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
24 #include "qemu/osdep.h"
25 #include "block/trace.h"
26 #include "block/block_int.h"
27 #include "block/blockjob.h"
28 #include "block/nbd.h"
29 #include "qemu/error-report.h"
30 #include "module_block.h"
31 #include "qemu/module.h"
32 #include "qapi/qmp/qerror.h"
33 #include "qapi/qmp/qbool.h"
34 #include "qapi/qmp/qjson.h"
35 #include "sysemu/block-backend.h"
36 #include "sysemu/sysemu.h"
37 #include "qemu/notify.h"
38 #include "qemu/coroutine.h"
39 #include "block/qapi.h"
40 #include "qmp-commands.h"
41 #include "qemu/timer.h"
42 #include "qapi-event.h"
43 #include "qemu/cutils.h"
47 #include <sys/ioctl.h>
48 #include <sys/queue.h>
58 #define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */
60 static QTAILQ_HEAD(, BlockDriverState
) graph_bdrv_states
=
61 QTAILQ_HEAD_INITIALIZER(graph_bdrv_states
);
63 static QTAILQ_HEAD(, BlockDriverState
) all_bdrv_states
=
64 QTAILQ_HEAD_INITIALIZER(all_bdrv_states
);
66 static QLIST_HEAD(, BlockDriver
) bdrv_drivers
=
67 QLIST_HEAD_INITIALIZER(bdrv_drivers
);
69 static BlockDriverState
*bdrv_open_inherit(const char *filename
,
70 const char *reference
,
71 QDict
*options
, int flags
,
72 BlockDriverState
*parent
,
73 const BdrvChildRole
*child_role
,
76 /* If non-zero, use only whitelisted block drivers */
77 static int use_bdrv_whitelist
;
80 static int is_windows_drive_prefix(const char *filename
)
82 return (((filename
[0] >= 'a' && filename
[0] <= 'z') ||
83 (filename
[0] >= 'A' && filename
[0] <= 'Z')) &&
87 int is_windows_drive(const char *filename
)
89 if (is_windows_drive_prefix(filename
) &&
92 if (strstart(filename
, "\\\\.\\", NULL
) ||
93 strstart(filename
, "//./", NULL
))
99 size_t bdrv_opt_mem_align(BlockDriverState
*bs
)
101 if (!bs
|| !bs
->drv
) {
102 /* page size or 4k (hdd sector size) should be on the safe side */
103 return MAX(4096, getpagesize());
106 return bs
->bl
.opt_mem_alignment
;
109 size_t bdrv_min_mem_align(BlockDriverState
*bs
)
111 if (!bs
|| !bs
->drv
) {
112 /* page size or 4k (hdd sector size) should be on the safe side */
113 return MAX(4096, getpagesize());
116 return bs
->bl
.min_mem_alignment
;
119 /* check if the path starts with "<protocol>:" */
120 int path_has_protocol(const char *path
)
125 if (is_windows_drive(path
) ||
126 is_windows_drive_prefix(path
)) {
129 p
= path
+ strcspn(path
, ":/\\");
131 p
= path
+ strcspn(path
, ":/");
137 int path_is_absolute(const char *path
)
140 /* specific case for names like: "\\.\d:" */
141 if (is_windows_drive(path
) || is_windows_drive_prefix(path
)) {
144 return (*path
== '/' || *path
== '\\');
146 return (*path
== '/');
150 /* if filename is absolute, just copy it to dest. Otherwise, build a
151 path to it by considering it is relative to base_path. URL are
153 void path_combine(char *dest
, int dest_size
,
154 const char *base_path
,
155 const char *filename
)
162 if (path_is_absolute(filename
)) {
163 pstrcpy(dest
, dest_size
, filename
);
165 const char *protocol_stripped
= NULL
;
167 if (path_has_protocol(base_path
)) {
168 protocol_stripped
= strchr(base_path
, ':');
169 if (protocol_stripped
) {
173 p
= protocol_stripped
?: base_path
;
175 p1
= strrchr(base_path
, '/');
179 p2
= strrchr(base_path
, '\\');
191 if (len
> dest_size
- 1)
193 memcpy(dest
, base_path
, len
);
195 pstrcat(dest
, dest_size
, filename
);
200 * Helper function for bdrv_parse_filename() implementations to remove optional
201 * protocol prefixes (especially "file:") from a filename and for putting the
202 * stripped filename into the options QDict if there is such a prefix.
204 void bdrv_parse_filename_strip_prefix(const char *filename
, const char *prefix
,
207 if (strstart(filename
, prefix
, &filename
)) {
208 /* Stripping the explicit protocol prefix may result in a protocol
209 * prefix being (wrongly) detected (if the filename contains a colon) */
210 if (path_has_protocol(filename
)) {
211 QString
*fat_filename
;
213 /* This means there is some colon before the first slash; therefore,
214 * this cannot be an absolute path */
215 assert(!path_is_absolute(filename
));
217 /* And we can thus fix the protocol detection issue by prefixing it
219 fat_filename
= qstring_from_str("./");
220 qstring_append(fat_filename
, filename
);
222 assert(!path_has_protocol(qstring_get_str(fat_filename
)));
224 qdict_put(options
, "filename", fat_filename
);
226 /* If no protocol prefix was detected, we can use the shortened
228 qdict_put_str(options
, "filename", filename
);
234 /* Returns whether the image file is opened as read-only. Note that this can
235 * return false and writing to the image file is still not possible because the
236 * image is inactivated. */
237 bool bdrv_is_read_only(BlockDriverState
*bs
)
239 return bs
->read_only
;
242 int bdrv_can_set_read_only(BlockDriverState
*bs
, bool read_only
,
243 bool ignore_allow_rdw
, Error
**errp
)
245 /* Do not set read_only if copy_on_read is enabled */
246 if (bs
->copy_on_read
&& read_only
) {
247 error_setg(errp
, "Can't set node '%s' to r/o with copy-on-read enabled",
248 bdrv_get_device_or_node_name(bs
));
252 /* Do not clear read_only if it is prohibited */
253 if (!read_only
&& !(bs
->open_flags
& BDRV_O_ALLOW_RDWR
) &&
256 error_setg(errp
, "Node '%s' is read only",
257 bdrv_get_device_or_node_name(bs
));
264 /* TODO Remove (deprecated since 2.11)
265 * Block drivers are not supposed to automatically change bs->read_only.
266 * Instead, they should just check whether they can provide what the user
267 * explicitly requested and error out if read-write is requested, but they can
268 * only provide read-only access. */
269 int bdrv_set_read_only(BlockDriverState
*bs
, bool read_only
, Error
**errp
)
273 ret
= bdrv_can_set_read_only(bs
, read_only
, false, errp
);
278 bs
->read_only
= read_only
;
282 void bdrv_get_full_backing_filename_from_filename(const char *backed
,
284 char *dest
, size_t sz
,
287 if (backing
[0] == '\0' || path_has_protocol(backing
) ||
288 path_is_absolute(backing
))
290 pstrcpy(dest
, sz
, backing
);
291 } else if (backed
[0] == '\0' || strstart(backed
, "json:", NULL
)) {
292 error_setg(errp
, "Cannot use relative backing file names for '%s'",
295 path_combine(dest
, sz
, backed
, backing
);
299 void bdrv_get_full_backing_filename(BlockDriverState
*bs
, char *dest
, size_t sz
,
302 char *backed
= bs
->exact_filename
[0] ? bs
->exact_filename
: bs
->filename
;
304 bdrv_get_full_backing_filename_from_filename(backed
, bs
->backing_file
,
308 void bdrv_register(BlockDriver
*bdrv
)
310 QLIST_INSERT_HEAD(&bdrv_drivers
, bdrv
, list
);
313 BlockDriverState
*bdrv_new(void)
315 BlockDriverState
*bs
;
318 bs
= g_new0(BlockDriverState
, 1);
319 QLIST_INIT(&bs
->dirty_bitmaps
);
320 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
321 QLIST_INIT(&bs
->op_blockers
[i
]);
323 notifier_with_return_list_init(&bs
->before_write_notifiers
);
324 qemu_co_mutex_init(&bs
->reqs_lock
);
325 qemu_mutex_init(&bs
->dirty_bitmap_mutex
);
327 bs
->aio_context
= qemu_get_aio_context();
329 qemu_co_queue_init(&bs
->flush_queue
);
331 QTAILQ_INSERT_TAIL(&all_bdrv_states
, bs
, bs_list
);
336 static BlockDriver
*bdrv_do_find_format(const char *format_name
)
340 QLIST_FOREACH(drv1
, &bdrv_drivers
, list
) {
341 if (!strcmp(drv1
->format_name
, format_name
)) {
349 BlockDriver
*bdrv_find_format(const char *format_name
)
354 drv1
= bdrv_do_find_format(format_name
);
359 /* The driver isn't registered, maybe we need to load a module */
360 for (i
= 0; i
< (int)ARRAY_SIZE(block_driver_modules
); ++i
) {
361 if (!strcmp(block_driver_modules
[i
].format_name
, format_name
)) {
362 block_module_load_one(block_driver_modules
[i
].library_name
);
367 return bdrv_do_find_format(format_name
);
370 static int bdrv_is_whitelisted(BlockDriver
*drv
, bool read_only
)
372 static const char *whitelist_rw
[] = {
373 CONFIG_BDRV_RW_WHITELIST
375 static const char *whitelist_ro
[] = {
376 CONFIG_BDRV_RO_WHITELIST
380 if (!whitelist_rw
[0] && !whitelist_ro
[0]) {
381 return 1; /* no whitelist, anything goes */
384 for (p
= whitelist_rw
; *p
; p
++) {
385 if (!strcmp(drv
->format_name
, *p
)) {
390 for (p
= whitelist_ro
; *p
; p
++) {
391 if (!strcmp(drv
->format_name
, *p
)) {
399 bool bdrv_uses_whitelist(void)
401 return use_bdrv_whitelist
;
404 typedef struct CreateCo
{
412 static void coroutine_fn
bdrv_create_co_entry(void *opaque
)
414 Error
*local_err
= NULL
;
417 CreateCo
*cco
= opaque
;
420 ret
= cco
->drv
->bdrv_create(cco
->filename
, cco
->opts
, &local_err
);
421 error_propagate(&cco
->err
, local_err
);
425 int bdrv_create(BlockDriver
*drv
, const char* filename
,
426 QemuOpts
*opts
, Error
**errp
)
433 .filename
= g_strdup(filename
),
439 if (!drv
->bdrv_create
) {
440 error_setg(errp
, "Driver '%s' does not support image creation", drv
->format_name
);
445 if (qemu_in_coroutine()) {
446 /* Fast-path if already in coroutine context */
447 bdrv_create_co_entry(&cco
);
449 co
= qemu_coroutine_create(bdrv_create_co_entry
, &cco
);
450 qemu_coroutine_enter(co
);
451 while (cco
.ret
== NOT_DONE
) {
452 aio_poll(qemu_get_aio_context(), true);
459 error_propagate(errp
, cco
.err
);
461 error_setg_errno(errp
, -ret
, "Could not create image");
466 g_free(cco
.filename
);
470 int bdrv_create_file(const char *filename
, QemuOpts
*opts
, Error
**errp
)
473 Error
*local_err
= NULL
;
476 drv
= bdrv_find_protocol(filename
, true, errp
);
481 ret
= bdrv_create(drv
, filename
, opts
, &local_err
);
482 error_propagate(errp
, local_err
);
487 * Try to get @bs's logical and physical block size.
488 * On success, store them in @bsz struct and return 0.
489 * On failure return -errno.
490 * @bs must not be empty.
492 int bdrv_probe_blocksizes(BlockDriverState
*bs
, BlockSizes
*bsz
)
494 BlockDriver
*drv
= bs
->drv
;
496 if (drv
&& drv
->bdrv_probe_blocksizes
) {
497 return drv
->bdrv_probe_blocksizes(bs
, bsz
);
498 } else if (drv
&& drv
->is_filter
&& bs
->file
) {
499 return bdrv_probe_blocksizes(bs
->file
->bs
, bsz
);
506 * Try to get @bs's geometry (cyls, heads, sectors).
507 * On success, store them in @geo struct and return 0.
508 * On failure return -errno.
509 * @bs must not be empty.
511 int bdrv_probe_geometry(BlockDriverState
*bs
, HDGeometry
*geo
)
513 BlockDriver
*drv
= bs
->drv
;
515 if (drv
&& drv
->bdrv_probe_geometry
) {
516 return drv
->bdrv_probe_geometry(bs
, geo
);
517 } else if (drv
&& drv
->is_filter
&& bs
->file
) {
518 return bdrv_probe_geometry(bs
->file
->bs
, geo
);
525 * Create a uniquely-named empty temporary file.
526 * Return 0 upon success, otherwise a negative errno value.
528 int get_tmp_filename(char *filename
, int size
)
531 char temp_dir
[MAX_PATH
];
532 /* GetTempFileName requires that its output buffer (4th param)
533 have length MAX_PATH or greater. */
534 assert(size
>= MAX_PATH
);
535 return (GetTempPath(MAX_PATH
, temp_dir
)
536 && GetTempFileName(temp_dir
, "qem", 0, filename
)
537 ? 0 : -GetLastError());
541 tmpdir
= getenv("TMPDIR");
545 if (snprintf(filename
, size
, "%s/vl.XXXXXX", tmpdir
) >= size
) {
548 fd
= mkstemp(filename
);
552 if (close(fd
) != 0) {
561 * Detect host devices. By convention, /dev/cdrom[N] is always
562 * recognized as a host CDROM.
564 static BlockDriver
*find_hdev_driver(const char *filename
)
566 int score_max
= 0, score
;
567 BlockDriver
*drv
= NULL
, *d
;
569 QLIST_FOREACH(d
, &bdrv_drivers
, list
) {
570 if (d
->bdrv_probe_device
) {
571 score
= d
->bdrv_probe_device(filename
);
572 if (score
> score_max
) {
582 static BlockDriver
*bdrv_do_find_protocol(const char *protocol
)
586 QLIST_FOREACH(drv1
, &bdrv_drivers
, list
) {
587 if (drv1
->protocol_name
&& !strcmp(drv1
->protocol_name
, protocol
)) {
595 BlockDriver
*bdrv_find_protocol(const char *filename
,
596 bool allow_protocol_prefix
,
605 /* TODO Drivers without bdrv_file_open must be specified explicitly */
608 * XXX(hch): we really should not let host device detection
609 * override an explicit protocol specification, but moving this
610 * later breaks access to device names with colons in them.
611 * Thanks to the brain-dead persistent naming schemes on udev-
612 * based Linux systems those actually are quite common.
614 drv1
= find_hdev_driver(filename
);
619 if (!path_has_protocol(filename
) || !allow_protocol_prefix
) {
623 p
= strchr(filename
, ':');
626 if (len
> sizeof(protocol
) - 1)
627 len
= sizeof(protocol
) - 1;
628 memcpy(protocol
, filename
, len
);
629 protocol
[len
] = '\0';
631 drv1
= bdrv_do_find_protocol(protocol
);
636 for (i
= 0; i
< (int)ARRAY_SIZE(block_driver_modules
); ++i
) {
637 if (block_driver_modules
[i
].protocol_name
&&
638 !strcmp(block_driver_modules
[i
].protocol_name
, protocol
)) {
639 block_module_load_one(block_driver_modules
[i
].library_name
);
644 drv1
= bdrv_do_find_protocol(protocol
);
646 error_setg(errp
, "Unknown protocol '%s'", protocol
);
652 * Guess image format by probing its contents.
653 * This is not a good idea when your image is raw (CVE-2008-2004), but
654 * we do it anyway for backward compatibility.
656 * @buf contains the image's first @buf_size bytes.
657 * @buf_size is the buffer size in bytes (generally BLOCK_PROBE_BUF_SIZE,
658 * but can be smaller if the image file is smaller)
659 * @filename is its filename.
661 * For all block drivers, call the bdrv_probe() method to get its
663 * Return the first block driver with the highest probing score.
665 BlockDriver
*bdrv_probe_all(const uint8_t *buf
, int buf_size
,
666 const char *filename
)
668 int score_max
= 0, score
;
669 BlockDriver
*drv
= NULL
, *d
;
671 QLIST_FOREACH(d
, &bdrv_drivers
, list
) {
673 score
= d
->bdrv_probe(buf
, buf_size
, filename
);
674 if (score
> score_max
) {
684 static int find_image_format(BlockBackend
*file
, const char *filename
,
685 BlockDriver
**pdrv
, Error
**errp
)
688 uint8_t buf
[BLOCK_PROBE_BUF_SIZE
];
691 /* Return the raw BlockDriver * to scsi-generic devices or empty drives */
692 if (blk_is_sg(file
) || !blk_is_inserted(file
) || blk_getlength(file
) == 0) {
697 ret
= blk_pread(file
, 0, buf
, sizeof(buf
));
699 error_setg_errno(errp
, -ret
, "Could not read image for determining its "
705 drv
= bdrv_probe_all(buf
, ret
, filename
);
707 error_setg(errp
, "Could not determine image format: No compatible "
716 * Set the current 'total_sectors' value
717 * Return 0 on success, -errno on error.
719 static int refresh_total_sectors(BlockDriverState
*bs
, int64_t hint
)
721 BlockDriver
*drv
= bs
->drv
;
727 /* Do not attempt drv->bdrv_getlength() on scsi-generic devices */
731 /* query actual device if possible, otherwise just trust the hint */
732 if (drv
->bdrv_getlength
) {
733 int64_t length
= drv
->bdrv_getlength(bs
);
737 hint
= DIV_ROUND_UP(length
, BDRV_SECTOR_SIZE
);
740 bs
->total_sectors
= hint
;
745 * Combines a QDict of new block driver @options with any missing options taken
746 * from @old_options, so that leaving out an option defaults to its old value.
748 static void bdrv_join_options(BlockDriverState
*bs
, QDict
*options
,
751 if (bs
->drv
&& bs
->drv
->bdrv_join_options
) {
752 bs
->drv
->bdrv_join_options(options
, old_options
);
754 qdict_join(options
, old_options
, false);
759 * Set open flags for a given discard mode
761 * Return 0 on success, -1 if the discard mode was invalid.
763 int bdrv_parse_discard_flags(const char *mode
, int *flags
)
765 *flags
&= ~BDRV_O_UNMAP
;
767 if (!strcmp(mode
, "off") || !strcmp(mode
, "ignore")) {
769 } else if (!strcmp(mode
, "on") || !strcmp(mode
, "unmap")) {
770 *flags
|= BDRV_O_UNMAP
;
779 * Set open flags for a given cache mode
781 * Return 0 on success, -1 if the cache mode was invalid.
783 int bdrv_parse_cache_mode(const char *mode
, int *flags
, bool *writethrough
)
785 *flags
&= ~BDRV_O_CACHE_MASK
;
787 if (!strcmp(mode
, "off") || !strcmp(mode
, "none")) {
788 *writethrough
= false;
789 *flags
|= BDRV_O_NOCACHE
;
790 } else if (!strcmp(mode
, "directsync")) {
791 *writethrough
= true;
792 *flags
|= BDRV_O_NOCACHE
;
793 } else if (!strcmp(mode
, "writeback")) {
794 *writethrough
= false;
795 } else if (!strcmp(mode
, "unsafe")) {
796 *writethrough
= false;
797 *flags
|= BDRV_O_NO_FLUSH
;
798 } else if (!strcmp(mode
, "writethrough")) {
799 *writethrough
= true;
807 static char *bdrv_child_get_parent_desc(BdrvChild
*c
)
809 BlockDriverState
*parent
= c
->opaque
;
810 return g_strdup(bdrv_get_device_or_node_name(parent
));
813 static void bdrv_child_cb_drained_begin(BdrvChild
*child
)
815 BlockDriverState
*bs
= child
->opaque
;
816 bdrv_drained_begin(bs
);
819 static void bdrv_child_cb_drained_end(BdrvChild
*child
)
821 BlockDriverState
*bs
= child
->opaque
;
822 bdrv_drained_end(bs
);
825 static int bdrv_child_cb_inactivate(BdrvChild
*child
)
827 BlockDriverState
*bs
= child
->opaque
;
828 assert(bs
->open_flags
& BDRV_O_INACTIVE
);
833 * Returns the options and flags that a temporary snapshot should get, based on
834 * the originally requested flags (the originally requested image will have
835 * flags like a backing file)
837 static void bdrv_temp_snapshot_options(int *child_flags
, QDict
*child_options
,
838 int parent_flags
, QDict
*parent_options
)
840 *child_flags
= (parent_flags
& ~BDRV_O_SNAPSHOT
) | BDRV_O_TEMPORARY
;
842 /* For temporary files, unconditional cache=unsafe is fine */
843 qdict_set_default_str(child_options
, BDRV_OPT_CACHE_DIRECT
, "off");
844 qdict_set_default_str(child_options
, BDRV_OPT_CACHE_NO_FLUSH
, "on");
846 /* Copy the read-only option from the parent */
847 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_READ_ONLY
);
849 /* aio=native doesn't work for cache.direct=off, so disable it for the
850 * temporary snapshot */
851 *child_flags
&= ~BDRV_O_NATIVE_AIO
;
855 * Returns the options and flags that bs->file should get if a protocol driver
856 * is expected, based on the given options and flags for the parent BDS
858 static void bdrv_inherited_options(int *child_flags
, QDict
*child_options
,
859 int parent_flags
, QDict
*parent_options
)
861 int flags
= parent_flags
;
863 /* Enable protocol handling, disable format probing for bs->file */
864 flags
|= BDRV_O_PROTOCOL
;
866 /* If the cache mode isn't explicitly set, inherit direct and no-flush from
868 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_CACHE_DIRECT
);
869 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_CACHE_NO_FLUSH
);
870 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_FORCE_SHARE
);
872 /* Inherit the read-only option from the parent if it's not set */
873 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_READ_ONLY
);
875 /* Our block drivers take care to send flushes and respect unmap policy,
876 * so we can default to enable both on lower layers regardless of the
877 * corresponding parent options. */
878 qdict_set_default_str(child_options
, BDRV_OPT_DISCARD
, "unmap");
880 /* Clear flags that only apply to the top layer */
881 flags
&= ~(BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
| BDRV_O_COPY_ON_READ
|
884 *child_flags
= flags
;
887 const BdrvChildRole child_file
= {
888 .get_parent_desc
= bdrv_child_get_parent_desc
,
889 .inherit_options
= bdrv_inherited_options
,
890 .drained_begin
= bdrv_child_cb_drained_begin
,
891 .drained_end
= bdrv_child_cb_drained_end
,
892 .inactivate
= bdrv_child_cb_inactivate
,
896 * Returns the options and flags that bs->file should get if the use of formats
897 * (and not only protocols) is permitted for it, based on the given options and
898 * flags for the parent BDS
900 static void bdrv_inherited_fmt_options(int *child_flags
, QDict
*child_options
,
901 int parent_flags
, QDict
*parent_options
)
903 child_file
.inherit_options(child_flags
, child_options
,
904 parent_flags
, parent_options
);
906 *child_flags
&= ~(BDRV_O_PROTOCOL
| BDRV_O_NO_IO
);
909 const BdrvChildRole child_format
= {
910 .get_parent_desc
= bdrv_child_get_parent_desc
,
911 .inherit_options
= bdrv_inherited_fmt_options
,
912 .drained_begin
= bdrv_child_cb_drained_begin
,
913 .drained_end
= bdrv_child_cb_drained_end
,
914 .inactivate
= bdrv_child_cb_inactivate
,
917 static void bdrv_backing_attach(BdrvChild
*c
)
919 BlockDriverState
*parent
= c
->opaque
;
920 BlockDriverState
*backing_hd
= c
->bs
;
922 assert(!parent
->backing_blocker
);
923 error_setg(&parent
->backing_blocker
,
924 "node is used as backing hd of '%s'",
925 bdrv_get_device_or_node_name(parent
));
927 parent
->open_flags
&= ~BDRV_O_NO_BACKING
;
928 pstrcpy(parent
->backing_file
, sizeof(parent
->backing_file
),
929 backing_hd
->filename
);
930 pstrcpy(parent
->backing_format
, sizeof(parent
->backing_format
),
931 backing_hd
->drv
? backing_hd
->drv
->format_name
: "");
933 bdrv_op_block_all(backing_hd
, parent
->backing_blocker
);
934 /* Otherwise we won't be able to commit or stream */
935 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_COMMIT_TARGET
,
936 parent
->backing_blocker
);
937 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_STREAM
,
938 parent
->backing_blocker
);
940 * We do backup in 3 ways:
942 * The target bs is new opened, and the source is top BDS
944 * Both the source and the target are top BDSes.
945 * 3. internal backup(used for block replication)
946 * Both the source and the target are backing file
948 * In case 1 and 2, neither the source nor the target is the backing file.
949 * In case 3, we will block the top BDS, so there is only one block job
950 * for the top BDS and its backing chain.
952 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_BACKUP_SOURCE
,
953 parent
->backing_blocker
);
954 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_BACKUP_TARGET
,
955 parent
->backing_blocker
);
958 static void bdrv_backing_detach(BdrvChild
*c
)
960 BlockDriverState
*parent
= c
->opaque
;
962 assert(parent
->backing_blocker
);
963 bdrv_op_unblock_all(c
->bs
, parent
->backing_blocker
);
964 error_free(parent
->backing_blocker
);
965 parent
->backing_blocker
= NULL
;
969 * Returns the options and flags that bs->backing should get, based on the
970 * given options and flags for the parent BDS
972 static void bdrv_backing_options(int *child_flags
, QDict
*child_options
,
973 int parent_flags
, QDict
*parent_options
)
975 int flags
= parent_flags
;
977 /* The cache mode is inherited unmodified for backing files; except WCE,
978 * which is only applied on the top level (BlockBackend) */
979 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_CACHE_DIRECT
);
980 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_CACHE_NO_FLUSH
);
981 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_FORCE_SHARE
);
983 /* backing files always opened read-only */
984 qdict_set_default_str(child_options
, BDRV_OPT_READ_ONLY
, "on");
985 flags
&= ~BDRV_O_COPY_ON_READ
;
987 /* snapshot=on is handled on the top layer */
988 flags
&= ~(BDRV_O_SNAPSHOT
| BDRV_O_TEMPORARY
);
990 *child_flags
= flags
;
993 static int bdrv_backing_update_filename(BdrvChild
*c
, BlockDriverState
*base
,
994 const char *filename
, Error
**errp
)
996 BlockDriverState
*parent
= c
->opaque
;
997 int orig_flags
= bdrv_get_flags(parent
);
1000 if (!(orig_flags
& BDRV_O_RDWR
)) {
1001 ret
= bdrv_reopen(parent
, orig_flags
| BDRV_O_RDWR
, errp
);
1007 ret
= bdrv_change_backing_file(parent
, filename
,
1008 base
->drv
? base
->drv
->format_name
: "");
1010 error_setg_errno(errp
, -ret
, "Could not update backing file link");
1013 if (!(orig_flags
& BDRV_O_RDWR
)) {
1014 bdrv_reopen(parent
, orig_flags
, NULL
);
1020 const BdrvChildRole child_backing
= {
1021 .get_parent_desc
= bdrv_child_get_parent_desc
,
1022 .attach
= bdrv_backing_attach
,
1023 .detach
= bdrv_backing_detach
,
1024 .inherit_options
= bdrv_backing_options
,
1025 .drained_begin
= bdrv_child_cb_drained_begin
,
1026 .drained_end
= bdrv_child_cb_drained_end
,
1027 .inactivate
= bdrv_child_cb_inactivate
,
1028 .update_filename
= bdrv_backing_update_filename
,
1031 static int bdrv_open_flags(BlockDriverState
*bs
, int flags
)
1033 int open_flags
= flags
;
1036 * Clear flags that are internal to the block layer before opening the
1039 open_flags
&= ~(BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
| BDRV_O_PROTOCOL
);
1042 * Snapshots should be writable.
1044 if (flags
& BDRV_O_TEMPORARY
) {
1045 open_flags
|= BDRV_O_RDWR
;
1051 static void update_flags_from_options(int *flags
, QemuOpts
*opts
)
1053 *flags
&= ~BDRV_O_CACHE_MASK
;
1055 assert(qemu_opt_find(opts
, BDRV_OPT_CACHE_NO_FLUSH
));
1056 if (qemu_opt_get_bool(opts
, BDRV_OPT_CACHE_NO_FLUSH
, false)) {
1057 *flags
|= BDRV_O_NO_FLUSH
;
1060 assert(qemu_opt_find(opts
, BDRV_OPT_CACHE_DIRECT
));
1061 if (qemu_opt_get_bool(opts
, BDRV_OPT_CACHE_DIRECT
, false)) {
1062 *flags
|= BDRV_O_NOCACHE
;
1065 *flags
&= ~BDRV_O_RDWR
;
1067 assert(qemu_opt_find(opts
, BDRV_OPT_READ_ONLY
));
1068 if (!qemu_opt_get_bool(opts
, BDRV_OPT_READ_ONLY
, false)) {
1069 *flags
|= BDRV_O_RDWR
;
1074 static void update_options_from_flags(QDict
*options
, int flags
)
1076 if (!qdict_haskey(options
, BDRV_OPT_CACHE_DIRECT
)) {
1077 qdict_put_bool(options
, BDRV_OPT_CACHE_DIRECT
, flags
& BDRV_O_NOCACHE
);
1079 if (!qdict_haskey(options
, BDRV_OPT_CACHE_NO_FLUSH
)) {
1080 qdict_put_bool(options
, BDRV_OPT_CACHE_NO_FLUSH
,
1081 flags
& BDRV_O_NO_FLUSH
);
1083 if (!qdict_haskey(options
, BDRV_OPT_READ_ONLY
)) {
1084 qdict_put_bool(options
, BDRV_OPT_READ_ONLY
, !(flags
& BDRV_O_RDWR
));
1088 static void bdrv_assign_node_name(BlockDriverState
*bs
,
1089 const char *node_name
,
1092 char *gen_node_name
= NULL
;
1095 node_name
= gen_node_name
= id_generate(ID_BLOCK
);
1096 } else if (!id_wellformed(node_name
)) {
1098 * Check for empty string or invalid characters, but not if it is
1099 * generated (generated names use characters not available to the user)
1101 error_setg(errp
, "Invalid node name");
1105 /* takes care of avoiding namespaces collisions */
1106 if (blk_by_name(node_name
)) {
1107 error_setg(errp
, "node-name=%s is conflicting with a device id",
1112 /* takes care of avoiding duplicates node names */
1113 if (bdrv_find_node(node_name
)) {
1114 error_setg(errp
, "Duplicate node name");
1118 /* copy node name into the bs and insert it into the graph list */
1119 pstrcpy(bs
->node_name
, sizeof(bs
->node_name
), node_name
);
1120 QTAILQ_INSERT_TAIL(&graph_bdrv_states
, bs
, node_list
);
1122 g_free(gen_node_name
);
1125 static int bdrv_open_driver(BlockDriverState
*bs
, BlockDriver
*drv
,
1126 const char *node_name
, QDict
*options
,
1127 int open_flags
, Error
**errp
)
1129 Error
*local_err
= NULL
;
1132 bdrv_assign_node_name(bs
, node_name
, &local_err
);
1134 error_propagate(errp
, local_err
);
1139 bs
->read_only
= !(bs
->open_flags
& BDRV_O_RDWR
);
1140 bs
->opaque
= g_malloc0(drv
->instance_size
);
1142 if (drv
->bdrv_file_open
) {
1143 assert(!drv
->bdrv_needs_filename
|| bs
->filename
[0]);
1144 ret
= drv
->bdrv_file_open(bs
, options
, open_flags
, &local_err
);
1145 } else if (drv
->bdrv_open
) {
1146 ret
= drv
->bdrv_open(bs
, options
, open_flags
, &local_err
);
1153 error_propagate(errp
, local_err
);
1154 } else if (bs
->filename
[0]) {
1155 error_setg_errno(errp
, -ret
, "Could not open '%s'", bs
->filename
);
1157 error_setg_errno(errp
, -ret
, "Could not open image");
1162 ret
= refresh_total_sectors(bs
, bs
->total_sectors
);
1164 error_setg_errno(errp
, -ret
, "Could not refresh total sector count");
1168 bdrv_refresh_limits(bs
, &local_err
);
1170 error_propagate(errp
, local_err
);
1174 assert(bdrv_opt_mem_align(bs
) != 0);
1175 assert(bdrv_min_mem_align(bs
) != 0);
1176 assert(is_power_of_2(bs
->bl
.request_alignment
));
1181 if (bs
->file
!= NULL
) {
1182 bdrv_unref_child(bs
, bs
->file
);
1190 BlockDriverState
*bdrv_new_open_driver(BlockDriver
*drv
, const char *node_name
,
1191 int flags
, Error
**errp
)
1193 BlockDriverState
*bs
;
1197 bs
->open_flags
= flags
;
1198 bs
->explicit_options
= qdict_new();
1199 bs
->options
= qdict_new();
1202 update_options_from_flags(bs
->options
, flags
);
1204 ret
= bdrv_open_driver(bs
, drv
, node_name
, bs
->options
, flags
, errp
);
1206 QDECREF(bs
->explicit_options
);
1207 bs
->explicit_options
= NULL
;
1208 QDECREF(bs
->options
);
1217 QemuOptsList bdrv_runtime_opts
= {
1218 .name
= "bdrv_common",
1219 .head
= QTAILQ_HEAD_INITIALIZER(bdrv_runtime_opts
.head
),
1222 .name
= "node-name",
1223 .type
= QEMU_OPT_STRING
,
1224 .help
= "Node name of the block device node",
1228 .type
= QEMU_OPT_STRING
,
1229 .help
= "Block driver to use for the node",
1232 .name
= BDRV_OPT_CACHE_DIRECT
,
1233 .type
= QEMU_OPT_BOOL
,
1234 .help
= "Bypass software writeback cache on the host",
1237 .name
= BDRV_OPT_CACHE_NO_FLUSH
,
1238 .type
= QEMU_OPT_BOOL
,
1239 .help
= "Ignore flush requests",
1242 .name
= BDRV_OPT_READ_ONLY
,
1243 .type
= QEMU_OPT_BOOL
,
1244 .help
= "Node is opened in read-only mode",
1247 .name
= "detect-zeroes",
1248 .type
= QEMU_OPT_STRING
,
1249 .help
= "try to optimize zero writes (off, on, unmap)",
1253 .type
= QEMU_OPT_STRING
,
1254 .help
= "discard operation (ignore/off, unmap/on)",
1257 .name
= BDRV_OPT_FORCE_SHARE
,
1258 .type
= QEMU_OPT_BOOL
,
1259 .help
= "always accept other writers (default: off)",
1261 { /* end of list */ }
1266 * Common part for opening disk images and files
1268 * Removes all processed options from *options.
1270 static int bdrv_open_common(BlockDriverState
*bs
, BlockBackend
*file
,
1271 QDict
*options
, Error
**errp
)
1273 int ret
, open_flags
;
1274 const char *filename
;
1275 const char *driver_name
= NULL
;
1276 const char *node_name
= NULL
;
1277 const char *discard
;
1278 const char *detect_zeroes
;
1281 Error
*local_err
= NULL
;
1283 assert(bs
->file
== NULL
);
1284 assert(options
!= NULL
&& bs
->options
!= options
);
1286 opts
= qemu_opts_create(&bdrv_runtime_opts
, NULL
, 0, &error_abort
);
1287 qemu_opts_absorb_qdict(opts
, options
, &local_err
);
1289 error_propagate(errp
, local_err
);
1294 update_flags_from_options(&bs
->open_flags
, opts
);
1296 driver_name
= qemu_opt_get(opts
, "driver");
1297 drv
= bdrv_find_format(driver_name
);
1298 assert(drv
!= NULL
);
1300 bs
->force_share
= qemu_opt_get_bool(opts
, BDRV_OPT_FORCE_SHARE
, false);
1302 if (bs
->force_share
&& (bs
->open_flags
& BDRV_O_RDWR
)) {
1304 BDRV_OPT_FORCE_SHARE
1305 "=on can only be used with read-only images");
1311 filename
= blk_bs(file
)->filename
;
1314 * Caution: while qdict_get_try_str() is fine, getting
1315 * non-string types would require more care. When @options
1316 * come from -blockdev or blockdev_add, its members are typed
1317 * according to the QAPI schema, but when they come from
1318 * -drive, they're all QString.
1320 filename
= qdict_get_try_str(options
, "filename");
1323 if (drv
->bdrv_needs_filename
&& (!filename
|| !filename
[0])) {
1324 error_setg(errp
, "The '%s' block driver requires a file name",
1330 trace_bdrv_open_common(bs
, filename
?: "", bs
->open_flags
,
1333 bs
->read_only
= !(bs
->open_flags
& BDRV_O_RDWR
);
1335 if (use_bdrv_whitelist
&& !bdrv_is_whitelisted(drv
, bs
->read_only
)) {
1337 !bs
->read_only
&& bdrv_is_whitelisted(drv
, true)
1338 ? "Driver '%s' can only be used for read-only devices"
1339 : "Driver '%s' is not whitelisted",
1345 /* bdrv_new() and bdrv_close() make it so */
1346 assert(atomic_read(&bs
->copy_on_read
) == 0);
1348 if (bs
->open_flags
& BDRV_O_COPY_ON_READ
) {
1349 if (!bs
->read_only
) {
1350 bdrv_enable_copy_on_read(bs
);
1352 error_setg(errp
, "Can't use copy-on-read on read-only device");
1358 discard
= qemu_opt_get(opts
, "discard");
1359 if (discard
!= NULL
) {
1360 if (bdrv_parse_discard_flags(discard
, &bs
->open_flags
) != 0) {
1361 error_setg(errp
, "Invalid discard option");
1367 detect_zeroes
= qemu_opt_get(opts
, "detect-zeroes");
1368 if (detect_zeroes
) {
1369 BlockdevDetectZeroesOptions value
=
1370 qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup
,
1372 BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF
,
1375 error_propagate(errp
, local_err
);
1380 if (value
== BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP
&&
1381 !(bs
->open_flags
& BDRV_O_UNMAP
))
1383 error_setg(errp
, "setting detect-zeroes to unmap is not allowed "
1384 "without setting discard operation to unmap");
1389 bs
->detect_zeroes
= value
;
1392 if (filename
!= NULL
) {
1393 pstrcpy(bs
->filename
, sizeof(bs
->filename
), filename
);
1395 bs
->filename
[0] = '\0';
1397 pstrcpy(bs
->exact_filename
, sizeof(bs
->exact_filename
), bs
->filename
);
1399 /* Open the image, either directly or using a protocol */
1400 open_flags
= bdrv_open_flags(bs
, bs
->open_flags
);
1401 node_name
= qemu_opt_get(opts
, "node-name");
1403 assert(!drv
->bdrv_file_open
|| file
== NULL
);
1404 ret
= bdrv_open_driver(bs
, drv
, node_name
, options
, open_flags
, errp
);
1409 qemu_opts_del(opts
);
1413 qemu_opts_del(opts
);
1417 static GCC_FMT_ATTR(1, 0)
1418 QDict
*parse_json_filename(const char *filename
, Error
**errp
)
1420 QObject
*options_obj
;
1424 ret
= strstart(filename
, "json:", &filename
);
1427 options_obj
= qobject_from_json(filename
, errp
);
1429 /* Work around qobject_from_json() lossage TODO fix that */
1430 if (errp
&& !*errp
) {
1431 error_setg(errp
, "Could not parse the JSON options");
1434 error_prepend(errp
, "Could not parse the JSON options: ");
1438 options
= qobject_to_qdict(options_obj
);
1440 qobject_decref(options_obj
);
1441 error_setg(errp
, "Invalid JSON object given");
1445 qdict_flatten(options
);
1450 static void parse_json_protocol(QDict
*options
, const char **pfilename
,
1453 QDict
*json_options
;
1454 Error
*local_err
= NULL
;
1456 /* Parse json: pseudo-protocol */
1457 if (!*pfilename
|| !g_str_has_prefix(*pfilename
, "json:")) {
1461 json_options
= parse_json_filename(*pfilename
, &local_err
);
1463 error_propagate(errp
, local_err
);
1467 /* Options given in the filename have lower priority than options
1468 * specified directly */
1469 qdict_join(options
, json_options
, false);
1470 QDECREF(json_options
);
1475 * Fills in default options for opening images and converts the legacy
1476 * filename/flags pair to option QDict entries.
1477 * The BDRV_O_PROTOCOL flag in *flags will be set or cleared accordingly if a
1478 * block driver has been specified explicitly.
1480 static int bdrv_fill_options(QDict
**options
, const char *filename
,
1481 int *flags
, Error
**errp
)
1483 const char *drvname
;
1484 bool protocol
= *flags
& BDRV_O_PROTOCOL
;
1485 bool parse_filename
= false;
1486 BlockDriver
*drv
= NULL
;
1487 Error
*local_err
= NULL
;
1490 * Caution: while qdict_get_try_str() is fine, getting non-string
1491 * types would require more care. When @options come from
1492 * -blockdev or blockdev_add, its members are typed according to
1493 * the QAPI schema, but when they come from -drive, they're all
1496 drvname
= qdict_get_try_str(*options
, "driver");
1498 drv
= bdrv_find_format(drvname
);
1500 error_setg(errp
, "Unknown driver '%s'", drvname
);
1503 /* If the user has explicitly specified the driver, this choice should
1504 * override the BDRV_O_PROTOCOL flag */
1505 protocol
= drv
->bdrv_file_open
;
1509 *flags
|= BDRV_O_PROTOCOL
;
1511 *flags
&= ~BDRV_O_PROTOCOL
;
1514 /* Translate cache options from flags into options */
1515 update_options_from_flags(*options
, *flags
);
1517 /* Fetch the file name from the options QDict if necessary */
1518 if (protocol
&& filename
) {
1519 if (!qdict_haskey(*options
, "filename")) {
1520 qdict_put_str(*options
, "filename", filename
);
1521 parse_filename
= true;
1523 error_setg(errp
, "Can't specify 'file' and 'filename' options at "
1529 /* Find the right block driver */
1530 /* See cautionary note on accessing @options above */
1531 filename
= qdict_get_try_str(*options
, "filename");
1533 if (!drvname
&& protocol
) {
1535 drv
= bdrv_find_protocol(filename
, parse_filename
, errp
);
1540 drvname
= drv
->format_name
;
1541 qdict_put_str(*options
, "driver", drvname
);
1543 error_setg(errp
, "Must specify either driver or file");
1548 assert(drv
|| !protocol
);
1550 /* Driver-specific filename parsing */
1551 if (drv
&& drv
->bdrv_parse_filename
&& parse_filename
) {
1552 drv
->bdrv_parse_filename(filename
, *options
, &local_err
);
1554 error_propagate(errp
, local_err
);
1558 if (!drv
->bdrv_needs_filename
) {
1559 qdict_del(*options
, "filename");
1566 static int bdrv_child_check_perm(BdrvChild
*c
, BlockReopenQueue
*q
,
1567 uint64_t perm
, uint64_t shared
,
1568 GSList
*ignore_children
, Error
**errp
);
1569 static void bdrv_child_abort_perm_update(BdrvChild
*c
);
1570 static void bdrv_child_set_perm(BdrvChild
*c
, uint64_t perm
, uint64_t shared
);
1572 typedef struct BlockReopenQueueEntry
{
1574 BDRVReopenState state
;
1575 QSIMPLEQ_ENTRY(BlockReopenQueueEntry
) entry
;
1576 } BlockReopenQueueEntry
;
1579 * Return the flags that @bs will have after the reopens in @q have
1580 * successfully completed. If @q is NULL (or @bs is not contained in @q),
1581 * return the current flags.
1583 static int bdrv_reopen_get_flags(BlockReopenQueue
*q
, BlockDriverState
*bs
)
1585 BlockReopenQueueEntry
*entry
;
1588 QSIMPLEQ_FOREACH(entry
, q
, entry
) {
1589 if (entry
->state
.bs
== bs
) {
1590 return entry
->state
.flags
;
1595 return bs
->open_flags
;
1598 /* Returns whether the image file can be written to after the reopen queue @q
1599 * has been successfully applied, or right now if @q is NULL. */
1600 static bool bdrv_is_writable(BlockDriverState
*bs
, BlockReopenQueue
*q
)
1602 int flags
= bdrv_reopen_get_flags(q
, bs
);
1604 return (flags
& (BDRV_O_RDWR
| BDRV_O_INACTIVE
)) == BDRV_O_RDWR
;
1607 static void bdrv_child_perm(BlockDriverState
*bs
, BlockDriverState
*child_bs
,
1608 BdrvChild
*c
, const BdrvChildRole
*role
,
1609 BlockReopenQueue
*reopen_queue
,
1610 uint64_t parent_perm
, uint64_t parent_shared
,
1611 uint64_t *nperm
, uint64_t *nshared
)
1613 if (bs
->drv
&& bs
->drv
->bdrv_child_perm
) {
1614 bs
->drv
->bdrv_child_perm(bs
, c
, role
, reopen_queue
,
1615 parent_perm
, parent_shared
,
1618 /* TODO Take force_share from reopen_queue */
1619 if (child_bs
&& child_bs
->force_share
) {
1620 *nshared
= BLK_PERM_ALL
;
1625 * Check whether permissions on this node can be changed in a way that
1626 * @cumulative_perms and @cumulative_shared_perms are the new cumulative
1627 * permissions of all its parents. This involves checking whether all necessary
1628 * permission changes to child nodes can be performed.
1630 * A call to this function must always be followed by a call to bdrv_set_perm()
1631 * or bdrv_abort_perm_update().
1633 static int bdrv_check_perm(BlockDriverState
*bs
, BlockReopenQueue
*q
,
1634 uint64_t cumulative_perms
,
1635 uint64_t cumulative_shared_perms
,
1636 GSList
*ignore_children
, Error
**errp
)
1638 BlockDriver
*drv
= bs
->drv
;
1642 /* Write permissions never work with read-only images */
1643 if ((cumulative_perms
& (BLK_PERM_WRITE
| BLK_PERM_WRITE_UNCHANGED
)) &&
1644 !bdrv_is_writable(bs
, q
))
1646 error_setg(errp
, "Block node is read-only");
1650 /* Check this node */
1655 if (drv
->bdrv_check_perm
) {
1656 return drv
->bdrv_check_perm(bs
, cumulative_perms
,
1657 cumulative_shared_perms
, errp
);
1660 /* Drivers that never have children can omit .bdrv_child_perm() */
1661 if (!drv
->bdrv_child_perm
) {
1662 assert(QLIST_EMPTY(&bs
->children
));
1666 /* Check all children */
1667 QLIST_FOREACH(c
, &bs
->children
, next
) {
1668 uint64_t cur_perm
, cur_shared
;
1669 bdrv_child_perm(bs
, c
->bs
, c
, c
->role
, q
,
1670 cumulative_perms
, cumulative_shared_perms
,
1671 &cur_perm
, &cur_shared
);
1672 ret
= bdrv_child_check_perm(c
, q
, cur_perm
, cur_shared
,
1673 ignore_children
, errp
);
1683 * Notifies drivers that after a previous bdrv_check_perm() call, the
1684 * permission update is not performed and any preparations made for it (e.g.
1685 * taken file locks) need to be undone.
1687 * This function recursively notifies all child nodes.
1689 static void bdrv_abort_perm_update(BlockDriverState
*bs
)
1691 BlockDriver
*drv
= bs
->drv
;
1698 if (drv
->bdrv_abort_perm_update
) {
1699 drv
->bdrv_abort_perm_update(bs
);
1702 QLIST_FOREACH(c
, &bs
->children
, next
) {
1703 bdrv_child_abort_perm_update(c
);
1707 static void bdrv_set_perm(BlockDriverState
*bs
, uint64_t cumulative_perms
,
1708 uint64_t cumulative_shared_perms
)
1710 BlockDriver
*drv
= bs
->drv
;
1717 /* Update this node */
1718 if (drv
->bdrv_set_perm
) {
1719 drv
->bdrv_set_perm(bs
, cumulative_perms
, cumulative_shared_perms
);
1722 /* Drivers that never have children can omit .bdrv_child_perm() */
1723 if (!drv
->bdrv_child_perm
) {
1724 assert(QLIST_EMPTY(&bs
->children
));
1728 /* Update all children */
1729 QLIST_FOREACH(c
, &bs
->children
, next
) {
1730 uint64_t cur_perm
, cur_shared
;
1731 bdrv_child_perm(bs
, c
->bs
, c
, c
->role
, NULL
,
1732 cumulative_perms
, cumulative_shared_perms
,
1733 &cur_perm
, &cur_shared
);
1734 bdrv_child_set_perm(c
, cur_perm
, cur_shared
);
1738 static void bdrv_get_cumulative_perm(BlockDriverState
*bs
, uint64_t *perm
,
1739 uint64_t *shared_perm
)
1742 uint64_t cumulative_perms
= 0;
1743 uint64_t cumulative_shared_perms
= BLK_PERM_ALL
;
1745 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
1746 cumulative_perms
|= c
->perm
;
1747 cumulative_shared_perms
&= c
->shared_perm
;
1750 *perm
= cumulative_perms
;
1751 *shared_perm
= cumulative_shared_perms
;
1754 static char *bdrv_child_user_desc(BdrvChild
*c
)
1756 if (c
->role
->get_parent_desc
) {
1757 return c
->role
->get_parent_desc(c
);
1760 return g_strdup("another user");
1763 char *bdrv_perm_names(uint64_t perm
)
1769 { BLK_PERM_CONSISTENT_READ
, "consistent read" },
1770 { BLK_PERM_WRITE
, "write" },
1771 { BLK_PERM_WRITE_UNCHANGED
, "write unchanged" },
1772 { BLK_PERM_RESIZE
, "resize" },
1773 { BLK_PERM_GRAPH_MOD
, "change children" },
1777 char *result
= g_strdup("");
1778 struct perm_name
*p
;
1780 for (p
= permissions
; p
->name
; p
++) {
1781 if (perm
& p
->perm
) {
1783 result
= g_strdup_printf("%s%s%s", old
, *old
? ", " : "", p
->name
);
1792 * Checks whether a new reference to @bs can be added if the new user requires
1793 * @new_used_perm/@new_shared_perm as its permissions. If @ignore_children is
1794 * set, the BdrvChild objects in this list are ignored in the calculations;
1795 * this allows checking permission updates for an existing reference.
1797 * Needs to be followed by a call to either bdrv_set_perm() or
1798 * bdrv_abort_perm_update(). */
1799 static int bdrv_check_update_perm(BlockDriverState
*bs
, BlockReopenQueue
*q
,
1800 uint64_t new_used_perm
,
1801 uint64_t new_shared_perm
,
1802 GSList
*ignore_children
, Error
**errp
)
1805 uint64_t cumulative_perms
= new_used_perm
;
1806 uint64_t cumulative_shared_perms
= new_shared_perm
;
1808 /* There is no reason why anyone couldn't tolerate write_unchanged */
1809 assert(new_shared_perm
& BLK_PERM_WRITE_UNCHANGED
);
1811 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
1812 if (g_slist_find(ignore_children
, c
)) {
1816 if ((new_used_perm
& c
->shared_perm
) != new_used_perm
) {
1817 char *user
= bdrv_child_user_desc(c
);
1818 char *perm_names
= bdrv_perm_names(new_used_perm
& ~c
->shared_perm
);
1819 error_setg(errp
, "Conflicts with use by %s as '%s', which does not "
1821 user
, c
->name
, perm_names
, bdrv_get_node_name(c
->bs
));
1827 if ((c
->perm
& new_shared_perm
) != c
->perm
) {
1828 char *user
= bdrv_child_user_desc(c
);
1829 char *perm_names
= bdrv_perm_names(c
->perm
& ~new_shared_perm
);
1830 error_setg(errp
, "Conflicts with use by %s as '%s', which uses "
1832 user
, c
->name
, perm_names
, bdrv_get_node_name(c
->bs
));
1838 cumulative_perms
|= c
->perm
;
1839 cumulative_shared_perms
&= c
->shared_perm
;
1842 return bdrv_check_perm(bs
, q
, cumulative_perms
, cumulative_shared_perms
,
1843 ignore_children
, errp
);
1846 /* Needs to be followed by a call to either bdrv_child_set_perm() or
1847 * bdrv_child_abort_perm_update(). */
1848 static int bdrv_child_check_perm(BdrvChild
*c
, BlockReopenQueue
*q
,
1849 uint64_t perm
, uint64_t shared
,
1850 GSList
*ignore_children
, Error
**errp
)
1854 ignore_children
= g_slist_prepend(g_slist_copy(ignore_children
), c
);
1855 ret
= bdrv_check_update_perm(c
->bs
, q
, perm
, shared
, ignore_children
, errp
);
1856 g_slist_free(ignore_children
);
1861 static void bdrv_child_set_perm(BdrvChild
*c
, uint64_t perm
, uint64_t shared
)
1863 uint64_t cumulative_perms
, cumulative_shared_perms
;
1866 c
->shared_perm
= shared
;
1868 bdrv_get_cumulative_perm(c
->bs
, &cumulative_perms
,
1869 &cumulative_shared_perms
);
1870 bdrv_set_perm(c
->bs
, cumulative_perms
, cumulative_shared_perms
);
1873 static void bdrv_child_abort_perm_update(BdrvChild
*c
)
1875 bdrv_abort_perm_update(c
->bs
);
1878 int bdrv_child_try_set_perm(BdrvChild
*c
, uint64_t perm
, uint64_t shared
,
1883 ret
= bdrv_child_check_perm(c
, NULL
, perm
, shared
, NULL
, errp
);
1885 bdrv_child_abort_perm_update(c
);
1889 bdrv_child_set_perm(c
, perm
, shared
);
1894 #define DEFAULT_PERM_PASSTHROUGH (BLK_PERM_CONSISTENT_READ \
1896 | BLK_PERM_WRITE_UNCHANGED \
1898 #define DEFAULT_PERM_UNCHANGED (BLK_PERM_ALL & ~DEFAULT_PERM_PASSTHROUGH)
1900 void bdrv_filter_default_perms(BlockDriverState
*bs
, BdrvChild
*c
,
1901 const BdrvChildRole
*role
,
1902 BlockReopenQueue
*reopen_queue
,
1903 uint64_t perm
, uint64_t shared
,
1904 uint64_t *nperm
, uint64_t *nshared
)
1907 *nperm
= perm
& DEFAULT_PERM_PASSTHROUGH
;
1908 *nshared
= (shared
& DEFAULT_PERM_PASSTHROUGH
) | DEFAULT_PERM_UNCHANGED
;
1912 *nperm
= (perm
& DEFAULT_PERM_PASSTHROUGH
) |
1913 (c
->perm
& DEFAULT_PERM_UNCHANGED
);
1914 *nshared
= (shared
& DEFAULT_PERM_PASSTHROUGH
) |
1915 (c
->shared_perm
& DEFAULT_PERM_UNCHANGED
);
1918 void bdrv_format_default_perms(BlockDriverState
*bs
, BdrvChild
*c
,
1919 const BdrvChildRole
*role
,
1920 BlockReopenQueue
*reopen_queue
,
1921 uint64_t perm
, uint64_t shared
,
1922 uint64_t *nperm
, uint64_t *nshared
)
1924 bool backing
= (role
== &child_backing
);
1925 assert(role
== &child_backing
|| role
== &child_file
);
1928 /* Apart from the modifications below, the same permissions are
1929 * forwarded and left alone as for filters */
1930 bdrv_filter_default_perms(bs
, c
, role
, reopen_queue
, perm
, shared
,
1933 /* Format drivers may touch metadata even if the guest doesn't write */
1934 if (bdrv_is_writable(bs
, reopen_queue
)) {
1935 perm
|= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
1938 /* bs->file always needs to be consistent because of the metadata. We
1939 * can never allow other users to resize or write to it. */
1940 perm
|= BLK_PERM_CONSISTENT_READ
;
1941 shared
&= ~(BLK_PERM_WRITE
| BLK_PERM_RESIZE
);
1943 /* We want consistent read from backing files if the parent needs it.
1944 * No other operations are performed on backing files. */
1945 perm
&= BLK_PERM_CONSISTENT_READ
;
1947 /* If the parent can deal with changing data, we're okay with a
1948 * writable and resizable backing file. */
1949 /* TODO Require !(perm & BLK_PERM_CONSISTENT_READ), too? */
1950 if (shared
& BLK_PERM_WRITE
) {
1951 shared
= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
1956 shared
|= BLK_PERM_CONSISTENT_READ
| BLK_PERM_GRAPH_MOD
|
1957 BLK_PERM_WRITE_UNCHANGED
;
1960 if (bs
->open_flags
& BDRV_O_INACTIVE
) {
1961 shared
|= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
1968 static void bdrv_replace_child_noperm(BdrvChild
*child
,
1969 BlockDriverState
*new_bs
)
1971 BlockDriverState
*old_bs
= child
->bs
;
1973 if (old_bs
&& new_bs
) {
1974 assert(bdrv_get_aio_context(old_bs
) == bdrv_get_aio_context(new_bs
));
1977 if (old_bs
->quiesce_counter
&& child
->role
->drained_end
) {
1978 child
->role
->drained_end(child
);
1980 if (child
->role
->detach
) {
1981 child
->role
->detach(child
);
1983 QLIST_REMOVE(child
, next_parent
);
1989 QLIST_INSERT_HEAD(&new_bs
->parents
, child
, next_parent
);
1990 if (new_bs
->quiesce_counter
&& child
->role
->drained_begin
) {
1991 child
->role
->drained_begin(child
);
1994 if (child
->role
->attach
) {
1995 child
->role
->attach(child
);
2001 * Updates @child to change its reference to point to @new_bs, including
2002 * checking and applying the necessary permisson updates both to the old node
2005 * NULL is passed as @new_bs for removing the reference before freeing @child.
2007 * If @new_bs is not NULL, bdrv_check_perm() must be called beforehand, as this
2008 * function uses bdrv_set_perm() to update the permissions according to the new
2009 * reference that @new_bs gets.
2011 static void bdrv_replace_child(BdrvChild
*child
, BlockDriverState
*new_bs
)
2013 BlockDriverState
*old_bs
= child
->bs
;
2014 uint64_t perm
, shared_perm
;
2016 bdrv_replace_child_noperm(child
, new_bs
);
2019 /* Update permissions for old node. This is guaranteed to succeed
2020 * because we're just taking a parent away, so we're loosening
2022 bdrv_get_cumulative_perm(old_bs
, &perm
, &shared_perm
);
2023 bdrv_check_perm(old_bs
, NULL
, perm
, shared_perm
, NULL
, &error_abort
);
2024 bdrv_set_perm(old_bs
, perm
, shared_perm
);
2028 bdrv_get_cumulative_perm(new_bs
, &perm
, &shared_perm
);
2029 bdrv_set_perm(new_bs
, perm
, shared_perm
);
2033 BdrvChild
*bdrv_root_attach_child(BlockDriverState
*child_bs
,
2034 const char *child_name
,
2035 const BdrvChildRole
*child_role
,
2036 uint64_t perm
, uint64_t shared_perm
,
2037 void *opaque
, Error
**errp
)
2042 ret
= bdrv_check_update_perm(child_bs
, NULL
, perm
, shared_perm
, NULL
, errp
);
2044 bdrv_abort_perm_update(child_bs
);
2048 child
= g_new(BdrvChild
, 1);
2049 *child
= (BdrvChild
) {
2051 .name
= g_strdup(child_name
),
2054 .shared_perm
= shared_perm
,
2058 /* This performs the matching bdrv_set_perm() for the above check. */
2059 bdrv_replace_child(child
, child_bs
);
2064 BdrvChild
*bdrv_attach_child(BlockDriverState
*parent_bs
,
2065 BlockDriverState
*child_bs
,
2066 const char *child_name
,
2067 const BdrvChildRole
*child_role
,
2071 uint64_t perm
, shared_perm
;
2073 bdrv_get_cumulative_perm(parent_bs
, &perm
, &shared_perm
);
2075 assert(parent_bs
->drv
);
2076 assert(bdrv_get_aio_context(parent_bs
) == bdrv_get_aio_context(child_bs
));
2077 bdrv_child_perm(parent_bs
, child_bs
, NULL
, child_role
, NULL
,
2078 perm
, shared_perm
, &perm
, &shared_perm
);
2080 child
= bdrv_root_attach_child(child_bs
, child_name
, child_role
,
2081 perm
, shared_perm
, parent_bs
, errp
);
2082 if (child
== NULL
) {
2086 QLIST_INSERT_HEAD(&parent_bs
->children
, child
, next
);
2090 static void bdrv_detach_child(BdrvChild
*child
)
2092 if (child
->next
.le_prev
) {
2093 QLIST_REMOVE(child
, next
);
2094 child
->next
.le_prev
= NULL
;
2097 bdrv_replace_child(child
, NULL
);
2099 g_free(child
->name
);
2103 void bdrv_root_unref_child(BdrvChild
*child
)
2105 BlockDriverState
*child_bs
;
2107 child_bs
= child
->bs
;
2108 bdrv_detach_child(child
);
2109 bdrv_unref(child_bs
);
2112 void bdrv_unref_child(BlockDriverState
*parent
, BdrvChild
*child
)
2114 if (child
== NULL
) {
2118 if (child
->bs
->inherits_from
== parent
) {
2121 /* Remove inherits_from only when the last reference between parent and
2122 * child->bs goes away. */
2123 QLIST_FOREACH(c
, &parent
->children
, next
) {
2124 if (c
!= child
&& c
->bs
== child
->bs
) {
2129 child
->bs
->inherits_from
= NULL
;
2133 bdrv_root_unref_child(child
);
2137 static void bdrv_parent_cb_change_media(BlockDriverState
*bs
, bool load
)
2140 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
2141 if (c
->role
->change_media
) {
2142 c
->role
->change_media(c
, load
);
2147 static void bdrv_parent_cb_resize(BlockDriverState
*bs
)
2150 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
2151 if (c
->role
->resize
) {
2158 * Sets the backing file link of a BDS. A new reference is created; callers
2159 * which don't need their own reference any more must call bdrv_unref().
2161 void bdrv_set_backing_hd(BlockDriverState
*bs
, BlockDriverState
*backing_hd
,
2165 bdrv_ref(backing_hd
);
2169 bdrv_unref_child(bs
, bs
->backing
);
2177 bs
->backing
= bdrv_attach_child(bs
, backing_hd
, "backing", &child_backing
,
2180 bdrv_unref(backing_hd
);
2183 bdrv_refresh_filename(bs
);
2186 bdrv_refresh_limits(bs
, NULL
);
2190 * Opens the backing file for a BlockDriverState if not yet open
2192 * bdref_key specifies the key for the image's BlockdevRef in the options QDict.
2193 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
2194 * itself, all options starting with "${bdref_key}." are considered part of the
2197 * TODO Can this be unified with bdrv_open_image()?
2199 int bdrv_open_backing_file(BlockDriverState
*bs
, QDict
*parent_options
,
2200 const char *bdref_key
, Error
**errp
)
2202 char *backing_filename
= g_malloc0(PATH_MAX
);
2203 char *bdref_key_dot
;
2204 const char *reference
= NULL
;
2206 BlockDriverState
*backing_hd
;
2208 QDict
*tmp_parent_options
= NULL
;
2209 Error
*local_err
= NULL
;
2211 if (bs
->backing
!= NULL
) {
2215 /* NULL means an empty set of options */
2216 if (parent_options
== NULL
) {
2217 tmp_parent_options
= qdict_new();
2218 parent_options
= tmp_parent_options
;
2221 bs
->open_flags
&= ~BDRV_O_NO_BACKING
;
2223 bdref_key_dot
= g_strdup_printf("%s.", bdref_key
);
2224 qdict_extract_subqdict(parent_options
, &options
, bdref_key_dot
);
2225 g_free(bdref_key_dot
);
2228 * Caution: while qdict_get_try_str() is fine, getting non-string
2229 * types would require more care. When @parent_options come from
2230 * -blockdev or blockdev_add, its members are typed according to
2231 * the QAPI schema, but when they come from -drive, they're all
2234 reference
= qdict_get_try_str(parent_options
, bdref_key
);
2235 if (reference
|| qdict_haskey(options
, "file.filename")) {
2236 backing_filename
[0] = '\0';
2237 } else if (bs
->backing_file
[0] == '\0' && qdict_size(options
) == 0) {
2241 bdrv_get_full_backing_filename(bs
, backing_filename
, PATH_MAX
,
2245 error_propagate(errp
, local_err
);
2251 if (!bs
->drv
|| !bs
->drv
->supports_backing
) {
2253 error_setg(errp
, "Driver doesn't support backing files");
2259 bs
->backing_format
[0] != '\0' && !qdict_haskey(options
, "driver")) {
2260 qdict_put_str(options
, "driver", bs
->backing_format
);
2263 backing_hd
= bdrv_open_inherit(*backing_filename
? backing_filename
: NULL
,
2264 reference
, options
, 0, bs
, &child_backing
,
2267 bs
->open_flags
|= BDRV_O_NO_BACKING
;
2268 error_prepend(errp
, "Could not open backing file: ");
2272 bdrv_set_aio_context(backing_hd
, bdrv_get_aio_context(bs
));
2274 /* Hook up the backing file link; drop our reference, bs owns the
2275 * backing_hd reference now */
2276 bdrv_set_backing_hd(bs
, backing_hd
, &local_err
);
2277 bdrv_unref(backing_hd
);
2279 error_propagate(errp
, local_err
);
2284 qdict_del(parent_options
, bdref_key
);
2287 g_free(backing_filename
);
2288 QDECREF(tmp_parent_options
);
2292 static BlockDriverState
*
2293 bdrv_open_child_bs(const char *filename
, QDict
*options
, const char *bdref_key
,
2294 BlockDriverState
*parent
, const BdrvChildRole
*child_role
,
2295 bool allow_none
, Error
**errp
)
2297 BlockDriverState
*bs
= NULL
;
2298 QDict
*image_options
;
2299 char *bdref_key_dot
;
2300 const char *reference
;
2302 assert(child_role
!= NULL
);
2304 bdref_key_dot
= g_strdup_printf("%s.", bdref_key
);
2305 qdict_extract_subqdict(options
, &image_options
, bdref_key_dot
);
2306 g_free(bdref_key_dot
);
2309 * Caution: while qdict_get_try_str() is fine, getting non-string
2310 * types would require more care. When @options come from
2311 * -blockdev or blockdev_add, its members are typed according to
2312 * the QAPI schema, but when they come from -drive, they're all
2315 reference
= qdict_get_try_str(options
, bdref_key
);
2316 if (!filename
&& !reference
&& !qdict_size(image_options
)) {
2318 error_setg(errp
, "A block device must be specified for \"%s\"",
2321 QDECREF(image_options
);
2325 bs
= bdrv_open_inherit(filename
, reference
, image_options
, 0,
2326 parent
, child_role
, errp
);
2332 qdict_del(options
, bdref_key
);
2337 * Opens a disk image whose options are given as BlockdevRef in another block
2340 * If allow_none is true, no image will be opened if filename is false and no
2341 * BlockdevRef is given. NULL will be returned, but errp remains unset.
2343 * bdrev_key specifies the key for the image's BlockdevRef in the options QDict.
2344 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
2345 * itself, all options starting with "${bdref_key}." are considered part of the
2348 * The BlockdevRef will be removed from the options QDict.
2350 BdrvChild
*bdrv_open_child(const char *filename
,
2351 QDict
*options
, const char *bdref_key
,
2352 BlockDriverState
*parent
,
2353 const BdrvChildRole
*child_role
,
2354 bool allow_none
, Error
**errp
)
2357 BlockDriverState
*bs
;
2359 bs
= bdrv_open_child_bs(filename
, options
, bdref_key
, parent
, child_role
,
2365 c
= bdrv_attach_child(parent
, bs
, bdref_key
, child_role
, errp
);
2374 static BlockDriverState
*bdrv_append_temp_snapshot(BlockDriverState
*bs
,
2376 QDict
*snapshot_options
,
2379 /* TODO: extra byte is a hack to ensure MAX_PATH space on Windows. */
2380 char *tmp_filename
= g_malloc0(PATH_MAX
+ 1);
2382 QemuOpts
*opts
= NULL
;
2383 BlockDriverState
*bs_snapshot
= NULL
;
2384 Error
*local_err
= NULL
;
2387 /* if snapshot, we create a temporary backing file and open it
2388 instead of opening 'filename' directly */
2390 /* Get the required size from the image */
2391 total_size
= bdrv_getlength(bs
);
2392 if (total_size
< 0) {
2393 error_setg_errno(errp
, -total_size
, "Could not get image size");
2397 /* Create the temporary image */
2398 ret
= get_tmp_filename(tmp_filename
, PATH_MAX
+ 1);
2400 error_setg_errno(errp
, -ret
, "Could not get temporary filename");
2404 opts
= qemu_opts_create(bdrv_qcow2
.create_opts
, NULL
, 0,
2406 qemu_opt_set_number(opts
, BLOCK_OPT_SIZE
, total_size
, &error_abort
);
2407 ret
= bdrv_create(&bdrv_qcow2
, tmp_filename
, opts
, errp
);
2408 qemu_opts_del(opts
);
2410 error_prepend(errp
, "Could not create temporary overlay '%s': ",
2415 /* Prepare options QDict for the temporary file */
2416 qdict_put_str(snapshot_options
, "file.driver", "file");
2417 qdict_put_str(snapshot_options
, "file.filename", tmp_filename
);
2418 qdict_put_str(snapshot_options
, "driver", "qcow2");
2420 bs_snapshot
= bdrv_open(NULL
, NULL
, snapshot_options
, flags
, errp
);
2421 snapshot_options
= NULL
;
2426 /* bdrv_append() consumes a strong reference to bs_snapshot
2427 * (i.e. it will call bdrv_unref() on it) even on error, so in
2428 * order to be able to return one, we have to increase
2429 * bs_snapshot's refcount here */
2430 bdrv_ref(bs_snapshot
);
2431 bdrv_append(bs_snapshot
, bs
, &local_err
);
2433 error_propagate(errp
, local_err
);
2439 QDECREF(snapshot_options
);
2440 g_free(tmp_filename
);
2445 * Opens a disk image (raw, qcow2, vmdk, ...)
2447 * options is a QDict of options to pass to the block drivers, or NULL for an
2448 * empty set of options. The reference to the QDict belongs to the block layer
2449 * after the call (even on failure), so if the caller intends to reuse the
2450 * dictionary, it needs to use QINCREF() before calling bdrv_open.
2452 * If *pbs is NULL, a new BDS will be created with a pointer to it stored there.
2453 * If it is not NULL, the referenced BDS will be reused.
2455 * The reference parameter may be used to specify an existing block device which
2456 * should be opened. If specified, neither options nor a filename may be given,
2457 * nor can an existing BDS be reused (that is, *pbs has to be NULL).
2459 static BlockDriverState
*bdrv_open_inherit(const char *filename
,
2460 const char *reference
,
2461 QDict
*options
, int flags
,
2462 BlockDriverState
*parent
,
2463 const BdrvChildRole
*child_role
,
2467 BlockBackend
*file
= NULL
;
2468 BlockDriverState
*bs
;
2469 BlockDriver
*drv
= NULL
;
2470 const char *drvname
;
2471 const char *backing
;
2472 Error
*local_err
= NULL
;
2473 QDict
*snapshot_options
= NULL
;
2474 int snapshot_flags
= 0;
2476 assert(!child_role
|| !flags
);
2477 assert(!child_role
== !parent
);
2480 bool options_non_empty
= options
? qdict_size(options
) : false;
2483 if (filename
|| options_non_empty
) {
2484 error_setg(errp
, "Cannot reference an existing block device with "
2485 "additional options or a new filename");
2489 bs
= bdrv_lookup_bs(reference
, reference
, errp
);
2500 /* NULL means an empty set of options */
2501 if (options
== NULL
) {
2502 options
= qdict_new();
2505 /* json: syntax counts as explicit options, as if in the QDict */
2506 parse_json_protocol(options
, &filename
, &local_err
);
2511 bs
->explicit_options
= qdict_clone_shallow(options
);
2514 bs
->inherits_from
= parent
;
2515 child_role
->inherit_options(&flags
, options
,
2516 parent
->open_flags
, parent
->options
);
2519 ret
= bdrv_fill_options(&options
, filename
, &flags
, &local_err
);
2525 * Set the BDRV_O_RDWR and BDRV_O_ALLOW_RDWR flags.
2526 * Caution: getting a boolean member of @options requires care.
2527 * When @options come from -blockdev or blockdev_add, members are
2528 * typed according to the QAPI schema, but when they come from
2529 * -drive, they're all QString.
2531 if (g_strcmp0(qdict_get_try_str(options
, BDRV_OPT_READ_ONLY
), "on") &&
2532 !qdict_get_try_bool(options
, BDRV_OPT_READ_ONLY
, false)) {
2533 flags
|= (BDRV_O_RDWR
| BDRV_O_ALLOW_RDWR
);
2535 flags
&= ~BDRV_O_RDWR
;
2538 if (flags
& BDRV_O_SNAPSHOT
) {
2539 snapshot_options
= qdict_new();
2540 bdrv_temp_snapshot_options(&snapshot_flags
, snapshot_options
,
2542 /* Let bdrv_backing_options() override "read-only" */
2543 qdict_del(options
, BDRV_OPT_READ_ONLY
);
2544 bdrv_backing_options(&flags
, options
, flags
, options
);
2547 bs
->open_flags
= flags
;
2548 bs
->options
= options
;
2549 options
= qdict_clone_shallow(options
);
2551 /* Find the right image format driver */
2552 /* See cautionary note on accessing @options above */
2553 drvname
= qdict_get_try_str(options
, "driver");
2555 drv
= bdrv_find_format(drvname
);
2557 error_setg(errp
, "Unknown driver: '%s'", drvname
);
2562 assert(drvname
|| !(flags
& BDRV_O_PROTOCOL
));
2564 /* See cautionary note on accessing @options above */
2565 backing
= qdict_get_try_str(options
, "backing");
2566 if (backing
&& *backing
== '\0') {
2567 flags
|= BDRV_O_NO_BACKING
;
2568 qdict_del(options
, "backing");
2571 /* Open image file without format layer. This BlockBackend is only used for
2572 * probing, the block drivers will do their own bdrv_open_child() for the
2573 * same BDS, which is why we put the node name back into options. */
2574 if ((flags
& BDRV_O_PROTOCOL
) == 0) {
2575 BlockDriverState
*file_bs
;
2577 file_bs
= bdrv_open_child_bs(filename
, options
, "file", bs
,
2578 &child_file
, true, &local_err
);
2582 if (file_bs
!= NULL
) {
2583 /* Not requesting BLK_PERM_CONSISTENT_READ because we're only
2584 * looking at the header to guess the image format. This works even
2585 * in cases where a guest would not see a consistent state. */
2586 file
= blk_new(0, BLK_PERM_ALL
);
2587 blk_insert_bs(file
, file_bs
, &local_err
);
2588 bdrv_unref(file_bs
);
2593 qdict_put_str(options
, "file", bdrv_get_node_name(file_bs
));
2597 /* Image format probing */
2600 ret
= find_image_format(file
, filename
, &drv
, &local_err
);
2605 * This option update would logically belong in bdrv_fill_options(),
2606 * but we first need to open bs->file for the probing to work, while
2607 * opening bs->file already requires the (mostly) final set of options
2608 * so that cache mode etc. can be inherited.
2610 * Adding the driver later is somewhat ugly, but it's not an option
2611 * that would ever be inherited, so it's correct. We just need to make
2612 * sure to update both bs->options (which has the full effective
2613 * options for bs) and options (which has file.* already removed).
2615 qdict_put_str(bs
->options
, "driver", drv
->format_name
);
2616 qdict_put_str(options
, "driver", drv
->format_name
);
2618 error_setg(errp
, "Must specify either driver or file");
2622 /* BDRV_O_PROTOCOL must be set iff a protocol BDS is about to be created */
2623 assert(!!(flags
& BDRV_O_PROTOCOL
) == !!drv
->bdrv_file_open
);
2624 /* file must be NULL if a protocol BDS is about to be created
2625 * (the inverse results in an error message from bdrv_open_common()) */
2626 assert(!(flags
& BDRV_O_PROTOCOL
) || !file
);
2628 /* Open the image */
2629 ret
= bdrv_open_common(bs
, file
, options
, &local_err
);
2639 /* If there is a backing file, use it */
2640 if ((flags
& BDRV_O_NO_BACKING
) == 0) {
2641 ret
= bdrv_open_backing_file(bs
, options
, "backing", &local_err
);
2643 goto close_and_fail
;
2647 bdrv_refresh_filename(bs
);
2649 /* Check if any unknown options were used */
2650 if (qdict_size(options
) != 0) {
2651 const QDictEntry
*entry
= qdict_first(options
);
2652 if (flags
& BDRV_O_PROTOCOL
) {
2653 error_setg(errp
, "Block protocol '%s' doesn't support the option "
2654 "'%s'", drv
->format_name
, entry
->key
);
2657 "Block format '%s' does not support the option '%s'",
2658 drv
->format_name
, entry
->key
);
2661 goto close_and_fail
;
2664 bdrv_parent_cb_change_media(bs
, true);
2668 /* For snapshot=on, create a temporary qcow2 overlay. bs points to the
2669 * temporary snapshot afterwards. */
2670 if (snapshot_flags
) {
2671 BlockDriverState
*snapshot_bs
;
2672 snapshot_bs
= bdrv_append_temp_snapshot(bs
, snapshot_flags
,
2673 snapshot_options
, &local_err
);
2674 snapshot_options
= NULL
;
2676 goto close_and_fail
;
2678 /* We are not going to return bs but the overlay on top of it
2679 * (snapshot_bs); thus, we have to drop the strong reference to bs
2680 * (which we obtained by calling bdrv_new()). bs will not be deleted,
2681 * though, because the overlay still has a reference to it. */
2690 QDECREF(snapshot_options
);
2691 QDECREF(bs
->explicit_options
);
2692 QDECREF(bs
->options
);
2695 bs
->explicit_options
= NULL
;
2697 error_propagate(errp
, local_err
);
2702 QDECREF(snapshot_options
);
2704 error_propagate(errp
, local_err
);
2708 BlockDriverState
*bdrv_open(const char *filename
, const char *reference
,
2709 QDict
*options
, int flags
, Error
**errp
)
2711 return bdrv_open_inherit(filename
, reference
, options
, flags
, NULL
,
2716 * Adds a BlockDriverState to a simple queue for an atomic, transactional
2717 * reopen of multiple devices.
2719 * bs_queue can either be an existing BlockReopenQueue that has had QSIMPLE_INIT
2720 * already performed, or alternatively may be NULL a new BlockReopenQueue will
2721 * be created and initialized. This newly created BlockReopenQueue should be
2722 * passed back in for subsequent calls that are intended to be of the same
2725 * bs is the BlockDriverState to add to the reopen queue.
2727 * options contains the changed options for the associated bs
2728 * (the BlockReopenQueue takes ownership)
2730 * flags contains the open flags for the associated bs
2732 * returns a pointer to bs_queue, which is either the newly allocated
2733 * bs_queue, or the existing bs_queue being used.
2736 static BlockReopenQueue
*bdrv_reopen_queue_child(BlockReopenQueue
*bs_queue
,
2737 BlockDriverState
*bs
,
2740 const BdrvChildRole
*role
,
2741 QDict
*parent_options
,
2746 BlockReopenQueueEntry
*bs_entry
;
2748 QDict
*old_options
, *explicit_options
;
2750 if (bs_queue
== NULL
) {
2751 bs_queue
= g_new0(BlockReopenQueue
, 1);
2752 QSIMPLEQ_INIT(bs_queue
);
2756 options
= qdict_new();
2759 /* Check if this BlockDriverState is already in the queue */
2760 QSIMPLEQ_FOREACH(bs_entry
, bs_queue
, entry
) {
2761 if (bs
== bs_entry
->state
.bs
) {
2767 * Precedence of options:
2768 * 1. Explicitly passed in options (highest)
2769 * 2. Set in flags (only for top level)
2770 * 3. Retained from explicitly set options of bs
2771 * 4. Inherited from parent node
2772 * 5. Retained from effective options of bs
2775 if (!parent_options
) {
2777 * Any setting represented by flags is always updated. If the
2778 * corresponding QDict option is set, it takes precedence. Otherwise
2779 * the flag is translated into a QDict option. The old setting of bs is
2782 update_options_from_flags(options
, flags
);
2785 /* Old explicitly set values (don't overwrite by inherited value) */
2787 old_options
= qdict_clone_shallow(bs_entry
->state
.explicit_options
);
2789 old_options
= qdict_clone_shallow(bs
->explicit_options
);
2791 bdrv_join_options(bs
, options
, old_options
);
2792 QDECREF(old_options
);
2794 explicit_options
= qdict_clone_shallow(options
);
2796 /* Inherit from parent node */
2797 if (parent_options
) {
2799 role
->inherit_options(&flags
, options
, parent_flags
, parent_options
);
2802 /* Old values are used for options that aren't set yet */
2803 old_options
= qdict_clone_shallow(bs
->options
);
2804 bdrv_join_options(bs
, options
, old_options
);
2805 QDECREF(old_options
);
2807 /* bdrv_open_inherit() sets and clears some additional flags internally */
2808 flags
&= ~BDRV_O_PROTOCOL
;
2809 if (flags
& BDRV_O_RDWR
) {
2810 flags
|= BDRV_O_ALLOW_RDWR
;
2814 bs_entry
= g_new0(BlockReopenQueueEntry
, 1);
2815 QSIMPLEQ_INSERT_TAIL(bs_queue
, bs_entry
, entry
);
2817 QDECREF(bs_entry
->state
.options
);
2818 QDECREF(bs_entry
->state
.explicit_options
);
2821 bs_entry
->state
.bs
= bs
;
2822 bs_entry
->state
.options
= options
;
2823 bs_entry
->state
.explicit_options
= explicit_options
;
2824 bs_entry
->state
.flags
= flags
;
2826 /* This needs to be overwritten in bdrv_reopen_prepare() */
2827 bs_entry
->state
.perm
= UINT64_MAX
;
2828 bs_entry
->state
.shared_perm
= 0;
2830 QLIST_FOREACH(child
, &bs
->children
, next
) {
2831 QDict
*new_child_options
;
2832 char *child_key_dot
;
2834 /* reopen can only change the options of block devices that were
2835 * implicitly created and inherited options. For other (referenced)
2836 * block devices, a syntax like "backing.foo" results in an error. */
2837 if (child
->bs
->inherits_from
!= bs
) {
2841 child_key_dot
= g_strdup_printf("%s.", child
->name
);
2842 qdict_extract_subqdict(options
, &new_child_options
, child_key_dot
);
2843 g_free(child_key_dot
);
2845 bdrv_reopen_queue_child(bs_queue
, child
->bs
, new_child_options
, 0,
2846 child
->role
, options
, flags
);
2852 BlockReopenQueue
*bdrv_reopen_queue(BlockReopenQueue
*bs_queue
,
2853 BlockDriverState
*bs
,
2854 QDict
*options
, int flags
)
2856 return bdrv_reopen_queue_child(bs_queue
, bs
, options
, flags
,
2861 * Reopen multiple BlockDriverStates atomically & transactionally.
2863 * The queue passed in (bs_queue) must have been built up previous
2864 * via bdrv_reopen_queue().
2866 * Reopens all BDS specified in the queue, with the appropriate
2867 * flags. All devices are prepared for reopen, and failure of any
2868 * device will cause all device changes to be abandonded, and intermediate
2871 * If all devices prepare successfully, then the changes are committed
2875 int bdrv_reopen_multiple(AioContext
*ctx
, BlockReopenQueue
*bs_queue
, Error
**errp
)
2878 BlockReopenQueueEntry
*bs_entry
, *next
;
2879 Error
*local_err
= NULL
;
2881 assert(bs_queue
!= NULL
);
2883 aio_context_release(ctx
);
2884 bdrv_drain_all_begin();
2885 aio_context_acquire(ctx
);
2887 QSIMPLEQ_FOREACH(bs_entry
, bs_queue
, entry
) {
2888 if (bdrv_reopen_prepare(&bs_entry
->state
, bs_queue
, &local_err
)) {
2889 error_propagate(errp
, local_err
);
2892 bs_entry
->prepared
= true;
2895 /* If we reach this point, we have success and just need to apply the
2898 QSIMPLEQ_FOREACH(bs_entry
, bs_queue
, entry
) {
2899 bdrv_reopen_commit(&bs_entry
->state
);
2905 QSIMPLEQ_FOREACH_SAFE(bs_entry
, bs_queue
, entry
, next
) {
2906 if (ret
&& bs_entry
->prepared
) {
2907 bdrv_reopen_abort(&bs_entry
->state
);
2909 QDECREF(bs_entry
->state
.explicit_options
);
2911 QDECREF(bs_entry
->state
.options
);
2916 bdrv_drain_all_end();
2922 /* Reopen a single BlockDriverState with the specified flags. */
2923 int bdrv_reopen(BlockDriverState
*bs
, int bdrv_flags
, Error
**errp
)
2926 Error
*local_err
= NULL
;
2927 BlockReopenQueue
*queue
= bdrv_reopen_queue(NULL
, bs
, NULL
, bdrv_flags
);
2929 ret
= bdrv_reopen_multiple(bdrv_get_aio_context(bs
), queue
, &local_err
);
2930 if (local_err
!= NULL
) {
2931 error_propagate(errp
, local_err
);
2936 static BlockReopenQueueEntry
*find_parent_in_reopen_queue(BlockReopenQueue
*q
,
2939 BlockReopenQueueEntry
*entry
;
2941 QSIMPLEQ_FOREACH(entry
, q
, entry
) {
2942 BlockDriverState
*bs
= entry
->state
.bs
;
2945 QLIST_FOREACH(child
, &bs
->children
, next
) {
2955 static void bdrv_reopen_perm(BlockReopenQueue
*q
, BlockDriverState
*bs
,
2956 uint64_t *perm
, uint64_t *shared
)
2959 BlockReopenQueueEntry
*parent
;
2960 uint64_t cumulative_perms
= 0;
2961 uint64_t cumulative_shared_perms
= BLK_PERM_ALL
;
2963 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
2964 parent
= find_parent_in_reopen_queue(q
, c
);
2966 cumulative_perms
|= c
->perm
;
2967 cumulative_shared_perms
&= c
->shared_perm
;
2969 uint64_t nperm
, nshared
;
2971 bdrv_child_perm(parent
->state
.bs
, bs
, c
, c
->role
, q
,
2972 parent
->state
.perm
, parent
->state
.shared_perm
,
2975 cumulative_perms
|= nperm
;
2976 cumulative_shared_perms
&= nshared
;
2979 *perm
= cumulative_perms
;
2980 *shared
= cumulative_shared_perms
;
2984 * Prepares a BlockDriverState for reopen. All changes are staged in the
2985 * 'opaque' field of the BDRVReopenState, which is used and allocated by
2986 * the block driver layer .bdrv_reopen_prepare()
2988 * bs is the BlockDriverState to reopen
2989 * flags are the new open flags
2990 * queue is the reopen queue
2992 * Returns 0 on success, non-zero on error. On error errp will be set
2995 * On failure, bdrv_reopen_abort() will be called to clean up any data.
2996 * It is the responsibility of the caller to then call the abort() or
2997 * commit() for any other BDS that have been left in a prepare() state
3000 int bdrv_reopen_prepare(BDRVReopenState
*reopen_state
, BlockReopenQueue
*queue
,
3004 Error
*local_err
= NULL
;
3010 assert(reopen_state
!= NULL
);
3011 assert(reopen_state
->bs
->drv
!= NULL
);
3012 drv
= reopen_state
->bs
->drv
;
3014 /* Process generic block layer options */
3015 opts
= qemu_opts_create(&bdrv_runtime_opts
, NULL
, 0, &error_abort
);
3016 qemu_opts_absorb_qdict(opts
, reopen_state
->options
, &local_err
);
3018 error_propagate(errp
, local_err
);
3023 update_flags_from_options(&reopen_state
->flags
, opts
);
3025 /* node-name and driver must be unchanged. Put them back into the QDict, so
3026 * that they are checked at the end of this function. */
3027 value
= qemu_opt_get(opts
, "node-name");
3029 qdict_put_str(reopen_state
->options
, "node-name", value
);
3032 value
= qemu_opt_get(opts
, "driver");
3034 qdict_put_str(reopen_state
->options
, "driver", value
);
3037 /* If we are to stay read-only, do not allow permission change
3038 * to r/w. Attempting to set to r/w may fail if either BDRV_O_ALLOW_RDWR is
3039 * not set, or if the BDS still has copy_on_read enabled */
3040 read_only
= !(reopen_state
->flags
& BDRV_O_RDWR
);
3041 ret
= bdrv_can_set_read_only(reopen_state
->bs
, read_only
, true, &local_err
);
3043 error_propagate(errp
, local_err
);
3047 /* Calculate required permissions after reopening */
3048 bdrv_reopen_perm(queue
, reopen_state
->bs
,
3049 &reopen_state
->perm
, &reopen_state
->shared_perm
);
3051 ret
= bdrv_flush(reopen_state
->bs
);
3053 error_setg_errno(errp
, -ret
, "Error flushing drive");
3057 if (drv
->bdrv_reopen_prepare
) {
3058 ret
= drv
->bdrv_reopen_prepare(reopen_state
, queue
, &local_err
);
3060 if (local_err
!= NULL
) {
3061 error_propagate(errp
, local_err
);
3063 error_setg(errp
, "failed while preparing to reopen image '%s'",
3064 reopen_state
->bs
->filename
);
3069 /* It is currently mandatory to have a bdrv_reopen_prepare()
3070 * handler for each supported drv. */
3071 error_setg(errp
, "Block format '%s' used by node '%s' "
3072 "does not support reopening files", drv
->format_name
,
3073 bdrv_get_device_or_node_name(reopen_state
->bs
));
3078 /* Options that are not handled are only okay if they are unchanged
3079 * compared to the old state. It is expected that some options are only
3080 * used for the initial open, but not reopen (e.g. filename) */
3081 if (qdict_size(reopen_state
->options
)) {
3082 const QDictEntry
*entry
= qdict_first(reopen_state
->options
);
3085 QObject
*new = entry
->value
;
3086 QObject
*old
= qdict_get(reopen_state
->bs
->options
, entry
->key
);
3089 * TODO: When using -drive to specify blockdev options, all values
3090 * will be strings; however, when using -blockdev, blockdev-add or
3091 * filenames using the json:{} pseudo-protocol, they will be
3093 * In contrast, reopening options are (currently) always strings
3094 * (because you can only specify them through qemu-io; all other
3095 * callers do not specify any options).
3096 * Therefore, when using anything other than -drive to create a BDS,
3097 * this cannot detect non-string options as unchanged, because
3098 * qobject_is_equal() always returns false for objects of different
3099 * type. In the future, this should be remedied by correctly typing
3100 * all options. For now, this is not too big of an issue because
3101 * the user can simply omit options which cannot be changed anyway,
3102 * so they will stay unchanged.
3104 if (!qobject_is_equal(new, old
)) {
3105 error_setg(errp
, "Cannot change the option '%s'", entry
->key
);
3109 } while ((entry
= qdict_next(reopen_state
->options
, entry
)));
3112 ret
= bdrv_check_perm(reopen_state
->bs
, queue
, reopen_state
->perm
,
3113 reopen_state
->shared_perm
, NULL
, errp
);
3121 qemu_opts_del(opts
);
3126 * Takes the staged changes for the reopen from bdrv_reopen_prepare(), and
3127 * makes them final by swapping the staging BlockDriverState contents into
3128 * the active BlockDriverState contents.
3130 void bdrv_reopen_commit(BDRVReopenState
*reopen_state
)
3133 BlockDriverState
*bs
;
3134 bool old_can_write
, new_can_write
;
3136 assert(reopen_state
!= NULL
);
3137 bs
= reopen_state
->bs
;
3139 assert(drv
!= NULL
);
3142 !bdrv_is_read_only(bs
) && !(bdrv_get_flags(bs
) & BDRV_O_INACTIVE
);
3144 /* If there are any driver level actions to take */
3145 if (drv
->bdrv_reopen_commit
) {
3146 drv
->bdrv_reopen_commit(reopen_state
);
3149 /* set BDS specific flags now */
3150 QDECREF(bs
->explicit_options
);
3152 bs
->explicit_options
= reopen_state
->explicit_options
;
3153 bs
->open_flags
= reopen_state
->flags
;
3154 bs
->read_only
= !(reopen_state
->flags
& BDRV_O_RDWR
);
3156 bdrv_refresh_limits(bs
, NULL
);
3158 bdrv_set_perm(reopen_state
->bs
, reopen_state
->perm
,
3159 reopen_state
->shared_perm
);
3162 !bdrv_is_read_only(bs
) && !(bdrv_get_flags(bs
) & BDRV_O_INACTIVE
);
3163 if (!old_can_write
&& new_can_write
&& drv
->bdrv_reopen_bitmaps_rw
) {
3164 Error
*local_err
= NULL
;
3165 if (drv
->bdrv_reopen_bitmaps_rw(bs
, &local_err
) < 0) {
3166 /* This is not fatal, bitmaps just left read-only, so all following
3167 * writes will fail. User can remove read-only bitmaps to unblock
3170 error_reportf_err(local_err
,
3171 "%s: Failed to make dirty bitmaps writable: ",
3172 bdrv_get_node_name(bs
));
3178 * Abort the reopen, and delete and free the staged changes in
3181 void bdrv_reopen_abort(BDRVReopenState
*reopen_state
)
3185 assert(reopen_state
!= NULL
);
3186 drv
= reopen_state
->bs
->drv
;
3187 assert(drv
!= NULL
);
3189 if (drv
->bdrv_reopen_abort
) {
3190 drv
->bdrv_reopen_abort(reopen_state
);
3193 QDECREF(reopen_state
->explicit_options
);
3195 bdrv_abort_perm_update(reopen_state
->bs
);
3199 static void bdrv_close(BlockDriverState
*bs
)
3201 BdrvAioNotifier
*ban
, *ban_next
;
3202 BdrvChild
*child
, *next
;
3205 assert(!bs
->refcnt
);
3207 bdrv_drained_begin(bs
); /* complete I/O */
3209 bdrv_drain(bs
); /* in case flush left pending I/O */
3212 bs
->drv
->bdrv_close(bs
);
3216 bdrv_set_backing_hd(bs
, NULL
, &error_abort
);
3218 if (bs
->file
!= NULL
) {
3219 bdrv_unref_child(bs
, bs
->file
);
3223 QLIST_FOREACH_SAFE(child
, &bs
->children
, next
, next
) {
3224 /* TODO Remove bdrv_unref() from drivers' close function and use
3225 * bdrv_unref_child() here */
3226 if (child
->bs
->inherits_from
== bs
) {
3227 child
->bs
->inherits_from
= NULL
;
3229 bdrv_detach_child(child
);
3234 atomic_set(&bs
->copy_on_read
, 0);
3235 bs
->backing_file
[0] = '\0';
3236 bs
->backing_format
[0] = '\0';
3237 bs
->total_sectors
= 0;
3238 bs
->encrypted
= false;
3240 QDECREF(bs
->options
);
3241 QDECREF(bs
->explicit_options
);
3243 bs
->explicit_options
= NULL
;
3244 QDECREF(bs
->full_open_options
);
3245 bs
->full_open_options
= NULL
;
3247 bdrv_release_named_dirty_bitmaps(bs
);
3248 assert(QLIST_EMPTY(&bs
->dirty_bitmaps
));
3250 QLIST_FOREACH_SAFE(ban
, &bs
->aio_notifiers
, list
, ban_next
) {
3253 QLIST_INIT(&bs
->aio_notifiers
);
3254 bdrv_drained_end(bs
);
3257 void bdrv_close_all(void)
3259 block_job_cancel_sync_all();
3260 nbd_export_close_all();
3262 /* Drop references from requests still in flight, such as canceled block
3263 * jobs whose AIO context has not been polled yet */
3266 blk_remove_all_bs();
3267 blockdev_close_all_bdrv_states();
3269 assert(QTAILQ_EMPTY(&all_bdrv_states
));
3272 static bool should_update_child(BdrvChild
*c
, BlockDriverState
*to
)
3276 if (c
->role
->stay_at_node
) {
3280 if (c
->role
== &child_backing
) {
3281 /* If @from is a backing file of @to, ignore the child to avoid
3282 * creating a loop. We only want to change the pointer of other
3284 QLIST_FOREACH(to_c
, &to
->children
, next
) {
3297 void bdrv_replace_node(BlockDriverState
*from
, BlockDriverState
*to
,
3300 BdrvChild
*c
, *next
;
3301 GSList
*list
= NULL
, *p
;
3302 uint64_t old_perm
, old_shared
;
3303 uint64_t perm
= 0, shared
= BLK_PERM_ALL
;
3306 assert(!atomic_read(&from
->in_flight
));
3307 assert(!atomic_read(&to
->in_flight
));
3309 /* Make sure that @from doesn't go away until we have successfully attached
3310 * all of its parents to @to. */
3313 /* Put all parents into @list and calculate their cumulative permissions */
3314 QLIST_FOREACH_SAFE(c
, &from
->parents
, next_parent
, next
) {
3315 if (!should_update_child(c
, to
)) {
3318 list
= g_slist_prepend(list
, c
);
3320 shared
&= c
->shared_perm
;
3323 /* Check whether the required permissions can be granted on @to, ignoring
3324 * all BdrvChild in @list so that they can't block themselves. */
3325 ret
= bdrv_check_update_perm(to
, NULL
, perm
, shared
, list
, errp
);
3327 bdrv_abort_perm_update(to
);
3331 /* Now actually perform the change. We performed the permission check for
3332 * all elements of @list at once, so set the permissions all at once at the
3334 for (p
= list
; p
!= NULL
; p
= p
->next
) {
3338 bdrv_replace_child_noperm(c
, to
);
3342 bdrv_get_cumulative_perm(to
, &old_perm
, &old_shared
);
3343 bdrv_set_perm(to
, old_perm
| perm
, old_shared
| shared
);
3351 * Add new bs contents at the top of an image chain while the chain is
3352 * live, while keeping required fields on the top layer.
3354 * This will modify the BlockDriverState fields, and swap contents
3355 * between bs_new and bs_top. Both bs_new and bs_top are modified.
3357 * bs_new must not be attached to a BlockBackend.
3359 * This function does not create any image files.
3361 * bdrv_append() takes ownership of a bs_new reference and unrefs it because
3362 * that's what the callers commonly need. bs_new will be referenced by the old
3363 * parents of bs_top after bdrv_append() returns. If the caller needs to keep a
3364 * reference of its own, it must call bdrv_ref().
3366 void bdrv_append(BlockDriverState
*bs_new
, BlockDriverState
*bs_top
,
3369 Error
*local_err
= NULL
;
3371 bdrv_set_backing_hd(bs_new
, bs_top
, &local_err
);
3373 error_propagate(errp
, local_err
);
3377 bdrv_replace_node(bs_top
, bs_new
, &local_err
);
3379 error_propagate(errp
, local_err
);
3380 bdrv_set_backing_hd(bs_new
, NULL
, &error_abort
);
3384 /* bs_new is now referenced by its new parents, we don't need the
3385 * additional reference any more. */
3390 static void bdrv_delete(BlockDriverState
*bs
)
3393 assert(bdrv_op_blocker_is_empty(bs
));
3394 assert(!bs
->refcnt
);
3398 /* remove from list, if necessary */
3399 if (bs
->node_name
[0] != '\0') {
3400 QTAILQ_REMOVE(&graph_bdrv_states
, bs
, node_list
);
3402 QTAILQ_REMOVE(&all_bdrv_states
, bs
, bs_list
);
3408 * Run consistency checks on an image
3410 * Returns 0 if the check could be completed (it doesn't mean that the image is
3411 * free of errors) or -errno when an internal error occurred. The results of the
3412 * check are stored in res.
3414 int bdrv_check(BlockDriverState
*bs
, BdrvCheckResult
*res
, BdrvCheckMode fix
)
3416 if (bs
->drv
== NULL
) {
3419 if (bs
->drv
->bdrv_check
== NULL
) {
3423 memset(res
, 0, sizeof(*res
));
3424 return bs
->drv
->bdrv_check(bs
, res
, fix
);
3430 * -EINVAL - backing format specified, but no file
3431 * -ENOSPC - can't update the backing file because no space is left in the
3433 * -ENOTSUP - format driver doesn't support changing the backing file
3435 int bdrv_change_backing_file(BlockDriverState
*bs
,
3436 const char *backing_file
, const char *backing_fmt
)
3438 BlockDriver
*drv
= bs
->drv
;
3445 /* Backing file format doesn't make sense without a backing file */
3446 if (backing_fmt
&& !backing_file
) {
3450 if (drv
->bdrv_change_backing_file
!= NULL
) {
3451 ret
= drv
->bdrv_change_backing_file(bs
, backing_file
, backing_fmt
);
3457 pstrcpy(bs
->backing_file
, sizeof(bs
->backing_file
), backing_file
?: "");
3458 pstrcpy(bs
->backing_format
, sizeof(bs
->backing_format
), backing_fmt
?: "");
3464 * Finds the image layer in the chain that has 'bs' as its backing file.
3466 * active is the current topmost image.
3468 * Returns NULL if bs is not found in active's image chain,
3469 * or if active == bs.
3471 * Returns the bottommost base image if bs == NULL.
3473 BlockDriverState
*bdrv_find_overlay(BlockDriverState
*active
,
3474 BlockDriverState
*bs
)
3476 while (active
&& bs
!= backing_bs(active
)) {
3477 active
= backing_bs(active
);
3483 /* Given a BDS, searches for the base layer. */
3484 BlockDriverState
*bdrv_find_base(BlockDriverState
*bs
)
3486 return bdrv_find_overlay(bs
, NULL
);
3490 * Drops images above 'base' up to and including 'top', and sets the image
3491 * above 'top' to have base as its backing file.
3493 * Requires that the overlay to 'top' is opened r/w, so that the backing file
3494 * information in 'bs' can be properly updated.
3496 * E.g., this will convert the following chain:
3497 * bottom <- base <- intermediate <- top <- active
3501 * bottom <- base <- active
3503 * It is allowed for bottom==base, in which case it converts:
3505 * base <- intermediate <- top <- active
3511 * If backing_file_str is non-NULL, it will be used when modifying top's
3512 * overlay image metadata.
3515 * if active == top, that is considered an error
3518 int bdrv_drop_intermediate(BlockDriverState
*top
, BlockDriverState
*base
,
3519 const char *backing_file_str
)
3521 BdrvChild
*c
, *next
;
3522 Error
*local_err
= NULL
;
3527 if (!top
->drv
|| !base
->drv
) {
3531 /* Make sure that base is in the backing chain of top */
3532 if (!bdrv_chain_contains(top
, base
)) {
3536 /* success - we can delete the intermediate states, and link top->base */
3537 /* TODO Check graph modification op blockers (BLK_PERM_GRAPH_MOD) once
3538 * we've figured out how they should work. */
3539 backing_file_str
= backing_file_str
? backing_file_str
: base
->filename
;
3541 QLIST_FOREACH_SAFE(c
, &top
->parents
, next_parent
, next
) {
3542 /* Check whether we are allowed to switch c from top to base */
3543 GSList
*ignore_children
= g_slist_prepend(NULL
, c
);
3544 bdrv_check_update_perm(base
, NULL
, c
->perm
, c
->shared_perm
,
3545 ignore_children
, &local_err
);
3548 error_report_err(local_err
);
3551 g_slist_free(ignore_children
);
3553 /* If so, update the backing file path in the image file */
3554 if (c
->role
->update_filename
) {
3555 ret
= c
->role
->update_filename(c
, base
, backing_file_str
,
3558 bdrv_abort_perm_update(base
);
3559 error_report_err(local_err
);
3564 /* Do the actual switch in the in-memory graph.
3565 * Completes bdrv_check_update_perm() transaction internally. */
3567 bdrv_replace_child(c
, base
);
3578 * Truncate file to 'offset' bytes (needed only for file protocols)
3580 int bdrv_truncate(BdrvChild
*child
, int64_t offset
, PreallocMode prealloc
,
3583 BlockDriverState
*bs
= child
->bs
;
3584 BlockDriver
*drv
= bs
->drv
;
3587 assert(child
->perm
& BLK_PERM_RESIZE
);
3589 /* if bs->drv == NULL, bs is closed, so there's nothing to do here */
3591 error_setg(errp
, "No medium inserted");
3594 if (!drv
->bdrv_truncate
) {
3595 if (bs
->file
&& drv
->is_filter
) {
3596 return bdrv_truncate(bs
->file
, offset
, prealloc
, errp
);
3598 error_setg(errp
, "Image format driver does not support resize");
3601 if (bs
->read_only
) {
3602 error_setg(errp
, "Image is read-only");
3606 assert(!(bs
->open_flags
& BDRV_O_INACTIVE
));
3608 ret
= drv
->bdrv_truncate(bs
, offset
, prealloc
, errp
);
3612 ret
= refresh_total_sectors(bs
, offset
>> BDRV_SECTOR_BITS
);
3614 error_setg_errno(errp
, -ret
, "Could not refresh total sector count");
3616 offset
= bs
->total_sectors
* BDRV_SECTOR_SIZE
;
3618 bdrv_dirty_bitmap_truncate(bs
, offset
);
3619 bdrv_parent_cb_resize(bs
);
3620 atomic_inc(&bs
->write_gen
);
3625 * Length of a allocated file in bytes. Sparse files are counted by actual
3626 * allocated space. Return < 0 if error or unknown.
3628 int64_t bdrv_get_allocated_file_size(BlockDriverState
*bs
)
3630 BlockDriver
*drv
= bs
->drv
;
3634 if (drv
->bdrv_get_allocated_file_size
) {
3635 return drv
->bdrv_get_allocated_file_size(bs
);
3638 return bdrv_get_allocated_file_size(bs
->file
->bs
);
3645 * @drv: Format driver
3646 * @opts: Creation options for new image
3647 * @in_bs: Existing image containing data for new image (may be NULL)
3648 * @errp: Error object
3649 * Returns: A #BlockMeasureInfo (free using qapi_free_BlockMeasureInfo())
3652 * Calculate file size required to create a new image.
3654 * If @in_bs is given then space for allocated clusters and zero clusters
3655 * from that image are included in the calculation. If @opts contains a
3656 * backing file that is shared by @in_bs then backing clusters may be omitted
3657 * from the calculation.
3659 * If @in_bs is NULL then the calculation includes no allocated clusters
3660 * unless a preallocation option is given in @opts.
3662 * Note that @in_bs may use a different BlockDriver from @drv.
3664 * If an error occurs the @errp pointer is set.
3666 BlockMeasureInfo
*bdrv_measure(BlockDriver
*drv
, QemuOpts
*opts
,
3667 BlockDriverState
*in_bs
, Error
**errp
)
3669 if (!drv
->bdrv_measure
) {
3670 error_setg(errp
, "Block driver '%s' does not support size measurement",
3675 return drv
->bdrv_measure(opts
, in_bs
, errp
);
3679 * Return number of sectors on success, -errno on error.
3681 int64_t bdrv_nb_sectors(BlockDriverState
*bs
)
3683 BlockDriver
*drv
= bs
->drv
;
3688 if (drv
->has_variable_length
) {
3689 int ret
= refresh_total_sectors(bs
, bs
->total_sectors
);
3694 return bs
->total_sectors
;
3698 * Return length in bytes on success, -errno on error.
3699 * The length is always a multiple of BDRV_SECTOR_SIZE.
3701 int64_t bdrv_getlength(BlockDriverState
*bs
)
3703 int64_t ret
= bdrv_nb_sectors(bs
);
3705 ret
= ret
> INT64_MAX
/ BDRV_SECTOR_SIZE
? -EFBIG
: ret
;
3706 return ret
< 0 ? ret
: ret
* BDRV_SECTOR_SIZE
;
3709 /* return 0 as number of sectors if no device present or error */
3710 void bdrv_get_geometry(BlockDriverState
*bs
, uint64_t *nb_sectors_ptr
)
3712 int64_t nb_sectors
= bdrv_nb_sectors(bs
);
3714 *nb_sectors_ptr
= nb_sectors
< 0 ? 0 : nb_sectors
;
3717 bool bdrv_is_sg(BlockDriverState
*bs
)
3722 bool bdrv_is_encrypted(BlockDriverState
*bs
)
3724 if (bs
->backing
&& bs
->backing
->bs
->encrypted
) {
3727 return bs
->encrypted
;
3730 const char *bdrv_get_format_name(BlockDriverState
*bs
)
3732 return bs
->drv
? bs
->drv
->format_name
: NULL
;
3735 static int qsort_strcmp(const void *a
, const void *b
)
3737 return strcmp(*(char *const *)a
, *(char *const *)b
);
3740 void bdrv_iterate_format(void (*it
)(void *opaque
, const char *name
),
3746 const char **formats
= NULL
;
3748 QLIST_FOREACH(drv
, &bdrv_drivers
, list
) {
3749 if (drv
->format_name
) {
3752 while (formats
&& i
&& !found
) {
3753 found
= !strcmp(formats
[--i
], drv
->format_name
);
3757 formats
= g_renew(const char *, formats
, count
+ 1);
3758 formats
[count
++] = drv
->format_name
;
3763 for (i
= 0; i
< (int)ARRAY_SIZE(block_driver_modules
); i
++) {
3764 const char *format_name
= block_driver_modules
[i
].format_name
;
3770 while (formats
&& j
&& !found
) {
3771 found
= !strcmp(formats
[--j
], format_name
);
3775 formats
= g_renew(const char *, formats
, count
+ 1);
3776 formats
[count
++] = format_name
;
3781 qsort(formats
, count
, sizeof(formats
[0]), qsort_strcmp
);
3783 for (i
= 0; i
< count
; i
++) {
3784 it(opaque
, formats
[i
]);
3790 /* This function is to find a node in the bs graph */
3791 BlockDriverState
*bdrv_find_node(const char *node_name
)
3793 BlockDriverState
*bs
;
3797 QTAILQ_FOREACH(bs
, &graph_bdrv_states
, node_list
) {
3798 if (!strcmp(node_name
, bs
->node_name
)) {
3805 /* Put this QMP function here so it can access the static graph_bdrv_states. */
3806 BlockDeviceInfoList
*bdrv_named_nodes_list(Error
**errp
)
3808 BlockDeviceInfoList
*list
, *entry
;
3809 BlockDriverState
*bs
;
3812 QTAILQ_FOREACH(bs
, &graph_bdrv_states
, node_list
) {
3813 BlockDeviceInfo
*info
= bdrv_block_device_info(NULL
, bs
, errp
);
3815 qapi_free_BlockDeviceInfoList(list
);
3818 entry
= g_malloc0(sizeof(*entry
));
3819 entry
->value
= info
;
3827 BlockDriverState
*bdrv_lookup_bs(const char *device
,
3828 const char *node_name
,
3832 BlockDriverState
*bs
;
3835 blk
= blk_by_name(device
);
3840 error_setg(errp
, "Device '%s' has no medium", device
);
3848 bs
= bdrv_find_node(node_name
);
3855 error_setg(errp
, "Cannot find device=%s nor node_name=%s",
3856 device
? device
: "",
3857 node_name
? node_name
: "");
3861 /* If 'base' is in the same chain as 'top', return true. Otherwise,
3862 * return false. If either argument is NULL, return false. */
3863 bool bdrv_chain_contains(BlockDriverState
*top
, BlockDriverState
*base
)
3865 while (top
&& top
!= base
) {
3866 top
= backing_bs(top
);
3872 BlockDriverState
*bdrv_next_node(BlockDriverState
*bs
)
3875 return QTAILQ_FIRST(&graph_bdrv_states
);
3877 return QTAILQ_NEXT(bs
, node_list
);
3880 const char *bdrv_get_node_name(const BlockDriverState
*bs
)
3882 return bs
->node_name
;
3885 const char *bdrv_get_parent_name(const BlockDriverState
*bs
)
3890 /* If multiple parents have a name, just pick the first one. */
3891 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
3892 if (c
->role
->get_name
) {
3893 name
= c
->role
->get_name(c
);
3894 if (name
&& *name
) {
3903 /* TODO check what callers really want: bs->node_name or blk_name() */
3904 const char *bdrv_get_device_name(const BlockDriverState
*bs
)
3906 return bdrv_get_parent_name(bs
) ?: "";
3909 /* This can be used to identify nodes that might not have a device
3910 * name associated. Since node and device names live in the same
3911 * namespace, the result is unambiguous. The exception is if both are
3912 * absent, then this returns an empty (non-null) string. */
3913 const char *bdrv_get_device_or_node_name(const BlockDriverState
*bs
)
3915 return bdrv_get_parent_name(bs
) ?: bs
->node_name
;
3918 int bdrv_get_flags(BlockDriverState
*bs
)
3920 return bs
->open_flags
;
3923 int bdrv_has_zero_init_1(BlockDriverState
*bs
)
3928 int bdrv_has_zero_init(BlockDriverState
*bs
)
3934 /* If BS is a copy on write image, it is initialized to
3935 the contents of the base image, which may not be zeroes. */
3939 if (bs
->drv
->bdrv_has_zero_init
) {
3940 return bs
->drv
->bdrv_has_zero_init(bs
);
3942 if (bs
->file
&& bs
->drv
->is_filter
) {
3943 return bdrv_has_zero_init(bs
->file
->bs
);
3950 bool bdrv_unallocated_blocks_are_zero(BlockDriverState
*bs
)
3952 BlockDriverInfo bdi
;
3958 if (bdrv_get_info(bs
, &bdi
) == 0) {
3959 return bdi
.unallocated_blocks_are_zero
;
3965 bool bdrv_can_write_zeroes_with_unmap(BlockDriverState
*bs
)
3967 BlockDriverInfo bdi
;
3969 if (!(bs
->open_flags
& BDRV_O_UNMAP
)) {
3973 if (bdrv_get_info(bs
, &bdi
) == 0) {
3974 return bdi
.can_write_zeroes_with_unmap
;
3980 const char *bdrv_get_encrypted_filename(BlockDriverState
*bs
)
3982 if (bs
->backing
&& bs
->backing
->bs
->encrypted
)
3983 return bs
->backing_file
;
3984 else if (bs
->encrypted
)
3985 return bs
->filename
;
3990 void bdrv_get_backing_filename(BlockDriverState
*bs
,
3991 char *filename
, int filename_size
)
3993 pstrcpy(filename
, filename_size
, bs
->backing_file
);
3996 int bdrv_get_info(BlockDriverState
*bs
, BlockDriverInfo
*bdi
)
3998 BlockDriver
*drv
= bs
->drv
;
3999 /* if bs->drv == NULL, bs is closed, so there's nothing to do here */
4003 if (!drv
->bdrv_get_info
) {
4004 if (bs
->file
&& drv
->is_filter
) {
4005 return bdrv_get_info(bs
->file
->bs
, bdi
);
4009 memset(bdi
, 0, sizeof(*bdi
));
4010 return drv
->bdrv_get_info(bs
, bdi
);
4013 ImageInfoSpecific
*bdrv_get_specific_info(BlockDriverState
*bs
)
4015 BlockDriver
*drv
= bs
->drv
;
4016 if (drv
&& drv
->bdrv_get_specific_info
) {
4017 return drv
->bdrv_get_specific_info(bs
);
4022 void bdrv_debug_event(BlockDriverState
*bs
, BlkdebugEvent event
)
4024 if (!bs
|| !bs
->drv
|| !bs
->drv
->bdrv_debug_event
) {
4028 bs
->drv
->bdrv_debug_event(bs
, event
);
4031 int bdrv_debug_breakpoint(BlockDriverState
*bs
, const char *event
,
4034 while (bs
&& bs
->drv
&& !bs
->drv
->bdrv_debug_breakpoint
) {
4035 bs
= bs
->file
? bs
->file
->bs
: NULL
;
4038 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_breakpoint
) {
4039 return bs
->drv
->bdrv_debug_breakpoint(bs
, event
, tag
);
4045 int bdrv_debug_remove_breakpoint(BlockDriverState
*bs
, const char *tag
)
4047 while (bs
&& bs
->drv
&& !bs
->drv
->bdrv_debug_remove_breakpoint
) {
4048 bs
= bs
->file
? bs
->file
->bs
: NULL
;
4051 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_remove_breakpoint
) {
4052 return bs
->drv
->bdrv_debug_remove_breakpoint(bs
, tag
);
4058 int bdrv_debug_resume(BlockDriverState
*bs
, const char *tag
)
4060 while (bs
&& (!bs
->drv
|| !bs
->drv
->bdrv_debug_resume
)) {
4061 bs
= bs
->file
? bs
->file
->bs
: NULL
;
4064 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_resume
) {
4065 return bs
->drv
->bdrv_debug_resume(bs
, tag
);
4071 bool bdrv_debug_is_suspended(BlockDriverState
*bs
, const char *tag
)
4073 while (bs
&& bs
->drv
&& !bs
->drv
->bdrv_debug_is_suspended
) {
4074 bs
= bs
->file
? bs
->file
->bs
: NULL
;
4077 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_is_suspended
) {
4078 return bs
->drv
->bdrv_debug_is_suspended(bs
, tag
);
4084 /* backing_file can either be relative, or absolute, or a protocol. If it is
4085 * relative, it must be relative to the chain. So, passing in bs->filename
4086 * from a BDS as backing_file should not be done, as that may be relative to
4087 * the CWD rather than the chain. */
4088 BlockDriverState
*bdrv_find_backing_image(BlockDriverState
*bs
,
4089 const char *backing_file
)
4091 char *filename_full
= NULL
;
4092 char *backing_file_full
= NULL
;
4093 char *filename_tmp
= NULL
;
4094 int is_protocol
= 0;
4095 BlockDriverState
*curr_bs
= NULL
;
4096 BlockDriverState
*retval
= NULL
;
4097 Error
*local_error
= NULL
;
4099 if (!bs
|| !bs
->drv
|| !backing_file
) {
4103 filename_full
= g_malloc(PATH_MAX
);
4104 backing_file_full
= g_malloc(PATH_MAX
);
4105 filename_tmp
= g_malloc(PATH_MAX
);
4107 is_protocol
= path_has_protocol(backing_file
);
4109 for (curr_bs
= bs
; curr_bs
->backing
; curr_bs
= curr_bs
->backing
->bs
) {
4111 /* If either of the filename paths is actually a protocol, then
4112 * compare unmodified paths; otherwise make paths relative */
4113 if (is_protocol
|| path_has_protocol(curr_bs
->backing_file
)) {
4114 if (strcmp(backing_file
, curr_bs
->backing_file
) == 0) {
4115 retval
= curr_bs
->backing
->bs
;
4118 /* Also check against the full backing filename for the image */
4119 bdrv_get_full_backing_filename(curr_bs
, backing_file_full
, PATH_MAX
,
4121 if (local_error
== NULL
) {
4122 if (strcmp(backing_file
, backing_file_full
) == 0) {
4123 retval
= curr_bs
->backing
->bs
;
4127 error_free(local_error
);
4131 /* If not an absolute filename path, make it relative to the current
4132 * image's filename path */
4133 path_combine(filename_tmp
, PATH_MAX
, curr_bs
->filename
,
4136 /* We are going to compare absolute pathnames */
4137 if (!realpath(filename_tmp
, filename_full
)) {
4141 /* We need to make sure the backing filename we are comparing against
4142 * is relative to the current image filename (or absolute) */
4143 path_combine(filename_tmp
, PATH_MAX
, curr_bs
->filename
,
4144 curr_bs
->backing_file
);
4146 if (!realpath(filename_tmp
, backing_file_full
)) {
4150 if (strcmp(backing_file_full
, filename_full
) == 0) {
4151 retval
= curr_bs
->backing
->bs
;
4157 g_free(filename_full
);
4158 g_free(backing_file_full
);
4159 g_free(filename_tmp
);
4163 void bdrv_init(void)
4165 module_call_init(MODULE_INIT_BLOCK
);
4168 void bdrv_init_with_whitelist(void)
4170 use_bdrv_whitelist
= 1;
4174 void bdrv_invalidate_cache(BlockDriverState
*bs
, Error
**errp
)
4176 BdrvChild
*child
, *parent
;
4177 uint64_t perm
, shared_perm
;
4178 Error
*local_err
= NULL
;
4185 if (!(bs
->open_flags
& BDRV_O_INACTIVE
)) {
4189 QLIST_FOREACH(child
, &bs
->children
, next
) {
4190 bdrv_invalidate_cache(child
->bs
, &local_err
);
4192 error_propagate(errp
, local_err
);
4198 * Update permissions, they may differ for inactive nodes.
4200 * Note that the required permissions of inactive images are always a
4201 * subset of the permissions required after activating the image. This
4202 * allows us to just get the permissions upfront without restricting
4203 * drv->bdrv_invalidate_cache().
4205 * It also means that in error cases, we don't have to try and revert to
4206 * the old permissions (which is an operation that could fail, too). We can
4207 * just keep the extended permissions for the next time that an activation
4208 * of the image is tried.
4210 bs
->open_flags
&= ~BDRV_O_INACTIVE
;
4211 bdrv_get_cumulative_perm(bs
, &perm
, &shared_perm
);
4212 ret
= bdrv_check_perm(bs
, NULL
, perm
, shared_perm
, NULL
, &local_err
);
4214 bs
->open_flags
|= BDRV_O_INACTIVE
;
4215 error_propagate(errp
, local_err
);
4218 bdrv_set_perm(bs
, perm
, shared_perm
);
4220 if (bs
->drv
->bdrv_invalidate_cache
) {
4221 bs
->drv
->bdrv_invalidate_cache(bs
, &local_err
);
4223 bs
->open_flags
|= BDRV_O_INACTIVE
;
4224 error_propagate(errp
, local_err
);
4229 ret
= refresh_total_sectors(bs
, bs
->total_sectors
);
4231 bs
->open_flags
|= BDRV_O_INACTIVE
;
4232 error_setg_errno(errp
, -ret
, "Could not refresh total sector count");
4236 QLIST_FOREACH(parent
, &bs
->parents
, next_parent
) {
4237 if (parent
->role
->activate
) {
4238 parent
->role
->activate(parent
, &local_err
);
4240 error_propagate(errp
, local_err
);
4247 void bdrv_invalidate_cache_all(Error
**errp
)
4249 BlockDriverState
*bs
;
4250 Error
*local_err
= NULL
;
4251 BdrvNextIterator it
;
4253 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
4254 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
4256 aio_context_acquire(aio_context
);
4257 bdrv_invalidate_cache(bs
, &local_err
);
4258 aio_context_release(aio_context
);
4260 error_propagate(errp
, local_err
);
4261 bdrv_next_cleanup(&it
);
4267 static int bdrv_inactivate_recurse(BlockDriverState
*bs
,
4270 BdrvChild
*child
, *parent
;
4277 if (!setting_flag
&& bs
->drv
->bdrv_inactivate
) {
4278 ret
= bs
->drv
->bdrv_inactivate(bs
);
4284 if (setting_flag
&& !(bs
->open_flags
& BDRV_O_INACTIVE
)) {
4285 uint64_t perm
, shared_perm
;
4287 QLIST_FOREACH(parent
, &bs
->parents
, next_parent
) {
4288 if (parent
->role
->inactivate
) {
4289 ret
= parent
->role
->inactivate(parent
);
4296 bs
->open_flags
|= BDRV_O_INACTIVE
;
4298 /* Update permissions, they may differ for inactive nodes */
4299 bdrv_get_cumulative_perm(bs
, &perm
, &shared_perm
);
4300 bdrv_check_perm(bs
, NULL
, perm
, shared_perm
, NULL
, &error_abort
);
4301 bdrv_set_perm(bs
, perm
, shared_perm
);
4304 QLIST_FOREACH(child
, &bs
->children
, next
) {
4305 ret
= bdrv_inactivate_recurse(child
->bs
, setting_flag
);
4311 /* At this point persistent bitmaps should be already stored by the format
4313 bdrv_release_persistent_dirty_bitmaps(bs
);
4318 int bdrv_inactivate_all(void)
4320 BlockDriverState
*bs
= NULL
;
4321 BdrvNextIterator it
;
4325 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
4326 aio_context_acquire(bdrv_get_aio_context(bs
));
4329 /* We do two passes of inactivation. The first pass calls to drivers'
4330 * .bdrv_inactivate callbacks recursively so all cache is flushed to disk;
4331 * the second pass sets the BDRV_O_INACTIVE flag so that no further write
4333 for (pass
= 0; pass
< 2; pass
++) {
4334 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
4335 ret
= bdrv_inactivate_recurse(bs
, pass
);
4337 bdrv_next_cleanup(&it
);
4344 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
4345 aio_context_release(bdrv_get_aio_context(bs
));
4351 /**************************************************************/
4352 /* removable device support */
4355 * Return TRUE if the media is present
4357 bool bdrv_is_inserted(BlockDriverState
*bs
)
4359 BlockDriver
*drv
= bs
->drv
;
4365 if (drv
->bdrv_is_inserted
) {
4366 return drv
->bdrv_is_inserted(bs
);
4368 QLIST_FOREACH(child
, &bs
->children
, next
) {
4369 if (!bdrv_is_inserted(child
->bs
)) {
4377 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
4379 void bdrv_eject(BlockDriverState
*bs
, bool eject_flag
)
4381 BlockDriver
*drv
= bs
->drv
;
4383 if (drv
&& drv
->bdrv_eject
) {
4384 drv
->bdrv_eject(bs
, eject_flag
);
4389 * Lock or unlock the media (if it is locked, the user won't be able
4390 * to eject it manually).
4392 void bdrv_lock_medium(BlockDriverState
*bs
, bool locked
)
4394 BlockDriver
*drv
= bs
->drv
;
4396 trace_bdrv_lock_medium(bs
, locked
);
4398 if (drv
&& drv
->bdrv_lock_medium
) {
4399 drv
->bdrv_lock_medium(bs
, locked
);
4403 /* Get a reference to bs */
4404 void bdrv_ref(BlockDriverState
*bs
)
4409 /* Release a previously grabbed reference to bs.
4410 * If after releasing, reference count is zero, the BlockDriverState is
4412 void bdrv_unref(BlockDriverState
*bs
)
4417 assert(bs
->refcnt
> 0);
4418 if (--bs
->refcnt
== 0) {
4423 struct BdrvOpBlocker
{
4425 QLIST_ENTRY(BdrvOpBlocker
) list
;
4428 bool bdrv_op_is_blocked(BlockDriverState
*bs
, BlockOpType op
, Error
**errp
)
4430 BdrvOpBlocker
*blocker
;
4431 assert((int) op
>= 0 && op
< BLOCK_OP_TYPE_MAX
);
4432 if (!QLIST_EMPTY(&bs
->op_blockers
[op
])) {
4433 blocker
= QLIST_FIRST(&bs
->op_blockers
[op
]);
4434 error_propagate(errp
, error_copy(blocker
->reason
));
4435 error_prepend(errp
, "Node '%s' is busy: ",
4436 bdrv_get_device_or_node_name(bs
));
4442 void bdrv_op_block(BlockDriverState
*bs
, BlockOpType op
, Error
*reason
)
4444 BdrvOpBlocker
*blocker
;
4445 assert((int) op
>= 0 && op
< BLOCK_OP_TYPE_MAX
);
4447 blocker
= g_new0(BdrvOpBlocker
, 1);
4448 blocker
->reason
= reason
;
4449 QLIST_INSERT_HEAD(&bs
->op_blockers
[op
], blocker
, list
);
4452 void bdrv_op_unblock(BlockDriverState
*bs
, BlockOpType op
, Error
*reason
)
4454 BdrvOpBlocker
*blocker
, *next
;
4455 assert((int) op
>= 0 && op
< BLOCK_OP_TYPE_MAX
);
4456 QLIST_FOREACH_SAFE(blocker
, &bs
->op_blockers
[op
], list
, next
) {
4457 if (blocker
->reason
== reason
) {
4458 QLIST_REMOVE(blocker
, list
);
4464 void bdrv_op_block_all(BlockDriverState
*bs
, Error
*reason
)
4467 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
4468 bdrv_op_block(bs
, i
, reason
);
4472 void bdrv_op_unblock_all(BlockDriverState
*bs
, Error
*reason
)
4475 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
4476 bdrv_op_unblock(bs
, i
, reason
);
4480 bool bdrv_op_blocker_is_empty(BlockDriverState
*bs
)
4484 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
4485 if (!QLIST_EMPTY(&bs
->op_blockers
[i
])) {
4492 void bdrv_img_create(const char *filename
, const char *fmt
,
4493 const char *base_filename
, const char *base_fmt
,
4494 char *options
, uint64_t img_size
, int flags
, bool quiet
,
4497 QemuOptsList
*create_opts
= NULL
;
4498 QemuOpts
*opts
= NULL
;
4499 const char *backing_fmt
, *backing_file
;
4501 BlockDriver
*drv
, *proto_drv
;
4502 Error
*local_err
= NULL
;
4505 /* Find driver and parse its options */
4506 drv
= bdrv_find_format(fmt
);
4508 error_setg(errp
, "Unknown file format '%s'", fmt
);
4512 proto_drv
= bdrv_find_protocol(filename
, true, errp
);
4517 if (!drv
->create_opts
) {
4518 error_setg(errp
, "Format driver '%s' does not support image creation",
4523 if (!proto_drv
->create_opts
) {
4524 error_setg(errp
, "Protocol driver '%s' does not support image creation",
4525 proto_drv
->format_name
);
4529 create_opts
= qemu_opts_append(create_opts
, drv
->create_opts
);
4530 create_opts
= qemu_opts_append(create_opts
, proto_drv
->create_opts
);
4532 /* Create parameter list with default values */
4533 opts
= qemu_opts_create(create_opts
, NULL
, 0, &error_abort
);
4534 qemu_opt_set_number(opts
, BLOCK_OPT_SIZE
, img_size
, &error_abort
);
4536 /* Parse -o options */
4538 qemu_opts_do_parse(opts
, options
, NULL
, &local_err
);
4540 error_report_err(local_err
);
4542 error_setg(errp
, "Invalid options for file format '%s'", fmt
);
4547 if (base_filename
) {
4548 qemu_opt_set(opts
, BLOCK_OPT_BACKING_FILE
, base_filename
, &local_err
);
4550 error_setg(errp
, "Backing file not supported for file format '%s'",
4557 qemu_opt_set(opts
, BLOCK_OPT_BACKING_FMT
, base_fmt
, &local_err
);
4559 error_setg(errp
, "Backing file format not supported for file "
4560 "format '%s'", fmt
);
4565 backing_file
= qemu_opt_get(opts
, BLOCK_OPT_BACKING_FILE
);
4567 if (!strcmp(filename
, backing_file
)) {
4568 error_setg(errp
, "Error: Trying to create an image with the "
4569 "same filename as the backing file");
4574 backing_fmt
= qemu_opt_get(opts
, BLOCK_OPT_BACKING_FMT
);
4576 /* The size for the image must always be specified, unless we have a backing
4577 * file and we have not been forbidden from opening it. */
4578 size
= qemu_opt_get_size(opts
, BLOCK_OPT_SIZE
, img_size
);
4579 if (backing_file
&& !(flags
& BDRV_O_NO_BACKING
)) {
4580 BlockDriverState
*bs
;
4581 char *full_backing
= g_new0(char, PATH_MAX
);
4583 QDict
*backing_options
= NULL
;
4585 bdrv_get_full_backing_filename_from_filename(filename
, backing_file
,
4586 full_backing
, PATH_MAX
,
4589 g_free(full_backing
);
4593 /* backing files always opened read-only */
4595 back_flags
&= ~(BDRV_O_RDWR
| BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
);
4598 backing_options
= qdict_new();
4599 qdict_put_str(backing_options
, "driver", backing_fmt
);
4602 bs
= bdrv_open(full_backing
, NULL
, backing_options
, back_flags
,
4604 g_free(full_backing
);
4605 if (!bs
&& size
!= -1) {
4606 /* Couldn't open BS, but we have a size, so it's nonfatal */
4607 warn_reportf_err(local_err
,
4608 "Could not verify backing image. "
4609 "This may become an error in future versions.\n");
4612 /* Couldn't open bs, do not have size */
4613 error_append_hint(&local_err
,
4614 "Could not open backing image to determine size.\n");
4618 /* Opened BS, have no size */
4619 size
= bdrv_getlength(bs
);
4621 error_setg_errno(errp
, -size
, "Could not get size of '%s'",
4626 qemu_opt_set_number(opts
, BLOCK_OPT_SIZE
, size
, &error_abort
);
4630 } /* (backing_file && !(flags & BDRV_O_NO_BACKING)) */
4633 error_setg(errp
, "Image creation needs a size parameter");
4638 printf("Formatting '%s', fmt=%s ", filename
, fmt
);
4639 qemu_opts_print(opts
, " ");
4643 ret
= bdrv_create(drv
, filename
, opts
, &local_err
);
4645 if (ret
== -EFBIG
) {
4646 /* This is generally a better message than whatever the driver would
4647 * deliver (especially because of the cluster_size_hint), since that
4648 * is most probably not much different from "image too large". */
4649 const char *cluster_size_hint
= "";
4650 if (qemu_opt_get_size(opts
, BLOCK_OPT_CLUSTER_SIZE
, 0)) {
4651 cluster_size_hint
= " (try using a larger cluster size)";
4653 error_setg(errp
, "The image size is too large for file format '%s'"
4654 "%s", fmt
, cluster_size_hint
);
4655 error_free(local_err
);
4660 qemu_opts_del(opts
);
4661 qemu_opts_free(create_opts
);
4662 error_propagate(errp
, local_err
);
4665 AioContext
*bdrv_get_aio_context(BlockDriverState
*bs
)
4667 return bs
->aio_context
;
4670 void bdrv_coroutine_enter(BlockDriverState
*bs
, Coroutine
*co
)
4672 aio_co_enter(bdrv_get_aio_context(bs
), co
);
4675 static void bdrv_do_remove_aio_context_notifier(BdrvAioNotifier
*ban
)
4677 QLIST_REMOVE(ban
, list
);
4681 void bdrv_detach_aio_context(BlockDriverState
*bs
)
4683 BdrvAioNotifier
*baf
, *baf_tmp
;
4690 assert(!bs
->walking_aio_notifiers
);
4691 bs
->walking_aio_notifiers
= true;
4692 QLIST_FOREACH_SAFE(baf
, &bs
->aio_notifiers
, list
, baf_tmp
) {
4694 bdrv_do_remove_aio_context_notifier(baf
);
4696 baf
->detach_aio_context(baf
->opaque
);
4699 /* Never mind iterating again to check for ->deleted. bdrv_close() will
4700 * remove remaining aio notifiers if we aren't called again.
4702 bs
->walking_aio_notifiers
= false;
4704 if (bs
->drv
->bdrv_detach_aio_context
) {
4705 bs
->drv
->bdrv_detach_aio_context(bs
);
4707 QLIST_FOREACH(child
, &bs
->children
, next
) {
4708 bdrv_detach_aio_context(child
->bs
);
4711 bs
->aio_context
= NULL
;
4714 void bdrv_attach_aio_context(BlockDriverState
*bs
,
4715 AioContext
*new_context
)
4717 BdrvAioNotifier
*ban
, *ban_tmp
;
4724 bs
->aio_context
= new_context
;
4726 QLIST_FOREACH(child
, &bs
->children
, next
) {
4727 bdrv_attach_aio_context(child
->bs
, new_context
);
4729 if (bs
->drv
->bdrv_attach_aio_context
) {
4730 bs
->drv
->bdrv_attach_aio_context(bs
, new_context
);
4733 assert(!bs
->walking_aio_notifiers
);
4734 bs
->walking_aio_notifiers
= true;
4735 QLIST_FOREACH_SAFE(ban
, &bs
->aio_notifiers
, list
, ban_tmp
) {
4737 bdrv_do_remove_aio_context_notifier(ban
);
4739 ban
->attached_aio_context(new_context
, ban
->opaque
);
4742 bs
->walking_aio_notifiers
= false;
4745 void bdrv_set_aio_context(BlockDriverState
*bs
, AioContext
*new_context
)
4747 AioContext
*ctx
= bdrv_get_aio_context(bs
);
4749 aio_disable_external(ctx
);
4750 bdrv_parent_drained_begin(bs
);
4751 bdrv_drain(bs
); /* ensure there are no in-flight requests */
4753 while (aio_poll(ctx
, false)) {
4754 /* wait for all bottom halves to execute */
4757 bdrv_detach_aio_context(bs
);
4759 /* This function executes in the old AioContext so acquire the new one in
4760 * case it runs in a different thread.
4762 aio_context_acquire(new_context
);
4763 bdrv_attach_aio_context(bs
, new_context
);
4764 bdrv_parent_drained_end(bs
);
4765 aio_enable_external(ctx
);
4766 aio_context_release(new_context
);
4769 void bdrv_add_aio_context_notifier(BlockDriverState
*bs
,
4770 void (*attached_aio_context
)(AioContext
*new_context
, void *opaque
),
4771 void (*detach_aio_context
)(void *opaque
), void *opaque
)
4773 BdrvAioNotifier
*ban
= g_new(BdrvAioNotifier
, 1);
4774 *ban
= (BdrvAioNotifier
){
4775 .attached_aio_context
= attached_aio_context
,
4776 .detach_aio_context
= detach_aio_context
,
4780 QLIST_INSERT_HEAD(&bs
->aio_notifiers
, ban
, list
);
4783 void bdrv_remove_aio_context_notifier(BlockDriverState
*bs
,
4784 void (*attached_aio_context
)(AioContext
*,
4786 void (*detach_aio_context
)(void *),
4789 BdrvAioNotifier
*ban
, *ban_next
;
4791 QLIST_FOREACH_SAFE(ban
, &bs
->aio_notifiers
, list
, ban_next
) {
4792 if (ban
->attached_aio_context
== attached_aio_context
&&
4793 ban
->detach_aio_context
== detach_aio_context
&&
4794 ban
->opaque
== opaque
&&
4795 ban
->deleted
== false)
4797 if (bs
->walking_aio_notifiers
) {
4798 ban
->deleted
= true;
4800 bdrv_do_remove_aio_context_notifier(ban
);
4809 int bdrv_amend_options(BlockDriverState
*bs
, QemuOpts
*opts
,
4810 BlockDriverAmendStatusCB
*status_cb
, void *cb_opaque
)
4815 if (!bs
->drv
->bdrv_amend_options
) {
4818 return bs
->drv
->bdrv_amend_options(bs
, opts
, status_cb
, cb_opaque
);
4821 /* This function will be called by the bdrv_recurse_is_first_non_filter method
4822 * of block filter and by bdrv_is_first_non_filter.
4823 * It is used to test if the given bs is the candidate or recurse more in the
4826 bool bdrv_recurse_is_first_non_filter(BlockDriverState
*bs
,
4827 BlockDriverState
*candidate
)
4829 /* return false if basic checks fails */
4830 if (!bs
|| !bs
->drv
) {
4834 /* the code reached a non block filter driver -> check if the bs is
4835 * the same as the candidate. It's the recursion termination condition.
4837 if (!bs
->drv
->is_filter
) {
4838 return bs
== candidate
;
4840 /* Down this path the driver is a block filter driver */
4842 /* If the block filter recursion method is defined use it to recurse down
4845 if (bs
->drv
->bdrv_recurse_is_first_non_filter
) {
4846 return bs
->drv
->bdrv_recurse_is_first_non_filter(bs
, candidate
);
4849 /* the driver is a block filter but don't allow to recurse -> return false
4854 /* This function checks if the candidate is the first non filter bs down it's
4855 * bs chain. Since we don't have pointers to parents it explore all bs chains
4856 * from the top. Some filters can choose not to pass down the recursion.
4858 bool bdrv_is_first_non_filter(BlockDriverState
*candidate
)
4860 BlockDriverState
*bs
;
4861 BdrvNextIterator it
;
4863 /* walk down the bs forest recursively */
4864 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
4867 /* try to recurse in this top level bs */
4868 perm
= bdrv_recurse_is_first_non_filter(bs
, candidate
);
4870 /* candidate is the first non filter */
4872 bdrv_next_cleanup(&it
);
4880 BlockDriverState
*check_to_replace_node(BlockDriverState
*parent_bs
,
4881 const char *node_name
, Error
**errp
)
4883 BlockDriverState
*to_replace_bs
= bdrv_find_node(node_name
);
4884 AioContext
*aio_context
;
4886 if (!to_replace_bs
) {
4887 error_setg(errp
, "Node name '%s' not found", node_name
);
4891 aio_context
= bdrv_get_aio_context(to_replace_bs
);
4892 aio_context_acquire(aio_context
);
4894 if (bdrv_op_is_blocked(to_replace_bs
, BLOCK_OP_TYPE_REPLACE
, errp
)) {
4895 to_replace_bs
= NULL
;
4899 /* We don't want arbitrary node of the BDS chain to be replaced only the top
4900 * most non filter in order to prevent data corruption.
4901 * Another benefit is that this tests exclude backing files which are
4902 * blocked by the backing blockers.
4904 if (!bdrv_recurse_is_first_non_filter(parent_bs
, to_replace_bs
)) {
4905 error_setg(errp
, "Only top most non filter can be replaced");
4906 to_replace_bs
= NULL
;
4911 aio_context_release(aio_context
);
4912 return to_replace_bs
;
4915 static bool append_open_options(QDict
*d
, BlockDriverState
*bs
)
4917 const QDictEntry
*entry
;
4920 bool found_any
= false;
4923 for (entry
= qdict_first(bs
->options
); entry
;
4924 entry
= qdict_next(bs
->options
, entry
))
4926 /* Exclude options for children */
4927 QLIST_FOREACH(child
, &bs
->children
, next
) {
4928 if (strstart(qdict_entry_key(entry
), child
->name
, &p
)
4929 && (!*p
|| *p
== '.'))
4938 /* And exclude all non-driver-specific options */
4939 for (desc
= bdrv_runtime_opts
.desc
; desc
->name
; desc
++) {
4940 if (!strcmp(qdict_entry_key(entry
), desc
->name
)) {
4948 qobject_incref(qdict_entry_value(entry
));
4949 qdict_put_obj(d
, qdict_entry_key(entry
), qdict_entry_value(entry
));
4956 /* Updates the following BDS fields:
4957 * - exact_filename: A filename which may be used for opening a block device
4958 * which (mostly) equals the given BDS (even without any
4959 * other options; so reading and writing must return the same
4960 * results, but caching etc. may be different)
4961 * - full_open_options: Options which, when given when opening a block device
4962 * (without a filename), result in a BDS (mostly)
4963 * equalling the given one
4964 * - filename: If exact_filename is set, it is copied here. Otherwise,
4965 * full_open_options is converted to a JSON object, prefixed with
4966 * "json:" (for use through the JSON pseudo protocol) and put here.
4968 void bdrv_refresh_filename(BlockDriverState
*bs
)
4970 BlockDriver
*drv
= bs
->drv
;
4977 /* This BDS's file name will most probably depend on its file's name, so
4978 * refresh that first */
4980 bdrv_refresh_filename(bs
->file
->bs
);
4983 if (drv
->bdrv_refresh_filename
) {
4984 /* Obsolete information is of no use here, so drop the old file name
4985 * information before refreshing it */
4986 bs
->exact_filename
[0] = '\0';
4987 if (bs
->full_open_options
) {
4988 QDECREF(bs
->full_open_options
);
4989 bs
->full_open_options
= NULL
;
4993 append_open_options(opts
, bs
);
4994 drv
->bdrv_refresh_filename(bs
, opts
);
4996 } else if (bs
->file
) {
4997 /* Try to reconstruct valid information from the underlying file */
4998 bool has_open_options
;
5000 bs
->exact_filename
[0] = '\0';
5001 if (bs
->full_open_options
) {
5002 QDECREF(bs
->full_open_options
);
5003 bs
->full_open_options
= NULL
;
5007 has_open_options
= append_open_options(opts
, bs
);
5009 /* If no specific options have been given for this BDS, the filename of
5010 * the underlying file should suffice for this one as well */
5011 if (bs
->file
->bs
->exact_filename
[0] && !has_open_options
) {
5012 strcpy(bs
->exact_filename
, bs
->file
->bs
->exact_filename
);
5014 /* Reconstructing the full options QDict is simple for most format block
5015 * drivers, as long as the full options are known for the underlying
5016 * file BDS. The full options QDict of that file BDS should somehow
5017 * contain a representation of the filename, therefore the following
5018 * suffices without querying the (exact_)filename of this BDS. */
5019 if (bs
->file
->bs
->full_open_options
) {
5020 qdict_put_str(opts
, "driver", drv
->format_name
);
5021 QINCREF(bs
->file
->bs
->full_open_options
);
5022 qdict_put(opts
, "file", bs
->file
->bs
->full_open_options
);
5024 bs
->full_open_options
= opts
;
5028 } else if (!bs
->full_open_options
&& qdict_size(bs
->options
)) {
5029 /* There is no underlying file BDS (at least referenced by BDS.file),
5030 * so the full options QDict should be equal to the options given
5031 * specifically for this block device when it was opened (plus the
5032 * driver specification).
5033 * Because those options don't change, there is no need to update
5034 * full_open_options when it's already set. */
5037 append_open_options(opts
, bs
);
5038 qdict_put_str(opts
, "driver", drv
->format_name
);
5040 if (bs
->exact_filename
[0]) {
5041 /* This may not work for all block protocol drivers (some may
5042 * require this filename to be parsed), but we have to find some
5043 * default solution here, so just include it. If some block driver
5044 * does not support pure options without any filename at all or
5045 * needs some special format of the options QDict, it needs to
5046 * implement the driver-specific bdrv_refresh_filename() function.
5048 qdict_put_str(opts
, "filename", bs
->exact_filename
);
5051 bs
->full_open_options
= opts
;
5054 if (bs
->exact_filename
[0]) {
5055 pstrcpy(bs
->filename
, sizeof(bs
->filename
), bs
->exact_filename
);
5056 } else if (bs
->full_open_options
) {
5057 QString
*json
= qobject_to_json(QOBJECT(bs
->full_open_options
));
5058 snprintf(bs
->filename
, sizeof(bs
->filename
), "json:%s",
5059 qstring_get_str(json
));
5065 * Hot add/remove a BDS's child. So the user can take a child offline when
5066 * it is broken and take a new child online
5068 void bdrv_add_child(BlockDriverState
*parent_bs
, BlockDriverState
*child_bs
,
5072 if (!parent_bs
->drv
|| !parent_bs
->drv
->bdrv_add_child
) {
5073 error_setg(errp
, "The node %s does not support adding a child",
5074 bdrv_get_device_or_node_name(parent_bs
));
5078 if (!QLIST_EMPTY(&child_bs
->parents
)) {
5079 error_setg(errp
, "The node %s already has a parent",
5080 child_bs
->node_name
);
5084 parent_bs
->drv
->bdrv_add_child(parent_bs
, child_bs
, errp
);
5087 void bdrv_del_child(BlockDriverState
*parent_bs
, BdrvChild
*child
, Error
**errp
)
5091 if (!parent_bs
->drv
|| !parent_bs
->drv
->bdrv_del_child
) {
5092 error_setg(errp
, "The node %s does not support removing a child",
5093 bdrv_get_device_or_node_name(parent_bs
));
5097 QLIST_FOREACH(tmp
, &parent_bs
->children
, next
) {
5104 error_setg(errp
, "The node %s does not have a child named %s",
5105 bdrv_get_device_or_node_name(parent_bs
),
5106 bdrv_get_device_or_node_name(child
->bs
));
5110 parent_bs
->drv
->bdrv_del_child(parent_bs
, child
, errp
);
5113 bool bdrv_can_store_new_dirty_bitmap(BlockDriverState
*bs
, const char *name
,
5114 uint32_t granularity
, Error
**errp
)
5116 BlockDriver
*drv
= bs
->drv
;
5119 error_setg_errno(errp
, ENOMEDIUM
,
5120 "Can't store persistent bitmaps to %s",
5121 bdrv_get_device_or_node_name(bs
));
5125 if (!drv
->bdrv_can_store_new_dirty_bitmap
) {
5126 error_setg_errno(errp
, ENOTSUP
,
5127 "Can't store persistent bitmaps to %s",
5128 bdrv_get_device_or_node_name(bs
));
5132 return drv
->bdrv_can_store_new_dirty_bitmap(bs
, name
, granularity
, errp
);