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/dirty-bitmap.h"
31 #include "block/fuse.h"
32 #include "block/nbd.h"
33 #include "block/qdict.h"
34 #include "qemu/error-report.h"
35 #include "block/module_block.h"
36 #include "qemu/main-loop.h"
37 #include "qemu/module.h"
38 #include "qapi/error.h"
39 #include "qapi/qmp/qdict.h"
40 #include "qapi/qmp/qjson.h"
41 #include "qapi/qmp/qnull.h"
42 #include "qapi/qmp/qstring.h"
43 #include "qapi/qobject-output-visitor.h"
44 #include "qapi/qapi-visit-block-core.h"
45 #include "sysemu/block-backend.h"
46 #include "qemu/notify.h"
47 #include "qemu/option.h"
48 #include "qemu/coroutine.h"
49 #include "block/qapi.h"
50 #include "qemu/timer.h"
51 #include "qemu/cutils.h"
53 #include "qemu/range.h"
55 #include "block/coroutines.h"
58 #include <sys/ioctl.h>
59 #include <sys/queue.h>
60 #if defined(HAVE_SYS_DISK_H)
69 #define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */
71 /* Protected by BQL */
72 static QTAILQ_HEAD(, BlockDriverState
) graph_bdrv_states
=
73 QTAILQ_HEAD_INITIALIZER(graph_bdrv_states
);
75 /* Protected by BQL */
76 static QTAILQ_HEAD(, BlockDriverState
) all_bdrv_states
=
77 QTAILQ_HEAD_INITIALIZER(all_bdrv_states
);
79 /* Protected by BQL */
80 static QLIST_HEAD(, BlockDriver
) bdrv_drivers
=
81 QLIST_HEAD_INITIALIZER(bdrv_drivers
);
83 static BlockDriverState
*bdrv_open_inherit(const char *filename
,
84 const char *reference
,
85 QDict
*options
, int flags
,
86 BlockDriverState
*parent
,
87 const BdrvChildClass
*child_class
,
88 BdrvChildRole child_role
,
91 static bool bdrv_recurse_has_child(BlockDriverState
*bs
,
92 BlockDriverState
*child
);
94 static void GRAPH_WRLOCK
95 bdrv_replace_child_noperm(BdrvChild
*child
, BlockDriverState
*new_bs
);
97 static void GRAPH_WRLOCK
98 bdrv_remove_child(BdrvChild
*child
, Transaction
*tran
);
100 static int bdrv_reopen_prepare(BDRVReopenState
*reopen_state
,
101 BlockReopenQueue
*queue
,
102 Transaction
*change_child_tran
, Error
**errp
);
103 static void bdrv_reopen_commit(BDRVReopenState
*reopen_state
);
104 static void bdrv_reopen_abort(BDRVReopenState
*reopen_state
);
106 static bool bdrv_backing_overridden(BlockDriverState
*bs
);
108 static bool bdrv_change_aio_context(BlockDriverState
*bs
, AioContext
*ctx
,
109 GHashTable
*visited
, Transaction
*tran
,
112 /* If non-zero, use only whitelisted block drivers */
113 static int use_bdrv_whitelist
;
116 static int is_windows_drive_prefix(const char *filename
)
118 return (((filename
[0] >= 'a' && filename
[0] <= 'z') ||
119 (filename
[0] >= 'A' && filename
[0] <= 'Z')) &&
123 int is_windows_drive(const char *filename
)
125 if (is_windows_drive_prefix(filename
) &&
128 if (strstart(filename
, "\\\\.\\", NULL
) ||
129 strstart(filename
, "//./", NULL
))
135 size_t bdrv_opt_mem_align(BlockDriverState
*bs
)
137 if (!bs
|| !bs
->drv
) {
138 /* page size or 4k (hdd sector size) should be on the safe side */
139 return MAX(4096, qemu_real_host_page_size());
143 return bs
->bl
.opt_mem_alignment
;
146 size_t bdrv_min_mem_align(BlockDriverState
*bs
)
148 if (!bs
|| !bs
->drv
) {
149 /* page size or 4k (hdd sector size) should be on the safe side */
150 return MAX(4096, qemu_real_host_page_size());
154 return bs
->bl
.min_mem_alignment
;
157 /* check if the path starts with "<protocol>:" */
158 int path_has_protocol(const char *path
)
163 if (is_windows_drive(path
) ||
164 is_windows_drive_prefix(path
)) {
167 p
= path
+ strcspn(path
, ":/\\");
169 p
= path
+ strcspn(path
, ":/");
175 int path_is_absolute(const char *path
)
178 /* specific case for names like: "\\.\d:" */
179 if (is_windows_drive(path
) || is_windows_drive_prefix(path
)) {
182 return (*path
== '/' || *path
== '\\');
184 return (*path
== '/');
188 /* if filename is absolute, just return its duplicate. Otherwise, build a
189 path to it by considering it is relative to base_path. URL are
191 char *path_combine(const char *base_path
, const char *filename
)
193 const char *protocol_stripped
= NULL
;
198 if (path_is_absolute(filename
)) {
199 return g_strdup(filename
);
202 if (path_has_protocol(base_path
)) {
203 protocol_stripped
= strchr(base_path
, ':');
204 if (protocol_stripped
) {
208 p
= protocol_stripped
?: base_path
;
210 p1
= strrchr(base_path
, '/');
214 p2
= strrchr(base_path
, '\\');
215 if (!p1
|| p2
> p1
) {
230 result
= g_malloc(len
+ strlen(filename
) + 1);
231 memcpy(result
, base_path
, len
);
232 strcpy(result
+ len
, filename
);
238 * Helper function for bdrv_parse_filename() implementations to remove optional
239 * protocol prefixes (especially "file:") from a filename and for putting the
240 * stripped filename into the options QDict if there is such a prefix.
242 void bdrv_parse_filename_strip_prefix(const char *filename
, const char *prefix
,
245 if (strstart(filename
, prefix
, &filename
)) {
246 /* Stripping the explicit protocol prefix may result in a protocol
247 * prefix being (wrongly) detected (if the filename contains a colon) */
248 if (path_has_protocol(filename
)) {
249 GString
*fat_filename
;
251 /* This means there is some colon before the first slash; therefore,
252 * this cannot be an absolute path */
253 assert(!path_is_absolute(filename
));
255 /* And we can thus fix the protocol detection issue by prefixing it
257 fat_filename
= g_string_new("./");
258 g_string_append(fat_filename
, filename
);
260 assert(!path_has_protocol(fat_filename
->str
));
262 qdict_put(options
, "filename",
263 qstring_from_gstring(fat_filename
));
265 /* If no protocol prefix was detected, we can use the shortened
267 qdict_put_str(options
, "filename", filename
);
273 /* Returns whether the image file is opened as read-only. Note that this can
274 * return false and writing to the image file is still not possible because the
275 * image is inactivated. */
276 bool bdrv_is_read_only(BlockDriverState
*bs
)
279 return !(bs
->open_flags
& BDRV_O_RDWR
);
282 static int bdrv_can_set_read_only(BlockDriverState
*bs
, bool read_only
,
283 bool ignore_allow_rdw
, Error
**errp
)
287 /* Do not set read_only if copy_on_read is enabled */
288 if (bs
->copy_on_read
&& read_only
) {
289 error_setg(errp
, "Can't set node '%s' to r/o with copy-on-read enabled",
290 bdrv_get_device_or_node_name(bs
));
294 /* Do not clear read_only if it is prohibited */
295 if (!read_only
&& !(bs
->open_flags
& BDRV_O_ALLOW_RDWR
) &&
298 error_setg(errp
, "Node '%s' is read only",
299 bdrv_get_device_or_node_name(bs
));
307 * Called by a driver that can only provide a read-only image.
309 * Returns 0 if the node is already read-only or it could switch the node to
310 * read-only because BDRV_O_AUTO_RDONLY is set.
312 * Returns -EACCES if the node is read-write and BDRV_O_AUTO_RDONLY is not set
313 * or bdrv_can_set_read_only() forbids making the node read-only. If @errmsg
314 * is not NULL, it is used as the error message for the Error object.
316 int bdrv_apply_auto_read_only(BlockDriverState
*bs
, const char *errmsg
,
322 if (!(bs
->open_flags
& BDRV_O_RDWR
)) {
325 if (!(bs
->open_flags
& BDRV_O_AUTO_RDONLY
)) {
329 ret
= bdrv_can_set_read_only(bs
, true, false, NULL
);
334 bs
->open_flags
&= ~BDRV_O_RDWR
;
339 error_setg(errp
, "%s", errmsg
?: "Image is read-only");
344 * If @backing is empty, this function returns NULL without setting
345 * @errp. In all other cases, NULL will only be returned with @errp
348 * Therefore, a return value of NULL without @errp set means that
349 * there is no backing file; if @errp is set, there is one but its
350 * absolute filename cannot be generated.
352 char *bdrv_get_full_backing_filename_from_filename(const char *backed
,
356 if (backing
[0] == '\0') {
358 } else if (path_has_protocol(backing
) || path_is_absolute(backing
)) {
359 return g_strdup(backing
);
360 } else if (backed
[0] == '\0' || strstart(backed
, "json:", NULL
)) {
361 error_setg(errp
, "Cannot use relative backing file names for '%s'",
365 return path_combine(backed
, backing
);
370 * If @filename is empty or NULL, this function returns NULL without
371 * setting @errp. In all other cases, NULL will only be returned with
374 static char *bdrv_make_absolute_filename(BlockDriverState
*relative_to
,
375 const char *filename
, Error
**errp
)
377 char *dir
, *full_name
;
379 if (!filename
|| filename
[0] == '\0') {
381 } else if (path_has_protocol(filename
) || path_is_absolute(filename
)) {
382 return g_strdup(filename
);
385 dir
= bdrv_dirname(relative_to
, errp
);
390 full_name
= g_strconcat(dir
, filename
, NULL
);
395 char *bdrv_get_full_backing_filename(BlockDriverState
*bs
, Error
**errp
)
398 return bdrv_make_absolute_filename(bs
, bs
->backing_file
, errp
);
401 void bdrv_register(BlockDriver
*bdrv
)
403 assert(bdrv
->format_name
);
405 QLIST_INSERT_HEAD(&bdrv_drivers
, bdrv
, list
);
408 BlockDriverState
*bdrv_new(void)
410 BlockDriverState
*bs
;
415 bs
= g_new0(BlockDriverState
, 1);
416 QLIST_INIT(&bs
->dirty_bitmaps
);
417 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
418 QLIST_INIT(&bs
->op_blockers
[i
]);
420 qemu_mutex_init(&bs
->reqs_lock
);
421 qemu_mutex_init(&bs
->dirty_bitmap_mutex
);
423 bs
->aio_context
= qemu_get_aio_context();
425 qemu_co_queue_init(&bs
->flush_queue
);
427 qemu_co_mutex_init(&bs
->bsc_modify_lock
);
428 bs
->block_status_cache
= g_new0(BdrvBlockStatusCache
, 1);
430 for (i
= 0; i
< bdrv_drain_all_count
; i
++) {
431 bdrv_drained_begin(bs
);
434 QTAILQ_INSERT_TAIL(&all_bdrv_states
, bs
, bs_list
);
439 static BlockDriver
*bdrv_do_find_format(const char *format_name
)
444 QLIST_FOREACH(drv1
, &bdrv_drivers
, list
) {
445 if (!strcmp(drv1
->format_name
, format_name
)) {
453 BlockDriver
*bdrv_find_format(const char *format_name
)
460 drv1
= bdrv_do_find_format(format_name
);
465 /* The driver isn't registered, maybe we need to load a module */
466 for (i
= 0; i
< (int)ARRAY_SIZE(block_driver_modules
); ++i
) {
467 if (!strcmp(block_driver_modules
[i
].format_name
, format_name
)) {
468 Error
*local_err
= NULL
;
469 int rv
= block_module_load(block_driver_modules
[i
].library_name
,
472 return bdrv_do_find_format(format_name
);
474 error_report_err(local_err
);
482 static int bdrv_format_is_whitelisted(const char *format_name
, bool read_only
)
484 static const char *whitelist_rw
[] = {
485 CONFIG_BDRV_RW_WHITELIST
488 static const char *whitelist_ro
[] = {
489 CONFIG_BDRV_RO_WHITELIST
494 if (!whitelist_rw
[0] && !whitelist_ro
[0]) {
495 return 1; /* no whitelist, anything goes */
498 for (p
= whitelist_rw
; *p
; p
++) {
499 if (!strcmp(format_name
, *p
)) {
504 for (p
= whitelist_ro
; *p
; p
++) {
505 if (!strcmp(format_name
, *p
)) {
513 int bdrv_is_whitelisted(BlockDriver
*drv
, bool read_only
)
516 return bdrv_format_is_whitelisted(drv
->format_name
, read_only
);
519 bool bdrv_uses_whitelist(void)
521 return use_bdrv_whitelist
;
524 typedef struct CreateCo
{
532 int coroutine_fn
bdrv_co_create(BlockDriver
*drv
, const char *filename
,
533 QemuOpts
*opts
, Error
**errp
)
539 if (!drv
->bdrv_co_create_opts
) {
540 error_setg(errp
, "Driver '%s' does not support image creation",
545 ret
= drv
->bdrv_co_create_opts(drv
, filename
, opts
, errp
);
546 if (ret
< 0 && !*errp
) {
547 error_setg_errno(errp
, -ret
, "Could not create image");
554 * Helper function for bdrv_create_file_fallback(): Resize @blk to at
555 * least the given @minimum_size.
557 * On success, return @blk's actual length.
558 * Otherwise, return -errno.
560 static int64_t coroutine_fn GRAPH_UNLOCKED
561 create_file_fallback_truncate(BlockBackend
*blk
, int64_t minimum_size
,
564 Error
*local_err
= NULL
;
570 ret
= blk_co_truncate(blk
, minimum_size
, false, PREALLOC_MODE_OFF
, 0,
572 if (ret
< 0 && ret
!= -ENOTSUP
) {
573 error_propagate(errp
, local_err
);
577 size
= blk_co_getlength(blk
);
579 error_free(local_err
);
580 error_setg_errno(errp
, -size
,
581 "Failed to inquire the new image file's length");
585 if (size
< minimum_size
) {
586 /* Need to grow the image, but we failed to do that */
587 error_propagate(errp
, local_err
);
591 error_free(local_err
);
598 * Helper function for bdrv_create_file_fallback(): Zero the first
599 * sector to remove any potentially pre-existing image header.
601 static int coroutine_fn
602 create_file_fallback_zero_first_sector(BlockBackend
*blk
,
603 int64_t current_size
,
606 int64_t bytes_to_clear
;
611 bytes_to_clear
= MIN(current_size
, BDRV_SECTOR_SIZE
);
612 if (bytes_to_clear
) {
613 ret
= blk_co_pwrite_zeroes(blk
, 0, bytes_to_clear
, BDRV_REQ_MAY_UNMAP
);
615 error_setg_errno(errp
, -ret
,
616 "Failed to clear the new image's first sector");
625 * Simple implementation of bdrv_co_create_opts for protocol drivers
626 * which only support creation via opening a file
627 * (usually existing raw storage device)
629 int coroutine_fn
bdrv_co_create_opts_simple(BlockDriver
*drv
,
630 const char *filename
,
638 PreallocMode prealloc
;
639 Error
*local_err
= NULL
;
644 size
= qemu_opt_get_size_del(opts
, BLOCK_OPT_SIZE
, 0);
645 buf
= qemu_opt_get_del(opts
, BLOCK_OPT_PREALLOC
);
646 prealloc
= qapi_enum_parse(&PreallocMode_lookup
, buf
,
647 PREALLOC_MODE_OFF
, &local_err
);
650 error_propagate(errp
, local_err
);
654 if (prealloc
!= PREALLOC_MODE_OFF
) {
655 error_setg(errp
, "Unsupported preallocation mode '%s'",
656 PreallocMode_str(prealloc
));
660 options
= qdict_new();
661 qdict_put_str(options
, "driver", drv
->format_name
);
663 blk
= blk_co_new_open(filename
, NULL
, options
,
664 BDRV_O_RDWR
| BDRV_O_RESIZE
, errp
);
666 error_prepend(errp
, "Protocol driver '%s' does not support creating "
667 "new images, so an existing image must be selected as "
668 "the target; however, opening the given target as an "
669 "existing image failed: ",
674 size
= create_file_fallback_truncate(blk
, size
, errp
);
680 ret
= create_file_fallback_zero_first_sector(blk
, size
, errp
);
691 int coroutine_fn
bdrv_co_create_file(const char *filename
, QemuOpts
*opts
,
694 QemuOpts
*protocol_opts
;
701 drv
= bdrv_find_protocol(filename
, true, errp
);
706 if (!drv
->create_opts
) {
707 error_setg(errp
, "Driver '%s' does not support image creation",
713 * 'opts' contains a QemuOptsList with a combination of format and protocol
716 * The format properly removes its options, but the default values remain
717 * in 'opts->list'. So if the protocol has options with the same name
718 * (e.g. rbd has 'cluster_size' as qcow2), it will see the default values
719 * of the format, since for overlapping options, the format wins.
721 * To avoid this issue, lets convert QemuOpts to QDict, in this way we take
722 * only the set options, and then convert it back to QemuOpts, using the
723 * create_opts of the protocol. So the new QemuOpts, will contain only the
726 qdict
= qemu_opts_to_qdict(opts
, NULL
);
727 protocol_opts
= qemu_opts_from_qdict(drv
->create_opts
, qdict
, errp
);
728 if (protocol_opts
== NULL
) {
733 ret
= bdrv_co_create(drv
, filename
, protocol_opts
, errp
);
735 qemu_opts_del(protocol_opts
);
736 qobject_unref(qdict
);
740 int coroutine_fn
bdrv_co_delete_file(BlockDriverState
*bs
, Error
**errp
)
742 Error
*local_err
= NULL
;
747 assert_bdrv_graph_readable();
750 error_setg(errp
, "Block node '%s' is not opened", bs
->filename
);
754 if (!bs
->drv
->bdrv_co_delete_file
) {
755 error_setg(errp
, "Driver '%s' does not support image deletion",
756 bs
->drv
->format_name
);
760 ret
= bs
->drv
->bdrv_co_delete_file(bs
, &local_err
);
762 error_propagate(errp
, local_err
);
768 void coroutine_fn
bdrv_co_delete_file_noerr(BlockDriverState
*bs
)
770 Error
*local_err
= NULL
;
778 ret
= bdrv_co_delete_file(bs
, &local_err
);
780 * ENOTSUP will happen if the block driver doesn't support
781 * the 'bdrv_co_delete_file' interface. This is a predictable
782 * scenario and shouldn't be reported back to the user.
784 if (ret
== -ENOTSUP
) {
785 error_free(local_err
);
786 } else if (ret
< 0) {
787 error_report_err(local_err
);
792 * Try to get @bs's logical and physical block size.
793 * On success, store them in @bsz struct and return 0.
794 * On failure return -errno.
795 * @bs must not be empty.
797 int bdrv_probe_blocksizes(BlockDriverState
*bs
, BlockSizes
*bsz
)
799 BlockDriver
*drv
= bs
->drv
;
800 BlockDriverState
*filtered
= bdrv_filter_bs(bs
);
803 if (drv
&& drv
->bdrv_probe_blocksizes
) {
804 return drv
->bdrv_probe_blocksizes(bs
, bsz
);
805 } else if (filtered
) {
806 return bdrv_probe_blocksizes(filtered
, bsz
);
813 * Try to get @bs's geometry (cyls, heads, sectors).
814 * On success, store them in @geo struct and return 0.
815 * On failure return -errno.
816 * @bs must not be empty.
818 int bdrv_probe_geometry(BlockDriverState
*bs
, HDGeometry
*geo
)
820 BlockDriver
*drv
= bs
->drv
;
821 BlockDriverState
*filtered
= bdrv_filter_bs(bs
);
824 if (drv
&& drv
->bdrv_probe_geometry
) {
825 return drv
->bdrv_probe_geometry(bs
, geo
);
826 } else if (filtered
) {
827 return bdrv_probe_geometry(filtered
, geo
);
834 * Create a uniquely-named empty temporary file.
835 * Return the actual file name used upon success, otherwise NULL.
836 * This string should be freed with g_free() when not needed any longer.
838 * Note: creating a temporary file for the caller to (re)open is
839 * inherently racy. Use g_file_open_tmp() instead whenever practical.
841 char *create_tmp_file(Error
**errp
)
845 g_autofree
char *filename
= NULL
;
847 tmpdir
= g_get_tmp_dir();
850 * See commit 69bef79 ("block: use /var/tmp instead of /tmp for -snapshot")
852 * This function is used to create temporary disk images (like -snapshot),
853 * so the files can become very large. /tmp is often a tmpfs where as
854 * /var/tmp is usually on a disk, so more appropriate for disk images.
856 if (!g_strcmp0(tmpdir
, "/tmp")) {
861 filename
= g_strdup_printf("%s/vl.XXXXXX", tmpdir
);
862 fd
= g_mkstemp(filename
);
864 error_setg_errno(errp
, errno
, "Could not open temporary file '%s'",
870 return g_steal_pointer(&filename
);
874 * Detect host devices. By convention, /dev/cdrom[N] is always
875 * recognized as a host CDROM.
877 static BlockDriver
*find_hdev_driver(const char *filename
)
879 int score_max
= 0, score
;
880 BlockDriver
*drv
= NULL
, *d
;
883 QLIST_FOREACH(d
, &bdrv_drivers
, list
) {
884 if (d
->bdrv_probe_device
) {
885 score
= d
->bdrv_probe_device(filename
);
886 if (score
> score_max
) {
896 static BlockDriver
*bdrv_do_find_protocol(const char *protocol
)
901 QLIST_FOREACH(drv1
, &bdrv_drivers
, list
) {
902 if (drv1
->protocol_name
&& !strcmp(drv1
->protocol_name
, protocol
)) {
910 BlockDriver
*bdrv_find_protocol(const char *filename
,
911 bool allow_protocol_prefix
,
921 /* TODO Drivers without bdrv_file_open must be specified explicitly */
924 * XXX(hch): we really should not let host device detection
925 * override an explicit protocol specification, but moving this
926 * later breaks access to device names with colons in them.
927 * Thanks to the brain-dead persistent naming schemes on udev-
928 * based Linux systems those actually are quite common.
930 drv1
= find_hdev_driver(filename
);
935 if (!path_has_protocol(filename
) || !allow_protocol_prefix
) {
939 p
= strchr(filename
, ':');
942 if (len
> sizeof(protocol
) - 1)
943 len
= sizeof(protocol
) - 1;
944 memcpy(protocol
, filename
, len
);
945 protocol
[len
] = '\0';
947 drv1
= bdrv_do_find_protocol(protocol
);
952 for (i
= 0; i
< (int)ARRAY_SIZE(block_driver_modules
); ++i
) {
953 if (block_driver_modules
[i
].protocol_name
&&
954 !strcmp(block_driver_modules
[i
].protocol_name
, protocol
)) {
955 int rv
= block_module_load(block_driver_modules
[i
].library_name
, errp
);
957 drv1
= bdrv_do_find_protocol(protocol
);
966 error_setg(errp
, "Unknown protocol '%s'", protocol
);
972 * Guess image format by probing its contents.
973 * This is not a good idea when your image is raw (CVE-2008-2004), but
974 * we do it anyway for backward compatibility.
976 * @buf contains the image's first @buf_size bytes.
977 * @buf_size is the buffer size in bytes (generally BLOCK_PROBE_BUF_SIZE,
978 * but can be smaller if the image file is smaller)
979 * @filename is its filename.
981 * For all block drivers, call the bdrv_probe() method to get its
983 * Return the first block driver with the highest probing score.
985 BlockDriver
*bdrv_probe_all(const uint8_t *buf
, int buf_size
,
986 const char *filename
)
988 int score_max
= 0, score
;
989 BlockDriver
*drv
= NULL
, *d
;
992 QLIST_FOREACH(d
, &bdrv_drivers
, list
) {
994 score
= d
->bdrv_probe(buf
, buf_size
, filename
);
995 if (score
> score_max
) {
1005 static int find_image_format(BlockBackend
*file
, const char *filename
,
1006 BlockDriver
**pdrv
, Error
**errp
)
1009 uint8_t buf
[BLOCK_PROBE_BUF_SIZE
];
1012 GLOBAL_STATE_CODE();
1014 /* Return the raw BlockDriver * to scsi-generic devices or empty drives */
1015 if (blk_is_sg(file
) || !blk_is_inserted(file
) || blk_getlength(file
) == 0) {
1020 ret
= blk_pread(file
, 0, sizeof(buf
), buf
, 0);
1022 error_setg_errno(errp
, -ret
, "Could not read image for determining its "
1028 drv
= bdrv_probe_all(buf
, sizeof(buf
), filename
);
1030 error_setg(errp
, "Could not determine image format: No compatible "
1041 * Set the current 'total_sectors' value
1042 * Return 0 on success, -errno on error.
1044 int coroutine_fn
bdrv_co_refresh_total_sectors(BlockDriverState
*bs
,
1047 BlockDriver
*drv
= bs
->drv
;
1049 assert_bdrv_graph_readable();
1055 /* Do not attempt drv->bdrv_co_getlength() on scsi-generic devices */
1059 /* query actual device if possible, otherwise just trust the hint */
1060 if (drv
->bdrv_co_getlength
) {
1061 int64_t length
= drv
->bdrv_co_getlength(bs
);
1065 hint
= DIV_ROUND_UP(length
, BDRV_SECTOR_SIZE
);
1068 bs
->total_sectors
= hint
;
1070 if (bs
->total_sectors
* BDRV_SECTOR_SIZE
> BDRV_MAX_LENGTH
) {
1078 * Combines a QDict of new block driver @options with any missing options taken
1079 * from @old_options, so that leaving out an option defaults to its old value.
1081 static void bdrv_join_options(BlockDriverState
*bs
, QDict
*options
,
1084 GLOBAL_STATE_CODE();
1085 if (bs
->drv
&& bs
->drv
->bdrv_join_options
) {
1086 bs
->drv
->bdrv_join_options(options
, old_options
);
1088 qdict_join(options
, old_options
, false);
1092 static BlockdevDetectZeroesOptions
bdrv_parse_detect_zeroes(QemuOpts
*opts
,
1096 Error
*local_err
= NULL
;
1097 char *value
= qemu_opt_get_del(opts
, "detect-zeroes");
1098 BlockdevDetectZeroesOptions detect_zeroes
=
1099 qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup
, value
,
1100 BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF
, &local_err
);
1101 GLOBAL_STATE_CODE();
1104 error_propagate(errp
, local_err
);
1105 return detect_zeroes
;
1108 if (detect_zeroes
== BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP
&&
1109 !(open_flags
& BDRV_O_UNMAP
))
1111 error_setg(errp
, "setting detect-zeroes to unmap is not allowed "
1112 "without setting discard operation to unmap");
1115 return detect_zeroes
;
1119 * Set open flags for aio engine
1121 * Return 0 on success, -1 if the engine specified is invalid
1123 int bdrv_parse_aio(const char *mode
, int *flags
)
1125 if (!strcmp(mode
, "threads")) {
1126 /* do nothing, default */
1127 } else if (!strcmp(mode
, "native")) {
1128 *flags
|= BDRV_O_NATIVE_AIO
;
1129 #ifdef CONFIG_LINUX_IO_URING
1130 } else if (!strcmp(mode
, "io_uring")) {
1131 *flags
|= BDRV_O_IO_URING
;
1141 * Set open flags for a given discard mode
1143 * Return 0 on success, -1 if the discard mode was invalid.
1145 int bdrv_parse_discard_flags(const char *mode
, int *flags
)
1147 *flags
&= ~BDRV_O_UNMAP
;
1149 if (!strcmp(mode
, "off") || !strcmp(mode
, "ignore")) {
1151 } else if (!strcmp(mode
, "on") || !strcmp(mode
, "unmap")) {
1152 *flags
|= BDRV_O_UNMAP
;
1161 * Set open flags for a given cache mode
1163 * Return 0 on success, -1 if the cache mode was invalid.
1165 int bdrv_parse_cache_mode(const char *mode
, int *flags
, bool *writethrough
)
1167 *flags
&= ~BDRV_O_CACHE_MASK
;
1169 if (!strcmp(mode
, "off") || !strcmp(mode
, "none")) {
1170 *writethrough
= false;
1171 *flags
|= BDRV_O_NOCACHE
;
1172 } else if (!strcmp(mode
, "directsync")) {
1173 *writethrough
= true;
1174 *flags
|= BDRV_O_NOCACHE
;
1175 } else if (!strcmp(mode
, "writeback")) {
1176 *writethrough
= false;
1177 } else if (!strcmp(mode
, "unsafe")) {
1178 *writethrough
= false;
1179 *flags
|= BDRV_O_NO_FLUSH
;
1180 } else if (!strcmp(mode
, "writethrough")) {
1181 *writethrough
= true;
1189 static char *bdrv_child_get_parent_desc(BdrvChild
*c
)
1191 BlockDriverState
*parent
= c
->opaque
;
1192 return g_strdup_printf("node '%s'", bdrv_get_node_name(parent
));
1195 static void bdrv_child_cb_drained_begin(BdrvChild
*child
)
1197 BlockDriverState
*bs
= child
->opaque
;
1198 bdrv_do_drained_begin_quiesce(bs
, NULL
);
1201 static bool bdrv_child_cb_drained_poll(BdrvChild
*child
)
1203 BlockDriverState
*bs
= child
->opaque
;
1204 return bdrv_drain_poll(bs
, NULL
, false);
1207 static void bdrv_child_cb_drained_end(BdrvChild
*child
)
1209 BlockDriverState
*bs
= child
->opaque
;
1210 bdrv_drained_end(bs
);
1213 static int bdrv_child_cb_inactivate(BdrvChild
*child
)
1215 BlockDriverState
*bs
= child
->opaque
;
1216 GLOBAL_STATE_CODE();
1217 assert(bs
->open_flags
& BDRV_O_INACTIVE
);
1221 static bool bdrv_child_cb_change_aio_ctx(BdrvChild
*child
, AioContext
*ctx
,
1222 GHashTable
*visited
, Transaction
*tran
,
1225 BlockDriverState
*bs
= child
->opaque
;
1226 return bdrv_change_aio_context(bs
, ctx
, visited
, tran
, errp
);
1230 * Returns the options and flags that a temporary snapshot should get, based on
1231 * the originally requested flags (the originally requested image will have
1232 * flags like a backing file)
1234 static void bdrv_temp_snapshot_options(int *child_flags
, QDict
*child_options
,
1235 int parent_flags
, QDict
*parent_options
)
1237 GLOBAL_STATE_CODE();
1238 *child_flags
= (parent_flags
& ~BDRV_O_SNAPSHOT
) | BDRV_O_TEMPORARY
;
1240 /* For temporary files, unconditional cache=unsafe is fine */
1241 qdict_set_default_str(child_options
, BDRV_OPT_CACHE_DIRECT
, "off");
1242 qdict_set_default_str(child_options
, BDRV_OPT_CACHE_NO_FLUSH
, "on");
1244 /* Copy the read-only and discard options from the parent */
1245 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_READ_ONLY
);
1246 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_DISCARD
);
1248 /* aio=native doesn't work for cache.direct=off, so disable it for the
1249 * temporary snapshot */
1250 *child_flags
&= ~BDRV_O_NATIVE_AIO
;
1253 static void bdrv_backing_attach(BdrvChild
*c
)
1255 BlockDriverState
*parent
= c
->opaque
;
1256 BlockDriverState
*backing_hd
= c
->bs
;
1258 GLOBAL_STATE_CODE();
1259 assert(!parent
->backing_blocker
);
1260 error_setg(&parent
->backing_blocker
,
1261 "node is used as backing hd of '%s'",
1262 bdrv_get_device_or_node_name(parent
));
1264 bdrv_refresh_filename(backing_hd
);
1266 parent
->open_flags
&= ~BDRV_O_NO_BACKING
;
1268 bdrv_op_block_all(backing_hd
, parent
->backing_blocker
);
1269 /* Otherwise we won't be able to commit or stream */
1270 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_COMMIT_TARGET
,
1271 parent
->backing_blocker
);
1272 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_STREAM
,
1273 parent
->backing_blocker
);
1275 * We do backup in 3 ways:
1277 * The target bs is new opened, and the source is top BDS
1278 * 2. blockdev backup
1279 * Both the source and the target are top BDSes.
1280 * 3. internal backup(used for block replication)
1281 * Both the source and the target are backing file
1283 * In case 1 and 2, neither the source nor the target is the backing file.
1284 * In case 3, we will block the top BDS, so there is only one block job
1285 * for the top BDS and its backing chain.
1287 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_BACKUP_SOURCE
,
1288 parent
->backing_blocker
);
1289 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_BACKUP_TARGET
,
1290 parent
->backing_blocker
);
1293 static void bdrv_backing_detach(BdrvChild
*c
)
1295 BlockDriverState
*parent
= c
->opaque
;
1297 GLOBAL_STATE_CODE();
1298 assert(parent
->backing_blocker
);
1299 bdrv_op_unblock_all(c
->bs
, parent
->backing_blocker
);
1300 error_free(parent
->backing_blocker
);
1301 parent
->backing_blocker
= NULL
;
1304 static int bdrv_backing_update_filename(BdrvChild
*c
, BlockDriverState
*base
,
1305 const char *filename
, Error
**errp
)
1307 BlockDriverState
*parent
= c
->opaque
;
1308 bool read_only
= bdrv_is_read_only(parent
);
1310 GLOBAL_STATE_CODE();
1313 ret
= bdrv_reopen_set_read_only(parent
, false, errp
);
1319 ret
= bdrv_change_backing_file(parent
, filename
,
1320 base
->drv
? base
->drv
->format_name
: "",
1323 error_setg_errno(errp
, -ret
, "Could not update backing file link");
1327 bdrv_reopen_set_read_only(parent
, true, NULL
);
1334 * Returns the options and flags that a generic child of a BDS should
1335 * get, based on the given options and flags for the parent BDS.
1337 static void bdrv_inherited_options(BdrvChildRole role
, bool parent_is_format
,
1338 int *child_flags
, QDict
*child_options
,
1339 int parent_flags
, QDict
*parent_options
)
1341 int flags
= parent_flags
;
1342 GLOBAL_STATE_CODE();
1345 * First, decide whether to set, clear, or leave BDRV_O_PROTOCOL.
1346 * Generally, the question to answer is: Should this child be
1347 * format-probed by default?
1351 * Pure and non-filtered data children of non-format nodes should
1352 * be probed by default (even when the node itself has BDRV_O_PROTOCOL
1353 * set). This only affects a very limited set of drivers (namely
1354 * quorum and blkverify when this comment was written).
1355 * Force-clear BDRV_O_PROTOCOL then.
1357 if (!parent_is_format
&&
1358 (role
& BDRV_CHILD_DATA
) &&
1359 !(role
& (BDRV_CHILD_METADATA
| BDRV_CHILD_FILTERED
)))
1361 flags
&= ~BDRV_O_PROTOCOL
;
1365 * All children of format nodes (except for COW children) and all
1366 * metadata children in general should never be format-probed.
1367 * Force-set BDRV_O_PROTOCOL then.
1369 if ((parent_is_format
&& !(role
& BDRV_CHILD_COW
)) ||
1370 (role
& BDRV_CHILD_METADATA
))
1372 flags
|= BDRV_O_PROTOCOL
;
1376 * If the cache mode isn't explicitly set, inherit direct and no-flush from
1379 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_CACHE_DIRECT
);
1380 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_CACHE_NO_FLUSH
);
1381 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_FORCE_SHARE
);
1383 if (role
& BDRV_CHILD_COW
) {
1384 /* backing files are opened read-only by default */
1385 qdict_set_default_str(child_options
, BDRV_OPT_READ_ONLY
, "on");
1386 qdict_set_default_str(child_options
, BDRV_OPT_AUTO_READ_ONLY
, "off");
1388 /* Inherit the read-only option from the parent if it's not set */
1389 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_READ_ONLY
);
1390 qdict_copy_default(child_options
, parent_options
,
1391 BDRV_OPT_AUTO_READ_ONLY
);
1395 * bdrv_co_pdiscard() respects unmap policy for the parent, so we
1396 * can default to enable it on lower layers regardless of the
1399 qdict_set_default_str(child_options
, BDRV_OPT_DISCARD
, "unmap");
1401 /* Clear flags that only apply to the top layer */
1402 flags
&= ~(BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
| BDRV_O_COPY_ON_READ
);
1404 if (role
& BDRV_CHILD_METADATA
) {
1405 flags
&= ~BDRV_O_NO_IO
;
1407 if (role
& BDRV_CHILD_COW
) {
1408 flags
&= ~BDRV_O_TEMPORARY
;
1411 *child_flags
= flags
;
1414 static void GRAPH_WRLOCK
bdrv_child_cb_attach(BdrvChild
*child
)
1416 BlockDriverState
*bs
= child
->opaque
;
1418 assert_bdrv_graph_writable();
1419 QLIST_INSERT_HEAD(&bs
->children
, child
, next
);
1420 if (bs
->drv
->is_filter
|| (child
->role
& BDRV_CHILD_FILTERED
)) {
1422 * Here we handle filters and block/raw-format.c when it behave like
1423 * filter. They generally have a single PRIMARY child, which is also the
1424 * FILTERED child, and that they may have multiple more children, which
1425 * are neither PRIMARY nor FILTERED. And never we have a COW child here.
1426 * So bs->file will be the PRIMARY child, unless the PRIMARY child goes
1427 * into bs->backing on exceptional cases; and bs->backing will be
1430 assert(!(child
->role
& BDRV_CHILD_COW
));
1431 if (child
->role
& BDRV_CHILD_PRIMARY
) {
1432 assert(child
->role
& BDRV_CHILD_FILTERED
);
1433 assert(!bs
->backing
);
1436 if (bs
->drv
->filtered_child_is_backing
) {
1437 bs
->backing
= child
;
1442 assert(!(child
->role
& BDRV_CHILD_FILTERED
));
1444 } else if (child
->role
& BDRV_CHILD_COW
) {
1445 assert(bs
->drv
->supports_backing
);
1446 assert(!(child
->role
& BDRV_CHILD_PRIMARY
));
1447 assert(!bs
->backing
);
1448 bs
->backing
= child
;
1449 bdrv_backing_attach(child
);
1450 } else if (child
->role
& BDRV_CHILD_PRIMARY
) {
1456 static void GRAPH_WRLOCK
bdrv_child_cb_detach(BdrvChild
*child
)
1458 BlockDriverState
*bs
= child
->opaque
;
1460 if (child
->role
& BDRV_CHILD_COW
) {
1461 bdrv_backing_detach(child
);
1464 assert_bdrv_graph_writable();
1465 QLIST_REMOVE(child
, next
);
1466 if (child
== bs
->backing
) {
1467 assert(child
!= bs
->file
);
1469 } else if (child
== bs
->file
) {
1474 static int bdrv_child_cb_update_filename(BdrvChild
*c
, BlockDriverState
*base
,
1475 const char *filename
, Error
**errp
)
1477 if (c
->role
& BDRV_CHILD_COW
) {
1478 return bdrv_backing_update_filename(c
, base
, filename
, errp
);
1483 AioContext
*child_of_bds_get_parent_aio_context(BdrvChild
*c
)
1485 BlockDriverState
*bs
= c
->opaque
;
1488 return bdrv_get_aio_context(bs
);
1491 const BdrvChildClass child_of_bds
= {
1492 .parent_is_bds
= true,
1493 .get_parent_desc
= bdrv_child_get_parent_desc
,
1494 .inherit_options
= bdrv_inherited_options
,
1495 .drained_begin
= bdrv_child_cb_drained_begin
,
1496 .drained_poll
= bdrv_child_cb_drained_poll
,
1497 .drained_end
= bdrv_child_cb_drained_end
,
1498 .attach
= bdrv_child_cb_attach
,
1499 .detach
= bdrv_child_cb_detach
,
1500 .inactivate
= bdrv_child_cb_inactivate
,
1501 .change_aio_ctx
= bdrv_child_cb_change_aio_ctx
,
1502 .update_filename
= bdrv_child_cb_update_filename
,
1503 .get_parent_aio_context
= child_of_bds_get_parent_aio_context
,
1506 AioContext
*bdrv_child_get_parent_aio_context(BdrvChild
*c
)
1509 return c
->klass
->get_parent_aio_context(c
);
1512 static int bdrv_open_flags(BlockDriverState
*bs
, int flags
)
1514 int open_flags
= flags
;
1515 GLOBAL_STATE_CODE();
1518 * Clear flags that are internal to the block layer before opening the
1521 open_flags
&= ~(BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
| BDRV_O_PROTOCOL
);
1526 static void update_flags_from_options(int *flags
, QemuOpts
*opts
)
1528 GLOBAL_STATE_CODE();
1530 *flags
&= ~(BDRV_O_CACHE_MASK
| BDRV_O_RDWR
| BDRV_O_AUTO_RDONLY
);
1532 if (qemu_opt_get_bool_del(opts
, BDRV_OPT_CACHE_NO_FLUSH
, false)) {
1533 *flags
|= BDRV_O_NO_FLUSH
;
1536 if (qemu_opt_get_bool_del(opts
, BDRV_OPT_CACHE_DIRECT
, false)) {
1537 *flags
|= BDRV_O_NOCACHE
;
1540 if (!qemu_opt_get_bool_del(opts
, BDRV_OPT_READ_ONLY
, false)) {
1541 *flags
|= BDRV_O_RDWR
;
1544 if (qemu_opt_get_bool_del(opts
, BDRV_OPT_AUTO_READ_ONLY
, false)) {
1545 *flags
|= BDRV_O_AUTO_RDONLY
;
1549 static void update_options_from_flags(QDict
*options
, int flags
)
1551 GLOBAL_STATE_CODE();
1552 if (!qdict_haskey(options
, BDRV_OPT_CACHE_DIRECT
)) {
1553 qdict_put_bool(options
, BDRV_OPT_CACHE_DIRECT
, flags
& BDRV_O_NOCACHE
);
1555 if (!qdict_haskey(options
, BDRV_OPT_CACHE_NO_FLUSH
)) {
1556 qdict_put_bool(options
, BDRV_OPT_CACHE_NO_FLUSH
,
1557 flags
& BDRV_O_NO_FLUSH
);
1559 if (!qdict_haskey(options
, BDRV_OPT_READ_ONLY
)) {
1560 qdict_put_bool(options
, BDRV_OPT_READ_ONLY
, !(flags
& BDRV_O_RDWR
));
1562 if (!qdict_haskey(options
, BDRV_OPT_AUTO_READ_ONLY
)) {
1563 qdict_put_bool(options
, BDRV_OPT_AUTO_READ_ONLY
,
1564 flags
& BDRV_O_AUTO_RDONLY
);
1568 static void bdrv_assign_node_name(BlockDriverState
*bs
,
1569 const char *node_name
,
1572 char *gen_node_name
= NULL
;
1573 GLOBAL_STATE_CODE();
1576 node_name
= gen_node_name
= id_generate(ID_BLOCK
);
1577 } else if (!id_wellformed(node_name
)) {
1579 * Check for empty string or invalid characters, but not if it is
1580 * generated (generated names use characters not available to the user)
1582 error_setg(errp
, "Invalid node-name: '%s'", node_name
);
1586 /* takes care of avoiding namespaces collisions */
1587 if (blk_by_name(node_name
)) {
1588 error_setg(errp
, "node-name=%s is conflicting with a device id",
1593 /* takes care of avoiding duplicates node names */
1594 if (bdrv_find_node(node_name
)) {
1595 error_setg(errp
, "Duplicate nodes with node-name='%s'", node_name
);
1599 /* Make sure that the node name isn't truncated */
1600 if (strlen(node_name
) >= sizeof(bs
->node_name
)) {
1601 error_setg(errp
, "Node name too long");
1605 /* copy node name into the bs and insert it into the graph list */
1606 pstrcpy(bs
->node_name
, sizeof(bs
->node_name
), node_name
);
1607 QTAILQ_INSERT_TAIL(&graph_bdrv_states
, bs
, node_list
);
1609 g_free(gen_node_name
);
1613 * The caller must always hold @bs AioContext lock, because this function calls
1614 * bdrv_refresh_total_sectors() which polls when called from non-coroutine
1617 static int no_coroutine_fn GRAPH_UNLOCKED
1618 bdrv_open_driver(BlockDriverState
*bs
, BlockDriver
*drv
, const char *node_name
,
1619 QDict
*options
, int open_flags
, Error
**errp
)
1622 Error
*local_err
= NULL
;
1624 GLOBAL_STATE_CODE();
1626 bdrv_assign_node_name(bs
, node_name
, &local_err
);
1628 error_propagate(errp
, local_err
);
1633 bs
->opaque
= g_malloc0(drv
->instance_size
);
1635 if (drv
->bdrv_file_open
) {
1636 assert(!drv
->bdrv_needs_filename
|| bs
->filename
[0]);
1637 ret
= drv
->bdrv_file_open(bs
, options
, open_flags
, &local_err
);
1638 } else if (drv
->bdrv_open
) {
1639 ret
= drv
->bdrv_open(bs
, options
, open_flags
, &local_err
);
1646 error_propagate(errp
, local_err
);
1647 } else if (bs
->filename
[0]) {
1648 error_setg_errno(errp
, -ret
, "Could not open '%s'", bs
->filename
);
1650 error_setg_errno(errp
, -ret
, "Could not open image");
1655 assert(!(bs
->supported_read_flags
& ~BDRV_REQ_MASK
));
1656 assert(!(bs
->supported_write_flags
& ~BDRV_REQ_MASK
));
1659 * Always allow the BDRV_REQ_REGISTERED_BUF optimization hint. This saves
1660 * drivers that pass read/write requests through to a child the trouble of
1661 * declaring support explicitly.
1663 * Drivers must not propagate this flag accidentally when they initiate I/O
1664 * to a bounce buffer. That case should be rare though.
1666 bs
->supported_read_flags
|= BDRV_REQ_REGISTERED_BUF
;
1667 bs
->supported_write_flags
|= BDRV_REQ_REGISTERED_BUF
;
1669 /* Get the context after .bdrv_open, it can change the context */
1670 ctx
= bdrv_get_aio_context(bs
);
1671 aio_context_acquire(ctx
);
1673 ret
= bdrv_refresh_total_sectors(bs
, bs
->total_sectors
);
1675 error_setg_errno(errp
, -ret
, "Could not refresh total sector count");
1676 aio_context_release(ctx
);
1680 bdrv_graph_rdlock_main_loop();
1681 bdrv_refresh_limits(bs
, NULL
, &local_err
);
1682 bdrv_graph_rdunlock_main_loop();
1683 aio_context_release(ctx
);
1686 error_propagate(errp
, local_err
);
1690 assert(bdrv_opt_mem_align(bs
) != 0);
1691 assert(bdrv_min_mem_align(bs
) != 0);
1692 assert(is_power_of_2(bs
->bl
.request_alignment
));
1694 for (i
= 0; i
< bs
->quiesce_counter
; i
++) {
1695 if (drv
->bdrv_drain_begin
) {
1696 drv
->bdrv_drain_begin(bs
);
1703 if (bs
->file
!= NULL
) {
1704 bdrv_unref_child(bs
, bs
->file
);
1713 * Create and open a block node.
1715 * @options is a QDict of options to pass to the block drivers, or NULL for an
1716 * empty set of options. The reference to the QDict belongs to the block layer
1717 * after the call (even on failure), so if the caller intends to reuse the
1718 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
1720 BlockDriverState
*bdrv_new_open_driver_opts(BlockDriver
*drv
,
1721 const char *node_name
,
1722 QDict
*options
, int flags
,
1725 BlockDriverState
*bs
;
1728 GLOBAL_STATE_CODE();
1731 bs
->open_flags
= flags
;
1732 bs
->options
= options
?: qdict_new();
1733 bs
->explicit_options
= qdict_clone_shallow(bs
->options
);
1736 update_options_from_flags(bs
->options
, flags
);
1738 ret
= bdrv_open_driver(bs
, drv
, node_name
, bs
->options
, flags
, errp
);
1740 qobject_unref(bs
->explicit_options
);
1741 bs
->explicit_options
= NULL
;
1742 qobject_unref(bs
->options
);
1751 /* Create and open a block node. */
1752 BlockDriverState
*bdrv_new_open_driver(BlockDriver
*drv
, const char *node_name
,
1753 int flags
, Error
**errp
)
1755 GLOBAL_STATE_CODE();
1756 return bdrv_new_open_driver_opts(drv
, node_name
, NULL
, flags
, errp
);
1759 QemuOptsList bdrv_runtime_opts
= {
1760 .name
= "bdrv_common",
1761 .head
= QTAILQ_HEAD_INITIALIZER(bdrv_runtime_opts
.head
),
1764 .name
= "node-name",
1765 .type
= QEMU_OPT_STRING
,
1766 .help
= "Node name of the block device node",
1770 .type
= QEMU_OPT_STRING
,
1771 .help
= "Block driver to use for the node",
1774 .name
= BDRV_OPT_CACHE_DIRECT
,
1775 .type
= QEMU_OPT_BOOL
,
1776 .help
= "Bypass software writeback cache on the host",
1779 .name
= BDRV_OPT_CACHE_NO_FLUSH
,
1780 .type
= QEMU_OPT_BOOL
,
1781 .help
= "Ignore flush requests",
1784 .name
= BDRV_OPT_READ_ONLY
,
1785 .type
= QEMU_OPT_BOOL
,
1786 .help
= "Node is opened in read-only mode",
1789 .name
= BDRV_OPT_AUTO_READ_ONLY
,
1790 .type
= QEMU_OPT_BOOL
,
1791 .help
= "Node can become read-only if opening read-write fails",
1794 .name
= "detect-zeroes",
1795 .type
= QEMU_OPT_STRING
,
1796 .help
= "try to optimize zero writes (off, on, unmap)",
1799 .name
= BDRV_OPT_DISCARD
,
1800 .type
= QEMU_OPT_STRING
,
1801 .help
= "discard operation (ignore/off, unmap/on)",
1804 .name
= BDRV_OPT_FORCE_SHARE
,
1805 .type
= QEMU_OPT_BOOL
,
1806 .help
= "always accept other writers (default: off)",
1808 { /* end of list */ }
1812 QemuOptsList bdrv_create_opts_simple
= {
1813 .name
= "simple-create-opts",
1814 .head
= QTAILQ_HEAD_INITIALIZER(bdrv_create_opts_simple
.head
),
1817 .name
= BLOCK_OPT_SIZE
,
1818 .type
= QEMU_OPT_SIZE
,
1819 .help
= "Virtual disk size"
1822 .name
= BLOCK_OPT_PREALLOC
,
1823 .type
= QEMU_OPT_STRING
,
1824 .help
= "Preallocation mode (allowed values: off)"
1826 { /* end of list */ }
1831 * Common part for opening disk images and files
1833 * Removes all processed options from *options.
1835 static int bdrv_open_common(BlockDriverState
*bs
, BlockBackend
*file
,
1836 QDict
*options
, Error
**errp
)
1838 int ret
, open_flags
;
1839 const char *filename
;
1840 const char *driver_name
= NULL
;
1841 const char *node_name
= NULL
;
1842 const char *discard
;
1845 Error
*local_err
= NULL
;
1848 assert(bs
->file
== NULL
);
1849 assert(options
!= NULL
&& bs
->options
!= options
);
1850 GLOBAL_STATE_CODE();
1852 opts
= qemu_opts_create(&bdrv_runtime_opts
, NULL
, 0, &error_abort
);
1853 if (!qemu_opts_absorb_qdict(opts
, options
, errp
)) {
1858 update_flags_from_options(&bs
->open_flags
, opts
);
1860 driver_name
= qemu_opt_get(opts
, "driver");
1861 drv
= bdrv_find_format(driver_name
);
1862 assert(drv
!= NULL
);
1864 bs
->force_share
= qemu_opt_get_bool(opts
, BDRV_OPT_FORCE_SHARE
, false);
1866 if (bs
->force_share
&& (bs
->open_flags
& BDRV_O_RDWR
)) {
1868 BDRV_OPT_FORCE_SHARE
1869 "=on can only be used with read-only images");
1875 bdrv_refresh_filename(blk_bs(file
));
1876 filename
= blk_bs(file
)->filename
;
1879 * Caution: while qdict_get_try_str() is fine, getting
1880 * non-string types would require more care. When @options
1881 * come from -blockdev or blockdev_add, its members are typed
1882 * according to the QAPI schema, but when they come from
1883 * -drive, they're all QString.
1885 filename
= qdict_get_try_str(options
, "filename");
1888 if (drv
->bdrv_needs_filename
&& (!filename
|| !filename
[0])) {
1889 error_setg(errp
, "The '%s' block driver requires a file name",
1895 trace_bdrv_open_common(bs
, filename
?: "", bs
->open_flags
,
1898 ro
= bdrv_is_read_only(bs
);
1900 if (use_bdrv_whitelist
&& !bdrv_is_whitelisted(drv
, ro
)) {
1901 if (!ro
&& bdrv_is_whitelisted(drv
, true)) {
1902 ret
= bdrv_apply_auto_read_only(bs
, NULL
, NULL
);
1908 !ro
&& bdrv_is_whitelisted(drv
, true)
1909 ? "Driver '%s' can only be used for read-only devices"
1910 : "Driver '%s' is not whitelisted",
1916 /* bdrv_new() and bdrv_close() make it so */
1917 assert(qatomic_read(&bs
->copy_on_read
) == 0);
1919 if (bs
->open_flags
& BDRV_O_COPY_ON_READ
) {
1921 bdrv_enable_copy_on_read(bs
);
1923 error_setg(errp
, "Can't use copy-on-read on read-only device");
1929 discard
= qemu_opt_get(opts
, BDRV_OPT_DISCARD
);
1930 if (discard
!= NULL
) {
1931 if (bdrv_parse_discard_flags(discard
, &bs
->open_flags
) != 0) {
1932 error_setg(errp
, "Invalid discard option");
1939 bdrv_parse_detect_zeroes(opts
, bs
->open_flags
, &local_err
);
1941 error_propagate(errp
, local_err
);
1946 if (filename
!= NULL
) {
1947 pstrcpy(bs
->filename
, sizeof(bs
->filename
), filename
);
1949 bs
->filename
[0] = '\0';
1951 pstrcpy(bs
->exact_filename
, sizeof(bs
->exact_filename
), bs
->filename
);
1953 /* Open the image, either directly or using a protocol */
1954 open_flags
= bdrv_open_flags(bs
, bs
->open_flags
);
1955 node_name
= qemu_opt_get(opts
, "node-name");
1957 assert(!drv
->bdrv_file_open
|| file
== NULL
);
1958 ret
= bdrv_open_driver(bs
, drv
, node_name
, options
, open_flags
, errp
);
1963 qemu_opts_del(opts
);
1967 qemu_opts_del(opts
);
1971 static QDict
*parse_json_filename(const char *filename
, Error
**errp
)
1973 QObject
*options_obj
;
1976 GLOBAL_STATE_CODE();
1978 ret
= strstart(filename
, "json:", &filename
);
1981 options_obj
= qobject_from_json(filename
, errp
);
1983 error_prepend(errp
, "Could not parse the JSON options: ");
1987 options
= qobject_to(QDict
, options_obj
);
1989 qobject_unref(options_obj
);
1990 error_setg(errp
, "Invalid JSON object given");
1994 qdict_flatten(options
);
1999 static void parse_json_protocol(QDict
*options
, const char **pfilename
,
2002 QDict
*json_options
;
2003 Error
*local_err
= NULL
;
2004 GLOBAL_STATE_CODE();
2006 /* Parse json: pseudo-protocol */
2007 if (!*pfilename
|| !g_str_has_prefix(*pfilename
, "json:")) {
2011 json_options
= parse_json_filename(*pfilename
, &local_err
);
2013 error_propagate(errp
, local_err
);
2017 /* Options given in the filename have lower priority than options
2018 * specified directly */
2019 qdict_join(options
, json_options
, false);
2020 qobject_unref(json_options
);
2025 * Fills in default options for opening images and converts the legacy
2026 * filename/flags pair to option QDict entries.
2027 * The BDRV_O_PROTOCOL flag in *flags will be set or cleared accordingly if a
2028 * block driver has been specified explicitly.
2030 static int bdrv_fill_options(QDict
**options
, const char *filename
,
2031 int *flags
, Error
**errp
)
2033 const char *drvname
;
2034 bool protocol
= *flags
& BDRV_O_PROTOCOL
;
2035 bool parse_filename
= false;
2036 BlockDriver
*drv
= NULL
;
2037 Error
*local_err
= NULL
;
2039 GLOBAL_STATE_CODE();
2042 * Caution: while qdict_get_try_str() is fine, getting non-string
2043 * types would require more care. When @options come from
2044 * -blockdev or blockdev_add, its members are typed according to
2045 * the QAPI schema, but when they come from -drive, they're all
2048 drvname
= qdict_get_try_str(*options
, "driver");
2050 drv
= bdrv_find_format(drvname
);
2052 error_setg(errp
, "Unknown driver '%s'", drvname
);
2055 /* If the user has explicitly specified the driver, this choice should
2056 * override the BDRV_O_PROTOCOL flag */
2057 protocol
= drv
->bdrv_file_open
;
2061 *flags
|= BDRV_O_PROTOCOL
;
2063 *flags
&= ~BDRV_O_PROTOCOL
;
2066 /* Translate cache options from flags into options */
2067 update_options_from_flags(*options
, *flags
);
2069 /* Fetch the file name from the options QDict if necessary */
2070 if (protocol
&& filename
) {
2071 if (!qdict_haskey(*options
, "filename")) {
2072 qdict_put_str(*options
, "filename", filename
);
2073 parse_filename
= true;
2075 error_setg(errp
, "Can't specify 'file' and 'filename' options at "
2081 /* Find the right block driver */
2082 /* See cautionary note on accessing @options above */
2083 filename
= qdict_get_try_str(*options
, "filename");
2085 if (!drvname
&& protocol
) {
2087 drv
= bdrv_find_protocol(filename
, parse_filename
, errp
);
2092 drvname
= drv
->format_name
;
2093 qdict_put_str(*options
, "driver", drvname
);
2095 error_setg(errp
, "Must specify either driver or file");
2100 assert(drv
|| !protocol
);
2102 /* Driver-specific filename parsing */
2103 if (drv
&& drv
->bdrv_parse_filename
&& parse_filename
) {
2104 drv
->bdrv_parse_filename(filename
, *options
, &local_err
);
2106 error_propagate(errp
, local_err
);
2110 if (!drv
->bdrv_needs_filename
) {
2111 qdict_del(*options
, "filename");
2118 typedef struct BlockReopenQueueEntry
{
2120 BDRVReopenState state
;
2121 QTAILQ_ENTRY(BlockReopenQueueEntry
) entry
;
2122 } BlockReopenQueueEntry
;
2125 * Return the flags that @bs will have after the reopens in @q have
2126 * successfully completed. If @q is NULL (or @bs is not contained in @q),
2127 * return the current flags.
2129 static int bdrv_reopen_get_flags(BlockReopenQueue
*q
, BlockDriverState
*bs
)
2131 BlockReopenQueueEntry
*entry
;
2134 QTAILQ_FOREACH(entry
, q
, entry
) {
2135 if (entry
->state
.bs
== bs
) {
2136 return entry
->state
.flags
;
2141 return bs
->open_flags
;
2144 /* Returns whether the image file can be written to after the reopen queue @q
2145 * has been successfully applied, or right now if @q is NULL. */
2146 static bool bdrv_is_writable_after_reopen(BlockDriverState
*bs
,
2147 BlockReopenQueue
*q
)
2149 int flags
= bdrv_reopen_get_flags(q
, bs
);
2151 return (flags
& (BDRV_O_RDWR
| BDRV_O_INACTIVE
)) == BDRV_O_RDWR
;
2155 * Return whether the BDS can be written to. This is not necessarily
2156 * the same as !bdrv_is_read_only(bs), as inactivated images may not
2157 * be written to but do not count as read-only images.
2159 bool bdrv_is_writable(BlockDriverState
*bs
)
2162 return bdrv_is_writable_after_reopen(bs
, NULL
);
2165 static char *bdrv_child_user_desc(BdrvChild
*c
)
2167 GLOBAL_STATE_CODE();
2168 return c
->klass
->get_parent_desc(c
);
2172 * Check that @a allows everything that @b needs. @a and @b must reference same
2175 static bool bdrv_a_allow_b(BdrvChild
*a
, BdrvChild
*b
, Error
**errp
)
2177 const char *child_bs_name
;
2178 g_autofree
char *a_user
= NULL
;
2179 g_autofree
char *b_user
= NULL
;
2180 g_autofree
char *perms
= NULL
;
2183 assert(a
->bs
== b
->bs
);
2184 GLOBAL_STATE_CODE();
2186 if ((b
->perm
& a
->shared_perm
) == b
->perm
) {
2190 child_bs_name
= bdrv_get_node_name(b
->bs
);
2191 a_user
= bdrv_child_user_desc(a
);
2192 b_user
= bdrv_child_user_desc(b
);
2193 perms
= bdrv_perm_names(b
->perm
& ~a
->shared_perm
);
2195 error_setg(errp
, "Permission conflict on node '%s': permissions '%s' are "
2196 "both required by %s (uses node '%s' as '%s' child) and "
2197 "unshared by %s (uses node '%s' as '%s' child).",
2198 child_bs_name
, perms
,
2199 b_user
, child_bs_name
, b
->name
,
2200 a_user
, child_bs_name
, a
->name
);
2205 static bool GRAPH_RDLOCK
2206 bdrv_parent_perms_conflict(BlockDriverState
*bs
, Error
**errp
)
2209 GLOBAL_STATE_CODE();
2212 * During the loop we'll look at each pair twice. That's correct because
2213 * bdrv_a_allow_b() is asymmetric and we should check each pair in both
2216 QLIST_FOREACH(a
, &bs
->parents
, next_parent
) {
2217 QLIST_FOREACH(b
, &bs
->parents
, next_parent
) {
2222 if (!bdrv_a_allow_b(a
, b
, errp
)) {
2231 static void GRAPH_RDLOCK
2232 bdrv_child_perm(BlockDriverState
*bs
, BlockDriverState
*child_bs
,
2233 BdrvChild
*c
, BdrvChildRole role
,
2234 BlockReopenQueue
*reopen_queue
,
2235 uint64_t parent_perm
, uint64_t parent_shared
,
2236 uint64_t *nperm
, uint64_t *nshared
)
2238 assert(bs
->drv
&& bs
->drv
->bdrv_child_perm
);
2239 GLOBAL_STATE_CODE();
2240 bs
->drv
->bdrv_child_perm(bs
, c
, role
, reopen_queue
,
2241 parent_perm
, parent_shared
,
2243 /* TODO Take force_share from reopen_queue */
2244 if (child_bs
&& child_bs
->force_share
) {
2245 *nshared
= BLK_PERM_ALL
;
2250 * Adds the whole subtree of @bs (including @bs itself) to the @list (except for
2251 * nodes that are already in the @list, of course) so that final list is
2252 * topologically sorted. Return the result (GSList @list object is updated, so
2253 * don't use old reference after function call).
2255 * On function start @list must be already topologically sorted and for any node
2256 * in the @list the whole subtree of the node must be in the @list as well. The
2257 * simplest way to satisfy this criteria: use only result of
2258 * bdrv_topological_dfs() or NULL as @list parameter.
2260 static GSList
* GRAPH_RDLOCK
2261 bdrv_topological_dfs(GSList
*list
, GHashTable
*found
, BlockDriverState
*bs
)
2264 g_autoptr(GHashTable
) local_found
= NULL
;
2266 GLOBAL_STATE_CODE();
2270 found
= local_found
= g_hash_table_new(NULL
, NULL
);
2273 if (g_hash_table_contains(found
, bs
)) {
2276 g_hash_table_add(found
, bs
);
2278 QLIST_FOREACH(child
, &bs
->children
, next
) {
2279 list
= bdrv_topological_dfs(list
, found
, child
->bs
);
2282 return g_slist_prepend(list
, bs
);
2285 typedef struct BdrvChildSetPermState
{
2288 uint64_t old_shared_perm
;
2289 } BdrvChildSetPermState
;
2291 static void bdrv_child_set_perm_abort(void *opaque
)
2293 BdrvChildSetPermState
*s
= opaque
;
2295 GLOBAL_STATE_CODE();
2297 s
->child
->perm
= s
->old_perm
;
2298 s
->child
->shared_perm
= s
->old_shared_perm
;
2301 static TransactionActionDrv bdrv_child_set_pem_drv
= {
2302 .abort
= bdrv_child_set_perm_abort
,
2306 static void bdrv_child_set_perm(BdrvChild
*c
, uint64_t perm
,
2307 uint64_t shared
, Transaction
*tran
)
2309 BdrvChildSetPermState
*s
= g_new(BdrvChildSetPermState
, 1);
2310 GLOBAL_STATE_CODE();
2312 *s
= (BdrvChildSetPermState
) {
2314 .old_perm
= c
->perm
,
2315 .old_shared_perm
= c
->shared_perm
,
2319 c
->shared_perm
= shared
;
2321 tran_add(tran
, &bdrv_child_set_pem_drv
, s
);
2324 static void GRAPH_RDLOCK
bdrv_drv_set_perm_commit(void *opaque
)
2326 BlockDriverState
*bs
= opaque
;
2327 uint64_t cumulative_perms
, cumulative_shared_perms
;
2328 GLOBAL_STATE_CODE();
2330 if (bs
->drv
->bdrv_set_perm
) {
2331 bdrv_get_cumulative_perm(bs
, &cumulative_perms
,
2332 &cumulative_shared_perms
);
2333 bs
->drv
->bdrv_set_perm(bs
, cumulative_perms
, cumulative_shared_perms
);
2337 static void GRAPH_RDLOCK
bdrv_drv_set_perm_abort(void *opaque
)
2339 BlockDriverState
*bs
= opaque
;
2340 GLOBAL_STATE_CODE();
2342 if (bs
->drv
->bdrv_abort_perm_update
) {
2343 bs
->drv
->bdrv_abort_perm_update(bs
);
2347 TransactionActionDrv bdrv_drv_set_perm_drv
= {
2348 .abort
= bdrv_drv_set_perm_abort
,
2349 .commit
= bdrv_drv_set_perm_commit
,
2353 * After calling this function, the transaction @tran may only be completed
2354 * while holding a reader lock for the graph.
2356 static int GRAPH_RDLOCK
2357 bdrv_drv_set_perm(BlockDriverState
*bs
, uint64_t perm
, uint64_t shared_perm
,
2358 Transaction
*tran
, Error
**errp
)
2360 GLOBAL_STATE_CODE();
2365 if (bs
->drv
->bdrv_check_perm
) {
2366 int ret
= bs
->drv
->bdrv_check_perm(bs
, perm
, shared_perm
, errp
);
2373 tran_add(tran
, &bdrv_drv_set_perm_drv
, bs
);
2379 typedef struct BdrvReplaceChildState
{
2381 BlockDriverState
*old_bs
;
2382 } BdrvReplaceChildState
;
2384 static void GRAPH_WRLOCK
bdrv_replace_child_commit(void *opaque
)
2386 BdrvReplaceChildState
*s
= opaque
;
2387 GLOBAL_STATE_CODE();
2389 bdrv_schedule_unref(s
->old_bs
);
2392 static void GRAPH_WRLOCK
bdrv_replace_child_abort(void *opaque
)
2394 BdrvReplaceChildState
*s
= opaque
;
2395 BlockDriverState
*new_bs
= s
->child
->bs
;
2397 GLOBAL_STATE_CODE();
2398 assert_bdrv_graph_writable();
2400 /* old_bs reference is transparently moved from @s to @s->child */
2401 if (!s
->child
->bs
) {
2403 * The parents were undrained when removing old_bs from the child. New
2404 * requests can't have been made, though, because the child was empty.
2406 * TODO Make bdrv_replace_child_noperm() transactionable to avoid
2407 * undraining the parent in the first place. Once this is done, having
2408 * new_bs drained when calling bdrv_replace_child_tran() is not a
2409 * requirement any more.
2411 bdrv_parent_drained_begin_single(s
->child
);
2412 assert(!bdrv_parent_drained_poll_single(s
->child
));
2414 assert(s
->child
->quiesced_parent
);
2415 bdrv_replace_child_noperm(s
->child
, s
->old_bs
);
2420 static TransactionActionDrv bdrv_replace_child_drv
= {
2421 .commit
= bdrv_replace_child_commit
,
2422 .abort
= bdrv_replace_child_abort
,
2427 * bdrv_replace_child_tran
2429 * Note: real unref of old_bs is done only on commit.
2431 * Both @child->bs and @new_bs (if non-NULL) must be drained. @new_bs must be
2432 * kept drained until the transaction is completed.
2434 * After calling this function, the transaction @tran may only be completed
2435 * while holding a writer lock for the graph.
2437 * The function doesn't update permissions, caller is responsible for this.
2439 static void GRAPH_WRLOCK
2440 bdrv_replace_child_tran(BdrvChild
*child
, BlockDriverState
*new_bs
,
2443 BdrvReplaceChildState
*s
= g_new(BdrvReplaceChildState
, 1);
2445 assert(child
->quiesced_parent
);
2446 assert(!new_bs
|| new_bs
->quiesce_counter
);
2448 *s
= (BdrvReplaceChildState
) {
2450 .old_bs
= child
->bs
,
2452 tran_add(tran
, &bdrv_replace_child_drv
, s
);
2458 bdrv_replace_child_noperm(child
, new_bs
);
2459 /* old_bs reference is transparently moved from @child to @s */
2463 * Refresh permissions in @bs subtree. The function is intended to be called
2464 * after some graph modification that was done without permission update.
2466 * After calling this function, the transaction @tran may only be completed
2467 * while holding a reader lock for the graph.
2469 static int GRAPH_RDLOCK
2470 bdrv_node_refresh_perm(BlockDriverState
*bs
, BlockReopenQueue
*q
,
2471 Transaction
*tran
, Error
**errp
)
2473 BlockDriver
*drv
= bs
->drv
;
2476 uint64_t cumulative_perms
, cumulative_shared_perms
;
2477 GLOBAL_STATE_CODE();
2479 bdrv_get_cumulative_perm(bs
, &cumulative_perms
, &cumulative_shared_perms
);
2481 /* Write permissions never work with read-only images */
2482 if ((cumulative_perms
& (BLK_PERM_WRITE
| BLK_PERM_WRITE_UNCHANGED
)) &&
2483 !bdrv_is_writable_after_reopen(bs
, q
))
2485 if (!bdrv_is_writable_after_reopen(bs
, NULL
)) {
2486 error_setg(errp
, "Block node is read-only");
2488 error_setg(errp
, "Read-only block node '%s' cannot support "
2489 "read-write users", bdrv_get_node_name(bs
));
2496 * Unaligned requests will automatically be aligned to bl.request_alignment
2497 * and without RESIZE we can't extend requests to write to space beyond the
2498 * end of the image, so it's required that the image size is aligned.
2500 if ((cumulative_perms
& (BLK_PERM_WRITE
| BLK_PERM_WRITE_UNCHANGED
)) &&
2501 !(cumulative_perms
& BLK_PERM_RESIZE
))
2503 if ((bs
->total_sectors
* BDRV_SECTOR_SIZE
) % bs
->bl
.request_alignment
) {
2504 error_setg(errp
, "Cannot get 'write' permission without 'resize': "
2505 "Image size is not a multiple of request "
2511 /* Check this node */
2516 ret
= bdrv_drv_set_perm(bs
, cumulative_perms
, cumulative_shared_perms
, tran
,
2522 /* Drivers that never have children can omit .bdrv_child_perm() */
2523 if (!drv
->bdrv_child_perm
) {
2524 assert(QLIST_EMPTY(&bs
->children
));
2528 /* Check all children */
2529 QLIST_FOREACH(c
, &bs
->children
, next
) {
2530 uint64_t cur_perm
, cur_shared
;
2532 bdrv_child_perm(bs
, c
->bs
, c
, c
->role
, q
,
2533 cumulative_perms
, cumulative_shared_perms
,
2534 &cur_perm
, &cur_shared
);
2535 bdrv_child_set_perm(c
, cur_perm
, cur_shared
, tran
);
2542 * @list is a product of bdrv_topological_dfs() (may be called several times) -
2543 * a topologically sorted subgraph.
2545 * After calling this function, the transaction @tran may only be completed
2546 * while holding a reader lock for the graph.
2548 static int GRAPH_RDLOCK
2549 bdrv_do_refresh_perms(GSList
*list
, BlockReopenQueue
*q
, Transaction
*tran
,
2553 BlockDriverState
*bs
;
2554 GLOBAL_STATE_CODE();
2556 for ( ; list
; list
= list
->next
) {
2559 if (bdrv_parent_perms_conflict(bs
, errp
)) {
2563 ret
= bdrv_node_refresh_perm(bs
, q
, tran
, errp
);
2573 * @list is any list of nodes. List is completed by all subtrees and
2574 * topologically sorted. It's not a problem if some node occurs in the @list
2577 * After calling this function, the transaction @tran may only be completed
2578 * while holding a reader lock for the graph.
2580 static int GRAPH_RDLOCK
2581 bdrv_list_refresh_perms(GSList
*list
, BlockReopenQueue
*q
, Transaction
*tran
,
2584 g_autoptr(GHashTable
) found
= g_hash_table_new(NULL
, NULL
);
2585 g_autoptr(GSList
) refresh_list
= NULL
;
2587 for ( ; list
; list
= list
->next
) {
2588 refresh_list
= bdrv_topological_dfs(refresh_list
, found
, list
->data
);
2591 return bdrv_do_refresh_perms(refresh_list
, q
, tran
, errp
);
2594 void bdrv_get_cumulative_perm(BlockDriverState
*bs
, uint64_t *perm
,
2595 uint64_t *shared_perm
)
2598 uint64_t cumulative_perms
= 0;
2599 uint64_t cumulative_shared_perms
= BLK_PERM_ALL
;
2601 GLOBAL_STATE_CODE();
2603 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
2604 cumulative_perms
|= c
->perm
;
2605 cumulative_shared_perms
&= c
->shared_perm
;
2608 *perm
= cumulative_perms
;
2609 *shared_perm
= cumulative_shared_perms
;
2612 char *bdrv_perm_names(uint64_t perm
)
2618 { BLK_PERM_CONSISTENT_READ
, "consistent read" },
2619 { BLK_PERM_WRITE
, "write" },
2620 { BLK_PERM_WRITE_UNCHANGED
, "write unchanged" },
2621 { BLK_PERM_RESIZE
, "resize" },
2625 GString
*result
= g_string_sized_new(30);
2626 struct perm_name
*p
;
2628 for (p
= permissions
; p
->name
; p
++) {
2629 if (perm
& p
->perm
) {
2630 if (result
->len
> 0) {
2631 g_string_append(result
, ", ");
2633 g_string_append(result
, p
->name
);
2637 return g_string_free(result
, FALSE
);
2642 * @tran is allowed to be NULL. In this case no rollback is possible.
2644 * After calling this function, the transaction @tran may only be completed
2645 * while holding a reader lock for the graph.
2647 static int GRAPH_RDLOCK
2648 bdrv_refresh_perms(BlockDriverState
*bs
, Transaction
*tran
, Error
**errp
)
2651 Transaction
*local_tran
= NULL
;
2652 g_autoptr(GSList
) list
= bdrv_topological_dfs(NULL
, NULL
, bs
);
2653 GLOBAL_STATE_CODE();
2656 tran
= local_tran
= tran_new();
2659 ret
= bdrv_do_refresh_perms(list
, NULL
, tran
, errp
);
2662 tran_finalize(local_tran
, ret
);
2668 int bdrv_child_try_set_perm(BdrvChild
*c
, uint64_t perm
, uint64_t shared
,
2671 Error
*local_err
= NULL
;
2672 Transaction
*tran
= tran_new();
2675 GLOBAL_STATE_CODE();
2677 bdrv_child_set_perm(c
, perm
, shared
, tran
);
2679 ret
= bdrv_refresh_perms(c
->bs
, tran
, &local_err
);
2681 tran_finalize(tran
, ret
);
2684 if ((perm
& ~c
->perm
) || (c
->shared_perm
& ~shared
)) {
2685 /* tighten permissions */
2686 error_propagate(errp
, local_err
);
2689 * Our caller may intend to only loosen restrictions and
2690 * does not expect this function to fail. Errors are not
2691 * fatal in such a case, so we can just hide them from our
2694 error_free(local_err
);
2702 int bdrv_child_refresh_perms(BlockDriverState
*bs
, BdrvChild
*c
, Error
**errp
)
2704 uint64_t parent_perms
, parent_shared
;
2705 uint64_t perms
, shared
;
2707 GLOBAL_STATE_CODE();
2709 bdrv_get_cumulative_perm(bs
, &parent_perms
, &parent_shared
);
2710 bdrv_child_perm(bs
, c
->bs
, c
, c
->role
, NULL
,
2711 parent_perms
, parent_shared
, &perms
, &shared
);
2713 return bdrv_child_try_set_perm(c
, perms
, shared
, errp
);
2717 * Default implementation for .bdrv_child_perm() for block filters:
2718 * Forward CONSISTENT_READ, WRITE, WRITE_UNCHANGED, and RESIZE to the
2721 static void bdrv_filter_default_perms(BlockDriverState
*bs
, BdrvChild
*c
,
2723 BlockReopenQueue
*reopen_queue
,
2724 uint64_t perm
, uint64_t shared
,
2725 uint64_t *nperm
, uint64_t *nshared
)
2727 GLOBAL_STATE_CODE();
2728 *nperm
= perm
& DEFAULT_PERM_PASSTHROUGH
;
2729 *nshared
= (shared
& DEFAULT_PERM_PASSTHROUGH
) | DEFAULT_PERM_UNCHANGED
;
2732 static void bdrv_default_perms_for_cow(BlockDriverState
*bs
, BdrvChild
*c
,
2734 BlockReopenQueue
*reopen_queue
,
2735 uint64_t perm
, uint64_t shared
,
2736 uint64_t *nperm
, uint64_t *nshared
)
2738 assert(role
& BDRV_CHILD_COW
);
2739 GLOBAL_STATE_CODE();
2742 * We want consistent read from backing files if the parent needs it.
2743 * No other operations are performed on backing files.
2745 perm
&= BLK_PERM_CONSISTENT_READ
;
2748 * If the parent can deal with changing data, we're okay with a
2749 * writable and resizable backing file.
2750 * TODO Require !(perm & BLK_PERM_CONSISTENT_READ), too?
2752 if (shared
& BLK_PERM_WRITE
) {
2753 shared
= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2758 shared
|= BLK_PERM_CONSISTENT_READ
| BLK_PERM_WRITE_UNCHANGED
;
2760 if (bs
->open_flags
& BDRV_O_INACTIVE
) {
2761 shared
|= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2768 static void bdrv_default_perms_for_storage(BlockDriverState
*bs
, BdrvChild
*c
,
2770 BlockReopenQueue
*reopen_queue
,
2771 uint64_t perm
, uint64_t shared
,
2772 uint64_t *nperm
, uint64_t *nshared
)
2776 GLOBAL_STATE_CODE();
2777 assert(role
& (BDRV_CHILD_METADATA
| BDRV_CHILD_DATA
));
2779 flags
= bdrv_reopen_get_flags(reopen_queue
, bs
);
2782 * Apart from the modifications below, the same permissions are
2783 * forwarded and left alone as for filters
2785 bdrv_filter_default_perms(bs
, c
, role
, reopen_queue
,
2786 perm
, shared
, &perm
, &shared
);
2788 if (role
& BDRV_CHILD_METADATA
) {
2789 /* Format drivers may touch metadata even if the guest doesn't write */
2790 if (bdrv_is_writable_after_reopen(bs
, reopen_queue
)) {
2791 perm
|= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2795 * bs->file always needs to be consistent because of the
2796 * metadata. We can never allow other users to resize or write
2799 if (!(flags
& BDRV_O_NO_IO
)) {
2800 perm
|= BLK_PERM_CONSISTENT_READ
;
2802 shared
&= ~(BLK_PERM_WRITE
| BLK_PERM_RESIZE
);
2805 if (role
& BDRV_CHILD_DATA
) {
2807 * Technically, everything in this block is a subset of the
2808 * BDRV_CHILD_METADATA path taken above, and so this could
2809 * be an "else if" branch. However, that is not obvious, and
2810 * this function is not performance critical, therefore we let
2811 * this be an independent "if".
2815 * We cannot allow other users to resize the file because the
2816 * format driver might have some assumptions about the size
2817 * (e.g. because it is stored in metadata, or because the file
2818 * is split into fixed-size data files).
2820 shared
&= ~BLK_PERM_RESIZE
;
2823 * WRITE_UNCHANGED often cannot be performed as such on the
2824 * data file. For example, the qcow2 driver may still need to
2825 * write copied clusters on copy-on-read.
2827 if (perm
& BLK_PERM_WRITE_UNCHANGED
) {
2828 perm
|= BLK_PERM_WRITE
;
2832 * If the data file is written to, the format driver may
2833 * expect to be able to resize it by writing beyond the EOF.
2835 if (perm
& BLK_PERM_WRITE
) {
2836 perm
|= BLK_PERM_RESIZE
;
2840 if (bs
->open_flags
& BDRV_O_INACTIVE
) {
2841 shared
|= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2848 void bdrv_default_perms(BlockDriverState
*bs
, BdrvChild
*c
,
2849 BdrvChildRole role
, BlockReopenQueue
*reopen_queue
,
2850 uint64_t perm
, uint64_t shared
,
2851 uint64_t *nperm
, uint64_t *nshared
)
2853 GLOBAL_STATE_CODE();
2854 if (role
& BDRV_CHILD_FILTERED
) {
2855 assert(!(role
& (BDRV_CHILD_DATA
| BDRV_CHILD_METADATA
|
2857 bdrv_filter_default_perms(bs
, c
, role
, reopen_queue
,
2858 perm
, shared
, nperm
, nshared
);
2859 } else if (role
& BDRV_CHILD_COW
) {
2860 assert(!(role
& (BDRV_CHILD_DATA
| BDRV_CHILD_METADATA
)));
2861 bdrv_default_perms_for_cow(bs
, c
, role
, reopen_queue
,
2862 perm
, shared
, nperm
, nshared
);
2863 } else if (role
& (BDRV_CHILD_METADATA
| BDRV_CHILD_DATA
)) {
2864 bdrv_default_perms_for_storage(bs
, c
, role
, reopen_queue
,
2865 perm
, shared
, nperm
, nshared
);
2867 g_assert_not_reached();
2871 uint64_t bdrv_qapi_perm_to_blk_perm(BlockPermission qapi_perm
)
2873 static const uint64_t permissions
[] = {
2874 [BLOCK_PERMISSION_CONSISTENT_READ
] = BLK_PERM_CONSISTENT_READ
,
2875 [BLOCK_PERMISSION_WRITE
] = BLK_PERM_WRITE
,
2876 [BLOCK_PERMISSION_WRITE_UNCHANGED
] = BLK_PERM_WRITE_UNCHANGED
,
2877 [BLOCK_PERMISSION_RESIZE
] = BLK_PERM_RESIZE
,
2880 QEMU_BUILD_BUG_ON(ARRAY_SIZE(permissions
) != BLOCK_PERMISSION__MAX
);
2881 QEMU_BUILD_BUG_ON(1UL << ARRAY_SIZE(permissions
) != BLK_PERM_ALL
+ 1);
2883 assert(qapi_perm
< BLOCK_PERMISSION__MAX
);
2885 return permissions
[qapi_perm
];
2889 * Replaces the node that a BdrvChild points to without updating permissions.
2891 * If @new_bs is non-NULL, the parent of @child must already be drained through
2892 * @child and the caller must hold the AioContext lock for @new_bs.
2894 static void GRAPH_WRLOCK
2895 bdrv_replace_child_noperm(BdrvChild
*child
, BlockDriverState
*new_bs
)
2897 BlockDriverState
*old_bs
= child
->bs
;
2898 int new_bs_quiesce_counter
;
2900 assert(!child
->frozen
);
2903 * If we want to change the BdrvChild to point to a drained node as its new
2904 * child->bs, we need to make sure that its new parent is drained, too. In
2905 * other words, either child->quiesce_parent must already be true or we must
2906 * be able to set it and keep the parent's quiesce_counter consistent with
2907 * that, but without polling or starting new requests (this function
2908 * guarantees that it doesn't poll, and starting new requests would be
2909 * against the invariants of drain sections).
2911 * To keep things simple, we pick the first option (child->quiesce_parent
2912 * must already be true). We also generalise the rule a bit to make it
2913 * easier to verify in callers and more likely to be covered in test cases:
2914 * The parent must be quiesced through this child even if new_bs isn't
2915 * currently drained.
2917 * The only exception is for callers that always pass new_bs == NULL. In
2918 * this case, we obviously never need to consider the case of a drained
2919 * new_bs, so we can keep the callers simpler by allowing them not to drain
2922 assert(!new_bs
|| child
->quiesced_parent
);
2923 assert(old_bs
!= new_bs
);
2924 GLOBAL_STATE_CODE();
2926 if (old_bs
&& new_bs
) {
2927 assert(bdrv_get_aio_context(old_bs
) == bdrv_get_aio_context(new_bs
));
2931 if (child
->klass
->detach
) {
2932 child
->klass
->detach(child
);
2934 QLIST_REMOVE(child
, next_parent
);
2940 QLIST_INSERT_HEAD(&new_bs
->parents
, child
, next_parent
);
2941 if (child
->klass
->attach
) {
2942 child
->klass
->attach(child
);
2947 * If the parent was drained through this BdrvChild previously, but new_bs
2948 * is not drained, allow requests to come in only after the new node has
2951 new_bs_quiesce_counter
= (new_bs
? new_bs
->quiesce_counter
: 0);
2952 if (!new_bs_quiesce_counter
&& child
->quiesced_parent
) {
2953 bdrv_parent_drained_end_single(child
);
2958 * Free the given @child.
2960 * The child must be empty (i.e. `child->bs == NULL`) and it must be
2961 * unused (i.e. not in a children list).
2963 static void bdrv_child_free(BdrvChild
*child
)
2966 GLOBAL_STATE_CODE();
2967 assert(!child
->next
.le_prev
); /* not in children list */
2969 g_free(child
->name
);
2973 typedef struct BdrvAttachChildCommonState
{
2975 AioContext
*old_parent_ctx
;
2976 AioContext
*old_child_ctx
;
2977 } BdrvAttachChildCommonState
;
2979 static void GRAPH_WRLOCK
bdrv_attach_child_common_abort(void *opaque
)
2981 BdrvAttachChildCommonState
*s
= opaque
;
2982 BlockDriverState
*bs
= s
->child
->bs
;
2984 GLOBAL_STATE_CODE();
2985 assert_bdrv_graph_writable();
2987 bdrv_replace_child_noperm(s
->child
, NULL
);
2989 if (bdrv_get_aio_context(bs
) != s
->old_child_ctx
) {
2990 bdrv_try_change_aio_context(bs
, s
->old_child_ctx
, NULL
, &error_abort
);
2993 if (bdrv_child_get_parent_aio_context(s
->child
) != s
->old_parent_ctx
) {
2995 GHashTable
*visited
;
3000 /* No need to visit `child`, because it has been detached already */
3001 visited
= g_hash_table_new(NULL
, NULL
);
3002 ret
= s
->child
->klass
->change_aio_ctx(s
->child
, s
->old_parent_ctx
,
3003 visited
, tran
, &error_abort
);
3004 g_hash_table_destroy(visited
);
3006 /* transaction is supposed to always succeed */
3007 assert(ret
== true);
3011 bdrv_schedule_unref(bs
);
3012 bdrv_child_free(s
->child
);
3015 static TransactionActionDrv bdrv_attach_child_common_drv
= {
3016 .abort
= bdrv_attach_child_common_abort
,
3021 * Common part of attaching bdrv child to bs or to blk or to job
3023 * Function doesn't update permissions, caller is responsible for this.
3025 * After calling this function, the transaction @tran may only be completed
3026 * while holding a writer lock for the graph.
3028 * Returns new created child.
3030 * The caller must hold the AioContext lock for @child_bs. Both @parent_bs and
3031 * @child_bs can move to a different AioContext in this function. Callers must
3032 * make sure that their AioContext locking is still correct after this.
3034 static BdrvChild
* GRAPH_WRLOCK
3035 bdrv_attach_child_common(BlockDriverState
*child_bs
,
3036 const char *child_name
,
3037 const BdrvChildClass
*child_class
,
3038 BdrvChildRole child_role
,
3039 uint64_t perm
, uint64_t shared_perm
,
3041 Transaction
*tran
, Error
**errp
)
3043 BdrvChild
*new_child
;
3044 AioContext
*parent_ctx
, *new_child_ctx
;
3045 AioContext
*child_ctx
= bdrv_get_aio_context(child_bs
);
3047 assert(child_class
->get_parent_desc
);
3048 GLOBAL_STATE_CODE();
3050 new_child
= g_new(BdrvChild
, 1);
3051 *new_child
= (BdrvChild
) {
3053 .name
= g_strdup(child_name
),
3054 .klass
= child_class
,
3057 .shared_perm
= shared_perm
,
3062 * If the AioContexts don't match, first try to move the subtree of
3063 * child_bs into the AioContext of the new parent. If this doesn't work,
3064 * try moving the parent into the AioContext of child_bs instead.
3066 parent_ctx
= bdrv_child_get_parent_aio_context(new_child
);
3067 if (child_ctx
!= parent_ctx
) {
3068 Error
*local_err
= NULL
;
3069 int ret
= bdrv_try_change_aio_context(child_bs
, parent_ctx
, NULL
,
3072 if (ret
< 0 && child_class
->change_aio_ctx
) {
3073 Transaction
*tran
= tran_new();
3074 GHashTable
*visited
= g_hash_table_new(NULL
, NULL
);
3077 g_hash_table_add(visited
, new_child
);
3078 ret_child
= child_class
->change_aio_ctx(new_child
, child_ctx
,
3079 visited
, tran
, NULL
);
3080 if (ret_child
== true) {
3081 error_free(local_err
);
3084 tran_finalize(tran
, ret_child
== true ? 0 : -1);
3085 g_hash_table_destroy(visited
);
3089 error_propagate(errp
, local_err
);
3090 bdrv_child_free(new_child
);
3095 new_child_ctx
= bdrv_get_aio_context(child_bs
);
3096 if (new_child_ctx
!= child_ctx
) {
3097 aio_context_release(child_ctx
);
3098 aio_context_acquire(new_child_ctx
);
3103 * Let every new BdrvChild start with a drained parent. Inserting the child
3104 * in the graph with bdrv_replace_child_noperm() will undrain it if
3105 * @child_bs is not drained.
3107 * The child was only just created and is not yet visible in global state
3108 * until bdrv_replace_child_noperm() inserts it into the graph, so nobody
3109 * could have sent requests and polling is not necessary.
3111 * Note that this means that the parent isn't fully drained yet, we only
3112 * stop new requests from coming in. This is fine, we don't care about the
3113 * old requests here, they are not for this child. If another place enters a
3114 * drain section for the same parent, but wants it to be fully quiesced, it
3115 * will not run most of the the code in .drained_begin() again (which is not
3116 * a problem, we already did this), but it will still poll until the parent
3117 * is fully quiesced, so it will not be negatively affected either.
3119 bdrv_parent_drained_begin_single(new_child
);
3120 bdrv_replace_child_noperm(new_child
, child_bs
);
3122 BdrvAttachChildCommonState
*s
= g_new(BdrvAttachChildCommonState
, 1);
3123 *s
= (BdrvAttachChildCommonState
) {
3125 .old_parent_ctx
= parent_ctx
,
3126 .old_child_ctx
= child_ctx
,
3128 tran_add(tran
, &bdrv_attach_child_common_drv
, s
);
3130 if (new_child_ctx
!= child_ctx
) {
3131 aio_context_release(new_child_ctx
);
3132 aio_context_acquire(child_ctx
);
3139 * Function doesn't update permissions, caller is responsible for this.
3141 * The caller must hold the AioContext lock for @child_bs. Both @parent_bs and
3142 * @child_bs can move to a different AioContext in this function. Callers must
3143 * make sure that their AioContext locking is still correct after this.
3145 * After calling this function, the transaction @tran may only be completed
3146 * while holding a writer lock for the graph.
3148 static BdrvChild
* GRAPH_WRLOCK
3149 bdrv_attach_child_noperm(BlockDriverState
*parent_bs
,
3150 BlockDriverState
*child_bs
,
3151 const char *child_name
,
3152 const BdrvChildClass
*child_class
,
3153 BdrvChildRole child_role
,
3157 uint64_t perm
, shared_perm
;
3159 assert(parent_bs
->drv
);
3160 GLOBAL_STATE_CODE();
3162 if (bdrv_recurse_has_child(child_bs
, parent_bs
)) {
3163 error_setg(errp
, "Making '%s' a %s child of '%s' would create a cycle",
3164 child_bs
->node_name
, child_name
, parent_bs
->node_name
);
3168 bdrv_get_cumulative_perm(parent_bs
, &perm
, &shared_perm
);
3169 bdrv_child_perm(parent_bs
, child_bs
, NULL
, child_role
, NULL
,
3170 perm
, shared_perm
, &perm
, &shared_perm
);
3172 return bdrv_attach_child_common(child_bs
, child_name
, child_class
,
3173 child_role
, perm
, shared_perm
, parent_bs
,
3178 * This function steals the reference to child_bs from the caller.
3179 * That reference is later dropped by bdrv_root_unref_child().
3181 * On failure NULL is returned, errp is set and the reference to
3182 * child_bs is also dropped.
3184 * The caller must hold the AioContext lock @child_bs, but not that of @ctx
3185 * (unless @child_bs is already in @ctx).
3187 BdrvChild
*bdrv_root_attach_child(BlockDriverState
*child_bs
,
3188 const char *child_name
,
3189 const BdrvChildClass
*child_class
,
3190 BdrvChildRole child_role
,
3191 uint64_t perm
, uint64_t shared_perm
,
3192 void *opaque
, Error
**errp
)
3196 Transaction
*tran
= tran_new();
3198 GLOBAL_STATE_CODE();
3200 bdrv_graph_wrlock(child_bs
);
3202 child
= bdrv_attach_child_common(child_bs
, child_name
, child_class
,
3203 child_role
, perm
, shared_perm
, opaque
,
3210 ret
= bdrv_refresh_perms(child_bs
, tran
, errp
);
3213 tran_finalize(tran
, ret
);
3214 bdrv_graph_wrunlock();
3216 bdrv_unref(child_bs
);
3218 return ret
< 0 ? NULL
: child
;
3222 * This function transfers the reference to child_bs from the caller
3223 * to parent_bs. That reference is later dropped by parent_bs on
3224 * bdrv_close() or if someone calls bdrv_unref_child().
3226 * On failure NULL is returned, errp is set and the reference to
3227 * child_bs is also dropped.
3229 * If @parent_bs and @child_bs are in different AioContexts, the caller must
3230 * hold the AioContext lock for @child_bs, but not for @parent_bs.
3232 BdrvChild
*bdrv_attach_child(BlockDriverState
*parent_bs
,
3233 BlockDriverState
*child_bs
,
3234 const char *child_name
,
3235 const BdrvChildClass
*child_class
,
3236 BdrvChildRole child_role
,
3241 Transaction
*tran
= tran_new();
3243 GLOBAL_STATE_CODE();
3245 child
= bdrv_attach_child_noperm(parent_bs
, child_bs
, child_name
,
3246 child_class
, child_role
, tran
, errp
);
3252 ret
= bdrv_refresh_perms(parent_bs
, tran
, errp
);
3258 tran_finalize(tran
, ret
);
3260 bdrv_schedule_unref(child_bs
);
3262 return ret
< 0 ? NULL
: child
;
3265 /* Callers must ensure that child->frozen is false. */
3266 void bdrv_root_unref_child(BdrvChild
*child
)
3268 BlockDriverState
*child_bs
= child
->bs
;
3270 GLOBAL_STATE_CODE();
3271 bdrv_graph_wrlock(NULL
);
3272 bdrv_replace_child_noperm(child
, NULL
);
3273 bdrv_child_free(child
);
3277 * Update permissions for old node. We're just taking a parent away, so
3278 * we're loosening restrictions. Errors of permission update are not
3279 * fatal in this case, ignore them.
3281 bdrv_refresh_perms(child_bs
, NULL
, NULL
);
3284 * When the parent requiring a non-default AioContext is removed, the
3285 * node moves back to the main AioContext
3287 bdrv_try_change_aio_context(child_bs
, qemu_get_aio_context(), NULL
,
3291 bdrv_graph_wrunlock();
3292 bdrv_unref(child_bs
);
3295 typedef struct BdrvSetInheritsFrom
{
3296 BlockDriverState
*bs
;
3297 BlockDriverState
*old_inherits_from
;
3298 } BdrvSetInheritsFrom
;
3300 static void bdrv_set_inherits_from_abort(void *opaque
)
3302 BdrvSetInheritsFrom
*s
= opaque
;
3304 s
->bs
->inherits_from
= s
->old_inherits_from
;
3307 static TransactionActionDrv bdrv_set_inherits_from_drv
= {
3308 .abort
= bdrv_set_inherits_from_abort
,
3312 /* @tran is allowed to be NULL. In this case no rollback is possible */
3313 static void bdrv_set_inherits_from(BlockDriverState
*bs
,
3314 BlockDriverState
*new_inherits_from
,
3318 BdrvSetInheritsFrom
*s
= g_new(BdrvSetInheritsFrom
, 1);
3320 *s
= (BdrvSetInheritsFrom
) {
3322 .old_inherits_from
= bs
->inherits_from
,
3325 tran_add(tran
, &bdrv_set_inherits_from_drv
, s
);
3328 bs
->inherits_from
= new_inherits_from
;
3332 * Clear all inherits_from pointers from children and grandchildren of
3333 * @root that point to @root, where necessary.
3334 * @tran is allowed to be NULL. In this case no rollback is possible
3336 static void bdrv_unset_inherits_from(BlockDriverState
*root
, BdrvChild
*child
,
3341 if (child
->bs
->inherits_from
== root
) {
3343 * Remove inherits_from only when the last reference between root and
3344 * child->bs goes away.
3346 QLIST_FOREACH(c
, &root
->children
, next
) {
3347 if (c
!= child
&& c
->bs
== child
->bs
) {
3352 bdrv_set_inherits_from(child
->bs
, NULL
, tran
);
3356 QLIST_FOREACH(c
, &child
->bs
->children
, next
) {
3357 bdrv_unset_inherits_from(root
, c
, tran
);
3361 /* Callers must ensure that child->frozen is false. */
3362 void bdrv_unref_child(BlockDriverState
*parent
, BdrvChild
*child
)
3364 GLOBAL_STATE_CODE();
3365 if (child
== NULL
) {
3369 bdrv_unset_inherits_from(parent
, child
, NULL
);
3370 bdrv_root_unref_child(child
);
3374 static void GRAPH_RDLOCK
3375 bdrv_parent_cb_change_media(BlockDriverState
*bs
, bool load
)
3378 GLOBAL_STATE_CODE();
3379 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
3380 if (c
->klass
->change_media
) {
3381 c
->klass
->change_media(c
, load
);
3386 /* Return true if you can reach parent going through child->inherits_from
3387 * recursively. If parent or child are NULL, return false */
3388 static bool bdrv_inherits_from_recursive(BlockDriverState
*child
,
3389 BlockDriverState
*parent
)
3391 while (child
&& child
!= parent
) {
3392 child
= child
->inherits_from
;
3395 return child
!= NULL
;
3399 * Return the BdrvChildRole for @bs's backing child. bs->backing is
3400 * mostly used for COW backing children (role = COW), but also for
3401 * filtered children (role = FILTERED | PRIMARY).
3403 static BdrvChildRole
bdrv_backing_role(BlockDriverState
*bs
)
3405 if (bs
->drv
&& bs
->drv
->is_filter
) {
3406 return BDRV_CHILD_FILTERED
| BDRV_CHILD_PRIMARY
;
3408 return BDRV_CHILD_COW
;
3413 * Sets the bs->backing or bs->file link of a BDS. A new reference is created;
3414 * callers which don't need their own reference any more must call bdrv_unref().
3416 * If the respective child is already present (i.e. we're detaching a node),
3417 * that child node must be drained.
3419 * Function doesn't update permissions, caller is responsible for this.
3421 * The caller must hold the AioContext lock for @child_bs. Both @parent_bs and
3422 * @child_bs can move to a different AioContext in this function. Callers must
3423 * make sure that their AioContext locking is still correct after this.
3425 * After calling this function, the transaction @tran may only be completed
3426 * while holding a writer lock for the graph.
3428 static int GRAPH_WRLOCK
3429 bdrv_set_file_or_backing_noperm(BlockDriverState
*parent_bs
,
3430 BlockDriverState
*child_bs
,
3432 Transaction
*tran
, Error
**errp
)
3434 bool update_inherits_from
=
3435 bdrv_inherits_from_recursive(child_bs
, parent_bs
);
3436 BdrvChild
*child
= is_backing
? parent_bs
->backing
: parent_bs
->file
;
3439 GLOBAL_STATE_CODE();
3441 if (!parent_bs
->drv
) {
3443 * Node without drv is an object without a class :/. TODO: finally fix
3444 * qcow2 driver to never clear bs->drv and implement format corruption
3445 * handling in other way.
3447 error_setg(errp
, "Node corrupted");
3451 if (child
&& child
->frozen
) {
3452 error_setg(errp
, "Cannot change frozen '%s' link from '%s' to '%s'",
3453 child
->name
, parent_bs
->node_name
, child
->bs
->node_name
);
3457 if (is_backing
&& !parent_bs
->drv
->is_filter
&&
3458 !parent_bs
->drv
->supports_backing
)
3460 error_setg(errp
, "Driver '%s' of node '%s' does not support backing "
3461 "files", parent_bs
->drv
->format_name
, parent_bs
->node_name
);
3465 if (parent_bs
->drv
->is_filter
) {
3466 role
= BDRV_CHILD_FILTERED
| BDRV_CHILD_PRIMARY
;
3467 } else if (is_backing
) {
3468 role
= BDRV_CHILD_COW
;
3471 * We only can use same role as it is in existing child. We don't have
3472 * infrastructure to determine role of file child in generic way
3475 error_setg(errp
, "Cannot set file child to format node without "
3483 assert(child
->bs
->quiesce_counter
);
3484 bdrv_unset_inherits_from(parent_bs
, child
, tran
);
3485 bdrv_remove_child(child
, tran
);
3492 child
= bdrv_attach_child_noperm(parent_bs
, child_bs
,
3493 is_backing
? "backing" : "file",
3494 &child_of_bds
, role
,
3502 * If inherits_from pointed recursively to bs then let's update it to
3503 * point directly to bs (else it will become NULL).
3505 if (update_inherits_from
) {
3506 bdrv_set_inherits_from(child_bs
, parent_bs
, tran
);
3510 bdrv_refresh_limits(parent_bs
, tran
, NULL
);
3516 * The caller must hold the AioContext lock for @backing_hd. Both @bs and
3517 * @backing_hd can move to a different AioContext in this function. Callers must
3518 * make sure that their AioContext locking is still correct after this.
3520 * If a backing child is already present (i.e. we're detaching a node), that
3521 * child node must be drained.
3523 * After calling this function, the transaction @tran may only be completed
3524 * while holding a writer lock for the graph.
3526 static int GRAPH_WRLOCK
3527 bdrv_set_backing_noperm(BlockDriverState
*bs
,
3528 BlockDriverState
*backing_hd
,
3529 Transaction
*tran
, Error
**errp
)
3531 GLOBAL_STATE_CODE();
3532 return bdrv_set_file_or_backing_noperm(bs
, backing_hd
, true, tran
, errp
);
3535 int bdrv_set_backing_hd_drained(BlockDriverState
*bs
,
3536 BlockDriverState
*backing_hd
,
3540 Transaction
*tran
= tran_new();
3542 GLOBAL_STATE_CODE();
3543 assert(bs
->quiesce_counter
> 0);
3545 assert(bs
->backing
->bs
->quiesce_counter
> 0);
3547 bdrv_graph_wrlock(backing_hd
);
3549 ret
= bdrv_set_backing_noperm(bs
, backing_hd
, tran
, errp
);
3554 ret
= bdrv_refresh_perms(bs
, tran
, errp
);
3556 tran_finalize(tran
, ret
);
3557 bdrv_graph_wrunlock();
3561 int bdrv_set_backing_hd(BlockDriverState
*bs
, BlockDriverState
*backing_hd
,
3564 BlockDriverState
*drain_bs
= bs
->backing
? bs
->backing
->bs
: bs
;
3566 GLOBAL_STATE_CODE();
3569 bdrv_drained_begin(drain_bs
);
3570 ret
= bdrv_set_backing_hd_drained(bs
, backing_hd
, errp
);
3571 bdrv_drained_end(drain_bs
);
3572 bdrv_unref(drain_bs
);
3578 * Opens the backing file for a BlockDriverState if not yet open
3580 * bdref_key specifies the key for the image's BlockdevRef in the options QDict.
3581 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
3582 * itself, all options starting with "${bdref_key}." are considered part of the
3585 * The caller must hold the main AioContext lock.
3587 * TODO Can this be unified with bdrv_open_image()?
3589 int bdrv_open_backing_file(BlockDriverState
*bs
, QDict
*parent_options
,
3590 const char *bdref_key
, Error
**errp
)
3592 char *backing_filename
= NULL
;
3593 char *bdref_key_dot
;
3594 const char *reference
= NULL
;
3596 bool implicit_backing
= false;
3597 BlockDriverState
*backing_hd
;
3598 AioContext
*backing_hd_ctx
;
3600 QDict
*tmp_parent_options
= NULL
;
3601 Error
*local_err
= NULL
;
3603 GLOBAL_STATE_CODE();
3605 if (bs
->backing
!= NULL
) {
3609 /* NULL means an empty set of options */
3610 if (parent_options
== NULL
) {
3611 tmp_parent_options
= qdict_new();
3612 parent_options
= tmp_parent_options
;
3615 bs
->open_flags
&= ~BDRV_O_NO_BACKING
;
3617 bdref_key_dot
= g_strdup_printf("%s.", bdref_key
);
3618 qdict_extract_subqdict(parent_options
, &options
, bdref_key_dot
);
3619 g_free(bdref_key_dot
);
3622 * Caution: while qdict_get_try_str() is fine, getting non-string
3623 * types would require more care. When @parent_options come from
3624 * -blockdev or blockdev_add, its members are typed according to
3625 * the QAPI schema, but when they come from -drive, they're all
3628 reference
= qdict_get_try_str(parent_options
, bdref_key
);
3629 if (reference
|| qdict_haskey(options
, "file.filename")) {
3630 /* keep backing_filename NULL */
3631 } else if (bs
->backing_file
[0] == '\0' && qdict_size(options
) == 0) {
3632 qobject_unref(options
);
3635 if (qdict_size(options
) == 0) {
3636 /* If the user specifies options that do not modify the
3637 * backing file's behavior, we might still consider it the
3638 * implicit backing file. But it's easier this way, and
3639 * just specifying some of the backing BDS's options is
3640 * only possible with -drive anyway (otherwise the QAPI
3641 * schema forces the user to specify everything). */
3642 implicit_backing
= !strcmp(bs
->auto_backing_file
, bs
->backing_file
);
3645 backing_filename
= bdrv_get_full_backing_filename(bs
, &local_err
);
3648 error_propagate(errp
, local_err
);
3649 qobject_unref(options
);
3654 if (!bs
->drv
|| !bs
->drv
->supports_backing
) {
3656 error_setg(errp
, "Driver doesn't support backing files");
3657 qobject_unref(options
);
3662 bs
->backing_format
[0] != '\0' && !qdict_haskey(options
, "driver")) {
3663 qdict_put_str(options
, "driver", bs
->backing_format
);
3666 backing_hd
= bdrv_open_inherit(backing_filename
, reference
, options
, 0, bs
,
3667 &child_of_bds
, bdrv_backing_role(bs
), errp
);
3669 bs
->open_flags
|= BDRV_O_NO_BACKING
;
3670 error_prepend(errp
, "Could not open backing file: ");
3675 if (implicit_backing
) {
3676 bdrv_refresh_filename(backing_hd
);
3677 pstrcpy(bs
->auto_backing_file
, sizeof(bs
->auto_backing_file
),
3678 backing_hd
->filename
);
3681 /* Hook up the backing file link; drop our reference, bs owns the
3682 * backing_hd reference now */
3683 backing_hd_ctx
= bdrv_get_aio_context(backing_hd
);
3684 aio_context_acquire(backing_hd_ctx
);
3685 ret
= bdrv_set_backing_hd(bs
, backing_hd
, errp
);
3686 bdrv_unref(backing_hd
);
3687 aio_context_release(backing_hd_ctx
);
3693 qdict_del(parent_options
, bdref_key
);
3696 g_free(backing_filename
);
3697 qobject_unref(tmp_parent_options
);
3701 static BlockDriverState
*
3702 bdrv_open_child_bs(const char *filename
, QDict
*options
, const char *bdref_key
,
3703 BlockDriverState
*parent
, const BdrvChildClass
*child_class
,
3704 BdrvChildRole child_role
, bool allow_none
, Error
**errp
)
3706 BlockDriverState
*bs
= NULL
;
3707 QDict
*image_options
;
3708 char *bdref_key_dot
;
3709 const char *reference
;
3711 assert(child_class
!= NULL
);
3713 bdref_key_dot
= g_strdup_printf("%s.", bdref_key
);
3714 qdict_extract_subqdict(options
, &image_options
, bdref_key_dot
);
3715 g_free(bdref_key_dot
);
3718 * Caution: while qdict_get_try_str() is fine, getting non-string
3719 * types would require more care. When @options come from
3720 * -blockdev or blockdev_add, its members are typed according to
3721 * the QAPI schema, but when they come from -drive, they're all
3724 reference
= qdict_get_try_str(options
, bdref_key
);
3725 if (!filename
&& !reference
&& !qdict_size(image_options
)) {
3727 error_setg(errp
, "A block device must be specified for \"%s\"",
3730 qobject_unref(image_options
);
3734 bs
= bdrv_open_inherit(filename
, reference
, image_options
, 0,
3735 parent
, child_class
, child_role
, errp
);
3741 qdict_del(options
, bdref_key
);
3746 * Opens a disk image whose options are given as BlockdevRef in another block
3749 * If allow_none is true, no image will be opened if filename is false and no
3750 * BlockdevRef is given. NULL will be returned, but errp remains unset.
3752 * bdrev_key specifies the key for the image's BlockdevRef in the options QDict.
3753 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
3754 * itself, all options starting with "${bdref_key}." are considered part of the
3757 * The BlockdevRef will be removed from the options QDict.
3759 * The caller must hold the lock of the main AioContext and no other AioContext.
3760 * @parent can move to a different AioContext in this function. Callers must
3761 * make sure that their AioContext locking is still correct after this.
3763 BdrvChild
*bdrv_open_child(const char *filename
,
3764 QDict
*options
, const char *bdref_key
,
3765 BlockDriverState
*parent
,
3766 const BdrvChildClass
*child_class
,
3767 BdrvChildRole child_role
,
3768 bool allow_none
, Error
**errp
)
3770 BlockDriverState
*bs
;
3774 GLOBAL_STATE_CODE();
3776 bs
= bdrv_open_child_bs(filename
, options
, bdref_key
, parent
, child_class
,
3777 child_role
, allow_none
, errp
);
3782 bdrv_graph_wrlock(NULL
);
3783 ctx
= bdrv_get_aio_context(bs
);
3784 aio_context_acquire(ctx
);
3785 child
= bdrv_attach_child(parent
, bs
, bdref_key
, child_class
, child_role
,
3787 aio_context_release(ctx
);
3788 bdrv_graph_wrunlock();
3794 * Wrapper on bdrv_open_child() for most popular case: open primary child of bs.
3796 * The caller must hold the lock of the main AioContext and no other AioContext.
3797 * @parent can move to a different AioContext in this function. Callers must
3798 * make sure that their AioContext locking is still correct after this.
3800 int bdrv_open_file_child(const char *filename
,
3801 QDict
*options
, const char *bdref_key
,
3802 BlockDriverState
*parent
, Error
**errp
)
3806 /* commit_top and mirror_top don't use this function */
3807 assert(!parent
->drv
->filtered_child_is_backing
);
3808 role
= parent
->drv
->is_filter
?
3809 (BDRV_CHILD_FILTERED
| BDRV_CHILD_PRIMARY
) : BDRV_CHILD_IMAGE
;
3811 if (!bdrv_open_child(filename
, options
, bdref_key
, parent
,
3812 &child_of_bds
, role
, false, errp
))
3821 * TODO Future callers may need to specify parent/child_class in order for
3822 * option inheritance to work. Existing callers use it for the root node.
3824 BlockDriverState
*bdrv_open_blockdev_ref(BlockdevRef
*ref
, Error
**errp
)
3826 BlockDriverState
*bs
= NULL
;
3827 QObject
*obj
= NULL
;
3828 QDict
*qdict
= NULL
;
3829 const char *reference
= NULL
;
3832 GLOBAL_STATE_CODE();
3834 if (ref
->type
== QTYPE_QSTRING
) {
3835 reference
= ref
->u
.reference
;
3837 BlockdevOptions
*options
= &ref
->u
.definition
;
3838 assert(ref
->type
== QTYPE_QDICT
);
3840 v
= qobject_output_visitor_new(&obj
);
3841 visit_type_BlockdevOptions(v
, NULL
, &options
, &error_abort
);
3842 visit_complete(v
, &obj
);
3844 qdict
= qobject_to(QDict
, obj
);
3845 qdict_flatten(qdict
);
3847 /* bdrv_open_inherit() defaults to the values in bdrv_flags (for
3848 * compatibility with other callers) rather than what we want as the
3849 * real defaults. Apply the defaults here instead. */
3850 qdict_set_default_str(qdict
, BDRV_OPT_CACHE_DIRECT
, "off");
3851 qdict_set_default_str(qdict
, BDRV_OPT_CACHE_NO_FLUSH
, "off");
3852 qdict_set_default_str(qdict
, BDRV_OPT_READ_ONLY
, "off");
3853 qdict_set_default_str(qdict
, BDRV_OPT_AUTO_READ_ONLY
, "off");
3857 bs
= bdrv_open_inherit(NULL
, reference
, qdict
, 0, NULL
, NULL
, 0, errp
);
3864 static BlockDriverState
*bdrv_append_temp_snapshot(BlockDriverState
*bs
,
3866 QDict
*snapshot_options
,
3869 g_autofree
char *tmp_filename
= NULL
;
3871 QemuOpts
*opts
= NULL
;
3872 BlockDriverState
*bs_snapshot
= NULL
;
3873 AioContext
*ctx
= bdrv_get_aio_context(bs
);
3876 GLOBAL_STATE_CODE();
3878 /* if snapshot, we create a temporary backing file and open it
3879 instead of opening 'filename' directly */
3881 /* Get the required size from the image */
3882 aio_context_acquire(ctx
);
3883 total_size
= bdrv_getlength(bs
);
3884 aio_context_release(ctx
);
3886 if (total_size
< 0) {
3887 error_setg_errno(errp
, -total_size
, "Could not get image size");
3891 /* Create the temporary image */
3892 tmp_filename
= create_tmp_file(errp
);
3893 if (!tmp_filename
) {
3897 opts
= qemu_opts_create(bdrv_qcow2
.create_opts
, NULL
, 0,
3899 qemu_opt_set_number(opts
, BLOCK_OPT_SIZE
, total_size
, &error_abort
);
3900 ret
= bdrv_create(&bdrv_qcow2
, tmp_filename
, opts
, errp
);
3901 qemu_opts_del(opts
);
3903 error_prepend(errp
, "Could not create temporary overlay '%s': ",
3908 /* Prepare options QDict for the temporary file */
3909 qdict_put_str(snapshot_options
, "file.driver", "file");
3910 qdict_put_str(snapshot_options
, "file.filename", tmp_filename
);
3911 qdict_put_str(snapshot_options
, "driver", "qcow2");
3913 bs_snapshot
= bdrv_open(NULL
, NULL
, snapshot_options
, flags
, errp
);
3914 snapshot_options
= NULL
;
3919 aio_context_acquire(ctx
);
3920 ret
= bdrv_append(bs_snapshot
, bs
, errp
);
3921 aio_context_release(ctx
);
3929 qobject_unref(snapshot_options
);
3934 * Opens a disk image (raw, qcow2, vmdk, ...)
3936 * options is a QDict of options to pass to the block drivers, or NULL for an
3937 * empty set of options. The reference to the QDict belongs to the block layer
3938 * after the call (even on failure), so if the caller intends to reuse the
3939 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
3941 * If *pbs is NULL, a new BDS will be created with a pointer to it stored there.
3942 * If it is not NULL, the referenced BDS will be reused.
3944 * The reference parameter may be used to specify an existing block device which
3945 * should be opened. If specified, neither options nor a filename may be given,
3946 * nor can an existing BDS be reused (that is, *pbs has to be NULL).
3948 * The caller must always hold the main AioContext lock.
3950 static BlockDriverState
* no_coroutine_fn
3951 bdrv_open_inherit(const char *filename
, const char *reference
, QDict
*options
,
3952 int flags
, BlockDriverState
*parent
,
3953 const BdrvChildClass
*child_class
, BdrvChildRole child_role
,
3957 BlockBackend
*file
= NULL
;
3958 BlockDriverState
*bs
;
3959 BlockDriver
*drv
= NULL
;
3961 const char *drvname
;
3962 const char *backing
;
3963 Error
*local_err
= NULL
;
3964 QDict
*snapshot_options
= NULL
;
3965 int snapshot_flags
= 0;
3966 AioContext
*ctx
= qemu_get_aio_context();
3968 assert(!child_class
|| !flags
);
3969 assert(!child_class
== !parent
);
3970 GLOBAL_STATE_CODE();
3971 assert(!qemu_in_coroutine());
3973 /* TODO We'll eventually have to take a writer lock in this function */
3974 GRAPH_RDLOCK_GUARD_MAINLOOP();
3977 bool options_non_empty
= options
? qdict_size(options
) : false;
3978 qobject_unref(options
);
3980 if (filename
|| options_non_empty
) {
3981 error_setg(errp
, "Cannot reference an existing block device with "
3982 "additional options or a new filename");
3986 bs
= bdrv_lookup_bs(reference
, reference
, errp
);
3997 /* NULL means an empty set of options */
3998 if (options
== NULL
) {
3999 options
= qdict_new();
4002 /* json: syntax counts as explicit options, as if in the QDict */
4003 parse_json_protocol(options
, &filename
, &local_err
);
4008 bs
->explicit_options
= qdict_clone_shallow(options
);
4011 bool parent_is_format
;
4014 parent_is_format
= parent
->drv
->is_format
;
4017 * parent->drv is not set yet because this node is opened for
4018 * (potential) format probing. That means that @parent is going
4019 * to be a format node.
4021 parent_is_format
= true;
4024 bs
->inherits_from
= parent
;
4025 child_class
->inherit_options(child_role
, parent_is_format
,
4027 parent
->open_flags
, parent
->options
);
4030 ret
= bdrv_fill_options(&options
, filename
, &flags
, &local_err
);
4036 * Set the BDRV_O_RDWR and BDRV_O_ALLOW_RDWR flags.
4037 * Caution: getting a boolean member of @options requires care.
4038 * When @options come from -blockdev or blockdev_add, members are
4039 * typed according to the QAPI schema, but when they come from
4040 * -drive, they're all QString.
4042 if (g_strcmp0(qdict_get_try_str(options
, BDRV_OPT_READ_ONLY
), "on") &&
4043 !qdict_get_try_bool(options
, BDRV_OPT_READ_ONLY
, false)) {
4044 flags
|= (BDRV_O_RDWR
| BDRV_O_ALLOW_RDWR
);
4046 flags
&= ~BDRV_O_RDWR
;
4049 if (flags
& BDRV_O_SNAPSHOT
) {
4050 snapshot_options
= qdict_new();
4051 bdrv_temp_snapshot_options(&snapshot_flags
, snapshot_options
,
4053 /* Let bdrv_backing_options() override "read-only" */
4054 qdict_del(options
, BDRV_OPT_READ_ONLY
);
4055 bdrv_inherited_options(BDRV_CHILD_COW
, true,
4056 &flags
, options
, flags
, options
);
4059 bs
->open_flags
= flags
;
4060 bs
->options
= options
;
4061 options
= qdict_clone_shallow(options
);
4063 /* Find the right image format driver */
4064 /* See cautionary note on accessing @options above */
4065 drvname
= qdict_get_try_str(options
, "driver");
4067 drv
= bdrv_find_format(drvname
);
4069 error_setg(errp
, "Unknown driver: '%s'", drvname
);
4074 assert(drvname
|| !(flags
& BDRV_O_PROTOCOL
));
4076 /* See cautionary note on accessing @options above */
4077 backing
= qdict_get_try_str(options
, "backing");
4078 if (qobject_to(QNull
, qdict_get(options
, "backing")) != NULL
||
4079 (backing
&& *backing
== '\0'))
4082 warn_report("Use of \"backing\": \"\" is deprecated; "
4083 "use \"backing\": null instead");
4085 flags
|= BDRV_O_NO_BACKING
;
4086 qdict_del(bs
->explicit_options
, "backing");
4087 qdict_del(bs
->options
, "backing");
4088 qdict_del(options
, "backing");
4091 /* Open image file without format layer. This BlockBackend is only used for
4092 * probing, the block drivers will do their own bdrv_open_child() for the
4093 * same BDS, which is why we put the node name back into options. */
4094 if ((flags
& BDRV_O_PROTOCOL
) == 0) {
4095 BlockDriverState
*file_bs
;
4097 file_bs
= bdrv_open_child_bs(filename
, options
, "file", bs
,
4098 &child_of_bds
, BDRV_CHILD_IMAGE
,
4103 if (file_bs
!= NULL
) {
4104 /* Not requesting BLK_PERM_CONSISTENT_READ because we're only
4105 * looking at the header to guess the image format. This works even
4106 * in cases where a guest would not see a consistent state. */
4107 ctx
= bdrv_get_aio_context(file_bs
);
4108 aio_context_acquire(ctx
);
4109 file
= blk_new(ctx
, 0, BLK_PERM_ALL
);
4110 blk_insert_bs(file
, file_bs
, &local_err
);
4111 bdrv_unref(file_bs
);
4112 aio_context_release(ctx
);
4118 qdict_put_str(options
, "file", bdrv_get_node_name(file_bs
));
4122 /* Image format probing */
4125 ret
= find_image_format(file
, filename
, &drv
, &local_err
);
4130 * This option update would logically belong in bdrv_fill_options(),
4131 * but we first need to open bs->file for the probing to work, while
4132 * opening bs->file already requires the (mostly) final set of options
4133 * so that cache mode etc. can be inherited.
4135 * Adding the driver later is somewhat ugly, but it's not an option
4136 * that would ever be inherited, so it's correct. We just need to make
4137 * sure to update both bs->options (which has the full effective
4138 * options for bs) and options (which has file.* already removed).
4140 qdict_put_str(bs
->options
, "driver", drv
->format_name
);
4141 qdict_put_str(options
, "driver", drv
->format_name
);
4143 error_setg(errp
, "Must specify either driver or file");
4147 /* BDRV_O_PROTOCOL must be set iff a protocol BDS is about to be created */
4148 assert(!!(flags
& BDRV_O_PROTOCOL
) == !!drv
->bdrv_file_open
);
4149 /* file must be NULL if a protocol BDS is about to be created
4150 * (the inverse results in an error message from bdrv_open_common()) */
4151 assert(!(flags
& BDRV_O_PROTOCOL
) || !file
);
4153 /* Open the image */
4154 ret
= bdrv_open_common(bs
, file
, options
, &local_err
);
4159 /* The AioContext could have changed during bdrv_open_common() */
4160 ctx
= bdrv_get_aio_context(bs
);
4163 aio_context_acquire(ctx
);
4165 aio_context_release(ctx
);
4169 /* If there is a backing file, use it */
4170 if ((flags
& BDRV_O_NO_BACKING
) == 0) {
4171 ret
= bdrv_open_backing_file(bs
, options
, "backing", &local_err
);
4173 goto close_and_fail
;
4177 /* Remove all children options and references
4178 * from bs->options and bs->explicit_options */
4179 QLIST_FOREACH(child
, &bs
->children
, next
) {
4180 char *child_key_dot
;
4181 child_key_dot
= g_strdup_printf("%s.", child
->name
);
4182 qdict_extract_subqdict(bs
->explicit_options
, NULL
, child_key_dot
);
4183 qdict_extract_subqdict(bs
->options
, NULL
, child_key_dot
);
4184 qdict_del(bs
->explicit_options
, child
->name
);
4185 qdict_del(bs
->options
, child
->name
);
4186 g_free(child_key_dot
);
4189 /* Check if any unknown options were used */
4190 if (qdict_size(options
) != 0) {
4191 const QDictEntry
*entry
= qdict_first(options
);
4192 if (flags
& BDRV_O_PROTOCOL
) {
4193 error_setg(errp
, "Block protocol '%s' doesn't support the option "
4194 "'%s'", drv
->format_name
, entry
->key
);
4197 "Block format '%s' does not support the option '%s'",
4198 drv
->format_name
, entry
->key
);
4201 goto close_and_fail
;
4204 bdrv_parent_cb_change_media(bs
, true);
4206 qobject_unref(options
);
4209 /* For snapshot=on, create a temporary qcow2 overlay. bs points to the
4210 * temporary snapshot afterwards. */
4211 if (snapshot_flags
) {
4212 BlockDriverState
*snapshot_bs
;
4213 snapshot_bs
= bdrv_append_temp_snapshot(bs
, snapshot_flags
,
4214 snapshot_options
, &local_err
);
4215 snapshot_options
= NULL
;
4217 goto close_and_fail
;
4219 /* We are not going to return bs but the overlay on top of it
4220 * (snapshot_bs); thus, we have to drop the strong reference to bs
4221 * (which we obtained by calling bdrv_new()). bs will not be deleted,
4222 * though, because the overlay still has a reference to it. */
4223 aio_context_acquire(ctx
);
4225 aio_context_release(ctx
);
4232 aio_context_acquire(ctx
);
4234 qobject_unref(snapshot_options
);
4235 qobject_unref(bs
->explicit_options
);
4236 qobject_unref(bs
->options
);
4237 qobject_unref(options
);
4239 bs
->explicit_options
= NULL
;
4241 aio_context_release(ctx
);
4242 error_propagate(errp
, local_err
);
4246 aio_context_acquire(ctx
);
4248 aio_context_release(ctx
);
4249 qobject_unref(snapshot_options
);
4250 qobject_unref(options
);
4251 error_propagate(errp
, local_err
);
4255 /* The caller must always hold the main AioContext lock. */
4256 BlockDriverState
*bdrv_open(const char *filename
, const char *reference
,
4257 QDict
*options
, int flags
, Error
**errp
)
4259 GLOBAL_STATE_CODE();
4261 return bdrv_open_inherit(filename
, reference
, options
, flags
, NULL
,
4265 /* Return true if the NULL-terminated @list contains @str */
4266 static bool is_str_in_list(const char *str
, const char *const *list
)
4270 for (i
= 0; list
[i
] != NULL
; i
++) {
4271 if (!strcmp(str
, list
[i
])) {
4280 * Check that every option set in @bs->options is also set in
4283 * Options listed in the common_options list and in
4284 * @bs->drv->mutable_opts are skipped.
4286 * Return 0 on success, otherwise return -EINVAL and set @errp.
4288 static int bdrv_reset_options_allowed(BlockDriverState
*bs
,
4289 const QDict
*new_opts
, Error
**errp
)
4291 const QDictEntry
*e
;
4292 /* These options are common to all block drivers and are handled
4293 * in bdrv_reopen_prepare() so they can be left out of @new_opts */
4294 const char *const common_options
[] = {
4295 "node-name", "discard", "cache.direct", "cache.no-flush",
4296 "read-only", "auto-read-only", "detect-zeroes", NULL
4299 for (e
= qdict_first(bs
->options
); e
; e
= qdict_next(bs
->options
, e
)) {
4300 if (!qdict_haskey(new_opts
, e
->key
) &&
4301 !is_str_in_list(e
->key
, common_options
) &&
4302 !is_str_in_list(e
->key
, bs
->drv
->mutable_opts
)) {
4303 error_setg(errp
, "Option '%s' cannot be reset "
4304 "to its default value", e
->key
);
4313 * Returns true if @child can be reached recursively from @bs
4315 static bool bdrv_recurse_has_child(BlockDriverState
*bs
,
4316 BlockDriverState
*child
)
4324 QLIST_FOREACH(c
, &bs
->children
, next
) {
4325 if (bdrv_recurse_has_child(c
->bs
, child
)) {
4334 * Adds a BlockDriverState to a simple queue for an atomic, transactional
4335 * reopen of multiple devices.
4337 * bs_queue can either be an existing BlockReopenQueue that has had QTAILQ_INIT
4338 * already performed, or alternatively may be NULL a new BlockReopenQueue will
4339 * be created and initialized. This newly created BlockReopenQueue should be
4340 * passed back in for subsequent calls that are intended to be of the same
4343 * bs is the BlockDriverState to add to the reopen queue.
4345 * options contains the changed options for the associated bs
4346 * (the BlockReopenQueue takes ownership)
4348 * flags contains the open flags for the associated bs
4350 * returns a pointer to bs_queue, which is either the newly allocated
4351 * bs_queue, or the existing bs_queue being used.
4353 * bs is drained here and undrained by bdrv_reopen_queue_free().
4355 * To be called with bs->aio_context locked.
4357 static BlockReopenQueue
*bdrv_reopen_queue_child(BlockReopenQueue
*bs_queue
,
4358 BlockDriverState
*bs
,
4360 const BdrvChildClass
*klass
,
4362 bool parent_is_format
,
4363 QDict
*parent_options
,
4369 BlockReopenQueueEntry
*bs_entry
;
4371 QDict
*old_options
, *explicit_options
, *options_copy
;
4375 GLOBAL_STATE_CODE();
4377 bdrv_drained_begin(bs
);
4379 if (bs_queue
== NULL
) {
4380 bs_queue
= g_new0(BlockReopenQueue
, 1);
4381 QTAILQ_INIT(bs_queue
);
4385 options
= qdict_new();
4388 /* Check if this BlockDriverState is already in the queue */
4389 QTAILQ_FOREACH(bs_entry
, bs_queue
, entry
) {
4390 if (bs
== bs_entry
->state
.bs
) {
4396 * Precedence of options:
4397 * 1. Explicitly passed in options (highest)
4398 * 2. Retained from explicitly set options of bs
4399 * 3. Inherited from parent node
4400 * 4. Retained from effective options of bs
4403 /* Old explicitly set values (don't overwrite by inherited value) */
4404 if (bs_entry
|| keep_old_opts
) {
4405 old_options
= qdict_clone_shallow(bs_entry
?
4406 bs_entry
->state
.explicit_options
:
4407 bs
->explicit_options
);
4408 bdrv_join_options(bs
, options
, old_options
);
4409 qobject_unref(old_options
);
4412 explicit_options
= qdict_clone_shallow(options
);
4414 /* Inherit from parent node */
4415 if (parent_options
) {
4417 klass
->inherit_options(role
, parent_is_format
, &flags
, options
,
4418 parent_flags
, parent_options
);
4420 flags
= bdrv_get_flags(bs
);
4423 if (keep_old_opts
) {
4424 /* Old values are used for options that aren't set yet */
4425 old_options
= qdict_clone_shallow(bs
->options
);
4426 bdrv_join_options(bs
, options
, old_options
);
4427 qobject_unref(old_options
);
4430 /* We have the final set of options so let's update the flags */
4431 options_copy
= qdict_clone_shallow(options
);
4432 opts
= qemu_opts_create(&bdrv_runtime_opts
, NULL
, 0, &error_abort
);
4433 qemu_opts_absorb_qdict(opts
, options_copy
, NULL
);
4434 update_flags_from_options(&flags
, opts
);
4435 qemu_opts_del(opts
);
4436 qobject_unref(options_copy
);
4438 /* bdrv_open_inherit() sets and clears some additional flags internally */
4439 flags
&= ~BDRV_O_PROTOCOL
;
4440 if (flags
& BDRV_O_RDWR
) {
4441 flags
|= BDRV_O_ALLOW_RDWR
;
4445 bs_entry
= g_new0(BlockReopenQueueEntry
, 1);
4446 QTAILQ_INSERT_TAIL(bs_queue
, bs_entry
, entry
);
4448 qobject_unref(bs_entry
->state
.options
);
4449 qobject_unref(bs_entry
->state
.explicit_options
);
4452 bs_entry
->state
.bs
= bs
;
4453 bs_entry
->state
.options
= options
;
4454 bs_entry
->state
.explicit_options
= explicit_options
;
4455 bs_entry
->state
.flags
= flags
;
4458 * If keep_old_opts is false then it means that unspecified
4459 * options must be reset to their original value. We don't allow
4460 * resetting 'backing' but we need to know if the option is
4461 * missing in order to decide if we have to return an error.
4463 if (!keep_old_opts
) {
4464 bs_entry
->state
.backing_missing
=
4465 !qdict_haskey(options
, "backing") &&
4466 !qdict_haskey(options
, "backing.driver");
4469 QLIST_FOREACH(child
, &bs
->children
, next
) {
4470 QDict
*new_child_options
= NULL
;
4471 bool child_keep_old
= keep_old_opts
;
4473 /* reopen can only change the options of block devices that were
4474 * implicitly created and inherited options. For other (referenced)
4475 * block devices, a syntax like "backing.foo" results in an error. */
4476 if (child
->bs
->inherits_from
!= bs
) {
4480 /* Check if the options contain a child reference */
4481 if (qdict_haskey(options
, child
->name
)) {
4482 const char *childref
= qdict_get_try_str(options
, child
->name
);
4484 * The current child must not be reopened if the child
4485 * reference is null or points to a different node.
4487 if (g_strcmp0(childref
, child
->bs
->node_name
)) {
4491 * If the child reference points to the current child then
4492 * reopen it with its existing set of options (note that
4493 * it can still inherit new options from the parent).
4495 child_keep_old
= true;
4497 /* Extract child options ("child-name.*") */
4498 char *child_key_dot
= g_strdup_printf("%s.", child
->name
);
4499 qdict_extract_subqdict(explicit_options
, NULL
, child_key_dot
);
4500 qdict_extract_subqdict(options
, &new_child_options
, child_key_dot
);
4501 g_free(child_key_dot
);
4504 bdrv_reopen_queue_child(bs_queue
, child
->bs
, new_child_options
,
4505 child
->klass
, child
->role
, bs
->drv
->is_format
,
4506 options
, flags
, child_keep_old
);
4512 /* To be called with bs->aio_context locked */
4513 BlockReopenQueue
*bdrv_reopen_queue(BlockReopenQueue
*bs_queue
,
4514 BlockDriverState
*bs
,
4515 QDict
*options
, bool keep_old_opts
)
4517 GLOBAL_STATE_CODE();
4519 return bdrv_reopen_queue_child(bs_queue
, bs
, options
, NULL
, 0, false,
4520 NULL
, 0, keep_old_opts
);
4523 void bdrv_reopen_queue_free(BlockReopenQueue
*bs_queue
)
4525 GLOBAL_STATE_CODE();
4527 BlockReopenQueueEntry
*bs_entry
, *next
;
4528 QTAILQ_FOREACH_SAFE(bs_entry
, bs_queue
, entry
, next
) {
4529 AioContext
*ctx
= bdrv_get_aio_context(bs_entry
->state
.bs
);
4531 aio_context_acquire(ctx
);
4532 bdrv_drained_end(bs_entry
->state
.bs
);
4533 aio_context_release(ctx
);
4535 qobject_unref(bs_entry
->state
.explicit_options
);
4536 qobject_unref(bs_entry
->state
.options
);
4544 * Reopen multiple BlockDriverStates atomically & transactionally.
4546 * The queue passed in (bs_queue) must have been built up previous
4547 * via bdrv_reopen_queue().
4549 * Reopens all BDS specified in the queue, with the appropriate
4550 * flags. All devices are prepared for reopen, and failure of any
4551 * device will cause all device changes to be abandoned, and intermediate
4554 * If all devices prepare successfully, then the changes are committed
4557 * All affected nodes must be drained between bdrv_reopen_queue() and
4558 * bdrv_reopen_multiple().
4560 * To be called from the main thread, with all other AioContexts unlocked.
4562 int bdrv_reopen_multiple(BlockReopenQueue
*bs_queue
, Error
**errp
)
4565 BlockReopenQueueEntry
*bs_entry
, *next
;
4567 Transaction
*tran
= tran_new();
4568 g_autoptr(GSList
) refresh_list
= NULL
;
4570 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
4571 assert(bs_queue
!= NULL
);
4572 GLOBAL_STATE_CODE();
4574 QTAILQ_FOREACH(bs_entry
, bs_queue
, entry
) {
4575 ctx
= bdrv_get_aio_context(bs_entry
->state
.bs
);
4576 aio_context_acquire(ctx
);
4577 ret
= bdrv_flush(bs_entry
->state
.bs
);
4578 aio_context_release(ctx
);
4580 error_setg_errno(errp
, -ret
, "Error flushing drive");
4585 QTAILQ_FOREACH(bs_entry
, bs_queue
, entry
) {
4586 assert(bs_entry
->state
.bs
->quiesce_counter
> 0);
4587 ctx
= bdrv_get_aio_context(bs_entry
->state
.bs
);
4588 aio_context_acquire(ctx
);
4589 ret
= bdrv_reopen_prepare(&bs_entry
->state
, bs_queue
, tran
, errp
);
4590 aio_context_release(ctx
);
4594 bs_entry
->prepared
= true;
4597 QTAILQ_FOREACH(bs_entry
, bs_queue
, entry
) {
4598 BDRVReopenState
*state
= &bs_entry
->state
;
4600 refresh_list
= g_slist_prepend(refresh_list
, state
->bs
);
4601 if (state
->old_backing_bs
) {
4602 refresh_list
= g_slist_prepend(refresh_list
, state
->old_backing_bs
);
4604 if (state
->old_file_bs
) {
4605 refresh_list
= g_slist_prepend(refresh_list
, state
->old_file_bs
);
4610 * Note that file-posix driver rely on permission update done during reopen
4611 * (even if no permission changed), because it wants "new" permissions for
4612 * reconfiguring the fd and that's why it does it in raw_check_perm(), not
4613 * in raw_reopen_prepare() which is called with "old" permissions.
4615 bdrv_graph_rdlock_main_loop();
4616 ret
= bdrv_list_refresh_perms(refresh_list
, bs_queue
, tran
, errp
);
4617 bdrv_graph_rdunlock_main_loop();
4624 * If we reach this point, we have success and just need to apply the
4627 * Reverse order is used to comfort qcow2 driver: on commit it need to write
4628 * IN_USE flag to the image, to mark bitmaps in the image as invalid. But
4629 * children are usually goes after parents in reopen-queue, so go from last
4632 QTAILQ_FOREACH_REVERSE(bs_entry
, bs_queue
, entry
) {
4633 ctx
= bdrv_get_aio_context(bs_entry
->state
.bs
);
4634 aio_context_acquire(ctx
);
4635 bdrv_reopen_commit(&bs_entry
->state
);
4636 aio_context_release(ctx
);
4639 bdrv_graph_wrlock(NULL
);
4641 bdrv_graph_wrunlock();
4643 QTAILQ_FOREACH_REVERSE(bs_entry
, bs_queue
, entry
) {
4644 BlockDriverState
*bs
= bs_entry
->state
.bs
;
4646 if (bs
->drv
->bdrv_reopen_commit_post
) {
4647 ctx
= bdrv_get_aio_context(bs
);
4648 aio_context_acquire(ctx
);
4649 bs
->drv
->bdrv_reopen_commit_post(&bs_entry
->state
);
4650 aio_context_release(ctx
);
4658 bdrv_graph_wrlock(NULL
);
4660 bdrv_graph_wrunlock();
4662 QTAILQ_FOREACH_SAFE(bs_entry
, bs_queue
, entry
, next
) {
4663 if (bs_entry
->prepared
) {
4664 ctx
= bdrv_get_aio_context(bs_entry
->state
.bs
);
4665 aio_context_acquire(ctx
);
4666 bdrv_reopen_abort(&bs_entry
->state
);
4667 aio_context_release(ctx
);
4672 bdrv_reopen_queue_free(bs_queue
);
4677 int bdrv_reopen(BlockDriverState
*bs
, QDict
*opts
, bool keep_old_opts
,
4680 AioContext
*ctx
= bdrv_get_aio_context(bs
);
4681 BlockReopenQueue
*queue
;
4684 GLOBAL_STATE_CODE();
4686 queue
= bdrv_reopen_queue(NULL
, bs
, opts
, keep_old_opts
);
4688 if (ctx
!= qemu_get_aio_context()) {
4689 aio_context_release(ctx
);
4691 ret
= bdrv_reopen_multiple(queue
, errp
);
4693 if (ctx
!= qemu_get_aio_context()) {
4694 aio_context_acquire(ctx
);
4700 int bdrv_reopen_set_read_only(BlockDriverState
*bs
, bool read_only
,
4703 QDict
*opts
= qdict_new();
4705 GLOBAL_STATE_CODE();
4707 qdict_put_bool(opts
, BDRV_OPT_READ_ONLY
, read_only
);
4709 return bdrv_reopen(bs
, opts
, true, errp
);
4713 * Take a BDRVReopenState and check if the value of 'backing' in the
4714 * reopen_state->options QDict is valid or not.
4716 * If 'backing' is missing from the QDict then return 0.
4718 * If 'backing' contains the node name of the backing file of
4719 * reopen_state->bs then return 0.
4721 * If 'backing' contains a different node name (or is null) then check
4722 * whether the current backing file can be replaced with the new one.
4723 * If that's the case then reopen_state->replace_backing_bs is set to
4724 * true and reopen_state->new_backing_bs contains a pointer to the new
4725 * backing BlockDriverState (or NULL).
4727 * After calling this function, the transaction @tran may only be completed
4728 * while holding a writer lock for the graph.
4730 * Return 0 on success, otherwise return < 0 and set @errp.
4732 * The caller must hold the AioContext lock of @reopen_state->bs.
4733 * @reopen_state->bs can move to a different AioContext in this function.
4734 * Callers must make sure that their AioContext locking is still correct after
4737 static int bdrv_reopen_parse_file_or_backing(BDRVReopenState
*reopen_state
,
4738 bool is_backing
, Transaction
*tran
,
4741 BlockDriverState
*bs
= reopen_state
->bs
;
4742 BlockDriverState
*new_child_bs
;
4743 BlockDriverState
*old_child_bs
= is_backing
? child_bs(bs
->backing
) :
4745 const char *child_name
= is_backing
? "backing" : "file";
4748 AioContext
*ctx
, *old_ctx
;
4751 GLOBAL_STATE_CODE();
4753 value
= qdict_get(reopen_state
->options
, child_name
);
4754 if (value
== NULL
) {
4758 switch (qobject_type(value
)) {
4760 assert(is_backing
); /* The 'file' option does not allow a null value */
4761 new_child_bs
= NULL
;
4764 str
= qstring_get_str(qobject_to(QString
, value
));
4765 new_child_bs
= bdrv_lookup_bs(NULL
, str
, errp
);
4766 if (new_child_bs
== NULL
) {
4768 } else if (bdrv_recurse_has_child(new_child_bs
, bs
)) {
4769 error_setg(errp
, "Making '%s' a %s child of '%s' would create a "
4770 "cycle", str
, child_name
, bs
->node_name
);
4776 * The options QDict has been flattened, so 'backing' and 'file'
4777 * do not allow any other data type here.
4779 g_assert_not_reached();
4782 if (old_child_bs
== new_child_bs
) {
4787 if (bdrv_skip_implicit_filters(old_child_bs
) == new_child_bs
) {
4791 if (old_child_bs
->implicit
) {
4792 error_setg(errp
, "Cannot replace implicit %s child of %s",
4793 child_name
, bs
->node_name
);
4798 if (bs
->drv
->is_filter
&& !old_child_bs
) {
4800 * Filters always have a file or a backing child, so we are trying to
4801 * change wrong child
4803 error_setg(errp
, "'%s' is a %s filter node that does not support a "
4804 "%s child", bs
->node_name
, bs
->drv
->format_name
, child_name
);
4809 reopen_state
->old_backing_bs
= old_child_bs
;
4811 reopen_state
->old_file_bs
= old_child_bs
;
4815 bdrv_ref(old_child_bs
);
4816 bdrv_drained_begin(old_child_bs
);
4819 old_ctx
= bdrv_get_aio_context(bs
);
4820 ctx
= bdrv_get_aio_context(new_child_bs
);
4821 if (old_ctx
!= ctx
) {
4822 aio_context_release(old_ctx
);
4823 aio_context_acquire(ctx
);
4826 bdrv_graph_wrlock(new_child_bs
);
4828 ret
= bdrv_set_file_or_backing_noperm(bs
, new_child_bs
, is_backing
,
4831 bdrv_graph_wrunlock();
4833 if (old_ctx
!= ctx
) {
4834 aio_context_release(ctx
);
4835 aio_context_acquire(old_ctx
);
4839 bdrv_drained_end(old_child_bs
);
4840 bdrv_unref(old_child_bs
);
4847 * Prepares a BlockDriverState for reopen. All changes are staged in the
4848 * 'opaque' field of the BDRVReopenState, which is used and allocated by
4849 * the block driver layer .bdrv_reopen_prepare()
4851 * bs is the BlockDriverState to reopen
4852 * flags are the new open flags
4853 * queue is the reopen queue
4855 * Returns 0 on success, non-zero on error. On error errp will be set
4858 * On failure, bdrv_reopen_abort() will be called to clean up any data.
4859 * It is the responsibility of the caller to then call the abort() or
4860 * commit() for any other BDS that have been left in a prepare() state
4862 * The caller must hold the AioContext lock of @reopen_state->bs.
4864 * After calling this function, the transaction @change_child_tran may only be
4865 * completed while holding a writer lock for the graph.
4867 static int bdrv_reopen_prepare(BDRVReopenState
*reopen_state
,
4868 BlockReopenQueue
*queue
,
4869 Transaction
*change_child_tran
, Error
**errp
)
4873 Error
*local_err
= NULL
;
4876 QDict
*orig_reopen_opts
;
4877 char *discard
= NULL
;
4879 bool drv_prepared
= false;
4881 assert(reopen_state
!= NULL
);
4882 assert(reopen_state
->bs
->drv
!= NULL
);
4883 GLOBAL_STATE_CODE();
4884 drv
= reopen_state
->bs
->drv
;
4886 /* This function and each driver's bdrv_reopen_prepare() remove
4887 * entries from reopen_state->options as they are processed, so
4888 * we need to make a copy of the original QDict. */
4889 orig_reopen_opts
= qdict_clone_shallow(reopen_state
->options
);
4891 /* Process generic block layer options */
4892 opts
= qemu_opts_create(&bdrv_runtime_opts
, NULL
, 0, &error_abort
);
4893 if (!qemu_opts_absorb_qdict(opts
, reopen_state
->options
, errp
)) {
4898 /* This was already called in bdrv_reopen_queue_child() so the flags
4899 * are up-to-date. This time we simply want to remove the options from
4900 * QemuOpts in order to indicate that they have been processed. */
4901 old_flags
= reopen_state
->flags
;
4902 update_flags_from_options(&reopen_state
->flags
, opts
);
4903 assert(old_flags
== reopen_state
->flags
);
4905 discard
= qemu_opt_get_del(opts
, BDRV_OPT_DISCARD
);
4906 if (discard
!= NULL
) {
4907 if (bdrv_parse_discard_flags(discard
, &reopen_state
->flags
) != 0) {
4908 error_setg(errp
, "Invalid discard option");
4914 reopen_state
->detect_zeroes
=
4915 bdrv_parse_detect_zeroes(opts
, reopen_state
->flags
, &local_err
);
4917 error_propagate(errp
, local_err
);
4922 /* All other options (including node-name and driver) must be unchanged.
4923 * Put them back into the QDict, so that they are checked at the end
4924 * of this function. */
4925 qemu_opts_to_qdict(opts
, reopen_state
->options
);
4927 /* If we are to stay read-only, do not allow permission change
4928 * to r/w. Attempting to set to r/w may fail if either BDRV_O_ALLOW_RDWR is
4929 * not set, or if the BDS still has copy_on_read enabled */
4930 read_only
= !(reopen_state
->flags
& BDRV_O_RDWR
);
4931 ret
= bdrv_can_set_read_only(reopen_state
->bs
, read_only
, true, &local_err
);
4933 error_propagate(errp
, local_err
);
4937 if (drv
->bdrv_reopen_prepare
) {
4939 * If a driver-specific option is missing, it means that we
4940 * should reset it to its default value.
4941 * But not all options allow that, so we need to check it first.
4943 ret
= bdrv_reset_options_allowed(reopen_state
->bs
,
4944 reopen_state
->options
, errp
);
4949 ret
= drv
->bdrv_reopen_prepare(reopen_state
, queue
, &local_err
);
4951 if (local_err
!= NULL
) {
4952 error_propagate(errp
, local_err
);
4954 bdrv_refresh_filename(reopen_state
->bs
);
4955 error_setg(errp
, "failed while preparing to reopen image '%s'",
4956 reopen_state
->bs
->filename
);
4961 /* It is currently mandatory to have a bdrv_reopen_prepare()
4962 * handler for each supported drv. */
4963 error_setg(errp
, "Block format '%s' used by node '%s' "
4964 "does not support reopening files", drv
->format_name
,
4965 bdrv_get_device_or_node_name(reopen_state
->bs
));
4970 drv_prepared
= true;
4973 * We must provide the 'backing' option if the BDS has a backing
4974 * file or if the image file has a backing file name as part of
4975 * its metadata. Otherwise the 'backing' option can be omitted.
4977 if (drv
->supports_backing
&& reopen_state
->backing_missing
&&
4978 (reopen_state
->bs
->backing
|| reopen_state
->bs
->backing_file
[0])) {
4979 error_setg(errp
, "backing is missing for '%s'",
4980 reopen_state
->bs
->node_name
);
4986 * Allow changing the 'backing' option. The new value can be
4987 * either a reference to an existing node (using its node name)
4988 * or NULL to simply detach the current backing file.
4990 ret
= bdrv_reopen_parse_file_or_backing(reopen_state
, true,
4991 change_child_tran
, errp
);
4995 qdict_del(reopen_state
->options
, "backing");
4997 /* Allow changing the 'file' option. In this case NULL is not allowed */
4998 ret
= bdrv_reopen_parse_file_or_backing(reopen_state
, false,
4999 change_child_tran
, errp
);
5003 qdict_del(reopen_state
->options
, "file");
5005 /* Options that are not handled are only okay if they are unchanged
5006 * compared to the old state. It is expected that some options are only
5007 * used for the initial open, but not reopen (e.g. filename) */
5008 if (qdict_size(reopen_state
->options
)) {
5009 const QDictEntry
*entry
= qdict_first(reopen_state
->options
);
5012 QObject
*new = entry
->value
;
5013 QObject
*old
= qdict_get(reopen_state
->bs
->options
, entry
->key
);
5015 /* Allow child references (child_name=node_name) as long as they
5016 * point to the current child (i.e. everything stays the same). */
5017 if (qobject_type(new) == QTYPE_QSTRING
) {
5019 QLIST_FOREACH(child
, &reopen_state
->bs
->children
, next
) {
5020 if (!strcmp(child
->name
, entry
->key
)) {
5026 if (!strcmp(child
->bs
->node_name
,
5027 qstring_get_str(qobject_to(QString
, new)))) {
5028 continue; /* Found child with this name, skip option */
5034 * TODO: When using -drive to specify blockdev options, all values
5035 * will be strings; however, when using -blockdev, blockdev-add or
5036 * filenames using the json:{} pseudo-protocol, they will be
5038 * In contrast, reopening options are (currently) always strings
5039 * (because you can only specify them through qemu-io; all other
5040 * callers do not specify any options).
5041 * Therefore, when using anything other than -drive to create a BDS,
5042 * this cannot detect non-string options as unchanged, because
5043 * qobject_is_equal() always returns false for objects of different
5044 * type. In the future, this should be remedied by correctly typing
5045 * all options. For now, this is not too big of an issue because
5046 * the user can simply omit options which cannot be changed anyway,
5047 * so they will stay unchanged.
5049 if (!qobject_is_equal(new, old
)) {
5050 error_setg(errp
, "Cannot change the option '%s'", entry
->key
);
5054 } while ((entry
= qdict_next(reopen_state
->options
, entry
)));
5059 /* Restore the original reopen_state->options QDict */
5060 qobject_unref(reopen_state
->options
);
5061 reopen_state
->options
= qobject_ref(orig_reopen_opts
);
5064 if (ret
< 0 && drv_prepared
) {
5065 /* drv->bdrv_reopen_prepare() has succeeded, so we need to
5066 * call drv->bdrv_reopen_abort() before signaling an error
5067 * (bdrv_reopen_multiple() will not call bdrv_reopen_abort()
5068 * when the respective bdrv_reopen_prepare() has failed) */
5069 if (drv
->bdrv_reopen_abort
) {
5070 drv
->bdrv_reopen_abort(reopen_state
);
5073 qemu_opts_del(opts
);
5074 qobject_unref(orig_reopen_opts
);
5080 * Takes the staged changes for the reopen from bdrv_reopen_prepare(), and
5081 * makes them final by swapping the staging BlockDriverState contents into
5082 * the active BlockDriverState contents.
5084 static void bdrv_reopen_commit(BDRVReopenState
*reopen_state
)
5087 BlockDriverState
*bs
;
5090 assert(reopen_state
!= NULL
);
5091 bs
= reopen_state
->bs
;
5093 assert(drv
!= NULL
);
5094 GLOBAL_STATE_CODE();
5096 /* If there are any driver level actions to take */
5097 if (drv
->bdrv_reopen_commit
) {
5098 drv
->bdrv_reopen_commit(reopen_state
);
5101 /* set BDS specific flags now */
5102 qobject_unref(bs
->explicit_options
);
5103 qobject_unref(bs
->options
);
5104 qobject_ref(reopen_state
->explicit_options
);
5105 qobject_ref(reopen_state
->options
);
5107 bs
->explicit_options
= reopen_state
->explicit_options
;
5108 bs
->options
= reopen_state
->options
;
5109 bs
->open_flags
= reopen_state
->flags
;
5110 bs
->detect_zeroes
= reopen_state
->detect_zeroes
;
5112 /* Remove child references from bs->options and bs->explicit_options.
5113 * Child options were already removed in bdrv_reopen_queue_child() */
5114 QLIST_FOREACH(child
, &bs
->children
, next
) {
5115 qdict_del(bs
->explicit_options
, child
->name
);
5116 qdict_del(bs
->options
, child
->name
);
5118 /* backing is probably removed, so it's not handled by previous loop */
5119 qdict_del(bs
->explicit_options
, "backing");
5120 qdict_del(bs
->options
, "backing");
5122 bdrv_graph_rdlock_main_loop();
5123 bdrv_refresh_limits(bs
, NULL
, NULL
);
5124 bdrv_graph_rdunlock_main_loop();
5125 bdrv_refresh_total_sectors(bs
, bs
->total_sectors
);
5129 * Abort the reopen, and delete and free the staged changes in
5132 static void bdrv_reopen_abort(BDRVReopenState
*reopen_state
)
5136 assert(reopen_state
!= NULL
);
5137 drv
= reopen_state
->bs
->drv
;
5138 assert(drv
!= NULL
);
5139 GLOBAL_STATE_CODE();
5141 if (drv
->bdrv_reopen_abort
) {
5142 drv
->bdrv_reopen_abort(reopen_state
);
5147 static void bdrv_close(BlockDriverState
*bs
)
5149 BdrvAioNotifier
*ban
, *ban_next
;
5150 BdrvChild
*child
, *next
;
5152 GLOBAL_STATE_CODE();
5153 assert(!bs
->refcnt
);
5155 bdrv_drained_begin(bs
); /* complete I/O */
5157 bdrv_drain(bs
); /* in case flush left pending I/O */
5160 if (bs
->drv
->bdrv_close
) {
5161 /* Must unfreeze all children, so bdrv_unref_child() works */
5162 bs
->drv
->bdrv_close(bs
);
5167 QLIST_FOREACH_SAFE(child
, &bs
->children
, next
, next
) {
5168 bdrv_unref_child(bs
, child
);
5171 assert(!bs
->backing
);
5175 qatomic_set(&bs
->copy_on_read
, 0);
5176 bs
->backing_file
[0] = '\0';
5177 bs
->backing_format
[0] = '\0';
5178 bs
->total_sectors
= 0;
5179 bs
->encrypted
= false;
5181 qobject_unref(bs
->options
);
5182 qobject_unref(bs
->explicit_options
);
5184 bs
->explicit_options
= NULL
;
5185 qobject_unref(bs
->full_open_options
);
5186 bs
->full_open_options
= NULL
;
5187 g_free(bs
->block_status_cache
);
5188 bs
->block_status_cache
= NULL
;
5190 bdrv_release_named_dirty_bitmaps(bs
);
5191 assert(QLIST_EMPTY(&bs
->dirty_bitmaps
));
5193 QLIST_FOREACH_SAFE(ban
, &bs
->aio_notifiers
, list
, ban_next
) {
5196 QLIST_INIT(&bs
->aio_notifiers
);
5197 bdrv_drained_end(bs
);
5200 * If we're still inside some bdrv_drain_all_begin()/end() sections, end
5201 * them now since this BDS won't exist anymore when bdrv_drain_all_end()
5204 if (bs
->quiesce_counter
) {
5205 bdrv_drain_all_end_quiesce(bs
);
5209 void bdrv_close_all(void)
5211 GLOBAL_STATE_CODE();
5212 assert(job_next(NULL
) == NULL
);
5214 /* Drop references from requests still in flight, such as canceled block
5215 * jobs whose AIO context has not been polled yet */
5218 blk_remove_all_bs();
5219 blockdev_close_all_bdrv_states();
5221 assert(QTAILQ_EMPTY(&all_bdrv_states
));
5224 static bool GRAPH_RDLOCK
should_update_child(BdrvChild
*c
, BlockDriverState
*to
)
5230 if (c
->klass
->stay_at_node
) {
5234 /* If the child @c belongs to the BDS @to, replacing the current
5235 * c->bs by @to would mean to create a loop.
5237 * Such a case occurs when appending a BDS to a backing chain.
5238 * For instance, imagine the following chain:
5240 * guest device -> node A -> further backing chain...
5242 * Now we create a new BDS B which we want to put on top of this
5243 * chain, so we first attach A as its backing node:
5248 * guest device -> node A -> further backing chain...
5250 * Finally we want to replace A by B. When doing that, we want to
5251 * replace all pointers to A by pointers to B -- except for the
5252 * pointer from B because (1) that would create a loop, and (2)
5253 * that pointer should simply stay intact:
5255 * guest device -> node B
5258 * node A -> further backing chain...
5260 * In general, when replacing a node A (c->bs) by a node B (@to),
5261 * if A is a child of B, that means we cannot replace A by B there
5262 * because that would create a loop. Silently detaching A from B
5263 * is also not really an option. So overall just leaving A in
5264 * place there is the most sensible choice.
5266 * We would also create a loop in any cases where @c is only
5267 * indirectly referenced by @to. Prevent this by returning false
5268 * if @c is found (by breadth-first search) anywhere in the whole
5273 found
= g_hash_table_new(NULL
, NULL
);
5274 g_hash_table_add(found
, to
);
5275 queue
= g_queue_new();
5276 g_queue_push_tail(queue
, to
);
5278 while (!g_queue_is_empty(queue
)) {
5279 BlockDriverState
*v
= g_queue_pop_head(queue
);
5282 QLIST_FOREACH(c2
, &v
->children
, next
) {
5288 if (g_hash_table_contains(found
, c2
->bs
)) {
5292 g_queue_push_tail(queue
, c2
->bs
);
5293 g_hash_table_add(found
, c2
->bs
);
5297 g_queue_free(queue
);
5298 g_hash_table_destroy(found
);
5303 static void bdrv_remove_child_commit(void *opaque
)
5305 GLOBAL_STATE_CODE();
5306 bdrv_child_free(opaque
);
5309 static TransactionActionDrv bdrv_remove_child_drv
= {
5310 .commit
= bdrv_remove_child_commit
,
5314 * Function doesn't update permissions, caller is responsible for this.
5316 * @child->bs (if non-NULL) must be drained.
5318 * After calling this function, the transaction @tran may only be completed
5319 * while holding a writer lock for the graph.
5321 static void GRAPH_WRLOCK
bdrv_remove_child(BdrvChild
*child
, Transaction
*tran
)
5328 assert(child
->quiesced_parent
);
5329 bdrv_replace_child_tran(child
, NULL
, tran
);
5332 tran_add(tran
, &bdrv_remove_child_drv
, child
);
5336 * Both @from and @to (if non-NULL) must be drained. @to must be kept drained
5337 * until the transaction is completed.
5339 * After calling this function, the transaction @tran may only be completed
5340 * while holding a writer lock for the graph.
5342 static int GRAPH_WRLOCK
5343 bdrv_replace_node_noperm(BlockDriverState
*from
,
5344 BlockDriverState
*to
,
5345 bool auto_skip
, Transaction
*tran
,
5348 BdrvChild
*c
, *next
;
5350 GLOBAL_STATE_CODE();
5352 assert(from
->quiesce_counter
);
5353 assert(to
->quiesce_counter
);
5355 QLIST_FOREACH_SAFE(c
, &from
->parents
, next_parent
, next
) {
5356 assert(c
->bs
== from
);
5357 if (!should_update_child(c
, to
)) {
5361 error_setg(errp
, "Should not change '%s' link to '%s'",
5362 c
->name
, from
->node_name
);
5366 error_setg(errp
, "Cannot change '%s' link to '%s'",
5367 c
->name
, from
->node_name
);
5370 bdrv_replace_child_tran(c
, to
, tran
);
5377 * With auto_skip=true bdrv_replace_node_common skips updating from parents
5378 * if it creates a parent-child relation loop or if parent is block-job.
5380 * With auto_skip=false the error is returned if from has a parent which should
5383 * With @detach_subchain=true @to must be in a backing chain of @from. In this
5384 * case backing link of the cow-parent of @to is removed.
5386 static int bdrv_replace_node_common(BlockDriverState
*from
,
5387 BlockDriverState
*to
,
5388 bool auto_skip
, bool detach_subchain
,
5391 Transaction
*tran
= tran_new();
5392 g_autoptr(GSList
) refresh_list
= NULL
;
5393 BlockDriverState
*to_cow_parent
= NULL
;
5396 GLOBAL_STATE_CODE();
5398 if (detach_subchain
) {
5399 assert(bdrv_chain_contains(from
, to
));
5401 for (to_cow_parent
= from
;
5402 bdrv_filter_or_cow_bs(to_cow_parent
) != to
;
5403 to_cow_parent
= bdrv_filter_or_cow_bs(to_cow_parent
))
5409 /* Make sure that @from doesn't go away until we have successfully attached
5410 * all of its parents to @to. */
5413 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
5414 assert(bdrv_get_aio_context(from
) == bdrv_get_aio_context(to
));
5415 bdrv_drained_begin(from
);
5416 bdrv_drained_begin(to
);
5418 bdrv_graph_wrlock(to
);
5421 * Do the replacement without permission update.
5422 * Replacement may influence the permissions, we should calculate new
5423 * permissions based on new graph. If we fail, we'll roll-back the
5426 ret
= bdrv_replace_node_noperm(from
, to
, auto_skip
, tran
, errp
);
5431 if (detach_subchain
) {
5432 /* to_cow_parent is already drained because from is drained */
5433 bdrv_remove_child(bdrv_filter_or_cow_child(to_cow_parent
), tran
);
5436 refresh_list
= g_slist_prepend(refresh_list
, to
);
5437 refresh_list
= g_slist_prepend(refresh_list
, from
);
5439 ret
= bdrv_list_refresh_perms(refresh_list
, NULL
, tran
, errp
);
5447 tran_finalize(tran
, ret
);
5448 bdrv_graph_wrunlock();
5450 bdrv_drained_end(to
);
5451 bdrv_drained_end(from
);
5457 int bdrv_replace_node(BlockDriverState
*from
, BlockDriverState
*to
,
5460 GLOBAL_STATE_CODE();
5462 return bdrv_replace_node_common(from
, to
, true, false, errp
);
5465 int bdrv_drop_filter(BlockDriverState
*bs
, Error
**errp
)
5467 GLOBAL_STATE_CODE();
5469 return bdrv_replace_node_common(bs
, bdrv_filter_or_cow_bs(bs
), true, true,
5474 * Add new bs contents at the top of an image chain while the chain is
5475 * live, while keeping required fields on the top layer.
5477 * This will modify the BlockDriverState fields, and swap contents
5478 * between bs_new and bs_top. Both bs_new and bs_top are modified.
5480 * bs_new must not be attached to a BlockBackend and must not have backing
5483 * This function does not create any image files.
5485 * The caller must hold the AioContext lock for @bs_top.
5487 int bdrv_append(BlockDriverState
*bs_new
, BlockDriverState
*bs_top
,
5492 Transaction
*tran
= tran_new();
5493 AioContext
*old_context
, *new_context
= NULL
;
5495 GLOBAL_STATE_CODE();
5497 assert(!bs_new
->backing
);
5499 old_context
= bdrv_get_aio_context(bs_top
);
5500 bdrv_drained_begin(bs_top
);
5503 * bdrv_drained_begin() requires that only the AioContext of the drained
5504 * node is locked, and at this point it can still differ from the AioContext
5507 new_context
= bdrv_get_aio_context(bs_new
);
5508 aio_context_release(old_context
);
5509 aio_context_acquire(new_context
);
5510 bdrv_drained_begin(bs_new
);
5511 aio_context_release(new_context
);
5512 aio_context_acquire(old_context
);
5515 bdrv_graph_wrlock(bs_top
);
5517 child
= bdrv_attach_child_noperm(bs_new
, bs_top
, "backing",
5518 &child_of_bds
, bdrv_backing_role(bs_new
),
5526 * bdrv_attach_child_noperm could change the AioContext of bs_top and
5527 * bs_new, but at least they are in the same AioContext now. This is the
5528 * AioContext that we need to lock for the rest of the function.
5530 new_context
= bdrv_get_aio_context(bs_top
);
5532 if (old_context
!= new_context
) {
5533 aio_context_release(old_context
);
5534 aio_context_acquire(new_context
);
5537 ret
= bdrv_replace_node_noperm(bs_top
, bs_new
, true, tran
, errp
);
5542 ret
= bdrv_refresh_perms(bs_new
, tran
, errp
);
5544 tran_finalize(tran
, ret
);
5546 bdrv_refresh_limits(bs_top
, NULL
, NULL
);
5547 bdrv_graph_wrunlock();
5549 bdrv_drained_end(bs_top
);
5550 bdrv_drained_end(bs_new
);
5552 if (new_context
&& old_context
!= new_context
) {
5553 aio_context_release(new_context
);
5554 aio_context_acquire(old_context
);
5560 /* Not for empty child */
5561 int bdrv_replace_child_bs(BdrvChild
*child
, BlockDriverState
*new_bs
,
5565 Transaction
*tran
= tran_new();
5566 g_autoptr(GSList
) refresh_list
= NULL
;
5567 BlockDriverState
*old_bs
= child
->bs
;
5569 GLOBAL_STATE_CODE();
5572 bdrv_drained_begin(old_bs
);
5573 bdrv_drained_begin(new_bs
);
5574 bdrv_graph_wrlock(new_bs
);
5576 bdrv_replace_child_tran(child
, new_bs
, tran
);
5578 refresh_list
= g_slist_prepend(refresh_list
, old_bs
);
5579 refresh_list
= g_slist_prepend(refresh_list
, new_bs
);
5581 ret
= bdrv_list_refresh_perms(refresh_list
, NULL
, tran
, errp
);
5583 tran_finalize(tran
, ret
);
5585 bdrv_graph_wrunlock();
5586 bdrv_drained_end(old_bs
);
5587 bdrv_drained_end(new_bs
);
5593 static void bdrv_delete(BlockDriverState
*bs
)
5595 assert(bdrv_op_blocker_is_empty(bs
));
5596 assert(!bs
->refcnt
);
5597 GLOBAL_STATE_CODE();
5599 /* remove from list, if necessary */
5600 if (bs
->node_name
[0] != '\0') {
5601 QTAILQ_REMOVE(&graph_bdrv_states
, bs
, node_list
);
5603 QTAILQ_REMOVE(&all_bdrv_states
, bs
, bs_list
);
5607 qemu_mutex_destroy(&bs
->reqs_lock
);
5614 * Replace @bs by newly created block node.
5616 * @options is a QDict of options to pass to the block drivers, or NULL for an
5617 * empty set of options. The reference to the QDict belongs to the block layer
5618 * after the call (even on failure), so if the caller intends to reuse the
5619 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
5621 * The caller holds the AioContext lock for @bs. It must make sure that @bs
5622 * stays in the same AioContext, i.e. @options must not refer to nodes in a
5623 * different AioContext.
5625 BlockDriverState
*bdrv_insert_node(BlockDriverState
*bs
, QDict
*options
,
5626 int flags
, Error
**errp
)
5630 AioContext
*ctx
= bdrv_get_aio_context(bs
);
5631 BlockDriverState
*new_node_bs
= NULL
;
5632 const char *drvname
, *node_name
;
5635 drvname
= qdict_get_try_str(options
, "driver");
5637 error_setg(errp
, "driver is not specified");
5641 drv
= bdrv_find_format(drvname
);
5643 error_setg(errp
, "Unknown driver: '%s'", drvname
);
5647 node_name
= qdict_get_try_str(options
, "node-name");
5649 GLOBAL_STATE_CODE();
5651 aio_context_release(ctx
);
5652 aio_context_acquire(qemu_get_aio_context());
5653 new_node_bs
= bdrv_new_open_driver_opts(drv
, node_name
, options
, flags
,
5655 aio_context_release(qemu_get_aio_context());
5656 aio_context_acquire(ctx
);
5657 assert(bdrv_get_aio_context(bs
) == ctx
);
5659 options
= NULL
; /* bdrv_new_open_driver() eats options */
5661 error_prepend(errp
, "Could not create node: ");
5665 bdrv_drained_begin(bs
);
5666 ret
= bdrv_replace_node(bs
, new_node_bs
, errp
);
5667 bdrv_drained_end(bs
);
5670 error_prepend(errp
, "Could not replace node: ");
5677 qobject_unref(options
);
5678 bdrv_unref(new_node_bs
);
5683 * Run consistency checks on an image
5685 * Returns 0 if the check could be completed (it doesn't mean that the image is
5686 * free of errors) or -errno when an internal error occurred. The results of the
5687 * check are stored in res.
5689 int coroutine_fn
bdrv_co_check(BlockDriverState
*bs
,
5690 BdrvCheckResult
*res
, BdrvCheckMode fix
)
5693 assert_bdrv_graph_readable();
5694 if (bs
->drv
== NULL
) {
5697 if (bs
->drv
->bdrv_co_check
== NULL
) {
5701 memset(res
, 0, sizeof(*res
));
5702 return bs
->drv
->bdrv_co_check(bs
, res
, fix
);
5708 * -EINVAL - backing format specified, but no file
5709 * -ENOSPC - can't update the backing file because no space is left in the
5711 * -ENOTSUP - format driver doesn't support changing the backing file
5713 int bdrv_change_backing_file(BlockDriverState
*bs
, const char *backing_file
,
5714 const char *backing_fmt
, bool require
)
5716 BlockDriver
*drv
= bs
->drv
;
5719 GLOBAL_STATE_CODE();
5725 /* Backing file format doesn't make sense without a backing file */
5726 if (backing_fmt
&& !backing_file
) {
5730 if (require
&& backing_file
&& !backing_fmt
) {
5734 if (drv
->bdrv_change_backing_file
!= NULL
) {
5735 ret
= drv
->bdrv_change_backing_file(bs
, backing_file
, backing_fmt
);
5741 pstrcpy(bs
->backing_file
, sizeof(bs
->backing_file
), backing_file
?: "");
5742 pstrcpy(bs
->backing_format
, sizeof(bs
->backing_format
), backing_fmt
?: "");
5743 pstrcpy(bs
->auto_backing_file
, sizeof(bs
->auto_backing_file
),
5744 backing_file
?: "");
5750 * Finds the first non-filter node above bs in the chain between
5751 * active and bs. The returned node is either an immediate parent of
5752 * bs, or there are only filter nodes between the two.
5754 * Returns NULL if bs is not found in active's image chain,
5755 * or if active == bs.
5757 * Returns the bottommost base image if bs == NULL.
5759 BlockDriverState
*bdrv_find_overlay(BlockDriverState
*active
,
5760 BlockDriverState
*bs
)
5763 GLOBAL_STATE_CODE();
5765 bs
= bdrv_skip_filters(bs
);
5766 active
= bdrv_skip_filters(active
);
5769 BlockDriverState
*next
= bdrv_backing_chain_next(active
);
5779 /* Given a BDS, searches for the base layer. */
5780 BlockDriverState
*bdrv_find_base(BlockDriverState
*bs
)
5782 GLOBAL_STATE_CODE();
5784 return bdrv_find_overlay(bs
, NULL
);
5788 * Return true if at least one of the COW (backing) and filter links
5789 * between @bs and @base is frozen. @errp is set if that's the case.
5790 * @base must be reachable from @bs, or NULL.
5792 bool bdrv_is_backing_chain_frozen(BlockDriverState
*bs
, BlockDriverState
*base
,
5795 BlockDriverState
*i
;
5798 GLOBAL_STATE_CODE();
5800 for (i
= bs
; i
!= base
; i
= child_bs(child
)) {
5801 child
= bdrv_filter_or_cow_child(i
);
5803 if (child
&& child
->frozen
) {
5804 error_setg(errp
, "Cannot change '%s' link from '%s' to '%s'",
5805 child
->name
, i
->node_name
, child
->bs
->node_name
);
5814 * Freeze all COW (backing) and filter links between @bs and @base.
5815 * If any of the links is already frozen the operation is aborted and
5816 * none of the links are modified.
5817 * @base must be reachable from @bs, or NULL.
5818 * Returns 0 on success. On failure returns < 0 and sets @errp.
5820 int bdrv_freeze_backing_chain(BlockDriverState
*bs
, BlockDriverState
*base
,
5823 BlockDriverState
*i
;
5826 GLOBAL_STATE_CODE();
5828 if (bdrv_is_backing_chain_frozen(bs
, base
, errp
)) {
5832 for (i
= bs
; i
!= base
; i
= child_bs(child
)) {
5833 child
= bdrv_filter_or_cow_child(i
);
5834 if (child
&& child
->bs
->never_freeze
) {
5835 error_setg(errp
, "Cannot freeze '%s' link to '%s'",
5836 child
->name
, child
->bs
->node_name
);
5841 for (i
= bs
; i
!= base
; i
= child_bs(child
)) {
5842 child
= bdrv_filter_or_cow_child(i
);
5844 child
->frozen
= true;
5852 * Unfreeze all COW (backing) and filter links between @bs and @base.
5853 * The caller must ensure that all links are frozen before using this
5855 * @base must be reachable from @bs, or NULL.
5857 void bdrv_unfreeze_backing_chain(BlockDriverState
*bs
, BlockDriverState
*base
)
5859 BlockDriverState
*i
;
5862 GLOBAL_STATE_CODE();
5864 for (i
= bs
; i
!= base
; i
= child_bs(child
)) {
5865 child
= bdrv_filter_or_cow_child(i
);
5867 assert(child
->frozen
);
5868 child
->frozen
= false;
5874 * Drops images above 'base' up to and including 'top', and sets the image
5875 * above 'top' to have base as its backing file.
5877 * Requires that the overlay to 'top' is opened r/w, so that the backing file
5878 * information in 'bs' can be properly updated.
5880 * E.g., this will convert the following chain:
5881 * bottom <- base <- intermediate <- top <- active
5885 * bottom <- base <- active
5887 * It is allowed for bottom==base, in which case it converts:
5889 * base <- intermediate <- top <- active
5895 * If backing_file_str is non-NULL, it will be used when modifying top's
5896 * overlay image metadata.
5899 * if active == top, that is considered an error
5902 int bdrv_drop_intermediate(BlockDriverState
*top
, BlockDriverState
*base
,
5903 const char *backing_file_str
)
5905 BlockDriverState
*explicit_top
= top
;
5906 bool update_inherits_from
;
5908 Error
*local_err
= NULL
;
5910 g_autoptr(GSList
) updated_children
= NULL
;
5913 GLOBAL_STATE_CODE();
5916 bdrv_drained_begin(base
);
5918 if (!top
->drv
|| !base
->drv
) {
5922 /* Make sure that base is in the backing chain of top */
5923 if (!bdrv_chain_contains(top
, base
)) {
5927 /* If 'base' recursively inherits from 'top' then we should set
5928 * base->inherits_from to top->inherits_from after 'top' and all
5929 * other intermediate nodes have been dropped.
5930 * If 'top' is an implicit node (e.g. "commit_top") we should skip
5931 * it because no one inherits from it. We use explicit_top for that. */
5932 explicit_top
= bdrv_skip_implicit_filters(explicit_top
);
5933 update_inherits_from
= bdrv_inherits_from_recursive(base
, explicit_top
);
5935 /* success - we can delete the intermediate states, and link top->base */
5936 if (!backing_file_str
) {
5937 bdrv_refresh_filename(base
);
5938 backing_file_str
= base
->filename
;
5941 QLIST_FOREACH(c
, &top
->parents
, next_parent
) {
5942 updated_children
= g_slist_prepend(updated_children
, c
);
5946 * It seems correct to pass detach_subchain=true here, but it triggers
5947 * one more yet not fixed bug, when due to nested aio_poll loop we switch to
5948 * another drained section, which modify the graph (for example, removing
5949 * the child, which we keep in updated_children list). So, it's a TODO.
5951 * Note, bug triggered if pass detach_subchain=true here and run
5952 * test-bdrv-drain. test_drop_intermediate_poll() test-case will crash.
5955 bdrv_replace_node_common(top
, base
, false, false, &local_err
);
5957 error_report_err(local_err
);
5961 for (p
= updated_children
; p
; p
= p
->next
) {
5964 if (c
->klass
->update_filename
) {
5965 ret
= c
->klass
->update_filename(c
, base
, backing_file_str
,
5969 * TODO: Actually, we want to rollback all previous iterations
5970 * of this loop, and (which is almost impossible) previous
5971 * bdrv_replace_node()...
5973 * Note, that c->klass->update_filename may lead to permission
5974 * update, so it's a bad idea to call it inside permission
5975 * update transaction of bdrv_replace_node.
5977 error_report_err(local_err
);
5983 if (update_inherits_from
) {
5984 base
->inherits_from
= explicit_top
->inherits_from
;
5989 bdrv_drained_end(base
);
5995 * Implementation of BlockDriver.bdrv_co_get_allocated_file_size() that
5996 * sums the size of all data-bearing children. (This excludes backing
5999 static int64_t coroutine_fn GRAPH_RDLOCK
6000 bdrv_sum_allocated_file_size(BlockDriverState
*bs
)
6003 int64_t child_size
, sum
= 0;
6005 QLIST_FOREACH(child
, &bs
->children
, next
) {
6006 if (child
->role
& (BDRV_CHILD_DATA
| BDRV_CHILD_METADATA
|
6007 BDRV_CHILD_FILTERED
))
6009 child_size
= bdrv_co_get_allocated_file_size(child
->bs
);
6010 if (child_size
< 0) {
6021 * Length of a allocated file in bytes. Sparse files are counted by actual
6022 * allocated space. Return < 0 if error or unknown.
6024 int64_t coroutine_fn
bdrv_co_get_allocated_file_size(BlockDriverState
*bs
)
6026 BlockDriver
*drv
= bs
->drv
;
6028 assert_bdrv_graph_readable();
6033 if (drv
->bdrv_co_get_allocated_file_size
) {
6034 return drv
->bdrv_co_get_allocated_file_size(bs
);
6037 if (drv
->bdrv_file_open
) {
6039 * Protocol drivers default to -ENOTSUP (most of their data is
6040 * not stored in any of their children (if they even have any),
6041 * so there is no generic way to figure it out).
6044 } else if (drv
->is_filter
) {
6045 /* Filter drivers default to the size of their filtered child */
6046 return bdrv_co_get_allocated_file_size(bdrv_filter_bs(bs
));
6048 /* Other drivers default to summing their children's sizes */
6049 return bdrv_sum_allocated_file_size(bs
);
6055 * @drv: Format driver
6056 * @opts: Creation options for new image
6057 * @in_bs: Existing image containing data for new image (may be NULL)
6058 * @errp: Error object
6059 * Returns: A #BlockMeasureInfo (free using qapi_free_BlockMeasureInfo())
6062 * Calculate file size required to create a new image.
6064 * If @in_bs is given then space for allocated clusters and zero clusters
6065 * from that image are included in the calculation. If @opts contains a
6066 * backing file that is shared by @in_bs then backing clusters may be omitted
6067 * from the calculation.
6069 * If @in_bs is NULL then the calculation includes no allocated clusters
6070 * unless a preallocation option is given in @opts.
6072 * Note that @in_bs may use a different BlockDriver from @drv.
6074 * If an error occurs the @errp pointer is set.
6076 BlockMeasureInfo
*bdrv_measure(BlockDriver
*drv
, QemuOpts
*opts
,
6077 BlockDriverState
*in_bs
, Error
**errp
)
6080 if (!drv
->bdrv_measure
) {
6081 error_setg(errp
, "Block driver '%s' does not support size measurement",
6086 return drv
->bdrv_measure(opts
, in_bs
, errp
);
6090 * Return number of sectors on success, -errno on error.
6092 int64_t coroutine_fn
bdrv_co_nb_sectors(BlockDriverState
*bs
)
6094 BlockDriver
*drv
= bs
->drv
;
6096 assert_bdrv_graph_readable();
6101 if (bs
->bl
.has_variable_length
) {
6102 int ret
= bdrv_co_refresh_total_sectors(bs
, bs
->total_sectors
);
6107 return bs
->total_sectors
;
6111 * This wrapper is written by hand because this function is in the hot I/O path,
6112 * via blk_get_geometry.
6114 int64_t coroutine_mixed_fn
bdrv_nb_sectors(BlockDriverState
*bs
)
6116 BlockDriver
*drv
= bs
->drv
;
6122 if (bs
->bl
.has_variable_length
) {
6123 int ret
= bdrv_refresh_total_sectors(bs
, bs
->total_sectors
);
6129 return bs
->total_sectors
;
6133 * Return length in bytes on success, -errno on error.
6134 * The length is always a multiple of BDRV_SECTOR_SIZE.
6136 int64_t coroutine_fn
bdrv_co_getlength(BlockDriverState
*bs
)
6140 assert_bdrv_graph_readable();
6142 ret
= bdrv_co_nb_sectors(bs
);
6146 if (ret
> INT64_MAX
/ BDRV_SECTOR_SIZE
) {
6149 return ret
* BDRV_SECTOR_SIZE
;
6152 bool bdrv_is_sg(BlockDriverState
*bs
)
6159 * Return whether the given node supports compressed writes.
6161 bool bdrv_supports_compressed_writes(BlockDriverState
*bs
)
6163 BlockDriverState
*filtered
;
6166 if (!bs
->drv
|| !block_driver_can_compress(bs
->drv
)) {
6170 filtered
= bdrv_filter_bs(bs
);
6173 * Filters can only forward compressed writes, so we have to
6176 return bdrv_supports_compressed_writes(filtered
);
6182 const char *bdrv_get_format_name(BlockDriverState
*bs
)
6185 return bs
->drv
? bs
->drv
->format_name
: NULL
;
6188 static int qsort_strcmp(const void *a
, const void *b
)
6190 return strcmp(*(char *const *)a
, *(char *const *)b
);
6193 void bdrv_iterate_format(void (*it
)(void *opaque
, const char *name
),
6194 void *opaque
, bool read_only
)
6199 const char **formats
= NULL
;
6201 GLOBAL_STATE_CODE();
6203 QLIST_FOREACH(drv
, &bdrv_drivers
, list
) {
6204 if (drv
->format_name
) {
6208 if (use_bdrv_whitelist
&& !bdrv_is_whitelisted(drv
, read_only
)) {
6212 while (formats
&& i
&& !found
) {
6213 found
= !strcmp(formats
[--i
], drv
->format_name
);
6217 formats
= g_renew(const char *, formats
, count
+ 1);
6218 formats
[count
++] = drv
->format_name
;
6223 for (i
= 0; i
< (int)ARRAY_SIZE(block_driver_modules
); i
++) {
6224 const char *format_name
= block_driver_modules
[i
].format_name
;
6230 if (use_bdrv_whitelist
&&
6231 !bdrv_format_is_whitelisted(format_name
, read_only
)) {
6235 while (formats
&& j
&& !found
) {
6236 found
= !strcmp(formats
[--j
], format_name
);
6240 formats
= g_renew(const char *, formats
, count
+ 1);
6241 formats
[count
++] = format_name
;
6246 qsort(formats
, count
, sizeof(formats
[0]), qsort_strcmp
);
6248 for (i
= 0; i
< count
; i
++) {
6249 it(opaque
, formats
[i
]);
6255 /* This function is to find a node in the bs graph */
6256 BlockDriverState
*bdrv_find_node(const char *node_name
)
6258 BlockDriverState
*bs
;
6261 GLOBAL_STATE_CODE();
6263 QTAILQ_FOREACH(bs
, &graph_bdrv_states
, node_list
) {
6264 if (!strcmp(node_name
, bs
->node_name
)) {
6271 /* Put this QMP function here so it can access the static graph_bdrv_states. */
6272 BlockDeviceInfoList
*bdrv_named_nodes_list(bool flat
,
6275 BlockDeviceInfoList
*list
;
6276 BlockDriverState
*bs
;
6278 GLOBAL_STATE_CODE();
6281 QTAILQ_FOREACH(bs
, &graph_bdrv_states
, node_list
) {
6282 BlockDeviceInfo
*info
= bdrv_block_device_info(NULL
, bs
, flat
, errp
);
6284 qapi_free_BlockDeviceInfoList(list
);
6287 QAPI_LIST_PREPEND(list
, info
);
6293 typedef struct XDbgBlockGraphConstructor
{
6294 XDbgBlockGraph
*graph
;
6295 GHashTable
*graph_nodes
;
6296 } XDbgBlockGraphConstructor
;
6298 static XDbgBlockGraphConstructor
*xdbg_graph_new(void)
6300 XDbgBlockGraphConstructor
*gr
= g_new(XDbgBlockGraphConstructor
, 1);
6302 gr
->graph
= g_new0(XDbgBlockGraph
, 1);
6303 gr
->graph_nodes
= g_hash_table_new(NULL
, NULL
);
6308 static XDbgBlockGraph
*xdbg_graph_finalize(XDbgBlockGraphConstructor
*gr
)
6310 XDbgBlockGraph
*graph
= gr
->graph
;
6312 g_hash_table_destroy(gr
->graph_nodes
);
6318 static uintptr_t xdbg_graph_node_num(XDbgBlockGraphConstructor
*gr
, void *node
)
6320 uintptr_t ret
= (uintptr_t)g_hash_table_lookup(gr
->graph_nodes
, node
);
6327 * Start counting from 1, not 0, because 0 interferes with not-found (NULL)
6328 * answer of g_hash_table_lookup.
6330 ret
= g_hash_table_size(gr
->graph_nodes
) + 1;
6331 g_hash_table_insert(gr
->graph_nodes
, node
, (void *)ret
);
6336 static void xdbg_graph_add_node(XDbgBlockGraphConstructor
*gr
, void *node
,
6337 XDbgBlockGraphNodeType type
, const char *name
)
6339 XDbgBlockGraphNode
*n
;
6341 n
= g_new0(XDbgBlockGraphNode
, 1);
6343 n
->id
= xdbg_graph_node_num(gr
, node
);
6345 n
->name
= g_strdup(name
);
6347 QAPI_LIST_PREPEND(gr
->graph
->nodes
, n
);
6350 static void xdbg_graph_add_edge(XDbgBlockGraphConstructor
*gr
, void *parent
,
6351 const BdrvChild
*child
)
6353 BlockPermission qapi_perm
;
6354 XDbgBlockGraphEdge
*edge
;
6355 GLOBAL_STATE_CODE();
6357 edge
= g_new0(XDbgBlockGraphEdge
, 1);
6359 edge
->parent
= xdbg_graph_node_num(gr
, parent
);
6360 edge
->child
= xdbg_graph_node_num(gr
, child
->bs
);
6361 edge
->name
= g_strdup(child
->name
);
6363 for (qapi_perm
= 0; qapi_perm
< BLOCK_PERMISSION__MAX
; qapi_perm
++) {
6364 uint64_t flag
= bdrv_qapi_perm_to_blk_perm(qapi_perm
);
6366 if (flag
& child
->perm
) {
6367 QAPI_LIST_PREPEND(edge
->perm
, qapi_perm
);
6369 if (flag
& child
->shared_perm
) {
6370 QAPI_LIST_PREPEND(edge
->shared_perm
, qapi_perm
);
6374 QAPI_LIST_PREPEND(gr
->graph
->edges
, edge
);
6378 XDbgBlockGraph
*bdrv_get_xdbg_block_graph(Error
**errp
)
6382 BlockDriverState
*bs
;
6384 XDbgBlockGraphConstructor
*gr
= xdbg_graph_new();
6386 GLOBAL_STATE_CODE();
6388 for (blk
= blk_all_next(NULL
); blk
; blk
= blk_all_next(blk
)) {
6389 char *allocated_name
= NULL
;
6390 const char *name
= blk_name(blk
);
6393 name
= allocated_name
= blk_get_attached_dev_id(blk
);
6395 xdbg_graph_add_node(gr
, blk
, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_BACKEND
,
6397 g_free(allocated_name
);
6398 if (blk_root(blk
)) {
6399 xdbg_graph_add_edge(gr
, blk
, blk_root(blk
));
6403 WITH_JOB_LOCK_GUARD() {
6404 for (job
= block_job_next_locked(NULL
); job
;
6405 job
= block_job_next_locked(job
)) {
6408 xdbg_graph_add_node(gr
, job
, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_JOB
,
6410 for (el
= job
->nodes
; el
; el
= el
->next
) {
6411 xdbg_graph_add_edge(gr
, job
, (BdrvChild
*)el
->data
);
6416 QTAILQ_FOREACH(bs
, &graph_bdrv_states
, node_list
) {
6417 xdbg_graph_add_node(gr
, bs
, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_DRIVER
,
6419 QLIST_FOREACH(child
, &bs
->children
, next
) {
6420 xdbg_graph_add_edge(gr
, bs
, child
);
6424 return xdbg_graph_finalize(gr
);
6427 BlockDriverState
*bdrv_lookup_bs(const char *device
,
6428 const char *node_name
,
6432 BlockDriverState
*bs
;
6434 GLOBAL_STATE_CODE();
6437 blk
= blk_by_name(device
);
6442 error_setg(errp
, "Device '%s' has no medium", device
);
6450 bs
= bdrv_find_node(node_name
);
6457 error_setg(errp
, "Cannot find device=\'%s\' nor node-name=\'%s\'",
6458 device
? device
: "",
6459 node_name
? node_name
: "");
6463 /* If 'base' is in the same chain as 'top', return true. Otherwise,
6464 * return false. If either argument is NULL, return false. */
6465 bool bdrv_chain_contains(BlockDriverState
*top
, BlockDriverState
*base
)
6468 GLOBAL_STATE_CODE();
6470 while (top
&& top
!= base
) {
6471 top
= bdrv_filter_or_cow_bs(top
);
6477 BlockDriverState
*bdrv_next_node(BlockDriverState
*bs
)
6479 GLOBAL_STATE_CODE();
6481 return QTAILQ_FIRST(&graph_bdrv_states
);
6483 return QTAILQ_NEXT(bs
, node_list
);
6486 BlockDriverState
*bdrv_next_all_states(BlockDriverState
*bs
)
6488 GLOBAL_STATE_CODE();
6490 return QTAILQ_FIRST(&all_bdrv_states
);
6492 return QTAILQ_NEXT(bs
, bs_list
);
6495 const char *bdrv_get_node_name(const BlockDriverState
*bs
)
6498 return bs
->node_name
;
6501 const char *bdrv_get_parent_name(const BlockDriverState
*bs
)
6507 /* If multiple parents have a name, just pick the first one. */
6508 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
6509 if (c
->klass
->get_name
) {
6510 name
= c
->klass
->get_name(c
);
6511 if (name
&& *name
) {
6520 /* TODO check what callers really want: bs->node_name or blk_name() */
6521 const char *bdrv_get_device_name(const BlockDriverState
*bs
)
6524 return bdrv_get_parent_name(bs
) ?: "";
6527 /* This can be used to identify nodes that might not have a device
6528 * name associated. Since node and device names live in the same
6529 * namespace, the result is unambiguous. The exception is if both are
6530 * absent, then this returns an empty (non-null) string. */
6531 const char *bdrv_get_device_or_node_name(const BlockDriverState
*bs
)
6534 return bdrv_get_parent_name(bs
) ?: bs
->node_name
;
6537 int bdrv_get_flags(BlockDriverState
*bs
)
6540 return bs
->open_flags
;
6543 int bdrv_has_zero_init_1(BlockDriverState
*bs
)
6545 GLOBAL_STATE_CODE();
6549 int bdrv_has_zero_init(BlockDriverState
*bs
)
6551 BlockDriverState
*filtered
;
6552 GLOBAL_STATE_CODE();
6558 /* If BS is a copy on write image, it is initialized to
6559 the contents of the base image, which may not be zeroes. */
6560 if (bdrv_cow_child(bs
)) {
6563 if (bs
->drv
->bdrv_has_zero_init
) {
6564 return bs
->drv
->bdrv_has_zero_init(bs
);
6567 filtered
= bdrv_filter_bs(bs
);
6569 return bdrv_has_zero_init(filtered
);
6576 bool bdrv_can_write_zeroes_with_unmap(BlockDriverState
*bs
)
6579 if (!(bs
->open_flags
& BDRV_O_UNMAP
)) {
6583 return bs
->supported_zero_flags
& BDRV_REQ_MAY_UNMAP
;
6586 void bdrv_get_backing_filename(BlockDriverState
*bs
,
6587 char *filename
, int filename_size
)
6590 pstrcpy(filename
, filename_size
, bs
->backing_file
);
6593 int coroutine_fn
bdrv_co_get_info(BlockDriverState
*bs
, BlockDriverInfo
*bdi
)
6596 BlockDriver
*drv
= bs
->drv
;
6598 assert_bdrv_graph_readable();
6600 /* if bs->drv == NULL, bs is closed, so there's nothing to do here */
6604 if (!drv
->bdrv_co_get_info
) {
6605 BlockDriverState
*filtered
= bdrv_filter_bs(bs
);
6607 return bdrv_co_get_info(filtered
, bdi
);
6611 memset(bdi
, 0, sizeof(*bdi
));
6612 ret
= drv
->bdrv_co_get_info(bs
, bdi
);
6613 if (bdi
->subcluster_size
== 0) {
6615 * If the driver left this unset, subclusters are not supported.
6616 * Then it is safe to treat each cluster as having only one subcluster.
6618 bdi
->subcluster_size
= bdi
->cluster_size
;
6624 if (bdi
->cluster_size
> BDRV_MAX_ALIGNMENT
) {
6631 ImageInfoSpecific
*bdrv_get_specific_info(BlockDriverState
*bs
,
6634 BlockDriver
*drv
= bs
->drv
;
6636 if (drv
&& drv
->bdrv_get_specific_info
) {
6637 return drv
->bdrv_get_specific_info(bs
, errp
);
6642 BlockStatsSpecific
*bdrv_get_specific_stats(BlockDriverState
*bs
)
6644 BlockDriver
*drv
= bs
->drv
;
6646 if (!drv
|| !drv
->bdrv_get_specific_stats
) {
6649 return drv
->bdrv_get_specific_stats(bs
);
6652 void coroutine_fn
bdrv_co_debug_event(BlockDriverState
*bs
, BlkdebugEvent event
)
6655 assert_bdrv_graph_readable();
6657 if (!bs
|| !bs
->drv
|| !bs
->drv
->bdrv_co_debug_event
) {
6661 bs
->drv
->bdrv_co_debug_event(bs
, event
);
6664 static BlockDriverState
*bdrv_find_debug_node(BlockDriverState
*bs
)
6666 GLOBAL_STATE_CODE();
6667 while (bs
&& bs
->drv
&& !bs
->drv
->bdrv_debug_breakpoint
) {
6668 bs
= bdrv_primary_bs(bs
);
6671 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_breakpoint
) {
6672 assert(bs
->drv
->bdrv_debug_remove_breakpoint
);
6679 int bdrv_debug_breakpoint(BlockDriverState
*bs
, const char *event
,
6682 GLOBAL_STATE_CODE();
6683 bs
= bdrv_find_debug_node(bs
);
6685 return bs
->drv
->bdrv_debug_breakpoint(bs
, event
, tag
);
6691 int bdrv_debug_remove_breakpoint(BlockDriverState
*bs
, const char *tag
)
6693 GLOBAL_STATE_CODE();
6694 bs
= bdrv_find_debug_node(bs
);
6696 return bs
->drv
->bdrv_debug_remove_breakpoint(bs
, tag
);
6702 int bdrv_debug_resume(BlockDriverState
*bs
, const char *tag
)
6704 GLOBAL_STATE_CODE();
6705 while (bs
&& (!bs
->drv
|| !bs
->drv
->bdrv_debug_resume
)) {
6706 bs
= bdrv_primary_bs(bs
);
6709 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_resume
) {
6710 return bs
->drv
->bdrv_debug_resume(bs
, tag
);
6716 bool bdrv_debug_is_suspended(BlockDriverState
*bs
, const char *tag
)
6718 GLOBAL_STATE_CODE();
6719 while (bs
&& bs
->drv
&& !bs
->drv
->bdrv_debug_is_suspended
) {
6720 bs
= bdrv_primary_bs(bs
);
6723 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_is_suspended
) {
6724 return bs
->drv
->bdrv_debug_is_suspended(bs
, tag
);
6730 /* backing_file can either be relative, or absolute, or a protocol. If it is
6731 * relative, it must be relative to the chain. So, passing in bs->filename
6732 * from a BDS as backing_file should not be done, as that may be relative to
6733 * the CWD rather than the chain. */
6734 BlockDriverState
*bdrv_find_backing_image(BlockDriverState
*bs
,
6735 const char *backing_file
)
6737 char *filename_full
= NULL
;
6738 char *backing_file_full
= NULL
;
6739 char *filename_tmp
= NULL
;
6740 int is_protocol
= 0;
6741 bool filenames_refreshed
= false;
6742 BlockDriverState
*curr_bs
= NULL
;
6743 BlockDriverState
*retval
= NULL
;
6744 BlockDriverState
*bs_below
;
6746 GLOBAL_STATE_CODE();
6748 if (!bs
|| !bs
->drv
|| !backing_file
) {
6752 filename_full
= g_malloc(PATH_MAX
);
6753 backing_file_full
= g_malloc(PATH_MAX
);
6755 is_protocol
= path_has_protocol(backing_file
);
6758 * Being largely a legacy function, skip any filters here
6759 * (because filters do not have normal filenames, so they cannot
6760 * match anyway; and allowing json:{} filenames is a bit out of
6763 for (curr_bs
= bdrv_skip_filters(bs
);
6764 bdrv_cow_child(curr_bs
) != NULL
;
6767 bs_below
= bdrv_backing_chain_next(curr_bs
);
6769 if (bdrv_backing_overridden(curr_bs
)) {
6771 * If the backing file was overridden, we can only compare
6772 * directly against the backing node's filename.
6775 if (!filenames_refreshed
) {
6777 * This will automatically refresh all of the
6778 * filenames in the rest of the backing chain, so we
6779 * only need to do this once.
6781 bdrv_refresh_filename(bs_below
);
6782 filenames_refreshed
= true;
6785 if (strcmp(backing_file
, bs_below
->filename
) == 0) {
6789 } else if (is_protocol
|| path_has_protocol(curr_bs
->backing_file
)) {
6791 * If either of the filename paths is actually a protocol, then
6792 * compare unmodified paths; otherwise make paths relative.
6794 char *backing_file_full_ret
;
6796 if (strcmp(backing_file
, curr_bs
->backing_file
) == 0) {
6800 /* Also check against the full backing filename for the image */
6801 backing_file_full_ret
= bdrv_get_full_backing_filename(curr_bs
,
6803 if (backing_file_full_ret
) {
6804 bool equal
= strcmp(backing_file
, backing_file_full_ret
) == 0;
6805 g_free(backing_file_full_ret
);
6812 /* If not an absolute filename path, make it relative to the current
6813 * image's filename path */
6814 filename_tmp
= bdrv_make_absolute_filename(curr_bs
, backing_file
,
6816 /* We are going to compare canonicalized absolute pathnames */
6817 if (!filename_tmp
|| !realpath(filename_tmp
, filename_full
)) {
6818 g_free(filename_tmp
);
6821 g_free(filename_tmp
);
6823 /* We need to make sure the backing filename we are comparing against
6824 * is relative to the current image filename (or absolute) */
6825 filename_tmp
= bdrv_get_full_backing_filename(curr_bs
, NULL
);
6826 if (!filename_tmp
|| !realpath(filename_tmp
, backing_file_full
)) {
6827 g_free(filename_tmp
);
6830 g_free(filename_tmp
);
6832 if (strcmp(backing_file_full
, filename_full
) == 0) {
6839 g_free(filename_full
);
6840 g_free(backing_file_full
);
6844 void bdrv_init(void)
6846 #ifdef CONFIG_BDRV_WHITELIST_TOOLS
6847 use_bdrv_whitelist
= 1;
6849 module_call_init(MODULE_INIT_BLOCK
);
6852 void bdrv_init_with_whitelist(void)
6854 use_bdrv_whitelist
= 1;
6858 int bdrv_activate(BlockDriverState
*bs
, Error
**errp
)
6860 BdrvChild
*child
, *parent
;
6861 Error
*local_err
= NULL
;
6863 BdrvDirtyBitmap
*bm
;
6865 GLOBAL_STATE_CODE();
6866 GRAPH_RDLOCK_GUARD_MAINLOOP();
6872 QLIST_FOREACH(child
, &bs
->children
, next
) {
6873 bdrv_activate(child
->bs
, &local_err
);
6875 error_propagate(errp
, local_err
);
6881 * Update permissions, they may differ for inactive nodes.
6883 * Note that the required permissions of inactive images are always a
6884 * subset of the permissions required after activating the image. This
6885 * allows us to just get the permissions upfront without restricting
6886 * bdrv_co_invalidate_cache().
6888 * It also means that in error cases, we don't have to try and revert to
6889 * the old permissions (which is an operation that could fail, too). We can
6890 * just keep the extended permissions for the next time that an activation
6891 * of the image is tried.
6893 if (bs
->open_flags
& BDRV_O_INACTIVE
) {
6894 bs
->open_flags
&= ~BDRV_O_INACTIVE
;
6895 ret
= bdrv_refresh_perms(bs
, NULL
, errp
);
6897 bs
->open_flags
|= BDRV_O_INACTIVE
;
6901 ret
= bdrv_invalidate_cache(bs
, errp
);
6903 bs
->open_flags
|= BDRV_O_INACTIVE
;
6907 FOR_EACH_DIRTY_BITMAP(bs
, bm
) {
6908 bdrv_dirty_bitmap_skip_store(bm
, false);
6911 ret
= bdrv_refresh_total_sectors(bs
, bs
->total_sectors
);
6913 bs
->open_flags
|= BDRV_O_INACTIVE
;
6914 error_setg_errno(errp
, -ret
, "Could not refresh total sector count");
6919 QLIST_FOREACH(parent
, &bs
->parents
, next_parent
) {
6920 if (parent
->klass
->activate
) {
6921 parent
->klass
->activate(parent
, &local_err
);
6923 bs
->open_flags
|= BDRV_O_INACTIVE
;
6924 error_propagate(errp
, local_err
);
6933 int coroutine_fn
bdrv_co_invalidate_cache(BlockDriverState
*bs
, Error
**errp
)
6935 Error
*local_err
= NULL
;
6938 assert(!(bs
->open_flags
& BDRV_O_INACTIVE
));
6939 assert_bdrv_graph_readable();
6941 if (bs
->drv
->bdrv_co_invalidate_cache
) {
6942 bs
->drv
->bdrv_co_invalidate_cache(bs
, &local_err
);
6944 error_propagate(errp
, local_err
);
6952 void bdrv_activate_all(Error
**errp
)
6954 BlockDriverState
*bs
;
6955 BdrvNextIterator it
;
6957 GLOBAL_STATE_CODE();
6959 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
6960 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
6963 aio_context_acquire(aio_context
);
6964 ret
= bdrv_activate(bs
, errp
);
6965 aio_context_release(aio_context
);
6967 bdrv_next_cleanup(&it
);
6973 static bool bdrv_has_bds_parent(BlockDriverState
*bs
, bool only_active
)
6976 GLOBAL_STATE_CODE();
6978 QLIST_FOREACH(parent
, &bs
->parents
, next_parent
) {
6979 if (parent
->klass
->parent_is_bds
) {
6980 BlockDriverState
*parent_bs
= parent
->opaque
;
6981 if (!only_active
|| !(parent_bs
->open_flags
& BDRV_O_INACTIVE
)) {
6990 static int bdrv_inactivate_recurse(BlockDriverState
*bs
)
6992 BdrvChild
*child
, *parent
;
6994 uint64_t cumulative_perms
, cumulative_shared_perms
;
6996 GLOBAL_STATE_CODE();
6997 GRAPH_RDLOCK_GUARD_MAINLOOP();
7003 /* Make sure that we don't inactivate a child before its parent.
7004 * It will be covered by recursion from the yet active parent. */
7005 if (bdrv_has_bds_parent(bs
, true)) {
7009 assert(!(bs
->open_flags
& BDRV_O_INACTIVE
));
7011 /* Inactivate this node */
7012 if (bs
->drv
->bdrv_inactivate
) {
7013 ret
= bs
->drv
->bdrv_inactivate(bs
);
7019 QLIST_FOREACH(parent
, &bs
->parents
, next_parent
) {
7020 if (parent
->klass
->inactivate
) {
7021 ret
= parent
->klass
->inactivate(parent
);
7028 bdrv_get_cumulative_perm(bs
, &cumulative_perms
,
7029 &cumulative_shared_perms
);
7030 if (cumulative_perms
& (BLK_PERM_WRITE
| BLK_PERM_WRITE_UNCHANGED
)) {
7031 /* Our inactive parents still need write access. Inactivation failed. */
7035 bs
->open_flags
|= BDRV_O_INACTIVE
;
7038 * Update permissions, they may differ for inactive nodes.
7039 * We only tried to loosen restrictions, so errors are not fatal, ignore
7042 bdrv_refresh_perms(bs
, NULL
, NULL
);
7044 /* Recursively inactivate children */
7045 QLIST_FOREACH(child
, &bs
->children
, next
) {
7046 ret
= bdrv_inactivate_recurse(child
->bs
);
7055 int bdrv_inactivate_all(void)
7057 BlockDriverState
*bs
= NULL
;
7058 BdrvNextIterator it
;
7060 GSList
*aio_ctxs
= NULL
, *ctx
;
7062 GLOBAL_STATE_CODE();
7064 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
7065 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
7067 if (!g_slist_find(aio_ctxs
, aio_context
)) {
7068 aio_ctxs
= g_slist_prepend(aio_ctxs
, aio_context
);
7069 aio_context_acquire(aio_context
);
7073 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
7074 /* Nodes with BDS parents are covered by recursion from the last
7075 * parent that gets inactivated. Don't inactivate them a second
7076 * time if that has already happened. */
7077 if (bdrv_has_bds_parent(bs
, false)) {
7080 ret
= bdrv_inactivate_recurse(bs
);
7082 bdrv_next_cleanup(&it
);
7088 for (ctx
= aio_ctxs
; ctx
!= NULL
; ctx
= ctx
->next
) {
7089 AioContext
*aio_context
= ctx
->data
;
7090 aio_context_release(aio_context
);
7092 g_slist_free(aio_ctxs
);
7097 /**************************************************************/
7098 /* removable device support */
7101 * Return TRUE if the media is present
7103 bool coroutine_fn
bdrv_co_is_inserted(BlockDriverState
*bs
)
7105 BlockDriver
*drv
= bs
->drv
;
7108 assert_bdrv_graph_readable();
7113 if (drv
->bdrv_co_is_inserted
) {
7114 return drv
->bdrv_co_is_inserted(bs
);
7116 QLIST_FOREACH(child
, &bs
->children
, next
) {
7117 if (!bdrv_co_is_inserted(child
->bs
)) {
7125 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
7127 void coroutine_fn
bdrv_co_eject(BlockDriverState
*bs
, bool eject_flag
)
7129 BlockDriver
*drv
= bs
->drv
;
7131 assert_bdrv_graph_readable();
7133 if (drv
&& drv
->bdrv_co_eject
) {
7134 drv
->bdrv_co_eject(bs
, eject_flag
);
7139 * Lock or unlock the media (if it is locked, the user won't be able
7140 * to eject it manually).
7142 void coroutine_fn
bdrv_co_lock_medium(BlockDriverState
*bs
, bool locked
)
7144 BlockDriver
*drv
= bs
->drv
;
7146 assert_bdrv_graph_readable();
7147 trace_bdrv_lock_medium(bs
, locked
);
7149 if (drv
&& drv
->bdrv_co_lock_medium
) {
7150 drv
->bdrv_co_lock_medium(bs
, locked
);
7154 /* Get a reference to bs */
7155 void bdrv_ref(BlockDriverState
*bs
)
7157 GLOBAL_STATE_CODE();
7161 /* Release a previously grabbed reference to bs.
7162 * If after releasing, reference count is zero, the BlockDriverState is
7164 void bdrv_unref(BlockDriverState
*bs
)
7166 GLOBAL_STATE_CODE();
7170 assert(bs
->refcnt
> 0);
7171 if (--bs
->refcnt
== 0) {
7177 * Release a BlockDriverState reference while holding the graph write lock.
7179 * Calling bdrv_unref() directly is forbidden while holding the graph lock
7180 * because bdrv_close() both involves polling and taking the graph lock
7181 * internally. bdrv_schedule_unref() instead delays decreasing the refcount and
7182 * possibly closing @bs until the graph lock is released.
7184 void bdrv_schedule_unref(BlockDriverState
*bs
)
7189 aio_bh_schedule_oneshot(qemu_get_aio_context(),
7190 (QEMUBHFunc
*) bdrv_unref
, bs
);
7193 struct BdrvOpBlocker
{
7195 QLIST_ENTRY(BdrvOpBlocker
) list
;
7198 bool bdrv_op_is_blocked(BlockDriverState
*bs
, BlockOpType op
, Error
**errp
)
7200 BdrvOpBlocker
*blocker
;
7201 GLOBAL_STATE_CODE();
7202 assert((int) op
>= 0 && op
< BLOCK_OP_TYPE_MAX
);
7203 if (!QLIST_EMPTY(&bs
->op_blockers
[op
])) {
7204 blocker
= QLIST_FIRST(&bs
->op_blockers
[op
]);
7205 error_propagate_prepend(errp
, error_copy(blocker
->reason
),
7206 "Node '%s' is busy: ",
7207 bdrv_get_device_or_node_name(bs
));
7213 void bdrv_op_block(BlockDriverState
*bs
, BlockOpType op
, Error
*reason
)
7215 BdrvOpBlocker
*blocker
;
7216 GLOBAL_STATE_CODE();
7217 assert((int) op
>= 0 && op
< BLOCK_OP_TYPE_MAX
);
7219 blocker
= g_new0(BdrvOpBlocker
, 1);
7220 blocker
->reason
= reason
;
7221 QLIST_INSERT_HEAD(&bs
->op_blockers
[op
], blocker
, list
);
7224 void bdrv_op_unblock(BlockDriverState
*bs
, BlockOpType op
, Error
*reason
)
7226 BdrvOpBlocker
*blocker
, *next
;
7227 GLOBAL_STATE_CODE();
7228 assert((int) op
>= 0 && op
< BLOCK_OP_TYPE_MAX
);
7229 QLIST_FOREACH_SAFE(blocker
, &bs
->op_blockers
[op
], list
, next
) {
7230 if (blocker
->reason
== reason
) {
7231 QLIST_REMOVE(blocker
, list
);
7237 void bdrv_op_block_all(BlockDriverState
*bs
, Error
*reason
)
7240 GLOBAL_STATE_CODE();
7241 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
7242 bdrv_op_block(bs
, i
, reason
);
7246 void bdrv_op_unblock_all(BlockDriverState
*bs
, Error
*reason
)
7249 GLOBAL_STATE_CODE();
7250 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
7251 bdrv_op_unblock(bs
, i
, reason
);
7255 bool bdrv_op_blocker_is_empty(BlockDriverState
*bs
)
7258 GLOBAL_STATE_CODE();
7259 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
7260 if (!QLIST_EMPTY(&bs
->op_blockers
[i
])) {
7268 * Must not be called while holding the lock of an AioContext other than the
7271 void bdrv_img_create(const char *filename
, const char *fmt
,
7272 const char *base_filename
, const char *base_fmt
,
7273 char *options
, uint64_t img_size
, int flags
, bool quiet
,
7276 QemuOptsList
*create_opts
= NULL
;
7277 QemuOpts
*opts
= NULL
;
7278 const char *backing_fmt
, *backing_file
;
7280 BlockDriver
*drv
, *proto_drv
;
7281 Error
*local_err
= NULL
;
7284 GLOBAL_STATE_CODE();
7286 /* Find driver and parse its options */
7287 drv
= bdrv_find_format(fmt
);
7289 error_setg(errp
, "Unknown file format '%s'", fmt
);
7293 proto_drv
= bdrv_find_protocol(filename
, true, errp
);
7298 if (!drv
->create_opts
) {
7299 error_setg(errp
, "Format driver '%s' does not support image creation",
7304 if (!proto_drv
->create_opts
) {
7305 error_setg(errp
, "Protocol driver '%s' does not support image creation",
7306 proto_drv
->format_name
);
7310 aio_context_acquire(qemu_get_aio_context());
7312 /* Create parameter list */
7313 create_opts
= qemu_opts_append(create_opts
, drv
->create_opts
);
7314 create_opts
= qemu_opts_append(create_opts
, proto_drv
->create_opts
);
7316 opts
= qemu_opts_create(create_opts
, NULL
, 0, &error_abort
);
7318 /* Parse -o options */
7320 if (!qemu_opts_do_parse(opts
, options
, NULL
, errp
)) {
7325 if (!qemu_opt_get(opts
, BLOCK_OPT_SIZE
)) {
7326 qemu_opt_set_number(opts
, BLOCK_OPT_SIZE
, img_size
, &error_abort
);
7327 } else if (img_size
!= UINT64_C(-1)) {
7328 error_setg(errp
, "The image size must be specified only once");
7332 if (base_filename
) {
7333 if (!qemu_opt_set(opts
, BLOCK_OPT_BACKING_FILE
, base_filename
,
7335 error_setg(errp
, "Backing file not supported for file format '%s'",
7342 if (!qemu_opt_set(opts
, BLOCK_OPT_BACKING_FMT
, base_fmt
, NULL
)) {
7343 error_setg(errp
, "Backing file format not supported for file "
7344 "format '%s'", fmt
);
7349 backing_file
= qemu_opt_get(opts
, BLOCK_OPT_BACKING_FILE
);
7351 if (!strcmp(filename
, backing_file
)) {
7352 error_setg(errp
, "Error: Trying to create an image with the "
7353 "same filename as the backing file");
7356 if (backing_file
[0] == '\0') {
7357 error_setg(errp
, "Expected backing file name, got empty string");
7362 backing_fmt
= qemu_opt_get(opts
, BLOCK_OPT_BACKING_FMT
);
7364 /* The size for the image must always be specified, unless we have a backing
7365 * file and we have not been forbidden from opening it. */
7366 size
= qemu_opt_get_size(opts
, BLOCK_OPT_SIZE
, img_size
);
7367 if (backing_file
&& !(flags
& BDRV_O_NO_BACKING
)) {
7368 BlockDriverState
*bs
;
7371 QDict
*backing_options
= NULL
;
7374 bdrv_get_full_backing_filename_from_filename(filename
, backing_file
,
7379 assert(full_backing
);
7382 * No need to do I/O here, which allows us to open encrypted
7383 * backing images without needing the secret
7386 back_flags
&= ~(BDRV_O_RDWR
| BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
);
7387 back_flags
|= BDRV_O_NO_IO
;
7389 backing_options
= qdict_new();
7391 qdict_put_str(backing_options
, "driver", backing_fmt
);
7393 qdict_put_bool(backing_options
, BDRV_OPT_FORCE_SHARE
, true);
7395 bs
= bdrv_open(full_backing
, NULL
, backing_options
, back_flags
,
7397 g_free(full_backing
);
7399 error_append_hint(&local_err
, "Could not open backing image.\n");
7403 error_setg(&local_err
,
7404 "Backing file specified without backing format");
7405 error_append_hint(&local_err
, "Detected format of %s.\n",
7406 bs
->drv
->format_name
);
7410 /* Opened BS, have no size */
7411 size
= bdrv_getlength(bs
);
7413 error_setg_errno(errp
, -size
, "Could not get size of '%s'",
7418 qemu_opt_set_number(opts
, BLOCK_OPT_SIZE
, size
, &error_abort
);
7422 /* (backing_file && !(flags & BDRV_O_NO_BACKING)) */
7423 } else if (backing_file
&& !backing_fmt
) {
7424 error_setg(&local_err
,
7425 "Backing file specified without backing format");
7430 error_setg(errp
, "Image creation needs a size parameter");
7435 printf("Formatting '%s', fmt=%s ", filename
, fmt
);
7436 qemu_opts_print(opts
, " ");
7441 ret
= bdrv_create(drv
, filename
, opts
, &local_err
);
7443 if (ret
== -EFBIG
) {
7444 /* This is generally a better message than whatever the driver would
7445 * deliver (especially because of the cluster_size_hint), since that
7446 * is most probably not much different from "image too large". */
7447 const char *cluster_size_hint
= "";
7448 if (qemu_opt_get_size(opts
, BLOCK_OPT_CLUSTER_SIZE
, 0)) {
7449 cluster_size_hint
= " (try using a larger cluster size)";
7451 error_setg(errp
, "The image size is too large for file format '%s'"
7452 "%s", fmt
, cluster_size_hint
);
7453 error_free(local_err
);
7458 qemu_opts_del(opts
);
7459 qemu_opts_free(create_opts
);
7460 error_propagate(errp
, local_err
);
7461 aio_context_release(qemu_get_aio_context());
7464 AioContext
*bdrv_get_aio_context(BlockDriverState
*bs
)
7467 return bs
? bs
->aio_context
: qemu_get_aio_context();
7470 AioContext
*coroutine_fn
bdrv_co_enter(BlockDriverState
*bs
)
7472 Coroutine
*self
= qemu_coroutine_self();
7473 AioContext
*old_ctx
= qemu_coroutine_get_aio_context(self
);
7474 AioContext
*new_ctx
;
7478 * Increase bs->in_flight to ensure that this operation is completed before
7479 * moving the node to a different AioContext. Read new_ctx only afterwards.
7481 bdrv_inc_in_flight(bs
);
7483 new_ctx
= bdrv_get_aio_context(bs
);
7484 aio_co_reschedule_self(new_ctx
);
7488 void coroutine_fn
bdrv_co_leave(BlockDriverState
*bs
, AioContext
*old_ctx
)
7491 aio_co_reschedule_self(old_ctx
);
7492 bdrv_dec_in_flight(bs
);
7495 void coroutine_fn
bdrv_co_lock(BlockDriverState
*bs
)
7497 AioContext
*ctx
= bdrv_get_aio_context(bs
);
7499 /* In the main thread, bs->aio_context won't change concurrently */
7500 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
7503 * We're in coroutine context, so we already hold the lock of the main
7504 * loop AioContext. Don't lock it twice to avoid deadlocks.
7506 assert(qemu_in_coroutine());
7507 if (ctx
!= qemu_get_aio_context()) {
7508 aio_context_acquire(ctx
);
7512 void coroutine_fn
bdrv_co_unlock(BlockDriverState
*bs
)
7514 AioContext
*ctx
= bdrv_get_aio_context(bs
);
7516 assert(qemu_in_coroutine());
7517 if (ctx
!= qemu_get_aio_context()) {
7518 aio_context_release(ctx
);
7522 static void bdrv_do_remove_aio_context_notifier(BdrvAioNotifier
*ban
)
7524 GLOBAL_STATE_CODE();
7525 QLIST_REMOVE(ban
, list
);
7529 static void bdrv_detach_aio_context(BlockDriverState
*bs
)
7531 BdrvAioNotifier
*baf
, *baf_tmp
;
7533 assert(!bs
->walking_aio_notifiers
);
7534 GLOBAL_STATE_CODE();
7535 bs
->walking_aio_notifiers
= true;
7536 QLIST_FOREACH_SAFE(baf
, &bs
->aio_notifiers
, list
, baf_tmp
) {
7538 bdrv_do_remove_aio_context_notifier(baf
);
7540 baf
->detach_aio_context(baf
->opaque
);
7543 /* Never mind iterating again to check for ->deleted. bdrv_close() will
7544 * remove remaining aio notifiers if we aren't called again.
7546 bs
->walking_aio_notifiers
= false;
7548 if (bs
->drv
&& bs
->drv
->bdrv_detach_aio_context
) {
7549 bs
->drv
->bdrv_detach_aio_context(bs
);
7552 bs
->aio_context
= NULL
;
7555 static void bdrv_attach_aio_context(BlockDriverState
*bs
,
7556 AioContext
*new_context
)
7558 BdrvAioNotifier
*ban
, *ban_tmp
;
7559 GLOBAL_STATE_CODE();
7561 bs
->aio_context
= new_context
;
7563 if (bs
->drv
&& bs
->drv
->bdrv_attach_aio_context
) {
7564 bs
->drv
->bdrv_attach_aio_context(bs
, new_context
);
7567 assert(!bs
->walking_aio_notifiers
);
7568 bs
->walking_aio_notifiers
= true;
7569 QLIST_FOREACH_SAFE(ban
, &bs
->aio_notifiers
, list
, ban_tmp
) {
7571 bdrv_do_remove_aio_context_notifier(ban
);
7573 ban
->attached_aio_context(new_context
, ban
->opaque
);
7576 bs
->walking_aio_notifiers
= false;
7579 typedef struct BdrvStateSetAioContext
{
7580 AioContext
*new_ctx
;
7581 BlockDriverState
*bs
;
7582 } BdrvStateSetAioContext
;
7584 static bool bdrv_parent_change_aio_context(BdrvChild
*c
, AioContext
*ctx
,
7585 GHashTable
*visited
,
7589 GLOBAL_STATE_CODE();
7590 if (g_hash_table_contains(visited
, c
)) {
7593 g_hash_table_add(visited
, c
);
7596 * A BdrvChildClass that doesn't handle AioContext changes cannot
7597 * tolerate any AioContext changes
7599 if (!c
->klass
->change_aio_ctx
) {
7600 char *user
= bdrv_child_user_desc(c
);
7601 error_setg(errp
, "Changing iothreads is not supported by %s", user
);
7605 if (!c
->klass
->change_aio_ctx(c
, ctx
, visited
, tran
, errp
)) {
7606 assert(!errp
|| *errp
);
7612 bool bdrv_child_change_aio_context(BdrvChild
*c
, AioContext
*ctx
,
7613 GHashTable
*visited
, Transaction
*tran
,
7616 GLOBAL_STATE_CODE();
7617 if (g_hash_table_contains(visited
, c
)) {
7620 g_hash_table_add(visited
, c
);
7621 return bdrv_change_aio_context(c
->bs
, ctx
, visited
, tran
, errp
);
7624 static void bdrv_set_aio_context_clean(void *opaque
)
7626 BdrvStateSetAioContext
*state
= (BdrvStateSetAioContext
*) opaque
;
7627 BlockDriverState
*bs
= (BlockDriverState
*) state
->bs
;
7629 /* Paired with bdrv_drained_begin in bdrv_change_aio_context() */
7630 bdrv_drained_end(bs
);
7635 static void bdrv_set_aio_context_commit(void *opaque
)
7637 BdrvStateSetAioContext
*state
= (BdrvStateSetAioContext
*) opaque
;
7638 BlockDriverState
*bs
= (BlockDriverState
*) state
->bs
;
7639 AioContext
*new_context
= state
->new_ctx
;
7640 AioContext
*old_context
= bdrv_get_aio_context(bs
);
7643 * Take the old AioContex when detaching it from bs.
7644 * At this point, new_context lock is already acquired, and we are now
7645 * also taking old_context. This is safe as long as bdrv_detach_aio_context
7646 * does not call AIO_POLL_WHILE().
7648 if (old_context
!= qemu_get_aio_context()) {
7649 aio_context_acquire(old_context
);
7651 bdrv_detach_aio_context(bs
);
7652 if (old_context
!= qemu_get_aio_context()) {
7653 aio_context_release(old_context
);
7655 bdrv_attach_aio_context(bs
, new_context
);
7658 static TransactionActionDrv set_aio_context
= {
7659 .commit
= bdrv_set_aio_context_commit
,
7660 .clean
= bdrv_set_aio_context_clean
,
7664 * Changes the AioContext used for fd handlers, timers, and BHs by this
7665 * BlockDriverState and all its children and parents.
7667 * Must be called from the main AioContext.
7669 * The caller must own the AioContext lock for the old AioContext of bs, but it
7670 * must not own the AioContext lock for new_context (unless new_context is the
7671 * same as the current context of bs).
7673 * @visited will accumulate all visited BdrvChild objects. The caller is
7674 * responsible for freeing the list afterwards.
7676 static bool bdrv_change_aio_context(BlockDriverState
*bs
, AioContext
*ctx
,
7677 GHashTable
*visited
, Transaction
*tran
,
7681 BdrvStateSetAioContext
*state
;
7683 GLOBAL_STATE_CODE();
7685 if (bdrv_get_aio_context(bs
) == ctx
) {
7689 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
7690 if (!bdrv_parent_change_aio_context(c
, ctx
, visited
, tran
, errp
)) {
7695 QLIST_FOREACH(c
, &bs
->children
, next
) {
7696 if (!bdrv_child_change_aio_context(c
, ctx
, visited
, tran
, errp
)) {
7701 state
= g_new(BdrvStateSetAioContext
, 1);
7702 *state
= (BdrvStateSetAioContext
) {
7707 /* Paired with bdrv_drained_end in bdrv_set_aio_context_clean() */
7708 bdrv_drained_begin(bs
);
7710 tran_add(tran
, &set_aio_context
, state
);
7716 * Change bs's and recursively all of its parents' and children's AioContext
7717 * to the given new context, returning an error if that isn't possible.
7719 * If ignore_child is not NULL, that child (and its subgraph) will not
7722 * This function still requires the caller to take the bs current
7723 * AioContext lock, otherwise draining will fail since AIO_WAIT_WHILE
7724 * assumes the lock is always held if bs is in another AioContext.
7725 * For the same reason, it temporarily also holds the new AioContext, since
7726 * bdrv_drained_end calls BDRV_POLL_WHILE that assumes the lock is taken too.
7727 * Therefore the new AioContext lock must not be taken by the caller.
7729 int bdrv_try_change_aio_context(BlockDriverState
*bs
, AioContext
*ctx
,
7730 BdrvChild
*ignore_child
, Error
**errp
)
7733 GHashTable
*visited
;
7735 AioContext
*old_context
= bdrv_get_aio_context(bs
);
7736 GLOBAL_STATE_CODE();
7739 * Recursion phase: go through all nodes of the graph.
7740 * Take care of checking that all nodes support changing AioContext
7741 * and drain them, building a linear list of callbacks to run if everything
7742 * is successful (the transaction itself).
7745 visited
= g_hash_table_new(NULL
, NULL
);
7747 g_hash_table_add(visited
, ignore_child
);
7749 ret
= bdrv_change_aio_context(bs
, ctx
, visited
, tran
, errp
);
7750 g_hash_table_destroy(visited
);
7753 * Linear phase: go through all callbacks collected in the transaction.
7754 * Run all callbacks collected in the recursion to switch all nodes
7755 * AioContext lock (transaction commit), or undo all changes done in the
7756 * recursion (transaction abort).
7760 /* Just run clean() callbacks. No AioContext changed. */
7766 * Release old AioContext, it won't be needed anymore, as all
7767 * bdrv_drained_begin() have been called already.
7769 if (qemu_get_aio_context() != old_context
) {
7770 aio_context_release(old_context
);
7774 * Acquire new AioContext since bdrv_drained_end() is going to be called
7775 * after we switched all nodes in the new AioContext, and the function
7776 * assumes that the lock of the bs is always taken.
7778 if (qemu_get_aio_context() != ctx
) {
7779 aio_context_acquire(ctx
);
7784 if (qemu_get_aio_context() != ctx
) {
7785 aio_context_release(ctx
);
7788 /* Re-acquire the old AioContext, since the caller takes and releases it. */
7789 if (qemu_get_aio_context() != old_context
) {
7790 aio_context_acquire(old_context
);
7796 void bdrv_add_aio_context_notifier(BlockDriverState
*bs
,
7797 void (*attached_aio_context
)(AioContext
*new_context
, void *opaque
),
7798 void (*detach_aio_context
)(void *opaque
), void *opaque
)
7800 BdrvAioNotifier
*ban
= g_new(BdrvAioNotifier
, 1);
7801 *ban
= (BdrvAioNotifier
){
7802 .attached_aio_context
= attached_aio_context
,
7803 .detach_aio_context
= detach_aio_context
,
7806 GLOBAL_STATE_CODE();
7808 QLIST_INSERT_HEAD(&bs
->aio_notifiers
, ban
, list
);
7811 void bdrv_remove_aio_context_notifier(BlockDriverState
*bs
,
7812 void (*attached_aio_context
)(AioContext
*,
7814 void (*detach_aio_context
)(void *),
7817 BdrvAioNotifier
*ban
, *ban_next
;
7818 GLOBAL_STATE_CODE();
7820 QLIST_FOREACH_SAFE(ban
, &bs
->aio_notifiers
, list
, ban_next
) {
7821 if (ban
->attached_aio_context
== attached_aio_context
&&
7822 ban
->detach_aio_context
== detach_aio_context
&&
7823 ban
->opaque
== opaque
&&
7824 ban
->deleted
== false)
7826 if (bs
->walking_aio_notifiers
) {
7827 ban
->deleted
= true;
7829 bdrv_do_remove_aio_context_notifier(ban
);
7838 int bdrv_amend_options(BlockDriverState
*bs
, QemuOpts
*opts
,
7839 BlockDriverAmendStatusCB
*status_cb
, void *cb_opaque
,
7843 GLOBAL_STATE_CODE();
7845 error_setg(errp
, "Node is ejected");
7848 if (!bs
->drv
->bdrv_amend_options
) {
7849 error_setg(errp
, "Block driver '%s' does not support option amendment",
7850 bs
->drv
->format_name
);
7853 return bs
->drv
->bdrv_amend_options(bs
, opts
, status_cb
,
7854 cb_opaque
, force
, errp
);
7858 * This function checks whether the given @to_replace is allowed to be
7859 * replaced by a node that always shows the same data as @bs. This is
7860 * used for example to verify whether the mirror job can replace
7861 * @to_replace by the target mirrored from @bs.
7862 * To be replaceable, @bs and @to_replace may either be guaranteed to
7863 * always show the same data (because they are only connected through
7864 * filters), or some driver may allow replacing one of its children
7865 * because it can guarantee that this child's data is not visible at
7866 * all (for example, for dissenting quorum children that have no other
7869 bool bdrv_recurse_can_replace(BlockDriverState
*bs
,
7870 BlockDriverState
*to_replace
)
7872 BlockDriverState
*filtered
;
7874 GLOBAL_STATE_CODE();
7876 if (!bs
|| !bs
->drv
) {
7880 if (bs
== to_replace
) {
7884 /* See what the driver can do */
7885 if (bs
->drv
->bdrv_recurse_can_replace
) {
7886 return bs
->drv
->bdrv_recurse_can_replace(bs
, to_replace
);
7889 /* For filters without an own implementation, we can recurse on our own */
7890 filtered
= bdrv_filter_bs(bs
);
7892 return bdrv_recurse_can_replace(filtered
, to_replace
);
7900 * Check whether the given @node_name can be replaced by a node that
7901 * has the same data as @parent_bs. If so, return @node_name's BDS;
7904 * @node_name must be a (recursive) *child of @parent_bs (or this
7905 * function will return NULL).
7907 * The result (whether the node can be replaced or not) is only valid
7908 * for as long as no graph or permission changes occur.
7910 BlockDriverState
*check_to_replace_node(BlockDriverState
*parent_bs
,
7911 const char *node_name
, Error
**errp
)
7913 BlockDriverState
*to_replace_bs
= bdrv_find_node(node_name
);
7914 AioContext
*aio_context
;
7916 GLOBAL_STATE_CODE();
7918 if (!to_replace_bs
) {
7919 error_setg(errp
, "Failed to find node with node-name='%s'", node_name
);
7923 aio_context
= bdrv_get_aio_context(to_replace_bs
);
7924 aio_context_acquire(aio_context
);
7926 if (bdrv_op_is_blocked(to_replace_bs
, BLOCK_OP_TYPE_REPLACE
, errp
)) {
7927 to_replace_bs
= NULL
;
7931 /* We don't want arbitrary node of the BDS chain to be replaced only the top
7932 * most non filter in order to prevent data corruption.
7933 * Another benefit is that this tests exclude backing files which are
7934 * blocked by the backing blockers.
7936 if (!bdrv_recurse_can_replace(parent_bs
, to_replace_bs
)) {
7937 error_setg(errp
, "Cannot replace '%s' by a node mirrored from '%s', "
7938 "because it cannot be guaranteed that doing so would not "
7939 "lead to an abrupt change of visible data",
7940 node_name
, parent_bs
->node_name
);
7941 to_replace_bs
= NULL
;
7946 aio_context_release(aio_context
);
7947 return to_replace_bs
;
7951 * Iterates through the list of runtime option keys that are said to
7952 * be "strong" for a BDS. An option is called "strong" if it changes
7953 * a BDS's data. For example, the null block driver's "size" and
7954 * "read-zeroes" options are strong, but its "latency-ns" option is
7957 * If a key returned by this function ends with a dot, all options
7958 * starting with that prefix are strong.
7960 static const char *const *strong_options(BlockDriverState
*bs
,
7961 const char *const *curopt
)
7963 static const char *const global_options
[] = {
7964 "driver", "filename", NULL
7968 return &global_options
[0];
7972 if (curopt
== &global_options
[ARRAY_SIZE(global_options
) - 1] && bs
->drv
) {
7973 curopt
= bs
->drv
->strong_runtime_opts
;
7976 return (curopt
&& *curopt
) ? curopt
: NULL
;
7980 * Copies all strong runtime options from bs->options to the given
7981 * QDict. The set of strong option keys is determined by invoking
7984 * Returns true iff any strong option was present in bs->options (and
7985 * thus copied to the target QDict) with the exception of "filename"
7986 * and "driver". The caller is expected to use this value to decide
7987 * whether the existence of strong options prevents the generation of
7990 static bool append_strong_runtime_options(QDict
*d
, BlockDriverState
*bs
)
7992 bool found_any
= false;
7993 const char *const *option_name
= NULL
;
7999 while ((option_name
= strong_options(bs
, option_name
))) {
8000 bool option_given
= false;
8002 assert(strlen(*option_name
) > 0);
8003 if ((*option_name
)[strlen(*option_name
) - 1] != '.') {
8004 QObject
*entry
= qdict_get(bs
->options
, *option_name
);
8009 qdict_put_obj(d
, *option_name
, qobject_ref(entry
));
8010 option_given
= true;
8012 const QDictEntry
*entry
;
8013 for (entry
= qdict_first(bs
->options
); entry
;
8014 entry
= qdict_next(bs
->options
, entry
))
8016 if (strstart(qdict_entry_key(entry
), *option_name
, NULL
)) {
8017 qdict_put_obj(d
, qdict_entry_key(entry
),
8018 qobject_ref(qdict_entry_value(entry
)));
8019 option_given
= true;
8024 /* While "driver" and "filename" need to be included in a JSON filename,
8025 * their existence does not prohibit generation of a plain filename. */
8026 if (!found_any
&& option_given
&&
8027 strcmp(*option_name
, "driver") && strcmp(*option_name
, "filename"))
8033 if (!qdict_haskey(d
, "driver")) {
8034 /* Drivers created with bdrv_new_open_driver() may not have a
8035 * @driver option. Add it here. */
8036 qdict_put_str(d
, "driver", bs
->drv
->format_name
);
8042 /* Note: This function may return false positives; it may return true
8043 * even if opening the backing file specified by bs's image header
8044 * would result in exactly bs->backing. */
8045 static bool bdrv_backing_overridden(BlockDriverState
*bs
)
8047 GLOBAL_STATE_CODE();
8049 return strcmp(bs
->auto_backing_file
,
8050 bs
->backing
->bs
->filename
);
8052 /* No backing BDS, so if the image header reports any backing
8053 * file, it must have been suppressed */
8054 return bs
->auto_backing_file
[0] != '\0';
8058 /* Updates the following BDS fields:
8059 * - exact_filename: A filename which may be used for opening a block device
8060 * which (mostly) equals the given BDS (even without any
8061 * other options; so reading and writing must return the same
8062 * results, but caching etc. may be different)
8063 * - full_open_options: Options which, when given when opening a block device
8064 * (without a filename), result in a BDS (mostly)
8065 * equalling the given one
8066 * - filename: If exact_filename is set, it is copied here. Otherwise,
8067 * full_open_options is converted to a JSON object, prefixed with
8068 * "json:" (for use through the JSON pseudo protocol) and put here.
8070 void bdrv_refresh_filename(BlockDriverState
*bs
)
8072 BlockDriver
*drv
= bs
->drv
;
8074 BlockDriverState
*primary_child_bs
;
8076 bool backing_overridden
;
8077 bool generate_json_filename
; /* Whether our default implementation should
8078 fill exact_filename (false) or not (true) */
8080 GLOBAL_STATE_CODE();
8086 /* This BDS's file name may depend on any of its children's file names, so
8087 * refresh those first */
8088 QLIST_FOREACH(child
, &bs
->children
, next
) {
8089 bdrv_refresh_filename(child
->bs
);
8093 /* For implicit nodes, just copy everything from the single child */
8094 child
= QLIST_FIRST(&bs
->children
);
8095 assert(QLIST_NEXT(child
, next
) == NULL
);
8097 pstrcpy(bs
->exact_filename
, sizeof(bs
->exact_filename
),
8098 child
->bs
->exact_filename
);
8099 pstrcpy(bs
->filename
, sizeof(bs
->filename
), child
->bs
->filename
);
8101 qobject_unref(bs
->full_open_options
);
8102 bs
->full_open_options
= qobject_ref(child
->bs
->full_open_options
);
8107 backing_overridden
= bdrv_backing_overridden(bs
);
8109 if (bs
->open_flags
& BDRV_O_NO_IO
) {
8110 /* Without I/O, the backing file does not change anything.
8111 * Therefore, in such a case (primarily qemu-img), we can
8112 * pretend the backing file has not been overridden even if
8113 * it technically has been. */
8114 backing_overridden
= false;
8117 /* Gather the options QDict */
8119 generate_json_filename
= append_strong_runtime_options(opts
, bs
);
8120 generate_json_filename
|= backing_overridden
;
8122 if (drv
->bdrv_gather_child_options
) {
8123 /* Some block drivers may not want to present all of their children's
8124 * options, or name them differently from BdrvChild.name */
8125 drv
->bdrv_gather_child_options(bs
, opts
, backing_overridden
);
8127 QLIST_FOREACH(child
, &bs
->children
, next
) {
8128 if (child
== bs
->backing
&& !backing_overridden
) {
8129 /* We can skip the backing BDS if it has not been overridden */
8133 qdict_put(opts
, child
->name
,
8134 qobject_ref(child
->bs
->full_open_options
));
8137 if (backing_overridden
&& !bs
->backing
) {
8138 /* Force no backing file */
8139 qdict_put_null(opts
, "backing");
8143 qobject_unref(bs
->full_open_options
);
8144 bs
->full_open_options
= opts
;
8146 primary_child_bs
= bdrv_primary_bs(bs
);
8148 if (drv
->bdrv_refresh_filename
) {
8149 /* Obsolete information is of no use here, so drop the old file name
8150 * information before refreshing it */
8151 bs
->exact_filename
[0] = '\0';
8153 drv
->bdrv_refresh_filename(bs
);
8154 } else if (primary_child_bs
) {
8156 * Try to reconstruct valid information from the underlying
8157 * file -- this only works for format nodes (filter nodes
8158 * cannot be probed and as such must be selected by the user
8159 * either through an options dict, or through a special
8160 * filename which the filter driver must construct in its
8161 * .bdrv_refresh_filename() implementation).
8164 bs
->exact_filename
[0] = '\0';
8167 * We can use the underlying file's filename if:
8168 * - it has a filename,
8169 * - the current BDS is not a filter,
8170 * - the file is a protocol BDS, and
8171 * - opening that file (as this BDS's format) will automatically create
8172 * the BDS tree we have right now, that is:
8173 * - the user did not significantly change this BDS's behavior with
8174 * some explicit (strong) options
8175 * - no non-file child of this BDS has been overridden by the user
8176 * Both of these conditions are represented by generate_json_filename.
8178 if (primary_child_bs
->exact_filename
[0] &&
8179 primary_child_bs
->drv
->bdrv_file_open
&&
8180 !drv
->is_filter
&& !generate_json_filename
)
8182 strcpy(bs
->exact_filename
, primary_child_bs
->exact_filename
);
8186 if (bs
->exact_filename
[0]) {
8187 pstrcpy(bs
->filename
, sizeof(bs
->filename
), bs
->exact_filename
);
8189 GString
*json
= qobject_to_json(QOBJECT(bs
->full_open_options
));
8190 if (snprintf(bs
->filename
, sizeof(bs
->filename
), "json:%s",
8191 json
->str
) >= sizeof(bs
->filename
)) {
8192 /* Give user a hint if we truncated things. */
8193 strcpy(bs
->filename
+ sizeof(bs
->filename
) - 4, "...");
8195 g_string_free(json
, true);
8199 char *bdrv_dirname(BlockDriverState
*bs
, Error
**errp
)
8201 BlockDriver
*drv
= bs
->drv
;
8202 BlockDriverState
*child_bs
;
8204 GLOBAL_STATE_CODE();
8207 error_setg(errp
, "Node '%s' is ejected", bs
->node_name
);
8211 if (drv
->bdrv_dirname
) {
8212 return drv
->bdrv_dirname(bs
, errp
);
8215 child_bs
= bdrv_primary_bs(bs
);
8217 return bdrv_dirname(child_bs
, errp
);
8220 bdrv_refresh_filename(bs
);
8221 if (bs
->exact_filename
[0] != '\0') {
8222 return path_combine(bs
->exact_filename
, "");
8225 error_setg(errp
, "Cannot generate a base directory for %s nodes",
8231 * Hot add/remove a BDS's child. So the user can take a child offline when
8232 * it is broken and take a new child online
8234 void bdrv_add_child(BlockDriverState
*parent_bs
, BlockDriverState
*child_bs
,
8237 GLOBAL_STATE_CODE();
8238 if (!parent_bs
->drv
|| !parent_bs
->drv
->bdrv_add_child
) {
8239 error_setg(errp
, "The node %s does not support adding a child",
8240 bdrv_get_device_or_node_name(parent_bs
));
8245 * Non-zoned block drivers do not follow zoned storage constraints
8246 * (i.e. sequential writes to zones). Refuse mixing zoned and non-zoned
8247 * drivers in a graph.
8249 if (!parent_bs
->drv
->supports_zoned_children
&&
8250 child_bs
->bl
.zoned
== BLK_Z_HM
) {
8252 * The host-aware model allows zoned storage constraints and random
8253 * write. Allow mixing host-aware and non-zoned drivers. Using
8254 * host-aware device as a regular device.
8256 error_setg(errp
, "Cannot add a %s child to a %s parent",
8257 child_bs
->bl
.zoned
== BLK_Z_HM
? "zoned" : "non-zoned",
8258 parent_bs
->drv
->supports_zoned_children
?
8259 "support zoned children" : "not support zoned children");
8263 if (!QLIST_EMPTY(&child_bs
->parents
)) {
8264 error_setg(errp
, "The node %s already has a parent",
8265 child_bs
->node_name
);
8269 parent_bs
->drv
->bdrv_add_child(parent_bs
, child_bs
, errp
);
8272 void bdrv_del_child(BlockDriverState
*parent_bs
, BdrvChild
*child
, Error
**errp
)
8276 GLOBAL_STATE_CODE();
8277 if (!parent_bs
->drv
|| !parent_bs
->drv
->bdrv_del_child
) {
8278 error_setg(errp
, "The node %s does not support removing a child",
8279 bdrv_get_device_or_node_name(parent_bs
));
8283 QLIST_FOREACH(tmp
, &parent_bs
->children
, next
) {
8290 error_setg(errp
, "The node %s does not have a child named %s",
8291 bdrv_get_device_or_node_name(parent_bs
),
8292 bdrv_get_device_or_node_name(child
->bs
));
8296 parent_bs
->drv
->bdrv_del_child(parent_bs
, child
, errp
);
8299 int bdrv_make_empty(BdrvChild
*c
, Error
**errp
)
8301 BlockDriver
*drv
= c
->bs
->drv
;
8304 GLOBAL_STATE_CODE();
8305 assert(c
->perm
& (BLK_PERM_WRITE
| BLK_PERM_WRITE_UNCHANGED
));
8307 if (!drv
->bdrv_make_empty
) {
8308 error_setg(errp
, "%s does not support emptying nodes",
8313 ret
= drv
->bdrv_make_empty(c
->bs
);
8315 error_setg_errno(errp
, -ret
, "Failed to empty %s",
8324 * Return the child that @bs acts as an overlay for, and from which data may be
8325 * copied in COW or COR operations. Usually this is the backing file.
8327 BdrvChild
*bdrv_cow_child(BlockDriverState
*bs
)
8331 if (!bs
|| !bs
->drv
) {
8335 if (bs
->drv
->is_filter
) {
8343 assert(bs
->backing
->role
& BDRV_CHILD_COW
);
8348 * If @bs acts as a filter for exactly one of its children, return
8351 BdrvChild
*bdrv_filter_child(BlockDriverState
*bs
)
8356 if (!bs
|| !bs
->drv
) {
8360 if (!bs
->drv
->is_filter
) {
8364 /* Only one of @backing or @file may be used */
8365 assert(!(bs
->backing
&& bs
->file
));
8367 c
= bs
->backing
?: bs
->file
;
8372 assert(c
->role
& BDRV_CHILD_FILTERED
);
8377 * Return either the result of bdrv_cow_child() or bdrv_filter_child(),
8378 * whichever is non-NULL.
8380 * Return NULL if both are NULL.
8382 BdrvChild
*bdrv_filter_or_cow_child(BlockDriverState
*bs
)
8384 BdrvChild
*cow_child
= bdrv_cow_child(bs
);
8385 BdrvChild
*filter_child
= bdrv_filter_child(bs
);
8388 /* Filter nodes cannot have COW backing files */
8389 assert(!(cow_child
&& filter_child
));
8391 return cow_child
?: filter_child
;
8395 * Return the primary child of this node: For filters, that is the
8396 * filtered child. For other nodes, that is usually the child storing
8398 * (A generally more helpful description is that this is (usually) the
8399 * child that has the same filename as @bs.)
8401 * Drivers do not necessarily have a primary child; for example quorum
8404 BdrvChild
*bdrv_primary_child(BlockDriverState
*bs
)
8406 BdrvChild
*c
, *found
= NULL
;
8409 QLIST_FOREACH(c
, &bs
->children
, next
) {
8410 if (c
->role
& BDRV_CHILD_PRIMARY
) {
8419 static BlockDriverState
*bdrv_do_skip_filters(BlockDriverState
*bs
,
8420 bool stop_on_explicit_filter
)
8428 while (!(stop_on_explicit_filter
&& !bs
->implicit
)) {
8429 c
= bdrv_filter_child(bs
);
8432 * A filter that is embedded in a working block graph must
8433 * have a child. Assert this here so this function does
8434 * not return a filter node that is not expected by the
8437 assert(!bs
->drv
|| !bs
->drv
->is_filter
);
8443 * Note that this treats nodes with bs->drv == NULL as not being
8444 * filters (bs->drv == NULL should be replaced by something else
8446 * The advantage of this behavior is that this function will thus
8447 * always return a non-NULL value (given a non-NULL @bs).
8454 * Return the first BDS that has not been added implicitly or that
8455 * does not have a filtered child down the chain starting from @bs
8456 * (including @bs itself).
8458 BlockDriverState
*bdrv_skip_implicit_filters(BlockDriverState
*bs
)
8460 GLOBAL_STATE_CODE();
8461 return bdrv_do_skip_filters(bs
, true);
8465 * Return the first BDS that does not have a filtered child down the
8466 * chain starting from @bs (including @bs itself).
8468 BlockDriverState
*bdrv_skip_filters(BlockDriverState
*bs
)
8471 return bdrv_do_skip_filters(bs
, false);
8475 * For a backing chain, return the first non-filter backing image of
8476 * the first non-filter image.
8478 BlockDriverState
*bdrv_backing_chain_next(BlockDriverState
*bs
)
8481 return bdrv_skip_filters(bdrv_cow_bs(bdrv_skip_filters(bs
)));
8485 * Check whether [offset, offset + bytes) overlaps with the cached
8486 * block-status data region.
8488 * If so, and @pnum is not NULL, set *pnum to `bsc.data_end - offset`,
8489 * which is what bdrv_bsc_is_data()'s interface needs.
8490 * Otherwise, *pnum is not touched.
8492 static bool bdrv_bsc_range_overlaps_locked(BlockDriverState
*bs
,
8493 int64_t offset
, int64_t bytes
,
8496 BdrvBlockStatusCache
*bsc
= qatomic_rcu_read(&bs
->block_status_cache
);
8500 qatomic_read(&bsc
->valid
) &&
8501 ranges_overlap(offset
, bytes
, bsc
->data_start
,
8502 bsc
->data_end
- bsc
->data_start
);
8504 if (overlaps
&& pnum
) {
8505 *pnum
= bsc
->data_end
- offset
;
8512 * See block_int.h for this function's documentation.
8514 bool bdrv_bsc_is_data(BlockDriverState
*bs
, int64_t offset
, int64_t *pnum
)
8517 RCU_READ_LOCK_GUARD();
8518 return bdrv_bsc_range_overlaps_locked(bs
, offset
, 1, pnum
);
8522 * See block_int.h for this function's documentation.
8524 void bdrv_bsc_invalidate_range(BlockDriverState
*bs
,
8525 int64_t offset
, int64_t bytes
)
8528 RCU_READ_LOCK_GUARD();
8530 if (bdrv_bsc_range_overlaps_locked(bs
, offset
, bytes
, NULL
)) {
8531 qatomic_set(&bs
->block_status_cache
->valid
, false);
8536 * See block_int.h for this function's documentation.
8538 void bdrv_bsc_fill(BlockDriverState
*bs
, int64_t offset
, int64_t bytes
)
8540 BdrvBlockStatusCache
*new_bsc
= g_new(BdrvBlockStatusCache
, 1);
8541 BdrvBlockStatusCache
*old_bsc
;
8544 *new_bsc
= (BdrvBlockStatusCache
) {
8546 .data_start
= offset
,
8547 .data_end
= offset
+ bytes
,
8550 QEMU_LOCK_GUARD(&bs
->bsc_modify_lock
);
8552 old_bsc
= qatomic_rcu_read(&bs
->block_status_cache
);
8553 qatomic_rcu_set(&bs
->block_status_cache
, new_bsc
);
8555 g_free_rcu(old_bsc
, rcu
);