2 * QEMU System Emulator block driver
4 * Copyright (c) 2003 Fabrice Bellard
6 * Permission is hereby granted, free of charge, to any person obtaining a copy
7 * of this software and associated documentation files (the "Software"), to deal
8 * in the Software without restriction, including without limitation the rights
9 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10 * copies of the Software, and to permit persons to whom the Software is
11 * furnished to do so, subject to the following conditions:
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
25 #include "qemu/osdep.h"
26 #include "block/trace.h"
27 #include "block/block_int.h"
28 #include "block/blockjob.h"
29 #include "block/nbd.h"
30 #include "block/qdict.h"
31 #include "qemu/error-report.h"
32 #include "module_block.h"
33 #include "qemu/module.h"
34 #include "qapi/error.h"
35 #include "qapi/qmp/qdict.h"
36 #include "qapi/qmp/qjson.h"
37 #include "qapi/qmp/qnull.h"
38 #include "qapi/qmp/qstring.h"
39 #include "qapi/qobject-output-visitor.h"
40 #include "qapi/qapi-visit-block-core.h"
41 #include "sysemu/block-backend.h"
42 #include "sysemu/sysemu.h"
43 #include "qemu/notify.h"
44 #include "qemu/option.h"
45 #include "qemu/coroutine.h"
46 #include "block/qapi.h"
47 #include "qemu/timer.h"
48 #include "qemu/cutils.h"
52 #include <sys/ioctl.h>
53 #include <sys/queue.h>
63 #define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */
65 static QTAILQ_HEAD(, BlockDriverState
) graph_bdrv_states
=
66 QTAILQ_HEAD_INITIALIZER(graph_bdrv_states
);
68 static QTAILQ_HEAD(, BlockDriverState
) all_bdrv_states
=
69 QTAILQ_HEAD_INITIALIZER(all_bdrv_states
);
71 static QLIST_HEAD(, BlockDriver
) bdrv_drivers
=
72 QLIST_HEAD_INITIALIZER(bdrv_drivers
);
74 static BlockDriverState
*bdrv_open_inherit(const char *filename
,
75 const char *reference
,
76 QDict
*options
, int flags
,
77 BlockDriverState
*parent
,
78 const BdrvChildRole
*child_role
,
81 /* If non-zero, use only whitelisted block drivers */
82 static int use_bdrv_whitelist
;
85 static int is_windows_drive_prefix(const char *filename
)
87 return (((filename
[0] >= 'a' && filename
[0] <= 'z') ||
88 (filename
[0] >= 'A' && filename
[0] <= 'Z')) &&
92 int is_windows_drive(const char *filename
)
94 if (is_windows_drive_prefix(filename
) &&
97 if (strstart(filename
, "\\\\.\\", NULL
) ||
98 strstart(filename
, "//./", NULL
))
104 size_t bdrv_opt_mem_align(BlockDriverState
*bs
)
106 if (!bs
|| !bs
->drv
) {
107 /* page size or 4k (hdd sector size) should be on the safe side */
108 return MAX(4096, getpagesize());
111 return bs
->bl
.opt_mem_alignment
;
114 size_t bdrv_min_mem_align(BlockDriverState
*bs
)
116 if (!bs
|| !bs
->drv
) {
117 /* page size or 4k (hdd sector size) should be on the safe side */
118 return MAX(4096, getpagesize());
121 return bs
->bl
.min_mem_alignment
;
124 /* check if the path starts with "<protocol>:" */
125 int path_has_protocol(const char *path
)
130 if (is_windows_drive(path
) ||
131 is_windows_drive_prefix(path
)) {
134 p
= path
+ strcspn(path
, ":/\\");
136 p
= path
+ strcspn(path
, ":/");
142 int path_is_absolute(const char *path
)
145 /* specific case for names like: "\\.\d:" */
146 if (is_windows_drive(path
) || is_windows_drive_prefix(path
)) {
149 return (*path
== '/' || *path
== '\\');
151 return (*path
== '/');
155 /* if filename is absolute, just copy it to dest. Otherwise, build a
156 path to it by considering it is relative to base_path. URL are
158 void path_combine(char *dest
, int dest_size
,
159 const char *base_path
,
160 const char *filename
)
167 if (path_is_absolute(filename
)) {
168 pstrcpy(dest
, dest_size
, filename
);
170 const char *protocol_stripped
= NULL
;
172 if (path_has_protocol(base_path
)) {
173 protocol_stripped
= strchr(base_path
, ':');
174 if (protocol_stripped
) {
178 p
= protocol_stripped
?: base_path
;
180 p1
= strrchr(base_path
, '/');
184 p2
= strrchr(base_path
, '\\');
196 if (len
> dest_size
- 1)
198 memcpy(dest
, base_path
, len
);
200 pstrcat(dest
, dest_size
, filename
);
205 * Helper function for bdrv_parse_filename() implementations to remove optional
206 * protocol prefixes (especially "file:") from a filename and for putting the
207 * stripped filename into the options QDict if there is such a prefix.
209 void bdrv_parse_filename_strip_prefix(const char *filename
, const char *prefix
,
212 if (strstart(filename
, prefix
, &filename
)) {
213 /* Stripping the explicit protocol prefix may result in a protocol
214 * prefix being (wrongly) detected (if the filename contains a colon) */
215 if (path_has_protocol(filename
)) {
216 QString
*fat_filename
;
218 /* This means there is some colon before the first slash; therefore,
219 * this cannot be an absolute path */
220 assert(!path_is_absolute(filename
));
222 /* And we can thus fix the protocol detection issue by prefixing it
224 fat_filename
= qstring_from_str("./");
225 qstring_append(fat_filename
, filename
);
227 assert(!path_has_protocol(qstring_get_str(fat_filename
)));
229 qdict_put(options
, "filename", fat_filename
);
231 /* If no protocol prefix was detected, we can use the shortened
233 qdict_put_str(options
, "filename", filename
);
239 /* Returns whether the image file is opened as read-only. Note that this can
240 * return false and writing to the image file is still not possible because the
241 * image is inactivated. */
242 bool bdrv_is_read_only(BlockDriverState
*bs
)
244 return bs
->read_only
;
247 int bdrv_can_set_read_only(BlockDriverState
*bs
, bool read_only
,
248 bool ignore_allow_rdw
, Error
**errp
)
250 /* Do not set read_only if copy_on_read is enabled */
251 if (bs
->copy_on_read
&& read_only
) {
252 error_setg(errp
, "Can't set node '%s' to r/o with copy-on-read enabled",
253 bdrv_get_device_or_node_name(bs
));
257 /* Do not clear read_only if it is prohibited */
258 if (!read_only
&& !(bs
->open_flags
& BDRV_O_ALLOW_RDWR
) &&
261 error_setg(errp
, "Node '%s' is read only",
262 bdrv_get_device_or_node_name(bs
));
269 /* TODO Remove (deprecated since 2.11)
270 * Block drivers are not supposed to automatically change bs->read_only.
271 * Instead, they should just check whether they can provide what the user
272 * explicitly requested and error out if read-write is requested, but they can
273 * only provide read-only access. */
274 int bdrv_set_read_only(BlockDriverState
*bs
, bool read_only
, Error
**errp
)
278 ret
= bdrv_can_set_read_only(bs
, read_only
, false, errp
);
283 bs
->read_only
= read_only
;
287 void bdrv_get_full_backing_filename_from_filename(const char *backed
,
289 char *dest
, size_t sz
,
292 if (backing
[0] == '\0' || path_has_protocol(backing
) ||
293 path_is_absolute(backing
))
295 pstrcpy(dest
, sz
, backing
);
296 } else if (backed
[0] == '\0' || strstart(backed
, "json:", NULL
)) {
297 error_setg(errp
, "Cannot use relative backing file names for '%s'",
300 path_combine(dest
, sz
, backed
, backing
);
304 void bdrv_get_full_backing_filename(BlockDriverState
*bs
, char *dest
, size_t sz
,
307 char *backed
= bs
->exact_filename
[0] ? bs
->exact_filename
: bs
->filename
;
309 bdrv_get_full_backing_filename_from_filename(backed
, bs
->backing_file
,
313 void bdrv_register(BlockDriver
*bdrv
)
315 QLIST_INSERT_HEAD(&bdrv_drivers
, bdrv
, list
);
318 BlockDriverState
*bdrv_new(void)
320 BlockDriverState
*bs
;
323 bs
= g_new0(BlockDriverState
, 1);
324 QLIST_INIT(&bs
->dirty_bitmaps
);
325 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
326 QLIST_INIT(&bs
->op_blockers
[i
]);
328 notifier_with_return_list_init(&bs
->before_write_notifiers
);
329 qemu_co_mutex_init(&bs
->reqs_lock
);
330 qemu_mutex_init(&bs
->dirty_bitmap_mutex
);
332 bs
->aio_context
= qemu_get_aio_context();
334 qemu_co_queue_init(&bs
->flush_queue
);
336 for (i
= 0; i
< bdrv_drain_all_count
; i
++) {
337 bdrv_drained_begin(bs
);
340 QTAILQ_INSERT_TAIL(&all_bdrv_states
, bs
, bs_list
);
345 static BlockDriver
*bdrv_do_find_format(const char *format_name
)
349 QLIST_FOREACH(drv1
, &bdrv_drivers
, list
) {
350 if (!strcmp(drv1
->format_name
, format_name
)) {
358 BlockDriver
*bdrv_find_format(const char *format_name
)
363 drv1
= bdrv_do_find_format(format_name
);
368 /* The driver isn't registered, maybe we need to load a module */
369 for (i
= 0; i
< (int)ARRAY_SIZE(block_driver_modules
); ++i
) {
370 if (!strcmp(block_driver_modules
[i
].format_name
, format_name
)) {
371 block_module_load_one(block_driver_modules
[i
].library_name
);
376 return bdrv_do_find_format(format_name
);
379 int bdrv_is_whitelisted(BlockDriver
*drv
, bool read_only
)
381 static const char *whitelist_rw
[] = {
382 CONFIG_BDRV_RW_WHITELIST
384 static const char *whitelist_ro
[] = {
385 CONFIG_BDRV_RO_WHITELIST
389 if (!whitelist_rw
[0] && !whitelist_ro
[0]) {
390 return 1; /* no whitelist, anything goes */
393 for (p
= whitelist_rw
; *p
; p
++) {
394 if (!strcmp(drv
->format_name
, *p
)) {
399 for (p
= whitelist_ro
; *p
; p
++) {
400 if (!strcmp(drv
->format_name
, *p
)) {
408 bool bdrv_uses_whitelist(void)
410 return use_bdrv_whitelist
;
413 typedef struct CreateCo
{
421 static void coroutine_fn
bdrv_create_co_entry(void *opaque
)
423 Error
*local_err
= NULL
;
426 CreateCo
*cco
= opaque
;
429 ret
= cco
->drv
->bdrv_co_create_opts(cco
->filename
, cco
->opts
, &local_err
);
430 error_propagate(&cco
->err
, local_err
);
434 int bdrv_create(BlockDriver
*drv
, const char* filename
,
435 QemuOpts
*opts
, Error
**errp
)
442 .filename
= g_strdup(filename
),
448 if (!drv
->bdrv_co_create_opts
) {
449 error_setg(errp
, "Driver '%s' does not support image creation", drv
->format_name
);
454 if (qemu_in_coroutine()) {
455 /* Fast-path if already in coroutine context */
456 bdrv_create_co_entry(&cco
);
458 co
= qemu_coroutine_create(bdrv_create_co_entry
, &cco
);
459 qemu_coroutine_enter(co
);
460 while (cco
.ret
== NOT_DONE
) {
461 aio_poll(qemu_get_aio_context(), true);
468 error_propagate(errp
, cco
.err
);
470 error_setg_errno(errp
, -ret
, "Could not create image");
475 g_free(cco
.filename
);
479 int bdrv_create_file(const char *filename
, QemuOpts
*opts
, Error
**errp
)
482 Error
*local_err
= NULL
;
485 drv
= bdrv_find_protocol(filename
, true, errp
);
490 ret
= bdrv_create(drv
, filename
, opts
, &local_err
);
491 error_propagate(errp
, local_err
);
496 * Try to get @bs's logical and physical block size.
497 * On success, store them in @bsz struct and return 0.
498 * On failure return -errno.
499 * @bs must not be empty.
501 int bdrv_probe_blocksizes(BlockDriverState
*bs
, BlockSizes
*bsz
)
503 BlockDriver
*drv
= bs
->drv
;
505 if (drv
&& drv
->bdrv_probe_blocksizes
) {
506 return drv
->bdrv_probe_blocksizes(bs
, bsz
);
507 } else if (drv
&& drv
->is_filter
&& bs
->file
) {
508 return bdrv_probe_blocksizes(bs
->file
->bs
, bsz
);
515 * Try to get @bs's geometry (cyls, heads, sectors).
516 * On success, store them in @geo struct and return 0.
517 * On failure return -errno.
518 * @bs must not be empty.
520 int bdrv_probe_geometry(BlockDriverState
*bs
, HDGeometry
*geo
)
522 BlockDriver
*drv
= bs
->drv
;
524 if (drv
&& drv
->bdrv_probe_geometry
) {
525 return drv
->bdrv_probe_geometry(bs
, geo
);
526 } else if (drv
&& drv
->is_filter
&& bs
->file
) {
527 return bdrv_probe_geometry(bs
->file
->bs
, geo
);
534 * Create a uniquely-named empty temporary file.
535 * Return 0 upon success, otherwise a negative errno value.
537 int get_tmp_filename(char *filename
, int size
)
540 char temp_dir
[MAX_PATH
];
541 /* GetTempFileName requires that its output buffer (4th param)
542 have length MAX_PATH or greater. */
543 assert(size
>= MAX_PATH
);
544 return (GetTempPath(MAX_PATH
, temp_dir
)
545 && GetTempFileName(temp_dir
, "qem", 0, filename
)
546 ? 0 : -GetLastError());
550 tmpdir
= getenv("TMPDIR");
554 if (snprintf(filename
, size
, "%s/vl.XXXXXX", tmpdir
) >= size
) {
557 fd
= mkstemp(filename
);
561 if (close(fd
) != 0) {
570 * Detect host devices. By convention, /dev/cdrom[N] is always
571 * recognized as a host CDROM.
573 static BlockDriver
*find_hdev_driver(const char *filename
)
575 int score_max
= 0, score
;
576 BlockDriver
*drv
= NULL
, *d
;
578 QLIST_FOREACH(d
, &bdrv_drivers
, list
) {
579 if (d
->bdrv_probe_device
) {
580 score
= d
->bdrv_probe_device(filename
);
581 if (score
> score_max
) {
591 static BlockDriver
*bdrv_do_find_protocol(const char *protocol
)
595 QLIST_FOREACH(drv1
, &bdrv_drivers
, list
) {
596 if (drv1
->protocol_name
&& !strcmp(drv1
->protocol_name
, protocol
)) {
604 BlockDriver
*bdrv_find_protocol(const char *filename
,
605 bool allow_protocol_prefix
,
614 /* TODO Drivers without bdrv_file_open must be specified explicitly */
617 * XXX(hch): we really should not let host device detection
618 * override an explicit protocol specification, but moving this
619 * later breaks access to device names with colons in them.
620 * Thanks to the brain-dead persistent naming schemes on udev-
621 * based Linux systems those actually are quite common.
623 drv1
= find_hdev_driver(filename
);
628 if (!path_has_protocol(filename
) || !allow_protocol_prefix
) {
632 p
= strchr(filename
, ':');
635 if (len
> sizeof(protocol
) - 1)
636 len
= sizeof(protocol
) - 1;
637 memcpy(protocol
, filename
, len
);
638 protocol
[len
] = '\0';
640 drv1
= bdrv_do_find_protocol(protocol
);
645 for (i
= 0; i
< (int)ARRAY_SIZE(block_driver_modules
); ++i
) {
646 if (block_driver_modules
[i
].protocol_name
&&
647 !strcmp(block_driver_modules
[i
].protocol_name
, protocol
)) {
648 block_module_load_one(block_driver_modules
[i
].library_name
);
653 drv1
= bdrv_do_find_protocol(protocol
);
655 error_setg(errp
, "Unknown protocol '%s'", protocol
);
661 * Guess image format by probing its contents.
662 * This is not a good idea when your image is raw (CVE-2008-2004), but
663 * we do it anyway for backward compatibility.
665 * @buf contains the image's first @buf_size bytes.
666 * @buf_size is the buffer size in bytes (generally BLOCK_PROBE_BUF_SIZE,
667 * but can be smaller if the image file is smaller)
668 * @filename is its filename.
670 * For all block drivers, call the bdrv_probe() method to get its
672 * Return the first block driver with the highest probing score.
674 BlockDriver
*bdrv_probe_all(const uint8_t *buf
, int buf_size
,
675 const char *filename
)
677 int score_max
= 0, score
;
678 BlockDriver
*drv
= NULL
, *d
;
680 QLIST_FOREACH(d
, &bdrv_drivers
, list
) {
682 score
= d
->bdrv_probe(buf
, buf_size
, filename
);
683 if (score
> score_max
) {
693 static int find_image_format(BlockBackend
*file
, const char *filename
,
694 BlockDriver
**pdrv
, Error
**errp
)
697 uint8_t buf
[BLOCK_PROBE_BUF_SIZE
];
700 /* Return the raw BlockDriver * to scsi-generic devices or empty drives */
701 if (blk_is_sg(file
) || !blk_is_inserted(file
) || blk_getlength(file
) == 0) {
706 ret
= blk_pread(file
, 0, buf
, sizeof(buf
));
708 error_setg_errno(errp
, -ret
, "Could not read image for determining its "
714 drv
= bdrv_probe_all(buf
, ret
, filename
);
716 error_setg(errp
, "Could not determine image format: No compatible "
725 * Set the current 'total_sectors' value
726 * Return 0 on success, -errno on error.
728 int refresh_total_sectors(BlockDriverState
*bs
, int64_t hint
)
730 BlockDriver
*drv
= bs
->drv
;
736 /* Do not attempt drv->bdrv_getlength() on scsi-generic devices */
740 /* query actual device if possible, otherwise just trust the hint */
741 if (drv
->bdrv_getlength
) {
742 int64_t length
= drv
->bdrv_getlength(bs
);
746 hint
= DIV_ROUND_UP(length
, BDRV_SECTOR_SIZE
);
749 bs
->total_sectors
= hint
;
754 * Combines a QDict of new block driver @options with any missing options taken
755 * from @old_options, so that leaving out an option defaults to its old value.
757 static void bdrv_join_options(BlockDriverState
*bs
, QDict
*options
,
760 if (bs
->drv
&& bs
->drv
->bdrv_join_options
) {
761 bs
->drv
->bdrv_join_options(options
, old_options
);
763 qdict_join(options
, old_options
, false);
768 * Set open flags for a given discard mode
770 * Return 0 on success, -1 if the discard mode was invalid.
772 int bdrv_parse_discard_flags(const char *mode
, int *flags
)
774 *flags
&= ~BDRV_O_UNMAP
;
776 if (!strcmp(mode
, "off") || !strcmp(mode
, "ignore")) {
778 } else if (!strcmp(mode
, "on") || !strcmp(mode
, "unmap")) {
779 *flags
|= BDRV_O_UNMAP
;
788 * Set open flags for a given cache mode
790 * Return 0 on success, -1 if the cache mode was invalid.
792 int bdrv_parse_cache_mode(const char *mode
, int *flags
, bool *writethrough
)
794 *flags
&= ~BDRV_O_CACHE_MASK
;
796 if (!strcmp(mode
, "off") || !strcmp(mode
, "none")) {
797 *writethrough
= false;
798 *flags
|= BDRV_O_NOCACHE
;
799 } else if (!strcmp(mode
, "directsync")) {
800 *writethrough
= true;
801 *flags
|= BDRV_O_NOCACHE
;
802 } else if (!strcmp(mode
, "writeback")) {
803 *writethrough
= false;
804 } else if (!strcmp(mode
, "unsafe")) {
805 *writethrough
= false;
806 *flags
|= BDRV_O_NO_FLUSH
;
807 } else if (!strcmp(mode
, "writethrough")) {
808 *writethrough
= true;
816 static char *bdrv_child_get_parent_desc(BdrvChild
*c
)
818 BlockDriverState
*parent
= c
->opaque
;
819 return g_strdup(bdrv_get_device_or_node_name(parent
));
822 static void bdrv_child_cb_drained_begin(BdrvChild
*child
)
824 BlockDriverState
*bs
= child
->opaque
;
825 bdrv_do_drained_begin_quiesce(bs
, NULL
, false);
828 static bool bdrv_child_cb_drained_poll(BdrvChild
*child
)
830 BlockDriverState
*bs
= child
->opaque
;
831 return bdrv_drain_poll(bs
, false, NULL
, false);
834 static void bdrv_child_cb_drained_end(BdrvChild
*child
)
836 BlockDriverState
*bs
= child
->opaque
;
837 bdrv_drained_end(bs
);
840 static void bdrv_child_cb_attach(BdrvChild
*child
)
842 BlockDriverState
*bs
= child
->opaque
;
843 bdrv_apply_subtree_drain(child
, bs
);
846 static void bdrv_child_cb_detach(BdrvChild
*child
)
848 BlockDriverState
*bs
= child
->opaque
;
849 bdrv_unapply_subtree_drain(child
, bs
);
852 static int bdrv_child_cb_inactivate(BdrvChild
*child
)
854 BlockDriverState
*bs
= child
->opaque
;
855 assert(bs
->open_flags
& BDRV_O_INACTIVE
);
860 * Returns the options and flags that a temporary snapshot should get, based on
861 * the originally requested flags (the originally requested image will have
862 * flags like a backing file)
864 static void bdrv_temp_snapshot_options(int *child_flags
, QDict
*child_options
,
865 int parent_flags
, QDict
*parent_options
)
867 *child_flags
= (parent_flags
& ~BDRV_O_SNAPSHOT
) | BDRV_O_TEMPORARY
;
869 /* For temporary files, unconditional cache=unsafe is fine */
870 qdict_set_default_str(child_options
, BDRV_OPT_CACHE_DIRECT
, "off");
871 qdict_set_default_str(child_options
, BDRV_OPT_CACHE_NO_FLUSH
, "on");
873 /* Copy the read-only option from the parent */
874 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_READ_ONLY
);
876 /* aio=native doesn't work for cache.direct=off, so disable it for the
877 * temporary snapshot */
878 *child_flags
&= ~BDRV_O_NATIVE_AIO
;
882 * Returns the options and flags that bs->file should get if a protocol driver
883 * is expected, based on the given options and flags for the parent BDS
885 static void bdrv_inherited_options(int *child_flags
, QDict
*child_options
,
886 int parent_flags
, QDict
*parent_options
)
888 int flags
= parent_flags
;
890 /* Enable protocol handling, disable format probing for bs->file */
891 flags
|= BDRV_O_PROTOCOL
;
893 /* If the cache mode isn't explicitly set, inherit direct and no-flush from
895 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_CACHE_DIRECT
);
896 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_CACHE_NO_FLUSH
);
897 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_FORCE_SHARE
);
899 /* Inherit the read-only option from the parent if it's not set */
900 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_READ_ONLY
);
902 /* Our block drivers take care to send flushes and respect unmap policy,
903 * so we can default to enable both on lower layers regardless of the
904 * corresponding parent options. */
905 qdict_set_default_str(child_options
, BDRV_OPT_DISCARD
, "unmap");
907 /* Clear flags that only apply to the top layer */
908 flags
&= ~(BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
| BDRV_O_COPY_ON_READ
|
911 *child_flags
= flags
;
914 const BdrvChildRole child_file
= {
915 .parent_is_bds
= true,
916 .get_parent_desc
= bdrv_child_get_parent_desc
,
917 .inherit_options
= bdrv_inherited_options
,
918 .drained_begin
= bdrv_child_cb_drained_begin
,
919 .drained_poll
= bdrv_child_cb_drained_poll
,
920 .drained_end
= bdrv_child_cb_drained_end
,
921 .attach
= bdrv_child_cb_attach
,
922 .detach
= bdrv_child_cb_detach
,
923 .inactivate
= bdrv_child_cb_inactivate
,
927 * Returns the options and flags that bs->file should get if the use of formats
928 * (and not only protocols) is permitted for it, based on the given options and
929 * flags for the parent BDS
931 static void bdrv_inherited_fmt_options(int *child_flags
, QDict
*child_options
,
932 int parent_flags
, QDict
*parent_options
)
934 child_file
.inherit_options(child_flags
, child_options
,
935 parent_flags
, parent_options
);
937 *child_flags
&= ~(BDRV_O_PROTOCOL
| BDRV_O_NO_IO
);
940 const BdrvChildRole child_format
= {
941 .parent_is_bds
= true,
942 .get_parent_desc
= bdrv_child_get_parent_desc
,
943 .inherit_options
= bdrv_inherited_fmt_options
,
944 .drained_begin
= bdrv_child_cb_drained_begin
,
945 .drained_poll
= bdrv_child_cb_drained_poll
,
946 .drained_end
= bdrv_child_cb_drained_end
,
947 .attach
= bdrv_child_cb_attach
,
948 .detach
= bdrv_child_cb_detach
,
949 .inactivate
= bdrv_child_cb_inactivate
,
952 static void bdrv_backing_attach(BdrvChild
*c
)
954 BlockDriverState
*parent
= c
->opaque
;
955 BlockDriverState
*backing_hd
= c
->bs
;
957 assert(!parent
->backing_blocker
);
958 error_setg(&parent
->backing_blocker
,
959 "node is used as backing hd of '%s'",
960 bdrv_get_device_or_node_name(parent
));
962 parent
->open_flags
&= ~BDRV_O_NO_BACKING
;
963 pstrcpy(parent
->backing_file
, sizeof(parent
->backing_file
),
964 backing_hd
->filename
);
965 pstrcpy(parent
->backing_format
, sizeof(parent
->backing_format
),
966 backing_hd
->drv
? backing_hd
->drv
->format_name
: "");
968 bdrv_op_block_all(backing_hd
, parent
->backing_blocker
);
969 /* Otherwise we won't be able to commit or stream */
970 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_COMMIT_TARGET
,
971 parent
->backing_blocker
);
972 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_STREAM
,
973 parent
->backing_blocker
);
975 * We do backup in 3 ways:
977 * The target bs is new opened, and the source is top BDS
979 * Both the source and the target are top BDSes.
980 * 3. internal backup(used for block replication)
981 * Both the source and the target are backing file
983 * In case 1 and 2, neither the source nor the target is the backing file.
984 * In case 3, we will block the top BDS, so there is only one block job
985 * for the top BDS and its backing chain.
987 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_BACKUP_SOURCE
,
988 parent
->backing_blocker
);
989 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_BACKUP_TARGET
,
990 parent
->backing_blocker
);
992 bdrv_child_cb_attach(c
);
995 static void bdrv_backing_detach(BdrvChild
*c
)
997 BlockDriverState
*parent
= c
->opaque
;
999 assert(parent
->backing_blocker
);
1000 bdrv_op_unblock_all(c
->bs
, parent
->backing_blocker
);
1001 error_free(parent
->backing_blocker
);
1002 parent
->backing_blocker
= NULL
;
1004 bdrv_child_cb_detach(c
);
1008 * Returns the options and flags that bs->backing should get, based on the
1009 * given options and flags for the parent BDS
1011 static void bdrv_backing_options(int *child_flags
, QDict
*child_options
,
1012 int parent_flags
, QDict
*parent_options
)
1014 int flags
= parent_flags
;
1016 /* The cache mode is inherited unmodified for backing files; except WCE,
1017 * which is only applied on the top level (BlockBackend) */
1018 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_CACHE_DIRECT
);
1019 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_CACHE_NO_FLUSH
);
1020 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_FORCE_SHARE
);
1022 /* backing files always opened read-only */
1023 qdict_set_default_str(child_options
, BDRV_OPT_READ_ONLY
, "on");
1024 flags
&= ~BDRV_O_COPY_ON_READ
;
1026 /* snapshot=on is handled on the top layer */
1027 flags
&= ~(BDRV_O_SNAPSHOT
| BDRV_O_TEMPORARY
);
1029 *child_flags
= flags
;
1032 static int bdrv_backing_update_filename(BdrvChild
*c
, BlockDriverState
*base
,
1033 const char *filename
, Error
**errp
)
1035 BlockDriverState
*parent
= c
->opaque
;
1036 int orig_flags
= bdrv_get_flags(parent
);
1039 if (!(orig_flags
& BDRV_O_RDWR
)) {
1040 ret
= bdrv_reopen(parent
, orig_flags
| BDRV_O_RDWR
, errp
);
1046 ret
= bdrv_change_backing_file(parent
, filename
,
1047 base
->drv
? base
->drv
->format_name
: "");
1049 error_setg_errno(errp
, -ret
, "Could not update backing file link");
1052 if (!(orig_flags
& BDRV_O_RDWR
)) {
1053 bdrv_reopen(parent
, orig_flags
, NULL
);
1059 const BdrvChildRole child_backing
= {
1060 .parent_is_bds
= true,
1061 .get_parent_desc
= bdrv_child_get_parent_desc
,
1062 .attach
= bdrv_backing_attach
,
1063 .detach
= bdrv_backing_detach
,
1064 .inherit_options
= bdrv_backing_options
,
1065 .drained_begin
= bdrv_child_cb_drained_begin
,
1066 .drained_poll
= bdrv_child_cb_drained_poll
,
1067 .drained_end
= bdrv_child_cb_drained_end
,
1068 .inactivate
= bdrv_child_cb_inactivate
,
1069 .update_filename
= bdrv_backing_update_filename
,
1072 static int bdrv_open_flags(BlockDriverState
*bs
, int flags
)
1074 int open_flags
= flags
;
1077 * Clear flags that are internal to the block layer before opening the
1080 open_flags
&= ~(BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
| BDRV_O_PROTOCOL
);
1083 * Snapshots should be writable.
1085 if (flags
& BDRV_O_TEMPORARY
) {
1086 open_flags
|= BDRV_O_RDWR
;
1092 static void update_flags_from_options(int *flags
, QemuOpts
*opts
)
1094 *flags
&= ~BDRV_O_CACHE_MASK
;
1096 assert(qemu_opt_find(opts
, BDRV_OPT_CACHE_NO_FLUSH
));
1097 if (qemu_opt_get_bool_del(opts
, BDRV_OPT_CACHE_NO_FLUSH
, false)) {
1098 *flags
|= BDRV_O_NO_FLUSH
;
1101 assert(qemu_opt_find(opts
, BDRV_OPT_CACHE_DIRECT
));
1102 if (qemu_opt_get_bool_del(opts
, BDRV_OPT_CACHE_DIRECT
, false)) {
1103 *flags
|= BDRV_O_NOCACHE
;
1106 *flags
&= ~BDRV_O_RDWR
;
1108 assert(qemu_opt_find(opts
, BDRV_OPT_READ_ONLY
));
1109 if (!qemu_opt_get_bool_del(opts
, BDRV_OPT_READ_ONLY
, false)) {
1110 *flags
|= BDRV_O_RDWR
;
1115 static void update_options_from_flags(QDict
*options
, int flags
)
1117 if (!qdict_haskey(options
, BDRV_OPT_CACHE_DIRECT
)) {
1118 qdict_put_bool(options
, BDRV_OPT_CACHE_DIRECT
, flags
& BDRV_O_NOCACHE
);
1120 if (!qdict_haskey(options
, BDRV_OPT_CACHE_NO_FLUSH
)) {
1121 qdict_put_bool(options
, BDRV_OPT_CACHE_NO_FLUSH
,
1122 flags
& BDRV_O_NO_FLUSH
);
1124 if (!qdict_haskey(options
, BDRV_OPT_READ_ONLY
)) {
1125 qdict_put_bool(options
, BDRV_OPT_READ_ONLY
, !(flags
& BDRV_O_RDWR
));
1129 static void bdrv_assign_node_name(BlockDriverState
*bs
,
1130 const char *node_name
,
1133 char *gen_node_name
= NULL
;
1136 node_name
= gen_node_name
= id_generate(ID_BLOCK
);
1137 } else if (!id_wellformed(node_name
)) {
1139 * Check for empty string or invalid characters, but not if it is
1140 * generated (generated names use characters not available to the user)
1142 error_setg(errp
, "Invalid node name");
1146 /* takes care of avoiding namespaces collisions */
1147 if (blk_by_name(node_name
)) {
1148 error_setg(errp
, "node-name=%s is conflicting with a device id",
1153 /* takes care of avoiding duplicates node names */
1154 if (bdrv_find_node(node_name
)) {
1155 error_setg(errp
, "Duplicate node name");
1159 /* Make sure that the node name isn't truncated */
1160 if (strlen(node_name
) >= sizeof(bs
->node_name
)) {
1161 error_setg(errp
, "Node name too long");
1165 /* copy node name into the bs and insert it into the graph list */
1166 pstrcpy(bs
->node_name
, sizeof(bs
->node_name
), node_name
);
1167 QTAILQ_INSERT_TAIL(&graph_bdrv_states
, bs
, node_list
);
1169 g_free(gen_node_name
);
1172 static int bdrv_open_driver(BlockDriverState
*bs
, BlockDriver
*drv
,
1173 const char *node_name
, QDict
*options
,
1174 int open_flags
, Error
**errp
)
1176 Error
*local_err
= NULL
;
1179 bdrv_assign_node_name(bs
, node_name
, &local_err
);
1181 error_propagate(errp
, local_err
);
1186 bs
->read_only
= !(bs
->open_flags
& BDRV_O_RDWR
);
1187 bs
->opaque
= g_malloc0(drv
->instance_size
);
1189 if (drv
->bdrv_file_open
) {
1190 assert(!drv
->bdrv_needs_filename
|| bs
->filename
[0]);
1191 ret
= drv
->bdrv_file_open(bs
, options
, open_flags
, &local_err
);
1192 } else if (drv
->bdrv_open
) {
1193 ret
= drv
->bdrv_open(bs
, options
, open_flags
, &local_err
);
1200 error_propagate(errp
, local_err
);
1201 } else if (bs
->filename
[0]) {
1202 error_setg_errno(errp
, -ret
, "Could not open '%s'", bs
->filename
);
1204 error_setg_errno(errp
, -ret
, "Could not open image");
1209 ret
= refresh_total_sectors(bs
, bs
->total_sectors
);
1211 error_setg_errno(errp
, -ret
, "Could not refresh total sector count");
1215 bdrv_refresh_limits(bs
, &local_err
);
1217 error_propagate(errp
, local_err
);
1221 assert(bdrv_opt_mem_align(bs
) != 0);
1222 assert(bdrv_min_mem_align(bs
) != 0);
1223 assert(is_power_of_2(bs
->bl
.request_alignment
));
1225 for (i
= 0; i
< bs
->quiesce_counter
; i
++) {
1226 if (drv
->bdrv_co_drain_begin
) {
1227 drv
->bdrv_co_drain_begin(bs
);
1234 if (bs
->file
!= NULL
) {
1235 bdrv_unref_child(bs
, bs
->file
);
1243 BlockDriverState
*bdrv_new_open_driver(BlockDriver
*drv
, const char *node_name
,
1244 int flags
, Error
**errp
)
1246 BlockDriverState
*bs
;
1250 bs
->open_flags
= flags
;
1251 bs
->explicit_options
= qdict_new();
1252 bs
->options
= qdict_new();
1255 update_options_from_flags(bs
->options
, flags
);
1257 ret
= bdrv_open_driver(bs
, drv
, node_name
, bs
->options
, flags
, errp
);
1259 qobject_unref(bs
->explicit_options
);
1260 bs
->explicit_options
= NULL
;
1261 qobject_unref(bs
->options
);
1270 QemuOptsList bdrv_runtime_opts
= {
1271 .name
= "bdrv_common",
1272 .head
= QTAILQ_HEAD_INITIALIZER(bdrv_runtime_opts
.head
),
1275 .name
= "node-name",
1276 .type
= QEMU_OPT_STRING
,
1277 .help
= "Node name of the block device node",
1281 .type
= QEMU_OPT_STRING
,
1282 .help
= "Block driver to use for the node",
1285 .name
= BDRV_OPT_CACHE_DIRECT
,
1286 .type
= QEMU_OPT_BOOL
,
1287 .help
= "Bypass software writeback cache on the host",
1290 .name
= BDRV_OPT_CACHE_NO_FLUSH
,
1291 .type
= QEMU_OPT_BOOL
,
1292 .help
= "Ignore flush requests",
1295 .name
= BDRV_OPT_READ_ONLY
,
1296 .type
= QEMU_OPT_BOOL
,
1297 .help
= "Node is opened in read-only mode",
1300 .name
= "detect-zeroes",
1301 .type
= QEMU_OPT_STRING
,
1302 .help
= "try to optimize zero writes (off, on, unmap)",
1306 .type
= QEMU_OPT_STRING
,
1307 .help
= "discard operation (ignore/off, unmap/on)",
1310 .name
= BDRV_OPT_FORCE_SHARE
,
1311 .type
= QEMU_OPT_BOOL
,
1312 .help
= "always accept other writers (default: off)",
1314 { /* end of list */ }
1319 * Common part for opening disk images and files
1321 * Removes all processed options from *options.
1323 static int bdrv_open_common(BlockDriverState
*bs
, BlockBackend
*file
,
1324 QDict
*options
, Error
**errp
)
1326 int ret
, open_flags
;
1327 const char *filename
;
1328 const char *driver_name
= NULL
;
1329 const char *node_name
= NULL
;
1330 const char *discard
;
1331 const char *detect_zeroes
;
1334 Error
*local_err
= NULL
;
1336 assert(bs
->file
== NULL
);
1337 assert(options
!= NULL
&& bs
->options
!= options
);
1339 opts
= qemu_opts_create(&bdrv_runtime_opts
, NULL
, 0, &error_abort
);
1340 qemu_opts_absorb_qdict(opts
, options
, &local_err
);
1342 error_propagate(errp
, local_err
);
1347 update_flags_from_options(&bs
->open_flags
, opts
);
1349 driver_name
= qemu_opt_get(opts
, "driver");
1350 drv
= bdrv_find_format(driver_name
);
1351 assert(drv
!= NULL
);
1353 bs
->force_share
= qemu_opt_get_bool(opts
, BDRV_OPT_FORCE_SHARE
, false);
1355 if (bs
->force_share
&& (bs
->open_flags
& BDRV_O_RDWR
)) {
1357 BDRV_OPT_FORCE_SHARE
1358 "=on can only be used with read-only images");
1364 filename
= blk_bs(file
)->filename
;
1367 * Caution: while qdict_get_try_str() is fine, getting
1368 * non-string types would require more care. When @options
1369 * come from -blockdev or blockdev_add, its members are typed
1370 * according to the QAPI schema, but when they come from
1371 * -drive, they're all QString.
1373 filename
= qdict_get_try_str(options
, "filename");
1376 if (drv
->bdrv_needs_filename
&& (!filename
|| !filename
[0])) {
1377 error_setg(errp
, "The '%s' block driver requires a file name",
1383 trace_bdrv_open_common(bs
, filename
?: "", bs
->open_flags
,
1386 bs
->read_only
= !(bs
->open_flags
& BDRV_O_RDWR
);
1388 if (use_bdrv_whitelist
&& !bdrv_is_whitelisted(drv
, bs
->read_only
)) {
1390 !bs
->read_only
&& bdrv_is_whitelisted(drv
, true)
1391 ? "Driver '%s' can only be used for read-only devices"
1392 : "Driver '%s' is not whitelisted",
1398 /* bdrv_new() and bdrv_close() make it so */
1399 assert(atomic_read(&bs
->copy_on_read
) == 0);
1401 if (bs
->open_flags
& BDRV_O_COPY_ON_READ
) {
1402 if (!bs
->read_only
) {
1403 bdrv_enable_copy_on_read(bs
);
1405 error_setg(errp
, "Can't use copy-on-read on read-only device");
1411 discard
= qemu_opt_get(opts
, "discard");
1412 if (discard
!= NULL
) {
1413 if (bdrv_parse_discard_flags(discard
, &bs
->open_flags
) != 0) {
1414 error_setg(errp
, "Invalid discard option");
1420 detect_zeroes
= qemu_opt_get(opts
, "detect-zeroes");
1421 if (detect_zeroes
) {
1422 BlockdevDetectZeroesOptions value
=
1423 qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup
,
1425 BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF
,
1428 error_propagate(errp
, local_err
);
1433 if (value
== BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP
&&
1434 !(bs
->open_flags
& BDRV_O_UNMAP
))
1436 error_setg(errp
, "setting detect-zeroes to unmap is not allowed "
1437 "without setting discard operation to unmap");
1442 bs
->detect_zeroes
= value
;
1445 if (filename
!= NULL
) {
1446 pstrcpy(bs
->filename
, sizeof(bs
->filename
), filename
);
1448 bs
->filename
[0] = '\0';
1450 pstrcpy(bs
->exact_filename
, sizeof(bs
->exact_filename
), bs
->filename
);
1452 /* Open the image, either directly or using a protocol */
1453 open_flags
= bdrv_open_flags(bs
, bs
->open_flags
);
1454 node_name
= qemu_opt_get(opts
, "node-name");
1456 assert(!drv
->bdrv_file_open
|| file
== NULL
);
1457 ret
= bdrv_open_driver(bs
, drv
, node_name
, options
, open_flags
, errp
);
1462 qemu_opts_del(opts
);
1466 qemu_opts_del(opts
);
1470 static QDict
*parse_json_filename(const char *filename
, Error
**errp
)
1472 QObject
*options_obj
;
1476 ret
= strstart(filename
, "json:", &filename
);
1479 options_obj
= qobject_from_json(filename
, errp
);
1481 error_prepend(errp
, "Could not parse the JSON options: ");
1485 options
= qobject_to(QDict
, options_obj
);
1487 qobject_unref(options_obj
);
1488 error_setg(errp
, "Invalid JSON object given");
1492 qdict_flatten(options
);
1497 static void parse_json_protocol(QDict
*options
, const char **pfilename
,
1500 QDict
*json_options
;
1501 Error
*local_err
= NULL
;
1503 /* Parse json: pseudo-protocol */
1504 if (!*pfilename
|| !g_str_has_prefix(*pfilename
, "json:")) {
1508 json_options
= parse_json_filename(*pfilename
, &local_err
);
1510 error_propagate(errp
, local_err
);
1514 /* Options given in the filename have lower priority than options
1515 * specified directly */
1516 qdict_join(options
, json_options
, false);
1517 qobject_unref(json_options
);
1522 * Fills in default options for opening images and converts the legacy
1523 * filename/flags pair to option QDict entries.
1524 * The BDRV_O_PROTOCOL flag in *flags will be set or cleared accordingly if a
1525 * block driver has been specified explicitly.
1527 static int bdrv_fill_options(QDict
**options
, const char *filename
,
1528 int *flags
, Error
**errp
)
1530 const char *drvname
;
1531 bool protocol
= *flags
& BDRV_O_PROTOCOL
;
1532 bool parse_filename
= false;
1533 BlockDriver
*drv
= NULL
;
1534 Error
*local_err
= NULL
;
1537 * Caution: while qdict_get_try_str() is fine, getting non-string
1538 * types would require more care. When @options come from
1539 * -blockdev or blockdev_add, its members are typed according to
1540 * the QAPI schema, but when they come from -drive, they're all
1543 drvname
= qdict_get_try_str(*options
, "driver");
1545 drv
= bdrv_find_format(drvname
);
1547 error_setg(errp
, "Unknown driver '%s'", drvname
);
1550 /* If the user has explicitly specified the driver, this choice should
1551 * override the BDRV_O_PROTOCOL flag */
1552 protocol
= drv
->bdrv_file_open
;
1556 *flags
|= BDRV_O_PROTOCOL
;
1558 *flags
&= ~BDRV_O_PROTOCOL
;
1561 /* Translate cache options from flags into options */
1562 update_options_from_flags(*options
, *flags
);
1564 /* Fetch the file name from the options QDict if necessary */
1565 if (protocol
&& filename
) {
1566 if (!qdict_haskey(*options
, "filename")) {
1567 qdict_put_str(*options
, "filename", filename
);
1568 parse_filename
= true;
1570 error_setg(errp
, "Can't specify 'file' and 'filename' options at "
1576 /* Find the right block driver */
1577 /* See cautionary note on accessing @options above */
1578 filename
= qdict_get_try_str(*options
, "filename");
1580 if (!drvname
&& protocol
) {
1582 drv
= bdrv_find_protocol(filename
, parse_filename
, errp
);
1587 drvname
= drv
->format_name
;
1588 qdict_put_str(*options
, "driver", drvname
);
1590 error_setg(errp
, "Must specify either driver or file");
1595 assert(drv
|| !protocol
);
1597 /* Driver-specific filename parsing */
1598 if (drv
&& drv
->bdrv_parse_filename
&& parse_filename
) {
1599 drv
->bdrv_parse_filename(filename
, *options
, &local_err
);
1601 error_propagate(errp
, local_err
);
1605 if (!drv
->bdrv_needs_filename
) {
1606 qdict_del(*options
, "filename");
1613 static int bdrv_child_check_perm(BdrvChild
*c
, BlockReopenQueue
*q
,
1614 uint64_t perm
, uint64_t shared
,
1615 GSList
*ignore_children
, Error
**errp
);
1616 static void bdrv_child_abort_perm_update(BdrvChild
*c
);
1617 static void bdrv_child_set_perm(BdrvChild
*c
, uint64_t perm
, uint64_t shared
);
1619 typedef struct BlockReopenQueueEntry
{
1621 BDRVReopenState state
;
1622 QSIMPLEQ_ENTRY(BlockReopenQueueEntry
) entry
;
1623 } BlockReopenQueueEntry
;
1626 * Return the flags that @bs will have after the reopens in @q have
1627 * successfully completed. If @q is NULL (or @bs is not contained in @q),
1628 * return the current flags.
1630 static int bdrv_reopen_get_flags(BlockReopenQueue
*q
, BlockDriverState
*bs
)
1632 BlockReopenQueueEntry
*entry
;
1635 QSIMPLEQ_FOREACH(entry
, q
, entry
) {
1636 if (entry
->state
.bs
== bs
) {
1637 return entry
->state
.flags
;
1642 return bs
->open_flags
;
1645 /* Returns whether the image file can be written to after the reopen queue @q
1646 * has been successfully applied, or right now if @q is NULL. */
1647 static bool bdrv_is_writable_after_reopen(BlockDriverState
*bs
,
1648 BlockReopenQueue
*q
)
1650 int flags
= bdrv_reopen_get_flags(q
, bs
);
1652 return (flags
& (BDRV_O_RDWR
| BDRV_O_INACTIVE
)) == BDRV_O_RDWR
;
1656 * Return whether the BDS can be written to. This is not necessarily
1657 * the same as !bdrv_is_read_only(bs), as inactivated images may not
1658 * be written to but do not count as read-only images.
1660 bool bdrv_is_writable(BlockDriverState
*bs
)
1662 return bdrv_is_writable_after_reopen(bs
, NULL
);
1665 static void bdrv_child_perm(BlockDriverState
*bs
, BlockDriverState
*child_bs
,
1666 BdrvChild
*c
, const BdrvChildRole
*role
,
1667 BlockReopenQueue
*reopen_queue
,
1668 uint64_t parent_perm
, uint64_t parent_shared
,
1669 uint64_t *nperm
, uint64_t *nshared
)
1671 if (bs
->drv
&& bs
->drv
->bdrv_child_perm
) {
1672 bs
->drv
->bdrv_child_perm(bs
, c
, role
, reopen_queue
,
1673 parent_perm
, parent_shared
,
1676 /* TODO Take force_share from reopen_queue */
1677 if (child_bs
&& child_bs
->force_share
) {
1678 *nshared
= BLK_PERM_ALL
;
1683 * Check whether permissions on this node can be changed in a way that
1684 * @cumulative_perms and @cumulative_shared_perms are the new cumulative
1685 * permissions of all its parents. This involves checking whether all necessary
1686 * permission changes to child nodes can be performed.
1688 * A call to this function must always be followed by a call to bdrv_set_perm()
1689 * or bdrv_abort_perm_update().
1691 static int bdrv_check_perm(BlockDriverState
*bs
, BlockReopenQueue
*q
,
1692 uint64_t cumulative_perms
,
1693 uint64_t cumulative_shared_perms
,
1694 GSList
*ignore_children
, Error
**errp
)
1696 BlockDriver
*drv
= bs
->drv
;
1700 /* Write permissions never work with read-only images */
1701 if ((cumulative_perms
& (BLK_PERM_WRITE
| BLK_PERM_WRITE_UNCHANGED
)) &&
1702 !bdrv_is_writable_after_reopen(bs
, q
))
1704 error_setg(errp
, "Block node is read-only");
1708 /* Check this node */
1713 if (drv
->bdrv_check_perm
) {
1714 return drv
->bdrv_check_perm(bs
, cumulative_perms
,
1715 cumulative_shared_perms
, errp
);
1718 /* Drivers that never have children can omit .bdrv_child_perm() */
1719 if (!drv
->bdrv_child_perm
) {
1720 assert(QLIST_EMPTY(&bs
->children
));
1724 /* Check all children */
1725 QLIST_FOREACH(c
, &bs
->children
, next
) {
1726 uint64_t cur_perm
, cur_shared
;
1727 bdrv_child_perm(bs
, c
->bs
, c
, c
->role
, q
,
1728 cumulative_perms
, cumulative_shared_perms
,
1729 &cur_perm
, &cur_shared
);
1730 ret
= bdrv_child_check_perm(c
, q
, cur_perm
, cur_shared
,
1731 ignore_children
, errp
);
1741 * Notifies drivers that after a previous bdrv_check_perm() call, the
1742 * permission update is not performed and any preparations made for it (e.g.
1743 * taken file locks) need to be undone.
1745 * This function recursively notifies all child nodes.
1747 static void bdrv_abort_perm_update(BlockDriverState
*bs
)
1749 BlockDriver
*drv
= bs
->drv
;
1756 if (drv
->bdrv_abort_perm_update
) {
1757 drv
->bdrv_abort_perm_update(bs
);
1760 QLIST_FOREACH(c
, &bs
->children
, next
) {
1761 bdrv_child_abort_perm_update(c
);
1765 static void bdrv_set_perm(BlockDriverState
*bs
, uint64_t cumulative_perms
,
1766 uint64_t cumulative_shared_perms
)
1768 BlockDriver
*drv
= bs
->drv
;
1775 /* Update this node */
1776 if (drv
->bdrv_set_perm
) {
1777 drv
->bdrv_set_perm(bs
, cumulative_perms
, cumulative_shared_perms
);
1780 /* Drivers that never have children can omit .bdrv_child_perm() */
1781 if (!drv
->bdrv_child_perm
) {
1782 assert(QLIST_EMPTY(&bs
->children
));
1786 /* Update all children */
1787 QLIST_FOREACH(c
, &bs
->children
, next
) {
1788 uint64_t cur_perm
, cur_shared
;
1789 bdrv_child_perm(bs
, c
->bs
, c
, c
->role
, NULL
,
1790 cumulative_perms
, cumulative_shared_perms
,
1791 &cur_perm
, &cur_shared
);
1792 bdrv_child_set_perm(c
, cur_perm
, cur_shared
);
1796 static void bdrv_get_cumulative_perm(BlockDriverState
*bs
, uint64_t *perm
,
1797 uint64_t *shared_perm
)
1800 uint64_t cumulative_perms
= 0;
1801 uint64_t cumulative_shared_perms
= BLK_PERM_ALL
;
1803 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
1804 cumulative_perms
|= c
->perm
;
1805 cumulative_shared_perms
&= c
->shared_perm
;
1808 *perm
= cumulative_perms
;
1809 *shared_perm
= cumulative_shared_perms
;
1812 static char *bdrv_child_user_desc(BdrvChild
*c
)
1814 if (c
->role
->get_parent_desc
) {
1815 return c
->role
->get_parent_desc(c
);
1818 return g_strdup("another user");
1821 char *bdrv_perm_names(uint64_t perm
)
1827 { BLK_PERM_CONSISTENT_READ
, "consistent read" },
1828 { BLK_PERM_WRITE
, "write" },
1829 { BLK_PERM_WRITE_UNCHANGED
, "write unchanged" },
1830 { BLK_PERM_RESIZE
, "resize" },
1831 { BLK_PERM_GRAPH_MOD
, "change children" },
1835 char *result
= g_strdup("");
1836 struct perm_name
*p
;
1838 for (p
= permissions
; p
->name
; p
++) {
1839 if (perm
& p
->perm
) {
1841 result
= g_strdup_printf("%s%s%s", old
, *old
? ", " : "", p
->name
);
1850 * Checks whether a new reference to @bs can be added if the new user requires
1851 * @new_used_perm/@new_shared_perm as its permissions. If @ignore_children is
1852 * set, the BdrvChild objects in this list are ignored in the calculations;
1853 * this allows checking permission updates for an existing reference.
1855 * Needs to be followed by a call to either bdrv_set_perm() or
1856 * bdrv_abort_perm_update(). */
1857 static int bdrv_check_update_perm(BlockDriverState
*bs
, BlockReopenQueue
*q
,
1858 uint64_t new_used_perm
,
1859 uint64_t new_shared_perm
,
1860 GSList
*ignore_children
, Error
**errp
)
1863 uint64_t cumulative_perms
= new_used_perm
;
1864 uint64_t cumulative_shared_perms
= new_shared_perm
;
1866 /* There is no reason why anyone couldn't tolerate write_unchanged */
1867 assert(new_shared_perm
& BLK_PERM_WRITE_UNCHANGED
);
1869 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
1870 if (g_slist_find(ignore_children
, c
)) {
1874 if ((new_used_perm
& c
->shared_perm
) != new_used_perm
) {
1875 char *user
= bdrv_child_user_desc(c
);
1876 char *perm_names
= bdrv_perm_names(new_used_perm
& ~c
->shared_perm
);
1877 error_setg(errp
, "Conflicts with use by %s as '%s', which does not "
1879 user
, c
->name
, perm_names
, bdrv_get_node_name(c
->bs
));
1885 if ((c
->perm
& new_shared_perm
) != c
->perm
) {
1886 char *user
= bdrv_child_user_desc(c
);
1887 char *perm_names
= bdrv_perm_names(c
->perm
& ~new_shared_perm
);
1888 error_setg(errp
, "Conflicts with use by %s as '%s', which uses "
1890 user
, c
->name
, perm_names
, bdrv_get_node_name(c
->bs
));
1896 cumulative_perms
|= c
->perm
;
1897 cumulative_shared_perms
&= c
->shared_perm
;
1900 return bdrv_check_perm(bs
, q
, cumulative_perms
, cumulative_shared_perms
,
1901 ignore_children
, errp
);
1904 /* Needs to be followed by a call to either bdrv_child_set_perm() or
1905 * bdrv_child_abort_perm_update(). */
1906 static int bdrv_child_check_perm(BdrvChild
*c
, BlockReopenQueue
*q
,
1907 uint64_t perm
, uint64_t shared
,
1908 GSList
*ignore_children
, Error
**errp
)
1912 ignore_children
= g_slist_prepend(g_slist_copy(ignore_children
), c
);
1913 ret
= bdrv_check_update_perm(c
->bs
, q
, perm
, shared
, ignore_children
, errp
);
1914 g_slist_free(ignore_children
);
1919 static void bdrv_child_set_perm(BdrvChild
*c
, uint64_t perm
, uint64_t shared
)
1921 uint64_t cumulative_perms
, cumulative_shared_perms
;
1924 c
->shared_perm
= shared
;
1926 bdrv_get_cumulative_perm(c
->bs
, &cumulative_perms
,
1927 &cumulative_shared_perms
);
1928 bdrv_set_perm(c
->bs
, cumulative_perms
, cumulative_shared_perms
);
1931 static void bdrv_child_abort_perm_update(BdrvChild
*c
)
1933 bdrv_abort_perm_update(c
->bs
);
1936 int bdrv_child_try_set_perm(BdrvChild
*c
, uint64_t perm
, uint64_t shared
,
1941 ret
= bdrv_child_check_perm(c
, NULL
, perm
, shared
, NULL
, errp
);
1943 bdrv_child_abort_perm_update(c
);
1947 bdrv_child_set_perm(c
, perm
, shared
);
1952 void bdrv_filter_default_perms(BlockDriverState
*bs
, BdrvChild
*c
,
1953 const BdrvChildRole
*role
,
1954 BlockReopenQueue
*reopen_queue
,
1955 uint64_t perm
, uint64_t shared
,
1956 uint64_t *nperm
, uint64_t *nshared
)
1959 *nperm
= perm
& DEFAULT_PERM_PASSTHROUGH
;
1960 *nshared
= (shared
& DEFAULT_PERM_PASSTHROUGH
) | DEFAULT_PERM_UNCHANGED
;
1964 *nperm
= (perm
& DEFAULT_PERM_PASSTHROUGH
) |
1965 (c
->perm
& DEFAULT_PERM_UNCHANGED
);
1966 *nshared
= (shared
& DEFAULT_PERM_PASSTHROUGH
) |
1967 (c
->shared_perm
& DEFAULT_PERM_UNCHANGED
);
1970 void bdrv_format_default_perms(BlockDriverState
*bs
, BdrvChild
*c
,
1971 const BdrvChildRole
*role
,
1972 BlockReopenQueue
*reopen_queue
,
1973 uint64_t perm
, uint64_t shared
,
1974 uint64_t *nperm
, uint64_t *nshared
)
1976 bool backing
= (role
== &child_backing
);
1977 assert(role
== &child_backing
|| role
== &child_file
);
1980 int flags
= bdrv_reopen_get_flags(reopen_queue
, bs
);
1982 /* Apart from the modifications below, the same permissions are
1983 * forwarded and left alone as for filters */
1984 bdrv_filter_default_perms(bs
, c
, role
, reopen_queue
, perm
, shared
,
1987 /* Format drivers may touch metadata even if the guest doesn't write */
1988 if (bdrv_is_writable_after_reopen(bs
, reopen_queue
)) {
1989 perm
|= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
1992 /* bs->file always needs to be consistent because of the metadata. We
1993 * can never allow other users to resize or write to it. */
1994 if (!(flags
& BDRV_O_NO_IO
)) {
1995 perm
|= BLK_PERM_CONSISTENT_READ
;
1997 shared
&= ~(BLK_PERM_WRITE
| BLK_PERM_RESIZE
);
1999 /* We want consistent read from backing files if the parent needs it.
2000 * No other operations are performed on backing files. */
2001 perm
&= BLK_PERM_CONSISTENT_READ
;
2003 /* If the parent can deal with changing data, we're okay with a
2004 * writable and resizable backing file. */
2005 /* TODO Require !(perm & BLK_PERM_CONSISTENT_READ), too? */
2006 if (shared
& BLK_PERM_WRITE
) {
2007 shared
= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2012 shared
|= BLK_PERM_CONSISTENT_READ
| BLK_PERM_GRAPH_MOD
|
2013 BLK_PERM_WRITE_UNCHANGED
;
2016 if (bs
->open_flags
& BDRV_O_INACTIVE
) {
2017 shared
|= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2024 static void bdrv_replace_child_noperm(BdrvChild
*child
,
2025 BlockDriverState
*new_bs
)
2027 BlockDriverState
*old_bs
= child
->bs
;
2030 if (old_bs
&& new_bs
) {
2031 assert(bdrv_get_aio_context(old_bs
) == bdrv_get_aio_context(new_bs
));
2034 /* Detach first so that the recursive drain sections coming from @child
2035 * are already gone and we only end the drain sections that came from
2037 if (child
->role
->detach
) {
2038 child
->role
->detach(child
);
2040 if (old_bs
->quiesce_counter
&& child
->role
->drained_end
) {
2041 int num
= old_bs
->quiesce_counter
;
2042 if (child
->role
->parent_is_bds
) {
2043 num
-= bdrv_drain_all_count
;
2046 for (i
= 0; i
< num
; i
++) {
2047 child
->role
->drained_end(child
);
2050 QLIST_REMOVE(child
, next_parent
);
2056 QLIST_INSERT_HEAD(&new_bs
->parents
, child
, next_parent
);
2057 if (new_bs
->quiesce_counter
&& child
->role
->drained_begin
) {
2058 int num
= new_bs
->quiesce_counter
;
2059 if (child
->role
->parent_is_bds
) {
2060 num
-= bdrv_drain_all_count
;
2063 for (i
= 0; i
< num
; i
++) {
2064 bdrv_parent_drained_begin_single(child
, true);
2068 /* Attach only after starting new drained sections, so that recursive
2069 * drain sections coming from @child don't get an extra .drained_begin
2071 if (child
->role
->attach
) {
2072 child
->role
->attach(child
);
2078 * Updates @child to change its reference to point to @new_bs, including
2079 * checking and applying the necessary permisson updates both to the old node
2082 * NULL is passed as @new_bs for removing the reference before freeing @child.
2084 * If @new_bs is not NULL, bdrv_check_perm() must be called beforehand, as this
2085 * function uses bdrv_set_perm() to update the permissions according to the new
2086 * reference that @new_bs gets.
2088 static void bdrv_replace_child(BdrvChild
*child
, BlockDriverState
*new_bs
)
2090 BlockDriverState
*old_bs
= child
->bs
;
2091 uint64_t perm
, shared_perm
;
2093 bdrv_replace_child_noperm(child
, new_bs
);
2096 /* Update permissions for old node. This is guaranteed to succeed
2097 * because we're just taking a parent away, so we're loosening
2099 bdrv_get_cumulative_perm(old_bs
, &perm
, &shared_perm
);
2100 bdrv_check_perm(old_bs
, NULL
, perm
, shared_perm
, NULL
, &error_abort
);
2101 bdrv_set_perm(old_bs
, perm
, shared_perm
);
2105 bdrv_get_cumulative_perm(new_bs
, &perm
, &shared_perm
);
2106 bdrv_set_perm(new_bs
, perm
, shared_perm
);
2110 BdrvChild
*bdrv_root_attach_child(BlockDriverState
*child_bs
,
2111 const char *child_name
,
2112 const BdrvChildRole
*child_role
,
2113 uint64_t perm
, uint64_t shared_perm
,
2114 void *opaque
, Error
**errp
)
2119 ret
= bdrv_check_update_perm(child_bs
, NULL
, perm
, shared_perm
, NULL
, errp
);
2121 bdrv_abort_perm_update(child_bs
);
2125 child
= g_new(BdrvChild
, 1);
2126 *child
= (BdrvChild
) {
2128 .name
= g_strdup(child_name
),
2131 .shared_perm
= shared_perm
,
2135 /* This performs the matching bdrv_set_perm() for the above check. */
2136 bdrv_replace_child(child
, child_bs
);
2141 BdrvChild
*bdrv_attach_child(BlockDriverState
*parent_bs
,
2142 BlockDriverState
*child_bs
,
2143 const char *child_name
,
2144 const BdrvChildRole
*child_role
,
2148 uint64_t perm
, shared_perm
;
2150 bdrv_get_cumulative_perm(parent_bs
, &perm
, &shared_perm
);
2152 assert(parent_bs
->drv
);
2153 assert(bdrv_get_aio_context(parent_bs
) == bdrv_get_aio_context(child_bs
));
2154 bdrv_child_perm(parent_bs
, child_bs
, NULL
, child_role
, NULL
,
2155 perm
, shared_perm
, &perm
, &shared_perm
);
2157 child
= bdrv_root_attach_child(child_bs
, child_name
, child_role
,
2158 perm
, shared_perm
, parent_bs
, errp
);
2159 if (child
== NULL
) {
2163 QLIST_INSERT_HEAD(&parent_bs
->children
, child
, next
);
2167 static void bdrv_detach_child(BdrvChild
*child
)
2169 if (child
->next
.le_prev
) {
2170 QLIST_REMOVE(child
, next
);
2171 child
->next
.le_prev
= NULL
;
2174 bdrv_replace_child(child
, NULL
);
2176 g_free(child
->name
);
2180 void bdrv_root_unref_child(BdrvChild
*child
)
2182 BlockDriverState
*child_bs
;
2184 child_bs
= child
->bs
;
2185 bdrv_detach_child(child
);
2186 bdrv_unref(child_bs
);
2189 void bdrv_unref_child(BlockDriverState
*parent
, BdrvChild
*child
)
2191 if (child
== NULL
) {
2195 if (child
->bs
->inherits_from
== parent
) {
2198 /* Remove inherits_from only when the last reference between parent and
2199 * child->bs goes away. */
2200 QLIST_FOREACH(c
, &parent
->children
, next
) {
2201 if (c
!= child
&& c
->bs
== child
->bs
) {
2206 child
->bs
->inherits_from
= NULL
;
2210 bdrv_root_unref_child(child
);
2214 static void bdrv_parent_cb_change_media(BlockDriverState
*bs
, bool load
)
2217 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
2218 if (c
->role
->change_media
) {
2219 c
->role
->change_media(c
, load
);
2225 * Sets the backing file link of a BDS. A new reference is created; callers
2226 * which don't need their own reference any more must call bdrv_unref().
2228 void bdrv_set_backing_hd(BlockDriverState
*bs
, BlockDriverState
*backing_hd
,
2232 bdrv_ref(backing_hd
);
2236 bdrv_unref_child(bs
, bs
->backing
);
2244 bs
->backing
= bdrv_attach_child(bs
, backing_hd
, "backing", &child_backing
,
2247 bdrv_unref(backing_hd
);
2250 bdrv_refresh_filename(bs
);
2253 bdrv_refresh_limits(bs
, NULL
);
2257 * Opens the backing file for a BlockDriverState if not yet open
2259 * bdref_key specifies the key for the image's BlockdevRef in the options QDict.
2260 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
2261 * itself, all options starting with "${bdref_key}." are considered part of the
2264 * TODO Can this be unified with bdrv_open_image()?
2266 int bdrv_open_backing_file(BlockDriverState
*bs
, QDict
*parent_options
,
2267 const char *bdref_key
, Error
**errp
)
2269 char *backing_filename
= g_malloc0(PATH_MAX
);
2270 char *bdref_key_dot
;
2271 const char *reference
= NULL
;
2273 BlockDriverState
*backing_hd
;
2275 QDict
*tmp_parent_options
= NULL
;
2276 Error
*local_err
= NULL
;
2278 if (bs
->backing
!= NULL
) {
2282 /* NULL means an empty set of options */
2283 if (parent_options
== NULL
) {
2284 tmp_parent_options
= qdict_new();
2285 parent_options
= tmp_parent_options
;
2288 bs
->open_flags
&= ~BDRV_O_NO_BACKING
;
2290 bdref_key_dot
= g_strdup_printf("%s.", bdref_key
);
2291 qdict_extract_subqdict(parent_options
, &options
, bdref_key_dot
);
2292 g_free(bdref_key_dot
);
2295 * Caution: while qdict_get_try_str() is fine, getting non-string
2296 * types would require more care. When @parent_options come from
2297 * -blockdev or blockdev_add, its members are typed according to
2298 * the QAPI schema, but when they come from -drive, they're all
2301 reference
= qdict_get_try_str(parent_options
, bdref_key
);
2302 if (reference
|| qdict_haskey(options
, "file.filename")) {
2303 backing_filename
[0] = '\0';
2304 } else if (bs
->backing_file
[0] == '\0' && qdict_size(options
) == 0) {
2305 qobject_unref(options
);
2308 bdrv_get_full_backing_filename(bs
, backing_filename
, PATH_MAX
,
2312 error_propagate(errp
, local_err
);
2313 qobject_unref(options
);
2318 if (!bs
->drv
|| !bs
->drv
->supports_backing
) {
2320 error_setg(errp
, "Driver doesn't support backing files");
2321 qobject_unref(options
);
2326 bs
->backing_format
[0] != '\0' && !qdict_haskey(options
, "driver")) {
2327 qdict_put_str(options
, "driver", bs
->backing_format
);
2330 backing_hd
= bdrv_open_inherit(*backing_filename
? backing_filename
: NULL
,
2331 reference
, options
, 0, bs
, &child_backing
,
2334 bs
->open_flags
|= BDRV_O_NO_BACKING
;
2335 error_prepend(errp
, "Could not open backing file: ");
2339 bdrv_set_aio_context(backing_hd
, bdrv_get_aio_context(bs
));
2341 /* Hook up the backing file link; drop our reference, bs owns the
2342 * backing_hd reference now */
2343 bdrv_set_backing_hd(bs
, backing_hd
, &local_err
);
2344 bdrv_unref(backing_hd
);
2346 error_propagate(errp
, local_err
);
2351 qdict_del(parent_options
, bdref_key
);
2354 g_free(backing_filename
);
2355 qobject_unref(tmp_parent_options
);
2359 static BlockDriverState
*
2360 bdrv_open_child_bs(const char *filename
, QDict
*options
, const char *bdref_key
,
2361 BlockDriverState
*parent
, const BdrvChildRole
*child_role
,
2362 bool allow_none
, Error
**errp
)
2364 BlockDriverState
*bs
= NULL
;
2365 QDict
*image_options
;
2366 char *bdref_key_dot
;
2367 const char *reference
;
2369 assert(child_role
!= NULL
);
2371 bdref_key_dot
= g_strdup_printf("%s.", bdref_key
);
2372 qdict_extract_subqdict(options
, &image_options
, bdref_key_dot
);
2373 g_free(bdref_key_dot
);
2376 * Caution: while qdict_get_try_str() is fine, getting non-string
2377 * types would require more care. When @options come from
2378 * -blockdev or blockdev_add, its members are typed according to
2379 * the QAPI schema, but when they come from -drive, they're all
2382 reference
= qdict_get_try_str(options
, bdref_key
);
2383 if (!filename
&& !reference
&& !qdict_size(image_options
)) {
2385 error_setg(errp
, "A block device must be specified for \"%s\"",
2388 qobject_unref(image_options
);
2392 bs
= bdrv_open_inherit(filename
, reference
, image_options
, 0,
2393 parent
, child_role
, errp
);
2399 qdict_del(options
, bdref_key
);
2404 * Opens a disk image whose options are given as BlockdevRef in another block
2407 * If allow_none is true, no image will be opened if filename is false and no
2408 * BlockdevRef is given. NULL will be returned, but errp remains unset.
2410 * bdrev_key specifies the key for the image's BlockdevRef in the options QDict.
2411 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
2412 * itself, all options starting with "${bdref_key}." are considered part of the
2415 * The BlockdevRef will be removed from the options QDict.
2417 BdrvChild
*bdrv_open_child(const char *filename
,
2418 QDict
*options
, const char *bdref_key
,
2419 BlockDriverState
*parent
,
2420 const BdrvChildRole
*child_role
,
2421 bool allow_none
, Error
**errp
)
2424 BlockDriverState
*bs
;
2426 bs
= bdrv_open_child_bs(filename
, options
, bdref_key
, parent
, child_role
,
2432 c
= bdrv_attach_child(parent
, bs
, bdref_key
, child_role
, errp
);
2441 /* TODO Future callers may need to specify parent/child_role in order for
2442 * option inheritance to work. Existing callers use it for the root node. */
2443 BlockDriverState
*bdrv_open_blockdev_ref(BlockdevRef
*ref
, Error
**errp
)
2445 BlockDriverState
*bs
= NULL
;
2446 Error
*local_err
= NULL
;
2447 QObject
*obj
= NULL
;
2448 QDict
*qdict
= NULL
;
2449 const char *reference
= NULL
;
2452 if (ref
->type
== QTYPE_QSTRING
) {
2453 reference
= ref
->u
.reference
;
2455 BlockdevOptions
*options
= &ref
->u
.definition
;
2456 assert(ref
->type
== QTYPE_QDICT
);
2458 v
= qobject_output_visitor_new(&obj
);
2459 visit_type_BlockdevOptions(v
, NULL
, &options
, &local_err
);
2461 error_propagate(errp
, local_err
);
2464 visit_complete(v
, &obj
);
2466 qdict
= qobject_to(QDict
, obj
);
2467 qdict_flatten(qdict
);
2469 /* bdrv_open_inherit() defaults to the values in bdrv_flags (for
2470 * compatibility with other callers) rather than what we want as the
2471 * real defaults. Apply the defaults here instead. */
2472 qdict_set_default_str(qdict
, BDRV_OPT_CACHE_DIRECT
, "off");
2473 qdict_set_default_str(qdict
, BDRV_OPT_CACHE_NO_FLUSH
, "off");
2474 qdict_set_default_str(qdict
, BDRV_OPT_READ_ONLY
, "off");
2477 bs
= bdrv_open_inherit(NULL
, reference
, qdict
, 0, NULL
, NULL
, errp
);
2486 static BlockDriverState
*bdrv_append_temp_snapshot(BlockDriverState
*bs
,
2488 QDict
*snapshot_options
,
2491 /* TODO: extra byte is a hack to ensure MAX_PATH space on Windows. */
2492 char *tmp_filename
= g_malloc0(PATH_MAX
+ 1);
2494 QemuOpts
*opts
= NULL
;
2495 BlockDriverState
*bs_snapshot
= NULL
;
2496 Error
*local_err
= NULL
;
2499 /* if snapshot, we create a temporary backing file and open it
2500 instead of opening 'filename' directly */
2502 /* Get the required size from the image */
2503 total_size
= bdrv_getlength(bs
);
2504 if (total_size
< 0) {
2505 error_setg_errno(errp
, -total_size
, "Could not get image size");
2509 /* Create the temporary image */
2510 ret
= get_tmp_filename(tmp_filename
, PATH_MAX
+ 1);
2512 error_setg_errno(errp
, -ret
, "Could not get temporary filename");
2516 opts
= qemu_opts_create(bdrv_qcow2
.create_opts
, NULL
, 0,
2518 qemu_opt_set_number(opts
, BLOCK_OPT_SIZE
, total_size
, &error_abort
);
2519 ret
= bdrv_create(&bdrv_qcow2
, tmp_filename
, opts
, errp
);
2520 qemu_opts_del(opts
);
2522 error_prepend(errp
, "Could not create temporary overlay '%s': ",
2527 /* Prepare options QDict for the temporary file */
2528 qdict_put_str(snapshot_options
, "file.driver", "file");
2529 qdict_put_str(snapshot_options
, "file.filename", tmp_filename
);
2530 qdict_put_str(snapshot_options
, "driver", "qcow2");
2532 bs_snapshot
= bdrv_open(NULL
, NULL
, snapshot_options
, flags
, errp
);
2533 snapshot_options
= NULL
;
2538 /* bdrv_append() consumes a strong reference to bs_snapshot
2539 * (i.e. it will call bdrv_unref() on it) even on error, so in
2540 * order to be able to return one, we have to increase
2541 * bs_snapshot's refcount here */
2542 bdrv_ref(bs_snapshot
);
2543 bdrv_append(bs_snapshot
, bs
, &local_err
);
2545 error_propagate(errp
, local_err
);
2551 qobject_unref(snapshot_options
);
2552 g_free(tmp_filename
);
2557 * Opens a disk image (raw, qcow2, vmdk, ...)
2559 * options is a QDict of options to pass to the block drivers, or NULL for an
2560 * empty set of options. The reference to the QDict belongs to the block layer
2561 * after the call (even on failure), so if the caller intends to reuse the
2562 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
2564 * If *pbs is NULL, a new BDS will be created with a pointer to it stored there.
2565 * If it is not NULL, the referenced BDS will be reused.
2567 * The reference parameter may be used to specify an existing block device which
2568 * should be opened. If specified, neither options nor a filename may be given,
2569 * nor can an existing BDS be reused (that is, *pbs has to be NULL).
2571 static BlockDriverState
*bdrv_open_inherit(const char *filename
,
2572 const char *reference
,
2573 QDict
*options
, int flags
,
2574 BlockDriverState
*parent
,
2575 const BdrvChildRole
*child_role
,
2579 BlockBackend
*file
= NULL
;
2580 BlockDriverState
*bs
;
2581 BlockDriver
*drv
= NULL
;
2583 const char *drvname
;
2584 const char *backing
;
2585 Error
*local_err
= NULL
;
2586 QDict
*snapshot_options
= NULL
;
2587 int snapshot_flags
= 0;
2589 assert(!child_role
|| !flags
);
2590 assert(!child_role
== !parent
);
2593 bool options_non_empty
= options
? qdict_size(options
) : false;
2594 qobject_unref(options
);
2596 if (filename
|| options_non_empty
) {
2597 error_setg(errp
, "Cannot reference an existing block device with "
2598 "additional options or a new filename");
2602 bs
= bdrv_lookup_bs(reference
, reference
, errp
);
2613 /* NULL means an empty set of options */
2614 if (options
== NULL
) {
2615 options
= qdict_new();
2618 /* json: syntax counts as explicit options, as if in the QDict */
2619 parse_json_protocol(options
, &filename
, &local_err
);
2624 bs
->explicit_options
= qdict_clone_shallow(options
);
2627 bs
->inherits_from
= parent
;
2628 child_role
->inherit_options(&flags
, options
,
2629 parent
->open_flags
, parent
->options
);
2632 ret
= bdrv_fill_options(&options
, filename
, &flags
, &local_err
);
2638 * Set the BDRV_O_RDWR and BDRV_O_ALLOW_RDWR flags.
2639 * Caution: getting a boolean member of @options requires care.
2640 * When @options come from -blockdev or blockdev_add, members are
2641 * typed according to the QAPI schema, but when they come from
2642 * -drive, they're all QString.
2644 if (g_strcmp0(qdict_get_try_str(options
, BDRV_OPT_READ_ONLY
), "on") &&
2645 !qdict_get_try_bool(options
, BDRV_OPT_READ_ONLY
, false)) {
2646 flags
|= (BDRV_O_RDWR
| BDRV_O_ALLOW_RDWR
);
2648 flags
&= ~BDRV_O_RDWR
;
2651 if (flags
& BDRV_O_SNAPSHOT
) {
2652 snapshot_options
= qdict_new();
2653 bdrv_temp_snapshot_options(&snapshot_flags
, snapshot_options
,
2655 /* Let bdrv_backing_options() override "read-only" */
2656 qdict_del(options
, BDRV_OPT_READ_ONLY
);
2657 bdrv_backing_options(&flags
, options
, flags
, options
);
2660 bs
->open_flags
= flags
;
2661 bs
->options
= options
;
2662 options
= qdict_clone_shallow(options
);
2664 /* Find the right image format driver */
2665 /* See cautionary note on accessing @options above */
2666 drvname
= qdict_get_try_str(options
, "driver");
2668 drv
= bdrv_find_format(drvname
);
2670 error_setg(errp
, "Unknown driver: '%s'", drvname
);
2675 assert(drvname
|| !(flags
& BDRV_O_PROTOCOL
));
2677 /* See cautionary note on accessing @options above */
2678 backing
= qdict_get_try_str(options
, "backing");
2679 if (qobject_to(QNull
, qdict_get(options
, "backing")) != NULL
||
2680 (backing
&& *backing
== '\0'))
2683 warn_report("Use of \"backing\": \"\" is deprecated; "
2684 "use \"backing\": null instead");
2686 flags
|= BDRV_O_NO_BACKING
;
2687 qdict_del(options
, "backing");
2690 /* Open image file without format layer. This BlockBackend is only used for
2691 * probing, the block drivers will do their own bdrv_open_child() for the
2692 * same BDS, which is why we put the node name back into options. */
2693 if ((flags
& BDRV_O_PROTOCOL
) == 0) {
2694 BlockDriverState
*file_bs
;
2696 file_bs
= bdrv_open_child_bs(filename
, options
, "file", bs
,
2697 &child_file
, true, &local_err
);
2701 if (file_bs
!= NULL
) {
2702 /* Not requesting BLK_PERM_CONSISTENT_READ because we're only
2703 * looking at the header to guess the image format. This works even
2704 * in cases where a guest would not see a consistent state. */
2705 file
= blk_new(0, BLK_PERM_ALL
);
2706 blk_insert_bs(file
, file_bs
, &local_err
);
2707 bdrv_unref(file_bs
);
2712 qdict_put_str(options
, "file", bdrv_get_node_name(file_bs
));
2716 /* Image format probing */
2719 ret
= find_image_format(file
, filename
, &drv
, &local_err
);
2724 * This option update would logically belong in bdrv_fill_options(),
2725 * but we first need to open bs->file for the probing to work, while
2726 * opening bs->file already requires the (mostly) final set of options
2727 * so that cache mode etc. can be inherited.
2729 * Adding the driver later is somewhat ugly, but it's not an option
2730 * that would ever be inherited, so it's correct. We just need to make
2731 * sure to update both bs->options (which has the full effective
2732 * options for bs) and options (which has file.* already removed).
2734 qdict_put_str(bs
->options
, "driver", drv
->format_name
);
2735 qdict_put_str(options
, "driver", drv
->format_name
);
2737 error_setg(errp
, "Must specify either driver or file");
2741 /* BDRV_O_PROTOCOL must be set iff a protocol BDS is about to be created */
2742 assert(!!(flags
& BDRV_O_PROTOCOL
) == !!drv
->bdrv_file_open
);
2743 /* file must be NULL if a protocol BDS is about to be created
2744 * (the inverse results in an error message from bdrv_open_common()) */
2745 assert(!(flags
& BDRV_O_PROTOCOL
) || !file
);
2747 /* Open the image */
2748 ret
= bdrv_open_common(bs
, file
, options
, &local_err
);
2758 /* If there is a backing file, use it */
2759 if ((flags
& BDRV_O_NO_BACKING
) == 0) {
2760 ret
= bdrv_open_backing_file(bs
, options
, "backing", &local_err
);
2762 goto close_and_fail
;
2766 /* Remove all children options and references
2767 * from bs->options and bs->explicit_options */
2768 QLIST_FOREACH(child
, &bs
->children
, next
) {
2769 char *child_key_dot
;
2770 child_key_dot
= g_strdup_printf("%s.", child
->name
);
2771 qdict_extract_subqdict(bs
->explicit_options
, NULL
, child_key_dot
);
2772 qdict_extract_subqdict(bs
->options
, NULL
, child_key_dot
);
2773 qdict_del(bs
->explicit_options
, child
->name
);
2774 qdict_del(bs
->options
, child
->name
);
2775 g_free(child_key_dot
);
2778 bdrv_refresh_filename(bs
);
2780 /* Check if any unknown options were used */
2781 if (qdict_size(options
) != 0) {
2782 const QDictEntry
*entry
= qdict_first(options
);
2783 if (flags
& BDRV_O_PROTOCOL
) {
2784 error_setg(errp
, "Block protocol '%s' doesn't support the option "
2785 "'%s'", drv
->format_name
, entry
->key
);
2788 "Block format '%s' does not support the option '%s'",
2789 drv
->format_name
, entry
->key
);
2792 goto close_and_fail
;
2795 bdrv_parent_cb_change_media(bs
, true);
2797 qobject_unref(options
);
2800 /* For snapshot=on, create a temporary qcow2 overlay. bs points to the
2801 * temporary snapshot afterwards. */
2802 if (snapshot_flags
) {
2803 BlockDriverState
*snapshot_bs
;
2804 snapshot_bs
= bdrv_append_temp_snapshot(bs
, snapshot_flags
,
2805 snapshot_options
, &local_err
);
2806 snapshot_options
= NULL
;
2808 goto close_and_fail
;
2810 /* We are not going to return bs but the overlay on top of it
2811 * (snapshot_bs); thus, we have to drop the strong reference to bs
2812 * (which we obtained by calling bdrv_new()). bs will not be deleted,
2813 * though, because the overlay still has a reference to it. */
2822 qobject_unref(snapshot_options
);
2823 qobject_unref(bs
->explicit_options
);
2824 qobject_unref(bs
->options
);
2825 qobject_unref(options
);
2827 bs
->explicit_options
= NULL
;
2829 error_propagate(errp
, local_err
);
2834 qobject_unref(snapshot_options
);
2835 qobject_unref(options
);
2836 error_propagate(errp
, local_err
);
2840 BlockDriverState
*bdrv_open(const char *filename
, const char *reference
,
2841 QDict
*options
, int flags
, Error
**errp
)
2843 return bdrv_open_inherit(filename
, reference
, options
, flags
, NULL
,
2848 * Adds a BlockDriverState to a simple queue for an atomic, transactional
2849 * reopen of multiple devices.
2851 * bs_queue can either be an existing BlockReopenQueue that has had QSIMPLE_INIT
2852 * already performed, or alternatively may be NULL a new BlockReopenQueue will
2853 * be created and initialized. This newly created BlockReopenQueue should be
2854 * passed back in for subsequent calls that are intended to be of the same
2857 * bs is the BlockDriverState to add to the reopen queue.
2859 * options contains the changed options for the associated bs
2860 * (the BlockReopenQueue takes ownership)
2862 * flags contains the open flags for the associated bs
2864 * returns a pointer to bs_queue, which is either the newly allocated
2865 * bs_queue, or the existing bs_queue being used.
2867 * bs must be drained between bdrv_reopen_queue() and bdrv_reopen_multiple().
2869 static BlockReopenQueue
*bdrv_reopen_queue_child(BlockReopenQueue
*bs_queue
,
2870 BlockDriverState
*bs
,
2873 const BdrvChildRole
*role
,
2874 QDict
*parent_options
,
2879 BlockReopenQueueEntry
*bs_entry
;
2881 QDict
*old_options
, *explicit_options
;
2883 /* Make sure that the caller remembered to use a drained section. This is
2884 * important to avoid graph changes between the recursive queuing here and
2885 * bdrv_reopen_multiple(). */
2886 assert(bs
->quiesce_counter
> 0);
2888 if (bs_queue
== NULL
) {
2889 bs_queue
= g_new0(BlockReopenQueue
, 1);
2890 QSIMPLEQ_INIT(bs_queue
);
2894 options
= qdict_new();
2897 /* Check if this BlockDriverState is already in the queue */
2898 QSIMPLEQ_FOREACH(bs_entry
, bs_queue
, entry
) {
2899 if (bs
== bs_entry
->state
.bs
) {
2905 * Precedence of options:
2906 * 1. Explicitly passed in options (highest)
2907 * 2. Set in flags (only for top level)
2908 * 3. Retained from explicitly set options of bs
2909 * 4. Inherited from parent node
2910 * 5. Retained from effective options of bs
2913 if (!parent_options
) {
2915 * Any setting represented by flags is always updated. If the
2916 * corresponding QDict option is set, it takes precedence. Otherwise
2917 * the flag is translated into a QDict option. The old setting of bs is
2920 update_options_from_flags(options
, flags
);
2923 /* Old explicitly set values (don't overwrite by inherited value) */
2925 old_options
= qdict_clone_shallow(bs_entry
->state
.explicit_options
);
2927 old_options
= qdict_clone_shallow(bs
->explicit_options
);
2929 bdrv_join_options(bs
, options
, old_options
);
2930 qobject_unref(old_options
);
2932 explicit_options
= qdict_clone_shallow(options
);
2934 /* Inherit from parent node */
2935 if (parent_options
) {
2937 QDict
*options_copy
;
2939 role
->inherit_options(&flags
, options
, parent_flags
, parent_options
);
2940 options_copy
= qdict_clone_shallow(options
);
2941 opts
= qemu_opts_create(&bdrv_runtime_opts
, NULL
, 0, &error_abort
);
2942 qemu_opts_absorb_qdict(opts
, options_copy
, NULL
);
2943 update_flags_from_options(&flags
, opts
);
2944 qemu_opts_del(opts
);
2945 qobject_unref(options_copy
);
2948 /* Old values are used for options that aren't set yet */
2949 old_options
= qdict_clone_shallow(bs
->options
);
2950 bdrv_join_options(bs
, options
, old_options
);
2951 qobject_unref(old_options
);
2953 /* bdrv_open_inherit() sets and clears some additional flags internally */
2954 flags
&= ~BDRV_O_PROTOCOL
;
2955 if (flags
& BDRV_O_RDWR
) {
2956 flags
|= BDRV_O_ALLOW_RDWR
;
2960 bs_entry
= g_new0(BlockReopenQueueEntry
, 1);
2961 QSIMPLEQ_INSERT_TAIL(bs_queue
, bs_entry
, entry
);
2963 qobject_unref(bs_entry
->state
.options
);
2964 qobject_unref(bs_entry
->state
.explicit_options
);
2967 bs_entry
->state
.bs
= bs
;
2968 bs_entry
->state
.options
= options
;
2969 bs_entry
->state
.explicit_options
= explicit_options
;
2970 bs_entry
->state
.flags
= flags
;
2972 /* This needs to be overwritten in bdrv_reopen_prepare() */
2973 bs_entry
->state
.perm
= UINT64_MAX
;
2974 bs_entry
->state
.shared_perm
= 0;
2976 QLIST_FOREACH(child
, &bs
->children
, next
) {
2977 QDict
*new_child_options
;
2978 char *child_key_dot
;
2980 /* reopen can only change the options of block devices that were
2981 * implicitly created and inherited options. For other (referenced)
2982 * block devices, a syntax like "backing.foo" results in an error. */
2983 if (child
->bs
->inherits_from
!= bs
) {
2987 child_key_dot
= g_strdup_printf("%s.", child
->name
);
2988 qdict_extract_subqdict(explicit_options
, NULL
, child_key_dot
);
2989 qdict_extract_subqdict(options
, &new_child_options
, child_key_dot
);
2990 g_free(child_key_dot
);
2992 bdrv_reopen_queue_child(bs_queue
, child
->bs
, new_child_options
, 0,
2993 child
->role
, options
, flags
);
2999 BlockReopenQueue
*bdrv_reopen_queue(BlockReopenQueue
*bs_queue
,
3000 BlockDriverState
*bs
,
3001 QDict
*options
, int flags
)
3003 return bdrv_reopen_queue_child(bs_queue
, bs
, options
, flags
,
3008 * Reopen multiple BlockDriverStates atomically & transactionally.
3010 * The queue passed in (bs_queue) must have been built up previous
3011 * via bdrv_reopen_queue().
3013 * Reopens all BDS specified in the queue, with the appropriate
3014 * flags. All devices are prepared for reopen, and failure of any
3015 * device will cause all device changes to be abandoned, and intermediate
3018 * If all devices prepare successfully, then the changes are committed
3021 * All affected nodes must be drained between bdrv_reopen_queue() and
3022 * bdrv_reopen_multiple().
3024 int bdrv_reopen_multiple(AioContext
*ctx
, BlockReopenQueue
*bs_queue
, Error
**errp
)
3027 BlockReopenQueueEntry
*bs_entry
, *next
;
3028 Error
*local_err
= NULL
;
3030 assert(bs_queue
!= NULL
);
3032 QSIMPLEQ_FOREACH(bs_entry
, bs_queue
, entry
) {
3033 assert(bs_entry
->state
.bs
->quiesce_counter
> 0);
3034 if (bdrv_reopen_prepare(&bs_entry
->state
, bs_queue
, &local_err
)) {
3035 error_propagate(errp
, local_err
);
3038 bs_entry
->prepared
= true;
3041 /* If we reach this point, we have success and just need to apply the
3044 QSIMPLEQ_FOREACH(bs_entry
, bs_queue
, entry
) {
3045 bdrv_reopen_commit(&bs_entry
->state
);
3051 QSIMPLEQ_FOREACH_SAFE(bs_entry
, bs_queue
, entry
, next
) {
3053 if (bs_entry
->prepared
) {
3054 bdrv_reopen_abort(&bs_entry
->state
);
3056 qobject_unref(bs_entry
->state
.explicit_options
);
3057 qobject_unref(bs_entry
->state
.options
);
3067 /* Reopen a single BlockDriverState with the specified flags. */
3068 int bdrv_reopen(BlockDriverState
*bs
, int bdrv_flags
, Error
**errp
)
3071 Error
*local_err
= NULL
;
3072 BlockReopenQueue
*queue
;
3074 bdrv_subtree_drained_begin(bs
);
3076 queue
= bdrv_reopen_queue(NULL
, bs
, NULL
, bdrv_flags
);
3077 ret
= bdrv_reopen_multiple(bdrv_get_aio_context(bs
), queue
, &local_err
);
3078 if (local_err
!= NULL
) {
3079 error_propagate(errp
, local_err
);
3082 bdrv_subtree_drained_end(bs
);
3087 static BlockReopenQueueEntry
*find_parent_in_reopen_queue(BlockReopenQueue
*q
,
3090 BlockReopenQueueEntry
*entry
;
3092 QSIMPLEQ_FOREACH(entry
, q
, entry
) {
3093 BlockDriverState
*bs
= entry
->state
.bs
;
3096 QLIST_FOREACH(child
, &bs
->children
, next
) {
3106 static void bdrv_reopen_perm(BlockReopenQueue
*q
, BlockDriverState
*bs
,
3107 uint64_t *perm
, uint64_t *shared
)
3110 BlockReopenQueueEntry
*parent
;
3111 uint64_t cumulative_perms
= 0;
3112 uint64_t cumulative_shared_perms
= BLK_PERM_ALL
;
3114 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
3115 parent
= find_parent_in_reopen_queue(q
, c
);
3117 cumulative_perms
|= c
->perm
;
3118 cumulative_shared_perms
&= c
->shared_perm
;
3120 uint64_t nperm
, nshared
;
3122 bdrv_child_perm(parent
->state
.bs
, bs
, c
, c
->role
, q
,
3123 parent
->state
.perm
, parent
->state
.shared_perm
,
3126 cumulative_perms
|= nperm
;
3127 cumulative_shared_perms
&= nshared
;
3130 *perm
= cumulative_perms
;
3131 *shared
= cumulative_shared_perms
;
3135 * Prepares a BlockDriverState for reopen. All changes are staged in the
3136 * 'opaque' field of the BDRVReopenState, which is used and allocated by
3137 * the block driver layer .bdrv_reopen_prepare()
3139 * bs is the BlockDriverState to reopen
3140 * flags are the new open flags
3141 * queue is the reopen queue
3143 * Returns 0 on success, non-zero on error. On error errp will be set
3146 * On failure, bdrv_reopen_abort() will be called to clean up any data.
3147 * It is the responsibility of the caller to then call the abort() or
3148 * commit() for any other BDS that have been left in a prepare() state
3151 int bdrv_reopen_prepare(BDRVReopenState
*reopen_state
, BlockReopenQueue
*queue
,
3155 Error
*local_err
= NULL
;
3158 QDict
*orig_reopen_opts
;
3159 char *discard
= NULL
;
3162 assert(reopen_state
!= NULL
);
3163 assert(reopen_state
->bs
->drv
!= NULL
);
3164 drv
= reopen_state
->bs
->drv
;
3166 /* This function and each driver's bdrv_reopen_prepare() remove
3167 * entries from reopen_state->options as they are processed, so
3168 * we need to make a copy of the original QDict. */
3169 orig_reopen_opts
= qdict_clone_shallow(reopen_state
->options
);
3171 /* Process generic block layer options */
3172 opts
= qemu_opts_create(&bdrv_runtime_opts
, NULL
, 0, &error_abort
);
3173 qemu_opts_absorb_qdict(opts
, reopen_state
->options
, &local_err
);
3175 error_propagate(errp
, local_err
);
3180 update_flags_from_options(&reopen_state
->flags
, opts
);
3182 discard
= qemu_opt_get_del(opts
, "discard");
3183 if (discard
!= NULL
) {
3184 if (bdrv_parse_discard_flags(discard
, &reopen_state
->flags
) != 0) {
3185 error_setg(errp
, "Invalid discard option");
3191 /* All other options (including node-name and driver) must be unchanged.
3192 * Put them back into the QDict, so that they are checked at the end
3193 * of this function. */
3194 qemu_opts_to_qdict(opts
, reopen_state
->options
);
3196 /* If we are to stay read-only, do not allow permission change
3197 * to r/w. Attempting to set to r/w may fail if either BDRV_O_ALLOW_RDWR is
3198 * not set, or if the BDS still has copy_on_read enabled */
3199 read_only
= !(reopen_state
->flags
& BDRV_O_RDWR
);
3200 ret
= bdrv_can_set_read_only(reopen_state
->bs
, read_only
, true, &local_err
);
3202 error_propagate(errp
, local_err
);
3206 /* Calculate required permissions after reopening */
3207 bdrv_reopen_perm(queue
, reopen_state
->bs
,
3208 &reopen_state
->perm
, &reopen_state
->shared_perm
);
3210 ret
= bdrv_flush(reopen_state
->bs
);
3212 error_setg_errno(errp
, -ret
, "Error flushing drive");
3216 if (drv
->bdrv_reopen_prepare
) {
3217 ret
= drv
->bdrv_reopen_prepare(reopen_state
, queue
, &local_err
);
3219 if (local_err
!= NULL
) {
3220 error_propagate(errp
, local_err
);
3222 error_setg(errp
, "failed while preparing to reopen image '%s'",
3223 reopen_state
->bs
->filename
);
3228 /* It is currently mandatory to have a bdrv_reopen_prepare()
3229 * handler for each supported drv. */
3230 error_setg(errp
, "Block format '%s' used by node '%s' "
3231 "does not support reopening files", drv
->format_name
,
3232 bdrv_get_device_or_node_name(reopen_state
->bs
));
3237 /* Options that are not handled are only okay if they are unchanged
3238 * compared to the old state. It is expected that some options are only
3239 * used for the initial open, but not reopen (e.g. filename) */
3240 if (qdict_size(reopen_state
->options
)) {
3241 const QDictEntry
*entry
= qdict_first(reopen_state
->options
);
3244 QObject
*new = entry
->value
;
3245 QObject
*old
= qdict_get(reopen_state
->bs
->options
, entry
->key
);
3247 /* Allow child references (child_name=node_name) as long as they
3248 * point to the current child (i.e. everything stays the same). */
3249 if (qobject_type(new) == QTYPE_QSTRING
) {
3251 QLIST_FOREACH(child
, &reopen_state
->bs
->children
, next
) {
3252 if (!strcmp(child
->name
, entry
->key
)) {
3258 const char *str
= qobject_get_try_str(new);
3259 if (!strcmp(child
->bs
->node_name
, str
)) {
3260 continue; /* Found child with this name, skip option */
3266 * TODO: When using -drive to specify blockdev options, all values
3267 * will be strings; however, when using -blockdev, blockdev-add or
3268 * filenames using the json:{} pseudo-protocol, they will be
3270 * In contrast, reopening options are (currently) always strings
3271 * (because you can only specify them through qemu-io; all other
3272 * callers do not specify any options).
3273 * Therefore, when using anything other than -drive to create a BDS,
3274 * this cannot detect non-string options as unchanged, because
3275 * qobject_is_equal() always returns false for objects of different
3276 * type. In the future, this should be remedied by correctly typing
3277 * all options. For now, this is not too big of an issue because
3278 * the user can simply omit options which cannot be changed anyway,
3279 * so they will stay unchanged.
3281 if (!qobject_is_equal(new, old
)) {
3282 error_setg(errp
, "Cannot change the option '%s'", entry
->key
);
3286 } while ((entry
= qdict_next(reopen_state
->options
, entry
)));
3289 ret
= bdrv_check_perm(reopen_state
->bs
, queue
, reopen_state
->perm
,
3290 reopen_state
->shared_perm
, NULL
, errp
);
3297 /* Restore the original reopen_state->options QDict */
3298 qobject_unref(reopen_state
->options
);
3299 reopen_state
->options
= qobject_ref(orig_reopen_opts
);
3302 qemu_opts_del(opts
);
3303 qobject_unref(orig_reopen_opts
);
3309 * Takes the staged changes for the reopen from bdrv_reopen_prepare(), and
3310 * makes them final by swapping the staging BlockDriverState contents into
3311 * the active BlockDriverState contents.
3313 void bdrv_reopen_commit(BDRVReopenState
*reopen_state
)
3316 BlockDriverState
*bs
;
3318 bool old_can_write
, new_can_write
;
3320 assert(reopen_state
!= NULL
);
3321 bs
= reopen_state
->bs
;
3323 assert(drv
!= NULL
);
3326 !bdrv_is_read_only(bs
) && !(bdrv_get_flags(bs
) & BDRV_O_INACTIVE
);
3328 /* If there are any driver level actions to take */
3329 if (drv
->bdrv_reopen_commit
) {
3330 drv
->bdrv_reopen_commit(reopen_state
);
3333 /* set BDS specific flags now */
3334 qobject_unref(bs
->explicit_options
);
3335 qobject_unref(bs
->options
);
3337 bs
->explicit_options
= reopen_state
->explicit_options
;
3338 bs
->options
= reopen_state
->options
;
3339 bs
->open_flags
= reopen_state
->flags
;
3340 bs
->read_only
= !(reopen_state
->flags
& BDRV_O_RDWR
);
3342 /* Remove child references from bs->options and bs->explicit_options.
3343 * Child options were already removed in bdrv_reopen_queue_child() */
3344 QLIST_FOREACH(child
, &bs
->children
, next
) {
3345 qdict_del(bs
->explicit_options
, child
->name
);
3346 qdict_del(bs
->options
, child
->name
);
3349 bdrv_refresh_limits(bs
, NULL
);
3351 bdrv_set_perm(reopen_state
->bs
, reopen_state
->perm
,
3352 reopen_state
->shared_perm
);
3355 !bdrv_is_read_only(bs
) && !(bdrv_get_flags(bs
) & BDRV_O_INACTIVE
);
3356 if (!old_can_write
&& new_can_write
&& drv
->bdrv_reopen_bitmaps_rw
) {
3357 Error
*local_err
= NULL
;
3358 if (drv
->bdrv_reopen_bitmaps_rw(bs
, &local_err
) < 0) {
3359 /* This is not fatal, bitmaps just left read-only, so all following
3360 * writes will fail. User can remove read-only bitmaps to unblock
3363 error_reportf_err(local_err
,
3364 "%s: Failed to make dirty bitmaps writable: ",
3365 bdrv_get_node_name(bs
));
3371 * Abort the reopen, and delete and free the staged changes in
3374 void bdrv_reopen_abort(BDRVReopenState
*reopen_state
)
3378 assert(reopen_state
!= NULL
);
3379 drv
= reopen_state
->bs
->drv
;
3380 assert(drv
!= NULL
);
3382 if (drv
->bdrv_reopen_abort
) {
3383 drv
->bdrv_reopen_abort(reopen_state
);
3386 bdrv_abort_perm_update(reopen_state
->bs
);
3390 static void bdrv_close(BlockDriverState
*bs
)
3392 BdrvAioNotifier
*ban
, *ban_next
;
3393 BdrvChild
*child
, *next
;
3396 assert(!bs
->refcnt
);
3398 bdrv_drained_begin(bs
); /* complete I/O */
3400 bdrv_drain(bs
); /* in case flush left pending I/O */
3403 if (bs
->drv
->bdrv_close
) {
3404 bs
->drv
->bdrv_close(bs
);
3409 bdrv_set_backing_hd(bs
, NULL
, &error_abort
);
3411 if (bs
->file
!= NULL
) {
3412 bdrv_unref_child(bs
, bs
->file
);
3416 QLIST_FOREACH_SAFE(child
, &bs
->children
, next
, next
) {
3417 /* TODO Remove bdrv_unref() from drivers' close function and use
3418 * bdrv_unref_child() here */
3419 if (child
->bs
->inherits_from
== bs
) {
3420 child
->bs
->inherits_from
= NULL
;
3422 bdrv_detach_child(child
);
3427 atomic_set(&bs
->copy_on_read
, 0);
3428 bs
->backing_file
[0] = '\0';
3429 bs
->backing_format
[0] = '\0';
3430 bs
->total_sectors
= 0;
3431 bs
->encrypted
= false;
3433 qobject_unref(bs
->options
);
3434 qobject_unref(bs
->explicit_options
);
3436 bs
->explicit_options
= NULL
;
3437 qobject_unref(bs
->full_open_options
);
3438 bs
->full_open_options
= NULL
;
3440 bdrv_release_named_dirty_bitmaps(bs
);
3441 assert(QLIST_EMPTY(&bs
->dirty_bitmaps
));
3443 QLIST_FOREACH_SAFE(ban
, &bs
->aio_notifiers
, list
, ban_next
) {
3446 QLIST_INIT(&bs
->aio_notifiers
);
3447 bdrv_drained_end(bs
);
3450 void bdrv_close_all(void)
3452 assert(job_next(NULL
) == NULL
);
3453 nbd_export_close_all();
3455 /* Drop references from requests still in flight, such as canceled block
3456 * jobs whose AIO context has not been polled yet */
3459 blk_remove_all_bs();
3460 blockdev_close_all_bdrv_states();
3462 assert(QTAILQ_EMPTY(&all_bdrv_states
));
3465 static bool should_update_child(BdrvChild
*c
, BlockDriverState
*to
)
3469 if (c
->role
->stay_at_node
) {
3473 /* If the child @c belongs to the BDS @to, replacing the current
3474 * c->bs by @to would mean to create a loop.
3476 * Such a case occurs when appending a BDS to a backing chain.
3477 * For instance, imagine the following chain:
3479 * guest device -> node A -> further backing chain...
3481 * Now we create a new BDS B which we want to put on top of this
3482 * chain, so we first attach A as its backing node:
3487 * guest device -> node A -> further backing chain...
3489 * Finally we want to replace A by B. When doing that, we want to
3490 * replace all pointers to A by pointers to B -- except for the
3491 * pointer from B because (1) that would create a loop, and (2)
3492 * that pointer should simply stay intact:
3494 * guest device -> node B
3497 * node A -> further backing chain...
3499 * In general, when replacing a node A (c->bs) by a node B (@to),
3500 * if A is a child of B, that means we cannot replace A by B there
3501 * because that would create a loop. Silently detaching A from B
3502 * is also not really an option. So overall just leaving A in
3503 * place there is the most sensible choice. */
3504 QLIST_FOREACH(to_c
, &to
->children
, next
) {
3513 void bdrv_replace_node(BlockDriverState
*from
, BlockDriverState
*to
,
3516 BdrvChild
*c
, *next
;
3517 GSList
*list
= NULL
, *p
;
3518 uint64_t old_perm
, old_shared
;
3519 uint64_t perm
= 0, shared
= BLK_PERM_ALL
;
3522 assert(!atomic_read(&from
->in_flight
));
3523 assert(!atomic_read(&to
->in_flight
));
3525 /* Make sure that @from doesn't go away until we have successfully attached
3526 * all of its parents to @to. */
3529 /* Put all parents into @list and calculate their cumulative permissions */
3530 QLIST_FOREACH_SAFE(c
, &from
->parents
, next_parent
, next
) {
3531 assert(c
->bs
== from
);
3532 if (!should_update_child(c
, to
)) {
3535 list
= g_slist_prepend(list
, c
);
3537 shared
&= c
->shared_perm
;
3540 /* Check whether the required permissions can be granted on @to, ignoring
3541 * all BdrvChild in @list so that they can't block themselves. */
3542 ret
= bdrv_check_update_perm(to
, NULL
, perm
, shared
, list
, errp
);
3544 bdrv_abort_perm_update(to
);
3548 /* Now actually perform the change. We performed the permission check for
3549 * all elements of @list at once, so set the permissions all at once at the
3551 for (p
= list
; p
!= NULL
; p
= p
->next
) {
3555 bdrv_replace_child_noperm(c
, to
);
3559 bdrv_get_cumulative_perm(to
, &old_perm
, &old_shared
);
3560 bdrv_set_perm(to
, old_perm
| perm
, old_shared
| shared
);
3568 * Add new bs contents at the top of an image chain while the chain is
3569 * live, while keeping required fields on the top layer.
3571 * This will modify the BlockDriverState fields, and swap contents
3572 * between bs_new and bs_top. Both bs_new and bs_top are modified.
3574 * bs_new must not be attached to a BlockBackend.
3576 * This function does not create any image files.
3578 * bdrv_append() takes ownership of a bs_new reference and unrefs it because
3579 * that's what the callers commonly need. bs_new will be referenced by the old
3580 * parents of bs_top after bdrv_append() returns. If the caller needs to keep a
3581 * reference of its own, it must call bdrv_ref().
3583 void bdrv_append(BlockDriverState
*bs_new
, BlockDriverState
*bs_top
,
3586 Error
*local_err
= NULL
;
3588 bdrv_set_backing_hd(bs_new
, bs_top
, &local_err
);
3590 error_propagate(errp
, local_err
);
3594 bdrv_replace_node(bs_top
, bs_new
, &local_err
);
3596 error_propagate(errp
, local_err
);
3597 bdrv_set_backing_hd(bs_new
, NULL
, &error_abort
);
3601 /* bs_new is now referenced by its new parents, we don't need the
3602 * additional reference any more. */
3607 static void bdrv_delete(BlockDriverState
*bs
)
3610 assert(bdrv_op_blocker_is_empty(bs
));
3611 assert(!bs
->refcnt
);
3615 /* remove from list, if necessary */
3616 if (bs
->node_name
[0] != '\0') {
3617 QTAILQ_REMOVE(&graph_bdrv_states
, bs
, node_list
);
3619 QTAILQ_REMOVE(&all_bdrv_states
, bs
, bs_list
);
3625 * Run consistency checks on an image
3627 * Returns 0 if the check could be completed (it doesn't mean that the image is
3628 * free of errors) or -errno when an internal error occurred. The results of the
3629 * check are stored in res.
3631 static int coroutine_fn
bdrv_co_check(BlockDriverState
*bs
,
3632 BdrvCheckResult
*res
, BdrvCheckMode fix
)
3634 if (bs
->drv
== NULL
) {
3637 if (bs
->drv
->bdrv_co_check
== NULL
) {
3641 memset(res
, 0, sizeof(*res
));
3642 return bs
->drv
->bdrv_co_check(bs
, res
, fix
);
3645 typedef struct CheckCo
{
3646 BlockDriverState
*bs
;
3647 BdrvCheckResult
*res
;
3652 static void bdrv_check_co_entry(void *opaque
)
3654 CheckCo
*cco
= opaque
;
3655 cco
->ret
= bdrv_co_check(cco
->bs
, cco
->res
, cco
->fix
);
3658 int bdrv_check(BlockDriverState
*bs
,
3659 BdrvCheckResult
*res
, BdrvCheckMode fix
)
3665 .ret
= -EINPROGRESS
,
3669 if (qemu_in_coroutine()) {
3670 /* Fast-path if already in coroutine context */
3671 bdrv_check_co_entry(&cco
);
3673 co
= qemu_coroutine_create(bdrv_check_co_entry
, &cco
);
3674 qemu_coroutine_enter(co
);
3675 BDRV_POLL_WHILE(bs
, cco
.ret
== -EINPROGRESS
);
3684 * -EINVAL - backing format specified, but no file
3685 * -ENOSPC - can't update the backing file because no space is left in the
3687 * -ENOTSUP - format driver doesn't support changing the backing file
3689 int bdrv_change_backing_file(BlockDriverState
*bs
,
3690 const char *backing_file
, const char *backing_fmt
)
3692 BlockDriver
*drv
= bs
->drv
;
3699 /* Backing file format doesn't make sense without a backing file */
3700 if (backing_fmt
&& !backing_file
) {
3704 if (drv
->bdrv_change_backing_file
!= NULL
) {
3705 ret
= drv
->bdrv_change_backing_file(bs
, backing_file
, backing_fmt
);
3711 pstrcpy(bs
->backing_file
, sizeof(bs
->backing_file
), backing_file
?: "");
3712 pstrcpy(bs
->backing_format
, sizeof(bs
->backing_format
), backing_fmt
?: "");
3718 * Finds the image layer in the chain that has 'bs' as its backing file.
3720 * active is the current topmost image.
3722 * Returns NULL if bs is not found in active's image chain,
3723 * or if active == bs.
3725 * Returns the bottommost base image if bs == NULL.
3727 BlockDriverState
*bdrv_find_overlay(BlockDriverState
*active
,
3728 BlockDriverState
*bs
)
3730 while (active
&& bs
!= backing_bs(active
)) {
3731 active
= backing_bs(active
);
3737 /* Given a BDS, searches for the base layer. */
3738 BlockDriverState
*bdrv_find_base(BlockDriverState
*bs
)
3740 return bdrv_find_overlay(bs
, NULL
);
3744 * Drops images above 'base' up to and including 'top', and sets the image
3745 * above 'top' to have base as its backing file.
3747 * Requires that the overlay to 'top' is opened r/w, so that the backing file
3748 * information in 'bs' can be properly updated.
3750 * E.g., this will convert the following chain:
3751 * bottom <- base <- intermediate <- top <- active
3755 * bottom <- base <- active
3757 * It is allowed for bottom==base, in which case it converts:
3759 * base <- intermediate <- top <- active
3765 * If backing_file_str is non-NULL, it will be used when modifying top's
3766 * overlay image metadata.
3769 * if active == top, that is considered an error
3772 int bdrv_drop_intermediate(BlockDriverState
*top
, BlockDriverState
*base
,
3773 const char *backing_file_str
)
3775 BdrvChild
*c
, *next
;
3776 Error
*local_err
= NULL
;
3781 if (!top
->drv
|| !base
->drv
) {
3785 /* Make sure that base is in the backing chain of top */
3786 if (!bdrv_chain_contains(top
, base
)) {
3790 /* success - we can delete the intermediate states, and link top->base */
3791 /* TODO Check graph modification op blockers (BLK_PERM_GRAPH_MOD) once
3792 * we've figured out how they should work. */
3793 backing_file_str
= backing_file_str
? backing_file_str
: base
->filename
;
3795 QLIST_FOREACH_SAFE(c
, &top
->parents
, next_parent
, next
) {
3796 /* Check whether we are allowed to switch c from top to base */
3797 GSList
*ignore_children
= g_slist_prepend(NULL
, c
);
3798 bdrv_check_update_perm(base
, NULL
, c
->perm
, c
->shared_perm
,
3799 ignore_children
, &local_err
);
3800 g_slist_free(ignore_children
);
3803 error_report_err(local_err
);
3807 /* If so, update the backing file path in the image file */
3808 if (c
->role
->update_filename
) {
3809 ret
= c
->role
->update_filename(c
, base
, backing_file_str
,
3812 bdrv_abort_perm_update(base
);
3813 error_report_err(local_err
);
3818 /* Do the actual switch in the in-memory graph.
3819 * Completes bdrv_check_update_perm() transaction internally. */
3821 bdrv_replace_child(c
, base
);
3832 * Length of a allocated file in bytes. Sparse files are counted by actual
3833 * allocated space. Return < 0 if error or unknown.
3835 int64_t bdrv_get_allocated_file_size(BlockDriverState
*bs
)
3837 BlockDriver
*drv
= bs
->drv
;
3841 if (drv
->bdrv_get_allocated_file_size
) {
3842 return drv
->bdrv_get_allocated_file_size(bs
);
3845 return bdrv_get_allocated_file_size(bs
->file
->bs
);
3852 * @drv: Format driver
3853 * @opts: Creation options for new image
3854 * @in_bs: Existing image containing data for new image (may be NULL)
3855 * @errp: Error object
3856 * Returns: A #BlockMeasureInfo (free using qapi_free_BlockMeasureInfo())
3859 * Calculate file size required to create a new image.
3861 * If @in_bs is given then space for allocated clusters and zero clusters
3862 * from that image are included in the calculation. If @opts contains a
3863 * backing file that is shared by @in_bs then backing clusters may be omitted
3864 * from the calculation.
3866 * If @in_bs is NULL then the calculation includes no allocated clusters
3867 * unless a preallocation option is given in @opts.
3869 * Note that @in_bs may use a different BlockDriver from @drv.
3871 * If an error occurs the @errp pointer is set.
3873 BlockMeasureInfo
*bdrv_measure(BlockDriver
*drv
, QemuOpts
*opts
,
3874 BlockDriverState
*in_bs
, Error
**errp
)
3876 if (!drv
->bdrv_measure
) {
3877 error_setg(errp
, "Block driver '%s' does not support size measurement",
3882 return drv
->bdrv_measure(opts
, in_bs
, errp
);
3886 * Return number of sectors on success, -errno on error.
3888 int64_t bdrv_nb_sectors(BlockDriverState
*bs
)
3890 BlockDriver
*drv
= bs
->drv
;
3895 if (drv
->has_variable_length
) {
3896 int ret
= refresh_total_sectors(bs
, bs
->total_sectors
);
3901 return bs
->total_sectors
;
3905 * Return length in bytes on success, -errno on error.
3906 * The length is always a multiple of BDRV_SECTOR_SIZE.
3908 int64_t bdrv_getlength(BlockDriverState
*bs
)
3910 int64_t ret
= bdrv_nb_sectors(bs
);
3912 ret
= ret
> INT64_MAX
/ BDRV_SECTOR_SIZE
? -EFBIG
: ret
;
3913 return ret
< 0 ? ret
: ret
* BDRV_SECTOR_SIZE
;
3916 /* return 0 as number of sectors if no device present or error */
3917 void bdrv_get_geometry(BlockDriverState
*bs
, uint64_t *nb_sectors_ptr
)
3919 int64_t nb_sectors
= bdrv_nb_sectors(bs
);
3921 *nb_sectors_ptr
= nb_sectors
< 0 ? 0 : nb_sectors
;
3924 bool bdrv_is_sg(BlockDriverState
*bs
)
3929 bool bdrv_is_encrypted(BlockDriverState
*bs
)
3931 if (bs
->backing
&& bs
->backing
->bs
->encrypted
) {
3934 return bs
->encrypted
;
3937 const char *bdrv_get_format_name(BlockDriverState
*bs
)
3939 return bs
->drv
? bs
->drv
->format_name
: NULL
;
3942 static int qsort_strcmp(const void *a
, const void *b
)
3944 return strcmp(*(char *const *)a
, *(char *const *)b
);
3947 void bdrv_iterate_format(void (*it
)(void *opaque
, const char *name
),
3953 const char **formats
= NULL
;
3955 QLIST_FOREACH(drv
, &bdrv_drivers
, list
) {
3956 if (drv
->format_name
) {
3959 while (formats
&& i
&& !found
) {
3960 found
= !strcmp(formats
[--i
], drv
->format_name
);
3964 formats
= g_renew(const char *, formats
, count
+ 1);
3965 formats
[count
++] = drv
->format_name
;
3970 for (i
= 0; i
< (int)ARRAY_SIZE(block_driver_modules
); i
++) {
3971 const char *format_name
= block_driver_modules
[i
].format_name
;
3977 while (formats
&& j
&& !found
) {
3978 found
= !strcmp(formats
[--j
], format_name
);
3982 formats
= g_renew(const char *, formats
, count
+ 1);
3983 formats
[count
++] = format_name
;
3988 qsort(formats
, count
, sizeof(formats
[0]), qsort_strcmp
);
3990 for (i
= 0; i
< count
; i
++) {
3991 it(opaque
, formats
[i
]);
3997 /* This function is to find a node in the bs graph */
3998 BlockDriverState
*bdrv_find_node(const char *node_name
)
4000 BlockDriverState
*bs
;
4004 QTAILQ_FOREACH(bs
, &graph_bdrv_states
, node_list
) {
4005 if (!strcmp(node_name
, bs
->node_name
)) {
4012 /* Put this QMP function here so it can access the static graph_bdrv_states. */
4013 BlockDeviceInfoList
*bdrv_named_nodes_list(Error
**errp
)
4015 BlockDeviceInfoList
*list
, *entry
;
4016 BlockDriverState
*bs
;
4019 QTAILQ_FOREACH(bs
, &graph_bdrv_states
, node_list
) {
4020 BlockDeviceInfo
*info
= bdrv_block_device_info(NULL
, bs
, errp
);
4022 qapi_free_BlockDeviceInfoList(list
);
4025 entry
= g_malloc0(sizeof(*entry
));
4026 entry
->value
= info
;
4034 BlockDriverState
*bdrv_lookup_bs(const char *device
,
4035 const char *node_name
,
4039 BlockDriverState
*bs
;
4042 blk
= blk_by_name(device
);
4047 error_setg(errp
, "Device '%s' has no medium", device
);
4055 bs
= bdrv_find_node(node_name
);
4062 error_setg(errp
, "Cannot find device=%s nor node_name=%s",
4063 device
? device
: "",
4064 node_name
? node_name
: "");
4068 /* If 'base' is in the same chain as 'top', return true. Otherwise,
4069 * return false. If either argument is NULL, return false. */
4070 bool bdrv_chain_contains(BlockDriverState
*top
, BlockDriverState
*base
)
4072 while (top
&& top
!= base
) {
4073 top
= backing_bs(top
);
4079 BlockDriverState
*bdrv_next_node(BlockDriverState
*bs
)
4082 return QTAILQ_FIRST(&graph_bdrv_states
);
4084 return QTAILQ_NEXT(bs
, node_list
);
4087 BlockDriverState
*bdrv_next_all_states(BlockDriverState
*bs
)
4090 return QTAILQ_FIRST(&all_bdrv_states
);
4092 return QTAILQ_NEXT(bs
, bs_list
);
4095 const char *bdrv_get_node_name(const BlockDriverState
*bs
)
4097 return bs
->node_name
;
4100 const char *bdrv_get_parent_name(const BlockDriverState
*bs
)
4105 /* If multiple parents have a name, just pick the first one. */
4106 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
4107 if (c
->role
->get_name
) {
4108 name
= c
->role
->get_name(c
);
4109 if (name
&& *name
) {
4118 /* TODO check what callers really want: bs->node_name or blk_name() */
4119 const char *bdrv_get_device_name(const BlockDriverState
*bs
)
4121 return bdrv_get_parent_name(bs
) ?: "";
4124 /* This can be used to identify nodes that might not have a device
4125 * name associated. Since node and device names live in the same
4126 * namespace, the result is unambiguous. The exception is if both are
4127 * absent, then this returns an empty (non-null) string. */
4128 const char *bdrv_get_device_or_node_name(const BlockDriverState
*bs
)
4130 return bdrv_get_parent_name(bs
) ?: bs
->node_name
;
4133 int bdrv_get_flags(BlockDriverState
*bs
)
4135 return bs
->open_flags
;
4138 int bdrv_has_zero_init_1(BlockDriverState
*bs
)
4143 int bdrv_has_zero_init(BlockDriverState
*bs
)
4149 /* If BS is a copy on write image, it is initialized to
4150 the contents of the base image, which may not be zeroes. */
4154 if (bs
->drv
->bdrv_has_zero_init
) {
4155 return bs
->drv
->bdrv_has_zero_init(bs
);
4157 if (bs
->file
&& bs
->drv
->is_filter
) {
4158 return bdrv_has_zero_init(bs
->file
->bs
);
4165 bool bdrv_unallocated_blocks_are_zero(BlockDriverState
*bs
)
4167 BlockDriverInfo bdi
;
4173 if (bdrv_get_info(bs
, &bdi
) == 0) {
4174 return bdi
.unallocated_blocks_are_zero
;
4180 bool bdrv_can_write_zeroes_with_unmap(BlockDriverState
*bs
)
4182 if (!(bs
->open_flags
& BDRV_O_UNMAP
)) {
4186 return bs
->supported_zero_flags
& BDRV_REQ_MAY_UNMAP
;
4189 const char *bdrv_get_encrypted_filename(BlockDriverState
*bs
)
4191 if (bs
->backing
&& bs
->backing
->bs
->encrypted
)
4192 return bs
->backing_file
;
4193 else if (bs
->encrypted
)
4194 return bs
->filename
;
4199 void bdrv_get_backing_filename(BlockDriverState
*bs
,
4200 char *filename
, int filename_size
)
4202 pstrcpy(filename
, filename_size
, bs
->backing_file
);
4205 int bdrv_get_info(BlockDriverState
*bs
, BlockDriverInfo
*bdi
)
4207 BlockDriver
*drv
= bs
->drv
;
4208 /* if bs->drv == NULL, bs is closed, so there's nothing to do here */
4212 if (!drv
->bdrv_get_info
) {
4213 if (bs
->file
&& drv
->is_filter
) {
4214 return bdrv_get_info(bs
->file
->bs
, bdi
);
4218 memset(bdi
, 0, sizeof(*bdi
));
4219 return drv
->bdrv_get_info(bs
, bdi
);
4222 ImageInfoSpecific
*bdrv_get_specific_info(BlockDriverState
*bs
)
4224 BlockDriver
*drv
= bs
->drv
;
4225 if (drv
&& drv
->bdrv_get_specific_info
) {
4226 return drv
->bdrv_get_specific_info(bs
);
4231 void bdrv_debug_event(BlockDriverState
*bs
, BlkdebugEvent event
)
4233 if (!bs
|| !bs
->drv
|| !bs
->drv
->bdrv_debug_event
) {
4237 bs
->drv
->bdrv_debug_event(bs
, event
);
4240 int bdrv_debug_breakpoint(BlockDriverState
*bs
, const char *event
,
4243 while (bs
&& bs
->drv
&& !bs
->drv
->bdrv_debug_breakpoint
) {
4244 bs
= bs
->file
? bs
->file
->bs
: NULL
;
4247 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_breakpoint
) {
4248 return bs
->drv
->bdrv_debug_breakpoint(bs
, event
, tag
);
4254 int bdrv_debug_remove_breakpoint(BlockDriverState
*bs
, const char *tag
)
4256 while (bs
&& bs
->drv
&& !bs
->drv
->bdrv_debug_remove_breakpoint
) {
4257 bs
= bs
->file
? bs
->file
->bs
: NULL
;
4260 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_remove_breakpoint
) {
4261 return bs
->drv
->bdrv_debug_remove_breakpoint(bs
, tag
);
4267 int bdrv_debug_resume(BlockDriverState
*bs
, const char *tag
)
4269 while (bs
&& (!bs
->drv
|| !bs
->drv
->bdrv_debug_resume
)) {
4270 bs
= bs
->file
? bs
->file
->bs
: NULL
;
4273 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_resume
) {
4274 return bs
->drv
->bdrv_debug_resume(bs
, tag
);
4280 bool bdrv_debug_is_suspended(BlockDriverState
*bs
, const char *tag
)
4282 while (bs
&& bs
->drv
&& !bs
->drv
->bdrv_debug_is_suspended
) {
4283 bs
= bs
->file
? bs
->file
->bs
: NULL
;
4286 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_is_suspended
) {
4287 return bs
->drv
->bdrv_debug_is_suspended(bs
, tag
);
4293 /* backing_file can either be relative, or absolute, or a protocol. If it is
4294 * relative, it must be relative to the chain. So, passing in bs->filename
4295 * from a BDS as backing_file should not be done, as that may be relative to
4296 * the CWD rather than the chain. */
4297 BlockDriverState
*bdrv_find_backing_image(BlockDriverState
*bs
,
4298 const char *backing_file
)
4300 char *filename_full
= NULL
;
4301 char *backing_file_full
= NULL
;
4302 char *filename_tmp
= NULL
;
4303 int is_protocol
= 0;
4304 BlockDriverState
*curr_bs
= NULL
;
4305 BlockDriverState
*retval
= NULL
;
4306 Error
*local_error
= NULL
;
4308 if (!bs
|| !bs
->drv
|| !backing_file
) {
4312 filename_full
= g_malloc(PATH_MAX
);
4313 backing_file_full
= g_malloc(PATH_MAX
);
4314 filename_tmp
= g_malloc(PATH_MAX
);
4316 is_protocol
= path_has_protocol(backing_file
);
4318 for (curr_bs
= bs
; curr_bs
->backing
; curr_bs
= curr_bs
->backing
->bs
) {
4320 /* If either of the filename paths is actually a protocol, then
4321 * compare unmodified paths; otherwise make paths relative */
4322 if (is_protocol
|| path_has_protocol(curr_bs
->backing_file
)) {
4323 if (strcmp(backing_file
, curr_bs
->backing_file
) == 0) {
4324 retval
= curr_bs
->backing
->bs
;
4327 /* Also check against the full backing filename for the image */
4328 bdrv_get_full_backing_filename(curr_bs
, backing_file_full
, PATH_MAX
,
4330 if (local_error
== NULL
) {
4331 if (strcmp(backing_file
, backing_file_full
) == 0) {
4332 retval
= curr_bs
->backing
->bs
;
4336 error_free(local_error
);
4340 /* If not an absolute filename path, make it relative to the current
4341 * image's filename path */
4342 path_combine(filename_tmp
, PATH_MAX
, curr_bs
->filename
,
4345 /* We are going to compare absolute pathnames */
4346 if (!realpath(filename_tmp
, filename_full
)) {
4350 /* We need to make sure the backing filename we are comparing against
4351 * is relative to the current image filename (or absolute) */
4352 path_combine(filename_tmp
, PATH_MAX
, curr_bs
->filename
,
4353 curr_bs
->backing_file
);
4355 if (!realpath(filename_tmp
, backing_file_full
)) {
4359 if (strcmp(backing_file_full
, filename_full
) == 0) {
4360 retval
= curr_bs
->backing
->bs
;
4366 g_free(filename_full
);
4367 g_free(backing_file_full
);
4368 g_free(filename_tmp
);
4372 void bdrv_init(void)
4374 module_call_init(MODULE_INIT_BLOCK
);
4377 void bdrv_init_with_whitelist(void)
4379 use_bdrv_whitelist
= 1;
4383 static void coroutine_fn
bdrv_co_invalidate_cache(BlockDriverState
*bs
,
4386 BdrvChild
*child
, *parent
;
4387 uint64_t perm
, shared_perm
;
4388 Error
*local_err
= NULL
;
4395 if (!(bs
->open_flags
& BDRV_O_INACTIVE
)) {
4399 QLIST_FOREACH(child
, &bs
->children
, next
) {
4400 bdrv_co_invalidate_cache(child
->bs
, &local_err
);
4402 error_propagate(errp
, local_err
);
4408 * Update permissions, they may differ for inactive nodes.
4410 * Note that the required permissions of inactive images are always a
4411 * subset of the permissions required after activating the image. This
4412 * allows us to just get the permissions upfront without restricting
4413 * drv->bdrv_invalidate_cache().
4415 * It also means that in error cases, we don't have to try and revert to
4416 * the old permissions (which is an operation that could fail, too). We can
4417 * just keep the extended permissions for the next time that an activation
4418 * of the image is tried.
4420 bs
->open_flags
&= ~BDRV_O_INACTIVE
;
4421 bdrv_get_cumulative_perm(bs
, &perm
, &shared_perm
);
4422 ret
= bdrv_check_perm(bs
, NULL
, perm
, shared_perm
, NULL
, &local_err
);
4424 bs
->open_flags
|= BDRV_O_INACTIVE
;
4425 error_propagate(errp
, local_err
);
4428 bdrv_set_perm(bs
, perm
, shared_perm
);
4430 if (bs
->drv
->bdrv_co_invalidate_cache
) {
4431 bs
->drv
->bdrv_co_invalidate_cache(bs
, &local_err
);
4433 bs
->open_flags
|= BDRV_O_INACTIVE
;
4434 error_propagate(errp
, local_err
);
4439 ret
= refresh_total_sectors(bs
, bs
->total_sectors
);
4441 bs
->open_flags
|= BDRV_O_INACTIVE
;
4442 error_setg_errno(errp
, -ret
, "Could not refresh total sector count");
4446 QLIST_FOREACH(parent
, &bs
->parents
, next_parent
) {
4447 if (parent
->role
->activate
) {
4448 parent
->role
->activate(parent
, &local_err
);
4450 error_propagate(errp
, local_err
);
4457 typedef struct InvalidateCacheCo
{
4458 BlockDriverState
*bs
;
4461 } InvalidateCacheCo
;
4463 static void coroutine_fn
bdrv_invalidate_cache_co_entry(void *opaque
)
4465 InvalidateCacheCo
*ico
= opaque
;
4466 bdrv_co_invalidate_cache(ico
->bs
, ico
->errp
);
4470 void bdrv_invalidate_cache(BlockDriverState
*bs
, Error
**errp
)
4473 InvalidateCacheCo ico
= {
4479 if (qemu_in_coroutine()) {
4480 /* Fast-path if already in coroutine context */
4481 bdrv_invalidate_cache_co_entry(&ico
);
4483 co
= qemu_coroutine_create(bdrv_invalidate_cache_co_entry
, &ico
);
4484 qemu_coroutine_enter(co
);
4485 BDRV_POLL_WHILE(bs
, !ico
.done
);
4489 void bdrv_invalidate_cache_all(Error
**errp
)
4491 BlockDriverState
*bs
;
4492 Error
*local_err
= NULL
;
4493 BdrvNextIterator it
;
4495 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
4496 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
4498 aio_context_acquire(aio_context
);
4499 bdrv_invalidate_cache(bs
, &local_err
);
4500 aio_context_release(aio_context
);
4502 error_propagate(errp
, local_err
);
4503 bdrv_next_cleanup(&it
);
4509 static int bdrv_inactivate_recurse(BlockDriverState
*bs
,
4512 BdrvChild
*child
, *parent
;
4519 if (!setting_flag
&& bs
->drv
->bdrv_inactivate
) {
4520 ret
= bs
->drv
->bdrv_inactivate(bs
);
4526 if (setting_flag
&& !(bs
->open_flags
& BDRV_O_INACTIVE
)) {
4527 uint64_t perm
, shared_perm
;
4529 QLIST_FOREACH(parent
, &bs
->parents
, next_parent
) {
4530 if (parent
->role
->inactivate
) {
4531 ret
= parent
->role
->inactivate(parent
);
4538 bs
->open_flags
|= BDRV_O_INACTIVE
;
4540 /* Update permissions, they may differ for inactive nodes */
4541 bdrv_get_cumulative_perm(bs
, &perm
, &shared_perm
);
4542 bdrv_check_perm(bs
, NULL
, perm
, shared_perm
, NULL
, &error_abort
);
4543 bdrv_set_perm(bs
, perm
, shared_perm
);
4546 QLIST_FOREACH(child
, &bs
->children
, next
) {
4547 ret
= bdrv_inactivate_recurse(child
->bs
, setting_flag
);
4553 /* At this point persistent bitmaps should be already stored by the format
4555 bdrv_release_persistent_dirty_bitmaps(bs
);
4560 int bdrv_inactivate_all(void)
4562 BlockDriverState
*bs
= NULL
;
4563 BdrvNextIterator it
;
4566 GSList
*aio_ctxs
= NULL
, *ctx
;
4568 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
4569 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
4571 if (!g_slist_find(aio_ctxs
, aio_context
)) {
4572 aio_ctxs
= g_slist_prepend(aio_ctxs
, aio_context
);
4573 aio_context_acquire(aio_context
);
4577 /* We do two passes of inactivation. The first pass calls to drivers'
4578 * .bdrv_inactivate callbacks recursively so all cache is flushed to disk;
4579 * the second pass sets the BDRV_O_INACTIVE flag so that no further write
4581 for (pass
= 0; pass
< 2; pass
++) {
4582 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
4583 ret
= bdrv_inactivate_recurse(bs
, pass
);
4585 bdrv_next_cleanup(&it
);
4592 for (ctx
= aio_ctxs
; ctx
!= NULL
; ctx
= ctx
->next
) {
4593 AioContext
*aio_context
= ctx
->data
;
4594 aio_context_release(aio_context
);
4596 g_slist_free(aio_ctxs
);
4601 /**************************************************************/
4602 /* removable device support */
4605 * Return TRUE if the media is present
4607 bool bdrv_is_inserted(BlockDriverState
*bs
)
4609 BlockDriver
*drv
= bs
->drv
;
4615 if (drv
->bdrv_is_inserted
) {
4616 return drv
->bdrv_is_inserted(bs
);
4618 QLIST_FOREACH(child
, &bs
->children
, next
) {
4619 if (!bdrv_is_inserted(child
->bs
)) {
4627 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
4629 void bdrv_eject(BlockDriverState
*bs
, bool eject_flag
)
4631 BlockDriver
*drv
= bs
->drv
;
4633 if (drv
&& drv
->bdrv_eject
) {
4634 drv
->bdrv_eject(bs
, eject_flag
);
4639 * Lock or unlock the media (if it is locked, the user won't be able
4640 * to eject it manually).
4642 void bdrv_lock_medium(BlockDriverState
*bs
, bool locked
)
4644 BlockDriver
*drv
= bs
->drv
;
4646 trace_bdrv_lock_medium(bs
, locked
);
4648 if (drv
&& drv
->bdrv_lock_medium
) {
4649 drv
->bdrv_lock_medium(bs
, locked
);
4653 /* Get a reference to bs */
4654 void bdrv_ref(BlockDriverState
*bs
)
4659 /* Release a previously grabbed reference to bs.
4660 * If after releasing, reference count is zero, the BlockDriverState is
4662 void bdrv_unref(BlockDriverState
*bs
)
4667 assert(bs
->refcnt
> 0);
4668 if (--bs
->refcnt
== 0) {
4673 struct BdrvOpBlocker
{
4675 QLIST_ENTRY(BdrvOpBlocker
) list
;
4678 bool bdrv_op_is_blocked(BlockDriverState
*bs
, BlockOpType op
, Error
**errp
)
4680 BdrvOpBlocker
*blocker
;
4681 assert((int) op
>= 0 && op
< BLOCK_OP_TYPE_MAX
);
4682 if (!QLIST_EMPTY(&bs
->op_blockers
[op
])) {
4683 blocker
= QLIST_FIRST(&bs
->op_blockers
[op
]);
4684 error_propagate(errp
, error_copy(blocker
->reason
));
4685 error_prepend(errp
, "Node '%s' is busy: ",
4686 bdrv_get_device_or_node_name(bs
));
4692 void bdrv_op_block(BlockDriverState
*bs
, BlockOpType op
, Error
*reason
)
4694 BdrvOpBlocker
*blocker
;
4695 assert((int) op
>= 0 && op
< BLOCK_OP_TYPE_MAX
);
4697 blocker
= g_new0(BdrvOpBlocker
, 1);
4698 blocker
->reason
= reason
;
4699 QLIST_INSERT_HEAD(&bs
->op_blockers
[op
], blocker
, list
);
4702 void bdrv_op_unblock(BlockDriverState
*bs
, BlockOpType op
, Error
*reason
)
4704 BdrvOpBlocker
*blocker
, *next
;
4705 assert((int) op
>= 0 && op
< BLOCK_OP_TYPE_MAX
);
4706 QLIST_FOREACH_SAFE(blocker
, &bs
->op_blockers
[op
], list
, next
) {
4707 if (blocker
->reason
== reason
) {
4708 QLIST_REMOVE(blocker
, list
);
4714 void bdrv_op_block_all(BlockDriverState
*bs
, Error
*reason
)
4717 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
4718 bdrv_op_block(bs
, i
, reason
);
4722 void bdrv_op_unblock_all(BlockDriverState
*bs
, Error
*reason
)
4725 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
4726 bdrv_op_unblock(bs
, i
, reason
);
4730 bool bdrv_op_blocker_is_empty(BlockDriverState
*bs
)
4734 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
4735 if (!QLIST_EMPTY(&bs
->op_blockers
[i
])) {
4742 void bdrv_img_create(const char *filename
, const char *fmt
,
4743 const char *base_filename
, const char *base_fmt
,
4744 char *options
, uint64_t img_size
, int flags
, bool quiet
,
4747 QemuOptsList
*create_opts
= NULL
;
4748 QemuOpts
*opts
= NULL
;
4749 const char *backing_fmt
, *backing_file
;
4751 BlockDriver
*drv
, *proto_drv
;
4752 Error
*local_err
= NULL
;
4755 /* Find driver and parse its options */
4756 drv
= bdrv_find_format(fmt
);
4758 error_setg(errp
, "Unknown file format '%s'", fmt
);
4762 proto_drv
= bdrv_find_protocol(filename
, true, errp
);
4767 if (!drv
->create_opts
) {
4768 error_setg(errp
, "Format driver '%s' does not support image creation",
4773 if (!proto_drv
->create_opts
) {
4774 error_setg(errp
, "Protocol driver '%s' does not support image creation",
4775 proto_drv
->format_name
);
4779 create_opts
= qemu_opts_append(create_opts
, drv
->create_opts
);
4780 create_opts
= qemu_opts_append(create_opts
, proto_drv
->create_opts
);
4782 /* Create parameter list with default values */
4783 opts
= qemu_opts_create(create_opts
, NULL
, 0, &error_abort
);
4784 qemu_opt_set_number(opts
, BLOCK_OPT_SIZE
, img_size
, &error_abort
);
4786 /* Parse -o options */
4788 qemu_opts_do_parse(opts
, options
, NULL
, &local_err
);
4790 error_report_err(local_err
);
4792 error_setg(errp
, "Invalid options for file format '%s'", fmt
);
4797 if (base_filename
) {
4798 qemu_opt_set(opts
, BLOCK_OPT_BACKING_FILE
, base_filename
, &local_err
);
4800 error_setg(errp
, "Backing file not supported for file format '%s'",
4807 qemu_opt_set(opts
, BLOCK_OPT_BACKING_FMT
, base_fmt
, &local_err
);
4809 error_setg(errp
, "Backing file format not supported for file "
4810 "format '%s'", fmt
);
4815 backing_file
= qemu_opt_get(opts
, BLOCK_OPT_BACKING_FILE
);
4817 if (!strcmp(filename
, backing_file
)) {
4818 error_setg(errp
, "Error: Trying to create an image with the "
4819 "same filename as the backing file");
4824 backing_fmt
= qemu_opt_get(opts
, BLOCK_OPT_BACKING_FMT
);
4826 /* The size for the image must always be specified, unless we have a backing
4827 * file and we have not been forbidden from opening it. */
4828 size
= qemu_opt_get_size(opts
, BLOCK_OPT_SIZE
, img_size
);
4829 if (backing_file
&& !(flags
& BDRV_O_NO_BACKING
)) {
4830 BlockDriverState
*bs
;
4831 char *full_backing
= g_new0(char, PATH_MAX
);
4833 QDict
*backing_options
= NULL
;
4835 bdrv_get_full_backing_filename_from_filename(filename
, backing_file
,
4836 full_backing
, PATH_MAX
,
4839 g_free(full_backing
);
4843 /* backing files always opened read-only */
4845 back_flags
&= ~(BDRV_O_RDWR
| BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
);
4847 backing_options
= qdict_new();
4849 qdict_put_str(backing_options
, "driver", backing_fmt
);
4851 qdict_put_bool(backing_options
, BDRV_OPT_FORCE_SHARE
, true);
4853 bs
= bdrv_open(full_backing
, NULL
, backing_options
, back_flags
,
4855 g_free(full_backing
);
4856 if (!bs
&& size
!= -1) {
4857 /* Couldn't open BS, but we have a size, so it's nonfatal */
4858 warn_reportf_err(local_err
,
4859 "Could not verify backing image. "
4860 "This may become an error in future versions.\n");
4863 /* Couldn't open bs, do not have size */
4864 error_append_hint(&local_err
,
4865 "Could not open backing image to determine size.\n");
4869 /* Opened BS, have no size */
4870 size
= bdrv_getlength(bs
);
4872 error_setg_errno(errp
, -size
, "Could not get size of '%s'",
4877 qemu_opt_set_number(opts
, BLOCK_OPT_SIZE
, size
, &error_abort
);
4881 } /* (backing_file && !(flags & BDRV_O_NO_BACKING)) */
4884 error_setg(errp
, "Image creation needs a size parameter");
4889 printf("Formatting '%s', fmt=%s ", filename
, fmt
);
4890 qemu_opts_print(opts
, " ");
4894 ret
= bdrv_create(drv
, filename
, opts
, &local_err
);
4896 if (ret
== -EFBIG
) {
4897 /* This is generally a better message than whatever the driver would
4898 * deliver (especially because of the cluster_size_hint), since that
4899 * is most probably not much different from "image too large". */
4900 const char *cluster_size_hint
= "";
4901 if (qemu_opt_get_size(opts
, BLOCK_OPT_CLUSTER_SIZE
, 0)) {
4902 cluster_size_hint
= " (try using a larger cluster size)";
4904 error_setg(errp
, "The image size is too large for file format '%s'"
4905 "%s", fmt
, cluster_size_hint
);
4906 error_free(local_err
);
4911 qemu_opts_del(opts
);
4912 qemu_opts_free(create_opts
);
4913 error_propagate(errp
, local_err
);
4916 AioContext
*bdrv_get_aio_context(BlockDriverState
*bs
)
4918 return bs
? bs
->aio_context
: qemu_get_aio_context();
4921 void bdrv_coroutine_enter(BlockDriverState
*bs
, Coroutine
*co
)
4923 aio_co_enter(bdrv_get_aio_context(bs
), co
);
4926 static void bdrv_do_remove_aio_context_notifier(BdrvAioNotifier
*ban
)
4928 QLIST_REMOVE(ban
, list
);
4932 void bdrv_detach_aio_context(BlockDriverState
*bs
)
4934 BdrvAioNotifier
*baf
, *baf_tmp
;
4941 assert(!bs
->walking_aio_notifiers
);
4942 bs
->walking_aio_notifiers
= true;
4943 QLIST_FOREACH_SAFE(baf
, &bs
->aio_notifiers
, list
, baf_tmp
) {
4945 bdrv_do_remove_aio_context_notifier(baf
);
4947 baf
->detach_aio_context(baf
->opaque
);
4950 /* Never mind iterating again to check for ->deleted. bdrv_close() will
4951 * remove remaining aio notifiers if we aren't called again.
4953 bs
->walking_aio_notifiers
= false;
4955 if (bs
->drv
->bdrv_detach_aio_context
) {
4956 bs
->drv
->bdrv_detach_aio_context(bs
);
4958 QLIST_FOREACH(child
, &bs
->children
, next
) {
4959 bdrv_detach_aio_context(child
->bs
);
4962 bs
->aio_context
= NULL
;
4965 void bdrv_attach_aio_context(BlockDriverState
*bs
,
4966 AioContext
*new_context
)
4968 BdrvAioNotifier
*ban
, *ban_tmp
;
4975 bs
->aio_context
= new_context
;
4977 QLIST_FOREACH(child
, &bs
->children
, next
) {
4978 bdrv_attach_aio_context(child
->bs
, new_context
);
4980 if (bs
->drv
->bdrv_attach_aio_context
) {
4981 bs
->drv
->bdrv_attach_aio_context(bs
, new_context
);
4984 assert(!bs
->walking_aio_notifiers
);
4985 bs
->walking_aio_notifiers
= true;
4986 QLIST_FOREACH_SAFE(ban
, &bs
->aio_notifiers
, list
, ban_tmp
) {
4988 bdrv_do_remove_aio_context_notifier(ban
);
4990 ban
->attached_aio_context(new_context
, ban
->opaque
);
4993 bs
->walking_aio_notifiers
= false;
4996 void bdrv_set_aio_context(BlockDriverState
*bs
, AioContext
*new_context
)
4998 AioContext
*ctx
= bdrv_get_aio_context(bs
);
5000 aio_disable_external(ctx
);
5001 bdrv_parent_drained_begin(bs
, NULL
, false);
5002 bdrv_drain(bs
); /* ensure there are no in-flight requests */
5004 while (aio_poll(ctx
, false)) {
5005 /* wait for all bottom halves to execute */
5008 bdrv_detach_aio_context(bs
);
5010 /* This function executes in the old AioContext so acquire the new one in
5011 * case it runs in a different thread.
5013 aio_context_acquire(new_context
);
5014 bdrv_attach_aio_context(bs
, new_context
);
5015 bdrv_parent_drained_end(bs
, NULL
, false);
5016 aio_enable_external(ctx
);
5017 aio_context_release(new_context
);
5020 void bdrv_add_aio_context_notifier(BlockDriverState
*bs
,
5021 void (*attached_aio_context
)(AioContext
*new_context
, void *opaque
),
5022 void (*detach_aio_context
)(void *opaque
), void *opaque
)
5024 BdrvAioNotifier
*ban
= g_new(BdrvAioNotifier
, 1);
5025 *ban
= (BdrvAioNotifier
){
5026 .attached_aio_context
= attached_aio_context
,
5027 .detach_aio_context
= detach_aio_context
,
5031 QLIST_INSERT_HEAD(&bs
->aio_notifiers
, ban
, list
);
5034 void bdrv_remove_aio_context_notifier(BlockDriverState
*bs
,
5035 void (*attached_aio_context
)(AioContext
*,
5037 void (*detach_aio_context
)(void *),
5040 BdrvAioNotifier
*ban
, *ban_next
;
5042 QLIST_FOREACH_SAFE(ban
, &bs
->aio_notifiers
, list
, ban_next
) {
5043 if (ban
->attached_aio_context
== attached_aio_context
&&
5044 ban
->detach_aio_context
== detach_aio_context
&&
5045 ban
->opaque
== opaque
&&
5046 ban
->deleted
== false)
5048 if (bs
->walking_aio_notifiers
) {
5049 ban
->deleted
= true;
5051 bdrv_do_remove_aio_context_notifier(ban
);
5060 int bdrv_amend_options(BlockDriverState
*bs
, QemuOpts
*opts
,
5061 BlockDriverAmendStatusCB
*status_cb
, void *cb_opaque
,
5065 error_setg(errp
, "Node is ejected");
5068 if (!bs
->drv
->bdrv_amend_options
) {
5069 error_setg(errp
, "Block driver '%s' does not support option amendment",
5070 bs
->drv
->format_name
);
5073 return bs
->drv
->bdrv_amend_options(bs
, opts
, status_cb
, cb_opaque
, errp
);
5076 /* This function will be called by the bdrv_recurse_is_first_non_filter method
5077 * of block filter and by bdrv_is_first_non_filter.
5078 * It is used to test if the given bs is the candidate or recurse more in the
5081 bool bdrv_recurse_is_first_non_filter(BlockDriverState
*bs
,
5082 BlockDriverState
*candidate
)
5084 /* return false if basic checks fails */
5085 if (!bs
|| !bs
->drv
) {
5089 /* the code reached a non block filter driver -> check if the bs is
5090 * the same as the candidate. It's the recursion termination condition.
5092 if (!bs
->drv
->is_filter
) {
5093 return bs
== candidate
;
5095 /* Down this path the driver is a block filter driver */
5097 /* If the block filter recursion method is defined use it to recurse down
5100 if (bs
->drv
->bdrv_recurse_is_first_non_filter
) {
5101 return bs
->drv
->bdrv_recurse_is_first_non_filter(bs
, candidate
);
5104 /* the driver is a block filter but don't allow to recurse -> return false
5109 /* This function checks if the candidate is the first non filter bs down it's
5110 * bs chain. Since we don't have pointers to parents it explore all bs chains
5111 * from the top. Some filters can choose not to pass down the recursion.
5113 bool bdrv_is_first_non_filter(BlockDriverState
*candidate
)
5115 BlockDriverState
*bs
;
5116 BdrvNextIterator it
;
5118 /* walk down the bs forest recursively */
5119 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
5122 /* try to recurse in this top level bs */
5123 perm
= bdrv_recurse_is_first_non_filter(bs
, candidate
);
5125 /* candidate is the first non filter */
5127 bdrv_next_cleanup(&it
);
5135 BlockDriverState
*check_to_replace_node(BlockDriverState
*parent_bs
,
5136 const char *node_name
, Error
**errp
)
5138 BlockDriverState
*to_replace_bs
= bdrv_find_node(node_name
);
5139 AioContext
*aio_context
;
5141 if (!to_replace_bs
) {
5142 error_setg(errp
, "Node name '%s' not found", node_name
);
5146 aio_context
= bdrv_get_aio_context(to_replace_bs
);
5147 aio_context_acquire(aio_context
);
5149 if (bdrv_op_is_blocked(to_replace_bs
, BLOCK_OP_TYPE_REPLACE
, errp
)) {
5150 to_replace_bs
= NULL
;
5154 /* We don't want arbitrary node of the BDS chain to be replaced only the top
5155 * most non filter in order to prevent data corruption.
5156 * Another benefit is that this tests exclude backing files which are
5157 * blocked by the backing blockers.
5159 if (!bdrv_recurse_is_first_non_filter(parent_bs
, to_replace_bs
)) {
5160 error_setg(errp
, "Only top most non filter can be replaced");
5161 to_replace_bs
= NULL
;
5166 aio_context_release(aio_context
);
5167 return to_replace_bs
;
5170 static bool append_open_options(QDict
*d
, BlockDriverState
*bs
)
5172 const QDictEntry
*entry
;
5174 bool found_any
= false;
5176 for (entry
= qdict_first(bs
->options
); entry
;
5177 entry
= qdict_next(bs
->options
, entry
))
5179 /* Exclude all non-driver-specific options */
5180 for (desc
= bdrv_runtime_opts
.desc
; desc
->name
; desc
++) {
5181 if (!strcmp(qdict_entry_key(entry
), desc
->name
)) {
5189 qdict_put_obj(d
, qdict_entry_key(entry
),
5190 qobject_ref(qdict_entry_value(entry
)));
5197 /* Updates the following BDS fields:
5198 * - exact_filename: A filename which may be used for opening a block device
5199 * which (mostly) equals the given BDS (even without any
5200 * other options; so reading and writing must return the same
5201 * results, but caching etc. may be different)
5202 * - full_open_options: Options which, when given when opening a block device
5203 * (without a filename), result in a BDS (mostly)
5204 * equalling the given one
5205 * - filename: If exact_filename is set, it is copied here. Otherwise,
5206 * full_open_options is converted to a JSON object, prefixed with
5207 * "json:" (for use through the JSON pseudo protocol) and put here.
5209 void bdrv_refresh_filename(BlockDriverState
*bs
)
5211 BlockDriver
*drv
= bs
->drv
;
5218 /* This BDS's file name will most probably depend on its file's name, so
5219 * refresh that first */
5221 bdrv_refresh_filename(bs
->file
->bs
);
5224 if (drv
->bdrv_refresh_filename
) {
5225 /* Obsolete information is of no use here, so drop the old file name
5226 * information before refreshing it */
5227 bs
->exact_filename
[0] = '\0';
5228 if (bs
->full_open_options
) {
5229 qobject_unref(bs
->full_open_options
);
5230 bs
->full_open_options
= NULL
;
5234 append_open_options(opts
, bs
);
5235 drv
->bdrv_refresh_filename(bs
, opts
);
5236 qobject_unref(opts
);
5237 } else if (bs
->file
) {
5238 /* Try to reconstruct valid information from the underlying file */
5239 bool has_open_options
;
5241 bs
->exact_filename
[0] = '\0';
5242 if (bs
->full_open_options
) {
5243 qobject_unref(bs
->full_open_options
);
5244 bs
->full_open_options
= NULL
;
5248 has_open_options
= append_open_options(opts
, bs
);
5250 /* If no specific options have been given for this BDS, the filename of
5251 * the underlying file should suffice for this one as well */
5252 if (bs
->file
->bs
->exact_filename
[0] && !has_open_options
) {
5253 strcpy(bs
->exact_filename
, bs
->file
->bs
->exact_filename
);
5255 /* Reconstructing the full options QDict is simple for most format block
5256 * drivers, as long as the full options are known for the underlying
5257 * file BDS. The full options QDict of that file BDS should somehow
5258 * contain a representation of the filename, therefore the following
5259 * suffices without querying the (exact_)filename of this BDS. */
5260 if (bs
->file
->bs
->full_open_options
) {
5261 qdict_put_str(opts
, "driver", drv
->format_name
);
5262 qdict_put(opts
, "file",
5263 qobject_ref(bs
->file
->bs
->full_open_options
));
5265 bs
->full_open_options
= opts
;
5267 qobject_unref(opts
);
5269 } else if (!bs
->full_open_options
&& qdict_size(bs
->options
)) {
5270 /* There is no underlying file BDS (at least referenced by BDS.file),
5271 * so the full options QDict should be equal to the options given
5272 * specifically for this block device when it was opened (plus the
5273 * driver specification).
5274 * Because those options don't change, there is no need to update
5275 * full_open_options when it's already set. */
5278 append_open_options(opts
, bs
);
5279 qdict_put_str(opts
, "driver", drv
->format_name
);
5281 if (bs
->exact_filename
[0]) {
5282 /* This may not work for all block protocol drivers (some may
5283 * require this filename to be parsed), but we have to find some
5284 * default solution here, so just include it. If some block driver
5285 * does not support pure options without any filename at all or
5286 * needs some special format of the options QDict, it needs to
5287 * implement the driver-specific bdrv_refresh_filename() function.
5289 qdict_put_str(opts
, "filename", bs
->exact_filename
);
5292 bs
->full_open_options
= opts
;
5295 if (bs
->exact_filename
[0]) {
5296 pstrcpy(bs
->filename
, sizeof(bs
->filename
), bs
->exact_filename
);
5297 } else if (bs
->full_open_options
) {
5298 QString
*json
= qobject_to_json(QOBJECT(bs
->full_open_options
));
5299 snprintf(bs
->filename
, sizeof(bs
->filename
), "json:%s",
5300 qstring_get_str(json
));
5301 qobject_unref(json
);
5306 * Hot add/remove a BDS's child. So the user can take a child offline when
5307 * it is broken and take a new child online
5309 void bdrv_add_child(BlockDriverState
*parent_bs
, BlockDriverState
*child_bs
,
5313 if (!parent_bs
->drv
|| !parent_bs
->drv
->bdrv_add_child
) {
5314 error_setg(errp
, "The node %s does not support adding a child",
5315 bdrv_get_device_or_node_name(parent_bs
));
5319 if (!QLIST_EMPTY(&child_bs
->parents
)) {
5320 error_setg(errp
, "The node %s already has a parent",
5321 child_bs
->node_name
);
5325 parent_bs
->drv
->bdrv_add_child(parent_bs
, child_bs
, errp
);
5328 void bdrv_del_child(BlockDriverState
*parent_bs
, BdrvChild
*child
, Error
**errp
)
5332 if (!parent_bs
->drv
|| !parent_bs
->drv
->bdrv_del_child
) {
5333 error_setg(errp
, "The node %s does not support removing a child",
5334 bdrv_get_device_or_node_name(parent_bs
));
5338 QLIST_FOREACH(tmp
, &parent_bs
->children
, next
) {
5345 error_setg(errp
, "The node %s does not have a child named %s",
5346 bdrv_get_device_or_node_name(parent_bs
),
5347 bdrv_get_device_or_node_name(child
->bs
));
5351 parent_bs
->drv
->bdrv_del_child(parent_bs
, child
, errp
);
5354 bool bdrv_can_store_new_dirty_bitmap(BlockDriverState
*bs
, const char *name
,
5355 uint32_t granularity
, Error
**errp
)
5357 BlockDriver
*drv
= bs
->drv
;
5360 error_setg_errno(errp
, ENOMEDIUM
,
5361 "Can't store persistent bitmaps to %s",
5362 bdrv_get_device_or_node_name(bs
));
5366 if (!drv
->bdrv_can_store_new_dirty_bitmap
) {
5367 error_setg_errno(errp
, ENOTSUP
,
5368 "Can't store persistent bitmaps to %s",
5369 bdrv_get_device_or_node_name(bs
));
5373 return drv
->bdrv_can_store_new_dirty_bitmap(bs
, name
, granularity
, errp
);