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/fuse.h"
30 #include "block/nbd.h"
31 #include "block/qdict.h"
32 #include "qemu/error-report.h"
33 #include "block/module_block.h"
34 #include "qemu/main-loop.h"
35 #include "qemu/module.h"
36 #include "qapi/error.h"
37 #include "qapi/qmp/qdict.h"
38 #include "qapi/qmp/qjson.h"
39 #include "qapi/qmp/qnull.h"
40 #include "qapi/qmp/qstring.h"
41 #include "qapi/qobject-output-visitor.h"
42 #include "qapi/qapi-visit-block-core.h"
43 #include "sysemu/block-backend.h"
44 #include "sysemu/sysemu.h"
45 #include "qemu/notify.h"
46 #include "qemu/option.h"
47 #include "qemu/coroutine.h"
48 #include "block/qapi.h"
49 #include "qemu/timer.h"
50 #include "qemu/cutils.h"
52 #include "qemu/transactions.h"
53 #include "block/coroutines.h"
56 #include <sys/ioctl.h>
57 #include <sys/queue.h>
67 #define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */
69 static QTAILQ_HEAD(, BlockDriverState
) graph_bdrv_states
=
70 QTAILQ_HEAD_INITIALIZER(graph_bdrv_states
);
72 static QTAILQ_HEAD(, BlockDriverState
) all_bdrv_states
=
73 QTAILQ_HEAD_INITIALIZER(all_bdrv_states
);
75 static QLIST_HEAD(, BlockDriver
) bdrv_drivers
=
76 QLIST_HEAD_INITIALIZER(bdrv_drivers
);
78 static BlockDriverState
*bdrv_open_inherit(const char *filename
,
79 const char *reference
,
80 QDict
*options
, int flags
,
81 BlockDriverState
*parent
,
82 const BdrvChildClass
*child_class
,
83 BdrvChildRole child_role
,
86 static void bdrv_replace_child_noperm(BdrvChild
*child
,
87 BlockDriverState
*new_bs
);
88 static int bdrv_attach_child_noperm(BlockDriverState
*parent_bs
,
89 BlockDriverState
*child_bs
,
90 const char *child_name
,
91 const BdrvChildClass
*child_class
,
92 BdrvChildRole child_role
,
97 static int bdrv_reopen_prepare(BDRVReopenState
*reopen_state
, BlockReopenQueue
98 *queue
, Error
**errp
);
99 static void bdrv_reopen_commit(BDRVReopenState
*reopen_state
);
100 static void bdrv_reopen_abort(BDRVReopenState
*reopen_state
);
102 /* If non-zero, use only whitelisted block drivers */
103 static int use_bdrv_whitelist
;
106 static int is_windows_drive_prefix(const char *filename
)
108 return (((filename
[0] >= 'a' && filename
[0] <= 'z') ||
109 (filename
[0] >= 'A' && filename
[0] <= 'Z')) &&
113 int is_windows_drive(const char *filename
)
115 if (is_windows_drive_prefix(filename
) &&
118 if (strstart(filename
, "\\\\.\\", NULL
) ||
119 strstart(filename
, "//./", NULL
))
125 size_t bdrv_opt_mem_align(BlockDriverState
*bs
)
127 if (!bs
|| !bs
->drv
) {
128 /* page size or 4k (hdd sector size) should be on the safe side */
129 return MAX(4096, qemu_real_host_page_size
);
132 return bs
->bl
.opt_mem_alignment
;
135 size_t bdrv_min_mem_align(BlockDriverState
*bs
)
137 if (!bs
|| !bs
->drv
) {
138 /* page size or 4k (hdd sector size) should be on the safe side */
139 return MAX(4096, qemu_real_host_page_size
);
142 return bs
->bl
.min_mem_alignment
;
145 /* check if the path starts with "<protocol>:" */
146 int path_has_protocol(const char *path
)
151 if (is_windows_drive(path
) ||
152 is_windows_drive_prefix(path
)) {
155 p
= path
+ strcspn(path
, ":/\\");
157 p
= path
+ strcspn(path
, ":/");
163 int path_is_absolute(const char *path
)
166 /* specific case for names like: "\\.\d:" */
167 if (is_windows_drive(path
) || is_windows_drive_prefix(path
)) {
170 return (*path
== '/' || *path
== '\\');
172 return (*path
== '/');
176 /* if filename is absolute, just return its duplicate. Otherwise, build a
177 path to it by considering it is relative to base_path. URL are
179 char *path_combine(const char *base_path
, const char *filename
)
181 const char *protocol_stripped
= NULL
;
186 if (path_is_absolute(filename
)) {
187 return g_strdup(filename
);
190 if (path_has_protocol(base_path
)) {
191 protocol_stripped
= strchr(base_path
, ':');
192 if (protocol_stripped
) {
196 p
= protocol_stripped
?: base_path
;
198 p1
= strrchr(base_path
, '/');
202 p2
= strrchr(base_path
, '\\');
203 if (!p1
|| p2
> p1
) {
218 result
= g_malloc(len
+ strlen(filename
) + 1);
219 memcpy(result
, base_path
, len
);
220 strcpy(result
+ len
, filename
);
226 * Helper function for bdrv_parse_filename() implementations to remove optional
227 * protocol prefixes (especially "file:") from a filename and for putting the
228 * stripped filename into the options QDict if there is such a prefix.
230 void bdrv_parse_filename_strip_prefix(const char *filename
, const char *prefix
,
233 if (strstart(filename
, prefix
, &filename
)) {
234 /* Stripping the explicit protocol prefix may result in a protocol
235 * prefix being (wrongly) detected (if the filename contains a colon) */
236 if (path_has_protocol(filename
)) {
237 GString
*fat_filename
;
239 /* This means there is some colon before the first slash; therefore,
240 * this cannot be an absolute path */
241 assert(!path_is_absolute(filename
));
243 /* And we can thus fix the protocol detection issue by prefixing it
245 fat_filename
= g_string_new("./");
246 g_string_append(fat_filename
, filename
);
248 assert(!path_has_protocol(fat_filename
->str
));
250 qdict_put(options
, "filename",
251 qstring_from_gstring(fat_filename
));
253 /* If no protocol prefix was detected, we can use the shortened
255 qdict_put_str(options
, "filename", filename
);
261 /* Returns whether the image file is opened as read-only. Note that this can
262 * return false and writing to the image file is still not possible because the
263 * image is inactivated. */
264 bool bdrv_is_read_only(BlockDriverState
*bs
)
266 return bs
->read_only
;
269 int bdrv_can_set_read_only(BlockDriverState
*bs
, bool read_only
,
270 bool ignore_allow_rdw
, Error
**errp
)
272 /* Do not set read_only if copy_on_read is enabled */
273 if (bs
->copy_on_read
&& read_only
) {
274 error_setg(errp
, "Can't set node '%s' to r/o with copy-on-read enabled",
275 bdrv_get_device_or_node_name(bs
));
279 /* Do not clear read_only if it is prohibited */
280 if (!read_only
&& !(bs
->open_flags
& BDRV_O_ALLOW_RDWR
) &&
283 error_setg(errp
, "Node '%s' is read only",
284 bdrv_get_device_or_node_name(bs
));
292 * Called by a driver that can only provide a read-only image.
294 * Returns 0 if the node is already read-only or it could switch the node to
295 * read-only because BDRV_O_AUTO_RDONLY is set.
297 * Returns -EACCES if the node is read-write and BDRV_O_AUTO_RDONLY is not set
298 * or bdrv_can_set_read_only() forbids making the node read-only. If @errmsg
299 * is not NULL, it is used as the error message for the Error object.
301 int bdrv_apply_auto_read_only(BlockDriverState
*bs
, const char *errmsg
,
306 if (!(bs
->open_flags
& BDRV_O_RDWR
)) {
309 if (!(bs
->open_flags
& BDRV_O_AUTO_RDONLY
)) {
313 ret
= bdrv_can_set_read_only(bs
, true, false, NULL
);
318 bs
->read_only
= true;
319 bs
->open_flags
&= ~BDRV_O_RDWR
;
324 error_setg(errp
, "%s", errmsg
?: "Image is read-only");
329 * If @backing is empty, this function returns NULL without setting
330 * @errp. In all other cases, NULL will only be returned with @errp
333 * Therefore, a return value of NULL without @errp set means that
334 * there is no backing file; if @errp is set, there is one but its
335 * absolute filename cannot be generated.
337 char *bdrv_get_full_backing_filename_from_filename(const char *backed
,
341 if (backing
[0] == '\0') {
343 } else if (path_has_protocol(backing
) || path_is_absolute(backing
)) {
344 return g_strdup(backing
);
345 } else if (backed
[0] == '\0' || strstart(backed
, "json:", NULL
)) {
346 error_setg(errp
, "Cannot use relative backing file names for '%s'",
350 return path_combine(backed
, backing
);
355 * If @filename is empty or NULL, this function returns NULL without
356 * setting @errp. In all other cases, NULL will only be returned with
359 static char *bdrv_make_absolute_filename(BlockDriverState
*relative_to
,
360 const char *filename
, Error
**errp
)
362 char *dir
, *full_name
;
364 if (!filename
|| filename
[0] == '\0') {
366 } else if (path_has_protocol(filename
) || path_is_absolute(filename
)) {
367 return g_strdup(filename
);
370 dir
= bdrv_dirname(relative_to
, errp
);
375 full_name
= g_strconcat(dir
, filename
, NULL
);
380 char *bdrv_get_full_backing_filename(BlockDriverState
*bs
, Error
**errp
)
382 return bdrv_make_absolute_filename(bs
, bs
->backing_file
, errp
);
385 void bdrv_register(BlockDriver
*bdrv
)
387 assert(bdrv
->format_name
);
388 QLIST_INSERT_HEAD(&bdrv_drivers
, bdrv
, list
);
391 BlockDriverState
*bdrv_new(void)
393 BlockDriverState
*bs
;
396 bs
= g_new0(BlockDriverState
, 1);
397 QLIST_INIT(&bs
->dirty_bitmaps
);
398 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
399 QLIST_INIT(&bs
->op_blockers
[i
]);
401 notifier_with_return_list_init(&bs
->before_write_notifiers
);
402 qemu_co_mutex_init(&bs
->reqs_lock
);
403 qemu_mutex_init(&bs
->dirty_bitmap_mutex
);
405 bs
->aio_context
= qemu_get_aio_context();
407 qemu_co_queue_init(&bs
->flush_queue
);
409 for (i
= 0; i
< bdrv_drain_all_count
; i
++) {
410 bdrv_drained_begin(bs
);
413 QTAILQ_INSERT_TAIL(&all_bdrv_states
, bs
, bs_list
);
418 static BlockDriver
*bdrv_do_find_format(const char *format_name
)
422 QLIST_FOREACH(drv1
, &bdrv_drivers
, list
) {
423 if (!strcmp(drv1
->format_name
, format_name
)) {
431 BlockDriver
*bdrv_find_format(const char *format_name
)
436 drv1
= bdrv_do_find_format(format_name
);
441 /* The driver isn't registered, maybe we need to load a module */
442 for (i
= 0; i
< (int)ARRAY_SIZE(block_driver_modules
); ++i
) {
443 if (!strcmp(block_driver_modules
[i
].format_name
, format_name
)) {
444 block_module_load_one(block_driver_modules
[i
].library_name
);
449 return bdrv_do_find_format(format_name
);
452 static int bdrv_format_is_whitelisted(const char *format_name
, bool read_only
)
454 static const char *whitelist_rw
[] = {
455 CONFIG_BDRV_RW_WHITELIST
458 static const char *whitelist_ro
[] = {
459 CONFIG_BDRV_RO_WHITELIST
464 if (!whitelist_rw
[0] && !whitelist_ro
[0]) {
465 return 1; /* no whitelist, anything goes */
468 for (p
= whitelist_rw
; *p
; p
++) {
469 if (!strcmp(format_name
, *p
)) {
474 for (p
= whitelist_ro
; *p
; p
++) {
475 if (!strcmp(format_name
, *p
)) {
483 int bdrv_is_whitelisted(BlockDriver
*drv
, bool read_only
)
485 return bdrv_format_is_whitelisted(drv
->format_name
, read_only
);
488 bool bdrv_uses_whitelist(void)
490 return use_bdrv_whitelist
;
493 typedef struct CreateCo
{
501 static void coroutine_fn
bdrv_create_co_entry(void *opaque
)
503 Error
*local_err
= NULL
;
506 CreateCo
*cco
= opaque
;
509 ret
= cco
->drv
->bdrv_co_create_opts(cco
->drv
,
510 cco
->filename
, cco
->opts
, &local_err
);
511 error_propagate(&cco
->err
, local_err
);
515 int bdrv_create(BlockDriver
*drv
, const char* filename
,
516 QemuOpts
*opts
, Error
**errp
)
523 .filename
= g_strdup(filename
),
529 if (!drv
->bdrv_co_create_opts
) {
530 error_setg(errp
, "Driver '%s' does not support image creation", drv
->format_name
);
535 if (qemu_in_coroutine()) {
536 /* Fast-path if already in coroutine context */
537 bdrv_create_co_entry(&cco
);
539 co
= qemu_coroutine_create(bdrv_create_co_entry
, &cco
);
540 qemu_coroutine_enter(co
);
541 while (cco
.ret
== NOT_DONE
) {
542 aio_poll(qemu_get_aio_context(), true);
549 error_propagate(errp
, cco
.err
);
551 error_setg_errno(errp
, -ret
, "Could not create image");
556 g_free(cco
.filename
);
561 * Helper function for bdrv_create_file_fallback(): Resize @blk to at
562 * least the given @minimum_size.
564 * On success, return @blk's actual length.
565 * Otherwise, return -errno.
567 static int64_t create_file_fallback_truncate(BlockBackend
*blk
,
568 int64_t minimum_size
, Error
**errp
)
570 Error
*local_err
= NULL
;
574 ret
= blk_truncate(blk
, minimum_size
, false, PREALLOC_MODE_OFF
, 0,
576 if (ret
< 0 && ret
!= -ENOTSUP
) {
577 error_propagate(errp
, local_err
);
581 size
= blk_getlength(blk
);
583 error_free(local_err
);
584 error_setg_errno(errp
, -size
,
585 "Failed to inquire the new image file's length");
589 if (size
< minimum_size
) {
590 /* Need to grow the image, but we failed to do that */
591 error_propagate(errp
, local_err
);
595 error_free(local_err
);
602 * Helper function for bdrv_create_file_fallback(): Zero the first
603 * sector to remove any potentially pre-existing image header.
605 static int create_file_fallback_zero_first_sector(BlockBackend
*blk
,
606 int64_t current_size
,
609 int64_t bytes_to_clear
;
612 bytes_to_clear
= MIN(current_size
, BDRV_SECTOR_SIZE
);
613 if (bytes_to_clear
) {
614 ret
= blk_pwrite_zeroes(blk
, 0, bytes_to_clear
, BDRV_REQ_MAY_UNMAP
);
616 error_setg_errno(errp
, -ret
,
617 "Failed to clear the new image's first sector");
626 * Simple implementation of bdrv_co_create_opts for protocol drivers
627 * which only support creation via opening a file
628 * (usually existing raw storage device)
630 int coroutine_fn
bdrv_co_create_opts_simple(BlockDriver
*drv
,
631 const char *filename
,
639 PreallocMode prealloc
;
640 Error
*local_err
= NULL
;
643 size
= qemu_opt_get_size_del(opts
, BLOCK_OPT_SIZE
, 0);
644 buf
= qemu_opt_get_del(opts
, BLOCK_OPT_PREALLOC
);
645 prealloc
= qapi_enum_parse(&PreallocMode_lookup
, buf
,
646 PREALLOC_MODE_OFF
, &local_err
);
649 error_propagate(errp
, local_err
);
653 if (prealloc
!= PREALLOC_MODE_OFF
) {
654 error_setg(errp
, "Unsupported preallocation mode '%s'",
655 PreallocMode_str(prealloc
));
659 options
= qdict_new();
660 qdict_put_str(options
, "driver", drv
->format_name
);
662 blk
= blk_new_open(filename
, NULL
, options
,
663 BDRV_O_RDWR
| BDRV_O_RESIZE
, errp
);
665 error_prepend(errp
, "Protocol driver '%s' does not support image "
666 "creation, and opening the image failed: ",
671 size
= create_file_fallback_truncate(blk
, size
, errp
);
677 ret
= create_file_fallback_zero_first_sector(blk
, size
, errp
);
688 int bdrv_create_file(const char *filename
, QemuOpts
*opts
, Error
**errp
)
690 QemuOpts
*protocol_opts
;
695 drv
= bdrv_find_protocol(filename
, true, errp
);
700 if (!drv
->create_opts
) {
701 error_setg(errp
, "Driver '%s' does not support image creation",
707 * 'opts' contains a QemuOptsList with a combination of format and protocol
710 * The format properly removes its options, but the default values remain
711 * in 'opts->list'. So if the protocol has options with the same name
712 * (e.g. rbd has 'cluster_size' as qcow2), it will see the default values
713 * of the format, since for overlapping options, the format wins.
715 * To avoid this issue, lets convert QemuOpts to QDict, in this way we take
716 * only the set options, and then convert it back to QemuOpts, using the
717 * create_opts of the protocol. So the new QemuOpts, will contain only the
720 qdict
= qemu_opts_to_qdict(opts
, NULL
);
721 protocol_opts
= qemu_opts_from_qdict(drv
->create_opts
, qdict
, errp
);
722 if (protocol_opts
== NULL
) {
727 ret
= bdrv_create(drv
, filename
, protocol_opts
, errp
);
729 qemu_opts_del(protocol_opts
);
730 qobject_unref(qdict
);
734 int coroutine_fn
bdrv_co_delete_file(BlockDriverState
*bs
, Error
**errp
)
736 Error
*local_err
= NULL
;
742 error_setg(errp
, "Block node '%s' is not opened", bs
->filename
);
746 if (!bs
->drv
->bdrv_co_delete_file
) {
747 error_setg(errp
, "Driver '%s' does not support image deletion",
748 bs
->drv
->format_name
);
752 ret
= bs
->drv
->bdrv_co_delete_file(bs
, &local_err
);
754 error_propagate(errp
, local_err
);
760 void coroutine_fn
bdrv_co_delete_file_noerr(BlockDriverState
*bs
)
762 Error
*local_err
= NULL
;
769 ret
= bdrv_co_delete_file(bs
, &local_err
);
771 * ENOTSUP will happen if the block driver doesn't support
772 * the 'bdrv_co_delete_file' interface. This is a predictable
773 * scenario and shouldn't be reported back to the user.
775 if (ret
== -ENOTSUP
) {
776 error_free(local_err
);
777 } else if (ret
< 0) {
778 error_report_err(local_err
);
783 * Try to get @bs's logical and physical block size.
784 * On success, store them in @bsz struct and return 0.
785 * On failure return -errno.
786 * @bs must not be empty.
788 int bdrv_probe_blocksizes(BlockDriverState
*bs
, BlockSizes
*bsz
)
790 BlockDriver
*drv
= bs
->drv
;
791 BlockDriverState
*filtered
= bdrv_filter_bs(bs
);
793 if (drv
&& drv
->bdrv_probe_blocksizes
) {
794 return drv
->bdrv_probe_blocksizes(bs
, bsz
);
795 } else if (filtered
) {
796 return bdrv_probe_blocksizes(filtered
, bsz
);
803 * Try to get @bs's geometry (cyls, heads, sectors).
804 * On success, store them in @geo struct and return 0.
805 * On failure return -errno.
806 * @bs must not be empty.
808 int bdrv_probe_geometry(BlockDriverState
*bs
, HDGeometry
*geo
)
810 BlockDriver
*drv
= bs
->drv
;
811 BlockDriverState
*filtered
= bdrv_filter_bs(bs
);
813 if (drv
&& drv
->bdrv_probe_geometry
) {
814 return drv
->bdrv_probe_geometry(bs
, geo
);
815 } else if (filtered
) {
816 return bdrv_probe_geometry(filtered
, geo
);
823 * Create a uniquely-named empty temporary file.
824 * Return 0 upon success, otherwise a negative errno value.
826 int get_tmp_filename(char *filename
, int size
)
829 char temp_dir
[MAX_PATH
];
830 /* GetTempFileName requires that its output buffer (4th param)
831 have length MAX_PATH or greater. */
832 assert(size
>= MAX_PATH
);
833 return (GetTempPath(MAX_PATH
, temp_dir
)
834 && GetTempFileName(temp_dir
, "qem", 0, filename
)
835 ? 0 : -GetLastError());
839 tmpdir
= getenv("TMPDIR");
843 if (snprintf(filename
, size
, "%s/vl.XXXXXX", tmpdir
) >= size
) {
846 fd
= mkstemp(filename
);
850 if (close(fd
) != 0) {
859 * Detect host devices. By convention, /dev/cdrom[N] is always
860 * recognized as a host CDROM.
862 static BlockDriver
*find_hdev_driver(const char *filename
)
864 int score_max
= 0, score
;
865 BlockDriver
*drv
= NULL
, *d
;
867 QLIST_FOREACH(d
, &bdrv_drivers
, list
) {
868 if (d
->bdrv_probe_device
) {
869 score
= d
->bdrv_probe_device(filename
);
870 if (score
> score_max
) {
880 static BlockDriver
*bdrv_do_find_protocol(const char *protocol
)
884 QLIST_FOREACH(drv1
, &bdrv_drivers
, list
) {
885 if (drv1
->protocol_name
&& !strcmp(drv1
->protocol_name
, protocol
)) {
893 BlockDriver
*bdrv_find_protocol(const char *filename
,
894 bool allow_protocol_prefix
,
903 /* TODO Drivers without bdrv_file_open must be specified explicitly */
906 * XXX(hch): we really should not let host device detection
907 * override an explicit protocol specification, but moving this
908 * later breaks access to device names with colons in them.
909 * Thanks to the brain-dead persistent naming schemes on udev-
910 * based Linux systems those actually are quite common.
912 drv1
= find_hdev_driver(filename
);
917 if (!path_has_protocol(filename
) || !allow_protocol_prefix
) {
921 p
= strchr(filename
, ':');
924 if (len
> sizeof(protocol
) - 1)
925 len
= sizeof(protocol
) - 1;
926 memcpy(protocol
, filename
, len
);
927 protocol
[len
] = '\0';
929 drv1
= bdrv_do_find_protocol(protocol
);
934 for (i
= 0; i
< (int)ARRAY_SIZE(block_driver_modules
); ++i
) {
935 if (block_driver_modules
[i
].protocol_name
&&
936 !strcmp(block_driver_modules
[i
].protocol_name
, protocol
)) {
937 block_module_load_one(block_driver_modules
[i
].library_name
);
942 drv1
= bdrv_do_find_protocol(protocol
);
944 error_setg(errp
, "Unknown protocol '%s'", protocol
);
950 * Guess image format by probing its contents.
951 * This is not a good idea when your image is raw (CVE-2008-2004), but
952 * we do it anyway for backward compatibility.
954 * @buf contains the image's first @buf_size bytes.
955 * @buf_size is the buffer size in bytes (generally BLOCK_PROBE_BUF_SIZE,
956 * but can be smaller if the image file is smaller)
957 * @filename is its filename.
959 * For all block drivers, call the bdrv_probe() method to get its
961 * Return the first block driver with the highest probing score.
963 BlockDriver
*bdrv_probe_all(const uint8_t *buf
, int buf_size
,
964 const char *filename
)
966 int score_max
= 0, score
;
967 BlockDriver
*drv
= NULL
, *d
;
969 QLIST_FOREACH(d
, &bdrv_drivers
, list
) {
971 score
= d
->bdrv_probe(buf
, buf_size
, filename
);
972 if (score
> score_max
) {
982 static int find_image_format(BlockBackend
*file
, const char *filename
,
983 BlockDriver
**pdrv
, Error
**errp
)
986 uint8_t buf
[BLOCK_PROBE_BUF_SIZE
];
989 /* Return the raw BlockDriver * to scsi-generic devices or empty drives */
990 if (blk_is_sg(file
) || !blk_is_inserted(file
) || blk_getlength(file
) == 0) {
995 ret
= blk_pread(file
, 0, buf
, sizeof(buf
));
997 error_setg_errno(errp
, -ret
, "Could not read image for determining its "
1003 drv
= bdrv_probe_all(buf
, ret
, filename
);
1005 error_setg(errp
, "Could not determine image format: No compatible "
1014 * Set the current 'total_sectors' value
1015 * Return 0 on success, -errno on error.
1017 int refresh_total_sectors(BlockDriverState
*bs
, int64_t hint
)
1019 BlockDriver
*drv
= bs
->drv
;
1025 /* Do not attempt drv->bdrv_getlength() on scsi-generic devices */
1029 /* query actual device if possible, otherwise just trust the hint */
1030 if (drv
->bdrv_getlength
) {
1031 int64_t length
= drv
->bdrv_getlength(bs
);
1035 hint
= DIV_ROUND_UP(length
, BDRV_SECTOR_SIZE
);
1038 bs
->total_sectors
= hint
;
1040 if (bs
->total_sectors
* BDRV_SECTOR_SIZE
> BDRV_MAX_LENGTH
) {
1048 * Combines a QDict of new block driver @options with any missing options taken
1049 * from @old_options, so that leaving out an option defaults to its old value.
1051 static void bdrv_join_options(BlockDriverState
*bs
, QDict
*options
,
1054 if (bs
->drv
&& bs
->drv
->bdrv_join_options
) {
1055 bs
->drv
->bdrv_join_options(options
, old_options
);
1057 qdict_join(options
, old_options
, false);
1061 static BlockdevDetectZeroesOptions
bdrv_parse_detect_zeroes(QemuOpts
*opts
,
1065 Error
*local_err
= NULL
;
1066 char *value
= qemu_opt_get_del(opts
, "detect-zeroes");
1067 BlockdevDetectZeroesOptions detect_zeroes
=
1068 qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup
, value
,
1069 BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF
, &local_err
);
1072 error_propagate(errp
, local_err
);
1073 return detect_zeroes
;
1076 if (detect_zeroes
== BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP
&&
1077 !(open_flags
& BDRV_O_UNMAP
))
1079 error_setg(errp
, "setting detect-zeroes to unmap is not allowed "
1080 "without setting discard operation to unmap");
1083 return detect_zeroes
;
1087 * Set open flags for aio engine
1089 * Return 0 on success, -1 if the engine specified is invalid
1091 int bdrv_parse_aio(const char *mode
, int *flags
)
1093 if (!strcmp(mode
, "threads")) {
1094 /* do nothing, default */
1095 } else if (!strcmp(mode
, "native")) {
1096 *flags
|= BDRV_O_NATIVE_AIO
;
1097 #ifdef CONFIG_LINUX_IO_URING
1098 } else if (!strcmp(mode
, "io_uring")) {
1099 *flags
|= BDRV_O_IO_URING
;
1109 * Set open flags for a given discard mode
1111 * Return 0 on success, -1 if the discard mode was invalid.
1113 int bdrv_parse_discard_flags(const char *mode
, int *flags
)
1115 *flags
&= ~BDRV_O_UNMAP
;
1117 if (!strcmp(mode
, "off") || !strcmp(mode
, "ignore")) {
1119 } else if (!strcmp(mode
, "on") || !strcmp(mode
, "unmap")) {
1120 *flags
|= BDRV_O_UNMAP
;
1129 * Set open flags for a given cache mode
1131 * Return 0 on success, -1 if the cache mode was invalid.
1133 int bdrv_parse_cache_mode(const char *mode
, int *flags
, bool *writethrough
)
1135 *flags
&= ~BDRV_O_CACHE_MASK
;
1137 if (!strcmp(mode
, "off") || !strcmp(mode
, "none")) {
1138 *writethrough
= false;
1139 *flags
|= BDRV_O_NOCACHE
;
1140 } else if (!strcmp(mode
, "directsync")) {
1141 *writethrough
= true;
1142 *flags
|= BDRV_O_NOCACHE
;
1143 } else if (!strcmp(mode
, "writeback")) {
1144 *writethrough
= false;
1145 } else if (!strcmp(mode
, "unsafe")) {
1146 *writethrough
= false;
1147 *flags
|= BDRV_O_NO_FLUSH
;
1148 } else if (!strcmp(mode
, "writethrough")) {
1149 *writethrough
= true;
1157 static char *bdrv_child_get_parent_desc(BdrvChild
*c
)
1159 BlockDriverState
*parent
= c
->opaque
;
1160 return g_strdup(bdrv_get_device_or_node_name(parent
));
1163 static void bdrv_child_cb_drained_begin(BdrvChild
*child
)
1165 BlockDriverState
*bs
= child
->opaque
;
1166 bdrv_do_drained_begin_quiesce(bs
, NULL
, false);
1169 static bool bdrv_child_cb_drained_poll(BdrvChild
*child
)
1171 BlockDriverState
*bs
= child
->opaque
;
1172 return bdrv_drain_poll(bs
, false, NULL
, false);
1175 static void bdrv_child_cb_drained_end(BdrvChild
*child
,
1176 int *drained_end_counter
)
1178 BlockDriverState
*bs
= child
->opaque
;
1179 bdrv_drained_end_no_poll(bs
, drained_end_counter
);
1182 static int bdrv_child_cb_inactivate(BdrvChild
*child
)
1184 BlockDriverState
*bs
= child
->opaque
;
1185 assert(bs
->open_flags
& BDRV_O_INACTIVE
);
1189 static bool bdrv_child_cb_can_set_aio_ctx(BdrvChild
*child
, AioContext
*ctx
,
1190 GSList
**ignore
, Error
**errp
)
1192 BlockDriverState
*bs
= child
->opaque
;
1193 return bdrv_can_set_aio_context(bs
, ctx
, ignore
, errp
);
1196 static void bdrv_child_cb_set_aio_ctx(BdrvChild
*child
, AioContext
*ctx
,
1199 BlockDriverState
*bs
= child
->opaque
;
1200 return bdrv_set_aio_context_ignore(bs
, ctx
, ignore
);
1204 * Returns the options and flags that a temporary snapshot should get, based on
1205 * the originally requested flags (the originally requested image will have
1206 * flags like a backing file)
1208 static void bdrv_temp_snapshot_options(int *child_flags
, QDict
*child_options
,
1209 int parent_flags
, QDict
*parent_options
)
1211 *child_flags
= (parent_flags
& ~BDRV_O_SNAPSHOT
) | BDRV_O_TEMPORARY
;
1213 /* For temporary files, unconditional cache=unsafe is fine */
1214 qdict_set_default_str(child_options
, BDRV_OPT_CACHE_DIRECT
, "off");
1215 qdict_set_default_str(child_options
, BDRV_OPT_CACHE_NO_FLUSH
, "on");
1217 /* Copy the read-only and discard options from the parent */
1218 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_READ_ONLY
);
1219 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_DISCARD
);
1221 /* aio=native doesn't work for cache.direct=off, so disable it for the
1222 * temporary snapshot */
1223 *child_flags
&= ~BDRV_O_NATIVE_AIO
;
1226 static void bdrv_backing_attach(BdrvChild
*c
)
1228 BlockDriverState
*parent
= c
->opaque
;
1229 BlockDriverState
*backing_hd
= c
->bs
;
1231 assert(!parent
->backing_blocker
);
1232 error_setg(&parent
->backing_blocker
,
1233 "node is used as backing hd of '%s'",
1234 bdrv_get_device_or_node_name(parent
));
1236 bdrv_refresh_filename(backing_hd
);
1238 parent
->open_flags
&= ~BDRV_O_NO_BACKING
;
1240 bdrv_op_block_all(backing_hd
, parent
->backing_blocker
);
1241 /* Otherwise we won't be able to commit or stream */
1242 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_COMMIT_TARGET
,
1243 parent
->backing_blocker
);
1244 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_STREAM
,
1245 parent
->backing_blocker
);
1247 * We do backup in 3 ways:
1249 * The target bs is new opened, and the source is top BDS
1250 * 2. blockdev backup
1251 * Both the source and the target are top BDSes.
1252 * 3. internal backup(used for block replication)
1253 * Both the source and the target are backing file
1255 * In case 1 and 2, neither the source nor the target is the backing file.
1256 * In case 3, we will block the top BDS, so there is only one block job
1257 * for the top BDS and its backing chain.
1259 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_BACKUP_SOURCE
,
1260 parent
->backing_blocker
);
1261 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_BACKUP_TARGET
,
1262 parent
->backing_blocker
);
1265 static void bdrv_backing_detach(BdrvChild
*c
)
1267 BlockDriverState
*parent
= c
->opaque
;
1269 assert(parent
->backing_blocker
);
1270 bdrv_op_unblock_all(c
->bs
, parent
->backing_blocker
);
1271 error_free(parent
->backing_blocker
);
1272 parent
->backing_blocker
= NULL
;
1275 static int bdrv_backing_update_filename(BdrvChild
*c
, BlockDriverState
*base
,
1276 const char *filename
, Error
**errp
)
1278 BlockDriverState
*parent
= c
->opaque
;
1279 bool read_only
= bdrv_is_read_only(parent
);
1283 ret
= bdrv_reopen_set_read_only(parent
, false, errp
);
1289 ret
= bdrv_change_backing_file(parent
, filename
,
1290 base
->drv
? base
->drv
->format_name
: "",
1293 error_setg_errno(errp
, -ret
, "Could not update backing file link");
1297 bdrv_reopen_set_read_only(parent
, true, NULL
);
1304 * Returns the options and flags that a generic child of a BDS should
1305 * get, based on the given options and flags for the parent BDS.
1307 static void bdrv_inherited_options(BdrvChildRole role
, bool parent_is_format
,
1308 int *child_flags
, QDict
*child_options
,
1309 int parent_flags
, QDict
*parent_options
)
1311 int flags
= parent_flags
;
1314 * First, decide whether to set, clear, or leave BDRV_O_PROTOCOL.
1315 * Generally, the question to answer is: Should this child be
1316 * format-probed by default?
1320 * Pure and non-filtered data children of non-format nodes should
1321 * be probed by default (even when the node itself has BDRV_O_PROTOCOL
1322 * set). This only affects a very limited set of drivers (namely
1323 * quorum and blkverify when this comment was written).
1324 * Force-clear BDRV_O_PROTOCOL then.
1326 if (!parent_is_format
&&
1327 (role
& BDRV_CHILD_DATA
) &&
1328 !(role
& (BDRV_CHILD_METADATA
| BDRV_CHILD_FILTERED
)))
1330 flags
&= ~BDRV_O_PROTOCOL
;
1334 * All children of format nodes (except for COW children) and all
1335 * metadata children in general should never be format-probed.
1336 * Force-set BDRV_O_PROTOCOL then.
1338 if ((parent_is_format
&& !(role
& BDRV_CHILD_COW
)) ||
1339 (role
& BDRV_CHILD_METADATA
))
1341 flags
|= BDRV_O_PROTOCOL
;
1345 * If the cache mode isn't explicitly set, inherit direct and no-flush from
1348 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_CACHE_DIRECT
);
1349 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_CACHE_NO_FLUSH
);
1350 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_FORCE_SHARE
);
1352 if (role
& BDRV_CHILD_COW
) {
1353 /* backing files are opened read-only by default */
1354 qdict_set_default_str(child_options
, BDRV_OPT_READ_ONLY
, "on");
1355 qdict_set_default_str(child_options
, BDRV_OPT_AUTO_READ_ONLY
, "off");
1357 /* Inherit the read-only option from the parent if it's not set */
1358 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_READ_ONLY
);
1359 qdict_copy_default(child_options
, parent_options
,
1360 BDRV_OPT_AUTO_READ_ONLY
);
1364 * bdrv_co_pdiscard() respects unmap policy for the parent, so we
1365 * can default to enable it on lower layers regardless of the
1368 qdict_set_default_str(child_options
, BDRV_OPT_DISCARD
, "unmap");
1370 /* Clear flags that only apply to the top layer */
1371 flags
&= ~(BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
| BDRV_O_COPY_ON_READ
);
1373 if (role
& BDRV_CHILD_METADATA
) {
1374 flags
&= ~BDRV_O_NO_IO
;
1376 if (role
& BDRV_CHILD_COW
) {
1377 flags
&= ~BDRV_O_TEMPORARY
;
1380 *child_flags
= flags
;
1383 static void bdrv_child_cb_attach(BdrvChild
*child
)
1385 BlockDriverState
*bs
= child
->opaque
;
1387 if (child
->role
& BDRV_CHILD_COW
) {
1388 bdrv_backing_attach(child
);
1391 bdrv_apply_subtree_drain(child
, bs
);
1394 static void bdrv_child_cb_detach(BdrvChild
*child
)
1396 BlockDriverState
*bs
= child
->opaque
;
1398 if (child
->role
& BDRV_CHILD_COW
) {
1399 bdrv_backing_detach(child
);
1402 bdrv_unapply_subtree_drain(child
, bs
);
1405 static int bdrv_child_cb_update_filename(BdrvChild
*c
, BlockDriverState
*base
,
1406 const char *filename
, Error
**errp
)
1408 if (c
->role
& BDRV_CHILD_COW
) {
1409 return bdrv_backing_update_filename(c
, base
, filename
, errp
);
1414 static AioContext
*bdrv_child_cb_get_parent_aio_context(BdrvChild
*c
)
1416 BlockDriverState
*bs
= c
->opaque
;
1418 return bdrv_get_aio_context(bs
);
1421 const BdrvChildClass child_of_bds
= {
1422 .parent_is_bds
= true,
1423 .get_parent_desc
= bdrv_child_get_parent_desc
,
1424 .inherit_options
= bdrv_inherited_options
,
1425 .drained_begin
= bdrv_child_cb_drained_begin
,
1426 .drained_poll
= bdrv_child_cb_drained_poll
,
1427 .drained_end
= bdrv_child_cb_drained_end
,
1428 .attach
= bdrv_child_cb_attach
,
1429 .detach
= bdrv_child_cb_detach
,
1430 .inactivate
= bdrv_child_cb_inactivate
,
1431 .can_set_aio_ctx
= bdrv_child_cb_can_set_aio_ctx
,
1432 .set_aio_ctx
= bdrv_child_cb_set_aio_ctx
,
1433 .update_filename
= bdrv_child_cb_update_filename
,
1434 .get_parent_aio_context
= bdrv_child_cb_get_parent_aio_context
,
1437 AioContext
*bdrv_child_get_parent_aio_context(BdrvChild
*c
)
1439 return c
->klass
->get_parent_aio_context(c
);
1442 static int bdrv_open_flags(BlockDriverState
*bs
, int flags
)
1444 int open_flags
= flags
;
1447 * Clear flags that are internal to the block layer before opening the
1450 open_flags
&= ~(BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
| BDRV_O_PROTOCOL
);
1455 static void update_flags_from_options(int *flags
, QemuOpts
*opts
)
1457 *flags
&= ~(BDRV_O_CACHE_MASK
| BDRV_O_RDWR
| BDRV_O_AUTO_RDONLY
);
1459 if (qemu_opt_get_bool_del(opts
, BDRV_OPT_CACHE_NO_FLUSH
, false)) {
1460 *flags
|= BDRV_O_NO_FLUSH
;
1463 if (qemu_opt_get_bool_del(opts
, BDRV_OPT_CACHE_DIRECT
, false)) {
1464 *flags
|= BDRV_O_NOCACHE
;
1467 if (!qemu_opt_get_bool_del(opts
, BDRV_OPT_READ_ONLY
, false)) {
1468 *flags
|= BDRV_O_RDWR
;
1471 if (qemu_opt_get_bool_del(opts
, BDRV_OPT_AUTO_READ_ONLY
, false)) {
1472 *flags
|= BDRV_O_AUTO_RDONLY
;
1476 static void update_options_from_flags(QDict
*options
, int flags
)
1478 if (!qdict_haskey(options
, BDRV_OPT_CACHE_DIRECT
)) {
1479 qdict_put_bool(options
, BDRV_OPT_CACHE_DIRECT
, flags
& BDRV_O_NOCACHE
);
1481 if (!qdict_haskey(options
, BDRV_OPT_CACHE_NO_FLUSH
)) {
1482 qdict_put_bool(options
, BDRV_OPT_CACHE_NO_FLUSH
,
1483 flags
& BDRV_O_NO_FLUSH
);
1485 if (!qdict_haskey(options
, BDRV_OPT_READ_ONLY
)) {
1486 qdict_put_bool(options
, BDRV_OPT_READ_ONLY
, !(flags
& BDRV_O_RDWR
));
1488 if (!qdict_haskey(options
, BDRV_OPT_AUTO_READ_ONLY
)) {
1489 qdict_put_bool(options
, BDRV_OPT_AUTO_READ_ONLY
,
1490 flags
& BDRV_O_AUTO_RDONLY
);
1494 static void bdrv_assign_node_name(BlockDriverState
*bs
,
1495 const char *node_name
,
1498 char *gen_node_name
= NULL
;
1501 node_name
= gen_node_name
= id_generate(ID_BLOCK
);
1502 } else if (!id_wellformed(node_name
)) {
1504 * Check for empty string or invalid characters, but not if it is
1505 * generated (generated names use characters not available to the user)
1507 error_setg(errp
, "Invalid node-name: '%s'", node_name
);
1511 /* takes care of avoiding namespaces collisions */
1512 if (blk_by_name(node_name
)) {
1513 error_setg(errp
, "node-name=%s is conflicting with a device id",
1518 /* takes care of avoiding duplicates node names */
1519 if (bdrv_find_node(node_name
)) {
1520 error_setg(errp
, "Duplicate nodes with node-name='%s'", node_name
);
1524 /* Make sure that the node name isn't truncated */
1525 if (strlen(node_name
) >= sizeof(bs
->node_name
)) {
1526 error_setg(errp
, "Node name too long");
1530 /* copy node name into the bs and insert it into the graph list */
1531 pstrcpy(bs
->node_name
, sizeof(bs
->node_name
), node_name
);
1532 QTAILQ_INSERT_TAIL(&graph_bdrv_states
, bs
, node_list
);
1534 g_free(gen_node_name
);
1537 static int bdrv_open_driver(BlockDriverState
*bs
, BlockDriver
*drv
,
1538 const char *node_name
, QDict
*options
,
1539 int open_flags
, Error
**errp
)
1541 Error
*local_err
= NULL
;
1544 bdrv_assign_node_name(bs
, node_name
, &local_err
);
1546 error_propagate(errp
, local_err
);
1551 bs
->read_only
= !(bs
->open_flags
& BDRV_O_RDWR
);
1552 bs
->opaque
= g_malloc0(drv
->instance_size
);
1554 if (drv
->bdrv_file_open
) {
1555 assert(!drv
->bdrv_needs_filename
|| bs
->filename
[0]);
1556 ret
= drv
->bdrv_file_open(bs
, options
, open_flags
, &local_err
);
1557 } else if (drv
->bdrv_open
) {
1558 ret
= drv
->bdrv_open(bs
, options
, open_flags
, &local_err
);
1565 error_propagate(errp
, local_err
);
1566 } else if (bs
->filename
[0]) {
1567 error_setg_errno(errp
, -ret
, "Could not open '%s'", bs
->filename
);
1569 error_setg_errno(errp
, -ret
, "Could not open image");
1574 ret
= refresh_total_sectors(bs
, bs
->total_sectors
);
1576 error_setg_errno(errp
, -ret
, "Could not refresh total sector count");
1580 bdrv_refresh_limits(bs
, &local_err
);
1582 error_propagate(errp
, local_err
);
1586 assert(bdrv_opt_mem_align(bs
) != 0);
1587 assert(bdrv_min_mem_align(bs
) != 0);
1588 assert(is_power_of_2(bs
->bl
.request_alignment
));
1590 for (i
= 0; i
< bs
->quiesce_counter
; i
++) {
1591 if (drv
->bdrv_co_drain_begin
) {
1592 drv
->bdrv_co_drain_begin(bs
);
1599 if (bs
->file
!= NULL
) {
1600 bdrv_unref_child(bs
, bs
->file
);
1608 BlockDriverState
*bdrv_new_open_driver(BlockDriver
*drv
, const char *node_name
,
1609 int flags
, Error
**errp
)
1611 BlockDriverState
*bs
;
1615 bs
->open_flags
= flags
;
1616 bs
->explicit_options
= qdict_new();
1617 bs
->options
= qdict_new();
1620 update_options_from_flags(bs
->options
, flags
);
1622 ret
= bdrv_open_driver(bs
, drv
, node_name
, bs
->options
, flags
, errp
);
1624 qobject_unref(bs
->explicit_options
);
1625 bs
->explicit_options
= NULL
;
1626 qobject_unref(bs
->options
);
1635 QemuOptsList bdrv_runtime_opts
= {
1636 .name
= "bdrv_common",
1637 .head
= QTAILQ_HEAD_INITIALIZER(bdrv_runtime_opts
.head
),
1640 .name
= "node-name",
1641 .type
= QEMU_OPT_STRING
,
1642 .help
= "Node name of the block device node",
1646 .type
= QEMU_OPT_STRING
,
1647 .help
= "Block driver to use for the node",
1650 .name
= BDRV_OPT_CACHE_DIRECT
,
1651 .type
= QEMU_OPT_BOOL
,
1652 .help
= "Bypass software writeback cache on the host",
1655 .name
= BDRV_OPT_CACHE_NO_FLUSH
,
1656 .type
= QEMU_OPT_BOOL
,
1657 .help
= "Ignore flush requests",
1660 .name
= BDRV_OPT_READ_ONLY
,
1661 .type
= QEMU_OPT_BOOL
,
1662 .help
= "Node is opened in read-only mode",
1665 .name
= BDRV_OPT_AUTO_READ_ONLY
,
1666 .type
= QEMU_OPT_BOOL
,
1667 .help
= "Node can become read-only if opening read-write fails",
1670 .name
= "detect-zeroes",
1671 .type
= QEMU_OPT_STRING
,
1672 .help
= "try to optimize zero writes (off, on, unmap)",
1675 .name
= BDRV_OPT_DISCARD
,
1676 .type
= QEMU_OPT_STRING
,
1677 .help
= "discard operation (ignore/off, unmap/on)",
1680 .name
= BDRV_OPT_FORCE_SHARE
,
1681 .type
= QEMU_OPT_BOOL
,
1682 .help
= "always accept other writers (default: off)",
1684 { /* end of list */ }
1688 QemuOptsList bdrv_create_opts_simple
= {
1689 .name
= "simple-create-opts",
1690 .head
= QTAILQ_HEAD_INITIALIZER(bdrv_create_opts_simple
.head
),
1693 .name
= BLOCK_OPT_SIZE
,
1694 .type
= QEMU_OPT_SIZE
,
1695 .help
= "Virtual disk size"
1698 .name
= BLOCK_OPT_PREALLOC
,
1699 .type
= QEMU_OPT_STRING
,
1700 .help
= "Preallocation mode (allowed values: off)"
1702 { /* end of list */ }
1707 * Common part for opening disk images and files
1709 * Removes all processed options from *options.
1711 static int bdrv_open_common(BlockDriverState
*bs
, BlockBackend
*file
,
1712 QDict
*options
, Error
**errp
)
1714 int ret
, open_flags
;
1715 const char *filename
;
1716 const char *driver_name
= NULL
;
1717 const char *node_name
= NULL
;
1718 const char *discard
;
1721 Error
*local_err
= NULL
;
1723 assert(bs
->file
== NULL
);
1724 assert(options
!= NULL
&& bs
->options
!= options
);
1726 opts
= qemu_opts_create(&bdrv_runtime_opts
, NULL
, 0, &error_abort
);
1727 if (!qemu_opts_absorb_qdict(opts
, options
, errp
)) {
1732 update_flags_from_options(&bs
->open_flags
, opts
);
1734 driver_name
= qemu_opt_get(opts
, "driver");
1735 drv
= bdrv_find_format(driver_name
);
1736 assert(drv
!= NULL
);
1738 bs
->force_share
= qemu_opt_get_bool(opts
, BDRV_OPT_FORCE_SHARE
, false);
1740 if (bs
->force_share
&& (bs
->open_flags
& BDRV_O_RDWR
)) {
1742 BDRV_OPT_FORCE_SHARE
1743 "=on can only be used with read-only images");
1749 bdrv_refresh_filename(blk_bs(file
));
1750 filename
= blk_bs(file
)->filename
;
1753 * Caution: while qdict_get_try_str() is fine, getting
1754 * non-string types would require more care. When @options
1755 * come from -blockdev or blockdev_add, its members are typed
1756 * according to the QAPI schema, but when they come from
1757 * -drive, they're all QString.
1759 filename
= qdict_get_try_str(options
, "filename");
1762 if (drv
->bdrv_needs_filename
&& (!filename
|| !filename
[0])) {
1763 error_setg(errp
, "The '%s' block driver requires a file name",
1769 trace_bdrv_open_common(bs
, filename
?: "", bs
->open_flags
,
1772 bs
->read_only
= !(bs
->open_flags
& BDRV_O_RDWR
);
1774 if (use_bdrv_whitelist
&& !bdrv_is_whitelisted(drv
, bs
->read_only
)) {
1775 if (!bs
->read_only
&& bdrv_is_whitelisted(drv
, true)) {
1776 ret
= bdrv_apply_auto_read_only(bs
, NULL
, NULL
);
1782 !bs
->read_only
&& bdrv_is_whitelisted(drv
, true)
1783 ? "Driver '%s' can only be used for read-only devices"
1784 : "Driver '%s' is not whitelisted",
1790 /* bdrv_new() and bdrv_close() make it so */
1791 assert(qatomic_read(&bs
->copy_on_read
) == 0);
1793 if (bs
->open_flags
& BDRV_O_COPY_ON_READ
) {
1794 if (!bs
->read_only
) {
1795 bdrv_enable_copy_on_read(bs
);
1797 error_setg(errp
, "Can't use copy-on-read on read-only device");
1803 discard
= qemu_opt_get(opts
, BDRV_OPT_DISCARD
);
1804 if (discard
!= NULL
) {
1805 if (bdrv_parse_discard_flags(discard
, &bs
->open_flags
) != 0) {
1806 error_setg(errp
, "Invalid discard option");
1813 bdrv_parse_detect_zeroes(opts
, bs
->open_flags
, &local_err
);
1815 error_propagate(errp
, local_err
);
1820 if (filename
!= NULL
) {
1821 pstrcpy(bs
->filename
, sizeof(bs
->filename
), filename
);
1823 bs
->filename
[0] = '\0';
1825 pstrcpy(bs
->exact_filename
, sizeof(bs
->exact_filename
), bs
->filename
);
1827 /* Open the image, either directly or using a protocol */
1828 open_flags
= bdrv_open_flags(bs
, bs
->open_flags
);
1829 node_name
= qemu_opt_get(opts
, "node-name");
1831 assert(!drv
->bdrv_file_open
|| file
== NULL
);
1832 ret
= bdrv_open_driver(bs
, drv
, node_name
, options
, open_flags
, errp
);
1837 qemu_opts_del(opts
);
1841 qemu_opts_del(opts
);
1845 static QDict
*parse_json_filename(const char *filename
, Error
**errp
)
1847 QObject
*options_obj
;
1851 ret
= strstart(filename
, "json:", &filename
);
1854 options_obj
= qobject_from_json(filename
, errp
);
1856 error_prepend(errp
, "Could not parse the JSON options: ");
1860 options
= qobject_to(QDict
, options_obj
);
1862 qobject_unref(options_obj
);
1863 error_setg(errp
, "Invalid JSON object given");
1867 qdict_flatten(options
);
1872 static void parse_json_protocol(QDict
*options
, const char **pfilename
,
1875 QDict
*json_options
;
1876 Error
*local_err
= NULL
;
1878 /* Parse json: pseudo-protocol */
1879 if (!*pfilename
|| !g_str_has_prefix(*pfilename
, "json:")) {
1883 json_options
= parse_json_filename(*pfilename
, &local_err
);
1885 error_propagate(errp
, local_err
);
1889 /* Options given in the filename have lower priority than options
1890 * specified directly */
1891 qdict_join(options
, json_options
, false);
1892 qobject_unref(json_options
);
1897 * Fills in default options for opening images and converts the legacy
1898 * filename/flags pair to option QDict entries.
1899 * The BDRV_O_PROTOCOL flag in *flags will be set or cleared accordingly if a
1900 * block driver has been specified explicitly.
1902 static int bdrv_fill_options(QDict
**options
, const char *filename
,
1903 int *flags
, Error
**errp
)
1905 const char *drvname
;
1906 bool protocol
= *flags
& BDRV_O_PROTOCOL
;
1907 bool parse_filename
= false;
1908 BlockDriver
*drv
= NULL
;
1909 Error
*local_err
= NULL
;
1912 * Caution: while qdict_get_try_str() is fine, getting non-string
1913 * types would require more care. When @options come from
1914 * -blockdev or blockdev_add, its members are typed according to
1915 * the QAPI schema, but when they come from -drive, they're all
1918 drvname
= qdict_get_try_str(*options
, "driver");
1920 drv
= bdrv_find_format(drvname
);
1922 error_setg(errp
, "Unknown driver '%s'", drvname
);
1925 /* If the user has explicitly specified the driver, this choice should
1926 * override the BDRV_O_PROTOCOL flag */
1927 protocol
= drv
->bdrv_file_open
;
1931 *flags
|= BDRV_O_PROTOCOL
;
1933 *flags
&= ~BDRV_O_PROTOCOL
;
1936 /* Translate cache options from flags into options */
1937 update_options_from_flags(*options
, *flags
);
1939 /* Fetch the file name from the options QDict if necessary */
1940 if (protocol
&& filename
) {
1941 if (!qdict_haskey(*options
, "filename")) {
1942 qdict_put_str(*options
, "filename", filename
);
1943 parse_filename
= true;
1945 error_setg(errp
, "Can't specify 'file' and 'filename' options at "
1951 /* Find the right block driver */
1952 /* See cautionary note on accessing @options above */
1953 filename
= qdict_get_try_str(*options
, "filename");
1955 if (!drvname
&& protocol
) {
1957 drv
= bdrv_find_protocol(filename
, parse_filename
, errp
);
1962 drvname
= drv
->format_name
;
1963 qdict_put_str(*options
, "driver", drvname
);
1965 error_setg(errp
, "Must specify either driver or file");
1970 assert(drv
|| !protocol
);
1972 /* Driver-specific filename parsing */
1973 if (drv
&& drv
->bdrv_parse_filename
&& parse_filename
) {
1974 drv
->bdrv_parse_filename(filename
, *options
, &local_err
);
1976 error_propagate(errp
, local_err
);
1980 if (!drv
->bdrv_needs_filename
) {
1981 qdict_del(*options
, "filename");
1988 static int bdrv_check_update_perm(BlockDriverState
*bs
, BlockReopenQueue
*q
,
1989 uint64_t new_used_perm
,
1990 uint64_t new_shared_perm
,
1991 GSList
*ignore_children
,
1994 typedef struct BlockReopenQueueEntry
{
1997 BDRVReopenState state
;
1998 QTAILQ_ENTRY(BlockReopenQueueEntry
) entry
;
1999 } BlockReopenQueueEntry
;
2002 * Return the flags that @bs will have after the reopens in @q have
2003 * successfully completed. If @q is NULL (or @bs is not contained in @q),
2004 * return the current flags.
2006 static int bdrv_reopen_get_flags(BlockReopenQueue
*q
, BlockDriverState
*bs
)
2008 BlockReopenQueueEntry
*entry
;
2011 QTAILQ_FOREACH(entry
, q
, entry
) {
2012 if (entry
->state
.bs
== bs
) {
2013 return entry
->state
.flags
;
2018 return bs
->open_flags
;
2021 /* Returns whether the image file can be written to after the reopen queue @q
2022 * has been successfully applied, or right now if @q is NULL. */
2023 static bool bdrv_is_writable_after_reopen(BlockDriverState
*bs
,
2024 BlockReopenQueue
*q
)
2026 int flags
= bdrv_reopen_get_flags(q
, bs
);
2028 return (flags
& (BDRV_O_RDWR
| BDRV_O_INACTIVE
)) == BDRV_O_RDWR
;
2032 * Return whether the BDS can be written to. This is not necessarily
2033 * the same as !bdrv_is_read_only(bs), as inactivated images may not
2034 * be written to but do not count as read-only images.
2036 bool bdrv_is_writable(BlockDriverState
*bs
)
2038 return bdrv_is_writable_after_reopen(bs
, NULL
);
2041 static char *bdrv_child_user_desc(BdrvChild
*c
)
2043 if (c
->klass
->get_parent_desc
) {
2044 return c
->klass
->get_parent_desc(c
);
2047 return g_strdup("another user");
2050 static bool bdrv_a_allow_b(BdrvChild
*a
, BdrvChild
*b
, Error
**errp
)
2052 g_autofree
char *user
= NULL
;
2053 g_autofree
char *perm_names
= NULL
;
2055 if ((b
->perm
& a
->shared_perm
) == b
->perm
) {
2059 perm_names
= bdrv_perm_names(b
->perm
& ~a
->shared_perm
);
2060 user
= bdrv_child_user_desc(a
);
2061 error_setg(errp
, "Conflicts with use by %s as '%s', which does not "
2063 user
, a
->name
, perm_names
, bdrv_get_node_name(b
->bs
));
2068 static bool bdrv_parent_perms_conflict(BlockDriverState
*bs
,
2069 GSList
*ignore_children
,
2075 * During the loop we'll look at each pair twice. That's correct because
2076 * bdrv_a_allow_b() is asymmetric and we should check each pair in both
2079 QLIST_FOREACH(a
, &bs
->parents
, next_parent
) {
2080 if (g_slist_find(ignore_children
, a
)) {
2084 QLIST_FOREACH(b
, &bs
->parents
, next_parent
) {
2085 if (a
== b
|| g_slist_find(ignore_children
, b
)) {
2089 if (!bdrv_a_allow_b(a
, b
, errp
)) {
2098 static void bdrv_child_perm(BlockDriverState
*bs
, BlockDriverState
*child_bs
,
2099 BdrvChild
*c
, BdrvChildRole role
,
2100 BlockReopenQueue
*reopen_queue
,
2101 uint64_t parent_perm
, uint64_t parent_shared
,
2102 uint64_t *nperm
, uint64_t *nshared
)
2104 assert(bs
->drv
&& bs
->drv
->bdrv_child_perm
);
2105 bs
->drv
->bdrv_child_perm(bs
, c
, role
, reopen_queue
,
2106 parent_perm
, parent_shared
,
2108 /* TODO Take force_share from reopen_queue */
2109 if (child_bs
&& child_bs
->force_share
) {
2110 *nshared
= BLK_PERM_ALL
;
2115 * Adds the whole subtree of @bs (including @bs itself) to the @list (except for
2116 * nodes that are already in the @list, of course) so that final list is
2117 * topologically sorted. Return the result (GSList @list object is updated, so
2118 * don't use old reference after function call).
2120 * On function start @list must be already topologically sorted and for any node
2121 * in the @list the whole subtree of the node must be in the @list as well. The
2122 * simplest way to satisfy this criteria: use only result of
2123 * bdrv_topological_dfs() or NULL as @list parameter.
2125 static GSList
*bdrv_topological_dfs(GSList
*list
, GHashTable
*found
,
2126 BlockDriverState
*bs
)
2129 g_autoptr(GHashTable
) local_found
= NULL
;
2133 found
= local_found
= g_hash_table_new(NULL
, NULL
);
2136 if (g_hash_table_contains(found
, bs
)) {
2139 g_hash_table_add(found
, bs
);
2141 QLIST_FOREACH(child
, &bs
->children
, next
) {
2142 list
= bdrv_topological_dfs(list
, found
, child
->bs
);
2145 return g_slist_prepend(list
, bs
);
2148 static void bdrv_child_set_perm_commit(void *opaque
)
2150 BdrvChild
*c
= opaque
;
2152 c
->has_backup_perm
= false;
2155 static void bdrv_child_set_perm_abort(void *opaque
)
2157 BdrvChild
*c
= opaque
;
2159 * We may have child->has_backup_perm unset at this point, as in case of
2160 * _check_ stage of permission update failure we may _check_ not the whole
2161 * subtree. Still, _abort_ is called on the whole subtree anyway.
2163 if (c
->has_backup_perm
) {
2164 c
->perm
= c
->backup_perm
;
2165 c
->shared_perm
= c
->backup_shared_perm
;
2166 c
->has_backup_perm
= false;
2170 static TransactionActionDrv bdrv_child_set_pem_drv
= {
2171 .abort
= bdrv_child_set_perm_abort
,
2172 .commit
= bdrv_child_set_perm_commit
,
2176 * With tran=NULL needs to be followed by direct call to either
2177 * bdrv_child_set_perm_commit() or bdrv_child_set_perm_abort().
2179 * With non-NULL tran needs to be followed by tran_abort() or tran_commit()
2182 static void bdrv_child_set_perm_safe(BdrvChild
*c
, uint64_t perm
,
2183 uint64_t shared
, Transaction
*tran
)
2185 if (!c
->has_backup_perm
) {
2186 c
->has_backup_perm
= true;
2187 c
->backup_perm
= c
->perm
;
2188 c
->backup_shared_perm
= c
->shared_perm
;
2191 * Note: it's OK if c->has_backup_perm was already set, as we can find the
2192 * same c twice during check_perm procedure
2196 c
->shared_perm
= shared
;
2199 tran_add(tran
, &bdrv_child_set_pem_drv
, c
);
2203 static void bdrv_drv_set_perm_commit(void *opaque
)
2205 BlockDriverState
*bs
= opaque
;
2206 uint64_t cumulative_perms
, cumulative_shared_perms
;
2208 if (bs
->drv
->bdrv_set_perm
) {
2209 bdrv_get_cumulative_perm(bs
, &cumulative_perms
,
2210 &cumulative_shared_perms
);
2211 bs
->drv
->bdrv_set_perm(bs
, cumulative_perms
, cumulative_shared_perms
);
2215 static void bdrv_drv_set_perm_abort(void *opaque
)
2217 BlockDriverState
*bs
= opaque
;
2219 if (bs
->drv
->bdrv_abort_perm_update
) {
2220 bs
->drv
->bdrv_abort_perm_update(bs
);
2224 TransactionActionDrv bdrv_drv_set_perm_drv
= {
2225 .abort
= bdrv_drv_set_perm_abort
,
2226 .commit
= bdrv_drv_set_perm_commit
,
2229 static int bdrv_drv_set_perm(BlockDriverState
*bs
, uint64_t perm
,
2230 uint64_t shared_perm
, Transaction
*tran
,
2237 if (bs
->drv
->bdrv_check_perm
) {
2238 int ret
= bs
->drv
->bdrv_check_perm(bs
, perm
, shared_perm
, errp
);
2245 tran_add(tran
, &bdrv_drv_set_perm_drv
, bs
);
2251 typedef struct BdrvReplaceChildState
{
2253 BlockDriverState
*old_bs
;
2254 } BdrvReplaceChildState
;
2256 static void bdrv_replace_child_commit(void *opaque
)
2258 BdrvReplaceChildState
*s
= opaque
;
2260 bdrv_unref(s
->old_bs
);
2263 static void bdrv_replace_child_abort(void *opaque
)
2265 BdrvReplaceChildState
*s
= opaque
;
2266 BlockDriverState
*new_bs
= s
->child
->bs
;
2268 /* old_bs reference is transparently moved from @s to @s->child */
2269 bdrv_replace_child_noperm(s
->child
, s
->old_bs
);
2273 static TransactionActionDrv bdrv_replace_child_drv
= {
2274 .commit
= bdrv_replace_child_commit
,
2275 .abort
= bdrv_replace_child_abort
,
2280 * bdrv_replace_child_safe
2282 * Note: real unref of old_bs is done only on commit.
2284 static void bdrv_replace_child_safe(BdrvChild
*child
, BlockDriverState
*new_bs
,
2287 BdrvReplaceChildState
*s
= g_new(BdrvReplaceChildState
, 1);
2288 *s
= (BdrvReplaceChildState
) {
2290 .old_bs
= child
->bs
,
2292 tran_add(tran
, &bdrv_replace_child_drv
, s
);
2297 bdrv_replace_child_noperm(child
, new_bs
);
2298 /* old_bs reference is transparently moved from @child to @s */
2302 * Check whether permissions on this node can be changed in a way that
2303 * @cumulative_perms and @cumulative_shared_perms are the new cumulative
2304 * permissions of all its parents. This involves checking whether all necessary
2305 * permission changes to child nodes can be performed.
2307 * A call to this function must always be followed by a call to bdrv_set_perm()
2308 * or bdrv_abort_perm_update().
2310 static int bdrv_node_check_perm(BlockDriverState
*bs
, BlockReopenQueue
*q
,
2311 uint64_t cumulative_perms
,
2312 uint64_t cumulative_shared_perms
,
2313 GSList
*ignore_children
,
2314 Transaction
*tran
, Error
**errp
)
2316 BlockDriver
*drv
= bs
->drv
;
2320 /* Write permissions never work with read-only images */
2321 if ((cumulative_perms
& (BLK_PERM_WRITE
| BLK_PERM_WRITE_UNCHANGED
)) &&
2322 !bdrv_is_writable_after_reopen(bs
, q
))
2324 if (!bdrv_is_writable_after_reopen(bs
, NULL
)) {
2325 error_setg(errp
, "Block node is read-only");
2327 uint64_t current_perms
, current_shared
;
2328 bdrv_get_cumulative_perm(bs
, ¤t_perms
, ¤t_shared
);
2329 if (current_perms
& (BLK_PERM_WRITE
| BLK_PERM_WRITE_UNCHANGED
)) {
2330 error_setg(errp
, "Cannot make block node read-only, there is "
2333 error_setg(errp
, "Cannot make block node read-only and create "
2342 * Unaligned requests will automatically be aligned to bl.request_alignment
2343 * and without RESIZE we can't extend requests to write to space beyond the
2344 * end of the image, so it's required that the image size is aligned.
2346 if ((cumulative_perms
& (BLK_PERM_WRITE
| BLK_PERM_WRITE_UNCHANGED
)) &&
2347 !(cumulative_perms
& BLK_PERM_RESIZE
))
2349 if ((bs
->total_sectors
* BDRV_SECTOR_SIZE
) % bs
->bl
.request_alignment
) {
2350 error_setg(errp
, "Cannot get 'write' permission without 'resize': "
2351 "Image size is not a multiple of request "
2357 /* Check this node */
2362 ret
= bdrv_drv_set_perm(bs
, cumulative_perms
, cumulative_shared_perms
, tran
,
2368 /* Drivers that never have children can omit .bdrv_child_perm() */
2369 if (!drv
->bdrv_child_perm
) {
2370 assert(QLIST_EMPTY(&bs
->children
));
2374 /* Check all children */
2375 QLIST_FOREACH(c
, &bs
->children
, next
) {
2376 uint64_t cur_perm
, cur_shared
;
2378 bdrv_child_perm(bs
, c
->bs
, c
, c
->role
, q
,
2379 cumulative_perms
, cumulative_shared_perms
,
2380 &cur_perm
, &cur_shared
);
2381 bdrv_child_set_perm_safe(c
, cur_perm
, cur_shared
, tran
);
2388 * If use_cumulative_perms is true, use cumulative_perms and
2389 * cumulative_shared_perms for first element of the list. Otherwise just refresh
2392 static int bdrv_check_perm_common(GSList
*list
, BlockReopenQueue
*q
,
2393 bool use_cumulative_perms
,
2394 uint64_t cumulative_perms
,
2395 uint64_t cumulative_shared_perms
,
2396 GSList
*ignore_children
,
2397 Transaction
*tran
, Error
**errp
)
2400 BlockDriverState
*bs
;
2402 if (use_cumulative_perms
) {
2405 ret
= bdrv_node_check_perm(bs
, q
, cumulative_perms
,
2406 cumulative_shared_perms
,
2407 ignore_children
, tran
, errp
);
2415 for ( ; list
; list
= list
->next
) {
2418 if (bdrv_parent_perms_conflict(bs
, ignore_children
, errp
)) {
2422 bdrv_get_cumulative_perm(bs
, &cumulative_perms
,
2423 &cumulative_shared_perms
);
2425 ret
= bdrv_node_check_perm(bs
, q
, cumulative_perms
,
2426 cumulative_shared_perms
,
2427 ignore_children
, tran
, errp
);
2436 static int bdrv_check_perm(BlockDriverState
*bs
, BlockReopenQueue
*q
,
2437 uint64_t cumulative_perms
,
2438 uint64_t cumulative_shared_perms
,
2439 GSList
*ignore_children
, Error
**errp
)
2441 g_autoptr(GSList
) list
= bdrv_topological_dfs(NULL
, NULL
, bs
);
2442 return bdrv_check_perm_common(list
, q
, true, cumulative_perms
,
2443 cumulative_shared_perms
, ignore_children
,
2447 static int bdrv_list_refresh_perms(GSList
*list
, BlockReopenQueue
*q
,
2448 Transaction
*tran
, Error
**errp
)
2450 return bdrv_check_perm_common(list
, q
, false, 0, 0, NULL
, tran
, errp
);
2454 * Notifies drivers that after a previous bdrv_check_perm() call, the
2455 * permission update is not performed and any preparations made for it (e.g.
2456 * taken file locks) need to be undone.
2458 static void bdrv_node_abort_perm_update(BlockDriverState
*bs
)
2460 BlockDriver
*drv
= bs
->drv
;
2467 bdrv_drv_set_perm_abort(bs
);
2469 QLIST_FOREACH(c
, &bs
->children
, next
) {
2470 bdrv_child_set_perm_abort(c
);
2474 static void bdrv_list_abort_perm_update(GSList
*list
)
2476 for ( ; list
; list
= list
->next
) {
2477 bdrv_node_abort_perm_update((BlockDriverState
*)list
->data
);
2481 static void bdrv_abort_perm_update(BlockDriverState
*bs
)
2483 g_autoptr(GSList
) list
= bdrv_topological_dfs(NULL
, NULL
, bs
);
2484 return bdrv_list_abort_perm_update(list
);
2487 static void bdrv_node_set_perm(BlockDriverState
*bs
)
2489 BlockDriver
*drv
= bs
->drv
;
2496 bdrv_drv_set_perm_commit(bs
);
2498 /* Drivers that never have children can omit .bdrv_child_perm() */
2499 if (!drv
->bdrv_child_perm
) {
2500 assert(QLIST_EMPTY(&bs
->children
));
2504 /* Update all children */
2505 QLIST_FOREACH(c
, &bs
->children
, next
) {
2506 bdrv_child_set_perm_commit(c
);
2510 static void bdrv_list_set_perm(GSList
*list
)
2512 for ( ; list
; list
= list
->next
) {
2513 bdrv_node_set_perm((BlockDriverState
*)list
->data
);
2517 static void bdrv_set_perm(BlockDriverState
*bs
)
2519 g_autoptr(GSList
) list
= bdrv_topological_dfs(NULL
, NULL
, bs
);
2520 return bdrv_list_set_perm(list
);
2523 void bdrv_get_cumulative_perm(BlockDriverState
*bs
, uint64_t *perm
,
2524 uint64_t *shared_perm
)
2527 uint64_t cumulative_perms
= 0;
2528 uint64_t cumulative_shared_perms
= BLK_PERM_ALL
;
2530 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
2531 cumulative_perms
|= c
->perm
;
2532 cumulative_shared_perms
&= c
->shared_perm
;
2535 *perm
= cumulative_perms
;
2536 *shared_perm
= cumulative_shared_perms
;
2539 char *bdrv_perm_names(uint64_t perm
)
2545 { BLK_PERM_CONSISTENT_READ
, "consistent read" },
2546 { BLK_PERM_WRITE
, "write" },
2547 { BLK_PERM_WRITE_UNCHANGED
, "write unchanged" },
2548 { BLK_PERM_RESIZE
, "resize" },
2549 { BLK_PERM_GRAPH_MOD
, "change children" },
2553 GString
*result
= g_string_sized_new(30);
2554 struct perm_name
*p
;
2556 for (p
= permissions
; p
->name
; p
++) {
2557 if (perm
& p
->perm
) {
2558 if (result
->len
> 0) {
2559 g_string_append(result
, ", ");
2561 g_string_append(result
, p
->name
);
2565 return g_string_free(result
, FALSE
);
2569 * Checks whether a new reference to @bs can be added if the new user requires
2570 * @new_used_perm/@new_shared_perm as its permissions. If @ignore_children is
2571 * set, the BdrvChild objects in this list are ignored in the calculations;
2572 * this allows checking permission updates for an existing reference.
2574 * Needs to be followed by a call to either bdrv_set_perm() or
2575 * bdrv_abort_perm_update(). */
2576 static int bdrv_check_update_perm(BlockDriverState
*bs
, BlockReopenQueue
*q
,
2577 uint64_t new_used_perm
,
2578 uint64_t new_shared_perm
,
2579 GSList
*ignore_children
,
2583 uint64_t cumulative_perms
= new_used_perm
;
2584 uint64_t cumulative_shared_perms
= new_shared_perm
;
2587 /* There is no reason why anyone couldn't tolerate write_unchanged */
2588 assert(new_shared_perm
& BLK_PERM_WRITE_UNCHANGED
);
2590 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
2591 if (g_slist_find(ignore_children
, c
)) {
2595 if ((new_used_perm
& c
->shared_perm
) != new_used_perm
) {
2596 char *user
= bdrv_child_user_desc(c
);
2597 char *perm_names
= bdrv_perm_names(new_used_perm
& ~c
->shared_perm
);
2599 error_setg(errp
, "Conflicts with use by %s as '%s', which does not "
2601 user
, c
->name
, perm_names
, bdrv_get_node_name(c
->bs
));
2607 if ((c
->perm
& new_shared_perm
) != c
->perm
) {
2608 char *user
= bdrv_child_user_desc(c
);
2609 char *perm_names
= bdrv_perm_names(c
->perm
& ~new_shared_perm
);
2611 error_setg(errp
, "Conflicts with use by %s as '%s', which uses "
2613 user
, c
->name
, perm_names
, bdrv_get_node_name(c
->bs
));
2619 cumulative_perms
|= c
->perm
;
2620 cumulative_shared_perms
&= c
->shared_perm
;
2623 return bdrv_check_perm(bs
, q
, cumulative_perms
, cumulative_shared_perms
,
2624 ignore_children
, errp
);
2627 static int bdrv_refresh_perms(BlockDriverState
*bs
, Error
**errp
)
2630 Transaction
*tran
= tran_new();
2631 g_autoptr(GSList
) list
= bdrv_topological_dfs(NULL
, NULL
, bs
);
2633 ret
= bdrv_list_refresh_perms(list
, NULL
, tran
, errp
);
2634 tran_finalize(tran
, ret
);
2639 int bdrv_child_try_set_perm(BdrvChild
*c
, uint64_t perm
, uint64_t shared
,
2642 Error
*local_err
= NULL
;
2643 Transaction
*tran
= tran_new();
2646 bdrv_child_set_perm_safe(c
, perm
, shared
, tran
);
2648 ret
= bdrv_refresh_perms(c
->bs
, &local_err
);
2650 tran_finalize(tran
, ret
);
2653 if ((perm
& ~c
->perm
) || (c
->shared_perm
& ~shared
)) {
2654 /* tighten permissions */
2655 error_propagate(errp
, local_err
);
2658 * Our caller may intend to only loosen restrictions and
2659 * does not expect this function to fail. Errors are not
2660 * fatal in such a case, so we can just hide them from our
2663 error_free(local_err
);
2671 int bdrv_child_refresh_perms(BlockDriverState
*bs
, BdrvChild
*c
, Error
**errp
)
2673 uint64_t parent_perms
, parent_shared
;
2674 uint64_t perms
, shared
;
2676 bdrv_get_cumulative_perm(bs
, &parent_perms
, &parent_shared
);
2677 bdrv_child_perm(bs
, c
->bs
, c
, c
->role
, NULL
,
2678 parent_perms
, parent_shared
, &perms
, &shared
);
2680 return bdrv_child_try_set_perm(c
, perms
, shared
, errp
);
2684 * Default implementation for .bdrv_child_perm() for block filters:
2685 * Forward CONSISTENT_READ, WRITE, WRITE_UNCHANGED, and RESIZE to the
2688 static void bdrv_filter_default_perms(BlockDriverState
*bs
, BdrvChild
*c
,
2690 BlockReopenQueue
*reopen_queue
,
2691 uint64_t perm
, uint64_t shared
,
2692 uint64_t *nperm
, uint64_t *nshared
)
2694 *nperm
= perm
& DEFAULT_PERM_PASSTHROUGH
;
2695 *nshared
= (shared
& DEFAULT_PERM_PASSTHROUGH
) | DEFAULT_PERM_UNCHANGED
;
2698 static void bdrv_default_perms_for_cow(BlockDriverState
*bs
, BdrvChild
*c
,
2700 BlockReopenQueue
*reopen_queue
,
2701 uint64_t perm
, uint64_t shared
,
2702 uint64_t *nperm
, uint64_t *nshared
)
2704 assert(role
& BDRV_CHILD_COW
);
2707 * We want consistent read from backing files if the parent needs it.
2708 * No other operations are performed on backing files.
2710 perm
&= BLK_PERM_CONSISTENT_READ
;
2713 * If the parent can deal with changing data, we're okay with a
2714 * writable and resizable backing file.
2715 * TODO Require !(perm & BLK_PERM_CONSISTENT_READ), too?
2717 if (shared
& BLK_PERM_WRITE
) {
2718 shared
= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2723 shared
|= BLK_PERM_CONSISTENT_READ
| BLK_PERM_GRAPH_MOD
|
2724 BLK_PERM_WRITE_UNCHANGED
;
2726 if (bs
->open_flags
& BDRV_O_INACTIVE
) {
2727 shared
|= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2734 static void bdrv_default_perms_for_storage(BlockDriverState
*bs
, BdrvChild
*c
,
2736 BlockReopenQueue
*reopen_queue
,
2737 uint64_t perm
, uint64_t shared
,
2738 uint64_t *nperm
, uint64_t *nshared
)
2742 assert(role
& (BDRV_CHILD_METADATA
| BDRV_CHILD_DATA
));
2744 flags
= bdrv_reopen_get_flags(reopen_queue
, bs
);
2747 * Apart from the modifications below, the same permissions are
2748 * forwarded and left alone as for filters
2750 bdrv_filter_default_perms(bs
, c
, role
, reopen_queue
,
2751 perm
, shared
, &perm
, &shared
);
2753 if (role
& BDRV_CHILD_METADATA
) {
2754 /* Format drivers may touch metadata even if the guest doesn't write */
2755 if (bdrv_is_writable_after_reopen(bs
, reopen_queue
)) {
2756 perm
|= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2760 * bs->file always needs to be consistent because of the
2761 * metadata. We can never allow other users to resize or write
2764 if (!(flags
& BDRV_O_NO_IO
)) {
2765 perm
|= BLK_PERM_CONSISTENT_READ
;
2767 shared
&= ~(BLK_PERM_WRITE
| BLK_PERM_RESIZE
);
2770 if (role
& BDRV_CHILD_DATA
) {
2772 * Technically, everything in this block is a subset of the
2773 * BDRV_CHILD_METADATA path taken above, and so this could
2774 * be an "else if" branch. However, that is not obvious, and
2775 * this function is not performance critical, therefore we let
2776 * this be an independent "if".
2780 * We cannot allow other users to resize the file because the
2781 * format driver might have some assumptions about the size
2782 * (e.g. because it is stored in metadata, or because the file
2783 * is split into fixed-size data files).
2785 shared
&= ~BLK_PERM_RESIZE
;
2788 * WRITE_UNCHANGED often cannot be performed as such on the
2789 * data file. For example, the qcow2 driver may still need to
2790 * write copied clusters on copy-on-read.
2792 if (perm
& BLK_PERM_WRITE_UNCHANGED
) {
2793 perm
|= BLK_PERM_WRITE
;
2797 * If the data file is written to, the format driver may
2798 * expect to be able to resize it by writing beyond the EOF.
2800 if (perm
& BLK_PERM_WRITE
) {
2801 perm
|= BLK_PERM_RESIZE
;
2805 if (bs
->open_flags
& BDRV_O_INACTIVE
) {
2806 shared
|= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2813 void bdrv_default_perms(BlockDriverState
*bs
, BdrvChild
*c
,
2814 BdrvChildRole role
, BlockReopenQueue
*reopen_queue
,
2815 uint64_t perm
, uint64_t shared
,
2816 uint64_t *nperm
, uint64_t *nshared
)
2818 if (role
& BDRV_CHILD_FILTERED
) {
2819 assert(!(role
& (BDRV_CHILD_DATA
| BDRV_CHILD_METADATA
|
2821 bdrv_filter_default_perms(bs
, c
, role
, reopen_queue
,
2822 perm
, shared
, nperm
, nshared
);
2823 } else if (role
& BDRV_CHILD_COW
) {
2824 assert(!(role
& (BDRV_CHILD_DATA
| BDRV_CHILD_METADATA
)));
2825 bdrv_default_perms_for_cow(bs
, c
, role
, reopen_queue
,
2826 perm
, shared
, nperm
, nshared
);
2827 } else if (role
& (BDRV_CHILD_METADATA
| BDRV_CHILD_DATA
)) {
2828 bdrv_default_perms_for_storage(bs
, c
, role
, reopen_queue
,
2829 perm
, shared
, nperm
, nshared
);
2831 g_assert_not_reached();
2835 uint64_t bdrv_qapi_perm_to_blk_perm(BlockPermission qapi_perm
)
2837 static const uint64_t permissions
[] = {
2838 [BLOCK_PERMISSION_CONSISTENT_READ
] = BLK_PERM_CONSISTENT_READ
,
2839 [BLOCK_PERMISSION_WRITE
] = BLK_PERM_WRITE
,
2840 [BLOCK_PERMISSION_WRITE_UNCHANGED
] = BLK_PERM_WRITE_UNCHANGED
,
2841 [BLOCK_PERMISSION_RESIZE
] = BLK_PERM_RESIZE
,
2842 [BLOCK_PERMISSION_GRAPH_MOD
] = BLK_PERM_GRAPH_MOD
,
2845 QEMU_BUILD_BUG_ON(ARRAY_SIZE(permissions
) != BLOCK_PERMISSION__MAX
);
2846 QEMU_BUILD_BUG_ON(1UL << ARRAY_SIZE(permissions
) != BLK_PERM_ALL
+ 1);
2848 assert(qapi_perm
< BLOCK_PERMISSION__MAX
);
2850 return permissions
[qapi_perm
];
2853 static void bdrv_replace_child_noperm(BdrvChild
*child
,
2854 BlockDriverState
*new_bs
)
2856 BlockDriverState
*old_bs
= child
->bs
;
2857 int new_bs_quiesce_counter
;
2860 assert(!child
->frozen
);
2862 if (old_bs
&& new_bs
) {
2863 assert(bdrv_get_aio_context(old_bs
) == bdrv_get_aio_context(new_bs
));
2866 new_bs_quiesce_counter
= (new_bs
? new_bs
->quiesce_counter
: 0);
2867 drain_saldo
= new_bs_quiesce_counter
- child
->parent_quiesce_counter
;
2870 * If the new child node is drained but the old one was not, flush
2871 * all outstanding requests to the old child node.
2873 while (drain_saldo
> 0 && child
->klass
->drained_begin
) {
2874 bdrv_parent_drained_begin_single(child
, true);
2879 /* Detach first so that the recursive drain sections coming from @child
2880 * are already gone and we only end the drain sections that came from
2882 if (child
->klass
->detach
) {
2883 child
->klass
->detach(child
);
2885 QLIST_REMOVE(child
, next_parent
);
2891 QLIST_INSERT_HEAD(&new_bs
->parents
, child
, next_parent
);
2894 * Detaching the old node may have led to the new node's
2895 * quiesce_counter having been decreased. Not a problem, we
2896 * just need to recognize this here and then invoke
2897 * drained_end appropriately more often.
2899 assert(new_bs
->quiesce_counter
<= new_bs_quiesce_counter
);
2900 drain_saldo
+= new_bs
->quiesce_counter
- new_bs_quiesce_counter
;
2902 /* Attach only after starting new drained sections, so that recursive
2903 * drain sections coming from @child don't get an extra .drained_begin
2905 if (child
->klass
->attach
) {
2906 child
->klass
->attach(child
);
2911 * If the old child node was drained but the new one is not, allow
2912 * requests to come in only after the new node has been attached.
2914 while (drain_saldo
< 0 && child
->klass
->drained_end
) {
2915 bdrv_parent_drained_end_single(child
);
2921 * Updates @child to change its reference to point to @new_bs, including
2922 * checking and applying the necessary permission updates both to the old node
2925 * NULL is passed as @new_bs for removing the reference before freeing @child.
2927 * If @new_bs is not NULL, bdrv_check_perm() must be called beforehand, as this
2928 * function uses bdrv_set_perm() to update the permissions according to the new
2929 * reference that @new_bs gets.
2931 * Callers must ensure that child->frozen is false.
2933 static void bdrv_replace_child(BdrvChild
*child
, BlockDriverState
*new_bs
)
2935 BlockDriverState
*old_bs
= child
->bs
;
2937 /* Asserts that child->frozen == false */
2938 bdrv_replace_child_noperm(child
, new_bs
);
2941 * Start with the new node's permissions. If @new_bs is a (direct
2942 * or indirect) child of @old_bs, we must complete the permission
2943 * update on @new_bs before we loosen the restrictions on @old_bs.
2944 * Otherwise, bdrv_check_perm() on @old_bs would re-initiate
2945 * updating the permissions of @new_bs, and thus not purely loosen
2949 bdrv_set_perm(new_bs
);
2954 * Update permissions for old node. We're just taking a parent away, so
2955 * we're loosening restrictions. Errors of permission update are not
2956 * fatal in this case, ignore them.
2958 bdrv_refresh_perms(old_bs
, NULL
);
2960 /* When the parent requiring a non-default AioContext is removed, the
2961 * node moves back to the main AioContext */
2962 bdrv_try_set_aio_context(old_bs
, qemu_get_aio_context(), NULL
);
2966 static void bdrv_remove_empty_child(BdrvChild
*child
)
2969 QLIST_SAFE_REMOVE(child
, next
);
2970 g_free(child
->name
);
2974 typedef struct BdrvAttachChildCommonState
{
2976 AioContext
*old_parent_ctx
;
2977 AioContext
*old_child_ctx
;
2978 } BdrvAttachChildCommonState
;
2980 static void bdrv_attach_child_common_abort(void *opaque
)
2982 BdrvAttachChildCommonState
*s
= opaque
;
2983 BdrvChild
*child
= *s
->child
;
2984 BlockDriverState
*bs
= child
->bs
;
2986 bdrv_replace_child_noperm(child
, NULL
);
2988 if (bdrv_get_aio_context(bs
) != s
->old_child_ctx
) {
2989 bdrv_try_set_aio_context(bs
, s
->old_child_ctx
, &error_abort
);
2992 if (bdrv_child_get_parent_aio_context(child
) != s
->old_parent_ctx
) {
2993 GSList
*ignore
= g_slist_prepend(NULL
, child
);
2995 child
->klass
->can_set_aio_ctx(child
, s
->old_parent_ctx
, &ignore
,
2997 g_slist_free(ignore
);
2998 ignore
= g_slist_prepend(NULL
, child
);
2999 child
->klass
->set_aio_ctx(child
, s
->old_parent_ctx
, &ignore
);
3001 g_slist_free(ignore
);
3005 bdrv_remove_empty_child(child
);
3009 static TransactionActionDrv bdrv_attach_child_common_drv
= {
3010 .abort
= bdrv_attach_child_common_abort
,
3015 * Common part of attaching bdrv child to bs or to blk or to job
3017 static int bdrv_attach_child_common(BlockDriverState
*child_bs
,
3018 const char *child_name
,
3019 const BdrvChildClass
*child_class
,
3020 BdrvChildRole child_role
,
3021 uint64_t perm
, uint64_t shared_perm
,
3022 void *opaque
, BdrvChild
**child
,
3023 Transaction
*tran
, Error
**errp
)
3025 BdrvChild
*new_child
;
3026 AioContext
*parent_ctx
;
3027 AioContext
*child_ctx
= bdrv_get_aio_context(child_bs
);
3030 assert(*child
== NULL
);
3032 new_child
= g_new(BdrvChild
, 1);
3033 *new_child
= (BdrvChild
) {
3035 .name
= g_strdup(child_name
),
3036 .klass
= child_class
,
3039 .shared_perm
= shared_perm
,
3044 * If the AioContexts don't match, first try to move the subtree of
3045 * child_bs into the AioContext of the new parent. If this doesn't work,
3046 * try moving the parent into the AioContext of child_bs instead.
3048 parent_ctx
= bdrv_child_get_parent_aio_context(new_child
);
3049 if (child_ctx
!= parent_ctx
) {
3050 Error
*local_err
= NULL
;
3051 int ret
= bdrv_try_set_aio_context(child_bs
, parent_ctx
, &local_err
);
3053 if (ret
< 0 && child_class
->can_set_aio_ctx
) {
3054 GSList
*ignore
= g_slist_prepend(NULL
, new_child
);
3055 if (child_class
->can_set_aio_ctx(new_child
, child_ctx
, &ignore
,
3058 error_free(local_err
);
3060 g_slist_free(ignore
);
3061 ignore
= g_slist_prepend(NULL
, new_child
);
3062 child_class
->set_aio_ctx(new_child
, child_ctx
, &ignore
);
3064 g_slist_free(ignore
);
3068 error_propagate(errp
, local_err
);
3069 bdrv_remove_empty_child(new_child
);
3075 bdrv_replace_child_noperm(new_child
, child_bs
);
3079 BdrvAttachChildCommonState
*s
= g_new(BdrvAttachChildCommonState
, 1);
3080 *s
= (BdrvAttachChildCommonState
) {
3082 .old_parent_ctx
= parent_ctx
,
3083 .old_child_ctx
= child_ctx
,
3085 tran_add(tran
, &bdrv_attach_child_common_drv
, s
);
3090 static int bdrv_attach_child_noperm(BlockDriverState
*parent_bs
,
3091 BlockDriverState
*child_bs
,
3092 const char *child_name
,
3093 const BdrvChildClass
*child_class
,
3094 BdrvChildRole child_role
,
3100 uint64_t perm
, shared_perm
;
3102 assert(parent_bs
->drv
);
3104 bdrv_get_cumulative_perm(parent_bs
, &perm
, &shared_perm
);
3105 bdrv_child_perm(parent_bs
, child_bs
, NULL
, child_role
, NULL
,
3106 perm
, shared_perm
, &perm
, &shared_perm
);
3108 ret
= bdrv_attach_child_common(child_bs
, child_name
, child_class
,
3109 child_role
, perm
, shared_perm
, parent_bs
,
3115 QLIST_INSERT_HEAD(&parent_bs
->children
, *child
, next
);
3117 * child is removed in bdrv_attach_child_common_abort(), so don't care to
3118 * abort this change separately.
3124 static void bdrv_detach_child(BdrvChild
*child
)
3126 bdrv_replace_child(child
, NULL
);
3127 bdrv_remove_empty_child(child
);
3131 * This function steals the reference to child_bs from the caller.
3132 * That reference is later dropped by bdrv_root_unref_child().
3134 * On failure NULL is returned, errp is set and the reference to
3135 * child_bs is also dropped.
3137 * The caller must hold the AioContext lock @child_bs, but not that of @ctx
3138 * (unless @child_bs is already in @ctx).
3140 BdrvChild
*bdrv_root_attach_child(BlockDriverState
*child_bs
,
3141 const char *child_name
,
3142 const BdrvChildClass
*child_class
,
3143 BdrvChildRole child_role
,
3144 uint64_t perm
, uint64_t shared_perm
,
3145 void *opaque
, Error
**errp
)
3148 BdrvChild
*child
= NULL
;
3149 Transaction
*tran
= tran_new();
3151 ret
= bdrv_attach_child_common(child_bs
, child_name
, child_class
,
3152 child_role
, perm
, shared_perm
, opaque
,
3153 &child
, tran
, errp
);
3155 bdrv_unref(child_bs
);
3159 ret
= bdrv_refresh_perms(child_bs
, errp
);
3160 tran_finalize(tran
, ret
);
3162 bdrv_unref(child_bs
);
3167 * This function transfers the reference to child_bs from the caller
3168 * to parent_bs. That reference is later dropped by parent_bs on
3169 * bdrv_close() or if someone calls bdrv_unref_child().
3171 * On failure NULL is returned, errp is set and the reference to
3172 * child_bs is also dropped.
3174 * If @parent_bs and @child_bs are in different AioContexts, the caller must
3175 * hold the AioContext lock for @child_bs, but not for @parent_bs.
3177 BdrvChild
*bdrv_attach_child(BlockDriverState
*parent_bs
,
3178 BlockDriverState
*child_bs
,
3179 const char *child_name
,
3180 const BdrvChildClass
*child_class
,
3181 BdrvChildRole child_role
,
3185 BdrvChild
*child
= NULL
;
3186 Transaction
*tran
= tran_new();
3188 ret
= bdrv_attach_child_noperm(parent_bs
, child_bs
, child_name
, child_class
,
3189 child_role
, &child
, tran
, errp
);
3194 ret
= bdrv_refresh_perms(parent_bs
, errp
);
3200 tran_finalize(tran
, ret
);
3202 bdrv_unref(child_bs
);
3207 /* Callers must ensure that child->frozen is false. */
3208 void bdrv_root_unref_child(BdrvChild
*child
)
3210 BlockDriverState
*child_bs
;
3212 child_bs
= child
->bs
;
3213 bdrv_detach_child(child
);
3214 bdrv_unref(child_bs
);
3218 * Clear all inherits_from pointers from children and grandchildren of
3219 * @root that point to @root, where necessary.
3221 static void bdrv_unset_inherits_from(BlockDriverState
*root
, BdrvChild
*child
)
3225 if (child
->bs
->inherits_from
== root
) {
3227 * Remove inherits_from only when the last reference between root and
3228 * child->bs goes away.
3230 QLIST_FOREACH(c
, &root
->children
, next
) {
3231 if (c
!= child
&& c
->bs
== child
->bs
) {
3236 child
->bs
->inherits_from
= NULL
;
3240 QLIST_FOREACH(c
, &child
->bs
->children
, next
) {
3241 bdrv_unset_inherits_from(root
, c
);
3245 /* Callers must ensure that child->frozen is false. */
3246 void bdrv_unref_child(BlockDriverState
*parent
, BdrvChild
*child
)
3248 if (child
== NULL
) {
3252 bdrv_unset_inherits_from(parent
, child
);
3253 bdrv_root_unref_child(child
);
3257 static void bdrv_parent_cb_change_media(BlockDriverState
*bs
, bool load
)
3260 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
3261 if (c
->klass
->change_media
) {
3262 c
->klass
->change_media(c
, load
);
3267 /* Return true if you can reach parent going through child->inherits_from
3268 * recursively. If parent or child are NULL, return false */
3269 static bool bdrv_inherits_from_recursive(BlockDriverState
*child
,
3270 BlockDriverState
*parent
)
3272 while (child
&& child
!= parent
) {
3273 child
= child
->inherits_from
;
3276 return child
!= NULL
;
3280 * Return the BdrvChildRole for @bs's backing child. bs->backing is
3281 * mostly used for COW backing children (role = COW), but also for
3282 * filtered children (role = FILTERED | PRIMARY).
3284 static BdrvChildRole
bdrv_backing_role(BlockDriverState
*bs
)
3286 if (bs
->drv
&& bs
->drv
->is_filter
) {
3287 return BDRV_CHILD_FILTERED
| BDRV_CHILD_PRIMARY
;
3289 return BDRV_CHILD_COW
;
3294 * Sets the bs->backing link of a BDS. A new reference is created; callers
3295 * which don't need their own reference any more must call bdrv_unref().
3297 int bdrv_set_backing_hd(BlockDriverState
*bs
, BlockDriverState
*backing_hd
,
3301 bool update_inherits_from
= bdrv_chain_contains(bs
, backing_hd
) &&
3302 bdrv_inherits_from_recursive(backing_hd
, bs
);
3304 if (bdrv_is_backing_chain_frozen(bs
, child_bs(bs
->backing
), errp
)) {
3309 bdrv_ref(backing_hd
);
3313 /* Cannot be frozen, we checked that above */
3314 bdrv_unref_child(bs
, bs
->backing
);
3322 bs
->backing
= bdrv_attach_child(bs
, backing_hd
, "backing", &child_of_bds
,
3323 bdrv_backing_role(bs
), errp
);
3329 /* If backing_hd was already part of bs's backing chain, and
3330 * inherits_from pointed recursively to bs then let's update it to
3331 * point directly to bs (else it will become NULL). */
3332 if (update_inherits_from
) {
3333 backing_hd
->inherits_from
= bs
;
3337 bdrv_refresh_limits(bs
, NULL
);
3343 * Opens the backing file for a BlockDriverState if not yet open
3345 * bdref_key specifies the key for the image's BlockdevRef in the options QDict.
3346 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
3347 * itself, all options starting with "${bdref_key}." are considered part of the
3350 * TODO Can this be unified with bdrv_open_image()?
3352 int bdrv_open_backing_file(BlockDriverState
*bs
, QDict
*parent_options
,
3353 const char *bdref_key
, Error
**errp
)
3355 char *backing_filename
= NULL
;
3356 char *bdref_key_dot
;
3357 const char *reference
= NULL
;
3359 bool implicit_backing
= false;
3360 BlockDriverState
*backing_hd
;
3362 QDict
*tmp_parent_options
= NULL
;
3363 Error
*local_err
= NULL
;
3365 if (bs
->backing
!= NULL
) {
3369 /* NULL means an empty set of options */
3370 if (parent_options
== NULL
) {
3371 tmp_parent_options
= qdict_new();
3372 parent_options
= tmp_parent_options
;
3375 bs
->open_flags
&= ~BDRV_O_NO_BACKING
;
3377 bdref_key_dot
= g_strdup_printf("%s.", bdref_key
);
3378 qdict_extract_subqdict(parent_options
, &options
, bdref_key_dot
);
3379 g_free(bdref_key_dot
);
3382 * Caution: while qdict_get_try_str() is fine, getting non-string
3383 * types would require more care. When @parent_options come from
3384 * -blockdev or blockdev_add, its members are typed according to
3385 * the QAPI schema, but when they come from -drive, they're all
3388 reference
= qdict_get_try_str(parent_options
, bdref_key
);
3389 if (reference
|| qdict_haskey(options
, "file.filename")) {
3390 /* keep backing_filename NULL */
3391 } else if (bs
->backing_file
[0] == '\0' && qdict_size(options
) == 0) {
3392 qobject_unref(options
);
3395 if (qdict_size(options
) == 0) {
3396 /* If the user specifies options that do not modify the
3397 * backing file's behavior, we might still consider it the
3398 * implicit backing file. But it's easier this way, and
3399 * just specifying some of the backing BDS's options is
3400 * only possible with -drive anyway (otherwise the QAPI
3401 * schema forces the user to specify everything). */
3402 implicit_backing
= !strcmp(bs
->auto_backing_file
, bs
->backing_file
);
3405 backing_filename
= bdrv_get_full_backing_filename(bs
, &local_err
);
3408 error_propagate(errp
, local_err
);
3409 qobject_unref(options
);
3414 if (!bs
->drv
|| !bs
->drv
->supports_backing
) {
3416 error_setg(errp
, "Driver doesn't support backing files");
3417 qobject_unref(options
);
3422 bs
->backing_format
[0] != '\0' && !qdict_haskey(options
, "driver")) {
3423 qdict_put_str(options
, "driver", bs
->backing_format
);
3426 backing_hd
= bdrv_open_inherit(backing_filename
, reference
, options
, 0, bs
,
3427 &child_of_bds
, bdrv_backing_role(bs
), errp
);
3429 bs
->open_flags
|= BDRV_O_NO_BACKING
;
3430 error_prepend(errp
, "Could not open backing file: ");
3435 if (implicit_backing
) {
3436 bdrv_refresh_filename(backing_hd
);
3437 pstrcpy(bs
->auto_backing_file
, sizeof(bs
->auto_backing_file
),
3438 backing_hd
->filename
);
3441 /* Hook up the backing file link; drop our reference, bs owns the
3442 * backing_hd reference now */
3443 ret
= bdrv_set_backing_hd(bs
, backing_hd
, errp
);
3444 bdrv_unref(backing_hd
);
3449 qdict_del(parent_options
, bdref_key
);
3452 g_free(backing_filename
);
3453 qobject_unref(tmp_parent_options
);
3457 static BlockDriverState
*
3458 bdrv_open_child_bs(const char *filename
, QDict
*options
, const char *bdref_key
,
3459 BlockDriverState
*parent
, const BdrvChildClass
*child_class
,
3460 BdrvChildRole child_role
, bool allow_none
, Error
**errp
)
3462 BlockDriverState
*bs
= NULL
;
3463 QDict
*image_options
;
3464 char *bdref_key_dot
;
3465 const char *reference
;
3467 assert(child_class
!= NULL
);
3469 bdref_key_dot
= g_strdup_printf("%s.", bdref_key
);
3470 qdict_extract_subqdict(options
, &image_options
, bdref_key_dot
);
3471 g_free(bdref_key_dot
);
3474 * Caution: while qdict_get_try_str() is fine, getting non-string
3475 * types would require more care. When @options come from
3476 * -blockdev or blockdev_add, its members are typed according to
3477 * the QAPI schema, but when they come from -drive, they're all
3480 reference
= qdict_get_try_str(options
, bdref_key
);
3481 if (!filename
&& !reference
&& !qdict_size(image_options
)) {
3483 error_setg(errp
, "A block device must be specified for \"%s\"",
3486 qobject_unref(image_options
);
3490 bs
= bdrv_open_inherit(filename
, reference
, image_options
, 0,
3491 parent
, child_class
, child_role
, errp
);
3497 qdict_del(options
, bdref_key
);
3502 * Opens a disk image whose options are given as BlockdevRef in another block
3505 * If allow_none is true, no image will be opened if filename is false and no
3506 * BlockdevRef is given. NULL will be returned, but errp remains unset.
3508 * bdrev_key specifies the key for the image's BlockdevRef in the options QDict.
3509 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
3510 * itself, all options starting with "${bdref_key}." are considered part of the
3513 * The BlockdevRef will be removed from the options QDict.
3515 BdrvChild
*bdrv_open_child(const char *filename
,
3516 QDict
*options
, const char *bdref_key
,
3517 BlockDriverState
*parent
,
3518 const BdrvChildClass
*child_class
,
3519 BdrvChildRole child_role
,
3520 bool allow_none
, Error
**errp
)
3522 BlockDriverState
*bs
;
3524 bs
= bdrv_open_child_bs(filename
, options
, bdref_key
, parent
, child_class
,
3525 child_role
, allow_none
, errp
);
3530 return bdrv_attach_child(parent
, bs
, bdref_key
, child_class
, child_role
,
3535 * TODO Future callers may need to specify parent/child_class in order for
3536 * option inheritance to work. Existing callers use it for the root node.
3538 BlockDriverState
*bdrv_open_blockdev_ref(BlockdevRef
*ref
, Error
**errp
)
3540 BlockDriverState
*bs
= NULL
;
3541 QObject
*obj
= NULL
;
3542 QDict
*qdict
= NULL
;
3543 const char *reference
= NULL
;
3546 if (ref
->type
== QTYPE_QSTRING
) {
3547 reference
= ref
->u
.reference
;
3549 BlockdevOptions
*options
= &ref
->u
.definition
;
3550 assert(ref
->type
== QTYPE_QDICT
);
3552 v
= qobject_output_visitor_new(&obj
);
3553 visit_type_BlockdevOptions(v
, NULL
, &options
, &error_abort
);
3554 visit_complete(v
, &obj
);
3556 qdict
= qobject_to(QDict
, obj
);
3557 qdict_flatten(qdict
);
3559 /* bdrv_open_inherit() defaults to the values in bdrv_flags (for
3560 * compatibility with other callers) rather than what we want as the
3561 * real defaults. Apply the defaults here instead. */
3562 qdict_set_default_str(qdict
, BDRV_OPT_CACHE_DIRECT
, "off");
3563 qdict_set_default_str(qdict
, BDRV_OPT_CACHE_NO_FLUSH
, "off");
3564 qdict_set_default_str(qdict
, BDRV_OPT_READ_ONLY
, "off");
3565 qdict_set_default_str(qdict
, BDRV_OPT_AUTO_READ_ONLY
, "off");
3569 bs
= bdrv_open_inherit(NULL
, reference
, qdict
, 0, NULL
, NULL
, 0, errp
);
3576 static BlockDriverState
*bdrv_append_temp_snapshot(BlockDriverState
*bs
,
3578 QDict
*snapshot_options
,
3581 /* TODO: extra byte is a hack to ensure MAX_PATH space on Windows. */
3582 char *tmp_filename
= g_malloc0(PATH_MAX
+ 1);
3584 QemuOpts
*opts
= NULL
;
3585 BlockDriverState
*bs_snapshot
= NULL
;
3588 /* if snapshot, we create a temporary backing file and open it
3589 instead of opening 'filename' directly */
3591 /* Get the required size from the image */
3592 total_size
= bdrv_getlength(bs
);
3593 if (total_size
< 0) {
3594 error_setg_errno(errp
, -total_size
, "Could not get image size");
3598 /* Create the temporary image */
3599 ret
= get_tmp_filename(tmp_filename
, PATH_MAX
+ 1);
3601 error_setg_errno(errp
, -ret
, "Could not get temporary filename");
3605 opts
= qemu_opts_create(bdrv_qcow2
.create_opts
, NULL
, 0,
3607 qemu_opt_set_number(opts
, BLOCK_OPT_SIZE
, total_size
, &error_abort
);
3608 ret
= bdrv_create(&bdrv_qcow2
, tmp_filename
, opts
, errp
);
3609 qemu_opts_del(opts
);
3611 error_prepend(errp
, "Could not create temporary overlay '%s': ",
3616 /* Prepare options QDict for the temporary file */
3617 qdict_put_str(snapshot_options
, "file.driver", "file");
3618 qdict_put_str(snapshot_options
, "file.filename", tmp_filename
);
3619 qdict_put_str(snapshot_options
, "driver", "qcow2");
3621 bs_snapshot
= bdrv_open(NULL
, NULL
, snapshot_options
, flags
, errp
);
3622 snapshot_options
= NULL
;
3627 ret
= bdrv_append(bs_snapshot
, bs
, errp
);
3634 qobject_unref(snapshot_options
);
3635 g_free(tmp_filename
);
3640 * Opens a disk image (raw, qcow2, vmdk, ...)
3642 * options is a QDict of options to pass to the block drivers, or NULL for an
3643 * empty set of options. The reference to the QDict belongs to the block layer
3644 * after the call (even on failure), so if the caller intends to reuse the
3645 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
3647 * If *pbs is NULL, a new BDS will be created with a pointer to it stored there.
3648 * If it is not NULL, the referenced BDS will be reused.
3650 * The reference parameter may be used to specify an existing block device which
3651 * should be opened. If specified, neither options nor a filename may be given,
3652 * nor can an existing BDS be reused (that is, *pbs has to be NULL).
3654 static BlockDriverState
*bdrv_open_inherit(const char *filename
,
3655 const char *reference
,
3656 QDict
*options
, int flags
,
3657 BlockDriverState
*parent
,
3658 const BdrvChildClass
*child_class
,
3659 BdrvChildRole child_role
,
3663 BlockBackend
*file
= NULL
;
3664 BlockDriverState
*bs
;
3665 BlockDriver
*drv
= NULL
;
3667 const char *drvname
;
3668 const char *backing
;
3669 Error
*local_err
= NULL
;
3670 QDict
*snapshot_options
= NULL
;
3671 int snapshot_flags
= 0;
3673 assert(!child_class
|| !flags
);
3674 assert(!child_class
== !parent
);
3677 bool options_non_empty
= options
? qdict_size(options
) : false;
3678 qobject_unref(options
);
3680 if (filename
|| options_non_empty
) {
3681 error_setg(errp
, "Cannot reference an existing block device with "
3682 "additional options or a new filename");
3686 bs
= bdrv_lookup_bs(reference
, reference
, errp
);
3697 /* NULL means an empty set of options */
3698 if (options
== NULL
) {
3699 options
= qdict_new();
3702 /* json: syntax counts as explicit options, as if in the QDict */
3703 parse_json_protocol(options
, &filename
, &local_err
);
3708 bs
->explicit_options
= qdict_clone_shallow(options
);
3711 bool parent_is_format
;
3714 parent_is_format
= parent
->drv
->is_format
;
3717 * parent->drv is not set yet because this node is opened for
3718 * (potential) format probing. That means that @parent is going
3719 * to be a format node.
3721 parent_is_format
= true;
3724 bs
->inherits_from
= parent
;
3725 child_class
->inherit_options(child_role
, parent_is_format
,
3727 parent
->open_flags
, parent
->options
);
3730 ret
= bdrv_fill_options(&options
, filename
, &flags
, &local_err
);
3736 * Set the BDRV_O_RDWR and BDRV_O_ALLOW_RDWR flags.
3737 * Caution: getting a boolean member of @options requires care.
3738 * When @options come from -blockdev or blockdev_add, members are
3739 * typed according to the QAPI schema, but when they come from
3740 * -drive, they're all QString.
3742 if (g_strcmp0(qdict_get_try_str(options
, BDRV_OPT_READ_ONLY
), "on") &&
3743 !qdict_get_try_bool(options
, BDRV_OPT_READ_ONLY
, false)) {
3744 flags
|= (BDRV_O_RDWR
| BDRV_O_ALLOW_RDWR
);
3746 flags
&= ~BDRV_O_RDWR
;
3749 if (flags
& BDRV_O_SNAPSHOT
) {
3750 snapshot_options
= qdict_new();
3751 bdrv_temp_snapshot_options(&snapshot_flags
, snapshot_options
,
3753 /* Let bdrv_backing_options() override "read-only" */
3754 qdict_del(options
, BDRV_OPT_READ_ONLY
);
3755 bdrv_inherited_options(BDRV_CHILD_COW
, true,
3756 &flags
, options
, flags
, options
);
3759 bs
->open_flags
= flags
;
3760 bs
->options
= options
;
3761 options
= qdict_clone_shallow(options
);
3763 /* Find the right image format driver */
3764 /* See cautionary note on accessing @options above */
3765 drvname
= qdict_get_try_str(options
, "driver");
3767 drv
= bdrv_find_format(drvname
);
3769 error_setg(errp
, "Unknown driver: '%s'", drvname
);
3774 assert(drvname
|| !(flags
& BDRV_O_PROTOCOL
));
3776 /* See cautionary note on accessing @options above */
3777 backing
= qdict_get_try_str(options
, "backing");
3778 if (qobject_to(QNull
, qdict_get(options
, "backing")) != NULL
||
3779 (backing
&& *backing
== '\0'))
3782 warn_report("Use of \"backing\": \"\" is deprecated; "
3783 "use \"backing\": null instead");
3785 flags
|= BDRV_O_NO_BACKING
;
3786 qdict_del(bs
->explicit_options
, "backing");
3787 qdict_del(bs
->options
, "backing");
3788 qdict_del(options
, "backing");
3791 /* Open image file without format layer. This BlockBackend is only used for
3792 * probing, the block drivers will do their own bdrv_open_child() for the
3793 * same BDS, which is why we put the node name back into options. */
3794 if ((flags
& BDRV_O_PROTOCOL
) == 0) {
3795 BlockDriverState
*file_bs
;
3797 file_bs
= bdrv_open_child_bs(filename
, options
, "file", bs
,
3798 &child_of_bds
, BDRV_CHILD_IMAGE
,
3803 if (file_bs
!= NULL
) {
3804 /* Not requesting BLK_PERM_CONSISTENT_READ because we're only
3805 * looking at the header to guess the image format. This works even
3806 * in cases where a guest would not see a consistent state. */
3807 file
= blk_new(bdrv_get_aio_context(file_bs
), 0, BLK_PERM_ALL
);
3808 blk_insert_bs(file
, file_bs
, &local_err
);
3809 bdrv_unref(file_bs
);
3814 qdict_put_str(options
, "file", bdrv_get_node_name(file_bs
));
3818 /* Image format probing */
3821 ret
= find_image_format(file
, filename
, &drv
, &local_err
);
3826 * This option update would logically belong in bdrv_fill_options(),
3827 * but we first need to open bs->file for the probing to work, while
3828 * opening bs->file already requires the (mostly) final set of options
3829 * so that cache mode etc. can be inherited.
3831 * Adding the driver later is somewhat ugly, but it's not an option
3832 * that would ever be inherited, so it's correct. We just need to make
3833 * sure to update both bs->options (which has the full effective
3834 * options for bs) and options (which has file.* already removed).
3836 qdict_put_str(bs
->options
, "driver", drv
->format_name
);
3837 qdict_put_str(options
, "driver", drv
->format_name
);
3839 error_setg(errp
, "Must specify either driver or file");
3843 /* BDRV_O_PROTOCOL must be set iff a protocol BDS is about to be created */
3844 assert(!!(flags
& BDRV_O_PROTOCOL
) == !!drv
->bdrv_file_open
);
3845 /* file must be NULL if a protocol BDS is about to be created
3846 * (the inverse results in an error message from bdrv_open_common()) */
3847 assert(!(flags
& BDRV_O_PROTOCOL
) || !file
);
3849 /* Open the image */
3850 ret
= bdrv_open_common(bs
, file
, options
, &local_err
);
3860 /* If there is a backing file, use it */
3861 if ((flags
& BDRV_O_NO_BACKING
) == 0) {
3862 ret
= bdrv_open_backing_file(bs
, options
, "backing", &local_err
);
3864 goto close_and_fail
;
3868 /* Remove all children options and references
3869 * from bs->options and bs->explicit_options */
3870 QLIST_FOREACH(child
, &bs
->children
, next
) {
3871 char *child_key_dot
;
3872 child_key_dot
= g_strdup_printf("%s.", child
->name
);
3873 qdict_extract_subqdict(bs
->explicit_options
, NULL
, child_key_dot
);
3874 qdict_extract_subqdict(bs
->options
, NULL
, child_key_dot
);
3875 qdict_del(bs
->explicit_options
, child
->name
);
3876 qdict_del(bs
->options
, child
->name
);
3877 g_free(child_key_dot
);
3880 /* Check if any unknown options were used */
3881 if (qdict_size(options
) != 0) {
3882 const QDictEntry
*entry
= qdict_first(options
);
3883 if (flags
& BDRV_O_PROTOCOL
) {
3884 error_setg(errp
, "Block protocol '%s' doesn't support the option "
3885 "'%s'", drv
->format_name
, entry
->key
);
3888 "Block format '%s' does not support the option '%s'",
3889 drv
->format_name
, entry
->key
);
3892 goto close_and_fail
;
3895 bdrv_parent_cb_change_media(bs
, true);
3897 qobject_unref(options
);
3900 /* For snapshot=on, create a temporary qcow2 overlay. bs points to the
3901 * temporary snapshot afterwards. */
3902 if (snapshot_flags
) {
3903 BlockDriverState
*snapshot_bs
;
3904 snapshot_bs
= bdrv_append_temp_snapshot(bs
, snapshot_flags
,
3905 snapshot_options
, &local_err
);
3906 snapshot_options
= NULL
;
3908 goto close_and_fail
;
3910 /* We are not going to return bs but the overlay on top of it
3911 * (snapshot_bs); thus, we have to drop the strong reference to bs
3912 * (which we obtained by calling bdrv_new()). bs will not be deleted,
3913 * though, because the overlay still has a reference to it. */
3922 qobject_unref(snapshot_options
);
3923 qobject_unref(bs
->explicit_options
);
3924 qobject_unref(bs
->options
);
3925 qobject_unref(options
);
3927 bs
->explicit_options
= NULL
;
3929 error_propagate(errp
, local_err
);
3934 qobject_unref(snapshot_options
);
3935 qobject_unref(options
);
3936 error_propagate(errp
, local_err
);
3940 BlockDriverState
*bdrv_open(const char *filename
, const char *reference
,
3941 QDict
*options
, int flags
, Error
**errp
)
3943 return bdrv_open_inherit(filename
, reference
, options
, flags
, NULL
,
3947 /* Return true if the NULL-terminated @list contains @str */
3948 static bool is_str_in_list(const char *str
, const char *const *list
)
3952 for (i
= 0; list
[i
] != NULL
; i
++) {
3953 if (!strcmp(str
, list
[i
])) {
3962 * Check that every option set in @bs->options is also set in
3965 * Options listed in the common_options list and in
3966 * @bs->drv->mutable_opts are skipped.
3968 * Return 0 on success, otherwise return -EINVAL and set @errp.
3970 static int bdrv_reset_options_allowed(BlockDriverState
*bs
,
3971 const QDict
*new_opts
, Error
**errp
)
3973 const QDictEntry
*e
;
3974 /* These options are common to all block drivers and are handled
3975 * in bdrv_reopen_prepare() so they can be left out of @new_opts */
3976 const char *const common_options
[] = {
3977 "node-name", "discard", "cache.direct", "cache.no-flush",
3978 "read-only", "auto-read-only", "detect-zeroes", NULL
3981 for (e
= qdict_first(bs
->options
); e
; e
= qdict_next(bs
->options
, e
)) {
3982 if (!qdict_haskey(new_opts
, e
->key
) &&
3983 !is_str_in_list(e
->key
, common_options
) &&
3984 !is_str_in_list(e
->key
, bs
->drv
->mutable_opts
)) {
3985 error_setg(errp
, "Option '%s' cannot be reset "
3986 "to its default value", e
->key
);
3995 * Returns true if @child can be reached recursively from @bs
3997 static bool bdrv_recurse_has_child(BlockDriverState
*bs
,
3998 BlockDriverState
*child
)
4006 QLIST_FOREACH(c
, &bs
->children
, next
) {
4007 if (bdrv_recurse_has_child(c
->bs
, child
)) {
4016 * Adds a BlockDriverState to a simple queue for an atomic, transactional
4017 * reopen of multiple devices.
4019 * bs_queue can either be an existing BlockReopenQueue that has had QTAILQ_INIT
4020 * already performed, or alternatively may be NULL a new BlockReopenQueue will
4021 * be created and initialized. This newly created BlockReopenQueue should be
4022 * passed back in for subsequent calls that are intended to be of the same
4025 * bs is the BlockDriverState to add to the reopen queue.
4027 * options contains the changed options for the associated bs
4028 * (the BlockReopenQueue takes ownership)
4030 * flags contains the open flags for the associated bs
4032 * returns a pointer to bs_queue, which is either the newly allocated
4033 * bs_queue, or the existing bs_queue being used.
4035 * bs must be drained between bdrv_reopen_queue() and bdrv_reopen_multiple().
4037 static BlockReopenQueue
*bdrv_reopen_queue_child(BlockReopenQueue
*bs_queue
,
4038 BlockDriverState
*bs
,
4040 const BdrvChildClass
*klass
,
4042 bool parent_is_format
,
4043 QDict
*parent_options
,
4049 BlockReopenQueueEntry
*bs_entry
;
4051 QDict
*old_options
, *explicit_options
, *options_copy
;
4055 /* Make sure that the caller remembered to use a drained section. This is
4056 * important to avoid graph changes between the recursive queuing here and
4057 * bdrv_reopen_multiple(). */
4058 assert(bs
->quiesce_counter
> 0);
4060 if (bs_queue
== NULL
) {
4061 bs_queue
= g_new0(BlockReopenQueue
, 1);
4062 QTAILQ_INIT(bs_queue
);
4066 options
= qdict_new();
4069 /* Check if this BlockDriverState is already in the queue */
4070 QTAILQ_FOREACH(bs_entry
, bs_queue
, entry
) {
4071 if (bs
== bs_entry
->state
.bs
) {
4077 * Precedence of options:
4078 * 1. Explicitly passed in options (highest)
4079 * 2. Retained from explicitly set options of bs
4080 * 3. Inherited from parent node
4081 * 4. Retained from effective options of bs
4084 /* Old explicitly set values (don't overwrite by inherited value) */
4085 if (bs_entry
|| keep_old_opts
) {
4086 old_options
= qdict_clone_shallow(bs_entry
?
4087 bs_entry
->state
.explicit_options
:
4088 bs
->explicit_options
);
4089 bdrv_join_options(bs
, options
, old_options
);
4090 qobject_unref(old_options
);
4093 explicit_options
= qdict_clone_shallow(options
);
4095 /* Inherit from parent node */
4096 if (parent_options
) {
4098 klass
->inherit_options(role
, parent_is_format
, &flags
, options
,
4099 parent_flags
, parent_options
);
4101 flags
= bdrv_get_flags(bs
);
4104 if (keep_old_opts
) {
4105 /* Old values are used for options that aren't set yet */
4106 old_options
= qdict_clone_shallow(bs
->options
);
4107 bdrv_join_options(bs
, options
, old_options
);
4108 qobject_unref(old_options
);
4111 /* We have the final set of options so let's update the flags */
4112 options_copy
= qdict_clone_shallow(options
);
4113 opts
= qemu_opts_create(&bdrv_runtime_opts
, NULL
, 0, &error_abort
);
4114 qemu_opts_absorb_qdict(opts
, options_copy
, NULL
);
4115 update_flags_from_options(&flags
, opts
);
4116 qemu_opts_del(opts
);
4117 qobject_unref(options_copy
);
4119 /* bdrv_open_inherit() sets and clears some additional flags internally */
4120 flags
&= ~BDRV_O_PROTOCOL
;
4121 if (flags
& BDRV_O_RDWR
) {
4122 flags
|= BDRV_O_ALLOW_RDWR
;
4126 bs_entry
= g_new0(BlockReopenQueueEntry
, 1);
4127 QTAILQ_INSERT_TAIL(bs_queue
, bs_entry
, entry
);
4129 qobject_unref(bs_entry
->state
.options
);
4130 qobject_unref(bs_entry
->state
.explicit_options
);
4133 bs_entry
->state
.bs
= bs
;
4134 bs_entry
->state
.options
= options
;
4135 bs_entry
->state
.explicit_options
= explicit_options
;
4136 bs_entry
->state
.flags
= flags
;
4138 /* This needs to be overwritten in bdrv_reopen_prepare() */
4139 bs_entry
->state
.perm
= UINT64_MAX
;
4140 bs_entry
->state
.shared_perm
= 0;
4143 * If keep_old_opts is false then it means that unspecified
4144 * options must be reset to their original value. We don't allow
4145 * resetting 'backing' but we need to know if the option is
4146 * missing in order to decide if we have to return an error.
4148 if (!keep_old_opts
) {
4149 bs_entry
->state
.backing_missing
=
4150 !qdict_haskey(options
, "backing") &&
4151 !qdict_haskey(options
, "backing.driver");
4154 QLIST_FOREACH(child
, &bs
->children
, next
) {
4155 QDict
*new_child_options
= NULL
;
4156 bool child_keep_old
= keep_old_opts
;
4158 /* reopen can only change the options of block devices that were
4159 * implicitly created and inherited options. For other (referenced)
4160 * block devices, a syntax like "backing.foo" results in an error. */
4161 if (child
->bs
->inherits_from
!= bs
) {
4165 /* Check if the options contain a child reference */
4166 if (qdict_haskey(options
, child
->name
)) {
4167 const char *childref
= qdict_get_try_str(options
, child
->name
);
4169 * The current child must not be reopened if the child
4170 * reference is null or points to a different node.
4172 if (g_strcmp0(childref
, child
->bs
->node_name
)) {
4176 * If the child reference points to the current child then
4177 * reopen it with its existing set of options (note that
4178 * it can still inherit new options from the parent).
4180 child_keep_old
= true;
4182 /* Extract child options ("child-name.*") */
4183 char *child_key_dot
= g_strdup_printf("%s.", child
->name
);
4184 qdict_extract_subqdict(explicit_options
, NULL
, child_key_dot
);
4185 qdict_extract_subqdict(options
, &new_child_options
, child_key_dot
);
4186 g_free(child_key_dot
);
4189 bdrv_reopen_queue_child(bs_queue
, child
->bs
, new_child_options
,
4190 child
->klass
, child
->role
, bs
->drv
->is_format
,
4191 options
, flags
, child_keep_old
);
4197 BlockReopenQueue
*bdrv_reopen_queue(BlockReopenQueue
*bs_queue
,
4198 BlockDriverState
*bs
,
4199 QDict
*options
, bool keep_old_opts
)
4201 return bdrv_reopen_queue_child(bs_queue
, bs
, options
, NULL
, 0, false,
4202 NULL
, 0, keep_old_opts
);
4206 * Reopen multiple BlockDriverStates atomically & transactionally.
4208 * The queue passed in (bs_queue) must have been built up previous
4209 * via bdrv_reopen_queue().
4211 * Reopens all BDS specified in the queue, with the appropriate
4212 * flags. All devices are prepared for reopen, and failure of any
4213 * device will cause all device changes to be abandoned, and intermediate
4216 * If all devices prepare successfully, then the changes are committed
4219 * All affected nodes must be drained between bdrv_reopen_queue() and
4220 * bdrv_reopen_multiple().
4222 int bdrv_reopen_multiple(BlockReopenQueue
*bs_queue
, Error
**errp
)
4225 BlockReopenQueueEntry
*bs_entry
, *next
;
4227 assert(bs_queue
!= NULL
);
4229 QTAILQ_FOREACH(bs_entry
, bs_queue
, entry
) {
4230 assert(bs_entry
->state
.bs
->quiesce_counter
> 0);
4231 if (bdrv_reopen_prepare(&bs_entry
->state
, bs_queue
, errp
)) {
4234 bs_entry
->prepared
= true;
4237 QTAILQ_FOREACH(bs_entry
, bs_queue
, entry
) {
4238 BDRVReopenState
*state
= &bs_entry
->state
;
4239 ret
= bdrv_check_perm(state
->bs
, bs_queue
, state
->perm
,
4240 state
->shared_perm
, NULL
, errp
);
4244 /* Check if new_backing_bs would accept the new permissions */
4245 if (state
->replace_backing_bs
&& state
->new_backing_bs
) {
4246 uint64_t nperm
, nshared
;
4247 bdrv_child_perm(state
->bs
, state
->new_backing_bs
,
4248 NULL
, bdrv_backing_role(state
->bs
),
4249 bs_queue
, state
->perm
, state
->shared_perm
,
4251 ret
= bdrv_check_update_perm(state
->new_backing_bs
, NULL
,
4252 nperm
, nshared
, NULL
, errp
);
4257 bs_entry
->perms_checked
= true;
4261 * If we reach this point, we have success and just need to apply the
4264 * Reverse order is used to comfort qcow2 driver: on commit it need to write
4265 * IN_USE flag to the image, to mark bitmaps in the image as invalid. But
4266 * children are usually goes after parents in reopen-queue, so go from last
4269 QTAILQ_FOREACH_REVERSE(bs_entry
, bs_queue
, entry
) {
4270 bdrv_reopen_commit(&bs_entry
->state
);
4275 QTAILQ_FOREACH_SAFE(bs_entry
, bs_queue
, entry
, next
) {
4276 BDRVReopenState
*state
= &bs_entry
->state
;
4278 if (!bs_entry
->perms_checked
) {
4283 uint64_t perm
, shared
;
4285 bdrv_get_cumulative_perm(state
->bs
, &perm
, &shared
);
4286 assert(perm
== state
->perm
);
4287 assert(shared
== state
->shared_perm
);
4289 bdrv_set_perm(state
->bs
);
4291 bdrv_abort_perm_update(state
->bs
);
4292 if (state
->replace_backing_bs
&& state
->new_backing_bs
) {
4293 bdrv_abort_perm_update(state
->new_backing_bs
);
4299 QTAILQ_FOREACH_REVERSE(bs_entry
, bs_queue
, entry
) {
4300 BlockDriverState
*bs
= bs_entry
->state
.bs
;
4302 if (bs
->drv
->bdrv_reopen_commit_post
)
4303 bs
->drv
->bdrv_reopen_commit_post(&bs_entry
->state
);
4307 QTAILQ_FOREACH_SAFE(bs_entry
, bs_queue
, entry
, next
) {
4309 if (bs_entry
->prepared
) {
4310 bdrv_reopen_abort(&bs_entry
->state
);
4312 qobject_unref(bs_entry
->state
.explicit_options
);
4313 qobject_unref(bs_entry
->state
.options
);
4315 if (bs_entry
->state
.new_backing_bs
) {
4316 bdrv_unref(bs_entry
->state
.new_backing_bs
);
4325 int bdrv_reopen_set_read_only(BlockDriverState
*bs
, bool read_only
,
4329 BlockReopenQueue
*queue
;
4330 QDict
*opts
= qdict_new();
4332 qdict_put_bool(opts
, BDRV_OPT_READ_ONLY
, read_only
);
4334 bdrv_subtree_drained_begin(bs
);
4335 queue
= bdrv_reopen_queue(NULL
, bs
, opts
, true);
4336 ret
= bdrv_reopen_multiple(queue
, errp
);
4337 bdrv_subtree_drained_end(bs
);
4342 static BlockReopenQueueEntry
*find_parent_in_reopen_queue(BlockReopenQueue
*q
,
4345 BlockReopenQueueEntry
*entry
;
4347 QTAILQ_FOREACH(entry
, q
, entry
) {
4348 BlockDriverState
*bs
= entry
->state
.bs
;
4351 QLIST_FOREACH(child
, &bs
->children
, next
) {
4361 static void bdrv_reopen_perm(BlockReopenQueue
*q
, BlockDriverState
*bs
,
4362 uint64_t *perm
, uint64_t *shared
)
4365 BlockReopenQueueEntry
*parent
;
4366 uint64_t cumulative_perms
= 0;
4367 uint64_t cumulative_shared_perms
= BLK_PERM_ALL
;
4369 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
4370 parent
= find_parent_in_reopen_queue(q
, c
);
4372 cumulative_perms
|= c
->perm
;
4373 cumulative_shared_perms
&= c
->shared_perm
;
4375 uint64_t nperm
, nshared
;
4377 bdrv_child_perm(parent
->state
.bs
, bs
, c
, c
->role
, q
,
4378 parent
->state
.perm
, parent
->state
.shared_perm
,
4381 cumulative_perms
|= nperm
;
4382 cumulative_shared_perms
&= nshared
;
4385 *perm
= cumulative_perms
;
4386 *shared
= cumulative_shared_perms
;
4389 static bool bdrv_reopen_can_attach(BlockDriverState
*parent
,
4391 BlockDriverState
*new_child
,
4394 AioContext
*parent_ctx
= bdrv_get_aio_context(parent
);
4395 AioContext
*child_ctx
= bdrv_get_aio_context(new_child
);
4399 ignore
= g_slist_prepend(NULL
, child
);
4400 ret
= bdrv_can_set_aio_context(new_child
, parent_ctx
, &ignore
, NULL
);
4401 g_slist_free(ignore
);
4406 ignore
= g_slist_prepend(NULL
, child
);
4407 ret
= bdrv_can_set_aio_context(parent
, child_ctx
, &ignore
, errp
);
4408 g_slist_free(ignore
);
4413 * Take a BDRVReopenState and check if the value of 'backing' in the
4414 * reopen_state->options QDict is valid or not.
4416 * If 'backing' is missing from the QDict then return 0.
4418 * If 'backing' contains the node name of the backing file of
4419 * reopen_state->bs then return 0.
4421 * If 'backing' contains a different node name (or is null) then check
4422 * whether the current backing file can be replaced with the new one.
4423 * If that's the case then reopen_state->replace_backing_bs is set to
4424 * true and reopen_state->new_backing_bs contains a pointer to the new
4425 * backing BlockDriverState (or NULL).
4427 * Return 0 on success, otherwise return < 0 and set @errp.
4429 static int bdrv_reopen_parse_backing(BDRVReopenState
*reopen_state
,
4432 BlockDriverState
*bs
= reopen_state
->bs
;
4433 BlockDriverState
*overlay_bs
, *below_bs
, *new_backing_bs
;
4437 value
= qdict_get(reopen_state
->options
, "backing");
4438 if (value
== NULL
) {
4442 switch (qobject_type(value
)) {
4444 new_backing_bs
= NULL
;
4447 str
= qstring_get_str(qobject_to(QString
, value
));
4448 new_backing_bs
= bdrv_lookup_bs(NULL
, str
, errp
);
4449 if (new_backing_bs
== NULL
) {
4451 } else if (bdrv_recurse_has_child(new_backing_bs
, bs
)) {
4452 error_setg(errp
, "Making '%s' a backing file of '%s' "
4453 "would create a cycle", str
, bs
->node_name
);
4458 /* 'backing' does not allow any other data type */
4459 g_assert_not_reached();
4463 * Check AioContext compatibility so that the bdrv_set_backing_hd() call in
4464 * bdrv_reopen_commit() won't fail.
4466 if (new_backing_bs
) {
4467 if (!bdrv_reopen_can_attach(bs
, bs
->backing
, new_backing_bs
, errp
)) {
4473 * Ensure that @bs can really handle backing files, because we are
4474 * about to give it one (or swap the existing one)
4476 if (bs
->drv
->is_filter
) {
4477 /* Filters always have a file or a backing child */
4479 error_setg(errp
, "'%s' is a %s filter node that does not support a "
4480 "backing child", bs
->node_name
, bs
->drv
->format_name
);
4483 } else if (!bs
->drv
->supports_backing
) {
4484 error_setg(errp
, "Driver '%s' of node '%s' does not support backing "
4485 "files", bs
->drv
->format_name
, bs
->node_name
);
4490 * Find the "actual" backing file by skipping all links that point
4491 * to an implicit node, if any (e.g. a commit filter node).
4492 * We cannot use any of the bdrv_skip_*() functions here because
4493 * those return the first explicit node, while we are looking for
4497 for (below_bs
= bdrv_filter_or_cow_bs(overlay_bs
);
4498 below_bs
&& below_bs
->implicit
;
4499 below_bs
= bdrv_filter_or_cow_bs(overlay_bs
))
4501 overlay_bs
= below_bs
;
4504 /* If we want to replace the backing file we need some extra checks */
4505 if (new_backing_bs
!= bdrv_filter_or_cow_bs(overlay_bs
)) {
4506 /* Check for implicit nodes between bs and its backing file */
4507 if (bs
!= overlay_bs
) {
4508 error_setg(errp
, "Cannot change backing link if '%s' has "
4509 "an implicit backing file", bs
->node_name
);
4513 * Check if the backing link that we want to replace is frozen.
4515 * bdrv_filter_or_cow_child(overlay_bs) == overlay_bs->backing,
4516 * because we know that overlay_bs == bs, and that @bs
4517 * either is a filter that uses ->backing or a COW format BDS
4518 * with bs->drv->supports_backing == true.
4520 if (bdrv_is_backing_chain_frozen(overlay_bs
,
4521 child_bs(overlay_bs
->backing
), errp
))
4525 reopen_state
->replace_backing_bs
= true;
4526 if (new_backing_bs
) {
4527 bdrv_ref(new_backing_bs
);
4528 reopen_state
->new_backing_bs
= new_backing_bs
;
4536 * Prepares a BlockDriverState for reopen. All changes are staged in the
4537 * 'opaque' field of the BDRVReopenState, which is used and allocated by
4538 * the block driver layer .bdrv_reopen_prepare()
4540 * bs is the BlockDriverState to reopen
4541 * flags are the new open flags
4542 * queue is the reopen queue
4544 * Returns 0 on success, non-zero on error. On error errp will be set
4547 * On failure, bdrv_reopen_abort() will be called to clean up any data.
4548 * It is the responsibility of the caller to then call the abort() or
4549 * commit() for any other BDS that have been left in a prepare() state
4552 static int bdrv_reopen_prepare(BDRVReopenState
*reopen_state
,
4553 BlockReopenQueue
*queue
, Error
**errp
)
4557 Error
*local_err
= NULL
;
4560 QDict
*orig_reopen_opts
;
4561 char *discard
= NULL
;
4563 bool drv_prepared
= false;
4565 assert(reopen_state
!= NULL
);
4566 assert(reopen_state
->bs
->drv
!= NULL
);
4567 drv
= reopen_state
->bs
->drv
;
4569 /* This function and each driver's bdrv_reopen_prepare() remove
4570 * entries from reopen_state->options as they are processed, so
4571 * we need to make a copy of the original QDict. */
4572 orig_reopen_opts
= qdict_clone_shallow(reopen_state
->options
);
4574 /* Process generic block layer options */
4575 opts
= qemu_opts_create(&bdrv_runtime_opts
, NULL
, 0, &error_abort
);
4576 if (!qemu_opts_absorb_qdict(opts
, reopen_state
->options
, errp
)) {
4581 /* This was already called in bdrv_reopen_queue_child() so the flags
4582 * are up-to-date. This time we simply want to remove the options from
4583 * QemuOpts in order to indicate that they have been processed. */
4584 old_flags
= reopen_state
->flags
;
4585 update_flags_from_options(&reopen_state
->flags
, opts
);
4586 assert(old_flags
== reopen_state
->flags
);
4588 discard
= qemu_opt_get_del(opts
, BDRV_OPT_DISCARD
);
4589 if (discard
!= NULL
) {
4590 if (bdrv_parse_discard_flags(discard
, &reopen_state
->flags
) != 0) {
4591 error_setg(errp
, "Invalid discard option");
4597 reopen_state
->detect_zeroes
=
4598 bdrv_parse_detect_zeroes(opts
, reopen_state
->flags
, &local_err
);
4600 error_propagate(errp
, local_err
);
4605 /* All other options (including node-name and driver) must be unchanged.
4606 * Put them back into the QDict, so that they are checked at the end
4607 * of this function. */
4608 qemu_opts_to_qdict(opts
, reopen_state
->options
);
4610 /* If we are to stay read-only, do not allow permission change
4611 * to r/w. Attempting to set to r/w may fail if either BDRV_O_ALLOW_RDWR is
4612 * not set, or if the BDS still has copy_on_read enabled */
4613 read_only
= !(reopen_state
->flags
& BDRV_O_RDWR
);
4614 ret
= bdrv_can_set_read_only(reopen_state
->bs
, read_only
, true, &local_err
);
4616 error_propagate(errp
, local_err
);
4620 /* Calculate required permissions after reopening */
4621 bdrv_reopen_perm(queue
, reopen_state
->bs
,
4622 &reopen_state
->perm
, &reopen_state
->shared_perm
);
4624 ret
= bdrv_flush(reopen_state
->bs
);
4626 error_setg_errno(errp
, -ret
, "Error flushing drive");
4630 if (drv
->bdrv_reopen_prepare
) {
4632 * If a driver-specific option is missing, it means that we
4633 * should reset it to its default value.
4634 * But not all options allow that, so we need to check it first.
4636 ret
= bdrv_reset_options_allowed(reopen_state
->bs
,
4637 reopen_state
->options
, errp
);
4642 ret
= drv
->bdrv_reopen_prepare(reopen_state
, queue
, &local_err
);
4644 if (local_err
!= NULL
) {
4645 error_propagate(errp
, local_err
);
4647 bdrv_refresh_filename(reopen_state
->bs
);
4648 error_setg(errp
, "failed while preparing to reopen image '%s'",
4649 reopen_state
->bs
->filename
);
4654 /* It is currently mandatory to have a bdrv_reopen_prepare()
4655 * handler for each supported drv. */
4656 error_setg(errp
, "Block format '%s' used by node '%s' "
4657 "does not support reopening files", drv
->format_name
,
4658 bdrv_get_device_or_node_name(reopen_state
->bs
));
4663 drv_prepared
= true;
4666 * We must provide the 'backing' option if the BDS has a backing
4667 * file or if the image file has a backing file name as part of
4668 * its metadata. Otherwise the 'backing' option can be omitted.
4670 if (drv
->supports_backing
&& reopen_state
->backing_missing
&&
4671 (reopen_state
->bs
->backing
|| reopen_state
->bs
->backing_file
[0])) {
4672 error_setg(errp
, "backing is missing for '%s'",
4673 reopen_state
->bs
->node_name
);
4679 * Allow changing the 'backing' option. The new value can be
4680 * either a reference to an existing node (using its node name)
4681 * or NULL to simply detach the current backing file.
4683 ret
= bdrv_reopen_parse_backing(reopen_state
, errp
);
4687 qdict_del(reopen_state
->options
, "backing");
4689 /* Options that are not handled are only okay if they are unchanged
4690 * compared to the old state. It is expected that some options are only
4691 * used for the initial open, but not reopen (e.g. filename) */
4692 if (qdict_size(reopen_state
->options
)) {
4693 const QDictEntry
*entry
= qdict_first(reopen_state
->options
);
4696 QObject
*new = entry
->value
;
4697 QObject
*old
= qdict_get(reopen_state
->bs
->options
, entry
->key
);
4699 /* Allow child references (child_name=node_name) as long as they
4700 * point to the current child (i.e. everything stays the same). */
4701 if (qobject_type(new) == QTYPE_QSTRING
) {
4703 QLIST_FOREACH(child
, &reopen_state
->bs
->children
, next
) {
4704 if (!strcmp(child
->name
, entry
->key
)) {
4710 if (!strcmp(child
->bs
->node_name
,
4711 qstring_get_str(qobject_to(QString
, new)))) {
4712 continue; /* Found child with this name, skip option */
4718 * TODO: When using -drive to specify blockdev options, all values
4719 * will be strings; however, when using -blockdev, blockdev-add or
4720 * filenames using the json:{} pseudo-protocol, they will be
4722 * In contrast, reopening options are (currently) always strings
4723 * (because you can only specify them through qemu-io; all other
4724 * callers do not specify any options).
4725 * Therefore, when using anything other than -drive to create a BDS,
4726 * this cannot detect non-string options as unchanged, because
4727 * qobject_is_equal() always returns false for objects of different
4728 * type. In the future, this should be remedied by correctly typing
4729 * all options. For now, this is not too big of an issue because
4730 * the user can simply omit options which cannot be changed anyway,
4731 * so they will stay unchanged.
4733 if (!qobject_is_equal(new, old
)) {
4734 error_setg(errp
, "Cannot change the option '%s'", entry
->key
);
4738 } while ((entry
= qdict_next(reopen_state
->options
, entry
)));
4743 /* Restore the original reopen_state->options QDict */
4744 qobject_unref(reopen_state
->options
);
4745 reopen_state
->options
= qobject_ref(orig_reopen_opts
);
4748 if (ret
< 0 && drv_prepared
) {
4749 /* drv->bdrv_reopen_prepare() has succeeded, so we need to
4750 * call drv->bdrv_reopen_abort() before signaling an error
4751 * (bdrv_reopen_multiple() will not call bdrv_reopen_abort()
4752 * when the respective bdrv_reopen_prepare() has failed) */
4753 if (drv
->bdrv_reopen_abort
) {
4754 drv
->bdrv_reopen_abort(reopen_state
);
4757 qemu_opts_del(opts
);
4758 qobject_unref(orig_reopen_opts
);
4764 * Takes the staged changes for the reopen from bdrv_reopen_prepare(), and
4765 * makes them final by swapping the staging BlockDriverState contents into
4766 * the active BlockDriverState contents.
4768 static void bdrv_reopen_commit(BDRVReopenState
*reopen_state
)
4771 BlockDriverState
*bs
;
4774 assert(reopen_state
!= NULL
);
4775 bs
= reopen_state
->bs
;
4777 assert(drv
!= NULL
);
4779 /* If there are any driver level actions to take */
4780 if (drv
->bdrv_reopen_commit
) {
4781 drv
->bdrv_reopen_commit(reopen_state
);
4784 /* set BDS specific flags now */
4785 qobject_unref(bs
->explicit_options
);
4786 qobject_unref(bs
->options
);
4788 bs
->explicit_options
= reopen_state
->explicit_options
;
4789 bs
->options
= reopen_state
->options
;
4790 bs
->open_flags
= reopen_state
->flags
;
4791 bs
->read_only
= !(reopen_state
->flags
& BDRV_O_RDWR
);
4792 bs
->detect_zeroes
= reopen_state
->detect_zeroes
;
4794 if (reopen_state
->replace_backing_bs
) {
4795 qdict_del(bs
->explicit_options
, "backing");
4796 qdict_del(bs
->options
, "backing");
4799 /* Remove child references from bs->options and bs->explicit_options.
4800 * Child options were already removed in bdrv_reopen_queue_child() */
4801 QLIST_FOREACH(child
, &bs
->children
, next
) {
4802 qdict_del(bs
->explicit_options
, child
->name
);
4803 qdict_del(bs
->options
, child
->name
);
4807 * Change the backing file if a new one was specified. We do this
4808 * after updating bs->options, so bdrv_refresh_filename() (called
4809 * from bdrv_set_backing_hd()) has the new values.
4811 if (reopen_state
->replace_backing_bs
) {
4812 BlockDriverState
*old_backing_bs
= child_bs(bs
->backing
);
4813 assert(!old_backing_bs
|| !old_backing_bs
->implicit
);
4814 /* Abort the permission update on the backing bs we're detaching */
4815 if (old_backing_bs
) {
4816 bdrv_abort_perm_update(old_backing_bs
);
4818 bdrv_set_backing_hd(bs
, reopen_state
->new_backing_bs
, &error_abort
);
4821 bdrv_refresh_limits(bs
, NULL
);
4825 * Abort the reopen, and delete and free the staged changes in
4828 static void bdrv_reopen_abort(BDRVReopenState
*reopen_state
)
4832 assert(reopen_state
!= NULL
);
4833 drv
= reopen_state
->bs
->drv
;
4834 assert(drv
!= NULL
);
4836 if (drv
->bdrv_reopen_abort
) {
4837 drv
->bdrv_reopen_abort(reopen_state
);
4842 static void bdrv_close(BlockDriverState
*bs
)
4844 BdrvAioNotifier
*ban
, *ban_next
;
4845 BdrvChild
*child
, *next
;
4847 assert(!bs
->refcnt
);
4849 bdrv_drained_begin(bs
); /* complete I/O */
4851 bdrv_drain(bs
); /* in case flush left pending I/O */
4854 if (bs
->drv
->bdrv_close
) {
4855 /* Must unfreeze all children, so bdrv_unref_child() works */
4856 bs
->drv
->bdrv_close(bs
);
4861 QLIST_FOREACH_SAFE(child
, &bs
->children
, next
, next
) {
4862 bdrv_unref_child(bs
, child
);
4869 qatomic_set(&bs
->copy_on_read
, 0);
4870 bs
->backing_file
[0] = '\0';
4871 bs
->backing_format
[0] = '\0';
4872 bs
->total_sectors
= 0;
4873 bs
->encrypted
= false;
4875 qobject_unref(bs
->options
);
4876 qobject_unref(bs
->explicit_options
);
4878 bs
->explicit_options
= NULL
;
4879 qobject_unref(bs
->full_open_options
);
4880 bs
->full_open_options
= NULL
;
4882 bdrv_release_named_dirty_bitmaps(bs
);
4883 assert(QLIST_EMPTY(&bs
->dirty_bitmaps
));
4885 QLIST_FOREACH_SAFE(ban
, &bs
->aio_notifiers
, list
, ban_next
) {
4888 QLIST_INIT(&bs
->aio_notifiers
);
4889 bdrv_drained_end(bs
);
4892 * If we're still inside some bdrv_drain_all_begin()/end() sections, end
4893 * them now since this BDS won't exist anymore when bdrv_drain_all_end()
4896 if (bs
->quiesce_counter
) {
4897 bdrv_drain_all_end_quiesce(bs
);
4901 void bdrv_close_all(void)
4903 assert(job_next(NULL
) == NULL
);
4905 /* Drop references from requests still in flight, such as canceled block
4906 * jobs whose AIO context has not been polled yet */
4909 blk_remove_all_bs();
4910 blockdev_close_all_bdrv_states();
4912 assert(QTAILQ_EMPTY(&all_bdrv_states
));
4915 static bool should_update_child(BdrvChild
*c
, BlockDriverState
*to
)
4921 if (c
->klass
->stay_at_node
) {
4925 /* If the child @c belongs to the BDS @to, replacing the current
4926 * c->bs by @to would mean to create a loop.
4928 * Such a case occurs when appending a BDS to a backing chain.
4929 * For instance, imagine the following chain:
4931 * guest device -> node A -> further backing chain...
4933 * Now we create a new BDS B which we want to put on top of this
4934 * chain, so we first attach A as its backing node:
4939 * guest device -> node A -> further backing chain...
4941 * Finally we want to replace A by B. When doing that, we want to
4942 * replace all pointers to A by pointers to B -- except for the
4943 * pointer from B because (1) that would create a loop, and (2)
4944 * that pointer should simply stay intact:
4946 * guest device -> node B
4949 * node A -> further backing chain...
4951 * In general, when replacing a node A (c->bs) by a node B (@to),
4952 * if A is a child of B, that means we cannot replace A by B there
4953 * because that would create a loop. Silently detaching A from B
4954 * is also not really an option. So overall just leaving A in
4955 * place there is the most sensible choice.
4957 * We would also create a loop in any cases where @c is only
4958 * indirectly referenced by @to. Prevent this by returning false
4959 * if @c is found (by breadth-first search) anywhere in the whole
4964 found
= g_hash_table_new(NULL
, NULL
);
4965 g_hash_table_add(found
, to
);
4966 queue
= g_queue_new();
4967 g_queue_push_tail(queue
, to
);
4969 while (!g_queue_is_empty(queue
)) {
4970 BlockDriverState
*v
= g_queue_pop_head(queue
);
4973 QLIST_FOREACH(c2
, &v
->children
, next
) {
4979 if (g_hash_table_contains(found
, c2
->bs
)) {
4983 g_queue_push_tail(queue
, c2
->bs
);
4984 g_hash_table_add(found
, c2
->bs
);
4988 g_queue_free(queue
);
4989 g_hash_table_destroy(found
);
4994 static int bdrv_replace_node_noperm(BlockDriverState
*from
,
4995 BlockDriverState
*to
,
4996 bool auto_skip
, Transaction
*tran
,
4999 BdrvChild
*c
, *next
;
5001 QLIST_FOREACH_SAFE(c
, &from
->parents
, next_parent
, next
) {
5002 assert(c
->bs
== from
);
5003 if (!should_update_child(c
, to
)) {
5007 error_setg(errp
, "Should not change '%s' link to '%s'",
5008 c
->name
, from
->node_name
);
5012 error_setg(errp
, "Cannot change '%s' link to '%s'",
5013 c
->name
, from
->node_name
);
5016 bdrv_replace_child_safe(c
, to
, tran
);
5023 * With auto_skip=true bdrv_replace_node_common skips updating from parents
5024 * if it creates a parent-child relation loop or if parent is block-job.
5026 * With auto_skip=false the error is returned if from has a parent which should
5029 static int bdrv_replace_node_common(BlockDriverState
*from
,
5030 BlockDriverState
*to
,
5031 bool auto_skip
, Error
**errp
)
5033 Transaction
*tran
= tran_new();
5034 g_autoptr(GHashTable
) found
= NULL
;
5035 g_autoptr(GSList
) refresh_list
= NULL
;
5038 /* Make sure that @from doesn't go away until we have successfully attached
5039 * all of its parents to @to. */
5042 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
5043 assert(bdrv_get_aio_context(from
) == bdrv_get_aio_context(to
));
5044 bdrv_drained_begin(from
);
5047 * Do the replacement without permission update.
5048 * Replacement may influence the permissions, we should calculate new
5049 * permissions based on new graph. If we fail, we'll roll-back the
5052 ret
= bdrv_replace_node_noperm(from
, to
, auto_skip
, tran
, errp
);
5057 found
= g_hash_table_new(NULL
, NULL
);
5059 refresh_list
= bdrv_topological_dfs(refresh_list
, found
, to
);
5060 refresh_list
= bdrv_topological_dfs(refresh_list
, found
, from
);
5062 ret
= bdrv_list_refresh_perms(refresh_list
, NULL
, tran
, errp
);
5070 tran_finalize(tran
, ret
);
5072 bdrv_drained_end(from
);
5078 int bdrv_replace_node(BlockDriverState
*from
, BlockDriverState
*to
,
5081 return bdrv_replace_node_common(from
, to
, true, errp
);
5085 * Add new bs contents at the top of an image chain while the chain is
5086 * live, while keeping required fields on the top layer.
5088 * This will modify the BlockDriverState fields, and swap contents
5089 * between bs_new and bs_top. Both bs_new and bs_top are modified.
5091 * bs_new must not be attached to a BlockBackend and must not have backing
5094 * This function does not create any image files.
5096 int bdrv_append(BlockDriverState
*bs_new
, BlockDriverState
*bs_top
,
5100 Transaction
*tran
= tran_new();
5102 assert(!bs_new
->backing
);
5104 ret
= bdrv_attach_child_noperm(bs_new
, bs_top
, "backing",
5105 &child_of_bds
, bdrv_backing_role(bs_new
),
5106 &bs_new
->backing
, tran
, errp
);
5111 ret
= bdrv_replace_node_noperm(bs_top
, bs_new
, true, tran
, errp
);
5116 ret
= bdrv_refresh_perms(bs_new
, errp
);
5118 tran_finalize(tran
, ret
);
5120 bdrv_refresh_limits(bs_top
, NULL
);
5125 static void bdrv_delete(BlockDriverState
*bs
)
5127 assert(bdrv_op_blocker_is_empty(bs
));
5128 assert(!bs
->refcnt
);
5130 /* remove from list, if necessary */
5131 if (bs
->node_name
[0] != '\0') {
5132 QTAILQ_REMOVE(&graph_bdrv_states
, bs
, node_list
);
5134 QTAILQ_REMOVE(&all_bdrv_states
, bs
, bs_list
);
5141 BlockDriverState
*bdrv_insert_node(BlockDriverState
*bs
, QDict
*node_options
,
5142 int flags
, Error
**errp
)
5144 BlockDriverState
*new_node_bs
;
5145 Error
*local_err
= NULL
;
5147 new_node_bs
= bdrv_open(NULL
, NULL
, node_options
, flags
, errp
);
5148 if (new_node_bs
== NULL
) {
5149 error_prepend(errp
, "Could not create node: ");
5153 bdrv_drained_begin(bs
);
5154 bdrv_replace_node(bs
, new_node_bs
, &local_err
);
5155 bdrv_drained_end(bs
);
5158 bdrv_unref(new_node_bs
);
5159 error_propagate(errp
, local_err
);
5167 * Run consistency checks on an image
5169 * Returns 0 if the check could be completed (it doesn't mean that the image is
5170 * free of errors) or -errno when an internal error occurred. The results of the
5171 * check are stored in res.
5173 int coroutine_fn
bdrv_co_check(BlockDriverState
*bs
,
5174 BdrvCheckResult
*res
, BdrvCheckMode fix
)
5176 if (bs
->drv
== NULL
) {
5179 if (bs
->drv
->bdrv_co_check
== NULL
) {
5183 memset(res
, 0, sizeof(*res
));
5184 return bs
->drv
->bdrv_co_check(bs
, res
, fix
);
5190 * -EINVAL - backing format specified, but no file
5191 * -ENOSPC - can't update the backing file because no space is left in the
5193 * -ENOTSUP - format driver doesn't support changing the backing file
5195 int bdrv_change_backing_file(BlockDriverState
*bs
, const char *backing_file
,
5196 const char *backing_fmt
, bool warn
)
5198 BlockDriver
*drv
= bs
->drv
;
5205 /* Backing file format doesn't make sense without a backing file */
5206 if (backing_fmt
&& !backing_file
) {
5210 if (warn
&& backing_file
&& !backing_fmt
) {
5211 warn_report("Deprecated use of backing file without explicit "
5212 "backing format, use of this image requires "
5213 "potentially unsafe format probing");
5216 if (drv
->bdrv_change_backing_file
!= NULL
) {
5217 ret
= drv
->bdrv_change_backing_file(bs
, backing_file
, backing_fmt
);
5223 pstrcpy(bs
->backing_file
, sizeof(bs
->backing_file
), backing_file
?: "");
5224 pstrcpy(bs
->backing_format
, sizeof(bs
->backing_format
), backing_fmt
?: "");
5225 pstrcpy(bs
->auto_backing_file
, sizeof(bs
->auto_backing_file
),
5226 backing_file
?: "");
5232 * Finds the first non-filter node above bs in the chain between
5233 * active and bs. The returned node is either an immediate parent of
5234 * bs, or there are only filter nodes between the two.
5236 * Returns NULL if bs is not found in active's image chain,
5237 * or if active == bs.
5239 * Returns the bottommost base image if bs == NULL.
5241 BlockDriverState
*bdrv_find_overlay(BlockDriverState
*active
,
5242 BlockDriverState
*bs
)
5244 bs
= bdrv_skip_filters(bs
);
5245 active
= bdrv_skip_filters(active
);
5248 BlockDriverState
*next
= bdrv_backing_chain_next(active
);
5258 /* Given a BDS, searches for the base layer. */
5259 BlockDriverState
*bdrv_find_base(BlockDriverState
*bs
)
5261 return bdrv_find_overlay(bs
, NULL
);
5265 * Return true if at least one of the COW (backing) and filter links
5266 * between @bs and @base is frozen. @errp is set if that's the case.
5267 * @base must be reachable from @bs, or NULL.
5269 bool bdrv_is_backing_chain_frozen(BlockDriverState
*bs
, BlockDriverState
*base
,
5272 BlockDriverState
*i
;
5275 for (i
= bs
; i
!= base
; i
= child_bs(child
)) {
5276 child
= bdrv_filter_or_cow_child(i
);
5278 if (child
&& child
->frozen
) {
5279 error_setg(errp
, "Cannot change '%s' link from '%s' to '%s'",
5280 child
->name
, i
->node_name
, child
->bs
->node_name
);
5289 * Freeze all COW (backing) and filter links between @bs and @base.
5290 * If any of the links is already frozen the operation is aborted and
5291 * none of the links are modified.
5292 * @base must be reachable from @bs, or NULL.
5293 * Returns 0 on success. On failure returns < 0 and sets @errp.
5295 int bdrv_freeze_backing_chain(BlockDriverState
*bs
, BlockDriverState
*base
,
5298 BlockDriverState
*i
;
5301 if (bdrv_is_backing_chain_frozen(bs
, base
, errp
)) {
5305 for (i
= bs
; i
!= base
; i
= child_bs(child
)) {
5306 child
= bdrv_filter_or_cow_child(i
);
5307 if (child
&& child
->bs
->never_freeze
) {
5308 error_setg(errp
, "Cannot freeze '%s' link to '%s'",
5309 child
->name
, child
->bs
->node_name
);
5314 for (i
= bs
; i
!= base
; i
= child_bs(child
)) {
5315 child
= bdrv_filter_or_cow_child(i
);
5317 child
->frozen
= true;
5325 * Unfreeze all COW (backing) and filter links between @bs and @base.
5326 * The caller must ensure that all links are frozen before using this
5328 * @base must be reachable from @bs, or NULL.
5330 void bdrv_unfreeze_backing_chain(BlockDriverState
*bs
, BlockDriverState
*base
)
5332 BlockDriverState
*i
;
5335 for (i
= bs
; i
!= base
; i
= child_bs(child
)) {
5336 child
= bdrv_filter_or_cow_child(i
);
5338 assert(child
->frozen
);
5339 child
->frozen
= false;
5345 * Drops images above 'base' up to and including 'top', and sets the image
5346 * above 'top' to have base as its backing file.
5348 * Requires that the overlay to 'top' is opened r/w, so that the backing file
5349 * information in 'bs' can be properly updated.
5351 * E.g., this will convert the following chain:
5352 * bottom <- base <- intermediate <- top <- active
5356 * bottom <- base <- active
5358 * It is allowed for bottom==base, in which case it converts:
5360 * base <- intermediate <- top <- active
5366 * If backing_file_str is non-NULL, it will be used when modifying top's
5367 * overlay image metadata.
5370 * if active == top, that is considered an error
5373 int bdrv_drop_intermediate(BlockDriverState
*top
, BlockDriverState
*base
,
5374 const char *backing_file_str
)
5376 BlockDriverState
*explicit_top
= top
;
5377 bool update_inherits_from
;
5379 Error
*local_err
= NULL
;
5381 g_autoptr(GSList
) updated_children
= NULL
;
5385 bdrv_subtree_drained_begin(top
);
5387 if (!top
->drv
|| !base
->drv
) {
5391 /* Make sure that base is in the backing chain of top */
5392 if (!bdrv_chain_contains(top
, base
)) {
5396 /* If 'base' recursively inherits from 'top' then we should set
5397 * base->inherits_from to top->inherits_from after 'top' and all
5398 * other intermediate nodes have been dropped.
5399 * If 'top' is an implicit node (e.g. "commit_top") we should skip
5400 * it because no one inherits from it. We use explicit_top for that. */
5401 explicit_top
= bdrv_skip_implicit_filters(explicit_top
);
5402 update_inherits_from
= bdrv_inherits_from_recursive(base
, explicit_top
);
5404 /* success - we can delete the intermediate states, and link top->base */
5405 /* TODO Check graph modification op blockers (BLK_PERM_GRAPH_MOD) once
5406 * we've figured out how they should work. */
5407 if (!backing_file_str
) {
5408 bdrv_refresh_filename(base
);
5409 backing_file_str
= base
->filename
;
5412 QLIST_FOREACH(c
, &top
->parents
, next_parent
) {
5413 updated_children
= g_slist_prepend(updated_children
, c
);
5416 bdrv_replace_node_common(top
, base
, false, &local_err
);
5418 error_report_err(local_err
);
5422 for (p
= updated_children
; p
; p
= p
->next
) {
5425 if (c
->klass
->update_filename
) {
5426 ret
= c
->klass
->update_filename(c
, base
, backing_file_str
,
5430 * TODO: Actually, we want to rollback all previous iterations
5431 * of this loop, and (which is almost impossible) previous
5432 * bdrv_replace_node()...
5434 * Note, that c->klass->update_filename may lead to permission
5435 * update, so it's a bad idea to call it inside permission
5436 * update transaction of bdrv_replace_node.
5438 error_report_err(local_err
);
5444 if (update_inherits_from
) {
5445 base
->inherits_from
= explicit_top
->inherits_from
;
5450 bdrv_subtree_drained_end(top
);
5456 * Implementation of BlockDriver.bdrv_get_allocated_file_size() that
5457 * sums the size of all data-bearing children. (This excludes backing
5460 static int64_t bdrv_sum_allocated_file_size(BlockDriverState
*bs
)
5463 int64_t child_size
, sum
= 0;
5465 QLIST_FOREACH(child
, &bs
->children
, next
) {
5466 if (child
->role
& (BDRV_CHILD_DATA
| BDRV_CHILD_METADATA
|
5467 BDRV_CHILD_FILTERED
))
5469 child_size
= bdrv_get_allocated_file_size(child
->bs
);
5470 if (child_size
< 0) {
5481 * Length of a allocated file in bytes. Sparse files are counted by actual
5482 * allocated space. Return < 0 if error or unknown.
5484 int64_t bdrv_get_allocated_file_size(BlockDriverState
*bs
)
5486 BlockDriver
*drv
= bs
->drv
;
5490 if (drv
->bdrv_get_allocated_file_size
) {
5491 return drv
->bdrv_get_allocated_file_size(bs
);
5494 if (drv
->bdrv_file_open
) {
5496 * Protocol drivers default to -ENOTSUP (most of their data is
5497 * not stored in any of their children (if they even have any),
5498 * so there is no generic way to figure it out).
5501 } else if (drv
->is_filter
) {
5502 /* Filter drivers default to the size of their filtered child */
5503 return bdrv_get_allocated_file_size(bdrv_filter_bs(bs
));
5505 /* Other drivers default to summing their children's sizes */
5506 return bdrv_sum_allocated_file_size(bs
);
5512 * @drv: Format driver
5513 * @opts: Creation options for new image
5514 * @in_bs: Existing image containing data for new image (may be NULL)
5515 * @errp: Error object
5516 * Returns: A #BlockMeasureInfo (free using qapi_free_BlockMeasureInfo())
5519 * Calculate file size required to create a new image.
5521 * If @in_bs is given then space for allocated clusters and zero clusters
5522 * from that image are included in the calculation. If @opts contains a
5523 * backing file that is shared by @in_bs then backing clusters may be omitted
5524 * from the calculation.
5526 * If @in_bs is NULL then the calculation includes no allocated clusters
5527 * unless a preallocation option is given in @opts.
5529 * Note that @in_bs may use a different BlockDriver from @drv.
5531 * If an error occurs the @errp pointer is set.
5533 BlockMeasureInfo
*bdrv_measure(BlockDriver
*drv
, QemuOpts
*opts
,
5534 BlockDriverState
*in_bs
, Error
**errp
)
5536 if (!drv
->bdrv_measure
) {
5537 error_setg(errp
, "Block driver '%s' does not support size measurement",
5542 return drv
->bdrv_measure(opts
, in_bs
, errp
);
5546 * Return number of sectors on success, -errno on error.
5548 int64_t bdrv_nb_sectors(BlockDriverState
*bs
)
5550 BlockDriver
*drv
= bs
->drv
;
5555 if (drv
->has_variable_length
) {
5556 int ret
= refresh_total_sectors(bs
, bs
->total_sectors
);
5561 return bs
->total_sectors
;
5565 * Return length in bytes on success, -errno on error.
5566 * The length is always a multiple of BDRV_SECTOR_SIZE.
5568 int64_t bdrv_getlength(BlockDriverState
*bs
)
5570 int64_t ret
= bdrv_nb_sectors(bs
);
5575 if (ret
> INT64_MAX
/ BDRV_SECTOR_SIZE
) {
5578 return ret
* BDRV_SECTOR_SIZE
;
5581 /* return 0 as number of sectors if no device present or error */
5582 void bdrv_get_geometry(BlockDriverState
*bs
, uint64_t *nb_sectors_ptr
)
5584 int64_t nb_sectors
= bdrv_nb_sectors(bs
);
5586 *nb_sectors_ptr
= nb_sectors
< 0 ? 0 : nb_sectors
;
5589 bool bdrv_is_sg(BlockDriverState
*bs
)
5595 * Return whether the given node supports compressed writes.
5597 bool bdrv_supports_compressed_writes(BlockDriverState
*bs
)
5599 BlockDriverState
*filtered
;
5601 if (!bs
->drv
|| !block_driver_can_compress(bs
->drv
)) {
5605 filtered
= bdrv_filter_bs(bs
);
5608 * Filters can only forward compressed writes, so we have to
5611 return bdrv_supports_compressed_writes(filtered
);
5617 const char *bdrv_get_format_name(BlockDriverState
*bs
)
5619 return bs
->drv
? bs
->drv
->format_name
: NULL
;
5622 static int qsort_strcmp(const void *a
, const void *b
)
5624 return strcmp(*(char *const *)a
, *(char *const *)b
);
5627 void bdrv_iterate_format(void (*it
)(void *opaque
, const char *name
),
5628 void *opaque
, bool read_only
)
5633 const char **formats
= NULL
;
5635 QLIST_FOREACH(drv
, &bdrv_drivers
, list
) {
5636 if (drv
->format_name
) {
5640 if (use_bdrv_whitelist
&& !bdrv_is_whitelisted(drv
, read_only
)) {
5644 while (formats
&& i
&& !found
) {
5645 found
= !strcmp(formats
[--i
], drv
->format_name
);
5649 formats
= g_renew(const char *, formats
, count
+ 1);
5650 formats
[count
++] = drv
->format_name
;
5655 for (i
= 0; i
< (int)ARRAY_SIZE(block_driver_modules
); i
++) {
5656 const char *format_name
= block_driver_modules
[i
].format_name
;
5662 if (use_bdrv_whitelist
&&
5663 !bdrv_format_is_whitelisted(format_name
, read_only
)) {
5667 while (formats
&& j
&& !found
) {
5668 found
= !strcmp(formats
[--j
], format_name
);
5672 formats
= g_renew(const char *, formats
, count
+ 1);
5673 formats
[count
++] = format_name
;
5678 qsort(formats
, count
, sizeof(formats
[0]), qsort_strcmp
);
5680 for (i
= 0; i
< count
; i
++) {
5681 it(opaque
, formats
[i
]);
5687 /* This function is to find a node in the bs graph */
5688 BlockDriverState
*bdrv_find_node(const char *node_name
)
5690 BlockDriverState
*bs
;
5694 QTAILQ_FOREACH(bs
, &graph_bdrv_states
, node_list
) {
5695 if (!strcmp(node_name
, bs
->node_name
)) {
5702 /* Put this QMP function here so it can access the static graph_bdrv_states. */
5703 BlockDeviceInfoList
*bdrv_named_nodes_list(bool flat
,
5706 BlockDeviceInfoList
*list
;
5707 BlockDriverState
*bs
;
5710 QTAILQ_FOREACH(bs
, &graph_bdrv_states
, node_list
) {
5711 BlockDeviceInfo
*info
= bdrv_block_device_info(NULL
, bs
, flat
, errp
);
5713 qapi_free_BlockDeviceInfoList(list
);
5716 QAPI_LIST_PREPEND(list
, info
);
5722 typedef struct XDbgBlockGraphConstructor
{
5723 XDbgBlockGraph
*graph
;
5724 GHashTable
*graph_nodes
;
5725 } XDbgBlockGraphConstructor
;
5727 static XDbgBlockGraphConstructor
*xdbg_graph_new(void)
5729 XDbgBlockGraphConstructor
*gr
= g_new(XDbgBlockGraphConstructor
, 1);
5731 gr
->graph
= g_new0(XDbgBlockGraph
, 1);
5732 gr
->graph_nodes
= g_hash_table_new(NULL
, NULL
);
5737 static XDbgBlockGraph
*xdbg_graph_finalize(XDbgBlockGraphConstructor
*gr
)
5739 XDbgBlockGraph
*graph
= gr
->graph
;
5741 g_hash_table_destroy(gr
->graph_nodes
);
5747 static uintptr_t xdbg_graph_node_num(XDbgBlockGraphConstructor
*gr
, void *node
)
5749 uintptr_t ret
= (uintptr_t)g_hash_table_lookup(gr
->graph_nodes
, node
);
5756 * Start counting from 1, not 0, because 0 interferes with not-found (NULL)
5757 * answer of g_hash_table_lookup.
5759 ret
= g_hash_table_size(gr
->graph_nodes
) + 1;
5760 g_hash_table_insert(gr
->graph_nodes
, node
, (void *)ret
);
5765 static void xdbg_graph_add_node(XDbgBlockGraphConstructor
*gr
, void *node
,
5766 XDbgBlockGraphNodeType type
, const char *name
)
5768 XDbgBlockGraphNode
*n
;
5770 n
= g_new0(XDbgBlockGraphNode
, 1);
5772 n
->id
= xdbg_graph_node_num(gr
, node
);
5774 n
->name
= g_strdup(name
);
5776 QAPI_LIST_PREPEND(gr
->graph
->nodes
, n
);
5779 static void xdbg_graph_add_edge(XDbgBlockGraphConstructor
*gr
, void *parent
,
5780 const BdrvChild
*child
)
5782 BlockPermission qapi_perm
;
5783 XDbgBlockGraphEdge
*edge
;
5785 edge
= g_new0(XDbgBlockGraphEdge
, 1);
5787 edge
->parent
= xdbg_graph_node_num(gr
, parent
);
5788 edge
->child
= xdbg_graph_node_num(gr
, child
->bs
);
5789 edge
->name
= g_strdup(child
->name
);
5791 for (qapi_perm
= 0; qapi_perm
< BLOCK_PERMISSION__MAX
; qapi_perm
++) {
5792 uint64_t flag
= bdrv_qapi_perm_to_blk_perm(qapi_perm
);
5794 if (flag
& child
->perm
) {
5795 QAPI_LIST_PREPEND(edge
->perm
, qapi_perm
);
5797 if (flag
& child
->shared_perm
) {
5798 QAPI_LIST_PREPEND(edge
->shared_perm
, qapi_perm
);
5802 QAPI_LIST_PREPEND(gr
->graph
->edges
, edge
);
5806 XDbgBlockGraph
*bdrv_get_xdbg_block_graph(Error
**errp
)
5810 BlockDriverState
*bs
;
5812 XDbgBlockGraphConstructor
*gr
= xdbg_graph_new();
5814 for (blk
= blk_all_next(NULL
); blk
; blk
= blk_all_next(blk
)) {
5815 char *allocated_name
= NULL
;
5816 const char *name
= blk_name(blk
);
5819 name
= allocated_name
= blk_get_attached_dev_id(blk
);
5821 xdbg_graph_add_node(gr
, blk
, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_BACKEND
,
5823 g_free(allocated_name
);
5824 if (blk_root(blk
)) {
5825 xdbg_graph_add_edge(gr
, blk
, blk_root(blk
));
5829 for (job
= block_job_next(NULL
); job
; job
= block_job_next(job
)) {
5832 xdbg_graph_add_node(gr
, job
, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_JOB
,
5834 for (el
= job
->nodes
; el
; el
= el
->next
) {
5835 xdbg_graph_add_edge(gr
, job
, (BdrvChild
*)el
->data
);
5839 QTAILQ_FOREACH(bs
, &graph_bdrv_states
, node_list
) {
5840 xdbg_graph_add_node(gr
, bs
, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_DRIVER
,
5842 QLIST_FOREACH(child
, &bs
->children
, next
) {
5843 xdbg_graph_add_edge(gr
, bs
, child
);
5847 return xdbg_graph_finalize(gr
);
5850 BlockDriverState
*bdrv_lookup_bs(const char *device
,
5851 const char *node_name
,
5855 BlockDriverState
*bs
;
5858 blk
= blk_by_name(device
);
5863 error_setg(errp
, "Device '%s' has no medium", device
);
5871 bs
= bdrv_find_node(node_name
);
5878 error_setg(errp
, "Cannot find device=\'%s\' nor node-name=\'%s\'",
5879 device
? device
: "",
5880 node_name
? node_name
: "");
5884 /* If 'base' is in the same chain as 'top', return true. Otherwise,
5885 * return false. If either argument is NULL, return false. */
5886 bool bdrv_chain_contains(BlockDriverState
*top
, BlockDriverState
*base
)
5888 while (top
&& top
!= base
) {
5889 top
= bdrv_filter_or_cow_bs(top
);
5895 BlockDriverState
*bdrv_next_node(BlockDriverState
*bs
)
5898 return QTAILQ_FIRST(&graph_bdrv_states
);
5900 return QTAILQ_NEXT(bs
, node_list
);
5903 BlockDriverState
*bdrv_next_all_states(BlockDriverState
*bs
)
5906 return QTAILQ_FIRST(&all_bdrv_states
);
5908 return QTAILQ_NEXT(bs
, bs_list
);
5911 const char *bdrv_get_node_name(const BlockDriverState
*bs
)
5913 return bs
->node_name
;
5916 const char *bdrv_get_parent_name(const BlockDriverState
*bs
)
5921 /* If multiple parents have a name, just pick the first one. */
5922 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
5923 if (c
->klass
->get_name
) {
5924 name
= c
->klass
->get_name(c
);
5925 if (name
&& *name
) {
5934 /* TODO check what callers really want: bs->node_name or blk_name() */
5935 const char *bdrv_get_device_name(const BlockDriverState
*bs
)
5937 return bdrv_get_parent_name(bs
) ?: "";
5940 /* This can be used to identify nodes that might not have a device
5941 * name associated. Since node and device names live in the same
5942 * namespace, the result is unambiguous. The exception is if both are
5943 * absent, then this returns an empty (non-null) string. */
5944 const char *bdrv_get_device_or_node_name(const BlockDriverState
*bs
)
5946 return bdrv_get_parent_name(bs
) ?: bs
->node_name
;
5949 int bdrv_get_flags(BlockDriverState
*bs
)
5951 return bs
->open_flags
;
5954 int bdrv_has_zero_init_1(BlockDriverState
*bs
)
5959 int bdrv_has_zero_init(BlockDriverState
*bs
)
5961 BlockDriverState
*filtered
;
5967 /* If BS is a copy on write image, it is initialized to
5968 the contents of the base image, which may not be zeroes. */
5969 if (bdrv_cow_child(bs
)) {
5972 if (bs
->drv
->bdrv_has_zero_init
) {
5973 return bs
->drv
->bdrv_has_zero_init(bs
);
5976 filtered
= bdrv_filter_bs(bs
);
5978 return bdrv_has_zero_init(filtered
);
5985 bool bdrv_can_write_zeroes_with_unmap(BlockDriverState
*bs
)
5987 if (!(bs
->open_flags
& BDRV_O_UNMAP
)) {
5991 return bs
->supported_zero_flags
& BDRV_REQ_MAY_UNMAP
;
5994 void bdrv_get_backing_filename(BlockDriverState
*bs
,
5995 char *filename
, int filename_size
)
5997 pstrcpy(filename
, filename_size
, bs
->backing_file
);
6000 int bdrv_get_info(BlockDriverState
*bs
, BlockDriverInfo
*bdi
)
6003 BlockDriver
*drv
= bs
->drv
;
6004 /* if bs->drv == NULL, bs is closed, so there's nothing to do here */
6008 if (!drv
->bdrv_get_info
) {
6009 BlockDriverState
*filtered
= bdrv_filter_bs(bs
);
6011 return bdrv_get_info(filtered
, bdi
);
6015 memset(bdi
, 0, sizeof(*bdi
));
6016 ret
= drv
->bdrv_get_info(bs
, bdi
);
6021 if (bdi
->cluster_size
> BDRV_MAX_ALIGNMENT
) {
6028 ImageInfoSpecific
*bdrv_get_specific_info(BlockDriverState
*bs
,
6031 BlockDriver
*drv
= bs
->drv
;
6032 if (drv
&& drv
->bdrv_get_specific_info
) {
6033 return drv
->bdrv_get_specific_info(bs
, errp
);
6038 BlockStatsSpecific
*bdrv_get_specific_stats(BlockDriverState
*bs
)
6040 BlockDriver
*drv
= bs
->drv
;
6041 if (!drv
|| !drv
->bdrv_get_specific_stats
) {
6044 return drv
->bdrv_get_specific_stats(bs
);
6047 void bdrv_debug_event(BlockDriverState
*bs
, BlkdebugEvent event
)
6049 if (!bs
|| !bs
->drv
|| !bs
->drv
->bdrv_debug_event
) {
6053 bs
->drv
->bdrv_debug_event(bs
, event
);
6056 static BlockDriverState
*bdrv_find_debug_node(BlockDriverState
*bs
)
6058 while (bs
&& bs
->drv
&& !bs
->drv
->bdrv_debug_breakpoint
) {
6059 bs
= bdrv_primary_bs(bs
);
6062 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_breakpoint
) {
6063 assert(bs
->drv
->bdrv_debug_remove_breakpoint
);
6070 int bdrv_debug_breakpoint(BlockDriverState
*bs
, const char *event
,
6073 bs
= bdrv_find_debug_node(bs
);
6075 return bs
->drv
->bdrv_debug_breakpoint(bs
, event
, tag
);
6081 int bdrv_debug_remove_breakpoint(BlockDriverState
*bs
, const char *tag
)
6083 bs
= bdrv_find_debug_node(bs
);
6085 return bs
->drv
->bdrv_debug_remove_breakpoint(bs
, tag
);
6091 int bdrv_debug_resume(BlockDriverState
*bs
, const char *tag
)
6093 while (bs
&& (!bs
->drv
|| !bs
->drv
->bdrv_debug_resume
)) {
6094 bs
= bdrv_primary_bs(bs
);
6097 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_resume
) {
6098 return bs
->drv
->bdrv_debug_resume(bs
, tag
);
6104 bool bdrv_debug_is_suspended(BlockDriverState
*bs
, const char *tag
)
6106 while (bs
&& bs
->drv
&& !bs
->drv
->bdrv_debug_is_suspended
) {
6107 bs
= bdrv_primary_bs(bs
);
6110 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_is_suspended
) {
6111 return bs
->drv
->bdrv_debug_is_suspended(bs
, tag
);
6117 /* backing_file can either be relative, or absolute, or a protocol. If it is
6118 * relative, it must be relative to the chain. So, passing in bs->filename
6119 * from a BDS as backing_file should not be done, as that may be relative to
6120 * the CWD rather than the chain. */
6121 BlockDriverState
*bdrv_find_backing_image(BlockDriverState
*bs
,
6122 const char *backing_file
)
6124 char *filename_full
= NULL
;
6125 char *backing_file_full
= NULL
;
6126 char *filename_tmp
= NULL
;
6127 int is_protocol
= 0;
6128 bool filenames_refreshed
= false;
6129 BlockDriverState
*curr_bs
= NULL
;
6130 BlockDriverState
*retval
= NULL
;
6131 BlockDriverState
*bs_below
;
6133 if (!bs
|| !bs
->drv
|| !backing_file
) {
6137 filename_full
= g_malloc(PATH_MAX
);
6138 backing_file_full
= g_malloc(PATH_MAX
);
6140 is_protocol
= path_has_protocol(backing_file
);
6143 * Being largely a legacy function, skip any filters here
6144 * (because filters do not have normal filenames, so they cannot
6145 * match anyway; and allowing json:{} filenames is a bit out of
6148 for (curr_bs
= bdrv_skip_filters(bs
);
6149 bdrv_cow_child(curr_bs
) != NULL
;
6152 bs_below
= bdrv_backing_chain_next(curr_bs
);
6154 if (bdrv_backing_overridden(curr_bs
)) {
6156 * If the backing file was overridden, we can only compare
6157 * directly against the backing node's filename.
6160 if (!filenames_refreshed
) {
6162 * This will automatically refresh all of the
6163 * filenames in the rest of the backing chain, so we
6164 * only need to do this once.
6166 bdrv_refresh_filename(bs_below
);
6167 filenames_refreshed
= true;
6170 if (strcmp(backing_file
, bs_below
->filename
) == 0) {
6174 } else if (is_protocol
|| path_has_protocol(curr_bs
->backing_file
)) {
6176 * If either of the filename paths is actually a protocol, then
6177 * compare unmodified paths; otherwise make paths relative.
6179 char *backing_file_full_ret
;
6181 if (strcmp(backing_file
, curr_bs
->backing_file
) == 0) {
6185 /* Also check against the full backing filename for the image */
6186 backing_file_full_ret
= bdrv_get_full_backing_filename(curr_bs
,
6188 if (backing_file_full_ret
) {
6189 bool equal
= strcmp(backing_file
, backing_file_full_ret
) == 0;
6190 g_free(backing_file_full_ret
);
6197 /* If not an absolute filename path, make it relative to the current
6198 * image's filename path */
6199 filename_tmp
= bdrv_make_absolute_filename(curr_bs
, backing_file
,
6201 /* We are going to compare canonicalized absolute pathnames */
6202 if (!filename_tmp
|| !realpath(filename_tmp
, filename_full
)) {
6203 g_free(filename_tmp
);
6206 g_free(filename_tmp
);
6208 /* We need to make sure the backing filename we are comparing against
6209 * is relative to the current image filename (or absolute) */
6210 filename_tmp
= bdrv_get_full_backing_filename(curr_bs
, NULL
);
6211 if (!filename_tmp
|| !realpath(filename_tmp
, backing_file_full
)) {
6212 g_free(filename_tmp
);
6215 g_free(filename_tmp
);
6217 if (strcmp(backing_file_full
, filename_full
) == 0) {
6224 g_free(filename_full
);
6225 g_free(backing_file_full
);
6229 void bdrv_init(void)
6231 module_call_init(MODULE_INIT_BLOCK
);
6234 void bdrv_init_with_whitelist(void)
6236 use_bdrv_whitelist
= 1;
6240 int coroutine_fn
bdrv_co_invalidate_cache(BlockDriverState
*bs
, Error
**errp
)
6242 BdrvChild
*child
, *parent
;
6243 Error
*local_err
= NULL
;
6245 BdrvDirtyBitmap
*bm
;
6251 QLIST_FOREACH(child
, &bs
->children
, next
) {
6252 bdrv_co_invalidate_cache(child
->bs
, &local_err
);
6254 error_propagate(errp
, local_err
);
6260 * Update permissions, they may differ for inactive nodes.
6262 * Note that the required permissions of inactive images are always a
6263 * subset of the permissions required after activating the image. This
6264 * allows us to just get the permissions upfront without restricting
6265 * drv->bdrv_invalidate_cache().
6267 * It also means that in error cases, we don't have to try and revert to
6268 * the old permissions (which is an operation that could fail, too). We can
6269 * just keep the extended permissions for the next time that an activation
6270 * of the image is tried.
6272 if (bs
->open_flags
& BDRV_O_INACTIVE
) {
6273 bs
->open_flags
&= ~BDRV_O_INACTIVE
;
6274 ret
= bdrv_refresh_perms(bs
, errp
);
6276 bs
->open_flags
|= BDRV_O_INACTIVE
;
6280 if (bs
->drv
->bdrv_co_invalidate_cache
) {
6281 bs
->drv
->bdrv_co_invalidate_cache(bs
, &local_err
);
6283 bs
->open_flags
|= BDRV_O_INACTIVE
;
6284 error_propagate(errp
, local_err
);
6289 FOR_EACH_DIRTY_BITMAP(bs
, bm
) {
6290 bdrv_dirty_bitmap_skip_store(bm
, false);
6293 ret
= refresh_total_sectors(bs
, bs
->total_sectors
);
6295 bs
->open_flags
|= BDRV_O_INACTIVE
;
6296 error_setg_errno(errp
, -ret
, "Could not refresh total sector count");
6301 QLIST_FOREACH(parent
, &bs
->parents
, next_parent
) {
6302 if (parent
->klass
->activate
) {
6303 parent
->klass
->activate(parent
, &local_err
);
6305 bs
->open_flags
|= BDRV_O_INACTIVE
;
6306 error_propagate(errp
, local_err
);
6315 void bdrv_invalidate_cache_all(Error
**errp
)
6317 BlockDriverState
*bs
;
6318 BdrvNextIterator it
;
6320 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
6321 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
6324 aio_context_acquire(aio_context
);
6325 ret
= bdrv_invalidate_cache(bs
, errp
);
6326 aio_context_release(aio_context
);
6328 bdrv_next_cleanup(&it
);
6334 static bool bdrv_has_bds_parent(BlockDriverState
*bs
, bool only_active
)
6338 QLIST_FOREACH(parent
, &bs
->parents
, next_parent
) {
6339 if (parent
->klass
->parent_is_bds
) {
6340 BlockDriverState
*parent_bs
= parent
->opaque
;
6341 if (!only_active
|| !(parent_bs
->open_flags
& BDRV_O_INACTIVE
)) {
6350 static int bdrv_inactivate_recurse(BlockDriverState
*bs
)
6352 BdrvChild
*child
, *parent
;
6359 /* Make sure that we don't inactivate a child before its parent.
6360 * It will be covered by recursion from the yet active parent. */
6361 if (bdrv_has_bds_parent(bs
, true)) {
6365 assert(!(bs
->open_flags
& BDRV_O_INACTIVE
));
6367 /* Inactivate this node */
6368 if (bs
->drv
->bdrv_inactivate
) {
6369 ret
= bs
->drv
->bdrv_inactivate(bs
);
6375 QLIST_FOREACH(parent
, &bs
->parents
, next_parent
) {
6376 if (parent
->klass
->inactivate
) {
6377 ret
= parent
->klass
->inactivate(parent
);
6384 bs
->open_flags
|= BDRV_O_INACTIVE
;
6387 * Update permissions, they may differ for inactive nodes.
6388 * We only tried to loosen restrictions, so errors are not fatal, ignore
6391 bdrv_refresh_perms(bs
, NULL
);
6393 /* Recursively inactivate children */
6394 QLIST_FOREACH(child
, &bs
->children
, next
) {
6395 ret
= bdrv_inactivate_recurse(child
->bs
);
6404 int bdrv_inactivate_all(void)
6406 BlockDriverState
*bs
= NULL
;
6407 BdrvNextIterator it
;
6409 GSList
*aio_ctxs
= NULL
, *ctx
;
6411 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
6412 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
6414 if (!g_slist_find(aio_ctxs
, aio_context
)) {
6415 aio_ctxs
= g_slist_prepend(aio_ctxs
, aio_context
);
6416 aio_context_acquire(aio_context
);
6420 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
6421 /* Nodes with BDS parents are covered by recursion from the last
6422 * parent that gets inactivated. Don't inactivate them a second
6423 * time if that has already happened. */
6424 if (bdrv_has_bds_parent(bs
, false)) {
6427 ret
= bdrv_inactivate_recurse(bs
);
6429 bdrv_next_cleanup(&it
);
6435 for (ctx
= aio_ctxs
; ctx
!= NULL
; ctx
= ctx
->next
) {
6436 AioContext
*aio_context
= ctx
->data
;
6437 aio_context_release(aio_context
);
6439 g_slist_free(aio_ctxs
);
6444 /**************************************************************/
6445 /* removable device support */
6448 * Return TRUE if the media is present
6450 bool bdrv_is_inserted(BlockDriverState
*bs
)
6452 BlockDriver
*drv
= bs
->drv
;
6458 if (drv
->bdrv_is_inserted
) {
6459 return drv
->bdrv_is_inserted(bs
);
6461 QLIST_FOREACH(child
, &bs
->children
, next
) {
6462 if (!bdrv_is_inserted(child
->bs
)) {
6470 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
6472 void bdrv_eject(BlockDriverState
*bs
, bool eject_flag
)
6474 BlockDriver
*drv
= bs
->drv
;
6476 if (drv
&& drv
->bdrv_eject
) {
6477 drv
->bdrv_eject(bs
, eject_flag
);
6482 * Lock or unlock the media (if it is locked, the user won't be able
6483 * to eject it manually).
6485 void bdrv_lock_medium(BlockDriverState
*bs
, bool locked
)
6487 BlockDriver
*drv
= bs
->drv
;
6489 trace_bdrv_lock_medium(bs
, locked
);
6491 if (drv
&& drv
->bdrv_lock_medium
) {
6492 drv
->bdrv_lock_medium(bs
, locked
);
6496 /* Get a reference to bs */
6497 void bdrv_ref(BlockDriverState
*bs
)
6502 /* Release a previously grabbed reference to bs.
6503 * If after releasing, reference count is zero, the BlockDriverState is
6505 void bdrv_unref(BlockDriverState
*bs
)
6510 assert(bs
->refcnt
> 0);
6511 if (--bs
->refcnt
== 0) {
6516 struct BdrvOpBlocker
{
6518 QLIST_ENTRY(BdrvOpBlocker
) list
;
6521 bool bdrv_op_is_blocked(BlockDriverState
*bs
, BlockOpType op
, Error
**errp
)
6523 BdrvOpBlocker
*blocker
;
6524 assert((int) op
>= 0 && op
< BLOCK_OP_TYPE_MAX
);
6525 if (!QLIST_EMPTY(&bs
->op_blockers
[op
])) {
6526 blocker
= QLIST_FIRST(&bs
->op_blockers
[op
]);
6527 error_propagate_prepend(errp
, error_copy(blocker
->reason
),
6528 "Node '%s' is busy: ",
6529 bdrv_get_device_or_node_name(bs
));
6535 void bdrv_op_block(BlockDriverState
*bs
, BlockOpType op
, Error
*reason
)
6537 BdrvOpBlocker
*blocker
;
6538 assert((int) op
>= 0 && op
< BLOCK_OP_TYPE_MAX
);
6540 blocker
= g_new0(BdrvOpBlocker
, 1);
6541 blocker
->reason
= reason
;
6542 QLIST_INSERT_HEAD(&bs
->op_blockers
[op
], blocker
, list
);
6545 void bdrv_op_unblock(BlockDriverState
*bs
, BlockOpType op
, Error
*reason
)
6547 BdrvOpBlocker
*blocker
, *next
;
6548 assert((int) op
>= 0 && op
< BLOCK_OP_TYPE_MAX
);
6549 QLIST_FOREACH_SAFE(blocker
, &bs
->op_blockers
[op
], list
, next
) {
6550 if (blocker
->reason
== reason
) {
6551 QLIST_REMOVE(blocker
, list
);
6557 void bdrv_op_block_all(BlockDriverState
*bs
, Error
*reason
)
6560 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
6561 bdrv_op_block(bs
, i
, reason
);
6565 void bdrv_op_unblock_all(BlockDriverState
*bs
, Error
*reason
)
6568 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
6569 bdrv_op_unblock(bs
, i
, reason
);
6573 bool bdrv_op_blocker_is_empty(BlockDriverState
*bs
)
6577 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
6578 if (!QLIST_EMPTY(&bs
->op_blockers
[i
])) {
6585 void bdrv_img_create(const char *filename
, const char *fmt
,
6586 const char *base_filename
, const char *base_fmt
,
6587 char *options
, uint64_t img_size
, int flags
, bool quiet
,
6590 QemuOptsList
*create_opts
= NULL
;
6591 QemuOpts
*opts
= NULL
;
6592 const char *backing_fmt
, *backing_file
;
6594 BlockDriver
*drv
, *proto_drv
;
6595 Error
*local_err
= NULL
;
6598 /* Find driver and parse its options */
6599 drv
= bdrv_find_format(fmt
);
6601 error_setg(errp
, "Unknown file format '%s'", fmt
);
6605 proto_drv
= bdrv_find_protocol(filename
, true, errp
);
6610 if (!drv
->create_opts
) {
6611 error_setg(errp
, "Format driver '%s' does not support image creation",
6616 if (!proto_drv
->create_opts
) {
6617 error_setg(errp
, "Protocol driver '%s' does not support image creation",
6618 proto_drv
->format_name
);
6622 /* Create parameter list */
6623 create_opts
= qemu_opts_append(create_opts
, drv
->create_opts
);
6624 create_opts
= qemu_opts_append(create_opts
, proto_drv
->create_opts
);
6626 opts
= qemu_opts_create(create_opts
, NULL
, 0, &error_abort
);
6628 /* Parse -o options */
6630 if (!qemu_opts_do_parse(opts
, options
, NULL
, errp
)) {
6635 if (!qemu_opt_get(opts
, BLOCK_OPT_SIZE
)) {
6636 qemu_opt_set_number(opts
, BLOCK_OPT_SIZE
, img_size
, &error_abort
);
6637 } else if (img_size
!= UINT64_C(-1)) {
6638 error_setg(errp
, "The image size must be specified only once");
6642 if (base_filename
) {
6643 if (!qemu_opt_set(opts
, BLOCK_OPT_BACKING_FILE
, base_filename
,
6645 error_setg(errp
, "Backing file not supported for file format '%s'",
6652 if (!qemu_opt_set(opts
, BLOCK_OPT_BACKING_FMT
, base_fmt
, NULL
)) {
6653 error_setg(errp
, "Backing file format not supported for file "
6654 "format '%s'", fmt
);
6659 backing_file
= qemu_opt_get(opts
, BLOCK_OPT_BACKING_FILE
);
6661 if (!strcmp(filename
, backing_file
)) {
6662 error_setg(errp
, "Error: Trying to create an image with the "
6663 "same filename as the backing file");
6666 if (backing_file
[0] == '\0') {
6667 error_setg(errp
, "Expected backing file name, got empty string");
6672 backing_fmt
= qemu_opt_get(opts
, BLOCK_OPT_BACKING_FMT
);
6674 /* The size for the image must always be specified, unless we have a backing
6675 * file and we have not been forbidden from opening it. */
6676 size
= qemu_opt_get_size(opts
, BLOCK_OPT_SIZE
, img_size
);
6677 if (backing_file
&& !(flags
& BDRV_O_NO_BACKING
)) {
6678 BlockDriverState
*bs
;
6681 QDict
*backing_options
= NULL
;
6684 bdrv_get_full_backing_filename_from_filename(filename
, backing_file
,
6689 assert(full_backing
);
6691 /* backing files always opened read-only */
6693 back_flags
&= ~(BDRV_O_RDWR
| BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
);
6695 backing_options
= qdict_new();
6697 qdict_put_str(backing_options
, "driver", backing_fmt
);
6699 qdict_put_bool(backing_options
, BDRV_OPT_FORCE_SHARE
, true);
6701 bs
= bdrv_open(full_backing
, NULL
, backing_options
, back_flags
,
6703 g_free(full_backing
);
6705 error_append_hint(&local_err
, "Could not open backing image.\n");
6709 warn_report("Deprecated use of backing file without explicit "
6710 "backing format (detected format of %s)",
6711 bs
->drv
->format_name
);
6712 if (bs
->drv
!= &bdrv_raw
) {
6714 * A probe of raw deserves the most attention:
6715 * leaving the backing format out of the image
6716 * will ensure bs->probed is set (ensuring we
6717 * don't accidentally commit into the backing
6718 * file), and allow more spots to warn the users
6719 * to fix their toolchain when opening this image
6720 * later. For other images, we can safely record
6721 * the format that we probed.
6723 backing_fmt
= bs
->drv
->format_name
;
6724 qemu_opt_set(opts
, BLOCK_OPT_BACKING_FMT
, backing_fmt
,
6729 /* Opened BS, have no size */
6730 size
= bdrv_getlength(bs
);
6732 error_setg_errno(errp
, -size
, "Could not get size of '%s'",
6737 qemu_opt_set_number(opts
, BLOCK_OPT_SIZE
, size
, &error_abort
);
6741 /* (backing_file && !(flags & BDRV_O_NO_BACKING)) */
6742 } else if (backing_file
&& !backing_fmt
) {
6743 warn_report("Deprecated use of unopened backing file without "
6744 "explicit backing format, use of this image requires "
6745 "potentially unsafe format probing");
6749 error_setg(errp
, "Image creation needs a size parameter");
6754 printf("Formatting '%s', fmt=%s ", filename
, fmt
);
6755 qemu_opts_print(opts
, " ");
6760 ret
= bdrv_create(drv
, filename
, opts
, &local_err
);
6762 if (ret
== -EFBIG
) {
6763 /* This is generally a better message than whatever the driver would
6764 * deliver (especially because of the cluster_size_hint), since that
6765 * is most probably not much different from "image too large". */
6766 const char *cluster_size_hint
= "";
6767 if (qemu_opt_get_size(opts
, BLOCK_OPT_CLUSTER_SIZE
, 0)) {
6768 cluster_size_hint
= " (try using a larger cluster size)";
6770 error_setg(errp
, "The image size is too large for file format '%s'"
6771 "%s", fmt
, cluster_size_hint
);
6772 error_free(local_err
);
6777 qemu_opts_del(opts
);
6778 qemu_opts_free(create_opts
);
6779 error_propagate(errp
, local_err
);
6782 AioContext
*bdrv_get_aio_context(BlockDriverState
*bs
)
6784 return bs
? bs
->aio_context
: qemu_get_aio_context();
6787 AioContext
*coroutine_fn
bdrv_co_enter(BlockDriverState
*bs
)
6789 Coroutine
*self
= qemu_coroutine_self();
6790 AioContext
*old_ctx
= qemu_coroutine_get_aio_context(self
);
6791 AioContext
*new_ctx
;
6794 * Increase bs->in_flight to ensure that this operation is completed before
6795 * moving the node to a different AioContext. Read new_ctx only afterwards.
6797 bdrv_inc_in_flight(bs
);
6799 new_ctx
= bdrv_get_aio_context(bs
);
6800 aio_co_reschedule_self(new_ctx
);
6804 void coroutine_fn
bdrv_co_leave(BlockDriverState
*bs
, AioContext
*old_ctx
)
6806 aio_co_reschedule_self(old_ctx
);
6807 bdrv_dec_in_flight(bs
);
6810 void coroutine_fn
bdrv_co_lock(BlockDriverState
*bs
)
6812 AioContext
*ctx
= bdrv_get_aio_context(bs
);
6814 /* In the main thread, bs->aio_context won't change concurrently */
6815 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
6818 * We're in coroutine context, so we already hold the lock of the main
6819 * loop AioContext. Don't lock it twice to avoid deadlocks.
6821 assert(qemu_in_coroutine());
6822 if (ctx
!= qemu_get_aio_context()) {
6823 aio_context_acquire(ctx
);
6827 void coroutine_fn
bdrv_co_unlock(BlockDriverState
*bs
)
6829 AioContext
*ctx
= bdrv_get_aio_context(bs
);
6831 assert(qemu_in_coroutine());
6832 if (ctx
!= qemu_get_aio_context()) {
6833 aio_context_release(ctx
);
6837 void bdrv_coroutine_enter(BlockDriverState
*bs
, Coroutine
*co
)
6839 aio_co_enter(bdrv_get_aio_context(bs
), co
);
6842 static void bdrv_do_remove_aio_context_notifier(BdrvAioNotifier
*ban
)
6844 QLIST_REMOVE(ban
, list
);
6848 static void bdrv_detach_aio_context(BlockDriverState
*bs
)
6850 BdrvAioNotifier
*baf
, *baf_tmp
;
6852 assert(!bs
->walking_aio_notifiers
);
6853 bs
->walking_aio_notifiers
= true;
6854 QLIST_FOREACH_SAFE(baf
, &bs
->aio_notifiers
, list
, baf_tmp
) {
6856 bdrv_do_remove_aio_context_notifier(baf
);
6858 baf
->detach_aio_context(baf
->opaque
);
6861 /* Never mind iterating again to check for ->deleted. bdrv_close() will
6862 * remove remaining aio notifiers if we aren't called again.
6864 bs
->walking_aio_notifiers
= false;
6866 if (bs
->drv
&& bs
->drv
->bdrv_detach_aio_context
) {
6867 bs
->drv
->bdrv_detach_aio_context(bs
);
6870 if (bs
->quiesce_counter
) {
6871 aio_enable_external(bs
->aio_context
);
6873 bs
->aio_context
= NULL
;
6876 static void bdrv_attach_aio_context(BlockDriverState
*bs
,
6877 AioContext
*new_context
)
6879 BdrvAioNotifier
*ban
, *ban_tmp
;
6881 if (bs
->quiesce_counter
) {
6882 aio_disable_external(new_context
);
6885 bs
->aio_context
= new_context
;
6887 if (bs
->drv
&& bs
->drv
->bdrv_attach_aio_context
) {
6888 bs
->drv
->bdrv_attach_aio_context(bs
, new_context
);
6891 assert(!bs
->walking_aio_notifiers
);
6892 bs
->walking_aio_notifiers
= true;
6893 QLIST_FOREACH_SAFE(ban
, &bs
->aio_notifiers
, list
, ban_tmp
) {
6895 bdrv_do_remove_aio_context_notifier(ban
);
6897 ban
->attached_aio_context(new_context
, ban
->opaque
);
6900 bs
->walking_aio_notifiers
= false;
6904 * Changes the AioContext used for fd handlers, timers, and BHs by this
6905 * BlockDriverState and all its children and parents.
6907 * Must be called from the main AioContext.
6909 * The caller must own the AioContext lock for the old AioContext of bs, but it
6910 * must not own the AioContext lock for new_context (unless new_context is the
6911 * same as the current context of bs).
6913 * @ignore will accumulate all visited BdrvChild object. The caller is
6914 * responsible for freeing the list afterwards.
6916 void bdrv_set_aio_context_ignore(BlockDriverState
*bs
,
6917 AioContext
*new_context
, GSList
**ignore
)
6919 AioContext
*old_context
= bdrv_get_aio_context(bs
);
6920 GSList
*children_to_process
= NULL
;
6921 GSList
*parents_to_process
= NULL
;
6923 BdrvChild
*child
, *parent
;
6925 g_assert(qemu_get_current_aio_context() == qemu_get_aio_context());
6927 if (old_context
== new_context
) {
6931 bdrv_drained_begin(bs
);
6933 QLIST_FOREACH(child
, &bs
->children
, next
) {
6934 if (g_slist_find(*ignore
, child
)) {
6937 *ignore
= g_slist_prepend(*ignore
, child
);
6938 children_to_process
= g_slist_prepend(children_to_process
, child
);
6941 QLIST_FOREACH(parent
, &bs
->parents
, next_parent
) {
6942 if (g_slist_find(*ignore
, parent
)) {
6945 *ignore
= g_slist_prepend(*ignore
, parent
);
6946 parents_to_process
= g_slist_prepend(parents_to_process
, parent
);
6949 for (entry
= children_to_process
;
6951 entry
= g_slist_next(entry
)) {
6952 child
= entry
->data
;
6953 bdrv_set_aio_context_ignore(child
->bs
, new_context
, ignore
);
6955 g_slist_free(children_to_process
);
6957 for (entry
= parents_to_process
;
6959 entry
= g_slist_next(entry
)) {
6960 parent
= entry
->data
;
6961 assert(parent
->klass
->set_aio_ctx
);
6962 parent
->klass
->set_aio_ctx(parent
, new_context
, ignore
);
6964 g_slist_free(parents_to_process
);
6966 bdrv_detach_aio_context(bs
);
6968 /* Acquire the new context, if necessary */
6969 if (qemu_get_aio_context() != new_context
) {
6970 aio_context_acquire(new_context
);
6973 bdrv_attach_aio_context(bs
, new_context
);
6976 * If this function was recursively called from
6977 * bdrv_set_aio_context_ignore(), there may be nodes in the
6978 * subtree that have not yet been moved to the new AioContext.
6979 * Release the old one so bdrv_drained_end() can poll them.
6981 if (qemu_get_aio_context() != old_context
) {
6982 aio_context_release(old_context
);
6985 bdrv_drained_end(bs
);
6987 if (qemu_get_aio_context() != old_context
) {
6988 aio_context_acquire(old_context
);
6990 if (qemu_get_aio_context() != new_context
) {
6991 aio_context_release(new_context
);
6995 static bool bdrv_parent_can_set_aio_context(BdrvChild
*c
, AioContext
*ctx
,
6996 GSList
**ignore
, Error
**errp
)
6998 if (g_slist_find(*ignore
, c
)) {
7001 *ignore
= g_slist_prepend(*ignore
, c
);
7004 * A BdrvChildClass that doesn't handle AioContext changes cannot
7005 * tolerate any AioContext changes
7007 if (!c
->klass
->can_set_aio_ctx
) {
7008 char *user
= bdrv_child_user_desc(c
);
7009 error_setg(errp
, "Changing iothreads is not supported by %s", user
);
7013 if (!c
->klass
->can_set_aio_ctx(c
, ctx
, ignore
, errp
)) {
7014 assert(!errp
|| *errp
);
7020 bool bdrv_child_can_set_aio_context(BdrvChild
*c
, AioContext
*ctx
,
7021 GSList
**ignore
, Error
**errp
)
7023 if (g_slist_find(*ignore
, c
)) {
7026 *ignore
= g_slist_prepend(*ignore
, c
);
7027 return bdrv_can_set_aio_context(c
->bs
, ctx
, ignore
, errp
);
7030 /* @ignore will accumulate all visited BdrvChild object. The caller is
7031 * responsible for freeing the list afterwards. */
7032 bool bdrv_can_set_aio_context(BlockDriverState
*bs
, AioContext
*ctx
,
7033 GSList
**ignore
, Error
**errp
)
7037 if (bdrv_get_aio_context(bs
) == ctx
) {
7041 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
7042 if (!bdrv_parent_can_set_aio_context(c
, ctx
, ignore
, errp
)) {
7046 QLIST_FOREACH(c
, &bs
->children
, next
) {
7047 if (!bdrv_child_can_set_aio_context(c
, ctx
, ignore
, errp
)) {
7055 int bdrv_child_try_set_aio_context(BlockDriverState
*bs
, AioContext
*ctx
,
7056 BdrvChild
*ignore_child
, Error
**errp
)
7061 ignore
= ignore_child
? g_slist_prepend(NULL
, ignore_child
) : NULL
;
7062 ret
= bdrv_can_set_aio_context(bs
, ctx
, &ignore
, errp
);
7063 g_slist_free(ignore
);
7069 ignore
= ignore_child
? g_slist_prepend(NULL
, ignore_child
) : NULL
;
7070 bdrv_set_aio_context_ignore(bs
, ctx
, &ignore
);
7071 g_slist_free(ignore
);
7076 int bdrv_try_set_aio_context(BlockDriverState
*bs
, AioContext
*ctx
,
7079 return bdrv_child_try_set_aio_context(bs
, ctx
, NULL
, errp
);
7082 void bdrv_add_aio_context_notifier(BlockDriverState
*bs
,
7083 void (*attached_aio_context
)(AioContext
*new_context
, void *opaque
),
7084 void (*detach_aio_context
)(void *opaque
), void *opaque
)
7086 BdrvAioNotifier
*ban
= g_new(BdrvAioNotifier
, 1);
7087 *ban
= (BdrvAioNotifier
){
7088 .attached_aio_context
= attached_aio_context
,
7089 .detach_aio_context
= detach_aio_context
,
7093 QLIST_INSERT_HEAD(&bs
->aio_notifiers
, ban
, list
);
7096 void bdrv_remove_aio_context_notifier(BlockDriverState
*bs
,
7097 void (*attached_aio_context
)(AioContext
*,
7099 void (*detach_aio_context
)(void *),
7102 BdrvAioNotifier
*ban
, *ban_next
;
7104 QLIST_FOREACH_SAFE(ban
, &bs
->aio_notifiers
, list
, ban_next
) {
7105 if (ban
->attached_aio_context
== attached_aio_context
&&
7106 ban
->detach_aio_context
== detach_aio_context
&&
7107 ban
->opaque
== opaque
&&
7108 ban
->deleted
== false)
7110 if (bs
->walking_aio_notifiers
) {
7111 ban
->deleted
= true;
7113 bdrv_do_remove_aio_context_notifier(ban
);
7122 int bdrv_amend_options(BlockDriverState
*bs
, QemuOpts
*opts
,
7123 BlockDriverAmendStatusCB
*status_cb
, void *cb_opaque
,
7128 error_setg(errp
, "Node is ejected");
7131 if (!bs
->drv
->bdrv_amend_options
) {
7132 error_setg(errp
, "Block driver '%s' does not support option amendment",
7133 bs
->drv
->format_name
);
7136 return bs
->drv
->bdrv_amend_options(bs
, opts
, status_cb
,
7137 cb_opaque
, force
, errp
);
7141 * This function checks whether the given @to_replace is allowed to be
7142 * replaced by a node that always shows the same data as @bs. This is
7143 * used for example to verify whether the mirror job can replace
7144 * @to_replace by the target mirrored from @bs.
7145 * To be replaceable, @bs and @to_replace may either be guaranteed to
7146 * always show the same data (because they are only connected through
7147 * filters), or some driver may allow replacing one of its children
7148 * because it can guarantee that this child's data is not visible at
7149 * all (for example, for dissenting quorum children that have no other
7152 bool bdrv_recurse_can_replace(BlockDriverState
*bs
,
7153 BlockDriverState
*to_replace
)
7155 BlockDriverState
*filtered
;
7157 if (!bs
|| !bs
->drv
) {
7161 if (bs
== to_replace
) {
7165 /* See what the driver can do */
7166 if (bs
->drv
->bdrv_recurse_can_replace
) {
7167 return bs
->drv
->bdrv_recurse_can_replace(bs
, to_replace
);
7170 /* For filters without an own implementation, we can recurse on our own */
7171 filtered
= bdrv_filter_bs(bs
);
7173 return bdrv_recurse_can_replace(filtered
, to_replace
);
7181 * Check whether the given @node_name can be replaced by a node that
7182 * has the same data as @parent_bs. If so, return @node_name's BDS;
7185 * @node_name must be a (recursive) *child of @parent_bs (or this
7186 * function will return NULL).
7188 * The result (whether the node can be replaced or not) is only valid
7189 * for as long as no graph or permission changes occur.
7191 BlockDriverState
*check_to_replace_node(BlockDriverState
*parent_bs
,
7192 const char *node_name
, Error
**errp
)
7194 BlockDriverState
*to_replace_bs
= bdrv_find_node(node_name
);
7195 AioContext
*aio_context
;
7197 if (!to_replace_bs
) {
7198 error_setg(errp
, "Failed to find node with node-name='%s'", node_name
);
7202 aio_context
= bdrv_get_aio_context(to_replace_bs
);
7203 aio_context_acquire(aio_context
);
7205 if (bdrv_op_is_blocked(to_replace_bs
, BLOCK_OP_TYPE_REPLACE
, errp
)) {
7206 to_replace_bs
= NULL
;
7210 /* We don't want arbitrary node of the BDS chain to be replaced only the top
7211 * most non filter in order to prevent data corruption.
7212 * Another benefit is that this tests exclude backing files which are
7213 * blocked by the backing blockers.
7215 if (!bdrv_recurse_can_replace(parent_bs
, to_replace_bs
)) {
7216 error_setg(errp
, "Cannot replace '%s' by a node mirrored from '%s', "
7217 "because it cannot be guaranteed that doing so would not "
7218 "lead to an abrupt change of visible data",
7219 node_name
, parent_bs
->node_name
);
7220 to_replace_bs
= NULL
;
7225 aio_context_release(aio_context
);
7226 return to_replace_bs
;
7230 * Iterates through the list of runtime option keys that are said to
7231 * be "strong" for a BDS. An option is called "strong" if it changes
7232 * a BDS's data. For example, the null block driver's "size" and
7233 * "read-zeroes" options are strong, but its "latency-ns" option is
7236 * If a key returned by this function ends with a dot, all options
7237 * starting with that prefix are strong.
7239 static const char *const *strong_options(BlockDriverState
*bs
,
7240 const char *const *curopt
)
7242 static const char *const global_options
[] = {
7243 "driver", "filename", NULL
7247 return &global_options
[0];
7251 if (curopt
== &global_options
[ARRAY_SIZE(global_options
) - 1] && bs
->drv
) {
7252 curopt
= bs
->drv
->strong_runtime_opts
;
7255 return (curopt
&& *curopt
) ? curopt
: NULL
;
7259 * Copies all strong runtime options from bs->options to the given
7260 * QDict. The set of strong option keys is determined by invoking
7263 * Returns true iff any strong option was present in bs->options (and
7264 * thus copied to the target QDict) with the exception of "filename"
7265 * and "driver". The caller is expected to use this value to decide
7266 * whether the existence of strong options prevents the generation of
7269 static bool append_strong_runtime_options(QDict
*d
, BlockDriverState
*bs
)
7271 bool found_any
= false;
7272 const char *const *option_name
= NULL
;
7278 while ((option_name
= strong_options(bs
, option_name
))) {
7279 bool option_given
= false;
7281 assert(strlen(*option_name
) > 0);
7282 if ((*option_name
)[strlen(*option_name
) - 1] != '.') {
7283 QObject
*entry
= qdict_get(bs
->options
, *option_name
);
7288 qdict_put_obj(d
, *option_name
, qobject_ref(entry
));
7289 option_given
= true;
7291 const QDictEntry
*entry
;
7292 for (entry
= qdict_first(bs
->options
); entry
;
7293 entry
= qdict_next(bs
->options
, entry
))
7295 if (strstart(qdict_entry_key(entry
), *option_name
, NULL
)) {
7296 qdict_put_obj(d
, qdict_entry_key(entry
),
7297 qobject_ref(qdict_entry_value(entry
)));
7298 option_given
= true;
7303 /* While "driver" and "filename" need to be included in a JSON filename,
7304 * their existence does not prohibit generation of a plain filename. */
7305 if (!found_any
&& option_given
&&
7306 strcmp(*option_name
, "driver") && strcmp(*option_name
, "filename"))
7312 if (!qdict_haskey(d
, "driver")) {
7313 /* Drivers created with bdrv_new_open_driver() may not have a
7314 * @driver option. Add it here. */
7315 qdict_put_str(d
, "driver", bs
->drv
->format_name
);
7321 /* Note: This function may return false positives; it may return true
7322 * even if opening the backing file specified by bs's image header
7323 * would result in exactly bs->backing. */
7324 bool bdrv_backing_overridden(BlockDriverState
*bs
)
7327 return strcmp(bs
->auto_backing_file
,
7328 bs
->backing
->bs
->filename
);
7330 /* No backing BDS, so if the image header reports any backing
7331 * file, it must have been suppressed */
7332 return bs
->auto_backing_file
[0] != '\0';
7336 /* Updates the following BDS fields:
7337 * - exact_filename: A filename which may be used for opening a block device
7338 * which (mostly) equals the given BDS (even without any
7339 * other options; so reading and writing must return the same
7340 * results, but caching etc. may be different)
7341 * - full_open_options: Options which, when given when opening a block device
7342 * (without a filename), result in a BDS (mostly)
7343 * equalling the given one
7344 * - filename: If exact_filename is set, it is copied here. Otherwise,
7345 * full_open_options is converted to a JSON object, prefixed with
7346 * "json:" (for use through the JSON pseudo protocol) and put here.
7348 void bdrv_refresh_filename(BlockDriverState
*bs
)
7350 BlockDriver
*drv
= bs
->drv
;
7352 BlockDriverState
*primary_child_bs
;
7354 bool backing_overridden
;
7355 bool generate_json_filename
; /* Whether our default implementation should
7356 fill exact_filename (false) or not (true) */
7362 /* This BDS's file name may depend on any of its children's file names, so
7363 * refresh those first */
7364 QLIST_FOREACH(child
, &bs
->children
, next
) {
7365 bdrv_refresh_filename(child
->bs
);
7369 /* For implicit nodes, just copy everything from the single child */
7370 child
= QLIST_FIRST(&bs
->children
);
7371 assert(QLIST_NEXT(child
, next
) == NULL
);
7373 pstrcpy(bs
->exact_filename
, sizeof(bs
->exact_filename
),
7374 child
->bs
->exact_filename
);
7375 pstrcpy(bs
->filename
, sizeof(bs
->filename
), child
->bs
->filename
);
7377 qobject_unref(bs
->full_open_options
);
7378 bs
->full_open_options
= qobject_ref(child
->bs
->full_open_options
);
7383 backing_overridden
= bdrv_backing_overridden(bs
);
7385 if (bs
->open_flags
& BDRV_O_NO_IO
) {
7386 /* Without I/O, the backing file does not change anything.
7387 * Therefore, in such a case (primarily qemu-img), we can
7388 * pretend the backing file has not been overridden even if
7389 * it technically has been. */
7390 backing_overridden
= false;
7393 /* Gather the options QDict */
7395 generate_json_filename
= append_strong_runtime_options(opts
, bs
);
7396 generate_json_filename
|= backing_overridden
;
7398 if (drv
->bdrv_gather_child_options
) {
7399 /* Some block drivers may not want to present all of their children's
7400 * options, or name them differently from BdrvChild.name */
7401 drv
->bdrv_gather_child_options(bs
, opts
, backing_overridden
);
7403 QLIST_FOREACH(child
, &bs
->children
, next
) {
7404 if (child
== bs
->backing
&& !backing_overridden
) {
7405 /* We can skip the backing BDS if it has not been overridden */
7409 qdict_put(opts
, child
->name
,
7410 qobject_ref(child
->bs
->full_open_options
));
7413 if (backing_overridden
&& !bs
->backing
) {
7414 /* Force no backing file */
7415 qdict_put_null(opts
, "backing");
7419 qobject_unref(bs
->full_open_options
);
7420 bs
->full_open_options
= opts
;
7422 primary_child_bs
= bdrv_primary_bs(bs
);
7424 if (drv
->bdrv_refresh_filename
) {
7425 /* Obsolete information is of no use here, so drop the old file name
7426 * information before refreshing it */
7427 bs
->exact_filename
[0] = '\0';
7429 drv
->bdrv_refresh_filename(bs
);
7430 } else if (primary_child_bs
) {
7432 * Try to reconstruct valid information from the underlying
7433 * file -- this only works for format nodes (filter nodes
7434 * cannot be probed and as such must be selected by the user
7435 * either through an options dict, or through a special
7436 * filename which the filter driver must construct in its
7437 * .bdrv_refresh_filename() implementation).
7440 bs
->exact_filename
[0] = '\0';
7443 * We can use the underlying file's filename if:
7444 * - it has a filename,
7445 * - the current BDS is not a filter,
7446 * - the file is a protocol BDS, and
7447 * - opening that file (as this BDS's format) will automatically create
7448 * the BDS tree we have right now, that is:
7449 * - the user did not significantly change this BDS's behavior with
7450 * some explicit (strong) options
7451 * - no non-file child of this BDS has been overridden by the user
7452 * Both of these conditions are represented by generate_json_filename.
7454 if (primary_child_bs
->exact_filename
[0] &&
7455 primary_child_bs
->drv
->bdrv_file_open
&&
7456 !drv
->is_filter
&& !generate_json_filename
)
7458 strcpy(bs
->exact_filename
, primary_child_bs
->exact_filename
);
7462 if (bs
->exact_filename
[0]) {
7463 pstrcpy(bs
->filename
, sizeof(bs
->filename
), bs
->exact_filename
);
7465 GString
*json
= qobject_to_json(QOBJECT(bs
->full_open_options
));
7466 if (snprintf(bs
->filename
, sizeof(bs
->filename
), "json:%s",
7467 json
->str
) >= sizeof(bs
->filename
)) {
7468 /* Give user a hint if we truncated things. */
7469 strcpy(bs
->filename
+ sizeof(bs
->filename
) - 4, "...");
7471 g_string_free(json
, true);
7475 char *bdrv_dirname(BlockDriverState
*bs
, Error
**errp
)
7477 BlockDriver
*drv
= bs
->drv
;
7478 BlockDriverState
*child_bs
;
7481 error_setg(errp
, "Node '%s' is ejected", bs
->node_name
);
7485 if (drv
->bdrv_dirname
) {
7486 return drv
->bdrv_dirname(bs
, errp
);
7489 child_bs
= bdrv_primary_bs(bs
);
7491 return bdrv_dirname(child_bs
, errp
);
7494 bdrv_refresh_filename(bs
);
7495 if (bs
->exact_filename
[0] != '\0') {
7496 return path_combine(bs
->exact_filename
, "");
7499 error_setg(errp
, "Cannot generate a base directory for %s nodes",
7505 * Hot add/remove a BDS's child. So the user can take a child offline when
7506 * it is broken and take a new child online
7508 void bdrv_add_child(BlockDriverState
*parent_bs
, BlockDriverState
*child_bs
,
7512 if (!parent_bs
->drv
|| !parent_bs
->drv
->bdrv_add_child
) {
7513 error_setg(errp
, "The node %s does not support adding a child",
7514 bdrv_get_device_or_node_name(parent_bs
));
7518 if (!QLIST_EMPTY(&child_bs
->parents
)) {
7519 error_setg(errp
, "The node %s already has a parent",
7520 child_bs
->node_name
);
7524 parent_bs
->drv
->bdrv_add_child(parent_bs
, child_bs
, errp
);
7527 void bdrv_del_child(BlockDriverState
*parent_bs
, BdrvChild
*child
, Error
**errp
)
7531 if (!parent_bs
->drv
|| !parent_bs
->drv
->bdrv_del_child
) {
7532 error_setg(errp
, "The node %s does not support removing a child",
7533 bdrv_get_device_or_node_name(parent_bs
));
7537 QLIST_FOREACH(tmp
, &parent_bs
->children
, next
) {
7544 error_setg(errp
, "The node %s does not have a child named %s",
7545 bdrv_get_device_or_node_name(parent_bs
),
7546 bdrv_get_device_or_node_name(child
->bs
));
7550 parent_bs
->drv
->bdrv_del_child(parent_bs
, child
, errp
);
7553 int bdrv_make_empty(BdrvChild
*c
, Error
**errp
)
7555 BlockDriver
*drv
= c
->bs
->drv
;
7558 assert(c
->perm
& (BLK_PERM_WRITE
| BLK_PERM_WRITE_UNCHANGED
));
7560 if (!drv
->bdrv_make_empty
) {
7561 error_setg(errp
, "%s does not support emptying nodes",
7566 ret
= drv
->bdrv_make_empty(c
->bs
);
7568 error_setg_errno(errp
, -ret
, "Failed to empty %s",
7577 * Return the child that @bs acts as an overlay for, and from which data may be
7578 * copied in COW or COR operations. Usually this is the backing file.
7580 BdrvChild
*bdrv_cow_child(BlockDriverState
*bs
)
7582 if (!bs
|| !bs
->drv
) {
7586 if (bs
->drv
->is_filter
) {
7594 assert(bs
->backing
->role
& BDRV_CHILD_COW
);
7599 * If @bs acts as a filter for exactly one of its children, return
7602 BdrvChild
*bdrv_filter_child(BlockDriverState
*bs
)
7606 if (!bs
|| !bs
->drv
) {
7610 if (!bs
->drv
->is_filter
) {
7614 /* Only one of @backing or @file may be used */
7615 assert(!(bs
->backing
&& bs
->file
));
7617 c
= bs
->backing
?: bs
->file
;
7622 assert(c
->role
& BDRV_CHILD_FILTERED
);
7627 * Return either the result of bdrv_cow_child() or bdrv_filter_child(),
7628 * whichever is non-NULL.
7630 * Return NULL if both are NULL.
7632 BdrvChild
*bdrv_filter_or_cow_child(BlockDriverState
*bs
)
7634 BdrvChild
*cow_child
= bdrv_cow_child(bs
);
7635 BdrvChild
*filter_child
= bdrv_filter_child(bs
);
7637 /* Filter nodes cannot have COW backing files */
7638 assert(!(cow_child
&& filter_child
));
7640 return cow_child
?: filter_child
;
7644 * Return the primary child of this node: For filters, that is the
7645 * filtered child. For other nodes, that is usually the child storing
7647 * (A generally more helpful description is that this is (usually) the
7648 * child that has the same filename as @bs.)
7650 * Drivers do not necessarily have a primary child; for example quorum
7653 BdrvChild
*bdrv_primary_child(BlockDriverState
*bs
)
7655 BdrvChild
*c
, *found
= NULL
;
7657 QLIST_FOREACH(c
, &bs
->children
, next
) {
7658 if (c
->role
& BDRV_CHILD_PRIMARY
) {
7667 static BlockDriverState
*bdrv_do_skip_filters(BlockDriverState
*bs
,
7668 bool stop_on_explicit_filter
)
7676 while (!(stop_on_explicit_filter
&& !bs
->implicit
)) {
7677 c
= bdrv_filter_child(bs
);
7680 * A filter that is embedded in a working block graph must
7681 * have a child. Assert this here so this function does
7682 * not return a filter node that is not expected by the
7685 assert(!bs
->drv
|| !bs
->drv
->is_filter
);
7691 * Note that this treats nodes with bs->drv == NULL as not being
7692 * filters (bs->drv == NULL should be replaced by something else
7694 * The advantage of this behavior is that this function will thus
7695 * always return a non-NULL value (given a non-NULL @bs).
7702 * Return the first BDS that has not been added implicitly or that
7703 * does not have a filtered child down the chain starting from @bs
7704 * (including @bs itself).
7706 BlockDriverState
*bdrv_skip_implicit_filters(BlockDriverState
*bs
)
7708 return bdrv_do_skip_filters(bs
, true);
7712 * Return the first BDS that does not have a filtered child down the
7713 * chain starting from @bs (including @bs itself).
7715 BlockDriverState
*bdrv_skip_filters(BlockDriverState
*bs
)
7717 return bdrv_do_skip_filters(bs
, false);
7721 * For a backing chain, return the first non-filter backing image of
7722 * the first non-filter image.
7724 BlockDriverState
*bdrv_backing_chain_next(BlockDriverState
*bs
)
7726 return bdrv_skip_filters(bdrv_cow_bs(bdrv_skip_filters(bs
)));