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 "qemu/error-report.h"
31 #include "module_block.h"
32 #include "qemu/module.h"
33 #include "qapi/error.h"
34 #include "qapi/qmp/qdict.h"
35 #include "qapi/qmp/qjson.h"
36 #include "qapi/qmp/qnull.h"
37 #include "qapi/qmp/qstring.h"
38 #include "qapi/qobject-output-visitor.h"
39 #include "qapi/qapi-visit-block-core.h"
40 #include "sysemu/block-backend.h"
41 #include "sysemu/sysemu.h"
42 #include "qemu/notify.h"
43 #include "qemu/option.h"
44 #include "qemu/coroutine.h"
45 #include "block/qapi.h"
46 #include "qemu/timer.h"
47 #include "qemu/cutils.h"
51 #include <sys/ioctl.h>
52 #include <sys/queue.h>
62 #define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */
64 static QTAILQ_HEAD(, BlockDriverState
) graph_bdrv_states
=
65 QTAILQ_HEAD_INITIALIZER(graph_bdrv_states
);
67 static QTAILQ_HEAD(, BlockDriverState
) all_bdrv_states
=
68 QTAILQ_HEAD_INITIALIZER(all_bdrv_states
);
70 static QLIST_HEAD(, BlockDriver
) bdrv_drivers
=
71 QLIST_HEAD_INITIALIZER(bdrv_drivers
);
73 static BlockDriverState
*bdrv_open_inherit(const char *filename
,
74 const char *reference
,
75 QDict
*options
, int flags
,
76 BlockDriverState
*parent
,
77 const BdrvChildRole
*child_role
,
80 /* If non-zero, use only whitelisted block drivers */
81 static int use_bdrv_whitelist
;
84 static int is_windows_drive_prefix(const char *filename
)
86 return (((filename
[0] >= 'a' && filename
[0] <= 'z') ||
87 (filename
[0] >= 'A' && filename
[0] <= 'Z')) &&
91 int is_windows_drive(const char *filename
)
93 if (is_windows_drive_prefix(filename
) &&
96 if (strstart(filename
, "\\\\.\\", NULL
) ||
97 strstart(filename
, "//./", NULL
))
103 size_t bdrv_opt_mem_align(BlockDriverState
*bs
)
105 if (!bs
|| !bs
->drv
) {
106 /* page size or 4k (hdd sector size) should be on the safe side */
107 return MAX(4096, getpagesize());
110 return bs
->bl
.opt_mem_alignment
;
113 size_t bdrv_min_mem_align(BlockDriverState
*bs
)
115 if (!bs
|| !bs
->drv
) {
116 /* page size or 4k (hdd sector size) should be on the safe side */
117 return MAX(4096, getpagesize());
120 return bs
->bl
.min_mem_alignment
;
123 /* check if the path starts with "<protocol>:" */
124 int path_has_protocol(const char *path
)
129 if (is_windows_drive(path
) ||
130 is_windows_drive_prefix(path
)) {
133 p
= path
+ strcspn(path
, ":/\\");
135 p
= path
+ strcspn(path
, ":/");
141 int path_is_absolute(const char *path
)
144 /* specific case for names like: "\\.\d:" */
145 if (is_windows_drive(path
) || is_windows_drive_prefix(path
)) {
148 return (*path
== '/' || *path
== '\\');
150 return (*path
== '/');
154 /* if filename is absolute, just copy it to dest. Otherwise, build a
155 path to it by considering it is relative to base_path. URL are
157 void path_combine(char *dest
, int dest_size
,
158 const char *base_path
,
159 const char *filename
)
166 if (path_is_absolute(filename
)) {
167 pstrcpy(dest
, dest_size
, filename
);
169 const char *protocol_stripped
= NULL
;
171 if (path_has_protocol(base_path
)) {
172 protocol_stripped
= strchr(base_path
, ':');
173 if (protocol_stripped
) {
177 p
= protocol_stripped
?: base_path
;
179 p1
= strrchr(base_path
, '/');
183 p2
= strrchr(base_path
, '\\');
195 if (len
> dest_size
- 1)
197 memcpy(dest
, base_path
, len
);
199 pstrcat(dest
, dest_size
, filename
);
204 * Helper function for bdrv_parse_filename() implementations to remove optional
205 * protocol prefixes (especially "file:") from a filename and for putting the
206 * stripped filename into the options QDict if there is such a prefix.
208 void bdrv_parse_filename_strip_prefix(const char *filename
, const char *prefix
,
211 if (strstart(filename
, prefix
, &filename
)) {
212 /* Stripping the explicit protocol prefix may result in a protocol
213 * prefix being (wrongly) detected (if the filename contains a colon) */
214 if (path_has_protocol(filename
)) {
215 QString
*fat_filename
;
217 /* This means there is some colon before the first slash; therefore,
218 * this cannot be an absolute path */
219 assert(!path_is_absolute(filename
));
221 /* And we can thus fix the protocol detection issue by prefixing it
223 fat_filename
= qstring_from_str("./");
224 qstring_append(fat_filename
, filename
);
226 assert(!path_has_protocol(qstring_get_str(fat_filename
)));
228 qdict_put(options
, "filename", fat_filename
);
230 /* If no protocol prefix was detected, we can use the shortened
232 qdict_put_str(options
, "filename", filename
);
238 /* Returns whether the image file is opened as read-only. Note that this can
239 * return false and writing to the image file is still not possible because the
240 * image is inactivated. */
241 bool bdrv_is_read_only(BlockDriverState
*bs
)
243 return bs
->read_only
;
246 int bdrv_can_set_read_only(BlockDriverState
*bs
, bool read_only
,
247 bool ignore_allow_rdw
, Error
**errp
)
249 /* Do not set read_only if copy_on_read is enabled */
250 if (bs
->copy_on_read
&& read_only
) {
251 error_setg(errp
, "Can't set node '%s' to r/o with copy-on-read enabled",
252 bdrv_get_device_or_node_name(bs
));
256 /* Do not clear read_only if it is prohibited */
257 if (!read_only
&& !(bs
->open_flags
& BDRV_O_ALLOW_RDWR
) &&
260 error_setg(errp
, "Node '%s' is read only",
261 bdrv_get_device_or_node_name(bs
));
268 /* TODO Remove (deprecated since 2.11)
269 * Block drivers are not supposed to automatically change bs->read_only.
270 * Instead, they should just check whether they can provide what the user
271 * explicitly requested and error out if read-write is requested, but they can
272 * only provide read-only access. */
273 int bdrv_set_read_only(BlockDriverState
*bs
, bool read_only
, Error
**errp
)
277 ret
= bdrv_can_set_read_only(bs
, read_only
, false, errp
);
282 bs
->read_only
= read_only
;
286 void bdrv_get_full_backing_filename_from_filename(const char *backed
,
288 char *dest
, size_t sz
,
291 if (backing
[0] == '\0' || path_has_protocol(backing
) ||
292 path_is_absolute(backing
))
294 pstrcpy(dest
, sz
, backing
);
295 } else if (backed
[0] == '\0' || strstart(backed
, "json:", NULL
)) {
296 error_setg(errp
, "Cannot use relative backing file names for '%s'",
299 path_combine(dest
, sz
, backed
, backing
);
303 void bdrv_get_full_backing_filename(BlockDriverState
*bs
, char *dest
, size_t sz
,
306 char *backed
= bs
->exact_filename
[0] ? bs
->exact_filename
: bs
->filename
;
308 bdrv_get_full_backing_filename_from_filename(backed
, bs
->backing_file
,
312 void bdrv_register(BlockDriver
*bdrv
)
314 QLIST_INSERT_HEAD(&bdrv_drivers
, bdrv
, list
);
317 BlockDriverState
*bdrv_new(void)
319 BlockDriverState
*bs
;
322 bs
= g_new0(BlockDriverState
, 1);
323 QLIST_INIT(&bs
->dirty_bitmaps
);
324 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
325 QLIST_INIT(&bs
->op_blockers
[i
]);
327 notifier_with_return_list_init(&bs
->before_write_notifiers
);
328 qemu_co_mutex_init(&bs
->reqs_lock
);
329 qemu_mutex_init(&bs
->dirty_bitmap_mutex
);
331 bs
->aio_context
= qemu_get_aio_context();
333 qemu_co_queue_init(&bs
->flush_queue
);
335 QTAILQ_INSERT_TAIL(&all_bdrv_states
, bs
, bs_list
);
340 static BlockDriver
*bdrv_do_find_format(const char *format_name
)
344 QLIST_FOREACH(drv1
, &bdrv_drivers
, list
) {
345 if (!strcmp(drv1
->format_name
, format_name
)) {
353 BlockDriver
*bdrv_find_format(const char *format_name
)
358 drv1
= bdrv_do_find_format(format_name
);
363 /* The driver isn't registered, maybe we need to load a module */
364 for (i
= 0; i
< (int)ARRAY_SIZE(block_driver_modules
); ++i
) {
365 if (!strcmp(block_driver_modules
[i
].format_name
, format_name
)) {
366 block_module_load_one(block_driver_modules
[i
].library_name
);
371 return bdrv_do_find_format(format_name
);
374 int bdrv_is_whitelisted(BlockDriver
*drv
, bool read_only
)
376 static const char *whitelist_rw
[] = {
377 CONFIG_BDRV_RW_WHITELIST
379 static const char *whitelist_ro
[] = {
380 CONFIG_BDRV_RO_WHITELIST
384 if (!whitelist_rw
[0] && !whitelist_ro
[0]) {
385 return 1; /* no whitelist, anything goes */
388 for (p
= whitelist_rw
; *p
; p
++) {
389 if (!strcmp(drv
->format_name
, *p
)) {
394 for (p
= whitelist_ro
; *p
; p
++) {
395 if (!strcmp(drv
->format_name
, *p
)) {
403 bool bdrv_uses_whitelist(void)
405 return use_bdrv_whitelist
;
408 typedef struct CreateCo
{
416 static void coroutine_fn
bdrv_create_co_entry(void *opaque
)
418 Error
*local_err
= NULL
;
421 CreateCo
*cco
= opaque
;
424 ret
= cco
->drv
->bdrv_co_create_opts(cco
->filename
, cco
->opts
, &local_err
);
425 error_propagate(&cco
->err
, local_err
);
429 int bdrv_create(BlockDriver
*drv
, const char* filename
,
430 QemuOpts
*opts
, Error
**errp
)
437 .filename
= g_strdup(filename
),
443 if (!drv
->bdrv_co_create_opts
) {
444 error_setg(errp
, "Driver '%s' does not support image creation", drv
->format_name
);
449 if (qemu_in_coroutine()) {
450 /* Fast-path if already in coroutine context */
451 bdrv_create_co_entry(&cco
);
453 co
= qemu_coroutine_create(bdrv_create_co_entry
, &cco
);
454 qemu_coroutine_enter(co
);
455 while (cco
.ret
== NOT_DONE
) {
456 aio_poll(qemu_get_aio_context(), true);
463 error_propagate(errp
, cco
.err
);
465 error_setg_errno(errp
, -ret
, "Could not create image");
470 g_free(cco
.filename
);
474 int bdrv_create_file(const char *filename
, QemuOpts
*opts
, Error
**errp
)
477 Error
*local_err
= NULL
;
480 drv
= bdrv_find_protocol(filename
, true, errp
);
485 ret
= bdrv_create(drv
, filename
, opts
, &local_err
);
486 error_propagate(errp
, local_err
);
491 * Try to get @bs's logical and physical block size.
492 * On success, store them in @bsz struct and return 0.
493 * On failure return -errno.
494 * @bs must not be empty.
496 int bdrv_probe_blocksizes(BlockDriverState
*bs
, BlockSizes
*bsz
)
498 BlockDriver
*drv
= bs
->drv
;
500 if (drv
&& drv
->bdrv_probe_blocksizes
) {
501 return drv
->bdrv_probe_blocksizes(bs
, bsz
);
502 } else if (drv
&& drv
->is_filter
&& bs
->file
) {
503 return bdrv_probe_blocksizes(bs
->file
->bs
, bsz
);
510 * Try to get @bs's geometry (cyls, heads, sectors).
511 * On success, store them in @geo struct and return 0.
512 * On failure return -errno.
513 * @bs must not be empty.
515 int bdrv_probe_geometry(BlockDriverState
*bs
, HDGeometry
*geo
)
517 BlockDriver
*drv
= bs
->drv
;
519 if (drv
&& drv
->bdrv_probe_geometry
) {
520 return drv
->bdrv_probe_geometry(bs
, geo
);
521 } else if (drv
&& drv
->is_filter
&& bs
->file
) {
522 return bdrv_probe_geometry(bs
->file
->bs
, geo
);
529 * Create a uniquely-named empty temporary file.
530 * Return 0 upon success, otherwise a negative errno value.
532 int get_tmp_filename(char *filename
, int size
)
535 char temp_dir
[MAX_PATH
];
536 /* GetTempFileName requires that its output buffer (4th param)
537 have length MAX_PATH or greater. */
538 assert(size
>= MAX_PATH
);
539 return (GetTempPath(MAX_PATH
, temp_dir
)
540 && GetTempFileName(temp_dir
, "qem", 0, filename
)
541 ? 0 : -GetLastError());
545 tmpdir
= getenv("TMPDIR");
549 if (snprintf(filename
, size
, "%s/vl.XXXXXX", tmpdir
) >= size
) {
552 fd
= mkstemp(filename
);
556 if (close(fd
) != 0) {
565 * Detect host devices. By convention, /dev/cdrom[N] is always
566 * recognized as a host CDROM.
568 static BlockDriver
*find_hdev_driver(const char *filename
)
570 int score_max
= 0, score
;
571 BlockDriver
*drv
= NULL
, *d
;
573 QLIST_FOREACH(d
, &bdrv_drivers
, list
) {
574 if (d
->bdrv_probe_device
) {
575 score
= d
->bdrv_probe_device(filename
);
576 if (score
> score_max
) {
586 static BlockDriver
*bdrv_do_find_protocol(const char *protocol
)
590 QLIST_FOREACH(drv1
, &bdrv_drivers
, list
) {
591 if (drv1
->protocol_name
&& !strcmp(drv1
->protocol_name
, protocol
)) {
599 BlockDriver
*bdrv_find_protocol(const char *filename
,
600 bool allow_protocol_prefix
,
609 /* TODO Drivers without bdrv_file_open must be specified explicitly */
612 * XXX(hch): we really should not let host device detection
613 * override an explicit protocol specification, but moving this
614 * later breaks access to device names with colons in them.
615 * Thanks to the brain-dead persistent naming schemes on udev-
616 * based Linux systems those actually are quite common.
618 drv1
= find_hdev_driver(filename
);
623 if (!path_has_protocol(filename
) || !allow_protocol_prefix
) {
627 p
= strchr(filename
, ':');
630 if (len
> sizeof(protocol
) - 1)
631 len
= sizeof(protocol
) - 1;
632 memcpy(protocol
, filename
, len
);
633 protocol
[len
] = '\0';
635 drv1
= bdrv_do_find_protocol(protocol
);
640 for (i
= 0; i
< (int)ARRAY_SIZE(block_driver_modules
); ++i
) {
641 if (block_driver_modules
[i
].protocol_name
&&
642 !strcmp(block_driver_modules
[i
].protocol_name
, protocol
)) {
643 block_module_load_one(block_driver_modules
[i
].library_name
);
648 drv1
= bdrv_do_find_protocol(protocol
);
650 error_setg(errp
, "Unknown protocol '%s'", protocol
);
656 * Guess image format by probing its contents.
657 * This is not a good idea when your image is raw (CVE-2008-2004), but
658 * we do it anyway for backward compatibility.
660 * @buf contains the image's first @buf_size bytes.
661 * @buf_size is the buffer size in bytes (generally BLOCK_PROBE_BUF_SIZE,
662 * but can be smaller if the image file is smaller)
663 * @filename is its filename.
665 * For all block drivers, call the bdrv_probe() method to get its
667 * Return the first block driver with the highest probing score.
669 BlockDriver
*bdrv_probe_all(const uint8_t *buf
, int buf_size
,
670 const char *filename
)
672 int score_max
= 0, score
;
673 BlockDriver
*drv
= NULL
, *d
;
675 QLIST_FOREACH(d
, &bdrv_drivers
, list
) {
677 score
= d
->bdrv_probe(buf
, buf_size
, filename
);
678 if (score
> score_max
) {
688 static int find_image_format(BlockBackend
*file
, const char *filename
,
689 BlockDriver
**pdrv
, Error
**errp
)
692 uint8_t buf
[BLOCK_PROBE_BUF_SIZE
];
695 /* Return the raw BlockDriver * to scsi-generic devices or empty drives */
696 if (blk_is_sg(file
) || !blk_is_inserted(file
) || blk_getlength(file
) == 0) {
701 ret
= blk_pread(file
, 0, buf
, sizeof(buf
));
703 error_setg_errno(errp
, -ret
, "Could not read image for determining its "
709 drv
= bdrv_probe_all(buf
, ret
, filename
);
711 error_setg(errp
, "Could not determine image format: No compatible "
720 * Set the current 'total_sectors' value
721 * Return 0 on success, -errno on error.
723 static int refresh_total_sectors(BlockDriverState
*bs
, int64_t hint
)
725 BlockDriver
*drv
= bs
->drv
;
731 /* Do not attempt drv->bdrv_getlength() on scsi-generic devices */
735 /* query actual device if possible, otherwise just trust the hint */
736 if (drv
->bdrv_getlength
) {
737 int64_t length
= drv
->bdrv_getlength(bs
);
741 hint
= DIV_ROUND_UP(length
, BDRV_SECTOR_SIZE
);
744 bs
->total_sectors
= hint
;
749 * Combines a QDict of new block driver @options with any missing options taken
750 * from @old_options, so that leaving out an option defaults to its old value.
752 static void bdrv_join_options(BlockDriverState
*bs
, QDict
*options
,
755 if (bs
->drv
&& bs
->drv
->bdrv_join_options
) {
756 bs
->drv
->bdrv_join_options(options
, old_options
);
758 qdict_join(options
, old_options
, false);
763 * Set open flags for a given discard mode
765 * Return 0 on success, -1 if the discard mode was invalid.
767 int bdrv_parse_discard_flags(const char *mode
, int *flags
)
769 *flags
&= ~BDRV_O_UNMAP
;
771 if (!strcmp(mode
, "off") || !strcmp(mode
, "ignore")) {
773 } else if (!strcmp(mode
, "on") || !strcmp(mode
, "unmap")) {
774 *flags
|= BDRV_O_UNMAP
;
783 * Set open flags for a given cache mode
785 * Return 0 on success, -1 if the cache mode was invalid.
787 int bdrv_parse_cache_mode(const char *mode
, int *flags
, bool *writethrough
)
789 *flags
&= ~BDRV_O_CACHE_MASK
;
791 if (!strcmp(mode
, "off") || !strcmp(mode
, "none")) {
792 *writethrough
= false;
793 *flags
|= BDRV_O_NOCACHE
;
794 } else if (!strcmp(mode
, "directsync")) {
795 *writethrough
= true;
796 *flags
|= BDRV_O_NOCACHE
;
797 } else if (!strcmp(mode
, "writeback")) {
798 *writethrough
= false;
799 } else if (!strcmp(mode
, "unsafe")) {
800 *writethrough
= false;
801 *flags
|= BDRV_O_NO_FLUSH
;
802 } else if (!strcmp(mode
, "writethrough")) {
803 *writethrough
= true;
811 static char *bdrv_child_get_parent_desc(BdrvChild
*c
)
813 BlockDriverState
*parent
= c
->opaque
;
814 return g_strdup(bdrv_get_device_or_node_name(parent
));
817 static void bdrv_child_cb_drained_begin(BdrvChild
*child
)
819 BlockDriverState
*bs
= child
->opaque
;
820 bdrv_drained_begin(bs
);
823 static void bdrv_child_cb_drained_end(BdrvChild
*child
)
825 BlockDriverState
*bs
= child
->opaque
;
826 bdrv_drained_end(bs
);
829 static void bdrv_child_cb_attach(BdrvChild
*child
)
831 BlockDriverState
*bs
= child
->opaque
;
832 bdrv_apply_subtree_drain(child
, bs
);
835 static void bdrv_child_cb_detach(BdrvChild
*child
)
837 BlockDriverState
*bs
= child
->opaque
;
838 bdrv_unapply_subtree_drain(child
, bs
);
841 static int bdrv_child_cb_inactivate(BdrvChild
*child
)
843 BlockDriverState
*bs
= child
->opaque
;
844 assert(bs
->open_flags
& BDRV_O_INACTIVE
);
849 * Returns the options and flags that a temporary snapshot should get, based on
850 * the originally requested flags (the originally requested image will have
851 * flags like a backing file)
853 static void bdrv_temp_snapshot_options(int *child_flags
, QDict
*child_options
,
854 int parent_flags
, QDict
*parent_options
)
856 *child_flags
= (parent_flags
& ~BDRV_O_SNAPSHOT
) | BDRV_O_TEMPORARY
;
858 /* For temporary files, unconditional cache=unsafe is fine */
859 qdict_set_default_str(child_options
, BDRV_OPT_CACHE_DIRECT
, "off");
860 qdict_set_default_str(child_options
, BDRV_OPT_CACHE_NO_FLUSH
, "on");
862 /* Copy the read-only option from the parent */
863 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_READ_ONLY
);
865 /* aio=native doesn't work for cache.direct=off, so disable it for the
866 * temporary snapshot */
867 *child_flags
&= ~BDRV_O_NATIVE_AIO
;
871 * Returns the options and flags that bs->file should get if a protocol driver
872 * is expected, based on the given options and flags for the parent BDS
874 static void bdrv_inherited_options(int *child_flags
, QDict
*child_options
,
875 int parent_flags
, QDict
*parent_options
)
877 int flags
= parent_flags
;
879 /* Enable protocol handling, disable format probing for bs->file */
880 flags
|= BDRV_O_PROTOCOL
;
882 /* If the cache mode isn't explicitly set, inherit direct and no-flush from
884 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_CACHE_DIRECT
);
885 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_CACHE_NO_FLUSH
);
886 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_FORCE_SHARE
);
888 /* Inherit the read-only option from the parent if it's not set */
889 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_READ_ONLY
);
891 /* Our block drivers take care to send flushes and respect unmap policy,
892 * so we can default to enable both on lower layers regardless of the
893 * corresponding parent options. */
894 qdict_set_default_str(child_options
, BDRV_OPT_DISCARD
, "unmap");
896 /* Clear flags that only apply to the top layer */
897 flags
&= ~(BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
| BDRV_O_COPY_ON_READ
|
900 *child_flags
= flags
;
903 const BdrvChildRole child_file
= {
904 .get_parent_desc
= bdrv_child_get_parent_desc
,
905 .inherit_options
= bdrv_inherited_options
,
906 .drained_begin
= bdrv_child_cb_drained_begin
,
907 .drained_end
= bdrv_child_cb_drained_end
,
908 .attach
= bdrv_child_cb_attach
,
909 .detach
= bdrv_child_cb_detach
,
910 .inactivate
= bdrv_child_cb_inactivate
,
914 * Returns the options and flags that bs->file should get if the use of formats
915 * (and not only protocols) is permitted for it, based on the given options and
916 * flags for the parent BDS
918 static void bdrv_inherited_fmt_options(int *child_flags
, QDict
*child_options
,
919 int parent_flags
, QDict
*parent_options
)
921 child_file
.inherit_options(child_flags
, child_options
,
922 parent_flags
, parent_options
);
924 *child_flags
&= ~(BDRV_O_PROTOCOL
| BDRV_O_NO_IO
);
927 const BdrvChildRole child_format
= {
928 .get_parent_desc
= bdrv_child_get_parent_desc
,
929 .inherit_options
= bdrv_inherited_fmt_options
,
930 .drained_begin
= bdrv_child_cb_drained_begin
,
931 .drained_end
= bdrv_child_cb_drained_end
,
932 .attach
= bdrv_child_cb_attach
,
933 .detach
= bdrv_child_cb_detach
,
934 .inactivate
= bdrv_child_cb_inactivate
,
937 static void bdrv_backing_attach(BdrvChild
*c
)
939 BlockDriverState
*parent
= c
->opaque
;
940 BlockDriverState
*backing_hd
= c
->bs
;
942 assert(!parent
->backing_blocker
);
943 error_setg(&parent
->backing_blocker
,
944 "node is used as backing hd of '%s'",
945 bdrv_get_device_or_node_name(parent
));
947 parent
->open_flags
&= ~BDRV_O_NO_BACKING
;
948 pstrcpy(parent
->backing_file
, sizeof(parent
->backing_file
),
949 backing_hd
->filename
);
950 pstrcpy(parent
->backing_format
, sizeof(parent
->backing_format
),
951 backing_hd
->drv
? backing_hd
->drv
->format_name
: "");
953 bdrv_op_block_all(backing_hd
, parent
->backing_blocker
);
954 /* Otherwise we won't be able to commit or stream */
955 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_COMMIT_TARGET
,
956 parent
->backing_blocker
);
957 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_STREAM
,
958 parent
->backing_blocker
);
960 * We do backup in 3 ways:
962 * The target bs is new opened, and the source is top BDS
964 * Both the source and the target are top BDSes.
965 * 3. internal backup(used for block replication)
966 * Both the source and the target are backing file
968 * In case 1 and 2, neither the source nor the target is the backing file.
969 * In case 3, we will block the top BDS, so there is only one block job
970 * for the top BDS and its backing chain.
972 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_BACKUP_SOURCE
,
973 parent
->backing_blocker
);
974 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_BACKUP_TARGET
,
975 parent
->backing_blocker
);
977 bdrv_child_cb_attach(c
);
980 static void bdrv_backing_detach(BdrvChild
*c
)
982 BlockDriverState
*parent
= c
->opaque
;
984 assert(parent
->backing_blocker
);
985 bdrv_op_unblock_all(c
->bs
, parent
->backing_blocker
);
986 error_free(parent
->backing_blocker
);
987 parent
->backing_blocker
= NULL
;
989 bdrv_child_cb_detach(c
);
993 * Returns the options and flags that bs->backing should get, based on the
994 * given options and flags for the parent BDS
996 static void bdrv_backing_options(int *child_flags
, QDict
*child_options
,
997 int parent_flags
, QDict
*parent_options
)
999 int flags
= parent_flags
;
1001 /* The cache mode is inherited unmodified for backing files; except WCE,
1002 * which is only applied on the top level (BlockBackend) */
1003 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_CACHE_DIRECT
);
1004 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_CACHE_NO_FLUSH
);
1005 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_FORCE_SHARE
);
1007 /* backing files always opened read-only */
1008 qdict_set_default_str(child_options
, BDRV_OPT_READ_ONLY
, "on");
1009 flags
&= ~BDRV_O_COPY_ON_READ
;
1011 /* snapshot=on is handled on the top layer */
1012 flags
&= ~(BDRV_O_SNAPSHOT
| BDRV_O_TEMPORARY
);
1014 *child_flags
= flags
;
1017 static int bdrv_backing_update_filename(BdrvChild
*c
, BlockDriverState
*base
,
1018 const char *filename
, Error
**errp
)
1020 BlockDriverState
*parent
= c
->opaque
;
1021 int orig_flags
= bdrv_get_flags(parent
);
1024 if (!(orig_flags
& BDRV_O_RDWR
)) {
1025 ret
= bdrv_reopen(parent
, orig_flags
| BDRV_O_RDWR
, errp
);
1031 ret
= bdrv_change_backing_file(parent
, filename
,
1032 base
->drv
? base
->drv
->format_name
: "");
1034 error_setg_errno(errp
, -ret
, "Could not update backing file link");
1037 if (!(orig_flags
& BDRV_O_RDWR
)) {
1038 bdrv_reopen(parent
, orig_flags
, NULL
);
1044 const BdrvChildRole child_backing
= {
1045 .get_parent_desc
= bdrv_child_get_parent_desc
,
1046 .attach
= bdrv_backing_attach
,
1047 .detach
= bdrv_backing_detach
,
1048 .inherit_options
= bdrv_backing_options
,
1049 .drained_begin
= bdrv_child_cb_drained_begin
,
1050 .drained_end
= bdrv_child_cb_drained_end
,
1051 .inactivate
= bdrv_child_cb_inactivate
,
1052 .update_filename
= bdrv_backing_update_filename
,
1055 static int bdrv_open_flags(BlockDriverState
*bs
, int flags
)
1057 int open_flags
= flags
;
1060 * Clear flags that are internal to the block layer before opening the
1063 open_flags
&= ~(BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
| BDRV_O_PROTOCOL
);
1066 * Snapshots should be writable.
1068 if (flags
& BDRV_O_TEMPORARY
) {
1069 open_flags
|= BDRV_O_RDWR
;
1075 static void update_flags_from_options(int *flags
, QemuOpts
*opts
)
1077 *flags
&= ~BDRV_O_CACHE_MASK
;
1079 assert(qemu_opt_find(opts
, BDRV_OPT_CACHE_NO_FLUSH
));
1080 if (qemu_opt_get_bool(opts
, BDRV_OPT_CACHE_NO_FLUSH
, false)) {
1081 *flags
|= BDRV_O_NO_FLUSH
;
1084 assert(qemu_opt_find(opts
, BDRV_OPT_CACHE_DIRECT
));
1085 if (qemu_opt_get_bool(opts
, BDRV_OPT_CACHE_DIRECT
, false)) {
1086 *flags
|= BDRV_O_NOCACHE
;
1089 *flags
&= ~BDRV_O_RDWR
;
1091 assert(qemu_opt_find(opts
, BDRV_OPT_READ_ONLY
));
1092 if (!qemu_opt_get_bool(opts
, BDRV_OPT_READ_ONLY
, false)) {
1093 *flags
|= BDRV_O_RDWR
;
1098 static void update_options_from_flags(QDict
*options
, int flags
)
1100 if (!qdict_haskey(options
, BDRV_OPT_CACHE_DIRECT
)) {
1101 qdict_put_bool(options
, BDRV_OPT_CACHE_DIRECT
, flags
& BDRV_O_NOCACHE
);
1103 if (!qdict_haskey(options
, BDRV_OPT_CACHE_NO_FLUSH
)) {
1104 qdict_put_bool(options
, BDRV_OPT_CACHE_NO_FLUSH
,
1105 flags
& BDRV_O_NO_FLUSH
);
1107 if (!qdict_haskey(options
, BDRV_OPT_READ_ONLY
)) {
1108 qdict_put_bool(options
, BDRV_OPT_READ_ONLY
, !(flags
& BDRV_O_RDWR
));
1112 static void bdrv_assign_node_name(BlockDriverState
*bs
,
1113 const char *node_name
,
1116 char *gen_node_name
= NULL
;
1119 node_name
= gen_node_name
= id_generate(ID_BLOCK
);
1120 } else if (!id_wellformed(node_name
)) {
1122 * Check for empty string or invalid characters, but not if it is
1123 * generated (generated names use characters not available to the user)
1125 error_setg(errp
, "Invalid node name");
1129 /* takes care of avoiding namespaces collisions */
1130 if (blk_by_name(node_name
)) {
1131 error_setg(errp
, "node-name=%s is conflicting with a device id",
1136 /* takes care of avoiding duplicates node names */
1137 if (bdrv_find_node(node_name
)) {
1138 error_setg(errp
, "Duplicate node name");
1142 /* copy node name into the bs and insert it into the graph list */
1143 pstrcpy(bs
->node_name
, sizeof(bs
->node_name
), node_name
);
1144 QTAILQ_INSERT_TAIL(&graph_bdrv_states
, bs
, node_list
);
1146 g_free(gen_node_name
);
1149 static int bdrv_open_driver(BlockDriverState
*bs
, BlockDriver
*drv
,
1150 const char *node_name
, QDict
*options
,
1151 int open_flags
, Error
**errp
)
1153 Error
*local_err
= NULL
;
1156 bdrv_assign_node_name(bs
, node_name
, &local_err
);
1158 error_propagate(errp
, local_err
);
1163 bs
->read_only
= !(bs
->open_flags
& BDRV_O_RDWR
);
1164 bs
->opaque
= g_malloc0(drv
->instance_size
);
1166 if (drv
->bdrv_file_open
) {
1167 assert(!drv
->bdrv_needs_filename
|| bs
->filename
[0]);
1168 ret
= drv
->bdrv_file_open(bs
, options
, open_flags
, &local_err
);
1169 } else if (drv
->bdrv_open
) {
1170 ret
= drv
->bdrv_open(bs
, options
, open_flags
, &local_err
);
1177 error_propagate(errp
, local_err
);
1178 } else if (bs
->filename
[0]) {
1179 error_setg_errno(errp
, -ret
, "Could not open '%s'", bs
->filename
);
1181 error_setg_errno(errp
, -ret
, "Could not open image");
1186 ret
= refresh_total_sectors(bs
, bs
->total_sectors
);
1188 error_setg_errno(errp
, -ret
, "Could not refresh total sector count");
1192 bdrv_refresh_limits(bs
, &local_err
);
1194 error_propagate(errp
, local_err
);
1198 assert(bdrv_opt_mem_align(bs
) != 0);
1199 assert(bdrv_min_mem_align(bs
) != 0);
1200 assert(is_power_of_2(bs
->bl
.request_alignment
));
1205 if (bs
->file
!= NULL
) {
1206 bdrv_unref_child(bs
, bs
->file
);
1214 BlockDriverState
*bdrv_new_open_driver(BlockDriver
*drv
, const char *node_name
,
1215 int flags
, Error
**errp
)
1217 BlockDriverState
*bs
;
1221 bs
->open_flags
= flags
;
1222 bs
->explicit_options
= qdict_new();
1223 bs
->options
= qdict_new();
1226 update_options_from_flags(bs
->options
, flags
);
1228 ret
= bdrv_open_driver(bs
, drv
, node_name
, bs
->options
, flags
, errp
);
1230 QDECREF(bs
->explicit_options
);
1231 bs
->explicit_options
= NULL
;
1232 QDECREF(bs
->options
);
1241 QemuOptsList bdrv_runtime_opts
= {
1242 .name
= "bdrv_common",
1243 .head
= QTAILQ_HEAD_INITIALIZER(bdrv_runtime_opts
.head
),
1246 .name
= "node-name",
1247 .type
= QEMU_OPT_STRING
,
1248 .help
= "Node name of the block device node",
1252 .type
= QEMU_OPT_STRING
,
1253 .help
= "Block driver to use for the node",
1256 .name
= BDRV_OPT_CACHE_DIRECT
,
1257 .type
= QEMU_OPT_BOOL
,
1258 .help
= "Bypass software writeback cache on the host",
1261 .name
= BDRV_OPT_CACHE_NO_FLUSH
,
1262 .type
= QEMU_OPT_BOOL
,
1263 .help
= "Ignore flush requests",
1266 .name
= BDRV_OPT_READ_ONLY
,
1267 .type
= QEMU_OPT_BOOL
,
1268 .help
= "Node is opened in read-only mode",
1271 .name
= "detect-zeroes",
1272 .type
= QEMU_OPT_STRING
,
1273 .help
= "try to optimize zero writes (off, on, unmap)",
1277 .type
= QEMU_OPT_STRING
,
1278 .help
= "discard operation (ignore/off, unmap/on)",
1281 .name
= BDRV_OPT_FORCE_SHARE
,
1282 .type
= QEMU_OPT_BOOL
,
1283 .help
= "always accept other writers (default: off)",
1285 { /* end of list */ }
1290 * Common part for opening disk images and files
1292 * Removes all processed options from *options.
1294 static int bdrv_open_common(BlockDriverState
*bs
, BlockBackend
*file
,
1295 QDict
*options
, Error
**errp
)
1297 int ret
, open_flags
;
1298 const char *filename
;
1299 const char *driver_name
= NULL
;
1300 const char *node_name
= NULL
;
1301 const char *discard
;
1302 const char *detect_zeroes
;
1305 Error
*local_err
= NULL
;
1307 assert(bs
->file
== NULL
);
1308 assert(options
!= NULL
&& bs
->options
!= options
);
1310 opts
= qemu_opts_create(&bdrv_runtime_opts
, NULL
, 0, &error_abort
);
1311 qemu_opts_absorb_qdict(opts
, options
, &local_err
);
1313 error_propagate(errp
, local_err
);
1318 update_flags_from_options(&bs
->open_flags
, opts
);
1320 driver_name
= qemu_opt_get(opts
, "driver");
1321 drv
= bdrv_find_format(driver_name
);
1322 assert(drv
!= NULL
);
1324 bs
->force_share
= qemu_opt_get_bool(opts
, BDRV_OPT_FORCE_SHARE
, false);
1326 if (bs
->force_share
&& (bs
->open_flags
& BDRV_O_RDWR
)) {
1328 BDRV_OPT_FORCE_SHARE
1329 "=on can only be used with read-only images");
1335 filename
= blk_bs(file
)->filename
;
1338 * Caution: while qdict_get_try_str() is fine, getting
1339 * non-string types would require more care. When @options
1340 * come from -blockdev or blockdev_add, its members are typed
1341 * according to the QAPI schema, but when they come from
1342 * -drive, they're all QString.
1344 filename
= qdict_get_try_str(options
, "filename");
1347 if (drv
->bdrv_needs_filename
&& (!filename
|| !filename
[0])) {
1348 error_setg(errp
, "The '%s' block driver requires a file name",
1354 trace_bdrv_open_common(bs
, filename
?: "", bs
->open_flags
,
1357 bs
->read_only
= !(bs
->open_flags
& BDRV_O_RDWR
);
1359 if (use_bdrv_whitelist
&& !bdrv_is_whitelisted(drv
, bs
->read_only
)) {
1361 !bs
->read_only
&& bdrv_is_whitelisted(drv
, true)
1362 ? "Driver '%s' can only be used for read-only devices"
1363 : "Driver '%s' is not whitelisted",
1369 /* bdrv_new() and bdrv_close() make it so */
1370 assert(atomic_read(&bs
->copy_on_read
) == 0);
1372 if (bs
->open_flags
& BDRV_O_COPY_ON_READ
) {
1373 if (!bs
->read_only
) {
1374 bdrv_enable_copy_on_read(bs
);
1376 error_setg(errp
, "Can't use copy-on-read on read-only device");
1382 discard
= qemu_opt_get(opts
, "discard");
1383 if (discard
!= NULL
) {
1384 if (bdrv_parse_discard_flags(discard
, &bs
->open_flags
) != 0) {
1385 error_setg(errp
, "Invalid discard option");
1391 detect_zeroes
= qemu_opt_get(opts
, "detect-zeroes");
1392 if (detect_zeroes
) {
1393 BlockdevDetectZeroesOptions value
=
1394 qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup
,
1396 BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF
,
1399 error_propagate(errp
, local_err
);
1404 if (value
== BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP
&&
1405 !(bs
->open_flags
& BDRV_O_UNMAP
))
1407 error_setg(errp
, "setting detect-zeroes to unmap is not allowed "
1408 "without setting discard operation to unmap");
1413 bs
->detect_zeroes
= value
;
1416 if (filename
!= NULL
) {
1417 pstrcpy(bs
->filename
, sizeof(bs
->filename
), filename
);
1419 bs
->filename
[0] = '\0';
1421 pstrcpy(bs
->exact_filename
, sizeof(bs
->exact_filename
), bs
->filename
);
1423 /* Open the image, either directly or using a protocol */
1424 open_flags
= bdrv_open_flags(bs
, bs
->open_flags
);
1425 node_name
= qemu_opt_get(opts
, "node-name");
1427 assert(!drv
->bdrv_file_open
|| file
== NULL
);
1428 ret
= bdrv_open_driver(bs
, drv
, node_name
, options
, open_flags
, errp
);
1433 qemu_opts_del(opts
);
1437 qemu_opts_del(opts
);
1441 static QDict
*parse_json_filename(const char *filename
, Error
**errp
)
1443 QObject
*options_obj
;
1447 ret
= strstart(filename
, "json:", &filename
);
1450 options_obj
= qobject_from_json(filename
, errp
);
1452 /* Work around qobject_from_json() lossage TODO fix that */
1453 if (errp
&& !*errp
) {
1454 error_setg(errp
, "Could not parse the JSON options");
1457 error_prepend(errp
, "Could not parse the JSON options: ");
1461 options
= qobject_to(QDict
, options_obj
);
1463 qobject_decref(options_obj
);
1464 error_setg(errp
, "Invalid JSON object given");
1468 qdict_flatten(options
);
1473 static void parse_json_protocol(QDict
*options
, const char **pfilename
,
1476 QDict
*json_options
;
1477 Error
*local_err
= NULL
;
1479 /* Parse json: pseudo-protocol */
1480 if (!*pfilename
|| !g_str_has_prefix(*pfilename
, "json:")) {
1484 json_options
= parse_json_filename(*pfilename
, &local_err
);
1486 error_propagate(errp
, local_err
);
1490 /* Options given in the filename have lower priority than options
1491 * specified directly */
1492 qdict_join(options
, json_options
, false);
1493 QDECREF(json_options
);
1498 * Fills in default options for opening images and converts the legacy
1499 * filename/flags pair to option QDict entries.
1500 * The BDRV_O_PROTOCOL flag in *flags will be set or cleared accordingly if a
1501 * block driver has been specified explicitly.
1503 static int bdrv_fill_options(QDict
**options
, const char *filename
,
1504 int *flags
, Error
**errp
)
1506 const char *drvname
;
1507 bool protocol
= *flags
& BDRV_O_PROTOCOL
;
1508 bool parse_filename
= false;
1509 BlockDriver
*drv
= NULL
;
1510 Error
*local_err
= NULL
;
1513 * Caution: while qdict_get_try_str() is fine, getting non-string
1514 * types would require more care. When @options come from
1515 * -blockdev or blockdev_add, its members are typed according to
1516 * the QAPI schema, but when they come from -drive, they're all
1519 drvname
= qdict_get_try_str(*options
, "driver");
1521 drv
= bdrv_find_format(drvname
);
1523 error_setg(errp
, "Unknown driver '%s'", drvname
);
1526 /* If the user has explicitly specified the driver, this choice should
1527 * override the BDRV_O_PROTOCOL flag */
1528 protocol
= drv
->bdrv_file_open
;
1532 *flags
|= BDRV_O_PROTOCOL
;
1534 *flags
&= ~BDRV_O_PROTOCOL
;
1537 /* Translate cache options from flags into options */
1538 update_options_from_flags(*options
, *flags
);
1540 /* Fetch the file name from the options QDict if necessary */
1541 if (protocol
&& filename
) {
1542 if (!qdict_haskey(*options
, "filename")) {
1543 qdict_put_str(*options
, "filename", filename
);
1544 parse_filename
= true;
1546 error_setg(errp
, "Can't specify 'file' and 'filename' options at "
1552 /* Find the right block driver */
1553 /* See cautionary note on accessing @options above */
1554 filename
= qdict_get_try_str(*options
, "filename");
1556 if (!drvname
&& protocol
) {
1558 drv
= bdrv_find_protocol(filename
, parse_filename
, errp
);
1563 drvname
= drv
->format_name
;
1564 qdict_put_str(*options
, "driver", drvname
);
1566 error_setg(errp
, "Must specify either driver or file");
1571 assert(drv
|| !protocol
);
1573 /* Driver-specific filename parsing */
1574 if (drv
&& drv
->bdrv_parse_filename
&& parse_filename
) {
1575 drv
->bdrv_parse_filename(filename
, *options
, &local_err
);
1577 error_propagate(errp
, local_err
);
1581 if (!drv
->bdrv_needs_filename
) {
1582 qdict_del(*options
, "filename");
1589 static int bdrv_child_check_perm(BdrvChild
*c
, BlockReopenQueue
*q
,
1590 uint64_t perm
, uint64_t shared
,
1591 GSList
*ignore_children
, Error
**errp
);
1592 static void bdrv_child_abort_perm_update(BdrvChild
*c
);
1593 static void bdrv_child_set_perm(BdrvChild
*c
, uint64_t perm
, uint64_t shared
);
1595 typedef struct BlockReopenQueueEntry
{
1597 BDRVReopenState state
;
1598 QSIMPLEQ_ENTRY(BlockReopenQueueEntry
) entry
;
1599 } BlockReopenQueueEntry
;
1602 * Return the flags that @bs will have after the reopens in @q have
1603 * successfully completed. If @q is NULL (or @bs is not contained in @q),
1604 * return the current flags.
1606 static int bdrv_reopen_get_flags(BlockReopenQueue
*q
, BlockDriverState
*bs
)
1608 BlockReopenQueueEntry
*entry
;
1611 QSIMPLEQ_FOREACH(entry
, q
, entry
) {
1612 if (entry
->state
.bs
== bs
) {
1613 return entry
->state
.flags
;
1618 return bs
->open_flags
;
1621 /* Returns whether the image file can be written to after the reopen queue @q
1622 * has been successfully applied, or right now if @q is NULL. */
1623 static bool bdrv_is_writable(BlockDriverState
*bs
, BlockReopenQueue
*q
)
1625 int flags
= bdrv_reopen_get_flags(q
, bs
);
1627 return (flags
& (BDRV_O_RDWR
| BDRV_O_INACTIVE
)) == BDRV_O_RDWR
;
1630 static void bdrv_child_perm(BlockDriverState
*bs
, BlockDriverState
*child_bs
,
1631 BdrvChild
*c
, const BdrvChildRole
*role
,
1632 BlockReopenQueue
*reopen_queue
,
1633 uint64_t parent_perm
, uint64_t parent_shared
,
1634 uint64_t *nperm
, uint64_t *nshared
)
1636 if (bs
->drv
&& bs
->drv
->bdrv_child_perm
) {
1637 bs
->drv
->bdrv_child_perm(bs
, c
, role
, reopen_queue
,
1638 parent_perm
, parent_shared
,
1641 /* TODO Take force_share from reopen_queue */
1642 if (child_bs
&& child_bs
->force_share
) {
1643 *nshared
= BLK_PERM_ALL
;
1648 * Check whether permissions on this node can be changed in a way that
1649 * @cumulative_perms and @cumulative_shared_perms are the new cumulative
1650 * permissions of all its parents. This involves checking whether all necessary
1651 * permission changes to child nodes can be performed.
1653 * A call to this function must always be followed by a call to bdrv_set_perm()
1654 * or bdrv_abort_perm_update().
1656 static int bdrv_check_perm(BlockDriverState
*bs
, BlockReopenQueue
*q
,
1657 uint64_t cumulative_perms
,
1658 uint64_t cumulative_shared_perms
,
1659 GSList
*ignore_children
, Error
**errp
)
1661 BlockDriver
*drv
= bs
->drv
;
1665 /* Write permissions never work with read-only images */
1666 if ((cumulative_perms
& (BLK_PERM_WRITE
| BLK_PERM_WRITE_UNCHANGED
)) &&
1667 !bdrv_is_writable(bs
, q
))
1669 error_setg(errp
, "Block node is read-only");
1673 /* Check this node */
1678 if (drv
->bdrv_check_perm
) {
1679 return drv
->bdrv_check_perm(bs
, cumulative_perms
,
1680 cumulative_shared_perms
, errp
);
1683 /* Drivers that never have children can omit .bdrv_child_perm() */
1684 if (!drv
->bdrv_child_perm
) {
1685 assert(QLIST_EMPTY(&bs
->children
));
1689 /* Check all children */
1690 QLIST_FOREACH(c
, &bs
->children
, next
) {
1691 uint64_t cur_perm
, cur_shared
;
1692 bdrv_child_perm(bs
, c
->bs
, c
, c
->role
, q
,
1693 cumulative_perms
, cumulative_shared_perms
,
1694 &cur_perm
, &cur_shared
);
1695 ret
= bdrv_child_check_perm(c
, q
, cur_perm
, cur_shared
,
1696 ignore_children
, errp
);
1706 * Notifies drivers that after a previous bdrv_check_perm() call, the
1707 * permission update is not performed and any preparations made for it (e.g.
1708 * taken file locks) need to be undone.
1710 * This function recursively notifies all child nodes.
1712 static void bdrv_abort_perm_update(BlockDriverState
*bs
)
1714 BlockDriver
*drv
= bs
->drv
;
1721 if (drv
->bdrv_abort_perm_update
) {
1722 drv
->bdrv_abort_perm_update(bs
);
1725 QLIST_FOREACH(c
, &bs
->children
, next
) {
1726 bdrv_child_abort_perm_update(c
);
1730 static void bdrv_set_perm(BlockDriverState
*bs
, uint64_t cumulative_perms
,
1731 uint64_t cumulative_shared_perms
)
1733 BlockDriver
*drv
= bs
->drv
;
1740 /* Update this node */
1741 if (drv
->bdrv_set_perm
) {
1742 drv
->bdrv_set_perm(bs
, cumulative_perms
, cumulative_shared_perms
);
1745 /* Drivers that never have children can omit .bdrv_child_perm() */
1746 if (!drv
->bdrv_child_perm
) {
1747 assert(QLIST_EMPTY(&bs
->children
));
1751 /* Update all children */
1752 QLIST_FOREACH(c
, &bs
->children
, next
) {
1753 uint64_t cur_perm
, cur_shared
;
1754 bdrv_child_perm(bs
, c
->bs
, c
, c
->role
, NULL
,
1755 cumulative_perms
, cumulative_shared_perms
,
1756 &cur_perm
, &cur_shared
);
1757 bdrv_child_set_perm(c
, cur_perm
, cur_shared
);
1761 static void bdrv_get_cumulative_perm(BlockDriverState
*bs
, uint64_t *perm
,
1762 uint64_t *shared_perm
)
1765 uint64_t cumulative_perms
= 0;
1766 uint64_t cumulative_shared_perms
= BLK_PERM_ALL
;
1768 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
1769 cumulative_perms
|= c
->perm
;
1770 cumulative_shared_perms
&= c
->shared_perm
;
1773 *perm
= cumulative_perms
;
1774 *shared_perm
= cumulative_shared_perms
;
1777 static char *bdrv_child_user_desc(BdrvChild
*c
)
1779 if (c
->role
->get_parent_desc
) {
1780 return c
->role
->get_parent_desc(c
);
1783 return g_strdup("another user");
1786 char *bdrv_perm_names(uint64_t perm
)
1792 { BLK_PERM_CONSISTENT_READ
, "consistent read" },
1793 { BLK_PERM_WRITE
, "write" },
1794 { BLK_PERM_WRITE_UNCHANGED
, "write unchanged" },
1795 { BLK_PERM_RESIZE
, "resize" },
1796 { BLK_PERM_GRAPH_MOD
, "change children" },
1800 char *result
= g_strdup("");
1801 struct perm_name
*p
;
1803 for (p
= permissions
; p
->name
; p
++) {
1804 if (perm
& p
->perm
) {
1806 result
= g_strdup_printf("%s%s%s", old
, *old
? ", " : "", p
->name
);
1815 * Checks whether a new reference to @bs can be added if the new user requires
1816 * @new_used_perm/@new_shared_perm as its permissions. If @ignore_children is
1817 * set, the BdrvChild objects in this list are ignored in the calculations;
1818 * this allows checking permission updates for an existing reference.
1820 * Needs to be followed by a call to either bdrv_set_perm() or
1821 * bdrv_abort_perm_update(). */
1822 static int bdrv_check_update_perm(BlockDriverState
*bs
, BlockReopenQueue
*q
,
1823 uint64_t new_used_perm
,
1824 uint64_t new_shared_perm
,
1825 GSList
*ignore_children
, Error
**errp
)
1828 uint64_t cumulative_perms
= new_used_perm
;
1829 uint64_t cumulative_shared_perms
= new_shared_perm
;
1831 /* There is no reason why anyone couldn't tolerate write_unchanged */
1832 assert(new_shared_perm
& BLK_PERM_WRITE_UNCHANGED
);
1834 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
1835 if (g_slist_find(ignore_children
, c
)) {
1839 if ((new_used_perm
& c
->shared_perm
) != new_used_perm
) {
1840 char *user
= bdrv_child_user_desc(c
);
1841 char *perm_names
= bdrv_perm_names(new_used_perm
& ~c
->shared_perm
);
1842 error_setg(errp
, "Conflicts with use by %s as '%s', which does not "
1844 user
, c
->name
, perm_names
, bdrv_get_node_name(c
->bs
));
1850 if ((c
->perm
& new_shared_perm
) != c
->perm
) {
1851 char *user
= bdrv_child_user_desc(c
);
1852 char *perm_names
= bdrv_perm_names(c
->perm
& ~new_shared_perm
);
1853 error_setg(errp
, "Conflicts with use by %s as '%s', which uses "
1855 user
, c
->name
, perm_names
, bdrv_get_node_name(c
->bs
));
1861 cumulative_perms
|= c
->perm
;
1862 cumulative_shared_perms
&= c
->shared_perm
;
1865 return bdrv_check_perm(bs
, q
, cumulative_perms
, cumulative_shared_perms
,
1866 ignore_children
, errp
);
1869 /* Needs to be followed by a call to either bdrv_child_set_perm() or
1870 * bdrv_child_abort_perm_update(). */
1871 static int bdrv_child_check_perm(BdrvChild
*c
, BlockReopenQueue
*q
,
1872 uint64_t perm
, uint64_t shared
,
1873 GSList
*ignore_children
, Error
**errp
)
1877 ignore_children
= g_slist_prepend(g_slist_copy(ignore_children
), c
);
1878 ret
= bdrv_check_update_perm(c
->bs
, q
, perm
, shared
, ignore_children
, errp
);
1879 g_slist_free(ignore_children
);
1884 static void bdrv_child_set_perm(BdrvChild
*c
, uint64_t perm
, uint64_t shared
)
1886 uint64_t cumulative_perms
, cumulative_shared_perms
;
1889 c
->shared_perm
= shared
;
1891 bdrv_get_cumulative_perm(c
->bs
, &cumulative_perms
,
1892 &cumulative_shared_perms
);
1893 bdrv_set_perm(c
->bs
, cumulative_perms
, cumulative_shared_perms
);
1896 static void bdrv_child_abort_perm_update(BdrvChild
*c
)
1898 bdrv_abort_perm_update(c
->bs
);
1901 int bdrv_child_try_set_perm(BdrvChild
*c
, uint64_t perm
, uint64_t shared
,
1906 ret
= bdrv_child_check_perm(c
, NULL
, perm
, shared
, NULL
, errp
);
1908 bdrv_child_abort_perm_update(c
);
1912 bdrv_child_set_perm(c
, perm
, shared
);
1917 #define DEFAULT_PERM_PASSTHROUGH (BLK_PERM_CONSISTENT_READ \
1919 | BLK_PERM_WRITE_UNCHANGED \
1921 #define DEFAULT_PERM_UNCHANGED (BLK_PERM_ALL & ~DEFAULT_PERM_PASSTHROUGH)
1923 void bdrv_filter_default_perms(BlockDriverState
*bs
, BdrvChild
*c
,
1924 const BdrvChildRole
*role
,
1925 BlockReopenQueue
*reopen_queue
,
1926 uint64_t perm
, uint64_t shared
,
1927 uint64_t *nperm
, uint64_t *nshared
)
1930 *nperm
= perm
& DEFAULT_PERM_PASSTHROUGH
;
1931 *nshared
= (shared
& DEFAULT_PERM_PASSTHROUGH
) | DEFAULT_PERM_UNCHANGED
;
1935 *nperm
= (perm
& DEFAULT_PERM_PASSTHROUGH
) |
1936 (c
->perm
& DEFAULT_PERM_UNCHANGED
);
1937 *nshared
= (shared
& DEFAULT_PERM_PASSTHROUGH
) |
1938 (c
->shared_perm
& DEFAULT_PERM_UNCHANGED
);
1941 void bdrv_format_default_perms(BlockDriverState
*bs
, BdrvChild
*c
,
1942 const BdrvChildRole
*role
,
1943 BlockReopenQueue
*reopen_queue
,
1944 uint64_t perm
, uint64_t shared
,
1945 uint64_t *nperm
, uint64_t *nshared
)
1947 bool backing
= (role
== &child_backing
);
1948 assert(role
== &child_backing
|| role
== &child_file
);
1951 int flags
= bdrv_reopen_get_flags(reopen_queue
, bs
);
1953 /* Apart from the modifications below, the same permissions are
1954 * forwarded and left alone as for filters */
1955 bdrv_filter_default_perms(bs
, c
, role
, reopen_queue
, perm
, shared
,
1958 /* Format drivers may touch metadata even if the guest doesn't write */
1959 if (bdrv_is_writable(bs
, reopen_queue
)) {
1960 perm
|= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
1963 /* bs->file always needs to be consistent because of the metadata. We
1964 * can never allow other users to resize or write to it. */
1965 if (!(flags
& BDRV_O_NO_IO
)) {
1966 perm
|= BLK_PERM_CONSISTENT_READ
;
1968 shared
&= ~(BLK_PERM_WRITE
| BLK_PERM_RESIZE
);
1970 /* We want consistent read from backing files if the parent needs it.
1971 * No other operations are performed on backing files. */
1972 perm
&= BLK_PERM_CONSISTENT_READ
;
1974 /* If the parent can deal with changing data, we're okay with a
1975 * writable and resizable backing file. */
1976 /* TODO Require !(perm & BLK_PERM_CONSISTENT_READ), too? */
1977 if (shared
& BLK_PERM_WRITE
) {
1978 shared
= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
1983 shared
|= BLK_PERM_CONSISTENT_READ
| BLK_PERM_GRAPH_MOD
|
1984 BLK_PERM_WRITE_UNCHANGED
;
1987 if (bs
->open_flags
& BDRV_O_INACTIVE
) {
1988 shared
|= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
1995 static void bdrv_replace_child_noperm(BdrvChild
*child
,
1996 BlockDriverState
*new_bs
)
1998 BlockDriverState
*old_bs
= child
->bs
;
2001 if (old_bs
&& new_bs
) {
2002 assert(bdrv_get_aio_context(old_bs
) == bdrv_get_aio_context(new_bs
));
2005 /* Detach first so that the recursive drain sections coming from @child
2006 * are already gone and we only end the drain sections that came from
2008 if (child
->role
->detach
) {
2009 child
->role
->detach(child
);
2011 if (old_bs
->quiesce_counter
&& child
->role
->drained_end
) {
2012 for (i
= 0; i
< old_bs
->quiesce_counter
; i
++) {
2013 child
->role
->drained_end(child
);
2016 QLIST_REMOVE(child
, next_parent
);
2022 QLIST_INSERT_HEAD(&new_bs
->parents
, child
, next_parent
);
2023 if (new_bs
->quiesce_counter
&& child
->role
->drained_begin
) {
2024 for (i
= 0; i
< new_bs
->quiesce_counter
; i
++) {
2025 child
->role
->drained_begin(child
);
2029 /* Attach only after starting new drained sections, so that recursive
2030 * drain sections coming from @child don't get an extra .drained_begin
2032 if (child
->role
->attach
) {
2033 child
->role
->attach(child
);
2039 * Updates @child to change its reference to point to @new_bs, including
2040 * checking and applying the necessary permisson updates both to the old node
2043 * NULL is passed as @new_bs for removing the reference before freeing @child.
2045 * If @new_bs is not NULL, bdrv_check_perm() must be called beforehand, as this
2046 * function uses bdrv_set_perm() to update the permissions according to the new
2047 * reference that @new_bs gets.
2049 static void bdrv_replace_child(BdrvChild
*child
, BlockDriverState
*new_bs
)
2051 BlockDriverState
*old_bs
= child
->bs
;
2052 uint64_t perm
, shared_perm
;
2054 bdrv_replace_child_noperm(child
, new_bs
);
2057 /* Update permissions for old node. This is guaranteed to succeed
2058 * because we're just taking a parent away, so we're loosening
2060 bdrv_get_cumulative_perm(old_bs
, &perm
, &shared_perm
);
2061 bdrv_check_perm(old_bs
, NULL
, perm
, shared_perm
, NULL
, &error_abort
);
2062 bdrv_set_perm(old_bs
, perm
, shared_perm
);
2066 bdrv_get_cumulative_perm(new_bs
, &perm
, &shared_perm
);
2067 bdrv_set_perm(new_bs
, perm
, shared_perm
);
2071 BdrvChild
*bdrv_root_attach_child(BlockDriverState
*child_bs
,
2072 const char *child_name
,
2073 const BdrvChildRole
*child_role
,
2074 uint64_t perm
, uint64_t shared_perm
,
2075 void *opaque
, Error
**errp
)
2080 ret
= bdrv_check_update_perm(child_bs
, NULL
, perm
, shared_perm
, NULL
, errp
);
2082 bdrv_abort_perm_update(child_bs
);
2086 child
= g_new(BdrvChild
, 1);
2087 *child
= (BdrvChild
) {
2089 .name
= g_strdup(child_name
),
2092 .shared_perm
= shared_perm
,
2096 /* This performs the matching bdrv_set_perm() for the above check. */
2097 bdrv_replace_child(child
, child_bs
);
2102 BdrvChild
*bdrv_attach_child(BlockDriverState
*parent_bs
,
2103 BlockDriverState
*child_bs
,
2104 const char *child_name
,
2105 const BdrvChildRole
*child_role
,
2109 uint64_t perm
, shared_perm
;
2111 bdrv_get_cumulative_perm(parent_bs
, &perm
, &shared_perm
);
2113 assert(parent_bs
->drv
);
2114 assert(bdrv_get_aio_context(parent_bs
) == bdrv_get_aio_context(child_bs
));
2115 bdrv_child_perm(parent_bs
, child_bs
, NULL
, child_role
, NULL
,
2116 perm
, shared_perm
, &perm
, &shared_perm
);
2118 child
= bdrv_root_attach_child(child_bs
, child_name
, child_role
,
2119 perm
, shared_perm
, parent_bs
, errp
);
2120 if (child
== NULL
) {
2124 QLIST_INSERT_HEAD(&parent_bs
->children
, child
, next
);
2128 static void bdrv_detach_child(BdrvChild
*child
)
2130 if (child
->next
.le_prev
) {
2131 QLIST_REMOVE(child
, next
);
2132 child
->next
.le_prev
= NULL
;
2135 bdrv_replace_child(child
, NULL
);
2137 g_free(child
->name
);
2141 void bdrv_root_unref_child(BdrvChild
*child
)
2143 BlockDriverState
*child_bs
;
2145 child_bs
= child
->bs
;
2146 bdrv_detach_child(child
);
2147 bdrv_unref(child_bs
);
2150 void bdrv_unref_child(BlockDriverState
*parent
, BdrvChild
*child
)
2152 if (child
== NULL
) {
2156 if (child
->bs
->inherits_from
== parent
) {
2159 /* Remove inherits_from only when the last reference between parent and
2160 * child->bs goes away. */
2161 QLIST_FOREACH(c
, &parent
->children
, next
) {
2162 if (c
!= child
&& c
->bs
== child
->bs
) {
2167 child
->bs
->inherits_from
= NULL
;
2171 bdrv_root_unref_child(child
);
2175 static void bdrv_parent_cb_change_media(BlockDriverState
*bs
, bool load
)
2178 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
2179 if (c
->role
->change_media
) {
2180 c
->role
->change_media(c
, load
);
2185 static void bdrv_parent_cb_resize(BlockDriverState
*bs
)
2188 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
2189 if (c
->role
->resize
) {
2196 * Sets the backing file link of a BDS. A new reference is created; callers
2197 * which don't need their own reference any more must call bdrv_unref().
2199 void bdrv_set_backing_hd(BlockDriverState
*bs
, BlockDriverState
*backing_hd
,
2203 bdrv_ref(backing_hd
);
2207 bdrv_unref_child(bs
, bs
->backing
);
2215 bs
->backing
= bdrv_attach_child(bs
, backing_hd
, "backing", &child_backing
,
2218 bdrv_unref(backing_hd
);
2221 bdrv_refresh_filename(bs
);
2224 bdrv_refresh_limits(bs
, NULL
);
2228 * Opens the backing file for a BlockDriverState if not yet open
2230 * bdref_key specifies the key for the image's BlockdevRef in the options QDict.
2231 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
2232 * itself, all options starting with "${bdref_key}." are considered part of the
2235 * TODO Can this be unified with bdrv_open_image()?
2237 int bdrv_open_backing_file(BlockDriverState
*bs
, QDict
*parent_options
,
2238 const char *bdref_key
, Error
**errp
)
2240 char *backing_filename
= g_malloc0(PATH_MAX
);
2241 char *bdref_key_dot
;
2242 const char *reference
= NULL
;
2244 BlockDriverState
*backing_hd
;
2246 QDict
*tmp_parent_options
= NULL
;
2247 Error
*local_err
= NULL
;
2249 if (bs
->backing
!= NULL
) {
2253 /* NULL means an empty set of options */
2254 if (parent_options
== NULL
) {
2255 tmp_parent_options
= qdict_new();
2256 parent_options
= tmp_parent_options
;
2259 bs
->open_flags
&= ~BDRV_O_NO_BACKING
;
2261 bdref_key_dot
= g_strdup_printf("%s.", bdref_key
);
2262 qdict_extract_subqdict(parent_options
, &options
, bdref_key_dot
);
2263 g_free(bdref_key_dot
);
2266 * Caution: while qdict_get_try_str() is fine, getting non-string
2267 * types would require more care. When @parent_options come from
2268 * -blockdev or blockdev_add, its members are typed according to
2269 * the QAPI schema, but when they come from -drive, they're all
2272 reference
= qdict_get_try_str(parent_options
, bdref_key
);
2273 if (reference
|| qdict_haskey(options
, "file.filename")) {
2274 backing_filename
[0] = '\0';
2275 } else if (bs
->backing_file
[0] == '\0' && qdict_size(options
) == 0) {
2279 bdrv_get_full_backing_filename(bs
, backing_filename
, PATH_MAX
,
2283 error_propagate(errp
, local_err
);
2289 if (!bs
->drv
|| !bs
->drv
->supports_backing
) {
2291 error_setg(errp
, "Driver doesn't support backing files");
2297 bs
->backing_format
[0] != '\0' && !qdict_haskey(options
, "driver")) {
2298 qdict_put_str(options
, "driver", bs
->backing_format
);
2301 backing_hd
= bdrv_open_inherit(*backing_filename
? backing_filename
: NULL
,
2302 reference
, options
, 0, bs
, &child_backing
,
2305 bs
->open_flags
|= BDRV_O_NO_BACKING
;
2306 error_prepend(errp
, "Could not open backing file: ");
2310 bdrv_set_aio_context(backing_hd
, bdrv_get_aio_context(bs
));
2312 /* Hook up the backing file link; drop our reference, bs owns the
2313 * backing_hd reference now */
2314 bdrv_set_backing_hd(bs
, backing_hd
, &local_err
);
2315 bdrv_unref(backing_hd
);
2317 error_propagate(errp
, local_err
);
2322 qdict_del(parent_options
, bdref_key
);
2325 g_free(backing_filename
);
2326 QDECREF(tmp_parent_options
);
2330 static BlockDriverState
*
2331 bdrv_open_child_bs(const char *filename
, QDict
*options
, const char *bdref_key
,
2332 BlockDriverState
*parent
, const BdrvChildRole
*child_role
,
2333 bool allow_none
, Error
**errp
)
2335 BlockDriverState
*bs
= NULL
;
2336 QDict
*image_options
;
2337 char *bdref_key_dot
;
2338 const char *reference
;
2340 assert(child_role
!= NULL
);
2342 bdref_key_dot
= g_strdup_printf("%s.", bdref_key
);
2343 qdict_extract_subqdict(options
, &image_options
, bdref_key_dot
);
2344 g_free(bdref_key_dot
);
2347 * Caution: while qdict_get_try_str() is fine, getting non-string
2348 * types would require more care. When @options come from
2349 * -blockdev or blockdev_add, its members are typed according to
2350 * the QAPI schema, but when they come from -drive, they're all
2353 reference
= qdict_get_try_str(options
, bdref_key
);
2354 if (!filename
&& !reference
&& !qdict_size(image_options
)) {
2356 error_setg(errp
, "A block device must be specified for \"%s\"",
2359 QDECREF(image_options
);
2363 bs
= bdrv_open_inherit(filename
, reference
, image_options
, 0,
2364 parent
, child_role
, errp
);
2370 qdict_del(options
, bdref_key
);
2375 * Opens a disk image whose options are given as BlockdevRef in another block
2378 * If allow_none is true, no image will be opened if filename is false and no
2379 * BlockdevRef is given. NULL will be returned, but errp remains unset.
2381 * bdrev_key specifies the key for the image's BlockdevRef in the options QDict.
2382 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
2383 * itself, all options starting with "${bdref_key}." are considered part of the
2386 * The BlockdevRef will be removed from the options QDict.
2388 BdrvChild
*bdrv_open_child(const char *filename
,
2389 QDict
*options
, const char *bdref_key
,
2390 BlockDriverState
*parent
,
2391 const BdrvChildRole
*child_role
,
2392 bool allow_none
, Error
**errp
)
2395 BlockDriverState
*bs
;
2397 bs
= bdrv_open_child_bs(filename
, options
, bdref_key
, parent
, child_role
,
2403 c
= bdrv_attach_child(parent
, bs
, bdref_key
, child_role
, errp
);
2412 /* TODO Future callers may need to specify parent/child_role in order for
2413 * option inheritance to work. Existing callers use it for the root node. */
2414 BlockDriverState
*bdrv_open_blockdev_ref(BlockdevRef
*ref
, Error
**errp
)
2416 BlockDriverState
*bs
= NULL
;
2417 Error
*local_err
= NULL
;
2418 QObject
*obj
= NULL
;
2419 QDict
*qdict
= NULL
;
2420 const char *reference
= NULL
;
2423 if (ref
->type
== QTYPE_QSTRING
) {
2424 reference
= ref
->u
.reference
;
2426 BlockdevOptions
*options
= &ref
->u
.definition
;
2427 assert(ref
->type
== QTYPE_QDICT
);
2429 v
= qobject_output_visitor_new(&obj
);
2430 visit_type_BlockdevOptions(v
, NULL
, &options
, &local_err
);
2432 error_propagate(errp
, local_err
);
2435 visit_complete(v
, &obj
);
2437 qdict
= qobject_to(QDict
, obj
);
2438 qdict_flatten(qdict
);
2440 /* bdrv_open_inherit() defaults to the values in bdrv_flags (for
2441 * compatibility with other callers) rather than what we want as the
2442 * real defaults. Apply the defaults here instead. */
2443 qdict_set_default_str(qdict
, BDRV_OPT_CACHE_DIRECT
, "off");
2444 qdict_set_default_str(qdict
, BDRV_OPT_CACHE_NO_FLUSH
, "off");
2445 qdict_set_default_str(qdict
, BDRV_OPT_READ_ONLY
, "off");
2448 bs
= bdrv_open_inherit(NULL
, reference
, qdict
, 0, NULL
, NULL
, errp
);
2452 qobject_decref(obj
);
2457 static BlockDriverState
*bdrv_append_temp_snapshot(BlockDriverState
*bs
,
2459 QDict
*snapshot_options
,
2462 /* TODO: extra byte is a hack to ensure MAX_PATH space on Windows. */
2463 char *tmp_filename
= g_malloc0(PATH_MAX
+ 1);
2465 QemuOpts
*opts
= NULL
;
2466 BlockDriverState
*bs_snapshot
= NULL
;
2467 Error
*local_err
= NULL
;
2470 /* if snapshot, we create a temporary backing file and open it
2471 instead of opening 'filename' directly */
2473 /* Get the required size from the image */
2474 total_size
= bdrv_getlength(bs
);
2475 if (total_size
< 0) {
2476 error_setg_errno(errp
, -total_size
, "Could not get image size");
2480 /* Create the temporary image */
2481 ret
= get_tmp_filename(tmp_filename
, PATH_MAX
+ 1);
2483 error_setg_errno(errp
, -ret
, "Could not get temporary filename");
2487 opts
= qemu_opts_create(bdrv_qcow2
.create_opts
, NULL
, 0,
2489 qemu_opt_set_number(opts
, BLOCK_OPT_SIZE
, total_size
, &error_abort
);
2490 ret
= bdrv_create(&bdrv_qcow2
, tmp_filename
, opts
, errp
);
2491 qemu_opts_del(opts
);
2493 error_prepend(errp
, "Could not create temporary overlay '%s': ",
2498 /* Prepare options QDict for the temporary file */
2499 qdict_put_str(snapshot_options
, "file.driver", "file");
2500 qdict_put_str(snapshot_options
, "file.filename", tmp_filename
);
2501 qdict_put_str(snapshot_options
, "driver", "qcow2");
2503 bs_snapshot
= bdrv_open(NULL
, NULL
, snapshot_options
, flags
, errp
);
2504 snapshot_options
= NULL
;
2509 /* bdrv_append() consumes a strong reference to bs_snapshot
2510 * (i.e. it will call bdrv_unref() on it) even on error, so in
2511 * order to be able to return one, we have to increase
2512 * bs_snapshot's refcount here */
2513 bdrv_ref(bs_snapshot
);
2514 bdrv_append(bs_snapshot
, bs
, &local_err
);
2516 error_propagate(errp
, local_err
);
2522 QDECREF(snapshot_options
);
2523 g_free(tmp_filename
);
2528 * Opens a disk image (raw, qcow2, vmdk, ...)
2530 * options is a QDict of options to pass to the block drivers, or NULL for an
2531 * empty set of options. The reference to the QDict belongs to the block layer
2532 * after the call (even on failure), so if the caller intends to reuse the
2533 * dictionary, it needs to use QINCREF() before calling bdrv_open.
2535 * If *pbs is NULL, a new BDS will be created with a pointer to it stored there.
2536 * If it is not NULL, the referenced BDS will be reused.
2538 * The reference parameter may be used to specify an existing block device which
2539 * should be opened. If specified, neither options nor a filename may be given,
2540 * nor can an existing BDS be reused (that is, *pbs has to be NULL).
2542 static BlockDriverState
*bdrv_open_inherit(const char *filename
,
2543 const char *reference
,
2544 QDict
*options
, int flags
,
2545 BlockDriverState
*parent
,
2546 const BdrvChildRole
*child_role
,
2550 BlockBackend
*file
= NULL
;
2551 BlockDriverState
*bs
;
2552 BlockDriver
*drv
= NULL
;
2553 const char *drvname
;
2554 const char *backing
;
2555 Error
*local_err
= NULL
;
2556 QDict
*snapshot_options
= NULL
;
2557 int snapshot_flags
= 0;
2559 assert(!child_role
|| !flags
);
2560 assert(!child_role
== !parent
);
2563 bool options_non_empty
= options
? qdict_size(options
) : false;
2566 if (filename
|| options_non_empty
) {
2567 error_setg(errp
, "Cannot reference an existing block device with "
2568 "additional options or a new filename");
2572 bs
= bdrv_lookup_bs(reference
, reference
, errp
);
2583 /* NULL means an empty set of options */
2584 if (options
== NULL
) {
2585 options
= qdict_new();
2588 /* json: syntax counts as explicit options, as if in the QDict */
2589 parse_json_protocol(options
, &filename
, &local_err
);
2594 bs
->explicit_options
= qdict_clone_shallow(options
);
2597 bs
->inherits_from
= parent
;
2598 child_role
->inherit_options(&flags
, options
,
2599 parent
->open_flags
, parent
->options
);
2602 ret
= bdrv_fill_options(&options
, filename
, &flags
, &local_err
);
2608 * Set the BDRV_O_RDWR and BDRV_O_ALLOW_RDWR flags.
2609 * Caution: getting a boolean member of @options requires care.
2610 * When @options come from -blockdev or blockdev_add, members are
2611 * typed according to the QAPI schema, but when they come from
2612 * -drive, they're all QString.
2614 if (g_strcmp0(qdict_get_try_str(options
, BDRV_OPT_READ_ONLY
), "on") &&
2615 !qdict_get_try_bool(options
, BDRV_OPT_READ_ONLY
, false)) {
2616 flags
|= (BDRV_O_RDWR
| BDRV_O_ALLOW_RDWR
);
2618 flags
&= ~BDRV_O_RDWR
;
2621 if (flags
& BDRV_O_SNAPSHOT
) {
2622 snapshot_options
= qdict_new();
2623 bdrv_temp_snapshot_options(&snapshot_flags
, snapshot_options
,
2625 /* Let bdrv_backing_options() override "read-only" */
2626 qdict_del(options
, BDRV_OPT_READ_ONLY
);
2627 bdrv_backing_options(&flags
, options
, flags
, options
);
2630 bs
->open_flags
= flags
;
2631 bs
->options
= options
;
2632 options
= qdict_clone_shallow(options
);
2634 /* Find the right image format driver */
2635 /* See cautionary note on accessing @options above */
2636 drvname
= qdict_get_try_str(options
, "driver");
2638 drv
= bdrv_find_format(drvname
);
2640 error_setg(errp
, "Unknown driver: '%s'", drvname
);
2645 assert(drvname
|| !(flags
& BDRV_O_PROTOCOL
));
2647 /* See cautionary note on accessing @options above */
2648 backing
= qdict_get_try_str(options
, "backing");
2649 if (qobject_to(QNull
, qdict_get(options
, "backing")) != NULL
||
2650 (backing
&& *backing
== '\0'))
2652 flags
|= BDRV_O_NO_BACKING
;
2653 qdict_del(options
, "backing");
2656 /* Open image file without format layer. This BlockBackend is only used for
2657 * probing, the block drivers will do their own bdrv_open_child() for the
2658 * same BDS, which is why we put the node name back into options. */
2659 if ((flags
& BDRV_O_PROTOCOL
) == 0) {
2660 BlockDriverState
*file_bs
;
2662 file_bs
= bdrv_open_child_bs(filename
, options
, "file", bs
,
2663 &child_file
, true, &local_err
);
2667 if (file_bs
!= NULL
) {
2668 /* Not requesting BLK_PERM_CONSISTENT_READ because we're only
2669 * looking at the header to guess the image format. This works even
2670 * in cases where a guest would not see a consistent state. */
2671 file
= blk_new(0, BLK_PERM_ALL
);
2672 blk_insert_bs(file
, file_bs
, &local_err
);
2673 bdrv_unref(file_bs
);
2678 qdict_put_str(options
, "file", bdrv_get_node_name(file_bs
));
2682 /* Image format probing */
2685 ret
= find_image_format(file
, filename
, &drv
, &local_err
);
2690 * This option update would logically belong in bdrv_fill_options(),
2691 * but we first need to open bs->file for the probing to work, while
2692 * opening bs->file already requires the (mostly) final set of options
2693 * so that cache mode etc. can be inherited.
2695 * Adding the driver later is somewhat ugly, but it's not an option
2696 * that would ever be inherited, so it's correct. We just need to make
2697 * sure to update both bs->options (which has the full effective
2698 * options for bs) and options (which has file.* already removed).
2700 qdict_put_str(bs
->options
, "driver", drv
->format_name
);
2701 qdict_put_str(options
, "driver", drv
->format_name
);
2703 error_setg(errp
, "Must specify either driver or file");
2707 /* BDRV_O_PROTOCOL must be set iff a protocol BDS is about to be created */
2708 assert(!!(flags
& BDRV_O_PROTOCOL
) == !!drv
->bdrv_file_open
);
2709 /* file must be NULL if a protocol BDS is about to be created
2710 * (the inverse results in an error message from bdrv_open_common()) */
2711 assert(!(flags
& BDRV_O_PROTOCOL
) || !file
);
2713 /* Open the image */
2714 ret
= bdrv_open_common(bs
, file
, options
, &local_err
);
2724 /* If there is a backing file, use it */
2725 if ((flags
& BDRV_O_NO_BACKING
) == 0) {
2726 ret
= bdrv_open_backing_file(bs
, options
, "backing", &local_err
);
2728 goto close_and_fail
;
2732 bdrv_refresh_filename(bs
);
2734 /* Check if any unknown options were used */
2735 if (qdict_size(options
) != 0) {
2736 const QDictEntry
*entry
= qdict_first(options
);
2737 if (flags
& BDRV_O_PROTOCOL
) {
2738 error_setg(errp
, "Block protocol '%s' doesn't support the option "
2739 "'%s'", drv
->format_name
, entry
->key
);
2742 "Block format '%s' does not support the option '%s'",
2743 drv
->format_name
, entry
->key
);
2746 goto close_and_fail
;
2749 bdrv_parent_cb_change_media(bs
, true);
2753 /* For snapshot=on, create a temporary qcow2 overlay. bs points to the
2754 * temporary snapshot afterwards. */
2755 if (snapshot_flags
) {
2756 BlockDriverState
*snapshot_bs
;
2757 snapshot_bs
= bdrv_append_temp_snapshot(bs
, snapshot_flags
,
2758 snapshot_options
, &local_err
);
2759 snapshot_options
= NULL
;
2761 goto close_and_fail
;
2763 /* We are not going to return bs but the overlay on top of it
2764 * (snapshot_bs); thus, we have to drop the strong reference to bs
2765 * (which we obtained by calling bdrv_new()). bs will not be deleted,
2766 * though, because the overlay still has a reference to it. */
2775 QDECREF(snapshot_options
);
2776 QDECREF(bs
->explicit_options
);
2777 QDECREF(bs
->options
);
2780 bs
->explicit_options
= NULL
;
2782 error_propagate(errp
, local_err
);
2787 QDECREF(snapshot_options
);
2789 error_propagate(errp
, local_err
);
2793 BlockDriverState
*bdrv_open(const char *filename
, const char *reference
,
2794 QDict
*options
, int flags
, Error
**errp
)
2796 return bdrv_open_inherit(filename
, reference
, options
, flags
, NULL
,
2801 * Adds a BlockDriverState to a simple queue for an atomic, transactional
2802 * reopen of multiple devices.
2804 * bs_queue can either be an existing BlockReopenQueue that has had QSIMPLE_INIT
2805 * already performed, or alternatively may be NULL a new BlockReopenQueue will
2806 * be created and initialized. This newly created BlockReopenQueue should be
2807 * passed back in for subsequent calls that are intended to be of the same
2810 * bs is the BlockDriverState to add to the reopen queue.
2812 * options contains the changed options for the associated bs
2813 * (the BlockReopenQueue takes ownership)
2815 * flags contains the open flags for the associated bs
2817 * returns a pointer to bs_queue, which is either the newly allocated
2818 * bs_queue, or the existing bs_queue being used.
2820 * bs must be drained between bdrv_reopen_queue() and bdrv_reopen_multiple().
2822 static BlockReopenQueue
*bdrv_reopen_queue_child(BlockReopenQueue
*bs_queue
,
2823 BlockDriverState
*bs
,
2826 const BdrvChildRole
*role
,
2827 QDict
*parent_options
,
2832 BlockReopenQueueEntry
*bs_entry
;
2834 QDict
*old_options
, *explicit_options
;
2836 /* Make sure that the caller remembered to use a drained section. This is
2837 * important to avoid graph changes between the recursive queuing here and
2838 * bdrv_reopen_multiple(). */
2839 assert(bs
->quiesce_counter
> 0);
2841 if (bs_queue
== NULL
) {
2842 bs_queue
= g_new0(BlockReopenQueue
, 1);
2843 QSIMPLEQ_INIT(bs_queue
);
2847 options
= qdict_new();
2850 /* Check if this BlockDriverState is already in the queue */
2851 QSIMPLEQ_FOREACH(bs_entry
, bs_queue
, entry
) {
2852 if (bs
== bs_entry
->state
.bs
) {
2858 * Precedence of options:
2859 * 1. Explicitly passed in options (highest)
2860 * 2. Set in flags (only for top level)
2861 * 3. Retained from explicitly set options of bs
2862 * 4. Inherited from parent node
2863 * 5. Retained from effective options of bs
2866 if (!parent_options
) {
2868 * Any setting represented by flags is always updated. If the
2869 * corresponding QDict option is set, it takes precedence. Otherwise
2870 * the flag is translated into a QDict option. The old setting of bs is
2873 update_options_from_flags(options
, flags
);
2876 /* Old explicitly set values (don't overwrite by inherited value) */
2878 old_options
= qdict_clone_shallow(bs_entry
->state
.explicit_options
);
2880 old_options
= qdict_clone_shallow(bs
->explicit_options
);
2882 bdrv_join_options(bs
, options
, old_options
);
2883 QDECREF(old_options
);
2885 explicit_options
= qdict_clone_shallow(options
);
2887 /* Inherit from parent node */
2888 if (parent_options
) {
2890 QDict
*options_copy
;
2892 role
->inherit_options(&flags
, options
, parent_flags
, parent_options
);
2893 options_copy
= qdict_clone_shallow(options
);
2894 opts
= qemu_opts_create(&bdrv_runtime_opts
, NULL
, 0, &error_abort
);
2895 qemu_opts_absorb_qdict(opts
, options_copy
, NULL
);
2896 update_flags_from_options(&flags
, opts
);
2897 qemu_opts_del(opts
);
2898 QDECREF(options_copy
);
2901 /* Old values are used for options that aren't set yet */
2902 old_options
= qdict_clone_shallow(bs
->options
);
2903 bdrv_join_options(bs
, options
, old_options
);
2904 QDECREF(old_options
);
2906 /* bdrv_open_inherit() sets and clears some additional flags internally */
2907 flags
&= ~BDRV_O_PROTOCOL
;
2908 if (flags
& BDRV_O_RDWR
) {
2909 flags
|= BDRV_O_ALLOW_RDWR
;
2913 bs_entry
= g_new0(BlockReopenQueueEntry
, 1);
2914 QSIMPLEQ_INSERT_TAIL(bs_queue
, bs_entry
, entry
);
2916 QDECREF(bs_entry
->state
.options
);
2917 QDECREF(bs_entry
->state
.explicit_options
);
2920 bs_entry
->state
.bs
= bs
;
2921 bs_entry
->state
.options
= options
;
2922 bs_entry
->state
.explicit_options
= explicit_options
;
2923 bs_entry
->state
.flags
= flags
;
2925 /* This needs to be overwritten in bdrv_reopen_prepare() */
2926 bs_entry
->state
.perm
= UINT64_MAX
;
2927 bs_entry
->state
.shared_perm
= 0;
2929 QLIST_FOREACH(child
, &bs
->children
, next
) {
2930 QDict
*new_child_options
;
2931 char *child_key_dot
;
2933 /* reopen can only change the options of block devices that were
2934 * implicitly created and inherited options. For other (referenced)
2935 * block devices, a syntax like "backing.foo" results in an error. */
2936 if (child
->bs
->inherits_from
!= bs
) {
2940 child_key_dot
= g_strdup_printf("%s.", child
->name
);
2941 qdict_extract_subqdict(options
, &new_child_options
, child_key_dot
);
2942 g_free(child_key_dot
);
2944 bdrv_reopen_queue_child(bs_queue
, child
->bs
, new_child_options
, 0,
2945 child
->role
, options
, flags
);
2951 BlockReopenQueue
*bdrv_reopen_queue(BlockReopenQueue
*bs_queue
,
2952 BlockDriverState
*bs
,
2953 QDict
*options
, int flags
)
2955 return bdrv_reopen_queue_child(bs_queue
, bs
, options
, flags
,
2960 * Reopen multiple BlockDriverStates atomically & transactionally.
2962 * The queue passed in (bs_queue) must have been built up previous
2963 * via bdrv_reopen_queue().
2965 * Reopens all BDS specified in the queue, with the appropriate
2966 * flags. All devices are prepared for reopen, and failure of any
2967 * device will cause all device changes to be abandonded, and intermediate
2970 * If all devices prepare successfully, then the changes are committed
2973 * All affected nodes must be drained between bdrv_reopen_queue() and
2974 * bdrv_reopen_multiple().
2976 int bdrv_reopen_multiple(AioContext
*ctx
, BlockReopenQueue
*bs_queue
, Error
**errp
)
2979 BlockReopenQueueEntry
*bs_entry
, *next
;
2980 Error
*local_err
= NULL
;
2982 assert(bs_queue
!= NULL
);
2984 QSIMPLEQ_FOREACH(bs_entry
, bs_queue
, entry
) {
2985 assert(bs_entry
->state
.bs
->quiesce_counter
> 0);
2986 if (bdrv_reopen_prepare(&bs_entry
->state
, bs_queue
, &local_err
)) {
2987 error_propagate(errp
, local_err
);
2990 bs_entry
->prepared
= true;
2993 /* If we reach this point, we have success and just need to apply the
2996 QSIMPLEQ_FOREACH(bs_entry
, bs_queue
, entry
) {
2997 bdrv_reopen_commit(&bs_entry
->state
);
3003 QSIMPLEQ_FOREACH_SAFE(bs_entry
, bs_queue
, entry
, next
) {
3004 if (ret
&& bs_entry
->prepared
) {
3005 bdrv_reopen_abort(&bs_entry
->state
);
3007 QDECREF(bs_entry
->state
.explicit_options
);
3009 QDECREF(bs_entry
->state
.options
);
3018 /* Reopen a single BlockDriverState with the specified flags. */
3019 int bdrv_reopen(BlockDriverState
*bs
, int bdrv_flags
, Error
**errp
)
3022 Error
*local_err
= NULL
;
3023 BlockReopenQueue
*queue
;
3025 bdrv_subtree_drained_begin(bs
);
3027 queue
= bdrv_reopen_queue(NULL
, bs
, NULL
, bdrv_flags
);
3028 ret
= bdrv_reopen_multiple(bdrv_get_aio_context(bs
), queue
, &local_err
);
3029 if (local_err
!= NULL
) {
3030 error_propagate(errp
, local_err
);
3033 bdrv_subtree_drained_end(bs
);
3038 static BlockReopenQueueEntry
*find_parent_in_reopen_queue(BlockReopenQueue
*q
,
3041 BlockReopenQueueEntry
*entry
;
3043 QSIMPLEQ_FOREACH(entry
, q
, entry
) {
3044 BlockDriverState
*bs
= entry
->state
.bs
;
3047 QLIST_FOREACH(child
, &bs
->children
, next
) {
3057 static void bdrv_reopen_perm(BlockReopenQueue
*q
, BlockDriverState
*bs
,
3058 uint64_t *perm
, uint64_t *shared
)
3061 BlockReopenQueueEntry
*parent
;
3062 uint64_t cumulative_perms
= 0;
3063 uint64_t cumulative_shared_perms
= BLK_PERM_ALL
;
3065 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
3066 parent
= find_parent_in_reopen_queue(q
, c
);
3068 cumulative_perms
|= c
->perm
;
3069 cumulative_shared_perms
&= c
->shared_perm
;
3071 uint64_t nperm
, nshared
;
3073 bdrv_child_perm(parent
->state
.bs
, bs
, c
, c
->role
, q
,
3074 parent
->state
.perm
, parent
->state
.shared_perm
,
3077 cumulative_perms
|= nperm
;
3078 cumulative_shared_perms
&= nshared
;
3081 *perm
= cumulative_perms
;
3082 *shared
= cumulative_shared_perms
;
3086 * Prepares a BlockDriverState for reopen. All changes are staged in the
3087 * 'opaque' field of the BDRVReopenState, which is used and allocated by
3088 * the block driver layer .bdrv_reopen_prepare()
3090 * bs is the BlockDriverState to reopen
3091 * flags are the new open flags
3092 * queue is the reopen queue
3094 * Returns 0 on success, non-zero on error. On error errp will be set
3097 * On failure, bdrv_reopen_abort() will be called to clean up any data.
3098 * It is the responsibility of the caller to then call the abort() or
3099 * commit() for any other BDS that have been left in a prepare() state
3102 int bdrv_reopen_prepare(BDRVReopenState
*reopen_state
, BlockReopenQueue
*queue
,
3106 Error
*local_err
= NULL
;
3112 assert(reopen_state
!= NULL
);
3113 assert(reopen_state
->bs
->drv
!= NULL
);
3114 drv
= reopen_state
->bs
->drv
;
3116 /* Process generic block layer options */
3117 opts
= qemu_opts_create(&bdrv_runtime_opts
, NULL
, 0, &error_abort
);
3118 qemu_opts_absorb_qdict(opts
, reopen_state
->options
, &local_err
);
3120 error_propagate(errp
, local_err
);
3125 update_flags_from_options(&reopen_state
->flags
, opts
);
3127 /* node-name and driver must be unchanged. Put them back into the QDict, so
3128 * that they are checked at the end of this function. */
3129 value
= qemu_opt_get(opts
, "node-name");
3131 qdict_put_str(reopen_state
->options
, "node-name", value
);
3134 value
= qemu_opt_get(opts
, "driver");
3136 qdict_put_str(reopen_state
->options
, "driver", value
);
3139 /* If we are to stay read-only, do not allow permission change
3140 * to r/w. Attempting to set to r/w may fail if either BDRV_O_ALLOW_RDWR is
3141 * not set, or if the BDS still has copy_on_read enabled */
3142 read_only
= !(reopen_state
->flags
& BDRV_O_RDWR
);
3143 ret
= bdrv_can_set_read_only(reopen_state
->bs
, read_only
, true, &local_err
);
3145 error_propagate(errp
, local_err
);
3149 /* Calculate required permissions after reopening */
3150 bdrv_reopen_perm(queue
, reopen_state
->bs
,
3151 &reopen_state
->perm
, &reopen_state
->shared_perm
);
3153 ret
= bdrv_flush(reopen_state
->bs
);
3155 error_setg_errno(errp
, -ret
, "Error flushing drive");
3159 if (drv
->bdrv_reopen_prepare
) {
3160 ret
= drv
->bdrv_reopen_prepare(reopen_state
, queue
, &local_err
);
3162 if (local_err
!= NULL
) {
3163 error_propagate(errp
, local_err
);
3165 error_setg(errp
, "failed while preparing to reopen image '%s'",
3166 reopen_state
->bs
->filename
);
3171 /* It is currently mandatory to have a bdrv_reopen_prepare()
3172 * handler for each supported drv. */
3173 error_setg(errp
, "Block format '%s' used by node '%s' "
3174 "does not support reopening files", drv
->format_name
,
3175 bdrv_get_device_or_node_name(reopen_state
->bs
));
3180 /* Options that are not handled are only okay if they are unchanged
3181 * compared to the old state. It is expected that some options are only
3182 * used for the initial open, but not reopen (e.g. filename) */
3183 if (qdict_size(reopen_state
->options
)) {
3184 const QDictEntry
*entry
= qdict_first(reopen_state
->options
);
3187 QObject
*new = entry
->value
;
3188 QObject
*old
= qdict_get(reopen_state
->bs
->options
, entry
->key
);
3191 * TODO: When using -drive to specify blockdev options, all values
3192 * will be strings; however, when using -blockdev, blockdev-add or
3193 * filenames using the json:{} pseudo-protocol, they will be
3195 * In contrast, reopening options are (currently) always strings
3196 * (because you can only specify them through qemu-io; all other
3197 * callers do not specify any options).
3198 * Therefore, when using anything other than -drive to create a BDS,
3199 * this cannot detect non-string options as unchanged, because
3200 * qobject_is_equal() always returns false for objects of different
3201 * type. In the future, this should be remedied by correctly typing
3202 * all options. For now, this is not too big of an issue because
3203 * the user can simply omit options which cannot be changed anyway,
3204 * so they will stay unchanged.
3206 if (!qobject_is_equal(new, old
)) {
3207 error_setg(errp
, "Cannot change the option '%s'", entry
->key
);
3211 } while ((entry
= qdict_next(reopen_state
->options
, entry
)));
3214 ret
= bdrv_check_perm(reopen_state
->bs
, queue
, reopen_state
->perm
,
3215 reopen_state
->shared_perm
, NULL
, errp
);
3223 qemu_opts_del(opts
);
3228 * Takes the staged changes for the reopen from bdrv_reopen_prepare(), and
3229 * makes them final by swapping the staging BlockDriverState contents into
3230 * the active BlockDriverState contents.
3232 void bdrv_reopen_commit(BDRVReopenState
*reopen_state
)
3235 BlockDriverState
*bs
;
3236 bool old_can_write
, new_can_write
;
3238 assert(reopen_state
!= NULL
);
3239 bs
= reopen_state
->bs
;
3241 assert(drv
!= NULL
);
3244 !bdrv_is_read_only(bs
) && !(bdrv_get_flags(bs
) & BDRV_O_INACTIVE
);
3246 /* If there are any driver level actions to take */
3247 if (drv
->bdrv_reopen_commit
) {
3248 drv
->bdrv_reopen_commit(reopen_state
);
3251 /* set BDS specific flags now */
3252 QDECREF(bs
->explicit_options
);
3254 bs
->explicit_options
= reopen_state
->explicit_options
;
3255 bs
->open_flags
= reopen_state
->flags
;
3256 bs
->read_only
= !(reopen_state
->flags
& BDRV_O_RDWR
);
3258 bdrv_refresh_limits(bs
, NULL
);
3260 bdrv_set_perm(reopen_state
->bs
, reopen_state
->perm
,
3261 reopen_state
->shared_perm
);
3264 !bdrv_is_read_only(bs
) && !(bdrv_get_flags(bs
) & BDRV_O_INACTIVE
);
3265 if (!old_can_write
&& new_can_write
&& drv
->bdrv_reopen_bitmaps_rw
) {
3266 Error
*local_err
= NULL
;
3267 if (drv
->bdrv_reopen_bitmaps_rw(bs
, &local_err
) < 0) {
3268 /* This is not fatal, bitmaps just left read-only, so all following
3269 * writes will fail. User can remove read-only bitmaps to unblock
3272 error_reportf_err(local_err
,
3273 "%s: Failed to make dirty bitmaps writable: ",
3274 bdrv_get_node_name(bs
));
3280 * Abort the reopen, and delete and free the staged changes in
3283 void bdrv_reopen_abort(BDRVReopenState
*reopen_state
)
3287 assert(reopen_state
!= NULL
);
3288 drv
= reopen_state
->bs
->drv
;
3289 assert(drv
!= NULL
);
3291 if (drv
->bdrv_reopen_abort
) {
3292 drv
->bdrv_reopen_abort(reopen_state
);
3295 QDECREF(reopen_state
->explicit_options
);
3297 bdrv_abort_perm_update(reopen_state
->bs
);
3301 static void bdrv_close(BlockDriverState
*bs
)
3303 BdrvAioNotifier
*ban
, *ban_next
;
3304 BdrvChild
*child
, *next
;
3307 assert(!bs
->refcnt
);
3309 bdrv_drained_begin(bs
); /* complete I/O */
3311 bdrv_drain(bs
); /* in case flush left pending I/O */
3314 bs
->drv
->bdrv_close(bs
);
3318 bdrv_set_backing_hd(bs
, NULL
, &error_abort
);
3320 if (bs
->file
!= NULL
) {
3321 bdrv_unref_child(bs
, bs
->file
);
3325 QLIST_FOREACH_SAFE(child
, &bs
->children
, next
, next
) {
3326 /* TODO Remove bdrv_unref() from drivers' close function and use
3327 * bdrv_unref_child() here */
3328 if (child
->bs
->inherits_from
== bs
) {
3329 child
->bs
->inherits_from
= NULL
;
3331 bdrv_detach_child(child
);
3336 atomic_set(&bs
->copy_on_read
, 0);
3337 bs
->backing_file
[0] = '\0';
3338 bs
->backing_format
[0] = '\0';
3339 bs
->total_sectors
= 0;
3340 bs
->encrypted
= false;
3342 QDECREF(bs
->options
);
3343 QDECREF(bs
->explicit_options
);
3345 bs
->explicit_options
= NULL
;
3346 QDECREF(bs
->full_open_options
);
3347 bs
->full_open_options
= NULL
;
3349 bdrv_release_named_dirty_bitmaps(bs
);
3350 assert(QLIST_EMPTY(&bs
->dirty_bitmaps
));
3352 QLIST_FOREACH_SAFE(ban
, &bs
->aio_notifiers
, list
, ban_next
) {
3355 QLIST_INIT(&bs
->aio_notifiers
);
3356 bdrv_drained_end(bs
);
3359 void bdrv_close_all(void)
3361 block_job_cancel_sync_all();
3362 nbd_export_close_all();
3364 /* Drop references from requests still in flight, such as canceled block
3365 * jobs whose AIO context has not been polled yet */
3368 blk_remove_all_bs();
3369 blockdev_close_all_bdrv_states();
3371 assert(QTAILQ_EMPTY(&all_bdrv_states
));
3374 static bool should_update_child(BdrvChild
*c
, BlockDriverState
*to
)
3378 if (c
->role
->stay_at_node
) {
3382 if (c
->role
== &child_backing
) {
3383 /* If @from is a backing file of @to, ignore the child to avoid
3384 * creating a loop. We only want to change the pointer of other
3386 QLIST_FOREACH(to_c
, &to
->children
, next
) {
3399 void bdrv_replace_node(BlockDriverState
*from
, BlockDriverState
*to
,
3402 BdrvChild
*c
, *next
;
3403 GSList
*list
= NULL
, *p
;
3404 uint64_t old_perm
, old_shared
;
3405 uint64_t perm
= 0, shared
= BLK_PERM_ALL
;
3408 assert(!atomic_read(&from
->in_flight
));
3409 assert(!atomic_read(&to
->in_flight
));
3411 /* Make sure that @from doesn't go away until we have successfully attached
3412 * all of its parents to @to. */
3415 /* Put all parents into @list and calculate their cumulative permissions */
3416 QLIST_FOREACH_SAFE(c
, &from
->parents
, next_parent
, next
) {
3417 if (!should_update_child(c
, to
)) {
3420 list
= g_slist_prepend(list
, c
);
3422 shared
&= c
->shared_perm
;
3425 /* Check whether the required permissions can be granted on @to, ignoring
3426 * all BdrvChild in @list so that they can't block themselves. */
3427 ret
= bdrv_check_update_perm(to
, NULL
, perm
, shared
, list
, errp
);
3429 bdrv_abort_perm_update(to
);
3433 /* Now actually perform the change. We performed the permission check for
3434 * all elements of @list at once, so set the permissions all at once at the
3436 for (p
= list
; p
!= NULL
; p
= p
->next
) {
3440 bdrv_replace_child_noperm(c
, to
);
3444 bdrv_get_cumulative_perm(to
, &old_perm
, &old_shared
);
3445 bdrv_set_perm(to
, old_perm
| perm
, old_shared
| shared
);
3453 * Add new bs contents at the top of an image chain while the chain is
3454 * live, while keeping required fields on the top layer.
3456 * This will modify the BlockDriverState fields, and swap contents
3457 * between bs_new and bs_top. Both bs_new and bs_top are modified.
3459 * bs_new must not be attached to a BlockBackend.
3461 * This function does not create any image files.
3463 * bdrv_append() takes ownership of a bs_new reference and unrefs it because
3464 * that's what the callers commonly need. bs_new will be referenced by the old
3465 * parents of bs_top after bdrv_append() returns. If the caller needs to keep a
3466 * reference of its own, it must call bdrv_ref().
3468 void bdrv_append(BlockDriverState
*bs_new
, BlockDriverState
*bs_top
,
3471 Error
*local_err
= NULL
;
3473 bdrv_set_backing_hd(bs_new
, bs_top
, &local_err
);
3475 error_propagate(errp
, local_err
);
3479 bdrv_replace_node(bs_top
, bs_new
, &local_err
);
3481 error_propagate(errp
, local_err
);
3482 bdrv_set_backing_hd(bs_new
, NULL
, &error_abort
);
3486 /* bs_new is now referenced by its new parents, we don't need the
3487 * additional reference any more. */
3492 static void bdrv_delete(BlockDriverState
*bs
)
3495 assert(bdrv_op_blocker_is_empty(bs
));
3496 assert(!bs
->refcnt
);
3500 /* remove from list, if necessary */
3501 if (bs
->node_name
[0] != '\0') {
3502 QTAILQ_REMOVE(&graph_bdrv_states
, bs
, node_list
);
3504 QTAILQ_REMOVE(&all_bdrv_states
, bs
, bs_list
);
3510 * Run consistency checks on an image
3512 * Returns 0 if the check could be completed (it doesn't mean that the image is
3513 * free of errors) or -errno when an internal error occurred. The results of the
3514 * check are stored in res.
3516 static int coroutine_fn
bdrv_co_check(BlockDriverState
*bs
,
3517 BdrvCheckResult
*res
, BdrvCheckMode fix
)
3519 if (bs
->drv
== NULL
) {
3522 if (bs
->drv
->bdrv_co_check
== NULL
) {
3526 memset(res
, 0, sizeof(*res
));
3527 return bs
->drv
->bdrv_co_check(bs
, res
, fix
);
3530 typedef struct CheckCo
{
3531 BlockDriverState
*bs
;
3532 BdrvCheckResult
*res
;
3537 static void bdrv_check_co_entry(void *opaque
)
3539 CheckCo
*cco
= opaque
;
3540 cco
->ret
= bdrv_co_check(cco
->bs
, cco
->res
, cco
->fix
);
3543 int bdrv_check(BlockDriverState
*bs
,
3544 BdrvCheckResult
*res
, BdrvCheckMode fix
)
3550 .ret
= -EINPROGRESS
,
3554 if (qemu_in_coroutine()) {
3555 /* Fast-path if already in coroutine context */
3556 bdrv_check_co_entry(&cco
);
3558 co
= qemu_coroutine_create(bdrv_check_co_entry
, &cco
);
3559 qemu_coroutine_enter(co
);
3560 BDRV_POLL_WHILE(bs
, cco
.ret
== -EINPROGRESS
);
3569 * -EINVAL - backing format specified, but no file
3570 * -ENOSPC - can't update the backing file because no space is left in the
3572 * -ENOTSUP - format driver doesn't support changing the backing file
3574 int bdrv_change_backing_file(BlockDriverState
*bs
,
3575 const char *backing_file
, const char *backing_fmt
)
3577 BlockDriver
*drv
= bs
->drv
;
3584 /* Backing file format doesn't make sense without a backing file */
3585 if (backing_fmt
&& !backing_file
) {
3589 if (drv
->bdrv_change_backing_file
!= NULL
) {
3590 ret
= drv
->bdrv_change_backing_file(bs
, backing_file
, backing_fmt
);
3596 pstrcpy(bs
->backing_file
, sizeof(bs
->backing_file
), backing_file
?: "");
3597 pstrcpy(bs
->backing_format
, sizeof(bs
->backing_format
), backing_fmt
?: "");
3603 * Finds the image layer in the chain that has 'bs' as its backing file.
3605 * active is the current topmost image.
3607 * Returns NULL if bs is not found in active's image chain,
3608 * or if active == bs.
3610 * Returns the bottommost base image if bs == NULL.
3612 BlockDriverState
*bdrv_find_overlay(BlockDriverState
*active
,
3613 BlockDriverState
*bs
)
3615 while (active
&& bs
!= backing_bs(active
)) {
3616 active
= backing_bs(active
);
3622 /* Given a BDS, searches for the base layer. */
3623 BlockDriverState
*bdrv_find_base(BlockDriverState
*bs
)
3625 return bdrv_find_overlay(bs
, NULL
);
3629 * Drops images above 'base' up to and including 'top', and sets the image
3630 * above 'top' to have base as its backing file.
3632 * Requires that the overlay to 'top' is opened r/w, so that the backing file
3633 * information in 'bs' can be properly updated.
3635 * E.g., this will convert the following chain:
3636 * bottom <- base <- intermediate <- top <- active
3640 * bottom <- base <- active
3642 * It is allowed for bottom==base, in which case it converts:
3644 * base <- intermediate <- top <- active
3650 * If backing_file_str is non-NULL, it will be used when modifying top's
3651 * overlay image metadata.
3654 * if active == top, that is considered an error
3657 int bdrv_drop_intermediate(BlockDriverState
*top
, BlockDriverState
*base
,
3658 const char *backing_file_str
)
3660 BdrvChild
*c
, *next
;
3661 Error
*local_err
= NULL
;
3666 if (!top
->drv
|| !base
->drv
) {
3670 /* Make sure that base is in the backing chain of top */
3671 if (!bdrv_chain_contains(top
, base
)) {
3675 /* success - we can delete the intermediate states, and link top->base */
3676 /* TODO Check graph modification op blockers (BLK_PERM_GRAPH_MOD) once
3677 * we've figured out how they should work. */
3678 backing_file_str
= backing_file_str
? backing_file_str
: base
->filename
;
3680 QLIST_FOREACH_SAFE(c
, &top
->parents
, next_parent
, next
) {
3681 /* Check whether we are allowed to switch c from top to base */
3682 GSList
*ignore_children
= g_slist_prepend(NULL
, c
);
3683 bdrv_check_update_perm(base
, NULL
, c
->perm
, c
->shared_perm
,
3684 ignore_children
, &local_err
);
3685 g_slist_free(ignore_children
);
3688 error_report_err(local_err
);
3692 /* If so, update the backing file path in the image file */
3693 if (c
->role
->update_filename
) {
3694 ret
= c
->role
->update_filename(c
, base
, backing_file_str
,
3697 bdrv_abort_perm_update(base
);
3698 error_report_err(local_err
);
3703 /* Do the actual switch in the in-memory graph.
3704 * Completes bdrv_check_update_perm() transaction internally. */
3706 bdrv_replace_child(c
, base
);
3717 * Truncate file to 'offset' bytes (needed only for file protocols)
3719 int bdrv_truncate(BdrvChild
*child
, int64_t offset
, PreallocMode prealloc
,
3722 BlockDriverState
*bs
= child
->bs
;
3723 BlockDriver
*drv
= bs
->drv
;
3726 assert(child
->perm
& BLK_PERM_RESIZE
);
3728 /* if bs->drv == NULL, bs is closed, so there's nothing to do here */
3730 error_setg(errp
, "No medium inserted");
3734 error_setg(errp
, "Image size cannot be negative");
3738 if (!drv
->bdrv_truncate
) {
3739 if (bs
->file
&& drv
->is_filter
) {
3740 return bdrv_truncate(bs
->file
, offset
, prealloc
, errp
);
3742 error_setg(errp
, "Image format driver does not support resize");
3745 if (bs
->read_only
) {
3746 error_setg(errp
, "Image is read-only");
3750 assert(!(bs
->open_flags
& BDRV_O_INACTIVE
));
3752 ret
= drv
->bdrv_truncate(bs
, offset
, prealloc
, errp
);
3756 ret
= refresh_total_sectors(bs
, offset
>> BDRV_SECTOR_BITS
);
3758 error_setg_errno(errp
, -ret
, "Could not refresh total sector count");
3760 offset
= bs
->total_sectors
* BDRV_SECTOR_SIZE
;
3762 bdrv_dirty_bitmap_truncate(bs
, offset
);
3763 bdrv_parent_cb_resize(bs
);
3764 atomic_inc(&bs
->write_gen
);
3769 * Length of a allocated file in bytes. Sparse files are counted by actual
3770 * allocated space. Return < 0 if error or unknown.
3772 int64_t bdrv_get_allocated_file_size(BlockDriverState
*bs
)
3774 BlockDriver
*drv
= bs
->drv
;
3778 if (drv
->bdrv_get_allocated_file_size
) {
3779 return drv
->bdrv_get_allocated_file_size(bs
);
3782 return bdrv_get_allocated_file_size(bs
->file
->bs
);
3789 * @drv: Format driver
3790 * @opts: Creation options for new image
3791 * @in_bs: Existing image containing data for new image (may be NULL)
3792 * @errp: Error object
3793 * Returns: A #BlockMeasureInfo (free using qapi_free_BlockMeasureInfo())
3796 * Calculate file size required to create a new image.
3798 * If @in_bs is given then space for allocated clusters and zero clusters
3799 * from that image are included in the calculation. If @opts contains a
3800 * backing file that is shared by @in_bs then backing clusters may be omitted
3801 * from the calculation.
3803 * If @in_bs is NULL then the calculation includes no allocated clusters
3804 * unless a preallocation option is given in @opts.
3806 * Note that @in_bs may use a different BlockDriver from @drv.
3808 * If an error occurs the @errp pointer is set.
3810 BlockMeasureInfo
*bdrv_measure(BlockDriver
*drv
, QemuOpts
*opts
,
3811 BlockDriverState
*in_bs
, Error
**errp
)
3813 if (!drv
->bdrv_measure
) {
3814 error_setg(errp
, "Block driver '%s' does not support size measurement",
3819 return drv
->bdrv_measure(opts
, in_bs
, errp
);
3823 * Return number of sectors on success, -errno on error.
3825 int64_t bdrv_nb_sectors(BlockDriverState
*bs
)
3827 BlockDriver
*drv
= bs
->drv
;
3832 if (drv
->has_variable_length
) {
3833 int ret
= refresh_total_sectors(bs
, bs
->total_sectors
);
3838 return bs
->total_sectors
;
3842 * Return length in bytes on success, -errno on error.
3843 * The length is always a multiple of BDRV_SECTOR_SIZE.
3845 int64_t bdrv_getlength(BlockDriverState
*bs
)
3847 int64_t ret
= bdrv_nb_sectors(bs
);
3849 ret
= ret
> INT64_MAX
/ BDRV_SECTOR_SIZE
? -EFBIG
: ret
;
3850 return ret
< 0 ? ret
: ret
* BDRV_SECTOR_SIZE
;
3853 /* return 0 as number of sectors if no device present or error */
3854 void bdrv_get_geometry(BlockDriverState
*bs
, uint64_t *nb_sectors_ptr
)
3856 int64_t nb_sectors
= bdrv_nb_sectors(bs
);
3858 *nb_sectors_ptr
= nb_sectors
< 0 ? 0 : nb_sectors
;
3861 bool bdrv_is_sg(BlockDriverState
*bs
)
3866 bool bdrv_is_encrypted(BlockDriverState
*bs
)
3868 if (bs
->backing
&& bs
->backing
->bs
->encrypted
) {
3871 return bs
->encrypted
;
3874 const char *bdrv_get_format_name(BlockDriverState
*bs
)
3876 return bs
->drv
? bs
->drv
->format_name
: NULL
;
3879 static int qsort_strcmp(const void *a
, const void *b
)
3881 return strcmp(*(char *const *)a
, *(char *const *)b
);
3884 void bdrv_iterate_format(void (*it
)(void *opaque
, const char *name
),
3890 const char **formats
= NULL
;
3892 QLIST_FOREACH(drv
, &bdrv_drivers
, list
) {
3893 if (drv
->format_name
) {
3896 while (formats
&& i
&& !found
) {
3897 found
= !strcmp(formats
[--i
], drv
->format_name
);
3901 formats
= g_renew(const char *, formats
, count
+ 1);
3902 formats
[count
++] = drv
->format_name
;
3907 for (i
= 0; i
< (int)ARRAY_SIZE(block_driver_modules
); i
++) {
3908 const char *format_name
= block_driver_modules
[i
].format_name
;
3914 while (formats
&& j
&& !found
) {
3915 found
= !strcmp(formats
[--j
], format_name
);
3919 formats
= g_renew(const char *, formats
, count
+ 1);
3920 formats
[count
++] = format_name
;
3925 qsort(formats
, count
, sizeof(formats
[0]), qsort_strcmp
);
3927 for (i
= 0; i
< count
; i
++) {
3928 it(opaque
, formats
[i
]);
3934 /* This function is to find a node in the bs graph */
3935 BlockDriverState
*bdrv_find_node(const char *node_name
)
3937 BlockDriverState
*bs
;
3941 QTAILQ_FOREACH(bs
, &graph_bdrv_states
, node_list
) {
3942 if (!strcmp(node_name
, bs
->node_name
)) {
3949 /* Put this QMP function here so it can access the static graph_bdrv_states. */
3950 BlockDeviceInfoList
*bdrv_named_nodes_list(Error
**errp
)
3952 BlockDeviceInfoList
*list
, *entry
;
3953 BlockDriverState
*bs
;
3956 QTAILQ_FOREACH(bs
, &graph_bdrv_states
, node_list
) {
3957 BlockDeviceInfo
*info
= bdrv_block_device_info(NULL
, bs
, errp
);
3959 qapi_free_BlockDeviceInfoList(list
);
3962 entry
= g_malloc0(sizeof(*entry
));
3963 entry
->value
= info
;
3971 BlockDriverState
*bdrv_lookup_bs(const char *device
,
3972 const char *node_name
,
3976 BlockDriverState
*bs
;
3979 blk
= blk_by_name(device
);
3984 error_setg(errp
, "Device '%s' has no medium", device
);
3992 bs
= bdrv_find_node(node_name
);
3999 error_setg(errp
, "Cannot find device=%s nor node_name=%s",
4000 device
? device
: "",
4001 node_name
? node_name
: "");
4005 /* If 'base' is in the same chain as 'top', return true. Otherwise,
4006 * return false. If either argument is NULL, return false. */
4007 bool bdrv_chain_contains(BlockDriverState
*top
, BlockDriverState
*base
)
4009 while (top
&& top
!= base
) {
4010 top
= backing_bs(top
);
4016 BlockDriverState
*bdrv_next_node(BlockDriverState
*bs
)
4019 return QTAILQ_FIRST(&graph_bdrv_states
);
4021 return QTAILQ_NEXT(bs
, node_list
);
4024 const char *bdrv_get_node_name(const BlockDriverState
*bs
)
4026 return bs
->node_name
;
4029 const char *bdrv_get_parent_name(const BlockDriverState
*bs
)
4034 /* If multiple parents have a name, just pick the first one. */
4035 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
4036 if (c
->role
->get_name
) {
4037 name
= c
->role
->get_name(c
);
4038 if (name
&& *name
) {
4047 /* TODO check what callers really want: bs->node_name or blk_name() */
4048 const char *bdrv_get_device_name(const BlockDriverState
*bs
)
4050 return bdrv_get_parent_name(bs
) ?: "";
4053 /* This can be used to identify nodes that might not have a device
4054 * name associated. Since node and device names live in the same
4055 * namespace, the result is unambiguous. The exception is if both are
4056 * absent, then this returns an empty (non-null) string. */
4057 const char *bdrv_get_device_or_node_name(const BlockDriverState
*bs
)
4059 return bdrv_get_parent_name(bs
) ?: bs
->node_name
;
4062 int bdrv_get_flags(BlockDriverState
*bs
)
4064 return bs
->open_flags
;
4067 int bdrv_has_zero_init_1(BlockDriverState
*bs
)
4072 int bdrv_has_zero_init(BlockDriverState
*bs
)
4078 /* If BS is a copy on write image, it is initialized to
4079 the contents of the base image, which may not be zeroes. */
4083 if (bs
->drv
->bdrv_has_zero_init
) {
4084 return bs
->drv
->bdrv_has_zero_init(bs
);
4086 if (bs
->file
&& bs
->drv
->is_filter
) {
4087 return bdrv_has_zero_init(bs
->file
->bs
);
4094 bool bdrv_unallocated_blocks_are_zero(BlockDriverState
*bs
)
4096 BlockDriverInfo bdi
;
4102 if (bdrv_get_info(bs
, &bdi
) == 0) {
4103 return bdi
.unallocated_blocks_are_zero
;
4109 bool bdrv_can_write_zeroes_with_unmap(BlockDriverState
*bs
)
4111 if (!(bs
->open_flags
& BDRV_O_UNMAP
)) {
4115 return bs
->supported_zero_flags
& BDRV_REQ_MAY_UNMAP
;
4118 const char *bdrv_get_encrypted_filename(BlockDriverState
*bs
)
4120 if (bs
->backing
&& bs
->backing
->bs
->encrypted
)
4121 return bs
->backing_file
;
4122 else if (bs
->encrypted
)
4123 return bs
->filename
;
4128 void bdrv_get_backing_filename(BlockDriverState
*bs
,
4129 char *filename
, int filename_size
)
4131 pstrcpy(filename
, filename_size
, bs
->backing_file
);
4134 int bdrv_get_info(BlockDriverState
*bs
, BlockDriverInfo
*bdi
)
4136 BlockDriver
*drv
= bs
->drv
;
4137 /* if bs->drv == NULL, bs is closed, so there's nothing to do here */
4141 if (!drv
->bdrv_get_info
) {
4142 if (bs
->file
&& drv
->is_filter
) {
4143 return bdrv_get_info(bs
->file
->bs
, bdi
);
4147 memset(bdi
, 0, sizeof(*bdi
));
4148 return drv
->bdrv_get_info(bs
, bdi
);
4151 ImageInfoSpecific
*bdrv_get_specific_info(BlockDriverState
*bs
)
4153 BlockDriver
*drv
= bs
->drv
;
4154 if (drv
&& drv
->bdrv_get_specific_info
) {
4155 return drv
->bdrv_get_specific_info(bs
);
4160 void bdrv_debug_event(BlockDriverState
*bs
, BlkdebugEvent event
)
4162 if (!bs
|| !bs
->drv
|| !bs
->drv
->bdrv_debug_event
) {
4166 bs
->drv
->bdrv_debug_event(bs
, event
);
4169 int bdrv_debug_breakpoint(BlockDriverState
*bs
, const char *event
,
4172 while (bs
&& bs
->drv
&& !bs
->drv
->bdrv_debug_breakpoint
) {
4173 bs
= bs
->file
? bs
->file
->bs
: NULL
;
4176 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_breakpoint
) {
4177 return bs
->drv
->bdrv_debug_breakpoint(bs
, event
, tag
);
4183 int bdrv_debug_remove_breakpoint(BlockDriverState
*bs
, const char *tag
)
4185 while (bs
&& bs
->drv
&& !bs
->drv
->bdrv_debug_remove_breakpoint
) {
4186 bs
= bs
->file
? bs
->file
->bs
: NULL
;
4189 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_remove_breakpoint
) {
4190 return bs
->drv
->bdrv_debug_remove_breakpoint(bs
, tag
);
4196 int bdrv_debug_resume(BlockDriverState
*bs
, const char *tag
)
4198 while (bs
&& (!bs
->drv
|| !bs
->drv
->bdrv_debug_resume
)) {
4199 bs
= bs
->file
? bs
->file
->bs
: NULL
;
4202 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_resume
) {
4203 return bs
->drv
->bdrv_debug_resume(bs
, tag
);
4209 bool bdrv_debug_is_suspended(BlockDriverState
*bs
, const char *tag
)
4211 while (bs
&& bs
->drv
&& !bs
->drv
->bdrv_debug_is_suspended
) {
4212 bs
= bs
->file
? bs
->file
->bs
: NULL
;
4215 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_is_suspended
) {
4216 return bs
->drv
->bdrv_debug_is_suspended(bs
, tag
);
4222 /* backing_file can either be relative, or absolute, or a protocol. If it is
4223 * relative, it must be relative to the chain. So, passing in bs->filename
4224 * from a BDS as backing_file should not be done, as that may be relative to
4225 * the CWD rather than the chain. */
4226 BlockDriverState
*bdrv_find_backing_image(BlockDriverState
*bs
,
4227 const char *backing_file
)
4229 char *filename_full
= NULL
;
4230 char *backing_file_full
= NULL
;
4231 char *filename_tmp
= NULL
;
4232 int is_protocol
= 0;
4233 BlockDriverState
*curr_bs
= NULL
;
4234 BlockDriverState
*retval
= NULL
;
4235 Error
*local_error
= NULL
;
4237 if (!bs
|| !bs
->drv
|| !backing_file
) {
4241 filename_full
= g_malloc(PATH_MAX
);
4242 backing_file_full
= g_malloc(PATH_MAX
);
4243 filename_tmp
= g_malloc(PATH_MAX
);
4245 is_protocol
= path_has_protocol(backing_file
);
4247 for (curr_bs
= bs
; curr_bs
->backing
; curr_bs
= curr_bs
->backing
->bs
) {
4249 /* If either of the filename paths is actually a protocol, then
4250 * compare unmodified paths; otherwise make paths relative */
4251 if (is_protocol
|| path_has_protocol(curr_bs
->backing_file
)) {
4252 if (strcmp(backing_file
, curr_bs
->backing_file
) == 0) {
4253 retval
= curr_bs
->backing
->bs
;
4256 /* Also check against the full backing filename for the image */
4257 bdrv_get_full_backing_filename(curr_bs
, backing_file_full
, PATH_MAX
,
4259 if (local_error
== NULL
) {
4260 if (strcmp(backing_file
, backing_file_full
) == 0) {
4261 retval
= curr_bs
->backing
->bs
;
4265 error_free(local_error
);
4269 /* If not an absolute filename path, make it relative to the current
4270 * image's filename path */
4271 path_combine(filename_tmp
, PATH_MAX
, curr_bs
->filename
,
4274 /* We are going to compare absolute pathnames */
4275 if (!realpath(filename_tmp
, filename_full
)) {
4279 /* We need to make sure the backing filename we are comparing against
4280 * is relative to the current image filename (or absolute) */
4281 path_combine(filename_tmp
, PATH_MAX
, curr_bs
->filename
,
4282 curr_bs
->backing_file
);
4284 if (!realpath(filename_tmp
, backing_file_full
)) {
4288 if (strcmp(backing_file_full
, filename_full
) == 0) {
4289 retval
= curr_bs
->backing
->bs
;
4295 g_free(filename_full
);
4296 g_free(backing_file_full
);
4297 g_free(filename_tmp
);
4301 void bdrv_init(void)
4303 module_call_init(MODULE_INIT_BLOCK
);
4306 void bdrv_init_with_whitelist(void)
4308 use_bdrv_whitelist
= 1;
4312 static void coroutine_fn
bdrv_co_invalidate_cache(BlockDriverState
*bs
,
4315 BdrvChild
*child
, *parent
;
4316 uint64_t perm
, shared_perm
;
4317 Error
*local_err
= NULL
;
4324 if (!(bs
->open_flags
& BDRV_O_INACTIVE
)) {
4328 QLIST_FOREACH(child
, &bs
->children
, next
) {
4329 bdrv_co_invalidate_cache(child
->bs
, &local_err
);
4331 error_propagate(errp
, local_err
);
4337 * Update permissions, they may differ for inactive nodes.
4339 * Note that the required permissions of inactive images are always a
4340 * subset of the permissions required after activating the image. This
4341 * allows us to just get the permissions upfront without restricting
4342 * drv->bdrv_invalidate_cache().
4344 * It also means that in error cases, we don't have to try and revert to
4345 * the old permissions (which is an operation that could fail, too). We can
4346 * just keep the extended permissions for the next time that an activation
4347 * of the image is tried.
4349 bs
->open_flags
&= ~BDRV_O_INACTIVE
;
4350 bdrv_get_cumulative_perm(bs
, &perm
, &shared_perm
);
4351 ret
= bdrv_check_perm(bs
, NULL
, perm
, shared_perm
, NULL
, &local_err
);
4353 bs
->open_flags
|= BDRV_O_INACTIVE
;
4354 error_propagate(errp
, local_err
);
4357 bdrv_set_perm(bs
, perm
, shared_perm
);
4359 if (bs
->drv
->bdrv_co_invalidate_cache
) {
4360 bs
->drv
->bdrv_co_invalidate_cache(bs
, &local_err
);
4362 bs
->open_flags
|= BDRV_O_INACTIVE
;
4363 error_propagate(errp
, local_err
);
4368 ret
= refresh_total_sectors(bs
, bs
->total_sectors
);
4370 bs
->open_flags
|= BDRV_O_INACTIVE
;
4371 error_setg_errno(errp
, -ret
, "Could not refresh total sector count");
4375 QLIST_FOREACH(parent
, &bs
->parents
, next_parent
) {
4376 if (parent
->role
->activate
) {
4377 parent
->role
->activate(parent
, &local_err
);
4379 error_propagate(errp
, local_err
);
4386 typedef struct InvalidateCacheCo
{
4387 BlockDriverState
*bs
;
4390 } InvalidateCacheCo
;
4392 static void coroutine_fn
bdrv_invalidate_cache_co_entry(void *opaque
)
4394 InvalidateCacheCo
*ico
= opaque
;
4395 bdrv_co_invalidate_cache(ico
->bs
, ico
->errp
);
4399 void bdrv_invalidate_cache(BlockDriverState
*bs
, Error
**errp
)
4402 InvalidateCacheCo ico
= {
4408 if (qemu_in_coroutine()) {
4409 /* Fast-path if already in coroutine context */
4410 bdrv_invalidate_cache_co_entry(&ico
);
4412 co
= qemu_coroutine_create(bdrv_invalidate_cache_co_entry
, &ico
);
4413 qemu_coroutine_enter(co
);
4414 BDRV_POLL_WHILE(bs
, !ico
.done
);
4418 void bdrv_invalidate_cache_all(Error
**errp
)
4420 BlockDriverState
*bs
;
4421 Error
*local_err
= NULL
;
4422 BdrvNextIterator it
;
4424 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
4425 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
4427 aio_context_acquire(aio_context
);
4428 bdrv_invalidate_cache(bs
, &local_err
);
4429 aio_context_release(aio_context
);
4431 error_propagate(errp
, local_err
);
4432 bdrv_next_cleanup(&it
);
4438 static int bdrv_inactivate_recurse(BlockDriverState
*bs
,
4441 BdrvChild
*child
, *parent
;
4448 if (!setting_flag
&& bs
->drv
->bdrv_inactivate
) {
4449 ret
= bs
->drv
->bdrv_inactivate(bs
);
4455 if (setting_flag
&& !(bs
->open_flags
& BDRV_O_INACTIVE
)) {
4456 uint64_t perm
, shared_perm
;
4458 QLIST_FOREACH(parent
, &bs
->parents
, next_parent
) {
4459 if (parent
->role
->inactivate
) {
4460 ret
= parent
->role
->inactivate(parent
);
4467 bs
->open_flags
|= BDRV_O_INACTIVE
;
4469 /* Update permissions, they may differ for inactive nodes */
4470 bdrv_get_cumulative_perm(bs
, &perm
, &shared_perm
);
4471 bdrv_check_perm(bs
, NULL
, perm
, shared_perm
, NULL
, &error_abort
);
4472 bdrv_set_perm(bs
, perm
, shared_perm
);
4475 QLIST_FOREACH(child
, &bs
->children
, next
) {
4476 ret
= bdrv_inactivate_recurse(child
->bs
, setting_flag
);
4482 /* At this point persistent bitmaps should be already stored by the format
4484 bdrv_release_persistent_dirty_bitmaps(bs
);
4489 int bdrv_inactivate_all(void)
4491 BlockDriverState
*bs
= NULL
;
4492 BdrvNextIterator it
;
4495 GSList
*aio_ctxs
= NULL
, *ctx
;
4497 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
4498 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
4500 if (!g_slist_find(aio_ctxs
, aio_context
)) {
4501 aio_ctxs
= g_slist_prepend(aio_ctxs
, aio_context
);
4502 aio_context_acquire(aio_context
);
4506 /* We do two passes of inactivation. The first pass calls to drivers'
4507 * .bdrv_inactivate callbacks recursively so all cache is flushed to disk;
4508 * the second pass sets the BDRV_O_INACTIVE flag so that no further write
4510 for (pass
= 0; pass
< 2; pass
++) {
4511 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
4512 ret
= bdrv_inactivate_recurse(bs
, pass
);
4514 bdrv_next_cleanup(&it
);
4521 for (ctx
= aio_ctxs
; ctx
!= NULL
; ctx
= ctx
->next
) {
4522 AioContext
*aio_context
= ctx
->data
;
4523 aio_context_release(aio_context
);
4525 g_slist_free(aio_ctxs
);
4530 /**************************************************************/
4531 /* removable device support */
4534 * Return TRUE if the media is present
4536 bool bdrv_is_inserted(BlockDriverState
*bs
)
4538 BlockDriver
*drv
= bs
->drv
;
4544 if (drv
->bdrv_is_inserted
) {
4545 return drv
->bdrv_is_inserted(bs
);
4547 QLIST_FOREACH(child
, &bs
->children
, next
) {
4548 if (!bdrv_is_inserted(child
->bs
)) {
4556 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
4558 void bdrv_eject(BlockDriverState
*bs
, bool eject_flag
)
4560 BlockDriver
*drv
= bs
->drv
;
4562 if (drv
&& drv
->bdrv_eject
) {
4563 drv
->bdrv_eject(bs
, eject_flag
);
4568 * Lock or unlock the media (if it is locked, the user won't be able
4569 * to eject it manually).
4571 void bdrv_lock_medium(BlockDriverState
*bs
, bool locked
)
4573 BlockDriver
*drv
= bs
->drv
;
4575 trace_bdrv_lock_medium(bs
, locked
);
4577 if (drv
&& drv
->bdrv_lock_medium
) {
4578 drv
->bdrv_lock_medium(bs
, locked
);
4582 /* Get a reference to bs */
4583 void bdrv_ref(BlockDriverState
*bs
)
4588 /* Release a previously grabbed reference to bs.
4589 * If after releasing, reference count is zero, the BlockDriverState is
4591 void bdrv_unref(BlockDriverState
*bs
)
4596 assert(bs
->refcnt
> 0);
4597 if (--bs
->refcnt
== 0) {
4602 struct BdrvOpBlocker
{
4604 QLIST_ENTRY(BdrvOpBlocker
) list
;
4607 bool bdrv_op_is_blocked(BlockDriverState
*bs
, BlockOpType op
, Error
**errp
)
4609 BdrvOpBlocker
*blocker
;
4610 assert((int) op
>= 0 && op
< BLOCK_OP_TYPE_MAX
);
4611 if (!QLIST_EMPTY(&bs
->op_blockers
[op
])) {
4612 blocker
= QLIST_FIRST(&bs
->op_blockers
[op
]);
4613 error_propagate(errp
, error_copy(blocker
->reason
));
4614 error_prepend(errp
, "Node '%s' is busy: ",
4615 bdrv_get_device_or_node_name(bs
));
4621 void bdrv_op_block(BlockDriverState
*bs
, BlockOpType op
, Error
*reason
)
4623 BdrvOpBlocker
*blocker
;
4624 assert((int) op
>= 0 && op
< BLOCK_OP_TYPE_MAX
);
4626 blocker
= g_new0(BdrvOpBlocker
, 1);
4627 blocker
->reason
= reason
;
4628 QLIST_INSERT_HEAD(&bs
->op_blockers
[op
], blocker
, list
);
4631 void bdrv_op_unblock(BlockDriverState
*bs
, BlockOpType op
, Error
*reason
)
4633 BdrvOpBlocker
*blocker
, *next
;
4634 assert((int) op
>= 0 && op
< BLOCK_OP_TYPE_MAX
);
4635 QLIST_FOREACH_SAFE(blocker
, &bs
->op_blockers
[op
], list
, next
) {
4636 if (blocker
->reason
== reason
) {
4637 QLIST_REMOVE(blocker
, list
);
4643 void bdrv_op_block_all(BlockDriverState
*bs
, Error
*reason
)
4646 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
4647 bdrv_op_block(bs
, i
, reason
);
4651 void bdrv_op_unblock_all(BlockDriverState
*bs
, Error
*reason
)
4654 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
4655 bdrv_op_unblock(bs
, i
, reason
);
4659 bool bdrv_op_blocker_is_empty(BlockDriverState
*bs
)
4663 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
4664 if (!QLIST_EMPTY(&bs
->op_blockers
[i
])) {
4671 void bdrv_img_create(const char *filename
, const char *fmt
,
4672 const char *base_filename
, const char *base_fmt
,
4673 char *options
, uint64_t img_size
, int flags
, bool quiet
,
4676 QemuOptsList
*create_opts
= NULL
;
4677 QemuOpts
*opts
= NULL
;
4678 const char *backing_fmt
, *backing_file
;
4680 BlockDriver
*drv
, *proto_drv
;
4681 Error
*local_err
= NULL
;
4684 /* Find driver and parse its options */
4685 drv
= bdrv_find_format(fmt
);
4687 error_setg(errp
, "Unknown file format '%s'", fmt
);
4691 proto_drv
= bdrv_find_protocol(filename
, true, errp
);
4696 if (!drv
->create_opts
) {
4697 error_setg(errp
, "Format driver '%s' does not support image creation",
4702 if (!proto_drv
->create_opts
) {
4703 error_setg(errp
, "Protocol driver '%s' does not support image creation",
4704 proto_drv
->format_name
);
4708 create_opts
= qemu_opts_append(create_opts
, drv
->create_opts
);
4709 create_opts
= qemu_opts_append(create_opts
, proto_drv
->create_opts
);
4711 /* Create parameter list with default values */
4712 opts
= qemu_opts_create(create_opts
, NULL
, 0, &error_abort
);
4713 qemu_opt_set_number(opts
, BLOCK_OPT_SIZE
, img_size
, &error_abort
);
4715 /* Parse -o options */
4717 qemu_opts_do_parse(opts
, options
, NULL
, &local_err
);
4719 error_report_err(local_err
);
4721 error_setg(errp
, "Invalid options for file format '%s'", fmt
);
4726 if (base_filename
) {
4727 qemu_opt_set(opts
, BLOCK_OPT_BACKING_FILE
, base_filename
, &local_err
);
4729 error_setg(errp
, "Backing file not supported for file format '%s'",
4736 qemu_opt_set(opts
, BLOCK_OPT_BACKING_FMT
, base_fmt
, &local_err
);
4738 error_setg(errp
, "Backing file format not supported for file "
4739 "format '%s'", fmt
);
4744 backing_file
= qemu_opt_get(opts
, BLOCK_OPT_BACKING_FILE
);
4746 if (!strcmp(filename
, backing_file
)) {
4747 error_setg(errp
, "Error: Trying to create an image with the "
4748 "same filename as the backing file");
4753 backing_fmt
= qemu_opt_get(opts
, BLOCK_OPT_BACKING_FMT
);
4755 /* The size for the image must always be specified, unless we have a backing
4756 * file and we have not been forbidden from opening it. */
4757 size
= qemu_opt_get_size(opts
, BLOCK_OPT_SIZE
, img_size
);
4758 if (backing_file
&& !(flags
& BDRV_O_NO_BACKING
)) {
4759 BlockDriverState
*bs
;
4760 char *full_backing
= g_new0(char, PATH_MAX
);
4762 QDict
*backing_options
= NULL
;
4764 bdrv_get_full_backing_filename_from_filename(filename
, backing_file
,
4765 full_backing
, PATH_MAX
,
4768 g_free(full_backing
);
4772 /* backing files always opened read-only */
4774 back_flags
&= ~(BDRV_O_RDWR
| BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
);
4776 backing_options
= qdict_new();
4778 qdict_put_str(backing_options
, "driver", backing_fmt
);
4780 qdict_put_bool(backing_options
, BDRV_OPT_FORCE_SHARE
, true);
4782 bs
= bdrv_open(full_backing
, NULL
, backing_options
, back_flags
,
4784 g_free(full_backing
);
4785 if (!bs
&& size
!= -1) {
4786 /* Couldn't open BS, but we have a size, so it's nonfatal */
4787 warn_reportf_err(local_err
,
4788 "Could not verify backing image. "
4789 "This may become an error in future versions.\n");
4792 /* Couldn't open bs, do not have size */
4793 error_append_hint(&local_err
,
4794 "Could not open backing image to determine size.\n");
4798 /* Opened BS, have no size */
4799 size
= bdrv_getlength(bs
);
4801 error_setg_errno(errp
, -size
, "Could not get size of '%s'",
4806 qemu_opt_set_number(opts
, BLOCK_OPT_SIZE
, size
, &error_abort
);
4810 } /* (backing_file && !(flags & BDRV_O_NO_BACKING)) */
4813 error_setg(errp
, "Image creation needs a size parameter");
4818 printf("Formatting '%s', fmt=%s ", filename
, fmt
);
4819 qemu_opts_print(opts
, " ");
4823 ret
= bdrv_create(drv
, filename
, opts
, &local_err
);
4825 if (ret
== -EFBIG
) {
4826 /* This is generally a better message than whatever the driver would
4827 * deliver (especially because of the cluster_size_hint), since that
4828 * is most probably not much different from "image too large". */
4829 const char *cluster_size_hint
= "";
4830 if (qemu_opt_get_size(opts
, BLOCK_OPT_CLUSTER_SIZE
, 0)) {
4831 cluster_size_hint
= " (try using a larger cluster size)";
4833 error_setg(errp
, "The image size is too large for file format '%s'"
4834 "%s", fmt
, cluster_size_hint
);
4835 error_free(local_err
);
4840 qemu_opts_del(opts
);
4841 qemu_opts_free(create_opts
);
4842 error_propagate(errp
, local_err
);
4845 AioContext
*bdrv_get_aio_context(BlockDriverState
*bs
)
4847 return bs
? bs
->aio_context
: qemu_get_aio_context();
4850 AioWait
*bdrv_get_aio_wait(BlockDriverState
*bs
)
4852 return bs
? &bs
->wait
: NULL
;
4855 void bdrv_coroutine_enter(BlockDriverState
*bs
, Coroutine
*co
)
4857 aio_co_enter(bdrv_get_aio_context(bs
), co
);
4860 static void bdrv_do_remove_aio_context_notifier(BdrvAioNotifier
*ban
)
4862 QLIST_REMOVE(ban
, list
);
4866 void bdrv_detach_aio_context(BlockDriverState
*bs
)
4868 BdrvAioNotifier
*baf
, *baf_tmp
;
4875 assert(!bs
->walking_aio_notifiers
);
4876 bs
->walking_aio_notifiers
= true;
4877 QLIST_FOREACH_SAFE(baf
, &bs
->aio_notifiers
, list
, baf_tmp
) {
4879 bdrv_do_remove_aio_context_notifier(baf
);
4881 baf
->detach_aio_context(baf
->opaque
);
4884 /* Never mind iterating again to check for ->deleted. bdrv_close() will
4885 * remove remaining aio notifiers if we aren't called again.
4887 bs
->walking_aio_notifiers
= false;
4889 if (bs
->drv
->bdrv_detach_aio_context
) {
4890 bs
->drv
->bdrv_detach_aio_context(bs
);
4892 QLIST_FOREACH(child
, &bs
->children
, next
) {
4893 bdrv_detach_aio_context(child
->bs
);
4896 bs
->aio_context
= NULL
;
4899 void bdrv_attach_aio_context(BlockDriverState
*bs
,
4900 AioContext
*new_context
)
4902 BdrvAioNotifier
*ban
, *ban_tmp
;
4909 bs
->aio_context
= new_context
;
4911 QLIST_FOREACH(child
, &bs
->children
, next
) {
4912 bdrv_attach_aio_context(child
->bs
, new_context
);
4914 if (bs
->drv
->bdrv_attach_aio_context
) {
4915 bs
->drv
->bdrv_attach_aio_context(bs
, new_context
);
4918 assert(!bs
->walking_aio_notifiers
);
4919 bs
->walking_aio_notifiers
= true;
4920 QLIST_FOREACH_SAFE(ban
, &bs
->aio_notifiers
, list
, ban_tmp
) {
4922 bdrv_do_remove_aio_context_notifier(ban
);
4924 ban
->attached_aio_context(new_context
, ban
->opaque
);
4927 bs
->walking_aio_notifiers
= false;
4930 void bdrv_set_aio_context(BlockDriverState
*bs
, AioContext
*new_context
)
4932 AioContext
*ctx
= bdrv_get_aio_context(bs
);
4934 aio_disable_external(ctx
);
4935 bdrv_parent_drained_begin(bs
, NULL
);
4936 bdrv_drain(bs
); /* ensure there are no in-flight requests */
4938 while (aio_poll(ctx
, false)) {
4939 /* wait for all bottom halves to execute */
4942 bdrv_detach_aio_context(bs
);
4944 /* This function executes in the old AioContext so acquire the new one in
4945 * case it runs in a different thread.
4947 aio_context_acquire(new_context
);
4948 bdrv_attach_aio_context(bs
, new_context
);
4949 bdrv_parent_drained_end(bs
, NULL
);
4950 aio_enable_external(ctx
);
4951 aio_context_release(new_context
);
4954 void bdrv_add_aio_context_notifier(BlockDriverState
*bs
,
4955 void (*attached_aio_context
)(AioContext
*new_context
, void *opaque
),
4956 void (*detach_aio_context
)(void *opaque
), void *opaque
)
4958 BdrvAioNotifier
*ban
= g_new(BdrvAioNotifier
, 1);
4959 *ban
= (BdrvAioNotifier
){
4960 .attached_aio_context
= attached_aio_context
,
4961 .detach_aio_context
= detach_aio_context
,
4965 QLIST_INSERT_HEAD(&bs
->aio_notifiers
, ban
, list
);
4968 void bdrv_remove_aio_context_notifier(BlockDriverState
*bs
,
4969 void (*attached_aio_context
)(AioContext
*,
4971 void (*detach_aio_context
)(void *),
4974 BdrvAioNotifier
*ban
, *ban_next
;
4976 QLIST_FOREACH_SAFE(ban
, &bs
->aio_notifiers
, list
, ban_next
) {
4977 if (ban
->attached_aio_context
== attached_aio_context
&&
4978 ban
->detach_aio_context
== detach_aio_context
&&
4979 ban
->opaque
== opaque
&&
4980 ban
->deleted
== false)
4982 if (bs
->walking_aio_notifiers
) {
4983 ban
->deleted
= true;
4985 bdrv_do_remove_aio_context_notifier(ban
);
4994 int bdrv_amend_options(BlockDriverState
*bs
, QemuOpts
*opts
,
4995 BlockDriverAmendStatusCB
*status_cb
, void *cb_opaque
)
5000 if (!bs
->drv
->bdrv_amend_options
) {
5003 return bs
->drv
->bdrv_amend_options(bs
, opts
, status_cb
, cb_opaque
);
5006 /* This function will be called by the bdrv_recurse_is_first_non_filter method
5007 * of block filter and by bdrv_is_first_non_filter.
5008 * It is used to test if the given bs is the candidate or recurse more in the
5011 bool bdrv_recurse_is_first_non_filter(BlockDriverState
*bs
,
5012 BlockDriverState
*candidate
)
5014 /* return false if basic checks fails */
5015 if (!bs
|| !bs
->drv
) {
5019 /* the code reached a non block filter driver -> check if the bs is
5020 * the same as the candidate. It's the recursion termination condition.
5022 if (!bs
->drv
->is_filter
) {
5023 return bs
== candidate
;
5025 /* Down this path the driver is a block filter driver */
5027 /* If the block filter recursion method is defined use it to recurse down
5030 if (bs
->drv
->bdrv_recurse_is_first_non_filter
) {
5031 return bs
->drv
->bdrv_recurse_is_first_non_filter(bs
, candidate
);
5034 /* the driver is a block filter but don't allow to recurse -> return false
5039 /* This function checks if the candidate is the first non filter bs down it's
5040 * bs chain. Since we don't have pointers to parents it explore all bs chains
5041 * from the top. Some filters can choose not to pass down the recursion.
5043 bool bdrv_is_first_non_filter(BlockDriverState
*candidate
)
5045 BlockDriverState
*bs
;
5046 BdrvNextIterator it
;
5048 /* walk down the bs forest recursively */
5049 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
5052 /* try to recurse in this top level bs */
5053 perm
= bdrv_recurse_is_first_non_filter(bs
, candidate
);
5055 /* candidate is the first non filter */
5057 bdrv_next_cleanup(&it
);
5065 BlockDriverState
*check_to_replace_node(BlockDriverState
*parent_bs
,
5066 const char *node_name
, Error
**errp
)
5068 BlockDriverState
*to_replace_bs
= bdrv_find_node(node_name
);
5069 AioContext
*aio_context
;
5071 if (!to_replace_bs
) {
5072 error_setg(errp
, "Node name '%s' not found", node_name
);
5076 aio_context
= bdrv_get_aio_context(to_replace_bs
);
5077 aio_context_acquire(aio_context
);
5079 if (bdrv_op_is_blocked(to_replace_bs
, BLOCK_OP_TYPE_REPLACE
, errp
)) {
5080 to_replace_bs
= NULL
;
5084 /* We don't want arbitrary node of the BDS chain to be replaced only the top
5085 * most non filter in order to prevent data corruption.
5086 * Another benefit is that this tests exclude backing files which are
5087 * blocked by the backing blockers.
5089 if (!bdrv_recurse_is_first_non_filter(parent_bs
, to_replace_bs
)) {
5090 error_setg(errp
, "Only top most non filter can be replaced");
5091 to_replace_bs
= NULL
;
5096 aio_context_release(aio_context
);
5097 return to_replace_bs
;
5100 static bool append_open_options(QDict
*d
, BlockDriverState
*bs
)
5102 const QDictEntry
*entry
;
5105 bool found_any
= false;
5108 for (entry
= qdict_first(bs
->options
); entry
;
5109 entry
= qdict_next(bs
->options
, entry
))
5111 /* Exclude options for children */
5112 QLIST_FOREACH(child
, &bs
->children
, next
) {
5113 if (strstart(qdict_entry_key(entry
), child
->name
, &p
)
5114 && (!*p
|| *p
== '.'))
5123 /* And exclude all non-driver-specific options */
5124 for (desc
= bdrv_runtime_opts
.desc
; desc
->name
; desc
++) {
5125 if (!strcmp(qdict_entry_key(entry
), desc
->name
)) {
5133 qobject_incref(qdict_entry_value(entry
));
5134 qdict_put_obj(d
, qdict_entry_key(entry
), qdict_entry_value(entry
));
5141 /* Updates the following BDS fields:
5142 * - exact_filename: A filename which may be used for opening a block device
5143 * which (mostly) equals the given BDS (even without any
5144 * other options; so reading and writing must return the same
5145 * results, but caching etc. may be different)
5146 * - full_open_options: Options which, when given when opening a block device
5147 * (without a filename), result in a BDS (mostly)
5148 * equalling the given one
5149 * - filename: If exact_filename is set, it is copied here. Otherwise,
5150 * full_open_options is converted to a JSON object, prefixed with
5151 * "json:" (for use through the JSON pseudo protocol) and put here.
5153 void bdrv_refresh_filename(BlockDriverState
*bs
)
5155 BlockDriver
*drv
= bs
->drv
;
5162 /* This BDS's file name will most probably depend on its file's name, so
5163 * refresh that first */
5165 bdrv_refresh_filename(bs
->file
->bs
);
5168 if (drv
->bdrv_refresh_filename
) {
5169 /* Obsolete information is of no use here, so drop the old file name
5170 * information before refreshing it */
5171 bs
->exact_filename
[0] = '\0';
5172 if (bs
->full_open_options
) {
5173 QDECREF(bs
->full_open_options
);
5174 bs
->full_open_options
= NULL
;
5178 append_open_options(opts
, bs
);
5179 drv
->bdrv_refresh_filename(bs
, opts
);
5181 } else if (bs
->file
) {
5182 /* Try to reconstruct valid information from the underlying file */
5183 bool has_open_options
;
5185 bs
->exact_filename
[0] = '\0';
5186 if (bs
->full_open_options
) {
5187 QDECREF(bs
->full_open_options
);
5188 bs
->full_open_options
= NULL
;
5192 has_open_options
= append_open_options(opts
, bs
);
5194 /* If no specific options have been given for this BDS, the filename of
5195 * the underlying file should suffice for this one as well */
5196 if (bs
->file
->bs
->exact_filename
[0] && !has_open_options
) {
5197 strcpy(bs
->exact_filename
, bs
->file
->bs
->exact_filename
);
5199 /* Reconstructing the full options QDict is simple for most format block
5200 * drivers, as long as the full options are known for the underlying
5201 * file BDS. The full options QDict of that file BDS should somehow
5202 * contain a representation of the filename, therefore the following
5203 * suffices without querying the (exact_)filename of this BDS. */
5204 if (bs
->file
->bs
->full_open_options
) {
5205 qdict_put_str(opts
, "driver", drv
->format_name
);
5206 QINCREF(bs
->file
->bs
->full_open_options
);
5207 qdict_put(opts
, "file", bs
->file
->bs
->full_open_options
);
5209 bs
->full_open_options
= opts
;
5213 } else if (!bs
->full_open_options
&& qdict_size(bs
->options
)) {
5214 /* There is no underlying file BDS (at least referenced by BDS.file),
5215 * so the full options QDict should be equal to the options given
5216 * specifically for this block device when it was opened (plus the
5217 * driver specification).
5218 * Because those options don't change, there is no need to update
5219 * full_open_options when it's already set. */
5222 append_open_options(opts
, bs
);
5223 qdict_put_str(opts
, "driver", drv
->format_name
);
5225 if (bs
->exact_filename
[0]) {
5226 /* This may not work for all block protocol drivers (some may
5227 * require this filename to be parsed), but we have to find some
5228 * default solution here, so just include it. If some block driver
5229 * does not support pure options without any filename at all or
5230 * needs some special format of the options QDict, it needs to
5231 * implement the driver-specific bdrv_refresh_filename() function.
5233 qdict_put_str(opts
, "filename", bs
->exact_filename
);
5236 bs
->full_open_options
= opts
;
5239 if (bs
->exact_filename
[0]) {
5240 pstrcpy(bs
->filename
, sizeof(bs
->filename
), bs
->exact_filename
);
5241 } else if (bs
->full_open_options
) {
5242 QString
*json
= qobject_to_json(QOBJECT(bs
->full_open_options
));
5243 snprintf(bs
->filename
, sizeof(bs
->filename
), "json:%s",
5244 qstring_get_str(json
));
5250 * Hot add/remove a BDS's child. So the user can take a child offline when
5251 * it is broken and take a new child online
5253 void bdrv_add_child(BlockDriverState
*parent_bs
, BlockDriverState
*child_bs
,
5257 if (!parent_bs
->drv
|| !parent_bs
->drv
->bdrv_add_child
) {
5258 error_setg(errp
, "The node %s does not support adding a child",
5259 bdrv_get_device_or_node_name(parent_bs
));
5263 if (!QLIST_EMPTY(&child_bs
->parents
)) {
5264 error_setg(errp
, "The node %s already has a parent",
5265 child_bs
->node_name
);
5269 parent_bs
->drv
->bdrv_add_child(parent_bs
, child_bs
, errp
);
5272 void bdrv_del_child(BlockDriverState
*parent_bs
, BdrvChild
*child
, Error
**errp
)
5276 if (!parent_bs
->drv
|| !parent_bs
->drv
->bdrv_del_child
) {
5277 error_setg(errp
, "The node %s does not support removing a child",
5278 bdrv_get_device_or_node_name(parent_bs
));
5282 QLIST_FOREACH(tmp
, &parent_bs
->children
, next
) {
5289 error_setg(errp
, "The node %s does not have a child named %s",
5290 bdrv_get_device_or_node_name(parent_bs
),
5291 bdrv_get_device_or_node_name(child
->bs
));
5295 parent_bs
->drv
->bdrv_del_child(parent_bs
, child
, errp
);
5298 bool bdrv_can_store_new_dirty_bitmap(BlockDriverState
*bs
, const char *name
,
5299 uint32_t granularity
, Error
**errp
)
5301 BlockDriver
*drv
= bs
->drv
;
5304 error_setg_errno(errp
, ENOMEDIUM
,
5305 "Can't store persistent bitmaps to %s",
5306 bdrv_get_device_or_node_name(bs
));
5310 if (!drv
->bdrv_can_store_new_dirty_bitmap
) {
5311 error_setg_errno(errp
, ENOTSUP
,
5312 "Can't store persistent bitmaps to %s",
5313 bdrv_get_device_or_node_name(bs
));
5317 return drv
->bdrv_can_store_new_dirty_bitmap(bs
, name
, granularity
, errp
);