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
));
270 * Called by a driver that can only provide a read-only image.
272 * Returns 0 if the node is already read-only or it could switch the node to
273 * read-only because BDRV_O_AUTO_RDONLY is set.
275 * Returns -EACCES if the node is read-write and BDRV_O_AUTO_RDONLY is not set
276 * or bdrv_can_set_read_only() forbids making the node read-only. If @errmsg
277 * is not NULL, it is used as the error message for the Error object.
279 int bdrv_apply_auto_read_only(BlockDriverState
*bs
, const char *errmsg
,
284 if (!(bs
->open_flags
& BDRV_O_RDWR
)) {
287 if (!(bs
->open_flags
& BDRV_O_AUTO_RDONLY
)) {
291 ret
= bdrv_can_set_read_only(bs
, true, false, NULL
);
296 bs
->read_only
= true;
297 bs
->open_flags
&= ~BDRV_O_RDWR
;
302 error_setg(errp
, "%s", errmsg
?: "Image is read-only");
306 void bdrv_get_full_backing_filename_from_filename(const char *backed
,
308 char *dest
, size_t sz
,
311 if (backing
[0] == '\0' || path_has_protocol(backing
) ||
312 path_is_absolute(backing
))
314 pstrcpy(dest
, sz
, backing
);
315 } else if (backed
[0] == '\0' || strstart(backed
, "json:", NULL
)) {
316 error_setg(errp
, "Cannot use relative backing file names for '%s'",
319 path_combine(dest
, sz
, backed
, backing
);
323 void bdrv_get_full_backing_filename(BlockDriverState
*bs
, char *dest
, size_t sz
,
326 char *backed
= bs
->exact_filename
[0] ? bs
->exact_filename
: bs
->filename
;
328 bdrv_get_full_backing_filename_from_filename(backed
, bs
->backing_file
,
332 void bdrv_register(BlockDriver
*bdrv
)
334 QLIST_INSERT_HEAD(&bdrv_drivers
, bdrv
, list
);
337 BlockDriverState
*bdrv_new(void)
339 BlockDriverState
*bs
;
342 bs
= g_new0(BlockDriverState
, 1);
343 QLIST_INIT(&bs
->dirty_bitmaps
);
344 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
345 QLIST_INIT(&bs
->op_blockers
[i
]);
347 notifier_with_return_list_init(&bs
->before_write_notifiers
);
348 qemu_co_mutex_init(&bs
->reqs_lock
);
349 qemu_mutex_init(&bs
->dirty_bitmap_mutex
);
351 bs
->aio_context
= qemu_get_aio_context();
353 qemu_co_queue_init(&bs
->flush_queue
);
355 for (i
= 0; i
< bdrv_drain_all_count
; i
++) {
356 bdrv_drained_begin(bs
);
359 QTAILQ_INSERT_TAIL(&all_bdrv_states
, bs
, bs_list
);
364 static BlockDriver
*bdrv_do_find_format(const char *format_name
)
368 QLIST_FOREACH(drv1
, &bdrv_drivers
, list
) {
369 if (!strcmp(drv1
->format_name
, format_name
)) {
377 BlockDriver
*bdrv_find_format(const char *format_name
)
382 drv1
= bdrv_do_find_format(format_name
);
387 /* The driver isn't registered, maybe we need to load a module */
388 for (i
= 0; i
< (int)ARRAY_SIZE(block_driver_modules
); ++i
) {
389 if (!strcmp(block_driver_modules
[i
].format_name
, format_name
)) {
390 block_module_load_one(block_driver_modules
[i
].library_name
);
395 return bdrv_do_find_format(format_name
);
398 int bdrv_is_whitelisted(BlockDriver
*drv
, bool read_only
)
400 static const char *whitelist_rw
[] = {
401 CONFIG_BDRV_RW_WHITELIST
403 static const char *whitelist_ro
[] = {
404 CONFIG_BDRV_RO_WHITELIST
408 if (!whitelist_rw
[0] && !whitelist_ro
[0]) {
409 return 1; /* no whitelist, anything goes */
412 for (p
= whitelist_rw
; *p
; p
++) {
413 if (!strcmp(drv
->format_name
, *p
)) {
418 for (p
= whitelist_ro
; *p
; p
++) {
419 if (!strcmp(drv
->format_name
, *p
)) {
427 bool bdrv_uses_whitelist(void)
429 return use_bdrv_whitelist
;
432 typedef struct CreateCo
{
440 static void coroutine_fn
bdrv_create_co_entry(void *opaque
)
442 Error
*local_err
= NULL
;
445 CreateCo
*cco
= opaque
;
448 ret
= cco
->drv
->bdrv_co_create_opts(cco
->filename
, cco
->opts
, &local_err
);
449 error_propagate(&cco
->err
, local_err
);
453 int bdrv_create(BlockDriver
*drv
, const char* filename
,
454 QemuOpts
*opts
, Error
**errp
)
461 .filename
= g_strdup(filename
),
467 if (!drv
->bdrv_co_create_opts
) {
468 error_setg(errp
, "Driver '%s' does not support image creation", drv
->format_name
);
473 if (qemu_in_coroutine()) {
474 /* Fast-path if already in coroutine context */
475 bdrv_create_co_entry(&cco
);
477 co
= qemu_coroutine_create(bdrv_create_co_entry
, &cco
);
478 qemu_coroutine_enter(co
);
479 while (cco
.ret
== NOT_DONE
) {
480 aio_poll(qemu_get_aio_context(), true);
487 error_propagate(errp
, cco
.err
);
489 error_setg_errno(errp
, -ret
, "Could not create image");
494 g_free(cco
.filename
);
498 int bdrv_create_file(const char *filename
, QemuOpts
*opts
, Error
**errp
)
501 Error
*local_err
= NULL
;
504 drv
= bdrv_find_protocol(filename
, true, errp
);
509 ret
= bdrv_create(drv
, filename
, opts
, &local_err
);
510 error_propagate(errp
, local_err
);
515 * Try to get @bs's logical and physical block size.
516 * On success, store them in @bsz struct and return 0.
517 * On failure return -errno.
518 * @bs must not be empty.
520 int bdrv_probe_blocksizes(BlockDriverState
*bs
, BlockSizes
*bsz
)
522 BlockDriver
*drv
= bs
->drv
;
524 if (drv
&& drv
->bdrv_probe_blocksizes
) {
525 return drv
->bdrv_probe_blocksizes(bs
, bsz
);
526 } else if (drv
&& drv
->is_filter
&& bs
->file
) {
527 return bdrv_probe_blocksizes(bs
->file
->bs
, bsz
);
534 * Try to get @bs's geometry (cyls, heads, sectors).
535 * On success, store them in @geo struct and return 0.
536 * On failure return -errno.
537 * @bs must not be empty.
539 int bdrv_probe_geometry(BlockDriverState
*bs
, HDGeometry
*geo
)
541 BlockDriver
*drv
= bs
->drv
;
543 if (drv
&& drv
->bdrv_probe_geometry
) {
544 return drv
->bdrv_probe_geometry(bs
, geo
);
545 } else if (drv
&& drv
->is_filter
&& bs
->file
) {
546 return bdrv_probe_geometry(bs
->file
->bs
, geo
);
553 * Create a uniquely-named empty temporary file.
554 * Return 0 upon success, otherwise a negative errno value.
556 int get_tmp_filename(char *filename
, int size
)
559 char temp_dir
[MAX_PATH
];
560 /* GetTempFileName requires that its output buffer (4th param)
561 have length MAX_PATH or greater. */
562 assert(size
>= MAX_PATH
);
563 return (GetTempPath(MAX_PATH
, temp_dir
)
564 && GetTempFileName(temp_dir
, "qem", 0, filename
)
565 ? 0 : -GetLastError());
569 tmpdir
= getenv("TMPDIR");
573 if (snprintf(filename
, size
, "%s/vl.XXXXXX", tmpdir
) >= size
) {
576 fd
= mkstemp(filename
);
580 if (close(fd
) != 0) {
589 * Detect host devices. By convention, /dev/cdrom[N] is always
590 * recognized as a host CDROM.
592 static BlockDriver
*find_hdev_driver(const char *filename
)
594 int score_max
= 0, score
;
595 BlockDriver
*drv
= NULL
, *d
;
597 QLIST_FOREACH(d
, &bdrv_drivers
, list
) {
598 if (d
->bdrv_probe_device
) {
599 score
= d
->bdrv_probe_device(filename
);
600 if (score
> score_max
) {
610 static BlockDriver
*bdrv_do_find_protocol(const char *protocol
)
614 QLIST_FOREACH(drv1
, &bdrv_drivers
, list
) {
615 if (drv1
->protocol_name
&& !strcmp(drv1
->protocol_name
, protocol
)) {
623 BlockDriver
*bdrv_find_protocol(const char *filename
,
624 bool allow_protocol_prefix
,
633 /* TODO Drivers without bdrv_file_open must be specified explicitly */
636 * XXX(hch): we really should not let host device detection
637 * override an explicit protocol specification, but moving this
638 * later breaks access to device names with colons in them.
639 * Thanks to the brain-dead persistent naming schemes on udev-
640 * based Linux systems those actually are quite common.
642 drv1
= find_hdev_driver(filename
);
647 if (!path_has_protocol(filename
) || !allow_protocol_prefix
) {
651 p
= strchr(filename
, ':');
654 if (len
> sizeof(protocol
) - 1)
655 len
= sizeof(protocol
) - 1;
656 memcpy(protocol
, filename
, len
);
657 protocol
[len
] = '\0';
659 drv1
= bdrv_do_find_protocol(protocol
);
664 for (i
= 0; i
< (int)ARRAY_SIZE(block_driver_modules
); ++i
) {
665 if (block_driver_modules
[i
].protocol_name
&&
666 !strcmp(block_driver_modules
[i
].protocol_name
, protocol
)) {
667 block_module_load_one(block_driver_modules
[i
].library_name
);
672 drv1
= bdrv_do_find_protocol(protocol
);
674 error_setg(errp
, "Unknown protocol '%s'", protocol
);
680 * Guess image format by probing its contents.
681 * This is not a good idea when your image is raw (CVE-2008-2004), but
682 * we do it anyway for backward compatibility.
684 * @buf contains the image's first @buf_size bytes.
685 * @buf_size is the buffer size in bytes (generally BLOCK_PROBE_BUF_SIZE,
686 * but can be smaller if the image file is smaller)
687 * @filename is its filename.
689 * For all block drivers, call the bdrv_probe() method to get its
691 * Return the first block driver with the highest probing score.
693 BlockDriver
*bdrv_probe_all(const uint8_t *buf
, int buf_size
,
694 const char *filename
)
696 int score_max
= 0, score
;
697 BlockDriver
*drv
= NULL
, *d
;
699 QLIST_FOREACH(d
, &bdrv_drivers
, list
) {
701 score
= d
->bdrv_probe(buf
, buf_size
, filename
);
702 if (score
> score_max
) {
712 static int find_image_format(BlockBackend
*file
, const char *filename
,
713 BlockDriver
**pdrv
, Error
**errp
)
716 uint8_t buf
[BLOCK_PROBE_BUF_SIZE
];
719 /* Return the raw BlockDriver * to scsi-generic devices or empty drives */
720 if (blk_is_sg(file
) || !blk_is_inserted(file
) || blk_getlength(file
) == 0) {
725 ret
= blk_pread(file
, 0, buf
, sizeof(buf
));
727 error_setg_errno(errp
, -ret
, "Could not read image for determining its "
733 drv
= bdrv_probe_all(buf
, ret
, filename
);
735 error_setg(errp
, "Could not determine image format: No compatible "
744 * Set the current 'total_sectors' value
745 * Return 0 on success, -errno on error.
747 int refresh_total_sectors(BlockDriverState
*bs
, int64_t hint
)
749 BlockDriver
*drv
= bs
->drv
;
755 /* Do not attempt drv->bdrv_getlength() on scsi-generic devices */
759 /* query actual device if possible, otherwise just trust the hint */
760 if (drv
->bdrv_getlength
) {
761 int64_t length
= drv
->bdrv_getlength(bs
);
765 hint
= DIV_ROUND_UP(length
, BDRV_SECTOR_SIZE
);
768 bs
->total_sectors
= hint
;
773 * Combines a QDict of new block driver @options with any missing options taken
774 * from @old_options, so that leaving out an option defaults to its old value.
776 static void bdrv_join_options(BlockDriverState
*bs
, QDict
*options
,
779 if (bs
->drv
&& bs
->drv
->bdrv_join_options
) {
780 bs
->drv
->bdrv_join_options(options
, old_options
);
782 qdict_join(options
, old_options
, false);
786 static BlockdevDetectZeroesOptions
bdrv_parse_detect_zeroes(QemuOpts
*opts
,
790 Error
*local_err
= NULL
;
791 char *value
= qemu_opt_get_del(opts
, "detect-zeroes");
792 BlockdevDetectZeroesOptions detect_zeroes
=
793 qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup
, value
,
794 BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF
, &local_err
);
797 error_propagate(errp
, local_err
);
798 return detect_zeroes
;
801 if (detect_zeroes
== BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP
&&
802 !(open_flags
& BDRV_O_UNMAP
))
804 error_setg(errp
, "setting detect-zeroes to unmap is not allowed "
805 "without setting discard operation to unmap");
808 return detect_zeroes
;
812 * Set open flags for a given discard mode
814 * Return 0 on success, -1 if the discard mode was invalid.
816 int bdrv_parse_discard_flags(const char *mode
, int *flags
)
818 *flags
&= ~BDRV_O_UNMAP
;
820 if (!strcmp(mode
, "off") || !strcmp(mode
, "ignore")) {
822 } else if (!strcmp(mode
, "on") || !strcmp(mode
, "unmap")) {
823 *flags
|= BDRV_O_UNMAP
;
832 * Set open flags for a given cache mode
834 * Return 0 on success, -1 if the cache mode was invalid.
836 int bdrv_parse_cache_mode(const char *mode
, int *flags
, bool *writethrough
)
838 *flags
&= ~BDRV_O_CACHE_MASK
;
840 if (!strcmp(mode
, "off") || !strcmp(mode
, "none")) {
841 *writethrough
= false;
842 *flags
|= BDRV_O_NOCACHE
;
843 } else if (!strcmp(mode
, "directsync")) {
844 *writethrough
= true;
845 *flags
|= BDRV_O_NOCACHE
;
846 } else if (!strcmp(mode
, "writeback")) {
847 *writethrough
= false;
848 } else if (!strcmp(mode
, "unsafe")) {
849 *writethrough
= false;
850 *flags
|= BDRV_O_NO_FLUSH
;
851 } else if (!strcmp(mode
, "writethrough")) {
852 *writethrough
= true;
860 static char *bdrv_child_get_parent_desc(BdrvChild
*c
)
862 BlockDriverState
*parent
= c
->opaque
;
863 return g_strdup(bdrv_get_device_or_node_name(parent
));
866 static void bdrv_child_cb_drained_begin(BdrvChild
*child
)
868 BlockDriverState
*bs
= child
->opaque
;
869 bdrv_do_drained_begin_quiesce(bs
, NULL
, false);
872 static bool bdrv_child_cb_drained_poll(BdrvChild
*child
)
874 BlockDriverState
*bs
= child
->opaque
;
875 return bdrv_drain_poll(bs
, false, NULL
, false);
878 static void bdrv_child_cb_drained_end(BdrvChild
*child
)
880 BlockDriverState
*bs
= child
->opaque
;
881 bdrv_drained_end(bs
);
884 static void bdrv_child_cb_attach(BdrvChild
*child
)
886 BlockDriverState
*bs
= child
->opaque
;
887 bdrv_apply_subtree_drain(child
, bs
);
890 static void bdrv_child_cb_detach(BdrvChild
*child
)
892 BlockDriverState
*bs
= child
->opaque
;
893 bdrv_unapply_subtree_drain(child
, bs
);
896 static int bdrv_child_cb_inactivate(BdrvChild
*child
)
898 BlockDriverState
*bs
= child
->opaque
;
899 assert(bs
->open_flags
& BDRV_O_INACTIVE
);
904 * Returns the options and flags that a temporary snapshot should get, based on
905 * the originally requested flags (the originally requested image will have
906 * flags like a backing file)
908 static void bdrv_temp_snapshot_options(int *child_flags
, QDict
*child_options
,
909 int parent_flags
, QDict
*parent_options
)
911 *child_flags
= (parent_flags
& ~BDRV_O_SNAPSHOT
) | BDRV_O_TEMPORARY
;
913 /* For temporary files, unconditional cache=unsafe is fine */
914 qdict_set_default_str(child_options
, BDRV_OPT_CACHE_DIRECT
, "off");
915 qdict_set_default_str(child_options
, BDRV_OPT_CACHE_NO_FLUSH
, "on");
917 /* Copy the read-only option from the parent */
918 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_READ_ONLY
);
920 /* aio=native doesn't work for cache.direct=off, so disable it for the
921 * temporary snapshot */
922 *child_flags
&= ~BDRV_O_NATIVE_AIO
;
926 * Returns the options and flags that bs->file should get if a protocol driver
927 * is expected, based on the given options and flags for the parent BDS
929 static void bdrv_inherited_options(int *child_flags
, QDict
*child_options
,
930 int parent_flags
, QDict
*parent_options
)
932 int flags
= parent_flags
;
934 /* Enable protocol handling, disable format probing for bs->file */
935 flags
|= BDRV_O_PROTOCOL
;
937 /* If the cache mode isn't explicitly set, inherit direct and no-flush from
939 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_CACHE_DIRECT
);
940 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_CACHE_NO_FLUSH
);
941 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_FORCE_SHARE
);
943 /* Inherit the read-only option from the parent if it's not set */
944 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_READ_ONLY
);
945 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_AUTO_READ_ONLY
);
947 /* Our block drivers take care to send flushes and respect unmap policy,
948 * so we can default to enable both on lower layers regardless of the
949 * corresponding parent options. */
950 qdict_set_default_str(child_options
, BDRV_OPT_DISCARD
, "unmap");
952 /* Clear flags that only apply to the top layer */
953 flags
&= ~(BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
| BDRV_O_COPY_ON_READ
|
956 *child_flags
= flags
;
959 const BdrvChildRole child_file
= {
960 .parent_is_bds
= true,
961 .get_parent_desc
= bdrv_child_get_parent_desc
,
962 .inherit_options
= bdrv_inherited_options
,
963 .drained_begin
= bdrv_child_cb_drained_begin
,
964 .drained_poll
= bdrv_child_cb_drained_poll
,
965 .drained_end
= bdrv_child_cb_drained_end
,
966 .attach
= bdrv_child_cb_attach
,
967 .detach
= bdrv_child_cb_detach
,
968 .inactivate
= bdrv_child_cb_inactivate
,
972 * Returns the options and flags that bs->file should get if the use of formats
973 * (and not only protocols) is permitted for it, based on the given options and
974 * flags for the parent BDS
976 static void bdrv_inherited_fmt_options(int *child_flags
, QDict
*child_options
,
977 int parent_flags
, QDict
*parent_options
)
979 child_file
.inherit_options(child_flags
, child_options
,
980 parent_flags
, parent_options
);
982 *child_flags
&= ~(BDRV_O_PROTOCOL
| BDRV_O_NO_IO
);
985 const BdrvChildRole child_format
= {
986 .parent_is_bds
= true,
987 .get_parent_desc
= bdrv_child_get_parent_desc
,
988 .inherit_options
= bdrv_inherited_fmt_options
,
989 .drained_begin
= bdrv_child_cb_drained_begin
,
990 .drained_poll
= bdrv_child_cb_drained_poll
,
991 .drained_end
= bdrv_child_cb_drained_end
,
992 .attach
= bdrv_child_cb_attach
,
993 .detach
= bdrv_child_cb_detach
,
994 .inactivate
= bdrv_child_cb_inactivate
,
997 static void bdrv_backing_attach(BdrvChild
*c
)
999 BlockDriverState
*parent
= c
->opaque
;
1000 BlockDriverState
*backing_hd
= c
->bs
;
1002 assert(!parent
->backing_blocker
);
1003 error_setg(&parent
->backing_blocker
,
1004 "node is used as backing hd of '%s'",
1005 bdrv_get_device_or_node_name(parent
));
1007 parent
->open_flags
&= ~BDRV_O_NO_BACKING
;
1008 pstrcpy(parent
->backing_file
, sizeof(parent
->backing_file
),
1009 backing_hd
->filename
);
1010 pstrcpy(parent
->backing_format
, sizeof(parent
->backing_format
),
1011 backing_hd
->drv
? backing_hd
->drv
->format_name
: "");
1013 bdrv_op_block_all(backing_hd
, parent
->backing_blocker
);
1014 /* Otherwise we won't be able to commit or stream */
1015 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_COMMIT_TARGET
,
1016 parent
->backing_blocker
);
1017 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_STREAM
,
1018 parent
->backing_blocker
);
1020 * We do backup in 3 ways:
1022 * The target bs is new opened, and the source is top BDS
1023 * 2. blockdev backup
1024 * Both the source and the target are top BDSes.
1025 * 3. internal backup(used for block replication)
1026 * Both the source and the target are backing file
1028 * In case 1 and 2, neither the source nor the target is the backing file.
1029 * In case 3, we will block the top BDS, so there is only one block job
1030 * for the top BDS and its backing chain.
1032 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_BACKUP_SOURCE
,
1033 parent
->backing_blocker
);
1034 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_BACKUP_TARGET
,
1035 parent
->backing_blocker
);
1037 bdrv_child_cb_attach(c
);
1040 static void bdrv_backing_detach(BdrvChild
*c
)
1042 BlockDriverState
*parent
= c
->opaque
;
1044 assert(parent
->backing_blocker
);
1045 bdrv_op_unblock_all(c
->bs
, parent
->backing_blocker
);
1046 error_free(parent
->backing_blocker
);
1047 parent
->backing_blocker
= NULL
;
1049 bdrv_child_cb_detach(c
);
1053 * Returns the options and flags that bs->backing should get, based on the
1054 * given options and flags for the parent BDS
1056 static void bdrv_backing_options(int *child_flags
, QDict
*child_options
,
1057 int parent_flags
, QDict
*parent_options
)
1059 int flags
= parent_flags
;
1061 /* The cache mode is inherited unmodified for backing files; except WCE,
1062 * which is only applied on the top level (BlockBackend) */
1063 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_CACHE_DIRECT
);
1064 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_CACHE_NO_FLUSH
);
1065 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_FORCE_SHARE
);
1067 /* backing files always opened read-only */
1068 qdict_set_default_str(child_options
, BDRV_OPT_READ_ONLY
, "on");
1069 qdict_set_default_str(child_options
, BDRV_OPT_AUTO_READ_ONLY
, "off");
1070 flags
&= ~BDRV_O_COPY_ON_READ
;
1072 /* snapshot=on is handled on the top layer */
1073 flags
&= ~(BDRV_O_SNAPSHOT
| BDRV_O_TEMPORARY
);
1075 *child_flags
= flags
;
1078 static int bdrv_backing_update_filename(BdrvChild
*c
, BlockDriverState
*base
,
1079 const char *filename
, Error
**errp
)
1081 BlockDriverState
*parent
= c
->opaque
;
1082 int orig_flags
= bdrv_get_flags(parent
);
1085 if (!(orig_flags
& BDRV_O_RDWR
)) {
1086 ret
= bdrv_reopen(parent
, orig_flags
| BDRV_O_RDWR
, errp
);
1092 ret
= bdrv_change_backing_file(parent
, filename
,
1093 base
->drv
? base
->drv
->format_name
: "");
1095 error_setg_errno(errp
, -ret
, "Could not update backing file link");
1098 if (!(orig_flags
& BDRV_O_RDWR
)) {
1099 bdrv_reopen(parent
, orig_flags
, NULL
);
1105 const BdrvChildRole child_backing
= {
1106 .parent_is_bds
= true,
1107 .get_parent_desc
= bdrv_child_get_parent_desc
,
1108 .attach
= bdrv_backing_attach
,
1109 .detach
= bdrv_backing_detach
,
1110 .inherit_options
= bdrv_backing_options
,
1111 .drained_begin
= bdrv_child_cb_drained_begin
,
1112 .drained_poll
= bdrv_child_cb_drained_poll
,
1113 .drained_end
= bdrv_child_cb_drained_end
,
1114 .inactivate
= bdrv_child_cb_inactivate
,
1115 .update_filename
= bdrv_backing_update_filename
,
1118 static int bdrv_open_flags(BlockDriverState
*bs
, int flags
)
1120 int open_flags
= flags
;
1123 * Clear flags that are internal to the block layer before opening the
1126 open_flags
&= ~(BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
| BDRV_O_PROTOCOL
);
1129 * Snapshots should be writable.
1131 if (flags
& BDRV_O_TEMPORARY
) {
1132 open_flags
|= BDRV_O_RDWR
;
1138 static void update_flags_from_options(int *flags
, QemuOpts
*opts
)
1140 *flags
&= ~BDRV_O_CACHE_MASK
;
1142 assert(qemu_opt_find(opts
, BDRV_OPT_CACHE_NO_FLUSH
));
1143 if (qemu_opt_get_bool_del(opts
, BDRV_OPT_CACHE_NO_FLUSH
, false)) {
1144 *flags
|= BDRV_O_NO_FLUSH
;
1147 assert(qemu_opt_find(opts
, BDRV_OPT_CACHE_DIRECT
));
1148 if (qemu_opt_get_bool_del(opts
, BDRV_OPT_CACHE_DIRECT
, false)) {
1149 *flags
|= BDRV_O_NOCACHE
;
1152 *flags
&= ~BDRV_O_RDWR
;
1154 assert(qemu_opt_find(opts
, BDRV_OPT_READ_ONLY
));
1155 if (!qemu_opt_get_bool_del(opts
, BDRV_OPT_READ_ONLY
, false)) {
1156 *flags
|= BDRV_O_RDWR
;
1159 assert(qemu_opt_find(opts
, BDRV_OPT_AUTO_READ_ONLY
));
1160 if (qemu_opt_get_bool_del(opts
, BDRV_OPT_AUTO_READ_ONLY
, false)) {
1161 *flags
|= BDRV_O_AUTO_RDONLY
;
1165 static void update_options_from_flags(QDict
*options
, int flags
)
1167 if (!qdict_haskey(options
, BDRV_OPT_CACHE_DIRECT
)) {
1168 qdict_put_bool(options
, BDRV_OPT_CACHE_DIRECT
, flags
& BDRV_O_NOCACHE
);
1170 if (!qdict_haskey(options
, BDRV_OPT_CACHE_NO_FLUSH
)) {
1171 qdict_put_bool(options
, BDRV_OPT_CACHE_NO_FLUSH
,
1172 flags
& BDRV_O_NO_FLUSH
);
1174 if (!qdict_haskey(options
, BDRV_OPT_READ_ONLY
)) {
1175 qdict_put_bool(options
, BDRV_OPT_READ_ONLY
, !(flags
& BDRV_O_RDWR
));
1177 if (!qdict_haskey(options
, BDRV_OPT_AUTO_READ_ONLY
)) {
1178 qdict_put_bool(options
, BDRV_OPT_AUTO_READ_ONLY
,
1179 flags
& BDRV_O_AUTO_RDONLY
);
1183 static void bdrv_assign_node_name(BlockDriverState
*bs
,
1184 const char *node_name
,
1187 char *gen_node_name
= NULL
;
1190 node_name
= gen_node_name
= id_generate(ID_BLOCK
);
1191 } else if (!id_wellformed(node_name
)) {
1193 * Check for empty string or invalid characters, but not if it is
1194 * generated (generated names use characters not available to the user)
1196 error_setg(errp
, "Invalid node name");
1200 /* takes care of avoiding namespaces collisions */
1201 if (blk_by_name(node_name
)) {
1202 error_setg(errp
, "node-name=%s is conflicting with a device id",
1207 /* takes care of avoiding duplicates node names */
1208 if (bdrv_find_node(node_name
)) {
1209 error_setg(errp
, "Duplicate node name");
1213 /* Make sure that the node name isn't truncated */
1214 if (strlen(node_name
) >= sizeof(bs
->node_name
)) {
1215 error_setg(errp
, "Node name too long");
1219 /* copy node name into the bs and insert it into the graph list */
1220 pstrcpy(bs
->node_name
, sizeof(bs
->node_name
), node_name
);
1221 QTAILQ_INSERT_TAIL(&graph_bdrv_states
, bs
, node_list
);
1223 g_free(gen_node_name
);
1226 static int bdrv_open_driver(BlockDriverState
*bs
, BlockDriver
*drv
,
1227 const char *node_name
, QDict
*options
,
1228 int open_flags
, Error
**errp
)
1230 Error
*local_err
= NULL
;
1233 bdrv_assign_node_name(bs
, node_name
, &local_err
);
1235 error_propagate(errp
, local_err
);
1240 bs
->read_only
= !(bs
->open_flags
& BDRV_O_RDWR
);
1241 bs
->opaque
= g_malloc0(drv
->instance_size
);
1243 if (drv
->bdrv_file_open
) {
1244 assert(!drv
->bdrv_needs_filename
|| bs
->filename
[0]);
1245 ret
= drv
->bdrv_file_open(bs
, options
, open_flags
, &local_err
);
1246 } else if (drv
->bdrv_open
) {
1247 ret
= drv
->bdrv_open(bs
, options
, open_flags
, &local_err
);
1254 error_propagate(errp
, local_err
);
1255 } else if (bs
->filename
[0]) {
1256 error_setg_errno(errp
, -ret
, "Could not open '%s'", bs
->filename
);
1258 error_setg_errno(errp
, -ret
, "Could not open image");
1263 ret
= refresh_total_sectors(bs
, bs
->total_sectors
);
1265 error_setg_errno(errp
, -ret
, "Could not refresh total sector count");
1269 bdrv_refresh_limits(bs
, &local_err
);
1271 error_propagate(errp
, local_err
);
1275 assert(bdrv_opt_mem_align(bs
) != 0);
1276 assert(bdrv_min_mem_align(bs
) != 0);
1277 assert(is_power_of_2(bs
->bl
.request_alignment
));
1279 for (i
= 0; i
< bs
->quiesce_counter
; i
++) {
1280 if (drv
->bdrv_co_drain_begin
) {
1281 drv
->bdrv_co_drain_begin(bs
);
1288 if (bs
->file
!= NULL
) {
1289 bdrv_unref_child(bs
, bs
->file
);
1297 BlockDriverState
*bdrv_new_open_driver(BlockDriver
*drv
, const char *node_name
,
1298 int flags
, Error
**errp
)
1300 BlockDriverState
*bs
;
1304 bs
->open_flags
= flags
;
1305 bs
->explicit_options
= qdict_new();
1306 bs
->options
= qdict_new();
1309 update_options_from_flags(bs
->options
, flags
);
1311 ret
= bdrv_open_driver(bs
, drv
, node_name
, bs
->options
, flags
, errp
);
1313 qobject_unref(bs
->explicit_options
);
1314 bs
->explicit_options
= NULL
;
1315 qobject_unref(bs
->options
);
1324 QemuOptsList bdrv_runtime_opts
= {
1325 .name
= "bdrv_common",
1326 .head
= QTAILQ_HEAD_INITIALIZER(bdrv_runtime_opts
.head
),
1329 .name
= "node-name",
1330 .type
= QEMU_OPT_STRING
,
1331 .help
= "Node name of the block device node",
1335 .type
= QEMU_OPT_STRING
,
1336 .help
= "Block driver to use for the node",
1339 .name
= BDRV_OPT_CACHE_DIRECT
,
1340 .type
= QEMU_OPT_BOOL
,
1341 .help
= "Bypass software writeback cache on the host",
1344 .name
= BDRV_OPT_CACHE_NO_FLUSH
,
1345 .type
= QEMU_OPT_BOOL
,
1346 .help
= "Ignore flush requests",
1349 .name
= BDRV_OPT_READ_ONLY
,
1350 .type
= QEMU_OPT_BOOL
,
1351 .help
= "Node is opened in read-only mode",
1354 .name
= BDRV_OPT_AUTO_READ_ONLY
,
1355 .type
= QEMU_OPT_BOOL
,
1356 .help
= "Node can become read-only if opening read-write fails",
1359 .name
= "detect-zeroes",
1360 .type
= QEMU_OPT_STRING
,
1361 .help
= "try to optimize zero writes (off, on, unmap)",
1364 .name
= BDRV_OPT_DISCARD
,
1365 .type
= QEMU_OPT_STRING
,
1366 .help
= "discard operation (ignore/off, unmap/on)",
1369 .name
= BDRV_OPT_FORCE_SHARE
,
1370 .type
= QEMU_OPT_BOOL
,
1371 .help
= "always accept other writers (default: off)",
1373 { /* end of list */ }
1378 * Common part for opening disk images and files
1380 * Removes all processed options from *options.
1382 static int bdrv_open_common(BlockDriverState
*bs
, BlockBackend
*file
,
1383 QDict
*options
, Error
**errp
)
1385 int ret
, open_flags
;
1386 const char *filename
;
1387 const char *driver_name
= NULL
;
1388 const char *node_name
= NULL
;
1389 const char *discard
;
1392 Error
*local_err
= NULL
;
1394 assert(bs
->file
== NULL
);
1395 assert(options
!= NULL
&& bs
->options
!= options
);
1397 opts
= qemu_opts_create(&bdrv_runtime_opts
, NULL
, 0, &error_abort
);
1398 qemu_opts_absorb_qdict(opts
, options
, &local_err
);
1400 error_propagate(errp
, local_err
);
1405 update_flags_from_options(&bs
->open_flags
, opts
);
1407 driver_name
= qemu_opt_get(opts
, "driver");
1408 drv
= bdrv_find_format(driver_name
);
1409 assert(drv
!= NULL
);
1411 bs
->force_share
= qemu_opt_get_bool(opts
, BDRV_OPT_FORCE_SHARE
, false);
1413 if (bs
->force_share
&& (bs
->open_flags
& BDRV_O_RDWR
)) {
1415 BDRV_OPT_FORCE_SHARE
1416 "=on can only be used with read-only images");
1422 filename
= blk_bs(file
)->filename
;
1425 * Caution: while qdict_get_try_str() is fine, getting
1426 * non-string types would require more care. When @options
1427 * come from -blockdev or blockdev_add, its members are typed
1428 * according to the QAPI schema, but when they come from
1429 * -drive, they're all QString.
1431 filename
= qdict_get_try_str(options
, "filename");
1434 if (drv
->bdrv_needs_filename
&& (!filename
|| !filename
[0])) {
1435 error_setg(errp
, "The '%s' block driver requires a file name",
1441 trace_bdrv_open_common(bs
, filename
?: "", bs
->open_flags
,
1444 bs
->read_only
= !(bs
->open_flags
& BDRV_O_RDWR
);
1446 if (use_bdrv_whitelist
&& !bdrv_is_whitelisted(drv
, bs
->read_only
)) {
1448 !bs
->read_only
&& bdrv_is_whitelisted(drv
, true)
1449 ? "Driver '%s' can only be used for read-only devices"
1450 : "Driver '%s' is not whitelisted",
1456 /* bdrv_new() and bdrv_close() make it so */
1457 assert(atomic_read(&bs
->copy_on_read
) == 0);
1459 if (bs
->open_flags
& BDRV_O_COPY_ON_READ
) {
1460 if (!bs
->read_only
) {
1461 bdrv_enable_copy_on_read(bs
);
1463 error_setg(errp
, "Can't use copy-on-read on read-only device");
1469 discard
= qemu_opt_get(opts
, BDRV_OPT_DISCARD
);
1470 if (discard
!= NULL
) {
1471 if (bdrv_parse_discard_flags(discard
, &bs
->open_flags
) != 0) {
1472 error_setg(errp
, "Invalid discard option");
1479 bdrv_parse_detect_zeroes(opts
, bs
->open_flags
, &local_err
);
1481 error_propagate(errp
, local_err
);
1486 if (filename
!= NULL
) {
1487 pstrcpy(bs
->filename
, sizeof(bs
->filename
), filename
);
1489 bs
->filename
[0] = '\0';
1491 pstrcpy(bs
->exact_filename
, sizeof(bs
->exact_filename
), bs
->filename
);
1493 /* Open the image, either directly or using a protocol */
1494 open_flags
= bdrv_open_flags(bs
, bs
->open_flags
);
1495 node_name
= qemu_opt_get(opts
, "node-name");
1497 assert(!drv
->bdrv_file_open
|| file
== NULL
);
1498 ret
= bdrv_open_driver(bs
, drv
, node_name
, options
, open_flags
, errp
);
1503 qemu_opts_del(opts
);
1507 qemu_opts_del(opts
);
1511 static GCC_FMT_ATTR(1, 0)
1512 QDict
*parse_json_filename(const char *filename
, Error
**errp
)
1514 QObject
*options_obj
;
1518 ret
= strstart(filename
, "json:", &filename
);
1521 options_obj
= qobject_from_json(filename
, errp
);
1523 error_prepend(errp
, "Could not parse the JSON options: ");
1527 options
= qobject_to(QDict
, options_obj
);
1529 qobject_unref(options_obj
);
1530 error_setg(errp
, "Invalid JSON object given");
1534 qdict_flatten(options
);
1539 static void parse_json_protocol(QDict
*options
, const char **pfilename
,
1542 QDict
*json_options
;
1543 Error
*local_err
= NULL
;
1545 /* Parse json: pseudo-protocol */
1546 if (!*pfilename
|| !g_str_has_prefix(*pfilename
, "json:")) {
1550 json_options
= parse_json_filename(*pfilename
, &local_err
);
1552 error_propagate(errp
, local_err
);
1556 /* Options given in the filename have lower priority than options
1557 * specified directly */
1558 qdict_join(options
, json_options
, false);
1559 qobject_unref(json_options
);
1564 * Fills in default options for opening images and converts the legacy
1565 * filename/flags pair to option QDict entries.
1566 * The BDRV_O_PROTOCOL flag in *flags will be set or cleared accordingly if a
1567 * block driver has been specified explicitly.
1569 static int bdrv_fill_options(QDict
**options
, const char *filename
,
1570 int *flags
, Error
**errp
)
1572 const char *drvname
;
1573 bool protocol
= *flags
& BDRV_O_PROTOCOL
;
1574 bool parse_filename
= false;
1575 BlockDriver
*drv
= NULL
;
1576 Error
*local_err
= NULL
;
1579 * Caution: while qdict_get_try_str() is fine, getting non-string
1580 * types would require more care. When @options come from
1581 * -blockdev or blockdev_add, its members are typed according to
1582 * the QAPI schema, but when they come from -drive, they're all
1585 drvname
= qdict_get_try_str(*options
, "driver");
1587 drv
= bdrv_find_format(drvname
);
1589 error_setg(errp
, "Unknown driver '%s'", drvname
);
1592 /* If the user has explicitly specified the driver, this choice should
1593 * override the BDRV_O_PROTOCOL flag */
1594 protocol
= drv
->bdrv_file_open
;
1598 *flags
|= BDRV_O_PROTOCOL
;
1600 *flags
&= ~BDRV_O_PROTOCOL
;
1603 /* Translate cache options from flags into options */
1604 update_options_from_flags(*options
, *flags
);
1606 /* Fetch the file name from the options QDict if necessary */
1607 if (protocol
&& filename
) {
1608 if (!qdict_haskey(*options
, "filename")) {
1609 qdict_put_str(*options
, "filename", filename
);
1610 parse_filename
= true;
1612 error_setg(errp
, "Can't specify 'file' and 'filename' options at "
1618 /* Find the right block driver */
1619 /* See cautionary note on accessing @options above */
1620 filename
= qdict_get_try_str(*options
, "filename");
1622 if (!drvname
&& protocol
) {
1624 drv
= bdrv_find_protocol(filename
, parse_filename
, errp
);
1629 drvname
= drv
->format_name
;
1630 qdict_put_str(*options
, "driver", drvname
);
1632 error_setg(errp
, "Must specify either driver or file");
1637 assert(drv
|| !protocol
);
1639 /* Driver-specific filename parsing */
1640 if (drv
&& drv
->bdrv_parse_filename
&& parse_filename
) {
1641 drv
->bdrv_parse_filename(filename
, *options
, &local_err
);
1643 error_propagate(errp
, local_err
);
1647 if (!drv
->bdrv_needs_filename
) {
1648 qdict_del(*options
, "filename");
1655 static int bdrv_child_check_perm(BdrvChild
*c
, BlockReopenQueue
*q
,
1656 uint64_t perm
, uint64_t shared
,
1657 GSList
*ignore_children
, Error
**errp
);
1658 static void bdrv_child_abort_perm_update(BdrvChild
*c
);
1659 static void bdrv_child_set_perm(BdrvChild
*c
, uint64_t perm
, uint64_t shared
);
1661 typedef struct BlockReopenQueueEntry
{
1663 BDRVReopenState state
;
1664 QSIMPLEQ_ENTRY(BlockReopenQueueEntry
) entry
;
1665 } BlockReopenQueueEntry
;
1668 * Return the flags that @bs will have after the reopens in @q have
1669 * successfully completed. If @q is NULL (or @bs is not contained in @q),
1670 * return the current flags.
1672 static int bdrv_reopen_get_flags(BlockReopenQueue
*q
, BlockDriverState
*bs
)
1674 BlockReopenQueueEntry
*entry
;
1677 QSIMPLEQ_FOREACH(entry
, q
, entry
) {
1678 if (entry
->state
.bs
== bs
) {
1679 return entry
->state
.flags
;
1684 return bs
->open_flags
;
1687 /* Returns whether the image file can be written to after the reopen queue @q
1688 * has been successfully applied, or right now if @q is NULL. */
1689 static bool bdrv_is_writable_after_reopen(BlockDriverState
*bs
,
1690 BlockReopenQueue
*q
)
1692 int flags
= bdrv_reopen_get_flags(q
, bs
);
1694 return (flags
& (BDRV_O_RDWR
| BDRV_O_INACTIVE
)) == BDRV_O_RDWR
;
1698 * Return whether the BDS can be written to. This is not necessarily
1699 * the same as !bdrv_is_read_only(bs), as inactivated images may not
1700 * be written to but do not count as read-only images.
1702 bool bdrv_is_writable(BlockDriverState
*bs
)
1704 return bdrv_is_writable_after_reopen(bs
, NULL
);
1707 static void bdrv_child_perm(BlockDriverState
*bs
, BlockDriverState
*child_bs
,
1708 BdrvChild
*c
, const BdrvChildRole
*role
,
1709 BlockReopenQueue
*reopen_queue
,
1710 uint64_t parent_perm
, uint64_t parent_shared
,
1711 uint64_t *nperm
, uint64_t *nshared
)
1713 if (bs
->drv
&& bs
->drv
->bdrv_child_perm
) {
1714 bs
->drv
->bdrv_child_perm(bs
, c
, role
, reopen_queue
,
1715 parent_perm
, parent_shared
,
1718 /* TODO Take force_share from reopen_queue */
1719 if (child_bs
&& child_bs
->force_share
) {
1720 *nshared
= BLK_PERM_ALL
;
1725 * Check whether permissions on this node can be changed in a way that
1726 * @cumulative_perms and @cumulative_shared_perms are the new cumulative
1727 * permissions of all its parents. This involves checking whether all necessary
1728 * permission changes to child nodes can be performed.
1730 * A call to this function must always be followed by a call to bdrv_set_perm()
1731 * or bdrv_abort_perm_update().
1733 static int bdrv_check_perm(BlockDriverState
*bs
, BlockReopenQueue
*q
,
1734 uint64_t cumulative_perms
,
1735 uint64_t cumulative_shared_perms
,
1736 GSList
*ignore_children
, Error
**errp
)
1738 BlockDriver
*drv
= bs
->drv
;
1742 /* Write permissions never work with read-only images */
1743 if ((cumulative_perms
& (BLK_PERM_WRITE
| BLK_PERM_WRITE_UNCHANGED
)) &&
1744 !bdrv_is_writable_after_reopen(bs
, q
))
1746 error_setg(errp
, "Block node is read-only");
1750 /* Check this node */
1755 if (drv
->bdrv_check_perm
) {
1756 return drv
->bdrv_check_perm(bs
, cumulative_perms
,
1757 cumulative_shared_perms
, errp
);
1760 /* Drivers that never have children can omit .bdrv_child_perm() */
1761 if (!drv
->bdrv_child_perm
) {
1762 assert(QLIST_EMPTY(&bs
->children
));
1766 /* Check all children */
1767 QLIST_FOREACH(c
, &bs
->children
, next
) {
1768 uint64_t cur_perm
, cur_shared
;
1769 bdrv_child_perm(bs
, c
->bs
, c
, c
->role
, q
,
1770 cumulative_perms
, cumulative_shared_perms
,
1771 &cur_perm
, &cur_shared
);
1772 ret
= bdrv_child_check_perm(c
, q
, cur_perm
, cur_shared
,
1773 ignore_children
, errp
);
1783 * Notifies drivers that after a previous bdrv_check_perm() call, the
1784 * permission update is not performed and any preparations made for it (e.g.
1785 * taken file locks) need to be undone.
1787 * This function recursively notifies all child nodes.
1789 static void bdrv_abort_perm_update(BlockDriverState
*bs
)
1791 BlockDriver
*drv
= bs
->drv
;
1798 if (drv
->bdrv_abort_perm_update
) {
1799 drv
->bdrv_abort_perm_update(bs
);
1802 QLIST_FOREACH(c
, &bs
->children
, next
) {
1803 bdrv_child_abort_perm_update(c
);
1807 static void bdrv_set_perm(BlockDriverState
*bs
, uint64_t cumulative_perms
,
1808 uint64_t cumulative_shared_perms
)
1810 BlockDriver
*drv
= bs
->drv
;
1817 /* Update this node */
1818 if (drv
->bdrv_set_perm
) {
1819 drv
->bdrv_set_perm(bs
, cumulative_perms
, cumulative_shared_perms
);
1822 /* Drivers that never have children can omit .bdrv_child_perm() */
1823 if (!drv
->bdrv_child_perm
) {
1824 assert(QLIST_EMPTY(&bs
->children
));
1828 /* Update all children */
1829 QLIST_FOREACH(c
, &bs
->children
, next
) {
1830 uint64_t cur_perm
, cur_shared
;
1831 bdrv_child_perm(bs
, c
->bs
, c
, c
->role
, NULL
,
1832 cumulative_perms
, cumulative_shared_perms
,
1833 &cur_perm
, &cur_shared
);
1834 bdrv_child_set_perm(c
, cur_perm
, cur_shared
);
1838 static void bdrv_get_cumulative_perm(BlockDriverState
*bs
, uint64_t *perm
,
1839 uint64_t *shared_perm
)
1842 uint64_t cumulative_perms
= 0;
1843 uint64_t cumulative_shared_perms
= BLK_PERM_ALL
;
1845 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
1846 cumulative_perms
|= c
->perm
;
1847 cumulative_shared_perms
&= c
->shared_perm
;
1850 *perm
= cumulative_perms
;
1851 *shared_perm
= cumulative_shared_perms
;
1854 static char *bdrv_child_user_desc(BdrvChild
*c
)
1856 if (c
->role
->get_parent_desc
) {
1857 return c
->role
->get_parent_desc(c
);
1860 return g_strdup("another user");
1863 char *bdrv_perm_names(uint64_t perm
)
1869 { BLK_PERM_CONSISTENT_READ
, "consistent read" },
1870 { BLK_PERM_WRITE
, "write" },
1871 { BLK_PERM_WRITE_UNCHANGED
, "write unchanged" },
1872 { BLK_PERM_RESIZE
, "resize" },
1873 { BLK_PERM_GRAPH_MOD
, "change children" },
1877 char *result
= g_strdup("");
1878 struct perm_name
*p
;
1880 for (p
= permissions
; p
->name
; p
++) {
1881 if (perm
& p
->perm
) {
1883 result
= g_strdup_printf("%s%s%s", old
, *old
? ", " : "", p
->name
);
1892 * Checks whether a new reference to @bs can be added if the new user requires
1893 * @new_used_perm/@new_shared_perm as its permissions. If @ignore_children is
1894 * set, the BdrvChild objects in this list are ignored in the calculations;
1895 * this allows checking permission updates for an existing reference.
1897 * Needs to be followed by a call to either bdrv_set_perm() or
1898 * bdrv_abort_perm_update(). */
1899 static int bdrv_check_update_perm(BlockDriverState
*bs
, BlockReopenQueue
*q
,
1900 uint64_t new_used_perm
,
1901 uint64_t new_shared_perm
,
1902 GSList
*ignore_children
, Error
**errp
)
1905 uint64_t cumulative_perms
= new_used_perm
;
1906 uint64_t cumulative_shared_perms
= new_shared_perm
;
1908 /* There is no reason why anyone couldn't tolerate write_unchanged */
1909 assert(new_shared_perm
& BLK_PERM_WRITE_UNCHANGED
);
1911 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
1912 if (g_slist_find(ignore_children
, c
)) {
1916 if ((new_used_perm
& c
->shared_perm
) != new_used_perm
) {
1917 char *user
= bdrv_child_user_desc(c
);
1918 char *perm_names
= bdrv_perm_names(new_used_perm
& ~c
->shared_perm
);
1919 error_setg(errp
, "Conflicts with use by %s as '%s', which does not "
1921 user
, c
->name
, perm_names
, bdrv_get_node_name(c
->bs
));
1927 if ((c
->perm
& new_shared_perm
) != c
->perm
) {
1928 char *user
= bdrv_child_user_desc(c
);
1929 char *perm_names
= bdrv_perm_names(c
->perm
& ~new_shared_perm
);
1930 error_setg(errp
, "Conflicts with use by %s as '%s', which uses "
1932 user
, c
->name
, perm_names
, bdrv_get_node_name(c
->bs
));
1938 cumulative_perms
|= c
->perm
;
1939 cumulative_shared_perms
&= c
->shared_perm
;
1942 return bdrv_check_perm(bs
, q
, cumulative_perms
, cumulative_shared_perms
,
1943 ignore_children
, errp
);
1946 /* Needs to be followed by a call to either bdrv_child_set_perm() or
1947 * bdrv_child_abort_perm_update(). */
1948 static int bdrv_child_check_perm(BdrvChild
*c
, BlockReopenQueue
*q
,
1949 uint64_t perm
, uint64_t shared
,
1950 GSList
*ignore_children
, Error
**errp
)
1954 ignore_children
= g_slist_prepend(g_slist_copy(ignore_children
), c
);
1955 ret
= bdrv_check_update_perm(c
->bs
, q
, perm
, shared
, ignore_children
, errp
);
1956 g_slist_free(ignore_children
);
1961 static void bdrv_child_set_perm(BdrvChild
*c
, uint64_t perm
, uint64_t shared
)
1963 uint64_t cumulative_perms
, cumulative_shared_perms
;
1966 c
->shared_perm
= shared
;
1968 bdrv_get_cumulative_perm(c
->bs
, &cumulative_perms
,
1969 &cumulative_shared_perms
);
1970 bdrv_set_perm(c
->bs
, cumulative_perms
, cumulative_shared_perms
);
1973 static void bdrv_child_abort_perm_update(BdrvChild
*c
)
1975 bdrv_abort_perm_update(c
->bs
);
1978 int bdrv_child_try_set_perm(BdrvChild
*c
, uint64_t perm
, uint64_t shared
,
1983 ret
= bdrv_child_check_perm(c
, NULL
, perm
, shared
, NULL
, errp
);
1985 bdrv_child_abort_perm_update(c
);
1989 bdrv_child_set_perm(c
, perm
, shared
);
1994 void bdrv_filter_default_perms(BlockDriverState
*bs
, BdrvChild
*c
,
1995 const BdrvChildRole
*role
,
1996 BlockReopenQueue
*reopen_queue
,
1997 uint64_t perm
, uint64_t shared
,
1998 uint64_t *nperm
, uint64_t *nshared
)
2001 *nperm
= perm
& DEFAULT_PERM_PASSTHROUGH
;
2002 *nshared
= (shared
& DEFAULT_PERM_PASSTHROUGH
) | DEFAULT_PERM_UNCHANGED
;
2006 *nperm
= (perm
& DEFAULT_PERM_PASSTHROUGH
) |
2007 (c
->perm
& DEFAULT_PERM_UNCHANGED
);
2008 *nshared
= (shared
& DEFAULT_PERM_PASSTHROUGH
) |
2009 (c
->shared_perm
& DEFAULT_PERM_UNCHANGED
);
2012 void bdrv_format_default_perms(BlockDriverState
*bs
, BdrvChild
*c
,
2013 const BdrvChildRole
*role
,
2014 BlockReopenQueue
*reopen_queue
,
2015 uint64_t perm
, uint64_t shared
,
2016 uint64_t *nperm
, uint64_t *nshared
)
2018 bool backing
= (role
== &child_backing
);
2019 assert(role
== &child_backing
|| role
== &child_file
);
2022 int flags
= bdrv_reopen_get_flags(reopen_queue
, bs
);
2024 /* Apart from the modifications below, the same permissions are
2025 * forwarded and left alone as for filters */
2026 bdrv_filter_default_perms(bs
, c
, role
, reopen_queue
, perm
, shared
,
2029 /* Format drivers may touch metadata even if the guest doesn't write */
2030 if (bdrv_is_writable_after_reopen(bs
, reopen_queue
)) {
2031 perm
|= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2034 /* bs->file always needs to be consistent because of the metadata. We
2035 * can never allow other users to resize or write to it. */
2036 if (!(flags
& BDRV_O_NO_IO
)) {
2037 perm
|= BLK_PERM_CONSISTENT_READ
;
2039 shared
&= ~(BLK_PERM_WRITE
| BLK_PERM_RESIZE
);
2041 /* We want consistent read from backing files if the parent needs it.
2042 * No other operations are performed on backing files. */
2043 perm
&= BLK_PERM_CONSISTENT_READ
;
2045 /* If the parent can deal with changing data, we're okay with a
2046 * writable and resizable backing file. */
2047 /* TODO Require !(perm & BLK_PERM_CONSISTENT_READ), too? */
2048 if (shared
& BLK_PERM_WRITE
) {
2049 shared
= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2054 shared
|= BLK_PERM_CONSISTENT_READ
| BLK_PERM_GRAPH_MOD
|
2055 BLK_PERM_WRITE_UNCHANGED
;
2058 if (bs
->open_flags
& BDRV_O_INACTIVE
) {
2059 shared
|= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2066 static void bdrv_replace_child_noperm(BdrvChild
*child
,
2067 BlockDriverState
*new_bs
)
2069 BlockDriverState
*old_bs
= child
->bs
;
2072 if (old_bs
&& new_bs
) {
2073 assert(bdrv_get_aio_context(old_bs
) == bdrv_get_aio_context(new_bs
));
2076 /* Detach first so that the recursive drain sections coming from @child
2077 * are already gone and we only end the drain sections that came from
2079 if (child
->role
->detach
) {
2080 child
->role
->detach(child
);
2082 if (old_bs
->quiesce_counter
&& child
->role
->drained_end
) {
2083 int num
= old_bs
->quiesce_counter
;
2084 if (child
->role
->parent_is_bds
) {
2085 num
-= bdrv_drain_all_count
;
2088 for (i
= 0; i
< num
; i
++) {
2089 child
->role
->drained_end(child
);
2092 QLIST_REMOVE(child
, next_parent
);
2098 QLIST_INSERT_HEAD(&new_bs
->parents
, child
, next_parent
);
2099 if (new_bs
->quiesce_counter
&& child
->role
->drained_begin
) {
2100 int num
= new_bs
->quiesce_counter
;
2101 if (child
->role
->parent_is_bds
) {
2102 num
-= bdrv_drain_all_count
;
2105 for (i
= 0; i
< num
; i
++) {
2106 bdrv_parent_drained_begin_single(child
, true);
2110 /* Attach only after starting new drained sections, so that recursive
2111 * drain sections coming from @child don't get an extra .drained_begin
2113 if (child
->role
->attach
) {
2114 child
->role
->attach(child
);
2120 * Updates @child to change its reference to point to @new_bs, including
2121 * checking and applying the necessary permisson updates both to the old node
2124 * NULL is passed as @new_bs for removing the reference before freeing @child.
2126 * If @new_bs is not NULL, bdrv_check_perm() must be called beforehand, as this
2127 * function uses bdrv_set_perm() to update the permissions according to the new
2128 * reference that @new_bs gets.
2130 static void bdrv_replace_child(BdrvChild
*child
, BlockDriverState
*new_bs
)
2132 BlockDriverState
*old_bs
= child
->bs
;
2133 uint64_t perm
, shared_perm
;
2135 bdrv_replace_child_noperm(child
, new_bs
);
2138 /* Update permissions for old node. This is guaranteed to succeed
2139 * because we're just taking a parent away, so we're loosening
2141 bdrv_get_cumulative_perm(old_bs
, &perm
, &shared_perm
);
2142 bdrv_check_perm(old_bs
, NULL
, perm
, shared_perm
, NULL
, &error_abort
);
2143 bdrv_set_perm(old_bs
, perm
, shared_perm
);
2147 bdrv_get_cumulative_perm(new_bs
, &perm
, &shared_perm
);
2148 bdrv_set_perm(new_bs
, perm
, shared_perm
);
2152 BdrvChild
*bdrv_root_attach_child(BlockDriverState
*child_bs
,
2153 const char *child_name
,
2154 const BdrvChildRole
*child_role
,
2155 uint64_t perm
, uint64_t shared_perm
,
2156 void *opaque
, Error
**errp
)
2161 ret
= bdrv_check_update_perm(child_bs
, NULL
, perm
, shared_perm
, NULL
, errp
);
2163 bdrv_abort_perm_update(child_bs
);
2167 child
= g_new(BdrvChild
, 1);
2168 *child
= (BdrvChild
) {
2170 .name
= g_strdup(child_name
),
2173 .shared_perm
= shared_perm
,
2177 /* This performs the matching bdrv_set_perm() for the above check. */
2178 bdrv_replace_child(child
, child_bs
);
2183 BdrvChild
*bdrv_attach_child(BlockDriverState
*parent_bs
,
2184 BlockDriverState
*child_bs
,
2185 const char *child_name
,
2186 const BdrvChildRole
*child_role
,
2190 uint64_t perm
, shared_perm
;
2192 bdrv_get_cumulative_perm(parent_bs
, &perm
, &shared_perm
);
2194 assert(parent_bs
->drv
);
2195 assert(bdrv_get_aio_context(parent_bs
) == bdrv_get_aio_context(child_bs
));
2196 bdrv_child_perm(parent_bs
, child_bs
, NULL
, child_role
, NULL
,
2197 perm
, shared_perm
, &perm
, &shared_perm
);
2199 child
= bdrv_root_attach_child(child_bs
, child_name
, child_role
,
2200 perm
, shared_perm
, parent_bs
, errp
);
2201 if (child
== NULL
) {
2205 QLIST_INSERT_HEAD(&parent_bs
->children
, child
, next
);
2209 static void bdrv_detach_child(BdrvChild
*child
)
2211 if (child
->next
.le_prev
) {
2212 QLIST_REMOVE(child
, next
);
2213 child
->next
.le_prev
= NULL
;
2216 bdrv_replace_child(child
, NULL
);
2218 g_free(child
->name
);
2222 void bdrv_root_unref_child(BdrvChild
*child
)
2224 BlockDriverState
*child_bs
;
2226 child_bs
= child
->bs
;
2227 bdrv_detach_child(child
);
2228 bdrv_unref(child_bs
);
2231 void bdrv_unref_child(BlockDriverState
*parent
, BdrvChild
*child
)
2233 if (child
== NULL
) {
2237 if (child
->bs
->inherits_from
== parent
) {
2240 /* Remove inherits_from only when the last reference between parent and
2241 * child->bs goes away. */
2242 QLIST_FOREACH(c
, &parent
->children
, next
) {
2243 if (c
!= child
&& c
->bs
== child
->bs
) {
2248 child
->bs
->inherits_from
= NULL
;
2252 bdrv_root_unref_child(child
);
2256 static void bdrv_parent_cb_change_media(BlockDriverState
*bs
, bool load
)
2259 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
2260 if (c
->role
->change_media
) {
2261 c
->role
->change_media(c
, load
);
2267 * Sets the backing file link of a BDS. A new reference is created; callers
2268 * which don't need their own reference any more must call bdrv_unref().
2270 void bdrv_set_backing_hd(BlockDriverState
*bs
, BlockDriverState
*backing_hd
,
2274 bdrv_ref(backing_hd
);
2278 bdrv_unref_child(bs
, bs
->backing
);
2286 bs
->backing
= bdrv_attach_child(bs
, backing_hd
, "backing", &child_backing
,
2289 bdrv_unref(backing_hd
);
2292 bdrv_refresh_filename(bs
);
2295 bdrv_refresh_limits(bs
, NULL
);
2299 * Opens the backing file for a BlockDriverState if not yet open
2301 * bdref_key specifies the key for the image's BlockdevRef in the options QDict.
2302 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
2303 * itself, all options starting with "${bdref_key}." are considered part of the
2306 * TODO Can this be unified with bdrv_open_image()?
2308 int bdrv_open_backing_file(BlockDriverState
*bs
, QDict
*parent_options
,
2309 const char *bdref_key
, Error
**errp
)
2311 char *backing_filename
= g_malloc0(PATH_MAX
);
2312 char *bdref_key_dot
;
2313 const char *reference
= NULL
;
2315 BlockDriverState
*backing_hd
;
2317 QDict
*tmp_parent_options
= NULL
;
2318 Error
*local_err
= NULL
;
2320 if (bs
->backing
!= NULL
) {
2324 /* NULL means an empty set of options */
2325 if (parent_options
== NULL
) {
2326 tmp_parent_options
= qdict_new();
2327 parent_options
= tmp_parent_options
;
2330 bs
->open_flags
&= ~BDRV_O_NO_BACKING
;
2332 bdref_key_dot
= g_strdup_printf("%s.", bdref_key
);
2333 qdict_extract_subqdict(parent_options
, &options
, bdref_key_dot
);
2334 g_free(bdref_key_dot
);
2337 * Caution: while qdict_get_try_str() is fine, getting non-string
2338 * types would require more care. When @parent_options come from
2339 * -blockdev or blockdev_add, its members are typed according to
2340 * the QAPI schema, but when they come from -drive, they're all
2343 reference
= qdict_get_try_str(parent_options
, bdref_key
);
2344 if (reference
|| qdict_haskey(options
, "file.filename")) {
2345 backing_filename
[0] = '\0';
2346 } else if (bs
->backing_file
[0] == '\0' && qdict_size(options
) == 0) {
2347 qobject_unref(options
);
2350 bdrv_get_full_backing_filename(bs
, backing_filename
, PATH_MAX
,
2354 error_propagate(errp
, local_err
);
2355 qobject_unref(options
);
2360 if (!bs
->drv
|| !bs
->drv
->supports_backing
) {
2362 error_setg(errp
, "Driver doesn't support backing files");
2363 qobject_unref(options
);
2368 bs
->backing_format
[0] != '\0' && !qdict_haskey(options
, "driver")) {
2369 qdict_put_str(options
, "driver", bs
->backing_format
);
2372 backing_hd
= bdrv_open_inherit(*backing_filename
? backing_filename
: NULL
,
2373 reference
, options
, 0, bs
, &child_backing
,
2376 bs
->open_flags
|= BDRV_O_NO_BACKING
;
2377 error_prepend(errp
, "Could not open backing file: ");
2381 bdrv_set_aio_context(backing_hd
, bdrv_get_aio_context(bs
));
2383 /* Hook up the backing file link; drop our reference, bs owns the
2384 * backing_hd reference now */
2385 bdrv_set_backing_hd(bs
, backing_hd
, &local_err
);
2386 bdrv_unref(backing_hd
);
2388 error_propagate(errp
, local_err
);
2393 qdict_del(parent_options
, bdref_key
);
2396 g_free(backing_filename
);
2397 qobject_unref(tmp_parent_options
);
2401 static BlockDriverState
*
2402 bdrv_open_child_bs(const char *filename
, QDict
*options
, const char *bdref_key
,
2403 BlockDriverState
*parent
, const BdrvChildRole
*child_role
,
2404 bool allow_none
, Error
**errp
)
2406 BlockDriverState
*bs
= NULL
;
2407 QDict
*image_options
;
2408 char *bdref_key_dot
;
2409 const char *reference
;
2411 assert(child_role
!= NULL
);
2413 bdref_key_dot
= g_strdup_printf("%s.", bdref_key
);
2414 qdict_extract_subqdict(options
, &image_options
, bdref_key_dot
);
2415 g_free(bdref_key_dot
);
2418 * Caution: while qdict_get_try_str() is fine, getting non-string
2419 * types would require more care. When @options come from
2420 * -blockdev or blockdev_add, its members are typed according to
2421 * the QAPI schema, but when they come from -drive, they're all
2424 reference
= qdict_get_try_str(options
, bdref_key
);
2425 if (!filename
&& !reference
&& !qdict_size(image_options
)) {
2427 error_setg(errp
, "A block device must be specified for \"%s\"",
2430 qobject_unref(image_options
);
2434 bs
= bdrv_open_inherit(filename
, reference
, image_options
, 0,
2435 parent
, child_role
, errp
);
2441 qdict_del(options
, bdref_key
);
2446 * Opens a disk image whose options are given as BlockdevRef in another block
2449 * If allow_none is true, no image will be opened if filename is false and no
2450 * BlockdevRef is given. NULL will be returned, but errp remains unset.
2452 * bdrev_key specifies the key for the image's BlockdevRef in the options QDict.
2453 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
2454 * itself, all options starting with "${bdref_key}." are considered part of the
2457 * The BlockdevRef will be removed from the options QDict.
2459 BdrvChild
*bdrv_open_child(const char *filename
,
2460 QDict
*options
, const char *bdref_key
,
2461 BlockDriverState
*parent
,
2462 const BdrvChildRole
*child_role
,
2463 bool allow_none
, Error
**errp
)
2466 BlockDriverState
*bs
;
2468 bs
= bdrv_open_child_bs(filename
, options
, bdref_key
, parent
, child_role
,
2474 c
= bdrv_attach_child(parent
, bs
, bdref_key
, child_role
, errp
);
2483 /* TODO Future callers may need to specify parent/child_role in order for
2484 * option inheritance to work. Existing callers use it for the root node. */
2485 BlockDriverState
*bdrv_open_blockdev_ref(BlockdevRef
*ref
, Error
**errp
)
2487 BlockDriverState
*bs
= NULL
;
2488 Error
*local_err
= NULL
;
2489 QObject
*obj
= NULL
;
2490 QDict
*qdict
= NULL
;
2491 const char *reference
= NULL
;
2494 if (ref
->type
== QTYPE_QSTRING
) {
2495 reference
= ref
->u
.reference
;
2497 BlockdevOptions
*options
= &ref
->u
.definition
;
2498 assert(ref
->type
== QTYPE_QDICT
);
2500 v
= qobject_output_visitor_new(&obj
);
2501 visit_type_BlockdevOptions(v
, NULL
, &options
, &local_err
);
2503 error_propagate(errp
, local_err
);
2506 visit_complete(v
, &obj
);
2508 qdict
= qobject_to(QDict
, obj
);
2509 qdict_flatten(qdict
);
2511 /* bdrv_open_inherit() defaults to the values in bdrv_flags (for
2512 * compatibility with other callers) rather than what we want as the
2513 * real defaults. Apply the defaults here instead. */
2514 qdict_set_default_str(qdict
, BDRV_OPT_CACHE_DIRECT
, "off");
2515 qdict_set_default_str(qdict
, BDRV_OPT_CACHE_NO_FLUSH
, "off");
2516 qdict_set_default_str(qdict
, BDRV_OPT_READ_ONLY
, "off");
2517 qdict_set_default_str(qdict
, BDRV_OPT_AUTO_READ_ONLY
, "off");
2521 bs
= bdrv_open_inherit(NULL
, reference
, qdict
, 0, NULL
, NULL
, errp
);
2530 static BlockDriverState
*bdrv_append_temp_snapshot(BlockDriverState
*bs
,
2532 QDict
*snapshot_options
,
2535 /* TODO: extra byte is a hack to ensure MAX_PATH space on Windows. */
2536 char *tmp_filename
= g_malloc0(PATH_MAX
+ 1);
2538 QemuOpts
*opts
= NULL
;
2539 BlockDriverState
*bs_snapshot
= NULL
;
2540 Error
*local_err
= NULL
;
2543 /* if snapshot, we create a temporary backing file and open it
2544 instead of opening 'filename' directly */
2546 /* Get the required size from the image */
2547 total_size
= bdrv_getlength(bs
);
2548 if (total_size
< 0) {
2549 error_setg_errno(errp
, -total_size
, "Could not get image size");
2553 /* Create the temporary image */
2554 ret
= get_tmp_filename(tmp_filename
, PATH_MAX
+ 1);
2556 error_setg_errno(errp
, -ret
, "Could not get temporary filename");
2560 opts
= qemu_opts_create(bdrv_qcow2
.create_opts
, NULL
, 0,
2562 qemu_opt_set_number(opts
, BLOCK_OPT_SIZE
, total_size
, &error_abort
);
2563 ret
= bdrv_create(&bdrv_qcow2
, tmp_filename
, opts
, errp
);
2564 qemu_opts_del(opts
);
2566 error_prepend(errp
, "Could not create temporary overlay '%s': ",
2571 /* Prepare options QDict for the temporary file */
2572 qdict_put_str(snapshot_options
, "file.driver", "file");
2573 qdict_put_str(snapshot_options
, "file.filename", tmp_filename
);
2574 qdict_put_str(snapshot_options
, "driver", "qcow2");
2576 bs_snapshot
= bdrv_open(NULL
, NULL
, snapshot_options
, flags
, errp
);
2577 snapshot_options
= NULL
;
2582 /* bdrv_append() consumes a strong reference to bs_snapshot
2583 * (i.e. it will call bdrv_unref() on it) even on error, so in
2584 * order to be able to return one, we have to increase
2585 * bs_snapshot's refcount here */
2586 bdrv_ref(bs_snapshot
);
2587 bdrv_append(bs_snapshot
, bs
, &local_err
);
2589 error_propagate(errp
, local_err
);
2595 qobject_unref(snapshot_options
);
2596 g_free(tmp_filename
);
2601 * Opens a disk image (raw, qcow2, vmdk, ...)
2603 * options is a QDict of options to pass to the block drivers, or NULL for an
2604 * empty set of options. The reference to the QDict belongs to the block layer
2605 * after the call (even on failure), so if the caller intends to reuse the
2606 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
2608 * If *pbs is NULL, a new BDS will be created with a pointer to it stored there.
2609 * If it is not NULL, the referenced BDS will be reused.
2611 * The reference parameter may be used to specify an existing block device which
2612 * should be opened. If specified, neither options nor a filename may be given,
2613 * nor can an existing BDS be reused (that is, *pbs has to be NULL).
2615 static BlockDriverState
*bdrv_open_inherit(const char *filename
,
2616 const char *reference
,
2617 QDict
*options
, int flags
,
2618 BlockDriverState
*parent
,
2619 const BdrvChildRole
*child_role
,
2623 BlockBackend
*file
= NULL
;
2624 BlockDriverState
*bs
;
2625 BlockDriver
*drv
= NULL
;
2627 const char *drvname
;
2628 const char *backing
;
2629 Error
*local_err
= NULL
;
2630 QDict
*snapshot_options
= NULL
;
2631 int snapshot_flags
= 0;
2633 assert(!child_role
|| !flags
);
2634 assert(!child_role
== !parent
);
2637 bool options_non_empty
= options
? qdict_size(options
) : false;
2638 qobject_unref(options
);
2640 if (filename
|| options_non_empty
) {
2641 error_setg(errp
, "Cannot reference an existing block device with "
2642 "additional options or a new filename");
2646 bs
= bdrv_lookup_bs(reference
, reference
, errp
);
2657 /* NULL means an empty set of options */
2658 if (options
== NULL
) {
2659 options
= qdict_new();
2662 /* json: syntax counts as explicit options, as if in the QDict */
2663 parse_json_protocol(options
, &filename
, &local_err
);
2668 bs
->explicit_options
= qdict_clone_shallow(options
);
2671 bs
->inherits_from
= parent
;
2672 child_role
->inherit_options(&flags
, options
,
2673 parent
->open_flags
, parent
->options
);
2676 ret
= bdrv_fill_options(&options
, filename
, &flags
, &local_err
);
2682 * Set the BDRV_O_RDWR and BDRV_O_ALLOW_RDWR flags.
2683 * Caution: getting a boolean member of @options requires care.
2684 * When @options come from -blockdev or blockdev_add, members are
2685 * typed according to the QAPI schema, but when they come from
2686 * -drive, they're all QString.
2688 if (g_strcmp0(qdict_get_try_str(options
, BDRV_OPT_READ_ONLY
), "on") &&
2689 !qdict_get_try_bool(options
, BDRV_OPT_READ_ONLY
, false)) {
2690 flags
|= (BDRV_O_RDWR
| BDRV_O_ALLOW_RDWR
);
2692 flags
&= ~BDRV_O_RDWR
;
2695 if (flags
& BDRV_O_SNAPSHOT
) {
2696 snapshot_options
= qdict_new();
2697 bdrv_temp_snapshot_options(&snapshot_flags
, snapshot_options
,
2699 /* Let bdrv_backing_options() override "read-only" */
2700 qdict_del(options
, BDRV_OPT_READ_ONLY
);
2701 bdrv_backing_options(&flags
, options
, flags
, options
);
2704 bs
->open_flags
= flags
;
2705 bs
->options
= options
;
2706 options
= qdict_clone_shallow(options
);
2708 /* Find the right image format driver */
2709 /* See cautionary note on accessing @options above */
2710 drvname
= qdict_get_try_str(options
, "driver");
2712 drv
= bdrv_find_format(drvname
);
2714 error_setg(errp
, "Unknown driver: '%s'", drvname
);
2719 assert(drvname
|| !(flags
& BDRV_O_PROTOCOL
));
2721 /* See cautionary note on accessing @options above */
2722 backing
= qdict_get_try_str(options
, "backing");
2723 if (qobject_to(QNull
, qdict_get(options
, "backing")) != NULL
||
2724 (backing
&& *backing
== '\0'))
2727 warn_report("Use of \"backing\": \"\" is deprecated; "
2728 "use \"backing\": null instead");
2730 flags
|= BDRV_O_NO_BACKING
;
2731 qdict_del(options
, "backing");
2734 /* Open image file without format layer. This BlockBackend is only used for
2735 * probing, the block drivers will do their own bdrv_open_child() for the
2736 * same BDS, which is why we put the node name back into options. */
2737 if ((flags
& BDRV_O_PROTOCOL
) == 0) {
2738 BlockDriverState
*file_bs
;
2740 file_bs
= bdrv_open_child_bs(filename
, options
, "file", bs
,
2741 &child_file
, true, &local_err
);
2745 if (file_bs
!= NULL
) {
2746 /* Not requesting BLK_PERM_CONSISTENT_READ because we're only
2747 * looking at the header to guess the image format. This works even
2748 * in cases where a guest would not see a consistent state. */
2749 file
= blk_new(0, BLK_PERM_ALL
);
2750 blk_insert_bs(file
, file_bs
, &local_err
);
2751 bdrv_unref(file_bs
);
2756 qdict_put_str(options
, "file", bdrv_get_node_name(file_bs
));
2760 /* Image format probing */
2763 ret
= find_image_format(file
, filename
, &drv
, &local_err
);
2768 * This option update would logically belong in bdrv_fill_options(),
2769 * but we first need to open bs->file for the probing to work, while
2770 * opening bs->file already requires the (mostly) final set of options
2771 * so that cache mode etc. can be inherited.
2773 * Adding the driver later is somewhat ugly, but it's not an option
2774 * that would ever be inherited, so it's correct. We just need to make
2775 * sure to update both bs->options (which has the full effective
2776 * options for bs) and options (which has file.* already removed).
2778 qdict_put_str(bs
->options
, "driver", drv
->format_name
);
2779 qdict_put_str(options
, "driver", drv
->format_name
);
2781 error_setg(errp
, "Must specify either driver or file");
2785 /* BDRV_O_PROTOCOL must be set iff a protocol BDS is about to be created */
2786 assert(!!(flags
& BDRV_O_PROTOCOL
) == !!drv
->bdrv_file_open
);
2787 /* file must be NULL if a protocol BDS is about to be created
2788 * (the inverse results in an error message from bdrv_open_common()) */
2789 assert(!(flags
& BDRV_O_PROTOCOL
) || !file
);
2791 /* Open the image */
2792 ret
= bdrv_open_common(bs
, file
, options
, &local_err
);
2802 /* If there is a backing file, use it */
2803 if ((flags
& BDRV_O_NO_BACKING
) == 0) {
2804 ret
= bdrv_open_backing_file(bs
, options
, "backing", &local_err
);
2806 goto close_and_fail
;
2810 /* Remove all children options and references
2811 * from bs->options and bs->explicit_options */
2812 QLIST_FOREACH(child
, &bs
->children
, next
) {
2813 char *child_key_dot
;
2814 child_key_dot
= g_strdup_printf("%s.", child
->name
);
2815 qdict_extract_subqdict(bs
->explicit_options
, NULL
, child_key_dot
);
2816 qdict_extract_subqdict(bs
->options
, NULL
, child_key_dot
);
2817 qdict_del(bs
->explicit_options
, child
->name
);
2818 qdict_del(bs
->options
, child
->name
);
2819 g_free(child_key_dot
);
2822 bdrv_refresh_filename(bs
);
2824 /* Check if any unknown options were used */
2825 if (qdict_size(options
) != 0) {
2826 const QDictEntry
*entry
= qdict_first(options
);
2827 if (flags
& BDRV_O_PROTOCOL
) {
2828 error_setg(errp
, "Block protocol '%s' doesn't support the option "
2829 "'%s'", drv
->format_name
, entry
->key
);
2832 "Block format '%s' does not support the option '%s'",
2833 drv
->format_name
, entry
->key
);
2836 goto close_and_fail
;
2839 bdrv_parent_cb_change_media(bs
, true);
2841 qobject_unref(options
);
2844 /* For snapshot=on, create a temporary qcow2 overlay. bs points to the
2845 * temporary snapshot afterwards. */
2846 if (snapshot_flags
) {
2847 BlockDriverState
*snapshot_bs
;
2848 snapshot_bs
= bdrv_append_temp_snapshot(bs
, snapshot_flags
,
2849 snapshot_options
, &local_err
);
2850 snapshot_options
= NULL
;
2852 goto close_and_fail
;
2854 /* We are not going to return bs but the overlay on top of it
2855 * (snapshot_bs); thus, we have to drop the strong reference to bs
2856 * (which we obtained by calling bdrv_new()). bs will not be deleted,
2857 * though, because the overlay still has a reference to it. */
2866 qobject_unref(snapshot_options
);
2867 qobject_unref(bs
->explicit_options
);
2868 qobject_unref(bs
->options
);
2869 qobject_unref(options
);
2871 bs
->explicit_options
= NULL
;
2873 error_propagate(errp
, local_err
);
2878 qobject_unref(snapshot_options
);
2879 qobject_unref(options
);
2880 error_propagate(errp
, local_err
);
2884 BlockDriverState
*bdrv_open(const char *filename
, const char *reference
,
2885 QDict
*options
, int flags
, Error
**errp
)
2887 return bdrv_open_inherit(filename
, reference
, options
, flags
, NULL
,
2892 * Adds a BlockDriverState to a simple queue for an atomic, transactional
2893 * reopen of multiple devices.
2895 * bs_queue can either be an existing BlockReopenQueue that has had QSIMPLE_INIT
2896 * already performed, or alternatively may be NULL a new BlockReopenQueue will
2897 * be created and initialized. This newly created BlockReopenQueue should be
2898 * passed back in for subsequent calls that are intended to be of the same
2901 * bs is the BlockDriverState to add to the reopen queue.
2903 * options contains the changed options for the associated bs
2904 * (the BlockReopenQueue takes ownership)
2906 * flags contains the open flags for the associated bs
2908 * returns a pointer to bs_queue, which is either the newly allocated
2909 * bs_queue, or the existing bs_queue being used.
2911 * bs must be drained between bdrv_reopen_queue() and bdrv_reopen_multiple().
2913 static BlockReopenQueue
*bdrv_reopen_queue_child(BlockReopenQueue
*bs_queue
,
2914 BlockDriverState
*bs
,
2917 const BdrvChildRole
*role
,
2918 QDict
*parent_options
,
2923 BlockReopenQueueEntry
*bs_entry
;
2925 QDict
*old_options
, *explicit_options
;
2927 /* Make sure that the caller remembered to use a drained section. This is
2928 * important to avoid graph changes between the recursive queuing here and
2929 * bdrv_reopen_multiple(). */
2930 assert(bs
->quiesce_counter
> 0);
2932 if (bs_queue
== NULL
) {
2933 bs_queue
= g_new0(BlockReopenQueue
, 1);
2934 QSIMPLEQ_INIT(bs_queue
);
2938 options
= qdict_new();
2941 /* Check if this BlockDriverState is already in the queue */
2942 QSIMPLEQ_FOREACH(bs_entry
, bs_queue
, entry
) {
2943 if (bs
== bs_entry
->state
.bs
) {
2949 * Precedence of options:
2950 * 1. Explicitly passed in options (highest)
2951 * 2. Set in flags (only for top level)
2952 * 3. Retained from explicitly set options of bs
2953 * 4. Inherited from parent node
2954 * 5. Retained from effective options of bs
2957 if (!parent_options
) {
2959 * Any setting represented by flags is always updated. If the
2960 * corresponding QDict option is set, it takes precedence. Otherwise
2961 * the flag is translated into a QDict option. The old setting of bs is
2964 update_options_from_flags(options
, flags
);
2967 /* Old explicitly set values (don't overwrite by inherited value) */
2969 old_options
= qdict_clone_shallow(bs_entry
->state
.explicit_options
);
2971 old_options
= qdict_clone_shallow(bs
->explicit_options
);
2973 bdrv_join_options(bs
, options
, old_options
);
2974 qobject_unref(old_options
);
2976 explicit_options
= qdict_clone_shallow(options
);
2978 /* Inherit from parent node */
2979 if (parent_options
) {
2981 QDict
*options_copy
;
2983 role
->inherit_options(&flags
, options
, parent_flags
, parent_options
);
2984 options_copy
= qdict_clone_shallow(options
);
2985 opts
= qemu_opts_create(&bdrv_runtime_opts
, NULL
, 0, &error_abort
);
2986 qemu_opts_absorb_qdict(opts
, options_copy
, NULL
);
2987 update_flags_from_options(&flags
, opts
);
2988 qemu_opts_del(opts
);
2989 qobject_unref(options_copy
);
2992 /* Old values are used for options that aren't set yet */
2993 old_options
= qdict_clone_shallow(bs
->options
);
2994 bdrv_join_options(bs
, options
, old_options
);
2995 qobject_unref(old_options
);
2997 /* bdrv_open_inherit() sets and clears some additional flags internally */
2998 flags
&= ~BDRV_O_PROTOCOL
;
2999 if (flags
& BDRV_O_RDWR
) {
3000 flags
|= BDRV_O_ALLOW_RDWR
;
3004 bs_entry
= g_new0(BlockReopenQueueEntry
, 1);
3005 QSIMPLEQ_INSERT_TAIL(bs_queue
, bs_entry
, entry
);
3007 qobject_unref(bs_entry
->state
.options
);
3008 qobject_unref(bs_entry
->state
.explicit_options
);
3011 bs_entry
->state
.bs
= bs
;
3012 bs_entry
->state
.options
= options
;
3013 bs_entry
->state
.explicit_options
= explicit_options
;
3014 bs_entry
->state
.flags
= flags
;
3016 /* This needs to be overwritten in bdrv_reopen_prepare() */
3017 bs_entry
->state
.perm
= UINT64_MAX
;
3018 bs_entry
->state
.shared_perm
= 0;
3020 QLIST_FOREACH(child
, &bs
->children
, next
) {
3021 QDict
*new_child_options
;
3022 char *child_key_dot
;
3024 /* reopen can only change the options of block devices that were
3025 * implicitly created and inherited options. For other (referenced)
3026 * block devices, a syntax like "backing.foo" results in an error. */
3027 if (child
->bs
->inherits_from
!= bs
) {
3031 child_key_dot
= g_strdup_printf("%s.", child
->name
);
3032 qdict_extract_subqdict(explicit_options
, NULL
, child_key_dot
);
3033 qdict_extract_subqdict(options
, &new_child_options
, child_key_dot
);
3034 g_free(child_key_dot
);
3036 bdrv_reopen_queue_child(bs_queue
, child
->bs
, new_child_options
, 0,
3037 child
->role
, options
, flags
);
3043 BlockReopenQueue
*bdrv_reopen_queue(BlockReopenQueue
*bs_queue
,
3044 BlockDriverState
*bs
,
3045 QDict
*options
, int flags
)
3047 return bdrv_reopen_queue_child(bs_queue
, bs
, options
, flags
,
3052 * Reopen multiple BlockDriverStates atomically & transactionally.
3054 * The queue passed in (bs_queue) must have been built up previous
3055 * via bdrv_reopen_queue().
3057 * Reopens all BDS specified in the queue, with the appropriate
3058 * flags. All devices are prepared for reopen, and failure of any
3059 * device will cause all device changes to be abandoned, and intermediate
3062 * If all devices prepare successfully, then the changes are committed
3065 * All affected nodes must be drained between bdrv_reopen_queue() and
3066 * bdrv_reopen_multiple().
3068 int bdrv_reopen_multiple(AioContext
*ctx
, BlockReopenQueue
*bs_queue
, Error
**errp
)
3071 BlockReopenQueueEntry
*bs_entry
, *next
;
3072 Error
*local_err
= NULL
;
3074 assert(bs_queue
!= NULL
);
3076 QSIMPLEQ_FOREACH(bs_entry
, bs_queue
, entry
) {
3077 assert(bs_entry
->state
.bs
->quiesce_counter
> 0);
3078 if (bdrv_reopen_prepare(&bs_entry
->state
, bs_queue
, &local_err
)) {
3079 error_propagate(errp
, local_err
);
3082 bs_entry
->prepared
= true;
3085 /* If we reach this point, we have success and just need to apply the
3088 QSIMPLEQ_FOREACH(bs_entry
, bs_queue
, entry
) {
3089 bdrv_reopen_commit(&bs_entry
->state
);
3095 QSIMPLEQ_FOREACH_SAFE(bs_entry
, bs_queue
, entry
, next
) {
3097 if (bs_entry
->prepared
) {
3098 bdrv_reopen_abort(&bs_entry
->state
);
3100 qobject_unref(bs_entry
->state
.explicit_options
);
3101 qobject_unref(bs_entry
->state
.options
);
3111 /* Reopen a single BlockDriverState with the specified flags. */
3112 int bdrv_reopen(BlockDriverState
*bs
, int bdrv_flags
, Error
**errp
)
3115 Error
*local_err
= NULL
;
3116 BlockReopenQueue
*queue
;
3118 bdrv_subtree_drained_begin(bs
);
3120 queue
= bdrv_reopen_queue(NULL
, bs
, NULL
, bdrv_flags
);
3121 ret
= bdrv_reopen_multiple(bdrv_get_aio_context(bs
), queue
, &local_err
);
3122 if (local_err
!= NULL
) {
3123 error_propagate(errp
, local_err
);
3126 bdrv_subtree_drained_end(bs
);
3131 static BlockReopenQueueEntry
*find_parent_in_reopen_queue(BlockReopenQueue
*q
,
3134 BlockReopenQueueEntry
*entry
;
3136 QSIMPLEQ_FOREACH(entry
, q
, entry
) {
3137 BlockDriverState
*bs
= entry
->state
.bs
;
3140 QLIST_FOREACH(child
, &bs
->children
, next
) {
3150 static void bdrv_reopen_perm(BlockReopenQueue
*q
, BlockDriverState
*bs
,
3151 uint64_t *perm
, uint64_t *shared
)
3154 BlockReopenQueueEntry
*parent
;
3155 uint64_t cumulative_perms
= 0;
3156 uint64_t cumulative_shared_perms
= BLK_PERM_ALL
;
3158 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
3159 parent
= find_parent_in_reopen_queue(q
, c
);
3161 cumulative_perms
|= c
->perm
;
3162 cumulative_shared_perms
&= c
->shared_perm
;
3164 uint64_t nperm
, nshared
;
3166 bdrv_child_perm(parent
->state
.bs
, bs
, c
, c
->role
, q
,
3167 parent
->state
.perm
, parent
->state
.shared_perm
,
3170 cumulative_perms
|= nperm
;
3171 cumulative_shared_perms
&= nshared
;
3174 *perm
= cumulative_perms
;
3175 *shared
= cumulative_shared_perms
;
3179 * Prepares a BlockDriverState for reopen. All changes are staged in the
3180 * 'opaque' field of the BDRVReopenState, which is used and allocated by
3181 * the block driver layer .bdrv_reopen_prepare()
3183 * bs is the BlockDriverState to reopen
3184 * flags are the new open flags
3185 * queue is the reopen queue
3187 * Returns 0 on success, non-zero on error. On error errp will be set
3190 * On failure, bdrv_reopen_abort() will be called to clean up any data.
3191 * It is the responsibility of the caller to then call the abort() or
3192 * commit() for any other BDS that have been left in a prepare() state
3195 int bdrv_reopen_prepare(BDRVReopenState
*reopen_state
, BlockReopenQueue
*queue
,
3199 Error
*local_err
= NULL
;
3202 QDict
*orig_reopen_opts
;
3203 char *discard
= NULL
;
3206 assert(reopen_state
!= NULL
);
3207 assert(reopen_state
->bs
->drv
!= NULL
);
3208 drv
= reopen_state
->bs
->drv
;
3210 /* This function and each driver's bdrv_reopen_prepare() remove
3211 * entries from reopen_state->options as they are processed, so
3212 * we need to make a copy of the original QDict. */
3213 orig_reopen_opts
= qdict_clone_shallow(reopen_state
->options
);
3215 /* Process generic block layer options */
3216 opts
= qemu_opts_create(&bdrv_runtime_opts
, NULL
, 0, &error_abort
);
3217 qemu_opts_absorb_qdict(opts
, reopen_state
->options
, &local_err
);
3219 error_propagate(errp
, local_err
);
3224 update_flags_from_options(&reopen_state
->flags
, opts
);
3226 discard
= qemu_opt_get_del(opts
, BDRV_OPT_DISCARD
);
3227 if (discard
!= NULL
) {
3228 if (bdrv_parse_discard_flags(discard
, &reopen_state
->flags
) != 0) {
3229 error_setg(errp
, "Invalid discard option");
3235 reopen_state
->detect_zeroes
=
3236 bdrv_parse_detect_zeroes(opts
, reopen_state
->flags
, &local_err
);
3238 error_propagate(errp
, local_err
);
3243 /* All other options (including node-name and driver) must be unchanged.
3244 * Put them back into the QDict, so that they are checked at the end
3245 * of this function. */
3246 qemu_opts_to_qdict(opts
, reopen_state
->options
);
3248 /* If we are to stay read-only, do not allow permission change
3249 * to r/w. Attempting to set to r/w may fail if either BDRV_O_ALLOW_RDWR is
3250 * not set, or if the BDS still has copy_on_read enabled */
3251 read_only
= !(reopen_state
->flags
& BDRV_O_RDWR
);
3252 ret
= bdrv_can_set_read_only(reopen_state
->bs
, read_only
, true, &local_err
);
3254 error_propagate(errp
, local_err
);
3258 /* Calculate required permissions after reopening */
3259 bdrv_reopen_perm(queue
, reopen_state
->bs
,
3260 &reopen_state
->perm
, &reopen_state
->shared_perm
);
3262 ret
= bdrv_flush(reopen_state
->bs
);
3264 error_setg_errno(errp
, -ret
, "Error flushing drive");
3268 if (drv
->bdrv_reopen_prepare
) {
3269 ret
= drv
->bdrv_reopen_prepare(reopen_state
, queue
, &local_err
);
3271 if (local_err
!= NULL
) {
3272 error_propagate(errp
, local_err
);
3274 error_setg(errp
, "failed while preparing to reopen image '%s'",
3275 reopen_state
->bs
->filename
);
3280 /* It is currently mandatory to have a bdrv_reopen_prepare()
3281 * handler for each supported drv. */
3282 error_setg(errp
, "Block format '%s' used by node '%s' "
3283 "does not support reopening files", drv
->format_name
,
3284 bdrv_get_device_or_node_name(reopen_state
->bs
));
3289 /* Options that are not handled are only okay if they are unchanged
3290 * compared to the old state. It is expected that some options are only
3291 * used for the initial open, but not reopen (e.g. filename) */
3292 if (qdict_size(reopen_state
->options
)) {
3293 const QDictEntry
*entry
= qdict_first(reopen_state
->options
);
3296 QObject
*new = entry
->value
;
3297 QObject
*old
= qdict_get(reopen_state
->bs
->options
, entry
->key
);
3299 /* Allow child references (child_name=node_name) as long as they
3300 * point to the current child (i.e. everything stays the same). */
3301 if (qobject_type(new) == QTYPE_QSTRING
) {
3303 QLIST_FOREACH(child
, &reopen_state
->bs
->children
, next
) {
3304 if (!strcmp(child
->name
, entry
->key
)) {
3310 const char *str
= qobject_get_try_str(new);
3311 if (!strcmp(child
->bs
->node_name
, str
)) {
3312 continue; /* Found child with this name, skip option */
3318 * TODO: When using -drive to specify blockdev options, all values
3319 * will be strings; however, when using -blockdev, blockdev-add or
3320 * filenames using the json:{} pseudo-protocol, they will be
3322 * In contrast, reopening options are (currently) always strings
3323 * (because you can only specify them through qemu-io; all other
3324 * callers do not specify any options).
3325 * Therefore, when using anything other than -drive to create a BDS,
3326 * this cannot detect non-string options as unchanged, because
3327 * qobject_is_equal() always returns false for objects of different
3328 * type. In the future, this should be remedied by correctly typing
3329 * all options. For now, this is not too big of an issue because
3330 * the user can simply omit options which cannot be changed anyway,
3331 * so they will stay unchanged.
3333 if (!qobject_is_equal(new, old
)) {
3334 error_setg(errp
, "Cannot change the option '%s'", entry
->key
);
3338 } while ((entry
= qdict_next(reopen_state
->options
, entry
)));
3341 ret
= bdrv_check_perm(reopen_state
->bs
, queue
, reopen_state
->perm
,
3342 reopen_state
->shared_perm
, NULL
, errp
);
3349 /* Restore the original reopen_state->options QDict */
3350 qobject_unref(reopen_state
->options
);
3351 reopen_state
->options
= qobject_ref(orig_reopen_opts
);
3354 qemu_opts_del(opts
);
3355 qobject_unref(orig_reopen_opts
);
3361 * Takes the staged changes for the reopen from bdrv_reopen_prepare(), and
3362 * makes them final by swapping the staging BlockDriverState contents into
3363 * the active BlockDriverState contents.
3365 void bdrv_reopen_commit(BDRVReopenState
*reopen_state
)
3368 BlockDriverState
*bs
;
3370 bool old_can_write
, new_can_write
;
3372 assert(reopen_state
!= NULL
);
3373 bs
= reopen_state
->bs
;
3375 assert(drv
!= NULL
);
3378 !bdrv_is_read_only(bs
) && !(bdrv_get_flags(bs
) & BDRV_O_INACTIVE
);
3380 /* If there are any driver level actions to take */
3381 if (drv
->bdrv_reopen_commit
) {
3382 drv
->bdrv_reopen_commit(reopen_state
);
3385 /* set BDS specific flags now */
3386 qobject_unref(bs
->explicit_options
);
3387 qobject_unref(bs
->options
);
3389 bs
->explicit_options
= reopen_state
->explicit_options
;
3390 bs
->options
= reopen_state
->options
;
3391 bs
->open_flags
= reopen_state
->flags
;
3392 bs
->read_only
= !(reopen_state
->flags
& BDRV_O_RDWR
);
3393 bs
->detect_zeroes
= reopen_state
->detect_zeroes
;
3395 /* Remove child references from bs->options and bs->explicit_options.
3396 * Child options were already removed in bdrv_reopen_queue_child() */
3397 QLIST_FOREACH(child
, &bs
->children
, next
) {
3398 qdict_del(bs
->explicit_options
, child
->name
);
3399 qdict_del(bs
->options
, child
->name
);
3402 bdrv_refresh_limits(bs
, NULL
);
3404 bdrv_set_perm(reopen_state
->bs
, reopen_state
->perm
,
3405 reopen_state
->shared_perm
);
3408 !bdrv_is_read_only(bs
) && !(bdrv_get_flags(bs
) & BDRV_O_INACTIVE
);
3409 if (!old_can_write
&& new_can_write
&& drv
->bdrv_reopen_bitmaps_rw
) {
3410 Error
*local_err
= NULL
;
3411 if (drv
->bdrv_reopen_bitmaps_rw(bs
, &local_err
) < 0) {
3412 /* This is not fatal, bitmaps just left read-only, so all following
3413 * writes will fail. User can remove read-only bitmaps to unblock
3416 error_reportf_err(local_err
,
3417 "%s: Failed to make dirty bitmaps writable: ",
3418 bdrv_get_node_name(bs
));
3424 * Abort the reopen, and delete and free the staged changes in
3427 void bdrv_reopen_abort(BDRVReopenState
*reopen_state
)
3431 assert(reopen_state
!= NULL
);
3432 drv
= reopen_state
->bs
->drv
;
3433 assert(drv
!= NULL
);
3435 if (drv
->bdrv_reopen_abort
) {
3436 drv
->bdrv_reopen_abort(reopen_state
);
3439 bdrv_abort_perm_update(reopen_state
->bs
);
3443 static void bdrv_close(BlockDriverState
*bs
)
3445 BdrvAioNotifier
*ban
, *ban_next
;
3446 BdrvChild
*child
, *next
;
3449 assert(!bs
->refcnt
);
3451 bdrv_drained_begin(bs
); /* complete I/O */
3453 bdrv_drain(bs
); /* in case flush left pending I/O */
3456 if (bs
->drv
->bdrv_close
) {
3457 bs
->drv
->bdrv_close(bs
);
3462 bdrv_set_backing_hd(bs
, NULL
, &error_abort
);
3464 if (bs
->file
!= NULL
) {
3465 bdrv_unref_child(bs
, bs
->file
);
3469 QLIST_FOREACH_SAFE(child
, &bs
->children
, next
, next
) {
3470 /* TODO Remove bdrv_unref() from drivers' close function and use
3471 * bdrv_unref_child() here */
3472 if (child
->bs
->inherits_from
== bs
) {
3473 child
->bs
->inherits_from
= NULL
;
3475 bdrv_detach_child(child
);
3480 atomic_set(&bs
->copy_on_read
, 0);
3481 bs
->backing_file
[0] = '\0';
3482 bs
->backing_format
[0] = '\0';
3483 bs
->total_sectors
= 0;
3484 bs
->encrypted
= false;
3486 qobject_unref(bs
->options
);
3487 qobject_unref(bs
->explicit_options
);
3489 bs
->explicit_options
= NULL
;
3490 qobject_unref(bs
->full_open_options
);
3491 bs
->full_open_options
= NULL
;
3493 bdrv_release_named_dirty_bitmaps(bs
);
3494 assert(QLIST_EMPTY(&bs
->dirty_bitmaps
));
3496 QLIST_FOREACH_SAFE(ban
, &bs
->aio_notifiers
, list
, ban_next
) {
3499 QLIST_INIT(&bs
->aio_notifiers
);
3500 bdrv_drained_end(bs
);
3503 void bdrv_close_all(void)
3505 assert(job_next(NULL
) == NULL
);
3506 nbd_export_close_all();
3508 /* Drop references from requests still in flight, such as canceled block
3509 * jobs whose AIO context has not been polled yet */
3512 blk_remove_all_bs();
3513 blockdev_close_all_bdrv_states();
3515 assert(QTAILQ_EMPTY(&all_bdrv_states
));
3518 static bool should_update_child(BdrvChild
*c
, BlockDriverState
*to
)
3522 if (c
->role
->stay_at_node
) {
3526 /* If the child @c belongs to the BDS @to, replacing the current
3527 * c->bs by @to would mean to create a loop.
3529 * Such a case occurs when appending a BDS to a backing chain.
3530 * For instance, imagine the following chain:
3532 * guest device -> node A -> further backing chain...
3534 * Now we create a new BDS B which we want to put on top of this
3535 * chain, so we first attach A as its backing node:
3540 * guest device -> node A -> further backing chain...
3542 * Finally we want to replace A by B. When doing that, we want to
3543 * replace all pointers to A by pointers to B -- except for the
3544 * pointer from B because (1) that would create a loop, and (2)
3545 * that pointer should simply stay intact:
3547 * guest device -> node B
3550 * node A -> further backing chain...
3552 * In general, when replacing a node A (c->bs) by a node B (@to),
3553 * if A is a child of B, that means we cannot replace A by B there
3554 * because that would create a loop. Silently detaching A from B
3555 * is also not really an option. So overall just leaving A in
3556 * place there is the most sensible choice. */
3557 QLIST_FOREACH(to_c
, &to
->children
, next
) {
3566 void bdrv_replace_node(BlockDriverState
*from
, BlockDriverState
*to
,
3569 BdrvChild
*c
, *next
;
3570 GSList
*list
= NULL
, *p
;
3571 uint64_t old_perm
, old_shared
;
3572 uint64_t perm
= 0, shared
= BLK_PERM_ALL
;
3575 assert(!atomic_read(&from
->in_flight
));
3576 assert(!atomic_read(&to
->in_flight
));
3578 /* Make sure that @from doesn't go away until we have successfully attached
3579 * all of its parents to @to. */
3582 /* Put all parents into @list and calculate their cumulative permissions */
3583 QLIST_FOREACH_SAFE(c
, &from
->parents
, next_parent
, next
) {
3584 assert(c
->bs
== from
);
3585 if (!should_update_child(c
, to
)) {
3588 list
= g_slist_prepend(list
, c
);
3590 shared
&= c
->shared_perm
;
3593 /* Check whether the required permissions can be granted on @to, ignoring
3594 * all BdrvChild in @list so that they can't block themselves. */
3595 ret
= bdrv_check_update_perm(to
, NULL
, perm
, shared
, list
, errp
);
3597 bdrv_abort_perm_update(to
);
3601 /* Now actually perform the change. We performed the permission check for
3602 * all elements of @list at once, so set the permissions all at once at the
3604 for (p
= list
; p
!= NULL
; p
= p
->next
) {
3608 bdrv_replace_child_noperm(c
, to
);
3612 bdrv_get_cumulative_perm(to
, &old_perm
, &old_shared
);
3613 bdrv_set_perm(to
, old_perm
| perm
, old_shared
| shared
);
3621 * Add new bs contents at the top of an image chain while the chain is
3622 * live, while keeping required fields on the top layer.
3624 * This will modify the BlockDriverState fields, and swap contents
3625 * between bs_new and bs_top. Both bs_new and bs_top are modified.
3627 * bs_new must not be attached to a BlockBackend.
3629 * This function does not create any image files.
3631 * bdrv_append() takes ownership of a bs_new reference and unrefs it because
3632 * that's what the callers commonly need. bs_new will be referenced by the old
3633 * parents of bs_top after bdrv_append() returns. If the caller needs to keep a
3634 * reference of its own, it must call bdrv_ref().
3636 void bdrv_append(BlockDriverState
*bs_new
, BlockDriverState
*bs_top
,
3639 Error
*local_err
= NULL
;
3641 bdrv_set_backing_hd(bs_new
, bs_top
, &local_err
);
3643 error_propagate(errp
, local_err
);
3647 bdrv_replace_node(bs_top
, bs_new
, &local_err
);
3649 error_propagate(errp
, local_err
);
3650 bdrv_set_backing_hd(bs_new
, NULL
, &error_abort
);
3654 /* bs_new is now referenced by its new parents, we don't need the
3655 * additional reference any more. */
3660 static void bdrv_delete(BlockDriverState
*bs
)
3663 assert(bdrv_op_blocker_is_empty(bs
));
3664 assert(!bs
->refcnt
);
3668 /* remove from list, if necessary */
3669 if (bs
->node_name
[0] != '\0') {
3670 QTAILQ_REMOVE(&graph_bdrv_states
, bs
, node_list
);
3672 QTAILQ_REMOVE(&all_bdrv_states
, bs
, bs_list
);
3678 * Run consistency checks on an image
3680 * Returns 0 if the check could be completed (it doesn't mean that the image is
3681 * free of errors) or -errno when an internal error occurred. The results of the
3682 * check are stored in res.
3684 static int coroutine_fn
bdrv_co_check(BlockDriverState
*bs
,
3685 BdrvCheckResult
*res
, BdrvCheckMode fix
)
3687 if (bs
->drv
== NULL
) {
3690 if (bs
->drv
->bdrv_co_check
== NULL
) {
3694 memset(res
, 0, sizeof(*res
));
3695 return bs
->drv
->bdrv_co_check(bs
, res
, fix
);
3698 typedef struct CheckCo
{
3699 BlockDriverState
*bs
;
3700 BdrvCheckResult
*res
;
3705 static void bdrv_check_co_entry(void *opaque
)
3707 CheckCo
*cco
= opaque
;
3708 cco
->ret
= bdrv_co_check(cco
->bs
, cco
->res
, cco
->fix
);
3711 int bdrv_check(BlockDriverState
*bs
,
3712 BdrvCheckResult
*res
, BdrvCheckMode fix
)
3718 .ret
= -EINPROGRESS
,
3722 if (qemu_in_coroutine()) {
3723 /* Fast-path if already in coroutine context */
3724 bdrv_check_co_entry(&cco
);
3726 co
= qemu_coroutine_create(bdrv_check_co_entry
, &cco
);
3727 qemu_coroutine_enter(co
);
3728 BDRV_POLL_WHILE(bs
, cco
.ret
== -EINPROGRESS
);
3737 * -EINVAL - backing format specified, but no file
3738 * -ENOSPC - can't update the backing file because no space is left in the
3740 * -ENOTSUP - format driver doesn't support changing the backing file
3742 int bdrv_change_backing_file(BlockDriverState
*bs
,
3743 const char *backing_file
, const char *backing_fmt
)
3745 BlockDriver
*drv
= bs
->drv
;
3752 /* Backing file format doesn't make sense without a backing file */
3753 if (backing_fmt
&& !backing_file
) {
3757 if (drv
->bdrv_change_backing_file
!= NULL
) {
3758 ret
= drv
->bdrv_change_backing_file(bs
, backing_file
, backing_fmt
);
3764 pstrcpy(bs
->backing_file
, sizeof(bs
->backing_file
), backing_file
?: "");
3765 pstrcpy(bs
->backing_format
, sizeof(bs
->backing_format
), backing_fmt
?: "");
3771 * Finds the image layer in the chain that has 'bs' as its backing file.
3773 * active is the current topmost image.
3775 * Returns NULL if bs is not found in active's image chain,
3776 * or if active == bs.
3778 * Returns the bottommost base image if bs == NULL.
3780 BlockDriverState
*bdrv_find_overlay(BlockDriverState
*active
,
3781 BlockDriverState
*bs
)
3783 while (active
&& bs
!= backing_bs(active
)) {
3784 active
= backing_bs(active
);
3790 /* Given a BDS, searches for the base layer. */
3791 BlockDriverState
*bdrv_find_base(BlockDriverState
*bs
)
3793 return bdrv_find_overlay(bs
, NULL
);
3797 * Drops images above 'base' up to and including 'top', and sets the image
3798 * above 'top' to have base as its backing file.
3800 * Requires that the overlay to 'top' is opened r/w, so that the backing file
3801 * information in 'bs' can be properly updated.
3803 * E.g., this will convert the following chain:
3804 * bottom <- base <- intermediate <- top <- active
3808 * bottom <- base <- active
3810 * It is allowed for bottom==base, in which case it converts:
3812 * base <- intermediate <- top <- active
3818 * If backing_file_str is non-NULL, it will be used when modifying top's
3819 * overlay image metadata.
3822 * if active == top, that is considered an error
3825 int bdrv_drop_intermediate(BlockDriverState
*top
, BlockDriverState
*base
,
3826 const char *backing_file_str
)
3828 BdrvChild
*c
, *next
;
3829 Error
*local_err
= NULL
;
3834 if (!top
->drv
|| !base
->drv
) {
3838 /* Make sure that base is in the backing chain of top */
3839 if (!bdrv_chain_contains(top
, base
)) {
3843 /* success - we can delete the intermediate states, and link top->base */
3844 /* TODO Check graph modification op blockers (BLK_PERM_GRAPH_MOD) once
3845 * we've figured out how they should work. */
3846 backing_file_str
= backing_file_str
? backing_file_str
: base
->filename
;
3848 QLIST_FOREACH_SAFE(c
, &top
->parents
, next_parent
, next
) {
3849 /* Check whether we are allowed to switch c from top to base */
3850 GSList
*ignore_children
= g_slist_prepend(NULL
, c
);
3851 bdrv_check_update_perm(base
, NULL
, c
->perm
, c
->shared_perm
,
3852 ignore_children
, &local_err
);
3853 g_slist_free(ignore_children
);
3856 error_report_err(local_err
);
3860 /* If so, update the backing file path in the image file */
3861 if (c
->role
->update_filename
) {
3862 ret
= c
->role
->update_filename(c
, base
, backing_file_str
,
3865 bdrv_abort_perm_update(base
);
3866 error_report_err(local_err
);
3871 /* Do the actual switch in the in-memory graph.
3872 * Completes bdrv_check_update_perm() transaction internally. */
3874 bdrv_replace_child(c
, base
);
3885 * Length of a allocated file in bytes. Sparse files are counted by actual
3886 * allocated space. Return < 0 if error or unknown.
3888 int64_t bdrv_get_allocated_file_size(BlockDriverState
*bs
)
3890 BlockDriver
*drv
= bs
->drv
;
3894 if (drv
->bdrv_get_allocated_file_size
) {
3895 return drv
->bdrv_get_allocated_file_size(bs
);
3898 return bdrv_get_allocated_file_size(bs
->file
->bs
);
3905 * @drv: Format driver
3906 * @opts: Creation options for new image
3907 * @in_bs: Existing image containing data for new image (may be NULL)
3908 * @errp: Error object
3909 * Returns: A #BlockMeasureInfo (free using qapi_free_BlockMeasureInfo())
3912 * Calculate file size required to create a new image.
3914 * If @in_bs is given then space for allocated clusters and zero clusters
3915 * from that image are included in the calculation. If @opts contains a
3916 * backing file that is shared by @in_bs then backing clusters may be omitted
3917 * from the calculation.
3919 * If @in_bs is NULL then the calculation includes no allocated clusters
3920 * unless a preallocation option is given in @opts.
3922 * Note that @in_bs may use a different BlockDriver from @drv.
3924 * If an error occurs the @errp pointer is set.
3926 BlockMeasureInfo
*bdrv_measure(BlockDriver
*drv
, QemuOpts
*opts
,
3927 BlockDriverState
*in_bs
, Error
**errp
)
3929 if (!drv
->bdrv_measure
) {
3930 error_setg(errp
, "Block driver '%s' does not support size measurement",
3935 return drv
->bdrv_measure(opts
, in_bs
, errp
);
3939 * Return number of sectors on success, -errno on error.
3941 int64_t bdrv_nb_sectors(BlockDriverState
*bs
)
3943 BlockDriver
*drv
= bs
->drv
;
3948 if (drv
->has_variable_length
) {
3949 int ret
= refresh_total_sectors(bs
, bs
->total_sectors
);
3954 return bs
->total_sectors
;
3958 * Return length in bytes on success, -errno on error.
3959 * The length is always a multiple of BDRV_SECTOR_SIZE.
3961 int64_t bdrv_getlength(BlockDriverState
*bs
)
3963 int64_t ret
= bdrv_nb_sectors(bs
);
3965 ret
= ret
> INT64_MAX
/ BDRV_SECTOR_SIZE
? -EFBIG
: ret
;
3966 return ret
< 0 ? ret
: ret
* BDRV_SECTOR_SIZE
;
3969 /* return 0 as number of sectors if no device present or error */
3970 void bdrv_get_geometry(BlockDriverState
*bs
, uint64_t *nb_sectors_ptr
)
3972 int64_t nb_sectors
= bdrv_nb_sectors(bs
);
3974 *nb_sectors_ptr
= nb_sectors
< 0 ? 0 : nb_sectors
;
3977 bool bdrv_is_sg(BlockDriverState
*bs
)
3982 bool bdrv_is_encrypted(BlockDriverState
*bs
)
3984 if (bs
->backing
&& bs
->backing
->bs
->encrypted
) {
3987 return bs
->encrypted
;
3990 const char *bdrv_get_format_name(BlockDriverState
*bs
)
3992 return bs
->drv
? bs
->drv
->format_name
: NULL
;
3995 static int qsort_strcmp(const void *a
, const void *b
)
3997 return strcmp(*(char *const *)a
, *(char *const *)b
);
4000 void bdrv_iterate_format(void (*it
)(void *opaque
, const char *name
),
4006 const char **formats
= NULL
;
4008 QLIST_FOREACH(drv
, &bdrv_drivers
, list
) {
4009 if (drv
->format_name
) {
4012 while (formats
&& i
&& !found
) {
4013 found
= !strcmp(formats
[--i
], drv
->format_name
);
4017 formats
= g_renew(const char *, formats
, count
+ 1);
4018 formats
[count
++] = drv
->format_name
;
4023 for (i
= 0; i
< (int)ARRAY_SIZE(block_driver_modules
); i
++) {
4024 const char *format_name
= block_driver_modules
[i
].format_name
;
4030 while (formats
&& j
&& !found
) {
4031 found
= !strcmp(formats
[--j
], format_name
);
4035 formats
= g_renew(const char *, formats
, count
+ 1);
4036 formats
[count
++] = format_name
;
4041 qsort(formats
, count
, sizeof(formats
[0]), qsort_strcmp
);
4043 for (i
= 0; i
< count
; i
++) {
4044 it(opaque
, formats
[i
]);
4050 /* This function is to find a node in the bs graph */
4051 BlockDriverState
*bdrv_find_node(const char *node_name
)
4053 BlockDriverState
*bs
;
4057 QTAILQ_FOREACH(bs
, &graph_bdrv_states
, node_list
) {
4058 if (!strcmp(node_name
, bs
->node_name
)) {
4065 /* Put this QMP function here so it can access the static graph_bdrv_states. */
4066 BlockDeviceInfoList
*bdrv_named_nodes_list(Error
**errp
)
4068 BlockDeviceInfoList
*list
, *entry
;
4069 BlockDriverState
*bs
;
4072 QTAILQ_FOREACH(bs
, &graph_bdrv_states
, node_list
) {
4073 BlockDeviceInfo
*info
= bdrv_block_device_info(NULL
, bs
, errp
);
4075 qapi_free_BlockDeviceInfoList(list
);
4078 entry
= g_malloc0(sizeof(*entry
));
4079 entry
->value
= info
;
4087 BlockDriverState
*bdrv_lookup_bs(const char *device
,
4088 const char *node_name
,
4092 BlockDriverState
*bs
;
4095 blk
= blk_by_name(device
);
4100 error_setg(errp
, "Device '%s' has no medium", device
);
4108 bs
= bdrv_find_node(node_name
);
4115 error_setg(errp
, "Cannot find device=%s nor node_name=%s",
4116 device
? device
: "",
4117 node_name
? node_name
: "");
4121 /* If 'base' is in the same chain as 'top', return true. Otherwise,
4122 * return false. If either argument is NULL, return false. */
4123 bool bdrv_chain_contains(BlockDriverState
*top
, BlockDriverState
*base
)
4125 while (top
&& top
!= base
) {
4126 top
= backing_bs(top
);
4132 BlockDriverState
*bdrv_next_node(BlockDriverState
*bs
)
4135 return QTAILQ_FIRST(&graph_bdrv_states
);
4137 return QTAILQ_NEXT(bs
, node_list
);
4140 BlockDriverState
*bdrv_next_all_states(BlockDriverState
*bs
)
4143 return QTAILQ_FIRST(&all_bdrv_states
);
4145 return QTAILQ_NEXT(bs
, bs_list
);
4148 const char *bdrv_get_node_name(const BlockDriverState
*bs
)
4150 return bs
->node_name
;
4153 const char *bdrv_get_parent_name(const BlockDriverState
*bs
)
4158 /* If multiple parents have a name, just pick the first one. */
4159 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
4160 if (c
->role
->get_name
) {
4161 name
= c
->role
->get_name(c
);
4162 if (name
&& *name
) {
4171 /* TODO check what callers really want: bs->node_name or blk_name() */
4172 const char *bdrv_get_device_name(const BlockDriverState
*bs
)
4174 return bdrv_get_parent_name(bs
) ?: "";
4177 /* This can be used to identify nodes that might not have a device
4178 * name associated. Since node and device names live in the same
4179 * namespace, the result is unambiguous. The exception is if both are
4180 * absent, then this returns an empty (non-null) string. */
4181 const char *bdrv_get_device_or_node_name(const BlockDriverState
*bs
)
4183 return bdrv_get_parent_name(bs
) ?: bs
->node_name
;
4186 int bdrv_get_flags(BlockDriverState
*bs
)
4188 return bs
->open_flags
;
4191 int bdrv_has_zero_init_1(BlockDriverState
*bs
)
4196 int bdrv_has_zero_init(BlockDriverState
*bs
)
4202 /* If BS is a copy on write image, it is initialized to
4203 the contents of the base image, which may not be zeroes. */
4207 if (bs
->drv
->bdrv_has_zero_init
) {
4208 return bs
->drv
->bdrv_has_zero_init(bs
);
4210 if (bs
->file
&& bs
->drv
->is_filter
) {
4211 return bdrv_has_zero_init(bs
->file
->bs
);
4218 bool bdrv_unallocated_blocks_are_zero(BlockDriverState
*bs
)
4220 BlockDriverInfo bdi
;
4226 if (bdrv_get_info(bs
, &bdi
) == 0) {
4227 return bdi
.unallocated_blocks_are_zero
;
4233 bool bdrv_can_write_zeroes_with_unmap(BlockDriverState
*bs
)
4235 if (!(bs
->open_flags
& BDRV_O_UNMAP
)) {
4239 return bs
->supported_zero_flags
& BDRV_REQ_MAY_UNMAP
;
4242 const char *bdrv_get_encrypted_filename(BlockDriverState
*bs
)
4244 if (bs
->backing
&& bs
->backing
->bs
->encrypted
)
4245 return bs
->backing_file
;
4246 else if (bs
->encrypted
)
4247 return bs
->filename
;
4252 void bdrv_get_backing_filename(BlockDriverState
*bs
,
4253 char *filename
, int filename_size
)
4255 pstrcpy(filename
, filename_size
, bs
->backing_file
);
4258 int bdrv_get_info(BlockDriverState
*bs
, BlockDriverInfo
*bdi
)
4260 BlockDriver
*drv
= bs
->drv
;
4261 /* if bs->drv == NULL, bs is closed, so there's nothing to do here */
4265 if (!drv
->bdrv_get_info
) {
4266 if (bs
->file
&& drv
->is_filter
) {
4267 return bdrv_get_info(bs
->file
->bs
, bdi
);
4271 memset(bdi
, 0, sizeof(*bdi
));
4272 return drv
->bdrv_get_info(bs
, bdi
);
4275 ImageInfoSpecific
*bdrv_get_specific_info(BlockDriverState
*bs
)
4277 BlockDriver
*drv
= bs
->drv
;
4278 if (drv
&& drv
->bdrv_get_specific_info
) {
4279 return drv
->bdrv_get_specific_info(bs
);
4284 void bdrv_debug_event(BlockDriverState
*bs
, BlkdebugEvent event
)
4286 if (!bs
|| !bs
->drv
|| !bs
->drv
->bdrv_debug_event
) {
4290 bs
->drv
->bdrv_debug_event(bs
, event
);
4293 int bdrv_debug_breakpoint(BlockDriverState
*bs
, const char *event
,
4296 while (bs
&& bs
->drv
&& !bs
->drv
->bdrv_debug_breakpoint
) {
4297 bs
= bs
->file
? bs
->file
->bs
: NULL
;
4300 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_breakpoint
) {
4301 return bs
->drv
->bdrv_debug_breakpoint(bs
, event
, tag
);
4307 int bdrv_debug_remove_breakpoint(BlockDriverState
*bs
, const char *tag
)
4309 while (bs
&& bs
->drv
&& !bs
->drv
->bdrv_debug_remove_breakpoint
) {
4310 bs
= bs
->file
? bs
->file
->bs
: NULL
;
4313 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_remove_breakpoint
) {
4314 return bs
->drv
->bdrv_debug_remove_breakpoint(bs
, tag
);
4320 int bdrv_debug_resume(BlockDriverState
*bs
, const char *tag
)
4322 while (bs
&& (!bs
->drv
|| !bs
->drv
->bdrv_debug_resume
)) {
4323 bs
= bs
->file
? bs
->file
->bs
: NULL
;
4326 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_resume
) {
4327 return bs
->drv
->bdrv_debug_resume(bs
, tag
);
4333 bool bdrv_debug_is_suspended(BlockDriverState
*bs
, const char *tag
)
4335 while (bs
&& bs
->drv
&& !bs
->drv
->bdrv_debug_is_suspended
) {
4336 bs
= bs
->file
? bs
->file
->bs
: NULL
;
4339 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_is_suspended
) {
4340 return bs
->drv
->bdrv_debug_is_suspended(bs
, tag
);
4346 /* backing_file can either be relative, or absolute, or a protocol. If it is
4347 * relative, it must be relative to the chain. So, passing in bs->filename
4348 * from a BDS as backing_file should not be done, as that may be relative to
4349 * the CWD rather than the chain. */
4350 BlockDriverState
*bdrv_find_backing_image(BlockDriverState
*bs
,
4351 const char *backing_file
)
4353 char *filename_full
= NULL
;
4354 char *backing_file_full
= NULL
;
4355 char *filename_tmp
= NULL
;
4356 int is_protocol
= 0;
4357 BlockDriverState
*curr_bs
= NULL
;
4358 BlockDriverState
*retval
= NULL
;
4359 Error
*local_error
= NULL
;
4361 if (!bs
|| !bs
->drv
|| !backing_file
) {
4365 filename_full
= g_malloc(PATH_MAX
);
4366 backing_file_full
= g_malloc(PATH_MAX
);
4367 filename_tmp
= g_malloc(PATH_MAX
);
4369 is_protocol
= path_has_protocol(backing_file
);
4371 for (curr_bs
= bs
; curr_bs
->backing
; curr_bs
= curr_bs
->backing
->bs
) {
4373 /* If either of the filename paths is actually a protocol, then
4374 * compare unmodified paths; otherwise make paths relative */
4375 if (is_protocol
|| path_has_protocol(curr_bs
->backing_file
)) {
4376 if (strcmp(backing_file
, curr_bs
->backing_file
) == 0) {
4377 retval
= curr_bs
->backing
->bs
;
4380 /* Also check against the full backing filename for the image */
4381 bdrv_get_full_backing_filename(curr_bs
, backing_file_full
, PATH_MAX
,
4383 if (local_error
== NULL
) {
4384 if (strcmp(backing_file
, backing_file_full
) == 0) {
4385 retval
= curr_bs
->backing
->bs
;
4389 error_free(local_error
);
4393 /* If not an absolute filename path, make it relative to the current
4394 * image's filename path */
4395 path_combine(filename_tmp
, PATH_MAX
, curr_bs
->filename
,
4398 /* We are going to compare absolute pathnames */
4399 if (!realpath(filename_tmp
, filename_full
)) {
4403 /* We need to make sure the backing filename we are comparing against
4404 * is relative to the current image filename (or absolute) */
4405 path_combine(filename_tmp
, PATH_MAX
, curr_bs
->filename
,
4406 curr_bs
->backing_file
);
4408 if (!realpath(filename_tmp
, backing_file_full
)) {
4412 if (strcmp(backing_file_full
, filename_full
) == 0) {
4413 retval
= curr_bs
->backing
->bs
;
4419 g_free(filename_full
);
4420 g_free(backing_file_full
);
4421 g_free(filename_tmp
);
4425 void bdrv_init(void)
4427 module_call_init(MODULE_INIT_BLOCK
);
4430 void bdrv_init_with_whitelist(void)
4432 use_bdrv_whitelist
= 1;
4436 static void coroutine_fn
bdrv_co_invalidate_cache(BlockDriverState
*bs
,
4439 BdrvChild
*child
, *parent
;
4440 uint64_t perm
, shared_perm
;
4441 Error
*local_err
= NULL
;
4443 BdrvDirtyBitmap
*bm
;
4449 if (!(bs
->open_flags
& BDRV_O_INACTIVE
)) {
4453 QLIST_FOREACH(child
, &bs
->children
, next
) {
4454 bdrv_co_invalidate_cache(child
->bs
, &local_err
);
4456 error_propagate(errp
, local_err
);
4462 * Update permissions, they may differ for inactive nodes.
4464 * Note that the required permissions of inactive images are always a
4465 * subset of the permissions required after activating the image. This
4466 * allows us to just get the permissions upfront without restricting
4467 * drv->bdrv_invalidate_cache().
4469 * It also means that in error cases, we don't have to try and revert to
4470 * the old permissions (which is an operation that could fail, too). We can
4471 * just keep the extended permissions for the next time that an activation
4472 * of the image is tried.
4474 bs
->open_flags
&= ~BDRV_O_INACTIVE
;
4475 bdrv_get_cumulative_perm(bs
, &perm
, &shared_perm
);
4476 ret
= bdrv_check_perm(bs
, NULL
, perm
, shared_perm
, NULL
, &local_err
);
4478 bs
->open_flags
|= BDRV_O_INACTIVE
;
4479 error_propagate(errp
, local_err
);
4482 bdrv_set_perm(bs
, perm
, shared_perm
);
4484 if (bs
->drv
->bdrv_co_invalidate_cache
) {
4485 bs
->drv
->bdrv_co_invalidate_cache(bs
, &local_err
);
4487 bs
->open_flags
|= BDRV_O_INACTIVE
;
4488 error_propagate(errp
, local_err
);
4493 for (bm
= bdrv_dirty_bitmap_next(bs
, NULL
); bm
;
4494 bm
= bdrv_dirty_bitmap_next(bs
, bm
))
4496 bdrv_dirty_bitmap_set_migration(bm
, false);
4499 ret
= refresh_total_sectors(bs
, bs
->total_sectors
);
4501 bs
->open_flags
|= BDRV_O_INACTIVE
;
4502 error_setg_errno(errp
, -ret
, "Could not refresh total sector count");
4506 QLIST_FOREACH(parent
, &bs
->parents
, next_parent
) {
4507 if (parent
->role
->activate
) {
4508 parent
->role
->activate(parent
, &local_err
);
4510 error_propagate(errp
, local_err
);
4517 typedef struct InvalidateCacheCo
{
4518 BlockDriverState
*bs
;
4521 } InvalidateCacheCo
;
4523 static void coroutine_fn
bdrv_invalidate_cache_co_entry(void *opaque
)
4525 InvalidateCacheCo
*ico
= opaque
;
4526 bdrv_co_invalidate_cache(ico
->bs
, ico
->errp
);
4530 void bdrv_invalidate_cache(BlockDriverState
*bs
, Error
**errp
)
4533 InvalidateCacheCo ico
= {
4539 if (qemu_in_coroutine()) {
4540 /* Fast-path if already in coroutine context */
4541 bdrv_invalidate_cache_co_entry(&ico
);
4543 co
= qemu_coroutine_create(bdrv_invalidate_cache_co_entry
, &ico
);
4544 qemu_coroutine_enter(co
);
4545 BDRV_POLL_WHILE(bs
, !ico
.done
);
4549 void bdrv_invalidate_cache_all(Error
**errp
)
4551 BlockDriverState
*bs
;
4552 Error
*local_err
= NULL
;
4553 BdrvNextIterator it
;
4555 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
4556 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
4558 aio_context_acquire(aio_context
);
4559 bdrv_invalidate_cache(bs
, &local_err
);
4560 aio_context_release(aio_context
);
4562 error_propagate(errp
, local_err
);
4563 bdrv_next_cleanup(&it
);
4569 static int bdrv_inactivate_recurse(BlockDriverState
*bs
,
4572 BdrvChild
*child
, *parent
;
4579 if (!setting_flag
&& bs
->drv
->bdrv_inactivate
) {
4580 ret
= bs
->drv
->bdrv_inactivate(bs
);
4586 if (setting_flag
&& !(bs
->open_flags
& BDRV_O_INACTIVE
)) {
4587 uint64_t perm
, shared_perm
;
4589 QLIST_FOREACH(parent
, &bs
->parents
, next_parent
) {
4590 if (parent
->role
->inactivate
) {
4591 ret
= parent
->role
->inactivate(parent
);
4598 bs
->open_flags
|= BDRV_O_INACTIVE
;
4600 /* Update permissions, they may differ for inactive nodes */
4601 bdrv_get_cumulative_perm(bs
, &perm
, &shared_perm
);
4602 bdrv_check_perm(bs
, NULL
, perm
, shared_perm
, NULL
, &error_abort
);
4603 bdrv_set_perm(bs
, perm
, shared_perm
);
4606 QLIST_FOREACH(child
, &bs
->children
, next
) {
4607 ret
= bdrv_inactivate_recurse(child
->bs
, setting_flag
);
4616 int bdrv_inactivate_all(void)
4618 BlockDriverState
*bs
= NULL
;
4619 BdrvNextIterator it
;
4622 GSList
*aio_ctxs
= NULL
, *ctx
;
4624 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
4625 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
4627 if (!g_slist_find(aio_ctxs
, aio_context
)) {
4628 aio_ctxs
= g_slist_prepend(aio_ctxs
, aio_context
);
4629 aio_context_acquire(aio_context
);
4633 /* We do two passes of inactivation. The first pass calls to drivers'
4634 * .bdrv_inactivate callbacks recursively so all cache is flushed to disk;
4635 * the second pass sets the BDRV_O_INACTIVE flag so that no further write
4637 for (pass
= 0; pass
< 2; pass
++) {
4638 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
4639 ret
= bdrv_inactivate_recurse(bs
, pass
);
4641 bdrv_next_cleanup(&it
);
4648 for (ctx
= aio_ctxs
; ctx
!= NULL
; ctx
= ctx
->next
) {
4649 AioContext
*aio_context
= ctx
->data
;
4650 aio_context_release(aio_context
);
4652 g_slist_free(aio_ctxs
);
4657 /**************************************************************/
4658 /* removable device support */
4661 * Return TRUE if the media is present
4663 bool bdrv_is_inserted(BlockDriverState
*bs
)
4665 BlockDriver
*drv
= bs
->drv
;
4671 if (drv
->bdrv_is_inserted
) {
4672 return drv
->bdrv_is_inserted(bs
);
4674 QLIST_FOREACH(child
, &bs
->children
, next
) {
4675 if (!bdrv_is_inserted(child
->bs
)) {
4683 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
4685 void bdrv_eject(BlockDriverState
*bs
, bool eject_flag
)
4687 BlockDriver
*drv
= bs
->drv
;
4689 if (drv
&& drv
->bdrv_eject
) {
4690 drv
->bdrv_eject(bs
, eject_flag
);
4695 * Lock or unlock the media (if it is locked, the user won't be able
4696 * to eject it manually).
4698 void bdrv_lock_medium(BlockDriverState
*bs
, bool locked
)
4700 BlockDriver
*drv
= bs
->drv
;
4702 trace_bdrv_lock_medium(bs
, locked
);
4704 if (drv
&& drv
->bdrv_lock_medium
) {
4705 drv
->bdrv_lock_medium(bs
, locked
);
4709 /* Get a reference to bs */
4710 void bdrv_ref(BlockDriverState
*bs
)
4715 /* Release a previously grabbed reference to bs.
4716 * If after releasing, reference count is zero, the BlockDriverState is
4718 void bdrv_unref(BlockDriverState
*bs
)
4723 assert(bs
->refcnt
> 0);
4724 if (--bs
->refcnt
== 0) {
4729 struct BdrvOpBlocker
{
4731 QLIST_ENTRY(BdrvOpBlocker
) list
;
4734 bool bdrv_op_is_blocked(BlockDriverState
*bs
, BlockOpType op
, Error
**errp
)
4736 BdrvOpBlocker
*blocker
;
4737 assert((int) op
>= 0 && op
< BLOCK_OP_TYPE_MAX
);
4738 if (!QLIST_EMPTY(&bs
->op_blockers
[op
])) {
4739 blocker
= QLIST_FIRST(&bs
->op_blockers
[op
]);
4740 error_propagate_prepend(errp
, error_copy(blocker
->reason
),
4741 "Node '%s' is busy: ",
4742 bdrv_get_device_or_node_name(bs
));
4748 void bdrv_op_block(BlockDriverState
*bs
, BlockOpType op
, Error
*reason
)
4750 BdrvOpBlocker
*blocker
;
4751 assert((int) op
>= 0 && op
< BLOCK_OP_TYPE_MAX
);
4753 blocker
= g_new0(BdrvOpBlocker
, 1);
4754 blocker
->reason
= reason
;
4755 QLIST_INSERT_HEAD(&bs
->op_blockers
[op
], blocker
, list
);
4758 void bdrv_op_unblock(BlockDriverState
*bs
, BlockOpType op
, Error
*reason
)
4760 BdrvOpBlocker
*blocker
, *next
;
4761 assert((int) op
>= 0 && op
< BLOCK_OP_TYPE_MAX
);
4762 QLIST_FOREACH_SAFE(blocker
, &bs
->op_blockers
[op
], list
, next
) {
4763 if (blocker
->reason
== reason
) {
4764 QLIST_REMOVE(blocker
, list
);
4770 void bdrv_op_block_all(BlockDriverState
*bs
, Error
*reason
)
4773 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
4774 bdrv_op_block(bs
, i
, reason
);
4778 void bdrv_op_unblock_all(BlockDriverState
*bs
, Error
*reason
)
4781 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
4782 bdrv_op_unblock(bs
, i
, reason
);
4786 bool bdrv_op_blocker_is_empty(BlockDriverState
*bs
)
4790 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
4791 if (!QLIST_EMPTY(&bs
->op_blockers
[i
])) {
4798 void bdrv_img_create(const char *filename
, const char *fmt
,
4799 const char *base_filename
, const char *base_fmt
,
4800 char *options
, uint64_t img_size
, int flags
, bool quiet
,
4803 QemuOptsList
*create_opts
= NULL
;
4804 QemuOpts
*opts
= NULL
;
4805 const char *backing_fmt
, *backing_file
;
4807 BlockDriver
*drv
, *proto_drv
;
4808 Error
*local_err
= NULL
;
4811 /* Find driver and parse its options */
4812 drv
= bdrv_find_format(fmt
);
4814 error_setg(errp
, "Unknown file format '%s'", fmt
);
4818 proto_drv
= bdrv_find_protocol(filename
, true, errp
);
4823 if (!drv
->create_opts
) {
4824 error_setg(errp
, "Format driver '%s' does not support image creation",
4829 if (!proto_drv
->create_opts
) {
4830 error_setg(errp
, "Protocol driver '%s' does not support image creation",
4831 proto_drv
->format_name
);
4835 create_opts
= qemu_opts_append(create_opts
, drv
->create_opts
);
4836 create_opts
= qemu_opts_append(create_opts
, proto_drv
->create_opts
);
4838 /* Create parameter list with default values */
4839 opts
= qemu_opts_create(create_opts
, NULL
, 0, &error_abort
);
4840 qemu_opt_set_number(opts
, BLOCK_OPT_SIZE
, img_size
, &error_abort
);
4842 /* Parse -o options */
4844 qemu_opts_do_parse(opts
, options
, NULL
, &local_err
);
4850 if (base_filename
) {
4851 qemu_opt_set(opts
, BLOCK_OPT_BACKING_FILE
, base_filename
, &local_err
);
4853 error_setg(errp
, "Backing file not supported for file format '%s'",
4860 qemu_opt_set(opts
, BLOCK_OPT_BACKING_FMT
, base_fmt
, &local_err
);
4862 error_setg(errp
, "Backing file format not supported for file "
4863 "format '%s'", fmt
);
4868 backing_file
= qemu_opt_get(opts
, BLOCK_OPT_BACKING_FILE
);
4870 if (!strcmp(filename
, backing_file
)) {
4871 error_setg(errp
, "Error: Trying to create an image with the "
4872 "same filename as the backing file");
4877 backing_fmt
= qemu_opt_get(opts
, BLOCK_OPT_BACKING_FMT
);
4879 /* The size for the image must always be specified, unless we have a backing
4880 * file and we have not been forbidden from opening it. */
4881 size
= qemu_opt_get_size(opts
, BLOCK_OPT_SIZE
, img_size
);
4882 if (backing_file
&& !(flags
& BDRV_O_NO_BACKING
)) {
4883 BlockDriverState
*bs
;
4884 char *full_backing
= g_new0(char, PATH_MAX
);
4886 QDict
*backing_options
= NULL
;
4888 bdrv_get_full_backing_filename_from_filename(filename
, backing_file
,
4889 full_backing
, PATH_MAX
,
4892 g_free(full_backing
);
4896 /* backing files always opened read-only */
4898 back_flags
&= ~(BDRV_O_RDWR
| BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
);
4900 backing_options
= qdict_new();
4902 qdict_put_str(backing_options
, "driver", backing_fmt
);
4904 qdict_put_bool(backing_options
, BDRV_OPT_FORCE_SHARE
, true);
4906 bs
= bdrv_open(full_backing
, NULL
, backing_options
, back_flags
,
4908 g_free(full_backing
);
4909 if (!bs
&& size
!= -1) {
4910 /* Couldn't open BS, but we have a size, so it's nonfatal */
4911 warn_reportf_err(local_err
,
4912 "Could not verify backing image. "
4913 "This may become an error in future versions.\n");
4916 /* Couldn't open bs, do not have size */
4917 error_append_hint(&local_err
,
4918 "Could not open backing image to determine size.\n");
4922 /* Opened BS, have no size */
4923 size
= bdrv_getlength(bs
);
4925 error_setg_errno(errp
, -size
, "Could not get size of '%s'",
4930 qemu_opt_set_number(opts
, BLOCK_OPT_SIZE
, size
, &error_abort
);
4934 } /* (backing_file && !(flags & BDRV_O_NO_BACKING)) */
4937 error_setg(errp
, "Image creation needs a size parameter");
4942 printf("Formatting '%s', fmt=%s ", filename
, fmt
);
4943 qemu_opts_print(opts
, " ");
4947 ret
= bdrv_create(drv
, filename
, opts
, &local_err
);
4949 if (ret
== -EFBIG
) {
4950 /* This is generally a better message than whatever the driver would
4951 * deliver (especially because of the cluster_size_hint), since that
4952 * is most probably not much different from "image too large". */
4953 const char *cluster_size_hint
= "";
4954 if (qemu_opt_get_size(opts
, BLOCK_OPT_CLUSTER_SIZE
, 0)) {
4955 cluster_size_hint
= " (try using a larger cluster size)";
4957 error_setg(errp
, "The image size is too large for file format '%s'"
4958 "%s", fmt
, cluster_size_hint
);
4959 error_free(local_err
);
4964 qemu_opts_del(opts
);
4965 qemu_opts_free(create_opts
);
4966 error_propagate(errp
, local_err
);
4969 AioContext
*bdrv_get_aio_context(BlockDriverState
*bs
)
4971 return bs
? bs
->aio_context
: qemu_get_aio_context();
4974 void bdrv_coroutine_enter(BlockDriverState
*bs
, Coroutine
*co
)
4976 aio_co_enter(bdrv_get_aio_context(bs
), co
);
4979 static void bdrv_do_remove_aio_context_notifier(BdrvAioNotifier
*ban
)
4981 QLIST_REMOVE(ban
, list
);
4985 void bdrv_detach_aio_context(BlockDriverState
*bs
)
4987 BdrvAioNotifier
*baf
, *baf_tmp
;
4994 assert(!bs
->walking_aio_notifiers
);
4995 bs
->walking_aio_notifiers
= true;
4996 QLIST_FOREACH_SAFE(baf
, &bs
->aio_notifiers
, list
, baf_tmp
) {
4998 bdrv_do_remove_aio_context_notifier(baf
);
5000 baf
->detach_aio_context(baf
->opaque
);
5003 /* Never mind iterating again to check for ->deleted. bdrv_close() will
5004 * remove remaining aio notifiers if we aren't called again.
5006 bs
->walking_aio_notifiers
= false;
5008 if (bs
->drv
->bdrv_detach_aio_context
) {
5009 bs
->drv
->bdrv_detach_aio_context(bs
);
5011 QLIST_FOREACH(child
, &bs
->children
, next
) {
5012 bdrv_detach_aio_context(child
->bs
);
5015 bs
->aio_context
= NULL
;
5018 void bdrv_attach_aio_context(BlockDriverState
*bs
,
5019 AioContext
*new_context
)
5021 BdrvAioNotifier
*ban
, *ban_tmp
;
5028 bs
->aio_context
= new_context
;
5030 QLIST_FOREACH(child
, &bs
->children
, next
) {
5031 bdrv_attach_aio_context(child
->bs
, new_context
);
5033 if (bs
->drv
->bdrv_attach_aio_context
) {
5034 bs
->drv
->bdrv_attach_aio_context(bs
, new_context
);
5037 assert(!bs
->walking_aio_notifiers
);
5038 bs
->walking_aio_notifiers
= true;
5039 QLIST_FOREACH_SAFE(ban
, &bs
->aio_notifiers
, list
, ban_tmp
) {
5041 bdrv_do_remove_aio_context_notifier(ban
);
5043 ban
->attached_aio_context(new_context
, ban
->opaque
);
5046 bs
->walking_aio_notifiers
= false;
5049 void bdrv_set_aio_context(BlockDriverState
*bs
, AioContext
*new_context
)
5051 AioContext
*ctx
= bdrv_get_aio_context(bs
);
5053 aio_disable_external(ctx
);
5054 bdrv_parent_drained_begin(bs
, NULL
, false);
5055 bdrv_drain(bs
); /* ensure there are no in-flight requests */
5057 while (aio_poll(ctx
, false)) {
5058 /* wait for all bottom halves to execute */
5061 bdrv_detach_aio_context(bs
);
5063 /* This function executes in the old AioContext so acquire the new one in
5064 * case it runs in a different thread.
5066 aio_context_acquire(new_context
);
5067 bdrv_attach_aio_context(bs
, new_context
);
5068 bdrv_parent_drained_end(bs
, NULL
, false);
5069 aio_enable_external(ctx
);
5070 aio_context_release(new_context
);
5073 void bdrv_add_aio_context_notifier(BlockDriverState
*bs
,
5074 void (*attached_aio_context
)(AioContext
*new_context
, void *opaque
),
5075 void (*detach_aio_context
)(void *opaque
), void *opaque
)
5077 BdrvAioNotifier
*ban
= g_new(BdrvAioNotifier
, 1);
5078 *ban
= (BdrvAioNotifier
){
5079 .attached_aio_context
= attached_aio_context
,
5080 .detach_aio_context
= detach_aio_context
,
5084 QLIST_INSERT_HEAD(&bs
->aio_notifiers
, ban
, list
);
5087 void bdrv_remove_aio_context_notifier(BlockDriverState
*bs
,
5088 void (*attached_aio_context
)(AioContext
*,
5090 void (*detach_aio_context
)(void *),
5093 BdrvAioNotifier
*ban
, *ban_next
;
5095 QLIST_FOREACH_SAFE(ban
, &bs
->aio_notifiers
, list
, ban_next
) {
5096 if (ban
->attached_aio_context
== attached_aio_context
&&
5097 ban
->detach_aio_context
== detach_aio_context
&&
5098 ban
->opaque
== opaque
&&
5099 ban
->deleted
== false)
5101 if (bs
->walking_aio_notifiers
) {
5102 ban
->deleted
= true;
5104 bdrv_do_remove_aio_context_notifier(ban
);
5113 int bdrv_amend_options(BlockDriverState
*bs
, QemuOpts
*opts
,
5114 BlockDriverAmendStatusCB
*status_cb
, void *cb_opaque
,
5118 error_setg(errp
, "Node is ejected");
5121 if (!bs
->drv
->bdrv_amend_options
) {
5122 error_setg(errp
, "Block driver '%s' does not support option amendment",
5123 bs
->drv
->format_name
);
5126 return bs
->drv
->bdrv_amend_options(bs
, opts
, status_cb
, cb_opaque
, errp
);
5129 /* This function will be called by the bdrv_recurse_is_first_non_filter method
5130 * of block filter and by bdrv_is_first_non_filter.
5131 * It is used to test if the given bs is the candidate or recurse more in the
5134 bool bdrv_recurse_is_first_non_filter(BlockDriverState
*bs
,
5135 BlockDriverState
*candidate
)
5137 /* return false if basic checks fails */
5138 if (!bs
|| !bs
->drv
) {
5142 /* the code reached a non block filter driver -> check if the bs is
5143 * the same as the candidate. It's the recursion termination condition.
5145 if (!bs
->drv
->is_filter
) {
5146 return bs
== candidate
;
5148 /* Down this path the driver is a block filter driver */
5150 /* If the block filter recursion method is defined use it to recurse down
5153 if (bs
->drv
->bdrv_recurse_is_first_non_filter
) {
5154 return bs
->drv
->bdrv_recurse_is_first_non_filter(bs
, candidate
);
5157 /* the driver is a block filter but don't allow to recurse -> return false
5162 /* This function checks if the candidate is the first non filter bs down it's
5163 * bs chain. Since we don't have pointers to parents it explore all bs chains
5164 * from the top. Some filters can choose not to pass down the recursion.
5166 bool bdrv_is_first_non_filter(BlockDriverState
*candidate
)
5168 BlockDriverState
*bs
;
5169 BdrvNextIterator it
;
5171 /* walk down the bs forest recursively */
5172 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
5175 /* try to recurse in this top level bs */
5176 perm
= bdrv_recurse_is_first_non_filter(bs
, candidate
);
5178 /* candidate is the first non filter */
5180 bdrv_next_cleanup(&it
);
5188 BlockDriverState
*check_to_replace_node(BlockDriverState
*parent_bs
,
5189 const char *node_name
, Error
**errp
)
5191 BlockDriverState
*to_replace_bs
= bdrv_find_node(node_name
);
5192 AioContext
*aio_context
;
5194 if (!to_replace_bs
) {
5195 error_setg(errp
, "Node name '%s' not found", node_name
);
5199 aio_context
= bdrv_get_aio_context(to_replace_bs
);
5200 aio_context_acquire(aio_context
);
5202 if (bdrv_op_is_blocked(to_replace_bs
, BLOCK_OP_TYPE_REPLACE
, errp
)) {
5203 to_replace_bs
= NULL
;
5207 /* We don't want arbitrary node of the BDS chain to be replaced only the top
5208 * most non filter in order to prevent data corruption.
5209 * Another benefit is that this tests exclude backing files which are
5210 * blocked by the backing blockers.
5212 if (!bdrv_recurse_is_first_non_filter(parent_bs
, to_replace_bs
)) {
5213 error_setg(errp
, "Only top most non filter can be replaced");
5214 to_replace_bs
= NULL
;
5219 aio_context_release(aio_context
);
5220 return to_replace_bs
;
5223 static bool append_open_options(QDict
*d
, BlockDriverState
*bs
)
5225 const QDictEntry
*entry
;
5227 bool found_any
= false;
5229 for (entry
= qdict_first(bs
->options
); entry
;
5230 entry
= qdict_next(bs
->options
, entry
))
5232 /* Exclude all non-driver-specific options */
5233 for (desc
= bdrv_runtime_opts
.desc
; desc
->name
; desc
++) {
5234 if (!strcmp(qdict_entry_key(entry
), desc
->name
)) {
5242 qdict_put_obj(d
, qdict_entry_key(entry
),
5243 qobject_ref(qdict_entry_value(entry
)));
5250 /* Updates the following BDS fields:
5251 * - exact_filename: A filename which may be used for opening a block device
5252 * which (mostly) equals the given BDS (even without any
5253 * other options; so reading and writing must return the same
5254 * results, but caching etc. may be different)
5255 * - full_open_options: Options which, when given when opening a block device
5256 * (without a filename), result in a BDS (mostly)
5257 * equalling the given one
5258 * - filename: If exact_filename is set, it is copied here. Otherwise,
5259 * full_open_options is converted to a JSON object, prefixed with
5260 * "json:" (for use through the JSON pseudo protocol) and put here.
5262 void bdrv_refresh_filename(BlockDriverState
*bs
)
5264 BlockDriver
*drv
= bs
->drv
;
5271 /* This BDS's file name will most probably depend on its file's name, so
5272 * refresh that first */
5274 bdrv_refresh_filename(bs
->file
->bs
);
5277 if (drv
->bdrv_refresh_filename
) {
5278 /* Obsolete information is of no use here, so drop the old file name
5279 * information before refreshing it */
5280 bs
->exact_filename
[0] = '\0';
5281 if (bs
->full_open_options
) {
5282 qobject_unref(bs
->full_open_options
);
5283 bs
->full_open_options
= NULL
;
5287 append_open_options(opts
, bs
);
5288 drv
->bdrv_refresh_filename(bs
, opts
);
5289 qobject_unref(opts
);
5290 } else if (bs
->file
) {
5291 /* Try to reconstruct valid information from the underlying file */
5292 bool has_open_options
;
5294 bs
->exact_filename
[0] = '\0';
5295 if (bs
->full_open_options
) {
5296 qobject_unref(bs
->full_open_options
);
5297 bs
->full_open_options
= NULL
;
5301 has_open_options
= append_open_options(opts
, bs
);
5303 /* If no specific options have been given for this BDS, the filename of
5304 * the underlying file should suffice for this one as well */
5305 if (bs
->file
->bs
->exact_filename
[0] && !has_open_options
) {
5306 strcpy(bs
->exact_filename
, bs
->file
->bs
->exact_filename
);
5308 /* Reconstructing the full options QDict is simple for most format block
5309 * drivers, as long as the full options are known for the underlying
5310 * file BDS. The full options QDict of that file BDS should somehow
5311 * contain a representation of the filename, therefore the following
5312 * suffices without querying the (exact_)filename of this BDS. */
5313 if (bs
->file
->bs
->full_open_options
) {
5314 qdict_put_str(opts
, "driver", drv
->format_name
);
5315 qdict_put(opts
, "file",
5316 qobject_ref(bs
->file
->bs
->full_open_options
));
5318 bs
->full_open_options
= opts
;
5320 qobject_unref(opts
);
5322 } else if (!bs
->full_open_options
&& qdict_size(bs
->options
)) {
5323 /* There is no underlying file BDS (at least referenced by BDS.file),
5324 * so the full options QDict should be equal to the options given
5325 * specifically for this block device when it was opened (plus the
5326 * driver specification).
5327 * Because those options don't change, there is no need to update
5328 * full_open_options when it's already set. */
5331 append_open_options(opts
, bs
);
5332 qdict_put_str(opts
, "driver", drv
->format_name
);
5334 if (bs
->exact_filename
[0]) {
5335 /* This may not work for all block protocol drivers (some may
5336 * require this filename to be parsed), but we have to find some
5337 * default solution here, so just include it. If some block driver
5338 * does not support pure options without any filename at all or
5339 * needs some special format of the options QDict, it needs to
5340 * implement the driver-specific bdrv_refresh_filename() function.
5342 qdict_put_str(opts
, "filename", bs
->exact_filename
);
5345 bs
->full_open_options
= opts
;
5348 if (bs
->exact_filename
[0]) {
5349 pstrcpy(bs
->filename
, sizeof(bs
->filename
), bs
->exact_filename
);
5350 } else if (bs
->full_open_options
) {
5351 QString
*json
= qobject_to_json(QOBJECT(bs
->full_open_options
));
5352 snprintf(bs
->filename
, sizeof(bs
->filename
), "json:%s",
5353 qstring_get_str(json
));
5354 qobject_unref(json
);
5359 * Hot add/remove a BDS's child. So the user can take a child offline when
5360 * it is broken and take a new child online
5362 void bdrv_add_child(BlockDriverState
*parent_bs
, BlockDriverState
*child_bs
,
5366 if (!parent_bs
->drv
|| !parent_bs
->drv
->bdrv_add_child
) {
5367 error_setg(errp
, "The node %s does not support adding a child",
5368 bdrv_get_device_or_node_name(parent_bs
));
5372 if (!QLIST_EMPTY(&child_bs
->parents
)) {
5373 error_setg(errp
, "The node %s already has a parent",
5374 child_bs
->node_name
);
5378 parent_bs
->drv
->bdrv_add_child(parent_bs
, child_bs
, errp
);
5381 void bdrv_del_child(BlockDriverState
*parent_bs
, BdrvChild
*child
, Error
**errp
)
5385 if (!parent_bs
->drv
|| !parent_bs
->drv
->bdrv_del_child
) {
5386 error_setg(errp
, "The node %s does not support removing a child",
5387 bdrv_get_device_or_node_name(parent_bs
));
5391 QLIST_FOREACH(tmp
, &parent_bs
->children
, next
) {
5398 error_setg(errp
, "The node %s does not have a child named %s",
5399 bdrv_get_device_or_node_name(parent_bs
),
5400 bdrv_get_device_or_node_name(child
->bs
));
5404 parent_bs
->drv
->bdrv_del_child(parent_bs
, child
, errp
);
5407 bool bdrv_can_store_new_dirty_bitmap(BlockDriverState
*bs
, const char *name
,
5408 uint32_t granularity
, Error
**errp
)
5410 BlockDriver
*drv
= bs
->drv
;
5413 error_setg_errno(errp
, ENOMEDIUM
,
5414 "Can't store persistent bitmaps to %s",
5415 bdrv_get_device_or_node_name(bs
));
5419 if (!drv
->bdrv_can_store_new_dirty_bitmap
) {
5420 error_setg_errno(errp
, ENOTSUP
,
5421 "Can't store persistent bitmaps to %s",
5422 bdrv_get_device_or_node_name(bs
));
5426 return drv
->bdrv_can_store_new_dirty_bitmap(bs
, name
, granularity
, errp
);