2 * QEMU System Emulator block driver
4 * Copyright (c) 2003 Fabrice Bellard
5 * Copyright (c) 2020 Virtuozzo International GmbH.
7 * Permission is hereby granted, free of charge, to any person obtaining a copy
8 * of this software and associated documentation files (the "Software"), to deal
9 * in the Software without restriction, including without limitation the rights
10 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
11 * copies of the Software, and to permit persons to whom the Software is
12 * furnished to do so, subject to the following conditions:
14 * The above copyright notice and this permission notice shall be included in
15 * all copies or substantial portions of the Software.
17 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
18 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
19 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
20 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
21 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
22 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
26 #include "qemu/osdep.h"
27 #include "block/trace.h"
28 #include "block/block_int.h"
29 #include "block/blockjob.h"
30 #include "block/dirty-bitmap.h"
31 #include "block/fuse.h"
32 #include "block/nbd.h"
33 #include "block/qdict.h"
34 #include "qemu/error-report.h"
35 #include "block/module_block.h"
36 #include "qemu/main-loop.h"
37 #include "qemu/module.h"
38 #include "qapi/error.h"
39 #include "qapi/qmp/qdict.h"
40 #include "qapi/qmp/qjson.h"
41 #include "qapi/qmp/qnull.h"
42 #include "qapi/qmp/qstring.h"
43 #include "qapi/qobject-output-visitor.h"
44 #include "qapi/qapi-visit-block-core.h"
45 #include "sysemu/block-backend.h"
46 #include "qemu/notify.h"
47 #include "qemu/option.h"
48 #include "qemu/coroutine.h"
49 #include "block/qapi.h"
50 #include "qemu/timer.h"
51 #include "qemu/cutils.h"
53 #include "qemu/range.h"
55 #include "block/coroutines.h"
58 #include <sys/ioctl.h>
59 #include <sys/queue.h>
60 #if defined(HAVE_SYS_DISK_H)
69 #define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */
71 /* Protected by BQL */
72 static QTAILQ_HEAD(, BlockDriverState
) graph_bdrv_states
=
73 QTAILQ_HEAD_INITIALIZER(graph_bdrv_states
);
75 /* Protected by BQL */
76 static QTAILQ_HEAD(, BlockDriverState
) all_bdrv_states
=
77 QTAILQ_HEAD_INITIALIZER(all_bdrv_states
);
79 /* Protected by BQL */
80 static QLIST_HEAD(, BlockDriver
) bdrv_drivers
=
81 QLIST_HEAD_INITIALIZER(bdrv_drivers
);
83 static BlockDriverState
*bdrv_open_inherit(const char *filename
,
84 const char *reference
,
85 QDict
*options
, int flags
,
86 BlockDriverState
*parent
,
87 const BdrvChildClass
*child_class
,
88 BdrvChildRole child_role
,
91 static bool bdrv_recurse_has_child(BlockDriverState
*bs
,
92 BlockDriverState
*child
);
94 static void GRAPH_WRLOCK
95 bdrv_replace_child_noperm(BdrvChild
*child
, BlockDriverState
*new_bs
);
97 static void GRAPH_WRLOCK
98 bdrv_remove_child(BdrvChild
*child
, Transaction
*tran
);
100 static int bdrv_reopen_prepare(BDRVReopenState
*reopen_state
,
101 BlockReopenQueue
*queue
,
102 Transaction
*change_child_tran
, Error
**errp
);
103 static void bdrv_reopen_commit(BDRVReopenState
*reopen_state
);
104 static void bdrv_reopen_abort(BDRVReopenState
*reopen_state
);
106 static bool bdrv_backing_overridden(BlockDriverState
*bs
);
108 static bool bdrv_change_aio_context(BlockDriverState
*bs
, AioContext
*ctx
,
109 GHashTable
*visited
, Transaction
*tran
,
112 /* If non-zero, use only whitelisted block drivers */
113 static int use_bdrv_whitelist
;
116 static int is_windows_drive_prefix(const char *filename
)
118 return (((filename
[0] >= 'a' && filename
[0] <= 'z') ||
119 (filename
[0] >= 'A' && filename
[0] <= 'Z')) &&
123 int is_windows_drive(const char *filename
)
125 if (is_windows_drive_prefix(filename
) &&
128 if (strstart(filename
, "\\\\.\\", NULL
) ||
129 strstart(filename
, "//./", NULL
))
135 size_t bdrv_opt_mem_align(BlockDriverState
*bs
)
137 if (!bs
|| !bs
->drv
) {
138 /* page size or 4k (hdd sector size) should be on the safe side */
139 return MAX(4096, qemu_real_host_page_size());
143 return bs
->bl
.opt_mem_alignment
;
146 size_t bdrv_min_mem_align(BlockDriverState
*bs
)
148 if (!bs
|| !bs
->drv
) {
149 /* page size or 4k (hdd sector size) should be on the safe side */
150 return MAX(4096, qemu_real_host_page_size());
154 return bs
->bl
.min_mem_alignment
;
157 /* check if the path starts with "<protocol>:" */
158 int path_has_protocol(const char *path
)
163 if (is_windows_drive(path
) ||
164 is_windows_drive_prefix(path
)) {
167 p
= path
+ strcspn(path
, ":/\\");
169 p
= path
+ strcspn(path
, ":/");
175 int path_is_absolute(const char *path
)
178 /* specific case for names like: "\\.\d:" */
179 if (is_windows_drive(path
) || is_windows_drive_prefix(path
)) {
182 return (*path
== '/' || *path
== '\\');
184 return (*path
== '/');
188 /* if filename is absolute, just return its duplicate. Otherwise, build a
189 path to it by considering it is relative to base_path. URL are
191 char *path_combine(const char *base_path
, const char *filename
)
193 const char *protocol_stripped
= NULL
;
198 if (path_is_absolute(filename
)) {
199 return g_strdup(filename
);
202 if (path_has_protocol(base_path
)) {
203 protocol_stripped
= strchr(base_path
, ':');
204 if (protocol_stripped
) {
208 p
= protocol_stripped
?: base_path
;
210 p1
= strrchr(base_path
, '/');
214 p2
= strrchr(base_path
, '\\');
215 if (!p1
|| p2
> p1
) {
230 result
= g_malloc(len
+ strlen(filename
) + 1);
231 memcpy(result
, base_path
, len
);
232 strcpy(result
+ len
, filename
);
238 * Helper function for bdrv_parse_filename() implementations to remove optional
239 * protocol prefixes (especially "file:") from a filename and for putting the
240 * stripped filename into the options QDict if there is such a prefix.
242 void bdrv_parse_filename_strip_prefix(const char *filename
, const char *prefix
,
245 if (strstart(filename
, prefix
, &filename
)) {
246 /* Stripping the explicit protocol prefix may result in a protocol
247 * prefix being (wrongly) detected (if the filename contains a colon) */
248 if (path_has_protocol(filename
)) {
249 GString
*fat_filename
;
251 /* This means there is some colon before the first slash; therefore,
252 * this cannot be an absolute path */
253 assert(!path_is_absolute(filename
));
255 /* And we can thus fix the protocol detection issue by prefixing it
257 fat_filename
= g_string_new("./");
258 g_string_append(fat_filename
, filename
);
260 assert(!path_has_protocol(fat_filename
->str
));
262 qdict_put(options
, "filename",
263 qstring_from_gstring(fat_filename
));
265 /* If no protocol prefix was detected, we can use the shortened
267 qdict_put_str(options
, "filename", filename
);
273 /* Returns whether the image file is opened as read-only. Note that this can
274 * return false and writing to the image file is still not possible because the
275 * image is inactivated. */
276 bool bdrv_is_read_only(BlockDriverState
*bs
)
279 return !(bs
->open_flags
& BDRV_O_RDWR
);
282 static int GRAPH_RDLOCK
283 bdrv_can_set_read_only(BlockDriverState
*bs
, bool read_only
,
284 bool ignore_allow_rdw
, Error
**errp
)
288 /* Do not set read_only if copy_on_read is enabled */
289 if (bs
->copy_on_read
&& read_only
) {
290 error_setg(errp
, "Can't set node '%s' to r/o with copy-on-read enabled",
291 bdrv_get_device_or_node_name(bs
));
295 /* Do not clear read_only if it is prohibited */
296 if (!read_only
&& !(bs
->open_flags
& BDRV_O_ALLOW_RDWR
) &&
299 error_setg(errp
, "Node '%s' is read only",
300 bdrv_get_device_or_node_name(bs
));
308 * Called by a driver that can only provide a read-only image.
310 * Returns 0 if the node is already read-only or it could switch the node to
311 * read-only because BDRV_O_AUTO_RDONLY is set.
313 * Returns -EACCES if the node is read-write and BDRV_O_AUTO_RDONLY is not set
314 * or bdrv_can_set_read_only() forbids making the node read-only. If @errmsg
315 * is not NULL, it is used as the error message for the Error object.
317 int bdrv_apply_auto_read_only(BlockDriverState
*bs
, const char *errmsg
,
323 assume_graph_lock(); /* FIXME */
325 if (!(bs
->open_flags
& BDRV_O_RDWR
)) {
328 if (!(bs
->open_flags
& BDRV_O_AUTO_RDONLY
)) {
332 ret
= bdrv_can_set_read_only(bs
, true, false, NULL
);
337 bs
->open_flags
&= ~BDRV_O_RDWR
;
342 error_setg(errp
, "%s", errmsg
?: "Image is read-only");
347 * If @backing is empty, this function returns NULL without setting
348 * @errp. In all other cases, NULL will only be returned with @errp
351 * Therefore, a return value of NULL without @errp set means that
352 * there is no backing file; if @errp is set, there is one but its
353 * absolute filename cannot be generated.
355 char *bdrv_get_full_backing_filename_from_filename(const char *backed
,
359 if (backing
[0] == '\0') {
361 } else if (path_has_protocol(backing
) || path_is_absolute(backing
)) {
362 return g_strdup(backing
);
363 } else if (backed
[0] == '\0' || strstart(backed
, "json:", NULL
)) {
364 error_setg(errp
, "Cannot use relative backing file names for '%s'",
368 return path_combine(backed
, backing
);
373 * If @filename is empty or NULL, this function returns NULL without
374 * setting @errp. In all other cases, NULL will only be returned with
377 static char * GRAPH_RDLOCK
378 bdrv_make_absolute_filename(BlockDriverState
*relative_to
,
379 const char *filename
, Error
**errp
)
381 char *dir
, *full_name
;
383 if (!filename
|| filename
[0] == '\0') {
385 } else if (path_has_protocol(filename
) || path_is_absolute(filename
)) {
386 return g_strdup(filename
);
389 dir
= bdrv_dirname(relative_to
, errp
);
394 full_name
= g_strconcat(dir
, filename
, NULL
);
399 char *bdrv_get_full_backing_filename(BlockDriverState
*bs
, Error
**errp
)
402 return bdrv_make_absolute_filename(bs
, bs
->backing_file
, errp
);
405 void bdrv_register(BlockDriver
*bdrv
)
407 assert(bdrv
->format_name
);
409 QLIST_INSERT_HEAD(&bdrv_drivers
, bdrv
, list
);
412 BlockDriverState
*bdrv_new(void)
414 BlockDriverState
*bs
;
419 bs
= g_new0(BlockDriverState
, 1);
420 QLIST_INIT(&bs
->dirty_bitmaps
);
421 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
422 QLIST_INIT(&bs
->op_blockers
[i
]);
424 qemu_mutex_init(&bs
->reqs_lock
);
425 qemu_mutex_init(&bs
->dirty_bitmap_mutex
);
427 bs
->aio_context
= qemu_get_aio_context();
429 qemu_co_queue_init(&bs
->flush_queue
);
431 qemu_co_mutex_init(&bs
->bsc_modify_lock
);
432 bs
->block_status_cache
= g_new0(BdrvBlockStatusCache
, 1);
434 for (i
= 0; i
< bdrv_drain_all_count
; i
++) {
435 bdrv_drained_begin(bs
);
438 QTAILQ_INSERT_TAIL(&all_bdrv_states
, bs
, bs_list
);
443 static BlockDriver
*bdrv_do_find_format(const char *format_name
)
448 QLIST_FOREACH(drv1
, &bdrv_drivers
, list
) {
449 if (!strcmp(drv1
->format_name
, format_name
)) {
457 BlockDriver
*bdrv_find_format(const char *format_name
)
464 drv1
= bdrv_do_find_format(format_name
);
469 /* The driver isn't registered, maybe we need to load a module */
470 for (i
= 0; i
< (int)ARRAY_SIZE(block_driver_modules
); ++i
) {
471 if (!strcmp(block_driver_modules
[i
].format_name
, format_name
)) {
472 Error
*local_err
= NULL
;
473 int rv
= block_module_load(block_driver_modules
[i
].library_name
,
476 return bdrv_do_find_format(format_name
);
478 error_report_err(local_err
);
486 static int bdrv_format_is_whitelisted(const char *format_name
, bool read_only
)
488 static const char *whitelist_rw
[] = {
489 CONFIG_BDRV_RW_WHITELIST
492 static const char *whitelist_ro
[] = {
493 CONFIG_BDRV_RO_WHITELIST
498 if (!whitelist_rw
[0] && !whitelist_ro
[0]) {
499 return 1; /* no whitelist, anything goes */
502 for (p
= whitelist_rw
; *p
; p
++) {
503 if (!strcmp(format_name
, *p
)) {
508 for (p
= whitelist_ro
; *p
; p
++) {
509 if (!strcmp(format_name
, *p
)) {
517 int bdrv_is_whitelisted(BlockDriver
*drv
, bool read_only
)
520 return bdrv_format_is_whitelisted(drv
->format_name
, read_only
);
523 bool bdrv_uses_whitelist(void)
525 return use_bdrv_whitelist
;
528 typedef struct CreateCo
{
536 int coroutine_fn
bdrv_co_create(BlockDriver
*drv
, const char *filename
,
537 QemuOpts
*opts
, Error
**errp
)
543 if (!drv
->bdrv_co_create_opts
) {
544 error_setg(errp
, "Driver '%s' does not support image creation",
549 ret
= drv
->bdrv_co_create_opts(drv
, filename
, opts
, errp
);
550 if (ret
< 0 && !*errp
) {
551 error_setg_errno(errp
, -ret
, "Could not create image");
558 * Helper function for bdrv_create_file_fallback(): Resize @blk to at
559 * least the given @minimum_size.
561 * On success, return @blk's actual length.
562 * Otherwise, return -errno.
564 static int64_t coroutine_fn GRAPH_UNLOCKED
565 create_file_fallback_truncate(BlockBackend
*blk
, int64_t minimum_size
,
568 Error
*local_err
= NULL
;
574 ret
= blk_co_truncate(blk
, minimum_size
, false, PREALLOC_MODE_OFF
, 0,
576 if (ret
< 0 && ret
!= -ENOTSUP
) {
577 error_propagate(errp
, local_err
);
581 size
= blk_co_getlength(blk
);
583 error_free(local_err
);
584 error_setg_errno(errp
, -size
,
585 "Failed to inquire the new image file's length");
589 if (size
< minimum_size
) {
590 /* Need to grow the image, but we failed to do that */
591 error_propagate(errp
, local_err
);
595 error_free(local_err
);
602 * Helper function for bdrv_create_file_fallback(): Zero the first
603 * sector to remove any potentially pre-existing image header.
605 static int coroutine_fn
606 create_file_fallback_zero_first_sector(BlockBackend
*blk
,
607 int64_t current_size
,
610 int64_t bytes_to_clear
;
615 bytes_to_clear
= MIN(current_size
, BDRV_SECTOR_SIZE
);
616 if (bytes_to_clear
) {
617 ret
= blk_co_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
;
648 size
= qemu_opt_get_size_del(opts
, BLOCK_OPT_SIZE
, 0);
649 buf
= qemu_opt_get_del(opts
, BLOCK_OPT_PREALLOC
);
650 prealloc
= qapi_enum_parse(&PreallocMode_lookup
, buf
,
651 PREALLOC_MODE_OFF
, &local_err
);
654 error_propagate(errp
, local_err
);
658 if (prealloc
!= PREALLOC_MODE_OFF
) {
659 error_setg(errp
, "Unsupported preallocation mode '%s'",
660 PreallocMode_str(prealloc
));
664 options
= qdict_new();
665 qdict_put_str(options
, "driver", drv
->format_name
);
667 blk
= blk_co_new_open(filename
, NULL
, options
,
668 BDRV_O_RDWR
| BDRV_O_RESIZE
, errp
);
670 error_prepend(errp
, "Protocol driver '%s' does not support creating "
671 "new images, so an existing image must be selected as "
672 "the target; however, opening the given target as an "
673 "existing image failed: ",
678 size
= create_file_fallback_truncate(blk
, size
, errp
);
684 ret
= create_file_fallback_zero_first_sector(blk
, size
, errp
);
695 int coroutine_fn
bdrv_co_create_file(const char *filename
, QemuOpts
*opts
,
698 QemuOpts
*protocol_opts
;
705 drv
= bdrv_find_protocol(filename
, true, errp
);
710 if (!drv
->create_opts
) {
711 error_setg(errp
, "Driver '%s' does not support image creation",
717 * 'opts' contains a QemuOptsList with a combination of format and protocol
720 * The format properly removes its options, but the default values remain
721 * in 'opts->list'. So if the protocol has options with the same name
722 * (e.g. rbd has 'cluster_size' as qcow2), it will see the default values
723 * of the format, since for overlapping options, the format wins.
725 * To avoid this issue, lets convert QemuOpts to QDict, in this way we take
726 * only the set options, and then convert it back to QemuOpts, using the
727 * create_opts of the protocol. So the new QemuOpts, will contain only the
730 qdict
= qemu_opts_to_qdict(opts
, NULL
);
731 protocol_opts
= qemu_opts_from_qdict(drv
->create_opts
, qdict
, errp
);
732 if (protocol_opts
== NULL
) {
737 ret
= bdrv_co_create(drv
, filename
, protocol_opts
, errp
);
739 qemu_opts_del(protocol_opts
);
740 qobject_unref(qdict
);
744 int coroutine_fn
bdrv_co_delete_file(BlockDriverState
*bs
, Error
**errp
)
746 Error
*local_err
= NULL
;
751 assert_bdrv_graph_readable();
754 error_setg(errp
, "Block node '%s' is not opened", bs
->filename
);
758 if (!bs
->drv
->bdrv_co_delete_file
) {
759 error_setg(errp
, "Driver '%s' does not support image deletion",
760 bs
->drv
->format_name
);
764 ret
= bs
->drv
->bdrv_co_delete_file(bs
, &local_err
);
766 error_propagate(errp
, local_err
);
772 void coroutine_fn
bdrv_co_delete_file_noerr(BlockDriverState
*bs
)
774 Error
*local_err
= NULL
;
782 ret
= bdrv_co_delete_file(bs
, &local_err
);
784 * ENOTSUP will happen if the block driver doesn't support
785 * the 'bdrv_co_delete_file' interface. This is a predictable
786 * scenario and shouldn't be reported back to the user.
788 if (ret
== -ENOTSUP
) {
789 error_free(local_err
);
790 } else if (ret
< 0) {
791 error_report_err(local_err
);
796 * Try to get @bs's logical and physical block size.
797 * On success, store them in @bsz struct and return 0.
798 * On failure return -errno.
799 * @bs must not be empty.
801 int bdrv_probe_blocksizes(BlockDriverState
*bs
, BlockSizes
*bsz
)
803 BlockDriver
*drv
= bs
->drv
;
804 BlockDriverState
*filtered
= bdrv_filter_bs(bs
);
807 if (drv
&& drv
->bdrv_probe_blocksizes
) {
808 return drv
->bdrv_probe_blocksizes(bs
, bsz
);
809 } else if (filtered
) {
810 return bdrv_probe_blocksizes(filtered
, bsz
);
817 * Try to get @bs's geometry (cyls, heads, sectors).
818 * On success, store them in @geo struct and return 0.
819 * On failure return -errno.
820 * @bs must not be empty.
822 int bdrv_probe_geometry(BlockDriverState
*bs
, HDGeometry
*geo
)
824 BlockDriver
*drv
= bs
->drv
;
825 BlockDriverState
*filtered
= bdrv_filter_bs(bs
);
828 if (drv
&& drv
->bdrv_probe_geometry
) {
829 return drv
->bdrv_probe_geometry(bs
, geo
);
830 } else if (filtered
) {
831 return bdrv_probe_geometry(filtered
, geo
);
838 * Create a uniquely-named empty temporary file.
839 * Return the actual file name used upon success, otherwise NULL.
840 * This string should be freed with g_free() when not needed any longer.
842 * Note: creating a temporary file for the caller to (re)open is
843 * inherently racy. Use g_file_open_tmp() instead whenever practical.
845 char *create_tmp_file(Error
**errp
)
849 g_autofree
char *filename
= NULL
;
851 tmpdir
= g_get_tmp_dir();
854 * See commit 69bef79 ("block: use /var/tmp instead of /tmp for -snapshot")
856 * This function is used to create temporary disk images (like -snapshot),
857 * so the files can become very large. /tmp is often a tmpfs where as
858 * /var/tmp is usually on a disk, so more appropriate for disk images.
860 if (!g_strcmp0(tmpdir
, "/tmp")) {
865 filename
= g_strdup_printf("%s/vl.XXXXXX", tmpdir
);
866 fd
= g_mkstemp(filename
);
868 error_setg_errno(errp
, errno
, "Could not open temporary file '%s'",
874 return g_steal_pointer(&filename
);
878 * Detect host devices. By convention, /dev/cdrom[N] is always
879 * recognized as a host CDROM.
881 static BlockDriver
*find_hdev_driver(const char *filename
)
883 int score_max
= 0, score
;
884 BlockDriver
*drv
= NULL
, *d
;
887 QLIST_FOREACH(d
, &bdrv_drivers
, list
) {
888 if (d
->bdrv_probe_device
) {
889 score
= d
->bdrv_probe_device(filename
);
890 if (score
> score_max
) {
900 static BlockDriver
*bdrv_do_find_protocol(const char *protocol
)
905 QLIST_FOREACH(drv1
, &bdrv_drivers
, list
) {
906 if (drv1
->protocol_name
&& !strcmp(drv1
->protocol_name
, protocol
)) {
914 BlockDriver
*bdrv_find_protocol(const char *filename
,
915 bool allow_protocol_prefix
,
925 /* TODO Drivers without bdrv_file_open must be specified explicitly */
928 * XXX(hch): we really should not let host device detection
929 * override an explicit protocol specification, but moving this
930 * later breaks access to device names with colons in them.
931 * Thanks to the brain-dead persistent naming schemes on udev-
932 * based Linux systems those actually are quite common.
934 drv1
= find_hdev_driver(filename
);
939 if (!path_has_protocol(filename
) || !allow_protocol_prefix
) {
943 p
= strchr(filename
, ':');
946 if (len
> sizeof(protocol
) - 1)
947 len
= sizeof(protocol
) - 1;
948 memcpy(protocol
, filename
, len
);
949 protocol
[len
] = '\0';
951 drv1
= bdrv_do_find_protocol(protocol
);
956 for (i
= 0; i
< (int)ARRAY_SIZE(block_driver_modules
); ++i
) {
957 if (block_driver_modules
[i
].protocol_name
&&
958 !strcmp(block_driver_modules
[i
].protocol_name
, protocol
)) {
959 int rv
= block_module_load(block_driver_modules
[i
].library_name
, errp
);
961 drv1
= bdrv_do_find_protocol(protocol
);
970 error_setg(errp
, "Unknown protocol '%s'", protocol
);
976 * Guess image format by probing its contents.
977 * This is not a good idea when your image is raw (CVE-2008-2004), but
978 * we do it anyway for backward compatibility.
980 * @buf contains the image's first @buf_size bytes.
981 * @buf_size is the buffer size in bytes (generally BLOCK_PROBE_BUF_SIZE,
982 * but can be smaller if the image file is smaller)
983 * @filename is its filename.
985 * For all block drivers, call the bdrv_probe() method to get its
987 * Return the first block driver with the highest probing score.
989 BlockDriver
*bdrv_probe_all(const uint8_t *buf
, int buf_size
,
990 const char *filename
)
992 int score_max
= 0, score
;
993 BlockDriver
*drv
= NULL
, *d
;
996 QLIST_FOREACH(d
, &bdrv_drivers
, list
) {
998 score
= d
->bdrv_probe(buf
, buf_size
, filename
);
999 if (score
> score_max
) {
1009 static int find_image_format(BlockBackend
*file
, const char *filename
,
1010 BlockDriver
**pdrv
, Error
**errp
)
1013 uint8_t buf
[BLOCK_PROBE_BUF_SIZE
];
1016 GLOBAL_STATE_CODE();
1018 /* Return the raw BlockDriver * to scsi-generic devices or empty drives */
1019 if (blk_is_sg(file
) || !blk_is_inserted(file
) || blk_getlength(file
) == 0) {
1024 ret
= blk_pread(file
, 0, sizeof(buf
), buf
, 0);
1026 error_setg_errno(errp
, -ret
, "Could not read image for determining its "
1032 drv
= bdrv_probe_all(buf
, sizeof(buf
), filename
);
1034 error_setg(errp
, "Could not determine image format: No compatible "
1045 * Set the current 'total_sectors' value
1046 * Return 0 on success, -errno on error.
1048 int coroutine_fn
bdrv_co_refresh_total_sectors(BlockDriverState
*bs
,
1051 BlockDriver
*drv
= bs
->drv
;
1053 assert_bdrv_graph_readable();
1059 /* Do not attempt drv->bdrv_co_getlength() on scsi-generic devices */
1063 /* query actual device if possible, otherwise just trust the hint */
1064 if (drv
->bdrv_co_getlength
) {
1065 int64_t length
= drv
->bdrv_co_getlength(bs
);
1069 hint
= DIV_ROUND_UP(length
, BDRV_SECTOR_SIZE
);
1072 bs
->total_sectors
= hint
;
1074 if (bs
->total_sectors
* BDRV_SECTOR_SIZE
> BDRV_MAX_LENGTH
) {
1082 * Combines a QDict of new block driver @options with any missing options taken
1083 * from @old_options, so that leaving out an option defaults to its old value.
1085 static void bdrv_join_options(BlockDriverState
*bs
, QDict
*options
,
1088 GLOBAL_STATE_CODE();
1089 if (bs
->drv
&& bs
->drv
->bdrv_join_options
) {
1090 bs
->drv
->bdrv_join_options(options
, old_options
);
1092 qdict_join(options
, old_options
, false);
1096 static BlockdevDetectZeroesOptions
bdrv_parse_detect_zeroes(QemuOpts
*opts
,
1100 Error
*local_err
= NULL
;
1101 char *value
= qemu_opt_get_del(opts
, "detect-zeroes");
1102 BlockdevDetectZeroesOptions detect_zeroes
=
1103 qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup
, value
,
1104 BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF
, &local_err
);
1105 GLOBAL_STATE_CODE();
1108 error_propagate(errp
, local_err
);
1109 return detect_zeroes
;
1112 if (detect_zeroes
== BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP
&&
1113 !(open_flags
& BDRV_O_UNMAP
))
1115 error_setg(errp
, "setting detect-zeroes to unmap is not allowed "
1116 "without setting discard operation to unmap");
1119 return detect_zeroes
;
1123 * Set open flags for aio engine
1125 * Return 0 on success, -1 if the engine specified is invalid
1127 int bdrv_parse_aio(const char *mode
, int *flags
)
1129 if (!strcmp(mode
, "threads")) {
1130 /* do nothing, default */
1131 } else if (!strcmp(mode
, "native")) {
1132 *flags
|= BDRV_O_NATIVE_AIO
;
1133 #ifdef CONFIG_LINUX_IO_URING
1134 } else if (!strcmp(mode
, "io_uring")) {
1135 *flags
|= BDRV_O_IO_URING
;
1145 * Set open flags for a given discard mode
1147 * Return 0 on success, -1 if the discard mode was invalid.
1149 int bdrv_parse_discard_flags(const char *mode
, int *flags
)
1151 *flags
&= ~BDRV_O_UNMAP
;
1153 if (!strcmp(mode
, "off") || !strcmp(mode
, "ignore")) {
1155 } else if (!strcmp(mode
, "on") || !strcmp(mode
, "unmap")) {
1156 *flags
|= BDRV_O_UNMAP
;
1165 * Set open flags for a given cache mode
1167 * Return 0 on success, -1 if the cache mode was invalid.
1169 int bdrv_parse_cache_mode(const char *mode
, int *flags
, bool *writethrough
)
1171 *flags
&= ~BDRV_O_CACHE_MASK
;
1173 if (!strcmp(mode
, "off") || !strcmp(mode
, "none")) {
1174 *writethrough
= false;
1175 *flags
|= BDRV_O_NOCACHE
;
1176 } else if (!strcmp(mode
, "directsync")) {
1177 *writethrough
= true;
1178 *flags
|= BDRV_O_NOCACHE
;
1179 } else if (!strcmp(mode
, "writeback")) {
1180 *writethrough
= false;
1181 } else if (!strcmp(mode
, "unsafe")) {
1182 *writethrough
= false;
1183 *flags
|= BDRV_O_NO_FLUSH
;
1184 } else if (!strcmp(mode
, "writethrough")) {
1185 *writethrough
= true;
1193 static char *bdrv_child_get_parent_desc(BdrvChild
*c
)
1195 BlockDriverState
*parent
= c
->opaque
;
1196 return g_strdup_printf("node '%s'", bdrv_get_node_name(parent
));
1199 static void GRAPH_RDLOCK
bdrv_child_cb_drained_begin(BdrvChild
*child
)
1201 BlockDriverState
*bs
= child
->opaque
;
1202 bdrv_do_drained_begin_quiesce(bs
, NULL
);
1205 static bool GRAPH_RDLOCK
bdrv_child_cb_drained_poll(BdrvChild
*child
)
1207 BlockDriverState
*bs
= child
->opaque
;
1208 return bdrv_drain_poll(bs
, NULL
, false);
1211 static void GRAPH_RDLOCK
bdrv_child_cb_drained_end(BdrvChild
*child
)
1213 BlockDriverState
*bs
= child
->opaque
;
1214 bdrv_drained_end(bs
);
1217 static int bdrv_child_cb_inactivate(BdrvChild
*child
)
1219 BlockDriverState
*bs
= child
->opaque
;
1220 GLOBAL_STATE_CODE();
1221 assert(bs
->open_flags
& BDRV_O_INACTIVE
);
1225 static bool bdrv_child_cb_change_aio_ctx(BdrvChild
*child
, AioContext
*ctx
,
1226 GHashTable
*visited
, Transaction
*tran
,
1229 BlockDriverState
*bs
= child
->opaque
;
1230 return bdrv_change_aio_context(bs
, ctx
, visited
, tran
, errp
);
1234 * Returns the options and flags that a temporary snapshot should get, based on
1235 * the originally requested flags (the originally requested image will have
1236 * flags like a backing file)
1238 static void bdrv_temp_snapshot_options(int *child_flags
, QDict
*child_options
,
1239 int parent_flags
, QDict
*parent_options
)
1241 GLOBAL_STATE_CODE();
1242 *child_flags
= (parent_flags
& ~BDRV_O_SNAPSHOT
) | BDRV_O_TEMPORARY
;
1244 /* For temporary files, unconditional cache=unsafe is fine */
1245 qdict_set_default_str(child_options
, BDRV_OPT_CACHE_DIRECT
, "off");
1246 qdict_set_default_str(child_options
, BDRV_OPT_CACHE_NO_FLUSH
, "on");
1248 /* Copy the read-only and discard options from the parent */
1249 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_READ_ONLY
);
1250 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_DISCARD
);
1252 /* aio=native doesn't work for cache.direct=off, so disable it for the
1253 * temporary snapshot */
1254 *child_flags
&= ~BDRV_O_NATIVE_AIO
;
1257 static void GRAPH_WRLOCK
bdrv_backing_attach(BdrvChild
*c
)
1259 BlockDriverState
*parent
= c
->opaque
;
1260 BlockDriverState
*backing_hd
= c
->bs
;
1262 GLOBAL_STATE_CODE();
1263 assert(!parent
->backing_blocker
);
1264 error_setg(&parent
->backing_blocker
,
1265 "node is used as backing hd of '%s'",
1266 bdrv_get_device_or_node_name(parent
));
1268 bdrv_refresh_filename(backing_hd
);
1270 parent
->open_flags
&= ~BDRV_O_NO_BACKING
;
1272 bdrv_op_block_all(backing_hd
, parent
->backing_blocker
);
1273 /* Otherwise we won't be able to commit or stream */
1274 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_COMMIT_TARGET
,
1275 parent
->backing_blocker
);
1276 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_STREAM
,
1277 parent
->backing_blocker
);
1279 * We do backup in 3 ways:
1281 * The target bs is new opened, and the source is top BDS
1282 * 2. blockdev backup
1283 * Both the source and the target are top BDSes.
1284 * 3. internal backup(used for block replication)
1285 * Both the source and the target are backing file
1287 * In case 1 and 2, neither the source nor the target is the backing file.
1288 * In case 3, we will block the top BDS, so there is only one block job
1289 * for the top BDS and its backing chain.
1291 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_BACKUP_SOURCE
,
1292 parent
->backing_blocker
);
1293 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_BACKUP_TARGET
,
1294 parent
->backing_blocker
);
1297 static void bdrv_backing_detach(BdrvChild
*c
)
1299 BlockDriverState
*parent
= c
->opaque
;
1301 GLOBAL_STATE_CODE();
1302 assert(parent
->backing_blocker
);
1303 bdrv_op_unblock_all(c
->bs
, parent
->backing_blocker
);
1304 error_free(parent
->backing_blocker
);
1305 parent
->backing_blocker
= NULL
;
1308 static int bdrv_backing_update_filename(BdrvChild
*c
, BlockDriverState
*base
,
1309 const char *filename
, Error
**errp
)
1311 BlockDriverState
*parent
= c
->opaque
;
1312 bool read_only
= bdrv_is_read_only(parent
);
1314 GLOBAL_STATE_CODE();
1317 ret
= bdrv_reopen_set_read_only(parent
, false, errp
);
1323 ret
= bdrv_change_backing_file(parent
, filename
,
1324 base
->drv
? base
->drv
->format_name
: "",
1327 error_setg_errno(errp
, -ret
, "Could not update backing file link");
1331 bdrv_reopen_set_read_only(parent
, true, NULL
);
1338 * Returns the options and flags that a generic child of a BDS should
1339 * get, based on the given options and flags for the parent BDS.
1341 static void bdrv_inherited_options(BdrvChildRole role
, bool parent_is_format
,
1342 int *child_flags
, QDict
*child_options
,
1343 int parent_flags
, QDict
*parent_options
)
1345 int flags
= parent_flags
;
1346 GLOBAL_STATE_CODE();
1349 * First, decide whether to set, clear, or leave BDRV_O_PROTOCOL.
1350 * Generally, the question to answer is: Should this child be
1351 * format-probed by default?
1355 * Pure and non-filtered data children of non-format nodes should
1356 * be probed by default (even when the node itself has BDRV_O_PROTOCOL
1357 * set). This only affects a very limited set of drivers (namely
1358 * quorum and blkverify when this comment was written).
1359 * Force-clear BDRV_O_PROTOCOL then.
1361 if (!parent_is_format
&&
1362 (role
& BDRV_CHILD_DATA
) &&
1363 !(role
& (BDRV_CHILD_METADATA
| BDRV_CHILD_FILTERED
)))
1365 flags
&= ~BDRV_O_PROTOCOL
;
1369 * All children of format nodes (except for COW children) and all
1370 * metadata children in general should never be format-probed.
1371 * Force-set BDRV_O_PROTOCOL then.
1373 if ((parent_is_format
&& !(role
& BDRV_CHILD_COW
)) ||
1374 (role
& BDRV_CHILD_METADATA
))
1376 flags
|= BDRV_O_PROTOCOL
;
1380 * If the cache mode isn't explicitly set, inherit direct and no-flush from
1383 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_CACHE_DIRECT
);
1384 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_CACHE_NO_FLUSH
);
1385 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_FORCE_SHARE
);
1387 if (role
& BDRV_CHILD_COW
) {
1388 /* backing files are opened read-only by default */
1389 qdict_set_default_str(child_options
, BDRV_OPT_READ_ONLY
, "on");
1390 qdict_set_default_str(child_options
, BDRV_OPT_AUTO_READ_ONLY
, "off");
1392 /* Inherit the read-only option from the parent if it's not set */
1393 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_READ_ONLY
);
1394 qdict_copy_default(child_options
, parent_options
,
1395 BDRV_OPT_AUTO_READ_ONLY
);
1399 * bdrv_co_pdiscard() respects unmap policy for the parent, so we
1400 * can default to enable it on lower layers regardless of the
1403 qdict_set_default_str(child_options
, BDRV_OPT_DISCARD
, "unmap");
1405 /* Clear flags that only apply to the top layer */
1406 flags
&= ~(BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
| BDRV_O_COPY_ON_READ
);
1408 if (role
& BDRV_CHILD_METADATA
) {
1409 flags
&= ~BDRV_O_NO_IO
;
1411 if (role
& BDRV_CHILD_COW
) {
1412 flags
&= ~BDRV_O_TEMPORARY
;
1415 *child_flags
= flags
;
1418 static void GRAPH_WRLOCK
bdrv_child_cb_attach(BdrvChild
*child
)
1420 BlockDriverState
*bs
= child
->opaque
;
1422 assert_bdrv_graph_writable();
1423 QLIST_INSERT_HEAD(&bs
->children
, child
, next
);
1424 if (bs
->drv
->is_filter
|| (child
->role
& BDRV_CHILD_FILTERED
)) {
1426 * Here we handle filters and block/raw-format.c when it behave like
1427 * filter. They generally have a single PRIMARY child, which is also the
1428 * FILTERED child, and that they may have multiple more children, which
1429 * are neither PRIMARY nor FILTERED. And never we have a COW child here.
1430 * So bs->file will be the PRIMARY child, unless the PRIMARY child goes
1431 * into bs->backing on exceptional cases; and bs->backing will be
1434 assert(!(child
->role
& BDRV_CHILD_COW
));
1435 if (child
->role
& BDRV_CHILD_PRIMARY
) {
1436 assert(child
->role
& BDRV_CHILD_FILTERED
);
1437 assert(!bs
->backing
);
1440 if (bs
->drv
->filtered_child_is_backing
) {
1441 bs
->backing
= child
;
1446 assert(!(child
->role
& BDRV_CHILD_FILTERED
));
1448 } else if (child
->role
& BDRV_CHILD_COW
) {
1449 assert(bs
->drv
->supports_backing
);
1450 assert(!(child
->role
& BDRV_CHILD_PRIMARY
));
1451 assert(!bs
->backing
);
1452 bs
->backing
= child
;
1453 bdrv_backing_attach(child
);
1454 } else if (child
->role
& BDRV_CHILD_PRIMARY
) {
1460 static void GRAPH_WRLOCK
bdrv_child_cb_detach(BdrvChild
*child
)
1462 BlockDriverState
*bs
= child
->opaque
;
1464 if (child
->role
& BDRV_CHILD_COW
) {
1465 bdrv_backing_detach(child
);
1468 assert_bdrv_graph_writable();
1469 QLIST_REMOVE(child
, next
);
1470 if (child
== bs
->backing
) {
1471 assert(child
!= bs
->file
);
1473 } else if (child
== bs
->file
) {
1478 static int bdrv_child_cb_update_filename(BdrvChild
*c
, BlockDriverState
*base
,
1479 const char *filename
, Error
**errp
)
1481 if (c
->role
& BDRV_CHILD_COW
) {
1482 return bdrv_backing_update_filename(c
, base
, filename
, errp
);
1487 AioContext
*child_of_bds_get_parent_aio_context(BdrvChild
*c
)
1489 BlockDriverState
*bs
= c
->opaque
;
1492 return bdrv_get_aio_context(bs
);
1495 const BdrvChildClass child_of_bds
= {
1496 .parent_is_bds
= true,
1497 .get_parent_desc
= bdrv_child_get_parent_desc
,
1498 .inherit_options
= bdrv_inherited_options
,
1499 .drained_begin
= bdrv_child_cb_drained_begin
,
1500 .drained_poll
= bdrv_child_cb_drained_poll
,
1501 .drained_end
= bdrv_child_cb_drained_end
,
1502 .attach
= bdrv_child_cb_attach
,
1503 .detach
= bdrv_child_cb_detach
,
1504 .inactivate
= bdrv_child_cb_inactivate
,
1505 .change_aio_ctx
= bdrv_child_cb_change_aio_ctx
,
1506 .update_filename
= bdrv_child_cb_update_filename
,
1507 .get_parent_aio_context
= child_of_bds_get_parent_aio_context
,
1510 AioContext
*bdrv_child_get_parent_aio_context(BdrvChild
*c
)
1513 return c
->klass
->get_parent_aio_context(c
);
1516 static int bdrv_open_flags(BlockDriverState
*bs
, int flags
)
1518 int open_flags
= flags
;
1519 GLOBAL_STATE_CODE();
1522 * Clear flags that are internal to the block layer before opening the
1525 open_flags
&= ~(BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
| BDRV_O_PROTOCOL
);
1530 static void update_flags_from_options(int *flags
, QemuOpts
*opts
)
1532 GLOBAL_STATE_CODE();
1534 *flags
&= ~(BDRV_O_CACHE_MASK
| BDRV_O_RDWR
| BDRV_O_AUTO_RDONLY
);
1536 if (qemu_opt_get_bool_del(opts
, BDRV_OPT_CACHE_NO_FLUSH
, false)) {
1537 *flags
|= BDRV_O_NO_FLUSH
;
1540 if (qemu_opt_get_bool_del(opts
, BDRV_OPT_CACHE_DIRECT
, false)) {
1541 *flags
|= BDRV_O_NOCACHE
;
1544 if (!qemu_opt_get_bool_del(opts
, BDRV_OPT_READ_ONLY
, false)) {
1545 *flags
|= BDRV_O_RDWR
;
1548 if (qemu_opt_get_bool_del(opts
, BDRV_OPT_AUTO_READ_ONLY
, false)) {
1549 *flags
|= BDRV_O_AUTO_RDONLY
;
1553 static void update_options_from_flags(QDict
*options
, int flags
)
1555 GLOBAL_STATE_CODE();
1556 if (!qdict_haskey(options
, BDRV_OPT_CACHE_DIRECT
)) {
1557 qdict_put_bool(options
, BDRV_OPT_CACHE_DIRECT
, flags
& BDRV_O_NOCACHE
);
1559 if (!qdict_haskey(options
, BDRV_OPT_CACHE_NO_FLUSH
)) {
1560 qdict_put_bool(options
, BDRV_OPT_CACHE_NO_FLUSH
,
1561 flags
& BDRV_O_NO_FLUSH
);
1563 if (!qdict_haskey(options
, BDRV_OPT_READ_ONLY
)) {
1564 qdict_put_bool(options
, BDRV_OPT_READ_ONLY
, !(flags
& BDRV_O_RDWR
));
1566 if (!qdict_haskey(options
, BDRV_OPT_AUTO_READ_ONLY
)) {
1567 qdict_put_bool(options
, BDRV_OPT_AUTO_READ_ONLY
,
1568 flags
& BDRV_O_AUTO_RDONLY
);
1572 static void bdrv_assign_node_name(BlockDriverState
*bs
,
1573 const char *node_name
,
1576 char *gen_node_name
= NULL
;
1577 GLOBAL_STATE_CODE();
1580 node_name
= gen_node_name
= id_generate(ID_BLOCK
);
1581 } else if (!id_wellformed(node_name
)) {
1583 * Check for empty string or invalid characters, but not if it is
1584 * generated (generated names use characters not available to the user)
1586 error_setg(errp
, "Invalid node-name: '%s'", node_name
);
1590 /* takes care of avoiding namespaces collisions */
1591 if (blk_by_name(node_name
)) {
1592 error_setg(errp
, "node-name=%s is conflicting with a device id",
1597 /* takes care of avoiding duplicates node names */
1598 if (bdrv_find_node(node_name
)) {
1599 error_setg(errp
, "Duplicate nodes with node-name='%s'", node_name
);
1603 /* Make sure that the node name isn't truncated */
1604 if (strlen(node_name
) >= sizeof(bs
->node_name
)) {
1605 error_setg(errp
, "Node name too long");
1609 /* copy node name into the bs and insert it into the graph list */
1610 pstrcpy(bs
->node_name
, sizeof(bs
->node_name
), node_name
);
1611 QTAILQ_INSERT_TAIL(&graph_bdrv_states
, bs
, node_list
);
1613 g_free(gen_node_name
);
1617 * The caller must always hold @bs AioContext lock, because this function calls
1618 * bdrv_refresh_total_sectors() which polls when called from non-coroutine
1621 static int no_coroutine_fn GRAPH_UNLOCKED
1622 bdrv_open_driver(BlockDriverState
*bs
, BlockDriver
*drv
, const char *node_name
,
1623 QDict
*options
, int open_flags
, Error
**errp
)
1626 Error
*local_err
= NULL
;
1628 GLOBAL_STATE_CODE();
1630 bdrv_assign_node_name(bs
, node_name
, &local_err
);
1632 error_propagate(errp
, local_err
);
1637 bs
->opaque
= g_malloc0(drv
->instance_size
);
1639 if (drv
->bdrv_file_open
) {
1640 assert(!drv
->bdrv_needs_filename
|| bs
->filename
[0]);
1641 ret
= drv
->bdrv_file_open(bs
, options
, open_flags
, &local_err
);
1642 } else if (drv
->bdrv_open
) {
1643 ret
= drv
->bdrv_open(bs
, options
, open_flags
, &local_err
);
1650 error_propagate(errp
, local_err
);
1651 } else if (bs
->filename
[0]) {
1652 error_setg_errno(errp
, -ret
, "Could not open '%s'", bs
->filename
);
1654 error_setg_errno(errp
, -ret
, "Could not open image");
1659 assert(!(bs
->supported_read_flags
& ~BDRV_REQ_MASK
));
1660 assert(!(bs
->supported_write_flags
& ~BDRV_REQ_MASK
));
1663 * Always allow the BDRV_REQ_REGISTERED_BUF optimization hint. This saves
1664 * drivers that pass read/write requests through to a child the trouble of
1665 * declaring support explicitly.
1667 * Drivers must not propagate this flag accidentally when they initiate I/O
1668 * to a bounce buffer. That case should be rare though.
1670 bs
->supported_read_flags
|= BDRV_REQ_REGISTERED_BUF
;
1671 bs
->supported_write_flags
|= BDRV_REQ_REGISTERED_BUF
;
1673 /* Get the context after .bdrv_open, it can change the context */
1674 ctx
= bdrv_get_aio_context(bs
);
1675 aio_context_acquire(ctx
);
1677 ret
= bdrv_refresh_total_sectors(bs
, bs
->total_sectors
);
1679 error_setg_errno(errp
, -ret
, "Could not refresh total sector count");
1680 aio_context_release(ctx
);
1684 bdrv_graph_rdlock_main_loop();
1685 bdrv_refresh_limits(bs
, NULL
, &local_err
);
1686 bdrv_graph_rdunlock_main_loop();
1687 aio_context_release(ctx
);
1690 error_propagate(errp
, local_err
);
1694 assert(bdrv_opt_mem_align(bs
) != 0);
1695 assert(bdrv_min_mem_align(bs
) != 0);
1696 assert(is_power_of_2(bs
->bl
.request_alignment
));
1698 for (i
= 0; i
< bs
->quiesce_counter
; i
++) {
1699 if (drv
->bdrv_drain_begin
) {
1700 drv
->bdrv_drain_begin(bs
);
1707 if (bs
->file
!= NULL
) {
1708 bdrv_graph_wrlock(NULL
);
1709 bdrv_unref_child(bs
, bs
->file
);
1710 bdrv_graph_wrunlock();
1719 * Create and open a block node.
1721 * @options is a QDict of options to pass to the block drivers, or NULL for an
1722 * empty set of options. The reference to the QDict belongs to the block layer
1723 * after the call (even on failure), so if the caller intends to reuse the
1724 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
1726 BlockDriverState
*bdrv_new_open_driver_opts(BlockDriver
*drv
,
1727 const char *node_name
,
1728 QDict
*options
, int flags
,
1731 BlockDriverState
*bs
;
1734 GLOBAL_STATE_CODE();
1737 bs
->open_flags
= flags
;
1738 bs
->options
= options
?: qdict_new();
1739 bs
->explicit_options
= qdict_clone_shallow(bs
->options
);
1742 update_options_from_flags(bs
->options
, flags
);
1744 ret
= bdrv_open_driver(bs
, drv
, node_name
, bs
->options
, flags
, errp
);
1746 qobject_unref(bs
->explicit_options
);
1747 bs
->explicit_options
= NULL
;
1748 qobject_unref(bs
->options
);
1757 /* Create and open a block node. */
1758 BlockDriverState
*bdrv_new_open_driver(BlockDriver
*drv
, const char *node_name
,
1759 int flags
, Error
**errp
)
1761 GLOBAL_STATE_CODE();
1762 return bdrv_new_open_driver_opts(drv
, node_name
, NULL
, flags
, errp
);
1765 QemuOptsList bdrv_runtime_opts
= {
1766 .name
= "bdrv_common",
1767 .head
= QTAILQ_HEAD_INITIALIZER(bdrv_runtime_opts
.head
),
1770 .name
= "node-name",
1771 .type
= QEMU_OPT_STRING
,
1772 .help
= "Node name of the block device node",
1776 .type
= QEMU_OPT_STRING
,
1777 .help
= "Block driver to use for the node",
1780 .name
= BDRV_OPT_CACHE_DIRECT
,
1781 .type
= QEMU_OPT_BOOL
,
1782 .help
= "Bypass software writeback cache on the host",
1785 .name
= BDRV_OPT_CACHE_NO_FLUSH
,
1786 .type
= QEMU_OPT_BOOL
,
1787 .help
= "Ignore flush requests",
1790 .name
= BDRV_OPT_READ_ONLY
,
1791 .type
= QEMU_OPT_BOOL
,
1792 .help
= "Node is opened in read-only mode",
1795 .name
= BDRV_OPT_AUTO_READ_ONLY
,
1796 .type
= QEMU_OPT_BOOL
,
1797 .help
= "Node can become read-only if opening read-write fails",
1800 .name
= "detect-zeroes",
1801 .type
= QEMU_OPT_STRING
,
1802 .help
= "try to optimize zero writes (off, on, unmap)",
1805 .name
= BDRV_OPT_DISCARD
,
1806 .type
= QEMU_OPT_STRING
,
1807 .help
= "discard operation (ignore/off, unmap/on)",
1810 .name
= BDRV_OPT_FORCE_SHARE
,
1811 .type
= QEMU_OPT_BOOL
,
1812 .help
= "always accept other writers (default: off)",
1814 { /* end of list */ }
1818 QemuOptsList bdrv_create_opts_simple
= {
1819 .name
= "simple-create-opts",
1820 .head
= QTAILQ_HEAD_INITIALIZER(bdrv_create_opts_simple
.head
),
1823 .name
= BLOCK_OPT_SIZE
,
1824 .type
= QEMU_OPT_SIZE
,
1825 .help
= "Virtual disk size"
1828 .name
= BLOCK_OPT_PREALLOC
,
1829 .type
= QEMU_OPT_STRING
,
1830 .help
= "Preallocation mode (allowed values: off)"
1832 { /* end of list */ }
1837 * Common part for opening disk images and files
1839 * Removes all processed options from *options.
1841 static int bdrv_open_common(BlockDriverState
*bs
, BlockBackend
*file
,
1842 QDict
*options
, Error
**errp
)
1844 int ret
, open_flags
;
1845 const char *filename
;
1846 const char *driver_name
= NULL
;
1847 const char *node_name
= NULL
;
1848 const char *discard
;
1851 Error
*local_err
= NULL
;
1854 assert(bs
->file
== NULL
);
1855 assert(options
!= NULL
&& bs
->options
!= options
);
1856 GLOBAL_STATE_CODE();
1858 opts
= qemu_opts_create(&bdrv_runtime_opts
, NULL
, 0, &error_abort
);
1859 if (!qemu_opts_absorb_qdict(opts
, options
, errp
)) {
1864 update_flags_from_options(&bs
->open_flags
, opts
);
1866 driver_name
= qemu_opt_get(opts
, "driver");
1867 drv
= bdrv_find_format(driver_name
);
1868 assert(drv
!= NULL
);
1870 bs
->force_share
= qemu_opt_get_bool(opts
, BDRV_OPT_FORCE_SHARE
, false);
1872 if (bs
->force_share
&& (bs
->open_flags
& BDRV_O_RDWR
)) {
1874 BDRV_OPT_FORCE_SHARE
1875 "=on can only be used with read-only images");
1881 bdrv_graph_rdlock_main_loop();
1882 bdrv_refresh_filename(blk_bs(file
));
1883 bdrv_graph_rdunlock_main_loop();
1885 filename
= blk_bs(file
)->filename
;
1888 * Caution: while qdict_get_try_str() is fine, getting
1889 * non-string types would require more care. When @options
1890 * come from -blockdev or blockdev_add, its members are typed
1891 * according to the QAPI schema, but when they come from
1892 * -drive, they're all QString.
1894 filename
= qdict_get_try_str(options
, "filename");
1897 if (drv
->bdrv_needs_filename
&& (!filename
|| !filename
[0])) {
1898 error_setg(errp
, "The '%s' block driver requires a file name",
1904 trace_bdrv_open_common(bs
, filename
?: "", bs
->open_flags
,
1907 ro
= bdrv_is_read_only(bs
);
1909 if (use_bdrv_whitelist
&& !bdrv_is_whitelisted(drv
, ro
)) {
1910 if (!ro
&& bdrv_is_whitelisted(drv
, true)) {
1911 ret
= bdrv_apply_auto_read_only(bs
, NULL
, NULL
);
1917 !ro
&& bdrv_is_whitelisted(drv
, true)
1918 ? "Driver '%s' can only be used for read-only devices"
1919 : "Driver '%s' is not whitelisted",
1925 /* bdrv_new() and bdrv_close() make it so */
1926 assert(qatomic_read(&bs
->copy_on_read
) == 0);
1928 if (bs
->open_flags
& BDRV_O_COPY_ON_READ
) {
1930 bdrv_enable_copy_on_read(bs
);
1932 error_setg(errp
, "Can't use copy-on-read on read-only device");
1938 discard
= qemu_opt_get(opts
, BDRV_OPT_DISCARD
);
1939 if (discard
!= NULL
) {
1940 if (bdrv_parse_discard_flags(discard
, &bs
->open_flags
) != 0) {
1941 error_setg(errp
, "Invalid discard option");
1948 bdrv_parse_detect_zeroes(opts
, bs
->open_flags
, &local_err
);
1950 error_propagate(errp
, local_err
);
1955 if (filename
!= NULL
) {
1956 pstrcpy(bs
->filename
, sizeof(bs
->filename
), filename
);
1958 bs
->filename
[0] = '\0';
1960 pstrcpy(bs
->exact_filename
, sizeof(bs
->exact_filename
), bs
->filename
);
1962 /* Open the image, either directly or using a protocol */
1963 open_flags
= bdrv_open_flags(bs
, bs
->open_flags
);
1964 node_name
= qemu_opt_get(opts
, "node-name");
1966 assert(!drv
->bdrv_file_open
|| file
== NULL
);
1967 ret
= bdrv_open_driver(bs
, drv
, node_name
, options
, open_flags
, errp
);
1972 qemu_opts_del(opts
);
1976 qemu_opts_del(opts
);
1980 static QDict
*parse_json_filename(const char *filename
, Error
**errp
)
1982 QObject
*options_obj
;
1985 GLOBAL_STATE_CODE();
1987 ret
= strstart(filename
, "json:", &filename
);
1990 options_obj
= qobject_from_json(filename
, errp
);
1992 error_prepend(errp
, "Could not parse the JSON options: ");
1996 options
= qobject_to(QDict
, options_obj
);
1998 qobject_unref(options_obj
);
1999 error_setg(errp
, "Invalid JSON object given");
2003 qdict_flatten(options
);
2008 static void parse_json_protocol(QDict
*options
, const char **pfilename
,
2011 QDict
*json_options
;
2012 Error
*local_err
= NULL
;
2013 GLOBAL_STATE_CODE();
2015 /* Parse json: pseudo-protocol */
2016 if (!*pfilename
|| !g_str_has_prefix(*pfilename
, "json:")) {
2020 json_options
= parse_json_filename(*pfilename
, &local_err
);
2022 error_propagate(errp
, local_err
);
2026 /* Options given in the filename have lower priority than options
2027 * specified directly */
2028 qdict_join(options
, json_options
, false);
2029 qobject_unref(json_options
);
2034 * Fills in default options for opening images and converts the legacy
2035 * filename/flags pair to option QDict entries.
2036 * The BDRV_O_PROTOCOL flag in *flags will be set or cleared accordingly if a
2037 * block driver has been specified explicitly.
2039 static int bdrv_fill_options(QDict
**options
, const char *filename
,
2040 int *flags
, Error
**errp
)
2042 const char *drvname
;
2043 bool protocol
= *flags
& BDRV_O_PROTOCOL
;
2044 bool parse_filename
= false;
2045 BlockDriver
*drv
= NULL
;
2046 Error
*local_err
= NULL
;
2048 GLOBAL_STATE_CODE();
2051 * Caution: while qdict_get_try_str() is fine, getting non-string
2052 * types would require more care. When @options come from
2053 * -blockdev or blockdev_add, its members are typed according to
2054 * the QAPI schema, but when they come from -drive, they're all
2057 drvname
= qdict_get_try_str(*options
, "driver");
2059 drv
= bdrv_find_format(drvname
);
2061 error_setg(errp
, "Unknown driver '%s'", drvname
);
2064 /* If the user has explicitly specified the driver, this choice should
2065 * override the BDRV_O_PROTOCOL flag */
2066 protocol
= drv
->bdrv_file_open
;
2070 *flags
|= BDRV_O_PROTOCOL
;
2072 *flags
&= ~BDRV_O_PROTOCOL
;
2075 /* Translate cache options from flags into options */
2076 update_options_from_flags(*options
, *flags
);
2078 /* Fetch the file name from the options QDict if necessary */
2079 if (protocol
&& filename
) {
2080 if (!qdict_haskey(*options
, "filename")) {
2081 qdict_put_str(*options
, "filename", filename
);
2082 parse_filename
= true;
2084 error_setg(errp
, "Can't specify 'file' and 'filename' options at "
2090 /* Find the right block driver */
2091 /* See cautionary note on accessing @options above */
2092 filename
= qdict_get_try_str(*options
, "filename");
2094 if (!drvname
&& protocol
) {
2096 drv
= bdrv_find_protocol(filename
, parse_filename
, errp
);
2101 drvname
= drv
->format_name
;
2102 qdict_put_str(*options
, "driver", drvname
);
2104 error_setg(errp
, "Must specify either driver or file");
2109 assert(drv
|| !protocol
);
2111 /* Driver-specific filename parsing */
2112 if (drv
&& drv
->bdrv_parse_filename
&& parse_filename
) {
2113 drv
->bdrv_parse_filename(filename
, *options
, &local_err
);
2115 error_propagate(errp
, local_err
);
2119 if (!drv
->bdrv_needs_filename
) {
2120 qdict_del(*options
, "filename");
2127 typedef struct BlockReopenQueueEntry
{
2129 BDRVReopenState state
;
2130 QTAILQ_ENTRY(BlockReopenQueueEntry
) entry
;
2131 } BlockReopenQueueEntry
;
2134 * Return the flags that @bs will have after the reopens in @q have
2135 * successfully completed. If @q is NULL (or @bs is not contained in @q),
2136 * return the current flags.
2138 static int bdrv_reopen_get_flags(BlockReopenQueue
*q
, BlockDriverState
*bs
)
2140 BlockReopenQueueEntry
*entry
;
2143 QTAILQ_FOREACH(entry
, q
, entry
) {
2144 if (entry
->state
.bs
== bs
) {
2145 return entry
->state
.flags
;
2150 return bs
->open_flags
;
2153 /* Returns whether the image file can be written to after the reopen queue @q
2154 * has been successfully applied, or right now if @q is NULL. */
2155 static bool bdrv_is_writable_after_reopen(BlockDriverState
*bs
,
2156 BlockReopenQueue
*q
)
2158 int flags
= bdrv_reopen_get_flags(q
, bs
);
2160 return (flags
& (BDRV_O_RDWR
| BDRV_O_INACTIVE
)) == BDRV_O_RDWR
;
2164 * Return whether the BDS can be written to. This is not necessarily
2165 * the same as !bdrv_is_read_only(bs), as inactivated images may not
2166 * be written to but do not count as read-only images.
2168 bool bdrv_is_writable(BlockDriverState
*bs
)
2171 return bdrv_is_writable_after_reopen(bs
, NULL
);
2174 static char *bdrv_child_user_desc(BdrvChild
*c
)
2176 GLOBAL_STATE_CODE();
2177 return c
->klass
->get_parent_desc(c
);
2181 * Check that @a allows everything that @b needs. @a and @b must reference same
2184 static bool bdrv_a_allow_b(BdrvChild
*a
, BdrvChild
*b
, Error
**errp
)
2186 const char *child_bs_name
;
2187 g_autofree
char *a_user
= NULL
;
2188 g_autofree
char *b_user
= NULL
;
2189 g_autofree
char *perms
= NULL
;
2192 assert(a
->bs
== b
->bs
);
2193 GLOBAL_STATE_CODE();
2195 if ((b
->perm
& a
->shared_perm
) == b
->perm
) {
2199 child_bs_name
= bdrv_get_node_name(b
->bs
);
2200 a_user
= bdrv_child_user_desc(a
);
2201 b_user
= bdrv_child_user_desc(b
);
2202 perms
= bdrv_perm_names(b
->perm
& ~a
->shared_perm
);
2204 error_setg(errp
, "Permission conflict on node '%s': permissions '%s' are "
2205 "both required by %s (uses node '%s' as '%s' child) and "
2206 "unshared by %s (uses node '%s' as '%s' child).",
2207 child_bs_name
, perms
,
2208 b_user
, child_bs_name
, b
->name
,
2209 a_user
, child_bs_name
, a
->name
);
2214 static bool GRAPH_RDLOCK
2215 bdrv_parent_perms_conflict(BlockDriverState
*bs
, Error
**errp
)
2218 GLOBAL_STATE_CODE();
2221 * During the loop we'll look at each pair twice. That's correct because
2222 * bdrv_a_allow_b() is asymmetric and we should check each pair in both
2225 QLIST_FOREACH(a
, &bs
->parents
, next_parent
) {
2226 QLIST_FOREACH(b
, &bs
->parents
, next_parent
) {
2231 if (!bdrv_a_allow_b(a
, b
, errp
)) {
2240 static void GRAPH_RDLOCK
2241 bdrv_child_perm(BlockDriverState
*bs
, BlockDriverState
*child_bs
,
2242 BdrvChild
*c
, BdrvChildRole role
,
2243 BlockReopenQueue
*reopen_queue
,
2244 uint64_t parent_perm
, uint64_t parent_shared
,
2245 uint64_t *nperm
, uint64_t *nshared
)
2247 assert(bs
->drv
&& bs
->drv
->bdrv_child_perm
);
2248 GLOBAL_STATE_CODE();
2249 bs
->drv
->bdrv_child_perm(bs
, c
, role
, reopen_queue
,
2250 parent_perm
, parent_shared
,
2252 /* TODO Take force_share from reopen_queue */
2253 if (child_bs
&& child_bs
->force_share
) {
2254 *nshared
= BLK_PERM_ALL
;
2259 * Adds the whole subtree of @bs (including @bs itself) to the @list (except for
2260 * nodes that are already in the @list, of course) so that final list is
2261 * topologically sorted. Return the result (GSList @list object is updated, so
2262 * don't use old reference after function call).
2264 * On function start @list must be already topologically sorted and for any node
2265 * in the @list the whole subtree of the node must be in the @list as well. The
2266 * simplest way to satisfy this criteria: use only result of
2267 * bdrv_topological_dfs() or NULL as @list parameter.
2269 static GSList
* GRAPH_RDLOCK
2270 bdrv_topological_dfs(GSList
*list
, GHashTable
*found
, BlockDriverState
*bs
)
2273 g_autoptr(GHashTable
) local_found
= NULL
;
2275 GLOBAL_STATE_CODE();
2279 found
= local_found
= g_hash_table_new(NULL
, NULL
);
2282 if (g_hash_table_contains(found
, bs
)) {
2285 g_hash_table_add(found
, bs
);
2287 QLIST_FOREACH(child
, &bs
->children
, next
) {
2288 list
= bdrv_topological_dfs(list
, found
, child
->bs
);
2291 return g_slist_prepend(list
, bs
);
2294 typedef struct BdrvChildSetPermState
{
2297 uint64_t old_shared_perm
;
2298 } BdrvChildSetPermState
;
2300 static void bdrv_child_set_perm_abort(void *opaque
)
2302 BdrvChildSetPermState
*s
= opaque
;
2304 GLOBAL_STATE_CODE();
2306 s
->child
->perm
= s
->old_perm
;
2307 s
->child
->shared_perm
= s
->old_shared_perm
;
2310 static TransactionActionDrv bdrv_child_set_pem_drv
= {
2311 .abort
= bdrv_child_set_perm_abort
,
2315 static void bdrv_child_set_perm(BdrvChild
*c
, uint64_t perm
,
2316 uint64_t shared
, Transaction
*tran
)
2318 BdrvChildSetPermState
*s
= g_new(BdrvChildSetPermState
, 1);
2319 GLOBAL_STATE_CODE();
2321 *s
= (BdrvChildSetPermState
) {
2323 .old_perm
= c
->perm
,
2324 .old_shared_perm
= c
->shared_perm
,
2328 c
->shared_perm
= shared
;
2330 tran_add(tran
, &bdrv_child_set_pem_drv
, s
);
2333 static void GRAPH_RDLOCK
bdrv_drv_set_perm_commit(void *opaque
)
2335 BlockDriverState
*bs
= opaque
;
2336 uint64_t cumulative_perms
, cumulative_shared_perms
;
2337 GLOBAL_STATE_CODE();
2339 if (bs
->drv
->bdrv_set_perm
) {
2340 bdrv_get_cumulative_perm(bs
, &cumulative_perms
,
2341 &cumulative_shared_perms
);
2342 bs
->drv
->bdrv_set_perm(bs
, cumulative_perms
, cumulative_shared_perms
);
2346 static void GRAPH_RDLOCK
bdrv_drv_set_perm_abort(void *opaque
)
2348 BlockDriverState
*bs
= opaque
;
2349 GLOBAL_STATE_CODE();
2351 if (bs
->drv
->bdrv_abort_perm_update
) {
2352 bs
->drv
->bdrv_abort_perm_update(bs
);
2356 TransactionActionDrv bdrv_drv_set_perm_drv
= {
2357 .abort
= bdrv_drv_set_perm_abort
,
2358 .commit
= bdrv_drv_set_perm_commit
,
2362 * After calling this function, the transaction @tran may only be completed
2363 * while holding a reader lock for the graph.
2365 static int GRAPH_RDLOCK
2366 bdrv_drv_set_perm(BlockDriverState
*bs
, uint64_t perm
, uint64_t shared_perm
,
2367 Transaction
*tran
, Error
**errp
)
2369 GLOBAL_STATE_CODE();
2374 if (bs
->drv
->bdrv_check_perm
) {
2375 int ret
= bs
->drv
->bdrv_check_perm(bs
, perm
, shared_perm
, errp
);
2382 tran_add(tran
, &bdrv_drv_set_perm_drv
, bs
);
2388 typedef struct BdrvReplaceChildState
{
2390 BlockDriverState
*old_bs
;
2391 } BdrvReplaceChildState
;
2393 static void GRAPH_WRLOCK
bdrv_replace_child_commit(void *opaque
)
2395 BdrvReplaceChildState
*s
= opaque
;
2396 GLOBAL_STATE_CODE();
2398 bdrv_schedule_unref(s
->old_bs
);
2401 static void GRAPH_WRLOCK
bdrv_replace_child_abort(void *opaque
)
2403 BdrvReplaceChildState
*s
= opaque
;
2404 BlockDriverState
*new_bs
= s
->child
->bs
;
2406 GLOBAL_STATE_CODE();
2407 assert_bdrv_graph_writable();
2409 /* old_bs reference is transparently moved from @s to @s->child */
2410 if (!s
->child
->bs
) {
2412 * The parents were undrained when removing old_bs from the child. New
2413 * requests can't have been made, though, because the child was empty.
2415 * TODO Make bdrv_replace_child_noperm() transactionable to avoid
2416 * undraining the parent in the first place. Once this is done, having
2417 * new_bs drained when calling bdrv_replace_child_tran() is not a
2418 * requirement any more.
2420 bdrv_parent_drained_begin_single(s
->child
);
2421 assert(!bdrv_parent_drained_poll_single(s
->child
));
2423 assert(s
->child
->quiesced_parent
);
2424 bdrv_replace_child_noperm(s
->child
, s
->old_bs
);
2429 static TransactionActionDrv bdrv_replace_child_drv
= {
2430 .commit
= bdrv_replace_child_commit
,
2431 .abort
= bdrv_replace_child_abort
,
2436 * bdrv_replace_child_tran
2438 * Note: real unref of old_bs is done only on commit.
2440 * Both @child->bs and @new_bs (if non-NULL) must be drained. @new_bs must be
2441 * kept drained until the transaction is completed.
2443 * After calling this function, the transaction @tran may only be completed
2444 * while holding a writer lock for the graph.
2446 * The function doesn't update permissions, caller is responsible for this.
2448 static void GRAPH_WRLOCK
2449 bdrv_replace_child_tran(BdrvChild
*child
, BlockDriverState
*new_bs
,
2452 BdrvReplaceChildState
*s
= g_new(BdrvReplaceChildState
, 1);
2454 assert(child
->quiesced_parent
);
2455 assert(!new_bs
|| new_bs
->quiesce_counter
);
2457 *s
= (BdrvReplaceChildState
) {
2459 .old_bs
= child
->bs
,
2461 tran_add(tran
, &bdrv_replace_child_drv
, s
);
2467 bdrv_replace_child_noperm(child
, new_bs
);
2468 /* old_bs reference is transparently moved from @child to @s */
2472 * Refresh permissions in @bs subtree. The function is intended to be called
2473 * after some graph modification that was done without permission update.
2475 * After calling this function, the transaction @tran may only be completed
2476 * while holding a reader lock for the graph.
2478 static int GRAPH_RDLOCK
2479 bdrv_node_refresh_perm(BlockDriverState
*bs
, BlockReopenQueue
*q
,
2480 Transaction
*tran
, Error
**errp
)
2482 BlockDriver
*drv
= bs
->drv
;
2485 uint64_t cumulative_perms
, cumulative_shared_perms
;
2486 GLOBAL_STATE_CODE();
2488 bdrv_get_cumulative_perm(bs
, &cumulative_perms
, &cumulative_shared_perms
);
2490 /* Write permissions never work with read-only images */
2491 if ((cumulative_perms
& (BLK_PERM_WRITE
| BLK_PERM_WRITE_UNCHANGED
)) &&
2492 !bdrv_is_writable_after_reopen(bs
, q
))
2494 if (!bdrv_is_writable_after_reopen(bs
, NULL
)) {
2495 error_setg(errp
, "Block node is read-only");
2497 error_setg(errp
, "Read-only block node '%s' cannot support "
2498 "read-write users", bdrv_get_node_name(bs
));
2505 * Unaligned requests will automatically be aligned to bl.request_alignment
2506 * and without RESIZE we can't extend requests to write to space beyond the
2507 * end of the image, so it's required that the image size is aligned.
2509 if ((cumulative_perms
& (BLK_PERM_WRITE
| BLK_PERM_WRITE_UNCHANGED
)) &&
2510 !(cumulative_perms
& BLK_PERM_RESIZE
))
2512 if ((bs
->total_sectors
* BDRV_SECTOR_SIZE
) % bs
->bl
.request_alignment
) {
2513 error_setg(errp
, "Cannot get 'write' permission without 'resize': "
2514 "Image size is not a multiple of request "
2520 /* Check this node */
2525 ret
= bdrv_drv_set_perm(bs
, cumulative_perms
, cumulative_shared_perms
, tran
,
2531 /* Drivers that never have children can omit .bdrv_child_perm() */
2532 if (!drv
->bdrv_child_perm
) {
2533 assert(QLIST_EMPTY(&bs
->children
));
2537 /* Check all children */
2538 QLIST_FOREACH(c
, &bs
->children
, next
) {
2539 uint64_t cur_perm
, cur_shared
;
2541 bdrv_child_perm(bs
, c
->bs
, c
, c
->role
, q
,
2542 cumulative_perms
, cumulative_shared_perms
,
2543 &cur_perm
, &cur_shared
);
2544 bdrv_child_set_perm(c
, cur_perm
, cur_shared
, tran
);
2551 * @list is a product of bdrv_topological_dfs() (may be called several times) -
2552 * a topologically sorted subgraph.
2554 * After calling this function, the transaction @tran may only be completed
2555 * while holding a reader lock for the graph.
2557 static int GRAPH_RDLOCK
2558 bdrv_do_refresh_perms(GSList
*list
, BlockReopenQueue
*q
, Transaction
*tran
,
2562 BlockDriverState
*bs
;
2563 GLOBAL_STATE_CODE();
2565 for ( ; list
; list
= list
->next
) {
2568 if (bdrv_parent_perms_conflict(bs
, errp
)) {
2572 ret
= bdrv_node_refresh_perm(bs
, q
, tran
, errp
);
2582 * @list is any list of nodes. List is completed by all subtrees and
2583 * topologically sorted. It's not a problem if some node occurs in the @list
2586 * After calling this function, the transaction @tran may only be completed
2587 * while holding a reader lock for the graph.
2589 static int GRAPH_RDLOCK
2590 bdrv_list_refresh_perms(GSList
*list
, BlockReopenQueue
*q
, Transaction
*tran
,
2593 g_autoptr(GHashTable
) found
= g_hash_table_new(NULL
, NULL
);
2594 g_autoptr(GSList
) refresh_list
= NULL
;
2596 for ( ; list
; list
= list
->next
) {
2597 refresh_list
= bdrv_topological_dfs(refresh_list
, found
, list
->data
);
2600 return bdrv_do_refresh_perms(refresh_list
, q
, tran
, errp
);
2603 void bdrv_get_cumulative_perm(BlockDriverState
*bs
, uint64_t *perm
,
2604 uint64_t *shared_perm
)
2607 uint64_t cumulative_perms
= 0;
2608 uint64_t cumulative_shared_perms
= BLK_PERM_ALL
;
2610 GLOBAL_STATE_CODE();
2612 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
2613 cumulative_perms
|= c
->perm
;
2614 cumulative_shared_perms
&= c
->shared_perm
;
2617 *perm
= cumulative_perms
;
2618 *shared_perm
= cumulative_shared_perms
;
2621 char *bdrv_perm_names(uint64_t perm
)
2627 { BLK_PERM_CONSISTENT_READ
, "consistent read" },
2628 { BLK_PERM_WRITE
, "write" },
2629 { BLK_PERM_WRITE_UNCHANGED
, "write unchanged" },
2630 { BLK_PERM_RESIZE
, "resize" },
2634 GString
*result
= g_string_sized_new(30);
2635 struct perm_name
*p
;
2637 for (p
= permissions
; p
->name
; p
++) {
2638 if (perm
& p
->perm
) {
2639 if (result
->len
> 0) {
2640 g_string_append(result
, ", ");
2642 g_string_append(result
, p
->name
);
2646 return g_string_free(result
, FALSE
);
2651 * @tran is allowed to be NULL. In this case no rollback is possible.
2653 * After calling this function, the transaction @tran may only be completed
2654 * while holding a reader lock for the graph.
2656 static int GRAPH_RDLOCK
2657 bdrv_refresh_perms(BlockDriverState
*bs
, Transaction
*tran
, Error
**errp
)
2660 Transaction
*local_tran
= NULL
;
2661 g_autoptr(GSList
) list
= bdrv_topological_dfs(NULL
, NULL
, bs
);
2662 GLOBAL_STATE_CODE();
2665 tran
= local_tran
= tran_new();
2668 ret
= bdrv_do_refresh_perms(list
, NULL
, tran
, errp
);
2671 tran_finalize(local_tran
, ret
);
2677 int bdrv_child_try_set_perm(BdrvChild
*c
, uint64_t perm
, uint64_t shared
,
2680 Error
*local_err
= NULL
;
2681 Transaction
*tran
= tran_new();
2684 GLOBAL_STATE_CODE();
2686 bdrv_child_set_perm(c
, perm
, shared
, tran
);
2688 ret
= bdrv_refresh_perms(c
->bs
, tran
, &local_err
);
2690 tran_finalize(tran
, ret
);
2693 if ((perm
& ~c
->perm
) || (c
->shared_perm
& ~shared
)) {
2694 /* tighten permissions */
2695 error_propagate(errp
, local_err
);
2698 * Our caller may intend to only loosen restrictions and
2699 * does not expect this function to fail. Errors are not
2700 * fatal in such a case, so we can just hide them from our
2703 error_free(local_err
);
2711 int bdrv_child_refresh_perms(BlockDriverState
*bs
, BdrvChild
*c
, Error
**errp
)
2713 uint64_t parent_perms
, parent_shared
;
2714 uint64_t perms
, shared
;
2716 GLOBAL_STATE_CODE();
2718 bdrv_get_cumulative_perm(bs
, &parent_perms
, &parent_shared
);
2719 bdrv_child_perm(bs
, c
->bs
, c
, c
->role
, NULL
,
2720 parent_perms
, parent_shared
, &perms
, &shared
);
2722 return bdrv_child_try_set_perm(c
, perms
, shared
, errp
);
2726 * Default implementation for .bdrv_child_perm() for block filters:
2727 * Forward CONSISTENT_READ, WRITE, WRITE_UNCHANGED, and RESIZE to the
2730 static void bdrv_filter_default_perms(BlockDriverState
*bs
, BdrvChild
*c
,
2732 BlockReopenQueue
*reopen_queue
,
2733 uint64_t perm
, uint64_t shared
,
2734 uint64_t *nperm
, uint64_t *nshared
)
2736 GLOBAL_STATE_CODE();
2737 *nperm
= perm
& DEFAULT_PERM_PASSTHROUGH
;
2738 *nshared
= (shared
& DEFAULT_PERM_PASSTHROUGH
) | DEFAULT_PERM_UNCHANGED
;
2741 static void bdrv_default_perms_for_cow(BlockDriverState
*bs
, BdrvChild
*c
,
2743 BlockReopenQueue
*reopen_queue
,
2744 uint64_t perm
, uint64_t shared
,
2745 uint64_t *nperm
, uint64_t *nshared
)
2747 assert(role
& BDRV_CHILD_COW
);
2748 GLOBAL_STATE_CODE();
2751 * We want consistent read from backing files if the parent needs it.
2752 * No other operations are performed on backing files.
2754 perm
&= BLK_PERM_CONSISTENT_READ
;
2757 * If the parent can deal with changing data, we're okay with a
2758 * writable and resizable backing file.
2759 * TODO Require !(perm & BLK_PERM_CONSISTENT_READ), too?
2761 if (shared
& BLK_PERM_WRITE
) {
2762 shared
= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2767 shared
|= BLK_PERM_CONSISTENT_READ
| BLK_PERM_WRITE_UNCHANGED
;
2769 if (bs
->open_flags
& BDRV_O_INACTIVE
) {
2770 shared
|= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2777 static void bdrv_default_perms_for_storage(BlockDriverState
*bs
, BdrvChild
*c
,
2779 BlockReopenQueue
*reopen_queue
,
2780 uint64_t perm
, uint64_t shared
,
2781 uint64_t *nperm
, uint64_t *nshared
)
2785 GLOBAL_STATE_CODE();
2786 assert(role
& (BDRV_CHILD_METADATA
| BDRV_CHILD_DATA
));
2788 flags
= bdrv_reopen_get_flags(reopen_queue
, bs
);
2791 * Apart from the modifications below, the same permissions are
2792 * forwarded and left alone as for filters
2794 bdrv_filter_default_perms(bs
, c
, role
, reopen_queue
,
2795 perm
, shared
, &perm
, &shared
);
2797 if (role
& BDRV_CHILD_METADATA
) {
2798 /* Format drivers may touch metadata even if the guest doesn't write */
2799 if (bdrv_is_writable_after_reopen(bs
, reopen_queue
)) {
2800 perm
|= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2804 * bs->file always needs to be consistent because of the
2805 * metadata. We can never allow other users to resize or write
2808 if (!(flags
& BDRV_O_NO_IO
)) {
2809 perm
|= BLK_PERM_CONSISTENT_READ
;
2811 shared
&= ~(BLK_PERM_WRITE
| BLK_PERM_RESIZE
);
2814 if (role
& BDRV_CHILD_DATA
) {
2816 * Technically, everything in this block is a subset of the
2817 * BDRV_CHILD_METADATA path taken above, and so this could
2818 * be an "else if" branch. However, that is not obvious, and
2819 * this function is not performance critical, therefore we let
2820 * this be an independent "if".
2824 * We cannot allow other users to resize the file because the
2825 * format driver might have some assumptions about the size
2826 * (e.g. because it is stored in metadata, or because the file
2827 * is split into fixed-size data files).
2829 shared
&= ~BLK_PERM_RESIZE
;
2832 * WRITE_UNCHANGED often cannot be performed as such on the
2833 * data file. For example, the qcow2 driver may still need to
2834 * write copied clusters on copy-on-read.
2836 if (perm
& BLK_PERM_WRITE_UNCHANGED
) {
2837 perm
|= BLK_PERM_WRITE
;
2841 * If the data file is written to, the format driver may
2842 * expect to be able to resize it by writing beyond the EOF.
2844 if (perm
& BLK_PERM_WRITE
) {
2845 perm
|= BLK_PERM_RESIZE
;
2849 if (bs
->open_flags
& BDRV_O_INACTIVE
) {
2850 shared
|= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2857 void bdrv_default_perms(BlockDriverState
*bs
, BdrvChild
*c
,
2858 BdrvChildRole role
, BlockReopenQueue
*reopen_queue
,
2859 uint64_t perm
, uint64_t shared
,
2860 uint64_t *nperm
, uint64_t *nshared
)
2862 GLOBAL_STATE_CODE();
2863 if (role
& BDRV_CHILD_FILTERED
) {
2864 assert(!(role
& (BDRV_CHILD_DATA
| BDRV_CHILD_METADATA
|
2866 bdrv_filter_default_perms(bs
, c
, role
, reopen_queue
,
2867 perm
, shared
, nperm
, nshared
);
2868 } else if (role
& BDRV_CHILD_COW
) {
2869 assert(!(role
& (BDRV_CHILD_DATA
| BDRV_CHILD_METADATA
)));
2870 bdrv_default_perms_for_cow(bs
, c
, role
, reopen_queue
,
2871 perm
, shared
, nperm
, nshared
);
2872 } else if (role
& (BDRV_CHILD_METADATA
| BDRV_CHILD_DATA
)) {
2873 bdrv_default_perms_for_storage(bs
, c
, role
, reopen_queue
,
2874 perm
, shared
, nperm
, nshared
);
2876 g_assert_not_reached();
2880 uint64_t bdrv_qapi_perm_to_blk_perm(BlockPermission qapi_perm
)
2882 static const uint64_t permissions
[] = {
2883 [BLOCK_PERMISSION_CONSISTENT_READ
] = BLK_PERM_CONSISTENT_READ
,
2884 [BLOCK_PERMISSION_WRITE
] = BLK_PERM_WRITE
,
2885 [BLOCK_PERMISSION_WRITE_UNCHANGED
] = BLK_PERM_WRITE_UNCHANGED
,
2886 [BLOCK_PERMISSION_RESIZE
] = BLK_PERM_RESIZE
,
2889 QEMU_BUILD_BUG_ON(ARRAY_SIZE(permissions
) != BLOCK_PERMISSION__MAX
);
2890 QEMU_BUILD_BUG_ON(1UL << ARRAY_SIZE(permissions
) != BLK_PERM_ALL
+ 1);
2892 assert(qapi_perm
< BLOCK_PERMISSION__MAX
);
2894 return permissions
[qapi_perm
];
2898 * Replaces the node that a BdrvChild points to without updating permissions.
2900 * If @new_bs is non-NULL, the parent of @child must already be drained through
2901 * @child and the caller must hold the AioContext lock for @new_bs.
2903 static void GRAPH_WRLOCK
2904 bdrv_replace_child_noperm(BdrvChild
*child
, BlockDriverState
*new_bs
)
2906 BlockDriverState
*old_bs
= child
->bs
;
2907 int new_bs_quiesce_counter
;
2909 assert(!child
->frozen
);
2912 * If we want to change the BdrvChild to point to a drained node as its new
2913 * child->bs, we need to make sure that its new parent is drained, too. In
2914 * other words, either child->quiesce_parent must already be true or we must
2915 * be able to set it and keep the parent's quiesce_counter consistent with
2916 * that, but without polling or starting new requests (this function
2917 * guarantees that it doesn't poll, and starting new requests would be
2918 * against the invariants of drain sections).
2920 * To keep things simple, we pick the first option (child->quiesce_parent
2921 * must already be true). We also generalise the rule a bit to make it
2922 * easier to verify in callers and more likely to be covered in test cases:
2923 * The parent must be quiesced through this child even if new_bs isn't
2924 * currently drained.
2926 * The only exception is for callers that always pass new_bs == NULL. In
2927 * this case, we obviously never need to consider the case of a drained
2928 * new_bs, so we can keep the callers simpler by allowing them not to drain
2931 assert(!new_bs
|| child
->quiesced_parent
);
2932 assert(old_bs
!= new_bs
);
2933 GLOBAL_STATE_CODE();
2935 if (old_bs
&& new_bs
) {
2936 assert(bdrv_get_aio_context(old_bs
) == bdrv_get_aio_context(new_bs
));
2940 if (child
->klass
->detach
) {
2941 child
->klass
->detach(child
);
2943 QLIST_REMOVE(child
, next_parent
);
2949 QLIST_INSERT_HEAD(&new_bs
->parents
, child
, next_parent
);
2950 if (child
->klass
->attach
) {
2951 child
->klass
->attach(child
);
2956 * If the parent was drained through this BdrvChild previously, but new_bs
2957 * is not drained, allow requests to come in only after the new node has
2960 new_bs_quiesce_counter
= (new_bs
? new_bs
->quiesce_counter
: 0);
2961 if (!new_bs_quiesce_counter
&& child
->quiesced_parent
) {
2962 bdrv_parent_drained_end_single(child
);
2967 * Free the given @child.
2969 * The child must be empty (i.e. `child->bs == NULL`) and it must be
2970 * unused (i.e. not in a children list).
2972 static void bdrv_child_free(BdrvChild
*child
)
2975 GLOBAL_STATE_CODE();
2976 assert(!child
->next
.le_prev
); /* not in children list */
2978 g_free(child
->name
);
2982 typedef struct BdrvAttachChildCommonState
{
2984 AioContext
*old_parent_ctx
;
2985 AioContext
*old_child_ctx
;
2986 } BdrvAttachChildCommonState
;
2988 static void GRAPH_WRLOCK
bdrv_attach_child_common_abort(void *opaque
)
2990 BdrvAttachChildCommonState
*s
= opaque
;
2991 BlockDriverState
*bs
= s
->child
->bs
;
2993 GLOBAL_STATE_CODE();
2994 assert_bdrv_graph_writable();
2996 bdrv_replace_child_noperm(s
->child
, NULL
);
2998 if (bdrv_get_aio_context(bs
) != s
->old_child_ctx
) {
2999 bdrv_try_change_aio_context(bs
, s
->old_child_ctx
, NULL
, &error_abort
);
3002 if (bdrv_child_get_parent_aio_context(s
->child
) != s
->old_parent_ctx
) {
3004 GHashTable
*visited
;
3009 /* No need to visit `child`, because it has been detached already */
3010 visited
= g_hash_table_new(NULL
, NULL
);
3011 ret
= s
->child
->klass
->change_aio_ctx(s
->child
, s
->old_parent_ctx
,
3012 visited
, tran
, &error_abort
);
3013 g_hash_table_destroy(visited
);
3015 /* transaction is supposed to always succeed */
3016 assert(ret
== true);
3020 bdrv_schedule_unref(bs
);
3021 bdrv_child_free(s
->child
);
3024 static TransactionActionDrv bdrv_attach_child_common_drv
= {
3025 .abort
= bdrv_attach_child_common_abort
,
3030 * Common part of attaching bdrv child to bs or to blk or to job
3032 * Function doesn't update permissions, caller is responsible for this.
3034 * After calling this function, the transaction @tran may only be completed
3035 * while holding a writer lock for the graph.
3037 * Returns new created child.
3039 * The caller must hold the AioContext lock for @child_bs. Both @parent_bs and
3040 * @child_bs can move to a different AioContext in this function. Callers must
3041 * make sure that their AioContext locking is still correct after this.
3043 static BdrvChild
* GRAPH_WRLOCK
3044 bdrv_attach_child_common(BlockDriverState
*child_bs
,
3045 const char *child_name
,
3046 const BdrvChildClass
*child_class
,
3047 BdrvChildRole child_role
,
3048 uint64_t perm
, uint64_t shared_perm
,
3050 Transaction
*tran
, Error
**errp
)
3052 BdrvChild
*new_child
;
3053 AioContext
*parent_ctx
, *new_child_ctx
;
3054 AioContext
*child_ctx
= bdrv_get_aio_context(child_bs
);
3056 assert(child_class
->get_parent_desc
);
3057 GLOBAL_STATE_CODE();
3059 new_child
= g_new(BdrvChild
, 1);
3060 *new_child
= (BdrvChild
) {
3062 .name
= g_strdup(child_name
),
3063 .klass
= child_class
,
3066 .shared_perm
= shared_perm
,
3071 * If the AioContexts don't match, first try to move the subtree of
3072 * child_bs into the AioContext of the new parent. If this doesn't work,
3073 * try moving the parent into the AioContext of child_bs instead.
3075 parent_ctx
= bdrv_child_get_parent_aio_context(new_child
);
3076 if (child_ctx
!= parent_ctx
) {
3077 Error
*local_err
= NULL
;
3078 int ret
= bdrv_try_change_aio_context(child_bs
, parent_ctx
, NULL
,
3081 if (ret
< 0 && child_class
->change_aio_ctx
) {
3082 Transaction
*aio_ctx_tran
= tran_new();
3083 GHashTable
*visited
= g_hash_table_new(NULL
, NULL
);
3086 g_hash_table_add(visited
, new_child
);
3087 ret_child
= child_class
->change_aio_ctx(new_child
, child_ctx
,
3088 visited
, aio_ctx_tran
,
3090 if (ret_child
== true) {
3091 error_free(local_err
);
3094 tran_finalize(aio_ctx_tran
, ret_child
== true ? 0 : -1);
3095 g_hash_table_destroy(visited
);
3099 error_propagate(errp
, local_err
);
3100 bdrv_child_free(new_child
);
3105 new_child_ctx
= bdrv_get_aio_context(child_bs
);
3106 if (new_child_ctx
!= child_ctx
) {
3107 aio_context_release(child_ctx
);
3108 aio_context_acquire(new_child_ctx
);
3113 * Let every new BdrvChild start with a drained parent. Inserting the child
3114 * in the graph with bdrv_replace_child_noperm() will undrain it if
3115 * @child_bs is not drained.
3117 * The child was only just created and is not yet visible in global state
3118 * until bdrv_replace_child_noperm() inserts it into the graph, so nobody
3119 * could have sent requests and polling is not necessary.
3121 * Note that this means that the parent isn't fully drained yet, we only
3122 * stop new requests from coming in. This is fine, we don't care about the
3123 * old requests here, they are not for this child. If another place enters a
3124 * drain section for the same parent, but wants it to be fully quiesced, it
3125 * will not run most of the the code in .drained_begin() again (which is not
3126 * a problem, we already did this), but it will still poll until the parent
3127 * is fully quiesced, so it will not be negatively affected either.
3129 bdrv_parent_drained_begin_single(new_child
);
3130 bdrv_replace_child_noperm(new_child
, child_bs
);
3132 BdrvAttachChildCommonState
*s
= g_new(BdrvAttachChildCommonState
, 1);
3133 *s
= (BdrvAttachChildCommonState
) {
3135 .old_parent_ctx
= parent_ctx
,
3136 .old_child_ctx
= child_ctx
,
3138 tran_add(tran
, &bdrv_attach_child_common_drv
, s
);
3140 if (new_child_ctx
!= child_ctx
) {
3141 aio_context_release(new_child_ctx
);
3142 aio_context_acquire(child_ctx
);
3149 * Function doesn't update permissions, caller is responsible for this.
3151 * The caller must hold the AioContext lock for @child_bs. Both @parent_bs and
3152 * @child_bs can move to a different AioContext in this function. Callers must
3153 * make sure that their AioContext locking is still correct after this.
3155 * After calling this function, the transaction @tran may only be completed
3156 * while holding a writer lock for the graph.
3158 static BdrvChild
* GRAPH_WRLOCK
3159 bdrv_attach_child_noperm(BlockDriverState
*parent_bs
,
3160 BlockDriverState
*child_bs
,
3161 const char *child_name
,
3162 const BdrvChildClass
*child_class
,
3163 BdrvChildRole child_role
,
3167 uint64_t perm
, shared_perm
;
3169 assert(parent_bs
->drv
);
3170 GLOBAL_STATE_CODE();
3172 if (bdrv_recurse_has_child(child_bs
, parent_bs
)) {
3173 error_setg(errp
, "Making '%s' a %s child of '%s' would create a cycle",
3174 child_bs
->node_name
, child_name
, parent_bs
->node_name
);
3178 bdrv_get_cumulative_perm(parent_bs
, &perm
, &shared_perm
);
3179 bdrv_child_perm(parent_bs
, child_bs
, NULL
, child_role
, NULL
,
3180 perm
, shared_perm
, &perm
, &shared_perm
);
3182 return bdrv_attach_child_common(child_bs
, child_name
, child_class
,
3183 child_role
, perm
, shared_perm
, parent_bs
,
3188 * This function steals the reference to child_bs from the caller.
3189 * That reference is later dropped by bdrv_root_unref_child().
3191 * On failure NULL is returned, errp is set and the reference to
3192 * child_bs is also dropped.
3194 * The caller must hold the AioContext lock @child_bs, but not that of @ctx
3195 * (unless @child_bs is already in @ctx).
3197 BdrvChild
*bdrv_root_attach_child(BlockDriverState
*child_bs
,
3198 const char *child_name
,
3199 const BdrvChildClass
*child_class
,
3200 BdrvChildRole child_role
,
3201 uint64_t perm
, uint64_t shared_perm
,
3202 void *opaque
, Error
**errp
)
3206 Transaction
*tran
= tran_new();
3208 GLOBAL_STATE_CODE();
3210 bdrv_graph_wrlock(child_bs
);
3212 child
= bdrv_attach_child_common(child_bs
, child_name
, child_class
,
3213 child_role
, perm
, shared_perm
, opaque
,
3220 ret
= bdrv_refresh_perms(child_bs
, tran
, errp
);
3223 tran_finalize(tran
, ret
);
3224 bdrv_graph_wrunlock();
3226 bdrv_unref(child_bs
);
3228 return ret
< 0 ? NULL
: child
;
3232 * This function transfers the reference to child_bs from the caller
3233 * to parent_bs. That reference is later dropped by parent_bs on
3234 * bdrv_close() or if someone calls bdrv_unref_child().
3236 * On failure NULL is returned, errp is set and the reference to
3237 * child_bs is also dropped.
3239 * If @parent_bs and @child_bs are in different AioContexts, the caller must
3240 * hold the AioContext lock for @child_bs, but not for @parent_bs.
3242 BdrvChild
*bdrv_attach_child(BlockDriverState
*parent_bs
,
3243 BlockDriverState
*child_bs
,
3244 const char *child_name
,
3245 const BdrvChildClass
*child_class
,
3246 BdrvChildRole child_role
,
3251 Transaction
*tran
= tran_new();
3253 GLOBAL_STATE_CODE();
3255 child
= bdrv_attach_child_noperm(parent_bs
, child_bs
, child_name
,
3256 child_class
, child_role
, tran
, errp
);
3262 ret
= bdrv_refresh_perms(parent_bs
, tran
, errp
);
3268 tran_finalize(tran
, ret
);
3270 bdrv_schedule_unref(child_bs
);
3272 return ret
< 0 ? NULL
: child
;
3275 /* Callers must ensure that child->frozen is false. */
3276 void bdrv_root_unref_child(BdrvChild
*child
)
3278 BlockDriverState
*child_bs
= child
->bs
;
3280 GLOBAL_STATE_CODE();
3281 bdrv_replace_child_noperm(child
, NULL
);
3282 bdrv_child_free(child
);
3286 * Update permissions for old node. We're just taking a parent away, so
3287 * we're loosening restrictions. Errors of permission update are not
3288 * fatal in this case, ignore them.
3290 bdrv_refresh_perms(child_bs
, NULL
, NULL
);
3293 * When the parent requiring a non-default AioContext is removed, the
3294 * node moves back to the main AioContext
3296 bdrv_try_change_aio_context(child_bs
, qemu_get_aio_context(), NULL
,
3300 bdrv_schedule_unref(child_bs
);
3303 typedef struct BdrvSetInheritsFrom
{
3304 BlockDriverState
*bs
;
3305 BlockDriverState
*old_inherits_from
;
3306 } BdrvSetInheritsFrom
;
3308 static void bdrv_set_inherits_from_abort(void *opaque
)
3310 BdrvSetInheritsFrom
*s
= opaque
;
3312 s
->bs
->inherits_from
= s
->old_inherits_from
;
3315 static TransactionActionDrv bdrv_set_inherits_from_drv
= {
3316 .abort
= bdrv_set_inherits_from_abort
,
3320 /* @tran is allowed to be NULL. In this case no rollback is possible */
3321 static void bdrv_set_inherits_from(BlockDriverState
*bs
,
3322 BlockDriverState
*new_inherits_from
,
3326 BdrvSetInheritsFrom
*s
= g_new(BdrvSetInheritsFrom
, 1);
3328 *s
= (BdrvSetInheritsFrom
) {
3330 .old_inherits_from
= bs
->inherits_from
,
3333 tran_add(tran
, &bdrv_set_inherits_from_drv
, s
);
3336 bs
->inherits_from
= new_inherits_from
;
3340 * Clear all inherits_from pointers from children and grandchildren of
3341 * @root that point to @root, where necessary.
3342 * @tran is allowed to be NULL. In this case no rollback is possible
3344 static void GRAPH_WRLOCK
3345 bdrv_unset_inherits_from(BlockDriverState
*root
, BdrvChild
*child
,
3350 if (child
->bs
->inherits_from
== root
) {
3352 * Remove inherits_from only when the last reference between root and
3353 * child->bs goes away.
3355 QLIST_FOREACH(c
, &root
->children
, next
) {
3356 if (c
!= child
&& c
->bs
== child
->bs
) {
3361 bdrv_set_inherits_from(child
->bs
, NULL
, tran
);
3365 QLIST_FOREACH(c
, &child
->bs
->children
, next
) {
3366 bdrv_unset_inherits_from(root
, c
, tran
);
3370 /* Callers must ensure that child->frozen is false. */
3371 void bdrv_unref_child(BlockDriverState
*parent
, BdrvChild
*child
)
3373 GLOBAL_STATE_CODE();
3374 if (child
== NULL
) {
3378 bdrv_unset_inherits_from(parent
, child
, NULL
);
3379 bdrv_root_unref_child(child
);
3383 static void GRAPH_RDLOCK
3384 bdrv_parent_cb_change_media(BlockDriverState
*bs
, bool load
)
3387 GLOBAL_STATE_CODE();
3388 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
3389 if (c
->klass
->change_media
) {
3390 c
->klass
->change_media(c
, load
);
3395 /* Return true if you can reach parent going through child->inherits_from
3396 * recursively. If parent or child are NULL, return false */
3397 static bool bdrv_inherits_from_recursive(BlockDriverState
*child
,
3398 BlockDriverState
*parent
)
3400 while (child
&& child
!= parent
) {
3401 child
= child
->inherits_from
;
3404 return child
!= NULL
;
3408 * Return the BdrvChildRole for @bs's backing child. bs->backing is
3409 * mostly used for COW backing children (role = COW), but also for
3410 * filtered children (role = FILTERED | PRIMARY).
3412 static BdrvChildRole
bdrv_backing_role(BlockDriverState
*bs
)
3414 if (bs
->drv
&& bs
->drv
->is_filter
) {
3415 return BDRV_CHILD_FILTERED
| BDRV_CHILD_PRIMARY
;
3417 return BDRV_CHILD_COW
;
3422 * Sets the bs->backing or bs->file link of a BDS. A new reference is created;
3423 * callers which don't need their own reference any more must call bdrv_unref().
3425 * If the respective child is already present (i.e. we're detaching a node),
3426 * that child node must be drained.
3428 * Function doesn't update permissions, caller is responsible for this.
3430 * The caller must hold the AioContext lock for @child_bs. Both @parent_bs and
3431 * @child_bs can move to a different AioContext in this function. Callers must
3432 * make sure that their AioContext locking is still correct after this.
3434 * After calling this function, the transaction @tran may only be completed
3435 * while holding a writer lock for the graph.
3437 static int GRAPH_WRLOCK
3438 bdrv_set_file_or_backing_noperm(BlockDriverState
*parent_bs
,
3439 BlockDriverState
*child_bs
,
3441 Transaction
*tran
, Error
**errp
)
3443 bool update_inherits_from
=
3444 bdrv_inherits_from_recursive(child_bs
, parent_bs
);
3445 BdrvChild
*child
= is_backing
? parent_bs
->backing
: parent_bs
->file
;
3448 GLOBAL_STATE_CODE();
3450 if (!parent_bs
->drv
) {
3452 * Node without drv is an object without a class :/. TODO: finally fix
3453 * qcow2 driver to never clear bs->drv and implement format corruption
3454 * handling in other way.
3456 error_setg(errp
, "Node corrupted");
3460 if (child
&& child
->frozen
) {
3461 error_setg(errp
, "Cannot change frozen '%s' link from '%s' to '%s'",
3462 child
->name
, parent_bs
->node_name
, child
->bs
->node_name
);
3466 if (is_backing
&& !parent_bs
->drv
->is_filter
&&
3467 !parent_bs
->drv
->supports_backing
)
3469 error_setg(errp
, "Driver '%s' of node '%s' does not support backing "
3470 "files", parent_bs
->drv
->format_name
, parent_bs
->node_name
);
3474 if (parent_bs
->drv
->is_filter
) {
3475 role
= BDRV_CHILD_FILTERED
| BDRV_CHILD_PRIMARY
;
3476 } else if (is_backing
) {
3477 role
= BDRV_CHILD_COW
;
3480 * We only can use same role as it is in existing child. We don't have
3481 * infrastructure to determine role of file child in generic way
3484 error_setg(errp
, "Cannot set file child to format node without "
3492 assert(child
->bs
->quiesce_counter
);
3493 bdrv_unset_inherits_from(parent_bs
, child
, tran
);
3494 bdrv_remove_child(child
, tran
);
3501 child
= bdrv_attach_child_noperm(parent_bs
, child_bs
,
3502 is_backing
? "backing" : "file",
3503 &child_of_bds
, role
,
3511 * If inherits_from pointed recursively to bs then let's update it to
3512 * point directly to bs (else it will become NULL).
3514 if (update_inherits_from
) {
3515 bdrv_set_inherits_from(child_bs
, parent_bs
, tran
);
3519 bdrv_refresh_limits(parent_bs
, tran
, NULL
);
3525 * The caller must hold the AioContext lock for @backing_hd. Both @bs and
3526 * @backing_hd can move to a different AioContext in this function. Callers must
3527 * make sure that their AioContext locking is still correct after this.
3529 * If a backing child is already present (i.e. we're detaching a node), that
3530 * child node must be drained.
3532 * After calling this function, the transaction @tran may only be completed
3533 * while holding a writer lock for the graph.
3535 static int GRAPH_WRLOCK
3536 bdrv_set_backing_noperm(BlockDriverState
*bs
,
3537 BlockDriverState
*backing_hd
,
3538 Transaction
*tran
, Error
**errp
)
3540 GLOBAL_STATE_CODE();
3541 return bdrv_set_file_or_backing_noperm(bs
, backing_hd
, true, tran
, errp
);
3544 int bdrv_set_backing_hd_drained(BlockDriverState
*bs
,
3545 BlockDriverState
*backing_hd
,
3549 Transaction
*tran
= tran_new();
3551 GLOBAL_STATE_CODE();
3552 assert(bs
->quiesce_counter
> 0);
3554 assert(bs
->backing
->bs
->quiesce_counter
> 0);
3556 bdrv_graph_wrlock(backing_hd
);
3558 ret
= bdrv_set_backing_noperm(bs
, backing_hd
, tran
, errp
);
3563 ret
= bdrv_refresh_perms(bs
, tran
, errp
);
3565 tran_finalize(tran
, ret
);
3566 bdrv_graph_wrunlock();
3570 int bdrv_set_backing_hd(BlockDriverState
*bs
, BlockDriverState
*backing_hd
,
3573 BlockDriverState
*drain_bs
= bs
->backing
? bs
->backing
->bs
: bs
;
3575 GLOBAL_STATE_CODE();
3578 bdrv_drained_begin(drain_bs
);
3579 ret
= bdrv_set_backing_hd_drained(bs
, backing_hd
, errp
);
3580 bdrv_drained_end(drain_bs
);
3581 bdrv_unref(drain_bs
);
3587 * Opens the backing file for a BlockDriverState if not yet open
3589 * bdref_key specifies the key for the image's BlockdevRef in the options QDict.
3590 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
3591 * itself, all options starting with "${bdref_key}." are considered part of the
3594 * The caller must hold the main AioContext lock.
3596 * TODO Can this be unified with bdrv_open_image()?
3598 int bdrv_open_backing_file(BlockDriverState
*bs
, QDict
*parent_options
,
3599 const char *bdref_key
, Error
**errp
)
3601 char *backing_filename
= NULL
;
3602 char *bdref_key_dot
;
3603 const char *reference
= NULL
;
3605 bool implicit_backing
= false;
3606 BlockDriverState
*backing_hd
;
3607 AioContext
*backing_hd_ctx
;
3609 QDict
*tmp_parent_options
= NULL
;
3610 Error
*local_err
= NULL
;
3612 GLOBAL_STATE_CODE();
3614 if (bs
->backing
!= NULL
) {
3618 /* NULL means an empty set of options */
3619 if (parent_options
== NULL
) {
3620 tmp_parent_options
= qdict_new();
3621 parent_options
= tmp_parent_options
;
3624 bs
->open_flags
&= ~BDRV_O_NO_BACKING
;
3626 bdref_key_dot
= g_strdup_printf("%s.", bdref_key
);
3627 qdict_extract_subqdict(parent_options
, &options
, bdref_key_dot
);
3628 g_free(bdref_key_dot
);
3631 * Caution: while qdict_get_try_str() is fine, getting non-string
3632 * types would require more care. When @parent_options come from
3633 * -blockdev or blockdev_add, its members are typed according to
3634 * the QAPI schema, but when they come from -drive, they're all
3637 reference
= qdict_get_try_str(parent_options
, bdref_key
);
3638 if (reference
|| qdict_haskey(options
, "file.filename")) {
3639 /* keep backing_filename NULL */
3640 } else if (bs
->backing_file
[0] == '\0' && qdict_size(options
) == 0) {
3641 qobject_unref(options
);
3644 if (qdict_size(options
) == 0) {
3645 /* If the user specifies options that do not modify the
3646 * backing file's behavior, we might still consider it the
3647 * implicit backing file. But it's easier this way, and
3648 * just specifying some of the backing BDS's options is
3649 * only possible with -drive anyway (otherwise the QAPI
3650 * schema forces the user to specify everything). */
3651 implicit_backing
= !strcmp(bs
->auto_backing_file
, bs
->backing_file
);
3654 bdrv_graph_rdlock_main_loop();
3655 backing_filename
= bdrv_get_full_backing_filename(bs
, &local_err
);
3656 bdrv_graph_rdunlock_main_loop();
3660 error_propagate(errp
, local_err
);
3661 qobject_unref(options
);
3666 if (!bs
->drv
|| !bs
->drv
->supports_backing
) {
3668 error_setg(errp
, "Driver doesn't support backing files");
3669 qobject_unref(options
);
3674 bs
->backing_format
[0] != '\0' && !qdict_haskey(options
, "driver")) {
3675 qdict_put_str(options
, "driver", bs
->backing_format
);
3678 backing_hd
= bdrv_open_inherit(backing_filename
, reference
, options
, 0, bs
,
3679 &child_of_bds
, bdrv_backing_role(bs
), errp
);
3681 bs
->open_flags
|= BDRV_O_NO_BACKING
;
3682 error_prepend(errp
, "Could not open backing file: ");
3687 if (implicit_backing
) {
3688 bdrv_graph_rdlock_main_loop();
3689 bdrv_refresh_filename(backing_hd
);
3690 bdrv_graph_rdunlock_main_loop();
3691 pstrcpy(bs
->auto_backing_file
, sizeof(bs
->auto_backing_file
),
3692 backing_hd
->filename
);
3695 /* Hook up the backing file link; drop our reference, bs owns the
3696 * backing_hd reference now */
3697 backing_hd_ctx
= bdrv_get_aio_context(backing_hd
);
3698 aio_context_acquire(backing_hd_ctx
);
3699 ret
= bdrv_set_backing_hd(bs
, backing_hd
, errp
);
3700 bdrv_unref(backing_hd
);
3701 aio_context_release(backing_hd_ctx
);
3707 qdict_del(parent_options
, bdref_key
);
3710 g_free(backing_filename
);
3711 qobject_unref(tmp_parent_options
);
3715 static BlockDriverState
*
3716 bdrv_open_child_bs(const char *filename
, QDict
*options
, const char *bdref_key
,
3717 BlockDriverState
*parent
, const BdrvChildClass
*child_class
,
3718 BdrvChildRole child_role
, bool allow_none
, Error
**errp
)
3720 BlockDriverState
*bs
= NULL
;
3721 QDict
*image_options
;
3722 char *bdref_key_dot
;
3723 const char *reference
;
3725 assert(child_class
!= NULL
);
3727 bdref_key_dot
= g_strdup_printf("%s.", bdref_key
);
3728 qdict_extract_subqdict(options
, &image_options
, bdref_key_dot
);
3729 g_free(bdref_key_dot
);
3732 * Caution: while qdict_get_try_str() is fine, getting non-string
3733 * types would require more care. When @options come from
3734 * -blockdev or blockdev_add, its members are typed according to
3735 * the QAPI schema, but when they come from -drive, they're all
3738 reference
= qdict_get_try_str(options
, bdref_key
);
3739 if (!filename
&& !reference
&& !qdict_size(image_options
)) {
3741 error_setg(errp
, "A block device must be specified for \"%s\"",
3744 qobject_unref(image_options
);
3748 bs
= bdrv_open_inherit(filename
, reference
, image_options
, 0,
3749 parent
, child_class
, child_role
, errp
);
3755 qdict_del(options
, bdref_key
);
3760 * Opens a disk image whose options are given as BlockdevRef in another block
3763 * If allow_none is true, no image will be opened if filename is false and no
3764 * BlockdevRef is given. NULL will be returned, but errp remains unset.
3766 * bdrev_key specifies the key for the image's BlockdevRef in the options QDict.
3767 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
3768 * itself, all options starting with "${bdref_key}." are considered part of the
3771 * The BlockdevRef will be removed from the options QDict.
3773 * The caller must hold the lock of the main AioContext and no other AioContext.
3774 * @parent can move to a different AioContext in this function. Callers must
3775 * make sure that their AioContext locking is still correct after this.
3777 BdrvChild
*bdrv_open_child(const char *filename
,
3778 QDict
*options
, const char *bdref_key
,
3779 BlockDriverState
*parent
,
3780 const BdrvChildClass
*child_class
,
3781 BdrvChildRole child_role
,
3782 bool allow_none
, Error
**errp
)
3784 BlockDriverState
*bs
;
3788 GLOBAL_STATE_CODE();
3790 bs
= bdrv_open_child_bs(filename
, options
, bdref_key
, parent
, child_class
,
3791 child_role
, allow_none
, errp
);
3796 bdrv_graph_wrlock(NULL
);
3797 ctx
= bdrv_get_aio_context(bs
);
3798 aio_context_acquire(ctx
);
3799 child
= bdrv_attach_child(parent
, bs
, bdref_key
, child_class
, child_role
,
3801 aio_context_release(ctx
);
3802 bdrv_graph_wrunlock();
3808 * Wrapper on bdrv_open_child() for most popular case: open primary child of bs.
3810 * The caller must hold the lock of the main AioContext and no other AioContext.
3811 * @parent can move to a different AioContext in this function. Callers must
3812 * make sure that their AioContext locking is still correct after this.
3814 int bdrv_open_file_child(const char *filename
,
3815 QDict
*options
, const char *bdref_key
,
3816 BlockDriverState
*parent
, Error
**errp
)
3820 /* commit_top and mirror_top don't use this function */
3821 assert(!parent
->drv
->filtered_child_is_backing
);
3822 role
= parent
->drv
->is_filter
?
3823 (BDRV_CHILD_FILTERED
| BDRV_CHILD_PRIMARY
) : BDRV_CHILD_IMAGE
;
3825 if (!bdrv_open_child(filename
, options
, bdref_key
, parent
,
3826 &child_of_bds
, role
, false, errp
))
3835 * TODO Future callers may need to specify parent/child_class in order for
3836 * option inheritance to work. Existing callers use it for the root node.
3838 BlockDriverState
*bdrv_open_blockdev_ref(BlockdevRef
*ref
, Error
**errp
)
3840 BlockDriverState
*bs
= NULL
;
3841 QObject
*obj
= NULL
;
3842 QDict
*qdict
= NULL
;
3843 const char *reference
= NULL
;
3846 GLOBAL_STATE_CODE();
3848 if (ref
->type
== QTYPE_QSTRING
) {
3849 reference
= ref
->u
.reference
;
3851 BlockdevOptions
*options
= &ref
->u
.definition
;
3852 assert(ref
->type
== QTYPE_QDICT
);
3854 v
= qobject_output_visitor_new(&obj
);
3855 visit_type_BlockdevOptions(v
, NULL
, &options
, &error_abort
);
3856 visit_complete(v
, &obj
);
3858 qdict
= qobject_to(QDict
, obj
);
3859 qdict_flatten(qdict
);
3861 /* bdrv_open_inherit() defaults to the values in bdrv_flags (for
3862 * compatibility with other callers) rather than what we want as the
3863 * real defaults. Apply the defaults here instead. */
3864 qdict_set_default_str(qdict
, BDRV_OPT_CACHE_DIRECT
, "off");
3865 qdict_set_default_str(qdict
, BDRV_OPT_CACHE_NO_FLUSH
, "off");
3866 qdict_set_default_str(qdict
, BDRV_OPT_READ_ONLY
, "off");
3867 qdict_set_default_str(qdict
, BDRV_OPT_AUTO_READ_ONLY
, "off");
3871 bs
= bdrv_open_inherit(NULL
, reference
, qdict
, 0, NULL
, NULL
, 0, errp
);
3878 static BlockDriverState
*bdrv_append_temp_snapshot(BlockDriverState
*bs
,
3880 QDict
*snapshot_options
,
3883 g_autofree
char *tmp_filename
= NULL
;
3885 QemuOpts
*opts
= NULL
;
3886 BlockDriverState
*bs_snapshot
= NULL
;
3887 AioContext
*ctx
= bdrv_get_aio_context(bs
);
3890 GLOBAL_STATE_CODE();
3892 /* if snapshot, we create a temporary backing file and open it
3893 instead of opening 'filename' directly */
3895 /* Get the required size from the image */
3896 aio_context_acquire(ctx
);
3897 total_size
= bdrv_getlength(bs
);
3898 aio_context_release(ctx
);
3900 if (total_size
< 0) {
3901 error_setg_errno(errp
, -total_size
, "Could not get image size");
3905 /* Create the temporary image */
3906 tmp_filename
= create_tmp_file(errp
);
3907 if (!tmp_filename
) {
3911 opts
= qemu_opts_create(bdrv_qcow2
.create_opts
, NULL
, 0,
3913 qemu_opt_set_number(opts
, BLOCK_OPT_SIZE
, total_size
, &error_abort
);
3914 ret
= bdrv_create(&bdrv_qcow2
, tmp_filename
, opts
, errp
);
3915 qemu_opts_del(opts
);
3917 error_prepend(errp
, "Could not create temporary overlay '%s': ",
3922 /* Prepare options QDict for the temporary file */
3923 qdict_put_str(snapshot_options
, "file.driver", "file");
3924 qdict_put_str(snapshot_options
, "file.filename", tmp_filename
);
3925 qdict_put_str(snapshot_options
, "driver", "qcow2");
3927 bs_snapshot
= bdrv_open(NULL
, NULL
, snapshot_options
, flags
, errp
);
3928 snapshot_options
= NULL
;
3933 aio_context_acquire(ctx
);
3934 ret
= bdrv_append(bs_snapshot
, bs
, errp
);
3935 aio_context_release(ctx
);
3943 qobject_unref(snapshot_options
);
3948 * Opens a disk image (raw, qcow2, vmdk, ...)
3950 * options is a QDict of options to pass to the block drivers, or NULL for an
3951 * empty set of options. The reference to the QDict belongs to the block layer
3952 * after the call (even on failure), so if the caller intends to reuse the
3953 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
3955 * If *pbs is NULL, a new BDS will be created with a pointer to it stored there.
3956 * If it is not NULL, the referenced BDS will be reused.
3958 * The reference parameter may be used to specify an existing block device which
3959 * should be opened. If specified, neither options nor a filename may be given,
3960 * nor can an existing BDS be reused (that is, *pbs has to be NULL).
3962 * The caller must always hold the main AioContext lock.
3964 static BlockDriverState
* no_coroutine_fn
3965 bdrv_open_inherit(const char *filename
, const char *reference
, QDict
*options
,
3966 int flags
, BlockDriverState
*parent
,
3967 const BdrvChildClass
*child_class
, BdrvChildRole child_role
,
3971 BlockBackend
*file
= NULL
;
3972 BlockDriverState
*bs
;
3973 BlockDriver
*drv
= NULL
;
3975 const char *drvname
;
3976 const char *backing
;
3977 Error
*local_err
= NULL
;
3978 QDict
*snapshot_options
= NULL
;
3979 int snapshot_flags
= 0;
3980 AioContext
*ctx
= qemu_get_aio_context();
3982 assert(!child_class
|| !flags
);
3983 assert(!child_class
== !parent
);
3984 GLOBAL_STATE_CODE();
3985 assert(!qemu_in_coroutine());
3987 /* TODO We'll eventually have to take a writer lock in this function */
3988 GRAPH_RDLOCK_GUARD_MAINLOOP();
3991 bool options_non_empty
= options
? qdict_size(options
) : false;
3992 qobject_unref(options
);
3994 if (filename
|| options_non_empty
) {
3995 error_setg(errp
, "Cannot reference an existing block device with "
3996 "additional options or a new filename");
4000 bs
= bdrv_lookup_bs(reference
, reference
, errp
);
4011 /* NULL means an empty set of options */
4012 if (options
== NULL
) {
4013 options
= qdict_new();
4016 /* json: syntax counts as explicit options, as if in the QDict */
4017 parse_json_protocol(options
, &filename
, &local_err
);
4022 bs
->explicit_options
= qdict_clone_shallow(options
);
4025 bool parent_is_format
;
4028 parent_is_format
= parent
->drv
->is_format
;
4031 * parent->drv is not set yet because this node is opened for
4032 * (potential) format probing. That means that @parent is going
4033 * to be a format node.
4035 parent_is_format
= true;
4038 bs
->inherits_from
= parent
;
4039 child_class
->inherit_options(child_role
, parent_is_format
,
4041 parent
->open_flags
, parent
->options
);
4044 ret
= bdrv_fill_options(&options
, filename
, &flags
, &local_err
);
4050 * Set the BDRV_O_RDWR and BDRV_O_ALLOW_RDWR flags.
4051 * Caution: getting a boolean member of @options requires care.
4052 * When @options come from -blockdev or blockdev_add, members are
4053 * typed according to the QAPI schema, but when they come from
4054 * -drive, they're all QString.
4056 if (g_strcmp0(qdict_get_try_str(options
, BDRV_OPT_READ_ONLY
), "on") &&
4057 !qdict_get_try_bool(options
, BDRV_OPT_READ_ONLY
, false)) {
4058 flags
|= (BDRV_O_RDWR
| BDRV_O_ALLOW_RDWR
);
4060 flags
&= ~BDRV_O_RDWR
;
4063 if (flags
& BDRV_O_SNAPSHOT
) {
4064 snapshot_options
= qdict_new();
4065 bdrv_temp_snapshot_options(&snapshot_flags
, snapshot_options
,
4067 /* Let bdrv_backing_options() override "read-only" */
4068 qdict_del(options
, BDRV_OPT_READ_ONLY
);
4069 bdrv_inherited_options(BDRV_CHILD_COW
, true,
4070 &flags
, options
, flags
, options
);
4073 bs
->open_flags
= flags
;
4074 bs
->options
= options
;
4075 options
= qdict_clone_shallow(options
);
4077 /* Find the right image format driver */
4078 /* See cautionary note on accessing @options above */
4079 drvname
= qdict_get_try_str(options
, "driver");
4081 drv
= bdrv_find_format(drvname
);
4083 error_setg(errp
, "Unknown driver: '%s'", drvname
);
4088 assert(drvname
|| !(flags
& BDRV_O_PROTOCOL
));
4090 /* See cautionary note on accessing @options above */
4091 backing
= qdict_get_try_str(options
, "backing");
4092 if (qobject_to(QNull
, qdict_get(options
, "backing")) != NULL
||
4093 (backing
&& *backing
== '\0'))
4096 warn_report("Use of \"backing\": \"\" is deprecated; "
4097 "use \"backing\": null instead");
4099 flags
|= BDRV_O_NO_BACKING
;
4100 qdict_del(bs
->explicit_options
, "backing");
4101 qdict_del(bs
->options
, "backing");
4102 qdict_del(options
, "backing");
4105 /* Open image file without format layer. This BlockBackend is only used for
4106 * probing, the block drivers will do their own bdrv_open_child() for the
4107 * same BDS, which is why we put the node name back into options. */
4108 if ((flags
& BDRV_O_PROTOCOL
) == 0) {
4109 BlockDriverState
*file_bs
;
4111 file_bs
= bdrv_open_child_bs(filename
, options
, "file", bs
,
4112 &child_of_bds
, BDRV_CHILD_IMAGE
,
4117 if (file_bs
!= NULL
) {
4118 /* Not requesting BLK_PERM_CONSISTENT_READ because we're only
4119 * looking at the header to guess the image format. This works even
4120 * in cases where a guest would not see a consistent state. */
4121 ctx
= bdrv_get_aio_context(file_bs
);
4122 aio_context_acquire(ctx
);
4123 file
= blk_new(ctx
, 0, BLK_PERM_ALL
);
4124 blk_insert_bs(file
, file_bs
, &local_err
);
4125 bdrv_unref(file_bs
);
4126 aio_context_release(ctx
);
4132 qdict_put_str(options
, "file", bdrv_get_node_name(file_bs
));
4136 /* Image format probing */
4139 ret
= find_image_format(file
, filename
, &drv
, &local_err
);
4144 * This option update would logically belong in bdrv_fill_options(),
4145 * but we first need to open bs->file for the probing to work, while
4146 * opening bs->file already requires the (mostly) final set of options
4147 * so that cache mode etc. can be inherited.
4149 * Adding the driver later is somewhat ugly, but it's not an option
4150 * that would ever be inherited, so it's correct. We just need to make
4151 * sure to update both bs->options (which has the full effective
4152 * options for bs) and options (which has file.* already removed).
4154 qdict_put_str(bs
->options
, "driver", drv
->format_name
);
4155 qdict_put_str(options
, "driver", drv
->format_name
);
4157 error_setg(errp
, "Must specify either driver or file");
4161 /* BDRV_O_PROTOCOL must be set iff a protocol BDS is about to be created */
4162 assert(!!(flags
& BDRV_O_PROTOCOL
) == !!drv
->bdrv_file_open
);
4163 /* file must be NULL if a protocol BDS is about to be created
4164 * (the inverse results in an error message from bdrv_open_common()) */
4165 assert(!(flags
& BDRV_O_PROTOCOL
) || !file
);
4167 /* Open the image */
4168 ret
= bdrv_open_common(bs
, file
, options
, &local_err
);
4173 /* The AioContext could have changed during bdrv_open_common() */
4174 ctx
= bdrv_get_aio_context(bs
);
4177 aio_context_acquire(ctx
);
4179 aio_context_release(ctx
);
4183 /* If there is a backing file, use it */
4184 if ((flags
& BDRV_O_NO_BACKING
) == 0) {
4185 ret
= bdrv_open_backing_file(bs
, options
, "backing", &local_err
);
4187 goto close_and_fail
;
4191 /* Remove all children options and references
4192 * from bs->options and bs->explicit_options */
4193 QLIST_FOREACH(child
, &bs
->children
, next
) {
4194 char *child_key_dot
;
4195 child_key_dot
= g_strdup_printf("%s.", child
->name
);
4196 qdict_extract_subqdict(bs
->explicit_options
, NULL
, child_key_dot
);
4197 qdict_extract_subqdict(bs
->options
, NULL
, child_key_dot
);
4198 qdict_del(bs
->explicit_options
, child
->name
);
4199 qdict_del(bs
->options
, child
->name
);
4200 g_free(child_key_dot
);
4203 /* Check if any unknown options were used */
4204 if (qdict_size(options
) != 0) {
4205 const QDictEntry
*entry
= qdict_first(options
);
4206 if (flags
& BDRV_O_PROTOCOL
) {
4207 error_setg(errp
, "Block protocol '%s' doesn't support the option "
4208 "'%s'", drv
->format_name
, entry
->key
);
4211 "Block format '%s' does not support the option '%s'",
4212 drv
->format_name
, entry
->key
);
4215 goto close_and_fail
;
4218 bdrv_parent_cb_change_media(bs
, true);
4220 qobject_unref(options
);
4223 /* For snapshot=on, create a temporary qcow2 overlay. bs points to the
4224 * temporary snapshot afterwards. */
4225 if (snapshot_flags
) {
4226 BlockDriverState
*snapshot_bs
;
4227 snapshot_bs
= bdrv_append_temp_snapshot(bs
, snapshot_flags
,
4228 snapshot_options
, &local_err
);
4229 snapshot_options
= NULL
;
4231 goto close_and_fail
;
4233 /* We are not going to return bs but the overlay on top of it
4234 * (snapshot_bs); thus, we have to drop the strong reference to bs
4235 * (which we obtained by calling bdrv_new()). bs will not be deleted,
4236 * though, because the overlay still has a reference to it. */
4237 aio_context_acquire(ctx
);
4239 aio_context_release(ctx
);
4246 aio_context_acquire(ctx
);
4248 qobject_unref(snapshot_options
);
4249 qobject_unref(bs
->explicit_options
);
4250 qobject_unref(bs
->options
);
4251 qobject_unref(options
);
4253 bs
->explicit_options
= NULL
;
4255 aio_context_release(ctx
);
4256 error_propagate(errp
, local_err
);
4260 aio_context_acquire(ctx
);
4262 aio_context_release(ctx
);
4263 qobject_unref(snapshot_options
);
4264 qobject_unref(options
);
4265 error_propagate(errp
, local_err
);
4269 /* The caller must always hold the main AioContext lock. */
4270 BlockDriverState
*bdrv_open(const char *filename
, const char *reference
,
4271 QDict
*options
, int flags
, Error
**errp
)
4273 GLOBAL_STATE_CODE();
4275 return bdrv_open_inherit(filename
, reference
, options
, flags
, NULL
,
4279 /* Return true if the NULL-terminated @list contains @str */
4280 static bool is_str_in_list(const char *str
, const char *const *list
)
4284 for (i
= 0; list
[i
] != NULL
; i
++) {
4285 if (!strcmp(str
, list
[i
])) {
4294 * Check that every option set in @bs->options is also set in
4297 * Options listed in the common_options list and in
4298 * @bs->drv->mutable_opts are skipped.
4300 * Return 0 on success, otherwise return -EINVAL and set @errp.
4302 static int bdrv_reset_options_allowed(BlockDriverState
*bs
,
4303 const QDict
*new_opts
, Error
**errp
)
4305 const QDictEntry
*e
;
4306 /* These options are common to all block drivers and are handled
4307 * in bdrv_reopen_prepare() so they can be left out of @new_opts */
4308 const char *const common_options
[] = {
4309 "node-name", "discard", "cache.direct", "cache.no-flush",
4310 "read-only", "auto-read-only", "detect-zeroes", NULL
4313 for (e
= qdict_first(bs
->options
); e
; e
= qdict_next(bs
->options
, e
)) {
4314 if (!qdict_haskey(new_opts
, e
->key
) &&
4315 !is_str_in_list(e
->key
, common_options
) &&
4316 !is_str_in_list(e
->key
, bs
->drv
->mutable_opts
)) {
4317 error_setg(errp
, "Option '%s' cannot be reset "
4318 "to its default value", e
->key
);
4327 * Returns true if @child can be reached recursively from @bs
4329 static bool GRAPH_RDLOCK
4330 bdrv_recurse_has_child(BlockDriverState
*bs
, BlockDriverState
*child
)
4338 QLIST_FOREACH(c
, &bs
->children
, next
) {
4339 if (bdrv_recurse_has_child(c
->bs
, child
)) {
4348 * Adds a BlockDriverState to a simple queue for an atomic, transactional
4349 * reopen of multiple devices.
4351 * bs_queue can either be an existing BlockReopenQueue that has had QTAILQ_INIT
4352 * already performed, or alternatively may be NULL a new BlockReopenQueue will
4353 * be created and initialized. This newly created BlockReopenQueue should be
4354 * passed back in for subsequent calls that are intended to be of the same
4357 * bs is the BlockDriverState to add to the reopen queue.
4359 * options contains the changed options for the associated bs
4360 * (the BlockReopenQueue takes ownership)
4362 * flags contains the open flags for the associated bs
4364 * returns a pointer to bs_queue, which is either the newly allocated
4365 * bs_queue, or the existing bs_queue being used.
4367 * bs is drained here and undrained by bdrv_reopen_queue_free().
4369 * To be called with bs->aio_context locked.
4371 static BlockReopenQueue
* GRAPH_RDLOCK
4372 bdrv_reopen_queue_child(BlockReopenQueue
*bs_queue
, BlockDriverState
*bs
,
4373 QDict
*options
, const BdrvChildClass
*klass
,
4374 BdrvChildRole role
, bool parent_is_format
,
4375 QDict
*parent_options
, int parent_flags
,
4380 BlockReopenQueueEntry
*bs_entry
;
4382 QDict
*old_options
, *explicit_options
, *options_copy
;
4386 GLOBAL_STATE_CODE();
4389 * Strictly speaking, draining is illegal under GRAPH_RDLOCK. We know that
4390 * we've been called with bdrv_graph_rdlock_main_loop(), though, so it's ok
4393 bdrv_drained_begin(bs
);
4395 if (bs_queue
== NULL
) {
4396 bs_queue
= g_new0(BlockReopenQueue
, 1);
4397 QTAILQ_INIT(bs_queue
);
4401 options
= qdict_new();
4404 /* Check if this BlockDriverState is already in the queue */
4405 QTAILQ_FOREACH(bs_entry
, bs_queue
, entry
) {
4406 if (bs
== bs_entry
->state
.bs
) {
4412 * Precedence of options:
4413 * 1. Explicitly passed in options (highest)
4414 * 2. Retained from explicitly set options of bs
4415 * 3. Inherited from parent node
4416 * 4. Retained from effective options of bs
4419 /* Old explicitly set values (don't overwrite by inherited value) */
4420 if (bs_entry
|| keep_old_opts
) {
4421 old_options
= qdict_clone_shallow(bs_entry
?
4422 bs_entry
->state
.explicit_options
:
4423 bs
->explicit_options
);
4424 bdrv_join_options(bs
, options
, old_options
);
4425 qobject_unref(old_options
);
4428 explicit_options
= qdict_clone_shallow(options
);
4430 /* Inherit from parent node */
4431 if (parent_options
) {
4433 klass
->inherit_options(role
, parent_is_format
, &flags
, options
,
4434 parent_flags
, parent_options
);
4436 flags
= bdrv_get_flags(bs
);
4439 if (keep_old_opts
) {
4440 /* Old values are used for options that aren't set yet */
4441 old_options
= qdict_clone_shallow(bs
->options
);
4442 bdrv_join_options(bs
, options
, old_options
);
4443 qobject_unref(old_options
);
4446 /* We have the final set of options so let's update the flags */
4447 options_copy
= qdict_clone_shallow(options
);
4448 opts
= qemu_opts_create(&bdrv_runtime_opts
, NULL
, 0, &error_abort
);
4449 qemu_opts_absorb_qdict(opts
, options_copy
, NULL
);
4450 update_flags_from_options(&flags
, opts
);
4451 qemu_opts_del(opts
);
4452 qobject_unref(options_copy
);
4454 /* bdrv_open_inherit() sets and clears some additional flags internally */
4455 flags
&= ~BDRV_O_PROTOCOL
;
4456 if (flags
& BDRV_O_RDWR
) {
4457 flags
|= BDRV_O_ALLOW_RDWR
;
4461 bs_entry
= g_new0(BlockReopenQueueEntry
, 1);
4462 QTAILQ_INSERT_TAIL(bs_queue
, bs_entry
, entry
);
4464 qobject_unref(bs_entry
->state
.options
);
4465 qobject_unref(bs_entry
->state
.explicit_options
);
4468 bs_entry
->state
.bs
= bs
;
4469 bs_entry
->state
.options
= options
;
4470 bs_entry
->state
.explicit_options
= explicit_options
;
4471 bs_entry
->state
.flags
= flags
;
4474 * If keep_old_opts is false then it means that unspecified
4475 * options must be reset to their original value. We don't allow
4476 * resetting 'backing' but we need to know if the option is
4477 * missing in order to decide if we have to return an error.
4479 if (!keep_old_opts
) {
4480 bs_entry
->state
.backing_missing
=
4481 !qdict_haskey(options
, "backing") &&
4482 !qdict_haskey(options
, "backing.driver");
4485 QLIST_FOREACH(child
, &bs
->children
, next
) {
4486 QDict
*new_child_options
= NULL
;
4487 bool child_keep_old
= keep_old_opts
;
4489 /* reopen can only change the options of block devices that were
4490 * implicitly created and inherited options. For other (referenced)
4491 * block devices, a syntax like "backing.foo" results in an error. */
4492 if (child
->bs
->inherits_from
!= bs
) {
4496 /* Check if the options contain a child reference */
4497 if (qdict_haskey(options
, child
->name
)) {
4498 const char *childref
= qdict_get_try_str(options
, child
->name
);
4500 * The current child must not be reopened if the child
4501 * reference is null or points to a different node.
4503 if (g_strcmp0(childref
, child
->bs
->node_name
)) {
4507 * If the child reference points to the current child then
4508 * reopen it with its existing set of options (note that
4509 * it can still inherit new options from the parent).
4511 child_keep_old
= true;
4513 /* Extract child options ("child-name.*") */
4514 char *child_key_dot
= g_strdup_printf("%s.", child
->name
);
4515 qdict_extract_subqdict(explicit_options
, NULL
, child_key_dot
);
4516 qdict_extract_subqdict(options
, &new_child_options
, child_key_dot
);
4517 g_free(child_key_dot
);
4520 bdrv_reopen_queue_child(bs_queue
, child
->bs
, new_child_options
,
4521 child
->klass
, child
->role
, bs
->drv
->is_format
,
4522 options
, flags
, child_keep_old
);
4528 /* To be called with bs->aio_context locked */
4529 BlockReopenQueue
*bdrv_reopen_queue(BlockReopenQueue
*bs_queue
,
4530 BlockDriverState
*bs
,
4531 QDict
*options
, bool keep_old_opts
)
4533 GLOBAL_STATE_CODE();
4534 GRAPH_RDLOCK_GUARD_MAINLOOP();
4536 return bdrv_reopen_queue_child(bs_queue
, bs
, options
, NULL
, 0, false,
4537 NULL
, 0, keep_old_opts
);
4540 void bdrv_reopen_queue_free(BlockReopenQueue
*bs_queue
)
4542 GLOBAL_STATE_CODE();
4544 BlockReopenQueueEntry
*bs_entry
, *next
;
4545 QTAILQ_FOREACH_SAFE(bs_entry
, bs_queue
, entry
, next
) {
4546 AioContext
*ctx
= bdrv_get_aio_context(bs_entry
->state
.bs
);
4548 aio_context_acquire(ctx
);
4549 bdrv_drained_end(bs_entry
->state
.bs
);
4550 aio_context_release(ctx
);
4552 qobject_unref(bs_entry
->state
.explicit_options
);
4553 qobject_unref(bs_entry
->state
.options
);
4561 * Reopen multiple BlockDriverStates atomically & transactionally.
4563 * The queue passed in (bs_queue) must have been built up previous
4564 * via bdrv_reopen_queue().
4566 * Reopens all BDS specified in the queue, with the appropriate
4567 * flags. All devices are prepared for reopen, and failure of any
4568 * device will cause all device changes to be abandoned, and intermediate
4571 * If all devices prepare successfully, then the changes are committed
4574 * All affected nodes must be drained between bdrv_reopen_queue() and
4575 * bdrv_reopen_multiple().
4577 * To be called from the main thread, with all other AioContexts unlocked.
4579 int bdrv_reopen_multiple(BlockReopenQueue
*bs_queue
, Error
**errp
)
4582 BlockReopenQueueEntry
*bs_entry
, *next
;
4584 Transaction
*tran
= tran_new();
4585 g_autoptr(GSList
) refresh_list
= NULL
;
4587 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
4588 assert(bs_queue
!= NULL
);
4589 GLOBAL_STATE_CODE();
4591 QTAILQ_FOREACH(bs_entry
, bs_queue
, entry
) {
4592 ctx
= bdrv_get_aio_context(bs_entry
->state
.bs
);
4593 aio_context_acquire(ctx
);
4594 ret
= bdrv_flush(bs_entry
->state
.bs
);
4595 aio_context_release(ctx
);
4597 error_setg_errno(errp
, -ret
, "Error flushing drive");
4602 QTAILQ_FOREACH(bs_entry
, bs_queue
, entry
) {
4603 assert(bs_entry
->state
.bs
->quiesce_counter
> 0);
4604 ctx
= bdrv_get_aio_context(bs_entry
->state
.bs
);
4605 aio_context_acquire(ctx
);
4606 ret
= bdrv_reopen_prepare(&bs_entry
->state
, bs_queue
, tran
, errp
);
4607 aio_context_release(ctx
);
4611 bs_entry
->prepared
= true;
4614 QTAILQ_FOREACH(bs_entry
, bs_queue
, entry
) {
4615 BDRVReopenState
*state
= &bs_entry
->state
;
4617 refresh_list
= g_slist_prepend(refresh_list
, state
->bs
);
4618 if (state
->old_backing_bs
) {
4619 refresh_list
= g_slist_prepend(refresh_list
, state
->old_backing_bs
);
4621 if (state
->old_file_bs
) {
4622 refresh_list
= g_slist_prepend(refresh_list
, state
->old_file_bs
);
4627 * Note that file-posix driver rely on permission update done during reopen
4628 * (even if no permission changed), because it wants "new" permissions for
4629 * reconfiguring the fd and that's why it does it in raw_check_perm(), not
4630 * in raw_reopen_prepare() which is called with "old" permissions.
4632 bdrv_graph_rdlock_main_loop();
4633 ret
= bdrv_list_refresh_perms(refresh_list
, bs_queue
, tran
, errp
);
4634 bdrv_graph_rdunlock_main_loop();
4641 * If we reach this point, we have success and just need to apply the
4644 * Reverse order is used to comfort qcow2 driver: on commit it need to write
4645 * IN_USE flag to the image, to mark bitmaps in the image as invalid. But
4646 * children are usually goes after parents in reopen-queue, so go from last
4649 QTAILQ_FOREACH_REVERSE(bs_entry
, bs_queue
, entry
) {
4650 ctx
= bdrv_get_aio_context(bs_entry
->state
.bs
);
4651 aio_context_acquire(ctx
);
4652 bdrv_reopen_commit(&bs_entry
->state
);
4653 aio_context_release(ctx
);
4656 bdrv_graph_wrlock(NULL
);
4658 bdrv_graph_wrunlock();
4660 QTAILQ_FOREACH_REVERSE(bs_entry
, bs_queue
, entry
) {
4661 BlockDriverState
*bs
= bs_entry
->state
.bs
;
4663 if (bs
->drv
->bdrv_reopen_commit_post
) {
4664 ctx
= bdrv_get_aio_context(bs
);
4665 aio_context_acquire(ctx
);
4666 bs
->drv
->bdrv_reopen_commit_post(&bs_entry
->state
);
4667 aio_context_release(ctx
);
4675 bdrv_graph_wrlock(NULL
);
4677 bdrv_graph_wrunlock();
4679 QTAILQ_FOREACH_SAFE(bs_entry
, bs_queue
, entry
, next
) {
4680 if (bs_entry
->prepared
) {
4681 ctx
= bdrv_get_aio_context(bs_entry
->state
.bs
);
4682 aio_context_acquire(ctx
);
4683 bdrv_reopen_abort(&bs_entry
->state
);
4684 aio_context_release(ctx
);
4689 bdrv_reopen_queue_free(bs_queue
);
4694 int bdrv_reopen(BlockDriverState
*bs
, QDict
*opts
, bool keep_old_opts
,
4697 AioContext
*ctx
= bdrv_get_aio_context(bs
);
4698 BlockReopenQueue
*queue
;
4701 GLOBAL_STATE_CODE();
4703 queue
= bdrv_reopen_queue(NULL
, bs
, opts
, keep_old_opts
);
4705 if (ctx
!= qemu_get_aio_context()) {
4706 aio_context_release(ctx
);
4708 ret
= bdrv_reopen_multiple(queue
, errp
);
4710 if (ctx
!= qemu_get_aio_context()) {
4711 aio_context_acquire(ctx
);
4717 int bdrv_reopen_set_read_only(BlockDriverState
*bs
, bool read_only
,
4720 QDict
*opts
= qdict_new();
4722 GLOBAL_STATE_CODE();
4724 qdict_put_bool(opts
, BDRV_OPT_READ_ONLY
, read_only
);
4726 return bdrv_reopen(bs
, opts
, true, errp
);
4730 * Take a BDRVReopenState and check if the value of 'backing' in the
4731 * reopen_state->options QDict is valid or not.
4733 * If 'backing' is missing from the QDict then return 0.
4735 * If 'backing' contains the node name of the backing file of
4736 * reopen_state->bs then return 0.
4738 * If 'backing' contains a different node name (or is null) then check
4739 * whether the current backing file can be replaced with the new one.
4740 * If that's the case then reopen_state->replace_backing_bs is set to
4741 * true and reopen_state->new_backing_bs contains a pointer to the new
4742 * backing BlockDriverState (or NULL).
4744 * After calling this function, the transaction @tran may only be completed
4745 * while holding a writer lock for the graph.
4747 * Return 0 on success, otherwise return < 0 and set @errp.
4749 * The caller must hold the AioContext lock of @reopen_state->bs.
4750 * @reopen_state->bs can move to a different AioContext in this function.
4751 * Callers must make sure that their AioContext locking is still correct after
4754 static int GRAPH_UNLOCKED
4755 bdrv_reopen_parse_file_or_backing(BDRVReopenState
*reopen_state
,
4756 bool is_backing
, Transaction
*tran
,
4759 BlockDriverState
*bs
= reopen_state
->bs
;
4760 BlockDriverState
*new_child_bs
;
4761 BlockDriverState
*old_child_bs
= is_backing
? child_bs(bs
->backing
) :
4763 const char *child_name
= is_backing
? "backing" : "file";
4766 AioContext
*ctx
, *old_ctx
;
4770 GLOBAL_STATE_CODE();
4772 value
= qdict_get(reopen_state
->options
, child_name
);
4773 if (value
== NULL
) {
4777 switch (qobject_type(value
)) {
4779 assert(is_backing
); /* The 'file' option does not allow a null value */
4780 new_child_bs
= NULL
;
4783 str
= qstring_get_str(qobject_to(QString
, value
));
4784 new_child_bs
= bdrv_lookup_bs(NULL
, str
, errp
);
4785 if (new_child_bs
== NULL
) {
4789 bdrv_graph_rdlock_main_loop();
4790 has_child
= bdrv_recurse_has_child(new_child_bs
, bs
);
4791 bdrv_graph_rdunlock_main_loop();
4794 error_setg(errp
, "Making '%s' a %s child of '%s' would create a "
4795 "cycle", str
, child_name
, bs
->node_name
);
4801 * The options QDict has been flattened, so 'backing' and 'file'
4802 * do not allow any other data type here.
4804 g_assert_not_reached();
4807 if (old_child_bs
== new_child_bs
) {
4812 if (bdrv_skip_implicit_filters(old_child_bs
) == new_child_bs
) {
4816 if (old_child_bs
->implicit
) {
4817 error_setg(errp
, "Cannot replace implicit %s child of %s",
4818 child_name
, bs
->node_name
);
4823 if (bs
->drv
->is_filter
&& !old_child_bs
) {
4825 * Filters always have a file or a backing child, so we are trying to
4826 * change wrong child
4828 error_setg(errp
, "'%s' is a %s filter node that does not support a "
4829 "%s child", bs
->node_name
, bs
->drv
->format_name
, child_name
);
4834 reopen_state
->old_backing_bs
= old_child_bs
;
4836 reopen_state
->old_file_bs
= old_child_bs
;
4840 bdrv_ref(old_child_bs
);
4841 bdrv_drained_begin(old_child_bs
);
4844 old_ctx
= bdrv_get_aio_context(bs
);
4845 ctx
= bdrv_get_aio_context(new_child_bs
);
4846 if (old_ctx
!= ctx
) {
4847 aio_context_release(old_ctx
);
4848 aio_context_acquire(ctx
);
4851 bdrv_graph_wrlock(new_child_bs
);
4853 ret
= bdrv_set_file_or_backing_noperm(bs
, new_child_bs
, is_backing
,
4856 bdrv_graph_wrunlock();
4858 if (old_ctx
!= ctx
) {
4859 aio_context_release(ctx
);
4860 aio_context_acquire(old_ctx
);
4864 bdrv_drained_end(old_child_bs
);
4865 bdrv_unref(old_child_bs
);
4872 * Prepares a BlockDriverState for reopen. All changes are staged in the
4873 * 'opaque' field of the BDRVReopenState, which is used and allocated by
4874 * the block driver layer .bdrv_reopen_prepare()
4876 * bs is the BlockDriverState to reopen
4877 * flags are the new open flags
4878 * queue is the reopen queue
4880 * Returns 0 on success, non-zero on error. On error errp will be set
4883 * On failure, bdrv_reopen_abort() will be called to clean up any data.
4884 * It is the responsibility of the caller to then call the abort() or
4885 * commit() for any other BDS that have been left in a prepare() state
4887 * The caller must hold the AioContext lock of @reopen_state->bs.
4889 * After calling this function, the transaction @change_child_tran may only be
4890 * completed while holding a writer lock for the graph.
4892 static int GRAPH_UNLOCKED
4893 bdrv_reopen_prepare(BDRVReopenState
*reopen_state
, BlockReopenQueue
*queue
,
4894 Transaction
*change_child_tran
, Error
**errp
)
4898 Error
*local_err
= NULL
;
4901 QDict
*orig_reopen_opts
;
4902 char *discard
= NULL
;
4904 bool drv_prepared
= false;
4906 assert(reopen_state
!= NULL
);
4907 assert(reopen_state
->bs
->drv
!= NULL
);
4908 GLOBAL_STATE_CODE();
4909 drv
= reopen_state
->bs
->drv
;
4911 /* This function and each driver's bdrv_reopen_prepare() remove
4912 * entries from reopen_state->options as they are processed, so
4913 * we need to make a copy of the original QDict. */
4914 orig_reopen_opts
= qdict_clone_shallow(reopen_state
->options
);
4916 /* Process generic block layer options */
4917 opts
= qemu_opts_create(&bdrv_runtime_opts
, NULL
, 0, &error_abort
);
4918 if (!qemu_opts_absorb_qdict(opts
, reopen_state
->options
, errp
)) {
4923 /* This was already called in bdrv_reopen_queue_child() so the flags
4924 * are up-to-date. This time we simply want to remove the options from
4925 * QemuOpts in order to indicate that they have been processed. */
4926 old_flags
= reopen_state
->flags
;
4927 update_flags_from_options(&reopen_state
->flags
, opts
);
4928 assert(old_flags
== reopen_state
->flags
);
4930 discard
= qemu_opt_get_del(opts
, BDRV_OPT_DISCARD
);
4931 if (discard
!= NULL
) {
4932 if (bdrv_parse_discard_flags(discard
, &reopen_state
->flags
) != 0) {
4933 error_setg(errp
, "Invalid discard option");
4939 reopen_state
->detect_zeroes
=
4940 bdrv_parse_detect_zeroes(opts
, reopen_state
->flags
, &local_err
);
4942 error_propagate(errp
, local_err
);
4947 /* All other options (including node-name and driver) must be unchanged.
4948 * Put them back into the QDict, so that they are checked at the end
4949 * of this function. */
4950 qemu_opts_to_qdict(opts
, reopen_state
->options
);
4952 /* If we are to stay read-only, do not allow permission change
4953 * to r/w. Attempting to set to r/w may fail if either BDRV_O_ALLOW_RDWR is
4954 * not set, or if the BDS still has copy_on_read enabled */
4955 read_only
= !(reopen_state
->flags
& BDRV_O_RDWR
);
4957 bdrv_graph_rdlock_main_loop();
4958 ret
= bdrv_can_set_read_only(reopen_state
->bs
, read_only
, true, &local_err
);
4959 bdrv_graph_rdunlock_main_loop();
4961 error_propagate(errp
, local_err
);
4965 if (drv
->bdrv_reopen_prepare
) {
4967 * If a driver-specific option is missing, it means that we
4968 * should reset it to its default value.
4969 * But not all options allow that, so we need to check it first.
4971 ret
= bdrv_reset_options_allowed(reopen_state
->bs
,
4972 reopen_state
->options
, errp
);
4977 ret
= drv
->bdrv_reopen_prepare(reopen_state
, queue
, &local_err
);
4979 if (local_err
!= NULL
) {
4980 error_propagate(errp
, local_err
);
4982 bdrv_graph_rdlock_main_loop();
4983 bdrv_refresh_filename(reopen_state
->bs
);
4984 bdrv_graph_rdunlock_main_loop();
4985 error_setg(errp
, "failed while preparing to reopen image '%s'",
4986 reopen_state
->bs
->filename
);
4991 /* It is currently mandatory to have a bdrv_reopen_prepare()
4992 * handler for each supported drv. */
4993 bdrv_graph_rdlock_main_loop();
4994 error_setg(errp
, "Block format '%s' used by node '%s' "
4995 "does not support reopening files", drv
->format_name
,
4996 bdrv_get_device_or_node_name(reopen_state
->bs
));
4997 bdrv_graph_rdunlock_main_loop();
5002 drv_prepared
= true;
5005 * We must provide the 'backing' option if the BDS has a backing
5006 * file or if the image file has a backing file name as part of
5007 * its metadata. Otherwise the 'backing' option can be omitted.
5009 if (drv
->supports_backing
&& reopen_state
->backing_missing
&&
5010 (reopen_state
->bs
->backing
|| reopen_state
->bs
->backing_file
[0])) {
5011 error_setg(errp
, "backing is missing for '%s'",
5012 reopen_state
->bs
->node_name
);
5018 * Allow changing the 'backing' option. The new value can be
5019 * either a reference to an existing node (using its node name)
5020 * or NULL to simply detach the current backing file.
5022 ret
= bdrv_reopen_parse_file_or_backing(reopen_state
, true,
5023 change_child_tran
, errp
);
5027 qdict_del(reopen_state
->options
, "backing");
5029 /* Allow changing the 'file' option. In this case NULL is not allowed */
5030 ret
= bdrv_reopen_parse_file_or_backing(reopen_state
, false,
5031 change_child_tran
, errp
);
5035 qdict_del(reopen_state
->options
, "file");
5037 /* Options that are not handled are only okay if they are unchanged
5038 * compared to the old state. It is expected that some options are only
5039 * used for the initial open, but not reopen (e.g. filename) */
5040 if (qdict_size(reopen_state
->options
)) {
5041 const QDictEntry
*entry
= qdict_first(reopen_state
->options
);
5043 GRAPH_RDLOCK_GUARD_MAINLOOP();
5046 QObject
*new = entry
->value
;
5047 QObject
*old
= qdict_get(reopen_state
->bs
->options
, entry
->key
);
5049 /* Allow child references (child_name=node_name) as long as they
5050 * point to the current child (i.e. everything stays the same). */
5051 if (qobject_type(new) == QTYPE_QSTRING
) {
5053 QLIST_FOREACH(child
, &reopen_state
->bs
->children
, next
) {
5054 if (!strcmp(child
->name
, entry
->key
)) {
5060 if (!strcmp(child
->bs
->node_name
,
5061 qstring_get_str(qobject_to(QString
, new)))) {
5062 continue; /* Found child with this name, skip option */
5068 * TODO: When using -drive to specify blockdev options, all values
5069 * will be strings; however, when using -blockdev, blockdev-add or
5070 * filenames using the json:{} pseudo-protocol, they will be
5072 * In contrast, reopening options are (currently) always strings
5073 * (because you can only specify them through qemu-io; all other
5074 * callers do not specify any options).
5075 * Therefore, when using anything other than -drive to create a BDS,
5076 * this cannot detect non-string options as unchanged, because
5077 * qobject_is_equal() always returns false for objects of different
5078 * type. In the future, this should be remedied by correctly typing
5079 * all options. For now, this is not too big of an issue because
5080 * the user can simply omit options which cannot be changed anyway,
5081 * so they will stay unchanged.
5083 if (!qobject_is_equal(new, old
)) {
5084 error_setg(errp
, "Cannot change the option '%s'", entry
->key
);
5088 } while ((entry
= qdict_next(reopen_state
->options
, entry
)));
5093 /* Restore the original reopen_state->options QDict */
5094 qobject_unref(reopen_state
->options
);
5095 reopen_state
->options
= qobject_ref(orig_reopen_opts
);
5098 if (ret
< 0 && drv_prepared
) {
5099 /* drv->bdrv_reopen_prepare() has succeeded, so we need to
5100 * call drv->bdrv_reopen_abort() before signaling an error
5101 * (bdrv_reopen_multiple() will not call bdrv_reopen_abort()
5102 * when the respective bdrv_reopen_prepare() has failed) */
5103 if (drv
->bdrv_reopen_abort
) {
5104 drv
->bdrv_reopen_abort(reopen_state
);
5107 qemu_opts_del(opts
);
5108 qobject_unref(orig_reopen_opts
);
5114 * Takes the staged changes for the reopen from bdrv_reopen_prepare(), and
5115 * makes them final by swapping the staging BlockDriverState contents into
5116 * the active BlockDriverState contents.
5118 static void GRAPH_UNLOCKED
bdrv_reopen_commit(BDRVReopenState
*reopen_state
)
5121 BlockDriverState
*bs
;
5124 assert(reopen_state
!= NULL
);
5125 bs
= reopen_state
->bs
;
5127 assert(drv
!= NULL
);
5128 GLOBAL_STATE_CODE();
5130 /* If there are any driver level actions to take */
5131 if (drv
->bdrv_reopen_commit
) {
5132 drv
->bdrv_reopen_commit(reopen_state
);
5135 GRAPH_RDLOCK_GUARD_MAINLOOP();
5137 /* set BDS specific flags now */
5138 qobject_unref(bs
->explicit_options
);
5139 qobject_unref(bs
->options
);
5140 qobject_ref(reopen_state
->explicit_options
);
5141 qobject_ref(reopen_state
->options
);
5143 bs
->explicit_options
= reopen_state
->explicit_options
;
5144 bs
->options
= reopen_state
->options
;
5145 bs
->open_flags
= reopen_state
->flags
;
5146 bs
->detect_zeroes
= reopen_state
->detect_zeroes
;
5148 /* Remove child references from bs->options and bs->explicit_options.
5149 * Child options were already removed in bdrv_reopen_queue_child() */
5150 QLIST_FOREACH(child
, &bs
->children
, next
) {
5151 qdict_del(bs
->explicit_options
, child
->name
);
5152 qdict_del(bs
->options
, child
->name
);
5154 /* backing is probably removed, so it's not handled by previous loop */
5155 qdict_del(bs
->explicit_options
, "backing");
5156 qdict_del(bs
->options
, "backing");
5158 bdrv_refresh_limits(bs
, NULL
, NULL
);
5159 bdrv_refresh_total_sectors(bs
, bs
->total_sectors
);
5163 * Abort the reopen, and delete and free the staged changes in
5166 static void GRAPH_UNLOCKED
bdrv_reopen_abort(BDRVReopenState
*reopen_state
)
5170 assert(reopen_state
!= NULL
);
5171 drv
= reopen_state
->bs
->drv
;
5172 assert(drv
!= NULL
);
5173 GLOBAL_STATE_CODE();
5175 if (drv
->bdrv_reopen_abort
) {
5176 drv
->bdrv_reopen_abort(reopen_state
);
5181 static void bdrv_close(BlockDriverState
*bs
)
5183 BdrvAioNotifier
*ban
, *ban_next
;
5184 BdrvChild
*child
, *next
;
5186 GLOBAL_STATE_CODE();
5187 assert(!bs
->refcnt
);
5189 bdrv_drained_begin(bs
); /* complete I/O */
5191 bdrv_drain(bs
); /* in case flush left pending I/O */
5194 if (bs
->drv
->bdrv_close
) {
5195 /* Must unfreeze all children, so bdrv_unref_child() works */
5196 bs
->drv
->bdrv_close(bs
);
5201 bdrv_graph_wrlock(NULL
);
5202 QLIST_FOREACH_SAFE(child
, &bs
->children
, next
, next
) {
5203 bdrv_unref_child(bs
, child
);
5205 bdrv_graph_wrunlock();
5207 assert(!bs
->backing
);
5211 qatomic_set(&bs
->copy_on_read
, 0);
5212 bs
->backing_file
[0] = '\0';
5213 bs
->backing_format
[0] = '\0';
5214 bs
->total_sectors
= 0;
5215 bs
->encrypted
= false;
5217 qobject_unref(bs
->options
);
5218 qobject_unref(bs
->explicit_options
);
5220 bs
->explicit_options
= NULL
;
5221 qobject_unref(bs
->full_open_options
);
5222 bs
->full_open_options
= NULL
;
5223 g_free(bs
->block_status_cache
);
5224 bs
->block_status_cache
= NULL
;
5226 bdrv_release_named_dirty_bitmaps(bs
);
5227 assert(QLIST_EMPTY(&bs
->dirty_bitmaps
));
5229 QLIST_FOREACH_SAFE(ban
, &bs
->aio_notifiers
, list
, ban_next
) {
5232 QLIST_INIT(&bs
->aio_notifiers
);
5233 bdrv_drained_end(bs
);
5236 * If we're still inside some bdrv_drain_all_begin()/end() sections, end
5237 * them now since this BDS won't exist anymore when bdrv_drain_all_end()
5240 if (bs
->quiesce_counter
) {
5241 bdrv_drain_all_end_quiesce(bs
);
5245 void bdrv_close_all(void)
5247 GLOBAL_STATE_CODE();
5248 assert(job_next(NULL
) == NULL
);
5250 /* Drop references from requests still in flight, such as canceled block
5251 * jobs whose AIO context has not been polled yet */
5254 blk_remove_all_bs();
5255 blockdev_close_all_bdrv_states();
5257 assert(QTAILQ_EMPTY(&all_bdrv_states
));
5260 static bool GRAPH_RDLOCK
should_update_child(BdrvChild
*c
, BlockDriverState
*to
)
5266 if (c
->klass
->stay_at_node
) {
5270 /* If the child @c belongs to the BDS @to, replacing the current
5271 * c->bs by @to would mean to create a loop.
5273 * Such a case occurs when appending a BDS to a backing chain.
5274 * For instance, imagine the following chain:
5276 * guest device -> node A -> further backing chain...
5278 * Now we create a new BDS B which we want to put on top of this
5279 * chain, so we first attach A as its backing node:
5284 * guest device -> node A -> further backing chain...
5286 * Finally we want to replace A by B. When doing that, we want to
5287 * replace all pointers to A by pointers to B -- except for the
5288 * pointer from B because (1) that would create a loop, and (2)
5289 * that pointer should simply stay intact:
5291 * guest device -> node B
5294 * node A -> further backing chain...
5296 * In general, when replacing a node A (c->bs) by a node B (@to),
5297 * if A is a child of B, that means we cannot replace A by B there
5298 * because that would create a loop. Silently detaching A from B
5299 * is also not really an option. So overall just leaving A in
5300 * place there is the most sensible choice.
5302 * We would also create a loop in any cases where @c is only
5303 * indirectly referenced by @to. Prevent this by returning false
5304 * if @c is found (by breadth-first search) anywhere in the whole
5309 found
= g_hash_table_new(NULL
, NULL
);
5310 g_hash_table_add(found
, to
);
5311 queue
= g_queue_new();
5312 g_queue_push_tail(queue
, to
);
5314 while (!g_queue_is_empty(queue
)) {
5315 BlockDriverState
*v
= g_queue_pop_head(queue
);
5318 QLIST_FOREACH(c2
, &v
->children
, next
) {
5324 if (g_hash_table_contains(found
, c2
->bs
)) {
5328 g_queue_push_tail(queue
, c2
->bs
);
5329 g_hash_table_add(found
, c2
->bs
);
5333 g_queue_free(queue
);
5334 g_hash_table_destroy(found
);
5339 static void bdrv_remove_child_commit(void *opaque
)
5341 GLOBAL_STATE_CODE();
5342 bdrv_child_free(opaque
);
5345 static TransactionActionDrv bdrv_remove_child_drv
= {
5346 .commit
= bdrv_remove_child_commit
,
5350 * Function doesn't update permissions, caller is responsible for this.
5352 * @child->bs (if non-NULL) must be drained.
5354 * After calling this function, the transaction @tran may only be completed
5355 * while holding a writer lock for the graph.
5357 static void GRAPH_WRLOCK
bdrv_remove_child(BdrvChild
*child
, Transaction
*tran
)
5364 assert(child
->quiesced_parent
);
5365 bdrv_replace_child_tran(child
, NULL
, tran
);
5368 tran_add(tran
, &bdrv_remove_child_drv
, child
);
5372 * Both @from and @to (if non-NULL) must be drained. @to must be kept drained
5373 * until the transaction is completed.
5375 * After calling this function, the transaction @tran may only be completed
5376 * while holding a writer lock for the graph.
5378 static int GRAPH_WRLOCK
5379 bdrv_replace_node_noperm(BlockDriverState
*from
,
5380 BlockDriverState
*to
,
5381 bool auto_skip
, Transaction
*tran
,
5384 BdrvChild
*c
, *next
;
5386 GLOBAL_STATE_CODE();
5388 assert(from
->quiesce_counter
);
5389 assert(to
->quiesce_counter
);
5391 QLIST_FOREACH_SAFE(c
, &from
->parents
, next_parent
, next
) {
5392 assert(c
->bs
== from
);
5393 if (!should_update_child(c
, to
)) {
5397 error_setg(errp
, "Should not change '%s' link to '%s'",
5398 c
->name
, from
->node_name
);
5402 error_setg(errp
, "Cannot change '%s' link to '%s'",
5403 c
->name
, from
->node_name
);
5406 bdrv_replace_child_tran(c
, to
, tran
);
5413 * With auto_skip=true bdrv_replace_node_common skips updating from parents
5414 * if it creates a parent-child relation loop or if parent is block-job.
5416 * With auto_skip=false the error is returned if from has a parent which should
5419 * With @detach_subchain=true @to must be in a backing chain of @from. In this
5420 * case backing link of the cow-parent of @to is removed.
5422 static int bdrv_replace_node_common(BlockDriverState
*from
,
5423 BlockDriverState
*to
,
5424 bool auto_skip
, bool detach_subchain
,
5427 Transaction
*tran
= tran_new();
5428 g_autoptr(GSList
) refresh_list
= NULL
;
5429 BlockDriverState
*to_cow_parent
= NULL
;
5432 GLOBAL_STATE_CODE();
5434 if (detach_subchain
) {
5435 assert(bdrv_chain_contains(from
, to
));
5437 for (to_cow_parent
= from
;
5438 bdrv_filter_or_cow_bs(to_cow_parent
) != to
;
5439 to_cow_parent
= bdrv_filter_or_cow_bs(to_cow_parent
))
5445 /* Make sure that @from doesn't go away until we have successfully attached
5446 * all of its parents to @to. */
5449 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
5450 assert(bdrv_get_aio_context(from
) == bdrv_get_aio_context(to
));
5451 bdrv_drained_begin(from
);
5452 bdrv_drained_begin(to
);
5454 bdrv_graph_wrlock(to
);
5457 * Do the replacement without permission update.
5458 * Replacement may influence the permissions, we should calculate new
5459 * permissions based on new graph. If we fail, we'll roll-back the
5462 ret
= bdrv_replace_node_noperm(from
, to
, auto_skip
, tran
, errp
);
5467 if (detach_subchain
) {
5468 /* to_cow_parent is already drained because from is drained */
5469 bdrv_remove_child(bdrv_filter_or_cow_child(to_cow_parent
), tran
);
5472 refresh_list
= g_slist_prepend(refresh_list
, to
);
5473 refresh_list
= g_slist_prepend(refresh_list
, from
);
5475 ret
= bdrv_list_refresh_perms(refresh_list
, NULL
, tran
, errp
);
5483 tran_finalize(tran
, ret
);
5484 bdrv_graph_wrunlock();
5486 bdrv_drained_end(to
);
5487 bdrv_drained_end(from
);
5493 int bdrv_replace_node(BlockDriverState
*from
, BlockDriverState
*to
,
5496 GLOBAL_STATE_CODE();
5498 return bdrv_replace_node_common(from
, to
, true, false, errp
);
5501 int bdrv_drop_filter(BlockDriverState
*bs
, Error
**errp
)
5503 GLOBAL_STATE_CODE();
5505 return bdrv_replace_node_common(bs
, bdrv_filter_or_cow_bs(bs
), true, true,
5510 * Add new bs contents at the top of an image chain while the chain is
5511 * live, while keeping required fields on the top layer.
5513 * This will modify the BlockDriverState fields, and swap contents
5514 * between bs_new and bs_top. Both bs_new and bs_top are modified.
5516 * bs_new must not be attached to a BlockBackend and must not have backing
5519 * This function does not create any image files.
5521 * The caller must hold the AioContext lock for @bs_top.
5523 int bdrv_append(BlockDriverState
*bs_new
, BlockDriverState
*bs_top
,
5528 Transaction
*tran
= tran_new();
5529 AioContext
*old_context
, *new_context
= NULL
;
5531 GLOBAL_STATE_CODE();
5533 assert(!bs_new
->backing
);
5535 old_context
= bdrv_get_aio_context(bs_top
);
5536 bdrv_drained_begin(bs_top
);
5539 * bdrv_drained_begin() requires that only the AioContext of the drained
5540 * node is locked, and at this point it can still differ from the AioContext
5543 new_context
= bdrv_get_aio_context(bs_new
);
5544 aio_context_release(old_context
);
5545 aio_context_acquire(new_context
);
5546 bdrv_drained_begin(bs_new
);
5547 aio_context_release(new_context
);
5548 aio_context_acquire(old_context
);
5551 bdrv_graph_wrlock(bs_top
);
5553 child
= bdrv_attach_child_noperm(bs_new
, bs_top
, "backing",
5554 &child_of_bds
, bdrv_backing_role(bs_new
),
5562 * bdrv_attach_child_noperm could change the AioContext of bs_top and
5563 * bs_new, but at least they are in the same AioContext now. This is the
5564 * AioContext that we need to lock for the rest of the function.
5566 new_context
= bdrv_get_aio_context(bs_top
);
5568 if (old_context
!= new_context
) {
5569 aio_context_release(old_context
);
5570 aio_context_acquire(new_context
);
5573 ret
= bdrv_replace_node_noperm(bs_top
, bs_new
, true, tran
, errp
);
5578 ret
= bdrv_refresh_perms(bs_new
, tran
, errp
);
5580 tran_finalize(tran
, ret
);
5582 bdrv_refresh_limits(bs_top
, NULL
, NULL
);
5583 bdrv_graph_wrunlock();
5585 bdrv_drained_end(bs_top
);
5586 bdrv_drained_end(bs_new
);
5588 if (new_context
&& old_context
!= new_context
) {
5589 aio_context_release(new_context
);
5590 aio_context_acquire(old_context
);
5596 /* Not for empty child */
5597 int bdrv_replace_child_bs(BdrvChild
*child
, BlockDriverState
*new_bs
,
5601 Transaction
*tran
= tran_new();
5602 g_autoptr(GSList
) refresh_list
= NULL
;
5603 BlockDriverState
*old_bs
= child
->bs
;
5605 GLOBAL_STATE_CODE();
5608 bdrv_drained_begin(old_bs
);
5609 bdrv_drained_begin(new_bs
);
5610 bdrv_graph_wrlock(new_bs
);
5612 bdrv_replace_child_tran(child
, new_bs
, tran
);
5614 refresh_list
= g_slist_prepend(refresh_list
, old_bs
);
5615 refresh_list
= g_slist_prepend(refresh_list
, new_bs
);
5617 ret
= bdrv_list_refresh_perms(refresh_list
, NULL
, tran
, errp
);
5619 tran_finalize(tran
, ret
);
5621 bdrv_graph_wrunlock();
5622 bdrv_drained_end(old_bs
);
5623 bdrv_drained_end(new_bs
);
5629 static void bdrv_delete(BlockDriverState
*bs
)
5631 assert(bdrv_op_blocker_is_empty(bs
));
5632 assert(!bs
->refcnt
);
5633 GLOBAL_STATE_CODE();
5635 /* remove from list, if necessary */
5636 if (bs
->node_name
[0] != '\0') {
5637 QTAILQ_REMOVE(&graph_bdrv_states
, bs
, node_list
);
5639 QTAILQ_REMOVE(&all_bdrv_states
, bs
, bs_list
);
5643 qemu_mutex_destroy(&bs
->reqs_lock
);
5650 * Replace @bs by newly created block node.
5652 * @options is a QDict of options to pass to the block drivers, or NULL for an
5653 * empty set of options. The reference to the QDict belongs to the block layer
5654 * after the call (even on failure), so if the caller intends to reuse the
5655 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
5657 * The caller holds the AioContext lock for @bs. It must make sure that @bs
5658 * stays in the same AioContext, i.e. @options must not refer to nodes in a
5659 * different AioContext.
5661 BlockDriverState
*bdrv_insert_node(BlockDriverState
*bs
, QDict
*options
,
5662 int flags
, Error
**errp
)
5666 AioContext
*ctx
= bdrv_get_aio_context(bs
);
5667 BlockDriverState
*new_node_bs
= NULL
;
5668 const char *drvname
, *node_name
;
5671 drvname
= qdict_get_try_str(options
, "driver");
5673 error_setg(errp
, "driver is not specified");
5677 drv
= bdrv_find_format(drvname
);
5679 error_setg(errp
, "Unknown driver: '%s'", drvname
);
5683 node_name
= qdict_get_try_str(options
, "node-name");
5685 GLOBAL_STATE_CODE();
5687 aio_context_release(ctx
);
5688 aio_context_acquire(qemu_get_aio_context());
5689 new_node_bs
= bdrv_new_open_driver_opts(drv
, node_name
, options
, flags
,
5691 aio_context_release(qemu_get_aio_context());
5692 aio_context_acquire(ctx
);
5693 assert(bdrv_get_aio_context(bs
) == ctx
);
5695 options
= NULL
; /* bdrv_new_open_driver() eats options */
5697 error_prepend(errp
, "Could not create node: ");
5701 bdrv_drained_begin(bs
);
5702 ret
= bdrv_replace_node(bs
, new_node_bs
, errp
);
5703 bdrv_drained_end(bs
);
5706 error_prepend(errp
, "Could not replace node: ");
5713 qobject_unref(options
);
5714 bdrv_unref(new_node_bs
);
5719 * Run consistency checks on an image
5721 * Returns 0 if the check could be completed (it doesn't mean that the image is
5722 * free of errors) or -errno when an internal error occurred. The results of the
5723 * check are stored in res.
5725 int coroutine_fn
bdrv_co_check(BlockDriverState
*bs
,
5726 BdrvCheckResult
*res
, BdrvCheckMode fix
)
5729 assert_bdrv_graph_readable();
5730 if (bs
->drv
== NULL
) {
5733 if (bs
->drv
->bdrv_co_check
== NULL
) {
5737 memset(res
, 0, sizeof(*res
));
5738 return bs
->drv
->bdrv_co_check(bs
, res
, fix
);
5744 * -EINVAL - backing format specified, but no file
5745 * -ENOSPC - can't update the backing file because no space is left in the
5747 * -ENOTSUP - format driver doesn't support changing the backing file
5749 int bdrv_change_backing_file(BlockDriverState
*bs
, const char *backing_file
,
5750 const char *backing_fmt
, bool require
)
5752 BlockDriver
*drv
= bs
->drv
;
5755 GLOBAL_STATE_CODE();
5761 /* Backing file format doesn't make sense without a backing file */
5762 if (backing_fmt
&& !backing_file
) {
5766 if (require
&& backing_file
&& !backing_fmt
) {
5770 if (drv
->bdrv_change_backing_file
!= NULL
) {
5771 ret
= drv
->bdrv_change_backing_file(bs
, backing_file
, backing_fmt
);
5777 pstrcpy(bs
->backing_file
, sizeof(bs
->backing_file
), backing_file
?: "");
5778 pstrcpy(bs
->backing_format
, sizeof(bs
->backing_format
), backing_fmt
?: "");
5779 pstrcpy(bs
->auto_backing_file
, sizeof(bs
->auto_backing_file
),
5780 backing_file
?: "");
5786 * Finds the first non-filter node above bs in the chain between
5787 * active and bs. The returned node is either an immediate parent of
5788 * bs, or there are only filter nodes between the two.
5790 * Returns NULL if bs is not found in active's image chain,
5791 * or if active == bs.
5793 * Returns the bottommost base image if bs == NULL.
5795 BlockDriverState
*bdrv_find_overlay(BlockDriverState
*active
,
5796 BlockDriverState
*bs
)
5799 GLOBAL_STATE_CODE();
5801 bs
= bdrv_skip_filters(bs
);
5802 active
= bdrv_skip_filters(active
);
5805 BlockDriverState
*next
= bdrv_backing_chain_next(active
);
5815 /* Given a BDS, searches for the base layer. */
5816 BlockDriverState
*bdrv_find_base(BlockDriverState
*bs
)
5818 GLOBAL_STATE_CODE();
5820 return bdrv_find_overlay(bs
, NULL
);
5824 * Return true if at least one of the COW (backing) and filter links
5825 * between @bs and @base is frozen. @errp is set if that's the case.
5826 * @base must be reachable from @bs, or NULL.
5828 bool bdrv_is_backing_chain_frozen(BlockDriverState
*bs
, BlockDriverState
*base
,
5831 BlockDriverState
*i
;
5834 GLOBAL_STATE_CODE();
5836 for (i
= bs
; i
!= base
; i
= child_bs(child
)) {
5837 child
= bdrv_filter_or_cow_child(i
);
5839 if (child
&& child
->frozen
) {
5840 error_setg(errp
, "Cannot change '%s' link from '%s' to '%s'",
5841 child
->name
, i
->node_name
, child
->bs
->node_name
);
5850 * Freeze all COW (backing) and filter links between @bs and @base.
5851 * If any of the links is already frozen the operation is aborted and
5852 * none of the links are modified.
5853 * @base must be reachable from @bs, or NULL.
5854 * Returns 0 on success. On failure returns < 0 and sets @errp.
5856 int bdrv_freeze_backing_chain(BlockDriverState
*bs
, BlockDriverState
*base
,
5859 BlockDriverState
*i
;
5862 GLOBAL_STATE_CODE();
5864 if (bdrv_is_backing_chain_frozen(bs
, base
, errp
)) {
5868 for (i
= bs
; i
!= base
; i
= child_bs(child
)) {
5869 child
= bdrv_filter_or_cow_child(i
);
5870 if (child
&& child
->bs
->never_freeze
) {
5871 error_setg(errp
, "Cannot freeze '%s' link to '%s'",
5872 child
->name
, child
->bs
->node_name
);
5877 for (i
= bs
; i
!= base
; i
= child_bs(child
)) {
5878 child
= bdrv_filter_or_cow_child(i
);
5880 child
->frozen
= true;
5888 * Unfreeze all COW (backing) and filter links between @bs and @base.
5889 * The caller must ensure that all links are frozen before using this
5891 * @base must be reachable from @bs, or NULL.
5893 void bdrv_unfreeze_backing_chain(BlockDriverState
*bs
, BlockDriverState
*base
)
5895 BlockDriverState
*i
;
5898 GLOBAL_STATE_CODE();
5900 for (i
= bs
; i
!= base
; i
= child_bs(child
)) {
5901 child
= bdrv_filter_or_cow_child(i
);
5903 assert(child
->frozen
);
5904 child
->frozen
= false;
5910 * Drops images above 'base' up to and including 'top', and sets the image
5911 * above 'top' to have base as its backing file.
5913 * Requires that the overlay to 'top' is opened r/w, so that the backing file
5914 * information in 'bs' can be properly updated.
5916 * E.g., this will convert the following chain:
5917 * bottom <- base <- intermediate <- top <- active
5921 * bottom <- base <- active
5923 * It is allowed for bottom==base, in which case it converts:
5925 * base <- intermediate <- top <- active
5931 * If backing_file_str is non-NULL, it will be used when modifying top's
5932 * overlay image metadata.
5935 * if active == top, that is considered an error
5938 int bdrv_drop_intermediate(BlockDriverState
*top
, BlockDriverState
*base
,
5939 const char *backing_file_str
)
5941 BlockDriverState
*explicit_top
= top
;
5942 bool update_inherits_from
;
5944 Error
*local_err
= NULL
;
5946 g_autoptr(GSList
) updated_children
= NULL
;
5949 GLOBAL_STATE_CODE();
5952 bdrv_drained_begin(base
);
5953 bdrv_graph_rdlock_main_loop();
5955 if (!top
->drv
|| !base
->drv
) {
5959 /* Make sure that base is in the backing chain of top */
5960 if (!bdrv_chain_contains(top
, base
)) {
5964 /* If 'base' recursively inherits from 'top' then we should set
5965 * base->inherits_from to top->inherits_from after 'top' and all
5966 * other intermediate nodes have been dropped.
5967 * If 'top' is an implicit node (e.g. "commit_top") we should skip
5968 * it because no one inherits from it. We use explicit_top for that. */
5969 explicit_top
= bdrv_skip_implicit_filters(explicit_top
);
5970 update_inherits_from
= bdrv_inherits_from_recursive(base
, explicit_top
);
5972 /* success - we can delete the intermediate states, and link top->base */
5973 if (!backing_file_str
) {
5974 bdrv_refresh_filename(base
);
5975 backing_file_str
= base
->filename
;
5978 QLIST_FOREACH(c
, &top
->parents
, next_parent
) {
5979 updated_children
= g_slist_prepend(updated_children
, c
);
5983 * It seems correct to pass detach_subchain=true here, but it triggers
5984 * one more yet not fixed bug, when due to nested aio_poll loop we switch to
5985 * another drained section, which modify the graph (for example, removing
5986 * the child, which we keep in updated_children list). So, it's a TODO.
5988 * Note, bug triggered if pass detach_subchain=true here and run
5989 * test-bdrv-drain. test_drop_intermediate_poll() test-case will crash.
5992 bdrv_replace_node_common(top
, base
, false, false, &local_err
);
5994 error_report_err(local_err
);
5998 for (p
= updated_children
; p
; p
= p
->next
) {
6001 if (c
->klass
->update_filename
) {
6002 ret
= c
->klass
->update_filename(c
, base
, backing_file_str
,
6006 * TODO: Actually, we want to rollback all previous iterations
6007 * of this loop, and (which is almost impossible) previous
6008 * bdrv_replace_node()...
6010 * Note, that c->klass->update_filename may lead to permission
6011 * update, so it's a bad idea to call it inside permission
6012 * update transaction of bdrv_replace_node.
6014 error_report_err(local_err
);
6020 if (update_inherits_from
) {
6021 base
->inherits_from
= explicit_top
->inherits_from
;
6026 bdrv_graph_rdunlock_main_loop();
6027 bdrv_drained_end(base
);
6033 * Implementation of BlockDriver.bdrv_co_get_allocated_file_size() that
6034 * sums the size of all data-bearing children. (This excludes backing
6037 static int64_t coroutine_fn GRAPH_RDLOCK
6038 bdrv_sum_allocated_file_size(BlockDriverState
*bs
)
6041 int64_t child_size
, sum
= 0;
6043 QLIST_FOREACH(child
, &bs
->children
, next
) {
6044 if (child
->role
& (BDRV_CHILD_DATA
| BDRV_CHILD_METADATA
|
6045 BDRV_CHILD_FILTERED
))
6047 child_size
= bdrv_co_get_allocated_file_size(child
->bs
);
6048 if (child_size
< 0) {
6059 * Length of a allocated file in bytes. Sparse files are counted by actual
6060 * allocated space. Return < 0 if error or unknown.
6062 int64_t coroutine_fn
bdrv_co_get_allocated_file_size(BlockDriverState
*bs
)
6064 BlockDriver
*drv
= bs
->drv
;
6066 assert_bdrv_graph_readable();
6071 if (drv
->bdrv_co_get_allocated_file_size
) {
6072 return drv
->bdrv_co_get_allocated_file_size(bs
);
6075 if (drv
->bdrv_file_open
) {
6077 * Protocol drivers default to -ENOTSUP (most of their data is
6078 * not stored in any of their children (if they even have any),
6079 * so there is no generic way to figure it out).
6082 } else if (drv
->is_filter
) {
6083 /* Filter drivers default to the size of their filtered child */
6084 return bdrv_co_get_allocated_file_size(bdrv_filter_bs(bs
));
6086 /* Other drivers default to summing their children's sizes */
6087 return bdrv_sum_allocated_file_size(bs
);
6093 * @drv: Format driver
6094 * @opts: Creation options for new image
6095 * @in_bs: Existing image containing data for new image (may be NULL)
6096 * @errp: Error object
6097 * Returns: A #BlockMeasureInfo (free using qapi_free_BlockMeasureInfo())
6100 * Calculate file size required to create a new image.
6102 * If @in_bs is given then space for allocated clusters and zero clusters
6103 * from that image are included in the calculation. If @opts contains a
6104 * backing file that is shared by @in_bs then backing clusters may be omitted
6105 * from the calculation.
6107 * If @in_bs is NULL then the calculation includes no allocated clusters
6108 * unless a preallocation option is given in @opts.
6110 * Note that @in_bs may use a different BlockDriver from @drv.
6112 * If an error occurs the @errp pointer is set.
6114 BlockMeasureInfo
*bdrv_measure(BlockDriver
*drv
, QemuOpts
*opts
,
6115 BlockDriverState
*in_bs
, Error
**errp
)
6118 if (!drv
->bdrv_measure
) {
6119 error_setg(errp
, "Block driver '%s' does not support size measurement",
6124 return drv
->bdrv_measure(opts
, in_bs
, errp
);
6128 * Return number of sectors on success, -errno on error.
6130 int64_t coroutine_fn
bdrv_co_nb_sectors(BlockDriverState
*bs
)
6132 BlockDriver
*drv
= bs
->drv
;
6134 assert_bdrv_graph_readable();
6139 if (bs
->bl
.has_variable_length
) {
6140 int ret
= bdrv_co_refresh_total_sectors(bs
, bs
->total_sectors
);
6145 return bs
->total_sectors
;
6149 * This wrapper is written by hand because this function is in the hot I/O path,
6150 * via blk_get_geometry.
6152 int64_t coroutine_mixed_fn
bdrv_nb_sectors(BlockDriverState
*bs
)
6154 BlockDriver
*drv
= bs
->drv
;
6160 if (bs
->bl
.has_variable_length
) {
6161 int ret
= bdrv_refresh_total_sectors(bs
, bs
->total_sectors
);
6167 return bs
->total_sectors
;
6171 * Return length in bytes on success, -errno on error.
6172 * The length is always a multiple of BDRV_SECTOR_SIZE.
6174 int64_t coroutine_fn
bdrv_co_getlength(BlockDriverState
*bs
)
6178 assert_bdrv_graph_readable();
6180 ret
= bdrv_co_nb_sectors(bs
);
6184 if (ret
> INT64_MAX
/ BDRV_SECTOR_SIZE
) {
6187 return ret
* BDRV_SECTOR_SIZE
;
6190 bool bdrv_is_sg(BlockDriverState
*bs
)
6197 * Return whether the given node supports compressed writes.
6199 bool bdrv_supports_compressed_writes(BlockDriverState
*bs
)
6201 BlockDriverState
*filtered
;
6204 if (!bs
->drv
|| !block_driver_can_compress(bs
->drv
)) {
6208 filtered
= bdrv_filter_bs(bs
);
6211 * Filters can only forward compressed writes, so we have to
6214 return bdrv_supports_compressed_writes(filtered
);
6220 const char *bdrv_get_format_name(BlockDriverState
*bs
)
6223 return bs
->drv
? bs
->drv
->format_name
: NULL
;
6226 static int qsort_strcmp(const void *a
, const void *b
)
6228 return strcmp(*(char *const *)a
, *(char *const *)b
);
6231 void bdrv_iterate_format(void (*it
)(void *opaque
, const char *name
),
6232 void *opaque
, bool read_only
)
6237 const char **formats
= NULL
;
6239 GLOBAL_STATE_CODE();
6241 QLIST_FOREACH(drv
, &bdrv_drivers
, list
) {
6242 if (drv
->format_name
) {
6245 if (use_bdrv_whitelist
&& !bdrv_is_whitelisted(drv
, read_only
)) {
6250 while (formats
&& i
&& !found
) {
6251 found
= !strcmp(formats
[--i
], drv
->format_name
);
6255 formats
= g_renew(const char *, formats
, count
+ 1);
6256 formats
[count
++] = drv
->format_name
;
6261 for (i
= 0; i
< (int)ARRAY_SIZE(block_driver_modules
); i
++) {
6262 const char *format_name
= block_driver_modules
[i
].format_name
;
6268 if (use_bdrv_whitelist
&&
6269 !bdrv_format_is_whitelisted(format_name
, read_only
)) {
6273 while (formats
&& j
&& !found
) {
6274 found
= !strcmp(formats
[--j
], format_name
);
6278 formats
= g_renew(const char *, formats
, count
+ 1);
6279 formats
[count
++] = format_name
;
6284 qsort(formats
, count
, sizeof(formats
[0]), qsort_strcmp
);
6286 for (i
= 0; i
< count
; i
++) {
6287 it(opaque
, formats
[i
]);
6293 /* This function is to find a node in the bs graph */
6294 BlockDriverState
*bdrv_find_node(const char *node_name
)
6296 BlockDriverState
*bs
;
6299 GLOBAL_STATE_CODE();
6301 QTAILQ_FOREACH(bs
, &graph_bdrv_states
, node_list
) {
6302 if (!strcmp(node_name
, bs
->node_name
)) {
6309 /* Put this QMP function here so it can access the static graph_bdrv_states. */
6310 BlockDeviceInfoList
*bdrv_named_nodes_list(bool flat
,
6313 BlockDeviceInfoList
*list
;
6314 BlockDriverState
*bs
;
6316 GLOBAL_STATE_CODE();
6317 GRAPH_RDLOCK_GUARD_MAINLOOP();
6320 QTAILQ_FOREACH(bs
, &graph_bdrv_states
, node_list
) {
6321 BlockDeviceInfo
*info
= bdrv_block_device_info(NULL
, bs
, flat
, errp
);
6323 qapi_free_BlockDeviceInfoList(list
);
6326 QAPI_LIST_PREPEND(list
, info
);
6332 typedef struct XDbgBlockGraphConstructor
{
6333 XDbgBlockGraph
*graph
;
6334 GHashTable
*graph_nodes
;
6335 } XDbgBlockGraphConstructor
;
6337 static XDbgBlockGraphConstructor
*xdbg_graph_new(void)
6339 XDbgBlockGraphConstructor
*gr
= g_new(XDbgBlockGraphConstructor
, 1);
6341 gr
->graph
= g_new0(XDbgBlockGraph
, 1);
6342 gr
->graph_nodes
= g_hash_table_new(NULL
, NULL
);
6347 static XDbgBlockGraph
*xdbg_graph_finalize(XDbgBlockGraphConstructor
*gr
)
6349 XDbgBlockGraph
*graph
= gr
->graph
;
6351 g_hash_table_destroy(gr
->graph_nodes
);
6357 static uintptr_t xdbg_graph_node_num(XDbgBlockGraphConstructor
*gr
, void *node
)
6359 uintptr_t ret
= (uintptr_t)g_hash_table_lookup(gr
->graph_nodes
, node
);
6366 * Start counting from 1, not 0, because 0 interferes with not-found (NULL)
6367 * answer of g_hash_table_lookup.
6369 ret
= g_hash_table_size(gr
->graph_nodes
) + 1;
6370 g_hash_table_insert(gr
->graph_nodes
, node
, (void *)ret
);
6375 static void xdbg_graph_add_node(XDbgBlockGraphConstructor
*gr
, void *node
,
6376 XDbgBlockGraphNodeType type
, const char *name
)
6378 XDbgBlockGraphNode
*n
;
6380 n
= g_new0(XDbgBlockGraphNode
, 1);
6382 n
->id
= xdbg_graph_node_num(gr
, node
);
6384 n
->name
= g_strdup(name
);
6386 QAPI_LIST_PREPEND(gr
->graph
->nodes
, n
);
6389 static void xdbg_graph_add_edge(XDbgBlockGraphConstructor
*gr
, void *parent
,
6390 const BdrvChild
*child
)
6392 BlockPermission qapi_perm
;
6393 XDbgBlockGraphEdge
*edge
;
6394 GLOBAL_STATE_CODE();
6396 edge
= g_new0(XDbgBlockGraphEdge
, 1);
6398 edge
->parent
= xdbg_graph_node_num(gr
, parent
);
6399 edge
->child
= xdbg_graph_node_num(gr
, child
->bs
);
6400 edge
->name
= g_strdup(child
->name
);
6402 for (qapi_perm
= 0; qapi_perm
< BLOCK_PERMISSION__MAX
; qapi_perm
++) {
6403 uint64_t flag
= bdrv_qapi_perm_to_blk_perm(qapi_perm
);
6405 if (flag
& child
->perm
) {
6406 QAPI_LIST_PREPEND(edge
->perm
, qapi_perm
);
6408 if (flag
& child
->shared_perm
) {
6409 QAPI_LIST_PREPEND(edge
->shared_perm
, qapi_perm
);
6413 QAPI_LIST_PREPEND(gr
->graph
->edges
, edge
);
6417 XDbgBlockGraph
*bdrv_get_xdbg_block_graph(Error
**errp
)
6421 BlockDriverState
*bs
;
6423 XDbgBlockGraphConstructor
*gr
= xdbg_graph_new();
6425 GLOBAL_STATE_CODE();
6427 for (blk
= blk_all_next(NULL
); blk
; blk
= blk_all_next(blk
)) {
6428 char *allocated_name
= NULL
;
6429 const char *name
= blk_name(blk
);
6432 name
= allocated_name
= blk_get_attached_dev_id(blk
);
6434 xdbg_graph_add_node(gr
, blk
, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_BACKEND
,
6436 g_free(allocated_name
);
6437 if (blk_root(blk
)) {
6438 xdbg_graph_add_edge(gr
, blk
, blk_root(blk
));
6442 WITH_JOB_LOCK_GUARD() {
6443 for (job
= block_job_next_locked(NULL
); job
;
6444 job
= block_job_next_locked(job
)) {
6447 xdbg_graph_add_node(gr
, job
, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_JOB
,
6449 for (el
= job
->nodes
; el
; el
= el
->next
) {
6450 xdbg_graph_add_edge(gr
, job
, (BdrvChild
*)el
->data
);
6455 QTAILQ_FOREACH(bs
, &graph_bdrv_states
, node_list
) {
6456 xdbg_graph_add_node(gr
, bs
, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_DRIVER
,
6458 QLIST_FOREACH(child
, &bs
->children
, next
) {
6459 xdbg_graph_add_edge(gr
, bs
, child
);
6463 return xdbg_graph_finalize(gr
);
6466 BlockDriverState
*bdrv_lookup_bs(const char *device
,
6467 const char *node_name
,
6471 BlockDriverState
*bs
;
6473 GLOBAL_STATE_CODE();
6476 blk
= blk_by_name(device
);
6481 error_setg(errp
, "Device '%s' has no medium", device
);
6489 bs
= bdrv_find_node(node_name
);
6496 error_setg(errp
, "Cannot find device=\'%s\' nor node-name=\'%s\'",
6497 device
? device
: "",
6498 node_name
? node_name
: "");
6502 /* If 'base' is in the same chain as 'top', return true. Otherwise,
6503 * return false. If either argument is NULL, return false. */
6504 bool bdrv_chain_contains(BlockDriverState
*top
, BlockDriverState
*base
)
6507 GLOBAL_STATE_CODE();
6509 while (top
&& top
!= base
) {
6510 top
= bdrv_filter_or_cow_bs(top
);
6516 BlockDriverState
*bdrv_next_node(BlockDriverState
*bs
)
6518 GLOBAL_STATE_CODE();
6520 return QTAILQ_FIRST(&graph_bdrv_states
);
6522 return QTAILQ_NEXT(bs
, node_list
);
6525 BlockDriverState
*bdrv_next_all_states(BlockDriverState
*bs
)
6527 GLOBAL_STATE_CODE();
6529 return QTAILQ_FIRST(&all_bdrv_states
);
6531 return QTAILQ_NEXT(bs
, bs_list
);
6534 const char *bdrv_get_node_name(const BlockDriverState
*bs
)
6537 return bs
->node_name
;
6540 const char *bdrv_get_parent_name(const BlockDriverState
*bs
)
6546 /* If multiple parents have a name, just pick the first one. */
6547 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
6548 if (c
->klass
->get_name
) {
6549 name
= c
->klass
->get_name(c
);
6550 if (name
&& *name
) {
6559 /* TODO check what callers really want: bs->node_name or blk_name() */
6560 const char *bdrv_get_device_name(const BlockDriverState
*bs
)
6563 return bdrv_get_parent_name(bs
) ?: "";
6566 /* This can be used to identify nodes that might not have a device
6567 * name associated. Since node and device names live in the same
6568 * namespace, the result is unambiguous. The exception is if both are
6569 * absent, then this returns an empty (non-null) string. */
6570 const char *bdrv_get_device_or_node_name(const BlockDriverState
*bs
)
6573 return bdrv_get_parent_name(bs
) ?: bs
->node_name
;
6576 int bdrv_get_flags(BlockDriverState
*bs
)
6579 return bs
->open_flags
;
6582 int bdrv_has_zero_init_1(BlockDriverState
*bs
)
6584 GLOBAL_STATE_CODE();
6588 int bdrv_has_zero_init(BlockDriverState
*bs
)
6590 BlockDriverState
*filtered
;
6591 GLOBAL_STATE_CODE();
6597 /* If BS is a copy on write image, it is initialized to
6598 the contents of the base image, which may not be zeroes. */
6599 if (bdrv_cow_child(bs
)) {
6602 if (bs
->drv
->bdrv_has_zero_init
) {
6603 return bs
->drv
->bdrv_has_zero_init(bs
);
6606 filtered
= bdrv_filter_bs(bs
);
6608 return bdrv_has_zero_init(filtered
);
6615 bool bdrv_can_write_zeroes_with_unmap(BlockDriverState
*bs
)
6618 if (!(bs
->open_flags
& BDRV_O_UNMAP
)) {
6622 return bs
->supported_zero_flags
& BDRV_REQ_MAY_UNMAP
;
6625 void bdrv_get_backing_filename(BlockDriverState
*bs
,
6626 char *filename
, int filename_size
)
6629 pstrcpy(filename
, filename_size
, bs
->backing_file
);
6632 int coroutine_fn
bdrv_co_get_info(BlockDriverState
*bs
, BlockDriverInfo
*bdi
)
6635 BlockDriver
*drv
= bs
->drv
;
6637 assert_bdrv_graph_readable();
6639 /* if bs->drv == NULL, bs is closed, so there's nothing to do here */
6643 if (!drv
->bdrv_co_get_info
) {
6644 BlockDriverState
*filtered
= bdrv_filter_bs(bs
);
6646 return bdrv_co_get_info(filtered
, bdi
);
6650 memset(bdi
, 0, sizeof(*bdi
));
6651 ret
= drv
->bdrv_co_get_info(bs
, bdi
);
6652 if (bdi
->subcluster_size
== 0) {
6654 * If the driver left this unset, subclusters are not supported.
6655 * Then it is safe to treat each cluster as having only one subcluster.
6657 bdi
->subcluster_size
= bdi
->cluster_size
;
6663 if (bdi
->cluster_size
> BDRV_MAX_ALIGNMENT
) {
6670 ImageInfoSpecific
*bdrv_get_specific_info(BlockDriverState
*bs
,
6673 BlockDriver
*drv
= bs
->drv
;
6675 if (drv
&& drv
->bdrv_get_specific_info
) {
6676 return drv
->bdrv_get_specific_info(bs
, errp
);
6681 BlockStatsSpecific
*bdrv_get_specific_stats(BlockDriverState
*bs
)
6683 BlockDriver
*drv
= bs
->drv
;
6685 if (!drv
|| !drv
->bdrv_get_specific_stats
) {
6688 return drv
->bdrv_get_specific_stats(bs
);
6691 void coroutine_fn
bdrv_co_debug_event(BlockDriverState
*bs
, BlkdebugEvent event
)
6694 assert_bdrv_graph_readable();
6696 if (!bs
|| !bs
->drv
|| !bs
->drv
->bdrv_co_debug_event
) {
6700 bs
->drv
->bdrv_co_debug_event(bs
, event
);
6703 static BlockDriverState
* GRAPH_RDLOCK
6704 bdrv_find_debug_node(BlockDriverState
*bs
)
6706 GLOBAL_STATE_CODE();
6707 while (bs
&& bs
->drv
&& !bs
->drv
->bdrv_debug_breakpoint
) {
6708 bs
= bdrv_primary_bs(bs
);
6711 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_breakpoint
) {
6712 assert(bs
->drv
->bdrv_debug_remove_breakpoint
);
6719 int bdrv_debug_breakpoint(BlockDriverState
*bs
, const char *event
,
6722 GLOBAL_STATE_CODE();
6723 GRAPH_RDLOCK_GUARD_MAINLOOP();
6725 bs
= bdrv_find_debug_node(bs
);
6727 return bs
->drv
->bdrv_debug_breakpoint(bs
, event
, tag
);
6733 int bdrv_debug_remove_breakpoint(BlockDriverState
*bs
, const char *tag
)
6735 GLOBAL_STATE_CODE();
6736 GRAPH_RDLOCK_GUARD_MAINLOOP();
6738 bs
= bdrv_find_debug_node(bs
);
6740 return bs
->drv
->bdrv_debug_remove_breakpoint(bs
, tag
);
6746 int bdrv_debug_resume(BlockDriverState
*bs
, const char *tag
)
6748 GLOBAL_STATE_CODE();
6749 GRAPH_RDLOCK_GUARD_MAINLOOP();
6751 while (bs
&& (!bs
->drv
|| !bs
->drv
->bdrv_debug_resume
)) {
6752 bs
= bdrv_primary_bs(bs
);
6755 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_resume
) {
6756 return bs
->drv
->bdrv_debug_resume(bs
, tag
);
6762 bool bdrv_debug_is_suspended(BlockDriverState
*bs
, const char *tag
)
6764 GLOBAL_STATE_CODE();
6765 GRAPH_RDLOCK_GUARD_MAINLOOP();
6767 while (bs
&& bs
->drv
&& !bs
->drv
->bdrv_debug_is_suspended
) {
6768 bs
= bdrv_primary_bs(bs
);
6771 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_is_suspended
) {
6772 return bs
->drv
->bdrv_debug_is_suspended(bs
, tag
);
6778 /* backing_file can either be relative, or absolute, or a protocol. If it is
6779 * relative, it must be relative to the chain. So, passing in bs->filename
6780 * from a BDS as backing_file should not be done, as that may be relative to
6781 * the CWD rather than the chain. */
6782 BlockDriverState
*bdrv_find_backing_image(BlockDriverState
*bs
,
6783 const char *backing_file
)
6785 char *filename_full
= NULL
;
6786 char *backing_file_full
= NULL
;
6787 char *filename_tmp
= NULL
;
6788 int is_protocol
= 0;
6789 bool filenames_refreshed
= false;
6790 BlockDriverState
*curr_bs
= NULL
;
6791 BlockDriverState
*retval
= NULL
;
6792 BlockDriverState
*bs_below
;
6794 GLOBAL_STATE_CODE();
6795 GRAPH_RDLOCK_GUARD_MAINLOOP();
6797 if (!bs
|| !bs
->drv
|| !backing_file
) {
6801 filename_full
= g_malloc(PATH_MAX
);
6802 backing_file_full
= g_malloc(PATH_MAX
);
6804 is_protocol
= path_has_protocol(backing_file
);
6807 * Being largely a legacy function, skip any filters here
6808 * (because filters do not have normal filenames, so they cannot
6809 * match anyway; and allowing json:{} filenames is a bit out of
6812 for (curr_bs
= bdrv_skip_filters(bs
);
6813 bdrv_cow_child(curr_bs
) != NULL
;
6816 bs_below
= bdrv_backing_chain_next(curr_bs
);
6818 if (bdrv_backing_overridden(curr_bs
)) {
6820 * If the backing file was overridden, we can only compare
6821 * directly against the backing node's filename.
6824 if (!filenames_refreshed
) {
6826 * This will automatically refresh all of the
6827 * filenames in the rest of the backing chain, so we
6828 * only need to do this once.
6830 bdrv_refresh_filename(bs_below
);
6831 filenames_refreshed
= true;
6834 if (strcmp(backing_file
, bs_below
->filename
) == 0) {
6838 } else if (is_protocol
|| path_has_protocol(curr_bs
->backing_file
)) {
6840 * If either of the filename paths is actually a protocol, then
6841 * compare unmodified paths; otherwise make paths relative.
6843 char *backing_file_full_ret
;
6845 if (strcmp(backing_file
, curr_bs
->backing_file
) == 0) {
6849 /* Also check against the full backing filename for the image */
6850 backing_file_full_ret
= bdrv_get_full_backing_filename(curr_bs
,
6852 if (backing_file_full_ret
) {
6853 bool equal
= strcmp(backing_file
, backing_file_full_ret
) == 0;
6854 g_free(backing_file_full_ret
);
6861 /* If not an absolute filename path, make it relative to the current
6862 * image's filename path */
6863 filename_tmp
= bdrv_make_absolute_filename(curr_bs
, backing_file
,
6865 /* We are going to compare canonicalized absolute pathnames */
6866 if (!filename_tmp
|| !realpath(filename_tmp
, filename_full
)) {
6867 g_free(filename_tmp
);
6870 g_free(filename_tmp
);
6872 /* We need to make sure the backing filename we are comparing against
6873 * is relative to the current image filename (or absolute) */
6874 filename_tmp
= bdrv_get_full_backing_filename(curr_bs
, NULL
);
6875 if (!filename_tmp
|| !realpath(filename_tmp
, backing_file_full
)) {
6876 g_free(filename_tmp
);
6879 g_free(filename_tmp
);
6881 if (strcmp(backing_file_full
, filename_full
) == 0) {
6888 g_free(filename_full
);
6889 g_free(backing_file_full
);
6893 void bdrv_init(void)
6895 #ifdef CONFIG_BDRV_WHITELIST_TOOLS
6896 use_bdrv_whitelist
= 1;
6898 module_call_init(MODULE_INIT_BLOCK
);
6901 void bdrv_init_with_whitelist(void)
6903 use_bdrv_whitelist
= 1;
6907 int bdrv_activate(BlockDriverState
*bs
, Error
**errp
)
6909 BdrvChild
*child
, *parent
;
6910 Error
*local_err
= NULL
;
6912 BdrvDirtyBitmap
*bm
;
6914 GLOBAL_STATE_CODE();
6915 GRAPH_RDLOCK_GUARD_MAINLOOP();
6921 QLIST_FOREACH(child
, &bs
->children
, next
) {
6922 bdrv_activate(child
->bs
, &local_err
);
6924 error_propagate(errp
, local_err
);
6930 * Update permissions, they may differ for inactive nodes.
6932 * Note that the required permissions of inactive images are always a
6933 * subset of the permissions required after activating the image. This
6934 * allows us to just get the permissions upfront without restricting
6935 * bdrv_co_invalidate_cache().
6937 * It also means that in error cases, we don't have to try and revert to
6938 * the old permissions (which is an operation that could fail, too). We can
6939 * just keep the extended permissions for the next time that an activation
6940 * of the image is tried.
6942 if (bs
->open_flags
& BDRV_O_INACTIVE
) {
6943 bs
->open_flags
&= ~BDRV_O_INACTIVE
;
6944 ret
= bdrv_refresh_perms(bs
, NULL
, errp
);
6946 bs
->open_flags
|= BDRV_O_INACTIVE
;
6950 ret
= bdrv_invalidate_cache(bs
, errp
);
6952 bs
->open_flags
|= BDRV_O_INACTIVE
;
6956 FOR_EACH_DIRTY_BITMAP(bs
, bm
) {
6957 bdrv_dirty_bitmap_skip_store(bm
, false);
6960 ret
= bdrv_refresh_total_sectors(bs
, bs
->total_sectors
);
6962 bs
->open_flags
|= BDRV_O_INACTIVE
;
6963 error_setg_errno(errp
, -ret
, "Could not refresh total sector count");
6968 QLIST_FOREACH(parent
, &bs
->parents
, next_parent
) {
6969 if (parent
->klass
->activate
) {
6970 parent
->klass
->activate(parent
, &local_err
);
6972 bs
->open_flags
|= BDRV_O_INACTIVE
;
6973 error_propagate(errp
, local_err
);
6982 int coroutine_fn
bdrv_co_invalidate_cache(BlockDriverState
*bs
, Error
**errp
)
6984 Error
*local_err
= NULL
;
6987 assert(!(bs
->open_flags
& BDRV_O_INACTIVE
));
6988 assert_bdrv_graph_readable();
6990 if (bs
->drv
->bdrv_co_invalidate_cache
) {
6991 bs
->drv
->bdrv_co_invalidate_cache(bs
, &local_err
);
6993 error_propagate(errp
, local_err
);
7001 void bdrv_activate_all(Error
**errp
)
7003 BlockDriverState
*bs
;
7004 BdrvNextIterator it
;
7006 GLOBAL_STATE_CODE();
7007 GRAPH_RDLOCK_GUARD_MAINLOOP();
7009 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
7010 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
7013 aio_context_acquire(aio_context
);
7014 ret
= bdrv_activate(bs
, errp
);
7015 aio_context_release(aio_context
);
7017 bdrv_next_cleanup(&it
);
7023 static bool GRAPH_RDLOCK
7024 bdrv_has_bds_parent(BlockDriverState
*bs
, bool only_active
)
7027 GLOBAL_STATE_CODE();
7029 QLIST_FOREACH(parent
, &bs
->parents
, next_parent
) {
7030 if (parent
->klass
->parent_is_bds
) {
7031 BlockDriverState
*parent_bs
= parent
->opaque
;
7032 if (!only_active
|| !(parent_bs
->open_flags
& BDRV_O_INACTIVE
)) {
7041 static int GRAPH_RDLOCK
bdrv_inactivate_recurse(BlockDriverState
*bs
)
7043 BdrvChild
*child
, *parent
;
7045 uint64_t cumulative_perms
, cumulative_shared_perms
;
7047 GLOBAL_STATE_CODE();
7053 /* Make sure that we don't inactivate a child before its parent.
7054 * It will be covered by recursion from the yet active parent. */
7055 if (bdrv_has_bds_parent(bs
, true)) {
7059 assert(!(bs
->open_flags
& BDRV_O_INACTIVE
));
7061 /* Inactivate this node */
7062 if (bs
->drv
->bdrv_inactivate
) {
7063 ret
= bs
->drv
->bdrv_inactivate(bs
);
7069 QLIST_FOREACH(parent
, &bs
->parents
, next_parent
) {
7070 if (parent
->klass
->inactivate
) {
7071 ret
= parent
->klass
->inactivate(parent
);
7078 bdrv_get_cumulative_perm(bs
, &cumulative_perms
,
7079 &cumulative_shared_perms
);
7080 if (cumulative_perms
& (BLK_PERM_WRITE
| BLK_PERM_WRITE_UNCHANGED
)) {
7081 /* Our inactive parents still need write access. Inactivation failed. */
7085 bs
->open_flags
|= BDRV_O_INACTIVE
;
7088 * Update permissions, they may differ for inactive nodes.
7089 * We only tried to loosen restrictions, so errors are not fatal, ignore
7092 bdrv_refresh_perms(bs
, NULL
, NULL
);
7094 /* Recursively inactivate children */
7095 QLIST_FOREACH(child
, &bs
->children
, next
) {
7096 ret
= bdrv_inactivate_recurse(child
->bs
);
7105 int bdrv_inactivate_all(void)
7107 BlockDriverState
*bs
= NULL
;
7108 BdrvNextIterator it
;
7110 GSList
*aio_ctxs
= NULL
, *ctx
;
7112 GLOBAL_STATE_CODE();
7113 GRAPH_RDLOCK_GUARD_MAINLOOP();
7115 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
7116 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
7118 if (!g_slist_find(aio_ctxs
, aio_context
)) {
7119 aio_ctxs
= g_slist_prepend(aio_ctxs
, aio_context
);
7120 aio_context_acquire(aio_context
);
7124 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
7125 /* Nodes with BDS parents are covered by recursion from the last
7126 * parent that gets inactivated. Don't inactivate them a second
7127 * time if that has already happened. */
7128 if (bdrv_has_bds_parent(bs
, false)) {
7131 ret
= bdrv_inactivate_recurse(bs
);
7133 bdrv_next_cleanup(&it
);
7139 for (ctx
= aio_ctxs
; ctx
!= NULL
; ctx
= ctx
->next
) {
7140 AioContext
*aio_context
= ctx
->data
;
7141 aio_context_release(aio_context
);
7143 g_slist_free(aio_ctxs
);
7148 /**************************************************************/
7149 /* removable device support */
7152 * Return TRUE if the media is present
7154 bool coroutine_fn
bdrv_co_is_inserted(BlockDriverState
*bs
)
7156 BlockDriver
*drv
= bs
->drv
;
7159 assert_bdrv_graph_readable();
7164 if (drv
->bdrv_co_is_inserted
) {
7165 return drv
->bdrv_co_is_inserted(bs
);
7167 QLIST_FOREACH(child
, &bs
->children
, next
) {
7168 if (!bdrv_co_is_inserted(child
->bs
)) {
7176 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
7178 void coroutine_fn
bdrv_co_eject(BlockDriverState
*bs
, bool eject_flag
)
7180 BlockDriver
*drv
= bs
->drv
;
7182 assert_bdrv_graph_readable();
7184 if (drv
&& drv
->bdrv_co_eject
) {
7185 drv
->bdrv_co_eject(bs
, eject_flag
);
7190 * Lock or unlock the media (if it is locked, the user won't be able
7191 * to eject it manually).
7193 void coroutine_fn
bdrv_co_lock_medium(BlockDriverState
*bs
, bool locked
)
7195 BlockDriver
*drv
= bs
->drv
;
7197 assert_bdrv_graph_readable();
7198 trace_bdrv_lock_medium(bs
, locked
);
7200 if (drv
&& drv
->bdrv_co_lock_medium
) {
7201 drv
->bdrv_co_lock_medium(bs
, locked
);
7205 /* Get a reference to bs */
7206 void bdrv_ref(BlockDriverState
*bs
)
7208 GLOBAL_STATE_CODE();
7212 /* Release a previously grabbed reference to bs.
7213 * If after releasing, reference count is zero, the BlockDriverState is
7215 void bdrv_unref(BlockDriverState
*bs
)
7217 GLOBAL_STATE_CODE();
7221 assert(bs
->refcnt
> 0);
7222 if (--bs
->refcnt
== 0) {
7228 * Release a BlockDriverState reference while holding the graph write lock.
7230 * Calling bdrv_unref() directly is forbidden while holding the graph lock
7231 * because bdrv_close() both involves polling and taking the graph lock
7232 * internally. bdrv_schedule_unref() instead delays decreasing the refcount and
7233 * possibly closing @bs until the graph lock is released.
7235 void bdrv_schedule_unref(BlockDriverState
*bs
)
7240 aio_bh_schedule_oneshot(qemu_get_aio_context(),
7241 (QEMUBHFunc
*) bdrv_unref
, bs
);
7244 struct BdrvOpBlocker
{
7246 QLIST_ENTRY(BdrvOpBlocker
) list
;
7249 bool bdrv_op_is_blocked(BlockDriverState
*bs
, BlockOpType op
, Error
**errp
)
7251 BdrvOpBlocker
*blocker
;
7252 GLOBAL_STATE_CODE();
7253 GRAPH_RDLOCK_GUARD_MAINLOOP();
7255 assert((int) op
>= 0 && op
< BLOCK_OP_TYPE_MAX
);
7256 if (!QLIST_EMPTY(&bs
->op_blockers
[op
])) {
7257 blocker
= QLIST_FIRST(&bs
->op_blockers
[op
]);
7258 error_propagate_prepend(errp
, error_copy(blocker
->reason
),
7259 "Node '%s' is busy: ",
7260 bdrv_get_device_or_node_name(bs
));
7266 void bdrv_op_block(BlockDriverState
*bs
, BlockOpType op
, Error
*reason
)
7268 BdrvOpBlocker
*blocker
;
7269 GLOBAL_STATE_CODE();
7270 assert((int) op
>= 0 && op
< BLOCK_OP_TYPE_MAX
);
7272 blocker
= g_new0(BdrvOpBlocker
, 1);
7273 blocker
->reason
= reason
;
7274 QLIST_INSERT_HEAD(&bs
->op_blockers
[op
], blocker
, list
);
7277 void bdrv_op_unblock(BlockDriverState
*bs
, BlockOpType op
, Error
*reason
)
7279 BdrvOpBlocker
*blocker
, *next
;
7280 GLOBAL_STATE_CODE();
7281 assert((int) op
>= 0 && op
< BLOCK_OP_TYPE_MAX
);
7282 QLIST_FOREACH_SAFE(blocker
, &bs
->op_blockers
[op
], list
, next
) {
7283 if (blocker
->reason
== reason
) {
7284 QLIST_REMOVE(blocker
, list
);
7290 void bdrv_op_block_all(BlockDriverState
*bs
, Error
*reason
)
7293 GLOBAL_STATE_CODE();
7294 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
7295 bdrv_op_block(bs
, i
, reason
);
7299 void bdrv_op_unblock_all(BlockDriverState
*bs
, Error
*reason
)
7302 GLOBAL_STATE_CODE();
7303 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
7304 bdrv_op_unblock(bs
, i
, reason
);
7308 bool bdrv_op_blocker_is_empty(BlockDriverState
*bs
)
7311 GLOBAL_STATE_CODE();
7312 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
7313 if (!QLIST_EMPTY(&bs
->op_blockers
[i
])) {
7321 * Must not be called while holding the lock of an AioContext other than the
7324 void bdrv_img_create(const char *filename
, const char *fmt
,
7325 const char *base_filename
, const char *base_fmt
,
7326 char *options
, uint64_t img_size
, int flags
, bool quiet
,
7329 QemuOptsList
*create_opts
= NULL
;
7330 QemuOpts
*opts
= NULL
;
7331 const char *backing_fmt
, *backing_file
;
7333 BlockDriver
*drv
, *proto_drv
;
7334 Error
*local_err
= NULL
;
7337 GLOBAL_STATE_CODE();
7339 /* Find driver and parse its options */
7340 drv
= bdrv_find_format(fmt
);
7342 error_setg(errp
, "Unknown file format '%s'", fmt
);
7346 proto_drv
= bdrv_find_protocol(filename
, true, errp
);
7351 if (!drv
->create_opts
) {
7352 error_setg(errp
, "Format driver '%s' does not support image creation",
7357 if (!proto_drv
->create_opts
) {
7358 error_setg(errp
, "Protocol driver '%s' does not support image creation",
7359 proto_drv
->format_name
);
7363 aio_context_acquire(qemu_get_aio_context());
7365 /* Create parameter list */
7366 create_opts
= qemu_opts_append(create_opts
, drv
->create_opts
);
7367 create_opts
= qemu_opts_append(create_opts
, proto_drv
->create_opts
);
7369 opts
= qemu_opts_create(create_opts
, NULL
, 0, &error_abort
);
7371 /* Parse -o options */
7373 if (!qemu_opts_do_parse(opts
, options
, NULL
, errp
)) {
7378 if (!qemu_opt_get(opts
, BLOCK_OPT_SIZE
)) {
7379 qemu_opt_set_number(opts
, BLOCK_OPT_SIZE
, img_size
, &error_abort
);
7380 } else if (img_size
!= UINT64_C(-1)) {
7381 error_setg(errp
, "The image size must be specified only once");
7385 if (base_filename
) {
7386 if (!qemu_opt_set(opts
, BLOCK_OPT_BACKING_FILE
, base_filename
,
7388 error_setg(errp
, "Backing file not supported for file format '%s'",
7395 if (!qemu_opt_set(opts
, BLOCK_OPT_BACKING_FMT
, base_fmt
, NULL
)) {
7396 error_setg(errp
, "Backing file format not supported for file "
7397 "format '%s'", fmt
);
7402 backing_file
= qemu_opt_get(opts
, BLOCK_OPT_BACKING_FILE
);
7404 if (!strcmp(filename
, backing_file
)) {
7405 error_setg(errp
, "Error: Trying to create an image with the "
7406 "same filename as the backing file");
7409 if (backing_file
[0] == '\0') {
7410 error_setg(errp
, "Expected backing file name, got empty string");
7415 backing_fmt
= qemu_opt_get(opts
, BLOCK_OPT_BACKING_FMT
);
7417 /* The size for the image must always be specified, unless we have a backing
7418 * file and we have not been forbidden from opening it. */
7419 size
= qemu_opt_get_size(opts
, BLOCK_OPT_SIZE
, img_size
);
7420 if (backing_file
&& !(flags
& BDRV_O_NO_BACKING
)) {
7421 BlockDriverState
*bs
;
7424 QDict
*backing_options
= NULL
;
7427 bdrv_get_full_backing_filename_from_filename(filename
, backing_file
,
7432 assert(full_backing
);
7435 * No need to do I/O here, which allows us to open encrypted
7436 * backing images without needing the secret
7439 back_flags
&= ~(BDRV_O_RDWR
| BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
);
7440 back_flags
|= BDRV_O_NO_IO
;
7442 backing_options
= qdict_new();
7444 qdict_put_str(backing_options
, "driver", backing_fmt
);
7446 qdict_put_bool(backing_options
, BDRV_OPT_FORCE_SHARE
, true);
7448 bs
= bdrv_open(full_backing
, NULL
, backing_options
, back_flags
,
7450 g_free(full_backing
);
7452 error_append_hint(&local_err
, "Could not open backing image.\n");
7456 error_setg(&local_err
,
7457 "Backing file specified without backing format");
7458 error_append_hint(&local_err
, "Detected format of %s.\n",
7459 bs
->drv
->format_name
);
7463 /* Opened BS, have no size */
7464 size
= bdrv_getlength(bs
);
7466 error_setg_errno(errp
, -size
, "Could not get size of '%s'",
7471 qemu_opt_set_number(opts
, BLOCK_OPT_SIZE
, size
, &error_abort
);
7475 /* (backing_file && !(flags & BDRV_O_NO_BACKING)) */
7476 } else if (backing_file
&& !backing_fmt
) {
7477 error_setg(&local_err
,
7478 "Backing file specified without backing format");
7483 error_setg(errp
, "Image creation needs a size parameter");
7488 printf("Formatting '%s', fmt=%s ", filename
, fmt
);
7489 qemu_opts_print(opts
, " ");
7494 ret
= bdrv_create(drv
, filename
, opts
, &local_err
);
7496 if (ret
== -EFBIG
) {
7497 /* This is generally a better message than whatever the driver would
7498 * deliver (especially because of the cluster_size_hint), since that
7499 * is most probably not much different from "image too large". */
7500 const char *cluster_size_hint
= "";
7501 if (qemu_opt_get_size(opts
, BLOCK_OPT_CLUSTER_SIZE
, 0)) {
7502 cluster_size_hint
= " (try using a larger cluster size)";
7504 error_setg(errp
, "The image size is too large for file format '%s'"
7505 "%s", fmt
, cluster_size_hint
);
7506 error_free(local_err
);
7511 qemu_opts_del(opts
);
7512 qemu_opts_free(create_opts
);
7513 error_propagate(errp
, local_err
);
7514 aio_context_release(qemu_get_aio_context());
7517 AioContext
*bdrv_get_aio_context(BlockDriverState
*bs
)
7520 return bs
? bs
->aio_context
: qemu_get_aio_context();
7523 AioContext
*coroutine_fn
bdrv_co_enter(BlockDriverState
*bs
)
7525 Coroutine
*self
= qemu_coroutine_self();
7526 AioContext
*old_ctx
= qemu_coroutine_get_aio_context(self
);
7527 AioContext
*new_ctx
;
7531 * Increase bs->in_flight to ensure that this operation is completed before
7532 * moving the node to a different AioContext. Read new_ctx only afterwards.
7534 bdrv_inc_in_flight(bs
);
7536 new_ctx
= bdrv_get_aio_context(bs
);
7537 aio_co_reschedule_self(new_ctx
);
7541 void coroutine_fn
bdrv_co_leave(BlockDriverState
*bs
, AioContext
*old_ctx
)
7544 aio_co_reschedule_self(old_ctx
);
7545 bdrv_dec_in_flight(bs
);
7548 void coroutine_fn
bdrv_co_lock(BlockDriverState
*bs
)
7550 AioContext
*ctx
= bdrv_get_aio_context(bs
);
7552 /* In the main thread, bs->aio_context won't change concurrently */
7553 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
7556 * We're in coroutine context, so we already hold the lock of the main
7557 * loop AioContext. Don't lock it twice to avoid deadlocks.
7559 assert(qemu_in_coroutine());
7560 if (ctx
!= qemu_get_aio_context()) {
7561 aio_context_acquire(ctx
);
7565 void coroutine_fn
bdrv_co_unlock(BlockDriverState
*bs
)
7567 AioContext
*ctx
= bdrv_get_aio_context(bs
);
7569 assert(qemu_in_coroutine());
7570 if (ctx
!= qemu_get_aio_context()) {
7571 aio_context_release(ctx
);
7575 static void bdrv_do_remove_aio_context_notifier(BdrvAioNotifier
*ban
)
7577 GLOBAL_STATE_CODE();
7578 QLIST_REMOVE(ban
, list
);
7582 static void bdrv_detach_aio_context(BlockDriverState
*bs
)
7584 BdrvAioNotifier
*baf
, *baf_tmp
;
7586 assert(!bs
->walking_aio_notifiers
);
7587 GLOBAL_STATE_CODE();
7588 bs
->walking_aio_notifiers
= true;
7589 QLIST_FOREACH_SAFE(baf
, &bs
->aio_notifiers
, list
, baf_tmp
) {
7591 bdrv_do_remove_aio_context_notifier(baf
);
7593 baf
->detach_aio_context(baf
->opaque
);
7596 /* Never mind iterating again to check for ->deleted. bdrv_close() will
7597 * remove remaining aio notifiers if we aren't called again.
7599 bs
->walking_aio_notifiers
= false;
7601 if (bs
->drv
&& bs
->drv
->bdrv_detach_aio_context
) {
7602 bs
->drv
->bdrv_detach_aio_context(bs
);
7605 bs
->aio_context
= NULL
;
7608 static void bdrv_attach_aio_context(BlockDriverState
*bs
,
7609 AioContext
*new_context
)
7611 BdrvAioNotifier
*ban
, *ban_tmp
;
7612 GLOBAL_STATE_CODE();
7614 bs
->aio_context
= new_context
;
7616 if (bs
->drv
&& bs
->drv
->bdrv_attach_aio_context
) {
7617 bs
->drv
->bdrv_attach_aio_context(bs
, new_context
);
7620 assert(!bs
->walking_aio_notifiers
);
7621 bs
->walking_aio_notifiers
= true;
7622 QLIST_FOREACH_SAFE(ban
, &bs
->aio_notifiers
, list
, ban_tmp
) {
7624 bdrv_do_remove_aio_context_notifier(ban
);
7626 ban
->attached_aio_context(new_context
, ban
->opaque
);
7629 bs
->walking_aio_notifiers
= false;
7632 typedef struct BdrvStateSetAioContext
{
7633 AioContext
*new_ctx
;
7634 BlockDriverState
*bs
;
7635 } BdrvStateSetAioContext
;
7637 static bool bdrv_parent_change_aio_context(BdrvChild
*c
, AioContext
*ctx
,
7638 GHashTable
*visited
,
7642 GLOBAL_STATE_CODE();
7643 if (g_hash_table_contains(visited
, c
)) {
7646 g_hash_table_add(visited
, c
);
7649 * A BdrvChildClass that doesn't handle AioContext changes cannot
7650 * tolerate any AioContext changes
7652 if (!c
->klass
->change_aio_ctx
) {
7653 char *user
= bdrv_child_user_desc(c
);
7654 error_setg(errp
, "Changing iothreads is not supported by %s", user
);
7658 if (!c
->klass
->change_aio_ctx(c
, ctx
, visited
, tran
, errp
)) {
7659 assert(!errp
|| *errp
);
7665 bool bdrv_child_change_aio_context(BdrvChild
*c
, AioContext
*ctx
,
7666 GHashTable
*visited
, Transaction
*tran
,
7669 GLOBAL_STATE_CODE();
7670 if (g_hash_table_contains(visited
, c
)) {
7673 g_hash_table_add(visited
, c
);
7674 return bdrv_change_aio_context(c
->bs
, ctx
, visited
, tran
, errp
);
7677 static void bdrv_set_aio_context_clean(void *opaque
)
7679 BdrvStateSetAioContext
*state
= (BdrvStateSetAioContext
*) opaque
;
7680 BlockDriverState
*bs
= (BlockDriverState
*) state
->bs
;
7682 /* Paired with bdrv_drained_begin in bdrv_change_aio_context() */
7683 bdrv_drained_end(bs
);
7688 static void bdrv_set_aio_context_commit(void *opaque
)
7690 BdrvStateSetAioContext
*state
= (BdrvStateSetAioContext
*) opaque
;
7691 BlockDriverState
*bs
= (BlockDriverState
*) state
->bs
;
7692 AioContext
*new_context
= state
->new_ctx
;
7693 AioContext
*old_context
= bdrv_get_aio_context(bs
);
7696 * Take the old AioContex when detaching it from bs.
7697 * At this point, new_context lock is already acquired, and we are now
7698 * also taking old_context. This is safe as long as bdrv_detach_aio_context
7699 * does not call AIO_POLL_WHILE().
7701 if (old_context
!= qemu_get_aio_context()) {
7702 aio_context_acquire(old_context
);
7704 bdrv_detach_aio_context(bs
);
7705 if (old_context
!= qemu_get_aio_context()) {
7706 aio_context_release(old_context
);
7708 bdrv_attach_aio_context(bs
, new_context
);
7711 static TransactionActionDrv set_aio_context
= {
7712 .commit
= bdrv_set_aio_context_commit
,
7713 .clean
= bdrv_set_aio_context_clean
,
7717 * Changes the AioContext used for fd handlers, timers, and BHs by this
7718 * BlockDriverState and all its children and parents.
7720 * Must be called from the main AioContext.
7722 * The caller must own the AioContext lock for the old AioContext of bs, but it
7723 * must not own the AioContext lock for new_context (unless new_context is the
7724 * same as the current context of bs).
7726 * @visited will accumulate all visited BdrvChild objects. The caller is
7727 * responsible for freeing the list afterwards.
7729 static bool bdrv_change_aio_context(BlockDriverState
*bs
, AioContext
*ctx
,
7730 GHashTable
*visited
, Transaction
*tran
,
7734 BdrvStateSetAioContext
*state
;
7736 GLOBAL_STATE_CODE();
7738 if (bdrv_get_aio_context(bs
) == ctx
) {
7742 bdrv_graph_rdlock_main_loop();
7743 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
7744 if (!bdrv_parent_change_aio_context(c
, ctx
, visited
, tran
, errp
)) {
7745 bdrv_graph_rdunlock_main_loop();
7750 QLIST_FOREACH(c
, &bs
->children
, next
) {
7751 if (!bdrv_child_change_aio_context(c
, ctx
, visited
, tran
, errp
)) {
7752 bdrv_graph_rdunlock_main_loop();
7756 bdrv_graph_rdunlock_main_loop();
7758 state
= g_new(BdrvStateSetAioContext
, 1);
7759 *state
= (BdrvStateSetAioContext
) {
7764 /* Paired with bdrv_drained_end in bdrv_set_aio_context_clean() */
7765 bdrv_drained_begin(bs
);
7767 tran_add(tran
, &set_aio_context
, state
);
7773 * Change bs's and recursively all of its parents' and children's AioContext
7774 * to the given new context, returning an error if that isn't possible.
7776 * If ignore_child is not NULL, that child (and its subgraph) will not
7779 * This function still requires the caller to take the bs current
7780 * AioContext lock, otherwise draining will fail since AIO_WAIT_WHILE
7781 * assumes the lock is always held if bs is in another AioContext.
7782 * For the same reason, it temporarily also holds the new AioContext, since
7783 * bdrv_drained_end calls BDRV_POLL_WHILE that assumes the lock is taken too.
7784 * Therefore the new AioContext lock must not be taken by the caller.
7786 int bdrv_try_change_aio_context(BlockDriverState
*bs
, AioContext
*ctx
,
7787 BdrvChild
*ignore_child
, Error
**errp
)
7790 GHashTable
*visited
;
7792 AioContext
*old_context
= bdrv_get_aio_context(bs
);
7793 GLOBAL_STATE_CODE();
7796 * Recursion phase: go through all nodes of the graph.
7797 * Take care of checking that all nodes support changing AioContext
7798 * and drain them, building a linear list of callbacks to run if everything
7799 * is successful (the transaction itself).
7802 visited
= g_hash_table_new(NULL
, NULL
);
7804 g_hash_table_add(visited
, ignore_child
);
7806 ret
= bdrv_change_aio_context(bs
, ctx
, visited
, tran
, errp
);
7807 g_hash_table_destroy(visited
);
7810 * Linear phase: go through all callbacks collected in the transaction.
7811 * Run all callbacks collected in the recursion to switch all nodes
7812 * AioContext lock (transaction commit), or undo all changes done in the
7813 * recursion (transaction abort).
7817 /* Just run clean() callbacks. No AioContext changed. */
7823 * Release old AioContext, it won't be needed anymore, as all
7824 * bdrv_drained_begin() have been called already.
7826 if (qemu_get_aio_context() != old_context
) {
7827 aio_context_release(old_context
);
7831 * Acquire new AioContext since bdrv_drained_end() is going to be called
7832 * after we switched all nodes in the new AioContext, and the function
7833 * assumes that the lock of the bs is always taken.
7835 if (qemu_get_aio_context() != ctx
) {
7836 aio_context_acquire(ctx
);
7841 if (qemu_get_aio_context() != ctx
) {
7842 aio_context_release(ctx
);
7845 /* Re-acquire the old AioContext, since the caller takes and releases it. */
7846 if (qemu_get_aio_context() != old_context
) {
7847 aio_context_acquire(old_context
);
7853 void bdrv_add_aio_context_notifier(BlockDriverState
*bs
,
7854 void (*attached_aio_context
)(AioContext
*new_context
, void *opaque
),
7855 void (*detach_aio_context
)(void *opaque
), void *opaque
)
7857 BdrvAioNotifier
*ban
= g_new(BdrvAioNotifier
, 1);
7858 *ban
= (BdrvAioNotifier
){
7859 .attached_aio_context
= attached_aio_context
,
7860 .detach_aio_context
= detach_aio_context
,
7863 GLOBAL_STATE_CODE();
7865 QLIST_INSERT_HEAD(&bs
->aio_notifiers
, ban
, list
);
7868 void bdrv_remove_aio_context_notifier(BlockDriverState
*bs
,
7869 void (*attached_aio_context
)(AioContext
*,
7871 void (*detach_aio_context
)(void *),
7874 BdrvAioNotifier
*ban
, *ban_next
;
7875 GLOBAL_STATE_CODE();
7877 QLIST_FOREACH_SAFE(ban
, &bs
->aio_notifiers
, list
, ban_next
) {
7878 if (ban
->attached_aio_context
== attached_aio_context
&&
7879 ban
->detach_aio_context
== detach_aio_context
&&
7880 ban
->opaque
== opaque
&&
7881 ban
->deleted
== false)
7883 if (bs
->walking_aio_notifiers
) {
7884 ban
->deleted
= true;
7886 bdrv_do_remove_aio_context_notifier(ban
);
7895 int bdrv_amend_options(BlockDriverState
*bs
, QemuOpts
*opts
,
7896 BlockDriverAmendStatusCB
*status_cb
, void *cb_opaque
,
7900 GLOBAL_STATE_CODE();
7902 error_setg(errp
, "Node is ejected");
7905 if (!bs
->drv
->bdrv_amend_options
) {
7906 error_setg(errp
, "Block driver '%s' does not support option amendment",
7907 bs
->drv
->format_name
);
7910 return bs
->drv
->bdrv_amend_options(bs
, opts
, status_cb
,
7911 cb_opaque
, force
, errp
);
7915 * This function checks whether the given @to_replace is allowed to be
7916 * replaced by a node that always shows the same data as @bs. This is
7917 * used for example to verify whether the mirror job can replace
7918 * @to_replace by the target mirrored from @bs.
7919 * To be replaceable, @bs and @to_replace may either be guaranteed to
7920 * always show the same data (because they are only connected through
7921 * filters), or some driver may allow replacing one of its children
7922 * because it can guarantee that this child's data is not visible at
7923 * all (for example, for dissenting quorum children that have no other
7926 bool bdrv_recurse_can_replace(BlockDriverState
*bs
,
7927 BlockDriverState
*to_replace
)
7929 BlockDriverState
*filtered
;
7931 GLOBAL_STATE_CODE();
7933 if (!bs
|| !bs
->drv
) {
7937 if (bs
== to_replace
) {
7941 /* See what the driver can do */
7942 if (bs
->drv
->bdrv_recurse_can_replace
) {
7943 return bs
->drv
->bdrv_recurse_can_replace(bs
, to_replace
);
7946 /* For filters without an own implementation, we can recurse on our own */
7947 filtered
= bdrv_filter_bs(bs
);
7949 return bdrv_recurse_can_replace(filtered
, to_replace
);
7957 * Check whether the given @node_name can be replaced by a node that
7958 * has the same data as @parent_bs. If so, return @node_name's BDS;
7961 * @node_name must be a (recursive) *child of @parent_bs (or this
7962 * function will return NULL).
7964 * The result (whether the node can be replaced or not) is only valid
7965 * for as long as no graph or permission changes occur.
7967 BlockDriverState
*check_to_replace_node(BlockDriverState
*parent_bs
,
7968 const char *node_name
, Error
**errp
)
7970 BlockDriverState
*to_replace_bs
= bdrv_find_node(node_name
);
7971 AioContext
*aio_context
;
7973 GLOBAL_STATE_CODE();
7975 if (!to_replace_bs
) {
7976 error_setg(errp
, "Failed to find node with node-name='%s'", node_name
);
7980 aio_context
= bdrv_get_aio_context(to_replace_bs
);
7981 aio_context_acquire(aio_context
);
7983 if (bdrv_op_is_blocked(to_replace_bs
, BLOCK_OP_TYPE_REPLACE
, errp
)) {
7984 to_replace_bs
= NULL
;
7988 /* We don't want arbitrary node of the BDS chain to be replaced only the top
7989 * most non filter in order to prevent data corruption.
7990 * Another benefit is that this tests exclude backing files which are
7991 * blocked by the backing blockers.
7993 if (!bdrv_recurse_can_replace(parent_bs
, to_replace_bs
)) {
7994 error_setg(errp
, "Cannot replace '%s' by a node mirrored from '%s', "
7995 "because it cannot be guaranteed that doing so would not "
7996 "lead to an abrupt change of visible data",
7997 node_name
, parent_bs
->node_name
);
7998 to_replace_bs
= NULL
;
8003 aio_context_release(aio_context
);
8004 return to_replace_bs
;
8008 * Iterates through the list of runtime option keys that are said to
8009 * be "strong" for a BDS. An option is called "strong" if it changes
8010 * a BDS's data. For example, the null block driver's "size" and
8011 * "read-zeroes" options are strong, but its "latency-ns" option is
8014 * If a key returned by this function ends with a dot, all options
8015 * starting with that prefix are strong.
8017 static const char *const *strong_options(BlockDriverState
*bs
,
8018 const char *const *curopt
)
8020 static const char *const global_options
[] = {
8021 "driver", "filename", NULL
8025 return &global_options
[0];
8029 if (curopt
== &global_options
[ARRAY_SIZE(global_options
) - 1] && bs
->drv
) {
8030 curopt
= bs
->drv
->strong_runtime_opts
;
8033 return (curopt
&& *curopt
) ? curopt
: NULL
;
8037 * Copies all strong runtime options from bs->options to the given
8038 * QDict. The set of strong option keys is determined by invoking
8041 * Returns true iff any strong option was present in bs->options (and
8042 * thus copied to the target QDict) with the exception of "filename"
8043 * and "driver". The caller is expected to use this value to decide
8044 * whether the existence of strong options prevents the generation of
8047 static bool append_strong_runtime_options(QDict
*d
, BlockDriverState
*bs
)
8049 bool found_any
= false;
8050 const char *const *option_name
= NULL
;
8056 while ((option_name
= strong_options(bs
, option_name
))) {
8057 bool option_given
= false;
8059 assert(strlen(*option_name
) > 0);
8060 if ((*option_name
)[strlen(*option_name
) - 1] != '.') {
8061 QObject
*entry
= qdict_get(bs
->options
, *option_name
);
8066 qdict_put_obj(d
, *option_name
, qobject_ref(entry
));
8067 option_given
= true;
8069 const QDictEntry
*entry
;
8070 for (entry
= qdict_first(bs
->options
); entry
;
8071 entry
= qdict_next(bs
->options
, entry
))
8073 if (strstart(qdict_entry_key(entry
), *option_name
, NULL
)) {
8074 qdict_put_obj(d
, qdict_entry_key(entry
),
8075 qobject_ref(qdict_entry_value(entry
)));
8076 option_given
= true;
8081 /* While "driver" and "filename" need to be included in a JSON filename,
8082 * their existence does not prohibit generation of a plain filename. */
8083 if (!found_any
&& option_given
&&
8084 strcmp(*option_name
, "driver") && strcmp(*option_name
, "filename"))
8090 if (!qdict_haskey(d
, "driver")) {
8091 /* Drivers created with bdrv_new_open_driver() may not have a
8092 * @driver option. Add it here. */
8093 qdict_put_str(d
, "driver", bs
->drv
->format_name
);
8099 /* Note: This function may return false positives; it may return true
8100 * even if opening the backing file specified by bs's image header
8101 * would result in exactly bs->backing. */
8102 static bool bdrv_backing_overridden(BlockDriverState
*bs
)
8104 GLOBAL_STATE_CODE();
8106 return strcmp(bs
->auto_backing_file
,
8107 bs
->backing
->bs
->filename
);
8109 /* No backing BDS, so if the image header reports any backing
8110 * file, it must have been suppressed */
8111 return bs
->auto_backing_file
[0] != '\0';
8115 /* Updates the following BDS fields:
8116 * - exact_filename: A filename which may be used for opening a block device
8117 * which (mostly) equals the given BDS (even without any
8118 * other options; so reading and writing must return the same
8119 * results, but caching etc. may be different)
8120 * - full_open_options: Options which, when given when opening a block device
8121 * (without a filename), result in a BDS (mostly)
8122 * equalling the given one
8123 * - filename: If exact_filename is set, it is copied here. Otherwise,
8124 * full_open_options is converted to a JSON object, prefixed with
8125 * "json:" (for use through the JSON pseudo protocol) and put here.
8127 void bdrv_refresh_filename(BlockDriverState
*bs
)
8129 BlockDriver
*drv
= bs
->drv
;
8131 BlockDriverState
*primary_child_bs
;
8133 bool backing_overridden
;
8134 bool generate_json_filename
; /* Whether our default implementation should
8135 fill exact_filename (false) or not (true) */
8137 GLOBAL_STATE_CODE();
8143 /* This BDS's file name may depend on any of its children's file names, so
8144 * refresh those first */
8145 QLIST_FOREACH(child
, &bs
->children
, next
) {
8146 bdrv_refresh_filename(child
->bs
);
8150 /* For implicit nodes, just copy everything from the single child */
8151 child
= QLIST_FIRST(&bs
->children
);
8152 assert(QLIST_NEXT(child
, next
) == NULL
);
8154 pstrcpy(bs
->exact_filename
, sizeof(bs
->exact_filename
),
8155 child
->bs
->exact_filename
);
8156 pstrcpy(bs
->filename
, sizeof(bs
->filename
), child
->bs
->filename
);
8158 qobject_unref(bs
->full_open_options
);
8159 bs
->full_open_options
= qobject_ref(child
->bs
->full_open_options
);
8164 backing_overridden
= bdrv_backing_overridden(bs
);
8166 if (bs
->open_flags
& BDRV_O_NO_IO
) {
8167 /* Without I/O, the backing file does not change anything.
8168 * Therefore, in such a case (primarily qemu-img), we can
8169 * pretend the backing file has not been overridden even if
8170 * it technically has been. */
8171 backing_overridden
= false;
8174 /* Gather the options QDict */
8176 generate_json_filename
= append_strong_runtime_options(opts
, bs
);
8177 generate_json_filename
|= backing_overridden
;
8179 if (drv
->bdrv_gather_child_options
) {
8180 /* Some block drivers may not want to present all of their children's
8181 * options, or name them differently from BdrvChild.name */
8182 drv
->bdrv_gather_child_options(bs
, opts
, backing_overridden
);
8184 QLIST_FOREACH(child
, &bs
->children
, next
) {
8185 if (child
== bs
->backing
&& !backing_overridden
) {
8186 /* We can skip the backing BDS if it has not been overridden */
8190 qdict_put(opts
, child
->name
,
8191 qobject_ref(child
->bs
->full_open_options
));
8194 if (backing_overridden
&& !bs
->backing
) {
8195 /* Force no backing file */
8196 qdict_put_null(opts
, "backing");
8200 qobject_unref(bs
->full_open_options
);
8201 bs
->full_open_options
= opts
;
8203 primary_child_bs
= bdrv_primary_bs(bs
);
8205 if (drv
->bdrv_refresh_filename
) {
8206 /* Obsolete information is of no use here, so drop the old file name
8207 * information before refreshing it */
8208 bs
->exact_filename
[0] = '\0';
8210 drv
->bdrv_refresh_filename(bs
);
8211 } else if (primary_child_bs
) {
8213 * Try to reconstruct valid information from the underlying
8214 * file -- this only works for format nodes (filter nodes
8215 * cannot be probed and as such must be selected by the user
8216 * either through an options dict, or through a special
8217 * filename which the filter driver must construct in its
8218 * .bdrv_refresh_filename() implementation).
8221 bs
->exact_filename
[0] = '\0';
8224 * We can use the underlying file's filename if:
8225 * - it has a filename,
8226 * - the current BDS is not a filter,
8227 * - the file is a protocol BDS, and
8228 * - opening that file (as this BDS's format) will automatically create
8229 * the BDS tree we have right now, that is:
8230 * - the user did not significantly change this BDS's behavior with
8231 * some explicit (strong) options
8232 * - no non-file child of this BDS has been overridden by the user
8233 * Both of these conditions are represented by generate_json_filename.
8235 if (primary_child_bs
->exact_filename
[0] &&
8236 primary_child_bs
->drv
->bdrv_file_open
&&
8237 !drv
->is_filter
&& !generate_json_filename
)
8239 strcpy(bs
->exact_filename
, primary_child_bs
->exact_filename
);
8243 if (bs
->exact_filename
[0]) {
8244 pstrcpy(bs
->filename
, sizeof(bs
->filename
), bs
->exact_filename
);
8246 GString
*json
= qobject_to_json(QOBJECT(bs
->full_open_options
));
8247 if (snprintf(bs
->filename
, sizeof(bs
->filename
), "json:%s",
8248 json
->str
) >= sizeof(bs
->filename
)) {
8249 /* Give user a hint if we truncated things. */
8250 strcpy(bs
->filename
+ sizeof(bs
->filename
) - 4, "...");
8252 g_string_free(json
, true);
8256 char *bdrv_dirname(BlockDriverState
*bs
, Error
**errp
)
8258 BlockDriver
*drv
= bs
->drv
;
8259 BlockDriverState
*child_bs
;
8261 GLOBAL_STATE_CODE();
8264 error_setg(errp
, "Node '%s' is ejected", bs
->node_name
);
8268 if (drv
->bdrv_dirname
) {
8269 return drv
->bdrv_dirname(bs
, errp
);
8272 child_bs
= bdrv_primary_bs(bs
);
8274 return bdrv_dirname(child_bs
, errp
);
8277 bdrv_refresh_filename(bs
);
8278 if (bs
->exact_filename
[0] != '\0') {
8279 return path_combine(bs
->exact_filename
, "");
8282 error_setg(errp
, "Cannot generate a base directory for %s nodes",
8288 * Hot add/remove a BDS's child. So the user can take a child offline when
8289 * it is broken and take a new child online
8291 void bdrv_add_child(BlockDriverState
*parent_bs
, BlockDriverState
*child_bs
,
8294 GLOBAL_STATE_CODE();
8295 if (!parent_bs
->drv
|| !parent_bs
->drv
->bdrv_add_child
) {
8296 error_setg(errp
, "The node %s does not support adding a child",
8297 bdrv_get_device_or_node_name(parent_bs
));
8302 * Non-zoned block drivers do not follow zoned storage constraints
8303 * (i.e. sequential writes to zones). Refuse mixing zoned and non-zoned
8304 * drivers in a graph.
8306 if (!parent_bs
->drv
->supports_zoned_children
&&
8307 child_bs
->bl
.zoned
== BLK_Z_HM
) {
8309 * The host-aware model allows zoned storage constraints and random
8310 * write. Allow mixing host-aware and non-zoned drivers. Using
8311 * host-aware device as a regular device.
8313 error_setg(errp
, "Cannot add a %s child to a %s parent",
8314 child_bs
->bl
.zoned
== BLK_Z_HM
? "zoned" : "non-zoned",
8315 parent_bs
->drv
->supports_zoned_children
?
8316 "support zoned children" : "not support zoned children");
8320 if (!QLIST_EMPTY(&child_bs
->parents
)) {
8321 error_setg(errp
, "The node %s already has a parent",
8322 child_bs
->node_name
);
8326 parent_bs
->drv
->bdrv_add_child(parent_bs
, child_bs
, errp
);
8329 void bdrv_del_child(BlockDriverState
*parent_bs
, BdrvChild
*child
, Error
**errp
)
8333 GLOBAL_STATE_CODE();
8334 if (!parent_bs
->drv
|| !parent_bs
->drv
->bdrv_del_child
) {
8335 error_setg(errp
, "The node %s does not support removing a child",
8336 bdrv_get_device_or_node_name(parent_bs
));
8340 QLIST_FOREACH(tmp
, &parent_bs
->children
, next
) {
8347 error_setg(errp
, "The node %s does not have a child named %s",
8348 bdrv_get_device_or_node_name(parent_bs
),
8349 bdrv_get_device_or_node_name(child
->bs
));
8353 parent_bs
->drv
->bdrv_del_child(parent_bs
, child
, errp
);
8356 int bdrv_make_empty(BdrvChild
*c
, Error
**errp
)
8358 BlockDriver
*drv
= c
->bs
->drv
;
8361 GLOBAL_STATE_CODE();
8362 assert(c
->perm
& (BLK_PERM_WRITE
| BLK_PERM_WRITE_UNCHANGED
));
8364 if (!drv
->bdrv_make_empty
) {
8365 error_setg(errp
, "%s does not support emptying nodes",
8370 ret
= drv
->bdrv_make_empty(c
->bs
);
8372 error_setg_errno(errp
, -ret
, "Failed to empty %s",
8381 * Return the child that @bs acts as an overlay for, and from which data may be
8382 * copied in COW or COR operations. Usually this is the backing file.
8384 BdrvChild
*bdrv_cow_child(BlockDriverState
*bs
)
8388 if (!bs
|| !bs
->drv
) {
8392 if (bs
->drv
->is_filter
) {
8400 assert(bs
->backing
->role
& BDRV_CHILD_COW
);
8405 * If @bs acts as a filter for exactly one of its children, return
8408 BdrvChild
*bdrv_filter_child(BlockDriverState
*bs
)
8413 if (!bs
|| !bs
->drv
) {
8417 if (!bs
->drv
->is_filter
) {
8421 /* Only one of @backing or @file may be used */
8422 assert(!(bs
->backing
&& bs
->file
));
8424 c
= bs
->backing
?: bs
->file
;
8429 assert(c
->role
& BDRV_CHILD_FILTERED
);
8434 * Return either the result of bdrv_cow_child() or bdrv_filter_child(),
8435 * whichever is non-NULL.
8437 * Return NULL if both are NULL.
8439 BdrvChild
*bdrv_filter_or_cow_child(BlockDriverState
*bs
)
8441 BdrvChild
*cow_child
= bdrv_cow_child(bs
);
8442 BdrvChild
*filter_child
= bdrv_filter_child(bs
);
8445 /* Filter nodes cannot have COW backing files */
8446 assert(!(cow_child
&& filter_child
));
8448 return cow_child
?: filter_child
;
8452 * Return the primary child of this node: For filters, that is the
8453 * filtered child. For other nodes, that is usually the child storing
8455 * (A generally more helpful description is that this is (usually) the
8456 * child that has the same filename as @bs.)
8458 * Drivers do not necessarily have a primary child; for example quorum
8461 BdrvChild
*bdrv_primary_child(BlockDriverState
*bs
)
8463 BdrvChild
*c
, *found
= NULL
;
8466 QLIST_FOREACH(c
, &bs
->children
, next
) {
8467 if (c
->role
& BDRV_CHILD_PRIMARY
) {
8476 static BlockDriverState
*bdrv_do_skip_filters(BlockDriverState
*bs
,
8477 bool stop_on_explicit_filter
)
8485 while (!(stop_on_explicit_filter
&& !bs
->implicit
)) {
8486 c
= bdrv_filter_child(bs
);
8489 * A filter that is embedded in a working block graph must
8490 * have a child. Assert this here so this function does
8491 * not return a filter node that is not expected by the
8494 assert(!bs
->drv
|| !bs
->drv
->is_filter
);
8500 * Note that this treats nodes with bs->drv == NULL as not being
8501 * filters (bs->drv == NULL should be replaced by something else
8503 * The advantage of this behavior is that this function will thus
8504 * always return a non-NULL value (given a non-NULL @bs).
8511 * Return the first BDS that has not been added implicitly or that
8512 * does not have a filtered child down the chain starting from @bs
8513 * (including @bs itself).
8515 BlockDriverState
*bdrv_skip_implicit_filters(BlockDriverState
*bs
)
8517 GLOBAL_STATE_CODE();
8518 return bdrv_do_skip_filters(bs
, true);
8522 * Return the first BDS that does not have a filtered child down the
8523 * chain starting from @bs (including @bs itself).
8525 BlockDriverState
*bdrv_skip_filters(BlockDriverState
*bs
)
8528 return bdrv_do_skip_filters(bs
, false);
8532 * For a backing chain, return the first non-filter backing image of
8533 * the first non-filter image.
8535 BlockDriverState
*bdrv_backing_chain_next(BlockDriverState
*bs
)
8538 return bdrv_skip_filters(bdrv_cow_bs(bdrv_skip_filters(bs
)));
8542 * Check whether [offset, offset + bytes) overlaps with the cached
8543 * block-status data region.
8545 * If so, and @pnum is not NULL, set *pnum to `bsc.data_end - offset`,
8546 * which is what bdrv_bsc_is_data()'s interface needs.
8547 * Otherwise, *pnum is not touched.
8549 static bool bdrv_bsc_range_overlaps_locked(BlockDriverState
*bs
,
8550 int64_t offset
, int64_t bytes
,
8553 BdrvBlockStatusCache
*bsc
= qatomic_rcu_read(&bs
->block_status_cache
);
8557 qatomic_read(&bsc
->valid
) &&
8558 ranges_overlap(offset
, bytes
, bsc
->data_start
,
8559 bsc
->data_end
- bsc
->data_start
);
8561 if (overlaps
&& pnum
) {
8562 *pnum
= bsc
->data_end
- offset
;
8569 * See block_int.h for this function's documentation.
8571 bool bdrv_bsc_is_data(BlockDriverState
*bs
, int64_t offset
, int64_t *pnum
)
8574 RCU_READ_LOCK_GUARD();
8575 return bdrv_bsc_range_overlaps_locked(bs
, offset
, 1, pnum
);
8579 * See block_int.h for this function's documentation.
8581 void bdrv_bsc_invalidate_range(BlockDriverState
*bs
,
8582 int64_t offset
, int64_t bytes
)
8585 RCU_READ_LOCK_GUARD();
8587 if (bdrv_bsc_range_overlaps_locked(bs
, offset
, bytes
, NULL
)) {
8588 qatomic_set(&bs
->block_status_cache
->valid
, false);
8593 * See block_int.h for this function's documentation.
8595 void bdrv_bsc_fill(BlockDriverState
*bs
, int64_t offset
, int64_t bytes
)
8597 BdrvBlockStatusCache
*new_bsc
= g_new(BdrvBlockStatusCache
, 1);
8598 BdrvBlockStatusCache
*old_bsc
;
8601 *new_bsc
= (BdrvBlockStatusCache
) {
8603 .data_start
= offset
,
8604 .data_end
= offset
+ bytes
,
8607 QEMU_LOCK_GUARD(&bs
->bsc_modify_lock
);
8609 old_bsc
= qatomic_rcu_read(&bs
->block_status_cache
);
8610 qatomic_rcu_set(&bs
->block_status_cache
, new_bsc
);
8612 g_free_rcu(old_bsc
, rcu
);