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 if (!(bs
->open_flags
& BDRV_O_RDWR
)) {
326 if (!(bs
->open_flags
& BDRV_O_AUTO_RDONLY
)) {
330 ret
= bdrv_can_set_read_only(bs
, true, false, NULL
);
335 bs
->open_flags
&= ~BDRV_O_RDWR
;
340 error_setg(errp
, "%s", errmsg
?: "Image is read-only");
345 * If @backing is empty, this function returns NULL without setting
346 * @errp. In all other cases, NULL will only be returned with @errp
349 * Therefore, a return value of NULL without @errp set means that
350 * there is no backing file; if @errp is set, there is one but its
351 * absolute filename cannot be generated.
353 char *bdrv_get_full_backing_filename_from_filename(const char *backed
,
357 if (backing
[0] == '\0') {
359 } else if (path_has_protocol(backing
) || path_is_absolute(backing
)) {
360 return g_strdup(backing
);
361 } else if (backed
[0] == '\0' || strstart(backed
, "json:", NULL
)) {
362 error_setg(errp
, "Cannot use relative backing file names for '%s'",
366 return path_combine(backed
, backing
);
371 * If @filename is empty or NULL, this function returns NULL without
372 * setting @errp. In all other cases, NULL will only be returned with
375 static char * GRAPH_RDLOCK
376 bdrv_make_absolute_filename(BlockDriverState
*relative_to
,
377 const char *filename
, Error
**errp
)
379 char *dir
, *full_name
;
381 if (!filename
|| filename
[0] == '\0') {
383 } else if (path_has_protocol(filename
) || path_is_absolute(filename
)) {
384 return g_strdup(filename
);
387 dir
= bdrv_dirname(relative_to
, errp
);
392 full_name
= g_strconcat(dir
, filename
, NULL
);
397 char *bdrv_get_full_backing_filename(BlockDriverState
*bs
, Error
**errp
)
400 return bdrv_make_absolute_filename(bs
, bs
->backing_file
, errp
);
403 void bdrv_register(BlockDriver
*bdrv
)
405 assert(bdrv
->format_name
);
407 QLIST_INSERT_HEAD(&bdrv_drivers
, bdrv
, list
);
410 BlockDriverState
*bdrv_new(void)
412 BlockDriverState
*bs
;
417 bs
= g_new0(BlockDriverState
, 1);
418 QLIST_INIT(&bs
->dirty_bitmaps
);
419 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
420 QLIST_INIT(&bs
->op_blockers
[i
]);
422 qemu_mutex_init(&bs
->reqs_lock
);
423 qemu_mutex_init(&bs
->dirty_bitmap_mutex
);
425 bs
->aio_context
= qemu_get_aio_context();
427 qemu_co_queue_init(&bs
->flush_queue
);
429 qemu_co_mutex_init(&bs
->bsc_modify_lock
);
430 bs
->block_status_cache
= g_new0(BdrvBlockStatusCache
, 1);
432 for (i
= 0; i
< bdrv_drain_all_count
; i
++) {
433 bdrv_drained_begin(bs
);
436 QTAILQ_INSERT_TAIL(&all_bdrv_states
, bs
, bs_list
);
441 static BlockDriver
*bdrv_do_find_format(const char *format_name
)
446 QLIST_FOREACH(drv1
, &bdrv_drivers
, list
) {
447 if (!strcmp(drv1
->format_name
, format_name
)) {
455 BlockDriver
*bdrv_find_format(const char *format_name
)
462 drv1
= bdrv_do_find_format(format_name
);
467 /* The driver isn't registered, maybe we need to load a module */
468 for (i
= 0; i
< (int)ARRAY_SIZE(block_driver_modules
); ++i
) {
469 if (!strcmp(block_driver_modules
[i
].format_name
, format_name
)) {
470 Error
*local_err
= NULL
;
471 int rv
= block_module_load(block_driver_modules
[i
].library_name
,
474 return bdrv_do_find_format(format_name
);
476 error_report_err(local_err
);
484 static int bdrv_format_is_whitelisted(const char *format_name
, bool read_only
)
486 static const char *whitelist_rw
[] = {
487 CONFIG_BDRV_RW_WHITELIST
490 static const char *whitelist_ro
[] = {
491 CONFIG_BDRV_RO_WHITELIST
496 if (!whitelist_rw
[0] && !whitelist_ro
[0]) {
497 return 1; /* no whitelist, anything goes */
500 for (p
= whitelist_rw
; *p
; p
++) {
501 if (!strcmp(format_name
, *p
)) {
506 for (p
= whitelist_ro
; *p
; p
++) {
507 if (!strcmp(format_name
, *p
)) {
515 int bdrv_is_whitelisted(BlockDriver
*drv
, bool read_only
)
518 return bdrv_format_is_whitelisted(drv
->format_name
, read_only
);
521 bool bdrv_uses_whitelist(void)
523 return use_bdrv_whitelist
;
526 typedef struct CreateCo
{
534 int coroutine_fn
bdrv_co_create(BlockDriver
*drv
, const char *filename
,
535 QemuOpts
*opts
, Error
**errp
)
541 if (!drv
->bdrv_co_create_opts
) {
542 error_setg(errp
, "Driver '%s' does not support image creation",
547 ret
= drv
->bdrv_co_create_opts(drv
, filename
, opts
, errp
);
548 if (ret
< 0 && !*errp
) {
549 error_setg_errno(errp
, -ret
, "Could not create image");
556 * Helper function for bdrv_create_file_fallback(): Resize @blk to at
557 * least the given @minimum_size.
559 * On success, return @blk's actual length.
560 * Otherwise, return -errno.
562 static int64_t coroutine_fn GRAPH_UNLOCKED
563 create_file_fallback_truncate(BlockBackend
*blk
, int64_t minimum_size
,
566 Error
*local_err
= NULL
;
572 ret
= blk_co_truncate(blk
, minimum_size
, false, PREALLOC_MODE_OFF
, 0,
574 if (ret
< 0 && ret
!= -ENOTSUP
) {
575 error_propagate(errp
, local_err
);
579 size
= blk_co_getlength(blk
);
581 error_free(local_err
);
582 error_setg_errno(errp
, -size
,
583 "Failed to inquire the new image file's length");
587 if (size
< minimum_size
) {
588 /* Need to grow the image, but we failed to do that */
589 error_propagate(errp
, local_err
);
593 error_free(local_err
);
600 * Helper function for bdrv_create_file_fallback(): Zero the first
601 * sector to remove any potentially pre-existing image header.
603 static int coroutine_fn
604 create_file_fallback_zero_first_sector(BlockBackend
*blk
,
605 int64_t current_size
,
608 int64_t bytes_to_clear
;
613 bytes_to_clear
= MIN(current_size
, BDRV_SECTOR_SIZE
);
614 if (bytes_to_clear
) {
615 ret
= blk_co_pwrite_zeroes(blk
, 0, bytes_to_clear
, BDRV_REQ_MAY_UNMAP
);
617 error_setg_errno(errp
, -ret
,
618 "Failed to clear the new image's first sector");
627 * Simple implementation of bdrv_co_create_opts for protocol drivers
628 * which only support creation via opening a file
629 * (usually existing raw storage device)
631 int coroutine_fn
bdrv_co_create_opts_simple(BlockDriver
*drv
,
632 const char *filename
,
640 PreallocMode prealloc
;
641 Error
*local_err
= NULL
;
646 size
= qemu_opt_get_size_del(opts
, BLOCK_OPT_SIZE
, 0);
647 buf
= qemu_opt_get_del(opts
, BLOCK_OPT_PREALLOC
);
648 prealloc
= qapi_enum_parse(&PreallocMode_lookup
, buf
,
649 PREALLOC_MODE_OFF
, &local_err
);
652 error_propagate(errp
, local_err
);
656 if (prealloc
!= PREALLOC_MODE_OFF
) {
657 error_setg(errp
, "Unsupported preallocation mode '%s'",
658 PreallocMode_str(prealloc
));
662 options
= qdict_new();
663 qdict_put_str(options
, "driver", drv
->format_name
);
665 blk
= blk_co_new_open(filename
, NULL
, options
,
666 BDRV_O_RDWR
| BDRV_O_RESIZE
, errp
);
668 error_prepend(errp
, "Protocol driver '%s' does not support creating "
669 "new images, so an existing image must be selected as "
670 "the target; however, opening the given target as an "
671 "existing image failed: ",
676 size
= create_file_fallback_truncate(blk
, size
, errp
);
682 ret
= create_file_fallback_zero_first_sector(blk
, size
, errp
);
693 int coroutine_fn
bdrv_co_create_file(const char *filename
, QemuOpts
*opts
,
696 QemuOpts
*protocol_opts
;
703 drv
= bdrv_find_protocol(filename
, true, errp
);
708 if (!drv
->create_opts
) {
709 error_setg(errp
, "Driver '%s' does not support image creation",
715 * 'opts' contains a QemuOptsList with a combination of format and protocol
718 * The format properly removes its options, but the default values remain
719 * in 'opts->list'. So if the protocol has options with the same name
720 * (e.g. rbd has 'cluster_size' as qcow2), it will see the default values
721 * of the format, since for overlapping options, the format wins.
723 * To avoid this issue, lets convert QemuOpts to QDict, in this way we take
724 * only the set options, and then convert it back to QemuOpts, using the
725 * create_opts of the protocol. So the new QemuOpts, will contain only the
728 qdict
= qemu_opts_to_qdict(opts
, NULL
);
729 protocol_opts
= qemu_opts_from_qdict(drv
->create_opts
, qdict
, errp
);
730 if (protocol_opts
== NULL
) {
735 ret
= bdrv_co_create(drv
, filename
, protocol_opts
, errp
);
737 qemu_opts_del(protocol_opts
);
738 qobject_unref(qdict
);
742 int coroutine_fn
bdrv_co_delete_file(BlockDriverState
*bs
, Error
**errp
)
744 Error
*local_err
= NULL
;
749 assert_bdrv_graph_readable();
752 error_setg(errp
, "Block node '%s' is not opened", bs
->filename
);
756 if (!bs
->drv
->bdrv_co_delete_file
) {
757 error_setg(errp
, "Driver '%s' does not support image deletion",
758 bs
->drv
->format_name
);
762 ret
= bs
->drv
->bdrv_co_delete_file(bs
, &local_err
);
764 error_propagate(errp
, local_err
);
770 void coroutine_fn
bdrv_co_delete_file_noerr(BlockDriverState
*bs
)
772 Error
*local_err
= NULL
;
780 ret
= bdrv_co_delete_file(bs
, &local_err
);
782 * ENOTSUP will happen if the block driver doesn't support
783 * the 'bdrv_co_delete_file' interface. This is a predictable
784 * scenario and shouldn't be reported back to the user.
786 if (ret
== -ENOTSUP
) {
787 error_free(local_err
);
788 } else if (ret
< 0) {
789 error_report_err(local_err
);
794 * Try to get @bs's logical and physical block size.
795 * On success, store them in @bsz struct and return 0.
796 * On failure return -errno.
797 * @bs must not be empty.
799 int bdrv_probe_blocksizes(BlockDriverState
*bs
, BlockSizes
*bsz
)
801 BlockDriver
*drv
= bs
->drv
;
802 BlockDriverState
*filtered
= bdrv_filter_bs(bs
);
805 if (drv
&& drv
->bdrv_probe_blocksizes
) {
806 return drv
->bdrv_probe_blocksizes(bs
, bsz
);
807 } else if (filtered
) {
808 return bdrv_probe_blocksizes(filtered
, bsz
);
815 * Try to get @bs's geometry (cyls, heads, sectors).
816 * On success, store them in @geo struct and return 0.
817 * On failure return -errno.
818 * @bs must not be empty.
820 int bdrv_probe_geometry(BlockDriverState
*bs
, HDGeometry
*geo
)
822 BlockDriver
*drv
= bs
->drv
;
823 BlockDriverState
*filtered
;
826 GRAPH_RDLOCK_GUARD_MAINLOOP();
828 if (drv
&& drv
->bdrv_probe_geometry
) {
829 return drv
->bdrv_probe_geometry(bs
, geo
);
832 filtered
= bdrv_filter_bs(bs
);
834 return bdrv_probe_geometry(filtered
, geo
);
841 * Create a uniquely-named empty temporary file.
842 * Return the actual file name used upon success, otherwise NULL.
843 * This string should be freed with g_free() when not needed any longer.
845 * Note: creating a temporary file for the caller to (re)open is
846 * inherently racy. Use g_file_open_tmp() instead whenever practical.
848 char *create_tmp_file(Error
**errp
)
852 g_autofree
char *filename
= NULL
;
854 tmpdir
= g_get_tmp_dir();
857 * See commit 69bef79 ("block: use /var/tmp instead of /tmp for -snapshot")
859 * This function is used to create temporary disk images (like -snapshot),
860 * so the files can become very large. /tmp is often a tmpfs where as
861 * /var/tmp is usually on a disk, so more appropriate for disk images.
863 if (!g_strcmp0(tmpdir
, "/tmp")) {
868 filename
= g_strdup_printf("%s/vl.XXXXXX", tmpdir
);
869 fd
= g_mkstemp(filename
);
871 error_setg_errno(errp
, errno
, "Could not open temporary file '%s'",
877 return g_steal_pointer(&filename
);
881 * Detect host devices. By convention, /dev/cdrom[N] is always
882 * recognized as a host CDROM.
884 static BlockDriver
*find_hdev_driver(const char *filename
)
886 int score_max
= 0, score
;
887 BlockDriver
*drv
= NULL
, *d
;
890 QLIST_FOREACH(d
, &bdrv_drivers
, list
) {
891 if (d
->bdrv_probe_device
) {
892 score
= d
->bdrv_probe_device(filename
);
893 if (score
> score_max
) {
903 static BlockDriver
*bdrv_do_find_protocol(const char *protocol
)
908 QLIST_FOREACH(drv1
, &bdrv_drivers
, list
) {
909 if (drv1
->protocol_name
&& !strcmp(drv1
->protocol_name
, protocol
)) {
917 BlockDriver
*bdrv_find_protocol(const char *filename
,
918 bool allow_protocol_prefix
,
928 /* TODO Drivers without bdrv_file_open must be specified explicitly */
931 * XXX(hch): we really should not let host device detection
932 * override an explicit protocol specification, but moving this
933 * later breaks access to device names with colons in them.
934 * Thanks to the brain-dead persistent naming schemes on udev-
935 * based Linux systems those actually are quite common.
937 drv1
= find_hdev_driver(filename
);
942 if (!path_has_protocol(filename
) || !allow_protocol_prefix
) {
946 p
= strchr(filename
, ':');
949 if (len
> sizeof(protocol
) - 1)
950 len
= sizeof(protocol
) - 1;
951 memcpy(protocol
, filename
, len
);
952 protocol
[len
] = '\0';
954 drv1
= bdrv_do_find_protocol(protocol
);
959 for (i
= 0; i
< (int)ARRAY_SIZE(block_driver_modules
); ++i
) {
960 if (block_driver_modules
[i
].protocol_name
&&
961 !strcmp(block_driver_modules
[i
].protocol_name
, protocol
)) {
962 int rv
= block_module_load(block_driver_modules
[i
].library_name
, errp
);
964 drv1
= bdrv_do_find_protocol(protocol
);
973 error_setg(errp
, "Unknown protocol '%s'", protocol
);
979 * Guess image format by probing its contents.
980 * This is not a good idea when your image is raw (CVE-2008-2004), but
981 * we do it anyway for backward compatibility.
983 * @buf contains the image's first @buf_size bytes.
984 * @buf_size is the buffer size in bytes (generally BLOCK_PROBE_BUF_SIZE,
985 * but can be smaller if the image file is smaller)
986 * @filename is its filename.
988 * For all block drivers, call the bdrv_probe() method to get its
990 * Return the first block driver with the highest probing score.
992 BlockDriver
*bdrv_probe_all(const uint8_t *buf
, int buf_size
,
993 const char *filename
)
995 int score_max
= 0, score
;
996 BlockDriver
*drv
= NULL
, *d
;
999 QLIST_FOREACH(d
, &bdrv_drivers
, list
) {
1000 if (d
->bdrv_probe
) {
1001 score
= d
->bdrv_probe(buf
, buf_size
, filename
);
1002 if (score
> score_max
) {
1012 static int find_image_format(BlockBackend
*file
, const char *filename
,
1013 BlockDriver
**pdrv
, Error
**errp
)
1016 uint8_t buf
[BLOCK_PROBE_BUF_SIZE
];
1019 GLOBAL_STATE_CODE();
1021 /* Return the raw BlockDriver * to scsi-generic devices or empty drives */
1022 if (blk_is_sg(file
) || !blk_is_inserted(file
) || blk_getlength(file
) == 0) {
1027 ret
= blk_pread(file
, 0, sizeof(buf
), buf
, 0);
1029 error_setg_errno(errp
, -ret
, "Could not read image for determining its "
1035 drv
= bdrv_probe_all(buf
, sizeof(buf
), filename
);
1037 error_setg(errp
, "Could not determine image format: No compatible "
1048 * Set the current 'total_sectors' value
1049 * Return 0 on success, -errno on error.
1051 int coroutine_fn
bdrv_co_refresh_total_sectors(BlockDriverState
*bs
,
1054 BlockDriver
*drv
= bs
->drv
;
1056 assert_bdrv_graph_readable();
1062 /* Do not attempt drv->bdrv_co_getlength() on scsi-generic devices */
1066 /* query actual device if possible, otherwise just trust the hint */
1067 if (drv
->bdrv_co_getlength
) {
1068 int64_t length
= drv
->bdrv_co_getlength(bs
);
1072 hint
= DIV_ROUND_UP(length
, BDRV_SECTOR_SIZE
);
1075 bs
->total_sectors
= hint
;
1077 if (bs
->total_sectors
* BDRV_SECTOR_SIZE
> BDRV_MAX_LENGTH
) {
1085 * Combines a QDict of new block driver @options with any missing options taken
1086 * from @old_options, so that leaving out an option defaults to its old value.
1088 static void bdrv_join_options(BlockDriverState
*bs
, QDict
*options
,
1091 GLOBAL_STATE_CODE();
1092 if (bs
->drv
&& bs
->drv
->bdrv_join_options
) {
1093 bs
->drv
->bdrv_join_options(options
, old_options
);
1095 qdict_join(options
, old_options
, false);
1099 static BlockdevDetectZeroesOptions
bdrv_parse_detect_zeroes(QemuOpts
*opts
,
1103 Error
*local_err
= NULL
;
1104 char *value
= qemu_opt_get_del(opts
, "detect-zeroes");
1105 BlockdevDetectZeroesOptions detect_zeroes
=
1106 qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup
, value
,
1107 BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF
, &local_err
);
1108 GLOBAL_STATE_CODE();
1111 error_propagate(errp
, local_err
);
1112 return detect_zeroes
;
1115 if (detect_zeroes
== BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP
&&
1116 !(open_flags
& BDRV_O_UNMAP
))
1118 error_setg(errp
, "setting detect-zeroes to unmap is not allowed "
1119 "without setting discard operation to unmap");
1122 return detect_zeroes
;
1126 * Set open flags for aio engine
1128 * Return 0 on success, -1 if the engine specified is invalid
1130 int bdrv_parse_aio(const char *mode
, int *flags
)
1132 if (!strcmp(mode
, "threads")) {
1133 /* do nothing, default */
1134 } else if (!strcmp(mode
, "native")) {
1135 *flags
|= BDRV_O_NATIVE_AIO
;
1136 #ifdef CONFIG_LINUX_IO_URING
1137 } else if (!strcmp(mode
, "io_uring")) {
1138 *flags
|= BDRV_O_IO_URING
;
1148 * Set open flags for a given discard mode
1150 * Return 0 on success, -1 if the discard mode was invalid.
1152 int bdrv_parse_discard_flags(const char *mode
, int *flags
)
1154 *flags
&= ~BDRV_O_UNMAP
;
1156 if (!strcmp(mode
, "off") || !strcmp(mode
, "ignore")) {
1158 } else if (!strcmp(mode
, "on") || !strcmp(mode
, "unmap")) {
1159 *flags
|= BDRV_O_UNMAP
;
1168 * Set open flags for a given cache mode
1170 * Return 0 on success, -1 if the cache mode was invalid.
1172 int bdrv_parse_cache_mode(const char *mode
, int *flags
, bool *writethrough
)
1174 *flags
&= ~BDRV_O_CACHE_MASK
;
1176 if (!strcmp(mode
, "off") || !strcmp(mode
, "none")) {
1177 *writethrough
= false;
1178 *flags
|= BDRV_O_NOCACHE
;
1179 } else if (!strcmp(mode
, "directsync")) {
1180 *writethrough
= true;
1181 *flags
|= BDRV_O_NOCACHE
;
1182 } else if (!strcmp(mode
, "writeback")) {
1183 *writethrough
= false;
1184 } else if (!strcmp(mode
, "unsafe")) {
1185 *writethrough
= false;
1186 *flags
|= BDRV_O_NO_FLUSH
;
1187 } else if (!strcmp(mode
, "writethrough")) {
1188 *writethrough
= true;
1196 static char *bdrv_child_get_parent_desc(BdrvChild
*c
)
1198 BlockDriverState
*parent
= c
->opaque
;
1199 return g_strdup_printf("node '%s'", bdrv_get_node_name(parent
));
1202 static void GRAPH_RDLOCK
bdrv_child_cb_drained_begin(BdrvChild
*child
)
1204 BlockDriverState
*bs
= child
->opaque
;
1205 bdrv_do_drained_begin_quiesce(bs
, NULL
);
1208 static bool GRAPH_RDLOCK
bdrv_child_cb_drained_poll(BdrvChild
*child
)
1210 BlockDriverState
*bs
= child
->opaque
;
1211 return bdrv_drain_poll(bs
, NULL
, false);
1214 static void GRAPH_RDLOCK
bdrv_child_cb_drained_end(BdrvChild
*child
)
1216 BlockDriverState
*bs
= child
->opaque
;
1217 bdrv_drained_end(bs
);
1220 static int bdrv_child_cb_inactivate(BdrvChild
*child
)
1222 BlockDriverState
*bs
= child
->opaque
;
1223 GLOBAL_STATE_CODE();
1224 assert(bs
->open_flags
& BDRV_O_INACTIVE
);
1228 static bool bdrv_child_cb_change_aio_ctx(BdrvChild
*child
, AioContext
*ctx
,
1229 GHashTable
*visited
, Transaction
*tran
,
1232 BlockDriverState
*bs
= child
->opaque
;
1233 return bdrv_change_aio_context(bs
, ctx
, visited
, tran
, errp
);
1237 * Returns the options and flags that a temporary snapshot should get, based on
1238 * the originally requested flags (the originally requested image will have
1239 * flags like a backing file)
1241 static void bdrv_temp_snapshot_options(int *child_flags
, QDict
*child_options
,
1242 int parent_flags
, QDict
*parent_options
)
1244 GLOBAL_STATE_CODE();
1245 *child_flags
= (parent_flags
& ~BDRV_O_SNAPSHOT
) | BDRV_O_TEMPORARY
;
1247 /* For temporary files, unconditional cache=unsafe is fine */
1248 qdict_set_default_str(child_options
, BDRV_OPT_CACHE_DIRECT
, "off");
1249 qdict_set_default_str(child_options
, BDRV_OPT_CACHE_NO_FLUSH
, "on");
1251 /* Copy the read-only and discard options from the parent */
1252 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_READ_ONLY
);
1253 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_DISCARD
);
1255 /* aio=native doesn't work for cache.direct=off, so disable it for the
1256 * temporary snapshot */
1257 *child_flags
&= ~BDRV_O_NATIVE_AIO
;
1260 static void GRAPH_WRLOCK
bdrv_backing_attach(BdrvChild
*c
)
1262 BlockDriverState
*parent
= c
->opaque
;
1263 BlockDriverState
*backing_hd
= c
->bs
;
1265 GLOBAL_STATE_CODE();
1266 assert(!parent
->backing_blocker
);
1267 error_setg(&parent
->backing_blocker
,
1268 "node is used as backing hd of '%s'",
1269 bdrv_get_device_or_node_name(parent
));
1271 bdrv_refresh_filename(backing_hd
);
1273 parent
->open_flags
&= ~BDRV_O_NO_BACKING
;
1275 bdrv_op_block_all(backing_hd
, parent
->backing_blocker
);
1276 /* Otherwise we won't be able to commit or stream */
1277 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_COMMIT_TARGET
,
1278 parent
->backing_blocker
);
1279 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_STREAM
,
1280 parent
->backing_blocker
);
1282 * We do backup in 3 ways:
1284 * The target bs is new opened, and the source is top BDS
1285 * 2. blockdev backup
1286 * Both the source and the target are top BDSes.
1287 * 3. internal backup(used for block replication)
1288 * Both the source and the target are backing file
1290 * In case 1 and 2, neither the source nor the target is the backing file.
1291 * In case 3, we will block the top BDS, so there is only one block job
1292 * for the top BDS and its backing chain.
1294 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_BACKUP_SOURCE
,
1295 parent
->backing_blocker
);
1296 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_BACKUP_TARGET
,
1297 parent
->backing_blocker
);
1300 static void bdrv_backing_detach(BdrvChild
*c
)
1302 BlockDriverState
*parent
= c
->opaque
;
1304 GLOBAL_STATE_CODE();
1305 assert(parent
->backing_blocker
);
1306 bdrv_op_unblock_all(c
->bs
, parent
->backing_blocker
);
1307 error_free(parent
->backing_blocker
);
1308 parent
->backing_blocker
= NULL
;
1311 static int bdrv_backing_update_filename(BdrvChild
*c
, BlockDriverState
*base
,
1312 const char *filename
, Error
**errp
)
1314 BlockDriverState
*parent
= c
->opaque
;
1315 bool read_only
= bdrv_is_read_only(parent
);
1317 GLOBAL_STATE_CODE();
1320 ret
= bdrv_reopen_set_read_only(parent
, false, errp
);
1326 ret
= bdrv_change_backing_file(parent
, filename
,
1327 base
->drv
? base
->drv
->format_name
: "",
1330 error_setg_errno(errp
, -ret
, "Could not update backing file link");
1334 bdrv_reopen_set_read_only(parent
, true, NULL
);
1341 * Returns the options and flags that a generic child of a BDS should
1342 * get, based on the given options and flags for the parent BDS.
1344 static void bdrv_inherited_options(BdrvChildRole role
, bool parent_is_format
,
1345 int *child_flags
, QDict
*child_options
,
1346 int parent_flags
, QDict
*parent_options
)
1348 int flags
= parent_flags
;
1349 GLOBAL_STATE_CODE();
1352 * First, decide whether to set, clear, or leave BDRV_O_PROTOCOL.
1353 * Generally, the question to answer is: Should this child be
1354 * format-probed by default?
1358 * Pure and non-filtered data children of non-format nodes should
1359 * be probed by default (even when the node itself has BDRV_O_PROTOCOL
1360 * set). This only affects a very limited set of drivers (namely
1361 * quorum and blkverify when this comment was written).
1362 * Force-clear BDRV_O_PROTOCOL then.
1364 if (!parent_is_format
&&
1365 (role
& BDRV_CHILD_DATA
) &&
1366 !(role
& (BDRV_CHILD_METADATA
| BDRV_CHILD_FILTERED
)))
1368 flags
&= ~BDRV_O_PROTOCOL
;
1372 * All children of format nodes (except for COW children) and all
1373 * metadata children in general should never be format-probed.
1374 * Force-set BDRV_O_PROTOCOL then.
1376 if ((parent_is_format
&& !(role
& BDRV_CHILD_COW
)) ||
1377 (role
& BDRV_CHILD_METADATA
))
1379 flags
|= BDRV_O_PROTOCOL
;
1383 * If the cache mode isn't explicitly set, inherit direct and no-flush from
1386 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_CACHE_DIRECT
);
1387 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_CACHE_NO_FLUSH
);
1388 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_FORCE_SHARE
);
1390 if (role
& BDRV_CHILD_COW
) {
1391 /* backing files are opened read-only by default */
1392 qdict_set_default_str(child_options
, BDRV_OPT_READ_ONLY
, "on");
1393 qdict_set_default_str(child_options
, BDRV_OPT_AUTO_READ_ONLY
, "off");
1395 /* Inherit the read-only option from the parent if it's not set */
1396 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_READ_ONLY
);
1397 qdict_copy_default(child_options
, parent_options
,
1398 BDRV_OPT_AUTO_READ_ONLY
);
1402 * bdrv_co_pdiscard() respects unmap policy for the parent, so we
1403 * can default to enable it on lower layers regardless of the
1406 qdict_set_default_str(child_options
, BDRV_OPT_DISCARD
, "unmap");
1408 /* Clear flags that only apply to the top layer */
1409 flags
&= ~(BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
| BDRV_O_COPY_ON_READ
);
1411 if (role
& BDRV_CHILD_METADATA
) {
1412 flags
&= ~BDRV_O_NO_IO
;
1414 if (role
& BDRV_CHILD_COW
) {
1415 flags
&= ~BDRV_O_TEMPORARY
;
1418 *child_flags
= flags
;
1421 static void GRAPH_WRLOCK
bdrv_child_cb_attach(BdrvChild
*child
)
1423 BlockDriverState
*bs
= child
->opaque
;
1425 assert_bdrv_graph_writable();
1426 QLIST_INSERT_HEAD(&bs
->children
, child
, next
);
1427 if (bs
->drv
->is_filter
|| (child
->role
& BDRV_CHILD_FILTERED
)) {
1429 * Here we handle filters and block/raw-format.c when it behave like
1430 * filter. They generally have a single PRIMARY child, which is also the
1431 * FILTERED child, and that they may have multiple more children, which
1432 * are neither PRIMARY nor FILTERED. And never we have a COW child here.
1433 * So bs->file will be the PRIMARY child, unless the PRIMARY child goes
1434 * into bs->backing on exceptional cases; and bs->backing will be
1437 assert(!(child
->role
& BDRV_CHILD_COW
));
1438 if (child
->role
& BDRV_CHILD_PRIMARY
) {
1439 assert(child
->role
& BDRV_CHILD_FILTERED
);
1440 assert(!bs
->backing
);
1443 if (bs
->drv
->filtered_child_is_backing
) {
1444 bs
->backing
= child
;
1449 assert(!(child
->role
& BDRV_CHILD_FILTERED
));
1451 } else if (child
->role
& BDRV_CHILD_COW
) {
1452 assert(bs
->drv
->supports_backing
);
1453 assert(!(child
->role
& BDRV_CHILD_PRIMARY
));
1454 assert(!bs
->backing
);
1455 bs
->backing
= child
;
1456 bdrv_backing_attach(child
);
1457 } else if (child
->role
& BDRV_CHILD_PRIMARY
) {
1463 static void GRAPH_WRLOCK
bdrv_child_cb_detach(BdrvChild
*child
)
1465 BlockDriverState
*bs
= child
->opaque
;
1467 if (child
->role
& BDRV_CHILD_COW
) {
1468 bdrv_backing_detach(child
);
1471 assert_bdrv_graph_writable();
1472 QLIST_REMOVE(child
, next
);
1473 if (child
== bs
->backing
) {
1474 assert(child
!= bs
->file
);
1476 } else if (child
== bs
->file
) {
1481 static int bdrv_child_cb_update_filename(BdrvChild
*c
, BlockDriverState
*base
,
1482 const char *filename
, Error
**errp
)
1484 if (c
->role
& BDRV_CHILD_COW
) {
1485 return bdrv_backing_update_filename(c
, base
, filename
, errp
);
1490 AioContext
*child_of_bds_get_parent_aio_context(BdrvChild
*c
)
1492 BlockDriverState
*bs
= c
->opaque
;
1495 return bdrv_get_aio_context(bs
);
1498 const BdrvChildClass child_of_bds
= {
1499 .parent_is_bds
= true,
1500 .get_parent_desc
= bdrv_child_get_parent_desc
,
1501 .inherit_options
= bdrv_inherited_options
,
1502 .drained_begin
= bdrv_child_cb_drained_begin
,
1503 .drained_poll
= bdrv_child_cb_drained_poll
,
1504 .drained_end
= bdrv_child_cb_drained_end
,
1505 .attach
= bdrv_child_cb_attach
,
1506 .detach
= bdrv_child_cb_detach
,
1507 .inactivate
= bdrv_child_cb_inactivate
,
1508 .change_aio_ctx
= bdrv_child_cb_change_aio_ctx
,
1509 .update_filename
= bdrv_child_cb_update_filename
,
1510 .get_parent_aio_context
= child_of_bds_get_parent_aio_context
,
1513 AioContext
*bdrv_child_get_parent_aio_context(BdrvChild
*c
)
1516 return c
->klass
->get_parent_aio_context(c
);
1519 static int bdrv_open_flags(BlockDriverState
*bs
, int flags
)
1521 int open_flags
= flags
;
1522 GLOBAL_STATE_CODE();
1525 * Clear flags that are internal to the block layer before opening the
1528 open_flags
&= ~(BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
| BDRV_O_PROTOCOL
);
1533 static void update_flags_from_options(int *flags
, QemuOpts
*opts
)
1535 GLOBAL_STATE_CODE();
1537 *flags
&= ~(BDRV_O_CACHE_MASK
| BDRV_O_RDWR
| BDRV_O_AUTO_RDONLY
);
1539 if (qemu_opt_get_bool_del(opts
, BDRV_OPT_CACHE_NO_FLUSH
, false)) {
1540 *flags
|= BDRV_O_NO_FLUSH
;
1543 if (qemu_opt_get_bool_del(opts
, BDRV_OPT_CACHE_DIRECT
, false)) {
1544 *flags
|= BDRV_O_NOCACHE
;
1547 if (!qemu_opt_get_bool_del(opts
, BDRV_OPT_READ_ONLY
, false)) {
1548 *flags
|= BDRV_O_RDWR
;
1551 if (qemu_opt_get_bool_del(opts
, BDRV_OPT_AUTO_READ_ONLY
, false)) {
1552 *flags
|= BDRV_O_AUTO_RDONLY
;
1556 static void update_options_from_flags(QDict
*options
, int flags
)
1558 GLOBAL_STATE_CODE();
1559 if (!qdict_haskey(options
, BDRV_OPT_CACHE_DIRECT
)) {
1560 qdict_put_bool(options
, BDRV_OPT_CACHE_DIRECT
, flags
& BDRV_O_NOCACHE
);
1562 if (!qdict_haskey(options
, BDRV_OPT_CACHE_NO_FLUSH
)) {
1563 qdict_put_bool(options
, BDRV_OPT_CACHE_NO_FLUSH
,
1564 flags
& BDRV_O_NO_FLUSH
);
1566 if (!qdict_haskey(options
, BDRV_OPT_READ_ONLY
)) {
1567 qdict_put_bool(options
, BDRV_OPT_READ_ONLY
, !(flags
& BDRV_O_RDWR
));
1569 if (!qdict_haskey(options
, BDRV_OPT_AUTO_READ_ONLY
)) {
1570 qdict_put_bool(options
, BDRV_OPT_AUTO_READ_ONLY
,
1571 flags
& BDRV_O_AUTO_RDONLY
);
1575 static void bdrv_assign_node_name(BlockDriverState
*bs
,
1576 const char *node_name
,
1579 char *gen_node_name
= NULL
;
1580 GLOBAL_STATE_CODE();
1583 node_name
= gen_node_name
= id_generate(ID_BLOCK
);
1584 } else if (!id_wellformed(node_name
)) {
1586 * Check for empty string or invalid characters, but not if it is
1587 * generated (generated names use characters not available to the user)
1589 error_setg(errp
, "Invalid node-name: '%s'", node_name
);
1593 /* takes care of avoiding namespaces collisions */
1594 if (blk_by_name(node_name
)) {
1595 error_setg(errp
, "node-name=%s is conflicting with a device id",
1600 /* takes care of avoiding duplicates node names */
1601 if (bdrv_find_node(node_name
)) {
1602 error_setg(errp
, "Duplicate nodes with node-name='%s'", node_name
);
1606 /* Make sure that the node name isn't truncated */
1607 if (strlen(node_name
) >= sizeof(bs
->node_name
)) {
1608 error_setg(errp
, "Node name too long");
1612 /* copy node name into the bs and insert it into the graph list */
1613 pstrcpy(bs
->node_name
, sizeof(bs
->node_name
), node_name
);
1614 QTAILQ_INSERT_TAIL(&graph_bdrv_states
, bs
, node_list
);
1616 g_free(gen_node_name
);
1620 * The caller must always hold @bs AioContext lock, because this function calls
1621 * bdrv_refresh_total_sectors() which polls when called from non-coroutine
1624 static int no_coroutine_fn GRAPH_UNLOCKED
1625 bdrv_open_driver(BlockDriverState
*bs
, BlockDriver
*drv
, const char *node_name
,
1626 QDict
*options
, int open_flags
, Error
**errp
)
1628 Error
*local_err
= NULL
;
1630 GLOBAL_STATE_CODE();
1632 bdrv_assign_node_name(bs
, node_name
, &local_err
);
1634 error_propagate(errp
, local_err
);
1639 bs
->opaque
= g_malloc0(drv
->instance_size
);
1641 if (drv
->bdrv_file_open
) {
1642 assert(!drv
->bdrv_needs_filename
|| bs
->filename
[0]);
1643 ret
= drv
->bdrv_file_open(bs
, options
, open_flags
, &local_err
);
1644 } else if (drv
->bdrv_open
) {
1645 ret
= drv
->bdrv_open(bs
, options
, open_flags
, &local_err
);
1652 error_propagate(errp
, local_err
);
1653 } else if (bs
->filename
[0]) {
1654 error_setg_errno(errp
, -ret
, "Could not open '%s'", bs
->filename
);
1656 error_setg_errno(errp
, -ret
, "Could not open image");
1661 assert(!(bs
->supported_read_flags
& ~BDRV_REQ_MASK
));
1662 assert(!(bs
->supported_write_flags
& ~BDRV_REQ_MASK
));
1665 * Always allow the BDRV_REQ_REGISTERED_BUF optimization hint. This saves
1666 * drivers that pass read/write requests through to a child the trouble of
1667 * declaring support explicitly.
1669 * Drivers must not propagate this flag accidentally when they initiate I/O
1670 * to a bounce buffer. That case should be rare though.
1672 bs
->supported_read_flags
|= BDRV_REQ_REGISTERED_BUF
;
1673 bs
->supported_write_flags
|= BDRV_REQ_REGISTERED_BUF
;
1675 ret
= bdrv_refresh_total_sectors(bs
, bs
->total_sectors
);
1677 error_setg_errno(errp
, -ret
, "Could not refresh total sector count");
1681 bdrv_graph_rdlock_main_loop();
1682 bdrv_refresh_limits(bs
, NULL
, &local_err
);
1683 bdrv_graph_rdunlock_main_loop();
1686 error_propagate(errp
, local_err
);
1690 assert(bdrv_opt_mem_align(bs
) != 0);
1691 assert(bdrv_min_mem_align(bs
) != 0);
1692 assert(is_power_of_2(bs
->bl
.request_alignment
));
1694 for (i
= 0; i
< bs
->quiesce_counter
; i
++) {
1695 if (drv
->bdrv_drain_begin
) {
1696 drv
->bdrv_drain_begin(bs
);
1704 bdrv_graph_wrlock();
1705 if (bs
->file
!= NULL
) {
1706 bdrv_unref_child(bs
, bs
->file
);
1709 bdrv_graph_wrunlock();
1717 * Create and open a block node.
1719 * @options is a QDict of options to pass to the block drivers, or NULL for an
1720 * empty set of options. The reference to the QDict belongs to the block layer
1721 * after the call (even on failure), so if the caller intends to reuse the
1722 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
1724 BlockDriverState
*bdrv_new_open_driver_opts(BlockDriver
*drv
,
1725 const char *node_name
,
1726 QDict
*options
, int flags
,
1729 BlockDriverState
*bs
;
1732 GLOBAL_STATE_CODE();
1735 bs
->open_flags
= flags
;
1736 bs
->options
= options
?: qdict_new();
1737 bs
->explicit_options
= qdict_clone_shallow(bs
->options
);
1740 update_options_from_flags(bs
->options
, flags
);
1742 ret
= bdrv_open_driver(bs
, drv
, node_name
, bs
->options
, flags
, errp
);
1744 qobject_unref(bs
->explicit_options
);
1745 bs
->explicit_options
= NULL
;
1746 qobject_unref(bs
->options
);
1755 /* Create and open a block node. */
1756 BlockDriverState
*bdrv_new_open_driver(BlockDriver
*drv
, const char *node_name
,
1757 int flags
, Error
**errp
)
1759 GLOBAL_STATE_CODE();
1760 return bdrv_new_open_driver_opts(drv
, node_name
, NULL
, flags
, errp
);
1763 QemuOptsList bdrv_runtime_opts
= {
1764 .name
= "bdrv_common",
1765 .head
= QTAILQ_HEAD_INITIALIZER(bdrv_runtime_opts
.head
),
1768 .name
= "node-name",
1769 .type
= QEMU_OPT_STRING
,
1770 .help
= "Node name of the block device node",
1774 .type
= QEMU_OPT_STRING
,
1775 .help
= "Block driver to use for the node",
1778 .name
= BDRV_OPT_CACHE_DIRECT
,
1779 .type
= QEMU_OPT_BOOL
,
1780 .help
= "Bypass software writeback cache on the host",
1783 .name
= BDRV_OPT_CACHE_NO_FLUSH
,
1784 .type
= QEMU_OPT_BOOL
,
1785 .help
= "Ignore flush requests",
1788 .name
= BDRV_OPT_READ_ONLY
,
1789 .type
= QEMU_OPT_BOOL
,
1790 .help
= "Node is opened in read-only mode",
1793 .name
= BDRV_OPT_AUTO_READ_ONLY
,
1794 .type
= QEMU_OPT_BOOL
,
1795 .help
= "Node can become read-only if opening read-write fails",
1798 .name
= "detect-zeroes",
1799 .type
= QEMU_OPT_STRING
,
1800 .help
= "try to optimize zero writes (off, on, unmap)",
1803 .name
= BDRV_OPT_DISCARD
,
1804 .type
= QEMU_OPT_STRING
,
1805 .help
= "discard operation (ignore/off, unmap/on)",
1808 .name
= BDRV_OPT_FORCE_SHARE
,
1809 .type
= QEMU_OPT_BOOL
,
1810 .help
= "always accept other writers (default: off)",
1812 { /* end of list */ }
1816 QemuOptsList bdrv_create_opts_simple
= {
1817 .name
= "simple-create-opts",
1818 .head
= QTAILQ_HEAD_INITIALIZER(bdrv_create_opts_simple
.head
),
1821 .name
= BLOCK_OPT_SIZE
,
1822 .type
= QEMU_OPT_SIZE
,
1823 .help
= "Virtual disk size"
1826 .name
= BLOCK_OPT_PREALLOC
,
1827 .type
= QEMU_OPT_STRING
,
1828 .help
= "Preallocation mode (allowed values: off)"
1830 { /* end of list */ }
1835 * Common part for opening disk images and files
1837 * Removes all processed options from *options.
1839 static int bdrv_open_common(BlockDriverState
*bs
, BlockBackend
*file
,
1840 QDict
*options
, Error
**errp
)
1842 int ret
, open_flags
;
1843 const char *filename
;
1844 const char *driver_name
= NULL
;
1845 const char *node_name
= NULL
;
1846 const char *discard
;
1849 Error
*local_err
= NULL
;
1852 GLOBAL_STATE_CODE();
1854 bdrv_graph_rdlock_main_loop();
1855 assert(bs
->file
== NULL
);
1856 assert(options
!= NULL
&& bs
->options
!= options
);
1857 bdrv_graph_rdunlock_main_loop();
1859 opts
= qemu_opts_create(&bdrv_runtime_opts
, NULL
, 0, &error_abort
);
1860 if (!qemu_opts_absorb_qdict(opts
, options
, errp
)) {
1865 update_flags_from_options(&bs
->open_flags
, opts
);
1867 driver_name
= qemu_opt_get(opts
, "driver");
1868 drv
= bdrv_find_format(driver_name
);
1869 assert(drv
!= NULL
);
1871 bs
->force_share
= qemu_opt_get_bool(opts
, BDRV_OPT_FORCE_SHARE
, false);
1873 if (bs
->force_share
&& (bs
->open_flags
& BDRV_O_RDWR
)) {
1875 BDRV_OPT_FORCE_SHARE
1876 "=on can only be used with read-only images");
1882 bdrv_graph_rdlock_main_loop();
1883 bdrv_refresh_filename(blk_bs(file
));
1884 bdrv_graph_rdunlock_main_loop();
1886 filename
= blk_bs(file
)->filename
;
1889 * Caution: while qdict_get_try_str() is fine, getting
1890 * non-string types would require more care. When @options
1891 * come from -blockdev or blockdev_add, its members are typed
1892 * according to the QAPI schema, but when they come from
1893 * -drive, they're all QString.
1895 filename
= qdict_get_try_str(options
, "filename");
1898 if (drv
->bdrv_needs_filename
&& (!filename
|| !filename
[0])) {
1899 error_setg(errp
, "The '%s' block driver requires a file name",
1905 trace_bdrv_open_common(bs
, filename
?: "", bs
->open_flags
,
1908 ro
= bdrv_is_read_only(bs
);
1910 if (use_bdrv_whitelist
&& !bdrv_is_whitelisted(drv
, ro
)) {
1911 if (!ro
&& bdrv_is_whitelisted(drv
, true)) {
1912 bdrv_graph_rdlock_main_loop();
1913 ret
= bdrv_apply_auto_read_only(bs
, NULL
, NULL
);
1914 bdrv_graph_rdunlock_main_loop();
1920 !ro
&& bdrv_is_whitelisted(drv
, true)
1921 ? "Driver '%s' can only be used for read-only devices"
1922 : "Driver '%s' is not whitelisted",
1928 /* bdrv_new() and bdrv_close() make it so */
1929 assert(qatomic_read(&bs
->copy_on_read
) == 0);
1931 if (bs
->open_flags
& BDRV_O_COPY_ON_READ
) {
1933 bdrv_enable_copy_on_read(bs
);
1935 error_setg(errp
, "Can't use copy-on-read on read-only device");
1941 discard
= qemu_opt_get(opts
, BDRV_OPT_DISCARD
);
1942 if (discard
!= NULL
) {
1943 if (bdrv_parse_discard_flags(discard
, &bs
->open_flags
) != 0) {
1944 error_setg(errp
, "Invalid discard option");
1951 bdrv_parse_detect_zeroes(opts
, bs
->open_flags
, &local_err
);
1953 error_propagate(errp
, local_err
);
1958 if (filename
!= NULL
) {
1959 pstrcpy(bs
->filename
, sizeof(bs
->filename
), filename
);
1961 bs
->filename
[0] = '\0';
1963 pstrcpy(bs
->exact_filename
, sizeof(bs
->exact_filename
), bs
->filename
);
1965 /* Open the image, either directly or using a protocol */
1966 open_flags
= bdrv_open_flags(bs
, bs
->open_flags
);
1967 node_name
= qemu_opt_get(opts
, "node-name");
1969 assert(!drv
->bdrv_file_open
|| file
== NULL
);
1970 ret
= bdrv_open_driver(bs
, drv
, node_name
, options
, open_flags
, errp
);
1975 qemu_opts_del(opts
);
1979 qemu_opts_del(opts
);
1983 static QDict
*parse_json_filename(const char *filename
, Error
**errp
)
1985 QObject
*options_obj
;
1988 GLOBAL_STATE_CODE();
1990 ret
= strstart(filename
, "json:", &filename
);
1993 options_obj
= qobject_from_json(filename
, errp
);
1995 error_prepend(errp
, "Could not parse the JSON options: ");
1999 options
= qobject_to(QDict
, options_obj
);
2001 qobject_unref(options_obj
);
2002 error_setg(errp
, "Invalid JSON object given");
2006 qdict_flatten(options
);
2011 static void parse_json_protocol(QDict
*options
, const char **pfilename
,
2014 QDict
*json_options
;
2015 Error
*local_err
= NULL
;
2016 GLOBAL_STATE_CODE();
2018 /* Parse json: pseudo-protocol */
2019 if (!*pfilename
|| !g_str_has_prefix(*pfilename
, "json:")) {
2023 json_options
= parse_json_filename(*pfilename
, &local_err
);
2025 error_propagate(errp
, local_err
);
2029 /* Options given in the filename have lower priority than options
2030 * specified directly */
2031 qdict_join(options
, json_options
, false);
2032 qobject_unref(json_options
);
2037 * Fills in default options for opening images and converts the legacy
2038 * filename/flags pair to option QDict entries.
2039 * The BDRV_O_PROTOCOL flag in *flags will be set or cleared accordingly if a
2040 * block driver has been specified explicitly.
2042 static int bdrv_fill_options(QDict
**options
, const char *filename
,
2043 int *flags
, Error
**errp
)
2045 const char *drvname
;
2046 bool protocol
= *flags
& BDRV_O_PROTOCOL
;
2047 bool parse_filename
= false;
2048 BlockDriver
*drv
= NULL
;
2049 Error
*local_err
= NULL
;
2051 GLOBAL_STATE_CODE();
2054 * Caution: while qdict_get_try_str() is fine, getting non-string
2055 * types would require more care. When @options come from
2056 * -blockdev or blockdev_add, its members are typed according to
2057 * the QAPI schema, but when they come from -drive, they're all
2060 drvname
= qdict_get_try_str(*options
, "driver");
2062 drv
= bdrv_find_format(drvname
);
2064 error_setg(errp
, "Unknown driver '%s'", drvname
);
2067 /* If the user has explicitly specified the driver, this choice should
2068 * override the BDRV_O_PROTOCOL flag */
2069 protocol
= drv
->bdrv_file_open
;
2073 *flags
|= BDRV_O_PROTOCOL
;
2075 *flags
&= ~BDRV_O_PROTOCOL
;
2078 /* Translate cache options from flags into options */
2079 update_options_from_flags(*options
, *flags
);
2081 /* Fetch the file name from the options QDict if necessary */
2082 if (protocol
&& filename
) {
2083 if (!qdict_haskey(*options
, "filename")) {
2084 qdict_put_str(*options
, "filename", filename
);
2085 parse_filename
= true;
2087 error_setg(errp
, "Can't specify 'file' and 'filename' options at "
2093 /* Find the right block driver */
2094 /* See cautionary note on accessing @options above */
2095 filename
= qdict_get_try_str(*options
, "filename");
2097 if (!drvname
&& protocol
) {
2099 drv
= bdrv_find_protocol(filename
, parse_filename
, errp
);
2104 drvname
= drv
->format_name
;
2105 qdict_put_str(*options
, "driver", drvname
);
2107 error_setg(errp
, "Must specify either driver or file");
2112 assert(drv
|| !protocol
);
2114 /* Driver-specific filename parsing */
2115 if (drv
&& drv
->bdrv_parse_filename
&& parse_filename
) {
2116 drv
->bdrv_parse_filename(filename
, *options
, &local_err
);
2118 error_propagate(errp
, local_err
);
2122 if (!drv
->bdrv_needs_filename
) {
2123 qdict_del(*options
, "filename");
2130 typedef struct BlockReopenQueueEntry
{
2132 BDRVReopenState state
;
2133 QTAILQ_ENTRY(BlockReopenQueueEntry
) entry
;
2134 } BlockReopenQueueEntry
;
2137 * Return the flags that @bs will have after the reopens in @q have
2138 * successfully completed. If @q is NULL (or @bs is not contained in @q),
2139 * return the current flags.
2141 static int bdrv_reopen_get_flags(BlockReopenQueue
*q
, BlockDriverState
*bs
)
2143 BlockReopenQueueEntry
*entry
;
2146 QTAILQ_FOREACH(entry
, q
, entry
) {
2147 if (entry
->state
.bs
== bs
) {
2148 return entry
->state
.flags
;
2153 return bs
->open_flags
;
2156 /* Returns whether the image file can be written to after the reopen queue @q
2157 * has been successfully applied, or right now if @q is NULL. */
2158 static bool bdrv_is_writable_after_reopen(BlockDriverState
*bs
,
2159 BlockReopenQueue
*q
)
2161 int flags
= bdrv_reopen_get_flags(q
, bs
);
2163 return (flags
& (BDRV_O_RDWR
| BDRV_O_INACTIVE
)) == BDRV_O_RDWR
;
2167 * Return whether the BDS can be written to. This is not necessarily
2168 * the same as !bdrv_is_read_only(bs), as inactivated images may not
2169 * be written to but do not count as read-only images.
2171 bool bdrv_is_writable(BlockDriverState
*bs
)
2174 return bdrv_is_writable_after_reopen(bs
, NULL
);
2177 static char *bdrv_child_user_desc(BdrvChild
*c
)
2179 GLOBAL_STATE_CODE();
2180 return c
->klass
->get_parent_desc(c
);
2184 * Check that @a allows everything that @b needs. @a and @b must reference same
2187 static bool bdrv_a_allow_b(BdrvChild
*a
, BdrvChild
*b
, Error
**errp
)
2189 const char *child_bs_name
;
2190 g_autofree
char *a_user
= NULL
;
2191 g_autofree
char *b_user
= NULL
;
2192 g_autofree
char *perms
= NULL
;
2195 assert(a
->bs
== b
->bs
);
2196 GLOBAL_STATE_CODE();
2198 if ((b
->perm
& a
->shared_perm
) == b
->perm
) {
2202 child_bs_name
= bdrv_get_node_name(b
->bs
);
2203 a_user
= bdrv_child_user_desc(a
);
2204 b_user
= bdrv_child_user_desc(b
);
2205 perms
= bdrv_perm_names(b
->perm
& ~a
->shared_perm
);
2207 error_setg(errp
, "Permission conflict on node '%s': permissions '%s' are "
2208 "both required by %s (uses node '%s' as '%s' child) and "
2209 "unshared by %s (uses node '%s' as '%s' child).",
2210 child_bs_name
, perms
,
2211 b_user
, child_bs_name
, b
->name
,
2212 a_user
, child_bs_name
, a
->name
);
2217 static bool GRAPH_RDLOCK
2218 bdrv_parent_perms_conflict(BlockDriverState
*bs
, Error
**errp
)
2221 GLOBAL_STATE_CODE();
2224 * During the loop we'll look at each pair twice. That's correct because
2225 * bdrv_a_allow_b() is asymmetric and we should check each pair in both
2228 QLIST_FOREACH(a
, &bs
->parents
, next_parent
) {
2229 QLIST_FOREACH(b
, &bs
->parents
, next_parent
) {
2234 if (!bdrv_a_allow_b(a
, b
, errp
)) {
2243 static void GRAPH_RDLOCK
2244 bdrv_child_perm(BlockDriverState
*bs
, BlockDriverState
*child_bs
,
2245 BdrvChild
*c
, BdrvChildRole role
,
2246 BlockReopenQueue
*reopen_queue
,
2247 uint64_t parent_perm
, uint64_t parent_shared
,
2248 uint64_t *nperm
, uint64_t *nshared
)
2250 assert(bs
->drv
&& bs
->drv
->bdrv_child_perm
);
2251 GLOBAL_STATE_CODE();
2252 bs
->drv
->bdrv_child_perm(bs
, c
, role
, reopen_queue
,
2253 parent_perm
, parent_shared
,
2255 /* TODO Take force_share from reopen_queue */
2256 if (child_bs
&& child_bs
->force_share
) {
2257 *nshared
= BLK_PERM_ALL
;
2262 * Adds the whole subtree of @bs (including @bs itself) to the @list (except for
2263 * nodes that are already in the @list, of course) so that final list is
2264 * topologically sorted. Return the result (GSList @list object is updated, so
2265 * don't use old reference after function call).
2267 * On function start @list must be already topologically sorted and for any node
2268 * in the @list the whole subtree of the node must be in the @list as well. The
2269 * simplest way to satisfy this criteria: use only result of
2270 * bdrv_topological_dfs() or NULL as @list parameter.
2272 static GSList
* GRAPH_RDLOCK
2273 bdrv_topological_dfs(GSList
*list
, GHashTable
*found
, BlockDriverState
*bs
)
2276 g_autoptr(GHashTable
) local_found
= NULL
;
2278 GLOBAL_STATE_CODE();
2282 found
= local_found
= g_hash_table_new(NULL
, NULL
);
2285 if (g_hash_table_contains(found
, bs
)) {
2288 g_hash_table_add(found
, bs
);
2290 QLIST_FOREACH(child
, &bs
->children
, next
) {
2291 list
= bdrv_topological_dfs(list
, found
, child
->bs
);
2294 return g_slist_prepend(list
, bs
);
2297 typedef struct BdrvChildSetPermState
{
2300 uint64_t old_shared_perm
;
2301 } BdrvChildSetPermState
;
2303 static void bdrv_child_set_perm_abort(void *opaque
)
2305 BdrvChildSetPermState
*s
= opaque
;
2307 GLOBAL_STATE_CODE();
2309 s
->child
->perm
= s
->old_perm
;
2310 s
->child
->shared_perm
= s
->old_shared_perm
;
2313 static TransactionActionDrv bdrv_child_set_pem_drv
= {
2314 .abort
= bdrv_child_set_perm_abort
,
2318 static void bdrv_child_set_perm(BdrvChild
*c
, uint64_t perm
,
2319 uint64_t shared
, Transaction
*tran
)
2321 BdrvChildSetPermState
*s
= g_new(BdrvChildSetPermState
, 1);
2322 GLOBAL_STATE_CODE();
2324 *s
= (BdrvChildSetPermState
) {
2326 .old_perm
= c
->perm
,
2327 .old_shared_perm
= c
->shared_perm
,
2331 c
->shared_perm
= shared
;
2333 tran_add(tran
, &bdrv_child_set_pem_drv
, s
);
2336 static void GRAPH_RDLOCK
bdrv_drv_set_perm_commit(void *opaque
)
2338 BlockDriverState
*bs
= opaque
;
2339 uint64_t cumulative_perms
, cumulative_shared_perms
;
2340 GLOBAL_STATE_CODE();
2342 if (bs
->drv
->bdrv_set_perm
) {
2343 bdrv_get_cumulative_perm(bs
, &cumulative_perms
,
2344 &cumulative_shared_perms
);
2345 bs
->drv
->bdrv_set_perm(bs
, cumulative_perms
, cumulative_shared_perms
);
2349 static void GRAPH_RDLOCK
bdrv_drv_set_perm_abort(void *opaque
)
2351 BlockDriverState
*bs
= opaque
;
2352 GLOBAL_STATE_CODE();
2354 if (bs
->drv
->bdrv_abort_perm_update
) {
2355 bs
->drv
->bdrv_abort_perm_update(bs
);
2359 TransactionActionDrv bdrv_drv_set_perm_drv
= {
2360 .abort
= bdrv_drv_set_perm_abort
,
2361 .commit
= bdrv_drv_set_perm_commit
,
2365 * After calling this function, the transaction @tran may only be completed
2366 * while holding a reader lock for the graph.
2368 static int GRAPH_RDLOCK
2369 bdrv_drv_set_perm(BlockDriverState
*bs
, uint64_t perm
, uint64_t shared_perm
,
2370 Transaction
*tran
, Error
**errp
)
2372 GLOBAL_STATE_CODE();
2377 if (bs
->drv
->bdrv_check_perm
) {
2378 int ret
= bs
->drv
->bdrv_check_perm(bs
, perm
, shared_perm
, errp
);
2385 tran_add(tran
, &bdrv_drv_set_perm_drv
, bs
);
2391 typedef struct BdrvReplaceChildState
{
2393 BlockDriverState
*old_bs
;
2394 } BdrvReplaceChildState
;
2396 static void GRAPH_WRLOCK
bdrv_replace_child_commit(void *opaque
)
2398 BdrvReplaceChildState
*s
= opaque
;
2399 GLOBAL_STATE_CODE();
2401 bdrv_schedule_unref(s
->old_bs
);
2404 static void GRAPH_WRLOCK
bdrv_replace_child_abort(void *opaque
)
2406 BdrvReplaceChildState
*s
= opaque
;
2407 BlockDriverState
*new_bs
= s
->child
->bs
;
2409 GLOBAL_STATE_CODE();
2410 assert_bdrv_graph_writable();
2412 /* old_bs reference is transparently moved from @s to @s->child */
2413 if (!s
->child
->bs
) {
2415 * The parents were undrained when removing old_bs from the child. New
2416 * requests can't have been made, though, because the child was empty.
2418 * TODO Make bdrv_replace_child_noperm() transactionable to avoid
2419 * undraining the parent in the first place. Once this is done, having
2420 * new_bs drained when calling bdrv_replace_child_tran() is not a
2421 * requirement any more.
2423 bdrv_parent_drained_begin_single(s
->child
);
2424 assert(!bdrv_parent_drained_poll_single(s
->child
));
2426 assert(s
->child
->quiesced_parent
);
2427 bdrv_replace_child_noperm(s
->child
, s
->old_bs
);
2432 static TransactionActionDrv bdrv_replace_child_drv
= {
2433 .commit
= bdrv_replace_child_commit
,
2434 .abort
= bdrv_replace_child_abort
,
2439 * bdrv_replace_child_tran
2441 * Note: real unref of old_bs is done only on commit.
2443 * Both @child->bs and @new_bs (if non-NULL) must be drained. @new_bs must be
2444 * kept drained until the transaction is completed.
2446 * After calling this function, the transaction @tran may only be completed
2447 * while holding a writer lock for the graph.
2449 * The function doesn't update permissions, caller is responsible for this.
2451 static void GRAPH_WRLOCK
2452 bdrv_replace_child_tran(BdrvChild
*child
, BlockDriverState
*new_bs
,
2455 BdrvReplaceChildState
*s
= g_new(BdrvReplaceChildState
, 1);
2457 assert(child
->quiesced_parent
);
2458 assert(!new_bs
|| new_bs
->quiesce_counter
);
2460 *s
= (BdrvReplaceChildState
) {
2462 .old_bs
= child
->bs
,
2464 tran_add(tran
, &bdrv_replace_child_drv
, s
);
2470 bdrv_replace_child_noperm(child
, new_bs
);
2471 /* old_bs reference is transparently moved from @child to @s */
2475 * Refresh permissions in @bs subtree. The function is intended to be called
2476 * after some graph modification that was done without permission update.
2478 * After calling this function, the transaction @tran may only be completed
2479 * while holding a reader lock for the graph.
2481 static int GRAPH_RDLOCK
2482 bdrv_node_refresh_perm(BlockDriverState
*bs
, BlockReopenQueue
*q
,
2483 Transaction
*tran
, Error
**errp
)
2485 BlockDriver
*drv
= bs
->drv
;
2488 uint64_t cumulative_perms
, cumulative_shared_perms
;
2489 GLOBAL_STATE_CODE();
2491 bdrv_get_cumulative_perm(bs
, &cumulative_perms
, &cumulative_shared_perms
);
2493 /* Write permissions never work with read-only images */
2494 if ((cumulative_perms
& (BLK_PERM_WRITE
| BLK_PERM_WRITE_UNCHANGED
)) &&
2495 !bdrv_is_writable_after_reopen(bs
, q
))
2497 if (!bdrv_is_writable_after_reopen(bs
, NULL
)) {
2498 error_setg(errp
, "Block node is read-only");
2500 error_setg(errp
, "Read-only block node '%s' cannot support "
2501 "read-write users", bdrv_get_node_name(bs
));
2508 * Unaligned requests will automatically be aligned to bl.request_alignment
2509 * and without RESIZE we can't extend requests to write to space beyond the
2510 * end of the image, so it's required that the image size is aligned.
2512 if ((cumulative_perms
& (BLK_PERM_WRITE
| BLK_PERM_WRITE_UNCHANGED
)) &&
2513 !(cumulative_perms
& BLK_PERM_RESIZE
))
2515 if ((bs
->total_sectors
* BDRV_SECTOR_SIZE
) % bs
->bl
.request_alignment
) {
2516 error_setg(errp
, "Cannot get 'write' permission without 'resize': "
2517 "Image size is not a multiple of request "
2523 /* Check this node */
2528 ret
= bdrv_drv_set_perm(bs
, cumulative_perms
, cumulative_shared_perms
, tran
,
2534 /* Drivers that never have children can omit .bdrv_child_perm() */
2535 if (!drv
->bdrv_child_perm
) {
2536 assert(QLIST_EMPTY(&bs
->children
));
2540 /* Check all children */
2541 QLIST_FOREACH(c
, &bs
->children
, next
) {
2542 uint64_t cur_perm
, cur_shared
;
2544 bdrv_child_perm(bs
, c
->bs
, c
, c
->role
, q
,
2545 cumulative_perms
, cumulative_shared_perms
,
2546 &cur_perm
, &cur_shared
);
2547 bdrv_child_set_perm(c
, cur_perm
, cur_shared
, tran
);
2554 * @list is a product of bdrv_topological_dfs() (may be called several times) -
2555 * a topologically sorted subgraph.
2557 * After calling this function, the transaction @tran may only be completed
2558 * while holding a reader lock for the graph.
2560 static int GRAPH_RDLOCK
2561 bdrv_do_refresh_perms(GSList
*list
, BlockReopenQueue
*q
, Transaction
*tran
,
2565 BlockDriverState
*bs
;
2566 GLOBAL_STATE_CODE();
2568 for ( ; list
; list
= list
->next
) {
2571 if (bdrv_parent_perms_conflict(bs
, errp
)) {
2575 ret
= bdrv_node_refresh_perm(bs
, q
, tran
, errp
);
2585 * @list is any list of nodes. List is completed by all subtrees and
2586 * topologically sorted. It's not a problem if some node occurs in the @list
2589 * After calling this function, the transaction @tran may only be completed
2590 * while holding a reader lock for the graph.
2592 static int GRAPH_RDLOCK
2593 bdrv_list_refresh_perms(GSList
*list
, BlockReopenQueue
*q
, Transaction
*tran
,
2596 g_autoptr(GHashTable
) found
= g_hash_table_new(NULL
, NULL
);
2597 g_autoptr(GSList
) refresh_list
= NULL
;
2599 for ( ; list
; list
= list
->next
) {
2600 refresh_list
= bdrv_topological_dfs(refresh_list
, found
, list
->data
);
2603 return bdrv_do_refresh_perms(refresh_list
, q
, tran
, errp
);
2606 void bdrv_get_cumulative_perm(BlockDriverState
*bs
, uint64_t *perm
,
2607 uint64_t *shared_perm
)
2610 uint64_t cumulative_perms
= 0;
2611 uint64_t cumulative_shared_perms
= BLK_PERM_ALL
;
2613 GLOBAL_STATE_CODE();
2615 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
2616 cumulative_perms
|= c
->perm
;
2617 cumulative_shared_perms
&= c
->shared_perm
;
2620 *perm
= cumulative_perms
;
2621 *shared_perm
= cumulative_shared_perms
;
2624 char *bdrv_perm_names(uint64_t perm
)
2630 { BLK_PERM_CONSISTENT_READ
, "consistent read" },
2631 { BLK_PERM_WRITE
, "write" },
2632 { BLK_PERM_WRITE_UNCHANGED
, "write unchanged" },
2633 { BLK_PERM_RESIZE
, "resize" },
2637 GString
*result
= g_string_sized_new(30);
2638 struct perm_name
*p
;
2640 for (p
= permissions
; p
->name
; p
++) {
2641 if (perm
& p
->perm
) {
2642 if (result
->len
> 0) {
2643 g_string_append(result
, ", ");
2645 g_string_append(result
, p
->name
);
2649 return g_string_free(result
, FALSE
);
2654 * @tran is allowed to be NULL. In this case no rollback is possible.
2656 * After calling this function, the transaction @tran may only be completed
2657 * while holding a reader lock for the graph.
2659 static int GRAPH_RDLOCK
2660 bdrv_refresh_perms(BlockDriverState
*bs
, Transaction
*tran
, Error
**errp
)
2663 Transaction
*local_tran
= NULL
;
2664 g_autoptr(GSList
) list
= bdrv_topological_dfs(NULL
, NULL
, bs
);
2665 GLOBAL_STATE_CODE();
2668 tran
= local_tran
= tran_new();
2671 ret
= bdrv_do_refresh_perms(list
, NULL
, tran
, errp
);
2674 tran_finalize(local_tran
, ret
);
2680 int bdrv_child_try_set_perm(BdrvChild
*c
, uint64_t perm
, uint64_t shared
,
2683 Error
*local_err
= NULL
;
2684 Transaction
*tran
= tran_new();
2687 GLOBAL_STATE_CODE();
2689 bdrv_child_set_perm(c
, perm
, shared
, tran
);
2691 ret
= bdrv_refresh_perms(c
->bs
, tran
, &local_err
);
2693 tran_finalize(tran
, ret
);
2696 if ((perm
& ~c
->perm
) || (c
->shared_perm
& ~shared
)) {
2697 /* tighten permissions */
2698 error_propagate(errp
, local_err
);
2701 * Our caller may intend to only loosen restrictions and
2702 * does not expect this function to fail. Errors are not
2703 * fatal in such a case, so we can just hide them from our
2706 error_free(local_err
);
2714 int bdrv_child_refresh_perms(BlockDriverState
*bs
, BdrvChild
*c
, Error
**errp
)
2716 uint64_t parent_perms
, parent_shared
;
2717 uint64_t perms
, shared
;
2719 GLOBAL_STATE_CODE();
2721 bdrv_get_cumulative_perm(bs
, &parent_perms
, &parent_shared
);
2722 bdrv_child_perm(bs
, c
->bs
, c
, c
->role
, NULL
,
2723 parent_perms
, parent_shared
, &perms
, &shared
);
2725 return bdrv_child_try_set_perm(c
, perms
, shared
, errp
);
2729 * Default implementation for .bdrv_child_perm() for block filters:
2730 * Forward CONSISTENT_READ, WRITE, WRITE_UNCHANGED, and RESIZE to the
2733 static void bdrv_filter_default_perms(BlockDriverState
*bs
, BdrvChild
*c
,
2735 BlockReopenQueue
*reopen_queue
,
2736 uint64_t perm
, uint64_t shared
,
2737 uint64_t *nperm
, uint64_t *nshared
)
2739 GLOBAL_STATE_CODE();
2740 *nperm
= perm
& DEFAULT_PERM_PASSTHROUGH
;
2741 *nshared
= (shared
& DEFAULT_PERM_PASSTHROUGH
) | DEFAULT_PERM_UNCHANGED
;
2744 static void bdrv_default_perms_for_cow(BlockDriverState
*bs
, BdrvChild
*c
,
2746 BlockReopenQueue
*reopen_queue
,
2747 uint64_t perm
, uint64_t shared
,
2748 uint64_t *nperm
, uint64_t *nshared
)
2750 assert(role
& BDRV_CHILD_COW
);
2751 GLOBAL_STATE_CODE();
2754 * We want consistent read from backing files if the parent needs it.
2755 * No other operations are performed on backing files.
2757 perm
&= BLK_PERM_CONSISTENT_READ
;
2760 * If the parent can deal with changing data, we're okay with a
2761 * writable and resizable backing file.
2762 * TODO Require !(perm & BLK_PERM_CONSISTENT_READ), too?
2764 if (shared
& BLK_PERM_WRITE
) {
2765 shared
= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2770 shared
|= BLK_PERM_CONSISTENT_READ
| BLK_PERM_WRITE_UNCHANGED
;
2772 if (bs
->open_flags
& BDRV_O_INACTIVE
) {
2773 shared
|= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2780 static void bdrv_default_perms_for_storage(BlockDriverState
*bs
, BdrvChild
*c
,
2782 BlockReopenQueue
*reopen_queue
,
2783 uint64_t perm
, uint64_t shared
,
2784 uint64_t *nperm
, uint64_t *nshared
)
2788 GLOBAL_STATE_CODE();
2789 assert(role
& (BDRV_CHILD_METADATA
| BDRV_CHILD_DATA
));
2791 flags
= bdrv_reopen_get_flags(reopen_queue
, bs
);
2794 * Apart from the modifications below, the same permissions are
2795 * forwarded and left alone as for filters
2797 bdrv_filter_default_perms(bs
, c
, role
, reopen_queue
,
2798 perm
, shared
, &perm
, &shared
);
2800 if (role
& BDRV_CHILD_METADATA
) {
2801 /* Format drivers may touch metadata even if the guest doesn't write */
2802 if (bdrv_is_writable_after_reopen(bs
, reopen_queue
)) {
2803 perm
|= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2807 * bs->file always needs to be consistent because of the
2808 * metadata. We can never allow other users to resize or write
2811 if (!(flags
& BDRV_O_NO_IO
)) {
2812 perm
|= BLK_PERM_CONSISTENT_READ
;
2814 shared
&= ~(BLK_PERM_WRITE
| BLK_PERM_RESIZE
);
2817 if (role
& BDRV_CHILD_DATA
) {
2819 * Technically, everything in this block is a subset of the
2820 * BDRV_CHILD_METADATA path taken above, and so this could
2821 * be an "else if" branch. However, that is not obvious, and
2822 * this function is not performance critical, therefore we let
2823 * this be an independent "if".
2827 * We cannot allow other users to resize the file because the
2828 * format driver might have some assumptions about the size
2829 * (e.g. because it is stored in metadata, or because the file
2830 * is split into fixed-size data files).
2832 shared
&= ~BLK_PERM_RESIZE
;
2835 * WRITE_UNCHANGED often cannot be performed as such on the
2836 * data file. For example, the qcow2 driver may still need to
2837 * write copied clusters on copy-on-read.
2839 if (perm
& BLK_PERM_WRITE_UNCHANGED
) {
2840 perm
|= BLK_PERM_WRITE
;
2844 * If the data file is written to, the format driver may
2845 * expect to be able to resize it by writing beyond the EOF.
2847 if (perm
& BLK_PERM_WRITE
) {
2848 perm
|= BLK_PERM_RESIZE
;
2852 if (bs
->open_flags
& BDRV_O_INACTIVE
) {
2853 shared
|= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2860 void bdrv_default_perms(BlockDriverState
*bs
, BdrvChild
*c
,
2861 BdrvChildRole role
, BlockReopenQueue
*reopen_queue
,
2862 uint64_t perm
, uint64_t shared
,
2863 uint64_t *nperm
, uint64_t *nshared
)
2865 GLOBAL_STATE_CODE();
2866 if (role
& BDRV_CHILD_FILTERED
) {
2867 assert(!(role
& (BDRV_CHILD_DATA
| BDRV_CHILD_METADATA
|
2869 bdrv_filter_default_perms(bs
, c
, role
, reopen_queue
,
2870 perm
, shared
, nperm
, nshared
);
2871 } else if (role
& BDRV_CHILD_COW
) {
2872 assert(!(role
& (BDRV_CHILD_DATA
| BDRV_CHILD_METADATA
)));
2873 bdrv_default_perms_for_cow(bs
, c
, role
, reopen_queue
,
2874 perm
, shared
, nperm
, nshared
);
2875 } else if (role
& (BDRV_CHILD_METADATA
| BDRV_CHILD_DATA
)) {
2876 bdrv_default_perms_for_storage(bs
, c
, role
, reopen_queue
,
2877 perm
, shared
, nperm
, nshared
);
2879 g_assert_not_reached();
2883 uint64_t bdrv_qapi_perm_to_blk_perm(BlockPermission qapi_perm
)
2885 static const uint64_t permissions
[] = {
2886 [BLOCK_PERMISSION_CONSISTENT_READ
] = BLK_PERM_CONSISTENT_READ
,
2887 [BLOCK_PERMISSION_WRITE
] = BLK_PERM_WRITE
,
2888 [BLOCK_PERMISSION_WRITE_UNCHANGED
] = BLK_PERM_WRITE_UNCHANGED
,
2889 [BLOCK_PERMISSION_RESIZE
] = BLK_PERM_RESIZE
,
2892 QEMU_BUILD_BUG_ON(ARRAY_SIZE(permissions
) != BLOCK_PERMISSION__MAX
);
2893 QEMU_BUILD_BUG_ON(1UL << ARRAY_SIZE(permissions
) != BLK_PERM_ALL
+ 1);
2895 assert(qapi_perm
< BLOCK_PERMISSION__MAX
);
2897 return permissions
[qapi_perm
];
2901 * Replaces the node that a BdrvChild points to without updating permissions.
2903 * If @new_bs is non-NULL, the parent of @child must already be drained through
2904 * @child and the caller must hold the AioContext lock for @new_bs.
2906 static void GRAPH_WRLOCK
2907 bdrv_replace_child_noperm(BdrvChild
*child
, BlockDriverState
*new_bs
)
2909 BlockDriverState
*old_bs
= child
->bs
;
2910 int new_bs_quiesce_counter
;
2912 assert(!child
->frozen
);
2915 * If we want to change the BdrvChild to point to a drained node as its new
2916 * child->bs, we need to make sure that its new parent is drained, too. In
2917 * other words, either child->quiesce_parent must already be true or we must
2918 * be able to set it and keep the parent's quiesce_counter consistent with
2919 * that, but without polling or starting new requests (this function
2920 * guarantees that it doesn't poll, and starting new requests would be
2921 * against the invariants of drain sections).
2923 * To keep things simple, we pick the first option (child->quiesce_parent
2924 * must already be true). We also generalise the rule a bit to make it
2925 * easier to verify in callers and more likely to be covered in test cases:
2926 * The parent must be quiesced through this child even if new_bs isn't
2927 * currently drained.
2929 * The only exception is for callers that always pass new_bs == NULL. In
2930 * this case, we obviously never need to consider the case of a drained
2931 * new_bs, so we can keep the callers simpler by allowing them not to drain
2934 assert(!new_bs
|| child
->quiesced_parent
);
2935 assert(old_bs
!= new_bs
);
2936 GLOBAL_STATE_CODE();
2938 if (old_bs
&& new_bs
) {
2939 assert(bdrv_get_aio_context(old_bs
) == bdrv_get_aio_context(new_bs
));
2943 if (child
->klass
->detach
) {
2944 child
->klass
->detach(child
);
2946 QLIST_REMOVE(child
, next_parent
);
2952 QLIST_INSERT_HEAD(&new_bs
->parents
, child
, next_parent
);
2953 if (child
->klass
->attach
) {
2954 child
->klass
->attach(child
);
2959 * If the parent was drained through this BdrvChild previously, but new_bs
2960 * is not drained, allow requests to come in only after the new node has
2963 new_bs_quiesce_counter
= (new_bs
? new_bs
->quiesce_counter
: 0);
2964 if (!new_bs_quiesce_counter
&& child
->quiesced_parent
) {
2965 bdrv_parent_drained_end_single(child
);
2970 * Free the given @child.
2972 * The child must be empty (i.e. `child->bs == NULL`) and it must be
2973 * unused (i.e. not in a children list).
2975 static void bdrv_child_free(BdrvChild
*child
)
2978 GLOBAL_STATE_CODE();
2979 GRAPH_RDLOCK_GUARD_MAINLOOP();
2981 assert(!child
->next
.le_prev
); /* not in children list */
2983 g_free(child
->name
);
2987 typedef struct BdrvAttachChildCommonState
{
2989 AioContext
*old_parent_ctx
;
2990 AioContext
*old_child_ctx
;
2991 } BdrvAttachChildCommonState
;
2993 static void GRAPH_WRLOCK
bdrv_attach_child_common_abort(void *opaque
)
2995 BdrvAttachChildCommonState
*s
= opaque
;
2996 BlockDriverState
*bs
= s
->child
->bs
;
2998 GLOBAL_STATE_CODE();
2999 assert_bdrv_graph_writable();
3001 bdrv_replace_child_noperm(s
->child
, NULL
);
3003 if (bdrv_get_aio_context(bs
) != s
->old_child_ctx
) {
3004 bdrv_try_change_aio_context(bs
, s
->old_child_ctx
, NULL
, &error_abort
);
3007 if (bdrv_child_get_parent_aio_context(s
->child
) != s
->old_parent_ctx
) {
3009 GHashTable
*visited
;
3014 /* No need to visit `child`, because it has been detached already */
3015 visited
= g_hash_table_new(NULL
, NULL
);
3016 ret
= s
->child
->klass
->change_aio_ctx(s
->child
, s
->old_parent_ctx
,
3017 visited
, tran
, &error_abort
);
3018 g_hash_table_destroy(visited
);
3020 /* transaction is supposed to always succeed */
3021 assert(ret
== true);
3025 bdrv_schedule_unref(bs
);
3026 bdrv_child_free(s
->child
);
3029 static TransactionActionDrv bdrv_attach_child_common_drv
= {
3030 .abort
= bdrv_attach_child_common_abort
,
3035 * Common part of attaching bdrv child to bs or to blk or to job
3037 * Function doesn't update permissions, caller is responsible for this.
3039 * After calling this function, the transaction @tran may only be completed
3040 * while holding a writer lock for the graph.
3042 * Returns new created child.
3044 * The caller must hold the AioContext lock for @child_bs. Both @parent_bs and
3045 * @child_bs can move to a different AioContext in this function. Callers must
3046 * make sure that their AioContext locking is still correct after this.
3048 static BdrvChild
* GRAPH_WRLOCK
3049 bdrv_attach_child_common(BlockDriverState
*child_bs
,
3050 const char *child_name
,
3051 const BdrvChildClass
*child_class
,
3052 BdrvChildRole child_role
,
3053 uint64_t perm
, uint64_t shared_perm
,
3055 Transaction
*tran
, Error
**errp
)
3057 BdrvChild
*new_child
;
3058 AioContext
*parent_ctx
;
3059 AioContext
*child_ctx
= bdrv_get_aio_context(child_bs
);
3061 assert(child_class
->get_parent_desc
);
3062 GLOBAL_STATE_CODE();
3064 new_child
= g_new(BdrvChild
, 1);
3065 *new_child
= (BdrvChild
) {
3067 .name
= g_strdup(child_name
),
3068 .klass
= child_class
,
3071 .shared_perm
= shared_perm
,
3076 * If the AioContexts don't match, first try to move the subtree of
3077 * child_bs into the AioContext of the new parent. If this doesn't work,
3078 * try moving the parent into the AioContext of child_bs instead.
3080 parent_ctx
= bdrv_child_get_parent_aio_context(new_child
);
3081 if (child_ctx
!= parent_ctx
) {
3082 Error
*local_err
= NULL
;
3083 int ret
= bdrv_try_change_aio_context(child_bs
, parent_ctx
, NULL
,
3086 if (ret
< 0 && child_class
->change_aio_ctx
) {
3087 Transaction
*aio_ctx_tran
= tran_new();
3088 GHashTable
*visited
= g_hash_table_new(NULL
, NULL
);
3091 g_hash_table_add(visited
, new_child
);
3092 ret_child
= child_class
->change_aio_ctx(new_child
, child_ctx
,
3093 visited
, aio_ctx_tran
,
3095 if (ret_child
== true) {
3096 error_free(local_err
);
3099 tran_finalize(aio_ctx_tran
, ret_child
== true ? 0 : -1);
3100 g_hash_table_destroy(visited
);
3104 error_propagate(errp
, local_err
);
3105 bdrv_child_free(new_child
);
3112 * Let every new BdrvChild start with a drained parent. Inserting the child
3113 * in the graph with bdrv_replace_child_noperm() will undrain it if
3114 * @child_bs is not drained.
3116 * The child was only just created and is not yet visible in global state
3117 * until bdrv_replace_child_noperm() inserts it into the graph, so nobody
3118 * could have sent requests and polling is not necessary.
3120 * Note that this means that the parent isn't fully drained yet, we only
3121 * stop new requests from coming in. This is fine, we don't care about the
3122 * old requests here, they are not for this child. If another place enters a
3123 * drain section for the same parent, but wants it to be fully quiesced, it
3124 * will not run most of the the code in .drained_begin() again (which is not
3125 * a problem, we already did this), but it will still poll until the parent
3126 * is fully quiesced, so it will not be negatively affected either.
3128 bdrv_parent_drained_begin_single(new_child
);
3129 bdrv_replace_child_noperm(new_child
, child_bs
);
3131 BdrvAttachChildCommonState
*s
= g_new(BdrvAttachChildCommonState
, 1);
3132 *s
= (BdrvAttachChildCommonState
) {
3134 .old_parent_ctx
= parent_ctx
,
3135 .old_child_ctx
= child_ctx
,
3137 tran_add(tran
, &bdrv_attach_child_common_drv
, s
);
3143 * Function doesn't update permissions, caller is responsible for this.
3145 * The caller must hold the AioContext lock for @child_bs. Both @parent_bs and
3146 * @child_bs can move to a different AioContext in this function. Callers must
3147 * make sure that their AioContext locking is still correct after this.
3149 * After calling this function, the transaction @tran may only be completed
3150 * while holding a writer lock for the graph.
3152 static BdrvChild
* GRAPH_WRLOCK
3153 bdrv_attach_child_noperm(BlockDriverState
*parent_bs
,
3154 BlockDriverState
*child_bs
,
3155 const char *child_name
,
3156 const BdrvChildClass
*child_class
,
3157 BdrvChildRole child_role
,
3161 uint64_t perm
, shared_perm
;
3163 assert(parent_bs
->drv
);
3164 GLOBAL_STATE_CODE();
3166 if (bdrv_recurse_has_child(child_bs
, parent_bs
)) {
3167 error_setg(errp
, "Making '%s' a %s child of '%s' would create a cycle",
3168 child_bs
->node_name
, child_name
, parent_bs
->node_name
);
3172 bdrv_get_cumulative_perm(parent_bs
, &perm
, &shared_perm
);
3173 bdrv_child_perm(parent_bs
, child_bs
, NULL
, child_role
, NULL
,
3174 perm
, shared_perm
, &perm
, &shared_perm
);
3176 return bdrv_attach_child_common(child_bs
, child_name
, child_class
,
3177 child_role
, perm
, shared_perm
, parent_bs
,
3182 * This function steals the reference to child_bs from the caller.
3183 * That reference is later dropped by bdrv_root_unref_child().
3185 * On failure NULL is returned, errp is set and the reference to
3186 * child_bs is also dropped.
3188 * The caller must hold the AioContext lock @child_bs, but not that of @ctx
3189 * (unless @child_bs is already in @ctx).
3191 BdrvChild
*bdrv_root_attach_child(BlockDriverState
*child_bs
,
3192 const char *child_name
,
3193 const BdrvChildClass
*child_class
,
3194 BdrvChildRole child_role
,
3195 uint64_t perm
, uint64_t shared_perm
,
3196 void *opaque
, Error
**errp
)
3200 Transaction
*tran
= tran_new();
3202 GLOBAL_STATE_CODE();
3204 child
= bdrv_attach_child_common(child_bs
, child_name
, child_class
,
3205 child_role
, perm
, shared_perm
, opaque
,
3212 ret
= bdrv_refresh_perms(child_bs
, tran
, errp
);
3215 tran_finalize(tran
, ret
);
3217 bdrv_schedule_unref(child_bs
);
3219 return ret
< 0 ? NULL
: child
;
3223 * This function transfers the reference to child_bs from the caller
3224 * to parent_bs. That reference is later dropped by parent_bs on
3225 * bdrv_close() or if someone calls bdrv_unref_child().
3227 * On failure NULL is returned, errp is set and the reference to
3228 * child_bs is also dropped.
3230 * If @parent_bs and @child_bs are in different AioContexts, the caller must
3231 * hold the AioContext lock for @child_bs, but not for @parent_bs.
3233 BdrvChild
*bdrv_attach_child(BlockDriverState
*parent_bs
,
3234 BlockDriverState
*child_bs
,
3235 const char *child_name
,
3236 const BdrvChildClass
*child_class
,
3237 BdrvChildRole child_role
,
3242 Transaction
*tran
= tran_new();
3244 GLOBAL_STATE_CODE();
3246 child
= bdrv_attach_child_noperm(parent_bs
, child_bs
, child_name
,
3247 child_class
, child_role
, tran
, errp
);
3253 ret
= bdrv_refresh_perms(parent_bs
, tran
, errp
);
3259 tran_finalize(tran
, ret
);
3261 bdrv_schedule_unref(child_bs
);
3263 return ret
< 0 ? NULL
: child
;
3266 /* Callers must ensure that child->frozen is false. */
3267 void bdrv_root_unref_child(BdrvChild
*child
)
3269 BlockDriverState
*child_bs
= child
->bs
;
3271 GLOBAL_STATE_CODE();
3272 bdrv_replace_child_noperm(child
, NULL
);
3273 bdrv_child_free(child
);
3277 * Update permissions for old node. We're just taking a parent away, so
3278 * we're loosening restrictions. Errors of permission update are not
3279 * fatal in this case, ignore them.
3281 bdrv_refresh_perms(child_bs
, NULL
, NULL
);
3284 * When the parent requiring a non-default AioContext is removed, the
3285 * node moves back to the main AioContext
3287 bdrv_try_change_aio_context(child_bs
, qemu_get_aio_context(), NULL
,
3291 bdrv_schedule_unref(child_bs
);
3294 typedef struct BdrvSetInheritsFrom
{
3295 BlockDriverState
*bs
;
3296 BlockDriverState
*old_inherits_from
;
3297 } BdrvSetInheritsFrom
;
3299 static void bdrv_set_inherits_from_abort(void *opaque
)
3301 BdrvSetInheritsFrom
*s
= opaque
;
3303 s
->bs
->inherits_from
= s
->old_inherits_from
;
3306 static TransactionActionDrv bdrv_set_inherits_from_drv
= {
3307 .abort
= bdrv_set_inherits_from_abort
,
3311 /* @tran is allowed to be NULL. In this case no rollback is possible */
3312 static void bdrv_set_inherits_from(BlockDriverState
*bs
,
3313 BlockDriverState
*new_inherits_from
,
3317 BdrvSetInheritsFrom
*s
= g_new(BdrvSetInheritsFrom
, 1);
3319 *s
= (BdrvSetInheritsFrom
) {
3321 .old_inherits_from
= bs
->inherits_from
,
3324 tran_add(tran
, &bdrv_set_inherits_from_drv
, s
);
3327 bs
->inherits_from
= new_inherits_from
;
3331 * Clear all inherits_from pointers from children and grandchildren of
3332 * @root that point to @root, where necessary.
3333 * @tran is allowed to be NULL. In this case no rollback is possible
3335 static void GRAPH_WRLOCK
3336 bdrv_unset_inherits_from(BlockDriverState
*root
, BdrvChild
*child
,
3341 if (child
->bs
->inherits_from
== root
) {
3343 * Remove inherits_from only when the last reference between root and
3344 * child->bs goes away.
3346 QLIST_FOREACH(c
, &root
->children
, next
) {
3347 if (c
!= child
&& c
->bs
== child
->bs
) {
3352 bdrv_set_inherits_from(child
->bs
, NULL
, tran
);
3356 QLIST_FOREACH(c
, &child
->bs
->children
, next
) {
3357 bdrv_unset_inherits_from(root
, c
, tran
);
3361 /* Callers must ensure that child->frozen is false. */
3362 void bdrv_unref_child(BlockDriverState
*parent
, BdrvChild
*child
)
3364 GLOBAL_STATE_CODE();
3365 if (child
== NULL
) {
3369 bdrv_unset_inherits_from(parent
, child
, NULL
);
3370 bdrv_root_unref_child(child
);
3374 static void GRAPH_RDLOCK
3375 bdrv_parent_cb_change_media(BlockDriverState
*bs
, bool load
)
3378 GLOBAL_STATE_CODE();
3379 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
3380 if (c
->klass
->change_media
) {
3381 c
->klass
->change_media(c
, load
);
3386 /* Return true if you can reach parent going through child->inherits_from
3387 * recursively. If parent or child are NULL, return false */
3388 static bool bdrv_inherits_from_recursive(BlockDriverState
*child
,
3389 BlockDriverState
*parent
)
3391 while (child
&& child
!= parent
) {
3392 child
= child
->inherits_from
;
3395 return child
!= NULL
;
3399 * Return the BdrvChildRole for @bs's backing child. bs->backing is
3400 * mostly used for COW backing children (role = COW), but also for
3401 * filtered children (role = FILTERED | PRIMARY).
3403 static BdrvChildRole
bdrv_backing_role(BlockDriverState
*bs
)
3405 if (bs
->drv
&& bs
->drv
->is_filter
) {
3406 return BDRV_CHILD_FILTERED
| BDRV_CHILD_PRIMARY
;
3408 return BDRV_CHILD_COW
;
3413 * Sets the bs->backing or bs->file link of a BDS. A new reference is created;
3414 * callers which don't need their own reference any more must call bdrv_unref().
3416 * If the respective child is already present (i.e. we're detaching a node),
3417 * that child node must be drained.
3419 * Function doesn't update permissions, caller is responsible for this.
3421 * The caller must hold the AioContext lock for @child_bs. Both @parent_bs and
3422 * @child_bs can move to a different AioContext in this function. Callers must
3423 * make sure that their AioContext locking is still correct after this.
3425 * After calling this function, the transaction @tran may only be completed
3426 * while holding a writer lock for the graph.
3428 static int GRAPH_WRLOCK
3429 bdrv_set_file_or_backing_noperm(BlockDriverState
*parent_bs
,
3430 BlockDriverState
*child_bs
,
3432 Transaction
*tran
, Error
**errp
)
3434 bool update_inherits_from
=
3435 bdrv_inherits_from_recursive(child_bs
, parent_bs
);
3436 BdrvChild
*child
= is_backing
? parent_bs
->backing
: parent_bs
->file
;
3439 GLOBAL_STATE_CODE();
3441 if (!parent_bs
->drv
) {
3443 * Node without drv is an object without a class :/. TODO: finally fix
3444 * qcow2 driver to never clear bs->drv and implement format corruption
3445 * handling in other way.
3447 error_setg(errp
, "Node corrupted");
3451 if (child
&& child
->frozen
) {
3452 error_setg(errp
, "Cannot change frozen '%s' link from '%s' to '%s'",
3453 child
->name
, parent_bs
->node_name
, child
->bs
->node_name
);
3457 if (is_backing
&& !parent_bs
->drv
->is_filter
&&
3458 !parent_bs
->drv
->supports_backing
)
3460 error_setg(errp
, "Driver '%s' of node '%s' does not support backing "
3461 "files", parent_bs
->drv
->format_name
, parent_bs
->node_name
);
3465 if (parent_bs
->drv
->is_filter
) {
3466 role
= BDRV_CHILD_FILTERED
| BDRV_CHILD_PRIMARY
;
3467 } else if (is_backing
) {
3468 role
= BDRV_CHILD_COW
;
3471 * We only can use same role as it is in existing child. We don't have
3472 * infrastructure to determine role of file child in generic way
3475 error_setg(errp
, "Cannot set file child to format node without "
3483 assert(child
->bs
->quiesce_counter
);
3484 bdrv_unset_inherits_from(parent_bs
, child
, tran
);
3485 bdrv_remove_child(child
, tran
);
3492 child
= bdrv_attach_child_noperm(parent_bs
, child_bs
,
3493 is_backing
? "backing" : "file",
3494 &child_of_bds
, role
,
3502 * If inherits_from pointed recursively to bs then let's update it to
3503 * point directly to bs (else it will become NULL).
3505 if (update_inherits_from
) {
3506 bdrv_set_inherits_from(child_bs
, parent_bs
, tran
);
3510 bdrv_refresh_limits(parent_bs
, tran
, NULL
);
3516 * The caller must hold the AioContext lock for @backing_hd. Both @bs and
3517 * @backing_hd can move to a different AioContext in this function. Callers must
3518 * make sure that their AioContext locking is still correct after this.
3520 * If a backing child is already present (i.e. we're detaching a node), that
3521 * child node must be drained.
3523 int bdrv_set_backing_hd_drained(BlockDriverState
*bs
,
3524 BlockDriverState
*backing_hd
,
3528 Transaction
*tran
= tran_new();
3530 GLOBAL_STATE_CODE();
3531 assert(bs
->quiesce_counter
> 0);
3533 assert(bs
->backing
->bs
->quiesce_counter
> 0);
3536 ret
= bdrv_set_file_or_backing_noperm(bs
, backing_hd
, true, tran
, errp
);
3541 ret
= bdrv_refresh_perms(bs
, tran
, errp
);
3543 tran_finalize(tran
, ret
);
3547 int bdrv_set_backing_hd(BlockDriverState
*bs
, BlockDriverState
*backing_hd
,
3550 BlockDriverState
*drain_bs
;
3552 GLOBAL_STATE_CODE();
3554 bdrv_graph_rdlock_main_loop();
3555 drain_bs
= bs
->backing
? bs
->backing
->bs
: bs
;
3556 bdrv_graph_rdunlock_main_loop();
3559 bdrv_drained_begin(drain_bs
);
3560 bdrv_graph_wrlock();
3561 ret
= bdrv_set_backing_hd_drained(bs
, backing_hd
, errp
);
3562 bdrv_graph_wrunlock();
3563 bdrv_drained_end(drain_bs
);
3564 bdrv_unref(drain_bs
);
3570 * Opens the backing file for a BlockDriverState if not yet open
3572 * bdref_key specifies the key for the image's BlockdevRef in the options QDict.
3573 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
3574 * itself, all options starting with "${bdref_key}." are considered part of the
3577 * The caller must hold the main AioContext lock.
3579 * TODO Can this be unified with bdrv_open_image()?
3581 int bdrv_open_backing_file(BlockDriverState
*bs
, QDict
*parent_options
,
3582 const char *bdref_key
, Error
**errp
)
3584 char *backing_filename
= NULL
;
3585 char *bdref_key_dot
;
3586 const char *reference
= NULL
;
3588 bool implicit_backing
= false;
3589 BlockDriverState
*backing_hd
;
3591 QDict
*tmp_parent_options
= NULL
;
3592 Error
*local_err
= NULL
;
3594 GLOBAL_STATE_CODE();
3595 GRAPH_RDLOCK_GUARD_MAINLOOP();
3597 if (bs
->backing
!= NULL
) {
3601 /* NULL means an empty set of options */
3602 if (parent_options
== NULL
) {
3603 tmp_parent_options
= qdict_new();
3604 parent_options
= tmp_parent_options
;
3607 bs
->open_flags
&= ~BDRV_O_NO_BACKING
;
3609 bdref_key_dot
= g_strdup_printf("%s.", bdref_key
);
3610 qdict_extract_subqdict(parent_options
, &options
, bdref_key_dot
);
3611 g_free(bdref_key_dot
);
3614 * Caution: while qdict_get_try_str() is fine, getting non-string
3615 * types would require more care. When @parent_options come from
3616 * -blockdev or blockdev_add, its members are typed according to
3617 * the QAPI schema, but when they come from -drive, they're all
3620 reference
= qdict_get_try_str(parent_options
, bdref_key
);
3621 if (reference
|| qdict_haskey(options
, "file.filename")) {
3622 /* keep backing_filename NULL */
3623 } else if (bs
->backing_file
[0] == '\0' && qdict_size(options
) == 0) {
3624 qobject_unref(options
);
3627 if (qdict_size(options
) == 0) {
3628 /* If the user specifies options that do not modify the
3629 * backing file's behavior, we might still consider it the
3630 * implicit backing file. But it's easier this way, and
3631 * just specifying some of the backing BDS's options is
3632 * only possible with -drive anyway (otherwise the QAPI
3633 * schema forces the user to specify everything). */
3634 implicit_backing
= !strcmp(bs
->auto_backing_file
, bs
->backing_file
);
3637 backing_filename
= bdrv_get_full_backing_filename(bs
, &local_err
);
3640 error_propagate(errp
, local_err
);
3641 qobject_unref(options
);
3646 if (!bs
->drv
|| !bs
->drv
->supports_backing
) {
3648 error_setg(errp
, "Driver doesn't support backing files");
3649 qobject_unref(options
);
3654 bs
->backing_format
[0] != '\0' && !qdict_haskey(options
, "driver")) {
3655 qdict_put_str(options
, "driver", bs
->backing_format
);
3658 backing_hd
= bdrv_open_inherit(backing_filename
, reference
, options
, 0, bs
,
3659 &child_of_bds
, bdrv_backing_role(bs
), errp
);
3661 bs
->open_flags
|= BDRV_O_NO_BACKING
;
3662 error_prepend(errp
, "Could not open backing file: ");
3667 if (implicit_backing
) {
3668 bdrv_refresh_filename(backing_hd
);
3669 pstrcpy(bs
->auto_backing_file
, sizeof(bs
->auto_backing_file
),
3670 backing_hd
->filename
);
3673 /* Hook up the backing file link; drop our reference, bs owns the
3674 * backing_hd reference now */
3675 ret
= bdrv_set_backing_hd(bs
, backing_hd
, errp
);
3676 bdrv_unref(backing_hd
);
3682 qdict_del(parent_options
, bdref_key
);
3685 g_free(backing_filename
);
3686 qobject_unref(tmp_parent_options
);
3690 static BlockDriverState
*
3691 bdrv_open_child_bs(const char *filename
, QDict
*options
, const char *bdref_key
,
3692 BlockDriverState
*parent
, const BdrvChildClass
*child_class
,
3693 BdrvChildRole child_role
, bool allow_none
, Error
**errp
)
3695 BlockDriverState
*bs
= NULL
;
3696 QDict
*image_options
;
3697 char *bdref_key_dot
;
3698 const char *reference
;
3700 assert(child_class
!= NULL
);
3702 bdref_key_dot
= g_strdup_printf("%s.", bdref_key
);
3703 qdict_extract_subqdict(options
, &image_options
, bdref_key_dot
);
3704 g_free(bdref_key_dot
);
3707 * Caution: while qdict_get_try_str() is fine, getting non-string
3708 * types would require more care. When @options come from
3709 * -blockdev or blockdev_add, its members are typed according to
3710 * the QAPI schema, but when they come from -drive, they're all
3713 reference
= qdict_get_try_str(options
, bdref_key
);
3714 if (!filename
&& !reference
&& !qdict_size(image_options
)) {
3716 error_setg(errp
, "A block device must be specified for \"%s\"",
3719 qobject_unref(image_options
);
3723 bs
= bdrv_open_inherit(filename
, reference
, image_options
, 0,
3724 parent
, child_class
, child_role
, errp
);
3730 qdict_del(options
, bdref_key
);
3735 * Opens a disk image whose options are given as BlockdevRef in another block
3738 * If allow_none is true, no image will be opened if filename is false and no
3739 * BlockdevRef is given. NULL will be returned, but errp remains unset.
3741 * bdrev_key specifies the key for the image's BlockdevRef in the options QDict.
3742 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
3743 * itself, all options starting with "${bdref_key}." are considered part of the
3746 * The BlockdevRef will be removed from the options QDict.
3748 * The caller must hold the lock of the main AioContext and no other AioContext.
3749 * @parent can move to a different AioContext in this function. Callers must
3750 * make sure that their AioContext locking is still correct after this.
3752 BdrvChild
*bdrv_open_child(const char *filename
,
3753 QDict
*options
, const char *bdref_key
,
3754 BlockDriverState
*parent
,
3755 const BdrvChildClass
*child_class
,
3756 BdrvChildRole child_role
,
3757 bool allow_none
, Error
**errp
)
3759 BlockDriverState
*bs
;
3762 GLOBAL_STATE_CODE();
3764 bs
= bdrv_open_child_bs(filename
, options
, bdref_key
, parent
, child_class
,
3765 child_role
, allow_none
, errp
);
3770 bdrv_graph_wrlock();
3771 child
= bdrv_attach_child(parent
, bs
, bdref_key
, child_class
, child_role
,
3773 bdrv_graph_wrunlock();
3779 * Wrapper on bdrv_open_child() for most popular case: open primary child of bs.
3781 * The caller must hold the lock of the main AioContext and no other AioContext.
3782 * @parent can move to a different AioContext in this function. Callers must
3783 * make sure that their AioContext locking is still correct after this.
3785 int bdrv_open_file_child(const char *filename
,
3786 QDict
*options
, const char *bdref_key
,
3787 BlockDriverState
*parent
, Error
**errp
)
3791 /* commit_top and mirror_top don't use this function */
3792 assert(!parent
->drv
->filtered_child_is_backing
);
3793 role
= parent
->drv
->is_filter
?
3794 (BDRV_CHILD_FILTERED
| BDRV_CHILD_PRIMARY
) : BDRV_CHILD_IMAGE
;
3796 if (!bdrv_open_child(filename
, options
, bdref_key
, parent
,
3797 &child_of_bds
, role
, false, errp
))
3806 * TODO Future callers may need to specify parent/child_class in order for
3807 * option inheritance to work. Existing callers use it for the root node.
3809 BlockDriverState
*bdrv_open_blockdev_ref(BlockdevRef
*ref
, Error
**errp
)
3811 BlockDriverState
*bs
= NULL
;
3812 QObject
*obj
= NULL
;
3813 QDict
*qdict
= NULL
;
3814 const char *reference
= NULL
;
3817 GLOBAL_STATE_CODE();
3819 if (ref
->type
== QTYPE_QSTRING
) {
3820 reference
= ref
->u
.reference
;
3822 BlockdevOptions
*options
= &ref
->u
.definition
;
3823 assert(ref
->type
== QTYPE_QDICT
);
3825 v
= qobject_output_visitor_new(&obj
);
3826 visit_type_BlockdevOptions(v
, NULL
, &options
, &error_abort
);
3827 visit_complete(v
, &obj
);
3829 qdict
= qobject_to(QDict
, obj
);
3830 qdict_flatten(qdict
);
3832 /* bdrv_open_inherit() defaults to the values in bdrv_flags (for
3833 * compatibility with other callers) rather than what we want as the
3834 * real defaults. Apply the defaults here instead. */
3835 qdict_set_default_str(qdict
, BDRV_OPT_CACHE_DIRECT
, "off");
3836 qdict_set_default_str(qdict
, BDRV_OPT_CACHE_NO_FLUSH
, "off");
3837 qdict_set_default_str(qdict
, BDRV_OPT_READ_ONLY
, "off");
3838 qdict_set_default_str(qdict
, BDRV_OPT_AUTO_READ_ONLY
, "off");
3842 bs
= bdrv_open_inherit(NULL
, reference
, qdict
, 0, NULL
, NULL
, 0, errp
);
3849 static BlockDriverState
*bdrv_append_temp_snapshot(BlockDriverState
*bs
,
3851 QDict
*snapshot_options
,
3854 g_autofree
char *tmp_filename
= NULL
;
3856 QemuOpts
*opts
= NULL
;
3857 BlockDriverState
*bs_snapshot
= NULL
;
3860 GLOBAL_STATE_CODE();
3862 /* if snapshot, we create a temporary backing file and open it
3863 instead of opening 'filename' directly */
3865 /* Get the required size from the image */
3866 total_size
= bdrv_getlength(bs
);
3868 if (total_size
< 0) {
3869 error_setg_errno(errp
, -total_size
, "Could not get image size");
3873 /* Create the temporary image */
3874 tmp_filename
= create_tmp_file(errp
);
3875 if (!tmp_filename
) {
3879 opts
= qemu_opts_create(bdrv_qcow2
.create_opts
, NULL
, 0,
3881 qemu_opt_set_number(opts
, BLOCK_OPT_SIZE
, total_size
, &error_abort
);
3882 ret
= bdrv_create(&bdrv_qcow2
, tmp_filename
, opts
, errp
);
3883 qemu_opts_del(opts
);
3885 error_prepend(errp
, "Could not create temporary overlay '%s': ",
3890 /* Prepare options QDict for the temporary file */
3891 qdict_put_str(snapshot_options
, "file.driver", "file");
3892 qdict_put_str(snapshot_options
, "file.filename", tmp_filename
);
3893 qdict_put_str(snapshot_options
, "driver", "qcow2");
3895 bs_snapshot
= bdrv_open(NULL
, NULL
, snapshot_options
, flags
, errp
);
3896 snapshot_options
= NULL
;
3901 ret
= bdrv_append(bs_snapshot
, bs
, errp
);
3908 qobject_unref(snapshot_options
);
3913 * Opens a disk image (raw, qcow2, vmdk, ...)
3915 * options is a QDict of options to pass to the block drivers, or NULL for an
3916 * empty set of options. The reference to the QDict belongs to the block layer
3917 * after the call (even on failure), so if the caller intends to reuse the
3918 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
3920 * If *pbs is NULL, a new BDS will be created with a pointer to it stored there.
3921 * If it is not NULL, the referenced BDS will be reused.
3923 * The reference parameter may be used to specify an existing block device which
3924 * should be opened. If specified, neither options nor a filename may be given,
3925 * nor can an existing BDS be reused (that is, *pbs has to be NULL).
3927 * The caller must always hold the main AioContext lock.
3929 static BlockDriverState
* no_coroutine_fn
3930 bdrv_open_inherit(const char *filename
, const char *reference
, QDict
*options
,
3931 int flags
, BlockDriverState
*parent
,
3932 const BdrvChildClass
*child_class
, BdrvChildRole child_role
,
3936 BlockBackend
*file
= NULL
;
3937 BlockDriverState
*bs
;
3938 BlockDriver
*drv
= NULL
;
3940 const char *drvname
;
3941 const char *backing
;
3942 Error
*local_err
= NULL
;
3943 QDict
*snapshot_options
= NULL
;
3944 int snapshot_flags
= 0;
3946 assert(!child_class
|| !flags
);
3947 assert(!child_class
== !parent
);
3948 GLOBAL_STATE_CODE();
3949 assert(!qemu_in_coroutine());
3951 /* TODO We'll eventually have to take a writer lock in this function */
3952 GRAPH_RDLOCK_GUARD_MAINLOOP();
3955 bool options_non_empty
= options
? qdict_size(options
) : false;
3956 qobject_unref(options
);
3958 if (filename
|| options_non_empty
) {
3959 error_setg(errp
, "Cannot reference an existing block device with "
3960 "additional options or a new filename");
3964 bs
= bdrv_lookup_bs(reference
, reference
, errp
);
3975 /* NULL means an empty set of options */
3976 if (options
== NULL
) {
3977 options
= qdict_new();
3980 /* json: syntax counts as explicit options, as if in the QDict */
3981 parse_json_protocol(options
, &filename
, &local_err
);
3986 bs
->explicit_options
= qdict_clone_shallow(options
);
3989 bool parent_is_format
;
3992 parent_is_format
= parent
->drv
->is_format
;
3995 * parent->drv is not set yet because this node is opened for
3996 * (potential) format probing. That means that @parent is going
3997 * to be a format node.
3999 parent_is_format
= true;
4002 bs
->inherits_from
= parent
;
4003 child_class
->inherit_options(child_role
, parent_is_format
,
4005 parent
->open_flags
, parent
->options
);
4008 ret
= bdrv_fill_options(&options
, filename
, &flags
, &local_err
);
4014 * Set the BDRV_O_RDWR and BDRV_O_ALLOW_RDWR flags.
4015 * Caution: getting a boolean member of @options requires care.
4016 * When @options come from -blockdev or blockdev_add, members are
4017 * typed according to the QAPI schema, but when they come from
4018 * -drive, they're all QString.
4020 if (g_strcmp0(qdict_get_try_str(options
, BDRV_OPT_READ_ONLY
), "on") &&
4021 !qdict_get_try_bool(options
, BDRV_OPT_READ_ONLY
, false)) {
4022 flags
|= (BDRV_O_RDWR
| BDRV_O_ALLOW_RDWR
);
4024 flags
&= ~BDRV_O_RDWR
;
4027 if (flags
& BDRV_O_SNAPSHOT
) {
4028 snapshot_options
= qdict_new();
4029 bdrv_temp_snapshot_options(&snapshot_flags
, snapshot_options
,
4031 /* Let bdrv_backing_options() override "read-only" */
4032 qdict_del(options
, BDRV_OPT_READ_ONLY
);
4033 bdrv_inherited_options(BDRV_CHILD_COW
, true,
4034 &flags
, options
, flags
, options
);
4037 bs
->open_flags
= flags
;
4038 bs
->options
= options
;
4039 options
= qdict_clone_shallow(options
);
4041 /* Find the right image format driver */
4042 /* See cautionary note on accessing @options above */
4043 drvname
= qdict_get_try_str(options
, "driver");
4045 drv
= bdrv_find_format(drvname
);
4047 error_setg(errp
, "Unknown driver: '%s'", drvname
);
4052 assert(drvname
|| !(flags
& BDRV_O_PROTOCOL
));
4054 /* See cautionary note on accessing @options above */
4055 backing
= qdict_get_try_str(options
, "backing");
4056 if (qobject_to(QNull
, qdict_get(options
, "backing")) != NULL
||
4057 (backing
&& *backing
== '\0'))
4060 warn_report("Use of \"backing\": \"\" is deprecated; "
4061 "use \"backing\": null instead");
4063 flags
|= BDRV_O_NO_BACKING
;
4064 qdict_del(bs
->explicit_options
, "backing");
4065 qdict_del(bs
->options
, "backing");
4066 qdict_del(options
, "backing");
4069 /* Open image file without format layer. This BlockBackend is only used for
4070 * probing, the block drivers will do their own bdrv_open_child() for the
4071 * same BDS, which is why we put the node name back into options. */
4072 if ((flags
& BDRV_O_PROTOCOL
) == 0) {
4073 BlockDriverState
*file_bs
;
4075 file_bs
= bdrv_open_child_bs(filename
, options
, "file", bs
,
4076 &child_of_bds
, BDRV_CHILD_IMAGE
,
4081 if (file_bs
!= NULL
) {
4082 /* Not requesting BLK_PERM_CONSISTENT_READ because we're only
4083 * looking at the header to guess the image format. This works even
4084 * in cases where a guest would not see a consistent state. */
4085 AioContext
*ctx
= bdrv_get_aio_context(file_bs
);
4086 file
= blk_new(ctx
, 0, BLK_PERM_ALL
);
4087 blk_insert_bs(file
, file_bs
, &local_err
);
4088 bdrv_unref(file_bs
);
4094 qdict_put_str(options
, "file", bdrv_get_node_name(file_bs
));
4098 /* Image format probing */
4101 ret
= find_image_format(file
, filename
, &drv
, &local_err
);
4106 * This option update would logically belong in bdrv_fill_options(),
4107 * but we first need to open bs->file for the probing to work, while
4108 * opening bs->file already requires the (mostly) final set of options
4109 * so that cache mode etc. can be inherited.
4111 * Adding the driver later is somewhat ugly, but it's not an option
4112 * that would ever be inherited, so it's correct. We just need to make
4113 * sure to update both bs->options (which has the full effective
4114 * options for bs) and options (which has file.* already removed).
4116 qdict_put_str(bs
->options
, "driver", drv
->format_name
);
4117 qdict_put_str(options
, "driver", drv
->format_name
);
4119 error_setg(errp
, "Must specify either driver or file");
4123 /* BDRV_O_PROTOCOL must be set iff a protocol BDS is about to be created */
4124 assert(!!(flags
& BDRV_O_PROTOCOL
) == !!drv
->bdrv_file_open
);
4125 /* file must be NULL if a protocol BDS is about to be created
4126 * (the inverse results in an error message from bdrv_open_common()) */
4127 assert(!(flags
& BDRV_O_PROTOCOL
) || !file
);
4129 /* Open the image */
4130 ret
= bdrv_open_common(bs
, file
, options
, &local_err
);
4140 /* If there is a backing file, use it */
4141 if ((flags
& BDRV_O_NO_BACKING
) == 0) {
4142 ret
= bdrv_open_backing_file(bs
, options
, "backing", &local_err
);
4144 goto close_and_fail
;
4148 /* Remove all children options and references
4149 * from bs->options and bs->explicit_options */
4150 QLIST_FOREACH(child
, &bs
->children
, next
) {
4151 char *child_key_dot
;
4152 child_key_dot
= g_strdup_printf("%s.", child
->name
);
4153 qdict_extract_subqdict(bs
->explicit_options
, NULL
, child_key_dot
);
4154 qdict_extract_subqdict(bs
->options
, NULL
, child_key_dot
);
4155 qdict_del(bs
->explicit_options
, child
->name
);
4156 qdict_del(bs
->options
, child
->name
);
4157 g_free(child_key_dot
);
4160 /* Check if any unknown options were used */
4161 if (qdict_size(options
) != 0) {
4162 const QDictEntry
*entry
= qdict_first(options
);
4163 if (flags
& BDRV_O_PROTOCOL
) {
4164 error_setg(errp
, "Block protocol '%s' doesn't support the option "
4165 "'%s'", drv
->format_name
, entry
->key
);
4168 "Block format '%s' does not support the option '%s'",
4169 drv
->format_name
, entry
->key
);
4172 goto close_and_fail
;
4175 bdrv_parent_cb_change_media(bs
, true);
4177 qobject_unref(options
);
4180 /* For snapshot=on, create a temporary qcow2 overlay. bs points to the
4181 * temporary snapshot afterwards. */
4182 if (snapshot_flags
) {
4183 BlockDriverState
*snapshot_bs
;
4184 snapshot_bs
= bdrv_append_temp_snapshot(bs
, snapshot_flags
,
4185 snapshot_options
, &local_err
);
4186 snapshot_options
= NULL
;
4188 goto close_and_fail
;
4190 /* We are not going to return bs but the overlay on top of it
4191 * (snapshot_bs); thus, we have to drop the strong reference to bs
4192 * (which we obtained by calling bdrv_new()). bs will not be deleted,
4193 * though, because the overlay still has a reference to it. */
4202 qobject_unref(snapshot_options
);
4203 qobject_unref(bs
->explicit_options
);
4204 qobject_unref(bs
->options
);
4205 qobject_unref(options
);
4207 bs
->explicit_options
= NULL
;
4209 error_propagate(errp
, local_err
);
4214 qobject_unref(snapshot_options
);
4215 qobject_unref(options
);
4216 error_propagate(errp
, local_err
);
4220 /* The caller must always hold the main AioContext lock. */
4221 BlockDriverState
*bdrv_open(const char *filename
, const char *reference
,
4222 QDict
*options
, int flags
, Error
**errp
)
4224 GLOBAL_STATE_CODE();
4226 return bdrv_open_inherit(filename
, reference
, options
, flags
, NULL
,
4230 /* Return true if the NULL-terminated @list contains @str */
4231 static bool is_str_in_list(const char *str
, const char *const *list
)
4235 for (i
= 0; list
[i
] != NULL
; i
++) {
4236 if (!strcmp(str
, list
[i
])) {
4245 * Check that every option set in @bs->options is also set in
4248 * Options listed in the common_options list and in
4249 * @bs->drv->mutable_opts are skipped.
4251 * Return 0 on success, otherwise return -EINVAL and set @errp.
4253 static int bdrv_reset_options_allowed(BlockDriverState
*bs
,
4254 const QDict
*new_opts
, Error
**errp
)
4256 const QDictEntry
*e
;
4257 /* These options are common to all block drivers and are handled
4258 * in bdrv_reopen_prepare() so they can be left out of @new_opts */
4259 const char *const common_options
[] = {
4260 "node-name", "discard", "cache.direct", "cache.no-flush",
4261 "read-only", "auto-read-only", "detect-zeroes", NULL
4264 for (e
= qdict_first(bs
->options
); e
; e
= qdict_next(bs
->options
, e
)) {
4265 if (!qdict_haskey(new_opts
, e
->key
) &&
4266 !is_str_in_list(e
->key
, common_options
) &&
4267 !is_str_in_list(e
->key
, bs
->drv
->mutable_opts
)) {
4268 error_setg(errp
, "Option '%s' cannot be reset "
4269 "to its default value", e
->key
);
4278 * Returns true if @child can be reached recursively from @bs
4280 static bool GRAPH_RDLOCK
4281 bdrv_recurse_has_child(BlockDriverState
*bs
, BlockDriverState
*child
)
4289 QLIST_FOREACH(c
, &bs
->children
, next
) {
4290 if (bdrv_recurse_has_child(c
->bs
, child
)) {
4299 * Adds a BlockDriverState to a simple queue for an atomic, transactional
4300 * reopen of multiple devices.
4302 * bs_queue can either be an existing BlockReopenQueue that has had QTAILQ_INIT
4303 * already performed, or alternatively may be NULL a new BlockReopenQueue will
4304 * be created and initialized. This newly created BlockReopenQueue should be
4305 * passed back in for subsequent calls that are intended to be of the same
4308 * bs is the BlockDriverState to add to the reopen queue.
4310 * options contains the changed options for the associated bs
4311 * (the BlockReopenQueue takes ownership)
4313 * flags contains the open flags for the associated bs
4315 * returns a pointer to bs_queue, which is either the newly allocated
4316 * bs_queue, or the existing bs_queue being used.
4318 * bs is drained here and undrained by bdrv_reopen_queue_free().
4320 * To be called with bs->aio_context locked.
4322 static BlockReopenQueue
* GRAPH_RDLOCK
4323 bdrv_reopen_queue_child(BlockReopenQueue
*bs_queue
, BlockDriverState
*bs
,
4324 QDict
*options
, const BdrvChildClass
*klass
,
4325 BdrvChildRole role
, bool parent_is_format
,
4326 QDict
*parent_options
, int parent_flags
,
4331 BlockReopenQueueEntry
*bs_entry
;
4333 QDict
*old_options
, *explicit_options
, *options_copy
;
4337 GLOBAL_STATE_CODE();
4340 * Strictly speaking, draining is illegal under GRAPH_RDLOCK. We know that
4341 * we've been called with bdrv_graph_rdlock_main_loop(), though, so it's ok
4344 bdrv_drained_begin(bs
);
4346 if (bs_queue
== NULL
) {
4347 bs_queue
= g_new0(BlockReopenQueue
, 1);
4348 QTAILQ_INIT(bs_queue
);
4352 options
= qdict_new();
4355 /* Check if this BlockDriverState is already in the queue */
4356 QTAILQ_FOREACH(bs_entry
, bs_queue
, entry
) {
4357 if (bs
== bs_entry
->state
.bs
) {
4363 * Precedence of options:
4364 * 1. Explicitly passed in options (highest)
4365 * 2. Retained from explicitly set options of bs
4366 * 3. Inherited from parent node
4367 * 4. Retained from effective options of bs
4370 /* Old explicitly set values (don't overwrite by inherited value) */
4371 if (bs_entry
|| keep_old_opts
) {
4372 old_options
= qdict_clone_shallow(bs_entry
?
4373 bs_entry
->state
.explicit_options
:
4374 bs
->explicit_options
);
4375 bdrv_join_options(bs
, options
, old_options
);
4376 qobject_unref(old_options
);
4379 explicit_options
= qdict_clone_shallow(options
);
4381 /* Inherit from parent node */
4382 if (parent_options
) {
4384 klass
->inherit_options(role
, parent_is_format
, &flags
, options
,
4385 parent_flags
, parent_options
);
4387 flags
= bdrv_get_flags(bs
);
4390 if (keep_old_opts
) {
4391 /* Old values are used for options that aren't set yet */
4392 old_options
= qdict_clone_shallow(bs
->options
);
4393 bdrv_join_options(bs
, options
, old_options
);
4394 qobject_unref(old_options
);
4397 /* We have the final set of options so let's update the flags */
4398 options_copy
= qdict_clone_shallow(options
);
4399 opts
= qemu_opts_create(&bdrv_runtime_opts
, NULL
, 0, &error_abort
);
4400 qemu_opts_absorb_qdict(opts
, options_copy
, NULL
);
4401 update_flags_from_options(&flags
, opts
);
4402 qemu_opts_del(opts
);
4403 qobject_unref(options_copy
);
4405 /* bdrv_open_inherit() sets and clears some additional flags internally */
4406 flags
&= ~BDRV_O_PROTOCOL
;
4407 if (flags
& BDRV_O_RDWR
) {
4408 flags
|= BDRV_O_ALLOW_RDWR
;
4412 bs_entry
= g_new0(BlockReopenQueueEntry
, 1);
4413 QTAILQ_INSERT_TAIL(bs_queue
, bs_entry
, entry
);
4415 qobject_unref(bs_entry
->state
.options
);
4416 qobject_unref(bs_entry
->state
.explicit_options
);
4419 bs_entry
->state
.bs
= bs
;
4420 bs_entry
->state
.options
= options
;
4421 bs_entry
->state
.explicit_options
= explicit_options
;
4422 bs_entry
->state
.flags
= flags
;
4425 * If keep_old_opts is false then it means that unspecified
4426 * options must be reset to their original value. We don't allow
4427 * resetting 'backing' but we need to know if the option is
4428 * missing in order to decide if we have to return an error.
4430 if (!keep_old_opts
) {
4431 bs_entry
->state
.backing_missing
=
4432 !qdict_haskey(options
, "backing") &&
4433 !qdict_haskey(options
, "backing.driver");
4436 QLIST_FOREACH(child
, &bs
->children
, next
) {
4437 QDict
*new_child_options
= NULL
;
4438 bool child_keep_old
= keep_old_opts
;
4440 /* reopen can only change the options of block devices that were
4441 * implicitly created and inherited options. For other (referenced)
4442 * block devices, a syntax like "backing.foo" results in an error. */
4443 if (child
->bs
->inherits_from
!= bs
) {
4447 /* Check if the options contain a child reference */
4448 if (qdict_haskey(options
, child
->name
)) {
4449 const char *childref
= qdict_get_try_str(options
, child
->name
);
4451 * The current child must not be reopened if the child
4452 * reference is null or points to a different node.
4454 if (g_strcmp0(childref
, child
->bs
->node_name
)) {
4458 * If the child reference points to the current child then
4459 * reopen it with its existing set of options (note that
4460 * it can still inherit new options from the parent).
4462 child_keep_old
= true;
4464 /* Extract child options ("child-name.*") */
4465 char *child_key_dot
= g_strdup_printf("%s.", child
->name
);
4466 qdict_extract_subqdict(explicit_options
, NULL
, child_key_dot
);
4467 qdict_extract_subqdict(options
, &new_child_options
, child_key_dot
);
4468 g_free(child_key_dot
);
4471 bdrv_reopen_queue_child(bs_queue
, child
->bs
, new_child_options
,
4472 child
->klass
, child
->role
, bs
->drv
->is_format
,
4473 options
, flags
, child_keep_old
);
4479 /* To be called with bs->aio_context locked */
4480 BlockReopenQueue
*bdrv_reopen_queue(BlockReopenQueue
*bs_queue
,
4481 BlockDriverState
*bs
,
4482 QDict
*options
, bool keep_old_opts
)
4484 GLOBAL_STATE_CODE();
4485 GRAPH_RDLOCK_GUARD_MAINLOOP();
4487 return bdrv_reopen_queue_child(bs_queue
, bs
, options
, NULL
, 0, false,
4488 NULL
, 0, keep_old_opts
);
4491 void bdrv_reopen_queue_free(BlockReopenQueue
*bs_queue
)
4493 GLOBAL_STATE_CODE();
4495 BlockReopenQueueEntry
*bs_entry
, *next
;
4496 QTAILQ_FOREACH_SAFE(bs_entry
, bs_queue
, entry
, next
) {
4497 bdrv_drained_end(bs_entry
->state
.bs
);
4498 qobject_unref(bs_entry
->state
.explicit_options
);
4499 qobject_unref(bs_entry
->state
.options
);
4507 * Reopen multiple BlockDriverStates atomically & transactionally.
4509 * The queue passed in (bs_queue) must have been built up previous
4510 * via bdrv_reopen_queue().
4512 * Reopens all BDS specified in the queue, with the appropriate
4513 * flags. All devices are prepared for reopen, and failure of any
4514 * device will cause all device changes to be abandoned, and intermediate
4517 * If all devices prepare successfully, then the changes are committed
4520 * All affected nodes must be drained between bdrv_reopen_queue() and
4521 * bdrv_reopen_multiple().
4523 * To be called from the main thread, with all other AioContexts unlocked.
4525 int bdrv_reopen_multiple(BlockReopenQueue
*bs_queue
, Error
**errp
)
4528 BlockReopenQueueEntry
*bs_entry
, *next
;
4529 Transaction
*tran
= tran_new();
4530 g_autoptr(GSList
) refresh_list
= NULL
;
4532 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
4533 assert(bs_queue
!= NULL
);
4534 GLOBAL_STATE_CODE();
4536 QTAILQ_FOREACH(bs_entry
, bs_queue
, entry
) {
4537 ret
= bdrv_flush(bs_entry
->state
.bs
);
4539 error_setg_errno(errp
, -ret
, "Error flushing drive");
4544 QTAILQ_FOREACH(bs_entry
, bs_queue
, entry
) {
4545 assert(bs_entry
->state
.bs
->quiesce_counter
> 0);
4546 ret
= bdrv_reopen_prepare(&bs_entry
->state
, bs_queue
, tran
, errp
);
4550 bs_entry
->prepared
= true;
4553 QTAILQ_FOREACH(bs_entry
, bs_queue
, entry
) {
4554 BDRVReopenState
*state
= &bs_entry
->state
;
4556 refresh_list
= g_slist_prepend(refresh_list
, state
->bs
);
4557 if (state
->old_backing_bs
) {
4558 refresh_list
= g_slist_prepend(refresh_list
, state
->old_backing_bs
);
4560 if (state
->old_file_bs
) {
4561 refresh_list
= g_slist_prepend(refresh_list
, state
->old_file_bs
);
4566 * Note that file-posix driver rely on permission update done during reopen
4567 * (even if no permission changed), because it wants "new" permissions for
4568 * reconfiguring the fd and that's why it does it in raw_check_perm(), not
4569 * in raw_reopen_prepare() which is called with "old" permissions.
4571 bdrv_graph_rdlock_main_loop();
4572 ret
= bdrv_list_refresh_perms(refresh_list
, bs_queue
, tran
, errp
);
4573 bdrv_graph_rdunlock_main_loop();
4580 * If we reach this point, we have success and just need to apply the
4583 * Reverse order is used to comfort qcow2 driver: on commit it need to write
4584 * IN_USE flag to the image, to mark bitmaps in the image as invalid. But
4585 * children are usually goes after parents in reopen-queue, so go from last
4588 QTAILQ_FOREACH_REVERSE(bs_entry
, bs_queue
, entry
) {
4589 bdrv_reopen_commit(&bs_entry
->state
);
4592 bdrv_graph_wrlock();
4594 bdrv_graph_wrunlock();
4596 QTAILQ_FOREACH_REVERSE(bs_entry
, bs_queue
, entry
) {
4597 BlockDriverState
*bs
= bs_entry
->state
.bs
;
4599 if (bs
->drv
->bdrv_reopen_commit_post
) {
4600 bs
->drv
->bdrv_reopen_commit_post(&bs_entry
->state
);
4608 bdrv_graph_wrlock();
4610 bdrv_graph_wrunlock();
4612 QTAILQ_FOREACH_SAFE(bs_entry
, bs_queue
, entry
, next
) {
4613 if (bs_entry
->prepared
) {
4614 bdrv_reopen_abort(&bs_entry
->state
);
4619 bdrv_reopen_queue_free(bs_queue
);
4624 int bdrv_reopen(BlockDriverState
*bs
, QDict
*opts
, bool keep_old_opts
,
4627 BlockReopenQueue
*queue
;
4629 GLOBAL_STATE_CODE();
4631 queue
= bdrv_reopen_queue(NULL
, bs
, opts
, keep_old_opts
);
4633 return bdrv_reopen_multiple(queue
, errp
);
4636 int bdrv_reopen_set_read_only(BlockDriverState
*bs
, bool read_only
,
4639 QDict
*opts
= qdict_new();
4641 GLOBAL_STATE_CODE();
4643 qdict_put_bool(opts
, BDRV_OPT_READ_ONLY
, read_only
);
4645 return bdrv_reopen(bs
, opts
, true, errp
);
4649 * Take a BDRVReopenState and check if the value of 'backing' in the
4650 * reopen_state->options QDict is valid or not.
4652 * If 'backing' is missing from the QDict then return 0.
4654 * If 'backing' contains the node name of the backing file of
4655 * reopen_state->bs then return 0.
4657 * If 'backing' contains a different node name (or is null) then check
4658 * whether the current backing file can be replaced with the new one.
4659 * If that's the case then reopen_state->replace_backing_bs is set to
4660 * true and reopen_state->new_backing_bs contains a pointer to the new
4661 * backing BlockDriverState (or NULL).
4663 * After calling this function, the transaction @tran may only be completed
4664 * while holding a writer lock for the graph.
4666 * Return 0 on success, otherwise return < 0 and set @errp.
4668 * The caller must hold the AioContext lock of @reopen_state->bs.
4669 * @reopen_state->bs can move to a different AioContext in this function.
4670 * Callers must make sure that their AioContext locking is still correct after
4673 static int GRAPH_UNLOCKED
4674 bdrv_reopen_parse_file_or_backing(BDRVReopenState
*reopen_state
,
4675 bool is_backing
, Transaction
*tran
,
4678 BlockDriverState
*bs
= reopen_state
->bs
;
4679 BlockDriverState
*new_child_bs
;
4680 BlockDriverState
*old_child_bs
;
4682 const char *child_name
= is_backing
? "backing" : "file";
4688 GLOBAL_STATE_CODE();
4690 value
= qdict_get(reopen_state
->options
, child_name
);
4691 if (value
== NULL
) {
4695 bdrv_graph_rdlock_main_loop();
4697 switch (qobject_type(value
)) {
4699 assert(is_backing
); /* The 'file' option does not allow a null value */
4700 new_child_bs
= NULL
;
4703 str
= qstring_get_str(qobject_to(QString
, value
));
4704 new_child_bs
= bdrv_lookup_bs(NULL
, str
, errp
);
4705 if (new_child_bs
== NULL
) {
4710 has_child
= bdrv_recurse_has_child(new_child_bs
, bs
);
4712 error_setg(errp
, "Making '%s' a %s child of '%s' would create a "
4713 "cycle", str
, child_name
, bs
->node_name
);
4720 * The options QDict has been flattened, so 'backing' and 'file'
4721 * do not allow any other data type here.
4723 g_assert_not_reached();
4726 old_child_bs
= is_backing
? child_bs(bs
->backing
) : child_bs(bs
->file
);
4727 if (old_child_bs
== new_child_bs
) {
4733 if (bdrv_skip_implicit_filters(old_child_bs
) == new_child_bs
) {
4738 if (old_child_bs
->implicit
) {
4739 error_setg(errp
, "Cannot replace implicit %s child of %s",
4740 child_name
, bs
->node_name
);
4746 if (bs
->drv
->is_filter
&& !old_child_bs
) {
4748 * Filters always have a file or a backing child, so we are trying to
4749 * change wrong child
4751 error_setg(errp
, "'%s' is a %s filter node that does not support a "
4752 "%s child", bs
->node_name
, bs
->drv
->format_name
, child_name
);
4758 reopen_state
->old_backing_bs
= old_child_bs
;
4760 reopen_state
->old_file_bs
= old_child_bs
;
4764 bdrv_ref(old_child_bs
);
4765 bdrv_drained_begin(old_child_bs
);
4768 bdrv_graph_rdunlock_main_loop();
4769 bdrv_graph_wrlock();
4771 ret
= bdrv_set_file_or_backing_noperm(bs
, new_child_bs
, is_backing
,
4774 bdrv_graph_wrunlock();
4777 bdrv_drained_end(old_child_bs
);
4778 bdrv_unref(old_child_bs
);
4784 bdrv_graph_rdunlock_main_loop();
4789 * Prepares a BlockDriverState for reopen. All changes are staged in the
4790 * 'opaque' field of the BDRVReopenState, which is used and allocated by
4791 * the block driver layer .bdrv_reopen_prepare()
4793 * bs is the BlockDriverState to reopen
4794 * flags are the new open flags
4795 * queue is the reopen queue
4797 * Returns 0 on success, non-zero on error. On error errp will be set
4800 * On failure, bdrv_reopen_abort() will be called to clean up any data.
4801 * It is the responsibility of the caller to then call the abort() or
4802 * commit() for any other BDS that have been left in a prepare() state
4804 * The caller must hold the AioContext lock of @reopen_state->bs.
4806 * After calling this function, the transaction @change_child_tran may only be
4807 * completed while holding a writer lock for the graph.
4809 static int GRAPH_UNLOCKED
4810 bdrv_reopen_prepare(BDRVReopenState
*reopen_state
, BlockReopenQueue
*queue
,
4811 Transaction
*change_child_tran
, Error
**errp
)
4815 Error
*local_err
= NULL
;
4818 QDict
*orig_reopen_opts
;
4819 char *discard
= NULL
;
4821 bool drv_prepared
= false;
4823 assert(reopen_state
!= NULL
);
4824 assert(reopen_state
->bs
->drv
!= NULL
);
4825 GLOBAL_STATE_CODE();
4826 drv
= reopen_state
->bs
->drv
;
4828 /* This function and each driver's bdrv_reopen_prepare() remove
4829 * entries from reopen_state->options as they are processed, so
4830 * we need to make a copy of the original QDict. */
4831 orig_reopen_opts
= qdict_clone_shallow(reopen_state
->options
);
4833 /* Process generic block layer options */
4834 opts
= qemu_opts_create(&bdrv_runtime_opts
, NULL
, 0, &error_abort
);
4835 if (!qemu_opts_absorb_qdict(opts
, reopen_state
->options
, errp
)) {
4840 /* This was already called in bdrv_reopen_queue_child() so the flags
4841 * are up-to-date. This time we simply want to remove the options from
4842 * QemuOpts in order to indicate that they have been processed. */
4843 old_flags
= reopen_state
->flags
;
4844 update_flags_from_options(&reopen_state
->flags
, opts
);
4845 assert(old_flags
== reopen_state
->flags
);
4847 discard
= qemu_opt_get_del(opts
, BDRV_OPT_DISCARD
);
4848 if (discard
!= NULL
) {
4849 if (bdrv_parse_discard_flags(discard
, &reopen_state
->flags
) != 0) {
4850 error_setg(errp
, "Invalid discard option");
4856 reopen_state
->detect_zeroes
=
4857 bdrv_parse_detect_zeroes(opts
, reopen_state
->flags
, &local_err
);
4859 error_propagate(errp
, local_err
);
4864 /* All other options (including node-name and driver) must be unchanged.
4865 * Put them back into the QDict, so that they are checked at the end
4866 * of this function. */
4867 qemu_opts_to_qdict(opts
, reopen_state
->options
);
4869 /* If we are to stay read-only, do not allow permission change
4870 * to r/w. Attempting to set to r/w may fail if either BDRV_O_ALLOW_RDWR is
4871 * not set, or if the BDS still has copy_on_read enabled */
4872 read_only
= !(reopen_state
->flags
& BDRV_O_RDWR
);
4874 bdrv_graph_rdlock_main_loop();
4875 ret
= bdrv_can_set_read_only(reopen_state
->bs
, read_only
, true, &local_err
);
4876 bdrv_graph_rdunlock_main_loop();
4878 error_propagate(errp
, local_err
);
4882 if (drv
->bdrv_reopen_prepare
) {
4884 * If a driver-specific option is missing, it means that we
4885 * should reset it to its default value.
4886 * But not all options allow that, so we need to check it first.
4888 ret
= bdrv_reset_options_allowed(reopen_state
->bs
,
4889 reopen_state
->options
, errp
);
4894 ret
= drv
->bdrv_reopen_prepare(reopen_state
, queue
, &local_err
);
4896 if (local_err
!= NULL
) {
4897 error_propagate(errp
, local_err
);
4899 bdrv_graph_rdlock_main_loop();
4900 bdrv_refresh_filename(reopen_state
->bs
);
4901 bdrv_graph_rdunlock_main_loop();
4902 error_setg(errp
, "failed while preparing to reopen image '%s'",
4903 reopen_state
->bs
->filename
);
4908 /* It is currently mandatory to have a bdrv_reopen_prepare()
4909 * handler for each supported drv. */
4910 bdrv_graph_rdlock_main_loop();
4911 error_setg(errp
, "Block format '%s' used by node '%s' "
4912 "does not support reopening files", drv
->format_name
,
4913 bdrv_get_device_or_node_name(reopen_state
->bs
));
4914 bdrv_graph_rdunlock_main_loop();
4919 drv_prepared
= true;
4922 * We must provide the 'backing' option if the BDS has a backing
4923 * file or if the image file has a backing file name as part of
4924 * its metadata. Otherwise the 'backing' option can be omitted.
4926 bdrv_graph_rdlock_main_loop();
4927 if (drv
->supports_backing
&& reopen_state
->backing_missing
&&
4928 (reopen_state
->bs
->backing
|| reopen_state
->bs
->backing_file
[0])) {
4929 error_setg(errp
, "backing is missing for '%s'",
4930 reopen_state
->bs
->node_name
);
4931 bdrv_graph_rdunlock_main_loop();
4935 bdrv_graph_rdunlock_main_loop();
4938 * Allow changing the 'backing' option. The new value can be
4939 * either a reference to an existing node (using its node name)
4940 * or NULL to simply detach the current backing file.
4942 ret
= bdrv_reopen_parse_file_or_backing(reopen_state
, true,
4943 change_child_tran
, errp
);
4947 qdict_del(reopen_state
->options
, "backing");
4949 /* Allow changing the 'file' option. In this case NULL is not allowed */
4950 ret
= bdrv_reopen_parse_file_or_backing(reopen_state
, false,
4951 change_child_tran
, errp
);
4955 qdict_del(reopen_state
->options
, "file");
4957 /* Options that are not handled are only okay if they are unchanged
4958 * compared to the old state. It is expected that some options are only
4959 * used for the initial open, but not reopen (e.g. filename) */
4960 if (qdict_size(reopen_state
->options
)) {
4961 const QDictEntry
*entry
= qdict_first(reopen_state
->options
);
4963 GRAPH_RDLOCK_GUARD_MAINLOOP();
4966 QObject
*new = entry
->value
;
4967 QObject
*old
= qdict_get(reopen_state
->bs
->options
, entry
->key
);
4969 /* Allow child references (child_name=node_name) as long as they
4970 * point to the current child (i.e. everything stays the same). */
4971 if (qobject_type(new) == QTYPE_QSTRING
) {
4973 QLIST_FOREACH(child
, &reopen_state
->bs
->children
, next
) {
4974 if (!strcmp(child
->name
, entry
->key
)) {
4980 if (!strcmp(child
->bs
->node_name
,
4981 qstring_get_str(qobject_to(QString
, new)))) {
4982 continue; /* Found child with this name, skip option */
4988 * TODO: When using -drive to specify blockdev options, all values
4989 * will be strings; however, when using -blockdev, blockdev-add or
4990 * filenames using the json:{} pseudo-protocol, they will be
4992 * In contrast, reopening options are (currently) always strings
4993 * (because you can only specify them through qemu-io; all other
4994 * callers do not specify any options).
4995 * Therefore, when using anything other than -drive to create a BDS,
4996 * this cannot detect non-string options as unchanged, because
4997 * qobject_is_equal() always returns false for objects of different
4998 * type. In the future, this should be remedied by correctly typing
4999 * all options. For now, this is not too big of an issue because
5000 * the user can simply omit options which cannot be changed anyway,
5001 * so they will stay unchanged.
5003 if (!qobject_is_equal(new, old
)) {
5004 error_setg(errp
, "Cannot change the option '%s'", entry
->key
);
5008 } while ((entry
= qdict_next(reopen_state
->options
, entry
)));
5013 /* Restore the original reopen_state->options QDict */
5014 qobject_unref(reopen_state
->options
);
5015 reopen_state
->options
= qobject_ref(orig_reopen_opts
);
5018 if (ret
< 0 && drv_prepared
) {
5019 /* drv->bdrv_reopen_prepare() has succeeded, so we need to
5020 * call drv->bdrv_reopen_abort() before signaling an error
5021 * (bdrv_reopen_multiple() will not call bdrv_reopen_abort()
5022 * when the respective bdrv_reopen_prepare() has failed) */
5023 if (drv
->bdrv_reopen_abort
) {
5024 drv
->bdrv_reopen_abort(reopen_state
);
5027 qemu_opts_del(opts
);
5028 qobject_unref(orig_reopen_opts
);
5034 * Takes the staged changes for the reopen from bdrv_reopen_prepare(), and
5035 * makes them final by swapping the staging BlockDriverState contents into
5036 * the active BlockDriverState contents.
5038 static void GRAPH_UNLOCKED
bdrv_reopen_commit(BDRVReopenState
*reopen_state
)
5041 BlockDriverState
*bs
;
5044 assert(reopen_state
!= NULL
);
5045 bs
= reopen_state
->bs
;
5047 assert(drv
!= NULL
);
5048 GLOBAL_STATE_CODE();
5050 /* If there are any driver level actions to take */
5051 if (drv
->bdrv_reopen_commit
) {
5052 drv
->bdrv_reopen_commit(reopen_state
);
5055 GRAPH_RDLOCK_GUARD_MAINLOOP();
5057 /* set BDS specific flags now */
5058 qobject_unref(bs
->explicit_options
);
5059 qobject_unref(bs
->options
);
5060 qobject_ref(reopen_state
->explicit_options
);
5061 qobject_ref(reopen_state
->options
);
5063 bs
->explicit_options
= reopen_state
->explicit_options
;
5064 bs
->options
= reopen_state
->options
;
5065 bs
->open_flags
= reopen_state
->flags
;
5066 bs
->detect_zeroes
= reopen_state
->detect_zeroes
;
5068 /* Remove child references from bs->options and bs->explicit_options.
5069 * Child options were already removed in bdrv_reopen_queue_child() */
5070 QLIST_FOREACH(child
, &bs
->children
, next
) {
5071 qdict_del(bs
->explicit_options
, child
->name
);
5072 qdict_del(bs
->options
, child
->name
);
5074 /* backing is probably removed, so it's not handled by previous loop */
5075 qdict_del(bs
->explicit_options
, "backing");
5076 qdict_del(bs
->options
, "backing");
5078 bdrv_refresh_limits(bs
, NULL
, NULL
);
5079 bdrv_refresh_total_sectors(bs
, bs
->total_sectors
);
5083 * Abort the reopen, and delete and free the staged changes in
5086 static void GRAPH_UNLOCKED
bdrv_reopen_abort(BDRVReopenState
*reopen_state
)
5090 assert(reopen_state
!= NULL
);
5091 drv
= reopen_state
->bs
->drv
;
5092 assert(drv
!= NULL
);
5093 GLOBAL_STATE_CODE();
5095 if (drv
->bdrv_reopen_abort
) {
5096 drv
->bdrv_reopen_abort(reopen_state
);
5101 static void bdrv_close(BlockDriverState
*bs
)
5103 BdrvAioNotifier
*ban
, *ban_next
;
5104 BdrvChild
*child
, *next
;
5106 GLOBAL_STATE_CODE();
5107 assert(!bs
->refcnt
);
5109 bdrv_drained_begin(bs
); /* complete I/O */
5111 bdrv_drain(bs
); /* in case flush left pending I/O */
5114 if (bs
->drv
->bdrv_close
) {
5115 /* Must unfreeze all children, so bdrv_unref_child() works */
5116 bs
->drv
->bdrv_close(bs
);
5121 bdrv_graph_wrlock();
5122 QLIST_FOREACH_SAFE(child
, &bs
->children
, next
, next
) {
5123 bdrv_unref_child(bs
, child
);
5126 assert(!bs
->backing
);
5128 bdrv_graph_wrunlock();
5132 qatomic_set(&bs
->copy_on_read
, 0);
5133 bs
->backing_file
[0] = '\0';
5134 bs
->backing_format
[0] = '\0';
5135 bs
->total_sectors
= 0;
5136 bs
->encrypted
= false;
5138 qobject_unref(bs
->options
);
5139 qobject_unref(bs
->explicit_options
);
5141 bs
->explicit_options
= NULL
;
5142 qobject_unref(bs
->full_open_options
);
5143 bs
->full_open_options
= NULL
;
5144 g_free(bs
->block_status_cache
);
5145 bs
->block_status_cache
= NULL
;
5147 bdrv_release_named_dirty_bitmaps(bs
);
5148 assert(QLIST_EMPTY(&bs
->dirty_bitmaps
));
5150 QLIST_FOREACH_SAFE(ban
, &bs
->aio_notifiers
, list
, ban_next
) {
5153 QLIST_INIT(&bs
->aio_notifiers
);
5154 bdrv_drained_end(bs
);
5157 * If we're still inside some bdrv_drain_all_begin()/end() sections, end
5158 * them now since this BDS won't exist anymore when bdrv_drain_all_end()
5161 if (bs
->quiesce_counter
) {
5162 bdrv_drain_all_end_quiesce(bs
);
5166 void bdrv_close_all(void)
5168 GLOBAL_STATE_CODE();
5169 assert(job_next(NULL
) == NULL
);
5171 /* Drop references from requests still in flight, such as canceled block
5172 * jobs whose AIO context has not been polled yet */
5175 blk_remove_all_bs();
5176 blockdev_close_all_bdrv_states();
5178 assert(QTAILQ_EMPTY(&all_bdrv_states
));
5181 static bool GRAPH_RDLOCK
should_update_child(BdrvChild
*c
, BlockDriverState
*to
)
5187 if (c
->klass
->stay_at_node
) {
5191 /* If the child @c belongs to the BDS @to, replacing the current
5192 * c->bs by @to would mean to create a loop.
5194 * Such a case occurs when appending a BDS to a backing chain.
5195 * For instance, imagine the following chain:
5197 * guest device -> node A -> further backing chain...
5199 * Now we create a new BDS B which we want to put on top of this
5200 * chain, so we first attach A as its backing node:
5205 * guest device -> node A -> further backing chain...
5207 * Finally we want to replace A by B. When doing that, we want to
5208 * replace all pointers to A by pointers to B -- except for the
5209 * pointer from B because (1) that would create a loop, and (2)
5210 * that pointer should simply stay intact:
5212 * guest device -> node B
5215 * node A -> further backing chain...
5217 * In general, when replacing a node A (c->bs) by a node B (@to),
5218 * if A is a child of B, that means we cannot replace A by B there
5219 * because that would create a loop. Silently detaching A from B
5220 * is also not really an option. So overall just leaving A in
5221 * place there is the most sensible choice.
5223 * We would also create a loop in any cases where @c is only
5224 * indirectly referenced by @to. Prevent this by returning false
5225 * if @c is found (by breadth-first search) anywhere in the whole
5230 found
= g_hash_table_new(NULL
, NULL
);
5231 g_hash_table_add(found
, to
);
5232 queue
= g_queue_new();
5233 g_queue_push_tail(queue
, to
);
5235 while (!g_queue_is_empty(queue
)) {
5236 BlockDriverState
*v
= g_queue_pop_head(queue
);
5239 QLIST_FOREACH(c2
, &v
->children
, next
) {
5245 if (g_hash_table_contains(found
, c2
->bs
)) {
5249 g_queue_push_tail(queue
, c2
->bs
);
5250 g_hash_table_add(found
, c2
->bs
);
5254 g_queue_free(queue
);
5255 g_hash_table_destroy(found
);
5260 static void bdrv_remove_child_commit(void *opaque
)
5262 GLOBAL_STATE_CODE();
5263 bdrv_child_free(opaque
);
5266 static TransactionActionDrv bdrv_remove_child_drv
= {
5267 .commit
= bdrv_remove_child_commit
,
5271 * Function doesn't update permissions, caller is responsible for this.
5273 * @child->bs (if non-NULL) must be drained.
5275 * After calling this function, the transaction @tran may only be completed
5276 * while holding a writer lock for the graph.
5278 static void GRAPH_WRLOCK
bdrv_remove_child(BdrvChild
*child
, Transaction
*tran
)
5285 assert(child
->quiesced_parent
);
5286 bdrv_replace_child_tran(child
, NULL
, tran
);
5289 tran_add(tran
, &bdrv_remove_child_drv
, child
);
5293 * Both @from and @to (if non-NULL) must be drained. @to must be kept drained
5294 * until the transaction is completed.
5296 * After calling this function, the transaction @tran may only be completed
5297 * while holding a writer lock for the graph.
5299 static int GRAPH_WRLOCK
5300 bdrv_replace_node_noperm(BlockDriverState
*from
,
5301 BlockDriverState
*to
,
5302 bool auto_skip
, Transaction
*tran
,
5305 BdrvChild
*c
, *next
;
5307 GLOBAL_STATE_CODE();
5309 assert(from
->quiesce_counter
);
5310 assert(to
->quiesce_counter
);
5312 QLIST_FOREACH_SAFE(c
, &from
->parents
, next_parent
, next
) {
5313 assert(c
->bs
== from
);
5314 if (!should_update_child(c
, to
)) {
5318 error_setg(errp
, "Should not change '%s' link to '%s'",
5319 c
->name
, from
->node_name
);
5323 error_setg(errp
, "Cannot change '%s' link to '%s'",
5324 c
->name
, from
->node_name
);
5327 bdrv_replace_child_tran(c
, to
, tran
);
5334 * Switch all parents of @from to point to @to instead. @from and @to must be in
5335 * the same AioContext and both must be drained.
5337 * With auto_skip=true bdrv_replace_node_common skips updating from parents
5338 * if it creates a parent-child relation loop or if parent is block-job.
5340 * With auto_skip=false the error is returned if from has a parent which should
5343 * With @detach_subchain=true @to must be in a backing chain of @from. In this
5344 * case backing link of the cow-parent of @to is removed.
5346 static int GRAPH_WRLOCK
5347 bdrv_replace_node_common(BlockDriverState
*from
, BlockDriverState
*to
,
5348 bool auto_skip
, bool detach_subchain
, Error
**errp
)
5350 Transaction
*tran
= tran_new();
5351 g_autoptr(GSList
) refresh_list
= NULL
;
5352 BlockDriverState
*to_cow_parent
= NULL
;
5355 GLOBAL_STATE_CODE();
5357 assert(from
->quiesce_counter
);
5358 assert(to
->quiesce_counter
);
5359 assert(bdrv_get_aio_context(from
) == bdrv_get_aio_context(to
));
5361 if (detach_subchain
) {
5362 assert(bdrv_chain_contains(from
, to
));
5364 for (to_cow_parent
= from
;
5365 bdrv_filter_or_cow_bs(to_cow_parent
) != to
;
5366 to_cow_parent
= bdrv_filter_or_cow_bs(to_cow_parent
))
5373 * Do the replacement without permission update.
5374 * Replacement may influence the permissions, we should calculate new
5375 * permissions based on new graph. If we fail, we'll roll-back the
5378 ret
= bdrv_replace_node_noperm(from
, to
, auto_skip
, tran
, errp
);
5383 if (detach_subchain
) {
5384 /* to_cow_parent is already drained because from is drained */
5385 bdrv_remove_child(bdrv_filter_or_cow_child(to_cow_parent
), tran
);
5388 refresh_list
= g_slist_prepend(refresh_list
, to
);
5389 refresh_list
= g_slist_prepend(refresh_list
, from
);
5391 ret
= bdrv_list_refresh_perms(refresh_list
, NULL
, tran
, errp
);
5399 tran_finalize(tran
, ret
);
5403 int bdrv_replace_node(BlockDriverState
*from
, BlockDriverState
*to
,
5406 return bdrv_replace_node_common(from
, to
, true, false, errp
);
5409 int bdrv_drop_filter(BlockDriverState
*bs
, Error
**errp
)
5411 BlockDriverState
*child_bs
;
5414 GLOBAL_STATE_CODE();
5416 bdrv_graph_rdlock_main_loop();
5417 child_bs
= bdrv_filter_or_cow_bs(bs
);
5418 bdrv_graph_rdunlock_main_loop();
5420 bdrv_drained_begin(child_bs
);
5421 bdrv_graph_wrlock();
5422 ret
= bdrv_replace_node_common(bs
, child_bs
, true, true, errp
);
5423 bdrv_graph_wrunlock();
5424 bdrv_drained_end(child_bs
);
5430 * Add new bs contents at the top of an image chain while the chain is
5431 * live, while keeping required fields on the top layer.
5433 * This will modify the BlockDriverState fields, and swap contents
5434 * between bs_new and bs_top. Both bs_new and bs_top are modified.
5436 * bs_new must not be attached to a BlockBackend and must not have backing
5439 * This function does not create any image files.
5441 * The caller must hold the AioContext lock for @bs_top.
5443 int bdrv_append(BlockDriverState
*bs_new
, BlockDriverState
*bs_top
,
5448 Transaction
*tran
= tran_new();
5450 GLOBAL_STATE_CODE();
5452 bdrv_graph_rdlock_main_loop();
5453 assert(!bs_new
->backing
);
5454 bdrv_graph_rdunlock_main_loop();
5456 bdrv_drained_begin(bs_top
);
5457 bdrv_drained_begin(bs_new
);
5459 bdrv_graph_wrlock();
5461 child
= bdrv_attach_child_noperm(bs_new
, bs_top
, "backing",
5462 &child_of_bds
, bdrv_backing_role(bs_new
),
5469 ret
= bdrv_replace_node_noperm(bs_top
, bs_new
, true, tran
, errp
);
5474 ret
= bdrv_refresh_perms(bs_new
, tran
, errp
);
5476 tran_finalize(tran
, ret
);
5478 bdrv_refresh_limits(bs_top
, NULL
, NULL
);
5479 bdrv_graph_wrunlock();
5481 bdrv_drained_end(bs_top
);
5482 bdrv_drained_end(bs_new
);
5487 /* Not for empty child */
5488 int bdrv_replace_child_bs(BdrvChild
*child
, BlockDriverState
*new_bs
,
5492 Transaction
*tran
= tran_new();
5493 g_autoptr(GSList
) refresh_list
= NULL
;
5494 BlockDriverState
*old_bs
= child
->bs
;
5496 GLOBAL_STATE_CODE();
5499 bdrv_drained_begin(old_bs
);
5500 bdrv_drained_begin(new_bs
);
5501 bdrv_graph_wrlock();
5503 bdrv_replace_child_tran(child
, new_bs
, tran
);
5505 refresh_list
= g_slist_prepend(refresh_list
, old_bs
);
5506 refresh_list
= g_slist_prepend(refresh_list
, new_bs
);
5508 ret
= bdrv_list_refresh_perms(refresh_list
, NULL
, tran
, errp
);
5510 tran_finalize(tran
, ret
);
5512 bdrv_graph_wrunlock();
5513 bdrv_drained_end(old_bs
);
5514 bdrv_drained_end(new_bs
);
5520 static void bdrv_delete(BlockDriverState
*bs
)
5522 assert(bdrv_op_blocker_is_empty(bs
));
5523 assert(!bs
->refcnt
);
5524 GLOBAL_STATE_CODE();
5526 /* remove from list, if necessary */
5527 if (bs
->node_name
[0] != '\0') {
5528 QTAILQ_REMOVE(&graph_bdrv_states
, bs
, node_list
);
5530 QTAILQ_REMOVE(&all_bdrv_states
, bs
, bs_list
);
5534 qemu_mutex_destroy(&bs
->reqs_lock
);
5541 * Replace @bs by newly created block node.
5543 * @options is a QDict of options to pass to the block drivers, or NULL for an
5544 * empty set of options. The reference to the QDict belongs to the block layer
5545 * after the call (even on failure), so if the caller intends to reuse the
5546 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
5548 * The caller holds the AioContext lock for @bs. It must make sure that @bs
5549 * stays in the same AioContext, i.e. @options must not refer to nodes in a
5550 * different AioContext.
5552 BlockDriverState
*bdrv_insert_node(BlockDriverState
*bs
, QDict
*options
,
5553 int flags
, Error
**errp
)
5557 AioContext
*ctx
= bdrv_get_aio_context(bs
);
5558 BlockDriverState
*new_node_bs
= NULL
;
5559 const char *drvname
, *node_name
;
5562 drvname
= qdict_get_try_str(options
, "driver");
5564 error_setg(errp
, "driver is not specified");
5568 drv
= bdrv_find_format(drvname
);
5570 error_setg(errp
, "Unknown driver: '%s'", drvname
);
5574 node_name
= qdict_get_try_str(options
, "node-name");
5576 GLOBAL_STATE_CODE();
5578 new_node_bs
= bdrv_new_open_driver_opts(drv
, node_name
, options
, flags
,
5580 assert(bdrv_get_aio_context(bs
) == ctx
);
5582 options
= NULL
; /* bdrv_new_open_driver() eats options */
5584 error_prepend(errp
, "Could not create node: ");
5589 * Make sure that @bs doesn't go away until we have successfully attached
5590 * all of its parents to @new_node_bs and undrained it again.
5593 bdrv_drained_begin(bs
);
5594 bdrv_drained_begin(new_node_bs
);
5595 bdrv_graph_wrlock();
5596 ret
= bdrv_replace_node(bs
, new_node_bs
, errp
);
5597 bdrv_graph_wrunlock();
5598 bdrv_drained_end(new_node_bs
);
5599 bdrv_drained_end(bs
);
5603 error_prepend(errp
, "Could not replace node: ");
5610 qobject_unref(options
);
5611 bdrv_unref(new_node_bs
);
5616 * Run consistency checks on an image
5618 * Returns 0 if the check could be completed (it doesn't mean that the image is
5619 * free of errors) or -errno when an internal error occurred. The results of the
5620 * check are stored in res.
5622 int coroutine_fn
bdrv_co_check(BlockDriverState
*bs
,
5623 BdrvCheckResult
*res
, BdrvCheckMode fix
)
5626 assert_bdrv_graph_readable();
5627 if (bs
->drv
== NULL
) {
5630 if (bs
->drv
->bdrv_co_check
== NULL
) {
5634 memset(res
, 0, sizeof(*res
));
5635 return bs
->drv
->bdrv_co_check(bs
, res
, fix
);
5641 * -EINVAL - backing format specified, but no file
5642 * -ENOSPC - can't update the backing file because no space is left in the
5644 * -ENOTSUP - format driver doesn't support changing the backing file
5647 bdrv_co_change_backing_file(BlockDriverState
*bs
, const char *backing_file
,
5648 const char *backing_fmt
, bool require
)
5650 BlockDriver
*drv
= bs
->drv
;
5659 /* Backing file format doesn't make sense without a backing file */
5660 if (backing_fmt
&& !backing_file
) {
5664 if (require
&& backing_file
&& !backing_fmt
) {
5668 if (drv
->bdrv_co_change_backing_file
!= NULL
) {
5669 ret
= drv
->bdrv_co_change_backing_file(bs
, backing_file
, backing_fmt
);
5675 pstrcpy(bs
->backing_file
, sizeof(bs
->backing_file
), backing_file
?: "");
5676 pstrcpy(bs
->backing_format
, sizeof(bs
->backing_format
), backing_fmt
?: "");
5677 pstrcpy(bs
->auto_backing_file
, sizeof(bs
->auto_backing_file
),
5678 backing_file
?: "");
5684 * Finds the first non-filter node above bs in the chain between
5685 * active and bs. The returned node is either an immediate parent of
5686 * bs, or there are only filter nodes between the two.
5688 * Returns NULL if bs is not found in active's image chain,
5689 * or if active == bs.
5691 * Returns the bottommost base image if bs == NULL.
5693 BlockDriverState
*bdrv_find_overlay(BlockDriverState
*active
,
5694 BlockDriverState
*bs
)
5697 GLOBAL_STATE_CODE();
5699 bs
= bdrv_skip_filters(bs
);
5700 active
= bdrv_skip_filters(active
);
5703 BlockDriverState
*next
= bdrv_backing_chain_next(active
);
5713 /* Given a BDS, searches for the base layer. */
5714 BlockDriverState
*bdrv_find_base(BlockDriverState
*bs
)
5716 GLOBAL_STATE_CODE();
5718 return bdrv_find_overlay(bs
, NULL
);
5722 * Return true if at least one of the COW (backing) and filter links
5723 * between @bs and @base is frozen. @errp is set if that's the case.
5724 * @base must be reachable from @bs, or NULL.
5726 static bool GRAPH_RDLOCK
5727 bdrv_is_backing_chain_frozen(BlockDriverState
*bs
, BlockDriverState
*base
,
5730 BlockDriverState
*i
;
5733 GLOBAL_STATE_CODE();
5735 for (i
= bs
; i
!= base
; i
= child_bs(child
)) {
5736 child
= bdrv_filter_or_cow_child(i
);
5738 if (child
&& child
->frozen
) {
5739 error_setg(errp
, "Cannot change '%s' link from '%s' to '%s'",
5740 child
->name
, i
->node_name
, child
->bs
->node_name
);
5749 * Freeze all COW (backing) and filter links between @bs and @base.
5750 * If any of the links is already frozen the operation is aborted and
5751 * none of the links are modified.
5752 * @base must be reachable from @bs, or NULL.
5753 * Returns 0 on success. On failure returns < 0 and sets @errp.
5755 int bdrv_freeze_backing_chain(BlockDriverState
*bs
, BlockDriverState
*base
,
5758 BlockDriverState
*i
;
5761 GLOBAL_STATE_CODE();
5763 if (bdrv_is_backing_chain_frozen(bs
, base
, errp
)) {
5767 for (i
= bs
; i
!= base
; i
= child_bs(child
)) {
5768 child
= bdrv_filter_or_cow_child(i
);
5769 if (child
&& child
->bs
->never_freeze
) {
5770 error_setg(errp
, "Cannot freeze '%s' link to '%s'",
5771 child
->name
, child
->bs
->node_name
);
5776 for (i
= bs
; i
!= base
; i
= child_bs(child
)) {
5777 child
= bdrv_filter_or_cow_child(i
);
5779 child
->frozen
= true;
5787 * Unfreeze all COW (backing) and filter links between @bs and @base.
5788 * The caller must ensure that all links are frozen before using this
5790 * @base must be reachable from @bs, or NULL.
5792 void bdrv_unfreeze_backing_chain(BlockDriverState
*bs
, BlockDriverState
*base
)
5794 BlockDriverState
*i
;
5797 GLOBAL_STATE_CODE();
5799 for (i
= bs
; i
!= base
; i
= child_bs(child
)) {
5800 child
= bdrv_filter_or_cow_child(i
);
5802 assert(child
->frozen
);
5803 child
->frozen
= false;
5809 * Drops images above 'base' up to and including 'top', and sets the image
5810 * above 'top' to have base as its backing file.
5812 * Requires that the overlay to 'top' is opened r/w, so that the backing file
5813 * information in 'bs' can be properly updated.
5815 * E.g., this will convert the following chain:
5816 * bottom <- base <- intermediate <- top <- active
5820 * bottom <- base <- active
5822 * It is allowed for bottom==base, in which case it converts:
5824 * base <- intermediate <- top <- active
5830 * If backing_file_str is non-NULL, it will be used when modifying top's
5831 * overlay image metadata.
5834 * if active == top, that is considered an error
5837 int bdrv_drop_intermediate(BlockDriverState
*top
, BlockDriverState
*base
,
5838 const char *backing_file_str
)
5840 BlockDriverState
*explicit_top
= top
;
5841 bool update_inherits_from
;
5843 Error
*local_err
= NULL
;
5845 g_autoptr(GSList
) updated_children
= NULL
;
5848 GLOBAL_STATE_CODE();
5851 bdrv_drained_begin(base
);
5852 bdrv_graph_wrlock();
5854 if (!top
->drv
|| !base
->drv
) {
5858 /* Make sure that base is in the backing chain of top */
5859 if (!bdrv_chain_contains(top
, base
)) {
5863 /* If 'base' recursively inherits from 'top' then we should set
5864 * base->inherits_from to top->inherits_from after 'top' and all
5865 * other intermediate nodes have been dropped.
5866 * If 'top' is an implicit node (e.g. "commit_top") we should skip
5867 * it because no one inherits from it. We use explicit_top for that. */
5868 explicit_top
= bdrv_skip_implicit_filters(explicit_top
);
5869 update_inherits_from
= bdrv_inherits_from_recursive(base
, explicit_top
);
5871 /* success - we can delete the intermediate states, and link top->base */
5872 if (!backing_file_str
) {
5873 bdrv_refresh_filename(base
);
5874 backing_file_str
= base
->filename
;
5877 QLIST_FOREACH(c
, &top
->parents
, next_parent
) {
5878 updated_children
= g_slist_prepend(updated_children
, c
);
5882 * It seems correct to pass detach_subchain=true here, but it triggers
5883 * one more yet not fixed bug, when due to nested aio_poll loop we switch to
5884 * another drained section, which modify the graph (for example, removing
5885 * the child, which we keep in updated_children list). So, it's a TODO.
5887 * Note, bug triggered if pass detach_subchain=true here and run
5888 * test-bdrv-drain. test_drop_intermediate_poll() test-case will crash.
5891 bdrv_replace_node_common(top
, base
, false, false, &local_err
);
5892 bdrv_graph_wrunlock();
5895 error_report_err(local_err
);
5899 for (p
= updated_children
; p
; p
= p
->next
) {
5902 if (c
->klass
->update_filename
) {
5903 ret
= c
->klass
->update_filename(c
, base
, backing_file_str
,
5907 * TODO: Actually, we want to rollback all previous iterations
5908 * of this loop, and (which is almost impossible) previous
5909 * bdrv_replace_node()...
5911 * Note, that c->klass->update_filename may lead to permission
5912 * update, so it's a bad idea to call it inside permission
5913 * update transaction of bdrv_replace_node.
5915 error_report_err(local_err
);
5921 if (update_inherits_from
) {
5922 base
->inherits_from
= explicit_top
->inherits_from
;
5929 bdrv_graph_wrunlock();
5931 bdrv_drained_end(base
);
5937 * Implementation of BlockDriver.bdrv_co_get_allocated_file_size() that
5938 * sums the size of all data-bearing children. (This excludes backing
5941 static int64_t coroutine_fn GRAPH_RDLOCK
5942 bdrv_sum_allocated_file_size(BlockDriverState
*bs
)
5945 int64_t child_size
, sum
= 0;
5947 QLIST_FOREACH(child
, &bs
->children
, next
) {
5948 if (child
->role
& (BDRV_CHILD_DATA
| BDRV_CHILD_METADATA
|
5949 BDRV_CHILD_FILTERED
))
5951 child_size
= bdrv_co_get_allocated_file_size(child
->bs
);
5952 if (child_size
< 0) {
5963 * Length of a allocated file in bytes. Sparse files are counted by actual
5964 * allocated space. Return < 0 if error or unknown.
5966 int64_t coroutine_fn
bdrv_co_get_allocated_file_size(BlockDriverState
*bs
)
5968 BlockDriver
*drv
= bs
->drv
;
5970 assert_bdrv_graph_readable();
5975 if (drv
->bdrv_co_get_allocated_file_size
) {
5976 return drv
->bdrv_co_get_allocated_file_size(bs
);
5979 if (drv
->bdrv_file_open
) {
5981 * Protocol drivers default to -ENOTSUP (most of their data is
5982 * not stored in any of their children (if they even have any),
5983 * so there is no generic way to figure it out).
5986 } else if (drv
->is_filter
) {
5987 /* Filter drivers default to the size of their filtered child */
5988 return bdrv_co_get_allocated_file_size(bdrv_filter_bs(bs
));
5990 /* Other drivers default to summing their children's sizes */
5991 return bdrv_sum_allocated_file_size(bs
);
5997 * @drv: Format driver
5998 * @opts: Creation options for new image
5999 * @in_bs: Existing image containing data for new image (may be NULL)
6000 * @errp: Error object
6001 * Returns: A #BlockMeasureInfo (free using qapi_free_BlockMeasureInfo())
6004 * Calculate file size required to create a new image.
6006 * If @in_bs is given then space for allocated clusters and zero clusters
6007 * from that image are included in the calculation. If @opts contains a
6008 * backing file that is shared by @in_bs then backing clusters may be omitted
6009 * from the calculation.
6011 * If @in_bs is NULL then the calculation includes no allocated clusters
6012 * unless a preallocation option is given in @opts.
6014 * Note that @in_bs may use a different BlockDriver from @drv.
6016 * If an error occurs the @errp pointer is set.
6018 BlockMeasureInfo
*bdrv_measure(BlockDriver
*drv
, QemuOpts
*opts
,
6019 BlockDriverState
*in_bs
, Error
**errp
)
6022 if (!drv
->bdrv_measure
) {
6023 error_setg(errp
, "Block driver '%s' does not support size measurement",
6028 return drv
->bdrv_measure(opts
, in_bs
, errp
);
6032 * Return number of sectors on success, -errno on error.
6034 int64_t coroutine_fn
bdrv_co_nb_sectors(BlockDriverState
*bs
)
6036 BlockDriver
*drv
= bs
->drv
;
6038 assert_bdrv_graph_readable();
6043 if (bs
->bl
.has_variable_length
) {
6044 int ret
= bdrv_co_refresh_total_sectors(bs
, bs
->total_sectors
);
6049 return bs
->total_sectors
;
6053 * This wrapper is written by hand because this function is in the hot I/O path,
6054 * via blk_get_geometry.
6056 int64_t coroutine_mixed_fn
bdrv_nb_sectors(BlockDriverState
*bs
)
6058 BlockDriver
*drv
= bs
->drv
;
6064 if (bs
->bl
.has_variable_length
) {
6065 int ret
= bdrv_refresh_total_sectors(bs
, bs
->total_sectors
);
6071 return bs
->total_sectors
;
6075 * Return length in bytes on success, -errno on error.
6076 * The length is always a multiple of BDRV_SECTOR_SIZE.
6078 int64_t coroutine_fn
bdrv_co_getlength(BlockDriverState
*bs
)
6082 assert_bdrv_graph_readable();
6084 ret
= bdrv_co_nb_sectors(bs
);
6088 if (ret
> INT64_MAX
/ BDRV_SECTOR_SIZE
) {
6091 return ret
* BDRV_SECTOR_SIZE
;
6094 bool bdrv_is_sg(BlockDriverState
*bs
)
6101 * Return whether the given node supports compressed writes.
6103 bool bdrv_supports_compressed_writes(BlockDriverState
*bs
)
6105 BlockDriverState
*filtered
;
6108 if (!bs
->drv
|| !block_driver_can_compress(bs
->drv
)) {
6112 filtered
= bdrv_filter_bs(bs
);
6115 * Filters can only forward compressed writes, so we have to
6118 return bdrv_supports_compressed_writes(filtered
);
6124 const char *bdrv_get_format_name(BlockDriverState
*bs
)
6127 return bs
->drv
? bs
->drv
->format_name
: NULL
;
6130 static int qsort_strcmp(const void *a
, const void *b
)
6132 return strcmp(*(char *const *)a
, *(char *const *)b
);
6135 void bdrv_iterate_format(void (*it
)(void *opaque
, const char *name
),
6136 void *opaque
, bool read_only
)
6141 const char **formats
= NULL
;
6143 GLOBAL_STATE_CODE();
6145 QLIST_FOREACH(drv
, &bdrv_drivers
, list
) {
6146 if (drv
->format_name
) {
6149 if (use_bdrv_whitelist
&& !bdrv_is_whitelisted(drv
, read_only
)) {
6154 while (formats
&& i
&& !found
) {
6155 found
= !strcmp(formats
[--i
], drv
->format_name
);
6159 formats
= g_renew(const char *, formats
, count
+ 1);
6160 formats
[count
++] = drv
->format_name
;
6165 for (i
= 0; i
< (int)ARRAY_SIZE(block_driver_modules
); i
++) {
6166 const char *format_name
= block_driver_modules
[i
].format_name
;
6172 if (use_bdrv_whitelist
&&
6173 !bdrv_format_is_whitelisted(format_name
, read_only
)) {
6177 while (formats
&& j
&& !found
) {
6178 found
= !strcmp(formats
[--j
], format_name
);
6182 formats
= g_renew(const char *, formats
, count
+ 1);
6183 formats
[count
++] = format_name
;
6188 qsort(formats
, count
, sizeof(formats
[0]), qsort_strcmp
);
6190 for (i
= 0; i
< count
; i
++) {
6191 it(opaque
, formats
[i
]);
6197 /* This function is to find a node in the bs graph */
6198 BlockDriverState
*bdrv_find_node(const char *node_name
)
6200 BlockDriverState
*bs
;
6203 GLOBAL_STATE_CODE();
6205 QTAILQ_FOREACH(bs
, &graph_bdrv_states
, node_list
) {
6206 if (!strcmp(node_name
, bs
->node_name
)) {
6213 /* Put this QMP function here so it can access the static graph_bdrv_states. */
6214 BlockDeviceInfoList
*bdrv_named_nodes_list(bool flat
,
6217 BlockDeviceInfoList
*list
;
6218 BlockDriverState
*bs
;
6220 GLOBAL_STATE_CODE();
6221 GRAPH_RDLOCK_GUARD_MAINLOOP();
6224 QTAILQ_FOREACH(bs
, &graph_bdrv_states
, node_list
) {
6225 BlockDeviceInfo
*info
= bdrv_block_device_info(NULL
, bs
, flat
, errp
);
6227 qapi_free_BlockDeviceInfoList(list
);
6230 QAPI_LIST_PREPEND(list
, info
);
6236 typedef struct XDbgBlockGraphConstructor
{
6237 XDbgBlockGraph
*graph
;
6238 GHashTable
*graph_nodes
;
6239 } XDbgBlockGraphConstructor
;
6241 static XDbgBlockGraphConstructor
*xdbg_graph_new(void)
6243 XDbgBlockGraphConstructor
*gr
= g_new(XDbgBlockGraphConstructor
, 1);
6245 gr
->graph
= g_new0(XDbgBlockGraph
, 1);
6246 gr
->graph_nodes
= g_hash_table_new(NULL
, NULL
);
6251 static XDbgBlockGraph
*xdbg_graph_finalize(XDbgBlockGraphConstructor
*gr
)
6253 XDbgBlockGraph
*graph
= gr
->graph
;
6255 g_hash_table_destroy(gr
->graph_nodes
);
6261 static uintptr_t xdbg_graph_node_num(XDbgBlockGraphConstructor
*gr
, void *node
)
6263 uintptr_t ret
= (uintptr_t)g_hash_table_lookup(gr
->graph_nodes
, node
);
6270 * Start counting from 1, not 0, because 0 interferes with not-found (NULL)
6271 * answer of g_hash_table_lookup.
6273 ret
= g_hash_table_size(gr
->graph_nodes
) + 1;
6274 g_hash_table_insert(gr
->graph_nodes
, node
, (void *)ret
);
6279 static void xdbg_graph_add_node(XDbgBlockGraphConstructor
*gr
, void *node
,
6280 XDbgBlockGraphNodeType type
, const char *name
)
6282 XDbgBlockGraphNode
*n
;
6284 n
= g_new0(XDbgBlockGraphNode
, 1);
6286 n
->id
= xdbg_graph_node_num(gr
, node
);
6288 n
->name
= g_strdup(name
);
6290 QAPI_LIST_PREPEND(gr
->graph
->nodes
, n
);
6293 static void xdbg_graph_add_edge(XDbgBlockGraphConstructor
*gr
, void *parent
,
6294 const BdrvChild
*child
)
6296 BlockPermission qapi_perm
;
6297 XDbgBlockGraphEdge
*edge
;
6298 GLOBAL_STATE_CODE();
6300 edge
= g_new0(XDbgBlockGraphEdge
, 1);
6302 edge
->parent
= xdbg_graph_node_num(gr
, parent
);
6303 edge
->child
= xdbg_graph_node_num(gr
, child
->bs
);
6304 edge
->name
= g_strdup(child
->name
);
6306 for (qapi_perm
= 0; qapi_perm
< BLOCK_PERMISSION__MAX
; qapi_perm
++) {
6307 uint64_t flag
= bdrv_qapi_perm_to_blk_perm(qapi_perm
);
6309 if (flag
& child
->perm
) {
6310 QAPI_LIST_PREPEND(edge
->perm
, qapi_perm
);
6312 if (flag
& child
->shared_perm
) {
6313 QAPI_LIST_PREPEND(edge
->shared_perm
, qapi_perm
);
6317 QAPI_LIST_PREPEND(gr
->graph
->edges
, edge
);
6321 XDbgBlockGraph
*bdrv_get_xdbg_block_graph(Error
**errp
)
6325 BlockDriverState
*bs
;
6327 XDbgBlockGraphConstructor
*gr
= xdbg_graph_new();
6329 GLOBAL_STATE_CODE();
6331 for (blk
= blk_all_next(NULL
); blk
; blk
= blk_all_next(blk
)) {
6332 char *allocated_name
= NULL
;
6333 const char *name
= blk_name(blk
);
6336 name
= allocated_name
= blk_get_attached_dev_id(blk
);
6338 xdbg_graph_add_node(gr
, blk
, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_BACKEND
,
6340 g_free(allocated_name
);
6341 if (blk_root(blk
)) {
6342 xdbg_graph_add_edge(gr
, blk
, blk_root(blk
));
6346 WITH_JOB_LOCK_GUARD() {
6347 for (job
= block_job_next_locked(NULL
); job
;
6348 job
= block_job_next_locked(job
)) {
6351 xdbg_graph_add_node(gr
, job
, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_JOB
,
6353 for (el
= job
->nodes
; el
; el
= el
->next
) {
6354 xdbg_graph_add_edge(gr
, job
, (BdrvChild
*)el
->data
);
6359 QTAILQ_FOREACH(bs
, &graph_bdrv_states
, node_list
) {
6360 xdbg_graph_add_node(gr
, bs
, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_DRIVER
,
6362 QLIST_FOREACH(child
, &bs
->children
, next
) {
6363 xdbg_graph_add_edge(gr
, bs
, child
);
6367 return xdbg_graph_finalize(gr
);
6370 BlockDriverState
*bdrv_lookup_bs(const char *device
,
6371 const char *node_name
,
6375 BlockDriverState
*bs
;
6377 GLOBAL_STATE_CODE();
6380 blk
= blk_by_name(device
);
6385 error_setg(errp
, "Device '%s' has no medium", device
);
6393 bs
= bdrv_find_node(node_name
);
6400 error_setg(errp
, "Cannot find device=\'%s\' nor node-name=\'%s\'",
6401 device
? device
: "",
6402 node_name
? node_name
: "");
6406 /* If 'base' is in the same chain as 'top', return true. Otherwise,
6407 * return false. If either argument is NULL, return false. */
6408 bool bdrv_chain_contains(BlockDriverState
*top
, BlockDriverState
*base
)
6411 GLOBAL_STATE_CODE();
6413 while (top
&& top
!= base
) {
6414 top
= bdrv_filter_or_cow_bs(top
);
6420 BlockDriverState
*bdrv_next_node(BlockDriverState
*bs
)
6422 GLOBAL_STATE_CODE();
6424 return QTAILQ_FIRST(&graph_bdrv_states
);
6426 return QTAILQ_NEXT(bs
, node_list
);
6429 BlockDriverState
*bdrv_next_all_states(BlockDriverState
*bs
)
6431 GLOBAL_STATE_CODE();
6433 return QTAILQ_FIRST(&all_bdrv_states
);
6435 return QTAILQ_NEXT(bs
, bs_list
);
6438 const char *bdrv_get_node_name(const BlockDriverState
*bs
)
6441 return bs
->node_name
;
6444 const char *bdrv_get_parent_name(const BlockDriverState
*bs
)
6450 /* If multiple parents have a name, just pick the first one. */
6451 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
6452 if (c
->klass
->get_name
) {
6453 name
= c
->klass
->get_name(c
);
6454 if (name
&& *name
) {
6463 /* TODO check what callers really want: bs->node_name or blk_name() */
6464 const char *bdrv_get_device_name(const BlockDriverState
*bs
)
6467 return bdrv_get_parent_name(bs
) ?: "";
6470 /* This can be used to identify nodes that might not have a device
6471 * name associated. Since node and device names live in the same
6472 * namespace, the result is unambiguous. The exception is if both are
6473 * absent, then this returns an empty (non-null) string. */
6474 const char *bdrv_get_device_or_node_name(const BlockDriverState
*bs
)
6477 return bdrv_get_parent_name(bs
) ?: bs
->node_name
;
6480 int bdrv_get_flags(BlockDriverState
*bs
)
6483 return bs
->open_flags
;
6486 int bdrv_has_zero_init_1(BlockDriverState
*bs
)
6488 GLOBAL_STATE_CODE();
6492 int coroutine_mixed_fn
bdrv_has_zero_init(BlockDriverState
*bs
)
6494 BlockDriverState
*filtered
;
6495 GLOBAL_STATE_CODE();
6501 /* If BS is a copy on write image, it is initialized to
6502 the contents of the base image, which may not be zeroes. */
6503 if (bdrv_cow_child(bs
)) {
6506 if (bs
->drv
->bdrv_has_zero_init
) {
6507 return bs
->drv
->bdrv_has_zero_init(bs
);
6510 filtered
= bdrv_filter_bs(bs
);
6512 return bdrv_has_zero_init(filtered
);
6519 bool bdrv_can_write_zeroes_with_unmap(BlockDriverState
*bs
)
6522 if (!(bs
->open_flags
& BDRV_O_UNMAP
)) {
6526 return bs
->supported_zero_flags
& BDRV_REQ_MAY_UNMAP
;
6529 void bdrv_get_backing_filename(BlockDriverState
*bs
,
6530 char *filename
, int filename_size
)
6533 pstrcpy(filename
, filename_size
, bs
->backing_file
);
6536 int coroutine_fn
bdrv_co_get_info(BlockDriverState
*bs
, BlockDriverInfo
*bdi
)
6539 BlockDriver
*drv
= bs
->drv
;
6541 assert_bdrv_graph_readable();
6543 /* if bs->drv == NULL, bs is closed, so there's nothing to do here */
6547 if (!drv
->bdrv_co_get_info
) {
6548 BlockDriverState
*filtered
= bdrv_filter_bs(bs
);
6550 return bdrv_co_get_info(filtered
, bdi
);
6554 memset(bdi
, 0, sizeof(*bdi
));
6555 ret
= drv
->bdrv_co_get_info(bs
, bdi
);
6556 if (bdi
->subcluster_size
== 0) {
6558 * If the driver left this unset, subclusters are not supported.
6559 * Then it is safe to treat each cluster as having only one subcluster.
6561 bdi
->subcluster_size
= bdi
->cluster_size
;
6567 if (bdi
->cluster_size
> BDRV_MAX_ALIGNMENT
) {
6574 ImageInfoSpecific
*bdrv_get_specific_info(BlockDriverState
*bs
,
6577 BlockDriver
*drv
= bs
->drv
;
6579 if (drv
&& drv
->bdrv_get_specific_info
) {
6580 return drv
->bdrv_get_specific_info(bs
, errp
);
6585 BlockStatsSpecific
*bdrv_get_specific_stats(BlockDriverState
*bs
)
6587 BlockDriver
*drv
= bs
->drv
;
6589 if (!drv
|| !drv
->bdrv_get_specific_stats
) {
6592 return drv
->bdrv_get_specific_stats(bs
);
6595 void coroutine_fn
bdrv_co_debug_event(BlockDriverState
*bs
, BlkdebugEvent event
)
6598 assert_bdrv_graph_readable();
6600 if (!bs
|| !bs
->drv
|| !bs
->drv
->bdrv_co_debug_event
) {
6604 bs
->drv
->bdrv_co_debug_event(bs
, event
);
6607 static BlockDriverState
* GRAPH_RDLOCK
6608 bdrv_find_debug_node(BlockDriverState
*bs
)
6610 GLOBAL_STATE_CODE();
6611 while (bs
&& bs
->drv
&& !bs
->drv
->bdrv_debug_breakpoint
) {
6612 bs
= bdrv_primary_bs(bs
);
6615 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_breakpoint
) {
6616 assert(bs
->drv
->bdrv_debug_remove_breakpoint
);
6623 int bdrv_debug_breakpoint(BlockDriverState
*bs
, const char *event
,
6626 GLOBAL_STATE_CODE();
6627 GRAPH_RDLOCK_GUARD_MAINLOOP();
6629 bs
= bdrv_find_debug_node(bs
);
6631 return bs
->drv
->bdrv_debug_breakpoint(bs
, event
, tag
);
6637 int bdrv_debug_remove_breakpoint(BlockDriverState
*bs
, const char *tag
)
6639 GLOBAL_STATE_CODE();
6640 GRAPH_RDLOCK_GUARD_MAINLOOP();
6642 bs
= bdrv_find_debug_node(bs
);
6644 return bs
->drv
->bdrv_debug_remove_breakpoint(bs
, tag
);
6650 int bdrv_debug_resume(BlockDriverState
*bs
, const char *tag
)
6652 GLOBAL_STATE_CODE();
6653 GRAPH_RDLOCK_GUARD_MAINLOOP();
6655 while (bs
&& (!bs
->drv
|| !bs
->drv
->bdrv_debug_resume
)) {
6656 bs
= bdrv_primary_bs(bs
);
6659 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_resume
) {
6660 return bs
->drv
->bdrv_debug_resume(bs
, tag
);
6666 bool bdrv_debug_is_suspended(BlockDriverState
*bs
, const char *tag
)
6668 GLOBAL_STATE_CODE();
6669 GRAPH_RDLOCK_GUARD_MAINLOOP();
6671 while (bs
&& bs
->drv
&& !bs
->drv
->bdrv_debug_is_suspended
) {
6672 bs
= bdrv_primary_bs(bs
);
6675 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_is_suspended
) {
6676 return bs
->drv
->bdrv_debug_is_suspended(bs
, tag
);
6682 /* backing_file can either be relative, or absolute, or a protocol. If it is
6683 * relative, it must be relative to the chain. So, passing in bs->filename
6684 * from a BDS as backing_file should not be done, as that may be relative to
6685 * the CWD rather than the chain. */
6686 BlockDriverState
*bdrv_find_backing_image(BlockDriverState
*bs
,
6687 const char *backing_file
)
6689 char *filename_full
= NULL
;
6690 char *backing_file_full
= NULL
;
6691 char *filename_tmp
= NULL
;
6692 int is_protocol
= 0;
6693 bool filenames_refreshed
= false;
6694 BlockDriverState
*curr_bs
= NULL
;
6695 BlockDriverState
*retval
= NULL
;
6696 BlockDriverState
*bs_below
;
6698 GLOBAL_STATE_CODE();
6699 GRAPH_RDLOCK_GUARD_MAINLOOP();
6701 if (!bs
|| !bs
->drv
|| !backing_file
) {
6705 filename_full
= g_malloc(PATH_MAX
);
6706 backing_file_full
= g_malloc(PATH_MAX
);
6708 is_protocol
= path_has_protocol(backing_file
);
6711 * Being largely a legacy function, skip any filters here
6712 * (because filters do not have normal filenames, so they cannot
6713 * match anyway; and allowing json:{} filenames is a bit out of
6716 for (curr_bs
= bdrv_skip_filters(bs
);
6717 bdrv_cow_child(curr_bs
) != NULL
;
6720 bs_below
= bdrv_backing_chain_next(curr_bs
);
6722 if (bdrv_backing_overridden(curr_bs
)) {
6724 * If the backing file was overridden, we can only compare
6725 * directly against the backing node's filename.
6728 if (!filenames_refreshed
) {
6730 * This will automatically refresh all of the
6731 * filenames in the rest of the backing chain, so we
6732 * only need to do this once.
6734 bdrv_refresh_filename(bs_below
);
6735 filenames_refreshed
= true;
6738 if (strcmp(backing_file
, bs_below
->filename
) == 0) {
6742 } else if (is_protocol
|| path_has_protocol(curr_bs
->backing_file
)) {
6744 * If either of the filename paths is actually a protocol, then
6745 * compare unmodified paths; otherwise make paths relative.
6747 char *backing_file_full_ret
;
6749 if (strcmp(backing_file
, curr_bs
->backing_file
) == 0) {
6753 /* Also check against the full backing filename for the image */
6754 backing_file_full_ret
= bdrv_get_full_backing_filename(curr_bs
,
6756 if (backing_file_full_ret
) {
6757 bool equal
= strcmp(backing_file
, backing_file_full_ret
) == 0;
6758 g_free(backing_file_full_ret
);
6765 /* If not an absolute filename path, make it relative to the current
6766 * image's filename path */
6767 filename_tmp
= bdrv_make_absolute_filename(curr_bs
, backing_file
,
6769 /* We are going to compare canonicalized absolute pathnames */
6770 if (!filename_tmp
|| !realpath(filename_tmp
, filename_full
)) {
6771 g_free(filename_tmp
);
6774 g_free(filename_tmp
);
6776 /* We need to make sure the backing filename we are comparing against
6777 * is relative to the current image filename (or absolute) */
6778 filename_tmp
= bdrv_get_full_backing_filename(curr_bs
, NULL
);
6779 if (!filename_tmp
|| !realpath(filename_tmp
, backing_file_full
)) {
6780 g_free(filename_tmp
);
6783 g_free(filename_tmp
);
6785 if (strcmp(backing_file_full
, filename_full
) == 0) {
6792 g_free(filename_full
);
6793 g_free(backing_file_full
);
6797 void bdrv_init(void)
6799 #ifdef CONFIG_BDRV_WHITELIST_TOOLS
6800 use_bdrv_whitelist
= 1;
6802 module_call_init(MODULE_INIT_BLOCK
);
6805 void bdrv_init_with_whitelist(void)
6807 use_bdrv_whitelist
= 1;
6811 int bdrv_activate(BlockDriverState
*bs
, Error
**errp
)
6813 BdrvChild
*child
, *parent
;
6814 Error
*local_err
= NULL
;
6816 BdrvDirtyBitmap
*bm
;
6818 GLOBAL_STATE_CODE();
6819 GRAPH_RDLOCK_GUARD_MAINLOOP();
6825 QLIST_FOREACH(child
, &bs
->children
, next
) {
6826 bdrv_activate(child
->bs
, &local_err
);
6828 error_propagate(errp
, local_err
);
6834 * Update permissions, they may differ for inactive nodes.
6836 * Note that the required permissions of inactive images are always a
6837 * subset of the permissions required after activating the image. This
6838 * allows us to just get the permissions upfront without restricting
6839 * bdrv_co_invalidate_cache().
6841 * It also means that in error cases, we don't have to try and revert to
6842 * the old permissions (which is an operation that could fail, too). We can
6843 * just keep the extended permissions for the next time that an activation
6844 * of the image is tried.
6846 if (bs
->open_flags
& BDRV_O_INACTIVE
) {
6847 bs
->open_flags
&= ~BDRV_O_INACTIVE
;
6848 ret
= bdrv_refresh_perms(bs
, NULL
, errp
);
6850 bs
->open_flags
|= BDRV_O_INACTIVE
;
6854 ret
= bdrv_invalidate_cache(bs
, errp
);
6856 bs
->open_flags
|= BDRV_O_INACTIVE
;
6860 FOR_EACH_DIRTY_BITMAP(bs
, bm
) {
6861 bdrv_dirty_bitmap_skip_store(bm
, false);
6864 ret
= bdrv_refresh_total_sectors(bs
, bs
->total_sectors
);
6866 bs
->open_flags
|= BDRV_O_INACTIVE
;
6867 error_setg_errno(errp
, -ret
, "Could not refresh total sector count");
6872 QLIST_FOREACH(parent
, &bs
->parents
, next_parent
) {
6873 if (parent
->klass
->activate
) {
6874 parent
->klass
->activate(parent
, &local_err
);
6876 bs
->open_flags
|= BDRV_O_INACTIVE
;
6877 error_propagate(errp
, local_err
);
6886 int coroutine_fn
bdrv_co_invalidate_cache(BlockDriverState
*bs
, Error
**errp
)
6888 Error
*local_err
= NULL
;
6891 assert(!(bs
->open_flags
& BDRV_O_INACTIVE
));
6892 assert_bdrv_graph_readable();
6894 if (bs
->drv
->bdrv_co_invalidate_cache
) {
6895 bs
->drv
->bdrv_co_invalidate_cache(bs
, &local_err
);
6897 error_propagate(errp
, local_err
);
6905 void bdrv_activate_all(Error
**errp
)
6907 BlockDriverState
*bs
;
6908 BdrvNextIterator it
;
6910 GLOBAL_STATE_CODE();
6911 GRAPH_RDLOCK_GUARD_MAINLOOP();
6913 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
6916 ret
= bdrv_activate(bs
, errp
);
6918 bdrv_next_cleanup(&it
);
6924 static bool GRAPH_RDLOCK
6925 bdrv_has_bds_parent(BlockDriverState
*bs
, bool only_active
)
6928 GLOBAL_STATE_CODE();
6930 QLIST_FOREACH(parent
, &bs
->parents
, next_parent
) {
6931 if (parent
->klass
->parent_is_bds
) {
6932 BlockDriverState
*parent_bs
= parent
->opaque
;
6933 if (!only_active
|| !(parent_bs
->open_flags
& BDRV_O_INACTIVE
)) {
6942 static int GRAPH_RDLOCK
bdrv_inactivate_recurse(BlockDriverState
*bs
)
6944 BdrvChild
*child
, *parent
;
6946 uint64_t cumulative_perms
, cumulative_shared_perms
;
6948 GLOBAL_STATE_CODE();
6954 /* Make sure that we don't inactivate a child before its parent.
6955 * It will be covered by recursion from the yet active parent. */
6956 if (bdrv_has_bds_parent(bs
, true)) {
6960 assert(!(bs
->open_flags
& BDRV_O_INACTIVE
));
6962 /* Inactivate this node */
6963 if (bs
->drv
->bdrv_inactivate
) {
6964 ret
= bs
->drv
->bdrv_inactivate(bs
);
6970 QLIST_FOREACH(parent
, &bs
->parents
, next_parent
) {
6971 if (parent
->klass
->inactivate
) {
6972 ret
= parent
->klass
->inactivate(parent
);
6979 bdrv_get_cumulative_perm(bs
, &cumulative_perms
,
6980 &cumulative_shared_perms
);
6981 if (cumulative_perms
& (BLK_PERM_WRITE
| BLK_PERM_WRITE_UNCHANGED
)) {
6982 /* Our inactive parents still need write access. Inactivation failed. */
6986 bs
->open_flags
|= BDRV_O_INACTIVE
;
6989 * Update permissions, they may differ for inactive nodes.
6990 * We only tried to loosen restrictions, so errors are not fatal, ignore
6993 bdrv_refresh_perms(bs
, NULL
, NULL
);
6995 /* Recursively inactivate children */
6996 QLIST_FOREACH(child
, &bs
->children
, next
) {
6997 ret
= bdrv_inactivate_recurse(child
->bs
);
7006 int bdrv_inactivate_all(void)
7008 BlockDriverState
*bs
= NULL
;
7009 BdrvNextIterator it
;
7012 GLOBAL_STATE_CODE();
7013 GRAPH_RDLOCK_GUARD_MAINLOOP();
7015 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
7016 /* Nodes with BDS parents are covered by recursion from the last
7017 * parent that gets inactivated. Don't inactivate them a second
7018 * time if that has already happened. */
7019 if (bdrv_has_bds_parent(bs
, false)) {
7022 ret
= bdrv_inactivate_recurse(bs
);
7024 bdrv_next_cleanup(&it
);
7032 /**************************************************************/
7033 /* removable device support */
7036 * Return TRUE if the media is present
7038 bool coroutine_fn
bdrv_co_is_inserted(BlockDriverState
*bs
)
7040 BlockDriver
*drv
= bs
->drv
;
7043 assert_bdrv_graph_readable();
7048 if (drv
->bdrv_co_is_inserted
) {
7049 return drv
->bdrv_co_is_inserted(bs
);
7051 QLIST_FOREACH(child
, &bs
->children
, next
) {
7052 if (!bdrv_co_is_inserted(child
->bs
)) {
7060 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
7062 void coroutine_fn
bdrv_co_eject(BlockDriverState
*bs
, bool eject_flag
)
7064 BlockDriver
*drv
= bs
->drv
;
7066 assert_bdrv_graph_readable();
7068 if (drv
&& drv
->bdrv_co_eject
) {
7069 drv
->bdrv_co_eject(bs
, eject_flag
);
7074 * Lock or unlock the media (if it is locked, the user won't be able
7075 * to eject it manually).
7077 void coroutine_fn
bdrv_co_lock_medium(BlockDriverState
*bs
, bool locked
)
7079 BlockDriver
*drv
= bs
->drv
;
7081 assert_bdrv_graph_readable();
7082 trace_bdrv_lock_medium(bs
, locked
);
7084 if (drv
&& drv
->bdrv_co_lock_medium
) {
7085 drv
->bdrv_co_lock_medium(bs
, locked
);
7089 /* Get a reference to bs */
7090 void bdrv_ref(BlockDriverState
*bs
)
7092 GLOBAL_STATE_CODE();
7096 /* Release a previously grabbed reference to bs.
7097 * If after releasing, reference count is zero, the BlockDriverState is
7099 void bdrv_unref(BlockDriverState
*bs
)
7101 GLOBAL_STATE_CODE();
7105 assert(bs
->refcnt
> 0);
7106 if (--bs
->refcnt
== 0) {
7111 static void bdrv_schedule_unref_bh(void *opaque
)
7113 BlockDriverState
*bs
= opaque
;
7119 * Release a BlockDriverState reference while holding the graph write lock.
7121 * Calling bdrv_unref() directly is forbidden while holding the graph lock
7122 * because bdrv_close() both involves polling and taking the graph lock
7123 * internally. bdrv_schedule_unref() instead delays decreasing the refcount and
7124 * possibly closing @bs until the graph lock is released.
7126 void bdrv_schedule_unref(BlockDriverState
*bs
)
7131 aio_bh_schedule_oneshot(qemu_get_aio_context(), bdrv_schedule_unref_bh
, bs
);
7134 struct BdrvOpBlocker
{
7136 QLIST_ENTRY(BdrvOpBlocker
) list
;
7139 bool bdrv_op_is_blocked(BlockDriverState
*bs
, BlockOpType op
, Error
**errp
)
7141 BdrvOpBlocker
*blocker
;
7142 GLOBAL_STATE_CODE();
7144 assert((int) op
>= 0 && op
< BLOCK_OP_TYPE_MAX
);
7145 if (!QLIST_EMPTY(&bs
->op_blockers
[op
])) {
7146 blocker
= QLIST_FIRST(&bs
->op_blockers
[op
]);
7147 error_propagate_prepend(errp
, error_copy(blocker
->reason
),
7148 "Node '%s' is busy: ",
7149 bdrv_get_device_or_node_name(bs
));
7155 void bdrv_op_block(BlockDriverState
*bs
, BlockOpType op
, Error
*reason
)
7157 BdrvOpBlocker
*blocker
;
7158 GLOBAL_STATE_CODE();
7159 assert((int) op
>= 0 && op
< BLOCK_OP_TYPE_MAX
);
7161 blocker
= g_new0(BdrvOpBlocker
, 1);
7162 blocker
->reason
= reason
;
7163 QLIST_INSERT_HEAD(&bs
->op_blockers
[op
], blocker
, list
);
7166 void bdrv_op_unblock(BlockDriverState
*bs
, BlockOpType op
, Error
*reason
)
7168 BdrvOpBlocker
*blocker
, *next
;
7169 GLOBAL_STATE_CODE();
7170 assert((int) op
>= 0 && op
< BLOCK_OP_TYPE_MAX
);
7171 QLIST_FOREACH_SAFE(blocker
, &bs
->op_blockers
[op
], list
, next
) {
7172 if (blocker
->reason
== reason
) {
7173 QLIST_REMOVE(blocker
, list
);
7179 void bdrv_op_block_all(BlockDriverState
*bs
, Error
*reason
)
7182 GLOBAL_STATE_CODE();
7183 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
7184 bdrv_op_block(bs
, i
, reason
);
7188 void bdrv_op_unblock_all(BlockDriverState
*bs
, Error
*reason
)
7191 GLOBAL_STATE_CODE();
7192 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
7193 bdrv_op_unblock(bs
, i
, reason
);
7197 bool bdrv_op_blocker_is_empty(BlockDriverState
*bs
)
7200 GLOBAL_STATE_CODE();
7201 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
7202 if (!QLIST_EMPTY(&bs
->op_blockers
[i
])) {
7210 * Must not be called while holding the lock of an AioContext other than the
7213 void bdrv_img_create(const char *filename
, const char *fmt
,
7214 const char *base_filename
, const char *base_fmt
,
7215 char *options
, uint64_t img_size
, int flags
, bool quiet
,
7218 QemuOptsList
*create_opts
= NULL
;
7219 QemuOpts
*opts
= NULL
;
7220 const char *backing_fmt
, *backing_file
;
7222 BlockDriver
*drv
, *proto_drv
;
7223 Error
*local_err
= NULL
;
7226 GLOBAL_STATE_CODE();
7228 /* Find driver and parse its options */
7229 drv
= bdrv_find_format(fmt
);
7231 error_setg(errp
, "Unknown file format '%s'", fmt
);
7235 proto_drv
= bdrv_find_protocol(filename
, true, errp
);
7240 if (!drv
->create_opts
) {
7241 error_setg(errp
, "Format driver '%s' does not support image creation",
7246 if (!proto_drv
->create_opts
) {
7247 error_setg(errp
, "Protocol driver '%s' does not support image creation",
7248 proto_drv
->format_name
);
7252 /* Create parameter list */
7253 create_opts
= qemu_opts_append(create_opts
, drv
->create_opts
);
7254 create_opts
= qemu_opts_append(create_opts
, proto_drv
->create_opts
);
7256 opts
= qemu_opts_create(create_opts
, NULL
, 0, &error_abort
);
7258 /* Parse -o options */
7260 if (!qemu_opts_do_parse(opts
, options
, NULL
, errp
)) {
7265 if (!qemu_opt_get(opts
, BLOCK_OPT_SIZE
)) {
7266 qemu_opt_set_number(opts
, BLOCK_OPT_SIZE
, img_size
, &error_abort
);
7267 } else if (img_size
!= UINT64_C(-1)) {
7268 error_setg(errp
, "The image size must be specified only once");
7272 if (base_filename
) {
7273 if (!qemu_opt_set(opts
, BLOCK_OPT_BACKING_FILE
, base_filename
,
7275 error_setg(errp
, "Backing file not supported for file format '%s'",
7282 if (!qemu_opt_set(opts
, BLOCK_OPT_BACKING_FMT
, base_fmt
, NULL
)) {
7283 error_setg(errp
, "Backing file format not supported for file "
7284 "format '%s'", fmt
);
7289 backing_file
= qemu_opt_get(opts
, BLOCK_OPT_BACKING_FILE
);
7291 if (!strcmp(filename
, backing_file
)) {
7292 error_setg(errp
, "Error: Trying to create an image with the "
7293 "same filename as the backing file");
7296 if (backing_file
[0] == '\0') {
7297 error_setg(errp
, "Expected backing file name, got empty string");
7302 backing_fmt
= qemu_opt_get(opts
, BLOCK_OPT_BACKING_FMT
);
7304 /* The size for the image must always be specified, unless we have a backing
7305 * file and we have not been forbidden from opening it. */
7306 size
= qemu_opt_get_size(opts
, BLOCK_OPT_SIZE
, img_size
);
7307 if (backing_file
&& !(flags
& BDRV_O_NO_BACKING
)) {
7308 BlockDriverState
*bs
;
7311 QDict
*backing_options
= NULL
;
7314 bdrv_get_full_backing_filename_from_filename(filename
, backing_file
,
7319 assert(full_backing
);
7322 * No need to do I/O here, which allows us to open encrypted
7323 * backing images without needing the secret
7326 back_flags
&= ~(BDRV_O_RDWR
| BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
);
7327 back_flags
|= BDRV_O_NO_IO
;
7329 backing_options
= qdict_new();
7331 qdict_put_str(backing_options
, "driver", backing_fmt
);
7333 qdict_put_bool(backing_options
, BDRV_OPT_FORCE_SHARE
, true);
7335 bs
= bdrv_open(full_backing
, NULL
, backing_options
, back_flags
,
7337 g_free(full_backing
);
7339 error_append_hint(&local_err
, "Could not open backing image.\n");
7343 error_setg(&local_err
,
7344 "Backing file specified without backing format");
7345 error_append_hint(&local_err
, "Detected format of %s.\n",
7346 bs
->drv
->format_name
);
7350 /* Opened BS, have no size */
7351 size
= bdrv_getlength(bs
);
7353 error_setg_errno(errp
, -size
, "Could not get size of '%s'",
7358 qemu_opt_set_number(opts
, BLOCK_OPT_SIZE
, size
, &error_abort
);
7362 /* (backing_file && !(flags & BDRV_O_NO_BACKING)) */
7363 } else if (backing_file
&& !backing_fmt
) {
7364 error_setg(&local_err
,
7365 "Backing file specified without backing format");
7370 error_setg(errp
, "Image creation needs a size parameter");
7375 printf("Formatting '%s', fmt=%s ", filename
, fmt
);
7376 qemu_opts_print(opts
, " ");
7381 ret
= bdrv_create(drv
, filename
, opts
, &local_err
);
7383 if (ret
== -EFBIG
) {
7384 /* This is generally a better message than whatever the driver would
7385 * deliver (especially because of the cluster_size_hint), since that
7386 * is most probably not much different from "image too large". */
7387 const char *cluster_size_hint
= "";
7388 if (qemu_opt_get_size(opts
, BLOCK_OPT_CLUSTER_SIZE
, 0)) {
7389 cluster_size_hint
= " (try using a larger cluster size)";
7391 error_setg(errp
, "The image size is too large for file format '%s'"
7392 "%s", fmt
, cluster_size_hint
);
7393 error_free(local_err
);
7398 qemu_opts_del(opts
);
7399 qemu_opts_free(create_opts
);
7400 error_propagate(errp
, local_err
);
7403 AioContext
*bdrv_get_aio_context(BlockDriverState
*bs
)
7406 return bs
? bs
->aio_context
: qemu_get_aio_context();
7409 AioContext
*coroutine_fn
bdrv_co_enter(BlockDriverState
*bs
)
7411 Coroutine
*self
= qemu_coroutine_self();
7412 AioContext
*old_ctx
= qemu_coroutine_get_aio_context(self
);
7413 AioContext
*new_ctx
;
7417 * Increase bs->in_flight to ensure that this operation is completed before
7418 * moving the node to a different AioContext. Read new_ctx only afterwards.
7420 bdrv_inc_in_flight(bs
);
7422 new_ctx
= bdrv_get_aio_context(bs
);
7423 aio_co_reschedule_self(new_ctx
);
7427 void coroutine_fn
bdrv_co_leave(BlockDriverState
*bs
, AioContext
*old_ctx
)
7430 aio_co_reschedule_self(old_ctx
);
7431 bdrv_dec_in_flight(bs
);
7434 void coroutine_fn
bdrv_co_lock(BlockDriverState
*bs
)
7436 /* TODO removed in next patch */
7439 void coroutine_fn
bdrv_co_unlock(BlockDriverState
*bs
)
7441 /* TODO removed in next patch */
7444 static void bdrv_do_remove_aio_context_notifier(BdrvAioNotifier
*ban
)
7446 GLOBAL_STATE_CODE();
7447 QLIST_REMOVE(ban
, list
);
7451 static void bdrv_detach_aio_context(BlockDriverState
*bs
)
7453 BdrvAioNotifier
*baf
, *baf_tmp
;
7455 assert(!bs
->walking_aio_notifiers
);
7456 GLOBAL_STATE_CODE();
7457 bs
->walking_aio_notifiers
= true;
7458 QLIST_FOREACH_SAFE(baf
, &bs
->aio_notifiers
, list
, baf_tmp
) {
7460 bdrv_do_remove_aio_context_notifier(baf
);
7462 baf
->detach_aio_context(baf
->opaque
);
7465 /* Never mind iterating again to check for ->deleted. bdrv_close() will
7466 * remove remaining aio notifiers if we aren't called again.
7468 bs
->walking_aio_notifiers
= false;
7470 if (bs
->drv
&& bs
->drv
->bdrv_detach_aio_context
) {
7471 bs
->drv
->bdrv_detach_aio_context(bs
);
7474 bs
->aio_context
= NULL
;
7477 static void bdrv_attach_aio_context(BlockDriverState
*bs
,
7478 AioContext
*new_context
)
7480 BdrvAioNotifier
*ban
, *ban_tmp
;
7481 GLOBAL_STATE_CODE();
7483 bs
->aio_context
= new_context
;
7485 if (bs
->drv
&& bs
->drv
->bdrv_attach_aio_context
) {
7486 bs
->drv
->bdrv_attach_aio_context(bs
, new_context
);
7489 assert(!bs
->walking_aio_notifiers
);
7490 bs
->walking_aio_notifiers
= true;
7491 QLIST_FOREACH_SAFE(ban
, &bs
->aio_notifiers
, list
, ban_tmp
) {
7493 bdrv_do_remove_aio_context_notifier(ban
);
7495 ban
->attached_aio_context(new_context
, ban
->opaque
);
7498 bs
->walking_aio_notifiers
= false;
7501 typedef struct BdrvStateSetAioContext
{
7502 AioContext
*new_ctx
;
7503 BlockDriverState
*bs
;
7504 } BdrvStateSetAioContext
;
7506 static bool bdrv_parent_change_aio_context(BdrvChild
*c
, AioContext
*ctx
,
7507 GHashTable
*visited
,
7511 GLOBAL_STATE_CODE();
7512 if (g_hash_table_contains(visited
, c
)) {
7515 g_hash_table_add(visited
, c
);
7518 * A BdrvChildClass that doesn't handle AioContext changes cannot
7519 * tolerate any AioContext changes
7521 if (!c
->klass
->change_aio_ctx
) {
7522 char *user
= bdrv_child_user_desc(c
);
7523 error_setg(errp
, "Changing iothreads is not supported by %s", user
);
7527 if (!c
->klass
->change_aio_ctx(c
, ctx
, visited
, tran
, errp
)) {
7528 assert(!errp
|| *errp
);
7534 bool bdrv_child_change_aio_context(BdrvChild
*c
, AioContext
*ctx
,
7535 GHashTable
*visited
, Transaction
*tran
,
7538 GLOBAL_STATE_CODE();
7539 if (g_hash_table_contains(visited
, c
)) {
7542 g_hash_table_add(visited
, c
);
7543 return bdrv_change_aio_context(c
->bs
, ctx
, visited
, tran
, errp
);
7546 static void bdrv_set_aio_context_clean(void *opaque
)
7548 BdrvStateSetAioContext
*state
= (BdrvStateSetAioContext
*) opaque
;
7549 BlockDriverState
*bs
= (BlockDriverState
*) state
->bs
;
7551 /* Paired with bdrv_drained_begin in bdrv_change_aio_context() */
7552 bdrv_drained_end(bs
);
7557 static void bdrv_set_aio_context_commit(void *opaque
)
7559 BdrvStateSetAioContext
*state
= (BdrvStateSetAioContext
*) opaque
;
7560 BlockDriverState
*bs
= (BlockDriverState
*) state
->bs
;
7561 AioContext
*new_context
= state
->new_ctx
;
7563 bdrv_detach_aio_context(bs
);
7564 bdrv_attach_aio_context(bs
, new_context
);
7567 static TransactionActionDrv set_aio_context
= {
7568 .commit
= bdrv_set_aio_context_commit
,
7569 .clean
= bdrv_set_aio_context_clean
,
7573 * Changes the AioContext used for fd handlers, timers, and BHs by this
7574 * BlockDriverState and all its children and parents.
7576 * Must be called from the main AioContext.
7578 * The caller must own the AioContext lock for the old AioContext of bs, but it
7579 * must not own the AioContext lock for new_context (unless new_context is the
7580 * same as the current context of bs).
7582 * @visited will accumulate all visited BdrvChild objects. The caller is
7583 * responsible for freeing the list afterwards.
7585 static bool bdrv_change_aio_context(BlockDriverState
*bs
, AioContext
*ctx
,
7586 GHashTable
*visited
, Transaction
*tran
,
7590 BdrvStateSetAioContext
*state
;
7592 GLOBAL_STATE_CODE();
7594 if (bdrv_get_aio_context(bs
) == ctx
) {
7598 bdrv_graph_rdlock_main_loop();
7599 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
7600 if (!bdrv_parent_change_aio_context(c
, ctx
, visited
, tran
, errp
)) {
7601 bdrv_graph_rdunlock_main_loop();
7606 QLIST_FOREACH(c
, &bs
->children
, next
) {
7607 if (!bdrv_child_change_aio_context(c
, ctx
, visited
, tran
, errp
)) {
7608 bdrv_graph_rdunlock_main_loop();
7612 bdrv_graph_rdunlock_main_loop();
7614 state
= g_new(BdrvStateSetAioContext
, 1);
7615 *state
= (BdrvStateSetAioContext
) {
7620 /* Paired with bdrv_drained_end in bdrv_set_aio_context_clean() */
7621 bdrv_drained_begin(bs
);
7623 tran_add(tran
, &set_aio_context
, state
);
7629 * Change bs's and recursively all of its parents' and children's AioContext
7630 * to the given new context, returning an error if that isn't possible.
7632 * If ignore_child is not NULL, that child (and its subgraph) will not
7635 * This function still requires the caller to take the bs current
7636 * AioContext lock, otherwise draining will fail since AIO_WAIT_WHILE
7637 * assumes the lock is always held if bs is in another AioContext.
7638 * For the same reason, it temporarily also holds the new AioContext, since
7639 * bdrv_drained_end calls BDRV_POLL_WHILE that assumes the lock is taken too.
7640 * Therefore the new AioContext lock must not be taken by the caller.
7642 int bdrv_try_change_aio_context(BlockDriverState
*bs
, AioContext
*ctx
,
7643 BdrvChild
*ignore_child
, Error
**errp
)
7646 GHashTable
*visited
;
7648 GLOBAL_STATE_CODE();
7651 * Recursion phase: go through all nodes of the graph.
7652 * Take care of checking that all nodes support changing AioContext
7653 * and drain them, building a linear list of callbacks to run if everything
7654 * is successful (the transaction itself).
7657 visited
= g_hash_table_new(NULL
, NULL
);
7659 g_hash_table_add(visited
, ignore_child
);
7661 ret
= bdrv_change_aio_context(bs
, ctx
, visited
, tran
, errp
);
7662 g_hash_table_destroy(visited
);
7665 * Linear phase: go through all callbacks collected in the transaction.
7666 * Run all callbacks collected in the recursion to switch all nodes
7667 * AioContext lock (transaction commit), or undo all changes done in the
7668 * recursion (transaction abort).
7672 /* Just run clean() callbacks. No AioContext changed. */
7681 void bdrv_add_aio_context_notifier(BlockDriverState
*bs
,
7682 void (*attached_aio_context
)(AioContext
*new_context
, void *opaque
),
7683 void (*detach_aio_context
)(void *opaque
), void *opaque
)
7685 BdrvAioNotifier
*ban
= g_new(BdrvAioNotifier
, 1);
7686 *ban
= (BdrvAioNotifier
){
7687 .attached_aio_context
= attached_aio_context
,
7688 .detach_aio_context
= detach_aio_context
,
7691 GLOBAL_STATE_CODE();
7693 QLIST_INSERT_HEAD(&bs
->aio_notifiers
, ban
, list
);
7696 void bdrv_remove_aio_context_notifier(BlockDriverState
*bs
,
7697 void (*attached_aio_context
)(AioContext
*,
7699 void (*detach_aio_context
)(void *),
7702 BdrvAioNotifier
*ban
, *ban_next
;
7703 GLOBAL_STATE_CODE();
7705 QLIST_FOREACH_SAFE(ban
, &bs
->aio_notifiers
, list
, ban_next
) {
7706 if (ban
->attached_aio_context
== attached_aio_context
&&
7707 ban
->detach_aio_context
== detach_aio_context
&&
7708 ban
->opaque
== opaque
&&
7709 ban
->deleted
== false)
7711 if (bs
->walking_aio_notifiers
) {
7712 ban
->deleted
= true;
7714 bdrv_do_remove_aio_context_notifier(ban
);
7723 int bdrv_amend_options(BlockDriverState
*bs
, QemuOpts
*opts
,
7724 BlockDriverAmendStatusCB
*status_cb
, void *cb_opaque
,
7728 GLOBAL_STATE_CODE();
7730 error_setg(errp
, "Node is ejected");
7733 if (!bs
->drv
->bdrv_amend_options
) {
7734 error_setg(errp
, "Block driver '%s' does not support option amendment",
7735 bs
->drv
->format_name
);
7738 return bs
->drv
->bdrv_amend_options(bs
, opts
, status_cb
,
7739 cb_opaque
, force
, errp
);
7743 * This function checks whether the given @to_replace is allowed to be
7744 * replaced by a node that always shows the same data as @bs. This is
7745 * used for example to verify whether the mirror job can replace
7746 * @to_replace by the target mirrored from @bs.
7747 * To be replaceable, @bs and @to_replace may either be guaranteed to
7748 * always show the same data (because they are only connected through
7749 * filters), or some driver may allow replacing one of its children
7750 * because it can guarantee that this child's data is not visible at
7751 * all (for example, for dissenting quorum children that have no other
7754 bool bdrv_recurse_can_replace(BlockDriverState
*bs
,
7755 BlockDriverState
*to_replace
)
7757 BlockDriverState
*filtered
;
7759 GLOBAL_STATE_CODE();
7761 if (!bs
|| !bs
->drv
) {
7765 if (bs
== to_replace
) {
7769 /* See what the driver can do */
7770 if (bs
->drv
->bdrv_recurse_can_replace
) {
7771 return bs
->drv
->bdrv_recurse_can_replace(bs
, to_replace
);
7774 /* For filters without an own implementation, we can recurse on our own */
7775 filtered
= bdrv_filter_bs(bs
);
7777 return bdrv_recurse_can_replace(filtered
, to_replace
);
7785 * Check whether the given @node_name can be replaced by a node that
7786 * has the same data as @parent_bs. If so, return @node_name's BDS;
7789 * @node_name must be a (recursive) *child of @parent_bs (or this
7790 * function will return NULL).
7792 * The result (whether the node can be replaced or not) is only valid
7793 * for as long as no graph or permission changes occur.
7795 BlockDriverState
*check_to_replace_node(BlockDriverState
*parent_bs
,
7796 const char *node_name
, Error
**errp
)
7798 BlockDriverState
*to_replace_bs
= bdrv_find_node(node_name
);
7800 GLOBAL_STATE_CODE();
7802 if (!to_replace_bs
) {
7803 error_setg(errp
, "Failed to find node with node-name='%s'", node_name
);
7807 if (bdrv_op_is_blocked(to_replace_bs
, BLOCK_OP_TYPE_REPLACE
, errp
)) {
7811 /* We don't want arbitrary node of the BDS chain to be replaced only the top
7812 * most non filter in order to prevent data corruption.
7813 * Another benefit is that this tests exclude backing files which are
7814 * blocked by the backing blockers.
7816 if (!bdrv_recurse_can_replace(parent_bs
, to_replace_bs
)) {
7817 error_setg(errp
, "Cannot replace '%s' by a node mirrored from '%s', "
7818 "because it cannot be guaranteed that doing so would not "
7819 "lead to an abrupt change of visible data",
7820 node_name
, parent_bs
->node_name
);
7824 return to_replace_bs
;
7828 * Iterates through the list of runtime option keys that are said to
7829 * be "strong" for a BDS. An option is called "strong" if it changes
7830 * a BDS's data. For example, the null block driver's "size" and
7831 * "read-zeroes" options are strong, but its "latency-ns" option is
7834 * If a key returned by this function ends with a dot, all options
7835 * starting with that prefix are strong.
7837 static const char *const *strong_options(BlockDriverState
*bs
,
7838 const char *const *curopt
)
7840 static const char *const global_options
[] = {
7841 "driver", "filename", NULL
7845 return &global_options
[0];
7849 if (curopt
== &global_options
[ARRAY_SIZE(global_options
) - 1] && bs
->drv
) {
7850 curopt
= bs
->drv
->strong_runtime_opts
;
7853 return (curopt
&& *curopt
) ? curopt
: NULL
;
7857 * Copies all strong runtime options from bs->options to the given
7858 * QDict. The set of strong option keys is determined by invoking
7861 * Returns true iff any strong option was present in bs->options (and
7862 * thus copied to the target QDict) with the exception of "filename"
7863 * and "driver". The caller is expected to use this value to decide
7864 * whether the existence of strong options prevents the generation of
7867 static bool append_strong_runtime_options(QDict
*d
, BlockDriverState
*bs
)
7869 bool found_any
= false;
7870 const char *const *option_name
= NULL
;
7876 while ((option_name
= strong_options(bs
, option_name
))) {
7877 bool option_given
= false;
7879 assert(strlen(*option_name
) > 0);
7880 if ((*option_name
)[strlen(*option_name
) - 1] != '.') {
7881 QObject
*entry
= qdict_get(bs
->options
, *option_name
);
7886 qdict_put_obj(d
, *option_name
, qobject_ref(entry
));
7887 option_given
= true;
7889 const QDictEntry
*entry
;
7890 for (entry
= qdict_first(bs
->options
); entry
;
7891 entry
= qdict_next(bs
->options
, entry
))
7893 if (strstart(qdict_entry_key(entry
), *option_name
, NULL
)) {
7894 qdict_put_obj(d
, qdict_entry_key(entry
),
7895 qobject_ref(qdict_entry_value(entry
)));
7896 option_given
= true;
7901 /* While "driver" and "filename" need to be included in a JSON filename,
7902 * their existence does not prohibit generation of a plain filename. */
7903 if (!found_any
&& option_given
&&
7904 strcmp(*option_name
, "driver") && strcmp(*option_name
, "filename"))
7910 if (!qdict_haskey(d
, "driver")) {
7911 /* Drivers created with bdrv_new_open_driver() may not have a
7912 * @driver option. Add it here. */
7913 qdict_put_str(d
, "driver", bs
->drv
->format_name
);
7919 /* Note: This function may return false positives; it may return true
7920 * even if opening the backing file specified by bs's image header
7921 * would result in exactly bs->backing. */
7922 static bool GRAPH_RDLOCK
bdrv_backing_overridden(BlockDriverState
*bs
)
7924 GLOBAL_STATE_CODE();
7926 return strcmp(bs
->auto_backing_file
,
7927 bs
->backing
->bs
->filename
);
7929 /* No backing BDS, so if the image header reports any backing
7930 * file, it must have been suppressed */
7931 return bs
->auto_backing_file
[0] != '\0';
7935 /* Updates the following BDS fields:
7936 * - exact_filename: A filename which may be used for opening a block device
7937 * which (mostly) equals the given BDS (even without any
7938 * other options; so reading and writing must return the same
7939 * results, but caching etc. may be different)
7940 * - full_open_options: Options which, when given when opening a block device
7941 * (without a filename), result in a BDS (mostly)
7942 * equalling the given one
7943 * - filename: If exact_filename is set, it is copied here. Otherwise,
7944 * full_open_options is converted to a JSON object, prefixed with
7945 * "json:" (for use through the JSON pseudo protocol) and put here.
7947 void bdrv_refresh_filename(BlockDriverState
*bs
)
7949 BlockDriver
*drv
= bs
->drv
;
7951 BlockDriverState
*primary_child_bs
;
7953 bool backing_overridden
;
7954 bool generate_json_filename
; /* Whether our default implementation should
7955 fill exact_filename (false) or not (true) */
7957 GLOBAL_STATE_CODE();
7963 /* This BDS's file name may depend on any of its children's file names, so
7964 * refresh those first */
7965 QLIST_FOREACH(child
, &bs
->children
, next
) {
7966 bdrv_refresh_filename(child
->bs
);
7970 /* For implicit nodes, just copy everything from the single child */
7971 child
= QLIST_FIRST(&bs
->children
);
7972 assert(QLIST_NEXT(child
, next
) == NULL
);
7974 pstrcpy(bs
->exact_filename
, sizeof(bs
->exact_filename
),
7975 child
->bs
->exact_filename
);
7976 pstrcpy(bs
->filename
, sizeof(bs
->filename
), child
->bs
->filename
);
7978 qobject_unref(bs
->full_open_options
);
7979 bs
->full_open_options
= qobject_ref(child
->bs
->full_open_options
);
7984 backing_overridden
= bdrv_backing_overridden(bs
);
7986 if (bs
->open_flags
& BDRV_O_NO_IO
) {
7987 /* Without I/O, the backing file does not change anything.
7988 * Therefore, in such a case (primarily qemu-img), we can
7989 * pretend the backing file has not been overridden even if
7990 * it technically has been. */
7991 backing_overridden
= false;
7994 /* Gather the options QDict */
7996 generate_json_filename
= append_strong_runtime_options(opts
, bs
);
7997 generate_json_filename
|= backing_overridden
;
7999 if (drv
->bdrv_gather_child_options
) {
8000 /* Some block drivers may not want to present all of their children's
8001 * options, or name them differently from BdrvChild.name */
8002 drv
->bdrv_gather_child_options(bs
, opts
, backing_overridden
);
8004 QLIST_FOREACH(child
, &bs
->children
, next
) {
8005 if (child
== bs
->backing
&& !backing_overridden
) {
8006 /* We can skip the backing BDS if it has not been overridden */
8010 qdict_put(opts
, child
->name
,
8011 qobject_ref(child
->bs
->full_open_options
));
8014 if (backing_overridden
&& !bs
->backing
) {
8015 /* Force no backing file */
8016 qdict_put_null(opts
, "backing");
8020 qobject_unref(bs
->full_open_options
);
8021 bs
->full_open_options
= opts
;
8023 primary_child_bs
= bdrv_primary_bs(bs
);
8025 if (drv
->bdrv_refresh_filename
) {
8026 /* Obsolete information is of no use here, so drop the old file name
8027 * information before refreshing it */
8028 bs
->exact_filename
[0] = '\0';
8030 drv
->bdrv_refresh_filename(bs
);
8031 } else if (primary_child_bs
) {
8033 * Try to reconstruct valid information from the underlying
8034 * file -- this only works for format nodes (filter nodes
8035 * cannot be probed and as such must be selected by the user
8036 * either through an options dict, or through a special
8037 * filename which the filter driver must construct in its
8038 * .bdrv_refresh_filename() implementation).
8041 bs
->exact_filename
[0] = '\0';
8044 * We can use the underlying file's filename if:
8045 * - it has a filename,
8046 * - the current BDS is not a filter,
8047 * - the file is a protocol BDS, and
8048 * - opening that file (as this BDS's format) will automatically create
8049 * the BDS tree we have right now, that is:
8050 * - the user did not significantly change this BDS's behavior with
8051 * some explicit (strong) options
8052 * - no non-file child of this BDS has been overridden by the user
8053 * Both of these conditions are represented by generate_json_filename.
8055 if (primary_child_bs
->exact_filename
[0] &&
8056 primary_child_bs
->drv
->bdrv_file_open
&&
8057 !drv
->is_filter
&& !generate_json_filename
)
8059 strcpy(bs
->exact_filename
, primary_child_bs
->exact_filename
);
8063 if (bs
->exact_filename
[0]) {
8064 pstrcpy(bs
->filename
, sizeof(bs
->filename
), bs
->exact_filename
);
8066 GString
*json
= qobject_to_json(QOBJECT(bs
->full_open_options
));
8067 if (snprintf(bs
->filename
, sizeof(bs
->filename
), "json:%s",
8068 json
->str
) >= sizeof(bs
->filename
)) {
8069 /* Give user a hint if we truncated things. */
8070 strcpy(bs
->filename
+ sizeof(bs
->filename
) - 4, "...");
8072 g_string_free(json
, true);
8076 char *bdrv_dirname(BlockDriverState
*bs
, Error
**errp
)
8078 BlockDriver
*drv
= bs
->drv
;
8079 BlockDriverState
*child_bs
;
8081 GLOBAL_STATE_CODE();
8084 error_setg(errp
, "Node '%s' is ejected", bs
->node_name
);
8088 if (drv
->bdrv_dirname
) {
8089 return drv
->bdrv_dirname(bs
, errp
);
8092 child_bs
= bdrv_primary_bs(bs
);
8094 return bdrv_dirname(child_bs
, errp
);
8097 bdrv_refresh_filename(bs
);
8098 if (bs
->exact_filename
[0] != '\0') {
8099 return path_combine(bs
->exact_filename
, "");
8102 error_setg(errp
, "Cannot generate a base directory for %s nodes",
8108 * Hot add/remove a BDS's child. So the user can take a child offline when
8109 * it is broken and take a new child online
8111 void bdrv_add_child(BlockDriverState
*parent_bs
, BlockDriverState
*child_bs
,
8114 GLOBAL_STATE_CODE();
8115 if (!parent_bs
->drv
|| !parent_bs
->drv
->bdrv_add_child
) {
8116 error_setg(errp
, "The node %s does not support adding a child",
8117 bdrv_get_device_or_node_name(parent_bs
));
8122 * Non-zoned block drivers do not follow zoned storage constraints
8123 * (i.e. sequential writes to zones). Refuse mixing zoned and non-zoned
8124 * drivers in a graph.
8126 if (!parent_bs
->drv
->supports_zoned_children
&&
8127 child_bs
->bl
.zoned
== BLK_Z_HM
) {
8129 * The host-aware model allows zoned storage constraints and random
8130 * write. Allow mixing host-aware and non-zoned drivers. Using
8131 * host-aware device as a regular device.
8133 error_setg(errp
, "Cannot add a %s child to a %s parent",
8134 child_bs
->bl
.zoned
== BLK_Z_HM
? "zoned" : "non-zoned",
8135 parent_bs
->drv
->supports_zoned_children
?
8136 "support zoned children" : "not support zoned children");
8140 if (!QLIST_EMPTY(&child_bs
->parents
)) {
8141 error_setg(errp
, "The node %s already has a parent",
8142 child_bs
->node_name
);
8146 parent_bs
->drv
->bdrv_add_child(parent_bs
, child_bs
, errp
);
8149 void bdrv_del_child(BlockDriverState
*parent_bs
, BdrvChild
*child
, Error
**errp
)
8153 GLOBAL_STATE_CODE();
8154 if (!parent_bs
->drv
|| !parent_bs
->drv
->bdrv_del_child
) {
8155 error_setg(errp
, "The node %s does not support removing a child",
8156 bdrv_get_device_or_node_name(parent_bs
));
8160 QLIST_FOREACH(tmp
, &parent_bs
->children
, next
) {
8167 error_setg(errp
, "The node %s does not have a child named %s",
8168 bdrv_get_device_or_node_name(parent_bs
),
8169 bdrv_get_device_or_node_name(child
->bs
));
8173 parent_bs
->drv
->bdrv_del_child(parent_bs
, child
, errp
);
8176 int bdrv_make_empty(BdrvChild
*c
, Error
**errp
)
8178 BlockDriver
*drv
= c
->bs
->drv
;
8181 GLOBAL_STATE_CODE();
8182 assert(c
->perm
& (BLK_PERM_WRITE
| BLK_PERM_WRITE_UNCHANGED
));
8184 if (!drv
->bdrv_make_empty
) {
8185 error_setg(errp
, "%s does not support emptying nodes",
8190 ret
= drv
->bdrv_make_empty(c
->bs
);
8192 error_setg_errno(errp
, -ret
, "Failed to empty %s",
8201 * Return the child that @bs acts as an overlay for, and from which data may be
8202 * copied in COW or COR operations. Usually this is the backing file.
8204 BdrvChild
*bdrv_cow_child(BlockDriverState
*bs
)
8208 if (!bs
|| !bs
->drv
) {
8212 if (bs
->drv
->is_filter
) {
8220 assert(bs
->backing
->role
& BDRV_CHILD_COW
);
8225 * If @bs acts as a filter for exactly one of its children, return
8228 BdrvChild
*bdrv_filter_child(BlockDriverState
*bs
)
8233 if (!bs
|| !bs
->drv
) {
8237 if (!bs
->drv
->is_filter
) {
8241 /* Only one of @backing or @file may be used */
8242 assert(!(bs
->backing
&& bs
->file
));
8244 c
= bs
->backing
?: bs
->file
;
8249 assert(c
->role
& BDRV_CHILD_FILTERED
);
8254 * Return either the result of bdrv_cow_child() or bdrv_filter_child(),
8255 * whichever is non-NULL.
8257 * Return NULL if both are NULL.
8259 BdrvChild
*bdrv_filter_or_cow_child(BlockDriverState
*bs
)
8261 BdrvChild
*cow_child
= bdrv_cow_child(bs
);
8262 BdrvChild
*filter_child
= bdrv_filter_child(bs
);
8265 /* Filter nodes cannot have COW backing files */
8266 assert(!(cow_child
&& filter_child
));
8268 return cow_child
?: filter_child
;
8272 * Return the primary child of this node: For filters, that is the
8273 * filtered child. For other nodes, that is usually the child storing
8275 * (A generally more helpful description is that this is (usually) the
8276 * child that has the same filename as @bs.)
8278 * Drivers do not necessarily have a primary child; for example quorum
8281 BdrvChild
*bdrv_primary_child(BlockDriverState
*bs
)
8283 BdrvChild
*c
, *found
= NULL
;
8286 QLIST_FOREACH(c
, &bs
->children
, next
) {
8287 if (c
->role
& BDRV_CHILD_PRIMARY
) {
8296 static BlockDriverState
* GRAPH_RDLOCK
8297 bdrv_do_skip_filters(BlockDriverState
*bs
, bool stop_on_explicit_filter
)
8305 while (!(stop_on_explicit_filter
&& !bs
->implicit
)) {
8306 c
= bdrv_filter_child(bs
);
8309 * A filter that is embedded in a working block graph must
8310 * have a child. Assert this here so this function does
8311 * not return a filter node that is not expected by the
8314 assert(!bs
->drv
|| !bs
->drv
->is_filter
);
8320 * Note that this treats nodes with bs->drv == NULL as not being
8321 * filters (bs->drv == NULL should be replaced by something else
8323 * The advantage of this behavior is that this function will thus
8324 * always return a non-NULL value (given a non-NULL @bs).
8331 * Return the first BDS that has not been added implicitly or that
8332 * does not have a filtered child down the chain starting from @bs
8333 * (including @bs itself).
8335 BlockDriverState
*bdrv_skip_implicit_filters(BlockDriverState
*bs
)
8337 GLOBAL_STATE_CODE();
8338 return bdrv_do_skip_filters(bs
, true);
8342 * Return the first BDS that does not have a filtered child down the
8343 * chain starting from @bs (including @bs itself).
8345 BlockDriverState
*bdrv_skip_filters(BlockDriverState
*bs
)
8348 return bdrv_do_skip_filters(bs
, false);
8352 * For a backing chain, return the first non-filter backing image of
8353 * the first non-filter image.
8355 BlockDriverState
*bdrv_backing_chain_next(BlockDriverState
*bs
)
8358 return bdrv_skip_filters(bdrv_cow_bs(bdrv_skip_filters(bs
)));
8362 * Check whether [offset, offset + bytes) overlaps with the cached
8363 * block-status data region.
8365 * If so, and @pnum is not NULL, set *pnum to `bsc.data_end - offset`,
8366 * which is what bdrv_bsc_is_data()'s interface needs.
8367 * Otherwise, *pnum is not touched.
8369 static bool bdrv_bsc_range_overlaps_locked(BlockDriverState
*bs
,
8370 int64_t offset
, int64_t bytes
,
8373 BdrvBlockStatusCache
*bsc
= qatomic_rcu_read(&bs
->block_status_cache
);
8377 qatomic_read(&bsc
->valid
) &&
8378 ranges_overlap(offset
, bytes
, bsc
->data_start
,
8379 bsc
->data_end
- bsc
->data_start
);
8381 if (overlaps
&& pnum
) {
8382 *pnum
= bsc
->data_end
- offset
;
8389 * See block_int.h for this function's documentation.
8391 bool bdrv_bsc_is_data(BlockDriverState
*bs
, int64_t offset
, int64_t *pnum
)
8394 RCU_READ_LOCK_GUARD();
8395 return bdrv_bsc_range_overlaps_locked(bs
, offset
, 1, pnum
);
8399 * See block_int.h for this function's documentation.
8401 void bdrv_bsc_invalidate_range(BlockDriverState
*bs
,
8402 int64_t offset
, int64_t bytes
)
8405 RCU_READ_LOCK_GUARD();
8407 if (bdrv_bsc_range_overlaps_locked(bs
, offset
, bytes
, NULL
)) {
8408 qatomic_set(&bs
->block_status_cache
->valid
, false);
8413 * See block_int.h for this function's documentation.
8415 void bdrv_bsc_fill(BlockDriverState
*bs
, int64_t offset
, int64_t bytes
)
8417 BdrvBlockStatusCache
*new_bsc
= g_new(BdrvBlockStatusCache
, 1);
8418 BdrvBlockStatusCache
*old_bsc
;
8421 *new_bsc
= (BdrvBlockStatusCache
) {
8423 .data_start
= offset
,
8424 .data_end
= offset
+ bytes
,
8427 QEMU_LOCK_GUARD(&bs
->bsc_modify_lock
);
8429 old_bsc
= qatomic_rcu_read(&bs
->block_status_cache
);
8430 qatomic_rcu_set(&bs
->block_status_cache
, new_bsc
);
8432 g_free_rcu(old_bsc
, rcu
);