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
,
915 int *drained_end_counter
)
917 BlockDriverState
*bs
= child
->opaque
;
918 bdrv_drained_end_no_poll(bs
, drained_end_counter
);
921 static void bdrv_child_cb_attach(BdrvChild
*child
)
923 BlockDriverState
*bs
= child
->opaque
;
924 bdrv_apply_subtree_drain(child
, bs
);
927 static void bdrv_child_cb_detach(BdrvChild
*child
)
929 BlockDriverState
*bs
= child
->opaque
;
930 bdrv_unapply_subtree_drain(child
, bs
);
933 static int bdrv_child_cb_inactivate(BdrvChild
*child
)
935 BlockDriverState
*bs
= child
->opaque
;
936 assert(bs
->open_flags
& BDRV_O_INACTIVE
);
940 static bool bdrv_child_cb_can_set_aio_ctx(BdrvChild
*child
, AioContext
*ctx
,
941 GSList
**ignore
, Error
**errp
)
943 BlockDriverState
*bs
= child
->opaque
;
944 return bdrv_can_set_aio_context(bs
, ctx
, ignore
, errp
);
947 static void bdrv_child_cb_set_aio_ctx(BdrvChild
*child
, AioContext
*ctx
,
950 BlockDriverState
*bs
= child
->opaque
;
951 return bdrv_set_aio_context_ignore(bs
, ctx
, ignore
);
955 * Returns the options and flags that a temporary snapshot should get, based on
956 * the originally requested flags (the originally requested image will have
957 * flags like a backing file)
959 static void bdrv_temp_snapshot_options(int *child_flags
, QDict
*child_options
,
960 int parent_flags
, QDict
*parent_options
)
962 *child_flags
= (parent_flags
& ~BDRV_O_SNAPSHOT
) | BDRV_O_TEMPORARY
;
964 /* For temporary files, unconditional cache=unsafe is fine */
965 qdict_set_default_str(child_options
, BDRV_OPT_CACHE_DIRECT
, "off");
966 qdict_set_default_str(child_options
, BDRV_OPT_CACHE_NO_FLUSH
, "on");
968 /* Copy the read-only and discard options from the parent */
969 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_READ_ONLY
);
970 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_DISCARD
);
972 /* aio=native doesn't work for cache.direct=off, so disable it for the
973 * temporary snapshot */
974 *child_flags
&= ~BDRV_O_NATIVE_AIO
;
978 * Returns the options and flags that bs->file should get if a protocol driver
979 * is expected, based on the given options and flags for the parent BDS
981 static void bdrv_inherited_options(int *child_flags
, QDict
*child_options
,
982 int parent_flags
, QDict
*parent_options
)
984 int flags
= parent_flags
;
986 /* Enable protocol handling, disable format probing for bs->file */
987 flags
|= BDRV_O_PROTOCOL
;
989 /* If the cache mode isn't explicitly set, inherit direct and no-flush from
991 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_CACHE_DIRECT
);
992 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_CACHE_NO_FLUSH
);
993 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_FORCE_SHARE
);
995 /* Inherit the read-only option from the parent if it's not set */
996 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_READ_ONLY
);
997 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_AUTO_READ_ONLY
);
999 /* Our block drivers take care to send flushes and respect unmap policy,
1000 * so we can default to enable both on lower layers regardless of the
1001 * corresponding parent options. */
1002 qdict_set_default_str(child_options
, BDRV_OPT_DISCARD
, "unmap");
1004 /* Clear flags that only apply to the top layer */
1005 flags
&= ~(BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
| BDRV_O_COPY_ON_READ
|
1008 *child_flags
= flags
;
1011 const BdrvChildRole child_file
= {
1012 .parent_is_bds
= true,
1013 .get_parent_desc
= bdrv_child_get_parent_desc
,
1014 .inherit_options
= bdrv_inherited_options
,
1015 .drained_begin
= bdrv_child_cb_drained_begin
,
1016 .drained_poll
= bdrv_child_cb_drained_poll
,
1017 .drained_end
= bdrv_child_cb_drained_end
,
1018 .attach
= bdrv_child_cb_attach
,
1019 .detach
= bdrv_child_cb_detach
,
1020 .inactivate
= bdrv_child_cb_inactivate
,
1021 .can_set_aio_ctx
= bdrv_child_cb_can_set_aio_ctx
,
1022 .set_aio_ctx
= bdrv_child_cb_set_aio_ctx
,
1026 * Returns the options and flags that bs->file should get if the use of formats
1027 * (and not only protocols) is permitted for it, based on the given options and
1028 * flags for the parent BDS
1030 static void bdrv_inherited_fmt_options(int *child_flags
, QDict
*child_options
,
1031 int parent_flags
, QDict
*parent_options
)
1033 child_file
.inherit_options(child_flags
, child_options
,
1034 parent_flags
, parent_options
);
1036 *child_flags
&= ~(BDRV_O_PROTOCOL
| BDRV_O_NO_IO
);
1039 const BdrvChildRole child_format
= {
1040 .parent_is_bds
= true,
1041 .get_parent_desc
= bdrv_child_get_parent_desc
,
1042 .inherit_options
= bdrv_inherited_fmt_options
,
1043 .drained_begin
= bdrv_child_cb_drained_begin
,
1044 .drained_poll
= bdrv_child_cb_drained_poll
,
1045 .drained_end
= bdrv_child_cb_drained_end
,
1046 .attach
= bdrv_child_cb_attach
,
1047 .detach
= bdrv_child_cb_detach
,
1048 .inactivate
= bdrv_child_cb_inactivate
,
1049 .can_set_aio_ctx
= bdrv_child_cb_can_set_aio_ctx
,
1050 .set_aio_ctx
= bdrv_child_cb_set_aio_ctx
,
1053 static void bdrv_backing_attach(BdrvChild
*c
)
1055 BlockDriverState
*parent
= c
->opaque
;
1056 BlockDriverState
*backing_hd
= c
->bs
;
1058 assert(!parent
->backing_blocker
);
1059 error_setg(&parent
->backing_blocker
,
1060 "node is used as backing hd of '%s'",
1061 bdrv_get_device_or_node_name(parent
));
1063 bdrv_refresh_filename(backing_hd
);
1065 parent
->open_flags
&= ~BDRV_O_NO_BACKING
;
1066 pstrcpy(parent
->backing_file
, sizeof(parent
->backing_file
),
1067 backing_hd
->filename
);
1068 pstrcpy(parent
->backing_format
, sizeof(parent
->backing_format
),
1069 backing_hd
->drv
? backing_hd
->drv
->format_name
: "");
1071 bdrv_op_block_all(backing_hd
, parent
->backing_blocker
);
1072 /* Otherwise we won't be able to commit or stream */
1073 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_COMMIT_TARGET
,
1074 parent
->backing_blocker
);
1075 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_STREAM
,
1076 parent
->backing_blocker
);
1078 * We do backup in 3 ways:
1080 * The target bs is new opened, and the source is top BDS
1081 * 2. blockdev backup
1082 * Both the source and the target are top BDSes.
1083 * 3. internal backup(used for block replication)
1084 * Both the source and the target are backing file
1086 * In case 1 and 2, neither the source nor the target is the backing file.
1087 * In case 3, we will block the top BDS, so there is only one block job
1088 * for the top BDS and its backing chain.
1090 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_BACKUP_SOURCE
,
1091 parent
->backing_blocker
);
1092 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_BACKUP_TARGET
,
1093 parent
->backing_blocker
);
1095 bdrv_child_cb_attach(c
);
1098 static void bdrv_backing_detach(BdrvChild
*c
)
1100 BlockDriverState
*parent
= c
->opaque
;
1102 assert(parent
->backing_blocker
);
1103 bdrv_op_unblock_all(c
->bs
, parent
->backing_blocker
);
1104 error_free(parent
->backing_blocker
);
1105 parent
->backing_blocker
= NULL
;
1107 bdrv_child_cb_detach(c
);
1111 * Returns the options and flags that bs->backing should get, based on the
1112 * given options and flags for the parent BDS
1114 static void bdrv_backing_options(int *child_flags
, QDict
*child_options
,
1115 int parent_flags
, QDict
*parent_options
)
1117 int flags
= parent_flags
;
1119 /* The cache mode is inherited unmodified for backing files; except WCE,
1120 * which is only applied on the top level (BlockBackend) */
1121 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_CACHE_DIRECT
);
1122 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_CACHE_NO_FLUSH
);
1123 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_FORCE_SHARE
);
1125 /* backing files always opened read-only */
1126 qdict_set_default_str(child_options
, BDRV_OPT_READ_ONLY
, "on");
1127 qdict_set_default_str(child_options
, BDRV_OPT_AUTO_READ_ONLY
, "off");
1128 flags
&= ~BDRV_O_COPY_ON_READ
;
1130 /* snapshot=on is handled on the top layer */
1131 flags
&= ~(BDRV_O_SNAPSHOT
| BDRV_O_TEMPORARY
);
1133 *child_flags
= flags
;
1136 static int bdrv_backing_update_filename(BdrvChild
*c
, BlockDriverState
*base
,
1137 const char *filename
, Error
**errp
)
1139 BlockDriverState
*parent
= c
->opaque
;
1140 bool read_only
= bdrv_is_read_only(parent
);
1144 ret
= bdrv_reopen_set_read_only(parent
, false, errp
);
1150 ret
= bdrv_change_backing_file(parent
, filename
,
1151 base
->drv
? base
->drv
->format_name
: "");
1153 error_setg_errno(errp
, -ret
, "Could not update backing file link");
1157 bdrv_reopen_set_read_only(parent
, true, NULL
);
1163 const BdrvChildRole child_backing
= {
1164 .parent_is_bds
= true,
1165 .get_parent_desc
= bdrv_child_get_parent_desc
,
1166 .attach
= bdrv_backing_attach
,
1167 .detach
= bdrv_backing_detach
,
1168 .inherit_options
= bdrv_backing_options
,
1169 .drained_begin
= bdrv_child_cb_drained_begin
,
1170 .drained_poll
= bdrv_child_cb_drained_poll
,
1171 .drained_end
= bdrv_child_cb_drained_end
,
1172 .inactivate
= bdrv_child_cb_inactivate
,
1173 .update_filename
= bdrv_backing_update_filename
,
1174 .can_set_aio_ctx
= bdrv_child_cb_can_set_aio_ctx
,
1175 .set_aio_ctx
= bdrv_child_cb_set_aio_ctx
,
1178 static int bdrv_open_flags(BlockDriverState
*bs
, int flags
)
1180 int open_flags
= flags
;
1183 * Clear flags that are internal to the block layer before opening the
1186 open_flags
&= ~(BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
| BDRV_O_PROTOCOL
);
1191 static void update_flags_from_options(int *flags
, QemuOpts
*opts
)
1193 *flags
&= ~(BDRV_O_CACHE_MASK
| BDRV_O_RDWR
| BDRV_O_AUTO_RDONLY
);
1195 if (qemu_opt_get_bool_del(opts
, BDRV_OPT_CACHE_NO_FLUSH
, false)) {
1196 *flags
|= BDRV_O_NO_FLUSH
;
1199 if (qemu_opt_get_bool_del(opts
, BDRV_OPT_CACHE_DIRECT
, false)) {
1200 *flags
|= BDRV_O_NOCACHE
;
1203 if (!qemu_opt_get_bool_del(opts
, BDRV_OPT_READ_ONLY
, false)) {
1204 *flags
|= BDRV_O_RDWR
;
1207 if (qemu_opt_get_bool_del(opts
, BDRV_OPT_AUTO_READ_ONLY
, false)) {
1208 *flags
|= BDRV_O_AUTO_RDONLY
;
1212 static void update_options_from_flags(QDict
*options
, int flags
)
1214 if (!qdict_haskey(options
, BDRV_OPT_CACHE_DIRECT
)) {
1215 qdict_put_bool(options
, BDRV_OPT_CACHE_DIRECT
, flags
& BDRV_O_NOCACHE
);
1217 if (!qdict_haskey(options
, BDRV_OPT_CACHE_NO_FLUSH
)) {
1218 qdict_put_bool(options
, BDRV_OPT_CACHE_NO_FLUSH
,
1219 flags
& BDRV_O_NO_FLUSH
);
1221 if (!qdict_haskey(options
, BDRV_OPT_READ_ONLY
)) {
1222 qdict_put_bool(options
, BDRV_OPT_READ_ONLY
, !(flags
& BDRV_O_RDWR
));
1224 if (!qdict_haskey(options
, BDRV_OPT_AUTO_READ_ONLY
)) {
1225 qdict_put_bool(options
, BDRV_OPT_AUTO_READ_ONLY
,
1226 flags
& BDRV_O_AUTO_RDONLY
);
1230 static void bdrv_assign_node_name(BlockDriverState
*bs
,
1231 const char *node_name
,
1234 char *gen_node_name
= NULL
;
1237 node_name
= gen_node_name
= id_generate(ID_BLOCK
);
1238 } else if (!id_wellformed(node_name
)) {
1240 * Check for empty string or invalid characters, but not if it is
1241 * generated (generated names use characters not available to the user)
1243 error_setg(errp
, "Invalid node name");
1247 /* takes care of avoiding namespaces collisions */
1248 if (blk_by_name(node_name
)) {
1249 error_setg(errp
, "node-name=%s is conflicting with a device id",
1254 /* takes care of avoiding duplicates node names */
1255 if (bdrv_find_node(node_name
)) {
1256 error_setg(errp
, "Duplicate node name");
1260 /* Make sure that the node name isn't truncated */
1261 if (strlen(node_name
) >= sizeof(bs
->node_name
)) {
1262 error_setg(errp
, "Node name too long");
1266 /* copy node name into the bs and insert it into the graph list */
1267 pstrcpy(bs
->node_name
, sizeof(bs
->node_name
), node_name
);
1268 QTAILQ_INSERT_TAIL(&graph_bdrv_states
, bs
, node_list
);
1270 g_free(gen_node_name
);
1273 static int bdrv_open_driver(BlockDriverState
*bs
, BlockDriver
*drv
,
1274 const char *node_name
, QDict
*options
,
1275 int open_flags
, Error
**errp
)
1277 Error
*local_err
= NULL
;
1280 bdrv_assign_node_name(bs
, node_name
, &local_err
);
1282 error_propagate(errp
, local_err
);
1287 bs
->read_only
= !(bs
->open_flags
& BDRV_O_RDWR
);
1288 bs
->opaque
= g_malloc0(drv
->instance_size
);
1290 if (drv
->bdrv_file_open
) {
1291 assert(!drv
->bdrv_needs_filename
|| bs
->filename
[0]);
1292 ret
= drv
->bdrv_file_open(bs
, options
, open_flags
, &local_err
);
1293 } else if (drv
->bdrv_open
) {
1294 ret
= drv
->bdrv_open(bs
, options
, open_flags
, &local_err
);
1301 error_propagate(errp
, local_err
);
1302 } else if (bs
->filename
[0]) {
1303 error_setg_errno(errp
, -ret
, "Could not open '%s'", bs
->filename
);
1305 error_setg_errno(errp
, -ret
, "Could not open image");
1310 ret
= refresh_total_sectors(bs
, bs
->total_sectors
);
1312 error_setg_errno(errp
, -ret
, "Could not refresh total sector count");
1316 bdrv_refresh_limits(bs
, &local_err
);
1318 error_propagate(errp
, local_err
);
1322 assert(bdrv_opt_mem_align(bs
) != 0);
1323 assert(bdrv_min_mem_align(bs
) != 0);
1324 assert(is_power_of_2(bs
->bl
.request_alignment
));
1326 for (i
= 0; i
< bs
->quiesce_counter
; i
++) {
1327 if (drv
->bdrv_co_drain_begin
) {
1328 drv
->bdrv_co_drain_begin(bs
);
1335 if (bs
->file
!= NULL
) {
1336 bdrv_unref_child(bs
, bs
->file
);
1344 BlockDriverState
*bdrv_new_open_driver(BlockDriver
*drv
, const char *node_name
,
1345 int flags
, Error
**errp
)
1347 BlockDriverState
*bs
;
1351 bs
->open_flags
= flags
;
1352 bs
->explicit_options
= qdict_new();
1353 bs
->options
= qdict_new();
1356 update_options_from_flags(bs
->options
, flags
);
1358 ret
= bdrv_open_driver(bs
, drv
, node_name
, bs
->options
, flags
, errp
);
1360 qobject_unref(bs
->explicit_options
);
1361 bs
->explicit_options
= NULL
;
1362 qobject_unref(bs
->options
);
1371 QemuOptsList bdrv_runtime_opts
= {
1372 .name
= "bdrv_common",
1373 .head
= QTAILQ_HEAD_INITIALIZER(bdrv_runtime_opts
.head
),
1376 .name
= "node-name",
1377 .type
= QEMU_OPT_STRING
,
1378 .help
= "Node name of the block device node",
1382 .type
= QEMU_OPT_STRING
,
1383 .help
= "Block driver to use for the node",
1386 .name
= BDRV_OPT_CACHE_DIRECT
,
1387 .type
= QEMU_OPT_BOOL
,
1388 .help
= "Bypass software writeback cache on the host",
1391 .name
= BDRV_OPT_CACHE_NO_FLUSH
,
1392 .type
= QEMU_OPT_BOOL
,
1393 .help
= "Ignore flush requests",
1396 .name
= BDRV_OPT_READ_ONLY
,
1397 .type
= QEMU_OPT_BOOL
,
1398 .help
= "Node is opened in read-only mode",
1401 .name
= BDRV_OPT_AUTO_READ_ONLY
,
1402 .type
= QEMU_OPT_BOOL
,
1403 .help
= "Node can become read-only if opening read-write fails",
1406 .name
= "detect-zeroes",
1407 .type
= QEMU_OPT_STRING
,
1408 .help
= "try to optimize zero writes (off, on, unmap)",
1411 .name
= BDRV_OPT_DISCARD
,
1412 .type
= QEMU_OPT_STRING
,
1413 .help
= "discard operation (ignore/off, unmap/on)",
1416 .name
= BDRV_OPT_FORCE_SHARE
,
1417 .type
= QEMU_OPT_BOOL
,
1418 .help
= "always accept other writers (default: off)",
1420 { /* end of list */ }
1425 * Common part for opening disk images and files
1427 * Removes all processed options from *options.
1429 static int bdrv_open_common(BlockDriverState
*bs
, BlockBackend
*file
,
1430 QDict
*options
, Error
**errp
)
1432 int ret
, open_flags
;
1433 const char *filename
;
1434 const char *driver_name
= NULL
;
1435 const char *node_name
= NULL
;
1436 const char *discard
;
1439 Error
*local_err
= NULL
;
1441 assert(bs
->file
== NULL
);
1442 assert(options
!= NULL
&& bs
->options
!= options
);
1444 opts
= qemu_opts_create(&bdrv_runtime_opts
, NULL
, 0, &error_abort
);
1445 qemu_opts_absorb_qdict(opts
, options
, &local_err
);
1447 error_propagate(errp
, local_err
);
1452 update_flags_from_options(&bs
->open_flags
, opts
);
1454 driver_name
= qemu_opt_get(opts
, "driver");
1455 drv
= bdrv_find_format(driver_name
);
1456 assert(drv
!= NULL
);
1458 bs
->force_share
= qemu_opt_get_bool(opts
, BDRV_OPT_FORCE_SHARE
, false);
1460 if (bs
->force_share
&& (bs
->open_flags
& BDRV_O_RDWR
)) {
1462 BDRV_OPT_FORCE_SHARE
1463 "=on can only be used with read-only images");
1469 bdrv_refresh_filename(blk_bs(file
));
1470 filename
= blk_bs(file
)->filename
;
1473 * Caution: while qdict_get_try_str() is fine, getting
1474 * non-string types would require more care. When @options
1475 * come from -blockdev or blockdev_add, its members are typed
1476 * according to the QAPI schema, but when they come from
1477 * -drive, they're all QString.
1479 filename
= qdict_get_try_str(options
, "filename");
1482 if (drv
->bdrv_needs_filename
&& (!filename
|| !filename
[0])) {
1483 error_setg(errp
, "The '%s' block driver requires a file name",
1489 trace_bdrv_open_common(bs
, filename
?: "", bs
->open_flags
,
1492 bs
->read_only
= !(bs
->open_flags
& BDRV_O_RDWR
);
1494 if (use_bdrv_whitelist
&& !bdrv_is_whitelisted(drv
, bs
->read_only
)) {
1495 if (!bs
->read_only
&& bdrv_is_whitelisted(drv
, true)) {
1496 ret
= bdrv_apply_auto_read_only(bs
, NULL
, NULL
);
1502 !bs
->read_only
&& bdrv_is_whitelisted(drv
, true)
1503 ? "Driver '%s' can only be used for read-only devices"
1504 : "Driver '%s' is not whitelisted",
1510 /* bdrv_new() and bdrv_close() make it so */
1511 assert(atomic_read(&bs
->copy_on_read
) == 0);
1513 if (bs
->open_flags
& BDRV_O_COPY_ON_READ
) {
1514 if (!bs
->read_only
) {
1515 bdrv_enable_copy_on_read(bs
);
1517 error_setg(errp
, "Can't use copy-on-read on read-only device");
1523 discard
= qemu_opt_get(opts
, BDRV_OPT_DISCARD
);
1524 if (discard
!= NULL
) {
1525 if (bdrv_parse_discard_flags(discard
, &bs
->open_flags
) != 0) {
1526 error_setg(errp
, "Invalid discard option");
1533 bdrv_parse_detect_zeroes(opts
, bs
->open_flags
, &local_err
);
1535 error_propagate(errp
, local_err
);
1540 if (filename
!= NULL
) {
1541 pstrcpy(bs
->filename
, sizeof(bs
->filename
), filename
);
1543 bs
->filename
[0] = '\0';
1545 pstrcpy(bs
->exact_filename
, sizeof(bs
->exact_filename
), bs
->filename
);
1547 /* Open the image, either directly or using a protocol */
1548 open_flags
= bdrv_open_flags(bs
, bs
->open_flags
);
1549 node_name
= qemu_opt_get(opts
, "node-name");
1551 assert(!drv
->bdrv_file_open
|| file
== NULL
);
1552 ret
= bdrv_open_driver(bs
, drv
, node_name
, options
, open_flags
, errp
);
1557 qemu_opts_del(opts
);
1561 qemu_opts_del(opts
);
1565 static QDict
*parse_json_filename(const char *filename
, Error
**errp
)
1567 QObject
*options_obj
;
1571 ret
= strstart(filename
, "json:", &filename
);
1574 options_obj
= qobject_from_json(filename
, errp
);
1576 error_prepend(errp
, "Could not parse the JSON options: ");
1580 options
= qobject_to(QDict
, options_obj
);
1582 qobject_unref(options_obj
);
1583 error_setg(errp
, "Invalid JSON object given");
1587 qdict_flatten(options
);
1592 static void parse_json_protocol(QDict
*options
, const char **pfilename
,
1595 QDict
*json_options
;
1596 Error
*local_err
= NULL
;
1598 /* Parse json: pseudo-protocol */
1599 if (!*pfilename
|| !g_str_has_prefix(*pfilename
, "json:")) {
1603 json_options
= parse_json_filename(*pfilename
, &local_err
);
1605 error_propagate(errp
, local_err
);
1609 /* Options given in the filename have lower priority than options
1610 * specified directly */
1611 qdict_join(options
, json_options
, false);
1612 qobject_unref(json_options
);
1617 * Fills in default options for opening images and converts the legacy
1618 * filename/flags pair to option QDict entries.
1619 * The BDRV_O_PROTOCOL flag in *flags will be set or cleared accordingly if a
1620 * block driver has been specified explicitly.
1622 static int bdrv_fill_options(QDict
**options
, const char *filename
,
1623 int *flags
, Error
**errp
)
1625 const char *drvname
;
1626 bool protocol
= *flags
& BDRV_O_PROTOCOL
;
1627 bool parse_filename
= false;
1628 BlockDriver
*drv
= NULL
;
1629 Error
*local_err
= NULL
;
1632 * Caution: while qdict_get_try_str() is fine, getting non-string
1633 * types would require more care. When @options come from
1634 * -blockdev or blockdev_add, its members are typed according to
1635 * the QAPI schema, but when they come from -drive, they're all
1638 drvname
= qdict_get_try_str(*options
, "driver");
1640 drv
= bdrv_find_format(drvname
);
1642 error_setg(errp
, "Unknown driver '%s'", drvname
);
1645 /* If the user has explicitly specified the driver, this choice should
1646 * override the BDRV_O_PROTOCOL flag */
1647 protocol
= drv
->bdrv_file_open
;
1651 *flags
|= BDRV_O_PROTOCOL
;
1653 *flags
&= ~BDRV_O_PROTOCOL
;
1656 /* Translate cache options from flags into options */
1657 update_options_from_flags(*options
, *flags
);
1659 /* Fetch the file name from the options QDict if necessary */
1660 if (protocol
&& filename
) {
1661 if (!qdict_haskey(*options
, "filename")) {
1662 qdict_put_str(*options
, "filename", filename
);
1663 parse_filename
= true;
1665 error_setg(errp
, "Can't specify 'file' and 'filename' options at "
1671 /* Find the right block driver */
1672 /* See cautionary note on accessing @options above */
1673 filename
= qdict_get_try_str(*options
, "filename");
1675 if (!drvname
&& protocol
) {
1677 drv
= bdrv_find_protocol(filename
, parse_filename
, errp
);
1682 drvname
= drv
->format_name
;
1683 qdict_put_str(*options
, "driver", drvname
);
1685 error_setg(errp
, "Must specify either driver or file");
1690 assert(drv
|| !protocol
);
1692 /* Driver-specific filename parsing */
1693 if (drv
&& drv
->bdrv_parse_filename
&& parse_filename
) {
1694 drv
->bdrv_parse_filename(filename
, *options
, &local_err
);
1696 error_propagate(errp
, local_err
);
1700 if (!drv
->bdrv_needs_filename
) {
1701 qdict_del(*options
, "filename");
1708 static int bdrv_child_check_perm(BdrvChild
*c
, BlockReopenQueue
*q
,
1709 uint64_t perm
, uint64_t shared
,
1710 GSList
*ignore_children
,
1711 bool *tighten_restrictions
, Error
**errp
);
1712 static void bdrv_child_abort_perm_update(BdrvChild
*c
);
1713 static void bdrv_child_set_perm(BdrvChild
*c
, uint64_t perm
, uint64_t shared
);
1714 static void bdrv_get_cumulative_perm(BlockDriverState
*bs
, uint64_t *perm
,
1715 uint64_t *shared_perm
);
1717 typedef struct BlockReopenQueueEntry
{
1720 BDRVReopenState state
;
1721 QSIMPLEQ_ENTRY(BlockReopenQueueEntry
) entry
;
1722 } BlockReopenQueueEntry
;
1725 * Return the flags that @bs will have after the reopens in @q have
1726 * successfully completed. If @q is NULL (or @bs is not contained in @q),
1727 * return the current flags.
1729 static int bdrv_reopen_get_flags(BlockReopenQueue
*q
, BlockDriverState
*bs
)
1731 BlockReopenQueueEntry
*entry
;
1734 QSIMPLEQ_FOREACH(entry
, q
, entry
) {
1735 if (entry
->state
.bs
== bs
) {
1736 return entry
->state
.flags
;
1741 return bs
->open_flags
;
1744 /* Returns whether the image file can be written to after the reopen queue @q
1745 * has been successfully applied, or right now if @q is NULL. */
1746 static bool bdrv_is_writable_after_reopen(BlockDriverState
*bs
,
1747 BlockReopenQueue
*q
)
1749 int flags
= bdrv_reopen_get_flags(q
, bs
);
1751 return (flags
& (BDRV_O_RDWR
| BDRV_O_INACTIVE
)) == BDRV_O_RDWR
;
1755 * Return whether the BDS can be written to. This is not necessarily
1756 * the same as !bdrv_is_read_only(bs), as inactivated images may not
1757 * be written to but do not count as read-only images.
1759 bool bdrv_is_writable(BlockDriverState
*bs
)
1761 return bdrv_is_writable_after_reopen(bs
, NULL
);
1764 static void bdrv_child_perm(BlockDriverState
*bs
, BlockDriverState
*child_bs
,
1765 BdrvChild
*c
, const BdrvChildRole
*role
,
1766 BlockReopenQueue
*reopen_queue
,
1767 uint64_t parent_perm
, uint64_t parent_shared
,
1768 uint64_t *nperm
, uint64_t *nshared
)
1770 assert(bs
->drv
&& bs
->drv
->bdrv_child_perm
);
1771 bs
->drv
->bdrv_child_perm(bs
, c
, role
, reopen_queue
,
1772 parent_perm
, parent_shared
,
1774 /* TODO Take force_share from reopen_queue */
1775 if (child_bs
&& child_bs
->force_share
) {
1776 *nshared
= BLK_PERM_ALL
;
1781 * Check whether permissions on this node can be changed in a way that
1782 * @cumulative_perms and @cumulative_shared_perms are the new cumulative
1783 * permissions of all its parents. This involves checking whether all necessary
1784 * permission changes to child nodes can be performed.
1786 * Will set *tighten_restrictions to true if and only if new permissions have to
1787 * be taken or currently shared permissions are to be unshared. Otherwise,
1788 * errors are not fatal as long as the caller accepts that the restrictions
1789 * remain tighter than they need to be. The caller still has to abort the
1791 * @tighten_restrictions cannot be used together with @q: When reopening, we may
1792 * encounter fatal errors even though no restrictions are to be tightened. For
1793 * example, changing a node from RW to RO will fail if the WRITE permission is
1796 * A call to this function must always be followed by a call to bdrv_set_perm()
1797 * or bdrv_abort_perm_update().
1799 static int bdrv_check_perm(BlockDriverState
*bs
, BlockReopenQueue
*q
,
1800 uint64_t cumulative_perms
,
1801 uint64_t cumulative_shared_perms
,
1802 GSList
*ignore_children
,
1803 bool *tighten_restrictions
, Error
**errp
)
1805 BlockDriver
*drv
= bs
->drv
;
1809 assert(!q
|| !tighten_restrictions
);
1811 if (tighten_restrictions
) {
1812 uint64_t current_perms
, current_shared
;
1813 uint64_t added_perms
, removed_shared_perms
;
1815 bdrv_get_cumulative_perm(bs
, ¤t_perms
, ¤t_shared
);
1817 added_perms
= cumulative_perms
& ~current_perms
;
1818 removed_shared_perms
= current_shared
& ~cumulative_shared_perms
;
1820 *tighten_restrictions
= added_perms
|| removed_shared_perms
;
1823 /* Write permissions never work with read-only images */
1824 if ((cumulative_perms
& (BLK_PERM_WRITE
| BLK_PERM_WRITE_UNCHANGED
)) &&
1825 !bdrv_is_writable_after_reopen(bs
, q
))
1827 if (!bdrv_is_writable_after_reopen(bs
, NULL
)) {
1828 error_setg(errp
, "Block node is read-only");
1830 uint64_t current_perms
, current_shared
;
1831 bdrv_get_cumulative_perm(bs
, ¤t_perms
, ¤t_shared
);
1832 if (current_perms
& (BLK_PERM_WRITE
| BLK_PERM_WRITE_UNCHANGED
)) {
1833 error_setg(errp
, "Cannot make block node read-only, there is "
1836 error_setg(errp
, "Cannot make block node read-only and create "
1844 /* Check this node */
1849 if (drv
->bdrv_check_perm
) {
1850 return drv
->bdrv_check_perm(bs
, cumulative_perms
,
1851 cumulative_shared_perms
, errp
);
1854 /* Drivers that never have children can omit .bdrv_child_perm() */
1855 if (!drv
->bdrv_child_perm
) {
1856 assert(QLIST_EMPTY(&bs
->children
));
1860 /* Check all children */
1861 QLIST_FOREACH(c
, &bs
->children
, next
) {
1862 uint64_t cur_perm
, cur_shared
;
1863 bool child_tighten_restr
;
1865 bdrv_child_perm(bs
, c
->bs
, c
, c
->role
, q
,
1866 cumulative_perms
, cumulative_shared_perms
,
1867 &cur_perm
, &cur_shared
);
1868 ret
= bdrv_child_check_perm(c
, q
, cur_perm
, cur_shared
, ignore_children
,
1869 tighten_restrictions
? &child_tighten_restr
1872 if (tighten_restrictions
) {
1873 *tighten_restrictions
|= child_tighten_restr
;
1884 * Notifies drivers that after a previous bdrv_check_perm() call, the
1885 * permission update is not performed and any preparations made for it (e.g.
1886 * taken file locks) need to be undone.
1888 * This function recursively notifies all child nodes.
1890 static void bdrv_abort_perm_update(BlockDriverState
*bs
)
1892 BlockDriver
*drv
= bs
->drv
;
1899 if (drv
->bdrv_abort_perm_update
) {
1900 drv
->bdrv_abort_perm_update(bs
);
1903 QLIST_FOREACH(c
, &bs
->children
, next
) {
1904 bdrv_child_abort_perm_update(c
);
1908 static void bdrv_set_perm(BlockDriverState
*bs
, uint64_t cumulative_perms
,
1909 uint64_t cumulative_shared_perms
)
1911 BlockDriver
*drv
= bs
->drv
;
1918 /* Update this node */
1919 if (drv
->bdrv_set_perm
) {
1920 drv
->bdrv_set_perm(bs
, cumulative_perms
, cumulative_shared_perms
);
1923 /* Drivers that never have children can omit .bdrv_child_perm() */
1924 if (!drv
->bdrv_child_perm
) {
1925 assert(QLIST_EMPTY(&bs
->children
));
1929 /* Update all children */
1930 QLIST_FOREACH(c
, &bs
->children
, next
) {
1931 uint64_t cur_perm
, cur_shared
;
1932 bdrv_child_perm(bs
, c
->bs
, c
, c
->role
, NULL
,
1933 cumulative_perms
, cumulative_shared_perms
,
1934 &cur_perm
, &cur_shared
);
1935 bdrv_child_set_perm(c
, cur_perm
, cur_shared
);
1939 static void bdrv_get_cumulative_perm(BlockDriverState
*bs
, uint64_t *perm
,
1940 uint64_t *shared_perm
)
1943 uint64_t cumulative_perms
= 0;
1944 uint64_t cumulative_shared_perms
= BLK_PERM_ALL
;
1946 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
1947 cumulative_perms
|= c
->perm
;
1948 cumulative_shared_perms
&= c
->shared_perm
;
1951 *perm
= cumulative_perms
;
1952 *shared_perm
= cumulative_shared_perms
;
1955 static char *bdrv_child_user_desc(BdrvChild
*c
)
1957 if (c
->role
->get_parent_desc
) {
1958 return c
->role
->get_parent_desc(c
);
1961 return g_strdup("another user");
1964 char *bdrv_perm_names(uint64_t perm
)
1970 { BLK_PERM_CONSISTENT_READ
, "consistent read" },
1971 { BLK_PERM_WRITE
, "write" },
1972 { BLK_PERM_WRITE_UNCHANGED
, "write unchanged" },
1973 { BLK_PERM_RESIZE
, "resize" },
1974 { BLK_PERM_GRAPH_MOD
, "change children" },
1978 char *result
= g_strdup("");
1979 struct perm_name
*p
;
1981 for (p
= permissions
; p
->name
; p
++) {
1982 if (perm
& p
->perm
) {
1984 result
= g_strdup_printf("%s%s%s", old
, *old
? ", " : "", p
->name
);
1993 * Checks whether a new reference to @bs can be added if the new user requires
1994 * @new_used_perm/@new_shared_perm as its permissions. If @ignore_children is
1995 * set, the BdrvChild objects in this list are ignored in the calculations;
1996 * this allows checking permission updates for an existing reference.
1998 * See bdrv_check_perm() for the semantics of @tighten_restrictions.
2000 * Needs to be followed by a call to either bdrv_set_perm() or
2001 * bdrv_abort_perm_update(). */
2002 static int bdrv_check_update_perm(BlockDriverState
*bs
, BlockReopenQueue
*q
,
2003 uint64_t new_used_perm
,
2004 uint64_t new_shared_perm
,
2005 GSList
*ignore_children
,
2006 bool *tighten_restrictions
,
2010 uint64_t cumulative_perms
= new_used_perm
;
2011 uint64_t cumulative_shared_perms
= new_shared_perm
;
2013 assert(!q
|| !tighten_restrictions
);
2015 /* There is no reason why anyone couldn't tolerate write_unchanged */
2016 assert(new_shared_perm
& BLK_PERM_WRITE_UNCHANGED
);
2018 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
2019 if (g_slist_find(ignore_children
, c
)) {
2023 if ((new_used_perm
& c
->shared_perm
) != new_used_perm
) {
2024 char *user
= bdrv_child_user_desc(c
);
2025 char *perm_names
= bdrv_perm_names(new_used_perm
& ~c
->shared_perm
);
2027 if (tighten_restrictions
) {
2028 *tighten_restrictions
= true;
2031 error_setg(errp
, "Conflicts with use by %s as '%s', which does not "
2033 user
, c
->name
, perm_names
, bdrv_get_node_name(c
->bs
));
2039 if ((c
->perm
& new_shared_perm
) != c
->perm
) {
2040 char *user
= bdrv_child_user_desc(c
);
2041 char *perm_names
= bdrv_perm_names(c
->perm
& ~new_shared_perm
);
2043 if (tighten_restrictions
) {
2044 *tighten_restrictions
= true;
2047 error_setg(errp
, "Conflicts with use by %s as '%s', which uses "
2049 user
, c
->name
, perm_names
, bdrv_get_node_name(c
->bs
));
2055 cumulative_perms
|= c
->perm
;
2056 cumulative_shared_perms
&= c
->shared_perm
;
2059 return bdrv_check_perm(bs
, q
, cumulative_perms
, cumulative_shared_perms
,
2060 ignore_children
, tighten_restrictions
, errp
);
2063 /* Needs to be followed by a call to either bdrv_child_set_perm() or
2064 * bdrv_child_abort_perm_update(). */
2065 static int bdrv_child_check_perm(BdrvChild
*c
, BlockReopenQueue
*q
,
2066 uint64_t perm
, uint64_t shared
,
2067 GSList
*ignore_children
,
2068 bool *tighten_restrictions
, Error
**errp
)
2072 ignore_children
= g_slist_prepend(g_slist_copy(ignore_children
), c
);
2073 ret
= bdrv_check_update_perm(c
->bs
, q
, perm
, shared
, ignore_children
,
2074 tighten_restrictions
, errp
);
2075 g_slist_free(ignore_children
);
2081 if (!c
->has_backup_perm
) {
2082 c
->has_backup_perm
= true;
2083 c
->backup_perm
= c
->perm
;
2084 c
->backup_shared_perm
= c
->shared_perm
;
2087 * Note: it's OK if c->has_backup_perm was already set, as we can find the
2088 * same child twice during check_perm procedure
2092 c
->shared_perm
= shared
;
2097 static void bdrv_child_set_perm(BdrvChild
*c
, uint64_t perm
, uint64_t shared
)
2099 uint64_t cumulative_perms
, cumulative_shared_perms
;
2101 c
->has_backup_perm
= false;
2104 c
->shared_perm
= shared
;
2106 bdrv_get_cumulative_perm(c
->bs
, &cumulative_perms
,
2107 &cumulative_shared_perms
);
2108 bdrv_set_perm(c
->bs
, cumulative_perms
, cumulative_shared_perms
);
2111 static void bdrv_child_abort_perm_update(BdrvChild
*c
)
2113 if (c
->has_backup_perm
) {
2114 c
->perm
= c
->backup_perm
;
2115 c
->shared_perm
= c
->backup_shared_perm
;
2116 c
->has_backup_perm
= false;
2119 bdrv_abort_perm_update(c
->bs
);
2122 int bdrv_child_try_set_perm(BdrvChild
*c
, uint64_t perm
, uint64_t shared
,
2125 Error
*local_err
= NULL
;
2127 bool tighten_restrictions
;
2129 ret
= bdrv_child_check_perm(c
, NULL
, perm
, shared
, NULL
,
2130 &tighten_restrictions
, &local_err
);
2132 bdrv_child_abort_perm_update(c
);
2133 if (tighten_restrictions
) {
2134 error_propagate(errp
, local_err
);
2137 * Our caller may intend to only loosen restrictions and
2138 * does not expect this function to fail. Errors are not
2139 * fatal in such a case, so we can just hide them from our
2142 error_free(local_err
);
2148 bdrv_child_set_perm(c
, perm
, shared
);
2153 int bdrv_child_refresh_perms(BlockDriverState
*bs
, BdrvChild
*c
, Error
**errp
)
2155 uint64_t parent_perms
, parent_shared
;
2156 uint64_t perms
, shared
;
2158 bdrv_get_cumulative_perm(bs
, &parent_perms
, &parent_shared
);
2159 bdrv_child_perm(bs
, c
->bs
, c
, c
->role
, NULL
, parent_perms
, parent_shared
,
2162 return bdrv_child_try_set_perm(c
, perms
, shared
, errp
);
2165 void bdrv_filter_default_perms(BlockDriverState
*bs
, BdrvChild
*c
,
2166 const BdrvChildRole
*role
,
2167 BlockReopenQueue
*reopen_queue
,
2168 uint64_t perm
, uint64_t shared
,
2169 uint64_t *nperm
, uint64_t *nshared
)
2171 *nperm
= perm
& DEFAULT_PERM_PASSTHROUGH
;
2172 *nshared
= (shared
& DEFAULT_PERM_PASSTHROUGH
) | DEFAULT_PERM_UNCHANGED
;
2175 void bdrv_format_default_perms(BlockDriverState
*bs
, BdrvChild
*c
,
2176 const BdrvChildRole
*role
,
2177 BlockReopenQueue
*reopen_queue
,
2178 uint64_t perm
, uint64_t shared
,
2179 uint64_t *nperm
, uint64_t *nshared
)
2181 bool backing
= (role
== &child_backing
);
2182 assert(role
== &child_backing
|| role
== &child_file
);
2185 int flags
= bdrv_reopen_get_flags(reopen_queue
, bs
);
2187 /* Apart from the modifications below, the same permissions are
2188 * forwarded and left alone as for filters */
2189 bdrv_filter_default_perms(bs
, c
, role
, reopen_queue
, perm
, shared
,
2192 /* Format drivers may touch metadata even if the guest doesn't write */
2193 if (bdrv_is_writable_after_reopen(bs
, reopen_queue
)) {
2194 perm
|= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2197 /* bs->file always needs to be consistent because of the metadata. We
2198 * can never allow other users to resize or write to it. */
2199 if (!(flags
& BDRV_O_NO_IO
)) {
2200 perm
|= BLK_PERM_CONSISTENT_READ
;
2202 shared
&= ~(BLK_PERM_WRITE
| BLK_PERM_RESIZE
);
2204 /* We want consistent read from backing files if the parent needs it.
2205 * No other operations are performed on backing files. */
2206 perm
&= BLK_PERM_CONSISTENT_READ
;
2208 /* If the parent can deal with changing data, we're okay with a
2209 * writable and resizable backing file. */
2210 /* TODO Require !(perm & BLK_PERM_CONSISTENT_READ), too? */
2211 if (shared
& BLK_PERM_WRITE
) {
2212 shared
= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2217 shared
|= BLK_PERM_CONSISTENT_READ
| BLK_PERM_GRAPH_MOD
|
2218 BLK_PERM_WRITE_UNCHANGED
;
2221 if (bs
->open_flags
& BDRV_O_INACTIVE
) {
2222 shared
|= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2229 static void bdrv_replace_child_noperm(BdrvChild
*child
,
2230 BlockDriverState
*new_bs
)
2232 BlockDriverState
*old_bs
= child
->bs
;
2235 assert(!child
->frozen
);
2237 if (old_bs
&& new_bs
) {
2238 assert(bdrv_get_aio_context(old_bs
) == bdrv_get_aio_context(new_bs
));
2241 /* Detach first so that the recursive drain sections coming from @child
2242 * are already gone and we only end the drain sections that came from
2244 if (child
->role
->detach
) {
2245 child
->role
->detach(child
);
2247 while (child
->parent_quiesce_counter
) {
2248 bdrv_parent_drained_end_single(child
);
2250 QLIST_REMOVE(child
, next_parent
);
2252 assert(child
->parent_quiesce_counter
== 0);
2258 QLIST_INSERT_HEAD(&new_bs
->parents
, child
, next_parent
);
2259 if (new_bs
->quiesce_counter
) {
2260 int num
= new_bs
->quiesce_counter
;
2261 if (child
->role
->parent_is_bds
) {
2262 num
-= bdrv_drain_all_count
;
2265 for (i
= 0; i
< num
; i
++) {
2266 bdrv_parent_drained_begin_single(child
, true);
2270 /* Attach only after starting new drained sections, so that recursive
2271 * drain sections coming from @child don't get an extra .drained_begin
2273 if (child
->role
->attach
) {
2274 child
->role
->attach(child
);
2280 * Updates @child to change its reference to point to @new_bs, including
2281 * checking and applying the necessary permisson updates both to the old node
2284 * NULL is passed as @new_bs for removing the reference before freeing @child.
2286 * If @new_bs is not NULL, bdrv_check_perm() must be called beforehand, as this
2287 * function uses bdrv_set_perm() to update the permissions according to the new
2288 * reference that @new_bs gets.
2290 static void bdrv_replace_child(BdrvChild
*child
, BlockDriverState
*new_bs
)
2292 BlockDriverState
*old_bs
= child
->bs
;
2293 uint64_t perm
, shared_perm
;
2295 bdrv_replace_child_noperm(child
, new_bs
);
2298 * Start with the new node's permissions. If @new_bs is a (direct
2299 * or indirect) child of @old_bs, we must complete the permission
2300 * update on @new_bs before we loosen the restrictions on @old_bs.
2301 * Otherwise, bdrv_check_perm() on @old_bs would re-initiate
2302 * updating the permissions of @new_bs, and thus not purely loosen
2306 bdrv_get_cumulative_perm(new_bs
, &perm
, &shared_perm
);
2307 bdrv_set_perm(new_bs
, perm
, shared_perm
);
2311 /* Update permissions for old node. This is guaranteed to succeed
2312 * because we're just taking a parent away, so we're loosening
2314 bool tighten_restrictions
;
2317 bdrv_get_cumulative_perm(old_bs
, &perm
, &shared_perm
);
2318 ret
= bdrv_check_perm(old_bs
, NULL
, perm
, shared_perm
, NULL
,
2319 &tighten_restrictions
, NULL
);
2320 assert(tighten_restrictions
== false);
2322 /* We only tried to loosen restrictions, so errors are not fatal */
2323 bdrv_abort_perm_update(old_bs
);
2325 bdrv_set_perm(old_bs
, perm
, shared_perm
);
2328 /* When the parent requiring a non-default AioContext is removed, the
2329 * node moves back to the main AioContext */
2330 bdrv_try_set_aio_context(old_bs
, qemu_get_aio_context(), NULL
);
2335 * This function steals the reference to child_bs from the caller.
2336 * That reference is later dropped by bdrv_root_unref_child().
2338 * On failure NULL is returned, errp is set and the reference to
2339 * child_bs is also dropped.
2341 * The caller must hold the AioContext lock @child_bs, but not that of @ctx
2342 * (unless @child_bs is already in @ctx).
2344 BdrvChild
*bdrv_root_attach_child(BlockDriverState
*child_bs
,
2345 const char *child_name
,
2346 const BdrvChildRole
*child_role
,
2348 uint64_t perm
, uint64_t shared_perm
,
2349 void *opaque
, Error
**errp
)
2352 Error
*local_err
= NULL
;
2355 ret
= bdrv_check_update_perm(child_bs
, NULL
, perm
, shared_perm
, NULL
, NULL
,
2358 bdrv_abort_perm_update(child_bs
);
2359 bdrv_unref(child_bs
);
2363 child
= g_new(BdrvChild
, 1);
2364 *child
= (BdrvChild
) {
2366 .name
= g_strdup(child_name
),
2369 .shared_perm
= shared_perm
,
2373 /* If the AioContexts don't match, first try to move the subtree of
2374 * child_bs into the AioContext of the new parent. If this doesn't work,
2375 * try moving the parent into the AioContext of child_bs instead. */
2376 if (bdrv_get_aio_context(child_bs
) != ctx
) {
2377 ret
= bdrv_try_set_aio_context(child_bs
, ctx
, &local_err
);
2378 if (ret
< 0 && child_role
->can_set_aio_ctx
) {
2379 GSList
*ignore
= g_slist_prepend(NULL
, child
);;
2380 ctx
= bdrv_get_aio_context(child_bs
);
2381 if (child_role
->can_set_aio_ctx(child
, ctx
, &ignore
, NULL
)) {
2382 error_free(local_err
);
2384 g_slist_free(ignore
);
2385 ignore
= g_slist_prepend(NULL
, child
);;
2386 child_role
->set_aio_ctx(child
, ctx
, &ignore
);
2388 g_slist_free(ignore
);
2391 error_propagate(errp
, local_err
);
2393 bdrv_abort_perm_update(child_bs
);
2398 /* This performs the matching bdrv_set_perm() for the above check. */
2399 bdrv_replace_child(child
, child_bs
);
2405 * This function transfers the reference to child_bs from the caller
2406 * to parent_bs. That reference is later dropped by parent_bs on
2407 * bdrv_close() or if someone calls bdrv_unref_child().
2409 * On failure NULL is returned, errp is set and the reference to
2410 * child_bs is also dropped.
2412 * If @parent_bs and @child_bs are in different AioContexts, the caller must
2413 * hold the AioContext lock for @child_bs, but not for @parent_bs.
2415 BdrvChild
*bdrv_attach_child(BlockDriverState
*parent_bs
,
2416 BlockDriverState
*child_bs
,
2417 const char *child_name
,
2418 const BdrvChildRole
*child_role
,
2422 uint64_t perm
, shared_perm
;
2424 bdrv_get_cumulative_perm(parent_bs
, &perm
, &shared_perm
);
2426 assert(parent_bs
->drv
);
2427 bdrv_child_perm(parent_bs
, child_bs
, NULL
, child_role
, NULL
,
2428 perm
, shared_perm
, &perm
, &shared_perm
);
2430 child
= bdrv_root_attach_child(child_bs
, child_name
, child_role
,
2431 bdrv_get_aio_context(parent_bs
),
2432 perm
, shared_perm
, parent_bs
, errp
);
2433 if (child
== NULL
) {
2437 QLIST_INSERT_HEAD(&parent_bs
->children
, child
, next
);
2441 static void bdrv_detach_child(BdrvChild
*child
)
2443 if (child
->next
.le_prev
) {
2444 QLIST_REMOVE(child
, next
);
2445 child
->next
.le_prev
= NULL
;
2448 bdrv_replace_child(child
, NULL
);
2450 g_free(child
->name
);
2454 void bdrv_root_unref_child(BdrvChild
*child
)
2456 BlockDriverState
*child_bs
;
2458 child_bs
= child
->bs
;
2459 bdrv_detach_child(child
);
2460 bdrv_unref(child_bs
);
2464 * Clear all inherits_from pointers from children and grandchildren of
2465 * @root that point to @root, where necessary.
2467 static void bdrv_unset_inherits_from(BlockDriverState
*root
, BdrvChild
*child
)
2471 if (child
->bs
->inherits_from
== root
) {
2473 * Remove inherits_from only when the last reference between root and
2474 * child->bs goes away.
2476 QLIST_FOREACH(c
, &root
->children
, next
) {
2477 if (c
!= child
&& c
->bs
== child
->bs
) {
2482 child
->bs
->inherits_from
= NULL
;
2486 QLIST_FOREACH(c
, &child
->bs
->children
, next
) {
2487 bdrv_unset_inherits_from(root
, c
);
2491 void bdrv_unref_child(BlockDriverState
*parent
, BdrvChild
*child
)
2493 if (child
== NULL
) {
2497 bdrv_unset_inherits_from(parent
, child
);
2498 bdrv_root_unref_child(child
);
2502 static void bdrv_parent_cb_change_media(BlockDriverState
*bs
, bool load
)
2505 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
2506 if (c
->role
->change_media
) {
2507 c
->role
->change_media(c
, load
);
2512 /* Return true if you can reach parent going through child->inherits_from
2513 * recursively. If parent or child are NULL, return false */
2514 static bool bdrv_inherits_from_recursive(BlockDriverState
*child
,
2515 BlockDriverState
*parent
)
2517 while (child
&& child
!= parent
) {
2518 child
= child
->inherits_from
;
2521 return child
!= NULL
;
2525 * Sets the backing file link of a BDS. A new reference is created; callers
2526 * which don't need their own reference any more must call bdrv_unref().
2528 void bdrv_set_backing_hd(BlockDriverState
*bs
, BlockDriverState
*backing_hd
,
2531 bool update_inherits_from
= bdrv_chain_contains(bs
, backing_hd
) &&
2532 bdrv_inherits_from_recursive(backing_hd
, bs
);
2534 if (bdrv_is_backing_chain_frozen(bs
, backing_bs(bs
), errp
)) {
2539 bdrv_ref(backing_hd
);
2543 bdrv_unref_child(bs
, bs
->backing
);
2551 bs
->backing
= bdrv_attach_child(bs
, backing_hd
, "backing", &child_backing
,
2553 /* If backing_hd was already part of bs's backing chain, and
2554 * inherits_from pointed recursively to bs then let's update it to
2555 * point directly to bs (else it will become NULL). */
2556 if (bs
->backing
&& update_inherits_from
) {
2557 backing_hd
->inherits_from
= bs
;
2561 bdrv_refresh_limits(bs
, NULL
);
2565 * Opens the backing file for a BlockDriverState if not yet open
2567 * bdref_key specifies the key for the image's BlockdevRef in the options QDict.
2568 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
2569 * itself, all options starting with "${bdref_key}." are considered part of the
2572 * TODO Can this be unified with bdrv_open_image()?
2574 int bdrv_open_backing_file(BlockDriverState
*bs
, QDict
*parent_options
,
2575 const char *bdref_key
, Error
**errp
)
2577 char *backing_filename
= NULL
;
2578 char *bdref_key_dot
;
2579 const char *reference
= NULL
;
2581 bool implicit_backing
= false;
2582 BlockDriverState
*backing_hd
;
2584 QDict
*tmp_parent_options
= NULL
;
2585 Error
*local_err
= NULL
;
2587 if (bs
->backing
!= NULL
) {
2591 /* NULL means an empty set of options */
2592 if (parent_options
== NULL
) {
2593 tmp_parent_options
= qdict_new();
2594 parent_options
= tmp_parent_options
;
2597 bs
->open_flags
&= ~BDRV_O_NO_BACKING
;
2599 bdref_key_dot
= g_strdup_printf("%s.", bdref_key
);
2600 qdict_extract_subqdict(parent_options
, &options
, bdref_key_dot
);
2601 g_free(bdref_key_dot
);
2604 * Caution: while qdict_get_try_str() is fine, getting non-string
2605 * types would require more care. When @parent_options come from
2606 * -blockdev or blockdev_add, its members are typed according to
2607 * the QAPI schema, but when they come from -drive, they're all
2610 reference
= qdict_get_try_str(parent_options
, bdref_key
);
2611 if (reference
|| qdict_haskey(options
, "file.filename")) {
2612 /* keep backing_filename NULL */
2613 } else if (bs
->backing_file
[0] == '\0' && qdict_size(options
) == 0) {
2614 qobject_unref(options
);
2617 if (qdict_size(options
) == 0) {
2618 /* If the user specifies options that do not modify the
2619 * backing file's behavior, we might still consider it the
2620 * implicit backing file. But it's easier this way, and
2621 * just specifying some of the backing BDS's options is
2622 * only possible with -drive anyway (otherwise the QAPI
2623 * schema forces the user to specify everything). */
2624 implicit_backing
= !strcmp(bs
->auto_backing_file
, bs
->backing_file
);
2627 backing_filename
= bdrv_get_full_backing_filename(bs
, &local_err
);
2630 error_propagate(errp
, local_err
);
2631 qobject_unref(options
);
2636 if (!bs
->drv
|| !bs
->drv
->supports_backing
) {
2638 error_setg(errp
, "Driver doesn't support backing files");
2639 qobject_unref(options
);
2644 bs
->backing_format
[0] != '\0' && !qdict_haskey(options
, "driver")) {
2645 qdict_put_str(options
, "driver", bs
->backing_format
);
2648 backing_hd
= bdrv_open_inherit(backing_filename
, reference
, options
, 0, bs
,
2649 &child_backing
, errp
);
2651 bs
->open_flags
|= BDRV_O_NO_BACKING
;
2652 error_prepend(errp
, "Could not open backing file: ");
2657 if (implicit_backing
) {
2658 bdrv_refresh_filename(backing_hd
);
2659 pstrcpy(bs
->auto_backing_file
, sizeof(bs
->auto_backing_file
),
2660 backing_hd
->filename
);
2663 /* Hook up the backing file link; drop our reference, bs owns the
2664 * backing_hd reference now */
2665 bdrv_set_backing_hd(bs
, backing_hd
, &local_err
);
2666 bdrv_unref(backing_hd
);
2668 error_propagate(errp
, local_err
);
2673 qdict_del(parent_options
, bdref_key
);
2676 g_free(backing_filename
);
2677 qobject_unref(tmp_parent_options
);
2681 static BlockDriverState
*
2682 bdrv_open_child_bs(const char *filename
, QDict
*options
, const char *bdref_key
,
2683 BlockDriverState
*parent
, const BdrvChildRole
*child_role
,
2684 bool allow_none
, Error
**errp
)
2686 BlockDriverState
*bs
= NULL
;
2687 QDict
*image_options
;
2688 char *bdref_key_dot
;
2689 const char *reference
;
2691 assert(child_role
!= NULL
);
2693 bdref_key_dot
= g_strdup_printf("%s.", bdref_key
);
2694 qdict_extract_subqdict(options
, &image_options
, bdref_key_dot
);
2695 g_free(bdref_key_dot
);
2698 * Caution: while qdict_get_try_str() is fine, getting non-string
2699 * types would require more care. When @options come from
2700 * -blockdev or blockdev_add, its members are typed according to
2701 * the QAPI schema, but when they come from -drive, they're all
2704 reference
= qdict_get_try_str(options
, bdref_key
);
2705 if (!filename
&& !reference
&& !qdict_size(image_options
)) {
2707 error_setg(errp
, "A block device must be specified for \"%s\"",
2710 qobject_unref(image_options
);
2714 bs
= bdrv_open_inherit(filename
, reference
, image_options
, 0,
2715 parent
, child_role
, errp
);
2721 qdict_del(options
, bdref_key
);
2726 * Opens a disk image whose options are given as BlockdevRef in another block
2729 * If allow_none is true, no image will be opened if filename is false and no
2730 * BlockdevRef is given. NULL will be returned, but errp remains unset.
2732 * bdrev_key specifies the key for the image's BlockdevRef in the options QDict.
2733 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
2734 * itself, all options starting with "${bdref_key}." are considered part of the
2737 * The BlockdevRef will be removed from the options QDict.
2739 BdrvChild
*bdrv_open_child(const char *filename
,
2740 QDict
*options
, const char *bdref_key
,
2741 BlockDriverState
*parent
,
2742 const BdrvChildRole
*child_role
,
2743 bool allow_none
, Error
**errp
)
2745 BlockDriverState
*bs
;
2747 bs
= bdrv_open_child_bs(filename
, options
, bdref_key
, parent
, child_role
,
2753 return bdrv_attach_child(parent
, bs
, bdref_key
, child_role
, errp
);
2756 /* TODO Future callers may need to specify parent/child_role in order for
2757 * option inheritance to work. Existing callers use it for the root node. */
2758 BlockDriverState
*bdrv_open_blockdev_ref(BlockdevRef
*ref
, Error
**errp
)
2760 BlockDriverState
*bs
= NULL
;
2761 Error
*local_err
= NULL
;
2762 QObject
*obj
= NULL
;
2763 QDict
*qdict
= NULL
;
2764 const char *reference
= NULL
;
2767 if (ref
->type
== QTYPE_QSTRING
) {
2768 reference
= ref
->u
.reference
;
2770 BlockdevOptions
*options
= &ref
->u
.definition
;
2771 assert(ref
->type
== QTYPE_QDICT
);
2773 v
= qobject_output_visitor_new(&obj
);
2774 visit_type_BlockdevOptions(v
, NULL
, &options
, &local_err
);
2776 error_propagate(errp
, local_err
);
2779 visit_complete(v
, &obj
);
2781 qdict
= qobject_to(QDict
, obj
);
2782 qdict_flatten(qdict
);
2784 /* bdrv_open_inherit() defaults to the values in bdrv_flags (for
2785 * compatibility with other callers) rather than what we want as the
2786 * real defaults. Apply the defaults here instead. */
2787 qdict_set_default_str(qdict
, BDRV_OPT_CACHE_DIRECT
, "off");
2788 qdict_set_default_str(qdict
, BDRV_OPT_CACHE_NO_FLUSH
, "off");
2789 qdict_set_default_str(qdict
, BDRV_OPT_READ_ONLY
, "off");
2790 qdict_set_default_str(qdict
, BDRV_OPT_AUTO_READ_ONLY
, "off");
2794 bs
= bdrv_open_inherit(NULL
, reference
, qdict
, 0, NULL
, NULL
, errp
);
2803 static BlockDriverState
*bdrv_append_temp_snapshot(BlockDriverState
*bs
,
2805 QDict
*snapshot_options
,
2808 /* TODO: extra byte is a hack to ensure MAX_PATH space on Windows. */
2809 char *tmp_filename
= g_malloc0(PATH_MAX
+ 1);
2811 QemuOpts
*opts
= NULL
;
2812 BlockDriverState
*bs_snapshot
= NULL
;
2813 Error
*local_err
= NULL
;
2816 /* if snapshot, we create a temporary backing file and open it
2817 instead of opening 'filename' directly */
2819 /* Get the required size from the image */
2820 total_size
= bdrv_getlength(bs
);
2821 if (total_size
< 0) {
2822 error_setg_errno(errp
, -total_size
, "Could not get image size");
2826 /* Create the temporary image */
2827 ret
= get_tmp_filename(tmp_filename
, PATH_MAX
+ 1);
2829 error_setg_errno(errp
, -ret
, "Could not get temporary filename");
2833 opts
= qemu_opts_create(bdrv_qcow2
.create_opts
, NULL
, 0,
2835 qemu_opt_set_number(opts
, BLOCK_OPT_SIZE
, total_size
, &error_abort
);
2836 ret
= bdrv_create(&bdrv_qcow2
, tmp_filename
, opts
, errp
);
2837 qemu_opts_del(opts
);
2839 error_prepend(errp
, "Could not create temporary overlay '%s': ",
2844 /* Prepare options QDict for the temporary file */
2845 qdict_put_str(snapshot_options
, "file.driver", "file");
2846 qdict_put_str(snapshot_options
, "file.filename", tmp_filename
);
2847 qdict_put_str(snapshot_options
, "driver", "qcow2");
2849 bs_snapshot
= bdrv_open(NULL
, NULL
, snapshot_options
, flags
, errp
);
2850 snapshot_options
= NULL
;
2855 /* bdrv_append() consumes a strong reference to bs_snapshot
2856 * (i.e. it will call bdrv_unref() on it) even on error, so in
2857 * order to be able to return one, we have to increase
2858 * bs_snapshot's refcount here */
2859 bdrv_ref(bs_snapshot
);
2860 bdrv_append(bs_snapshot
, bs
, &local_err
);
2862 error_propagate(errp
, local_err
);
2868 qobject_unref(snapshot_options
);
2869 g_free(tmp_filename
);
2874 * Opens a disk image (raw, qcow2, vmdk, ...)
2876 * options is a QDict of options to pass to the block drivers, or NULL for an
2877 * empty set of options. The reference to the QDict belongs to the block layer
2878 * after the call (even on failure), so if the caller intends to reuse the
2879 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
2881 * If *pbs is NULL, a new BDS will be created with a pointer to it stored there.
2882 * If it is not NULL, the referenced BDS will be reused.
2884 * The reference parameter may be used to specify an existing block device which
2885 * should be opened. If specified, neither options nor a filename may be given,
2886 * nor can an existing BDS be reused (that is, *pbs has to be NULL).
2888 static BlockDriverState
*bdrv_open_inherit(const char *filename
,
2889 const char *reference
,
2890 QDict
*options
, int flags
,
2891 BlockDriverState
*parent
,
2892 const BdrvChildRole
*child_role
,
2896 BlockBackend
*file
= NULL
;
2897 BlockDriverState
*bs
;
2898 BlockDriver
*drv
= NULL
;
2900 const char *drvname
;
2901 const char *backing
;
2902 Error
*local_err
= NULL
;
2903 QDict
*snapshot_options
= NULL
;
2904 int snapshot_flags
= 0;
2906 assert(!child_role
|| !flags
);
2907 assert(!child_role
== !parent
);
2910 bool options_non_empty
= options
? qdict_size(options
) : false;
2911 qobject_unref(options
);
2913 if (filename
|| options_non_empty
) {
2914 error_setg(errp
, "Cannot reference an existing block device with "
2915 "additional options or a new filename");
2919 bs
= bdrv_lookup_bs(reference
, reference
, errp
);
2930 /* NULL means an empty set of options */
2931 if (options
== NULL
) {
2932 options
= qdict_new();
2935 /* json: syntax counts as explicit options, as if in the QDict */
2936 parse_json_protocol(options
, &filename
, &local_err
);
2941 bs
->explicit_options
= qdict_clone_shallow(options
);
2944 bs
->inherits_from
= parent
;
2945 child_role
->inherit_options(&flags
, options
,
2946 parent
->open_flags
, parent
->options
);
2949 ret
= bdrv_fill_options(&options
, filename
, &flags
, &local_err
);
2955 * Set the BDRV_O_RDWR and BDRV_O_ALLOW_RDWR flags.
2956 * Caution: getting a boolean member of @options requires care.
2957 * When @options come from -blockdev or blockdev_add, members are
2958 * typed according to the QAPI schema, but when they come from
2959 * -drive, they're all QString.
2961 if (g_strcmp0(qdict_get_try_str(options
, BDRV_OPT_READ_ONLY
), "on") &&
2962 !qdict_get_try_bool(options
, BDRV_OPT_READ_ONLY
, false)) {
2963 flags
|= (BDRV_O_RDWR
| BDRV_O_ALLOW_RDWR
);
2965 flags
&= ~BDRV_O_RDWR
;
2968 if (flags
& BDRV_O_SNAPSHOT
) {
2969 snapshot_options
= qdict_new();
2970 bdrv_temp_snapshot_options(&snapshot_flags
, snapshot_options
,
2972 /* Let bdrv_backing_options() override "read-only" */
2973 qdict_del(options
, BDRV_OPT_READ_ONLY
);
2974 bdrv_backing_options(&flags
, options
, flags
, options
);
2977 bs
->open_flags
= flags
;
2978 bs
->options
= options
;
2979 options
= qdict_clone_shallow(options
);
2981 /* Find the right image format driver */
2982 /* See cautionary note on accessing @options above */
2983 drvname
= qdict_get_try_str(options
, "driver");
2985 drv
= bdrv_find_format(drvname
);
2987 error_setg(errp
, "Unknown driver: '%s'", drvname
);
2992 assert(drvname
|| !(flags
& BDRV_O_PROTOCOL
));
2994 /* See cautionary note on accessing @options above */
2995 backing
= qdict_get_try_str(options
, "backing");
2996 if (qobject_to(QNull
, qdict_get(options
, "backing")) != NULL
||
2997 (backing
&& *backing
== '\0'))
3000 warn_report("Use of \"backing\": \"\" is deprecated; "
3001 "use \"backing\": null instead");
3003 flags
|= BDRV_O_NO_BACKING
;
3004 qdict_del(options
, "backing");
3007 /* Open image file without format layer. This BlockBackend is only used for
3008 * probing, the block drivers will do their own bdrv_open_child() for the
3009 * same BDS, which is why we put the node name back into options. */
3010 if ((flags
& BDRV_O_PROTOCOL
) == 0) {
3011 BlockDriverState
*file_bs
;
3013 file_bs
= bdrv_open_child_bs(filename
, options
, "file", bs
,
3014 &child_file
, true, &local_err
);
3018 if (file_bs
!= NULL
) {
3019 /* Not requesting BLK_PERM_CONSISTENT_READ because we're only
3020 * looking at the header to guess the image format. This works even
3021 * in cases where a guest would not see a consistent state. */
3022 file
= blk_new(bdrv_get_aio_context(file_bs
), 0, BLK_PERM_ALL
);
3023 blk_insert_bs(file
, file_bs
, &local_err
);
3024 bdrv_unref(file_bs
);
3029 qdict_put_str(options
, "file", bdrv_get_node_name(file_bs
));
3033 /* Image format probing */
3036 ret
= find_image_format(file
, filename
, &drv
, &local_err
);
3041 * This option update would logically belong in bdrv_fill_options(),
3042 * but we first need to open bs->file for the probing to work, while
3043 * opening bs->file already requires the (mostly) final set of options
3044 * so that cache mode etc. can be inherited.
3046 * Adding the driver later is somewhat ugly, but it's not an option
3047 * that would ever be inherited, so it's correct. We just need to make
3048 * sure to update both bs->options (which has the full effective
3049 * options for bs) and options (which has file.* already removed).
3051 qdict_put_str(bs
->options
, "driver", drv
->format_name
);
3052 qdict_put_str(options
, "driver", drv
->format_name
);
3054 error_setg(errp
, "Must specify either driver or file");
3058 /* BDRV_O_PROTOCOL must be set iff a protocol BDS is about to be created */
3059 assert(!!(flags
& BDRV_O_PROTOCOL
) == !!drv
->bdrv_file_open
);
3060 /* file must be NULL if a protocol BDS is about to be created
3061 * (the inverse results in an error message from bdrv_open_common()) */
3062 assert(!(flags
& BDRV_O_PROTOCOL
) || !file
);
3064 /* Open the image */
3065 ret
= bdrv_open_common(bs
, file
, options
, &local_err
);
3075 /* If there is a backing file, use it */
3076 if ((flags
& BDRV_O_NO_BACKING
) == 0) {
3077 ret
= bdrv_open_backing_file(bs
, options
, "backing", &local_err
);
3079 goto close_and_fail
;
3083 /* Remove all children options and references
3084 * from bs->options and bs->explicit_options */
3085 QLIST_FOREACH(child
, &bs
->children
, next
) {
3086 char *child_key_dot
;
3087 child_key_dot
= g_strdup_printf("%s.", child
->name
);
3088 qdict_extract_subqdict(bs
->explicit_options
, NULL
, child_key_dot
);
3089 qdict_extract_subqdict(bs
->options
, NULL
, child_key_dot
);
3090 qdict_del(bs
->explicit_options
, child
->name
);
3091 qdict_del(bs
->options
, child
->name
);
3092 g_free(child_key_dot
);
3095 /* Check if any unknown options were used */
3096 if (qdict_size(options
) != 0) {
3097 const QDictEntry
*entry
= qdict_first(options
);
3098 if (flags
& BDRV_O_PROTOCOL
) {
3099 error_setg(errp
, "Block protocol '%s' doesn't support the option "
3100 "'%s'", drv
->format_name
, entry
->key
);
3103 "Block format '%s' does not support the option '%s'",
3104 drv
->format_name
, entry
->key
);
3107 goto close_and_fail
;
3110 bdrv_parent_cb_change_media(bs
, true);
3112 qobject_unref(options
);
3115 /* For snapshot=on, create a temporary qcow2 overlay. bs points to the
3116 * temporary snapshot afterwards. */
3117 if (snapshot_flags
) {
3118 BlockDriverState
*snapshot_bs
;
3119 snapshot_bs
= bdrv_append_temp_snapshot(bs
, snapshot_flags
,
3120 snapshot_options
, &local_err
);
3121 snapshot_options
= NULL
;
3123 goto close_and_fail
;
3125 /* We are not going to return bs but the overlay on top of it
3126 * (snapshot_bs); thus, we have to drop the strong reference to bs
3127 * (which we obtained by calling bdrv_new()). bs will not be deleted,
3128 * though, because the overlay still has a reference to it. */
3137 qobject_unref(snapshot_options
);
3138 qobject_unref(bs
->explicit_options
);
3139 qobject_unref(bs
->options
);
3140 qobject_unref(options
);
3142 bs
->explicit_options
= NULL
;
3144 error_propagate(errp
, local_err
);
3149 qobject_unref(snapshot_options
);
3150 qobject_unref(options
);
3151 error_propagate(errp
, local_err
);
3155 BlockDriverState
*bdrv_open(const char *filename
, const char *reference
,
3156 QDict
*options
, int flags
, Error
**errp
)
3158 return bdrv_open_inherit(filename
, reference
, options
, flags
, NULL
,
3162 /* Return true if the NULL-terminated @list contains @str */
3163 static bool is_str_in_list(const char *str
, const char *const *list
)
3167 for (i
= 0; list
[i
] != NULL
; i
++) {
3168 if (!strcmp(str
, list
[i
])) {
3177 * Check that every option set in @bs->options is also set in
3180 * Options listed in the common_options list and in
3181 * @bs->drv->mutable_opts are skipped.
3183 * Return 0 on success, otherwise return -EINVAL and set @errp.
3185 static int bdrv_reset_options_allowed(BlockDriverState
*bs
,
3186 const QDict
*new_opts
, Error
**errp
)
3188 const QDictEntry
*e
;
3189 /* These options are common to all block drivers and are handled
3190 * in bdrv_reopen_prepare() so they can be left out of @new_opts */
3191 const char *const common_options
[] = {
3192 "node-name", "discard", "cache.direct", "cache.no-flush",
3193 "read-only", "auto-read-only", "detect-zeroes", NULL
3196 for (e
= qdict_first(bs
->options
); e
; e
= qdict_next(bs
->options
, e
)) {
3197 if (!qdict_haskey(new_opts
, e
->key
) &&
3198 !is_str_in_list(e
->key
, common_options
) &&
3199 !is_str_in_list(e
->key
, bs
->drv
->mutable_opts
)) {
3200 error_setg(errp
, "Option '%s' cannot be reset "
3201 "to its default value", e
->key
);
3210 * Returns true if @child can be reached recursively from @bs
3212 static bool bdrv_recurse_has_child(BlockDriverState
*bs
,
3213 BlockDriverState
*child
)
3221 QLIST_FOREACH(c
, &bs
->children
, next
) {
3222 if (bdrv_recurse_has_child(c
->bs
, child
)) {
3231 * Adds a BlockDriverState to a simple queue for an atomic, transactional
3232 * reopen of multiple devices.
3234 * bs_queue can either be an existing BlockReopenQueue that has had QSIMPLE_INIT
3235 * already performed, or alternatively may be NULL a new BlockReopenQueue will
3236 * be created and initialized. This newly created BlockReopenQueue should be
3237 * passed back in for subsequent calls that are intended to be of the same
3240 * bs is the BlockDriverState to add to the reopen queue.
3242 * options contains the changed options for the associated bs
3243 * (the BlockReopenQueue takes ownership)
3245 * flags contains the open flags for the associated bs
3247 * returns a pointer to bs_queue, which is either the newly allocated
3248 * bs_queue, or the existing bs_queue being used.
3250 * bs must be drained between bdrv_reopen_queue() and bdrv_reopen_multiple().
3252 static BlockReopenQueue
*bdrv_reopen_queue_child(BlockReopenQueue
*bs_queue
,
3253 BlockDriverState
*bs
,
3255 const BdrvChildRole
*role
,
3256 QDict
*parent_options
,
3262 BlockReopenQueueEntry
*bs_entry
;
3264 QDict
*old_options
, *explicit_options
, *options_copy
;
3268 /* Make sure that the caller remembered to use a drained section. This is
3269 * important to avoid graph changes between the recursive queuing here and
3270 * bdrv_reopen_multiple(). */
3271 assert(bs
->quiesce_counter
> 0);
3273 if (bs_queue
== NULL
) {
3274 bs_queue
= g_new0(BlockReopenQueue
, 1);
3275 QSIMPLEQ_INIT(bs_queue
);
3279 options
= qdict_new();
3282 /* Check if this BlockDriverState is already in the queue */
3283 QSIMPLEQ_FOREACH(bs_entry
, bs_queue
, entry
) {
3284 if (bs
== bs_entry
->state
.bs
) {
3290 * Precedence of options:
3291 * 1. Explicitly passed in options (highest)
3292 * 2. Retained from explicitly set options of bs
3293 * 3. Inherited from parent node
3294 * 4. Retained from effective options of bs
3297 /* Old explicitly set values (don't overwrite by inherited value) */
3298 if (bs_entry
|| keep_old_opts
) {
3299 old_options
= qdict_clone_shallow(bs_entry
?
3300 bs_entry
->state
.explicit_options
:
3301 bs
->explicit_options
);
3302 bdrv_join_options(bs
, options
, old_options
);
3303 qobject_unref(old_options
);
3306 explicit_options
= qdict_clone_shallow(options
);
3308 /* Inherit from parent node */
3309 if (parent_options
) {
3311 role
->inherit_options(&flags
, options
, parent_flags
, parent_options
);
3313 flags
= bdrv_get_flags(bs
);
3316 if (keep_old_opts
) {
3317 /* Old values are used for options that aren't set yet */
3318 old_options
= qdict_clone_shallow(bs
->options
);
3319 bdrv_join_options(bs
, options
, old_options
);
3320 qobject_unref(old_options
);
3323 /* We have the final set of options so let's update the flags */
3324 options_copy
= qdict_clone_shallow(options
);
3325 opts
= qemu_opts_create(&bdrv_runtime_opts
, NULL
, 0, &error_abort
);
3326 qemu_opts_absorb_qdict(opts
, options_copy
, NULL
);
3327 update_flags_from_options(&flags
, opts
);
3328 qemu_opts_del(opts
);
3329 qobject_unref(options_copy
);
3331 /* bdrv_open_inherit() sets and clears some additional flags internally */
3332 flags
&= ~BDRV_O_PROTOCOL
;
3333 if (flags
& BDRV_O_RDWR
) {
3334 flags
|= BDRV_O_ALLOW_RDWR
;
3338 bs_entry
= g_new0(BlockReopenQueueEntry
, 1);
3339 QSIMPLEQ_INSERT_TAIL(bs_queue
, bs_entry
, entry
);
3341 qobject_unref(bs_entry
->state
.options
);
3342 qobject_unref(bs_entry
->state
.explicit_options
);
3345 bs_entry
->state
.bs
= bs
;
3346 bs_entry
->state
.options
= options
;
3347 bs_entry
->state
.explicit_options
= explicit_options
;
3348 bs_entry
->state
.flags
= flags
;
3350 /* This needs to be overwritten in bdrv_reopen_prepare() */
3351 bs_entry
->state
.perm
= UINT64_MAX
;
3352 bs_entry
->state
.shared_perm
= 0;
3355 * If keep_old_opts is false then it means that unspecified
3356 * options must be reset to their original value. We don't allow
3357 * resetting 'backing' but we need to know if the option is
3358 * missing in order to decide if we have to return an error.
3360 if (!keep_old_opts
) {
3361 bs_entry
->state
.backing_missing
=
3362 !qdict_haskey(options
, "backing") &&
3363 !qdict_haskey(options
, "backing.driver");
3366 QLIST_FOREACH(child
, &bs
->children
, next
) {
3367 QDict
*new_child_options
= NULL
;
3368 bool child_keep_old
= keep_old_opts
;
3370 /* reopen can only change the options of block devices that were
3371 * implicitly created and inherited options. For other (referenced)
3372 * block devices, a syntax like "backing.foo" results in an error. */
3373 if (child
->bs
->inherits_from
!= bs
) {
3377 /* Check if the options contain a child reference */
3378 if (qdict_haskey(options
, child
->name
)) {
3379 const char *childref
= qdict_get_try_str(options
, child
->name
);
3381 * The current child must not be reopened if the child
3382 * reference is null or points to a different node.
3384 if (g_strcmp0(childref
, child
->bs
->node_name
)) {
3388 * If the child reference points to the current child then
3389 * reopen it with its existing set of options (note that
3390 * it can still inherit new options from the parent).
3392 child_keep_old
= true;
3394 /* Extract child options ("child-name.*") */
3395 char *child_key_dot
= g_strdup_printf("%s.", child
->name
);
3396 qdict_extract_subqdict(explicit_options
, NULL
, child_key_dot
);
3397 qdict_extract_subqdict(options
, &new_child_options
, child_key_dot
);
3398 g_free(child_key_dot
);
3401 bdrv_reopen_queue_child(bs_queue
, child
->bs
, new_child_options
,
3402 child
->role
, options
, flags
, child_keep_old
);
3408 BlockReopenQueue
*bdrv_reopen_queue(BlockReopenQueue
*bs_queue
,
3409 BlockDriverState
*bs
,
3410 QDict
*options
, bool keep_old_opts
)
3412 return bdrv_reopen_queue_child(bs_queue
, bs
, options
, NULL
, NULL
, 0,
3417 * Reopen multiple BlockDriverStates atomically & transactionally.
3419 * The queue passed in (bs_queue) must have been built up previous
3420 * via bdrv_reopen_queue().
3422 * Reopens all BDS specified in the queue, with the appropriate
3423 * flags. All devices are prepared for reopen, and failure of any
3424 * device will cause all device changes to be abandoned, and intermediate
3427 * If all devices prepare successfully, then the changes are committed
3430 * All affected nodes must be drained between bdrv_reopen_queue() and
3431 * bdrv_reopen_multiple().
3433 int bdrv_reopen_multiple(BlockReopenQueue
*bs_queue
, Error
**errp
)
3436 BlockReopenQueueEntry
*bs_entry
, *next
;
3438 assert(bs_queue
!= NULL
);
3440 QSIMPLEQ_FOREACH(bs_entry
, bs_queue
, entry
) {
3441 assert(bs_entry
->state
.bs
->quiesce_counter
> 0);
3442 if (bdrv_reopen_prepare(&bs_entry
->state
, bs_queue
, errp
)) {
3445 bs_entry
->prepared
= true;
3448 QSIMPLEQ_FOREACH(bs_entry
, bs_queue
, entry
) {
3449 BDRVReopenState
*state
= &bs_entry
->state
;
3450 ret
= bdrv_check_perm(state
->bs
, bs_queue
, state
->perm
,
3451 state
->shared_perm
, NULL
, NULL
, errp
);
3455 /* Check if new_backing_bs would accept the new permissions */
3456 if (state
->replace_backing_bs
&& state
->new_backing_bs
) {
3457 uint64_t nperm
, nshared
;
3458 bdrv_child_perm(state
->bs
, state
->new_backing_bs
,
3459 NULL
, &child_backing
, bs_queue
,
3460 state
->perm
, state
->shared_perm
,
3462 ret
= bdrv_check_update_perm(state
->new_backing_bs
, NULL
,
3463 nperm
, nshared
, NULL
, NULL
, errp
);
3468 bs_entry
->perms_checked
= true;
3471 /* If we reach this point, we have success and just need to apply the
3474 QSIMPLEQ_FOREACH(bs_entry
, bs_queue
, entry
) {
3475 bdrv_reopen_commit(&bs_entry
->state
);
3480 QSIMPLEQ_FOREACH_SAFE(bs_entry
, bs_queue
, entry
, next
) {
3481 BDRVReopenState
*state
= &bs_entry
->state
;
3483 if (!bs_entry
->perms_checked
) {
3488 bdrv_set_perm(state
->bs
, state
->perm
, state
->shared_perm
);
3490 bdrv_abort_perm_update(state
->bs
);
3491 if (state
->replace_backing_bs
&& state
->new_backing_bs
) {
3492 bdrv_abort_perm_update(state
->new_backing_bs
);
3497 QSIMPLEQ_FOREACH_SAFE(bs_entry
, bs_queue
, entry
, next
) {
3499 if (bs_entry
->prepared
) {
3500 bdrv_reopen_abort(&bs_entry
->state
);
3502 qobject_unref(bs_entry
->state
.explicit_options
);
3503 qobject_unref(bs_entry
->state
.options
);
3505 if (bs_entry
->state
.new_backing_bs
) {
3506 bdrv_unref(bs_entry
->state
.new_backing_bs
);
3515 int bdrv_reopen_set_read_only(BlockDriverState
*bs
, bool read_only
,
3519 BlockReopenQueue
*queue
;
3520 QDict
*opts
= qdict_new();
3522 qdict_put_bool(opts
, BDRV_OPT_READ_ONLY
, read_only
);
3524 bdrv_subtree_drained_begin(bs
);
3525 queue
= bdrv_reopen_queue(NULL
, bs
, opts
, true);
3526 ret
= bdrv_reopen_multiple(queue
, errp
);
3527 bdrv_subtree_drained_end(bs
);
3532 static BlockReopenQueueEntry
*find_parent_in_reopen_queue(BlockReopenQueue
*q
,
3535 BlockReopenQueueEntry
*entry
;
3537 QSIMPLEQ_FOREACH(entry
, q
, entry
) {
3538 BlockDriverState
*bs
= entry
->state
.bs
;
3541 QLIST_FOREACH(child
, &bs
->children
, next
) {
3551 static void bdrv_reopen_perm(BlockReopenQueue
*q
, BlockDriverState
*bs
,
3552 uint64_t *perm
, uint64_t *shared
)
3555 BlockReopenQueueEntry
*parent
;
3556 uint64_t cumulative_perms
= 0;
3557 uint64_t cumulative_shared_perms
= BLK_PERM_ALL
;
3559 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
3560 parent
= find_parent_in_reopen_queue(q
, c
);
3562 cumulative_perms
|= c
->perm
;
3563 cumulative_shared_perms
&= c
->shared_perm
;
3565 uint64_t nperm
, nshared
;
3567 bdrv_child_perm(parent
->state
.bs
, bs
, c
, c
->role
, q
,
3568 parent
->state
.perm
, parent
->state
.shared_perm
,
3571 cumulative_perms
|= nperm
;
3572 cumulative_shared_perms
&= nshared
;
3575 *perm
= cumulative_perms
;
3576 *shared
= cumulative_shared_perms
;
3580 * Take a BDRVReopenState and check if the value of 'backing' in the
3581 * reopen_state->options QDict is valid or not.
3583 * If 'backing' is missing from the QDict then return 0.
3585 * If 'backing' contains the node name of the backing file of
3586 * reopen_state->bs then return 0.
3588 * If 'backing' contains a different node name (or is null) then check
3589 * whether the current backing file can be replaced with the new one.
3590 * If that's the case then reopen_state->replace_backing_bs is set to
3591 * true and reopen_state->new_backing_bs contains a pointer to the new
3592 * backing BlockDriverState (or NULL).
3594 * Return 0 on success, otherwise return < 0 and set @errp.
3596 static int bdrv_reopen_parse_backing(BDRVReopenState
*reopen_state
,
3599 BlockDriverState
*bs
= reopen_state
->bs
;
3600 BlockDriverState
*overlay_bs
, *new_backing_bs
;
3604 value
= qdict_get(reopen_state
->options
, "backing");
3605 if (value
== NULL
) {
3609 switch (qobject_type(value
)) {
3611 new_backing_bs
= NULL
;
3614 str
= qobject_get_try_str(value
);
3615 new_backing_bs
= bdrv_lookup_bs(NULL
, str
, errp
);
3616 if (new_backing_bs
== NULL
) {
3618 } else if (bdrv_recurse_has_child(new_backing_bs
, bs
)) {
3619 error_setg(errp
, "Making '%s' a backing file of '%s' "
3620 "would create a cycle", str
, bs
->node_name
);
3625 /* 'backing' does not allow any other data type */
3626 g_assert_not_reached();
3630 * TODO: before removing the x- prefix from x-blockdev-reopen we
3631 * should move the new backing file into the right AioContext
3632 * instead of returning an error.
3634 if (new_backing_bs
) {
3635 if (bdrv_get_aio_context(new_backing_bs
) != bdrv_get_aio_context(bs
)) {
3636 error_setg(errp
, "Cannot use a new backing file "
3637 "with a different AioContext");
3643 * Find the "actual" backing file by skipping all links that point
3644 * to an implicit node, if any (e.g. a commit filter node).
3647 while (backing_bs(overlay_bs
) && backing_bs(overlay_bs
)->implicit
) {
3648 overlay_bs
= backing_bs(overlay_bs
);
3651 /* If we want to replace the backing file we need some extra checks */
3652 if (new_backing_bs
!= backing_bs(overlay_bs
)) {
3653 /* Check for implicit nodes between bs and its backing file */
3654 if (bs
!= overlay_bs
) {
3655 error_setg(errp
, "Cannot change backing link if '%s' has "
3656 "an implicit backing file", bs
->node_name
);
3659 /* Check if the backing link that we want to replace is frozen */
3660 if (bdrv_is_backing_chain_frozen(overlay_bs
, backing_bs(overlay_bs
),
3664 reopen_state
->replace_backing_bs
= true;
3665 if (new_backing_bs
) {
3666 bdrv_ref(new_backing_bs
);
3667 reopen_state
->new_backing_bs
= new_backing_bs
;
3675 * Prepares a BlockDriverState for reopen. All changes are staged in the
3676 * 'opaque' field of the BDRVReopenState, which is used and allocated by
3677 * the block driver layer .bdrv_reopen_prepare()
3679 * bs is the BlockDriverState to reopen
3680 * flags are the new open flags
3681 * queue is the reopen queue
3683 * Returns 0 on success, non-zero on error. On error errp will be set
3686 * On failure, bdrv_reopen_abort() will be called to clean up any data.
3687 * It is the responsibility of the caller to then call the abort() or
3688 * commit() for any other BDS that have been left in a prepare() state
3691 int bdrv_reopen_prepare(BDRVReopenState
*reopen_state
, BlockReopenQueue
*queue
,
3696 Error
*local_err
= NULL
;
3699 QDict
*orig_reopen_opts
;
3700 char *discard
= NULL
;
3702 bool drv_prepared
= false;
3704 assert(reopen_state
!= NULL
);
3705 assert(reopen_state
->bs
->drv
!= NULL
);
3706 drv
= reopen_state
->bs
->drv
;
3708 /* This function and each driver's bdrv_reopen_prepare() remove
3709 * entries from reopen_state->options as they are processed, so
3710 * we need to make a copy of the original QDict. */
3711 orig_reopen_opts
= qdict_clone_shallow(reopen_state
->options
);
3713 /* Process generic block layer options */
3714 opts
= qemu_opts_create(&bdrv_runtime_opts
, NULL
, 0, &error_abort
);
3715 qemu_opts_absorb_qdict(opts
, reopen_state
->options
, &local_err
);
3717 error_propagate(errp
, local_err
);
3722 /* This was already called in bdrv_reopen_queue_child() so the flags
3723 * are up-to-date. This time we simply want to remove the options from
3724 * QemuOpts in order to indicate that they have been processed. */
3725 old_flags
= reopen_state
->flags
;
3726 update_flags_from_options(&reopen_state
->flags
, opts
);
3727 assert(old_flags
== reopen_state
->flags
);
3729 discard
= qemu_opt_get_del(opts
, BDRV_OPT_DISCARD
);
3730 if (discard
!= NULL
) {
3731 if (bdrv_parse_discard_flags(discard
, &reopen_state
->flags
) != 0) {
3732 error_setg(errp
, "Invalid discard option");
3738 reopen_state
->detect_zeroes
=
3739 bdrv_parse_detect_zeroes(opts
, reopen_state
->flags
, &local_err
);
3741 error_propagate(errp
, local_err
);
3746 /* All other options (including node-name and driver) must be unchanged.
3747 * Put them back into the QDict, so that they are checked at the end
3748 * of this function. */
3749 qemu_opts_to_qdict(opts
, reopen_state
->options
);
3751 /* If we are to stay read-only, do not allow permission change
3752 * to r/w. Attempting to set to r/w may fail if either BDRV_O_ALLOW_RDWR is
3753 * not set, or if the BDS still has copy_on_read enabled */
3754 read_only
= !(reopen_state
->flags
& BDRV_O_RDWR
);
3755 ret
= bdrv_can_set_read_only(reopen_state
->bs
, read_only
, true, &local_err
);
3757 error_propagate(errp
, local_err
);
3761 /* Calculate required permissions after reopening */
3762 bdrv_reopen_perm(queue
, reopen_state
->bs
,
3763 &reopen_state
->perm
, &reopen_state
->shared_perm
);
3765 ret
= bdrv_flush(reopen_state
->bs
);
3767 error_setg_errno(errp
, -ret
, "Error flushing drive");
3771 if (drv
->bdrv_reopen_prepare
) {
3773 * If a driver-specific option is missing, it means that we
3774 * should reset it to its default value.
3775 * But not all options allow that, so we need to check it first.
3777 ret
= bdrv_reset_options_allowed(reopen_state
->bs
,
3778 reopen_state
->options
, errp
);
3783 ret
= drv
->bdrv_reopen_prepare(reopen_state
, queue
, &local_err
);
3785 if (local_err
!= NULL
) {
3786 error_propagate(errp
, local_err
);
3788 bdrv_refresh_filename(reopen_state
->bs
);
3789 error_setg(errp
, "failed while preparing to reopen image '%s'",
3790 reopen_state
->bs
->filename
);
3795 /* It is currently mandatory to have a bdrv_reopen_prepare()
3796 * handler for each supported drv. */
3797 error_setg(errp
, "Block format '%s' used by node '%s' "
3798 "does not support reopening files", drv
->format_name
,
3799 bdrv_get_device_or_node_name(reopen_state
->bs
));
3804 drv_prepared
= true;
3807 * We must provide the 'backing' option if the BDS has a backing
3808 * file or if the image file has a backing file name as part of
3809 * its metadata. Otherwise the 'backing' option can be omitted.
3811 if (drv
->supports_backing
&& reopen_state
->backing_missing
&&
3812 (backing_bs(reopen_state
->bs
) || reopen_state
->bs
->backing_file
[0])) {
3813 error_setg(errp
, "backing is missing for '%s'",
3814 reopen_state
->bs
->node_name
);
3820 * Allow changing the 'backing' option. The new value can be
3821 * either a reference to an existing node (using its node name)
3822 * or NULL to simply detach the current backing file.
3824 ret
= bdrv_reopen_parse_backing(reopen_state
, errp
);
3828 qdict_del(reopen_state
->options
, "backing");
3830 /* Options that are not handled are only okay if they are unchanged
3831 * compared to the old state. It is expected that some options are only
3832 * used for the initial open, but not reopen (e.g. filename) */
3833 if (qdict_size(reopen_state
->options
)) {
3834 const QDictEntry
*entry
= qdict_first(reopen_state
->options
);
3837 QObject
*new = entry
->value
;
3838 QObject
*old
= qdict_get(reopen_state
->bs
->options
, entry
->key
);
3840 /* Allow child references (child_name=node_name) as long as they
3841 * point to the current child (i.e. everything stays the same). */
3842 if (qobject_type(new) == QTYPE_QSTRING
) {
3844 QLIST_FOREACH(child
, &reopen_state
->bs
->children
, next
) {
3845 if (!strcmp(child
->name
, entry
->key
)) {
3851 const char *str
= qobject_get_try_str(new);
3852 if (!strcmp(child
->bs
->node_name
, str
)) {
3853 continue; /* Found child with this name, skip option */
3859 * TODO: When using -drive to specify blockdev options, all values
3860 * will be strings; however, when using -blockdev, blockdev-add or
3861 * filenames using the json:{} pseudo-protocol, they will be
3863 * In contrast, reopening options are (currently) always strings
3864 * (because you can only specify them through qemu-io; all other
3865 * callers do not specify any options).
3866 * Therefore, when using anything other than -drive to create a BDS,
3867 * this cannot detect non-string options as unchanged, because
3868 * qobject_is_equal() always returns false for objects of different
3869 * type. In the future, this should be remedied by correctly typing
3870 * all options. For now, this is not too big of an issue because
3871 * the user can simply omit options which cannot be changed anyway,
3872 * so they will stay unchanged.
3874 if (!qobject_is_equal(new, old
)) {
3875 error_setg(errp
, "Cannot change the option '%s'", entry
->key
);
3879 } while ((entry
= qdict_next(reopen_state
->options
, entry
)));
3884 /* Restore the original reopen_state->options QDict */
3885 qobject_unref(reopen_state
->options
);
3886 reopen_state
->options
= qobject_ref(orig_reopen_opts
);
3889 if (ret
< 0 && drv_prepared
) {
3890 /* drv->bdrv_reopen_prepare() has succeeded, so we need to
3891 * call drv->bdrv_reopen_abort() before signaling an error
3892 * (bdrv_reopen_multiple() will not call bdrv_reopen_abort()
3893 * when the respective bdrv_reopen_prepare() has failed) */
3894 if (drv
->bdrv_reopen_abort
) {
3895 drv
->bdrv_reopen_abort(reopen_state
);
3898 qemu_opts_del(opts
);
3899 qobject_unref(orig_reopen_opts
);
3905 * Takes the staged changes for the reopen from bdrv_reopen_prepare(), and
3906 * makes them final by swapping the staging BlockDriverState contents into
3907 * the active BlockDriverState contents.
3909 void bdrv_reopen_commit(BDRVReopenState
*reopen_state
)
3912 BlockDriverState
*bs
;
3914 bool old_can_write
, new_can_write
;
3916 assert(reopen_state
!= NULL
);
3917 bs
= reopen_state
->bs
;
3919 assert(drv
!= NULL
);
3922 !bdrv_is_read_only(bs
) && !(bdrv_get_flags(bs
) & BDRV_O_INACTIVE
);
3924 /* If there are any driver level actions to take */
3925 if (drv
->bdrv_reopen_commit
) {
3926 drv
->bdrv_reopen_commit(reopen_state
);
3929 /* set BDS specific flags now */
3930 qobject_unref(bs
->explicit_options
);
3931 qobject_unref(bs
->options
);
3933 bs
->explicit_options
= reopen_state
->explicit_options
;
3934 bs
->options
= reopen_state
->options
;
3935 bs
->open_flags
= reopen_state
->flags
;
3936 bs
->read_only
= !(reopen_state
->flags
& BDRV_O_RDWR
);
3937 bs
->detect_zeroes
= reopen_state
->detect_zeroes
;
3939 if (reopen_state
->replace_backing_bs
) {
3940 qdict_del(bs
->explicit_options
, "backing");
3941 qdict_del(bs
->options
, "backing");
3944 /* Remove child references from bs->options and bs->explicit_options.
3945 * Child options were already removed in bdrv_reopen_queue_child() */
3946 QLIST_FOREACH(child
, &bs
->children
, next
) {
3947 qdict_del(bs
->explicit_options
, child
->name
);
3948 qdict_del(bs
->options
, child
->name
);
3952 * Change the backing file if a new one was specified. We do this
3953 * after updating bs->options, so bdrv_refresh_filename() (called
3954 * from bdrv_set_backing_hd()) has the new values.
3956 if (reopen_state
->replace_backing_bs
) {
3957 BlockDriverState
*old_backing_bs
= backing_bs(bs
);
3958 assert(!old_backing_bs
|| !old_backing_bs
->implicit
);
3959 /* Abort the permission update on the backing bs we're detaching */
3960 if (old_backing_bs
) {
3961 bdrv_abort_perm_update(old_backing_bs
);
3963 bdrv_set_backing_hd(bs
, reopen_state
->new_backing_bs
, &error_abort
);
3966 bdrv_refresh_limits(bs
, NULL
);
3969 !bdrv_is_read_only(bs
) && !(bdrv_get_flags(bs
) & BDRV_O_INACTIVE
);
3970 if (!old_can_write
&& new_can_write
&& drv
->bdrv_reopen_bitmaps_rw
) {
3971 Error
*local_err
= NULL
;
3972 if (drv
->bdrv_reopen_bitmaps_rw(bs
, &local_err
) < 0) {
3973 /* This is not fatal, bitmaps just left read-only, so all following
3974 * writes will fail. User can remove read-only bitmaps to unblock
3977 error_reportf_err(local_err
,
3978 "%s: Failed to make dirty bitmaps writable: ",
3979 bdrv_get_node_name(bs
));
3985 * Abort the reopen, and delete and free the staged changes in
3988 void bdrv_reopen_abort(BDRVReopenState
*reopen_state
)
3992 assert(reopen_state
!= NULL
);
3993 drv
= reopen_state
->bs
->drv
;
3994 assert(drv
!= NULL
);
3996 if (drv
->bdrv_reopen_abort
) {
3997 drv
->bdrv_reopen_abort(reopen_state
);
4002 static void bdrv_close(BlockDriverState
*bs
)
4004 BdrvAioNotifier
*ban
, *ban_next
;
4005 BdrvChild
*child
, *next
;
4007 assert(!bs
->refcnt
);
4009 bdrv_drained_begin(bs
); /* complete I/O */
4011 bdrv_drain(bs
); /* in case flush left pending I/O */
4014 if (bs
->drv
->bdrv_close
) {
4015 bs
->drv
->bdrv_close(bs
);
4020 QLIST_FOREACH_SAFE(child
, &bs
->children
, next
, next
) {
4021 bdrv_unref_child(bs
, child
);
4028 atomic_set(&bs
->copy_on_read
, 0);
4029 bs
->backing_file
[0] = '\0';
4030 bs
->backing_format
[0] = '\0';
4031 bs
->total_sectors
= 0;
4032 bs
->encrypted
= false;
4034 qobject_unref(bs
->options
);
4035 qobject_unref(bs
->explicit_options
);
4037 bs
->explicit_options
= NULL
;
4038 qobject_unref(bs
->full_open_options
);
4039 bs
->full_open_options
= NULL
;
4041 bdrv_release_named_dirty_bitmaps(bs
);
4042 assert(QLIST_EMPTY(&bs
->dirty_bitmaps
));
4044 QLIST_FOREACH_SAFE(ban
, &bs
->aio_notifiers
, list
, ban_next
) {
4047 QLIST_INIT(&bs
->aio_notifiers
);
4048 bdrv_drained_end(bs
);
4051 void bdrv_close_all(void)
4053 assert(job_next(NULL
) == NULL
);
4054 nbd_export_close_all();
4056 /* Drop references from requests still in flight, such as canceled block
4057 * jobs whose AIO context has not been polled yet */
4060 blk_remove_all_bs();
4061 blockdev_close_all_bdrv_states();
4063 assert(QTAILQ_EMPTY(&all_bdrv_states
));
4066 static bool should_update_child(BdrvChild
*c
, BlockDriverState
*to
)
4072 if (c
->role
->stay_at_node
) {
4076 /* If the child @c belongs to the BDS @to, replacing the current
4077 * c->bs by @to would mean to create a loop.
4079 * Such a case occurs when appending a BDS to a backing chain.
4080 * For instance, imagine the following chain:
4082 * guest device -> node A -> further backing chain...
4084 * Now we create a new BDS B which we want to put on top of this
4085 * chain, so we first attach A as its backing node:
4090 * guest device -> node A -> further backing chain...
4092 * Finally we want to replace A by B. When doing that, we want to
4093 * replace all pointers to A by pointers to B -- except for the
4094 * pointer from B because (1) that would create a loop, and (2)
4095 * that pointer should simply stay intact:
4097 * guest device -> node B
4100 * node A -> further backing chain...
4102 * In general, when replacing a node A (c->bs) by a node B (@to),
4103 * if A is a child of B, that means we cannot replace A by B there
4104 * because that would create a loop. Silently detaching A from B
4105 * is also not really an option. So overall just leaving A in
4106 * place there is the most sensible choice.
4108 * We would also create a loop in any cases where @c is only
4109 * indirectly referenced by @to. Prevent this by returning false
4110 * if @c is found (by breadth-first search) anywhere in the whole
4115 found
= g_hash_table_new(NULL
, NULL
);
4116 g_hash_table_add(found
, to
);
4117 queue
= g_queue_new();
4118 g_queue_push_tail(queue
, to
);
4120 while (!g_queue_is_empty(queue
)) {
4121 BlockDriverState
*v
= g_queue_pop_head(queue
);
4124 QLIST_FOREACH(c2
, &v
->children
, next
) {
4130 if (g_hash_table_contains(found
, c2
->bs
)) {
4134 g_queue_push_tail(queue
, c2
->bs
);
4135 g_hash_table_add(found
, c2
->bs
);
4139 g_queue_free(queue
);
4140 g_hash_table_destroy(found
);
4145 void bdrv_replace_node(BlockDriverState
*from
, BlockDriverState
*to
,
4148 BdrvChild
*c
, *next
;
4149 GSList
*list
= NULL
, *p
;
4150 uint64_t old_perm
, old_shared
;
4151 uint64_t perm
= 0, shared
= BLK_PERM_ALL
;
4154 /* Make sure that @from doesn't go away until we have successfully attached
4155 * all of its parents to @to. */
4158 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
4159 bdrv_drained_begin(from
);
4161 /* Put all parents into @list and calculate their cumulative permissions */
4162 QLIST_FOREACH_SAFE(c
, &from
->parents
, next_parent
, next
) {
4163 assert(c
->bs
== from
);
4164 if (!should_update_child(c
, to
)) {
4168 error_setg(errp
, "Cannot change '%s' link to '%s'",
4169 c
->name
, from
->node_name
);
4172 list
= g_slist_prepend(list
, c
);
4174 shared
&= c
->shared_perm
;
4177 /* Check whether the required permissions can be granted on @to, ignoring
4178 * all BdrvChild in @list so that they can't block themselves. */
4179 ret
= bdrv_check_update_perm(to
, NULL
, perm
, shared
, list
, NULL
, errp
);
4181 bdrv_abort_perm_update(to
);
4185 /* Now actually perform the change. We performed the permission check for
4186 * all elements of @list at once, so set the permissions all at once at the
4188 for (p
= list
; p
!= NULL
; p
= p
->next
) {
4192 bdrv_replace_child_noperm(c
, to
);
4196 bdrv_get_cumulative_perm(to
, &old_perm
, &old_shared
);
4197 bdrv_set_perm(to
, old_perm
| perm
, old_shared
| shared
);
4201 bdrv_drained_end(from
);
4206 * Add new bs contents at the top of an image chain while the chain is
4207 * live, while keeping required fields on the top layer.
4209 * This will modify the BlockDriverState fields, and swap contents
4210 * between bs_new and bs_top. Both bs_new and bs_top are modified.
4212 * bs_new must not be attached to a BlockBackend.
4214 * This function does not create any image files.
4216 * bdrv_append() takes ownership of a bs_new reference and unrefs it because
4217 * that's what the callers commonly need. bs_new will be referenced by the old
4218 * parents of bs_top after bdrv_append() returns. If the caller needs to keep a
4219 * reference of its own, it must call bdrv_ref().
4221 void bdrv_append(BlockDriverState
*bs_new
, BlockDriverState
*bs_top
,
4224 Error
*local_err
= NULL
;
4226 bdrv_set_backing_hd(bs_new
, bs_top
, &local_err
);
4228 error_propagate(errp
, local_err
);
4232 bdrv_replace_node(bs_top
, bs_new
, &local_err
);
4234 error_propagate(errp
, local_err
);
4235 bdrv_set_backing_hd(bs_new
, NULL
, &error_abort
);
4239 /* bs_new is now referenced by its new parents, we don't need the
4240 * additional reference any more. */
4245 static void bdrv_delete(BlockDriverState
*bs
)
4247 assert(bdrv_op_blocker_is_empty(bs
));
4248 assert(!bs
->refcnt
);
4250 /* remove from list, if necessary */
4251 if (bs
->node_name
[0] != '\0') {
4252 QTAILQ_REMOVE(&graph_bdrv_states
, bs
, node_list
);
4254 QTAILQ_REMOVE(&all_bdrv_states
, bs
, bs_list
);
4262 * Run consistency checks on an image
4264 * Returns 0 if the check could be completed (it doesn't mean that the image is
4265 * free of errors) or -errno when an internal error occurred. The results of the
4266 * check are stored in res.
4268 static int coroutine_fn
bdrv_co_check(BlockDriverState
*bs
,
4269 BdrvCheckResult
*res
, BdrvCheckMode fix
)
4271 if (bs
->drv
== NULL
) {
4274 if (bs
->drv
->bdrv_co_check
== NULL
) {
4278 memset(res
, 0, sizeof(*res
));
4279 return bs
->drv
->bdrv_co_check(bs
, res
, fix
);
4282 typedef struct CheckCo
{
4283 BlockDriverState
*bs
;
4284 BdrvCheckResult
*res
;
4289 static void coroutine_fn
bdrv_check_co_entry(void *opaque
)
4291 CheckCo
*cco
= opaque
;
4292 cco
->ret
= bdrv_co_check(cco
->bs
, cco
->res
, cco
->fix
);
4296 int bdrv_check(BlockDriverState
*bs
,
4297 BdrvCheckResult
*res
, BdrvCheckMode fix
)
4303 .ret
= -EINPROGRESS
,
4307 if (qemu_in_coroutine()) {
4308 /* Fast-path if already in coroutine context */
4309 bdrv_check_co_entry(&cco
);
4311 co
= qemu_coroutine_create(bdrv_check_co_entry
, &cco
);
4312 bdrv_coroutine_enter(bs
, co
);
4313 BDRV_POLL_WHILE(bs
, cco
.ret
== -EINPROGRESS
);
4322 * -EINVAL - backing format specified, but no file
4323 * -ENOSPC - can't update the backing file because no space is left in the
4325 * -ENOTSUP - format driver doesn't support changing the backing file
4327 int bdrv_change_backing_file(BlockDriverState
*bs
,
4328 const char *backing_file
, const char *backing_fmt
)
4330 BlockDriver
*drv
= bs
->drv
;
4337 /* Backing file format doesn't make sense without a backing file */
4338 if (backing_fmt
&& !backing_file
) {
4342 if (drv
->bdrv_change_backing_file
!= NULL
) {
4343 ret
= drv
->bdrv_change_backing_file(bs
, backing_file
, backing_fmt
);
4349 pstrcpy(bs
->backing_file
, sizeof(bs
->backing_file
), backing_file
?: "");
4350 pstrcpy(bs
->backing_format
, sizeof(bs
->backing_format
), backing_fmt
?: "");
4351 pstrcpy(bs
->auto_backing_file
, sizeof(bs
->auto_backing_file
),
4352 backing_file
?: "");
4358 * Finds the image layer in the chain that has 'bs' as its backing file.
4360 * active is the current topmost image.
4362 * Returns NULL if bs is not found in active's image chain,
4363 * or if active == bs.
4365 * Returns the bottommost base image if bs == NULL.
4367 BlockDriverState
*bdrv_find_overlay(BlockDriverState
*active
,
4368 BlockDriverState
*bs
)
4370 while (active
&& bs
!= backing_bs(active
)) {
4371 active
= backing_bs(active
);
4377 /* Given a BDS, searches for the base layer. */
4378 BlockDriverState
*bdrv_find_base(BlockDriverState
*bs
)
4380 return bdrv_find_overlay(bs
, NULL
);
4384 * Return true if at least one of the backing links between @bs and
4385 * @base is frozen. @errp is set if that's the case.
4386 * @base must be reachable from @bs, or NULL.
4388 bool bdrv_is_backing_chain_frozen(BlockDriverState
*bs
, BlockDriverState
*base
,
4391 BlockDriverState
*i
;
4393 for (i
= bs
; i
!= base
; i
= backing_bs(i
)) {
4394 if (i
->backing
&& i
->backing
->frozen
) {
4395 error_setg(errp
, "Cannot change '%s' link from '%s' to '%s'",
4396 i
->backing
->name
, i
->node_name
,
4397 backing_bs(i
)->node_name
);
4406 * Freeze all backing links between @bs and @base.
4407 * If any of the links is already frozen the operation is aborted and
4408 * none of the links are modified.
4409 * @base must be reachable from @bs, or NULL.
4410 * Returns 0 on success. On failure returns < 0 and sets @errp.
4412 int bdrv_freeze_backing_chain(BlockDriverState
*bs
, BlockDriverState
*base
,
4415 BlockDriverState
*i
;
4417 if (bdrv_is_backing_chain_frozen(bs
, base
, errp
)) {
4421 for (i
= bs
; i
!= base
; i
= backing_bs(i
)) {
4422 if (i
->backing
&& backing_bs(i
)->never_freeze
) {
4423 error_setg(errp
, "Cannot freeze '%s' link to '%s'",
4424 i
->backing
->name
, backing_bs(i
)->node_name
);
4429 for (i
= bs
; i
!= base
; i
= backing_bs(i
)) {
4431 i
->backing
->frozen
= true;
4439 * Unfreeze all backing links between @bs and @base. The caller must
4440 * ensure that all links are frozen before using this function.
4441 * @base must be reachable from @bs, or NULL.
4443 void bdrv_unfreeze_backing_chain(BlockDriverState
*bs
, BlockDriverState
*base
)
4445 BlockDriverState
*i
;
4447 for (i
= bs
; i
!= base
; i
= backing_bs(i
)) {
4449 assert(i
->backing
->frozen
);
4450 i
->backing
->frozen
= false;
4456 * Drops images above 'base' up to and including 'top', and sets the image
4457 * above 'top' to have base as its backing file.
4459 * Requires that the overlay to 'top' is opened r/w, so that the backing file
4460 * information in 'bs' can be properly updated.
4462 * E.g., this will convert the following chain:
4463 * bottom <- base <- intermediate <- top <- active
4467 * bottom <- base <- active
4469 * It is allowed for bottom==base, in which case it converts:
4471 * base <- intermediate <- top <- active
4477 * If backing_file_str is non-NULL, it will be used when modifying top's
4478 * overlay image metadata.
4481 * if active == top, that is considered an error
4484 int bdrv_drop_intermediate(BlockDriverState
*top
, BlockDriverState
*base
,
4485 const char *backing_file_str
)
4487 BlockDriverState
*explicit_top
= top
;
4488 bool update_inherits_from
;
4489 BdrvChild
*c
, *next
;
4490 Error
*local_err
= NULL
;
4494 bdrv_subtree_drained_begin(top
);
4496 if (!top
->drv
|| !base
->drv
) {
4500 /* Make sure that base is in the backing chain of top */
4501 if (!bdrv_chain_contains(top
, base
)) {
4505 /* This function changes all links that point to top and makes
4506 * them point to base. Check that none of them is frozen. */
4507 QLIST_FOREACH(c
, &top
->parents
, next_parent
) {
4513 /* If 'base' recursively inherits from 'top' then we should set
4514 * base->inherits_from to top->inherits_from after 'top' and all
4515 * other intermediate nodes have been dropped.
4516 * If 'top' is an implicit node (e.g. "commit_top") we should skip
4517 * it because no one inherits from it. We use explicit_top for that. */
4518 while (explicit_top
&& explicit_top
->implicit
) {
4519 explicit_top
= backing_bs(explicit_top
);
4521 update_inherits_from
= bdrv_inherits_from_recursive(base
, explicit_top
);
4523 /* success - we can delete the intermediate states, and link top->base */
4524 /* TODO Check graph modification op blockers (BLK_PERM_GRAPH_MOD) once
4525 * we've figured out how they should work. */
4526 if (!backing_file_str
) {
4527 bdrv_refresh_filename(base
);
4528 backing_file_str
= base
->filename
;
4531 QLIST_FOREACH_SAFE(c
, &top
->parents
, next_parent
, next
) {
4532 /* Check whether we are allowed to switch c from top to base */
4533 GSList
*ignore_children
= g_slist_prepend(NULL
, c
);
4534 ret
= bdrv_check_update_perm(base
, NULL
, c
->perm
, c
->shared_perm
,
4535 ignore_children
, NULL
, &local_err
);
4536 g_slist_free(ignore_children
);
4538 error_report_err(local_err
);
4542 /* If so, update the backing file path in the image file */
4543 if (c
->role
->update_filename
) {
4544 ret
= c
->role
->update_filename(c
, base
, backing_file_str
,
4547 bdrv_abort_perm_update(base
);
4548 error_report_err(local_err
);
4553 /* Do the actual switch in the in-memory graph.
4554 * Completes bdrv_check_update_perm() transaction internally. */
4556 bdrv_replace_child(c
, base
);
4560 if (update_inherits_from
) {
4561 base
->inherits_from
= explicit_top
->inherits_from
;
4566 bdrv_subtree_drained_end(top
);
4572 * Length of a allocated file in bytes. Sparse files are counted by actual
4573 * allocated space. Return < 0 if error or unknown.
4575 int64_t bdrv_get_allocated_file_size(BlockDriverState
*bs
)
4577 BlockDriver
*drv
= bs
->drv
;
4581 if (drv
->bdrv_get_allocated_file_size
) {
4582 return drv
->bdrv_get_allocated_file_size(bs
);
4585 return bdrv_get_allocated_file_size(bs
->file
->bs
);
4592 * @drv: Format driver
4593 * @opts: Creation options for new image
4594 * @in_bs: Existing image containing data for new image (may be NULL)
4595 * @errp: Error object
4596 * Returns: A #BlockMeasureInfo (free using qapi_free_BlockMeasureInfo())
4599 * Calculate file size required to create a new image.
4601 * If @in_bs is given then space for allocated clusters and zero clusters
4602 * from that image are included in the calculation. If @opts contains a
4603 * backing file that is shared by @in_bs then backing clusters may be omitted
4604 * from the calculation.
4606 * If @in_bs is NULL then the calculation includes no allocated clusters
4607 * unless a preallocation option is given in @opts.
4609 * Note that @in_bs may use a different BlockDriver from @drv.
4611 * If an error occurs the @errp pointer is set.
4613 BlockMeasureInfo
*bdrv_measure(BlockDriver
*drv
, QemuOpts
*opts
,
4614 BlockDriverState
*in_bs
, Error
**errp
)
4616 if (!drv
->bdrv_measure
) {
4617 error_setg(errp
, "Block driver '%s' does not support size measurement",
4622 return drv
->bdrv_measure(opts
, in_bs
, errp
);
4626 * Return number of sectors on success, -errno on error.
4628 int64_t bdrv_nb_sectors(BlockDriverState
*bs
)
4630 BlockDriver
*drv
= bs
->drv
;
4635 if (drv
->has_variable_length
) {
4636 int ret
= refresh_total_sectors(bs
, bs
->total_sectors
);
4641 return bs
->total_sectors
;
4645 * Return length in bytes on success, -errno on error.
4646 * The length is always a multiple of BDRV_SECTOR_SIZE.
4648 int64_t bdrv_getlength(BlockDriverState
*bs
)
4650 int64_t ret
= bdrv_nb_sectors(bs
);
4652 ret
= ret
> INT64_MAX
/ BDRV_SECTOR_SIZE
? -EFBIG
: ret
;
4653 return ret
< 0 ? ret
: ret
* BDRV_SECTOR_SIZE
;
4656 /* return 0 as number of sectors if no device present or error */
4657 void bdrv_get_geometry(BlockDriverState
*bs
, uint64_t *nb_sectors_ptr
)
4659 int64_t nb_sectors
= bdrv_nb_sectors(bs
);
4661 *nb_sectors_ptr
= nb_sectors
< 0 ? 0 : nb_sectors
;
4664 bool bdrv_is_sg(BlockDriverState
*bs
)
4669 bool bdrv_is_encrypted(BlockDriverState
*bs
)
4671 if (bs
->backing
&& bs
->backing
->bs
->encrypted
) {
4674 return bs
->encrypted
;
4677 const char *bdrv_get_format_name(BlockDriverState
*bs
)
4679 return bs
->drv
? bs
->drv
->format_name
: NULL
;
4682 static int qsort_strcmp(const void *a
, const void *b
)
4684 return strcmp(*(char *const *)a
, *(char *const *)b
);
4687 void bdrv_iterate_format(void (*it
)(void *opaque
, const char *name
),
4688 void *opaque
, bool read_only
)
4693 const char **formats
= NULL
;
4695 QLIST_FOREACH(drv
, &bdrv_drivers
, list
) {
4696 if (drv
->format_name
) {
4700 if (use_bdrv_whitelist
&& !bdrv_is_whitelisted(drv
, read_only
)) {
4704 while (formats
&& i
&& !found
) {
4705 found
= !strcmp(formats
[--i
], drv
->format_name
);
4709 formats
= g_renew(const char *, formats
, count
+ 1);
4710 formats
[count
++] = drv
->format_name
;
4715 for (i
= 0; i
< (int)ARRAY_SIZE(block_driver_modules
); i
++) {
4716 const char *format_name
= block_driver_modules
[i
].format_name
;
4722 if (use_bdrv_whitelist
&&
4723 !bdrv_format_is_whitelisted(format_name
, read_only
)) {
4727 while (formats
&& j
&& !found
) {
4728 found
= !strcmp(formats
[--j
], format_name
);
4732 formats
= g_renew(const char *, formats
, count
+ 1);
4733 formats
[count
++] = format_name
;
4738 qsort(formats
, count
, sizeof(formats
[0]), qsort_strcmp
);
4740 for (i
= 0; i
< count
; i
++) {
4741 it(opaque
, formats
[i
]);
4747 /* This function is to find a node in the bs graph */
4748 BlockDriverState
*bdrv_find_node(const char *node_name
)
4750 BlockDriverState
*bs
;
4754 QTAILQ_FOREACH(bs
, &graph_bdrv_states
, node_list
) {
4755 if (!strcmp(node_name
, bs
->node_name
)) {
4762 /* Put this QMP function here so it can access the static graph_bdrv_states. */
4763 BlockDeviceInfoList
*bdrv_named_nodes_list(Error
**errp
)
4765 BlockDeviceInfoList
*list
, *entry
;
4766 BlockDriverState
*bs
;
4769 QTAILQ_FOREACH(bs
, &graph_bdrv_states
, node_list
) {
4770 BlockDeviceInfo
*info
= bdrv_block_device_info(NULL
, bs
, errp
);
4772 qapi_free_BlockDeviceInfoList(list
);
4775 entry
= g_malloc0(sizeof(*entry
));
4776 entry
->value
= info
;
4784 #define QAPI_LIST_ADD(list, element) do { \
4785 typeof(list) _tmp = g_new(typeof(*(list)), 1); \
4786 _tmp->value = (element); \
4787 _tmp->next = (list); \
4791 typedef struct XDbgBlockGraphConstructor
{
4792 XDbgBlockGraph
*graph
;
4793 GHashTable
*graph_nodes
;
4794 } XDbgBlockGraphConstructor
;
4796 static XDbgBlockGraphConstructor
*xdbg_graph_new(void)
4798 XDbgBlockGraphConstructor
*gr
= g_new(XDbgBlockGraphConstructor
, 1);
4800 gr
->graph
= g_new0(XDbgBlockGraph
, 1);
4801 gr
->graph_nodes
= g_hash_table_new(NULL
, NULL
);
4806 static XDbgBlockGraph
*xdbg_graph_finalize(XDbgBlockGraphConstructor
*gr
)
4808 XDbgBlockGraph
*graph
= gr
->graph
;
4810 g_hash_table_destroy(gr
->graph_nodes
);
4816 static uintptr_t xdbg_graph_node_num(XDbgBlockGraphConstructor
*gr
, void *node
)
4818 uintptr_t ret
= (uintptr_t)g_hash_table_lookup(gr
->graph_nodes
, node
);
4825 * Start counting from 1, not 0, because 0 interferes with not-found (NULL)
4826 * answer of g_hash_table_lookup.
4828 ret
= g_hash_table_size(gr
->graph_nodes
) + 1;
4829 g_hash_table_insert(gr
->graph_nodes
, node
, (void *)ret
);
4834 static void xdbg_graph_add_node(XDbgBlockGraphConstructor
*gr
, void *node
,
4835 XDbgBlockGraphNodeType type
, const char *name
)
4837 XDbgBlockGraphNode
*n
;
4839 n
= g_new0(XDbgBlockGraphNode
, 1);
4841 n
->id
= xdbg_graph_node_num(gr
, node
);
4843 n
->name
= g_strdup(name
);
4845 QAPI_LIST_ADD(gr
->graph
->nodes
, n
);
4848 static void xdbg_graph_add_edge(XDbgBlockGraphConstructor
*gr
, void *parent
,
4849 const BdrvChild
*child
)
4853 BlockPermission num
;
4856 static const PermissionMap permissions
[] = {
4857 { BLK_PERM_CONSISTENT_READ
, BLOCK_PERMISSION_CONSISTENT_READ
},
4858 { BLK_PERM_WRITE
, BLOCK_PERMISSION_WRITE
},
4859 { BLK_PERM_WRITE_UNCHANGED
, BLOCK_PERMISSION_WRITE_UNCHANGED
},
4860 { BLK_PERM_RESIZE
, BLOCK_PERMISSION_RESIZE
},
4861 { BLK_PERM_GRAPH_MOD
, BLOCK_PERMISSION_GRAPH_MOD
},
4864 const PermissionMap
*p
;
4865 XDbgBlockGraphEdge
*edge
;
4867 QEMU_BUILD_BUG_ON(1UL << (ARRAY_SIZE(permissions
) - 1) != BLK_PERM_ALL
+ 1);
4869 edge
= g_new0(XDbgBlockGraphEdge
, 1);
4871 edge
->parent
= xdbg_graph_node_num(gr
, parent
);
4872 edge
->child
= xdbg_graph_node_num(gr
, child
->bs
);
4873 edge
->name
= g_strdup(child
->name
);
4875 for (p
= permissions
; p
->flag
; p
++) {
4876 if (p
->flag
& child
->perm
) {
4877 QAPI_LIST_ADD(edge
->perm
, p
->num
);
4879 if (p
->flag
& child
->shared_perm
) {
4880 QAPI_LIST_ADD(edge
->shared_perm
, p
->num
);
4884 QAPI_LIST_ADD(gr
->graph
->edges
, edge
);
4888 XDbgBlockGraph
*bdrv_get_xdbg_block_graph(Error
**errp
)
4892 BlockDriverState
*bs
;
4894 XDbgBlockGraphConstructor
*gr
= xdbg_graph_new();
4896 for (blk
= blk_all_next(NULL
); blk
; blk
= blk_all_next(blk
)) {
4897 char *allocated_name
= NULL
;
4898 const char *name
= blk_name(blk
);
4901 name
= allocated_name
= blk_get_attached_dev_id(blk
);
4903 xdbg_graph_add_node(gr
, blk
, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_BACKEND
,
4905 g_free(allocated_name
);
4906 if (blk_root(blk
)) {
4907 xdbg_graph_add_edge(gr
, blk
, blk_root(blk
));
4911 for (job
= block_job_next(NULL
); job
; job
= block_job_next(job
)) {
4914 xdbg_graph_add_node(gr
, job
, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_JOB
,
4916 for (el
= job
->nodes
; el
; el
= el
->next
) {
4917 xdbg_graph_add_edge(gr
, job
, (BdrvChild
*)el
->data
);
4921 QTAILQ_FOREACH(bs
, &graph_bdrv_states
, node_list
) {
4922 xdbg_graph_add_node(gr
, bs
, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_DRIVER
,
4924 QLIST_FOREACH(child
, &bs
->children
, next
) {
4925 xdbg_graph_add_edge(gr
, bs
, child
);
4929 return xdbg_graph_finalize(gr
);
4932 BlockDriverState
*bdrv_lookup_bs(const char *device
,
4933 const char *node_name
,
4937 BlockDriverState
*bs
;
4940 blk
= blk_by_name(device
);
4945 error_setg(errp
, "Device '%s' has no medium", device
);
4953 bs
= bdrv_find_node(node_name
);
4960 error_setg(errp
, "Cannot find device=%s nor node_name=%s",
4961 device
? device
: "",
4962 node_name
? node_name
: "");
4966 /* If 'base' is in the same chain as 'top', return true. Otherwise,
4967 * return false. If either argument is NULL, return false. */
4968 bool bdrv_chain_contains(BlockDriverState
*top
, BlockDriverState
*base
)
4970 while (top
&& top
!= base
) {
4971 top
= backing_bs(top
);
4977 BlockDriverState
*bdrv_next_node(BlockDriverState
*bs
)
4980 return QTAILQ_FIRST(&graph_bdrv_states
);
4982 return QTAILQ_NEXT(bs
, node_list
);
4985 BlockDriverState
*bdrv_next_all_states(BlockDriverState
*bs
)
4988 return QTAILQ_FIRST(&all_bdrv_states
);
4990 return QTAILQ_NEXT(bs
, bs_list
);
4993 const char *bdrv_get_node_name(const BlockDriverState
*bs
)
4995 return bs
->node_name
;
4998 const char *bdrv_get_parent_name(const BlockDriverState
*bs
)
5003 /* If multiple parents have a name, just pick the first one. */
5004 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
5005 if (c
->role
->get_name
) {
5006 name
= c
->role
->get_name(c
);
5007 if (name
&& *name
) {
5016 /* TODO check what callers really want: bs->node_name or blk_name() */
5017 const char *bdrv_get_device_name(const BlockDriverState
*bs
)
5019 return bdrv_get_parent_name(bs
) ?: "";
5022 /* This can be used to identify nodes that might not have a device
5023 * name associated. Since node and device names live in the same
5024 * namespace, the result is unambiguous. The exception is if both are
5025 * absent, then this returns an empty (non-null) string. */
5026 const char *bdrv_get_device_or_node_name(const BlockDriverState
*bs
)
5028 return bdrv_get_parent_name(bs
) ?: bs
->node_name
;
5031 int bdrv_get_flags(BlockDriverState
*bs
)
5033 return bs
->open_flags
;
5036 int bdrv_has_zero_init_1(BlockDriverState
*bs
)
5041 int bdrv_has_zero_init(BlockDriverState
*bs
)
5047 /* If BS is a copy on write image, it is initialized to
5048 the contents of the base image, which may not be zeroes. */
5052 if (bs
->drv
->bdrv_has_zero_init
) {
5053 return bs
->drv
->bdrv_has_zero_init(bs
);
5055 if (bs
->file
&& bs
->drv
->is_filter
) {
5056 return bdrv_has_zero_init(bs
->file
->bs
);
5063 bool bdrv_unallocated_blocks_are_zero(BlockDriverState
*bs
)
5065 BlockDriverInfo bdi
;
5071 if (bdrv_get_info(bs
, &bdi
) == 0) {
5072 return bdi
.unallocated_blocks_are_zero
;
5078 bool bdrv_can_write_zeroes_with_unmap(BlockDriverState
*bs
)
5080 if (!(bs
->open_flags
& BDRV_O_UNMAP
)) {
5084 return bs
->supported_zero_flags
& BDRV_REQ_MAY_UNMAP
;
5087 void bdrv_get_backing_filename(BlockDriverState
*bs
,
5088 char *filename
, int filename_size
)
5090 pstrcpy(filename
, filename_size
, bs
->backing_file
);
5093 int bdrv_get_info(BlockDriverState
*bs
, BlockDriverInfo
*bdi
)
5095 BlockDriver
*drv
= bs
->drv
;
5096 /* if bs->drv == NULL, bs is closed, so there's nothing to do here */
5100 if (!drv
->bdrv_get_info
) {
5101 if (bs
->file
&& drv
->is_filter
) {
5102 return bdrv_get_info(bs
->file
->bs
, bdi
);
5106 memset(bdi
, 0, sizeof(*bdi
));
5107 return drv
->bdrv_get_info(bs
, bdi
);
5110 ImageInfoSpecific
*bdrv_get_specific_info(BlockDriverState
*bs
,
5113 BlockDriver
*drv
= bs
->drv
;
5114 if (drv
&& drv
->bdrv_get_specific_info
) {
5115 return drv
->bdrv_get_specific_info(bs
, errp
);
5120 void bdrv_debug_event(BlockDriverState
*bs
, BlkdebugEvent event
)
5122 if (!bs
|| !bs
->drv
|| !bs
->drv
->bdrv_debug_event
) {
5126 bs
->drv
->bdrv_debug_event(bs
, event
);
5129 int bdrv_debug_breakpoint(BlockDriverState
*bs
, const char *event
,
5132 while (bs
&& bs
->drv
&& !bs
->drv
->bdrv_debug_breakpoint
) {
5133 bs
= bs
->file
? bs
->file
->bs
: NULL
;
5136 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_breakpoint
) {
5137 return bs
->drv
->bdrv_debug_breakpoint(bs
, event
, tag
);
5143 int bdrv_debug_remove_breakpoint(BlockDriverState
*bs
, const char *tag
)
5145 while (bs
&& bs
->drv
&& !bs
->drv
->bdrv_debug_remove_breakpoint
) {
5146 bs
= bs
->file
? bs
->file
->bs
: NULL
;
5149 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_remove_breakpoint
) {
5150 return bs
->drv
->bdrv_debug_remove_breakpoint(bs
, tag
);
5156 int bdrv_debug_resume(BlockDriverState
*bs
, const char *tag
)
5158 while (bs
&& (!bs
->drv
|| !bs
->drv
->bdrv_debug_resume
)) {
5159 bs
= bs
->file
? bs
->file
->bs
: NULL
;
5162 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_resume
) {
5163 return bs
->drv
->bdrv_debug_resume(bs
, tag
);
5169 bool bdrv_debug_is_suspended(BlockDriverState
*bs
, const char *tag
)
5171 while (bs
&& bs
->drv
&& !bs
->drv
->bdrv_debug_is_suspended
) {
5172 bs
= bs
->file
? bs
->file
->bs
: NULL
;
5175 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_is_suspended
) {
5176 return bs
->drv
->bdrv_debug_is_suspended(bs
, tag
);
5182 /* backing_file can either be relative, or absolute, or a protocol. If it is
5183 * relative, it must be relative to the chain. So, passing in bs->filename
5184 * from a BDS as backing_file should not be done, as that may be relative to
5185 * the CWD rather than the chain. */
5186 BlockDriverState
*bdrv_find_backing_image(BlockDriverState
*bs
,
5187 const char *backing_file
)
5189 char *filename_full
= NULL
;
5190 char *backing_file_full
= NULL
;
5191 char *filename_tmp
= NULL
;
5192 int is_protocol
= 0;
5193 BlockDriverState
*curr_bs
= NULL
;
5194 BlockDriverState
*retval
= NULL
;
5196 if (!bs
|| !bs
->drv
|| !backing_file
) {
5200 filename_full
= g_malloc(PATH_MAX
);
5201 backing_file_full
= g_malloc(PATH_MAX
);
5203 is_protocol
= path_has_protocol(backing_file
);
5205 for (curr_bs
= bs
; curr_bs
->backing
; curr_bs
= curr_bs
->backing
->bs
) {
5207 /* If either of the filename paths is actually a protocol, then
5208 * compare unmodified paths; otherwise make paths relative */
5209 if (is_protocol
|| path_has_protocol(curr_bs
->backing_file
)) {
5210 char *backing_file_full_ret
;
5212 if (strcmp(backing_file
, curr_bs
->backing_file
) == 0) {
5213 retval
= curr_bs
->backing
->bs
;
5216 /* Also check against the full backing filename for the image */
5217 backing_file_full_ret
= bdrv_get_full_backing_filename(curr_bs
,
5219 if (backing_file_full_ret
) {
5220 bool equal
= strcmp(backing_file
, backing_file_full_ret
) == 0;
5221 g_free(backing_file_full_ret
);
5223 retval
= curr_bs
->backing
->bs
;
5228 /* If not an absolute filename path, make it relative to the current
5229 * image's filename path */
5230 filename_tmp
= bdrv_make_absolute_filename(curr_bs
, backing_file
,
5232 /* We are going to compare canonicalized absolute pathnames */
5233 if (!filename_tmp
|| !realpath(filename_tmp
, filename_full
)) {
5234 g_free(filename_tmp
);
5237 g_free(filename_tmp
);
5239 /* We need to make sure the backing filename we are comparing against
5240 * is relative to the current image filename (or absolute) */
5241 filename_tmp
= bdrv_get_full_backing_filename(curr_bs
, NULL
);
5242 if (!filename_tmp
|| !realpath(filename_tmp
, backing_file_full
)) {
5243 g_free(filename_tmp
);
5246 g_free(filename_tmp
);
5248 if (strcmp(backing_file_full
, filename_full
) == 0) {
5249 retval
= curr_bs
->backing
->bs
;
5255 g_free(filename_full
);
5256 g_free(backing_file_full
);
5260 void bdrv_init(void)
5262 module_call_init(MODULE_INIT_BLOCK
);
5265 void bdrv_init_with_whitelist(void)
5267 use_bdrv_whitelist
= 1;
5271 static void coroutine_fn
bdrv_co_invalidate_cache(BlockDriverState
*bs
,
5274 BdrvChild
*child
, *parent
;
5275 uint64_t perm
, shared_perm
;
5276 Error
*local_err
= NULL
;
5278 BdrvDirtyBitmap
*bm
;
5284 if (!(bs
->open_flags
& BDRV_O_INACTIVE
)) {
5288 QLIST_FOREACH(child
, &bs
->children
, next
) {
5289 bdrv_co_invalidate_cache(child
->bs
, &local_err
);
5291 error_propagate(errp
, local_err
);
5297 * Update permissions, they may differ for inactive nodes.
5299 * Note that the required permissions of inactive images are always a
5300 * subset of the permissions required after activating the image. This
5301 * allows us to just get the permissions upfront without restricting
5302 * drv->bdrv_invalidate_cache().
5304 * It also means that in error cases, we don't have to try and revert to
5305 * the old permissions (which is an operation that could fail, too). We can
5306 * just keep the extended permissions for the next time that an activation
5307 * of the image is tried.
5309 bs
->open_flags
&= ~BDRV_O_INACTIVE
;
5310 bdrv_get_cumulative_perm(bs
, &perm
, &shared_perm
);
5311 ret
= bdrv_check_perm(bs
, NULL
, perm
, shared_perm
, NULL
, NULL
, &local_err
);
5313 bs
->open_flags
|= BDRV_O_INACTIVE
;
5314 error_propagate(errp
, local_err
);
5317 bdrv_set_perm(bs
, perm
, shared_perm
);
5319 if (bs
->drv
->bdrv_co_invalidate_cache
) {
5320 bs
->drv
->bdrv_co_invalidate_cache(bs
, &local_err
);
5322 bs
->open_flags
|= BDRV_O_INACTIVE
;
5323 error_propagate(errp
, local_err
);
5328 for (bm
= bdrv_dirty_bitmap_next(bs
, NULL
); bm
;
5329 bm
= bdrv_dirty_bitmap_next(bs
, bm
))
5331 bdrv_dirty_bitmap_set_migration(bm
, false);
5334 ret
= refresh_total_sectors(bs
, bs
->total_sectors
);
5336 bs
->open_flags
|= BDRV_O_INACTIVE
;
5337 error_setg_errno(errp
, -ret
, "Could not refresh total sector count");
5341 QLIST_FOREACH(parent
, &bs
->parents
, next_parent
) {
5342 if (parent
->role
->activate
) {
5343 parent
->role
->activate(parent
, &local_err
);
5345 bs
->open_flags
|= BDRV_O_INACTIVE
;
5346 error_propagate(errp
, local_err
);
5353 typedef struct InvalidateCacheCo
{
5354 BlockDriverState
*bs
;
5357 } InvalidateCacheCo
;
5359 static void coroutine_fn
bdrv_invalidate_cache_co_entry(void *opaque
)
5361 InvalidateCacheCo
*ico
= opaque
;
5362 bdrv_co_invalidate_cache(ico
->bs
, ico
->errp
);
5367 void bdrv_invalidate_cache(BlockDriverState
*bs
, Error
**errp
)
5370 InvalidateCacheCo ico
= {
5376 if (qemu_in_coroutine()) {
5377 /* Fast-path if already in coroutine context */
5378 bdrv_invalidate_cache_co_entry(&ico
);
5380 co
= qemu_coroutine_create(bdrv_invalidate_cache_co_entry
, &ico
);
5381 bdrv_coroutine_enter(bs
, co
);
5382 BDRV_POLL_WHILE(bs
, !ico
.done
);
5386 void bdrv_invalidate_cache_all(Error
**errp
)
5388 BlockDriverState
*bs
;
5389 Error
*local_err
= NULL
;
5390 BdrvNextIterator it
;
5392 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
5393 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
5395 aio_context_acquire(aio_context
);
5396 bdrv_invalidate_cache(bs
, &local_err
);
5397 aio_context_release(aio_context
);
5399 error_propagate(errp
, local_err
);
5400 bdrv_next_cleanup(&it
);
5406 static bool bdrv_has_bds_parent(BlockDriverState
*bs
, bool only_active
)
5410 QLIST_FOREACH(parent
, &bs
->parents
, next_parent
) {
5411 if (parent
->role
->parent_is_bds
) {
5412 BlockDriverState
*parent_bs
= parent
->opaque
;
5413 if (!only_active
|| !(parent_bs
->open_flags
& BDRV_O_INACTIVE
)) {
5422 static int bdrv_inactivate_recurse(BlockDriverState
*bs
)
5424 BdrvChild
*child
, *parent
;
5425 bool tighten_restrictions
;
5426 uint64_t perm
, shared_perm
;
5433 /* Make sure that we don't inactivate a child before its parent.
5434 * It will be covered by recursion from the yet active parent. */
5435 if (bdrv_has_bds_parent(bs
, true)) {
5439 assert(!(bs
->open_flags
& BDRV_O_INACTIVE
));
5441 /* Inactivate this node */
5442 if (bs
->drv
->bdrv_inactivate
) {
5443 ret
= bs
->drv
->bdrv_inactivate(bs
);
5449 QLIST_FOREACH(parent
, &bs
->parents
, next_parent
) {
5450 if (parent
->role
->inactivate
) {
5451 ret
= parent
->role
->inactivate(parent
);
5458 bs
->open_flags
|= BDRV_O_INACTIVE
;
5460 /* Update permissions, they may differ for inactive nodes */
5461 bdrv_get_cumulative_perm(bs
, &perm
, &shared_perm
);
5462 ret
= bdrv_check_perm(bs
, NULL
, perm
, shared_perm
, NULL
,
5463 &tighten_restrictions
, NULL
);
5464 assert(tighten_restrictions
== false);
5466 /* We only tried to loosen restrictions, so errors are not fatal */
5467 bdrv_abort_perm_update(bs
);
5469 bdrv_set_perm(bs
, perm
, shared_perm
);
5473 /* Recursively inactivate children */
5474 QLIST_FOREACH(child
, &bs
->children
, next
) {
5475 ret
= bdrv_inactivate_recurse(child
->bs
);
5484 int bdrv_inactivate_all(void)
5486 BlockDriverState
*bs
= NULL
;
5487 BdrvNextIterator it
;
5489 GSList
*aio_ctxs
= NULL
, *ctx
;
5491 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
5492 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
5494 if (!g_slist_find(aio_ctxs
, aio_context
)) {
5495 aio_ctxs
= g_slist_prepend(aio_ctxs
, aio_context
);
5496 aio_context_acquire(aio_context
);
5500 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
5501 /* Nodes with BDS parents are covered by recursion from the last
5502 * parent that gets inactivated. Don't inactivate them a second
5503 * time if that has already happened. */
5504 if (bdrv_has_bds_parent(bs
, false)) {
5507 ret
= bdrv_inactivate_recurse(bs
);
5509 bdrv_next_cleanup(&it
);
5515 for (ctx
= aio_ctxs
; ctx
!= NULL
; ctx
= ctx
->next
) {
5516 AioContext
*aio_context
= ctx
->data
;
5517 aio_context_release(aio_context
);
5519 g_slist_free(aio_ctxs
);
5524 /**************************************************************/
5525 /* removable device support */
5528 * Return TRUE if the media is present
5530 bool bdrv_is_inserted(BlockDriverState
*bs
)
5532 BlockDriver
*drv
= bs
->drv
;
5538 if (drv
->bdrv_is_inserted
) {
5539 return drv
->bdrv_is_inserted(bs
);
5541 QLIST_FOREACH(child
, &bs
->children
, next
) {
5542 if (!bdrv_is_inserted(child
->bs
)) {
5550 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
5552 void bdrv_eject(BlockDriverState
*bs
, bool eject_flag
)
5554 BlockDriver
*drv
= bs
->drv
;
5556 if (drv
&& drv
->bdrv_eject
) {
5557 drv
->bdrv_eject(bs
, eject_flag
);
5562 * Lock or unlock the media (if it is locked, the user won't be able
5563 * to eject it manually).
5565 void bdrv_lock_medium(BlockDriverState
*bs
, bool locked
)
5567 BlockDriver
*drv
= bs
->drv
;
5569 trace_bdrv_lock_medium(bs
, locked
);
5571 if (drv
&& drv
->bdrv_lock_medium
) {
5572 drv
->bdrv_lock_medium(bs
, locked
);
5576 /* Get a reference to bs */
5577 void bdrv_ref(BlockDriverState
*bs
)
5582 /* Release a previously grabbed reference to bs.
5583 * If after releasing, reference count is zero, the BlockDriverState is
5585 void bdrv_unref(BlockDriverState
*bs
)
5590 assert(bs
->refcnt
> 0);
5591 if (--bs
->refcnt
== 0) {
5596 struct BdrvOpBlocker
{
5598 QLIST_ENTRY(BdrvOpBlocker
) list
;
5601 bool bdrv_op_is_blocked(BlockDriverState
*bs
, BlockOpType op
, Error
**errp
)
5603 BdrvOpBlocker
*blocker
;
5604 assert((int) op
>= 0 && op
< BLOCK_OP_TYPE_MAX
);
5605 if (!QLIST_EMPTY(&bs
->op_blockers
[op
])) {
5606 blocker
= QLIST_FIRST(&bs
->op_blockers
[op
]);
5607 error_propagate_prepend(errp
, error_copy(blocker
->reason
),
5608 "Node '%s' is busy: ",
5609 bdrv_get_device_or_node_name(bs
));
5615 void bdrv_op_block(BlockDriverState
*bs
, BlockOpType op
, Error
*reason
)
5617 BdrvOpBlocker
*blocker
;
5618 assert((int) op
>= 0 && op
< BLOCK_OP_TYPE_MAX
);
5620 blocker
= g_new0(BdrvOpBlocker
, 1);
5621 blocker
->reason
= reason
;
5622 QLIST_INSERT_HEAD(&bs
->op_blockers
[op
], blocker
, list
);
5625 void bdrv_op_unblock(BlockDriverState
*bs
, BlockOpType op
, Error
*reason
)
5627 BdrvOpBlocker
*blocker
, *next
;
5628 assert((int) op
>= 0 && op
< BLOCK_OP_TYPE_MAX
);
5629 QLIST_FOREACH_SAFE(blocker
, &bs
->op_blockers
[op
], list
, next
) {
5630 if (blocker
->reason
== reason
) {
5631 QLIST_REMOVE(blocker
, list
);
5637 void bdrv_op_block_all(BlockDriverState
*bs
, Error
*reason
)
5640 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
5641 bdrv_op_block(bs
, i
, reason
);
5645 void bdrv_op_unblock_all(BlockDriverState
*bs
, Error
*reason
)
5648 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
5649 bdrv_op_unblock(bs
, i
, reason
);
5653 bool bdrv_op_blocker_is_empty(BlockDriverState
*bs
)
5657 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
5658 if (!QLIST_EMPTY(&bs
->op_blockers
[i
])) {
5665 void bdrv_img_create(const char *filename
, const char *fmt
,
5666 const char *base_filename
, const char *base_fmt
,
5667 char *options
, uint64_t img_size
, int flags
, bool quiet
,
5670 QemuOptsList
*create_opts
= NULL
;
5671 QemuOpts
*opts
= NULL
;
5672 const char *backing_fmt
, *backing_file
;
5674 BlockDriver
*drv
, *proto_drv
;
5675 Error
*local_err
= NULL
;
5678 /* Find driver and parse its options */
5679 drv
= bdrv_find_format(fmt
);
5681 error_setg(errp
, "Unknown file format '%s'", fmt
);
5685 proto_drv
= bdrv_find_protocol(filename
, true, errp
);
5690 if (!drv
->create_opts
) {
5691 error_setg(errp
, "Format driver '%s' does not support image creation",
5696 if (!proto_drv
->create_opts
) {
5697 error_setg(errp
, "Protocol driver '%s' does not support image creation",
5698 proto_drv
->format_name
);
5702 create_opts
= qemu_opts_append(create_opts
, drv
->create_opts
);
5703 create_opts
= qemu_opts_append(create_opts
, proto_drv
->create_opts
);
5705 /* Create parameter list with default values */
5706 opts
= qemu_opts_create(create_opts
, NULL
, 0, &error_abort
);
5707 qemu_opt_set_number(opts
, BLOCK_OPT_SIZE
, img_size
, &error_abort
);
5709 /* Parse -o options */
5711 qemu_opts_do_parse(opts
, options
, NULL
, &local_err
);
5717 if (base_filename
) {
5718 qemu_opt_set(opts
, BLOCK_OPT_BACKING_FILE
, base_filename
, &local_err
);
5720 error_setg(errp
, "Backing file not supported for file format '%s'",
5727 qemu_opt_set(opts
, BLOCK_OPT_BACKING_FMT
, base_fmt
, &local_err
);
5729 error_setg(errp
, "Backing file format not supported for file "
5730 "format '%s'", fmt
);
5735 backing_file
= qemu_opt_get(opts
, BLOCK_OPT_BACKING_FILE
);
5737 if (!strcmp(filename
, backing_file
)) {
5738 error_setg(errp
, "Error: Trying to create an image with the "
5739 "same filename as the backing file");
5744 backing_fmt
= qemu_opt_get(opts
, BLOCK_OPT_BACKING_FMT
);
5746 /* The size for the image must always be specified, unless we have a backing
5747 * file and we have not been forbidden from opening it. */
5748 size
= qemu_opt_get_size(opts
, BLOCK_OPT_SIZE
, img_size
);
5749 if (backing_file
&& !(flags
& BDRV_O_NO_BACKING
)) {
5750 BlockDriverState
*bs
;
5753 QDict
*backing_options
= NULL
;
5756 bdrv_get_full_backing_filename_from_filename(filename
, backing_file
,
5761 assert(full_backing
);
5763 /* backing files always opened read-only */
5765 back_flags
&= ~(BDRV_O_RDWR
| BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
);
5767 backing_options
= qdict_new();
5769 qdict_put_str(backing_options
, "driver", backing_fmt
);
5771 qdict_put_bool(backing_options
, BDRV_OPT_FORCE_SHARE
, true);
5773 bs
= bdrv_open(full_backing
, NULL
, backing_options
, back_flags
,
5775 g_free(full_backing
);
5776 if (!bs
&& size
!= -1) {
5777 /* Couldn't open BS, but we have a size, so it's nonfatal */
5778 warn_reportf_err(local_err
,
5779 "Could not verify backing image. "
5780 "This may become an error in future versions.\n");
5783 /* Couldn't open bs, do not have size */
5784 error_append_hint(&local_err
,
5785 "Could not open backing image to determine size.\n");
5789 /* Opened BS, have no size */
5790 size
= bdrv_getlength(bs
);
5792 error_setg_errno(errp
, -size
, "Could not get size of '%s'",
5797 qemu_opt_set_number(opts
, BLOCK_OPT_SIZE
, size
, &error_abort
);
5801 } /* (backing_file && !(flags & BDRV_O_NO_BACKING)) */
5804 error_setg(errp
, "Image creation needs a size parameter");
5809 printf("Formatting '%s', fmt=%s ", filename
, fmt
);
5810 qemu_opts_print(opts
, " ");
5814 ret
= bdrv_create(drv
, filename
, opts
, &local_err
);
5816 if (ret
== -EFBIG
) {
5817 /* This is generally a better message than whatever the driver would
5818 * deliver (especially because of the cluster_size_hint), since that
5819 * is most probably not much different from "image too large". */
5820 const char *cluster_size_hint
= "";
5821 if (qemu_opt_get_size(opts
, BLOCK_OPT_CLUSTER_SIZE
, 0)) {
5822 cluster_size_hint
= " (try using a larger cluster size)";
5824 error_setg(errp
, "The image size is too large for file format '%s'"
5825 "%s", fmt
, cluster_size_hint
);
5826 error_free(local_err
);
5831 qemu_opts_del(opts
);
5832 qemu_opts_free(create_opts
);
5833 error_propagate(errp
, local_err
);
5836 AioContext
*bdrv_get_aio_context(BlockDriverState
*bs
)
5838 return bs
? bs
->aio_context
: qemu_get_aio_context();
5841 void bdrv_coroutine_enter(BlockDriverState
*bs
, Coroutine
*co
)
5843 aio_co_enter(bdrv_get_aio_context(bs
), co
);
5846 static void bdrv_do_remove_aio_context_notifier(BdrvAioNotifier
*ban
)
5848 QLIST_REMOVE(ban
, list
);
5852 static void bdrv_detach_aio_context(BlockDriverState
*bs
)
5854 BdrvAioNotifier
*baf
, *baf_tmp
;
5856 assert(!bs
->walking_aio_notifiers
);
5857 bs
->walking_aio_notifiers
= true;
5858 QLIST_FOREACH_SAFE(baf
, &bs
->aio_notifiers
, list
, baf_tmp
) {
5860 bdrv_do_remove_aio_context_notifier(baf
);
5862 baf
->detach_aio_context(baf
->opaque
);
5865 /* Never mind iterating again to check for ->deleted. bdrv_close() will
5866 * remove remaining aio notifiers if we aren't called again.
5868 bs
->walking_aio_notifiers
= false;
5870 if (bs
->drv
&& bs
->drv
->bdrv_detach_aio_context
) {
5871 bs
->drv
->bdrv_detach_aio_context(bs
);
5874 if (bs
->quiesce_counter
) {
5875 aio_enable_external(bs
->aio_context
);
5877 bs
->aio_context
= NULL
;
5880 static void bdrv_attach_aio_context(BlockDriverState
*bs
,
5881 AioContext
*new_context
)
5883 BdrvAioNotifier
*ban
, *ban_tmp
;
5885 if (bs
->quiesce_counter
) {
5886 aio_disable_external(new_context
);
5889 bs
->aio_context
= new_context
;
5891 if (bs
->drv
&& bs
->drv
->bdrv_attach_aio_context
) {
5892 bs
->drv
->bdrv_attach_aio_context(bs
, new_context
);
5895 assert(!bs
->walking_aio_notifiers
);
5896 bs
->walking_aio_notifiers
= true;
5897 QLIST_FOREACH_SAFE(ban
, &bs
->aio_notifiers
, list
, ban_tmp
) {
5899 bdrv_do_remove_aio_context_notifier(ban
);
5901 ban
->attached_aio_context(new_context
, ban
->opaque
);
5904 bs
->walking_aio_notifiers
= false;
5908 * Changes the AioContext used for fd handlers, timers, and BHs by this
5909 * BlockDriverState and all its children and parents.
5911 * Must be called from the main AioContext.
5913 * The caller must own the AioContext lock for the old AioContext of bs, but it
5914 * must not own the AioContext lock for new_context (unless new_context is the
5915 * same as the current context of bs).
5917 * @ignore will accumulate all visited BdrvChild object. The caller is
5918 * responsible for freeing the list afterwards.
5920 void bdrv_set_aio_context_ignore(BlockDriverState
*bs
,
5921 AioContext
*new_context
, GSList
**ignore
)
5923 AioContext
*old_context
= bdrv_get_aio_context(bs
);
5926 g_assert(qemu_get_current_aio_context() == qemu_get_aio_context());
5928 if (old_context
== new_context
) {
5932 bdrv_drained_begin(bs
);
5934 QLIST_FOREACH(child
, &bs
->children
, next
) {
5935 if (g_slist_find(*ignore
, child
)) {
5938 *ignore
= g_slist_prepend(*ignore
, child
);
5939 bdrv_set_aio_context_ignore(child
->bs
, new_context
, ignore
);
5941 QLIST_FOREACH(child
, &bs
->parents
, next_parent
) {
5942 if (g_slist_find(*ignore
, child
)) {
5945 assert(child
->role
->set_aio_ctx
);
5946 *ignore
= g_slist_prepend(*ignore
, child
);
5947 child
->role
->set_aio_ctx(child
, new_context
, ignore
);
5950 bdrv_detach_aio_context(bs
);
5952 /* Acquire the new context, if necessary */
5953 if (qemu_get_aio_context() != new_context
) {
5954 aio_context_acquire(new_context
);
5957 bdrv_attach_aio_context(bs
, new_context
);
5960 * If this function was recursively called from
5961 * bdrv_set_aio_context_ignore(), there may be nodes in the
5962 * subtree that have not yet been moved to the new AioContext.
5963 * Release the old one so bdrv_drained_end() can poll them.
5965 if (qemu_get_aio_context() != old_context
) {
5966 aio_context_release(old_context
);
5969 bdrv_drained_end(bs
);
5971 if (qemu_get_aio_context() != old_context
) {
5972 aio_context_acquire(old_context
);
5974 if (qemu_get_aio_context() != new_context
) {
5975 aio_context_release(new_context
);
5979 static bool bdrv_parent_can_set_aio_context(BdrvChild
*c
, AioContext
*ctx
,
5980 GSList
**ignore
, Error
**errp
)
5982 if (g_slist_find(*ignore
, c
)) {
5985 *ignore
= g_slist_prepend(*ignore
, c
);
5987 /* A BdrvChildRole that doesn't handle AioContext changes cannot
5988 * tolerate any AioContext changes */
5989 if (!c
->role
->can_set_aio_ctx
) {
5990 char *user
= bdrv_child_user_desc(c
);
5991 error_setg(errp
, "Changing iothreads is not supported by %s", user
);
5995 if (!c
->role
->can_set_aio_ctx(c
, ctx
, ignore
, errp
)) {
5996 assert(!errp
|| *errp
);
6002 bool bdrv_child_can_set_aio_context(BdrvChild
*c
, AioContext
*ctx
,
6003 GSList
**ignore
, Error
**errp
)
6005 if (g_slist_find(*ignore
, c
)) {
6008 *ignore
= g_slist_prepend(*ignore
, c
);
6009 return bdrv_can_set_aio_context(c
->bs
, ctx
, ignore
, errp
);
6012 /* @ignore will accumulate all visited BdrvChild object. The caller is
6013 * responsible for freeing the list afterwards. */
6014 bool bdrv_can_set_aio_context(BlockDriverState
*bs
, AioContext
*ctx
,
6015 GSList
**ignore
, Error
**errp
)
6019 if (bdrv_get_aio_context(bs
) == ctx
) {
6023 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
6024 if (!bdrv_parent_can_set_aio_context(c
, ctx
, ignore
, errp
)) {
6028 QLIST_FOREACH(c
, &bs
->children
, next
) {
6029 if (!bdrv_child_can_set_aio_context(c
, ctx
, ignore
, errp
)) {
6037 int bdrv_child_try_set_aio_context(BlockDriverState
*bs
, AioContext
*ctx
,
6038 BdrvChild
*ignore_child
, Error
**errp
)
6043 ignore
= ignore_child
? g_slist_prepend(NULL
, ignore_child
) : NULL
;
6044 ret
= bdrv_can_set_aio_context(bs
, ctx
, &ignore
, errp
);
6045 g_slist_free(ignore
);
6051 ignore
= ignore_child
? g_slist_prepend(NULL
, ignore_child
) : NULL
;
6052 bdrv_set_aio_context_ignore(bs
, ctx
, &ignore
);
6053 g_slist_free(ignore
);
6058 int bdrv_try_set_aio_context(BlockDriverState
*bs
, AioContext
*ctx
,
6061 return bdrv_child_try_set_aio_context(bs
, ctx
, NULL
, errp
);
6064 void bdrv_add_aio_context_notifier(BlockDriverState
*bs
,
6065 void (*attached_aio_context
)(AioContext
*new_context
, void *opaque
),
6066 void (*detach_aio_context
)(void *opaque
), void *opaque
)
6068 BdrvAioNotifier
*ban
= g_new(BdrvAioNotifier
, 1);
6069 *ban
= (BdrvAioNotifier
){
6070 .attached_aio_context
= attached_aio_context
,
6071 .detach_aio_context
= detach_aio_context
,
6075 QLIST_INSERT_HEAD(&bs
->aio_notifiers
, ban
, list
);
6078 void bdrv_remove_aio_context_notifier(BlockDriverState
*bs
,
6079 void (*attached_aio_context
)(AioContext
*,
6081 void (*detach_aio_context
)(void *),
6084 BdrvAioNotifier
*ban
, *ban_next
;
6086 QLIST_FOREACH_SAFE(ban
, &bs
->aio_notifiers
, list
, ban_next
) {
6087 if (ban
->attached_aio_context
== attached_aio_context
&&
6088 ban
->detach_aio_context
== detach_aio_context
&&
6089 ban
->opaque
== opaque
&&
6090 ban
->deleted
== false)
6092 if (bs
->walking_aio_notifiers
) {
6093 ban
->deleted
= true;
6095 bdrv_do_remove_aio_context_notifier(ban
);
6104 int bdrv_amend_options(BlockDriverState
*bs
, QemuOpts
*opts
,
6105 BlockDriverAmendStatusCB
*status_cb
, void *cb_opaque
,
6109 error_setg(errp
, "Node is ejected");
6112 if (!bs
->drv
->bdrv_amend_options
) {
6113 error_setg(errp
, "Block driver '%s' does not support option amendment",
6114 bs
->drv
->format_name
);
6117 return bs
->drv
->bdrv_amend_options(bs
, opts
, status_cb
, cb_opaque
, errp
);
6120 /* This function will be called by the bdrv_recurse_is_first_non_filter method
6121 * of block filter and by bdrv_is_first_non_filter.
6122 * It is used to test if the given bs is the candidate or recurse more in the
6125 bool bdrv_recurse_is_first_non_filter(BlockDriverState
*bs
,
6126 BlockDriverState
*candidate
)
6128 /* return false if basic checks fails */
6129 if (!bs
|| !bs
->drv
) {
6133 /* the code reached a non block filter driver -> check if the bs is
6134 * the same as the candidate. It's the recursion termination condition.
6136 if (!bs
->drv
->is_filter
) {
6137 return bs
== candidate
;
6139 /* Down this path the driver is a block filter driver */
6141 /* If the block filter recursion method is defined use it to recurse down
6144 if (bs
->drv
->bdrv_recurse_is_first_non_filter
) {
6145 return bs
->drv
->bdrv_recurse_is_first_non_filter(bs
, candidate
);
6148 /* the driver is a block filter but don't allow to recurse -> return false
6153 /* This function checks if the candidate is the first non filter bs down it's
6154 * bs chain. Since we don't have pointers to parents it explore all bs chains
6155 * from the top. Some filters can choose not to pass down the recursion.
6157 bool bdrv_is_first_non_filter(BlockDriverState
*candidate
)
6159 BlockDriverState
*bs
;
6160 BdrvNextIterator it
;
6162 /* walk down the bs forest recursively */
6163 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
6166 /* try to recurse in this top level bs */
6167 perm
= bdrv_recurse_is_first_non_filter(bs
, candidate
);
6169 /* candidate is the first non filter */
6171 bdrv_next_cleanup(&it
);
6179 BlockDriverState
*check_to_replace_node(BlockDriverState
*parent_bs
,
6180 const char *node_name
, Error
**errp
)
6182 BlockDriverState
*to_replace_bs
= bdrv_find_node(node_name
);
6183 AioContext
*aio_context
;
6185 if (!to_replace_bs
) {
6186 error_setg(errp
, "Node name '%s' not found", node_name
);
6190 aio_context
= bdrv_get_aio_context(to_replace_bs
);
6191 aio_context_acquire(aio_context
);
6193 if (bdrv_op_is_blocked(to_replace_bs
, BLOCK_OP_TYPE_REPLACE
, errp
)) {
6194 to_replace_bs
= NULL
;
6198 /* We don't want arbitrary node of the BDS chain to be replaced only the top
6199 * most non filter in order to prevent data corruption.
6200 * Another benefit is that this tests exclude backing files which are
6201 * blocked by the backing blockers.
6203 if (!bdrv_recurse_is_first_non_filter(parent_bs
, to_replace_bs
)) {
6204 error_setg(errp
, "Only top most non filter can be replaced");
6205 to_replace_bs
= NULL
;
6210 aio_context_release(aio_context
);
6211 return to_replace_bs
;
6215 * Iterates through the list of runtime option keys that are said to
6216 * be "strong" for a BDS. An option is called "strong" if it changes
6217 * a BDS's data. For example, the null block driver's "size" and
6218 * "read-zeroes" options are strong, but its "latency-ns" option is
6221 * If a key returned by this function ends with a dot, all options
6222 * starting with that prefix are strong.
6224 static const char *const *strong_options(BlockDriverState
*bs
,
6225 const char *const *curopt
)
6227 static const char *const global_options
[] = {
6228 "driver", "filename", NULL
6232 return &global_options
[0];
6236 if (curopt
== &global_options
[ARRAY_SIZE(global_options
) - 1] && bs
->drv
) {
6237 curopt
= bs
->drv
->strong_runtime_opts
;
6240 return (curopt
&& *curopt
) ? curopt
: NULL
;
6244 * Copies all strong runtime options from bs->options to the given
6245 * QDict. The set of strong option keys is determined by invoking
6248 * Returns true iff any strong option was present in bs->options (and
6249 * thus copied to the target QDict) with the exception of "filename"
6250 * and "driver". The caller is expected to use this value to decide
6251 * whether the existence of strong options prevents the generation of
6254 static bool append_strong_runtime_options(QDict
*d
, BlockDriverState
*bs
)
6256 bool found_any
= false;
6257 const char *const *option_name
= NULL
;
6263 while ((option_name
= strong_options(bs
, option_name
))) {
6264 bool option_given
= false;
6266 assert(strlen(*option_name
) > 0);
6267 if ((*option_name
)[strlen(*option_name
) - 1] != '.') {
6268 QObject
*entry
= qdict_get(bs
->options
, *option_name
);
6273 qdict_put_obj(d
, *option_name
, qobject_ref(entry
));
6274 option_given
= true;
6276 const QDictEntry
*entry
;
6277 for (entry
= qdict_first(bs
->options
); entry
;
6278 entry
= qdict_next(bs
->options
, entry
))
6280 if (strstart(qdict_entry_key(entry
), *option_name
, NULL
)) {
6281 qdict_put_obj(d
, qdict_entry_key(entry
),
6282 qobject_ref(qdict_entry_value(entry
)));
6283 option_given
= true;
6288 /* While "driver" and "filename" need to be included in a JSON filename,
6289 * their existence does not prohibit generation of a plain filename. */
6290 if (!found_any
&& option_given
&&
6291 strcmp(*option_name
, "driver") && strcmp(*option_name
, "filename"))
6297 if (!qdict_haskey(d
, "driver")) {
6298 /* Drivers created with bdrv_new_open_driver() may not have a
6299 * @driver option. Add it here. */
6300 qdict_put_str(d
, "driver", bs
->drv
->format_name
);
6306 /* Note: This function may return false positives; it may return true
6307 * even if opening the backing file specified by bs's image header
6308 * would result in exactly bs->backing. */
6309 static bool bdrv_backing_overridden(BlockDriverState
*bs
)
6312 return strcmp(bs
->auto_backing_file
,
6313 bs
->backing
->bs
->filename
);
6315 /* No backing BDS, so if the image header reports any backing
6316 * file, it must have been suppressed */
6317 return bs
->auto_backing_file
[0] != '\0';
6321 /* Updates the following BDS fields:
6322 * - exact_filename: A filename which may be used for opening a block device
6323 * which (mostly) equals the given BDS (even without any
6324 * other options; so reading and writing must return the same
6325 * results, but caching etc. may be different)
6326 * - full_open_options: Options which, when given when opening a block device
6327 * (without a filename), result in a BDS (mostly)
6328 * equalling the given one
6329 * - filename: If exact_filename is set, it is copied here. Otherwise,
6330 * full_open_options is converted to a JSON object, prefixed with
6331 * "json:" (for use through the JSON pseudo protocol) and put here.
6333 void bdrv_refresh_filename(BlockDriverState
*bs
)
6335 BlockDriver
*drv
= bs
->drv
;
6338 bool backing_overridden
;
6339 bool generate_json_filename
; /* Whether our default implementation should
6340 fill exact_filename (false) or not (true) */
6346 /* This BDS's file name may depend on any of its children's file names, so
6347 * refresh those first */
6348 QLIST_FOREACH(child
, &bs
->children
, next
) {
6349 bdrv_refresh_filename(child
->bs
);
6353 /* For implicit nodes, just copy everything from the single child */
6354 child
= QLIST_FIRST(&bs
->children
);
6355 assert(QLIST_NEXT(child
, next
) == NULL
);
6357 pstrcpy(bs
->exact_filename
, sizeof(bs
->exact_filename
),
6358 child
->bs
->exact_filename
);
6359 pstrcpy(bs
->filename
, sizeof(bs
->filename
), child
->bs
->filename
);
6361 bs
->full_open_options
= qobject_ref(child
->bs
->full_open_options
);
6366 backing_overridden
= bdrv_backing_overridden(bs
);
6368 if (bs
->open_flags
& BDRV_O_NO_IO
) {
6369 /* Without I/O, the backing file does not change anything.
6370 * Therefore, in such a case (primarily qemu-img), we can
6371 * pretend the backing file has not been overridden even if
6372 * it technically has been. */
6373 backing_overridden
= false;
6376 /* Gather the options QDict */
6378 generate_json_filename
= append_strong_runtime_options(opts
, bs
);
6379 generate_json_filename
|= backing_overridden
;
6381 if (drv
->bdrv_gather_child_options
) {
6382 /* Some block drivers may not want to present all of their children's
6383 * options, or name them differently from BdrvChild.name */
6384 drv
->bdrv_gather_child_options(bs
, opts
, backing_overridden
);
6386 QLIST_FOREACH(child
, &bs
->children
, next
) {
6387 if (child
->role
== &child_backing
&& !backing_overridden
) {
6388 /* We can skip the backing BDS if it has not been overridden */
6392 qdict_put(opts
, child
->name
,
6393 qobject_ref(child
->bs
->full_open_options
));
6396 if (backing_overridden
&& !bs
->backing
) {
6397 /* Force no backing file */
6398 qdict_put_null(opts
, "backing");
6402 qobject_unref(bs
->full_open_options
);
6403 bs
->full_open_options
= opts
;
6405 if (drv
->bdrv_refresh_filename
) {
6406 /* Obsolete information is of no use here, so drop the old file name
6407 * information before refreshing it */
6408 bs
->exact_filename
[0] = '\0';
6410 drv
->bdrv_refresh_filename(bs
);
6411 } else if (bs
->file
) {
6412 /* Try to reconstruct valid information from the underlying file */
6414 bs
->exact_filename
[0] = '\0';
6417 * We can use the underlying file's filename if:
6418 * - it has a filename,
6419 * - the file is a protocol BDS, and
6420 * - opening that file (as this BDS's format) will automatically create
6421 * the BDS tree we have right now, that is:
6422 * - the user did not significantly change this BDS's behavior with
6423 * some explicit (strong) options
6424 * - no non-file child of this BDS has been overridden by the user
6425 * Both of these conditions are represented by generate_json_filename.
6427 if (bs
->file
->bs
->exact_filename
[0] &&
6428 bs
->file
->bs
->drv
->bdrv_file_open
&&
6429 !generate_json_filename
)
6431 strcpy(bs
->exact_filename
, bs
->file
->bs
->exact_filename
);
6435 if (bs
->exact_filename
[0]) {
6436 pstrcpy(bs
->filename
, sizeof(bs
->filename
), bs
->exact_filename
);
6438 QString
*json
= qobject_to_json(QOBJECT(bs
->full_open_options
));
6439 snprintf(bs
->filename
, sizeof(bs
->filename
), "json:%s",
6440 qstring_get_str(json
));
6441 qobject_unref(json
);
6445 char *bdrv_dirname(BlockDriverState
*bs
, Error
**errp
)
6447 BlockDriver
*drv
= bs
->drv
;
6450 error_setg(errp
, "Node '%s' is ejected", bs
->node_name
);
6454 if (drv
->bdrv_dirname
) {
6455 return drv
->bdrv_dirname(bs
, errp
);
6459 return bdrv_dirname(bs
->file
->bs
, errp
);
6462 bdrv_refresh_filename(bs
);
6463 if (bs
->exact_filename
[0] != '\0') {
6464 return path_combine(bs
->exact_filename
, "");
6467 error_setg(errp
, "Cannot generate a base directory for %s nodes",
6473 * Hot add/remove a BDS's child. So the user can take a child offline when
6474 * it is broken and take a new child online
6476 void bdrv_add_child(BlockDriverState
*parent_bs
, BlockDriverState
*child_bs
,
6480 if (!parent_bs
->drv
|| !parent_bs
->drv
->bdrv_add_child
) {
6481 error_setg(errp
, "The node %s does not support adding a child",
6482 bdrv_get_device_or_node_name(parent_bs
));
6486 if (!QLIST_EMPTY(&child_bs
->parents
)) {
6487 error_setg(errp
, "The node %s already has a parent",
6488 child_bs
->node_name
);
6492 parent_bs
->drv
->bdrv_add_child(parent_bs
, child_bs
, errp
);
6495 void bdrv_del_child(BlockDriverState
*parent_bs
, BdrvChild
*child
, Error
**errp
)
6499 if (!parent_bs
->drv
|| !parent_bs
->drv
->bdrv_del_child
) {
6500 error_setg(errp
, "The node %s does not support removing a child",
6501 bdrv_get_device_or_node_name(parent_bs
));
6505 QLIST_FOREACH(tmp
, &parent_bs
->children
, next
) {
6512 error_setg(errp
, "The node %s does not have a child named %s",
6513 bdrv_get_device_or_node_name(parent_bs
),
6514 bdrv_get_device_or_node_name(child
->bs
));
6518 parent_bs
->drv
->bdrv_del_child(parent_bs
, child
, errp
);
6521 bool bdrv_can_store_new_dirty_bitmap(BlockDriverState
*bs
, const char *name
,
6522 uint32_t granularity
, Error
**errp
)
6524 BlockDriver
*drv
= bs
->drv
;
6527 error_setg_errno(errp
, ENOMEDIUM
,
6528 "Can't store persistent bitmaps to %s",
6529 bdrv_get_device_or_node_name(bs
));
6533 if (!drv
->bdrv_can_store_new_dirty_bitmap
) {
6534 error_setg_errno(errp
, ENOTSUP
,
6535 "Can't store persistent bitmaps to %s",
6536 bdrv_get_device_or_node_name(bs
));
6540 return drv
->bdrv_can_store_new_dirty_bitmap(bs
, name
, granularity
, errp
);