2 * QEMU System Emulator block driver
4 * Copyright (c) 2003 Fabrice Bellard
5 * Copyright (c) 2020 Virtuozzo International GmbH.
7 * Permission is hereby granted, free of charge, to any person obtaining a copy
8 * of this software and associated documentation files (the "Software"), to deal
9 * in the Software without restriction, including without limitation the rights
10 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
11 * copies of the Software, and to permit persons to whom the Software is
12 * furnished to do so, subject to the following conditions:
14 * The above copyright notice and this permission notice shall be included in
15 * all copies or substantial portions of the Software.
17 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
18 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
19 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
20 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
21 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
22 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
26 #include "qemu/osdep.h"
27 #include "block/trace.h"
28 #include "block/block_int.h"
29 #include "block/blockjob.h"
30 #include "block/fuse.h"
31 #include "block/nbd.h"
32 #include "block/qdict.h"
33 #include "qemu/error-report.h"
34 #include "block/module_block.h"
35 #include "qemu/main-loop.h"
36 #include "qemu/module.h"
37 #include "qapi/error.h"
38 #include "qapi/qmp/qdict.h"
39 #include "qapi/qmp/qjson.h"
40 #include "qapi/qmp/qnull.h"
41 #include "qapi/qmp/qstring.h"
42 #include "qapi/qobject-output-visitor.h"
43 #include "qapi/qapi-visit-block-core.h"
44 #include "sysemu/block-backend.h"
45 #include "qemu/notify.h"
46 #include "qemu/option.h"
47 #include "qemu/coroutine.h"
48 #include "block/qapi.h"
49 #include "qemu/timer.h"
50 #include "qemu/cutils.h"
52 #include "qemu/range.h"
54 #include "block/coroutines.h"
57 #include <sys/ioctl.h>
58 #include <sys/queue.h>
59 #if defined(HAVE_SYS_DISK_H)
68 #define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */
70 /* Protected by BQL */
71 static QTAILQ_HEAD(, BlockDriverState
) graph_bdrv_states
=
72 QTAILQ_HEAD_INITIALIZER(graph_bdrv_states
);
74 /* Protected by BQL */
75 static QTAILQ_HEAD(, BlockDriverState
) all_bdrv_states
=
76 QTAILQ_HEAD_INITIALIZER(all_bdrv_states
);
78 /* Protected by BQL */
79 static QLIST_HEAD(, BlockDriver
) bdrv_drivers
=
80 QLIST_HEAD_INITIALIZER(bdrv_drivers
);
82 static BlockDriverState
*bdrv_open_inherit(const char *filename
,
83 const char *reference
,
84 QDict
*options
, int flags
,
85 BlockDriverState
*parent
,
86 const BdrvChildClass
*child_class
,
87 BdrvChildRole child_role
,
90 static bool bdrv_recurse_has_child(BlockDriverState
*bs
,
91 BlockDriverState
*child
);
93 static void bdrv_replace_child_noperm(BdrvChild
*child
,
94 BlockDriverState
*new_bs
);
95 static void bdrv_remove_child(BdrvChild
*child
, Transaction
*tran
);
96 static void bdrv_remove_filter_or_cow_child(BlockDriverState
*bs
,
99 static int bdrv_reopen_prepare(BDRVReopenState
*reopen_state
,
100 BlockReopenQueue
*queue
,
101 Transaction
*change_child_tran
, Error
**errp
);
102 static void bdrv_reopen_commit(BDRVReopenState
*reopen_state
);
103 static void bdrv_reopen_abort(BDRVReopenState
*reopen_state
);
105 static bool bdrv_backing_overridden(BlockDriverState
*bs
);
107 static bool bdrv_change_aio_context(BlockDriverState
*bs
, AioContext
*ctx
,
108 GHashTable
*visited
, Transaction
*tran
,
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 Error
*local_err
= NULL
;
468 int rv
= block_module_load(block_driver_modules
[i
].library_name
,
471 return bdrv_do_find_format(format_name
);
473 error_report_err(local_err
);
481 static int bdrv_format_is_whitelisted(const char *format_name
, bool read_only
)
483 static const char *whitelist_rw
[] = {
484 CONFIG_BDRV_RW_WHITELIST
487 static const char *whitelist_ro
[] = {
488 CONFIG_BDRV_RO_WHITELIST
493 if (!whitelist_rw
[0] && !whitelist_ro
[0]) {
494 return 1; /* no whitelist, anything goes */
497 for (p
= whitelist_rw
; *p
; p
++) {
498 if (!strcmp(format_name
, *p
)) {
503 for (p
= whitelist_ro
; *p
; p
++) {
504 if (!strcmp(format_name
, *p
)) {
512 int bdrv_is_whitelisted(BlockDriver
*drv
, bool read_only
)
515 return bdrv_format_is_whitelisted(drv
->format_name
, read_only
);
518 bool bdrv_uses_whitelist(void)
520 return use_bdrv_whitelist
;
523 typedef struct CreateCo
{
531 static void coroutine_fn
bdrv_create_co_entry(void *opaque
)
533 Error
*local_err
= NULL
;
536 CreateCo
*cco
= opaque
;
540 ret
= cco
->drv
->bdrv_co_create_opts(cco
->drv
,
541 cco
->filename
, cco
->opts
, &local_err
);
542 error_propagate(&cco
->err
, local_err
);
546 int bdrv_create(BlockDriver
*drv
, const char* filename
,
547 QemuOpts
*opts
, Error
**errp
)
556 .filename
= g_strdup(filename
),
562 if (!drv
->bdrv_co_create_opts
) {
563 error_setg(errp
, "Driver '%s' does not support image creation", drv
->format_name
);
568 if (qemu_in_coroutine()) {
569 /* Fast-path if already in coroutine context */
570 bdrv_create_co_entry(&cco
);
572 co
= qemu_coroutine_create(bdrv_create_co_entry
, &cco
);
573 qemu_coroutine_enter(co
);
574 while (cco
.ret
== NOT_DONE
) {
575 aio_poll(qemu_get_aio_context(), true);
582 error_propagate(errp
, cco
.err
);
584 error_setg_errno(errp
, -ret
, "Could not create image");
589 g_free(cco
.filename
);
594 * Helper function for bdrv_create_file_fallback(): Resize @blk to at
595 * least the given @minimum_size.
597 * On success, return @blk's actual length.
598 * Otherwise, return -errno.
600 static int64_t create_file_fallback_truncate(BlockBackend
*blk
,
601 int64_t minimum_size
, Error
**errp
)
603 Error
*local_err
= NULL
;
609 ret
= blk_truncate(blk
, minimum_size
, false, PREALLOC_MODE_OFF
, 0,
611 if (ret
< 0 && ret
!= -ENOTSUP
) {
612 error_propagate(errp
, local_err
);
616 size
= blk_getlength(blk
);
618 error_free(local_err
);
619 error_setg_errno(errp
, -size
,
620 "Failed to inquire the new image file's length");
624 if (size
< minimum_size
) {
625 /* Need to grow the image, but we failed to do that */
626 error_propagate(errp
, local_err
);
630 error_free(local_err
);
637 * Helper function for bdrv_create_file_fallback(): Zero the first
638 * sector to remove any potentially pre-existing image header.
640 static int coroutine_fn
641 create_file_fallback_zero_first_sector(BlockBackend
*blk
,
642 int64_t current_size
,
645 int64_t bytes_to_clear
;
650 bytes_to_clear
= MIN(current_size
, BDRV_SECTOR_SIZE
);
651 if (bytes_to_clear
) {
652 ret
= blk_co_pwrite_zeroes(blk
, 0, bytes_to_clear
, BDRV_REQ_MAY_UNMAP
);
654 error_setg_errno(errp
, -ret
,
655 "Failed to clear the new image's first sector");
664 * Simple implementation of bdrv_co_create_opts for protocol drivers
665 * which only support creation via opening a file
666 * (usually existing raw storage device)
668 int coroutine_fn
bdrv_co_create_opts_simple(BlockDriver
*drv
,
669 const char *filename
,
677 PreallocMode prealloc
;
678 Error
*local_err
= NULL
;
683 size
= qemu_opt_get_size_del(opts
, BLOCK_OPT_SIZE
, 0);
684 buf
= qemu_opt_get_del(opts
, BLOCK_OPT_PREALLOC
);
685 prealloc
= qapi_enum_parse(&PreallocMode_lookup
, buf
,
686 PREALLOC_MODE_OFF
, &local_err
);
689 error_propagate(errp
, local_err
);
693 if (prealloc
!= PREALLOC_MODE_OFF
) {
694 error_setg(errp
, "Unsupported preallocation mode '%s'",
695 PreallocMode_str(prealloc
));
699 options
= qdict_new();
700 qdict_put_str(options
, "driver", drv
->format_name
);
702 blk
= blk_new_open(filename
, NULL
, options
,
703 BDRV_O_RDWR
| BDRV_O_RESIZE
, errp
);
705 error_prepend(errp
, "Protocol driver '%s' does not support image "
706 "creation, and opening the image failed: ",
711 size
= create_file_fallback_truncate(blk
, size
, errp
);
717 ret
= create_file_fallback_zero_first_sector(blk
, size
, errp
);
728 int bdrv_create_file(const char *filename
, QemuOpts
*opts
, Error
**errp
)
730 QemuOpts
*protocol_opts
;
737 drv
= bdrv_find_protocol(filename
, true, errp
);
742 if (!drv
->create_opts
) {
743 error_setg(errp
, "Driver '%s' does not support image creation",
749 * 'opts' contains a QemuOptsList with a combination of format and protocol
752 * The format properly removes its options, but the default values remain
753 * in 'opts->list'. So if the protocol has options with the same name
754 * (e.g. rbd has 'cluster_size' as qcow2), it will see the default values
755 * of the format, since for overlapping options, the format wins.
757 * To avoid this issue, lets convert QemuOpts to QDict, in this way we take
758 * only the set options, and then convert it back to QemuOpts, using the
759 * create_opts of the protocol. So the new QemuOpts, will contain only the
762 qdict
= qemu_opts_to_qdict(opts
, NULL
);
763 protocol_opts
= qemu_opts_from_qdict(drv
->create_opts
, qdict
, errp
);
764 if (protocol_opts
== NULL
) {
769 ret
= bdrv_create(drv
, filename
, protocol_opts
, errp
);
771 qemu_opts_del(protocol_opts
);
772 qobject_unref(qdict
);
776 int coroutine_fn
bdrv_co_delete_file(BlockDriverState
*bs
, Error
**errp
)
778 Error
*local_err
= NULL
;
785 error_setg(errp
, "Block node '%s' is not opened", bs
->filename
);
789 if (!bs
->drv
->bdrv_co_delete_file
) {
790 error_setg(errp
, "Driver '%s' does not support image deletion",
791 bs
->drv
->format_name
);
795 ret
= bs
->drv
->bdrv_co_delete_file(bs
, &local_err
);
797 error_propagate(errp
, local_err
);
803 void coroutine_fn
bdrv_co_delete_file_noerr(BlockDriverState
*bs
)
805 Error
*local_err
= NULL
;
813 ret
= bdrv_co_delete_file(bs
, &local_err
);
815 * ENOTSUP will happen if the block driver doesn't support
816 * the 'bdrv_co_delete_file' interface. This is a predictable
817 * scenario and shouldn't be reported back to the user.
819 if (ret
== -ENOTSUP
) {
820 error_free(local_err
);
821 } else if (ret
< 0) {
822 error_report_err(local_err
);
827 * Try to get @bs's logical and physical block size.
828 * On success, store them in @bsz struct and return 0.
829 * On failure return -errno.
830 * @bs must not be empty.
832 int bdrv_probe_blocksizes(BlockDriverState
*bs
, BlockSizes
*bsz
)
834 BlockDriver
*drv
= bs
->drv
;
835 BlockDriverState
*filtered
= bdrv_filter_bs(bs
);
838 if (drv
&& drv
->bdrv_probe_blocksizes
) {
839 return drv
->bdrv_probe_blocksizes(bs
, bsz
);
840 } else if (filtered
) {
841 return bdrv_probe_blocksizes(filtered
, bsz
);
848 * Try to get @bs's geometry (cyls, heads, sectors).
849 * On success, store them in @geo struct and return 0.
850 * On failure return -errno.
851 * @bs must not be empty.
853 int bdrv_probe_geometry(BlockDriverState
*bs
, HDGeometry
*geo
)
855 BlockDriver
*drv
= bs
->drv
;
856 BlockDriverState
*filtered
= bdrv_filter_bs(bs
);
859 if (drv
&& drv
->bdrv_probe_geometry
) {
860 return drv
->bdrv_probe_geometry(bs
, geo
);
861 } else if (filtered
) {
862 return bdrv_probe_geometry(filtered
, geo
);
869 * Create a uniquely-named empty temporary file.
870 * Return the actual file name used upon success, otherwise NULL.
871 * This string should be freed with g_free() when not needed any longer.
873 * Note: creating a temporary file for the caller to (re)open is
874 * inherently racy. Use g_file_open_tmp() instead whenever practical.
876 char *create_tmp_file(Error
**errp
)
880 g_autofree
char *filename
= NULL
;
882 tmpdir
= g_get_tmp_dir();
885 * See commit 69bef79 ("block: use /var/tmp instead of /tmp for -snapshot")
887 * This function is used to create temporary disk images (like -snapshot),
888 * so the files can become very large. /tmp is often a tmpfs where as
889 * /var/tmp is usually on a disk, so more appropriate for disk images.
891 if (!g_strcmp0(tmpdir
, "/tmp")) {
896 filename
= g_strdup_printf("%s/vl.XXXXXX", tmpdir
);
897 fd
= g_mkstemp(filename
);
899 error_setg_errno(errp
, errno
, "Could not open temporary file '%s'",
905 return g_steal_pointer(&filename
);
909 * Detect host devices. By convention, /dev/cdrom[N] is always
910 * recognized as a host CDROM.
912 static BlockDriver
*find_hdev_driver(const char *filename
)
914 int score_max
= 0, score
;
915 BlockDriver
*drv
= NULL
, *d
;
918 QLIST_FOREACH(d
, &bdrv_drivers
, list
) {
919 if (d
->bdrv_probe_device
) {
920 score
= d
->bdrv_probe_device(filename
);
921 if (score
> score_max
) {
931 static BlockDriver
*bdrv_do_find_protocol(const char *protocol
)
936 QLIST_FOREACH(drv1
, &bdrv_drivers
, list
) {
937 if (drv1
->protocol_name
&& !strcmp(drv1
->protocol_name
, protocol
)) {
945 BlockDriver
*bdrv_find_protocol(const char *filename
,
946 bool allow_protocol_prefix
,
956 /* TODO Drivers without bdrv_file_open must be specified explicitly */
959 * XXX(hch): we really should not let host device detection
960 * override an explicit protocol specification, but moving this
961 * later breaks access to device names with colons in them.
962 * Thanks to the brain-dead persistent naming schemes on udev-
963 * based Linux systems those actually are quite common.
965 drv1
= find_hdev_driver(filename
);
970 if (!path_has_protocol(filename
) || !allow_protocol_prefix
) {
974 p
= strchr(filename
, ':');
977 if (len
> sizeof(protocol
) - 1)
978 len
= sizeof(protocol
) - 1;
979 memcpy(protocol
, filename
, len
);
980 protocol
[len
] = '\0';
982 drv1
= bdrv_do_find_protocol(protocol
);
987 for (i
= 0; i
< (int)ARRAY_SIZE(block_driver_modules
); ++i
) {
988 if (block_driver_modules
[i
].protocol_name
&&
989 !strcmp(block_driver_modules
[i
].protocol_name
, protocol
)) {
990 int rv
= block_module_load(block_driver_modules
[i
].library_name
, errp
);
992 drv1
= bdrv_do_find_protocol(protocol
);
1001 error_setg(errp
, "Unknown protocol '%s'", protocol
);
1007 * Guess image format by probing its contents.
1008 * This is not a good idea when your image is raw (CVE-2008-2004), but
1009 * we do it anyway for backward compatibility.
1011 * @buf contains the image's first @buf_size bytes.
1012 * @buf_size is the buffer size in bytes (generally BLOCK_PROBE_BUF_SIZE,
1013 * but can be smaller if the image file is smaller)
1014 * @filename is its filename.
1016 * For all block drivers, call the bdrv_probe() method to get its
1018 * Return the first block driver with the highest probing score.
1020 BlockDriver
*bdrv_probe_all(const uint8_t *buf
, int buf_size
,
1021 const char *filename
)
1023 int score_max
= 0, score
;
1024 BlockDriver
*drv
= NULL
, *d
;
1027 QLIST_FOREACH(d
, &bdrv_drivers
, list
) {
1028 if (d
->bdrv_probe
) {
1029 score
= d
->bdrv_probe(buf
, buf_size
, filename
);
1030 if (score
> score_max
) {
1040 static int find_image_format(BlockBackend
*file
, const char *filename
,
1041 BlockDriver
**pdrv
, Error
**errp
)
1044 uint8_t buf
[BLOCK_PROBE_BUF_SIZE
];
1047 GLOBAL_STATE_CODE();
1049 /* Return the raw BlockDriver * to scsi-generic devices or empty drives */
1050 if (blk_is_sg(file
) || !blk_is_inserted(file
) || blk_getlength(file
) == 0) {
1055 ret
= blk_pread(file
, 0, sizeof(buf
), buf
, 0);
1057 error_setg_errno(errp
, -ret
, "Could not read image for determining its "
1063 drv
= bdrv_probe_all(buf
, sizeof(buf
), filename
);
1065 error_setg(errp
, "Could not determine image format: No compatible "
1076 * Set the current 'total_sectors' value
1077 * Return 0 on success, -errno on error.
1079 int refresh_total_sectors(BlockDriverState
*bs
, int64_t hint
)
1081 BlockDriver
*drv
= bs
->drv
;
1088 /* Do not attempt drv->bdrv_getlength() on scsi-generic devices */
1092 /* query actual device if possible, otherwise just trust the hint */
1093 if (drv
->bdrv_getlength
) {
1094 int64_t length
= drv
->bdrv_getlength(bs
);
1098 hint
= DIV_ROUND_UP(length
, BDRV_SECTOR_SIZE
);
1101 bs
->total_sectors
= hint
;
1103 if (bs
->total_sectors
* BDRV_SECTOR_SIZE
> BDRV_MAX_LENGTH
) {
1111 * Combines a QDict of new block driver @options with any missing options taken
1112 * from @old_options, so that leaving out an option defaults to its old value.
1114 static void bdrv_join_options(BlockDriverState
*bs
, QDict
*options
,
1117 GLOBAL_STATE_CODE();
1118 if (bs
->drv
&& bs
->drv
->bdrv_join_options
) {
1119 bs
->drv
->bdrv_join_options(options
, old_options
);
1121 qdict_join(options
, old_options
, false);
1125 static BlockdevDetectZeroesOptions
bdrv_parse_detect_zeroes(QemuOpts
*opts
,
1129 Error
*local_err
= NULL
;
1130 char *value
= qemu_opt_get_del(opts
, "detect-zeroes");
1131 BlockdevDetectZeroesOptions detect_zeroes
=
1132 qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup
, value
,
1133 BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF
, &local_err
);
1134 GLOBAL_STATE_CODE();
1137 error_propagate(errp
, local_err
);
1138 return detect_zeroes
;
1141 if (detect_zeroes
== BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP
&&
1142 !(open_flags
& BDRV_O_UNMAP
))
1144 error_setg(errp
, "setting detect-zeroes to unmap is not allowed "
1145 "without setting discard operation to unmap");
1148 return detect_zeroes
;
1152 * Set open flags for aio engine
1154 * Return 0 on success, -1 if the engine specified is invalid
1156 int bdrv_parse_aio(const char *mode
, int *flags
)
1158 if (!strcmp(mode
, "threads")) {
1159 /* do nothing, default */
1160 } else if (!strcmp(mode
, "native")) {
1161 *flags
|= BDRV_O_NATIVE_AIO
;
1162 #ifdef CONFIG_LINUX_IO_URING
1163 } else if (!strcmp(mode
, "io_uring")) {
1164 *flags
|= BDRV_O_IO_URING
;
1174 * Set open flags for a given discard mode
1176 * Return 0 on success, -1 if the discard mode was invalid.
1178 int bdrv_parse_discard_flags(const char *mode
, int *flags
)
1180 *flags
&= ~BDRV_O_UNMAP
;
1182 if (!strcmp(mode
, "off") || !strcmp(mode
, "ignore")) {
1184 } else if (!strcmp(mode
, "on") || !strcmp(mode
, "unmap")) {
1185 *flags
|= BDRV_O_UNMAP
;
1194 * Set open flags for a given cache mode
1196 * Return 0 on success, -1 if the cache mode was invalid.
1198 int bdrv_parse_cache_mode(const char *mode
, int *flags
, bool *writethrough
)
1200 *flags
&= ~BDRV_O_CACHE_MASK
;
1202 if (!strcmp(mode
, "off") || !strcmp(mode
, "none")) {
1203 *writethrough
= false;
1204 *flags
|= BDRV_O_NOCACHE
;
1205 } else if (!strcmp(mode
, "directsync")) {
1206 *writethrough
= true;
1207 *flags
|= BDRV_O_NOCACHE
;
1208 } else if (!strcmp(mode
, "writeback")) {
1209 *writethrough
= false;
1210 } else if (!strcmp(mode
, "unsafe")) {
1211 *writethrough
= false;
1212 *flags
|= BDRV_O_NO_FLUSH
;
1213 } else if (!strcmp(mode
, "writethrough")) {
1214 *writethrough
= true;
1222 static char *bdrv_child_get_parent_desc(BdrvChild
*c
)
1224 BlockDriverState
*parent
= c
->opaque
;
1225 return g_strdup_printf("node '%s'", bdrv_get_node_name(parent
));
1228 static void bdrv_child_cb_drained_begin(BdrvChild
*child
)
1230 BlockDriverState
*bs
= child
->opaque
;
1231 bdrv_do_drained_begin_quiesce(bs
, NULL
, false);
1234 static bool bdrv_child_cb_drained_poll(BdrvChild
*child
)
1236 BlockDriverState
*bs
= child
->opaque
;
1237 return bdrv_drain_poll(bs
, false, NULL
, false);
1240 static void bdrv_child_cb_drained_end(BdrvChild
*child
,
1241 int *drained_end_counter
)
1243 BlockDriverState
*bs
= child
->opaque
;
1244 bdrv_drained_end_no_poll(bs
, drained_end_counter
);
1247 static int bdrv_child_cb_inactivate(BdrvChild
*child
)
1249 BlockDriverState
*bs
= child
->opaque
;
1250 GLOBAL_STATE_CODE();
1251 assert(bs
->open_flags
& BDRV_O_INACTIVE
);
1255 static bool bdrv_child_cb_change_aio_ctx(BdrvChild
*child
, AioContext
*ctx
,
1256 GHashTable
*visited
, Transaction
*tran
,
1259 BlockDriverState
*bs
= child
->opaque
;
1260 return bdrv_change_aio_context(bs
, ctx
, visited
, tran
, errp
);
1264 * Returns the options and flags that a temporary snapshot should get, based on
1265 * the originally requested flags (the originally requested image will have
1266 * flags like a backing file)
1268 static void bdrv_temp_snapshot_options(int *child_flags
, QDict
*child_options
,
1269 int parent_flags
, QDict
*parent_options
)
1271 GLOBAL_STATE_CODE();
1272 *child_flags
= (parent_flags
& ~BDRV_O_SNAPSHOT
) | BDRV_O_TEMPORARY
;
1274 /* For temporary files, unconditional cache=unsafe is fine */
1275 qdict_set_default_str(child_options
, BDRV_OPT_CACHE_DIRECT
, "off");
1276 qdict_set_default_str(child_options
, BDRV_OPT_CACHE_NO_FLUSH
, "on");
1278 /* Copy the read-only and discard options from the parent */
1279 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_READ_ONLY
);
1280 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_DISCARD
);
1282 /* aio=native doesn't work for cache.direct=off, so disable it for the
1283 * temporary snapshot */
1284 *child_flags
&= ~BDRV_O_NATIVE_AIO
;
1287 static void bdrv_backing_attach(BdrvChild
*c
)
1289 BlockDriverState
*parent
= c
->opaque
;
1290 BlockDriverState
*backing_hd
= c
->bs
;
1292 GLOBAL_STATE_CODE();
1293 assert(!parent
->backing_blocker
);
1294 error_setg(&parent
->backing_blocker
,
1295 "node is used as backing hd of '%s'",
1296 bdrv_get_device_or_node_name(parent
));
1298 bdrv_refresh_filename(backing_hd
);
1300 parent
->open_flags
&= ~BDRV_O_NO_BACKING
;
1302 bdrv_op_block_all(backing_hd
, parent
->backing_blocker
);
1303 /* Otherwise we won't be able to commit or stream */
1304 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_COMMIT_TARGET
,
1305 parent
->backing_blocker
);
1306 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_STREAM
,
1307 parent
->backing_blocker
);
1309 * We do backup in 3 ways:
1311 * The target bs is new opened, and the source is top BDS
1312 * 2. blockdev backup
1313 * Both the source and the target are top BDSes.
1314 * 3. internal backup(used for block replication)
1315 * Both the source and the target are backing file
1317 * In case 1 and 2, neither the source nor the target is the backing file.
1318 * In case 3, we will block the top BDS, so there is only one block job
1319 * for the top BDS and its backing chain.
1321 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_BACKUP_SOURCE
,
1322 parent
->backing_blocker
);
1323 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_BACKUP_TARGET
,
1324 parent
->backing_blocker
);
1327 static void bdrv_backing_detach(BdrvChild
*c
)
1329 BlockDriverState
*parent
= c
->opaque
;
1331 GLOBAL_STATE_CODE();
1332 assert(parent
->backing_blocker
);
1333 bdrv_op_unblock_all(c
->bs
, parent
->backing_blocker
);
1334 error_free(parent
->backing_blocker
);
1335 parent
->backing_blocker
= NULL
;
1338 static int bdrv_backing_update_filename(BdrvChild
*c
, BlockDriverState
*base
,
1339 const char *filename
, Error
**errp
)
1341 BlockDriverState
*parent
= c
->opaque
;
1342 bool read_only
= bdrv_is_read_only(parent
);
1344 GLOBAL_STATE_CODE();
1347 ret
= bdrv_reopen_set_read_only(parent
, false, errp
);
1353 ret
= bdrv_change_backing_file(parent
, filename
,
1354 base
->drv
? base
->drv
->format_name
: "",
1357 error_setg_errno(errp
, -ret
, "Could not update backing file link");
1361 bdrv_reopen_set_read_only(parent
, true, NULL
);
1368 * Returns the options and flags that a generic child of a BDS should
1369 * get, based on the given options and flags for the parent BDS.
1371 static void bdrv_inherited_options(BdrvChildRole role
, bool parent_is_format
,
1372 int *child_flags
, QDict
*child_options
,
1373 int parent_flags
, QDict
*parent_options
)
1375 int flags
= parent_flags
;
1376 GLOBAL_STATE_CODE();
1379 * First, decide whether to set, clear, or leave BDRV_O_PROTOCOL.
1380 * Generally, the question to answer is: Should this child be
1381 * format-probed by default?
1385 * Pure and non-filtered data children of non-format nodes should
1386 * be probed by default (even when the node itself has BDRV_O_PROTOCOL
1387 * set). This only affects a very limited set of drivers (namely
1388 * quorum and blkverify when this comment was written).
1389 * Force-clear BDRV_O_PROTOCOL then.
1391 if (!parent_is_format
&&
1392 (role
& BDRV_CHILD_DATA
) &&
1393 !(role
& (BDRV_CHILD_METADATA
| BDRV_CHILD_FILTERED
)))
1395 flags
&= ~BDRV_O_PROTOCOL
;
1399 * All children of format nodes (except for COW children) and all
1400 * metadata children in general should never be format-probed.
1401 * Force-set BDRV_O_PROTOCOL then.
1403 if ((parent_is_format
&& !(role
& BDRV_CHILD_COW
)) ||
1404 (role
& BDRV_CHILD_METADATA
))
1406 flags
|= BDRV_O_PROTOCOL
;
1410 * If the cache mode isn't explicitly set, inherit direct and no-flush from
1413 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_CACHE_DIRECT
);
1414 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_CACHE_NO_FLUSH
);
1415 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_FORCE_SHARE
);
1417 if (role
& BDRV_CHILD_COW
) {
1418 /* backing files are opened read-only by default */
1419 qdict_set_default_str(child_options
, BDRV_OPT_READ_ONLY
, "on");
1420 qdict_set_default_str(child_options
, BDRV_OPT_AUTO_READ_ONLY
, "off");
1422 /* Inherit the read-only option from the parent if it's not set */
1423 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_READ_ONLY
);
1424 qdict_copy_default(child_options
, parent_options
,
1425 BDRV_OPT_AUTO_READ_ONLY
);
1429 * bdrv_co_pdiscard() respects unmap policy for the parent, so we
1430 * can default to enable it on lower layers regardless of the
1433 qdict_set_default_str(child_options
, BDRV_OPT_DISCARD
, "unmap");
1435 /* Clear flags that only apply to the top layer */
1436 flags
&= ~(BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
| BDRV_O_COPY_ON_READ
);
1438 if (role
& BDRV_CHILD_METADATA
) {
1439 flags
&= ~BDRV_O_NO_IO
;
1441 if (role
& BDRV_CHILD_COW
) {
1442 flags
&= ~BDRV_O_TEMPORARY
;
1445 *child_flags
= flags
;
1448 static void bdrv_child_cb_attach(BdrvChild
*child
)
1450 BlockDriverState
*bs
= child
->opaque
;
1452 assert_bdrv_graph_writable(bs
);
1453 QLIST_INSERT_HEAD(&bs
->children
, child
, next
);
1454 if (bs
->drv
->is_filter
|| (child
->role
& BDRV_CHILD_FILTERED
)) {
1456 * Here we handle filters and block/raw-format.c when it behave like
1457 * filter. They generally have a single PRIMARY child, which is also the
1458 * FILTERED child, and that they may have multiple more children, which
1459 * are neither PRIMARY nor FILTERED. And never we have a COW child here.
1460 * So bs->file will be the PRIMARY child, unless the PRIMARY child goes
1461 * into bs->backing on exceptional cases; and bs->backing will be
1464 assert(!(child
->role
& BDRV_CHILD_COW
));
1465 if (child
->role
& BDRV_CHILD_PRIMARY
) {
1466 assert(child
->role
& BDRV_CHILD_FILTERED
);
1467 assert(!bs
->backing
);
1470 if (bs
->drv
->filtered_child_is_backing
) {
1471 bs
->backing
= child
;
1476 assert(!(child
->role
& BDRV_CHILD_FILTERED
));
1478 } else if (child
->role
& BDRV_CHILD_COW
) {
1479 assert(bs
->drv
->supports_backing
);
1480 assert(!(child
->role
& BDRV_CHILD_PRIMARY
));
1481 assert(!bs
->backing
);
1482 bs
->backing
= child
;
1483 bdrv_backing_attach(child
);
1484 } else if (child
->role
& BDRV_CHILD_PRIMARY
) {
1489 bdrv_apply_subtree_drain(child
, bs
);
1492 static void bdrv_child_cb_detach(BdrvChild
*child
)
1494 BlockDriverState
*bs
= child
->opaque
;
1496 if (child
->role
& BDRV_CHILD_COW
) {
1497 bdrv_backing_detach(child
);
1500 bdrv_unapply_subtree_drain(child
, bs
);
1502 assert_bdrv_graph_writable(bs
);
1503 QLIST_REMOVE(child
, next
);
1504 if (child
== bs
->backing
) {
1505 assert(child
!= bs
->file
);
1507 } else if (child
== bs
->file
) {
1512 static int bdrv_child_cb_update_filename(BdrvChild
*c
, BlockDriverState
*base
,
1513 const char *filename
, Error
**errp
)
1515 if (c
->role
& BDRV_CHILD_COW
) {
1516 return bdrv_backing_update_filename(c
, base
, filename
, errp
);
1521 AioContext
*child_of_bds_get_parent_aio_context(BdrvChild
*c
)
1523 BlockDriverState
*bs
= c
->opaque
;
1526 return bdrv_get_aio_context(bs
);
1529 const BdrvChildClass child_of_bds
= {
1530 .parent_is_bds
= true,
1531 .get_parent_desc
= bdrv_child_get_parent_desc
,
1532 .inherit_options
= bdrv_inherited_options
,
1533 .drained_begin
= bdrv_child_cb_drained_begin
,
1534 .drained_poll
= bdrv_child_cb_drained_poll
,
1535 .drained_end
= bdrv_child_cb_drained_end
,
1536 .attach
= bdrv_child_cb_attach
,
1537 .detach
= bdrv_child_cb_detach
,
1538 .inactivate
= bdrv_child_cb_inactivate
,
1539 .change_aio_ctx
= bdrv_child_cb_change_aio_ctx
,
1540 .update_filename
= bdrv_child_cb_update_filename
,
1541 .get_parent_aio_context
= child_of_bds_get_parent_aio_context
,
1544 AioContext
*bdrv_child_get_parent_aio_context(BdrvChild
*c
)
1546 GLOBAL_STATE_CODE();
1547 return c
->klass
->get_parent_aio_context(c
);
1550 static int bdrv_open_flags(BlockDriverState
*bs
, int flags
)
1552 int open_flags
= flags
;
1553 GLOBAL_STATE_CODE();
1556 * Clear flags that are internal to the block layer before opening the
1559 open_flags
&= ~(BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
| BDRV_O_PROTOCOL
);
1564 static void update_flags_from_options(int *flags
, QemuOpts
*opts
)
1566 GLOBAL_STATE_CODE();
1568 *flags
&= ~(BDRV_O_CACHE_MASK
| BDRV_O_RDWR
| BDRV_O_AUTO_RDONLY
);
1570 if (qemu_opt_get_bool_del(opts
, BDRV_OPT_CACHE_NO_FLUSH
, false)) {
1571 *flags
|= BDRV_O_NO_FLUSH
;
1574 if (qemu_opt_get_bool_del(opts
, BDRV_OPT_CACHE_DIRECT
, false)) {
1575 *flags
|= BDRV_O_NOCACHE
;
1578 if (!qemu_opt_get_bool_del(opts
, BDRV_OPT_READ_ONLY
, false)) {
1579 *flags
|= BDRV_O_RDWR
;
1582 if (qemu_opt_get_bool_del(opts
, BDRV_OPT_AUTO_READ_ONLY
, false)) {
1583 *flags
|= BDRV_O_AUTO_RDONLY
;
1587 static void update_options_from_flags(QDict
*options
, int flags
)
1589 GLOBAL_STATE_CODE();
1590 if (!qdict_haskey(options
, BDRV_OPT_CACHE_DIRECT
)) {
1591 qdict_put_bool(options
, BDRV_OPT_CACHE_DIRECT
, flags
& BDRV_O_NOCACHE
);
1593 if (!qdict_haskey(options
, BDRV_OPT_CACHE_NO_FLUSH
)) {
1594 qdict_put_bool(options
, BDRV_OPT_CACHE_NO_FLUSH
,
1595 flags
& BDRV_O_NO_FLUSH
);
1597 if (!qdict_haskey(options
, BDRV_OPT_READ_ONLY
)) {
1598 qdict_put_bool(options
, BDRV_OPT_READ_ONLY
, !(flags
& BDRV_O_RDWR
));
1600 if (!qdict_haskey(options
, BDRV_OPT_AUTO_READ_ONLY
)) {
1601 qdict_put_bool(options
, BDRV_OPT_AUTO_READ_ONLY
,
1602 flags
& BDRV_O_AUTO_RDONLY
);
1606 static void bdrv_assign_node_name(BlockDriverState
*bs
,
1607 const char *node_name
,
1610 char *gen_node_name
= NULL
;
1611 GLOBAL_STATE_CODE();
1614 node_name
= gen_node_name
= id_generate(ID_BLOCK
);
1615 } else if (!id_wellformed(node_name
)) {
1617 * Check for empty string or invalid characters, but not if it is
1618 * generated (generated names use characters not available to the user)
1620 error_setg(errp
, "Invalid node-name: '%s'", node_name
);
1624 /* takes care of avoiding namespaces collisions */
1625 if (blk_by_name(node_name
)) {
1626 error_setg(errp
, "node-name=%s is conflicting with a device id",
1631 /* takes care of avoiding duplicates node names */
1632 if (bdrv_find_node(node_name
)) {
1633 error_setg(errp
, "Duplicate nodes with node-name='%s'", node_name
);
1637 /* Make sure that the node name isn't truncated */
1638 if (strlen(node_name
) >= sizeof(bs
->node_name
)) {
1639 error_setg(errp
, "Node name too long");
1643 /* copy node name into the bs and insert it into the graph list */
1644 pstrcpy(bs
->node_name
, sizeof(bs
->node_name
), node_name
);
1645 QTAILQ_INSERT_TAIL(&graph_bdrv_states
, bs
, node_list
);
1647 g_free(gen_node_name
);
1650 static int bdrv_open_driver(BlockDriverState
*bs
, BlockDriver
*drv
,
1651 const char *node_name
, QDict
*options
,
1652 int open_flags
, Error
**errp
)
1654 Error
*local_err
= NULL
;
1656 GLOBAL_STATE_CODE();
1658 bdrv_assign_node_name(bs
, node_name
, &local_err
);
1660 error_propagate(errp
, local_err
);
1665 bs
->opaque
= g_malloc0(drv
->instance_size
);
1667 if (drv
->bdrv_file_open
) {
1668 assert(!drv
->bdrv_needs_filename
|| bs
->filename
[0]);
1669 ret
= drv
->bdrv_file_open(bs
, options
, open_flags
, &local_err
);
1670 } else if (drv
->bdrv_open
) {
1671 ret
= drv
->bdrv_open(bs
, options
, open_flags
, &local_err
);
1678 error_propagate(errp
, local_err
);
1679 } else if (bs
->filename
[0]) {
1680 error_setg_errno(errp
, -ret
, "Could not open '%s'", bs
->filename
);
1682 error_setg_errno(errp
, -ret
, "Could not open image");
1687 assert(!(bs
->supported_read_flags
& ~BDRV_REQ_MASK
));
1688 assert(!(bs
->supported_write_flags
& ~BDRV_REQ_MASK
));
1691 * Always allow the BDRV_REQ_REGISTERED_BUF optimization hint. This saves
1692 * drivers that pass read/write requests through to a child the trouble of
1693 * declaring support explicitly.
1695 * Drivers must not propagate this flag accidentally when they initiate I/O
1696 * to a bounce buffer. That case should be rare though.
1698 bs
->supported_read_flags
|= BDRV_REQ_REGISTERED_BUF
;
1699 bs
->supported_write_flags
|= BDRV_REQ_REGISTERED_BUF
;
1701 ret
= refresh_total_sectors(bs
, bs
->total_sectors
);
1703 error_setg_errno(errp
, -ret
, "Could not refresh total sector count");
1707 bdrv_refresh_limits(bs
, NULL
, &local_err
);
1709 error_propagate(errp
, local_err
);
1713 assert(bdrv_opt_mem_align(bs
) != 0);
1714 assert(bdrv_min_mem_align(bs
) != 0);
1715 assert(is_power_of_2(bs
->bl
.request_alignment
));
1717 for (i
= 0; i
< bs
->quiesce_counter
; i
++) {
1718 if (drv
->bdrv_co_drain_begin
) {
1719 drv
->bdrv_co_drain_begin(bs
);
1726 if (bs
->file
!= NULL
) {
1727 bdrv_unref_child(bs
, bs
->file
);
1736 * Create and open a block node.
1738 * @options is a QDict of options to pass to the block drivers, or NULL for an
1739 * empty set of options. The reference to the QDict belongs to the block layer
1740 * after the call (even on failure), so if the caller intends to reuse the
1741 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
1743 BlockDriverState
*bdrv_new_open_driver_opts(BlockDriver
*drv
,
1744 const char *node_name
,
1745 QDict
*options
, int flags
,
1748 BlockDriverState
*bs
;
1751 GLOBAL_STATE_CODE();
1754 bs
->open_flags
= flags
;
1755 bs
->options
= options
?: qdict_new();
1756 bs
->explicit_options
= qdict_clone_shallow(bs
->options
);
1759 update_options_from_flags(bs
->options
, flags
);
1761 ret
= bdrv_open_driver(bs
, drv
, node_name
, bs
->options
, flags
, errp
);
1763 qobject_unref(bs
->explicit_options
);
1764 bs
->explicit_options
= NULL
;
1765 qobject_unref(bs
->options
);
1774 /* Create and open a block node. */
1775 BlockDriverState
*bdrv_new_open_driver(BlockDriver
*drv
, const char *node_name
,
1776 int flags
, Error
**errp
)
1778 GLOBAL_STATE_CODE();
1779 return bdrv_new_open_driver_opts(drv
, node_name
, NULL
, flags
, errp
);
1782 QemuOptsList bdrv_runtime_opts
= {
1783 .name
= "bdrv_common",
1784 .head
= QTAILQ_HEAD_INITIALIZER(bdrv_runtime_opts
.head
),
1787 .name
= "node-name",
1788 .type
= QEMU_OPT_STRING
,
1789 .help
= "Node name of the block device node",
1793 .type
= QEMU_OPT_STRING
,
1794 .help
= "Block driver to use for the node",
1797 .name
= BDRV_OPT_CACHE_DIRECT
,
1798 .type
= QEMU_OPT_BOOL
,
1799 .help
= "Bypass software writeback cache on the host",
1802 .name
= BDRV_OPT_CACHE_NO_FLUSH
,
1803 .type
= QEMU_OPT_BOOL
,
1804 .help
= "Ignore flush requests",
1807 .name
= BDRV_OPT_READ_ONLY
,
1808 .type
= QEMU_OPT_BOOL
,
1809 .help
= "Node is opened in read-only mode",
1812 .name
= BDRV_OPT_AUTO_READ_ONLY
,
1813 .type
= QEMU_OPT_BOOL
,
1814 .help
= "Node can become read-only if opening read-write fails",
1817 .name
= "detect-zeroes",
1818 .type
= QEMU_OPT_STRING
,
1819 .help
= "try to optimize zero writes (off, on, unmap)",
1822 .name
= BDRV_OPT_DISCARD
,
1823 .type
= QEMU_OPT_STRING
,
1824 .help
= "discard operation (ignore/off, unmap/on)",
1827 .name
= BDRV_OPT_FORCE_SHARE
,
1828 .type
= QEMU_OPT_BOOL
,
1829 .help
= "always accept other writers (default: off)",
1831 { /* end of list */ }
1835 QemuOptsList bdrv_create_opts_simple
= {
1836 .name
= "simple-create-opts",
1837 .head
= QTAILQ_HEAD_INITIALIZER(bdrv_create_opts_simple
.head
),
1840 .name
= BLOCK_OPT_SIZE
,
1841 .type
= QEMU_OPT_SIZE
,
1842 .help
= "Virtual disk size"
1845 .name
= BLOCK_OPT_PREALLOC
,
1846 .type
= QEMU_OPT_STRING
,
1847 .help
= "Preallocation mode (allowed values: off)"
1849 { /* end of list */ }
1854 * Common part for opening disk images and files
1856 * Removes all processed options from *options.
1858 static int bdrv_open_common(BlockDriverState
*bs
, BlockBackend
*file
,
1859 QDict
*options
, Error
**errp
)
1861 int ret
, open_flags
;
1862 const char *filename
;
1863 const char *driver_name
= NULL
;
1864 const char *node_name
= NULL
;
1865 const char *discard
;
1868 Error
*local_err
= NULL
;
1871 assert(bs
->file
== NULL
);
1872 assert(options
!= NULL
&& bs
->options
!= options
);
1873 GLOBAL_STATE_CODE();
1875 opts
= qemu_opts_create(&bdrv_runtime_opts
, NULL
, 0, &error_abort
);
1876 if (!qemu_opts_absorb_qdict(opts
, options
, errp
)) {
1881 update_flags_from_options(&bs
->open_flags
, opts
);
1883 driver_name
= qemu_opt_get(opts
, "driver");
1884 drv
= bdrv_find_format(driver_name
);
1885 assert(drv
!= NULL
);
1887 bs
->force_share
= qemu_opt_get_bool(opts
, BDRV_OPT_FORCE_SHARE
, false);
1889 if (bs
->force_share
&& (bs
->open_flags
& BDRV_O_RDWR
)) {
1891 BDRV_OPT_FORCE_SHARE
1892 "=on can only be used with read-only images");
1898 bdrv_refresh_filename(blk_bs(file
));
1899 filename
= blk_bs(file
)->filename
;
1902 * Caution: while qdict_get_try_str() is fine, getting
1903 * non-string types would require more care. When @options
1904 * come from -blockdev or blockdev_add, its members are typed
1905 * according to the QAPI schema, but when they come from
1906 * -drive, they're all QString.
1908 filename
= qdict_get_try_str(options
, "filename");
1911 if (drv
->bdrv_needs_filename
&& (!filename
|| !filename
[0])) {
1912 error_setg(errp
, "The '%s' block driver requires a file name",
1918 trace_bdrv_open_common(bs
, filename
?: "", bs
->open_flags
,
1921 ro
= bdrv_is_read_only(bs
);
1923 if (use_bdrv_whitelist
&& !bdrv_is_whitelisted(drv
, ro
)) {
1924 if (!ro
&& bdrv_is_whitelisted(drv
, true)) {
1925 ret
= bdrv_apply_auto_read_only(bs
, NULL
, NULL
);
1931 !ro
&& bdrv_is_whitelisted(drv
, true)
1932 ? "Driver '%s' can only be used for read-only devices"
1933 : "Driver '%s' is not whitelisted",
1939 /* bdrv_new() and bdrv_close() make it so */
1940 assert(qatomic_read(&bs
->copy_on_read
) == 0);
1942 if (bs
->open_flags
& BDRV_O_COPY_ON_READ
) {
1944 bdrv_enable_copy_on_read(bs
);
1946 error_setg(errp
, "Can't use copy-on-read on read-only device");
1952 discard
= qemu_opt_get(opts
, BDRV_OPT_DISCARD
);
1953 if (discard
!= NULL
) {
1954 if (bdrv_parse_discard_flags(discard
, &bs
->open_flags
) != 0) {
1955 error_setg(errp
, "Invalid discard option");
1962 bdrv_parse_detect_zeroes(opts
, bs
->open_flags
, &local_err
);
1964 error_propagate(errp
, local_err
);
1969 if (filename
!= NULL
) {
1970 pstrcpy(bs
->filename
, sizeof(bs
->filename
), filename
);
1972 bs
->filename
[0] = '\0';
1974 pstrcpy(bs
->exact_filename
, sizeof(bs
->exact_filename
), bs
->filename
);
1976 /* Open the image, either directly or using a protocol */
1977 open_flags
= bdrv_open_flags(bs
, bs
->open_flags
);
1978 node_name
= qemu_opt_get(opts
, "node-name");
1980 assert(!drv
->bdrv_file_open
|| file
== NULL
);
1981 ret
= bdrv_open_driver(bs
, drv
, node_name
, options
, open_flags
, errp
);
1986 qemu_opts_del(opts
);
1990 qemu_opts_del(opts
);
1994 static G_GNUC_PRINTF(1, 0)
1995 QDict
*parse_json_filename(const char *filename
, Error
**errp
)
1997 QObject
*options_obj
;
2000 GLOBAL_STATE_CODE();
2002 ret
= strstart(filename
, "json:", &filename
);
2005 options_obj
= qobject_from_json(filename
, errp
);
2007 error_prepend(errp
, "Could not parse the JSON options: ");
2011 options
= qobject_to(QDict
, options_obj
);
2013 qobject_unref(options_obj
);
2014 error_setg(errp
, "Invalid JSON object given");
2018 qdict_flatten(options
);
2023 static void parse_json_protocol(QDict
*options
, const char **pfilename
,
2026 QDict
*json_options
;
2027 Error
*local_err
= NULL
;
2028 GLOBAL_STATE_CODE();
2030 /* Parse json: pseudo-protocol */
2031 if (!*pfilename
|| !g_str_has_prefix(*pfilename
, "json:")) {
2035 json_options
= parse_json_filename(*pfilename
, &local_err
);
2037 error_propagate(errp
, local_err
);
2041 /* Options given in the filename have lower priority than options
2042 * specified directly */
2043 qdict_join(options
, json_options
, false);
2044 qobject_unref(json_options
);
2049 * Fills in default options for opening images and converts the legacy
2050 * filename/flags pair to option QDict entries.
2051 * The BDRV_O_PROTOCOL flag in *flags will be set or cleared accordingly if a
2052 * block driver has been specified explicitly.
2054 static int bdrv_fill_options(QDict
**options
, const char *filename
,
2055 int *flags
, Error
**errp
)
2057 const char *drvname
;
2058 bool protocol
= *flags
& BDRV_O_PROTOCOL
;
2059 bool parse_filename
= false;
2060 BlockDriver
*drv
= NULL
;
2061 Error
*local_err
= NULL
;
2063 GLOBAL_STATE_CODE();
2066 * Caution: while qdict_get_try_str() is fine, getting non-string
2067 * types would require more care. When @options come from
2068 * -blockdev or blockdev_add, its members are typed according to
2069 * the QAPI schema, but when they come from -drive, they're all
2072 drvname
= qdict_get_try_str(*options
, "driver");
2074 drv
= bdrv_find_format(drvname
);
2076 error_setg(errp
, "Unknown driver '%s'", drvname
);
2079 /* If the user has explicitly specified the driver, this choice should
2080 * override the BDRV_O_PROTOCOL flag */
2081 protocol
= drv
->bdrv_file_open
;
2085 *flags
|= BDRV_O_PROTOCOL
;
2087 *flags
&= ~BDRV_O_PROTOCOL
;
2090 /* Translate cache options from flags into options */
2091 update_options_from_flags(*options
, *flags
);
2093 /* Fetch the file name from the options QDict if necessary */
2094 if (protocol
&& filename
) {
2095 if (!qdict_haskey(*options
, "filename")) {
2096 qdict_put_str(*options
, "filename", filename
);
2097 parse_filename
= true;
2099 error_setg(errp
, "Can't specify 'file' and 'filename' options at "
2105 /* Find the right block driver */
2106 /* See cautionary note on accessing @options above */
2107 filename
= qdict_get_try_str(*options
, "filename");
2109 if (!drvname
&& protocol
) {
2111 drv
= bdrv_find_protocol(filename
, parse_filename
, errp
);
2116 drvname
= drv
->format_name
;
2117 qdict_put_str(*options
, "driver", drvname
);
2119 error_setg(errp
, "Must specify either driver or file");
2124 assert(drv
|| !protocol
);
2126 /* Driver-specific filename parsing */
2127 if (drv
&& drv
->bdrv_parse_filename
&& parse_filename
) {
2128 drv
->bdrv_parse_filename(filename
, *options
, &local_err
);
2130 error_propagate(errp
, local_err
);
2134 if (!drv
->bdrv_needs_filename
) {
2135 qdict_del(*options
, "filename");
2142 typedef struct BlockReopenQueueEntry
{
2145 BDRVReopenState state
;
2146 QTAILQ_ENTRY(BlockReopenQueueEntry
) entry
;
2147 } BlockReopenQueueEntry
;
2150 * Return the flags that @bs will have after the reopens in @q have
2151 * successfully completed. If @q is NULL (or @bs is not contained in @q),
2152 * return the current flags.
2154 static int bdrv_reopen_get_flags(BlockReopenQueue
*q
, BlockDriverState
*bs
)
2156 BlockReopenQueueEntry
*entry
;
2159 QTAILQ_FOREACH(entry
, q
, entry
) {
2160 if (entry
->state
.bs
== bs
) {
2161 return entry
->state
.flags
;
2166 return bs
->open_flags
;
2169 /* Returns whether the image file can be written to after the reopen queue @q
2170 * has been successfully applied, or right now if @q is NULL. */
2171 static bool bdrv_is_writable_after_reopen(BlockDriverState
*bs
,
2172 BlockReopenQueue
*q
)
2174 int flags
= bdrv_reopen_get_flags(q
, bs
);
2176 return (flags
& (BDRV_O_RDWR
| BDRV_O_INACTIVE
)) == BDRV_O_RDWR
;
2180 * Return whether the BDS can be written to. This is not necessarily
2181 * the same as !bdrv_is_read_only(bs), as inactivated images may not
2182 * be written to but do not count as read-only images.
2184 bool bdrv_is_writable(BlockDriverState
*bs
)
2187 return bdrv_is_writable_after_reopen(bs
, NULL
);
2190 static char *bdrv_child_user_desc(BdrvChild
*c
)
2192 GLOBAL_STATE_CODE();
2193 return c
->klass
->get_parent_desc(c
);
2197 * Check that @a allows everything that @b needs. @a and @b must reference same
2200 static bool bdrv_a_allow_b(BdrvChild
*a
, BdrvChild
*b
, Error
**errp
)
2202 const char *child_bs_name
;
2203 g_autofree
char *a_user
= NULL
;
2204 g_autofree
char *b_user
= NULL
;
2205 g_autofree
char *perms
= NULL
;
2208 assert(a
->bs
== b
->bs
);
2209 GLOBAL_STATE_CODE();
2211 if ((b
->perm
& a
->shared_perm
) == b
->perm
) {
2215 child_bs_name
= bdrv_get_node_name(b
->bs
);
2216 a_user
= bdrv_child_user_desc(a
);
2217 b_user
= bdrv_child_user_desc(b
);
2218 perms
= bdrv_perm_names(b
->perm
& ~a
->shared_perm
);
2220 error_setg(errp
, "Permission conflict on node '%s': permissions '%s' are "
2221 "both required by %s (uses node '%s' as '%s' child) and "
2222 "unshared by %s (uses node '%s' as '%s' child).",
2223 child_bs_name
, perms
,
2224 b_user
, child_bs_name
, b
->name
,
2225 a_user
, child_bs_name
, a
->name
);
2230 static bool bdrv_parent_perms_conflict(BlockDriverState
*bs
, Error
**errp
)
2233 GLOBAL_STATE_CODE();
2236 * During the loop we'll look at each pair twice. That's correct because
2237 * bdrv_a_allow_b() is asymmetric and we should check each pair in both
2240 QLIST_FOREACH(a
, &bs
->parents
, next_parent
) {
2241 QLIST_FOREACH(b
, &bs
->parents
, next_parent
) {
2246 if (!bdrv_a_allow_b(a
, b
, errp
)) {
2255 static void bdrv_child_perm(BlockDriverState
*bs
, BlockDriverState
*child_bs
,
2256 BdrvChild
*c
, BdrvChildRole role
,
2257 BlockReopenQueue
*reopen_queue
,
2258 uint64_t parent_perm
, uint64_t parent_shared
,
2259 uint64_t *nperm
, uint64_t *nshared
)
2261 assert(bs
->drv
&& bs
->drv
->bdrv_child_perm
);
2262 GLOBAL_STATE_CODE();
2263 bs
->drv
->bdrv_child_perm(bs
, c
, role
, reopen_queue
,
2264 parent_perm
, parent_shared
,
2266 /* TODO Take force_share from reopen_queue */
2267 if (child_bs
&& child_bs
->force_share
) {
2268 *nshared
= BLK_PERM_ALL
;
2273 * Adds the whole subtree of @bs (including @bs itself) to the @list (except for
2274 * nodes that are already in the @list, of course) so that final list is
2275 * topologically sorted. Return the result (GSList @list object is updated, so
2276 * don't use old reference after function call).
2278 * On function start @list must be already topologically sorted and for any node
2279 * in the @list the whole subtree of the node must be in the @list as well. The
2280 * simplest way to satisfy this criteria: use only result of
2281 * bdrv_topological_dfs() or NULL as @list parameter.
2283 static GSList
*bdrv_topological_dfs(GSList
*list
, GHashTable
*found
,
2284 BlockDriverState
*bs
)
2287 g_autoptr(GHashTable
) local_found
= NULL
;
2289 GLOBAL_STATE_CODE();
2293 found
= local_found
= g_hash_table_new(NULL
, NULL
);
2296 if (g_hash_table_contains(found
, bs
)) {
2299 g_hash_table_add(found
, bs
);
2301 QLIST_FOREACH(child
, &bs
->children
, next
) {
2302 list
= bdrv_topological_dfs(list
, found
, child
->bs
);
2305 return g_slist_prepend(list
, bs
);
2308 typedef struct BdrvChildSetPermState
{
2311 uint64_t old_shared_perm
;
2312 } BdrvChildSetPermState
;
2314 static void bdrv_child_set_perm_abort(void *opaque
)
2316 BdrvChildSetPermState
*s
= opaque
;
2318 GLOBAL_STATE_CODE();
2320 s
->child
->perm
= s
->old_perm
;
2321 s
->child
->shared_perm
= s
->old_shared_perm
;
2324 static TransactionActionDrv bdrv_child_set_pem_drv
= {
2325 .abort
= bdrv_child_set_perm_abort
,
2329 static void bdrv_child_set_perm(BdrvChild
*c
, uint64_t perm
,
2330 uint64_t shared
, Transaction
*tran
)
2332 BdrvChildSetPermState
*s
= g_new(BdrvChildSetPermState
, 1);
2333 GLOBAL_STATE_CODE();
2335 *s
= (BdrvChildSetPermState
) {
2337 .old_perm
= c
->perm
,
2338 .old_shared_perm
= c
->shared_perm
,
2342 c
->shared_perm
= shared
;
2344 tran_add(tran
, &bdrv_child_set_pem_drv
, s
);
2347 static void bdrv_drv_set_perm_commit(void *opaque
)
2349 BlockDriverState
*bs
= opaque
;
2350 uint64_t cumulative_perms
, cumulative_shared_perms
;
2351 GLOBAL_STATE_CODE();
2353 if (bs
->drv
->bdrv_set_perm
) {
2354 bdrv_get_cumulative_perm(bs
, &cumulative_perms
,
2355 &cumulative_shared_perms
);
2356 bs
->drv
->bdrv_set_perm(bs
, cumulative_perms
, cumulative_shared_perms
);
2360 static void bdrv_drv_set_perm_abort(void *opaque
)
2362 BlockDriverState
*bs
= opaque
;
2363 GLOBAL_STATE_CODE();
2365 if (bs
->drv
->bdrv_abort_perm_update
) {
2366 bs
->drv
->bdrv_abort_perm_update(bs
);
2370 TransactionActionDrv bdrv_drv_set_perm_drv
= {
2371 .abort
= bdrv_drv_set_perm_abort
,
2372 .commit
= bdrv_drv_set_perm_commit
,
2375 static int bdrv_drv_set_perm(BlockDriverState
*bs
, uint64_t perm
,
2376 uint64_t shared_perm
, Transaction
*tran
,
2379 GLOBAL_STATE_CODE();
2384 if (bs
->drv
->bdrv_check_perm
) {
2385 int ret
= bs
->drv
->bdrv_check_perm(bs
, perm
, shared_perm
, errp
);
2392 tran_add(tran
, &bdrv_drv_set_perm_drv
, bs
);
2398 typedef struct BdrvReplaceChildState
{
2400 BlockDriverState
*old_bs
;
2401 } BdrvReplaceChildState
;
2403 static void bdrv_replace_child_commit(void *opaque
)
2405 BdrvReplaceChildState
*s
= opaque
;
2406 GLOBAL_STATE_CODE();
2408 bdrv_unref(s
->old_bs
);
2411 static void bdrv_replace_child_abort(void *opaque
)
2413 BdrvReplaceChildState
*s
= opaque
;
2414 BlockDriverState
*new_bs
= s
->child
->bs
;
2416 GLOBAL_STATE_CODE();
2417 /* old_bs reference is transparently moved from @s to @s->child */
2418 bdrv_replace_child_noperm(s
->child
, s
->old_bs
);
2422 static TransactionActionDrv bdrv_replace_child_drv
= {
2423 .commit
= bdrv_replace_child_commit
,
2424 .abort
= bdrv_replace_child_abort
,
2429 * bdrv_replace_child_tran
2431 * Note: real unref of old_bs is done only on commit.
2433 * The function doesn't update permissions, caller is responsible for this.
2435 static void bdrv_replace_child_tran(BdrvChild
*child
, BlockDriverState
*new_bs
,
2438 BdrvReplaceChildState
*s
= g_new(BdrvReplaceChildState
, 1);
2439 *s
= (BdrvReplaceChildState
) {
2441 .old_bs
= child
->bs
,
2443 tran_add(tran
, &bdrv_replace_child_drv
, s
);
2448 bdrv_replace_child_noperm(child
, new_bs
);
2449 /* old_bs reference is transparently moved from @child to @s */
2453 * Refresh permissions in @bs subtree. The function is intended to be called
2454 * after some graph modification that was done without permission update.
2456 static int bdrv_node_refresh_perm(BlockDriverState
*bs
, BlockReopenQueue
*q
,
2457 Transaction
*tran
, Error
**errp
)
2459 BlockDriver
*drv
= bs
->drv
;
2462 uint64_t cumulative_perms
, cumulative_shared_perms
;
2463 GLOBAL_STATE_CODE();
2465 bdrv_get_cumulative_perm(bs
, &cumulative_perms
, &cumulative_shared_perms
);
2467 /* Write permissions never work with read-only images */
2468 if ((cumulative_perms
& (BLK_PERM_WRITE
| BLK_PERM_WRITE_UNCHANGED
)) &&
2469 !bdrv_is_writable_after_reopen(bs
, q
))
2471 if (!bdrv_is_writable_after_reopen(bs
, NULL
)) {
2472 error_setg(errp
, "Block node is read-only");
2474 error_setg(errp
, "Read-only block node '%s' cannot support "
2475 "read-write users", bdrv_get_node_name(bs
));
2482 * Unaligned requests will automatically be aligned to bl.request_alignment
2483 * and without RESIZE we can't extend requests to write to space beyond the
2484 * end of the image, so it's required that the image size is aligned.
2486 if ((cumulative_perms
& (BLK_PERM_WRITE
| BLK_PERM_WRITE_UNCHANGED
)) &&
2487 !(cumulative_perms
& BLK_PERM_RESIZE
))
2489 if ((bs
->total_sectors
* BDRV_SECTOR_SIZE
) % bs
->bl
.request_alignment
) {
2490 error_setg(errp
, "Cannot get 'write' permission without 'resize': "
2491 "Image size is not a multiple of request "
2497 /* Check this node */
2502 ret
= bdrv_drv_set_perm(bs
, cumulative_perms
, cumulative_shared_perms
, tran
,
2508 /* Drivers that never have children can omit .bdrv_child_perm() */
2509 if (!drv
->bdrv_child_perm
) {
2510 assert(QLIST_EMPTY(&bs
->children
));
2514 /* Check all children */
2515 QLIST_FOREACH(c
, &bs
->children
, next
) {
2516 uint64_t cur_perm
, cur_shared
;
2518 bdrv_child_perm(bs
, c
->bs
, c
, c
->role
, q
,
2519 cumulative_perms
, cumulative_shared_perms
,
2520 &cur_perm
, &cur_shared
);
2521 bdrv_child_set_perm(c
, cur_perm
, cur_shared
, tran
);
2527 static int bdrv_list_refresh_perms(GSList
*list
, BlockReopenQueue
*q
,
2528 Transaction
*tran
, Error
**errp
)
2531 BlockDriverState
*bs
;
2532 GLOBAL_STATE_CODE();
2534 for ( ; list
; list
= list
->next
) {
2537 if (bdrv_parent_perms_conflict(bs
, errp
)) {
2541 ret
= bdrv_node_refresh_perm(bs
, q
, tran
, errp
);
2550 void bdrv_get_cumulative_perm(BlockDriverState
*bs
, uint64_t *perm
,
2551 uint64_t *shared_perm
)
2554 uint64_t cumulative_perms
= 0;
2555 uint64_t cumulative_shared_perms
= BLK_PERM_ALL
;
2557 GLOBAL_STATE_CODE();
2559 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
2560 cumulative_perms
|= c
->perm
;
2561 cumulative_shared_perms
&= c
->shared_perm
;
2564 *perm
= cumulative_perms
;
2565 *shared_perm
= cumulative_shared_perms
;
2568 char *bdrv_perm_names(uint64_t perm
)
2574 { BLK_PERM_CONSISTENT_READ
, "consistent read" },
2575 { BLK_PERM_WRITE
, "write" },
2576 { BLK_PERM_WRITE_UNCHANGED
, "write unchanged" },
2577 { BLK_PERM_RESIZE
, "resize" },
2581 GString
*result
= g_string_sized_new(30);
2582 struct perm_name
*p
;
2584 for (p
= permissions
; p
->name
; p
++) {
2585 if (perm
& p
->perm
) {
2586 if (result
->len
> 0) {
2587 g_string_append(result
, ", ");
2589 g_string_append(result
, p
->name
);
2593 return g_string_free(result
, FALSE
);
2597 static int bdrv_refresh_perms(BlockDriverState
*bs
, Error
**errp
)
2600 Transaction
*tran
= tran_new();
2601 g_autoptr(GSList
) list
= bdrv_topological_dfs(NULL
, NULL
, bs
);
2602 GLOBAL_STATE_CODE();
2604 ret
= bdrv_list_refresh_perms(list
, NULL
, tran
, errp
);
2605 tran_finalize(tran
, ret
);
2610 int bdrv_child_try_set_perm(BdrvChild
*c
, uint64_t perm
, uint64_t shared
,
2613 Error
*local_err
= NULL
;
2614 Transaction
*tran
= tran_new();
2617 GLOBAL_STATE_CODE();
2619 bdrv_child_set_perm(c
, perm
, shared
, tran
);
2621 ret
= bdrv_refresh_perms(c
->bs
, &local_err
);
2623 tran_finalize(tran
, ret
);
2626 if ((perm
& ~c
->perm
) || (c
->shared_perm
& ~shared
)) {
2627 /* tighten permissions */
2628 error_propagate(errp
, local_err
);
2631 * Our caller may intend to only loosen restrictions and
2632 * does not expect this function to fail. Errors are not
2633 * fatal in such a case, so we can just hide them from our
2636 error_free(local_err
);
2644 int bdrv_child_refresh_perms(BlockDriverState
*bs
, BdrvChild
*c
, Error
**errp
)
2646 uint64_t parent_perms
, parent_shared
;
2647 uint64_t perms
, shared
;
2649 GLOBAL_STATE_CODE();
2651 bdrv_get_cumulative_perm(bs
, &parent_perms
, &parent_shared
);
2652 bdrv_child_perm(bs
, c
->bs
, c
, c
->role
, NULL
,
2653 parent_perms
, parent_shared
, &perms
, &shared
);
2655 return bdrv_child_try_set_perm(c
, perms
, shared
, errp
);
2659 * Default implementation for .bdrv_child_perm() for block filters:
2660 * Forward CONSISTENT_READ, WRITE, WRITE_UNCHANGED, and RESIZE to the
2663 static void bdrv_filter_default_perms(BlockDriverState
*bs
, BdrvChild
*c
,
2665 BlockReopenQueue
*reopen_queue
,
2666 uint64_t perm
, uint64_t shared
,
2667 uint64_t *nperm
, uint64_t *nshared
)
2669 GLOBAL_STATE_CODE();
2670 *nperm
= perm
& DEFAULT_PERM_PASSTHROUGH
;
2671 *nshared
= (shared
& DEFAULT_PERM_PASSTHROUGH
) | DEFAULT_PERM_UNCHANGED
;
2674 static void bdrv_default_perms_for_cow(BlockDriverState
*bs
, BdrvChild
*c
,
2676 BlockReopenQueue
*reopen_queue
,
2677 uint64_t perm
, uint64_t shared
,
2678 uint64_t *nperm
, uint64_t *nshared
)
2680 assert(role
& BDRV_CHILD_COW
);
2681 GLOBAL_STATE_CODE();
2684 * We want consistent read from backing files if the parent needs it.
2685 * No other operations are performed on backing files.
2687 perm
&= BLK_PERM_CONSISTENT_READ
;
2690 * If the parent can deal with changing data, we're okay with a
2691 * writable and resizable backing file.
2692 * TODO Require !(perm & BLK_PERM_CONSISTENT_READ), too?
2694 if (shared
& BLK_PERM_WRITE
) {
2695 shared
= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2700 shared
|= BLK_PERM_CONSISTENT_READ
| BLK_PERM_WRITE_UNCHANGED
;
2702 if (bs
->open_flags
& BDRV_O_INACTIVE
) {
2703 shared
|= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2710 static void bdrv_default_perms_for_storage(BlockDriverState
*bs
, BdrvChild
*c
,
2712 BlockReopenQueue
*reopen_queue
,
2713 uint64_t perm
, uint64_t shared
,
2714 uint64_t *nperm
, uint64_t *nshared
)
2718 GLOBAL_STATE_CODE();
2719 assert(role
& (BDRV_CHILD_METADATA
| BDRV_CHILD_DATA
));
2721 flags
= bdrv_reopen_get_flags(reopen_queue
, bs
);
2724 * Apart from the modifications below, the same permissions are
2725 * forwarded and left alone as for filters
2727 bdrv_filter_default_perms(bs
, c
, role
, reopen_queue
,
2728 perm
, shared
, &perm
, &shared
);
2730 if (role
& BDRV_CHILD_METADATA
) {
2731 /* Format drivers may touch metadata even if the guest doesn't write */
2732 if (bdrv_is_writable_after_reopen(bs
, reopen_queue
)) {
2733 perm
|= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2737 * bs->file always needs to be consistent because of the
2738 * metadata. We can never allow other users to resize or write
2741 if (!(flags
& BDRV_O_NO_IO
)) {
2742 perm
|= BLK_PERM_CONSISTENT_READ
;
2744 shared
&= ~(BLK_PERM_WRITE
| BLK_PERM_RESIZE
);
2747 if (role
& BDRV_CHILD_DATA
) {
2749 * Technically, everything in this block is a subset of the
2750 * BDRV_CHILD_METADATA path taken above, and so this could
2751 * be an "else if" branch. However, that is not obvious, and
2752 * this function is not performance critical, therefore we let
2753 * this be an independent "if".
2757 * We cannot allow other users to resize the file because the
2758 * format driver might have some assumptions about the size
2759 * (e.g. because it is stored in metadata, or because the file
2760 * is split into fixed-size data files).
2762 shared
&= ~BLK_PERM_RESIZE
;
2765 * WRITE_UNCHANGED often cannot be performed as such on the
2766 * data file. For example, the qcow2 driver may still need to
2767 * write copied clusters on copy-on-read.
2769 if (perm
& BLK_PERM_WRITE_UNCHANGED
) {
2770 perm
|= BLK_PERM_WRITE
;
2774 * If the data file is written to, the format driver may
2775 * expect to be able to resize it by writing beyond the EOF.
2777 if (perm
& BLK_PERM_WRITE
) {
2778 perm
|= BLK_PERM_RESIZE
;
2782 if (bs
->open_flags
& BDRV_O_INACTIVE
) {
2783 shared
|= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2790 void bdrv_default_perms(BlockDriverState
*bs
, BdrvChild
*c
,
2791 BdrvChildRole role
, BlockReopenQueue
*reopen_queue
,
2792 uint64_t perm
, uint64_t shared
,
2793 uint64_t *nperm
, uint64_t *nshared
)
2795 GLOBAL_STATE_CODE();
2796 if (role
& BDRV_CHILD_FILTERED
) {
2797 assert(!(role
& (BDRV_CHILD_DATA
| BDRV_CHILD_METADATA
|
2799 bdrv_filter_default_perms(bs
, c
, role
, reopen_queue
,
2800 perm
, shared
, nperm
, nshared
);
2801 } else if (role
& BDRV_CHILD_COW
) {
2802 assert(!(role
& (BDRV_CHILD_DATA
| BDRV_CHILD_METADATA
)));
2803 bdrv_default_perms_for_cow(bs
, c
, role
, reopen_queue
,
2804 perm
, shared
, nperm
, nshared
);
2805 } else if (role
& (BDRV_CHILD_METADATA
| BDRV_CHILD_DATA
)) {
2806 bdrv_default_perms_for_storage(bs
, c
, role
, reopen_queue
,
2807 perm
, shared
, nperm
, nshared
);
2809 g_assert_not_reached();
2813 uint64_t bdrv_qapi_perm_to_blk_perm(BlockPermission qapi_perm
)
2815 static const uint64_t permissions
[] = {
2816 [BLOCK_PERMISSION_CONSISTENT_READ
] = BLK_PERM_CONSISTENT_READ
,
2817 [BLOCK_PERMISSION_WRITE
] = BLK_PERM_WRITE
,
2818 [BLOCK_PERMISSION_WRITE_UNCHANGED
] = BLK_PERM_WRITE_UNCHANGED
,
2819 [BLOCK_PERMISSION_RESIZE
] = BLK_PERM_RESIZE
,
2822 QEMU_BUILD_BUG_ON(ARRAY_SIZE(permissions
) != BLOCK_PERMISSION__MAX
);
2823 QEMU_BUILD_BUG_ON(1UL << ARRAY_SIZE(permissions
) != BLK_PERM_ALL
+ 1);
2825 assert(qapi_perm
< BLOCK_PERMISSION__MAX
);
2827 return permissions
[qapi_perm
];
2830 static void bdrv_replace_child_noperm(BdrvChild
*child
,
2831 BlockDriverState
*new_bs
)
2833 BlockDriverState
*old_bs
= child
->bs
;
2834 int new_bs_quiesce_counter
;
2837 assert(!child
->frozen
);
2838 assert(old_bs
!= new_bs
);
2839 GLOBAL_STATE_CODE();
2841 if (old_bs
&& new_bs
) {
2842 assert(bdrv_get_aio_context(old_bs
) == bdrv_get_aio_context(new_bs
));
2845 new_bs_quiesce_counter
= (new_bs
? new_bs
->quiesce_counter
: 0);
2846 drain_saldo
= new_bs_quiesce_counter
- child
->parent_quiesce_counter
;
2849 * If the new child node is drained but the old one was not, flush
2850 * all outstanding requests to the old child node.
2852 while (drain_saldo
> 0 && child
->klass
->drained_begin
) {
2853 bdrv_parent_drained_begin_single(child
, true);
2858 /* Detach first so that the recursive drain sections coming from @child
2859 * are already gone and we only end the drain sections that came from
2861 if (child
->klass
->detach
) {
2862 child
->klass
->detach(child
);
2864 assert_bdrv_graph_writable(old_bs
);
2865 QLIST_REMOVE(child
, next_parent
);
2871 assert_bdrv_graph_writable(new_bs
);
2872 QLIST_INSERT_HEAD(&new_bs
->parents
, child
, next_parent
);
2875 * Detaching the old node may have led to the new node's
2876 * quiesce_counter having been decreased. Not a problem, we
2877 * just need to recognize this here and then invoke
2878 * drained_end appropriately more often.
2880 assert(new_bs
->quiesce_counter
<= new_bs_quiesce_counter
);
2881 drain_saldo
+= new_bs
->quiesce_counter
- new_bs_quiesce_counter
;
2883 /* Attach only after starting new drained sections, so that recursive
2884 * drain sections coming from @child don't get an extra .drained_begin
2886 if (child
->klass
->attach
) {
2887 child
->klass
->attach(child
);
2892 * If the old child node was drained but the new one is not, allow
2893 * requests to come in only after the new node has been attached.
2895 while (drain_saldo
< 0 && child
->klass
->drained_end
) {
2896 bdrv_parent_drained_end_single(child
);
2902 * Free the given @child.
2904 * The child must be empty (i.e. `child->bs == NULL`) and it must be
2905 * unused (i.e. not in a children list).
2907 static void bdrv_child_free(BdrvChild
*child
)
2910 GLOBAL_STATE_CODE();
2911 assert(!child
->next
.le_prev
); /* not in children list */
2913 g_free(child
->name
);
2917 typedef struct BdrvAttachChildCommonState
{
2919 AioContext
*old_parent_ctx
;
2920 AioContext
*old_child_ctx
;
2921 } BdrvAttachChildCommonState
;
2923 static void bdrv_attach_child_common_abort(void *opaque
)
2925 BdrvAttachChildCommonState
*s
= opaque
;
2926 BlockDriverState
*bs
= s
->child
->bs
;
2928 GLOBAL_STATE_CODE();
2929 bdrv_replace_child_noperm(s
->child
, NULL
);
2931 if (bdrv_get_aio_context(bs
) != s
->old_child_ctx
) {
2932 bdrv_try_change_aio_context(bs
, s
->old_child_ctx
, NULL
, &error_abort
);
2935 if (bdrv_child_get_parent_aio_context(s
->child
) != s
->old_parent_ctx
) {
2937 GHashTable
*visited
;
2942 /* No need to visit `child`, because it has been detached already */
2943 visited
= g_hash_table_new(NULL
, NULL
);
2944 ret
= s
->child
->klass
->change_aio_ctx(s
->child
, s
->old_parent_ctx
,
2945 visited
, tran
, &error_abort
);
2946 g_hash_table_destroy(visited
);
2948 /* transaction is supposed to always succeed */
2949 assert(ret
== true);
2954 bdrv_child_free(s
->child
);
2957 static TransactionActionDrv bdrv_attach_child_common_drv
= {
2958 .abort
= bdrv_attach_child_common_abort
,
2963 * Common part of attaching bdrv child to bs or to blk or to job
2965 * Function doesn't update permissions, caller is responsible for this.
2967 * Returns new created child.
2969 static BdrvChild
*bdrv_attach_child_common(BlockDriverState
*child_bs
,
2970 const char *child_name
,
2971 const BdrvChildClass
*child_class
,
2972 BdrvChildRole child_role
,
2973 uint64_t perm
, uint64_t shared_perm
,
2975 Transaction
*tran
, Error
**errp
)
2977 BdrvChild
*new_child
;
2978 AioContext
*parent_ctx
;
2979 AioContext
*child_ctx
= bdrv_get_aio_context(child_bs
);
2981 assert(child_class
->get_parent_desc
);
2982 GLOBAL_STATE_CODE();
2984 new_child
= g_new(BdrvChild
, 1);
2985 *new_child
= (BdrvChild
) {
2987 .name
= g_strdup(child_name
),
2988 .klass
= child_class
,
2991 .shared_perm
= shared_perm
,
2996 * If the AioContexts don't match, first try to move the subtree of
2997 * child_bs into the AioContext of the new parent. If this doesn't work,
2998 * try moving the parent into the AioContext of child_bs instead.
3000 parent_ctx
= bdrv_child_get_parent_aio_context(new_child
);
3001 if (child_ctx
!= parent_ctx
) {
3002 Error
*local_err
= NULL
;
3003 int ret
= bdrv_try_change_aio_context(child_bs
, parent_ctx
, NULL
,
3006 if (ret
< 0 && child_class
->change_aio_ctx
) {
3007 Transaction
*tran
= tran_new();
3008 GHashTable
*visited
= g_hash_table_new(NULL
, NULL
);
3011 g_hash_table_add(visited
, new_child
);
3012 ret_child
= child_class
->change_aio_ctx(new_child
, child_ctx
,
3013 visited
, tran
, NULL
);
3014 if (ret_child
== true) {
3015 error_free(local_err
);
3018 tran_finalize(tran
, ret_child
== true ? 0 : -1);
3019 g_hash_table_destroy(visited
);
3023 error_propagate(errp
, local_err
);
3024 bdrv_child_free(new_child
);
3030 bdrv_replace_child_noperm(new_child
, child_bs
);
3032 BdrvAttachChildCommonState
*s
= g_new(BdrvAttachChildCommonState
, 1);
3033 *s
= (BdrvAttachChildCommonState
) {
3035 .old_parent_ctx
= parent_ctx
,
3036 .old_child_ctx
= child_ctx
,
3038 tran_add(tran
, &bdrv_attach_child_common_drv
, s
);
3044 * Function doesn't update permissions, caller is responsible for this.
3046 static BdrvChild
*bdrv_attach_child_noperm(BlockDriverState
*parent_bs
,
3047 BlockDriverState
*child_bs
,
3048 const char *child_name
,
3049 const BdrvChildClass
*child_class
,
3050 BdrvChildRole child_role
,
3054 uint64_t perm
, shared_perm
;
3056 assert(parent_bs
->drv
);
3057 GLOBAL_STATE_CODE();
3059 if (bdrv_recurse_has_child(child_bs
, parent_bs
)) {
3060 error_setg(errp
, "Making '%s' a %s child of '%s' would create a cycle",
3061 child_bs
->node_name
, child_name
, parent_bs
->node_name
);
3065 bdrv_get_cumulative_perm(parent_bs
, &perm
, &shared_perm
);
3066 bdrv_child_perm(parent_bs
, child_bs
, NULL
, child_role
, NULL
,
3067 perm
, shared_perm
, &perm
, &shared_perm
);
3069 return bdrv_attach_child_common(child_bs
, child_name
, child_class
,
3070 child_role
, perm
, shared_perm
, parent_bs
,
3074 static void bdrv_detach_child(BdrvChild
*child
)
3076 BlockDriverState
*old_bs
= child
->bs
;
3078 GLOBAL_STATE_CODE();
3079 bdrv_replace_child_noperm(child
, NULL
);
3080 bdrv_child_free(child
);
3084 * Update permissions for old node. We're just taking a parent away, so
3085 * we're loosening restrictions. Errors of permission update are not
3086 * fatal in this case, ignore them.
3088 bdrv_refresh_perms(old_bs
, NULL
);
3091 * When the parent requiring a non-default AioContext is removed, the
3092 * node moves back to the main AioContext
3094 bdrv_try_change_aio_context(old_bs
, qemu_get_aio_context(), NULL
, NULL
);
3099 * This function steals the reference to child_bs from the caller.
3100 * That reference is later dropped by bdrv_root_unref_child().
3102 * On failure NULL is returned, errp is set and the reference to
3103 * child_bs is also dropped.
3105 * The caller must hold the AioContext lock @child_bs, but not that of @ctx
3106 * (unless @child_bs is already in @ctx).
3108 BdrvChild
*bdrv_root_attach_child(BlockDriverState
*child_bs
,
3109 const char *child_name
,
3110 const BdrvChildClass
*child_class
,
3111 BdrvChildRole child_role
,
3112 uint64_t perm
, uint64_t shared_perm
,
3113 void *opaque
, Error
**errp
)
3117 Transaction
*tran
= tran_new();
3119 GLOBAL_STATE_CODE();
3121 child
= bdrv_attach_child_common(child_bs
, child_name
, child_class
,
3122 child_role
, perm
, shared_perm
, opaque
,
3129 ret
= bdrv_refresh_perms(child_bs
, errp
);
3132 tran_finalize(tran
, ret
);
3134 bdrv_unref(child_bs
);
3136 return ret
< 0 ? NULL
: child
;
3140 * This function transfers the reference to child_bs from the caller
3141 * to parent_bs. That reference is later dropped by parent_bs on
3142 * bdrv_close() or if someone calls bdrv_unref_child().
3144 * On failure NULL is returned, errp is set and the reference to
3145 * child_bs is also dropped.
3147 * If @parent_bs and @child_bs are in different AioContexts, the caller must
3148 * hold the AioContext lock for @child_bs, but not for @parent_bs.
3150 BdrvChild
*bdrv_attach_child(BlockDriverState
*parent_bs
,
3151 BlockDriverState
*child_bs
,
3152 const char *child_name
,
3153 const BdrvChildClass
*child_class
,
3154 BdrvChildRole child_role
,
3159 Transaction
*tran
= tran_new();
3161 GLOBAL_STATE_CODE();
3163 child
= bdrv_attach_child_noperm(parent_bs
, child_bs
, child_name
,
3164 child_class
, child_role
, tran
, errp
);
3170 ret
= bdrv_refresh_perms(parent_bs
, errp
);
3176 tran_finalize(tran
, ret
);
3178 bdrv_unref(child_bs
);
3180 return ret
< 0 ? NULL
: child
;
3183 /* Callers must ensure that child->frozen is false. */
3184 void bdrv_root_unref_child(BdrvChild
*child
)
3186 BlockDriverState
*child_bs
;
3188 GLOBAL_STATE_CODE();
3190 child_bs
= child
->bs
;
3191 bdrv_detach_child(child
);
3192 bdrv_unref(child_bs
);
3195 typedef struct BdrvSetInheritsFrom
{
3196 BlockDriverState
*bs
;
3197 BlockDriverState
*old_inherits_from
;
3198 } BdrvSetInheritsFrom
;
3200 static void bdrv_set_inherits_from_abort(void *opaque
)
3202 BdrvSetInheritsFrom
*s
= opaque
;
3204 s
->bs
->inherits_from
= s
->old_inherits_from
;
3207 static TransactionActionDrv bdrv_set_inherits_from_drv
= {
3208 .abort
= bdrv_set_inherits_from_abort
,
3212 /* @tran is allowed to be NULL. In this case no rollback is possible */
3213 static void bdrv_set_inherits_from(BlockDriverState
*bs
,
3214 BlockDriverState
*new_inherits_from
,
3218 BdrvSetInheritsFrom
*s
= g_new(BdrvSetInheritsFrom
, 1);
3220 *s
= (BdrvSetInheritsFrom
) {
3222 .old_inherits_from
= bs
->inherits_from
,
3225 tran_add(tran
, &bdrv_set_inherits_from_drv
, s
);
3228 bs
->inherits_from
= new_inherits_from
;
3232 * Clear all inherits_from pointers from children and grandchildren of
3233 * @root that point to @root, where necessary.
3234 * @tran is allowed to be NULL. In this case no rollback is possible
3236 static void bdrv_unset_inherits_from(BlockDriverState
*root
, BdrvChild
*child
,
3241 if (child
->bs
->inherits_from
== root
) {
3243 * Remove inherits_from only when the last reference between root and
3244 * child->bs goes away.
3246 QLIST_FOREACH(c
, &root
->children
, next
) {
3247 if (c
!= child
&& c
->bs
== child
->bs
) {
3252 bdrv_set_inherits_from(child
->bs
, NULL
, tran
);
3256 QLIST_FOREACH(c
, &child
->bs
->children
, next
) {
3257 bdrv_unset_inherits_from(root
, c
, tran
);
3261 /* Callers must ensure that child->frozen is false. */
3262 void bdrv_unref_child(BlockDriverState
*parent
, BdrvChild
*child
)
3264 GLOBAL_STATE_CODE();
3265 if (child
== NULL
) {
3269 bdrv_unset_inherits_from(parent
, child
, NULL
);
3270 bdrv_root_unref_child(child
);
3274 static void bdrv_parent_cb_change_media(BlockDriverState
*bs
, bool load
)
3277 GLOBAL_STATE_CODE();
3278 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
3279 if (c
->klass
->change_media
) {
3280 c
->klass
->change_media(c
, load
);
3285 /* Return true if you can reach parent going through child->inherits_from
3286 * recursively. If parent or child are NULL, return false */
3287 static bool bdrv_inherits_from_recursive(BlockDriverState
*child
,
3288 BlockDriverState
*parent
)
3290 while (child
&& child
!= parent
) {
3291 child
= child
->inherits_from
;
3294 return child
!= NULL
;
3298 * Return the BdrvChildRole for @bs's backing child. bs->backing is
3299 * mostly used for COW backing children (role = COW), but also for
3300 * filtered children (role = FILTERED | PRIMARY).
3302 static BdrvChildRole
bdrv_backing_role(BlockDriverState
*bs
)
3304 if (bs
->drv
&& bs
->drv
->is_filter
) {
3305 return BDRV_CHILD_FILTERED
| BDRV_CHILD_PRIMARY
;
3307 return BDRV_CHILD_COW
;
3312 * Sets the bs->backing or bs->file link of a BDS. A new reference is created;
3313 * callers which don't need their own reference any more must call bdrv_unref().
3315 * Function doesn't update permissions, caller is responsible for this.
3317 static int bdrv_set_file_or_backing_noperm(BlockDriverState
*parent_bs
,
3318 BlockDriverState
*child_bs
,
3320 Transaction
*tran
, Error
**errp
)
3322 bool update_inherits_from
=
3323 bdrv_inherits_from_recursive(child_bs
, parent_bs
);
3324 BdrvChild
*child
= is_backing
? parent_bs
->backing
: parent_bs
->file
;
3327 GLOBAL_STATE_CODE();
3329 if (!parent_bs
->drv
) {
3331 * Node without drv is an object without a class :/. TODO: finally fix
3332 * qcow2 driver to never clear bs->drv and implement format corruption
3333 * handling in other way.
3335 error_setg(errp
, "Node corrupted");
3339 if (child
&& child
->frozen
) {
3340 error_setg(errp
, "Cannot change frozen '%s' link from '%s' to '%s'",
3341 child
->name
, parent_bs
->node_name
, child
->bs
->node_name
);
3345 if (is_backing
&& !parent_bs
->drv
->is_filter
&&
3346 !parent_bs
->drv
->supports_backing
)
3348 error_setg(errp
, "Driver '%s' of node '%s' does not support backing "
3349 "files", parent_bs
->drv
->format_name
, parent_bs
->node_name
);
3353 if (parent_bs
->drv
->is_filter
) {
3354 role
= BDRV_CHILD_FILTERED
| BDRV_CHILD_PRIMARY
;
3355 } else if (is_backing
) {
3356 role
= BDRV_CHILD_COW
;
3359 * We only can use same role as it is in existing child. We don't have
3360 * infrastructure to determine role of file child in generic way
3363 error_setg(errp
, "Cannot set file child to format node without "
3371 bdrv_unset_inherits_from(parent_bs
, child
, tran
);
3372 bdrv_remove_child(child
, tran
);
3379 child
= bdrv_attach_child_noperm(parent_bs
, child_bs
,
3380 is_backing
? "backing" : "file",
3381 &child_of_bds
, role
,
3389 * If inherits_from pointed recursively to bs then let's update it to
3390 * point directly to bs (else it will become NULL).
3392 if (update_inherits_from
) {
3393 bdrv_set_inherits_from(child_bs
, parent_bs
, tran
);
3397 bdrv_refresh_limits(parent_bs
, tran
, NULL
);
3402 static int bdrv_set_backing_noperm(BlockDriverState
*bs
,
3403 BlockDriverState
*backing_hd
,
3404 Transaction
*tran
, Error
**errp
)
3406 GLOBAL_STATE_CODE();
3407 return bdrv_set_file_or_backing_noperm(bs
, backing_hd
, true, tran
, errp
);
3410 int bdrv_set_backing_hd(BlockDriverState
*bs
, BlockDriverState
*backing_hd
,
3414 Transaction
*tran
= tran_new();
3416 GLOBAL_STATE_CODE();
3417 bdrv_drained_begin(bs
);
3419 ret
= bdrv_set_backing_noperm(bs
, backing_hd
, tran
, errp
);
3424 ret
= bdrv_refresh_perms(bs
, errp
);
3426 tran_finalize(tran
, ret
);
3428 bdrv_drained_end(bs
);
3434 * Opens the backing file for a BlockDriverState if not yet open
3436 * bdref_key specifies the key for the image's BlockdevRef in the options QDict.
3437 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
3438 * itself, all options starting with "${bdref_key}." are considered part of the
3441 * TODO Can this be unified with bdrv_open_image()?
3443 int bdrv_open_backing_file(BlockDriverState
*bs
, QDict
*parent_options
,
3444 const char *bdref_key
, Error
**errp
)
3446 char *backing_filename
= NULL
;
3447 char *bdref_key_dot
;
3448 const char *reference
= NULL
;
3450 bool implicit_backing
= false;
3451 BlockDriverState
*backing_hd
;
3453 QDict
*tmp_parent_options
= NULL
;
3454 Error
*local_err
= NULL
;
3456 GLOBAL_STATE_CODE();
3458 if (bs
->backing
!= NULL
) {
3462 /* NULL means an empty set of options */
3463 if (parent_options
== NULL
) {
3464 tmp_parent_options
= qdict_new();
3465 parent_options
= tmp_parent_options
;
3468 bs
->open_flags
&= ~BDRV_O_NO_BACKING
;
3470 bdref_key_dot
= g_strdup_printf("%s.", bdref_key
);
3471 qdict_extract_subqdict(parent_options
, &options
, bdref_key_dot
);
3472 g_free(bdref_key_dot
);
3475 * Caution: while qdict_get_try_str() is fine, getting non-string
3476 * types would require more care. When @parent_options come from
3477 * -blockdev or blockdev_add, its members are typed according to
3478 * the QAPI schema, but when they come from -drive, they're all
3481 reference
= qdict_get_try_str(parent_options
, bdref_key
);
3482 if (reference
|| qdict_haskey(options
, "file.filename")) {
3483 /* keep backing_filename NULL */
3484 } else if (bs
->backing_file
[0] == '\0' && qdict_size(options
) == 0) {
3485 qobject_unref(options
);
3488 if (qdict_size(options
) == 0) {
3489 /* If the user specifies options that do not modify the
3490 * backing file's behavior, we might still consider it the
3491 * implicit backing file. But it's easier this way, and
3492 * just specifying some of the backing BDS's options is
3493 * only possible with -drive anyway (otherwise the QAPI
3494 * schema forces the user to specify everything). */
3495 implicit_backing
= !strcmp(bs
->auto_backing_file
, bs
->backing_file
);
3498 backing_filename
= bdrv_get_full_backing_filename(bs
, &local_err
);
3501 error_propagate(errp
, local_err
);
3502 qobject_unref(options
);
3507 if (!bs
->drv
|| !bs
->drv
->supports_backing
) {
3509 error_setg(errp
, "Driver doesn't support backing files");
3510 qobject_unref(options
);
3515 bs
->backing_format
[0] != '\0' && !qdict_haskey(options
, "driver")) {
3516 qdict_put_str(options
, "driver", bs
->backing_format
);
3519 backing_hd
= bdrv_open_inherit(backing_filename
, reference
, options
, 0, bs
,
3520 &child_of_bds
, bdrv_backing_role(bs
), errp
);
3522 bs
->open_flags
|= BDRV_O_NO_BACKING
;
3523 error_prepend(errp
, "Could not open backing file: ");
3528 if (implicit_backing
) {
3529 bdrv_refresh_filename(backing_hd
);
3530 pstrcpy(bs
->auto_backing_file
, sizeof(bs
->auto_backing_file
),
3531 backing_hd
->filename
);
3534 /* Hook up the backing file link; drop our reference, bs owns the
3535 * backing_hd reference now */
3536 ret
= bdrv_set_backing_hd(bs
, backing_hd
, errp
);
3537 bdrv_unref(backing_hd
);
3542 qdict_del(parent_options
, bdref_key
);
3545 g_free(backing_filename
);
3546 qobject_unref(tmp_parent_options
);
3550 static BlockDriverState
*
3551 bdrv_open_child_bs(const char *filename
, QDict
*options
, const char *bdref_key
,
3552 BlockDriverState
*parent
, const BdrvChildClass
*child_class
,
3553 BdrvChildRole child_role
, bool allow_none
, Error
**errp
)
3555 BlockDriverState
*bs
= NULL
;
3556 QDict
*image_options
;
3557 char *bdref_key_dot
;
3558 const char *reference
;
3560 assert(child_class
!= NULL
);
3562 bdref_key_dot
= g_strdup_printf("%s.", bdref_key
);
3563 qdict_extract_subqdict(options
, &image_options
, bdref_key_dot
);
3564 g_free(bdref_key_dot
);
3567 * Caution: while qdict_get_try_str() is fine, getting non-string
3568 * types would require more care. When @options come from
3569 * -blockdev or blockdev_add, its members are typed according to
3570 * the QAPI schema, but when they come from -drive, they're all
3573 reference
= qdict_get_try_str(options
, bdref_key
);
3574 if (!filename
&& !reference
&& !qdict_size(image_options
)) {
3576 error_setg(errp
, "A block device must be specified for \"%s\"",
3579 qobject_unref(image_options
);
3583 bs
= bdrv_open_inherit(filename
, reference
, image_options
, 0,
3584 parent
, child_class
, child_role
, errp
);
3590 qdict_del(options
, bdref_key
);
3595 * Opens a disk image whose options are given as BlockdevRef in another block
3598 * If allow_none is true, no image will be opened if filename is false and no
3599 * BlockdevRef is given. NULL will be returned, but errp remains unset.
3601 * bdrev_key specifies the key for the image's BlockdevRef in the options QDict.
3602 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
3603 * itself, all options starting with "${bdref_key}." are considered part of the
3606 * The BlockdevRef will be removed from the options QDict.
3608 BdrvChild
*bdrv_open_child(const char *filename
,
3609 QDict
*options
, const char *bdref_key
,
3610 BlockDriverState
*parent
,
3611 const BdrvChildClass
*child_class
,
3612 BdrvChildRole child_role
,
3613 bool allow_none
, Error
**errp
)
3615 BlockDriverState
*bs
;
3617 GLOBAL_STATE_CODE();
3619 bs
= bdrv_open_child_bs(filename
, options
, bdref_key
, parent
, child_class
,
3620 child_role
, allow_none
, errp
);
3625 return bdrv_attach_child(parent
, bs
, bdref_key
, child_class
, child_role
,
3630 * Wrapper on bdrv_open_child() for most popular case: open primary child of bs.
3632 int bdrv_open_file_child(const char *filename
,
3633 QDict
*options
, const char *bdref_key
,
3634 BlockDriverState
*parent
, Error
**errp
)
3638 /* commit_top and mirror_top don't use this function */
3639 assert(!parent
->drv
->filtered_child_is_backing
);
3640 role
= parent
->drv
->is_filter
?
3641 (BDRV_CHILD_FILTERED
| BDRV_CHILD_PRIMARY
) : BDRV_CHILD_IMAGE
;
3643 if (!bdrv_open_child(filename
, options
, bdref_key
, parent
,
3644 &child_of_bds
, role
, false, errp
))
3653 * TODO Future callers may need to specify parent/child_class in order for
3654 * option inheritance to work. Existing callers use it for the root node.
3656 BlockDriverState
*bdrv_open_blockdev_ref(BlockdevRef
*ref
, Error
**errp
)
3658 BlockDriverState
*bs
= NULL
;
3659 QObject
*obj
= NULL
;
3660 QDict
*qdict
= NULL
;
3661 const char *reference
= NULL
;
3664 GLOBAL_STATE_CODE();
3666 if (ref
->type
== QTYPE_QSTRING
) {
3667 reference
= ref
->u
.reference
;
3669 BlockdevOptions
*options
= &ref
->u
.definition
;
3670 assert(ref
->type
== QTYPE_QDICT
);
3672 v
= qobject_output_visitor_new(&obj
);
3673 visit_type_BlockdevOptions(v
, NULL
, &options
, &error_abort
);
3674 visit_complete(v
, &obj
);
3676 qdict
= qobject_to(QDict
, obj
);
3677 qdict_flatten(qdict
);
3679 /* bdrv_open_inherit() defaults to the values in bdrv_flags (for
3680 * compatibility with other callers) rather than what we want as the
3681 * real defaults. Apply the defaults here instead. */
3682 qdict_set_default_str(qdict
, BDRV_OPT_CACHE_DIRECT
, "off");
3683 qdict_set_default_str(qdict
, BDRV_OPT_CACHE_NO_FLUSH
, "off");
3684 qdict_set_default_str(qdict
, BDRV_OPT_READ_ONLY
, "off");
3685 qdict_set_default_str(qdict
, BDRV_OPT_AUTO_READ_ONLY
, "off");
3689 bs
= bdrv_open_inherit(NULL
, reference
, qdict
, 0, NULL
, NULL
, 0, errp
);
3696 static BlockDriverState
*bdrv_append_temp_snapshot(BlockDriverState
*bs
,
3698 QDict
*snapshot_options
,
3701 g_autofree
char *tmp_filename
= NULL
;
3703 QemuOpts
*opts
= NULL
;
3704 BlockDriverState
*bs_snapshot
= NULL
;
3707 GLOBAL_STATE_CODE();
3709 /* if snapshot, we create a temporary backing file and open it
3710 instead of opening 'filename' directly */
3712 /* Get the required size from the image */
3713 total_size
= bdrv_getlength(bs
);
3714 if (total_size
< 0) {
3715 error_setg_errno(errp
, -total_size
, "Could not get image size");
3719 /* Create the temporary image */
3720 tmp_filename
= create_tmp_file(errp
);
3721 if (!tmp_filename
) {
3725 opts
= qemu_opts_create(bdrv_qcow2
.create_opts
, NULL
, 0,
3727 qemu_opt_set_number(opts
, BLOCK_OPT_SIZE
, total_size
, &error_abort
);
3728 ret
= bdrv_create(&bdrv_qcow2
, tmp_filename
, opts
, errp
);
3729 qemu_opts_del(opts
);
3731 error_prepend(errp
, "Could not create temporary overlay '%s': ",
3736 /* Prepare options QDict for the temporary file */
3737 qdict_put_str(snapshot_options
, "file.driver", "file");
3738 qdict_put_str(snapshot_options
, "file.filename", tmp_filename
);
3739 qdict_put_str(snapshot_options
, "driver", "qcow2");
3741 bs_snapshot
= bdrv_open(NULL
, NULL
, snapshot_options
, flags
, errp
);
3742 snapshot_options
= NULL
;
3747 ret
= bdrv_append(bs_snapshot
, bs
, errp
);
3754 qobject_unref(snapshot_options
);
3759 * Opens a disk image (raw, qcow2, vmdk, ...)
3761 * options is a QDict of options to pass to the block drivers, or NULL for an
3762 * empty set of options. The reference to the QDict belongs to the block layer
3763 * after the call (even on failure), so if the caller intends to reuse the
3764 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
3766 * If *pbs is NULL, a new BDS will be created with a pointer to it stored there.
3767 * If it is not NULL, the referenced BDS will be reused.
3769 * The reference parameter may be used to specify an existing block device which
3770 * should be opened. If specified, neither options nor a filename may be given,
3771 * nor can an existing BDS be reused (that is, *pbs has to be NULL).
3773 static BlockDriverState
*bdrv_open_inherit(const char *filename
,
3774 const char *reference
,
3775 QDict
*options
, int flags
,
3776 BlockDriverState
*parent
,
3777 const BdrvChildClass
*child_class
,
3778 BdrvChildRole child_role
,
3782 BlockBackend
*file
= NULL
;
3783 BlockDriverState
*bs
;
3784 BlockDriver
*drv
= NULL
;
3786 const char *drvname
;
3787 const char *backing
;
3788 Error
*local_err
= NULL
;
3789 QDict
*snapshot_options
= NULL
;
3790 int snapshot_flags
= 0;
3792 assert(!child_class
|| !flags
);
3793 assert(!child_class
== !parent
);
3794 GLOBAL_STATE_CODE();
3797 bool options_non_empty
= options
? qdict_size(options
) : false;
3798 qobject_unref(options
);
3800 if (filename
|| options_non_empty
) {
3801 error_setg(errp
, "Cannot reference an existing block device with "
3802 "additional options or a new filename");
3806 bs
= bdrv_lookup_bs(reference
, reference
, errp
);
3817 /* NULL means an empty set of options */
3818 if (options
== NULL
) {
3819 options
= qdict_new();
3822 /* json: syntax counts as explicit options, as if in the QDict */
3823 parse_json_protocol(options
, &filename
, &local_err
);
3828 bs
->explicit_options
= qdict_clone_shallow(options
);
3831 bool parent_is_format
;
3834 parent_is_format
= parent
->drv
->is_format
;
3837 * parent->drv is not set yet because this node is opened for
3838 * (potential) format probing. That means that @parent is going
3839 * to be a format node.
3841 parent_is_format
= true;
3844 bs
->inherits_from
= parent
;
3845 child_class
->inherit_options(child_role
, parent_is_format
,
3847 parent
->open_flags
, parent
->options
);
3850 ret
= bdrv_fill_options(&options
, filename
, &flags
, &local_err
);
3856 * Set the BDRV_O_RDWR and BDRV_O_ALLOW_RDWR flags.
3857 * Caution: getting a boolean member of @options requires care.
3858 * When @options come from -blockdev or blockdev_add, members are
3859 * typed according to the QAPI schema, but when they come from
3860 * -drive, they're all QString.
3862 if (g_strcmp0(qdict_get_try_str(options
, BDRV_OPT_READ_ONLY
), "on") &&
3863 !qdict_get_try_bool(options
, BDRV_OPT_READ_ONLY
, false)) {
3864 flags
|= (BDRV_O_RDWR
| BDRV_O_ALLOW_RDWR
);
3866 flags
&= ~BDRV_O_RDWR
;
3869 if (flags
& BDRV_O_SNAPSHOT
) {
3870 snapshot_options
= qdict_new();
3871 bdrv_temp_snapshot_options(&snapshot_flags
, snapshot_options
,
3873 /* Let bdrv_backing_options() override "read-only" */
3874 qdict_del(options
, BDRV_OPT_READ_ONLY
);
3875 bdrv_inherited_options(BDRV_CHILD_COW
, true,
3876 &flags
, options
, flags
, options
);
3879 bs
->open_flags
= flags
;
3880 bs
->options
= options
;
3881 options
= qdict_clone_shallow(options
);
3883 /* Find the right image format driver */
3884 /* See cautionary note on accessing @options above */
3885 drvname
= qdict_get_try_str(options
, "driver");
3887 drv
= bdrv_find_format(drvname
);
3889 error_setg(errp
, "Unknown driver: '%s'", drvname
);
3894 assert(drvname
|| !(flags
& BDRV_O_PROTOCOL
));
3896 /* See cautionary note on accessing @options above */
3897 backing
= qdict_get_try_str(options
, "backing");
3898 if (qobject_to(QNull
, qdict_get(options
, "backing")) != NULL
||
3899 (backing
&& *backing
== '\0'))
3902 warn_report("Use of \"backing\": \"\" is deprecated; "
3903 "use \"backing\": null instead");
3905 flags
|= BDRV_O_NO_BACKING
;
3906 qdict_del(bs
->explicit_options
, "backing");
3907 qdict_del(bs
->options
, "backing");
3908 qdict_del(options
, "backing");
3911 /* Open image file without format layer. This BlockBackend is only used for
3912 * probing, the block drivers will do their own bdrv_open_child() for the
3913 * same BDS, which is why we put the node name back into options. */
3914 if ((flags
& BDRV_O_PROTOCOL
) == 0) {
3915 BlockDriverState
*file_bs
;
3917 file_bs
= bdrv_open_child_bs(filename
, options
, "file", bs
,
3918 &child_of_bds
, BDRV_CHILD_IMAGE
,
3923 if (file_bs
!= NULL
) {
3924 /* Not requesting BLK_PERM_CONSISTENT_READ because we're only
3925 * looking at the header to guess the image format. This works even
3926 * in cases where a guest would not see a consistent state. */
3927 file
= blk_new(bdrv_get_aio_context(file_bs
), 0, BLK_PERM_ALL
);
3928 blk_insert_bs(file
, file_bs
, &local_err
);
3929 bdrv_unref(file_bs
);
3934 qdict_put_str(options
, "file", bdrv_get_node_name(file_bs
));
3938 /* Image format probing */
3941 ret
= find_image_format(file
, filename
, &drv
, &local_err
);
3946 * This option update would logically belong in bdrv_fill_options(),
3947 * but we first need to open bs->file for the probing to work, while
3948 * opening bs->file already requires the (mostly) final set of options
3949 * so that cache mode etc. can be inherited.
3951 * Adding the driver later is somewhat ugly, but it's not an option
3952 * that would ever be inherited, so it's correct. We just need to make
3953 * sure to update both bs->options (which has the full effective
3954 * options for bs) and options (which has file.* already removed).
3956 qdict_put_str(bs
->options
, "driver", drv
->format_name
);
3957 qdict_put_str(options
, "driver", drv
->format_name
);
3959 error_setg(errp
, "Must specify either driver or file");
3963 /* BDRV_O_PROTOCOL must be set iff a protocol BDS is about to be created */
3964 assert(!!(flags
& BDRV_O_PROTOCOL
) == !!drv
->bdrv_file_open
);
3965 /* file must be NULL if a protocol BDS is about to be created
3966 * (the inverse results in an error message from bdrv_open_common()) */
3967 assert(!(flags
& BDRV_O_PROTOCOL
) || !file
);
3969 /* Open the image */
3970 ret
= bdrv_open_common(bs
, file
, options
, &local_err
);
3980 /* If there is a backing file, use it */
3981 if ((flags
& BDRV_O_NO_BACKING
) == 0) {
3982 ret
= bdrv_open_backing_file(bs
, options
, "backing", &local_err
);
3984 goto close_and_fail
;
3988 /* Remove all children options and references
3989 * from bs->options and bs->explicit_options */
3990 QLIST_FOREACH(child
, &bs
->children
, next
) {
3991 char *child_key_dot
;
3992 child_key_dot
= g_strdup_printf("%s.", child
->name
);
3993 qdict_extract_subqdict(bs
->explicit_options
, NULL
, child_key_dot
);
3994 qdict_extract_subqdict(bs
->options
, NULL
, child_key_dot
);
3995 qdict_del(bs
->explicit_options
, child
->name
);
3996 qdict_del(bs
->options
, child
->name
);
3997 g_free(child_key_dot
);
4000 /* Check if any unknown options were used */
4001 if (qdict_size(options
) != 0) {
4002 const QDictEntry
*entry
= qdict_first(options
);
4003 if (flags
& BDRV_O_PROTOCOL
) {
4004 error_setg(errp
, "Block protocol '%s' doesn't support the option "
4005 "'%s'", drv
->format_name
, entry
->key
);
4008 "Block format '%s' does not support the option '%s'",
4009 drv
->format_name
, entry
->key
);
4012 goto close_and_fail
;
4015 bdrv_parent_cb_change_media(bs
, true);
4017 qobject_unref(options
);
4020 /* For snapshot=on, create a temporary qcow2 overlay. bs points to the
4021 * temporary snapshot afterwards. */
4022 if (snapshot_flags
) {
4023 BlockDriverState
*snapshot_bs
;
4024 snapshot_bs
= bdrv_append_temp_snapshot(bs
, snapshot_flags
,
4025 snapshot_options
, &local_err
);
4026 snapshot_options
= NULL
;
4028 goto close_and_fail
;
4030 /* We are not going to return bs but the overlay on top of it
4031 * (snapshot_bs); thus, we have to drop the strong reference to bs
4032 * (which we obtained by calling bdrv_new()). bs will not be deleted,
4033 * though, because the overlay still has a reference to it. */
4042 qobject_unref(snapshot_options
);
4043 qobject_unref(bs
->explicit_options
);
4044 qobject_unref(bs
->options
);
4045 qobject_unref(options
);
4047 bs
->explicit_options
= NULL
;
4049 error_propagate(errp
, local_err
);
4054 qobject_unref(snapshot_options
);
4055 qobject_unref(options
);
4056 error_propagate(errp
, local_err
);
4060 BlockDriverState
*bdrv_open(const char *filename
, const char *reference
,
4061 QDict
*options
, int flags
, Error
**errp
)
4063 GLOBAL_STATE_CODE();
4065 return bdrv_open_inherit(filename
, reference
, options
, flags
, NULL
,
4069 /* Return true if the NULL-terminated @list contains @str */
4070 static bool is_str_in_list(const char *str
, const char *const *list
)
4074 for (i
= 0; list
[i
] != NULL
; i
++) {
4075 if (!strcmp(str
, list
[i
])) {
4084 * Check that every option set in @bs->options is also set in
4087 * Options listed in the common_options list and in
4088 * @bs->drv->mutable_opts are skipped.
4090 * Return 0 on success, otherwise return -EINVAL and set @errp.
4092 static int bdrv_reset_options_allowed(BlockDriverState
*bs
,
4093 const QDict
*new_opts
, Error
**errp
)
4095 const QDictEntry
*e
;
4096 /* These options are common to all block drivers and are handled
4097 * in bdrv_reopen_prepare() so they can be left out of @new_opts */
4098 const char *const common_options
[] = {
4099 "node-name", "discard", "cache.direct", "cache.no-flush",
4100 "read-only", "auto-read-only", "detect-zeroes", NULL
4103 for (e
= qdict_first(bs
->options
); e
; e
= qdict_next(bs
->options
, e
)) {
4104 if (!qdict_haskey(new_opts
, e
->key
) &&
4105 !is_str_in_list(e
->key
, common_options
) &&
4106 !is_str_in_list(e
->key
, bs
->drv
->mutable_opts
)) {
4107 error_setg(errp
, "Option '%s' cannot be reset "
4108 "to its default value", e
->key
);
4117 * Returns true if @child can be reached recursively from @bs
4119 static bool bdrv_recurse_has_child(BlockDriverState
*bs
,
4120 BlockDriverState
*child
)
4128 QLIST_FOREACH(c
, &bs
->children
, next
) {
4129 if (bdrv_recurse_has_child(c
->bs
, child
)) {
4138 * Adds a BlockDriverState to a simple queue for an atomic, transactional
4139 * reopen of multiple devices.
4141 * bs_queue can either be an existing BlockReopenQueue that has had QTAILQ_INIT
4142 * already performed, or alternatively may be NULL a new BlockReopenQueue will
4143 * be created and initialized. This newly created BlockReopenQueue should be
4144 * passed back in for subsequent calls that are intended to be of the same
4147 * bs is the BlockDriverState to add to the reopen queue.
4149 * options contains the changed options for the associated bs
4150 * (the BlockReopenQueue takes ownership)
4152 * flags contains the open flags for the associated bs
4154 * returns a pointer to bs_queue, which is either the newly allocated
4155 * bs_queue, or the existing bs_queue being used.
4157 * bs must be drained between bdrv_reopen_queue() and bdrv_reopen_multiple().
4159 static BlockReopenQueue
*bdrv_reopen_queue_child(BlockReopenQueue
*bs_queue
,
4160 BlockDriverState
*bs
,
4162 const BdrvChildClass
*klass
,
4164 bool parent_is_format
,
4165 QDict
*parent_options
,
4171 BlockReopenQueueEntry
*bs_entry
;
4173 QDict
*old_options
, *explicit_options
, *options_copy
;
4177 /* Make sure that the caller remembered to use a drained section. This is
4178 * important to avoid graph changes between the recursive queuing here and
4179 * bdrv_reopen_multiple(). */
4180 assert(bs
->quiesce_counter
> 0);
4181 GLOBAL_STATE_CODE();
4183 if (bs_queue
== NULL
) {
4184 bs_queue
= g_new0(BlockReopenQueue
, 1);
4185 QTAILQ_INIT(bs_queue
);
4189 options
= qdict_new();
4192 /* Check if this BlockDriverState is already in the queue */
4193 QTAILQ_FOREACH(bs_entry
, bs_queue
, entry
) {
4194 if (bs
== bs_entry
->state
.bs
) {
4200 * Precedence of options:
4201 * 1. Explicitly passed in options (highest)
4202 * 2. Retained from explicitly set options of bs
4203 * 3. Inherited from parent node
4204 * 4. Retained from effective options of bs
4207 /* Old explicitly set values (don't overwrite by inherited value) */
4208 if (bs_entry
|| keep_old_opts
) {
4209 old_options
= qdict_clone_shallow(bs_entry
?
4210 bs_entry
->state
.explicit_options
:
4211 bs
->explicit_options
);
4212 bdrv_join_options(bs
, options
, old_options
);
4213 qobject_unref(old_options
);
4216 explicit_options
= qdict_clone_shallow(options
);
4218 /* Inherit from parent node */
4219 if (parent_options
) {
4221 klass
->inherit_options(role
, parent_is_format
, &flags
, options
,
4222 parent_flags
, parent_options
);
4224 flags
= bdrv_get_flags(bs
);
4227 if (keep_old_opts
) {
4228 /* Old values are used for options that aren't set yet */
4229 old_options
= qdict_clone_shallow(bs
->options
);
4230 bdrv_join_options(bs
, options
, old_options
);
4231 qobject_unref(old_options
);
4234 /* We have the final set of options so let's update the flags */
4235 options_copy
= qdict_clone_shallow(options
);
4236 opts
= qemu_opts_create(&bdrv_runtime_opts
, NULL
, 0, &error_abort
);
4237 qemu_opts_absorb_qdict(opts
, options_copy
, NULL
);
4238 update_flags_from_options(&flags
, opts
);
4239 qemu_opts_del(opts
);
4240 qobject_unref(options_copy
);
4242 /* bdrv_open_inherit() sets and clears some additional flags internally */
4243 flags
&= ~BDRV_O_PROTOCOL
;
4244 if (flags
& BDRV_O_RDWR
) {
4245 flags
|= BDRV_O_ALLOW_RDWR
;
4249 bs_entry
= g_new0(BlockReopenQueueEntry
, 1);
4250 QTAILQ_INSERT_TAIL(bs_queue
, bs_entry
, entry
);
4252 qobject_unref(bs_entry
->state
.options
);
4253 qobject_unref(bs_entry
->state
.explicit_options
);
4256 bs_entry
->state
.bs
= bs
;
4257 bs_entry
->state
.options
= options
;
4258 bs_entry
->state
.explicit_options
= explicit_options
;
4259 bs_entry
->state
.flags
= flags
;
4262 * If keep_old_opts is false then it means that unspecified
4263 * options must be reset to their original value. We don't allow
4264 * resetting 'backing' but we need to know if the option is
4265 * missing in order to decide if we have to return an error.
4267 if (!keep_old_opts
) {
4268 bs_entry
->state
.backing_missing
=
4269 !qdict_haskey(options
, "backing") &&
4270 !qdict_haskey(options
, "backing.driver");
4273 QLIST_FOREACH(child
, &bs
->children
, next
) {
4274 QDict
*new_child_options
= NULL
;
4275 bool child_keep_old
= keep_old_opts
;
4277 /* reopen can only change the options of block devices that were
4278 * implicitly created and inherited options. For other (referenced)
4279 * block devices, a syntax like "backing.foo" results in an error. */
4280 if (child
->bs
->inherits_from
!= bs
) {
4284 /* Check if the options contain a child reference */
4285 if (qdict_haskey(options
, child
->name
)) {
4286 const char *childref
= qdict_get_try_str(options
, child
->name
);
4288 * The current child must not be reopened if the child
4289 * reference is null or points to a different node.
4291 if (g_strcmp0(childref
, child
->bs
->node_name
)) {
4295 * If the child reference points to the current child then
4296 * reopen it with its existing set of options (note that
4297 * it can still inherit new options from the parent).
4299 child_keep_old
= true;
4301 /* Extract child options ("child-name.*") */
4302 char *child_key_dot
= g_strdup_printf("%s.", child
->name
);
4303 qdict_extract_subqdict(explicit_options
, NULL
, child_key_dot
);
4304 qdict_extract_subqdict(options
, &new_child_options
, child_key_dot
);
4305 g_free(child_key_dot
);
4308 bdrv_reopen_queue_child(bs_queue
, child
->bs
, new_child_options
,
4309 child
->klass
, child
->role
, bs
->drv
->is_format
,
4310 options
, flags
, child_keep_old
);
4316 BlockReopenQueue
*bdrv_reopen_queue(BlockReopenQueue
*bs_queue
,
4317 BlockDriverState
*bs
,
4318 QDict
*options
, bool keep_old_opts
)
4320 GLOBAL_STATE_CODE();
4322 return bdrv_reopen_queue_child(bs_queue
, bs
, options
, NULL
, 0, false,
4323 NULL
, 0, keep_old_opts
);
4326 void bdrv_reopen_queue_free(BlockReopenQueue
*bs_queue
)
4328 GLOBAL_STATE_CODE();
4330 BlockReopenQueueEntry
*bs_entry
, *next
;
4331 QTAILQ_FOREACH_SAFE(bs_entry
, bs_queue
, entry
, next
) {
4332 qobject_unref(bs_entry
->state
.explicit_options
);
4333 qobject_unref(bs_entry
->state
.options
);
4341 * Reopen multiple BlockDriverStates atomically & transactionally.
4343 * The queue passed in (bs_queue) must have been built up previous
4344 * via bdrv_reopen_queue().
4346 * Reopens all BDS specified in the queue, with the appropriate
4347 * flags. All devices are prepared for reopen, and failure of any
4348 * device will cause all device changes to be abandoned, and intermediate
4351 * If all devices prepare successfully, then the changes are committed
4354 * All affected nodes must be drained between bdrv_reopen_queue() and
4355 * bdrv_reopen_multiple().
4357 * To be called from the main thread, with all other AioContexts unlocked.
4359 int bdrv_reopen_multiple(BlockReopenQueue
*bs_queue
, Error
**errp
)
4362 BlockReopenQueueEntry
*bs_entry
, *next
;
4364 Transaction
*tran
= tran_new();
4365 g_autoptr(GHashTable
) found
= NULL
;
4366 g_autoptr(GSList
) refresh_list
= NULL
;
4368 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
4369 assert(bs_queue
!= NULL
);
4370 GLOBAL_STATE_CODE();
4372 QTAILQ_FOREACH(bs_entry
, bs_queue
, entry
) {
4373 ctx
= bdrv_get_aio_context(bs_entry
->state
.bs
);
4374 aio_context_acquire(ctx
);
4375 ret
= bdrv_flush(bs_entry
->state
.bs
);
4376 aio_context_release(ctx
);
4378 error_setg_errno(errp
, -ret
, "Error flushing drive");
4383 QTAILQ_FOREACH(bs_entry
, bs_queue
, entry
) {
4384 assert(bs_entry
->state
.bs
->quiesce_counter
> 0);
4385 ctx
= bdrv_get_aio_context(bs_entry
->state
.bs
);
4386 aio_context_acquire(ctx
);
4387 ret
= bdrv_reopen_prepare(&bs_entry
->state
, bs_queue
, tran
, errp
);
4388 aio_context_release(ctx
);
4392 bs_entry
->prepared
= true;
4395 found
= g_hash_table_new(NULL
, NULL
);
4396 QTAILQ_FOREACH(bs_entry
, bs_queue
, entry
) {
4397 BDRVReopenState
*state
= &bs_entry
->state
;
4399 refresh_list
= bdrv_topological_dfs(refresh_list
, found
, state
->bs
);
4400 if (state
->old_backing_bs
) {
4401 refresh_list
= bdrv_topological_dfs(refresh_list
, found
,
4402 state
->old_backing_bs
);
4404 if (state
->old_file_bs
) {
4405 refresh_list
= bdrv_topological_dfs(refresh_list
, found
,
4406 state
->old_file_bs
);
4411 * Note that file-posix driver rely on permission update done during reopen
4412 * (even if no permission changed), because it wants "new" permissions for
4413 * reconfiguring the fd and that's why it does it in raw_check_perm(), not
4414 * in raw_reopen_prepare() which is called with "old" permissions.
4416 ret
= bdrv_list_refresh_perms(refresh_list
, bs_queue
, tran
, errp
);
4422 * If we reach this point, we have success and just need to apply the
4425 * Reverse order is used to comfort qcow2 driver: on commit it need to write
4426 * IN_USE flag to the image, to mark bitmaps in the image as invalid. But
4427 * children are usually goes after parents in reopen-queue, so go from last
4430 QTAILQ_FOREACH_REVERSE(bs_entry
, bs_queue
, entry
) {
4431 ctx
= bdrv_get_aio_context(bs_entry
->state
.bs
);
4432 aio_context_acquire(ctx
);
4433 bdrv_reopen_commit(&bs_entry
->state
);
4434 aio_context_release(ctx
);
4439 QTAILQ_FOREACH_REVERSE(bs_entry
, bs_queue
, entry
) {
4440 BlockDriverState
*bs
= bs_entry
->state
.bs
;
4442 if (bs
->drv
->bdrv_reopen_commit_post
) {
4443 ctx
= bdrv_get_aio_context(bs
);
4444 aio_context_acquire(ctx
);
4445 bs
->drv
->bdrv_reopen_commit_post(&bs_entry
->state
);
4446 aio_context_release(ctx
);
4455 QTAILQ_FOREACH_SAFE(bs_entry
, bs_queue
, entry
, next
) {
4456 if (bs_entry
->prepared
) {
4457 ctx
= bdrv_get_aio_context(bs_entry
->state
.bs
);
4458 aio_context_acquire(ctx
);
4459 bdrv_reopen_abort(&bs_entry
->state
);
4460 aio_context_release(ctx
);
4465 bdrv_reopen_queue_free(bs_queue
);
4470 int bdrv_reopen(BlockDriverState
*bs
, QDict
*opts
, bool keep_old_opts
,
4473 AioContext
*ctx
= bdrv_get_aio_context(bs
);
4474 BlockReopenQueue
*queue
;
4477 GLOBAL_STATE_CODE();
4479 bdrv_subtree_drained_begin(bs
);
4480 if (ctx
!= qemu_get_aio_context()) {
4481 aio_context_release(ctx
);
4484 queue
= bdrv_reopen_queue(NULL
, bs
, opts
, keep_old_opts
);
4485 ret
= bdrv_reopen_multiple(queue
, errp
);
4487 if (ctx
!= qemu_get_aio_context()) {
4488 aio_context_acquire(ctx
);
4490 bdrv_subtree_drained_end(bs
);
4495 int bdrv_reopen_set_read_only(BlockDriverState
*bs
, bool read_only
,
4498 QDict
*opts
= qdict_new();
4500 GLOBAL_STATE_CODE();
4502 qdict_put_bool(opts
, BDRV_OPT_READ_ONLY
, read_only
);
4504 return bdrv_reopen(bs
, opts
, true, errp
);
4508 * Take a BDRVReopenState and check if the value of 'backing' in the
4509 * reopen_state->options QDict is valid or not.
4511 * If 'backing' is missing from the QDict then return 0.
4513 * If 'backing' contains the node name of the backing file of
4514 * reopen_state->bs then return 0.
4516 * If 'backing' contains a different node name (or is null) then check
4517 * whether the current backing file can be replaced with the new one.
4518 * If that's the case then reopen_state->replace_backing_bs is set to
4519 * true and reopen_state->new_backing_bs contains a pointer to the new
4520 * backing BlockDriverState (or NULL).
4522 * Return 0 on success, otherwise return < 0 and set @errp.
4524 static int bdrv_reopen_parse_file_or_backing(BDRVReopenState
*reopen_state
,
4525 bool is_backing
, Transaction
*tran
,
4528 BlockDriverState
*bs
= reopen_state
->bs
;
4529 BlockDriverState
*new_child_bs
;
4530 BlockDriverState
*old_child_bs
= is_backing
? child_bs(bs
->backing
) :
4532 const char *child_name
= is_backing
? "backing" : "file";
4536 GLOBAL_STATE_CODE();
4538 value
= qdict_get(reopen_state
->options
, child_name
);
4539 if (value
== NULL
) {
4543 switch (qobject_type(value
)) {
4545 assert(is_backing
); /* The 'file' option does not allow a null value */
4546 new_child_bs
= NULL
;
4549 str
= qstring_get_str(qobject_to(QString
, value
));
4550 new_child_bs
= bdrv_lookup_bs(NULL
, str
, errp
);
4551 if (new_child_bs
== NULL
) {
4553 } else if (bdrv_recurse_has_child(new_child_bs
, bs
)) {
4554 error_setg(errp
, "Making '%s' a %s child of '%s' would create a "
4555 "cycle", str
, child_name
, bs
->node_name
);
4561 * The options QDict has been flattened, so 'backing' and 'file'
4562 * do not allow any other data type here.
4564 g_assert_not_reached();
4567 if (old_child_bs
== new_child_bs
) {
4572 if (bdrv_skip_implicit_filters(old_child_bs
) == new_child_bs
) {
4576 if (old_child_bs
->implicit
) {
4577 error_setg(errp
, "Cannot replace implicit %s child of %s",
4578 child_name
, bs
->node_name
);
4583 if (bs
->drv
->is_filter
&& !old_child_bs
) {
4585 * Filters always have a file or a backing child, so we are trying to
4586 * change wrong child
4588 error_setg(errp
, "'%s' is a %s filter node that does not support a "
4589 "%s child", bs
->node_name
, bs
->drv
->format_name
, child_name
);
4594 reopen_state
->old_backing_bs
= old_child_bs
;
4596 reopen_state
->old_file_bs
= old_child_bs
;
4599 return bdrv_set_file_or_backing_noperm(bs
, new_child_bs
, is_backing
,
4604 * Prepares a BlockDriverState for reopen. All changes are staged in the
4605 * 'opaque' field of the BDRVReopenState, which is used and allocated by
4606 * the block driver layer .bdrv_reopen_prepare()
4608 * bs is the BlockDriverState to reopen
4609 * flags are the new open flags
4610 * queue is the reopen queue
4612 * Returns 0 on success, non-zero on error. On error errp will be set
4615 * On failure, bdrv_reopen_abort() will be called to clean up any data.
4616 * It is the responsibility of the caller to then call the abort() or
4617 * commit() for any other BDS that have been left in a prepare() state
4620 static int bdrv_reopen_prepare(BDRVReopenState
*reopen_state
,
4621 BlockReopenQueue
*queue
,
4622 Transaction
*change_child_tran
, Error
**errp
)
4626 Error
*local_err
= NULL
;
4629 QDict
*orig_reopen_opts
;
4630 char *discard
= NULL
;
4632 bool drv_prepared
= false;
4634 assert(reopen_state
!= NULL
);
4635 assert(reopen_state
->bs
->drv
!= NULL
);
4636 GLOBAL_STATE_CODE();
4637 drv
= reopen_state
->bs
->drv
;
4639 /* This function and each driver's bdrv_reopen_prepare() remove
4640 * entries from reopen_state->options as they are processed, so
4641 * we need to make a copy of the original QDict. */
4642 orig_reopen_opts
= qdict_clone_shallow(reopen_state
->options
);
4644 /* Process generic block layer options */
4645 opts
= qemu_opts_create(&bdrv_runtime_opts
, NULL
, 0, &error_abort
);
4646 if (!qemu_opts_absorb_qdict(opts
, reopen_state
->options
, errp
)) {
4651 /* This was already called in bdrv_reopen_queue_child() so the flags
4652 * are up-to-date. This time we simply want to remove the options from
4653 * QemuOpts in order to indicate that they have been processed. */
4654 old_flags
= reopen_state
->flags
;
4655 update_flags_from_options(&reopen_state
->flags
, opts
);
4656 assert(old_flags
== reopen_state
->flags
);
4658 discard
= qemu_opt_get_del(opts
, BDRV_OPT_DISCARD
);
4659 if (discard
!= NULL
) {
4660 if (bdrv_parse_discard_flags(discard
, &reopen_state
->flags
) != 0) {
4661 error_setg(errp
, "Invalid discard option");
4667 reopen_state
->detect_zeroes
=
4668 bdrv_parse_detect_zeroes(opts
, reopen_state
->flags
, &local_err
);
4670 error_propagate(errp
, local_err
);
4675 /* All other options (including node-name and driver) must be unchanged.
4676 * Put them back into the QDict, so that they are checked at the end
4677 * of this function. */
4678 qemu_opts_to_qdict(opts
, reopen_state
->options
);
4680 /* If we are to stay read-only, do not allow permission change
4681 * to r/w. Attempting to set to r/w may fail if either BDRV_O_ALLOW_RDWR is
4682 * not set, or if the BDS still has copy_on_read enabled */
4683 read_only
= !(reopen_state
->flags
& BDRV_O_RDWR
);
4684 ret
= bdrv_can_set_read_only(reopen_state
->bs
, read_only
, true, &local_err
);
4686 error_propagate(errp
, local_err
);
4690 if (drv
->bdrv_reopen_prepare
) {
4692 * If a driver-specific option is missing, it means that we
4693 * should reset it to its default value.
4694 * But not all options allow that, so we need to check it first.
4696 ret
= bdrv_reset_options_allowed(reopen_state
->bs
,
4697 reopen_state
->options
, errp
);
4702 ret
= drv
->bdrv_reopen_prepare(reopen_state
, queue
, &local_err
);
4704 if (local_err
!= NULL
) {
4705 error_propagate(errp
, local_err
);
4707 bdrv_refresh_filename(reopen_state
->bs
);
4708 error_setg(errp
, "failed while preparing to reopen image '%s'",
4709 reopen_state
->bs
->filename
);
4714 /* It is currently mandatory to have a bdrv_reopen_prepare()
4715 * handler for each supported drv. */
4716 error_setg(errp
, "Block format '%s' used by node '%s' "
4717 "does not support reopening files", drv
->format_name
,
4718 bdrv_get_device_or_node_name(reopen_state
->bs
));
4723 drv_prepared
= true;
4726 * We must provide the 'backing' option if the BDS has a backing
4727 * file or if the image file has a backing file name as part of
4728 * its metadata. Otherwise the 'backing' option can be omitted.
4730 if (drv
->supports_backing
&& reopen_state
->backing_missing
&&
4731 (reopen_state
->bs
->backing
|| reopen_state
->bs
->backing_file
[0])) {
4732 error_setg(errp
, "backing is missing for '%s'",
4733 reopen_state
->bs
->node_name
);
4739 * Allow changing the 'backing' option. The new value can be
4740 * either a reference to an existing node (using its node name)
4741 * or NULL to simply detach the current backing file.
4743 ret
= bdrv_reopen_parse_file_or_backing(reopen_state
, true,
4744 change_child_tran
, errp
);
4748 qdict_del(reopen_state
->options
, "backing");
4750 /* Allow changing the 'file' option. In this case NULL is not allowed */
4751 ret
= bdrv_reopen_parse_file_or_backing(reopen_state
, false,
4752 change_child_tran
, errp
);
4756 qdict_del(reopen_state
->options
, "file");
4758 /* Options that are not handled are only okay if they are unchanged
4759 * compared to the old state. It is expected that some options are only
4760 * used for the initial open, but not reopen (e.g. filename) */
4761 if (qdict_size(reopen_state
->options
)) {
4762 const QDictEntry
*entry
= qdict_first(reopen_state
->options
);
4765 QObject
*new = entry
->value
;
4766 QObject
*old
= qdict_get(reopen_state
->bs
->options
, entry
->key
);
4768 /* Allow child references (child_name=node_name) as long as they
4769 * point to the current child (i.e. everything stays the same). */
4770 if (qobject_type(new) == QTYPE_QSTRING
) {
4772 QLIST_FOREACH(child
, &reopen_state
->bs
->children
, next
) {
4773 if (!strcmp(child
->name
, entry
->key
)) {
4779 if (!strcmp(child
->bs
->node_name
,
4780 qstring_get_str(qobject_to(QString
, new)))) {
4781 continue; /* Found child with this name, skip option */
4787 * TODO: When using -drive to specify blockdev options, all values
4788 * will be strings; however, when using -blockdev, blockdev-add or
4789 * filenames using the json:{} pseudo-protocol, they will be
4791 * In contrast, reopening options are (currently) always strings
4792 * (because you can only specify them through qemu-io; all other
4793 * callers do not specify any options).
4794 * Therefore, when using anything other than -drive to create a BDS,
4795 * this cannot detect non-string options as unchanged, because
4796 * qobject_is_equal() always returns false for objects of different
4797 * type. In the future, this should be remedied by correctly typing
4798 * all options. For now, this is not too big of an issue because
4799 * the user can simply omit options which cannot be changed anyway,
4800 * so they will stay unchanged.
4802 if (!qobject_is_equal(new, old
)) {
4803 error_setg(errp
, "Cannot change the option '%s'", entry
->key
);
4807 } while ((entry
= qdict_next(reopen_state
->options
, entry
)));
4812 /* Restore the original reopen_state->options QDict */
4813 qobject_unref(reopen_state
->options
);
4814 reopen_state
->options
= qobject_ref(orig_reopen_opts
);
4817 if (ret
< 0 && drv_prepared
) {
4818 /* drv->bdrv_reopen_prepare() has succeeded, so we need to
4819 * call drv->bdrv_reopen_abort() before signaling an error
4820 * (bdrv_reopen_multiple() will not call bdrv_reopen_abort()
4821 * when the respective bdrv_reopen_prepare() has failed) */
4822 if (drv
->bdrv_reopen_abort
) {
4823 drv
->bdrv_reopen_abort(reopen_state
);
4826 qemu_opts_del(opts
);
4827 qobject_unref(orig_reopen_opts
);
4833 * Takes the staged changes for the reopen from bdrv_reopen_prepare(), and
4834 * makes them final by swapping the staging BlockDriverState contents into
4835 * the active BlockDriverState contents.
4837 static void bdrv_reopen_commit(BDRVReopenState
*reopen_state
)
4840 BlockDriverState
*bs
;
4843 assert(reopen_state
!= NULL
);
4844 bs
= reopen_state
->bs
;
4846 assert(drv
!= NULL
);
4847 GLOBAL_STATE_CODE();
4849 /* If there are any driver level actions to take */
4850 if (drv
->bdrv_reopen_commit
) {
4851 drv
->bdrv_reopen_commit(reopen_state
);
4854 /* set BDS specific flags now */
4855 qobject_unref(bs
->explicit_options
);
4856 qobject_unref(bs
->options
);
4857 qobject_ref(reopen_state
->explicit_options
);
4858 qobject_ref(reopen_state
->options
);
4860 bs
->explicit_options
= reopen_state
->explicit_options
;
4861 bs
->options
= reopen_state
->options
;
4862 bs
->open_flags
= reopen_state
->flags
;
4863 bs
->detect_zeroes
= reopen_state
->detect_zeroes
;
4865 /* Remove child references from bs->options and bs->explicit_options.
4866 * Child options were already removed in bdrv_reopen_queue_child() */
4867 QLIST_FOREACH(child
, &bs
->children
, next
) {
4868 qdict_del(bs
->explicit_options
, child
->name
);
4869 qdict_del(bs
->options
, child
->name
);
4871 /* backing is probably removed, so it's not handled by previous loop */
4872 qdict_del(bs
->explicit_options
, "backing");
4873 qdict_del(bs
->options
, "backing");
4875 bdrv_refresh_limits(bs
, NULL
, NULL
);
4879 * Abort the reopen, and delete and free the staged changes in
4882 static void bdrv_reopen_abort(BDRVReopenState
*reopen_state
)
4886 assert(reopen_state
!= NULL
);
4887 drv
= reopen_state
->bs
->drv
;
4888 assert(drv
!= NULL
);
4889 GLOBAL_STATE_CODE();
4891 if (drv
->bdrv_reopen_abort
) {
4892 drv
->bdrv_reopen_abort(reopen_state
);
4897 static void bdrv_close(BlockDriverState
*bs
)
4899 BdrvAioNotifier
*ban
, *ban_next
;
4900 BdrvChild
*child
, *next
;
4902 GLOBAL_STATE_CODE();
4903 assert(!bs
->refcnt
);
4905 bdrv_drained_begin(bs
); /* complete I/O */
4907 bdrv_drain(bs
); /* in case flush left pending I/O */
4910 if (bs
->drv
->bdrv_close
) {
4911 /* Must unfreeze all children, so bdrv_unref_child() works */
4912 bs
->drv
->bdrv_close(bs
);
4917 QLIST_FOREACH_SAFE(child
, &bs
->children
, next
, next
) {
4918 bdrv_unref_child(bs
, child
);
4921 assert(!bs
->backing
);
4925 qatomic_set(&bs
->copy_on_read
, 0);
4926 bs
->backing_file
[0] = '\0';
4927 bs
->backing_format
[0] = '\0';
4928 bs
->total_sectors
= 0;
4929 bs
->encrypted
= false;
4931 qobject_unref(bs
->options
);
4932 qobject_unref(bs
->explicit_options
);
4934 bs
->explicit_options
= NULL
;
4935 qobject_unref(bs
->full_open_options
);
4936 bs
->full_open_options
= NULL
;
4937 g_free(bs
->block_status_cache
);
4938 bs
->block_status_cache
= NULL
;
4940 bdrv_release_named_dirty_bitmaps(bs
);
4941 assert(QLIST_EMPTY(&bs
->dirty_bitmaps
));
4943 QLIST_FOREACH_SAFE(ban
, &bs
->aio_notifiers
, list
, ban_next
) {
4946 QLIST_INIT(&bs
->aio_notifiers
);
4947 bdrv_drained_end(bs
);
4950 * If we're still inside some bdrv_drain_all_begin()/end() sections, end
4951 * them now since this BDS won't exist anymore when bdrv_drain_all_end()
4954 if (bs
->quiesce_counter
) {
4955 bdrv_drain_all_end_quiesce(bs
);
4959 void bdrv_close_all(void)
4961 GLOBAL_STATE_CODE();
4962 assert(job_next(NULL
) == NULL
);
4964 /* Drop references from requests still in flight, such as canceled block
4965 * jobs whose AIO context has not been polled yet */
4968 blk_remove_all_bs();
4969 blockdev_close_all_bdrv_states();
4971 assert(QTAILQ_EMPTY(&all_bdrv_states
));
4974 static bool should_update_child(BdrvChild
*c
, BlockDriverState
*to
)
4980 if (c
->klass
->stay_at_node
) {
4984 /* If the child @c belongs to the BDS @to, replacing the current
4985 * c->bs by @to would mean to create a loop.
4987 * Such a case occurs when appending a BDS to a backing chain.
4988 * For instance, imagine the following chain:
4990 * guest device -> node A -> further backing chain...
4992 * Now we create a new BDS B which we want to put on top of this
4993 * chain, so we first attach A as its backing node:
4998 * guest device -> node A -> further backing chain...
5000 * Finally we want to replace A by B. When doing that, we want to
5001 * replace all pointers to A by pointers to B -- except for the
5002 * pointer from B because (1) that would create a loop, and (2)
5003 * that pointer should simply stay intact:
5005 * guest device -> node B
5008 * node A -> further backing chain...
5010 * In general, when replacing a node A (c->bs) by a node B (@to),
5011 * if A is a child of B, that means we cannot replace A by B there
5012 * because that would create a loop. Silently detaching A from B
5013 * is also not really an option. So overall just leaving A in
5014 * place there is the most sensible choice.
5016 * We would also create a loop in any cases where @c is only
5017 * indirectly referenced by @to. Prevent this by returning false
5018 * if @c is found (by breadth-first search) anywhere in the whole
5023 found
= g_hash_table_new(NULL
, NULL
);
5024 g_hash_table_add(found
, to
);
5025 queue
= g_queue_new();
5026 g_queue_push_tail(queue
, to
);
5028 while (!g_queue_is_empty(queue
)) {
5029 BlockDriverState
*v
= g_queue_pop_head(queue
);
5032 QLIST_FOREACH(c2
, &v
->children
, next
) {
5038 if (g_hash_table_contains(found
, c2
->bs
)) {
5042 g_queue_push_tail(queue
, c2
->bs
);
5043 g_hash_table_add(found
, c2
->bs
);
5047 g_queue_free(queue
);
5048 g_hash_table_destroy(found
);
5053 static void bdrv_remove_child_commit(void *opaque
)
5055 GLOBAL_STATE_CODE();
5056 bdrv_child_free(opaque
);
5059 static TransactionActionDrv bdrv_remove_child_drv
= {
5060 .commit
= bdrv_remove_child_commit
,
5063 /* Function doesn't update permissions, caller is responsible for this. */
5064 static void bdrv_remove_child(BdrvChild
*child
, Transaction
*tran
)
5071 bdrv_replace_child_tran(child
, NULL
, tran
);
5074 tran_add(tran
, &bdrv_remove_child_drv
, child
);
5078 * A function to remove backing-chain child of @bs if exists: cow child for
5079 * format nodes (always .backing) and filter child for filters (may be .file or
5082 static void bdrv_remove_filter_or_cow_child(BlockDriverState
*bs
,
5085 bdrv_remove_child(bdrv_filter_or_cow_child(bs
), tran
);
5088 static int bdrv_replace_node_noperm(BlockDriverState
*from
,
5089 BlockDriverState
*to
,
5090 bool auto_skip
, Transaction
*tran
,
5093 BdrvChild
*c
, *next
;
5095 GLOBAL_STATE_CODE();
5097 QLIST_FOREACH_SAFE(c
, &from
->parents
, next_parent
, next
) {
5098 assert(c
->bs
== from
);
5099 if (!should_update_child(c
, to
)) {
5103 error_setg(errp
, "Should not change '%s' link to '%s'",
5104 c
->name
, from
->node_name
);
5108 error_setg(errp
, "Cannot change '%s' link to '%s'",
5109 c
->name
, from
->node_name
);
5112 bdrv_replace_child_tran(c
, to
, tran
);
5119 * With auto_skip=true bdrv_replace_node_common skips updating from parents
5120 * if it creates a parent-child relation loop or if parent is block-job.
5122 * With auto_skip=false the error is returned if from has a parent which should
5125 * With @detach_subchain=true @to must be in a backing chain of @from. In this
5126 * case backing link of the cow-parent of @to is removed.
5128 static int bdrv_replace_node_common(BlockDriverState
*from
,
5129 BlockDriverState
*to
,
5130 bool auto_skip
, bool detach_subchain
,
5133 Transaction
*tran
= tran_new();
5134 g_autoptr(GHashTable
) found
= NULL
;
5135 g_autoptr(GSList
) refresh_list
= NULL
;
5136 BlockDriverState
*to_cow_parent
= NULL
;
5139 GLOBAL_STATE_CODE();
5141 if (detach_subchain
) {
5142 assert(bdrv_chain_contains(from
, to
));
5144 for (to_cow_parent
= from
;
5145 bdrv_filter_or_cow_bs(to_cow_parent
) != to
;
5146 to_cow_parent
= bdrv_filter_or_cow_bs(to_cow_parent
))
5152 /* Make sure that @from doesn't go away until we have successfully attached
5153 * all of its parents to @to. */
5156 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
5157 assert(bdrv_get_aio_context(from
) == bdrv_get_aio_context(to
));
5158 bdrv_drained_begin(from
);
5161 * Do the replacement without permission update.
5162 * Replacement may influence the permissions, we should calculate new
5163 * permissions based on new graph. If we fail, we'll roll-back the
5166 ret
= bdrv_replace_node_noperm(from
, to
, auto_skip
, tran
, errp
);
5171 if (detach_subchain
) {
5172 bdrv_remove_filter_or_cow_child(to_cow_parent
, tran
);
5175 found
= g_hash_table_new(NULL
, NULL
);
5177 refresh_list
= bdrv_topological_dfs(refresh_list
, found
, to
);
5178 refresh_list
= bdrv_topological_dfs(refresh_list
, found
, from
);
5180 ret
= bdrv_list_refresh_perms(refresh_list
, NULL
, tran
, errp
);
5188 tran_finalize(tran
, ret
);
5190 bdrv_drained_end(from
);
5196 int bdrv_replace_node(BlockDriverState
*from
, BlockDriverState
*to
,
5199 GLOBAL_STATE_CODE();
5201 return bdrv_replace_node_common(from
, to
, true, false, errp
);
5204 int bdrv_drop_filter(BlockDriverState
*bs
, Error
**errp
)
5206 GLOBAL_STATE_CODE();
5208 return bdrv_replace_node_common(bs
, bdrv_filter_or_cow_bs(bs
), true, true,
5213 * Add new bs contents at the top of an image chain while the chain is
5214 * live, while keeping required fields on the top layer.
5216 * This will modify the BlockDriverState fields, and swap contents
5217 * between bs_new and bs_top. Both bs_new and bs_top are modified.
5219 * bs_new must not be attached to a BlockBackend and must not have backing
5222 * This function does not create any image files.
5224 int bdrv_append(BlockDriverState
*bs_new
, BlockDriverState
*bs_top
,
5229 Transaction
*tran
= tran_new();
5231 GLOBAL_STATE_CODE();
5233 assert(!bs_new
->backing
);
5235 child
= bdrv_attach_child_noperm(bs_new
, bs_top
, "backing",
5236 &child_of_bds
, bdrv_backing_role(bs_new
),
5243 ret
= bdrv_replace_node_noperm(bs_top
, bs_new
, true, tran
, errp
);
5248 ret
= bdrv_refresh_perms(bs_new
, errp
);
5250 tran_finalize(tran
, ret
);
5252 bdrv_refresh_limits(bs_top
, NULL
, NULL
);
5257 /* Not for empty child */
5258 int bdrv_replace_child_bs(BdrvChild
*child
, BlockDriverState
*new_bs
,
5262 Transaction
*tran
= tran_new();
5263 g_autoptr(GHashTable
) found
= NULL
;
5264 g_autoptr(GSList
) refresh_list
= NULL
;
5265 BlockDriverState
*old_bs
= child
->bs
;
5267 GLOBAL_STATE_CODE();
5270 bdrv_drained_begin(old_bs
);
5271 bdrv_drained_begin(new_bs
);
5273 bdrv_replace_child_tran(child
, new_bs
, tran
);
5275 found
= g_hash_table_new(NULL
, NULL
);
5276 refresh_list
= bdrv_topological_dfs(refresh_list
, found
, old_bs
);
5277 refresh_list
= bdrv_topological_dfs(refresh_list
, found
, new_bs
);
5279 ret
= bdrv_list_refresh_perms(refresh_list
, NULL
, tran
, errp
);
5281 tran_finalize(tran
, ret
);
5283 bdrv_drained_end(old_bs
);
5284 bdrv_drained_end(new_bs
);
5290 static void bdrv_delete(BlockDriverState
*bs
)
5292 assert(bdrv_op_blocker_is_empty(bs
));
5293 assert(!bs
->refcnt
);
5294 GLOBAL_STATE_CODE();
5296 /* remove from list, if necessary */
5297 if (bs
->node_name
[0] != '\0') {
5298 QTAILQ_REMOVE(&graph_bdrv_states
, bs
, node_list
);
5300 QTAILQ_REMOVE(&all_bdrv_states
, bs
, bs_list
);
5309 * Replace @bs by newly created block node.
5311 * @options is a QDict of options to pass to the block drivers, or NULL for an
5312 * empty set of options. The reference to the QDict belongs to the block layer
5313 * after the call (even on failure), so if the caller intends to reuse the
5314 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
5316 BlockDriverState
*bdrv_insert_node(BlockDriverState
*bs
, QDict
*options
,
5317 int flags
, Error
**errp
)
5321 BlockDriverState
*new_node_bs
= NULL
;
5322 const char *drvname
, *node_name
;
5325 drvname
= qdict_get_try_str(options
, "driver");
5327 error_setg(errp
, "driver is not specified");
5331 drv
= bdrv_find_format(drvname
);
5333 error_setg(errp
, "Unknown driver: '%s'", drvname
);
5337 node_name
= qdict_get_try_str(options
, "node-name");
5339 GLOBAL_STATE_CODE();
5341 new_node_bs
= bdrv_new_open_driver_opts(drv
, node_name
, options
, flags
,
5343 options
= NULL
; /* bdrv_new_open_driver() eats options */
5345 error_prepend(errp
, "Could not create node: ");
5349 bdrv_drained_begin(bs
);
5350 ret
= bdrv_replace_node(bs
, new_node_bs
, errp
);
5351 bdrv_drained_end(bs
);
5354 error_prepend(errp
, "Could not replace node: ");
5361 qobject_unref(options
);
5362 bdrv_unref(new_node_bs
);
5367 * Run consistency checks on an image
5369 * Returns 0 if the check could be completed (it doesn't mean that the image is
5370 * free of errors) or -errno when an internal error occurred. The results of the
5371 * check are stored in res.
5373 int coroutine_fn
bdrv_co_check(BlockDriverState
*bs
,
5374 BdrvCheckResult
*res
, BdrvCheckMode fix
)
5377 if (bs
->drv
== NULL
) {
5380 if (bs
->drv
->bdrv_co_check
== NULL
) {
5384 memset(res
, 0, sizeof(*res
));
5385 return bs
->drv
->bdrv_co_check(bs
, res
, fix
);
5391 * -EINVAL - backing format specified, but no file
5392 * -ENOSPC - can't update the backing file because no space is left in the
5394 * -ENOTSUP - format driver doesn't support changing the backing file
5396 int bdrv_change_backing_file(BlockDriverState
*bs
, const char *backing_file
,
5397 const char *backing_fmt
, bool require
)
5399 BlockDriver
*drv
= bs
->drv
;
5402 GLOBAL_STATE_CODE();
5408 /* Backing file format doesn't make sense without a backing file */
5409 if (backing_fmt
&& !backing_file
) {
5413 if (require
&& backing_file
&& !backing_fmt
) {
5417 if (drv
->bdrv_change_backing_file
!= NULL
) {
5418 ret
= drv
->bdrv_change_backing_file(bs
, backing_file
, backing_fmt
);
5424 pstrcpy(bs
->backing_file
, sizeof(bs
->backing_file
), backing_file
?: "");
5425 pstrcpy(bs
->backing_format
, sizeof(bs
->backing_format
), backing_fmt
?: "");
5426 pstrcpy(bs
->auto_backing_file
, sizeof(bs
->auto_backing_file
),
5427 backing_file
?: "");
5433 * Finds the first non-filter node above bs in the chain between
5434 * active and bs. The returned node is either an immediate parent of
5435 * bs, or there are only filter nodes between the two.
5437 * Returns NULL if bs is not found in active's image chain,
5438 * or if active == bs.
5440 * Returns the bottommost base image if bs == NULL.
5442 BlockDriverState
*bdrv_find_overlay(BlockDriverState
*active
,
5443 BlockDriverState
*bs
)
5446 GLOBAL_STATE_CODE();
5448 bs
= bdrv_skip_filters(bs
);
5449 active
= bdrv_skip_filters(active
);
5452 BlockDriverState
*next
= bdrv_backing_chain_next(active
);
5462 /* Given a BDS, searches for the base layer. */
5463 BlockDriverState
*bdrv_find_base(BlockDriverState
*bs
)
5465 GLOBAL_STATE_CODE();
5467 return bdrv_find_overlay(bs
, NULL
);
5471 * Return true if at least one of the COW (backing) and filter links
5472 * between @bs and @base is frozen. @errp is set if that's the case.
5473 * @base must be reachable from @bs, or NULL.
5475 bool bdrv_is_backing_chain_frozen(BlockDriverState
*bs
, BlockDriverState
*base
,
5478 BlockDriverState
*i
;
5481 GLOBAL_STATE_CODE();
5483 for (i
= bs
; i
!= base
; i
= child_bs(child
)) {
5484 child
= bdrv_filter_or_cow_child(i
);
5486 if (child
&& child
->frozen
) {
5487 error_setg(errp
, "Cannot change '%s' link from '%s' to '%s'",
5488 child
->name
, i
->node_name
, child
->bs
->node_name
);
5497 * Freeze all COW (backing) and filter links between @bs and @base.
5498 * If any of the links is already frozen the operation is aborted and
5499 * none of the links are modified.
5500 * @base must be reachable from @bs, or NULL.
5501 * Returns 0 on success. On failure returns < 0 and sets @errp.
5503 int bdrv_freeze_backing_chain(BlockDriverState
*bs
, BlockDriverState
*base
,
5506 BlockDriverState
*i
;
5509 GLOBAL_STATE_CODE();
5511 if (bdrv_is_backing_chain_frozen(bs
, base
, errp
)) {
5515 for (i
= bs
; i
!= base
; i
= child_bs(child
)) {
5516 child
= bdrv_filter_or_cow_child(i
);
5517 if (child
&& child
->bs
->never_freeze
) {
5518 error_setg(errp
, "Cannot freeze '%s' link to '%s'",
5519 child
->name
, child
->bs
->node_name
);
5524 for (i
= bs
; i
!= base
; i
= child_bs(child
)) {
5525 child
= bdrv_filter_or_cow_child(i
);
5527 child
->frozen
= true;
5535 * Unfreeze all COW (backing) and filter links between @bs and @base.
5536 * The caller must ensure that all links are frozen before using this
5538 * @base must be reachable from @bs, or NULL.
5540 void bdrv_unfreeze_backing_chain(BlockDriverState
*bs
, BlockDriverState
*base
)
5542 BlockDriverState
*i
;
5545 GLOBAL_STATE_CODE();
5547 for (i
= bs
; i
!= base
; i
= child_bs(child
)) {
5548 child
= bdrv_filter_or_cow_child(i
);
5550 assert(child
->frozen
);
5551 child
->frozen
= false;
5557 * Drops images above 'base' up to and including 'top', and sets the image
5558 * above 'top' to have base as its backing file.
5560 * Requires that the overlay to 'top' is opened r/w, so that the backing file
5561 * information in 'bs' can be properly updated.
5563 * E.g., this will convert the following chain:
5564 * bottom <- base <- intermediate <- top <- active
5568 * bottom <- base <- active
5570 * It is allowed for bottom==base, in which case it converts:
5572 * base <- intermediate <- top <- active
5578 * If backing_file_str is non-NULL, it will be used when modifying top's
5579 * overlay image metadata.
5582 * if active == top, that is considered an error
5585 int bdrv_drop_intermediate(BlockDriverState
*top
, BlockDriverState
*base
,
5586 const char *backing_file_str
)
5588 BlockDriverState
*explicit_top
= top
;
5589 bool update_inherits_from
;
5591 Error
*local_err
= NULL
;
5593 g_autoptr(GSList
) updated_children
= NULL
;
5596 GLOBAL_STATE_CODE();
5599 bdrv_subtree_drained_begin(top
);
5601 if (!top
->drv
|| !base
->drv
) {
5605 /* Make sure that base is in the backing chain of top */
5606 if (!bdrv_chain_contains(top
, base
)) {
5610 /* If 'base' recursively inherits from 'top' then we should set
5611 * base->inherits_from to top->inherits_from after 'top' and all
5612 * other intermediate nodes have been dropped.
5613 * If 'top' is an implicit node (e.g. "commit_top") we should skip
5614 * it because no one inherits from it. We use explicit_top for that. */
5615 explicit_top
= bdrv_skip_implicit_filters(explicit_top
);
5616 update_inherits_from
= bdrv_inherits_from_recursive(base
, explicit_top
);
5618 /* success - we can delete the intermediate states, and link top->base */
5619 if (!backing_file_str
) {
5620 bdrv_refresh_filename(base
);
5621 backing_file_str
= base
->filename
;
5624 QLIST_FOREACH(c
, &top
->parents
, next_parent
) {
5625 updated_children
= g_slist_prepend(updated_children
, c
);
5629 * It seems correct to pass detach_subchain=true here, but it triggers
5630 * one more yet not fixed bug, when due to nested aio_poll loop we switch to
5631 * another drained section, which modify the graph (for example, removing
5632 * the child, which we keep in updated_children list). So, it's a TODO.
5634 * Note, bug triggered if pass detach_subchain=true here and run
5635 * test-bdrv-drain. test_drop_intermediate_poll() test-case will crash.
5638 bdrv_replace_node_common(top
, base
, false, false, &local_err
);
5640 error_report_err(local_err
);
5644 for (p
= updated_children
; p
; p
= p
->next
) {
5647 if (c
->klass
->update_filename
) {
5648 ret
= c
->klass
->update_filename(c
, base
, backing_file_str
,
5652 * TODO: Actually, we want to rollback all previous iterations
5653 * of this loop, and (which is almost impossible) previous
5654 * bdrv_replace_node()...
5656 * Note, that c->klass->update_filename may lead to permission
5657 * update, so it's a bad idea to call it inside permission
5658 * update transaction of bdrv_replace_node.
5660 error_report_err(local_err
);
5666 if (update_inherits_from
) {
5667 base
->inherits_from
= explicit_top
->inherits_from
;
5672 bdrv_subtree_drained_end(top
);
5678 * Implementation of BlockDriver.bdrv_get_allocated_file_size() that
5679 * sums the size of all data-bearing children. (This excludes backing
5682 static int64_t bdrv_sum_allocated_file_size(BlockDriverState
*bs
)
5685 int64_t child_size
, sum
= 0;
5687 QLIST_FOREACH(child
, &bs
->children
, next
) {
5688 if (child
->role
& (BDRV_CHILD_DATA
| BDRV_CHILD_METADATA
|
5689 BDRV_CHILD_FILTERED
))
5691 child_size
= bdrv_get_allocated_file_size(child
->bs
);
5692 if (child_size
< 0) {
5703 * Length of a allocated file in bytes. Sparse files are counted by actual
5704 * allocated space. Return < 0 if error or unknown.
5706 int64_t bdrv_get_allocated_file_size(BlockDriverState
*bs
)
5708 BlockDriver
*drv
= bs
->drv
;
5714 if (drv
->bdrv_get_allocated_file_size
) {
5715 return drv
->bdrv_get_allocated_file_size(bs
);
5718 if (drv
->bdrv_file_open
) {
5720 * Protocol drivers default to -ENOTSUP (most of their data is
5721 * not stored in any of their children (if they even have any),
5722 * so there is no generic way to figure it out).
5725 } else if (drv
->is_filter
) {
5726 /* Filter drivers default to the size of their filtered child */
5727 return bdrv_get_allocated_file_size(bdrv_filter_bs(bs
));
5729 /* Other drivers default to summing their children's sizes */
5730 return bdrv_sum_allocated_file_size(bs
);
5736 * @drv: Format driver
5737 * @opts: Creation options for new image
5738 * @in_bs: Existing image containing data for new image (may be NULL)
5739 * @errp: Error object
5740 * Returns: A #BlockMeasureInfo (free using qapi_free_BlockMeasureInfo())
5743 * Calculate file size required to create a new image.
5745 * If @in_bs is given then space for allocated clusters and zero clusters
5746 * from that image are included in the calculation. If @opts contains a
5747 * backing file that is shared by @in_bs then backing clusters may be omitted
5748 * from the calculation.
5750 * If @in_bs is NULL then the calculation includes no allocated clusters
5751 * unless a preallocation option is given in @opts.
5753 * Note that @in_bs may use a different BlockDriver from @drv.
5755 * If an error occurs the @errp pointer is set.
5757 BlockMeasureInfo
*bdrv_measure(BlockDriver
*drv
, QemuOpts
*opts
,
5758 BlockDriverState
*in_bs
, Error
**errp
)
5761 if (!drv
->bdrv_measure
) {
5762 error_setg(errp
, "Block driver '%s' does not support size measurement",
5767 return drv
->bdrv_measure(opts
, in_bs
, errp
);
5771 * Return number of sectors on success, -errno on error.
5773 int64_t bdrv_nb_sectors(BlockDriverState
*bs
)
5775 BlockDriver
*drv
= bs
->drv
;
5781 if (drv
->has_variable_length
) {
5782 int ret
= refresh_total_sectors(bs
, bs
->total_sectors
);
5787 return bs
->total_sectors
;
5791 * Return length in bytes on success, -errno on error.
5792 * The length is always a multiple of BDRV_SECTOR_SIZE.
5794 int64_t bdrv_getlength(BlockDriverState
*bs
)
5796 int64_t ret
= bdrv_nb_sectors(bs
);
5802 if (ret
> INT64_MAX
/ BDRV_SECTOR_SIZE
) {
5805 return ret
* BDRV_SECTOR_SIZE
;
5808 /* return 0 as number of sectors if no device present or error */
5809 void bdrv_get_geometry(BlockDriverState
*bs
, uint64_t *nb_sectors_ptr
)
5811 int64_t nb_sectors
= bdrv_nb_sectors(bs
);
5814 *nb_sectors_ptr
= nb_sectors
< 0 ? 0 : nb_sectors
;
5817 bool bdrv_is_sg(BlockDriverState
*bs
)
5824 * Return whether the given node supports compressed writes.
5826 bool bdrv_supports_compressed_writes(BlockDriverState
*bs
)
5828 BlockDriverState
*filtered
;
5831 if (!bs
->drv
|| !block_driver_can_compress(bs
->drv
)) {
5835 filtered
= bdrv_filter_bs(bs
);
5838 * Filters can only forward compressed writes, so we have to
5841 return bdrv_supports_compressed_writes(filtered
);
5847 const char *bdrv_get_format_name(BlockDriverState
*bs
)
5850 return bs
->drv
? bs
->drv
->format_name
: NULL
;
5853 static int qsort_strcmp(const void *a
, const void *b
)
5855 return strcmp(*(char *const *)a
, *(char *const *)b
);
5858 void bdrv_iterate_format(void (*it
)(void *opaque
, const char *name
),
5859 void *opaque
, bool read_only
)
5864 const char **formats
= NULL
;
5866 GLOBAL_STATE_CODE();
5868 QLIST_FOREACH(drv
, &bdrv_drivers
, list
) {
5869 if (drv
->format_name
) {
5873 if (use_bdrv_whitelist
&& !bdrv_is_whitelisted(drv
, read_only
)) {
5877 while (formats
&& i
&& !found
) {
5878 found
= !strcmp(formats
[--i
], drv
->format_name
);
5882 formats
= g_renew(const char *, formats
, count
+ 1);
5883 formats
[count
++] = drv
->format_name
;
5888 for (i
= 0; i
< (int)ARRAY_SIZE(block_driver_modules
); i
++) {
5889 const char *format_name
= block_driver_modules
[i
].format_name
;
5895 if (use_bdrv_whitelist
&&
5896 !bdrv_format_is_whitelisted(format_name
, read_only
)) {
5900 while (formats
&& j
&& !found
) {
5901 found
= !strcmp(formats
[--j
], format_name
);
5905 formats
= g_renew(const char *, formats
, count
+ 1);
5906 formats
[count
++] = format_name
;
5911 qsort(formats
, count
, sizeof(formats
[0]), qsort_strcmp
);
5913 for (i
= 0; i
< count
; i
++) {
5914 it(opaque
, formats
[i
]);
5920 /* This function is to find a node in the bs graph */
5921 BlockDriverState
*bdrv_find_node(const char *node_name
)
5923 BlockDriverState
*bs
;
5926 GLOBAL_STATE_CODE();
5928 QTAILQ_FOREACH(bs
, &graph_bdrv_states
, node_list
) {
5929 if (!strcmp(node_name
, bs
->node_name
)) {
5936 /* Put this QMP function here so it can access the static graph_bdrv_states. */
5937 BlockDeviceInfoList
*bdrv_named_nodes_list(bool flat
,
5940 BlockDeviceInfoList
*list
;
5941 BlockDriverState
*bs
;
5943 GLOBAL_STATE_CODE();
5946 QTAILQ_FOREACH(bs
, &graph_bdrv_states
, node_list
) {
5947 BlockDeviceInfo
*info
= bdrv_block_device_info(NULL
, bs
, flat
, errp
);
5949 qapi_free_BlockDeviceInfoList(list
);
5952 QAPI_LIST_PREPEND(list
, info
);
5958 typedef struct XDbgBlockGraphConstructor
{
5959 XDbgBlockGraph
*graph
;
5960 GHashTable
*graph_nodes
;
5961 } XDbgBlockGraphConstructor
;
5963 static XDbgBlockGraphConstructor
*xdbg_graph_new(void)
5965 XDbgBlockGraphConstructor
*gr
= g_new(XDbgBlockGraphConstructor
, 1);
5967 gr
->graph
= g_new0(XDbgBlockGraph
, 1);
5968 gr
->graph_nodes
= g_hash_table_new(NULL
, NULL
);
5973 static XDbgBlockGraph
*xdbg_graph_finalize(XDbgBlockGraphConstructor
*gr
)
5975 XDbgBlockGraph
*graph
= gr
->graph
;
5977 g_hash_table_destroy(gr
->graph_nodes
);
5983 static uintptr_t xdbg_graph_node_num(XDbgBlockGraphConstructor
*gr
, void *node
)
5985 uintptr_t ret
= (uintptr_t)g_hash_table_lookup(gr
->graph_nodes
, node
);
5992 * Start counting from 1, not 0, because 0 interferes with not-found (NULL)
5993 * answer of g_hash_table_lookup.
5995 ret
= g_hash_table_size(gr
->graph_nodes
) + 1;
5996 g_hash_table_insert(gr
->graph_nodes
, node
, (void *)ret
);
6001 static void xdbg_graph_add_node(XDbgBlockGraphConstructor
*gr
, void *node
,
6002 XDbgBlockGraphNodeType type
, const char *name
)
6004 XDbgBlockGraphNode
*n
;
6006 n
= g_new0(XDbgBlockGraphNode
, 1);
6008 n
->id
= xdbg_graph_node_num(gr
, node
);
6010 n
->name
= g_strdup(name
);
6012 QAPI_LIST_PREPEND(gr
->graph
->nodes
, n
);
6015 static void xdbg_graph_add_edge(XDbgBlockGraphConstructor
*gr
, void *parent
,
6016 const BdrvChild
*child
)
6018 BlockPermission qapi_perm
;
6019 XDbgBlockGraphEdge
*edge
;
6020 GLOBAL_STATE_CODE();
6022 edge
= g_new0(XDbgBlockGraphEdge
, 1);
6024 edge
->parent
= xdbg_graph_node_num(gr
, parent
);
6025 edge
->child
= xdbg_graph_node_num(gr
, child
->bs
);
6026 edge
->name
= g_strdup(child
->name
);
6028 for (qapi_perm
= 0; qapi_perm
< BLOCK_PERMISSION__MAX
; qapi_perm
++) {
6029 uint64_t flag
= bdrv_qapi_perm_to_blk_perm(qapi_perm
);
6031 if (flag
& child
->perm
) {
6032 QAPI_LIST_PREPEND(edge
->perm
, qapi_perm
);
6034 if (flag
& child
->shared_perm
) {
6035 QAPI_LIST_PREPEND(edge
->shared_perm
, qapi_perm
);
6039 QAPI_LIST_PREPEND(gr
->graph
->edges
, edge
);
6043 XDbgBlockGraph
*bdrv_get_xdbg_block_graph(Error
**errp
)
6047 BlockDriverState
*bs
;
6049 XDbgBlockGraphConstructor
*gr
= xdbg_graph_new();
6051 GLOBAL_STATE_CODE();
6053 for (blk
= blk_all_next(NULL
); blk
; blk
= blk_all_next(blk
)) {
6054 char *allocated_name
= NULL
;
6055 const char *name
= blk_name(blk
);
6058 name
= allocated_name
= blk_get_attached_dev_id(blk
);
6060 xdbg_graph_add_node(gr
, blk
, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_BACKEND
,
6062 g_free(allocated_name
);
6063 if (blk_root(blk
)) {
6064 xdbg_graph_add_edge(gr
, blk
, blk_root(blk
));
6068 WITH_JOB_LOCK_GUARD() {
6069 for (job
= block_job_next_locked(NULL
); job
;
6070 job
= block_job_next_locked(job
)) {
6073 xdbg_graph_add_node(gr
, job
, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_JOB
,
6075 for (el
= job
->nodes
; el
; el
= el
->next
) {
6076 xdbg_graph_add_edge(gr
, job
, (BdrvChild
*)el
->data
);
6081 QTAILQ_FOREACH(bs
, &graph_bdrv_states
, node_list
) {
6082 xdbg_graph_add_node(gr
, bs
, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_DRIVER
,
6084 QLIST_FOREACH(child
, &bs
->children
, next
) {
6085 xdbg_graph_add_edge(gr
, bs
, child
);
6089 return xdbg_graph_finalize(gr
);
6092 BlockDriverState
*bdrv_lookup_bs(const char *device
,
6093 const char *node_name
,
6097 BlockDriverState
*bs
;
6099 GLOBAL_STATE_CODE();
6102 blk
= blk_by_name(device
);
6107 error_setg(errp
, "Device '%s' has no medium", device
);
6115 bs
= bdrv_find_node(node_name
);
6122 error_setg(errp
, "Cannot find device=\'%s\' nor node-name=\'%s\'",
6123 device
? device
: "",
6124 node_name
? node_name
: "");
6128 /* If 'base' is in the same chain as 'top', return true. Otherwise,
6129 * return false. If either argument is NULL, return false. */
6130 bool bdrv_chain_contains(BlockDriverState
*top
, BlockDriverState
*base
)
6133 GLOBAL_STATE_CODE();
6135 while (top
&& top
!= base
) {
6136 top
= bdrv_filter_or_cow_bs(top
);
6142 BlockDriverState
*bdrv_next_node(BlockDriverState
*bs
)
6144 GLOBAL_STATE_CODE();
6146 return QTAILQ_FIRST(&graph_bdrv_states
);
6148 return QTAILQ_NEXT(bs
, node_list
);
6151 BlockDriverState
*bdrv_next_all_states(BlockDriverState
*bs
)
6153 GLOBAL_STATE_CODE();
6155 return QTAILQ_FIRST(&all_bdrv_states
);
6157 return QTAILQ_NEXT(bs
, bs_list
);
6160 const char *bdrv_get_node_name(const BlockDriverState
*bs
)
6163 return bs
->node_name
;
6166 const char *bdrv_get_parent_name(const BlockDriverState
*bs
)
6172 /* If multiple parents have a name, just pick the first one. */
6173 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
6174 if (c
->klass
->get_name
) {
6175 name
= c
->klass
->get_name(c
);
6176 if (name
&& *name
) {
6185 /* TODO check what callers really want: bs->node_name or blk_name() */
6186 const char *bdrv_get_device_name(const BlockDriverState
*bs
)
6189 return bdrv_get_parent_name(bs
) ?: "";
6192 /* This can be used to identify nodes that might not have a device
6193 * name associated. Since node and device names live in the same
6194 * namespace, the result is unambiguous. The exception is if both are
6195 * absent, then this returns an empty (non-null) string. */
6196 const char *bdrv_get_device_or_node_name(const BlockDriverState
*bs
)
6199 return bdrv_get_parent_name(bs
) ?: bs
->node_name
;
6202 int bdrv_get_flags(BlockDriverState
*bs
)
6205 return bs
->open_flags
;
6208 int bdrv_has_zero_init_1(BlockDriverState
*bs
)
6210 GLOBAL_STATE_CODE();
6214 int bdrv_has_zero_init(BlockDriverState
*bs
)
6216 BlockDriverState
*filtered
;
6217 GLOBAL_STATE_CODE();
6223 /* If BS is a copy on write image, it is initialized to
6224 the contents of the base image, which may not be zeroes. */
6225 if (bdrv_cow_child(bs
)) {
6228 if (bs
->drv
->bdrv_has_zero_init
) {
6229 return bs
->drv
->bdrv_has_zero_init(bs
);
6232 filtered
= bdrv_filter_bs(bs
);
6234 return bdrv_has_zero_init(filtered
);
6241 bool bdrv_can_write_zeroes_with_unmap(BlockDriverState
*bs
)
6244 if (!(bs
->open_flags
& BDRV_O_UNMAP
)) {
6248 return bs
->supported_zero_flags
& BDRV_REQ_MAY_UNMAP
;
6251 void bdrv_get_backing_filename(BlockDriverState
*bs
,
6252 char *filename
, int filename_size
)
6255 pstrcpy(filename
, filename_size
, bs
->backing_file
);
6258 int bdrv_get_info(BlockDriverState
*bs
, BlockDriverInfo
*bdi
)
6261 BlockDriver
*drv
= bs
->drv
;
6263 /* if bs->drv == NULL, bs is closed, so there's nothing to do here */
6267 if (!drv
->bdrv_get_info
) {
6268 BlockDriverState
*filtered
= bdrv_filter_bs(bs
);
6270 return bdrv_get_info(filtered
, bdi
);
6274 memset(bdi
, 0, sizeof(*bdi
));
6275 ret
= drv
->bdrv_get_info(bs
, bdi
);
6280 if (bdi
->cluster_size
> BDRV_MAX_ALIGNMENT
) {
6287 ImageInfoSpecific
*bdrv_get_specific_info(BlockDriverState
*bs
,
6290 BlockDriver
*drv
= bs
->drv
;
6292 if (drv
&& drv
->bdrv_get_specific_info
) {
6293 return drv
->bdrv_get_specific_info(bs
, errp
);
6298 BlockStatsSpecific
*bdrv_get_specific_stats(BlockDriverState
*bs
)
6300 BlockDriver
*drv
= bs
->drv
;
6302 if (!drv
|| !drv
->bdrv_get_specific_stats
) {
6305 return drv
->bdrv_get_specific_stats(bs
);
6308 void bdrv_debug_event(BlockDriverState
*bs
, BlkdebugEvent event
)
6311 if (!bs
|| !bs
->drv
|| !bs
->drv
->bdrv_debug_event
) {
6315 bs
->drv
->bdrv_debug_event(bs
, event
);
6318 static BlockDriverState
*bdrv_find_debug_node(BlockDriverState
*bs
)
6320 GLOBAL_STATE_CODE();
6321 while (bs
&& bs
->drv
&& !bs
->drv
->bdrv_debug_breakpoint
) {
6322 bs
= bdrv_primary_bs(bs
);
6325 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_breakpoint
) {
6326 assert(bs
->drv
->bdrv_debug_remove_breakpoint
);
6333 int bdrv_debug_breakpoint(BlockDriverState
*bs
, const char *event
,
6336 GLOBAL_STATE_CODE();
6337 bs
= bdrv_find_debug_node(bs
);
6339 return bs
->drv
->bdrv_debug_breakpoint(bs
, event
, tag
);
6345 int bdrv_debug_remove_breakpoint(BlockDriverState
*bs
, const char *tag
)
6347 GLOBAL_STATE_CODE();
6348 bs
= bdrv_find_debug_node(bs
);
6350 return bs
->drv
->bdrv_debug_remove_breakpoint(bs
, tag
);
6356 int bdrv_debug_resume(BlockDriverState
*bs
, const char *tag
)
6358 GLOBAL_STATE_CODE();
6359 while (bs
&& (!bs
->drv
|| !bs
->drv
->bdrv_debug_resume
)) {
6360 bs
= bdrv_primary_bs(bs
);
6363 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_resume
) {
6364 return bs
->drv
->bdrv_debug_resume(bs
, tag
);
6370 bool bdrv_debug_is_suspended(BlockDriverState
*bs
, const char *tag
)
6372 GLOBAL_STATE_CODE();
6373 while (bs
&& bs
->drv
&& !bs
->drv
->bdrv_debug_is_suspended
) {
6374 bs
= bdrv_primary_bs(bs
);
6377 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_is_suspended
) {
6378 return bs
->drv
->bdrv_debug_is_suspended(bs
, tag
);
6384 /* backing_file can either be relative, or absolute, or a protocol. If it is
6385 * relative, it must be relative to the chain. So, passing in bs->filename
6386 * from a BDS as backing_file should not be done, as that may be relative to
6387 * the CWD rather than the chain. */
6388 BlockDriverState
*bdrv_find_backing_image(BlockDriverState
*bs
,
6389 const char *backing_file
)
6391 char *filename_full
= NULL
;
6392 char *backing_file_full
= NULL
;
6393 char *filename_tmp
= NULL
;
6394 int is_protocol
= 0;
6395 bool filenames_refreshed
= false;
6396 BlockDriverState
*curr_bs
= NULL
;
6397 BlockDriverState
*retval
= NULL
;
6398 BlockDriverState
*bs_below
;
6400 GLOBAL_STATE_CODE();
6402 if (!bs
|| !bs
->drv
|| !backing_file
) {
6406 filename_full
= g_malloc(PATH_MAX
);
6407 backing_file_full
= g_malloc(PATH_MAX
);
6409 is_protocol
= path_has_protocol(backing_file
);
6412 * Being largely a legacy function, skip any filters here
6413 * (because filters do not have normal filenames, so they cannot
6414 * match anyway; and allowing json:{} filenames is a bit out of
6417 for (curr_bs
= bdrv_skip_filters(bs
);
6418 bdrv_cow_child(curr_bs
) != NULL
;
6421 bs_below
= bdrv_backing_chain_next(curr_bs
);
6423 if (bdrv_backing_overridden(curr_bs
)) {
6425 * If the backing file was overridden, we can only compare
6426 * directly against the backing node's filename.
6429 if (!filenames_refreshed
) {
6431 * This will automatically refresh all of the
6432 * filenames in the rest of the backing chain, so we
6433 * only need to do this once.
6435 bdrv_refresh_filename(bs_below
);
6436 filenames_refreshed
= true;
6439 if (strcmp(backing_file
, bs_below
->filename
) == 0) {
6443 } else if (is_protocol
|| path_has_protocol(curr_bs
->backing_file
)) {
6445 * If either of the filename paths is actually a protocol, then
6446 * compare unmodified paths; otherwise make paths relative.
6448 char *backing_file_full_ret
;
6450 if (strcmp(backing_file
, curr_bs
->backing_file
) == 0) {
6454 /* Also check against the full backing filename for the image */
6455 backing_file_full_ret
= bdrv_get_full_backing_filename(curr_bs
,
6457 if (backing_file_full_ret
) {
6458 bool equal
= strcmp(backing_file
, backing_file_full_ret
) == 0;
6459 g_free(backing_file_full_ret
);
6466 /* If not an absolute filename path, make it relative to the current
6467 * image's filename path */
6468 filename_tmp
= bdrv_make_absolute_filename(curr_bs
, backing_file
,
6470 /* We are going to compare canonicalized absolute pathnames */
6471 if (!filename_tmp
|| !realpath(filename_tmp
, filename_full
)) {
6472 g_free(filename_tmp
);
6475 g_free(filename_tmp
);
6477 /* We need to make sure the backing filename we are comparing against
6478 * is relative to the current image filename (or absolute) */
6479 filename_tmp
= bdrv_get_full_backing_filename(curr_bs
, NULL
);
6480 if (!filename_tmp
|| !realpath(filename_tmp
, backing_file_full
)) {
6481 g_free(filename_tmp
);
6484 g_free(filename_tmp
);
6486 if (strcmp(backing_file_full
, filename_full
) == 0) {
6493 g_free(filename_full
);
6494 g_free(backing_file_full
);
6498 void bdrv_init(void)
6500 #ifdef CONFIG_BDRV_WHITELIST_TOOLS
6501 use_bdrv_whitelist
= 1;
6503 module_call_init(MODULE_INIT_BLOCK
);
6506 void bdrv_init_with_whitelist(void)
6508 use_bdrv_whitelist
= 1;
6512 int bdrv_activate(BlockDriverState
*bs
, Error
**errp
)
6514 BdrvChild
*child
, *parent
;
6515 Error
*local_err
= NULL
;
6517 BdrvDirtyBitmap
*bm
;
6519 GLOBAL_STATE_CODE();
6525 QLIST_FOREACH(child
, &bs
->children
, next
) {
6526 bdrv_activate(child
->bs
, &local_err
);
6528 error_propagate(errp
, local_err
);
6534 * Update permissions, they may differ for inactive nodes.
6536 * Note that the required permissions of inactive images are always a
6537 * subset of the permissions required after activating the image. This
6538 * allows us to just get the permissions upfront without restricting
6539 * bdrv_co_invalidate_cache().
6541 * It also means that in error cases, we don't have to try and revert to
6542 * the old permissions (which is an operation that could fail, too). We can
6543 * just keep the extended permissions for the next time that an activation
6544 * of the image is tried.
6546 if (bs
->open_flags
& BDRV_O_INACTIVE
) {
6547 bs
->open_flags
&= ~BDRV_O_INACTIVE
;
6548 ret
= bdrv_refresh_perms(bs
, errp
);
6550 bs
->open_flags
|= BDRV_O_INACTIVE
;
6554 ret
= bdrv_invalidate_cache(bs
, errp
);
6556 bs
->open_flags
|= BDRV_O_INACTIVE
;
6560 FOR_EACH_DIRTY_BITMAP(bs
, bm
) {
6561 bdrv_dirty_bitmap_skip_store(bm
, false);
6564 ret
= refresh_total_sectors(bs
, bs
->total_sectors
);
6566 bs
->open_flags
|= BDRV_O_INACTIVE
;
6567 error_setg_errno(errp
, -ret
, "Could not refresh total sector count");
6572 QLIST_FOREACH(parent
, &bs
->parents
, next_parent
) {
6573 if (parent
->klass
->activate
) {
6574 parent
->klass
->activate(parent
, &local_err
);
6576 bs
->open_flags
|= BDRV_O_INACTIVE
;
6577 error_propagate(errp
, local_err
);
6586 int coroutine_fn
bdrv_co_invalidate_cache(BlockDriverState
*bs
, Error
**errp
)
6588 Error
*local_err
= NULL
;
6591 assert(!(bs
->open_flags
& BDRV_O_INACTIVE
));
6593 if (bs
->drv
->bdrv_co_invalidate_cache
) {
6594 bs
->drv
->bdrv_co_invalidate_cache(bs
, &local_err
);
6596 error_propagate(errp
, local_err
);
6604 void bdrv_activate_all(Error
**errp
)
6606 BlockDriverState
*bs
;
6607 BdrvNextIterator it
;
6609 GLOBAL_STATE_CODE();
6611 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
6612 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
6615 aio_context_acquire(aio_context
);
6616 ret
= bdrv_activate(bs
, errp
);
6617 aio_context_release(aio_context
);
6619 bdrv_next_cleanup(&it
);
6625 static bool bdrv_has_bds_parent(BlockDriverState
*bs
, bool only_active
)
6628 GLOBAL_STATE_CODE();
6630 QLIST_FOREACH(parent
, &bs
->parents
, next_parent
) {
6631 if (parent
->klass
->parent_is_bds
) {
6632 BlockDriverState
*parent_bs
= parent
->opaque
;
6633 if (!only_active
|| !(parent_bs
->open_flags
& BDRV_O_INACTIVE
)) {
6642 static int bdrv_inactivate_recurse(BlockDriverState
*bs
)
6644 BdrvChild
*child
, *parent
;
6646 uint64_t cumulative_perms
, cumulative_shared_perms
;
6648 GLOBAL_STATE_CODE();
6654 /* Make sure that we don't inactivate a child before its parent.
6655 * It will be covered by recursion from the yet active parent. */
6656 if (bdrv_has_bds_parent(bs
, true)) {
6660 assert(!(bs
->open_flags
& BDRV_O_INACTIVE
));
6662 /* Inactivate this node */
6663 if (bs
->drv
->bdrv_inactivate
) {
6664 ret
= bs
->drv
->bdrv_inactivate(bs
);
6670 QLIST_FOREACH(parent
, &bs
->parents
, next_parent
) {
6671 if (parent
->klass
->inactivate
) {
6672 ret
= parent
->klass
->inactivate(parent
);
6679 bdrv_get_cumulative_perm(bs
, &cumulative_perms
,
6680 &cumulative_shared_perms
);
6681 if (cumulative_perms
& (BLK_PERM_WRITE
| BLK_PERM_WRITE_UNCHANGED
)) {
6682 /* Our inactive parents still need write access. Inactivation failed. */
6686 bs
->open_flags
|= BDRV_O_INACTIVE
;
6689 * Update permissions, they may differ for inactive nodes.
6690 * We only tried to loosen restrictions, so errors are not fatal, ignore
6693 bdrv_refresh_perms(bs
, NULL
);
6695 /* Recursively inactivate children */
6696 QLIST_FOREACH(child
, &bs
->children
, next
) {
6697 ret
= bdrv_inactivate_recurse(child
->bs
);
6706 int bdrv_inactivate_all(void)
6708 BlockDriverState
*bs
= NULL
;
6709 BdrvNextIterator it
;
6711 GSList
*aio_ctxs
= NULL
, *ctx
;
6713 GLOBAL_STATE_CODE();
6715 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
6716 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
6718 if (!g_slist_find(aio_ctxs
, aio_context
)) {
6719 aio_ctxs
= g_slist_prepend(aio_ctxs
, aio_context
);
6720 aio_context_acquire(aio_context
);
6724 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
6725 /* Nodes with BDS parents are covered by recursion from the last
6726 * parent that gets inactivated. Don't inactivate them a second
6727 * time if that has already happened. */
6728 if (bdrv_has_bds_parent(bs
, false)) {
6731 ret
= bdrv_inactivate_recurse(bs
);
6733 bdrv_next_cleanup(&it
);
6739 for (ctx
= aio_ctxs
; ctx
!= NULL
; ctx
= ctx
->next
) {
6740 AioContext
*aio_context
= ctx
->data
;
6741 aio_context_release(aio_context
);
6743 g_slist_free(aio_ctxs
);
6748 /**************************************************************/
6749 /* removable device support */
6752 * Return TRUE if the media is present
6754 bool bdrv_is_inserted(BlockDriverState
*bs
)
6756 BlockDriver
*drv
= bs
->drv
;
6763 if (drv
->bdrv_is_inserted
) {
6764 return drv
->bdrv_is_inserted(bs
);
6766 QLIST_FOREACH(child
, &bs
->children
, next
) {
6767 if (!bdrv_is_inserted(child
->bs
)) {
6775 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
6777 void bdrv_eject(BlockDriverState
*bs
, bool eject_flag
)
6779 BlockDriver
*drv
= bs
->drv
;
6782 if (drv
&& drv
->bdrv_eject
) {
6783 drv
->bdrv_eject(bs
, eject_flag
);
6788 * Lock or unlock the media (if it is locked, the user won't be able
6789 * to eject it manually).
6791 void bdrv_lock_medium(BlockDriverState
*bs
, bool locked
)
6793 BlockDriver
*drv
= bs
->drv
;
6795 trace_bdrv_lock_medium(bs
, locked
);
6797 if (drv
&& drv
->bdrv_lock_medium
) {
6798 drv
->bdrv_lock_medium(bs
, locked
);
6802 /* Get a reference to bs */
6803 void bdrv_ref(BlockDriverState
*bs
)
6805 GLOBAL_STATE_CODE();
6809 /* Release a previously grabbed reference to bs.
6810 * If after releasing, reference count is zero, the BlockDriverState is
6812 void bdrv_unref(BlockDriverState
*bs
)
6814 GLOBAL_STATE_CODE();
6818 assert(bs
->refcnt
> 0);
6819 if (--bs
->refcnt
== 0) {
6824 struct BdrvOpBlocker
{
6826 QLIST_ENTRY(BdrvOpBlocker
) list
;
6829 bool bdrv_op_is_blocked(BlockDriverState
*bs
, BlockOpType op
, Error
**errp
)
6831 BdrvOpBlocker
*blocker
;
6832 GLOBAL_STATE_CODE();
6833 assert((int) op
>= 0 && op
< BLOCK_OP_TYPE_MAX
);
6834 if (!QLIST_EMPTY(&bs
->op_blockers
[op
])) {
6835 blocker
= QLIST_FIRST(&bs
->op_blockers
[op
]);
6836 error_propagate_prepend(errp
, error_copy(blocker
->reason
),
6837 "Node '%s' is busy: ",
6838 bdrv_get_device_or_node_name(bs
));
6844 void bdrv_op_block(BlockDriverState
*bs
, BlockOpType op
, Error
*reason
)
6846 BdrvOpBlocker
*blocker
;
6847 GLOBAL_STATE_CODE();
6848 assert((int) op
>= 0 && op
< BLOCK_OP_TYPE_MAX
);
6850 blocker
= g_new0(BdrvOpBlocker
, 1);
6851 blocker
->reason
= reason
;
6852 QLIST_INSERT_HEAD(&bs
->op_blockers
[op
], blocker
, list
);
6855 void bdrv_op_unblock(BlockDriverState
*bs
, BlockOpType op
, Error
*reason
)
6857 BdrvOpBlocker
*blocker
, *next
;
6858 GLOBAL_STATE_CODE();
6859 assert((int) op
>= 0 && op
< BLOCK_OP_TYPE_MAX
);
6860 QLIST_FOREACH_SAFE(blocker
, &bs
->op_blockers
[op
], list
, next
) {
6861 if (blocker
->reason
== reason
) {
6862 QLIST_REMOVE(blocker
, list
);
6868 void bdrv_op_block_all(BlockDriverState
*bs
, Error
*reason
)
6871 GLOBAL_STATE_CODE();
6872 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
6873 bdrv_op_block(bs
, i
, reason
);
6877 void bdrv_op_unblock_all(BlockDriverState
*bs
, Error
*reason
)
6880 GLOBAL_STATE_CODE();
6881 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
6882 bdrv_op_unblock(bs
, i
, reason
);
6886 bool bdrv_op_blocker_is_empty(BlockDriverState
*bs
)
6889 GLOBAL_STATE_CODE();
6890 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
6891 if (!QLIST_EMPTY(&bs
->op_blockers
[i
])) {
6898 void bdrv_img_create(const char *filename
, const char *fmt
,
6899 const char *base_filename
, const char *base_fmt
,
6900 char *options
, uint64_t img_size
, int flags
, bool quiet
,
6903 QemuOptsList
*create_opts
= NULL
;
6904 QemuOpts
*opts
= NULL
;
6905 const char *backing_fmt
, *backing_file
;
6907 BlockDriver
*drv
, *proto_drv
;
6908 Error
*local_err
= NULL
;
6911 GLOBAL_STATE_CODE();
6913 /* Find driver and parse its options */
6914 drv
= bdrv_find_format(fmt
);
6916 error_setg(errp
, "Unknown file format '%s'", fmt
);
6920 proto_drv
= bdrv_find_protocol(filename
, true, errp
);
6925 if (!drv
->create_opts
) {
6926 error_setg(errp
, "Format driver '%s' does not support image creation",
6931 if (!proto_drv
->create_opts
) {
6932 error_setg(errp
, "Protocol driver '%s' does not support image creation",
6933 proto_drv
->format_name
);
6937 /* Create parameter list */
6938 create_opts
= qemu_opts_append(create_opts
, drv
->create_opts
);
6939 create_opts
= qemu_opts_append(create_opts
, proto_drv
->create_opts
);
6941 opts
= qemu_opts_create(create_opts
, NULL
, 0, &error_abort
);
6943 /* Parse -o options */
6945 if (!qemu_opts_do_parse(opts
, options
, NULL
, errp
)) {
6950 if (!qemu_opt_get(opts
, BLOCK_OPT_SIZE
)) {
6951 qemu_opt_set_number(opts
, BLOCK_OPT_SIZE
, img_size
, &error_abort
);
6952 } else if (img_size
!= UINT64_C(-1)) {
6953 error_setg(errp
, "The image size must be specified only once");
6957 if (base_filename
) {
6958 if (!qemu_opt_set(opts
, BLOCK_OPT_BACKING_FILE
, base_filename
,
6960 error_setg(errp
, "Backing file not supported for file format '%s'",
6967 if (!qemu_opt_set(opts
, BLOCK_OPT_BACKING_FMT
, base_fmt
, NULL
)) {
6968 error_setg(errp
, "Backing file format not supported for file "
6969 "format '%s'", fmt
);
6974 backing_file
= qemu_opt_get(opts
, BLOCK_OPT_BACKING_FILE
);
6976 if (!strcmp(filename
, backing_file
)) {
6977 error_setg(errp
, "Error: Trying to create an image with the "
6978 "same filename as the backing file");
6981 if (backing_file
[0] == '\0') {
6982 error_setg(errp
, "Expected backing file name, got empty string");
6987 backing_fmt
= qemu_opt_get(opts
, BLOCK_OPT_BACKING_FMT
);
6989 /* The size for the image must always be specified, unless we have a backing
6990 * file and we have not been forbidden from opening it. */
6991 size
= qemu_opt_get_size(opts
, BLOCK_OPT_SIZE
, img_size
);
6992 if (backing_file
&& !(flags
& BDRV_O_NO_BACKING
)) {
6993 BlockDriverState
*bs
;
6996 QDict
*backing_options
= NULL
;
6999 bdrv_get_full_backing_filename_from_filename(filename
, backing_file
,
7004 assert(full_backing
);
7007 * No need to do I/O here, which allows us to open encrypted
7008 * backing images without needing the secret
7011 back_flags
&= ~(BDRV_O_RDWR
| BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
);
7012 back_flags
|= BDRV_O_NO_IO
;
7014 backing_options
= qdict_new();
7016 qdict_put_str(backing_options
, "driver", backing_fmt
);
7018 qdict_put_bool(backing_options
, BDRV_OPT_FORCE_SHARE
, true);
7020 bs
= bdrv_open(full_backing
, NULL
, backing_options
, back_flags
,
7022 g_free(full_backing
);
7024 error_append_hint(&local_err
, "Could not open backing image.\n");
7028 error_setg(&local_err
,
7029 "Backing file specified without backing format");
7030 error_append_hint(&local_err
, "Detected format of %s.",
7031 bs
->drv
->format_name
);
7035 /* Opened BS, have no size */
7036 size
= bdrv_getlength(bs
);
7038 error_setg_errno(errp
, -size
, "Could not get size of '%s'",
7043 qemu_opt_set_number(opts
, BLOCK_OPT_SIZE
, size
, &error_abort
);
7047 /* (backing_file && !(flags & BDRV_O_NO_BACKING)) */
7048 } else if (backing_file
&& !backing_fmt
) {
7049 error_setg(&local_err
,
7050 "Backing file specified without backing format");
7055 error_setg(errp
, "Image creation needs a size parameter");
7060 printf("Formatting '%s', fmt=%s ", filename
, fmt
);
7061 qemu_opts_print(opts
, " ");
7066 ret
= bdrv_create(drv
, filename
, opts
, &local_err
);
7068 if (ret
== -EFBIG
) {
7069 /* This is generally a better message than whatever the driver would
7070 * deliver (especially because of the cluster_size_hint), since that
7071 * is most probably not much different from "image too large". */
7072 const char *cluster_size_hint
= "";
7073 if (qemu_opt_get_size(opts
, BLOCK_OPT_CLUSTER_SIZE
, 0)) {
7074 cluster_size_hint
= " (try using a larger cluster size)";
7076 error_setg(errp
, "The image size is too large for file format '%s'"
7077 "%s", fmt
, cluster_size_hint
);
7078 error_free(local_err
);
7083 qemu_opts_del(opts
);
7084 qemu_opts_free(create_opts
);
7085 error_propagate(errp
, local_err
);
7088 AioContext
*bdrv_get_aio_context(BlockDriverState
*bs
)
7091 return bs
? bs
->aio_context
: qemu_get_aio_context();
7094 AioContext
*coroutine_fn
bdrv_co_enter(BlockDriverState
*bs
)
7096 Coroutine
*self
= qemu_coroutine_self();
7097 AioContext
*old_ctx
= qemu_coroutine_get_aio_context(self
);
7098 AioContext
*new_ctx
;
7102 * Increase bs->in_flight to ensure that this operation is completed before
7103 * moving the node to a different AioContext. Read new_ctx only afterwards.
7105 bdrv_inc_in_flight(bs
);
7107 new_ctx
= bdrv_get_aio_context(bs
);
7108 aio_co_reschedule_self(new_ctx
);
7112 void coroutine_fn
bdrv_co_leave(BlockDriverState
*bs
, AioContext
*old_ctx
)
7115 aio_co_reschedule_self(old_ctx
);
7116 bdrv_dec_in_flight(bs
);
7119 void coroutine_fn
bdrv_co_lock(BlockDriverState
*bs
)
7121 AioContext
*ctx
= bdrv_get_aio_context(bs
);
7123 /* In the main thread, bs->aio_context won't change concurrently */
7124 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
7127 * We're in coroutine context, so we already hold the lock of the main
7128 * loop AioContext. Don't lock it twice to avoid deadlocks.
7130 assert(qemu_in_coroutine());
7131 if (ctx
!= qemu_get_aio_context()) {
7132 aio_context_acquire(ctx
);
7136 void coroutine_fn
bdrv_co_unlock(BlockDriverState
*bs
)
7138 AioContext
*ctx
= bdrv_get_aio_context(bs
);
7140 assert(qemu_in_coroutine());
7141 if (ctx
!= qemu_get_aio_context()) {
7142 aio_context_release(ctx
);
7146 void bdrv_coroutine_enter(BlockDriverState
*bs
, Coroutine
*co
)
7149 aio_co_enter(bdrv_get_aio_context(bs
), co
);
7152 static void bdrv_do_remove_aio_context_notifier(BdrvAioNotifier
*ban
)
7154 GLOBAL_STATE_CODE();
7155 QLIST_REMOVE(ban
, list
);
7159 static void bdrv_detach_aio_context(BlockDriverState
*bs
)
7161 BdrvAioNotifier
*baf
, *baf_tmp
;
7163 assert(!bs
->walking_aio_notifiers
);
7164 GLOBAL_STATE_CODE();
7165 bs
->walking_aio_notifiers
= true;
7166 QLIST_FOREACH_SAFE(baf
, &bs
->aio_notifiers
, list
, baf_tmp
) {
7168 bdrv_do_remove_aio_context_notifier(baf
);
7170 baf
->detach_aio_context(baf
->opaque
);
7173 /* Never mind iterating again to check for ->deleted. bdrv_close() will
7174 * remove remaining aio notifiers if we aren't called again.
7176 bs
->walking_aio_notifiers
= false;
7178 if (bs
->drv
&& bs
->drv
->bdrv_detach_aio_context
) {
7179 bs
->drv
->bdrv_detach_aio_context(bs
);
7182 if (bs
->quiesce_counter
) {
7183 aio_enable_external(bs
->aio_context
);
7185 assert_bdrv_graph_writable(bs
);
7186 bs
->aio_context
= NULL
;
7189 static void bdrv_attach_aio_context(BlockDriverState
*bs
,
7190 AioContext
*new_context
)
7192 BdrvAioNotifier
*ban
, *ban_tmp
;
7193 GLOBAL_STATE_CODE();
7195 if (bs
->quiesce_counter
) {
7196 aio_disable_external(new_context
);
7199 assert_bdrv_graph_writable(bs
);
7200 bs
->aio_context
= new_context
;
7202 if (bs
->drv
&& bs
->drv
->bdrv_attach_aio_context
) {
7203 bs
->drv
->bdrv_attach_aio_context(bs
, new_context
);
7206 assert(!bs
->walking_aio_notifiers
);
7207 bs
->walking_aio_notifiers
= true;
7208 QLIST_FOREACH_SAFE(ban
, &bs
->aio_notifiers
, list
, ban_tmp
) {
7210 bdrv_do_remove_aio_context_notifier(ban
);
7212 ban
->attached_aio_context(new_context
, ban
->opaque
);
7215 bs
->walking_aio_notifiers
= false;
7218 typedef struct BdrvStateSetAioContext
{
7219 AioContext
*new_ctx
;
7220 BlockDriverState
*bs
;
7221 } BdrvStateSetAioContext
;
7223 static bool bdrv_parent_change_aio_context(BdrvChild
*c
, AioContext
*ctx
,
7224 GHashTable
*visited
,
7228 GLOBAL_STATE_CODE();
7229 if (g_hash_table_contains(visited
, c
)) {
7232 g_hash_table_add(visited
, c
);
7235 * A BdrvChildClass that doesn't handle AioContext changes cannot
7236 * tolerate any AioContext changes
7238 if (!c
->klass
->change_aio_ctx
) {
7239 char *user
= bdrv_child_user_desc(c
);
7240 error_setg(errp
, "Changing iothreads is not supported by %s", user
);
7244 if (!c
->klass
->change_aio_ctx(c
, ctx
, visited
, tran
, errp
)) {
7245 assert(!errp
|| *errp
);
7251 bool bdrv_child_change_aio_context(BdrvChild
*c
, AioContext
*ctx
,
7252 GHashTable
*visited
, Transaction
*tran
,
7255 GLOBAL_STATE_CODE();
7256 if (g_hash_table_contains(visited
, c
)) {
7259 g_hash_table_add(visited
, c
);
7260 return bdrv_change_aio_context(c
->bs
, ctx
, visited
, tran
, errp
);
7263 static void bdrv_set_aio_context_clean(void *opaque
)
7265 BdrvStateSetAioContext
*state
= (BdrvStateSetAioContext
*) opaque
;
7266 BlockDriverState
*bs
= (BlockDriverState
*) state
->bs
;
7268 /* Paired with bdrv_drained_begin in bdrv_change_aio_context() */
7269 bdrv_drained_end(bs
);
7274 static void bdrv_set_aio_context_commit(void *opaque
)
7276 BdrvStateSetAioContext
*state
= (BdrvStateSetAioContext
*) opaque
;
7277 BlockDriverState
*bs
= (BlockDriverState
*) state
->bs
;
7278 AioContext
*new_context
= state
->new_ctx
;
7279 AioContext
*old_context
= bdrv_get_aio_context(bs
);
7280 assert_bdrv_graph_writable(bs
);
7283 * Take the old AioContex when detaching it from bs.
7284 * At this point, new_context lock is already acquired, and we are now
7285 * also taking old_context. This is safe as long as bdrv_detach_aio_context
7286 * does not call AIO_POLL_WHILE().
7288 if (old_context
!= qemu_get_aio_context()) {
7289 aio_context_acquire(old_context
);
7291 bdrv_detach_aio_context(bs
);
7292 if (old_context
!= qemu_get_aio_context()) {
7293 aio_context_release(old_context
);
7295 bdrv_attach_aio_context(bs
, new_context
);
7298 static TransactionActionDrv set_aio_context
= {
7299 .commit
= bdrv_set_aio_context_commit
,
7300 .clean
= bdrv_set_aio_context_clean
,
7304 * Changes the AioContext used for fd handlers, timers, and BHs by this
7305 * BlockDriverState and all its children and parents.
7307 * Must be called from the main AioContext.
7309 * The caller must own the AioContext lock for the old AioContext of bs, but it
7310 * must not own the AioContext lock for new_context (unless new_context is the
7311 * same as the current context of bs).
7313 * @visited will accumulate all visited BdrvChild objects. The caller is
7314 * responsible for freeing the list afterwards.
7316 static bool bdrv_change_aio_context(BlockDriverState
*bs
, AioContext
*ctx
,
7317 GHashTable
*visited
, Transaction
*tran
,
7321 BdrvStateSetAioContext
*state
;
7323 GLOBAL_STATE_CODE();
7325 if (bdrv_get_aio_context(bs
) == ctx
) {
7329 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
7330 if (!bdrv_parent_change_aio_context(c
, ctx
, visited
, tran
, errp
)) {
7335 QLIST_FOREACH(c
, &bs
->children
, next
) {
7336 if (!bdrv_child_change_aio_context(c
, ctx
, visited
, tran
, errp
)) {
7341 state
= g_new(BdrvStateSetAioContext
, 1);
7342 *state
= (BdrvStateSetAioContext
) {
7347 /* Paired with bdrv_drained_end in bdrv_set_aio_context_clean() */
7348 bdrv_drained_begin(bs
);
7350 tran_add(tran
, &set_aio_context
, state
);
7356 * Change bs's and recursively all of its parents' and children's AioContext
7357 * to the given new context, returning an error if that isn't possible.
7359 * If ignore_child is not NULL, that child (and its subgraph) will not
7362 * This function still requires the caller to take the bs current
7363 * AioContext lock, otherwise draining will fail since AIO_WAIT_WHILE
7364 * assumes the lock is always held if bs is in another AioContext.
7365 * For the same reason, it temporarily also holds the new AioContext, since
7366 * bdrv_drained_end calls BDRV_POLL_WHILE that assumes the lock is taken too.
7367 * Therefore the new AioContext lock must not be taken by the caller.
7369 int bdrv_try_change_aio_context(BlockDriverState
*bs
, AioContext
*ctx
,
7370 BdrvChild
*ignore_child
, Error
**errp
)
7373 GHashTable
*visited
;
7375 AioContext
*old_context
= bdrv_get_aio_context(bs
);
7376 GLOBAL_STATE_CODE();
7379 * Recursion phase: go through all nodes of the graph.
7380 * Take care of checking that all nodes support changing AioContext
7381 * and drain them, builing a linear list of callbacks to run if everything
7382 * is successful (the transaction itself).
7385 visited
= g_hash_table_new(NULL
, NULL
);
7387 g_hash_table_add(visited
, ignore_child
);
7389 ret
= bdrv_change_aio_context(bs
, ctx
, visited
, tran
, errp
);
7390 g_hash_table_destroy(visited
);
7393 * Linear phase: go through all callbacks collected in the transaction.
7394 * Run all callbacks collected in the recursion to switch all nodes
7395 * AioContext lock (transaction commit), or undo all changes done in the
7396 * recursion (transaction abort).
7400 /* Just run clean() callbacks. No AioContext changed. */
7406 * Release old AioContext, it won't be needed anymore, as all
7407 * bdrv_drained_begin() have been called already.
7409 if (qemu_get_aio_context() != old_context
) {
7410 aio_context_release(old_context
);
7414 * Acquire new AioContext since bdrv_drained_end() is going to be called
7415 * after we switched all nodes in the new AioContext, and the function
7416 * assumes that the lock of the bs is always taken.
7418 if (qemu_get_aio_context() != ctx
) {
7419 aio_context_acquire(ctx
);
7424 if (qemu_get_aio_context() != ctx
) {
7425 aio_context_release(ctx
);
7428 /* Re-acquire the old AioContext, since the caller takes and releases it. */
7429 if (qemu_get_aio_context() != old_context
) {
7430 aio_context_acquire(old_context
);
7436 void bdrv_add_aio_context_notifier(BlockDriverState
*bs
,
7437 void (*attached_aio_context
)(AioContext
*new_context
, void *opaque
),
7438 void (*detach_aio_context
)(void *opaque
), void *opaque
)
7440 BdrvAioNotifier
*ban
= g_new(BdrvAioNotifier
, 1);
7441 *ban
= (BdrvAioNotifier
){
7442 .attached_aio_context
= attached_aio_context
,
7443 .detach_aio_context
= detach_aio_context
,
7446 GLOBAL_STATE_CODE();
7448 QLIST_INSERT_HEAD(&bs
->aio_notifiers
, ban
, list
);
7451 void bdrv_remove_aio_context_notifier(BlockDriverState
*bs
,
7452 void (*attached_aio_context
)(AioContext
*,
7454 void (*detach_aio_context
)(void *),
7457 BdrvAioNotifier
*ban
, *ban_next
;
7458 GLOBAL_STATE_CODE();
7460 QLIST_FOREACH_SAFE(ban
, &bs
->aio_notifiers
, list
, ban_next
) {
7461 if (ban
->attached_aio_context
== attached_aio_context
&&
7462 ban
->detach_aio_context
== detach_aio_context
&&
7463 ban
->opaque
== opaque
&&
7464 ban
->deleted
== false)
7466 if (bs
->walking_aio_notifiers
) {
7467 ban
->deleted
= true;
7469 bdrv_do_remove_aio_context_notifier(ban
);
7478 int bdrv_amend_options(BlockDriverState
*bs
, QemuOpts
*opts
,
7479 BlockDriverAmendStatusCB
*status_cb
, void *cb_opaque
,
7483 GLOBAL_STATE_CODE();
7485 error_setg(errp
, "Node is ejected");
7488 if (!bs
->drv
->bdrv_amend_options
) {
7489 error_setg(errp
, "Block driver '%s' does not support option amendment",
7490 bs
->drv
->format_name
);
7493 return bs
->drv
->bdrv_amend_options(bs
, opts
, status_cb
,
7494 cb_opaque
, force
, errp
);
7498 * This function checks whether the given @to_replace is allowed to be
7499 * replaced by a node that always shows the same data as @bs. This is
7500 * used for example to verify whether the mirror job can replace
7501 * @to_replace by the target mirrored from @bs.
7502 * To be replaceable, @bs and @to_replace may either be guaranteed to
7503 * always show the same data (because they are only connected through
7504 * filters), or some driver may allow replacing one of its children
7505 * because it can guarantee that this child's data is not visible at
7506 * all (for example, for dissenting quorum children that have no other
7509 bool bdrv_recurse_can_replace(BlockDriverState
*bs
,
7510 BlockDriverState
*to_replace
)
7512 BlockDriverState
*filtered
;
7514 GLOBAL_STATE_CODE();
7516 if (!bs
|| !bs
->drv
) {
7520 if (bs
== to_replace
) {
7524 /* See what the driver can do */
7525 if (bs
->drv
->bdrv_recurse_can_replace
) {
7526 return bs
->drv
->bdrv_recurse_can_replace(bs
, to_replace
);
7529 /* For filters without an own implementation, we can recurse on our own */
7530 filtered
= bdrv_filter_bs(bs
);
7532 return bdrv_recurse_can_replace(filtered
, to_replace
);
7540 * Check whether the given @node_name can be replaced by a node that
7541 * has the same data as @parent_bs. If so, return @node_name's BDS;
7544 * @node_name must be a (recursive) *child of @parent_bs (or this
7545 * function will return NULL).
7547 * The result (whether the node can be replaced or not) is only valid
7548 * for as long as no graph or permission changes occur.
7550 BlockDriverState
*check_to_replace_node(BlockDriverState
*parent_bs
,
7551 const char *node_name
, Error
**errp
)
7553 BlockDriverState
*to_replace_bs
= bdrv_find_node(node_name
);
7554 AioContext
*aio_context
;
7556 GLOBAL_STATE_CODE();
7558 if (!to_replace_bs
) {
7559 error_setg(errp
, "Failed to find node with node-name='%s'", node_name
);
7563 aio_context
= bdrv_get_aio_context(to_replace_bs
);
7564 aio_context_acquire(aio_context
);
7566 if (bdrv_op_is_blocked(to_replace_bs
, BLOCK_OP_TYPE_REPLACE
, errp
)) {
7567 to_replace_bs
= NULL
;
7571 /* We don't want arbitrary node of the BDS chain to be replaced only the top
7572 * most non filter in order to prevent data corruption.
7573 * Another benefit is that this tests exclude backing files which are
7574 * blocked by the backing blockers.
7576 if (!bdrv_recurse_can_replace(parent_bs
, to_replace_bs
)) {
7577 error_setg(errp
, "Cannot replace '%s' by a node mirrored from '%s', "
7578 "because it cannot be guaranteed that doing so would not "
7579 "lead to an abrupt change of visible data",
7580 node_name
, parent_bs
->node_name
);
7581 to_replace_bs
= NULL
;
7586 aio_context_release(aio_context
);
7587 return to_replace_bs
;
7591 * Iterates through the list of runtime option keys that are said to
7592 * be "strong" for a BDS. An option is called "strong" if it changes
7593 * a BDS's data. For example, the null block driver's "size" and
7594 * "read-zeroes" options are strong, but its "latency-ns" option is
7597 * If a key returned by this function ends with a dot, all options
7598 * starting with that prefix are strong.
7600 static const char *const *strong_options(BlockDriverState
*bs
,
7601 const char *const *curopt
)
7603 static const char *const global_options
[] = {
7604 "driver", "filename", NULL
7608 return &global_options
[0];
7612 if (curopt
== &global_options
[ARRAY_SIZE(global_options
) - 1] && bs
->drv
) {
7613 curopt
= bs
->drv
->strong_runtime_opts
;
7616 return (curopt
&& *curopt
) ? curopt
: NULL
;
7620 * Copies all strong runtime options from bs->options to the given
7621 * QDict. The set of strong option keys is determined by invoking
7624 * Returns true iff any strong option was present in bs->options (and
7625 * thus copied to the target QDict) with the exception of "filename"
7626 * and "driver". The caller is expected to use this value to decide
7627 * whether the existence of strong options prevents the generation of
7630 static bool append_strong_runtime_options(QDict
*d
, BlockDriverState
*bs
)
7632 bool found_any
= false;
7633 const char *const *option_name
= NULL
;
7639 while ((option_name
= strong_options(bs
, option_name
))) {
7640 bool option_given
= false;
7642 assert(strlen(*option_name
) > 0);
7643 if ((*option_name
)[strlen(*option_name
) - 1] != '.') {
7644 QObject
*entry
= qdict_get(bs
->options
, *option_name
);
7649 qdict_put_obj(d
, *option_name
, qobject_ref(entry
));
7650 option_given
= true;
7652 const QDictEntry
*entry
;
7653 for (entry
= qdict_first(bs
->options
); entry
;
7654 entry
= qdict_next(bs
->options
, entry
))
7656 if (strstart(qdict_entry_key(entry
), *option_name
, NULL
)) {
7657 qdict_put_obj(d
, qdict_entry_key(entry
),
7658 qobject_ref(qdict_entry_value(entry
)));
7659 option_given
= true;
7664 /* While "driver" and "filename" need to be included in a JSON filename,
7665 * their existence does not prohibit generation of a plain filename. */
7666 if (!found_any
&& option_given
&&
7667 strcmp(*option_name
, "driver") && strcmp(*option_name
, "filename"))
7673 if (!qdict_haskey(d
, "driver")) {
7674 /* Drivers created with bdrv_new_open_driver() may not have a
7675 * @driver option. Add it here. */
7676 qdict_put_str(d
, "driver", bs
->drv
->format_name
);
7682 /* Note: This function may return false positives; it may return true
7683 * even if opening the backing file specified by bs's image header
7684 * would result in exactly bs->backing. */
7685 static bool bdrv_backing_overridden(BlockDriverState
*bs
)
7687 GLOBAL_STATE_CODE();
7689 return strcmp(bs
->auto_backing_file
,
7690 bs
->backing
->bs
->filename
);
7692 /* No backing BDS, so if the image header reports any backing
7693 * file, it must have been suppressed */
7694 return bs
->auto_backing_file
[0] != '\0';
7698 /* Updates the following BDS fields:
7699 * - exact_filename: A filename which may be used for opening a block device
7700 * which (mostly) equals the given BDS (even without any
7701 * other options; so reading and writing must return the same
7702 * results, but caching etc. may be different)
7703 * - full_open_options: Options which, when given when opening a block device
7704 * (without a filename), result in a BDS (mostly)
7705 * equalling the given one
7706 * - filename: If exact_filename is set, it is copied here. Otherwise,
7707 * full_open_options is converted to a JSON object, prefixed with
7708 * "json:" (for use through the JSON pseudo protocol) and put here.
7710 void bdrv_refresh_filename(BlockDriverState
*bs
)
7712 BlockDriver
*drv
= bs
->drv
;
7714 BlockDriverState
*primary_child_bs
;
7716 bool backing_overridden
;
7717 bool generate_json_filename
; /* Whether our default implementation should
7718 fill exact_filename (false) or not (true) */
7720 GLOBAL_STATE_CODE();
7726 /* This BDS's file name may depend on any of its children's file names, so
7727 * refresh those first */
7728 QLIST_FOREACH(child
, &bs
->children
, next
) {
7729 bdrv_refresh_filename(child
->bs
);
7733 /* For implicit nodes, just copy everything from the single child */
7734 child
= QLIST_FIRST(&bs
->children
);
7735 assert(QLIST_NEXT(child
, next
) == NULL
);
7737 pstrcpy(bs
->exact_filename
, sizeof(bs
->exact_filename
),
7738 child
->bs
->exact_filename
);
7739 pstrcpy(bs
->filename
, sizeof(bs
->filename
), child
->bs
->filename
);
7741 qobject_unref(bs
->full_open_options
);
7742 bs
->full_open_options
= qobject_ref(child
->bs
->full_open_options
);
7747 backing_overridden
= bdrv_backing_overridden(bs
);
7749 if (bs
->open_flags
& BDRV_O_NO_IO
) {
7750 /* Without I/O, the backing file does not change anything.
7751 * Therefore, in such a case (primarily qemu-img), we can
7752 * pretend the backing file has not been overridden even if
7753 * it technically has been. */
7754 backing_overridden
= false;
7757 /* Gather the options QDict */
7759 generate_json_filename
= append_strong_runtime_options(opts
, bs
);
7760 generate_json_filename
|= backing_overridden
;
7762 if (drv
->bdrv_gather_child_options
) {
7763 /* Some block drivers may not want to present all of their children's
7764 * options, or name them differently from BdrvChild.name */
7765 drv
->bdrv_gather_child_options(bs
, opts
, backing_overridden
);
7767 QLIST_FOREACH(child
, &bs
->children
, next
) {
7768 if (child
== bs
->backing
&& !backing_overridden
) {
7769 /* We can skip the backing BDS if it has not been overridden */
7773 qdict_put(opts
, child
->name
,
7774 qobject_ref(child
->bs
->full_open_options
));
7777 if (backing_overridden
&& !bs
->backing
) {
7778 /* Force no backing file */
7779 qdict_put_null(opts
, "backing");
7783 qobject_unref(bs
->full_open_options
);
7784 bs
->full_open_options
= opts
;
7786 primary_child_bs
= bdrv_primary_bs(bs
);
7788 if (drv
->bdrv_refresh_filename
) {
7789 /* Obsolete information is of no use here, so drop the old file name
7790 * information before refreshing it */
7791 bs
->exact_filename
[0] = '\0';
7793 drv
->bdrv_refresh_filename(bs
);
7794 } else if (primary_child_bs
) {
7796 * Try to reconstruct valid information from the underlying
7797 * file -- this only works for format nodes (filter nodes
7798 * cannot be probed and as such must be selected by the user
7799 * either through an options dict, or through a special
7800 * filename which the filter driver must construct in its
7801 * .bdrv_refresh_filename() implementation).
7804 bs
->exact_filename
[0] = '\0';
7807 * We can use the underlying file's filename if:
7808 * - it has a filename,
7809 * - the current BDS is not a filter,
7810 * - the file is a protocol BDS, and
7811 * - opening that file (as this BDS's format) will automatically create
7812 * the BDS tree we have right now, that is:
7813 * - the user did not significantly change this BDS's behavior with
7814 * some explicit (strong) options
7815 * - no non-file child of this BDS has been overridden by the user
7816 * Both of these conditions are represented by generate_json_filename.
7818 if (primary_child_bs
->exact_filename
[0] &&
7819 primary_child_bs
->drv
->bdrv_file_open
&&
7820 !drv
->is_filter
&& !generate_json_filename
)
7822 strcpy(bs
->exact_filename
, primary_child_bs
->exact_filename
);
7826 if (bs
->exact_filename
[0]) {
7827 pstrcpy(bs
->filename
, sizeof(bs
->filename
), bs
->exact_filename
);
7829 GString
*json
= qobject_to_json(QOBJECT(bs
->full_open_options
));
7830 if (snprintf(bs
->filename
, sizeof(bs
->filename
), "json:%s",
7831 json
->str
) >= sizeof(bs
->filename
)) {
7832 /* Give user a hint if we truncated things. */
7833 strcpy(bs
->filename
+ sizeof(bs
->filename
) - 4, "...");
7835 g_string_free(json
, true);
7839 char *bdrv_dirname(BlockDriverState
*bs
, Error
**errp
)
7841 BlockDriver
*drv
= bs
->drv
;
7842 BlockDriverState
*child_bs
;
7844 GLOBAL_STATE_CODE();
7847 error_setg(errp
, "Node '%s' is ejected", bs
->node_name
);
7851 if (drv
->bdrv_dirname
) {
7852 return drv
->bdrv_dirname(bs
, errp
);
7855 child_bs
= bdrv_primary_bs(bs
);
7857 return bdrv_dirname(child_bs
, errp
);
7860 bdrv_refresh_filename(bs
);
7861 if (bs
->exact_filename
[0] != '\0') {
7862 return path_combine(bs
->exact_filename
, "");
7865 error_setg(errp
, "Cannot generate a base directory for %s nodes",
7871 * Hot add/remove a BDS's child. So the user can take a child offline when
7872 * it is broken and take a new child online
7874 void bdrv_add_child(BlockDriverState
*parent_bs
, BlockDriverState
*child_bs
,
7877 GLOBAL_STATE_CODE();
7878 if (!parent_bs
->drv
|| !parent_bs
->drv
->bdrv_add_child
) {
7879 error_setg(errp
, "The node %s does not support adding a child",
7880 bdrv_get_device_or_node_name(parent_bs
));
7884 if (!QLIST_EMPTY(&child_bs
->parents
)) {
7885 error_setg(errp
, "The node %s already has a parent",
7886 child_bs
->node_name
);
7890 parent_bs
->drv
->bdrv_add_child(parent_bs
, child_bs
, errp
);
7893 void bdrv_del_child(BlockDriverState
*parent_bs
, BdrvChild
*child
, Error
**errp
)
7897 GLOBAL_STATE_CODE();
7898 if (!parent_bs
->drv
|| !parent_bs
->drv
->bdrv_del_child
) {
7899 error_setg(errp
, "The node %s does not support removing a child",
7900 bdrv_get_device_or_node_name(parent_bs
));
7904 QLIST_FOREACH(tmp
, &parent_bs
->children
, next
) {
7911 error_setg(errp
, "The node %s does not have a child named %s",
7912 bdrv_get_device_or_node_name(parent_bs
),
7913 bdrv_get_device_or_node_name(child
->bs
));
7917 parent_bs
->drv
->bdrv_del_child(parent_bs
, child
, errp
);
7920 int bdrv_make_empty(BdrvChild
*c
, Error
**errp
)
7922 BlockDriver
*drv
= c
->bs
->drv
;
7925 GLOBAL_STATE_CODE();
7926 assert(c
->perm
& (BLK_PERM_WRITE
| BLK_PERM_WRITE_UNCHANGED
));
7928 if (!drv
->bdrv_make_empty
) {
7929 error_setg(errp
, "%s does not support emptying nodes",
7934 ret
= drv
->bdrv_make_empty(c
->bs
);
7936 error_setg_errno(errp
, -ret
, "Failed to empty %s",
7945 * Return the child that @bs acts as an overlay for, and from which data may be
7946 * copied in COW or COR operations. Usually this is the backing file.
7948 BdrvChild
*bdrv_cow_child(BlockDriverState
*bs
)
7952 if (!bs
|| !bs
->drv
) {
7956 if (bs
->drv
->is_filter
) {
7964 assert(bs
->backing
->role
& BDRV_CHILD_COW
);
7969 * If @bs acts as a filter for exactly one of its children, return
7972 BdrvChild
*bdrv_filter_child(BlockDriverState
*bs
)
7977 if (!bs
|| !bs
->drv
) {
7981 if (!bs
->drv
->is_filter
) {
7985 /* Only one of @backing or @file may be used */
7986 assert(!(bs
->backing
&& bs
->file
));
7988 c
= bs
->backing
?: bs
->file
;
7993 assert(c
->role
& BDRV_CHILD_FILTERED
);
7998 * Return either the result of bdrv_cow_child() or bdrv_filter_child(),
7999 * whichever is non-NULL.
8001 * Return NULL if both are NULL.
8003 BdrvChild
*bdrv_filter_or_cow_child(BlockDriverState
*bs
)
8005 BdrvChild
*cow_child
= bdrv_cow_child(bs
);
8006 BdrvChild
*filter_child
= bdrv_filter_child(bs
);
8009 /* Filter nodes cannot have COW backing files */
8010 assert(!(cow_child
&& filter_child
));
8012 return cow_child
?: filter_child
;
8016 * Return the primary child of this node: For filters, that is the
8017 * filtered child. For other nodes, that is usually the child storing
8019 * (A generally more helpful description is that this is (usually) the
8020 * child that has the same filename as @bs.)
8022 * Drivers do not necessarily have a primary child; for example quorum
8025 BdrvChild
*bdrv_primary_child(BlockDriverState
*bs
)
8027 BdrvChild
*c
, *found
= NULL
;
8030 QLIST_FOREACH(c
, &bs
->children
, next
) {
8031 if (c
->role
& BDRV_CHILD_PRIMARY
) {
8040 static BlockDriverState
*bdrv_do_skip_filters(BlockDriverState
*bs
,
8041 bool stop_on_explicit_filter
)
8049 while (!(stop_on_explicit_filter
&& !bs
->implicit
)) {
8050 c
= bdrv_filter_child(bs
);
8053 * A filter that is embedded in a working block graph must
8054 * have a child. Assert this here so this function does
8055 * not return a filter node that is not expected by the
8058 assert(!bs
->drv
|| !bs
->drv
->is_filter
);
8064 * Note that this treats nodes with bs->drv == NULL as not being
8065 * filters (bs->drv == NULL should be replaced by something else
8067 * The advantage of this behavior is that this function will thus
8068 * always return a non-NULL value (given a non-NULL @bs).
8075 * Return the first BDS that has not been added implicitly or that
8076 * does not have a filtered child down the chain starting from @bs
8077 * (including @bs itself).
8079 BlockDriverState
*bdrv_skip_implicit_filters(BlockDriverState
*bs
)
8081 GLOBAL_STATE_CODE();
8082 return bdrv_do_skip_filters(bs
, true);
8086 * Return the first BDS that does not have a filtered child down the
8087 * chain starting from @bs (including @bs itself).
8089 BlockDriverState
*bdrv_skip_filters(BlockDriverState
*bs
)
8092 return bdrv_do_skip_filters(bs
, false);
8096 * For a backing chain, return the first non-filter backing image of
8097 * the first non-filter image.
8099 BlockDriverState
*bdrv_backing_chain_next(BlockDriverState
*bs
)
8102 return bdrv_skip_filters(bdrv_cow_bs(bdrv_skip_filters(bs
)));
8106 * Check whether [offset, offset + bytes) overlaps with the cached
8107 * block-status data region.
8109 * If so, and @pnum is not NULL, set *pnum to `bsc.data_end - offset`,
8110 * which is what bdrv_bsc_is_data()'s interface needs.
8111 * Otherwise, *pnum is not touched.
8113 static bool bdrv_bsc_range_overlaps_locked(BlockDriverState
*bs
,
8114 int64_t offset
, int64_t bytes
,
8117 BdrvBlockStatusCache
*bsc
= qatomic_rcu_read(&bs
->block_status_cache
);
8121 qatomic_read(&bsc
->valid
) &&
8122 ranges_overlap(offset
, bytes
, bsc
->data_start
,
8123 bsc
->data_end
- bsc
->data_start
);
8125 if (overlaps
&& pnum
) {
8126 *pnum
= bsc
->data_end
- offset
;
8133 * See block_int.h for this function's documentation.
8135 bool bdrv_bsc_is_data(BlockDriverState
*bs
, int64_t offset
, int64_t *pnum
)
8138 RCU_READ_LOCK_GUARD();
8139 return bdrv_bsc_range_overlaps_locked(bs
, offset
, 1, pnum
);
8143 * See block_int.h for this function's documentation.
8145 void bdrv_bsc_invalidate_range(BlockDriverState
*bs
,
8146 int64_t offset
, int64_t bytes
)
8149 RCU_READ_LOCK_GUARD();
8151 if (bdrv_bsc_range_overlaps_locked(bs
, offset
, bytes
, NULL
)) {
8152 qatomic_set(&bs
->block_status_cache
->valid
, false);
8157 * See block_int.h for this function's documentation.
8159 void bdrv_bsc_fill(BlockDriverState
*bs
, int64_t offset
, int64_t bytes
)
8161 BdrvBlockStatusCache
*new_bsc
= g_new(BdrvBlockStatusCache
, 1);
8162 BdrvBlockStatusCache
*old_bsc
;
8165 *new_bsc
= (BdrvBlockStatusCache
) {
8167 .data_start
= offset
,
8168 .data_end
= offset
+ bytes
,
8171 QEMU_LOCK_GUARD(&bs
->bsc_modify_lock
);
8173 old_bsc
= qatomic_rcu_read(&bs
->block_status_cache
);
8174 qatomic_rcu_set(&bs
->block_status_cache
, new_bsc
);
8176 g_free_rcu(old_bsc
, rcu
);