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 static QTAILQ_HEAD(, BlockDriverState
) graph_bdrv_states
=
71 QTAILQ_HEAD_INITIALIZER(graph_bdrv_states
);
73 static QTAILQ_HEAD(, BlockDriverState
) all_bdrv_states
=
74 QTAILQ_HEAD_INITIALIZER(all_bdrv_states
);
76 static QLIST_HEAD(, BlockDriver
) bdrv_drivers
=
77 QLIST_HEAD_INITIALIZER(bdrv_drivers
);
79 static BlockDriverState
*bdrv_open_inherit(const char *filename
,
80 const char *reference
,
81 QDict
*options
, int flags
,
82 BlockDriverState
*parent
,
83 const BdrvChildClass
*child_class
,
84 BdrvChildRole child_role
,
87 static bool bdrv_recurse_has_child(BlockDriverState
*bs
,
88 BlockDriverState
*child
);
90 static void bdrv_replace_child_noperm(BdrvChild
*child
,
91 BlockDriverState
*new_bs
);
92 static void bdrv_remove_file_or_backing_child(BlockDriverState
*bs
,
95 static void bdrv_remove_filter_or_cow_child(BlockDriverState
*bs
,
98 static int bdrv_reopen_prepare(BDRVReopenState
*reopen_state
,
99 BlockReopenQueue
*queue
,
100 Transaction
*change_child_tran
, Error
**errp
);
101 static void bdrv_reopen_commit(BDRVReopenState
*reopen_state
);
102 static void bdrv_reopen_abort(BDRVReopenState
*reopen_state
);
104 /* If non-zero, use only whitelisted block drivers */
105 static int use_bdrv_whitelist
;
108 static int is_windows_drive_prefix(const char *filename
)
110 return (((filename
[0] >= 'a' && filename
[0] <= 'z') ||
111 (filename
[0] >= 'A' && filename
[0] <= 'Z')) &&
115 int is_windows_drive(const char *filename
)
117 if (is_windows_drive_prefix(filename
) &&
120 if (strstart(filename
, "\\\\.\\", NULL
) ||
121 strstart(filename
, "//./", NULL
))
127 size_t bdrv_opt_mem_align(BlockDriverState
*bs
)
129 if (!bs
|| !bs
->drv
) {
130 /* page size or 4k (hdd sector size) should be on the safe side */
131 return MAX(4096, qemu_real_host_page_size
);
134 return bs
->bl
.opt_mem_alignment
;
137 size_t bdrv_min_mem_align(BlockDriverState
*bs
)
139 if (!bs
|| !bs
->drv
) {
140 /* page size or 4k (hdd sector size) should be on the safe side */
141 return MAX(4096, qemu_real_host_page_size
);
144 return bs
->bl
.min_mem_alignment
;
147 /* check if the path starts with "<protocol>:" */
148 int path_has_protocol(const char *path
)
153 if (is_windows_drive(path
) ||
154 is_windows_drive_prefix(path
)) {
157 p
= path
+ strcspn(path
, ":/\\");
159 p
= path
+ strcspn(path
, ":/");
165 int path_is_absolute(const char *path
)
168 /* specific case for names like: "\\.\d:" */
169 if (is_windows_drive(path
) || is_windows_drive_prefix(path
)) {
172 return (*path
== '/' || *path
== '\\');
174 return (*path
== '/');
178 /* if filename is absolute, just return its duplicate. Otherwise, build a
179 path to it by considering it is relative to base_path. URL are
181 char *path_combine(const char *base_path
, const char *filename
)
183 const char *protocol_stripped
= NULL
;
188 if (path_is_absolute(filename
)) {
189 return g_strdup(filename
);
192 if (path_has_protocol(base_path
)) {
193 protocol_stripped
= strchr(base_path
, ':');
194 if (protocol_stripped
) {
198 p
= protocol_stripped
?: base_path
;
200 p1
= strrchr(base_path
, '/');
204 p2
= strrchr(base_path
, '\\');
205 if (!p1
|| p2
> p1
) {
220 result
= g_malloc(len
+ strlen(filename
) + 1);
221 memcpy(result
, base_path
, len
);
222 strcpy(result
+ len
, filename
);
228 * Helper function for bdrv_parse_filename() implementations to remove optional
229 * protocol prefixes (especially "file:") from a filename and for putting the
230 * stripped filename into the options QDict if there is such a prefix.
232 void bdrv_parse_filename_strip_prefix(const char *filename
, const char *prefix
,
235 if (strstart(filename
, prefix
, &filename
)) {
236 /* Stripping the explicit protocol prefix may result in a protocol
237 * prefix being (wrongly) detected (if the filename contains a colon) */
238 if (path_has_protocol(filename
)) {
239 GString
*fat_filename
;
241 /* This means there is some colon before the first slash; therefore,
242 * this cannot be an absolute path */
243 assert(!path_is_absolute(filename
));
245 /* And we can thus fix the protocol detection issue by prefixing it
247 fat_filename
= g_string_new("./");
248 g_string_append(fat_filename
, filename
);
250 assert(!path_has_protocol(fat_filename
->str
));
252 qdict_put(options
, "filename",
253 qstring_from_gstring(fat_filename
));
255 /* If no protocol prefix was detected, we can use the shortened
257 qdict_put_str(options
, "filename", filename
);
263 /* Returns whether the image file is opened as read-only. Note that this can
264 * return false and writing to the image file is still not possible because the
265 * image is inactivated. */
266 bool bdrv_is_read_only(BlockDriverState
*bs
)
268 return !(bs
->open_flags
& BDRV_O_RDWR
);
271 int bdrv_can_set_read_only(BlockDriverState
*bs
, bool read_only
,
272 bool ignore_allow_rdw
, Error
**errp
)
274 /* Do not set read_only if copy_on_read is enabled */
275 if (bs
->copy_on_read
&& read_only
) {
276 error_setg(errp
, "Can't set node '%s' to r/o with copy-on-read enabled",
277 bdrv_get_device_or_node_name(bs
));
281 /* Do not clear read_only if it is prohibited */
282 if (!read_only
&& !(bs
->open_flags
& BDRV_O_ALLOW_RDWR
) &&
285 error_setg(errp
, "Node '%s' is read only",
286 bdrv_get_device_or_node_name(bs
));
294 * Called by a driver that can only provide a read-only image.
296 * Returns 0 if the node is already read-only or it could switch the node to
297 * read-only because BDRV_O_AUTO_RDONLY is set.
299 * Returns -EACCES if the node is read-write and BDRV_O_AUTO_RDONLY is not set
300 * or bdrv_can_set_read_only() forbids making the node read-only. If @errmsg
301 * is not NULL, it is used as the error message for the Error object.
303 int bdrv_apply_auto_read_only(BlockDriverState
*bs
, const char *errmsg
,
308 if (!(bs
->open_flags
& BDRV_O_RDWR
)) {
311 if (!(bs
->open_flags
& BDRV_O_AUTO_RDONLY
)) {
315 ret
= bdrv_can_set_read_only(bs
, true, false, NULL
);
320 bs
->open_flags
&= ~BDRV_O_RDWR
;
325 error_setg(errp
, "%s", errmsg
?: "Image is read-only");
330 * If @backing is empty, this function returns NULL without setting
331 * @errp. In all other cases, NULL will only be returned with @errp
334 * Therefore, a return value of NULL without @errp set means that
335 * there is no backing file; if @errp is set, there is one but its
336 * absolute filename cannot be generated.
338 char *bdrv_get_full_backing_filename_from_filename(const char *backed
,
342 if (backing
[0] == '\0') {
344 } else if (path_has_protocol(backing
) || path_is_absolute(backing
)) {
345 return g_strdup(backing
);
346 } else if (backed
[0] == '\0' || strstart(backed
, "json:", NULL
)) {
347 error_setg(errp
, "Cannot use relative backing file names for '%s'",
351 return path_combine(backed
, backing
);
356 * If @filename is empty or NULL, this function returns NULL without
357 * setting @errp. In all other cases, NULL will only be returned with
360 static char *bdrv_make_absolute_filename(BlockDriverState
*relative_to
,
361 const char *filename
, Error
**errp
)
363 char *dir
, *full_name
;
365 if (!filename
|| filename
[0] == '\0') {
367 } else if (path_has_protocol(filename
) || path_is_absolute(filename
)) {
368 return g_strdup(filename
);
371 dir
= bdrv_dirname(relative_to
, errp
);
376 full_name
= g_strconcat(dir
, filename
, NULL
);
381 char *bdrv_get_full_backing_filename(BlockDriverState
*bs
, Error
**errp
)
383 return bdrv_make_absolute_filename(bs
, bs
->backing_file
, errp
);
386 void bdrv_register(BlockDriver
*bdrv
)
388 assert(bdrv
->format_name
);
389 QLIST_INSERT_HEAD(&bdrv_drivers
, bdrv
, list
);
392 BlockDriverState
*bdrv_new(void)
394 BlockDriverState
*bs
;
397 bs
= g_new0(BlockDriverState
, 1);
398 QLIST_INIT(&bs
->dirty_bitmaps
);
399 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
400 QLIST_INIT(&bs
->op_blockers
[i
]);
402 qemu_co_mutex_init(&bs
->reqs_lock
);
403 qemu_mutex_init(&bs
->dirty_bitmap_mutex
);
405 bs
->aio_context
= qemu_get_aio_context();
407 qemu_co_queue_init(&bs
->flush_queue
);
409 qemu_co_mutex_init(&bs
->bsc_modify_lock
);
410 bs
->block_status_cache
= g_new0(BdrvBlockStatusCache
, 1);
412 for (i
= 0; i
< bdrv_drain_all_count
; i
++) {
413 bdrv_drained_begin(bs
);
416 QTAILQ_INSERT_TAIL(&all_bdrv_states
, bs
, bs_list
);
421 static BlockDriver
*bdrv_do_find_format(const char *format_name
)
425 QLIST_FOREACH(drv1
, &bdrv_drivers
, list
) {
426 if (!strcmp(drv1
->format_name
, format_name
)) {
434 BlockDriver
*bdrv_find_format(const char *format_name
)
439 drv1
= bdrv_do_find_format(format_name
);
444 /* The driver isn't registered, maybe we need to load a module */
445 for (i
= 0; i
< (int)ARRAY_SIZE(block_driver_modules
); ++i
) {
446 if (!strcmp(block_driver_modules
[i
].format_name
, format_name
)) {
447 block_module_load_one(block_driver_modules
[i
].library_name
);
452 return bdrv_do_find_format(format_name
);
455 static int bdrv_format_is_whitelisted(const char *format_name
, bool read_only
)
457 static const char *whitelist_rw
[] = {
458 CONFIG_BDRV_RW_WHITELIST
461 static const char *whitelist_ro
[] = {
462 CONFIG_BDRV_RO_WHITELIST
467 if (!whitelist_rw
[0] && !whitelist_ro
[0]) {
468 return 1; /* no whitelist, anything goes */
471 for (p
= whitelist_rw
; *p
; p
++) {
472 if (!strcmp(format_name
, *p
)) {
477 for (p
= whitelist_ro
; *p
; p
++) {
478 if (!strcmp(format_name
, *p
)) {
486 int bdrv_is_whitelisted(BlockDriver
*drv
, bool read_only
)
488 return bdrv_format_is_whitelisted(drv
->format_name
, read_only
);
491 bool bdrv_uses_whitelist(void)
493 return use_bdrv_whitelist
;
496 typedef struct CreateCo
{
504 static void coroutine_fn
bdrv_create_co_entry(void *opaque
)
506 Error
*local_err
= NULL
;
509 CreateCo
*cco
= opaque
;
512 ret
= cco
->drv
->bdrv_co_create_opts(cco
->drv
,
513 cco
->filename
, cco
->opts
, &local_err
);
514 error_propagate(&cco
->err
, local_err
);
518 int bdrv_create(BlockDriver
*drv
, const char* filename
,
519 QemuOpts
*opts
, Error
**errp
)
526 .filename
= g_strdup(filename
),
532 if (!drv
->bdrv_co_create_opts
) {
533 error_setg(errp
, "Driver '%s' does not support image creation", drv
->format_name
);
538 if (qemu_in_coroutine()) {
539 /* Fast-path if already in coroutine context */
540 bdrv_create_co_entry(&cco
);
542 co
= qemu_coroutine_create(bdrv_create_co_entry
, &cco
);
543 qemu_coroutine_enter(co
);
544 while (cco
.ret
== NOT_DONE
) {
545 aio_poll(qemu_get_aio_context(), true);
552 error_propagate(errp
, cco
.err
);
554 error_setg_errno(errp
, -ret
, "Could not create image");
559 g_free(cco
.filename
);
564 * Helper function for bdrv_create_file_fallback(): Resize @blk to at
565 * least the given @minimum_size.
567 * On success, return @blk's actual length.
568 * Otherwise, return -errno.
570 static int64_t create_file_fallback_truncate(BlockBackend
*blk
,
571 int64_t minimum_size
, Error
**errp
)
573 Error
*local_err
= NULL
;
577 ret
= blk_truncate(blk
, minimum_size
, false, PREALLOC_MODE_OFF
, 0,
579 if (ret
< 0 && ret
!= -ENOTSUP
) {
580 error_propagate(errp
, local_err
);
584 size
= blk_getlength(blk
);
586 error_free(local_err
);
587 error_setg_errno(errp
, -size
,
588 "Failed to inquire the new image file's length");
592 if (size
< minimum_size
) {
593 /* Need to grow the image, but we failed to do that */
594 error_propagate(errp
, local_err
);
598 error_free(local_err
);
605 * Helper function for bdrv_create_file_fallback(): Zero the first
606 * sector to remove any potentially pre-existing image header.
608 static int create_file_fallback_zero_first_sector(BlockBackend
*blk
,
609 int64_t current_size
,
612 int64_t bytes_to_clear
;
615 bytes_to_clear
= MIN(current_size
, BDRV_SECTOR_SIZE
);
616 if (bytes_to_clear
) {
617 ret
= blk_pwrite_zeroes(blk
, 0, bytes_to_clear
, BDRV_REQ_MAY_UNMAP
);
619 error_setg_errno(errp
, -ret
,
620 "Failed to clear the new image's first sector");
629 * Simple implementation of bdrv_co_create_opts for protocol drivers
630 * which only support creation via opening a file
631 * (usually existing raw storage device)
633 int coroutine_fn
bdrv_co_create_opts_simple(BlockDriver
*drv
,
634 const char *filename
,
642 PreallocMode prealloc
;
643 Error
*local_err
= NULL
;
646 size
= qemu_opt_get_size_del(opts
, BLOCK_OPT_SIZE
, 0);
647 buf
= qemu_opt_get_del(opts
, BLOCK_OPT_PREALLOC
);
648 prealloc
= qapi_enum_parse(&PreallocMode_lookup
, buf
,
649 PREALLOC_MODE_OFF
, &local_err
);
652 error_propagate(errp
, local_err
);
656 if (prealloc
!= PREALLOC_MODE_OFF
) {
657 error_setg(errp
, "Unsupported preallocation mode '%s'",
658 PreallocMode_str(prealloc
));
662 options
= qdict_new();
663 qdict_put_str(options
, "driver", drv
->format_name
);
665 blk
= blk_new_open(filename
, NULL
, options
,
666 BDRV_O_RDWR
| BDRV_O_RESIZE
, errp
);
668 error_prepend(errp
, "Protocol driver '%s' does not support image "
669 "creation, and opening the image failed: ",
674 size
= create_file_fallback_truncate(blk
, size
, errp
);
680 ret
= create_file_fallback_zero_first_sector(blk
, size
, errp
);
691 int bdrv_create_file(const char *filename
, QemuOpts
*opts
, Error
**errp
)
693 QemuOpts
*protocol_opts
;
698 drv
= bdrv_find_protocol(filename
, true, errp
);
703 if (!drv
->create_opts
) {
704 error_setg(errp
, "Driver '%s' does not support image creation",
710 * 'opts' contains a QemuOptsList with a combination of format and protocol
713 * The format properly removes its options, but the default values remain
714 * in 'opts->list'. So if the protocol has options with the same name
715 * (e.g. rbd has 'cluster_size' as qcow2), it will see the default values
716 * of the format, since for overlapping options, the format wins.
718 * To avoid this issue, lets convert QemuOpts to QDict, in this way we take
719 * only the set options, and then convert it back to QemuOpts, using the
720 * create_opts of the protocol. So the new QemuOpts, will contain only the
723 qdict
= qemu_opts_to_qdict(opts
, NULL
);
724 protocol_opts
= qemu_opts_from_qdict(drv
->create_opts
, qdict
, errp
);
725 if (protocol_opts
== NULL
) {
730 ret
= bdrv_create(drv
, filename
, protocol_opts
, errp
);
732 qemu_opts_del(protocol_opts
);
733 qobject_unref(qdict
);
737 int coroutine_fn
bdrv_co_delete_file(BlockDriverState
*bs
, Error
**errp
)
739 Error
*local_err
= NULL
;
745 error_setg(errp
, "Block node '%s' is not opened", bs
->filename
);
749 if (!bs
->drv
->bdrv_co_delete_file
) {
750 error_setg(errp
, "Driver '%s' does not support image deletion",
751 bs
->drv
->format_name
);
755 ret
= bs
->drv
->bdrv_co_delete_file(bs
, &local_err
);
757 error_propagate(errp
, local_err
);
763 void coroutine_fn
bdrv_co_delete_file_noerr(BlockDriverState
*bs
)
765 Error
*local_err
= NULL
;
772 ret
= bdrv_co_delete_file(bs
, &local_err
);
774 * ENOTSUP will happen if the block driver doesn't support
775 * the 'bdrv_co_delete_file' interface. This is a predictable
776 * scenario and shouldn't be reported back to the user.
778 if (ret
== -ENOTSUP
) {
779 error_free(local_err
);
780 } else if (ret
< 0) {
781 error_report_err(local_err
);
786 * Try to get @bs's logical and physical block size.
787 * On success, store them in @bsz struct and return 0.
788 * On failure return -errno.
789 * @bs must not be empty.
791 int bdrv_probe_blocksizes(BlockDriverState
*bs
, BlockSizes
*bsz
)
793 BlockDriver
*drv
= bs
->drv
;
794 BlockDriverState
*filtered
= bdrv_filter_bs(bs
);
796 if (drv
&& drv
->bdrv_probe_blocksizes
) {
797 return drv
->bdrv_probe_blocksizes(bs
, bsz
);
798 } else if (filtered
) {
799 return bdrv_probe_blocksizes(filtered
, bsz
);
806 * Try to get @bs's geometry (cyls, heads, sectors).
807 * On success, store them in @geo struct and return 0.
808 * On failure return -errno.
809 * @bs must not be empty.
811 int bdrv_probe_geometry(BlockDriverState
*bs
, HDGeometry
*geo
)
813 BlockDriver
*drv
= bs
->drv
;
814 BlockDriverState
*filtered
= bdrv_filter_bs(bs
);
816 if (drv
&& drv
->bdrv_probe_geometry
) {
817 return drv
->bdrv_probe_geometry(bs
, geo
);
818 } else if (filtered
) {
819 return bdrv_probe_geometry(filtered
, geo
);
826 * Create a uniquely-named empty temporary file.
827 * Return 0 upon success, otherwise a negative errno value.
829 int get_tmp_filename(char *filename
, int size
)
832 char temp_dir
[MAX_PATH
];
833 /* GetTempFileName requires that its output buffer (4th param)
834 have length MAX_PATH or greater. */
835 assert(size
>= MAX_PATH
);
836 return (GetTempPath(MAX_PATH
, temp_dir
)
837 && GetTempFileName(temp_dir
, "qem", 0, filename
)
838 ? 0 : -GetLastError());
842 tmpdir
= getenv("TMPDIR");
846 if (snprintf(filename
, size
, "%s/vl.XXXXXX", tmpdir
) >= size
) {
849 fd
= mkstemp(filename
);
853 if (close(fd
) != 0) {
862 * Detect host devices. By convention, /dev/cdrom[N] is always
863 * recognized as a host CDROM.
865 static BlockDriver
*find_hdev_driver(const char *filename
)
867 int score_max
= 0, score
;
868 BlockDriver
*drv
= NULL
, *d
;
870 QLIST_FOREACH(d
, &bdrv_drivers
, list
) {
871 if (d
->bdrv_probe_device
) {
872 score
= d
->bdrv_probe_device(filename
);
873 if (score
> score_max
) {
883 static BlockDriver
*bdrv_do_find_protocol(const char *protocol
)
887 QLIST_FOREACH(drv1
, &bdrv_drivers
, list
) {
888 if (drv1
->protocol_name
&& !strcmp(drv1
->protocol_name
, protocol
)) {
896 BlockDriver
*bdrv_find_protocol(const char *filename
,
897 bool allow_protocol_prefix
,
906 /* TODO Drivers without bdrv_file_open must be specified explicitly */
909 * XXX(hch): we really should not let host device detection
910 * override an explicit protocol specification, but moving this
911 * later breaks access to device names with colons in them.
912 * Thanks to the brain-dead persistent naming schemes on udev-
913 * based Linux systems those actually are quite common.
915 drv1
= find_hdev_driver(filename
);
920 if (!path_has_protocol(filename
) || !allow_protocol_prefix
) {
924 p
= strchr(filename
, ':');
927 if (len
> sizeof(protocol
) - 1)
928 len
= sizeof(protocol
) - 1;
929 memcpy(protocol
, filename
, len
);
930 protocol
[len
] = '\0';
932 drv1
= bdrv_do_find_protocol(protocol
);
937 for (i
= 0; i
< (int)ARRAY_SIZE(block_driver_modules
); ++i
) {
938 if (block_driver_modules
[i
].protocol_name
&&
939 !strcmp(block_driver_modules
[i
].protocol_name
, protocol
)) {
940 block_module_load_one(block_driver_modules
[i
].library_name
);
945 drv1
= bdrv_do_find_protocol(protocol
);
947 error_setg(errp
, "Unknown protocol '%s'", protocol
);
953 * Guess image format by probing its contents.
954 * This is not a good idea when your image is raw (CVE-2008-2004), but
955 * we do it anyway for backward compatibility.
957 * @buf contains the image's first @buf_size bytes.
958 * @buf_size is the buffer size in bytes (generally BLOCK_PROBE_BUF_SIZE,
959 * but can be smaller if the image file is smaller)
960 * @filename is its filename.
962 * For all block drivers, call the bdrv_probe() method to get its
964 * Return the first block driver with the highest probing score.
966 BlockDriver
*bdrv_probe_all(const uint8_t *buf
, int buf_size
,
967 const char *filename
)
969 int score_max
= 0, score
;
970 BlockDriver
*drv
= NULL
, *d
;
972 QLIST_FOREACH(d
, &bdrv_drivers
, list
) {
974 score
= d
->bdrv_probe(buf
, buf_size
, filename
);
975 if (score
> score_max
) {
985 static int find_image_format(BlockBackend
*file
, const char *filename
,
986 BlockDriver
**pdrv
, Error
**errp
)
989 uint8_t buf
[BLOCK_PROBE_BUF_SIZE
];
992 /* Return the raw BlockDriver * to scsi-generic devices or empty drives */
993 if (blk_is_sg(file
) || !blk_is_inserted(file
) || blk_getlength(file
) == 0) {
998 ret
= blk_pread(file
, 0, buf
, sizeof(buf
));
1000 error_setg_errno(errp
, -ret
, "Could not read image for determining its "
1006 drv
= bdrv_probe_all(buf
, ret
, filename
);
1008 error_setg(errp
, "Could not determine image format: No compatible "
1017 * Set the current 'total_sectors' value
1018 * Return 0 on success, -errno on error.
1020 int refresh_total_sectors(BlockDriverState
*bs
, int64_t hint
)
1022 BlockDriver
*drv
= bs
->drv
;
1028 /* Do not attempt drv->bdrv_getlength() on scsi-generic devices */
1032 /* query actual device if possible, otherwise just trust the hint */
1033 if (drv
->bdrv_getlength
) {
1034 int64_t length
= drv
->bdrv_getlength(bs
);
1038 hint
= DIV_ROUND_UP(length
, BDRV_SECTOR_SIZE
);
1041 bs
->total_sectors
= hint
;
1043 if (bs
->total_sectors
* BDRV_SECTOR_SIZE
> BDRV_MAX_LENGTH
) {
1051 * Combines a QDict of new block driver @options with any missing options taken
1052 * from @old_options, so that leaving out an option defaults to its old value.
1054 static void bdrv_join_options(BlockDriverState
*bs
, QDict
*options
,
1057 if (bs
->drv
&& bs
->drv
->bdrv_join_options
) {
1058 bs
->drv
->bdrv_join_options(options
, old_options
);
1060 qdict_join(options
, old_options
, false);
1064 static BlockdevDetectZeroesOptions
bdrv_parse_detect_zeroes(QemuOpts
*opts
,
1068 Error
*local_err
= NULL
;
1069 char *value
= qemu_opt_get_del(opts
, "detect-zeroes");
1070 BlockdevDetectZeroesOptions detect_zeroes
=
1071 qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup
, value
,
1072 BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF
, &local_err
);
1075 error_propagate(errp
, local_err
);
1076 return detect_zeroes
;
1079 if (detect_zeroes
== BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP
&&
1080 !(open_flags
& BDRV_O_UNMAP
))
1082 error_setg(errp
, "setting detect-zeroes to unmap is not allowed "
1083 "without setting discard operation to unmap");
1086 return detect_zeroes
;
1090 * Set open flags for aio engine
1092 * Return 0 on success, -1 if the engine specified is invalid
1094 int bdrv_parse_aio(const char *mode
, int *flags
)
1096 if (!strcmp(mode
, "threads")) {
1097 /* do nothing, default */
1098 } else if (!strcmp(mode
, "native")) {
1099 *flags
|= BDRV_O_NATIVE_AIO
;
1100 #ifdef CONFIG_LINUX_IO_URING
1101 } else if (!strcmp(mode
, "io_uring")) {
1102 *flags
|= BDRV_O_IO_URING
;
1112 * Set open flags for a given discard mode
1114 * Return 0 on success, -1 if the discard mode was invalid.
1116 int bdrv_parse_discard_flags(const char *mode
, int *flags
)
1118 *flags
&= ~BDRV_O_UNMAP
;
1120 if (!strcmp(mode
, "off") || !strcmp(mode
, "ignore")) {
1122 } else if (!strcmp(mode
, "on") || !strcmp(mode
, "unmap")) {
1123 *flags
|= BDRV_O_UNMAP
;
1132 * Set open flags for a given cache mode
1134 * Return 0 on success, -1 if the cache mode was invalid.
1136 int bdrv_parse_cache_mode(const char *mode
, int *flags
, bool *writethrough
)
1138 *flags
&= ~BDRV_O_CACHE_MASK
;
1140 if (!strcmp(mode
, "off") || !strcmp(mode
, "none")) {
1141 *writethrough
= false;
1142 *flags
|= BDRV_O_NOCACHE
;
1143 } else if (!strcmp(mode
, "directsync")) {
1144 *writethrough
= true;
1145 *flags
|= BDRV_O_NOCACHE
;
1146 } else if (!strcmp(mode
, "writeback")) {
1147 *writethrough
= false;
1148 } else if (!strcmp(mode
, "unsafe")) {
1149 *writethrough
= false;
1150 *flags
|= BDRV_O_NO_FLUSH
;
1151 } else if (!strcmp(mode
, "writethrough")) {
1152 *writethrough
= true;
1160 static char *bdrv_child_get_parent_desc(BdrvChild
*c
)
1162 BlockDriverState
*parent
= c
->opaque
;
1163 return g_strdup_printf("node '%s'", bdrv_get_node_name(parent
));
1166 static void bdrv_child_cb_drained_begin(BdrvChild
*child
)
1168 BlockDriverState
*bs
= child
->opaque
;
1169 bdrv_do_drained_begin_quiesce(bs
, NULL
, false);
1172 static bool bdrv_child_cb_drained_poll(BdrvChild
*child
)
1174 BlockDriverState
*bs
= child
->opaque
;
1175 return bdrv_drain_poll(bs
, false, NULL
, false);
1178 static void bdrv_child_cb_drained_end(BdrvChild
*child
,
1179 int *drained_end_counter
)
1181 BlockDriverState
*bs
= child
->opaque
;
1182 bdrv_drained_end_no_poll(bs
, drained_end_counter
);
1185 static int bdrv_child_cb_inactivate(BdrvChild
*child
)
1187 BlockDriverState
*bs
= child
->opaque
;
1188 assert(bs
->open_flags
& BDRV_O_INACTIVE
);
1192 static bool bdrv_child_cb_can_set_aio_ctx(BdrvChild
*child
, AioContext
*ctx
,
1193 GSList
**ignore
, Error
**errp
)
1195 BlockDriverState
*bs
= child
->opaque
;
1196 return bdrv_can_set_aio_context(bs
, ctx
, ignore
, errp
);
1199 static void bdrv_child_cb_set_aio_ctx(BdrvChild
*child
, AioContext
*ctx
,
1202 BlockDriverState
*bs
= child
->opaque
;
1203 return bdrv_set_aio_context_ignore(bs
, ctx
, ignore
);
1207 * Returns the options and flags that a temporary snapshot should get, based on
1208 * the originally requested flags (the originally requested image will have
1209 * flags like a backing file)
1211 static void bdrv_temp_snapshot_options(int *child_flags
, QDict
*child_options
,
1212 int parent_flags
, QDict
*parent_options
)
1214 *child_flags
= (parent_flags
& ~BDRV_O_SNAPSHOT
) | BDRV_O_TEMPORARY
;
1216 /* For temporary files, unconditional cache=unsafe is fine */
1217 qdict_set_default_str(child_options
, BDRV_OPT_CACHE_DIRECT
, "off");
1218 qdict_set_default_str(child_options
, BDRV_OPT_CACHE_NO_FLUSH
, "on");
1220 /* Copy the read-only and discard options from the parent */
1221 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_READ_ONLY
);
1222 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_DISCARD
);
1224 /* aio=native doesn't work for cache.direct=off, so disable it for the
1225 * temporary snapshot */
1226 *child_flags
&= ~BDRV_O_NATIVE_AIO
;
1229 static void bdrv_backing_attach(BdrvChild
*c
)
1231 BlockDriverState
*parent
= c
->opaque
;
1232 BlockDriverState
*backing_hd
= c
->bs
;
1234 assert(!parent
->backing_blocker
);
1235 error_setg(&parent
->backing_blocker
,
1236 "node is used as backing hd of '%s'",
1237 bdrv_get_device_or_node_name(parent
));
1239 bdrv_refresh_filename(backing_hd
);
1241 parent
->open_flags
&= ~BDRV_O_NO_BACKING
;
1243 bdrv_op_block_all(backing_hd
, parent
->backing_blocker
);
1244 /* Otherwise we won't be able to commit or stream */
1245 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_COMMIT_TARGET
,
1246 parent
->backing_blocker
);
1247 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_STREAM
,
1248 parent
->backing_blocker
);
1250 * We do backup in 3 ways:
1252 * The target bs is new opened, and the source is top BDS
1253 * 2. blockdev backup
1254 * Both the source and the target are top BDSes.
1255 * 3. internal backup(used for block replication)
1256 * Both the source and the target are backing file
1258 * In case 1 and 2, neither the source nor the target is the backing file.
1259 * In case 3, we will block the top BDS, so there is only one block job
1260 * for the top BDS and its backing chain.
1262 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_BACKUP_SOURCE
,
1263 parent
->backing_blocker
);
1264 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_BACKUP_TARGET
,
1265 parent
->backing_blocker
);
1268 static void bdrv_backing_detach(BdrvChild
*c
)
1270 BlockDriverState
*parent
= c
->opaque
;
1272 assert(parent
->backing_blocker
);
1273 bdrv_op_unblock_all(c
->bs
, parent
->backing_blocker
);
1274 error_free(parent
->backing_blocker
);
1275 parent
->backing_blocker
= NULL
;
1278 static int bdrv_backing_update_filename(BdrvChild
*c
, BlockDriverState
*base
,
1279 const char *filename
, Error
**errp
)
1281 BlockDriverState
*parent
= c
->opaque
;
1282 bool read_only
= bdrv_is_read_only(parent
);
1286 ret
= bdrv_reopen_set_read_only(parent
, false, errp
);
1292 ret
= bdrv_change_backing_file(parent
, filename
,
1293 base
->drv
? base
->drv
->format_name
: "",
1296 error_setg_errno(errp
, -ret
, "Could not update backing file link");
1300 bdrv_reopen_set_read_only(parent
, true, NULL
);
1307 * Returns the options and flags that a generic child of a BDS should
1308 * get, based on the given options and flags for the parent BDS.
1310 static void bdrv_inherited_options(BdrvChildRole role
, bool parent_is_format
,
1311 int *child_flags
, QDict
*child_options
,
1312 int parent_flags
, QDict
*parent_options
)
1314 int flags
= parent_flags
;
1317 * First, decide whether to set, clear, or leave BDRV_O_PROTOCOL.
1318 * Generally, the question to answer is: Should this child be
1319 * format-probed by default?
1323 * Pure and non-filtered data children of non-format nodes should
1324 * be probed by default (even when the node itself has BDRV_O_PROTOCOL
1325 * set). This only affects a very limited set of drivers (namely
1326 * quorum and blkverify when this comment was written).
1327 * Force-clear BDRV_O_PROTOCOL then.
1329 if (!parent_is_format
&&
1330 (role
& BDRV_CHILD_DATA
) &&
1331 !(role
& (BDRV_CHILD_METADATA
| BDRV_CHILD_FILTERED
)))
1333 flags
&= ~BDRV_O_PROTOCOL
;
1337 * All children of format nodes (except for COW children) and all
1338 * metadata children in general should never be format-probed.
1339 * Force-set BDRV_O_PROTOCOL then.
1341 if ((parent_is_format
&& !(role
& BDRV_CHILD_COW
)) ||
1342 (role
& BDRV_CHILD_METADATA
))
1344 flags
|= BDRV_O_PROTOCOL
;
1348 * If the cache mode isn't explicitly set, inherit direct and no-flush from
1351 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_CACHE_DIRECT
);
1352 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_CACHE_NO_FLUSH
);
1353 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_FORCE_SHARE
);
1355 if (role
& BDRV_CHILD_COW
) {
1356 /* backing files are opened read-only by default */
1357 qdict_set_default_str(child_options
, BDRV_OPT_READ_ONLY
, "on");
1358 qdict_set_default_str(child_options
, BDRV_OPT_AUTO_READ_ONLY
, "off");
1360 /* Inherit the read-only option from the parent if it's not set */
1361 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_READ_ONLY
);
1362 qdict_copy_default(child_options
, parent_options
,
1363 BDRV_OPT_AUTO_READ_ONLY
);
1367 * bdrv_co_pdiscard() respects unmap policy for the parent, so we
1368 * can default to enable it on lower layers regardless of the
1371 qdict_set_default_str(child_options
, BDRV_OPT_DISCARD
, "unmap");
1373 /* Clear flags that only apply to the top layer */
1374 flags
&= ~(BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
| BDRV_O_COPY_ON_READ
);
1376 if (role
& BDRV_CHILD_METADATA
) {
1377 flags
&= ~BDRV_O_NO_IO
;
1379 if (role
& BDRV_CHILD_COW
) {
1380 flags
&= ~BDRV_O_TEMPORARY
;
1383 *child_flags
= flags
;
1386 static void bdrv_child_cb_attach(BdrvChild
*child
)
1388 BlockDriverState
*bs
= child
->opaque
;
1390 QLIST_INSERT_HEAD(&bs
->children
, child
, next
);
1392 if (child
->role
& BDRV_CHILD_COW
) {
1393 bdrv_backing_attach(child
);
1396 bdrv_apply_subtree_drain(child
, bs
);
1399 static void bdrv_child_cb_detach(BdrvChild
*child
)
1401 BlockDriverState
*bs
= child
->opaque
;
1403 if (child
->role
& BDRV_CHILD_COW
) {
1404 bdrv_backing_detach(child
);
1407 bdrv_unapply_subtree_drain(child
, bs
);
1409 QLIST_REMOVE(child
, next
);
1412 static int bdrv_child_cb_update_filename(BdrvChild
*c
, BlockDriverState
*base
,
1413 const char *filename
, Error
**errp
)
1415 if (c
->role
& BDRV_CHILD_COW
) {
1416 return bdrv_backing_update_filename(c
, base
, filename
, errp
);
1421 AioContext
*child_of_bds_get_parent_aio_context(BdrvChild
*c
)
1423 BlockDriverState
*bs
= c
->opaque
;
1425 return bdrv_get_aio_context(bs
);
1428 const BdrvChildClass child_of_bds
= {
1429 .parent_is_bds
= true,
1430 .get_parent_desc
= bdrv_child_get_parent_desc
,
1431 .inherit_options
= bdrv_inherited_options
,
1432 .drained_begin
= bdrv_child_cb_drained_begin
,
1433 .drained_poll
= bdrv_child_cb_drained_poll
,
1434 .drained_end
= bdrv_child_cb_drained_end
,
1435 .attach
= bdrv_child_cb_attach
,
1436 .detach
= bdrv_child_cb_detach
,
1437 .inactivate
= bdrv_child_cb_inactivate
,
1438 .can_set_aio_ctx
= bdrv_child_cb_can_set_aio_ctx
,
1439 .set_aio_ctx
= bdrv_child_cb_set_aio_ctx
,
1440 .update_filename
= bdrv_child_cb_update_filename
,
1441 .get_parent_aio_context
= child_of_bds_get_parent_aio_context
,
1444 AioContext
*bdrv_child_get_parent_aio_context(BdrvChild
*c
)
1446 return c
->klass
->get_parent_aio_context(c
);
1449 static int bdrv_open_flags(BlockDriverState
*bs
, int flags
)
1451 int open_flags
= flags
;
1454 * Clear flags that are internal to the block layer before opening the
1457 open_flags
&= ~(BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
| BDRV_O_PROTOCOL
);
1462 static void update_flags_from_options(int *flags
, QemuOpts
*opts
)
1464 *flags
&= ~(BDRV_O_CACHE_MASK
| BDRV_O_RDWR
| BDRV_O_AUTO_RDONLY
);
1466 if (qemu_opt_get_bool_del(opts
, BDRV_OPT_CACHE_NO_FLUSH
, false)) {
1467 *flags
|= BDRV_O_NO_FLUSH
;
1470 if (qemu_opt_get_bool_del(opts
, BDRV_OPT_CACHE_DIRECT
, false)) {
1471 *flags
|= BDRV_O_NOCACHE
;
1474 if (!qemu_opt_get_bool_del(opts
, BDRV_OPT_READ_ONLY
, false)) {
1475 *flags
|= BDRV_O_RDWR
;
1478 if (qemu_opt_get_bool_del(opts
, BDRV_OPT_AUTO_READ_ONLY
, false)) {
1479 *flags
|= BDRV_O_AUTO_RDONLY
;
1483 static void update_options_from_flags(QDict
*options
, int flags
)
1485 if (!qdict_haskey(options
, BDRV_OPT_CACHE_DIRECT
)) {
1486 qdict_put_bool(options
, BDRV_OPT_CACHE_DIRECT
, flags
& BDRV_O_NOCACHE
);
1488 if (!qdict_haskey(options
, BDRV_OPT_CACHE_NO_FLUSH
)) {
1489 qdict_put_bool(options
, BDRV_OPT_CACHE_NO_FLUSH
,
1490 flags
& BDRV_O_NO_FLUSH
);
1492 if (!qdict_haskey(options
, BDRV_OPT_READ_ONLY
)) {
1493 qdict_put_bool(options
, BDRV_OPT_READ_ONLY
, !(flags
& BDRV_O_RDWR
));
1495 if (!qdict_haskey(options
, BDRV_OPT_AUTO_READ_ONLY
)) {
1496 qdict_put_bool(options
, BDRV_OPT_AUTO_READ_ONLY
,
1497 flags
& BDRV_O_AUTO_RDONLY
);
1501 static void bdrv_assign_node_name(BlockDriverState
*bs
,
1502 const char *node_name
,
1505 char *gen_node_name
= NULL
;
1508 node_name
= gen_node_name
= id_generate(ID_BLOCK
);
1509 } else if (!id_wellformed(node_name
)) {
1511 * Check for empty string or invalid characters, but not if it is
1512 * generated (generated names use characters not available to the user)
1514 error_setg(errp
, "Invalid node-name: '%s'", node_name
);
1518 /* takes care of avoiding namespaces collisions */
1519 if (blk_by_name(node_name
)) {
1520 error_setg(errp
, "node-name=%s is conflicting with a device id",
1525 /* takes care of avoiding duplicates node names */
1526 if (bdrv_find_node(node_name
)) {
1527 error_setg(errp
, "Duplicate nodes with node-name='%s'", node_name
);
1531 /* Make sure that the node name isn't truncated */
1532 if (strlen(node_name
) >= sizeof(bs
->node_name
)) {
1533 error_setg(errp
, "Node name too long");
1537 /* copy node name into the bs and insert it into the graph list */
1538 pstrcpy(bs
->node_name
, sizeof(bs
->node_name
), node_name
);
1539 QTAILQ_INSERT_TAIL(&graph_bdrv_states
, bs
, node_list
);
1541 g_free(gen_node_name
);
1544 static int bdrv_open_driver(BlockDriverState
*bs
, BlockDriver
*drv
,
1545 const char *node_name
, QDict
*options
,
1546 int open_flags
, Error
**errp
)
1548 Error
*local_err
= NULL
;
1551 bdrv_assign_node_name(bs
, node_name
, &local_err
);
1553 error_propagate(errp
, local_err
);
1558 bs
->opaque
= g_malloc0(drv
->instance_size
);
1560 if (drv
->bdrv_file_open
) {
1561 assert(!drv
->bdrv_needs_filename
|| bs
->filename
[0]);
1562 ret
= drv
->bdrv_file_open(bs
, options
, open_flags
, &local_err
);
1563 } else if (drv
->bdrv_open
) {
1564 ret
= drv
->bdrv_open(bs
, options
, open_flags
, &local_err
);
1571 error_propagate(errp
, local_err
);
1572 } else if (bs
->filename
[0]) {
1573 error_setg_errno(errp
, -ret
, "Could not open '%s'", bs
->filename
);
1575 error_setg_errno(errp
, -ret
, "Could not open image");
1580 ret
= refresh_total_sectors(bs
, bs
->total_sectors
);
1582 error_setg_errno(errp
, -ret
, "Could not refresh total sector count");
1586 bdrv_refresh_limits(bs
, NULL
, &local_err
);
1588 error_propagate(errp
, local_err
);
1592 assert(bdrv_opt_mem_align(bs
) != 0);
1593 assert(bdrv_min_mem_align(bs
) != 0);
1594 assert(is_power_of_2(bs
->bl
.request_alignment
));
1596 for (i
= 0; i
< bs
->quiesce_counter
; i
++) {
1597 if (drv
->bdrv_co_drain_begin
) {
1598 drv
->bdrv_co_drain_begin(bs
);
1605 if (bs
->file
!= NULL
) {
1606 bdrv_unref_child(bs
, bs
->file
);
1615 * Create and open a block node.
1617 * @options is a QDict of options to pass to the block drivers, or NULL for an
1618 * empty set of options. The reference to the QDict belongs to the block layer
1619 * after the call (even on failure), so if the caller intends to reuse the
1620 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
1622 BlockDriverState
*bdrv_new_open_driver_opts(BlockDriver
*drv
,
1623 const char *node_name
,
1624 QDict
*options
, int flags
,
1627 BlockDriverState
*bs
;
1631 bs
->open_flags
= flags
;
1632 bs
->options
= options
?: qdict_new();
1633 bs
->explicit_options
= qdict_clone_shallow(bs
->options
);
1636 update_options_from_flags(bs
->options
, flags
);
1638 ret
= bdrv_open_driver(bs
, drv
, node_name
, bs
->options
, flags
, errp
);
1640 qobject_unref(bs
->explicit_options
);
1641 bs
->explicit_options
= NULL
;
1642 qobject_unref(bs
->options
);
1651 /* Create and open a block node. */
1652 BlockDriverState
*bdrv_new_open_driver(BlockDriver
*drv
, const char *node_name
,
1653 int flags
, Error
**errp
)
1655 return bdrv_new_open_driver_opts(drv
, node_name
, NULL
, flags
, errp
);
1658 QemuOptsList bdrv_runtime_opts
= {
1659 .name
= "bdrv_common",
1660 .head
= QTAILQ_HEAD_INITIALIZER(bdrv_runtime_opts
.head
),
1663 .name
= "node-name",
1664 .type
= QEMU_OPT_STRING
,
1665 .help
= "Node name of the block device node",
1669 .type
= QEMU_OPT_STRING
,
1670 .help
= "Block driver to use for the node",
1673 .name
= BDRV_OPT_CACHE_DIRECT
,
1674 .type
= QEMU_OPT_BOOL
,
1675 .help
= "Bypass software writeback cache on the host",
1678 .name
= BDRV_OPT_CACHE_NO_FLUSH
,
1679 .type
= QEMU_OPT_BOOL
,
1680 .help
= "Ignore flush requests",
1683 .name
= BDRV_OPT_READ_ONLY
,
1684 .type
= QEMU_OPT_BOOL
,
1685 .help
= "Node is opened in read-only mode",
1688 .name
= BDRV_OPT_AUTO_READ_ONLY
,
1689 .type
= QEMU_OPT_BOOL
,
1690 .help
= "Node can become read-only if opening read-write fails",
1693 .name
= "detect-zeroes",
1694 .type
= QEMU_OPT_STRING
,
1695 .help
= "try to optimize zero writes (off, on, unmap)",
1698 .name
= BDRV_OPT_DISCARD
,
1699 .type
= QEMU_OPT_STRING
,
1700 .help
= "discard operation (ignore/off, unmap/on)",
1703 .name
= BDRV_OPT_FORCE_SHARE
,
1704 .type
= QEMU_OPT_BOOL
,
1705 .help
= "always accept other writers (default: off)",
1707 { /* end of list */ }
1711 QemuOptsList bdrv_create_opts_simple
= {
1712 .name
= "simple-create-opts",
1713 .head
= QTAILQ_HEAD_INITIALIZER(bdrv_create_opts_simple
.head
),
1716 .name
= BLOCK_OPT_SIZE
,
1717 .type
= QEMU_OPT_SIZE
,
1718 .help
= "Virtual disk size"
1721 .name
= BLOCK_OPT_PREALLOC
,
1722 .type
= QEMU_OPT_STRING
,
1723 .help
= "Preallocation mode (allowed values: off)"
1725 { /* end of list */ }
1730 * Common part for opening disk images and files
1732 * Removes all processed options from *options.
1734 static int bdrv_open_common(BlockDriverState
*bs
, BlockBackend
*file
,
1735 QDict
*options
, Error
**errp
)
1737 int ret
, open_flags
;
1738 const char *filename
;
1739 const char *driver_name
= NULL
;
1740 const char *node_name
= NULL
;
1741 const char *discard
;
1744 Error
*local_err
= NULL
;
1747 assert(bs
->file
== NULL
);
1748 assert(options
!= NULL
&& bs
->options
!= options
);
1750 opts
= qemu_opts_create(&bdrv_runtime_opts
, NULL
, 0, &error_abort
);
1751 if (!qemu_opts_absorb_qdict(opts
, options
, errp
)) {
1756 update_flags_from_options(&bs
->open_flags
, opts
);
1758 driver_name
= qemu_opt_get(opts
, "driver");
1759 drv
= bdrv_find_format(driver_name
);
1760 assert(drv
!= NULL
);
1762 bs
->force_share
= qemu_opt_get_bool(opts
, BDRV_OPT_FORCE_SHARE
, false);
1764 if (bs
->force_share
&& (bs
->open_flags
& BDRV_O_RDWR
)) {
1766 BDRV_OPT_FORCE_SHARE
1767 "=on can only be used with read-only images");
1773 bdrv_refresh_filename(blk_bs(file
));
1774 filename
= blk_bs(file
)->filename
;
1777 * Caution: while qdict_get_try_str() is fine, getting
1778 * non-string types would require more care. When @options
1779 * come from -blockdev or blockdev_add, its members are typed
1780 * according to the QAPI schema, but when they come from
1781 * -drive, they're all QString.
1783 filename
= qdict_get_try_str(options
, "filename");
1786 if (drv
->bdrv_needs_filename
&& (!filename
|| !filename
[0])) {
1787 error_setg(errp
, "The '%s' block driver requires a file name",
1793 trace_bdrv_open_common(bs
, filename
?: "", bs
->open_flags
,
1796 ro
= bdrv_is_read_only(bs
);
1798 if (use_bdrv_whitelist
&& !bdrv_is_whitelisted(drv
, ro
)) {
1799 if (!ro
&& bdrv_is_whitelisted(drv
, true)) {
1800 ret
= bdrv_apply_auto_read_only(bs
, NULL
, NULL
);
1806 !ro
&& bdrv_is_whitelisted(drv
, true)
1807 ? "Driver '%s' can only be used for read-only devices"
1808 : "Driver '%s' is not whitelisted",
1814 /* bdrv_new() and bdrv_close() make it so */
1815 assert(qatomic_read(&bs
->copy_on_read
) == 0);
1817 if (bs
->open_flags
& BDRV_O_COPY_ON_READ
) {
1819 bdrv_enable_copy_on_read(bs
);
1821 error_setg(errp
, "Can't use copy-on-read on read-only device");
1827 discard
= qemu_opt_get(opts
, BDRV_OPT_DISCARD
);
1828 if (discard
!= NULL
) {
1829 if (bdrv_parse_discard_flags(discard
, &bs
->open_flags
) != 0) {
1830 error_setg(errp
, "Invalid discard option");
1837 bdrv_parse_detect_zeroes(opts
, bs
->open_flags
, &local_err
);
1839 error_propagate(errp
, local_err
);
1844 if (filename
!= NULL
) {
1845 pstrcpy(bs
->filename
, sizeof(bs
->filename
), filename
);
1847 bs
->filename
[0] = '\0';
1849 pstrcpy(bs
->exact_filename
, sizeof(bs
->exact_filename
), bs
->filename
);
1851 /* Open the image, either directly or using a protocol */
1852 open_flags
= bdrv_open_flags(bs
, bs
->open_flags
);
1853 node_name
= qemu_opt_get(opts
, "node-name");
1855 assert(!drv
->bdrv_file_open
|| file
== NULL
);
1856 ret
= bdrv_open_driver(bs
, drv
, node_name
, options
, open_flags
, errp
);
1861 qemu_opts_del(opts
);
1865 qemu_opts_del(opts
);
1869 static QDict
*parse_json_filename(const char *filename
, Error
**errp
)
1871 QObject
*options_obj
;
1875 ret
= strstart(filename
, "json:", &filename
);
1878 options_obj
= qobject_from_json(filename
, errp
);
1880 error_prepend(errp
, "Could not parse the JSON options: ");
1884 options
= qobject_to(QDict
, options_obj
);
1886 qobject_unref(options_obj
);
1887 error_setg(errp
, "Invalid JSON object given");
1891 qdict_flatten(options
);
1896 static void parse_json_protocol(QDict
*options
, const char **pfilename
,
1899 QDict
*json_options
;
1900 Error
*local_err
= NULL
;
1902 /* Parse json: pseudo-protocol */
1903 if (!*pfilename
|| !g_str_has_prefix(*pfilename
, "json:")) {
1907 json_options
= parse_json_filename(*pfilename
, &local_err
);
1909 error_propagate(errp
, local_err
);
1913 /* Options given in the filename have lower priority than options
1914 * specified directly */
1915 qdict_join(options
, json_options
, false);
1916 qobject_unref(json_options
);
1921 * Fills in default options for opening images and converts the legacy
1922 * filename/flags pair to option QDict entries.
1923 * The BDRV_O_PROTOCOL flag in *flags will be set or cleared accordingly if a
1924 * block driver has been specified explicitly.
1926 static int bdrv_fill_options(QDict
**options
, const char *filename
,
1927 int *flags
, Error
**errp
)
1929 const char *drvname
;
1930 bool protocol
= *flags
& BDRV_O_PROTOCOL
;
1931 bool parse_filename
= false;
1932 BlockDriver
*drv
= NULL
;
1933 Error
*local_err
= NULL
;
1936 * Caution: while qdict_get_try_str() is fine, getting non-string
1937 * types would require more care. When @options come from
1938 * -blockdev or blockdev_add, its members are typed according to
1939 * the QAPI schema, but when they come from -drive, they're all
1942 drvname
= qdict_get_try_str(*options
, "driver");
1944 drv
= bdrv_find_format(drvname
);
1946 error_setg(errp
, "Unknown driver '%s'", drvname
);
1949 /* If the user has explicitly specified the driver, this choice should
1950 * override the BDRV_O_PROTOCOL flag */
1951 protocol
= drv
->bdrv_file_open
;
1955 *flags
|= BDRV_O_PROTOCOL
;
1957 *flags
&= ~BDRV_O_PROTOCOL
;
1960 /* Translate cache options from flags into options */
1961 update_options_from_flags(*options
, *flags
);
1963 /* Fetch the file name from the options QDict if necessary */
1964 if (protocol
&& filename
) {
1965 if (!qdict_haskey(*options
, "filename")) {
1966 qdict_put_str(*options
, "filename", filename
);
1967 parse_filename
= true;
1969 error_setg(errp
, "Can't specify 'file' and 'filename' options at "
1975 /* Find the right block driver */
1976 /* See cautionary note on accessing @options above */
1977 filename
= qdict_get_try_str(*options
, "filename");
1979 if (!drvname
&& protocol
) {
1981 drv
= bdrv_find_protocol(filename
, parse_filename
, errp
);
1986 drvname
= drv
->format_name
;
1987 qdict_put_str(*options
, "driver", drvname
);
1989 error_setg(errp
, "Must specify either driver or file");
1994 assert(drv
|| !protocol
);
1996 /* Driver-specific filename parsing */
1997 if (drv
&& drv
->bdrv_parse_filename
&& parse_filename
) {
1998 drv
->bdrv_parse_filename(filename
, *options
, &local_err
);
2000 error_propagate(errp
, local_err
);
2004 if (!drv
->bdrv_needs_filename
) {
2005 qdict_del(*options
, "filename");
2012 typedef struct BlockReopenQueueEntry
{
2015 BDRVReopenState state
;
2016 QTAILQ_ENTRY(BlockReopenQueueEntry
) entry
;
2017 } BlockReopenQueueEntry
;
2020 * Return the flags that @bs will have after the reopens in @q have
2021 * successfully completed. If @q is NULL (or @bs is not contained in @q),
2022 * return the current flags.
2024 static int bdrv_reopen_get_flags(BlockReopenQueue
*q
, BlockDriverState
*bs
)
2026 BlockReopenQueueEntry
*entry
;
2029 QTAILQ_FOREACH(entry
, q
, entry
) {
2030 if (entry
->state
.bs
== bs
) {
2031 return entry
->state
.flags
;
2036 return bs
->open_flags
;
2039 /* Returns whether the image file can be written to after the reopen queue @q
2040 * has been successfully applied, or right now if @q is NULL. */
2041 static bool bdrv_is_writable_after_reopen(BlockDriverState
*bs
,
2042 BlockReopenQueue
*q
)
2044 int flags
= bdrv_reopen_get_flags(q
, bs
);
2046 return (flags
& (BDRV_O_RDWR
| BDRV_O_INACTIVE
)) == BDRV_O_RDWR
;
2050 * Return whether the BDS can be written to. This is not necessarily
2051 * the same as !bdrv_is_read_only(bs), as inactivated images may not
2052 * be written to but do not count as read-only images.
2054 bool bdrv_is_writable(BlockDriverState
*bs
)
2056 return bdrv_is_writable_after_reopen(bs
, NULL
);
2059 static char *bdrv_child_user_desc(BdrvChild
*c
)
2061 return c
->klass
->get_parent_desc(c
);
2065 * Check that @a allows everything that @b needs. @a and @b must reference same
2068 static bool bdrv_a_allow_b(BdrvChild
*a
, BdrvChild
*b
, Error
**errp
)
2070 const char *child_bs_name
;
2071 g_autofree
char *a_user
= NULL
;
2072 g_autofree
char *b_user
= NULL
;
2073 g_autofree
char *perms
= NULL
;
2076 assert(a
->bs
== b
->bs
);
2078 if ((b
->perm
& a
->shared_perm
) == b
->perm
) {
2082 child_bs_name
= bdrv_get_node_name(b
->bs
);
2083 a_user
= bdrv_child_user_desc(a
);
2084 b_user
= bdrv_child_user_desc(b
);
2085 perms
= bdrv_perm_names(b
->perm
& ~a
->shared_perm
);
2087 error_setg(errp
, "Permission conflict on node '%s': permissions '%s' are "
2088 "both required by %s (uses node '%s' as '%s' child) and "
2089 "unshared by %s (uses node '%s' as '%s' child).",
2090 child_bs_name
, perms
,
2091 b_user
, child_bs_name
, b
->name
,
2092 a_user
, child_bs_name
, a
->name
);
2097 static bool bdrv_parent_perms_conflict(BlockDriverState
*bs
, Error
**errp
)
2102 * During the loop we'll look at each pair twice. That's correct because
2103 * bdrv_a_allow_b() is asymmetric and we should check each pair in both
2106 QLIST_FOREACH(a
, &bs
->parents
, next_parent
) {
2107 QLIST_FOREACH(b
, &bs
->parents
, next_parent
) {
2112 if (!bdrv_a_allow_b(a
, b
, errp
)) {
2121 static void bdrv_child_perm(BlockDriverState
*bs
, BlockDriverState
*child_bs
,
2122 BdrvChild
*c
, BdrvChildRole role
,
2123 BlockReopenQueue
*reopen_queue
,
2124 uint64_t parent_perm
, uint64_t parent_shared
,
2125 uint64_t *nperm
, uint64_t *nshared
)
2127 assert(bs
->drv
&& bs
->drv
->bdrv_child_perm
);
2128 bs
->drv
->bdrv_child_perm(bs
, c
, role
, reopen_queue
,
2129 parent_perm
, parent_shared
,
2131 /* TODO Take force_share from reopen_queue */
2132 if (child_bs
&& child_bs
->force_share
) {
2133 *nshared
= BLK_PERM_ALL
;
2138 * Adds the whole subtree of @bs (including @bs itself) to the @list (except for
2139 * nodes that are already in the @list, of course) so that final list is
2140 * topologically sorted. Return the result (GSList @list object is updated, so
2141 * don't use old reference after function call).
2143 * On function start @list must be already topologically sorted and for any node
2144 * in the @list the whole subtree of the node must be in the @list as well. The
2145 * simplest way to satisfy this criteria: use only result of
2146 * bdrv_topological_dfs() or NULL as @list parameter.
2148 static GSList
*bdrv_topological_dfs(GSList
*list
, GHashTable
*found
,
2149 BlockDriverState
*bs
)
2152 g_autoptr(GHashTable
) local_found
= NULL
;
2156 found
= local_found
= g_hash_table_new(NULL
, NULL
);
2159 if (g_hash_table_contains(found
, bs
)) {
2162 g_hash_table_add(found
, bs
);
2164 QLIST_FOREACH(child
, &bs
->children
, next
) {
2165 list
= bdrv_topological_dfs(list
, found
, child
->bs
);
2168 return g_slist_prepend(list
, bs
);
2171 typedef struct BdrvChildSetPermState
{
2174 uint64_t old_shared_perm
;
2175 } BdrvChildSetPermState
;
2177 static void bdrv_child_set_perm_abort(void *opaque
)
2179 BdrvChildSetPermState
*s
= opaque
;
2181 s
->child
->perm
= s
->old_perm
;
2182 s
->child
->shared_perm
= s
->old_shared_perm
;
2185 static TransactionActionDrv bdrv_child_set_pem_drv
= {
2186 .abort
= bdrv_child_set_perm_abort
,
2190 static void bdrv_child_set_perm(BdrvChild
*c
, uint64_t perm
,
2191 uint64_t shared
, Transaction
*tran
)
2193 BdrvChildSetPermState
*s
= g_new(BdrvChildSetPermState
, 1);
2195 *s
= (BdrvChildSetPermState
) {
2197 .old_perm
= c
->perm
,
2198 .old_shared_perm
= c
->shared_perm
,
2202 c
->shared_perm
= shared
;
2204 tran_add(tran
, &bdrv_child_set_pem_drv
, s
);
2207 static void bdrv_drv_set_perm_commit(void *opaque
)
2209 BlockDriverState
*bs
= opaque
;
2210 uint64_t cumulative_perms
, cumulative_shared_perms
;
2212 if (bs
->drv
->bdrv_set_perm
) {
2213 bdrv_get_cumulative_perm(bs
, &cumulative_perms
,
2214 &cumulative_shared_perms
);
2215 bs
->drv
->bdrv_set_perm(bs
, cumulative_perms
, cumulative_shared_perms
);
2219 static void bdrv_drv_set_perm_abort(void *opaque
)
2221 BlockDriverState
*bs
= opaque
;
2223 if (bs
->drv
->bdrv_abort_perm_update
) {
2224 bs
->drv
->bdrv_abort_perm_update(bs
);
2228 TransactionActionDrv bdrv_drv_set_perm_drv
= {
2229 .abort
= bdrv_drv_set_perm_abort
,
2230 .commit
= bdrv_drv_set_perm_commit
,
2233 static int bdrv_drv_set_perm(BlockDriverState
*bs
, uint64_t perm
,
2234 uint64_t shared_perm
, Transaction
*tran
,
2241 if (bs
->drv
->bdrv_check_perm
) {
2242 int ret
= bs
->drv
->bdrv_check_perm(bs
, perm
, shared_perm
, errp
);
2249 tran_add(tran
, &bdrv_drv_set_perm_drv
, bs
);
2255 typedef struct BdrvReplaceChildState
{
2257 BlockDriverState
*old_bs
;
2258 } BdrvReplaceChildState
;
2260 static void bdrv_replace_child_commit(void *opaque
)
2262 BdrvReplaceChildState
*s
= opaque
;
2264 bdrv_unref(s
->old_bs
);
2267 static void bdrv_replace_child_abort(void *opaque
)
2269 BdrvReplaceChildState
*s
= opaque
;
2270 BlockDriverState
*new_bs
= s
->child
->bs
;
2272 /* old_bs reference is transparently moved from @s to @s->child */
2273 bdrv_replace_child_noperm(s
->child
, s
->old_bs
);
2277 static TransactionActionDrv bdrv_replace_child_drv
= {
2278 .commit
= bdrv_replace_child_commit
,
2279 .abort
= bdrv_replace_child_abort
,
2284 * bdrv_replace_child_tran
2286 * Note: real unref of old_bs is done only on commit.
2288 * The function doesn't update permissions, caller is responsible for this.
2290 static void bdrv_replace_child_tran(BdrvChild
*child
, BlockDriverState
*new_bs
,
2293 BdrvReplaceChildState
*s
= g_new(BdrvReplaceChildState
, 1);
2294 *s
= (BdrvReplaceChildState
) {
2296 .old_bs
= child
->bs
,
2298 tran_add(tran
, &bdrv_replace_child_drv
, s
);
2303 bdrv_replace_child_noperm(child
, new_bs
);
2304 /* old_bs reference is transparently moved from @child to @s */
2308 * Refresh permissions in @bs subtree. The function is intended to be called
2309 * after some graph modification that was done without permission update.
2311 static int bdrv_node_refresh_perm(BlockDriverState
*bs
, BlockReopenQueue
*q
,
2312 Transaction
*tran
, Error
**errp
)
2314 BlockDriver
*drv
= bs
->drv
;
2317 uint64_t cumulative_perms
, cumulative_shared_perms
;
2319 bdrv_get_cumulative_perm(bs
, &cumulative_perms
, &cumulative_shared_perms
);
2321 /* Write permissions never work with read-only images */
2322 if ((cumulative_perms
& (BLK_PERM_WRITE
| BLK_PERM_WRITE_UNCHANGED
)) &&
2323 !bdrv_is_writable_after_reopen(bs
, q
))
2325 if (!bdrv_is_writable_after_reopen(bs
, NULL
)) {
2326 error_setg(errp
, "Block node is read-only");
2328 error_setg(errp
, "Read-only block node '%s' cannot support "
2329 "read-write users", bdrv_get_node_name(bs
));
2336 * Unaligned requests will automatically be aligned to bl.request_alignment
2337 * and without RESIZE we can't extend requests to write to space beyond the
2338 * end of the image, so it's required that the image size is aligned.
2340 if ((cumulative_perms
& (BLK_PERM_WRITE
| BLK_PERM_WRITE_UNCHANGED
)) &&
2341 !(cumulative_perms
& BLK_PERM_RESIZE
))
2343 if ((bs
->total_sectors
* BDRV_SECTOR_SIZE
) % bs
->bl
.request_alignment
) {
2344 error_setg(errp
, "Cannot get 'write' permission without 'resize': "
2345 "Image size is not a multiple of request "
2351 /* Check this node */
2356 ret
= bdrv_drv_set_perm(bs
, cumulative_perms
, cumulative_shared_perms
, tran
,
2362 /* Drivers that never have children can omit .bdrv_child_perm() */
2363 if (!drv
->bdrv_child_perm
) {
2364 assert(QLIST_EMPTY(&bs
->children
));
2368 /* Check all children */
2369 QLIST_FOREACH(c
, &bs
->children
, next
) {
2370 uint64_t cur_perm
, cur_shared
;
2372 bdrv_child_perm(bs
, c
->bs
, c
, c
->role
, q
,
2373 cumulative_perms
, cumulative_shared_perms
,
2374 &cur_perm
, &cur_shared
);
2375 bdrv_child_set_perm(c
, cur_perm
, cur_shared
, tran
);
2381 static int bdrv_list_refresh_perms(GSList
*list
, BlockReopenQueue
*q
,
2382 Transaction
*tran
, Error
**errp
)
2385 BlockDriverState
*bs
;
2387 for ( ; list
; list
= list
->next
) {
2390 if (bdrv_parent_perms_conflict(bs
, errp
)) {
2394 ret
= bdrv_node_refresh_perm(bs
, q
, tran
, errp
);
2403 void bdrv_get_cumulative_perm(BlockDriverState
*bs
, uint64_t *perm
,
2404 uint64_t *shared_perm
)
2407 uint64_t cumulative_perms
= 0;
2408 uint64_t cumulative_shared_perms
= BLK_PERM_ALL
;
2410 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
2411 cumulative_perms
|= c
->perm
;
2412 cumulative_shared_perms
&= c
->shared_perm
;
2415 *perm
= cumulative_perms
;
2416 *shared_perm
= cumulative_shared_perms
;
2419 char *bdrv_perm_names(uint64_t perm
)
2425 { BLK_PERM_CONSISTENT_READ
, "consistent read" },
2426 { BLK_PERM_WRITE
, "write" },
2427 { BLK_PERM_WRITE_UNCHANGED
, "write unchanged" },
2428 { BLK_PERM_RESIZE
, "resize" },
2429 { BLK_PERM_GRAPH_MOD
, "change children" },
2433 GString
*result
= g_string_sized_new(30);
2434 struct perm_name
*p
;
2436 for (p
= permissions
; p
->name
; p
++) {
2437 if (perm
& p
->perm
) {
2438 if (result
->len
> 0) {
2439 g_string_append(result
, ", ");
2441 g_string_append(result
, p
->name
);
2445 return g_string_free(result
, FALSE
);
2449 static int bdrv_refresh_perms(BlockDriverState
*bs
, Error
**errp
)
2452 Transaction
*tran
= tran_new();
2453 g_autoptr(GSList
) list
= bdrv_topological_dfs(NULL
, NULL
, bs
);
2455 ret
= bdrv_list_refresh_perms(list
, NULL
, tran
, errp
);
2456 tran_finalize(tran
, ret
);
2461 int bdrv_child_try_set_perm(BdrvChild
*c
, uint64_t perm
, uint64_t shared
,
2464 Error
*local_err
= NULL
;
2465 Transaction
*tran
= tran_new();
2468 bdrv_child_set_perm(c
, perm
, shared
, tran
);
2470 ret
= bdrv_refresh_perms(c
->bs
, &local_err
);
2472 tran_finalize(tran
, ret
);
2475 if ((perm
& ~c
->perm
) || (c
->shared_perm
& ~shared
)) {
2476 /* tighten permissions */
2477 error_propagate(errp
, local_err
);
2480 * Our caller may intend to only loosen restrictions and
2481 * does not expect this function to fail. Errors are not
2482 * fatal in such a case, so we can just hide them from our
2485 error_free(local_err
);
2493 int bdrv_child_refresh_perms(BlockDriverState
*bs
, BdrvChild
*c
, Error
**errp
)
2495 uint64_t parent_perms
, parent_shared
;
2496 uint64_t perms
, shared
;
2498 bdrv_get_cumulative_perm(bs
, &parent_perms
, &parent_shared
);
2499 bdrv_child_perm(bs
, c
->bs
, c
, c
->role
, NULL
,
2500 parent_perms
, parent_shared
, &perms
, &shared
);
2502 return bdrv_child_try_set_perm(c
, perms
, shared
, errp
);
2506 * Default implementation for .bdrv_child_perm() for block filters:
2507 * Forward CONSISTENT_READ, WRITE, WRITE_UNCHANGED, and RESIZE to the
2510 static void bdrv_filter_default_perms(BlockDriverState
*bs
, BdrvChild
*c
,
2512 BlockReopenQueue
*reopen_queue
,
2513 uint64_t perm
, uint64_t shared
,
2514 uint64_t *nperm
, uint64_t *nshared
)
2516 *nperm
= perm
& DEFAULT_PERM_PASSTHROUGH
;
2517 *nshared
= (shared
& DEFAULT_PERM_PASSTHROUGH
) | DEFAULT_PERM_UNCHANGED
;
2520 static void bdrv_default_perms_for_cow(BlockDriverState
*bs
, BdrvChild
*c
,
2522 BlockReopenQueue
*reopen_queue
,
2523 uint64_t perm
, uint64_t shared
,
2524 uint64_t *nperm
, uint64_t *nshared
)
2526 assert(role
& BDRV_CHILD_COW
);
2529 * We want consistent read from backing files if the parent needs it.
2530 * No other operations are performed on backing files.
2532 perm
&= BLK_PERM_CONSISTENT_READ
;
2535 * If the parent can deal with changing data, we're okay with a
2536 * writable and resizable backing file.
2537 * TODO Require !(perm & BLK_PERM_CONSISTENT_READ), too?
2539 if (shared
& BLK_PERM_WRITE
) {
2540 shared
= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2545 shared
|= BLK_PERM_CONSISTENT_READ
| BLK_PERM_GRAPH_MOD
|
2546 BLK_PERM_WRITE_UNCHANGED
;
2548 if (bs
->open_flags
& BDRV_O_INACTIVE
) {
2549 shared
|= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2556 static void bdrv_default_perms_for_storage(BlockDriverState
*bs
, BdrvChild
*c
,
2558 BlockReopenQueue
*reopen_queue
,
2559 uint64_t perm
, uint64_t shared
,
2560 uint64_t *nperm
, uint64_t *nshared
)
2564 assert(role
& (BDRV_CHILD_METADATA
| BDRV_CHILD_DATA
));
2566 flags
= bdrv_reopen_get_flags(reopen_queue
, bs
);
2569 * Apart from the modifications below, the same permissions are
2570 * forwarded and left alone as for filters
2572 bdrv_filter_default_perms(bs
, c
, role
, reopen_queue
,
2573 perm
, shared
, &perm
, &shared
);
2575 if (role
& BDRV_CHILD_METADATA
) {
2576 /* Format drivers may touch metadata even if the guest doesn't write */
2577 if (bdrv_is_writable_after_reopen(bs
, reopen_queue
)) {
2578 perm
|= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2582 * bs->file always needs to be consistent because of the
2583 * metadata. We can never allow other users to resize or write
2586 if (!(flags
& BDRV_O_NO_IO
)) {
2587 perm
|= BLK_PERM_CONSISTENT_READ
;
2589 shared
&= ~(BLK_PERM_WRITE
| BLK_PERM_RESIZE
);
2592 if (role
& BDRV_CHILD_DATA
) {
2594 * Technically, everything in this block is a subset of the
2595 * BDRV_CHILD_METADATA path taken above, and so this could
2596 * be an "else if" branch. However, that is not obvious, and
2597 * this function is not performance critical, therefore we let
2598 * this be an independent "if".
2602 * We cannot allow other users to resize the file because the
2603 * format driver might have some assumptions about the size
2604 * (e.g. because it is stored in metadata, or because the file
2605 * is split into fixed-size data files).
2607 shared
&= ~BLK_PERM_RESIZE
;
2610 * WRITE_UNCHANGED often cannot be performed as such on the
2611 * data file. For example, the qcow2 driver may still need to
2612 * write copied clusters on copy-on-read.
2614 if (perm
& BLK_PERM_WRITE_UNCHANGED
) {
2615 perm
|= BLK_PERM_WRITE
;
2619 * If the data file is written to, the format driver may
2620 * expect to be able to resize it by writing beyond the EOF.
2622 if (perm
& BLK_PERM_WRITE
) {
2623 perm
|= BLK_PERM_RESIZE
;
2627 if (bs
->open_flags
& BDRV_O_INACTIVE
) {
2628 shared
|= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2635 void bdrv_default_perms(BlockDriverState
*bs
, BdrvChild
*c
,
2636 BdrvChildRole role
, BlockReopenQueue
*reopen_queue
,
2637 uint64_t perm
, uint64_t shared
,
2638 uint64_t *nperm
, uint64_t *nshared
)
2640 if (role
& BDRV_CHILD_FILTERED
) {
2641 assert(!(role
& (BDRV_CHILD_DATA
| BDRV_CHILD_METADATA
|
2643 bdrv_filter_default_perms(bs
, c
, role
, reopen_queue
,
2644 perm
, shared
, nperm
, nshared
);
2645 } else if (role
& BDRV_CHILD_COW
) {
2646 assert(!(role
& (BDRV_CHILD_DATA
| BDRV_CHILD_METADATA
)));
2647 bdrv_default_perms_for_cow(bs
, c
, role
, reopen_queue
,
2648 perm
, shared
, nperm
, nshared
);
2649 } else if (role
& (BDRV_CHILD_METADATA
| BDRV_CHILD_DATA
)) {
2650 bdrv_default_perms_for_storage(bs
, c
, role
, reopen_queue
,
2651 perm
, shared
, nperm
, nshared
);
2653 g_assert_not_reached();
2657 uint64_t bdrv_qapi_perm_to_blk_perm(BlockPermission qapi_perm
)
2659 static const uint64_t permissions
[] = {
2660 [BLOCK_PERMISSION_CONSISTENT_READ
] = BLK_PERM_CONSISTENT_READ
,
2661 [BLOCK_PERMISSION_WRITE
] = BLK_PERM_WRITE
,
2662 [BLOCK_PERMISSION_WRITE_UNCHANGED
] = BLK_PERM_WRITE_UNCHANGED
,
2663 [BLOCK_PERMISSION_RESIZE
] = BLK_PERM_RESIZE
,
2664 [BLOCK_PERMISSION_GRAPH_MOD
] = BLK_PERM_GRAPH_MOD
,
2667 QEMU_BUILD_BUG_ON(ARRAY_SIZE(permissions
) != BLOCK_PERMISSION__MAX
);
2668 QEMU_BUILD_BUG_ON(1UL << ARRAY_SIZE(permissions
) != BLK_PERM_ALL
+ 1);
2670 assert(qapi_perm
< BLOCK_PERMISSION__MAX
);
2672 return permissions
[qapi_perm
];
2675 static void bdrv_replace_child_noperm(BdrvChild
*child
,
2676 BlockDriverState
*new_bs
)
2678 BlockDriverState
*old_bs
= child
->bs
;
2679 int new_bs_quiesce_counter
;
2682 assert(!child
->frozen
);
2683 assert(old_bs
!= new_bs
);
2685 if (old_bs
&& new_bs
) {
2686 assert(bdrv_get_aio_context(old_bs
) == bdrv_get_aio_context(new_bs
));
2689 new_bs_quiesce_counter
= (new_bs
? new_bs
->quiesce_counter
: 0);
2690 drain_saldo
= new_bs_quiesce_counter
- child
->parent_quiesce_counter
;
2693 * If the new child node is drained but the old one was not, flush
2694 * all outstanding requests to the old child node.
2696 while (drain_saldo
> 0 && child
->klass
->drained_begin
) {
2697 bdrv_parent_drained_begin_single(child
, true);
2702 /* Detach first so that the recursive drain sections coming from @child
2703 * are already gone and we only end the drain sections that came from
2705 if (child
->klass
->detach
) {
2706 child
->klass
->detach(child
);
2708 QLIST_REMOVE(child
, next_parent
);
2714 QLIST_INSERT_HEAD(&new_bs
->parents
, child
, next_parent
);
2717 * Detaching the old node may have led to the new node's
2718 * quiesce_counter having been decreased. Not a problem, we
2719 * just need to recognize this here and then invoke
2720 * drained_end appropriately more often.
2722 assert(new_bs
->quiesce_counter
<= new_bs_quiesce_counter
);
2723 drain_saldo
+= new_bs
->quiesce_counter
- new_bs_quiesce_counter
;
2725 /* Attach only after starting new drained sections, so that recursive
2726 * drain sections coming from @child don't get an extra .drained_begin
2728 if (child
->klass
->attach
) {
2729 child
->klass
->attach(child
);
2734 * If the old child node was drained but the new one is not, allow
2735 * requests to come in only after the new node has been attached.
2737 while (drain_saldo
< 0 && child
->klass
->drained_end
) {
2738 bdrv_parent_drained_end_single(child
);
2744 * Free the given @child.
2746 * The child must be empty (i.e. `child->bs == NULL`) and it must be
2747 * unused (i.e. not in a children list).
2749 static void bdrv_child_free(BdrvChild
*child
)
2752 assert(!child
->next
.le_prev
); /* not in children list */
2754 g_free(child
->name
);
2758 typedef struct BdrvAttachChildCommonState
{
2760 AioContext
*old_parent_ctx
;
2761 AioContext
*old_child_ctx
;
2762 } BdrvAttachChildCommonState
;
2764 static void bdrv_attach_child_common_abort(void *opaque
)
2766 BdrvAttachChildCommonState
*s
= opaque
;
2767 BdrvChild
*child
= *s
->child
;
2768 BlockDriverState
*bs
= child
->bs
;
2770 bdrv_replace_child_noperm(child
, NULL
);
2772 if (bdrv_get_aio_context(bs
) != s
->old_child_ctx
) {
2773 bdrv_try_set_aio_context(bs
, s
->old_child_ctx
, &error_abort
);
2776 if (bdrv_child_get_parent_aio_context(child
) != s
->old_parent_ctx
) {
2777 GSList
*ignore
= g_slist_prepend(NULL
, child
);
2779 child
->klass
->can_set_aio_ctx(child
, s
->old_parent_ctx
, &ignore
,
2781 g_slist_free(ignore
);
2782 ignore
= g_slist_prepend(NULL
, child
);
2783 child
->klass
->set_aio_ctx(child
, s
->old_parent_ctx
, &ignore
);
2785 g_slist_free(ignore
);
2789 bdrv_child_free(child
);
2793 static TransactionActionDrv bdrv_attach_child_common_drv
= {
2794 .abort
= bdrv_attach_child_common_abort
,
2799 * Common part of attaching bdrv child to bs or to blk or to job
2801 * Resulting new child is returned through @child.
2802 * At start *@child must be NULL.
2803 * @child is saved to a new entry of @tran, so that *@child could be reverted to
2804 * NULL on abort(). So referenced variable must live at least until transaction
2807 * Function doesn't update permissions, caller is responsible for this.
2809 static int bdrv_attach_child_common(BlockDriverState
*child_bs
,
2810 const char *child_name
,
2811 const BdrvChildClass
*child_class
,
2812 BdrvChildRole child_role
,
2813 uint64_t perm
, uint64_t shared_perm
,
2814 void *opaque
, BdrvChild
**child
,
2815 Transaction
*tran
, Error
**errp
)
2817 BdrvChild
*new_child
;
2818 AioContext
*parent_ctx
;
2819 AioContext
*child_ctx
= bdrv_get_aio_context(child_bs
);
2822 assert(*child
== NULL
);
2823 assert(child_class
->get_parent_desc
);
2825 new_child
= g_new(BdrvChild
, 1);
2826 *new_child
= (BdrvChild
) {
2828 .name
= g_strdup(child_name
),
2829 .klass
= child_class
,
2832 .shared_perm
= shared_perm
,
2837 * If the AioContexts don't match, first try to move the subtree of
2838 * child_bs into the AioContext of the new parent. If this doesn't work,
2839 * try moving the parent into the AioContext of child_bs instead.
2841 parent_ctx
= bdrv_child_get_parent_aio_context(new_child
);
2842 if (child_ctx
!= parent_ctx
) {
2843 Error
*local_err
= NULL
;
2844 int ret
= bdrv_try_set_aio_context(child_bs
, parent_ctx
, &local_err
);
2846 if (ret
< 0 && child_class
->can_set_aio_ctx
) {
2847 GSList
*ignore
= g_slist_prepend(NULL
, new_child
);
2848 if (child_class
->can_set_aio_ctx(new_child
, child_ctx
, &ignore
,
2851 error_free(local_err
);
2853 g_slist_free(ignore
);
2854 ignore
= g_slist_prepend(NULL
, new_child
);
2855 child_class
->set_aio_ctx(new_child
, child_ctx
, &ignore
);
2857 g_slist_free(ignore
);
2861 error_propagate(errp
, local_err
);
2862 bdrv_child_free(new_child
);
2868 bdrv_replace_child_noperm(new_child
, child_bs
);
2872 BdrvAttachChildCommonState
*s
= g_new(BdrvAttachChildCommonState
, 1);
2873 *s
= (BdrvAttachChildCommonState
) {
2875 .old_parent_ctx
= parent_ctx
,
2876 .old_child_ctx
= child_ctx
,
2878 tran_add(tran
, &bdrv_attach_child_common_drv
, s
);
2884 * Variable referenced by @child must live at least until transaction end.
2885 * (see bdrv_attach_child_common() doc for details)
2887 * Function doesn't update permissions, caller is responsible for this.
2889 static int bdrv_attach_child_noperm(BlockDriverState
*parent_bs
,
2890 BlockDriverState
*child_bs
,
2891 const char *child_name
,
2892 const BdrvChildClass
*child_class
,
2893 BdrvChildRole child_role
,
2899 uint64_t perm
, shared_perm
;
2901 assert(parent_bs
->drv
);
2903 if (bdrv_recurse_has_child(child_bs
, parent_bs
)) {
2904 error_setg(errp
, "Making '%s' a %s child of '%s' would create a cycle",
2905 child_bs
->node_name
, child_name
, parent_bs
->node_name
);
2909 bdrv_get_cumulative_perm(parent_bs
, &perm
, &shared_perm
);
2910 bdrv_child_perm(parent_bs
, child_bs
, NULL
, child_role
, NULL
,
2911 perm
, shared_perm
, &perm
, &shared_perm
);
2913 ret
= bdrv_attach_child_common(child_bs
, child_name
, child_class
,
2914 child_role
, perm
, shared_perm
, parent_bs
,
2923 static void bdrv_detach_child(BdrvChild
*child
)
2925 BlockDriverState
*old_bs
= child
->bs
;
2927 bdrv_replace_child_noperm(child
, NULL
);
2928 bdrv_child_free(child
);
2932 * Update permissions for old node. We're just taking a parent away, so
2933 * we're loosening restrictions. Errors of permission update are not
2934 * fatal in this case, ignore them.
2936 bdrv_refresh_perms(old_bs
, NULL
);
2939 * When the parent requiring a non-default AioContext is removed, the
2940 * node moves back to the main AioContext
2942 bdrv_try_set_aio_context(old_bs
, qemu_get_aio_context(), NULL
);
2947 * This function steals the reference to child_bs from the caller.
2948 * That reference is later dropped by bdrv_root_unref_child().
2950 * On failure NULL is returned, errp is set and the reference to
2951 * child_bs is also dropped.
2953 * The caller must hold the AioContext lock @child_bs, but not that of @ctx
2954 * (unless @child_bs is already in @ctx).
2956 BdrvChild
*bdrv_root_attach_child(BlockDriverState
*child_bs
,
2957 const char *child_name
,
2958 const BdrvChildClass
*child_class
,
2959 BdrvChildRole child_role
,
2960 uint64_t perm
, uint64_t shared_perm
,
2961 void *opaque
, Error
**errp
)
2964 BdrvChild
*child
= NULL
;
2965 Transaction
*tran
= tran_new();
2967 ret
= bdrv_attach_child_common(child_bs
, child_name
, child_class
,
2968 child_role
, perm
, shared_perm
, opaque
,
2969 &child
, tran
, errp
);
2974 ret
= bdrv_refresh_perms(child_bs
, errp
);
2977 tran_finalize(tran
, ret
);
2978 /* child is unset on failure by bdrv_attach_child_common_abort() */
2979 assert((ret
< 0) == !child
);
2981 bdrv_unref(child_bs
);
2986 * This function transfers the reference to child_bs from the caller
2987 * to parent_bs. That reference is later dropped by parent_bs on
2988 * bdrv_close() or if someone calls bdrv_unref_child().
2990 * On failure NULL is returned, errp is set and the reference to
2991 * child_bs is also dropped.
2993 * If @parent_bs and @child_bs are in different AioContexts, the caller must
2994 * hold the AioContext lock for @child_bs, but not for @parent_bs.
2996 BdrvChild
*bdrv_attach_child(BlockDriverState
*parent_bs
,
2997 BlockDriverState
*child_bs
,
2998 const char *child_name
,
2999 const BdrvChildClass
*child_class
,
3000 BdrvChildRole child_role
,
3004 BdrvChild
*child
= NULL
;
3005 Transaction
*tran
= tran_new();
3007 ret
= bdrv_attach_child_noperm(parent_bs
, child_bs
, child_name
, child_class
,
3008 child_role
, &child
, tran
, errp
);
3013 ret
= bdrv_refresh_perms(parent_bs
, errp
);
3019 tran_finalize(tran
, ret
);
3020 /* child is unset on failure by bdrv_attach_child_common_abort() */
3021 assert((ret
< 0) == !child
);
3023 bdrv_unref(child_bs
);
3028 /* Callers must ensure that child->frozen is false. */
3029 void bdrv_root_unref_child(BdrvChild
*child
)
3031 BlockDriverState
*child_bs
;
3033 child_bs
= child
->bs
;
3034 bdrv_detach_child(child
);
3035 bdrv_unref(child_bs
);
3038 typedef struct BdrvSetInheritsFrom
{
3039 BlockDriverState
*bs
;
3040 BlockDriverState
*old_inherits_from
;
3041 } BdrvSetInheritsFrom
;
3043 static void bdrv_set_inherits_from_abort(void *opaque
)
3045 BdrvSetInheritsFrom
*s
= opaque
;
3047 s
->bs
->inherits_from
= s
->old_inherits_from
;
3050 static TransactionActionDrv bdrv_set_inherits_from_drv
= {
3051 .abort
= bdrv_set_inherits_from_abort
,
3055 /* @tran is allowed to be NULL. In this case no rollback is possible */
3056 static void bdrv_set_inherits_from(BlockDriverState
*bs
,
3057 BlockDriverState
*new_inherits_from
,
3061 BdrvSetInheritsFrom
*s
= g_new(BdrvSetInheritsFrom
, 1);
3063 *s
= (BdrvSetInheritsFrom
) {
3065 .old_inherits_from
= bs
->inherits_from
,
3068 tran_add(tran
, &bdrv_set_inherits_from_drv
, s
);
3071 bs
->inherits_from
= new_inherits_from
;
3075 * Clear all inherits_from pointers from children and grandchildren of
3076 * @root that point to @root, where necessary.
3077 * @tran is allowed to be NULL. In this case no rollback is possible
3079 static void bdrv_unset_inherits_from(BlockDriverState
*root
, BdrvChild
*child
,
3084 if (child
->bs
->inherits_from
== root
) {
3086 * Remove inherits_from only when the last reference between root and
3087 * child->bs goes away.
3089 QLIST_FOREACH(c
, &root
->children
, next
) {
3090 if (c
!= child
&& c
->bs
== child
->bs
) {
3095 bdrv_set_inherits_from(child
->bs
, NULL
, tran
);
3099 QLIST_FOREACH(c
, &child
->bs
->children
, next
) {
3100 bdrv_unset_inherits_from(root
, c
, tran
);
3104 /* Callers must ensure that child->frozen is false. */
3105 void bdrv_unref_child(BlockDriverState
*parent
, BdrvChild
*child
)
3107 if (child
== NULL
) {
3111 bdrv_unset_inherits_from(parent
, child
, NULL
);
3112 bdrv_root_unref_child(child
);
3116 static void bdrv_parent_cb_change_media(BlockDriverState
*bs
, bool load
)
3119 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
3120 if (c
->klass
->change_media
) {
3121 c
->klass
->change_media(c
, load
);
3126 /* Return true if you can reach parent going through child->inherits_from
3127 * recursively. If parent or child are NULL, return false */
3128 static bool bdrv_inherits_from_recursive(BlockDriverState
*child
,
3129 BlockDriverState
*parent
)
3131 while (child
&& child
!= parent
) {
3132 child
= child
->inherits_from
;
3135 return child
!= NULL
;
3139 * Return the BdrvChildRole for @bs's backing child. bs->backing is
3140 * mostly used for COW backing children (role = COW), but also for
3141 * filtered children (role = FILTERED | PRIMARY).
3143 static BdrvChildRole
bdrv_backing_role(BlockDriverState
*bs
)
3145 if (bs
->drv
&& bs
->drv
->is_filter
) {
3146 return BDRV_CHILD_FILTERED
| BDRV_CHILD_PRIMARY
;
3148 return BDRV_CHILD_COW
;
3153 * Sets the bs->backing or bs->file link of a BDS. A new reference is created;
3154 * callers which don't need their own reference any more must call bdrv_unref().
3156 * Function doesn't update permissions, caller is responsible for this.
3158 static int bdrv_set_file_or_backing_noperm(BlockDriverState
*parent_bs
,
3159 BlockDriverState
*child_bs
,
3161 Transaction
*tran
, Error
**errp
)
3164 bool update_inherits_from
=
3165 bdrv_inherits_from_recursive(child_bs
, parent_bs
);
3166 BdrvChild
*child
= is_backing
? parent_bs
->backing
: parent_bs
->file
;
3169 if (!parent_bs
->drv
) {
3171 * Node without drv is an object without a class :/. TODO: finally fix
3172 * qcow2 driver to never clear bs->drv and implement format corruption
3173 * handling in other way.
3175 error_setg(errp
, "Node corrupted");
3179 if (child
&& child
->frozen
) {
3180 error_setg(errp
, "Cannot change frozen '%s' link from '%s' to '%s'",
3181 child
->name
, parent_bs
->node_name
, child
->bs
->node_name
);
3185 if (is_backing
&& !parent_bs
->drv
->is_filter
&&
3186 !parent_bs
->drv
->supports_backing
)
3188 error_setg(errp
, "Driver '%s' of node '%s' does not support backing "
3189 "files", parent_bs
->drv
->format_name
, parent_bs
->node_name
);
3193 if (parent_bs
->drv
->is_filter
) {
3194 role
= BDRV_CHILD_FILTERED
| BDRV_CHILD_PRIMARY
;
3195 } else if (is_backing
) {
3196 role
= BDRV_CHILD_COW
;
3199 * We only can use same role as it is in existing child. We don't have
3200 * infrastructure to determine role of file child in generic way
3203 error_setg(errp
, "Cannot set file child to format node without "
3211 bdrv_unset_inherits_from(parent_bs
, child
, tran
);
3212 bdrv_remove_file_or_backing_child(parent_bs
, child
, tran
);
3219 ret
= bdrv_attach_child_noperm(parent_bs
, child_bs
,
3220 is_backing
? "backing" : "file",
3221 &child_of_bds
, role
,
3222 is_backing
? &parent_bs
->backing
:
3231 * If inherits_from pointed recursively to bs then let's update it to
3232 * point directly to bs (else it will become NULL).
3234 if (update_inherits_from
) {
3235 bdrv_set_inherits_from(child_bs
, parent_bs
, tran
);
3239 bdrv_refresh_limits(parent_bs
, tran
, NULL
);
3244 static int bdrv_set_backing_noperm(BlockDriverState
*bs
,
3245 BlockDriverState
*backing_hd
,
3246 Transaction
*tran
, Error
**errp
)
3248 return bdrv_set_file_or_backing_noperm(bs
, backing_hd
, true, tran
, errp
);
3251 int bdrv_set_backing_hd(BlockDriverState
*bs
, BlockDriverState
*backing_hd
,
3255 Transaction
*tran
= tran_new();
3257 ret
= bdrv_set_backing_noperm(bs
, backing_hd
, tran
, errp
);
3262 ret
= bdrv_refresh_perms(bs
, errp
);
3264 tran_finalize(tran
, ret
);
3270 * Opens the backing file for a BlockDriverState if not yet open
3272 * bdref_key specifies the key for the image's BlockdevRef in the options QDict.
3273 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
3274 * itself, all options starting with "${bdref_key}." are considered part of the
3277 * TODO Can this be unified with bdrv_open_image()?
3279 int bdrv_open_backing_file(BlockDriverState
*bs
, QDict
*parent_options
,
3280 const char *bdref_key
, Error
**errp
)
3282 char *backing_filename
= NULL
;
3283 char *bdref_key_dot
;
3284 const char *reference
= NULL
;
3286 bool implicit_backing
= false;
3287 BlockDriverState
*backing_hd
;
3289 QDict
*tmp_parent_options
= NULL
;
3290 Error
*local_err
= NULL
;
3292 if (bs
->backing
!= NULL
) {
3296 /* NULL means an empty set of options */
3297 if (parent_options
== NULL
) {
3298 tmp_parent_options
= qdict_new();
3299 parent_options
= tmp_parent_options
;
3302 bs
->open_flags
&= ~BDRV_O_NO_BACKING
;
3304 bdref_key_dot
= g_strdup_printf("%s.", bdref_key
);
3305 qdict_extract_subqdict(parent_options
, &options
, bdref_key_dot
);
3306 g_free(bdref_key_dot
);
3309 * Caution: while qdict_get_try_str() is fine, getting non-string
3310 * types would require more care. When @parent_options come from
3311 * -blockdev or blockdev_add, its members are typed according to
3312 * the QAPI schema, but when they come from -drive, they're all
3315 reference
= qdict_get_try_str(parent_options
, bdref_key
);
3316 if (reference
|| qdict_haskey(options
, "file.filename")) {
3317 /* keep backing_filename NULL */
3318 } else if (bs
->backing_file
[0] == '\0' && qdict_size(options
) == 0) {
3319 qobject_unref(options
);
3322 if (qdict_size(options
) == 0) {
3323 /* If the user specifies options that do not modify the
3324 * backing file's behavior, we might still consider it the
3325 * implicit backing file. But it's easier this way, and
3326 * just specifying some of the backing BDS's options is
3327 * only possible with -drive anyway (otherwise the QAPI
3328 * schema forces the user to specify everything). */
3329 implicit_backing
= !strcmp(bs
->auto_backing_file
, bs
->backing_file
);
3332 backing_filename
= bdrv_get_full_backing_filename(bs
, &local_err
);
3335 error_propagate(errp
, local_err
);
3336 qobject_unref(options
);
3341 if (!bs
->drv
|| !bs
->drv
->supports_backing
) {
3343 error_setg(errp
, "Driver doesn't support backing files");
3344 qobject_unref(options
);
3349 bs
->backing_format
[0] != '\0' && !qdict_haskey(options
, "driver")) {
3350 qdict_put_str(options
, "driver", bs
->backing_format
);
3353 backing_hd
= bdrv_open_inherit(backing_filename
, reference
, options
, 0, bs
,
3354 &child_of_bds
, bdrv_backing_role(bs
), errp
);
3356 bs
->open_flags
|= BDRV_O_NO_BACKING
;
3357 error_prepend(errp
, "Could not open backing file: ");
3362 if (implicit_backing
) {
3363 bdrv_refresh_filename(backing_hd
);
3364 pstrcpy(bs
->auto_backing_file
, sizeof(bs
->auto_backing_file
),
3365 backing_hd
->filename
);
3368 /* Hook up the backing file link; drop our reference, bs owns the
3369 * backing_hd reference now */
3370 ret
= bdrv_set_backing_hd(bs
, backing_hd
, errp
);
3371 bdrv_unref(backing_hd
);
3376 qdict_del(parent_options
, bdref_key
);
3379 g_free(backing_filename
);
3380 qobject_unref(tmp_parent_options
);
3384 static BlockDriverState
*
3385 bdrv_open_child_bs(const char *filename
, QDict
*options
, const char *bdref_key
,
3386 BlockDriverState
*parent
, const BdrvChildClass
*child_class
,
3387 BdrvChildRole child_role
, bool allow_none
, Error
**errp
)
3389 BlockDriverState
*bs
= NULL
;
3390 QDict
*image_options
;
3391 char *bdref_key_dot
;
3392 const char *reference
;
3394 assert(child_class
!= NULL
);
3396 bdref_key_dot
= g_strdup_printf("%s.", bdref_key
);
3397 qdict_extract_subqdict(options
, &image_options
, bdref_key_dot
);
3398 g_free(bdref_key_dot
);
3401 * Caution: while qdict_get_try_str() is fine, getting non-string
3402 * types would require more care. When @options come from
3403 * -blockdev or blockdev_add, its members are typed according to
3404 * the QAPI schema, but when they come from -drive, they're all
3407 reference
= qdict_get_try_str(options
, bdref_key
);
3408 if (!filename
&& !reference
&& !qdict_size(image_options
)) {
3410 error_setg(errp
, "A block device must be specified for \"%s\"",
3413 qobject_unref(image_options
);
3417 bs
= bdrv_open_inherit(filename
, reference
, image_options
, 0,
3418 parent
, child_class
, child_role
, errp
);
3424 qdict_del(options
, bdref_key
);
3429 * Opens a disk image whose options are given as BlockdevRef in another block
3432 * If allow_none is true, no image will be opened if filename is false and no
3433 * BlockdevRef is given. NULL will be returned, but errp remains unset.
3435 * bdrev_key specifies the key for the image's BlockdevRef in the options QDict.
3436 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
3437 * itself, all options starting with "${bdref_key}." are considered part of the
3440 * The BlockdevRef will be removed from the options QDict.
3442 BdrvChild
*bdrv_open_child(const char *filename
,
3443 QDict
*options
, const char *bdref_key
,
3444 BlockDriverState
*parent
,
3445 const BdrvChildClass
*child_class
,
3446 BdrvChildRole child_role
,
3447 bool allow_none
, Error
**errp
)
3449 BlockDriverState
*bs
;
3451 bs
= bdrv_open_child_bs(filename
, options
, bdref_key
, parent
, child_class
,
3452 child_role
, allow_none
, errp
);
3457 return bdrv_attach_child(parent
, bs
, bdref_key
, child_class
, child_role
,
3462 * TODO Future callers may need to specify parent/child_class in order for
3463 * option inheritance to work. Existing callers use it for the root node.
3465 BlockDriverState
*bdrv_open_blockdev_ref(BlockdevRef
*ref
, Error
**errp
)
3467 BlockDriverState
*bs
= NULL
;
3468 QObject
*obj
= NULL
;
3469 QDict
*qdict
= NULL
;
3470 const char *reference
= NULL
;
3473 if (ref
->type
== QTYPE_QSTRING
) {
3474 reference
= ref
->u
.reference
;
3476 BlockdevOptions
*options
= &ref
->u
.definition
;
3477 assert(ref
->type
== QTYPE_QDICT
);
3479 v
= qobject_output_visitor_new(&obj
);
3480 visit_type_BlockdevOptions(v
, NULL
, &options
, &error_abort
);
3481 visit_complete(v
, &obj
);
3483 qdict
= qobject_to(QDict
, obj
);
3484 qdict_flatten(qdict
);
3486 /* bdrv_open_inherit() defaults to the values in bdrv_flags (for
3487 * compatibility with other callers) rather than what we want as the
3488 * real defaults. Apply the defaults here instead. */
3489 qdict_set_default_str(qdict
, BDRV_OPT_CACHE_DIRECT
, "off");
3490 qdict_set_default_str(qdict
, BDRV_OPT_CACHE_NO_FLUSH
, "off");
3491 qdict_set_default_str(qdict
, BDRV_OPT_READ_ONLY
, "off");
3492 qdict_set_default_str(qdict
, BDRV_OPT_AUTO_READ_ONLY
, "off");
3496 bs
= bdrv_open_inherit(NULL
, reference
, qdict
, 0, NULL
, NULL
, 0, errp
);
3503 static BlockDriverState
*bdrv_append_temp_snapshot(BlockDriverState
*bs
,
3505 QDict
*snapshot_options
,
3508 /* TODO: extra byte is a hack to ensure MAX_PATH space on Windows. */
3509 char *tmp_filename
= g_malloc0(PATH_MAX
+ 1);
3511 QemuOpts
*opts
= NULL
;
3512 BlockDriverState
*bs_snapshot
= NULL
;
3515 /* if snapshot, we create a temporary backing file and open it
3516 instead of opening 'filename' directly */
3518 /* Get the required size from the image */
3519 total_size
= bdrv_getlength(bs
);
3520 if (total_size
< 0) {
3521 error_setg_errno(errp
, -total_size
, "Could not get image size");
3525 /* Create the temporary image */
3526 ret
= get_tmp_filename(tmp_filename
, PATH_MAX
+ 1);
3528 error_setg_errno(errp
, -ret
, "Could not get temporary filename");
3532 opts
= qemu_opts_create(bdrv_qcow2
.create_opts
, NULL
, 0,
3534 qemu_opt_set_number(opts
, BLOCK_OPT_SIZE
, total_size
, &error_abort
);
3535 ret
= bdrv_create(&bdrv_qcow2
, tmp_filename
, opts
, errp
);
3536 qemu_opts_del(opts
);
3538 error_prepend(errp
, "Could not create temporary overlay '%s': ",
3543 /* Prepare options QDict for the temporary file */
3544 qdict_put_str(snapshot_options
, "file.driver", "file");
3545 qdict_put_str(snapshot_options
, "file.filename", tmp_filename
);
3546 qdict_put_str(snapshot_options
, "driver", "qcow2");
3548 bs_snapshot
= bdrv_open(NULL
, NULL
, snapshot_options
, flags
, errp
);
3549 snapshot_options
= NULL
;
3554 ret
= bdrv_append(bs_snapshot
, bs
, errp
);
3561 qobject_unref(snapshot_options
);
3562 g_free(tmp_filename
);
3567 * Opens a disk image (raw, qcow2, vmdk, ...)
3569 * options is a QDict of options to pass to the block drivers, or NULL for an
3570 * empty set of options. The reference to the QDict belongs to the block layer
3571 * after the call (even on failure), so if the caller intends to reuse the
3572 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
3574 * If *pbs is NULL, a new BDS will be created with a pointer to it stored there.
3575 * If it is not NULL, the referenced BDS will be reused.
3577 * The reference parameter may be used to specify an existing block device which
3578 * should be opened. If specified, neither options nor a filename may be given,
3579 * nor can an existing BDS be reused (that is, *pbs has to be NULL).
3581 static BlockDriverState
*bdrv_open_inherit(const char *filename
,
3582 const char *reference
,
3583 QDict
*options
, int flags
,
3584 BlockDriverState
*parent
,
3585 const BdrvChildClass
*child_class
,
3586 BdrvChildRole child_role
,
3590 BlockBackend
*file
= NULL
;
3591 BlockDriverState
*bs
;
3592 BlockDriver
*drv
= NULL
;
3594 const char *drvname
;
3595 const char *backing
;
3596 Error
*local_err
= NULL
;
3597 QDict
*snapshot_options
= NULL
;
3598 int snapshot_flags
= 0;
3600 assert(!child_class
|| !flags
);
3601 assert(!child_class
== !parent
);
3604 bool options_non_empty
= options
? qdict_size(options
) : false;
3605 qobject_unref(options
);
3607 if (filename
|| options_non_empty
) {
3608 error_setg(errp
, "Cannot reference an existing block device with "
3609 "additional options or a new filename");
3613 bs
= bdrv_lookup_bs(reference
, reference
, errp
);
3624 /* NULL means an empty set of options */
3625 if (options
== NULL
) {
3626 options
= qdict_new();
3629 /* json: syntax counts as explicit options, as if in the QDict */
3630 parse_json_protocol(options
, &filename
, &local_err
);
3635 bs
->explicit_options
= qdict_clone_shallow(options
);
3638 bool parent_is_format
;
3641 parent_is_format
= parent
->drv
->is_format
;
3644 * parent->drv is not set yet because this node is opened for
3645 * (potential) format probing. That means that @parent is going
3646 * to be a format node.
3648 parent_is_format
= true;
3651 bs
->inherits_from
= parent
;
3652 child_class
->inherit_options(child_role
, parent_is_format
,
3654 parent
->open_flags
, parent
->options
);
3657 ret
= bdrv_fill_options(&options
, filename
, &flags
, &local_err
);
3663 * Set the BDRV_O_RDWR and BDRV_O_ALLOW_RDWR flags.
3664 * Caution: getting a boolean member of @options requires care.
3665 * When @options come from -blockdev or blockdev_add, members are
3666 * typed according to the QAPI schema, but when they come from
3667 * -drive, they're all QString.
3669 if (g_strcmp0(qdict_get_try_str(options
, BDRV_OPT_READ_ONLY
), "on") &&
3670 !qdict_get_try_bool(options
, BDRV_OPT_READ_ONLY
, false)) {
3671 flags
|= (BDRV_O_RDWR
| BDRV_O_ALLOW_RDWR
);
3673 flags
&= ~BDRV_O_RDWR
;
3676 if (flags
& BDRV_O_SNAPSHOT
) {
3677 snapshot_options
= qdict_new();
3678 bdrv_temp_snapshot_options(&snapshot_flags
, snapshot_options
,
3680 /* Let bdrv_backing_options() override "read-only" */
3681 qdict_del(options
, BDRV_OPT_READ_ONLY
);
3682 bdrv_inherited_options(BDRV_CHILD_COW
, true,
3683 &flags
, options
, flags
, options
);
3686 bs
->open_flags
= flags
;
3687 bs
->options
= options
;
3688 options
= qdict_clone_shallow(options
);
3690 /* Find the right image format driver */
3691 /* See cautionary note on accessing @options above */
3692 drvname
= qdict_get_try_str(options
, "driver");
3694 drv
= bdrv_find_format(drvname
);
3696 error_setg(errp
, "Unknown driver: '%s'", drvname
);
3701 assert(drvname
|| !(flags
& BDRV_O_PROTOCOL
));
3703 /* See cautionary note on accessing @options above */
3704 backing
= qdict_get_try_str(options
, "backing");
3705 if (qobject_to(QNull
, qdict_get(options
, "backing")) != NULL
||
3706 (backing
&& *backing
== '\0'))
3709 warn_report("Use of \"backing\": \"\" is deprecated; "
3710 "use \"backing\": null instead");
3712 flags
|= BDRV_O_NO_BACKING
;
3713 qdict_del(bs
->explicit_options
, "backing");
3714 qdict_del(bs
->options
, "backing");
3715 qdict_del(options
, "backing");
3718 /* Open image file without format layer. This BlockBackend is only used for
3719 * probing, the block drivers will do their own bdrv_open_child() for the
3720 * same BDS, which is why we put the node name back into options. */
3721 if ((flags
& BDRV_O_PROTOCOL
) == 0) {
3722 BlockDriverState
*file_bs
;
3724 file_bs
= bdrv_open_child_bs(filename
, options
, "file", bs
,
3725 &child_of_bds
, BDRV_CHILD_IMAGE
,
3730 if (file_bs
!= NULL
) {
3731 /* Not requesting BLK_PERM_CONSISTENT_READ because we're only
3732 * looking at the header to guess the image format. This works even
3733 * in cases where a guest would not see a consistent state. */
3734 file
= blk_new(bdrv_get_aio_context(file_bs
), 0, BLK_PERM_ALL
);
3735 blk_insert_bs(file
, file_bs
, &local_err
);
3736 bdrv_unref(file_bs
);
3741 qdict_put_str(options
, "file", bdrv_get_node_name(file_bs
));
3745 /* Image format probing */
3748 ret
= find_image_format(file
, filename
, &drv
, &local_err
);
3753 * This option update would logically belong in bdrv_fill_options(),
3754 * but we first need to open bs->file for the probing to work, while
3755 * opening bs->file already requires the (mostly) final set of options
3756 * so that cache mode etc. can be inherited.
3758 * Adding the driver later is somewhat ugly, but it's not an option
3759 * that would ever be inherited, so it's correct. We just need to make
3760 * sure to update both bs->options (which has the full effective
3761 * options for bs) and options (which has file.* already removed).
3763 qdict_put_str(bs
->options
, "driver", drv
->format_name
);
3764 qdict_put_str(options
, "driver", drv
->format_name
);
3766 error_setg(errp
, "Must specify either driver or file");
3770 /* BDRV_O_PROTOCOL must be set iff a protocol BDS is about to be created */
3771 assert(!!(flags
& BDRV_O_PROTOCOL
) == !!drv
->bdrv_file_open
);
3772 /* file must be NULL if a protocol BDS is about to be created
3773 * (the inverse results in an error message from bdrv_open_common()) */
3774 assert(!(flags
& BDRV_O_PROTOCOL
) || !file
);
3776 /* Open the image */
3777 ret
= bdrv_open_common(bs
, file
, options
, &local_err
);
3787 /* If there is a backing file, use it */
3788 if ((flags
& BDRV_O_NO_BACKING
) == 0) {
3789 ret
= bdrv_open_backing_file(bs
, options
, "backing", &local_err
);
3791 goto close_and_fail
;
3795 /* Remove all children options and references
3796 * from bs->options and bs->explicit_options */
3797 QLIST_FOREACH(child
, &bs
->children
, next
) {
3798 char *child_key_dot
;
3799 child_key_dot
= g_strdup_printf("%s.", child
->name
);
3800 qdict_extract_subqdict(bs
->explicit_options
, NULL
, child_key_dot
);
3801 qdict_extract_subqdict(bs
->options
, NULL
, child_key_dot
);
3802 qdict_del(bs
->explicit_options
, child
->name
);
3803 qdict_del(bs
->options
, child
->name
);
3804 g_free(child_key_dot
);
3807 /* Check if any unknown options were used */
3808 if (qdict_size(options
) != 0) {
3809 const QDictEntry
*entry
= qdict_first(options
);
3810 if (flags
& BDRV_O_PROTOCOL
) {
3811 error_setg(errp
, "Block protocol '%s' doesn't support the option "
3812 "'%s'", drv
->format_name
, entry
->key
);
3815 "Block format '%s' does not support the option '%s'",
3816 drv
->format_name
, entry
->key
);
3819 goto close_and_fail
;
3822 bdrv_parent_cb_change_media(bs
, true);
3824 qobject_unref(options
);
3827 /* For snapshot=on, create a temporary qcow2 overlay. bs points to the
3828 * temporary snapshot afterwards. */
3829 if (snapshot_flags
) {
3830 BlockDriverState
*snapshot_bs
;
3831 snapshot_bs
= bdrv_append_temp_snapshot(bs
, snapshot_flags
,
3832 snapshot_options
, &local_err
);
3833 snapshot_options
= NULL
;
3835 goto close_and_fail
;
3837 /* We are not going to return bs but the overlay on top of it
3838 * (snapshot_bs); thus, we have to drop the strong reference to bs
3839 * (which we obtained by calling bdrv_new()). bs will not be deleted,
3840 * though, because the overlay still has a reference to it. */
3849 qobject_unref(snapshot_options
);
3850 qobject_unref(bs
->explicit_options
);
3851 qobject_unref(bs
->options
);
3852 qobject_unref(options
);
3854 bs
->explicit_options
= NULL
;
3856 error_propagate(errp
, local_err
);
3861 qobject_unref(snapshot_options
);
3862 qobject_unref(options
);
3863 error_propagate(errp
, local_err
);
3867 BlockDriverState
*bdrv_open(const char *filename
, const char *reference
,
3868 QDict
*options
, int flags
, Error
**errp
)
3870 return bdrv_open_inherit(filename
, reference
, options
, flags
, NULL
,
3874 /* Return true if the NULL-terminated @list contains @str */
3875 static bool is_str_in_list(const char *str
, const char *const *list
)
3879 for (i
= 0; list
[i
] != NULL
; i
++) {
3880 if (!strcmp(str
, list
[i
])) {
3889 * Check that every option set in @bs->options is also set in
3892 * Options listed in the common_options list and in
3893 * @bs->drv->mutable_opts are skipped.
3895 * Return 0 on success, otherwise return -EINVAL and set @errp.
3897 static int bdrv_reset_options_allowed(BlockDriverState
*bs
,
3898 const QDict
*new_opts
, Error
**errp
)
3900 const QDictEntry
*e
;
3901 /* These options are common to all block drivers and are handled
3902 * in bdrv_reopen_prepare() so they can be left out of @new_opts */
3903 const char *const common_options
[] = {
3904 "node-name", "discard", "cache.direct", "cache.no-flush",
3905 "read-only", "auto-read-only", "detect-zeroes", NULL
3908 for (e
= qdict_first(bs
->options
); e
; e
= qdict_next(bs
->options
, e
)) {
3909 if (!qdict_haskey(new_opts
, e
->key
) &&
3910 !is_str_in_list(e
->key
, common_options
) &&
3911 !is_str_in_list(e
->key
, bs
->drv
->mutable_opts
)) {
3912 error_setg(errp
, "Option '%s' cannot be reset "
3913 "to its default value", e
->key
);
3922 * Returns true if @child can be reached recursively from @bs
3924 static bool bdrv_recurse_has_child(BlockDriverState
*bs
,
3925 BlockDriverState
*child
)
3933 QLIST_FOREACH(c
, &bs
->children
, next
) {
3934 if (bdrv_recurse_has_child(c
->bs
, child
)) {
3943 * Adds a BlockDriverState to a simple queue for an atomic, transactional
3944 * reopen of multiple devices.
3946 * bs_queue can either be an existing BlockReopenQueue that has had QTAILQ_INIT
3947 * already performed, or alternatively may be NULL a new BlockReopenQueue will
3948 * be created and initialized. This newly created BlockReopenQueue should be
3949 * passed back in for subsequent calls that are intended to be of the same
3952 * bs is the BlockDriverState to add to the reopen queue.
3954 * options contains the changed options for the associated bs
3955 * (the BlockReopenQueue takes ownership)
3957 * flags contains the open flags for the associated bs
3959 * returns a pointer to bs_queue, which is either the newly allocated
3960 * bs_queue, or the existing bs_queue being used.
3962 * bs must be drained between bdrv_reopen_queue() and bdrv_reopen_multiple().
3964 static BlockReopenQueue
*bdrv_reopen_queue_child(BlockReopenQueue
*bs_queue
,
3965 BlockDriverState
*bs
,
3967 const BdrvChildClass
*klass
,
3969 bool parent_is_format
,
3970 QDict
*parent_options
,
3976 BlockReopenQueueEntry
*bs_entry
;
3978 QDict
*old_options
, *explicit_options
, *options_copy
;
3982 /* Make sure that the caller remembered to use a drained section. This is
3983 * important to avoid graph changes between the recursive queuing here and
3984 * bdrv_reopen_multiple(). */
3985 assert(bs
->quiesce_counter
> 0);
3987 if (bs_queue
== NULL
) {
3988 bs_queue
= g_new0(BlockReopenQueue
, 1);
3989 QTAILQ_INIT(bs_queue
);
3993 options
= qdict_new();
3996 /* Check if this BlockDriverState is already in the queue */
3997 QTAILQ_FOREACH(bs_entry
, bs_queue
, entry
) {
3998 if (bs
== bs_entry
->state
.bs
) {
4004 * Precedence of options:
4005 * 1. Explicitly passed in options (highest)
4006 * 2. Retained from explicitly set options of bs
4007 * 3. Inherited from parent node
4008 * 4. Retained from effective options of bs
4011 /* Old explicitly set values (don't overwrite by inherited value) */
4012 if (bs_entry
|| keep_old_opts
) {
4013 old_options
= qdict_clone_shallow(bs_entry
?
4014 bs_entry
->state
.explicit_options
:
4015 bs
->explicit_options
);
4016 bdrv_join_options(bs
, options
, old_options
);
4017 qobject_unref(old_options
);
4020 explicit_options
= qdict_clone_shallow(options
);
4022 /* Inherit from parent node */
4023 if (parent_options
) {
4025 klass
->inherit_options(role
, parent_is_format
, &flags
, options
,
4026 parent_flags
, parent_options
);
4028 flags
= bdrv_get_flags(bs
);
4031 if (keep_old_opts
) {
4032 /* Old values are used for options that aren't set yet */
4033 old_options
= qdict_clone_shallow(bs
->options
);
4034 bdrv_join_options(bs
, options
, old_options
);
4035 qobject_unref(old_options
);
4038 /* We have the final set of options so let's update the flags */
4039 options_copy
= qdict_clone_shallow(options
);
4040 opts
= qemu_opts_create(&bdrv_runtime_opts
, NULL
, 0, &error_abort
);
4041 qemu_opts_absorb_qdict(opts
, options_copy
, NULL
);
4042 update_flags_from_options(&flags
, opts
);
4043 qemu_opts_del(opts
);
4044 qobject_unref(options_copy
);
4046 /* bdrv_open_inherit() sets and clears some additional flags internally */
4047 flags
&= ~BDRV_O_PROTOCOL
;
4048 if (flags
& BDRV_O_RDWR
) {
4049 flags
|= BDRV_O_ALLOW_RDWR
;
4053 bs_entry
= g_new0(BlockReopenQueueEntry
, 1);
4054 QTAILQ_INSERT_TAIL(bs_queue
, bs_entry
, entry
);
4056 qobject_unref(bs_entry
->state
.options
);
4057 qobject_unref(bs_entry
->state
.explicit_options
);
4060 bs_entry
->state
.bs
= bs
;
4061 bs_entry
->state
.options
= options
;
4062 bs_entry
->state
.explicit_options
= explicit_options
;
4063 bs_entry
->state
.flags
= flags
;
4066 * If keep_old_opts is false then it means that unspecified
4067 * options must be reset to their original value. We don't allow
4068 * resetting 'backing' but we need to know if the option is
4069 * missing in order to decide if we have to return an error.
4071 if (!keep_old_opts
) {
4072 bs_entry
->state
.backing_missing
=
4073 !qdict_haskey(options
, "backing") &&
4074 !qdict_haskey(options
, "backing.driver");
4077 QLIST_FOREACH(child
, &bs
->children
, next
) {
4078 QDict
*new_child_options
= NULL
;
4079 bool child_keep_old
= keep_old_opts
;
4081 /* reopen can only change the options of block devices that were
4082 * implicitly created and inherited options. For other (referenced)
4083 * block devices, a syntax like "backing.foo" results in an error. */
4084 if (child
->bs
->inherits_from
!= bs
) {
4088 /* Check if the options contain a child reference */
4089 if (qdict_haskey(options
, child
->name
)) {
4090 const char *childref
= qdict_get_try_str(options
, child
->name
);
4092 * The current child must not be reopened if the child
4093 * reference is null or points to a different node.
4095 if (g_strcmp0(childref
, child
->bs
->node_name
)) {
4099 * If the child reference points to the current child then
4100 * reopen it with its existing set of options (note that
4101 * it can still inherit new options from the parent).
4103 child_keep_old
= true;
4105 /* Extract child options ("child-name.*") */
4106 char *child_key_dot
= g_strdup_printf("%s.", child
->name
);
4107 qdict_extract_subqdict(explicit_options
, NULL
, child_key_dot
);
4108 qdict_extract_subqdict(options
, &new_child_options
, child_key_dot
);
4109 g_free(child_key_dot
);
4112 bdrv_reopen_queue_child(bs_queue
, child
->bs
, new_child_options
,
4113 child
->klass
, child
->role
, bs
->drv
->is_format
,
4114 options
, flags
, child_keep_old
);
4120 BlockReopenQueue
*bdrv_reopen_queue(BlockReopenQueue
*bs_queue
,
4121 BlockDriverState
*bs
,
4122 QDict
*options
, bool keep_old_opts
)
4124 return bdrv_reopen_queue_child(bs_queue
, bs
, options
, NULL
, 0, false,
4125 NULL
, 0, keep_old_opts
);
4128 void bdrv_reopen_queue_free(BlockReopenQueue
*bs_queue
)
4131 BlockReopenQueueEntry
*bs_entry
, *next
;
4132 QTAILQ_FOREACH_SAFE(bs_entry
, bs_queue
, entry
, next
) {
4133 qobject_unref(bs_entry
->state
.explicit_options
);
4134 qobject_unref(bs_entry
->state
.options
);
4142 * Reopen multiple BlockDriverStates atomically & transactionally.
4144 * The queue passed in (bs_queue) must have been built up previous
4145 * via bdrv_reopen_queue().
4147 * Reopens all BDS specified in the queue, with the appropriate
4148 * flags. All devices are prepared for reopen, and failure of any
4149 * device will cause all device changes to be abandoned, and intermediate
4152 * If all devices prepare successfully, then the changes are committed
4155 * All affected nodes must be drained between bdrv_reopen_queue() and
4156 * bdrv_reopen_multiple().
4158 * To be called from the main thread, with all other AioContexts unlocked.
4160 int bdrv_reopen_multiple(BlockReopenQueue
*bs_queue
, Error
**errp
)
4163 BlockReopenQueueEntry
*bs_entry
, *next
;
4165 Transaction
*tran
= tran_new();
4166 g_autoptr(GHashTable
) found
= NULL
;
4167 g_autoptr(GSList
) refresh_list
= NULL
;
4169 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
4170 assert(bs_queue
!= NULL
);
4172 QTAILQ_FOREACH(bs_entry
, bs_queue
, entry
) {
4173 ctx
= bdrv_get_aio_context(bs_entry
->state
.bs
);
4174 aio_context_acquire(ctx
);
4175 ret
= bdrv_flush(bs_entry
->state
.bs
);
4176 aio_context_release(ctx
);
4178 error_setg_errno(errp
, -ret
, "Error flushing drive");
4183 QTAILQ_FOREACH(bs_entry
, bs_queue
, entry
) {
4184 assert(bs_entry
->state
.bs
->quiesce_counter
> 0);
4185 ctx
= bdrv_get_aio_context(bs_entry
->state
.bs
);
4186 aio_context_acquire(ctx
);
4187 ret
= bdrv_reopen_prepare(&bs_entry
->state
, bs_queue
, tran
, errp
);
4188 aio_context_release(ctx
);
4192 bs_entry
->prepared
= true;
4195 found
= g_hash_table_new(NULL
, NULL
);
4196 QTAILQ_FOREACH(bs_entry
, bs_queue
, entry
) {
4197 BDRVReopenState
*state
= &bs_entry
->state
;
4199 refresh_list
= bdrv_topological_dfs(refresh_list
, found
, state
->bs
);
4200 if (state
->old_backing_bs
) {
4201 refresh_list
= bdrv_topological_dfs(refresh_list
, found
,
4202 state
->old_backing_bs
);
4204 if (state
->old_file_bs
) {
4205 refresh_list
= bdrv_topological_dfs(refresh_list
, found
,
4206 state
->old_file_bs
);
4211 * Note that file-posix driver rely on permission update done during reopen
4212 * (even if no permission changed), because it wants "new" permissions for
4213 * reconfiguring the fd and that's why it does it in raw_check_perm(), not
4214 * in raw_reopen_prepare() which is called with "old" permissions.
4216 ret
= bdrv_list_refresh_perms(refresh_list
, bs_queue
, tran
, errp
);
4222 * If we reach this point, we have success and just need to apply the
4225 * Reverse order is used to comfort qcow2 driver: on commit it need to write
4226 * IN_USE flag to the image, to mark bitmaps in the image as invalid. But
4227 * children are usually goes after parents in reopen-queue, so go from last
4230 QTAILQ_FOREACH_REVERSE(bs_entry
, bs_queue
, entry
) {
4231 ctx
= bdrv_get_aio_context(bs_entry
->state
.bs
);
4232 aio_context_acquire(ctx
);
4233 bdrv_reopen_commit(&bs_entry
->state
);
4234 aio_context_release(ctx
);
4239 QTAILQ_FOREACH_REVERSE(bs_entry
, bs_queue
, entry
) {
4240 BlockDriverState
*bs
= bs_entry
->state
.bs
;
4242 if (bs
->drv
->bdrv_reopen_commit_post
) {
4243 ctx
= bdrv_get_aio_context(bs
);
4244 aio_context_acquire(ctx
);
4245 bs
->drv
->bdrv_reopen_commit_post(&bs_entry
->state
);
4246 aio_context_release(ctx
);
4255 QTAILQ_FOREACH_SAFE(bs_entry
, bs_queue
, entry
, next
) {
4256 if (bs_entry
->prepared
) {
4257 ctx
= bdrv_get_aio_context(bs_entry
->state
.bs
);
4258 aio_context_acquire(ctx
);
4259 bdrv_reopen_abort(&bs_entry
->state
);
4260 aio_context_release(ctx
);
4265 bdrv_reopen_queue_free(bs_queue
);
4270 int bdrv_reopen(BlockDriverState
*bs
, QDict
*opts
, bool keep_old_opts
,
4273 AioContext
*ctx
= bdrv_get_aio_context(bs
);
4274 BlockReopenQueue
*queue
;
4277 bdrv_subtree_drained_begin(bs
);
4278 if (ctx
!= qemu_get_aio_context()) {
4279 aio_context_release(ctx
);
4282 queue
= bdrv_reopen_queue(NULL
, bs
, opts
, keep_old_opts
);
4283 ret
= bdrv_reopen_multiple(queue
, errp
);
4285 if (ctx
!= qemu_get_aio_context()) {
4286 aio_context_acquire(ctx
);
4288 bdrv_subtree_drained_end(bs
);
4293 int bdrv_reopen_set_read_only(BlockDriverState
*bs
, bool read_only
,
4296 QDict
*opts
= qdict_new();
4298 qdict_put_bool(opts
, BDRV_OPT_READ_ONLY
, read_only
);
4300 return bdrv_reopen(bs
, opts
, true, errp
);
4304 * Take a BDRVReopenState and check if the value of 'backing' in the
4305 * reopen_state->options QDict is valid or not.
4307 * If 'backing' is missing from the QDict then return 0.
4309 * If 'backing' contains the node name of the backing file of
4310 * reopen_state->bs then return 0.
4312 * If 'backing' contains a different node name (or is null) then check
4313 * whether the current backing file can be replaced with the new one.
4314 * If that's the case then reopen_state->replace_backing_bs is set to
4315 * true and reopen_state->new_backing_bs contains a pointer to the new
4316 * backing BlockDriverState (or NULL).
4318 * Return 0 on success, otherwise return < 0 and set @errp.
4320 static int bdrv_reopen_parse_file_or_backing(BDRVReopenState
*reopen_state
,
4321 bool is_backing
, Transaction
*tran
,
4324 BlockDriverState
*bs
= reopen_state
->bs
;
4325 BlockDriverState
*new_child_bs
;
4326 BlockDriverState
*old_child_bs
= is_backing
? child_bs(bs
->backing
) :
4328 const char *child_name
= is_backing
? "backing" : "file";
4332 value
= qdict_get(reopen_state
->options
, child_name
);
4333 if (value
== NULL
) {
4337 switch (qobject_type(value
)) {
4339 assert(is_backing
); /* The 'file' option does not allow a null value */
4340 new_child_bs
= NULL
;
4343 str
= qstring_get_str(qobject_to(QString
, value
));
4344 new_child_bs
= bdrv_lookup_bs(NULL
, str
, errp
);
4345 if (new_child_bs
== NULL
) {
4347 } else if (bdrv_recurse_has_child(new_child_bs
, bs
)) {
4348 error_setg(errp
, "Making '%s' a %s child of '%s' would create a "
4349 "cycle", str
, child_name
, bs
->node_name
);
4355 * The options QDict has been flattened, so 'backing' and 'file'
4356 * do not allow any other data type here.
4358 g_assert_not_reached();
4361 if (old_child_bs
== new_child_bs
) {
4366 if (bdrv_skip_implicit_filters(old_child_bs
) == new_child_bs
) {
4370 if (old_child_bs
->implicit
) {
4371 error_setg(errp
, "Cannot replace implicit %s child of %s",
4372 child_name
, bs
->node_name
);
4377 if (bs
->drv
->is_filter
&& !old_child_bs
) {
4379 * Filters always have a file or a backing child, so we are trying to
4380 * change wrong child
4382 error_setg(errp
, "'%s' is a %s filter node that does not support a "
4383 "%s child", bs
->node_name
, bs
->drv
->format_name
, child_name
);
4388 reopen_state
->old_backing_bs
= old_child_bs
;
4390 reopen_state
->old_file_bs
= old_child_bs
;
4393 return bdrv_set_file_or_backing_noperm(bs
, new_child_bs
, is_backing
,
4398 * Prepares a BlockDriverState for reopen. All changes are staged in the
4399 * 'opaque' field of the BDRVReopenState, which is used and allocated by
4400 * the block driver layer .bdrv_reopen_prepare()
4402 * bs is the BlockDriverState to reopen
4403 * flags are the new open flags
4404 * queue is the reopen queue
4406 * Returns 0 on success, non-zero on error. On error errp will be set
4409 * On failure, bdrv_reopen_abort() will be called to clean up any data.
4410 * It is the responsibility of the caller to then call the abort() or
4411 * commit() for any other BDS that have been left in a prepare() state
4414 static int bdrv_reopen_prepare(BDRVReopenState
*reopen_state
,
4415 BlockReopenQueue
*queue
,
4416 Transaction
*change_child_tran
, Error
**errp
)
4420 Error
*local_err
= NULL
;
4423 QDict
*orig_reopen_opts
;
4424 char *discard
= NULL
;
4426 bool drv_prepared
= false;
4428 assert(reopen_state
!= NULL
);
4429 assert(reopen_state
->bs
->drv
!= NULL
);
4430 drv
= reopen_state
->bs
->drv
;
4432 /* This function and each driver's bdrv_reopen_prepare() remove
4433 * entries from reopen_state->options as they are processed, so
4434 * we need to make a copy of the original QDict. */
4435 orig_reopen_opts
= qdict_clone_shallow(reopen_state
->options
);
4437 /* Process generic block layer options */
4438 opts
= qemu_opts_create(&bdrv_runtime_opts
, NULL
, 0, &error_abort
);
4439 if (!qemu_opts_absorb_qdict(opts
, reopen_state
->options
, errp
)) {
4444 /* This was already called in bdrv_reopen_queue_child() so the flags
4445 * are up-to-date. This time we simply want to remove the options from
4446 * QemuOpts in order to indicate that they have been processed. */
4447 old_flags
= reopen_state
->flags
;
4448 update_flags_from_options(&reopen_state
->flags
, opts
);
4449 assert(old_flags
== reopen_state
->flags
);
4451 discard
= qemu_opt_get_del(opts
, BDRV_OPT_DISCARD
);
4452 if (discard
!= NULL
) {
4453 if (bdrv_parse_discard_flags(discard
, &reopen_state
->flags
) != 0) {
4454 error_setg(errp
, "Invalid discard option");
4460 reopen_state
->detect_zeroes
=
4461 bdrv_parse_detect_zeroes(opts
, reopen_state
->flags
, &local_err
);
4463 error_propagate(errp
, local_err
);
4468 /* All other options (including node-name and driver) must be unchanged.
4469 * Put them back into the QDict, so that they are checked at the end
4470 * of this function. */
4471 qemu_opts_to_qdict(opts
, reopen_state
->options
);
4473 /* If we are to stay read-only, do not allow permission change
4474 * to r/w. Attempting to set to r/w may fail if either BDRV_O_ALLOW_RDWR is
4475 * not set, or if the BDS still has copy_on_read enabled */
4476 read_only
= !(reopen_state
->flags
& BDRV_O_RDWR
);
4477 ret
= bdrv_can_set_read_only(reopen_state
->bs
, read_only
, true, &local_err
);
4479 error_propagate(errp
, local_err
);
4483 if (drv
->bdrv_reopen_prepare
) {
4485 * If a driver-specific option is missing, it means that we
4486 * should reset it to its default value.
4487 * But not all options allow that, so we need to check it first.
4489 ret
= bdrv_reset_options_allowed(reopen_state
->bs
,
4490 reopen_state
->options
, errp
);
4495 ret
= drv
->bdrv_reopen_prepare(reopen_state
, queue
, &local_err
);
4497 if (local_err
!= NULL
) {
4498 error_propagate(errp
, local_err
);
4500 bdrv_refresh_filename(reopen_state
->bs
);
4501 error_setg(errp
, "failed while preparing to reopen image '%s'",
4502 reopen_state
->bs
->filename
);
4507 /* It is currently mandatory to have a bdrv_reopen_prepare()
4508 * handler for each supported drv. */
4509 error_setg(errp
, "Block format '%s' used by node '%s' "
4510 "does not support reopening files", drv
->format_name
,
4511 bdrv_get_device_or_node_name(reopen_state
->bs
));
4516 drv_prepared
= true;
4519 * We must provide the 'backing' option if the BDS has a backing
4520 * file or if the image file has a backing file name as part of
4521 * its metadata. Otherwise the 'backing' option can be omitted.
4523 if (drv
->supports_backing
&& reopen_state
->backing_missing
&&
4524 (reopen_state
->bs
->backing
|| reopen_state
->bs
->backing_file
[0])) {
4525 error_setg(errp
, "backing is missing for '%s'",
4526 reopen_state
->bs
->node_name
);
4532 * Allow changing the 'backing' option. The new value can be
4533 * either a reference to an existing node (using its node name)
4534 * or NULL to simply detach the current backing file.
4536 ret
= bdrv_reopen_parse_file_or_backing(reopen_state
, true,
4537 change_child_tran
, errp
);
4541 qdict_del(reopen_state
->options
, "backing");
4543 /* Allow changing the 'file' option. In this case NULL is not allowed */
4544 ret
= bdrv_reopen_parse_file_or_backing(reopen_state
, false,
4545 change_child_tran
, errp
);
4549 qdict_del(reopen_state
->options
, "file");
4551 /* Options that are not handled are only okay if they are unchanged
4552 * compared to the old state. It is expected that some options are only
4553 * used for the initial open, but not reopen (e.g. filename) */
4554 if (qdict_size(reopen_state
->options
)) {
4555 const QDictEntry
*entry
= qdict_first(reopen_state
->options
);
4558 QObject
*new = entry
->value
;
4559 QObject
*old
= qdict_get(reopen_state
->bs
->options
, entry
->key
);
4561 /* Allow child references (child_name=node_name) as long as they
4562 * point to the current child (i.e. everything stays the same). */
4563 if (qobject_type(new) == QTYPE_QSTRING
) {
4565 QLIST_FOREACH(child
, &reopen_state
->bs
->children
, next
) {
4566 if (!strcmp(child
->name
, entry
->key
)) {
4572 if (!strcmp(child
->bs
->node_name
,
4573 qstring_get_str(qobject_to(QString
, new)))) {
4574 continue; /* Found child with this name, skip option */
4580 * TODO: When using -drive to specify blockdev options, all values
4581 * will be strings; however, when using -blockdev, blockdev-add or
4582 * filenames using the json:{} pseudo-protocol, they will be
4584 * In contrast, reopening options are (currently) always strings
4585 * (because you can only specify them through qemu-io; all other
4586 * callers do not specify any options).
4587 * Therefore, when using anything other than -drive to create a BDS,
4588 * this cannot detect non-string options as unchanged, because
4589 * qobject_is_equal() always returns false for objects of different
4590 * type. In the future, this should be remedied by correctly typing
4591 * all options. For now, this is not too big of an issue because
4592 * the user can simply omit options which cannot be changed anyway,
4593 * so they will stay unchanged.
4595 if (!qobject_is_equal(new, old
)) {
4596 error_setg(errp
, "Cannot change the option '%s'", entry
->key
);
4600 } while ((entry
= qdict_next(reopen_state
->options
, entry
)));
4605 /* Restore the original reopen_state->options QDict */
4606 qobject_unref(reopen_state
->options
);
4607 reopen_state
->options
= qobject_ref(orig_reopen_opts
);
4610 if (ret
< 0 && drv_prepared
) {
4611 /* drv->bdrv_reopen_prepare() has succeeded, so we need to
4612 * call drv->bdrv_reopen_abort() before signaling an error
4613 * (bdrv_reopen_multiple() will not call bdrv_reopen_abort()
4614 * when the respective bdrv_reopen_prepare() has failed) */
4615 if (drv
->bdrv_reopen_abort
) {
4616 drv
->bdrv_reopen_abort(reopen_state
);
4619 qemu_opts_del(opts
);
4620 qobject_unref(orig_reopen_opts
);
4626 * Takes the staged changes for the reopen from bdrv_reopen_prepare(), and
4627 * makes them final by swapping the staging BlockDriverState contents into
4628 * the active BlockDriverState contents.
4630 static void bdrv_reopen_commit(BDRVReopenState
*reopen_state
)
4633 BlockDriverState
*bs
;
4636 assert(reopen_state
!= NULL
);
4637 bs
= reopen_state
->bs
;
4639 assert(drv
!= NULL
);
4641 /* If there are any driver level actions to take */
4642 if (drv
->bdrv_reopen_commit
) {
4643 drv
->bdrv_reopen_commit(reopen_state
);
4646 /* set BDS specific flags now */
4647 qobject_unref(bs
->explicit_options
);
4648 qobject_unref(bs
->options
);
4649 qobject_ref(reopen_state
->explicit_options
);
4650 qobject_ref(reopen_state
->options
);
4652 bs
->explicit_options
= reopen_state
->explicit_options
;
4653 bs
->options
= reopen_state
->options
;
4654 bs
->open_flags
= reopen_state
->flags
;
4655 bs
->detect_zeroes
= reopen_state
->detect_zeroes
;
4657 /* Remove child references from bs->options and bs->explicit_options.
4658 * Child options were already removed in bdrv_reopen_queue_child() */
4659 QLIST_FOREACH(child
, &bs
->children
, next
) {
4660 qdict_del(bs
->explicit_options
, child
->name
);
4661 qdict_del(bs
->options
, child
->name
);
4663 /* backing is probably removed, so it's not handled by previous loop */
4664 qdict_del(bs
->explicit_options
, "backing");
4665 qdict_del(bs
->options
, "backing");
4667 bdrv_refresh_limits(bs
, NULL
, NULL
);
4671 * Abort the reopen, and delete and free the staged changes in
4674 static void bdrv_reopen_abort(BDRVReopenState
*reopen_state
)
4678 assert(reopen_state
!= NULL
);
4679 drv
= reopen_state
->bs
->drv
;
4680 assert(drv
!= NULL
);
4682 if (drv
->bdrv_reopen_abort
) {
4683 drv
->bdrv_reopen_abort(reopen_state
);
4688 static void bdrv_close(BlockDriverState
*bs
)
4690 BdrvAioNotifier
*ban
, *ban_next
;
4691 BdrvChild
*child
, *next
;
4693 assert(!bs
->refcnt
);
4695 bdrv_drained_begin(bs
); /* complete I/O */
4697 bdrv_drain(bs
); /* in case flush left pending I/O */
4700 if (bs
->drv
->bdrv_close
) {
4701 /* Must unfreeze all children, so bdrv_unref_child() works */
4702 bs
->drv
->bdrv_close(bs
);
4707 QLIST_FOREACH_SAFE(child
, &bs
->children
, next
, next
) {
4708 bdrv_unref_child(bs
, child
);
4715 qatomic_set(&bs
->copy_on_read
, 0);
4716 bs
->backing_file
[0] = '\0';
4717 bs
->backing_format
[0] = '\0';
4718 bs
->total_sectors
= 0;
4719 bs
->encrypted
= false;
4721 qobject_unref(bs
->options
);
4722 qobject_unref(bs
->explicit_options
);
4724 bs
->explicit_options
= NULL
;
4725 qobject_unref(bs
->full_open_options
);
4726 bs
->full_open_options
= NULL
;
4727 g_free(bs
->block_status_cache
);
4728 bs
->block_status_cache
= NULL
;
4730 bdrv_release_named_dirty_bitmaps(bs
);
4731 assert(QLIST_EMPTY(&bs
->dirty_bitmaps
));
4733 QLIST_FOREACH_SAFE(ban
, &bs
->aio_notifiers
, list
, ban_next
) {
4736 QLIST_INIT(&bs
->aio_notifiers
);
4737 bdrv_drained_end(bs
);
4740 * If we're still inside some bdrv_drain_all_begin()/end() sections, end
4741 * them now since this BDS won't exist anymore when bdrv_drain_all_end()
4744 if (bs
->quiesce_counter
) {
4745 bdrv_drain_all_end_quiesce(bs
);
4749 void bdrv_close_all(void)
4751 assert(job_next(NULL
) == NULL
);
4753 /* Drop references from requests still in flight, such as canceled block
4754 * jobs whose AIO context has not been polled yet */
4757 blk_remove_all_bs();
4758 blockdev_close_all_bdrv_states();
4760 assert(QTAILQ_EMPTY(&all_bdrv_states
));
4763 static bool should_update_child(BdrvChild
*c
, BlockDriverState
*to
)
4769 if (c
->klass
->stay_at_node
) {
4773 /* If the child @c belongs to the BDS @to, replacing the current
4774 * c->bs by @to would mean to create a loop.
4776 * Such a case occurs when appending a BDS to a backing chain.
4777 * For instance, imagine the following chain:
4779 * guest device -> node A -> further backing chain...
4781 * Now we create a new BDS B which we want to put on top of this
4782 * chain, so we first attach A as its backing node:
4787 * guest device -> node A -> further backing chain...
4789 * Finally we want to replace A by B. When doing that, we want to
4790 * replace all pointers to A by pointers to B -- except for the
4791 * pointer from B because (1) that would create a loop, and (2)
4792 * that pointer should simply stay intact:
4794 * guest device -> node B
4797 * node A -> further backing chain...
4799 * In general, when replacing a node A (c->bs) by a node B (@to),
4800 * if A is a child of B, that means we cannot replace A by B there
4801 * because that would create a loop. Silently detaching A from B
4802 * is also not really an option. So overall just leaving A in
4803 * place there is the most sensible choice.
4805 * We would also create a loop in any cases where @c is only
4806 * indirectly referenced by @to. Prevent this by returning false
4807 * if @c is found (by breadth-first search) anywhere in the whole
4812 found
= g_hash_table_new(NULL
, NULL
);
4813 g_hash_table_add(found
, to
);
4814 queue
= g_queue_new();
4815 g_queue_push_tail(queue
, to
);
4817 while (!g_queue_is_empty(queue
)) {
4818 BlockDriverState
*v
= g_queue_pop_head(queue
);
4821 QLIST_FOREACH(c2
, &v
->children
, next
) {
4827 if (g_hash_table_contains(found
, c2
->bs
)) {
4831 g_queue_push_tail(queue
, c2
->bs
);
4832 g_hash_table_add(found
, c2
->bs
);
4836 g_queue_free(queue
);
4837 g_hash_table_destroy(found
);
4842 typedef struct BdrvRemoveFilterOrCowChild
{
4845 } BdrvRemoveFilterOrCowChild
;
4847 static void bdrv_remove_filter_or_cow_child_abort(void *opaque
)
4849 BdrvRemoveFilterOrCowChild
*s
= opaque
;
4850 BlockDriverState
*parent_bs
= s
->child
->opaque
;
4852 if (s
->is_backing
) {
4853 parent_bs
->backing
= s
->child
;
4855 parent_bs
->file
= s
->child
;
4859 * We don't have to restore child->bs here to undo bdrv_replace_child_tran()
4860 * because that function is transactionable and it registered own completion
4861 * entries in @tran, so .abort() for bdrv_replace_child_safe() will be
4862 * called automatically.
4866 static void bdrv_remove_filter_or_cow_child_commit(void *opaque
)
4868 BdrvRemoveFilterOrCowChild
*s
= opaque
;
4870 bdrv_child_free(s
->child
);
4873 static TransactionActionDrv bdrv_remove_filter_or_cow_child_drv
= {
4874 .abort
= bdrv_remove_filter_or_cow_child_abort
,
4875 .commit
= bdrv_remove_filter_or_cow_child_commit
,
4880 * A function to remove backing or file child of @bs.
4881 * Function doesn't update permissions, caller is responsible for this.
4883 static void bdrv_remove_file_or_backing_child(BlockDriverState
*bs
,
4887 BdrvRemoveFilterOrCowChild
*s
;
4889 assert(child
== bs
->backing
|| child
== bs
->file
);
4896 bdrv_replace_child_tran(child
, NULL
, tran
);
4899 s
= g_new(BdrvRemoveFilterOrCowChild
, 1);
4900 *s
= (BdrvRemoveFilterOrCowChild
) {
4902 .is_backing
= (child
== bs
->backing
),
4904 tran_add(tran
, &bdrv_remove_filter_or_cow_child_drv
, s
);
4906 if (s
->is_backing
) {
4914 * A function to remove backing-chain child of @bs if exists: cow child for
4915 * format nodes (always .backing) and filter child for filters (may be .file or
4918 static void bdrv_remove_filter_or_cow_child(BlockDriverState
*bs
,
4921 bdrv_remove_file_or_backing_child(bs
, bdrv_filter_or_cow_child(bs
), tran
);
4924 static int bdrv_replace_node_noperm(BlockDriverState
*from
,
4925 BlockDriverState
*to
,
4926 bool auto_skip
, Transaction
*tran
,
4929 BdrvChild
*c
, *next
;
4931 QLIST_FOREACH_SAFE(c
, &from
->parents
, next_parent
, next
) {
4932 assert(c
->bs
== from
);
4933 if (!should_update_child(c
, to
)) {
4937 error_setg(errp
, "Should not change '%s' link to '%s'",
4938 c
->name
, from
->node_name
);
4942 error_setg(errp
, "Cannot change '%s' link to '%s'",
4943 c
->name
, from
->node_name
);
4946 bdrv_replace_child_tran(c
, to
, tran
);
4953 * With auto_skip=true bdrv_replace_node_common skips updating from parents
4954 * if it creates a parent-child relation loop or if parent is block-job.
4956 * With auto_skip=false the error is returned if from has a parent which should
4959 * With @detach_subchain=true @to must be in a backing chain of @from. In this
4960 * case backing link of the cow-parent of @to is removed.
4962 static int bdrv_replace_node_common(BlockDriverState
*from
,
4963 BlockDriverState
*to
,
4964 bool auto_skip
, bool detach_subchain
,
4967 Transaction
*tran
= tran_new();
4968 g_autoptr(GHashTable
) found
= NULL
;
4969 g_autoptr(GSList
) refresh_list
= NULL
;
4970 BlockDriverState
*to_cow_parent
= NULL
;
4973 if (detach_subchain
) {
4974 assert(bdrv_chain_contains(from
, to
));
4976 for (to_cow_parent
= from
;
4977 bdrv_filter_or_cow_bs(to_cow_parent
) != to
;
4978 to_cow_parent
= bdrv_filter_or_cow_bs(to_cow_parent
))
4984 /* Make sure that @from doesn't go away until we have successfully attached
4985 * all of its parents to @to. */
4988 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
4989 assert(bdrv_get_aio_context(from
) == bdrv_get_aio_context(to
));
4990 bdrv_drained_begin(from
);
4993 * Do the replacement without permission update.
4994 * Replacement may influence the permissions, we should calculate new
4995 * permissions based on new graph. If we fail, we'll roll-back the
4998 ret
= bdrv_replace_node_noperm(from
, to
, auto_skip
, tran
, errp
);
5003 if (detach_subchain
) {
5004 bdrv_remove_filter_or_cow_child(to_cow_parent
, tran
);
5007 found
= g_hash_table_new(NULL
, NULL
);
5009 refresh_list
= bdrv_topological_dfs(refresh_list
, found
, to
);
5010 refresh_list
= bdrv_topological_dfs(refresh_list
, found
, from
);
5012 ret
= bdrv_list_refresh_perms(refresh_list
, NULL
, tran
, errp
);
5020 tran_finalize(tran
, ret
);
5022 bdrv_drained_end(from
);
5028 int bdrv_replace_node(BlockDriverState
*from
, BlockDriverState
*to
,
5031 return bdrv_replace_node_common(from
, to
, true, false, errp
);
5034 int bdrv_drop_filter(BlockDriverState
*bs
, Error
**errp
)
5036 return bdrv_replace_node_common(bs
, bdrv_filter_or_cow_bs(bs
), true, true,
5041 * Add new bs contents at the top of an image chain while the chain is
5042 * live, while keeping required fields on the top layer.
5044 * This will modify the BlockDriverState fields, and swap contents
5045 * between bs_new and bs_top. Both bs_new and bs_top are modified.
5047 * bs_new must not be attached to a BlockBackend and must not have backing
5050 * This function does not create any image files.
5052 int bdrv_append(BlockDriverState
*bs_new
, BlockDriverState
*bs_top
,
5056 Transaction
*tran
= tran_new();
5058 assert(!bs_new
->backing
);
5060 ret
= bdrv_attach_child_noperm(bs_new
, bs_top
, "backing",
5061 &child_of_bds
, bdrv_backing_role(bs_new
),
5062 &bs_new
->backing
, tran
, errp
);
5067 ret
= bdrv_replace_node_noperm(bs_top
, bs_new
, true, tran
, errp
);
5072 ret
= bdrv_refresh_perms(bs_new
, errp
);
5074 tran_finalize(tran
, ret
);
5076 bdrv_refresh_limits(bs_top
, NULL
, NULL
);
5081 /* Not for empty child */
5082 int bdrv_replace_child_bs(BdrvChild
*child
, BlockDriverState
*new_bs
,
5086 Transaction
*tran
= tran_new();
5087 g_autoptr(GHashTable
) found
= NULL
;
5088 g_autoptr(GSList
) refresh_list
= NULL
;
5089 BlockDriverState
*old_bs
= child
->bs
;
5092 bdrv_drained_begin(old_bs
);
5093 bdrv_drained_begin(new_bs
);
5095 bdrv_replace_child_tran(child
, new_bs
, tran
);
5097 found
= g_hash_table_new(NULL
, NULL
);
5098 refresh_list
= bdrv_topological_dfs(refresh_list
, found
, old_bs
);
5099 refresh_list
= bdrv_topological_dfs(refresh_list
, found
, new_bs
);
5101 ret
= bdrv_list_refresh_perms(refresh_list
, NULL
, tran
, errp
);
5103 tran_finalize(tran
, ret
);
5105 bdrv_drained_end(old_bs
);
5106 bdrv_drained_end(new_bs
);
5112 static void bdrv_delete(BlockDriverState
*bs
)
5114 assert(bdrv_op_blocker_is_empty(bs
));
5115 assert(!bs
->refcnt
);
5117 /* remove from list, if necessary */
5118 if (bs
->node_name
[0] != '\0') {
5119 QTAILQ_REMOVE(&graph_bdrv_states
, bs
, node_list
);
5121 QTAILQ_REMOVE(&all_bdrv_states
, bs
, bs_list
);
5130 * Replace @bs by newly created block node.
5132 * @options is a QDict of options to pass to the block drivers, or NULL for an
5133 * empty set of options. The reference to the QDict belongs to the block layer
5134 * after the call (even on failure), so if the caller intends to reuse the
5135 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
5137 BlockDriverState
*bdrv_insert_node(BlockDriverState
*bs
, QDict
*options
,
5138 int flags
, Error
**errp
)
5142 BlockDriverState
*new_node_bs
= NULL
;
5143 const char *drvname
, *node_name
;
5146 drvname
= qdict_get_try_str(options
, "driver");
5148 error_setg(errp
, "driver is not specified");
5152 drv
= bdrv_find_format(drvname
);
5154 error_setg(errp
, "Unknown driver: '%s'", drvname
);
5158 node_name
= qdict_get_try_str(options
, "node-name");
5160 new_node_bs
= bdrv_new_open_driver_opts(drv
, node_name
, options
, flags
,
5162 options
= NULL
; /* bdrv_new_open_driver() eats options */
5164 error_prepend(errp
, "Could not create node: ");
5168 bdrv_drained_begin(bs
);
5169 ret
= bdrv_replace_node(bs
, new_node_bs
, errp
);
5170 bdrv_drained_end(bs
);
5173 error_prepend(errp
, "Could not replace node: ");
5180 qobject_unref(options
);
5181 bdrv_unref(new_node_bs
);
5186 * Run consistency checks on an image
5188 * Returns 0 if the check could be completed (it doesn't mean that the image is
5189 * free of errors) or -errno when an internal error occurred. The results of the
5190 * check are stored in res.
5192 int coroutine_fn
bdrv_co_check(BlockDriverState
*bs
,
5193 BdrvCheckResult
*res
, BdrvCheckMode fix
)
5195 if (bs
->drv
== NULL
) {
5198 if (bs
->drv
->bdrv_co_check
== NULL
) {
5202 memset(res
, 0, sizeof(*res
));
5203 return bs
->drv
->bdrv_co_check(bs
, res
, fix
);
5209 * -EINVAL - backing format specified, but no file
5210 * -ENOSPC - can't update the backing file because no space is left in the
5212 * -ENOTSUP - format driver doesn't support changing the backing file
5214 int bdrv_change_backing_file(BlockDriverState
*bs
, const char *backing_file
,
5215 const char *backing_fmt
, bool require
)
5217 BlockDriver
*drv
= bs
->drv
;
5224 /* Backing file format doesn't make sense without a backing file */
5225 if (backing_fmt
&& !backing_file
) {
5229 if (require
&& backing_file
&& !backing_fmt
) {
5233 if (drv
->bdrv_change_backing_file
!= NULL
) {
5234 ret
= drv
->bdrv_change_backing_file(bs
, backing_file
, backing_fmt
);
5240 pstrcpy(bs
->backing_file
, sizeof(bs
->backing_file
), backing_file
?: "");
5241 pstrcpy(bs
->backing_format
, sizeof(bs
->backing_format
), backing_fmt
?: "");
5242 pstrcpy(bs
->auto_backing_file
, sizeof(bs
->auto_backing_file
),
5243 backing_file
?: "");
5249 * Finds the first non-filter node above bs in the chain between
5250 * active and bs. The returned node is either an immediate parent of
5251 * bs, or there are only filter nodes between the two.
5253 * Returns NULL if bs is not found in active's image chain,
5254 * or if active == bs.
5256 * Returns the bottommost base image if bs == NULL.
5258 BlockDriverState
*bdrv_find_overlay(BlockDriverState
*active
,
5259 BlockDriverState
*bs
)
5261 bs
= bdrv_skip_filters(bs
);
5262 active
= bdrv_skip_filters(active
);
5265 BlockDriverState
*next
= bdrv_backing_chain_next(active
);
5275 /* Given a BDS, searches for the base layer. */
5276 BlockDriverState
*bdrv_find_base(BlockDriverState
*bs
)
5278 return bdrv_find_overlay(bs
, NULL
);
5282 * Return true if at least one of the COW (backing) and filter links
5283 * between @bs and @base is frozen. @errp is set if that's the case.
5284 * @base must be reachable from @bs, or NULL.
5286 bool bdrv_is_backing_chain_frozen(BlockDriverState
*bs
, BlockDriverState
*base
,
5289 BlockDriverState
*i
;
5292 for (i
= bs
; i
!= base
; i
= child_bs(child
)) {
5293 child
= bdrv_filter_or_cow_child(i
);
5295 if (child
&& child
->frozen
) {
5296 error_setg(errp
, "Cannot change '%s' link from '%s' to '%s'",
5297 child
->name
, i
->node_name
, child
->bs
->node_name
);
5306 * Freeze all COW (backing) and filter links between @bs and @base.
5307 * If any of the links is already frozen the operation is aborted and
5308 * none of the links are modified.
5309 * @base must be reachable from @bs, or NULL.
5310 * Returns 0 on success. On failure returns < 0 and sets @errp.
5312 int bdrv_freeze_backing_chain(BlockDriverState
*bs
, BlockDriverState
*base
,
5315 BlockDriverState
*i
;
5318 if (bdrv_is_backing_chain_frozen(bs
, base
, errp
)) {
5322 for (i
= bs
; i
!= base
; i
= child_bs(child
)) {
5323 child
= bdrv_filter_or_cow_child(i
);
5324 if (child
&& child
->bs
->never_freeze
) {
5325 error_setg(errp
, "Cannot freeze '%s' link to '%s'",
5326 child
->name
, child
->bs
->node_name
);
5331 for (i
= bs
; i
!= base
; i
= child_bs(child
)) {
5332 child
= bdrv_filter_or_cow_child(i
);
5334 child
->frozen
= true;
5342 * Unfreeze all COW (backing) and filter links between @bs and @base.
5343 * The caller must ensure that all links are frozen before using this
5345 * @base must be reachable from @bs, or NULL.
5347 void bdrv_unfreeze_backing_chain(BlockDriverState
*bs
, BlockDriverState
*base
)
5349 BlockDriverState
*i
;
5352 for (i
= bs
; i
!= base
; i
= child_bs(child
)) {
5353 child
= bdrv_filter_or_cow_child(i
);
5355 assert(child
->frozen
);
5356 child
->frozen
= false;
5362 * Drops images above 'base' up to and including 'top', and sets the image
5363 * above 'top' to have base as its backing file.
5365 * Requires that the overlay to 'top' is opened r/w, so that the backing file
5366 * information in 'bs' can be properly updated.
5368 * E.g., this will convert the following chain:
5369 * bottom <- base <- intermediate <- top <- active
5373 * bottom <- base <- active
5375 * It is allowed for bottom==base, in which case it converts:
5377 * base <- intermediate <- top <- active
5383 * If backing_file_str is non-NULL, it will be used when modifying top's
5384 * overlay image metadata.
5387 * if active == top, that is considered an error
5390 int bdrv_drop_intermediate(BlockDriverState
*top
, BlockDriverState
*base
,
5391 const char *backing_file_str
)
5393 BlockDriverState
*explicit_top
= top
;
5394 bool update_inherits_from
;
5396 Error
*local_err
= NULL
;
5398 g_autoptr(GSList
) updated_children
= NULL
;
5402 bdrv_subtree_drained_begin(top
);
5404 if (!top
->drv
|| !base
->drv
) {
5408 /* Make sure that base is in the backing chain of top */
5409 if (!bdrv_chain_contains(top
, base
)) {
5413 /* If 'base' recursively inherits from 'top' then we should set
5414 * base->inherits_from to top->inherits_from after 'top' and all
5415 * other intermediate nodes have been dropped.
5416 * If 'top' is an implicit node (e.g. "commit_top") we should skip
5417 * it because no one inherits from it. We use explicit_top for that. */
5418 explicit_top
= bdrv_skip_implicit_filters(explicit_top
);
5419 update_inherits_from
= bdrv_inherits_from_recursive(base
, explicit_top
);
5421 /* success - we can delete the intermediate states, and link top->base */
5422 /* TODO Check graph modification op blockers (BLK_PERM_GRAPH_MOD) once
5423 * we've figured out how they should work. */
5424 if (!backing_file_str
) {
5425 bdrv_refresh_filename(base
);
5426 backing_file_str
= base
->filename
;
5429 QLIST_FOREACH(c
, &top
->parents
, next_parent
) {
5430 updated_children
= g_slist_prepend(updated_children
, c
);
5434 * It seems correct to pass detach_subchain=true here, but it triggers
5435 * one more yet not fixed bug, when due to nested aio_poll loop we switch to
5436 * another drained section, which modify the graph (for example, removing
5437 * the child, which we keep in updated_children list). So, it's a TODO.
5439 * Note, bug triggered if pass detach_subchain=true here and run
5440 * test-bdrv-drain. test_drop_intermediate_poll() test-case will crash.
5443 bdrv_replace_node_common(top
, base
, false, false, &local_err
);
5445 error_report_err(local_err
);
5449 for (p
= updated_children
; p
; p
= p
->next
) {
5452 if (c
->klass
->update_filename
) {
5453 ret
= c
->klass
->update_filename(c
, base
, backing_file_str
,
5457 * TODO: Actually, we want to rollback all previous iterations
5458 * of this loop, and (which is almost impossible) previous
5459 * bdrv_replace_node()...
5461 * Note, that c->klass->update_filename may lead to permission
5462 * update, so it's a bad idea to call it inside permission
5463 * update transaction of bdrv_replace_node.
5465 error_report_err(local_err
);
5471 if (update_inherits_from
) {
5472 base
->inherits_from
= explicit_top
->inherits_from
;
5477 bdrv_subtree_drained_end(top
);
5483 * Implementation of BlockDriver.bdrv_get_allocated_file_size() that
5484 * sums the size of all data-bearing children. (This excludes backing
5487 static int64_t bdrv_sum_allocated_file_size(BlockDriverState
*bs
)
5490 int64_t child_size
, sum
= 0;
5492 QLIST_FOREACH(child
, &bs
->children
, next
) {
5493 if (child
->role
& (BDRV_CHILD_DATA
| BDRV_CHILD_METADATA
|
5494 BDRV_CHILD_FILTERED
))
5496 child_size
= bdrv_get_allocated_file_size(child
->bs
);
5497 if (child_size
< 0) {
5508 * Length of a allocated file in bytes. Sparse files are counted by actual
5509 * allocated space. Return < 0 if error or unknown.
5511 int64_t bdrv_get_allocated_file_size(BlockDriverState
*bs
)
5513 BlockDriver
*drv
= bs
->drv
;
5517 if (drv
->bdrv_get_allocated_file_size
) {
5518 return drv
->bdrv_get_allocated_file_size(bs
);
5521 if (drv
->bdrv_file_open
) {
5523 * Protocol drivers default to -ENOTSUP (most of their data is
5524 * not stored in any of their children (if they even have any),
5525 * so there is no generic way to figure it out).
5528 } else if (drv
->is_filter
) {
5529 /* Filter drivers default to the size of their filtered child */
5530 return bdrv_get_allocated_file_size(bdrv_filter_bs(bs
));
5532 /* Other drivers default to summing their children's sizes */
5533 return bdrv_sum_allocated_file_size(bs
);
5539 * @drv: Format driver
5540 * @opts: Creation options for new image
5541 * @in_bs: Existing image containing data for new image (may be NULL)
5542 * @errp: Error object
5543 * Returns: A #BlockMeasureInfo (free using qapi_free_BlockMeasureInfo())
5546 * Calculate file size required to create a new image.
5548 * If @in_bs is given then space for allocated clusters and zero clusters
5549 * from that image are included in the calculation. If @opts contains a
5550 * backing file that is shared by @in_bs then backing clusters may be omitted
5551 * from the calculation.
5553 * If @in_bs is NULL then the calculation includes no allocated clusters
5554 * unless a preallocation option is given in @opts.
5556 * Note that @in_bs may use a different BlockDriver from @drv.
5558 * If an error occurs the @errp pointer is set.
5560 BlockMeasureInfo
*bdrv_measure(BlockDriver
*drv
, QemuOpts
*opts
,
5561 BlockDriverState
*in_bs
, Error
**errp
)
5563 if (!drv
->bdrv_measure
) {
5564 error_setg(errp
, "Block driver '%s' does not support size measurement",
5569 return drv
->bdrv_measure(opts
, in_bs
, errp
);
5573 * Return number of sectors on success, -errno on error.
5575 int64_t bdrv_nb_sectors(BlockDriverState
*bs
)
5577 BlockDriver
*drv
= bs
->drv
;
5582 if (drv
->has_variable_length
) {
5583 int ret
= refresh_total_sectors(bs
, bs
->total_sectors
);
5588 return bs
->total_sectors
;
5592 * Return length in bytes on success, -errno on error.
5593 * The length is always a multiple of BDRV_SECTOR_SIZE.
5595 int64_t bdrv_getlength(BlockDriverState
*bs
)
5597 int64_t ret
= bdrv_nb_sectors(bs
);
5602 if (ret
> INT64_MAX
/ BDRV_SECTOR_SIZE
) {
5605 return ret
* BDRV_SECTOR_SIZE
;
5608 /* return 0 as number of sectors if no device present or error */
5609 void bdrv_get_geometry(BlockDriverState
*bs
, uint64_t *nb_sectors_ptr
)
5611 int64_t nb_sectors
= bdrv_nb_sectors(bs
);
5613 *nb_sectors_ptr
= nb_sectors
< 0 ? 0 : nb_sectors
;
5616 bool bdrv_is_sg(BlockDriverState
*bs
)
5622 * Return whether the given node supports compressed writes.
5624 bool bdrv_supports_compressed_writes(BlockDriverState
*bs
)
5626 BlockDriverState
*filtered
;
5628 if (!bs
->drv
|| !block_driver_can_compress(bs
->drv
)) {
5632 filtered
= bdrv_filter_bs(bs
);
5635 * Filters can only forward compressed writes, so we have to
5638 return bdrv_supports_compressed_writes(filtered
);
5644 const char *bdrv_get_format_name(BlockDriverState
*bs
)
5646 return bs
->drv
? bs
->drv
->format_name
: NULL
;
5649 static int qsort_strcmp(const void *a
, const void *b
)
5651 return strcmp(*(char *const *)a
, *(char *const *)b
);
5654 void bdrv_iterate_format(void (*it
)(void *opaque
, const char *name
),
5655 void *opaque
, bool read_only
)
5660 const char **formats
= NULL
;
5662 QLIST_FOREACH(drv
, &bdrv_drivers
, list
) {
5663 if (drv
->format_name
) {
5667 if (use_bdrv_whitelist
&& !bdrv_is_whitelisted(drv
, read_only
)) {
5671 while (formats
&& i
&& !found
) {
5672 found
= !strcmp(formats
[--i
], drv
->format_name
);
5676 formats
= g_renew(const char *, formats
, count
+ 1);
5677 formats
[count
++] = drv
->format_name
;
5682 for (i
= 0; i
< (int)ARRAY_SIZE(block_driver_modules
); i
++) {
5683 const char *format_name
= block_driver_modules
[i
].format_name
;
5689 if (use_bdrv_whitelist
&&
5690 !bdrv_format_is_whitelisted(format_name
, read_only
)) {
5694 while (formats
&& j
&& !found
) {
5695 found
= !strcmp(formats
[--j
], format_name
);
5699 formats
= g_renew(const char *, formats
, count
+ 1);
5700 formats
[count
++] = format_name
;
5705 qsort(formats
, count
, sizeof(formats
[0]), qsort_strcmp
);
5707 for (i
= 0; i
< count
; i
++) {
5708 it(opaque
, formats
[i
]);
5714 /* This function is to find a node in the bs graph */
5715 BlockDriverState
*bdrv_find_node(const char *node_name
)
5717 BlockDriverState
*bs
;
5721 QTAILQ_FOREACH(bs
, &graph_bdrv_states
, node_list
) {
5722 if (!strcmp(node_name
, bs
->node_name
)) {
5729 /* Put this QMP function here so it can access the static graph_bdrv_states. */
5730 BlockDeviceInfoList
*bdrv_named_nodes_list(bool flat
,
5733 BlockDeviceInfoList
*list
;
5734 BlockDriverState
*bs
;
5737 QTAILQ_FOREACH(bs
, &graph_bdrv_states
, node_list
) {
5738 BlockDeviceInfo
*info
= bdrv_block_device_info(NULL
, bs
, flat
, errp
);
5740 qapi_free_BlockDeviceInfoList(list
);
5743 QAPI_LIST_PREPEND(list
, info
);
5749 typedef struct XDbgBlockGraphConstructor
{
5750 XDbgBlockGraph
*graph
;
5751 GHashTable
*graph_nodes
;
5752 } XDbgBlockGraphConstructor
;
5754 static XDbgBlockGraphConstructor
*xdbg_graph_new(void)
5756 XDbgBlockGraphConstructor
*gr
= g_new(XDbgBlockGraphConstructor
, 1);
5758 gr
->graph
= g_new0(XDbgBlockGraph
, 1);
5759 gr
->graph_nodes
= g_hash_table_new(NULL
, NULL
);
5764 static XDbgBlockGraph
*xdbg_graph_finalize(XDbgBlockGraphConstructor
*gr
)
5766 XDbgBlockGraph
*graph
= gr
->graph
;
5768 g_hash_table_destroy(gr
->graph_nodes
);
5774 static uintptr_t xdbg_graph_node_num(XDbgBlockGraphConstructor
*gr
, void *node
)
5776 uintptr_t ret
= (uintptr_t)g_hash_table_lookup(gr
->graph_nodes
, node
);
5783 * Start counting from 1, not 0, because 0 interferes with not-found (NULL)
5784 * answer of g_hash_table_lookup.
5786 ret
= g_hash_table_size(gr
->graph_nodes
) + 1;
5787 g_hash_table_insert(gr
->graph_nodes
, node
, (void *)ret
);
5792 static void xdbg_graph_add_node(XDbgBlockGraphConstructor
*gr
, void *node
,
5793 XDbgBlockGraphNodeType type
, const char *name
)
5795 XDbgBlockGraphNode
*n
;
5797 n
= g_new0(XDbgBlockGraphNode
, 1);
5799 n
->id
= xdbg_graph_node_num(gr
, node
);
5801 n
->name
= g_strdup(name
);
5803 QAPI_LIST_PREPEND(gr
->graph
->nodes
, n
);
5806 static void xdbg_graph_add_edge(XDbgBlockGraphConstructor
*gr
, void *parent
,
5807 const BdrvChild
*child
)
5809 BlockPermission qapi_perm
;
5810 XDbgBlockGraphEdge
*edge
;
5812 edge
= g_new0(XDbgBlockGraphEdge
, 1);
5814 edge
->parent
= xdbg_graph_node_num(gr
, parent
);
5815 edge
->child
= xdbg_graph_node_num(gr
, child
->bs
);
5816 edge
->name
= g_strdup(child
->name
);
5818 for (qapi_perm
= 0; qapi_perm
< BLOCK_PERMISSION__MAX
; qapi_perm
++) {
5819 uint64_t flag
= bdrv_qapi_perm_to_blk_perm(qapi_perm
);
5821 if (flag
& child
->perm
) {
5822 QAPI_LIST_PREPEND(edge
->perm
, qapi_perm
);
5824 if (flag
& child
->shared_perm
) {
5825 QAPI_LIST_PREPEND(edge
->shared_perm
, qapi_perm
);
5829 QAPI_LIST_PREPEND(gr
->graph
->edges
, edge
);
5833 XDbgBlockGraph
*bdrv_get_xdbg_block_graph(Error
**errp
)
5837 BlockDriverState
*bs
;
5839 XDbgBlockGraphConstructor
*gr
= xdbg_graph_new();
5841 for (blk
= blk_all_next(NULL
); blk
; blk
= blk_all_next(blk
)) {
5842 char *allocated_name
= NULL
;
5843 const char *name
= blk_name(blk
);
5846 name
= allocated_name
= blk_get_attached_dev_id(blk
);
5848 xdbg_graph_add_node(gr
, blk
, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_BACKEND
,
5850 g_free(allocated_name
);
5851 if (blk_root(blk
)) {
5852 xdbg_graph_add_edge(gr
, blk
, blk_root(blk
));
5856 for (job
= block_job_next(NULL
); job
; job
= block_job_next(job
)) {
5859 xdbg_graph_add_node(gr
, job
, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_JOB
,
5861 for (el
= job
->nodes
; el
; el
= el
->next
) {
5862 xdbg_graph_add_edge(gr
, job
, (BdrvChild
*)el
->data
);
5866 QTAILQ_FOREACH(bs
, &graph_bdrv_states
, node_list
) {
5867 xdbg_graph_add_node(gr
, bs
, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_DRIVER
,
5869 QLIST_FOREACH(child
, &bs
->children
, next
) {
5870 xdbg_graph_add_edge(gr
, bs
, child
);
5874 return xdbg_graph_finalize(gr
);
5877 BlockDriverState
*bdrv_lookup_bs(const char *device
,
5878 const char *node_name
,
5882 BlockDriverState
*bs
;
5885 blk
= blk_by_name(device
);
5890 error_setg(errp
, "Device '%s' has no medium", device
);
5898 bs
= bdrv_find_node(node_name
);
5905 error_setg(errp
, "Cannot find device=\'%s\' nor node-name=\'%s\'",
5906 device
? device
: "",
5907 node_name
? node_name
: "");
5911 /* If 'base' is in the same chain as 'top', return true. Otherwise,
5912 * return false. If either argument is NULL, return false. */
5913 bool bdrv_chain_contains(BlockDriverState
*top
, BlockDriverState
*base
)
5915 while (top
&& top
!= base
) {
5916 top
= bdrv_filter_or_cow_bs(top
);
5922 BlockDriverState
*bdrv_next_node(BlockDriverState
*bs
)
5925 return QTAILQ_FIRST(&graph_bdrv_states
);
5927 return QTAILQ_NEXT(bs
, node_list
);
5930 BlockDriverState
*bdrv_next_all_states(BlockDriverState
*bs
)
5933 return QTAILQ_FIRST(&all_bdrv_states
);
5935 return QTAILQ_NEXT(bs
, bs_list
);
5938 const char *bdrv_get_node_name(const BlockDriverState
*bs
)
5940 return bs
->node_name
;
5943 const char *bdrv_get_parent_name(const BlockDriverState
*bs
)
5948 /* If multiple parents have a name, just pick the first one. */
5949 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
5950 if (c
->klass
->get_name
) {
5951 name
= c
->klass
->get_name(c
);
5952 if (name
&& *name
) {
5961 /* TODO check what callers really want: bs->node_name or blk_name() */
5962 const char *bdrv_get_device_name(const BlockDriverState
*bs
)
5964 return bdrv_get_parent_name(bs
) ?: "";
5967 /* This can be used to identify nodes that might not have a device
5968 * name associated. Since node and device names live in the same
5969 * namespace, the result is unambiguous. The exception is if both are
5970 * absent, then this returns an empty (non-null) string. */
5971 const char *bdrv_get_device_or_node_name(const BlockDriverState
*bs
)
5973 return bdrv_get_parent_name(bs
) ?: bs
->node_name
;
5976 int bdrv_get_flags(BlockDriverState
*bs
)
5978 return bs
->open_flags
;
5981 int bdrv_has_zero_init_1(BlockDriverState
*bs
)
5986 int bdrv_has_zero_init(BlockDriverState
*bs
)
5988 BlockDriverState
*filtered
;
5994 /* If BS is a copy on write image, it is initialized to
5995 the contents of the base image, which may not be zeroes. */
5996 if (bdrv_cow_child(bs
)) {
5999 if (bs
->drv
->bdrv_has_zero_init
) {
6000 return bs
->drv
->bdrv_has_zero_init(bs
);
6003 filtered
= bdrv_filter_bs(bs
);
6005 return bdrv_has_zero_init(filtered
);
6012 bool bdrv_can_write_zeroes_with_unmap(BlockDriverState
*bs
)
6014 if (!(bs
->open_flags
& BDRV_O_UNMAP
)) {
6018 return bs
->supported_zero_flags
& BDRV_REQ_MAY_UNMAP
;
6021 void bdrv_get_backing_filename(BlockDriverState
*bs
,
6022 char *filename
, int filename_size
)
6024 pstrcpy(filename
, filename_size
, bs
->backing_file
);
6027 int bdrv_get_info(BlockDriverState
*bs
, BlockDriverInfo
*bdi
)
6030 BlockDriver
*drv
= bs
->drv
;
6031 /* if bs->drv == NULL, bs is closed, so there's nothing to do here */
6035 if (!drv
->bdrv_get_info
) {
6036 BlockDriverState
*filtered
= bdrv_filter_bs(bs
);
6038 return bdrv_get_info(filtered
, bdi
);
6042 memset(bdi
, 0, sizeof(*bdi
));
6043 ret
= drv
->bdrv_get_info(bs
, bdi
);
6048 if (bdi
->cluster_size
> BDRV_MAX_ALIGNMENT
) {
6055 ImageInfoSpecific
*bdrv_get_specific_info(BlockDriverState
*bs
,
6058 BlockDriver
*drv
= bs
->drv
;
6059 if (drv
&& drv
->bdrv_get_specific_info
) {
6060 return drv
->bdrv_get_specific_info(bs
, errp
);
6065 BlockStatsSpecific
*bdrv_get_specific_stats(BlockDriverState
*bs
)
6067 BlockDriver
*drv
= bs
->drv
;
6068 if (!drv
|| !drv
->bdrv_get_specific_stats
) {
6071 return drv
->bdrv_get_specific_stats(bs
);
6074 void bdrv_debug_event(BlockDriverState
*bs
, BlkdebugEvent event
)
6076 if (!bs
|| !bs
->drv
|| !bs
->drv
->bdrv_debug_event
) {
6080 bs
->drv
->bdrv_debug_event(bs
, event
);
6083 static BlockDriverState
*bdrv_find_debug_node(BlockDriverState
*bs
)
6085 while (bs
&& bs
->drv
&& !bs
->drv
->bdrv_debug_breakpoint
) {
6086 bs
= bdrv_primary_bs(bs
);
6089 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_breakpoint
) {
6090 assert(bs
->drv
->bdrv_debug_remove_breakpoint
);
6097 int bdrv_debug_breakpoint(BlockDriverState
*bs
, const char *event
,
6100 bs
= bdrv_find_debug_node(bs
);
6102 return bs
->drv
->bdrv_debug_breakpoint(bs
, event
, tag
);
6108 int bdrv_debug_remove_breakpoint(BlockDriverState
*bs
, const char *tag
)
6110 bs
= bdrv_find_debug_node(bs
);
6112 return bs
->drv
->bdrv_debug_remove_breakpoint(bs
, tag
);
6118 int bdrv_debug_resume(BlockDriverState
*bs
, const char *tag
)
6120 while (bs
&& (!bs
->drv
|| !bs
->drv
->bdrv_debug_resume
)) {
6121 bs
= bdrv_primary_bs(bs
);
6124 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_resume
) {
6125 return bs
->drv
->bdrv_debug_resume(bs
, tag
);
6131 bool bdrv_debug_is_suspended(BlockDriverState
*bs
, const char *tag
)
6133 while (bs
&& bs
->drv
&& !bs
->drv
->bdrv_debug_is_suspended
) {
6134 bs
= bdrv_primary_bs(bs
);
6137 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_is_suspended
) {
6138 return bs
->drv
->bdrv_debug_is_suspended(bs
, tag
);
6144 /* backing_file can either be relative, or absolute, or a protocol. If it is
6145 * relative, it must be relative to the chain. So, passing in bs->filename
6146 * from a BDS as backing_file should not be done, as that may be relative to
6147 * the CWD rather than the chain. */
6148 BlockDriverState
*bdrv_find_backing_image(BlockDriverState
*bs
,
6149 const char *backing_file
)
6151 char *filename_full
= NULL
;
6152 char *backing_file_full
= NULL
;
6153 char *filename_tmp
= NULL
;
6154 int is_protocol
= 0;
6155 bool filenames_refreshed
= false;
6156 BlockDriverState
*curr_bs
= NULL
;
6157 BlockDriverState
*retval
= NULL
;
6158 BlockDriverState
*bs_below
;
6160 if (!bs
|| !bs
->drv
|| !backing_file
) {
6164 filename_full
= g_malloc(PATH_MAX
);
6165 backing_file_full
= g_malloc(PATH_MAX
);
6167 is_protocol
= path_has_protocol(backing_file
);
6170 * Being largely a legacy function, skip any filters here
6171 * (because filters do not have normal filenames, so they cannot
6172 * match anyway; and allowing json:{} filenames is a bit out of
6175 for (curr_bs
= bdrv_skip_filters(bs
);
6176 bdrv_cow_child(curr_bs
) != NULL
;
6179 bs_below
= bdrv_backing_chain_next(curr_bs
);
6181 if (bdrv_backing_overridden(curr_bs
)) {
6183 * If the backing file was overridden, we can only compare
6184 * directly against the backing node's filename.
6187 if (!filenames_refreshed
) {
6189 * This will automatically refresh all of the
6190 * filenames in the rest of the backing chain, so we
6191 * only need to do this once.
6193 bdrv_refresh_filename(bs_below
);
6194 filenames_refreshed
= true;
6197 if (strcmp(backing_file
, bs_below
->filename
) == 0) {
6201 } else if (is_protocol
|| path_has_protocol(curr_bs
->backing_file
)) {
6203 * If either of the filename paths is actually a protocol, then
6204 * compare unmodified paths; otherwise make paths relative.
6206 char *backing_file_full_ret
;
6208 if (strcmp(backing_file
, curr_bs
->backing_file
) == 0) {
6212 /* Also check against the full backing filename for the image */
6213 backing_file_full_ret
= bdrv_get_full_backing_filename(curr_bs
,
6215 if (backing_file_full_ret
) {
6216 bool equal
= strcmp(backing_file
, backing_file_full_ret
) == 0;
6217 g_free(backing_file_full_ret
);
6224 /* If not an absolute filename path, make it relative to the current
6225 * image's filename path */
6226 filename_tmp
= bdrv_make_absolute_filename(curr_bs
, backing_file
,
6228 /* We are going to compare canonicalized absolute pathnames */
6229 if (!filename_tmp
|| !realpath(filename_tmp
, filename_full
)) {
6230 g_free(filename_tmp
);
6233 g_free(filename_tmp
);
6235 /* We need to make sure the backing filename we are comparing against
6236 * is relative to the current image filename (or absolute) */
6237 filename_tmp
= bdrv_get_full_backing_filename(curr_bs
, NULL
);
6238 if (!filename_tmp
|| !realpath(filename_tmp
, backing_file_full
)) {
6239 g_free(filename_tmp
);
6242 g_free(filename_tmp
);
6244 if (strcmp(backing_file_full
, filename_full
) == 0) {
6251 g_free(filename_full
);
6252 g_free(backing_file_full
);
6256 void bdrv_init(void)
6258 #ifdef CONFIG_BDRV_WHITELIST_TOOLS
6259 use_bdrv_whitelist
= 1;
6261 module_call_init(MODULE_INIT_BLOCK
);
6264 void bdrv_init_with_whitelist(void)
6266 use_bdrv_whitelist
= 1;
6270 int coroutine_fn
bdrv_co_invalidate_cache(BlockDriverState
*bs
, Error
**errp
)
6272 BdrvChild
*child
, *parent
;
6273 Error
*local_err
= NULL
;
6275 BdrvDirtyBitmap
*bm
;
6281 QLIST_FOREACH(child
, &bs
->children
, next
) {
6282 bdrv_co_invalidate_cache(child
->bs
, &local_err
);
6284 error_propagate(errp
, local_err
);
6290 * Update permissions, they may differ for inactive nodes.
6292 * Note that the required permissions of inactive images are always a
6293 * subset of the permissions required after activating the image. This
6294 * allows us to just get the permissions upfront without restricting
6295 * drv->bdrv_invalidate_cache().
6297 * It also means that in error cases, we don't have to try and revert to
6298 * the old permissions (which is an operation that could fail, too). We can
6299 * just keep the extended permissions for the next time that an activation
6300 * of the image is tried.
6302 if (bs
->open_flags
& BDRV_O_INACTIVE
) {
6303 bs
->open_flags
&= ~BDRV_O_INACTIVE
;
6304 ret
= bdrv_refresh_perms(bs
, errp
);
6306 bs
->open_flags
|= BDRV_O_INACTIVE
;
6310 if (bs
->drv
->bdrv_co_invalidate_cache
) {
6311 bs
->drv
->bdrv_co_invalidate_cache(bs
, &local_err
);
6313 bs
->open_flags
|= BDRV_O_INACTIVE
;
6314 error_propagate(errp
, local_err
);
6319 FOR_EACH_DIRTY_BITMAP(bs
, bm
) {
6320 bdrv_dirty_bitmap_skip_store(bm
, false);
6323 ret
= refresh_total_sectors(bs
, bs
->total_sectors
);
6325 bs
->open_flags
|= BDRV_O_INACTIVE
;
6326 error_setg_errno(errp
, -ret
, "Could not refresh total sector count");
6331 QLIST_FOREACH(parent
, &bs
->parents
, next_parent
) {
6332 if (parent
->klass
->activate
) {
6333 parent
->klass
->activate(parent
, &local_err
);
6335 bs
->open_flags
|= BDRV_O_INACTIVE
;
6336 error_propagate(errp
, local_err
);
6345 void bdrv_invalidate_cache_all(Error
**errp
)
6347 BlockDriverState
*bs
;
6348 BdrvNextIterator it
;
6350 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
6351 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
6354 aio_context_acquire(aio_context
);
6355 ret
= bdrv_invalidate_cache(bs
, errp
);
6356 aio_context_release(aio_context
);
6358 bdrv_next_cleanup(&it
);
6364 static bool bdrv_has_bds_parent(BlockDriverState
*bs
, bool only_active
)
6368 QLIST_FOREACH(parent
, &bs
->parents
, next_parent
) {
6369 if (parent
->klass
->parent_is_bds
) {
6370 BlockDriverState
*parent_bs
= parent
->opaque
;
6371 if (!only_active
|| !(parent_bs
->open_flags
& BDRV_O_INACTIVE
)) {
6380 static int bdrv_inactivate_recurse(BlockDriverState
*bs
)
6382 BdrvChild
*child
, *parent
;
6384 uint64_t cumulative_perms
, cumulative_shared_perms
;
6390 /* Make sure that we don't inactivate a child before its parent.
6391 * It will be covered by recursion from the yet active parent. */
6392 if (bdrv_has_bds_parent(bs
, true)) {
6396 assert(!(bs
->open_flags
& BDRV_O_INACTIVE
));
6398 /* Inactivate this node */
6399 if (bs
->drv
->bdrv_inactivate
) {
6400 ret
= bs
->drv
->bdrv_inactivate(bs
);
6406 QLIST_FOREACH(parent
, &bs
->parents
, next_parent
) {
6407 if (parent
->klass
->inactivate
) {
6408 ret
= parent
->klass
->inactivate(parent
);
6415 bdrv_get_cumulative_perm(bs
, &cumulative_perms
,
6416 &cumulative_shared_perms
);
6417 if (cumulative_perms
& (BLK_PERM_WRITE
| BLK_PERM_WRITE_UNCHANGED
)) {
6418 /* Our inactive parents still need write access. Inactivation failed. */
6422 bs
->open_flags
|= BDRV_O_INACTIVE
;
6425 * Update permissions, they may differ for inactive nodes.
6426 * We only tried to loosen restrictions, so errors are not fatal, ignore
6429 bdrv_refresh_perms(bs
, NULL
);
6431 /* Recursively inactivate children */
6432 QLIST_FOREACH(child
, &bs
->children
, next
) {
6433 ret
= bdrv_inactivate_recurse(child
->bs
);
6442 int bdrv_inactivate_all(void)
6444 BlockDriverState
*bs
= NULL
;
6445 BdrvNextIterator it
;
6447 GSList
*aio_ctxs
= NULL
, *ctx
;
6449 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
6450 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
6452 if (!g_slist_find(aio_ctxs
, aio_context
)) {
6453 aio_ctxs
= g_slist_prepend(aio_ctxs
, aio_context
);
6454 aio_context_acquire(aio_context
);
6458 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
6459 /* Nodes with BDS parents are covered by recursion from the last
6460 * parent that gets inactivated. Don't inactivate them a second
6461 * time if that has already happened. */
6462 if (bdrv_has_bds_parent(bs
, false)) {
6465 ret
= bdrv_inactivate_recurse(bs
);
6467 bdrv_next_cleanup(&it
);
6473 for (ctx
= aio_ctxs
; ctx
!= NULL
; ctx
= ctx
->next
) {
6474 AioContext
*aio_context
= ctx
->data
;
6475 aio_context_release(aio_context
);
6477 g_slist_free(aio_ctxs
);
6482 /**************************************************************/
6483 /* removable device support */
6486 * Return TRUE if the media is present
6488 bool bdrv_is_inserted(BlockDriverState
*bs
)
6490 BlockDriver
*drv
= bs
->drv
;
6496 if (drv
->bdrv_is_inserted
) {
6497 return drv
->bdrv_is_inserted(bs
);
6499 QLIST_FOREACH(child
, &bs
->children
, next
) {
6500 if (!bdrv_is_inserted(child
->bs
)) {
6508 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
6510 void bdrv_eject(BlockDriverState
*bs
, bool eject_flag
)
6512 BlockDriver
*drv
= bs
->drv
;
6514 if (drv
&& drv
->bdrv_eject
) {
6515 drv
->bdrv_eject(bs
, eject_flag
);
6520 * Lock or unlock the media (if it is locked, the user won't be able
6521 * to eject it manually).
6523 void bdrv_lock_medium(BlockDriverState
*bs
, bool locked
)
6525 BlockDriver
*drv
= bs
->drv
;
6527 trace_bdrv_lock_medium(bs
, locked
);
6529 if (drv
&& drv
->bdrv_lock_medium
) {
6530 drv
->bdrv_lock_medium(bs
, locked
);
6534 /* Get a reference to bs */
6535 void bdrv_ref(BlockDriverState
*bs
)
6540 /* Release a previously grabbed reference to bs.
6541 * If after releasing, reference count is zero, the BlockDriverState is
6543 void bdrv_unref(BlockDriverState
*bs
)
6548 assert(bs
->refcnt
> 0);
6549 if (--bs
->refcnt
== 0) {
6554 struct BdrvOpBlocker
{
6556 QLIST_ENTRY(BdrvOpBlocker
) list
;
6559 bool bdrv_op_is_blocked(BlockDriverState
*bs
, BlockOpType op
, Error
**errp
)
6561 BdrvOpBlocker
*blocker
;
6562 assert((int) op
>= 0 && op
< BLOCK_OP_TYPE_MAX
);
6563 if (!QLIST_EMPTY(&bs
->op_blockers
[op
])) {
6564 blocker
= QLIST_FIRST(&bs
->op_blockers
[op
]);
6565 error_propagate_prepend(errp
, error_copy(blocker
->reason
),
6566 "Node '%s' is busy: ",
6567 bdrv_get_device_or_node_name(bs
));
6573 void bdrv_op_block(BlockDriverState
*bs
, BlockOpType op
, Error
*reason
)
6575 BdrvOpBlocker
*blocker
;
6576 assert((int) op
>= 0 && op
< BLOCK_OP_TYPE_MAX
);
6578 blocker
= g_new0(BdrvOpBlocker
, 1);
6579 blocker
->reason
= reason
;
6580 QLIST_INSERT_HEAD(&bs
->op_blockers
[op
], blocker
, list
);
6583 void bdrv_op_unblock(BlockDriverState
*bs
, BlockOpType op
, Error
*reason
)
6585 BdrvOpBlocker
*blocker
, *next
;
6586 assert((int) op
>= 0 && op
< BLOCK_OP_TYPE_MAX
);
6587 QLIST_FOREACH_SAFE(blocker
, &bs
->op_blockers
[op
], list
, next
) {
6588 if (blocker
->reason
== reason
) {
6589 QLIST_REMOVE(blocker
, list
);
6595 void bdrv_op_block_all(BlockDriverState
*bs
, Error
*reason
)
6598 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
6599 bdrv_op_block(bs
, i
, reason
);
6603 void bdrv_op_unblock_all(BlockDriverState
*bs
, Error
*reason
)
6606 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
6607 bdrv_op_unblock(bs
, i
, reason
);
6611 bool bdrv_op_blocker_is_empty(BlockDriverState
*bs
)
6615 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
6616 if (!QLIST_EMPTY(&bs
->op_blockers
[i
])) {
6623 void bdrv_img_create(const char *filename
, const char *fmt
,
6624 const char *base_filename
, const char *base_fmt
,
6625 char *options
, uint64_t img_size
, int flags
, bool quiet
,
6628 QemuOptsList
*create_opts
= NULL
;
6629 QemuOpts
*opts
= NULL
;
6630 const char *backing_fmt
, *backing_file
;
6632 BlockDriver
*drv
, *proto_drv
;
6633 Error
*local_err
= NULL
;
6636 /* Find driver and parse its options */
6637 drv
= bdrv_find_format(fmt
);
6639 error_setg(errp
, "Unknown file format '%s'", fmt
);
6643 proto_drv
= bdrv_find_protocol(filename
, true, errp
);
6648 if (!drv
->create_opts
) {
6649 error_setg(errp
, "Format driver '%s' does not support image creation",
6654 if (!proto_drv
->create_opts
) {
6655 error_setg(errp
, "Protocol driver '%s' does not support image creation",
6656 proto_drv
->format_name
);
6660 /* Create parameter list */
6661 create_opts
= qemu_opts_append(create_opts
, drv
->create_opts
);
6662 create_opts
= qemu_opts_append(create_opts
, proto_drv
->create_opts
);
6664 opts
= qemu_opts_create(create_opts
, NULL
, 0, &error_abort
);
6666 /* Parse -o options */
6668 if (!qemu_opts_do_parse(opts
, options
, NULL
, errp
)) {
6673 if (!qemu_opt_get(opts
, BLOCK_OPT_SIZE
)) {
6674 qemu_opt_set_number(opts
, BLOCK_OPT_SIZE
, img_size
, &error_abort
);
6675 } else if (img_size
!= UINT64_C(-1)) {
6676 error_setg(errp
, "The image size must be specified only once");
6680 if (base_filename
) {
6681 if (!qemu_opt_set(opts
, BLOCK_OPT_BACKING_FILE
, base_filename
,
6683 error_setg(errp
, "Backing file not supported for file format '%s'",
6690 if (!qemu_opt_set(opts
, BLOCK_OPT_BACKING_FMT
, base_fmt
, NULL
)) {
6691 error_setg(errp
, "Backing file format not supported for file "
6692 "format '%s'", fmt
);
6697 backing_file
= qemu_opt_get(opts
, BLOCK_OPT_BACKING_FILE
);
6699 if (!strcmp(filename
, backing_file
)) {
6700 error_setg(errp
, "Error: Trying to create an image with the "
6701 "same filename as the backing file");
6704 if (backing_file
[0] == '\0') {
6705 error_setg(errp
, "Expected backing file name, got empty string");
6710 backing_fmt
= qemu_opt_get(opts
, BLOCK_OPT_BACKING_FMT
);
6712 /* The size for the image must always be specified, unless we have a backing
6713 * file and we have not been forbidden from opening it. */
6714 size
= qemu_opt_get_size(opts
, BLOCK_OPT_SIZE
, img_size
);
6715 if (backing_file
&& !(flags
& BDRV_O_NO_BACKING
)) {
6716 BlockDriverState
*bs
;
6719 QDict
*backing_options
= NULL
;
6722 bdrv_get_full_backing_filename_from_filename(filename
, backing_file
,
6727 assert(full_backing
);
6730 * No need to do I/O here, which allows us to open encrypted
6731 * backing images without needing the secret
6734 back_flags
&= ~(BDRV_O_RDWR
| BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
);
6735 back_flags
|= BDRV_O_NO_IO
;
6737 backing_options
= qdict_new();
6739 qdict_put_str(backing_options
, "driver", backing_fmt
);
6741 qdict_put_bool(backing_options
, BDRV_OPT_FORCE_SHARE
, true);
6743 bs
= bdrv_open(full_backing
, NULL
, backing_options
, back_flags
,
6745 g_free(full_backing
);
6747 error_append_hint(&local_err
, "Could not open backing image.\n");
6751 error_setg(&local_err
,
6752 "Backing file specified without backing format");
6753 error_append_hint(&local_err
, "Detected format of %s.",
6754 bs
->drv
->format_name
);
6758 /* Opened BS, have no size */
6759 size
= bdrv_getlength(bs
);
6761 error_setg_errno(errp
, -size
, "Could not get size of '%s'",
6766 qemu_opt_set_number(opts
, BLOCK_OPT_SIZE
, size
, &error_abort
);
6770 /* (backing_file && !(flags & BDRV_O_NO_BACKING)) */
6771 } else if (backing_file
&& !backing_fmt
) {
6772 error_setg(&local_err
,
6773 "Backing file specified without backing format");
6778 error_setg(errp
, "Image creation needs a size parameter");
6783 printf("Formatting '%s', fmt=%s ", filename
, fmt
);
6784 qemu_opts_print(opts
, " ");
6789 ret
= bdrv_create(drv
, filename
, opts
, &local_err
);
6791 if (ret
== -EFBIG
) {
6792 /* This is generally a better message than whatever the driver would
6793 * deliver (especially because of the cluster_size_hint), since that
6794 * is most probably not much different from "image too large". */
6795 const char *cluster_size_hint
= "";
6796 if (qemu_opt_get_size(opts
, BLOCK_OPT_CLUSTER_SIZE
, 0)) {
6797 cluster_size_hint
= " (try using a larger cluster size)";
6799 error_setg(errp
, "The image size is too large for file format '%s'"
6800 "%s", fmt
, cluster_size_hint
);
6801 error_free(local_err
);
6806 qemu_opts_del(opts
);
6807 qemu_opts_free(create_opts
);
6808 error_propagate(errp
, local_err
);
6811 AioContext
*bdrv_get_aio_context(BlockDriverState
*bs
)
6813 return bs
? bs
->aio_context
: qemu_get_aio_context();
6816 AioContext
*coroutine_fn
bdrv_co_enter(BlockDriverState
*bs
)
6818 Coroutine
*self
= qemu_coroutine_self();
6819 AioContext
*old_ctx
= qemu_coroutine_get_aio_context(self
);
6820 AioContext
*new_ctx
;
6823 * Increase bs->in_flight to ensure that this operation is completed before
6824 * moving the node to a different AioContext. Read new_ctx only afterwards.
6826 bdrv_inc_in_flight(bs
);
6828 new_ctx
= bdrv_get_aio_context(bs
);
6829 aio_co_reschedule_self(new_ctx
);
6833 void coroutine_fn
bdrv_co_leave(BlockDriverState
*bs
, AioContext
*old_ctx
)
6835 aio_co_reschedule_self(old_ctx
);
6836 bdrv_dec_in_flight(bs
);
6839 void coroutine_fn
bdrv_co_lock(BlockDriverState
*bs
)
6841 AioContext
*ctx
= bdrv_get_aio_context(bs
);
6843 /* In the main thread, bs->aio_context won't change concurrently */
6844 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
6847 * We're in coroutine context, so we already hold the lock of the main
6848 * loop AioContext. Don't lock it twice to avoid deadlocks.
6850 assert(qemu_in_coroutine());
6851 if (ctx
!= qemu_get_aio_context()) {
6852 aio_context_acquire(ctx
);
6856 void coroutine_fn
bdrv_co_unlock(BlockDriverState
*bs
)
6858 AioContext
*ctx
= bdrv_get_aio_context(bs
);
6860 assert(qemu_in_coroutine());
6861 if (ctx
!= qemu_get_aio_context()) {
6862 aio_context_release(ctx
);
6866 void bdrv_coroutine_enter(BlockDriverState
*bs
, Coroutine
*co
)
6868 aio_co_enter(bdrv_get_aio_context(bs
), co
);
6871 static void bdrv_do_remove_aio_context_notifier(BdrvAioNotifier
*ban
)
6873 QLIST_REMOVE(ban
, list
);
6877 static void bdrv_detach_aio_context(BlockDriverState
*bs
)
6879 BdrvAioNotifier
*baf
, *baf_tmp
;
6881 assert(!bs
->walking_aio_notifiers
);
6882 bs
->walking_aio_notifiers
= true;
6883 QLIST_FOREACH_SAFE(baf
, &bs
->aio_notifiers
, list
, baf_tmp
) {
6885 bdrv_do_remove_aio_context_notifier(baf
);
6887 baf
->detach_aio_context(baf
->opaque
);
6890 /* Never mind iterating again to check for ->deleted. bdrv_close() will
6891 * remove remaining aio notifiers if we aren't called again.
6893 bs
->walking_aio_notifiers
= false;
6895 if (bs
->drv
&& bs
->drv
->bdrv_detach_aio_context
) {
6896 bs
->drv
->bdrv_detach_aio_context(bs
);
6899 if (bs
->quiesce_counter
) {
6900 aio_enable_external(bs
->aio_context
);
6902 bs
->aio_context
= NULL
;
6905 static void bdrv_attach_aio_context(BlockDriverState
*bs
,
6906 AioContext
*new_context
)
6908 BdrvAioNotifier
*ban
, *ban_tmp
;
6910 if (bs
->quiesce_counter
) {
6911 aio_disable_external(new_context
);
6914 bs
->aio_context
= new_context
;
6916 if (bs
->drv
&& bs
->drv
->bdrv_attach_aio_context
) {
6917 bs
->drv
->bdrv_attach_aio_context(bs
, new_context
);
6920 assert(!bs
->walking_aio_notifiers
);
6921 bs
->walking_aio_notifiers
= true;
6922 QLIST_FOREACH_SAFE(ban
, &bs
->aio_notifiers
, list
, ban_tmp
) {
6924 bdrv_do_remove_aio_context_notifier(ban
);
6926 ban
->attached_aio_context(new_context
, ban
->opaque
);
6929 bs
->walking_aio_notifiers
= false;
6933 * Changes the AioContext used for fd handlers, timers, and BHs by this
6934 * BlockDriverState and all its children and parents.
6936 * Must be called from the main AioContext.
6938 * The caller must own the AioContext lock for the old AioContext of bs, but it
6939 * must not own the AioContext lock for new_context (unless new_context is the
6940 * same as the current context of bs).
6942 * @ignore will accumulate all visited BdrvChild object. The caller is
6943 * responsible for freeing the list afterwards.
6945 void bdrv_set_aio_context_ignore(BlockDriverState
*bs
,
6946 AioContext
*new_context
, GSList
**ignore
)
6948 AioContext
*old_context
= bdrv_get_aio_context(bs
);
6949 GSList
*children_to_process
= NULL
;
6950 GSList
*parents_to_process
= NULL
;
6952 BdrvChild
*child
, *parent
;
6954 g_assert(qemu_get_current_aio_context() == qemu_get_aio_context());
6956 if (old_context
== new_context
) {
6960 bdrv_drained_begin(bs
);
6962 QLIST_FOREACH(child
, &bs
->children
, next
) {
6963 if (g_slist_find(*ignore
, child
)) {
6966 *ignore
= g_slist_prepend(*ignore
, child
);
6967 children_to_process
= g_slist_prepend(children_to_process
, child
);
6970 QLIST_FOREACH(parent
, &bs
->parents
, next_parent
) {
6971 if (g_slist_find(*ignore
, parent
)) {
6974 *ignore
= g_slist_prepend(*ignore
, parent
);
6975 parents_to_process
= g_slist_prepend(parents_to_process
, parent
);
6978 for (entry
= children_to_process
;
6980 entry
= g_slist_next(entry
)) {
6981 child
= entry
->data
;
6982 bdrv_set_aio_context_ignore(child
->bs
, new_context
, ignore
);
6984 g_slist_free(children_to_process
);
6986 for (entry
= parents_to_process
;
6988 entry
= g_slist_next(entry
)) {
6989 parent
= entry
->data
;
6990 assert(parent
->klass
->set_aio_ctx
);
6991 parent
->klass
->set_aio_ctx(parent
, new_context
, ignore
);
6993 g_slist_free(parents_to_process
);
6995 bdrv_detach_aio_context(bs
);
6997 /* Acquire the new context, if necessary */
6998 if (qemu_get_aio_context() != new_context
) {
6999 aio_context_acquire(new_context
);
7002 bdrv_attach_aio_context(bs
, new_context
);
7005 * If this function was recursively called from
7006 * bdrv_set_aio_context_ignore(), there may be nodes in the
7007 * subtree that have not yet been moved to the new AioContext.
7008 * Release the old one so bdrv_drained_end() can poll them.
7010 if (qemu_get_aio_context() != old_context
) {
7011 aio_context_release(old_context
);
7014 bdrv_drained_end(bs
);
7016 if (qemu_get_aio_context() != old_context
) {
7017 aio_context_acquire(old_context
);
7019 if (qemu_get_aio_context() != new_context
) {
7020 aio_context_release(new_context
);
7024 static bool bdrv_parent_can_set_aio_context(BdrvChild
*c
, AioContext
*ctx
,
7025 GSList
**ignore
, Error
**errp
)
7027 if (g_slist_find(*ignore
, c
)) {
7030 *ignore
= g_slist_prepend(*ignore
, c
);
7033 * A BdrvChildClass that doesn't handle AioContext changes cannot
7034 * tolerate any AioContext changes
7036 if (!c
->klass
->can_set_aio_ctx
) {
7037 char *user
= bdrv_child_user_desc(c
);
7038 error_setg(errp
, "Changing iothreads is not supported by %s", user
);
7042 if (!c
->klass
->can_set_aio_ctx(c
, ctx
, ignore
, errp
)) {
7043 assert(!errp
|| *errp
);
7049 bool bdrv_child_can_set_aio_context(BdrvChild
*c
, AioContext
*ctx
,
7050 GSList
**ignore
, Error
**errp
)
7052 if (g_slist_find(*ignore
, c
)) {
7055 *ignore
= g_slist_prepend(*ignore
, c
);
7056 return bdrv_can_set_aio_context(c
->bs
, ctx
, ignore
, errp
);
7059 /* @ignore will accumulate all visited BdrvChild object. The caller is
7060 * responsible for freeing the list afterwards. */
7061 bool bdrv_can_set_aio_context(BlockDriverState
*bs
, AioContext
*ctx
,
7062 GSList
**ignore
, Error
**errp
)
7066 if (bdrv_get_aio_context(bs
) == ctx
) {
7070 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
7071 if (!bdrv_parent_can_set_aio_context(c
, ctx
, ignore
, errp
)) {
7075 QLIST_FOREACH(c
, &bs
->children
, next
) {
7076 if (!bdrv_child_can_set_aio_context(c
, ctx
, ignore
, errp
)) {
7084 int bdrv_child_try_set_aio_context(BlockDriverState
*bs
, AioContext
*ctx
,
7085 BdrvChild
*ignore_child
, Error
**errp
)
7090 ignore
= ignore_child
? g_slist_prepend(NULL
, ignore_child
) : NULL
;
7091 ret
= bdrv_can_set_aio_context(bs
, ctx
, &ignore
, errp
);
7092 g_slist_free(ignore
);
7098 ignore
= ignore_child
? g_slist_prepend(NULL
, ignore_child
) : NULL
;
7099 bdrv_set_aio_context_ignore(bs
, ctx
, &ignore
);
7100 g_slist_free(ignore
);
7105 int bdrv_try_set_aio_context(BlockDriverState
*bs
, AioContext
*ctx
,
7108 return bdrv_child_try_set_aio_context(bs
, ctx
, NULL
, errp
);
7111 void bdrv_add_aio_context_notifier(BlockDriverState
*bs
,
7112 void (*attached_aio_context
)(AioContext
*new_context
, void *opaque
),
7113 void (*detach_aio_context
)(void *opaque
), void *opaque
)
7115 BdrvAioNotifier
*ban
= g_new(BdrvAioNotifier
, 1);
7116 *ban
= (BdrvAioNotifier
){
7117 .attached_aio_context
= attached_aio_context
,
7118 .detach_aio_context
= detach_aio_context
,
7122 QLIST_INSERT_HEAD(&bs
->aio_notifiers
, ban
, list
);
7125 void bdrv_remove_aio_context_notifier(BlockDriverState
*bs
,
7126 void (*attached_aio_context
)(AioContext
*,
7128 void (*detach_aio_context
)(void *),
7131 BdrvAioNotifier
*ban
, *ban_next
;
7133 QLIST_FOREACH_SAFE(ban
, &bs
->aio_notifiers
, list
, ban_next
) {
7134 if (ban
->attached_aio_context
== attached_aio_context
&&
7135 ban
->detach_aio_context
== detach_aio_context
&&
7136 ban
->opaque
== opaque
&&
7137 ban
->deleted
== false)
7139 if (bs
->walking_aio_notifiers
) {
7140 ban
->deleted
= true;
7142 bdrv_do_remove_aio_context_notifier(ban
);
7151 int bdrv_amend_options(BlockDriverState
*bs
, QemuOpts
*opts
,
7152 BlockDriverAmendStatusCB
*status_cb
, void *cb_opaque
,
7157 error_setg(errp
, "Node is ejected");
7160 if (!bs
->drv
->bdrv_amend_options
) {
7161 error_setg(errp
, "Block driver '%s' does not support option amendment",
7162 bs
->drv
->format_name
);
7165 return bs
->drv
->bdrv_amend_options(bs
, opts
, status_cb
,
7166 cb_opaque
, force
, errp
);
7170 * This function checks whether the given @to_replace is allowed to be
7171 * replaced by a node that always shows the same data as @bs. This is
7172 * used for example to verify whether the mirror job can replace
7173 * @to_replace by the target mirrored from @bs.
7174 * To be replaceable, @bs and @to_replace may either be guaranteed to
7175 * always show the same data (because they are only connected through
7176 * filters), or some driver may allow replacing one of its children
7177 * because it can guarantee that this child's data is not visible at
7178 * all (for example, for dissenting quorum children that have no other
7181 bool bdrv_recurse_can_replace(BlockDriverState
*bs
,
7182 BlockDriverState
*to_replace
)
7184 BlockDriverState
*filtered
;
7186 if (!bs
|| !bs
->drv
) {
7190 if (bs
== to_replace
) {
7194 /* See what the driver can do */
7195 if (bs
->drv
->bdrv_recurse_can_replace
) {
7196 return bs
->drv
->bdrv_recurse_can_replace(bs
, to_replace
);
7199 /* For filters without an own implementation, we can recurse on our own */
7200 filtered
= bdrv_filter_bs(bs
);
7202 return bdrv_recurse_can_replace(filtered
, to_replace
);
7210 * Check whether the given @node_name can be replaced by a node that
7211 * has the same data as @parent_bs. If so, return @node_name's BDS;
7214 * @node_name must be a (recursive) *child of @parent_bs (or this
7215 * function will return NULL).
7217 * The result (whether the node can be replaced or not) is only valid
7218 * for as long as no graph or permission changes occur.
7220 BlockDriverState
*check_to_replace_node(BlockDriverState
*parent_bs
,
7221 const char *node_name
, Error
**errp
)
7223 BlockDriverState
*to_replace_bs
= bdrv_find_node(node_name
);
7224 AioContext
*aio_context
;
7226 if (!to_replace_bs
) {
7227 error_setg(errp
, "Failed to find node with node-name='%s'", node_name
);
7231 aio_context
= bdrv_get_aio_context(to_replace_bs
);
7232 aio_context_acquire(aio_context
);
7234 if (bdrv_op_is_blocked(to_replace_bs
, BLOCK_OP_TYPE_REPLACE
, errp
)) {
7235 to_replace_bs
= NULL
;
7239 /* We don't want arbitrary node of the BDS chain to be replaced only the top
7240 * most non filter in order to prevent data corruption.
7241 * Another benefit is that this tests exclude backing files which are
7242 * blocked by the backing blockers.
7244 if (!bdrv_recurse_can_replace(parent_bs
, to_replace_bs
)) {
7245 error_setg(errp
, "Cannot replace '%s' by a node mirrored from '%s', "
7246 "because it cannot be guaranteed that doing so would not "
7247 "lead to an abrupt change of visible data",
7248 node_name
, parent_bs
->node_name
);
7249 to_replace_bs
= NULL
;
7254 aio_context_release(aio_context
);
7255 return to_replace_bs
;
7259 * Iterates through the list of runtime option keys that are said to
7260 * be "strong" for a BDS. An option is called "strong" if it changes
7261 * a BDS's data. For example, the null block driver's "size" and
7262 * "read-zeroes" options are strong, but its "latency-ns" option is
7265 * If a key returned by this function ends with a dot, all options
7266 * starting with that prefix are strong.
7268 static const char *const *strong_options(BlockDriverState
*bs
,
7269 const char *const *curopt
)
7271 static const char *const global_options
[] = {
7272 "driver", "filename", NULL
7276 return &global_options
[0];
7280 if (curopt
== &global_options
[ARRAY_SIZE(global_options
) - 1] && bs
->drv
) {
7281 curopt
= bs
->drv
->strong_runtime_opts
;
7284 return (curopt
&& *curopt
) ? curopt
: NULL
;
7288 * Copies all strong runtime options from bs->options to the given
7289 * QDict. The set of strong option keys is determined by invoking
7292 * Returns true iff any strong option was present in bs->options (and
7293 * thus copied to the target QDict) with the exception of "filename"
7294 * and "driver". The caller is expected to use this value to decide
7295 * whether the existence of strong options prevents the generation of
7298 static bool append_strong_runtime_options(QDict
*d
, BlockDriverState
*bs
)
7300 bool found_any
= false;
7301 const char *const *option_name
= NULL
;
7307 while ((option_name
= strong_options(bs
, option_name
))) {
7308 bool option_given
= false;
7310 assert(strlen(*option_name
) > 0);
7311 if ((*option_name
)[strlen(*option_name
) - 1] != '.') {
7312 QObject
*entry
= qdict_get(bs
->options
, *option_name
);
7317 qdict_put_obj(d
, *option_name
, qobject_ref(entry
));
7318 option_given
= true;
7320 const QDictEntry
*entry
;
7321 for (entry
= qdict_first(bs
->options
); entry
;
7322 entry
= qdict_next(bs
->options
, entry
))
7324 if (strstart(qdict_entry_key(entry
), *option_name
, NULL
)) {
7325 qdict_put_obj(d
, qdict_entry_key(entry
),
7326 qobject_ref(qdict_entry_value(entry
)));
7327 option_given
= true;
7332 /* While "driver" and "filename" need to be included in a JSON filename,
7333 * their existence does not prohibit generation of a plain filename. */
7334 if (!found_any
&& option_given
&&
7335 strcmp(*option_name
, "driver") && strcmp(*option_name
, "filename"))
7341 if (!qdict_haskey(d
, "driver")) {
7342 /* Drivers created with bdrv_new_open_driver() may not have a
7343 * @driver option. Add it here. */
7344 qdict_put_str(d
, "driver", bs
->drv
->format_name
);
7350 /* Note: This function may return false positives; it may return true
7351 * even if opening the backing file specified by bs's image header
7352 * would result in exactly bs->backing. */
7353 bool bdrv_backing_overridden(BlockDriverState
*bs
)
7356 return strcmp(bs
->auto_backing_file
,
7357 bs
->backing
->bs
->filename
);
7359 /* No backing BDS, so if the image header reports any backing
7360 * file, it must have been suppressed */
7361 return bs
->auto_backing_file
[0] != '\0';
7365 /* Updates the following BDS fields:
7366 * - exact_filename: A filename which may be used for opening a block device
7367 * which (mostly) equals the given BDS (even without any
7368 * other options; so reading and writing must return the same
7369 * results, but caching etc. may be different)
7370 * - full_open_options: Options which, when given when opening a block device
7371 * (without a filename), result in a BDS (mostly)
7372 * equalling the given one
7373 * - filename: If exact_filename is set, it is copied here. Otherwise,
7374 * full_open_options is converted to a JSON object, prefixed with
7375 * "json:" (for use through the JSON pseudo protocol) and put here.
7377 void bdrv_refresh_filename(BlockDriverState
*bs
)
7379 BlockDriver
*drv
= bs
->drv
;
7381 BlockDriverState
*primary_child_bs
;
7383 bool backing_overridden
;
7384 bool generate_json_filename
; /* Whether our default implementation should
7385 fill exact_filename (false) or not (true) */
7391 /* This BDS's file name may depend on any of its children's file names, so
7392 * refresh those first */
7393 QLIST_FOREACH(child
, &bs
->children
, next
) {
7394 bdrv_refresh_filename(child
->bs
);
7398 /* For implicit nodes, just copy everything from the single child */
7399 child
= QLIST_FIRST(&bs
->children
);
7400 assert(QLIST_NEXT(child
, next
) == NULL
);
7402 pstrcpy(bs
->exact_filename
, sizeof(bs
->exact_filename
),
7403 child
->bs
->exact_filename
);
7404 pstrcpy(bs
->filename
, sizeof(bs
->filename
), child
->bs
->filename
);
7406 qobject_unref(bs
->full_open_options
);
7407 bs
->full_open_options
= qobject_ref(child
->bs
->full_open_options
);
7412 backing_overridden
= bdrv_backing_overridden(bs
);
7414 if (bs
->open_flags
& BDRV_O_NO_IO
) {
7415 /* Without I/O, the backing file does not change anything.
7416 * Therefore, in such a case (primarily qemu-img), we can
7417 * pretend the backing file has not been overridden even if
7418 * it technically has been. */
7419 backing_overridden
= false;
7422 /* Gather the options QDict */
7424 generate_json_filename
= append_strong_runtime_options(opts
, bs
);
7425 generate_json_filename
|= backing_overridden
;
7427 if (drv
->bdrv_gather_child_options
) {
7428 /* Some block drivers may not want to present all of their children's
7429 * options, or name them differently from BdrvChild.name */
7430 drv
->bdrv_gather_child_options(bs
, opts
, backing_overridden
);
7432 QLIST_FOREACH(child
, &bs
->children
, next
) {
7433 if (child
== bs
->backing
&& !backing_overridden
) {
7434 /* We can skip the backing BDS if it has not been overridden */
7438 qdict_put(opts
, child
->name
,
7439 qobject_ref(child
->bs
->full_open_options
));
7442 if (backing_overridden
&& !bs
->backing
) {
7443 /* Force no backing file */
7444 qdict_put_null(opts
, "backing");
7448 qobject_unref(bs
->full_open_options
);
7449 bs
->full_open_options
= opts
;
7451 primary_child_bs
= bdrv_primary_bs(bs
);
7453 if (drv
->bdrv_refresh_filename
) {
7454 /* Obsolete information is of no use here, so drop the old file name
7455 * information before refreshing it */
7456 bs
->exact_filename
[0] = '\0';
7458 drv
->bdrv_refresh_filename(bs
);
7459 } else if (primary_child_bs
) {
7461 * Try to reconstruct valid information from the underlying
7462 * file -- this only works for format nodes (filter nodes
7463 * cannot be probed and as such must be selected by the user
7464 * either through an options dict, or through a special
7465 * filename which the filter driver must construct in its
7466 * .bdrv_refresh_filename() implementation).
7469 bs
->exact_filename
[0] = '\0';
7472 * We can use the underlying file's filename if:
7473 * - it has a filename,
7474 * - the current BDS is not a filter,
7475 * - the file is a protocol BDS, and
7476 * - opening that file (as this BDS's format) will automatically create
7477 * the BDS tree we have right now, that is:
7478 * - the user did not significantly change this BDS's behavior with
7479 * some explicit (strong) options
7480 * - no non-file child of this BDS has been overridden by the user
7481 * Both of these conditions are represented by generate_json_filename.
7483 if (primary_child_bs
->exact_filename
[0] &&
7484 primary_child_bs
->drv
->bdrv_file_open
&&
7485 !drv
->is_filter
&& !generate_json_filename
)
7487 strcpy(bs
->exact_filename
, primary_child_bs
->exact_filename
);
7491 if (bs
->exact_filename
[0]) {
7492 pstrcpy(bs
->filename
, sizeof(bs
->filename
), bs
->exact_filename
);
7494 GString
*json
= qobject_to_json(QOBJECT(bs
->full_open_options
));
7495 if (snprintf(bs
->filename
, sizeof(bs
->filename
), "json:%s",
7496 json
->str
) >= sizeof(bs
->filename
)) {
7497 /* Give user a hint if we truncated things. */
7498 strcpy(bs
->filename
+ sizeof(bs
->filename
) - 4, "...");
7500 g_string_free(json
, true);
7504 char *bdrv_dirname(BlockDriverState
*bs
, Error
**errp
)
7506 BlockDriver
*drv
= bs
->drv
;
7507 BlockDriverState
*child_bs
;
7510 error_setg(errp
, "Node '%s' is ejected", bs
->node_name
);
7514 if (drv
->bdrv_dirname
) {
7515 return drv
->bdrv_dirname(bs
, errp
);
7518 child_bs
= bdrv_primary_bs(bs
);
7520 return bdrv_dirname(child_bs
, errp
);
7523 bdrv_refresh_filename(bs
);
7524 if (bs
->exact_filename
[0] != '\0') {
7525 return path_combine(bs
->exact_filename
, "");
7528 error_setg(errp
, "Cannot generate a base directory for %s nodes",
7534 * Hot add/remove a BDS's child. So the user can take a child offline when
7535 * it is broken and take a new child online
7537 void bdrv_add_child(BlockDriverState
*parent_bs
, BlockDriverState
*child_bs
,
7541 if (!parent_bs
->drv
|| !parent_bs
->drv
->bdrv_add_child
) {
7542 error_setg(errp
, "The node %s does not support adding a child",
7543 bdrv_get_device_or_node_name(parent_bs
));
7547 if (!QLIST_EMPTY(&child_bs
->parents
)) {
7548 error_setg(errp
, "The node %s already has a parent",
7549 child_bs
->node_name
);
7553 parent_bs
->drv
->bdrv_add_child(parent_bs
, child_bs
, errp
);
7556 void bdrv_del_child(BlockDriverState
*parent_bs
, BdrvChild
*child
, Error
**errp
)
7560 if (!parent_bs
->drv
|| !parent_bs
->drv
->bdrv_del_child
) {
7561 error_setg(errp
, "The node %s does not support removing a child",
7562 bdrv_get_device_or_node_name(parent_bs
));
7566 QLIST_FOREACH(tmp
, &parent_bs
->children
, next
) {
7573 error_setg(errp
, "The node %s does not have a child named %s",
7574 bdrv_get_device_or_node_name(parent_bs
),
7575 bdrv_get_device_or_node_name(child
->bs
));
7579 parent_bs
->drv
->bdrv_del_child(parent_bs
, child
, errp
);
7582 int bdrv_make_empty(BdrvChild
*c
, Error
**errp
)
7584 BlockDriver
*drv
= c
->bs
->drv
;
7587 assert(c
->perm
& (BLK_PERM_WRITE
| BLK_PERM_WRITE_UNCHANGED
));
7589 if (!drv
->bdrv_make_empty
) {
7590 error_setg(errp
, "%s does not support emptying nodes",
7595 ret
= drv
->bdrv_make_empty(c
->bs
);
7597 error_setg_errno(errp
, -ret
, "Failed to empty %s",
7606 * Return the child that @bs acts as an overlay for, and from which data may be
7607 * copied in COW or COR operations. Usually this is the backing file.
7609 BdrvChild
*bdrv_cow_child(BlockDriverState
*bs
)
7611 if (!bs
|| !bs
->drv
) {
7615 if (bs
->drv
->is_filter
) {
7623 assert(bs
->backing
->role
& BDRV_CHILD_COW
);
7628 * If @bs acts as a filter for exactly one of its children, return
7631 BdrvChild
*bdrv_filter_child(BlockDriverState
*bs
)
7635 if (!bs
|| !bs
->drv
) {
7639 if (!bs
->drv
->is_filter
) {
7643 /* Only one of @backing or @file may be used */
7644 assert(!(bs
->backing
&& bs
->file
));
7646 c
= bs
->backing
?: bs
->file
;
7651 assert(c
->role
& BDRV_CHILD_FILTERED
);
7656 * Return either the result of bdrv_cow_child() or bdrv_filter_child(),
7657 * whichever is non-NULL.
7659 * Return NULL if both are NULL.
7661 BdrvChild
*bdrv_filter_or_cow_child(BlockDriverState
*bs
)
7663 BdrvChild
*cow_child
= bdrv_cow_child(bs
);
7664 BdrvChild
*filter_child
= bdrv_filter_child(bs
);
7666 /* Filter nodes cannot have COW backing files */
7667 assert(!(cow_child
&& filter_child
));
7669 return cow_child
?: filter_child
;
7673 * Return the primary child of this node: For filters, that is the
7674 * filtered child. For other nodes, that is usually the child storing
7676 * (A generally more helpful description is that this is (usually) the
7677 * child that has the same filename as @bs.)
7679 * Drivers do not necessarily have a primary child; for example quorum
7682 BdrvChild
*bdrv_primary_child(BlockDriverState
*bs
)
7684 BdrvChild
*c
, *found
= NULL
;
7686 QLIST_FOREACH(c
, &bs
->children
, next
) {
7687 if (c
->role
& BDRV_CHILD_PRIMARY
) {
7696 static BlockDriverState
*bdrv_do_skip_filters(BlockDriverState
*bs
,
7697 bool stop_on_explicit_filter
)
7705 while (!(stop_on_explicit_filter
&& !bs
->implicit
)) {
7706 c
= bdrv_filter_child(bs
);
7709 * A filter that is embedded in a working block graph must
7710 * have a child. Assert this here so this function does
7711 * not return a filter node that is not expected by the
7714 assert(!bs
->drv
|| !bs
->drv
->is_filter
);
7720 * Note that this treats nodes with bs->drv == NULL as not being
7721 * filters (bs->drv == NULL should be replaced by something else
7723 * The advantage of this behavior is that this function will thus
7724 * always return a non-NULL value (given a non-NULL @bs).
7731 * Return the first BDS that has not been added implicitly or that
7732 * does not have a filtered child down the chain starting from @bs
7733 * (including @bs itself).
7735 BlockDriverState
*bdrv_skip_implicit_filters(BlockDriverState
*bs
)
7737 return bdrv_do_skip_filters(bs
, true);
7741 * Return the first BDS that does not have a filtered child down the
7742 * chain starting from @bs (including @bs itself).
7744 BlockDriverState
*bdrv_skip_filters(BlockDriverState
*bs
)
7746 return bdrv_do_skip_filters(bs
, false);
7750 * For a backing chain, return the first non-filter backing image of
7751 * the first non-filter image.
7753 BlockDriverState
*bdrv_backing_chain_next(BlockDriverState
*bs
)
7755 return bdrv_skip_filters(bdrv_cow_bs(bdrv_skip_filters(bs
)));
7759 * Check whether [offset, offset + bytes) overlaps with the cached
7760 * block-status data region.
7762 * If so, and @pnum is not NULL, set *pnum to `bsc.data_end - offset`,
7763 * which is what bdrv_bsc_is_data()'s interface needs.
7764 * Otherwise, *pnum is not touched.
7766 static bool bdrv_bsc_range_overlaps_locked(BlockDriverState
*bs
,
7767 int64_t offset
, int64_t bytes
,
7770 BdrvBlockStatusCache
*bsc
= qatomic_rcu_read(&bs
->block_status_cache
);
7774 qatomic_read(&bsc
->valid
) &&
7775 ranges_overlap(offset
, bytes
, bsc
->data_start
,
7776 bsc
->data_end
- bsc
->data_start
);
7778 if (overlaps
&& pnum
) {
7779 *pnum
= bsc
->data_end
- offset
;
7786 * See block_int.h for this function's documentation.
7788 bool bdrv_bsc_is_data(BlockDriverState
*bs
, int64_t offset
, int64_t *pnum
)
7790 RCU_READ_LOCK_GUARD();
7792 return bdrv_bsc_range_overlaps_locked(bs
, offset
, 1, pnum
);
7796 * See block_int.h for this function's documentation.
7798 void bdrv_bsc_invalidate_range(BlockDriverState
*bs
,
7799 int64_t offset
, int64_t bytes
)
7801 RCU_READ_LOCK_GUARD();
7803 if (bdrv_bsc_range_overlaps_locked(bs
, offset
, bytes
, NULL
)) {
7804 qatomic_set(&bs
->block_status_cache
->valid
, false);
7809 * See block_int.h for this function's documentation.
7811 void bdrv_bsc_fill(BlockDriverState
*bs
, int64_t offset
, int64_t bytes
)
7813 BdrvBlockStatusCache
*new_bsc
= g_new(BdrvBlockStatusCache
, 1);
7814 BdrvBlockStatusCache
*old_bsc
;
7816 *new_bsc
= (BdrvBlockStatusCache
) {
7818 .data_start
= offset
,
7819 .data_end
= offset
+ bytes
,
7822 QEMU_LOCK_GUARD(&bs
->bsc_modify_lock
);
7824 old_bsc
= qatomic_rcu_read(&bs
->block_status_cache
);
7825 qatomic_rcu_set(&bs
->block_status_cache
, new_bsc
);
7827 g_free_rcu(old_bsc
, rcu
);