2 * QEMU System Emulator block driver
4 * Copyright (c) 2003 Fabrice Bellard
6 * Permission is hereby granted, free of charge, to any person obtaining a copy
7 * of this software and associated documentation files (the "Software"), to deal
8 * in the Software without restriction, including without limitation the rights
9 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10 * copies of the Software, and to permit persons to whom the Software is
11 * furnished to do so, subject to the following conditions:
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
25 #include "qemu/osdep.h"
26 #include "block/trace.h"
27 #include "block/block_int.h"
28 #include "block/blockjob.h"
29 #include "block/nbd.h"
30 #include "block/qdict.h"
31 #include "qemu/error-report.h"
32 #include "module_block.h"
33 #include "qemu/module.h"
34 #include "qapi/error.h"
35 #include "qapi/qmp/qdict.h"
36 #include "qapi/qmp/qjson.h"
37 #include "qapi/qmp/qnull.h"
38 #include "qapi/qmp/qstring.h"
39 #include "qapi/qobject-output-visitor.h"
40 #include "qapi/qapi-visit-block-core.h"
41 #include "sysemu/block-backend.h"
42 #include "sysemu/sysemu.h"
43 #include "qemu/notify.h"
44 #include "qemu/option.h"
45 #include "qemu/coroutine.h"
46 #include "block/qapi.h"
47 #include "qemu/timer.h"
48 #include "qemu/cutils.h"
52 #include <sys/ioctl.h>
53 #include <sys/queue.h>
63 #define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */
65 static QTAILQ_HEAD(, BlockDriverState
) graph_bdrv_states
=
66 QTAILQ_HEAD_INITIALIZER(graph_bdrv_states
);
68 static QTAILQ_HEAD(, BlockDriverState
) all_bdrv_states
=
69 QTAILQ_HEAD_INITIALIZER(all_bdrv_states
);
71 static QLIST_HEAD(, BlockDriver
) bdrv_drivers
=
72 QLIST_HEAD_INITIALIZER(bdrv_drivers
);
74 static BlockDriverState
*bdrv_open_inherit(const char *filename
,
75 const char *reference
,
76 QDict
*options
, int flags
,
77 BlockDriverState
*parent
,
78 const BdrvChildRole
*child_role
,
81 /* If non-zero, use only whitelisted block drivers */
82 static int use_bdrv_whitelist
;
85 static int is_windows_drive_prefix(const char *filename
)
87 return (((filename
[0] >= 'a' && filename
[0] <= 'z') ||
88 (filename
[0] >= 'A' && filename
[0] <= 'Z')) &&
92 int is_windows_drive(const char *filename
)
94 if (is_windows_drive_prefix(filename
) &&
97 if (strstart(filename
, "\\\\.\\", NULL
) ||
98 strstart(filename
, "//./", NULL
))
104 size_t bdrv_opt_mem_align(BlockDriverState
*bs
)
106 if (!bs
|| !bs
->drv
) {
107 /* page size or 4k (hdd sector size) should be on the safe side */
108 return MAX(4096, getpagesize());
111 return bs
->bl
.opt_mem_alignment
;
114 size_t bdrv_min_mem_align(BlockDriverState
*bs
)
116 if (!bs
|| !bs
->drv
) {
117 /* page size or 4k (hdd sector size) should be on the safe side */
118 return MAX(4096, getpagesize());
121 return bs
->bl
.min_mem_alignment
;
124 /* check if the path starts with "<protocol>:" */
125 int path_has_protocol(const char *path
)
130 if (is_windows_drive(path
) ||
131 is_windows_drive_prefix(path
)) {
134 p
= path
+ strcspn(path
, ":/\\");
136 p
= path
+ strcspn(path
, ":/");
142 int path_is_absolute(const char *path
)
145 /* specific case for names like: "\\.\d:" */
146 if (is_windows_drive(path
) || is_windows_drive_prefix(path
)) {
149 return (*path
== '/' || *path
== '\\');
151 return (*path
== '/');
155 /* if filename is absolute, just return its duplicate. Otherwise, build a
156 path to it by considering it is relative to base_path. URL are
158 char *path_combine(const char *base_path
, const char *filename
)
160 const char *protocol_stripped
= NULL
;
165 if (path_is_absolute(filename
)) {
166 return g_strdup(filename
);
169 if (path_has_protocol(base_path
)) {
170 protocol_stripped
= strchr(base_path
, ':');
171 if (protocol_stripped
) {
175 p
= protocol_stripped
?: base_path
;
177 p1
= strrchr(base_path
, '/');
181 p2
= strrchr(base_path
, '\\');
182 if (!p1
|| p2
> p1
) {
197 result
= g_malloc(len
+ strlen(filename
) + 1);
198 memcpy(result
, base_path
, len
);
199 strcpy(result
+ len
, filename
);
205 * Helper function for bdrv_parse_filename() implementations to remove optional
206 * protocol prefixes (especially "file:") from a filename and for putting the
207 * stripped filename into the options QDict if there is such a prefix.
209 void bdrv_parse_filename_strip_prefix(const char *filename
, const char *prefix
,
212 if (strstart(filename
, prefix
, &filename
)) {
213 /* Stripping the explicit protocol prefix may result in a protocol
214 * prefix being (wrongly) detected (if the filename contains a colon) */
215 if (path_has_protocol(filename
)) {
216 QString
*fat_filename
;
218 /* This means there is some colon before the first slash; therefore,
219 * this cannot be an absolute path */
220 assert(!path_is_absolute(filename
));
222 /* And we can thus fix the protocol detection issue by prefixing it
224 fat_filename
= qstring_from_str("./");
225 qstring_append(fat_filename
, filename
);
227 assert(!path_has_protocol(qstring_get_str(fat_filename
)));
229 qdict_put(options
, "filename", fat_filename
);
231 /* If no protocol prefix was detected, we can use the shortened
233 qdict_put_str(options
, "filename", filename
);
239 /* Returns whether the image file is opened as read-only. Note that this can
240 * return false and writing to the image file is still not possible because the
241 * image is inactivated. */
242 bool bdrv_is_read_only(BlockDriverState
*bs
)
244 return bs
->read_only
;
247 int bdrv_can_set_read_only(BlockDriverState
*bs
, bool read_only
,
248 bool ignore_allow_rdw
, Error
**errp
)
250 /* Do not set read_only if copy_on_read is enabled */
251 if (bs
->copy_on_read
&& read_only
) {
252 error_setg(errp
, "Can't set node '%s' to r/o with copy-on-read enabled",
253 bdrv_get_device_or_node_name(bs
));
257 /* Do not clear read_only if it is prohibited */
258 if (!read_only
&& !(bs
->open_flags
& BDRV_O_ALLOW_RDWR
) &&
261 error_setg(errp
, "Node '%s' is read only",
262 bdrv_get_device_or_node_name(bs
));
270 * Called by a driver that can only provide a read-only image.
272 * Returns 0 if the node is already read-only or it could switch the node to
273 * read-only because BDRV_O_AUTO_RDONLY is set.
275 * Returns -EACCES if the node is read-write and BDRV_O_AUTO_RDONLY is not set
276 * or bdrv_can_set_read_only() forbids making the node read-only. If @errmsg
277 * is not NULL, it is used as the error message for the Error object.
279 int bdrv_apply_auto_read_only(BlockDriverState
*bs
, const char *errmsg
,
284 if (!(bs
->open_flags
& BDRV_O_RDWR
)) {
287 if (!(bs
->open_flags
& BDRV_O_AUTO_RDONLY
)) {
291 ret
= bdrv_can_set_read_only(bs
, true, false, NULL
);
296 bs
->read_only
= true;
297 bs
->open_flags
&= ~BDRV_O_RDWR
;
302 error_setg(errp
, "%s", errmsg
?: "Image is read-only");
307 * If @backing is empty, this function returns NULL without setting
308 * @errp. In all other cases, NULL will only be returned with @errp
311 * Therefore, a return value of NULL without @errp set means that
312 * there is no backing file; if @errp is set, there is one but its
313 * absolute filename cannot be generated.
315 char *bdrv_get_full_backing_filename_from_filename(const char *backed
,
319 if (backing
[0] == '\0') {
321 } else if (path_has_protocol(backing
) || path_is_absolute(backing
)) {
322 return g_strdup(backing
);
323 } else if (backed
[0] == '\0' || strstart(backed
, "json:", NULL
)) {
324 error_setg(errp
, "Cannot use relative backing file names for '%s'",
328 return path_combine(backed
, backing
);
333 * If @filename is empty or NULL, this function returns NULL without
334 * setting @errp. In all other cases, NULL will only be returned with
337 static char *bdrv_make_absolute_filename(BlockDriverState
*relative_to
,
338 const char *filename
, Error
**errp
)
340 char *dir
, *full_name
;
342 if (!filename
|| filename
[0] == '\0') {
344 } else if (path_has_protocol(filename
) || path_is_absolute(filename
)) {
345 return g_strdup(filename
);
348 dir
= bdrv_dirname(relative_to
, errp
);
353 full_name
= g_strconcat(dir
, filename
, NULL
);
358 char *bdrv_get_full_backing_filename(BlockDriverState
*bs
, Error
**errp
)
360 return bdrv_make_absolute_filename(bs
, bs
->backing_file
, errp
);
363 void bdrv_register(BlockDriver
*bdrv
)
365 QLIST_INSERT_HEAD(&bdrv_drivers
, bdrv
, list
);
368 BlockDriverState
*bdrv_new(void)
370 BlockDriverState
*bs
;
373 bs
= g_new0(BlockDriverState
, 1);
374 QLIST_INIT(&bs
->dirty_bitmaps
);
375 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
376 QLIST_INIT(&bs
->op_blockers
[i
]);
378 notifier_with_return_list_init(&bs
->before_write_notifiers
);
379 qemu_co_mutex_init(&bs
->reqs_lock
);
380 qemu_mutex_init(&bs
->dirty_bitmap_mutex
);
382 bs
->aio_context
= qemu_get_aio_context();
384 qemu_co_queue_init(&bs
->flush_queue
);
386 for (i
= 0; i
< bdrv_drain_all_count
; i
++) {
387 bdrv_drained_begin(bs
);
390 QTAILQ_INSERT_TAIL(&all_bdrv_states
, bs
, bs_list
);
395 static BlockDriver
*bdrv_do_find_format(const char *format_name
)
399 QLIST_FOREACH(drv1
, &bdrv_drivers
, list
) {
400 if (!strcmp(drv1
->format_name
, format_name
)) {
408 BlockDriver
*bdrv_find_format(const char *format_name
)
413 drv1
= bdrv_do_find_format(format_name
);
418 /* The driver isn't registered, maybe we need to load a module */
419 for (i
= 0; i
< (int)ARRAY_SIZE(block_driver_modules
); ++i
) {
420 if (!strcmp(block_driver_modules
[i
].format_name
, format_name
)) {
421 block_module_load_one(block_driver_modules
[i
].library_name
);
426 return bdrv_do_find_format(format_name
);
429 static int bdrv_format_is_whitelisted(const char *format_name
, bool read_only
)
431 static const char *whitelist_rw
[] = {
432 CONFIG_BDRV_RW_WHITELIST
434 static const char *whitelist_ro
[] = {
435 CONFIG_BDRV_RO_WHITELIST
439 if (!whitelist_rw
[0] && !whitelist_ro
[0]) {
440 return 1; /* no whitelist, anything goes */
443 for (p
= whitelist_rw
; *p
; p
++) {
444 if (!strcmp(format_name
, *p
)) {
449 for (p
= whitelist_ro
; *p
; p
++) {
450 if (!strcmp(format_name
, *p
)) {
458 int bdrv_is_whitelisted(BlockDriver
*drv
, bool read_only
)
460 return bdrv_format_is_whitelisted(drv
->format_name
, read_only
);
463 bool bdrv_uses_whitelist(void)
465 return use_bdrv_whitelist
;
468 typedef struct CreateCo
{
476 static void coroutine_fn
bdrv_create_co_entry(void *opaque
)
478 Error
*local_err
= NULL
;
481 CreateCo
*cco
= opaque
;
484 ret
= cco
->drv
->bdrv_co_create_opts(cco
->filename
, cco
->opts
, &local_err
);
485 error_propagate(&cco
->err
, local_err
);
489 int bdrv_create(BlockDriver
*drv
, const char* filename
,
490 QemuOpts
*opts
, Error
**errp
)
497 .filename
= g_strdup(filename
),
503 if (!drv
->bdrv_co_create_opts
) {
504 error_setg(errp
, "Driver '%s' does not support image creation", drv
->format_name
);
509 if (qemu_in_coroutine()) {
510 /* Fast-path if already in coroutine context */
511 bdrv_create_co_entry(&cco
);
513 co
= qemu_coroutine_create(bdrv_create_co_entry
, &cco
);
514 qemu_coroutine_enter(co
);
515 while (cco
.ret
== NOT_DONE
) {
516 aio_poll(qemu_get_aio_context(), true);
523 error_propagate(errp
, cco
.err
);
525 error_setg_errno(errp
, -ret
, "Could not create image");
530 g_free(cco
.filename
);
534 int bdrv_create_file(const char *filename
, QemuOpts
*opts
, Error
**errp
)
537 Error
*local_err
= NULL
;
540 drv
= bdrv_find_protocol(filename
, true, errp
);
545 ret
= bdrv_create(drv
, filename
, opts
, &local_err
);
546 error_propagate(errp
, local_err
);
551 * Try to get @bs's logical and physical block size.
552 * On success, store them in @bsz struct and return 0.
553 * On failure return -errno.
554 * @bs must not be empty.
556 int bdrv_probe_blocksizes(BlockDriverState
*bs
, BlockSizes
*bsz
)
558 BlockDriver
*drv
= bs
->drv
;
560 if (drv
&& drv
->bdrv_probe_blocksizes
) {
561 return drv
->bdrv_probe_blocksizes(bs
, bsz
);
562 } else if (drv
&& drv
->is_filter
&& bs
->file
) {
563 return bdrv_probe_blocksizes(bs
->file
->bs
, bsz
);
570 * Try to get @bs's geometry (cyls, heads, sectors).
571 * On success, store them in @geo struct and return 0.
572 * On failure return -errno.
573 * @bs must not be empty.
575 int bdrv_probe_geometry(BlockDriverState
*bs
, HDGeometry
*geo
)
577 BlockDriver
*drv
= bs
->drv
;
579 if (drv
&& drv
->bdrv_probe_geometry
) {
580 return drv
->bdrv_probe_geometry(bs
, geo
);
581 } else if (drv
&& drv
->is_filter
&& bs
->file
) {
582 return bdrv_probe_geometry(bs
->file
->bs
, geo
);
589 * Create a uniquely-named empty temporary file.
590 * Return 0 upon success, otherwise a negative errno value.
592 int get_tmp_filename(char *filename
, int size
)
595 char temp_dir
[MAX_PATH
];
596 /* GetTempFileName requires that its output buffer (4th param)
597 have length MAX_PATH or greater. */
598 assert(size
>= MAX_PATH
);
599 return (GetTempPath(MAX_PATH
, temp_dir
)
600 && GetTempFileName(temp_dir
, "qem", 0, filename
)
601 ? 0 : -GetLastError());
605 tmpdir
= getenv("TMPDIR");
609 if (snprintf(filename
, size
, "%s/vl.XXXXXX", tmpdir
) >= size
) {
612 fd
= mkstemp(filename
);
616 if (close(fd
) != 0) {
625 * Detect host devices. By convention, /dev/cdrom[N] is always
626 * recognized as a host CDROM.
628 static BlockDriver
*find_hdev_driver(const char *filename
)
630 int score_max
= 0, score
;
631 BlockDriver
*drv
= NULL
, *d
;
633 QLIST_FOREACH(d
, &bdrv_drivers
, list
) {
634 if (d
->bdrv_probe_device
) {
635 score
= d
->bdrv_probe_device(filename
);
636 if (score
> score_max
) {
646 static BlockDriver
*bdrv_do_find_protocol(const char *protocol
)
650 QLIST_FOREACH(drv1
, &bdrv_drivers
, list
) {
651 if (drv1
->protocol_name
&& !strcmp(drv1
->protocol_name
, protocol
)) {
659 BlockDriver
*bdrv_find_protocol(const char *filename
,
660 bool allow_protocol_prefix
,
669 /* TODO Drivers without bdrv_file_open must be specified explicitly */
672 * XXX(hch): we really should not let host device detection
673 * override an explicit protocol specification, but moving this
674 * later breaks access to device names with colons in them.
675 * Thanks to the brain-dead persistent naming schemes on udev-
676 * based Linux systems those actually are quite common.
678 drv1
= find_hdev_driver(filename
);
683 if (!path_has_protocol(filename
) || !allow_protocol_prefix
) {
687 p
= strchr(filename
, ':');
690 if (len
> sizeof(protocol
) - 1)
691 len
= sizeof(protocol
) - 1;
692 memcpy(protocol
, filename
, len
);
693 protocol
[len
] = '\0';
695 drv1
= bdrv_do_find_protocol(protocol
);
700 for (i
= 0; i
< (int)ARRAY_SIZE(block_driver_modules
); ++i
) {
701 if (block_driver_modules
[i
].protocol_name
&&
702 !strcmp(block_driver_modules
[i
].protocol_name
, protocol
)) {
703 block_module_load_one(block_driver_modules
[i
].library_name
);
708 drv1
= bdrv_do_find_protocol(protocol
);
710 error_setg(errp
, "Unknown protocol '%s'", protocol
);
716 * Guess image format by probing its contents.
717 * This is not a good idea when your image is raw (CVE-2008-2004), but
718 * we do it anyway for backward compatibility.
720 * @buf contains the image's first @buf_size bytes.
721 * @buf_size is the buffer size in bytes (generally BLOCK_PROBE_BUF_SIZE,
722 * but can be smaller if the image file is smaller)
723 * @filename is its filename.
725 * For all block drivers, call the bdrv_probe() method to get its
727 * Return the first block driver with the highest probing score.
729 BlockDriver
*bdrv_probe_all(const uint8_t *buf
, int buf_size
,
730 const char *filename
)
732 int score_max
= 0, score
;
733 BlockDriver
*drv
= NULL
, *d
;
735 QLIST_FOREACH(d
, &bdrv_drivers
, list
) {
737 score
= d
->bdrv_probe(buf
, buf_size
, filename
);
738 if (score
> score_max
) {
748 static int find_image_format(BlockBackend
*file
, const char *filename
,
749 BlockDriver
**pdrv
, Error
**errp
)
752 uint8_t buf
[BLOCK_PROBE_BUF_SIZE
];
755 /* Return the raw BlockDriver * to scsi-generic devices or empty drives */
756 if (blk_is_sg(file
) || !blk_is_inserted(file
) || blk_getlength(file
) == 0) {
761 ret
= blk_pread(file
, 0, buf
, sizeof(buf
));
763 error_setg_errno(errp
, -ret
, "Could not read image for determining its "
769 drv
= bdrv_probe_all(buf
, ret
, filename
);
771 error_setg(errp
, "Could not determine image format: No compatible "
780 * Set the current 'total_sectors' value
781 * Return 0 on success, -errno on error.
783 int refresh_total_sectors(BlockDriverState
*bs
, int64_t hint
)
785 BlockDriver
*drv
= bs
->drv
;
791 /* Do not attempt drv->bdrv_getlength() on scsi-generic devices */
795 /* query actual device if possible, otherwise just trust the hint */
796 if (drv
->bdrv_getlength
) {
797 int64_t length
= drv
->bdrv_getlength(bs
);
801 hint
= DIV_ROUND_UP(length
, BDRV_SECTOR_SIZE
);
804 bs
->total_sectors
= hint
;
809 * Combines a QDict of new block driver @options with any missing options taken
810 * from @old_options, so that leaving out an option defaults to its old value.
812 static void bdrv_join_options(BlockDriverState
*bs
, QDict
*options
,
815 if (bs
->drv
&& bs
->drv
->bdrv_join_options
) {
816 bs
->drv
->bdrv_join_options(options
, old_options
);
818 qdict_join(options
, old_options
, false);
822 static BlockdevDetectZeroesOptions
bdrv_parse_detect_zeroes(QemuOpts
*opts
,
826 Error
*local_err
= NULL
;
827 char *value
= qemu_opt_get_del(opts
, "detect-zeroes");
828 BlockdevDetectZeroesOptions detect_zeroes
=
829 qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup
, value
,
830 BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF
, &local_err
);
833 error_propagate(errp
, local_err
);
834 return detect_zeroes
;
837 if (detect_zeroes
== BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP
&&
838 !(open_flags
& BDRV_O_UNMAP
))
840 error_setg(errp
, "setting detect-zeroes to unmap is not allowed "
841 "without setting discard operation to unmap");
844 return detect_zeroes
;
848 * Set open flags for a given discard mode
850 * Return 0 on success, -1 if the discard mode was invalid.
852 int bdrv_parse_discard_flags(const char *mode
, int *flags
)
854 *flags
&= ~BDRV_O_UNMAP
;
856 if (!strcmp(mode
, "off") || !strcmp(mode
, "ignore")) {
858 } else if (!strcmp(mode
, "on") || !strcmp(mode
, "unmap")) {
859 *flags
|= BDRV_O_UNMAP
;
868 * Set open flags for a given cache mode
870 * Return 0 on success, -1 if the cache mode was invalid.
872 int bdrv_parse_cache_mode(const char *mode
, int *flags
, bool *writethrough
)
874 *flags
&= ~BDRV_O_CACHE_MASK
;
876 if (!strcmp(mode
, "off") || !strcmp(mode
, "none")) {
877 *writethrough
= false;
878 *flags
|= BDRV_O_NOCACHE
;
879 } else if (!strcmp(mode
, "directsync")) {
880 *writethrough
= true;
881 *flags
|= BDRV_O_NOCACHE
;
882 } else if (!strcmp(mode
, "writeback")) {
883 *writethrough
= false;
884 } else if (!strcmp(mode
, "unsafe")) {
885 *writethrough
= false;
886 *flags
|= BDRV_O_NO_FLUSH
;
887 } else if (!strcmp(mode
, "writethrough")) {
888 *writethrough
= true;
896 static char *bdrv_child_get_parent_desc(BdrvChild
*c
)
898 BlockDriverState
*parent
= c
->opaque
;
899 return g_strdup(bdrv_get_device_or_node_name(parent
));
902 static void bdrv_child_cb_drained_begin(BdrvChild
*child
)
904 BlockDriverState
*bs
= child
->opaque
;
905 bdrv_do_drained_begin_quiesce(bs
, NULL
, false);
908 static bool bdrv_child_cb_drained_poll(BdrvChild
*child
)
910 BlockDriverState
*bs
= child
->opaque
;
911 return bdrv_drain_poll(bs
, false, NULL
, false);
914 static void bdrv_child_cb_drained_end(BdrvChild
*child
)
916 BlockDriverState
*bs
= child
->opaque
;
917 bdrv_drained_end(bs
);
920 static void bdrv_child_cb_attach(BdrvChild
*child
)
922 BlockDriverState
*bs
= child
->opaque
;
923 bdrv_apply_subtree_drain(child
, bs
);
926 static void bdrv_child_cb_detach(BdrvChild
*child
)
928 BlockDriverState
*bs
= child
->opaque
;
929 bdrv_unapply_subtree_drain(child
, bs
);
932 static int bdrv_child_cb_inactivate(BdrvChild
*child
)
934 BlockDriverState
*bs
= child
->opaque
;
935 assert(bs
->open_flags
& BDRV_O_INACTIVE
);
939 static bool bdrv_child_cb_can_set_aio_ctx(BdrvChild
*child
, AioContext
*ctx
,
940 GSList
**ignore
, Error
**errp
)
942 BlockDriverState
*bs
= child
->opaque
;
943 return bdrv_can_set_aio_context(bs
, ctx
, ignore
, errp
);
946 static void bdrv_child_cb_set_aio_ctx(BdrvChild
*child
, AioContext
*ctx
,
949 BlockDriverState
*bs
= child
->opaque
;
950 return bdrv_set_aio_context_ignore(bs
, ctx
, ignore
);
954 * Returns the options and flags that a temporary snapshot should get, based on
955 * the originally requested flags (the originally requested image will have
956 * flags like a backing file)
958 static void bdrv_temp_snapshot_options(int *child_flags
, QDict
*child_options
,
959 int parent_flags
, QDict
*parent_options
)
961 *child_flags
= (parent_flags
& ~BDRV_O_SNAPSHOT
) | BDRV_O_TEMPORARY
;
963 /* For temporary files, unconditional cache=unsafe is fine */
964 qdict_set_default_str(child_options
, BDRV_OPT_CACHE_DIRECT
, "off");
965 qdict_set_default_str(child_options
, BDRV_OPT_CACHE_NO_FLUSH
, "on");
967 /* Copy the read-only and discard options from the parent */
968 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_READ_ONLY
);
969 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_DISCARD
);
971 /* aio=native doesn't work for cache.direct=off, so disable it for the
972 * temporary snapshot */
973 *child_flags
&= ~BDRV_O_NATIVE_AIO
;
977 * Returns the options and flags that bs->file should get if a protocol driver
978 * is expected, based on the given options and flags for the parent BDS
980 static void bdrv_inherited_options(int *child_flags
, QDict
*child_options
,
981 int parent_flags
, QDict
*parent_options
)
983 int flags
= parent_flags
;
985 /* Enable protocol handling, disable format probing for bs->file */
986 flags
|= BDRV_O_PROTOCOL
;
988 /* If the cache mode isn't explicitly set, inherit direct and no-flush from
990 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_CACHE_DIRECT
);
991 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_CACHE_NO_FLUSH
);
992 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_FORCE_SHARE
);
994 /* Inherit the read-only option from the parent if it's not set */
995 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_READ_ONLY
);
996 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_AUTO_READ_ONLY
);
998 /* Our block drivers take care to send flushes and respect unmap policy,
999 * so we can default to enable both on lower layers regardless of the
1000 * corresponding parent options. */
1001 qdict_set_default_str(child_options
, BDRV_OPT_DISCARD
, "unmap");
1003 /* Clear flags that only apply to the top layer */
1004 flags
&= ~(BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
| BDRV_O_COPY_ON_READ
|
1007 *child_flags
= flags
;
1010 const BdrvChildRole child_file
= {
1011 .parent_is_bds
= true,
1012 .get_parent_desc
= bdrv_child_get_parent_desc
,
1013 .inherit_options
= bdrv_inherited_options
,
1014 .drained_begin
= bdrv_child_cb_drained_begin
,
1015 .drained_poll
= bdrv_child_cb_drained_poll
,
1016 .drained_end
= bdrv_child_cb_drained_end
,
1017 .attach
= bdrv_child_cb_attach
,
1018 .detach
= bdrv_child_cb_detach
,
1019 .inactivate
= bdrv_child_cb_inactivate
,
1020 .can_set_aio_ctx
= bdrv_child_cb_can_set_aio_ctx
,
1021 .set_aio_ctx
= bdrv_child_cb_set_aio_ctx
,
1025 * Returns the options and flags that bs->file should get if the use of formats
1026 * (and not only protocols) is permitted for it, based on the given options and
1027 * flags for the parent BDS
1029 static void bdrv_inherited_fmt_options(int *child_flags
, QDict
*child_options
,
1030 int parent_flags
, QDict
*parent_options
)
1032 child_file
.inherit_options(child_flags
, child_options
,
1033 parent_flags
, parent_options
);
1035 *child_flags
&= ~(BDRV_O_PROTOCOL
| BDRV_O_NO_IO
);
1038 const BdrvChildRole child_format
= {
1039 .parent_is_bds
= true,
1040 .get_parent_desc
= bdrv_child_get_parent_desc
,
1041 .inherit_options
= bdrv_inherited_fmt_options
,
1042 .drained_begin
= bdrv_child_cb_drained_begin
,
1043 .drained_poll
= bdrv_child_cb_drained_poll
,
1044 .drained_end
= bdrv_child_cb_drained_end
,
1045 .attach
= bdrv_child_cb_attach
,
1046 .detach
= bdrv_child_cb_detach
,
1047 .inactivate
= bdrv_child_cb_inactivate
,
1048 .can_set_aio_ctx
= bdrv_child_cb_can_set_aio_ctx
,
1049 .set_aio_ctx
= bdrv_child_cb_set_aio_ctx
,
1052 static void bdrv_backing_attach(BdrvChild
*c
)
1054 BlockDriverState
*parent
= c
->opaque
;
1055 BlockDriverState
*backing_hd
= c
->bs
;
1057 assert(!parent
->backing_blocker
);
1058 error_setg(&parent
->backing_blocker
,
1059 "node is used as backing hd of '%s'",
1060 bdrv_get_device_or_node_name(parent
));
1062 bdrv_refresh_filename(backing_hd
);
1064 parent
->open_flags
&= ~BDRV_O_NO_BACKING
;
1065 pstrcpy(parent
->backing_file
, sizeof(parent
->backing_file
),
1066 backing_hd
->filename
);
1067 pstrcpy(parent
->backing_format
, sizeof(parent
->backing_format
),
1068 backing_hd
->drv
? backing_hd
->drv
->format_name
: "");
1070 bdrv_op_block_all(backing_hd
, parent
->backing_blocker
);
1071 /* Otherwise we won't be able to commit or stream */
1072 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_COMMIT_TARGET
,
1073 parent
->backing_blocker
);
1074 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_STREAM
,
1075 parent
->backing_blocker
);
1077 * We do backup in 3 ways:
1079 * The target bs is new opened, and the source is top BDS
1080 * 2. blockdev backup
1081 * Both the source and the target are top BDSes.
1082 * 3. internal backup(used for block replication)
1083 * Both the source and the target are backing file
1085 * In case 1 and 2, neither the source nor the target is the backing file.
1086 * In case 3, we will block the top BDS, so there is only one block job
1087 * for the top BDS and its backing chain.
1089 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_BACKUP_SOURCE
,
1090 parent
->backing_blocker
);
1091 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_BACKUP_TARGET
,
1092 parent
->backing_blocker
);
1094 bdrv_child_cb_attach(c
);
1097 static void bdrv_backing_detach(BdrvChild
*c
)
1099 BlockDriverState
*parent
= c
->opaque
;
1101 assert(parent
->backing_blocker
);
1102 bdrv_op_unblock_all(c
->bs
, parent
->backing_blocker
);
1103 error_free(parent
->backing_blocker
);
1104 parent
->backing_blocker
= NULL
;
1106 bdrv_child_cb_detach(c
);
1110 * Returns the options and flags that bs->backing should get, based on the
1111 * given options and flags for the parent BDS
1113 static void bdrv_backing_options(int *child_flags
, QDict
*child_options
,
1114 int parent_flags
, QDict
*parent_options
)
1116 int flags
= parent_flags
;
1118 /* The cache mode is inherited unmodified for backing files; except WCE,
1119 * which is only applied on the top level (BlockBackend) */
1120 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_CACHE_DIRECT
);
1121 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_CACHE_NO_FLUSH
);
1122 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_FORCE_SHARE
);
1124 /* backing files always opened read-only */
1125 qdict_set_default_str(child_options
, BDRV_OPT_READ_ONLY
, "on");
1126 qdict_set_default_str(child_options
, BDRV_OPT_AUTO_READ_ONLY
, "off");
1127 flags
&= ~BDRV_O_COPY_ON_READ
;
1129 /* snapshot=on is handled on the top layer */
1130 flags
&= ~(BDRV_O_SNAPSHOT
| BDRV_O_TEMPORARY
);
1132 *child_flags
= flags
;
1135 static int bdrv_backing_update_filename(BdrvChild
*c
, BlockDriverState
*base
,
1136 const char *filename
, Error
**errp
)
1138 BlockDriverState
*parent
= c
->opaque
;
1139 bool read_only
= bdrv_is_read_only(parent
);
1143 ret
= bdrv_reopen_set_read_only(parent
, false, errp
);
1149 ret
= bdrv_change_backing_file(parent
, filename
,
1150 base
->drv
? base
->drv
->format_name
: "");
1152 error_setg_errno(errp
, -ret
, "Could not update backing file link");
1156 bdrv_reopen_set_read_only(parent
, true, NULL
);
1162 const BdrvChildRole child_backing
= {
1163 .parent_is_bds
= true,
1164 .get_parent_desc
= bdrv_child_get_parent_desc
,
1165 .attach
= bdrv_backing_attach
,
1166 .detach
= bdrv_backing_detach
,
1167 .inherit_options
= bdrv_backing_options
,
1168 .drained_begin
= bdrv_child_cb_drained_begin
,
1169 .drained_poll
= bdrv_child_cb_drained_poll
,
1170 .drained_end
= bdrv_child_cb_drained_end
,
1171 .inactivate
= bdrv_child_cb_inactivate
,
1172 .update_filename
= bdrv_backing_update_filename
,
1173 .can_set_aio_ctx
= bdrv_child_cb_can_set_aio_ctx
,
1174 .set_aio_ctx
= bdrv_child_cb_set_aio_ctx
,
1177 static int bdrv_open_flags(BlockDriverState
*bs
, int flags
)
1179 int open_flags
= flags
;
1182 * Clear flags that are internal to the block layer before opening the
1185 open_flags
&= ~(BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
| BDRV_O_PROTOCOL
);
1190 static void update_flags_from_options(int *flags
, QemuOpts
*opts
)
1192 *flags
&= ~(BDRV_O_CACHE_MASK
| BDRV_O_RDWR
| BDRV_O_AUTO_RDONLY
);
1194 if (qemu_opt_get_bool_del(opts
, BDRV_OPT_CACHE_NO_FLUSH
, false)) {
1195 *flags
|= BDRV_O_NO_FLUSH
;
1198 if (qemu_opt_get_bool_del(opts
, BDRV_OPT_CACHE_DIRECT
, false)) {
1199 *flags
|= BDRV_O_NOCACHE
;
1202 if (!qemu_opt_get_bool_del(opts
, BDRV_OPT_READ_ONLY
, false)) {
1203 *flags
|= BDRV_O_RDWR
;
1206 if (qemu_opt_get_bool_del(opts
, BDRV_OPT_AUTO_READ_ONLY
, false)) {
1207 *flags
|= BDRV_O_AUTO_RDONLY
;
1211 static void update_options_from_flags(QDict
*options
, int flags
)
1213 if (!qdict_haskey(options
, BDRV_OPT_CACHE_DIRECT
)) {
1214 qdict_put_bool(options
, BDRV_OPT_CACHE_DIRECT
, flags
& BDRV_O_NOCACHE
);
1216 if (!qdict_haskey(options
, BDRV_OPT_CACHE_NO_FLUSH
)) {
1217 qdict_put_bool(options
, BDRV_OPT_CACHE_NO_FLUSH
,
1218 flags
& BDRV_O_NO_FLUSH
);
1220 if (!qdict_haskey(options
, BDRV_OPT_READ_ONLY
)) {
1221 qdict_put_bool(options
, BDRV_OPT_READ_ONLY
, !(flags
& BDRV_O_RDWR
));
1223 if (!qdict_haskey(options
, BDRV_OPT_AUTO_READ_ONLY
)) {
1224 qdict_put_bool(options
, BDRV_OPT_AUTO_READ_ONLY
,
1225 flags
& BDRV_O_AUTO_RDONLY
);
1229 static void bdrv_assign_node_name(BlockDriverState
*bs
,
1230 const char *node_name
,
1233 char *gen_node_name
= NULL
;
1236 node_name
= gen_node_name
= id_generate(ID_BLOCK
);
1237 } else if (!id_wellformed(node_name
)) {
1239 * Check for empty string or invalid characters, but not if it is
1240 * generated (generated names use characters not available to the user)
1242 error_setg(errp
, "Invalid node name");
1246 /* takes care of avoiding namespaces collisions */
1247 if (blk_by_name(node_name
)) {
1248 error_setg(errp
, "node-name=%s is conflicting with a device id",
1253 /* takes care of avoiding duplicates node names */
1254 if (bdrv_find_node(node_name
)) {
1255 error_setg(errp
, "Duplicate node name");
1259 /* Make sure that the node name isn't truncated */
1260 if (strlen(node_name
) >= sizeof(bs
->node_name
)) {
1261 error_setg(errp
, "Node name too long");
1265 /* copy node name into the bs and insert it into the graph list */
1266 pstrcpy(bs
->node_name
, sizeof(bs
->node_name
), node_name
);
1267 QTAILQ_INSERT_TAIL(&graph_bdrv_states
, bs
, node_list
);
1269 g_free(gen_node_name
);
1272 static int bdrv_open_driver(BlockDriverState
*bs
, BlockDriver
*drv
,
1273 const char *node_name
, QDict
*options
,
1274 int open_flags
, Error
**errp
)
1276 Error
*local_err
= NULL
;
1279 bdrv_assign_node_name(bs
, node_name
, &local_err
);
1281 error_propagate(errp
, local_err
);
1286 bs
->read_only
= !(bs
->open_flags
& BDRV_O_RDWR
);
1287 bs
->opaque
= g_malloc0(drv
->instance_size
);
1289 if (drv
->bdrv_file_open
) {
1290 assert(!drv
->bdrv_needs_filename
|| bs
->filename
[0]);
1291 ret
= drv
->bdrv_file_open(bs
, options
, open_flags
, &local_err
);
1292 } else if (drv
->bdrv_open
) {
1293 ret
= drv
->bdrv_open(bs
, options
, open_flags
, &local_err
);
1300 error_propagate(errp
, local_err
);
1301 } else if (bs
->filename
[0]) {
1302 error_setg_errno(errp
, -ret
, "Could not open '%s'", bs
->filename
);
1304 error_setg_errno(errp
, -ret
, "Could not open image");
1309 ret
= refresh_total_sectors(bs
, bs
->total_sectors
);
1311 error_setg_errno(errp
, -ret
, "Could not refresh total sector count");
1315 bdrv_refresh_limits(bs
, &local_err
);
1317 error_propagate(errp
, local_err
);
1321 assert(bdrv_opt_mem_align(bs
) != 0);
1322 assert(bdrv_min_mem_align(bs
) != 0);
1323 assert(is_power_of_2(bs
->bl
.request_alignment
));
1325 for (i
= 0; i
< bs
->quiesce_counter
; i
++) {
1326 if (drv
->bdrv_co_drain_begin
) {
1327 drv
->bdrv_co_drain_begin(bs
);
1334 if (bs
->file
!= NULL
) {
1335 bdrv_unref_child(bs
, bs
->file
);
1343 BlockDriverState
*bdrv_new_open_driver(BlockDriver
*drv
, const char *node_name
,
1344 int flags
, Error
**errp
)
1346 BlockDriverState
*bs
;
1350 bs
->open_flags
= flags
;
1351 bs
->explicit_options
= qdict_new();
1352 bs
->options
= qdict_new();
1355 update_options_from_flags(bs
->options
, flags
);
1357 ret
= bdrv_open_driver(bs
, drv
, node_name
, bs
->options
, flags
, errp
);
1359 qobject_unref(bs
->explicit_options
);
1360 bs
->explicit_options
= NULL
;
1361 qobject_unref(bs
->options
);
1370 QemuOptsList bdrv_runtime_opts
= {
1371 .name
= "bdrv_common",
1372 .head
= QTAILQ_HEAD_INITIALIZER(bdrv_runtime_opts
.head
),
1375 .name
= "node-name",
1376 .type
= QEMU_OPT_STRING
,
1377 .help
= "Node name of the block device node",
1381 .type
= QEMU_OPT_STRING
,
1382 .help
= "Block driver to use for the node",
1385 .name
= BDRV_OPT_CACHE_DIRECT
,
1386 .type
= QEMU_OPT_BOOL
,
1387 .help
= "Bypass software writeback cache on the host",
1390 .name
= BDRV_OPT_CACHE_NO_FLUSH
,
1391 .type
= QEMU_OPT_BOOL
,
1392 .help
= "Ignore flush requests",
1395 .name
= BDRV_OPT_READ_ONLY
,
1396 .type
= QEMU_OPT_BOOL
,
1397 .help
= "Node is opened in read-only mode",
1400 .name
= BDRV_OPT_AUTO_READ_ONLY
,
1401 .type
= QEMU_OPT_BOOL
,
1402 .help
= "Node can become read-only if opening read-write fails",
1405 .name
= "detect-zeroes",
1406 .type
= QEMU_OPT_STRING
,
1407 .help
= "try to optimize zero writes (off, on, unmap)",
1410 .name
= BDRV_OPT_DISCARD
,
1411 .type
= QEMU_OPT_STRING
,
1412 .help
= "discard operation (ignore/off, unmap/on)",
1415 .name
= BDRV_OPT_FORCE_SHARE
,
1416 .type
= QEMU_OPT_BOOL
,
1417 .help
= "always accept other writers (default: off)",
1419 { /* end of list */ }
1424 * Common part for opening disk images and files
1426 * Removes all processed options from *options.
1428 static int bdrv_open_common(BlockDriverState
*bs
, BlockBackend
*file
,
1429 QDict
*options
, Error
**errp
)
1431 int ret
, open_flags
;
1432 const char *filename
;
1433 const char *driver_name
= NULL
;
1434 const char *node_name
= NULL
;
1435 const char *discard
;
1438 Error
*local_err
= NULL
;
1440 assert(bs
->file
== NULL
);
1441 assert(options
!= NULL
&& bs
->options
!= options
);
1443 opts
= qemu_opts_create(&bdrv_runtime_opts
, NULL
, 0, &error_abort
);
1444 qemu_opts_absorb_qdict(opts
, options
, &local_err
);
1446 error_propagate(errp
, local_err
);
1451 update_flags_from_options(&bs
->open_flags
, opts
);
1453 driver_name
= qemu_opt_get(opts
, "driver");
1454 drv
= bdrv_find_format(driver_name
);
1455 assert(drv
!= NULL
);
1457 bs
->force_share
= qemu_opt_get_bool(opts
, BDRV_OPT_FORCE_SHARE
, false);
1459 if (bs
->force_share
&& (bs
->open_flags
& BDRV_O_RDWR
)) {
1461 BDRV_OPT_FORCE_SHARE
1462 "=on can only be used with read-only images");
1468 bdrv_refresh_filename(blk_bs(file
));
1469 filename
= blk_bs(file
)->filename
;
1472 * Caution: while qdict_get_try_str() is fine, getting
1473 * non-string types would require more care. When @options
1474 * come from -blockdev or blockdev_add, its members are typed
1475 * according to the QAPI schema, but when they come from
1476 * -drive, they're all QString.
1478 filename
= qdict_get_try_str(options
, "filename");
1481 if (drv
->bdrv_needs_filename
&& (!filename
|| !filename
[0])) {
1482 error_setg(errp
, "The '%s' block driver requires a file name",
1488 trace_bdrv_open_common(bs
, filename
?: "", bs
->open_flags
,
1491 bs
->read_only
= !(bs
->open_flags
& BDRV_O_RDWR
);
1493 if (use_bdrv_whitelist
&& !bdrv_is_whitelisted(drv
, bs
->read_only
)) {
1494 if (!bs
->read_only
&& bdrv_is_whitelisted(drv
, true)) {
1495 ret
= bdrv_apply_auto_read_only(bs
, NULL
, NULL
);
1501 !bs
->read_only
&& bdrv_is_whitelisted(drv
, true)
1502 ? "Driver '%s' can only be used for read-only devices"
1503 : "Driver '%s' is not whitelisted",
1509 /* bdrv_new() and bdrv_close() make it so */
1510 assert(atomic_read(&bs
->copy_on_read
) == 0);
1512 if (bs
->open_flags
& BDRV_O_COPY_ON_READ
) {
1513 if (!bs
->read_only
) {
1514 bdrv_enable_copy_on_read(bs
);
1516 error_setg(errp
, "Can't use copy-on-read on read-only device");
1522 discard
= qemu_opt_get(opts
, BDRV_OPT_DISCARD
);
1523 if (discard
!= NULL
) {
1524 if (bdrv_parse_discard_flags(discard
, &bs
->open_flags
) != 0) {
1525 error_setg(errp
, "Invalid discard option");
1532 bdrv_parse_detect_zeroes(opts
, bs
->open_flags
, &local_err
);
1534 error_propagate(errp
, local_err
);
1539 if (filename
!= NULL
) {
1540 pstrcpy(bs
->filename
, sizeof(bs
->filename
), filename
);
1542 bs
->filename
[0] = '\0';
1544 pstrcpy(bs
->exact_filename
, sizeof(bs
->exact_filename
), bs
->filename
);
1546 /* Open the image, either directly or using a protocol */
1547 open_flags
= bdrv_open_flags(bs
, bs
->open_flags
);
1548 node_name
= qemu_opt_get(opts
, "node-name");
1550 assert(!drv
->bdrv_file_open
|| file
== NULL
);
1551 ret
= bdrv_open_driver(bs
, drv
, node_name
, options
, open_flags
, errp
);
1556 qemu_opts_del(opts
);
1560 qemu_opts_del(opts
);
1564 static QDict
*parse_json_filename(const char *filename
, Error
**errp
)
1566 QObject
*options_obj
;
1570 ret
= strstart(filename
, "json:", &filename
);
1573 options_obj
= qobject_from_json(filename
, errp
);
1575 error_prepend(errp
, "Could not parse the JSON options: ");
1579 options
= qobject_to(QDict
, options_obj
);
1581 qobject_unref(options_obj
);
1582 error_setg(errp
, "Invalid JSON object given");
1586 qdict_flatten(options
);
1591 static void parse_json_protocol(QDict
*options
, const char **pfilename
,
1594 QDict
*json_options
;
1595 Error
*local_err
= NULL
;
1597 /* Parse json: pseudo-protocol */
1598 if (!*pfilename
|| !g_str_has_prefix(*pfilename
, "json:")) {
1602 json_options
= parse_json_filename(*pfilename
, &local_err
);
1604 error_propagate(errp
, local_err
);
1608 /* Options given in the filename have lower priority than options
1609 * specified directly */
1610 qdict_join(options
, json_options
, false);
1611 qobject_unref(json_options
);
1616 * Fills in default options for opening images and converts the legacy
1617 * filename/flags pair to option QDict entries.
1618 * The BDRV_O_PROTOCOL flag in *flags will be set or cleared accordingly if a
1619 * block driver has been specified explicitly.
1621 static int bdrv_fill_options(QDict
**options
, const char *filename
,
1622 int *flags
, Error
**errp
)
1624 const char *drvname
;
1625 bool protocol
= *flags
& BDRV_O_PROTOCOL
;
1626 bool parse_filename
= false;
1627 BlockDriver
*drv
= NULL
;
1628 Error
*local_err
= NULL
;
1631 * Caution: while qdict_get_try_str() is fine, getting non-string
1632 * types would require more care. When @options come from
1633 * -blockdev or blockdev_add, its members are typed according to
1634 * the QAPI schema, but when they come from -drive, they're all
1637 drvname
= qdict_get_try_str(*options
, "driver");
1639 drv
= bdrv_find_format(drvname
);
1641 error_setg(errp
, "Unknown driver '%s'", drvname
);
1644 /* If the user has explicitly specified the driver, this choice should
1645 * override the BDRV_O_PROTOCOL flag */
1646 protocol
= drv
->bdrv_file_open
;
1650 *flags
|= BDRV_O_PROTOCOL
;
1652 *flags
&= ~BDRV_O_PROTOCOL
;
1655 /* Translate cache options from flags into options */
1656 update_options_from_flags(*options
, *flags
);
1658 /* Fetch the file name from the options QDict if necessary */
1659 if (protocol
&& filename
) {
1660 if (!qdict_haskey(*options
, "filename")) {
1661 qdict_put_str(*options
, "filename", filename
);
1662 parse_filename
= true;
1664 error_setg(errp
, "Can't specify 'file' and 'filename' options at "
1670 /* Find the right block driver */
1671 /* See cautionary note on accessing @options above */
1672 filename
= qdict_get_try_str(*options
, "filename");
1674 if (!drvname
&& protocol
) {
1676 drv
= bdrv_find_protocol(filename
, parse_filename
, errp
);
1681 drvname
= drv
->format_name
;
1682 qdict_put_str(*options
, "driver", drvname
);
1684 error_setg(errp
, "Must specify either driver or file");
1689 assert(drv
|| !protocol
);
1691 /* Driver-specific filename parsing */
1692 if (drv
&& drv
->bdrv_parse_filename
&& parse_filename
) {
1693 drv
->bdrv_parse_filename(filename
, *options
, &local_err
);
1695 error_propagate(errp
, local_err
);
1699 if (!drv
->bdrv_needs_filename
) {
1700 qdict_del(*options
, "filename");
1707 static int bdrv_child_check_perm(BdrvChild
*c
, BlockReopenQueue
*q
,
1708 uint64_t perm
, uint64_t shared
,
1709 GSList
*ignore_children
, Error
**errp
);
1710 static void bdrv_child_abort_perm_update(BdrvChild
*c
);
1711 static void bdrv_child_set_perm(BdrvChild
*c
, uint64_t perm
, uint64_t shared
);
1713 typedef struct BlockReopenQueueEntry
{
1716 BDRVReopenState state
;
1717 QSIMPLEQ_ENTRY(BlockReopenQueueEntry
) entry
;
1718 } BlockReopenQueueEntry
;
1721 * Return the flags that @bs will have after the reopens in @q have
1722 * successfully completed. If @q is NULL (or @bs is not contained in @q),
1723 * return the current flags.
1725 static int bdrv_reopen_get_flags(BlockReopenQueue
*q
, BlockDriverState
*bs
)
1727 BlockReopenQueueEntry
*entry
;
1730 QSIMPLEQ_FOREACH(entry
, q
, entry
) {
1731 if (entry
->state
.bs
== bs
) {
1732 return entry
->state
.flags
;
1737 return bs
->open_flags
;
1740 /* Returns whether the image file can be written to after the reopen queue @q
1741 * has been successfully applied, or right now if @q is NULL. */
1742 static bool bdrv_is_writable_after_reopen(BlockDriverState
*bs
,
1743 BlockReopenQueue
*q
)
1745 int flags
= bdrv_reopen_get_flags(q
, bs
);
1747 return (flags
& (BDRV_O_RDWR
| BDRV_O_INACTIVE
)) == BDRV_O_RDWR
;
1751 * Return whether the BDS can be written to. This is not necessarily
1752 * the same as !bdrv_is_read_only(bs), as inactivated images may not
1753 * be written to but do not count as read-only images.
1755 bool bdrv_is_writable(BlockDriverState
*bs
)
1757 return bdrv_is_writable_after_reopen(bs
, NULL
);
1760 static void bdrv_child_perm(BlockDriverState
*bs
, BlockDriverState
*child_bs
,
1761 BdrvChild
*c
, const BdrvChildRole
*role
,
1762 BlockReopenQueue
*reopen_queue
,
1763 uint64_t parent_perm
, uint64_t parent_shared
,
1764 uint64_t *nperm
, uint64_t *nshared
)
1766 assert(bs
->drv
&& bs
->drv
->bdrv_child_perm
);
1767 bs
->drv
->bdrv_child_perm(bs
, c
, role
, reopen_queue
,
1768 parent_perm
, parent_shared
,
1770 /* TODO Take force_share from reopen_queue */
1771 if (child_bs
&& child_bs
->force_share
) {
1772 *nshared
= BLK_PERM_ALL
;
1777 * Check whether permissions on this node can be changed in a way that
1778 * @cumulative_perms and @cumulative_shared_perms are the new cumulative
1779 * permissions of all its parents. This involves checking whether all necessary
1780 * permission changes to child nodes can be performed.
1782 * A call to this function must always be followed by a call to bdrv_set_perm()
1783 * or bdrv_abort_perm_update().
1785 static int bdrv_check_perm(BlockDriverState
*bs
, BlockReopenQueue
*q
,
1786 uint64_t cumulative_perms
,
1787 uint64_t cumulative_shared_perms
,
1788 GSList
*ignore_children
, Error
**errp
)
1790 BlockDriver
*drv
= bs
->drv
;
1794 /* Write permissions never work with read-only images */
1795 if ((cumulative_perms
& (BLK_PERM_WRITE
| BLK_PERM_WRITE_UNCHANGED
)) &&
1796 !bdrv_is_writable_after_reopen(bs
, q
))
1798 error_setg(errp
, "Block node is read-only");
1802 /* Check this node */
1807 if (drv
->bdrv_check_perm
) {
1808 return drv
->bdrv_check_perm(bs
, cumulative_perms
,
1809 cumulative_shared_perms
, errp
);
1812 /* Drivers that never have children can omit .bdrv_child_perm() */
1813 if (!drv
->bdrv_child_perm
) {
1814 assert(QLIST_EMPTY(&bs
->children
));
1818 /* Check all children */
1819 QLIST_FOREACH(c
, &bs
->children
, next
) {
1820 uint64_t cur_perm
, cur_shared
;
1821 bdrv_child_perm(bs
, c
->bs
, c
, c
->role
, q
,
1822 cumulative_perms
, cumulative_shared_perms
,
1823 &cur_perm
, &cur_shared
);
1824 ret
= bdrv_child_check_perm(c
, q
, cur_perm
, cur_shared
,
1825 ignore_children
, errp
);
1835 * Notifies drivers that after a previous bdrv_check_perm() call, the
1836 * permission update is not performed and any preparations made for it (e.g.
1837 * taken file locks) need to be undone.
1839 * This function recursively notifies all child nodes.
1841 static void bdrv_abort_perm_update(BlockDriverState
*bs
)
1843 BlockDriver
*drv
= bs
->drv
;
1850 if (drv
->bdrv_abort_perm_update
) {
1851 drv
->bdrv_abort_perm_update(bs
);
1854 QLIST_FOREACH(c
, &bs
->children
, next
) {
1855 bdrv_child_abort_perm_update(c
);
1859 static void bdrv_set_perm(BlockDriverState
*bs
, uint64_t cumulative_perms
,
1860 uint64_t cumulative_shared_perms
)
1862 BlockDriver
*drv
= bs
->drv
;
1869 /* Update this node */
1870 if (drv
->bdrv_set_perm
) {
1871 drv
->bdrv_set_perm(bs
, cumulative_perms
, cumulative_shared_perms
);
1874 /* Drivers that never have children can omit .bdrv_child_perm() */
1875 if (!drv
->bdrv_child_perm
) {
1876 assert(QLIST_EMPTY(&bs
->children
));
1880 /* Update all children */
1881 QLIST_FOREACH(c
, &bs
->children
, next
) {
1882 uint64_t cur_perm
, cur_shared
;
1883 bdrv_child_perm(bs
, c
->bs
, c
, c
->role
, NULL
,
1884 cumulative_perms
, cumulative_shared_perms
,
1885 &cur_perm
, &cur_shared
);
1886 bdrv_child_set_perm(c
, cur_perm
, cur_shared
);
1890 static void bdrv_get_cumulative_perm(BlockDriverState
*bs
, uint64_t *perm
,
1891 uint64_t *shared_perm
)
1894 uint64_t cumulative_perms
= 0;
1895 uint64_t cumulative_shared_perms
= BLK_PERM_ALL
;
1897 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
1898 cumulative_perms
|= c
->perm
;
1899 cumulative_shared_perms
&= c
->shared_perm
;
1902 *perm
= cumulative_perms
;
1903 *shared_perm
= cumulative_shared_perms
;
1906 static char *bdrv_child_user_desc(BdrvChild
*c
)
1908 if (c
->role
->get_parent_desc
) {
1909 return c
->role
->get_parent_desc(c
);
1912 return g_strdup("another user");
1915 char *bdrv_perm_names(uint64_t perm
)
1921 { BLK_PERM_CONSISTENT_READ
, "consistent read" },
1922 { BLK_PERM_WRITE
, "write" },
1923 { BLK_PERM_WRITE_UNCHANGED
, "write unchanged" },
1924 { BLK_PERM_RESIZE
, "resize" },
1925 { BLK_PERM_GRAPH_MOD
, "change children" },
1929 char *result
= g_strdup("");
1930 struct perm_name
*p
;
1932 for (p
= permissions
; p
->name
; p
++) {
1933 if (perm
& p
->perm
) {
1935 result
= g_strdup_printf("%s%s%s", old
, *old
? ", " : "", p
->name
);
1944 * Checks whether a new reference to @bs can be added if the new user requires
1945 * @new_used_perm/@new_shared_perm as its permissions. If @ignore_children is
1946 * set, the BdrvChild objects in this list are ignored in the calculations;
1947 * this allows checking permission updates for an existing reference.
1949 * Needs to be followed by a call to either bdrv_set_perm() or
1950 * bdrv_abort_perm_update(). */
1951 static int bdrv_check_update_perm(BlockDriverState
*bs
, BlockReopenQueue
*q
,
1952 uint64_t new_used_perm
,
1953 uint64_t new_shared_perm
,
1954 GSList
*ignore_children
, Error
**errp
)
1957 uint64_t cumulative_perms
= new_used_perm
;
1958 uint64_t cumulative_shared_perms
= new_shared_perm
;
1960 /* There is no reason why anyone couldn't tolerate write_unchanged */
1961 assert(new_shared_perm
& BLK_PERM_WRITE_UNCHANGED
);
1963 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
1964 if (g_slist_find(ignore_children
, c
)) {
1968 if ((new_used_perm
& c
->shared_perm
) != new_used_perm
) {
1969 char *user
= bdrv_child_user_desc(c
);
1970 char *perm_names
= bdrv_perm_names(new_used_perm
& ~c
->shared_perm
);
1971 error_setg(errp
, "Conflicts with use by %s as '%s', which does not "
1973 user
, c
->name
, perm_names
, bdrv_get_node_name(c
->bs
));
1979 if ((c
->perm
& new_shared_perm
) != c
->perm
) {
1980 char *user
= bdrv_child_user_desc(c
);
1981 char *perm_names
= bdrv_perm_names(c
->perm
& ~new_shared_perm
);
1982 error_setg(errp
, "Conflicts with use by %s as '%s', which uses "
1984 user
, c
->name
, perm_names
, bdrv_get_node_name(c
->bs
));
1990 cumulative_perms
|= c
->perm
;
1991 cumulative_shared_perms
&= c
->shared_perm
;
1994 return bdrv_check_perm(bs
, q
, cumulative_perms
, cumulative_shared_perms
,
1995 ignore_children
, errp
);
1998 /* Needs to be followed by a call to either bdrv_child_set_perm() or
1999 * bdrv_child_abort_perm_update(). */
2000 static int bdrv_child_check_perm(BdrvChild
*c
, BlockReopenQueue
*q
,
2001 uint64_t perm
, uint64_t shared
,
2002 GSList
*ignore_children
, Error
**errp
)
2006 ignore_children
= g_slist_prepend(g_slist_copy(ignore_children
), c
);
2007 ret
= bdrv_check_update_perm(c
->bs
, q
, perm
, shared
, ignore_children
, errp
);
2008 g_slist_free(ignore_children
);
2014 if (!c
->has_backup_perm
) {
2015 c
->has_backup_perm
= true;
2016 c
->backup_perm
= c
->perm
;
2017 c
->backup_shared_perm
= c
->shared_perm
;
2020 * Note: it's OK if c->has_backup_perm was already set, as we can find the
2021 * same child twice during check_perm procedure
2025 c
->shared_perm
= shared
;
2030 static void bdrv_child_set_perm(BdrvChild
*c
, uint64_t perm
, uint64_t shared
)
2032 uint64_t cumulative_perms
, cumulative_shared_perms
;
2034 c
->has_backup_perm
= false;
2037 c
->shared_perm
= shared
;
2039 bdrv_get_cumulative_perm(c
->bs
, &cumulative_perms
,
2040 &cumulative_shared_perms
);
2041 bdrv_set_perm(c
->bs
, cumulative_perms
, cumulative_shared_perms
);
2044 static void bdrv_child_abort_perm_update(BdrvChild
*c
)
2046 if (c
->has_backup_perm
) {
2047 c
->perm
= c
->backup_perm
;
2048 c
->shared_perm
= c
->backup_shared_perm
;
2049 c
->has_backup_perm
= false;
2052 bdrv_abort_perm_update(c
->bs
);
2055 int bdrv_child_try_set_perm(BdrvChild
*c
, uint64_t perm
, uint64_t shared
,
2060 ret
= bdrv_child_check_perm(c
, NULL
, perm
, shared
, NULL
, errp
);
2062 bdrv_child_abort_perm_update(c
);
2066 bdrv_child_set_perm(c
, perm
, shared
);
2071 void bdrv_filter_default_perms(BlockDriverState
*bs
, BdrvChild
*c
,
2072 const BdrvChildRole
*role
,
2073 BlockReopenQueue
*reopen_queue
,
2074 uint64_t perm
, uint64_t shared
,
2075 uint64_t *nperm
, uint64_t *nshared
)
2078 *nperm
= perm
& DEFAULT_PERM_PASSTHROUGH
;
2079 *nshared
= (shared
& DEFAULT_PERM_PASSTHROUGH
) | DEFAULT_PERM_UNCHANGED
;
2083 *nperm
= (perm
& DEFAULT_PERM_PASSTHROUGH
) |
2084 (c
->perm
& DEFAULT_PERM_UNCHANGED
);
2085 *nshared
= (shared
& DEFAULT_PERM_PASSTHROUGH
) |
2086 (c
->shared_perm
& DEFAULT_PERM_UNCHANGED
);
2089 void bdrv_format_default_perms(BlockDriverState
*bs
, BdrvChild
*c
,
2090 const BdrvChildRole
*role
,
2091 BlockReopenQueue
*reopen_queue
,
2092 uint64_t perm
, uint64_t shared
,
2093 uint64_t *nperm
, uint64_t *nshared
)
2095 bool backing
= (role
== &child_backing
);
2096 assert(role
== &child_backing
|| role
== &child_file
);
2099 int flags
= bdrv_reopen_get_flags(reopen_queue
, bs
);
2101 /* Apart from the modifications below, the same permissions are
2102 * forwarded and left alone as for filters */
2103 bdrv_filter_default_perms(bs
, c
, role
, reopen_queue
, perm
, shared
,
2106 /* Format drivers may touch metadata even if the guest doesn't write */
2107 if (bdrv_is_writable_after_reopen(bs
, reopen_queue
)) {
2108 perm
|= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2111 /* bs->file always needs to be consistent because of the metadata. We
2112 * can never allow other users to resize or write to it. */
2113 if (!(flags
& BDRV_O_NO_IO
)) {
2114 perm
|= BLK_PERM_CONSISTENT_READ
;
2116 shared
&= ~(BLK_PERM_WRITE
| BLK_PERM_RESIZE
);
2118 /* We want consistent read from backing files if the parent needs it.
2119 * No other operations are performed on backing files. */
2120 perm
&= BLK_PERM_CONSISTENT_READ
;
2122 /* If the parent can deal with changing data, we're okay with a
2123 * writable and resizable backing file. */
2124 /* TODO Require !(perm & BLK_PERM_CONSISTENT_READ), too? */
2125 if (shared
& BLK_PERM_WRITE
) {
2126 shared
= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2131 shared
|= BLK_PERM_CONSISTENT_READ
| BLK_PERM_GRAPH_MOD
|
2132 BLK_PERM_WRITE_UNCHANGED
;
2135 if (bs
->open_flags
& BDRV_O_INACTIVE
) {
2136 shared
|= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2143 static void bdrv_replace_child_noperm(BdrvChild
*child
,
2144 BlockDriverState
*new_bs
)
2146 BlockDriverState
*old_bs
= child
->bs
;
2149 assert(!child
->frozen
);
2151 if (old_bs
&& new_bs
) {
2152 assert(bdrv_get_aio_context(old_bs
) == bdrv_get_aio_context(new_bs
));
2155 /* Detach first so that the recursive drain sections coming from @child
2156 * are already gone and we only end the drain sections that came from
2158 if (child
->role
->detach
) {
2159 child
->role
->detach(child
);
2161 if (old_bs
->quiesce_counter
&& child
->role
->drained_end
) {
2162 int num
= old_bs
->quiesce_counter
;
2163 if (child
->role
->parent_is_bds
) {
2164 num
-= bdrv_drain_all_count
;
2167 for (i
= 0; i
< num
; i
++) {
2168 child
->role
->drained_end(child
);
2171 QLIST_REMOVE(child
, next_parent
);
2177 QLIST_INSERT_HEAD(&new_bs
->parents
, child
, next_parent
);
2178 if (new_bs
->quiesce_counter
&& child
->role
->drained_begin
) {
2179 int num
= new_bs
->quiesce_counter
;
2180 if (child
->role
->parent_is_bds
) {
2181 num
-= bdrv_drain_all_count
;
2184 for (i
= 0; i
< num
; i
++) {
2185 bdrv_parent_drained_begin_single(child
, true);
2189 /* Attach only after starting new drained sections, so that recursive
2190 * drain sections coming from @child don't get an extra .drained_begin
2192 if (child
->role
->attach
) {
2193 child
->role
->attach(child
);
2199 * Updates @child to change its reference to point to @new_bs, including
2200 * checking and applying the necessary permisson updates both to the old node
2203 * NULL is passed as @new_bs for removing the reference before freeing @child.
2205 * If @new_bs is not NULL, bdrv_check_perm() must be called beforehand, as this
2206 * function uses bdrv_set_perm() to update the permissions according to the new
2207 * reference that @new_bs gets.
2209 static void bdrv_replace_child(BdrvChild
*child
, BlockDriverState
*new_bs
)
2211 BlockDriverState
*old_bs
= child
->bs
;
2212 uint64_t perm
, shared_perm
;
2214 bdrv_replace_child_noperm(child
, new_bs
);
2217 /* Update permissions for old node. This is guaranteed to succeed
2218 * because we're just taking a parent away, so we're loosening
2220 bdrv_get_cumulative_perm(old_bs
, &perm
, &shared_perm
);
2221 bdrv_check_perm(old_bs
, NULL
, perm
, shared_perm
, NULL
, &error_abort
);
2222 bdrv_set_perm(old_bs
, perm
, shared_perm
);
2226 bdrv_get_cumulative_perm(new_bs
, &perm
, &shared_perm
);
2227 bdrv_set_perm(new_bs
, perm
, shared_perm
);
2231 BdrvChild
*bdrv_root_attach_child(BlockDriverState
*child_bs
,
2232 const char *child_name
,
2233 const BdrvChildRole
*child_role
,
2234 uint64_t perm
, uint64_t shared_perm
,
2235 void *opaque
, Error
**errp
)
2240 ret
= bdrv_check_update_perm(child_bs
, NULL
, perm
, shared_perm
, NULL
, errp
);
2242 bdrv_abort_perm_update(child_bs
);
2246 child
= g_new(BdrvChild
, 1);
2247 *child
= (BdrvChild
) {
2249 .name
= g_strdup(child_name
),
2252 .shared_perm
= shared_perm
,
2256 /* This performs the matching bdrv_set_perm() for the above check. */
2257 bdrv_replace_child(child
, child_bs
);
2262 BdrvChild
*bdrv_attach_child(BlockDriverState
*parent_bs
,
2263 BlockDriverState
*child_bs
,
2264 const char *child_name
,
2265 const BdrvChildRole
*child_role
,
2269 uint64_t perm
, shared_perm
;
2271 bdrv_get_cumulative_perm(parent_bs
, &perm
, &shared_perm
);
2273 assert(parent_bs
->drv
);
2274 assert(bdrv_get_aio_context(parent_bs
) == bdrv_get_aio_context(child_bs
));
2275 bdrv_child_perm(parent_bs
, child_bs
, NULL
, child_role
, NULL
,
2276 perm
, shared_perm
, &perm
, &shared_perm
);
2278 child
= bdrv_root_attach_child(child_bs
, child_name
, child_role
,
2279 perm
, shared_perm
, parent_bs
, errp
);
2280 if (child
== NULL
) {
2284 QLIST_INSERT_HEAD(&parent_bs
->children
, child
, next
);
2288 static void bdrv_detach_child(BdrvChild
*child
)
2290 if (child
->next
.le_prev
) {
2291 QLIST_REMOVE(child
, next
);
2292 child
->next
.le_prev
= NULL
;
2295 bdrv_replace_child(child
, NULL
);
2297 g_free(child
->name
);
2301 void bdrv_root_unref_child(BdrvChild
*child
)
2303 BlockDriverState
*child_bs
;
2305 child_bs
= child
->bs
;
2306 bdrv_detach_child(child
);
2307 bdrv_unref(child_bs
);
2310 void bdrv_unref_child(BlockDriverState
*parent
, BdrvChild
*child
)
2312 if (child
== NULL
) {
2316 if (child
->bs
->inherits_from
== parent
) {
2319 /* Remove inherits_from only when the last reference between parent and
2320 * child->bs goes away. */
2321 QLIST_FOREACH(c
, &parent
->children
, next
) {
2322 if (c
!= child
&& c
->bs
== child
->bs
) {
2327 child
->bs
->inherits_from
= NULL
;
2331 bdrv_root_unref_child(child
);
2335 static void bdrv_parent_cb_change_media(BlockDriverState
*bs
, bool load
)
2338 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
2339 if (c
->role
->change_media
) {
2340 c
->role
->change_media(c
, load
);
2345 /* Return true if you can reach parent going through child->inherits_from
2346 * recursively. If parent or child are NULL, return false */
2347 static bool bdrv_inherits_from_recursive(BlockDriverState
*child
,
2348 BlockDriverState
*parent
)
2350 while (child
&& child
!= parent
) {
2351 child
= child
->inherits_from
;
2354 return child
!= NULL
;
2358 * Sets the backing file link of a BDS. A new reference is created; callers
2359 * which don't need their own reference any more must call bdrv_unref().
2361 void bdrv_set_backing_hd(BlockDriverState
*bs
, BlockDriverState
*backing_hd
,
2364 bool update_inherits_from
= bdrv_chain_contains(bs
, backing_hd
) &&
2365 bdrv_inherits_from_recursive(backing_hd
, bs
);
2367 if (bdrv_is_backing_chain_frozen(bs
, backing_bs(bs
), errp
)) {
2372 bdrv_ref(backing_hd
);
2376 bdrv_unref_child(bs
, bs
->backing
);
2384 bs
->backing
= bdrv_attach_child(bs
, backing_hd
, "backing", &child_backing
,
2386 /* If backing_hd was already part of bs's backing chain, and
2387 * inherits_from pointed recursively to bs then let's update it to
2388 * point directly to bs (else it will become NULL). */
2389 if (update_inherits_from
) {
2390 backing_hd
->inherits_from
= bs
;
2393 bdrv_unref(backing_hd
);
2397 bdrv_refresh_limits(bs
, NULL
);
2401 * Opens the backing file for a BlockDriverState if not yet open
2403 * bdref_key specifies the key for the image's BlockdevRef in the options QDict.
2404 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
2405 * itself, all options starting with "${bdref_key}." are considered part of the
2408 * TODO Can this be unified with bdrv_open_image()?
2410 int bdrv_open_backing_file(BlockDriverState
*bs
, QDict
*parent_options
,
2411 const char *bdref_key
, Error
**errp
)
2413 char *backing_filename
= NULL
;
2414 char *bdref_key_dot
;
2415 const char *reference
= NULL
;
2417 bool implicit_backing
= false;
2418 BlockDriverState
*backing_hd
;
2420 QDict
*tmp_parent_options
= NULL
;
2421 Error
*local_err
= NULL
;
2423 if (bs
->backing
!= NULL
) {
2427 /* NULL means an empty set of options */
2428 if (parent_options
== NULL
) {
2429 tmp_parent_options
= qdict_new();
2430 parent_options
= tmp_parent_options
;
2433 bs
->open_flags
&= ~BDRV_O_NO_BACKING
;
2435 bdref_key_dot
= g_strdup_printf("%s.", bdref_key
);
2436 qdict_extract_subqdict(parent_options
, &options
, bdref_key_dot
);
2437 g_free(bdref_key_dot
);
2440 * Caution: while qdict_get_try_str() is fine, getting non-string
2441 * types would require more care. When @parent_options come from
2442 * -blockdev or blockdev_add, its members are typed according to
2443 * the QAPI schema, but when they come from -drive, they're all
2446 reference
= qdict_get_try_str(parent_options
, bdref_key
);
2447 if (reference
|| qdict_haskey(options
, "file.filename")) {
2448 /* keep backing_filename NULL */
2449 } else if (bs
->backing_file
[0] == '\0' && qdict_size(options
) == 0) {
2450 qobject_unref(options
);
2453 if (qdict_size(options
) == 0) {
2454 /* If the user specifies options that do not modify the
2455 * backing file's behavior, we might still consider it the
2456 * implicit backing file. But it's easier this way, and
2457 * just specifying some of the backing BDS's options is
2458 * only possible with -drive anyway (otherwise the QAPI
2459 * schema forces the user to specify everything). */
2460 implicit_backing
= !strcmp(bs
->auto_backing_file
, bs
->backing_file
);
2463 backing_filename
= bdrv_get_full_backing_filename(bs
, &local_err
);
2466 error_propagate(errp
, local_err
);
2467 qobject_unref(options
);
2472 if (!bs
->drv
|| !bs
->drv
->supports_backing
) {
2474 error_setg(errp
, "Driver doesn't support backing files");
2475 qobject_unref(options
);
2480 bs
->backing_format
[0] != '\0' && !qdict_haskey(options
, "driver")) {
2481 qdict_put_str(options
, "driver", bs
->backing_format
);
2484 backing_hd
= bdrv_open_inherit(backing_filename
, reference
, options
, 0, bs
,
2485 &child_backing
, errp
);
2487 bs
->open_flags
|= BDRV_O_NO_BACKING
;
2488 error_prepend(errp
, "Could not open backing file: ");
2492 bdrv_set_aio_context(backing_hd
, bdrv_get_aio_context(bs
));
2494 if (implicit_backing
) {
2495 bdrv_refresh_filename(backing_hd
);
2496 pstrcpy(bs
->auto_backing_file
, sizeof(bs
->auto_backing_file
),
2497 backing_hd
->filename
);
2500 /* Hook up the backing file link; drop our reference, bs owns the
2501 * backing_hd reference now */
2502 bdrv_set_backing_hd(bs
, backing_hd
, &local_err
);
2503 bdrv_unref(backing_hd
);
2505 error_propagate(errp
, local_err
);
2510 qdict_del(parent_options
, bdref_key
);
2513 g_free(backing_filename
);
2514 qobject_unref(tmp_parent_options
);
2518 static BlockDriverState
*
2519 bdrv_open_child_bs(const char *filename
, QDict
*options
, const char *bdref_key
,
2520 BlockDriverState
*parent
, const BdrvChildRole
*child_role
,
2521 bool allow_none
, Error
**errp
)
2523 BlockDriverState
*bs
= NULL
;
2524 QDict
*image_options
;
2525 char *bdref_key_dot
;
2526 const char *reference
;
2528 assert(child_role
!= NULL
);
2530 bdref_key_dot
= g_strdup_printf("%s.", bdref_key
);
2531 qdict_extract_subqdict(options
, &image_options
, bdref_key_dot
);
2532 g_free(bdref_key_dot
);
2535 * Caution: while qdict_get_try_str() is fine, getting non-string
2536 * types would require more care. When @options come from
2537 * -blockdev or blockdev_add, its members are typed according to
2538 * the QAPI schema, but when they come from -drive, they're all
2541 reference
= qdict_get_try_str(options
, bdref_key
);
2542 if (!filename
&& !reference
&& !qdict_size(image_options
)) {
2544 error_setg(errp
, "A block device must be specified for \"%s\"",
2547 qobject_unref(image_options
);
2551 bs
= bdrv_open_inherit(filename
, reference
, image_options
, 0,
2552 parent
, child_role
, errp
);
2558 qdict_del(options
, bdref_key
);
2563 * Opens a disk image whose options are given as BlockdevRef in another block
2566 * If allow_none is true, no image will be opened if filename is false and no
2567 * BlockdevRef is given. NULL will be returned, but errp remains unset.
2569 * bdrev_key specifies the key for the image's BlockdevRef in the options QDict.
2570 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
2571 * itself, all options starting with "${bdref_key}." are considered part of the
2574 * The BlockdevRef will be removed from the options QDict.
2576 BdrvChild
*bdrv_open_child(const char *filename
,
2577 QDict
*options
, const char *bdref_key
,
2578 BlockDriverState
*parent
,
2579 const BdrvChildRole
*child_role
,
2580 bool allow_none
, Error
**errp
)
2583 BlockDriverState
*bs
;
2585 bs
= bdrv_open_child_bs(filename
, options
, bdref_key
, parent
, child_role
,
2591 c
= bdrv_attach_child(parent
, bs
, bdref_key
, child_role
, errp
);
2600 /* TODO Future callers may need to specify parent/child_role in order for
2601 * option inheritance to work. Existing callers use it for the root node. */
2602 BlockDriverState
*bdrv_open_blockdev_ref(BlockdevRef
*ref
, Error
**errp
)
2604 BlockDriverState
*bs
= NULL
;
2605 Error
*local_err
= NULL
;
2606 QObject
*obj
= NULL
;
2607 QDict
*qdict
= NULL
;
2608 const char *reference
= NULL
;
2611 if (ref
->type
== QTYPE_QSTRING
) {
2612 reference
= ref
->u
.reference
;
2614 BlockdevOptions
*options
= &ref
->u
.definition
;
2615 assert(ref
->type
== QTYPE_QDICT
);
2617 v
= qobject_output_visitor_new(&obj
);
2618 visit_type_BlockdevOptions(v
, NULL
, &options
, &local_err
);
2620 error_propagate(errp
, local_err
);
2623 visit_complete(v
, &obj
);
2625 qdict
= qobject_to(QDict
, obj
);
2626 qdict_flatten(qdict
);
2628 /* bdrv_open_inherit() defaults to the values in bdrv_flags (for
2629 * compatibility with other callers) rather than what we want as the
2630 * real defaults. Apply the defaults here instead. */
2631 qdict_set_default_str(qdict
, BDRV_OPT_CACHE_DIRECT
, "off");
2632 qdict_set_default_str(qdict
, BDRV_OPT_CACHE_NO_FLUSH
, "off");
2633 qdict_set_default_str(qdict
, BDRV_OPT_READ_ONLY
, "off");
2634 qdict_set_default_str(qdict
, BDRV_OPT_AUTO_READ_ONLY
, "off");
2638 bs
= bdrv_open_inherit(NULL
, reference
, qdict
, 0, NULL
, NULL
, errp
);
2647 static BlockDriverState
*bdrv_append_temp_snapshot(BlockDriverState
*bs
,
2649 QDict
*snapshot_options
,
2652 /* TODO: extra byte is a hack to ensure MAX_PATH space on Windows. */
2653 char *tmp_filename
= g_malloc0(PATH_MAX
+ 1);
2655 QemuOpts
*opts
= NULL
;
2656 BlockDriverState
*bs_snapshot
= NULL
;
2657 Error
*local_err
= NULL
;
2660 /* if snapshot, we create a temporary backing file and open it
2661 instead of opening 'filename' directly */
2663 /* Get the required size from the image */
2664 total_size
= bdrv_getlength(bs
);
2665 if (total_size
< 0) {
2666 error_setg_errno(errp
, -total_size
, "Could not get image size");
2670 /* Create the temporary image */
2671 ret
= get_tmp_filename(tmp_filename
, PATH_MAX
+ 1);
2673 error_setg_errno(errp
, -ret
, "Could not get temporary filename");
2677 opts
= qemu_opts_create(bdrv_qcow2
.create_opts
, NULL
, 0,
2679 qemu_opt_set_number(opts
, BLOCK_OPT_SIZE
, total_size
, &error_abort
);
2680 ret
= bdrv_create(&bdrv_qcow2
, tmp_filename
, opts
, errp
);
2681 qemu_opts_del(opts
);
2683 error_prepend(errp
, "Could not create temporary overlay '%s': ",
2688 /* Prepare options QDict for the temporary file */
2689 qdict_put_str(snapshot_options
, "file.driver", "file");
2690 qdict_put_str(snapshot_options
, "file.filename", tmp_filename
);
2691 qdict_put_str(snapshot_options
, "driver", "qcow2");
2693 bs_snapshot
= bdrv_open(NULL
, NULL
, snapshot_options
, flags
, errp
);
2694 snapshot_options
= NULL
;
2699 /* bdrv_append() consumes a strong reference to bs_snapshot
2700 * (i.e. it will call bdrv_unref() on it) even on error, so in
2701 * order to be able to return one, we have to increase
2702 * bs_snapshot's refcount here */
2703 bdrv_ref(bs_snapshot
);
2704 bdrv_append(bs_snapshot
, bs
, &local_err
);
2706 error_propagate(errp
, local_err
);
2712 qobject_unref(snapshot_options
);
2713 g_free(tmp_filename
);
2718 * Opens a disk image (raw, qcow2, vmdk, ...)
2720 * options is a QDict of options to pass to the block drivers, or NULL for an
2721 * empty set of options. The reference to the QDict belongs to the block layer
2722 * after the call (even on failure), so if the caller intends to reuse the
2723 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
2725 * If *pbs is NULL, a new BDS will be created with a pointer to it stored there.
2726 * If it is not NULL, the referenced BDS will be reused.
2728 * The reference parameter may be used to specify an existing block device which
2729 * should be opened. If specified, neither options nor a filename may be given,
2730 * nor can an existing BDS be reused (that is, *pbs has to be NULL).
2732 static BlockDriverState
*bdrv_open_inherit(const char *filename
,
2733 const char *reference
,
2734 QDict
*options
, int flags
,
2735 BlockDriverState
*parent
,
2736 const BdrvChildRole
*child_role
,
2740 BlockBackend
*file
= NULL
;
2741 BlockDriverState
*bs
;
2742 BlockDriver
*drv
= NULL
;
2744 const char *drvname
;
2745 const char *backing
;
2746 Error
*local_err
= NULL
;
2747 QDict
*snapshot_options
= NULL
;
2748 int snapshot_flags
= 0;
2750 assert(!child_role
|| !flags
);
2751 assert(!child_role
== !parent
);
2754 bool options_non_empty
= options
? qdict_size(options
) : false;
2755 qobject_unref(options
);
2757 if (filename
|| options_non_empty
) {
2758 error_setg(errp
, "Cannot reference an existing block device with "
2759 "additional options or a new filename");
2763 bs
= bdrv_lookup_bs(reference
, reference
, errp
);
2774 /* NULL means an empty set of options */
2775 if (options
== NULL
) {
2776 options
= qdict_new();
2779 /* json: syntax counts as explicit options, as if in the QDict */
2780 parse_json_protocol(options
, &filename
, &local_err
);
2785 bs
->explicit_options
= qdict_clone_shallow(options
);
2788 bs
->inherits_from
= parent
;
2789 child_role
->inherit_options(&flags
, options
,
2790 parent
->open_flags
, parent
->options
);
2793 ret
= bdrv_fill_options(&options
, filename
, &flags
, &local_err
);
2799 * Set the BDRV_O_RDWR and BDRV_O_ALLOW_RDWR flags.
2800 * Caution: getting a boolean member of @options requires care.
2801 * When @options come from -blockdev or blockdev_add, members are
2802 * typed according to the QAPI schema, but when they come from
2803 * -drive, they're all QString.
2805 if (g_strcmp0(qdict_get_try_str(options
, BDRV_OPT_READ_ONLY
), "on") &&
2806 !qdict_get_try_bool(options
, BDRV_OPT_READ_ONLY
, false)) {
2807 flags
|= (BDRV_O_RDWR
| BDRV_O_ALLOW_RDWR
);
2809 flags
&= ~BDRV_O_RDWR
;
2812 if (flags
& BDRV_O_SNAPSHOT
) {
2813 snapshot_options
= qdict_new();
2814 bdrv_temp_snapshot_options(&snapshot_flags
, snapshot_options
,
2816 /* Let bdrv_backing_options() override "read-only" */
2817 qdict_del(options
, BDRV_OPT_READ_ONLY
);
2818 bdrv_backing_options(&flags
, options
, flags
, options
);
2821 bs
->open_flags
= flags
;
2822 bs
->options
= options
;
2823 options
= qdict_clone_shallow(options
);
2825 /* Find the right image format driver */
2826 /* See cautionary note on accessing @options above */
2827 drvname
= qdict_get_try_str(options
, "driver");
2829 drv
= bdrv_find_format(drvname
);
2831 error_setg(errp
, "Unknown driver: '%s'", drvname
);
2836 assert(drvname
|| !(flags
& BDRV_O_PROTOCOL
));
2838 /* See cautionary note on accessing @options above */
2839 backing
= qdict_get_try_str(options
, "backing");
2840 if (qobject_to(QNull
, qdict_get(options
, "backing")) != NULL
||
2841 (backing
&& *backing
== '\0'))
2844 warn_report("Use of \"backing\": \"\" is deprecated; "
2845 "use \"backing\": null instead");
2847 flags
|= BDRV_O_NO_BACKING
;
2848 qdict_del(options
, "backing");
2851 /* Open image file without format layer. This BlockBackend is only used for
2852 * probing, the block drivers will do their own bdrv_open_child() for the
2853 * same BDS, which is why we put the node name back into options. */
2854 if ((flags
& BDRV_O_PROTOCOL
) == 0) {
2855 BlockDriverState
*file_bs
;
2857 file_bs
= bdrv_open_child_bs(filename
, options
, "file", bs
,
2858 &child_file
, true, &local_err
);
2862 if (file_bs
!= NULL
) {
2863 /* Not requesting BLK_PERM_CONSISTENT_READ because we're only
2864 * looking at the header to guess the image format. This works even
2865 * in cases where a guest would not see a consistent state. */
2866 file
= blk_new(0, BLK_PERM_ALL
);
2867 blk_insert_bs(file
, file_bs
, &local_err
);
2868 bdrv_unref(file_bs
);
2873 qdict_put_str(options
, "file", bdrv_get_node_name(file_bs
));
2877 /* Image format probing */
2880 ret
= find_image_format(file
, filename
, &drv
, &local_err
);
2885 * This option update would logically belong in bdrv_fill_options(),
2886 * but we first need to open bs->file for the probing to work, while
2887 * opening bs->file already requires the (mostly) final set of options
2888 * so that cache mode etc. can be inherited.
2890 * Adding the driver later is somewhat ugly, but it's not an option
2891 * that would ever be inherited, so it's correct. We just need to make
2892 * sure to update both bs->options (which has the full effective
2893 * options for bs) and options (which has file.* already removed).
2895 qdict_put_str(bs
->options
, "driver", drv
->format_name
);
2896 qdict_put_str(options
, "driver", drv
->format_name
);
2898 error_setg(errp
, "Must specify either driver or file");
2902 /* BDRV_O_PROTOCOL must be set iff a protocol BDS is about to be created */
2903 assert(!!(flags
& BDRV_O_PROTOCOL
) == !!drv
->bdrv_file_open
);
2904 /* file must be NULL if a protocol BDS is about to be created
2905 * (the inverse results in an error message from bdrv_open_common()) */
2906 assert(!(flags
& BDRV_O_PROTOCOL
) || !file
);
2908 /* Open the image */
2909 ret
= bdrv_open_common(bs
, file
, options
, &local_err
);
2919 /* If there is a backing file, use it */
2920 if ((flags
& BDRV_O_NO_BACKING
) == 0) {
2921 ret
= bdrv_open_backing_file(bs
, options
, "backing", &local_err
);
2923 goto close_and_fail
;
2927 /* Remove all children options and references
2928 * from bs->options and bs->explicit_options */
2929 QLIST_FOREACH(child
, &bs
->children
, next
) {
2930 char *child_key_dot
;
2931 child_key_dot
= g_strdup_printf("%s.", child
->name
);
2932 qdict_extract_subqdict(bs
->explicit_options
, NULL
, child_key_dot
);
2933 qdict_extract_subqdict(bs
->options
, NULL
, child_key_dot
);
2934 qdict_del(bs
->explicit_options
, child
->name
);
2935 qdict_del(bs
->options
, child
->name
);
2936 g_free(child_key_dot
);
2939 /* Check if any unknown options were used */
2940 if (qdict_size(options
) != 0) {
2941 const QDictEntry
*entry
= qdict_first(options
);
2942 if (flags
& BDRV_O_PROTOCOL
) {
2943 error_setg(errp
, "Block protocol '%s' doesn't support the option "
2944 "'%s'", drv
->format_name
, entry
->key
);
2947 "Block format '%s' does not support the option '%s'",
2948 drv
->format_name
, entry
->key
);
2951 goto close_and_fail
;
2954 bdrv_parent_cb_change_media(bs
, true);
2956 qobject_unref(options
);
2959 /* For snapshot=on, create a temporary qcow2 overlay. bs points to the
2960 * temporary snapshot afterwards. */
2961 if (snapshot_flags
) {
2962 BlockDriverState
*snapshot_bs
;
2963 snapshot_bs
= bdrv_append_temp_snapshot(bs
, snapshot_flags
,
2964 snapshot_options
, &local_err
);
2965 snapshot_options
= NULL
;
2967 goto close_and_fail
;
2969 /* We are not going to return bs but the overlay on top of it
2970 * (snapshot_bs); thus, we have to drop the strong reference to bs
2971 * (which we obtained by calling bdrv_new()). bs will not be deleted,
2972 * though, because the overlay still has a reference to it. */
2981 qobject_unref(snapshot_options
);
2982 qobject_unref(bs
->explicit_options
);
2983 qobject_unref(bs
->options
);
2984 qobject_unref(options
);
2986 bs
->explicit_options
= NULL
;
2988 error_propagate(errp
, local_err
);
2993 qobject_unref(snapshot_options
);
2994 qobject_unref(options
);
2995 error_propagate(errp
, local_err
);
2999 BlockDriverState
*bdrv_open(const char *filename
, const char *reference
,
3000 QDict
*options
, int flags
, Error
**errp
)
3002 return bdrv_open_inherit(filename
, reference
, options
, flags
, NULL
,
3006 /* Return true if the NULL-terminated @list contains @str */
3007 static bool is_str_in_list(const char *str
, const char *const *list
)
3011 for (i
= 0; list
[i
] != NULL
; i
++) {
3012 if (!strcmp(str
, list
[i
])) {
3021 * Check that every option set in @bs->options is also set in
3024 * Options listed in the common_options list and in
3025 * @bs->drv->mutable_opts are skipped.
3027 * Return 0 on success, otherwise return -EINVAL and set @errp.
3029 static int bdrv_reset_options_allowed(BlockDriverState
*bs
,
3030 const QDict
*new_opts
, Error
**errp
)
3032 const QDictEntry
*e
;
3033 /* These options are common to all block drivers and are handled
3034 * in bdrv_reopen_prepare() so they can be left out of @new_opts */
3035 const char *const common_options
[] = {
3036 "node-name", "discard", "cache.direct", "cache.no-flush",
3037 "read-only", "auto-read-only", "detect-zeroes", NULL
3040 for (e
= qdict_first(bs
->options
); e
; e
= qdict_next(bs
->options
, e
)) {
3041 if (!qdict_haskey(new_opts
, e
->key
) &&
3042 !is_str_in_list(e
->key
, common_options
) &&
3043 !is_str_in_list(e
->key
, bs
->drv
->mutable_opts
)) {
3044 error_setg(errp
, "Option '%s' cannot be reset "
3045 "to its default value", e
->key
);
3054 * Returns true if @child can be reached recursively from @bs
3056 static bool bdrv_recurse_has_child(BlockDriverState
*bs
,
3057 BlockDriverState
*child
)
3065 QLIST_FOREACH(c
, &bs
->children
, next
) {
3066 if (bdrv_recurse_has_child(c
->bs
, child
)) {
3075 * Adds a BlockDriverState to a simple queue for an atomic, transactional
3076 * reopen of multiple devices.
3078 * bs_queue can either be an existing BlockReopenQueue that has had QSIMPLE_INIT
3079 * already performed, or alternatively may be NULL a new BlockReopenQueue will
3080 * be created and initialized. This newly created BlockReopenQueue should be
3081 * passed back in for subsequent calls that are intended to be of the same
3084 * bs is the BlockDriverState to add to the reopen queue.
3086 * options contains the changed options for the associated bs
3087 * (the BlockReopenQueue takes ownership)
3089 * flags contains the open flags for the associated bs
3091 * returns a pointer to bs_queue, which is either the newly allocated
3092 * bs_queue, or the existing bs_queue being used.
3094 * bs must be drained between bdrv_reopen_queue() and bdrv_reopen_multiple().
3096 static BlockReopenQueue
*bdrv_reopen_queue_child(BlockReopenQueue
*bs_queue
,
3097 BlockDriverState
*bs
,
3099 const BdrvChildRole
*role
,
3100 QDict
*parent_options
,
3106 BlockReopenQueueEntry
*bs_entry
;
3108 QDict
*old_options
, *explicit_options
, *options_copy
;
3112 /* Make sure that the caller remembered to use a drained section. This is
3113 * important to avoid graph changes between the recursive queuing here and
3114 * bdrv_reopen_multiple(). */
3115 assert(bs
->quiesce_counter
> 0);
3117 if (bs_queue
== NULL
) {
3118 bs_queue
= g_new0(BlockReopenQueue
, 1);
3119 QSIMPLEQ_INIT(bs_queue
);
3123 options
= qdict_new();
3126 /* Check if this BlockDriverState is already in the queue */
3127 QSIMPLEQ_FOREACH(bs_entry
, bs_queue
, entry
) {
3128 if (bs
== bs_entry
->state
.bs
) {
3134 * Precedence of options:
3135 * 1. Explicitly passed in options (highest)
3136 * 2. Retained from explicitly set options of bs
3137 * 3. Inherited from parent node
3138 * 4. Retained from effective options of bs
3141 /* Old explicitly set values (don't overwrite by inherited value) */
3142 if (bs_entry
|| keep_old_opts
) {
3143 old_options
= qdict_clone_shallow(bs_entry
?
3144 bs_entry
->state
.explicit_options
:
3145 bs
->explicit_options
);
3146 bdrv_join_options(bs
, options
, old_options
);
3147 qobject_unref(old_options
);
3150 explicit_options
= qdict_clone_shallow(options
);
3152 /* Inherit from parent node */
3153 if (parent_options
) {
3155 role
->inherit_options(&flags
, options
, parent_flags
, parent_options
);
3157 flags
= bdrv_get_flags(bs
);
3160 if (keep_old_opts
) {
3161 /* Old values are used for options that aren't set yet */
3162 old_options
= qdict_clone_shallow(bs
->options
);
3163 bdrv_join_options(bs
, options
, old_options
);
3164 qobject_unref(old_options
);
3167 /* We have the final set of options so let's update the flags */
3168 options_copy
= qdict_clone_shallow(options
);
3169 opts
= qemu_opts_create(&bdrv_runtime_opts
, NULL
, 0, &error_abort
);
3170 qemu_opts_absorb_qdict(opts
, options_copy
, NULL
);
3171 update_flags_from_options(&flags
, opts
);
3172 qemu_opts_del(opts
);
3173 qobject_unref(options_copy
);
3175 /* bdrv_open_inherit() sets and clears some additional flags internally */
3176 flags
&= ~BDRV_O_PROTOCOL
;
3177 if (flags
& BDRV_O_RDWR
) {
3178 flags
|= BDRV_O_ALLOW_RDWR
;
3182 bs_entry
= g_new0(BlockReopenQueueEntry
, 1);
3183 QSIMPLEQ_INSERT_TAIL(bs_queue
, bs_entry
, entry
);
3185 qobject_unref(bs_entry
->state
.options
);
3186 qobject_unref(bs_entry
->state
.explicit_options
);
3189 bs_entry
->state
.bs
= bs
;
3190 bs_entry
->state
.options
= options
;
3191 bs_entry
->state
.explicit_options
= explicit_options
;
3192 bs_entry
->state
.flags
= flags
;
3194 /* This needs to be overwritten in bdrv_reopen_prepare() */
3195 bs_entry
->state
.perm
= UINT64_MAX
;
3196 bs_entry
->state
.shared_perm
= 0;
3199 * If keep_old_opts is false then it means that unspecified
3200 * options must be reset to their original value. We don't allow
3201 * resetting 'backing' but we need to know if the option is
3202 * missing in order to decide if we have to return an error.
3204 if (!keep_old_opts
) {
3205 bs_entry
->state
.backing_missing
=
3206 !qdict_haskey(options
, "backing") &&
3207 !qdict_haskey(options
, "backing.driver");
3210 QLIST_FOREACH(child
, &bs
->children
, next
) {
3211 QDict
*new_child_options
= NULL
;
3212 bool child_keep_old
= keep_old_opts
;
3214 /* reopen can only change the options of block devices that were
3215 * implicitly created and inherited options. For other (referenced)
3216 * block devices, a syntax like "backing.foo" results in an error. */
3217 if (child
->bs
->inherits_from
!= bs
) {
3221 /* Check if the options contain a child reference */
3222 if (qdict_haskey(options
, child
->name
)) {
3223 const char *childref
= qdict_get_try_str(options
, child
->name
);
3225 * The current child must not be reopened if the child
3226 * reference is null or points to a different node.
3228 if (g_strcmp0(childref
, child
->bs
->node_name
)) {
3232 * If the child reference points to the current child then
3233 * reopen it with its existing set of options (note that
3234 * it can still inherit new options from the parent).
3236 child_keep_old
= true;
3238 /* Extract child options ("child-name.*") */
3239 char *child_key_dot
= g_strdup_printf("%s.", child
->name
);
3240 qdict_extract_subqdict(explicit_options
, NULL
, child_key_dot
);
3241 qdict_extract_subqdict(options
, &new_child_options
, child_key_dot
);
3242 g_free(child_key_dot
);
3245 bdrv_reopen_queue_child(bs_queue
, child
->bs
, new_child_options
,
3246 child
->role
, options
, flags
, child_keep_old
);
3252 BlockReopenQueue
*bdrv_reopen_queue(BlockReopenQueue
*bs_queue
,
3253 BlockDriverState
*bs
,
3254 QDict
*options
, bool keep_old_opts
)
3256 return bdrv_reopen_queue_child(bs_queue
, bs
, options
, NULL
, NULL
, 0,
3261 * Reopen multiple BlockDriverStates atomically & transactionally.
3263 * The queue passed in (bs_queue) must have been built up previous
3264 * via bdrv_reopen_queue().
3266 * Reopens all BDS specified in the queue, with the appropriate
3267 * flags. All devices are prepared for reopen, and failure of any
3268 * device will cause all device changes to be abandoned, and intermediate
3271 * If all devices prepare successfully, then the changes are committed
3274 * All affected nodes must be drained between bdrv_reopen_queue() and
3275 * bdrv_reopen_multiple().
3277 int bdrv_reopen_multiple(BlockReopenQueue
*bs_queue
, Error
**errp
)
3280 BlockReopenQueueEntry
*bs_entry
, *next
;
3282 assert(bs_queue
!= NULL
);
3284 QSIMPLEQ_FOREACH(bs_entry
, bs_queue
, entry
) {
3285 assert(bs_entry
->state
.bs
->quiesce_counter
> 0);
3286 if (bdrv_reopen_prepare(&bs_entry
->state
, bs_queue
, errp
)) {
3289 bs_entry
->prepared
= true;
3292 QSIMPLEQ_FOREACH(bs_entry
, bs_queue
, entry
) {
3293 BDRVReopenState
*state
= &bs_entry
->state
;
3294 ret
= bdrv_check_perm(state
->bs
, bs_queue
, state
->perm
,
3295 state
->shared_perm
, NULL
, errp
);
3299 /* Check if new_backing_bs would accept the new permissions */
3300 if (state
->replace_backing_bs
&& state
->new_backing_bs
) {
3301 uint64_t nperm
, nshared
;
3302 bdrv_child_perm(state
->bs
, state
->new_backing_bs
,
3303 NULL
, &child_backing
, bs_queue
,
3304 state
->perm
, state
->shared_perm
,
3306 ret
= bdrv_check_update_perm(state
->new_backing_bs
, NULL
,
3307 nperm
, nshared
, NULL
, errp
);
3312 bs_entry
->perms_checked
= true;
3315 /* If we reach this point, we have success and just need to apply the
3318 QSIMPLEQ_FOREACH(bs_entry
, bs_queue
, entry
) {
3319 bdrv_reopen_commit(&bs_entry
->state
);
3324 QSIMPLEQ_FOREACH_SAFE(bs_entry
, bs_queue
, entry
, next
) {
3325 BDRVReopenState
*state
= &bs_entry
->state
;
3327 if (!bs_entry
->perms_checked
) {
3332 bdrv_set_perm(state
->bs
, state
->perm
, state
->shared_perm
);
3334 bdrv_abort_perm_update(state
->bs
);
3335 if (state
->replace_backing_bs
&& state
->new_backing_bs
) {
3336 bdrv_abort_perm_update(state
->new_backing_bs
);
3341 QSIMPLEQ_FOREACH_SAFE(bs_entry
, bs_queue
, entry
, next
) {
3343 if (bs_entry
->prepared
) {
3344 bdrv_reopen_abort(&bs_entry
->state
);
3346 qobject_unref(bs_entry
->state
.explicit_options
);
3347 qobject_unref(bs_entry
->state
.options
);
3349 if (bs_entry
->state
.new_backing_bs
) {
3350 bdrv_unref(bs_entry
->state
.new_backing_bs
);
3359 int bdrv_reopen_set_read_only(BlockDriverState
*bs
, bool read_only
,
3363 BlockReopenQueue
*queue
;
3364 QDict
*opts
= qdict_new();
3366 qdict_put_bool(opts
, BDRV_OPT_READ_ONLY
, read_only
);
3368 bdrv_subtree_drained_begin(bs
);
3369 queue
= bdrv_reopen_queue(NULL
, bs
, opts
, true);
3370 ret
= bdrv_reopen_multiple(queue
, errp
);
3371 bdrv_subtree_drained_end(bs
);
3376 static BlockReopenQueueEntry
*find_parent_in_reopen_queue(BlockReopenQueue
*q
,
3379 BlockReopenQueueEntry
*entry
;
3381 QSIMPLEQ_FOREACH(entry
, q
, entry
) {
3382 BlockDriverState
*bs
= entry
->state
.bs
;
3385 QLIST_FOREACH(child
, &bs
->children
, next
) {
3395 static void bdrv_reopen_perm(BlockReopenQueue
*q
, BlockDriverState
*bs
,
3396 uint64_t *perm
, uint64_t *shared
)
3399 BlockReopenQueueEntry
*parent
;
3400 uint64_t cumulative_perms
= 0;
3401 uint64_t cumulative_shared_perms
= BLK_PERM_ALL
;
3403 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
3404 parent
= find_parent_in_reopen_queue(q
, c
);
3406 cumulative_perms
|= c
->perm
;
3407 cumulative_shared_perms
&= c
->shared_perm
;
3409 uint64_t nperm
, nshared
;
3411 bdrv_child_perm(parent
->state
.bs
, bs
, c
, c
->role
, q
,
3412 parent
->state
.perm
, parent
->state
.shared_perm
,
3415 cumulative_perms
|= nperm
;
3416 cumulative_shared_perms
&= nshared
;
3419 *perm
= cumulative_perms
;
3420 *shared
= cumulative_shared_perms
;
3424 * Take a BDRVReopenState and check if the value of 'backing' in the
3425 * reopen_state->options QDict is valid or not.
3427 * If 'backing' is missing from the QDict then return 0.
3429 * If 'backing' contains the node name of the backing file of
3430 * reopen_state->bs then return 0.
3432 * If 'backing' contains a different node name (or is null) then check
3433 * whether the current backing file can be replaced with the new one.
3434 * If that's the case then reopen_state->replace_backing_bs is set to
3435 * true and reopen_state->new_backing_bs contains a pointer to the new
3436 * backing BlockDriverState (or NULL).
3438 * Return 0 on success, otherwise return < 0 and set @errp.
3440 static int bdrv_reopen_parse_backing(BDRVReopenState
*reopen_state
,
3443 BlockDriverState
*bs
= reopen_state
->bs
;
3444 BlockDriverState
*overlay_bs
, *new_backing_bs
;
3448 value
= qdict_get(reopen_state
->options
, "backing");
3449 if (value
== NULL
) {
3453 switch (qobject_type(value
)) {
3455 new_backing_bs
= NULL
;
3458 str
= qobject_get_try_str(value
);
3459 new_backing_bs
= bdrv_lookup_bs(NULL
, str
, errp
);
3460 if (new_backing_bs
== NULL
) {
3462 } else if (bdrv_recurse_has_child(new_backing_bs
, bs
)) {
3463 error_setg(errp
, "Making '%s' a backing file of '%s' "
3464 "would create a cycle", str
, bs
->node_name
);
3469 /* 'backing' does not allow any other data type */
3470 g_assert_not_reached();
3474 * TODO: before removing the x- prefix from x-blockdev-reopen we
3475 * should move the new backing file into the right AioContext
3476 * instead of returning an error.
3478 if (new_backing_bs
) {
3479 if (bdrv_get_aio_context(new_backing_bs
) != bdrv_get_aio_context(bs
)) {
3480 error_setg(errp
, "Cannot use a new backing file "
3481 "with a different AioContext");
3487 * Find the "actual" backing file by skipping all links that point
3488 * to an implicit node, if any (e.g. a commit filter node).
3491 while (backing_bs(overlay_bs
) && backing_bs(overlay_bs
)->implicit
) {
3492 overlay_bs
= backing_bs(overlay_bs
);
3495 /* If we want to replace the backing file we need some extra checks */
3496 if (new_backing_bs
!= backing_bs(overlay_bs
)) {
3497 /* Check for implicit nodes between bs and its backing file */
3498 if (bs
!= overlay_bs
) {
3499 error_setg(errp
, "Cannot change backing link if '%s' has "
3500 "an implicit backing file", bs
->node_name
);
3503 /* Check if the backing link that we want to replace is frozen */
3504 if (bdrv_is_backing_chain_frozen(overlay_bs
, backing_bs(overlay_bs
),
3508 reopen_state
->replace_backing_bs
= true;
3509 if (new_backing_bs
) {
3510 bdrv_ref(new_backing_bs
);
3511 reopen_state
->new_backing_bs
= new_backing_bs
;
3519 * Prepares a BlockDriverState for reopen. All changes are staged in the
3520 * 'opaque' field of the BDRVReopenState, which is used and allocated by
3521 * the block driver layer .bdrv_reopen_prepare()
3523 * bs is the BlockDriverState to reopen
3524 * flags are the new open flags
3525 * queue is the reopen queue
3527 * Returns 0 on success, non-zero on error. On error errp will be set
3530 * On failure, bdrv_reopen_abort() will be called to clean up any data.
3531 * It is the responsibility of the caller to then call the abort() or
3532 * commit() for any other BDS that have been left in a prepare() state
3535 int bdrv_reopen_prepare(BDRVReopenState
*reopen_state
, BlockReopenQueue
*queue
,
3540 Error
*local_err
= NULL
;
3543 QDict
*orig_reopen_opts
;
3544 char *discard
= NULL
;
3546 bool drv_prepared
= false;
3548 assert(reopen_state
!= NULL
);
3549 assert(reopen_state
->bs
->drv
!= NULL
);
3550 drv
= reopen_state
->bs
->drv
;
3552 /* This function and each driver's bdrv_reopen_prepare() remove
3553 * entries from reopen_state->options as they are processed, so
3554 * we need to make a copy of the original QDict. */
3555 orig_reopen_opts
= qdict_clone_shallow(reopen_state
->options
);
3557 /* Process generic block layer options */
3558 opts
= qemu_opts_create(&bdrv_runtime_opts
, NULL
, 0, &error_abort
);
3559 qemu_opts_absorb_qdict(opts
, reopen_state
->options
, &local_err
);
3561 error_propagate(errp
, local_err
);
3566 /* This was already called in bdrv_reopen_queue_child() so the flags
3567 * are up-to-date. This time we simply want to remove the options from
3568 * QemuOpts in order to indicate that they have been processed. */
3569 old_flags
= reopen_state
->flags
;
3570 update_flags_from_options(&reopen_state
->flags
, opts
);
3571 assert(old_flags
== reopen_state
->flags
);
3573 discard
= qemu_opt_get_del(opts
, BDRV_OPT_DISCARD
);
3574 if (discard
!= NULL
) {
3575 if (bdrv_parse_discard_flags(discard
, &reopen_state
->flags
) != 0) {
3576 error_setg(errp
, "Invalid discard option");
3582 reopen_state
->detect_zeroes
=
3583 bdrv_parse_detect_zeroes(opts
, reopen_state
->flags
, &local_err
);
3585 error_propagate(errp
, local_err
);
3590 /* All other options (including node-name and driver) must be unchanged.
3591 * Put them back into the QDict, so that they are checked at the end
3592 * of this function. */
3593 qemu_opts_to_qdict(opts
, reopen_state
->options
);
3595 /* If we are to stay read-only, do not allow permission change
3596 * to r/w. Attempting to set to r/w may fail if either BDRV_O_ALLOW_RDWR is
3597 * not set, or if the BDS still has copy_on_read enabled */
3598 read_only
= !(reopen_state
->flags
& BDRV_O_RDWR
);
3599 ret
= bdrv_can_set_read_only(reopen_state
->bs
, read_only
, true, &local_err
);
3601 error_propagate(errp
, local_err
);
3605 /* Calculate required permissions after reopening */
3606 bdrv_reopen_perm(queue
, reopen_state
->bs
,
3607 &reopen_state
->perm
, &reopen_state
->shared_perm
);
3609 ret
= bdrv_flush(reopen_state
->bs
);
3611 error_setg_errno(errp
, -ret
, "Error flushing drive");
3615 if (drv
->bdrv_reopen_prepare
) {
3617 * If a driver-specific option is missing, it means that we
3618 * should reset it to its default value.
3619 * But not all options allow that, so we need to check it first.
3621 ret
= bdrv_reset_options_allowed(reopen_state
->bs
,
3622 reopen_state
->options
, errp
);
3627 ret
= drv
->bdrv_reopen_prepare(reopen_state
, queue
, &local_err
);
3629 if (local_err
!= NULL
) {
3630 error_propagate(errp
, local_err
);
3632 bdrv_refresh_filename(reopen_state
->bs
);
3633 error_setg(errp
, "failed while preparing to reopen image '%s'",
3634 reopen_state
->bs
->filename
);
3639 /* It is currently mandatory to have a bdrv_reopen_prepare()
3640 * handler for each supported drv. */
3641 error_setg(errp
, "Block format '%s' used by node '%s' "
3642 "does not support reopening files", drv
->format_name
,
3643 bdrv_get_device_or_node_name(reopen_state
->bs
));
3648 drv_prepared
= true;
3651 * We must provide the 'backing' option if the BDS has a backing
3652 * file or if the image file has a backing file name as part of
3653 * its metadata. Otherwise the 'backing' option can be omitted.
3655 if (drv
->supports_backing
&& reopen_state
->backing_missing
&&
3656 (backing_bs(reopen_state
->bs
) || reopen_state
->bs
->backing_file
[0])) {
3657 error_setg(errp
, "backing is missing for '%s'",
3658 reopen_state
->bs
->node_name
);
3664 * Allow changing the 'backing' option. The new value can be
3665 * either a reference to an existing node (using its node name)
3666 * or NULL to simply detach the current backing file.
3668 ret
= bdrv_reopen_parse_backing(reopen_state
, errp
);
3672 qdict_del(reopen_state
->options
, "backing");
3674 /* Options that are not handled are only okay if they are unchanged
3675 * compared to the old state. It is expected that some options are only
3676 * used for the initial open, but not reopen (e.g. filename) */
3677 if (qdict_size(reopen_state
->options
)) {
3678 const QDictEntry
*entry
= qdict_first(reopen_state
->options
);
3681 QObject
*new = entry
->value
;
3682 QObject
*old
= qdict_get(reopen_state
->bs
->options
, entry
->key
);
3684 /* Allow child references (child_name=node_name) as long as they
3685 * point to the current child (i.e. everything stays the same). */
3686 if (qobject_type(new) == QTYPE_QSTRING
) {
3688 QLIST_FOREACH(child
, &reopen_state
->bs
->children
, next
) {
3689 if (!strcmp(child
->name
, entry
->key
)) {
3695 const char *str
= qobject_get_try_str(new);
3696 if (!strcmp(child
->bs
->node_name
, str
)) {
3697 continue; /* Found child with this name, skip option */
3703 * TODO: When using -drive to specify blockdev options, all values
3704 * will be strings; however, when using -blockdev, blockdev-add or
3705 * filenames using the json:{} pseudo-protocol, they will be
3707 * In contrast, reopening options are (currently) always strings
3708 * (because you can only specify them through qemu-io; all other
3709 * callers do not specify any options).
3710 * Therefore, when using anything other than -drive to create a BDS,
3711 * this cannot detect non-string options as unchanged, because
3712 * qobject_is_equal() always returns false for objects of different
3713 * type. In the future, this should be remedied by correctly typing
3714 * all options. For now, this is not too big of an issue because
3715 * the user can simply omit options which cannot be changed anyway,
3716 * so they will stay unchanged.
3718 if (!qobject_is_equal(new, old
)) {
3719 error_setg(errp
, "Cannot change the option '%s'", entry
->key
);
3723 } while ((entry
= qdict_next(reopen_state
->options
, entry
)));
3728 /* Restore the original reopen_state->options QDict */
3729 qobject_unref(reopen_state
->options
);
3730 reopen_state
->options
= qobject_ref(orig_reopen_opts
);
3733 if (ret
< 0 && drv_prepared
) {
3734 /* drv->bdrv_reopen_prepare() has succeeded, so we need to
3735 * call drv->bdrv_reopen_abort() before signaling an error
3736 * (bdrv_reopen_multiple() will not call bdrv_reopen_abort()
3737 * when the respective bdrv_reopen_prepare() has failed) */
3738 if (drv
->bdrv_reopen_abort
) {
3739 drv
->bdrv_reopen_abort(reopen_state
);
3742 qemu_opts_del(opts
);
3743 qobject_unref(orig_reopen_opts
);
3749 * Takes the staged changes for the reopen from bdrv_reopen_prepare(), and
3750 * makes them final by swapping the staging BlockDriverState contents into
3751 * the active BlockDriverState contents.
3753 void bdrv_reopen_commit(BDRVReopenState
*reopen_state
)
3756 BlockDriverState
*bs
;
3758 bool old_can_write
, new_can_write
;
3760 assert(reopen_state
!= NULL
);
3761 bs
= reopen_state
->bs
;
3763 assert(drv
!= NULL
);
3766 !bdrv_is_read_only(bs
) && !(bdrv_get_flags(bs
) & BDRV_O_INACTIVE
);
3768 /* If there are any driver level actions to take */
3769 if (drv
->bdrv_reopen_commit
) {
3770 drv
->bdrv_reopen_commit(reopen_state
);
3773 /* set BDS specific flags now */
3774 qobject_unref(bs
->explicit_options
);
3775 qobject_unref(bs
->options
);
3777 bs
->explicit_options
= reopen_state
->explicit_options
;
3778 bs
->options
= reopen_state
->options
;
3779 bs
->open_flags
= reopen_state
->flags
;
3780 bs
->read_only
= !(reopen_state
->flags
& BDRV_O_RDWR
);
3781 bs
->detect_zeroes
= reopen_state
->detect_zeroes
;
3783 if (reopen_state
->replace_backing_bs
) {
3784 qdict_del(bs
->explicit_options
, "backing");
3785 qdict_del(bs
->options
, "backing");
3788 /* Remove child references from bs->options and bs->explicit_options.
3789 * Child options were already removed in bdrv_reopen_queue_child() */
3790 QLIST_FOREACH(child
, &bs
->children
, next
) {
3791 qdict_del(bs
->explicit_options
, child
->name
);
3792 qdict_del(bs
->options
, child
->name
);
3796 * Change the backing file if a new one was specified. We do this
3797 * after updating bs->options, so bdrv_refresh_filename() (called
3798 * from bdrv_set_backing_hd()) has the new values.
3800 if (reopen_state
->replace_backing_bs
) {
3801 BlockDriverState
*old_backing_bs
= backing_bs(bs
);
3802 assert(!old_backing_bs
|| !old_backing_bs
->implicit
);
3803 /* Abort the permission update on the backing bs we're detaching */
3804 if (old_backing_bs
) {
3805 bdrv_abort_perm_update(old_backing_bs
);
3807 bdrv_set_backing_hd(bs
, reopen_state
->new_backing_bs
, &error_abort
);
3810 bdrv_refresh_limits(bs
, NULL
);
3813 !bdrv_is_read_only(bs
) && !(bdrv_get_flags(bs
) & BDRV_O_INACTIVE
);
3814 if (!old_can_write
&& new_can_write
&& drv
->bdrv_reopen_bitmaps_rw
) {
3815 Error
*local_err
= NULL
;
3816 if (drv
->bdrv_reopen_bitmaps_rw(bs
, &local_err
) < 0) {
3817 /* This is not fatal, bitmaps just left read-only, so all following
3818 * writes will fail. User can remove read-only bitmaps to unblock
3821 error_reportf_err(local_err
,
3822 "%s: Failed to make dirty bitmaps writable: ",
3823 bdrv_get_node_name(bs
));
3829 * Abort the reopen, and delete and free the staged changes in
3832 void bdrv_reopen_abort(BDRVReopenState
*reopen_state
)
3836 assert(reopen_state
!= NULL
);
3837 drv
= reopen_state
->bs
->drv
;
3838 assert(drv
!= NULL
);
3840 if (drv
->bdrv_reopen_abort
) {
3841 drv
->bdrv_reopen_abort(reopen_state
);
3846 static void bdrv_close(BlockDriverState
*bs
)
3848 BdrvAioNotifier
*ban
, *ban_next
;
3849 BdrvChild
*child
, *next
;
3852 assert(!bs
->refcnt
);
3854 bdrv_drained_begin(bs
); /* complete I/O */
3856 bdrv_drain(bs
); /* in case flush left pending I/O */
3859 if (bs
->drv
->bdrv_close
) {
3860 bs
->drv
->bdrv_close(bs
);
3865 bdrv_set_backing_hd(bs
, NULL
, &error_abort
);
3867 if (bs
->file
!= NULL
) {
3868 bdrv_unref_child(bs
, bs
->file
);
3872 QLIST_FOREACH_SAFE(child
, &bs
->children
, next
, next
) {
3873 /* TODO Remove bdrv_unref() from drivers' close function and use
3874 * bdrv_unref_child() here */
3875 if (child
->bs
->inherits_from
== bs
) {
3876 child
->bs
->inherits_from
= NULL
;
3878 bdrv_detach_child(child
);
3883 atomic_set(&bs
->copy_on_read
, 0);
3884 bs
->backing_file
[0] = '\0';
3885 bs
->backing_format
[0] = '\0';
3886 bs
->total_sectors
= 0;
3887 bs
->encrypted
= false;
3889 qobject_unref(bs
->options
);
3890 qobject_unref(bs
->explicit_options
);
3892 bs
->explicit_options
= NULL
;
3893 qobject_unref(bs
->full_open_options
);
3894 bs
->full_open_options
= NULL
;
3896 bdrv_release_named_dirty_bitmaps(bs
);
3897 assert(QLIST_EMPTY(&bs
->dirty_bitmaps
));
3899 QLIST_FOREACH_SAFE(ban
, &bs
->aio_notifiers
, list
, ban_next
) {
3902 QLIST_INIT(&bs
->aio_notifiers
);
3903 bdrv_drained_end(bs
);
3906 void bdrv_close_all(void)
3908 assert(job_next(NULL
) == NULL
);
3909 nbd_export_close_all();
3911 /* Drop references from requests still in flight, such as canceled block
3912 * jobs whose AIO context has not been polled yet */
3915 blk_remove_all_bs();
3916 blockdev_close_all_bdrv_states();
3918 assert(QTAILQ_EMPTY(&all_bdrv_states
));
3921 static bool should_update_child(BdrvChild
*c
, BlockDriverState
*to
)
3927 if (c
->role
->stay_at_node
) {
3931 /* If the child @c belongs to the BDS @to, replacing the current
3932 * c->bs by @to would mean to create a loop.
3934 * Such a case occurs when appending a BDS to a backing chain.
3935 * For instance, imagine the following chain:
3937 * guest device -> node A -> further backing chain...
3939 * Now we create a new BDS B which we want to put on top of this
3940 * chain, so we first attach A as its backing node:
3945 * guest device -> node A -> further backing chain...
3947 * Finally we want to replace A by B. When doing that, we want to
3948 * replace all pointers to A by pointers to B -- except for the
3949 * pointer from B because (1) that would create a loop, and (2)
3950 * that pointer should simply stay intact:
3952 * guest device -> node B
3955 * node A -> further backing chain...
3957 * In general, when replacing a node A (c->bs) by a node B (@to),
3958 * if A is a child of B, that means we cannot replace A by B there
3959 * because that would create a loop. Silently detaching A from B
3960 * is also not really an option. So overall just leaving A in
3961 * place there is the most sensible choice.
3963 * We would also create a loop in any cases where @c is only
3964 * indirectly referenced by @to. Prevent this by returning false
3965 * if @c is found (by breadth-first search) anywhere in the whole
3970 found
= g_hash_table_new(NULL
, NULL
);
3971 g_hash_table_add(found
, to
);
3972 queue
= g_queue_new();
3973 g_queue_push_tail(queue
, to
);
3975 while (!g_queue_is_empty(queue
)) {
3976 BlockDriverState
*v
= g_queue_pop_head(queue
);
3979 QLIST_FOREACH(c2
, &v
->children
, next
) {
3985 if (g_hash_table_contains(found
, c2
->bs
)) {
3989 g_queue_push_tail(queue
, c2
->bs
);
3990 g_hash_table_add(found
, c2
->bs
);
3994 g_queue_free(queue
);
3995 g_hash_table_destroy(found
);
4000 void bdrv_replace_node(BlockDriverState
*from
, BlockDriverState
*to
,
4003 BdrvChild
*c
, *next
;
4004 GSList
*list
= NULL
, *p
;
4005 uint64_t old_perm
, old_shared
;
4006 uint64_t perm
= 0, shared
= BLK_PERM_ALL
;
4009 assert(!atomic_read(&from
->in_flight
));
4010 assert(!atomic_read(&to
->in_flight
));
4012 /* Make sure that @from doesn't go away until we have successfully attached
4013 * all of its parents to @to. */
4016 /* Put all parents into @list and calculate their cumulative permissions */
4017 QLIST_FOREACH_SAFE(c
, &from
->parents
, next_parent
, next
) {
4018 assert(c
->bs
== from
);
4019 if (!should_update_child(c
, to
)) {
4023 error_setg(errp
, "Cannot change '%s' link to '%s'",
4024 c
->name
, from
->node_name
);
4027 list
= g_slist_prepend(list
, c
);
4029 shared
&= c
->shared_perm
;
4032 /* Check whether the required permissions can be granted on @to, ignoring
4033 * all BdrvChild in @list so that they can't block themselves. */
4034 ret
= bdrv_check_update_perm(to
, NULL
, perm
, shared
, list
, errp
);
4036 bdrv_abort_perm_update(to
);
4040 /* Now actually perform the change. We performed the permission check for
4041 * all elements of @list at once, so set the permissions all at once at the
4043 for (p
= list
; p
!= NULL
; p
= p
->next
) {
4047 bdrv_replace_child_noperm(c
, to
);
4051 bdrv_get_cumulative_perm(to
, &old_perm
, &old_shared
);
4052 bdrv_set_perm(to
, old_perm
| perm
, old_shared
| shared
);
4060 * Add new bs contents at the top of an image chain while the chain is
4061 * live, while keeping required fields on the top layer.
4063 * This will modify the BlockDriverState fields, and swap contents
4064 * between bs_new and bs_top. Both bs_new and bs_top are modified.
4066 * bs_new must not be attached to a BlockBackend.
4068 * This function does not create any image files.
4070 * bdrv_append() takes ownership of a bs_new reference and unrefs it because
4071 * that's what the callers commonly need. bs_new will be referenced by the old
4072 * parents of bs_top after bdrv_append() returns. If the caller needs to keep a
4073 * reference of its own, it must call bdrv_ref().
4075 void bdrv_append(BlockDriverState
*bs_new
, BlockDriverState
*bs_top
,
4078 Error
*local_err
= NULL
;
4080 bdrv_set_backing_hd(bs_new
, bs_top
, &local_err
);
4082 error_propagate(errp
, local_err
);
4086 bdrv_replace_node(bs_top
, bs_new
, &local_err
);
4088 error_propagate(errp
, local_err
);
4089 bdrv_set_backing_hd(bs_new
, NULL
, &error_abort
);
4093 /* bs_new is now referenced by its new parents, we don't need the
4094 * additional reference any more. */
4099 static void bdrv_delete(BlockDriverState
*bs
)
4102 assert(bdrv_op_blocker_is_empty(bs
));
4103 assert(!bs
->refcnt
);
4105 /* remove from list, if necessary */
4106 if (bs
->node_name
[0] != '\0') {
4107 QTAILQ_REMOVE(&graph_bdrv_states
, bs
, node_list
);
4109 QTAILQ_REMOVE(&all_bdrv_states
, bs
, bs_list
);
4117 * Run consistency checks on an image
4119 * Returns 0 if the check could be completed (it doesn't mean that the image is
4120 * free of errors) or -errno when an internal error occurred. The results of the
4121 * check are stored in res.
4123 static int coroutine_fn
bdrv_co_check(BlockDriverState
*bs
,
4124 BdrvCheckResult
*res
, BdrvCheckMode fix
)
4126 if (bs
->drv
== NULL
) {
4129 if (bs
->drv
->bdrv_co_check
== NULL
) {
4133 memset(res
, 0, sizeof(*res
));
4134 return bs
->drv
->bdrv_co_check(bs
, res
, fix
);
4137 typedef struct CheckCo
{
4138 BlockDriverState
*bs
;
4139 BdrvCheckResult
*res
;
4144 static void coroutine_fn
bdrv_check_co_entry(void *opaque
)
4146 CheckCo
*cco
= opaque
;
4147 cco
->ret
= bdrv_co_check(cco
->bs
, cco
->res
, cco
->fix
);
4151 int bdrv_check(BlockDriverState
*bs
,
4152 BdrvCheckResult
*res
, BdrvCheckMode fix
)
4158 .ret
= -EINPROGRESS
,
4162 if (qemu_in_coroutine()) {
4163 /* Fast-path if already in coroutine context */
4164 bdrv_check_co_entry(&cco
);
4166 co
= qemu_coroutine_create(bdrv_check_co_entry
, &cco
);
4167 bdrv_coroutine_enter(bs
, co
);
4168 BDRV_POLL_WHILE(bs
, cco
.ret
== -EINPROGRESS
);
4177 * -EINVAL - backing format specified, but no file
4178 * -ENOSPC - can't update the backing file because no space is left in the
4180 * -ENOTSUP - format driver doesn't support changing the backing file
4182 int bdrv_change_backing_file(BlockDriverState
*bs
,
4183 const char *backing_file
, const char *backing_fmt
)
4185 BlockDriver
*drv
= bs
->drv
;
4192 /* Backing file format doesn't make sense without a backing file */
4193 if (backing_fmt
&& !backing_file
) {
4197 if (drv
->bdrv_change_backing_file
!= NULL
) {
4198 ret
= drv
->bdrv_change_backing_file(bs
, backing_file
, backing_fmt
);
4204 pstrcpy(bs
->backing_file
, sizeof(bs
->backing_file
), backing_file
?: "");
4205 pstrcpy(bs
->backing_format
, sizeof(bs
->backing_format
), backing_fmt
?: "");
4206 pstrcpy(bs
->auto_backing_file
, sizeof(bs
->auto_backing_file
),
4207 backing_file
?: "");
4213 * Finds the image layer in the chain that has 'bs' as its backing file.
4215 * active is the current topmost image.
4217 * Returns NULL if bs is not found in active's image chain,
4218 * or if active == bs.
4220 * Returns the bottommost base image if bs == NULL.
4222 BlockDriverState
*bdrv_find_overlay(BlockDriverState
*active
,
4223 BlockDriverState
*bs
)
4225 while (active
&& bs
!= backing_bs(active
)) {
4226 active
= backing_bs(active
);
4232 /* Given a BDS, searches for the base layer. */
4233 BlockDriverState
*bdrv_find_base(BlockDriverState
*bs
)
4235 return bdrv_find_overlay(bs
, NULL
);
4239 * Return true if at least one of the backing links between @bs and
4240 * @base is frozen. @errp is set if that's the case.
4241 * @base must be reachable from @bs, or NULL.
4243 bool bdrv_is_backing_chain_frozen(BlockDriverState
*bs
, BlockDriverState
*base
,
4246 BlockDriverState
*i
;
4248 for (i
= bs
; i
!= base
; i
= backing_bs(i
)) {
4249 if (i
->backing
&& i
->backing
->frozen
) {
4250 error_setg(errp
, "Cannot change '%s' link from '%s' to '%s'",
4251 i
->backing
->name
, i
->node_name
,
4252 backing_bs(i
)->node_name
);
4261 * Freeze all backing links between @bs and @base.
4262 * If any of the links is already frozen the operation is aborted and
4263 * none of the links are modified.
4264 * @base must be reachable from @bs, or NULL.
4265 * Returns 0 on success. On failure returns < 0 and sets @errp.
4267 int bdrv_freeze_backing_chain(BlockDriverState
*bs
, BlockDriverState
*base
,
4270 BlockDriverState
*i
;
4272 if (bdrv_is_backing_chain_frozen(bs
, base
, errp
)) {
4276 for (i
= bs
; i
!= base
; i
= backing_bs(i
)) {
4278 i
->backing
->frozen
= true;
4286 * Unfreeze all backing links between @bs and @base. The caller must
4287 * ensure that all links are frozen before using this function.
4288 * @base must be reachable from @bs, or NULL.
4290 void bdrv_unfreeze_backing_chain(BlockDriverState
*bs
, BlockDriverState
*base
)
4292 BlockDriverState
*i
;
4294 for (i
= bs
; i
!= base
; i
= backing_bs(i
)) {
4296 assert(i
->backing
->frozen
);
4297 i
->backing
->frozen
= false;
4303 * Drops images above 'base' up to and including 'top', and sets the image
4304 * above 'top' to have base as its backing file.
4306 * Requires that the overlay to 'top' is opened r/w, so that the backing file
4307 * information in 'bs' can be properly updated.
4309 * E.g., this will convert the following chain:
4310 * bottom <- base <- intermediate <- top <- active
4314 * bottom <- base <- active
4316 * It is allowed for bottom==base, in which case it converts:
4318 * base <- intermediate <- top <- active
4324 * If backing_file_str is non-NULL, it will be used when modifying top's
4325 * overlay image metadata.
4328 * if active == top, that is considered an error
4331 int bdrv_drop_intermediate(BlockDriverState
*top
, BlockDriverState
*base
,
4332 const char *backing_file_str
)
4334 BlockDriverState
*explicit_top
= top
;
4335 bool update_inherits_from
;
4336 BdrvChild
*c
, *next
;
4337 Error
*local_err
= NULL
;
4342 if (!top
->drv
|| !base
->drv
) {
4346 /* Make sure that base is in the backing chain of top */
4347 if (!bdrv_chain_contains(top
, base
)) {
4351 /* This function changes all links that point to top and makes
4352 * them point to base. Check that none of them is frozen. */
4353 QLIST_FOREACH(c
, &top
->parents
, next_parent
) {
4359 /* If 'base' recursively inherits from 'top' then we should set
4360 * base->inherits_from to top->inherits_from after 'top' and all
4361 * other intermediate nodes have been dropped.
4362 * If 'top' is an implicit node (e.g. "commit_top") we should skip
4363 * it because no one inherits from it. We use explicit_top for that. */
4364 while (explicit_top
&& explicit_top
->implicit
) {
4365 explicit_top
= backing_bs(explicit_top
);
4367 update_inherits_from
= bdrv_inherits_from_recursive(base
, explicit_top
);
4369 /* success - we can delete the intermediate states, and link top->base */
4370 /* TODO Check graph modification op blockers (BLK_PERM_GRAPH_MOD) once
4371 * we've figured out how they should work. */
4372 if (!backing_file_str
) {
4373 bdrv_refresh_filename(base
);
4374 backing_file_str
= base
->filename
;
4377 QLIST_FOREACH_SAFE(c
, &top
->parents
, next_parent
, next
) {
4378 /* Check whether we are allowed to switch c from top to base */
4379 GSList
*ignore_children
= g_slist_prepend(NULL
, c
);
4380 ret
= bdrv_check_update_perm(base
, NULL
, c
->perm
, c
->shared_perm
,
4381 ignore_children
, &local_err
);
4382 g_slist_free(ignore_children
);
4384 error_report_err(local_err
);
4388 /* If so, update the backing file path in the image file */
4389 if (c
->role
->update_filename
) {
4390 ret
= c
->role
->update_filename(c
, base
, backing_file_str
,
4393 bdrv_abort_perm_update(base
);
4394 error_report_err(local_err
);
4399 /* Do the actual switch in the in-memory graph.
4400 * Completes bdrv_check_update_perm() transaction internally. */
4402 bdrv_replace_child(c
, base
);
4406 if (update_inherits_from
) {
4407 base
->inherits_from
= explicit_top
->inherits_from
;
4417 * Length of a allocated file in bytes. Sparse files are counted by actual
4418 * allocated space. Return < 0 if error or unknown.
4420 int64_t bdrv_get_allocated_file_size(BlockDriverState
*bs
)
4422 BlockDriver
*drv
= bs
->drv
;
4426 if (drv
->bdrv_get_allocated_file_size
) {
4427 return drv
->bdrv_get_allocated_file_size(bs
);
4430 return bdrv_get_allocated_file_size(bs
->file
->bs
);
4437 * @drv: Format driver
4438 * @opts: Creation options for new image
4439 * @in_bs: Existing image containing data for new image (may be NULL)
4440 * @errp: Error object
4441 * Returns: A #BlockMeasureInfo (free using qapi_free_BlockMeasureInfo())
4444 * Calculate file size required to create a new image.
4446 * If @in_bs is given then space for allocated clusters and zero clusters
4447 * from that image are included in the calculation. If @opts contains a
4448 * backing file that is shared by @in_bs then backing clusters may be omitted
4449 * from the calculation.
4451 * If @in_bs is NULL then the calculation includes no allocated clusters
4452 * unless a preallocation option is given in @opts.
4454 * Note that @in_bs may use a different BlockDriver from @drv.
4456 * If an error occurs the @errp pointer is set.
4458 BlockMeasureInfo
*bdrv_measure(BlockDriver
*drv
, QemuOpts
*opts
,
4459 BlockDriverState
*in_bs
, Error
**errp
)
4461 if (!drv
->bdrv_measure
) {
4462 error_setg(errp
, "Block driver '%s' does not support size measurement",
4467 return drv
->bdrv_measure(opts
, in_bs
, errp
);
4471 * Return number of sectors on success, -errno on error.
4473 int64_t bdrv_nb_sectors(BlockDriverState
*bs
)
4475 BlockDriver
*drv
= bs
->drv
;
4480 if (drv
->has_variable_length
) {
4481 int ret
= refresh_total_sectors(bs
, bs
->total_sectors
);
4486 return bs
->total_sectors
;
4490 * Return length in bytes on success, -errno on error.
4491 * The length is always a multiple of BDRV_SECTOR_SIZE.
4493 int64_t bdrv_getlength(BlockDriverState
*bs
)
4495 int64_t ret
= bdrv_nb_sectors(bs
);
4497 ret
= ret
> INT64_MAX
/ BDRV_SECTOR_SIZE
? -EFBIG
: ret
;
4498 return ret
< 0 ? ret
: ret
* BDRV_SECTOR_SIZE
;
4501 /* return 0 as number of sectors if no device present or error */
4502 void bdrv_get_geometry(BlockDriverState
*bs
, uint64_t *nb_sectors_ptr
)
4504 int64_t nb_sectors
= bdrv_nb_sectors(bs
);
4506 *nb_sectors_ptr
= nb_sectors
< 0 ? 0 : nb_sectors
;
4509 bool bdrv_is_sg(BlockDriverState
*bs
)
4514 bool bdrv_is_encrypted(BlockDriverState
*bs
)
4516 if (bs
->backing
&& bs
->backing
->bs
->encrypted
) {
4519 return bs
->encrypted
;
4522 const char *bdrv_get_format_name(BlockDriverState
*bs
)
4524 return bs
->drv
? bs
->drv
->format_name
: NULL
;
4527 static int qsort_strcmp(const void *a
, const void *b
)
4529 return strcmp(*(char *const *)a
, *(char *const *)b
);
4532 void bdrv_iterate_format(void (*it
)(void *opaque
, const char *name
),
4533 void *opaque
, bool read_only
)
4538 const char **formats
= NULL
;
4540 QLIST_FOREACH(drv
, &bdrv_drivers
, list
) {
4541 if (drv
->format_name
) {
4545 if (use_bdrv_whitelist
&& !bdrv_is_whitelisted(drv
, read_only
)) {
4549 while (formats
&& i
&& !found
) {
4550 found
= !strcmp(formats
[--i
], drv
->format_name
);
4554 formats
= g_renew(const char *, formats
, count
+ 1);
4555 formats
[count
++] = drv
->format_name
;
4560 for (i
= 0; i
< (int)ARRAY_SIZE(block_driver_modules
); i
++) {
4561 const char *format_name
= block_driver_modules
[i
].format_name
;
4567 if (use_bdrv_whitelist
&&
4568 !bdrv_format_is_whitelisted(format_name
, read_only
)) {
4572 while (formats
&& j
&& !found
) {
4573 found
= !strcmp(formats
[--j
], format_name
);
4577 formats
= g_renew(const char *, formats
, count
+ 1);
4578 formats
[count
++] = format_name
;
4583 qsort(formats
, count
, sizeof(formats
[0]), qsort_strcmp
);
4585 for (i
= 0; i
< count
; i
++) {
4586 it(opaque
, formats
[i
]);
4592 /* This function is to find a node in the bs graph */
4593 BlockDriverState
*bdrv_find_node(const char *node_name
)
4595 BlockDriverState
*bs
;
4599 QTAILQ_FOREACH(bs
, &graph_bdrv_states
, node_list
) {
4600 if (!strcmp(node_name
, bs
->node_name
)) {
4607 /* Put this QMP function here so it can access the static graph_bdrv_states. */
4608 BlockDeviceInfoList
*bdrv_named_nodes_list(Error
**errp
)
4610 BlockDeviceInfoList
*list
, *entry
;
4611 BlockDriverState
*bs
;
4614 QTAILQ_FOREACH(bs
, &graph_bdrv_states
, node_list
) {
4615 BlockDeviceInfo
*info
= bdrv_block_device_info(NULL
, bs
, errp
);
4617 qapi_free_BlockDeviceInfoList(list
);
4620 entry
= g_malloc0(sizeof(*entry
));
4621 entry
->value
= info
;
4629 #define QAPI_LIST_ADD(list, element) do { \
4630 typeof(list) _tmp = g_new(typeof(*(list)), 1); \
4631 _tmp->value = (element); \
4632 _tmp->next = (list); \
4636 typedef struct XDbgBlockGraphConstructor
{
4637 XDbgBlockGraph
*graph
;
4638 GHashTable
*graph_nodes
;
4639 } XDbgBlockGraphConstructor
;
4641 static XDbgBlockGraphConstructor
*xdbg_graph_new(void)
4643 XDbgBlockGraphConstructor
*gr
= g_new(XDbgBlockGraphConstructor
, 1);
4645 gr
->graph
= g_new0(XDbgBlockGraph
, 1);
4646 gr
->graph_nodes
= g_hash_table_new(NULL
, NULL
);
4651 static XDbgBlockGraph
*xdbg_graph_finalize(XDbgBlockGraphConstructor
*gr
)
4653 XDbgBlockGraph
*graph
= gr
->graph
;
4655 g_hash_table_destroy(gr
->graph_nodes
);
4661 static uintptr_t xdbg_graph_node_num(XDbgBlockGraphConstructor
*gr
, void *node
)
4663 uintptr_t ret
= (uintptr_t)g_hash_table_lookup(gr
->graph_nodes
, node
);
4670 * Start counting from 1, not 0, because 0 interferes with not-found (NULL)
4671 * answer of g_hash_table_lookup.
4673 ret
= g_hash_table_size(gr
->graph_nodes
) + 1;
4674 g_hash_table_insert(gr
->graph_nodes
, node
, (void *)ret
);
4679 static void xdbg_graph_add_node(XDbgBlockGraphConstructor
*gr
, void *node
,
4680 XDbgBlockGraphNodeType type
, const char *name
)
4682 XDbgBlockGraphNode
*n
;
4684 n
= g_new0(XDbgBlockGraphNode
, 1);
4686 n
->id
= xdbg_graph_node_num(gr
, node
);
4688 n
->name
= g_strdup(name
);
4690 QAPI_LIST_ADD(gr
->graph
->nodes
, n
);
4693 static void xdbg_graph_add_edge(XDbgBlockGraphConstructor
*gr
, void *parent
,
4694 const BdrvChild
*child
)
4698 BlockPermission num
;
4701 static const PermissionMap permissions
[] = {
4702 { BLK_PERM_CONSISTENT_READ
, BLOCK_PERMISSION_CONSISTENT_READ
},
4703 { BLK_PERM_WRITE
, BLOCK_PERMISSION_WRITE
},
4704 { BLK_PERM_WRITE_UNCHANGED
, BLOCK_PERMISSION_WRITE_UNCHANGED
},
4705 { BLK_PERM_RESIZE
, BLOCK_PERMISSION_RESIZE
},
4706 { BLK_PERM_GRAPH_MOD
, BLOCK_PERMISSION_GRAPH_MOD
},
4709 const PermissionMap
*p
;
4710 XDbgBlockGraphEdge
*edge
;
4712 QEMU_BUILD_BUG_ON(1UL << (ARRAY_SIZE(permissions
) - 1) != BLK_PERM_ALL
+ 1);
4714 edge
= g_new0(XDbgBlockGraphEdge
, 1);
4716 edge
->parent
= xdbg_graph_node_num(gr
, parent
);
4717 edge
->child
= xdbg_graph_node_num(gr
, child
->bs
);
4718 edge
->name
= g_strdup(child
->name
);
4720 for (p
= permissions
; p
->flag
; p
++) {
4721 if (p
->flag
& child
->perm
) {
4722 QAPI_LIST_ADD(edge
->perm
, p
->num
);
4724 if (p
->flag
& child
->shared_perm
) {
4725 QAPI_LIST_ADD(edge
->shared_perm
, p
->num
);
4729 QAPI_LIST_ADD(gr
->graph
->edges
, edge
);
4733 XDbgBlockGraph
*bdrv_get_xdbg_block_graph(Error
**errp
)
4737 BlockDriverState
*bs
;
4739 XDbgBlockGraphConstructor
*gr
= xdbg_graph_new();
4741 for (blk
= blk_all_next(NULL
); blk
; blk
= blk_all_next(blk
)) {
4742 char *allocated_name
= NULL
;
4743 const char *name
= blk_name(blk
);
4746 name
= allocated_name
= blk_get_attached_dev_id(blk
);
4748 xdbg_graph_add_node(gr
, blk
, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_BACKEND
,
4750 g_free(allocated_name
);
4751 if (blk_root(blk
)) {
4752 xdbg_graph_add_edge(gr
, blk
, blk_root(blk
));
4756 for (job
= block_job_next(NULL
); job
; job
= block_job_next(job
)) {
4759 xdbg_graph_add_node(gr
, job
, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_JOB
,
4761 for (el
= job
->nodes
; el
; el
= el
->next
) {
4762 xdbg_graph_add_edge(gr
, job
, (BdrvChild
*)el
->data
);
4766 QTAILQ_FOREACH(bs
, &graph_bdrv_states
, node_list
) {
4767 xdbg_graph_add_node(gr
, bs
, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_DRIVER
,
4769 QLIST_FOREACH(child
, &bs
->children
, next
) {
4770 xdbg_graph_add_edge(gr
, bs
, child
);
4774 return xdbg_graph_finalize(gr
);
4777 BlockDriverState
*bdrv_lookup_bs(const char *device
,
4778 const char *node_name
,
4782 BlockDriverState
*bs
;
4785 blk
= blk_by_name(device
);
4790 error_setg(errp
, "Device '%s' has no medium", device
);
4798 bs
= bdrv_find_node(node_name
);
4805 error_setg(errp
, "Cannot find device=%s nor node_name=%s",
4806 device
? device
: "",
4807 node_name
? node_name
: "");
4811 /* If 'base' is in the same chain as 'top', return true. Otherwise,
4812 * return false. If either argument is NULL, return false. */
4813 bool bdrv_chain_contains(BlockDriverState
*top
, BlockDriverState
*base
)
4815 while (top
&& top
!= base
) {
4816 top
= backing_bs(top
);
4822 BlockDriverState
*bdrv_next_node(BlockDriverState
*bs
)
4825 return QTAILQ_FIRST(&graph_bdrv_states
);
4827 return QTAILQ_NEXT(bs
, node_list
);
4830 BlockDriverState
*bdrv_next_all_states(BlockDriverState
*bs
)
4833 return QTAILQ_FIRST(&all_bdrv_states
);
4835 return QTAILQ_NEXT(bs
, bs_list
);
4838 const char *bdrv_get_node_name(const BlockDriverState
*bs
)
4840 return bs
->node_name
;
4843 const char *bdrv_get_parent_name(const BlockDriverState
*bs
)
4848 /* If multiple parents have a name, just pick the first one. */
4849 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
4850 if (c
->role
->get_name
) {
4851 name
= c
->role
->get_name(c
);
4852 if (name
&& *name
) {
4861 /* TODO check what callers really want: bs->node_name or blk_name() */
4862 const char *bdrv_get_device_name(const BlockDriverState
*bs
)
4864 return bdrv_get_parent_name(bs
) ?: "";
4867 /* This can be used to identify nodes that might not have a device
4868 * name associated. Since node and device names live in the same
4869 * namespace, the result is unambiguous. The exception is if both are
4870 * absent, then this returns an empty (non-null) string. */
4871 const char *bdrv_get_device_or_node_name(const BlockDriverState
*bs
)
4873 return bdrv_get_parent_name(bs
) ?: bs
->node_name
;
4876 int bdrv_get_flags(BlockDriverState
*bs
)
4878 return bs
->open_flags
;
4881 int bdrv_has_zero_init_1(BlockDriverState
*bs
)
4886 int bdrv_has_zero_init(BlockDriverState
*bs
)
4892 /* If BS is a copy on write image, it is initialized to
4893 the contents of the base image, which may not be zeroes. */
4897 if (bs
->drv
->bdrv_has_zero_init
) {
4898 return bs
->drv
->bdrv_has_zero_init(bs
);
4900 if (bs
->file
&& bs
->drv
->is_filter
) {
4901 return bdrv_has_zero_init(bs
->file
->bs
);
4908 bool bdrv_unallocated_blocks_are_zero(BlockDriverState
*bs
)
4910 BlockDriverInfo bdi
;
4916 if (bdrv_get_info(bs
, &bdi
) == 0) {
4917 return bdi
.unallocated_blocks_are_zero
;
4923 bool bdrv_can_write_zeroes_with_unmap(BlockDriverState
*bs
)
4925 if (!(bs
->open_flags
& BDRV_O_UNMAP
)) {
4929 return bs
->supported_zero_flags
& BDRV_REQ_MAY_UNMAP
;
4932 void bdrv_get_backing_filename(BlockDriverState
*bs
,
4933 char *filename
, int filename_size
)
4935 pstrcpy(filename
, filename_size
, bs
->backing_file
);
4938 int bdrv_get_info(BlockDriverState
*bs
, BlockDriverInfo
*bdi
)
4940 BlockDriver
*drv
= bs
->drv
;
4941 /* if bs->drv == NULL, bs is closed, so there's nothing to do here */
4945 if (!drv
->bdrv_get_info
) {
4946 if (bs
->file
&& drv
->is_filter
) {
4947 return bdrv_get_info(bs
->file
->bs
, bdi
);
4951 memset(bdi
, 0, sizeof(*bdi
));
4952 return drv
->bdrv_get_info(bs
, bdi
);
4955 ImageInfoSpecific
*bdrv_get_specific_info(BlockDriverState
*bs
,
4958 BlockDriver
*drv
= bs
->drv
;
4959 if (drv
&& drv
->bdrv_get_specific_info
) {
4960 return drv
->bdrv_get_specific_info(bs
, errp
);
4965 void bdrv_debug_event(BlockDriverState
*bs
, BlkdebugEvent event
)
4967 if (!bs
|| !bs
->drv
|| !bs
->drv
->bdrv_debug_event
) {
4971 bs
->drv
->bdrv_debug_event(bs
, event
);
4974 int bdrv_debug_breakpoint(BlockDriverState
*bs
, const char *event
,
4977 while (bs
&& bs
->drv
&& !bs
->drv
->bdrv_debug_breakpoint
) {
4978 bs
= bs
->file
? bs
->file
->bs
: NULL
;
4981 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_breakpoint
) {
4982 return bs
->drv
->bdrv_debug_breakpoint(bs
, event
, tag
);
4988 int bdrv_debug_remove_breakpoint(BlockDriverState
*bs
, const char *tag
)
4990 while (bs
&& bs
->drv
&& !bs
->drv
->bdrv_debug_remove_breakpoint
) {
4991 bs
= bs
->file
? bs
->file
->bs
: NULL
;
4994 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_remove_breakpoint
) {
4995 return bs
->drv
->bdrv_debug_remove_breakpoint(bs
, tag
);
5001 int bdrv_debug_resume(BlockDriverState
*bs
, const char *tag
)
5003 while (bs
&& (!bs
->drv
|| !bs
->drv
->bdrv_debug_resume
)) {
5004 bs
= bs
->file
? bs
->file
->bs
: NULL
;
5007 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_resume
) {
5008 return bs
->drv
->bdrv_debug_resume(bs
, tag
);
5014 bool bdrv_debug_is_suspended(BlockDriverState
*bs
, const char *tag
)
5016 while (bs
&& bs
->drv
&& !bs
->drv
->bdrv_debug_is_suspended
) {
5017 bs
= bs
->file
? bs
->file
->bs
: NULL
;
5020 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_is_suspended
) {
5021 return bs
->drv
->bdrv_debug_is_suspended(bs
, tag
);
5027 /* backing_file can either be relative, or absolute, or a protocol. If it is
5028 * relative, it must be relative to the chain. So, passing in bs->filename
5029 * from a BDS as backing_file should not be done, as that may be relative to
5030 * the CWD rather than the chain. */
5031 BlockDriverState
*bdrv_find_backing_image(BlockDriverState
*bs
,
5032 const char *backing_file
)
5034 char *filename_full
= NULL
;
5035 char *backing_file_full
= NULL
;
5036 char *filename_tmp
= NULL
;
5037 int is_protocol
= 0;
5038 BlockDriverState
*curr_bs
= NULL
;
5039 BlockDriverState
*retval
= NULL
;
5041 if (!bs
|| !bs
->drv
|| !backing_file
) {
5045 filename_full
= g_malloc(PATH_MAX
);
5046 backing_file_full
= g_malloc(PATH_MAX
);
5048 is_protocol
= path_has_protocol(backing_file
);
5050 for (curr_bs
= bs
; curr_bs
->backing
; curr_bs
= curr_bs
->backing
->bs
) {
5052 /* If either of the filename paths is actually a protocol, then
5053 * compare unmodified paths; otherwise make paths relative */
5054 if (is_protocol
|| path_has_protocol(curr_bs
->backing_file
)) {
5055 char *backing_file_full_ret
;
5057 if (strcmp(backing_file
, curr_bs
->backing_file
) == 0) {
5058 retval
= curr_bs
->backing
->bs
;
5061 /* Also check against the full backing filename for the image */
5062 backing_file_full_ret
= bdrv_get_full_backing_filename(curr_bs
,
5064 if (backing_file_full_ret
) {
5065 bool equal
= strcmp(backing_file
, backing_file_full_ret
) == 0;
5066 g_free(backing_file_full_ret
);
5068 retval
= curr_bs
->backing
->bs
;
5073 /* If not an absolute filename path, make it relative to the current
5074 * image's filename path */
5075 filename_tmp
= bdrv_make_absolute_filename(curr_bs
, backing_file
,
5077 /* We are going to compare canonicalized absolute pathnames */
5078 if (!filename_tmp
|| !realpath(filename_tmp
, filename_full
)) {
5079 g_free(filename_tmp
);
5082 g_free(filename_tmp
);
5084 /* We need to make sure the backing filename we are comparing against
5085 * is relative to the current image filename (or absolute) */
5086 filename_tmp
= bdrv_get_full_backing_filename(curr_bs
, NULL
);
5087 if (!filename_tmp
|| !realpath(filename_tmp
, backing_file_full
)) {
5088 g_free(filename_tmp
);
5091 g_free(filename_tmp
);
5093 if (strcmp(backing_file_full
, filename_full
) == 0) {
5094 retval
= curr_bs
->backing
->bs
;
5100 g_free(filename_full
);
5101 g_free(backing_file_full
);
5105 void bdrv_init(void)
5107 module_call_init(MODULE_INIT_BLOCK
);
5110 void bdrv_init_with_whitelist(void)
5112 use_bdrv_whitelist
= 1;
5116 static void coroutine_fn
bdrv_co_invalidate_cache(BlockDriverState
*bs
,
5119 BdrvChild
*child
, *parent
;
5120 uint64_t perm
, shared_perm
;
5121 Error
*local_err
= NULL
;
5123 BdrvDirtyBitmap
*bm
;
5129 if (!(bs
->open_flags
& BDRV_O_INACTIVE
)) {
5133 QLIST_FOREACH(child
, &bs
->children
, next
) {
5134 bdrv_co_invalidate_cache(child
->bs
, &local_err
);
5136 error_propagate(errp
, local_err
);
5142 * Update permissions, they may differ for inactive nodes.
5144 * Note that the required permissions of inactive images are always a
5145 * subset of the permissions required after activating the image. This
5146 * allows us to just get the permissions upfront without restricting
5147 * drv->bdrv_invalidate_cache().
5149 * It also means that in error cases, we don't have to try and revert to
5150 * the old permissions (which is an operation that could fail, too). We can
5151 * just keep the extended permissions for the next time that an activation
5152 * of the image is tried.
5154 bs
->open_flags
&= ~BDRV_O_INACTIVE
;
5155 bdrv_get_cumulative_perm(bs
, &perm
, &shared_perm
);
5156 ret
= bdrv_check_perm(bs
, NULL
, perm
, shared_perm
, NULL
, &local_err
);
5158 bs
->open_flags
|= BDRV_O_INACTIVE
;
5159 error_propagate(errp
, local_err
);
5162 bdrv_set_perm(bs
, perm
, shared_perm
);
5164 if (bs
->drv
->bdrv_co_invalidate_cache
) {
5165 bs
->drv
->bdrv_co_invalidate_cache(bs
, &local_err
);
5167 bs
->open_flags
|= BDRV_O_INACTIVE
;
5168 error_propagate(errp
, local_err
);
5173 for (bm
= bdrv_dirty_bitmap_next(bs
, NULL
); bm
;
5174 bm
= bdrv_dirty_bitmap_next(bs
, bm
))
5176 bdrv_dirty_bitmap_set_migration(bm
, false);
5179 ret
= refresh_total_sectors(bs
, bs
->total_sectors
);
5181 bs
->open_flags
|= BDRV_O_INACTIVE
;
5182 error_setg_errno(errp
, -ret
, "Could not refresh total sector count");
5186 QLIST_FOREACH(parent
, &bs
->parents
, next_parent
) {
5187 if (parent
->role
->activate
) {
5188 parent
->role
->activate(parent
, &local_err
);
5190 bs
->open_flags
|= BDRV_O_INACTIVE
;
5191 error_propagate(errp
, local_err
);
5198 typedef struct InvalidateCacheCo
{
5199 BlockDriverState
*bs
;
5202 } InvalidateCacheCo
;
5204 static void coroutine_fn
bdrv_invalidate_cache_co_entry(void *opaque
)
5206 InvalidateCacheCo
*ico
= opaque
;
5207 bdrv_co_invalidate_cache(ico
->bs
, ico
->errp
);
5212 void bdrv_invalidate_cache(BlockDriverState
*bs
, Error
**errp
)
5215 InvalidateCacheCo ico
= {
5221 if (qemu_in_coroutine()) {
5222 /* Fast-path if already in coroutine context */
5223 bdrv_invalidate_cache_co_entry(&ico
);
5225 co
= qemu_coroutine_create(bdrv_invalidate_cache_co_entry
, &ico
);
5226 bdrv_coroutine_enter(bs
, co
);
5227 BDRV_POLL_WHILE(bs
, !ico
.done
);
5231 void bdrv_invalidate_cache_all(Error
**errp
)
5233 BlockDriverState
*bs
;
5234 Error
*local_err
= NULL
;
5235 BdrvNextIterator it
;
5237 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
5238 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
5240 aio_context_acquire(aio_context
);
5241 bdrv_invalidate_cache(bs
, &local_err
);
5242 aio_context_release(aio_context
);
5244 error_propagate(errp
, local_err
);
5245 bdrv_next_cleanup(&it
);
5251 static bool bdrv_has_bds_parent(BlockDriverState
*bs
, bool only_active
)
5255 QLIST_FOREACH(parent
, &bs
->parents
, next_parent
) {
5256 if (parent
->role
->parent_is_bds
) {
5257 BlockDriverState
*parent_bs
= parent
->opaque
;
5258 if (!only_active
|| !(parent_bs
->open_flags
& BDRV_O_INACTIVE
)) {
5267 static int bdrv_inactivate_recurse(BlockDriverState
*bs
)
5269 BdrvChild
*child
, *parent
;
5270 uint64_t perm
, shared_perm
;
5277 /* Make sure that we don't inactivate a child before its parent.
5278 * It will be covered by recursion from the yet active parent. */
5279 if (bdrv_has_bds_parent(bs
, true)) {
5283 assert(!(bs
->open_flags
& BDRV_O_INACTIVE
));
5285 /* Inactivate this node */
5286 if (bs
->drv
->bdrv_inactivate
) {
5287 ret
= bs
->drv
->bdrv_inactivate(bs
);
5293 QLIST_FOREACH(parent
, &bs
->parents
, next_parent
) {
5294 if (parent
->role
->inactivate
) {
5295 ret
= parent
->role
->inactivate(parent
);
5302 bs
->open_flags
|= BDRV_O_INACTIVE
;
5304 /* Update permissions, they may differ for inactive nodes */
5305 bdrv_get_cumulative_perm(bs
, &perm
, &shared_perm
);
5306 bdrv_check_perm(bs
, NULL
, perm
, shared_perm
, NULL
, &error_abort
);
5307 bdrv_set_perm(bs
, perm
, shared_perm
);
5310 /* Recursively inactivate children */
5311 QLIST_FOREACH(child
, &bs
->children
, next
) {
5312 ret
= bdrv_inactivate_recurse(child
->bs
);
5321 int bdrv_inactivate_all(void)
5323 BlockDriverState
*bs
= NULL
;
5324 BdrvNextIterator it
;
5326 GSList
*aio_ctxs
= NULL
, *ctx
;
5328 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
5329 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
5331 if (!g_slist_find(aio_ctxs
, aio_context
)) {
5332 aio_ctxs
= g_slist_prepend(aio_ctxs
, aio_context
);
5333 aio_context_acquire(aio_context
);
5337 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
5338 /* Nodes with BDS parents are covered by recursion from the last
5339 * parent that gets inactivated. Don't inactivate them a second
5340 * time if that has already happened. */
5341 if (bdrv_has_bds_parent(bs
, false)) {
5344 ret
= bdrv_inactivate_recurse(bs
);
5346 bdrv_next_cleanup(&it
);
5352 for (ctx
= aio_ctxs
; ctx
!= NULL
; ctx
= ctx
->next
) {
5353 AioContext
*aio_context
= ctx
->data
;
5354 aio_context_release(aio_context
);
5356 g_slist_free(aio_ctxs
);
5361 /**************************************************************/
5362 /* removable device support */
5365 * Return TRUE if the media is present
5367 bool bdrv_is_inserted(BlockDriverState
*bs
)
5369 BlockDriver
*drv
= bs
->drv
;
5375 if (drv
->bdrv_is_inserted
) {
5376 return drv
->bdrv_is_inserted(bs
);
5378 QLIST_FOREACH(child
, &bs
->children
, next
) {
5379 if (!bdrv_is_inserted(child
->bs
)) {
5387 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
5389 void bdrv_eject(BlockDriverState
*bs
, bool eject_flag
)
5391 BlockDriver
*drv
= bs
->drv
;
5393 if (drv
&& drv
->bdrv_eject
) {
5394 drv
->bdrv_eject(bs
, eject_flag
);
5399 * Lock or unlock the media (if it is locked, the user won't be able
5400 * to eject it manually).
5402 void bdrv_lock_medium(BlockDriverState
*bs
, bool locked
)
5404 BlockDriver
*drv
= bs
->drv
;
5406 trace_bdrv_lock_medium(bs
, locked
);
5408 if (drv
&& drv
->bdrv_lock_medium
) {
5409 drv
->bdrv_lock_medium(bs
, locked
);
5413 /* Get a reference to bs */
5414 void bdrv_ref(BlockDriverState
*bs
)
5419 /* Release a previously grabbed reference to bs.
5420 * If after releasing, reference count is zero, the BlockDriverState is
5422 void bdrv_unref(BlockDriverState
*bs
)
5427 assert(bs
->refcnt
> 0);
5428 if (--bs
->refcnt
== 0) {
5433 struct BdrvOpBlocker
{
5435 QLIST_ENTRY(BdrvOpBlocker
) list
;
5438 bool bdrv_op_is_blocked(BlockDriverState
*bs
, BlockOpType op
, Error
**errp
)
5440 BdrvOpBlocker
*blocker
;
5441 assert((int) op
>= 0 && op
< BLOCK_OP_TYPE_MAX
);
5442 if (!QLIST_EMPTY(&bs
->op_blockers
[op
])) {
5443 blocker
= QLIST_FIRST(&bs
->op_blockers
[op
]);
5444 error_propagate_prepend(errp
, error_copy(blocker
->reason
),
5445 "Node '%s' is busy: ",
5446 bdrv_get_device_or_node_name(bs
));
5452 void bdrv_op_block(BlockDriverState
*bs
, BlockOpType op
, Error
*reason
)
5454 BdrvOpBlocker
*blocker
;
5455 assert((int) op
>= 0 && op
< BLOCK_OP_TYPE_MAX
);
5457 blocker
= g_new0(BdrvOpBlocker
, 1);
5458 blocker
->reason
= reason
;
5459 QLIST_INSERT_HEAD(&bs
->op_blockers
[op
], blocker
, list
);
5462 void bdrv_op_unblock(BlockDriverState
*bs
, BlockOpType op
, Error
*reason
)
5464 BdrvOpBlocker
*blocker
, *next
;
5465 assert((int) op
>= 0 && op
< BLOCK_OP_TYPE_MAX
);
5466 QLIST_FOREACH_SAFE(blocker
, &bs
->op_blockers
[op
], list
, next
) {
5467 if (blocker
->reason
== reason
) {
5468 QLIST_REMOVE(blocker
, list
);
5474 void bdrv_op_block_all(BlockDriverState
*bs
, Error
*reason
)
5477 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
5478 bdrv_op_block(bs
, i
, reason
);
5482 void bdrv_op_unblock_all(BlockDriverState
*bs
, Error
*reason
)
5485 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
5486 bdrv_op_unblock(bs
, i
, reason
);
5490 bool bdrv_op_blocker_is_empty(BlockDriverState
*bs
)
5494 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
5495 if (!QLIST_EMPTY(&bs
->op_blockers
[i
])) {
5502 void bdrv_img_create(const char *filename
, const char *fmt
,
5503 const char *base_filename
, const char *base_fmt
,
5504 char *options
, uint64_t img_size
, int flags
, bool quiet
,
5507 QemuOptsList
*create_opts
= NULL
;
5508 QemuOpts
*opts
= NULL
;
5509 const char *backing_fmt
, *backing_file
;
5511 BlockDriver
*drv
, *proto_drv
;
5512 Error
*local_err
= NULL
;
5515 /* Find driver and parse its options */
5516 drv
= bdrv_find_format(fmt
);
5518 error_setg(errp
, "Unknown file format '%s'", fmt
);
5522 proto_drv
= bdrv_find_protocol(filename
, true, errp
);
5527 if (!drv
->create_opts
) {
5528 error_setg(errp
, "Format driver '%s' does not support image creation",
5533 if (!proto_drv
->create_opts
) {
5534 error_setg(errp
, "Protocol driver '%s' does not support image creation",
5535 proto_drv
->format_name
);
5539 create_opts
= qemu_opts_append(create_opts
, drv
->create_opts
);
5540 create_opts
= qemu_opts_append(create_opts
, proto_drv
->create_opts
);
5542 /* Create parameter list with default values */
5543 opts
= qemu_opts_create(create_opts
, NULL
, 0, &error_abort
);
5544 qemu_opt_set_number(opts
, BLOCK_OPT_SIZE
, img_size
, &error_abort
);
5546 /* Parse -o options */
5548 qemu_opts_do_parse(opts
, options
, NULL
, &local_err
);
5554 if (base_filename
) {
5555 qemu_opt_set(opts
, BLOCK_OPT_BACKING_FILE
, base_filename
, &local_err
);
5557 error_setg(errp
, "Backing file not supported for file format '%s'",
5564 qemu_opt_set(opts
, BLOCK_OPT_BACKING_FMT
, base_fmt
, &local_err
);
5566 error_setg(errp
, "Backing file format not supported for file "
5567 "format '%s'", fmt
);
5572 backing_file
= qemu_opt_get(opts
, BLOCK_OPT_BACKING_FILE
);
5574 if (!strcmp(filename
, backing_file
)) {
5575 error_setg(errp
, "Error: Trying to create an image with the "
5576 "same filename as the backing file");
5581 backing_fmt
= qemu_opt_get(opts
, BLOCK_OPT_BACKING_FMT
);
5583 /* The size for the image must always be specified, unless we have a backing
5584 * file and we have not been forbidden from opening it. */
5585 size
= qemu_opt_get_size(opts
, BLOCK_OPT_SIZE
, img_size
);
5586 if (backing_file
&& !(flags
& BDRV_O_NO_BACKING
)) {
5587 BlockDriverState
*bs
;
5590 QDict
*backing_options
= NULL
;
5593 bdrv_get_full_backing_filename_from_filename(filename
, backing_file
,
5598 assert(full_backing
);
5600 /* backing files always opened read-only */
5602 back_flags
&= ~(BDRV_O_RDWR
| BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
);
5604 backing_options
= qdict_new();
5606 qdict_put_str(backing_options
, "driver", backing_fmt
);
5608 qdict_put_bool(backing_options
, BDRV_OPT_FORCE_SHARE
, true);
5610 bs
= bdrv_open(full_backing
, NULL
, backing_options
, back_flags
,
5612 g_free(full_backing
);
5613 if (!bs
&& size
!= -1) {
5614 /* Couldn't open BS, but we have a size, so it's nonfatal */
5615 warn_reportf_err(local_err
,
5616 "Could not verify backing image. "
5617 "This may become an error in future versions.\n");
5620 /* Couldn't open bs, do not have size */
5621 error_append_hint(&local_err
,
5622 "Could not open backing image to determine size.\n");
5626 /* Opened BS, have no size */
5627 size
= bdrv_getlength(bs
);
5629 error_setg_errno(errp
, -size
, "Could not get size of '%s'",
5634 qemu_opt_set_number(opts
, BLOCK_OPT_SIZE
, size
, &error_abort
);
5638 } /* (backing_file && !(flags & BDRV_O_NO_BACKING)) */
5641 error_setg(errp
, "Image creation needs a size parameter");
5646 printf("Formatting '%s', fmt=%s ", filename
, fmt
);
5647 qemu_opts_print(opts
, " ");
5651 ret
= bdrv_create(drv
, filename
, opts
, &local_err
);
5653 if (ret
== -EFBIG
) {
5654 /* This is generally a better message than whatever the driver would
5655 * deliver (especially because of the cluster_size_hint), since that
5656 * is most probably not much different from "image too large". */
5657 const char *cluster_size_hint
= "";
5658 if (qemu_opt_get_size(opts
, BLOCK_OPT_CLUSTER_SIZE
, 0)) {
5659 cluster_size_hint
= " (try using a larger cluster size)";
5661 error_setg(errp
, "The image size is too large for file format '%s'"
5662 "%s", fmt
, cluster_size_hint
);
5663 error_free(local_err
);
5668 qemu_opts_del(opts
);
5669 qemu_opts_free(create_opts
);
5670 error_propagate(errp
, local_err
);
5673 AioContext
*bdrv_get_aio_context(BlockDriverState
*bs
)
5675 return bs
? bs
->aio_context
: qemu_get_aio_context();
5678 void bdrv_coroutine_enter(BlockDriverState
*bs
, Coroutine
*co
)
5680 aio_co_enter(bdrv_get_aio_context(bs
), co
);
5683 static void bdrv_do_remove_aio_context_notifier(BdrvAioNotifier
*ban
)
5685 QLIST_REMOVE(ban
, list
);
5689 static void bdrv_detach_aio_context(BlockDriverState
*bs
)
5691 BdrvAioNotifier
*baf
, *baf_tmp
;
5693 assert(!bs
->walking_aio_notifiers
);
5694 bs
->walking_aio_notifiers
= true;
5695 QLIST_FOREACH_SAFE(baf
, &bs
->aio_notifiers
, list
, baf_tmp
) {
5697 bdrv_do_remove_aio_context_notifier(baf
);
5699 baf
->detach_aio_context(baf
->opaque
);
5702 /* Never mind iterating again to check for ->deleted. bdrv_close() will
5703 * remove remaining aio notifiers if we aren't called again.
5705 bs
->walking_aio_notifiers
= false;
5707 if (bs
->drv
&& bs
->drv
->bdrv_detach_aio_context
) {
5708 bs
->drv
->bdrv_detach_aio_context(bs
);
5711 if (bs
->quiesce_counter
) {
5712 aio_enable_external(bs
->aio_context
);
5714 bs
->aio_context
= NULL
;
5717 static void bdrv_attach_aio_context(BlockDriverState
*bs
,
5718 AioContext
*new_context
)
5720 BdrvAioNotifier
*ban
, *ban_tmp
;
5722 if (bs
->quiesce_counter
) {
5723 aio_disable_external(new_context
);
5726 bs
->aio_context
= new_context
;
5728 if (bs
->drv
&& bs
->drv
->bdrv_attach_aio_context
) {
5729 bs
->drv
->bdrv_attach_aio_context(bs
, new_context
);
5732 assert(!bs
->walking_aio_notifiers
);
5733 bs
->walking_aio_notifiers
= true;
5734 QLIST_FOREACH_SAFE(ban
, &bs
->aio_notifiers
, list
, ban_tmp
) {
5736 bdrv_do_remove_aio_context_notifier(ban
);
5738 ban
->attached_aio_context(new_context
, ban
->opaque
);
5741 bs
->walking_aio_notifiers
= false;
5744 /* @ignore will accumulate all visited BdrvChild object. The caller is
5745 * responsible for freeing the list afterwards. */
5746 void bdrv_set_aio_context_ignore(BlockDriverState
*bs
,
5747 AioContext
*new_context
, GSList
**ignore
)
5751 if (bdrv_get_aio_context(bs
) == new_context
) {
5755 bdrv_drained_begin(bs
);
5757 QLIST_FOREACH(child
, &bs
->children
, next
) {
5758 if (g_slist_find(*ignore
, child
)) {
5761 *ignore
= g_slist_prepend(*ignore
, child
);
5762 bdrv_set_aio_context_ignore(child
->bs
, new_context
, ignore
);
5764 QLIST_FOREACH(child
, &bs
->parents
, next_parent
) {
5765 if (g_slist_find(*ignore
, child
)) {
5768 if (child
->role
->set_aio_ctx
) {
5769 *ignore
= g_slist_prepend(*ignore
, child
);
5770 child
->role
->set_aio_ctx(child
, new_context
, ignore
);
5774 bdrv_detach_aio_context(bs
);
5776 /* This function executes in the old AioContext so acquire the new one in
5777 * case it runs in a different thread.
5779 aio_context_acquire(new_context
);
5780 bdrv_attach_aio_context(bs
, new_context
);
5781 bdrv_drained_end(bs
);
5782 aio_context_release(new_context
);
5785 /* The caller must own the AioContext lock for the old AioContext of bs, but it
5786 * must not own the AioContext lock for new_context (unless new_context is
5787 * the same as the current context of bs). */
5788 void bdrv_set_aio_context(BlockDriverState
*bs
, AioContext
*new_context
)
5790 GSList
*ignore_list
= NULL
;
5791 bdrv_set_aio_context_ignore(bs
, new_context
, &ignore_list
);
5792 g_slist_free(ignore_list
);
5795 static bool bdrv_parent_can_set_aio_context(BdrvChild
*c
, AioContext
*ctx
,
5796 GSList
**ignore
, Error
**errp
)
5798 if (g_slist_find(*ignore
, c
)) {
5801 *ignore
= g_slist_prepend(*ignore
, c
);
5803 /* A BdrvChildRole that doesn't handle AioContext changes cannot
5804 * tolerate any AioContext changes */
5805 if (!c
->role
->can_set_aio_ctx
) {
5806 char *user
= bdrv_child_user_desc(c
);
5807 error_setg(errp
, "Changing iothreads is not supported by %s", user
);
5811 if (!c
->role
->can_set_aio_ctx(c
, ctx
, ignore
, errp
)) {
5812 assert(!errp
|| *errp
);
5818 bool bdrv_child_can_set_aio_context(BdrvChild
*c
, AioContext
*ctx
,
5819 GSList
**ignore
, Error
**errp
)
5821 if (g_slist_find(*ignore
, c
)) {
5824 *ignore
= g_slist_prepend(*ignore
, c
);
5825 return bdrv_can_set_aio_context(c
->bs
, ctx
, ignore
, errp
);
5828 /* @ignore will accumulate all visited BdrvChild object. The caller is
5829 * responsible for freeing the list afterwards. */
5830 bool bdrv_can_set_aio_context(BlockDriverState
*bs
, AioContext
*ctx
,
5831 GSList
**ignore
, Error
**errp
)
5835 if (bdrv_get_aio_context(bs
) == ctx
) {
5839 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
5840 if (!bdrv_parent_can_set_aio_context(c
, ctx
, ignore
, errp
)) {
5844 QLIST_FOREACH(c
, &bs
->children
, next
) {
5845 if (!bdrv_child_can_set_aio_context(c
, ctx
, ignore
, errp
)) {
5853 int bdrv_child_try_set_aio_context(BlockDriverState
*bs
, AioContext
*ctx
,
5854 BdrvChild
*ignore_child
, Error
**errp
)
5859 ignore
= ignore_child
? g_slist_prepend(NULL
, ignore_child
) : NULL
;
5860 ret
= bdrv_can_set_aio_context(bs
, ctx
, &ignore
, errp
);
5861 g_slist_free(ignore
);
5867 ignore
= ignore_child
? g_slist_prepend(NULL
, ignore_child
) : NULL
;
5868 bdrv_set_aio_context_ignore(bs
, ctx
, &ignore
);
5869 g_slist_free(ignore
);
5874 int bdrv_try_set_aio_context(BlockDriverState
*bs
, AioContext
*ctx
,
5877 return bdrv_child_try_set_aio_context(bs
, ctx
, NULL
, errp
);
5880 void bdrv_add_aio_context_notifier(BlockDriverState
*bs
,
5881 void (*attached_aio_context
)(AioContext
*new_context
, void *opaque
),
5882 void (*detach_aio_context
)(void *opaque
), void *opaque
)
5884 BdrvAioNotifier
*ban
= g_new(BdrvAioNotifier
, 1);
5885 *ban
= (BdrvAioNotifier
){
5886 .attached_aio_context
= attached_aio_context
,
5887 .detach_aio_context
= detach_aio_context
,
5891 QLIST_INSERT_HEAD(&bs
->aio_notifiers
, ban
, list
);
5894 void bdrv_remove_aio_context_notifier(BlockDriverState
*bs
,
5895 void (*attached_aio_context
)(AioContext
*,
5897 void (*detach_aio_context
)(void *),
5900 BdrvAioNotifier
*ban
, *ban_next
;
5902 QLIST_FOREACH_SAFE(ban
, &bs
->aio_notifiers
, list
, ban_next
) {
5903 if (ban
->attached_aio_context
== attached_aio_context
&&
5904 ban
->detach_aio_context
== detach_aio_context
&&
5905 ban
->opaque
== opaque
&&
5906 ban
->deleted
== false)
5908 if (bs
->walking_aio_notifiers
) {
5909 ban
->deleted
= true;
5911 bdrv_do_remove_aio_context_notifier(ban
);
5920 int bdrv_amend_options(BlockDriverState
*bs
, QemuOpts
*opts
,
5921 BlockDriverAmendStatusCB
*status_cb
, void *cb_opaque
,
5925 error_setg(errp
, "Node is ejected");
5928 if (!bs
->drv
->bdrv_amend_options
) {
5929 error_setg(errp
, "Block driver '%s' does not support option amendment",
5930 bs
->drv
->format_name
);
5933 return bs
->drv
->bdrv_amend_options(bs
, opts
, status_cb
, cb_opaque
, errp
);
5936 /* This function will be called by the bdrv_recurse_is_first_non_filter method
5937 * of block filter and by bdrv_is_first_non_filter.
5938 * It is used to test if the given bs is the candidate or recurse more in the
5941 bool bdrv_recurse_is_first_non_filter(BlockDriverState
*bs
,
5942 BlockDriverState
*candidate
)
5944 /* return false if basic checks fails */
5945 if (!bs
|| !bs
->drv
) {
5949 /* the code reached a non block filter driver -> check if the bs is
5950 * the same as the candidate. It's the recursion termination condition.
5952 if (!bs
->drv
->is_filter
) {
5953 return bs
== candidate
;
5955 /* Down this path the driver is a block filter driver */
5957 /* If the block filter recursion method is defined use it to recurse down
5960 if (bs
->drv
->bdrv_recurse_is_first_non_filter
) {
5961 return bs
->drv
->bdrv_recurse_is_first_non_filter(bs
, candidate
);
5964 /* the driver is a block filter but don't allow to recurse -> return false
5969 /* This function checks if the candidate is the first non filter bs down it's
5970 * bs chain. Since we don't have pointers to parents it explore all bs chains
5971 * from the top. Some filters can choose not to pass down the recursion.
5973 bool bdrv_is_first_non_filter(BlockDriverState
*candidate
)
5975 BlockDriverState
*bs
;
5976 BdrvNextIterator it
;
5978 /* walk down the bs forest recursively */
5979 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
5982 /* try to recurse in this top level bs */
5983 perm
= bdrv_recurse_is_first_non_filter(bs
, candidate
);
5985 /* candidate is the first non filter */
5987 bdrv_next_cleanup(&it
);
5995 BlockDriverState
*check_to_replace_node(BlockDriverState
*parent_bs
,
5996 const char *node_name
, Error
**errp
)
5998 BlockDriverState
*to_replace_bs
= bdrv_find_node(node_name
);
5999 AioContext
*aio_context
;
6001 if (!to_replace_bs
) {
6002 error_setg(errp
, "Node name '%s' not found", node_name
);
6006 aio_context
= bdrv_get_aio_context(to_replace_bs
);
6007 aio_context_acquire(aio_context
);
6009 if (bdrv_op_is_blocked(to_replace_bs
, BLOCK_OP_TYPE_REPLACE
, errp
)) {
6010 to_replace_bs
= NULL
;
6014 /* We don't want arbitrary node of the BDS chain to be replaced only the top
6015 * most non filter in order to prevent data corruption.
6016 * Another benefit is that this tests exclude backing files which are
6017 * blocked by the backing blockers.
6019 if (!bdrv_recurse_is_first_non_filter(parent_bs
, to_replace_bs
)) {
6020 error_setg(errp
, "Only top most non filter can be replaced");
6021 to_replace_bs
= NULL
;
6026 aio_context_release(aio_context
);
6027 return to_replace_bs
;
6031 * Iterates through the list of runtime option keys that are said to
6032 * be "strong" for a BDS. An option is called "strong" if it changes
6033 * a BDS's data. For example, the null block driver's "size" and
6034 * "read-zeroes" options are strong, but its "latency-ns" option is
6037 * If a key returned by this function ends with a dot, all options
6038 * starting with that prefix are strong.
6040 static const char *const *strong_options(BlockDriverState
*bs
,
6041 const char *const *curopt
)
6043 static const char *const global_options
[] = {
6044 "driver", "filename", NULL
6048 return &global_options
[0];
6052 if (curopt
== &global_options
[ARRAY_SIZE(global_options
) - 1] && bs
->drv
) {
6053 curopt
= bs
->drv
->strong_runtime_opts
;
6056 return (curopt
&& *curopt
) ? curopt
: NULL
;
6060 * Copies all strong runtime options from bs->options to the given
6061 * QDict. The set of strong option keys is determined by invoking
6064 * Returns true iff any strong option was present in bs->options (and
6065 * thus copied to the target QDict) with the exception of "filename"
6066 * and "driver". The caller is expected to use this value to decide
6067 * whether the existence of strong options prevents the generation of
6070 static bool append_strong_runtime_options(QDict
*d
, BlockDriverState
*bs
)
6072 bool found_any
= false;
6073 const char *const *option_name
= NULL
;
6079 while ((option_name
= strong_options(bs
, option_name
))) {
6080 bool option_given
= false;
6082 assert(strlen(*option_name
) > 0);
6083 if ((*option_name
)[strlen(*option_name
) - 1] != '.') {
6084 QObject
*entry
= qdict_get(bs
->options
, *option_name
);
6089 qdict_put_obj(d
, *option_name
, qobject_ref(entry
));
6090 option_given
= true;
6092 const QDictEntry
*entry
;
6093 for (entry
= qdict_first(bs
->options
); entry
;
6094 entry
= qdict_next(bs
->options
, entry
))
6096 if (strstart(qdict_entry_key(entry
), *option_name
, NULL
)) {
6097 qdict_put_obj(d
, qdict_entry_key(entry
),
6098 qobject_ref(qdict_entry_value(entry
)));
6099 option_given
= true;
6104 /* While "driver" and "filename" need to be included in a JSON filename,
6105 * their existence does not prohibit generation of a plain filename. */
6106 if (!found_any
&& option_given
&&
6107 strcmp(*option_name
, "driver") && strcmp(*option_name
, "filename"))
6113 if (!qdict_haskey(d
, "driver")) {
6114 /* Drivers created with bdrv_new_open_driver() may not have a
6115 * @driver option. Add it here. */
6116 qdict_put_str(d
, "driver", bs
->drv
->format_name
);
6122 /* Note: This function may return false positives; it may return true
6123 * even if opening the backing file specified by bs's image header
6124 * would result in exactly bs->backing. */
6125 static bool bdrv_backing_overridden(BlockDriverState
*bs
)
6128 return strcmp(bs
->auto_backing_file
,
6129 bs
->backing
->bs
->filename
);
6131 /* No backing BDS, so if the image header reports any backing
6132 * file, it must have been suppressed */
6133 return bs
->auto_backing_file
[0] != '\0';
6137 /* Updates the following BDS fields:
6138 * - exact_filename: A filename which may be used for opening a block device
6139 * which (mostly) equals the given BDS (even without any
6140 * other options; so reading and writing must return the same
6141 * results, but caching etc. may be different)
6142 * - full_open_options: Options which, when given when opening a block device
6143 * (without a filename), result in a BDS (mostly)
6144 * equalling the given one
6145 * - filename: If exact_filename is set, it is copied here. Otherwise,
6146 * full_open_options is converted to a JSON object, prefixed with
6147 * "json:" (for use through the JSON pseudo protocol) and put here.
6149 void bdrv_refresh_filename(BlockDriverState
*bs
)
6151 BlockDriver
*drv
= bs
->drv
;
6154 bool backing_overridden
;
6155 bool generate_json_filename
; /* Whether our default implementation should
6156 fill exact_filename (false) or not (true) */
6162 /* This BDS's file name may depend on any of its children's file names, so
6163 * refresh those first */
6164 QLIST_FOREACH(child
, &bs
->children
, next
) {
6165 bdrv_refresh_filename(child
->bs
);
6169 /* For implicit nodes, just copy everything from the single child */
6170 child
= QLIST_FIRST(&bs
->children
);
6171 assert(QLIST_NEXT(child
, next
) == NULL
);
6173 pstrcpy(bs
->exact_filename
, sizeof(bs
->exact_filename
),
6174 child
->bs
->exact_filename
);
6175 pstrcpy(bs
->filename
, sizeof(bs
->filename
), child
->bs
->filename
);
6177 bs
->full_open_options
= qobject_ref(child
->bs
->full_open_options
);
6182 backing_overridden
= bdrv_backing_overridden(bs
);
6184 if (bs
->open_flags
& BDRV_O_NO_IO
) {
6185 /* Without I/O, the backing file does not change anything.
6186 * Therefore, in such a case (primarily qemu-img), we can
6187 * pretend the backing file has not been overridden even if
6188 * it technically has been. */
6189 backing_overridden
= false;
6192 /* Gather the options QDict */
6194 generate_json_filename
= append_strong_runtime_options(opts
, bs
);
6195 generate_json_filename
|= backing_overridden
;
6197 if (drv
->bdrv_gather_child_options
) {
6198 /* Some block drivers may not want to present all of their children's
6199 * options, or name them differently from BdrvChild.name */
6200 drv
->bdrv_gather_child_options(bs
, opts
, backing_overridden
);
6202 QLIST_FOREACH(child
, &bs
->children
, next
) {
6203 if (child
->role
== &child_backing
&& !backing_overridden
) {
6204 /* We can skip the backing BDS if it has not been overridden */
6208 qdict_put(opts
, child
->name
,
6209 qobject_ref(child
->bs
->full_open_options
));
6212 if (backing_overridden
&& !bs
->backing
) {
6213 /* Force no backing file */
6214 qdict_put_null(opts
, "backing");
6218 qobject_unref(bs
->full_open_options
);
6219 bs
->full_open_options
= opts
;
6221 if (drv
->bdrv_refresh_filename
) {
6222 /* Obsolete information is of no use here, so drop the old file name
6223 * information before refreshing it */
6224 bs
->exact_filename
[0] = '\0';
6226 drv
->bdrv_refresh_filename(bs
);
6227 } else if (bs
->file
) {
6228 /* Try to reconstruct valid information from the underlying file */
6230 bs
->exact_filename
[0] = '\0';
6233 * We can use the underlying file's filename if:
6234 * - it has a filename,
6235 * - the file is a protocol BDS, and
6236 * - opening that file (as this BDS's format) will automatically create
6237 * the BDS tree we have right now, that is:
6238 * - the user did not significantly change this BDS's behavior with
6239 * some explicit (strong) options
6240 * - no non-file child of this BDS has been overridden by the user
6241 * Both of these conditions are represented by generate_json_filename.
6243 if (bs
->file
->bs
->exact_filename
[0] &&
6244 bs
->file
->bs
->drv
->bdrv_file_open
&&
6245 !generate_json_filename
)
6247 strcpy(bs
->exact_filename
, bs
->file
->bs
->exact_filename
);
6251 if (bs
->exact_filename
[0]) {
6252 pstrcpy(bs
->filename
, sizeof(bs
->filename
), bs
->exact_filename
);
6254 QString
*json
= qobject_to_json(QOBJECT(bs
->full_open_options
));
6255 snprintf(bs
->filename
, sizeof(bs
->filename
), "json:%s",
6256 qstring_get_str(json
));
6257 qobject_unref(json
);
6261 char *bdrv_dirname(BlockDriverState
*bs
, Error
**errp
)
6263 BlockDriver
*drv
= bs
->drv
;
6266 error_setg(errp
, "Node '%s' is ejected", bs
->node_name
);
6270 if (drv
->bdrv_dirname
) {
6271 return drv
->bdrv_dirname(bs
, errp
);
6275 return bdrv_dirname(bs
->file
->bs
, errp
);
6278 bdrv_refresh_filename(bs
);
6279 if (bs
->exact_filename
[0] != '\0') {
6280 return path_combine(bs
->exact_filename
, "");
6283 error_setg(errp
, "Cannot generate a base directory for %s nodes",
6289 * Hot add/remove a BDS's child. So the user can take a child offline when
6290 * it is broken and take a new child online
6292 void bdrv_add_child(BlockDriverState
*parent_bs
, BlockDriverState
*child_bs
,
6296 if (!parent_bs
->drv
|| !parent_bs
->drv
->bdrv_add_child
) {
6297 error_setg(errp
, "The node %s does not support adding a child",
6298 bdrv_get_device_or_node_name(parent_bs
));
6302 if (!QLIST_EMPTY(&child_bs
->parents
)) {
6303 error_setg(errp
, "The node %s already has a parent",
6304 child_bs
->node_name
);
6308 parent_bs
->drv
->bdrv_add_child(parent_bs
, child_bs
, errp
);
6311 void bdrv_del_child(BlockDriverState
*parent_bs
, BdrvChild
*child
, Error
**errp
)
6315 if (!parent_bs
->drv
|| !parent_bs
->drv
->bdrv_del_child
) {
6316 error_setg(errp
, "The node %s does not support removing a child",
6317 bdrv_get_device_or_node_name(parent_bs
));
6321 QLIST_FOREACH(tmp
, &parent_bs
->children
, next
) {
6328 error_setg(errp
, "The node %s does not have a child named %s",
6329 bdrv_get_device_or_node_name(parent_bs
),
6330 bdrv_get_device_or_node_name(child
->bs
));
6334 parent_bs
->drv
->bdrv_del_child(parent_bs
, child
, errp
);
6337 bool bdrv_can_store_new_dirty_bitmap(BlockDriverState
*bs
, const char *name
,
6338 uint32_t granularity
, Error
**errp
)
6340 BlockDriver
*drv
= bs
->drv
;
6343 error_setg_errno(errp
, ENOMEDIUM
,
6344 "Can't store persistent bitmaps to %s",
6345 bdrv_get_device_or_node_name(bs
));
6349 if (!drv
->bdrv_can_store_new_dirty_bitmap
) {
6350 error_setg_errno(errp
, ENOTSUP
,
6351 "Can't store persistent bitmaps to %s",
6352 bdrv_get_device_or_node_name(bs
));
6356 return drv
->bdrv_can_store_new_dirty_bitmap(bs
, name
, granularity
, errp
);