2 * QEMU System Emulator block driver
4 * Copyright (c) 2003 Fabrice Bellard
5 * Copyright (c) 2020 Virtuozzo International GmbH.
7 * Permission is hereby granted, free of charge, to any person obtaining a copy
8 * of this software and associated documentation files (the "Software"), to deal
9 * in the Software without restriction, including without limitation the rights
10 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
11 * copies of the Software, and to permit persons to whom the Software is
12 * furnished to do so, subject to the following conditions:
14 * The above copyright notice and this permission notice shall be included in
15 * all copies or substantial portions of the Software.
17 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
18 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
19 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
20 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
21 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
22 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
26 #include "qemu/osdep.h"
27 #include "block/trace.h"
28 #include "block/block_int.h"
29 #include "block/blockjob.h"
30 #include "block/fuse.h"
31 #include "block/nbd.h"
32 #include "block/qdict.h"
33 #include "qemu/error-report.h"
34 #include "block/module_block.h"
35 #include "qemu/main-loop.h"
36 #include "qemu/module.h"
37 #include "qapi/error.h"
38 #include "qapi/qmp/qdict.h"
39 #include "qapi/qmp/qjson.h"
40 #include "qapi/qmp/qnull.h"
41 #include "qapi/qmp/qstring.h"
42 #include "qapi/qobject-output-visitor.h"
43 #include "qapi/qapi-visit-block-core.h"
44 #include "sysemu/block-backend.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 "block/coroutines.h"
55 #include <sys/ioctl.h>
56 #include <sys/queue.h>
57 #if defined(HAVE_SYS_DISK_H)
66 #define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */
68 static QTAILQ_HEAD(, BlockDriverState
) graph_bdrv_states
=
69 QTAILQ_HEAD_INITIALIZER(graph_bdrv_states
);
71 static QTAILQ_HEAD(, BlockDriverState
) all_bdrv_states
=
72 QTAILQ_HEAD_INITIALIZER(all_bdrv_states
);
74 static QLIST_HEAD(, BlockDriver
) bdrv_drivers
=
75 QLIST_HEAD_INITIALIZER(bdrv_drivers
);
77 static BlockDriverState
*bdrv_open_inherit(const char *filename
,
78 const char *reference
,
79 QDict
*options
, int flags
,
80 BlockDriverState
*parent
,
81 const BdrvChildClass
*child_class
,
82 BdrvChildRole child_role
,
85 static void bdrv_replace_child_noperm(BdrvChild
*child
,
86 BlockDriverState
*new_bs
);
87 static void bdrv_remove_file_or_backing_child(BlockDriverState
*bs
,
90 static void bdrv_remove_filter_or_cow_child(BlockDriverState
*bs
,
93 static int bdrv_reopen_prepare(BDRVReopenState
*reopen_state
,
94 BlockReopenQueue
*queue
,
95 Transaction
*set_backings_tran
, Error
**errp
);
96 static void bdrv_reopen_commit(BDRVReopenState
*reopen_state
);
97 static void bdrv_reopen_abort(BDRVReopenState
*reopen_state
);
99 /* If non-zero, use only whitelisted block drivers */
100 static int use_bdrv_whitelist
;
103 static int is_windows_drive_prefix(const char *filename
)
105 return (((filename
[0] >= 'a' && filename
[0] <= 'z') ||
106 (filename
[0] >= 'A' && filename
[0] <= 'Z')) &&
110 int is_windows_drive(const char *filename
)
112 if (is_windows_drive_prefix(filename
) &&
115 if (strstart(filename
, "\\\\.\\", NULL
) ||
116 strstart(filename
, "//./", NULL
))
122 size_t bdrv_opt_mem_align(BlockDriverState
*bs
)
124 if (!bs
|| !bs
->drv
) {
125 /* page size or 4k (hdd sector size) should be on the safe side */
126 return MAX(4096, qemu_real_host_page_size
);
129 return bs
->bl
.opt_mem_alignment
;
132 size_t bdrv_min_mem_align(BlockDriverState
*bs
)
134 if (!bs
|| !bs
->drv
) {
135 /* page size or 4k (hdd sector size) should be on the safe side */
136 return MAX(4096, qemu_real_host_page_size
);
139 return bs
->bl
.min_mem_alignment
;
142 /* check if the path starts with "<protocol>:" */
143 int path_has_protocol(const char *path
)
148 if (is_windows_drive(path
) ||
149 is_windows_drive_prefix(path
)) {
152 p
= path
+ strcspn(path
, ":/\\");
154 p
= path
+ strcspn(path
, ":/");
160 int path_is_absolute(const char *path
)
163 /* specific case for names like: "\\.\d:" */
164 if (is_windows_drive(path
) || is_windows_drive_prefix(path
)) {
167 return (*path
== '/' || *path
== '\\');
169 return (*path
== '/');
173 /* if filename is absolute, just return its duplicate. Otherwise, build a
174 path to it by considering it is relative to base_path. URL are
176 char *path_combine(const char *base_path
, const char *filename
)
178 const char *protocol_stripped
= NULL
;
183 if (path_is_absolute(filename
)) {
184 return g_strdup(filename
);
187 if (path_has_protocol(base_path
)) {
188 protocol_stripped
= strchr(base_path
, ':');
189 if (protocol_stripped
) {
193 p
= protocol_stripped
?: base_path
;
195 p1
= strrchr(base_path
, '/');
199 p2
= strrchr(base_path
, '\\');
200 if (!p1
|| p2
> p1
) {
215 result
= g_malloc(len
+ strlen(filename
) + 1);
216 memcpy(result
, base_path
, len
);
217 strcpy(result
+ len
, filename
);
223 * Helper function for bdrv_parse_filename() implementations to remove optional
224 * protocol prefixes (especially "file:") from a filename and for putting the
225 * stripped filename into the options QDict if there is such a prefix.
227 void bdrv_parse_filename_strip_prefix(const char *filename
, const char *prefix
,
230 if (strstart(filename
, prefix
, &filename
)) {
231 /* Stripping the explicit protocol prefix may result in a protocol
232 * prefix being (wrongly) detected (if the filename contains a colon) */
233 if (path_has_protocol(filename
)) {
234 GString
*fat_filename
;
236 /* This means there is some colon before the first slash; therefore,
237 * this cannot be an absolute path */
238 assert(!path_is_absolute(filename
));
240 /* And we can thus fix the protocol detection issue by prefixing it
242 fat_filename
= g_string_new("./");
243 g_string_append(fat_filename
, filename
);
245 assert(!path_has_protocol(fat_filename
->str
));
247 qdict_put(options
, "filename",
248 qstring_from_gstring(fat_filename
));
250 /* If no protocol prefix was detected, we can use the shortened
252 qdict_put_str(options
, "filename", filename
);
258 /* Returns whether the image file is opened as read-only. Note that this can
259 * return false and writing to the image file is still not possible because the
260 * image is inactivated. */
261 bool bdrv_is_read_only(BlockDriverState
*bs
)
263 return !(bs
->open_flags
& BDRV_O_RDWR
);
266 int bdrv_can_set_read_only(BlockDriverState
*bs
, bool read_only
,
267 bool ignore_allow_rdw
, Error
**errp
)
269 /* Do not set read_only if copy_on_read is enabled */
270 if (bs
->copy_on_read
&& read_only
) {
271 error_setg(errp
, "Can't set node '%s' to r/o with copy-on-read enabled",
272 bdrv_get_device_or_node_name(bs
));
276 /* Do not clear read_only if it is prohibited */
277 if (!read_only
&& !(bs
->open_flags
& BDRV_O_ALLOW_RDWR
) &&
280 error_setg(errp
, "Node '%s' is read only",
281 bdrv_get_device_or_node_name(bs
));
289 * Called by a driver that can only provide a read-only image.
291 * Returns 0 if the node is already read-only or it could switch the node to
292 * read-only because BDRV_O_AUTO_RDONLY is set.
294 * Returns -EACCES if the node is read-write and BDRV_O_AUTO_RDONLY is not set
295 * or bdrv_can_set_read_only() forbids making the node read-only. If @errmsg
296 * is not NULL, it is used as the error message for the Error object.
298 int bdrv_apply_auto_read_only(BlockDriverState
*bs
, const char *errmsg
,
303 if (!(bs
->open_flags
& BDRV_O_RDWR
)) {
306 if (!(bs
->open_flags
& BDRV_O_AUTO_RDONLY
)) {
310 ret
= bdrv_can_set_read_only(bs
, true, false, NULL
);
315 bs
->open_flags
&= ~BDRV_O_RDWR
;
320 error_setg(errp
, "%s", errmsg
?: "Image is read-only");
325 * If @backing is empty, this function returns NULL without setting
326 * @errp. In all other cases, NULL will only be returned with @errp
329 * Therefore, a return value of NULL without @errp set means that
330 * there is no backing file; if @errp is set, there is one but its
331 * absolute filename cannot be generated.
333 char *bdrv_get_full_backing_filename_from_filename(const char *backed
,
337 if (backing
[0] == '\0') {
339 } else if (path_has_protocol(backing
) || path_is_absolute(backing
)) {
340 return g_strdup(backing
);
341 } else if (backed
[0] == '\0' || strstart(backed
, "json:", NULL
)) {
342 error_setg(errp
, "Cannot use relative backing file names for '%s'",
346 return path_combine(backed
, backing
);
351 * If @filename is empty or NULL, this function returns NULL without
352 * setting @errp. In all other cases, NULL will only be returned with
355 static char *bdrv_make_absolute_filename(BlockDriverState
*relative_to
,
356 const char *filename
, Error
**errp
)
358 char *dir
, *full_name
;
360 if (!filename
|| filename
[0] == '\0') {
362 } else if (path_has_protocol(filename
) || path_is_absolute(filename
)) {
363 return g_strdup(filename
);
366 dir
= bdrv_dirname(relative_to
, errp
);
371 full_name
= g_strconcat(dir
, filename
, NULL
);
376 char *bdrv_get_full_backing_filename(BlockDriverState
*bs
, Error
**errp
)
378 return bdrv_make_absolute_filename(bs
, bs
->backing_file
, errp
);
381 void bdrv_register(BlockDriver
*bdrv
)
383 assert(bdrv
->format_name
);
384 QLIST_INSERT_HEAD(&bdrv_drivers
, bdrv
, list
);
387 BlockDriverState
*bdrv_new(void)
389 BlockDriverState
*bs
;
392 bs
= g_new0(BlockDriverState
, 1);
393 QLIST_INIT(&bs
->dirty_bitmaps
);
394 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
395 QLIST_INIT(&bs
->op_blockers
[i
]);
397 qemu_co_mutex_init(&bs
->reqs_lock
);
398 qemu_mutex_init(&bs
->dirty_bitmap_mutex
);
400 bs
->aio_context
= qemu_get_aio_context();
402 qemu_co_queue_init(&bs
->flush_queue
);
404 for (i
= 0; i
< bdrv_drain_all_count
; i
++) {
405 bdrv_drained_begin(bs
);
408 QTAILQ_INSERT_TAIL(&all_bdrv_states
, bs
, bs_list
);
413 static BlockDriver
*bdrv_do_find_format(const char *format_name
)
417 QLIST_FOREACH(drv1
, &bdrv_drivers
, list
) {
418 if (!strcmp(drv1
->format_name
, format_name
)) {
426 BlockDriver
*bdrv_find_format(const char *format_name
)
431 drv1
= bdrv_do_find_format(format_name
);
436 /* The driver isn't registered, maybe we need to load a module */
437 for (i
= 0; i
< (int)ARRAY_SIZE(block_driver_modules
); ++i
) {
438 if (!strcmp(block_driver_modules
[i
].format_name
, format_name
)) {
439 block_module_load_one(block_driver_modules
[i
].library_name
);
444 return bdrv_do_find_format(format_name
);
447 static int bdrv_format_is_whitelisted(const char *format_name
, bool read_only
)
449 static const char *whitelist_rw
[] = {
450 CONFIG_BDRV_RW_WHITELIST
453 static const char *whitelist_ro
[] = {
454 CONFIG_BDRV_RO_WHITELIST
459 if (!whitelist_rw
[0] && !whitelist_ro
[0]) {
460 return 1; /* no whitelist, anything goes */
463 for (p
= whitelist_rw
; *p
; p
++) {
464 if (!strcmp(format_name
, *p
)) {
469 for (p
= whitelist_ro
; *p
; p
++) {
470 if (!strcmp(format_name
, *p
)) {
478 int bdrv_is_whitelisted(BlockDriver
*drv
, bool read_only
)
480 return bdrv_format_is_whitelisted(drv
->format_name
, read_only
);
483 bool bdrv_uses_whitelist(void)
485 return use_bdrv_whitelist
;
488 typedef struct CreateCo
{
496 static void coroutine_fn
bdrv_create_co_entry(void *opaque
)
498 Error
*local_err
= NULL
;
501 CreateCo
*cco
= opaque
;
504 ret
= cco
->drv
->bdrv_co_create_opts(cco
->drv
,
505 cco
->filename
, cco
->opts
, &local_err
);
506 error_propagate(&cco
->err
, local_err
);
510 int bdrv_create(BlockDriver
*drv
, const char* filename
,
511 QemuOpts
*opts
, Error
**errp
)
518 .filename
= g_strdup(filename
),
524 if (!drv
->bdrv_co_create_opts
) {
525 error_setg(errp
, "Driver '%s' does not support image creation", drv
->format_name
);
530 if (qemu_in_coroutine()) {
531 /* Fast-path if already in coroutine context */
532 bdrv_create_co_entry(&cco
);
534 co
= qemu_coroutine_create(bdrv_create_co_entry
, &cco
);
535 qemu_coroutine_enter(co
);
536 while (cco
.ret
== NOT_DONE
) {
537 aio_poll(qemu_get_aio_context(), true);
544 error_propagate(errp
, cco
.err
);
546 error_setg_errno(errp
, -ret
, "Could not create image");
551 g_free(cco
.filename
);
556 * Helper function for bdrv_create_file_fallback(): Resize @blk to at
557 * least the given @minimum_size.
559 * On success, return @blk's actual length.
560 * Otherwise, return -errno.
562 static int64_t create_file_fallback_truncate(BlockBackend
*blk
,
563 int64_t minimum_size
, Error
**errp
)
565 Error
*local_err
= NULL
;
569 ret
= blk_truncate(blk
, minimum_size
, false, PREALLOC_MODE_OFF
, 0,
571 if (ret
< 0 && ret
!= -ENOTSUP
) {
572 error_propagate(errp
, local_err
);
576 size
= blk_getlength(blk
);
578 error_free(local_err
);
579 error_setg_errno(errp
, -size
,
580 "Failed to inquire the new image file's length");
584 if (size
< minimum_size
) {
585 /* Need to grow the image, but we failed to do that */
586 error_propagate(errp
, local_err
);
590 error_free(local_err
);
597 * Helper function for bdrv_create_file_fallback(): Zero the first
598 * sector to remove any potentially pre-existing image header.
600 static int create_file_fallback_zero_first_sector(BlockBackend
*blk
,
601 int64_t current_size
,
604 int64_t bytes_to_clear
;
607 bytes_to_clear
= MIN(current_size
, BDRV_SECTOR_SIZE
);
608 if (bytes_to_clear
) {
609 ret
= blk_pwrite_zeroes(blk
, 0, bytes_to_clear
, BDRV_REQ_MAY_UNMAP
);
611 error_setg_errno(errp
, -ret
,
612 "Failed to clear the new image's first sector");
621 * Simple implementation of bdrv_co_create_opts for protocol drivers
622 * which only support creation via opening a file
623 * (usually existing raw storage device)
625 int coroutine_fn
bdrv_co_create_opts_simple(BlockDriver
*drv
,
626 const char *filename
,
634 PreallocMode prealloc
;
635 Error
*local_err
= NULL
;
638 size
= qemu_opt_get_size_del(opts
, BLOCK_OPT_SIZE
, 0);
639 buf
= qemu_opt_get_del(opts
, BLOCK_OPT_PREALLOC
);
640 prealloc
= qapi_enum_parse(&PreallocMode_lookup
, buf
,
641 PREALLOC_MODE_OFF
, &local_err
);
644 error_propagate(errp
, local_err
);
648 if (prealloc
!= PREALLOC_MODE_OFF
) {
649 error_setg(errp
, "Unsupported preallocation mode '%s'",
650 PreallocMode_str(prealloc
));
654 options
= qdict_new();
655 qdict_put_str(options
, "driver", drv
->format_name
);
657 blk
= blk_new_open(filename
, NULL
, options
,
658 BDRV_O_RDWR
| BDRV_O_RESIZE
, errp
);
660 error_prepend(errp
, "Protocol driver '%s' does not support image "
661 "creation, and opening the image failed: ",
666 size
= create_file_fallback_truncate(blk
, size
, errp
);
672 ret
= create_file_fallback_zero_first_sector(blk
, size
, errp
);
683 int bdrv_create_file(const char *filename
, QemuOpts
*opts
, Error
**errp
)
685 QemuOpts
*protocol_opts
;
690 drv
= bdrv_find_protocol(filename
, true, errp
);
695 if (!drv
->create_opts
) {
696 error_setg(errp
, "Driver '%s' does not support image creation",
702 * 'opts' contains a QemuOptsList with a combination of format and protocol
705 * The format properly removes its options, but the default values remain
706 * in 'opts->list'. So if the protocol has options with the same name
707 * (e.g. rbd has 'cluster_size' as qcow2), it will see the default values
708 * of the format, since for overlapping options, the format wins.
710 * To avoid this issue, lets convert QemuOpts to QDict, in this way we take
711 * only the set options, and then convert it back to QemuOpts, using the
712 * create_opts of the protocol. So the new QemuOpts, will contain only the
715 qdict
= qemu_opts_to_qdict(opts
, NULL
);
716 protocol_opts
= qemu_opts_from_qdict(drv
->create_opts
, qdict
, errp
);
717 if (protocol_opts
== NULL
) {
722 ret
= bdrv_create(drv
, filename
, protocol_opts
, errp
);
724 qemu_opts_del(protocol_opts
);
725 qobject_unref(qdict
);
729 int coroutine_fn
bdrv_co_delete_file(BlockDriverState
*bs
, Error
**errp
)
731 Error
*local_err
= NULL
;
737 error_setg(errp
, "Block node '%s' is not opened", bs
->filename
);
741 if (!bs
->drv
->bdrv_co_delete_file
) {
742 error_setg(errp
, "Driver '%s' does not support image deletion",
743 bs
->drv
->format_name
);
747 ret
= bs
->drv
->bdrv_co_delete_file(bs
, &local_err
);
749 error_propagate(errp
, local_err
);
755 void coroutine_fn
bdrv_co_delete_file_noerr(BlockDriverState
*bs
)
757 Error
*local_err
= NULL
;
764 ret
= bdrv_co_delete_file(bs
, &local_err
);
766 * ENOTSUP will happen if the block driver doesn't support
767 * the 'bdrv_co_delete_file' interface. This is a predictable
768 * scenario and shouldn't be reported back to the user.
770 if (ret
== -ENOTSUP
) {
771 error_free(local_err
);
772 } else if (ret
< 0) {
773 error_report_err(local_err
);
778 * Try to get @bs's logical and physical block size.
779 * On success, store them in @bsz struct and return 0.
780 * On failure return -errno.
781 * @bs must not be empty.
783 int bdrv_probe_blocksizes(BlockDriverState
*bs
, BlockSizes
*bsz
)
785 BlockDriver
*drv
= bs
->drv
;
786 BlockDriverState
*filtered
= bdrv_filter_bs(bs
);
788 if (drv
&& drv
->bdrv_probe_blocksizes
) {
789 return drv
->bdrv_probe_blocksizes(bs
, bsz
);
790 } else if (filtered
) {
791 return bdrv_probe_blocksizes(filtered
, bsz
);
798 * Try to get @bs's geometry (cyls, heads, sectors).
799 * On success, store them in @geo struct and return 0.
800 * On failure return -errno.
801 * @bs must not be empty.
803 int bdrv_probe_geometry(BlockDriverState
*bs
, HDGeometry
*geo
)
805 BlockDriver
*drv
= bs
->drv
;
806 BlockDriverState
*filtered
= bdrv_filter_bs(bs
);
808 if (drv
&& drv
->bdrv_probe_geometry
) {
809 return drv
->bdrv_probe_geometry(bs
, geo
);
810 } else if (filtered
) {
811 return bdrv_probe_geometry(filtered
, geo
);
818 * Create a uniquely-named empty temporary file.
819 * Return 0 upon success, otherwise a negative errno value.
821 int get_tmp_filename(char *filename
, int size
)
824 char temp_dir
[MAX_PATH
];
825 /* GetTempFileName requires that its output buffer (4th param)
826 have length MAX_PATH or greater. */
827 assert(size
>= MAX_PATH
);
828 return (GetTempPath(MAX_PATH
, temp_dir
)
829 && GetTempFileName(temp_dir
, "qem", 0, filename
)
830 ? 0 : -GetLastError());
834 tmpdir
= getenv("TMPDIR");
838 if (snprintf(filename
, size
, "%s/vl.XXXXXX", tmpdir
) >= size
) {
841 fd
= mkstemp(filename
);
845 if (close(fd
) != 0) {
854 * Detect host devices. By convention, /dev/cdrom[N] is always
855 * recognized as a host CDROM.
857 static BlockDriver
*find_hdev_driver(const char *filename
)
859 int score_max
= 0, score
;
860 BlockDriver
*drv
= NULL
, *d
;
862 QLIST_FOREACH(d
, &bdrv_drivers
, list
) {
863 if (d
->bdrv_probe_device
) {
864 score
= d
->bdrv_probe_device(filename
);
865 if (score
> score_max
) {
875 static BlockDriver
*bdrv_do_find_protocol(const char *protocol
)
879 QLIST_FOREACH(drv1
, &bdrv_drivers
, list
) {
880 if (drv1
->protocol_name
&& !strcmp(drv1
->protocol_name
, protocol
)) {
888 BlockDriver
*bdrv_find_protocol(const char *filename
,
889 bool allow_protocol_prefix
,
898 /* TODO Drivers without bdrv_file_open must be specified explicitly */
901 * XXX(hch): we really should not let host device detection
902 * override an explicit protocol specification, but moving this
903 * later breaks access to device names with colons in them.
904 * Thanks to the brain-dead persistent naming schemes on udev-
905 * based Linux systems those actually are quite common.
907 drv1
= find_hdev_driver(filename
);
912 if (!path_has_protocol(filename
) || !allow_protocol_prefix
) {
916 p
= strchr(filename
, ':');
919 if (len
> sizeof(protocol
) - 1)
920 len
= sizeof(protocol
) - 1;
921 memcpy(protocol
, filename
, len
);
922 protocol
[len
] = '\0';
924 drv1
= bdrv_do_find_protocol(protocol
);
929 for (i
= 0; i
< (int)ARRAY_SIZE(block_driver_modules
); ++i
) {
930 if (block_driver_modules
[i
].protocol_name
&&
931 !strcmp(block_driver_modules
[i
].protocol_name
, protocol
)) {
932 block_module_load_one(block_driver_modules
[i
].library_name
);
937 drv1
= bdrv_do_find_protocol(protocol
);
939 error_setg(errp
, "Unknown protocol '%s'", protocol
);
945 * Guess image format by probing its contents.
946 * This is not a good idea when your image is raw (CVE-2008-2004), but
947 * we do it anyway for backward compatibility.
949 * @buf contains the image's first @buf_size bytes.
950 * @buf_size is the buffer size in bytes (generally BLOCK_PROBE_BUF_SIZE,
951 * but can be smaller if the image file is smaller)
952 * @filename is its filename.
954 * For all block drivers, call the bdrv_probe() method to get its
956 * Return the first block driver with the highest probing score.
958 BlockDriver
*bdrv_probe_all(const uint8_t *buf
, int buf_size
,
959 const char *filename
)
961 int score_max
= 0, score
;
962 BlockDriver
*drv
= NULL
, *d
;
964 QLIST_FOREACH(d
, &bdrv_drivers
, list
) {
966 score
= d
->bdrv_probe(buf
, buf_size
, filename
);
967 if (score
> score_max
) {
977 static int find_image_format(BlockBackend
*file
, const char *filename
,
978 BlockDriver
**pdrv
, Error
**errp
)
981 uint8_t buf
[BLOCK_PROBE_BUF_SIZE
];
984 /* Return the raw BlockDriver * to scsi-generic devices or empty drives */
985 if (blk_is_sg(file
) || !blk_is_inserted(file
) || blk_getlength(file
) == 0) {
990 ret
= blk_pread(file
, 0, buf
, sizeof(buf
));
992 error_setg_errno(errp
, -ret
, "Could not read image for determining its "
998 drv
= bdrv_probe_all(buf
, ret
, filename
);
1000 error_setg(errp
, "Could not determine image format: No compatible "
1009 * Set the current 'total_sectors' value
1010 * Return 0 on success, -errno on error.
1012 int refresh_total_sectors(BlockDriverState
*bs
, int64_t hint
)
1014 BlockDriver
*drv
= bs
->drv
;
1020 /* Do not attempt drv->bdrv_getlength() on scsi-generic devices */
1024 /* query actual device if possible, otherwise just trust the hint */
1025 if (drv
->bdrv_getlength
) {
1026 int64_t length
= drv
->bdrv_getlength(bs
);
1030 hint
= DIV_ROUND_UP(length
, BDRV_SECTOR_SIZE
);
1033 bs
->total_sectors
= hint
;
1035 if (bs
->total_sectors
* BDRV_SECTOR_SIZE
> BDRV_MAX_LENGTH
) {
1043 * Combines a QDict of new block driver @options with any missing options taken
1044 * from @old_options, so that leaving out an option defaults to its old value.
1046 static void bdrv_join_options(BlockDriverState
*bs
, QDict
*options
,
1049 if (bs
->drv
&& bs
->drv
->bdrv_join_options
) {
1050 bs
->drv
->bdrv_join_options(options
, old_options
);
1052 qdict_join(options
, old_options
, false);
1056 static BlockdevDetectZeroesOptions
bdrv_parse_detect_zeroes(QemuOpts
*opts
,
1060 Error
*local_err
= NULL
;
1061 char *value
= qemu_opt_get_del(opts
, "detect-zeroes");
1062 BlockdevDetectZeroesOptions detect_zeroes
=
1063 qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup
, value
,
1064 BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF
, &local_err
);
1067 error_propagate(errp
, local_err
);
1068 return detect_zeroes
;
1071 if (detect_zeroes
== BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP
&&
1072 !(open_flags
& BDRV_O_UNMAP
))
1074 error_setg(errp
, "setting detect-zeroes to unmap is not allowed "
1075 "without setting discard operation to unmap");
1078 return detect_zeroes
;
1082 * Set open flags for aio engine
1084 * Return 0 on success, -1 if the engine specified is invalid
1086 int bdrv_parse_aio(const char *mode
, int *flags
)
1088 if (!strcmp(mode
, "threads")) {
1089 /* do nothing, default */
1090 } else if (!strcmp(mode
, "native")) {
1091 *flags
|= BDRV_O_NATIVE_AIO
;
1092 #ifdef CONFIG_LINUX_IO_URING
1093 } else if (!strcmp(mode
, "io_uring")) {
1094 *flags
|= BDRV_O_IO_URING
;
1104 * Set open flags for a given discard mode
1106 * Return 0 on success, -1 if the discard mode was invalid.
1108 int bdrv_parse_discard_flags(const char *mode
, int *flags
)
1110 *flags
&= ~BDRV_O_UNMAP
;
1112 if (!strcmp(mode
, "off") || !strcmp(mode
, "ignore")) {
1114 } else if (!strcmp(mode
, "on") || !strcmp(mode
, "unmap")) {
1115 *flags
|= BDRV_O_UNMAP
;
1124 * Set open flags for a given cache mode
1126 * Return 0 on success, -1 if the cache mode was invalid.
1128 int bdrv_parse_cache_mode(const char *mode
, int *flags
, bool *writethrough
)
1130 *flags
&= ~BDRV_O_CACHE_MASK
;
1132 if (!strcmp(mode
, "off") || !strcmp(mode
, "none")) {
1133 *writethrough
= false;
1134 *flags
|= BDRV_O_NOCACHE
;
1135 } else if (!strcmp(mode
, "directsync")) {
1136 *writethrough
= true;
1137 *flags
|= BDRV_O_NOCACHE
;
1138 } else if (!strcmp(mode
, "writeback")) {
1139 *writethrough
= false;
1140 } else if (!strcmp(mode
, "unsafe")) {
1141 *writethrough
= false;
1142 *flags
|= BDRV_O_NO_FLUSH
;
1143 } else if (!strcmp(mode
, "writethrough")) {
1144 *writethrough
= true;
1152 static char *bdrv_child_get_parent_desc(BdrvChild
*c
)
1154 BlockDriverState
*parent
= c
->opaque
;
1155 return g_strdup_printf("node '%s'", bdrv_get_node_name(parent
));
1158 static void bdrv_child_cb_drained_begin(BdrvChild
*child
)
1160 BlockDriverState
*bs
= child
->opaque
;
1161 bdrv_do_drained_begin_quiesce(bs
, NULL
, false);
1164 static bool bdrv_child_cb_drained_poll(BdrvChild
*child
)
1166 BlockDriverState
*bs
= child
->opaque
;
1167 return bdrv_drain_poll(bs
, false, NULL
, false);
1170 static void bdrv_child_cb_drained_end(BdrvChild
*child
,
1171 int *drained_end_counter
)
1173 BlockDriverState
*bs
= child
->opaque
;
1174 bdrv_drained_end_no_poll(bs
, drained_end_counter
);
1177 static int bdrv_child_cb_inactivate(BdrvChild
*child
)
1179 BlockDriverState
*bs
= child
->opaque
;
1180 assert(bs
->open_flags
& BDRV_O_INACTIVE
);
1184 static bool bdrv_child_cb_can_set_aio_ctx(BdrvChild
*child
, AioContext
*ctx
,
1185 GSList
**ignore
, Error
**errp
)
1187 BlockDriverState
*bs
= child
->opaque
;
1188 return bdrv_can_set_aio_context(bs
, ctx
, ignore
, errp
);
1191 static void bdrv_child_cb_set_aio_ctx(BdrvChild
*child
, AioContext
*ctx
,
1194 BlockDriverState
*bs
= child
->opaque
;
1195 return bdrv_set_aio_context_ignore(bs
, ctx
, ignore
);
1199 * Returns the options and flags that a temporary snapshot should get, based on
1200 * the originally requested flags (the originally requested image will have
1201 * flags like a backing file)
1203 static void bdrv_temp_snapshot_options(int *child_flags
, QDict
*child_options
,
1204 int parent_flags
, QDict
*parent_options
)
1206 *child_flags
= (parent_flags
& ~BDRV_O_SNAPSHOT
) | BDRV_O_TEMPORARY
;
1208 /* For temporary files, unconditional cache=unsafe is fine */
1209 qdict_set_default_str(child_options
, BDRV_OPT_CACHE_DIRECT
, "off");
1210 qdict_set_default_str(child_options
, BDRV_OPT_CACHE_NO_FLUSH
, "on");
1212 /* Copy the read-only and discard options from the parent */
1213 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_READ_ONLY
);
1214 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_DISCARD
);
1216 /* aio=native doesn't work for cache.direct=off, so disable it for the
1217 * temporary snapshot */
1218 *child_flags
&= ~BDRV_O_NATIVE_AIO
;
1221 static void bdrv_backing_attach(BdrvChild
*c
)
1223 BlockDriverState
*parent
= c
->opaque
;
1224 BlockDriverState
*backing_hd
= c
->bs
;
1226 assert(!parent
->backing_blocker
);
1227 error_setg(&parent
->backing_blocker
,
1228 "node is used as backing hd of '%s'",
1229 bdrv_get_device_or_node_name(parent
));
1231 bdrv_refresh_filename(backing_hd
);
1233 parent
->open_flags
&= ~BDRV_O_NO_BACKING
;
1235 bdrv_op_block_all(backing_hd
, parent
->backing_blocker
);
1236 /* Otherwise we won't be able to commit or stream */
1237 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_COMMIT_TARGET
,
1238 parent
->backing_blocker
);
1239 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_STREAM
,
1240 parent
->backing_blocker
);
1242 * We do backup in 3 ways:
1244 * The target bs is new opened, and the source is top BDS
1245 * 2. blockdev backup
1246 * Both the source and the target are top BDSes.
1247 * 3. internal backup(used for block replication)
1248 * Both the source and the target are backing file
1250 * In case 1 and 2, neither the source nor the target is the backing file.
1251 * In case 3, we will block the top BDS, so there is only one block job
1252 * for the top BDS and its backing chain.
1254 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_BACKUP_SOURCE
,
1255 parent
->backing_blocker
);
1256 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_BACKUP_TARGET
,
1257 parent
->backing_blocker
);
1260 static void bdrv_backing_detach(BdrvChild
*c
)
1262 BlockDriverState
*parent
= c
->opaque
;
1264 assert(parent
->backing_blocker
);
1265 bdrv_op_unblock_all(c
->bs
, parent
->backing_blocker
);
1266 error_free(parent
->backing_blocker
);
1267 parent
->backing_blocker
= NULL
;
1270 static int bdrv_backing_update_filename(BdrvChild
*c
, BlockDriverState
*base
,
1271 const char *filename
, Error
**errp
)
1273 BlockDriverState
*parent
= c
->opaque
;
1274 bool read_only
= bdrv_is_read_only(parent
);
1278 ret
= bdrv_reopen_set_read_only(parent
, false, errp
);
1284 ret
= bdrv_change_backing_file(parent
, filename
,
1285 base
->drv
? base
->drv
->format_name
: "",
1288 error_setg_errno(errp
, -ret
, "Could not update backing file link");
1292 bdrv_reopen_set_read_only(parent
, true, NULL
);
1299 * Returns the options and flags that a generic child of a BDS should
1300 * get, based on the given options and flags for the parent BDS.
1302 static void bdrv_inherited_options(BdrvChildRole role
, bool parent_is_format
,
1303 int *child_flags
, QDict
*child_options
,
1304 int parent_flags
, QDict
*parent_options
)
1306 int flags
= parent_flags
;
1309 * First, decide whether to set, clear, or leave BDRV_O_PROTOCOL.
1310 * Generally, the question to answer is: Should this child be
1311 * format-probed by default?
1315 * Pure and non-filtered data children of non-format nodes should
1316 * be probed by default (even when the node itself has BDRV_O_PROTOCOL
1317 * set). This only affects a very limited set of drivers (namely
1318 * quorum and blkverify when this comment was written).
1319 * Force-clear BDRV_O_PROTOCOL then.
1321 if (!parent_is_format
&&
1322 (role
& BDRV_CHILD_DATA
) &&
1323 !(role
& (BDRV_CHILD_METADATA
| BDRV_CHILD_FILTERED
)))
1325 flags
&= ~BDRV_O_PROTOCOL
;
1329 * All children of format nodes (except for COW children) and all
1330 * metadata children in general should never be format-probed.
1331 * Force-set BDRV_O_PROTOCOL then.
1333 if ((parent_is_format
&& !(role
& BDRV_CHILD_COW
)) ||
1334 (role
& BDRV_CHILD_METADATA
))
1336 flags
|= BDRV_O_PROTOCOL
;
1340 * If the cache mode isn't explicitly set, inherit direct and no-flush from
1343 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_CACHE_DIRECT
);
1344 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_CACHE_NO_FLUSH
);
1345 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_FORCE_SHARE
);
1347 if (role
& BDRV_CHILD_COW
) {
1348 /* backing files are opened read-only by default */
1349 qdict_set_default_str(child_options
, BDRV_OPT_READ_ONLY
, "on");
1350 qdict_set_default_str(child_options
, BDRV_OPT_AUTO_READ_ONLY
, "off");
1352 /* Inherit the read-only option from the parent if it's not set */
1353 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_READ_ONLY
);
1354 qdict_copy_default(child_options
, parent_options
,
1355 BDRV_OPT_AUTO_READ_ONLY
);
1359 * bdrv_co_pdiscard() respects unmap policy for the parent, so we
1360 * can default to enable it on lower layers regardless of the
1363 qdict_set_default_str(child_options
, BDRV_OPT_DISCARD
, "unmap");
1365 /* Clear flags that only apply to the top layer */
1366 flags
&= ~(BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
| BDRV_O_COPY_ON_READ
);
1368 if (role
& BDRV_CHILD_METADATA
) {
1369 flags
&= ~BDRV_O_NO_IO
;
1371 if (role
& BDRV_CHILD_COW
) {
1372 flags
&= ~BDRV_O_TEMPORARY
;
1375 *child_flags
= flags
;
1378 static void bdrv_child_cb_attach(BdrvChild
*child
)
1380 BlockDriverState
*bs
= child
->opaque
;
1382 if (child
->role
& BDRV_CHILD_COW
) {
1383 bdrv_backing_attach(child
);
1386 bdrv_apply_subtree_drain(child
, bs
);
1389 static void bdrv_child_cb_detach(BdrvChild
*child
)
1391 BlockDriverState
*bs
= child
->opaque
;
1393 if (child
->role
& BDRV_CHILD_COW
) {
1394 bdrv_backing_detach(child
);
1397 bdrv_unapply_subtree_drain(child
, bs
);
1400 static int bdrv_child_cb_update_filename(BdrvChild
*c
, BlockDriverState
*base
,
1401 const char *filename
, Error
**errp
)
1403 if (c
->role
& BDRV_CHILD_COW
) {
1404 return bdrv_backing_update_filename(c
, base
, filename
, errp
);
1409 AioContext
*child_of_bds_get_parent_aio_context(BdrvChild
*c
)
1411 BlockDriverState
*bs
= c
->opaque
;
1413 return bdrv_get_aio_context(bs
);
1416 const BdrvChildClass child_of_bds
= {
1417 .parent_is_bds
= true,
1418 .get_parent_desc
= bdrv_child_get_parent_desc
,
1419 .inherit_options
= bdrv_inherited_options
,
1420 .drained_begin
= bdrv_child_cb_drained_begin
,
1421 .drained_poll
= bdrv_child_cb_drained_poll
,
1422 .drained_end
= bdrv_child_cb_drained_end
,
1423 .attach
= bdrv_child_cb_attach
,
1424 .detach
= bdrv_child_cb_detach
,
1425 .inactivate
= bdrv_child_cb_inactivate
,
1426 .can_set_aio_ctx
= bdrv_child_cb_can_set_aio_ctx
,
1427 .set_aio_ctx
= bdrv_child_cb_set_aio_ctx
,
1428 .update_filename
= bdrv_child_cb_update_filename
,
1429 .get_parent_aio_context
= child_of_bds_get_parent_aio_context
,
1432 AioContext
*bdrv_child_get_parent_aio_context(BdrvChild
*c
)
1434 return c
->klass
->get_parent_aio_context(c
);
1437 static int bdrv_open_flags(BlockDriverState
*bs
, int flags
)
1439 int open_flags
= flags
;
1442 * Clear flags that are internal to the block layer before opening the
1445 open_flags
&= ~(BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
| BDRV_O_PROTOCOL
);
1450 static void update_flags_from_options(int *flags
, QemuOpts
*opts
)
1452 *flags
&= ~(BDRV_O_CACHE_MASK
| BDRV_O_RDWR
| BDRV_O_AUTO_RDONLY
);
1454 if (qemu_opt_get_bool_del(opts
, BDRV_OPT_CACHE_NO_FLUSH
, false)) {
1455 *flags
|= BDRV_O_NO_FLUSH
;
1458 if (qemu_opt_get_bool_del(opts
, BDRV_OPT_CACHE_DIRECT
, false)) {
1459 *flags
|= BDRV_O_NOCACHE
;
1462 if (!qemu_opt_get_bool_del(opts
, BDRV_OPT_READ_ONLY
, false)) {
1463 *flags
|= BDRV_O_RDWR
;
1466 if (qemu_opt_get_bool_del(opts
, BDRV_OPT_AUTO_READ_ONLY
, false)) {
1467 *flags
|= BDRV_O_AUTO_RDONLY
;
1471 static void update_options_from_flags(QDict
*options
, int flags
)
1473 if (!qdict_haskey(options
, BDRV_OPT_CACHE_DIRECT
)) {
1474 qdict_put_bool(options
, BDRV_OPT_CACHE_DIRECT
, flags
& BDRV_O_NOCACHE
);
1476 if (!qdict_haskey(options
, BDRV_OPT_CACHE_NO_FLUSH
)) {
1477 qdict_put_bool(options
, BDRV_OPT_CACHE_NO_FLUSH
,
1478 flags
& BDRV_O_NO_FLUSH
);
1480 if (!qdict_haskey(options
, BDRV_OPT_READ_ONLY
)) {
1481 qdict_put_bool(options
, BDRV_OPT_READ_ONLY
, !(flags
& BDRV_O_RDWR
));
1483 if (!qdict_haskey(options
, BDRV_OPT_AUTO_READ_ONLY
)) {
1484 qdict_put_bool(options
, BDRV_OPT_AUTO_READ_ONLY
,
1485 flags
& BDRV_O_AUTO_RDONLY
);
1489 static void bdrv_assign_node_name(BlockDriverState
*bs
,
1490 const char *node_name
,
1493 char *gen_node_name
= NULL
;
1496 node_name
= gen_node_name
= id_generate(ID_BLOCK
);
1497 } else if (!id_wellformed(node_name
)) {
1499 * Check for empty string or invalid characters, but not if it is
1500 * generated (generated names use characters not available to the user)
1502 error_setg(errp
, "Invalid node-name: '%s'", node_name
);
1506 /* takes care of avoiding namespaces collisions */
1507 if (blk_by_name(node_name
)) {
1508 error_setg(errp
, "node-name=%s is conflicting with a device id",
1513 /* takes care of avoiding duplicates node names */
1514 if (bdrv_find_node(node_name
)) {
1515 error_setg(errp
, "Duplicate nodes with node-name='%s'", node_name
);
1519 /* Make sure that the node name isn't truncated */
1520 if (strlen(node_name
) >= sizeof(bs
->node_name
)) {
1521 error_setg(errp
, "Node name too long");
1525 /* copy node name into the bs and insert it into the graph list */
1526 pstrcpy(bs
->node_name
, sizeof(bs
->node_name
), node_name
);
1527 QTAILQ_INSERT_TAIL(&graph_bdrv_states
, bs
, node_list
);
1529 g_free(gen_node_name
);
1532 static int bdrv_open_driver(BlockDriverState
*bs
, BlockDriver
*drv
,
1533 const char *node_name
, QDict
*options
,
1534 int open_flags
, Error
**errp
)
1536 Error
*local_err
= NULL
;
1539 bdrv_assign_node_name(bs
, node_name
, &local_err
);
1541 error_propagate(errp
, local_err
);
1546 bs
->opaque
= g_malloc0(drv
->instance_size
);
1548 if (drv
->bdrv_file_open
) {
1549 assert(!drv
->bdrv_needs_filename
|| bs
->filename
[0]);
1550 ret
= drv
->bdrv_file_open(bs
, options
, open_flags
, &local_err
);
1551 } else if (drv
->bdrv_open
) {
1552 ret
= drv
->bdrv_open(bs
, options
, open_flags
, &local_err
);
1559 error_propagate(errp
, local_err
);
1560 } else if (bs
->filename
[0]) {
1561 error_setg_errno(errp
, -ret
, "Could not open '%s'", bs
->filename
);
1563 error_setg_errno(errp
, -ret
, "Could not open image");
1568 ret
= refresh_total_sectors(bs
, bs
->total_sectors
);
1570 error_setg_errno(errp
, -ret
, "Could not refresh total sector count");
1574 bdrv_refresh_limits(bs
, NULL
, &local_err
);
1576 error_propagate(errp
, local_err
);
1580 assert(bdrv_opt_mem_align(bs
) != 0);
1581 assert(bdrv_min_mem_align(bs
) != 0);
1582 assert(is_power_of_2(bs
->bl
.request_alignment
));
1584 for (i
= 0; i
< bs
->quiesce_counter
; i
++) {
1585 if (drv
->bdrv_co_drain_begin
) {
1586 drv
->bdrv_co_drain_begin(bs
);
1593 if (bs
->file
!= NULL
) {
1594 bdrv_unref_child(bs
, bs
->file
);
1602 BlockDriverState
*bdrv_new_open_driver(BlockDriver
*drv
, const char *node_name
,
1603 int flags
, Error
**errp
)
1605 BlockDriverState
*bs
;
1609 bs
->open_flags
= flags
;
1610 bs
->explicit_options
= qdict_new();
1611 bs
->options
= qdict_new();
1614 update_options_from_flags(bs
->options
, flags
);
1616 ret
= bdrv_open_driver(bs
, drv
, node_name
, bs
->options
, flags
, errp
);
1618 qobject_unref(bs
->explicit_options
);
1619 bs
->explicit_options
= NULL
;
1620 qobject_unref(bs
->options
);
1629 QemuOptsList bdrv_runtime_opts
= {
1630 .name
= "bdrv_common",
1631 .head
= QTAILQ_HEAD_INITIALIZER(bdrv_runtime_opts
.head
),
1634 .name
= "node-name",
1635 .type
= QEMU_OPT_STRING
,
1636 .help
= "Node name of the block device node",
1640 .type
= QEMU_OPT_STRING
,
1641 .help
= "Block driver to use for the node",
1644 .name
= BDRV_OPT_CACHE_DIRECT
,
1645 .type
= QEMU_OPT_BOOL
,
1646 .help
= "Bypass software writeback cache on the host",
1649 .name
= BDRV_OPT_CACHE_NO_FLUSH
,
1650 .type
= QEMU_OPT_BOOL
,
1651 .help
= "Ignore flush requests",
1654 .name
= BDRV_OPT_READ_ONLY
,
1655 .type
= QEMU_OPT_BOOL
,
1656 .help
= "Node is opened in read-only mode",
1659 .name
= BDRV_OPT_AUTO_READ_ONLY
,
1660 .type
= QEMU_OPT_BOOL
,
1661 .help
= "Node can become read-only if opening read-write fails",
1664 .name
= "detect-zeroes",
1665 .type
= QEMU_OPT_STRING
,
1666 .help
= "try to optimize zero writes (off, on, unmap)",
1669 .name
= BDRV_OPT_DISCARD
,
1670 .type
= QEMU_OPT_STRING
,
1671 .help
= "discard operation (ignore/off, unmap/on)",
1674 .name
= BDRV_OPT_FORCE_SHARE
,
1675 .type
= QEMU_OPT_BOOL
,
1676 .help
= "always accept other writers (default: off)",
1678 { /* end of list */ }
1682 QemuOptsList bdrv_create_opts_simple
= {
1683 .name
= "simple-create-opts",
1684 .head
= QTAILQ_HEAD_INITIALIZER(bdrv_create_opts_simple
.head
),
1687 .name
= BLOCK_OPT_SIZE
,
1688 .type
= QEMU_OPT_SIZE
,
1689 .help
= "Virtual disk size"
1692 .name
= BLOCK_OPT_PREALLOC
,
1693 .type
= QEMU_OPT_STRING
,
1694 .help
= "Preallocation mode (allowed values: off)"
1696 { /* end of list */ }
1701 * Common part for opening disk images and files
1703 * Removes all processed options from *options.
1705 static int bdrv_open_common(BlockDriverState
*bs
, BlockBackend
*file
,
1706 QDict
*options
, Error
**errp
)
1708 int ret
, open_flags
;
1709 const char *filename
;
1710 const char *driver_name
= NULL
;
1711 const char *node_name
= NULL
;
1712 const char *discard
;
1715 Error
*local_err
= NULL
;
1718 assert(bs
->file
== NULL
);
1719 assert(options
!= NULL
&& bs
->options
!= options
);
1721 opts
= qemu_opts_create(&bdrv_runtime_opts
, NULL
, 0, &error_abort
);
1722 if (!qemu_opts_absorb_qdict(opts
, options
, errp
)) {
1727 update_flags_from_options(&bs
->open_flags
, opts
);
1729 driver_name
= qemu_opt_get(opts
, "driver");
1730 drv
= bdrv_find_format(driver_name
);
1731 assert(drv
!= NULL
);
1733 bs
->force_share
= qemu_opt_get_bool(opts
, BDRV_OPT_FORCE_SHARE
, false);
1735 if (bs
->force_share
&& (bs
->open_flags
& BDRV_O_RDWR
)) {
1737 BDRV_OPT_FORCE_SHARE
1738 "=on can only be used with read-only images");
1744 bdrv_refresh_filename(blk_bs(file
));
1745 filename
= blk_bs(file
)->filename
;
1748 * Caution: while qdict_get_try_str() is fine, getting
1749 * non-string types would require more care. When @options
1750 * come from -blockdev or blockdev_add, its members are typed
1751 * according to the QAPI schema, but when they come from
1752 * -drive, they're all QString.
1754 filename
= qdict_get_try_str(options
, "filename");
1757 if (drv
->bdrv_needs_filename
&& (!filename
|| !filename
[0])) {
1758 error_setg(errp
, "The '%s' block driver requires a file name",
1764 trace_bdrv_open_common(bs
, filename
?: "", bs
->open_flags
,
1767 ro
= bdrv_is_read_only(bs
);
1769 if (use_bdrv_whitelist
&& !bdrv_is_whitelisted(drv
, ro
)) {
1770 if (!ro
&& bdrv_is_whitelisted(drv
, true)) {
1771 ret
= bdrv_apply_auto_read_only(bs
, NULL
, NULL
);
1777 !ro
&& bdrv_is_whitelisted(drv
, true)
1778 ? "Driver '%s' can only be used for read-only devices"
1779 : "Driver '%s' is not whitelisted",
1785 /* bdrv_new() and bdrv_close() make it so */
1786 assert(qatomic_read(&bs
->copy_on_read
) == 0);
1788 if (bs
->open_flags
& BDRV_O_COPY_ON_READ
) {
1790 bdrv_enable_copy_on_read(bs
);
1792 error_setg(errp
, "Can't use copy-on-read on read-only device");
1798 discard
= qemu_opt_get(opts
, BDRV_OPT_DISCARD
);
1799 if (discard
!= NULL
) {
1800 if (bdrv_parse_discard_flags(discard
, &bs
->open_flags
) != 0) {
1801 error_setg(errp
, "Invalid discard option");
1808 bdrv_parse_detect_zeroes(opts
, bs
->open_flags
, &local_err
);
1810 error_propagate(errp
, local_err
);
1815 if (filename
!= NULL
) {
1816 pstrcpy(bs
->filename
, sizeof(bs
->filename
), filename
);
1818 bs
->filename
[0] = '\0';
1820 pstrcpy(bs
->exact_filename
, sizeof(bs
->exact_filename
), bs
->filename
);
1822 /* Open the image, either directly or using a protocol */
1823 open_flags
= bdrv_open_flags(bs
, bs
->open_flags
);
1824 node_name
= qemu_opt_get(opts
, "node-name");
1826 assert(!drv
->bdrv_file_open
|| file
== NULL
);
1827 ret
= bdrv_open_driver(bs
, drv
, node_name
, options
, open_flags
, errp
);
1832 qemu_opts_del(opts
);
1836 qemu_opts_del(opts
);
1840 static QDict
*parse_json_filename(const char *filename
, Error
**errp
)
1842 QObject
*options_obj
;
1846 ret
= strstart(filename
, "json:", &filename
);
1849 options_obj
= qobject_from_json(filename
, errp
);
1851 error_prepend(errp
, "Could not parse the JSON options: ");
1855 options
= qobject_to(QDict
, options_obj
);
1857 qobject_unref(options_obj
);
1858 error_setg(errp
, "Invalid JSON object given");
1862 qdict_flatten(options
);
1867 static void parse_json_protocol(QDict
*options
, const char **pfilename
,
1870 QDict
*json_options
;
1871 Error
*local_err
= NULL
;
1873 /* Parse json: pseudo-protocol */
1874 if (!*pfilename
|| !g_str_has_prefix(*pfilename
, "json:")) {
1878 json_options
= parse_json_filename(*pfilename
, &local_err
);
1880 error_propagate(errp
, local_err
);
1884 /* Options given in the filename have lower priority than options
1885 * specified directly */
1886 qdict_join(options
, json_options
, false);
1887 qobject_unref(json_options
);
1892 * Fills in default options for opening images and converts the legacy
1893 * filename/flags pair to option QDict entries.
1894 * The BDRV_O_PROTOCOL flag in *flags will be set or cleared accordingly if a
1895 * block driver has been specified explicitly.
1897 static int bdrv_fill_options(QDict
**options
, const char *filename
,
1898 int *flags
, Error
**errp
)
1900 const char *drvname
;
1901 bool protocol
= *flags
& BDRV_O_PROTOCOL
;
1902 bool parse_filename
= false;
1903 BlockDriver
*drv
= NULL
;
1904 Error
*local_err
= NULL
;
1907 * Caution: while qdict_get_try_str() is fine, getting non-string
1908 * types would require more care. When @options come from
1909 * -blockdev or blockdev_add, its members are typed according to
1910 * the QAPI schema, but when they come from -drive, they're all
1913 drvname
= qdict_get_try_str(*options
, "driver");
1915 drv
= bdrv_find_format(drvname
);
1917 error_setg(errp
, "Unknown driver '%s'", drvname
);
1920 /* If the user has explicitly specified the driver, this choice should
1921 * override the BDRV_O_PROTOCOL flag */
1922 protocol
= drv
->bdrv_file_open
;
1926 *flags
|= BDRV_O_PROTOCOL
;
1928 *flags
&= ~BDRV_O_PROTOCOL
;
1931 /* Translate cache options from flags into options */
1932 update_options_from_flags(*options
, *flags
);
1934 /* Fetch the file name from the options QDict if necessary */
1935 if (protocol
&& filename
) {
1936 if (!qdict_haskey(*options
, "filename")) {
1937 qdict_put_str(*options
, "filename", filename
);
1938 parse_filename
= true;
1940 error_setg(errp
, "Can't specify 'file' and 'filename' options at "
1946 /* Find the right block driver */
1947 /* See cautionary note on accessing @options above */
1948 filename
= qdict_get_try_str(*options
, "filename");
1950 if (!drvname
&& protocol
) {
1952 drv
= bdrv_find_protocol(filename
, parse_filename
, errp
);
1957 drvname
= drv
->format_name
;
1958 qdict_put_str(*options
, "driver", drvname
);
1960 error_setg(errp
, "Must specify either driver or file");
1965 assert(drv
|| !protocol
);
1967 /* Driver-specific filename parsing */
1968 if (drv
&& drv
->bdrv_parse_filename
&& parse_filename
) {
1969 drv
->bdrv_parse_filename(filename
, *options
, &local_err
);
1971 error_propagate(errp
, local_err
);
1975 if (!drv
->bdrv_needs_filename
) {
1976 qdict_del(*options
, "filename");
1983 typedef struct BlockReopenQueueEntry
{
1986 BDRVReopenState state
;
1987 QTAILQ_ENTRY(BlockReopenQueueEntry
) entry
;
1988 } BlockReopenQueueEntry
;
1991 * Return the flags that @bs will have after the reopens in @q have
1992 * successfully completed. If @q is NULL (or @bs is not contained in @q),
1993 * return the current flags.
1995 static int bdrv_reopen_get_flags(BlockReopenQueue
*q
, BlockDriverState
*bs
)
1997 BlockReopenQueueEntry
*entry
;
2000 QTAILQ_FOREACH(entry
, q
, entry
) {
2001 if (entry
->state
.bs
== bs
) {
2002 return entry
->state
.flags
;
2007 return bs
->open_flags
;
2010 /* Returns whether the image file can be written to after the reopen queue @q
2011 * has been successfully applied, or right now if @q is NULL. */
2012 static bool bdrv_is_writable_after_reopen(BlockDriverState
*bs
,
2013 BlockReopenQueue
*q
)
2015 int flags
= bdrv_reopen_get_flags(q
, bs
);
2017 return (flags
& (BDRV_O_RDWR
| BDRV_O_INACTIVE
)) == BDRV_O_RDWR
;
2021 * Return whether the BDS can be written to. This is not necessarily
2022 * the same as !bdrv_is_read_only(bs), as inactivated images may not
2023 * be written to but do not count as read-only images.
2025 bool bdrv_is_writable(BlockDriverState
*bs
)
2027 return bdrv_is_writable_after_reopen(bs
, NULL
);
2030 static char *bdrv_child_user_desc(BdrvChild
*c
)
2032 return c
->klass
->get_parent_desc(c
);
2036 * Check that @a allows everything that @b needs. @a and @b must reference same
2039 static bool bdrv_a_allow_b(BdrvChild
*a
, BdrvChild
*b
, Error
**errp
)
2041 const char *child_bs_name
;
2042 g_autofree
char *a_user
= NULL
;
2043 g_autofree
char *b_user
= NULL
;
2044 g_autofree
char *perms
= NULL
;
2047 assert(a
->bs
== b
->bs
);
2049 if ((b
->perm
& a
->shared_perm
) == b
->perm
) {
2053 child_bs_name
= bdrv_get_node_name(b
->bs
);
2054 a_user
= bdrv_child_user_desc(a
);
2055 b_user
= bdrv_child_user_desc(b
);
2056 perms
= bdrv_perm_names(b
->perm
& ~a
->shared_perm
);
2058 error_setg(errp
, "Permission conflict on node '%s': permissions '%s' are "
2059 "both required by %s (uses node '%s' as '%s' child) and "
2060 "unshared by %s (uses node '%s' as '%s' child).",
2061 child_bs_name
, perms
,
2062 b_user
, child_bs_name
, b
->name
,
2063 a_user
, child_bs_name
, a
->name
);
2068 static bool bdrv_parent_perms_conflict(BlockDriverState
*bs
, Error
**errp
)
2073 * During the loop we'll look at each pair twice. That's correct because
2074 * bdrv_a_allow_b() is asymmetric and we should check each pair in both
2077 QLIST_FOREACH(a
, &bs
->parents
, next_parent
) {
2078 QLIST_FOREACH(b
, &bs
->parents
, next_parent
) {
2083 if (!bdrv_a_allow_b(a
, b
, errp
)) {
2092 static void bdrv_child_perm(BlockDriverState
*bs
, BlockDriverState
*child_bs
,
2093 BdrvChild
*c
, BdrvChildRole role
,
2094 BlockReopenQueue
*reopen_queue
,
2095 uint64_t parent_perm
, uint64_t parent_shared
,
2096 uint64_t *nperm
, uint64_t *nshared
)
2098 assert(bs
->drv
&& bs
->drv
->bdrv_child_perm
);
2099 bs
->drv
->bdrv_child_perm(bs
, c
, role
, reopen_queue
,
2100 parent_perm
, parent_shared
,
2102 /* TODO Take force_share from reopen_queue */
2103 if (child_bs
&& child_bs
->force_share
) {
2104 *nshared
= BLK_PERM_ALL
;
2109 * Adds the whole subtree of @bs (including @bs itself) to the @list (except for
2110 * nodes that are already in the @list, of course) so that final list is
2111 * topologically sorted. Return the result (GSList @list object is updated, so
2112 * don't use old reference after function call).
2114 * On function start @list must be already topologically sorted and for any node
2115 * in the @list the whole subtree of the node must be in the @list as well. The
2116 * simplest way to satisfy this criteria: use only result of
2117 * bdrv_topological_dfs() or NULL as @list parameter.
2119 static GSList
*bdrv_topological_dfs(GSList
*list
, GHashTable
*found
,
2120 BlockDriverState
*bs
)
2123 g_autoptr(GHashTable
) local_found
= NULL
;
2127 found
= local_found
= g_hash_table_new(NULL
, NULL
);
2130 if (g_hash_table_contains(found
, bs
)) {
2133 g_hash_table_add(found
, bs
);
2135 QLIST_FOREACH(child
, &bs
->children
, next
) {
2136 list
= bdrv_topological_dfs(list
, found
, child
->bs
);
2139 return g_slist_prepend(list
, bs
);
2142 typedef struct BdrvChildSetPermState
{
2145 uint64_t old_shared_perm
;
2146 } BdrvChildSetPermState
;
2148 static void bdrv_child_set_perm_abort(void *opaque
)
2150 BdrvChildSetPermState
*s
= opaque
;
2152 s
->child
->perm
= s
->old_perm
;
2153 s
->child
->shared_perm
= s
->old_shared_perm
;
2156 static TransactionActionDrv bdrv_child_set_pem_drv
= {
2157 .abort
= bdrv_child_set_perm_abort
,
2161 static void bdrv_child_set_perm(BdrvChild
*c
, uint64_t perm
,
2162 uint64_t shared
, Transaction
*tran
)
2164 BdrvChildSetPermState
*s
= g_new(BdrvChildSetPermState
, 1);
2166 *s
= (BdrvChildSetPermState
) {
2168 .old_perm
= c
->perm
,
2169 .old_shared_perm
= c
->shared_perm
,
2173 c
->shared_perm
= shared
;
2175 tran_add(tran
, &bdrv_child_set_pem_drv
, s
);
2178 static void bdrv_drv_set_perm_commit(void *opaque
)
2180 BlockDriverState
*bs
= opaque
;
2181 uint64_t cumulative_perms
, cumulative_shared_perms
;
2183 if (bs
->drv
->bdrv_set_perm
) {
2184 bdrv_get_cumulative_perm(bs
, &cumulative_perms
,
2185 &cumulative_shared_perms
);
2186 bs
->drv
->bdrv_set_perm(bs
, cumulative_perms
, cumulative_shared_perms
);
2190 static void bdrv_drv_set_perm_abort(void *opaque
)
2192 BlockDriverState
*bs
= opaque
;
2194 if (bs
->drv
->bdrv_abort_perm_update
) {
2195 bs
->drv
->bdrv_abort_perm_update(bs
);
2199 TransactionActionDrv bdrv_drv_set_perm_drv
= {
2200 .abort
= bdrv_drv_set_perm_abort
,
2201 .commit
= bdrv_drv_set_perm_commit
,
2204 static int bdrv_drv_set_perm(BlockDriverState
*bs
, uint64_t perm
,
2205 uint64_t shared_perm
, Transaction
*tran
,
2212 if (bs
->drv
->bdrv_check_perm
) {
2213 int ret
= bs
->drv
->bdrv_check_perm(bs
, perm
, shared_perm
, errp
);
2220 tran_add(tran
, &bdrv_drv_set_perm_drv
, bs
);
2226 typedef struct BdrvReplaceChildState
{
2228 BlockDriverState
*old_bs
;
2229 } BdrvReplaceChildState
;
2231 static void bdrv_replace_child_commit(void *opaque
)
2233 BdrvReplaceChildState
*s
= opaque
;
2235 bdrv_unref(s
->old_bs
);
2238 static void bdrv_replace_child_abort(void *opaque
)
2240 BdrvReplaceChildState
*s
= opaque
;
2241 BlockDriverState
*new_bs
= s
->child
->bs
;
2243 /* old_bs reference is transparently moved from @s to @s->child */
2244 bdrv_replace_child_noperm(s
->child
, s
->old_bs
);
2248 static TransactionActionDrv bdrv_replace_child_drv
= {
2249 .commit
= bdrv_replace_child_commit
,
2250 .abort
= bdrv_replace_child_abort
,
2255 * bdrv_replace_child_tran
2257 * Note: real unref of old_bs is done only on commit.
2259 * The function doesn't update permissions, caller is responsible for this.
2261 static void bdrv_replace_child_tran(BdrvChild
*child
, BlockDriverState
*new_bs
,
2264 BdrvReplaceChildState
*s
= g_new(BdrvReplaceChildState
, 1);
2265 *s
= (BdrvReplaceChildState
) {
2267 .old_bs
= child
->bs
,
2269 tran_add(tran
, &bdrv_replace_child_drv
, s
);
2274 bdrv_replace_child_noperm(child
, new_bs
);
2275 /* old_bs reference is transparently moved from @child to @s */
2279 * Refresh permissions in @bs subtree. The function is intended to be called
2280 * after some graph modification that was done without permission update.
2282 static int bdrv_node_refresh_perm(BlockDriverState
*bs
, BlockReopenQueue
*q
,
2283 Transaction
*tran
, Error
**errp
)
2285 BlockDriver
*drv
= bs
->drv
;
2288 uint64_t cumulative_perms
, cumulative_shared_perms
;
2290 bdrv_get_cumulative_perm(bs
, &cumulative_perms
, &cumulative_shared_perms
);
2292 /* Write permissions never work with read-only images */
2293 if ((cumulative_perms
& (BLK_PERM_WRITE
| BLK_PERM_WRITE_UNCHANGED
)) &&
2294 !bdrv_is_writable_after_reopen(bs
, q
))
2296 if (!bdrv_is_writable_after_reopen(bs
, NULL
)) {
2297 error_setg(errp
, "Block node is read-only");
2299 error_setg(errp
, "Read-only block node '%s' cannot support "
2300 "read-write users", bdrv_get_node_name(bs
));
2307 * Unaligned requests will automatically be aligned to bl.request_alignment
2308 * and without RESIZE we can't extend requests to write to space beyond the
2309 * end of the image, so it's required that the image size is aligned.
2311 if ((cumulative_perms
& (BLK_PERM_WRITE
| BLK_PERM_WRITE_UNCHANGED
)) &&
2312 !(cumulative_perms
& BLK_PERM_RESIZE
))
2314 if ((bs
->total_sectors
* BDRV_SECTOR_SIZE
) % bs
->bl
.request_alignment
) {
2315 error_setg(errp
, "Cannot get 'write' permission without 'resize': "
2316 "Image size is not a multiple of request "
2322 /* Check this node */
2327 ret
= bdrv_drv_set_perm(bs
, cumulative_perms
, cumulative_shared_perms
, tran
,
2333 /* Drivers that never have children can omit .bdrv_child_perm() */
2334 if (!drv
->bdrv_child_perm
) {
2335 assert(QLIST_EMPTY(&bs
->children
));
2339 /* Check all children */
2340 QLIST_FOREACH(c
, &bs
->children
, next
) {
2341 uint64_t cur_perm
, cur_shared
;
2343 bdrv_child_perm(bs
, c
->bs
, c
, c
->role
, q
,
2344 cumulative_perms
, cumulative_shared_perms
,
2345 &cur_perm
, &cur_shared
);
2346 bdrv_child_set_perm(c
, cur_perm
, cur_shared
, tran
);
2352 static int bdrv_list_refresh_perms(GSList
*list
, BlockReopenQueue
*q
,
2353 Transaction
*tran
, Error
**errp
)
2356 BlockDriverState
*bs
;
2358 for ( ; list
; list
= list
->next
) {
2361 if (bdrv_parent_perms_conflict(bs
, errp
)) {
2365 ret
= bdrv_node_refresh_perm(bs
, q
, tran
, errp
);
2374 void bdrv_get_cumulative_perm(BlockDriverState
*bs
, uint64_t *perm
,
2375 uint64_t *shared_perm
)
2378 uint64_t cumulative_perms
= 0;
2379 uint64_t cumulative_shared_perms
= BLK_PERM_ALL
;
2381 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
2382 cumulative_perms
|= c
->perm
;
2383 cumulative_shared_perms
&= c
->shared_perm
;
2386 *perm
= cumulative_perms
;
2387 *shared_perm
= cumulative_shared_perms
;
2390 char *bdrv_perm_names(uint64_t perm
)
2396 { BLK_PERM_CONSISTENT_READ
, "consistent read" },
2397 { BLK_PERM_WRITE
, "write" },
2398 { BLK_PERM_WRITE_UNCHANGED
, "write unchanged" },
2399 { BLK_PERM_RESIZE
, "resize" },
2400 { BLK_PERM_GRAPH_MOD
, "change children" },
2404 GString
*result
= g_string_sized_new(30);
2405 struct perm_name
*p
;
2407 for (p
= permissions
; p
->name
; p
++) {
2408 if (perm
& p
->perm
) {
2409 if (result
->len
> 0) {
2410 g_string_append(result
, ", ");
2412 g_string_append(result
, p
->name
);
2416 return g_string_free(result
, FALSE
);
2420 static int bdrv_refresh_perms(BlockDriverState
*bs
, Error
**errp
)
2423 Transaction
*tran
= tran_new();
2424 g_autoptr(GSList
) list
= bdrv_topological_dfs(NULL
, NULL
, bs
);
2426 ret
= bdrv_list_refresh_perms(list
, NULL
, tran
, errp
);
2427 tran_finalize(tran
, ret
);
2432 int bdrv_child_try_set_perm(BdrvChild
*c
, uint64_t perm
, uint64_t shared
,
2435 Error
*local_err
= NULL
;
2436 Transaction
*tran
= tran_new();
2439 bdrv_child_set_perm(c
, perm
, shared
, tran
);
2441 ret
= bdrv_refresh_perms(c
->bs
, &local_err
);
2443 tran_finalize(tran
, ret
);
2446 if ((perm
& ~c
->perm
) || (c
->shared_perm
& ~shared
)) {
2447 /* tighten permissions */
2448 error_propagate(errp
, local_err
);
2451 * Our caller may intend to only loosen restrictions and
2452 * does not expect this function to fail. Errors are not
2453 * fatal in such a case, so we can just hide them from our
2456 error_free(local_err
);
2464 int bdrv_child_refresh_perms(BlockDriverState
*bs
, BdrvChild
*c
, Error
**errp
)
2466 uint64_t parent_perms
, parent_shared
;
2467 uint64_t perms
, shared
;
2469 bdrv_get_cumulative_perm(bs
, &parent_perms
, &parent_shared
);
2470 bdrv_child_perm(bs
, c
->bs
, c
, c
->role
, NULL
,
2471 parent_perms
, parent_shared
, &perms
, &shared
);
2473 return bdrv_child_try_set_perm(c
, perms
, shared
, errp
);
2477 * Default implementation for .bdrv_child_perm() for block filters:
2478 * Forward CONSISTENT_READ, WRITE, WRITE_UNCHANGED, and RESIZE to the
2481 static void bdrv_filter_default_perms(BlockDriverState
*bs
, BdrvChild
*c
,
2483 BlockReopenQueue
*reopen_queue
,
2484 uint64_t perm
, uint64_t shared
,
2485 uint64_t *nperm
, uint64_t *nshared
)
2487 *nperm
= perm
& DEFAULT_PERM_PASSTHROUGH
;
2488 *nshared
= (shared
& DEFAULT_PERM_PASSTHROUGH
) | DEFAULT_PERM_UNCHANGED
;
2491 static void bdrv_default_perms_for_cow(BlockDriverState
*bs
, BdrvChild
*c
,
2493 BlockReopenQueue
*reopen_queue
,
2494 uint64_t perm
, uint64_t shared
,
2495 uint64_t *nperm
, uint64_t *nshared
)
2497 assert(role
& BDRV_CHILD_COW
);
2500 * We want consistent read from backing files if the parent needs it.
2501 * No other operations are performed on backing files.
2503 perm
&= BLK_PERM_CONSISTENT_READ
;
2506 * If the parent can deal with changing data, we're okay with a
2507 * writable and resizable backing file.
2508 * TODO Require !(perm & BLK_PERM_CONSISTENT_READ), too?
2510 if (shared
& BLK_PERM_WRITE
) {
2511 shared
= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2516 shared
|= BLK_PERM_CONSISTENT_READ
| BLK_PERM_GRAPH_MOD
|
2517 BLK_PERM_WRITE_UNCHANGED
;
2519 if (bs
->open_flags
& BDRV_O_INACTIVE
) {
2520 shared
|= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2527 static void bdrv_default_perms_for_storage(BlockDriverState
*bs
, BdrvChild
*c
,
2529 BlockReopenQueue
*reopen_queue
,
2530 uint64_t perm
, uint64_t shared
,
2531 uint64_t *nperm
, uint64_t *nshared
)
2535 assert(role
& (BDRV_CHILD_METADATA
| BDRV_CHILD_DATA
));
2537 flags
= bdrv_reopen_get_flags(reopen_queue
, bs
);
2540 * Apart from the modifications below, the same permissions are
2541 * forwarded and left alone as for filters
2543 bdrv_filter_default_perms(bs
, c
, role
, reopen_queue
,
2544 perm
, shared
, &perm
, &shared
);
2546 if (role
& BDRV_CHILD_METADATA
) {
2547 /* Format drivers may touch metadata even if the guest doesn't write */
2548 if (bdrv_is_writable_after_reopen(bs
, reopen_queue
)) {
2549 perm
|= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2553 * bs->file always needs to be consistent because of the
2554 * metadata. We can never allow other users to resize or write
2557 if (!(flags
& BDRV_O_NO_IO
)) {
2558 perm
|= BLK_PERM_CONSISTENT_READ
;
2560 shared
&= ~(BLK_PERM_WRITE
| BLK_PERM_RESIZE
);
2563 if (role
& BDRV_CHILD_DATA
) {
2565 * Technically, everything in this block is a subset of the
2566 * BDRV_CHILD_METADATA path taken above, and so this could
2567 * be an "else if" branch. However, that is not obvious, and
2568 * this function is not performance critical, therefore we let
2569 * this be an independent "if".
2573 * We cannot allow other users to resize the file because the
2574 * format driver might have some assumptions about the size
2575 * (e.g. because it is stored in metadata, or because the file
2576 * is split into fixed-size data files).
2578 shared
&= ~BLK_PERM_RESIZE
;
2581 * WRITE_UNCHANGED often cannot be performed as such on the
2582 * data file. For example, the qcow2 driver may still need to
2583 * write copied clusters on copy-on-read.
2585 if (perm
& BLK_PERM_WRITE_UNCHANGED
) {
2586 perm
|= BLK_PERM_WRITE
;
2590 * If the data file is written to, the format driver may
2591 * expect to be able to resize it by writing beyond the EOF.
2593 if (perm
& BLK_PERM_WRITE
) {
2594 perm
|= BLK_PERM_RESIZE
;
2598 if (bs
->open_flags
& BDRV_O_INACTIVE
) {
2599 shared
|= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2606 void bdrv_default_perms(BlockDriverState
*bs
, BdrvChild
*c
,
2607 BdrvChildRole role
, BlockReopenQueue
*reopen_queue
,
2608 uint64_t perm
, uint64_t shared
,
2609 uint64_t *nperm
, uint64_t *nshared
)
2611 if (role
& BDRV_CHILD_FILTERED
) {
2612 assert(!(role
& (BDRV_CHILD_DATA
| BDRV_CHILD_METADATA
|
2614 bdrv_filter_default_perms(bs
, c
, role
, reopen_queue
,
2615 perm
, shared
, nperm
, nshared
);
2616 } else if (role
& BDRV_CHILD_COW
) {
2617 assert(!(role
& (BDRV_CHILD_DATA
| BDRV_CHILD_METADATA
)));
2618 bdrv_default_perms_for_cow(bs
, c
, role
, reopen_queue
,
2619 perm
, shared
, nperm
, nshared
);
2620 } else if (role
& (BDRV_CHILD_METADATA
| BDRV_CHILD_DATA
)) {
2621 bdrv_default_perms_for_storage(bs
, c
, role
, reopen_queue
,
2622 perm
, shared
, nperm
, nshared
);
2624 g_assert_not_reached();
2628 uint64_t bdrv_qapi_perm_to_blk_perm(BlockPermission qapi_perm
)
2630 static const uint64_t permissions
[] = {
2631 [BLOCK_PERMISSION_CONSISTENT_READ
] = BLK_PERM_CONSISTENT_READ
,
2632 [BLOCK_PERMISSION_WRITE
] = BLK_PERM_WRITE
,
2633 [BLOCK_PERMISSION_WRITE_UNCHANGED
] = BLK_PERM_WRITE_UNCHANGED
,
2634 [BLOCK_PERMISSION_RESIZE
] = BLK_PERM_RESIZE
,
2635 [BLOCK_PERMISSION_GRAPH_MOD
] = BLK_PERM_GRAPH_MOD
,
2638 QEMU_BUILD_BUG_ON(ARRAY_SIZE(permissions
) != BLOCK_PERMISSION__MAX
);
2639 QEMU_BUILD_BUG_ON(1UL << ARRAY_SIZE(permissions
) != BLK_PERM_ALL
+ 1);
2641 assert(qapi_perm
< BLOCK_PERMISSION__MAX
);
2643 return permissions
[qapi_perm
];
2646 static void bdrv_replace_child_noperm(BdrvChild
*child
,
2647 BlockDriverState
*new_bs
)
2649 BlockDriverState
*old_bs
= child
->bs
;
2650 int new_bs_quiesce_counter
;
2653 assert(!child
->frozen
);
2655 if (old_bs
&& new_bs
) {
2656 assert(bdrv_get_aio_context(old_bs
) == bdrv_get_aio_context(new_bs
));
2659 new_bs_quiesce_counter
= (new_bs
? new_bs
->quiesce_counter
: 0);
2660 drain_saldo
= new_bs_quiesce_counter
- child
->parent_quiesce_counter
;
2663 * If the new child node is drained but the old one was not, flush
2664 * all outstanding requests to the old child node.
2666 while (drain_saldo
> 0 && child
->klass
->drained_begin
) {
2667 bdrv_parent_drained_begin_single(child
, true);
2672 /* Detach first so that the recursive drain sections coming from @child
2673 * are already gone and we only end the drain sections that came from
2675 if (child
->klass
->detach
) {
2676 child
->klass
->detach(child
);
2678 QLIST_REMOVE(child
, next_parent
);
2684 QLIST_INSERT_HEAD(&new_bs
->parents
, child
, next_parent
);
2687 * Detaching the old node may have led to the new node's
2688 * quiesce_counter having been decreased. Not a problem, we
2689 * just need to recognize this here and then invoke
2690 * drained_end appropriately more often.
2692 assert(new_bs
->quiesce_counter
<= new_bs_quiesce_counter
);
2693 drain_saldo
+= new_bs
->quiesce_counter
- new_bs_quiesce_counter
;
2695 /* Attach only after starting new drained sections, so that recursive
2696 * drain sections coming from @child don't get an extra .drained_begin
2698 if (child
->klass
->attach
) {
2699 child
->klass
->attach(child
);
2704 * If the old child node was drained but the new one is not, allow
2705 * requests to come in only after the new node has been attached.
2707 while (drain_saldo
< 0 && child
->klass
->drained_end
) {
2708 bdrv_parent_drained_end_single(child
);
2713 static void bdrv_child_free(void *opaque
)
2715 BdrvChild
*c
= opaque
;
2721 static void bdrv_remove_empty_child(BdrvChild
*child
)
2724 QLIST_SAFE_REMOVE(child
, next
);
2725 bdrv_child_free(child
);
2728 typedef struct BdrvAttachChildCommonState
{
2730 AioContext
*old_parent_ctx
;
2731 AioContext
*old_child_ctx
;
2732 } BdrvAttachChildCommonState
;
2734 static void bdrv_attach_child_common_abort(void *opaque
)
2736 BdrvAttachChildCommonState
*s
= opaque
;
2737 BdrvChild
*child
= *s
->child
;
2738 BlockDriverState
*bs
= child
->bs
;
2740 bdrv_replace_child_noperm(child
, NULL
);
2742 if (bdrv_get_aio_context(bs
) != s
->old_child_ctx
) {
2743 bdrv_try_set_aio_context(bs
, s
->old_child_ctx
, &error_abort
);
2746 if (bdrv_child_get_parent_aio_context(child
) != s
->old_parent_ctx
) {
2747 GSList
*ignore
= g_slist_prepend(NULL
, child
);
2749 child
->klass
->can_set_aio_ctx(child
, s
->old_parent_ctx
, &ignore
,
2751 g_slist_free(ignore
);
2752 ignore
= g_slist_prepend(NULL
, child
);
2753 child
->klass
->set_aio_ctx(child
, s
->old_parent_ctx
, &ignore
);
2755 g_slist_free(ignore
);
2759 bdrv_remove_empty_child(child
);
2763 static TransactionActionDrv bdrv_attach_child_common_drv
= {
2764 .abort
= bdrv_attach_child_common_abort
,
2769 * Common part of attaching bdrv child to bs or to blk or to job
2771 * Resulting new child is returned through @child.
2772 * At start *@child must be NULL.
2773 * @child is saved to a new entry of @tran, so that *@child could be reverted to
2774 * NULL on abort(). So referenced variable must live at least until transaction
2777 * Function doesn't update permissions, caller is responsible for this.
2779 static int bdrv_attach_child_common(BlockDriverState
*child_bs
,
2780 const char *child_name
,
2781 const BdrvChildClass
*child_class
,
2782 BdrvChildRole child_role
,
2783 uint64_t perm
, uint64_t shared_perm
,
2784 void *opaque
, BdrvChild
**child
,
2785 Transaction
*tran
, Error
**errp
)
2787 BdrvChild
*new_child
;
2788 AioContext
*parent_ctx
;
2789 AioContext
*child_ctx
= bdrv_get_aio_context(child_bs
);
2792 assert(*child
== NULL
);
2793 assert(child_class
->get_parent_desc
);
2795 new_child
= g_new(BdrvChild
, 1);
2796 *new_child
= (BdrvChild
) {
2798 .name
= g_strdup(child_name
),
2799 .klass
= child_class
,
2802 .shared_perm
= shared_perm
,
2807 * If the AioContexts don't match, first try to move the subtree of
2808 * child_bs into the AioContext of the new parent. If this doesn't work,
2809 * try moving the parent into the AioContext of child_bs instead.
2811 parent_ctx
= bdrv_child_get_parent_aio_context(new_child
);
2812 if (child_ctx
!= parent_ctx
) {
2813 Error
*local_err
= NULL
;
2814 int ret
= bdrv_try_set_aio_context(child_bs
, parent_ctx
, &local_err
);
2816 if (ret
< 0 && child_class
->can_set_aio_ctx
) {
2817 GSList
*ignore
= g_slist_prepend(NULL
, new_child
);
2818 if (child_class
->can_set_aio_ctx(new_child
, child_ctx
, &ignore
,
2821 error_free(local_err
);
2823 g_slist_free(ignore
);
2824 ignore
= g_slist_prepend(NULL
, new_child
);
2825 child_class
->set_aio_ctx(new_child
, child_ctx
, &ignore
);
2827 g_slist_free(ignore
);
2831 error_propagate(errp
, local_err
);
2832 bdrv_remove_empty_child(new_child
);
2838 bdrv_replace_child_noperm(new_child
, child_bs
);
2842 BdrvAttachChildCommonState
*s
= g_new(BdrvAttachChildCommonState
, 1);
2843 *s
= (BdrvAttachChildCommonState
) {
2845 .old_parent_ctx
= parent_ctx
,
2846 .old_child_ctx
= child_ctx
,
2848 tran_add(tran
, &bdrv_attach_child_common_drv
, s
);
2854 * Variable referenced by @child must live at least until transaction end.
2855 * (see bdrv_attach_child_common() doc for details)
2857 * Function doesn't update permissions, caller is responsible for this.
2859 static int bdrv_attach_child_noperm(BlockDriverState
*parent_bs
,
2860 BlockDriverState
*child_bs
,
2861 const char *child_name
,
2862 const BdrvChildClass
*child_class
,
2863 BdrvChildRole child_role
,
2869 uint64_t perm
, shared_perm
;
2871 assert(parent_bs
->drv
);
2873 bdrv_get_cumulative_perm(parent_bs
, &perm
, &shared_perm
);
2874 bdrv_child_perm(parent_bs
, child_bs
, NULL
, child_role
, NULL
,
2875 perm
, shared_perm
, &perm
, &shared_perm
);
2877 ret
= bdrv_attach_child_common(child_bs
, child_name
, child_class
,
2878 child_role
, perm
, shared_perm
, parent_bs
,
2884 QLIST_INSERT_HEAD(&parent_bs
->children
, *child
, next
);
2886 * child is removed in bdrv_attach_child_common_abort(), so don't care to
2887 * abort this change separately.
2893 static void bdrv_detach_child(BdrvChild
*child
)
2895 BlockDriverState
*old_bs
= child
->bs
;
2897 bdrv_replace_child_noperm(child
, NULL
);
2898 bdrv_remove_empty_child(child
);
2902 * Update permissions for old node. We're just taking a parent away, so
2903 * we're loosening restrictions. Errors of permission update are not
2904 * fatal in this case, ignore them.
2906 bdrv_refresh_perms(old_bs
, NULL
);
2909 * When the parent requiring a non-default AioContext is removed, the
2910 * node moves back to the main AioContext
2912 bdrv_try_set_aio_context(old_bs
, qemu_get_aio_context(), NULL
);
2917 * This function steals the reference to child_bs from the caller.
2918 * That reference is later dropped by bdrv_root_unref_child().
2920 * On failure NULL is returned, errp is set and the reference to
2921 * child_bs is also dropped.
2923 * The caller must hold the AioContext lock @child_bs, but not that of @ctx
2924 * (unless @child_bs is already in @ctx).
2926 BdrvChild
*bdrv_root_attach_child(BlockDriverState
*child_bs
,
2927 const char *child_name
,
2928 const BdrvChildClass
*child_class
,
2929 BdrvChildRole child_role
,
2930 uint64_t perm
, uint64_t shared_perm
,
2931 void *opaque
, Error
**errp
)
2934 BdrvChild
*child
= NULL
;
2935 Transaction
*tran
= tran_new();
2937 ret
= bdrv_attach_child_common(child_bs
, child_name
, child_class
,
2938 child_role
, perm
, shared_perm
, opaque
,
2939 &child
, tran
, errp
);
2944 ret
= bdrv_refresh_perms(child_bs
, errp
);
2947 tran_finalize(tran
, ret
);
2948 /* child is unset on failure by bdrv_attach_child_common_abort() */
2949 assert((ret
< 0) == !child
);
2951 bdrv_unref(child_bs
);
2956 * This function transfers the reference to child_bs from the caller
2957 * to parent_bs. That reference is later dropped by parent_bs on
2958 * bdrv_close() or if someone calls bdrv_unref_child().
2960 * On failure NULL is returned, errp is set and the reference to
2961 * child_bs is also dropped.
2963 * If @parent_bs and @child_bs are in different AioContexts, the caller must
2964 * hold the AioContext lock for @child_bs, but not for @parent_bs.
2966 BdrvChild
*bdrv_attach_child(BlockDriverState
*parent_bs
,
2967 BlockDriverState
*child_bs
,
2968 const char *child_name
,
2969 const BdrvChildClass
*child_class
,
2970 BdrvChildRole child_role
,
2974 BdrvChild
*child
= NULL
;
2975 Transaction
*tran
= tran_new();
2977 ret
= bdrv_attach_child_noperm(parent_bs
, child_bs
, child_name
, child_class
,
2978 child_role
, &child
, tran
, errp
);
2983 ret
= bdrv_refresh_perms(parent_bs
, errp
);
2989 tran_finalize(tran
, ret
);
2990 /* child is unset on failure by bdrv_attach_child_common_abort() */
2991 assert((ret
< 0) == !child
);
2993 bdrv_unref(child_bs
);
2998 /* Callers must ensure that child->frozen is false. */
2999 void bdrv_root_unref_child(BdrvChild
*child
)
3001 BlockDriverState
*child_bs
;
3003 child_bs
= child
->bs
;
3004 bdrv_detach_child(child
);
3005 bdrv_unref(child_bs
);
3008 typedef struct BdrvSetInheritsFrom
{
3009 BlockDriverState
*bs
;
3010 BlockDriverState
*old_inherits_from
;
3011 } BdrvSetInheritsFrom
;
3013 static void bdrv_set_inherits_from_abort(void *opaque
)
3015 BdrvSetInheritsFrom
*s
= opaque
;
3017 s
->bs
->inherits_from
= s
->old_inherits_from
;
3020 static TransactionActionDrv bdrv_set_inherits_from_drv
= {
3021 .abort
= bdrv_set_inherits_from_abort
,
3025 /* @tran is allowed to be NULL. In this case no rollback is possible */
3026 static void bdrv_set_inherits_from(BlockDriverState
*bs
,
3027 BlockDriverState
*new_inherits_from
,
3031 BdrvSetInheritsFrom
*s
= g_new(BdrvSetInheritsFrom
, 1);
3033 *s
= (BdrvSetInheritsFrom
) {
3035 .old_inherits_from
= bs
->inherits_from
,
3038 tran_add(tran
, &bdrv_set_inherits_from_drv
, s
);
3041 bs
->inherits_from
= new_inherits_from
;
3045 * Clear all inherits_from pointers from children and grandchildren of
3046 * @root that point to @root, where necessary.
3047 * @tran is allowed to be NULL. In this case no rollback is possible
3049 static void bdrv_unset_inherits_from(BlockDriverState
*root
, BdrvChild
*child
,
3054 if (child
->bs
->inherits_from
== root
) {
3056 * Remove inherits_from only when the last reference between root and
3057 * child->bs goes away.
3059 QLIST_FOREACH(c
, &root
->children
, next
) {
3060 if (c
!= child
&& c
->bs
== child
->bs
) {
3065 bdrv_set_inherits_from(child
->bs
, NULL
, tran
);
3069 QLIST_FOREACH(c
, &child
->bs
->children
, next
) {
3070 bdrv_unset_inherits_from(root
, c
, tran
);
3074 /* Callers must ensure that child->frozen is false. */
3075 void bdrv_unref_child(BlockDriverState
*parent
, BdrvChild
*child
)
3077 if (child
== NULL
) {
3081 bdrv_unset_inherits_from(parent
, child
, NULL
);
3082 bdrv_root_unref_child(child
);
3086 static void bdrv_parent_cb_change_media(BlockDriverState
*bs
, bool load
)
3089 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
3090 if (c
->klass
->change_media
) {
3091 c
->klass
->change_media(c
, load
);
3096 /* Return true if you can reach parent going through child->inherits_from
3097 * recursively. If parent or child are NULL, return false */
3098 static bool bdrv_inherits_from_recursive(BlockDriverState
*child
,
3099 BlockDriverState
*parent
)
3101 while (child
&& child
!= parent
) {
3102 child
= child
->inherits_from
;
3105 return child
!= NULL
;
3109 * Return the BdrvChildRole for @bs's backing child. bs->backing is
3110 * mostly used for COW backing children (role = COW), but also for
3111 * filtered children (role = FILTERED | PRIMARY).
3113 static BdrvChildRole
bdrv_backing_role(BlockDriverState
*bs
)
3115 if (bs
->drv
&& bs
->drv
->is_filter
) {
3116 return BDRV_CHILD_FILTERED
| BDRV_CHILD_PRIMARY
;
3118 return BDRV_CHILD_COW
;
3123 * Sets the bs->backing or bs->file link of a BDS. A new reference is created;
3124 * callers which don't need their own reference any more must call bdrv_unref().
3126 * Function doesn't update permissions, caller is responsible for this.
3128 static int bdrv_set_file_or_backing_noperm(BlockDriverState
*parent_bs
,
3129 BlockDriverState
*child_bs
,
3131 Transaction
*tran
, Error
**errp
)
3134 bool update_inherits_from
=
3135 bdrv_inherits_from_recursive(child_bs
, parent_bs
);
3136 BdrvChild
*child
= is_backing
? parent_bs
->backing
: parent_bs
->file
;
3139 if (!parent_bs
->drv
) {
3141 * Node without drv is an object without a class :/. TODO: finally fix
3142 * qcow2 driver to never clear bs->drv and implement format corruption
3143 * handling in other way.
3145 error_setg(errp
, "Node corrupted");
3149 if (child
&& child
->frozen
) {
3150 error_setg(errp
, "Cannot change frozen '%s' link from '%s' to '%s'",
3151 child
->name
, parent_bs
->node_name
, child
->bs
->node_name
);
3155 if (parent_bs
->drv
->is_filter
) {
3156 role
= BDRV_CHILD_FILTERED
| BDRV_CHILD_PRIMARY
;
3157 } else if (is_backing
) {
3158 role
= BDRV_CHILD_COW
;
3161 * We only can use same role as it is in existing child. We don't have
3162 * infrastructure to determine role of file child in generic way
3165 error_setg(errp
, "Cannot set file child to format node without "
3173 bdrv_unset_inherits_from(parent_bs
, child
, tran
);
3174 bdrv_remove_file_or_backing_child(parent_bs
, child
, tran
);
3181 ret
= bdrv_attach_child_noperm(parent_bs
, child_bs
,
3182 is_backing
? "backing" : "file",
3183 &child_of_bds
, role
,
3184 is_backing
? &parent_bs
->backing
:
3193 * If inherits_from pointed recursively to bs then let's update it to
3194 * point directly to bs (else it will become NULL).
3196 if (update_inherits_from
) {
3197 bdrv_set_inherits_from(child_bs
, parent_bs
, tran
);
3201 bdrv_refresh_limits(parent_bs
, tran
, NULL
);
3206 static int bdrv_set_backing_noperm(BlockDriverState
*bs
,
3207 BlockDriverState
*backing_hd
,
3208 Transaction
*tran
, Error
**errp
)
3210 return bdrv_set_file_or_backing_noperm(bs
, backing_hd
, true, tran
, errp
);
3213 int bdrv_set_backing_hd(BlockDriverState
*bs
, BlockDriverState
*backing_hd
,
3217 Transaction
*tran
= tran_new();
3219 ret
= bdrv_set_backing_noperm(bs
, backing_hd
, tran
, errp
);
3224 ret
= bdrv_refresh_perms(bs
, errp
);
3226 tran_finalize(tran
, ret
);
3232 * Opens the backing file for a BlockDriverState if not yet open
3234 * bdref_key specifies the key for the image's BlockdevRef in the options QDict.
3235 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
3236 * itself, all options starting with "${bdref_key}." are considered part of the
3239 * TODO Can this be unified with bdrv_open_image()?
3241 int bdrv_open_backing_file(BlockDriverState
*bs
, QDict
*parent_options
,
3242 const char *bdref_key
, Error
**errp
)
3244 char *backing_filename
= NULL
;
3245 char *bdref_key_dot
;
3246 const char *reference
= NULL
;
3248 bool implicit_backing
= false;
3249 BlockDriverState
*backing_hd
;
3251 QDict
*tmp_parent_options
= NULL
;
3252 Error
*local_err
= NULL
;
3254 if (bs
->backing
!= NULL
) {
3258 /* NULL means an empty set of options */
3259 if (parent_options
== NULL
) {
3260 tmp_parent_options
= qdict_new();
3261 parent_options
= tmp_parent_options
;
3264 bs
->open_flags
&= ~BDRV_O_NO_BACKING
;
3266 bdref_key_dot
= g_strdup_printf("%s.", bdref_key
);
3267 qdict_extract_subqdict(parent_options
, &options
, bdref_key_dot
);
3268 g_free(bdref_key_dot
);
3271 * Caution: while qdict_get_try_str() is fine, getting non-string
3272 * types would require more care. When @parent_options come from
3273 * -blockdev or blockdev_add, its members are typed according to
3274 * the QAPI schema, but when they come from -drive, they're all
3277 reference
= qdict_get_try_str(parent_options
, bdref_key
);
3278 if (reference
|| qdict_haskey(options
, "file.filename")) {
3279 /* keep backing_filename NULL */
3280 } else if (bs
->backing_file
[0] == '\0' && qdict_size(options
) == 0) {
3281 qobject_unref(options
);
3284 if (qdict_size(options
) == 0) {
3285 /* If the user specifies options that do not modify the
3286 * backing file's behavior, we might still consider it the
3287 * implicit backing file. But it's easier this way, and
3288 * just specifying some of the backing BDS's options is
3289 * only possible with -drive anyway (otherwise the QAPI
3290 * schema forces the user to specify everything). */
3291 implicit_backing
= !strcmp(bs
->auto_backing_file
, bs
->backing_file
);
3294 backing_filename
= bdrv_get_full_backing_filename(bs
, &local_err
);
3297 error_propagate(errp
, local_err
);
3298 qobject_unref(options
);
3303 if (!bs
->drv
|| !bs
->drv
->supports_backing
) {
3305 error_setg(errp
, "Driver doesn't support backing files");
3306 qobject_unref(options
);
3311 bs
->backing_format
[0] != '\0' && !qdict_haskey(options
, "driver")) {
3312 qdict_put_str(options
, "driver", bs
->backing_format
);
3315 backing_hd
= bdrv_open_inherit(backing_filename
, reference
, options
, 0, bs
,
3316 &child_of_bds
, bdrv_backing_role(bs
), errp
);
3318 bs
->open_flags
|= BDRV_O_NO_BACKING
;
3319 error_prepend(errp
, "Could not open backing file: ");
3324 if (implicit_backing
) {
3325 bdrv_refresh_filename(backing_hd
);
3326 pstrcpy(bs
->auto_backing_file
, sizeof(bs
->auto_backing_file
),
3327 backing_hd
->filename
);
3330 /* Hook up the backing file link; drop our reference, bs owns the
3331 * backing_hd reference now */
3332 ret
= bdrv_set_backing_hd(bs
, backing_hd
, errp
);
3333 bdrv_unref(backing_hd
);
3338 qdict_del(parent_options
, bdref_key
);
3341 g_free(backing_filename
);
3342 qobject_unref(tmp_parent_options
);
3346 static BlockDriverState
*
3347 bdrv_open_child_bs(const char *filename
, QDict
*options
, const char *bdref_key
,
3348 BlockDriverState
*parent
, const BdrvChildClass
*child_class
,
3349 BdrvChildRole child_role
, bool allow_none
, Error
**errp
)
3351 BlockDriverState
*bs
= NULL
;
3352 QDict
*image_options
;
3353 char *bdref_key_dot
;
3354 const char *reference
;
3356 assert(child_class
!= NULL
);
3358 bdref_key_dot
= g_strdup_printf("%s.", bdref_key
);
3359 qdict_extract_subqdict(options
, &image_options
, bdref_key_dot
);
3360 g_free(bdref_key_dot
);
3363 * Caution: while qdict_get_try_str() is fine, getting non-string
3364 * types would require more care. When @options come from
3365 * -blockdev or blockdev_add, its members are typed according to
3366 * the QAPI schema, but when they come from -drive, they're all
3369 reference
= qdict_get_try_str(options
, bdref_key
);
3370 if (!filename
&& !reference
&& !qdict_size(image_options
)) {
3372 error_setg(errp
, "A block device must be specified for \"%s\"",
3375 qobject_unref(image_options
);
3379 bs
= bdrv_open_inherit(filename
, reference
, image_options
, 0,
3380 parent
, child_class
, child_role
, errp
);
3386 qdict_del(options
, bdref_key
);
3391 * Opens a disk image whose options are given as BlockdevRef in another block
3394 * If allow_none is true, no image will be opened if filename is false and no
3395 * BlockdevRef is given. NULL will be returned, but errp remains unset.
3397 * bdrev_key specifies the key for the image's BlockdevRef in the options QDict.
3398 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
3399 * itself, all options starting with "${bdref_key}." are considered part of the
3402 * The BlockdevRef will be removed from the options QDict.
3404 BdrvChild
*bdrv_open_child(const char *filename
,
3405 QDict
*options
, const char *bdref_key
,
3406 BlockDriverState
*parent
,
3407 const BdrvChildClass
*child_class
,
3408 BdrvChildRole child_role
,
3409 bool allow_none
, Error
**errp
)
3411 BlockDriverState
*bs
;
3413 bs
= bdrv_open_child_bs(filename
, options
, bdref_key
, parent
, child_class
,
3414 child_role
, allow_none
, errp
);
3419 return bdrv_attach_child(parent
, bs
, bdref_key
, child_class
, child_role
,
3424 * TODO Future callers may need to specify parent/child_class in order for
3425 * option inheritance to work. Existing callers use it for the root node.
3427 BlockDriverState
*bdrv_open_blockdev_ref(BlockdevRef
*ref
, Error
**errp
)
3429 BlockDriverState
*bs
= NULL
;
3430 QObject
*obj
= NULL
;
3431 QDict
*qdict
= NULL
;
3432 const char *reference
= NULL
;
3435 if (ref
->type
== QTYPE_QSTRING
) {
3436 reference
= ref
->u
.reference
;
3438 BlockdevOptions
*options
= &ref
->u
.definition
;
3439 assert(ref
->type
== QTYPE_QDICT
);
3441 v
= qobject_output_visitor_new(&obj
);
3442 visit_type_BlockdevOptions(v
, NULL
, &options
, &error_abort
);
3443 visit_complete(v
, &obj
);
3445 qdict
= qobject_to(QDict
, obj
);
3446 qdict_flatten(qdict
);
3448 /* bdrv_open_inherit() defaults to the values in bdrv_flags (for
3449 * compatibility with other callers) rather than what we want as the
3450 * real defaults. Apply the defaults here instead. */
3451 qdict_set_default_str(qdict
, BDRV_OPT_CACHE_DIRECT
, "off");
3452 qdict_set_default_str(qdict
, BDRV_OPT_CACHE_NO_FLUSH
, "off");
3453 qdict_set_default_str(qdict
, BDRV_OPT_READ_ONLY
, "off");
3454 qdict_set_default_str(qdict
, BDRV_OPT_AUTO_READ_ONLY
, "off");
3458 bs
= bdrv_open_inherit(NULL
, reference
, qdict
, 0, NULL
, NULL
, 0, errp
);
3465 static BlockDriverState
*bdrv_append_temp_snapshot(BlockDriverState
*bs
,
3467 QDict
*snapshot_options
,
3470 /* TODO: extra byte is a hack to ensure MAX_PATH space on Windows. */
3471 char *tmp_filename
= g_malloc0(PATH_MAX
+ 1);
3473 QemuOpts
*opts
= NULL
;
3474 BlockDriverState
*bs_snapshot
= NULL
;
3477 /* if snapshot, we create a temporary backing file and open it
3478 instead of opening 'filename' directly */
3480 /* Get the required size from the image */
3481 total_size
= bdrv_getlength(bs
);
3482 if (total_size
< 0) {
3483 error_setg_errno(errp
, -total_size
, "Could not get image size");
3487 /* Create the temporary image */
3488 ret
= get_tmp_filename(tmp_filename
, PATH_MAX
+ 1);
3490 error_setg_errno(errp
, -ret
, "Could not get temporary filename");
3494 opts
= qemu_opts_create(bdrv_qcow2
.create_opts
, NULL
, 0,
3496 qemu_opt_set_number(opts
, BLOCK_OPT_SIZE
, total_size
, &error_abort
);
3497 ret
= bdrv_create(&bdrv_qcow2
, tmp_filename
, opts
, errp
);
3498 qemu_opts_del(opts
);
3500 error_prepend(errp
, "Could not create temporary overlay '%s': ",
3505 /* Prepare options QDict for the temporary file */
3506 qdict_put_str(snapshot_options
, "file.driver", "file");
3507 qdict_put_str(snapshot_options
, "file.filename", tmp_filename
);
3508 qdict_put_str(snapshot_options
, "driver", "qcow2");
3510 bs_snapshot
= bdrv_open(NULL
, NULL
, snapshot_options
, flags
, errp
);
3511 snapshot_options
= NULL
;
3516 ret
= bdrv_append(bs_snapshot
, bs
, errp
);
3523 qobject_unref(snapshot_options
);
3524 g_free(tmp_filename
);
3529 * Opens a disk image (raw, qcow2, vmdk, ...)
3531 * options is a QDict of options to pass to the block drivers, or NULL for an
3532 * empty set of options. The reference to the QDict belongs to the block layer
3533 * after the call (even on failure), so if the caller intends to reuse the
3534 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
3536 * If *pbs is NULL, a new BDS will be created with a pointer to it stored there.
3537 * If it is not NULL, the referenced BDS will be reused.
3539 * The reference parameter may be used to specify an existing block device which
3540 * should be opened. If specified, neither options nor a filename may be given,
3541 * nor can an existing BDS be reused (that is, *pbs has to be NULL).
3543 static BlockDriverState
*bdrv_open_inherit(const char *filename
,
3544 const char *reference
,
3545 QDict
*options
, int flags
,
3546 BlockDriverState
*parent
,
3547 const BdrvChildClass
*child_class
,
3548 BdrvChildRole child_role
,
3552 BlockBackend
*file
= NULL
;
3553 BlockDriverState
*bs
;
3554 BlockDriver
*drv
= NULL
;
3556 const char *drvname
;
3557 const char *backing
;
3558 Error
*local_err
= NULL
;
3559 QDict
*snapshot_options
= NULL
;
3560 int snapshot_flags
= 0;
3562 assert(!child_class
|| !flags
);
3563 assert(!child_class
== !parent
);
3566 bool options_non_empty
= options
? qdict_size(options
) : false;
3567 qobject_unref(options
);
3569 if (filename
|| options_non_empty
) {
3570 error_setg(errp
, "Cannot reference an existing block device with "
3571 "additional options or a new filename");
3575 bs
= bdrv_lookup_bs(reference
, reference
, errp
);
3586 /* NULL means an empty set of options */
3587 if (options
== NULL
) {
3588 options
= qdict_new();
3591 /* json: syntax counts as explicit options, as if in the QDict */
3592 parse_json_protocol(options
, &filename
, &local_err
);
3597 bs
->explicit_options
= qdict_clone_shallow(options
);
3600 bool parent_is_format
;
3603 parent_is_format
= parent
->drv
->is_format
;
3606 * parent->drv is not set yet because this node is opened for
3607 * (potential) format probing. That means that @parent is going
3608 * to be a format node.
3610 parent_is_format
= true;
3613 bs
->inherits_from
= parent
;
3614 child_class
->inherit_options(child_role
, parent_is_format
,
3616 parent
->open_flags
, parent
->options
);
3619 ret
= bdrv_fill_options(&options
, filename
, &flags
, &local_err
);
3625 * Set the BDRV_O_RDWR and BDRV_O_ALLOW_RDWR flags.
3626 * Caution: getting a boolean member of @options requires care.
3627 * When @options come from -blockdev or blockdev_add, members are
3628 * typed according to the QAPI schema, but when they come from
3629 * -drive, they're all QString.
3631 if (g_strcmp0(qdict_get_try_str(options
, BDRV_OPT_READ_ONLY
), "on") &&
3632 !qdict_get_try_bool(options
, BDRV_OPT_READ_ONLY
, false)) {
3633 flags
|= (BDRV_O_RDWR
| BDRV_O_ALLOW_RDWR
);
3635 flags
&= ~BDRV_O_RDWR
;
3638 if (flags
& BDRV_O_SNAPSHOT
) {
3639 snapshot_options
= qdict_new();
3640 bdrv_temp_snapshot_options(&snapshot_flags
, snapshot_options
,
3642 /* Let bdrv_backing_options() override "read-only" */
3643 qdict_del(options
, BDRV_OPT_READ_ONLY
);
3644 bdrv_inherited_options(BDRV_CHILD_COW
, true,
3645 &flags
, options
, flags
, options
);
3648 bs
->open_flags
= flags
;
3649 bs
->options
= options
;
3650 options
= qdict_clone_shallow(options
);
3652 /* Find the right image format driver */
3653 /* See cautionary note on accessing @options above */
3654 drvname
= qdict_get_try_str(options
, "driver");
3656 drv
= bdrv_find_format(drvname
);
3658 error_setg(errp
, "Unknown driver: '%s'", drvname
);
3663 assert(drvname
|| !(flags
& BDRV_O_PROTOCOL
));
3665 /* See cautionary note on accessing @options above */
3666 backing
= qdict_get_try_str(options
, "backing");
3667 if (qobject_to(QNull
, qdict_get(options
, "backing")) != NULL
||
3668 (backing
&& *backing
== '\0'))
3671 warn_report("Use of \"backing\": \"\" is deprecated; "
3672 "use \"backing\": null instead");
3674 flags
|= BDRV_O_NO_BACKING
;
3675 qdict_del(bs
->explicit_options
, "backing");
3676 qdict_del(bs
->options
, "backing");
3677 qdict_del(options
, "backing");
3680 /* Open image file without format layer. This BlockBackend is only used for
3681 * probing, the block drivers will do their own bdrv_open_child() for the
3682 * same BDS, which is why we put the node name back into options. */
3683 if ((flags
& BDRV_O_PROTOCOL
) == 0) {
3684 BlockDriverState
*file_bs
;
3686 file_bs
= bdrv_open_child_bs(filename
, options
, "file", bs
,
3687 &child_of_bds
, BDRV_CHILD_IMAGE
,
3692 if (file_bs
!= NULL
) {
3693 /* Not requesting BLK_PERM_CONSISTENT_READ because we're only
3694 * looking at the header to guess the image format. This works even
3695 * in cases where a guest would not see a consistent state. */
3696 file
= blk_new(bdrv_get_aio_context(file_bs
), 0, BLK_PERM_ALL
);
3697 blk_insert_bs(file
, file_bs
, &local_err
);
3698 bdrv_unref(file_bs
);
3703 qdict_put_str(options
, "file", bdrv_get_node_name(file_bs
));
3707 /* Image format probing */
3710 ret
= find_image_format(file
, filename
, &drv
, &local_err
);
3715 * This option update would logically belong in bdrv_fill_options(),
3716 * but we first need to open bs->file for the probing to work, while
3717 * opening bs->file already requires the (mostly) final set of options
3718 * so that cache mode etc. can be inherited.
3720 * Adding the driver later is somewhat ugly, but it's not an option
3721 * that would ever be inherited, so it's correct. We just need to make
3722 * sure to update both bs->options (which has the full effective
3723 * options for bs) and options (which has file.* already removed).
3725 qdict_put_str(bs
->options
, "driver", drv
->format_name
);
3726 qdict_put_str(options
, "driver", drv
->format_name
);
3728 error_setg(errp
, "Must specify either driver or file");
3732 /* BDRV_O_PROTOCOL must be set iff a protocol BDS is about to be created */
3733 assert(!!(flags
& BDRV_O_PROTOCOL
) == !!drv
->bdrv_file_open
);
3734 /* file must be NULL if a protocol BDS is about to be created
3735 * (the inverse results in an error message from bdrv_open_common()) */
3736 assert(!(flags
& BDRV_O_PROTOCOL
) || !file
);
3738 /* Open the image */
3739 ret
= bdrv_open_common(bs
, file
, options
, &local_err
);
3749 /* If there is a backing file, use it */
3750 if ((flags
& BDRV_O_NO_BACKING
) == 0) {
3751 ret
= bdrv_open_backing_file(bs
, options
, "backing", &local_err
);
3753 goto close_and_fail
;
3757 /* Remove all children options and references
3758 * from bs->options and bs->explicit_options */
3759 QLIST_FOREACH(child
, &bs
->children
, next
) {
3760 char *child_key_dot
;
3761 child_key_dot
= g_strdup_printf("%s.", child
->name
);
3762 qdict_extract_subqdict(bs
->explicit_options
, NULL
, child_key_dot
);
3763 qdict_extract_subqdict(bs
->options
, NULL
, child_key_dot
);
3764 qdict_del(bs
->explicit_options
, child
->name
);
3765 qdict_del(bs
->options
, child
->name
);
3766 g_free(child_key_dot
);
3769 /* Check if any unknown options were used */
3770 if (qdict_size(options
) != 0) {
3771 const QDictEntry
*entry
= qdict_first(options
);
3772 if (flags
& BDRV_O_PROTOCOL
) {
3773 error_setg(errp
, "Block protocol '%s' doesn't support the option "
3774 "'%s'", drv
->format_name
, entry
->key
);
3777 "Block format '%s' does not support the option '%s'",
3778 drv
->format_name
, entry
->key
);
3781 goto close_and_fail
;
3784 bdrv_parent_cb_change_media(bs
, true);
3786 qobject_unref(options
);
3789 /* For snapshot=on, create a temporary qcow2 overlay. bs points to the
3790 * temporary snapshot afterwards. */
3791 if (snapshot_flags
) {
3792 BlockDriverState
*snapshot_bs
;
3793 snapshot_bs
= bdrv_append_temp_snapshot(bs
, snapshot_flags
,
3794 snapshot_options
, &local_err
);
3795 snapshot_options
= NULL
;
3797 goto close_and_fail
;
3799 /* We are not going to return bs but the overlay on top of it
3800 * (snapshot_bs); thus, we have to drop the strong reference to bs
3801 * (which we obtained by calling bdrv_new()). bs will not be deleted,
3802 * though, because the overlay still has a reference to it. */
3811 qobject_unref(snapshot_options
);
3812 qobject_unref(bs
->explicit_options
);
3813 qobject_unref(bs
->options
);
3814 qobject_unref(options
);
3816 bs
->explicit_options
= NULL
;
3818 error_propagate(errp
, local_err
);
3823 qobject_unref(snapshot_options
);
3824 qobject_unref(options
);
3825 error_propagate(errp
, local_err
);
3829 BlockDriverState
*bdrv_open(const char *filename
, const char *reference
,
3830 QDict
*options
, int flags
, Error
**errp
)
3832 return bdrv_open_inherit(filename
, reference
, options
, flags
, NULL
,
3836 /* Return true if the NULL-terminated @list contains @str */
3837 static bool is_str_in_list(const char *str
, const char *const *list
)
3841 for (i
= 0; list
[i
] != NULL
; i
++) {
3842 if (!strcmp(str
, list
[i
])) {
3851 * Check that every option set in @bs->options is also set in
3854 * Options listed in the common_options list and in
3855 * @bs->drv->mutable_opts are skipped.
3857 * Return 0 on success, otherwise return -EINVAL and set @errp.
3859 static int bdrv_reset_options_allowed(BlockDriverState
*bs
,
3860 const QDict
*new_opts
, Error
**errp
)
3862 const QDictEntry
*e
;
3863 /* These options are common to all block drivers and are handled
3864 * in bdrv_reopen_prepare() so they can be left out of @new_opts */
3865 const char *const common_options
[] = {
3866 "node-name", "discard", "cache.direct", "cache.no-flush",
3867 "read-only", "auto-read-only", "detect-zeroes", NULL
3870 for (e
= qdict_first(bs
->options
); e
; e
= qdict_next(bs
->options
, e
)) {
3871 if (!qdict_haskey(new_opts
, e
->key
) &&
3872 !is_str_in_list(e
->key
, common_options
) &&
3873 !is_str_in_list(e
->key
, bs
->drv
->mutable_opts
)) {
3874 error_setg(errp
, "Option '%s' cannot be reset "
3875 "to its default value", e
->key
);
3884 * Returns true if @child can be reached recursively from @bs
3886 static bool bdrv_recurse_has_child(BlockDriverState
*bs
,
3887 BlockDriverState
*child
)
3895 QLIST_FOREACH(c
, &bs
->children
, next
) {
3896 if (bdrv_recurse_has_child(c
->bs
, child
)) {
3905 * Adds a BlockDriverState to a simple queue for an atomic, transactional
3906 * reopen of multiple devices.
3908 * bs_queue can either be an existing BlockReopenQueue that has had QTAILQ_INIT
3909 * already performed, or alternatively may be NULL a new BlockReopenQueue will
3910 * be created and initialized. This newly created BlockReopenQueue should be
3911 * passed back in for subsequent calls that are intended to be of the same
3914 * bs is the BlockDriverState to add to the reopen queue.
3916 * options contains the changed options for the associated bs
3917 * (the BlockReopenQueue takes ownership)
3919 * flags contains the open flags for the associated bs
3921 * returns a pointer to bs_queue, which is either the newly allocated
3922 * bs_queue, or the existing bs_queue being used.
3924 * bs must be drained between bdrv_reopen_queue() and bdrv_reopen_multiple().
3926 static BlockReopenQueue
*bdrv_reopen_queue_child(BlockReopenQueue
*bs_queue
,
3927 BlockDriverState
*bs
,
3929 const BdrvChildClass
*klass
,
3931 bool parent_is_format
,
3932 QDict
*parent_options
,
3938 BlockReopenQueueEntry
*bs_entry
;
3940 QDict
*old_options
, *explicit_options
, *options_copy
;
3944 /* Make sure that the caller remembered to use a drained section. This is
3945 * important to avoid graph changes between the recursive queuing here and
3946 * bdrv_reopen_multiple(). */
3947 assert(bs
->quiesce_counter
> 0);
3949 if (bs_queue
== NULL
) {
3950 bs_queue
= g_new0(BlockReopenQueue
, 1);
3951 QTAILQ_INIT(bs_queue
);
3955 options
= qdict_new();
3958 /* Check if this BlockDriverState is already in the queue */
3959 QTAILQ_FOREACH(bs_entry
, bs_queue
, entry
) {
3960 if (bs
== bs_entry
->state
.bs
) {
3966 * Precedence of options:
3967 * 1. Explicitly passed in options (highest)
3968 * 2. Retained from explicitly set options of bs
3969 * 3. Inherited from parent node
3970 * 4. Retained from effective options of bs
3973 /* Old explicitly set values (don't overwrite by inherited value) */
3974 if (bs_entry
|| keep_old_opts
) {
3975 old_options
= qdict_clone_shallow(bs_entry
?
3976 bs_entry
->state
.explicit_options
:
3977 bs
->explicit_options
);
3978 bdrv_join_options(bs
, options
, old_options
);
3979 qobject_unref(old_options
);
3982 explicit_options
= qdict_clone_shallow(options
);
3984 /* Inherit from parent node */
3985 if (parent_options
) {
3987 klass
->inherit_options(role
, parent_is_format
, &flags
, options
,
3988 parent_flags
, parent_options
);
3990 flags
= bdrv_get_flags(bs
);
3993 if (keep_old_opts
) {
3994 /* Old values are used for options that aren't set yet */
3995 old_options
= qdict_clone_shallow(bs
->options
);
3996 bdrv_join_options(bs
, options
, old_options
);
3997 qobject_unref(old_options
);
4000 /* We have the final set of options so let's update the flags */
4001 options_copy
= qdict_clone_shallow(options
);
4002 opts
= qemu_opts_create(&bdrv_runtime_opts
, NULL
, 0, &error_abort
);
4003 qemu_opts_absorb_qdict(opts
, options_copy
, NULL
);
4004 update_flags_from_options(&flags
, opts
);
4005 qemu_opts_del(opts
);
4006 qobject_unref(options_copy
);
4008 /* bdrv_open_inherit() sets and clears some additional flags internally */
4009 flags
&= ~BDRV_O_PROTOCOL
;
4010 if (flags
& BDRV_O_RDWR
) {
4011 flags
|= BDRV_O_ALLOW_RDWR
;
4015 bs_entry
= g_new0(BlockReopenQueueEntry
, 1);
4016 QTAILQ_INSERT_TAIL(bs_queue
, bs_entry
, entry
);
4018 qobject_unref(bs_entry
->state
.options
);
4019 qobject_unref(bs_entry
->state
.explicit_options
);
4022 bs_entry
->state
.bs
= bs
;
4023 bs_entry
->state
.options
= options
;
4024 bs_entry
->state
.explicit_options
= explicit_options
;
4025 bs_entry
->state
.flags
= flags
;
4028 * If keep_old_opts is false then it means that unspecified
4029 * options must be reset to their original value. We don't allow
4030 * resetting 'backing' but we need to know if the option is
4031 * missing in order to decide if we have to return an error.
4033 if (!keep_old_opts
) {
4034 bs_entry
->state
.backing_missing
=
4035 !qdict_haskey(options
, "backing") &&
4036 !qdict_haskey(options
, "backing.driver");
4039 QLIST_FOREACH(child
, &bs
->children
, next
) {
4040 QDict
*new_child_options
= NULL
;
4041 bool child_keep_old
= keep_old_opts
;
4043 /* reopen can only change the options of block devices that were
4044 * implicitly created and inherited options. For other (referenced)
4045 * block devices, a syntax like "backing.foo" results in an error. */
4046 if (child
->bs
->inherits_from
!= bs
) {
4050 /* Check if the options contain a child reference */
4051 if (qdict_haskey(options
, child
->name
)) {
4052 const char *childref
= qdict_get_try_str(options
, child
->name
);
4054 * The current child must not be reopened if the child
4055 * reference is null or points to a different node.
4057 if (g_strcmp0(childref
, child
->bs
->node_name
)) {
4061 * If the child reference points to the current child then
4062 * reopen it with its existing set of options (note that
4063 * it can still inherit new options from the parent).
4065 child_keep_old
= true;
4067 /* Extract child options ("child-name.*") */
4068 char *child_key_dot
= g_strdup_printf("%s.", child
->name
);
4069 qdict_extract_subqdict(explicit_options
, NULL
, child_key_dot
);
4070 qdict_extract_subqdict(options
, &new_child_options
, child_key_dot
);
4071 g_free(child_key_dot
);
4074 bdrv_reopen_queue_child(bs_queue
, child
->bs
, new_child_options
,
4075 child
->klass
, child
->role
, bs
->drv
->is_format
,
4076 options
, flags
, child_keep_old
);
4082 BlockReopenQueue
*bdrv_reopen_queue(BlockReopenQueue
*bs_queue
,
4083 BlockDriverState
*bs
,
4084 QDict
*options
, bool keep_old_opts
)
4086 return bdrv_reopen_queue_child(bs_queue
, bs
, options
, NULL
, 0, false,
4087 NULL
, 0, keep_old_opts
);
4091 * Reopen multiple BlockDriverStates atomically & transactionally.
4093 * The queue passed in (bs_queue) must have been built up previous
4094 * via bdrv_reopen_queue().
4096 * Reopens all BDS specified in the queue, with the appropriate
4097 * flags. All devices are prepared for reopen, and failure of any
4098 * device will cause all device changes to be abandoned, and intermediate
4101 * If all devices prepare successfully, then the changes are committed
4104 * All affected nodes must be drained between bdrv_reopen_queue() and
4105 * bdrv_reopen_multiple().
4107 int bdrv_reopen_multiple(BlockReopenQueue
*bs_queue
, Error
**errp
)
4110 BlockReopenQueueEntry
*bs_entry
, *next
;
4111 Transaction
*tran
= tran_new();
4112 g_autoptr(GHashTable
) found
= NULL
;
4113 g_autoptr(GSList
) refresh_list
= NULL
;
4115 assert(bs_queue
!= NULL
);
4117 QTAILQ_FOREACH(bs_entry
, bs_queue
, entry
) {
4118 ret
= bdrv_flush(bs_entry
->state
.bs
);
4120 error_setg_errno(errp
, -ret
, "Error flushing drive");
4125 QTAILQ_FOREACH(bs_entry
, bs_queue
, entry
) {
4126 assert(bs_entry
->state
.bs
->quiesce_counter
> 0);
4127 ret
= bdrv_reopen_prepare(&bs_entry
->state
, bs_queue
, tran
, errp
);
4131 bs_entry
->prepared
= true;
4134 found
= g_hash_table_new(NULL
, NULL
);
4135 QTAILQ_FOREACH(bs_entry
, bs_queue
, entry
) {
4136 BDRVReopenState
*state
= &bs_entry
->state
;
4138 refresh_list
= bdrv_topological_dfs(refresh_list
, found
, state
->bs
);
4139 if (state
->old_backing_bs
) {
4140 refresh_list
= bdrv_topological_dfs(refresh_list
, found
,
4141 state
->old_backing_bs
);
4146 * Note that file-posix driver rely on permission update done during reopen
4147 * (even if no permission changed), because it wants "new" permissions for
4148 * reconfiguring the fd and that's why it does it in raw_check_perm(), not
4149 * in raw_reopen_prepare() which is called with "old" permissions.
4151 ret
= bdrv_list_refresh_perms(refresh_list
, bs_queue
, tran
, errp
);
4157 * If we reach this point, we have success and just need to apply the
4160 * Reverse order is used to comfort qcow2 driver: on commit it need to write
4161 * IN_USE flag to the image, to mark bitmaps in the image as invalid. But
4162 * children are usually goes after parents in reopen-queue, so go from last
4165 QTAILQ_FOREACH_REVERSE(bs_entry
, bs_queue
, entry
) {
4166 bdrv_reopen_commit(&bs_entry
->state
);
4171 QTAILQ_FOREACH_REVERSE(bs_entry
, bs_queue
, entry
) {
4172 BlockDriverState
*bs
= bs_entry
->state
.bs
;
4174 if (bs
->drv
->bdrv_reopen_commit_post
) {
4175 bs
->drv
->bdrv_reopen_commit_post(&bs_entry
->state
);
4184 QTAILQ_FOREACH_SAFE(bs_entry
, bs_queue
, entry
, next
) {
4185 if (bs_entry
->prepared
) {
4186 bdrv_reopen_abort(&bs_entry
->state
);
4188 qobject_unref(bs_entry
->state
.explicit_options
);
4189 qobject_unref(bs_entry
->state
.options
);
4193 QTAILQ_FOREACH_SAFE(bs_entry
, bs_queue
, entry
, next
) {
4201 int bdrv_reopen_set_read_only(BlockDriverState
*bs
, bool read_only
,
4205 BlockReopenQueue
*queue
;
4206 QDict
*opts
= qdict_new();
4208 qdict_put_bool(opts
, BDRV_OPT_READ_ONLY
, read_only
);
4210 bdrv_subtree_drained_begin(bs
);
4211 queue
= bdrv_reopen_queue(NULL
, bs
, opts
, true);
4212 ret
= bdrv_reopen_multiple(queue
, errp
);
4213 bdrv_subtree_drained_end(bs
);
4218 static bool bdrv_reopen_can_attach(BlockDriverState
*parent
,
4220 BlockDriverState
*new_child
,
4223 AioContext
*parent_ctx
= bdrv_get_aio_context(parent
);
4224 AioContext
*child_ctx
= bdrv_get_aio_context(new_child
);
4228 ignore
= g_slist_prepend(NULL
, child
);
4229 ret
= bdrv_can_set_aio_context(new_child
, parent_ctx
, &ignore
, NULL
);
4230 g_slist_free(ignore
);
4235 ignore
= g_slist_prepend(NULL
, child
);
4236 ret
= bdrv_can_set_aio_context(parent
, child_ctx
, &ignore
, errp
);
4237 g_slist_free(ignore
);
4242 * Take a BDRVReopenState and check if the value of 'backing' in the
4243 * reopen_state->options QDict is valid or not.
4245 * If 'backing' is missing from the QDict then return 0.
4247 * If 'backing' contains the node name of the backing file of
4248 * reopen_state->bs then return 0.
4250 * If 'backing' contains a different node name (or is null) then check
4251 * whether the current backing file can be replaced with the new one.
4252 * If that's the case then reopen_state->replace_backing_bs is set to
4253 * true and reopen_state->new_backing_bs contains a pointer to the new
4254 * backing BlockDriverState (or NULL).
4256 * Return 0 on success, otherwise return < 0 and set @errp.
4258 static int bdrv_reopen_parse_backing(BDRVReopenState
*reopen_state
,
4259 Transaction
*set_backings_tran
,
4262 BlockDriverState
*bs
= reopen_state
->bs
;
4263 BlockDriverState
*overlay_bs
, *below_bs
, *new_backing_bs
;
4267 value
= qdict_get(reopen_state
->options
, "backing");
4268 if (value
== NULL
) {
4272 switch (qobject_type(value
)) {
4274 new_backing_bs
= NULL
;
4277 str
= qstring_get_str(qobject_to(QString
, value
));
4278 new_backing_bs
= bdrv_lookup_bs(NULL
, str
, errp
);
4279 if (new_backing_bs
== NULL
) {
4281 } else if (bdrv_recurse_has_child(new_backing_bs
, bs
)) {
4282 error_setg(errp
, "Making '%s' a backing file of '%s' "
4283 "would create a cycle", str
, bs
->node_name
);
4288 /* 'backing' does not allow any other data type */
4289 g_assert_not_reached();
4293 * Check AioContext compatibility so that the bdrv_set_backing_hd() call in
4294 * bdrv_reopen_commit() won't fail.
4296 if (new_backing_bs
) {
4297 if (!bdrv_reopen_can_attach(bs
, bs
->backing
, new_backing_bs
, errp
)) {
4303 * Ensure that @bs can really handle backing files, because we are
4304 * about to give it one (or swap the existing one)
4306 if (bs
->drv
->is_filter
) {
4307 /* Filters always have a file or a backing child */
4309 error_setg(errp
, "'%s' is a %s filter node that does not support a "
4310 "backing child", bs
->node_name
, bs
->drv
->format_name
);
4313 } else if (!bs
->drv
->supports_backing
) {
4314 error_setg(errp
, "Driver '%s' of node '%s' does not support backing "
4315 "files", bs
->drv
->format_name
, bs
->node_name
);
4320 * Find the "actual" backing file by skipping all links that point
4321 * to an implicit node, if any (e.g. a commit filter node).
4322 * We cannot use any of the bdrv_skip_*() functions here because
4323 * those return the first explicit node, while we are looking for
4327 for (below_bs
= bdrv_filter_or_cow_bs(overlay_bs
);
4328 below_bs
&& below_bs
->implicit
;
4329 below_bs
= bdrv_filter_or_cow_bs(overlay_bs
))
4331 overlay_bs
= below_bs
;
4334 /* If we want to replace the backing file we need some extra checks */
4335 if (new_backing_bs
!= bdrv_filter_or_cow_bs(overlay_bs
)) {
4338 /* Check for implicit nodes between bs and its backing file */
4339 if (bs
!= overlay_bs
) {
4340 error_setg(errp
, "Cannot change backing link if '%s' has "
4341 "an implicit backing file", bs
->node_name
);
4345 * Check if the backing link that we want to replace is frozen.
4347 * bdrv_filter_or_cow_child(overlay_bs) == overlay_bs->backing,
4348 * because we know that overlay_bs == bs, and that @bs
4349 * either is a filter that uses ->backing or a COW format BDS
4350 * with bs->drv->supports_backing == true.
4352 if (bdrv_is_backing_chain_frozen(overlay_bs
,
4353 child_bs(overlay_bs
->backing
), errp
))
4357 reopen_state
->replace_backing_bs
= true;
4358 reopen_state
->old_backing_bs
= bs
->backing
? bs
->backing
->bs
: NULL
;
4359 ret
= bdrv_set_backing_noperm(bs
, new_backing_bs
, set_backings_tran
,
4370 * Prepares a BlockDriverState for reopen. All changes are staged in the
4371 * 'opaque' field of the BDRVReopenState, which is used and allocated by
4372 * the block driver layer .bdrv_reopen_prepare()
4374 * bs is the BlockDriverState to reopen
4375 * flags are the new open flags
4376 * queue is the reopen queue
4378 * Returns 0 on success, non-zero on error. On error errp will be set
4381 * On failure, bdrv_reopen_abort() will be called to clean up any data.
4382 * It is the responsibility of the caller to then call the abort() or
4383 * commit() for any other BDS that have been left in a prepare() state
4386 static int bdrv_reopen_prepare(BDRVReopenState
*reopen_state
,
4387 BlockReopenQueue
*queue
,
4388 Transaction
*set_backings_tran
, Error
**errp
)
4392 Error
*local_err
= NULL
;
4395 QDict
*orig_reopen_opts
;
4396 char *discard
= NULL
;
4398 bool drv_prepared
= false;
4400 assert(reopen_state
!= NULL
);
4401 assert(reopen_state
->bs
->drv
!= NULL
);
4402 drv
= reopen_state
->bs
->drv
;
4404 /* This function and each driver's bdrv_reopen_prepare() remove
4405 * entries from reopen_state->options as they are processed, so
4406 * we need to make a copy of the original QDict. */
4407 orig_reopen_opts
= qdict_clone_shallow(reopen_state
->options
);
4409 /* Process generic block layer options */
4410 opts
= qemu_opts_create(&bdrv_runtime_opts
, NULL
, 0, &error_abort
);
4411 if (!qemu_opts_absorb_qdict(opts
, reopen_state
->options
, errp
)) {
4416 /* This was already called in bdrv_reopen_queue_child() so the flags
4417 * are up-to-date. This time we simply want to remove the options from
4418 * QemuOpts in order to indicate that they have been processed. */
4419 old_flags
= reopen_state
->flags
;
4420 update_flags_from_options(&reopen_state
->flags
, opts
);
4421 assert(old_flags
== reopen_state
->flags
);
4423 discard
= qemu_opt_get_del(opts
, BDRV_OPT_DISCARD
);
4424 if (discard
!= NULL
) {
4425 if (bdrv_parse_discard_flags(discard
, &reopen_state
->flags
) != 0) {
4426 error_setg(errp
, "Invalid discard option");
4432 reopen_state
->detect_zeroes
=
4433 bdrv_parse_detect_zeroes(opts
, reopen_state
->flags
, &local_err
);
4435 error_propagate(errp
, local_err
);
4440 /* All other options (including node-name and driver) must be unchanged.
4441 * Put them back into the QDict, so that they are checked at the end
4442 * of this function. */
4443 qemu_opts_to_qdict(opts
, reopen_state
->options
);
4445 /* If we are to stay read-only, do not allow permission change
4446 * to r/w. Attempting to set to r/w may fail if either BDRV_O_ALLOW_RDWR is
4447 * not set, or if the BDS still has copy_on_read enabled */
4448 read_only
= !(reopen_state
->flags
& BDRV_O_RDWR
);
4449 ret
= bdrv_can_set_read_only(reopen_state
->bs
, read_only
, true, &local_err
);
4451 error_propagate(errp
, local_err
);
4455 if (drv
->bdrv_reopen_prepare
) {
4457 * If a driver-specific option is missing, it means that we
4458 * should reset it to its default value.
4459 * But not all options allow that, so we need to check it first.
4461 ret
= bdrv_reset_options_allowed(reopen_state
->bs
,
4462 reopen_state
->options
, errp
);
4467 ret
= drv
->bdrv_reopen_prepare(reopen_state
, queue
, &local_err
);
4469 if (local_err
!= NULL
) {
4470 error_propagate(errp
, local_err
);
4472 bdrv_refresh_filename(reopen_state
->bs
);
4473 error_setg(errp
, "failed while preparing to reopen image '%s'",
4474 reopen_state
->bs
->filename
);
4479 /* It is currently mandatory to have a bdrv_reopen_prepare()
4480 * handler for each supported drv. */
4481 error_setg(errp
, "Block format '%s' used by node '%s' "
4482 "does not support reopening files", drv
->format_name
,
4483 bdrv_get_device_or_node_name(reopen_state
->bs
));
4488 drv_prepared
= true;
4491 * We must provide the 'backing' option if the BDS has a backing
4492 * file or if the image file has a backing file name as part of
4493 * its metadata. Otherwise the 'backing' option can be omitted.
4495 if (drv
->supports_backing
&& reopen_state
->backing_missing
&&
4496 (reopen_state
->bs
->backing
|| reopen_state
->bs
->backing_file
[0])) {
4497 error_setg(errp
, "backing is missing for '%s'",
4498 reopen_state
->bs
->node_name
);
4504 * Allow changing the 'backing' option. The new value can be
4505 * either a reference to an existing node (using its node name)
4506 * or NULL to simply detach the current backing file.
4508 ret
= bdrv_reopen_parse_backing(reopen_state
, set_backings_tran
, errp
);
4512 qdict_del(reopen_state
->options
, "backing");
4514 /* Options that are not handled are only okay if they are unchanged
4515 * compared to the old state. It is expected that some options are only
4516 * used for the initial open, but not reopen (e.g. filename) */
4517 if (qdict_size(reopen_state
->options
)) {
4518 const QDictEntry
*entry
= qdict_first(reopen_state
->options
);
4521 QObject
*new = entry
->value
;
4522 QObject
*old
= qdict_get(reopen_state
->bs
->options
, entry
->key
);
4524 /* Allow child references (child_name=node_name) as long as they
4525 * point to the current child (i.e. everything stays the same). */
4526 if (qobject_type(new) == QTYPE_QSTRING
) {
4528 QLIST_FOREACH(child
, &reopen_state
->bs
->children
, next
) {
4529 if (!strcmp(child
->name
, entry
->key
)) {
4535 if (!strcmp(child
->bs
->node_name
,
4536 qstring_get_str(qobject_to(QString
, new)))) {
4537 continue; /* Found child with this name, skip option */
4543 * TODO: When using -drive to specify blockdev options, all values
4544 * will be strings; however, when using -blockdev, blockdev-add or
4545 * filenames using the json:{} pseudo-protocol, they will be
4547 * In contrast, reopening options are (currently) always strings
4548 * (because you can only specify them through qemu-io; all other
4549 * callers do not specify any options).
4550 * Therefore, when using anything other than -drive to create a BDS,
4551 * this cannot detect non-string options as unchanged, because
4552 * qobject_is_equal() always returns false for objects of different
4553 * type. In the future, this should be remedied by correctly typing
4554 * all options. For now, this is not too big of an issue because
4555 * the user can simply omit options which cannot be changed anyway,
4556 * so they will stay unchanged.
4558 if (!qobject_is_equal(new, old
)) {
4559 error_setg(errp
, "Cannot change the option '%s'", entry
->key
);
4563 } while ((entry
= qdict_next(reopen_state
->options
, entry
)));
4568 /* Restore the original reopen_state->options QDict */
4569 qobject_unref(reopen_state
->options
);
4570 reopen_state
->options
= qobject_ref(orig_reopen_opts
);
4573 if (ret
< 0 && drv_prepared
) {
4574 /* drv->bdrv_reopen_prepare() has succeeded, so we need to
4575 * call drv->bdrv_reopen_abort() before signaling an error
4576 * (bdrv_reopen_multiple() will not call bdrv_reopen_abort()
4577 * when the respective bdrv_reopen_prepare() has failed) */
4578 if (drv
->bdrv_reopen_abort
) {
4579 drv
->bdrv_reopen_abort(reopen_state
);
4582 qemu_opts_del(opts
);
4583 qobject_unref(orig_reopen_opts
);
4589 * Takes the staged changes for the reopen from bdrv_reopen_prepare(), and
4590 * makes them final by swapping the staging BlockDriverState contents into
4591 * the active BlockDriverState contents.
4593 static void bdrv_reopen_commit(BDRVReopenState
*reopen_state
)
4596 BlockDriverState
*bs
;
4599 assert(reopen_state
!= NULL
);
4600 bs
= reopen_state
->bs
;
4602 assert(drv
!= NULL
);
4604 /* If there are any driver level actions to take */
4605 if (drv
->bdrv_reopen_commit
) {
4606 drv
->bdrv_reopen_commit(reopen_state
);
4609 /* set BDS specific flags now */
4610 qobject_unref(bs
->explicit_options
);
4611 qobject_unref(bs
->options
);
4613 bs
->explicit_options
= reopen_state
->explicit_options
;
4614 bs
->options
= reopen_state
->options
;
4615 bs
->open_flags
= reopen_state
->flags
;
4616 bs
->detect_zeroes
= reopen_state
->detect_zeroes
;
4618 if (reopen_state
->replace_backing_bs
) {
4619 qdict_del(bs
->explicit_options
, "backing");
4620 qdict_del(bs
->options
, "backing");
4623 /* Remove child references from bs->options and bs->explicit_options.
4624 * Child options were already removed in bdrv_reopen_queue_child() */
4625 QLIST_FOREACH(child
, &bs
->children
, next
) {
4626 qdict_del(bs
->explicit_options
, child
->name
);
4627 qdict_del(bs
->options
, child
->name
);
4629 bdrv_refresh_limits(bs
, NULL
, NULL
);
4633 * Abort the reopen, and delete and free the staged changes in
4636 static void bdrv_reopen_abort(BDRVReopenState
*reopen_state
)
4640 assert(reopen_state
!= NULL
);
4641 drv
= reopen_state
->bs
->drv
;
4642 assert(drv
!= NULL
);
4644 if (drv
->bdrv_reopen_abort
) {
4645 drv
->bdrv_reopen_abort(reopen_state
);
4650 static void bdrv_close(BlockDriverState
*bs
)
4652 BdrvAioNotifier
*ban
, *ban_next
;
4653 BdrvChild
*child
, *next
;
4655 assert(!bs
->refcnt
);
4657 bdrv_drained_begin(bs
); /* complete I/O */
4659 bdrv_drain(bs
); /* in case flush left pending I/O */
4662 if (bs
->drv
->bdrv_close
) {
4663 /* Must unfreeze all children, so bdrv_unref_child() works */
4664 bs
->drv
->bdrv_close(bs
);
4669 QLIST_FOREACH_SAFE(child
, &bs
->children
, next
, next
) {
4670 bdrv_unref_child(bs
, child
);
4677 qatomic_set(&bs
->copy_on_read
, 0);
4678 bs
->backing_file
[0] = '\0';
4679 bs
->backing_format
[0] = '\0';
4680 bs
->total_sectors
= 0;
4681 bs
->encrypted
= false;
4683 qobject_unref(bs
->options
);
4684 qobject_unref(bs
->explicit_options
);
4686 bs
->explicit_options
= NULL
;
4687 qobject_unref(bs
->full_open_options
);
4688 bs
->full_open_options
= NULL
;
4690 bdrv_release_named_dirty_bitmaps(bs
);
4691 assert(QLIST_EMPTY(&bs
->dirty_bitmaps
));
4693 QLIST_FOREACH_SAFE(ban
, &bs
->aio_notifiers
, list
, ban_next
) {
4696 QLIST_INIT(&bs
->aio_notifiers
);
4697 bdrv_drained_end(bs
);
4700 * If we're still inside some bdrv_drain_all_begin()/end() sections, end
4701 * them now since this BDS won't exist anymore when bdrv_drain_all_end()
4704 if (bs
->quiesce_counter
) {
4705 bdrv_drain_all_end_quiesce(bs
);
4709 void bdrv_close_all(void)
4711 assert(job_next(NULL
) == NULL
);
4713 /* Drop references from requests still in flight, such as canceled block
4714 * jobs whose AIO context has not been polled yet */
4717 blk_remove_all_bs();
4718 blockdev_close_all_bdrv_states();
4720 assert(QTAILQ_EMPTY(&all_bdrv_states
));
4723 static bool should_update_child(BdrvChild
*c
, BlockDriverState
*to
)
4729 if (c
->klass
->stay_at_node
) {
4733 /* If the child @c belongs to the BDS @to, replacing the current
4734 * c->bs by @to would mean to create a loop.
4736 * Such a case occurs when appending a BDS to a backing chain.
4737 * For instance, imagine the following chain:
4739 * guest device -> node A -> further backing chain...
4741 * Now we create a new BDS B which we want to put on top of this
4742 * chain, so we first attach A as its backing node:
4747 * guest device -> node A -> further backing chain...
4749 * Finally we want to replace A by B. When doing that, we want to
4750 * replace all pointers to A by pointers to B -- except for the
4751 * pointer from B because (1) that would create a loop, and (2)
4752 * that pointer should simply stay intact:
4754 * guest device -> node B
4757 * node A -> further backing chain...
4759 * In general, when replacing a node A (c->bs) by a node B (@to),
4760 * if A is a child of B, that means we cannot replace A by B there
4761 * because that would create a loop. Silently detaching A from B
4762 * is also not really an option. So overall just leaving A in
4763 * place there is the most sensible choice.
4765 * We would also create a loop in any cases where @c is only
4766 * indirectly referenced by @to. Prevent this by returning false
4767 * if @c is found (by breadth-first search) anywhere in the whole
4772 found
= g_hash_table_new(NULL
, NULL
);
4773 g_hash_table_add(found
, to
);
4774 queue
= g_queue_new();
4775 g_queue_push_tail(queue
, to
);
4777 while (!g_queue_is_empty(queue
)) {
4778 BlockDriverState
*v
= g_queue_pop_head(queue
);
4781 QLIST_FOREACH(c2
, &v
->children
, next
) {
4787 if (g_hash_table_contains(found
, c2
->bs
)) {
4791 g_queue_push_tail(queue
, c2
->bs
);
4792 g_hash_table_add(found
, c2
->bs
);
4796 g_queue_free(queue
);
4797 g_hash_table_destroy(found
);
4802 typedef struct BdrvRemoveFilterOrCowChild
{
4805 } BdrvRemoveFilterOrCowChild
;
4807 static void bdrv_remove_filter_or_cow_child_abort(void *opaque
)
4809 BdrvRemoveFilterOrCowChild
*s
= opaque
;
4810 BlockDriverState
*parent_bs
= s
->child
->opaque
;
4812 QLIST_INSERT_HEAD(&parent_bs
->children
, s
->child
, next
);
4813 if (s
->is_backing
) {
4814 parent_bs
->backing
= s
->child
;
4816 parent_bs
->file
= s
->child
;
4820 * We don't have to restore child->bs here to undo bdrv_replace_child_tran()
4821 * because that function is transactionable and it registered own completion
4822 * entries in @tran, so .abort() for bdrv_replace_child_safe() will be
4823 * called automatically.
4827 static void bdrv_remove_filter_or_cow_child_commit(void *opaque
)
4829 BdrvRemoveFilterOrCowChild
*s
= opaque
;
4831 bdrv_child_free(s
->child
);
4834 static TransactionActionDrv bdrv_remove_filter_or_cow_child_drv
= {
4835 .abort
= bdrv_remove_filter_or_cow_child_abort
,
4836 .commit
= bdrv_remove_filter_or_cow_child_commit
,
4841 * A function to remove backing or file child of @bs.
4842 * Function doesn't update permissions, caller is responsible for this.
4844 static void bdrv_remove_file_or_backing_child(BlockDriverState
*bs
,
4848 BdrvRemoveFilterOrCowChild
*s
;
4850 assert(child
== bs
->backing
|| child
== bs
->file
);
4857 bdrv_replace_child_tran(child
, NULL
, tran
);
4860 s
= g_new(BdrvRemoveFilterOrCowChild
, 1);
4861 *s
= (BdrvRemoveFilterOrCowChild
) {
4863 .is_backing
= (child
== bs
->backing
),
4865 tran_add(tran
, &bdrv_remove_filter_or_cow_child_drv
, s
);
4867 QLIST_SAFE_REMOVE(child
, next
);
4868 if (s
->is_backing
) {
4876 * A function to remove backing-chain child of @bs if exists: cow child for
4877 * format nodes (always .backing) and filter child for filters (may be .file or
4880 static void bdrv_remove_filter_or_cow_child(BlockDriverState
*bs
,
4883 bdrv_remove_file_or_backing_child(bs
, bdrv_filter_or_cow_child(bs
), tran
);
4886 static int bdrv_replace_node_noperm(BlockDriverState
*from
,
4887 BlockDriverState
*to
,
4888 bool auto_skip
, Transaction
*tran
,
4891 BdrvChild
*c
, *next
;
4893 QLIST_FOREACH_SAFE(c
, &from
->parents
, next_parent
, next
) {
4894 assert(c
->bs
== from
);
4895 if (!should_update_child(c
, to
)) {
4899 error_setg(errp
, "Should not change '%s' link to '%s'",
4900 c
->name
, from
->node_name
);
4904 error_setg(errp
, "Cannot change '%s' link to '%s'",
4905 c
->name
, from
->node_name
);
4908 bdrv_replace_child_tran(c
, to
, tran
);
4915 * With auto_skip=true bdrv_replace_node_common skips updating from parents
4916 * if it creates a parent-child relation loop or if parent is block-job.
4918 * With auto_skip=false the error is returned if from has a parent which should
4921 * With @detach_subchain=true @to must be in a backing chain of @from. In this
4922 * case backing link of the cow-parent of @to is removed.
4924 static int bdrv_replace_node_common(BlockDriverState
*from
,
4925 BlockDriverState
*to
,
4926 bool auto_skip
, bool detach_subchain
,
4929 Transaction
*tran
= tran_new();
4930 g_autoptr(GHashTable
) found
= NULL
;
4931 g_autoptr(GSList
) refresh_list
= NULL
;
4932 BlockDriverState
*to_cow_parent
= NULL
;
4935 if (detach_subchain
) {
4936 assert(bdrv_chain_contains(from
, to
));
4938 for (to_cow_parent
= from
;
4939 bdrv_filter_or_cow_bs(to_cow_parent
) != to
;
4940 to_cow_parent
= bdrv_filter_or_cow_bs(to_cow_parent
))
4946 /* Make sure that @from doesn't go away until we have successfully attached
4947 * all of its parents to @to. */
4950 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
4951 assert(bdrv_get_aio_context(from
) == bdrv_get_aio_context(to
));
4952 bdrv_drained_begin(from
);
4955 * Do the replacement without permission update.
4956 * Replacement may influence the permissions, we should calculate new
4957 * permissions based on new graph. If we fail, we'll roll-back the
4960 ret
= bdrv_replace_node_noperm(from
, to
, auto_skip
, tran
, errp
);
4965 if (detach_subchain
) {
4966 bdrv_remove_filter_or_cow_child(to_cow_parent
, tran
);
4969 found
= g_hash_table_new(NULL
, NULL
);
4971 refresh_list
= bdrv_topological_dfs(refresh_list
, found
, to
);
4972 refresh_list
= bdrv_topological_dfs(refresh_list
, found
, from
);
4974 ret
= bdrv_list_refresh_perms(refresh_list
, NULL
, tran
, errp
);
4982 tran_finalize(tran
, ret
);
4984 bdrv_drained_end(from
);
4990 int bdrv_replace_node(BlockDriverState
*from
, BlockDriverState
*to
,
4993 return bdrv_replace_node_common(from
, to
, true, false, errp
);
4996 int bdrv_drop_filter(BlockDriverState
*bs
, Error
**errp
)
4998 return bdrv_replace_node_common(bs
, bdrv_filter_or_cow_bs(bs
), true, true,
5003 * Add new bs contents at the top of an image chain while the chain is
5004 * live, while keeping required fields on the top layer.
5006 * This will modify the BlockDriverState fields, and swap contents
5007 * between bs_new and bs_top. Both bs_new and bs_top are modified.
5009 * bs_new must not be attached to a BlockBackend and must not have backing
5012 * This function does not create any image files.
5014 int bdrv_append(BlockDriverState
*bs_new
, BlockDriverState
*bs_top
,
5018 Transaction
*tran
= tran_new();
5020 assert(!bs_new
->backing
);
5022 ret
= bdrv_attach_child_noperm(bs_new
, bs_top
, "backing",
5023 &child_of_bds
, bdrv_backing_role(bs_new
),
5024 &bs_new
->backing
, tran
, errp
);
5029 ret
= bdrv_replace_node_noperm(bs_top
, bs_new
, true, tran
, errp
);
5034 ret
= bdrv_refresh_perms(bs_new
, errp
);
5036 tran_finalize(tran
, ret
);
5038 bdrv_refresh_limits(bs_top
, NULL
, NULL
);
5043 static void bdrv_delete(BlockDriverState
*bs
)
5045 assert(bdrv_op_blocker_is_empty(bs
));
5046 assert(!bs
->refcnt
);
5048 /* remove from list, if necessary */
5049 if (bs
->node_name
[0] != '\0') {
5050 QTAILQ_REMOVE(&graph_bdrv_states
, bs
, node_list
);
5052 QTAILQ_REMOVE(&all_bdrv_states
, bs
, bs_list
);
5059 BlockDriverState
*bdrv_insert_node(BlockDriverState
*bs
, QDict
*node_options
,
5060 int flags
, Error
**errp
)
5062 BlockDriverState
*new_node_bs
;
5063 Error
*local_err
= NULL
;
5065 new_node_bs
= bdrv_open(NULL
, NULL
, node_options
, flags
, errp
);
5066 if (new_node_bs
== NULL
) {
5067 error_prepend(errp
, "Could not create node: ");
5071 bdrv_drained_begin(bs
);
5072 bdrv_replace_node(bs
, new_node_bs
, &local_err
);
5073 bdrv_drained_end(bs
);
5076 bdrv_unref(new_node_bs
);
5077 error_propagate(errp
, local_err
);
5085 * Run consistency checks on an image
5087 * Returns 0 if the check could be completed (it doesn't mean that the image is
5088 * free of errors) or -errno when an internal error occurred. The results of the
5089 * check are stored in res.
5091 int coroutine_fn
bdrv_co_check(BlockDriverState
*bs
,
5092 BdrvCheckResult
*res
, BdrvCheckMode fix
)
5094 if (bs
->drv
== NULL
) {
5097 if (bs
->drv
->bdrv_co_check
== NULL
) {
5101 memset(res
, 0, sizeof(*res
));
5102 return bs
->drv
->bdrv_co_check(bs
, res
, fix
);
5108 * -EINVAL - backing format specified, but no file
5109 * -ENOSPC - can't update the backing file because no space is left in the
5111 * -ENOTSUP - format driver doesn't support changing the backing file
5113 int bdrv_change_backing_file(BlockDriverState
*bs
, const char *backing_file
,
5114 const char *backing_fmt
, bool warn
)
5116 BlockDriver
*drv
= bs
->drv
;
5123 /* Backing file format doesn't make sense without a backing file */
5124 if (backing_fmt
&& !backing_file
) {
5128 if (warn
&& backing_file
&& !backing_fmt
) {
5129 warn_report("Deprecated use of backing file without explicit "
5130 "backing format, use of this image requires "
5131 "potentially unsafe format probing");
5134 if (drv
->bdrv_change_backing_file
!= NULL
) {
5135 ret
= drv
->bdrv_change_backing_file(bs
, backing_file
, backing_fmt
);
5141 pstrcpy(bs
->backing_file
, sizeof(bs
->backing_file
), backing_file
?: "");
5142 pstrcpy(bs
->backing_format
, sizeof(bs
->backing_format
), backing_fmt
?: "");
5143 pstrcpy(bs
->auto_backing_file
, sizeof(bs
->auto_backing_file
),
5144 backing_file
?: "");
5150 * Finds the first non-filter node above bs in the chain between
5151 * active and bs. The returned node is either an immediate parent of
5152 * bs, or there are only filter nodes between the two.
5154 * Returns NULL if bs is not found in active's image chain,
5155 * or if active == bs.
5157 * Returns the bottommost base image if bs == NULL.
5159 BlockDriverState
*bdrv_find_overlay(BlockDriverState
*active
,
5160 BlockDriverState
*bs
)
5162 bs
= bdrv_skip_filters(bs
);
5163 active
= bdrv_skip_filters(active
);
5166 BlockDriverState
*next
= bdrv_backing_chain_next(active
);
5176 /* Given a BDS, searches for the base layer. */
5177 BlockDriverState
*bdrv_find_base(BlockDriverState
*bs
)
5179 return bdrv_find_overlay(bs
, NULL
);
5183 * Return true if at least one of the COW (backing) and filter links
5184 * between @bs and @base is frozen. @errp is set if that's the case.
5185 * @base must be reachable from @bs, or NULL.
5187 bool bdrv_is_backing_chain_frozen(BlockDriverState
*bs
, BlockDriverState
*base
,
5190 BlockDriverState
*i
;
5193 for (i
= bs
; i
!= base
; i
= child_bs(child
)) {
5194 child
= bdrv_filter_or_cow_child(i
);
5196 if (child
&& child
->frozen
) {
5197 error_setg(errp
, "Cannot change '%s' link from '%s' to '%s'",
5198 child
->name
, i
->node_name
, child
->bs
->node_name
);
5207 * Freeze all COW (backing) and filter links between @bs and @base.
5208 * If any of the links is already frozen the operation is aborted and
5209 * none of the links are modified.
5210 * @base must be reachable from @bs, or NULL.
5211 * Returns 0 on success. On failure returns < 0 and sets @errp.
5213 int bdrv_freeze_backing_chain(BlockDriverState
*bs
, BlockDriverState
*base
,
5216 BlockDriverState
*i
;
5219 if (bdrv_is_backing_chain_frozen(bs
, base
, errp
)) {
5223 for (i
= bs
; i
!= base
; i
= child_bs(child
)) {
5224 child
= bdrv_filter_or_cow_child(i
);
5225 if (child
&& child
->bs
->never_freeze
) {
5226 error_setg(errp
, "Cannot freeze '%s' link to '%s'",
5227 child
->name
, child
->bs
->node_name
);
5232 for (i
= bs
; i
!= base
; i
= child_bs(child
)) {
5233 child
= bdrv_filter_or_cow_child(i
);
5235 child
->frozen
= true;
5243 * Unfreeze all COW (backing) and filter links between @bs and @base.
5244 * The caller must ensure that all links are frozen before using this
5246 * @base must be reachable from @bs, or NULL.
5248 void bdrv_unfreeze_backing_chain(BlockDriverState
*bs
, BlockDriverState
*base
)
5250 BlockDriverState
*i
;
5253 for (i
= bs
; i
!= base
; i
= child_bs(child
)) {
5254 child
= bdrv_filter_or_cow_child(i
);
5256 assert(child
->frozen
);
5257 child
->frozen
= false;
5263 * Drops images above 'base' up to and including 'top', and sets the image
5264 * above 'top' to have base as its backing file.
5266 * Requires that the overlay to 'top' is opened r/w, so that the backing file
5267 * information in 'bs' can be properly updated.
5269 * E.g., this will convert the following chain:
5270 * bottom <- base <- intermediate <- top <- active
5274 * bottom <- base <- active
5276 * It is allowed for bottom==base, in which case it converts:
5278 * base <- intermediate <- top <- active
5284 * If backing_file_str is non-NULL, it will be used when modifying top's
5285 * overlay image metadata.
5288 * if active == top, that is considered an error
5291 int bdrv_drop_intermediate(BlockDriverState
*top
, BlockDriverState
*base
,
5292 const char *backing_file_str
)
5294 BlockDriverState
*explicit_top
= top
;
5295 bool update_inherits_from
;
5297 Error
*local_err
= NULL
;
5299 g_autoptr(GSList
) updated_children
= NULL
;
5303 bdrv_subtree_drained_begin(top
);
5305 if (!top
->drv
|| !base
->drv
) {
5309 /* Make sure that base is in the backing chain of top */
5310 if (!bdrv_chain_contains(top
, base
)) {
5314 /* If 'base' recursively inherits from 'top' then we should set
5315 * base->inherits_from to top->inherits_from after 'top' and all
5316 * other intermediate nodes have been dropped.
5317 * If 'top' is an implicit node (e.g. "commit_top") we should skip
5318 * it because no one inherits from it. We use explicit_top for that. */
5319 explicit_top
= bdrv_skip_implicit_filters(explicit_top
);
5320 update_inherits_from
= bdrv_inherits_from_recursive(base
, explicit_top
);
5322 /* success - we can delete the intermediate states, and link top->base */
5323 /* TODO Check graph modification op blockers (BLK_PERM_GRAPH_MOD) once
5324 * we've figured out how they should work. */
5325 if (!backing_file_str
) {
5326 bdrv_refresh_filename(base
);
5327 backing_file_str
= base
->filename
;
5330 QLIST_FOREACH(c
, &top
->parents
, next_parent
) {
5331 updated_children
= g_slist_prepend(updated_children
, c
);
5335 * It seems correct to pass detach_subchain=true here, but it triggers
5336 * one more yet not fixed bug, when due to nested aio_poll loop we switch to
5337 * another drained section, which modify the graph (for example, removing
5338 * the child, which we keep in updated_children list). So, it's a TODO.
5340 * Note, bug triggered if pass detach_subchain=true here and run
5341 * test-bdrv-drain. test_drop_intermediate_poll() test-case will crash.
5344 bdrv_replace_node_common(top
, base
, false, false, &local_err
);
5346 error_report_err(local_err
);
5350 for (p
= updated_children
; p
; p
= p
->next
) {
5353 if (c
->klass
->update_filename
) {
5354 ret
= c
->klass
->update_filename(c
, base
, backing_file_str
,
5358 * TODO: Actually, we want to rollback all previous iterations
5359 * of this loop, and (which is almost impossible) previous
5360 * bdrv_replace_node()...
5362 * Note, that c->klass->update_filename may lead to permission
5363 * update, so it's a bad idea to call it inside permission
5364 * update transaction of bdrv_replace_node.
5366 error_report_err(local_err
);
5372 if (update_inherits_from
) {
5373 base
->inherits_from
= explicit_top
->inherits_from
;
5378 bdrv_subtree_drained_end(top
);
5384 * Implementation of BlockDriver.bdrv_get_allocated_file_size() that
5385 * sums the size of all data-bearing children. (This excludes backing
5388 static int64_t bdrv_sum_allocated_file_size(BlockDriverState
*bs
)
5391 int64_t child_size
, sum
= 0;
5393 QLIST_FOREACH(child
, &bs
->children
, next
) {
5394 if (child
->role
& (BDRV_CHILD_DATA
| BDRV_CHILD_METADATA
|
5395 BDRV_CHILD_FILTERED
))
5397 child_size
= bdrv_get_allocated_file_size(child
->bs
);
5398 if (child_size
< 0) {
5409 * Length of a allocated file in bytes. Sparse files are counted by actual
5410 * allocated space. Return < 0 if error or unknown.
5412 int64_t bdrv_get_allocated_file_size(BlockDriverState
*bs
)
5414 BlockDriver
*drv
= bs
->drv
;
5418 if (drv
->bdrv_get_allocated_file_size
) {
5419 return drv
->bdrv_get_allocated_file_size(bs
);
5422 if (drv
->bdrv_file_open
) {
5424 * Protocol drivers default to -ENOTSUP (most of their data is
5425 * not stored in any of their children (if they even have any),
5426 * so there is no generic way to figure it out).
5429 } else if (drv
->is_filter
) {
5430 /* Filter drivers default to the size of their filtered child */
5431 return bdrv_get_allocated_file_size(bdrv_filter_bs(bs
));
5433 /* Other drivers default to summing their children's sizes */
5434 return bdrv_sum_allocated_file_size(bs
);
5440 * @drv: Format driver
5441 * @opts: Creation options for new image
5442 * @in_bs: Existing image containing data for new image (may be NULL)
5443 * @errp: Error object
5444 * Returns: A #BlockMeasureInfo (free using qapi_free_BlockMeasureInfo())
5447 * Calculate file size required to create a new image.
5449 * If @in_bs is given then space for allocated clusters and zero clusters
5450 * from that image are included in the calculation. If @opts contains a
5451 * backing file that is shared by @in_bs then backing clusters may be omitted
5452 * from the calculation.
5454 * If @in_bs is NULL then the calculation includes no allocated clusters
5455 * unless a preallocation option is given in @opts.
5457 * Note that @in_bs may use a different BlockDriver from @drv.
5459 * If an error occurs the @errp pointer is set.
5461 BlockMeasureInfo
*bdrv_measure(BlockDriver
*drv
, QemuOpts
*opts
,
5462 BlockDriverState
*in_bs
, Error
**errp
)
5464 if (!drv
->bdrv_measure
) {
5465 error_setg(errp
, "Block driver '%s' does not support size measurement",
5470 return drv
->bdrv_measure(opts
, in_bs
, errp
);
5474 * Return number of sectors on success, -errno on error.
5476 int64_t bdrv_nb_sectors(BlockDriverState
*bs
)
5478 BlockDriver
*drv
= bs
->drv
;
5483 if (drv
->has_variable_length
) {
5484 int ret
= refresh_total_sectors(bs
, bs
->total_sectors
);
5489 return bs
->total_sectors
;
5493 * Return length in bytes on success, -errno on error.
5494 * The length is always a multiple of BDRV_SECTOR_SIZE.
5496 int64_t bdrv_getlength(BlockDriverState
*bs
)
5498 int64_t ret
= bdrv_nb_sectors(bs
);
5503 if (ret
> INT64_MAX
/ BDRV_SECTOR_SIZE
) {
5506 return ret
* BDRV_SECTOR_SIZE
;
5509 /* return 0 as number of sectors if no device present or error */
5510 void bdrv_get_geometry(BlockDriverState
*bs
, uint64_t *nb_sectors_ptr
)
5512 int64_t nb_sectors
= bdrv_nb_sectors(bs
);
5514 *nb_sectors_ptr
= nb_sectors
< 0 ? 0 : nb_sectors
;
5517 bool bdrv_is_sg(BlockDriverState
*bs
)
5523 * Return whether the given node supports compressed writes.
5525 bool bdrv_supports_compressed_writes(BlockDriverState
*bs
)
5527 BlockDriverState
*filtered
;
5529 if (!bs
->drv
|| !block_driver_can_compress(bs
->drv
)) {
5533 filtered
= bdrv_filter_bs(bs
);
5536 * Filters can only forward compressed writes, so we have to
5539 return bdrv_supports_compressed_writes(filtered
);
5545 const char *bdrv_get_format_name(BlockDriverState
*bs
)
5547 return bs
->drv
? bs
->drv
->format_name
: NULL
;
5550 static int qsort_strcmp(const void *a
, const void *b
)
5552 return strcmp(*(char *const *)a
, *(char *const *)b
);
5555 void bdrv_iterate_format(void (*it
)(void *opaque
, const char *name
),
5556 void *opaque
, bool read_only
)
5561 const char **formats
= NULL
;
5563 QLIST_FOREACH(drv
, &bdrv_drivers
, list
) {
5564 if (drv
->format_name
) {
5568 if (use_bdrv_whitelist
&& !bdrv_is_whitelisted(drv
, read_only
)) {
5572 while (formats
&& i
&& !found
) {
5573 found
= !strcmp(formats
[--i
], drv
->format_name
);
5577 formats
= g_renew(const char *, formats
, count
+ 1);
5578 formats
[count
++] = drv
->format_name
;
5583 for (i
= 0; i
< (int)ARRAY_SIZE(block_driver_modules
); i
++) {
5584 const char *format_name
= block_driver_modules
[i
].format_name
;
5590 if (use_bdrv_whitelist
&&
5591 !bdrv_format_is_whitelisted(format_name
, read_only
)) {
5595 while (formats
&& j
&& !found
) {
5596 found
= !strcmp(formats
[--j
], format_name
);
5600 formats
= g_renew(const char *, formats
, count
+ 1);
5601 formats
[count
++] = format_name
;
5606 qsort(formats
, count
, sizeof(formats
[0]), qsort_strcmp
);
5608 for (i
= 0; i
< count
; i
++) {
5609 it(opaque
, formats
[i
]);
5615 /* This function is to find a node in the bs graph */
5616 BlockDriverState
*bdrv_find_node(const char *node_name
)
5618 BlockDriverState
*bs
;
5622 QTAILQ_FOREACH(bs
, &graph_bdrv_states
, node_list
) {
5623 if (!strcmp(node_name
, bs
->node_name
)) {
5630 /* Put this QMP function here so it can access the static graph_bdrv_states. */
5631 BlockDeviceInfoList
*bdrv_named_nodes_list(bool flat
,
5634 BlockDeviceInfoList
*list
;
5635 BlockDriverState
*bs
;
5638 QTAILQ_FOREACH(bs
, &graph_bdrv_states
, node_list
) {
5639 BlockDeviceInfo
*info
= bdrv_block_device_info(NULL
, bs
, flat
, errp
);
5641 qapi_free_BlockDeviceInfoList(list
);
5644 QAPI_LIST_PREPEND(list
, info
);
5650 typedef struct XDbgBlockGraphConstructor
{
5651 XDbgBlockGraph
*graph
;
5652 GHashTable
*graph_nodes
;
5653 } XDbgBlockGraphConstructor
;
5655 static XDbgBlockGraphConstructor
*xdbg_graph_new(void)
5657 XDbgBlockGraphConstructor
*gr
= g_new(XDbgBlockGraphConstructor
, 1);
5659 gr
->graph
= g_new0(XDbgBlockGraph
, 1);
5660 gr
->graph_nodes
= g_hash_table_new(NULL
, NULL
);
5665 static XDbgBlockGraph
*xdbg_graph_finalize(XDbgBlockGraphConstructor
*gr
)
5667 XDbgBlockGraph
*graph
= gr
->graph
;
5669 g_hash_table_destroy(gr
->graph_nodes
);
5675 static uintptr_t xdbg_graph_node_num(XDbgBlockGraphConstructor
*gr
, void *node
)
5677 uintptr_t ret
= (uintptr_t)g_hash_table_lookup(gr
->graph_nodes
, node
);
5684 * Start counting from 1, not 0, because 0 interferes with not-found (NULL)
5685 * answer of g_hash_table_lookup.
5687 ret
= g_hash_table_size(gr
->graph_nodes
) + 1;
5688 g_hash_table_insert(gr
->graph_nodes
, node
, (void *)ret
);
5693 static void xdbg_graph_add_node(XDbgBlockGraphConstructor
*gr
, void *node
,
5694 XDbgBlockGraphNodeType type
, const char *name
)
5696 XDbgBlockGraphNode
*n
;
5698 n
= g_new0(XDbgBlockGraphNode
, 1);
5700 n
->id
= xdbg_graph_node_num(gr
, node
);
5702 n
->name
= g_strdup(name
);
5704 QAPI_LIST_PREPEND(gr
->graph
->nodes
, n
);
5707 static void xdbg_graph_add_edge(XDbgBlockGraphConstructor
*gr
, void *parent
,
5708 const BdrvChild
*child
)
5710 BlockPermission qapi_perm
;
5711 XDbgBlockGraphEdge
*edge
;
5713 edge
= g_new0(XDbgBlockGraphEdge
, 1);
5715 edge
->parent
= xdbg_graph_node_num(gr
, parent
);
5716 edge
->child
= xdbg_graph_node_num(gr
, child
->bs
);
5717 edge
->name
= g_strdup(child
->name
);
5719 for (qapi_perm
= 0; qapi_perm
< BLOCK_PERMISSION__MAX
; qapi_perm
++) {
5720 uint64_t flag
= bdrv_qapi_perm_to_blk_perm(qapi_perm
);
5722 if (flag
& child
->perm
) {
5723 QAPI_LIST_PREPEND(edge
->perm
, qapi_perm
);
5725 if (flag
& child
->shared_perm
) {
5726 QAPI_LIST_PREPEND(edge
->shared_perm
, qapi_perm
);
5730 QAPI_LIST_PREPEND(gr
->graph
->edges
, edge
);
5734 XDbgBlockGraph
*bdrv_get_xdbg_block_graph(Error
**errp
)
5738 BlockDriverState
*bs
;
5740 XDbgBlockGraphConstructor
*gr
= xdbg_graph_new();
5742 for (blk
= blk_all_next(NULL
); blk
; blk
= blk_all_next(blk
)) {
5743 char *allocated_name
= NULL
;
5744 const char *name
= blk_name(blk
);
5747 name
= allocated_name
= blk_get_attached_dev_id(blk
);
5749 xdbg_graph_add_node(gr
, blk
, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_BACKEND
,
5751 g_free(allocated_name
);
5752 if (blk_root(blk
)) {
5753 xdbg_graph_add_edge(gr
, blk
, blk_root(blk
));
5757 for (job
= block_job_next(NULL
); job
; job
= block_job_next(job
)) {
5760 xdbg_graph_add_node(gr
, job
, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_JOB
,
5762 for (el
= job
->nodes
; el
; el
= el
->next
) {
5763 xdbg_graph_add_edge(gr
, job
, (BdrvChild
*)el
->data
);
5767 QTAILQ_FOREACH(bs
, &graph_bdrv_states
, node_list
) {
5768 xdbg_graph_add_node(gr
, bs
, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_DRIVER
,
5770 QLIST_FOREACH(child
, &bs
->children
, next
) {
5771 xdbg_graph_add_edge(gr
, bs
, child
);
5775 return xdbg_graph_finalize(gr
);
5778 BlockDriverState
*bdrv_lookup_bs(const char *device
,
5779 const char *node_name
,
5783 BlockDriverState
*bs
;
5786 blk
= blk_by_name(device
);
5791 error_setg(errp
, "Device '%s' has no medium", device
);
5799 bs
= bdrv_find_node(node_name
);
5806 error_setg(errp
, "Cannot find device=\'%s\' nor node-name=\'%s\'",
5807 device
? device
: "",
5808 node_name
? node_name
: "");
5812 /* If 'base' is in the same chain as 'top', return true. Otherwise,
5813 * return false. If either argument is NULL, return false. */
5814 bool bdrv_chain_contains(BlockDriverState
*top
, BlockDriverState
*base
)
5816 while (top
&& top
!= base
) {
5817 top
= bdrv_filter_or_cow_bs(top
);
5823 BlockDriverState
*bdrv_next_node(BlockDriverState
*bs
)
5826 return QTAILQ_FIRST(&graph_bdrv_states
);
5828 return QTAILQ_NEXT(bs
, node_list
);
5831 BlockDriverState
*bdrv_next_all_states(BlockDriverState
*bs
)
5834 return QTAILQ_FIRST(&all_bdrv_states
);
5836 return QTAILQ_NEXT(bs
, bs_list
);
5839 const char *bdrv_get_node_name(const BlockDriverState
*bs
)
5841 return bs
->node_name
;
5844 const char *bdrv_get_parent_name(const BlockDriverState
*bs
)
5849 /* If multiple parents have a name, just pick the first one. */
5850 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
5851 if (c
->klass
->get_name
) {
5852 name
= c
->klass
->get_name(c
);
5853 if (name
&& *name
) {
5862 /* TODO check what callers really want: bs->node_name or blk_name() */
5863 const char *bdrv_get_device_name(const BlockDriverState
*bs
)
5865 return bdrv_get_parent_name(bs
) ?: "";
5868 /* This can be used to identify nodes that might not have a device
5869 * name associated. Since node and device names live in the same
5870 * namespace, the result is unambiguous. The exception is if both are
5871 * absent, then this returns an empty (non-null) string. */
5872 const char *bdrv_get_device_or_node_name(const BlockDriverState
*bs
)
5874 return bdrv_get_parent_name(bs
) ?: bs
->node_name
;
5877 int bdrv_get_flags(BlockDriverState
*bs
)
5879 return bs
->open_flags
;
5882 int bdrv_has_zero_init_1(BlockDriverState
*bs
)
5887 int bdrv_has_zero_init(BlockDriverState
*bs
)
5889 BlockDriverState
*filtered
;
5895 /* If BS is a copy on write image, it is initialized to
5896 the contents of the base image, which may not be zeroes. */
5897 if (bdrv_cow_child(bs
)) {
5900 if (bs
->drv
->bdrv_has_zero_init
) {
5901 return bs
->drv
->bdrv_has_zero_init(bs
);
5904 filtered
= bdrv_filter_bs(bs
);
5906 return bdrv_has_zero_init(filtered
);
5913 bool bdrv_can_write_zeroes_with_unmap(BlockDriverState
*bs
)
5915 if (!(bs
->open_flags
& BDRV_O_UNMAP
)) {
5919 return bs
->supported_zero_flags
& BDRV_REQ_MAY_UNMAP
;
5922 void bdrv_get_backing_filename(BlockDriverState
*bs
,
5923 char *filename
, int filename_size
)
5925 pstrcpy(filename
, filename_size
, bs
->backing_file
);
5928 int bdrv_get_info(BlockDriverState
*bs
, BlockDriverInfo
*bdi
)
5931 BlockDriver
*drv
= bs
->drv
;
5932 /* if bs->drv == NULL, bs is closed, so there's nothing to do here */
5936 if (!drv
->bdrv_get_info
) {
5937 BlockDriverState
*filtered
= bdrv_filter_bs(bs
);
5939 return bdrv_get_info(filtered
, bdi
);
5943 memset(bdi
, 0, sizeof(*bdi
));
5944 ret
= drv
->bdrv_get_info(bs
, bdi
);
5949 if (bdi
->cluster_size
> BDRV_MAX_ALIGNMENT
) {
5956 ImageInfoSpecific
*bdrv_get_specific_info(BlockDriverState
*bs
,
5959 BlockDriver
*drv
= bs
->drv
;
5960 if (drv
&& drv
->bdrv_get_specific_info
) {
5961 return drv
->bdrv_get_specific_info(bs
, errp
);
5966 BlockStatsSpecific
*bdrv_get_specific_stats(BlockDriverState
*bs
)
5968 BlockDriver
*drv
= bs
->drv
;
5969 if (!drv
|| !drv
->bdrv_get_specific_stats
) {
5972 return drv
->bdrv_get_specific_stats(bs
);
5975 void bdrv_debug_event(BlockDriverState
*bs
, BlkdebugEvent event
)
5977 if (!bs
|| !bs
->drv
|| !bs
->drv
->bdrv_debug_event
) {
5981 bs
->drv
->bdrv_debug_event(bs
, event
);
5984 static BlockDriverState
*bdrv_find_debug_node(BlockDriverState
*bs
)
5986 while (bs
&& bs
->drv
&& !bs
->drv
->bdrv_debug_breakpoint
) {
5987 bs
= bdrv_primary_bs(bs
);
5990 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_breakpoint
) {
5991 assert(bs
->drv
->bdrv_debug_remove_breakpoint
);
5998 int bdrv_debug_breakpoint(BlockDriverState
*bs
, const char *event
,
6001 bs
= bdrv_find_debug_node(bs
);
6003 return bs
->drv
->bdrv_debug_breakpoint(bs
, event
, tag
);
6009 int bdrv_debug_remove_breakpoint(BlockDriverState
*bs
, const char *tag
)
6011 bs
= bdrv_find_debug_node(bs
);
6013 return bs
->drv
->bdrv_debug_remove_breakpoint(bs
, tag
);
6019 int bdrv_debug_resume(BlockDriverState
*bs
, const char *tag
)
6021 while (bs
&& (!bs
->drv
|| !bs
->drv
->bdrv_debug_resume
)) {
6022 bs
= bdrv_primary_bs(bs
);
6025 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_resume
) {
6026 return bs
->drv
->bdrv_debug_resume(bs
, tag
);
6032 bool bdrv_debug_is_suspended(BlockDriverState
*bs
, const char *tag
)
6034 while (bs
&& bs
->drv
&& !bs
->drv
->bdrv_debug_is_suspended
) {
6035 bs
= bdrv_primary_bs(bs
);
6038 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_is_suspended
) {
6039 return bs
->drv
->bdrv_debug_is_suspended(bs
, tag
);
6045 /* backing_file can either be relative, or absolute, or a protocol. If it is
6046 * relative, it must be relative to the chain. So, passing in bs->filename
6047 * from a BDS as backing_file should not be done, as that may be relative to
6048 * the CWD rather than the chain. */
6049 BlockDriverState
*bdrv_find_backing_image(BlockDriverState
*bs
,
6050 const char *backing_file
)
6052 char *filename_full
= NULL
;
6053 char *backing_file_full
= NULL
;
6054 char *filename_tmp
= NULL
;
6055 int is_protocol
= 0;
6056 bool filenames_refreshed
= false;
6057 BlockDriverState
*curr_bs
= NULL
;
6058 BlockDriverState
*retval
= NULL
;
6059 BlockDriverState
*bs_below
;
6061 if (!bs
|| !bs
->drv
|| !backing_file
) {
6065 filename_full
= g_malloc(PATH_MAX
);
6066 backing_file_full
= g_malloc(PATH_MAX
);
6068 is_protocol
= path_has_protocol(backing_file
);
6071 * Being largely a legacy function, skip any filters here
6072 * (because filters do not have normal filenames, so they cannot
6073 * match anyway; and allowing json:{} filenames is a bit out of
6076 for (curr_bs
= bdrv_skip_filters(bs
);
6077 bdrv_cow_child(curr_bs
) != NULL
;
6080 bs_below
= bdrv_backing_chain_next(curr_bs
);
6082 if (bdrv_backing_overridden(curr_bs
)) {
6084 * If the backing file was overridden, we can only compare
6085 * directly against the backing node's filename.
6088 if (!filenames_refreshed
) {
6090 * This will automatically refresh all of the
6091 * filenames in the rest of the backing chain, so we
6092 * only need to do this once.
6094 bdrv_refresh_filename(bs_below
);
6095 filenames_refreshed
= true;
6098 if (strcmp(backing_file
, bs_below
->filename
) == 0) {
6102 } else if (is_protocol
|| path_has_protocol(curr_bs
->backing_file
)) {
6104 * If either of the filename paths is actually a protocol, then
6105 * compare unmodified paths; otherwise make paths relative.
6107 char *backing_file_full_ret
;
6109 if (strcmp(backing_file
, curr_bs
->backing_file
) == 0) {
6113 /* Also check against the full backing filename for the image */
6114 backing_file_full_ret
= bdrv_get_full_backing_filename(curr_bs
,
6116 if (backing_file_full_ret
) {
6117 bool equal
= strcmp(backing_file
, backing_file_full_ret
) == 0;
6118 g_free(backing_file_full_ret
);
6125 /* If not an absolute filename path, make it relative to the current
6126 * image's filename path */
6127 filename_tmp
= bdrv_make_absolute_filename(curr_bs
, backing_file
,
6129 /* We are going to compare canonicalized absolute pathnames */
6130 if (!filename_tmp
|| !realpath(filename_tmp
, filename_full
)) {
6131 g_free(filename_tmp
);
6134 g_free(filename_tmp
);
6136 /* We need to make sure the backing filename we are comparing against
6137 * is relative to the current image filename (or absolute) */
6138 filename_tmp
= bdrv_get_full_backing_filename(curr_bs
, NULL
);
6139 if (!filename_tmp
|| !realpath(filename_tmp
, backing_file_full
)) {
6140 g_free(filename_tmp
);
6143 g_free(filename_tmp
);
6145 if (strcmp(backing_file_full
, filename_full
) == 0) {
6152 g_free(filename_full
);
6153 g_free(backing_file_full
);
6157 void bdrv_init(void)
6159 module_call_init(MODULE_INIT_BLOCK
);
6162 void bdrv_init_with_whitelist(void)
6164 use_bdrv_whitelist
= 1;
6168 int coroutine_fn
bdrv_co_invalidate_cache(BlockDriverState
*bs
, Error
**errp
)
6170 BdrvChild
*child
, *parent
;
6171 Error
*local_err
= NULL
;
6173 BdrvDirtyBitmap
*bm
;
6179 QLIST_FOREACH(child
, &bs
->children
, next
) {
6180 bdrv_co_invalidate_cache(child
->bs
, &local_err
);
6182 error_propagate(errp
, local_err
);
6188 * Update permissions, they may differ for inactive nodes.
6190 * Note that the required permissions of inactive images are always a
6191 * subset of the permissions required after activating the image. This
6192 * allows us to just get the permissions upfront without restricting
6193 * drv->bdrv_invalidate_cache().
6195 * It also means that in error cases, we don't have to try and revert to
6196 * the old permissions (which is an operation that could fail, too). We can
6197 * just keep the extended permissions for the next time that an activation
6198 * of the image is tried.
6200 if (bs
->open_flags
& BDRV_O_INACTIVE
) {
6201 bs
->open_flags
&= ~BDRV_O_INACTIVE
;
6202 ret
= bdrv_refresh_perms(bs
, errp
);
6204 bs
->open_flags
|= BDRV_O_INACTIVE
;
6208 if (bs
->drv
->bdrv_co_invalidate_cache
) {
6209 bs
->drv
->bdrv_co_invalidate_cache(bs
, &local_err
);
6211 bs
->open_flags
|= BDRV_O_INACTIVE
;
6212 error_propagate(errp
, local_err
);
6217 FOR_EACH_DIRTY_BITMAP(bs
, bm
) {
6218 bdrv_dirty_bitmap_skip_store(bm
, false);
6221 ret
= refresh_total_sectors(bs
, bs
->total_sectors
);
6223 bs
->open_flags
|= BDRV_O_INACTIVE
;
6224 error_setg_errno(errp
, -ret
, "Could not refresh total sector count");
6229 QLIST_FOREACH(parent
, &bs
->parents
, next_parent
) {
6230 if (parent
->klass
->activate
) {
6231 parent
->klass
->activate(parent
, &local_err
);
6233 bs
->open_flags
|= BDRV_O_INACTIVE
;
6234 error_propagate(errp
, local_err
);
6243 void bdrv_invalidate_cache_all(Error
**errp
)
6245 BlockDriverState
*bs
;
6246 BdrvNextIterator it
;
6248 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
6249 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
6252 aio_context_acquire(aio_context
);
6253 ret
= bdrv_invalidate_cache(bs
, errp
);
6254 aio_context_release(aio_context
);
6256 bdrv_next_cleanup(&it
);
6262 static bool bdrv_has_bds_parent(BlockDriverState
*bs
, bool only_active
)
6266 QLIST_FOREACH(parent
, &bs
->parents
, next_parent
) {
6267 if (parent
->klass
->parent_is_bds
) {
6268 BlockDriverState
*parent_bs
= parent
->opaque
;
6269 if (!only_active
|| !(parent_bs
->open_flags
& BDRV_O_INACTIVE
)) {
6278 static int bdrv_inactivate_recurse(BlockDriverState
*bs
)
6280 BdrvChild
*child
, *parent
;
6287 /* Make sure that we don't inactivate a child before its parent.
6288 * It will be covered by recursion from the yet active parent. */
6289 if (bdrv_has_bds_parent(bs
, true)) {
6293 assert(!(bs
->open_flags
& BDRV_O_INACTIVE
));
6295 /* Inactivate this node */
6296 if (bs
->drv
->bdrv_inactivate
) {
6297 ret
= bs
->drv
->bdrv_inactivate(bs
);
6303 QLIST_FOREACH(parent
, &bs
->parents
, next_parent
) {
6304 if (parent
->klass
->inactivate
) {
6305 ret
= parent
->klass
->inactivate(parent
);
6312 bs
->open_flags
|= BDRV_O_INACTIVE
;
6315 * Update permissions, they may differ for inactive nodes.
6316 * We only tried to loosen restrictions, so errors are not fatal, ignore
6319 bdrv_refresh_perms(bs
, NULL
);
6321 /* Recursively inactivate children */
6322 QLIST_FOREACH(child
, &bs
->children
, next
) {
6323 ret
= bdrv_inactivate_recurse(child
->bs
);
6332 int bdrv_inactivate_all(void)
6334 BlockDriverState
*bs
= NULL
;
6335 BdrvNextIterator it
;
6337 GSList
*aio_ctxs
= NULL
, *ctx
;
6339 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
6340 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
6342 if (!g_slist_find(aio_ctxs
, aio_context
)) {
6343 aio_ctxs
= g_slist_prepend(aio_ctxs
, aio_context
);
6344 aio_context_acquire(aio_context
);
6348 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
6349 /* Nodes with BDS parents are covered by recursion from the last
6350 * parent that gets inactivated. Don't inactivate them a second
6351 * time if that has already happened. */
6352 if (bdrv_has_bds_parent(bs
, false)) {
6355 ret
= bdrv_inactivate_recurse(bs
);
6357 bdrv_next_cleanup(&it
);
6363 for (ctx
= aio_ctxs
; ctx
!= NULL
; ctx
= ctx
->next
) {
6364 AioContext
*aio_context
= ctx
->data
;
6365 aio_context_release(aio_context
);
6367 g_slist_free(aio_ctxs
);
6372 /**************************************************************/
6373 /* removable device support */
6376 * Return TRUE if the media is present
6378 bool bdrv_is_inserted(BlockDriverState
*bs
)
6380 BlockDriver
*drv
= bs
->drv
;
6386 if (drv
->bdrv_is_inserted
) {
6387 return drv
->bdrv_is_inserted(bs
);
6389 QLIST_FOREACH(child
, &bs
->children
, next
) {
6390 if (!bdrv_is_inserted(child
->bs
)) {
6398 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
6400 void bdrv_eject(BlockDriverState
*bs
, bool eject_flag
)
6402 BlockDriver
*drv
= bs
->drv
;
6404 if (drv
&& drv
->bdrv_eject
) {
6405 drv
->bdrv_eject(bs
, eject_flag
);
6410 * Lock or unlock the media (if it is locked, the user won't be able
6411 * to eject it manually).
6413 void bdrv_lock_medium(BlockDriverState
*bs
, bool locked
)
6415 BlockDriver
*drv
= bs
->drv
;
6417 trace_bdrv_lock_medium(bs
, locked
);
6419 if (drv
&& drv
->bdrv_lock_medium
) {
6420 drv
->bdrv_lock_medium(bs
, locked
);
6424 /* Get a reference to bs */
6425 void bdrv_ref(BlockDriverState
*bs
)
6430 /* Release a previously grabbed reference to bs.
6431 * If after releasing, reference count is zero, the BlockDriverState is
6433 void bdrv_unref(BlockDriverState
*bs
)
6438 assert(bs
->refcnt
> 0);
6439 if (--bs
->refcnt
== 0) {
6444 struct BdrvOpBlocker
{
6446 QLIST_ENTRY(BdrvOpBlocker
) list
;
6449 bool bdrv_op_is_blocked(BlockDriverState
*bs
, BlockOpType op
, Error
**errp
)
6451 BdrvOpBlocker
*blocker
;
6452 assert((int) op
>= 0 && op
< BLOCK_OP_TYPE_MAX
);
6453 if (!QLIST_EMPTY(&bs
->op_blockers
[op
])) {
6454 blocker
= QLIST_FIRST(&bs
->op_blockers
[op
]);
6455 error_propagate_prepend(errp
, error_copy(blocker
->reason
),
6456 "Node '%s' is busy: ",
6457 bdrv_get_device_or_node_name(bs
));
6463 void bdrv_op_block(BlockDriverState
*bs
, BlockOpType op
, Error
*reason
)
6465 BdrvOpBlocker
*blocker
;
6466 assert((int) op
>= 0 && op
< BLOCK_OP_TYPE_MAX
);
6468 blocker
= g_new0(BdrvOpBlocker
, 1);
6469 blocker
->reason
= reason
;
6470 QLIST_INSERT_HEAD(&bs
->op_blockers
[op
], blocker
, list
);
6473 void bdrv_op_unblock(BlockDriverState
*bs
, BlockOpType op
, Error
*reason
)
6475 BdrvOpBlocker
*blocker
, *next
;
6476 assert((int) op
>= 0 && op
< BLOCK_OP_TYPE_MAX
);
6477 QLIST_FOREACH_SAFE(blocker
, &bs
->op_blockers
[op
], list
, next
) {
6478 if (blocker
->reason
== reason
) {
6479 QLIST_REMOVE(blocker
, list
);
6485 void bdrv_op_block_all(BlockDriverState
*bs
, Error
*reason
)
6488 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
6489 bdrv_op_block(bs
, i
, reason
);
6493 void bdrv_op_unblock_all(BlockDriverState
*bs
, Error
*reason
)
6496 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
6497 bdrv_op_unblock(bs
, i
, reason
);
6501 bool bdrv_op_blocker_is_empty(BlockDriverState
*bs
)
6505 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
6506 if (!QLIST_EMPTY(&bs
->op_blockers
[i
])) {
6513 void bdrv_img_create(const char *filename
, const char *fmt
,
6514 const char *base_filename
, const char *base_fmt
,
6515 char *options
, uint64_t img_size
, int flags
, bool quiet
,
6518 QemuOptsList
*create_opts
= NULL
;
6519 QemuOpts
*opts
= NULL
;
6520 const char *backing_fmt
, *backing_file
;
6522 BlockDriver
*drv
, *proto_drv
;
6523 Error
*local_err
= NULL
;
6526 /* Find driver and parse its options */
6527 drv
= bdrv_find_format(fmt
);
6529 error_setg(errp
, "Unknown file format '%s'", fmt
);
6533 proto_drv
= bdrv_find_protocol(filename
, true, errp
);
6538 if (!drv
->create_opts
) {
6539 error_setg(errp
, "Format driver '%s' does not support image creation",
6544 if (!proto_drv
->create_opts
) {
6545 error_setg(errp
, "Protocol driver '%s' does not support image creation",
6546 proto_drv
->format_name
);
6550 /* Create parameter list */
6551 create_opts
= qemu_opts_append(create_opts
, drv
->create_opts
);
6552 create_opts
= qemu_opts_append(create_opts
, proto_drv
->create_opts
);
6554 opts
= qemu_opts_create(create_opts
, NULL
, 0, &error_abort
);
6556 /* Parse -o options */
6558 if (!qemu_opts_do_parse(opts
, options
, NULL
, errp
)) {
6563 if (!qemu_opt_get(opts
, BLOCK_OPT_SIZE
)) {
6564 qemu_opt_set_number(opts
, BLOCK_OPT_SIZE
, img_size
, &error_abort
);
6565 } else if (img_size
!= UINT64_C(-1)) {
6566 error_setg(errp
, "The image size must be specified only once");
6570 if (base_filename
) {
6571 if (!qemu_opt_set(opts
, BLOCK_OPT_BACKING_FILE
, base_filename
,
6573 error_setg(errp
, "Backing file not supported for file format '%s'",
6580 if (!qemu_opt_set(opts
, BLOCK_OPT_BACKING_FMT
, base_fmt
, NULL
)) {
6581 error_setg(errp
, "Backing file format not supported for file "
6582 "format '%s'", fmt
);
6587 backing_file
= qemu_opt_get(opts
, BLOCK_OPT_BACKING_FILE
);
6589 if (!strcmp(filename
, backing_file
)) {
6590 error_setg(errp
, "Error: Trying to create an image with the "
6591 "same filename as the backing file");
6594 if (backing_file
[0] == '\0') {
6595 error_setg(errp
, "Expected backing file name, got empty string");
6600 backing_fmt
= qemu_opt_get(opts
, BLOCK_OPT_BACKING_FMT
);
6602 /* The size for the image must always be specified, unless we have a backing
6603 * file and we have not been forbidden from opening it. */
6604 size
= qemu_opt_get_size(opts
, BLOCK_OPT_SIZE
, img_size
);
6605 if (backing_file
&& !(flags
& BDRV_O_NO_BACKING
)) {
6606 BlockDriverState
*bs
;
6609 QDict
*backing_options
= NULL
;
6612 bdrv_get_full_backing_filename_from_filename(filename
, backing_file
,
6617 assert(full_backing
);
6620 * No need to do I/O here, which allows us to open encrypted
6621 * backing images without needing the secret
6624 back_flags
&= ~(BDRV_O_RDWR
| BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
);
6625 back_flags
|= BDRV_O_NO_IO
;
6627 backing_options
= qdict_new();
6629 qdict_put_str(backing_options
, "driver", backing_fmt
);
6631 qdict_put_bool(backing_options
, BDRV_OPT_FORCE_SHARE
, true);
6633 bs
= bdrv_open(full_backing
, NULL
, backing_options
, back_flags
,
6635 g_free(full_backing
);
6637 error_append_hint(&local_err
, "Could not open backing image.\n");
6641 warn_report("Deprecated use of backing file without explicit "
6642 "backing format (detected format of %s)",
6643 bs
->drv
->format_name
);
6644 if (bs
->drv
!= &bdrv_raw
) {
6646 * A probe of raw deserves the most attention:
6647 * leaving the backing format out of the image
6648 * will ensure bs->probed is set (ensuring we
6649 * don't accidentally commit into the backing
6650 * file), and allow more spots to warn the users
6651 * to fix their toolchain when opening this image
6652 * later. For other images, we can safely record
6653 * the format that we probed.
6655 backing_fmt
= bs
->drv
->format_name
;
6656 qemu_opt_set(opts
, BLOCK_OPT_BACKING_FMT
, backing_fmt
,
6661 /* Opened BS, have no size */
6662 size
= bdrv_getlength(bs
);
6664 error_setg_errno(errp
, -size
, "Could not get size of '%s'",
6669 qemu_opt_set_number(opts
, BLOCK_OPT_SIZE
, size
, &error_abort
);
6673 /* (backing_file && !(flags & BDRV_O_NO_BACKING)) */
6674 } else if (backing_file
&& !backing_fmt
) {
6675 warn_report("Deprecated use of unopened backing file without "
6676 "explicit backing format, use of this image requires "
6677 "potentially unsafe format probing");
6681 error_setg(errp
, "Image creation needs a size parameter");
6686 printf("Formatting '%s', fmt=%s ", filename
, fmt
);
6687 qemu_opts_print(opts
, " ");
6692 ret
= bdrv_create(drv
, filename
, opts
, &local_err
);
6694 if (ret
== -EFBIG
) {
6695 /* This is generally a better message than whatever the driver would
6696 * deliver (especially because of the cluster_size_hint), since that
6697 * is most probably not much different from "image too large". */
6698 const char *cluster_size_hint
= "";
6699 if (qemu_opt_get_size(opts
, BLOCK_OPT_CLUSTER_SIZE
, 0)) {
6700 cluster_size_hint
= " (try using a larger cluster size)";
6702 error_setg(errp
, "The image size is too large for file format '%s'"
6703 "%s", fmt
, cluster_size_hint
);
6704 error_free(local_err
);
6709 qemu_opts_del(opts
);
6710 qemu_opts_free(create_opts
);
6711 error_propagate(errp
, local_err
);
6714 AioContext
*bdrv_get_aio_context(BlockDriverState
*bs
)
6716 return bs
? bs
->aio_context
: qemu_get_aio_context();
6719 AioContext
*coroutine_fn
bdrv_co_enter(BlockDriverState
*bs
)
6721 Coroutine
*self
= qemu_coroutine_self();
6722 AioContext
*old_ctx
= qemu_coroutine_get_aio_context(self
);
6723 AioContext
*new_ctx
;
6726 * Increase bs->in_flight to ensure that this operation is completed before
6727 * moving the node to a different AioContext. Read new_ctx only afterwards.
6729 bdrv_inc_in_flight(bs
);
6731 new_ctx
= bdrv_get_aio_context(bs
);
6732 aio_co_reschedule_self(new_ctx
);
6736 void coroutine_fn
bdrv_co_leave(BlockDriverState
*bs
, AioContext
*old_ctx
)
6738 aio_co_reschedule_self(old_ctx
);
6739 bdrv_dec_in_flight(bs
);
6742 void coroutine_fn
bdrv_co_lock(BlockDriverState
*bs
)
6744 AioContext
*ctx
= bdrv_get_aio_context(bs
);
6746 /* In the main thread, bs->aio_context won't change concurrently */
6747 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
6750 * We're in coroutine context, so we already hold the lock of the main
6751 * loop AioContext. Don't lock it twice to avoid deadlocks.
6753 assert(qemu_in_coroutine());
6754 if (ctx
!= qemu_get_aio_context()) {
6755 aio_context_acquire(ctx
);
6759 void coroutine_fn
bdrv_co_unlock(BlockDriverState
*bs
)
6761 AioContext
*ctx
= bdrv_get_aio_context(bs
);
6763 assert(qemu_in_coroutine());
6764 if (ctx
!= qemu_get_aio_context()) {
6765 aio_context_release(ctx
);
6769 void bdrv_coroutine_enter(BlockDriverState
*bs
, Coroutine
*co
)
6771 aio_co_enter(bdrv_get_aio_context(bs
), co
);
6774 static void bdrv_do_remove_aio_context_notifier(BdrvAioNotifier
*ban
)
6776 QLIST_REMOVE(ban
, list
);
6780 static void bdrv_detach_aio_context(BlockDriverState
*bs
)
6782 BdrvAioNotifier
*baf
, *baf_tmp
;
6784 assert(!bs
->walking_aio_notifiers
);
6785 bs
->walking_aio_notifiers
= true;
6786 QLIST_FOREACH_SAFE(baf
, &bs
->aio_notifiers
, list
, baf_tmp
) {
6788 bdrv_do_remove_aio_context_notifier(baf
);
6790 baf
->detach_aio_context(baf
->opaque
);
6793 /* Never mind iterating again to check for ->deleted. bdrv_close() will
6794 * remove remaining aio notifiers if we aren't called again.
6796 bs
->walking_aio_notifiers
= false;
6798 if (bs
->drv
&& bs
->drv
->bdrv_detach_aio_context
) {
6799 bs
->drv
->bdrv_detach_aio_context(bs
);
6802 if (bs
->quiesce_counter
) {
6803 aio_enable_external(bs
->aio_context
);
6805 bs
->aio_context
= NULL
;
6808 static void bdrv_attach_aio_context(BlockDriverState
*bs
,
6809 AioContext
*new_context
)
6811 BdrvAioNotifier
*ban
, *ban_tmp
;
6813 if (bs
->quiesce_counter
) {
6814 aio_disable_external(new_context
);
6817 bs
->aio_context
= new_context
;
6819 if (bs
->drv
&& bs
->drv
->bdrv_attach_aio_context
) {
6820 bs
->drv
->bdrv_attach_aio_context(bs
, new_context
);
6823 assert(!bs
->walking_aio_notifiers
);
6824 bs
->walking_aio_notifiers
= true;
6825 QLIST_FOREACH_SAFE(ban
, &bs
->aio_notifiers
, list
, ban_tmp
) {
6827 bdrv_do_remove_aio_context_notifier(ban
);
6829 ban
->attached_aio_context(new_context
, ban
->opaque
);
6832 bs
->walking_aio_notifiers
= false;
6836 * Changes the AioContext used for fd handlers, timers, and BHs by this
6837 * BlockDriverState and all its children and parents.
6839 * Must be called from the main AioContext.
6841 * The caller must own the AioContext lock for the old AioContext of bs, but it
6842 * must not own the AioContext lock for new_context (unless new_context is the
6843 * same as the current context of bs).
6845 * @ignore will accumulate all visited BdrvChild object. The caller is
6846 * responsible for freeing the list afterwards.
6848 void bdrv_set_aio_context_ignore(BlockDriverState
*bs
,
6849 AioContext
*new_context
, GSList
**ignore
)
6851 AioContext
*old_context
= bdrv_get_aio_context(bs
);
6852 GSList
*children_to_process
= NULL
;
6853 GSList
*parents_to_process
= NULL
;
6855 BdrvChild
*child
, *parent
;
6857 g_assert(qemu_get_current_aio_context() == qemu_get_aio_context());
6859 if (old_context
== new_context
) {
6863 bdrv_drained_begin(bs
);
6865 QLIST_FOREACH(child
, &bs
->children
, next
) {
6866 if (g_slist_find(*ignore
, child
)) {
6869 *ignore
= g_slist_prepend(*ignore
, child
);
6870 children_to_process
= g_slist_prepend(children_to_process
, child
);
6873 QLIST_FOREACH(parent
, &bs
->parents
, next_parent
) {
6874 if (g_slist_find(*ignore
, parent
)) {
6877 *ignore
= g_slist_prepend(*ignore
, parent
);
6878 parents_to_process
= g_slist_prepend(parents_to_process
, parent
);
6881 for (entry
= children_to_process
;
6883 entry
= g_slist_next(entry
)) {
6884 child
= entry
->data
;
6885 bdrv_set_aio_context_ignore(child
->bs
, new_context
, ignore
);
6887 g_slist_free(children_to_process
);
6889 for (entry
= parents_to_process
;
6891 entry
= g_slist_next(entry
)) {
6892 parent
= entry
->data
;
6893 assert(parent
->klass
->set_aio_ctx
);
6894 parent
->klass
->set_aio_ctx(parent
, new_context
, ignore
);
6896 g_slist_free(parents_to_process
);
6898 bdrv_detach_aio_context(bs
);
6900 /* Acquire the new context, if necessary */
6901 if (qemu_get_aio_context() != new_context
) {
6902 aio_context_acquire(new_context
);
6905 bdrv_attach_aio_context(bs
, new_context
);
6908 * If this function was recursively called from
6909 * bdrv_set_aio_context_ignore(), there may be nodes in the
6910 * subtree that have not yet been moved to the new AioContext.
6911 * Release the old one so bdrv_drained_end() can poll them.
6913 if (qemu_get_aio_context() != old_context
) {
6914 aio_context_release(old_context
);
6917 bdrv_drained_end(bs
);
6919 if (qemu_get_aio_context() != old_context
) {
6920 aio_context_acquire(old_context
);
6922 if (qemu_get_aio_context() != new_context
) {
6923 aio_context_release(new_context
);
6927 static bool bdrv_parent_can_set_aio_context(BdrvChild
*c
, AioContext
*ctx
,
6928 GSList
**ignore
, Error
**errp
)
6930 if (g_slist_find(*ignore
, c
)) {
6933 *ignore
= g_slist_prepend(*ignore
, c
);
6936 * A BdrvChildClass that doesn't handle AioContext changes cannot
6937 * tolerate any AioContext changes
6939 if (!c
->klass
->can_set_aio_ctx
) {
6940 char *user
= bdrv_child_user_desc(c
);
6941 error_setg(errp
, "Changing iothreads is not supported by %s", user
);
6945 if (!c
->klass
->can_set_aio_ctx(c
, ctx
, ignore
, errp
)) {
6946 assert(!errp
|| *errp
);
6952 bool bdrv_child_can_set_aio_context(BdrvChild
*c
, AioContext
*ctx
,
6953 GSList
**ignore
, Error
**errp
)
6955 if (g_slist_find(*ignore
, c
)) {
6958 *ignore
= g_slist_prepend(*ignore
, c
);
6959 return bdrv_can_set_aio_context(c
->bs
, ctx
, ignore
, errp
);
6962 /* @ignore will accumulate all visited BdrvChild object. The caller is
6963 * responsible for freeing the list afterwards. */
6964 bool bdrv_can_set_aio_context(BlockDriverState
*bs
, AioContext
*ctx
,
6965 GSList
**ignore
, Error
**errp
)
6969 if (bdrv_get_aio_context(bs
) == ctx
) {
6973 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
6974 if (!bdrv_parent_can_set_aio_context(c
, ctx
, ignore
, errp
)) {
6978 QLIST_FOREACH(c
, &bs
->children
, next
) {
6979 if (!bdrv_child_can_set_aio_context(c
, ctx
, ignore
, errp
)) {
6987 int bdrv_child_try_set_aio_context(BlockDriverState
*bs
, AioContext
*ctx
,
6988 BdrvChild
*ignore_child
, Error
**errp
)
6993 ignore
= ignore_child
? g_slist_prepend(NULL
, ignore_child
) : NULL
;
6994 ret
= bdrv_can_set_aio_context(bs
, ctx
, &ignore
, errp
);
6995 g_slist_free(ignore
);
7001 ignore
= ignore_child
? g_slist_prepend(NULL
, ignore_child
) : NULL
;
7002 bdrv_set_aio_context_ignore(bs
, ctx
, &ignore
);
7003 g_slist_free(ignore
);
7008 int bdrv_try_set_aio_context(BlockDriverState
*bs
, AioContext
*ctx
,
7011 return bdrv_child_try_set_aio_context(bs
, ctx
, NULL
, errp
);
7014 void bdrv_add_aio_context_notifier(BlockDriverState
*bs
,
7015 void (*attached_aio_context
)(AioContext
*new_context
, void *opaque
),
7016 void (*detach_aio_context
)(void *opaque
), void *opaque
)
7018 BdrvAioNotifier
*ban
= g_new(BdrvAioNotifier
, 1);
7019 *ban
= (BdrvAioNotifier
){
7020 .attached_aio_context
= attached_aio_context
,
7021 .detach_aio_context
= detach_aio_context
,
7025 QLIST_INSERT_HEAD(&bs
->aio_notifiers
, ban
, list
);
7028 void bdrv_remove_aio_context_notifier(BlockDriverState
*bs
,
7029 void (*attached_aio_context
)(AioContext
*,
7031 void (*detach_aio_context
)(void *),
7034 BdrvAioNotifier
*ban
, *ban_next
;
7036 QLIST_FOREACH_SAFE(ban
, &bs
->aio_notifiers
, list
, ban_next
) {
7037 if (ban
->attached_aio_context
== attached_aio_context
&&
7038 ban
->detach_aio_context
== detach_aio_context
&&
7039 ban
->opaque
== opaque
&&
7040 ban
->deleted
== false)
7042 if (bs
->walking_aio_notifiers
) {
7043 ban
->deleted
= true;
7045 bdrv_do_remove_aio_context_notifier(ban
);
7054 int bdrv_amend_options(BlockDriverState
*bs
, QemuOpts
*opts
,
7055 BlockDriverAmendStatusCB
*status_cb
, void *cb_opaque
,
7060 error_setg(errp
, "Node is ejected");
7063 if (!bs
->drv
->bdrv_amend_options
) {
7064 error_setg(errp
, "Block driver '%s' does not support option amendment",
7065 bs
->drv
->format_name
);
7068 return bs
->drv
->bdrv_amend_options(bs
, opts
, status_cb
,
7069 cb_opaque
, force
, errp
);
7073 * This function checks whether the given @to_replace is allowed to be
7074 * replaced by a node that always shows the same data as @bs. This is
7075 * used for example to verify whether the mirror job can replace
7076 * @to_replace by the target mirrored from @bs.
7077 * To be replaceable, @bs and @to_replace may either be guaranteed to
7078 * always show the same data (because they are only connected through
7079 * filters), or some driver may allow replacing one of its children
7080 * because it can guarantee that this child's data is not visible at
7081 * all (for example, for dissenting quorum children that have no other
7084 bool bdrv_recurse_can_replace(BlockDriverState
*bs
,
7085 BlockDriverState
*to_replace
)
7087 BlockDriverState
*filtered
;
7089 if (!bs
|| !bs
->drv
) {
7093 if (bs
== to_replace
) {
7097 /* See what the driver can do */
7098 if (bs
->drv
->bdrv_recurse_can_replace
) {
7099 return bs
->drv
->bdrv_recurse_can_replace(bs
, to_replace
);
7102 /* For filters without an own implementation, we can recurse on our own */
7103 filtered
= bdrv_filter_bs(bs
);
7105 return bdrv_recurse_can_replace(filtered
, to_replace
);
7113 * Check whether the given @node_name can be replaced by a node that
7114 * has the same data as @parent_bs. If so, return @node_name's BDS;
7117 * @node_name must be a (recursive) *child of @parent_bs (or this
7118 * function will return NULL).
7120 * The result (whether the node can be replaced or not) is only valid
7121 * for as long as no graph or permission changes occur.
7123 BlockDriverState
*check_to_replace_node(BlockDriverState
*parent_bs
,
7124 const char *node_name
, Error
**errp
)
7126 BlockDriverState
*to_replace_bs
= bdrv_find_node(node_name
);
7127 AioContext
*aio_context
;
7129 if (!to_replace_bs
) {
7130 error_setg(errp
, "Failed to find node with node-name='%s'", node_name
);
7134 aio_context
= bdrv_get_aio_context(to_replace_bs
);
7135 aio_context_acquire(aio_context
);
7137 if (bdrv_op_is_blocked(to_replace_bs
, BLOCK_OP_TYPE_REPLACE
, errp
)) {
7138 to_replace_bs
= NULL
;
7142 /* We don't want arbitrary node of the BDS chain to be replaced only the top
7143 * most non filter in order to prevent data corruption.
7144 * Another benefit is that this tests exclude backing files which are
7145 * blocked by the backing blockers.
7147 if (!bdrv_recurse_can_replace(parent_bs
, to_replace_bs
)) {
7148 error_setg(errp
, "Cannot replace '%s' by a node mirrored from '%s', "
7149 "because it cannot be guaranteed that doing so would not "
7150 "lead to an abrupt change of visible data",
7151 node_name
, parent_bs
->node_name
);
7152 to_replace_bs
= NULL
;
7157 aio_context_release(aio_context
);
7158 return to_replace_bs
;
7162 * Iterates through the list of runtime option keys that are said to
7163 * be "strong" for a BDS. An option is called "strong" if it changes
7164 * a BDS's data. For example, the null block driver's "size" and
7165 * "read-zeroes" options are strong, but its "latency-ns" option is
7168 * If a key returned by this function ends with a dot, all options
7169 * starting with that prefix are strong.
7171 static const char *const *strong_options(BlockDriverState
*bs
,
7172 const char *const *curopt
)
7174 static const char *const global_options
[] = {
7175 "driver", "filename", NULL
7179 return &global_options
[0];
7183 if (curopt
== &global_options
[ARRAY_SIZE(global_options
) - 1] && bs
->drv
) {
7184 curopt
= bs
->drv
->strong_runtime_opts
;
7187 return (curopt
&& *curopt
) ? curopt
: NULL
;
7191 * Copies all strong runtime options from bs->options to the given
7192 * QDict. The set of strong option keys is determined by invoking
7195 * Returns true iff any strong option was present in bs->options (and
7196 * thus copied to the target QDict) with the exception of "filename"
7197 * and "driver". The caller is expected to use this value to decide
7198 * whether the existence of strong options prevents the generation of
7201 static bool append_strong_runtime_options(QDict
*d
, BlockDriverState
*bs
)
7203 bool found_any
= false;
7204 const char *const *option_name
= NULL
;
7210 while ((option_name
= strong_options(bs
, option_name
))) {
7211 bool option_given
= false;
7213 assert(strlen(*option_name
) > 0);
7214 if ((*option_name
)[strlen(*option_name
) - 1] != '.') {
7215 QObject
*entry
= qdict_get(bs
->options
, *option_name
);
7220 qdict_put_obj(d
, *option_name
, qobject_ref(entry
));
7221 option_given
= true;
7223 const QDictEntry
*entry
;
7224 for (entry
= qdict_first(bs
->options
); entry
;
7225 entry
= qdict_next(bs
->options
, entry
))
7227 if (strstart(qdict_entry_key(entry
), *option_name
, NULL
)) {
7228 qdict_put_obj(d
, qdict_entry_key(entry
),
7229 qobject_ref(qdict_entry_value(entry
)));
7230 option_given
= true;
7235 /* While "driver" and "filename" need to be included in a JSON filename,
7236 * their existence does not prohibit generation of a plain filename. */
7237 if (!found_any
&& option_given
&&
7238 strcmp(*option_name
, "driver") && strcmp(*option_name
, "filename"))
7244 if (!qdict_haskey(d
, "driver")) {
7245 /* Drivers created with bdrv_new_open_driver() may not have a
7246 * @driver option. Add it here. */
7247 qdict_put_str(d
, "driver", bs
->drv
->format_name
);
7253 /* Note: This function may return false positives; it may return true
7254 * even if opening the backing file specified by bs's image header
7255 * would result in exactly bs->backing. */
7256 bool bdrv_backing_overridden(BlockDriverState
*bs
)
7259 return strcmp(bs
->auto_backing_file
,
7260 bs
->backing
->bs
->filename
);
7262 /* No backing BDS, so if the image header reports any backing
7263 * file, it must have been suppressed */
7264 return bs
->auto_backing_file
[0] != '\0';
7268 /* Updates the following BDS fields:
7269 * - exact_filename: A filename which may be used for opening a block device
7270 * which (mostly) equals the given BDS (even without any
7271 * other options; so reading and writing must return the same
7272 * results, but caching etc. may be different)
7273 * - full_open_options: Options which, when given when opening a block device
7274 * (without a filename), result in a BDS (mostly)
7275 * equalling the given one
7276 * - filename: If exact_filename is set, it is copied here. Otherwise,
7277 * full_open_options is converted to a JSON object, prefixed with
7278 * "json:" (for use through the JSON pseudo protocol) and put here.
7280 void bdrv_refresh_filename(BlockDriverState
*bs
)
7282 BlockDriver
*drv
= bs
->drv
;
7284 BlockDriverState
*primary_child_bs
;
7286 bool backing_overridden
;
7287 bool generate_json_filename
; /* Whether our default implementation should
7288 fill exact_filename (false) or not (true) */
7294 /* This BDS's file name may depend on any of its children's file names, so
7295 * refresh those first */
7296 QLIST_FOREACH(child
, &bs
->children
, next
) {
7297 bdrv_refresh_filename(child
->bs
);
7301 /* For implicit nodes, just copy everything from the single child */
7302 child
= QLIST_FIRST(&bs
->children
);
7303 assert(QLIST_NEXT(child
, next
) == NULL
);
7305 pstrcpy(bs
->exact_filename
, sizeof(bs
->exact_filename
),
7306 child
->bs
->exact_filename
);
7307 pstrcpy(bs
->filename
, sizeof(bs
->filename
), child
->bs
->filename
);
7309 qobject_unref(bs
->full_open_options
);
7310 bs
->full_open_options
= qobject_ref(child
->bs
->full_open_options
);
7315 backing_overridden
= bdrv_backing_overridden(bs
);
7317 if (bs
->open_flags
& BDRV_O_NO_IO
) {
7318 /* Without I/O, the backing file does not change anything.
7319 * Therefore, in such a case (primarily qemu-img), we can
7320 * pretend the backing file has not been overridden even if
7321 * it technically has been. */
7322 backing_overridden
= false;
7325 /* Gather the options QDict */
7327 generate_json_filename
= append_strong_runtime_options(opts
, bs
);
7328 generate_json_filename
|= backing_overridden
;
7330 if (drv
->bdrv_gather_child_options
) {
7331 /* Some block drivers may not want to present all of their children's
7332 * options, or name them differently from BdrvChild.name */
7333 drv
->bdrv_gather_child_options(bs
, opts
, backing_overridden
);
7335 QLIST_FOREACH(child
, &bs
->children
, next
) {
7336 if (child
== bs
->backing
&& !backing_overridden
) {
7337 /* We can skip the backing BDS if it has not been overridden */
7341 qdict_put(opts
, child
->name
,
7342 qobject_ref(child
->bs
->full_open_options
));
7345 if (backing_overridden
&& !bs
->backing
) {
7346 /* Force no backing file */
7347 qdict_put_null(opts
, "backing");
7351 qobject_unref(bs
->full_open_options
);
7352 bs
->full_open_options
= opts
;
7354 primary_child_bs
= bdrv_primary_bs(bs
);
7356 if (drv
->bdrv_refresh_filename
) {
7357 /* Obsolete information is of no use here, so drop the old file name
7358 * information before refreshing it */
7359 bs
->exact_filename
[0] = '\0';
7361 drv
->bdrv_refresh_filename(bs
);
7362 } else if (primary_child_bs
) {
7364 * Try to reconstruct valid information from the underlying
7365 * file -- this only works for format nodes (filter nodes
7366 * cannot be probed and as such must be selected by the user
7367 * either through an options dict, or through a special
7368 * filename which the filter driver must construct in its
7369 * .bdrv_refresh_filename() implementation).
7372 bs
->exact_filename
[0] = '\0';
7375 * We can use the underlying file's filename if:
7376 * - it has a filename,
7377 * - the current BDS is not a filter,
7378 * - the file is a protocol BDS, and
7379 * - opening that file (as this BDS's format) will automatically create
7380 * the BDS tree we have right now, that is:
7381 * - the user did not significantly change this BDS's behavior with
7382 * some explicit (strong) options
7383 * - no non-file child of this BDS has been overridden by the user
7384 * Both of these conditions are represented by generate_json_filename.
7386 if (primary_child_bs
->exact_filename
[0] &&
7387 primary_child_bs
->drv
->bdrv_file_open
&&
7388 !drv
->is_filter
&& !generate_json_filename
)
7390 strcpy(bs
->exact_filename
, primary_child_bs
->exact_filename
);
7394 if (bs
->exact_filename
[0]) {
7395 pstrcpy(bs
->filename
, sizeof(bs
->filename
), bs
->exact_filename
);
7397 GString
*json
= qobject_to_json(QOBJECT(bs
->full_open_options
));
7398 if (snprintf(bs
->filename
, sizeof(bs
->filename
), "json:%s",
7399 json
->str
) >= sizeof(bs
->filename
)) {
7400 /* Give user a hint if we truncated things. */
7401 strcpy(bs
->filename
+ sizeof(bs
->filename
) - 4, "...");
7403 g_string_free(json
, true);
7407 char *bdrv_dirname(BlockDriverState
*bs
, Error
**errp
)
7409 BlockDriver
*drv
= bs
->drv
;
7410 BlockDriverState
*child_bs
;
7413 error_setg(errp
, "Node '%s' is ejected", bs
->node_name
);
7417 if (drv
->bdrv_dirname
) {
7418 return drv
->bdrv_dirname(bs
, errp
);
7421 child_bs
= bdrv_primary_bs(bs
);
7423 return bdrv_dirname(child_bs
, errp
);
7426 bdrv_refresh_filename(bs
);
7427 if (bs
->exact_filename
[0] != '\0') {
7428 return path_combine(bs
->exact_filename
, "");
7431 error_setg(errp
, "Cannot generate a base directory for %s nodes",
7437 * Hot add/remove a BDS's child. So the user can take a child offline when
7438 * it is broken and take a new child online
7440 void bdrv_add_child(BlockDriverState
*parent_bs
, BlockDriverState
*child_bs
,
7444 if (!parent_bs
->drv
|| !parent_bs
->drv
->bdrv_add_child
) {
7445 error_setg(errp
, "The node %s does not support adding a child",
7446 bdrv_get_device_or_node_name(parent_bs
));
7450 if (!QLIST_EMPTY(&child_bs
->parents
)) {
7451 error_setg(errp
, "The node %s already has a parent",
7452 child_bs
->node_name
);
7456 parent_bs
->drv
->bdrv_add_child(parent_bs
, child_bs
, errp
);
7459 void bdrv_del_child(BlockDriverState
*parent_bs
, BdrvChild
*child
, Error
**errp
)
7463 if (!parent_bs
->drv
|| !parent_bs
->drv
->bdrv_del_child
) {
7464 error_setg(errp
, "The node %s does not support removing a child",
7465 bdrv_get_device_or_node_name(parent_bs
));
7469 QLIST_FOREACH(tmp
, &parent_bs
->children
, next
) {
7476 error_setg(errp
, "The node %s does not have a child named %s",
7477 bdrv_get_device_or_node_name(parent_bs
),
7478 bdrv_get_device_or_node_name(child
->bs
));
7482 parent_bs
->drv
->bdrv_del_child(parent_bs
, child
, errp
);
7485 int bdrv_make_empty(BdrvChild
*c
, Error
**errp
)
7487 BlockDriver
*drv
= c
->bs
->drv
;
7490 assert(c
->perm
& (BLK_PERM_WRITE
| BLK_PERM_WRITE_UNCHANGED
));
7492 if (!drv
->bdrv_make_empty
) {
7493 error_setg(errp
, "%s does not support emptying nodes",
7498 ret
= drv
->bdrv_make_empty(c
->bs
);
7500 error_setg_errno(errp
, -ret
, "Failed to empty %s",
7509 * Return the child that @bs acts as an overlay for, and from which data may be
7510 * copied in COW or COR operations. Usually this is the backing file.
7512 BdrvChild
*bdrv_cow_child(BlockDriverState
*bs
)
7514 if (!bs
|| !bs
->drv
) {
7518 if (bs
->drv
->is_filter
) {
7526 assert(bs
->backing
->role
& BDRV_CHILD_COW
);
7531 * If @bs acts as a filter for exactly one of its children, return
7534 BdrvChild
*bdrv_filter_child(BlockDriverState
*bs
)
7538 if (!bs
|| !bs
->drv
) {
7542 if (!bs
->drv
->is_filter
) {
7546 /* Only one of @backing or @file may be used */
7547 assert(!(bs
->backing
&& bs
->file
));
7549 c
= bs
->backing
?: bs
->file
;
7554 assert(c
->role
& BDRV_CHILD_FILTERED
);
7559 * Return either the result of bdrv_cow_child() or bdrv_filter_child(),
7560 * whichever is non-NULL.
7562 * Return NULL if both are NULL.
7564 BdrvChild
*bdrv_filter_or_cow_child(BlockDriverState
*bs
)
7566 BdrvChild
*cow_child
= bdrv_cow_child(bs
);
7567 BdrvChild
*filter_child
= bdrv_filter_child(bs
);
7569 /* Filter nodes cannot have COW backing files */
7570 assert(!(cow_child
&& filter_child
));
7572 return cow_child
?: filter_child
;
7576 * Return the primary child of this node: For filters, that is the
7577 * filtered child. For other nodes, that is usually the child storing
7579 * (A generally more helpful description is that this is (usually) the
7580 * child that has the same filename as @bs.)
7582 * Drivers do not necessarily have a primary child; for example quorum
7585 BdrvChild
*bdrv_primary_child(BlockDriverState
*bs
)
7587 BdrvChild
*c
, *found
= NULL
;
7589 QLIST_FOREACH(c
, &bs
->children
, next
) {
7590 if (c
->role
& BDRV_CHILD_PRIMARY
) {
7599 static BlockDriverState
*bdrv_do_skip_filters(BlockDriverState
*bs
,
7600 bool stop_on_explicit_filter
)
7608 while (!(stop_on_explicit_filter
&& !bs
->implicit
)) {
7609 c
= bdrv_filter_child(bs
);
7612 * A filter that is embedded in a working block graph must
7613 * have a child. Assert this here so this function does
7614 * not return a filter node that is not expected by the
7617 assert(!bs
->drv
|| !bs
->drv
->is_filter
);
7623 * Note that this treats nodes with bs->drv == NULL as not being
7624 * filters (bs->drv == NULL should be replaced by something else
7626 * The advantage of this behavior is that this function will thus
7627 * always return a non-NULL value (given a non-NULL @bs).
7634 * Return the first BDS that has not been added implicitly or that
7635 * does not have a filtered child down the chain starting from @bs
7636 * (including @bs itself).
7638 BlockDriverState
*bdrv_skip_implicit_filters(BlockDriverState
*bs
)
7640 return bdrv_do_skip_filters(bs
, true);
7644 * Return the first BDS that does not have a filtered child down the
7645 * chain starting from @bs (including @bs itself).
7647 BlockDriverState
*bdrv_skip_filters(BlockDriverState
*bs
)
7649 return bdrv_do_skip_filters(bs
, false);
7653 * For a backing chain, return the first non-filter backing image of
7654 * the first non-filter image.
7656 BlockDriverState
*bdrv_backing_chain_next(BlockDriverState
*bs
)
7658 return bdrv_skip_filters(bdrv_cow_bs(bdrv_skip_filters(bs
)));