2 * QEMU System Emulator block driver
4 * Copyright (c) 2003 Fabrice Bellard
6 * Permission is hereby granted, free of charge, to any person obtaining a copy
7 * of this software and associated documentation files (the "Software"), to deal
8 * in the Software without restriction, including without limitation the rights
9 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10 * copies of the Software, and to permit persons to whom the Software is
11 * furnished to do so, subject to the following conditions:
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
25 #include "qemu/osdep.h"
26 #include "block/trace.h"
27 #include "block/block_int.h"
28 #include "block/blockjob.h"
29 #include "block/nbd.h"
30 #include "qemu/error-report.h"
31 #include "module_block.h"
32 #include "qemu/module.h"
33 #include "qapi/error.h"
34 #include "qapi/qmp/qdict.h"
35 #include "qapi/qmp/qjson.h"
36 #include "qapi/qmp/qstring.h"
37 #include "qapi/qobject-output-visitor.h"
38 #include "qapi/qapi-visit-block-core.h"
39 #include "sysemu/block-backend.h"
40 #include "sysemu/sysemu.h"
41 #include "qemu/notify.h"
42 #include "qemu/option.h"
43 #include "qemu/coroutine.h"
44 #include "block/qapi.h"
45 #include "qemu/timer.h"
46 #include "qemu/cutils.h"
50 #include <sys/ioctl.h>
51 #include <sys/queue.h>
61 #define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */
63 static QTAILQ_HEAD(, BlockDriverState
) graph_bdrv_states
=
64 QTAILQ_HEAD_INITIALIZER(graph_bdrv_states
);
66 static QTAILQ_HEAD(, BlockDriverState
) all_bdrv_states
=
67 QTAILQ_HEAD_INITIALIZER(all_bdrv_states
);
69 static QLIST_HEAD(, BlockDriver
) bdrv_drivers
=
70 QLIST_HEAD_INITIALIZER(bdrv_drivers
);
72 static BlockDriverState
*bdrv_open_inherit(const char *filename
,
73 const char *reference
,
74 QDict
*options
, int flags
,
75 BlockDriverState
*parent
,
76 const BdrvChildRole
*child_role
,
79 /* If non-zero, use only whitelisted block drivers */
80 static int use_bdrv_whitelist
;
83 static int is_windows_drive_prefix(const char *filename
)
85 return (((filename
[0] >= 'a' && filename
[0] <= 'z') ||
86 (filename
[0] >= 'A' && filename
[0] <= 'Z')) &&
90 int is_windows_drive(const char *filename
)
92 if (is_windows_drive_prefix(filename
) &&
95 if (strstart(filename
, "\\\\.\\", NULL
) ||
96 strstart(filename
, "//./", NULL
))
102 size_t bdrv_opt_mem_align(BlockDriverState
*bs
)
104 if (!bs
|| !bs
->drv
) {
105 /* page size or 4k (hdd sector size) should be on the safe side */
106 return MAX(4096, getpagesize());
109 return bs
->bl
.opt_mem_alignment
;
112 size_t bdrv_min_mem_align(BlockDriverState
*bs
)
114 if (!bs
|| !bs
->drv
) {
115 /* page size or 4k (hdd sector size) should be on the safe side */
116 return MAX(4096, getpagesize());
119 return bs
->bl
.min_mem_alignment
;
122 /* check if the path starts with "<protocol>:" */
123 int path_has_protocol(const char *path
)
128 if (is_windows_drive(path
) ||
129 is_windows_drive_prefix(path
)) {
132 p
= path
+ strcspn(path
, ":/\\");
134 p
= path
+ strcspn(path
, ":/");
140 int path_is_absolute(const char *path
)
143 /* specific case for names like: "\\.\d:" */
144 if (is_windows_drive(path
) || is_windows_drive_prefix(path
)) {
147 return (*path
== '/' || *path
== '\\');
149 return (*path
== '/');
153 /* if filename is absolute, just copy it to dest. Otherwise, build a
154 path to it by considering it is relative to base_path. URL are
156 void path_combine(char *dest
, int dest_size
,
157 const char *base_path
,
158 const char *filename
)
165 if (path_is_absolute(filename
)) {
166 pstrcpy(dest
, dest_size
, filename
);
168 const char *protocol_stripped
= NULL
;
170 if (path_has_protocol(base_path
)) {
171 protocol_stripped
= strchr(base_path
, ':');
172 if (protocol_stripped
) {
176 p
= protocol_stripped
?: base_path
;
178 p1
= strrchr(base_path
, '/');
182 p2
= strrchr(base_path
, '\\');
194 if (len
> dest_size
- 1)
196 memcpy(dest
, base_path
, len
);
198 pstrcat(dest
, dest_size
, filename
);
203 * Helper function for bdrv_parse_filename() implementations to remove optional
204 * protocol prefixes (especially "file:") from a filename and for putting the
205 * stripped filename into the options QDict if there is such a prefix.
207 void bdrv_parse_filename_strip_prefix(const char *filename
, const char *prefix
,
210 if (strstart(filename
, prefix
, &filename
)) {
211 /* Stripping the explicit protocol prefix may result in a protocol
212 * prefix being (wrongly) detected (if the filename contains a colon) */
213 if (path_has_protocol(filename
)) {
214 QString
*fat_filename
;
216 /* This means there is some colon before the first slash; therefore,
217 * this cannot be an absolute path */
218 assert(!path_is_absolute(filename
));
220 /* And we can thus fix the protocol detection issue by prefixing it
222 fat_filename
= qstring_from_str("./");
223 qstring_append(fat_filename
, filename
);
225 assert(!path_has_protocol(qstring_get_str(fat_filename
)));
227 qdict_put(options
, "filename", fat_filename
);
229 /* If no protocol prefix was detected, we can use the shortened
231 qdict_put_str(options
, "filename", filename
);
237 /* Returns whether the image file is opened as read-only. Note that this can
238 * return false and writing to the image file is still not possible because the
239 * image is inactivated. */
240 bool bdrv_is_read_only(BlockDriverState
*bs
)
242 return bs
->read_only
;
245 int bdrv_can_set_read_only(BlockDriverState
*bs
, bool read_only
,
246 bool ignore_allow_rdw
, Error
**errp
)
248 /* Do not set read_only if copy_on_read is enabled */
249 if (bs
->copy_on_read
&& read_only
) {
250 error_setg(errp
, "Can't set node '%s' to r/o with copy-on-read enabled",
251 bdrv_get_device_or_node_name(bs
));
255 /* Do not clear read_only if it is prohibited */
256 if (!read_only
&& !(bs
->open_flags
& BDRV_O_ALLOW_RDWR
) &&
259 error_setg(errp
, "Node '%s' is read only",
260 bdrv_get_device_or_node_name(bs
));
267 /* TODO Remove (deprecated since 2.11)
268 * Block drivers are not supposed to automatically change bs->read_only.
269 * Instead, they should just check whether they can provide what the user
270 * explicitly requested and error out if read-write is requested, but they can
271 * only provide read-only access. */
272 int bdrv_set_read_only(BlockDriverState
*bs
, bool read_only
, Error
**errp
)
276 ret
= bdrv_can_set_read_only(bs
, read_only
, false, errp
);
281 bs
->read_only
= read_only
;
285 void bdrv_get_full_backing_filename_from_filename(const char *backed
,
287 char *dest
, size_t sz
,
290 if (backing
[0] == '\0' || path_has_protocol(backing
) ||
291 path_is_absolute(backing
))
293 pstrcpy(dest
, sz
, backing
);
294 } else if (backed
[0] == '\0' || strstart(backed
, "json:", NULL
)) {
295 error_setg(errp
, "Cannot use relative backing file names for '%s'",
298 path_combine(dest
, sz
, backed
, backing
);
302 void bdrv_get_full_backing_filename(BlockDriverState
*bs
, char *dest
, size_t sz
,
305 char *backed
= bs
->exact_filename
[0] ? bs
->exact_filename
: bs
->filename
;
307 bdrv_get_full_backing_filename_from_filename(backed
, bs
->backing_file
,
311 void bdrv_register(BlockDriver
*bdrv
)
313 QLIST_INSERT_HEAD(&bdrv_drivers
, bdrv
, list
);
316 BlockDriverState
*bdrv_new(void)
318 BlockDriverState
*bs
;
321 bs
= g_new0(BlockDriverState
, 1);
322 QLIST_INIT(&bs
->dirty_bitmaps
);
323 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
324 QLIST_INIT(&bs
->op_blockers
[i
]);
326 notifier_with_return_list_init(&bs
->before_write_notifiers
);
327 qemu_co_mutex_init(&bs
->reqs_lock
);
328 qemu_mutex_init(&bs
->dirty_bitmap_mutex
);
330 bs
->aio_context
= qemu_get_aio_context();
332 qemu_co_queue_init(&bs
->flush_queue
);
334 QTAILQ_INSERT_TAIL(&all_bdrv_states
, bs
, bs_list
);
339 static BlockDriver
*bdrv_do_find_format(const char *format_name
)
343 QLIST_FOREACH(drv1
, &bdrv_drivers
, list
) {
344 if (!strcmp(drv1
->format_name
, format_name
)) {
352 BlockDriver
*bdrv_find_format(const char *format_name
)
357 drv1
= bdrv_do_find_format(format_name
);
362 /* The driver isn't registered, maybe we need to load a module */
363 for (i
= 0; i
< (int)ARRAY_SIZE(block_driver_modules
); ++i
) {
364 if (!strcmp(block_driver_modules
[i
].format_name
, format_name
)) {
365 block_module_load_one(block_driver_modules
[i
].library_name
);
370 return bdrv_do_find_format(format_name
);
373 int bdrv_is_whitelisted(BlockDriver
*drv
, bool read_only
)
375 static const char *whitelist_rw
[] = {
376 CONFIG_BDRV_RW_WHITELIST
378 static const char *whitelist_ro
[] = {
379 CONFIG_BDRV_RO_WHITELIST
383 if (!whitelist_rw
[0] && !whitelist_ro
[0]) {
384 return 1; /* no whitelist, anything goes */
387 for (p
= whitelist_rw
; *p
; p
++) {
388 if (!strcmp(drv
->format_name
, *p
)) {
393 for (p
= whitelist_ro
; *p
; p
++) {
394 if (!strcmp(drv
->format_name
, *p
)) {
402 bool bdrv_uses_whitelist(void)
404 return use_bdrv_whitelist
;
407 typedef struct CreateCo
{
415 static void coroutine_fn
bdrv_create_co_entry(void *opaque
)
417 Error
*local_err
= NULL
;
420 CreateCo
*cco
= opaque
;
423 ret
= cco
->drv
->bdrv_co_create_opts(cco
->filename
, cco
->opts
, &local_err
);
424 error_propagate(&cco
->err
, local_err
);
428 int bdrv_create(BlockDriver
*drv
, const char* filename
,
429 QemuOpts
*opts
, Error
**errp
)
436 .filename
= g_strdup(filename
),
442 if (!drv
->bdrv_co_create_opts
) {
443 error_setg(errp
, "Driver '%s' does not support image creation", drv
->format_name
);
448 if (qemu_in_coroutine()) {
449 /* Fast-path if already in coroutine context */
450 bdrv_create_co_entry(&cco
);
452 co
= qemu_coroutine_create(bdrv_create_co_entry
, &cco
);
453 qemu_coroutine_enter(co
);
454 while (cco
.ret
== NOT_DONE
) {
455 aio_poll(qemu_get_aio_context(), true);
462 error_propagate(errp
, cco
.err
);
464 error_setg_errno(errp
, -ret
, "Could not create image");
469 g_free(cco
.filename
);
473 int bdrv_create_file(const char *filename
, QemuOpts
*opts
, Error
**errp
)
476 Error
*local_err
= NULL
;
479 drv
= bdrv_find_protocol(filename
, true, errp
);
484 ret
= bdrv_create(drv
, filename
, opts
, &local_err
);
485 error_propagate(errp
, local_err
);
490 * Try to get @bs's logical and physical block size.
491 * On success, store them in @bsz struct and return 0.
492 * On failure return -errno.
493 * @bs must not be empty.
495 int bdrv_probe_blocksizes(BlockDriverState
*bs
, BlockSizes
*bsz
)
497 BlockDriver
*drv
= bs
->drv
;
499 if (drv
&& drv
->bdrv_probe_blocksizes
) {
500 return drv
->bdrv_probe_blocksizes(bs
, bsz
);
501 } else if (drv
&& drv
->is_filter
&& bs
->file
) {
502 return bdrv_probe_blocksizes(bs
->file
->bs
, bsz
);
509 * Try to get @bs's geometry (cyls, heads, sectors).
510 * On success, store them in @geo struct and return 0.
511 * On failure return -errno.
512 * @bs must not be empty.
514 int bdrv_probe_geometry(BlockDriverState
*bs
, HDGeometry
*geo
)
516 BlockDriver
*drv
= bs
->drv
;
518 if (drv
&& drv
->bdrv_probe_geometry
) {
519 return drv
->bdrv_probe_geometry(bs
, geo
);
520 } else if (drv
&& drv
->is_filter
&& bs
->file
) {
521 return bdrv_probe_geometry(bs
->file
->bs
, geo
);
528 * Create a uniquely-named empty temporary file.
529 * Return 0 upon success, otherwise a negative errno value.
531 int get_tmp_filename(char *filename
, int size
)
534 char temp_dir
[MAX_PATH
];
535 /* GetTempFileName requires that its output buffer (4th param)
536 have length MAX_PATH or greater. */
537 assert(size
>= MAX_PATH
);
538 return (GetTempPath(MAX_PATH
, temp_dir
)
539 && GetTempFileName(temp_dir
, "qem", 0, filename
)
540 ? 0 : -GetLastError());
544 tmpdir
= getenv("TMPDIR");
548 if (snprintf(filename
, size
, "%s/vl.XXXXXX", tmpdir
) >= size
) {
551 fd
= mkstemp(filename
);
555 if (close(fd
) != 0) {
564 * Detect host devices. By convention, /dev/cdrom[N] is always
565 * recognized as a host CDROM.
567 static BlockDriver
*find_hdev_driver(const char *filename
)
569 int score_max
= 0, score
;
570 BlockDriver
*drv
= NULL
, *d
;
572 QLIST_FOREACH(d
, &bdrv_drivers
, list
) {
573 if (d
->bdrv_probe_device
) {
574 score
= d
->bdrv_probe_device(filename
);
575 if (score
> score_max
) {
585 static BlockDriver
*bdrv_do_find_protocol(const char *protocol
)
589 QLIST_FOREACH(drv1
, &bdrv_drivers
, list
) {
590 if (drv1
->protocol_name
&& !strcmp(drv1
->protocol_name
, protocol
)) {
598 BlockDriver
*bdrv_find_protocol(const char *filename
,
599 bool allow_protocol_prefix
,
608 /* TODO Drivers without bdrv_file_open must be specified explicitly */
611 * XXX(hch): we really should not let host device detection
612 * override an explicit protocol specification, but moving this
613 * later breaks access to device names with colons in them.
614 * Thanks to the brain-dead persistent naming schemes on udev-
615 * based Linux systems those actually are quite common.
617 drv1
= find_hdev_driver(filename
);
622 if (!path_has_protocol(filename
) || !allow_protocol_prefix
) {
626 p
= strchr(filename
, ':');
629 if (len
> sizeof(protocol
) - 1)
630 len
= sizeof(protocol
) - 1;
631 memcpy(protocol
, filename
, len
);
632 protocol
[len
] = '\0';
634 drv1
= bdrv_do_find_protocol(protocol
);
639 for (i
= 0; i
< (int)ARRAY_SIZE(block_driver_modules
); ++i
) {
640 if (block_driver_modules
[i
].protocol_name
&&
641 !strcmp(block_driver_modules
[i
].protocol_name
, protocol
)) {
642 block_module_load_one(block_driver_modules
[i
].library_name
);
647 drv1
= bdrv_do_find_protocol(protocol
);
649 error_setg(errp
, "Unknown protocol '%s'", protocol
);
655 * Guess image format by probing its contents.
656 * This is not a good idea when your image is raw (CVE-2008-2004), but
657 * we do it anyway for backward compatibility.
659 * @buf contains the image's first @buf_size bytes.
660 * @buf_size is the buffer size in bytes (generally BLOCK_PROBE_BUF_SIZE,
661 * but can be smaller if the image file is smaller)
662 * @filename is its filename.
664 * For all block drivers, call the bdrv_probe() method to get its
666 * Return the first block driver with the highest probing score.
668 BlockDriver
*bdrv_probe_all(const uint8_t *buf
, int buf_size
,
669 const char *filename
)
671 int score_max
= 0, score
;
672 BlockDriver
*drv
= NULL
, *d
;
674 QLIST_FOREACH(d
, &bdrv_drivers
, list
) {
676 score
= d
->bdrv_probe(buf
, buf_size
, filename
);
677 if (score
> score_max
) {
687 static int find_image_format(BlockBackend
*file
, const char *filename
,
688 BlockDriver
**pdrv
, Error
**errp
)
691 uint8_t buf
[BLOCK_PROBE_BUF_SIZE
];
694 /* Return the raw BlockDriver * to scsi-generic devices or empty drives */
695 if (blk_is_sg(file
) || !blk_is_inserted(file
) || blk_getlength(file
) == 0) {
700 ret
= blk_pread(file
, 0, buf
, sizeof(buf
));
702 error_setg_errno(errp
, -ret
, "Could not read image for determining its "
708 drv
= bdrv_probe_all(buf
, ret
, filename
);
710 error_setg(errp
, "Could not determine image format: No compatible "
719 * Set the current 'total_sectors' value
720 * Return 0 on success, -errno on error.
722 static int refresh_total_sectors(BlockDriverState
*bs
, int64_t hint
)
724 BlockDriver
*drv
= bs
->drv
;
730 /* Do not attempt drv->bdrv_getlength() on scsi-generic devices */
734 /* query actual device if possible, otherwise just trust the hint */
735 if (drv
->bdrv_getlength
) {
736 int64_t length
= drv
->bdrv_getlength(bs
);
740 hint
= DIV_ROUND_UP(length
, BDRV_SECTOR_SIZE
);
743 bs
->total_sectors
= hint
;
748 * Combines a QDict of new block driver @options with any missing options taken
749 * from @old_options, so that leaving out an option defaults to its old value.
751 static void bdrv_join_options(BlockDriverState
*bs
, QDict
*options
,
754 if (bs
->drv
&& bs
->drv
->bdrv_join_options
) {
755 bs
->drv
->bdrv_join_options(options
, old_options
);
757 qdict_join(options
, old_options
, false);
762 * Set open flags for a given discard mode
764 * Return 0 on success, -1 if the discard mode was invalid.
766 int bdrv_parse_discard_flags(const char *mode
, int *flags
)
768 *flags
&= ~BDRV_O_UNMAP
;
770 if (!strcmp(mode
, "off") || !strcmp(mode
, "ignore")) {
772 } else if (!strcmp(mode
, "on") || !strcmp(mode
, "unmap")) {
773 *flags
|= BDRV_O_UNMAP
;
782 * Set open flags for a given cache mode
784 * Return 0 on success, -1 if the cache mode was invalid.
786 int bdrv_parse_cache_mode(const char *mode
, int *flags
, bool *writethrough
)
788 *flags
&= ~BDRV_O_CACHE_MASK
;
790 if (!strcmp(mode
, "off") || !strcmp(mode
, "none")) {
791 *writethrough
= false;
792 *flags
|= BDRV_O_NOCACHE
;
793 } else if (!strcmp(mode
, "directsync")) {
794 *writethrough
= true;
795 *flags
|= BDRV_O_NOCACHE
;
796 } else if (!strcmp(mode
, "writeback")) {
797 *writethrough
= false;
798 } else if (!strcmp(mode
, "unsafe")) {
799 *writethrough
= false;
800 *flags
|= BDRV_O_NO_FLUSH
;
801 } else if (!strcmp(mode
, "writethrough")) {
802 *writethrough
= true;
810 static char *bdrv_child_get_parent_desc(BdrvChild
*c
)
812 BlockDriverState
*parent
= c
->opaque
;
813 return g_strdup(bdrv_get_device_or_node_name(parent
));
816 static void bdrv_child_cb_drained_begin(BdrvChild
*child
)
818 BlockDriverState
*bs
= child
->opaque
;
819 bdrv_drained_begin(bs
);
822 static void bdrv_child_cb_drained_end(BdrvChild
*child
)
824 BlockDriverState
*bs
= child
->opaque
;
825 bdrv_drained_end(bs
);
828 static void bdrv_child_cb_attach(BdrvChild
*child
)
830 BlockDriverState
*bs
= child
->opaque
;
831 bdrv_apply_subtree_drain(child
, bs
);
834 static void bdrv_child_cb_detach(BdrvChild
*child
)
836 BlockDriverState
*bs
= child
->opaque
;
837 bdrv_unapply_subtree_drain(child
, bs
);
840 static int bdrv_child_cb_inactivate(BdrvChild
*child
)
842 BlockDriverState
*bs
= child
->opaque
;
843 assert(bs
->open_flags
& BDRV_O_INACTIVE
);
848 * Returns the options and flags that a temporary snapshot should get, based on
849 * the originally requested flags (the originally requested image will have
850 * flags like a backing file)
852 static void bdrv_temp_snapshot_options(int *child_flags
, QDict
*child_options
,
853 int parent_flags
, QDict
*parent_options
)
855 *child_flags
= (parent_flags
& ~BDRV_O_SNAPSHOT
) | BDRV_O_TEMPORARY
;
857 /* For temporary files, unconditional cache=unsafe is fine */
858 qdict_set_default_str(child_options
, BDRV_OPT_CACHE_DIRECT
, "off");
859 qdict_set_default_str(child_options
, BDRV_OPT_CACHE_NO_FLUSH
, "on");
861 /* Copy the read-only option from the parent */
862 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_READ_ONLY
);
864 /* aio=native doesn't work for cache.direct=off, so disable it for the
865 * temporary snapshot */
866 *child_flags
&= ~BDRV_O_NATIVE_AIO
;
870 * Returns the options and flags that bs->file should get if a protocol driver
871 * is expected, based on the given options and flags for the parent BDS
873 static void bdrv_inherited_options(int *child_flags
, QDict
*child_options
,
874 int parent_flags
, QDict
*parent_options
)
876 int flags
= parent_flags
;
878 /* Enable protocol handling, disable format probing for bs->file */
879 flags
|= BDRV_O_PROTOCOL
;
881 /* If the cache mode isn't explicitly set, inherit direct and no-flush from
883 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_CACHE_DIRECT
);
884 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_CACHE_NO_FLUSH
);
885 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_FORCE_SHARE
);
887 /* Inherit the read-only option from the parent if it's not set */
888 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_READ_ONLY
);
890 /* Our block drivers take care to send flushes and respect unmap policy,
891 * so we can default to enable both on lower layers regardless of the
892 * corresponding parent options. */
893 qdict_set_default_str(child_options
, BDRV_OPT_DISCARD
, "unmap");
895 /* Clear flags that only apply to the top layer */
896 flags
&= ~(BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
| BDRV_O_COPY_ON_READ
|
899 *child_flags
= flags
;
902 const BdrvChildRole child_file
= {
903 .get_parent_desc
= bdrv_child_get_parent_desc
,
904 .inherit_options
= bdrv_inherited_options
,
905 .drained_begin
= bdrv_child_cb_drained_begin
,
906 .drained_end
= bdrv_child_cb_drained_end
,
907 .attach
= bdrv_child_cb_attach
,
908 .detach
= bdrv_child_cb_detach
,
909 .inactivate
= bdrv_child_cb_inactivate
,
913 * Returns the options and flags that bs->file should get if the use of formats
914 * (and not only protocols) is permitted for it, based on the given options and
915 * flags for the parent BDS
917 static void bdrv_inherited_fmt_options(int *child_flags
, QDict
*child_options
,
918 int parent_flags
, QDict
*parent_options
)
920 child_file
.inherit_options(child_flags
, child_options
,
921 parent_flags
, parent_options
);
923 *child_flags
&= ~(BDRV_O_PROTOCOL
| BDRV_O_NO_IO
);
926 const BdrvChildRole child_format
= {
927 .get_parent_desc
= bdrv_child_get_parent_desc
,
928 .inherit_options
= bdrv_inherited_fmt_options
,
929 .drained_begin
= bdrv_child_cb_drained_begin
,
930 .drained_end
= bdrv_child_cb_drained_end
,
931 .attach
= bdrv_child_cb_attach
,
932 .detach
= bdrv_child_cb_detach
,
933 .inactivate
= bdrv_child_cb_inactivate
,
936 static void bdrv_backing_attach(BdrvChild
*c
)
938 BlockDriverState
*parent
= c
->opaque
;
939 BlockDriverState
*backing_hd
= c
->bs
;
941 assert(!parent
->backing_blocker
);
942 error_setg(&parent
->backing_blocker
,
943 "node is used as backing hd of '%s'",
944 bdrv_get_device_or_node_name(parent
));
946 parent
->open_flags
&= ~BDRV_O_NO_BACKING
;
947 pstrcpy(parent
->backing_file
, sizeof(parent
->backing_file
),
948 backing_hd
->filename
);
949 pstrcpy(parent
->backing_format
, sizeof(parent
->backing_format
),
950 backing_hd
->drv
? backing_hd
->drv
->format_name
: "");
952 bdrv_op_block_all(backing_hd
, parent
->backing_blocker
);
953 /* Otherwise we won't be able to commit or stream */
954 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_COMMIT_TARGET
,
955 parent
->backing_blocker
);
956 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_STREAM
,
957 parent
->backing_blocker
);
959 * We do backup in 3 ways:
961 * The target bs is new opened, and the source is top BDS
963 * Both the source and the target are top BDSes.
964 * 3. internal backup(used for block replication)
965 * Both the source and the target are backing file
967 * In case 1 and 2, neither the source nor the target is the backing file.
968 * In case 3, we will block the top BDS, so there is only one block job
969 * for the top BDS and its backing chain.
971 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_BACKUP_SOURCE
,
972 parent
->backing_blocker
);
973 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_BACKUP_TARGET
,
974 parent
->backing_blocker
);
976 bdrv_child_cb_attach(c
);
979 static void bdrv_backing_detach(BdrvChild
*c
)
981 BlockDriverState
*parent
= c
->opaque
;
983 assert(parent
->backing_blocker
);
984 bdrv_op_unblock_all(c
->bs
, parent
->backing_blocker
);
985 error_free(parent
->backing_blocker
);
986 parent
->backing_blocker
= NULL
;
988 bdrv_child_cb_detach(c
);
992 * Returns the options and flags that bs->backing should get, based on the
993 * given options and flags for the parent BDS
995 static void bdrv_backing_options(int *child_flags
, QDict
*child_options
,
996 int parent_flags
, QDict
*parent_options
)
998 int flags
= parent_flags
;
1000 /* The cache mode is inherited unmodified for backing files; except WCE,
1001 * which is only applied on the top level (BlockBackend) */
1002 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_CACHE_DIRECT
);
1003 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_CACHE_NO_FLUSH
);
1004 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_FORCE_SHARE
);
1006 /* backing files always opened read-only */
1007 qdict_set_default_str(child_options
, BDRV_OPT_READ_ONLY
, "on");
1008 flags
&= ~BDRV_O_COPY_ON_READ
;
1010 /* snapshot=on is handled on the top layer */
1011 flags
&= ~(BDRV_O_SNAPSHOT
| BDRV_O_TEMPORARY
);
1013 *child_flags
= flags
;
1016 static int bdrv_backing_update_filename(BdrvChild
*c
, BlockDriverState
*base
,
1017 const char *filename
, Error
**errp
)
1019 BlockDriverState
*parent
= c
->opaque
;
1020 int orig_flags
= bdrv_get_flags(parent
);
1023 if (!(orig_flags
& BDRV_O_RDWR
)) {
1024 ret
= bdrv_reopen(parent
, orig_flags
| BDRV_O_RDWR
, errp
);
1030 ret
= bdrv_change_backing_file(parent
, filename
,
1031 base
->drv
? base
->drv
->format_name
: "");
1033 error_setg_errno(errp
, -ret
, "Could not update backing file link");
1036 if (!(orig_flags
& BDRV_O_RDWR
)) {
1037 bdrv_reopen(parent
, orig_flags
, NULL
);
1043 const BdrvChildRole child_backing
= {
1044 .get_parent_desc
= bdrv_child_get_parent_desc
,
1045 .attach
= bdrv_backing_attach
,
1046 .detach
= bdrv_backing_detach
,
1047 .inherit_options
= bdrv_backing_options
,
1048 .drained_begin
= bdrv_child_cb_drained_begin
,
1049 .drained_end
= bdrv_child_cb_drained_end
,
1050 .inactivate
= bdrv_child_cb_inactivate
,
1051 .update_filename
= bdrv_backing_update_filename
,
1054 static int bdrv_open_flags(BlockDriverState
*bs
, int flags
)
1056 int open_flags
= flags
;
1059 * Clear flags that are internal to the block layer before opening the
1062 open_flags
&= ~(BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
| BDRV_O_PROTOCOL
);
1065 * Snapshots should be writable.
1067 if (flags
& BDRV_O_TEMPORARY
) {
1068 open_flags
|= BDRV_O_RDWR
;
1074 static void update_flags_from_options(int *flags
, QemuOpts
*opts
)
1076 *flags
&= ~BDRV_O_CACHE_MASK
;
1078 assert(qemu_opt_find(opts
, BDRV_OPT_CACHE_NO_FLUSH
));
1079 if (qemu_opt_get_bool(opts
, BDRV_OPT_CACHE_NO_FLUSH
, false)) {
1080 *flags
|= BDRV_O_NO_FLUSH
;
1083 assert(qemu_opt_find(opts
, BDRV_OPT_CACHE_DIRECT
));
1084 if (qemu_opt_get_bool(opts
, BDRV_OPT_CACHE_DIRECT
, false)) {
1085 *flags
|= BDRV_O_NOCACHE
;
1088 *flags
&= ~BDRV_O_RDWR
;
1090 assert(qemu_opt_find(opts
, BDRV_OPT_READ_ONLY
));
1091 if (!qemu_opt_get_bool(opts
, BDRV_OPT_READ_ONLY
, false)) {
1092 *flags
|= BDRV_O_RDWR
;
1097 static void update_options_from_flags(QDict
*options
, int flags
)
1099 if (!qdict_haskey(options
, BDRV_OPT_CACHE_DIRECT
)) {
1100 qdict_put_bool(options
, BDRV_OPT_CACHE_DIRECT
, flags
& BDRV_O_NOCACHE
);
1102 if (!qdict_haskey(options
, BDRV_OPT_CACHE_NO_FLUSH
)) {
1103 qdict_put_bool(options
, BDRV_OPT_CACHE_NO_FLUSH
,
1104 flags
& BDRV_O_NO_FLUSH
);
1106 if (!qdict_haskey(options
, BDRV_OPT_READ_ONLY
)) {
1107 qdict_put_bool(options
, BDRV_OPT_READ_ONLY
, !(flags
& BDRV_O_RDWR
));
1111 static void bdrv_assign_node_name(BlockDriverState
*bs
,
1112 const char *node_name
,
1115 char *gen_node_name
= NULL
;
1118 node_name
= gen_node_name
= id_generate(ID_BLOCK
);
1119 } else if (!id_wellformed(node_name
)) {
1121 * Check for empty string or invalid characters, but not if it is
1122 * generated (generated names use characters not available to the user)
1124 error_setg(errp
, "Invalid node name");
1128 /* takes care of avoiding namespaces collisions */
1129 if (blk_by_name(node_name
)) {
1130 error_setg(errp
, "node-name=%s is conflicting with a device id",
1135 /* takes care of avoiding duplicates node names */
1136 if (bdrv_find_node(node_name
)) {
1137 error_setg(errp
, "Duplicate node name");
1141 /* copy node name into the bs and insert it into the graph list */
1142 pstrcpy(bs
->node_name
, sizeof(bs
->node_name
), node_name
);
1143 QTAILQ_INSERT_TAIL(&graph_bdrv_states
, bs
, node_list
);
1145 g_free(gen_node_name
);
1148 static int bdrv_open_driver(BlockDriverState
*bs
, BlockDriver
*drv
,
1149 const char *node_name
, QDict
*options
,
1150 int open_flags
, Error
**errp
)
1152 Error
*local_err
= NULL
;
1155 bdrv_assign_node_name(bs
, node_name
, &local_err
);
1157 error_propagate(errp
, local_err
);
1162 bs
->read_only
= !(bs
->open_flags
& BDRV_O_RDWR
);
1163 bs
->opaque
= g_malloc0(drv
->instance_size
);
1165 if (drv
->bdrv_file_open
) {
1166 assert(!drv
->bdrv_needs_filename
|| bs
->filename
[0]);
1167 ret
= drv
->bdrv_file_open(bs
, options
, open_flags
, &local_err
);
1168 } else if (drv
->bdrv_open
) {
1169 ret
= drv
->bdrv_open(bs
, options
, open_flags
, &local_err
);
1176 error_propagate(errp
, local_err
);
1177 } else if (bs
->filename
[0]) {
1178 error_setg_errno(errp
, -ret
, "Could not open '%s'", bs
->filename
);
1180 error_setg_errno(errp
, -ret
, "Could not open image");
1185 ret
= refresh_total_sectors(bs
, bs
->total_sectors
);
1187 error_setg_errno(errp
, -ret
, "Could not refresh total sector count");
1191 bdrv_refresh_limits(bs
, &local_err
);
1193 error_propagate(errp
, local_err
);
1197 assert(bdrv_opt_mem_align(bs
) != 0);
1198 assert(bdrv_min_mem_align(bs
) != 0);
1199 assert(is_power_of_2(bs
->bl
.request_alignment
));
1204 if (bs
->file
!= NULL
) {
1205 bdrv_unref_child(bs
, bs
->file
);
1213 BlockDriverState
*bdrv_new_open_driver(BlockDriver
*drv
, const char *node_name
,
1214 int flags
, Error
**errp
)
1216 BlockDriverState
*bs
;
1220 bs
->open_flags
= flags
;
1221 bs
->explicit_options
= qdict_new();
1222 bs
->options
= qdict_new();
1225 update_options_from_flags(bs
->options
, flags
);
1227 ret
= bdrv_open_driver(bs
, drv
, node_name
, bs
->options
, flags
, errp
);
1229 QDECREF(bs
->explicit_options
);
1230 bs
->explicit_options
= NULL
;
1231 QDECREF(bs
->options
);
1240 QemuOptsList bdrv_runtime_opts
= {
1241 .name
= "bdrv_common",
1242 .head
= QTAILQ_HEAD_INITIALIZER(bdrv_runtime_opts
.head
),
1245 .name
= "node-name",
1246 .type
= QEMU_OPT_STRING
,
1247 .help
= "Node name of the block device node",
1251 .type
= QEMU_OPT_STRING
,
1252 .help
= "Block driver to use for the node",
1255 .name
= BDRV_OPT_CACHE_DIRECT
,
1256 .type
= QEMU_OPT_BOOL
,
1257 .help
= "Bypass software writeback cache on the host",
1260 .name
= BDRV_OPT_CACHE_NO_FLUSH
,
1261 .type
= QEMU_OPT_BOOL
,
1262 .help
= "Ignore flush requests",
1265 .name
= BDRV_OPT_READ_ONLY
,
1266 .type
= QEMU_OPT_BOOL
,
1267 .help
= "Node is opened in read-only mode",
1270 .name
= "detect-zeroes",
1271 .type
= QEMU_OPT_STRING
,
1272 .help
= "try to optimize zero writes (off, on, unmap)",
1276 .type
= QEMU_OPT_STRING
,
1277 .help
= "discard operation (ignore/off, unmap/on)",
1280 .name
= BDRV_OPT_FORCE_SHARE
,
1281 .type
= QEMU_OPT_BOOL
,
1282 .help
= "always accept other writers (default: off)",
1284 { /* end of list */ }
1289 * Common part for opening disk images and files
1291 * Removes all processed options from *options.
1293 static int bdrv_open_common(BlockDriverState
*bs
, BlockBackend
*file
,
1294 QDict
*options
, Error
**errp
)
1296 int ret
, open_flags
;
1297 const char *filename
;
1298 const char *driver_name
= NULL
;
1299 const char *node_name
= NULL
;
1300 const char *discard
;
1301 const char *detect_zeroes
;
1304 Error
*local_err
= NULL
;
1306 assert(bs
->file
== NULL
);
1307 assert(options
!= NULL
&& bs
->options
!= options
);
1309 opts
= qemu_opts_create(&bdrv_runtime_opts
, NULL
, 0, &error_abort
);
1310 qemu_opts_absorb_qdict(opts
, options
, &local_err
);
1312 error_propagate(errp
, local_err
);
1317 update_flags_from_options(&bs
->open_flags
, opts
);
1319 driver_name
= qemu_opt_get(opts
, "driver");
1320 drv
= bdrv_find_format(driver_name
);
1321 assert(drv
!= NULL
);
1323 bs
->force_share
= qemu_opt_get_bool(opts
, BDRV_OPT_FORCE_SHARE
, false);
1325 if (bs
->force_share
&& (bs
->open_flags
& BDRV_O_RDWR
)) {
1327 BDRV_OPT_FORCE_SHARE
1328 "=on can only be used with read-only images");
1334 filename
= blk_bs(file
)->filename
;
1337 * Caution: while qdict_get_try_str() is fine, getting
1338 * non-string types would require more care. When @options
1339 * come from -blockdev or blockdev_add, its members are typed
1340 * according to the QAPI schema, but when they come from
1341 * -drive, they're all QString.
1343 filename
= qdict_get_try_str(options
, "filename");
1346 if (drv
->bdrv_needs_filename
&& (!filename
|| !filename
[0])) {
1347 error_setg(errp
, "The '%s' block driver requires a file name",
1353 trace_bdrv_open_common(bs
, filename
?: "", bs
->open_flags
,
1356 bs
->read_only
= !(bs
->open_flags
& BDRV_O_RDWR
);
1358 if (use_bdrv_whitelist
&& !bdrv_is_whitelisted(drv
, bs
->read_only
)) {
1360 !bs
->read_only
&& bdrv_is_whitelisted(drv
, true)
1361 ? "Driver '%s' can only be used for read-only devices"
1362 : "Driver '%s' is not whitelisted",
1368 /* bdrv_new() and bdrv_close() make it so */
1369 assert(atomic_read(&bs
->copy_on_read
) == 0);
1371 if (bs
->open_flags
& BDRV_O_COPY_ON_READ
) {
1372 if (!bs
->read_only
) {
1373 bdrv_enable_copy_on_read(bs
);
1375 error_setg(errp
, "Can't use copy-on-read on read-only device");
1381 discard
= qemu_opt_get(opts
, "discard");
1382 if (discard
!= NULL
) {
1383 if (bdrv_parse_discard_flags(discard
, &bs
->open_flags
) != 0) {
1384 error_setg(errp
, "Invalid discard option");
1390 detect_zeroes
= qemu_opt_get(opts
, "detect-zeroes");
1391 if (detect_zeroes
) {
1392 BlockdevDetectZeroesOptions value
=
1393 qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup
,
1395 BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF
,
1398 error_propagate(errp
, local_err
);
1403 if (value
== BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP
&&
1404 !(bs
->open_flags
& BDRV_O_UNMAP
))
1406 error_setg(errp
, "setting detect-zeroes to unmap is not allowed "
1407 "without setting discard operation to unmap");
1412 bs
->detect_zeroes
= value
;
1415 if (filename
!= NULL
) {
1416 pstrcpy(bs
->filename
, sizeof(bs
->filename
), filename
);
1418 bs
->filename
[0] = '\0';
1420 pstrcpy(bs
->exact_filename
, sizeof(bs
->exact_filename
), bs
->filename
);
1422 /* Open the image, either directly or using a protocol */
1423 open_flags
= bdrv_open_flags(bs
, bs
->open_flags
);
1424 node_name
= qemu_opt_get(opts
, "node-name");
1426 assert(!drv
->bdrv_file_open
|| file
== NULL
);
1427 ret
= bdrv_open_driver(bs
, drv
, node_name
, options
, open_flags
, errp
);
1432 qemu_opts_del(opts
);
1436 qemu_opts_del(opts
);
1440 static QDict
*parse_json_filename(const char *filename
, Error
**errp
)
1442 QObject
*options_obj
;
1446 ret
= strstart(filename
, "json:", &filename
);
1449 options_obj
= qobject_from_json(filename
, errp
);
1451 /* Work around qobject_from_json() lossage TODO fix that */
1452 if (errp
&& !*errp
) {
1453 error_setg(errp
, "Could not parse the JSON options");
1456 error_prepend(errp
, "Could not parse the JSON options: ");
1460 options
= qobject_to_qdict(options_obj
);
1462 qobject_decref(options_obj
);
1463 error_setg(errp
, "Invalid JSON object given");
1467 qdict_flatten(options
);
1472 static void parse_json_protocol(QDict
*options
, const char **pfilename
,
1475 QDict
*json_options
;
1476 Error
*local_err
= NULL
;
1478 /* Parse json: pseudo-protocol */
1479 if (!*pfilename
|| !g_str_has_prefix(*pfilename
, "json:")) {
1483 json_options
= parse_json_filename(*pfilename
, &local_err
);
1485 error_propagate(errp
, local_err
);
1489 /* Options given in the filename have lower priority than options
1490 * specified directly */
1491 qdict_join(options
, json_options
, false);
1492 QDECREF(json_options
);
1497 * Fills in default options for opening images and converts the legacy
1498 * filename/flags pair to option QDict entries.
1499 * The BDRV_O_PROTOCOL flag in *flags will be set or cleared accordingly if a
1500 * block driver has been specified explicitly.
1502 static int bdrv_fill_options(QDict
**options
, const char *filename
,
1503 int *flags
, Error
**errp
)
1505 const char *drvname
;
1506 bool protocol
= *flags
& BDRV_O_PROTOCOL
;
1507 bool parse_filename
= false;
1508 BlockDriver
*drv
= NULL
;
1509 Error
*local_err
= NULL
;
1512 * Caution: while qdict_get_try_str() is fine, getting non-string
1513 * types would require more care. When @options come from
1514 * -blockdev or blockdev_add, its members are typed according to
1515 * the QAPI schema, but when they come from -drive, they're all
1518 drvname
= qdict_get_try_str(*options
, "driver");
1520 drv
= bdrv_find_format(drvname
);
1522 error_setg(errp
, "Unknown driver '%s'", drvname
);
1525 /* If the user has explicitly specified the driver, this choice should
1526 * override the BDRV_O_PROTOCOL flag */
1527 protocol
= drv
->bdrv_file_open
;
1531 *flags
|= BDRV_O_PROTOCOL
;
1533 *flags
&= ~BDRV_O_PROTOCOL
;
1536 /* Translate cache options from flags into options */
1537 update_options_from_flags(*options
, *flags
);
1539 /* Fetch the file name from the options QDict if necessary */
1540 if (protocol
&& filename
) {
1541 if (!qdict_haskey(*options
, "filename")) {
1542 qdict_put_str(*options
, "filename", filename
);
1543 parse_filename
= true;
1545 error_setg(errp
, "Can't specify 'file' and 'filename' options at "
1551 /* Find the right block driver */
1552 /* See cautionary note on accessing @options above */
1553 filename
= qdict_get_try_str(*options
, "filename");
1555 if (!drvname
&& protocol
) {
1557 drv
= bdrv_find_protocol(filename
, parse_filename
, errp
);
1562 drvname
= drv
->format_name
;
1563 qdict_put_str(*options
, "driver", drvname
);
1565 error_setg(errp
, "Must specify either driver or file");
1570 assert(drv
|| !protocol
);
1572 /* Driver-specific filename parsing */
1573 if (drv
&& drv
->bdrv_parse_filename
&& parse_filename
) {
1574 drv
->bdrv_parse_filename(filename
, *options
, &local_err
);
1576 error_propagate(errp
, local_err
);
1580 if (!drv
->bdrv_needs_filename
) {
1581 qdict_del(*options
, "filename");
1588 static int bdrv_child_check_perm(BdrvChild
*c
, BlockReopenQueue
*q
,
1589 uint64_t perm
, uint64_t shared
,
1590 GSList
*ignore_children
, Error
**errp
);
1591 static void bdrv_child_abort_perm_update(BdrvChild
*c
);
1592 static void bdrv_child_set_perm(BdrvChild
*c
, uint64_t perm
, uint64_t shared
);
1594 typedef struct BlockReopenQueueEntry
{
1596 BDRVReopenState state
;
1597 QSIMPLEQ_ENTRY(BlockReopenQueueEntry
) entry
;
1598 } BlockReopenQueueEntry
;
1601 * Return the flags that @bs will have after the reopens in @q have
1602 * successfully completed. If @q is NULL (or @bs is not contained in @q),
1603 * return the current flags.
1605 static int bdrv_reopen_get_flags(BlockReopenQueue
*q
, BlockDriverState
*bs
)
1607 BlockReopenQueueEntry
*entry
;
1610 QSIMPLEQ_FOREACH(entry
, q
, entry
) {
1611 if (entry
->state
.bs
== bs
) {
1612 return entry
->state
.flags
;
1617 return bs
->open_flags
;
1620 /* Returns whether the image file can be written to after the reopen queue @q
1621 * has been successfully applied, or right now if @q is NULL. */
1622 static bool bdrv_is_writable(BlockDriverState
*bs
, BlockReopenQueue
*q
)
1624 int flags
= bdrv_reopen_get_flags(q
, bs
);
1626 return (flags
& (BDRV_O_RDWR
| BDRV_O_INACTIVE
)) == BDRV_O_RDWR
;
1629 static void bdrv_child_perm(BlockDriverState
*bs
, BlockDriverState
*child_bs
,
1630 BdrvChild
*c
, const BdrvChildRole
*role
,
1631 BlockReopenQueue
*reopen_queue
,
1632 uint64_t parent_perm
, uint64_t parent_shared
,
1633 uint64_t *nperm
, uint64_t *nshared
)
1635 if (bs
->drv
&& bs
->drv
->bdrv_child_perm
) {
1636 bs
->drv
->bdrv_child_perm(bs
, c
, role
, reopen_queue
,
1637 parent_perm
, parent_shared
,
1640 /* TODO Take force_share from reopen_queue */
1641 if (child_bs
&& child_bs
->force_share
) {
1642 *nshared
= BLK_PERM_ALL
;
1647 * Check whether permissions on this node can be changed in a way that
1648 * @cumulative_perms and @cumulative_shared_perms are the new cumulative
1649 * permissions of all its parents. This involves checking whether all necessary
1650 * permission changes to child nodes can be performed.
1652 * A call to this function must always be followed by a call to bdrv_set_perm()
1653 * or bdrv_abort_perm_update().
1655 static int bdrv_check_perm(BlockDriverState
*bs
, BlockReopenQueue
*q
,
1656 uint64_t cumulative_perms
,
1657 uint64_t cumulative_shared_perms
,
1658 GSList
*ignore_children
, Error
**errp
)
1660 BlockDriver
*drv
= bs
->drv
;
1664 /* Write permissions never work with read-only images */
1665 if ((cumulative_perms
& (BLK_PERM_WRITE
| BLK_PERM_WRITE_UNCHANGED
)) &&
1666 !bdrv_is_writable(bs
, q
))
1668 error_setg(errp
, "Block node is read-only");
1672 /* Check this node */
1677 if (drv
->bdrv_check_perm
) {
1678 return drv
->bdrv_check_perm(bs
, cumulative_perms
,
1679 cumulative_shared_perms
, errp
);
1682 /* Drivers that never have children can omit .bdrv_child_perm() */
1683 if (!drv
->bdrv_child_perm
) {
1684 assert(QLIST_EMPTY(&bs
->children
));
1688 /* Check all children */
1689 QLIST_FOREACH(c
, &bs
->children
, next
) {
1690 uint64_t cur_perm
, cur_shared
;
1691 bdrv_child_perm(bs
, c
->bs
, c
, c
->role
, q
,
1692 cumulative_perms
, cumulative_shared_perms
,
1693 &cur_perm
, &cur_shared
);
1694 ret
= bdrv_child_check_perm(c
, q
, cur_perm
, cur_shared
,
1695 ignore_children
, errp
);
1705 * Notifies drivers that after a previous bdrv_check_perm() call, the
1706 * permission update is not performed and any preparations made for it (e.g.
1707 * taken file locks) need to be undone.
1709 * This function recursively notifies all child nodes.
1711 static void bdrv_abort_perm_update(BlockDriverState
*bs
)
1713 BlockDriver
*drv
= bs
->drv
;
1720 if (drv
->bdrv_abort_perm_update
) {
1721 drv
->bdrv_abort_perm_update(bs
);
1724 QLIST_FOREACH(c
, &bs
->children
, next
) {
1725 bdrv_child_abort_perm_update(c
);
1729 static void bdrv_set_perm(BlockDriverState
*bs
, uint64_t cumulative_perms
,
1730 uint64_t cumulative_shared_perms
)
1732 BlockDriver
*drv
= bs
->drv
;
1739 /* Update this node */
1740 if (drv
->bdrv_set_perm
) {
1741 drv
->bdrv_set_perm(bs
, cumulative_perms
, cumulative_shared_perms
);
1744 /* Drivers that never have children can omit .bdrv_child_perm() */
1745 if (!drv
->bdrv_child_perm
) {
1746 assert(QLIST_EMPTY(&bs
->children
));
1750 /* Update all children */
1751 QLIST_FOREACH(c
, &bs
->children
, next
) {
1752 uint64_t cur_perm
, cur_shared
;
1753 bdrv_child_perm(bs
, c
->bs
, c
, c
->role
, NULL
,
1754 cumulative_perms
, cumulative_shared_perms
,
1755 &cur_perm
, &cur_shared
);
1756 bdrv_child_set_perm(c
, cur_perm
, cur_shared
);
1760 static void bdrv_get_cumulative_perm(BlockDriverState
*bs
, uint64_t *perm
,
1761 uint64_t *shared_perm
)
1764 uint64_t cumulative_perms
= 0;
1765 uint64_t cumulative_shared_perms
= BLK_PERM_ALL
;
1767 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
1768 cumulative_perms
|= c
->perm
;
1769 cumulative_shared_perms
&= c
->shared_perm
;
1772 *perm
= cumulative_perms
;
1773 *shared_perm
= cumulative_shared_perms
;
1776 static char *bdrv_child_user_desc(BdrvChild
*c
)
1778 if (c
->role
->get_parent_desc
) {
1779 return c
->role
->get_parent_desc(c
);
1782 return g_strdup("another user");
1785 char *bdrv_perm_names(uint64_t perm
)
1791 { BLK_PERM_CONSISTENT_READ
, "consistent read" },
1792 { BLK_PERM_WRITE
, "write" },
1793 { BLK_PERM_WRITE_UNCHANGED
, "write unchanged" },
1794 { BLK_PERM_RESIZE
, "resize" },
1795 { BLK_PERM_GRAPH_MOD
, "change children" },
1799 char *result
= g_strdup("");
1800 struct perm_name
*p
;
1802 for (p
= permissions
; p
->name
; p
++) {
1803 if (perm
& p
->perm
) {
1805 result
= g_strdup_printf("%s%s%s", old
, *old
? ", " : "", p
->name
);
1814 * Checks whether a new reference to @bs can be added if the new user requires
1815 * @new_used_perm/@new_shared_perm as its permissions. If @ignore_children is
1816 * set, the BdrvChild objects in this list are ignored in the calculations;
1817 * this allows checking permission updates for an existing reference.
1819 * Needs to be followed by a call to either bdrv_set_perm() or
1820 * bdrv_abort_perm_update(). */
1821 static int bdrv_check_update_perm(BlockDriverState
*bs
, BlockReopenQueue
*q
,
1822 uint64_t new_used_perm
,
1823 uint64_t new_shared_perm
,
1824 GSList
*ignore_children
, Error
**errp
)
1827 uint64_t cumulative_perms
= new_used_perm
;
1828 uint64_t cumulative_shared_perms
= new_shared_perm
;
1830 /* There is no reason why anyone couldn't tolerate write_unchanged */
1831 assert(new_shared_perm
& BLK_PERM_WRITE_UNCHANGED
);
1833 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
1834 if (g_slist_find(ignore_children
, c
)) {
1838 if ((new_used_perm
& c
->shared_perm
) != new_used_perm
) {
1839 char *user
= bdrv_child_user_desc(c
);
1840 char *perm_names
= bdrv_perm_names(new_used_perm
& ~c
->shared_perm
);
1841 error_setg(errp
, "Conflicts with use by %s as '%s', which does not "
1843 user
, c
->name
, perm_names
, bdrv_get_node_name(c
->bs
));
1849 if ((c
->perm
& new_shared_perm
) != c
->perm
) {
1850 char *user
= bdrv_child_user_desc(c
);
1851 char *perm_names
= bdrv_perm_names(c
->perm
& ~new_shared_perm
);
1852 error_setg(errp
, "Conflicts with use by %s as '%s', which uses "
1854 user
, c
->name
, perm_names
, bdrv_get_node_name(c
->bs
));
1860 cumulative_perms
|= c
->perm
;
1861 cumulative_shared_perms
&= c
->shared_perm
;
1864 return bdrv_check_perm(bs
, q
, cumulative_perms
, cumulative_shared_perms
,
1865 ignore_children
, errp
);
1868 /* Needs to be followed by a call to either bdrv_child_set_perm() or
1869 * bdrv_child_abort_perm_update(). */
1870 static int bdrv_child_check_perm(BdrvChild
*c
, BlockReopenQueue
*q
,
1871 uint64_t perm
, uint64_t shared
,
1872 GSList
*ignore_children
, Error
**errp
)
1876 ignore_children
= g_slist_prepend(g_slist_copy(ignore_children
), c
);
1877 ret
= bdrv_check_update_perm(c
->bs
, q
, perm
, shared
, ignore_children
, errp
);
1878 g_slist_free(ignore_children
);
1883 static void bdrv_child_set_perm(BdrvChild
*c
, uint64_t perm
, uint64_t shared
)
1885 uint64_t cumulative_perms
, cumulative_shared_perms
;
1888 c
->shared_perm
= shared
;
1890 bdrv_get_cumulative_perm(c
->bs
, &cumulative_perms
,
1891 &cumulative_shared_perms
);
1892 bdrv_set_perm(c
->bs
, cumulative_perms
, cumulative_shared_perms
);
1895 static void bdrv_child_abort_perm_update(BdrvChild
*c
)
1897 bdrv_abort_perm_update(c
->bs
);
1900 int bdrv_child_try_set_perm(BdrvChild
*c
, uint64_t perm
, uint64_t shared
,
1905 ret
= bdrv_child_check_perm(c
, NULL
, perm
, shared
, NULL
, errp
);
1907 bdrv_child_abort_perm_update(c
);
1911 bdrv_child_set_perm(c
, perm
, shared
);
1916 #define DEFAULT_PERM_PASSTHROUGH (BLK_PERM_CONSISTENT_READ \
1918 | BLK_PERM_WRITE_UNCHANGED \
1920 #define DEFAULT_PERM_UNCHANGED (BLK_PERM_ALL & ~DEFAULT_PERM_PASSTHROUGH)
1922 void bdrv_filter_default_perms(BlockDriverState
*bs
, BdrvChild
*c
,
1923 const BdrvChildRole
*role
,
1924 BlockReopenQueue
*reopen_queue
,
1925 uint64_t perm
, uint64_t shared
,
1926 uint64_t *nperm
, uint64_t *nshared
)
1929 *nperm
= perm
& DEFAULT_PERM_PASSTHROUGH
;
1930 *nshared
= (shared
& DEFAULT_PERM_PASSTHROUGH
) | DEFAULT_PERM_UNCHANGED
;
1934 *nperm
= (perm
& DEFAULT_PERM_PASSTHROUGH
) |
1935 (c
->perm
& DEFAULT_PERM_UNCHANGED
);
1936 *nshared
= (shared
& DEFAULT_PERM_PASSTHROUGH
) |
1937 (c
->shared_perm
& DEFAULT_PERM_UNCHANGED
);
1940 void bdrv_format_default_perms(BlockDriverState
*bs
, BdrvChild
*c
,
1941 const BdrvChildRole
*role
,
1942 BlockReopenQueue
*reopen_queue
,
1943 uint64_t perm
, uint64_t shared
,
1944 uint64_t *nperm
, uint64_t *nshared
)
1946 bool backing
= (role
== &child_backing
);
1947 assert(role
== &child_backing
|| role
== &child_file
);
1950 int flags
= bdrv_reopen_get_flags(reopen_queue
, bs
);
1952 /* Apart from the modifications below, the same permissions are
1953 * forwarded and left alone as for filters */
1954 bdrv_filter_default_perms(bs
, c
, role
, reopen_queue
, perm
, shared
,
1957 /* Format drivers may touch metadata even if the guest doesn't write */
1958 if (bdrv_is_writable(bs
, reopen_queue
)) {
1959 perm
|= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
1962 /* bs->file always needs to be consistent because of the metadata. We
1963 * can never allow other users to resize or write to it. */
1964 if (!(flags
& BDRV_O_NO_IO
)) {
1965 perm
|= BLK_PERM_CONSISTENT_READ
;
1967 shared
&= ~(BLK_PERM_WRITE
| BLK_PERM_RESIZE
);
1969 /* We want consistent read from backing files if the parent needs it.
1970 * No other operations are performed on backing files. */
1971 perm
&= BLK_PERM_CONSISTENT_READ
;
1973 /* If the parent can deal with changing data, we're okay with a
1974 * writable and resizable backing file. */
1975 /* TODO Require !(perm & BLK_PERM_CONSISTENT_READ), too? */
1976 if (shared
& BLK_PERM_WRITE
) {
1977 shared
= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
1982 shared
|= BLK_PERM_CONSISTENT_READ
| BLK_PERM_GRAPH_MOD
|
1983 BLK_PERM_WRITE_UNCHANGED
;
1986 if (bs
->open_flags
& BDRV_O_INACTIVE
) {
1987 shared
|= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
1994 static void bdrv_replace_child_noperm(BdrvChild
*child
,
1995 BlockDriverState
*new_bs
)
1997 BlockDriverState
*old_bs
= child
->bs
;
2000 if (old_bs
&& new_bs
) {
2001 assert(bdrv_get_aio_context(old_bs
) == bdrv_get_aio_context(new_bs
));
2004 /* Detach first so that the recursive drain sections coming from @child
2005 * are already gone and we only end the drain sections that came from
2007 if (child
->role
->detach
) {
2008 child
->role
->detach(child
);
2010 if (old_bs
->quiesce_counter
&& child
->role
->drained_end
) {
2011 for (i
= 0; i
< old_bs
->quiesce_counter
; i
++) {
2012 child
->role
->drained_end(child
);
2015 QLIST_REMOVE(child
, next_parent
);
2021 QLIST_INSERT_HEAD(&new_bs
->parents
, child
, next_parent
);
2022 if (new_bs
->quiesce_counter
&& child
->role
->drained_begin
) {
2023 for (i
= 0; i
< new_bs
->quiesce_counter
; i
++) {
2024 child
->role
->drained_begin(child
);
2028 /* Attach only after starting new drained sections, so that recursive
2029 * drain sections coming from @child don't get an extra .drained_begin
2031 if (child
->role
->attach
) {
2032 child
->role
->attach(child
);
2038 * Updates @child to change its reference to point to @new_bs, including
2039 * checking and applying the necessary permisson updates both to the old node
2042 * NULL is passed as @new_bs for removing the reference before freeing @child.
2044 * If @new_bs is not NULL, bdrv_check_perm() must be called beforehand, as this
2045 * function uses bdrv_set_perm() to update the permissions according to the new
2046 * reference that @new_bs gets.
2048 static void bdrv_replace_child(BdrvChild
*child
, BlockDriverState
*new_bs
)
2050 BlockDriverState
*old_bs
= child
->bs
;
2051 uint64_t perm
, shared_perm
;
2053 bdrv_replace_child_noperm(child
, new_bs
);
2056 /* Update permissions for old node. This is guaranteed to succeed
2057 * because we're just taking a parent away, so we're loosening
2059 bdrv_get_cumulative_perm(old_bs
, &perm
, &shared_perm
);
2060 bdrv_check_perm(old_bs
, NULL
, perm
, shared_perm
, NULL
, &error_abort
);
2061 bdrv_set_perm(old_bs
, perm
, shared_perm
);
2065 bdrv_get_cumulative_perm(new_bs
, &perm
, &shared_perm
);
2066 bdrv_set_perm(new_bs
, perm
, shared_perm
);
2070 BdrvChild
*bdrv_root_attach_child(BlockDriverState
*child_bs
,
2071 const char *child_name
,
2072 const BdrvChildRole
*child_role
,
2073 uint64_t perm
, uint64_t shared_perm
,
2074 void *opaque
, Error
**errp
)
2079 ret
= bdrv_check_update_perm(child_bs
, NULL
, perm
, shared_perm
, NULL
, errp
);
2081 bdrv_abort_perm_update(child_bs
);
2085 child
= g_new(BdrvChild
, 1);
2086 *child
= (BdrvChild
) {
2088 .name
= g_strdup(child_name
),
2091 .shared_perm
= shared_perm
,
2095 /* This performs the matching bdrv_set_perm() for the above check. */
2096 bdrv_replace_child(child
, child_bs
);
2101 BdrvChild
*bdrv_attach_child(BlockDriverState
*parent_bs
,
2102 BlockDriverState
*child_bs
,
2103 const char *child_name
,
2104 const BdrvChildRole
*child_role
,
2108 uint64_t perm
, shared_perm
;
2110 bdrv_get_cumulative_perm(parent_bs
, &perm
, &shared_perm
);
2112 assert(parent_bs
->drv
);
2113 assert(bdrv_get_aio_context(parent_bs
) == bdrv_get_aio_context(child_bs
));
2114 bdrv_child_perm(parent_bs
, child_bs
, NULL
, child_role
, NULL
,
2115 perm
, shared_perm
, &perm
, &shared_perm
);
2117 child
= bdrv_root_attach_child(child_bs
, child_name
, child_role
,
2118 perm
, shared_perm
, parent_bs
, errp
);
2119 if (child
== NULL
) {
2123 QLIST_INSERT_HEAD(&parent_bs
->children
, child
, next
);
2127 static void bdrv_detach_child(BdrvChild
*child
)
2129 if (child
->next
.le_prev
) {
2130 QLIST_REMOVE(child
, next
);
2131 child
->next
.le_prev
= NULL
;
2134 bdrv_replace_child(child
, NULL
);
2136 g_free(child
->name
);
2140 void bdrv_root_unref_child(BdrvChild
*child
)
2142 BlockDriverState
*child_bs
;
2144 child_bs
= child
->bs
;
2145 bdrv_detach_child(child
);
2146 bdrv_unref(child_bs
);
2149 void bdrv_unref_child(BlockDriverState
*parent
, BdrvChild
*child
)
2151 if (child
== NULL
) {
2155 if (child
->bs
->inherits_from
== parent
) {
2158 /* Remove inherits_from only when the last reference between parent and
2159 * child->bs goes away. */
2160 QLIST_FOREACH(c
, &parent
->children
, next
) {
2161 if (c
!= child
&& c
->bs
== child
->bs
) {
2166 child
->bs
->inherits_from
= NULL
;
2170 bdrv_root_unref_child(child
);
2174 static void bdrv_parent_cb_change_media(BlockDriverState
*bs
, bool load
)
2177 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
2178 if (c
->role
->change_media
) {
2179 c
->role
->change_media(c
, load
);
2184 static void bdrv_parent_cb_resize(BlockDriverState
*bs
)
2187 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
2188 if (c
->role
->resize
) {
2195 * Sets the backing file link of a BDS. A new reference is created; callers
2196 * which don't need their own reference any more must call bdrv_unref().
2198 void bdrv_set_backing_hd(BlockDriverState
*bs
, BlockDriverState
*backing_hd
,
2202 bdrv_ref(backing_hd
);
2206 bdrv_unref_child(bs
, bs
->backing
);
2214 bs
->backing
= bdrv_attach_child(bs
, backing_hd
, "backing", &child_backing
,
2217 bdrv_unref(backing_hd
);
2220 bdrv_refresh_filename(bs
);
2223 bdrv_refresh_limits(bs
, NULL
);
2227 * Opens the backing file for a BlockDriverState if not yet open
2229 * bdref_key specifies the key for the image's BlockdevRef in the options QDict.
2230 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
2231 * itself, all options starting with "${bdref_key}." are considered part of the
2234 * TODO Can this be unified with bdrv_open_image()?
2236 int bdrv_open_backing_file(BlockDriverState
*bs
, QDict
*parent_options
,
2237 const char *bdref_key
, Error
**errp
)
2239 char *backing_filename
= g_malloc0(PATH_MAX
);
2240 char *bdref_key_dot
;
2241 const char *reference
= NULL
;
2243 BlockDriverState
*backing_hd
;
2245 QDict
*tmp_parent_options
= NULL
;
2246 Error
*local_err
= NULL
;
2248 if (bs
->backing
!= NULL
) {
2252 /* NULL means an empty set of options */
2253 if (parent_options
== NULL
) {
2254 tmp_parent_options
= qdict_new();
2255 parent_options
= tmp_parent_options
;
2258 bs
->open_flags
&= ~BDRV_O_NO_BACKING
;
2260 bdref_key_dot
= g_strdup_printf("%s.", bdref_key
);
2261 qdict_extract_subqdict(parent_options
, &options
, bdref_key_dot
);
2262 g_free(bdref_key_dot
);
2265 * Caution: while qdict_get_try_str() is fine, getting non-string
2266 * types would require more care. When @parent_options come from
2267 * -blockdev or blockdev_add, its members are typed according to
2268 * the QAPI schema, but when they come from -drive, they're all
2271 reference
= qdict_get_try_str(parent_options
, bdref_key
);
2272 if (reference
|| qdict_haskey(options
, "file.filename")) {
2273 backing_filename
[0] = '\0';
2274 } else if (bs
->backing_file
[0] == '\0' && qdict_size(options
) == 0) {
2278 bdrv_get_full_backing_filename(bs
, backing_filename
, PATH_MAX
,
2282 error_propagate(errp
, local_err
);
2288 if (!bs
->drv
|| !bs
->drv
->supports_backing
) {
2290 error_setg(errp
, "Driver doesn't support backing files");
2296 bs
->backing_format
[0] != '\0' && !qdict_haskey(options
, "driver")) {
2297 qdict_put_str(options
, "driver", bs
->backing_format
);
2300 backing_hd
= bdrv_open_inherit(*backing_filename
? backing_filename
: NULL
,
2301 reference
, options
, 0, bs
, &child_backing
,
2304 bs
->open_flags
|= BDRV_O_NO_BACKING
;
2305 error_prepend(errp
, "Could not open backing file: ");
2309 bdrv_set_aio_context(backing_hd
, bdrv_get_aio_context(bs
));
2311 /* Hook up the backing file link; drop our reference, bs owns the
2312 * backing_hd reference now */
2313 bdrv_set_backing_hd(bs
, backing_hd
, &local_err
);
2314 bdrv_unref(backing_hd
);
2316 error_propagate(errp
, local_err
);
2321 qdict_del(parent_options
, bdref_key
);
2324 g_free(backing_filename
);
2325 QDECREF(tmp_parent_options
);
2329 static BlockDriverState
*
2330 bdrv_open_child_bs(const char *filename
, QDict
*options
, const char *bdref_key
,
2331 BlockDriverState
*parent
, const BdrvChildRole
*child_role
,
2332 bool allow_none
, Error
**errp
)
2334 BlockDriverState
*bs
= NULL
;
2335 QDict
*image_options
;
2336 char *bdref_key_dot
;
2337 const char *reference
;
2339 assert(child_role
!= NULL
);
2341 bdref_key_dot
= g_strdup_printf("%s.", bdref_key
);
2342 qdict_extract_subqdict(options
, &image_options
, bdref_key_dot
);
2343 g_free(bdref_key_dot
);
2346 * Caution: while qdict_get_try_str() is fine, getting non-string
2347 * types would require more care. When @options come from
2348 * -blockdev or blockdev_add, its members are typed according to
2349 * the QAPI schema, but when they come from -drive, they're all
2352 reference
= qdict_get_try_str(options
, bdref_key
);
2353 if (!filename
&& !reference
&& !qdict_size(image_options
)) {
2355 error_setg(errp
, "A block device must be specified for \"%s\"",
2358 QDECREF(image_options
);
2362 bs
= bdrv_open_inherit(filename
, reference
, image_options
, 0,
2363 parent
, child_role
, errp
);
2369 qdict_del(options
, bdref_key
);
2374 * Opens a disk image whose options are given as BlockdevRef in another block
2377 * If allow_none is true, no image will be opened if filename is false and no
2378 * BlockdevRef is given. NULL will be returned, but errp remains unset.
2380 * bdrev_key specifies the key for the image's BlockdevRef in the options QDict.
2381 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
2382 * itself, all options starting with "${bdref_key}." are considered part of the
2385 * The BlockdevRef will be removed from the options QDict.
2387 BdrvChild
*bdrv_open_child(const char *filename
,
2388 QDict
*options
, const char *bdref_key
,
2389 BlockDriverState
*parent
,
2390 const BdrvChildRole
*child_role
,
2391 bool allow_none
, Error
**errp
)
2394 BlockDriverState
*bs
;
2396 bs
= bdrv_open_child_bs(filename
, options
, bdref_key
, parent
, child_role
,
2402 c
= bdrv_attach_child(parent
, bs
, bdref_key
, child_role
, errp
);
2411 /* TODO Future callers may need to specify parent/child_role in order for
2412 * option inheritance to work. Existing callers use it for the root node. */
2413 BlockDriverState
*bdrv_open_blockdev_ref(BlockdevRef
*ref
, Error
**errp
)
2415 BlockDriverState
*bs
= NULL
;
2416 Error
*local_err
= NULL
;
2417 QObject
*obj
= NULL
;
2418 QDict
*qdict
= NULL
;
2419 const char *reference
= NULL
;
2422 if (ref
->type
== QTYPE_QSTRING
) {
2423 reference
= ref
->u
.reference
;
2425 BlockdevOptions
*options
= &ref
->u
.definition
;
2426 assert(ref
->type
== QTYPE_QDICT
);
2428 v
= qobject_output_visitor_new(&obj
);
2429 visit_type_BlockdevOptions(v
, NULL
, &options
, &local_err
);
2431 error_propagate(errp
, local_err
);
2434 visit_complete(v
, &obj
);
2436 qdict
= qobject_to_qdict(obj
);
2437 qdict_flatten(qdict
);
2439 /* bdrv_open_inherit() defaults to the values in bdrv_flags (for
2440 * compatibility with other callers) rather than what we want as the
2441 * real defaults. Apply the defaults here instead. */
2442 qdict_set_default_str(qdict
, BDRV_OPT_CACHE_DIRECT
, "off");
2443 qdict_set_default_str(qdict
, BDRV_OPT_CACHE_NO_FLUSH
, "off");
2444 qdict_set_default_str(qdict
, BDRV_OPT_READ_ONLY
, "off");
2447 bs
= bdrv_open_inherit(NULL
, reference
, qdict
, 0, NULL
, NULL
, errp
);
2451 qobject_decref(obj
);
2456 static BlockDriverState
*bdrv_append_temp_snapshot(BlockDriverState
*bs
,
2458 QDict
*snapshot_options
,
2461 /* TODO: extra byte is a hack to ensure MAX_PATH space on Windows. */
2462 char *tmp_filename
= g_malloc0(PATH_MAX
+ 1);
2464 QemuOpts
*opts
= NULL
;
2465 BlockDriverState
*bs_snapshot
= NULL
;
2466 Error
*local_err
= NULL
;
2469 /* if snapshot, we create a temporary backing file and open it
2470 instead of opening 'filename' directly */
2472 /* Get the required size from the image */
2473 total_size
= bdrv_getlength(bs
);
2474 if (total_size
< 0) {
2475 error_setg_errno(errp
, -total_size
, "Could not get image size");
2479 /* Create the temporary image */
2480 ret
= get_tmp_filename(tmp_filename
, PATH_MAX
+ 1);
2482 error_setg_errno(errp
, -ret
, "Could not get temporary filename");
2486 opts
= qemu_opts_create(bdrv_qcow2
.create_opts
, NULL
, 0,
2488 qemu_opt_set_number(opts
, BLOCK_OPT_SIZE
, total_size
, &error_abort
);
2489 ret
= bdrv_create(&bdrv_qcow2
, tmp_filename
, opts
, errp
);
2490 qemu_opts_del(opts
);
2492 error_prepend(errp
, "Could not create temporary overlay '%s': ",
2497 /* Prepare options QDict for the temporary file */
2498 qdict_put_str(snapshot_options
, "file.driver", "file");
2499 qdict_put_str(snapshot_options
, "file.filename", tmp_filename
);
2500 qdict_put_str(snapshot_options
, "driver", "qcow2");
2502 bs_snapshot
= bdrv_open(NULL
, NULL
, snapshot_options
, flags
, errp
);
2503 snapshot_options
= NULL
;
2508 /* bdrv_append() consumes a strong reference to bs_snapshot
2509 * (i.e. it will call bdrv_unref() on it) even on error, so in
2510 * order to be able to return one, we have to increase
2511 * bs_snapshot's refcount here */
2512 bdrv_ref(bs_snapshot
);
2513 bdrv_append(bs_snapshot
, bs
, &local_err
);
2515 error_propagate(errp
, local_err
);
2521 QDECREF(snapshot_options
);
2522 g_free(tmp_filename
);
2527 * Opens a disk image (raw, qcow2, vmdk, ...)
2529 * options is a QDict of options to pass to the block drivers, or NULL for an
2530 * empty set of options. The reference to the QDict belongs to the block layer
2531 * after the call (even on failure), so if the caller intends to reuse the
2532 * dictionary, it needs to use QINCREF() before calling bdrv_open.
2534 * If *pbs is NULL, a new BDS will be created with a pointer to it stored there.
2535 * If it is not NULL, the referenced BDS will be reused.
2537 * The reference parameter may be used to specify an existing block device which
2538 * should be opened. If specified, neither options nor a filename may be given,
2539 * nor can an existing BDS be reused (that is, *pbs has to be NULL).
2541 static BlockDriverState
*bdrv_open_inherit(const char *filename
,
2542 const char *reference
,
2543 QDict
*options
, int flags
,
2544 BlockDriverState
*parent
,
2545 const BdrvChildRole
*child_role
,
2549 BlockBackend
*file
= NULL
;
2550 BlockDriverState
*bs
;
2551 BlockDriver
*drv
= NULL
;
2552 const char *drvname
;
2553 const char *backing
;
2554 Error
*local_err
= NULL
;
2555 QDict
*snapshot_options
= NULL
;
2556 int snapshot_flags
= 0;
2558 assert(!child_role
|| !flags
);
2559 assert(!child_role
== !parent
);
2562 bool options_non_empty
= options
? qdict_size(options
) : false;
2565 if (filename
|| options_non_empty
) {
2566 error_setg(errp
, "Cannot reference an existing block device with "
2567 "additional options or a new filename");
2571 bs
= bdrv_lookup_bs(reference
, reference
, errp
);
2582 /* NULL means an empty set of options */
2583 if (options
== NULL
) {
2584 options
= qdict_new();
2587 /* json: syntax counts as explicit options, as if in the QDict */
2588 parse_json_protocol(options
, &filename
, &local_err
);
2593 bs
->explicit_options
= qdict_clone_shallow(options
);
2596 bs
->inherits_from
= parent
;
2597 child_role
->inherit_options(&flags
, options
,
2598 parent
->open_flags
, parent
->options
);
2601 ret
= bdrv_fill_options(&options
, filename
, &flags
, &local_err
);
2607 * Set the BDRV_O_RDWR and BDRV_O_ALLOW_RDWR flags.
2608 * Caution: getting a boolean member of @options requires care.
2609 * When @options come from -blockdev or blockdev_add, members are
2610 * typed according to the QAPI schema, but when they come from
2611 * -drive, they're all QString.
2613 if (g_strcmp0(qdict_get_try_str(options
, BDRV_OPT_READ_ONLY
), "on") &&
2614 !qdict_get_try_bool(options
, BDRV_OPT_READ_ONLY
, false)) {
2615 flags
|= (BDRV_O_RDWR
| BDRV_O_ALLOW_RDWR
);
2617 flags
&= ~BDRV_O_RDWR
;
2620 if (flags
& BDRV_O_SNAPSHOT
) {
2621 snapshot_options
= qdict_new();
2622 bdrv_temp_snapshot_options(&snapshot_flags
, snapshot_options
,
2624 /* Let bdrv_backing_options() override "read-only" */
2625 qdict_del(options
, BDRV_OPT_READ_ONLY
);
2626 bdrv_backing_options(&flags
, options
, flags
, options
);
2629 bs
->open_flags
= flags
;
2630 bs
->options
= options
;
2631 options
= qdict_clone_shallow(options
);
2633 /* Find the right image format driver */
2634 /* See cautionary note on accessing @options above */
2635 drvname
= qdict_get_try_str(options
, "driver");
2637 drv
= bdrv_find_format(drvname
);
2639 error_setg(errp
, "Unknown driver: '%s'", drvname
);
2644 assert(drvname
|| !(flags
& BDRV_O_PROTOCOL
));
2646 /* See cautionary note on accessing @options above */
2647 backing
= qdict_get_try_str(options
, "backing");
2648 if (backing
&& *backing
== '\0') {
2649 flags
|= BDRV_O_NO_BACKING
;
2650 qdict_del(options
, "backing");
2653 /* Open image file without format layer. This BlockBackend is only used for
2654 * probing, the block drivers will do their own bdrv_open_child() for the
2655 * same BDS, which is why we put the node name back into options. */
2656 if ((flags
& BDRV_O_PROTOCOL
) == 0) {
2657 BlockDriverState
*file_bs
;
2659 file_bs
= bdrv_open_child_bs(filename
, options
, "file", bs
,
2660 &child_file
, true, &local_err
);
2664 if (file_bs
!= NULL
) {
2665 /* Not requesting BLK_PERM_CONSISTENT_READ because we're only
2666 * looking at the header to guess the image format. This works even
2667 * in cases where a guest would not see a consistent state. */
2668 file
= blk_new(0, BLK_PERM_ALL
);
2669 blk_insert_bs(file
, file_bs
, &local_err
);
2670 bdrv_unref(file_bs
);
2675 qdict_put_str(options
, "file", bdrv_get_node_name(file_bs
));
2679 /* Image format probing */
2682 ret
= find_image_format(file
, filename
, &drv
, &local_err
);
2687 * This option update would logically belong in bdrv_fill_options(),
2688 * but we first need to open bs->file for the probing to work, while
2689 * opening bs->file already requires the (mostly) final set of options
2690 * so that cache mode etc. can be inherited.
2692 * Adding the driver later is somewhat ugly, but it's not an option
2693 * that would ever be inherited, so it's correct. We just need to make
2694 * sure to update both bs->options (which has the full effective
2695 * options for bs) and options (which has file.* already removed).
2697 qdict_put_str(bs
->options
, "driver", drv
->format_name
);
2698 qdict_put_str(options
, "driver", drv
->format_name
);
2700 error_setg(errp
, "Must specify either driver or file");
2704 /* BDRV_O_PROTOCOL must be set iff a protocol BDS is about to be created */
2705 assert(!!(flags
& BDRV_O_PROTOCOL
) == !!drv
->bdrv_file_open
);
2706 /* file must be NULL if a protocol BDS is about to be created
2707 * (the inverse results in an error message from bdrv_open_common()) */
2708 assert(!(flags
& BDRV_O_PROTOCOL
) || !file
);
2710 /* Open the image */
2711 ret
= bdrv_open_common(bs
, file
, options
, &local_err
);
2721 /* If there is a backing file, use it */
2722 if ((flags
& BDRV_O_NO_BACKING
) == 0) {
2723 ret
= bdrv_open_backing_file(bs
, options
, "backing", &local_err
);
2725 goto close_and_fail
;
2729 bdrv_refresh_filename(bs
);
2731 /* Check if any unknown options were used */
2732 if (qdict_size(options
) != 0) {
2733 const QDictEntry
*entry
= qdict_first(options
);
2734 if (flags
& BDRV_O_PROTOCOL
) {
2735 error_setg(errp
, "Block protocol '%s' doesn't support the option "
2736 "'%s'", drv
->format_name
, entry
->key
);
2739 "Block format '%s' does not support the option '%s'",
2740 drv
->format_name
, entry
->key
);
2743 goto close_and_fail
;
2746 bdrv_parent_cb_change_media(bs
, true);
2750 /* For snapshot=on, create a temporary qcow2 overlay. bs points to the
2751 * temporary snapshot afterwards. */
2752 if (snapshot_flags
) {
2753 BlockDriverState
*snapshot_bs
;
2754 snapshot_bs
= bdrv_append_temp_snapshot(bs
, snapshot_flags
,
2755 snapshot_options
, &local_err
);
2756 snapshot_options
= NULL
;
2758 goto close_and_fail
;
2760 /* We are not going to return bs but the overlay on top of it
2761 * (snapshot_bs); thus, we have to drop the strong reference to bs
2762 * (which we obtained by calling bdrv_new()). bs will not be deleted,
2763 * though, because the overlay still has a reference to it. */
2772 QDECREF(snapshot_options
);
2773 QDECREF(bs
->explicit_options
);
2774 QDECREF(bs
->options
);
2777 bs
->explicit_options
= NULL
;
2779 error_propagate(errp
, local_err
);
2784 QDECREF(snapshot_options
);
2786 error_propagate(errp
, local_err
);
2790 BlockDriverState
*bdrv_open(const char *filename
, const char *reference
,
2791 QDict
*options
, int flags
, Error
**errp
)
2793 return bdrv_open_inherit(filename
, reference
, options
, flags
, NULL
,
2798 * Adds a BlockDriverState to a simple queue for an atomic, transactional
2799 * reopen of multiple devices.
2801 * bs_queue can either be an existing BlockReopenQueue that has had QSIMPLE_INIT
2802 * already performed, or alternatively may be NULL a new BlockReopenQueue will
2803 * be created and initialized. This newly created BlockReopenQueue should be
2804 * passed back in for subsequent calls that are intended to be of the same
2807 * bs is the BlockDriverState to add to the reopen queue.
2809 * options contains the changed options for the associated bs
2810 * (the BlockReopenQueue takes ownership)
2812 * flags contains the open flags for the associated bs
2814 * returns a pointer to bs_queue, which is either the newly allocated
2815 * bs_queue, or the existing bs_queue being used.
2817 * bs must be drained between bdrv_reopen_queue() and bdrv_reopen_multiple().
2819 static BlockReopenQueue
*bdrv_reopen_queue_child(BlockReopenQueue
*bs_queue
,
2820 BlockDriverState
*bs
,
2823 const BdrvChildRole
*role
,
2824 QDict
*parent_options
,
2829 BlockReopenQueueEntry
*bs_entry
;
2831 QDict
*old_options
, *explicit_options
;
2833 /* Make sure that the caller remembered to use a drained section. This is
2834 * important to avoid graph changes between the recursive queuing here and
2835 * bdrv_reopen_multiple(). */
2836 assert(bs
->quiesce_counter
> 0);
2838 if (bs_queue
== NULL
) {
2839 bs_queue
= g_new0(BlockReopenQueue
, 1);
2840 QSIMPLEQ_INIT(bs_queue
);
2844 options
= qdict_new();
2847 /* Check if this BlockDriverState is already in the queue */
2848 QSIMPLEQ_FOREACH(bs_entry
, bs_queue
, entry
) {
2849 if (bs
== bs_entry
->state
.bs
) {
2855 * Precedence of options:
2856 * 1. Explicitly passed in options (highest)
2857 * 2. Set in flags (only for top level)
2858 * 3. Retained from explicitly set options of bs
2859 * 4. Inherited from parent node
2860 * 5. Retained from effective options of bs
2863 if (!parent_options
) {
2865 * Any setting represented by flags is always updated. If the
2866 * corresponding QDict option is set, it takes precedence. Otherwise
2867 * the flag is translated into a QDict option. The old setting of bs is
2870 update_options_from_flags(options
, flags
);
2873 /* Old explicitly set values (don't overwrite by inherited value) */
2875 old_options
= qdict_clone_shallow(bs_entry
->state
.explicit_options
);
2877 old_options
= qdict_clone_shallow(bs
->explicit_options
);
2879 bdrv_join_options(bs
, options
, old_options
);
2880 QDECREF(old_options
);
2882 explicit_options
= qdict_clone_shallow(options
);
2884 /* Inherit from parent node */
2885 if (parent_options
) {
2887 role
->inherit_options(&flags
, options
, parent_flags
, parent_options
);
2890 /* Old values are used for options that aren't set yet */
2891 old_options
= qdict_clone_shallow(bs
->options
);
2892 bdrv_join_options(bs
, options
, old_options
);
2893 QDECREF(old_options
);
2895 /* bdrv_open_inherit() sets and clears some additional flags internally */
2896 flags
&= ~BDRV_O_PROTOCOL
;
2897 if (flags
& BDRV_O_RDWR
) {
2898 flags
|= BDRV_O_ALLOW_RDWR
;
2902 bs_entry
= g_new0(BlockReopenQueueEntry
, 1);
2903 QSIMPLEQ_INSERT_TAIL(bs_queue
, bs_entry
, entry
);
2905 QDECREF(bs_entry
->state
.options
);
2906 QDECREF(bs_entry
->state
.explicit_options
);
2909 bs_entry
->state
.bs
= bs
;
2910 bs_entry
->state
.options
= options
;
2911 bs_entry
->state
.explicit_options
= explicit_options
;
2912 bs_entry
->state
.flags
= flags
;
2914 /* This needs to be overwritten in bdrv_reopen_prepare() */
2915 bs_entry
->state
.perm
= UINT64_MAX
;
2916 bs_entry
->state
.shared_perm
= 0;
2918 QLIST_FOREACH(child
, &bs
->children
, next
) {
2919 QDict
*new_child_options
;
2920 char *child_key_dot
;
2922 /* reopen can only change the options of block devices that were
2923 * implicitly created and inherited options. For other (referenced)
2924 * block devices, a syntax like "backing.foo" results in an error. */
2925 if (child
->bs
->inherits_from
!= bs
) {
2929 child_key_dot
= g_strdup_printf("%s.", child
->name
);
2930 qdict_extract_subqdict(options
, &new_child_options
, child_key_dot
);
2931 g_free(child_key_dot
);
2933 bdrv_reopen_queue_child(bs_queue
, child
->bs
, new_child_options
, 0,
2934 child
->role
, options
, flags
);
2940 BlockReopenQueue
*bdrv_reopen_queue(BlockReopenQueue
*bs_queue
,
2941 BlockDriverState
*bs
,
2942 QDict
*options
, int flags
)
2944 return bdrv_reopen_queue_child(bs_queue
, bs
, options
, flags
,
2949 * Reopen multiple BlockDriverStates atomically & transactionally.
2951 * The queue passed in (bs_queue) must have been built up previous
2952 * via bdrv_reopen_queue().
2954 * Reopens all BDS specified in the queue, with the appropriate
2955 * flags. All devices are prepared for reopen, and failure of any
2956 * device will cause all device changes to be abandonded, and intermediate
2959 * If all devices prepare successfully, then the changes are committed
2962 * All affected nodes must be drained between bdrv_reopen_queue() and
2963 * bdrv_reopen_multiple().
2965 int bdrv_reopen_multiple(AioContext
*ctx
, BlockReopenQueue
*bs_queue
, Error
**errp
)
2968 BlockReopenQueueEntry
*bs_entry
, *next
;
2969 Error
*local_err
= NULL
;
2971 assert(bs_queue
!= NULL
);
2973 QSIMPLEQ_FOREACH(bs_entry
, bs_queue
, entry
) {
2974 assert(bs_entry
->state
.bs
->quiesce_counter
> 0);
2975 if (bdrv_reopen_prepare(&bs_entry
->state
, bs_queue
, &local_err
)) {
2976 error_propagate(errp
, local_err
);
2979 bs_entry
->prepared
= true;
2982 /* If we reach this point, we have success and just need to apply the
2985 QSIMPLEQ_FOREACH(bs_entry
, bs_queue
, entry
) {
2986 bdrv_reopen_commit(&bs_entry
->state
);
2992 QSIMPLEQ_FOREACH_SAFE(bs_entry
, bs_queue
, entry
, next
) {
2993 if (ret
&& bs_entry
->prepared
) {
2994 bdrv_reopen_abort(&bs_entry
->state
);
2996 QDECREF(bs_entry
->state
.explicit_options
);
2998 QDECREF(bs_entry
->state
.options
);
3007 /* Reopen a single BlockDriverState with the specified flags. */
3008 int bdrv_reopen(BlockDriverState
*bs
, int bdrv_flags
, Error
**errp
)
3011 Error
*local_err
= NULL
;
3012 BlockReopenQueue
*queue
;
3014 bdrv_subtree_drained_begin(bs
);
3016 queue
= bdrv_reopen_queue(NULL
, bs
, NULL
, bdrv_flags
);
3017 ret
= bdrv_reopen_multiple(bdrv_get_aio_context(bs
), queue
, &local_err
);
3018 if (local_err
!= NULL
) {
3019 error_propagate(errp
, local_err
);
3022 bdrv_subtree_drained_end(bs
);
3027 static BlockReopenQueueEntry
*find_parent_in_reopen_queue(BlockReopenQueue
*q
,
3030 BlockReopenQueueEntry
*entry
;
3032 QSIMPLEQ_FOREACH(entry
, q
, entry
) {
3033 BlockDriverState
*bs
= entry
->state
.bs
;
3036 QLIST_FOREACH(child
, &bs
->children
, next
) {
3046 static void bdrv_reopen_perm(BlockReopenQueue
*q
, BlockDriverState
*bs
,
3047 uint64_t *perm
, uint64_t *shared
)
3050 BlockReopenQueueEntry
*parent
;
3051 uint64_t cumulative_perms
= 0;
3052 uint64_t cumulative_shared_perms
= BLK_PERM_ALL
;
3054 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
3055 parent
= find_parent_in_reopen_queue(q
, c
);
3057 cumulative_perms
|= c
->perm
;
3058 cumulative_shared_perms
&= c
->shared_perm
;
3060 uint64_t nperm
, nshared
;
3062 bdrv_child_perm(parent
->state
.bs
, bs
, c
, c
->role
, q
,
3063 parent
->state
.perm
, parent
->state
.shared_perm
,
3066 cumulative_perms
|= nperm
;
3067 cumulative_shared_perms
&= nshared
;
3070 *perm
= cumulative_perms
;
3071 *shared
= cumulative_shared_perms
;
3075 * Prepares a BlockDriverState for reopen. All changes are staged in the
3076 * 'opaque' field of the BDRVReopenState, which is used and allocated by
3077 * the block driver layer .bdrv_reopen_prepare()
3079 * bs is the BlockDriverState to reopen
3080 * flags are the new open flags
3081 * queue is the reopen queue
3083 * Returns 0 on success, non-zero on error. On error errp will be set
3086 * On failure, bdrv_reopen_abort() will be called to clean up any data.
3087 * It is the responsibility of the caller to then call the abort() or
3088 * commit() for any other BDS that have been left in a prepare() state
3091 int bdrv_reopen_prepare(BDRVReopenState
*reopen_state
, BlockReopenQueue
*queue
,
3095 Error
*local_err
= NULL
;
3101 assert(reopen_state
!= NULL
);
3102 assert(reopen_state
->bs
->drv
!= NULL
);
3103 drv
= reopen_state
->bs
->drv
;
3105 /* Process generic block layer options */
3106 opts
= qemu_opts_create(&bdrv_runtime_opts
, NULL
, 0, &error_abort
);
3107 qemu_opts_absorb_qdict(opts
, reopen_state
->options
, &local_err
);
3109 error_propagate(errp
, local_err
);
3114 update_flags_from_options(&reopen_state
->flags
, opts
);
3116 /* node-name and driver must be unchanged. Put them back into the QDict, so
3117 * that they are checked at the end of this function. */
3118 value
= qemu_opt_get(opts
, "node-name");
3120 qdict_put_str(reopen_state
->options
, "node-name", value
);
3123 value
= qemu_opt_get(opts
, "driver");
3125 qdict_put_str(reopen_state
->options
, "driver", value
);
3128 /* If we are to stay read-only, do not allow permission change
3129 * to r/w. Attempting to set to r/w may fail if either BDRV_O_ALLOW_RDWR is
3130 * not set, or if the BDS still has copy_on_read enabled */
3131 read_only
= !(reopen_state
->flags
& BDRV_O_RDWR
);
3132 ret
= bdrv_can_set_read_only(reopen_state
->bs
, read_only
, true, &local_err
);
3134 error_propagate(errp
, local_err
);
3138 /* Calculate required permissions after reopening */
3139 bdrv_reopen_perm(queue
, reopen_state
->bs
,
3140 &reopen_state
->perm
, &reopen_state
->shared_perm
);
3142 ret
= bdrv_flush(reopen_state
->bs
);
3144 error_setg_errno(errp
, -ret
, "Error flushing drive");
3148 if (drv
->bdrv_reopen_prepare
) {
3149 ret
= drv
->bdrv_reopen_prepare(reopen_state
, queue
, &local_err
);
3151 if (local_err
!= NULL
) {
3152 error_propagate(errp
, local_err
);
3154 error_setg(errp
, "failed while preparing to reopen image '%s'",
3155 reopen_state
->bs
->filename
);
3160 /* It is currently mandatory to have a bdrv_reopen_prepare()
3161 * handler for each supported drv. */
3162 error_setg(errp
, "Block format '%s' used by node '%s' "
3163 "does not support reopening files", drv
->format_name
,
3164 bdrv_get_device_or_node_name(reopen_state
->bs
));
3169 /* Options that are not handled are only okay if they are unchanged
3170 * compared to the old state. It is expected that some options are only
3171 * used for the initial open, but not reopen (e.g. filename) */
3172 if (qdict_size(reopen_state
->options
)) {
3173 const QDictEntry
*entry
= qdict_first(reopen_state
->options
);
3176 QObject
*new = entry
->value
;
3177 QObject
*old
= qdict_get(reopen_state
->bs
->options
, entry
->key
);
3180 * TODO: When using -drive to specify blockdev options, all values
3181 * will be strings; however, when using -blockdev, blockdev-add or
3182 * filenames using the json:{} pseudo-protocol, they will be
3184 * In contrast, reopening options are (currently) always strings
3185 * (because you can only specify them through qemu-io; all other
3186 * callers do not specify any options).
3187 * Therefore, when using anything other than -drive to create a BDS,
3188 * this cannot detect non-string options as unchanged, because
3189 * qobject_is_equal() always returns false for objects of different
3190 * type. In the future, this should be remedied by correctly typing
3191 * all options. For now, this is not too big of an issue because
3192 * the user can simply omit options which cannot be changed anyway,
3193 * so they will stay unchanged.
3195 if (!qobject_is_equal(new, old
)) {
3196 error_setg(errp
, "Cannot change the option '%s'", entry
->key
);
3200 } while ((entry
= qdict_next(reopen_state
->options
, entry
)));
3203 ret
= bdrv_check_perm(reopen_state
->bs
, queue
, reopen_state
->perm
,
3204 reopen_state
->shared_perm
, NULL
, errp
);
3212 qemu_opts_del(opts
);
3217 * Takes the staged changes for the reopen from bdrv_reopen_prepare(), and
3218 * makes them final by swapping the staging BlockDriverState contents into
3219 * the active BlockDriverState contents.
3221 void bdrv_reopen_commit(BDRVReopenState
*reopen_state
)
3224 BlockDriverState
*bs
;
3225 bool old_can_write
, new_can_write
;
3227 assert(reopen_state
!= NULL
);
3228 bs
= reopen_state
->bs
;
3230 assert(drv
!= NULL
);
3233 !bdrv_is_read_only(bs
) && !(bdrv_get_flags(bs
) & BDRV_O_INACTIVE
);
3235 /* If there are any driver level actions to take */
3236 if (drv
->bdrv_reopen_commit
) {
3237 drv
->bdrv_reopen_commit(reopen_state
);
3240 /* set BDS specific flags now */
3241 QDECREF(bs
->explicit_options
);
3243 bs
->explicit_options
= reopen_state
->explicit_options
;
3244 bs
->open_flags
= reopen_state
->flags
;
3245 bs
->read_only
= !(reopen_state
->flags
& BDRV_O_RDWR
);
3247 bdrv_refresh_limits(bs
, NULL
);
3249 bdrv_set_perm(reopen_state
->bs
, reopen_state
->perm
,
3250 reopen_state
->shared_perm
);
3253 !bdrv_is_read_only(bs
) && !(bdrv_get_flags(bs
) & BDRV_O_INACTIVE
);
3254 if (!old_can_write
&& new_can_write
&& drv
->bdrv_reopen_bitmaps_rw
) {
3255 Error
*local_err
= NULL
;
3256 if (drv
->bdrv_reopen_bitmaps_rw(bs
, &local_err
) < 0) {
3257 /* This is not fatal, bitmaps just left read-only, so all following
3258 * writes will fail. User can remove read-only bitmaps to unblock
3261 error_reportf_err(local_err
,
3262 "%s: Failed to make dirty bitmaps writable: ",
3263 bdrv_get_node_name(bs
));
3269 * Abort the reopen, and delete and free the staged changes in
3272 void bdrv_reopen_abort(BDRVReopenState
*reopen_state
)
3276 assert(reopen_state
!= NULL
);
3277 drv
= reopen_state
->bs
->drv
;
3278 assert(drv
!= NULL
);
3280 if (drv
->bdrv_reopen_abort
) {
3281 drv
->bdrv_reopen_abort(reopen_state
);
3284 QDECREF(reopen_state
->explicit_options
);
3286 bdrv_abort_perm_update(reopen_state
->bs
);
3290 static void bdrv_close(BlockDriverState
*bs
)
3292 BdrvAioNotifier
*ban
, *ban_next
;
3293 BdrvChild
*child
, *next
;
3296 assert(!bs
->refcnt
);
3298 bdrv_drained_begin(bs
); /* complete I/O */
3300 bdrv_drain(bs
); /* in case flush left pending I/O */
3303 bs
->drv
->bdrv_close(bs
);
3307 bdrv_set_backing_hd(bs
, NULL
, &error_abort
);
3309 if (bs
->file
!= NULL
) {
3310 bdrv_unref_child(bs
, bs
->file
);
3314 QLIST_FOREACH_SAFE(child
, &bs
->children
, next
, next
) {
3315 /* TODO Remove bdrv_unref() from drivers' close function and use
3316 * bdrv_unref_child() here */
3317 if (child
->bs
->inherits_from
== bs
) {
3318 child
->bs
->inherits_from
= NULL
;
3320 bdrv_detach_child(child
);
3325 atomic_set(&bs
->copy_on_read
, 0);
3326 bs
->backing_file
[0] = '\0';
3327 bs
->backing_format
[0] = '\0';
3328 bs
->total_sectors
= 0;
3329 bs
->encrypted
= false;
3331 QDECREF(bs
->options
);
3332 QDECREF(bs
->explicit_options
);
3334 bs
->explicit_options
= NULL
;
3335 QDECREF(bs
->full_open_options
);
3336 bs
->full_open_options
= NULL
;
3338 bdrv_release_named_dirty_bitmaps(bs
);
3339 assert(QLIST_EMPTY(&bs
->dirty_bitmaps
));
3341 QLIST_FOREACH_SAFE(ban
, &bs
->aio_notifiers
, list
, ban_next
) {
3344 QLIST_INIT(&bs
->aio_notifiers
);
3345 bdrv_drained_end(bs
);
3348 void bdrv_close_all(void)
3350 block_job_cancel_sync_all();
3351 nbd_export_close_all();
3353 /* Drop references from requests still in flight, such as canceled block
3354 * jobs whose AIO context has not been polled yet */
3357 blk_remove_all_bs();
3358 blockdev_close_all_bdrv_states();
3360 assert(QTAILQ_EMPTY(&all_bdrv_states
));
3363 static bool should_update_child(BdrvChild
*c
, BlockDriverState
*to
)
3367 if (c
->role
->stay_at_node
) {
3371 if (c
->role
== &child_backing
) {
3372 /* If @from is a backing file of @to, ignore the child to avoid
3373 * creating a loop. We only want to change the pointer of other
3375 QLIST_FOREACH(to_c
, &to
->children
, next
) {
3388 void bdrv_replace_node(BlockDriverState
*from
, BlockDriverState
*to
,
3391 BdrvChild
*c
, *next
;
3392 GSList
*list
= NULL
, *p
;
3393 uint64_t old_perm
, old_shared
;
3394 uint64_t perm
= 0, shared
= BLK_PERM_ALL
;
3397 assert(!atomic_read(&from
->in_flight
));
3398 assert(!atomic_read(&to
->in_flight
));
3400 /* Make sure that @from doesn't go away until we have successfully attached
3401 * all of its parents to @to. */
3404 /* Put all parents into @list and calculate their cumulative permissions */
3405 QLIST_FOREACH_SAFE(c
, &from
->parents
, next_parent
, next
) {
3406 if (!should_update_child(c
, to
)) {
3409 list
= g_slist_prepend(list
, c
);
3411 shared
&= c
->shared_perm
;
3414 /* Check whether the required permissions can be granted on @to, ignoring
3415 * all BdrvChild in @list so that they can't block themselves. */
3416 ret
= bdrv_check_update_perm(to
, NULL
, perm
, shared
, list
, errp
);
3418 bdrv_abort_perm_update(to
);
3422 /* Now actually perform the change. We performed the permission check for
3423 * all elements of @list at once, so set the permissions all at once at the
3425 for (p
= list
; p
!= NULL
; p
= p
->next
) {
3429 bdrv_replace_child_noperm(c
, to
);
3433 bdrv_get_cumulative_perm(to
, &old_perm
, &old_shared
);
3434 bdrv_set_perm(to
, old_perm
| perm
, old_shared
| shared
);
3442 * Add new bs contents at the top of an image chain while the chain is
3443 * live, while keeping required fields on the top layer.
3445 * This will modify the BlockDriverState fields, and swap contents
3446 * between bs_new and bs_top. Both bs_new and bs_top are modified.
3448 * bs_new must not be attached to a BlockBackend.
3450 * This function does not create any image files.
3452 * bdrv_append() takes ownership of a bs_new reference and unrefs it because
3453 * that's what the callers commonly need. bs_new will be referenced by the old
3454 * parents of bs_top after bdrv_append() returns. If the caller needs to keep a
3455 * reference of its own, it must call bdrv_ref().
3457 void bdrv_append(BlockDriverState
*bs_new
, BlockDriverState
*bs_top
,
3460 Error
*local_err
= NULL
;
3462 bdrv_set_backing_hd(bs_new
, bs_top
, &local_err
);
3464 error_propagate(errp
, local_err
);
3468 bdrv_replace_node(bs_top
, bs_new
, &local_err
);
3470 error_propagate(errp
, local_err
);
3471 bdrv_set_backing_hd(bs_new
, NULL
, &error_abort
);
3475 /* bs_new is now referenced by its new parents, we don't need the
3476 * additional reference any more. */
3481 static void bdrv_delete(BlockDriverState
*bs
)
3484 assert(bdrv_op_blocker_is_empty(bs
));
3485 assert(!bs
->refcnt
);
3489 /* remove from list, if necessary */
3490 if (bs
->node_name
[0] != '\0') {
3491 QTAILQ_REMOVE(&graph_bdrv_states
, bs
, node_list
);
3493 QTAILQ_REMOVE(&all_bdrv_states
, bs
, bs_list
);
3499 * Run consistency checks on an image
3501 * Returns 0 if the check could be completed (it doesn't mean that the image is
3502 * free of errors) or -errno when an internal error occurred. The results of the
3503 * check are stored in res.
3505 static int coroutine_fn
bdrv_co_check(BlockDriverState
*bs
,
3506 BdrvCheckResult
*res
, BdrvCheckMode fix
)
3508 if (bs
->drv
== NULL
) {
3511 if (bs
->drv
->bdrv_co_check
== NULL
) {
3515 memset(res
, 0, sizeof(*res
));
3516 return bs
->drv
->bdrv_co_check(bs
, res
, fix
);
3519 typedef struct CheckCo
{
3520 BlockDriverState
*bs
;
3521 BdrvCheckResult
*res
;
3526 static void bdrv_check_co_entry(void *opaque
)
3528 CheckCo
*cco
= opaque
;
3529 cco
->ret
= bdrv_co_check(cco
->bs
, cco
->res
, cco
->fix
);
3532 int bdrv_check(BlockDriverState
*bs
,
3533 BdrvCheckResult
*res
, BdrvCheckMode fix
)
3539 .ret
= -EINPROGRESS
,
3543 if (qemu_in_coroutine()) {
3544 /* Fast-path if already in coroutine context */
3545 bdrv_check_co_entry(&cco
);
3547 co
= qemu_coroutine_create(bdrv_check_co_entry
, &cco
);
3548 qemu_coroutine_enter(co
);
3549 BDRV_POLL_WHILE(bs
, cco
.ret
== -EINPROGRESS
);
3558 * -EINVAL - backing format specified, but no file
3559 * -ENOSPC - can't update the backing file because no space is left in the
3561 * -ENOTSUP - format driver doesn't support changing the backing file
3563 int bdrv_change_backing_file(BlockDriverState
*bs
,
3564 const char *backing_file
, const char *backing_fmt
)
3566 BlockDriver
*drv
= bs
->drv
;
3573 /* Backing file format doesn't make sense without a backing file */
3574 if (backing_fmt
&& !backing_file
) {
3578 if (drv
->bdrv_change_backing_file
!= NULL
) {
3579 ret
= drv
->bdrv_change_backing_file(bs
, backing_file
, backing_fmt
);
3585 pstrcpy(bs
->backing_file
, sizeof(bs
->backing_file
), backing_file
?: "");
3586 pstrcpy(bs
->backing_format
, sizeof(bs
->backing_format
), backing_fmt
?: "");
3592 * Finds the image layer in the chain that has 'bs' as its backing file.
3594 * active is the current topmost image.
3596 * Returns NULL if bs is not found in active's image chain,
3597 * or if active == bs.
3599 * Returns the bottommost base image if bs == NULL.
3601 BlockDriverState
*bdrv_find_overlay(BlockDriverState
*active
,
3602 BlockDriverState
*bs
)
3604 while (active
&& bs
!= backing_bs(active
)) {
3605 active
= backing_bs(active
);
3611 /* Given a BDS, searches for the base layer. */
3612 BlockDriverState
*bdrv_find_base(BlockDriverState
*bs
)
3614 return bdrv_find_overlay(bs
, NULL
);
3618 * Drops images above 'base' up to and including 'top', and sets the image
3619 * above 'top' to have base as its backing file.
3621 * Requires that the overlay to 'top' is opened r/w, so that the backing file
3622 * information in 'bs' can be properly updated.
3624 * E.g., this will convert the following chain:
3625 * bottom <- base <- intermediate <- top <- active
3629 * bottom <- base <- active
3631 * It is allowed for bottom==base, in which case it converts:
3633 * base <- intermediate <- top <- active
3639 * If backing_file_str is non-NULL, it will be used when modifying top's
3640 * overlay image metadata.
3643 * if active == top, that is considered an error
3646 int bdrv_drop_intermediate(BlockDriverState
*top
, BlockDriverState
*base
,
3647 const char *backing_file_str
)
3649 BdrvChild
*c
, *next
;
3650 Error
*local_err
= NULL
;
3655 if (!top
->drv
|| !base
->drv
) {
3659 /* Make sure that base is in the backing chain of top */
3660 if (!bdrv_chain_contains(top
, base
)) {
3664 /* success - we can delete the intermediate states, and link top->base */
3665 /* TODO Check graph modification op blockers (BLK_PERM_GRAPH_MOD) once
3666 * we've figured out how they should work. */
3667 backing_file_str
= backing_file_str
? backing_file_str
: base
->filename
;
3669 QLIST_FOREACH_SAFE(c
, &top
->parents
, next_parent
, next
) {
3670 /* Check whether we are allowed to switch c from top to base */
3671 GSList
*ignore_children
= g_slist_prepend(NULL
, c
);
3672 bdrv_check_update_perm(base
, NULL
, c
->perm
, c
->shared_perm
,
3673 ignore_children
, &local_err
);
3676 error_report_err(local_err
);
3679 g_slist_free(ignore_children
);
3681 /* If so, update the backing file path in the image file */
3682 if (c
->role
->update_filename
) {
3683 ret
= c
->role
->update_filename(c
, base
, backing_file_str
,
3686 bdrv_abort_perm_update(base
);
3687 error_report_err(local_err
);
3692 /* Do the actual switch in the in-memory graph.
3693 * Completes bdrv_check_update_perm() transaction internally. */
3695 bdrv_replace_child(c
, base
);
3706 * Truncate file to 'offset' bytes (needed only for file protocols)
3708 int bdrv_truncate(BdrvChild
*child
, int64_t offset
, PreallocMode prealloc
,
3711 BlockDriverState
*bs
= child
->bs
;
3712 BlockDriver
*drv
= bs
->drv
;
3715 assert(child
->perm
& BLK_PERM_RESIZE
);
3717 /* if bs->drv == NULL, bs is closed, so there's nothing to do here */
3719 error_setg(errp
, "No medium inserted");
3723 error_setg(errp
, "Image size cannot be negative");
3727 if (!drv
->bdrv_truncate
) {
3728 if (bs
->file
&& drv
->is_filter
) {
3729 return bdrv_truncate(bs
->file
, offset
, prealloc
, errp
);
3731 error_setg(errp
, "Image format driver does not support resize");
3734 if (bs
->read_only
) {
3735 error_setg(errp
, "Image is read-only");
3739 assert(!(bs
->open_flags
& BDRV_O_INACTIVE
));
3741 ret
= drv
->bdrv_truncate(bs
, offset
, prealloc
, errp
);
3745 ret
= refresh_total_sectors(bs
, offset
>> BDRV_SECTOR_BITS
);
3747 error_setg_errno(errp
, -ret
, "Could not refresh total sector count");
3749 offset
= bs
->total_sectors
* BDRV_SECTOR_SIZE
;
3751 bdrv_dirty_bitmap_truncate(bs
, offset
);
3752 bdrv_parent_cb_resize(bs
);
3753 atomic_inc(&bs
->write_gen
);
3758 * Length of a allocated file in bytes. Sparse files are counted by actual
3759 * allocated space. Return < 0 if error or unknown.
3761 int64_t bdrv_get_allocated_file_size(BlockDriverState
*bs
)
3763 BlockDriver
*drv
= bs
->drv
;
3767 if (drv
->bdrv_get_allocated_file_size
) {
3768 return drv
->bdrv_get_allocated_file_size(bs
);
3771 return bdrv_get_allocated_file_size(bs
->file
->bs
);
3778 * @drv: Format driver
3779 * @opts: Creation options for new image
3780 * @in_bs: Existing image containing data for new image (may be NULL)
3781 * @errp: Error object
3782 * Returns: A #BlockMeasureInfo (free using qapi_free_BlockMeasureInfo())
3785 * Calculate file size required to create a new image.
3787 * If @in_bs is given then space for allocated clusters and zero clusters
3788 * from that image are included in the calculation. If @opts contains a
3789 * backing file that is shared by @in_bs then backing clusters may be omitted
3790 * from the calculation.
3792 * If @in_bs is NULL then the calculation includes no allocated clusters
3793 * unless a preallocation option is given in @opts.
3795 * Note that @in_bs may use a different BlockDriver from @drv.
3797 * If an error occurs the @errp pointer is set.
3799 BlockMeasureInfo
*bdrv_measure(BlockDriver
*drv
, QemuOpts
*opts
,
3800 BlockDriverState
*in_bs
, Error
**errp
)
3802 if (!drv
->bdrv_measure
) {
3803 error_setg(errp
, "Block driver '%s' does not support size measurement",
3808 return drv
->bdrv_measure(opts
, in_bs
, errp
);
3812 * Return number of sectors on success, -errno on error.
3814 int64_t bdrv_nb_sectors(BlockDriverState
*bs
)
3816 BlockDriver
*drv
= bs
->drv
;
3821 if (drv
->has_variable_length
) {
3822 int ret
= refresh_total_sectors(bs
, bs
->total_sectors
);
3827 return bs
->total_sectors
;
3831 * Return length in bytes on success, -errno on error.
3832 * The length is always a multiple of BDRV_SECTOR_SIZE.
3834 int64_t bdrv_getlength(BlockDriverState
*bs
)
3836 int64_t ret
= bdrv_nb_sectors(bs
);
3838 ret
= ret
> INT64_MAX
/ BDRV_SECTOR_SIZE
? -EFBIG
: ret
;
3839 return ret
< 0 ? ret
: ret
* BDRV_SECTOR_SIZE
;
3842 /* return 0 as number of sectors if no device present or error */
3843 void bdrv_get_geometry(BlockDriverState
*bs
, uint64_t *nb_sectors_ptr
)
3845 int64_t nb_sectors
= bdrv_nb_sectors(bs
);
3847 *nb_sectors_ptr
= nb_sectors
< 0 ? 0 : nb_sectors
;
3850 bool bdrv_is_sg(BlockDriverState
*bs
)
3855 bool bdrv_is_encrypted(BlockDriverState
*bs
)
3857 if (bs
->backing
&& bs
->backing
->bs
->encrypted
) {
3860 return bs
->encrypted
;
3863 const char *bdrv_get_format_name(BlockDriverState
*bs
)
3865 return bs
->drv
? bs
->drv
->format_name
: NULL
;
3868 static int qsort_strcmp(const void *a
, const void *b
)
3870 return strcmp(*(char *const *)a
, *(char *const *)b
);
3873 void bdrv_iterate_format(void (*it
)(void *opaque
, const char *name
),
3879 const char **formats
= NULL
;
3881 QLIST_FOREACH(drv
, &bdrv_drivers
, list
) {
3882 if (drv
->format_name
) {
3885 while (formats
&& i
&& !found
) {
3886 found
= !strcmp(formats
[--i
], drv
->format_name
);
3890 formats
= g_renew(const char *, formats
, count
+ 1);
3891 formats
[count
++] = drv
->format_name
;
3896 for (i
= 0; i
< (int)ARRAY_SIZE(block_driver_modules
); i
++) {
3897 const char *format_name
= block_driver_modules
[i
].format_name
;
3903 while (formats
&& j
&& !found
) {
3904 found
= !strcmp(formats
[--j
], format_name
);
3908 formats
= g_renew(const char *, formats
, count
+ 1);
3909 formats
[count
++] = format_name
;
3914 qsort(formats
, count
, sizeof(formats
[0]), qsort_strcmp
);
3916 for (i
= 0; i
< count
; i
++) {
3917 it(opaque
, formats
[i
]);
3923 /* This function is to find a node in the bs graph */
3924 BlockDriverState
*bdrv_find_node(const char *node_name
)
3926 BlockDriverState
*bs
;
3930 QTAILQ_FOREACH(bs
, &graph_bdrv_states
, node_list
) {
3931 if (!strcmp(node_name
, bs
->node_name
)) {
3938 /* Put this QMP function here so it can access the static graph_bdrv_states. */
3939 BlockDeviceInfoList
*bdrv_named_nodes_list(Error
**errp
)
3941 BlockDeviceInfoList
*list
, *entry
;
3942 BlockDriverState
*bs
;
3945 QTAILQ_FOREACH(bs
, &graph_bdrv_states
, node_list
) {
3946 BlockDeviceInfo
*info
= bdrv_block_device_info(NULL
, bs
, errp
);
3948 qapi_free_BlockDeviceInfoList(list
);
3951 entry
= g_malloc0(sizeof(*entry
));
3952 entry
->value
= info
;
3960 BlockDriverState
*bdrv_lookup_bs(const char *device
,
3961 const char *node_name
,
3965 BlockDriverState
*bs
;
3968 blk
= blk_by_name(device
);
3973 error_setg(errp
, "Device '%s' has no medium", device
);
3981 bs
= bdrv_find_node(node_name
);
3988 error_setg(errp
, "Cannot find device=%s nor node_name=%s",
3989 device
? device
: "",
3990 node_name
? node_name
: "");
3994 /* If 'base' is in the same chain as 'top', return true. Otherwise,
3995 * return false. If either argument is NULL, return false. */
3996 bool bdrv_chain_contains(BlockDriverState
*top
, BlockDriverState
*base
)
3998 while (top
&& top
!= base
) {
3999 top
= backing_bs(top
);
4005 BlockDriverState
*bdrv_next_node(BlockDriverState
*bs
)
4008 return QTAILQ_FIRST(&graph_bdrv_states
);
4010 return QTAILQ_NEXT(bs
, node_list
);
4013 const char *bdrv_get_node_name(const BlockDriverState
*bs
)
4015 return bs
->node_name
;
4018 const char *bdrv_get_parent_name(const BlockDriverState
*bs
)
4023 /* If multiple parents have a name, just pick the first one. */
4024 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
4025 if (c
->role
->get_name
) {
4026 name
= c
->role
->get_name(c
);
4027 if (name
&& *name
) {
4036 /* TODO check what callers really want: bs->node_name or blk_name() */
4037 const char *bdrv_get_device_name(const BlockDriverState
*bs
)
4039 return bdrv_get_parent_name(bs
) ?: "";
4042 /* This can be used to identify nodes that might not have a device
4043 * name associated. Since node and device names live in the same
4044 * namespace, the result is unambiguous. The exception is if both are
4045 * absent, then this returns an empty (non-null) string. */
4046 const char *bdrv_get_device_or_node_name(const BlockDriverState
*bs
)
4048 return bdrv_get_parent_name(bs
) ?: bs
->node_name
;
4051 int bdrv_get_flags(BlockDriverState
*bs
)
4053 return bs
->open_flags
;
4056 int bdrv_has_zero_init_1(BlockDriverState
*bs
)
4061 int bdrv_has_zero_init(BlockDriverState
*bs
)
4067 /* If BS is a copy on write image, it is initialized to
4068 the contents of the base image, which may not be zeroes. */
4072 if (bs
->drv
->bdrv_has_zero_init
) {
4073 return bs
->drv
->bdrv_has_zero_init(bs
);
4075 if (bs
->file
&& bs
->drv
->is_filter
) {
4076 return bdrv_has_zero_init(bs
->file
->bs
);
4083 bool bdrv_unallocated_blocks_are_zero(BlockDriverState
*bs
)
4085 BlockDriverInfo bdi
;
4091 if (bdrv_get_info(bs
, &bdi
) == 0) {
4092 return bdi
.unallocated_blocks_are_zero
;
4098 bool bdrv_can_write_zeroes_with_unmap(BlockDriverState
*bs
)
4100 if (!(bs
->open_flags
& BDRV_O_UNMAP
)) {
4104 return bs
->supported_zero_flags
& BDRV_REQ_MAY_UNMAP
;
4107 const char *bdrv_get_encrypted_filename(BlockDriverState
*bs
)
4109 if (bs
->backing
&& bs
->backing
->bs
->encrypted
)
4110 return bs
->backing_file
;
4111 else if (bs
->encrypted
)
4112 return bs
->filename
;
4117 void bdrv_get_backing_filename(BlockDriverState
*bs
,
4118 char *filename
, int filename_size
)
4120 pstrcpy(filename
, filename_size
, bs
->backing_file
);
4123 int bdrv_get_info(BlockDriverState
*bs
, BlockDriverInfo
*bdi
)
4125 BlockDriver
*drv
= bs
->drv
;
4126 /* if bs->drv == NULL, bs is closed, so there's nothing to do here */
4130 if (!drv
->bdrv_get_info
) {
4131 if (bs
->file
&& drv
->is_filter
) {
4132 return bdrv_get_info(bs
->file
->bs
, bdi
);
4136 memset(bdi
, 0, sizeof(*bdi
));
4137 return drv
->bdrv_get_info(bs
, bdi
);
4140 ImageInfoSpecific
*bdrv_get_specific_info(BlockDriverState
*bs
)
4142 BlockDriver
*drv
= bs
->drv
;
4143 if (drv
&& drv
->bdrv_get_specific_info
) {
4144 return drv
->bdrv_get_specific_info(bs
);
4149 void bdrv_debug_event(BlockDriverState
*bs
, BlkdebugEvent event
)
4151 if (!bs
|| !bs
->drv
|| !bs
->drv
->bdrv_debug_event
) {
4155 bs
->drv
->bdrv_debug_event(bs
, event
);
4158 int bdrv_debug_breakpoint(BlockDriverState
*bs
, const char *event
,
4161 while (bs
&& bs
->drv
&& !bs
->drv
->bdrv_debug_breakpoint
) {
4162 bs
= bs
->file
? bs
->file
->bs
: NULL
;
4165 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_breakpoint
) {
4166 return bs
->drv
->bdrv_debug_breakpoint(bs
, event
, tag
);
4172 int bdrv_debug_remove_breakpoint(BlockDriverState
*bs
, const char *tag
)
4174 while (bs
&& bs
->drv
&& !bs
->drv
->bdrv_debug_remove_breakpoint
) {
4175 bs
= bs
->file
? bs
->file
->bs
: NULL
;
4178 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_remove_breakpoint
) {
4179 return bs
->drv
->bdrv_debug_remove_breakpoint(bs
, tag
);
4185 int bdrv_debug_resume(BlockDriverState
*bs
, const char *tag
)
4187 while (bs
&& (!bs
->drv
|| !bs
->drv
->bdrv_debug_resume
)) {
4188 bs
= bs
->file
? bs
->file
->bs
: NULL
;
4191 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_resume
) {
4192 return bs
->drv
->bdrv_debug_resume(bs
, tag
);
4198 bool bdrv_debug_is_suspended(BlockDriverState
*bs
, const char *tag
)
4200 while (bs
&& bs
->drv
&& !bs
->drv
->bdrv_debug_is_suspended
) {
4201 bs
= bs
->file
? bs
->file
->bs
: NULL
;
4204 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_is_suspended
) {
4205 return bs
->drv
->bdrv_debug_is_suspended(bs
, tag
);
4211 /* backing_file can either be relative, or absolute, or a protocol. If it is
4212 * relative, it must be relative to the chain. So, passing in bs->filename
4213 * from a BDS as backing_file should not be done, as that may be relative to
4214 * the CWD rather than the chain. */
4215 BlockDriverState
*bdrv_find_backing_image(BlockDriverState
*bs
,
4216 const char *backing_file
)
4218 char *filename_full
= NULL
;
4219 char *backing_file_full
= NULL
;
4220 char *filename_tmp
= NULL
;
4221 int is_protocol
= 0;
4222 BlockDriverState
*curr_bs
= NULL
;
4223 BlockDriverState
*retval
= NULL
;
4224 Error
*local_error
= NULL
;
4226 if (!bs
|| !bs
->drv
|| !backing_file
) {
4230 filename_full
= g_malloc(PATH_MAX
);
4231 backing_file_full
= g_malloc(PATH_MAX
);
4232 filename_tmp
= g_malloc(PATH_MAX
);
4234 is_protocol
= path_has_protocol(backing_file
);
4236 for (curr_bs
= bs
; curr_bs
->backing
; curr_bs
= curr_bs
->backing
->bs
) {
4238 /* If either of the filename paths is actually a protocol, then
4239 * compare unmodified paths; otherwise make paths relative */
4240 if (is_protocol
|| path_has_protocol(curr_bs
->backing_file
)) {
4241 if (strcmp(backing_file
, curr_bs
->backing_file
) == 0) {
4242 retval
= curr_bs
->backing
->bs
;
4245 /* Also check against the full backing filename for the image */
4246 bdrv_get_full_backing_filename(curr_bs
, backing_file_full
, PATH_MAX
,
4248 if (local_error
== NULL
) {
4249 if (strcmp(backing_file
, backing_file_full
) == 0) {
4250 retval
= curr_bs
->backing
->bs
;
4254 error_free(local_error
);
4258 /* If not an absolute filename path, make it relative to the current
4259 * image's filename path */
4260 path_combine(filename_tmp
, PATH_MAX
, curr_bs
->filename
,
4263 /* We are going to compare absolute pathnames */
4264 if (!realpath(filename_tmp
, filename_full
)) {
4268 /* We need to make sure the backing filename we are comparing against
4269 * is relative to the current image filename (or absolute) */
4270 path_combine(filename_tmp
, PATH_MAX
, curr_bs
->filename
,
4271 curr_bs
->backing_file
);
4273 if (!realpath(filename_tmp
, backing_file_full
)) {
4277 if (strcmp(backing_file_full
, filename_full
) == 0) {
4278 retval
= curr_bs
->backing
->bs
;
4284 g_free(filename_full
);
4285 g_free(backing_file_full
);
4286 g_free(filename_tmp
);
4290 void bdrv_init(void)
4292 module_call_init(MODULE_INIT_BLOCK
);
4295 void bdrv_init_with_whitelist(void)
4297 use_bdrv_whitelist
= 1;
4301 static void coroutine_fn
bdrv_co_invalidate_cache(BlockDriverState
*bs
,
4304 BdrvChild
*child
, *parent
;
4305 uint64_t perm
, shared_perm
;
4306 Error
*local_err
= NULL
;
4313 if (!(bs
->open_flags
& BDRV_O_INACTIVE
)) {
4317 QLIST_FOREACH(child
, &bs
->children
, next
) {
4318 bdrv_co_invalidate_cache(child
->bs
, &local_err
);
4320 error_propagate(errp
, local_err
);
4326 * Update permissions, they may differ for inactive nodes.
4328 * Note that the required permissions of inactive images are always a
4329 * subset of the permissions required after activating the image. This
4330 * allows us to just get the permissions upfront without restricting
4331 * drv->bdrv_invalidate_cache().
4333 * It also means that in error cases, we don't have to try and revert to
4334 * the old permissions (which is an operation that could fail, too). We can
4335 * just keep the extended permissions for the next time that an activation
4336 * of the image is tried.
4338 bs
->open_flags
&= ~BDRV_O_INACTIVE
;
4339 bdrv_get_cumulative_perm(bs
, &perm
, &shared_perm
);
4340 ret
= bdrv_check_perm(bs
, NULL
, perm
, shared_perm
, NULL
, &local_err
);
4342 bs
->open_flags
|= BDRV_O_INACTIVE
;
4343 error_propagate(errp
, local_err
);
4346 bdrv_set_perm(bs
, perm
, shared_perm
);
4348 if (bs
->drv
->bdrv_co_invalidate_cache
) {
4349 bs
->drv
->bdrv_co_invalidate_cache(bs
, &local_err
);
4351 bs
->open_flags
|= BDRV_O_INACTIVE
;
4352 error_propagate(errp
, local_err
);
4357 ret
= refresh_total_sectors(bs
, bs
->total_sectors
);
4359 bs
->open_flags
|= BDRV_O_INACTIVE
;
4360 error_setg_errno(errp
, -ret
, "Could not refresh total sector count");
4364 QLIST_FOREACH(parent
, &bs
->parents
, next_parent
) {
4365 if (parent
->role
->activate
) {
4366 parent
->role
->activate(parent
, &local_err
);
4368 error_propagate(errp
, local_err
);
4375 typedef struct InvalidateCacheCo
{
4376 BlockDriverState
*bs
;
4379 } InvalidateCacheCo
;
4381 static void coroutine_fn
bdrv_invalidate_cache_co_entry(void *opaque
)
4383 InvalidateCacheCo
*ico
= opaque
;
4384 bdrv_co_invalidate_cache(ico
->bs
, ico
->errp
);
4388 void bdrv_invalidate_cache(BlockDriverState
*bs
, Error
**errp
)
4391 InvalidateCacheCo ico
= {
4397 if (qemu_in_coroutine()) {
4398 /* Fast-path if already in coroutine context */
4399 bdrv_invalidate_cache_co_entry(&ico
);
4401 co
= qemu_coroutine_create(bdrv_invalidate_cache_co_entry
, &ico
);
4402 qemu_coroutine_enter(co
);
4403 BDRV_POLL_WHILE(bs
, !ico
.done
);
4407 void bdrv_invalidate_cache_all(Error
**errp
)
4409 BlockDriverState
*bs
;
4410 Error
*local_err
= NULL
;
4411 BdrvNextIterator it
;
4413 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
4414 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
4416 aio_context_acquire(aio_context
);
4417 bdrv_invalidate_cache(bs
, &local_err
);
4418 aio_context_release(aio_context
);
4420 error_propagate(errp
, local_err
);
4421 bdrv_next_cleanup(&it
);
4427 static int bdrv_inactivate_recurse(BlockDriverState
*bs
,
4430 BdrvChild
*child
, *parent
;
4437 if (!setting_flag
&& bs
->drv
->bdrv_inactivate
) {
4438 ret
= bs
->drv
->bdrv_inactivate(bs
);
4444 if (setting_flag
&& !(bs
->open_flags
& BDRV_O_INACTIVE
)) {
4445 uint64_t perm
, shared_perm
;
4447 QLIST_FOREACH(parent
, &bs
->parents
, next_parent
) {
4448 if (parent
->role
->inactivate
) {
4449 ret
= parent
->role
->inactivate(parent
);
4456 bs
->open_flags
|= BDRV_O_INACTIVE
;
4458 /* Update permissions, they may differ for inactive nodes */
4459 bdrv_get_cumulative_perm(bs
, &perm
, &shared_perm
);
4460 bdrv_check_perm(bs
, NULL
, perm
, shared_perm
, NULL
, &error_abort
);
4461 bdrv_set_perm(bs
, perm
, shared_perm
);
4464 QLIST_FOREACH(child
, &bs
->children
, next
) {
4465 ret
= bdrv_inactivate_recurse(child
->bs
, setting_flag
);
4471 /* At this point persistent bitmaps should be already stored by the format
4473 bdrv_release_persistent_dirty_bitmaps(bs
);
4478 int bdrv_inactivate_all(void)
4480 BlockDriverState
*bs
= NULL
;
4481 BdrvNextIterator it
;
4484 GSList
*aio_ctxs
= NULL
, *ctx
;
4486 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
4487 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
4489 if (!g_slist_find(aio_ctxs
, aio_context
)) {
4490 aio_ctxs
= g_slist_prepend(aio_ctxs
, aio_context
);
4491 aio_context_acquire(aio_context
);
4495 /* We do two passes of inactivation. The first pass calls to drivers'
4496 * .bdrv_inactivate callbacks recursively so all cache is flushed to disk;
4497 * the second pass sets the BDRV_O_INACTIVE flag so that no further write
4499 for (pass
= 0; pass
< 2; pass
++) {
4500 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
4501 ret
= bdrv_inactivate_recurse(bs
, pass
);
4503 bdrv_next_cleanup(&it
);
4510 for (ctx
= aio_ctxs
; ctx
!= NULL
; ctx
= ctx
->next
) {
4511 AioContext
*aio_context
= ctx
->data
;
4512 aio_context_release(aio_context
);
4514 g_slist_free(aio_ctxs
);
4519 /**************************************************************/
4520 /* removable device support */
4523 * Return TRUE if the media is present
4525 bool bdrv_is_inserted(BlockDriverState
*bs
)
4527 BlockDriver
*drv
= bs
->drv
;
4533 if (drv
->bdrv_is_inserted
) {
4534 return drv
->bdrv_is_inserted(bs
);
4536 QLIST_FOREACH(child
, &bs
->children
, next
) {
4537 if (!bdrv_is_inserted(child
->bs
)) {
4545 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
4547 void bdrv_eject(BlockDriverState
*bs
, bool eject_flag
)
4549 BlockDriver
*drv
= bs
->drv
;
4551 if (drv
&& drv
->bdrv_eject
) {
4552 drv
->bdrv_eject(bs
, eject_flag
);
4557 * Lock or unlock the media (if it is locked, the user won't be able
4558 * to eject it manually).
4560 void bdrv_lock_medium(BlockDriverState
*bs
, bool locked
)
4562 BlockDriver
*drv
= bs
->drv
;
4564 trace_bdrv_lock_medium(bs
, locked
);
4566 if (drv
&& drv
->bdrv_lock_medium
) {
4567 drv
->bdrv_lock_medium(bs
, locked
);
4571 /* Get a reference to bs */
4572 void bdrv_ref(BlockDriverState
*bs
)
4577 /* Release a previously grabbed reference to bs.
4578 * If after releasing, reference count is zero, the BlockDriverState is
4580 void bdrv_unref(BlockDriverState
*bs
)
4585 assert(bs
->refcnt
> 0);
4586 if (--bs
->refcnt
== 0) {
4591 struct BdrvOpBlocker
{
4593 QLIST_ENTRY(BdrvOpBlocker
) list
;
4596 bool bdrv_op_is_blocked(BlockDriverState
*bs
, BlockOpType op
, Error
**errp
)
4598 BdrvOpBlocker
*blocker
;
4599 assert((int) op
>= 0 && op
< BLOCK_OP_TYPE_MAX
);
4600 if (!QLIST_EMPTY(&bs
->op_blockers
[op
])) {
4601 blocker
= QLIST_FIRST(&bs
->op_blockers
[op
]);
4602 error_propagate(errp
, error_copy(blocker
->reason
));
4603 error_prepend(errp
, "Node '%s' is busy: ",
4604 bdrv_get_device_or_node_name(bs
));
4610 void bdrv_op_block(BlockDriverState
*bs
, BlockOpType op
, Error
*reason
)
4612 BdrvOpBlocker
*blocker
;
4613 assert((int) op
>= 0 && op
< BLOCK_OP_TYPE_MAX
);
4615 blocker
= g_new0(BdrvOpBlocker
, 1);
4616 blocker
->reason
= reason
;
4617 QLIST_INSERT_HEAD(&bs
->op_blockers
[op
], blocker
, list
);
4620 void bdrv_op_unblock(BlockDriverState
*bs
, BlockOpType op
, Error
*reason
)
4622 BdrvOpBlocker
*blocker
, *next
;
4623 assert((int) op
>= 0 && op
< BLOCK_OP_TYPE_MAX
);
4624 QLIST_FOREACH_SAFE(blocker
, &bs
->op_blockers
[op
], list
, next
) {
4625 if (blocker
->reason
== reason
) {
4626 QLIST_REMOVE(blocker
, list
);
4632 void bdrv_op_block_all(BlockDriverState
*bs
, Error
*reason
)
4635 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
4636 bdrv_op_block(bs
, i
, reason
);
4640 void bdrv_op_unblock_all(BlockDriverState
*bs
, Error
*reason
)
4643 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
4644 bdrv_op_unblock(bs
, i
, reason
);
4648 bool bdrv_op_blocker_is_empty(BlockDriverState
*bs
)
4652 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
4653 if (!QLIST_EMPTY(&bs
->op_blockers
[i
])) {
4660 void bdrv_img_create(const char *filename
, const char *fmt
,
4661 const char *base_filename
, const char *base_fmt
,
4662 char *options
, uint64_t img_size
, int flags
, bool quiet
,
4665 QemuOptsList
*create_opts
= NULL
;
4666 QemuOpts
*opts
= NULL
;
4667 const char *backing_fmt
, *backing_file
;
4669 BlockDriver
*drv
, *proto_drv
;
4670 Error
*local_err
= NULL
;
4673 /* Find driver and parse its options */
4674 drv
= bdrv_find_format(fmt
);
4676 error_setg(errp
, "Unknown file format '%s'", fmt
);
4680 proto_drv
= bdrv_find_protocol(filename
, true, errp
);
4685 if (!drv
->create_opts
) {
4686 error_setg(errp
, "Format driver '%s' does not support image creation",
4691 if (!proto_drv
->create_opts
) {
4692 error_setg(errp
, "Protocol driver '%s' does not support image creation",
4693 proto_drv
->format_name
);
4697 create_opts
= qemu_opts_append(create_opts
, drv
->create_opts
);
4698 create_opts
= qemu_opts_append(create_opts
, proto_drv
->create_opts
);
4700 /* Create parameter list with default values */
4701 opts
= qemu_opts_create(create_opts
, NULL
, 0, &error_abort
);
4702 qemu_opt_set_number(opts
, BLOCK_OPT_SIZE
, img_size
, &error_abort
);
4704 /* Parse -o options */
4706 qemu_opts_do_parse(opts
, options
, NULL
, &local_err
);
4708 error_report_err(local_err
);
4710 error_setg(errp
, "Invalid options for file format '%s'", fmt
);
4715 if (base_filename
) {
4716 qemu_opt_set(opts
, BLOCK_OPT_BACKING_FILE
, base_filename
, &local_err
);
4718 error_setg(errp
, "Backing file not supported for file format '%s'",
4725 qemu_opt_set(opts
, BLOCK_OPT_BACKING_FMT
, base_fmt
, &local_err
);
4727 error_setg(errp
, "Backing file format not supported for file "
4728 "format '%s'", fmt
);
4733 backing_file
= qemu_opt_get(opts
, BLOCK_OPT_BACKING_FILE
);
4735 if (!strcmp(filename
, backing_file
)) {
4736 error_setg(errp
, "Error: Trying to create an image with the "
4737 "same filename as the backing file");
4742 backing_fmt
= qemu_opt_get(opts
, BLOCK_OPT_BACKING_FMT
);
4744 /* The size for the image must always be specified, unless we have a backing
4745 * file and we have not been forbidden from opening it. */
4746 size
= qemu_opt_get_size(opts
, BLOCK_OPT_SIZE
, img_size
);
4747 if (backing_file
&& !(flags
& BDRV_O_NO_BACKING
)) {
4748 BlockDriverState
*bs
;
4749 char *full_backing
= g_new0(char, PATH_MAX
);
4751 QDict
*backing_options
= NULL
;
4753 bdrv_get_full_backing_filename_from_filename(filename
, backing_file
,
4754 full_backing
, PATH_MAX
,
4757 g_free(full_backing
);
4761 /* backing files always opened read-only */
4763 back_flags
&= ~(BDRV_O_RDWR
| BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
);
4765 backing_options
= qdict_new();
4767 qdict_put_str(backing_options
, "driver", backing_fmt
);
4769 qdict_put_bool(backing_options
, BDRV_OPT_FORCE_SHARE
, true);
4771 bs
= bdrv_open(full_backing
, NULL
, backing_options
, back_flags
,
4773 g_free(full_backing
);
4774 if (!bs
&& size
!= -1) {
4775 /* Couldn't open BS, but we have a size, so it's nonfatal */
4776 warn_reportf_err(local_err
,
4777 "Could not verify backing image. "
4778 "This may become an error in future versions.\n");
4781 /* Couldn't open bs, do not have size */
4782 error_append_hint(&local_err
,
4783 "Could not open backing image to determine size.\n");
4787 /* Opened BS, have no size */
4788 size
= bdrv_getlength(bs
);
4790 error_setg_errno(errp
, -size
, "Could not get size of '%s'",
4795 qemu_opt_set_number(opts
, BLOCK_OPT_SIZE
, size
, &error_abort
);
4799 } /* (backing_file && !(flags & BDRV_O_NO_BACKING)) */
4802 error_setg(errp
, "Image creation needs a size parameter");
4807 printf("Formatting '%s', fmt=%s ", filename
, fmt
);
4808 qemu_opts_print(opts
, " ");
4812 ret
= bdrv_create(drv
, filename
, opts
, &local_err
);
4814 if (ret
== -EFBIG
) {
4815 /* This is generally a better message than whatever the driver would
4816 * deliver (especially because of the cluster_size_hint), since that
4817 * is most probably not much different from "image too large". */
4818 const char *cluster_size_hint
= "";
4819 if (qemu_opt_get_size(opts
, BLOCK_OPT_CLUSTER_SIZE
, 0)) {
4820 cluster_size_hint
= " (try using a larger cluster size)";
4822 error_setg(errp
, "The image size is too large for file format '%s'"
4823 "%s", fmt
, cluster_size_hint
);
4824 error_free(local_err
);
4829 qemu_opts_del(opts
);
4830 qemu_opts_free(create_opts
);
4831 error_propagate(errp
, local_err
);
4834 AioContext
*bdrv_get_aio_context(BlockDriverState
*bs
)
4836 return bs
? bs
->aio_context
: qemu_get_aio_context();
4839 AioWait
*bdrv_get_aio_wait(BlockDriverState
*bs
)
4841 return bs
? &bs
->wait
: NULL
;
4844 void bdrv_coroutine_enter(BlockDriverState
*bs
, Coroutine
*co
)
4846 aio_co_enter(bdrv_get_aio_context(bs
), co
);
4849 static void bdrv_do_remove_aio_context_notifier(BdrvAioNotifier
*ban
)
4851 QLIST_REMOVE(ban
, list
);
4855 void bdrv_detach_aio_context(BlockDriverState
*bs
)
4857 BdrvAioNotifier
*baf
, *baf_tmp
;
4864 assert(!bs
->walking_aio_notifiers
);
4865 bs
->walking_aio_notifiers
= true;
4866 QLIST_FOREACH_SAFE(baf
, &bs
->aio_notifiers
, list
, baf_tmp
) {
4868 bdrv_do_remove_aio_context_notifier(baf
);
4870 baf
->detach_aio_context(baf
->opaque
);
4873 /* Never mind iterating again to check for ->deleted. bdrv_close() will
4874 * remove remaining aio notifiers if we aren't called again.
4876 bs
->walking_aio_notifiers
= false;
4878 if (bs
->drv
->bdrv_detach_aio_context
) {
4879 bs
->drv
->bdrv_detach_aio_context(bs
);
4881 QLIST_FOREACH(child
, &bs
->children
, next
) {
4882 bdrv_detach_aio_context(child
->bs
);
4885 bs
->aio_context
= NULL
;
4888 void bdrv_attach_aio_context(BlockDriverState
*bs
,
4889 AioContext
*new_context
)
4891 BdrvAioNotifier
*ban
, *ban_tmp
;
4898 bs
->aio_context
= new_context
;
4900 QLIST_FOREACH(child
, &bs
->children
, next
) {
4901 bdrv_attach_aio_context(child
->bs
, new_context
);
4903 if (bs
->drv
->bdrv_attach_aio_context
) {
4904 bs
->drv
->bdrv_attach_aio_context(bs
, new_context
);
4907 assert(!bs
->walking_aio_notifiers
);
4908 bs
->walking_aio_notifiers
= true;
4909 QLIST_FOREACH_SAFE(ban
, &bs
->aio_notifiers
, list
, ban_tmp
) {
4911 bdrv_do_remove_aio_context_notifier(ban
);
4913 ban
->attached_aio_context(new_context
, ban
->opaque
);
4916 bs
->walking_aio_notifiers
= false;
4919 void bdrv_set_aio_context(BlockDriverState
*bs
, AioContext
*new_context
)
4921 AioContext
*ctx
= bdrv_get_aio_context(bs
);
4923 aio_disable_external(ctx
);
4924 bdrv_parent_drained_begin(bs
, NULL
);
4925 bdrv_drain(bs
); /* ensure there are no in-flight requests */
4927 while (aio_poll(ctx
, false)) {
4928 /* wait for all bottom halves to execute */
4931 bdrv_detach_aio_context(bs
);
4933 /* This function executes in the old AioContext so acquire the new one in
4934 * case it runs in a different thread.
4936 aio_context_acquire(new_context
);
4937 bdrv_attach_aio_context(bs
, new_context
);
4938 bdrv_parent_drained_end(bs
, NULL
);
4939 aio_enable_external(ctx
);
4940 aio_context_release(new_context
);
4943 void bdrv_add_aio_context_notifier(BlockDriverState
*bs
,
4944 void (*attached_aio_context
)(AioContext
*new_context
, void *opaque
),
4945 void (*detach_aio_context
)(void *opaque
), void *opaque
)
4947 BdrvAioNotifier
*ban
= g_new(BdrvAioNotifier
, 1);
4948 *ban
= (BdrvAioNotifier
){
4949 .attached_aio_context
= attached_aio_context
,
4950 .detach_aio_context
= detach_aio_context
,
4954 QLIST_INSERT_HEAD(&bs
->aio_notifiers
, ban
, list
);
4957 void bdrv_remove_aio_context_notifier(BlockDriverState
*bs
,
4958 void (*attached_aio_context
)(AioContext
*,
4960 void (*detach_aio_context
)(void *),
4963 BdrvAioNotifier
*ban
, *ban_next
;
4965 QLIST_FOREACH_SAFE(ban
, &bs
->aio_notifiers
, list
, ban_next
) {
4966 if (ban
->attached_aio_context
== attached_aio_context
&&
4967 ban
->detach_aio_context
== detach_aio_context
&&
4968 ban
->opaque
== opaque
&&
4969 ban
->deleted
== false)
4971 if (bs
->walking_aio_notifiers
) {
4972 ban
->deleted
= true;
4974 bdrv_do_remove_aio_context_notifier(ban
);
4983 int bdrv_amend_options(BlockDriverState
*bs
, QemuOpts
*opts
,
4984 BlockDriverAmendStatusCB
*status_cb
, void *cb_opaque
)
4989 if (!bs
->drv
->bdrv_amend_options
) {
4992 return bs
->drv
->bdrv_amend_options(bs
, opts
, status_cb
, cb_opaque
);
4995 /* This function will be called by the bdrv_recurse_is_first_non_filter method
4996 * of block filter and by bdrv_is_first_non_filter.
4997 * It is used to test if the given bs is the candidate or recurse more in the
5000 bool bdrv_recurse_is_first_non_filter(BlockDriverState
*bs
,
5001 BlockDriverState
*candidate
)
5003 /* return false if basic checks fails */
5004 if (!bs
|| !bs
->drv
) {
5008 /* the code reached a non block filter driver -> check if the bs is
5009 * the same as the candidate. It's the recursion termination condition.
5011 if (!bs
->drv
->is_filter
) {
5012 return bs
== candidate
;
5014 /* Down this path the driver is a block filter driver */
5016 /* If the block filter recursion method is defined use it to recurse down
5019 if (bs
->drv
->bdrv_recurse_is_first_non_filter
) {
5020 return bs
->drv
->bdrv_recurse_is_first_non_filter(bs
, candidate
);
5023 /* the driver is a block filter but don't allow to recurse -> return false
5028 /* This function checks if the candidate is the first non filter bs down it's
5029 * bs chain. Since we don't have pointers to parents it explore all bs chains
5030 * from the top. Some filters can choose not to pass down the recursion.
5032 bool bdrv_is_first_non_filter(BlockDriverState
*candidate
)
5034 BlockDriverState
*bs
;
5035 BdrvNextIterator it
;
5037 /* walk down the bs forest recursively */
5038 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
5041 /* try to recurse in this top level bs */
5042 perm
= bdrv_recurse_is_first_non_filter(bs
, candidate
);
5044 /* candidate is the first non filter */
5046 bdrv_next_cleanup(&it
);
5054 BlockDriverState
*check_to_replace_node(BlockDriverState
*parent_bs
,
5055 const char *node_name
, Error
**errp
)
5057 BlockDriverState
*to_replace_bs
= bdrv_find_node(node_name
);
5058 AioContext
*aio_context
;
5060 if (!to_replace_bs
) {
5061 error_setg(errp
, "Node name '%s' not found", node_name
);
5065 aio_context
= bdrv_get_aio_context(to_replace_bs
);
5066 aio_context_acquire(aio_context
);
5068 if (bdrv_op_is_blocked(to_replace_bs
, BLOCK_OP_TYPE_REPLACE
, errp
)) {
5069 to_replace_bs
= NULL
;
5073 /* We don't want arbitrary node of the BDS chain to be replaced only the top
5074 * most non filter in order to prevent data corruption.
5075 * Another benefit is that this tests exclude backing files which are
5076 * blocked by the backing blockers.
5078 if (!bdrv_recurse_is_first_non_filter(parent_bs
, to_replace_bs
)) {
5079 error_setg(errp
, "Only top most non filter can be replaced");
5080 to_replace_bs
= NULL
;
5085 aio_context_release(aio_context
);
5086 return to_replace_bs
;
5089 static bool append_open_options(QDict
*d
, BlockDriverState
*bs
)
5091 const QDictEntry
*entry
;
5094 bool found_any
= false;
5097 for (entry
= qdict_first(bs
->options
); entry
;
5098 entry
= qdict_next(bs
->options
, entry
))
5100 /* Exclude options for children */
5101 QLIST_FOREACH(child
, &bs
->children
, next
) {
5102 if (strstart(qdict_entry_key(entry
), child
->name
, &p
)
5103 && (!*p
|| *p
== '.'))
5112 /* And exclude all non-driver-specific options */
5113 for (desc
= bdrv_runtime_opts
.desc
; desc
->name
; desc
++) {
5114 if (!strcmp(qdict_entry_key(entry
), desc
->name
)) {
5122 qobject_incref(qdict_entry_value(entry
));
5123 qdict_put_obj(d
, qdict_entry_key(entry
), qdict_entry_value(entry
));
5130 /* Updates the following BDS fields:
5131 * - exact_filename: A filename which may be used for opening a block device
5132 * which (mostly) equals the given BDS (even without any
5133 * other options; so reading and writing must return the same
5134 * results, but caching etc. may be different)
5135 * - full_open_options: Options which, when given when opening a block device
5136 * (without a filename), result in a BDS (mostly)
5137 * equalling the given one
5138 * - filename: If exact_filename is set, it is copied here. Otherwise,
5139 * full_open_options is converted to a JSON object, prefixed with
5140 * "json:" (for use through the JSON pseudo protocol) and put here.
5142 void bdrv_refresh_filename(BlockDriverState
*bs
)
5144 BlockDriver
*drv
= bs
->drv
;
5151 /* This BDS's file name will most probably depend on its file's name, so
5152 * refresh that first */
5154 bdrv_refresh_filename(bs
->file
->bs
);
5157 if (drv
->bdrv_refresh_filename
) {
5158 /* Obsolete information is of no use here, so drop the old file name
5159 * information before refreshing it */
5160 bs
->exact_filename
[0] = '\0';
5161 if (bs
->full_open_options
) {
5162 QDECREF(bs
->full_open_options
);
5163 bs
->full_open_options
= NULL
;
5167 append_open_options(opts
, bs
);
5168 drv
->bdrv_refresh_filename(bs
, opts
);
5170 } else if (bs
->file
) {
5171 /* Try to reconstruct valid information from the underlying file */
5172 bool has_open_options
;
5174 bs
->exact_filename
[0] = '\0';
5175 if (bs
->full_open_options
) {
5176 QDECREF(bs
->full_open_options
);
5177 bs
->full_open_options
= NULL
;
5181 has_open_options
= append_open_options(opts
, bs
);
5183 /* If no specific options have been given for this BDS, the filename of
5184 * the underlying file should suffice for this one as well */
5185 if (bs
->file
->bs
->exact_filename
[0] && !has_open_options
) {
5186 strcpy(bs
->exact_filename
, bs
->file
->bs
->exact_filename
);
5188 /* Reconstructing the full options QDict is simple for most format block
5189 * drivers, as long as the full options are known for the underlying
5190 * file BDS. The full options QDict of that file BDS should somehow
5191 * contain a representation of the filename, therefore the following
5192 * suffices without querying the (exact_)filename of this BDS. */
5193 if (bs
->file
->bs
->full_open_options
) {
5194 qdict_put_str(opts
, "driver", drv
->format_name
);
5195 QINCREF(bs
->file
->bs
->full_open_options
);
5196 qdict_put(opts
, "file", bs
->file
->bs
->full_open_options
);
5198 bs
->full_open_options
= opts
;
5202 } else if (!bs
->full_open_options
&& qdict_size(bs
->options
)) {
5203 /* There is no underlying file BDS (at least referenced by BDS.file),
5204 * so the full options QDict should be equal to the options given
5205 * specifically for this block device when it was opened (plus the
5206 * driver specification).
5207 * Because those options don't change, there is no need to update
5208 * full_open_options when it's already set. */
5211 append_open_options(opts
, bs
);
5212 qdict_put_str(opts
, "driver", drv
->format_name
);
5214 if (bs
->exact_filename
[0]) {
5215 /* This may not work for all block protocol drivers (some may
5216 * require this filename to be parsed), but we have to find some
5217 * default solution here, so just include it. If some block driver
5218 * does not support pure options without any filename at all or
5219 * needs some special format of the options QDict, it needs to
5220 * implement the driver-specific bdrv_refresh_filename() function.
5222 qdict_put_str(opts
, "filename", bs
->exact_filename
);
5225 bs
->full_open_options
= opts
;
5228 if (bs
->exact_filename
[0]) {
5229 pstrcpy(bs
->filename
, sizeof(bs
->filename
), bs
->exact_filename
);
5230 } else if (bs
->full_open_options
) {
5231 QString
*json
= qobject_to_json(QOBJECT(bs
->full_open_options
));
5232 snprintf(bs
->filename
, sizeof(bs
->filename
), "json:%s",
5233 qstring_get_str(json
));
5239 * Hot add/remove a BDS's child. So the user can take a child offline when
5240 * it is broken and take a new child online
5242 void bdrv_add_child(BlockDriverState
*parent_bs
, BlockDriverState
*child_bs
,
5246 if (!parent_bs
->drv
|| !parent_bs
->drv
->bdrv_add_child
) {
5247 error_setg(errp
, "The node %s does not support adding a child",
5248 bdrv_get_device_or_node_name(parent_bs
));
5252 if (!QLIST_EMPTY(&child_bs
->parents
)) {
5253 error_setg(errp
, "The node %s already has a parent",
5254 child_bs
->node_name
);
5258 parent_bs
->drv
->bdrv_add_child(parent_bs
, child_bs
, errp
);
5261 void bdrv_del_child(BlockDriverState
*parent_bs
, BdrvChild
*child
, Error
**errp
)
5265 if (!parent_bs
->drv
|| !parent_bs
->drv
->bdrv_del_child
) {
5266 error_setg(errp
, "The node %s does not support removing a child",
5267 bdrv_get_device_or_node_name(parent_bs
));
5271 QLIST_FOREACH(tmp
, &parent_bs
->children
, next
) {
5278 error_setg(errp
, "The node %s does not have a child named %s",
5279 bdrv_get_device_or_node_name(parent_bs
),
5280 bdrv_get_device_or_node_name(child
->bs
));
5284 parent_bs
->drv
->bdrv_del_child(parent_bs
, child
, errp
);
5287 bool bdrv_can_store_new_dirty_bitmap(BlockDriverState
*bs
, const char *name
,
5288 uint32_t granularity
, Error
**errp
)
5290 BlockDriver
*drv
= bs
->drv
;
5293 error_setg_errno(errp
, ENOMEDIUM
,
5294 "Can't store persistent bitmaps to %s",
5295 bdrv_get_device_or_node_name(bs
));
5299 if (!drv
->bdrv_can_store_new_dirty_bitmap
) {
5300 error_setg_errno(errp
, ENOTSUP
,
5301 "Can't store persistent bitmaps to %s",
5302 bdrv_get_device_or_node_name(bs
));
5306 return drv
->bdrv_can_store_new_dirty_bitmap(bs
, name
, granularity
, errp
);