio_context(BlockDriverState
*bs
, AioContext
*ctx
,
6025 BdrvChild
*ignore_child
, Error
**errp
)
6030 ignore
= ignore_child
? g_slist_prepend(NULL
, ignore_child
) : NULL
;
6031 ret
= bdrv_can_set_aio_context(bs
, ctx
, &ignore
, errp
);
6032 g_slist_free(ignore
);
6038 ignore
= ignore_child
? g_slist_prepend(NULL
, ignore_child
) : NULL
;
6039 bdrv_set_aio_context_ignore(bs
, ctx
, &ignore
);
6040 g_slist_free(ignore
);
6045 int bdrv_try_set_aio_context(BlockDriverState
*bs
, AioContext
*ctx
,
6048 return bdrv_child_try_set_aio_context(bs
, ctx
, NULL
, errp
);
6051 void bdrv_add_aio_context_notifier(BlockDriverState
*bs
,
6052 void (*attached_aio_context
)(AioContext
*new_context
, void *opaque
),
6053 void (*detach_aio_context
)(void *opaque
), void *opaque
)
6055 BdrvAioNotifier
*ban
= g_new(BdrvAioNotifier
, 1);
6056 *ban
= (BdrvAioNotifier
){
6057 .attached_aio_context
= attached_aio_context
,
6058 .detach_aio_context
= detach_aio_context
,
6062 QLIST_INSERT_HEAD(&bs
->aio_notifiers
, ban
, list
);
6065 void bdrv_remove_aio_context_notifier(BlockDriverState
*bs
,
6066 void (*attached_aio_context
)(AioContext
*,
6068 void (*detach_aio_context
)(void *),
6071 BdrvAioNotifier
*ban
, *ban_next
;
6073 QLIST_FOREACH_SAFE(ban
, &bs
->aio_notifiers
, list
, ban_next
) {
6074 if (ban
->attached_aio_context
== attached_aio_context
&&
6075 ban
->detach_aio_context
== detach_aio_context
&&
6076 ban
->opaque
== opaque
&&
6077 ban
->deleted
== false)
6079 if (bs
->walking_aio_notifiers
) {
6080 ban
->deleted
= true;
6082 bdrv_do_remove_aio_context_notifier(ban
);
6091 int bdrv_amend_options(BlockDriverState
*bs
, QemuOpts
*opts
,
6092 BlockDriverAmendStatusCB
*status_cb
, void *cb_opaque
,
6096 error_setg(errp
, "Node is ejected");
6099 if (!bs
->drv
->bdrv_amend_options
) {
6100 error_setg(errp
, "Block driver '%s' does not support option amendment",
6101 bs
->drv
->format_name
);
6104 return bs
->drv
->bdrv_amend_options(bs
, opts
, status_cb
, cb_opaque
, errp
);
6107 /* This function will be called by the bdrv_recurse_is_first_non_filter method
6108 * of block filter and by bdrv_is_first_non_filter.
6109 * It is used to test if the given bs is the candidate or recurse more in the
6112 bool bdrv_recurse_is_first_non_filter(BlockDriverState
*bs
,
6113 BlockDriverState
*candidate
)
6115 /* return false if basic checks fails */
6116 if (!bs
|| !bs
->drv
) {
6120 /* the code reached a non block filter driver -> check if the bs is
6121 * the same as the candidate. It's the recursion termination condition.
6123 if (!bs
->drv
->is_filter
) {
6124 return bs
== candidate
;
6126 /* Down this path the driver is a block filter driver */
6128 /* If the block filter recursion method is defined use it to recurse down
6131 if (bs
->drv
->bdrv_recurse_is_first_non_filter
) {
6132 return bs
->drv
->bdrv_recurse_is_first_non_filter(bs
, candidate
);
6135 /* the driver is a block filter but don't allow to recurse -> return false
6140 /* This function checks if the candidate is the first non filter bs down it's
6141 * bs chain. Since we don't have pointers to parents it explore all bs chains
6142 * from the top. Some filters can choose not to pass down the recursion.
6144 bool bdrv_is_first_non_filter(BlockDriverState
*candidate
)
6146 BlockDriverState
*bs
;
6147 BdrvNextIterator it
;
6149 /* walk down the bs forest recursively */
6150 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
6153 /* try to recurse in this top level bs */
6154 perm
= bdrv_recurse_is_first_non_filter(bs
, candidate
);
6156 /* candidate is the first non filter */
6158 bdrv_next_cleanup(&it
);
6166 BlockDriverState
*check_to_replace_node(BlockDriverState
*parent_bs
,
6167 const char *node_name
, Error
**errp
)
6169 BlockDriverState
*to_replace_bs
= bdrv_find_node(node_name
);
6170 AioContext
*aio_context
;
6172 if (!to_replace_bs
) {
6173 error_setg(errp
, "Node name '%s' not found", node_name
);
6177 aio_context
= bdrv_get_aio_context(to_replace_bs
);
6178 aio_context_acquire(aio_context
);
6180 if (bdrv_op_is_blocked(to_replace_bs
, BLOCK_OP_TYPE_REPLACE
, errp
)) {
6181 to_replace_bs
= NULL
;
6185 /* We don't want arbitrary node of the BDS chain to be replaced only the top
6186 * most non filter in order to prevent data corruption.
6187 * Another benefit is that this tests exclude backing files which are
6188 * blocked by the backing blockers.
6190 if (!bdrv_recurse_is_first_non_filter(parent_bs
, to_replace_bs
)) {
6191 error_setg(errp
, "Only top most non filter can be replaced");
6192 to_replace_bs
= NULL
;
6197 aio_context_release(aio_context
);
6198 return to_replace_bs
;
6202 * Iterates through the list of runtime option keys that are said to
6203 * be "strong" for a BDS. An option is called "strong" if it changes
6204 * a BDS's data. For example, the null block driver's "size" and
6205 * "read-zeroes" options are strong, but its "latency-ns" option is
6208 * If a key returned by this function ends with a dot, all options
6209 * starting with that prefix are strong.
6211 static const char *const *strong_options(BlockDriverState
*bs
,
6212 const char *const *curopt
)
6214 static const char *const global_options
[] = {
6215 "driver", "filename", NULL
6219 return &global_options
[0];
6223 if (curopt
== &global_options
[ARRAY_SIZE(global_options
) - 1] && bs
->drv
) {
6224 curopt
= bs
->drv
->strong_runtime_opts
;
6227 return (curopt
&& *curopt
) ? curopt
: NULL
;
6231 * Copies all strong runtime options from bs->options to the given
6232 * QDict. The set of strong option keys is determined by invoking
6235 * Returns true iff any strong option was present in bs->options (and
6236 * thus copied to the target QDict) with the exception of "filename"
6237 * and "driver". The caller is expected to use this value to decide
6238 * whether the existence of strong options prevents the generation of
6241 static bool append_strong_runtime_options(QDict
*d
, BlockDriverState
*bs
)
6243 bool found_any
= false;
6244 const char *const *option_name
= NULL
;
6250 while ((option_name
= strong_options(bs
, option_name
))) {
6251 bool option_given
= false;
6253 assert(strlen(*option_name
) > 0);
6254 if ((*option_name
)[strlen(*option_name
) - 1] != '.') {
6255 QObject
*entry
= qdict_get(bs
->options
, *option_name
);
6260 qdict_put_obj(d
, *option_name
, qobject_ref(entry
));
6261 option_given
= true;
6263 const QDictEntry
*entry
;
6264 for (entry
= qdict_first(bs
->options
); entry
;
6265 entry
= qdict_next(bs
->options
, entry
))
6267 if (strstart(qdict_entry_key(entry
), *option_name
, NULL
)) {
6268 qdict_put_obj(d
, qdict_entry_key(entry
),
6269 qobject_ref(qdict_entry_value(entry
)));
6270 option_given
= true;
6275 /* While "driver" and "filename" need to be included in a JSON filename,
6276 * their existence does not prohibit generation of a plain filename. */
6277 if (!found_any
&& option_given
&&
6278 strcmp(*option_name
, "driver") && strcmp(*option_name
, "filename"))
6284 if (!qdict_haskey(d
, "driver")) {
6285 /* Drivers created with bdrv_new_open_driver() may not have a
6286 * @driver option. Add it here. */
6287 qdict_put_str(d
, "driver", bs
->drv
->format_name
);
6293 /* Note: This function may return false positives; it may return true
6294 * even if opening the backing file specified by bs's image header
6295 * would result in exactly bs->backing. */
6296 static bool bdrv_backing_overridden(BlockDriverState
*bs
)
6299 return strcmp(bs
->auto_backing_file
,
6300 bs
->backing
->bs
->filename
);
6302 /* No backing BDS, so if the image header reports any backing
6303 * file, it must have been suppressed */
6304 return bs
->auto_backing_file
[0] != '\0';
6308 /* Updates the following BDS fields:
6309 * - exact_filename: A filename which may be used for opening a block device
6310 * which (mostly) equals the given BDS (even without any
6311 * other options; so reading and writing must return the same
6312 * results, but caching etc. may be different)
6313 * - full_open_options: Options which, when given when opening a block device
6314 * (without a filename), result in a BDS (mostly)
6315 * equalling the given one
6316 * - filename: If exact_filename is set, it is copied here. Otherwise,
6317 * full_open_options is converted to a JSON object, prefixed with
6318 * "json:" (for use through the JSON pseudo protocol) and put here.
6320 void bdrv_refresh_filename(BlockDriverState
*bs
)
6322 BlockDriver
*drv
= bs
->drv
;
6325 bool backing_overridden
;
6326 bool generate_json_filename
; /* Whether our default implementation should
6327 fill exact_filename (false) or not (true) */
6333 /* This BDS's file name may depend on any of its children's file names, so
6334 * refresh those first */
6335 QLIST_FOREACH(child
, &bs
->children
, next
) {
6336 bdrv_refresh_filename(child
->bs
);
6340 /* For implicit nodes, just copy everything from the single child */
6341 child
= QLIST_FIRST(&bs
->children
);
6342 assert(QLIST_NEXT(child
, next
) == NULL
);
6344 pstrcpy(bs
->exact_filename
, sizeof(bs
->exact_filename
),
6345 child
->bs
->exact_filename
);
6346 pstrcpy(bs
->filename
, sizeof(bs
->filename
), child
->bs
->filename
);
6348 bs
->full_open_options
= qobject_ref(child
->bs
->full_open_options
);
6353 backing_overridden
= bdrv_backing_overridden(bs
);
6355 if (bs
->open_flags
& BDRV_O_NO_IO
) {
6356 /* Without I/O, the backing file does not change anything.
6357 * Therefore, in such a case (primarily qemu-img), we can
6358 * pretend the backing file has not been overridden even if
6359 * it technically has been. */
6360 backing_overridden
= false;
6363 /* Gather the options QDict */
6365 generate_json_filename
= append_strong_runtime_options(opts
, bs
);
6366 generate_json_filename
|= backing_overridden
;
6368 if (drv
->bdrv_gather_child_options
) {
6369 /* Some block drivers may not want to present all of their children's
6370 * options, or name them differently from BdrvChild.name */
6371 drv
->bdrv_gather_child_options(bs
, opts
, backing_overridden
);
6373 QLIST_FOREACH(child
, &bs
->children
, next
) {
6374 if (child
->role
== &child_backing
&& !backing_overridden
) {
6375 /* We can skip the backing BDS if it has not been overridden */
6379 qdict_put(opts
, child
->name
,
6380 qobject_ref(child
->bs
->full_open_options
));
6383 if (backing_overridden
&& !bs
->backing
) {
6384 /* Force no backing file */
6385 qdict_put_null(opts
, "backing");
6389 qobject_unref(bs
->full_open_options
);
6390 bs
->full_open_options
= opts
;
6392 if (drv
->bdrv_refresh_filename
) {
6393 /* Obsolete information is of no use here, so drop the old file name
6394 * information before refreshing it */
6395 bs
->exact_filename
[0] = '\0';
6397 drv
->bdrv_refresh_filename(bs
);
6398 } else if (bs
->file
) {
6399 /* Try to reconstruct valid information from the underlying file */
6401 bs
->exact_filename
[0] = '\0';
6404 * We can use the underlying file's filename if:
6405 * - it has a filename,
6406 * - the file is a protocol BDS, and
6407 * - opening that file (as this BDS's format) will automatically create
6408 * the BDS tree we have right now, that is:
6409 * - the user did not significantly change this BDS's behavior with
6410 * some explicit (strong) options
6411 * - no non-file child of this BDS has been overridden by the user
6412 * Both of these conditions are represented by generate_json_filename.
6414 if (bs
->file
->bs
->exact_filename
[0] &&
6415 bs
->file
->bs
->drv
->bdrv_file_open
&&
6416 !generate_json_filename
)
6418 strcpy(bs
->exact_filename
, bs
->file
->bs
->exact_filename
);
6422 if (bs
->exact_filename
[0]) {
6423 pstrcpy(bs
->filename
, sizeof(bs
->filename
), bs
->exact_filename
);
6425 QString
*json
= qobject_to_json(QOBJECT(bs
->full_open_options
));
6426 snprintf(bs
->filename
, sizeof(bs
->filename
), "json:%s",
6427 qstring_get_str(json
));
6428 qobject_unref(json
);
6432 char *bdrv_dirname(BlockDriverState
*bs
, Error
**errp
)
6434 BlockDriver
*drv
= bs
->drv
;
6437 error_setg(errp
, "Node '%s' is ejected", bs
->node_name
);
6441 if (drv
->bdrv_dirname
) {
6442 return drv
->bdrv_dirname(bs
, errp
);
6446 return bdrv_dirname(bs
->file
->bs
, errp
);
6449 bdrv_refresh_filename(bs
);
6450 if (bs
->exact_filename
[0] != '\0') {
6451 return path_combine(bs
->exact_filename
, "");
6454 error_setg(errp
, "Cannot generate a base directory for %s nodes",
6460 * Hot add/remove a BDS's child. So the user can take a child offline when
6461 * it is broken and take a new child online
6463 void bdrv_add_child(BlockDriverState
*parent_bs
, BlockDriverState
*child_bs
,
6467 if (!parent_bs
->drv
|| !parent_bs
->drv
->bdrv_add_child
) {
6468 error_setg(errp
, "The node %s does not support adding a child",
6469 bdrv_get_device_or_node_name(parent_bs
));
6473 if (!QLIST_EMPTY(&child_bs
->parents
)) {
6474 error_setg(errp
, "The node %s already has a parent",
6475 child_bs
->node_name
);
6479 parent_bs
->drv
->bdrv_add_child(parent_bs
, child_bs
, errp
);
6482 void bdrv_del_child(BlockDriverState
*parent_bs
, BdrvChild
*child
, Error
**errp
)
6486 if (!parent_bs
->drv
|| !parent_bs
->drv
->bdrv_del_child
) {
6487 error_setg(errp
, "The node %s does not support removing a child",
6488 bdrv_get_device_or_node_name(parent_bs
));
6492 QLIST_FOREACH(tmp
, &parent_bs
->children
, next
) {
6499 error_setg(errp
, "The node %s does not have a child named %s",
6500 bdrv_get_device_or_node_name(parent_bs
),
6501 bdrv_get_device_or_node_name(child
->bs
));
6505 parent_bs
->drv
->bdrv_del_child(parent_bs
, child
, errp
);
6508 bool bdrv_can_store_new_dirty_bitmap(BlockDriverState
*bs
, const char *name
,
6509 uint32_t granularity
, Error
**errp
)
6511 BlockDriver
*drv
= bs
->drv
;
6514 error_setg_errno(errp
, ENOMEDIUM
,
6515 "Can't store persistent bitmaps to %s",
6516 bdrv_get_device_or_node_name(bs
));
6520 if (!drv
->bdrv_can_store_new_dirty_bitmap
) {
6521 error_setg_errno(errp
, ENOTSUP
,
6522 "Can't store persistent bitmaps to %s",
6523 bdrv_get_device_or_node_name(bs
));
6527 return drv
->bdrv_can_store_new_dirty_bitmap(bs
, name
, granularity
, errp
);