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_child_free(BdrvChild
*child
);
94 static void bdrv_replace_child_noperm(BdrvChild
**child
,
95 BlockDriverState
*new_bs
,
96 bool free_empty_child
);
97 static void bdrv_remove_file_or_backing_child(BlockDriverState
*bs
,
100 static void bdrv_remove_filter_or_cow_child(BlockDriverState
*bs
,
103 static int bdrv_reopen_prepare(BDRVReopenState
*reopen_state
,
104 BlockReopenQueue
*queue
,
105 Transaction
*change_child_tran
, Error
**errp
);
106 static void bdrv_reopen_commit(BDRVReopenState
*reopen_state
);
107 static void bdrv_reopen_abort(BDRVReopenState
*reopen_state
);
109 static bool bdrv_backing_overridden(BlockDriverState
*bs
);
111 /* If non-zero, use only whitelisted block drivers */
112 static int use_bdrv_whitelist
;
115 static int is_windows_drive_prefix(const char *filename
)
117 return (((filename
[0] >= 'a' && filename
[0] <= 'z') ||
118 (filename
[0] >= 'A' && filename
[0] <= 'Z')) &&
122 int is_windows_drive(const char *filename
)
124 if (is_windows_drive_prefix(filename
) &&
127 if (strstart(filename
, "\\\\.\\", NULL
) ||
128 strstart(filename
, "//./", NULL
))
134 size_t bdrv_opt_mem_align(BlockDriverState
*bs
)
136 if (!bs
|| !bs
->drv
) {
137 /* page size or 4k (hdd sector size) should be on the safe side */
138 return MAX(4096, qemu_real_host_page_size());
142 return bs
->bl
.opt_mem_alignment
;
145 size_t bdrv_min_mem_align(BlockDriverState
*bs
)
147 if (!bs
|| !bs
->drv
) {
148 /* page size or 4k (hdd sector size) should be on the safe side */
149 return MAX(4096, qemu_real_host_page_size());
153 return bs
->bl
.min_mem_alignment
;
156 /* check if the path starts with "<protocol>:" */
157 int path_has_protocol(const char *path
)
162 if (is_windows_drive(path
) ||
163 is_windows_drive_prefix(path
)) {
166 p
= path
+ strcspn(path
, ":/\\");
168 p
= path
+ strcspn(path
, ":/");
174 int path_is_absolute(const char *path
)
177 /* specific case for names like: "\\.\d:" */
178 if (is_windows_drive(path
) || is_windows_drive_prefix(path
)) {
181 return (*path
== '/' || *path
== '\\');
183 return (*path
== '/');
187 /* if filename is absolute, just return its duplicate. Otherwise, build a
188 path to it by considering it is relative to base_path. URL are
190 char *path_combine(const char *base_path
, const char *filename
)
192 const char *protocol_stripped
= NULL
;
197 if (path_is_absolute(filename
)) {
198 return g_strdup(filename
);
201 if (path_has_protocol(base_path
)) {
202 protocol_stripped
= strchr(base_path
, ':');
203 if (protocol_stripped
) {
207 p
= protocol_stripped
?: base_path
;
209 p1
= strrchr(base_path
, '/');
213 p2
= strrchr(base_path
, '\\');
214 if (!p1
|| p2
> p1
) {
229 result
= g_malloc(len
+ strlen(filename
) + 1);
230 memcpy(result
, base_path
, len
);
231 strcpy(result
+ len
, filename
);
237 * Helper function for bdrv_parse_filename() implementations to remove optional
238 * protocol prefixes (especially "file:") from a filename and for putting the
239 * stripped filename into the options QDict if there is such a prefix.
241 void bdrv_parse_filename_strip_prefix(const char *filename
, const char *prefix
,
244 if (strstart(filename
, prefix
, &filename
)) {
245 /* Stripping the explicit protocol prefix may result in a protocol
246 * prefix being (wrongly) detected (if the filename contains a colon) */
247 if (path_has_protocol(filename
)) {
248 GString
*fat_filename
;
250 /* This means there is some colon before the first slash; therefore,
251 * this cannot be an absolute path */
252 assert(!path_is_absolute(filename
));
254 /* And we can thus fix the protocol detection issue by prefixing it
256 fat_filename
= g_string_new("./");
257 g_string_append(fat_filename
, filename
);
259 assert(!path_has_protocol(fat_filename
->str
));
261 qdict_put(options
, "filename",
262 qstring_from_gstring(fat_filename
));
264 /* If no protocol prefix was detected, we can use the shortened
266 qdict_put_str(options
, "filename", filename
);
272 /* Returns whether the image file is opened as read-only. Note that this can
273 * return false and writing to the image file is still not possible because the
274 * image is inactivated. */
275 bool bdrv_is_read_only(BlockDriverState
*bs
)
278 return !(bs
->open_flags
& BDRV_O_RDWR
);
281 int bdrv_can_set_read_only(BlockDriverState
*bs
, bool read_only
,
282 bool ignore_allow_rdw
, Error
**errp
)
286 /* Do not set read_only if copy_on_read is enabled */
287 if (bs
->copy_on_read
&& read_only
) {
288 error_setg(errp
, "Can't set node '%s' to r/o with copy-on-read enabled",
289 bdrv_get_device_or_node_name(bs
));
293 /* Do not clear read_only if it is prohibited */
294 if (!read_only
&& !(bs
->open_flags
& BDRV_O_ALLOW_RDWR
) &&
297 error_setg(errp
, "Node '%s' is read only",
298 bdrv_get_device_or_node_name(bs
));
306 * Called by a driver that can only provide a read-only image.
308 * Returns 0 if the node is already read-only or it could switch the node to
309 * read-only because BDRV_O_AUTO_RDONLY is set.
311 * Returns -EACCES if the node is read-write and BDRV_O_AUTO_RDONLY is not set
312 * or bdrv_can_set_read_only() forbids making the node read-only. If @errmsg
313 * is not NULL, it is used as the error message for the Error object.
315 int bdrv_apply_auto_read_only(BlockDriverState
*bs
, const char *errmsg
,
321 if (!(bs
->open_flags
& BDRV_O_RDWR
)) {
324 if (!(bs
->open_flags
& BDRV_O_AUTO_RDONLY
)) {
328 ret
= bdrv_can_set_read_only(bs
, true, false, NULL
);
333 bs
->open_flags
&= ~BDRV_O_RDWR
;
338 error_setg(errp
, "%s", errmsg
?: "Image is read-only");
343 * If @backing is empty, this function returns NULL without setting
344 * @errp. In all other cases, NULL will only be returned with @errp
347 * Therefore, a return value of NULL without @errp set means that
348 * there is no backing file; if @errp is set, there is one but its
349 * absolute filename cannot be generated.
351 char *bdrv_get_full_backing_filename_from_filename(const char *backed
,
355 if (backing
[0] == '\0') {
357 } else if (path_has_protocol(backing
) || path_is_absolute(backing
)) {
358 return g_strdup(backing
);
359 } else if (backed
[0] == '\0' || strstart(backed
, "json:", NULL
)) {
360 error_setg(errp
, "Cannot use relative backing file names for '%s'",
364 return path_combine(backed
, backing
);
369 * If @filename is empty or NULL, this function returns NULL without
370 * setting @errp. In all other cases, NULL will only be returned with
373 static char *bdrv_make_absolute_filename(BlockDriverState
*relative_to
,
374 const char *filename
, Error
**errp
)
376 char *dir
, *full_name
;
378 if (!filename
|| filename
[0] == '\0') {
380 } else if (path_has_protocol(filename
) || path_is_absolute(filename
)) {
381 return g_strdup(filename
);
384 dir
= bdrv_dirname(relative_to
, errp
);
389 full_name
= g_strconcat(dir
, filename
, NULL
);
394 char *bdrv_get_full_backing_filename(BlockDriverState
*bs
, Error
**errp
)
397 return bdrv_make_absolute_filename(bs
, bs
->backing_file
, errp
);
400 void bdrv_register(BlockDriver
*bdrv
)
402 assert(bdrv
->format_name
);
404 QLIST_INSERT_HEAD(&bdrv_drivers
, bdrv
, list
);
407 BlockDriverState
*bdrv_new(void)
409 BlockDriverState
*bs
;
414 bs
= g_new0(BlockDriverState
, 1);
415 QLIST_INIT(&bs
->dirty_bitmaps
);
416 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
417 QLIST_INIT(&bs
->op_blockers
[i
]);
419 qemu_co_mutex_init(&bs
->reqs_lock
);
420 qemu_mutex_init(&bs
->dirty_bitmap_mutex
);
422 bs
->aio_context
= qemu_get_aio_context();
424 qemu_co_queue_init(&bs
->flush_queue
);
426 qemu_co_mutex_init(&bs
->bsc_modify_lock
);
427 bs
->block_status_cache
= g_new0(BdrvBlockStatusCache
, 1);
429 for (i
= 0; i
< bdrv_drain_all_count
; i
++) {
430 bdrv_drained_begin(bs
);
433 QTAILQ_INSERT_TAIL(&all_bdrv_states
, bs
, bs_list
);
438 static BlockDriver
*bdrv_do_find_format(const char *format_name
)
443 QLIST_FOREACH(drv1
, &bdrv_drivers
, list
) {
444 if (!strcmp(drv1
->format_name
, format_name
)) {
452 BlockDriver
*bdrv_find_format(const char *format_name
)
459 drv1
= bdrv_do_find_format(format_name
);
464 /* The driver isn't registered, maybe we need to load a module */
465 for (i
= 0; i
< (int)ARRAY_SIZE(block_driver_modules
); ++i
) {
466 if (!strcmp(block_driver_modules
[i
].format_name
, format_name
)) {
467 block_module_load_one(block_driver_modules
[i
].library_name
);
472 return bdrv_do_find_format(format_name
);
475 static int bdrv_format_is_whitelisted(const char *format_name
, bool read_only
)
477 static const char *whitelist_rw
[] = {
478 CONFIG_BDRV_RW_WHITELIST
481 static const char *whitelist_ro
[] = {
482 CONFIG_BDRV_RO_WHITELIST
487 if (!whitelist_rw
[0] && !whitelist_ro
[0]) {
488 return 1; /* no whitelist, anything goes */
491 for (p
= whitelist_rw
; *p
; p
++) {
492 if (!strcmp(format_name
, *p
)) {
497 for (p
= whitelist_ro
; *p
; p
++) {
498 if (!strcmp(format_name
, *p
)) {
506 int bdrv_is_whitelisted(BlockDriver
*drv
, bool read_only
)
509 return bdrv_format_is_whitelisted(drv
->format_name
, read_only
);
512 bool bdrv_uses_whitelist(void)
514 return use_bdrv_whitelist
;
517 typedef struct CreateCo
{
525 static void coroutine_fn
bdrv_create_co_entry(void *opaque
)
527 Error
*local_err
= NULL
;
530 CreateCo
*cco
= opaque
;
534 ret
= cco
->drv
->bdrv_co_create_opts(cco
->drv
,
535 cco
->filename
, cco
->opts
, &local_err
);
536 error_propagate(&cco
->err
, local_err
);
540 int bdrv_create(BlockDriver
*drv
, const char* filename
,
541 QemuOpts
*opts
, Error
**errp
)
550 .filename
= g_strdup(filename
),
556 if (!drv
->bdrv_co_create_opts
) {
557 error_setg(errp
, "Driver '%s' does not support image creation", drv
->format_name
);
562 if (qemu_in_coroutine()) {
563 /* Fast-path if already in coroutine context */
564 bdrv_create_co_entry(&cco
);
566 co
= qemu_coroutine_create(bdrv_create_co_entry
, &cco
);
567 qemu_coroutine_enter(co
);
568 while (cco
.ret
== NOT_DONE
) {
569 aio_poll(qemu_get_aio_context(), true);
576 error_propagate(errp
, cco
.err
);
578 error_setg_errno(errp
, -ret
, "Could not create image");
583 g_free(cco
.filename
);
588 * Helper function for bdrv_create_file_fallback(): Resize @blk to at
589 * least the given @minimum_size.
591 * On success, return @blk's actual length.
592 * Otherwise, return -errno.
594 static int64_t create_file_fallback_truncate(BlockBackend
*blk
,
595 int64_t minimum_size
, Error
**errp
)
597 Error
*local_err
= NULL
;
603 ret
= blk_truncate(blk
, minimum_size
, false, PREALLOC_MODE_OFF
, 0,
605 if (ret
< 0 && ret
!= -ENOTSUP
) {
606 error_propagate(errp
, local_err
);
610 size
= blk_getlength(blk
);
612 error_free(local_err
);
613 error_setg_errno(errp
, -size
,
614 "Failed to inquire the new image file's length");
618 if (size
< minimum_size
) {
619 /* Need to grow the image, but we failed to do that */
620 error_propagate(errp
, local_err
);
624 error_free(local_err
);
631 * Helper function for bdrv_create_file_fallback(): Zero the first
632 * sector to remove any potentially pre-existing image header.
634 static int coroutine_fn
635 create_file_fallback_zero_first_sector(BlockBackend
*blk
,
636 int64_t current_size
,
639 int64_t bytes_to_clear
;
644 bytes_to_clear
= MIN(current_size
, BDRV_SECTOR_SIZE
);
645 if (bytes_to_clear
) {
646 ret
= blk_pwrite_zeroes(blk
, 0, bytes_to_clear
, BDRV_REQ_MAY_UNMAP
);
648 error_setg_errno(errp
, -ret
,
649 "Failed to clear the new image's first sector");
658 * Simple implementation of bdrv_co_create_opts for protocol drivers
659 * which only support creation via opening a file
660 * (usually existing raw storage device)
662 int coroutine_fn
bdrv_co_create_opts_simple(BlockDriver
*drv
,
663 const char *filename
,
671 PreallocMode prealloc
;
672 Error
*local_err
= NULL
;
677 size
= qemu_opt_get_size_del(opts
, BLOCK_OPT_SIZE
, 0);
678 buf
= qemu_opt_get_del(opts
, BLOCK_OPT_PREALLOC
);
679 prealloc
= qapi_enum_parse(&PreallocMode_lookup
, buf
,
680 PREALLOC_MODE_OFF
, &local_err
);
683 error_propagate(errp
, local_err
);
687 if (prealloc
!= PREALLOC_MODE_OFF
) {
688 error_setg(errp
, "Unsupported preallocation mode '%s'",
689 PreallocMode_str(prealloc
));
693 options
= qdict_new();
694 qdict_put_str(options
, "driver", drv
->format_name
);
696 blk
= blk_new_open(filename
, NULL
, options
,
697 BDRV_O_RDWR
| BDRV_O_RESIZE
, errp
);
699 error_prepend(errp
, "Protocol driver '%s' does not support image "
700 "creation, and opening the image failed: ",
705 size
= create_file_fallback_truncate(blk
, size
, errp
);
711 ret
= create_file_fallback_zero_first_sector(blk
, size
, errp
);
722 int bdrv_create_file(const char *filename
, QemuOpts
*opts
, Error
**errp
)
724 QemuOpts
*protocol_opts
;
731 drv
= bdrv_find_protocol(filename
, true, errp
);
736 if (!drv
->create_opts
) {
737 error_setg(errp
, "Driver '%s' does not support image creation",
743 * 'opts' contains a QemuOptsList with a combination of format and protocol
746 * The format properly removes its options, but the default values remain
747 * in 'opts->list'. So if the protocol has options with the same name
748 * (e.g. rbd has 'cluster_size' as qcow2), it will see the default values
749 * of the format, since for overlapping options, the format wins.
751 * To avoid this issue, lets convert QemuOpts to QDict, in this way we take
752 * only the set options, and then convert it back to QemuOpts, using the
753 * create_opts of the protocol. So the new QemuOpts, will contain only the
756 qdict
= qemu_opts_to_qdict(opts
, NULL
);
757 protocol_opts
= qemu_opts_from_qdict(drv
->create_opts
, qdict
, errp
);
758 if (protocol_opts
== NULL
) {
763 ret
= bdrv_create(drv
, filename
, protocol_opts
, errp
);
765 qemu_opts_del(protocol_opts
);
766 qobject_unref(qdict
);
770 int coroutine_fn
bdrv_co_delete_file(BlockDriverState
*bs
, Error
**errp
)
772 Error
*local_err
= NULL
;
779 error_setg(errp
, "Block node '%s' is not opened", bs
->filename
);
783 if (!bs
->drv
->bdrv_co_delete_file
) {
784 error_setg(errp
, "Driver '%s' does not support image deletion",
785 bs
->drv
->format_name
);
789 ret
= bs
->drv
->bdrv_co_delete_file(bs
, &local_err
);
791 error_propagate(errp
, local_err
);
797 void coroutine_fn
bdrv_co_delete_file_noerr(BlockDriverState
*bs
)
799 Error
*local_err
= NULL
;
807 ret
= bdrv_co_delete_file(bs
, &local_err
);
809 * ENOTSUP will happen if the block driver doesn't support
810 * the 'bdrv_co_delete_file' interface. This is a predictable
811 * scenario and shouldn't be reported back to the user.
813 if (ret
== -ENOTSUP
) {
814 error_free(local_err
);
815 } else if (ret
< 0) {
816 error_report_err(local_err
);
821 * Try to get @bs's logical and physical block size.
822 * On success, store them in @bsz struct and return 0.
823 * On failure return -errno.
824 * @bs must not be empty.
826 int bdrv_probe_blocksizes(BlockDriverState
*bs
, BlockSizes
*bsz
)
828 BlockDriver
*drv
= bs
->drv
;
829 BlockDriverState
*filtered
= bdrv_filter_bs(bs
);
832 if (drv
&& drv
->bdrv_probe_blocksizes
) {
833 return drv
->bdrv_probe_blocksizes(bs
, bsz
);
834 } else if (filtered
) {
835 return bdrv_probe_blocksizes(filtered
, bsz
);
842 * Try to get @bs's geometry (cyls, heads, sectors).
843 * On success, store them in @geo struct and return 0.
844 * On failure return -errno.
845 * @bs must not be empty.
847 int bdrv_probe_geometry(BlockDriverState
*bs
, HDGeometry
*geo
)
849 BlockDriver
*drv
= bs
->drv
;
850 BlockDriverState
*filtered
= bdrv_filter_bs(bs
);
853 if (drv
&& drv
->bdrv_probe_geometry
) {
854 return drv
->bdrv_probe_geometry(bs
, geo
);
855 } else if (filtered
) {
856 return bdrv_probe_geometry(filtered
, geo
);
863 * Create a uniquely-named empty temporary file.
864 * Return the actual file name used upon success, otherwise NULL.
865 * This string should be freed with g_free() when not needed any longer.
867 * Note: creating a temporary file for the caller to (re)open is
868 * inherently racy. Use g_file_open_tmp() instead whenever practical.
870 char *create_tmp_file(Error
**errp
)
874 g_autofree
char *filename
= NULL
;
876 tmpdir
= g_get_tmp_dir();
879 * See commit 69bef79 ("block: use /var/tmp instead of /tmp for -snapshot")
881 * This function is used to create temporary disk images (like -snapshot),
882 * so the files can become very large. /tmp is often a tmpfs where as
883 * /var/tmp is usually on a disk, so more appropriate for disk images.
885 if (!g_strcmp0(tmpdir
, "/tmp")) {
890 filename
= g_strdup_printf("%s/vl.XXXXXX", tmpdir
);
891 fd
= g_mkstemp(filename
);
893 error_setg_errno(errp
, errno
, "Could not open temporary file '%s'",
899 return g_steal_pointer(&filename
);
903 * Detect host devices. By convention, /dev/cdrom[N] is always
904 * recognized as a host CDROM.
906 static BlockDriver
*find_hdev_driver(const char *filename
)
908 int score_max
= 0, score
;
909 BlockDriver
*drv
= NULL
, *d
;
912 QLIST_FOREACH(d
, &bdrv_drivers
, list
) {
913 if (d
->bdrv_probe_device
) {
914 score
= d
->bdrv_probe_device(filename
);
915 if (score
> score_max
) {
925 static BlockDriver
*bdrv_do_find_protocol(const char *protocol
)
930 QLIST_FOREACH(drv1
, &bdrv_drivers
, list
) {
931 if (drv1
->protocol_name
&& !strcmp(drv1
->protocol_name
, protocol
)) {
939 BlockDriver
*bdrv_find_protocol(const char *filename
,
940 bool allow_protocol_prefix
,
950 /* TODO Drivers without bdrv_file_open must be specified explicitly */
953 * XXX(hch): we really should not let host device detection
954 * override an explicit protocol specification, but moving this
955 * later breaks access to device names with colons in them.
956 * Thanks to the brain-dead persistent naming schemes on udev-
957 * based Linux systems those actually are quite common.
959 drv1
= find_hdev_driver(filename
);
964 if (!path_has_protocol(filename
) || !allow_protocol_prefix
) {
968 p
= strchr(filename
, ':');
971 if (len
> sizeof(protocol
) - 1)
972 len
= sizeof(protocol
) - 1;
973 memcpy(protocol
, filename
, len
);
974 protocol
[len
] = '\0';
976 drv1
= bdrv_do_find_protocol(protocol
);
981 for (i
= 0; i
< (int)ARRAY_SIZE(block_driver_modules
); ++i
) {
982 if (block_driver_modules
[i
].protocol_name
&&
983 !strcmp(block_driver_modules
[i
].protocol_name
, protocol
)) {
984 block_module_load_one(block_driver_modules
[i
].library_name
);
989 drv1
= bdrv_do_find_protocol(protocol
);
991 error_setg(errp
, "Unknown protocol '%s'", protocol
);
997 * Guess image format by probing its contents.
998 * This is not a good idea when your image is raw (CVE-2008-2004), but
999 * we do it anyway for backward compatibility.
1001 * @buf contains the image's first @buf_size bytes.
1002 * @buf_size is the buffer size in bytes (generally BLOCK_PROBE_BUF_SIZE,
1003 * but can be smaller if the image file is smaller)
1004 * @filename is its filename.
1006 * For all block drivers, call the bdrv_probe() method to get its
1008 * Return the first block driver with the highest probing score.
1010 BlockDriver
*bdrv_probe_all(const uint8_t *buf
, int buf_size
,
1011 const char *filename
)
1013 int score_max
= 0, score
;
1014 BlockDriver
*drv
= NULL
, *d
;
1017 QLIST_FOREACH(d
, &bdrv_drivers
, list
) {
1018 if (d
->bdrv_probe
) {
1019 score
= d
->bdrv_probe(buf
, buf_size
, filename
);
1020 if (score
> score_max
) {
1030 static int find_image_format(BlockBackend
*file
, const char *filename
,
1031 BlockDriver
**pdrv
, Error
**errp
)
1034 uint8_t buf
[BLOCK_PROBE_BUF_SIZE
];
1037 GLOBAL_STATE_CODE();
1039 /* Return the raw BlockDriver * to scsi-generic devices or empty drives */
1040 if (blk_is_sg(file
) || !blk_is_inserted(file
) || blk_getlength(file
) == 0) {
1045 ret
= blk_pread(file
, 0, sizeof(buf
), buf
, 0);
1047 error_setg_errno(errp
, -ret
, "Could not read image for determining its "
1053 drv
= bdrv_probe_all(buf
, sizeof(buf
), filename
);
1055 error_setg(errp
, "Could not determine image format: No compatible "
1066 * Set the current 'total_sectors' value
1067 * Return 0 on success, -errno on error.
1069 int refresh_total_sectors(BlockDriverState
*bs
, int64_t hint
)
1071 BlockDriver
*drv
= bs
->drv
;
1078 /* Do not attempt drv->bdrv_getlength() on scsi-generic devices */
1082 /* query actual device if possible, otherwise just trust the hint */
1083 if (drv
->bdrv_getlength
) {
1084 int64_t length
= drv
->bdrv_getlength(bs
);
1088 hint
= DIV_ROUND_UP(length
, BDRV_SECTOR_SIZE
);
1091 bs
->total_sectors
= hint
;
1093 if (bs
->total_sectors
* BDRV_SECTOR_SIZE
> BDRV_MAX_LENGTH
) {
1101 * Combines a QDict of new block driver @options with any missing options taken
1102 * from @old_options, so that leaving out an option defaults to its old value.
1104 static void bdrv_join_options(BlockDriverState
*bs
, QDict
*options
,
1107 GLOBAL_STATE_CODE();
1108 if (bs
->drv
&& bs
->drv
->bdrv_join_options
) {
1109 bs
->drv
->bdrv_join_options(options
, old_options
);
1111 qdict_join(options
, old_options
, false);
1115 static BlockdevDetectZeroesOptions
bdrv_parse_detect_zeroes(QemuOpts
*opts
,
1119 Error
*local_err
= NULL
;
1120 char *value
= qemu_opt_get_del(opts
, "detect-zeroes");
1121 BlockdevDetectZeroesOptions detect_zeroes
=
1122 qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup
, value
,
1123 BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF
, &local_err
);
1124 GLOBAL_STATE_CODE();
1127 error_propagate(errp
, local_err
);
1128 return detect_zeroes
;
1131 if (detect_zeroes
== BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP
&&
1132 !(open_flags
& BDRV_O_UNMAP
))
1134 error_setg(errp
, "setting detect-zeroes to unmap is not allowed "
1135 "without setting discard operation to unmap");
1138 return detect_zeroes
;
1142 * Set open flags for aio engine
1144 * Return 0 on success, -1 if the engine specified is invalid
1146 int bdrv_parse_aio(const char *mode
, int *flags
)
1148 if (!strcmp(mode
, "threads")) {
1149 /* do nothing, default */
1150 } else if (!strcmp(mode
, "native")) {
1151 *flags
|= BDRV_O_NATIVE_AIO
;
1152 #ifdef CONFIG_LINUX_IO_URING
1153 } else if (!strcmp(mode
, "io_uring")) {
1154 *flags
|= BDRV_O_IO_URING
;
1164 * Set open flags for a given discard mode
1166 * Return 0 on success, -1 if the discard mode was invalid.
1168 int bdrv_parse_discard_flags(const char *mode
, int *flags
)
1170 *flags
&= ~BDRV_O_UNMAP
;
1172 if (!strcmp(mode
, "off") || !strcmp(mode
, "ignore")) {
1174 } else if (!strcmp(mode
, "on") || !strcmp(mode
, "unmap")) {
1175 *flags
|= BDRV_O_UNMAP
;
1184 * Set open flags for a given cache mode
1186 * Return 0 on success, -1 if the cache mode was invalid.
1188 int bdrv_parse_cache_mode(const char *mode
, int *flags
, bool *writethrough
)
1190 *flags
&= ~BDRV_O_CACHE_MASK
;
1192 if (!strcmp(mode
, "off") || !strcmp(mode
, "none")) {
1193 *writethrough
= false;
1194 *flags
|= BDRV_O_NOCACHE
;
1195 } else if (!strcmp(mode
, "directsync")) {
1196 *writethrough
= true;
1197 *flags
|= BDRV_O_NOCACHE
;
1198 } else if (!strcmp(mode
, "writeback")) {
1199 *writethrough
= false;
1200 } else if (!strcmp(mode
, "unsafe")) {
1201 *writethrough
= false;
1202 *flags
|= BDRV_O_NO_FLUSH
;
1203 } else if (!strcmp(mode
, "writethrough")) {
1204 *writethrough
= true;
1212 static char *bdrv_child_get_parent_desc(BdrvChild
*c
)
1214 BlockDriverState
*parent
= c
->opaque
;
1215 return g_strdup_printf("node '%s'", bdrv_get_node_name(parent
));
1218 static void bdrv_child_cb_drained_begin(BdrvChild
*child
)
1220 BlockDriverState
*bs
= child
->opaque
;
1221 bdrv_do_drained_begin_quiesce(bs
, NULL
, false);
1224 static bool bdrv_child_cb_drained_poll(BdrvChild
*child
)
1226 BlockDriverState
*bs
= child
->opaque
;
1227 return bdrv_drain_poll(bs
, false, NULL
, false);
1230 static void bdrv_child_cb_drained_end(BdrvChild
*child
,
1231 int *drained_end_counter
)
1233 BlockDriverState
*bs
= child
->opaque
;
1234 bdrv_drained_end_no_poll(bs
, drained_end_counter
);
1237 static int bdrv_child_cb_inactivate(BdrvChild
*child
)
1239 BlockDriverState
*bs
= child
->opaque
;
1240 GLOBAL_STATE_CODE();
1241 assert(bs
->open_flags
& BDRV_O_INACTIVE
);
1245 static bool bdrv_child_cb_can_set_aio_ctx(BdrvChild
*child
, AioContext
*ctx
,
1246 GSList
**ignore
, Error
**errp
)
1248 BlockDriverState
*bs
= child
->opaque
;
1249 return bdrv_can_set_aio_context(bs
, ctx
, ignore
, errp
);
1252 static void bdrv_child_cb_set_aio_ctx(BdrvChild
*child
, AioContext
*ctx
,
1255 BlockDriverState
*bs
= child
->opaque
;
1256 return bdrv_set_aio_context_ignore(bs
, ctx
, ignore
);
1260 * Returns the options and flags that a temporary snapshot should get, based on
1261 * the originally requested flags (the originally requested image will have
1262 * flags like a backing file)
1264 static void bdrv_temp_snapshot_options(int *child_flags
, QDict
*child_options
,
1265 int parent_flags
, QDict
*parent_options
)
1267 GLOBAL_STATE_CODE();
1268 *child_flags
= (parent_flags
& ~BDRV_O_SNAPSHOT
) | BDRV_O_TEMPORARY
;
1270 /* For temporary files, unconditional cache=unsafe is fine */
1271 qdict_set_default_str(child_options
, BDRV_OPT_CACHE_DIRECT
, "off");
1272 qdict_set_default_str(child_options
, BDRV_OPT_CACHE_NO_FLUSH
, "on");
1274 /* Copy the read-only and discard options from the parent */
1275 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_READ_ONLY
);
1276 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_DISCARD
);
1278 /* aio=native doesn't work for cache.direct=off, so disable it for the
1279 * temporary snapshot */
1280 *child_flags
&= ~BDRV_O_NATIVE_AIO
;
1283 static void bdrv_backing_attach(BdrvChild
*c
)
1285 BlockDriverState
*parent
= c
->opaque
;
1286 BlockDriverState
*backing_hd
= c
->bs
;
1288 GLOBAL_STATE_CODE();
1289 assert(!parent
->backing_blocker
);
1290 error_setg(&parent
->backing_blocker
,
1291 "node is used as backing hd of '%s'",
1292 bdrv_get_device_or_node_name(parent
));
1294 bdrv_refresh_filename(backing_hd
);
1296 parent
->open_flags
&= ~BDRV_O_NO_BACKING
;
1298 bdrv_op_block_all(backing_hd
, parent
->backing_blocker
);
1299 /* Otherwise we won't be able to commit or stream */
1300 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_COMMIT_TARGET
,
1301 parent
->backing_blocker
);
1302 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_STREAM
,
1303 parent
->backing_blocker
);
1305 * We do backup in 3 ways:
1307 * The target bs is new opened, and the source is top BDS
1308 * 2. blockdev backup
1309 * Both the source and the target are top BDSes.
1310 * 3. internal backup(used for block replication)
1311 * Both the source and the target are backing file
1313 * In case 1 and 2, neither the source nor the target is the backing file.
1314 * In case 3, we will block the top BDS, so there is only one block job
1315 * for the top BDS and its backing chain.
1317 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_BACKUP_SOURCE
,
1318 parent
->backing_blocker
);
1319 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_BACKUP_TARGET
,
1320 parent
->backing_blocker
);
1323 static void bdrv_backing_detach(BdrvChild
*c
)
1325 BlockDriverState
*parent
= c
->opaque
;
1327 GLOBAL_STATE_CODE();
1328 assert(parent
->backing_blocker
);
1329 bdrv_op_unblock_all(c
->bs
, parent
->backing_blocker
);
1330 error_free(parent
->backing_blocker
);
1331 parent
->backing_blocker
= NULL
;
1334 static int bdrv_backing_update_filename(BdrvChild
*c
, BlockDriverState
*base
,
1335 const char *filename
, Error
**errp
)
1337 BlockDriverState
*parent
= c
->opaque
;
1338 bool read_only
= bdrv_is_read_only(parent
);
1340 GLOBAL_STATE_CODE();
1343 ret
= bdrv_reopen_set_read_only(parent
, false, errp
);
1349 ret
= bdrv_change_backing_file(parent
, filename
,
1350 base
->drv
? base
->drv
->format_name
: "",
1353 error_setg_errno(errp
, -ret
, "Could not update backing file link");
1357 bdrv_reopen_set_read_only(parent
, true, NULL
);
1364 * Returns the options and flags that a generic child of a BDS should
1365 * get, based on the given options and flags for the parent BDS.
1367 static void bdrv_inherited_options(BdrvChildRole role
, bool parent_is_format
,
1368 int *child_flags
, QDict
*child_options
,
1369 int parent_flags
, QDict
*parent_options
)
1371 int flags
= parent_flags
;
1372 GLOBAL_STATE_CODE();
1375 * First, decide whether to set, clear, or leave BDRV_O_PROTOCOL.
1376 * Generally, the question to answer is: Should this child be
1377 * format-probed by default?
1381 * Pure and non-filtered data children of non-format nodes should
1382 * be probed by default (even when the node itself has BDRV_O_PROTOCOL
1383 * set). This only affects a very limited set of drivers (namely
1384 * quorum and blkverify when this comment was written).
1385 * Force-clear BDRV_O_PROTOCOL then.
1387 if (!parent_is_format
&&
1388 (role
& BDRV_CHILD_DATA
) &&
1389 !(role
& (BDRV_CHILD_METADATA
| BDRV_CHILD_FILTERED
)))
1391 flags
&= ~BDRV_O_PROTOCOL
;
1395 * All children of format nodes (except for COW children) and all
1396 * metadata children in general should never be format-probed.
1397 * Force-set BDRV_O_PROTOCOL then.
1399 if ((parent_is_format
&& !(role
& BDRV_CHILD_COW
)) ||
1400 (role
& BDRV_CHILD_METADATA
))
1402 flags
|= BDRV_O_PROTOCOL
;
1406 * If the cache mode isn't explicitly set, inherit direct and no-flush from
1409 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_CACHE_DIRECT
);
1410 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_CACHE_NO_FLUSH
);
1411 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_FORCE_SHARE
);
1413 if (role
& BDRV_CHILD_COW
) {
1414 /* backing files are opened read-only by default */
1415 qdict_set_default_str(child_options
, BDRV_OPT_READ_ONLY
, "on");
1416 qdict_set_default_str(child_options
, BDRV_OPT_AUTO_READ_ONLY
, "off");
1418 /* Inherit the read-only option from the parent if it's not set */
1419 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_READ_ONLY
);
1420 qdict_copy_default(child_options
, parent_options
,
1421 BDRV_OPT_AUTO_READ_ONLY
);
1425 * bdrv_co_pdiscard() respects unmap policy for the parent, so we
1426 * can default to enable it on lower layers regardless of the
1429 qdict_set_default_str(child_options
, BDRV_OPT_DISCARD
, "unmap");
1431 /* Clear flags that only apply to the top layer */
1432 flags
&= ~(BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
| BDRV_O_COPY_ON_READ
);
1434 if (role
& BDRV_CHILD_METADATA
) {
1435 flags
&= ~BDRV_O_NO_IO
;
1437 if (role
& BDRV_CHILD_COW
) {
1438 flags
&= ~BDRV_O_TEMPORARY
;
1441 *child_flags
= flags
;
1444 static void bdrv_child_cb_attach(BdrvChild
*child
)
1446 BlockDriverState
*bs
= child
->opaque
;
1448 assert_bdrv_graph_writable(bs
);
1449 QLIST_INSERT_HEAD(&bs
->children
, child
, next
);
1451 if (child
->role
& BDRV_CHILD_COW
) {
1452 bdrv_backing_attach(child
);
1455 bdrv_apply_subtree_drain(child
, bs
);
1458 static void bdrv_child_cb_detach(BdrvChild
*child
)
1460 BlockDriverState
*bs
= child
->opaque
;
1462 if (child
->role
& BDRV_CHILD_COW
) {
1463 bdrv_backing_detach(child
);
1466 bdrv_unapply_subtree_drain(child
, bs
);
1468 assert_bdrv_graph_writable(bs
);
1469 QLIST_REMOVE(child
, next
);
1472 static int bdrv_child_cb_update_filename(BdrvChild
*c
, BlockDriverState
*base
,
1473 const char *filename
, Error
**errp
)
1475 if (c
->role
& BDRV_CHILD_COW
) {
1476 return bdrv_backing_update_filename(c
, base
, filename
, errp
);
1481 AioContext
*child_of_bds_get_parent_aio_context(BdrvChild
*c
)
1483 BlockDriverState
*bs
= c
->opaque
;
1486 return bdrv_get_aio_context(bs
);
1489 const BdrvChildClass child_of_bds
= {
1490 .parent_is_bds
= true,
1491 .get_parent_desc
= bdrv_child_get_parent_desc
,
1492 .inherit_options
= bdrv_inherited_options
,
1493 .drained_begin
= bdrv_child_cb_drained_begin
,
1494 .drained_poll
= bdrv_child_cb_drained_poll
,
1495 .drained_end
= bdrv_child_cb_drained_end
,
1496 .attach
= bdrv_child_cb_attach
,
1497 .detach
= bdrv_child_cb_detach
,
1498 .inactivate
= bdrv_child_cb_inactivate
,
1499 .can_set_aio_ctx
= bdrv_child_cb_can_set_aio_ctx
,
1500 .set_aio_ctx
= bdrv_child_cb_set_aio_ctx
,
1501 .update_filename
= bdrv_child_cb_update_filename
,
1502 .get_parent_aio_context
= child_of_bds_get_parent_aio_context
,
1505 AioContext
*bdrv_child_get_parent_aio_context(BdrvChild
*c
)
1507 GLOBAL_STATE_CODE();
1508 return c
->klass
->get_parent_aio_context(c
);
1511 static int bdrv_open_flags(BlockDriverState
*bs
, int flags
)
1513 int open_flags
= flags
;
1514 GLOBAL_STATE_CODE();
1517 * Clear flags that are internal to the block layer before opening the
1520 open_flags
&= ~(BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
| BDRV_O_PROTOCOL
);
1525 static void update_flags_from_options(int *flags
, QemuOpts
*opts
)
1527 GLOBAL_STATE_CODE();
1529 *flags
&= ~(BDRV_O_CACHE_MASK
| BDRV_O_RDWR
| BDRV_O_AUTO_RDONLY
);
1531 if (qemu_opt_get_bool_del(opts
, BDRV_OPT_CACHE_NO_FLUSH
, false)) {
1532 *flags
|= BDRV_O_NO_FLUSH
;
1535 if (qemu_opt_get_bool_del(opts
, BDRV_OPT_CACHE_DIRECT
, false)) {
1536 *flags
|= BDRV_O_NOCACHE
;
1539 if (!qemu_opt_get_bool_del(opts
, BDRV_OPT_READ_ONLY
, false)) {
1540 *flags
|= BDRV_O_RDWR
;
1543 if (qemu_opt_get_bool_del(opts
, BDRV_OPT_AUTO_READ_ONLY
, false)) {
1544 *flags
|= BDRV_O_AUTO_RDONLY
;
1548 static void update_options_from_flags(QDict
*options
, int flags
)
1550 GLOBAL_STATE_CODE();
1551 if (!qdict_haskey(options
, BDRV_OPT_CACHE_DIRECT
)) {
1552 qdict_put_bool(options
, BDRV_OPT_CACHE_DIRECT
, flags
& BDRV_O_NOCACHE
);
1554 if (!qdict_haskey(options
, BDRV_OPT_CACHE_NO_FLUSH
)) {
1555 qdict_put_bool(options
, BDRV_OPT_CACHE_NO_FLUSH
,
1556 flags
& BDRV_O_NO_FLUSH
);
1558 if (!qdict_haskey(options
, BDRV_OPT_READ_ONLY
)) {
1559 qdict_put_bool(options
, BDRV_OPT_READ_ONLY
, !(flags
& BDRV_O_RDWR
));
1561 if (!qdict_haskey(options
, BDRV_OPT_AUTO_READ_ONLY
)) {
1562 qdict_put_bool(options
, BDRV_OPT_AUTO_READ_ONLY
,
1563 flags
& BDRV_O_AUTO_RDONLY
);
1567 static void bdrv_assign_node_name(BlockDriverState
*bs
,
1568 const char *node_name
,
1571 char *gen_node_name
= NULL
;
1572 GLOBAL_STATE_CODE();
1575 node_name
= gen_node_name
= id_generate(ID_BLOCK
);
1576 } else if (!id_wellformed(node_name
)) {
1578 * Check for empty string or invalid characters, but not if it is
1579 * generated (generated names use characters not available to the user)
1581 error_setg(errp
, "Invalid node-name: '%s'", node_name
);
1585 /* takes care of avoiding namespaces collisions */
1586 if (blk_by_name(node_name
)) {
1587 error_setg(errp
, "node-name=%s is conflicting with a device id",
1592 /* takes care of avoiding duplicates node names */
1593 if (bdrv_find_node(node_name
)) {
1594 error_setg(errp
, "Duplicate nodes with node-name='%s'", node_name
);
1598 /* Make sure that the node name isn't truncated */
1599 if (strlen(node_name
) >= sizeof(bs
->node_name
)) {
1600 error_setg(errp
, "Node name too long");
1604 /* copy node name into the bs and insert it into the graph list */
1605 pstrcpy(bs
->node_name
, sizeof(bs
->node_name
), node_name
);
1606 QTAILQ_INSERT_TAIL(&graph_bdrv_states
, bs
, node_list
);
1608 g_free(gen_node_name
);
1611 static int bdrv_open_driver(BlockDriverState
*bs
, BlockDriver
*drv
,
1612 const char *node_name
, QDict
*options
,
1613 int open_flags
, Error
**errp
)
1615 Error
*local_err
= NULL
;
1617 GLOBAL_STATE_CODE();
1619 bdrv_assign_node_name(bs
, node_name
, &local_err
);
1621 error_propagate(errp
, local_err
);
1626 bs
->opaque
= g_malloc0(drv
->instance_size
);
1628 if (drv
->bdrv_file_open
) {
1629 assert(!drv
->bdrv_needs_filename
|| bs
->filename
[0]);
1630 ret
= drv
->bdrv_file_open(bs
, options
, open_flags
, &local_err
);
1631 } else if (drv
->bdrv_open
) {
1632 ret
= drv
->bdrv_open(bs
, options
, open_flags
, &local_err
);
1639 error_propagate(errp
, local_err
);
1640 } else if (bs
->filename
[0]) {
1641 error_setg_errno(errp
, -ret
, "Could not open '%s'", bs
->filename
);
1643 error_setg_errno(errp
, -ret
, "Could not open image");
1648 ret
= refresh_total_sectors(bs
, bs
->total_sectors
);
1650 error_setg_errno(errp
, -ret
, "Could not refresh total sector count");
1654 bdrv_refresh_limits(bs
, NULL
, &local_err
);
1656 error_propagate(errp
, local_err
);
1660 assert(bdrv_opt_mem_align(bs
) != 0);
1661 assert(bdrv_min_mem_align(bs
) != 0);
1662 assert(is_power_of_2(bs
->bl
.request_alignment
));
1664 for (i
= 0; i
< bs
->quiesce_counter
; i
++) {
1665 if (drv
->bdrv_co_drain_begin
) {
1666 drv
->bdrv_co_drain_begin(bs
);
1673 if (bs
->file
!= NULL
) {
1674 bdrv_unref_child(bs
, bs
->file
);
1683 * Create and open a block node.
1685 * @options is a QDict of options to pass to the block drivers, or NULL for an
1686 * empty set of options. The reference to the QDict belongs to the block layer
1687 * after the call (even on failure), so if the caller intends to reuse the
1688 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
1690 BlockDriverState
*bdrv_new_open_driver_opts(BlockDriver
*drv
,
1691 const char *node_name
,
1692 QDict
*options
, int flags
,
1695 BlockDriverState
*bs
;
1698 GLOBAL_STATE_CODE();
1701 bs
->open_flags
= flags
;
1702 bs
->options
= options
?: qdict_new();
1703 bs
->explicit_options
= qdict_clone_shallow(bs
->options
);
1706 update_options_from_flags(bs
->options
, flags
);
1708 ret
= bdrv_open_driver(bs
, drv
, node_name
, bs
->options
, flags
, errp
);
1710 qobject_unref(bs
->explicit_options
);
1711 bs
->explicit_options
= NULL
;
1712 qobject_unref(bs
->options
);
1721 /* Create and open a block node. */
1722 BlockDriverState
*bdrv_new_open_driver(BlockDriver
*drv
, const char *node_name
,
1723 int flags
, Error
**errp
)
1725 GLOBAL_STATE_CODE();
1726 return bdrv_new_open_driver_opts(drv
, node_name
, NULL
, flags
, errp
);
1729 QemuOptsList bdrv_runtime_opts
= {
1730 .name
= "bdrv_common",
1731 .head
= QTAILQ_HEAD_INITIALIZER(bdrv_runtime_opts
.head
),
1734 .name
= "node-name",
1735 .type
= QEMU_OPT_STRING
,
1736 .help
= "Node name of the block device node",
1740 .type
= QEMU_OPT_STRING
,
1741 .help
= "Block driver to use for the node",
1744 .name
= BDRV_OPT_CACHE_DIRECT
,
1745 .type
= QEMU_OPT_BOOL
,
1746 .help
= "Bypass software writeback cache on the host",
1749 .name
= BDRV_OPT_CACHE_NO_FLUSH
,
1750 .type
= QEMU_OPT_BOOL
,
1751 .help
= "Ignore flush requests",
1754 .name
= BDRV_OPT_READ_ONLY
,
1755 .type
= QEMU_OPT_BOOL
,
1756 .help
= "Node is opened in read-only mode",
1759 .name
= BDRV_OPT_AUTO_READ_ONLY
,
1760 .type
= QEMU_OPT_BOOL
,
1761 .help
= "Node can become read-only if opening read-write fails",
1764 .name
= "detect-zeroes",
1765 .type
= QEMU_OPT_STRING
,
1766 .help
= "try to optimize zero writes (off, on, unmap)",
1769 .name
= BDRV_OPT_DISCARD
,
1770 .type
= QEMU_OPT_STRING
,
1771 .help
= "discard operation (ignore/off, unmap/on)",
1774 .name
= BDRV_OPT_FORCE_SHARE
,
1775 .type
= QEMU_OPT_BOOL
,
1776 .help
= "always accept other writers (default: off)",
1778 { /* end of list */ }
1782 QemuOptsList bdrv_create_opts_simple
= {
1783 .name
= "simple-create-opts",
1784 .head
= QTAILQ_HEAD_INITIALIZER(bdrv_create_opts_simple
.head
),
1787 .name
= BLOCK_OPT_SIZE
,
1788 .type
= QEMU_OPT_SIZE
,
1789 .help
= "Virtual disk size"
1792 .name
= BLOCK_OPT_PREALLOC
,
1793 .type
= QEMU_OPT_STRING
,
1794 .help
= "Preallocation mode (allowed values: off)"
1796 { /* end of list */ }
1801 * Common part for opening disk images and files
1803 * Removes all processed options from *options.
1805 static int bdrv_open_common(BlockDriverState
*bs
, BlockBackend
*file
,
1806 QDict
*options
, Error
**errp
)
1808 int ret
, open_flags
;
1809 const char *filename
;
1810 const char *driver_name
= NULL
;
1811 const char *node_name
= NULL
;
1812 const char *discard
;
1815 Error
*local_err
= NULL
;
1818 assert(bs
->file
== NULL
);
1819 assert(options
!= NULL
&& bs
->options
!= options
);
1820 GLOBAL_STATE_CODE();
1822 opts
= qemu_opts_create(&bdrv_runtime_opts
, NULL
, 0, &error_abort
);
1823 if (!qemu_opts_absorb_qdict(opts
, options
, errp
)) {
1828 update_flags_from_options(&bs
->open_flags
, opts
);
1830 driver_name
= qemu_opt_get(opts
, "driver");
1831 drv
= bdrv_find_format(driver_name
);
1832 assert(drv
!= NULL
);
1834 bs
->force_share
= qemu_opt_get_bool(opts
, BDRV_OPT_FORCE_SHARE
, false);
1836 if (bs
->force_share
&& (bs
->open_flags
& BDRV_O_RDWR
)) {
1838 BDRV_OPT_FORCE_SHARE
1839 "=on can only be used with read-only images");
1845 bdrv_refresh_filename(blk_bs(file
));
1846 filename
= blk_bs(file
)->filename
;
1849 * Caution: while qdict_get_try_str() is fine, getting
1850 * non-string types would require more care. When @options
1851 * come from -blockdev or blockdev_add, its members are typed
1852 * according to the QAPI schema, but when they come from
1853 * -drive, they're all QString.
1855 filename
= qdict_get_try_str(options
, "filename");
1858 if (drv
->bdrv_needs_filename
&& (!filename
|| !filename
[0])) {
1859 error_setg(errp
, "The '%s' block driver requires a file name",
1865 trace_bdrv_open_common(bs
, filename
?: "", bs
->open_flags
,
1868 ro
= bdrv_is_read_only(bs
);
1870 if (use_bdrv_whitelist
&& !bdrv_is_whitelisted(drv
, ro
)) {
1871 if (!ro
&& bdrv_is_whitelisted(drv
, true)) {
1872 ret
= bdrv_apply_auto_read_only(bs
, NULL
, NULL
);
1878 !ro
&& bdrv_is_whitelisted(drv
, true)
1879 ? "Driver '%s' can only be used for read-only devices"
1880 : "Driver '%s' is not whitelisted",
1886 /* bdrv_new() and bdrv_close() make it so */
1887 assert(qatomic_read(&bs
->copy_on_read
) == 0);
1889 if (bs
->open_flags
& BDRV_O_COPY_ON_READ
) {
1891 bdrv_enable_copy_on_read(bs
);
1893 error_setg(errp
, "Can't use copy-on-read on read-only device");
1899 discard
= qemu_opt_get(opts
, BDRV_OPT_DISCARD
);
1900 if (discard
!= NULL
) {
1901 if (bdrv_parse_discard_flags(discard
, &bs
->open_flags
) != 0) {
1902 error_setg(errp
, "Invalid discard option");
1909 bdrv_parse_detect_zeroes(opts
, bs
->open_flags
, &local_err
);
1911 error_propagate(errp
, local_err
);
1916 if (filename
!= NULL
) {
1917 pstrcpy(bs
->filename
, sizeof(bs
->filename
), filename
);
1919 bs
->filename
[0] = '\0';
1921 pstrcpy(bs
->exact_filename
, sizeof(bs
->exact_filename
), bs
->filename
);
1923 /* Open the image, either directly or using a protocol */
1924 open_flags
= bdrv_open_flags(bs
, bs
->open_flags
);
1925 node_name
= qemu_opt_get(opts
, "node-name");
1927 assert(!drv
->bdrv_file_open
|| file
== NULL
);
1928 ret
= bdrv_open_driver(bs
, drv
, node_name
, options
, open_flags
, errp
);
1933 qemu_opts_del(opts
);
1937 qemu_opts_del(opts
);
1941 static QDict
*parse_json_filename(const char *filename
, Error
**errp
)
1943 QObject
*options_obj
;
1946 GLOBAL_STATE_CODE();
1948 ret
= strstart(filename
, "json:", &filename
);
1951 options_obj
= qobject_from_json(filename
, errp
);
1953 error_prepend(errp
, "Could not parse the JSON options: ");
1957 options
= qobject_to(QDict
, options_obj
);
1959 qobject_unref(options_obj
);
1960 error_setg(errp
, "Invalid JSON object given");
1964 qdict_flatten(options
);
1969 static void parse_json_protocol(QDict
*options
, const char **pfilename
,
1972 QDict
*json_options
;
1973 Error
*local_err
= NULL
;
1974 GLOBAL_STATE_CODE();
1976 /* Parse json: pseudo-protocol */
1977 if (!*pfilename
|| !g_str_has_prefix(*pfilename
, "json:")) {
1981 json_options
= parse_json_filename(*pfilename
, &local_err
);
1983 error_propagate(errp
, local_err
);
1987 /* Options given in the filename have lower priority than options
1988 * specified directly */
1989 qdict_join(options
, json_options
, false);
1990 qobject_unref(json_options
);
1995 * Fills in default options for opening images and converts the legacy
1996 * filename/flags pair to option QDict entries.
1997 * The BDRV_O_PROTOCOL flag in *flags will be set or cleared accordingly if a
1998 * block driver has been specified explicitly.
2000 static int bdrv_fill_options(QDict
**options
, const char *filename
,
2001 int *flags
, Error
**errp
)
2003 const char *drvname
;
2004 bool protocol
= *flags
& BDRV_O_PROTOCOL
;
2005 bool parse_filename
= false;
2006 BlockDriver
*drv
= NULL
;
2007 Error
*local_err
= NULL
;
2009 GLOBAL_STATE_CODE();
2012 * Caution: while qdict_get_try_str() is fine, getting non-string
2013 * types would require more care. When @options come from
2014 * -blockdev or blockdev_add, its members are typed according to
2015 * the QAPI schema, but when they come from -drive, they're all
2018 drvname
= qdict_get_try_str(*options
, "driver");
2020 drv
= bdrv_find_format(drvname
);
2022 error_setg(errp
, "Unknown driver '%s'", drvname
);
2025 /* If the user has explicitly specified the driver, this choice should
2026 * override the BDRV_O_PROTOCOL flag */
2027 protocol
= drv
->bdrv_file_open
;
2031 *flags
|= BDRV_O_PROTOCOL
;
2033 *flags
&= ~BDRV_O_PROTOCOL
;
2036 /* Translate cache options from flags into options */
2037 update_options_from_flags(*options
, *flags
);
2039 /* Fetch the file name from the options QDict if necessary */
2040 if (protocol
&& filename
) {
2041 if (!qdict_haskey(*options
, "filename")) {
2042 qdict_put_str(*options
, "filename", filename
);
2043 parse_filename
= true;
2045 error_setg(errp
, "Can't specify 'file' and 'filename' options at "
2051 /* Find the right block driver */
2052 /* See cautionary note on accessing @options above */
2053 filename
= qdict_get_try_str(*options
, "filename");
2055 if (!drvname
&& protocol
) {
2057 drv
= bdrv_find_protocol(filename
, parse_filename
, errp
);
2062 drvname
= drv
->format_name
;
2063 qdict_put_str(*options
, "driver", drvname
);
2065 error_setg(errp
, "Must specify either driver or file");
2070 assert(drv
|| !protocol
);
2072 /* Driver-specific filename parsing */
2073 if (drv
&& drv
->bdrv_parse_filename
&& parse_filename
) {
2074 drv
->bdrv_parse_filename(filename
, *options
, &local_err
);
2076 error_propagate(errp
, local_err
);
2080 if (!drv
->bdrv_needs_filename
) {
2081 qdict_del(*options
, "filename");
2088 typedef struct BlockReopenQueueEntry
{
2091 BDRVReopenState state
;
2092 QTAILQ_ENTRY(BlockReopenQueueEntry
) entry
;
2093 } BlockReopenQueueEntry
;
2096 * Return the flags that @bs will have after the reopens in @q have
2097 * successfully completed. If @q is NULL (or @bs is not contained in @q),
2098 * return the current flags.
2100 static int bdrv_reopen_get_flags(BlockReopenQueue
*q
, BlockDriverState
*bs
)
2102 BlockReopenQueueEntry
*entry
;
2105 QTAILQ_FOREACH(entry
, q
, entry
) {
2106 if (entry
->state
.bs
== bs
) {
2107 return entry
->state
.flags
;
2112 return bs
->open_flags
;
2115 /* Returns whether the image file can be written to after the reopen queue @q
2116 * has been successfully applied, or right now if @q is NULL. */
2117 static bool bdrv_is_writable_after_reopen(BlockDriverState
*bs
,
2118 BlockReopenQueue
*q
)
2120 int flags
= bdrv_reopen_get_flags(q
, bs
);
2122 return (flags
& (BDRV_O_RDWR
| BDRV_O_INACTIVE
)) == BDRV_O_RDWR
;
2126 * Return whether the BDS can be written to. This is not necessarily
2127 * the same as !bdrv_is_read_only(bs), as inactivated images may not
2128 * be written to but do not count as read-only images.
2130 bool bdrv_is_writable(BlockDriverState
*bs
)
2133 return bdrv_is_writable_after_reopen(bs
, NULL
);
2136 static char *bdrv_child_user_desc(BdrvChild
*c
)
2138 GLOBAL_STATE_CODE();
2139 return c
->klass
->get_parent_desc(c
);
2143 * Check that @a allows everything that @b needs. @a and @b must reference same
2146 static bool bdrv_a_allow_b(BdrvChild
*a
, BdrvChild
*b
, Error
**errp
)
2148 const char *child_bs_name
;
2149 g_autofree
char *a_user
= NULL
;
2150 g_autofree
char *b_user
= NULL
;
2151 g_autofree
char *perms
= NULL
;
2154 assert(a
->bs
== b
->bs
);
2155 GLOBAL_STATE_CODE();
2157 if ((b
->perm
& a
->shared_perm
) == b
->perm
) {
2161 child_bs_name
= bdrv_get_node_name(b
->bs
);
2162 a_user
= bdrv_child_user_desc(a
);
2163 b_user
= bdrv_child_user_desc(b
);
2164 perms
= bdrv_perm_names(b
->perm
& ~a
->shared_perm
);
2166 error_setg(errp
, "Permission conflict on node '%s': permissions '%s' are "
2167 "both required by %s (uses node '%s' as '%s' child) and "
2168 "unshared by %s (uses node '%s' as '%s' child).",
2169 child_bs_name
, perms
,
2170 b_user
, child_bs_name
, b
->name
,
2171 a_user
, child_bs_name
, a
->name
);
2176 static bool bdrv_parent_perms_conflict(BlockDriverState
*bs
, Error
**errp
)
2179 GLOBAL_STATE_CODE();
2182 * During the loop we'll look at each pair twice. That's correct because
2183 * bdrv_a_allow_b() is asymmetric and we should check each pair in both
2186 QLIST_FOREACH(a
, &bs
->parents
, next_parent
) {
2187 QLIST_FOREACH(b
, &bs
->parents
, next_parent
) {
2192 if (!bdrv_a_allow_b(a
, b
, errp
)) {
2201 static void bdrv_child_perm(BlockDriverState
*bs
, BlockDriverState
*child_bs
,
2202 BdrvChild
*c
, BdrvChildRole role
,
2203 BlockReopenQueue
*reopen_queue
,
2204 uint64_t parent_perm
, uint64_t parent_shared
,
2205 uint64_t *nperm
, uint64_t *nshared
)
2207 assert(bs
->drv
&& bs
->drv
->bdrv_child_perm
);
2208 GLOBAL_STATE_CODE();
2209 bs
->drv
->bdrv_child_perm(bs
, c
, role
, reopen_queue
,
2210 parent_perm
, parent_shared
,
2212 /* TODO Take force_share from reopen_queue */
2213 if (child_bs
&& child_bs
->force_share
) {
2214 *nshared
= BLK_PERM_ALL
;
2219 * Adds the whole subtree of @bs (including @bs itself) to the @list (except for
2220 * nodes that are already in the @list, of course) so that final list is
2221 * topologically sorted. Return the result (GSList @list object is updated, so
2222 * don't use old reference after function call).
2224 * On function start @list must be already topologically sorted and for any node
2225 * in the @list the whole subtree of the node must be in the @list as well. The
2226 * simplest way to satisfy this criteria: use only result of
2227 * bdrv_topological_dfs() or NULL as @list parameter.
2229 static GSList
*bdrv_topological_dfs(GSList
*list
, GHashTable
*found
,
2230 BlockDriverState
*bs
)
2233 g_autoptr(GHashTable
) local_found
= NULL
;
2235 GLOBAL_STATE_CODE();
2239 found
= local_found
= g_hash_table_new(NULL
, NULL
);
2242 if (g_hash_table_contains(found
, bs
)) {
2245 g_hash_table_add(found
, bs
);
2247 QLIST_FOREACH(child
, &bs
->children
, next
) {
2248 list
= bdrv_topological_dfs(list
, found
, child
->bs
);
2251 return g_slist_prepend(list
, bs
);
2254 typedef struct BdrvChildSetPermState
{
2257 uint64_t old_shared_perm
;
2258 } BdrvChildSetPermState
;
2260 static void bdrv_child_set_perm_abort(void *opaque
)
2262 BdrvChildSetPermState
*s
= opaque
;
2264 GLOBAL_STATE_CODE();
2266 s
->child
->perm
= s
->old_perm
;
2267 s
->child
->shared_perm
= s
->old_shared_perm
;
2270 static TransactionActionDrv bdrv_child_set_pem_drv
= {
2271 .abort
= bdrv_child_set_perm_abort
,
2275 static void bdrv_child_set_perm(BdrvChild
*c
, uint64_t perm
,
2276 uint64_t shared
, Transaction
*tran
)
2278 BdrvChildSetPermState
*s
= g_new(BdrvChildSetPermState
, 1);
2279 GLOBAL_STATE_CODE();
2281 *s
= (BdrvChildSetPermState
) {
2283 .old_perm
= c
->perm
,
2284 .old_shared_perm
= c
->shared_perm
,
2288 c
->shared_perm
= shared
;
2290 tran_add(tran
, &bdrv_child_set_pem_drv
, s
);
2293 static void bdrv_drv_set_perm_commit(void *opaque
)
2295 BlockDriverState
*bs
= opaque
;
2296 uint64_t cumulative_perms
, cumulative_shared_perms
;
2297 GLOBAL_STATE_CODE();
2299 if (bs
->drv
->bdrv_set_perm
) {
2300 bdrv_get_cumulative_perm(bs
, &cumulative_perms
,
2301 &cumulative_shared_perms
);
2302 bs
->drv
->bdrv_set_perm(bs
, cumulative_perms
, cumulative_shared_perms
);
2306 static void bdrv_drv_set_perm_abort(void *opaque
)
2308 BlockDriverState
*bs
= opaque
;
2309 GLOBAL_STATE_CODE();
2311 if (bs
->drv
->bdrv_abort_perm_update
) {
2312 bs
->drv
->bdrv_abort_perm_update(bs
);
2316 TransactionActionDrv bdrv_drv_set_perm_drv
= {
2317 .abort
= bdrv_drv_set_perm_abort
,
2318 .commit
= bdrv_drv_set_perm_commit
,
2321 static int bdrv_drv_set_perm(BlockDriverState
*bs
, uint64_t perm
,
2322 uint64_t shared_perm
, Transaction
*tran
,
2325 GLOBAL_STATE_CODE();
2330 if (bs
->drv
->bdrv_check_perm
) {
2331 int ret
= bs
->drv
->bdrv_check_perm(bs
, perm
, shared_perm
, errp
);
2338 tran_add(tran
, &bdrv_drv_set_perm_drv
, bs
);
2344 typedef struct BdrvReplaceChildState
{
2347 BlockDriverState
*old_bs
;
2348 bool free_empty_child
;
2349 } BdrvReplaceChildState
;
2351 static void bdrv_replace_child_commit(void *opaque
)
2353 BdrvReplaceChildState
*s
= opaque
;
2354 GLOBAL_STATE_CODE();
2356 if (s
->free_empty_child
&& !s
->child
->bs
) {
2357 bdrv_child_free(s
->child
);
2359 bdrv_unref(s
->old_bs
);
2362 static void bdrv_replace_child_abort(void *opaque
)
2364 BdrvReplaceChildState
*s
= opaque
;
2365 BlockDriverState
*new_bs
= s
->child
->bs
;
2367 GLOBAL_STATE_CODE();
2369 * old_bs reference is transparently moved from @s to s->child.
2371 * Pass &s->child here instead of s->childp, because:
2372 * (1) s->old_bs must be non-NULL, so bdrv_replace_child_noperm() will not
2373 * modify the BdrvChild * pointer we indirectly pass to it, i.e. it
2374 * will not modify s->child. From that perspective, it does not matter
2375 * whether we pass s->childp or &s->child.
2376 * (2) If new_bs is not NULL, s->childp will be NULL. We then cannot use
2378 * (3) If new_bs is NULL, *s->childp will have been NULLed by
2379 * bdrv_replace_child_tran()'s bdrv_replace_child_noperm() call, and we
2380 * must not pass a NULL *s->childp here.
2382 * So whether new_bs was NULL or not, we cannot pass s->childp here; and in
2383 * any case, there is no reason to pass it anyway.
2385 bdrv_replace_child_noperm(&s
->child
, s
->old_bs
, true);
2387 * The child was pre-existing, so s->old_bs must be non-NULL, and
2388 * s->child thus must not have been freed
2390 assert(s
->child
!= NULL
);
2392 /* As described above, *s->childp was cleared, so restore it */
2393 assert(s
->childp
!= NULL
);
2394 *s
->childp
= s
->child
;
2399 static TransactionActionDrv bdrv_replace_child_drv
= {
2400 .commit
= bdrv_replace_child_commit
,
2401 .abort
= bdrv_replace_child_abort
,
2406 * bdrv_replace_child_tran
2408 * Note: real unref of old_bs is done only on commit.
2410 * The function doesn't update permissions, caller is responsible for this.
2412 * (*childp)->bs must not be NULL.
2414 * Note that if new_bs == NULL, @childp is stored in a state object attached
2415 * to @tran, so that the old child can be reinstated in the abort handler.
2416 * Therefore, if @new_bs can be NULL, @childp must stay valid until the
2417 * transaction is committed or aborted.
2419 * If @free_empty_child is true and @new_bs is NULL, the BdrvChild is
2420 * freed (on commit). @free_empty_child should only be false if the
2421 * caller will free the BDrvChild themselves (which may be important
2422 * if this is in turn called in another transactional context).
2424 static void bdrv_replace_child_tran(BdrvChild
**childp
,
2425 BlockDriverState
*new_bs
,
2427 bool free_empty_child
)
2429 BdrvReplaceChildState
*s
= g_new(BdrvReplaceChildState
, 1);
2430 *s
= (BdrvReplaceChildState
) {
2432 .childp
= new_bs
== NULL
? childp
: NULL
,
2433 .old_bs
= (*childp
)->bs
,
2434 .free_empty_child
= free_empty_child
,
2436 tran_add(tran
, &bdrv_replace_child_drv
, s
);
2438 /* The abort handler relies on this */
2439 assert(s
->old_bs
!= NULL
);
2445 * Pass free_empty_child=false, we will free the child (if
2446 * necessary) in bdrv_replace_child_commit() (if our
2447 * @free_empty_child parameter was true).
2449 bdrv_replace_child_noperm(childp
, new_bs
, false);
2450 /* old_bs reference is transparently moved from *childp to @s */
2454 * Refresh permissions in @bs subtree. The function is intended to be called
2455 * after some graph modification that was done without permission update.
2457 static int bdrv_node_refresh_perm(BlockDriverState
*bs
, BlockReopenQueue
*q
,
2458 Transaction
*tran
, Error
**errp
)
2460 BlockDriver
*drv
= bs
->drv
;
2463 uint64_t cumulative_perms
, cumulative_shared_perms
;
2464 GLOBAL_STATE_CODE();
2466 bdrv_get_cumulative_perm(bs
, &cumulative_perms
, &cumulative_shared_perms
);
2468 /* Write permissions never work with read-only images */
2469 if ((cumulative_perms
& (BLK_PERM_WRITE
| BLK_PERM_WRITE_UNCHANGED
)) &&
2470 !bdrv_is_writable_after_reopen(bs
, q
))
2472 if (!bdrv_is_writable_after_reopen(bs
, NULL
)) {
2473 error_setg(errp
, "Block node is read-only");
2475 error_setg(errp
, "Read-only block node '%s' cannot support "
2476 "read-write users", bdrv_get_node_name(bs
));
2483 * Unaligned requests will automatically be aligned to bl.request_alignment
2484 * and without RESIZE we can't extend requests to write to space beyond the
2485 * end of the image, so it's required that the image size is aligned.
2487 if ((cumulative_perms
& (BLK_PERM_WRITE
| BLK_PERM_WRITE_UNCHANGED
)) &&
2488 !(cumulative_perms
& BLK_PERM_RESIZE
))
2490 if ((bs
->total_sectors
* BDRV_SECTOR_SIZE
) % bs
->bl
.request_alignment
) {
2491 error_setg(errp
, "Cannot get 'write' permission without 'resize': "
2492 "Image size is not a multiple of request "
2498 /* Check this node */
2503 ret
= bdrv_drv_set_perm(bs
, cumulative_perms
, cumulative_shared_perms
, tran
,
2509 /* Drivers that never have children can omit .bdrv_child_perm() */
2510 if (!drv
->bdrv_child_perm
) {
2511 assert(QLIST_EMPTY(&bs
->children
));
2515 /* Check all children */
2516 QLIST_FOREACH(c
, &bs
->children
, next
) {
2517 uint64_t cur_perm
, cur_shared
;
2519 bdrv_child_perm(bs
, c
->bs
, c
, c
->role
, q
,
2520 cumulative_perms
, cumulative_shared_perms
,
2521 &cur_perm
, &cur_shared
);
2522 bdrv_child_set_perm(c
, cur_perm
, cur_shared
, tran
);
2528 static int bdrv_list_refresh_perms(GSList
*list
, BlockReopenQueue
*q
,
2529 Transaction
*tran
, Error
**errp
)
2532 BlockDriverState
*bs
;
2533 GLOBAL_STATE_CODE();
2535 for ( ; list
; list
= list
->next
) {
2538 if (bdrv_parent_perms_conflict(bs
, errp
)) {
2542 ret
= bdrv_node_refresh_perm(bs
, q
, tran
, errp
);
2551 void bdrv_get_cumulative_perm(BlockDriverState
*bs
, uint64_t *perm
,
2552 uint64_t *shared_perm
)
2555 uint64_t cumulative_perms
= 0;
2556 uint64_t cumulative_shared_perms
= BLK_PERM_ALL
;
2558 GLOBAL_STATE_CODE();
2560 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
2561 cumulative_perms
|= c
->perm
;
2562 cumulative_shared_perms
&= c
->shared_perm
;
2565 *perm
= cumulative_perms
;
2566 *shared_perm
= cumulative_shared_perms
;
2569 char *bdrv_perm_names(uint64_t perm
)
2575 { BLK_PERM_CONSISTENT_READ
, "consistent read" },
2576 { BLK_PERM_WRITE
, "write" },
2577 { BLK_PERM_WRITE_UNCHANGED
, "write unchanged" },
2578 { BLK_PERM_RESIZE
, "resize" },
2582 GString
*result
= g_string_sized_new(30);
2583 struct perm_name
*p
;
2585 for (p
= permissions
; p
->name
; p
++) {
2586 if (perm
& p
->perm
) {
2587 if (result
->len
> 0) {
2588 g_string_append(result
, ", ");
2590 g_string_append(result
, p
->name
);
2594 return g_string_free(result
, FALSE
);
2598 static int bdrv_refresh_perms(BlockDriverState
*bs
, Error
**errp
)
2601 Transaction
*tran
= tran_new();
2602 g_autoptr(GSList
) list
= bdrv_topological_dfs(NULL
, NULL
, bs
);
2603 GLOBAL_STATE_CODE();
2605 ret
= bdrv_list_refresh_perms(list
, NULL
, tran
, errp
);
2606 tran_finalize(tran
, ret
);
2611 int bdrv_child_try_set_perm(BdrvChild
*c
, uint64_t perm
, uint64_t shared
,
2614 Error
*local_err
= NULL
;
2615 Transaction
*tran
= tran_new();
2618 GLOBAL_STATE_CODE();
2620 bdrv_child_set_perm(c
, perm
, shared
, tran
);
2622 ret
= bdrv_refresh_perms(c
->bs
, &local_err
);
2624 tran_finalize(tran
, ret
);
2627 if ((perm
& ~c
->perm
) || (c
->shared_perm
& ~shared
)) {
2628 /* tighten permissions */
2629 error_propagate(errp
, local_err
);
2632 * Our caller may intend to only loosen restrictions and
2633 * does not expect this function to fail. Errors are not
2634 * fatal in such a case, so we can just hide them from our
2637 error_free(local_err
);
2645 int bdrv_child_refresh_perms(BlockDriverState
*bs
, BdrvChild
*c
, Error
**errp
)
2647 uint64_t parent_perms
, parent_shared
;
2648 uint64_t perms
, shared
;
2650 GLOBAL_STATE_CODE();
2652 bdrv_get_cumulative_perm(bs
, &parent_perms
, &parent_shared
);
2653 bdrv_child_perm(bs
, c
->bs
, c
, c
->role
, NULL
,
2654 parent_perms
, parent_shared
, &perms
, &shared
);
2656 return bdrv_child_try_set_perm(c
, perms
, shared
, errp
);
2660 * Default implementation for .bdrv_child_perm() for block filters:
2661 * Forward CONSISTENT_READ, WRITE, WRITE_UNCHANGED, and RESIZE to the
2664 static void bdrv_filter_default_perms(BlockDriverState
*bs
, BdrvChild
*c
,
2666 BlockReopenQueue
*reopen_queue
,
2667 uint64_t perm
, uint64_t shared
,
2668 uint64_t *nperm
, uint64_t *nshared
)
2670 GLOBAL_STATE_CODE();
2671 *nperm
= perm
& DEFAULT_PERM_PASSTHROUGH
;
2672 *nshared
= (shared
& DEFAULT_PERM_PASSTHROUGH
) | DEFAULT_PERM_UNCHANGED
;
2675 static void bdrv_default_perms_for_cow(BlockDriverState
*bs
, BdrvChild
*c
,
2677 BlockReopenQueue
*reopen_queue
,
2678 uint64_t perm
, uint64_t shared
,
2679 uint64_t *nperm
, uint64_t *nshared
)
2681 assert(role
& BDRV_CHILD_COW
);
2682 GLOBAL_STATE_CODE();
2685 * We want consistent read from backing files if the parent needs it.
2686 * No other operations are performed on backing files.
2688 perm
&= BLK_PERM_CONSISTENT_READ
;
2691 * If the parent can deal with changing data, we're okay with a
2692 * writable and resizable backing file.
2693 * TODO Require !(perm & BLK_PERM_CONSISTENT_READ), too?
2695 if (shared
& BLK_PERM_WRITE
) {
2696 shared
= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2701 shared
|= BLK_PERM_CONSISTENT_READ
| BLK_PERM_WRITE_UNCHANGED
;
2703 if (bs
->open_flags
& BDRV_O_INACTIVE
) {
2704 shared
|= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2711 static void bdrv_default_perms_for_storage(BlockDriverState
*bs
, BdrvChild
*c
,
2713 BlockReopenQueue
*reopen_queue
,
2714 uint64_t perm
, uint64_t shared
,
2715 uint64_t *nperm
, uint64_t *nshared
)
2719 GLOBAL_STATE_CODE();
2720 assert(role
& (BDRV_CHILD_METADATA
| BDRV_CHILD_DATA
));
2722 flags
= bdrv_reopen_get_flags(reopen_queue
, bs
);
2725 * Apart from the modifications below, the same permissions are
2726 * forwarded and left alone as for filters
2728 bdrv_filter_default_perms(bs
, c
, role
, reopen_queue
,
2729 perm
, shared
, &perm
, &shared
);
2731 if (role
& BDRV_CHILD_METADATA
) {
2732 /* Format drivers may touch metadata even if the guest doesn't write */
2733 if (bdrv_is_writable_after_reopen(bs
, reopen_queue
)) {
2734 perm
|= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2738 * bs->file always needs to be consistent because of the
2739 * metadata. We can never allow other users to resize or write
2742 if (!(flags
& BDRV_O_NO_IO
)) {
2743 perm
|= BLK_PERM_CONSISTENT_READ
;
2745 shared
&= ~(BLK_PERM_WRITE
| BLK_PERM_RESIZE
);
2748 if (role
& BDRV_CHILD_DATA
) {
2750 * Technically, everything in this block is a subset of the
2751 * BDRV_CHILD_METADATA path taken above, and so this could
2752 * be an "else if" branch. However, that is not obvious, and
2753 * this function is not performance critical, therefore we let
2754 * this be an independent "if".
2758 * We cannot allow other users to resize the file because the
2759 * format driver might have some assumptions about the size
2760 * (e.g. because it is stored in metadata, or because the file
2761 * is split into fixed-size data files).
2763 shared
&= ~BLK_PERM_RESIZE
;
2766 * WRITE_UNCHANGED often cannot be performed as such on the
2767 * data file. For example, the qcow2 driver may still need to
2768 * write copied clusters on copy-on-read.
2770 if (perm
& BLK_PERM_WRITE_UNCHANGED
) {
2771 perm
|= BLK_PERM_WRITE
;
2775 * If the data file is written to, the format driver may
2776 * expect to be able to resize it by writing beyond the EOF.
2778 if (perm
& BLK_PERM_WRITE
) {
2779 perm
|= BLK_PERM_RESIZE
;
2783 if (bs
->open_flags
& BDRV_O_INACTIVE
) {
2784 shared
|= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2791 void bdrv_default_perms(BlockDriverState
*bs
, BdrvChild
*c
,
2792 BdrvChildRole role
, BlockReopenQueue
*reopen_queue
,
2793 uint64_t perm
, uint64_t shared
,
2794 uint64_t *nperm
, uint64_t *nshared
)
2796 GLOBAL_STATE_CODE();
2797 if (role
& BDRV_CHILD_FILTERED
) {
2798 assert(!(role
& (BDRV_CHILD_DATA
| BDRV_CHILD_METADATA
|
2800 bdrv_filter_default_perms(bs
, c
, role
, reopen_queue
,
2801 perm
, shared
, nperm
, nshared
);
2802 } else if (role
& BDRV_CHILD_COW
) {
2803 assert(!(role
& (BDRV_CHILD_DATA
| BDRV_CHILD_METADATA
)));
2804 bdrv_default_perms_for_cow(bs
, c
, role
, reopen_queue
,
2805 perm
, shared
, nperm
, nshared
);
2806 } else if (role
& (BDRV_CHILD_METADATA
| BDRV_CHILD_DATA
)) {
2807 bdrv_default_perms_for_storage(bs
, c
, role
, reopen_queue
,
2808 perm
, shared
, nperm
, nshared
);
2810 g_assert_not_reached();
2814 uint64_t bdrv_qapi_perm_to_blk_perm(BlockPermission qapi_perm
)
2816 static const uint64_t permissions
[] = {
2817 [BLOCK_PERMISSION_CONSISTENT_READ
] = BLK_PERM_CONSISTENT_READ
,
2818 [BLOCK_PERMISSION_WRITE
] = BLK_PERM_WRITE
,
2819 [BLOCK_PERMISSION_WRITE_UNCHANGED
] = BLK_PERM_WRITE_UNCHANGED
,
2820 [BLOCK_PERMISSION_RESIZE
] = BLK_PERM_RESIZE
,
2823 QEMU_BUILD_BUG_ON(ARRAY_SIZE(permissions
) != BLOCK_PERMISSION__MAX
);
2824 QEMU_BUILD_BUG_ON(1UL << ARRAY_SIZE(permissions
) != BLK_PERM_ALL
+ 1);
2826 assert(qapi_perm
< BLOCK_PERMISSION__MAX
);
2828 return permissions
[qapi_perm
];
2832 * Replace (*childp)->bs by @new_bs.
2834 * If @new_bs is NULL, *childp will be set to NULL, too: BDS parents
2835 * generally cannot handle a BdrvChild with .bs == NULL, so clearing
2836 * BdrvChild.bs should generally immediately be followed by the
2837 * BdrvChild pointer being cleared as well.
2839 * If @free_empty_child is true and @new_bs is NULL, the BdrvChild is
2840 * freed. @free_empty_child should only be false if the caller will
2841 * free the BdrvChild themselves (this may be important in a
2842 * transactional context, where it may only be freed on commit).
2844 static void bdrv_replace_child_noperm(BdrvChild
**childp
,
2845 BlockDriverState
*new_bs
,
2846 bool free_empty_child
)
2848 BdrvChild
*child
= *childp
;
2849 BlockDriverState
*old_bs
= child
->bs
;
2850 int new_bs_quiesce_counter
;
2853 assert(!child
->frozen
);
2854 assert(old_bs
!= new_bs
);
2855 GLOBAL_STATE_CODE();
2857 if (old_bs
&& new_bs
) {
2858 assert(bdrv_get_aio_context(old_bs
) == bdrv_get_aio_context(new_bs
));
2861 new_bs_quiesce_counter
= (new_bs
? new_bs
->quiesce_counter
: 0);
2862 drain_saldo
= new_bs_quiesce_counter
- child
->parent_quiesce_counter
;
2865 * If the new child node is drained but the old one was not, flush
2866 * all outstanding requests to the old child node.
2868 while (drain_saldo
> 0 && child
->klass
->drained_begin
) {
2869 bdrv_parent_drained_begin_single(child
, true);
2874 /* Detach first so that the recursive drain sections coming from @child
2875 * are already gone and we only end the drain sections that came from
2877 if (child
->klass
->detach
) {
2878 child
->klass
->detach(child
);
2880 assert_bdrv_graph_writable(old_bs
);
2881 QLIST_REMOVE(child
, next_parent
);
2890 assert_bdrv_graph_writable(new_bs
);
2891 QLIST_INSERT_HEAD(&new_bs
->parents
, child
, next_parent
);
2894 * Detaching the old node may have led to the new node's
2895 * quiesce_counter having been decreased. Not a problem, we
2896 * just need to recognize this here and then invoke
2897 * drained_end appropriately more often.
2899 assert(new_bs
->quiesce_counter
<= new_bs_quiesce_counter
);
2900 drain_saldo
+= new_bs
->quiesce_counter
- new_bs_quiesce_counter
;
2902 /* Attach only after starting new drained sections, so that recursive
2903 * drain sections coming from @child don't get an extra .drained_begin
2905 if (child
->klass
->attach
) {
2906 child
->klass
->attach(child
);
2911 * If the old child node was drained but the new one is not, allow
2912 * requests to come in only after the new node has been attached.
2914 while (drain_saldo
< 0 && child
->klass
->drained_end
) {
2915 bdrv_parent_drained_end_single(child
);
2919 if (free_empty_child
&& !child
->bs
) {
2920 bdrv_child_free(child
);
2925 * Free the given @child.
2927 * The child must be empty (i.e. `child->bs == NULL`) and it must be
2928 * unused (i.e. not in a children list).
2930 static void bdrv_child_free(BdrvChild
*child
)
2933 GLOBAL_STATE_CODE();
2934 assert(!child
->next
.le_prev
); /* not in children list */
2936 g_free(child
->name
);
2940 typedef struct BdrvAttachChildCommonState
{
2942 AioContext
*old_parent_ctx
;
2943 AioContext
*old_child_ctx
;
2944 } BdrvAttachChildCommonState
;
2946 static void bdrv_attach_child_common_abort(void *opaque
)
2948 BdrvAttachChildCommonState
*s
= opaque
;
2949 BdrvChild
*child
= *s
->child
;
2950 BlockDriverState
*bs
= child
->bs
;
2952 GLOBAL_STATE_CODE();
2954 * Pass free_empty_child=false, because we still need the child
2955 * for the AioContext operations on the parent below; those
2956 * BdrvChildClass methods all work on a BdrvChild object, so we
2957 * need to keep it as an empty shell (after this function, it will
2958 * not be attached to any parent, and it will not have a .bs).
2960 bdrv_replace_child_noperm(s
->child
, NULL
, false);
2962 if (bdrv_get_aio_context(bs
) != s
->old_child_ctx
) {
2963 bdrv_try_set_aio_context(bs
, s
->old_child_ctx
, &error_abort
);
2966 if (bdrv_child_get_parent_aio_context(child
) != s
->old_parent_ctx
) {
2969 /* No need to ignore `child`, because it has been detached already */
2971 child
->klass
->can_set_aio_ctx(child
, s
->old_parent_ctx
, &ignore
,
2973 g_slist_free(ignore
);
2976 child
->klass
->set_aio_ctx(child
, s
->old_parent_ctx
, &ignore
);
2977 g_slist_free(ignore
);
2981 bdrv_child_free(child
);
2984 static TransactionActionDrv bdrv_attach_child_common_drv
= {
2985 .abort
= bdrv_attach_child_common_abort
,
2990 * Common part of attaching bdrv child to bs or to blk or to job
2992 * Resulting new child is returned through @child.
2993 * At start *@child must be NULL.
2994 * @child is saved to a new entry of @tran, so that *@child could be reverted to
2995 * NULL on abort(). So referenced variable must live at least until transaction
2998 * Function doesn't update permissions, caller is responsible for this.
3000 static int bdrv_attach_child_common(BlockDriverState
*child_bs
,
3001 const char *child_name
,
3002 const BdrvChildClass
*child_class
,
3003 BdrvChildRole child_role
,
3004 uint64_t perm
, uint64_t shared_perm
,
3005 void *opaque
, BdrvChild
**child
,
3006 Transaction
*tran
, Error
**errp
)
3008 BdrvChild
*new_child
;
3009 AioContext
*parent_ctx
;
3010 AioContext
*child_ctx
= bdrv_get_aio_context(child_bs
);
3013 assert(*child
== NULL
);
3014 assert(child_class
->get_parent_desc
);
3015 GLOBAL_STATE_CODE();
3017 new_child
= g_new(BdrvChild
, 1);
3018 *new_child
= (BdrvChild
) {
3020 .name
= g_strdup(child_name
),
3021 .klass
= child_class
,
3024 .shared_perm
= shared_perm
,
3029 * If the AioContexts don't match, first try to move the subtree of
3030 * child_bs into the AioContext of the new parent. If this doesn't work,
3031 * try moving the parent into the AioContext of child_bs instead.
3033 parent_ctx
= bdrv_child_get_parent_aio_context(new_child
);
3034 if (child_ctx
!= parent_ctx
) {
3035 Error
*local_err
= NULL
;
3036 int ret
= bdrv_try_set_aio_context(child_bs
, parent_ctx
, &local_err
);
3038 if (ret
< 0 && child_class
->can_set_aio_ctx
) {
3039 GSList
*ignore
= g_slist_prepend(NULL
, new_child
);
3040 if (child_class
->can_set_aio_ctx(new_child
, child_ctx
, &ignore
,
3043 error_free(local_err
);
3045 g_slist_free(ignore
);
3046 ignore
= g_slist_prepend(NULL
, new_child
);
3047 child_class
->set_aio_ctx(new_child
, child_ctx
, &ignore
);
3049 g_slist_free(ignore
);
3053 error_propagate(errp
, local_err
);
3054 bdrv_child_free(new_child
);
3060 bdrv_replace_child_noperm(&new_child
, child_bs
, true);
3061 /* child_bs was non-NULL, so new_child must not have been freed */
3062 assert(new_child
!= NULL
);
3066 BdrvAttachChildCommonState
*s
= g_new(BdrvAttachChildCommonState
, 1);
3067 *s
= (BdrvAttachChildCommonState
) {
3069 .old_parent_ctx
= parent_ctx
,
3070 .old_child_ctx
= child_ctx
,
3072 tran_add(tran
, &bdrv_attach_child_common_drv
, s
);
3078 * Variable referenced by @child must live at least until transaction end.
3079 * (see bdrv_attach_child_common() doc for details)
3081 * Function doesn't update permissions, caller is responsible for this.
3083 static int bdrv_attach_child_noperm(BlockDriverState
*parent_bs
,
3084 BlockDriverState
*child_bs
,
3085 const char *child_name
,
3086 const BdrvChildClass
*child_class
,
3087 BdrvChildRole child_role
,
3093 uint64_t perm
, shared_perm
;
3095 assert(parent_bs
->drv
);
3096 GLOBAL_STATE_CODE();
3098 if (bdrv_recurse_has_child(child_bs
, parent_bs
)) {
3099 error_setg(errp
, "Making '%s' a %s child of '%s' would create a cycle",
3100 child_bs
->node_name
, child_name
, parent_bs
->node_name
);
3104 bdrv_get_cumulative_perm(parent_bs
, &perm
, &shared_perm
);
3105 bdrv_child_perm(parent_bs
, child_bs
, NULL
, child_role
, NULL
,
3106 perm
, shared_perm
, &perm
, &shared_perm
);
3108 ret
= bdrv_attach_child_common(child_bs
, child_name
, child_class
,
3109 child_role
, perm
, shared_perm
, parent_bs
,
3118 static void bdrv_detach_child(BdrvChild
**childp
)
3120 BlockDriverState
*old_bs
= (*childp
)->bs
;
3122 GLOBAL_STATE_CODE();
3123 bdrv_replace_child_noperm(childp
, NULL
, true);
3127 * Update permissions for old node. We're just taking a parent away, so
3128 * we're loosening restrictions. Errors of permission update are not
3129 * fatal in this case, ignore them.
3131 bdrv_refresh_perms(old_bs
, NULL
);
3134 * When the parent requiring a non-default AioContext is removed, the
3135 * node moves back to the main AioContext
3137 bdrv_try_set_aio_context(old_bs
, qemu_get_aio_context(), NULL
);
3142 * This function steals the reference to child_bs from the caller.
3143 * That reference is later dropped by bdrv_root_unref_child().
3145 * On failure NULL is returned, errp is set and the reference to
3146 * child_bs is also dropped.
3148 * The caller must hold the AioContext lock @child_bs, but not that of @ctx
3149 * (unless @child_bs is already in @ctx).
3151 BdrvChild
*bdrv_root_attach_child(BlockDriverState
*child_bs
,
3152 const char *child_name
,
3153 const BdrvChildClass
*child_class
,
3154 BdrvChildRole child_role
,
3155 uint64_t perm
, uint64_t shared_perm
,
3156 void *opaque
, Error
**errp
)
3159 BdrvChild
*child
= NULL
;
3160 Transaction
*tran
= tran_new();
3162 GLOBAL_STATE_CODE();
3164 ret
= bdrv_attach_child_common(child_bs
, child_name
, child_class
,
3165 child_role
, perm
, shared_perm
, opaque
,
3166 &child
, tran
, errp
);
3171 ret
= bdrv_refresh_perms(child_bs
, errp
);
3174 tran_finalize(tran
, ret
);
3175 /* child is unset on failure by bdrv_attach_child_common_abort() */
3176 assert((ret
< 0) == !child
);
3178 bdrv_unref(child_bs
);
3183 * This function transfers the reference to child_bs from the caller
3184 * to parent_bs. That reference is later dropped by parent_bs on
3185 * bdrv_close() or if someone calls bdrv_unref_child().
3187 * On failure NULL is returned, errp is set and the reference to
3188 * child_bs is also dropped.
3190 * If @parent_bs and @child_bs are in different AioContexts, the caller must
3191 * hold the AioContext lock for @child_bs, but not for @parent_bs.
3193 BdrvChild
*bdrv_attach_child(BlockDriverState
*parent_bs
,
3194 BlockDriverState
*child_bs
,
3195 const char *child_name
,
3196 const BdrvChildClass
*child_class
,
3197 BdrvChildRole child_role
,
3201 BdrvChild
*child
= NULL
;
3202 Transaction
*tran
= tran_new();
3204 GLOBAL_STATE_CODE();
3206 ret
= bdrv_attach_child_noperm(parent_bs
, child_bs
, child_name
, child_class
,
3207 child_role
, &child
, tran
, errp
);
3212 ret
= bdrv_refresh_perms(parent_bs
, errp
);
3218 tran_finalize(tran
, ret
);
3219 /* child is unset on failure by bdrv_attach_child_common_abort() */
3220 assert((ret
< 0) == !child
);
3222 bdrv_unref(child_bs
);
3227 /* Callers must ensure that child->frozen is false. */
3228 void bdrv_root_unref_child(BdrvChild
*child
)
3230 BlockDriverState
*child_bs
;
3232 GLOBAL_STATE_CODE();
3234 child_bs
= child
->bs
;
3235 bdrv_detach_child(&child
);
3236 bdrv_unref(child_bs
);
3239 typedef struct BdrvSetInheritsFrom
{
3240 BlockDriverState
*bs
;
3241 BlockDriverState
*old_inherits_from
;
3242 } BdrvSetInheritsFrom
;
3244 static void bdrv_set_inherits_from_abort(void *opaque
)
3246 BdrvSetInheritsFrom
*s
= opaque
;
3248 s
->bs
->inherits_from
= s
->old_inherits_from
;
3251 static TransactionActionDrv bdrv_set_inherits_from_drv
= {
3252 .abort
= bdrv_set_inherits_from_abort
,
3256 /* @tran is allowed to be NULL. In this case no rollback is possible */
3257 static void bdrv_set_inherits_from(BlockDriverState
*bs
,
3258 BlockDriverState
*new_inherits_from
,
3262 BdrvSetInheritsFrom
*s
= g_new(BdrvSetInheritsFrom
, 1);
3264 *s
= (BdrvSetInheritsFrom
) {
3266 .old_inherits_from
= bs
->inherits_from
,
3269 tran_add(tran
, &bdrv_set_inherits_from_drv
, s
);
3272 bs
->inherits_from
= new_inherits_from
;
3276 * Clear all inherits_from pointers from children and grandchildren of
3277 * @root that point to @root, where necessary.
3278 * @tran is allowed to be NULL. In this case no rollback is possible
3280 static void bdrv_unset_inherits_from(BlockDriverState
*root
, BdrvChild
*child
,
3285 if (child
->bs
->inherits_from
== root
) {
3287 * Remove inherits_from only when the last reference between root and
3288 * child->bs goes away.
3290 QLIST_FOREACH(c
, &root
->children
, next
) {
3291 if (c
!= child
&& c
->bs
== child
->bs
) {
3296 bdrv_set_inherits_from(child
->bs
, NULL
, tran
);
3300 QLIST_FOREACH(c
, &child
->bs
->children
, next
) {
3301 bdrv_unset_inherits_from(root
, c
, tran
);
3305 /* Callers must ensure that child->frozen is false. */
3306 void bdrv_unref_child(BlockDriverState
*parent
, BdrvChild
*child
)
3308 GLOBAL_STATE_CODE();
3309 if (child
== NULL
) {
3313 bdrv_unset_inherits_from(parent
, child
, NULL
);
3314 bdrv_root_unref_child(child
);
3318 static void bdrv_parent_cb_change_media(BlockDriverState
*bs
, bool load
)
3321 GLOBAL_STATE_CODE();
3322 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
3323 if (c
->klass
->change_media
) {
3324 c
->klass
->change_media(c
, load
);
3329 /* Return true if you can reach parent going through child->inherits_from
3330 * recursively. If parent or child are NULL, return false */
3331 static bool bdrv_inherits_from_recursive(BlockDriverState
*child
,
3332 BlockDriverState
*parent
)
3334 while (child
&& child
!= parent
) {
3335 child
= child
->inherits_from
;
3338 return child
!= NULL
;
3342 * Return the BdrvChildRole for @bs's backing child. bs->backing is
3343 * mostly used for COW backing children (role = COW), but also for
3344 * filtered children (role = FILTERED | PRIMARY).
3346 static BdrvChildRole
bdrv_backing_role(BlockDriverState
*bs
)
3348 if (bs
->drv
&& bs
->drv
->is_filter
) {
3349 return BDRV_CHILD_FILTERED
| BDRV_CHILD_PRIMARY
;
3351 return BDRV_CHILD_COW
;
3356 * Sets the bs->backing or bs->file link of a BDS. A new reference is created;
3357 * callers which don't need their own reference any more must call bdrv_unref().
3359 * Function doesn't update permissions, caller is responsible for this.
3361 static int bdrv_set_file_or_backing_noperm(BlockDriverState
*parent_bs
,
3362 BlockDriverState
*child_bs
,
3364 Transaction
*tran
, Error
**errp
)
3367 bool update_inherits_from
=
3368 bdrv_inherits_from_recursive(child_bs
, parent_bs
);
3369 BdrvChild
*child
= is_backing
? parent_bs
->backing
: parent_bs
->file
;
3372 GLOBAL_STATE_CODE();
3374 if (!parent_bs
->drv
) {
3376 * Node without drv is an object without a class :/. TODO: finally fix
3377 * qcow2 driver to never clear bs->drv and implement format corruption
3378 * handling in other way.
3380 error_setg(errp
, "Node corrupted");
3384 if (child
&& child
->frozen
) {
3385 error_setg(errp
, "Cannot change frozen '%s' link from '%s' to '%s'",
3386 child
->name
, parent_bs
->node_name
, child
->bs
->node_name
);
3390 if (is_backing
&& !parent_bs
->drv
->is_filter
&&
3391 !parent_bs
->drv
->supports_backing
)
3393 error_setg(errp
, "Driver '%s' of node '%s' does not support backing "
3394 "files", parent_bs
->drv
->format_name
, parent_bs
->node_name
);
3398 if (parent_bs
->drv
->is_filter
) {
3399 role
= BDRV_CHILD_FILTERED
| BDRV_CHILD_PRIMARY
;
3400 } else if (is_backing
) {
3401 role
= BDRV_CHILD_COW
;
3404 * We only can use same role as it is in existing child. We don't have
3405 * infrastructure to determine role of file child in generic way
3408 error_setg(errp
, "Cannot set file child to format node without "
3416 bdrv_unset_inherits_from(parent_bs
, child
, tran
);
3417 bdrv_remove_file_or_backing_child(parent_bs
, child
, tran
);
3424 ret
= bdrv_attach_child_noperm(parent_bs
, child_bs
,
3425 is_backing
? "backing" : "file",
3426 &child_of_bds
, role
,
3427 is_backing
? &parent_bs
->backing
:
3436 * If inherits_from pointed recursively to bs then let's update it to
3437 * point directly to bs (else it will become NULL).
3439 if (update_inherits_from
) {
3440 bdrv_set_inherits_from(child_bs
, parent_bs
, tran
);
3444 bdrv_refresh_limits(parent_bs
, tran
, NULL
);
3449 static int bdrv_set_backing_noperm(BlockDriverState
*bs
,
3450 BlockDriverState
*backing_hd
,
3451 Transaction
*tran
, Error
**errp
)
3453 GLOBAL_STATE_CODE();
3454 return bdrv_set_file_or_backing_noperm(bs
, backing_hd
, true, tran
, errp
);
3457 int bdrv_set_backing_hd(BlockDriverState
*bs
, BlockDriverState
*backing_hd
,
3461 Transaction
*tran
= tran_new();
3463 GLOBAL_STATE_CODE();
3464 bdrv_drained_begin(bs
);
3466 ret
= bdrv_set_backing_noperm(bs
, backing_hd
, tran
, errp
);
3471 ret
= bdrv_refresh_perms(bs
, errp
);
3473 tran_finalize(tran
, ret
);
3475 bdrv_drained_end(bs
);
3481 * Opens the backing file for a BlockDriverState if not yet open
3483 * bdref_key specifies the key for the image's BlockdevRef in the options QDict.
3484 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
3485 * itself, all options starting with "${bdref_key}." are considered part of the
3488 * TODO Can this be unified with bdrv_open_image()?
3490 int bdrv_open_backing_file(BlockDriverState
*bs
, QDict
*parent_options
,
3491 const char *bdref_key
, Error
**errp
)
3493 char *backing_filename
= NULL
;
3494 char *bdref_key_dot
;
3495 const char *reference
= NULL
;
3497 bool implicit_backing
= false;
3498 BlockDriverState
*backing_hd
;
3500 QDict
*tmp_parent_options
= NULL
;
3501 Error
*local_err
= NULL
;
3503 GLOBAL_STATE_CODE();
3505 if (bs
->backing
!= NULL
) {
3509 /* NULL means an empty set of options */
3510 if (parent_options
== NULL
) {
3511 tmp_parent_options
= qdict_new();
3512 parent_options
= tmp_parent_options
;
3515 bs
->open_flags
&= ~BDRV_O_NO_BACKING
;
3517 bdref_key_dot
= g_strdup_printf("%s.", bdref_key
);
3518 qdict_extract_subqdict(parent_options
, &options
, bdref_key_dot
);
3519 g_free(bdref_key_dot
);
3522 * Caution: while qdict_get_try_str() is fine, getting non-string
3523 * types would require more care. When @parent_options come from
3524 * -blockdev or blockdev_add, its members are typed according to
3525 * the QAPI schema, but when they come from -drive, they're all
3528 reference
= qdict_get_try_str(parent_options
, bdref_key
);
3529 if (reference
|| qdict_haskey(options
, "file.filename")) {
3530 /* keep backing_filename NULL */
3531 } else if (bs
->backing_file
[0] == '\0' && qdict_size(options
) == 0) {
3532 qobject_unref(options
);
3535 if (qdict_size(options
) == 0) {
3536 /* If the user specifies options that do not modify the
3537 * backing file's behavior, we might still consider it the
3538 * implicit backing file. But it's easier this way, and
3539 * just specifying some of the backing BDS's options is
3540 * only possible with -drive anyway (otherwise the QAPI
3541 * schema forces the user to specify everything). */
3542 implicit_backing
= !strcmp(bs
->auto_backing_file
, bs
->backing_file
);
3545 backing_filename
= bdrv_get_full_backing_filename(bs
, &local_err
);
3548 error_propagate(errp
, local_err
);
3549 qobject_unref(options
);
3554 if (!bs
->drv
|| !bs
->drv
->supports_backing
) {
3556 error_setg(errp
, "Driver doesn't support backing files");
3557 qobject_unref(options
);
3562 bs
->backing_format
[0] != '\0' && !qdict_haskey(options
, "driver")) {
3563 qdict_put_str(options
, "driver", bs
->backing_format
);
3566 backing_hd
= bdrv_open_inherit(backing_filename
, reference
, options
, 0, bs
,
3567 &child_of_bds
, bdrv_backing_role(bs
), errp
);
3569 bs
->open_flags
|= BDRV_O_NO_BACKING
;
3570 error_prepend(errp
, "Could not open backing file: ");
3575 if (implicit_backing
) {
3576 bdrv_refresh_filename(backing_hd
);
3577 pstrcpy(bs
->auto_backing_file
, sizeof(bs
->auto_backing_file
),
3578 backing_hd
->filename
);
3581 /* Hook up the backing file link; drop our reference, bs owns the
3582 * backing_hd reference now */
3583 ret
= bdrv_set_backing_hd(bs
, backing_hd
, errp
);
3584 bdrv_unref(backing_hd
);
3589 qdict_del(parent_options
, bdref_key
);
3592 g_free(backing_filename
);
3593 qobject_unref(tmp_parent_options
);
3597 static BlockDriverState
*
3598 bdrv_open_child_bs(const char *filename
, QDict
*options
, const char *bdref_key
,
3599 BlockDriverState
*parent
, const BdrvChildClass
*child_class
,
3600 BdrvChildRole child_role
, bool allow_none
, Error
**errp
)
3602 BlockDriverState
*bs
= NULL
;
3603 QDict
*image_options
;
3604 char *bdref_key_dot
;
3605 const char *reference
;
3607 assert(child_class
!= NULL
);
3609 bdref_key_dot
= g_strdup_printf("%s.", bdref_key
);
3610 qdict_extract_subqdict(options
, &image_options
, bdref_key_dot
);
3611 g_free(bdref_key_dot
);
3614 * Caution: while qdict_get_try_str() is fine, getting non-string
3615 * types would require more care. When @options come from
3616 * -blockdev or blockdev_add, its members are typed according to
3617 * the QAPI schema, but when they come from -drive, they're all
3620 reference
= qdict_get_try_str(options
, bdref_key
);
3621 if (!filename
&& !reference
&& !qdict_size(image_options
)) {
3623 error_setg(errp
, "A block device must be specified for \"%s\"",
3626 qobject_unref(image_options
);
3630 bs
= bdrv_open_inherit(filename
, reference
, image_options
, 0,
3631 parent
, child_class
, child_role
, errp
);
3637 qdict_del(options
, bdref_key
);
3642 * Opens a disk image whose options are given as BlockdevRef in another block
3645 * If allow_none is true, no image will be opened if filename is false and no
3646 * BlockdevRef is given. NULL will be returned, but errp remains unset.
3648 * bdrev_key specifies the key for the image's BlockdevRef in the options QDict.
3649 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
3650 * itself, all options starting with "${bdref_key}." are considered part of the
3653 * The BlockdevRef will be removed from the options QDict.
3655 BdrvChild
*bdrv_open_child(const char *filename
,
3656 QDict
*options
, const char *bdref_key
,
3657 BlockDriverState
*parent
,
3658 const BdrvChildClass
*child_class
,
3659 BdrvChildRole child_role
,
3660 bool allow_none
, Error
**errp
)
3662 BlockDriverState
*bs
;
3664 GLOBAL_STATE_CODE();
3666 bs
= bdrv_open_child_bs(filename
, options
, bdref_key
, parent
, child_class
,
3667 child_role
, allow_none
, errp
);
3672 return bdrv_attach_child(parent
, bs
, bdref_key
, child_class
, child_role
,
3677 * TODO Future callers may need to specify parent/child_class in order for
3678 * option inheritance to work. Existing callers use it for the root node.
3680 BlockDriverState
*bdrv_open_blockdev_ref(BlockdevRef
*ref
, Error
**errp
)
3682 BlockDriverState
*bs
= NULL
;
3683 QObject
*obj
= NULL
;
3684 QDict
*qdict
= NULL
;
3685 const char *reference
= NULL
;
3688 GLOBAL_STATE_CODE();
3690 if (ref
->type
== QTYPE_QSTRING
) {
3691 reference
= ref
->u
.reference
;
3693 BlockdevOptions
*options
= &ref
->u
.definition
;
3694 assert(ref
->type
== QTYPE_QDICT
);
3696 v
= qobject_output_visitor_new(&obj
);
3697 visit_type_BlockdevOptions(v
, NULL
, &options
, &error_abort
);
3698 visit_complete(v
, &obj
);
3700 qdict
= qobject_to(QDict
, obj
);
3701 qdict_flatten(qdict
);
3703 /* bdrv_open_inherit() defaults to the values in bdrv_flags (for
3704 * compatibility with other callers) rather than what we want as the
3705 * real defaults. Apply the defaults here instead. */
3706 qdict_set_default_str(qdict
, BDRV_OPT_CACHE_DIRECT
, "off");
3707 qdict_set_default_str(qdict
, BDRV_OPT_CACHE_NO_FLUSH
, "off");
3708 qdict_set_default_str(qdict
, BDRV_OPT_READ_ONLY
, "off");
3709 qdict_set_default_str(qdict
, BDRV_OPT_AUTO_READ_ONLY
, "off");
3713 bs
= bdrv_open_inherit(NULL
, reference
, qdict
, 0, NULL
, NULL
, 0, errp
);
3720 static BlockDriverState
*bdrv_append_temp_snapshot(BlockDriverState
*bs
,
3722 QDict
*snapshot_options
,
3725 g_autofree
char *tmp_filename
= NULL
;
3727 QemuOpts
*opts
= NULL
;
3728 BlockDriverState
*bs_snapshot
= NULL
;
3731 GLOBAL_STATE_CODE();
3733 /* if snapshot, we create a temporary backing file and open it
3734 instead of opening 'filename' directly */
3736 /* Get the required size from the image */
3737 total_size
= bdrv_getlength(bs
);
3738 if (total_size
< 0) {
3739 error_setg_errno(errp
, -total_size
, "Could not get image size");
3743 /* Create the temporary image */
3744 tmp_filename
= create_tmp_file(errp
);
3745 if (!tmp_filename
) {
3749 opts
= qemu_opts_create(bdrv_qcow2
.create_opts
, NULL
, 0,
3751 qemu_opt_set_number(opts
, BLOCK_OPT_SIZE
, total_size
, &error_abort
);
3752 ret
= bdrv_create(&bdrv_qcow2
, tmp_filename
, opts
, errp
);
3753 qemu_opts_del(opts
);
3755 error_prepend(errp
, "Could not create temporary overlay '%s': ",
3760 /* Prepare options QDict for the temporary file */
3761 qdict_put_str(snapshot_options
, "file.driver", "file");
3762 qdict_put_str(snapshot_options
, "file.filename", tmp_filename
);
3763 qdict_put_str(snapshot_options
, "driver", "qcow2");
3765 bs_snapshot
= bdrv_open(NULL
, NULL
, snapshot_options
, flags
, errp
);
3766 snapshot_options
= NULL
;
3771 ret
= bdrv_append(bs_snapshot
, bs
, errp
);
3778 qobject_unref(snapshot_options
);
3783 * Opens a disk image (raw, qcow2, vmdk, ...)
3785 * options is a QDict of options to pass to the block drivers, or NULL for an
3786 * empty set of options. The reference to the QDict belongs to the block layer
3787 * after the call (even on failure), so if the caller intends to reuse the
3788 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
3790 * If *pbs is NULL, a new BDS will be created with a pointer to it stored there.
3791 * If it is not NULL, the referenced BDS will be reused.
3793 * The reference parameter may be used to specify an existing block device which
3794 * should be opened. If specified, neither options nor a filename may be given,
3795 * nor can an existing BDS be reused (that is, *pbs has to be NULL).
3797 static BlockDriverState
*bdrv_open_inherit(const char *filename
,
3798 const char *reference
,
3799 QDict
*options
, int flags
,
3800 BlockDriverState
*parent
,
3801 const BdrvChildClass
*child_class
,
3802 BdrvChildRole child_role
,
3806 BlockBackend
*file
= NULL
;
3807 BlockDriverState
*bs
;
3808 BlockDriver
*drv
= NULL
;
3810 const char *drvname
;
3811 const char *backing
;
3812 Error
*local_err
= NULL
;
3813 QDict
*snapshot_options
= NULL
;
3814 int snapshot_flags
= 0;
3816 assert(!child_class
|| !flags
);
3817 assert(!child_class
== !parent
);
3818 GLOBAL_STATE_CODE();
3821 bool options_non_empty
= options
? qdict_size(options
) : false;
3822 qobject_unref(options
);
3824 if (filename
|| options_non_empty
) {
3825 error_setg(errp
, "Cannot reference an existing block device with "
3826 "additional options or a new filename");
3830 bs
= bdrv_lookup_bs(reference
, reference
, errp
);
3841 /* NULL means an empty set of options */
3842 if (options
== NULL
) {
3843 options
= qdict_new();
3846 /* json: syntax counts as explicit options, as if in the QDict */
3847 parse_json_protocol(options
, &filename
, &local_err
);
3852 bs
->explicit_options
= qdict_clone_shallow(options
);
3855 bool parent_is_format
;
3858 parent_is_format
= parent
->drv
->is_format
;
3861 * parent->drv is not set yet because this node is opened for
3862 * (potential) format probing. That means that @parent is going
3863 * to be a format node.
3865 parent_is_format
= true;
3868 bs
->inherits_from
= parent
;
3869 child_class
->inherit_options(child_role
, parent_is_format
,
3871 parent
->open_flags
, parent
->options
);
3874 ret
= bdrv_fill_options(&options
, filename
, &flags
, &local_err
);
3880 * Set the BDRV_O_RDWR and BDRV_O_ALLOW_RDWR flags.
3881 * Caution: getting a boolean member of @options requires care.
3882 * When @options come from -blockdev or blockdev_add, members are
3883 * typed according to the QAPI schema, but when they come from
3884 * -drive, they're all QString.
3886 if (g_strcmp0(qdict_get_try_str(options
, BDRV_OPT_READ_ONLY
), "on") &&
3887 !qdict_get_try_bool(options
, BDRV_OPT_READ_ONLY
, false)) {
3888 flags
|= (BDRV_O_RDWR
| BDRV_O_ALLOW_RDWR
);
3890 flags
&= ~BDRV_O_RDWR
;
3893 if (flags
& BDRV_O_SNAPSHOT
) {
3894 snapshot_options
= qdict_new();
3895 bdrv_temp_snapshot_options(&snapshot_flags
, snapshot_options
,
3897 /* Let bdrv_backing_options() override "read-only" */
3898 qdict_del(options
, BDRV_OPT_READ_ONLY
);
3899 bdrv_inherited_options(BDRV_CHILD_COW
, true,
3900 &flags
, options
, flags
, options
);
3903 bs
->open_flags
= flags
;
3904 bs
->options
= options
;
3905 options
= qdict_clone_shallow(options
);
3907 /* Find the right image format driver */
3908 /* See cautionary note on accessing @options above */
3909 drvname
= qdict_get_try_str(options
, "driver");
3911 drv
= bdrv_find_format(drvname
);
3913 error_setg(errp
, "Unknown driver: '%s'", drvname
);
3918 assert(drvname
|| !(flags
& BDRV_O_PROTOCOL
));
3920 /* See cautionary note on accessing @options above */
3921 backing
= qdict_get_try_str(options
, "backing");
3922 if (qobject_to(QNull
, qdict_get(options
, "backing")) != NULL
||
3923 (backing
&& *backing
== '\0'))
3926 warn_report("Use of \"backing\": \"\" is deprecated; "
3927 "use \"backing\": null instead");
3929 flags
|= BDRV_O_NO_BACKING
;
3930 qdict_del(bs
->explicit_options
, "backing");
3931 qdict_del(bs
->options
, "backing");
3932 qdict_del(options
, "backing");
3935 /* Open image file without format layer. This BlockBackend is only used for
3936 * probing, the block drivers will do their own bdrv_open_child() for the
3937 * same BDS, which is why we put the node name back into options. */
3938 if ((flags
& BDRV_O_PROTOCOL
) == 0) {
3939 BlockDriverState
*file_bs
;
3941 file_bs
= bdrv_open_child_bs(filename
, options
, "file", bs
,
3942 &child_of_bds
, BDRV_CHILD_IMAGE
,
3947 if (file_bs
!= NULL
) {
3948 /* Not requesting BLK_PERM_CONSISTENT_READ because we're only
3949 * looking at the header to guess the image format. This works even
3950 * in cases where a guest would not see a consistent state. */
3951 file
= blk_new(bdrv_get_aio_context(file_bs
), 0, BLK_PERM_ALL
);
3952 blk_insert_bs(file
, file_bs
, &local_err
);
3953 bdrv_unref(file_bs
);
3958 qdict_put_str(options
, "file", bdrv_get_node_name(file_bs
));
3962 /* Image format probing */
3965 ret
= find_image_format(file
, filename
, &drv
, &local_err
);
3970 * This option update would logically belong in bdrv_fill_options(),
3971 * but we first need to open bs->file for the probing to work, while
3972 * opening bs->file already requires the (mostly) final set of options
3973 * so that cache mode etc. can be inherited.
3975 * Adding the driver later is somewhat ugly, but it's not an option
3976 * that would ever be inherited, so it's correct. We just need to make
3977 * sure to update both bs->options (which has the full effective
3978 * options for bs) and options (which has file.* already removed).
3980 qdict_put_str(bs
->options
, "driver", drv
->format_name
);
3981 qdict_put_str(options
, "driver", drv
->format_name
);
3983 error_setg(errp
, "Must specify either driver or file");
3987 /* BDRV_O_PROTOCOL must be set iff a protocol BDS is about to be created */
3988 assert(!!(flags
& BDRV_O_PROTOCOL
) == !!drv
->bdrv_file_open
);
3989 /* file must be NULL if a protocol BDS is about to be created
3990 * (the inverse results in an error message from bdrv_open_common()) */
3991 assert(!(flags
& BDRV_O_PROTOCOL
) || !file
);
3993 /* Open the image */
3994 ret
= bdrv_open_common(bs
, file
, options
, &local_err
);
4004 /* If there is a backing file, use it */
4005 if ((flags
& BDRV_O_NO_BACKING
) == 0) {
4006 ret
= bdrv_open_backing_file(bs
, options
, "backing", &local_err
);
4008 goto close_and_fail
;
4012 /* Remove all children options and references
4013 * from bs->options and bs->explicit_options */
4014 QLIST_FOREACH(child
, &bs
->children
, next
) {
4015 char *child_key_dot
;
4016 child_key_dot
= g_strdup_printf("%s.", child
->name
);
4017 qdict_extract_subqdict(bs
->explicit_options
, NULL
, child_key_dot
);
4018 qdict_extract_subqdict(bs
->options
, NULL
, child_key_dot
);
4019 qdict_del(bs
->explicit_options
, child
->name
);
4020 qdict_del(bs
->options
, child
->name
);
4021 g_free(child_key_dot
);
4024 /* Check if any unknown options were used */
4025 if (qdict_size(options
) != 0) {
4026 const QDictEntry
*entry
= qdict_first(options
);
4027 if (flags
& BDRV_O_PROTOCOL
) {
4028 error_setg(errp
, "Block protocol '%s' doesn't support the option "
4029 "'%s'", drv
->format_name
, entry
->key
);
4032 "Block format '%s' does not support the option '%s'",
4033 drv
->format_name
, entry
->key
);
4036 goto close_and_fail
;
4039 bdrv_parent_cb_change_media(bs
, true);
4041 qobject_unref(options
);
4044 /* For snapshot=on, create a temporary qcow2 overlay. bs points to the
4045 * temporary snapshot afterwards. */
4046 if (snapshot_flags
) {
4047 BlockDriverState
*snapshot_bs
;
4048 snapshot_bs
= bdrv_append_temp_snapshot(bs
, snapshot_flags
,
4049 snapshot_options
, &local_err
);
4050 snapshot_options
= NULL
;
4052 goto close_and_fail
;
4054 /* We are not going to return bs but the overlay on top of it
4055 * (snapshot_bs); thus, we have to drop the strong reference to bs
4056 * (which we obtained by calling bdrv_new()). bs will not be deleted,
4057 * though, because the overlay still has a reference to it. */
4066 qobject_unref(snapshot_options
);
4067 qobject_unref(bs
->explicit_options
);
4068 qobject_unref(bs
->options
);
4069 qobject_unref(options
);
4071 bs
->explicit_options
= NULL
;
4073 error_propagate(errp
, local_err
);
4078 qobject_unref(snapshot_options
);
4079 qobject_unref(options
);
4080 error_propagate(errp
, local_err
);
4084 BlockDriverState
*bdrv_open(const char *filename
, const char *reference
,
4085 QDict
*options
, int flags
, Error
**errp
)
4087 GLOBAL_STATE_CODE();
4089 return bdrv_open_inherit(filename
, reference
, options
, flags
, NULL
,
4093 /* Return true if the NULL-terminated @list contains @str */
4094 static bool is_str_in_list(const char *str
, const char *const *list
)
4098 for (i
= 0; list
[i
] != NULL
; i
++) {
4099 if (!strcmp(str
, list
[i
])) {
4108 * Check that every option set in @bs->options is also set in
4111 * Options listed in the common_options list and in
4112 * @bs->drv->mutable_opts are skipped.
4114 * Return 0 on success, otherwise return -EINVAL and set @errp.
4116 static int bdrv_reset_options_allowed(BlockDriverState
*bs
,
4117 const QDict
*new_opts
, Error
**errp
)
4119 const QDictEntry
*e
;
4120 /* These options are common to all block drivers and are handled
4121 * in bdrv_reopen_prepare() so they can be left out of @new_opts */
4122 const char *const common_options
[] = {
4123 "node-name", "discard", "cache.direct", "cache.no-flush",
4124 "read-only", "auto-read-only", "detect-zeroes", NULL
4127 for (e
= qdict_first(bs
->options
); e
; e
= qdict_next(bs
->options
, e
)) {
4128 if (!qdict_haskey(new_opts
, e
->key
) &&
4129 !is_str_in_list(e
->key
, common_options
) &&
4130 !is_str_in_list(e
->key
, bs
->drv
->mutable_opts
)) {
4131 error_setg(errp
, "Option '%s' cannot be reset "
4132 "to its default value", e
->key
);
4141 * Returns true if @child can be reached recursively from @bs
4143 static bool bdrv_recurse_has_child(BlockDriverState
*bs
,
4144 BlockDriverState
*child
)
4152 QLIST_FOREACH(c
, &bs
->children
, next
) {
4153 if (bdrv_recurse_has_child(c
->bs
, child
)) {
4162 * Adds a BlockDriverState to a simple queue for an atomic, transactional
4163 * reopen of multiple devices.
4165 * bs_queue can either be an existing BlockReopenQueue that has had QTAILQ_INIT
4166 * already performed, or alternatively may be NULL a new BlockReopenQueue will
4167 * be created and initialized. This newly created BlockReopenQueue should be
4168 * passed back in for subsequent calls that are intended to be of the same
4171 * bs is the BlockDriverState to add to the reopen queue.
4173 * options contains the changed options for the associated bs
4174 * (the BlockReopenQueue takes ownership)
4176 * flags contains the open flags for the associated bs
4178 * returns a pointer to bs_queue, which is either the newly allocated
4179 * bs_queue, or the existing bs_queue being used.
4181 * bs must be drained between bdrv_reopen_queue() and bdrv_reopen_multiple().
4183 static BlockReopenQueue
*bdrv_reopen_queue_child(BlockReopenQueue
*bs_queue
,
4184 BlockDriverState
*bs
,
4186 const BdrvChildClass
*klass
,
4188 bool parent_is_format
,
4189 QDict
*parent_options
,
4195 BlockReopenQueueEntry
*bs_entry
;
4197 QDict
*old_options
, *explicit_options
, *options_copy
;
4201 /* Make sure that the caller remembered to use a drained section. This is
4202 * important to avoid graph changes between the recursive queuing here and
4203 * bdrv_reopen_multiple(). */
4204 assert(bs
->quiesce_counter
> 0);
4205 GLOBAL_STATE_CODE();
4207 if (bs_queue
== NULL
) {
4208 bs_queue
= g_new0(BlockReopenQueue
, 1);
4209 QTAILQ_INIT(bs_queue
);
4213 options
= qdict_new();
4216 /* Check if this BlockDriverState is already in the queue */
4217 QTAILQ_FOREACH(bs_entry
, bs_queue
, entry
) {
4218 if (bs
== bs_entry
->state
.bs
) {
4224 * Precedence of options:
4225 * 1. Explicitly passed in options (highest)
4226 * 2. Retained from explicitly set options of bs
4227 * 3. Inherited from parent node
4228 * 4. Retained from effective options of bs
4231 /* Old explicitly set values (don't overwrite by inherited value) */
4232 if (bs_entry
|| keep_old_opts
) {
4233 old_options
= qdict_clone_shallow(bs_entry
?
4234 bs_entry
->state
.explicit_options
:
4235 bs
->explicit_options
);
4236 bdrv_join_options(bs
, options
, old_options
);
4237 qobject_unref(old_options
);
4240 explicit_options
= qdict_clone_shallow(options
);
4242 /* Inherit from parent node */
4243 if (parent_options
) {
4245 klass
->inherit_options(role
, parent_is_format
, &flags
, options
,
4246 parent_flags
, parent_options
);
4248 flags
= bdrv_get_flags(bs
);
4251 if (keep_old_opts
) {
4252 /* Old values are used for options that aren't set yet */
4253 old_options
= qdict_clone_shallow(bs
->options
);
4254 bdrv_join_options(bs
, options
, old_options
);
4255 qobject_unref(old_options
);
4258 /* We have the final set of options so let's update the flags */
4259 options_copy
= qdict_clone_shallow(options
);
4260 opts
= qemu_opts_create(&bdrv_runtime_opts
, NULL
, 0, &error_abort
);
4261 qemu_opts_absorb_qdict(opts
, options_copy
, NULL
);
4262 update_flags_from_options(&flags
, opts
);
4263 qemu_opts_del(opts
);
4264 qobject_unref(options_copy
);
4266 /* bdrv_open_inherit() sets and clears some additional flags internally */
4267 flags
&= ~BDRV_O_PROTOCOL
;
4268 if (flags
& BDRV_O_RDWR
) {
4269 flags
|= BDRV_O_ALLOW_RDWR
;
4273 bs_entry
= g_new0(BlockReopenQueueEntry
, 1);
4274 QTAILQ_INSERT_TAIL(bs_queue
, bs_entry
, entry
);
4276 qobject_unref(bs_entry
->state
.options
);
4277 qobject_unref(bs_entry
->state
.explicit_options
);
4280 bs_entry
->state
.bs
= bs
;
4281 bs_entry
->state
.options
= options
;
4282 bs_entry
->state
.explicit_options
= explicit_options
;
4283 bs_entry
->state
.flags
= flags
;
4286 * If keep_old_opts is false then it means that unspecified
4287 * options must be reset to their original value. We don't allow
4288 * resetting 'backing' but we need to know if the option is
4289 * missing in order to decide if we have to return an error.
4291 if (!keep_old_opts
) {
4292 bs_entry
->state
.backing_missing
=
4293 !qdict_haskey(options
, "backing") &&
4294 !qdict_haskey(options
, "backing.driver");
4297 QLIST_FOREACH(child
, &bs
->children
, next
) {
4298 QDict
*new_child_options
= NULL
;
4299 bool child_keep_old
= keep_old_opts
;
4301 /* reopen can only change the options of block devices that were
4302 * implicitly created and inherited options. For other (referenced)
4303 * block devices, a syntax like "backing.foo" results in an error. */
4304 if (child
->bs
->inherits_from
!= bs
) {
4308 /* Check if the options contain a child reference */
4309 if (qdict_haskey(options
, child
->name
)) {
4310 const char *childref
= qdict_get_try_str(options
, child
->name
);
4312 * The current child must not be reopened if the child
4313 * reference is null or points to a different node.
4315 if (g_strcmp0(childref
, child
->bs
->node_name
)) {
4319 * If the child reference points to the current child then
4320 * reopen it with its existing set of options (note that
4321 * it can still inherit new options from the parent).
4323 child_keep_old
= true;
4325 /* Extract child options ("child-name.*") */
4326 char *child_key_dot
= g_strdup_printf("%s.", child
->name
);
4327 qdict_extract_subqdict(explicit_options
, NULL
, child_key_dot
);
4328 qdict_extract_subqdict(options
, &new_child_options
, child_key_dot
);
4329 g_free(child_key_dot
);
4332 bdrv_reopen_queue_child(bs_queue
, child
->bs
, new_child_options
,
4333 child
->klass
, child
->role
, bs
->drv
->is_format
,
4334 options
, flags
, child_keep_old
);
4340 BlockReopenQueue
*bdrv_reopen_queue(BlockReopenQueue
*bs_queue
,
4341 BlockDriverState
*bs
,
4342 QDict
*options
, bool keep_old_opts
)
4344 GLOBAL_STATE_CODE();
4346 return bdrv_reopen_queue_child(bs_queue
, bs
, options
, NULL
, 0, false,
4347 NULL
, 0, keep_old_opts
);
4350 void bdrv_reopen_queue_free(BlockReopenQueue
*bs_queue
)
4352 GLOBAL_STATE_CODE();
4354 BlockReopenQueueEntry
*bs_entry
, *next
;
4355 QTAILQ_FOREACH_SAFE(bs_entry
, bs_queue
, entry
, next
) {
4356 qobject_unref(bs_entry
->state
.explicit_options
);
4357 qobject_unref(bs_entry
->state
.options
);
4365 * Reopen multiple BlockDriverStates atomically & transactionally.
4367 * The queue passed in (bs_queue) must have been built up previous
4368 * via bdrv_reopen_queue().
4370 * Reopens all BDS specified in the queue, with the appropriate
4371 * flags. All devices are prepared for reopen, and failure of any
4372 * device will cause all device changes to be abandoned, and intermediate
4375 * If all devices prepare successfully, then the changes are committed
4378 * All affected nodes must be drained between bdrv_reopen_queue() and
4379 * bdrv_reopen_multiple().
4381 * To be called from the main thread, with all other AioContexts unlocked.
4383 int bdrv_reopen_multiple(BlockReopenQueue
*bs_queue
, Error
**errp
)
4386 BlockReopenQueueEntry
*bs_entry
, *next
;
4388 Transaction
*tran
= tran_new();
4389 g_autoptr(GHashTable
) found
= NULL
;
4390 g_autoptr(GSList
) refresh_list
= NULL
;
4392 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
4393 assert(bs_queue
!= NULL
);
4394 GLOBAL_STATE_CODE();
4396 QTAILQ_FOREACH(bs_entry
, bs_queue
, entry
) {
4397 ctx
= bdrv_get_aio_context(bs_entry
->state
.bs
);
4398 aio_context_acquire(ctx
);
4399 ret
= bdrv_flush(bs_entry
->state
.bs
);
4400 aio_context_release(ctx
);
4402 error_setg_errno(errp
, -ret
, "Error flushing drive");
4407 QTAILQ_FOREACH(bs_entry
, bs_queue
, entry
) {
4408 assert(bs_entry
->state
.bs
->quiesce_counter
> 0);
4409 ctx
= bdrv_get_aio_context(bs_entry
->state
.bs
);
4410 aio_context_acquire(ctx
);
4411 ret
= bdrv_reopen_prepare(&bs_entry
->state
, bs_queue
, tran
, errp
);
4412 aio_context_release(ctx
);
4416 bs_entry
->prepared
= true;
4419 found
= g_hash_table_new(NULL
, NULL
);
4420 QTAILQ_FOREACH(bs_entry
, bs_queue
, entry
) {
4421 BDRVReopenState
*state
= &bs_entry
->state
;
4423 refresh_list
= bdrv_topological_dfs(refresh_list
, found
, state
->bs
);
4424 if (state
->old_backing_bs
) {
4425 refresh_list
= bdrv_topological_dfs(refresh_list
, found
,
4426 state
->old_backing_bs
);
4428 if (state
->old_file_bs
) {
4429 refresh_list
= bdrv_topological_dfs(refresh_list
, found
,
4430 state
->old_file_bs
);
4435 * Note that file-posix driver rely on permission update done during reopen
4436 * (even if no permission changed), because it wants "new" permissions for
4437 * reconfiguring the fd and that's why it does it in raw_check_perm(), not
4438 * in raw_reopen_prepare() which is called with "old" permissions.
4440 ret
= bdrv_list_refresh_perms(refresh_list
, bs_queue
, tran
, errp
);
4446 * If we reach this point, we have success and just need to apply the
4449 * Reverse order is used to comfort qcow2 driver: on commit it need to write
4450 * IN_USE flag to the image, to mark bitmaps in the image as invalid. But
4451 * children are usually goes after parents in reopen-queue, so go from last
4454 QTAILQ_FOREACH_REVERSE(bs_entry
, bs_queue
, entry
) {
4455 ctx
= bdrv_get_aio_context(bs_entry
->state
.bs
);
4456 aio_context_acquire(ctx
);
4457 bdrv_reopen_commit(&bs_entry
->state
);
4458 aio_context_release(ctx
);
4463 QTAILQ_FOREACH_REVERSE(bs_entry
, bs_queue
, entry
) {
4464 BlockDriverState
*bs
= bs_entry
->state
.bs
;
4466 if (bs
->drv
->bdrv_reopen_commit_post
) {
4467 ctx
= bdrv_get_aio_context(bs
);
4468 aio_context_acquire(ctx
);
4469 bs
->drv
->bdrv_reopen_commit_post(&bs_entry
->state
);
4470 aio_context_release(ctx
);
4479 QTAILQ_FOREACH_SAFE(bs_entry
, bs_queue
, entry
, next
) {
4480 if (bs_entry
->prepared
) {
4481 ctx
= bdrv_get_aio_context(bs_entry
->state
.bs
);
4482 aio_context_acquire(ctx
);
4483 bdrv_reopen_abort(&bs_entry
->state
);
4484 aio_context_release(ctx
);
4489 bdrv_reopen_queue_free(bs_queue
);
4494 int bdrv_reopen(BlockDriverState
*bs
, QDict
*opts
, bool keep_old_opts
,
4497 AioContext
*ctx
= bdrv_get_aio_context(bs
);
4498 BlockReopenQueue
*queue
;
4501 GLOBAL_STATE_CODE();
4503 bdrv_subtree_drained_begin(bs
);
4504 if (ctx
!= qemu_get_aio_context()) {
4505 aio_context_release(ctx
);
4508 queue
= bdrv_reopen_queue(NULL
, bs
, opts
, keep_old_opts
);
4509 ret
= bdrv_reopen_multiple(queue
, errp
);
4511 if (ctx
!= qemu_get_aio_context()) {
4512 aio_context_acquire(ctx
);
4514 bdrv_subtree_drained_end(bs
);
4519 int bdrv_reopen_set_read_only(BlockDriverState
*bs
, bool read_only
,
4522 QDict
*opts
= qdict_new();
4524 GLOBAL_STATE_CODE();
4526 qdict_put_bool(opts
, BDRV_OPT_READ_ONLY
, read_only
);
4528 return bdrv_reopen(bs
, opts
, true, errp
);
4532 * Take a BDRVReopenState and check if the value of 'backing' in the
4533 * reopen_state->options QDict is valid or not.
4535 * If 'backing' is missing from the QDict then return 0.
4537 * If 'backing' contains the node name of the backing file of
4538 * reopen_state->bs then return 0.
4540 * If 'backing' contains a different node name (or is null) then check
4541 * whether the current backing file can be replaced with the new one.
4542 * If that's the case then reopen_state->replace_backing_bs is set to
4543 * true and reopen_state->new_backing_bs contains a pointer to the new
4544 * backing BlockDriverState (or NULL).
4546 * Return 0 on success, otherwise return < 0 and set @errp.
4548 static int bdrv_reopen_parse_file_or_backing(BDRVReopenState
*reopen_state
,
4549 bool is_backing
, Transaction
*tran
,
4552 BlockDriverState
*bs
= reopen_state
->bs
;
4553 BlockDriverState
*new_child_bs
;
4554 BlockDriverState
*old_child_bs
= is_backing
? child_bs(bs
->backing
) :
4556 const char *child_name
= is_backing
? "backing" : "file";
4560 GLOBAL_STATE_CODE();
4562 value
= qdict_get(reopen_state
->options
, child_name
);
4563 if (value
== NULL
) {
4567 switch (qobject_type(value
)) {
4569 assert(is_backing
); /* The 'file' option does not allow a null value */
4570 new_child_bs
= NULL
;
4573 str
= qstring_get_str(qobject_to(QString
, value
));
4574 new_child_bs
= bdrv_lookup_bs(NULL
, str
, errp
);
4575 if (new_child_bs
== NULL
) {
4577 } else if (bdrv_recurse_has_child(new_child_bs
, bs
)) {
4578 error_setg(errp
, "Making '%s' a %s child of '%s' would create a "
4579 "cycle", str
, child_name
, bs
->node_name
);
4585 * The options QDict has been flattened, so 'backing' and 'file'
4586 * do not allow any other data type here.
4588 g_assert_not_reached();
4591 if (old_child_bs
== new_child_bs
) {
4596 if (bdrv_skip_implicit_filters(old_child_bs
) == new_child_bs
) {
4600 if (old_child_bs
->implicit
) {
4601 error_setg(errp
, "Cannot replace implicit %s child of %s",
4602 child_name
, bs
->node_name
);
4607 if (bs
->drv
->is_filter
&& !old_child_bs
) {
4609 * Filters always have a file or a backing child, so we are trying to
4610 * change wrong child
4612 error_setg(errp
, "'%s' is a %s filter node that does not support a "
4613 "%s child", bs
->node_name
, bs
->drv
->format_name
, child_name
);
4618 reopen_state
->old_backing_bs
= old_child_bs
;
4620 reopen_state
->old_file_bs
= old_child_bs
;
4623 return bdrv_set_file_or_backing_noperm(bs
, new_child_bs
, is_backing
,
4628 * Prepares a BlockDriverState for reopen. All changes are staged in the
4629 * 'opaque' field of the BDRVReopenState, which is used and allocated by
4630 * the block driver layer .bdrv_reopen_prepare()
4632 * bs is the BlockDriverState to reopen
4633 * flags are the new open flags
4634 * queue is the reopen queue
4636 * Returns 0 on success, non-zero on error. On error errp will be set
4639 * On failure, bdrv_reopen_abort() will be called to clean up any data.
4640 * It is the responsibility of the caller to then call the abort() or
4641 * commit() for any other BDS that have been left in a prepare() state
4644 static int bdrv_reopen_prepare(BDRVReopenState
*reopen_state
,
4645 BlockReopenQueue
*queue
,
4646 Transaction
*change_child_tran
, Error
**errp
)
4650 Error
*local_err
= NULL
;
4653 QDict
*orig_reopen_opts
;
4654 char *discard
= NULL
;
4656 bool drv_prepared
= false;
4658 assert(reopen_state
!= NULL
);
4659 assert(reopen_state
->bs
->drv
!= NULL
);
4660 GLOBAL_STATE_CODE();
4661 drv
= reopen_state
->bs
->drv
;
4663 /* This function and each driver's bdrv_reopen_prepare() remove
4664 * entries from reopen_state->options as they are processed, so
4665 * we need to make a copy of the original QDict. */
4666 orig_reopen_opts
= qdict_clone_shallow(reopen_state
->options
);
4668 /* Process generic block layer options */
4669 opts
= qemu_opts_create(&bdrv_runtime_opts
, NULL
, 0, &error_abort
);
4670 if (!qemu_opts_absorb_qdict(opts
, reopen_state
->options
, errp
)) {
4675 /* This was already called in bdrv_reopen_queue_child() so the flags
4676 * are up-to-date. This time we simply want to remove the options from
4677 * QemuOpts in order to indicate that they have been processed. */
4678 old_flags
= reopen_state
->flags
;
4679 update_flags_from_options(&reopen_state
->flags
, opts
);
4680 assert(old_flags
== reopen_state
->flags
);
4682 discard
= qemu_opt_get_del(opts
, BDRV_OPT_DISCARD
);
4683 if (discard
!= NULL
) {
4684 if (bdrv_parse_discard_flags(discard
, &reopen_state
->flags
) != 0) {
4685 error_setg(errp
, "Invalid discard option");
4691 reopen_state
->detect_zeroes
=
4692 bdrv_parse_detect_zeroes(opts
, reopen_state
->flags
, &local_err
);
4694 error_propagate(errp
, local_err
);
4699 /* All other options (including node-name and driver) must be unchanged.
4700 * Put them back into the QDict, so that they are checked at the end
4701 * of this function. */
4702 qemu_opts_to_qdict(opts
, reopen_state
->options
);
4704 /* If we are to stay read-only, do not allow permission change
4705 * to r/w. Attempting to set to r/w may fail if either BDRV_O_ALLOW_RDWR is
4706 * not set, or if the BDS still has copy_on_read enabled */
4707 read_only
= !(reopen_state
->flags
& BDRV_O_RDWR
);
4708 ret
= bdrv_can_set_read_only(reopen_state
->bs
, read_only
, true, &local_err
);
4710 error_propagate(errp
, local_err
);
4714 if (drv
->bdrv_reopen_prepare
) {
4716 * If a driver-specific option is missing, it means that we
4717 * should reset it to its default value.
4718 * But not all options allow that, so we need to check it first.
4720 ret
= bdrv_reset_options_allowed(reopen_state
->bs
,
4721 reopen_state
->options
, errp
);
4726 ret
= drv
->bdrv_reopen_prepare(reopen_state
, queue
, &local_err
);
4728 if (local_err
!= NULL
) {
4729 error_propagate(errp
, local_err
);
4731 bdrv_refresh_filename(reopen_state
->bs
);
4732 error_setg(errp
, "failed while preparing to reopen image '%s'",
4733 reopen_state
->bs
->filename
);
4738 /* It is currently mandatory to have a bdrv_reopen_prepare()
4739 * handler for each supported drv. */
4740 error_setg(errp
, "Block format '%s' used by node '%s' "
4741 "does not support reopening files", drv
->format_name
,
4742 bdrv_get_device_or_node_name(reopen_state
->bs
));
4747 drv_prepared
= true;
4750 * We must provide the 'backing' option if the BDS has a backing
4751 * file or if the image file has a backing file name as part of
4752 * its metadata. Otherwise the 'backing' option can be omitted.
4754 if (drv
->supports_backing
&& reopen_state
->backing_missing
&&
4755 (reopen_state
->bs
->backing
|| reopen_state
->bs
->backing_file
[0])) {
4756 error_setg(errp
, "backing is missing for '%s'",
4757 reopen_state
->bs
->node_name
);
4763 * Allow changing the 'backing' option. The new value can be
4764 * either a reference to an existing node (using its node name)
4765 * or NULL to simply detach the current backing file.
4767 ret
= bdrv_reopen_parse_file_or_backing(reopen_state
, true,
4768 change_child_tran
, errp
);
4772 qdict_del(reopen_state
->options
, "backing");
4774 /* Allow changing the 'file' option. In this case NULL is not allowed */
4775 ret
= bdrv_reopen_parse_file_or_backing(reopen_state
, false,
4776 change_child_tran
, errp
);
4780 qdict_del(reopen_state
->options
, "file");
4782 /* Options that are not handled are only okay if they are unchanged
4783 * compared to the old state. It is expected that some options are only
4784 * used for the initial open, but not reopen (e.g. filename) */
4785 if (qdict_size(reopen_state
->options
)) {
4786 const QDictEntry
*entry
= qdict_first(reopen_state
->options
);
4789 QObject
*new = entry
->value
;
4790 QObject
*old
= qdict_get(reopen_state
->bs
->options
, entry
->key
);
4792 /* Allow child references (child_name=node_name) as long as they
4793 * point to the current child (i.e. everything stays the same). */
4794 if (qobject_type(new) == QTYPE_QSTRING
) {
4796 QLIST_FOREACH(child
, &reopen_state
->bs
->children
, next
) {
4797 if (!strcmp(child
->name
, entry
->key
)) {
4803 if (!strcmp(child
->bs
->node_name
,
4804 qstring_get_str(qobject_to(QString
, new)))) {
4805 continue; /* Found child with this name, skip option */
4811 * TODO: When using -drive to specify blockdev options, all values
4812 * will be strings; however, when using -blockdev, blockdev-add or
4813 * filenames using the json:{} pseudo-protocol, they will be
4815 * In contrast, reopening options are (currently) always strings
4816 * (because you can only specify them through qemu-io; all other
4817 * callers do not specify any options).
4818 * Therefore, when using anything other than -drive to create a BDS,
4819 * this cannot detect non-string options as unchanged, because
4820 * qobject_is_equal() always returns false for objects of different
4821 * type. In the future, this should be remedied by correctly typing
4822 * all options. For now, this is not too big of an issue because
4823 * the user can simply omit options which cannot be changed anyway,
4824 * so they will stay unchanged.
4826 if (!qobject_is_equal(new, old
)) {
4827 error_setg(errp
, "Cannot change the option '%s'", entry
->key
);
4831 } while ((entry
= qdict_next(reopen_state
->options
, entry
)));
4836 /* Restore the original reopen_state->options QDict */
4837 qobject_unref(reopen_state
->options
);
4838 reopen_state
->options
= qobject_ref(orig_reopen_opts
);
4841 if (ret
< 0 && drv_prepared
) {
4842 /* drv->bdrv_reopen_prepare() has succeeded, so we need to
4843 * call drv->bdrv_reopen_abort() before signaling an error
4844 * (bdrv_reopen_multiple() will not call bdrv_reopen_abort()
4845 * when the respective bdrv_reopen_prepare() has failed) */
4846 if (drv
->bdrv_reopen_abort
) {
4847 drv
->bdrv_reopen_abort(reopen_state
);
4850 qemu_opts_del(opts
);
4851 qobject_unref(orig_reopen_opts
);
4857 * Takes the staged changes for the reopen from bdrv_reopen_prepare(), and
4858 * makes them final by swapping the staging BlockDriverState contents into
4859 * the active BlockDriverState contents.
4861 static void bdrv_reopen_commit(BDRVReopenState
*reopen_state
)
4864 BlockDriverState
*bs
;
4867 assert(reopen_state
!= NULL
);
4868 bs
= reopen_state
->bs
;
4870 assert(drv
!= NULL
);
4871 GLOBAL_STATE_CODE();
4873 /* If there are any driver level actions to take */
4874 if (drv
->bdrv_reopen_commit
) {
4875 drv
->bdrv_reopen_commit(reopen_state
);
4878 /* set BDS specific flags now */
4879 qobject_unref(bs
->explicit_options
);
4880 qobject_unref(bs
->options
);
4881 qobject_ref(reopen_state
->explicit_options
);
4882 qobject_ref(reopen_state
->options
);
4884 bs
->explicit_options
= reopen_state
->explicit_options
;
4885 bs
->options
= reopen_state
->options
;
4886 bs
->open_flags
= reopen_state
->flags
;
4887 bs
->detect_zeroes
= reopen_state
->detect_zeroes
;
4889 /* Remove child references from bs->options and bs->explicit_options.
4890 * Child options were already removed in bdrv_reopen_queue_child() */
4891 QLIST_FOREACH(child
, &bs
->children
, next
) {
4892 qdict_del(bs
->explicit_options
, child
->name
);
4893 qdict_del(bs
->options
, child
->name
);
4895 /* backing is probably removed, so it's not handled by previous loop */
4896 qdict_del(bs
->explicit_options
, "backing");
4897 qdict_del(bs
->options
, "backing");
4899 bdrv_refresh_limits(bs
, NULL
, NULL
);
4903 * Abort the reopen, and delete and free the staged changes in
4906 static void bdrv_reopen_abort(BDRVReopenState
*reopen_state
)
4910 assert(reopen_state
!= NULL
);
4911 drv
= reopen_state
->bs
->drv
;
4912 assert(drv
!= NULL
);
4913 GLOBAL_STATE_CODE();
4915 if (drv
->bdrv_reopen_abort
) {
4916 drv
->bdrv_reopen_abort(reopen_state
);
4921 static void bdrv_close(BlockDriverState
*bs
)
4923 BdrvAioNotifier
*ban
, *ban_next
;
4924 BdrvChild
*child
, *next
;
4926 GLOBAL_STATE_CODE();
4927 assert(!bs
->refcnt
);
4929 bdrv_drained_begin(bs
); /* complete I/O */
4931 bdrv_drain(bs
); /* in case flush left pending I/O */
4934 if (bs
->drv
->bdrv_close
) {
4935 /* Must unfreeze all children, so bdrv_unref_child() works */
4936 bs
->drv
->bdrv_close(bs
);
4941 QLIST_FOREACH_SAFE(child
, &bs
->children
, next
, next
) {
4942 bdrv_unref_child(bs
, child
);
4949 qatomic_set(&bs
->copy_on_read
, 0);
4950 bs
->backing_file
[0] = '\0';
4951 bs
->backing_format
[0] = '\0';
4952 bs
->total_sectors
= 0;
4953 bs
->encrypted
= false;
4955 qobject_unref(bs
->options
);
4956 qobject_unref(bs
->explicit_options
);
4958 bs
->explicit_options
= NULL
;
4959 qobject_unref(bs
->full_open_options
);
4960 bs
->full_open_options
= NULL
;
4961 g_free(bs
->block_status_cache
);
4962 bs
->block_status_cache
= NULL
;
4964 bdrv_release_named_dirty_bitmaps(bs
);
4965 assert(QLIST_EMPTY(&bs
->dirty_bitmaps
));
4967 QLIST_FOREACH_SAFE(ban
, &bs
->aio_notifiers
, list
, ban_next
) {
4970 QLIST_INIT(&bs
->aio_notifiers
);
4971 bdrv_drained_end(bs
);
4974 * If we're still inside some bdrv_drain_all_begin()/end() sections, end
4975 * them now since this BDS won't exist anymore when bdrv_drain_all_end()
4978 if (bs
->quiesce_counter
) {
4979 bdrv_drain_all_end_quiesce(bs
);
4983 void bdrv_close_all(void)
4985 GLOBAL_STATE_CODE();
4986 assert(job_next(NULL
) == NULL
);
4988 /* Drop references from requests still in flight, such as canceled block
4989 * jobs whose AIO context has not been polled yet */
4992 blk_remove_all_bs();
4993 blockdev_close_all_bdrv_states();
4995 assert(QTAILQ_EMPTY(&all_bdrv_states
));
4998 static bool should_update_child(BdrvChild
*c
, BlockDriverState
*to
)
5004 if (c
->klass
->stay_at_node
) {
5008 /* If the child @c belongs to the BDS @to, replacing the current
5009 * c->bs by @to would mean to create a loop.
5011 * Such a case occurs when appending a BDS to a backing chain.
5012 * For instance, imagine the following chain:
5014 * guest device -> node A -> further backing chain...
5016 * Now we create a new BDS B which we want to put on top of this
5017 * chain, so we first attach A as its backing node:
5022 * guest device -> node A -> further backing chain...
5024 * Finally we want to replace A by B. When doing that, we want to
5025 * replace all pointers to A by pointers to B -- except for the
5026 * pointer from B because (1) that would create a loop, and (2)
5027 * that pointer should simply stay intact:
5029 * guest device -> node B
5032 * node A -> further backing chain...
5034 * In general, when replacing a node A (c->bs) by a node B (@to),
5035 * if A is a child of B, that means we cannot replace A by B there
5036 * because that would create a loop. Silently detaching A from B
5037 * is also not really an option. So overall just leaving A in
5038 * place there is the most sensible choice.
5040 * We would also create a loop in any cases where @c is only
5041 * indirectly referenced by @to. Prevent this by returning false
5042 * if @c is found (by breadth-first search) anywhere in the whole
5047 found
= g_hash_table_new(NULL
, NULL
);
5048 g_hash_table_add(found
, to
);
5049 queue
= g_queue_new();
5050 g_queue_push_tail(queue
, to
);
5052 while (!g_queue_is_empty(queue
)) {
5053 BlockDriverState
*v
= g_queue_pop_head(queue
);
5056 QLIST_FOREACH(c2
, &v
->children
, next
) {
5062 if (g_hash_table_contains(found
, c2
->bs
)) {
5066 g_queue_push_tail(queue
, c2
->bs
);
5067 g_hash_table_add(found
, c2
->bs
);
5071 g_queue_free(queue
);
5072 g_hash_table_destroy(found
);
5077 typedef struct BdrvRemoveFilterOrCowChild
{
5079 BlockDriverState
*bs
;
5081 } BdrvRemoveFilterOrCowChild
;
5083 static void bdrv_remove_filter_or_cow_child_abort(void *opaque
)
5085 BdrvRemoveFilterOrCowChild
*s
= opaque
;
5086 BlockDriverState
*parent_bs
= s
->child
->opaque
;
5088 if (s
->is_backing
) {
5089 parent_bs
->backing
= s
->child
;
5091 parent_bs
->file
= s
->child
;
5095 * We don't have to restore child->bs here to undo bdrv_replace_child_tran()
5096 * because that function is transactionable and it registered own completion
5097 * entries in @tran, so .abort() for bdrv_replace_child_safe() will be
5098 * called automatically.
5102 static void bdrv_remove_filter_or_cow_child_commit(void *opaque
)
5104 BdrvRemoveFilterOrCowChild
*s
= opaque
;
5105 GLOBAL_STATE_CODE();
5106 bdrv_child_free(s
->child
);
5109 static void bdrv_remove_filter_or_cow_child_clean(void *opaque
)
5111 BdrvRemoveFilterOrCowChild
*s
= opaque
;
5113 /* Drop the bs reference after the transaction is done */
5118 static TransactionActionDrv bdrv_remove_filter_or_cow_child_drv
= {
5119 .abort
= bdrv_remove_filter_or_cow_child_abort
,
5120 .commit
= bdrv_remove_filter_or_cow_child_commit
,
5121 .clean
= bdrv_remove_filter_or_cow_child_clean
,
5125 * A function to remove backing or file child of @bs.
5126 * Function doesn't update permissions, caller is responsible for this.
5128 static void bdrv_remove_file_or_backing_child(BlockDriverState
*bs
,
5133 BdrvRemoveFilterOrCowChild
*s
;
5140 * Keep a reference to @bs so @childp will stay valid throughout the
5141 * transaction (required by bdrv_replace_child_tran())
5144 if (child
== bs
->backing
) {
5145 childp
= &bs
->backing
;
5146 } else if (child
== bs
->file
) {
5149 g_assert_not_reached();
5154 * Pass free_empty_child=false, we will free the child in
5155 * bdrv_remove_filter_or_cow_child_commit()
5157 bdrv_replace_child_tran(childp
, NULL
, tran
, false);
5160 s
= g_new(BdrvRemoveFilterOrCowChild
, 1);
5161 *s
= (BdrvRemoveFilterOrCowChild
) {
5164 .is_backing
= (childp
== &bs
->backing
),
5166 tran_add(tran
, &bdrv_remove_filter_or_cow_child_drv
, s
);
5170 * A function to remove backing-chain child of @bs if exists: cow child for
5171 * format nodes (always .backing) and filter child for filters (may be .file or
5174 static void bdrv_remove_filter_or_cow_child(BlockDriverState
*bs
,
5177 bdrv_remove_file_or_backing_child(bs
, bdrv_filter_or_cow_child(bs
), tran
);
5180 static int bdrv_replace_node_noperm(BlockDriverState
*from
,
5181 BlockDriverState
*to
,
5182 bool auto_skip
, Transaction
*tran
,
5185 BdrvChild
*c
, *next
;
5188 GLOBAL_STATE_CODE();
5190 QLIST_FOREACH_SAFE(c
, &from
->parents
, next_parent
, next
) {
5191 assert(c
->bs
== from
);
5192 if (!should_update_child(c
, to
)) {
5196 error_setg(errp
, "Should not change '%s' link to '%s'",
5197 c
->name
, from
->node_name
);
5201 error_setg(errp
, "Cannot change '%s' link to '%s'",
5202 c
->name
, from
->node_name
);
5207 * Passing a pointer to the local variable @c is fine here, because
5208 * @to is not NULL, and so &c will not be attached to the transaction.
5210 bdrv_replace_child_tran(&c
, to
, tran
, true);
5217 * With auto_skip=true bdrv_replace_node_common skips updating from parents
5218 * if it creates a parent-child relation loop or if parent is block-job.
5220 * With auto_skip=false the error is returned if from has a parent which should
5223 * With @detach_subchain=true @to must be in a backing chain of @from. In this
5224 * case backing link of the cow-parent of @to is removed.
5226 * @to must not be NULL.
5228 static int bdrv_replace_node_common(BlockDriverState
*from
,
5229 BlockDriverState
*to
,
5230 bool auto_skip
, bool detach_subchain
,
5233 Transaction
*tran
= tran_new();
5234 g_autoptr(GHashTable
) found
= NULL
;
5235 g_autoptr(GSList
) refresh_list
= NULL
;
5236 BlockDriverState
*to_cow_parent
= NULL
;
5239 GLOBAL_STATE_CODE();
5242 if (detach_subchain
) {
5243 assert(bdrv_chain_contains(from
, to
));
5245 for (to_cow_parent
= from
;
5246 bdrv_filter_or_cow_bs(to_cow_parent
) != to
;
5247 to_cow_parent
= bdrv_filter_or_cow_bs(to_cow_parent
))
5253 /* Make sure that @from doesn't go away until we have successfully attached
5254 * all of its parents to @to. */
5257 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
5258 assert(bdrv_get_aio_context(from
) == bdrv_get_aio_context(to
));
5259 bdrv_drained_begin(from
);
5262 * Do the replacement without permission update.
5263 * Replacement may influence the permissions, we should calculate new
5264 * permissions based on new graph. If we fail, we'll roll-back the
5267 ret
= bdrv_replace_node_noperm(from
, to
, auto_skip
, tran
, errp
);
5272 if (detach_subchain
) {
5273 bdrv_remove_filter_or_cow_child(to_cow_parent
, tran
);
5276 found
= g_hash_table_new(NULL
, NULL
);
5278 refresh_list
= bdrv_topological_dfs(refresh_list
, found
, to
);
5279 refresh_list
= bdrv_topological_dfs(refresh_list
, found
, from
);
5281 ret
= bdrv_list_refresh_perms(refresh_list
, NULL
, tran
, errp
);
5289 tran_finalize(tran
, ret
);
5291 bdrv_drained_end(from
);
5298 * Replace node @from by @to (where neither may be NULL).
5300 int bdrv_replace_node(BlockDriverState
*from
, BlockDriverState
*to
,
5303 GLOBAL_STATE_CODE();
5305 return bdrv_replace_node_common(from
, to
, true, false, errp
);
5308 int bdrv_drop_filter(BlockDriverState
*bs
, Error
**errp
)
5310 GLOBAL_STATE_CODE();
5312 return bdrv_replace_node_common(bs
, bdrv_filter_or_cow_bs(bs
), true, true,
5317 * Add new bs contents at the top of an image chain while the chain is
5318 * live, while keeping required fields on the top layer.
5320 * This will modify the BlockDriverState fields, and swap contents
5321 * between bs_new and bs_top. Both bs_new and bs_top are modified.
5323 * bs_new must not be attached to a BlockBackend and must not have backing
5326 * This function does not create any image files.
5328 int bdrv_append(BlockDriverState
*bs_new
, BlockDriverState
*bs_top
,
5332 Transaction
*tran
= tran_new();
5334 GLOBAL_STATE_CODE();
5336 assert(!bs_new
->backing
);
5338 ret
= bdrv_attach_child_noperm(bs_new
, bs_top
, "backing",
5339 &child_of_bds
, bdrv_backing_role(bs_new
),
5340 &bs_new
->backing
, tran
, errp
);
5345 ret
= bdrv_replace_node_noperm(bs_top
, bs_new
, true, tran
, errp
);
5350 ret
= bdrv_refresh_perms(bs_new
, errp
);
5352 tran_finalize(tran
, ret
);
5354 bdrv_refresh_limits(bs_top
, NULL
, NULL
);
5359 /* Not for empty child */
5360 int bdrv_replace_child_bs(BdrvChild
*child
, BlockDriverState
*new_bs
,
5364 Transaction
*tran
= tran_new();
5365 g_autoptr(GHashTable
) found
= NULL
;
5366 g_autoptr(GSList
) refresh_list
= NULL
;
5367 BlockDriverState
*old_bs
= child
->bs
;
5369 GLOBAL_STATE_CODE();
5372 bdrv_drained_begin(old_bs
);
5373 bdrv_drained_begin(new_bs
);
5375 bdrv_replace_child_tran(&child
, new_bs
, tran
, true);
5376 /* @new_bs must have been non-NULL, so @child must not have been freed */
5377 assert(child
!= NULL
);
5379 found
= g_hash_table_new(NULL
, NULL
);
5380 refresh_list
= bdrv_topological_dfs(refresh_list
, found
, old_bs
);
5381 refresh_list
= bdrv_topological_dfs(refresh_list
, found
, new_bs
);
5383 ret
= bdrv_list_refresh_perms(refresh_list
, NULL
, tran
, errp
);
5385 tran_finalize(tran
, ret
);
5387 bdrv_drained_end(old_bs
);
5388 bdrv_drained_end(new_bs
);
5394 static void bdrv_delete(BlockDriverState
*bs
)
5396 assert(bdrv_op_blocker_is_empty(bs
));
5397 assert(!bs
->refcnt
);
5398 GLOBAL_STATE_CODE();
5400 /* remove from list, if necessary */
5401 if (bs
->node_name
[0] != '\0') {
5402 QTAILQ_REMOVE(&graph_bdrv_states
, bs
, node_list
);
5404 QTAILQ_REMOVE(&all_bdrv_states
, bs
, bs_list
);
5413 * Replace @bs by newly created block node.
5415 * @options is a QDict of options to pass to the block drivers, or NULL for an
5416 * empty set of options. The reference to the QDict belongs to the block layer
5417 * after the call (even on failure), so if the caller intends to reuse the
5418 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
5420 BlockDriverState
*bdrv_insert_node(BlockDriverState
*bs
, QDict
*options
,
5421 int flags
, Error
**errp
)
5425 BlockDriverState
*new_node_bs
= NULL
;
5426 const char *drvname
, *node_name
;
5429 drvname
= qdict_get_try_str(options
, "driver");
5431 error_setg(errp
, "driver is not specified");
5435 drv
= bdrv_find_format(drvname
);
5437 error_setg(errp
, "Unknown driver: '%s'", drvname
);
5441 node_name
= qdict_get_try_str(options
, "node-name");
5443 GLOBAL_STATE_CODE();
5445 new_node_bs
= bdrv_new_open_driver_opts(drv
, node_name
, options
, flags
,
5447 options
= NULL
; /* bdrv_new_open_driver() eats options */
5449 error_prepend(errp
, "Could not create node: ");
5453 bdrv_drained_begin(bs
);
5454 ret
= bdrv_replace_node(bs
, new_node_bs
, errp
);
5455 bdrv_drained_end(bs
);
5458 error_prepend(errp
, "Could not replace node: ");
5465 qobject_unref(options
);
5466 bdrv_unref(new_node_bs
);
5471 * Run consistency checks on an image
5473 * Returns 0 if the check could be completed (it doesn't mean that the image is
5474 * free of errors) or -errno when an internal error occurred. The results of the
5475 * check are stored in res.
5477 int coroutine_fn
bdrv_co_check(BlockDriverState
*bs
,
5478 BdrvCheckResult
*res
, BdrvCheckMode fix
)
5481 if (bs
->drv
== NULL
) {
5484 if (bs
->drv
->bdrv_co_check
== NULL
) {
5488 memset(res
, 0, sizeof(*res
));
5489 return bs
->drv
->bdrv_co_check(bs
, res
, fix
);
5495 * -EINVAL - backing format specified, but no file
5496 * -ENOSPC - can't update the backing file because no space is left in the
5498 * -ENOTSUP - format driver doesn't support changing the backing file
5500 int bdrv_change_backing_file(BlockDriverState
*bs
, const char *backing_file
,
5501 const char *backing_fmt
, bool require
)
5503 BlockDriver
*drv
= bs
->drv
;
5506 GLOBAL_STATE_CODE();
5512 /* Backing file format doesn't make sense without a backing file */
5513 if (backing_fmt
&& !backing_file
) {
5517 if (require
&& backing_file
&& !backing_fmt
) {
5521 if (drv
->bdrv_change_backing_file
!= NULL
) {
5522 ret
= drv
->bdrv_change_backing_file(bs
, backing_file
, backing_fmt
);
5528 pstrcpy(bs
->backing_file
, sizeof(bs
->backing_file
), backing_file
?: "");
5529 pstrcpy(bs
->backing_format
, sizeof(bs
->backing_format
), backing_fmt
?: "");
5530 pstrcpy(bs
->auto_backing_file
, sizeof(bs
->auto_backing_file
),
5531 backing_file
?: "");
5537 * Finds the first non-filter node above bs in the chain between
5538 * active and bs. The returned node is either an immediate parent of
5539 * bs, or there are only filter nodes between the two.
5541 * Returns NULL if bs is not found in active's image chain,
5542 * or if active == bs.
5544 * Returns the bottommost base image if bs == NULL.
5546 BlockDriverState
*bdrv_find_overlay(BlockDriverState
*active
,
5547 BlockDriverState
*bs
)
5550 GLOBAL_STATE_CODE();
5552 bs
= bdrv_skip_filters(bs
);
5553 active
= bdrv_skip_filters(active
);
5556 BlockDriverState
*next
= bdrv_backing_chain_next(active
);
5566 /* Given a BDS, searches for the base layer. */
5567 BlockDriverState
*bdrv_find_base(BlockDriverState
*bs
)
5569 GLOBAL_STATE_CODE();
5571 return bdrv_find_overlay(bs
, NULL
);
5575 * Return true if at least one of the COW (backing) and filter links
5576 * between @bs and @base is frozen. @errp is set if that's the case.
5577 * @base must be reachable from @bs, or NULL.
5579 bool bdrv_is_backing_chain_frozen(BlockDriverState
*bs
, BlockDriverState
*base
,
5582 BlockDriverState
*i
;
5585 GLOBAL_STATE_CODE();
5587 for (i
= bs
; i
!= base
; i
= child_bs(child
)) {
5588 child
= bdrv_filter_or_cow_child(i
);
5590 if (child
&& child
->frozen
) {
5591 error_setg(errp
, "Cannot change '%s' link from '%s' to '%s'",
5592 child
->name
, i
->node_name
, child
->bs
->node_name
);
5601 * Freeze all COW (backing) and filter links between @bs and @base.
5602 * If any of the links is already frozen the operation is aborted and
5603 * none of the links are modified.
5604 * @base must be reachable from @bs, or NULL.
5605 * Returns 0 on success. On failure returns < 0 and sets @errp.
5607 int bdrv_freeze_backing_chain(BlockDriverState
*bs
, BlockDriverState
*base
,
5610 BlockDriverState
*i
;
5613 GLOBAL_STATE_CODE();
5615 if (bdrv_is_backing_chain_frozen(bs
, base
, errp
)) {
5619 for (i
= bs
; i
!= base
; i
= child_bs(child
)) {
5620 child
= bdrv_filter_or_cow_child(i
);
5621 if (child
&& child
->bs
->never_freeze
) {
5622 error_setg(errp
, "Cannot freeze '%s' link to '%s'",
5623 child
->name
, child
->bs
->node_name
);
5628 for (i
= bs
; i
!= base
; i
= child_bs(child
)) {
5629 child
= bdrv_filter_or_cow_child(i
);
5631 child
->frozen
= true;
5639 * Unfreeze all COW (backing) and filter links between @bs and @base.
5640 * The caller must ensure that all links are frozen before using this
5642 * @base must be reachable from @bs, or NULL.
5644 void bdrv_unfreeze_backing_chain(BlockDriverState
*bs
, BlockDriverState
*base
)
5646 BlockDriverState
*i
;
5649 GLOBAL_STATE_CODE();
5651 for (i
= bs
; i
!= base
; i
= child_bs(child
)) {
5652 child
= bdrv_filter_or_cow_child(i
);
5654 assert(child
->frozen
);
5655 child
->frozen
= false;
5661 * Drops images above 'base' up to and including 'top', and sets the image
5662 * above 'top' to have base as its backing file.
5664 * Requires that the overlay to 'top' is opened r/w, so that the backing file
5665 * information in 'bs' can be properly updated.
5667 * E.g., this will convert the following chain:
5668 * bottom <- base <- intermediate <- top <- active
5672 * bottom <- base <- active
5674 * It is allowed for bottom==base, in which case it converts:
5676 * base <- intermediate <- top <- active
5682 * If backing_file_str is non-NULL, it will be used when modifying top's
5683 * overlay image metadata.
5686 * if active == top, that is considered an error
5689 int bdrv_drop_intermediate(BlockDriverState
*top
, BlockDriverState
*base
,
5690 const char *backing_file_str
)
5692 BlockDriverState
*explicit_top
= top
;
5693 bool update_inherits_from
;
5695 Error
*local_err
= NULL
;
5697 g_autoptr(GSList
) updated_children
= NULL
;
5700 GLOBAL_STATE_CODE();
5703 bdrv_subtree_drained_begin(top
);
5705 if (!top
->drv
|| !base
->drv
) {
5709 /* Make sure that base is in the backing chain of top */
5710 if (!bdrv_chain_contains(top
, base
)) {
5714 /* If 'base' recursively inherits from 'top' then we should set
5715 * base->inherits_from to top->inherits_from after 'top' and all
5716 * other intermediate nodes have been dropped.
5717 * If 'top' is an implicit node (e.g. "commit_top") we should skip
5718 * it because no one inherits from it. We use explicit_top for that. */
5719 explicit_top
= bdrv_skip_implicit_filters(explicit_top
);
5720 update_inherits_from
= bdrv_inherits_from_recursive(base
, explicit_top
);
5722 /* success - we can delete the intermediate states, and link top->base */
5723 if (!backing_file_str
) {
5724 bdrv_refresh_filename(base
);
5725 backing_file_str
= base
->filename
;
5728 QLIST_FOREACH(c
, &top
->parents
, next_parent
) {
5729 updated_children
= g_slist_prepend(updated_children
, c
);
5733 * It seems correct to pass detach_subchain=true here, but it triggers
5734 * one more yet not fixed bug, when due to nested aio_poll loop we switch to
5735 * another drained section, which modify the graph (for example, removing
5736 * the child, which we keep in updated_children list). So, it's a TODO.
5738 * Note, bug triggered if pass detach_subchain=true here and run
5739 * test-bdrv-drain. test_drop_intermediate_poll() test-case will crash.
5742 bdrv_replace_node_common(top
, base
, false, false, &local_err
);
5744 error_report_err(local_err
);
5748 for (p
= updated_children
; p
; p
= p
->next
) {
5751 if (c
->klass
->update_filename
) {
5752 ret
= c
->klass
->update_filename(c
, base
, backing_file_str
,
5756 * TODO: Actually, we want to rollback all previous iterations
5757 * of this loop, and (which is almost impossible) previous
5758 * bdrv_replace_node()...
5760 * Note, that c->klass->update_filename may lead to permission
5761 * update, so it's a bad idea to call it inside permission
5762 * update transaction of bdrv_replace_node.
5764 error_report_err(local_err
);
5770 if (update_inherits_from
) {
5771 base
->inherits_from
= explicit_top
->inherits_from
;
5776 bdrv_subtree_drained_end(top
);
5782 * Implementation of BlockDriver.bdrv_get_allocated_file_size() that
5783 * sums the size of all data-bearing children. (This excludes backing
5786 static int64_t bdrv_sum_allocated_file_size(BlockDriverState
*bs
)
5789 int64_t child_size
, sum
= 0;
5791 QLIST_FOREACH(child
, &bs
->children
, next
) {
5792 if (child
->role
& (BDRV_CHILD_DATA
| BDRV_CHILD_METADATA
|
5793 BDRV_CHILD_FILTERED
))
5795 child_size
= bdrv_get_allocated_file_size(child
->bs
);
5796 if (child_size
< 0) {
5807 * Length of a allocated file in bytes. Sparse files are counted by actual
5808 * allocated space. Return < 0 if error or unknown.
5810 int64_t bdrv_get_allocated_file_size(BlockDriverState
*bs
)
5812 BlockDriver
*drv
= bs
->drv
;
5818 if (drv
->bdrv_get_allocated_file_size
) {
5819 return drv
->bdrv_get_allocated_file_size(bs
);
5822 if (drv
->bdrv_file_open
) {
5824 * Protocol drivers default to -ENOTSUP (most of their data is
5825 * not stored in any of their children (if they even have any),
5826 * so there is no generic way to figure it out).
5829 } else if (drv
->is_filter
) {
5830 /* Filter drivers default to the size of their filtered child */
5831 return bdrv_get_allocated_file_size(bdrv_filter_bs(bs
));
5833 /* Other drivers default to summing their children's sizes */
5834 return bdrv_sum_allocated_file_size(bs
);
5840 * @drv: Format driver
5841 * @opts: Creation options for new image
5842 * @in_bs: Existing image containing data for new image (may be NULL)
5843 * @errp: Error object
5844 * Returns: A #BlockMeasureInfo (free using qapi_free_BlockMeasureInfo())
5847 * Calculate file size required to create a new image.
5849 * If @in_bs is given then space for allocated clusters and zero clusters
5850 * from that image are included in the calculation. If @opts contains a
5851 * backing file that is shared by @in_bs then backing clusters may be omitted
5852 * from the calculation.
5854 * If @in_bs is NULL then the calculation includes no allocated clusters
5855 * unless a preallocation option is given in @opts.
5857 * Note that @in_bs may use a different BlockDriver from @drv.
5859 * If an error occurs the @errp pointer is set.
5861 BlockMeasureInfo
*bdrv_measure(BlockDriver
*drv
, QemuOpts
*opts
,
5862 BlockDriverState
*in_bs
, Error
**errp
)
5865 if (!drv
->bdrv_measure
) {
5866 error_setg(errp
, "Block driver '%s' does not support size measurement",
5871 return drv
->bdrv_measure(opts
, in_bs
, errp
);
5875 * Return number of sectors on success, -errno on error.
5877 int64_t bdrv_nb_sectors(BlockDriverState
*bs
)
5879 BlockDriver
*drv
= bs
->drv
;
5885 if (drv
->has_variable_length
) {
5886 int ret
= refresh_total_sectors(bs
, bs
->total_sectors
);
5891 return bs
->total_sectors
;
5895 * Return length in bytes on success, -errno on error.
5896 * The length is always a multiple of BDRV_SECTOR_SIZE.
5898 int64_t bdrv_getlength(BlockDriverState
*bs
)
5900 int64_t ret
= bdrv_nb_sectors(bs
);
5906 if (ret
> INT64_MAX
/ BDRV_SECTOR_SIZE
) {
5909 return ret
* BDRV_SECTOR_SIZE
;
5912 /* return 0 as number of sectors if no device present or error */
5913 void bdrv_get_geometry(BlockDriverState
*bs
, uint64_t *nb_sectors_ptr
)
5915 int64_t nb_sectors
= bdrv_nb_sectors(bs
);
5918 *nb_sectors_ptr
= nb_sectors
< 0 ? 0 : nb_sectors
;
5921 bool bdrv_is_sg(BlockDriverState
*bs
)
5928 * Return whether the given node supports compressed writes.
5930 bool bdrv_supports_compressed_writes(BlockDriverState
*bs
)
5932 BlockDriverState
*filtered
;
5935 if (!bs
->drv
|| !block_driver_can_compress(bs
->drv
)) {
5939 filtered
= bdrv_filter_bs(bs
);
5942 * Filters can only forward compressed writes, so we have to
5945 return bdrv_supports_compressed_writes(filtered
);
5951 const char *bdrv_get_format_name(BlockDriverState
*bs
)
5954 return bs
->drv
? bs
->drv
->format_name
: NULL
;
5957 static int qsort_strcmp(const void *a
, const void *b
)
5959 return strcmp(*(char *const *)a
, *(char *const *)b
);
5962 void bdrv_iterate_format(void (*it
)(void *opaque
, const char *name
),
5963 void *opaque
, bool read_only
)
5968 const char **formats
= NULL
;
5970 GLOBAL_STATE_CODE();
5972 QLIST_FOREACH(drv
, &bdrv_drivers
, list
) {
5973 if (drv
->format_name
) {
5977 if (use_bdrv_whitelist
&& !bdrv_is_whitelisted(drv
, read_only
)) {
5981 while (formats
&& i
&& !found
) {
5982 found
= !strcmp(formats
[--i
], drv
->format_name
);
5986 formats
= g_renew(const char *, formats
, count
+ 1);
5987 formats
[count
++] = drv
->format_name
;
5992 for (i
= 0; i
< (int)ARRAY_SIZE(block_driver_modules
); i
++) {
5993 const char *format_name
= block_driver_modules
[i
].format_name
;
5999 if (use_bdrv_whitelist
&&
6000 !bdrv_format_is_whitelisted(format_name
, read_only
)) {
6004 while (formats
&& j
&& !found
) {
6005 found
= !strcmp(formats
[--j
], format_name
);
6009 formats
= g_renew(const char *, formats
, count
+ 1);
6010 formats
[count
++] = format_name
;
6015 qsort(formats
, count
, sizeof(formats
[0]), qsort_strcmp
);
6017 for (i
= 0; i
< count
; i
++) {
6018 it(opaque
, formats
[i
]);
6024 /* This function is to find a node in the bs graph */
6025 BlockDriverState
*bdrv_find_node(const char *node_name
)
6027 BlockDriverState
*bs
;
6030 GLOBAL_STATE_CODE();
6032 QTAILQ_FOREACH(bs
, &graph_bdrv_states
, node_list
) {
6033 if (!strcmp(node_name
, bs
->node_name
)) {
6040 /* Put this QMP function here so it can access the static graph_bdrv_states. */
6041 BlockDeviceInfoList
*bdrv_named_nodes_list(bool flat
,
6044 BlockDeviceInfoList
*list
;
6045 BlockDriverState
*bs
;
6047 GLOBAL_STATE_CODE();
6050 QTAILQ_FOREACH(bs
, &graph_bdrv_states
, node_list
) {
6051 BlockDeviceInfo
*info
= bdrv_block_device_info(NULL
, bs
, flat
, errp
);
6053 qapi_free_BlockDeviceInfoList(list
);
6056 QAPI_LIST_PREPEND(list
, info
);
6062 typedef struct XDbgBlockGraphConstructor
{
6063 XDbgBlockGraph
*graph
;
6064 GHashTable
*graph_nodes
;
6065 } XDbgBlockGraphConstructor
;
6067 static XDbgBlockGraphConstructor
*xdbg_graph_new(void)
6069 XDbgBlockGraphConstructor
*gr
= g_new(XDbgBlockGraphConstructor
, 1);
6071 gr
->graph
= g_new0(XDbgBlockGraph
, 1);
6072 gr
->graph_nodes
= g_hash_table_new(NULL
, NULL
);
6077 static XDbgBlockGraph
*xdbg_graph_finalize(XDbgBlockGraphConstructor
*gr
)
6079 XDbgBlockGraph
*graph
= gr
->graph
;
6081 g_hash_table_destroy(gr
->graph_nodes
);
6087 static uintptr_t xdbg_graph_node_num(XDbgBlockGraphConstructor
*gr
, void *node
)
6089 uintptr_t ret
= (uintptr_t)g_hash_table_lookup(gr
->graph_nodes
, node
);
6096 * Start counting from 1, not 0, because 0 interferes with not-found (NULL)
6097 * answer of g_hash_table_lookup.
6099 ret
= g_hash_table_size(gr
->graph_nodes
) + 1;
6100 g_hash_table_insert(gr
->graph_nodes
, node
, (void *)ret
);
6105 static void xdbg_graph_add_node(XDbgBlockGraphConstructor
*gr
, void *node
,
6106 XDbgBlockGraphNodeType type
, const char *name
)
6108 XDbgBlockGraphNode
*n
;
6110 n
= g_new0(XDbgBlockGraphNode
, 1);
6112 n
->id
= xdbg_graph_node_num(gr
, node
);
6114 n
->name
= g_strdup(name
);
6116 QAPI_LIST_PREPEND(gr
->graph
->nodes
, n
);
6119 static void xdbg_graph_add_edge(XDbgBlockGraphConstructor
*gr
, void *parent
,
6120 const BdrvChild
*child
)
6122 BlockPermission qapi_perm
;
6123 XDbgBlockGraphEdge
*edge
;
6124 GLOBAL_STATE_CODE();
6126 edge
= g_new0(XDbgBlockGraphEdge
, 1);
6128 edge
->parent
= xdbg_graph_node_num(gr
, parent
);
6129 edge
->child
= xdbg_graph_node_num(gr
, child
->bs
);
6130 edge
->name
= g_strdup(child
->name
);
6132 for (qapi_perm
= 0; qapi_perm
< BLOCK_PERMISSION__MAX
; qapi_perm
++) {
6133 uint64_t flag
= bdrv_qapi_perm_to_blk_perm(qapi_perm
);
6135 if (flag
& child
->perm
) {
6136 QAPI_LIST_PREPEND(edge
->perm
, qapi_perm
);
6138 if (flag
& child
->shared_perm
) {
6139 QAPI_LIST_PREPEND(edge
->shared_perm
, qapi_perm
);
6143 QAPI_LIST_PREPEND(gr
->graph
->edges
, edge
);
6147 XDbgBlockGraph
*bdrv_get_xdbg_block_graph(Error
**errp
)
6151 BlockDriverState
*bs
;
6153 XDbgBlockGraphConstructor
*gr
= xdbg_graph_new();
6155 GLOBAL_STATE_CODE();
6157 for (blk
= blk_all_next(NULL
); blk
; blk
= blk_all_next(blk
)) {
6158 char *allocated_name
= NULL
;
6159 const char *name
= blk_name(blk
);
6162 name
= allocated_name
= blk_get_attached_dev_id(blk
);
6164 xdbg_graph_add_node(gr
, blk
, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_BACKEND
,
6166 g_free(allocated_name
);
6167 if (blk_root(blk
)) {
6168 xdbg_graph_add_edge(gr
, blk
, blk_root(blk
));
6172 WITH_JOB_LOCK_GUARD() {
6173 for (job
= block_job_next_locked(NULL
); job
;
6174 job
= block_job_next_locked(job
)) {
6177 xdbg_graph_add_node(gr
, job
, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_JOB
,
6179 for (el
= job
->nodes
; el
; el
= el
->next
) {
6180 xdbg_graph_add_edge(gr
, job
, (BdrvChild
*)el
->data
);
6185 QTAILQ_FOREACH(bs
, &graph_bdrv_states
, node_list
) {
6186 xdbg_graph_add_node(gr
, bs
, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_DRIVER
,
6188 QLIST_FOREACH(child
, &bs
->children
, next
) {
6189 xdbg_graph_add_edge(gr
, bs
, child
);
6193 return xdbg_graph_finalize(gr
);
6196 BlockDriverState
*bdrv_lookup_bs(const char *device
,
6197 const char *node_name
,
6201 BlockDriverState
*bs
;
6203 GLOBAL_STATE_CODE();
6206 blk
= blk_by_name(device
);
6211 error_setg(errp
, "Device '%s' has no medium", device
);
6219 bs
= bdrv_find_node(node_name
);
6226 error_setg(errp
, "Cannot find device=\'%s\' nor node-name=\'%s\'",
6227 device
? device
: "",
6228 node_name
? node_name
: "");
6232 /* If 'base' is in the same chain as 'top', return true. Otherwise,
6233 * return false. If either argument is NULL, return false. */
6234 bool bdrv_chain_contains(BlockDriverState
*top
, BlockDriverState
*base
)
6237 GLOBAL_STATE_CODE();
6239 while (top
&& top
!= base
) {
6240 top
= bdrv_filter_or_cow_bs(top
);
6246 BlockDriverState
*bdrv_next_node(BlockDriverState
*bs
)
6248 GLOBAL_STATE_CODE();
6250 return QTAILQ_FIRST(&graph_bdrv_states
);
6252 return QTAILQ_NEXT(bs
, node_list
);
6255 BlockDriverState
*bdrv_next_all_states(BlockDriverState
*bs
)
6257 GLOBAL_STATE_CODE();
6259 return QTAILQ_FIRST(&all_bdrv_states
);
6261 return QTAILQ_NEXT(bs
, bs_list
);
6264 const char *bdrv_get_node_name(const BlockDriverState
*bs
)
6267 return bs
->node_name
;
6270 const char *bdrv_get_parent_name(const BlockDriverState
*bs
)
6276 /* If multiple parents have a name, just pick the first one. */
6277 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
6278 if (c
->klass
->get_name
) {
6279 name
= c
->klass
->get_name(c
);
6280 if (name
&& *name
) {
6289 /* TODO check what callers really want: bs->node_name or blk_name() */
6290 const char *bdrv_get_device_name(const BlockDriverState
*bs
)
6293 return bdrv_get_parent_name(bs
) ?: "";
6296 /* This can be used to identify nodes that might not have a device
6297 * name associated. Since node and device names live in the same
6298 * namespace, the result is unambiguous. The exception is if both are
6299 * absent, then this returns an empty (non-null) string. */
6300 const char *bdrv_get_device_or_node_name(const BlockDriverState
*bs
)
6303 return bdrv_get_parent_name(bs
) ?: bs
->node_name
;
6306 int bdrv_get_flags(BlockDriverState
*bs
)
6309 return bs
->open_flags
;
6312 int bdrv_has_zero_init_1(BlockDriverState
*bs
)
6314 GLOBAL_STATE_CODE();
6318 int bdrv_has_zero_init(BlockDriverState
*bs
)
6320 BlockDriverState
*filtered
;
6321 GLOBAL_STATE_CODE();
6327 /* If BS is a copy on write image, it is initialized to
6328 the contents of the base image, which may not be zeroes. */
6329 if (bdrv_cow_child(bs
)) {
6332 if (bs
->drv
->bdrv_has_zero_init
) {
6333 return bs
->drv
->bdrv_has_zero_init(bs
);
6336 filtered
= bdrv_filter_bs(bs
);
6338 return bdrv_has_zero_init(filtered
);
6345 bool bdrv_can_write_zeroes_with_unmap(BlockDriverState
*bs
)
6348 if (!(bs
->open_flags
& BDRV_O_UNMAP
)) {
6352 return bs
->supported_zero_flags
& BDRV_REQ_MAY_UNMAP
;
6355 void bdrv_get_backing_filename(BlockDriverState
*bs
,
6356 char *filename
, int filename_size
)
6359 pstrcpy(filename
, filename_size
, bs
->backing_file
);
6362 int bdrv_get_info(BlockDriverState
*bs
, BlockDriverInfo
*bdi
)
6365 BlockDriver
*drv
= bs
->drv
;
6367 /* if bs->drv == NULL, bs is closed, so there's nothing to do here */
6371 if (!drv
->bdrv_get_info
) {
6372 BlockDriverState
*filtered
= bdrv_filter_bs(bs
);
6374 return bdrv_get_info(filtered
, bdi
);
6378 memset(bdi
, 0, sizeof(*bdi
));
6379 ret
= drv
->bdrv_get_info(bs
, bdi
);
6384 if (bdi
->cluster_size
> BDRV_MAX_ALIGNMENT
) {
6391 ImageInfoSpecific
*bdrv_get_specific_info(BlockDriverState
*bs
,
6394 BlockDriver
*drv
= bs
->drv
;
6396 if (drv
&& drv
->bdrv_get_specific_info
) {
6397 return drv
->bdrv_get_specific_info(bs
, errp
);
6402 BlockStatsSpecific
*bdrv_get_specific_stats(BlockDriverState
*bs
)
6404 BlockDriver
*drv
= bs
->drv
;
6406 if (!drv
|| !drv
->bdrv_get_specific_stats
) {
6409 return drv
->bdrv_get_specific_stats(bs
);
6412 void bdrv_debug_event(BlockDriverState
*bs
, BlkdebugEvent event
)
6415 if (!bs
|| !bs
->drv
|| !bs
->drv
->bdrv_debug_event
) {
6419 bs
->drv
->bdrv_debug_event(bs
, event
);
6422 static BlockDriverState
*bdrv_find_debug_node(BlockDriverState
*bs
)
6424 GLOBAL_STATE_CODE();
6425 while (bs
&& bs
->drv
&& !bs
->drv
->bdrv_debug_breakpoint
) {
6426 bs
= bdrv_primary_bs(bs
);
6429 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_breakpoint
) {
6430 assert(bs
->drv
->bdrv_debug_remove_breakpoint
);
6437 int bdrv_debug_breakpoint(BlockDriverState
*bs
, const char *event
,
6440 GLOBAL_STATE_CODE();
6441 bs
= bdrv_find_debug_node(bs
);
6443 return bs
->drv
->bdrv_debug_breakpoint(bs
, event
, tag
);
6449 int bdrv_debug_remove_breakpoint(BlockDriverState
*bs
, const char *tag
)
6451 GLOBAL_STATE_CODE();
6452 bs
= bdrv_find_debug_node(bs
);
6454 return bs
->drv
->bdrv_debug_remove_breakpoint(bs
, tag
);
6460 int bdrv_debug_resume(BlockDriverState
*bs
, const char *tag
)
6462 GLOBAL_STATE_CODE();
6463 while (bs
&& (!bs
->drv
|| !bs
->drv
->bdrv_debug_resume
)) {
6464 bs
= bdrv_primary_bs(bs
);
6467 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_resume
) {
6468 return bs
->drv
->bdrv_debug_resume(bs
, tag
);
6474 bool bdrv_debug_is_suspended(BlockDriverState
*bs
, const char *tag
)
6476 GLOBAL_STATE_CODE();
6477 while (bs
&& bs
->drv
&& !bs
->drv
->bdrv_debug_is_suspended
) {
6478 bs
= bdrv_primary_bs(bs
);
6481 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_is_suspended
) {
6482 return bs
->drv
->bdrv_debug_is_suspended(bs
, tag
);
6488 /* backing_file can either be relative, or absolute, or a protocol. If it is
6489 * relative, it must be relative to the chain. So, passing in bs->filename
6490 * from a BDS as backing_file should not be done, as that may be relative to
6491 * the CWD rather than the chain. */
6492 BlockDriverState
*bdrv_find_backing_image(BlockDriverState
*bs
,
6493 const char *backing_file
)
6495 char *filename_full
= NULL
;
6496 char *backing_file_full
= NULL
;
6497 char *filename_tmp
= NULL
;
6498 int is_protocol
= 0;
6499 bool filenames_refreshed
= false;
6500 BlockDriverState
*curr_bs
= NULL
;
6501 BlockDriverState
*retval
= NULL
;
6502 BlockDriverState
*bs_below
;
6504 GLOBAL_STATE_CODE();
6506 if (!bs
|| !bs
->drv
|| !backing_file
) {
6510 filename_full
= g_malloc(PATH_MAX
);
6511 backing_file_full
= g_malloc(PATH_MAX
);
6513 is_protocol
= path_has_protocol(backing_file
);
6516 * Being largely a legacy function, skip any filters here
6517 * (because filters do not have normal filenames, so they cannot
6518 * match anyway; and allowing json:{} filenames is a bit out of
6521 for (curr_bs
= bdrv_skip_filters(bs
);
6522 bdrv_cow_child(curr_bs
) != NULL
;
6525 bs_below
= bdrv_backing_chain_next(curr_bs
);
6527 if (bdrv_backing_overridden(curr_bs
)) {
6529 * If the backing file was overridden, we can only compare
6530 * directly against the backing node's filename.
6533 if (!filenames_refreshed
) {
6535 * This will automatically refresh all of the
6536 * filenames in the rest of the backing chain, so we
6537 * only need to do this once.
6539 bdrv_refresh_filename(bs_below
);
6540 filenames_refreshed
= true;
6543 if (strcmp(backing_file
, bs_below
->filename
) == 0) {
6547 } else if (is_protocol
|| path_has_protocol(curr_bs
->backing_file
)) {
6549 * If either of the filename paths is actually a protocol, then
6550 * compare unmodified paths; otherwise make paths relative.
6552 char *backing_file_full_ret
;
6554 if (strcmp(backing_file
, curr_bs
->backing_file
) == 0) {
6558 /* Also check against the full backing filename for the image */
6559 backing_file_full_ret
= bdrv_get_full_backing_filename(curr_bs
,
6561 if (backing_file_full_ret
) {
6562 bool equal
= strcmp(backing_file
, backing_file_full_ret
) == 0;
6563 g_free(backing_file_full_ret
);
6570 /* If not an absolute filename path, make it relative to the current
6571 * image's filename path */
6572 filename_tmp
= bdrv_make_absolute_filename(curr_bs
, backing_file
,
6574 /* We are going to compare canonicalized absolute pathnames */
6575 if (!filename_tmp
|| !realpath(filename_tmp
, filename_full
)) {
6576 g_free(filename_tmp
);
6579 g_free(filename_tmp
);
6581 /* We need to make sure the backing filename we are comparing against
6582 * is relative to the current image filename (or absolute) */
6583 filename_tmp
= bdrv_get_full_backing_filename(curr_bs
, NULL
);
6584 if (!filename_tmp
|| !realpath(filename_tmp
, backing_file_full
)) {
6585 g_free(filename_tmp
);
6588 g_free(filename_tmp
);
6590 if (strcmp(backing_file_full
, filename_full
) == 0) {
6597 g_free(filename_full
);
6598 g_free(backing_file_full
);
6602 void bdrv_init(void)
6604 #ifdef CONFIG_BDRV_WHITELIST_TOOLS
6605 use_bdrv_whitelist
= 1;
6607 module_call_init(MODULE_INIT_BLOCK
);
6610 void bdrv_init_with_whitelist(void)
6612 use_bdrv_whitelist
= 1;
6616 int bdrv_activate(BlockDriverState
*bs
, Error
**errp
)
6618 BdrvChild
*child
, *parent
;
6619 Error
*local_err
= NULL
;
6621 BdrvDirtyBitmap
*bm
;
6623 GLOBAL_STATE_CODE();
6629 QLIST_FOREACH(child
, &bs
->children
, next
) {
6630 bdrv_activate(child
->bs
, &local_err
);
6632 error_propagate(errp
, local_err
);
6638 * Update permissions, they may differ for inactive nodes.
6640 * Note that the required permissions of inactive images are always a
6641 * subset of the permissions required after activating the image. This
6642 * allows us to just get the permissions upfront without restricting
6643 * bdrv_co_invalidate_cache().
6645 * It also means that in error cases, we don't have to try and revert to
6646 * the old permissions (which is an operation that could fail, too). We can
6647 * just keep the extended permissions for the next time that an activation
6648 * of the image is tried.
6650 if (bs
->open_flags
& BDRV_O_INACTIVE
) {
6651 bs
->open_flags
&= ~BDRV_O_INACTIVE
;
6652 ret
= bdrv_refresh_perms(bs
, errp
);
6654 bs
->open_flags
|= BDRV_O_INACTIVE
;
6658 ret
= bdrv_invalidate_cache(bs
, errp
);
6660 bs
->open_flags
|= BDRV_O_INACTIVE
;
6664 FOR_EACH_DIRTY_BITMAP(bs
, bm
) {
6665 bdrv_dirty_bitmap_skip_store(bm
, false);
6668 ret
= refresh_total_sectors(bs
, bs
->total_sectors
);
6670 bs
->open_flags
|= BDRV_O_INACTIVE
;
6671 error_setg_errno(errp
, -ret
, "Could not refresh total sector count");
6676 QLIST_FOREACH(parent
, &bs
->parents
, next_parent
) {
6677 if (parent
->klass
->activate
) {
6678 parent
->klass
->activate(parent
, &local_err
);
6680 bs
->open_flags
|= BDRV_O_INACTIVE
;
6681 error_propagate(errp
, local_err
);
6690 int coroutine_fn
bdrv_co_invalidate_cache(BlockDriverState
*bs
, Error
**errp
)
6692 Error
*local_err
= NULL
;
6695 assert(!(bs
->open_flags
& BDRV_O_INACTIVE
));
6697 if (bs
->drv
->bdrv_co_invalidate_cache
) {
6698 bs
->drv
->bdrv_co_invalidate_cache(bs
, &local_err
);
6700 error_propagate(errp
, local_err
);
6708 void bdrv_activate_all(Error
**errp
)
6710 BlockDriverState
*bs
;
6711 BdrvNextIterator it
;
6713 GLOBAL_STATE_CODE();
6715 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
6716 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
6719 aio_context_acquire(aio_context
);
6720 ret
= bdrv_activate(bs
, errp
);
6721 aio_context_release(aio_context
);
6723 bdrv_next_cleanup(&it
);
6729 static bool bdrv_has_bds_parent(BlockDriverState
*bs
, bool only_active
)
6732 GLOBAL_STATE_CODE();
6734 QLIST_FOREACH(parent
, &bs
->parents
, next_parent
) {
6735 if (parent
->klass
->parent_is_bds
) {
6736 BlockDriverState
*parent_bs
= parent
->opaque
;
6737 if (!only_active
|| !(parent_bs
->open_flags
& BDRV_O_INACTIVE
)) {
6746 static int bdrv_inactivate_recurse(BlockDriverState
*bs
)
6748 BdrvChild
*child
, *parent
;
6750 uint64_t cumulative_perms
, cumulative_shared_perms
;
6752 GLOBAL_STATE_CODE();
6758 /* Make sure that we don't inactivate a child before its parent.
6759 * It will be covered by recursion from the yet active parent. */
6760 if (bdrv_has_bds_parent(bs
, true)) {
6764 assert(!(bs
->open_flags
& BDRV_O_INACTIVE
));
6766 /* Inactivate this node */
6767 if (bs
->drv
->bdrv_inactivate
) {
6768 ret
= bs
->drv
->bdrv_inactivate(bs
);
6774 QLIST_FOREACH(parent
, &bs
->parents
, next_parent
) {
6775 if (parent
->klass
->inactivate
) {
6776 ret
= parent
->klass
->inactivate(parent
);
6783 bdrv_get_cumulative_perm(bs
, &cumulative_perms
,
6784 &cumulative_shared_perms
);
6785 if (cumulative_perms
& (BLK_PERM_WRITE
| BLK_PERM_WRITE_UNCHANGED
)) {
6786 /* Our inactive parents still need write access. Inactivation failed. */
6790 bs
->open_flags
|= BDRV_O_INACTIVE
;
6793 * Update permissions, they may differ for inactive nodes.
6794 * We only tried to loosen restrictions, so errors are not fatal, ignore
6797 bdrv_refresh_perms(bs
, NULL
);
6799 /* Recursively inactivate children */
6800 QLIST_FOREACH(child
, &bs
->children
, next
) {
6801 ret
= bdrv_inactivate_recurse(child
->bs
);
6810 int bdrv_inactivate_all(void)
6812 BlockDriverState
*bs
= NULL
;
6813 BdrvNextIterator it
;
6815 GSList
*aio_ctxs
= NULL
, *ctx
;
6817 GLOBAL_STATE_CODE();
6819 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
6820 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
6822 if (!g_slist_find(aio_ctxs
, aio_context
)) {
6823 aio_ctxs
= g_slist_prepend(aio_ctxs
, aio_context
);
6824 aio_context_acquire(aio_context
);
6828 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
6829 /* Nodes with BDS parents are covered by recursion from the last
6830 * parent that gets inactivated. Don't inactivate them a second
6831 * time if that has already happened. */
6832 if (bdrv_has_bds_parent(bs
, false)) {
6835 ret
= bdrv_inactivate_recurse(bs
);
6837 bdrv_next_cleanup(&it
);
6843 for (ctx
= aio_ctxs
; ctx
!= NULL
; ctx
= ctx
->next
) {
6844 AioContext
*aio_context
= ctx
->data
;
6845 aio_context_release(aio_context
);
6847 g_slist_free(aio_ctxs
);
6852 /**************************************************************/
6853 /* removable device support */
6856 * Return TRUE if the media is present
6858 bool bdrv_is_inserted(BlockDriverState
*bs
)
6860 BlockDriver
*drv
= bs
->drv
;
6867 if (drv
->bdrv_is_inserted
) {
6868 return drv
->bdrv_is_inserted(bs
);
6870 QLIST_FOREACH(child
, &bs
->children
, next
) {
6871 if (!bdrv_is_inserted(child
->bs
)) {
6879 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
6881 void bdrv_eject(BlockDriverState
*bs
, bool eject_flag
)
6883 BlockDriver
*drv
= bs
->drv
;
6886 if (drv
&& drv
->bdrv_eject
) {
6887 drv
->bdrv_eject(bs
, eject_flag
);
6892 * Lock or unlock the media (if it is locked, the user won't be able
6893 * to eject it manually).
6895 void bdrv_lock_medium(BlockDriverState
*bs
, bool locked
)
6897 BlockDriver
*drv
= bs
->drv
;
6899 trace_bdrv_lock_medium(bs
, locked
);
6901 if (drv
&& drv
->bdrv_lock_medium
) {
6902 drv
->bdrv_lock_medium(bs
, locked
);
6906 /* Get a reference to bs */
6907 void bdrv_ref(BlockDriverState
*bs
)
6909 GLOBAL_STATE_CODE();
6913 /* Release a previously grabbed reference to bs.
6914 * If after releasing, reference count is zero, the BlockDriverState is
6916 void bdrv_unref(BlockDriverState
*bs
)
6918 GLOBAL_STATE_CODE();
6922 assert(bs
->refcnt
> 0);
6923 if (--bs
->refcnt
== 0) {
6928 struct BdrvOpBlocker
{
6930 QLIST_ENTRY(BdrvOpBlocker
) list
;
6933 bool bdrv_op_is_blocked(BlockDriverState
*bs
, BlockOpType op
, Error
**errp
)
6935 BdrvOpBlocker
*blocker
;
6936 GLOBAL_STATE_CODE();
6937 assert((int) op
>= 0 && op
< BLOCK_OP_TYPE_MAX
);
6938 if (!QLIST_EMPTY(&bs
->op_blockers
[op
])) {
6939 blocker
= QLIST_FIRST(&bs
->op_blockers
[op
]);
6940 error_propagate_prepend(errp
, error_copy(blocker
->reason
),
6941 "Node '%s' is busy: ",
6942 bdrv_get_device_or_node_name(bs
));
6948 void bdrv_op_block(BlockDriverState
*bs
, BlockOpType op
, Error
*reason
)
6950 BdrvOpBlocker
*blocker
;
6951 GLOBAL_STATE_CODE();
6952 assert((int) op
>= 0 && op
< BLOCK_OP_TYPE_MAX
);
6954 blocker
= g_new0(BdrvOpBlocker
, 1);
6955 blocker
->reason
= reason
;
6956 QLIST_INSERT_HEAD(&bs
->op_blockers
[op
], blocker
, list
);
6959 void bdrv_op_unblock(BlockDriverState
*bs
, BlockOpType op
, Error
*reason
)
6961 BdrvOpBlocker
*blocker
, *next
;
6962 GLOBAL_STATE_CODE();
6963 assert((int) op
>= 0 && op
< BLOCK_OP_TYPE_MAX
);
6964 QLIST_FOREACH_SAFE(blocker
, &bs
->op_blockers
[op
], list
, next
) {
6965 if (blocker
->reason
== reason
) {
6966 QLIST_REMOVE(blocker
, list
);
6972 void bdrv_op_block_all(BlockDriverState
*bs
, Error
*reason
)
6975 GLOBAL_STATE_CODE();
6976 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
6977 bdrv_op_block(bs
, i
, reason
);
6981 void bdrv_op_unblock_all(BlockDriverState
*bs
, Error
*reason
)
6984 GLOBAL_STATE_CODE();
6985 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
6986 bdrv_op_unblock(bs
, i
, reason
);
6990 bool bdrv_op_blocker_is_empty(BlockDriverState
*bs
)
6993 GLOBAL_STATE_CODE();
6994 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
6995 if (!QLIST_EMPTY(&bs
->op_blockers
[i
])) {
7002 void bdrv_img_create(const char *filename
, const char *fmt
,
7003 const char *base_filename
, const char *base_fmt
,
7004 char *options
, uint64_t img_size
, int flags
, bool quiet
,
7007 QemuOptsList
*create_opts
= NULL
;
7008 QemuOpts
*opts
= NULL
;
7009 const char *backing_fmt
, *backing_file
;
7011 BlockDriver
*drv
, *proto_drv
;
7012 Error
*local_err
= NULL
;
7015 GLOBAL_STATE_CODE();
7017 /* Find driver and parse its options */
7018 drv
= bdrv_find_format(fmt
);
7020 error_setg(errp
, "Unknown file format '%s'", fmt
);
7024 proto_drv
= bdrv_find_protocol(filename
, true, errp
);
7029 if (!drv
->create_opts
) {
7030 error_setg(errp
, "Format driver '%s' does not support image creation",
7035 if (!proto_drv
->create_opts
) {
7036 error_setg(errp
, "Protocol driver '%s' does not support image creation",
7037 proto_drv
->format_name
);
7041 /* Create parameter list */
7042 create_opts
= qemu_opts_append(create_opts
, drv
->create_opts
);
7043 create_opts
= qemu_opts_append(create_opts
, proto_drv
->create_opts
);
7045 opts
= qemu_opts_create(create_opts
, NULL
, 0, &error_abort
);
7047 /* Parse -o options */
7049 if (!qemu_opts_do_parse(opts
, options
, NULL
, errp
)) {
7054 if (!qemu_opt_get(opts
, BLOCK_OPT_SIZE
)) {
7055 qemu_opt_set_number(opts
, BLOCK_OPT_SIZE
, img_size
, &error_abort
);
7056 } else if (img_size
!= UINT64_C(-1)) {
7057 error_setg(errp
, "The image size must be specified only once");
7061 if (base_filename
) {
7062 if (!qemu_opt_set(opts
, BLOCK_OPT_BACKING_FILE
, base_filename
,
7064 error_setg(errp
, "Backing file not supported for file format '%s'",
7071 if (!qemu_opt_set(opts
, BLOCK_OPT_BACKING_FMT
, base_fmt
, NULL
)) {
7072 error_setg(errp
, "Backing file format not supported for file "
7073 "format '%s'", fmt
);
7078 backing_file
= qemu_opt_get(opts
, BLOCK_OPT_BACKING_FILE
);
7080 if (!strcmp(filename
, backing_file
)) {
7081 error_setg(errp
, "Error: Trying to create an image with the "
7082 "same filename as the backing file");
7085 if (backing_file
[0] == '\0') {
7086 error_setg(errp
, "Expected backing file name, got empty string");
7091 backing_fmt
= qemu_opt_get(opts
, BLOCK_OPT_BACKING_FMT
);
7093 /* The size for the image must always be specified, unless we have a backing
7094 * file and we have not been forbidden from opening it. */
7095 size
= qemu_opt_get_size(opts
, BLOCK_OPT_SIZE
, img_size
);
7096 if (backing_file
&& !(flags
& BDRV_O_NO_BACKING
)) {
7097 BlockDriverState
*bs
;
7100 QDict
*backing_options
= NULL
;
7103 bdrv_get_full_backing_filename_from_filename(filename
, backing_file
,
7108 assert(full_backing
);
7111 * No need to do I/O here, which allows us to open encrypted
7112 * backing images without needing the secret
7115 back_flags
&= ~(BDRV_O_RDWR
| BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
);
7116 back_flags
|= BDRV_O_NO_IO
;
7118 backing_options
= qdict_new();
7120 qdict_put_str(backing_options
, "driver", backing_fmt
);
7122 qdict_put_bool(backing_options
, BDRV_OPT_FORCE_SHARE
, true);
7124 bs
= bdrv_open(full_backing
, NULL
, backing_options
, back_flags
,
7126 g_free(full_backing
);
7128 error_append_hint(&local_err
, "Could not open backing image.\n");
7132 error_setg(&local_err
,
7133 "Backing file specified without backing format");
7134 error_append_hint(&local_err
, "Detected format of %s.",
7135 bs
->drv
->format_name
);
7139 /* Opened BS, have no size */
7140 size
= bdrv_getlength(bs
);
7142 error_setg_errno(errp
, -size
, "Could not get size of '%s'",
7147 qemu_opt_set_number(opts
, BLOCK_OPT_SIZE
, size
, &error_abort
);
7151 /* (backing_file && !(flags & BDRV_O_NO_BACKING)) */
7152 } else if (backing_file
&& !backing_fmt
) {
7153 error_setg(&local_err
,
7154 "Backing file specified without backing format");
7159 error_setg(errp
, "Image creation needs a size parameter");
7164 printf("Formatting '%s', fmt=%s ", filename
, fmt
);
7165 qemu_opts_print(opts
, " ");
7170 ret
= bdrv_create(drv
, filename
, opts
, &local_err
);
7172 if (ret
== -EFBIG
) {
7173 /* This is generally a better message than whatever the driver would
7174 * deliver (especially because of the cluster_size_hint), since that
7175 * is most probably not much different from "image too large". */
7176 const char *cluster_size_hint
= "";
7177 if (qemu_opt_get_size(opts
, BLOCK_OPT_CLUSTER_SIZE
, 0)) {
7178 cluster_size_hint
= " (try using a larger cluster size)";
7180 error_setg(errp
, "The image size is too large for file format '%s'"
7181 "%s", fmt
, cluster_size_hint
);
7182 error_free(local_err
);
7187 qemu_opts_del(opts
);
7188 qemu_opts_free(create_opts
);
7189 error_propagate(errp
, local_err
);
7192 AioContext
*bdrv_get_aio_context(BlockDriverState
*bs
)
7195 return bs
? bs
->aio_context
: qemu_get_aio_context();
7198 AioContext
*coroutine_fn
bdrv_co_enter(BlockDriverState
*bs
)
7200 Coroutine
*self
= qemu_coroutine_self();
7201 AioContext
*old_ctx
= qemu_coroutine_get_aio_context(self
);
7202 AioContext
*new_ctx
;
7206 * Increase bs->in_flight to ensure that this operation is completed before
7207 * moving the node to a different AioContext. Read new_ctx only afterwards.
7209 bdrv_inc_in_flight(bs
);
7211 new_ctx
= bdrv_get_aio_context(bs
);
7212 aio_co_reschedule_self(new_ctx
);
7216 void coroutine_fn
bdrv_co_leave(BlockDriverState
*bs
, AioContext
*old_ctx
)
7219 aio_co_reschedule_self(old_ctx
);
7220 bdrv_dec_in_flight(bs
);
7223 void coroutine_fn
bdrv_co_lock(BlockDriverState
*bs
)
7225 AioContext
*ctx
= bdrv_get_aio_context(bs
);
7227 /* In the main thread, bs->aio_context won't change concurrently */
7228 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
7231 * We're in coroutine context, so we already hold the lock of the main
7232 * loop AioContext. Don't lock it twice to avoid deadlocks.
7234 assert(qemu_in_coroutine());
7235 if (ctx
!= qemu_get_aio_context()) {
7236 aio_context_acquire(ctx
);
7240 void coroutine_fn
bdrv_co_unlock(BlockDriverState
*bs
)
7242 AioContext
*ctx
= bdrv_get_aio_context(bs
);
7244 assert(qemu_in_coroutine());
7245 if (ctx
!= qemu_get_aio_context()) {
7246 aio_context_release(ctx
);
7250 void bdrv_coroutine_enter(BlockDriverState
*bs
, Coroutine
*co
)
7253 aio_co_enter(bdrv_get_aio_context(bs
), co
);
7256 static void bdrv_do_remove_aio_context_notifier(BdrvAioNotifier
*ban
)
7258 GLOBAL_STATE_CODE();
7259 QLIST_REMOVE(ban
, list
);
7263 static void bdrv_detach_aio_context(BlockDriverState
*bs
)
7265 BdrvAioNotifier
*baf
, *baf_tmp
;
7267 assert(!bs
->walking_aio_notifiers
);
7268 GLOBAL_STATE_CODE();
7269 bs
->walking_aio_notifiers
= true;
7270 QLIST_FOREACH_SAFE(baf
, &bs
->aio_notifiers
, list
, baf_tmp
) {
7272 bdrv_do_remove_aio_context_notifier(baf
);
7274 baf
->detach_aio_context(baf
->opaque
);
7277 /* Never mind iterating again to check for ->deleted. bdrv_close() will
7278 * remove remaining aio notifiers if we aren't called again.
7280 bs
->walking_aio_notifiers
= false;
7282 if (bs
->drv
&& bs
->drv
->bdrv_detach_aio_context
) {
7283 bs
->drv
->bdrv_detach_aio_context(bs
);
7286 if (bs
->quiesce_counter
) {
7287 aio_enable_external(bs
->aio_context
);
7289 bs
->aio_context
= NULL
;
7292 static void bdrv_attach_aio_context(BlockDriverState
*bs
,
7293 AioContext
*new_context
)
7295 BdrvAioNotifier
*ban
, *ban_tmp
;
7296 GLOBAL_STATE_CODE();
7298 if (bs
->quiesce_counter
) {
7299 aio_disable_external(new_context
);
7302 bs
->aio_context
= new_context
;
7304 if (bs
->drv
&& bs
->drv
->bdrv_attach_aio_context
) {
7305 bs
->drv
->bdrv_attach_aio_context(bs
, new_context
);
7308 assert(!bs
->walking_aio_notifiers
);
7309 bs
->walking_aio_notifiers
= true;
7310 QLIST_FOREACH_SAFE(ban
, &bs
->aio_notifiers
, list
, ban_tmp
) {
7312 bdrv_do_remove_aio_context_notifier(ban
);
7314 ban
->attached_aio_context(new_context
, ban
->opaque
);
7317 bs
->walking_aio_notifiers
= false;
7321 * Changes the AioContext used for fd handlers, timers, and BHs by this
7322 * BlockDriverState and all its children and parents.
7324 * Must be called from the main AioContext.
7326 * The caller must own the AioContext lock for the old AioContext of bs, but it
7327 * must not own the AioContext lock for new_context (unless new_context is the
7328 * same as the current context of bs).
7330 * @ignore will accumulate all visited BdrvChild object. The caller is
7331 * responsible for freeing the list afterwards.
7333 void bdrv_set_aio_context_ignore(BlockDriverState
*bs
,
7334 AioContext
*new_context
, GSList
**ignore
)
7336 AioContext
*old_context
= bdrv_get_aio_context(bs
);
7337 GSList
*children_to_process
= NULL
;
7338 GSList
*parents_to_process
= NULL
;
7340 BdrvChild
*child
, *parent
;
7342 g_assert(qemu_get_current_aio_context() == qemu_get_aio_context());
7343 GLOBAL_STATE_CODE();
7345 if (old_context
== new_context
) {
7349 bdrv_drained_begin(bs
);
7351 QLIST_FOREACH(child
, &bs
->children
, next
) {
7352 if (g_slist_find(*ignore
, child
)) {
7355 *ignore
= g_slist_prepend(*ignore
, child
);
7356 children_to_process
= g_slist_prepend(children_to_process
, child
);
7359 QLIST_FOREACH(parent
, &bs
->parents
, next_parent
) {
7360 if (g_slist_find(*ignore
, parent
)) {
7363 *ignore
= g_slist_prepend(*ignore
, parent
);
7364 parents_to_process
= g_slist_prepend(parents_to_process
, parent
);
7367 for (entry
= children_to_process
;
7369 entry
= g_slist_next(entry
)) {
7370 child
= entry
->data
;
7371 bdrv_set_aio_context_ignore(child
->bs
, new_context
, ignore
);
7373 g_slist_free(children_to_process
);
7375 for (entry
= parents_to_process
;
7377 entry
= g_slist_next(entry
)) {
7378 parent
= entry
->data
;
7379 assert(parent
->klass
->set_aio_ctx
);
7380 parent
->klass
->set_aio_ctx(parent
, new_context
, ignore
);
7382 g_slist_free(parents_to_process
);
7384 bdrv_detach_aio_context(bs
);
7386 /* Acquire the new context, if necessary */
7387 if (qemu_get_aio_context() != new_context
) {
7388 aio_context_acquire(new_context
);
7391 bdrv_attach_aio_context(bs
, new_context
);
7394 * If this function was recursively called from
7395 * bdrv_set_aio_context_ignore(), there may be nodes in the
7396 * subtree that have not yet been moved to the new AioContext.
7397 * Release the old one so bdrv_drained_end() can poll them.
7399 if (qemu_get_aio_context() != old_context
) {
7400 aio_context_release(old_context
);
7403 bdrv_drained_end(bs
);
7405 if (qemu_get_aio_context() != old_context
) {
7406 aio_context_acquire(old_context
);
7408 if (qemu_get_aio_context() != new_context
) {
7409 aio_context_release(new_context
);
7413 static bool bdrv_parent_can_set_aio_context(BdrvChild
*c
, AioContext
*ctx
,
7414 GSList
**ignore
, Error
**errp
)
7416 GLOBAL_STATE_CODE();
7417 if (g_slist_find(*ignore
, c
)) {
7420 *ignore
= g_slist_prepend(*ignore
, c
);
7423 * A BdrvChildClass that doesn't handle AioContext changes cannot
7424 * tolerate any AioContext changes
7426 if (!c
->klass
->can_set_aio_ctx
) {
7427 char *user
= bdrv_child_user_desc(c
);
7428 error_setg(errp
, "Changing iothreads is not supported by %s", user
);
7432 if (!c
->klass
->can_set_aio_ctx(c
, ctx
, ignore
, errp
)) {
7433 assert(!errp
|| *errp
);
7439 bool bdrv_child_can_set_aio_context(BdrvChild
*c
, AioContext
*ctx
,
7440 GSList
**ignore
, Error
**errp
)
7442 GLOBAL_STATE_CODE();
7443 if (g_slist_find(*ignore
, c
)) {
7446 *ignore
= g_slist_prepend(*ignore
, c
);
7447 return bdrv_can_set_aio_context(c
->bs
, ctx
, ignore
, errp
);
7450 /* @ignore will accumulate all visited BdrvChild object. The caller is
7451 * responsible for freeing the list afterwards. */
7452 bool bdrv_can_set_aio_context(BlockDriverState
*bs
, AioContext
*ctx
,
7453 GSList
**ignore
, Error
**errp
)
7457 if (bdrv_get_aio_context(bs
) == ctx
) {
7461 GLOBAL_STATE_CODE();
7463 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
7464 if (!bdrv_parent_can_set_aio_context(c
, ctx
, ignore
, errp
)) {
7468 QLIST_FOREACH(c
, &bs
->children
, next
) {
7469 if (!bdrv_child_can_set_aio_context(c
, ctx
, ignore
, errp
)) {
7477 int bdrv_child_try_set_aio_context(BlockDriverState
*bs
, AioContext
*ctx
,
7478 BdrvChild
*ignore_child
, Error
**errp
)
7483 GLOBAL_STATE_CODE();
7485 ignore
= ignore_child
? g_slist_prepend(NULL
, ignore_child
) : NULL
;
7486 ret
= bdrv_can_set_aio_context(bs
, ctx
, &ignore
, errp
);
7487 g_slist_free(ignore
);
7493 ignore
= ignore_child
? g_slist_prepend(NULL
, ignore_child
) : NULL
;
7494 bdrv_set_aio_context_ignore(bs
, ctx
, &ignore
);
7495 g_slist_free(ignore
);
7500 int bdrv_try_set_aio_context(BlockDriverState
*bs
, AioContext
*ctx
,
7503 GLOBAL_STATE_CODE();
7504 return bdrv_child_try_set_aio_context(bs
, ctx
, NULL
, errp
);
7507 void bdrv_add_aio_context_notifier(BlockDriverState
*bs
,
7508 void (*attached_aio_context
)(AioContext
*new_context
, void *opaque
),
7509 void (*detach_aio_context
)(void *opaque
), void *opaque
)
7511 BdrvAioNotifier
*ban
= g_new(BdrvAioNotifier
, 1);
7512 *ban
= (BdrvAioNotifier
){
7513 .attached_aio_context
= attached_aio_context
,
7514 .detach_aio_context
= detach_aio_context
,
7517 GLOBAL_STATE_CODE();
7519 QLIST_INSERT_HEAD(&bs
->aio_notifiers
, ban
, list
);
7522 void bdrv_remove_aio_context_notifier(BlockDriverState
*bs
,
7523 void (*attached_aio_context
)(AioContext
*,
7525 void (*detach_aio_context
)(void *),
7528 BdrvAioNotifier
*ban
, *ban_next
;
7529 GLOBAL_STATE_CODE();
7531 QLIST_FOREACH_SAFE(ban
, &bs
->aio_notifiers
, list
, ban_next
) {
7532 if (ban
->attached_aio_context
== attached_aio_context
&&
7533 ban
->detach_aio_context
== detach_aio_context
&&
7534 ban
->opaque
== opaque
&&
7535 ban
->deleted
== false)
7537 if (bs
->walking_aio_notifiers
) {
7538 ban
->deleted
= true;
7540 bdrv_do_remove_aio_context_notifier(ban
);
7549 int bdrv_amend_options(BlockDriverState
*bs
, QemuOpts
*opts
,
7550 BlockDriverAmendStatusCB
*status_cb
, void *cb_opaque
,
7554 GLOBAL_STATE_CODE();
7556 error_setg(errp
, "Node is ejected");
7559 if (!bs
->drv
->bdrv_amend_options
) {
7560 error_setg(errp
, "Block driver '%s' does not support option amendment",
7561 bs
->drv
->format_name
);
7564 return bs
->drv
->bdrv_amend_options(bs
, opts
, status_cb
,
7565 cb_opaque
, force
, errp
);
7569 * This function checks whether the given @to_replace is allowed to be
7570 * replaced by a node that always shows the same data as @bs. This is
7571 * used for example to verify whether the mirror job can replace
7572 * @to_replace by the target mirrored from @bs.
7573 * To be replaceable, @bs and @to_replace may either be guaranteed to
7574 * always show the same data (because they are only connected through
7575 * filters), or some driver may allow replacing one of its children
7576 * because it can guarantee that this child's data is not visible at
7577 * all (for example, for dissenting quorum children that have no other
7580 bool bdrv_recurse_can_replace(BlockDriverState
*bs
,
7581 BlockDriverState
*to_replace
)
7583 BlockDriverState
*filtered
;
7585 GLOBAL_STATE_CODE();
7587 if (!bs
|| !bs
->drv
) {
7591 if (bs
== to_replace
) {
7595 /* See what the driver can do */
7596 if (bs
->drv
->bdrv_recurse_can_replace
) {
7597 return bs
->drv
->bdrv_recurse_can_replace(bs
, to_replace
);
7600 /* For filters without an own implementation, we can recurse on our own */
7601 filtered
= bdrv_filter_bs(bs
);
7603 return bdrv_recurse_can_replace(filtered
, to_replace
);
7611 * Check whether the given @node_name can be replaced by a node that
7612 * has the same data as @parent_bs. If so, return @node_name's BDS;
7615 * @node_name must be a (recursive) *child of @parent_bs (or this
7616 * function will return NULL).
7618 * The result (whether the node can be replaced or not) is only valid
7619 * for as long as no graph or permission changes occur.
7621 BlockDriverState
*check_to_replace_node(BlockDriverState
*parent_bs
,
7622 const char *node_name
, Error
**errp
)
7624 BlockDriverState
*to_replace_bs
= bdrv_find_node(node_name
);
7625 AioContext
*aio_context
;
7627 GLOBAL_STATE_CODE();
7629 if (!to_replace_bs
) {
7630 error_setg(errp
, "Failed to find node with node-name='%s'", node_name
);
7634 aio_context
= bdrv_get_aio_context(to_replace_bs
);
7635 aio_context_acquire(aio_context
);
7637 if (bdrv_op_is_blocked(to_replace_bs
, BLOCK_OP_TYPE_REPLACE
, errp
)) {
7638 to_replace_bs
= NULL
;
7642 /* We don't want arbitrary node of the BDS chain to be replaced only the top
7643 * most non filter in order to prevent data corruption.
7644 * Another benefit is that this tests exclude backing files which are
7645 * blocked by the backing blockers.
7647 if (!bdrv_recurse_can_replace(parent_bs
, to_replace_bs
)) {
7648 error_setg(errp
, "Cannot replace '%s' by a node mirrored from '%s', "
7649 "because it cannot be guaranteed that doing so would not "
7650 "lead to an abrupt change of visible data",
7651 node_name
, parent_bs
->node_name
);
7652 to_replace_bs
= NULL
;
7657 aio_context_release(aio_context
);
7658 return to_replace_bs
;
7662 * Iterates through the list of runtime option keys that are said to
7663 * be "strong" for a BDS. An option is called "strong" if it changes
7664 * a BDS's data. For example, the null block driver's "size" and
7665 * "read-zeroes" options are strong, but its "latency-ns" option is
7668 * If a key returned by this function ends with a dot, all options
7669 * starting with that prefix are strong.
7671 static const char *const *strong_options(BlockDriverState
*bs
,
7672 const char *const *curopt
)
7674 static const char *const global_options
[] = {
7675 "driver", "filename", NULL
7679 return &global_options
[0];
7683 if (curopt
== &global_options
[ARRAY_SIZE(global_options
) - 1] && bs
->drv
) {
7684 curopt
= bs
->drv
->strong_runtime_opts
;
7687 return (curopt
&& *curopt
) ? curopt
: NULL
;
7691 * Copies all strong runtime options from bs->options to the given
7692 * QDict. The set of strong option keys is determined by invoking
7695 * Returns true iff any strong option was present in bs->options (and
7696 * thus copied to the target QDict) with the exception of "filename"
7697 * and "driver". The caller is expected to use this value to decide
7698 * whether the existence of strong options prevents the generation of
7701 static bool append_strong_runtime_options(QDict
*d
, BlockDriverState
*bs
)
7703 bool found_any
= false;
7704 const char *const *option_name
= NULL
;
7710 while ((option_name
= strong_options(bs
, option_name
))) {
7711 bool option_given
= false;
7713 assert(strlen(*option_name
) > 0);
7714 if ((*option_name
)[strlen(*option_name
) - 1] != '.') {
7715 QObject
*entry
= qdict_get(bs
->options
, *option_name
);
7720 qdict_put_obj(d
, *option_name
, qobject_ref(entry
));
7721 option_given
= true;
7723 const QDictEntry
*entry
;
7724 for (entry
= qdict_first(bs
->options
); entry
;
7725 entry
= qdict_next(bs
->options
, entry
))
7727 if (strstart(qdict_entry_key(entry
), *option_name
, NULL
)) {
7728 qdict_put_obj(d
, qdict_entry_key(entry
),
7729 qobject_ref(qdict_entry_value(entry
)));
7730 option_given
= true;
7735 /* While "driver" and "filename" need to be included in a JSON filename,
7736 * their existence does not prohibit generation of a plain filename. */
7737 if (!found_any
&& option_given
&&
7738 strcmp(*option_name
, "driver") && strcmp(*option_name
, "filename"))
7744 if (!qdict_haskey(d
, "driver")) {
7745 /* Drivers created with bdrv_new_open_driver() may not have a
7746 * @driver option. Add it here. */
7747 qdict_put_str(d
, "driver", bs
->drv
->format_name
);
7753 /* Note: This function may return false positives; it may return true
7754 * even if opening the backing file specified by bs's image header
7755 * would result in exactly bs->backing. */
7756 static bool bdrv_backing_overridden(BlockDriverState
*bs
)
7758 GLOBAL_STATE_CODE();
7760 return strcmp(bs
->auto_backing_file
,
7761 bs
->backing
->bs
->filename
);
7763 /* No backing BDS, so if the image header reports any backing
7764 * file, it must have been suppressed */
7765 return bs
->auto_backing_file
[0] != '\0';
7769 /* Updates the following BDS fields:
7770 * - exact_filename: A filename which may be used for opening a block device
7771 * which (mostly) equals the given BDS (even without any
7772 * other options; so reading and writing must return the same
7773 * results, but caching etc. may be different)
7774 * - full_open_options: Options which, when given when opening a block device
7775 * (without a filename), result in a BDS (mostly)
7776 * equalling the given one
7777 * - filename: If exact_filename is set, it is copied here. Otherwise,
7778 * full_open_options is converted to a JSON object, prefixed with
7779 * "json:" (for use through the JSON pseudo protocol) and put here.
7781 void bdrv_refresh_filename(BlockDriverState
*bs
)
7783 BlockDriver
*drv
= bs
->drv
;
7785 BlockDriverState
*primary_child_bs
;
7787 bool backing_overridden
;
7788 bool generate_json_filename
; /* Whether our default implementation should
7789 fill exact_filename (false) or not (true) */
7791 GLOBAL_STATE_CODE();
7797 /* This BDS's file name may depend on any of its children's file names, so
7798 * refresh those first */
7799 QLIST_FOREACH(child
, &bs
->children
, next
) {
7800 bdrv_refresh_filename(child
->bs
);
7804 /* For implicit nodes, just copy everything from the single child */
7805 child
= QLIST_FIRST(&bs
->children
);
7806 assert(QLIST_NEXT(child
, next
) == NULL
);
7808 pstrcpy(bs
->exact_filename
, sizeof(bs
->exact_filename
),
7809 child
->bs
->exact_filename
);
7810 pstrcpy(bs
->filename
, sizeof(bs
->filename
), child
->bs
->filename
);
7812 qobject_unref(bs
->full_open_options
);
7813 bs
->full_open_options
= qobject_ref(child
->bs
->full_open_options
);
7818 backing_overridden
= bdrv_backing_overridden(bs
);
7820 if (bs
->open_flags
& BDRV_O_NO_IO
) {
7821 /* Without I/O, the backing file does not change anything.
7822 * Therefore, in such a case (primarily qemu-img), we can
7823 * pretend the backing file has not been overridden even if
7824 * it technically has been. */
7825 backing_overridden
= false;
7828 /* Gather the options QDict */
7830 generate_json_filename
= append_strong_runtime_options(opts
, bs
);
7831 generate_json_filename
|= backing_overridden
;
7833 if (drv
->bdrv_gather_child_options
) {
7834 /* Some block drivers may not want to present all of their children's
7835 * options, or name them differently from BdrvChild.name */
7836 drv
->bdrv_gather_child_options(bs
, opts
, backing_overridden
);
7838 QLIST_FOREACH(child
, &bs
->children
, next
) {
7839 if (child
== bs
->backing
&& !backing_overridden
) {
7840 /* We can skip the backing BDS if it has not been overridden */
7844 qdict_put(opts
, child
->name
,
7845 qobject_ref(child
->bs
->full_open_options
));
7848 if (backing_overridden
&& !bs
->backing
) {
7849 /* Force no backing file */
7850 qdict_put_null(opts
, "backing");
7854 qobject_unref(bs
->full_open_options
);
7855 bs
->full_open_options
= opts
;
7857 primary_child_bs
= bdrv_primary_bs(bs
);
7859 if (drv
->bdrv_refresh_filename
) {
7860 /* Obsolete information is of no use here, so drop the old file name
7861 * information before refreshing it */
7862 bs
->exact_filename
[0] = '\0';
7864 drv
->bdrv_refresh_filename(bs
);
7865 } else if (primary_child_bs
) {
7867 * Try to reconstruct valid information from the underlying
7868 * file -- this only works for format nodes (filter nodes
7869 * cannot be probed and as such must be selected by the user
7870 * either through an options dict, or through a special
7871 * filename which the filter driver must construct in its
7872 * .bdrv_refresh_filename() implementation).
7875 bs
->exact_filename
[0] = '\0';
7878 * We can use the underlying file's filename if:
7879 * - it has a filename,
7880 * - the current BDS is not a filter,
7881 * - the file is a protocol BDS, and
7882 * - opening that file (as this BDS's format) will automatically create
7883 * the BDS tree we have right now, that is:
7884 * - the user did not significantly change this BDS's behavior with
7885 * some explicit (strong) options
7886 * - no non-file child of this BDS has been overridden by the user
7887 * Both of these conditions are represented by generate_json_filename.
7889 if (primary_child_bs
->exact_filename
[0] &&
7890 primary_child_bs
->drv
->bdrv_file_open
&&
7891 !drv
->is_filter
&& !generate_json_filename
)
7893 strcpy(bs
->exact_filename
, primary_child_bs
->exact_filename
);
7897 if (bs
->exact_filename
[0]) {
7898 pstrcpy(bs
->filename
, sizeof(bs
->filename
), bs
->exact_filename
);
7900 GString
*json
= qobject_to_json(QOBJECT(bs
->full_open_options
));
7901 if (snprintf(bs
->filename
, sizeof(bs
->filename
), "json:%s",
7902 json
->str
) >= sizeof(bs
->filename
)) {
7903 /* Give user a hint if we truncated things. */
7904 strcpy(bs
->filename
+ sizeof(bs
->filename
) - 4, "...");
7906 g_string_free(json
, true);
7910 char *bdrv_dirname(BlockDriverState
*bs
, Error
**errp
)
7912 BlockDriver
*drv
= bs
->drv
;
7913 BlockDriverState
*child_bs
;
7915 GLOBAL_STATE_CODE();
7918 error_setg(errp
, "Node '%s' is ejected", bs
->node_name
);
7922 if (drv
->bdrv_dirname
) {
7923 return drv
->bdrv_dirname(bs
, errp
);
7926 child_bs
= bdrv_primary_bs(bs
);
7928 return bdrv_dirname(child_bs
, errp
);
7931 bdrv_refresh_filename(bs
);
7932 if (bs
->exact_filename
[0] != '\0') {
7933 return path_combine(bs
->exact_filename
, "");
7936 error_setg(errp
, "Cannot generate a base directory for %s nodes",
7942 * Hot add/remove a BDS's child. So the user can take a child offline when
7943 * it is broken and take a new child online
7945 void bdrv_add_child(BlockDriverState
*parent_bs
, BlockDriverState
*child_bs
,
7948 GLOBAL_STATE_CODE();
7949 if (!parent_bs
->drv
|| !parent_bs
->drv
->bdrv_add_child
) {
7950 error_setg(errp
, "The node %s does not support adding a child",
7951 bdrv_get_device_or_node_name(parent_bs
));
7955 if (!QLIST_EMPTY(&child_bs
->parents
)) {
7956 error_setg(errp
, "The node %s already has a parent",
7957 child_bs
->node_name
);
7961 parent_bs
->drv
->bdrv_add_child(parent_bs
, child_bs
, errp
);
7964 void bdrv_del_child(BlockDriverState
*parent_bs
, BdrvChild
*child
, Error
**errp
)
7968 GLOBAL_STATE_CODE();
7969 if (!parent_bs
->drv
|| !parent_bs
->drv
->bdrv_del_child
) {
7970 error_setg(errp
, "The node %s does not support removing a child",
7971 bdrv_get_device_or_node_name(parent_bs
));
7975 QLIST_FOREACH(tmp
, &parent_bs
->children
, next
) {
7982 error_setg(errp
, "The node %s does not have a child named %s",
7983 bdrv_get_device_or_node_name(parent_bs
),
7984 bdrv_get_device_or_node_name(child
->bs
));
7988 parent_bs
->drv
->bdrv_del_child(parent_bs
, child
, errp
);
7991 int bdrv_make_empty(BdrvChild
*c
, Error
**errp
)
7993 BlockDriver
*drv
= c
->bs
->drv
;
7996 GLOBAL_STATE_CODE();
7997 assert(c
->perm
& (BLK_PERM_WRITE
| BLK_PERM_WRITE_UNCHANGED
));
7999 if (!drv
->bdrv_make_empty
) {
8000 error_setg(errp
, "%s does not support emptying nodes",
8005 ret
= drv
->bdrv_make_empty(c
->bs
);
8007 error_setg_errno(errp
, -ret
, "Failed to empty %s",
8016 * Return the child that @bs acts as an overlay for, and from which data may be
8017 * copied in COW or COR operations. Usually this is the backing file.
8019 BdrvChild
*bdrv_cow_child(BlockDriverState
*bs
)
8023 if (!bs
|| !bs
->drv
) {
8027 if (bs
->drv
->is_filter
) {
8035 assert(bs
->backing
->role
& BDRV_CHILD_COW
);
8040 * If @bs acts as a filter for exactly one of its children, return
8043 BdrvChild
*bdrv_filter_child(BlockDriverState
*bs
)
8048 if (!bs
|| !bs
->drv
) {
8052 if (!bs
->drv
->is_filter
) {
8056 /* Only one of @backing or @file may be used */
8057 assert(!(bs
->backing
&& bs
->file
));
8059 c
= bs
->backing
?: bs
->file
;
8064 assert(c
->role
& BDRV_CHILD_FILTERED
);
8069 * Return either the result of bdrv_cow_child() or bdrv_filter_child(),
8070 * whichever is non-NULL.
8072 * Return NULL if both are NULL.
8074 BdrvChild
*bdrv_filter_or_cow_child(BlockDriverState
*bs
)
8076 BdrvChild
*cow_child
= bdrv_cow_child(bs
);
8077 BdrvChild
*filter_child
= bdrv_filter_child(bs
);
8080 /* Filter nodes cannot have COW backing files */
8081 assert(!(cow_child
&& filter_child
));
8083 return cow_child
?: filter_child
;
8087 * Return the primary child of this node: For filters, that is the
8088 * filtered child. For other nodes, that is usually the child storing
8090 * (A generally more helpful description is that this is (usually) the
8091 * child that has the same filename as @bs.)
8093 * Drivers do not necessarily have a primary child; for example quorum
8096 BdrvChild
*bdrv_primary_child(BlockDriverState
*bs
)
8098 BdrvChild
*c
, *found
= NULL
;
8101 QLIST_FOREACH(c
, &bs
->children
, next
) {
8102 if (c
->role
& BDRV_CHILD_PRIMARY
) {
8111 static BlockDriverState
*bdrv_do_skip_filters(BlockDriverState
*bs
,
8112 bool stop_on_explicit_filter
)
8120 while (!(stop_on_explicit_filter
&& !bs
->implicit
)) {
8121 c
= bdrv_filter_child(bs
);
8124 * A filter that is embedded in a working block graph must
8125 * have a child. Assert this here so this function does
8126 * not return a filter node that is not expected by the
8129 assert(!bs
->drv
|| !bs
->drv
->is_filter
);
8135 * Note that this treats nodes with bs->drv == NULL as not being
8136 * filters (bs->drv == NULL should be replaced by something else
8138 * The advantage of this behavior is that this function will thus
8139 * always return a non-NULL value (given a non-NULL @bs).
8146 * Return the first BDS that has not been added implicitly or that
8147 * does not have a filtered child down the chain starting from @bs
8148 * (including @bs itself).
8150 BlockDriverState
*bdrv_skip_implicit_filters(BlockDriverState
*bs
)
8152 GLOBAL_STATE_CODE();
8153 return bdrv_do_skip_filters(bs
, true);
8157 * Return the first BDS that does not have a filtered child down the
8158 * chain starting from @bs (including @bs itself).
8160 BlockDriverState
*bdrv_skip_filters(BlockDriverState
*bs
)
8163 return bdrv_do_skip_filters(bs
, false);
8167 * For a backing chain, return the first non-filter backing image of
8168 * the first non-filter image.
8170 BlockDriverState
*bdrv_backing_chain_next(BlockDriverState
*bs
)
8173 return bdrv_skip_filters(bdrv_cow_bs(bdrv_skip_filters(bs
)));
8177 * Check whether [offset, offset + bytes) overlaps with the cached
8178 * block-status data region.
8180 * If so, and @pnum is not NULL, set *pnum to `bsc.data_end - offset`,
8181 * which is what bdrv_bsc_is_data()'s interface needs.
8182 * Otherwise, *pnum is not touched.
8184 static bool bdrv_bsc_range_overlaps_locked(BlockDriverState
*bs
,
8185 int64_t offset
, int64_t bytes
,
8188 BdrvBlockStatusCache
*bsc
= qatomic_rcu_read(&bs
->block_status_cache
);
8192 qatomic_read(&bsc
->valid
) &&
8193 ranges_overlap(offset
, bytes
, bsc
->data_start
,
8194 bsc
->data_end
- bsc
->data_start
);
8196 if (overlaps
&& pnum
) {
8197 *pnum
= bsc
->data_end
- offset
;
8204 * See block_int.h for this function's documentation.
8206 bool bdrv_bsc_is_data(BlockDriverState
*bs
, int64_t offset
, int64_t *pnum
)
8209 RCU_READ_LOCK_GUARD();
8210 return bdrv_bsc_range_overlaps_locked(bs
, offset
, 1, pnum
);
8214 * See block_int.h for this function's documentation.
8216 void bdrv_bsc_invalidate_range(BlockDriverState
*bs
,
8217 int64_t offset
, int64_t bytes
)
8220 RCU_READ_LOCK_GUARD();
8222 if (bdrv_bsc_range_overlaps_locked(bs
, offset
, bytes
, NULL
)) {
8223 qatomic_set(&bs
->block_status_cache
->valid
, false);
8228 * See block_int.h for this function's documentation.
8230 void bdrv_bsc_fill(BlockDriverState
*bs
, int64_t offset
, int64_t bytes
)
8232 BdrvBlockStatusCache
*new_bsc
= g_new(BdrvBlockStatusCache
, 1);
8233 BdrvBlockStatusCache
*old_bsc
;
8236 *new_bsc
= (BdrvBlockStatusCache
) {
8238 .data_start
= offset
,
8239 .data_end
= offset
+ bytes
,
8242 QEMU_LOCK_GUARD(&bs
->bsc_modify_lock
);
8244 old_bsc
= qatomic_rcu_read(&bs
->block_status_cache
);
8245 qatomic_rcu_set(&bs
->block_status_cache
, new_bsc
);
8247 g_free_rcu(old_bsc
, rcu
);