2 * QEMU System Emulator block driver
4 * Copyright (c) 2003 Fabrice Bellard
5 * Copyright (c) 2020 Virtuozzo International GmbH.
7 * Permission is hereby granted, free of charge, to any person obtaining a copy
8 * of this software and associated documentation files (the "Software"), to deal
9 * in the Software without restriction, including without limitation the rights
10 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
11 * copies of the Software, and to permit persons to whom the Software is
12 * furnished to do so, subject to the following conditions:
14 * The above copyright notice and this permission notice shall be included in
15 * all copies or substantial portions of the Software.
17 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
18 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
19 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
20 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
21 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
22 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
26 #include "qemu/osdep.h"
27 #include "block/trace.h"
28 #include "block/block_int.h"
29 #include "block/blockjob.h"
30 #include "block/fuse.h"
31 #include "block/nbd.h"
32 #include "block/qdict.h"
33 #include "qemu/error-report.h"
34 #include "block/module_block.h"
35 #include "qemu/main-loop.h"
36 #include "qemu/module.h"
37 #include "qapi/error.h"
38 #include "qapi/qmp/qdict.h"
39 #include "qapi/qmp/qjson.h"
40 #include "qapi/qmp/qnull.h"
41 #include "qapi/qmp/qstring.h"
42 #include "qapi/qobject-output-visitor.h"
43 #include "qapi/qapi-visit-block-core.h"
44 #include "sysemu/block-backend.h"
45 #include "qemu/notify.h"
46 #include "qemu/option.h"
47 #include "qemu/coroutine.h"
48 #include "block/qapi.h"
49 #include "qemu/timer.h"
50 #include "qemu/cutils.h"
52 #include "qemu/range.h"
54 #include "block/coroutines.h"
57 #include <sys/ioctl.h>
58 #include <sys/queue.h>
59 #if defined(HAVE_SYS_DISK_H)
68 #define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */
70 /* Protected by BQL */
71 static QTAILQ_HEAD(, BlockDriverState
) graph_bdrv_states
=
72 QTAILQ_HEAD_INITIALIZER(graph_bdrv_states
);
74 /* Protected by BQL */
75 static QTAILQ_HEAD(, BlockDriverState
) all_bdrv_states
=
76 QTAILQ_HEAD_INITIALIZER(all_bdrv_states
);
78 /* Protected by BQL */
79 static QLIST_HEAD(, BlockDriver
) bdrv_drivers
=
80 QLIST_HEAD_INITIALIZER(bdrv_drivers
);
82 static BlockDriverState
*bdrv_open_inherit(const char *filename
,
83 const char *reference
,
84 QDict
*options
, int flags
,
85 BlockDriverState
*parent
,
86 const BdrvChildClass
*child_class
,
87 BdrvChildRole child_role
,
90 static bool bdrv_recurse_has_child(BlockDriverState
*bs
,
91 BlockDriverState
*child
);
93 static void bdrv_replace_child_noperm(BdrvChild
*child
,
94 BlockDriverState
*new_bs
);
95 static void bdrv_remove_child(BdrvChild
*child
, Transaction
*tran
);
97 static int bdrv_reopen_prepare(BDRVReopenState
*reopen_state
,
98 BlockReopenQueue
*queue
,
99 Transaction
*change_child_tran
, Error
**errp
);
100 static void bdrv_reopen_commit(BDRVReopenState
*reopen_state
);
101 static void bdrv_reopen_abort(BDRVReopenState
*reopen_state
);
103 static bool bdrv_backing_overridden(BlockDriverState
*bs
);
105 static bool bdrv_change_aio_context(BlockDriverState
*bs
, AioContext
*ctx
,
106 GHashTable
*visited
, Transaction
*tran
,
109 /* If non-zero, use only whitelisted block drivers */
110 static int use_bdrv_whitelist
;
113 static int is_windows_drive_prefix(const char *filename
)
115 return (((filename
[0] >= 'a' && filename
[0] <= 'z') ||
116 (filename
[0] >= 'A' && filename
[0] <= 'Z')) &&
120 int is_windows_drive(const char *filename
)
122 if (is_windows_drive_prefix(filename
) &&
125 if (strstart(filename
, "\\\\.\\", NULL
) ||
126 strstart(filename
, "//./", NULL
))
132 size_t bdrv_opt_mem_align(BlockDriverState
*bs
)
134 if (!bs
|| !bs
->drv
) {
135 /* page size or 4k (hdd sector size) should be on the safe side */
136 return MAX(4096, qemu_real_host_page_size());
140 return bs
->bl
.opt_mem_alignment
;
143 size_t bdrv_min_mem_align(BlockDriverState
*bs
)
145 if (!bs
|| !bs
->drv
) {
146 /* page size or 4k (hdd sector size) should be on the safe side */
147 return MAX(4096, qemu_real_host_page_size());
151 return bs
->bl
.min_mem_alignment
;
154 /* check if the path starts with "<protocol>:" */
155 int path_has_protocol(const char *path
)
160 if (is_windows_drive(path
) ||
161 is_windows_drive_prefix(path
)) {
164 p
= path
+ strcspn(path
, ":/\\");
166 p
= path
+ strcspn(path
, ":/");
172 int path_is_absolute(const char *path
)
175 /* specific case for names like: "\\.\d:" */
176 if (is_windows_drive(path
) || is_windows_drive_prefix(path
)) {
179 return (*path
== '/' || *path
== '\\');
181 return (*path
== '/');
185 /* if filename is absolute, just return its duplicate. Otherwise, build a
186 path to it by considering it is relative to base_path. URL are
188 char *path_combine(const char *base_path
, const char *filename
)
190 const char *protocol_stripped
= NULL
;
195 if (path_is_absolute(filename
)) {
196 return g_strdup(filename
);
199 if (path_has_protocol(base_path
)) {
200 protocol_stripped
= strchr(base_path
, ':');
201 if (protocol_stripped
) {
205 p
= protocol_stripped
?: base_path
;
207 p1
= strrchr(base_path
, '/');
211 p2
= strrchr(base_path
, '\\');
212 if (!p1
|| p2
> p1
) {
227 result
= g_malloc(len
+ strlen(filename
) + 1);
228 memcpy(result
, base_path
, len
);
229 strcpy(result
+ len
, filename
);
235 * Helper function for bdrv_parse_filename() implementations to remove optional
236 * protocol prefixes (especially "file:") from a filename and for putting the
237 * stripped filename into the options QDict if there is such a prefix.
239 void bdrv_parse_filename_strip_prefix(const char *filename
, const char *prefix
,
242 if (strstart(filename
, prefix
, &filename
)) {
243 /* Stripping the explicit protocol prefix may result in a protocol
244 * prefix being (wrongly) detected (if the filename contains a colon) */
245 if (path_has_protocol(filename
)) {
246 GString
*fat_filename
;
248 /* This means there is some colon before the first slash; therefore,
249 * this cannot be an absolute path */
250 assert(!path_is_absolute(filename
));
252 /* And we can thus fix the protocol detection issue by prefixing it
254 fat_filename
= g_string_new("./");
255 g_string_append(fat_filename
, filename
);
257 assert(!path_has_protocol(fat_filename
->str
));
259 qdict_put(options
, "filename",
260 qstring_from_gstring(fat_filename
));
262 /* If no protocol prefix was detected, we can use the shortened
264 qdict_put_str(options
, "filename", filename
);
270 /* Returns whether the image file is opened as read-only. Note that this can
271 * return false and writing to the image file is still not possible because the
272 * image is inactivated. */
273 bool bdrv_is_read_only(BlockDriverState
*bs
)
276 return !(bs
->open_flags
& BDRV_O_RDWR
);
279 int bdrv_can_set_read_only(BlockDriverState
*bs
, bool read_only
,
280 bool ignore_allow_rdw
, Error
**errp
)
284 /* Do not set read_only if copy_on_read is enabled */
285 if (bs
->copy_on_read
&& read_only
) {
286 error_setg(errp
, "Can't set node '%s' to r/o with copy-on-read enabled",
287 bdrv_get_device_or_node_name(bs
));
291 /* Do not clear read_only if it is prohibited */
292 if (!read_only
&& !(bs
->open_flags
& BDRV_O_ALLOW_RDWR
) &&
295 error_setg(errp
, "Node '%s' is read only",
296 bdrv_get_device_or_node_name(bs
));
304 * Called by a driver that can only provide a read-only image.
306 * Returns 0 if the node is already read-only or it could switch the node to
307 * read-only because BDRV_O_AUTO_RDONLY is set.
309 * Returns -EACCES if the node is read-write and BDRV_O_AUTO_RDONLY is not set
310 * or bdrv_can_set_read_only() forbids making the node read-only. If @errmsg
311 * is not NULL, it is used as the error message for the Error object.
313 int bdrv_apply_auto_read_only(BlockDriverState
*bs
, const char *errmsg
,
319 if (!(bs
->open_flags
& BDRV_O_RDWR
)) {
322 if (!(bs
->open_flags
& BDRV_O_AUTO_RDONLY
)) {
326 ret
= bdrv_can_set_read_only(bs
, true, false, NULL
);
331 bs
->open_flags
&= ~BDRV_O_RDWR
;
336 error_setg(errp
, "%s", errmsg
?: "Image is read-only");
341 * If @backing is empty, this function returns NULL without setting
342 * @errp. In all other cases, NULL will only be returned with @errp
345 * Therefore, a return value of NULL without @errp set means that
346 * there is no backing file; if @errp is set, there is one but its
347 * absolute filename cannot be generated.
349 char *bdrv_get_full_backing_filename_from_filename(const char *backed
,
353 if (backing
[0] == '\0') {
355 } else if (path_has_protocol(backing
) || path_is_absolute(backing
)) {
356 return g_strdup(backing
);
357 } else if (backed
[0] == '\0' || strstart(backed
, "json:", NULL
)) {
358 error_setg(errp
, "Cannot use relative backing file names for '%s'",
362 return path_combine(backed
, backing
);
367 * If @filename is empty or NULL, this function returns NULL without
368 * setting @errp. In all other cases, NULL will only be returned with
371 static char *bdrv_make_absolute_filename(BlockDriverState
*relative_to
,
372 const char *filename
, Error
**errp
)
374 char *dir
, *full_name
;
376 if (!filename
|| filename
[0] == '\0') {
378 } else if (path_has_protocol(filename
) || path_is_absolute(filename
)) {
379 return g_strdup(filename
);
382 dir
= bdrv_dirname(relative_to
, errp
);
387 full_name
= g_strconcat(dir
, filename
, NULL
);
392 char *bdrv_get_full_backing_filename(BlockDriverState
*bs
, Error
**errp
)
395 return bdrv_make_absolute_filename(bs
, bs
->backing_file
, errp
);
398 void bdrv_register(BlockDriver
*bdrv
)
400 assert(bdrv
->format_name
);
402 QLIST_INSERT_HEAD(&bdrv_drivers
, bdrv
, list
);
405 BlockDriverState
*bdrv_new(void)
407 BlockDriverState
*bs
;
412 bs
= g_new0(BlockDriverState
, 1);
413 QLIST_INIT(&bs
->dirty_bitmaps
);
414 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
415 QLIST_INIT(&bs
->op_blockers
[i
]);
417 qemu_co_mutex_init(&bs
->reqs_lock
);
418 qemu_mutex_init(&bs
->dirty_bitmap_mutex
);
420 bs
->aio_context
= qemu_get_aio_context();
422 qemu_co_queue_init(&bs
->flush_queue
);
424 qemu_co_mutex_init(&bs
->bsc_modify_lock
);
425 bs
->block_status_cache
= g_new0(BdrvBlockStatusCache
, 1);
427 for (i
= 0; i
< bdrv_drain_all_count
; i
++) {
428 bdrv_drained_begin(bs
);
431 QTAILQ_INSERT_TAIL(&all_bdrv_states
, bs
, bs_list
);
436 static BlockDriver
*bdrv_do_find_format(const char *format_name
)
441 QLIST_FOREACH(drv1
, &bdrv_drivers
, list
) {
442 if (!strcmp(drv1
->format_name
, format_name
)) {
450 BlockDriver
*bdrv_find_format(const char *format_name
)
457 drv1
= bdrv_do_find_format(format_name
);
462 /* The driver isn't registered, maybe we need to load a module */
463 for (i
= 0; i
< (int)ARRAY_SIZE(block_driver_modules
); ++i
) {
464 if (!strcmp(block_driver_modules
[i
].format_name
, format_name
)) {
465 Error
*local_err
= NULL
;
466 int rv
= block_module_load(block_driver_modules
[i
].library_name
,
469 return bdrv_do_find_format(format_name
);
471 error_report_err(local_err
);
479 static int bdrv_format_is_whitelisted(const char *format_name
, bool read_only
)
481 static const char *whitelist_rw
[] = {
482 CONFIG_BDRV_RW_WHITELIST
485 static const char *whitelist_ro
[] = {
486 CONFIG_BDRV_RO_WHITELIST
491 if (!whitelist_rw
[0] && !whitelist_ro
[0]) {
492 return 1; /* no whitelist, anything goes */
495 for (p
= whitelist_rw
; *p
; p
++) {
496 if (!strcmp(format_name
, *p
)) {
501 for (p
= whitelist_ro
; *p
; p
++) {
502 if (!strcmp(format_name
, *p
)) {
510 int bdrv_is_whitelisted(BlockDriver
*drv
, bool read_only
)
513 return bdrv_format_is_whitelisted(drv
->format_name
, read_only
);
516 bool bdrv_uses_whitelist(void)
518 return use_bdrv_whitelist
;
521 typedef struct CreateCo
{
529 static void coroutine_fn
bdrv_create_co_entry(void *opaque
)
531 Error
*local_err
= NULL
;
534 CreateCo
*cco
= opaque
;
538 ret
= cco
->drv
->bdrv_co_create_opts(cco
->drv
,
539 cco
->filename
, cco
->opts
, &local_err
);
540 error_propagate(&cco
->err
, local_err
);
544 int bdrv_create(BlockDriver
*drv
, const char* filename
,
545 QemuOpts
*opts
, Error
**errp
)
554 .filename
= g_strdup(filename
),
560 if (!drv
->bdrv_co_create_opts
) {
561 error_setg(errp
, "Driver '%s' does not support image creation", drv
->format_name
);
566 if (qemu_in_coroutine()) {
567 /* Fast-path if already in coroutine context */
568 bdrv_create_co_entry(&cco
);
570 co
= qemu_coroutine_create(bdrv_create_co_entry
, &cco
);
571 qemu_coroutine_enter(co
);
572 while (cco
.ret
== NOT_DONE
) {
573 aio_poll(qemu_get_aio_context(), true);
580 error_propagate(errp
, cco
.err
);
582 error_setg_errno(errp
, -ret
, "Could not create image");
587 g_free(cco
.filename
);
592 * Helper function for bdrv_create_file_fallback(): Resize @blk to at
593 * least the given @minimum_size.
595 * On success, return @blk's actual length.
596 * Otherwise, return -errno.
598 static int64_t create_file_fallback_truncate(BlockBackend
*blk
,
599 int64_t minimum_size
, Error
**errp
)
601 Error
*local_err
= NULL
;
607 ret
= blk_truncate(blk
, minimum_size
, false, PREALLOC_MODE_OFF
, 0,
609 if (ret
< 0 && ret
!= -ENOTSUP
) {
610 error_propagate(errp
, local_err
);
614 size
= blk_getlength(blk
);
616 error_free(local_err
);
617 error_setg_errno(errp
, -size
,
618 "Failed to inquire the new image file's length");
622 if (size
< minimum_size
) {
623 /* Need to grow the image, but we failed to do that */
624 error_propagate(errp
, local_err
);
628 error_free(local_err
);
635 * Helper function for bdrv_create_file_fallback(): Zero the first
636 * sector to remove any potentially pre-existing image header.
638 static int coroutine_fn
639 create_file_fallback_zero_first_sector(BlockBackend
*blk
,
640 int64_t current_size
,
643 int64_t bytes_to_clear
;
648 bytes_to_clear
= MIN(current_size
, BDRV_SECTOR_SIZE
);
649 if (bytes_to_clear
) {
650 ret
= blk_co_pwrite_zeroes(blk
, 0, bytes_to_clear
, BDRV_REQ_MAY_UNMAP
);
652 error_setg_errno(errp
, -ret
,
653 "Failed to clear the new image's first sector");
662 * Simple implementation of bdrv_co_create_opts for protocol drivers
663 * which only support creation via opening a file
664 * (usually existing raw storage device)
666 int coroutine_fn
bdrv_co_create_opts_simple(BlockDriver
*drv
,
667 const char *filename
,
675 PreallocMode prealloc
;
676 Error
*local_err
= NULL
;
681 size
= qemu_opt_get_size_del(opts
, BLOCK_OPT_SIZE
, 0);
682 buf
= qemu_opt_get_del(opts
, BLOCK_OPT_PREALLOC
);
683 prealloc
= qapi_enum_parse(&PreallocMode_lookup
, buf
,
684 PREALLOC_MODE_OFF
, &local_err
);
687 error_propagate(errp
, local_err
);
691 if (prealloc
!= PREALLOC_MODE_OFF
) {
692 error_setg(errp
, "Unsupported preallocation mode '%s'",
693 PreallocMode_str(prealloc
));
697 options
= qdict_new();
698 qdict_put_str(options
, "driver", drv
->format_name
);
700 blk
= blk_new_open(filename
, NULL
, options
,
701 BDRV_O_RDWR
| BDRV_O_RESIZE
, errp
);
703 error_prepend(errp
, "Protocol driver '%s' does not support image "
704 "creation, and opening the image failed: ",
709 size
= create_file_fallback_truncate(blk
, size
, errp
);
715 ret
= create_file_fallback_zero_first_sector(blk
, size
, errp
);
726 int bdrv_create_file(const char *filename
, QemuOpts
*opts
, Error
**errp
)
728 QemuOpts
*protocol_opts
;
735 drv
= bdrv_find_protocol(filename
, true, errp
);
740 if (!drv
->create_opts
) {
741 error_setg(errp
, "Driver '%s' does not support image creation",
747 * 'opts' contains a QemuOptsList with a combination of format and protocol
750 * The format properly removes its options, but the default values remain
751 * in 'opts->list'. So if the protocol has options with the same name
752 * (e.g. rbd has 'cluster_size' as qcow2), it will see the default values
753 * of the format, since for overlapping options, the format wins.
755 * To avoid this issue, lets convert QemuOpts to QDict, in this way we take
756 * only the set options, and then convert it back to QemuOpts, using the
757 * create_opts of the protocol. So the new QemuOpts, will contain only the
760 qdict
= qemu_opts_to_qdict(opts
, NULL
);
761 protocol_opts
= qemu_opts_from_qdict(drv
->create_opts
, qdict
, errp
);
762 if (protocol_opts
== NULL
) {
767 ret
= bdrv_create(drv
, filename
, protocol_opts
, errp
);
769 qemu_opts_del(protocol_opts
);
770 qobject_unref(qdict
);
774 int coroutine_fn
bdrv_co_delete_file(BlockDriverState
*bs
, Error
**errp
)
776 Error
*local_err
= NULL
;
783 error_setg(errp
, "Block node '%s' is not opened", bs
->filename
);
787 if (!bs
->drv
->bdrv_co_delete_file
) {
788 error_setg(errp
, "Driver '%s' does not support image deletion",
789 bs
->drv
->format_name
);
793 ret
= bs
->drv
->bdrv_co_delete_file(bs
, &local_err
);
795 error_propagate(errp
, local_err
);
801 void coroutine_fn
bdrv_co_delete_file_noerr(BlockDriverState
*bs
)
803 Error
*local_err
= NULL
;
811 ret
= bdrv_co_delete_file(bs
, &local_err
);
813 * ENOTSUP will happen if the block driver doesn't support
814 * the 'bdrv_co_delete_file' interface. This is a predictable
815 * scenario and shouldn't be reported back to the user.
817 if (ret
== -ENOTSUP
) {
818 error_free(local_err
);
819 } else if (ret
< 0) {
820 error_report_err(local_err
);
825 * Try to get @bs's logical and physical block size.
826 * On success, store them in @bsz struct and return 0.
827 * On failure return -errno.
828 * @bs must not be empty.
830 int bdrv_probe_blocksizes(BlockDriverState
*bs
, BlockSizes
*bsz
)
832 BlockDriver
*drv
= bs
->drv
;
833 BlockDriverState
*filtered
= bdrv_filter_bs(bs
);
836 if (drv
&& drv
->bdrv_probe_blocksizes
) {
837 return drv
->bdrv_probe_blocksizes(bs
, bsz
);
838 } else if (filtered
) {
839 return bdrv_probe_blocksizes(filtered
, bsz
);
846 * Try to get @bs's geometry (cyls, heads, sectors).
847 * On success, store them in @geo struct and return 0.
848 * On failure return -errno.
849 * @bs must not be empty.
851 int bdrv_probe_geometry(BlockDriverState
*bs
, HDGeometry
*geo
)
853 BlockDriver
*drv
= bs
->drv
;
854 BlockDriverState
*filtered
= bdrv_filter_bs(bs
);
857 if (drv
&& drv
->bdrv_probe_geometry
) {
858 return drv
->bdrv_probe_geometry(bs
, geo
);
859 } else if (filtered
) {
860 return bdrv_probe_geometry(filtered
, geo
);
867 * Create a uniquely-named empty temporary file.
868 * Return the actual file name used upon success, otherwise NULL.
869 * This string should be freed with g_free() when not needed any longer.
871 * Note: creating a temporary file for the caller to (re)open is
872 * inherently racy. Use g_file_open_tmp() instead whenever practical.
874 char *create_tmp_file(Error
**errp
)
878 g_autofree
char *filename
= NULL
;
880 tmpdir
= g_get_tmp_dir();
883 * See commit 69bef79 ("block: use /var/tmp instead of /tmp for -snapshot")
885 * This function is used to create temporary disk images (like -snapshot),
886 * so the files can become very large. /tmp is often a tmpfs where as
887 * /var/tmp is usually on a disk, so more appropriate for disk images.
889 if (!g_strcmp0(tmpdir
, "/tmp")) {
894 filename
= g_strdup_printf("%s/vl.XXXXXX", tmpdir
);
895 fd
= g_mkstemp(filename
);
897 error_setg_errno(errp
, errno
, "Could not open temporary file '%s'",
903 return g_steal_pointer(&filename
);
907 * Detect host devices. By convention, /dev/cdrom[N] is always
908 * recognized as a host CDROM.
910 static BlockDriver
*find_hdev_driver(const char *filename
)
912 int score_max
= 0, score
;
913 BlockDriver
*drv
= NULL
, *d
;
916 QLIST_FOREACH(d
, &bdrv_drivers
, list
) {
917 if (d
->bdrv_probe_device
) {
918 score
= d
->bdrv_probe_device(filename
);
919 if (score
> score_max
) {
929 static BlockDriver
*bdrv_do_find_protocol(const char *protocol
)
934 QLIST_FOREACH(drv1
, &bdrv_drivers
, list
) {
935 if (drv1
->protocol_name
&& !strcmp(drv1
->protocol_name
, protocol
)) {
943 BlockDriver
*bdrv_find_protocol(const char *filename
,
944 bool allow_protocol_prefix
,
954 /* TODO Drivers without bdrv_file_open must be specified explicitly */
957 * XXX(hch): we really should not let host device detection
958 * override an explicit protocol specification, but moving this
959 * later breaks access to device names with colons in them.
960 * Thanks to the brain-dead persistent naming schemes on udev-
961 * based Linux systems those actually are quite common.
963 drv1
= find_hdev_driver(filename
);
968 if (!path_has_protocol(filename
) || !allow_protocol_prefix
) {
972 p
= strchr(filename
, ':');
975 if (len
> sizeof(protocol
) - 1)
976 len
= sizeof(protocol
) - 1;
977 memcpy(protocol
, filename
, len
);
978 protocol
[len
] = '\0';
980 drv1
= bdrv_do_find_protocol(protocol
);
985 for (i
= 0; i
< (int)ARRAY_SIZE(block_driver_modules
); ++i
) {
986 if (block_driver_modules
[i
].protocol_name
&&
987 !strcmp(block_driver_modules
[i
].protocol_name
, protocol
)) {
988 int rv
= block_module_load(block_driver_modules
[i
].library_name
, errp
);
990 drv1
= bdrv_do_find_protocol(protocol
);
999 error_setg(errp
, "Unknown protocol '%s'", protocol
);
1005 * Guess image format by probing its contents.
1006 * This is not a good idea when your image is raw (CVE-2008-2004), but
1007 * we do it anyway for backward compatibility.
1009 * @buf contains the image's first @buf_size bytes.
1010 * @buf_size is the buffer size in bytes (generally BLOCK_PROBE_BUF_SIZE,
1011 * but can be smaller if the image file is smaller)
1012 * @filename is its filename.
1014 * For all block drivers, call the bdrv_probe() method to get its
1016 * Return the first block driver with the highest probing score.
1018 BlockDriver
*bdrv_probe_all(const uint8_t *buf
, int buf_size
,
1019 const char *filename
)
1021 int score_max
= 0, score
;
1022 BlockDriver
*drv
= NULL
, *d
;
1025 QLIST_FOREACH(d
, &bdrv_drivers
, list
) {
1026 if (d
->bdrv_probe
) {
1027 score
= d
->bdrv_probe(buf
, buf_size
, filename
);
1028 if (score
> score_max
) {
1038 static int find_image_format(BlockBackend
*file
, const char *filename
,
1039 BlockDriver
**pdrv
, Error
**errp
)
1042 uint8_t buf
[BLOCK_PROBE_BUF_SIZE
];
1045 GLOBAL_STATE_CODE();
1047 /* Return the raw BlockDriver * to scsi-generic devices or empty drives */
1048 if (blk_is_sg(file
) || !blk_is_inserted(file
) || blk_getlength(file
) == 0) {
1053 ret
= blk_pread(file
, 0, sizeof(buf
), buf
, 0);
1055 error_setg_errno(errp
, -ret
, "Could not read image for determining its "
1061 drv
= bdrv_probe_all(buf
, sizeof(buf
), filename
);
1063 error_setg(errp
, "Could not determine image format: No compatible "
1074 * Set the current 'total_sectors' value
1075 * Return 0 on success, -errno on error.
1077 int refresh_total_sectors(BlockDriverState
*bs
, int64_t hint
)
1079 BlockDriver
*drv
= bs
->drv
;
1086 /* Do not attempt drv->bdrv_getlength() on scsi-generic devices */
1090 /* query actual device if possible, otherwise just trust the hint */
1091 if (drv
->bdrv_getlength
) {
1092 int64_t length
= drv
->bdrv_getlength(bs
);
1096 hint
= DIV_ROUND_UP(length
, BDRV_SECTOR_SIZE
);
1099 bs
->total_sectors
= hint
;
1101 if (bs
->total_sectors
* BDRV_SECTOR_SIZE
> BDRV_MAX_LENGTH
) {
1109 * Combines a QDict of new block driver @options with any missing options taken
1110 * from @old_options, so that leaving out an option defaults to its old value.
1112 static void bdrv_join_options(BlockDriverState
*bs
, QDict
*options
,
1115 GLOBAL_STATE_CODE();
1116 if (bs
->drv
&& bs
->drv
->bdrv_join_options
) {
1117 bs
->drv
->bdrv_join_options(options
, old_options
);
1119 qdict_join(options
, old_options
, false);
1123 static BlockdevDetectZeroesOptions
bdrv_parse_detect_zeroes(QemuOpts
*opts
,
1127 Error
*local_err
= NULL
;
1128 char *value
= qemu_opt_get_del(opts
, "detect-zeroes");
1129 BlockdevDetectZeroesOptions detect_zeroes
=
1130 qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup
, value
,
1131 BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF
, &local_err
);
1132 GLOBAL_STATE_CODE();
1135 error_propagate(errp
, local_err
);
1136 return detect_zeroes
;
1139 if (detect_zeroes
== BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP
&&
1140 !(open_flags
& BDRV_O_UNMAP
))
1142 error_setg(errp
, "setting detect-zeroes to unmap is not allowed "
1143 "without setting discard operation to unmap");
1146 return detect_zeroes
;
1150 * Set open flags for aio engine
1152 * Return 0 on success, -1 if the engine specified is invalid
1154 int bdrv_parse_aio(const char *mode
, int *flags
)
1156 if (!strcmp(mode
, "threads")) {
1157 /* do nothing, default */
1158 } else if (!strcmp(mode
, "native")) {
1159 *flags
|= BDRV_O_NATIVE_AIO
;
1160 #ifdef CONFIG_LINUX_IO_URING
1161 } else if (!strcmp(mode
, "io_uring")) {
1162 *flags
|= BDRV_O_IO_URING
;
1172 * Set open flags for a given discard mode
1174 * Return 0 on success, -1 if the discard mode was invalid.
1176 int bdrv_parse_discard_flags(const char *mode
, int *flags
)
1178 *flags
&= ~BDRV_O_UNMAP
;
1180 if (!strcmp(mode
, "off") || !strcmp(mode
, "ignore")) {
1182 } else if (!strcmp(mode
, "on") || !strcmp(mode
, "unmap")) {
1183 *flags
|= BDRV_O_UNMAP
;
1192 * Set open flags for a given cache mode
1194 * Return 0 on success, -1 if the cache mode was invalid.
1196 int bdrv_parse_cache_mode(const char *mode
, int *flags
, bool *writethrough
)
1198 *flags
&= ~BDRV_O_CACHE_MASK
;
1200 if (!strcmp(mode
, "off") || !strcmp(mode
, "none")) {
1201 *writethrough
= false;
1202 *flags
|= BDRV_O_NOCACHE
;
1203 } else if (!strcmp(mode
, "directsync")) {
1204 *writethrough
= true;
1205 *flags
|= BDRV_O_NOCACHE
;
1206 } else if (!strcmp(mode
, "writeback")) {
1207 *writethrough
= false;
1208 } else if (!strcmp(mode
, "unsafe")) {
1209 *writethrough
= false;
1210 *flags
|= BDRV_O_NO_FLUSH
;
1211 } else if (!strcmp(mode
, "writethrough")) {
1212 *writethrough
= true;
1220 static char *bdrv_child_get_parent_desc(BdrvChild
*c
)
1222 BlockDriverState
*parent
= c
->opaque
;
1223 return g_strdup_printf("node '%s'", bdrv_get_node_name(parent
));
1226 static void bdrv_child_cb_drained_begin(BdrvChild
*child
)
1228 BlockDriverState
*bs
= child
->opaque
;
1229 bdrv_do_drained_begin_quiesce(bs
, NULL
);
1232 static bool bdrv_child_cb_drained_poll(BdrvChild
*child
)
1234 BlockDriverState
*bs
= child
->opaque
;
1235 return bdrv_drain_poll(bs
, NULL
, false);
1238 static void bdrv_child_cb_drained_end(BdrvChild
*child
)
1240 BlockDriverState
*bs
= child
->opaque
;
1241 bdrv_drained_end(bs
);
1244 static int bdrv_child_cb_inactivate(BdrvChild
*child
)
1246 BlockDriverState
*bs
= child
->opaque
;
1247 GLOBAL_STATE_CODE();
1248 assert(bs
->open_flags
& BDRV_O_INACTIVE
);
1252 static bool bdrv_child_cb_change_aio_ctx(BdrvChild
*child
, AioContext
*ctx
,
1253 GHashTable
*visited
, Transaction
*tran
,
1256 BlockDriverState
*bs
= child
->opaque
;
1257 return bdrv_change_aio_context(bs
, ctx
, visited
, tran
, errp
);
1261 * Returns the options and flags that a temporary snapshot should get, based on
1262 * the originally requested flags (the originally requested image will have
1263 * flags like a backing file)
1265 static void bdrv_temp_snapshot_options(int *child_flags
, QDict
*child_options
,
1266 int parent_flags
, QDict
*parent_options
)
1268 GLOBAL_STATE_CODE();
1269 *child_flags
= (parent_flags
& ~BDRV_O_SNAPSHOT
) | BDRV_O_TEMPORARY
;
1271 /* For temporary files, unconditional cache=unsafe is fine */
1272 qdict_set_default_str(child_options
, BDRV_OPT_CACHE_DIRECT
, "off");
1273 qdict_set_default_str(child_options
, BDRV_OPT_CACHE_NO_FLUSH
, "on");
1275 /* Copy the read-only and discard options from the parent */
1276 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_READ_ONLY
);
1277 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_DISCARD
);
1279 /* aio=native doesn't work for cache.direct=off, so disable it for the
1280 * temporary snapshot */
1281 *child_flags
&= ~BDRV_O_NATIVE_AIO
;
1284 static void bdrv_backing_attach(BdrvChild
*c
)
1286 BlockDriverState
*parent
= c
->opaque
;
1287 BlockDriverState
*backing_hd
= c
->bs
;
1289 GLOBAL_STATE_CODE();
1290 assert(!parent
->backing_blocker
);
1291 error_setg(&parent
->backing_blocker
,
1292 "node is used as backing hd of '%s'",
1293 bdrv_get_device_or_node_name(parent
));
1295 bdrv_refresh_filename(backing_hd
);
1297 parent
->open_flags
&= ~BDRV_O_NO_BACKING
;
1299 bdrv_op_block_all(backing_hd
, parent
->backing_blocker
);
1300 /* Otherwise we won't be able to commit or stream */
1301 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_COMMIT_TARGET
,
1302 parent
->backing_blocker
);
1303 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_STREAM
,
1304 parent
->backing_blocker
);
1306 * We do backup in 3 ways:
1308 * The target bs is new opened, and the source is top BDS
1309 * 2. blockdev backup
1310 * Both the source and the target are top BDSes.
1311 * 3. internal backup(used for block replication)
1312 * Both the source and the target are backing file
1314 * In case 1 and 2, neither the source nor the target is the backing file.
1315 * In case 3, we will block the top BDS, so there is only one block job
1316 * for the top BDS and its backing chain.
1318 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_BACKUP_SOURCE
,
1319 parent
->backing_blocker
);
1320 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_BACKUP_TARGET
,
1321 parent
->backing_blocker
);
1324 static void bdrv_backing_detach(BdrvChild
*c
)
1326 BlockDriverState
*parent
= c
->opaque
;
1328 GLOBAL_STATE_CODE();
1329 assert(parent
->backing_blocker
);
1330 bdrv_op_unblock_all(c
->bs
, parent
->backing_blocker
);
1331 error_free(parent
->backing_blocker
);
1332 parent
->backing_blocker
= NULL
;
1335 static int bdrv_backing_update_filename(BdrvChild
*c
, BlockDriverState
*base
,
1336 const char *filename
, Error
**errp
)
1338 BlockDriverState
*parent
= c
->opaque
;
1339 bool read_only
= bdrv_is_read_only(parent
);
1341 GLOBAL_STATE_CODE();
1344 ret
= bdrv_reopen_set_read_only(parent
, false, errp
);
1350 ret
= bdrv_change_backing_file(parent
, filename
,
1351 base
->drv
? base
->drv
->format_name
: "",
1354 error_setg_errno(errp
, -ret
, "Could not update backing file link");
1358 bdrv_reopen_set_read_only(parent
, true, NULL
);
1365 * Returns the options and flags that a generic child of a BDS should
1366 * get, based on the given options and flags for the parent BDS.
1368 static void bdrv_inherited_options(BdrvChildRole role
, bool parent_is_format
,
1369 int *child_flags
, QDict
*child_options
,
1370 int parent_flags
, QDict
*parent_options
)
1372 int flags
= parent_flags
;
1373 GLOBAL_STATE_CODE();
1376 * First, decide whether to set, clear, or leave BDRV_O_PROTOCOL.
1377 * Generally, the question to answer is: Should this child be
1378 * format-probed by default?
1382 * Pure and non-filtered data children of non-format nodes should
1383 * be probed by default (even when the node itself has BDRV_O_PROTOCOL
1384 * set). This only affects a very limited set of drivers (namely
1385 * quorum and blkverify when this comment was written).
1386 * Force-clear BDRV_O_PROTOCOL then.
1388 if (!parent_is_format
&&
1389 (role
& BDRV_CHILD_DATA
) &&
1390 !(role
& (BDRV_CHILD_METADATA
| BDRV_CHILD_FILTERED
)))
1392 flags
&= ~BDRV_O_PROTOCOL
;
1396 * All children of format nodes (except for COW children) and all
1397 * metadata children in general should never be format-probed.
1398 * Force-set BDRV_O_PROTOCOL then.
1400 if ((parent_is_format
&& !(role
& BDRV_CHILD_COW
)) ||
1401 (role
& BDRV_CHILD_METADATA
))
1403 flags
|= BDRV_O_PROTOCOL
;
1407 * If the cache mode isn't explicitly set, inherit direct and no-flush from
1410 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_CACHE_DIRECT
);
1411 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_CACHE_NO_FLUSH
);
1412 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_FORCE_SHARE
);
1414 if (role
& BDRV_CHILD_COW
) {
1415 /* backing files are opened read-only by default */
1416 qdict_set_default_str(child_options
, BDRV_OPT_READ_ONLY
, "on");
1417 qdict_set_default_str(child_options
, BDRV_OPT_AUTO_READ_ONLY
, "off");
1419 /* Inherit the read-only option from the parent if it's not set */
1420 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_READ_ONLY
);
1421 qdict_copy_default(child_options
, parent_options
,
1422 BDRV_OPT_AUTO_READ_ONLY
);
1426 * bdrv_co_pdiscard() respects unmap policy for the parent, so we
1427 * can default to enable it on lower layers regardless of the
1430 qdict_set_default_str(child_options
, BDRV_OPT_DISCARD
, "unmap");
1432 /* Clear flags that only apply to the top layer */
1433 flags
&= ~(BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
| BDRV_O_COPY_ON_READ
);
1435 if (role
& BDRV_CHILD_METADATA
) {
1436 flags
&= ~BDRV_O_NO_IO
;
1438 if (role
& BDRV_CHILD_COW
) {
1439 flags
&= ~BDRV_O_TEMPORARY
;
1442 *child_flags
= flags
;
1445 static void bdrv_child_cb_attach(BdrvChild
*child
)
1447 BlockDriverState
*bs
= child
->opaque
;
1449 assert_bdrv_graph_writable(bs
);
1450 QLIST_INSERT_HEAD(&bs
->children
, child
, next
);
1451 if (bs
->drv
->is_filter
|| (child
->role
& BDRV_CHILD_FILTERED
)) {
1453 * Here we handle filters and block/raw-format.c when it behave like
1454 * filter. They generally have a single PRIMARY child, which is also the
1455 * FILTERED child, and that they may have multiple more children, which
1456 * are neither PRIMARY nor FILTERED. And never we have a COW child here.
1457 * So bs->file will be the PRIMARY child, unless the PRIMARY child goes
1458 * into bs->backing on exceptional cases; and bs->backing will be
1461 assert(!(child
->role
& BDRV_CHILD_COW
));
1462 if (child
->role
& BDRV_CHILD_PRIMARY
) {
1463 assert(child
->role
& BDRV_CHILD_FILTERED
);
1464 assert(!bs
->backing
);
1467 if (bs
->drv
->filtered_child_is_backing
) {
1468 bs
->backing
= child
;
1473 assert(!(child
->role
& BDRV_CHILD_FILTERED
));
1475 } else if (child
->role
& BDRV_CHILD_COW
) {
1476 assert(bs
->drv
->supports_backing
);
1477 assert(!(child
->role
& BDRV_CHILD_PRIMARY
));
1478 assert(!bs
->backing
);
1479 bs
->backing
= child
;
1480 bdrv_backing_attach(child
);
1481 } else if (child
->role
& BDRV_CHILD_PRIMARY
) {
1487 static void bdrv_child_cb_detach(BdrvChild
*child
)
1489 BlockDriverState
*bs
= child
->opaque
;
1491 if (child
->role
& BDRV_CHILD_COW
) {
1492 bdrv_backing_detach(child
);
1495 assert_bdrv_graph_writable(bs
);
1496 QLIST_REMOVE(child
, next
);
1497 if (child
== bs
->backing
) {
1498 assert(child
!= bs
->file
);
1500 } else if (child
== bs
->file
) {
1505 static int bdrv_child_cb_update_filename(BdrvChild
*c
, BlockDriverState
*base
,
1506 const char *filename
, Error
**errp
)
1508 if (c
->role
& BDRV_CHILD_COW
) {
1509 return bdrv_backing_update_filename(c
, base
, filename
, errp
);
1514 AioContext
*child_of_bds_get_parent_aio_context(BdrvChild
*c
)
1516 BlockDriverState
*bs
= c
->opaque
;
1519 return bdrv_get_aio_context(bs
);
1522 const BdrvChildClass child_of_bds
= {
1523 .parent_is_bds
= true,
1524 .get_parent_desc
= bdrv_child_get_parent_desc
,
1525 .inherit_options
= bdrv_inherited_options
,
1526 .drained_begin
= bdrv_child_cb_drained_begin
,
1527 .drained_poll
= bdrv_child_cb_drained_poll
,
1528 .drained_end
= bdrv_child_cb_drained_end
,
1529 .attach
= bdrv_child_cb_attach
,
1530 .detach
= bdrv_child_cb_detach
,
1531 .inactivate
= bdrv_child_cb_inactivate
,
1532 .change_aio_ctx
= bdrv_child_cb_change_aio_ctx
,
1533 .update_filename
= bdrv_child_cb_update_filename
,
1534 .get_parent_aio_context
= child_of_bds_get_parent_aio_context
,
1537 AioContext
*bdrv_child_get_parent_aio_context(BdrvChild
*c
)
1540 return c
->klass
->get_parent_aio_context(c
);
1543 static int bdrv_open_flags(BlockDriverState
*bs
, int flags
)
1545 int open_flags
= flags
;
1546 GLOBAL_STATE_CODE();
1549 * Clear flags that are internal to the block layer before opening the
1552 open_flags
&= ~(BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
| BDRV_O_PROTOCOL
);
1557 static void update_flags_from_options(int *flags
, QemuOpts
*opts
)
1559 GLOBAL_STATE_CODE();
1561 *flags
&= ~(BDRV_O_CACHE_MASK
| BDRV_O_RDWR
| BDRV_O_AUTO_RDONLY
);
1563 if (qemu_opt_get_bool_del(opts
, BDRV_OPT_CACHE_NO_FLUSH
, false)) {
1564 *flags
|= BDRV_O_NO_FLUSH
;
1567 if (qemu_opt_get_bool_del(opts
, BDRV_OPT_CACHE_DIRECT
, false)) {
1568 *flags
|= BDRV_O_NOCACHE
;
1571 if (!qemu_opt_get_bool_del(opts
, BDRV_OPT_READ_ONLY
, false)) {
1572 *flags
|= BDRV_O_RDWR
;
1575 if (qemu_opt_get_bool_del(opts
, BDRV_OPT_AUTO_READ_ONLY
, false)) {
1576 *flags
|= BDRV_O_AUTO_RDONLY
;
1580 static void update_options_from_flags(QDict
*options
, int flags
)
1582 GLOBAL_STATE_CODE();
1583 if (!qdict_haskey(options
, BDRV_OPT_CACHE_DIRECT
)) {
1584 qdict_put_bool(options
, BDRV_OPT_CACHE_DIRECT
, flags
& BDRV_O_NOCACHE
);
1586 if (!qdict_haskey(options
, BDRV_OPT_CACHE_NO_FLUSH
)) {
1587 qdict_put_bool(options
, BDRV_OPT_CACHE_NO_FLUSH
,
1588 flags
& BDRV_O_NO_FLUSH
);
1590 if (!qdict_haskey(options
, BDRV_OPT_READ_ONLY
)) {
1591 qdict_put_bool(options
, BDRV_OPT_READ_ONLY
, !(flags
& BDRV_O_RDWR
));
1593 if (!qdict_haskey(options
, BDRV_OPT_AUTO_READ_ONLY
)) {
1594 qdict_put_bool(options
, BDRV_OPT_AUTO_READ_ONLY
,
1595 flags
& BDRV_O_AUTO_RDONLY
);
1599 static void bdrv_assign_node_name(BlockDriverState
*bs
,
1600 const char *node_name
,
1603 char *gen_node_name
= NULL
;
1604 GLOBAL_STATE_CODE();
1607 node_name
= gen_node_name
= id_generate(ID_BLOCK
);
1608 } else if (!id_wellformed(node_name
)) {
1610 * Check for empty string or invalid characters, but not if it is
1611 * generated (generated names use characters not available to the user)
1613 error_setg(errp
, "Invalid node-name: '%s'", node_name
);
1617 /* takes care of avoiding namespaces collisions */
1618 if (blk_by_name(node_name
)) {
1619 error_setg(errp
, "node-name=%s is conflicting with a device id",
1624 /* takes care of avoiding duplicates node names */
1625 if (bdrv_find_node(node_name
)) {
1626 error_setg(errp
, "Duplicate nodes with node-name='%s'", node_name
);
1630 /* Make sure that the node name isn't truncated */
1631 if (strlen(node_name
) >= sizeof(bs
->node_name
)) {
1632 error_setg(errp
, "Node name too long");
1636 /* copy node name into the bs and insert it into the graph list */
1637 pstrcpy(bs
->node_name
, sizeof(bs
->node_name
), node_name
);
1638 QTAILQ_INSERT_TAIL(&graph_bdrv_states
, bs
, node_list
);
1640 g_free(gen_node_name
);
1643 static int bdrv_open_driver(BlockDriverState
*bs
, BlockDriver
*drv
,
1644 const char *node_name
, QDict
*options
,
1645 int open_flags
, Error
**errp
)
1647 Error
*local_err
= NULL
;
1649 GLOBAL_STATE_CODE();
1651 bdrv_assign_node_name(bs
, node_name
, &local_err
);
1653 error_propagate(errp
, local_err
);
1658 bs
->opaque
= g_malloc0(drv
->instance_size
);
1660 if (drv
->bdrv_file_open
) {
1661 assert(!drv
->bdrv_needs_filename
|| bs
->filename
[0]);
1662 ret
= drv
->bdrv_file_open(bs
, options
, open_flags
, &local_err
);
1663 } else if (drv
->bdrv_open
) {
1664 ret
= drv
->bdrv_open(bs
, options
, open_flags
, &local_err
);
1671 error_propagate(errp
, local_err
);
1672 } else if (bs
->filename
[0]) {
1673 error_setg_errno(errp
, -ret
, "Could not open '%s'", bs
->filename
);
1675 error_setg_errno(errp
, -ret
, "Could not open image");
1680 assert(!(bs
->supported_read_flags
& ~BDRV_REQ_MASK
));
1681 assert(!(bs
->supported_write_flags
& ~BDRV_REQ_MASK
));
1684 * Always allow the BDRV_REQ_REGISTERED_BUF optimization hint. This saves
1685 * drivers that pass read/write requests through to a child the trouble of
1686 * declaring support explicitly.
1688 * Drivers must not propagate this flag accidentally when they initiate I/O
1689 * to a bounce buffer. That case should be rare though.
1691 bs
->supported_read_flags
|= BDRV_REQ_REGISTERED_BUF
;
1692 bs
->supported_write_flags
|= BDRV_REQ_REGISTERED_BUF
;
1694 ret
= refresh_total_sectors(bs
, bs
->total_sectors
);
1696 error_setg_errno(errp
, -ret
, "Could not refresh total sector count");
1700 bdrv_refresh_limits(bs
, NULL
, &local_err
);
1702 error_propagate(errp
, local_err
);
1706 assert(bdrv_opt_mem_align(bs
) != 0);
1707 assert(bdrv_min_mem_align(bs
) != 0);
1708 assert(is_power_of_2(bs
->bl
.request_alignment
));
1710 for (i
= 0; i
< bs
->quiesce_counter
; i
++) {
1711 if (drv
->bdrv_drain_begin
) {
1712 drv
->bdrv_drain_begin(bs
);
1719 if (bs
->file
!= NULL
) {
1720 bdrv_unref_child(bs
, bs
->file
);
1729 * Create and open a block node.
1731 * @options is a QDict of options to pass to the block drivers, or NULL for an
1732 * empty set of options. The reference to the QDict belongs to the block layer
1733 * after the call (even on failure), so if the caller intends to reuse the
1734 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
1736 BlockDriverState
*bdrv_new_open_driver_opts(BlockDriver
*drv
,
1737 const char *node_name
,
1738 QDict
*options
, int flags
,
1741 BlockDriverState
*bs
;
1744 GLOBAL_STATE_CODE();
1747 bs
->open_flags
= flags
;
1748 bs
->options
= options
?: qdict_new();
1749 bs
->explicit_options
= qdict_clone_shallow(bs
->options
);
1752 update_options_from_flags(bs
->options
, flags
);
1754 ret
= bdrv_open_driver(bs
, drv
, node_name
, bs
->options
, flags
, errp
);
1756 qobject_unref(bs
->explicit_options
);
1757 bs
->explicit_options
= NULL
;
1758 qobject_unref(bs
->options
);
1767 /* Create and open a block node. */
1768 BlockDriverState
*bdrv_new_open_driver(BlockDriver
*drv
, const char *node_name
,
1769 int flags
, Error
**errp
)
1771 GLOBAL_STATE_CODE();
1772 return bdrv_new_open_driver_opts(drv
, node_name
, NULL
, flags
, errp
);
1775 QemuOptsList bdrv_runtime_opts
= {
1776 .name
= "bdrv_common",
1777 .head
= QTAILQ_HEAD_INITIALIZER(bdrv_runtime_opts
.head
),
1780 .name
= "node-name",
1781 .type
= QEMU_OPT_STRING
,
1782 .help
= "Node name of the block device node",
1786 .type
= QEMU_OPT_STRING
,
1787 .help
= "Block driver to use for the node",
1790 .name
= BDRV_OPT_CACHE_DIRECT
,
1791 .type
= QEMU_OPT_BOOL
,
1792 .help
= "Bypass software writeback cache on the host",
1795 .name
= BDRV_OPT_CACHE_NO_FLUSH
,
1796 .type
= QEMU_OPT_BOOL
,
1797 .help
= "Ignore flush requests",
1800 .name
= BDRV_OPT_READ_ONLY
,
1801 .type
= QEMU_OPT_BOOL
,
1802 .help
= "Node is opened in read-only mode",
1805 .name
= BDRV_OPT_AUTO_READ_ONLY
,
1806 .type
= QEMU_OPT_BOOL
,
1807 .help
= "Node can become read-only if opening read-write fails",
1810 .name
= "detect-zeroes",
1811 .type
= QEMU_OPT_STRING
,
1812 .help
= "try to optimize zero writes (off, on, unmap)",
1815 .name
= BDRV_OPT_DISCARD
,
1816 .type
= QEMU_OPT_STRING
,
1817 .help
= "discard operation (ignore/off, unmap/on)",
1820 .name
= BDRV_OPT_FORCE_SHARE
,
1821 .type
= QEMU_OPT_BOOL
,
1822 .help
= "always accept other writers (default: off)",
1824 { /* end of list */ }
1828 QemuOptsList bdrv_create_opts_simple
= {
1829 .name
= "simple-create-opts",
1830 .head
= QTAILQ_HEAD_INITIALIZER(bdrv_create_opts_simple
.head
),
1833 .name
= BLOCK_OPT_SIZE
,
1834 .type
= QEMU_OPT_SIZE
,
1835 .help
= "Virtual disk size"
1838 .name
= BLOCK_OPT_PREALLOC
,
1839 .type
= QEMU_OPT_STRING
,
1840 .help
= "Preallocation mode (allowed values: off)"
1842 { /* end of list */ }
1847 * Common part for opening disk images and files
1849 * Removes all processed options from *options.
1851 static int bdrv_open_common(BlockDriverState
*bs
, BlockBackend
*file
,
1852 QDict
*options
, Error
**errp
)
1854 int ret
, open_flags
;
1855 const char *filename
;
1856 const char *driver_name
= NULL
;
1857 const char *node_name
= NULL
;
1858 const char *discard
;
1861 Error
*local_err
= NULL
;
1864 assert(bs
->file
== NULL
);
1865 assert(options
!= NULL
&& bs
->options
!= options
);
1866 GLOBAL_STATE_CODE();
1868 opts
= qemu_opts_create(&bdrv_runtime_opts
, NULL
, 0, &error_abort
);
1869 if (!qemu_opts_absorb_qdict(opts
, options
, errp
)) {
1874 update_flags_from_options(&bs
->open_flags
, opts
);
1876 driver_name
= qemu_opt_get(opts
, "driver");
1877 drv
= bdrv_find_format(driver_name
);
1878 assert(drv
!= NULL
);
1880 bs
->force_share
= qemu_opt_get_bool(opts
, BDRV_OPT_FORCE_SHARE
, false);
1882 if (bs
->force_share
&& (bs
->open_flags
& BDRV_O_RDWR
)) {
1884 BDRV_OPT_FORCE_SHARE
1885 "=on can only be used with read-only images");
1891 bdrv_refresh_filename(blk_bs(file
));
1892 filename
= blk_bs(file
)->filename
;
1895 * Caution: while qdict_get_try_str() is fine, getting
1896 * non-string types would require more care. When @options
1897 * come from -blockdev or blockdev_add, its members are typed
1898 * according to the QAPI schema, but when they come from
1899 * -drive, they're all QString.
1901 filename
= qdict_get_try_str(options
, "filename");
1904 if (drv
->bdrv_needs_filename
&& (!filename
|| !filename
[0])) {
1905 error_setg(errp
, "The '%s' block driver requires a file name",
1911 trace_bdrv_open_common(bs
, filename
?: "", bs
->open_flags
,
1914 ro
= bdrv_is_read_only(bs
);
1916 if (use_bdrv_whitelist
&& !bdrv_is_whitelisted(drv
, ro
)) {
1917 if (!ro
&& bdrv_is_whitelisted(drv
, true)) {
1918 ret
= bdrv_apply_auto_read_only(bs
, NULL
, NULL
);
1924 !ro
&& bdrv_is_whitelisted(drv
, true)
1925 ? "Driver '%s' can only be used for read-only devices"
1926 : "Driver '%s' is not whitelisted",
1932 /* bdrv_new() and bdrv_close() make it so */
1933 assert(qatomic_read(&bs
->copy_on_read
) == 0);
1935 if (bs
->open_flags
& BDRV_O_COPY_ON_READ
) {
1937 bdrv_enable_copy_on_read(bs
);
1939 error_setg(errp
, "Can't use copy-on-read on read-only device");
1945 discard
= qemu_opt_get(opts
, BDRV_OPT_DISCARD
);
1946 if (discard
!= NULL
) {
1947 if (bdrv_parse_discard_flags(discard
, &bs
->open_flags
) != 0) {
1948 error_setg(errp
, "Invalid discard option");
1955 bdrv_parse_detect_zeroes(opts
, bs
->open_flags
, &local_err
);
1957 error_propagate(errp
, local_err
);
1962 if (filename
!= NULL
) {
1963 pstrcpy(bs
->filename
, sizeof(bs
->filename
), filename
);
1965 bs
->filename
[0] = '\0';
1967 pstrcpy(bs
->exact_filename
, sizeof(bs
->exact_filename
), bs
->filename
);
1969 /* Open the image, either directly or using a protocol */
1970 open_flags
= bdrv_open_flags(bs
, bs
->open_flags
);
1971 node_name
= qemu_opt_get(opts
, "node-name");
1973 assert(!drv
->bdrv_file_open
|| file
== NULL
);
1974 ret
= bdrv_open_driver(bs
, drv
, node_name
, options
, open_flags
, errp
);
1979 qemu_opts_del(opts
);
1983 qemu_opts_del(opts
);
1987 static QDict
*parse_json_filename(const char *filename
, Error
**errp
)
1989 QObject
*options_obj
;
1992 GLOBAL_STATE_CODE();
1994 ret
= strstart(filename
, "json:", &filename
);
1997 options_obj
= qobject_from_json(filename
, errp
);
1999 error_prepend(errp
, "Could not parse the JSON options: ");
2003 options
= qobject_to(QDict
, options_obj
);
2005 qobject_unref(options_obj
);
2006 error_setg(errp
, "Invalid JSON object given");
2010 qdict_flatten(options
);
2015 static void parse_json_protocol(QDict
*options
, const char **pfilename
,
2018 QDict
*json_options
;
2019 Error
*local_err
= NULL
;
2020 GLOBAL_STATE_CODE();
2022 /* Parse json: pseudo-protocol */
2023 if (!*pfilename
|| !g_str_has_prefix(*pfilename
, "json:")) {
2027 json_options
= parse_json_filename(*pfilename
, &local_err
);
2029 error_propagate(errp
, local_err
);
2033 /* Options given in the filename have lower priority than options
2034 * specified directly */
2035 qdict_join(options
, json_options
, false);
2036 qobject_unref(json_options
);
2041 * Fills in default options for opening images and converts the legacy
2042 * filename/flags pair to option QDict entries.
2043 * The BDRV_O_PROTOCOL flag in *flags will be set or cleared accordingly if a
2044 * block driver has been specified explicitly.
2046 static int bdrv_fill_options(QDict
**options
, const char *filename
,
2047 int *flags
, Error
**errp
)
2049 const char *drvname
;
2050 bool protocol
= *flags
& BDRV_O_PROTOCOL
;
2051 bool parse_filename
= false;
2052 BlockDriver
*drv
= NULL
;
2053 Error
*local_err
= NULL
;
2055 GLOBAL_STATE_CODE();
2058 * Caution: while qdict_get_try_str() is fine, getting non-string
2059 * types would require more care. When @options come from
2060 * -blockdev or blockdev_add, its members are typed according to
2061 * the QAPI schema, but when they come from -drive, they're all
2064 drvname
= qdict_get_try_str(*options
, "driver");
2066 drv
= bdrv_find_format(drvname
);
2068 error_setg(errp
, "Unknown driver '%s'", drvname
);
2071 /* If the user has explicitly specified the driver, this choice should
2072 * override the BDRV_O_PROTOCOL flag */
2073 protocol
= drv
->bdrv_file_open
;
2077 *flags
|= BDRV_O_PROTOCOL
;
2079 *flags
&= ~BDRV_O_PROTOCOL
;
2082 /* Translate cache options from flags into options */
2083 update_options_from_flags(*options
, *flags
);
2085 /* Fetch the file name from the options QDict if necessary */
2086 if (protocol
&& filename
) {
2087 if (!qdict_haskey(*options
, "filename")) {
2088 qdict_put_str(*options
, "filename", filename
);
2089 parse_filename
= true;
2091 error_setg(errp
, "Can't specify 'file' and 'filename' options at "
2097 /* Find the right block driver */
2098 /* See cautionary note on accessing @options above */
2099 filename
= qdict_get_try_str(*options
, "filename");
2101 if (!drvname
&& protocol
) {
2103 drv
= bdrv_find_protocol(filename
, parse_filename
, errp
);
2108 drvname
= drv
->format_name
;
2109 qdict_put_str(*options
, "driver", drvname
);
2111 error_setg(errp
, "Must specify either driver or file");
2116 assert(drv
|| !protocol
);
2118 /* Driver-specific filename parsing */
2119 if (drv
&& drv
->bdrv_parse_filename
&& parse_filename
) {
2120 drv
->bdrv_parse_filename(filename
, *options
, &local_err
);
2122 error_propagate(errp
, local_err
);
2126 if (!drv
->bdrv_needs_filename
) {
2127 qdict_del(*options
, "filename");
2134 typedef struct BlockReopenQueueEntry
{
2137 BDRVReopenState state
;
2138 QTAILQ_ENTRY(BlockReopenQueueEntry
) entry
;
2139 } BlockReopenQueueEntry
;
2142 * Return the flags that @bs will have after the reopens in @q have
2143 * successfully completed. If @q is NULL (or @bs is not contained in @q),
2144 * return the current flags.
2146 static int bdrv_reopen_get_flags(BlockReopenQueue
*q
, BlockDriverState
*bs
)
2148 BlockReopenQueueEntry
*entry
;
2151 QTAILQ_FOREACH(entry
, q
, entry
) {
2152 if (entry
->state
.bs
== bs
) {
2153 return entry
->state
.flags
;
2158 return bs
->open_flags
;
2161 /* Returns whether the image file can be written to after the reopen queue @q
2162 * has been successfully applied, or right now if @q is NULL. */
2163 static bool bdrv_is_writable_after_reopen(BlockDriverState
*bs
,
2164 BlockReopenQueue
*q
)
2166 int flags
= bdrv_reopen_get_flags(q
, bs
);
2168 return (flags
& (BDRV_O_RDWR
| BDRV_O_INACTIVE
)) == BDRV_O_RDWR
;
2172 * Return whether the BDS can be written to. This is not necessarily
2173 * the same as !bdrv_is_read_only(bs), as inactivated images may not
2174 * be written to but do not count as read-only images.
2176 bool bdrv_is_writable(BlockDriverState
*bs
)
2179 return bdrv_is_writable_after_reopen(bs
, NULL
);
2182 static char *bdrv_child_user_desc(BdrvChild
*c
)
2184 GLOBAL_STATE_CODE();
2185 return c
->klass
->get_parent_desc(c
);
2189 * Check that @a allows everything that @b needs. @a and @b must reference same
2192 static bool bdrv_a_allow_b(BdrvChild
*a
, BdrvChild
*b
, Error
**errp
)
2194 const char *child_bs_name
;
2195 g_autofree
char *a_user
= NULL
;
2196 g_autofree
char *b_user
= NULL
;
2197 g_autofree
char *perms
= NULL
;
2200 assert(a
->bs
== b
->bs
);
2201 GLOBAL_STATE_CODE();
2203 if ((b
->perm
& a
->shared_perm
) == b
->perm
) {
2207 child_bs_name
= bdrv_get_node_name(b
->bs
);
2208 a_user
= bdrv_child_user_desc(a
);
2209 b_user
= bdrv_child_user_desc(b
);
2210 perms
= bdrv_perm_names(b
->perm
& ~a
->shared_perm
);
2212 error_setg(errp
, "Permission conflict on node '%s': permissions '%s' are "
2213 "both required by %s (uses node '%s' as '%s' child) and "
2214 "unshared by %s (uses node '%s' as '%s' child).",
2215 child_bs_name
, perms
,
2216 b_user
, child_bs_name
, b
->name
,
2217 a_user
, child_bs_name
, a
->name
);
2222 static bool bdrv_parent_perms_conflict(BlockDriverState
*bs
, Error
**errp
)
2225 GLOBAL_STATE_CODE();
2228 * During the loop we'll look at each pair twice. That's correct because
2229 * bdrv_a_allow_b() is asymmetric and we should check each pair in both
2232 QLIST_FOREACH(a
, &bs
->parents
, next_parent
) {
2233 QLIST_FOREACH(b
, &bs
->parents
, next_parent
) {
2238 if (!bdrv_a_allow_b(a
, b
, errp
)) {
2247 static void bdrv_child_perm(BlockDriverState
*bs
, BlockDriverState
*child_bs
,
2248 BdrvChild
*c
, BdrvChildRole role
,
2249 BlockReopenQueue
*reopen_queue
,
2250 uint64_t parent_perm
, uint64_t parent_shared
,
2251 uint64_t *nperm
, uint64_t *nshared
)
2253 assert(bs
->drv
&& bs
->drv
->bdrv_child_perm
);
2254 GLOBAL_STATE_CODE();
2255 bs
->drv
->bdrv_child_perm(bs
, c
, role
, reopen_queue
,
2256 parent_perm
, parent_shared
,
2258 /* TODO Take force_share from reopen_queue */
2259 if (child_bs
&& child_bs
->force_share
) {
2260 *nshared
= BLK_PERM_ALL
;
2265 * Adds the whole subtree of @bs (including @bs itself) to the @list (except for
2266 * nodes that are already in the @list, of course) so that final list is
2267 * topologically sorted. Return the result (GSList @list object is updated, so
2268 * don't use old reference after function call).
2270 * On function start @list must be already topologically sorted and for any node
2271 * in the @list the whole subtree of the node must be in the @list as well. The
2272 * simplest way to satisfy this criteria: use only result of
2273 * bdrv_topological_dfs() or NULL as @list parameter.
2275 static GSList
*bdrv_topological_dfs(GSList
*list
, GHashTable
*found
,
2276 BlockDriverState
*bs
)
2279 g_autoptr(GHashTable
) local_found
= NULL
;
2281 GLOBAL_STATE_CODE();
2285 found
= local_found
= g_hash_table_new(NULL
, NULL
);
2288 if (g_hash_table_contains(found
, bs
)) {
2291 g_hash_table_add(found
, bs
);
2293 QLIST_FOREACH(child
, &bs
->children
, next
) {
2294 list
= bdrv_topological_dfs(list
, found
, child
->bs
);
2297 return g_slist_prepend(list
, bs
);
2300 typedef struct BdrvChildSetPermState
{
2303 uint64_t old_shared_perm
;
2304 } BdrvChildSetPermState
;
2306 static void bdrv_child_set_perm_abort(void *opaque
)
2308 BdrvChildSetPermState
*s
= opaque
;
2310 GLOBAL_STATE_CODE();
2312 s
->child
->perm
= s
->old_perm
;
2313 s
->child
->shared_perm
= s
->old_shared_perm
;
2316 static TransactionActionDrv bdrv_child_set_pem_drv
= {
2317 .abort
= bdrv_child_set_perm_abort
,
2321 static void bdrv_child_set_perm(BdrvChild
*c
, uint64_t perm
,
2322 uint64_t shared
, Transaction
*tran
)
2324 BdrvChildSetPermState
*s
= g_new(BdrvChildSetPermState
, 1);
2325 GLOBAL_STATE_CODE();
2327 *s
= (BdrvChildSetPermState
) {
2329 .old_perm
= c
->perm
,
2330 .old_shared_perm
= c
->shared_perm
,
2334 c
->shared_perm
= shared
;
2336 tran_add(tran
, &bdrv_child_set_pem_drv
, s
);
2339 static void bdrv_drv_set_perm_commit(void *opaque
)
2341 BlockDriverState
*bs
= opaque
;
2342 uint64_t cumulative_perms
, cumulative_shared_perms
;
2343 GLOBAL_STATE_CODE();
2345 if (bs
->drv
->bdrv_set_perm
) {
2346 bdrv_get_cumulative_perm(bs
, &cumulative_perms
,
2347 &cumulative_shared_perms
);
2348 bs
->drv
->bdrv_set_perm(bs
, cumulative_perms
, cumulative_shared_perms
);
2352 static void bdrv_drv_set_perm_abort(void *opaque
)
2354 BlockDriverState
*bs
= opaque
;
2355 GLOBAL_STATE_CODE();
2357 if (bs
->drv
->bdrv_abort_perm_update
) {
2358 bs
->drv
->bdrv_abort_perm_update(bs
);
2362 TransactionActionDrv bdrv_drv_set_perm_drv
= {
2363 .abort
= bdrv_drv_set_perm_abort
,
2364 .commit
= bdrv_drv_set_perm_commit
,
2367 static int bdrv_drv_set_perm(BlockDriverState
*bs
, uint64_t perm
,
2368 uint64_t shared_perm
, Transaction
*tran
,
2371 GLOBAL_STATE_CODE();
2376 if (bs
->drv
->bdrv_check_perm
) {
2377 int ret
= bs
->drv
->bdrv_check_perm(bs
, perm
, shared_perm
, errp
);
2384 tran_add(tran
, &bdrv_drv_set_perm_drv
, bs
);
2390 typedef struct BdrvReplaceChildState
{
2392 BlockDriverState
*old_bs
;
2393 } BdrvReplaceChildState
;
2395 static void bdrv_replace_child_commit(void *opaque
)
2397 BdrvReplaceChildState
*s
= opaque
;
2398 GLOBAL_STATE_CODE();
2400 bdrv_unref(s
->old_bs
);
2403 static void bdrv_replace_child_abort(void *opaque
)
2405 BdrvReplaceChildState
*s
= opaque
;
2406 BlockDriverState
*new_bs
= s
->child
->bs
;
2408 GLOBAL_STATE_CODE();
2409 /* old_bs reference is transparently moved from @s to @s->child */
2410 bdrv_replace_child_noperm(s
->child
, s
->old_bs
);
2414 static TransactionActionDrv bdrv_replace_child_drv
= {
2415 .commit
= bdrv_replace_child_commit
,
2416 .abort
= bdrv_replace_child_abort
,
2421 * bdrv_replace_child_tran
2423 * Note: real unref of old_bs is done only on commit.
2425 * The function doesn't update permissions, caller is responsible for this.
2427 static void bdrv_replace_child_tran(BdrvChild
*child
, BlockDriverState
*new_bs
,
2430 BdrvReplaceChildState
*s
= g_new(BdrvReplaceChildState
, 1);
2431 *s
= (BdrvReplaceChildState
) {
2433 .old_bs
= child
->bs
,
2435 tran_add(tran
, &bdrv_replace_child_drv
, s
);
2440 bdrv_replace_child_noperm(child
, new_bs
);
2441 /* old_bs reference is transparently moved from @child to @s */
2445 * Refresh permissions in @bs subtree. The function is intended to be called
2446 * after some graph modification that was done without permission update.
2448 static int bdrv_node_refresh_perm(BlockDriverState
*bs
, BlockReopenQueue
*q
,
2449 Transaction
*tran
, Error
**errp
)
2451 BlockDriver
*drv
= bs
->drv
;
2454 uint64_t cumulative_perms
, cumulative_shared_perms
;
2455 GLOBAL_STATE_CODE();
2457 bdrv_get_cumulative_perm(bs
, &cumulative_perms
, &cumulative_shared_perms
);
2459 /* Write permissions never work with read-only images */
2460 if ((cumulative_perms
& (BLK_PERM_WRITE
| BLK_PERM_WRITE_UNCHANGED
)) &&
2461 !bdrv_is_writable_after_reopen(bs
, q
))
2463 if (!bdrv_is_writable_after_reopen(bs
, NULL
)) {
2464 error_setg(errp
, "Block node is read-only");
2466 error_setg(errp
, "Read-only block node '%s' cannot support "
2467 "read-write users", bdrv_get_node_name(bs
));
2474 * Unaligned requests will automatically be aligned to bl.request_alignment
2475 * and without RESIZE we can't extend requests to write to space beyond the
2476 * end of the image, so it's required that the image size is aligned.
2478 if ((cumulative_perms
& (BLK_PERM_WRITE
| BLK_PERM_WRITE_UNCHANGED
)) &&
2479 !(cumulative_perms
& BLK_PERM_RESIZE
))
2481 if ((bs
->total_sectors
* BDRV_SECTOR_SIZE
) % bs
->bl
.request_alignment
) {
2482 error_setg(errp
, "Cannot get 'write' permission without 'resize': "
2483 "Image size is not a multiple of request "
2489 /* Check this node */
2494 ret
= bdrv_drv_set_perm(bs
, cumulative_perms
, cumulative_shared_perms
, tran
,
2500 /* Drivers that never have children can omit .bdrv_child_perm() */
2501 if (!drv
->bdrv_child_perm
) {
2502 assert(QLIST_EMPTY(&bs
->children
));
2506 /* Check all children */
2507 QLIST_FOREACH(c
, &bs
->children
, next
) {
2508 uint64_t cur_perm
, cur_shared
;
2510 bdrv_child_perm(bs
, c
->bs
, c
, c
->role
, q
,
2511 cumulative_perms
, cumulative_shared_perms
,
2512 &cur_perm
, &cur_shared
);
2513 bdrv_child_set_perm(c
, cur_perm
, cur_shared
, tran
);
2520 * @list is a product of bdrv_topological_dfs() (may be called several times) -
2521 * a topologically sorted subgraph.
2523 static int bdrv_do_refresh_perms(GSList
*list
, BlockReopenQueue
*q
,
2524 Transaction
*tran
, Error
**errp
)
2527 BlockDriverState
*bs
;
2528 GLOBAL_STATE_CODE();
2530 for ( ; list
; list
= list
->next
) {
2533 if (bdrv_parent_perms_conflict(bs
, errp
)) {
2537 ret
= bdrv_node_refresh_perm(bs
, q
, tran
, errp
);
2547 * @list is any list of nodes. List is completed by all subtrees and
2548 * topologically sorted. It's not a problem if some node occurs in the @list
2551 static int bdrv_list_refresh_perms(GSList
*list
, BlockReopenQueue
*q
,
2552 Transaction
*tran
, Error
**errp
)
2554 g_autoptr(GHashTable
) found
= g_hash_table_new(NULL
, NULL
);
2555 g_autoptr(GSList
) refresh_list
= NULL
;
2557 for ( ; list
; list
= list
->next
) {
2558 refresh_list
= bdrv_topological_dfs(refresh_list
, found
, list
->data
);
2561 return bdrv_do_refresh_perms(refresh_list
, q
, tran
, errp
);
2564 void bdrv_get_cumulative_perm(BlockDriverState
*bs
, uint64_t *perm
,
2565 uint64_t *shared_perm
)
2568 uint64_t cumulative_perms
= 0;
2569 uint64_t cumulative_shared_perms
= BLK_PERM_ALL
;
2571 GLOBAL_STATE_CODE();
2573 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
2574 cumulative_perms
|= c
->perm
;
2575 cumulative_shared_perms
&= c
->shared_perm
;
2578 *perm
= cumulative_perms
;
2579 *shared_perm
= cumulative_shared_perms
;
2582 char *bdrv_perm_names(uint64_t perm
)
2588 { BLK_PERM_CONSISTENT_READ
, "consistent read" },
2589 { BLK_PERM_WRITE
, "write" },
2590 { BLK_PERM_WRITE_UNCHANGED
, "write unchanged" },
2591 { BLK_PERM_RESIZE
, "resize" },
2595 GString
*result
= g_string_sized_new(30);
2596 struct perm_name
*p
;
2598 for (p
= permissions
; p
->name
; p
++) {
2599 if (perm
& p
->perm
) {
2600 if (result
->len
> 0) {
2601 g_string_append(result
, ", ");
2603 g_string_append(result
, p
->name
);
2607 return g_string_free(result
, FALSE
);
2611 /* @tran is allowed to be NULL. In this case no rollback is possible */
2612 static int bdrv_refresh_perms(BlockDriverState
*bs
, Transaction
*tran
,
2616 Transaction
*local_tran
= NULL
;
2617 g_autoptr(GSList
) list
= bdrv_topological_dfs(NULL
, NULL
, bs
);
2618 GLOBAL_STATE_CODE();
2621 tran
= local_tran
= tran_new();
2624 ret
= bdrv_do_refresh_perms(list
, NULL
, tran
, errp
);
2627 tran_finalize(local_tran
, ret
);
2633 int bdrv_child_try_set_perm(BdrvChild
*c
, uint64_t perm
, uint64_t shared
,
2636 Error
*local_err
= NULL
;
2637 Transaction
*tran
= tran_new();
2640 GLOBAL_STATE_CODE();
2642 bdrv_child_set_perm(c
, perm
, shared
, tran
);
2644 ret
= bdrv_refresh_perms(c
->bs
, tran
, &local_err
);
2646 tran_finalize(tran
, ret
);
2649 if ((perm
& ~c
->perm
) || (c
->shared_perm
& ~shared
)) {
2650 /* tighten permissions */
2651 error_propagate(errp
, local_err
);
2654 * Our caller may intend to only loosen restrictions and
2655 * does not expect this function to fail. Errors are not
2656 * fatal in such a case, so we can just hide them from our
2659 error_free(local_err
);
2667 int bdrv_child_refresh_perms(BlockDriverState
*bs
, BdrvChild
*c
, Error
**errp
)
2669 uint64_t parent_perms
, parent_shared
;
2670 uint64_t perms
, shared
;
2672 GLOBAL_STATE_CODE();
2674 bdrv_get_cumulative_perm(bs
, &parent_perms
, &parent_shared
);
2675 bdrv_child_perm(bs
, c
->bs
, c
, c
->role
, NULL
,
2676 parent_perms
, parent_shared
, &perms
, &shared
);
2678 return bdrv_child_try_set_perm(c
, perms
, shared
, errp
);
2682 * Default implementation for .bdrv_child_perm() for block filters:
2683 * Forward CONSISTENT_READ, WRITE, WRITE_UNCHANGED, and RESIZE to the
2686 static void bdrv_filter_default_perms(BlockDriverState
*bs
, BdrvChild
*c
,
2688 BlockReopenQueue
*reopen_queue
,
2689 uint64_t perm
, uint64_t shared
,
2690 uint64_t *nperm
, uint64_t *nshared
)
2692 GLOBAL_STATE_CODE();
2693 *nperm
= perm
& DEFAULT_PERM_PASSTHROUGH
;
2694 *nshared
= (shared
& DEFAULT_PERM_PASSTHROUGH
) | DEFAULT_PERM_UNCHANGED
;
2697 static void bdrv_default_perms_for_cow(BlockDriverState
*bs
, BdrvChild
*c
,
2699 BlockReopenQueue
*reopen_queue
,
2700 uint64_t perm
, uint64_t shared
,
2701 uint64_t *nperm
, uint64_t *nshared
)
2703 assert(role
& BDRV_CHILD_COW
);
2704 GLOBAL_STATE_CODE();
2707 * We want consistent read from backing files if the parent needs it.
2708 * No other operations are performed on backing files.
2710 perm
&= BLK_PERM_CONSISTENT_READ
;
2713 * If the parent can deal with changing data, we're okay with a
2714 * writable and resizable backing file.
2715 * TODO Require !(perm & BLK_PERM_CONSISTENT_READ), too?
2717 if (shared
& BLK_PERM_WRITE
) {
2718 shared
= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2723 shared
|= BLK_PERM_CONSISTENT_READ
| BLK_PERM_WRITE_UNCHANGED
;
2725 if (bs
->open_flags
& BDRV_O_INACTIVE
) {
2726 shared
|= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2733 static void bdrv_default_perms_for_storage(BlockDriverState
*bs
, BdrvChild
*c
,
2735 BlockReopenQueue
*reopen_queue
,
2736 uint64_t perm
, uint64_t shared
,
2737 uint64_t *nperm
, uint64_t *nshared
)
2741 GLOBAL_STATE_CODE();
2742 assert(role
& (BDRV_CHILD_METADATA
| BDRV_CHILD_DATA
));
2744 flags
= bdrv_reopen_get_flags(reopen_queue
, bs
);
2747 * Apart from the modifications below, the same permissions are
2748 * forwarded and left alone as for filters
2750 bdrv_filter_default_perms(bs
, c
, role
, reopen_queue
,
2751 perm
, shared
, &perm
, &shared
);
2753 if (role
& BDRV_CHILD_METADATA
) {
2754 /* Format drivers may touch metadata even if the guest doesn't write */
2755 if (bdrv_is_writable_after_reopen(bs
, reopen_queue
)) {
2756 perm
|= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2760 * bs->file always needs to be consistent because of the
2761 * metadata. We can never allow other users to resize or write
2764 if (!(flags
& BDRV_O_NO_IO
)) {
2765 perm
|= BLK_PERM_CONSISTENT_READ
;
2767 shared
&= ~(BLK_PERM_WRITE
| BLK_PERM_RESIZE
);
2770 if (role
& BDRV_CHILD_DATA
) {
2772 * Technically, everything in this block is a subset of the
2773 * BDRV_CHILD_METADATA path taken above, and so this could
2774 * be an "else if" branch. However, that is not obvious, and
2775 * this function is not performance critical, therefore we let
2776 * this be an independent "if".
2780 * We cannot allow other users to resize the file because the
2781 * format driver might have some assumptions about the size
2782 * (e.g. because it is stored in metadata, or because the file
2783 * is split into fixed-size data files).
2785 shared
&= ~BLK_PERM_RESIZE
;
2788 * WRITE_UNCHANGED often cannot be performed as such on the
2789 * data file. For example, the qcow2 driver may still need to
2790 * write copied clusters on copy-on-read.
2792 if (perm
& BLK_PERM_WRITE_UNCHANGED
) {
2793 perm
|= BLK_PERM_WRITE
;
2797 * If the data file is written to, the format driver may
2798 * expect to be able to resize it by writing beyond the EOF.
2800 if (perm
& BLK_PERM_WRITE
) {
2801 perm
|= BLK_PERM_RESIZE
;
2805 if (bs
->open_flags
& BDRV_O_INACTIVE
) {
2806 shared
|= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2813 void bdrv_default_perms(BlockDriverState
*bs
, BdrvChild
*c
,
2814 BdrvChildRole role
, BlockReopenQueue
*reopen_queue
,
2815 uint64_t perm
, uint64_t shared
,
2816 uint64_t *nperm
, uint64_t *nshared
)
2818 GLOBAL_STATE_CODE();
2819 if (role
& BDRV_CHILD_FILTERED
) {
2820 assert(!(role
& (BDRV_CHILD_DATA
| BDRV_CHILD_METADATA
|
2822 bdrv_filter_default_perms(bs
, c
, role
, reopen_queue
,
2823 perm
, shared
, nperm
, nshared
);
2824 } else if (role
& BDRV_CHILD_COW
) {
2825 assert(!(role
& (BDRV_CHILD_DATA
| BDRV_CHILD_METADATA
)));
2826 bdrv_default_perms_for_cow(bs
, c
, role
, reopen_queue
,
2827 perm
, shared
, nperm
, nshared
);
2828 } else if (role
& (BDRV_CHILD_METADATA
| BDRV_CHILD_DATA
)) {
2829 bdrv_default_perms_for_storage(bs
, c
, role
, reopen_queue
,
2830 perm
, shared
, nperm
, nshared
);
2832 g_assert_not_reached();
2836 uint64_t bdrv_qapi_perm_to_blk_perm(BlockPermission qapi_perm
)
2838 static const uint64_t permissions
[] = {
2839 [BLOCK_PERMISSION_CONSISTENT_READ
] = BLK_PERM_CONSISTENT_READ
,
2840 [BLOCK_PERMISSION_WRITE
] = BLK_PERM_WRITE
,
2841 [BLOCK_PERMISSION_WRITE_UNCHANGED
] = BLK_PERM_WRITE_UNCHANGED
,
2842 [BLOCK_PERMISSION_RESIZE
] = BLK_PERM_RESIZE
,
2845 QEMU_BUILD_BUG_ON(ARRAY_SIZE(permissions
) != BLOCK_PERMISSION__MAX
);
2846 QEMU_BUILD_BUG_ON(1UL << ARRAY_SIZE(permissions
) != BLK_PERM_ALL
+ 1);
2848 assert(qapi_perm
< BLOCK_PERMISSION__MAX
);
2850 return permissions
[qapi_perm
];
2853 static void bdrv_replace_child_noperm(BdrvChild
*child
,
2854 BlockDriverState
*new_bs
)
2856 BlockDriverState
*old_bs
= child
->bs
;
2857 int new_bs_quiesce_counter
;
2859 assert(!child
->frozen
);
2860 assert(old_bs
!= new_bs
);
2861 GLOBAL_STATE_CODE();
2863 if (old_bs
&& new_bs
) {
2864 assert(bdrv_get_aio_context(old_bs
) == bdrv_get_aio_context(new_bs
));
2868 * If the new child node is drained but the old one was not, flush
2869 * all outstanding requests to the old child node.
2871 new_bs_quiesce_counter
= (new_bs
? new_bs
->quiesce_counter
: 0);
2872 if (new_bs_quiesce_counter
&& !child
->quiesced_parent
) {
2873 bdrv_parent_drained_begin_single(child
, true);
2877 if (child
->klass
->detach
) {
2878 child
->klass
->detach(child
);
2880 assert_bdrv_graph_writable(old_bs
);
2881 QLIST_REMOVE(child
, next_parent
);
2887 assert_bdrv_graph_writable(new_bs
);
2888 QLIST_INSERT_HEAD(&new_bs
->parents
, child
, next_parent
);
2889 if (child
->klass
->attach
) {
2890 child
->klass
->attach(child
);
2895 * If the old child node was drained but the new one is not, allow
2896 * requests to come in only after the new node has been attached.
2898 * Update new_bs_quiesce_counter because bdrv_parent_drained_begin_single()
2899 * polls, which could have changed the value.
2901 new_bs_quiesce_counter
= (new_bs
? new_bs
->quiesce_counter
: 0);
2902 if (!new_bs_quiesce_counter
&& child
->quiesced_parent
) {
2903 bdrv_parent_drained_end_single(child
);
2908 * Free the given @child.
2910 * The child must be empty (i.e. `child->bs == NULL`) and it must be
2911 * unused (i.e. not in a children list).
2913 static void bdrv_child_free(BdrvChild
*child
)
2916 GLOBAL_STATE_CODE();
2917 assert(!child
->next
.le_prev
); /* not in children list */
2919 g_free(child
->name
);
2923 typedef struct BdrvAttachChildCommonState
{
2925 AioContext
*old_parent_ctx
;
2926 AioContext
*old_child_ctx
;
2927 } BdrvAttachChildCommonState
;
2929 static void bdrv_attach_child_common_abort(void *opaque
)
2931 BdrvAttachChildCommonState
*s
= opaque
;
2932 BlockDriverState
*bs
= s
->child
->bs
;
2934 GLOBAL_STATE_CODE();
2935 bdrv_replace_child_noperm(s
->child
, NULL
);
2937 if (bdrv_get_aio_context(bs
) != s
->old_child_ctx
) {
2938 bdrv_try_change_aio_context(bs
, s
->old_child_ctx
, NULL
, &error_abort
);
2941 if (bdrv_child_get_parent_aio_context(s
->child
) != s
->old_parent_ctx
) {
2943 GHashTable
*visited
;
2948 /* No need to visit `child`, because it has been detached already */
2949 visited
= g_hash_table_new(NULL
, NULL
);
2950 ret
= s
->child
->klass
->change_aio_ctx(s
->child
, s
->old_parent_ctx
,
2951 visited
, tran
, &error_abort
);
2952 g_hash_table_destroy(visited
);
2954 /* transaction is supposed to always succeed */
2955 assert(ret
== true);
2960 bdrv_child_free(s
->child
);
2963 static TransactionActionDrv bdrv_attach_child_common_drv
= {
2964 .abort
= bdrv_attach_child_common_abort
,
2969 * Common part of attaching bdrv child to bs or to blk or to job
2971 * Function doesn't update permissions, caller is responsible for this.
2973 * Returns new created child.
2975 static BdrvChild
*bdrv_attach_child_common(BlockDriverState
*child_bs
,
2976 const char *child_name
,
2977 const BdrvChildClass
*child_class
,
2978 BdrvChildRole child_role
,
2979 uint64_t perm
, uint64_t shared_perm
,
2981 Transaction
*tran
, Error
**errp
)
2983 BdrvChild
*new_child
;
2984 AioContext
*parent_ctx
;
2985 AioContext
*child_ctx
= bdrv_get_aio_context(child_bs
);
2987 assert(child_class
->get_parent_desc
);
2988 GLOBAL_STATE_CODE();
2990 new_child
= g_new(BdrvChild
, 1);
2991 *new_child
= (BdrvChild
) {
2993 .name
= g_strdup(child_name
),
2994 .klass
= child_class
,
2997 .shared_perm
= shared_perm
,
3002 * If the AioContexts don't match, first try to move the subtree of
3003 * child_bs into the AioContext of the new parent. If this doesn't work,
3004 * try moving the parent into the AioContext of child_bs instead.
3006 parent_ctx
= bdrv_child_get_parent_aio_context(new_child
);
3007 if (child_ctx
!= parent_ctx
) {
3008 Error
*local_err
= NULL
;
3009 int ret
= bdrv_try_change_aio_context(child_bs
, parent_ctx
, NULL
,
3012 if (ret
< 0 && child_class
->change_aio_ctx
) {
3013 Transaction
*tran
= tran_new();
3014 GHashTable
*visited
= g_hash_table_new(NULL
, NULL
);
3017 g_hash_table_add(visited
, new_child
);
3018 ret_child
= child_class
->change_aio_ctx(new_child
, child_ctx
,
3019 visited
, tran
, NULL
);
3020 if (ret_child
== true) {
3021 error_free(local_err
);
3024 tran_finalize(tran
, ret_child
== true ? 0 : -1);
3025 g_hash_table_destroy(visited
);
3029 error_propagate(errp
, local_err
);
3030 bdrv_child_free(new_child
);
3036 bdrv_replace_child_noperm(new_child
, child_bs
);
3038 BdrvAttachChildCommonState
*s
= g_new(BdrvAttachChildCommonState
, 1);
3039 *s
= (BdrvAttachChildCommonState
) {
3041 .old_parent_ctx
= parent_ctx
,
3042 .old_child_ctx
= child_ctx
,
3044 tran_add(tran
, &bdrv_attach_child_common_drv
, s
);
3050 * Function doesn't update permissions, caller is responsible for this.
3052 static BdrvChild
*bdrv_attach_child_noperm(BlockDriverState
*parent_bs
,
3053 BlockDriverState
*child_bs
,
3054 const char *child_name
,
3055 const BdrvChildClass
*child_class
,
3056 BdrvChildRole child_role
,
3060 uint64_t perm
, shared_perm
;
3062 assert(parent_bs
->drv
);
3063 GLOBAL_STATE_CODE();
3065 if (bdrv_recurse_has_child(child_bs
, parent_bs
)) {
3066 error_setg(errp
, "Making '%s' a %s child of '%s' would create a cycle",
3067 child_bs
->node_name
, child_name
, parent_bs
->node_name
);
3071 bdrv_get_cumulative_perm(parent_bs
, &perm
, &shared_perm
);
3072 bdrv_child_perm(parent_bs
, child_bs
, NULL
, child_role
, NULL
,
3073 perm
, shared_perm
, &perm
, &shared_perm
);
3075 return bdrv_attach_child_common(child_bs
, child_name
, child_class
,
3076 child_role
, perm
, shared_perm
, parent_bs
,
3081 * This function steals the reference to child_bs from the caller.
3082 * That reference is later dropped by bdrv_root_unref_child().
3084 * On failure NULL is returned, errp is set and the reference to
3085 * child_bs is also dropped.
3087 * The caller must hold the AioContext lock @child_bs, but not that of @ctx
3088 * (unless @child_bs is already in @ctx).
3090 BdrvChild
*bdrv_root_attach_child(BlockDriverState
*child_bs
,
3091 const char *child_name
,
3092 const BdrvChildClass
*child_class
,
3093 BdrvChildRole child_role
,
3094 uint64_t perm
, uint64_t shared_perm
,
3095 void *opaque
, Error
**errp
)
3099 Transaction
*tran
= tran_new();
3101 GLOBAL_STATE_CODE();
3103 child
= bdrv_attach_child_common(child_bs
, child_name
, child_class
,
3104 child_role
, perm
, shared_perm
, opaque
,
3111 ret
= bdrv_refresh_perms(child_bs
, tran
, errp
);
3114 tran_finalize(tran
, ret
);
3116 bdrv_unref(child_bs
);
3118 return ret
< 0 ? NULL
: child
;
3122 * This function transfers the reference to child_bs from the caller
3123 * to parent_bs. That reference is later dropped by parent_bs on
3124 * bdrv_close() or if someone calls bdrv_unref_child().
3126 * On failure NULL is returned, errp is set and the reference to
3127 * child_bs is also dropped.
3129 * If @parent_bs and @child_bs are in different AioContexts, the caller must
3130 * hold the AioContext lock for @child_bs, but not for @parent_bs.
3132 BdrvChild
*bdrv_attach_child(BlockDriverState
*parent_bs
,
3133 BlockDriverState
*child_bs
,
3134 const char *child_name
,
3135 const BdrvChildClass
*child_class
,
3136 BdrvChildRole child_role
,
3141 Transaction
*tran
= tran_new();
3143 GLOBAL_STATE_CODE();
3145 child
= bdrv_attach_child_noperm(parent_bs
, child_bs
, child_name
,
3146 child_class
, child_role
, tran
, errp
);
3152 ret
= bdrv_refresh_perms(parent_bs
, tran
, errp
);
3158 tran_finalize(tran
, ret
);
3160 bdrv_unref(child_bs
);
3162 return ret
< 0 ? NULL
: child
;
3165 /* Callers must ensure that child->frozen is false. */
3166 void bdrv_root_unref_child(BdrvChild
*child
)
3168 BlockDriverState
*child_bs
= child
->bs
;
3170 GLOBAL_STATE_CODE();
3171 bdrv_replace_child_noperm(child
, NULL
);
3172 bdrv_child_free(child
);
3176 * Update permissions for old node. We're just taking a parent away, so
3177 * we're loosening restrictions. Errors of permission update are not
3178 * fatal in this case, ignore them.
3180 bdrv_refresh_perms(child_bs
, NULL
, NULL
);
3183 * When the parent requiring a non-default AioContext is removed, the
3184 * node moves back to the main AioContext
3186 bdrv_try_change_aio_context(child_bs
, qemu_get_aio_context(), NULL
,
3190 bdrv_unref(child_bs
);
3193 typedef struct BdrvSetInheritsFrom
{
3194 BlockDriverState
*bs
;
3195 BlockDriverState
*old_inherits_from
;
3196 } BdrvSetInheritsFrom
;
3198 static void bdrv_set_inherits_from_abort(void *opaque
)
3200 BdrvSetInheritsFrom
*s
= opaque
;
3202 s
->bs
->inherits_from
= s
->old_inherits_from
;
3205 static TransactionActionDrv bdrv_set_inherits_from_drv
= {
3206 .abort
= bdrv_set_inherits_from_abort
,
3210 /* @tran is allowed to be NULL. In this case no rollback is possible */
3211 static void bdrv_set_inherits_from(BlockDriverState
*bs
,
3212 BlockDriverState
*new_inherits_from
,
3216 BdrvSetInheritsFrom
*s
= g_new(BdrvSetInheritsFrom
, 1);
3218 *s
= (BdrvSetInheritsFrom
) {
3220 .old_inherits_from
= bs
->inherits_from
,
3223 tran_add(tran
, &bdrv_set_inherits_from_drv
, s
);
3226 bs
->inherits_from
= new_inherits_from
;
3230 * Clear all inherits_from pointers from children and grandchildren of
3231 * @root that point to @root, where necessary.
3232 * @tran is allowed to be NULL. In this case no rollback is possible
3234 static void bdrv_unset_inherits_from(BlockDriverState
*root
, BdrvChild
*child
,
3239 if (child
->bs
->inherits_from
== root
) {
3241 * Remove inherits_from only when the last reference between root and
3242 * child->bs goes away.
3244 QLIST_FOREACH(c
, &root
->children
, next
) {
3245 if (c
!= child
&& c
->bs
== child
->bs
) {
3250 bdrv_set_inherits_from(child
->bs
, NULL
, tran
);
3254 QLIST_FOREACH(c
, &child
->bs
->children
, next
) {
3255 bdrv_unset_inherits_from(root
, c
, tran
);
3259 /* Callers must ensure that child->frozen is false. */
3260 void bdrv_unref_child(BlockDriverState
*parent
, BdrvChild
*child
)
3262 GLOBAL_STATE_CODE();
3263 if (child
== NULL
) {
3267 bdrv_unset_inherits_from(parent
, child
, NULL
);
3268 bdrv_root_unref_child(child
);
3272 static void bdrv_parent_cb_change_media(BlockDriverState
*bs
, bool load
)
3275 GLOBAL_STATE_CODE();
3276 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
3277 if (c
->klass
->change_media
) {
3278 c
->klass
->change_media(c
, load
);
3283 /* Return true if you can reach parent going through child->inherits_from
3284 * recursively. If parent or child are NULL, return false */
3285 static bool bdrv_inherits_from_recursive(BlockDriverState
*child
,
3286 BlockDriverState
*parent
)
3288 while (child
&& child
!= parent
) {
3289 child
= child
->inherits_from
;
3292 return child
!= NULL
;
3296 * Return the BdrvChildRole for @bs's backing child. bs->backing is
3297 * mostly used for COW backing children (role = COW), but also for
3298 * filtered children (role = FILTERED | PRIMARY).
3300 static BdrvChildRole
bdrv_backing_role(BlockDriverState
*bs
)
3302 if (bs
->drv
&& bs
->drv
->is_filter
) {
3303 return BDRV_CHILD_FILTERED
| BDRV_CHILD_PRIMARY
;
3305 return BDRV_CHILD_COW
;
3310 * Sets the bs->backing or bs->file link of a BDS. A new reference is created;
3311 * callers which don't need their own reference any more must call bdrv_unref().
3313 * Function doesn't update permissions, caller is responsible for this.
3315 static int bdrv_set_file_or_backing_noperm(BlockDriverState
*parent_bs
,
3316 BlockDriverState
*child_bs
,
3318 Transaction
*tran
, Error
**errp
)
3320 bool update_inherits_from
=
3321 bdrv_inherits_from_recursive(child_bs
, parent_bs
);
3322 BdrvChild
*child
= is_backing
? parent_bs
->backing
: parent_bs
->file
;
3325 GLOBAL_STATE_CODE();
3327 if (!parent_bs
->drv
) {
3329 * Node without drv is an object without a class :/. TODO: finally fix
3330 * qcow2 driver to never clear bs->drv and implement format corruption
3331 * handling in other way.
3333 error_setg(errp
, "Node corrupted");
3337 if (child
&& child
->frozen
) {
3338 error_setg(errp
, "Cannot change frozen '%s' link from '%s' to '%s'",
3339 child
->name
, parent_bs
->node_name
, child
->bs
->node_name
);
3343 if (is_backing
&& !parent_bs
->drv
->is_filter
&&
3344 !parent_bs
->drv
->supports_backing
)
3346 error_setg(errp
, "Driver '%s' of node '%s' does not support backing "
3347 "files", parent_bs
->drv
->format_name
, parent_bs
->node_name
);
3351 if (parent_bs
->drv
->is_filter
) {
3352 role
= BDRV_CHILD_FILTERED
| BDRV_CHILD_PRIMARY
;
3353 } else if (is_backing
) {
3354 role
= BDRV_CHILD_COW
;
3357 * We only can use same role as it is in existing child. We don't have
3358 * infrastructure to determine role of file child in generic way
3361 error_setg(errp
, "Cannot set file child to format node without "
3369 bdrv_unset_inherits_from(parent_bs
, child
, tran
);
3370 bdrv_remove_child(child
, tran
);
3377 child
= bdrv_attach_child_noperm(parent_bs
, child_bs
,
3378 is_backing
? "backing" : "file",
3379 &child_of_bds
, role
,
3387 * If inherits_from pointed recursively to bs then let's update it to
3388 * point directly to bs (else it will become NULL).
3390 if (update_inherits_from
) {
3391 bdrv_set_inherits_from(child_bs
, parent_bs
, tran
);
3395 bdrv_refresh_limits(parent_bs
, tran
, NULL
);
3400 static int bdrv_set_backing_noperm(BlockDriverState
*bs
,
3401 BlockDriverState
*backing_hd
,
3402 Transaction
*tran
, Error
**errp
)
3404 GLOBAL_STATE_CODE();
3405 return bdrv_set_file_or_backing_noperm(bs
, backing_hd
, true, tran
, errp
);
3408 int bdrv_set_backing_hd_drained(BlockDriverState
*bs
,
3409 BlockDriverState
*backing_hd
,
3413 Transaction
*tran
= tran_new();
3415 GLOBAL_STATE_CODE();
3416 assert(bs
->quiesce_counter
> 0);
3418 ret
= bdrv_set_backing_noperm(bs
, backing_hd
, tran
, errp
);
3423 ret
= bdrv_refresh_perms(bs
, tran
, errp
);
3425 tran_finalize(tran
, ret
);
3429 int bdrv_set_backing_hd(BlockDriverState
*bs
, BlockDriverState
*backing_hd
,
3433 GLOBAL_STATE_CODE();
3435 bdrv_drained_begin(bs
);
3436 ret
= bdrv_set_backing_hd_drained(bs
, backing_hd
, errp
);
3437 bdrv_drained_end(bs
);
3443 * Opens the backing file for a BlockDriverState if not yet open
3445 * bdref_key specifies the key for the image's BlockdevRef in the options QDict.
3446 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
3447 * itself, all options starting with "${bdref_key}." are considered part of the
3450 * TODO Can this be unified with bdrv_open_image()?
3452 int bdrv_open_backing_file(BlockDriverState
*bs
, QDict
*parent_options
,
3453 const char *bdref_key
, Error
**errp
)
3455 char *backing_filename
= NULL
;
3456 char *bdref_key_dot
;
3457 const char *reference
= NULL
;
3459 bool implicit_backing
= false;
3460 BlockDriverState
*backing_hd
;
3462 QDict
*tmp_parent_options
= NULL
;
3463 Error
*local_err
= NULL
;
3465 GLOBAL_STATE_CODE();
3467 if (bs
->backing
!= NULL
) {
3471 /* NULL means an empty set of options */
3472 if (parent_options
== NULL
) {
3473 tmp_parent_options
= qdict_new();
3474 parent_options
= tmp_parent_options
;
3477 bs
->open_flags
&= ~BDRV_O_NO_BACKING
;
3479 bdref_key_dot
= g_strdup_printf("%s.", bdref_key
);
3480 qdict_extract_subqdict(parent_options
, &options
, bdref_key_dot
);
3481 g_free(bdref_key_dot
);
3484 * Caution: while qdict_get_try_str() is fine, getting non-string
3485 * types would require more care. When @parent_options come from
3486 * -blockdev or blockdev_add, its members are typed according to
3487 * the QAPI schema, but when they come from -drive, they're all
3490 reference
= qdict_get_try_str(parent_options
, bdref_key
);
3491 if (reference
|| qdict_haskey(options
, "file.filename")) {
3492 /* keep backing_filename NULL */
3493 } else if (bs
->backing_file
[0] == '\0' && qdict_size(options
) == 0) {
3494 qobject_unref(options
);
3497 if (qdict_size(options
) == 0) {
3498 /* If the user specifies options that do not modify the
3499 * backing file's behavior, we might still consider it the
3500 * implicit backing file. But it's easier this way, and
3501 * just specifying some of the backing BDS's options is
3502 * only possible with -drive anyway (otherwise the QAPI
3503 * schema forces the user to specify everything). */
3504 implicit_backing
= !strcmp(bs
->auto_backing_file
, bs
->backing_file
);
3507 backing_filename
= bdrv_get_full_backing_filename(bs
, &local_err
);
3510 error_propagate(errp
, local_err
);
3511 qobject_unref(options
);
3516 if (!bs
->drv
|| !bs
->drv
->supports_backing
) {
3518 error_setg(errp
, "Driver doesn't support backing files");
3519 qobject_unref(options
);
3524 bs
->backing_format
[0] != '\0' && !qdict_haskey(options
, "driver")) {
3525 qdict_put_str(options
, "driver", bs
->backing_format
);
3528 backing_hd
= bdrv_open_inherit(backing_filename
, reference
, options
, 0, bs
,
3529 &child_of_bds
, bdrv_backing_role(bs
), errp
);
3531 bs
->open_flags
|= BDRV_O_NO_BACKING
;
3532 error_prepend(errp
, "Could not open backing file: ");
3537 if (implicit_backing
) {
3538 bdrv_refresh_filename(backing_hd
);
3539 pstrcpy(bs
->auto_backing_file
, sizeof(bs
->auto_backing_file
),
3540 backing_hd
->filename
);
3543 /* Hook up the backing file link; drop our reference, bs owns the
3544 * backing_hd reference now */
3545 ret
= bdrv_set_backing_hd(bs
, backing_hd
, errp
);
3546 bdrv_unref(backing_hd
);
3551 qdict_del(parent_options
, bdref_key
);
3554 g_free(backing_filename
);
3555 qobject_unref(tmp_parent_options
);
3559 static BlockDriverState
*
3560 bdrv_open_child_bs(const char *filename
, QDict
*options
, const char *bdref_key
,
3561 BlockDriverState
*parent
, const BdrvChildClass
*child_class
,
3562 BdrvChildRole child_role
, bool allow_none
, Error
**errp
)
3564 BlockDriverState
*bs
= NULL
;
3565 QDict
*image_options
;
3566 char *bdref_key_dot
;
3567 const char *reference
;
3569 assert(child_class
!= NULL
);
3571 bdref_key_dot
= g_strdup_printf("%s.", bdref_key
);
3572 qdict_extract_subqdict(options
, &image_options
, bdref_key_dot
);
3573 g_free(bdref_key_dot
);
3576 * Caution: while qdict_get_try_str() is fine, getting non-string
3577 * types would require more care. When @options come from
3578 * -blockdev or blockdev_add, its members are typed according to
3579 * the QAPI schema, but when they come from -drive, they're all
3582 reference
= qdict_get_try_str(options
, bdref_key
);
3583 if (!filename
&& !reference
&& !qdict_size(image_options
)) {
3585 error_setg(errp
, "A block device must be specified for \"%s\"",
3588 qobject_unref(image_options
);
3592 bs
= bdrv_open_inherit(filename
, reference
, image_options
, 0,
3593 parent
, child_class
, child_role
, errp
);
3599 qdict_del(options
, bdref_key
);
3604 * Opens a disk image whose options are given as BlockdevRef in another block
3607 * If allow_none is true, no image will be opened if filename is false and no
3608 * BlockdevRef is given. NULL will be returned, but errp remains unset.
3610 * bdrev_key specifies the key for the image's BlockdevRef in the options QDict.
3611 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
3612 * itself, all options starting with "${bdref_key}." are considered part of the
3615 * The BlockdevRef will be removed from the options QDict.
3617 BdrvChild
*bdrv_open_child(const char *filename
,
3618 QDict
*options
, const char *bdref_key
,
3619 BlockDriverState
*parent
,
3620 const BdrvChildClass
*child_class
,
3621 BdrvChildRole child_role
,
3622 bool allow_none
, Error
**errp
)
3624 BlockDriverState
*bs
;
3626 GLOBAL_STATE_CODE();
3628 bs
= bdrv_open_child_bs(filename
, options
, bdref_key
, parent
, child_class
,
3629 child_role
, allow_none
, errp
);
3634 return bdrv_attach_child(parent
, bs
, bdref_key
, child_class
, child_role
,
3639 * Wrapper on bdrv_open_child() for most popular case: open primary child of bs.
3641 int bdrv_open_file_child(const char *filename
,
3642 QDict
*options
, const char *bdref_key
,
3643 BlockDriverState
*parent
, Error
**errp
)
3647 /* commit_top and mirror_top don't use this function */
3648 assert(!parent
->drv
->filtered_child_is_backing
);
3649 role
= parent
->drv
->is_filter
?
3650 (BDRV_CHILD_FILTERED
| BDRV_CHILD_PRIMARY
) : BDRV_CHILD_IMAGE
;
3652 if (!bdrv_open_child(filename
, options
, bdref_key
, parent
,
3653 &child_of_bds
, role
, false, errp
))
3662 * TODO Future callers may need to specify parent/child_class in order for
3663 * option inheritance to work. Existing callers use it for the root node.
3665 BlockDriverState
*bdrv_open_blockdev_ref(BlockdevRef
*ref
, Error
**errp
)
3667 BlockDriverState
*bs
= NULL
;
3668 QObject
*obj
= NULL
;
3669 QDict
*qdict
= NULL
;
3670 const char *reference
= NULL
;
3673 GLOBAL_STATE_CODE();
3675 if (ref
->type
== QTYPE_QSTRING
) {
3676 reference
= ref
->u
.reference
;
3678 BlockdevOptions
*options
= &ref
->u
.definition
;
3679 assert(ref
->type
== QTYPE_QDICT
);
3681 v
= qobject_output_visitor_new(&obj
);
3682 visit_type_BlockdevOptions(v
, NULL
, &options
, &error_abort
);
3683 visit_complete(v
, &obj
);
3685 qdict
= qobject_to(QDict
, obj
);
3686 qdict_flatten(qdict
);
3688 /* bdrv_open_inherit() defaults to the values in bdrv_flags (for
3689 * compatibility with other callers) rather than what we want as the
3690 * real defaults. Apply the defaults here instead. */
3691 qdict_set_default_str(qdict
, BDRV_OPT_CACHE_DIRECT
, "off");
3692 qdict_set_default_str(qdict
, BDRV_OPT_CACHE_NO_FLUSH
, "off");
3693 qdict_set_default_str(qdict
, BDRV_OPT_READ_ONLY
, "off");
3694 qdict_set_default_str(qdict
, BDRV_OPT_AUTO_READ_ONLY
, "off");
3698 bs
= bdrv_open_inherit(NULL
, reference
, qdict
, 0, NULL
, NULL
, 0, errp
);
3705 static BlockDriverState
*bdrv_append_temp_snapshot(BlockDriverState
*bs
,
3707 QDict
*snapshot_options
,
3710 g_autofree
char *tmp_filename
= NULL
;
3712 QemuOpts
*opts
= NULL
;
3713 BlockDriverState
*bs_snapshot
= NULL
;
3716 GLOBAL_STATE_CODE();
3718 /* if snapshot, we create a temporary backing file and open it
3719 instead of opening 'filename' directly */
3721 /* Get the required size from the image */
3722 total_size
= bdrv_getlength(bs
);
3723 if (total_size
< 0) {
3724 error_setg_errno(errp
, -total_size
, "Could not get image size");
3728 /* Create the temporary image */
3729 tmp_filename
= create_tmp_file(errp
);
3730 if (!tmp_filename
) {
3734 opts
= qemu_opts_create(bdrv_qcow2
.create_opts
, NULL
, 0,
3736 qemu_opt_set_number(opts
, BLOCK_OPT_SIZE
, total_size
, &error_abort
);
3737 ret
= bdrv_create(&bdrv_qcow2
, tmp_filename
, opts
, errp
);
3738 qemu_opts_del(opts
);
3740 error_prepend(errp
, "Could not create temporary overlay '%s': ",
3745 /* Prepare options QDict for the temporary file */
3746 qdict_put_str(snapshot_options
, "file.driver", "file");
3747 qdict_put_str(snapshot_options
, "file.filename", tmp_filename
);
3748 qdict_put_str(snapshot_options
, "driver", "qcow2");
3750 bs_snapshot
= bdrv_open(NULL
, NULL
, snapshot_options
, flags
, errp
);
3751 snapshot_options
= NULL
;
3756 ret
= bdrv_append(bs_snapshot
, bs
, errp
);
3763 qobject_unref(snapshot_options
);
3768 * Opens a disk image (raw, qcow2, vmdk, ...)
3770 * options is a QDict of options to pass to the block drivers, or NULL for an
3771 * empty set of options. The reference to the QDict belongs to the block layer
3772 * after the call (even on failure), so if the caller intends to reuse the
3773 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
3775 * If *pbs is NULL, a new BDS will be created with a pointer to it stored there.
3776 * If it is not NULL, the referenced BDS will be reused.
3778 * The reference parameter may be used to specify an existing block device which
3779 * should be opened. If specified, neither options nor a filename may be given,
3780 * nor can an existing BDS be reused (that is, *pbs has to be NULL).
3782 static BlockDriverState
*bdrv_open_inherit(const char *filename
,
3783 const char *reference
,
3784 QDict
*options
, int flags
,
3785 BlockDriverState
*parent
,
3786 const BdrvChildClass
*child_class
,
3787 BdrvChildRole child_role
,
3791 BlockBackend
*file
= NULL
;
3792 BlockDriverState
*bs
;
3793 BlockDriver
*drv
= NULL
;
3795 const char *drvname
;
3796 const char *backing
;
3797 Error
*local_err
= NULL
;
3798 QDict
*snapshot_options
= NULL
;
3799 int snapshot_flags
= 0;
3801 assert(!child_class
|| !flags
);
3802 assert(!child_class
== !parent
);
3803 GLOBAL_STATE_CODE();
3806 bool options_non_empty
= options
? qdict_size(options
) : false;
3807 qobject_unref(options
);
3809 if (filename
|| options_non_empty
) {
3810 error_setg(errp
, "Cannot reference an existing block device with "
3811 "additional options or a new filename");
3815 bs
= bdrv_lookup_bs(reference
, reference
, errp
);
3826 /* NULL means an empty set of options */
3827 if (options
== NULL
) {
3828 options
= qdict_new();
3831 /* json: syntax counts as explicit options, as if in the QDict */
3832 parse_json_protocol(options
, &filename
, &local_err
);
3837 bs
->explicit_options
= qdict_clone_shallow(options
);
3840 bool parent_is_format
;
3843 parent_is_format
= parent
->drv
->is_format
;
3846 * parent->drv is not set yet because this node is opened for
3847 * (potential) format probing. That means that @parent is going
3848 * to be a format node.
3850 parent_is_format
= true;
3853 bs
->inherits_from
= parent
;
3854 child_class
->inherit_options(child_role
, parent_is_format
,
3856 parent
->open_flags
, parent
->options
);
3859 ret
= bdrv_fill_options(&options
, filename
, &flags
, &local_err
);
3865 * Set the BDRV_O_RDWR and BDRV_O_ALLOW_RDWR flags.
3866 * Caution: getting a boolean member of @options requires care.
3867 * When @options come from -blockdev or blockdev_add, members are
3868 * typed according to the QAPI schema, but when they come from
3869 * -drive, they're all QString.
3871 if (g_strcmp0(qdict_get_try_str(options
, BDRV_OPT_READ_ONLY
), "on") &&
3872 !qdict_get_try_bool(options
, BDRV_OPT_READ_ONLY
, false)) {
3873 flags
|= (BDRV_O_RDWR
| BDRV_O_ALLOW_RDWR
);
3875 flags
&= ~BDRV_O_RDWR
;
3878 if (flags
& BDRV_O_SNAPSHOT
) {
3879 snapshot_options
= qdict_new();
3880 bdrv_temp_snapshot_options(&snapshot_flags
, snapshot_options
,
3882 /* Let bdrv_backing_options() override "read-only" */
3883 qdict_del(options
, BDRV_OPT_READ_ONLY
);
3884 bdrv_inherited_options(BDRV_CHILD_COW
, true,
3885 &flags
, options
, flags
, options
);
3888 bs
->open_flags
= flags
;
3889 bs
->options
= options
;
3890 options
= qdict_clone_shallow(options
);
3892 /* Find the right image format driver */
3893 /* See cautionary note on accessing @options above */
3894 drvname
= qdict_get_try_str(options
, "driver");
3896 drv
= bdrv_find_format(drvname
);
3898 error_setg(errp
, "Unknown driver: '%s'", drvname
);
3903 assert(drvname
|| !(flags
& BDRV_O_PROTOCOL
));
3905 /* See cautionary note on accessing @options above */
3906 backing
= qdict_get_try_str(options
, "backing");
3907 if (qobject_to(QNull
, qdict_get(options
, "backing")) != NULL
||
3908 (backing
&& *backing
== '\0'))
3911 warn_report("Use of \"backing\": \"\" is deprecated; "
3912 "use \"backing\": null instead");
3914 flags
|= BDRV_O_NO_BACKING
;
3915 qdict_del(bs
->explicit_options
, "backing");
3916 qdict_del(bs
->options
, "backing");
3917 qdict_del(options
, "backing");
3920 /* Open image file without format layer. This BlockBackend is only used for
3921 * probing, the block drivers will do their own bdrv_open_child() for the
3922 * same BDS, which is why we put the node name back into options. */
3923 if ((flags
& BDRV_O_PROTOCOL
) == 0) {
3924 BlockDriverState
*file_bs
;
3926 file_bs
= bdrv_open_child_bs(filename
, options
, "file", bs
,
3927 &child_of_bds
, BDRV_CHILD_IMAGE
,
3932 if (file_bs
!= NULL
) {
3933 /* Not requesting BLK_PERM_CONSISTENT_READ because we're only
3934 * looking at the header to guess the image format. This works even
3935 * in cases where a guest would not see a consistent state. */
3936 file
= blk_new(bdrv_get_aio_context(file_bs
), 0, BLK_PERM_ALL
);
3937 blk_insert_bs(file
, file_bs
, &local_err
);
3938 bdrv_unref(file_bs
);
3943 qdict_put_str(options
, "file", bdrv_get_node_name(file_bs
));
3947 /* Image format probing */
3950 ret
= find_image_format(file
, filename
, &drv
, &local_err
);
3955 * This option update would logically belong in bdrv_fill_options(),
3956 * but we first need to open bs->file for the probing to work, while
3957 * opening bs->file already requires the (mostly) final set of options
3958 * so that cache mode etc. can be inherited.
3960 * Adding the driver later is somewhat ugly, but it's not an option
3961 * that would ever be inherited, so it's correct. We just need to make
3962 * sure to update both bs->options (which has the full effective
3963 * options for bs) and options (which has file.* already removed).
3965 qdict_put_str(bs
->options
, "driver", drv
->format_name
);
3966 qdict_put_str(options
, "driver", drv
->format_name
);
3968 error_setg(errp
, "Must specify either driver or file");
3972 /* BDRV_O_PROTOCOL must be set iff a protocol BDS is about to be created */
3973 assert(!!(flags
& BDRV_O_PROTOCOL
) == !!drv
->bdrv_file_open
);
3974 /* file must be NULL if a protocol BDS is about to be created
3975 * (the inverse results in an error message from bdrv_open_common()) */
3976 assert(!(flags
& BDRV_O_PROTOCOL
) || !file
);
3978 /* Open the image */
3979 ret
= bdrv_open_common(bs
, file
, options
, &local_err
);
3989 /* If there is a backing file, use it */
3990 if ((flags
& BDRV_O_NO_BACKING
) == 0) {
3991 ret
= bdrv_open_backing_file(bs
, options
, "backing", &local_err
);
3993 goto close_and_fail
;
3997 /* Remove all children options and references
3998 * from bs->options and bs->explicit_options */
3999 QLIST_FOREACH(child
, &bs
->children
, next
) {
4000 char *child_key_dot
;
4001 child_key_dot
= g_strdup_printf("%s.", child
->name
);
4002 qdict_extract_subqdict(bs
->explicit_options
, NULL
, child_key_dot
);
4003 qdict_extract_subqdict(bs
->options
, NULL
, child_key_dot
);
4004 qdict_del(bs
->explicit_options
, child
->name
);
4005 qdict_del(bs
->options
, child
->name
);
4006 g_free(child_key_dot
);
4009 /* Check if any unknown options were used */
4010 if (qdict_size(options
) != 0) {
4011 const QDictEntry
*entry
= qdict_first(options
);
4012 if (flags
& BDRV_O_PROTOCOL
) {
4013 error_setg(errp
, "Block protocol '%s' doesn't support the option "
4014 "'%s'", drv
->format_name
, entry
->key
);
4017 "Block format '%s' does not support the option '%s'",
4018 drv
->format_name
, entry
->key
);
4021 goto close_and_fail
;
4024 bdrv_parent_cb_change_media(bs
, true);
4026 qobject_unref(options
);
4029 /* For snapshot=on, create a temporary qcow2 overlay. bs points to the
4030 * temporary snapshot afterwards. */
4031 if (snapshot_flags
) {
4032 BlockDriverState
*snapshot_bs
;
4033 snapshot_bs
= bdrv_append_temp_snapshot(bs
, snapshot_flags
,
4034 snapshot_options
, &local_err
);
4035 snapshot_options
= NULL
;
4037 goto close_and_fail
;
4039 /* We are not going to return bs but the overlay on top of it
4040 * (snapshot_bs); thus, we have to drop the strong reference to bs
4041 * (which we obtained by calling bdrv_new()). bs will not be deleted,
4042 * though, because the overlay still has a reference to it. */
4051 qobject_unref(snapshot_options
);
4052 qobject_unref(bs
->explicit_options
);
4053 qobject_unref(bs
->options
);
4054 qobject_unref(options
);
4056 bs
->explicit_options
= NULL
;
4058 error_propagate(errp
, local_err
);
4063 qobject_unref(snapshot_options
);
4064 qobject_unref(options
);
4065 error_propagate(errp
, local_err
);
4069 BlockDriverState
*bdrv_open(const char *filename
, const char *reference
,
4070 QDict
*options
, int flags
, Error
**errp
)
4072 GLOBAL_STATE_CODE();
4074 return bdrv_open_inherit(filename
, reference
, options
, flags
, NULL
,
4078 /* Return true if the NULL-terminated @list contains @str */
4079 static bool is_str_in_list(const char *str
, const char *const *list
)
4083 for (i
= 0; list
[i
] != NULL
; i
++) {
4084 if (!strcmp(str
, list
[i
])) {
4093 * Check that every option set in @bs->options is also set in
4096 * Options listed in the common_options list and in
4097 * @bs->drv->mutable_opts are skipped.
4099 * Return 0 on success, otherwise return -EINVAL and set @errp.
4101 static int bdrv_reset_options_allowed(BlockDriverState
*bs
,
4102 const QDict
*new_opts
, Error
**errp
)
4104 const QDictEntry
*e
;
4105 /* These options are common to all block drivers and are handled
4106 * in bdrv_reopen_prepare() so they can be left out of @new_opts */
4107 const char *const common_options
[] = {
4108 "node-name", "discard", "cache.direct", "cache.no-flush",
4109 "read-only", "auto-read-only", "detect-zeroes", NULL
4112 for (e
= qdict_first(bs
->options
); e
; e
= qdict_next(bs
->options
, e
)) {
4113 if (!qdict_haskey(new_opts
, e
->key
) &&
4114 !is_str_in_list(e
->key
, common_options
) &&
4115 !is_str_in_list(e
->key
, bs
->drv
->mutable_opts
)) {
4116 error_setg(errp
, "Option '%s' cannot be reset "
4117 "to its default value", e
->key
);
4126 * Returns true if @child can be reached recursively from @bs
4128 static bool bdrv_recurse_has_child(BlockDriverState
*bs
,
4129 BlockDriverState
*child
)
4137 QLIST_FOREACH(c
, &bs
->children
, next
) {
4138 if (bdrv_recurse_has_child(c
->bs
, child
)) {
4147 * Adds a BlockDriverState to a simple queue for an atomic, transactional
4148 * reopen of multiple devices.
4150 * bs_queue can either be an existing BlockReopenQueue that has had QTAILQ_INIT
4151 * already performed, or alternatively may be NULL a new BlockReopenQueue will
4152 * be created and initialized. This newly created BlockReopenQueue should be
4153 * passed back in for subsequent calls that are intended to be of the same
4156 * bs is the BlockDriverState to add to the reopen queue.
4158 * options contains the changed options for the associated bs
4159 * (the BlockReopenQueue takes ownership)
4161 * flags contains the open flags for the associated bs
4163 * returns a pointer to bs_queue, which is either the newly allocated
4164 * bs_queue, or the existing bs_queue being used.
4166 * bs is drained here and undrained by bdrv_reopen_queue_free().
4168 * To be called with bs->aio_context locked.
4170 static BlockReopenQueue
*bdrv_reopen_queue_child(BlockReopenQueue
*bs_queue
,
4171 BlockDriverState
*bs
,
4173 const BdrvChildClass
*klass
,
4175 bool parent_is_format
,
4176 QDict
*parent_options
,
4182 BlockReopenQueueEntry
*bs_entry
;
4184 QDict
*old_options
, *explicit_options
, *options_copy
;
4188 GLOBAL_STATE_CODE();
4190 bdrv_drained_begin(bs
);
4192 if (bs_queue
== NULL
) {
4193 bs_queue
= g_new0(BlockReopenQueue
, 1);
4194 QTAILQ_INIT(bs_queue
);
4198 options
= qdict_new();
4201 /* Check if this BlockDriverState is already in the queue */
4202 QTAILQ_FOREACH(bs_entry
, bs_queue
, entry
) {
4203 if (bs
== bs_entry
->state
.bs
) {
4209 * Precedence of options:
4210 * 1. Explicitly passed in options (highest)
4211 * 2. Retained from explicitly set options of bs
4212 * 3. Inherited from parent node
4213 * 4. Retained from effective options of bs
4216 /* Old explicitly set values (don't overwrite by inherited value) */
4217 if (bs_entry
|| keep_old_opts
) {
4218 old_options
= qdict_clone_shallow(bs_entry
?
4219 bs_entry
->state
.explicit_options
:
4220 bs
->explicit_options
);
4221 bdrv_join_options(bs
, options
, old_options
);
4222 qobject_unref(old_options
);
4225 explicit_options
= qdict_clone_shallow(options
);
4227 /* Inherit from parent node */
4228 if (parent_options
) {
4230 klass
->inherit_options(role
, parent_is_format
, &flags
, options
,
4231 parent_flags
, parent_options
);
4233 flags
= bdrv_get_flags(bs
);
4236 if (keep_old_opts
) {
4237 /* Old values are used for options that aren't set yet */
4238 old_options
= qdict_clone_shallow(bs
->options
);
4239 bdrv_join_options(bs
, options
, old_options
);
4240 qobject_unref(old_options
);
4243 /* We have the final set of options so let's update the flags */
4244 options_copy
= qdict_clone_shallow(options
);
4245 opts
= qemu_opts_create(&bdrv_runtime_opts
, NULL
, 0, &error_abort
);
4246 qemu_opts_absorb_qdict(opts
, options_copy
, NULL
);
4247 update_flags_from_options(&flags
, opts
);
4248 qemu_opts_del(opts
);
4249 qobject_unref(options_copy
);
4251 /* bdrv_open_inherit() sets and clears some additional flags internally */
4252 flags
&= ~BDRV_O_PROTOCOL
;
4253 if (flags
& BDRV_O_RDWR
) {
4254 flags
|= BDRV_O_ALLOW_RDWR
;
4258 bs_entry
= g_new0(BlockReopenQueueEntry
, 1);
4259 QTAILQ_INSERT_TAIL(bs_queue
, bs_entry
, entry
);
4261 qobject_unref(bs_entry
->state
.options
);
4262 qobject_unref(bs_entry
->state
.explicit_options
);
4265 bs_entry
->state
.bs
= bs
;
4266 bs_entry
->state
.options
= options
;
4267 bs_entry
->state
.explicit_options
= explicit_options
;
4268 bs_entry
->state
.flags
= flags
;
4271 * If keep_old_opts is false then it means that unspecified
4272 * options must be reset to their original value. We don't allow
4273 * resetting 'backing' but we need to know if the option is
4274 * missing in order to decide if we have to return an error.
4276 if (!keep_old_opts
) {
4277 bs_entry
->state
.backing_missing
=
4278 !qdict_haskey(options
, "backing") &&
4279 !qdict_haskey(options
, "backing.driver");
4282 QLIST_FOREACH(child
, &bs
->children
, next
) {
4283 QDict
*new_child_options
= NULL
;
4284 bool child_keep_old
= keep_old_opts
;
4286 /* reopen can only change the options of block devices that were
4287 * implicitly created and inherited options. For other (referenced)
4288 * block devices, a syntax like "backing.foo" results in an error. */
4289 if (child
->bs
->inherits_from
!= bs
) {
4293 /* Check if the options contain a child reference */
4294 if (qdict_haskey(options
, child
->name
)) {
4295 const char *childref
= qdict_get_try_str(options
, child
->name
);
4297 * The current child must not be reopened if the child
4298 * reference is null or points to a different node.
4300 if (g_strcmp0(childref
, child
->bs
->node_name
)) {
4304 * If the child reference points to the current child then
4305 * reopen it with its existing set of options (note that
4306 * it can still inherit new options from the parent).
4308 child_keep_old
= true;
4310 /* Extract child options ("child-name.*") */
4311 char *child_key_dot
= g_strdup_printf("%s.", child
->name
);
4312 qdict_extract_subqdict(explicit_options
, NULL
, child_key_dot
);
4313 qdict_extract_subqdict(options
, &new_child_options
, child_key_dot
);
4314 g_free(child_key_dot
);
4317 bdrv_reopen_queue_child(bs_queue
, child
->bs
, new_child_options
,
4318 child
->klass
, child
->role
, bs
->drv
->is_format
,
4319 options
, flags
, child_keep_old
);
4325 /* To be called with bs->aio_context locked */
4326 BlockReopenQueue
*bdrv_reopen_queue(BlockReopenQueue
*bs_queue
,
4327 BlockDriverState
*bs
,
4328 QDict
*options
, bool keep_old_opts
)
4330 GLOBAL_STATE_CODE();
4332 return bdrv_reopen_queue_child(bs_queue
, bs
, options
, NULL
, 0, false,
4333 NULL
, 0, keep_old_opts
);
4336 void bdrv_reopen_queue_free(BlockReopenQueue
*bs_queue
)
4338 GLOBAL_STATE_CODE();
4340 BlockReopenQueueEntry
*bs_entry
, *next
;
4341 QTAILQ_FOREACH_SAFE(bs_entry
, bs_queue
, entry
, next
) {
4342 AioContext
*ctx
= bdrv_get_aio_context(bs_entry
->state
.bs
);
4344 aio_context_acquire(ctx
);
4345 bdrv_drained_end(bs_entry
->state
.bs
);
4346 aio_context_release(ctx
);
4348 qobject_unref(bs_entry
->state
.explicit_options
);
4349 qobject_unref(bs_entry
->state
.options
);
4357 * Reopen multiple BlockDriverStates atomically & transactionally.
4359 * The queue passed in (bs_queue) must have been built up previous
4360 * via bdrv_reopen_queue().
4362 * Reopens all BDS specified in the queue, with the appropriate
4363 * flags. All devices are prepared for reopen, and failure of any
4364 * device will cause all device changes to be abandoned, and intermediate
4367 * If all devices prepare successfully, then the changes are committed
4370 * All affected nodes must be drained between bdrv_reopen_queue() and
4371 * bdrv_reopen_multiple().
4373 * To be called from the main thread, with all other AioContexts unlocked.
4375 int bdrv_reopen_multiple(BlockReopenQueue
*bs_queue
, Error
**errp
)
4378 BlockReopenQueueEntry
*bs_entry
, *next
;
4380 Transaction
*tran
= tran_new();
4381 g_autoptr(GSList
) refresh_list
= NULL
;
4383 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
4384 assert(bs_queue
!= NULL
);
4385 GLOBAL_STATE_CODE();
4387 QTAILQ_FOREACH(bs_entry
, bs_queue
, entry
) {
4388 ctx
= bdrv_get_aio_context(bs_entry
->state
.bs
);
4389 aio_context_acquire(ctx
);
4390 ret
= bdrv_flush(bs_entry
->state
.bs
);
4391 aio_context_release(ctx
);
4393 error_setg_errno(errp
, -ret
, "Error flushing drive");
4398 QTAILQ_FOREACH(bs_entry
, bs_queue
, entry
) {
4399 assert(bs_entry
->state
.bs
->quiesce_counter
> 0);
4400 ctx
= bdrv_get_aio_context(bs_entry
->state
.bs
);
4401 aio_context_acquire(ctx
);
4402 ret
= bdrv_reopen_prepare(&bs_entry
->state
, bs_queue
, tran
, errp
);
4403 aio_context_release(ctx
);
4407 bs_entry
->prepared
= true;
4410 QTAILQ_FOREACH(bs_entry
, bs_queue
, entry
) {
4411 BDRVReopenState
*state
= &bs_entry
->state
;
4413 refresh_list
= g_slist_prepend(refresh_list
, state
->bs
);
4414 if (state
->old_backing_bs
) {
4415 refresh_list
= g_slist_prepend(refresh_list
, state
->old_backing_bs
);
4417 if (state
->old_file_bs
) {
4418 refresh_list
= g_slist_prepend(refresh_list
, state
->old_file_bs
);
4423 * Note that file-posix driver rely on permission update done during reopen
4424 * (even if no permission changed), because it wants "new" permissions for
4425 * reconfiguring the fd and that's why it does it in raw_check_perm(), not
4426 * in raw_reopen_prepare() which is called with "old" permissions.
4428 ret
= bdrv_list_refresh_perms(refresh_list
, bs_queue
, tran
, errp
);
4434 * If we reach this point, we have success and just need to apply the
4437 * Reverse order is used to comfort qcow2 driver: on commit it need to write
4438 * IN_USE flag to the image, to mark bitmaps in the image as invalid. But
4439 * children are usually goes after parents in reopen-queue, so go from last
4442 QTAILQ_FOREACH_REVERSE(bs_entry
, bs_queue
, entry
) {
4443 ctx
= bdrv_get_aio_context(bs_entry
->state
.bs
);
4444 aio_context_acquire(ctx
);
4445 bdrv_reopen_commit(&bs_entry
->state
);
4446 aio_context_release(ctx
);
4451 QTAILQ_FOREACH_REVERSE(bs_entry
, bs_queue
, entry
) {
4452 BlockDriverState
*bs
= bs_entry
->state
.bs
;
4454 if (bs
->drv
->bdrv_reopen_commit_post
) {
4455 ctx
= bdrv_get_aio_context(bs
);
4456 aio_context_acquire(ctx
);
4457 bs
->drv
->bdrv_reopen_commit_post(&bs_entry
->state
);
4458 aio_context_release(ctx
);
4467 QTAILQ_FOREACH_SAFE(bs_entry
, bs_queue
, entry
, next
) {
4468 if (bs_entry
->prepared
) {
4469 ctx
= bdrv_get_aio_context(bs_entry
->state
.bs
);
4470 aio_context_acquire(ctx
);
4471 bdrv_reopen_abort(&bs_entry
->state
);
4472 aio_context_release(ctx
);
4477 bdrv_reopen_queue_free(bs_queue
);
4482 int bdrv_reopen(BlockDriverState
*bs
, QDict
*opts
, bool keep_old_opts
,
4485 AioContext
*ctx
= bdrv_get_aio_context(bs
);
4486 BlockReopenQueue
*queue
;
4489 GLOBAL_STATE_CODE();
4491 queue
= bdrv_reopen_queue(NULL
, bs
, opts
, keep_old_opts
);
4493 if (ctx
!= qemu_get_aio_context()) {
4494 aio_context_release(ctx
);
4496 ret
= bdrv_reopen_multiple(queue
, errp
);
4498 if (ctx
!= qemu_get_aio_context()) {
4499 aio_context_acquire(ctx
);
4505 int bdrv_reopen_set_read_only(BlockDriverState
*bs
, bool read_only
,
4508 QDict
*opts
= qdict_new();
4510 GLOBAL_STATE_CODE();
4512 qdict_put_bool(opts
, BDRV_OPT_READ_ONLY
, read_only
);
4514 return bdrv_reopen(bs
, opts
, true, errp
);
4518 * Take a BDRVReopenState and check if the value of 'backing' in the
4519 * reopen_state->options QDict is valid or not.
4521 * If 'backing' is missing from the QDict then return 0.
4523 * If 'backing' contains the node name of the backing file of
4524 * reopen_state->bs then return 0.
4526 * If 'backing' contains a different node name (or is null) then check
4527 * whether the current backing file can be replaced with the new one.
4528 * If that's the case then reopen_state->replace_backing_bs is set to
4529 * true and reopen_state->new_backing_bs contains a pointer to the new
4530 * backing BlockDriverState (or NULL).
4532 * Return 0 on success, otherwise return < 0 and set @errp.
4534 static int bdrv_reopen_parse_file_or_backing(BDRVReopenState
*reopen_state
,
4535 bool is_backing
, Transaction
*tran
,
4538 BlockDriverState
*bs
= reopen_state
->bs
;
4539 BlockDriverState
*new_child_bs
;
4540 BlockDriverState
*old_child_bs
= is_backing
? child_bs(bs
->backing
) :
4542 const char *child_name
= is_backing
? "backing" : "file";
4546 GLOBAL_STATE_CODE();
4548 value
= qdict_get(reopen_state
->options
, child_name
);
4549 if (value
== NULL
) {
4553 switch (qobject_type(value
)) {
4555 assert(is_backing
); /* The 'file' option does not allow a null value */
4556 new_child_bs
= NULL
;
4559 str
= qstring_get_str(qobject_to(QString
, value
));
4560 new_child_bs
= bdrv_lookup_bs(NULL
, str
, errp
);
4561 if (new_child_bs
== NULL
) {
4563 } else if (bdrv_recurse_has_child(new_child_bs
, bs
)) {
4564 error_setg(errp
, "Making '%s' a %s child of '%s' would create a "
4565 "cycle", str
, child_name
, bs
->node_name
);
4571 * The options QDict has been flattened, so 'backing' and 'file'
4572 * do not allow any other data type here.
4574 g_assert_not_reached();
4577 if (old_child_bs
== new_child_bs
) {
4582 if (bdrv_skip_implicit_filters(old_child_bs
) == new_child_bs
) {
4586 if (old_child_bs
->implicit
) {
4587 error_setg(errp
, "Cannot replace implicit %s child of %s",
4588 child_name
, bs
->node_name
);
4593 if (bs
->drv
->is_filter
&& !old_child_bs
) {
4595 * Filters always have a file or a backing child, so we are trying to
4596 * change wrong child
4598 error_setg(errp
, "'%s' is a %s filter node that does not support a "
4599 "%s child", bs
->node_name
, bs
->drv
->format_name
, child_name
);
4604 reopen_state
->old_backing_bs
= old_child_bs
;
4606 reopen_state
->old_file_bs
= old_child_bs
;
4609 return bdrv_set_file_or_backing_noperm(bs
, new_child_bs
, is_backing
,
4614 * Prepares a BlockDriverState for reopen. All changes are staged in the
4615 * 'opaque' field of the BDRVReopenState, which is used and allocated by
4616 * the block driver layer .bdrv_reopen_prepare()
4618 * bs is the BlockDriverState to reopen
4619 * flags are the new open flags
4620 * queue is the reopen queue
4622 * Returns 0 on success, non-zero on error. On error errp will be set
4625 * On failure, bdrv_reopen_abort() will be called to clean up any data.
4626 * It is the responsibility of the caller to then call the abort() or
4627 * commit() for any other BDS that have been left in a prepare() state
4630 static int bdrv_reopen_prepare(BDRVReopenState
*reopen_state
,
4631 BlockReopenQueue
*queue
,
4632 Transaction
*change_child_tran
, Error
**errp
)
4636 Error
*local_err
= NULL
;
4639 QDict
*orig_reopen_opts
;
4640 char *discard
= NULL
;
4642 bool drv_prepared
= false;
4644 assert(reopen_state
!= NULL
);
4645 assert(reopen_state
->bs
->drv
!= NULL
);
4646 GLOBAL_STATE_CODE();
4647 drv
= reopen_state
->bs
->drv
;
4649 /* This function and each driver's bdrv_reopen_prepare() remove
4650 * entries from reopen_state->options as they are processed, so
4651 * we need to make a copy of the original QDict. */
4652 orig_reopen_opts
= qdict_clone_shallow(reopen_state
->options
);
4654 /* Process generic block layer options */
4655 opts
= qemu_opts_create(&bdrv_runtime_opts
, NULL
, 0, &error_abort
);
4656 if (!qemu_opts_absorb_qdict(opts
, reopen_state
->options
, errp
)) {
4661 /* This was already called in bdrv_reopen_queue_child() so the flags
4662 * are up-to-date. This time we simply want to remove the options from
4663 * QemuOpts in order to indicate that they have been processed. */
4664 old_flags
= reopen_state
->flags
;
4665 update_flags_from_options(&reopen_state
->flags
, opts
);
4666 assert(old_flags
== reopen_state
->flags
);
4668 discard
= qemu_opt_get_del(opts
, BDRV_OPT_DISCARD
);
4669 if (discard
!= NULL
) {
4670 if (bdrv_parse_discard_flags(discard
, &reopen_state
->flags
) != 0) {
4671 error_setg(errp
, "Invalid discard option");
4677 reopen_state
->detect_zeroes
=
4678 bdrv_parse_detect_zeroes(opts
, reopen_state
->flags
, &local_err
);
4680 error_propagate(errp
, local_err
);
4685 /* All other options (including node-name and driver) must be unchanged.
4686 * Put them back into the QDict, so that they are checked at the end
4687 * of this function. */
4688 qemu_opts_to_qdict(opts
, reopen_state
->options
);
4690 /* If we are to stay read-only, do not allow permission change
4691 * to r/w. Attempting to set to r/w may fail if either BDRV_O_ALLOW_RDWR is
4692 * not set, or if the BDS still has copy_on_read enabled */
4693 read_only
= !(reopen_state
->flags
& BDRV_O_RDWR
);
4694 ret
= bdrv_can_set_read_only(reopen_state
->bs
, read_only
, true, &local_err
);
4696 error_propagate(errp
, local_err
);
4700 if (drv
->bdrv_reopen_prepare
) {
4702 * If a driver-specific option is missing, it means that we
4703 * should reset it to its default value.
4704 * But not all options allow that, so we need to check it first.
4706 ret
= bdrv_reset_options_allowed(reopen_state
->bs
,
4707 reopen_state
->options
, errp
);
4712 ret
= drv
->bdrv_reopen_prepare(reopen_state
, queue
, &local_err
);
4714 if (local_err
!= NULL
) {
4715 error_propagate(errp
, local_err
);
4717 bdrv_refresh_filename(reopen_state
->bs
);
4718 error_setg(errp
, "failed while preparing to reopen image '%s'",
4719 reopen_state
->bs
->filename
);
4724 /* It is currently mandatory to have a bdrv_reopen_prepare()
4725 * handler for each supported drv. */
4726 error_setg(errp
, "Block format '%s' used by node '%s' "
4727 "does not support reopening files", drv
->format_name
,
4728 bdrv_get_device_or_node_name(reopen_state
->bs
));
4733 drv_prepared
= true;
4736 * We must provide the 'backing' option if the BDS has a backing
4737 * file or if the image file has a backing file name as part of
4738 * its metadata. Otherwise the 'backing' option can be omitted.
4740 if (drv
->supports_backing
&& reopen_state
->backing_missing
&&
4741 (reopen_state
->bs
->backing
|| reopen_state
->bs
->backing_file
[0])) {
4742 error_setg(errp
, "backing is missing for '%s'",
4743 reopen_state
->bs
->node_name
);
4749 * Allow changing the 'backing' option. The new value can be
4750 * either a reference to an existing node (using its node name)
4751 * or NULL to simply detach the current backing file.
4753 ret
= bdrv_reopen_parse_file_or_backing(reopen_state
, true,
4754 change_child_tran
, errp
);
4758 qdict_del(reopen_state
->options
, "backing");
4760 /* Allow changing the 'file' option. In this case NULL is not allowed */
4761 ret
= bdrv_reopen_parse_file_or_backing(reopen_state
, false,
4762 change_child_tran
, errp
);
4766 qdict_del(reopen_state
->options
, "file");
4768 /* Options that are not handled are only okay if they are unchanged
4769 * compared to the old state. It is expected that some options are only
4770 * used for the initial open, but not reopen (e.g. filename) */
4771 if (qdict_size(reopen_state
->options
)) {
4772 const QDictEntry
*entry
= qdict_first(reopen_state
->options
);
4775 QObject
*new = entry
->value
;
4776 QObject
*old
= qdict_get(reopen_state
->bs
->options
, entry
->key
);
4778 /* Allow child references (child_name=node_name) as long as they
4779 * point to the current child (i.e. everything stays the same). */
4780 if (qobject_type(new) == QTYPE_QSTRING
) {
4782 QLIST_FOREACH(child
, &reopen_state
->bs
->children
, next
) {
4783 if (!strcmp(child
->name
, entry
->key
)) {
4789 if (!strcmp(child
->bs
->node_name
,
4790 qstring_get_str(qobject_to(QString
, new)))) {
4791 continue; /* Found child with this name, skip option */
4797 * TODO: When using -drive to specify blockdev options, all values
4798 * will be strings; however, when using -blockdev, blockdev-add or
4799 * filenames using the json:{} pseudo-protocol, they will be
4801 * In contrast, reopening options are (currently) always strings
4802 * (because you can only specify them through qemu-io; all other
4803 * callers do not specify any options).
4804 * Therefore, when using anything other than -drive to create a BDS,
4805 * this cannot detect non-string options as unchanged, because
4806 * qobject_is_equal() always returns false for objects of different
4807 * type. In the future, this should be remedied by correctly typing
4808 * all options. For now, this is not too big of an issue because
4809 * the user can simply omit options which cannot be changed anyway,
4810 * so they will stay unchanged.
4812 if (!qobject_is_equal(new, old
)) {
4813 error_setg(errp
, "Cannot change the option '%s'", entry
->key
);
4817 } while ((entry
= qdict_next(reopen_state
->options
, entry
)));
4822 /* Restore the original reopen_state->options QDict */
4823 qobject_unref(reopen_state
->options
);
4824 reopen_state
->options
= qobject_ref(orig_reopen_opts
);
4827 if (ret
< 0 && drv_prepared
) {
4828 /* drv->bdrv_reopen_prepare() has succeeded, so we need to
4829 * call drv->bdrv_reopen_abort() before signaling an error
4830 * (bdrv_reopen_multiple() will not call bdrv_reopen_abort()
4831 * when the respective bdrv_reopen_prepare() has failed) */
4832 if (drv
->bdrv_reopen_abort
) {
4833 drv
->bdrv_reopen_abort(reopen_state
);
4836 qemu_opts_del(opts
);
4837 qobject_unref(orig_reopen_opts
);
4843 * Takes the staged changes for the reopen from bdrv_reopen_prepare(), and
4844 * makes them final by swapping the staging BlockDriverState contents into
4845 * the active BlockDriverState contents.
4847 static void bdrv_reopen_commit(BDRVReopenState
*reopen_state
)
4850 BlockDriverState
*bs
;
4853 assert(reopen_state
!= NULL
);
4854 bs
= reopen_state
->bs
;
4856 assert(drv
!= NULL
);
4857 GLOBAL_STATE_CODE();
4859 /* If there are any driver level actions to take */
4860 if (drv
->bdrv_reopen_commit
) {
4861 drv
->bdrv_reopen_commit(reopen_state
);
4864 /* set BDS specific flags now */
4865 qobject_unref(bs
->explicit_options
);
4866 qobject_unref(bs
->options
);
4867 qobject_ref(reopen_state
->explicit_options
);
4868 qobject_ref(reopen_state
->options
);
4870 bs
->explicit_options
= reopen_state
->explicit_options
;
4871 bs
->options
= reopen_state
->options
;
4872 bs
->open_flags
= reopen_state
->flags
;
4873 bs
->detect_zeroes
= reopen_state
->detect_zeroes
;
4875 /* Remove child references from bs->options and bs->explicit_options.
4876 * Child options were already removed in bdrv_reopen_queue_child() */
4877 QLIST_FOREACH(child
, &bs
->children
, next
) {
4878 qdict_del(bs
->explicit_options
, child
->name
);
4879 qdict_del(bs
->options
, child
->name
);
4881 /* backing is probably removed, so it's not handled by previous loop */
4882 qdict_del(bs
->explicit_options
, "backing");
4883 qdict_del(bs
->options
, "backing");
4885 bdrv_refresh_limits(bs
, NULL
, NULL
);
4889 * Abort the reopen, and delete and free the staged changes in
4892 static void bdrv_reopen_abort(BDRVReopenState
*reopen_state
)
4896 assert(reopen_state
!= NULL
);
4897 drv
= reopen_state
->bs
->drv
;
4898 assert(drv
!= NULL
);
4899 GLOBAL_STATE_CODE();
4901 if (drv
->bdrv_reopen_abort
) {
4902 drv
->bdrv_reopen_abort(reopen_state
);
4907 static void bdrv_close(BlockDriverState
*bs
)
4909 BdrvAioNotifier
*ban
, *ban_next
;
4910 BdrvChild
*child
, *next
;
4912 GLOBAL_STATE_CODE();
4913 assert(!bs
->refcnt
);
4915 bdrv_drained_begin(bs
); /* complete I/O */
4917 bdrv_drain(bs
); /* in case flush left pending I/O */
4920 if (bs
->drv
->bdrv_close
) {
4921 /* Must unfreeze all children, so bdrv_unref_child() works */
4922 bs
->drv
->bdrv_close(bs
);
4927 QLIST_FOREACH_SAFE(child
, &bs
->children
, next
, next
) {
4928 bdrv_unref_child(bs
, child
);
4931 assert(!bs
->backing
);
4935 qatomic_set(&bs
->copy_on_read
, 0);
4936 bs
->backing_file
[0] = '\0';
4937 bs
->backing_format
[0] = '\0';
4938 bs
->total_sectors
= 0;
4939 bs
->encrypted
= false;
4941 qobject_unref(bs
->options
);
4942 qobject_unref(bs
->explicit_options
);
4944 bs
->explicit_options
= NULL
;
4945 qobject_unref(bs
->full_open_options
);
4946 bs
->full_open_options
= NULL
;
4947 g_free(bs
->block_status_cache
);
4948 bs
->block_status_cache
= NULL
;
4950 bdrv_release_named_dirty_bitmaps(bs
);
4951 assert(QLIST_EMPTY(&bs
->dirty_bitmaps
));
4953 QLIST_FOREACH_SAFE(ban
, &bs
->aio_notifiers
, list
, ban_next
) {
4956 QLIST_INIT(&bs
->aio_notifiers
);
4957 bdrv_drained_end(bs
);
4960 * If we're still inside some bdrv_drain_all_begin()/end() sections, end
4961 * them now since this BDS won't exist anymore when bdrv_drain_all_end()
4964 if (bs
->quiesce_counter
) {
4965 bdrv_drain_all_end_quiesce(bs
);
4969 void bdrv_close_all(void)
4971 GLOBAL_STATE_CODE();
4972 assert(job_next(NULL
) == NULL
);
4974 /* Drop references from requests still in flight, such as canceled block
4975 * jobs whose AIO context has not been polled yet */
4978 blk_remove_all_bs();
4979 blockdev_close_all_bdrv_states();
4981 assert(QTAILQ_EMPTY(&all_bdrv_states
));
4984 static bool should_update_child(BdrvChild
*c
, BlockDriverState
*to
)
4990 if (c
->klass
->stay_at_node
) {
4994 /* If the child @c belongs to the BDS @to, replacing the current
4995 * c->bs by @to would mean to create a loop.
4997 * Such a case occurs when appending a BDS to a backing chain.
4998 * For instance, imagine the following chain:
5000 * guest device -> node A -> further backing chain...
5002 * Now we create a new BDS B which we want to put on top of this
5003 * chain, so we first attach A as its backing node:
5008 * guest device -> node A -> further backing chain...
5010 * Finally we want to replace A by B. When doing that, we want to
5011 * replace all pointers to A by pointers to B -- except for the
5012 * pointer from B because (1) that would create a loop, and (2)
5013 * that pointer should simply stay intact:
5015 * guest device -> node B
5018 * node A -> further backing chain...
5020 * In general, when replacing a node A (c->bs) by a node B (@to),
5021 * if A is a child of B, that means we cannot replace A by B there
5022 * because that would create a loop. Silently detaching A from B
5023 * is also not really an option. So overall just leaving A in
5024 * place there is the most sensible choice.
5026 * We would also create a loop in any cases where @c is only
5027 * indirectly referenced by @to. Prevent this by returning false
5028 * if @c is found (by breadth-first search) anywhere in the whole
5033 found
= g_hash_table_new(NULL
, NULL
);
5034 g_hash_table_add(found
, to
);
5035 queue
= g_queue_new();
5036 g_queue_push_tail(queue
, to
);
5038 while (!g_queue_is_empty(queue
)) {
5039 BlockDriverState
*v
= g_queue_pop_head(queue
);
5042 QLIST_FOREACH(c2
, &v
->children
, next
) {
5048 if (g_hash_table_contains(found
, c2
->bs
)) {
5052 g_queue_push_tail(queue
, c2
->bs
);
5053 g_hash_table_add(found
, c2
->bs
);
5057 g_queue_free(queue
);
5058 g_hash_table_destroy(found
);
5063 static void bdrv_remove_child_commit(void *opaque
)
5065 GLOBAL_STATE_CODE();
5066 bdrv_child_free(opaque
);
5069 static TransactionActionDrv bdrv_remove_child_drv
= {
5070 .commit
= bdrv_remove_child_commit
,
5073 /* Function doesn't update permissions, caller is responsible for this. */
5074 static void bdrv_remove_child(BdrvChild
*child
, Transaction
*tran
)
5081 bdrv_replace_child_tran(child
, NULL
, tran
);
5084 tran_add(tran
, &bdrv_remove_child_drv
, child
);
5087 static int bdrv_replace_node_noperm(BlockDriverState
*from
,
5088 BlockDriverState
*to
,
5089 bool auto_skip
, Transaction
*tran
,
5092 BdrvChild
*c
, *next
;
5094 GLOBAL_STATE_CODE();
5096 QLIST_FOREACH_SAFE(c
, &from
->parents
, next_parent
, next
) {
5097 assert(c
->bs
== from
);
5098 if (!should_update_child(c
, to
)) {
5102 error_setg(errp
, "Should not change '%s' link to '%s'",
5103 c
->name
, from
->node_name
);
5107 error_setg(errp
, "Cannot change '%s' link to '%s'",
5108 c
->name
, from
->node_name
);
5111 bdrv_replace_child_tran(c
, to
, tran
);
5118 * With auto_skip=true bdrv_replace_node_common skips updating from parents
5119 * if it creates a parent-child relation loop or if parent is block-job.
5121 * With auto_skip=false the error is returned if from has a parent which should
5124 * With @detach_subchain=true @to must be in a backing chain of @from. In this
5125 * case backing link of the cow-parent of @to is removed.
5127 static int bdrv_replace_node_common(BlockDriverState
*from
,
5128 BlockDriverState
*to
,
5129 bool auto_skip
, bool detach_subchain
,
5132 Transaction
*tran
= tran_new();
5133 g_autoptr(GSList
) refresh_list
= NULL
;
5134 BlockDriverState
*to_cow_parent
= NULL
;
5137 GLOBAL_STATE_CODE();
5139 if (detach_subchain
) {
5140 assert(bdrv_chain_contains(from
, to
));
5142 for (to_cow_parent
= from
;
5143 bdrv_filter_or_cow_bs(to_cow_parent
) != to
;
5144 to_cow_parent
= bdrv_filter_or_cow_bs(to_cow_parent
))
5150 /* Make sure that @from doesn't go away until we have successfully attached
5151 * all of its parents to @to. */
5154 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
5155 assert(bdrv_get_aio_context(from
) == bdrv_get_aio_context(to
));
5156 bdrv_drained_begin(from
);
5159 * Do the replacement without permission update.
5160 * Replacement may influence the permissions, we should calculate new
5161 * permissions based on new graph. If we fail, we'll roll-back the
5164 ret
= bdrv_replace_node_noperm(from
, to
, auto_skip
, tran
, errp
);
5169 if (detach_subchain
) {
5170 bdrv_remove_child(bdrv_filter_or_cow_child(to_cow_parent
), tran
);
5173 refresh_list
= g_slist_prepend(refresh_list
, to
);
5174 refresh_list
= g_slist_prepend(refresh_list
, from
);
5176 ret
= bdrv_list_refresh_perms(refresh_list
, NULL
, tran
, errp
);
5184 tran_finalize(tran
, ret
);
5186 bdrv_drained_end(from
);
5192 int bdrv_replace_node(BlockDriverState
*from
, BlockDriverState
*to
,
5195 GLOBAL_STATE_CODE();
5197 return bdrv_replace_node_common(from
, to
, true, false, errp
);
5200 int bdrv_drop_filter(BlockDriverState
*bs
, Error
**errp
)
5202 GLOBAL_STATE_CODE();
5204 return bdrv_replace_node_common(bs
, bdrv_filter_or_cow_bs(bs
), true, true,
5209 * Add new bs contents at the top of an image chain while the chain is
5210 * live, while keeping required fields on the top layer.
5212 * This will modify the BlockDriverState fields, and swap contents
5213 * between bs_new and bs_top. Both bs_new and bs_top are modified.
5215 * bs_new must not be attached to a BlockBackend and must not have backing
5218 * This function does not create any image files.
5220 int bdrv_append(BlockDriverState
*bs_new
, BlockDriverState
*bs_top
,
5225 Transaction
*tran
= tran_new();
5227 GLOBAL_STATE_CODE();
5229 assert(!bs_new
->backing
);
5231 child
= bdrv_attach_child_noperm(bs_new
, bs_top
, "backing",
5232 &child_of_bds
, bdrv_backing_role(bs_new
),
5239 ret
= bdrv_replace_node_noperm(bs_top
, bs_new
, true, tran
, errp
);
5244 ret
= bdrv_refresh_perms(bs_new
, tran
, errp
);
5246 tran_finalize(tran
, ret
);
5248 bdrv_refresh_limits(bs_top
, NULL
, NULL
);
5253 /* Not for empty child */
5254 int bdrv_replace_child_bs(BdrvChild
*child
, BlockDriverState
*new_bs
,
5258 Transaction
*tran
= tran_new();
5259 g_autoptr(GSList
) refresh_list
= NULL
;
5260 BlockDriverState
*old_bs
= child
->bs
;
5262 GLOBAL_STATE_CODE();
5265 bdrv_drained_begin(old_bs
);
5266 bdrv_drained_begin(new_bs
);
5268 bdrv_replace_child_tran(child
, new_bs
, tran
);
5270 refresh_list
= g_slist_prepend(refresh_list
, old_bs
);
5271 refresh_list
= g_slist_prepend(refresh_list
, new_bs
);
5273 ret
= bdrv_list_refresh_perms(refresh_list
, NULL
, tran
, errp
);
5275 tran_finalize(tran
, ret
);
5277 bdrv_drained_end(old_bs
);
5278 bdrv_drained_end(new_bs
);
5284 static void bdrv_delete(BlockDriverState
*bs
)
5286 assert(bdrv_op_blocker_is_empty(bs
));
5287 assert(!bs
->refcnt
);
5288 GLOBAL_STATE_CODE();
5290 /* remove from list, if necessary */
5291 if (bs
->node_name
[0] != '\0') {
5292 QTAILQ_REMOVE(&graph_bdrv_states
, bs
, node_list
);
5294 QTAILQ_REMOVE(&all_bdrv_states
, bs
, bs_list
);
5303 * Replace @bs by newly created block node.
5305 * @options is a QDict of options to pass to the block drivers, or NULL for an
5306 * empty set of options. The reference to the QDict belongs to the block layer
5307 * after the call (even on failure), so if the caller intends to reuse the
5308 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
5310 BlockDriverState
*bdrv_insert_node(BlockDriverState
*bs
, QDict
*options
,
5311 int flags
, Error
**errp
)
5315 BlockDriverState
*new_node_bs
= NULL
;
5316 const char *drvname
, *node_name
;
5319 drvname
= qdict_get_try_str(options
, "driver");
5321 error_setg(errp
, "driver is not specified");
5325 drv
= bdrv_find_format(drvname
);
5327 error_setg(errp
, "Unknown driver: '%s'", drvname
);
5331 node_name
= qdict_get_try_str(options
, "node-name");
5333 GLOBAL_STATE_CODE();
5335 new_node_bs
= bdrv_new_open_driver_opts(drv
, node_name
, options
, flags
,
5337 options
= NULL
; /* bdrv_new_open_driver() eats options */
5339 error_prepend(errp
, "Could not create node: ");
5343 bdrv_drained_begin(bs
);
5344 ret
= bdrv_replace_node(bs
, new_node_bs
, errp
);
5345 bdrv_drained_end(bs
);
5348 error_prepend(errp
, "Could not replace node: ");
5355 qobject_unref(options
);
5356 bdrv_unref(new_node_bs
);
5361 * Run consistency checks on an image
5363 * Returns 0 if the check could be completed (it doesn't mean that the image is
5364 * free of errors) or -errno when an internal error occurred. The results of the
5365 * check are stored in res.
5367 int coroutine_fn
bdrv_co_check(BlockDriverState
*bs
,
5368 BdrvCheckResult
*res
, BdrvCheckMode fix
)
5371 if (bs
->drv
== NULL
) {
5374 if (bs
->drv
->bdrv_co_check
== NULL
) {
5378 memset(res
, 0, sizeof(*res
));
5379 return bs
->drv
->bdrv_co_check(bs
, res
, fix
);
5385 * -EINVAL - backing format specified, but no file
5386 * -ENOSPC - can't update the backing file because no space is left in the
5388 * -ENOTSUP - format driver doesn't support changing the backing file
5390 int bdrv_change_backing_file(BlockDriverState
*bs
, const char *backing_file
,
5391 const char *backing_fmt
, bool require
)
5393 BlockDriver
*drv
= bs
->drv
;
5396 GLOBAL_STATE_CODE();
5402 /* Backing file format doesn't make sense without a backing file */
5403 if (backing_fmt
&& !backing_file
) {
5407 if (require
&& backing_file
&& !backing_fmt
) {
5411 if (drv
->bdrv_change_backing_file
!= NULL
) {
5412 ret
= drv
->bdrv_change_backing_file(bs
, backing_file
, backing_fmt
);
5418 pstrcpy(bs
->backing_file
, sizeof(bs
->backing_file
), backing_file
?: "");
5419 pstrcpy(bs
->backing_format
, sizeof(bs
->backing_format
), backing_fmt
?: "");
5420 pstrcpy(bs
->auto_backing_file
, sizeof(bs
->auto_backing_file
),
5421 backing_file
?: "");
5427 * Finds the first non-filter node above bs in the chain between
5428 * active and bs. The returned node is either an immediate parent of
5429 * bs, or there are only filter nodes between the two.
5431 * Returns NULL if bs is not found in active's image chain,
5432 * or if active == bs.
5434 * Returns the bottommost base image if bs == NULL.
5436 BlockDriverState
*bdrv_find_overlay(BlockDriverState
*active
,
5437 BlockDriverState
*bs
)
5440 GLOBAL_STATE_CODE();
5442 bs
= bdrv_skip_filters(bs
);
5443 active
= bdrv_skip_filters(active
);
5446 BlockDriverState
*next
= bdrv_backing_chain_next(active
);
5456 /* Given a BDS, searches for the base layer. */
5457 BlockDriverState
*bdrv_find_base(BlockDriverState
*bs
)
5459 GLOBAL_STATE_CODE();
5461 return bdrv_find_overlay(bs
, NULL
);
5465 * Return true if at least one of the COW (backing) and filter links
5466 * between @bs and @base is frozen. @errp is set if that's the case.
5467 * @base must be reachable from @bs, or NULL.
5469 bool bdrv_is_backing_chain_frozen(BlockDriverState
*bs
, BlockDriverState
*base
,
5472 BlockDriverState
*i
;
5475 GLOBAL_STATE_CODE();
5477 for (i
= bs
; i
!= base
; i
= child_bs(child
)) {
5478 child
= bdrv_filter_or_cow_child(i
);
5480 if (child
&& child
->frozen
) {
5481 error_setg(errp
, "Cannot change '%s' link from '%s' to '%s'",
5482 child
->name
, i
->node_name
, child
->bs
->node_name
);
5491 * Freeze all COW (backing) and filter links between @bs and @base.
5492 * If any of the links is already frozen the operation is aborted and
5493 * none of the links are modified.
5494 * @base must be reachable from @bs, or NULL.
5495 * Returns 0 on success. On failure returns < 0 and sets @errp.
5497 int bdrv_freeze_backing_chain(BlockDriverState
*bs
, BlockDriverState
*base
,
5500 BlockDriverState
*i
;
5503 GLOBAL_STATE_CODE();
5505 if (bdrv_is_backing_chain_frozen(bs
, base
, errp
)) {
5509 for (i
= bs
; i
!= base
; i
= child_bs(child
)) {
5510 child
= bdrv_filter_or_cow_child(i
);
5511 if (child
&& child
->bs
->never_freeze
) {
5512 error_setg(errp
, "Cannot freeze '%s' link to '%s'",
5513 child
->name
, child
->bs
->node_name
);
5518 for (i
= bs
; i
!= base
; i
= child_bs(child
)) {
5519 child
= bdrv_filter_or_cow_child(i
);
5521 child
->frozen
= true;
5529 * Unfreeze all COW (backing) and filter links between @bs and @base.
5530 * The caller must ensure that all links are frozen before using this
5532 * @base must be reachable from @bs, or NULL.
5534 void bdrv_unfreeze_backing_chain(BlockDriverState
*bs
, BlockDriverState
*base
)
5536 BlockDriverState
*i
;
5539 GLOBAL_STATE_CODE();
5541 for (i
= bs
; i
!= base
; i
= child_bs(child
)) {
5542 child
= bdrv_filter_or_cow_child(i
);
5544 assert(child
->frozen
);
5545 child
->frozen
= false;
5551 * Drops images above 'base' up to and including 'top', and sets the image
5552 * above 'top' to have base as its backing file.
5554 * Requires that the overlay to 'top' is opened r/w, so that the backing file
5555 * information in 'bs' can be properly updated.
5557 * E.g., this will convert the following chain:
5558 * bottom <- base <- intermediate <- top <- active
5562 * bottom <- base <- active
5564 * It is allowed for bottom==base, in which case it converts:
5566 * base <- intermediate <- top <- active
5572 * If backing_file_str is non-NULL, it will be used when modifying top's
5573 * overlay image metadata.
5576 * if active == top, that is considered an error
5579 int bdrv_drop_intermediate(BlockDriverState
*top
, BlockDriverState
*base
,
5580 const char *backing_file_str
)
5582 BlockDriverState
*explicit_top
= top
;
5583 bool update_inherits_from
;
5585 Error
*local_err
= NULL
;
5587 g_autoptr(GSList
) updated_children
= NULL
;
5590 GLOBAL_STATE_CODE();
5593 bdrv_drained_begin(base
);
5595 if (!top
->drv
|| !base
->drv
) {
5599 /* Make sure that base is in the backing chain of top */
5600 if (!bdrv_chain_contains(top
, base
)) {
5604 /* If 'base' recursively inherits from 'top' then we should set
5605 * base->inherits_from to top->inherits_from after 'top' and all
5606 * other intermediate nodes have been dropped.
5607 * If 'top' is an implicit node (e.g. "commit_top") we should skip
5608 * it because no one inherits from it. We use explicit_top for that. */
5609 explicit_top
= bdrv_skip_implicit_filters(explicit_top
);
5610 update_inherits_from
= bdrv_inherits_from_recursive(base
, explicit_top
);
5612 /* success - we can delete the intermediate states, and link top->base */
5613 if (!backing_file_str
) {
5614 bdrv_refresh_filename(base
);
5615 backing_file_str
= base
->filename
;
5618 QLIST_FOREACH(c
, &top
->parents
, next_parent
) {
5619 updated_children
= g_slist_prepend(updated_children
, c
);
5623 * It seems correct to pass detach_subchain=true here, but it triggers
5624 * one more yet not fixed bug, when due to nested aio_poll loop we switch to
5625 * another drained section, which modify the graph (for example, removing
5626 * the child, which we keep in updated_children list). So, it's a TODO.
5628 * Note, bug triggered if pass detach_subchain=true here and run
5629 * test-bdrv-drain. test_drop_intermediate_poll() test-case will crash.
5632 bdrv_replace_node_common(top
, base
, false, false, &local_err
);
5634 error_report_err(local_err
);
5638 for (p
= updated_children
; p
; p
= p
->next
) {
5641 if (c
->klass
->update_filename
) {
5642 ret
= c
->klass
->update_filename(c
, base
, backing_file_str
,
5646 * TODO: Actually, we want to rollback all previous iterations
5647 * of this loop, and (which is almost impossible) previous
5648 * bdrv_replace_node()...
5650 * Note, that c->klass->update_filename may lead to permission
5651 * update, so it's a bad idea to call it inside permission
5652 * update transaction of bdrv_replace_node.
5654 error_report_err(local_err
);
5660 if (update_inherits_from
) {
5661 base
->inherits_from
= explicit_top
->inherits_from
;
5666 bdrv_drained_end(base
);
5672 * Implementation of BlockDriver.bdrv_get_allocated_file_size() that
5673 * sums the size of all data-bearing children. (This excludes backing
5676 static int64_t bdrv_sum_allocated_file_size(BlockDriverState
*bs
)
5679 int64_t child_size
, sum
= 0;
5681 QLIST_FOREACH(child
, &bs
->children
, next
) {
5682 if (child
->role
& (BDRV_CHILD_DATA
| BDRV_CHILD_METADATA
|
5683 BDRV_CHILD_FILTERED
))
5685 child_size
= bdrv_get_allocated_file_size(child
->bs
);
5686 if (child_size
< 0) {
5697 * Length of a allocated file in bytes. Sparse files are counted by actual
5698 * allocated space. Return < 0 if error or unknown.
5700 int64_t bdrv_get_allocated_file_size(BlockDriverState
*bs
)
5702 BlockDriver
*drv
= bs
->drv
;
5708 if (drv
->bdrv_get_allocated_file_size
) {
5709 return drv
->bdrv_get_allocated_file_size(bs
);
5712 if (drv
->bdrv_file_open
) {
5714 * Protocol drivers default to -ENOTSUP (most of their data is
5715 * not stored in any of their children (if they even have any),
5716 * so there is no generic way to figure it out).
5719 } else if (drv
->is_filter
) {
5720 /* Filter drivers default to the size of their filtered child */
5721 return bdrv_get_allocated_file_size(bdrv_filter_bs(bs
));
5723 /* Other drivers default to summing their children's sizes */
5724 return bdrv_sum_allocated_file_size(bs
);
5730 * @drv: Format driver
5731 * @opts: Creation options for new image
5732 * @in_bs: Existing image containing data for new image (may be NULL)
5733 * @errp: Error object
5734 * Returns: A #BlockMeasureInfo (free using qapi_free_BlockMeasureInfo())
5737 * Calculate file size required to create a new image.
5739 * If @in_bs is given then space for allocated clusters and zero clusters
5740 * from that image are included in the calculation. If @opts contains a
5741 * backing file that is shared by @in_bs then backing clusters may be omitted
5742 * from the calculation.
5744 * If @in_bs is NULL then the calculation includes no allocated clusters
5745 * unless a preallocation option is given in @opts.
5747 * Note that @in_bs may use a different BlockDriver from @drv.
5749 * If an error occurs the @errp pointer is set.
5751 BlockMeasureInfo
*bdrv_measure(BlockDriver
*drv
, QemuOpts
*opts
,
5752 BlockDriverState
*in_bs
, Error
**errp
)
5755 if (!drv
->bdrv_measure
) {
5756 error_setg(errp
, "Block driver '%s' does not support size measurement",
5761 return drv
->bdrv_measure(opts
, in_bs
, errp
);
5765 * Return number of sectors on success, -errno on error.
5767 int64_t bdrv_nb_sectors(BlockDriverState
*bs
)
5769 BlockDriver
*drv
= bs
->drv
;
5775 if (drv
->has_variable_length
) {
5776 int ret
= refresh_total_sectors(bs
, bs
->total_sectors
);
5781 return bs
->total_sectors
;
5785 * Return length in bytes on success, -errno on error.
5786 * The length is always a multiple of BDRV_SECTOR_SIZE.
5788 int64_t bdrv_getlength(BlockDriverState
*bs
)
5790 int64_t ret
= bdrv_nb_sectors(bs
);
5796 if (ret
> INT64_MAX
/ BDRV_SECTOR_SIZE
) {
5799 return ret
* BDRV_SECTOR_SIZE
;
5802 /* return 0 as number of sectors if no device present or error */
5803 void bdrv_get_geometry(BlockDriverState
*bs
, uint64_t *nb_sectors_ptr
)
5805 int64_t nb_sectors
= bdrv_nb_sectors(bs
);
5808 *nb_sectors_ptr
= nb_sectors
< 0 ? 0 : nb_sectors
;
5811 bool bdrv_is_sg(BlockDriverState
*bs
)
5818 * Return whether the given node supports compressed writes.
5820 bool bdrv_supports_compressed_writes(BlockDriverState
*bs
)
5822 BlockDriverState
*filtered
;
5825 if (!bs
->drv
|| !block_driver_can_compress(bs
->drv
)) {
5829 filtered
= bdrv_filter_bs(bs
);
5832 * Filters can only forward compressed writes, so we have to
5835 return bdrv_supports_compressed_writes(filtered
);
5841 const char *bdrv_get_format_name(BlockDriverState
*bs
)
5844 return bs
->drv
? bs
->drv
->format_name
: NULL
;
5847 static int qsort_strcmp(const void *a
, const void *b
)
5849 return strcmp(*(char *const *)a
, *(char *const *)b
);
5852 void bdrv_iterate_format(void (*it
)(void *opaque
, const char *name
),
5853 void *opaque
, bool read_only
)
5858 const char **formats
= NULL
;
5860 GLOBAL_STATE_CODE();
5862 QLIST_FOREACH(drv
, &bdrv_drivers
, list
) {
5863 if (drv
->format_name
) {
5867 if (use_bdrv_whitelist
&& !bdrv_is_whitelisted(drv
, read_only
)) {
5871 while (formats
&& i
&& !found
) {
5872 found
= !strcmp(formats
[--i
], drv
->format_name
);
5876 formats
= g_renew(const char *, formats
, count
+ 1);
5877 formats
[count
++] = drv
->format_name
;
5882 for (i
= 0; i
< (int)ARRAY_SIZE(block_driver_modules
); i
++) {
5883 const char *format_name
= block_driver_modules
[i
].format_name
;
5889 if (use_bdrv_whitelist
&&
5890 !bdrv_format_is_whitelisted(format_name
, read_only
)) {
5894 while (formats
&& j
&& !found
) {
5895 found
= !strcmp(formats
[--j
], format_name
);
5899 formats
= g_renew(const char *, formats
, count
+ 1);
5900 formats
[count
++] = format_name
;
5905 qsort(formats
, count
, sizeof(formats
[0]), qsort_strcmp
);
5907 for (i
= 0; i
< count
; i
++) {
5908 it(opaque
, formats
[i
]);
5914 /* This function is to find a node in the bs graph */
5915 BlockDriverState
*bdrv_find_node(const char *node_name
)
5917 BlockDriverState
*bs
;
5920 GLOBAL_STATE_CODE();
5922 QTAILQ_FOREACH(bs
, &graph_bdrv_states
, node_list
) {
5923 if (!strcmp(node_name
, bs
->node_name
)) {
5930 /* Put this QMP function here so it can access the static graph_bdrv_states. */
5931 BlockDeviceInfoList
*bdrv_named_nodes_list(bool flat
,
5934 BlockDeviceInfoList
*list
;
5935 BlockDriverState
*bs
;
5937 GLOBAL_STATE_CODE();
5940 QTAILQ_FOREACH(bs
, &graph_bdrv_states
, node_list
) {
5941 BlockDeviceInfo
*info
= bdrv_block_device_info(NULL
, bs
, flat
, errp
);
5943 qapi_free_BlockDeviceInfoList(list
);
5946 QAPI_LIST_PREPEND(list
, info
);
5952 typedef struct XDbgBlockGraphConstructor
{
5953 XDbgBlockGraph
*graph
;
5954 GHashTable
*graph_nodes
;
5955 } XDbgBlockGraphConstructor
;
5957 static XDbgBlockGraphConstructor
*xdbg_graph_new(void)
5959 XDbgBlockGraphConstructor
*gr
= g_new(XDbgBlockGraphConstructor
, 1);
5961 gr
->graph
= g_new0(XDbgBlockGraph
, 1);
5962 gr
->graph_nodes
= g_hash_table_new(NULL
, NULL
);
5967 static XDbgBlockGraph
*xdbg_graph_finalize(XDbgBlockGraphConstructor
*gr
)
5969 XDbgBlockGraph
*graph
= gr
->graph
;
5971 g_hash_table_destroy(gr
->graph_nodes
);
5977 static uintptr_t xdbg_graph_node_num(XDbgBlockGraphConstructor
*gr
, void *node
)
5979 uintptr_t ret
= (uintptr_t)g_hash_table_lookup(gr
->graph_nodes
, node
);
5986 * Start counting from 1, not 0, because 0 interferes with not-found (NULL)
5987 * answer of g_hash_table_lookup.
5989 ret
= g_hash_table_size(gr
->graph_nodes
) + 1;
5990 g_hash_table_insert(gr
->graph_nodes
, node
, (void *)ret
);
5995 static void xdbg_graph_add_node(XDbgBlockGraphConstructor
*gr
, void *node
,
5996 XDbgBlockGraphNodeType type
, const char *name
)
5998 XDbgBlockGraphNode
*n
;
6000 n
= g_new0(XDbgBlockGraphNode
, 1);
6002 n
->id
= xdbg_graph_node_num(gr
, node
);
6004 n
->name
= g_strdup(name
);
6006 QAPI_LIST_PREPEND(gr
->graph
->nodes
, n
);
6009 static void xdbg_graph_add_edge(XDbgBlockGraphConstructor
*gr
, void *parent
,
6010 const BdrvChild
*child
)
6012 BlockPermission qapi_perm
;
6013 XDbgBlockGraphEdge
*edge
;
6014 GLOBAL_STATE_CODE();
6016 edge
= g_new0(XDbgBlockGraphEdge
, 1);
6018 edge
->parent
= xdbg_graph_node_num(gr
, parent
);
6019 edge
->child
= xdbg_graph_node_num(gr
, child
->bs
);
6020 edge
->name
= g_strdup(child
->name
);
6022 for (qapi_perm
= 0; qapi_perm
< BLOCK_PERMISSION__MAX
; qapi_perm
++) {
6023 uint64_t flag
= bdrv_qapi_perm_to_blk_perm(qapi_perm
);
6025 if (flag
& child
->perm
) {
6026 QAPI_LIST_PREPEND(edge
->perm
, qapi_perm
);
6028 if (flag
& child
->shared_perm
) {
6029 QAPI_LIST_PREPEND(edge
->shared_perm
, qapi_perm
);
6033 QAPI_LIST_PREPEND(gr
->graph
->edges
, edge
);
6037 XDbgBlockGraph
*bdrv_get_xdbg_block_graph(Error
**errp
)
6041 BlockDriverState
*bs
;
6043 XDbgBlockGraphConstructor
*gr
= xdbg_graph_new();
6045 GLOBAL_STATE_CODE();
6047 for (blk
= blk_all_next(NULL
); blk
; blk
= blk_all_next(blk
)) {
6048 char *allocated_name
= NULL
;
6049 const char *name
= blk_name(blk
);
6052 name
= allocated_name
= blk_get_attached_dev_id(blk
);
6054 xdbg_graph_add_node(gr
, blk
, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_BACKEND
,
6056 g_free(allocated_name
);
6057 if (blk_root(blk
)) {
6058 xdbg_graph_add_edge(gr
, blk
, blk_root(blk
));
6062 WITH_JOB_LOCK_GUARD() {
6063 for (job
= block_job_next_locked(NULL
); job
;
6064 job
= block_job_next_locked(job
)) {
6067 xdbg_graph_add_node(gr
, job
, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_JOB
,
6069 for (el
= job
->nodes
; el
; el
= el
->next
) {
6070 xdbg_graph_add_edge(gr
, job
, (BdrvChild
*)el
->data
);
6075 QTAILQ_FOREACH(bs
, &graph_bdrv_states
, node_list
) {
6076 xdbg_graph_add_node(gr
, bs
, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_DRIVER
,
6078 QLIST_FOREACH(child
, &bs
->children
, next
) {
6079 xdbg_graph_add_edge(gr
, bs
, child
);
6083 return xdbg_graph_finalize(gr
);
6086 BlockDriverState
*bdrv_lookup_bs(const char *device
,
6087 const char *node_name
,
6091 BlockDriverState
*bs
;
6093 GLOBAL_STATE_CODE();
6096 blk
= blk_by_name(device
);
6101 error_setg(errp
, "Device '%s' has no medium", device
);
6109 bs
= bdrv_find_node(node_name
);
6116 error_setg(errp
, "Cannot find device=\'%s\' nor node-name=\'%s\'",
6117 device
? device
: "",
6118 node_name
? node_name
: "");
6122 /* If 'base' is in the same chain as 'top', return true. Otherwise,
6123 * return false. If either argument is NULL, return false. */
6124 bool bdrv_chain_contains(BlockDriverState
*top
, BlockDriverState
*base
)
6127 GLOBAL_STATE_CODE();
6129 while (top
&& top
!= base
) {
6130 top
= bdrv_filter_or_cow_bs(top
);
6136 BlockDriverState
*bdrv_next_node(BlockDriverState
*bs
)
6138 GLOBAL_STATE_CODE();
6140 return QTAILQ_FIRST(&graph_bdrv_states
);
6142 return QTAILQ_NEXT(bs
, node_list
);
6145 BlockDriverState
*bdrv_next_all_states(BlockDriverState
*bs
)
6147 GLOBAL_STATE_CODE();
6149 return QTAILQ_FIRST(&all_bdrv_states
);
6151 return QTAILQ_NEXT(bs
, bs_list
);
6154 const char *bdrv_get_node_name(const BlockDriverState
*bs
)
6157 return bs
->node_name
;
6160 const char *bdrv_get_parent_name(const BlockDriverState
*bs
)
6166 /* If multiple parents have a name, just pick the first one. */
6167 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
6168 if (c
->klass
->get_name
) {
6169 name
= c
->klass
->get_name(c
);
6170 if (name
&& *name
) {
6179 /* TODO check what callers really want: bs->node_name or blk_name() */
6180 const char *bdrv_get_device_name(const BlockDriverState
*bs
)
6183 return bdrv_get_parent_name(bs
) ?: "";
6186 /* This can be used to identify nodes that might not have a device
6187 * name associated. Since node and device names live in the same
6188 * namespace, the result is unambiguous. The exception is if both are
6189 * absent, then this returns an empty (non-null) string. */
6190 const char *bdrv_get_device_or_node_name(const BlockDriverState
*bs
)
6193 return bdrv_get_parent_name(bs
) ?: bs
->node_name
;
6196 int bdrv_get_flags(BlockDriverState
*bs
)
6199 return bs
->open_flags
;
6202 int bdrv_has_zero_init_1(BlockDriverState
*bs
)
6204 GLOBAL_STATE_CODE();
6208 int bdrv_has_zero_init(BlockDriverState
*bs
)
6210 BlockDriverState
*filtered
;
6211 GLOBAL_STATE_CODE();
6217 /* If BS is a copy on write image, it is initialized to
6218 the contents of the base image, which may not be zeroes. */
6219 if (bdrv_cow_child(bs
)) {
6222 if (bs
->drv
->bdrv_has_zero_init
) {
6223 return bs
->drv
->bdrv_has_zero_init(bs
);
6226 filtered
= bdrv_filter_bs(bs
);
6228 return bdrv_has_zero_init(filtered
);
6235 bool bdrv_can_write_zeroes_with_unmap(BlockDriverState
*bs
)
6238 if (!(bs
->open_flags
& BDRV_O_UNMAP
)) {
6242 return bs
->supported_zero_flags
& BDRV_REQ_MAY_UNMAP
;
6245 void bdrv_get_backing_filename(BlockDriverState
*bs
,
6246 char *filename
, int filename_size
)
6249 pstrcpy(filename
, filename_size
, bs
->backing_file
);
6252 int bdrv_get_info(BlockDriverState
*bs
, BlockDriverInfo
*bdi
)
6255 BlockDriver
*drv
= bs
->drv
;
6257 /* if bs->drv == NULL, bs is closed, so there's nothing to do here */
6261 if (!drv
->bdrv_get_info
) {
6262 BlockDriverState
*filtered
= bdrv_filter_bs(bs
);
6264 return bdrv_get_info(filtered
, bdi
);
6268 memset(bdi
, 0, sizeof(*bdi
));
6269 ret
= drv
->bdrv_get_info(bs
, bdi
);
6274 if (bdi
->cluster_size
> BDRV_MAX_ALIGNMENT
) {
6281 ImageInfoSpecific
*bdrv_get_specific_info(BlockDriverState
*bs
,
6284 BlockDriver
*drv
= bs
->drv
;
6286 if (drv
&& drv
->bdrv_get_specific_info
) {
6287 return drv
->bdrv_get_specific_info(bs
, errp
);
6292 BlockStatsSpecific
*bdrv_get_specific_stats(BlockDriverState
*bs
)
6294 BlockDriver
*drv
= bs
->drv
;
6296 if (!drv
|| !drv
->bdrv_get_specific_stats
) {
6299 return drv
->bdrv_get_specific_stats(bs
);
6302 void bdrv_debug_event(BlockDriverState
*bs
, BlkdebugEvent event
)
6305 if (!bs
|| !bs
->drv
|| !bs
->drv
->bdrv_debug_event
) {
6309 bs
->drv
->bdrv_debug_event(bs
, event
);
6312 static BlockDriverState
*bdrv_find_debug_node(BlockDriverState
*bs
)
6314 GLOBAL_STATE_CODE();
6315 while (bs
&& bs
->drv
&& !bs
->drv
->bdrv_debug_breakpoint
) {
6316 bs
= bdrv_primary_bs(bs
);
6319 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_breakpoint
) {
6320 assert(bs
->drv
->bdrv_debug_remove_breakpoint
);
6327 int bdrv_debug_breakpoint(BlockDriverState
*bs
, const char *event
,
6330 GLOBAL_STATE_CODE();
6331 bs
= bdrv_find_debug_node(bs
);
6333 return bs
->drv
->bdrv_debug_breakpoint(bs
, event
, tag
);
6339 int bdrv_debug_remove_breakpoint(BlockDriverState
*bs
, const char *tag
)
6341 GLOBAL_STATE_CODE();
6342 bs
= bdrv_find_debug_node(bs
);
6344 return bs
->drv
->bdrv_debug_remove_breakpoint(bs
, tag
);
6350 int bdrv_debug_resume(BlockDriverState
*bs
, const char *tag
)
6352 GLOBAL_STATE_CODE();
6353 while (bs
&& (!bs
->drv
|| !bs
->drv
->bdrv_debug_resume
)) {
6354 bs
= bdrv_primary_bs(bs
);
6357 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_resume
) {
6358 return bs
->drv
->bdrv_debug_resume(bs
, tag
);
6364 bool bdrv_debug_is_suspended(BlockDriverState
*bs
, const char *tag
)
6366 GLOBAL_STATE_CODE();
6367 while (bs
&& bs
->drv
&& !bs
->drv
->bdrv_debug_is_suspended
) {
6368 bs
= bdrv_primary_bs(bs
);
6371 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_is_suspended
) {
6372 return bs
->drv
->bdrv_debug_is_suspended(bs
, tag
);
6378 /* backing_file can either be relative, or absolute, or a protocol. If it is
6379 * relative, it must be relative to the chain. So, passing in bs->filename
6380 * from a BDS as backing_file should not be done, as that may be relative to
6381 * the CWD rather than the chain. */
6382 BlockDriverState
*bdrv_find_backing_image(BlockDriverState
*bs
,
6383 const char *backing_file
)
6385 char *filename_full
= NULL
;
6386 char *backing_file_full
= NULL
;
6387 char *filename_tmp
= NULL
;
6388 int is_protocol
= 0;
6389 bool filenames_refreshed
= false;
6390 BlockDriverState
*curr_bs
= NULL
;
6391 BlockDriverState
*retval
= NULL
;
6392 BlockDriverState
*bs_below
;
6394 GLOBAL_STATE_CODE();
6396 if (!bs
|| !bs
->drv
|| !backing_file
) {
6400 filename_full
= g_malloc(PATH_MAX
);
6401 backing_file_full
= g_malloc(PATH_MAX
);
6403 is_protocol
= path_has_protocol(backing_file
);
6406 * Being largely a legacy function, skip any filters here
6407 * (because filters do not have normal filenames, so they cannot
6408 * match anyway; and allowing json:{} filenames is a bit out of
6411 for (curr_bs
= bdrv_skip_filters(bs
);
6412 bdrv_cow_child(curr_bs
) != NULL
;
6415 bs_below
= bdrv_backing_chain_next(curr_bs
);
6417 if (bdrv_backing_overridden(curr_bs
)) {
6419 * If the backing file was overridden, we can only compare
6420 * directly against the backing node's filename.
6423 if (!filenames_refreshed
) {
6425 * This will automatically refresh all of the
6426 * filenames in the rest of the backing chain, so we
6427 * only need to do this once.
6429 bdrv_refresh_filename(bs_below
);
6430 filenames_refreshed
= true;
6433 if (strcmp(backing_file
, bs_below
->filename
) == 0) {
6437 } else if (is_protocol
|| path_has_protocol(curr_bs
->backing_file
)) {
6439 * If either of the filename paths is actually a protocol, then
6440 * compare unmodified paths; otherwise make paths relative.
6442 char *backing_file_full_ret
;
6444 if (strcmp(backing_file
, curr_bs
->backing_file
) == 0) {
6448 /* Also check against the full backing filename for the image */
6449 backing_file_full_ret
= bdrv_get_full_backing_filename(curr_bs
,
6451 if (backing_file_full_ret
) {
6452 bool equal
= strcmp(backing_file
, backing_file_full_ret
) == 0;
6453 g_free(backing_file_full_ret
);
6460 /* If not an absolute filename path, make it relative to the current
6461 * image's filename path */
6462 filename_tmp
= bdrv_make_absolute_filename(curr_bs
, backing_file
,
6464 /* We are going to compare canonicalized absolute pathnames */
6465 if (!filename_tmp
|| !realpath(filename_tmp
, filename_full
)) {
6466 g_free(filename_tmp
);
6469 g_free(filename_tmp
);
6471 /* We need to make sure the backing filename we are comparing against
6472 * is relative to the current image filename (or absolute) */
6473 filename_tmp
= bdrv_get_full_backing_filename(curr_bs
, NULL
);
6474 if (!filename_tmp
|| !realpath(filename_tmp
, backing_file_full
)) {
6475 g_free(filename_tmp
);
6478 g_free(filename_tmp
);
6480 if (strcmp(backing_file_full
, filename_full
) == 0) {
6487 g_free(filename_full
);
6488 g_free(backing_file_full
);
6492 void bdrv_init(void)
6494 #ifdef CONFIG_BDRV_WHITELIST_TOOLS
6495 use_bdrv_whitelist
= 1;
6497 module_call_init(MODULE_INIT_BLOCK
);
6500 void bdrv_init_with_whitelist(void)
6502 use_bdrv_whitelist
= 1;
6506 int bdrv_activate(BlockDriverState
*bs
, Error
**errp
)
6508 BdrvChild
*child
, *parent
;
6509 Error
*local_err
= NULL
;
6511 BdrvDirtyBitmap
*bm
;
6513 GLOBAL_STATE_CODE();
6519 QLIST_FOREACH(child
, &bs
->children
, next
) {
6520 bdrv_activate(child
->bs
, &local_err
);
6522 error_propagate(errp
, local_err
);
6528 * Update permissions, they may differ for inactive nodes.
6530 * Note that the required permissions of inactive images are always a
6531 * subset of the permissions required after activating the image. This
6532 * allows us to just get the permissions upfront without restricting
6533 * bdrv_co_invalidate_cache().
6535 * It also means that in error cases, we don't have to try and revert to
6536 * the old permissions (which is an operation that could fail, too). We can
6537 * just keep the extended permissions for the next time that an activation
6538 * of the image is tried.
6540 if (bs
->open_flags
& BDRV_O_INACTIVE
) {
6541 bs
->open_flags
&= ~BDRV_O_INACTIVE
;
6542 ret
= bdrv_refresh_perms(bs
, NULL
, errp
);
6544 bs
->open_flags
|= BDRV_O_INACTIVE
;
6548 ret
= bdrv_invalidate_cache(bs
, errp
);
6550 bs
->open_flags
|= BDRV_O_INACTIVE
;
6554 FOR_EACH_DIRTY_BITMAP(bs
, bm
) {
6555 bdrv_dirty_bitmap_skip_store(bm
, false);
6558 ret
= refresh_total_sectors(bs
, bs
->total_sectors
);
6560 bs
->open_flags
|= BDRV_O_INACTIVE
;
6561 error_setg_errno(errp
, -ret
, "Could not refresh total sector count");
6566 QLIST_FOREACH(parent
, &bs
->parents
, next_parent
) {
6567 if (parent
->klass
->activate
) {
6568 parent
->klass
->activate(parent
, &local_err
);
6570 bs
->open_flags
|= BDRV_O_INACTIVE
;
6571 error_propagate(errp
, local_err
);
6580 int coroutine_fn
bdrv_co_invalidate_cache(BlockDriverState
*bs
, Error
**errp
)
6582 Error
*local_err
= NULL
;
6585 assert(!(bs
->open_flags
& BDRV_O_INACTIVE
));
6587 if (bs
->drv
->bdrv_co_invalidate_cache
) {
6588 bs
->drv
->bdrv_co_invalidate_cache(bs
, &local_err
);
6590 error_propagate(errp
, local_err
);
6598 void bdrv_activate_all(Error
**errp
)
6600 BlockDriverState
*bs
;
6601 BdrvNextIterator it
;
6603 GLOBAL_STATE_CODE();
6605 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
6606 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
6609 aio_context_acquire(aio_context
);
6610 ret
= bdrv_activate(bs
, errp
);
6611 aio_context_release(aio_context
);
6613 bdrv_next_cleanup(&it
);
6619 static bool bdrv_has_bds_parent(BlockDriverState
*bs
, bool only_active
)
6622 GLOBAL_STATE_CODE();
6624 QLIST_FOREACH(parent
, &bs
->parents
, next_parent
) {
6625 if (parent
->klass
->parent_is_bds
) {
6626 BlockDriverState
*parent_bs
= parent
->opaque
;
6627 if (!only_active
|| !(parent_bs
->open_flags
& BDRV_O_INACTIVE
)) {
6636 static int bdrv_inactivate_recurse(BlockDriverState
*bs
)
6638 BdrvChild
*child
, *parent
;
6640 uint64_t cumulative_perms
, cumulative_shared_perms
;
6642 GLOBAL_STATE_CODE();
6648 /* Make sure that we don't inactivate a child before its parent.
6649 * It will be covered by recursion from the yet active parent. */
6650 if (bdrv_has_bds_parent(bs
, true)) {
6654 assert(!(bs
->open_flags
& BDRV_O_INACTIVE
));
6656 /* Inactivate this node */
6657 if (bs
->drv
->bdrv_inactivate
) {
6658 ret
= bs
->drv
->bdrv_inactivate(bs
);
6664 QLIST_FOREACH(parent
, &bs
->parents
, next_parent
) {
6665 if (parent
->klass
->inactivate
) {
6666 ret
= parent
->klass
->inactivate(parent
);
6673 bdrv_get_cumulative_perm(bs
, &cumulative_perms
,
6674 &cumulative_shared_perms
);
6675 if (cumulative_perms
& (BLK_PERM_WRITE
| BLK_PERM_WRITE_UNCHANGED
)) {
6676 /* Our inactive parents still need write access. Inactivation failed. */
6680 bs
->open_flags
|= BDRV_O_INACTIVE
;
6683 * Update permissions, they may differ for inactive nodes.
6684 * We only tried to loosen restrictions, so errors are not fatal, ignore
6687 bdrv_refresh_perms(bs
, NULL
, NULL
);
6689 /* Recursively inactivate children */
6690 QLIST_FOREACH(child
, &bs
->children
, next
) {
6691 ret
= bdrv_inactivate_recurse(child
->bs
);
6700 int bdrv_inactivate_all(void)
6702 BlockDriverState
*bs
= NULL
;
6703 BdrvNextIterator it
;
6705 GSList
*aio_ctxs
= NULL
, *ctx
;
6707 GLOBAL_STATE_CODE();
6709 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
6710 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
6712 if (!g_slist_find(aio_ctxs
, aio_context
)) {
6713 aio_ctxs
= g_slist_prepend(aio_ctxs
, aio_context
);
6714 aio_context_acquire(aio_context
);
6718 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
6719 /* Nodes with BDS parents are covered by recursion from the last
6720 * parent that gets inactivated. Don't inactivate them a second
6721 * time if that has already happened. */
6722 if (bdrv_has_bds_parent(bs
, false)) {
6725 ret
= bdrv_inactivate_recurse(bs
);
6727 bdrv_next_cleanup(&it
);
6733 for (ctx
= aio_ctxs
; ctx
!= NULL
; ctx
= ctx
->next
) {
6734 AioContext
*aio_context
= ctx
->data
;
6735 aio_context_release(aio_context
);
6737 g_slist_free(aio_ctxs
);
6742 /**************************************************************/
6743 /* removable device support */
6746 * Return TRUE if the media is present
6748 bool bdrv_is_inserted(BlockDriverState
*bs
)
6750 BlockDriver
*drv
= bs
->drv
;
6757 if (drv
->bdrv_is_inserted
) {
6758 return drv
->bdrv_is_inserted(bs
);
6760 QLIST_FOREACH(child
, &bs
->children
, next
) {
6761 if (!bdrv_is_inserted(child
->bs
)) {
6769 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
6771 void bdrv_eject(BlockDriverState
*bs
, bool eject_flag
)
6773 BlockDriver
*drv
= bs
->drv
;
6776 if (drv
&& drv
->bdrv_eject
) {
6777 drv
->bdrv_eject(bs
, eject_flag
);
6782 * Lock or unlock the media (if it is locked, the user won't be able
6783 * to eject it manually).
6785 void bdrv_lock_medium(BlockDriverState
*bs
, bool locked
)
6787 BlockDriver
*drv
= bs
->drv
;
6789 trace_bdrv_lock_medium(bs
, locked
);
6791 if (drv
&& drv
->bdrv_lock_medium
) {
6792 drv
->bdrv_lock_medium(bs
, locked
);
6796 /* Get a reference to bs */
6797 void bdrv_ref(BlockDriverState
*bs
)
6799 GLOBAL_STATE_CODE();
6803 /* Release a previously grabbed reference to bs.
6804 * If after releasing, reference count is zero, the BlockDriverState is
6806 void bdrv_unref(BlockDriverState
*bs
)
6808 GLOBAL_STATE_CODE();
6812 assert(bs
->refcnt
> 0);
6813 if (--bs
->refcnt
== 0) {
6818 struct BdrvOpBlocker
{
6820 QLIST_ENTRY(BdrvOpBlocker
) list
;
6823 bool bdrv_op_is_blocked(BlockDriverState
*bs
, BlockOpType op
, Error
**errp
)
6825 BdrvOpBlocker
*blocker
;
6826 GLOBAL_STATE_CODE();
6827 assert((int) op
>= 0 && op
< BLOCK_OP_TYPE_MAX
);
6828 if (!QLIST_EMPTY(&bs
->op_blockers
[op
])) {
6829 blocker
= QLIST_FIRST(&bs
->op_blockers
[op
]);
6830 error_propagate_prepend(errp
, error_copy(blocker
->reason
),
6831 "Node '%s' is busy: ",
6832 bdrv_get_device_or_node_name(bs
));
6838 void bdrv_op_block(BlockDriverState
*bs
, BlockOpType op
, Error
*reason
)
6840 BdrvOpBlocker
*blocker
;
6841 GLOBAL_STATE_CODE();
6842 assert((int) op
>= 0 && op
< BLOCK_OP_TYPE_MAX
);
6844 blocker
= g_new0(BdrvOpBlocker
, 1);
6845 blocker
->reason
= reason
;
6846 QLIST_INSERT_HEAD(&bs
->op_blockers
[op
], blocker
, list
);
6849 void bdrv_op_unblock(BlockDriverState
*bs
, BlockOpType op
, Error
*reason
)
6851 BdrvOpBlocker
*blocker
, *next
;
6852 GLOBAL_STATE_CODE();
6853 assert((int) op
>= 0 && op
< BLOCK_OP_TYPE_MAX
);
6854 QLIST_FOREACH_SAFE(blocker
, &bs
->op_blockers
[op
], list
, next
) {
6855 if (blocker
->reason
== reason
) {
6856 QLIST_REMOVE(blocker
, list
);
6862 void bdrv_op_block_all(BlockDriverState
*bs
, Error
*reason
)
6865 GLOBAL_STATE_CODE();
6866 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
6867 bdrv_op_block(bs
, i
, reason
);
6871 void bdrv_op_unblock_all(BlockDriverState
*bs
, Error
*reason
)
6874 GLOBAL_STATE_CODE();
6875 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
6876 bdrv_op_unblock(bs
, i
, reason
);
6880 bool bdrv_op_blocker_is_empty(BlockDriverState
*bs
)
6883 GLOBAL_STATE_CODE();
6884 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
6885 if (!QLIST_EMPTY(&bs
->op_blockers
[i
])) {
6892 void bdrv_img_create(const char *filename
, const char *fmt
,
6893 const char *base_filename
, const char *base_fmt
,
6894 char *options
, uint64_t img_size
, int flags
, bool quiet
,
6897 QemuOptsList
*create_opts
= NULL
;
6898 QemuOpts
*opts
= NULL
;
6899 const char *backing_fmt
, *backing_file
;
6901 BlockDriver
*drv
, *proto_drv
;
6902 Error
*local_err
= NULL
;
6905 GLOBAL_STATE_CODE();
6907 /* Find driver and parse its options */
6908 drv
= bdrv_find_format(fmt
);
6910 error_setg(errp
, "Unknown file format '%s'", fmt
);
6914 proto_drv
= bdrv_find_protocol(filename
, true, errp
);
6919 if (!drv
->create_opts
) {
6920 error_setg(errp
, "Format driver '%s' does not support image creation",
6925 if (!proto_drv
->create_opts
) {
6926 error_setg(errp
, "Protocol driver '%s' does not support image creation",
6927 proto_drv
->format_name
);
6931 /* Create parameter list */
6932 create_opts
= qemu_opts_append(create_opts
, drv
->create_opts
);
6933 create_opts
= qemu_opts_append(create_opts
, proto_drv
->create_opts
);
6935 opts
= qemu_opts_create(create_opts
, NULL
, 0, &error_abort
);
6937 /* Parse -o options */
6939 if (!qemu_opts_do_parse(opts
, options
, NULL
, errp
)) {
6944 if (!qemu_opt_get(opts
, BLOCK_OPT_SIZE
)) {
6945 qemu_opt_set_number(opts
, BLOCK_OPT_SIZE
, img_size
, &error_abort
);
6946 } else if (img_size
!= UINT64_C(-1)) {
6947 error_setg(errp
, "The image size must be specified only once");
6951 if (base_filename
) {
6952 if (!qemu_opt_set(opts
, BLOCK_OPT_BACKING_FILE
, base_filename
,
6954 error_setg(errp
, "Backing file not supported for file format '%s'",
6961 if (!qemu_opt_set(opts
, BLOCK_OPT_BACKING_FMT
, base_fmt
, NULL
)) {
6962 error_setg(errp
, "Backing file format not supported for file "
6963 "format '%s'", fmt
);
6968 backing_file
= qemu_opt_get(opts
, BLOCK_OPT_BACKING_FILE
);
6970 if (!strcmp(filename
, backing_file
)) {
6971 error_setg(errp
, "Error: Trying to create an image with the "
6972 "same filename as the backing file");
6975 if (backing_file
[0] == '\0') {
6976 error_setg(errp
, "Expected backing file name, got empty string");
6981 backing_fmt
= qemu_opt_get(opts
, BLOCK_OPT_BACKING_FMT
);
6983 /* The size for the image must always be specified, unless we have a backing
6984 * file and we have not been forbidden from opening it. */
6985 size
= qemu_opt_get_size(opts
, BLOCK_OPT_SIZE
, img_size
);
6986 if (backing_file
&& !(flags
& BDRV_O_NO_BACKING
)) {
6987 BlockDriverState
*bs
;
6990 QDict
*backing_options
= NULL
;
6993 bdrv_get_full_backing_filename_from_filename(filename
, backing_file
,
6998 assert(full_backing
);
7001 * No need to do I/O here, which allows us to open encrypted
7002 * backing images without needing the secret
7005 back_flags
&= ~(BDRV_O_RDWR
| BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
);
7006 back_flags
|= BDRV_O_NO_IO
;
7008 backing_options
= qdict_new();
7010 qdict_put_str(backing_options
, "driver", backing_fmt
);
7012 qdict_put_bool(backing_options
, BDRV_OPT_FORCE_SHARE
, true);
7014 bs
= bdrv_open(full_backing
, NULL
, backing_options
, back_flags
,
7016 g_free(full_backing
);
7018 error_append_hint(&local_err
, "Could not open backing image.\n");
7022 error_setg(&local_err
,
7023 "Backing file specified without backing format");
7024 error_append_hint(&local_err
, "Detected format of %s.",
7025 bs
->drv
->format_name
);
7029 /* Opened BS, have no size */
7030 size
= bdrv_getlength(bs
);
7032 error_setg_errno(errp
, -size
, "Could not get size of '%s'",
7037 qemu_opt_set_number(opts
, BLOCK_OPT_SIZE
, size
, &error_abort
);
7041 /* (backing_file && !(flags & BDRV_O_NO_BACKING)) */
7042 } else if (backing_file
&& !backing_fmt
) {
7043 error_setg(&local_err
,
7044 "Backing file specified without backing format");
7049 error_setg(errp
, "Image creation needs a size parameter");
7054 printf("Formatting '%s', fmt=%s ", filename
, fmt
);
7055 qemu_opts_print(opts
, " ");
7060 ret
= bdrv_create(drv
, filename
, opts
, &local_err
);
7062 if (ret
== -EFBIG
) {
7063 /* This is generally a better message than whatever the driver would
7064 * deliver (especially because of the cluster_size_hint), since that
7065 * is most probably not much different from "image too large". */
7066 const char *cluster_size_hint
= "";
7067 if (qemu_opt_get_size(opts
, BLOCK_OPT_CLUSTER_SIZE
, 0)) {
7068 cluster_size_hint
= " (try using a larger cluster size)";
7070 error_setg(errp
, "The image size is too large for file format '%s'"
7071 "%s", fmt
, cluster_size_hint
);
7072 error_free(local_err
);
7077 qemu_opts_del(opts
);
7078 qemu_opts_free(create_opts
);
7079 error_propagate(errp
, local_err
);
7082 AioContext
*bdrv_get_aio_context(BlockDriverState
*bs
)
7085 return bs
? bs
->aio_context
: qemu_get_aio_context();
7088 AioContext
*coroutine_fn
bdrv_co_enter(BlockDriverState
*bs
)
7090 Coroutine
*self
= qemu_coroutine_self();
7091 AioContext
*old_ctx
= qemu_coroutine_get_aio_context(self
);
7092 AioContext
*new_ctx
;
7096 * Increase bs->in_flight to ensure that this operation is completed before
7097 * moving the node to a different AioContext. Read new_ctx only afterwards.
7099 bdrv_inc_in_flight(bs
);
7101 new_ctx
= bdrv_get_aio_context(bs
);
7102 aio_co_reschedule_self(new_ctx
);
7106 void coroutine_fn
bdrv_co_leave(BlockDriverState
*bs
, AioContext
*old_ctx
)
7109 aio_co_reschedule_self(old_ctx
);
7110 bdrv_dec_in_flight(bs
);
7113 void coroutine_fn
bdrv_co_lock(BlockDriverState
*bs
)
7115 AioContext
*ctx
= bdrv_get_aio_context(bs
);
7117 /* In the main thread, bs->aio_context won't change concurrently */
7118 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
7121 * We're in coroutine context, so we already hold the lock of the main
7122 * loop AioContext. Don't lock it twice to avoid deadlocks.
7124 assert(qemu_in_coroutine());
7125 if (ctx
!= qemu_get_aio_context()) {
7126 aio_context_acquire(ctx
);
7130 void coroutine_fn
bdrv_co_unlock(BlockDriverState
*bs
)
7132 AioContext
*ctx
= bdrv_get_aio_context(bs
);
7134 assert(qemu_in_coroutine());
7135 if (ctx
!= qemu_get_aio_context()) {
7136 aio_context_release(ctx
);
7140 void bdrv_coroutine_enter(BlockDriverState
*bs
, Coroutine
*co
)
7143 aio_co_enter(bdrv_get_aio_context(bs
), co
);
7146 static void bdrv_do_remove_aio_context_notifier(BdrvAioNotifier
*ban
)
7148 GLOBAL_STATE_CODE();
7149 QLIST_REMOVE(ban
, list
);
7153 static void bdrv_detach_aio_context(BlockDriverState
*bs
)
7155 BdrvAioNotifier
*baf
, *baf_tmp
;
7157 assert(!bs
->walking_aio_notifiers
);
7158 GLOBAL_STATE_CODE();
7159 bs
->walking_aio_notifiers
= true;
7160 QLIST_FOREACH_SAFE(baf
, &bs
->aio_notifiers
, list
, baf_tmp
) {
7162 bdrv_do_remove_aio_context_notifier(baf
);
7164 baf
->detach_aio_context(baf
->opaque
);
7167 /* Never mind iterating again to check for ->deleted. bdrv_close() will
7168 * remove remaining aio notifiers if we aren't called again.
7170 bs
->walking_aio_notifiers
= false;
7172 if (bs
->drv
&& bs
->drv
->bdrv_detach_aio_context
) {
7173 bs
->drv
->bdrv_detach_aio_context(bs
);
7176 if (bs
->quiesce_counter
) {
7177 aio_enable_external(bs
->aio_context
);
7179 assert_bdrv_graph_writable(bs
);
7180 bs
->aio_context
= NULL
;
7183 static void bdrv_attach_aio_context(BlockDriverState
*bs
,
7184 AioContext
*new_context
)
7186 BdrvAioNotifier
*ban
, *ban_tmp
;
7187 GLOBAL_STATE_CODE();
7189 if (bs
->quiesce_counter
) {
7190 aio_disable_external(new_context
);
7193 assert_bdrv_graph_writable(bs
);
7194 bs
->aio_context
= new_context
;
7196 if (bs
->drv
&& bs
->drv
->bdrv_attach_aio_context
) {
7197 bs
->drv
->bdrv_attach_aio_context(bs
, new_context
);
7200 assert(!bs
->walking_aio_notifiers
);
7201 bs
->walking_aio_notifiers
= true;
7202 QLIST_FOREACH_SAFE(ban
, &bs
->aio_notifiers
, list
, ban_tmp
) {
7204 bdrv_do_remove_aio_context_notifier(ban
);
7206 ban
->attached_aio_context(new_context
, ban
->opaque
);
7209 bs
->walking_aio_notifiers
= false;
7212 typedef struct BdrvStateSetAioContext
{
7213 AioContext
*new_ctx
;
7214 BlockDriverState
*bs
;
7215 } BdrvStateSetAioContext
;
7217 static bool bdrv_parent_change_aio_context(BdrvChild
*c
, AioContext
*ctx
,
7218 GHashTable
*visited
,
7222 GLOBAL_STATE_CODE();
7223 if (g_hash_table_contains(visited
, c
)) {
7226 g_hash_table_add(visited
, c
);
7229 * A BdrvChildClass that doesn't handle AioContext changes cannot
7230 * tolerate any AioContext changes
7232 if (!c
->klass
->change_aio_ctx
) {
7233 char *user
= bdrv_child_user_desc(c
);
7234 error_setg(errp
, "Changing iothreads is not supported by %s", user
);
7238 if (!c
->klass
->change_aio_ctx(c
, ctx
, visited
, tran
, errp
)) {
7239 assert(!errp
|| *errp
);
7245 bool bdrv_child_change_aio_context(BdrvChild
*c
, AioContext
*ctx
,
7246 GHashTable
*visited
, Transaction
*tran
,
7249 GLOBAL_STATE_CODE();
7250 if (g_hash_table_contains(visited
, c
)) {
7253 g_hash_table_add(visited
, c
);
7254 return bdrv_change_aio_context(c
->bs
, ctx
, visited
, tran
, errp
);
7257 static void bdrv_set_aio_context_clean(void *opaque
)
7259 BdrvStateSetAioContext
*state
= (BdrvStateSetAioContext
*) opaque
;
7260 BlockDriverState
*bs
= (BlockDriverState
*) state
->bs
;
7262 /* Paired with bdrv_drained_begin in bdrv_change_aio_context() */
7263 bdrv_drained_end(bs
);
7268 static void bdrv_set_aio_context_commit(void *opaque
)
7270 BdrvStateSetAioContext
*state
= (BdrvStateSetAioContext
*) opaque
;
7271 BlockDriverState
*bs
= (BlockDriverState
*) state
->bs
;
7272 AioContext
*new_context
= state
->new_ctx
;
7273 AioContext
*old_context
= bdrv_get_aio_context(bs
);
7274 assert_bdrv_graph_writable(bs
);
7277 * Take the old AioContex when detaching it from bs.
7278 * At this point, new_context lock is already acquired, and we are now
7279 * also taking old_context. This is safe as long as bdrv_detach_aio_context
7280 * does not call AIO_POLL_WHILE().
7282 if (old_context
!= qemu_get_aio_context()) {
7283 aio_context_acquire(old_context
);
7285 bdrv_detach_aio_context(bs
);
7286 if (old_context
!= qemu_get_aio_context()) {
7287 aio_context_release(old_context
);
7289 bdrv_attach_aio_context(bs
, new_context
);
7292 static TransactionActionDrv set_aio_context
= {
7293 .commit
= bdrv_set_aio_context_commit
,
7294 .clean
= bdrv_set_aio_context_clean
,
7298 * Changes the AioContext used for fd handlers, timers, and BHs by this
7299 * BlockDriverState and all its children and parents.
7301 * Must be called from the main AioContext.
7303 * The caller must own the AioContext lock for the old AioContext of bs, but it
7304 * must not own the AioContext lock for new_context (unless new_context is the
7305 * same as the current context of bs).
7307 * @visited will accumulate all visited BdrvChild objects. The caller is
7308 * responsible for freeing the list afterwards.
7310 static bool bdrv_change_aio_context(BlockDriverState
*bs
, AioContext
*ctx
,
7311 GHashTable
*visited
, Transaction
*tran
,
7315 BdrvStateSetAioContext
*state
;
7317 GLOBAL_STATE_CODE();
7319 if (bdrv_get_aio_context(bs
) == ctx
) {
7323 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
7324 if (!bdrv_parent_change_aio_context(c
, ctx
, visited
, tran
, errp
)) {
7329 QLIST_FOREACH(c
, &bs
->children
, next
) {
7330 if (!bdrv_child_change_aio_context(c
, ctx
, visited
, tran
, errp
)) {
7335 state
= g_new(BdrvStateSetAioContext
, 1);
7336 *state
= (BdrvStateSetAioContext
) {
7341 /* Paired with bdrv_drained_end in bdrv_set_aio_context_clean() */
7342 bdrv_drained_begin(bs
);
7344 tran_add(tran
, &set_aio_context
, state
);
7350 * Change bs's and recursively all of its parents' and children's AioContext
7351 * to the given new context, returning an error if that isn't possible.
7353 * If ignore_child is not NULL, that child (and its subgraph) will not
7356 * This function still requires the caller to take the bs current
7357 * AioContext lock, otherwise draining will fail since AIO_WAIT_WHILE
7358 * assumes the lock is always held if bs is in another AioContext.
7359 * For the same reason, it temporarily also holds the new AioContext, since
7360 * bdrv_drained_end calls BDRV_POLL_WHILE that assumes the lock is taken too.
7361 * Therefore the new AioContext lock must not be taken by the caller.
7363 int bdrv_try_change_aio_context(BlockDriverState
*bs
, AioContext
*ctx
,
7364 BdrvChild
*ignore_child
, Error
**errp
)
7367 GHashTable
*visited
;
7369 AioContext
*old_context
= bdrv_get_aio_context(bs
);
7370 GLOBAL_STATE_CODE();
7373 * Recursion phase: go through all nodes of the graph.
7374 * Take care of checking that all nodes support changing AioContext
7375 * and drain them, builing a linear list of callbacks to run if everything
7376 * is successful (the transaction itself).
7379 visited
= g_hash_table_new(NULL
, NULL
);
7381 g_hash_table_add(visited
, ignore_child
);
7383 ret
= bdrv_change_aio_context(bs
, ctx
, visited
, tran
, errp
);
7384 g_hash_table_destroy(visited
);
7387 * Linear phase: go through all callbacks collected in the transaction.
7388 * Run all callbacks collected in the recursion to switch all nodes
7389 * AioContext lock (transaction commit), or undo all changes done in the
7390 * recursion (transaction abort).
7394 /* Just run clean() callbacks. No AioContext changed. */
7400 * Release old AioContext, it won't be needed anymore, as all
7401 * bdrv_drained_begin() have been called already.
7403 if (qemu_get_aio_context() != old_context
) {
7404 aio_context_release(old_context
);
7408 * Acquire new AioContext since bdrv_drained_end() is going to be called
7409 * after we switched all nodes in the new AioContext, and the function
7410 * assumes that the lock of the bs is always taken.
7412 if (qemu_get_aio_context() != ctx
) {
7413 aio_context_acquire(ctx
);
7418 if (qemu_get_aio_context() != ctx
) {
7419 aio_context_release(ctx
);
7422 /* Re-acquire the old AioContext, since the caller takes and releases it. */
7423 if (qemu_get_aio_context() != old_context
) {
7424 aio_context_acquire(old_context
);
7430 void bdrv_add_aio_context_notifier(BlockDriverState
*bs
,
7431 void (*attached_aio_context
)(AioContext
*new_context
, void *opaque
),
7432 void (*detach_aio_context
)(void *opaque
), void *opaque
)
7434 BdrvAioNotifier
*ban
= g_new(BdrvAioNotifier
, 1);
7435 *ban
= (BdrvAioNotifier
){
7436 .attached_aio_context
= attached_aio_context
,
7437 .detach_aio_context
= detach_aio_context
,
7440 GLOBAL_STATE_CODE();
7442 QLIST_INSERT_HEAD(&bs
->aio_notifiers
, ban
, list
);
7445 void bdrv_remove_aio_context_notifier(BlockDriverState
*bs
,
7446 void (*attached_aio_context
)(AioContext
*,
7448 void (*detach_aio_context
)(void *),
7451 BdrvAioNotifier
*ban
, *ban_next
;
7452 GLOBAL_STATE_CODE();
7454 QLIST_FOREACH_SAFE(ban
, &bs
->aio_notifiers
, list
, ban_next
) {
7455 if (ban
->attached_aio_context
== attached_aio_context
&&
7456 ban
->detach_aio_context
== detach_aio_context
&&
7457 ban
->opaque
== opaque
&&
7458 ban
->deleted
== false)
7460 if (bs
->walking_aio_notifiers
) {
7461 ban
->deleted
= true;
7463 bdrv_do_remove_aio_context_notifier(ban
);
7472 int bdrv_amend_options(BlockDriverState
*bs
, QemuOpts
*opts
,
7473 BlockDriverAmendStatusCB
*status_cb
, void *cb_opaque
,
7477 GLOBAL_STATE_CODE();
7479 error_setg(errp
, "Node is ejected");
7482 if (!bs
->drv
->bdrv_amend_options
) {
7483 error_setg(errp
, "Block driver '%s' does not support option amendment",
7484 bs
->drv
->format_name
);
7487 return bs
->drv
->bdrv_amend_options(bs
, opts
, status_cb
,
7488 cb_opaque
, force
, errp
);
7492 * This function checks whether the given @to_replace is allowed to be
7493 * replaced by a node that always shows the same data as @bs. This is
7494 * used for example to verify whether the mirror job can replace
7495 * @to_replace by the target mirrored from @bs.
7496 * To be replaceable, @bs and @to_replace may either be guaranteed to
7497 * always show the same data (because they are only connected through
7498 * filters), or some driver may allow replacing one of its children
7499 * because it can guarantee that this child's data is not visible at
7500 * all (for example, for dissenting quorum children that have no other
7503 bool bdrv_recurse_can_replace(BlockDriverState
*bs
,
7504 BlockDriverState
*to_replace
)
7506 BlockDriverState
*filtered
;
7508 GLOBAL_STATE_CODE();
7510 if (!bs
|| !bs
->drv
) {
7514 if (bs
== to_replace
) {
7518 /* See what the driver can do */
7519 if (bs
->drv
->bdrv_recurse_can_replace
) {
7520 return bs
->drv
->bdrv_recurse_can_replace(bs
, to_replace
);
7523 /* For filters without an own implementation, we can recurse on our own */
7524 filtered
= bdrv_filter_bs(bs
);
7526 return bdrv_recurse_can_replace(filtered
, to_replace
);
7534 * Check whether the given @node_name can be replaced by a node that
7535 * has the same data as @parent_bs. If so, return @node_name's BDS;
7538 * @node_name must be a (recursive) *child of @parent_bs (or this
7539 * function will return NULL).
7541 * The result (whether the node can be replaced or not) is only valid
7542 * for as long as no graph or permission changes occur.
7544 BlockDriverState
*check_to_replace_node(BlockDriverState
*parent_bs
,
7545 const char *node_name
, Error
**errp
)
7547 BlockDriverState
*to_replace_bs
= bdrv_find_node(node_name
);
7548 AioContext
*aio_context
;
7550 GLOBAL_STATE_CODE();
7552 if (!to_replace_bs
) {
7553 error_setg(errp
, "Failed to find node with node-name='%s'", node_name
);
7557 aio_context
= bdrv_get_aio_context(to_replace_bs
);
7558 aio_context_acquire(aio_context
);
7560 if (bdrv_op_is_blocked(to_replace_bs
, BLOCK_OP_TYPE_REPLACE
, errp
)) {
7561 to_replace_bs
= NULL
;
7565 /* We don't want arbitrary node of the BDS chain to be replaced only the top
7566 * most non filter in order to prevent data corruption.
7567 * Another benefit is that this tests exclude backing files which are
7568 * blocked by the backing blockers.
7570 if (!bdrv_recurse_can_replace(parent_bs
, to_replace_bs
)) {
7571 error_setg(errp
, "Cannot replace '%s' by a node mirrored from '%s', "
7572 "because it cannot be guaranteed that doing so would not "
7573 "lead to an abrupt change of visible data",
7574 node_name
, parent_bs
->node_name
);
7575 to_replace_bs
= NULL
;
7580 aio_context_release(aio_context
);
7581 return to_replace_bs
;
7585 * Iterates through the list of runtime option keys that are said to
7586 * be "strong" for a BDS. An option is called "strong" if it changes
7587 * a BDS's data. For example, the null block driver's "size" and
7588 * "read-zeroes" options are strong, but its "latency-ns" option is
7591 * If a key returned by this function ends with a dot, all options
7592 * starting with that prefix are strong.
7594 static const char *const *strong_options(BlockDriverState
*bs
,
7595 const char *const *curopt
)
7597 static const char *const global_options
[] = {
7598 "driver", "filename", NULL
7602 return &global_options
[0];
7606 if (curopt
== &global_options
[ARRAY_SIZE(global_options
) - 1] && bs
->drv
) {
7607 curopt
= bs
->drv
->strong_runtime_opts
;
7610 return (curopt
&& *curopt
) ? curopt
: NULL
;
7614 * Copies all strong runtime options from bs->options to the given
7615 * QDict. The set of strong option keys is determined by invoking
7618 * Returns true iff any strong option was present in bs->options (and
7619 * thus copied to the target QDict) with the exception of "filename"
7620 * and "driver". The caller is expected to use this value to decide
7621 * whether the existence of strong options prevents the generation of
7624 static bool append_strong_runtime_options(QDict
*d
, BlockDriverState
*bs
)
7626 bool found_any
= false;
7627 const char *const *option_name
= NULL
;
7633 while ((option_name
= strong_options(bs
, option_name
))) {
7634 bool option_given
= false;
7636 assert(strlen(*option_name
) > 0);
7637 if ((*option_name
)[strlen(*option_name
) - 1] != '.') {
7638 QObject
*entry
= qdict_get(bs
->options
, *option_name
);
7643 qdict_put_obj(d
, *option_name
, qobject_ref(entry
));
7644 option_given
= true;
7646 const QDictEntry
*entry
;
7647 for (entry
= qdict_first(bs
->options
); entry
;
7648 entry
= qdict_next(bs
->options
, entry
))
7650 if (strstart(qdict_entry_key(entry
), *option_name
, NULL
)) {
7651 qdict_put_obj(d
, qdict_entry_key(entry
),
7652 qobject_ref(qdict_entry_value(entry
)));
7653 option_given
= true;
7658 /* While "driver" and "filename" need to be included in a JSON filename,
7659 * their existence does not prohibit generation of a plain filename. */
7660 if (!found_any
&& option_given
&&
7661 strcmp(*option_name
, "driver") && strcmp(*option_name
, "filename"))
7667 if (!qdict_haskey(d
, "driver")) {
7668 /* Drivers created with bdrv_new_open_driver() may not have a
7669 * @driver option. Add it here. */
7670 qdict_put_str(d
, "driver", bs
->drv
->format_name
);
7676 /* Note: This function may return false positives; it may return true
7677 * even if opening the backing file specified by bs's image header
7678 * would result in exactly bs->backing. */
7679 static bool bdrv_backing_overridden(BlockDriverState
*bs
)
7681 GLOBAL_STATE_CODE();
7683 return strcmp(bs
->auto_backing_file
,
7684 bs
->backing
->bs
->filename
);
7686 /* No backing BDS, so if the image header reports any backing
7687 * file, it must have been suppressed */
7688 return bs
->auto_backing_file
[0] != '\0';
7692 /* Updates the following BDS fields:
7693 * - exact_filename: A filename which may be used for opening a block device
7694 * which (mostly) equals the given BDS (even without any
7695 * other options; so reading and writing must return the same
7696 * results, but caching etc. may be different)
7697 * - full_open_options: Options which, when given when opening a block device
7698 * (without a filename), result in a BDS (mostly)
7699 * equalling the given one
7700 * - filename: If exact_filename is set, it is copied here. Otherwise,
7701 * full_open_options is converted to a JSON object, prefixed with
7702 * "json:" (for use through the JSON pseudo protocol) and put here.
7704 void bdrv_refresh_filename(BlockDriverState
*bs
)
7706 BlockDriver
*drv
= bs
->drv
;
7708 BlockDriverState
*primary_child_bs
;
7710 bool backing_overridden
;
7711 bool generate_json_filename
; /* Whether our default implementation should
7712 fill exact_filename (false) or not (true) */
7714 GLOBAL_STATE_CODE();
7720 /* This BDS's file name may depend on any of its children's file names, so
7721 * refresh those first */
7722 QLIST_FOREACH(child
, &bs
->children
, next
) {
7723 bdrv_refresh_filename(child
->bs
);
7727 /* For implicit nodes, just copy everything from the single child */
7728 child
= QLIST_FIRST(&bs
->children
);
7729 assert(QLIST_NEXT(child
, next
) == NULL
);
7731 pstrcpy(bs
->exact_filename
, sizeof(bs
->exact_filename
),
7732 child
->bs
->exact_filename
);
7733 pstrcpy(bs
->filename
, sizeof(bs
->filename
), child
->bs
->filename
);
7735 qobject_unref(bs
->full_open_options
);
7736 bs
->full_open_options
= qobject_ref(child
->bs
->full_open_options
);
7741 backing_overridden
= bdrv_backing_overridden(bs
);
7743 if (bs
->open_flags
& BDRV_O_NO_IO
) {
7744 /* Without I/O, the backing file does not change anything.
7745 * Therefore, in such a case (primarily qemu-img), we can
7746 * pretend the backing file has not been overridden even if
7747 * it technically has been. */
7748 backing_overridden
= false;
7751 /* Gather the options QDict */
7753 generate_json_filename
= append_strong_runtime_options(opts
, bs
);
7754 generate_json_filename
|= backing_overridden
;
7756 if (drv
->bdrv_gather_child_options
) {
7757 /* Some block drivers may not want to present all of their children's
7758 * options, or name them differently from BdrvChild.name */
7759 drv
->bdrv_gather_child_options(bs
, opts
, backing_overridden
);
7761 QLIST_FOREACH(child
, &bs
->children
, next
) {
7762 if (child
== bs
->backing
&& !backing_overridden
) {
7763 /* We can skip the backing BDS if it has not been overridden */
7767 qdict_put(opts
, child
->name
,
7768 qobject_ref(child
->bs
->full_open_options
));
7771 if (backing_overridden
&& !bs
->backing
) {
7772 /* Force no backing file */
7773 qdict_put_null(opts
, "backing");
7777 qobject_unref(bs
->full_open_options
);
7778 bs
->full_open_options
= opts
;
7780 primary_child_bs
= bdrv_primary_bs(bs
);
7782 if (drv
->bdrv_refresh_filename
) {
7783 /* Obsolete information is of no use here, so drop the old file name
7784 * information before refreshing it */
7785 bs
->exact_filename
[0] = '\0';
7787 drv
->bdrv_refresh_filename(bs
);
7788 } else if (primary_child_bs
) {
7790 * Try to reconstruct valid information from the underlying
7791 * file -- this only works for format nodes (filter nodes
7792 * cannot be probed and as such must be selected by the user
7793 * either through an options dict, or through a special
7794 * filename which the filter driver must construct in its
7795 * .bdrv_refresh_filename() implementation).
7798 bs
->exact_filename
[0] = '\0';
7801 * We can use the underlying file's filename if:
7802 * - it has a filename,
7803 * - the current BDS is not a filter,
7804 * - the file is a protocol BDS, and
7805 * - opening that file (as this BDS's format) will automatically create
7806 * the BDS tree we have right now, that is:
7807 * - the user did not significantly change this BDS's behavior with
7808 * some explicit (strong) options
7809 * - no non-file child of this BDS has been overridden by the user
7810 * Both of these conditions are represented by generate_json_filename.
7812 if (primary_child_bs
->exact_filename
[0] &&
7813 primary_child_bs
->drv
->bdrv_file_open
&&
7814 !drv
->is_filter
&& !generate_json_filename
)
7816 strcpy(bs
->exact_filename
, primary_child_bs
->exact_filename
);
7820 if (bs
->exact_filename
[0]) {
7821 pstrcpy(bs
->filename
, sizeof(bs
->filename
), bs
->exact_filename
);
7823 GString
*json
= qobject_to_json(QOBJECT(bs
->full_open_options
));
7824 if (snprintf(bs
->filename
, sizeof(bs
->filename
), "json:%s",
7825 json
->str
) >= sizeof(bs
->filename
)) {
7826 /* Give user a hint if we truncated things. */
7827 strcpy(bs
->filename
+ sizeof(bs
->filename
) - 4, "...");
7829 g_string_free(json
, true);
7833 char *bdrv_dirname(BlockDriverState
*bs
, Error
**errp
)
7835 BlockDriver
*drv
= bs
->drv
;
7836 BlockDriverState
*child_bs
;
7838 GLOBAL_STATE_CODE();
7841 error_setg(errp
, "Node '%s' is ejected", bs
->node_name
);
7845 if (drv
->bdrv_dirname
) {
7846 return drv
->bdrv_dirname(bs
, errp
);
7849 child_bs
= bdrv_primary_bs(bs
);
7851 return bdrv_dirname(child_bs
, errp
);
7854 bdrv_refresh_filename(bs
);
7855 if (bs
->exact_filename
[0] != '\0') {
7856 return path_combine(bs
->exact_filename
, "");
7859 error_setg(errp
, "Cannot generate a base directory for %s nodes",
7865 * Hot add/remove a BDS's child. So the user can take a child offline when
7866 * it is broken and take a new child online
7868 void bdrv_add_child(BlockDriverState
*parent_bs
, BlockDriverState
*child_bs
,
7871 GLOBAL_STATE_CODE();
7872 if (!parent_bs
->drv
|| !parent_bs
->drv
->bdrv_add_child
) {
7873 error_setg(errp
, "The node %s does not support adding a child",
7874 bdrv_get_device_or_node_name(parent_bs
));
7878 if (!QLIST_EMPTY(&child_bs
->parents
)) {
7879 error_setg(errp
, "The node %s already has a parent",
7880 child_bs
->node_name
);
7884 parent_bs
->drv
->bdrv_add_child(parent_bs
, child_bs
, errp
);
7887 void bdrv_del_child(BlockDriverState
*parent_bs
, BdrvChild
*child
, Error
**errp
)
7891 GLOBAL_STATE_CODE();
7892 if (!parent_bs
->drv
|| !parent_bs
->drv
->bdrv_del_child
) {
7893 error_setg(errp
, "The node %s does not support removing a child",
7894 bdrv_get_device_or_node_name(parent_bs
));
7898 QLIST_FOREACH(tmp
, &parent_bs
->children
, next
) {
7905 error_setg(errp
, "The node %s does not have a child named %s",
7906 bdrv_get_device_or_node_name(parent_bs
),
7907 bdrv_get_device_or_node_name(child
->bs
));
7911 parent_bs
->drv
->bdrv_del_child(parent_bs
, child
, errp
);
7914 int bdrv_make_empty(BdrvChild
*c
, Error
**errp
)
7916 BlockDriver
*drv
= c
->bs
->drv
;
7919 GLOBAL_STATE_CODE();
7920 assert(c
->perm
& (BLK_PERM_WRITE
| BLK_PERM_WRITE_UNCHANGED
));
7922 if (!drv
->bdrv_make_empty
) {
7923 error_setg(errp
, "%s does not support emptying nodes",
7928 ret
= drv
->bdrv_make_empty(c
->bs
);
7930 error_setg_errno(errp
, -ret
, "Failed to empty %s",
7939 * Return the child that @bs acts as an overlay for, and from which data may be
7940 * copied in COW or COR operations. Usually this is the backing file.
7942 BdrvChild
*bdrv_cow_child(BlockDriverState
*bs
)
7946 if (!bs
|| !bs
->drv
) {
7950 if (bs
->drv
->is_filter
) {
7958 assert(bs
->backing
->role
& BDRV_CHILD_COW
);
7963 * If @bs acts as a filter for exactly one of its children, return
7966 BdrvChild
*bdrv_filter_child(BlockDriverState
*bs
)
7971 if (!bs
|| !bs
->drv
) {
7975 if (!bs
->drv
->is_filter
) {
7979 /* Only one of @backing or @file may be used */
7980 assert(!(bs
->backing
&& bs
->file
));
7982 c
= bs
->backing
?: bs
->file
;
7987 assert(c
->role
& BDRV_CHILD_FILTERED
);
7992 * Return either the result of bdrv_cow_child() or bdrv_filter_child(),
7993 * whichever is non-NULL.
7995 * Return NULL if both are NULL.
7997 BdrvChild
*bdrv_filter_or_cow_child(BlockDriverState
*bs
)
7999 BdrvChild
*cow_child
= bdrv_cow_child(bs
);
8000 BdrvChild
*filter_child
= bdrv_filter_child(bs
);
8003 /* Filter nodes cannot have COW backing files */
8004 assert(!(cow_child
&& filter_child
));
8006 return cow_child
?: filter_child
;
8010 * Return the primary child of this node: For filters, that is the
8011 * filtered child. For other nodes, that is usually the child storing
8013 * (A generally more helpful description is that this is (usually) the
8014 * child that has the same filename as @bs.)
8016 * Drivers do not necessarily have a primary child; for example quorum
8019 BdrvChild
*bdrv_primary_child(BlockDriverState
*bs
)
8021 BdrvChild
*c
, *found
= NULL
;
8024 QLIST_FOREACH(c
, &bs
->children
, next
) {
8025 if (c
->role
& BDRV_CHILD_PRIMARY
) {
8034 static BlockDriverState
*bdrv_do_skip_filters(BlockDriverState
*bs
,
8035 bool stop_on_explicit_filter
)
8043 while (!(stop_on_explicit_filter
&& !bs
->implicit
)) {
8044 c
= bdrv_filter_child(bs
);
8047 * A filter that is embedded in a working block graph must
8048 * have a child. Assert this here so this function does
8049 * not return a filter node that is not expected by the
8052 assert(!bs
->drv
|| !bs
->drv
->is_filter
);
8058 * Note that this treats nodes with bs->drv == NULL as not being
8059 * filters (bs->drv == NULL should be replaced by something else
8061 * The advantage of this behavior is that this function will thus
8062 * always return a non-NULL value (given a non-NULL @bs).
8069 * Return the first BDS that has not been added implicitly or that
8070 * does not have a filtered child down the chain starting from @bs
8071 * (including @bs itself).
8073 BlockDriverState
*bdrv_skip_implicit_filters(BlockDriverState
*bs
)
8075 GLOBAL_STATE_CODE();
8076 return bdrv_do_skip_filters(bs
, true);
8080 * Return the first BDS that does not have a filtered child down the
8081 * chain starting from @bs (including @bs itself).
8083 BlockDriverState
*bdrv_skip_filters(BlockDriverState
*bs
)
8086 return bdrv_do_skip_filters(bs
, false);
8090 * For a backing chain, return the first non-filter backing image of
8091 * the first non-filter image.
8093 BlockDriverState
*bdrv_backing_chain_next(BlockDriverState
*bs
)
8096 return bdrv_skip_filters(bdrv_cow_bs(bdrv_skip_filters(bs
)));
8100 * Check whether [offset, offset + bytes) overlaps with the cached
8101 * block-status data region.
8103 * If so, and @pnum is not NULL, set *pnum to `bsc.data_end - offset`,
8104 * which is what bdrv_bsc_is_data()'s interface needs.
8105 * Otherwise, *pnum is not touched.
8107 static bool bdrv_bsc_range_overlaps_locked(BlockDriverState
*bs
,
8108 int64_t offset
, int64_t bytes
,
8111 BdrvBlockStatusCache
*bsc
= qatomic_rcu_read(&bs
->block_status_cache
);
8115 qatomic_read(&bsc
->valid
) &&
8116 ranges_overlap(offset
, bytes
, bsc
->data_start
,
8117 bsc
->data_end
- bsc
->data_start
);
8119 if (overlaps
&& pnum
) {
8120 *pnum
= bsc
->data_end
- offset
;
8127 * See block_int.h for this function's documentation.
8129 bool bdrv_bsc_is_data(BlockDriverState
*bs
, int64_t offset
, int64_t *pnum
)
8132 RCU_READ_LOCK_GUARD();
8133 return bdrv_bsc_range_overlaps_locked(bs
, offset
, 1, pnum
);
8137 * See block_int.h for this function's documentation.
8139 void bdrv_bsc_invalidate_range(BlockDriverState
*bs
,
8140 int64_t offset
, int64_t bytes
)
8143 RCU_READ_LOCK_GUARD();
8145 if (bdrv_bsc_range_overlaps_locked(bs
, offset
, bytes
, NULL
)) {
8146 qatomic_set(&bs
->block_status_cache
->valid
, false);
8151 * See block_int.h for this function's documentation.
8153 void bdrv_bsc_fill(BlockDriverState
*bs
, int64_t offset
, int64_t bytes
)
8155 BdrvBlockStatusCache
*new_bsc
= g_new(BdrvBlockStatusCache
, 1);
8156 BdrvBlockStatusCache
*old_bsc
;
8159 *new_bsc
= (BdrvBlockStatusCache
) {
8161 .data_start
= offset
,
8162 .data_end
= offset
+ bytes
,
8165 QEMU_LOCK_GUARD(&bs
->bsc_modify_lock
);
8167 old_bsc
= qatomic_rcu_read(&bs
->block_status_cache
);
8168 qatomic_rcu_set(&bs
->block_status_cache
, new_bsc
);
8170 g_free_rcu(old_bsc
, rcu
);