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 bdrv_replace_child_noperm(BdrvChild
*child
,
95 BlockDriverState
*new_bs
);
96 static void bdrv_remove_child(BdrvChild
*child
, Transaction
*tran
);
98 static int bdrv_reopen_prepare(BDRVReopenState
*reopen_state
,
99 BlockReopenQueue
*queue
,
100 Transaction
*change_child_tran
, Error
**errp
);
101 static void bdrv_reopen_commit(BDRVReopenState
*reopen_state
);
102 static void bdrv_reopen_abort(BDRVReopenState
*reopen_state
);
104 static bool bdrv_backing_overridden(BlockDriverState
*bs
);
106 static bool bdrv_change_aio_context(BlockDriverState
*bs
, AioContext
*ctx
,
107 GHashTable
*visited
, Transaction
*tran
,
110 /* If non-zero, use only whitelisted block drivers */
111 static int use_bdrv_whitelist
;
114 static int is_windows_drive_prefix(const char *filename
)
116 return (((filename
[0] >= 'a' && filename
[0] <= 'z') ||
117 (filename
[0] >= 'A' && filename
[0] <= 'Z')) &&
121 int is_windows_drive(const char *filename
)
123 if (is_windows_drive_prefix(filename
) &&
126 if (strstart(filename
, "\\\\.\\", NULL
) ||
127 strstart(filename
, "//./", NULL
))
133 size_t bdrv_opt_mem_align(BlockDriverState
*bs
)
135 if (!bs
|| !bs
->drv
) {
136 /* page size or 4k (hdd sector size) should be on the safe side */
137 return MAX(4096, qemu_real_host_page_size());
141 return bs
->bl
.opt_mem_alignment
;
144 size_t bdrv_min_mem_align(BlockDriverState
*bs
)
146 if (!bs
|| !bs
->drv
) {
147 /* page size or 4k (hdd sector size) should be on the safe side */
148 return MAX(4096, qemu_real_host_page_size());
152 return bs
->bl
.min_mem_alignment
;
155 /* check if the path starts with "<protocol>:" */
156 int path_has_protocol(const char *path
)
161 if (is_windows_drive(path
) ||
162 is_windows_drive_prefix(path
)) {
165 p
= path
+ strcspn(path
, ":/\\");
167 p
= path
+ strcspn(path
, ":/");
173 int path_is_absolute(const char *path
)
176 /* specific case for names like: "\\.\d:" */
177 if (is_windows_drive(path
) || is_windows_drive_prefix(path
)) {
180 return (*path
== '/' || *path
== '\\');
182 return (*path
== '/');
186 /* if filename is absolute, just return its duplicate. Otherwise, build a
187 path to it by considering it is relative to base_path. URL are
189 char *path_combine(const char *base_path
, const char *filename
)
191 const char *protocol_stripped
= NULL
;
196 if (path_is_absolute(filename
)) {
197 return g_strdup(filename
);
200 if (path_has_protocol(base_path
)) {
201 protocol_stripped
= strchr(base_path
, ':');
202 if (protocol_stripped
) {
206 p
= protocol_stripped
?: base_path
;
208 p1
= strrchr(base_path
, '/');
212 p2
= strrchr(base_path
, '\\');
213 if (!p1
|| p2
> p1
) {
228 result
= g_malloc(len
+ strlen(filename
) + 1);
229 memcpy(result
, base_path
, len
);
230 strcpy(result
+ len
, filename
);
236 * Helper function for bdrv_parse_filename() implementations to remove optional
237 * protocol prefixes (especially "file:") from a filename and for putting the
238 * stripped filename into the options QDict if there is such a prefix.
240 void bdrv_parse_filename_strip_prefix(const char *filename
, const char *prefix
,
243 if (strstart(filename
, prefix
, &filename
)) {
244 /* Stripping the explicit protocol prefix may result in a protocol
245 * prefix being (wrongly) detected (if the filename contains a colon) */
246 if (path_has_protocol(filename
)) {
247 GString
*fat_filename
;
249 /* This means there is some colon before the first slash; therefore,
250 * this cannot be an absolute path */
251 assert(!path_is_absolute(filename
));
253 /* And we can thus fix the protocol detection issue by prefixing it
255 fat_filename
= g_string_new("./");
256 g_string_append(fat_filename
, filename
);
258 assert(!path_has_protocol(fat_filename
->str
));
260 qdict_put(options
, "filename",
261 qstring_from_gstring(fat_filename
));
263 /* If no protocol prefix was detected, we can use the shortened
265 qdict_put_str(options
, "filename", filename
);
271 /* Returns whether the image file is opened as read-only. Note that this can
272 * return false and writing to the image file is still not possible because the
273 * image is inactivated. */
274 bool bdrv_is_read_only(BlockDriverState
*bs
)
277 return !(bs
->open_flags
& BDRV_O_RDWR
);
280 static int bdrv_can_set_read_only(BlockDriverState
*bs
, bool read_only
,
281 bool ignore_allow_rdw
, Error
**errp
)
285 /* Do not set read_only if copy_on_read is enabled */
286 if (bs
->copy_on_read
&& read_only
) {
287 error_setg(errp
, "Can't set node '%s' to r/o with copy-on-read enabled",
288 bdrv_get_device_or_node_name(bs
));
292 /* Do not clear read_only if it is prohibited */
293 if (!read_only
&& !(bs
->open_flags
& BDRV_O_ALLOW_RDWR
) &&
296 error_setg(errp
, "Node '%s' is read only",
297 bdrv_get_device_or_node_name(bs
));
305 * Called by a driver that can only provide a read-only image.
307 * Returns 0 if the node is already read-only or it could switch the node to
308 * read-only because BDRV_O_AUTO_RDONLY is set.
310 * Returns -EACCES if the node is read-write and BDRV_O_AUTO_RDONLY is not set
311 * or bdrv_can_set_read_only() forbids making the node read-only. If @errmsg
312 * is not NULL, it is used as the error message for the Error object.
314 int bdrv_apply_auto_read_only(BlockDriverState
*bs
, const char *errmsg
,
320 if (!(bs
->open_flags
& BDRV_O_RDWR
)) {
323 if (!(bs
->open_flags
& BDRV_O_AUTO_RDONLY
)) {
327 ret
= bdrv_can_set_read_only(bs
, true, false, NULL
);
332 bs
->open_flags
&= ~BDRV_O_RDWR
;
337 error_setg(errp
, "%s", errmsg
?: "Image is read-only");
342 * If @backing is empty, this function returns NULL without setting
343 * @errp. In all other cases, NULL will only be returned with @errp
346 * Therefore, a return value of NULL without @errp set means that
347 * there is no backing file; if @errp is set, there is one but its
348 * absolute filename cannot be generated.
350 char *bdrv_get_full_backing_filename_from_filename(const char *backed
,
354 if (backing
[0] == '\0') {
356 } else if (path_has_protocol(backing
) || path_is_absolute(backing
)) {
357 return g_strdup(backing
);
358 } else if (backed
[0] == '\0' || strstart(backed
, "json:", NULL
)) {
359 error_setg(errp
, "Cannot use relative backing file names for '%s'",
363 return path_combine(backed
, backing
);
368 * If @filename is empty or NULL, this function returns NULL without
369 * setting @errp. In all other cases, NULL will only be returned with
372 static char *bdrv_make_absolute_filename(BlockDriverState
*relative_to
,
373 const char *filename
, Error
**errp
)
375 char *dir
, *full_name
;
377 if (!filename
|| filename
[0] == '\0') {
379 } else if (path_has_protocol(filename
) || path_is_absolute(filename
)) {
380 return g_strdup(filename
);
383 dir
= bdrv_dirname(relative_to
, errp
);
388 full_name
= g_strconcat(dir
, filename
, NULL
);
393 char *bdrv_get_full_backing_filename(BlockDriverState
*bs
, Error
**errp
)
396 return bdrv_make_absolute_filename(bs
, bs
->backing_file
, errp
);
399 void bdrv_register(BlockDriver
*bdrv
)
401 assert(bdrv
->format_name
);
403 QLIST_INSERT_HEAD(&bdrv_drivers
, bdrv
, list
);
406 BlockDriverState
*bdrv_new(void)
408 BlockDriverState
*bs
;
413 bs
= g_new0(BlockDriverState
, 1);
414 QLIST_INIT(&bs
->dirty_bitmaps
);
415 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
416 QLIST_INIT(&bs
->op_blockers
[i
]);
418 qemu_co_mutex_init(&bs
->reqs_lock
);
419 qemu_mutex_init(&bs
->dirty_bitmap_mutex
);
421 bs
->aio_context
= qemu_get_aio_context();
423 qemu_co_queue_init(&bs
->flush_queue
);
425 qemu_co_mutex_init(&bs
->bsc_modify_lock
);
426 bs
->block_status_cache
= g_new0(BdrvBlockStatusCache
, 1);
428 for (i
= 0; i
< bdrv_drain_all_count
; i
++) {
429 bdrv_drained_begin(bs
);
432 QTAILQ_INSERT_TAIL(&all_bdrv_states
, bs
, bs_list
);
437 static BlockDriver
*bdrv_do_find_format(const char *format_name
)
442 QLIST_FOREACH(drv1
, &bdrv_drivers
, list
) {
443 if (!strcmp(drv1
->format_name
, format_name
)) {
451 BlockDriver
*bdrv_find_format(const char *format_name
)
458 drv1
= bdrv_do_find_format(format_name
);
463 /* The driver isn't registered, maybe we need to load a module */
464 for (i
= 0; i
< (int)ARRAY_SIZE(block_driver_modules
); ++i
) {
465 if (!strcmp(block_driver_modules
[i
].format_name
, format_name
)) {
466 Error
*local_err
= NULL
;
467 int rv
= block_module_load(block_driver_modules
[i
].library_name
,
470 return bdrv_do_find_format(format_name
);
472 error_report_err(local_err
);
480 static int bdrv_format_is_whitelisted(const char *format_name
, bool read_only
)
482 static const char *whitelist_rw
[] = {
483 CONFIG_BDRV_RW_WHITELIST
486 static const char *whitelist_ro
[] = {
487 CONFIG_BDRV_RO_WHITELIST
492 if (!whitelist_rw
[0] && !whitelist_ro
[0]) {
493 return 1; /* no whitelist, anything goes */
496 for (p
= whitelist_rw
; *p
; p
++) {
497 if (!strcmp(format_name
, *p
)) {
502 for (p
= whitelist_ro
; *p
; p
++) {
503 if (!strcmp(format_name
, *p
)) {
511 int bdrv_is_whitelisted(BlockDriver
*drv
, bool read_only
)
514 return bdrv_format_is_whitelisted(drv
->format_name
, read_only
);
517 bool bdrv_uses_whitelist(void)
519 return use_bdrv_whitelist
;
522 typedef struct CreateCo
{
530 int coroutine_fn
bdrv_co_create(BlockDriver
*drv
, const char *filename
,
531 QemuOpts
*opts
, Error
**errp
)
536 assert_bdrv_graph_readable();
538 if (!drv
->bdrv_co_create_opts
) {
539 error_setg(errp
, "Driver '%s' does not support image creation",
544 ret
= drv
->bdrv_co_create_opts(drv
, filename
, opts
, errp
);
545 if (ret
< 0 && !*errp
) {
546 error_setg_errno(errp
, -ret
, "Could not create image");
553 * Helper function for bdrv_create_file_fallback(): Resize @blk to at
554 * least the given @minimum_size.
556 * On success, return @blk's actual length.
557 * Otherwise, return -errno.
559 static int64_t create_file_fallback_truncate(BlockBackend
*blk
,
560 int64_t minimum_size
, Error
**errp
)
562 Error
*local_err
= NULL
;
568 ret
= blk_truncate(blk
, minimum_size
, false, PREALLOC_MODE_OFF
, 0,
570 if (ret
< 0 && ret
!= -ENOTSUP
) {
571 error_propagate(errp
, local_err
);
575 size
= blk_getlength(blk
);
577 error_free(local_err
);
578 error_setg_errno(errp
, -size
,
579 "Failed to inquire the new image file's length");
583 if (size
< minimum_size
) {
584 /* Need to grow the image, but we failed to do that */
585 error_propagate(errp
, local_err
);
589 error_free(local_err
);
596 * Helper function for bdrv_create_file_fallback(): Zero the first
597 * sector to remove any potentially pre-existing image header.
599 static int coroutine_fn
600 create_file_fallback_zero_first_sector(BlockBackend
*blk
,
601 int64_t current_size
,
604 int64_t bytes_to_clear
;
609 bytes_to_clear
= MIN(current_size
, BDRV_SECTOR_SIZE
);
610 if (bytes_to_clear
) {
611 ret
= blk_co_pwrite_zeroes(blk
, 0, bytes_to_clear
, BDRV_REQ_MAY_UNMAP
);
613 error_setg_errno(errp
, -ret
,
614 "Failed to clear the new image's first sector");
623 * Simple implementation of bdrv_co_create_opts for protocol drivers
624 * which only support creation via opening a file
625 * (usually existing raw storage device)
627 int coroutine_fn
bdrv_co_create_opts_simple(BlockDriver
*drv
,
628 const char *filename
,
636 PreallocMode prealloc
;
637 Error
*local_err
= NULL
;
642 size
= qemu_opt_get_size_del(opts
, BLOCK_OPT_SIZE
, 0);
643 buf
= qemu_opt_get_del(opts
, BLOCK_OPT_PREALLOC
);
644 prealloc
= qapi_enum_parse(&PreallocMode_lookup
, buf
,
645 PREALLOC_MODE_OFF
, &local_err
);
648 error_propagate(errp
, local_err
);
652 if (prealloc
!= PREALLOC_MODE_OFF
) {
653 error_setg(errp
, "Unsupported preallocation mode '%s'",
654 PreallocMode_str(prealloc
));
658 options
= qdict_new();
659 qdict_put_str(options
, "driver", drv
->format_name
);
661 blk
= blk_co_new_open(filename
, NULL
, options
,
662 BDRV_O_RDWR
| BDRV_O_RESIZE
, errp
);
664 error_prepend(errp
, "Protocol driver '%s' does not support image "
665 "creation, and opening the image failed: ",
670 size
= create_file_fallback_truncate(blk
, size
, errp
);
676 ret
= create_file_fallback_zero_first_sector(blk
, size
, errp
);
687 int coroutine_fn
bdrv_co_create_file(const char *filename
, QemuOpts
*opts
,
690 QemuOpts
*protocol_opts
;
697 drv
= bdrv_find_protocol(filename
, true, errp
);
702 if (!drv
->create_opts
) {
703 error_setg(errp
, "Driver '%s' does not support image creation",
709 * 'opts' contains a QemuOptsList with a combination of format and protocol
712 * The format properly removes its options, but the default values remain
713 * in 'opts->list'. So if the protocol has options with the same name
714 * (e.g. rbd has 'cluster_size' as qcow2), it will see the default values
715 * of the format, since for overlapping options, the format wins.
717 * To avoid this issue, lets convert QemuOpts to QDict, in this way we take
718 * only the set options, and then convert it back to QemuOpts, using the
719 * create_opts of the protocol. So the new QemuOpts, will contain only the
722 qdict
= qemu_opts_to_qdict(opts
, NULL
);
723 protocol_opts
= qemu_opts_from_qdict(drv
->create_opts
, qdict
, errp
);
724 if (protocol_opts
== NULL
) {
729 ret
= bdrv_co_create(drv
, filename
, protocol_opts
, errp
);
731 qemu_opts_del(protocol_opts
);
732 qobject_unref(qdict
);
736 int coroutine_fn
bdrv_co_delete_file(BlockDriverState
*bs
, Error
**errp
)
738 Error
*local_err
= NULL
;
745 error_setg(errp
, "Block node '%s' is not opened", bs
->filename
);
749 if (!bs
->drv
->bdrv_co_delete_file
) {
750 error_setg(errp
, "Driver '%s' does not support image deletion",
751 bs
->drv
->format_name
);
755 ret
= bs
->drv
->bdrv_co_delete_file(bs
, &local_err
);
757 error_propagate(errp
, local_err
);
763 void coroutine_fn
bdrv_co_delete_file_noerr(BlockDriverState
*bs
)
765 Error
*local_err
= NULL
;
773 ret
= bdrv_co_delete_file(bs
, &local_err
);
775 * ENOTSUP will happen if the block driver doesn't support
776 * the 'bdrv_co_delete_file' interface. This is a predictable
777 * scenario and shouldn't be reported back to the user.
779 if (ret
== -ENOTSUP
) {
780 error_free(local_err
);
781 } else if (ret
< 0) {
782 error_report_err(local_err
);
787 * Try to get @bs's logical and physical block size.
788 * On success, store them in @bsz struct and return 0.
789 * On failure return -errno.
790 * @bs must not be empty.
792 int bdrv_probe_blocksizes(BlockDriverState
*bs
, BlockSizes
*bsz
)
794 BlockDriver
*drv
= bs
->drv
;
795 BlockDriverState
*filtered
= bdrv_filter_bs(bs
);
798 if (drv
&& drv
->bdrv_probe_blocksizes
) {
799 return drv
->bdrv_probe_blocksizes(bs
, bsz
);
800 } else if (filtered
) {
801 return bdrv_probe_blocksizes(filtered
, bsz
);
808 * Try to get @bs's geometry (cyls, heads, sectors).
809 * On success, store them in @geo struct and return 0.
810 * On failure return -errno.
811 * @bs must not be empty.
813 int bdrv_probe_geometry(BlockDriverState
*bs
, HDGeometry
*geo
)
815 BlockDriver
*drv
= bs
->drv
;
816 BlockDriverState
*filtered
= bdrv_filter_bs(bs
);
819 if (drv
&& drv
->bdrv_probe_geometry
) {
820 return drv
->bdrv_probe_geometry(bs
, geo
);
821 } else if (filtered
) {
822 return bdrv_probe_geometry(filtered
, geo
);
829 * Create a uniquely-named empty temporary file.
830 * Return the actual file name used upon success, otherwise NULL.
831 * This string should be freed with g_free() when not needed any longer.
833 * Note: creating a temporary file for the caller to (re)open is
834 * inherently racy. Use g_file_open_tmp() instead whenever practical.
836 char *create_tmp_file(Error
**errp
)
840 g_autofree
char *filename
= NULL
;
842 tmpdir
= g_get_tmp_dir();
845 * See commit 69bef79 ("block: use /var/tmp instead of /tmp for -snapshot")
847 * This function is used to create temporary disk images (like -snapshot),
848 * so the files can become very large. /tmp is often a tmpfs where as
849 * /var/tmp is usually on a disk, so more appropriate for disk images.
851 if (!g_strcmp0(tmpdir
, "/tmp")) {
856 filename
= g_strdup_printf("%s/vl.XXXXXX", tmpdir
);
857 fd
= g_mkstemp(filename
);
859 error_setg_errno(errp
, errno
, "Could not open temporary file '%s'",
865 return g_steal_pointer(&filename
);
869 * Detect host devices. By convention, /dev/cdrom[N] is always
870 * recognized as a host CDROM.
872 static BlockDriver
*find_hdev_driver(const char *filename
)
874 int score_max
= 0, score
;
875 BlockDriver
*drv
= NULL
, *d
;
878 QLIST_FOREACH(d
, &bdrv_drivers
, list
) {
879 if (d
->bdrv_probe_device
) {
880 score
= d
->bdrv_probe_device(filename
);
881 if (score
> score_max
) {
891 static BlockDriver
*bdrv_do_find_protocol(const char *protocol
)
896 QLIST_FOREACH(drv1
, &bdrv_drivers
, list
) {
897 if (drv1
->protocol_name
&& !strcmp(drv1
->protocol_name
, protocol
)) {
905 BlockDriver
*bdrv_find_protocol(const char *filename
,
906 bool allow_protocol_prefix
,
916 /* TODO Drivers without bdrv_file_open must be specified explicitly */
919 * XXX(hch): we really should not let host device detection
920 * override an explicit protocol specification, but moving this
921 * later breaks access to device names with colons in them.
922 * Thanks to the brain-dead persistent naming schemes on udev-
923 * based Linux systems those actually are quite common.
925 drv1
= find_hdev_driver(filename
);
930 if (!path_has_protocol(filename
) || !allow_protocol_prefix
) {
934 p
= strchr(filename
, ':');
937 if (len
> sizeof(protocol
) - 1)
938 len
= sizeof(protocol
) - 1;
939 memcpy(protocol
, filename
, len
);
940 protocol
[len
] = '\0';
942 drv1
= bdrv_do_find_protocol(protocol
);
947 for (i
= 0; i
< (int)ARRAY_SIZE(block_driver_modules
); ++i
) {
948 if (block_driver_modules
[i
].protocol_name
&&
949 !strcmp(block_driver_modules
[i
].protocol_name
, protocol
)) {
950 int rv
= block_module_load(block_driver_modules
[i
].library_name
, errp
);
952 drv1
= bdrv_do_find_protocol(protocol
);
961 error_setg(errp
, "Unknown protocol '%s'", protocol
);
967 * Guess image format by probing its contents.
968 * This is not a good idea when your image is raw (CVE-2008-2004), but
969 * we do it anyway for backward compatibility.
971 * @buf contains the image's first @buf_size bytes.
972 * @buf_size is the buffer size in bytes (generally BLOCK_PROBE_BUF_SIZE,
973 * but can be smaller if the image file is smaller)
974 * @filename is its filename.
976 * For all block drivers, call the bdrv_probe() method to get its
978 * Return the first block driver with the highest probing score.
980 BlockDriver
*bdrv_probe_all(const uint8_t *buf
, int buf_size
,
981 const char *filename
)
983 int score_max
= 0, score
;
984 BlockDriver
*drv
= NULL
, *d
;
987 QLIST_FOREACH(d
, &bdrv_drivers
, list
) {
989 score
= d
->bdrv_probe(buf
, buf_size
, filename
);
990 if (score
> score_max
) {
1000 static int find_image_format(BlockBackend
*file
, const char *filename
,
1001 BlockDriver
**pdrv
, Error
**errp
)
1004 uint8_t buf
[BLOCK_PROBE_BUF_SIZE
];
1007 GLOBAL_STATE_CODE();
1009 /* Return the raw BlockDriver * to scsi-generic devices or empty drives */
1010 if (blk_is_sg(file
) || !blk_is_inserted(file
) || blk_getlength(file
) == 0) {
1015 ret
= blk_pread(file
, 0, sizeof(buf
), buf
, 0);
1017 error_setg_errno(errp
, -ret
, "Could not read image for determining its "
1023 drv
= bdrv_probe_all(buf
, sizeof(buf
), filename
);
1025 error_setg(errp
, "Could not determine image format: No compatible "
1036 * Set the current 'total_sectors' value
1037 * Return 0 on success, -errno on error.
1039 int coroutine_fn
bdrv_co_refresh_total_sectors(BlockDriverState
*bs
,
1042 BlockDriver
*drv
= bs
->drv
;
1049 /* Do not attempt drv->bdrv_co_getlength() on scsi-generic devices */
1053 /* query actual device if possible, otherwise just trust the hint */
1054 if (drv
->bdrv_co_getlength
) {
1055 int64_t length
= drv
->bdrv_co_getlength(bs
);
1059 hint
= DIV_ROUND_UP(length
, BDRV_SECTOR_SIZE
);
1062 bs
->total_sectors
= hint
;
1064 if (bs
->total_sectors
* BDRV_SECTOR_SIZE
> BDRV_MAX_LENGTH
) {
1072 * Combines a QDict of new block driver @options with any missing options taken
1073 * from @old_options, so that leaving out an option defaults to its old value.
1075 static void bdrv_join_options(BlockDriverState
*bs
, QDict
*options
,
1078 GLOBAL_STATE_CODE();
1079 if (bs
->drv
&& bs
->drv
->bdrv_join_options
) {
1080 bs
->drv
->bdrv_join_options(options
, old_options
);
1082 qdict_join(options
, old_options
, false);
1086 static BlockdevDetectZeroesOptions
bdrv_parse_detect_zeroes(QemuOpts
*opts
,
1090 Error
*local_err
= NULL
;
1091 char *value
= qemu_opt_get_del(opts
, "detect-zeroes");
1092 BlockdevDetectZeroesOptions detect_zeroes
=
1093 qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup
, value
,
1094 BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF
, &local_err
);
1095 GLOBAL_STATE_CODE();
1098 error_propagate(errp
, local_err
);
1099 return detect_zeroes
;
1102 if (detect_zeroes
== BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP
&&
1103 !(open_flags
& BDRV_O_UNMAP
))
1105 error_setg(errp
, "setting detect-zeroes to unmap is not allowed "
1106 "without setting discard operation to unmap");
1109 return detect_zeroes
;
1113 * Set open flags for aio engine
1115 * Return 0 on success, -1 if the engine specified is invalid
1117 int bdrv_parse_aio(const char *mode
, int *flags
)
1119 if (!strcmp(mode
, "threads")) {
1120 /* do nothing, default */
1121 } else if (!strcmp(mode
, "native")) {
1122 *flags
|= BDRV_O_NATIVE_AIO
;
1123 #ifdef CONFIG_LINUX_IO_URING
1124 } else if (!strcmp(mode
, "io_uring")) {
1125 *flags
|= BDRV_O_IO_URING
;
1135 * Set open flags for a given discard mode
1137 * Return 0 on success, -1 if the discard mode was invalid.
1139 int bdrv_parse_discard_flags(const char *mode
, int *flags
)
1141 *flags
&= ~BDRV_O_UNMAP
;
1143 if (!strcmp(mode
, "off") || !strcmp(mode
, "ignore")) {
1145 } else if (!strcmp(mode
, "on") || !strcmp(mode
, "unmap")) {
1146 *flags
|= BDRV_O_UNMAP
;
1155 * Set open flags for a given cache mode
1157 * Return 0 on success, -1 if the cache mode was invalid.
1159 int bdrv_parse_cache_mode(const char *mode
, int *flags
, bool *writethrough
)
1161 *flags
&= ~BDRV_O_CACHE_MASK
;
1163 if (!strcmp(mode
, "off") || !strcmp(mode
, "none")) {
1164 *writethrough
= false;
1165 *flags
|= BDRV_O_NOCACHE
;
1166 } else if (!strcmp(mode
, "directsync")) {
1167 *writethrough
= true;
1168 *flags
|= BDRV_O_NOCACHE
;
1169 } else if (!strcmp(mode
, "writeback")) {
1170 *writethrough
= false;
1171 } else if (!strcmp(mode
, "unsafe")) {
1172 *writethrough
= false;
1173 *flags
|= BDRV_O_NO_FLUSH
;
1174 } else if (!strcmp(mode
, "writethrough")) {
1175 *writethrough
= true;
1183 static char *bdrv_child_get_parent_desc(BdrvChild
*c
)
1185 BlockDriverState
*parent
= c
->opaque
;
1186 return g_strdup_printf("node '%s'", bdrv_get_node_name(parent
));
1189 static void bdrv_child_cb_drained_begin(BdrvChild
*child
)
1191 BlockDriverState
*bs
= child
->opaque
;
1192 bdrv_do_drained_begin_quiesce(bs
, NULL
);
1195 static bool bdrv_child_cb_drained_poll(BdrvChild
*child
)
1197 BlockDriverState
*bs
= child
->opaque
;
1198 return bdrv_drain_poll(bs
, NULL
, false);
1201 static void bdrv_child_cb_drained_end(BdrvChild
*child
)
1203 BlockDriverState
*bs
= child
->opaque
;
1204 bdrv_drained_end(bs
);
1207 static int bdrv_child_cb_inactivate(BdrvChild
*child
)
1209 BlockDriverState
*bs
= child
->opaque
;
1210 GLOBAL_STATE_CODE();
1211 assert(bs
->open_flags
& BDRV_O_INACTIVE
);
1215 static bool bdrv_child_cb_change_aio_ctx(BdrvChild
*child
, AioContext
*ctx
,
1216 GHashTable
*visited
, Transaction
*tran
,
1219 BlockDriverState
*bs
= child
->opaque
;
1220 return bdrv_change_aio_context(bs
, ctx
, visited
, tran
, errp
);
1224 * Returns the options and flags that a temporary snapshot should get, based on
1225 * the originally requested flags (the originally requested image will have
1226 * flags like a backing file)
1228 static void bdrv_temp_snapshot_options(int *child_flags
, QDict
*child_options
,
1229 int parent_flags
, QDict
*parent_options
)
1231 GLOBAL_STATE_CODE();
1232 *child_flags
= (parent_flags
& ~BDRV_O_SNAPSHOT
) | BDRV_O_TEMPORARY
;
1234 /* For temporary files, unconditional cache=unsafe is fine */
1235 qdict_set_default_str(child_options
, BDRV_OPT_CACHE_DIRECT
, "off");
1236 qdict_set_default_str(child_options
, BDRV_OPT_CACHE_NO_FLUSH
, "on");
1238 /* Copy the read-only and discard options from the parent */
1239 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_READ_ONLY
);
1240 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_DISCARD
);
1242 /* aio=native doesn't work for cache.direct=off, so disable it for the
1243 * temporary snapshot */
1244 *child_flags
&= ~BDRV_O_NATIVE_AIO
;
1247 static void bdrv_backing_attach(BdrvChild
*c
)
1249 BlockDriverState
*parent
= c
->opaque
;
1250 BlockDriverState
*backing_hd
= c
->bs
;
1252 GLOBAL_STATE_CODE();
1253 assert(!parent
->backing_blocker
);
1254 error_setg(&parent
->backing_blocker
,
1255 "node is used as backing hd of '%s'",
1256 bdrv_get_device_or_node_name(parent
));
1258 bdrv_refresh_filename(backing_hd
);
1260 parent
->open_flags
&= ~BDRV_O_NO_BACKING
;
1262 bdrv_op_block_all(backing_hd
, parent
->backing_blocker
);
1263 /* Otherwise we won't be able to commit or stream */
1264 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_COMMIT_TARGET
,
1265 parent
->backing_blocker
);
1266 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_STREAM
,
1267 parent
->backing_blocker
);
1269 * We do backup in 3 ways:
1271 * The target bs is new opened, and the source is top BDS
1272 * 2. blockdev backup
1273 * Both the source and the target are top BDSes.
1274 * 3. internal backup(used for block replication)
1275 * Both the source and the target are backing file
1277 * In case 1 and 2, neither the source nor the target is the backing file.
1278 * In case 3, we will block the top BDS, so there is only one block job
1279 * for the top BDS and its backing chain.
1281 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_BACKUP_SOURCE
,
1282 parent
->backing_blocker
);
1283 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_BACKUP_TARGET
,
1284 parent
->backing_blocker
);
1287 static void bdrv_backing_detach(BdrvChild
*c
)
1289 BlockDriverState
*parent
= c
->opaque
;
1291 GLOBAL_STATE_CODE();
1292 assert(parent
->backing_blocker
);
1293 bdrv_op_unblock_all(c
->bs
, parent
->backing_blocker
);
1294 error_free(parent
->backing_blocker
);
1295 parent
->backing_blocker
= NULL
;
1298 static int bdrv_backing_update_filename(BdrvChild
*c
, BlockDriverState
*base
,
1299 const char *filename
, Error
**errp
)
1301 BlockDriverState
*parent
= c
->opaque
;
1302 bool read_only
= bdrv_is_read_only(parent
);
1304 GLOBAL_STATE_CODE();
1307 ret
= bdrv_reopen_set_read_only(parent
, false, errp
);
1313 ret
= bdrv_change_backing_file(parent
, filename
,
1314 base
->drv
? base
->drv
->format_name
: "",
1317 error_setg_errno(errp
, -ret
, "Could not update backing file link");
1321 bdrv_reopen_set_read_only(parent
, true, NULL
);
1328 * Returns the options and flags that a generic child of a BDS should
1329 * get, based on the given options and flags for the parent BDS.
1331 static void bdrv_inherited_options(BdrvChildRole role
, bool parent_is_format
,
1332 int *child_flags
, QDict
*child_options
,
1333 int parent_flags
, QDict
*parent_options
)
1335 int flags
= parent_flags
;
1336 GLOBAL_STATE_CODE();
1339 * First, decide whether to set, clear, or leave BDRV_O_PROTOCOL.
1340 * Generally, the question to answer is: Should this child be
1341 * format-probed by default?
1345 * Pure and non-filtered data children of non-format nodes should
1346 * be probed by default (even when the node itself has BDRV_O_PROTOCOL
1347 * set). This only affects a very limited set of drivers (namely
1348 * quorum and blkverify when this comment was written).
1349 * Force-clear BDRV_O_PROTOCOL then.
1351 if (!parent_is_format
&&
1352 (role
& BDRV_CHILD_DATA
) &&
1353 !(role
& (BDRV_CHILD_METADATA
| BDRV_CHILD_FILTERED
)))
1355 flags
&= ~BDRV_O_PROTOCOL
;
1359 * All children of format nodes (except for COW children) and all
1360 * metadata children in general should never be format-probed.
1361 * Force-set BDRV_O_PROTOCOL then.
1363 if ((parent_is_format
&& !(role
& BDRV_CHILD_COW
)) ||
1364 (role
& BDRV_CHILD_METADATA
))
1366 flags
|= BDRV_O_PROTOCOL
;
1370 * If the cache mode isn't explicitly set, inherit direct and no-flush from
1373 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_CACHE_DIRECT
);
1374 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_CACHE_NO_FLUSH
);
1375 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_FORCE_SHARE
);
1377 if (role
& BDRV_CHILD_COW
) {
1378 /* backing files are opened read-only by default */
1379 qdict_set_default_str(child_options
, BDRV_OPT_READ_ONLY
, "on");
1380 qdict_set_default_str(child_options
, BDRV_OPT_AUTO_READ_ONLY
, "off");
1382 /* Inherit the read-only option from the parent if it's not set */
1383 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_READ_ONLY
);
1384 qdict_copy_default(child_options
, parent_options
,
1385 BDRV_OPT_AUTO_READ_ONLY
);
1389 * bdrv_co_pdiscard() respects unmap policy for the parent, so we
1390 * can default to enable it on lower layers regardless of the
1393 qdict_set_default_str(child_options
, BDRV_OPT_DISCARD
, "unmap");
1395 /* Clear flags that only apply to the top layer */
1396 flags
&= ~(BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
| BDRV_O_COPY_ON_READ
);
1398 if (role
& BDRV_CHILD_METADATA
) {
1399 flags
&= ~BDRV_O_NO_IO
;
1401 if (role
& BDRV_CHILD_COW
) {
1402 flags
&= ~BDRV_O_TEMPORARY
;
1405 *child_flags
= flags
;
1408 static void GRAPH_WRLOCK
bdrv_child_cb_attach(BdrvChild
*child
)
1410 BlockDriverState
*bs
= child
->opaque
;
1412 assert_bdrv_graph_writable();
1413 QLIST_INSERT_HEAD(&bs
->children
, child
, next
);
1414 if (bs
->drv
->is_filter
|| (child
->role
& BDRV_CHILD_FILTERED
)) {
1416 * Here we handle filters and block/raw-format.c when it behave like
1417 * filter. They generally have a single PRIMARY child, which is also the
1418 * FILTERED child, and that they may have multiple more children, which
1419 * are neither PRIMARY nor FILTERED. And never we have a COW child here.
1420 * So bs->file will be the PRIMARY child, unless the PRIMARY child goes
1421 * into bs->backing on exceptional cases; and bs->backing will be
1424 assert(!(child
->role
& BDRV_CHILD_COW
));
1425 if (child
->role
& BDRV_CHILD_PRIMARY
) {
1426 assert(child
->role
& BDRV_CHILD_FILTERED
);
1427 assert(!bs
->backing
);
1430 if (bs
->drv
->filtered_child_is_backing
) {
1431 bs
->backing
= child
;
1436 assert(!(child
->role
& BDRV_CHILD_FILTERED
));
1438 } else if (child
->role
& BDRV_CHILD_COW
) {
1439 assert(bs
->drv
->supports_backing
);
1440 assert(!(child
->role
& BDRV_CHILD_PRIMARY
));
1441 assert(!bs
->backing
);
1442 bs
->backing
= child
;
1443 bdrv_backing_attach(child
);
1444 } else if (child
->role
& BDRV_CHILD_PRIMARY
) {
1450 static void GRAPH_WRLOCK
bdrv_child_cb_detach(BdrvChild
*child
)
1452 BlockDriverState
*bs
= child
->opaque
;
1454 if (child
->role
& BDRV_CHILD_COW
) {
1455 bdrv_backing_detach(child
);
1458 assert_bdrv_graph_writable();
1459 QLIST_REMOVE(child
, next
);
1460 if (child
== bs
->backing
) {
1461 assert(child
!= bs
->file
);
1463 } else if (child
== bs
->file
) {
1468 static int bdrv_child_cb_update_filename(BdrvChild
*c
, BlockDriverState
*base
,
1469 const char *filename
, Error
**errp
)
1471 if (c
->role
& BDRV_CHILD_COW
) {
1472 return bdrv_backing_update_filename(c
, base
, filename
, errp
);
1477 AioContext
*child_of_bds_get_parent_aio_context(BdrvChild
*c
)
1479 BlockDriverState
*bs
= c
->opaque
;
1482 return bdrv_get_aio_context(bs
);
1485 const BdrvChildClass child_of_bds
= {
1486 .parent_is_bds
= true,
1487 .get_parent_desc
= bdrv_child_get_parent_desc
,
1488 .inherit_options
= bdrv_inherited_options
,
1489 .drained_begin
= bdrv_child_cb_drained_begin
,
1490 .drained_poll
= bdrv_child_cb_drained_poll
,
1491 .drained_end
= bdrv_child_cb_drained_end
,
1492 .attach
= bdrv_child_cb_attach
,
1493 .detach
= bdrv_child_cb_detach
,
1494 .inactivate
= bdrv_child_cb_inactivate
,
1495 .change_aio_ctx
= bdrv_child_cb_change_aio_ctx
,
1496 .update_filename
= bdrv_child_cb_update_filename
,
1497 .get_parent_aio_context
= child_of_bds_get_parent_aio_context
,
1500 AioContext
*bdrv_child_get_parent_aio_context(BdrvChild
*c
)
1503 return c
->klass
->get_parent_aio_context(c
);
1506 static int bdrv_open_flags(BlockDriverState
*bs
, int flags
)
1508 int open_flags
= flags
;
1509 GLOBAL_STATE_CODE();
1512 * Clear flags that are internal to the block layer before opening the
1515 open_flags
&= ~(BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
| BDRV_O_PROTOCOL
);
1520 static void update_flags_from_options(int *flags
, QemuOpts
*opts
)
1522 GLOBAL_STATE_CODE();
1524 *flags
&= ~(BDRV_O_CACHE_MASK
| BDRV_O_RDWR
| BDRV_O_AUTO_RDONLY
);
1526 if (qemu_opt_get_bool_del(opts
, BDRV_OPT_CACHE_NO_FLUSH
, false)) {
1527 *flags
|= BDRV_O_NO_FLUSH
;
1530 if (qemu_opt_get_bool_del(opts
, BDRV_OPT_CACHE_DIRECT
, false)) {
1531 *flags
|= BDRV_O_NOCACHE
;
1534 if (!qemu_opt_get_bool_del(opts
, BDRV_OPT_READ_ONLY
, false)) {
1535 *flags
|= BDRV_O_RDWR
;
1538 if (qemu_opt_get_bool_del(opts
, BDRV_OPT_AUTO_READ_ONLY
, false)) {
1539 *flags
|= BDRV_O_AUTO_RDONLY
;
1543 static void update_options_from_flags(QDict
*options
, int flags
)
1545 GLOBAL_STATE_CODE();
1546 if (!qdict_haskey(options
, BDRV_OPT_CACHE_DIRECT
)) {
1547 qdict_put_bool(options
, BDRV_OPT_CACHE_DIRECT
, flags
& BDRV_O_NOCACHE
);
1549 if (!qdict_haskey(options
, BDRV_OPT_CACHE_NO_FLUSH
)) {
1550 qdict_put_bool(options
, BDRV_OPT_CACHE_NO_FLUSH
,
1551 flags
& BDRV_O_NO_FLUSH
);
1553 if (!qdict_haskey(options
, BDRV_OPT_READ_ONLY
)) {
1554 qdict_put_bool(options
, BDRV_OPT_READ_ONLY
, !(flags
& BDRV_O_RDWR
));
1556 if (!qdict_haskey(options
, BDRV_OPT_AUTO_READ_ONLY
)) {
1557 qdict_put_bool(options
, BDRV_OPT_AUTO_READ_ONLY
,
1558 flags
& BDRV_O_AUTO_RDONLY
);
1562 static void bdrv_assign_node_name(BlockDriverState
*bs
,
1563 const char *node_name
,
1566 char *gen_node_name
= NULL
;
1567 GLOBAL_STATE_CODE();
1570 node_name
= gen_node_name
= id_generate(ID_BLOCK
);
1571 } else if (!id_wellformed(node_name
)) {
1573 * Check for empty string or invalid characters, but not if it is
1574 * generated (generated names use characters not available to the user)
1576 error_setg(errp
, "Invalid node-name: '%s'", node_name
);
1580 /* takes care of avoiding namespaces collisions */
1581 if (blk_by_name(node_name
)) {
1582 error_setg(errp
, "node-name=%s is conflicting with a device id",
1587 /* takes care of avoiding duplicates node names */
1588 if (bdrv_find_node(node_name
)) {
1589 error_setg(errp
, "Duplicate nodes with node-name='%s'", node_name
);
1593 /* Make sure that the node name isn't truncated */
1594 if (strlen(node_name
) >= sizeof(bs
->node_name
)) {
1595 error_setg(errp
, "Node name too long");
1599 /* copy node name into the bs and insert it into the graph list */
1600 pstrcpy(bs
->node_name
, sizeof(bs
->node_name
), node_name
);
1601 QTAILQ_INSERT_TAIL(&graph_bdrv_states
, bs
, node_list
);
1603 g_free(gen_node_name
);
1607 * The caller must always hold @bs AioContext lock, because this function calls
1608 * bdrv_refresh_total_sectors() which polls when called from non-coroutine
1611 static int bdrv_open_driver(BlockDriverState
*bs
, BlockDriver
*drv
,
1612 const char *node_name
, QDict
*options
,
1613 int open_flags
, Error
**errp
)
1615 Error
*local_err
= NULL
;
1617 GLOBAL_STATE_CODE();
1619 bdrv_assign_node_name(bs
, node_name
, &local_err
);
1621 error_propagate(errp
, local_err
);
1626 bs
->opaque
= g_malloc0(drv
->instance_size
);
1628 if (drv
->bdrv_file_open
) {
1629 assert(!drv
->bdrv_needs_filename
|| bs
->filename
[0]);
1630 ret
= drv
->bdrv_file_open(bs
, options
, open_flags
, &local_err
);
1631 } else if (drv
->bdrv_open
) {
1632 ret
= drv
->bdrv_open(bs
, options
, open_flags
, &local_err
);
1639 error_propagate(errp
, local_err
);
1640 } else if (bs
->filename
[0]) {
1641 error_setg_errno(errp
, -ret
, "Could not open '%s'", bs
->filename
);
1643 error_setg_errno(errp
, -ret
, "Could not open image");
1648 assert(!(bs
->supported_read_flags
& ~BDRV_REQ_MASK
));
1649 assert(!(bs
->supported_write_flags
& ~BDRV_REQ_MASK
));
1652 * Always allow the BDRV_REQ_REGISTERED_BUF optimization hint. This saves
1653 * drivers that pass read/write requests through to a child the trouble of
1654 * declaring support explicitly.
1656 * Drivers must not propagate this flag accidentally when they initiate I/O
1657 * to a bounce buffer. That case should be rare though.
1659 bs
->supported_read_flags
|= BDRV_REQ_REGISTERED_BUF
;
1660 bs
->supported_write_flags
|= BDRV_REQ_REGISTERED_BUF
;
1662 ret
= bdrv_refresh_total_sectors(bs
, bs
->total_sectors
);
1664 error_setg_errno(errp
, -ret
, "Could not refresh total sector count");
1668 bdrv_refresh_limits(bs
, NULL
, &local_err
);
1670 error_propagate(errp
, local_err
);
1674 assert(bdrv_opt_mem_align(bs
) != 0);
1675 assert(bdrv_min_mem_align(bs
) != 0);
1676 assert(is_power_of_2(bs
->bl
.request_alignment
));
1678 for (i
= 0; i
< bs
->quiesce_counter
; i
++) {
1679 if (drv
->bdrv_drain_begin
) {
1680 drv
->bdrv_drain_begin(bs
);
1687 if (bs
->file
!= NULL
) {
1688 bdrv_unref_child(bs
, bs
->file
);
1697 * Create and open a block node.
1699 * @options is a QDict of options to pass to the block drivers, or NULL for an
1700 * empty set of options. The reference to the QDict belongs to the block layer
1701 * after the call (even on failure), so if the caller intends to reuse the
1702 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
1704 BlockDriverState
*bdrv_new_open_driver_opts(BlockDriver
*drv
,
1705 const char *node_name
,
1706 QDict
*options
, int flags
,
1709 BlockDriverState
*bs
;
1712 GLOBAL_STATE_CODE();
1715 bs
->open_flags
= flags
;
1716 bs
->options
= options
?: qdict_new();
1717 bs
->explicit_options
= qdict_clone_shallow(bs
->options
);
1720 update_options_from_flags(bs
->options
, flags
);
1722 ret
= bdrv_open_driver(bs
, drv
, node_name
, bs
->options
, flags
, errp
);
1724 qobject_unref(bs
->explicit_options
);
1725 bs
->explicit_options
= NULL
;
1726 qobject_unref(bs
->options
);
1735 /* Create and open a block node. */
1736 BlockDriverState
*bdrv_new_open_driver(BlockDriver
*drv
, const char *node_name
,
1737 int flags
, Error
**errp
)
1739 GLOBAL_STATE_CODE();
1740 return bdrv_new_open_driver_opts(drv
, node_name
, NULL
, flags
, errp
);
1743 QemuOptsList bdrv_runtime_opts
= {
1744 .name
= "bdrv_common",
1745 .head
= QTAILQ_HEAD_INITIALIZER(bdrv_runtime_opts
.head
),
1748 .name
= "node-name",
1749 .type
= QEMU_OPT_STRING
,
1750 .help
= "Node name of the block device node",
1754 .type
= QEMU_OPT_STRING
,
1755 .help
= "Block driver to use for the node",
1758 .name
= BDRV_OPT_CACHE_DIRECT
,
1759 .type
= QEMU_OPT_BOOL
,
1760 .help
= "Bypass software writeback cache on the host",
1763 .name
= BDRV_OPT_CACHE_NO_FLUSH
,
1764 .type
= QEMU_OPT_BOOL
,
1765 .help
= "Ignore flush requests",
1768 .name
= BDRV_OPT_READ_ONLY
,
1769 .type
= QEMU_OPT_BOOL
,
1770 .help
= "Node is opened in read-only mode",
1773 .name
= BDRV_OPT_AUTO_READ_ONLY
,
1774 .type
= QEMU_OPT_BOOL
,
1775 .help
= "Node can become read-only if opening read-write fails",
1778 .name
= "detect-zeroes",
1779 .type
= QEMU_OPT_STRING
,
1780 .help
= "try to optimize zero writes (off, on, unmap)",
1783 .name
= BDRV_OPT_DISCARD
,
1784 .type
= QEMU_OPT_STRING
,
1785 .help
= "discard operation (ignore/off, unmap/on)",
1788 .name
= BDRV_OPT_FORCE_SHARE
,
1789 .type
= QEMU_OPT_BOOL
,
1790 .help
= "always accept other writers (default: off)",
1792 { /* end of list */ }
1796 QemuOptsList bdrv_create_opts_simple
= {
1797 .name
= "simple-create-opts",
1798 .head
= QTAILQ_HEAD_INITIALIZER(bdrv_create_opts_simple
.head
),
1801 .name
= BLOCK_OPT_SIZE
,
1802 .type
= QEMU_OPT_SIZE
,
1803 .help
= "Virtual disk size"
1806 .name
= BLOCK_OPT_PREALLOC
,
1807 .type
= QEMU_OPT_STRING
,
1808 .help
= "Preallocation mode (allowed values: off)"
1810 { /* end of list */ }
1815 * Common part for opening disk images and files
1817 * Removes all processed options from *options.
1819 static int bdrv_open_common(BlockDriverState
*bs
, BlockBackend
*file
,
1820 QDict
*options
, Error
**errp
)
1822 int ret
, open_flags
;
1823 const char *filename
;
1824 const char *driver_name
= NULL
;
1825 const char *node_name
= NULL
;
1826 const char *discard
;
1829 Error
*local_err
= NULL
;
1832 assert(bs
->file
== NULL
);
1833 assert(options
!= NULL
&& bs
->options
!= options
);
1834 GLOBAL_STATE_CODE();
1836 opts
= qemu_opts_create(&bdrv_runtime_opts
, NULL
, 0, &error_abort
);
1837 if (!qemu_opts_absorb_qdict(opts
, options
, errp
)) {
1842 update_flags_from_options(&bs
->open_flags
, opts
);
1844 driver_name
= qemu_opt_get(opts
, "driver");
1845 drv
= bdrv_find_format(driver_name
);
1846 assert(drv
!= NULL
);
1848 bs
->force_share
= qemu_opt_get_bool(opts
, BDRV_OPT_FORCE_SHARE
, false);
1850 if (bs
->force_share
&& (bs
->open_flags
& BDRV_O_RDWR
)) {
1852 BDRV_OPT_FORCE_SHARE
1853 "=on can only be used with read-only images");
1859 bdrv_refresh_filename(blk_bs(file
));
1860 filename
= blk_bs(file
)->filename
;
1863 * Caution: while qdict_get_try_str() is fine, getting
1864 * non-string types would require more care. When @options
1865 * come from -blockdev or blockdev_add, its members are typed
1866 * according to the QAPI schema, but when they come from
1867 * -drive, they're all QString.
1869 filename
= qdict_get_try_str(options
, "filename");
1872 if (drv
->bdrv_needs_filename
&& (!filename
|| !filename
[0])) {
1873 error_setg(errp
, "The '%s' block driver requires a file name",
1879 trace_bdrv_open_common(bs
, filename
?: "", bs
->open_flags
,
1882 ro
= bdrv_is_read_only(bs
);
1884 if (use_bdrv_whitelist
&& !bdrv_is_whitelisted(drv
, ro
)) {
1885 if (!ro
&& bdrv_is_whitelisted(drv
, true)) {
1886 ret
= bdrv_apply_auto_read_only(bs
, NULL
, NULL
);
1892 !ro
&& bdrv_is_whitelisted(drv
, true)
1893 ? "Driver '%s' can only be used for read-only devices"
1894 : "Driver '%s' is not whitelisted",
1900 /* bdrv_new() and bdrv_close() make it so */
1901 assert(qatomic_read(&bs
->copy_on_read
) == 0);
1903 if (bs
->open_flags
& BDRV_O_COPY_ON_READ
) {
1905 bdrv_enable_copy_on_read(bs
);
1907 error_setg(errp
, "Can't use copy-on-read on read-only device");
1913 discard
= qemu_opt_get(opts
, BDRV_OPT_DISCARD
);
1914 if (discard
!= NULL
) {
1915 if (bdrv_parse_discard_flags(discard
, &bs
->open_flags
) != 0) {
1916 error_setg(errp
, "Invalid discard option");
1923 bdrv_parse_detect_zeroes(opts
, bs
->open_flags
, &local_err
);
1925 error_propagate(errp
, local_err
);
1930 if (filename
!= NULL
) {
1931 pstrcpy(bs
->filename
, sizeof(bs
->filename
), filename
);
1933 bs
->filename
[0] = '\0';
1935 pstrcpy(bs
->exact_filename
, sizeof(bs
->exact_filename
), bs
->filename
);
1937 /* Open the image, either directly or using a protocol */
1938 open_flags
= bdrv_open_flags(bs
, bs
->open_flags
);
1939 node_name
= qemu_opt_get(opts
, "node-name");
1941 assert(!drv
->bdrv_file_open
|| file
== NULL
);
1942 ret
= bdrv_open_driver(bs
, drv
, node_name
, options
, open_flags
, errp
);
1947 qemu_opts_del(opts
);
1951 qemu_opts_del(opts
);
1955 static QDict
*parse_json_filename(const char *filename
, Error
**errp
)
1957 QObject
*options_obj
;
1960 GLOBAL_STATE_CODE();
1962 ret
= strstart(filename
, "json:", &filename
);
1965 options_obj
= qobject_from_json(filename
, errp
);
1967 error_prepend(errp
, "Could not parse the JSON options: ");
1971 options
= qobject_to(QDict
, options_obj
);
1973 qobject_unref(options_obj
);
1974 error_setg(errp
, "Invalid JSON object given");
1978 qdict_flatten(options
);
1983 static void parse_json_protocol(QDict
*options
, const char **pfilename
,
1986 QDict
*json_options
;
1987 Error
*local_err
= NULL
;
1988 GLOBAL_STATE_CODE();
1990 /* Parse json: pseudo-protocol */
1991 if (!*pfilename
|| !g_str_has_prefix(*pfilename
, "json:")) {
1995 json_options
= parse_json_filename(*pfilename
, &local_err
);
1997 error_propagate(errp
, local_err
);
2001 /* Options given in the filename have lower priority than options
2002 * specified directly */
2003 qdict_join(options
, json_options
, false);
2004 qobject_unref(json_options
);
2009 * Fills in default options for opening images and converts the legacy
2010 * filename/flags pair to option QDict entries.
2011 * The BDRV_O_PROTOCOL flag in *flags will be set or cleared accordingly if a
2012 * block driver has been specified explicitly.
2014 static int bdrv_fill_options(QDict
**options
, const char *filename
,
2015 int *flags
, Error
**errp
)
2017 const char *drvname
;
2018 bool protocol
= *flags
& BDRV_O_PROTOCOL
;
2019 bool parse_filename
= false;
2020 BlockDriver
*drv
= NULL
;
2021 Error
*local_err
= NULL
;
2023 GLOBAL_STATE_CODE();
2026 * Caution: while qdict_get_try_str() is fine, getting non-string
2027 * types would require more care. When @options come from
2028 * -blockdev or blockdev_add, its members are typed according to
2029 * the QAPI schema, but when they come from -drive, they're all
2032 drvname
= qdict_get_try_str(*options
, "driver");
2034 drv
= bdrv_find_format(drvname
);
2036 error_setg(errp
, "Unknown driver '%s'", drvname
);
2039 /* If the user has explicitly specified the driver, this choice should
2040 * override the BDRV_O_PROTOCOL flag */
2041 protocol
= drv
->bdrv_file_open
;
2045 *flags
|= BDRV_O_PROTOCOL
;
2047 *flags
&= ~BDRV_O_PROTOCOL
;
2050 /* Translate cache options from flags into options */
2051 update_options_from_flags(*options
, *flags
);
2053 /* Fetch the file name from the options QDict if necessary */
2054 if (protocol
&& filename
) {
2055 if (!qdict_haskey(*options
, "filename")) {
2056 qdict_put_str(*options
, "filename", filename
);
2057 parse_filename
= true;
2059 error_setg(errp
, "Can't specify 'file' and 'filename' options at "
2065 /* Find the right block driver */
2066 /* See cautionary note on accessing @options above */
2067 filename
= qdict_get_try_str(*options
, "filename");
2069 if (!drvname
&& protocol
) {
2071 drv
= bdrv_find_protocol(filename
, parse_filename
, errp
);
2076 drvname
= drv
->format_name
;
2077 qdict_put_str(*options
, "driver", drvname
);
2079 error_setg(errp
, "Must specify either driver or file");
2084 assert(drv
|| !protocol
);
2086 /* Driver-specific filename parsing */
2087 if (drv
&& drv
->bdrv_parse_filename
&& parse_filename
) {
2088 drv
->bdrv_parse_filename(filename
, *options
, &local_err
);
2090 error_propagate(errp
, local_err
);
2094 if (!drv
->bdrv_needs_filename
) {
2095 qdict_del(*options
, "filename");
2102 typedef struct BlockReopenQueueEntry
{
2105 BDRVReopenState state
;
2106 QTAILQ_ENTRY(BlockReopenQueueEntry
) entry
;
2107 } BlockReopenQueueEntry
;
2110 * Return the flags that @bs will have after the reopens in @q have
2111 * successfully completed. If @q is NULL (or @bs is not contained in @q),
2112 * return the current flags.
2114 static int bdrv_reopen_get_flags(BlockReopenQueue
*q
, BlockDriverState
*bs
)
2116 BlockReopenQueueEntry
*entry
;
2119 QTAILQ_FOREACH(entry
, q
, entry
) {
2120 if (entry
->state
.bs
== bs
) {
2121 return entry
->state
.flags
;
2126 return bs
->open_flags
;
2129 /* Returns whether the image file can be written to after the reopen queue @q
2130 * has been successfully applied, or right now if @q is NULL. */
2131 static bool bdrv_is_writable_after_reopen(BlockDriverState
*bs
,
2132 BlockReopenQueue
*q
)
2134 int flags
= bdrv_reopen_get_flags(q
, bs
);
2136 return (flags
& (BDRV_O_RDWR
| BDRV_O_INACTIVE
)) == BDRV_O_RDWR
;
2140 * Return whether the BDS can be written to. This is not necessarily
2141 * the same as !bdrv_is_read_only(bs), as inactivated images may not
2142 * be written to but do not count as read-only images.
2144 bool bdrv_is_writable(BlockDriverState
*bs
)
2147 return bdrv_is_writable_after_reopen(bs
, NULL
);
2150 static char *bdrv_child_user_desc(BdrvChild
*c
)
2152 GLOBAL_STATE_CODE();
2153 return c
->klass
->get_parent_desc(c
);
2157 * Check that @a allows everything that @b needs. @a and @b must reference same
2160 static bool bdrv_a_allow_b(BdrvChild
*a
, BdrvChild
*b
, Error
**errp
)
2162 const char *child_bs_name
;
2163 g_autofree
char *a_user
= NULL
;
2164 g_autofree
char *b_user
= NULL
;
2165 g_autofree
char *perms
= NULL
;
2168 assert(a
->bs
== b
->bs
);
2169 GLOBAL_STATE_CODE();
2171 if ((b
->perm
& a
->shared_perm
) == b
->perm
) {
2175 child_bs_name
= bdrv_get_node_name(b
->bs
);
2176 a_user
= bdrv_child_user_desc(a
);
2177 b_user
= bdrv_child_user_desc(b
);
2178 perms
= bdrv_perm_names(b
->perm
& ~a
->shared_perm
);
2180 error_setg(errp
, "Permission conflict on node '%s': permissions '%s' are "
2181 "both required by %s (uses node '%s' as '%s' child) and "
2182 "unshared by %s (uses node '%s' as '%s' child).",
2183 child_bs_name
, perms
,
2184 b_user
, child_bs_name
, b
->name
,
2185 a_user
, child_bs_name
, a
->name
);
2190 static bool bdrv_parent_perms_conflict(BlockDriverState
*bs
, Error
**errp
)
2193 GLOBAL_STATE_CODE();
2196 * During the loop we'll look at each pair twice. That's correct because
2197 * bdrv_a_allow_b() is asymmetric and we should check each pair in both
2200 QLIST_FOREACH(a
, &bs
->parents
, next_parent
) {
2201 QLIST_FOREACH(b
, &bs
->parents
, next_parent
) {
2206 if (!bdrv_a_allow_b(a
, b
, errp
)) {
2215 static void bdrv_child_perm(BlockDriverState
*bs
, BlockDriverState
*child_bs
,
2216 BdrvChild
*c
, BdrvChildRole role
,
2217 BlockReopenQueue
*reopen_queue
,
2218 uint64_t parent_perm
, uint64_t parent_shared
,
2219 uint64_t *nperm
, uint64_t *nshared
)
2221 assert(bs
->drv
&& bs
->drv
->bdrv_child_perm
);
2222 GLOBAL_STATE_CODE();
2223 bs
->drv
->bdrv_child_perm(bs
, c
, role
, reopen_queue
,
2224 parent_perm
, parent_shared
,
2226 /* TODO Take force_share from reopen_queue */
2227 if (child_bs
&& child_bs
->force_share
) {
2228 *nshared
= BLK_PERM_ALL
;
2233 * Adds the whole subtree of @bs (including @bs itself) to the @list (except for
2234 * nodes that are already in the @list, of course) so that final list is
2235 * topologically sorted. Return the result (GSList @list object is updated, so
2236 * don't use old reference after function call).
2238 * On function start @list must be already topologically sorted and for any node
2239 * in the @list the whole subtree of the node must be in the @list as well. The
2240 * simplest way to satisfy this criteria: use only result of
2241 * bdrv_topological_dfs() or NULL as @list parameter.
2243 static GSList
*bdrv_topological_dfs(GSList
*list
, GHashTable
*found
,
2244 BlockDriverState
*bs
)
2247 g_autoptr(GHashTable
) local_found
= NULL
;
2249 GLOBAL_STATE_CODE();
2253 found
= local_found
= g_hash_table_new(NULL
, NULL
);
2256 if (g_hash_table_contains(found
, bs
)) {
2259 g_hash_table_add(found
, bs
);
2261 QLIST_FOREACH(child
, &bs
->children
, next
) {
2262 list
= bdrv_topological_dfs(list
, found
, child
->bs
);
2265 return g_slist_prepend(list
, bs
);
2268 typedef struct BdrvChildSetPermState
{
2271 uint64_t old_shared_perm
;
2272 } BdrvChildSetPermState
;
2274 static void bdrv_child_set_perm_abort(void *opaque
)
2276 BdrvChildSetPermState
*s
= opaque
;
2278 GLOBAL_STATE_CODE();
2280 s
->child
->perm
= s
->old_perm
;
2281 s
->child
->shared_perm
= s
->old_shared_perm
;
2284 static TransactionActionDrv bdrv_child_set_pem_drv
= {
2285 .abort
= bdrv_child_set_perm_abort
,
2289 static void bdrv_child_set_perm(BdrvChild
*c
, uint64_t perm
,
2290 uint64_t shared
, Transaction
*tran
)
2292 BdrvChildSetPermState
*s
= g_new(BdrvChildSetPermState
, 1);
2293 GLOBAL_STATE_CODE();
2295 *s
= (BdrvChildSetPermState
) {
2297 .old_perm
= c
->perm
,
2298 .old_shared_perm
= c
->shared_perm
,
2302 c
->shared_perm
= shared
;
2304 tran_add(tran
, &bdrv_child_set_pem_drv
, s
);
2307 static void bdrv_drv_set_perm_commit(void *opaque
)
2309 BlockDriverState
*bs
= opaque
;
2310 uint64_t cumulative_perms
, cumulative_shared_perms
;
2311 GLOBAL_STATE_CODE();
2313 if (bs
->drv
->bdrv_set_perm
) {
2314 bdrv_get_cumulative_perm(bs
, &cumulative_perms
,
2315 &cumulative_shared_perms
);
2316 bs
->drv
->bdrv_set_perm(bs
, cumulative_perms
, cumulative_shared_perms
);
2320 static void bdrv_drv_set_perm_abort(void *opaque
)
2322 BlockDriverState
*bs
= opaque
;
2323 GLOBAL_STATE_CODE();
2325 if (bs
->drv
->bdrv_abort_perm_update
) {
2326 bs
->drv
->bdrv_abort_perm_update(bs
);
2330 TransactionActionDrv bdrv_drv_set_perm_drv
= {
2331 .abort
= bdrv_drv_set_perm_abort
,
2332 .commit
= bdrv_drv_set_perm_commit
,
2335 static int bdrv_drv_set_perm(BlockDriverState
*bs
, uint64_t perm
,
2336 uint64_t shared_perm
, Transaction
*tran
,
2339 GLOBAL_STATE_CODE();
2344 if (bs
->drv
->bdrv_check_perm
) {
2345 int ret
= bs
->drv
->bdrv_check_perm(bs
, perm
, shared_perm
, errp
);
2352 tran_add(tran
, &bdrv_drv_set_perm_drv
, bs
);
2358 typedef struct BdrvReplaceChildState
{
2360 BlockDriverState
*old_bs
;
2361 } BdrvReplaceChildState
;
2363 static void bdrv_replace_child_commit(void *opaque
)
2365 BdrvReplaceChildState
*s
= opaque
;
2366 GLOBAL_STATE_CODE();
2368 bdrv_unref(s
->old_bs
);
2371 static void bdrv_replace_child_abort(void *opaque
)
2373 BdrvReplaceChildState
*s
= opaque
;
2374 BlockDriverState
*new_bs
= s
->child
->bs
;
2376 GLOBAL_STATE_CODE();
2377 /* old_bs reference is transparently moved from @s to @s->child */
2378 if (!s
->child
->bs
) {
2380 * The parents were undrained when removing old_bs from the child. New
2381 * requests can't have been made, though, because the child was empty.
2383 * TODO Make bdrv_replace_child_noperm() transactionable to avoid
2384 * undraining the parent in the first place. Once this is done, having
2385 * new_bs drained when calling bdrv_replace_child_tran() is not a
2386 * requirement any more.
2388 bdrv_parent_drained_begin_single(s
->child
);
2389 assert(!bdrv_parent_drained_poll_single(s
->child
));
2391 assert(s
->child
->quiesced_parent
);
2392 bdrv_replace_child_noperm(s
->child
, s
->old_bs
);
2396 static TransactionActionDrv bdrv_replace_child_drv
= {
2397 .commit
= bdrv_replace_child_commit
,
2398 .abort
= bdrv_replace_child_abort
,
2403 * bdrv_replace_child_tran
2405 * Note: real unref of old_bs is done only on commit.
2407 * Both @child->bs and @new_bs (if non-NULL) must be drained. @new_bs must be
2408 * kept drained until the transaction is completed.
2410 * The function doesn't update permissions, caller is responsible for this.
2412 static void bdrv_replace_child_tran(BdrvChild
*child
, BlockDriverState
*new_bs
,
2415 BdrvReplaceChildState
*s
= g_new(BdrvReplaceChildState
, 1);
2417 assert(child
->quiesced_parent
);
2418 assert(!new_bs
|| new_bs
->quiesce_counter
);
2420 *s
= (BdrvReplaceChildState
) {
2422 .old_bs
= child
->bs
,
2424 tran_add(tran
, &bdrv_replace_child_drv
, s
);
2429 bdrv_replace_child_noperm(child
, new_bs
);
2430 /* old_bs reference is transparently moved from @child to @s */
2434 * Refresh permissions in @bs subtree. The function is intended to be called
2435 * after some graph modification that was done without permission update.
2437 static int bdrv_node_refresh_perm(BlockDriverState
*bs
, BlockReopenQueue
*q
,
2438 Transaction
*tran
, Error
**errp
)
2440 BlockDriver
*drv
= bs
->drv
;
2443 uint64_t cumulative_perms
, cumulative_shared_perms
;
2444 GLOBAL_STATE_CODE();
2446 bdrv_get_cumulative_perm(bs
, &cumulative_perms
, &cumulative_shared_perms
);
2448 /* Write permissions never work with read-only images */
2449 if ((cumulative_perms
& (BLK_PERM_WRITE
| BLK_PERM_WRITE_UNCHANGED
)) &&
2450 !bdrv_is_writable_after_reopen(bs
, q
))
2452 if (!bdrv_is_writable_after_reopen(bs
, NULL
)) {
2453 error_setg(errp
, "Block node is read-only");
2455 error_setg(errp
, "Read-only block node '%s' cannot support "
2456 "read-write users", bdrv_get_node_name(bs
));
2463 * Unaligned requests will automatically be aligned to bl.request_alignment
2464 * and without RESIZE we can't extend requests to write to space beyond the
2465 * end of the image, so it's required that the image size is aligned.
2467 if ((cumulative_perms
& (BLK_PERM_WRITE
| BLK_PERM_WRITE_UNCHANGED
)) &&
2468 !(cumulative_perms
& BLK_PERM_RESIZE
))
2470 if ((bs
->total_sectors
* BDRV_SECTOR_SIZE
) % bs
->bl
.request_alignment
) {
2471 error_setg(errp
, "Cannot get 'write' permission without 'resize': "
2472 "Image size is not a multiple of request "
2478 /* Check this node */
2483 ret
= bdrv_drv_set_perm(bs
, cumulative_perms
, cumulative_shared_perms
, tran
,
2489 /* Drivers that never have children can omit .bdrv_child_perm() */
2490 if (!drv
->bdrv_child_perm
) {
2491 assert(QLIST_EMPTY(&bs
->children
));
2495 /* Check all children */
2496 QLIST_FOREACH(c
, &bs
->children
, next
) {
2497 uint64_t cur_perm
, cur_shared
;
2499 bdrv_child_perm(bs
, c
->bs
, c
, c
->role
, q
,
2500 cumulative_perms
, cumulative_shared_perms
,
2501 &cur_perm
, &cur_shared
);
2502 bdrv_child_set_perm(c
, cur_perm
, cur_shared
, tran
);
2509 * @list is a product of bdrv_topological_dfs() (may be called several times) -
2510 * a topologically sorted subgraph.
2512 static int bdrv_do_refresh_perms(GSList
*list
, BlockReopenQueue
*q
,
2513 Transaction
*tran
, Error
**errp
)
2516 BlockDriverState
*bs
;
2517 GLOBAL_STATE_CODE();
2519 for ( ; list
; list
= list
->next
) {
2522 if (bdrv_parent_perms_conflict(bs
, errp
)) {
2526 ret
= bdrv_node_refresh_perm(bs
, q
, tran
, errp
);
2536 * @list is any list of nodes. List is completed by all subtrees and
2537 * topologically sorted. It's not a problem if some node occurs in the @list
2540 static int bdrv_list_refresh_perms(GSList
*list
, BlockReopenQueue
*q
,
2541 Transaction
*tran
, Error
**errp
)
2543 g_autoptr(GHashTable
) found
= g_hash_table_new(NULL
, NULL
);
2544 g_autoptr(GSList
) refresh_list
= NULL
;
2546 for ( ; list
; list
= list
->next
) {
2547 refresh_list
= bdrv_topological_dfs(refresh_list
, found
, list
->data
);
2550 return bdrv_do_refresh_perms(refresh_list
, q
, tran
, errp
);
2553 void bdrv_get_cumulative_perm(BlockDriverState
*bs
, uint64_t *perm
,
2554 uint64_t *shared_perm
)
2557 uint64_t cumulative_perms
= 0;
2558 uint64_t cumulative_shared_perms
= BLK_PERM_ALL
;
2560 GLOBAL_STATE_CODE();
2562 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
2563 cumulative_perms
|= c
->perm
;
2564 cumulative_shared_perms
&= c
->shared_perm
;
2567 *perm
= cumulative_perms
;
2568 *shared_perm
= cumulative_shared_perms
;
2571 char *bdrv_perm_names(uint64_t perm
)
2577 { BLK_PERM_CONSISTENT_READ
, "consistent read" },
2578 { BLK_PERM_WRITE
, "write" },
2579 { BLK_PERM_WRITE_UNCHANGED
, "write unchanged" },
2580 { BLK_PERM_RESIZE
, "resize" },
2584 GString
*result
= g_string_sized_new(30);
2585 struct perm_name
*p
;
2587 for (p
= permissions
; p
->name
; p
++) {
2588 if (perm
& p
->perm
) {
2589 if (result
->len
> 0) {
2590 g_string_append(result
, ", ");
2592 g_string_append(result
, p
->name
);
2596 return g_string_free(result
, FALSE
);
2600 /* @tran is allowed to be NULL. In this case no rollback is possible */
2601 static int bdrv_refresh_perms(BlockDriverState
*bs
, Transaction
*tran
,
2605 Transaction
*local_tran
= NULL
;
2606 g_autoptr(GSList
) list
= bdrv_topological_dfs(NULL
, NULL
, bs
);
2607 GLOBAL_STATE_CODE();
2610 tran
= local_tran
= tran_new();
2613 ret
= bdrv_do_refresh_perms(list
, NULL
, tran
, errp
);
2616 tran_finalize(local_tran
, ret
);
2622 int bdrv_child_try_set_perm(BdrvChild
*c
, uint64_t perm
, uint64_t shared
,
2625 Error
*local_err
= NULL
;
2626 Transaction
*tran
= tran_new();
2629 GLOBAL_STATE_CODE();
2631 bdrv_child_set_perm(c
, perm
, shared
, tran
);
2633 ret
= bdrv_refresh_perms(c
->bs
, tran
, &local_err
);
2635 tran_finalize(tran
, ret
);
2638 if ((perm
& ~c
->perm
) || (c
->shared_perm
& ~shared
)) {
2639 /* tighten permissions */
2640 error_propagate(errp
, local_err
);
2643 * Our caller may intend to only loosen restrictions and
2644 * does not expect this function to fail. Errors are not
2645 * fatal in such a case, so we can just hide them from our
2648 error_free(local_err
);
2656 int bdrv_child_refresh_perms(BlockDriverState
*bs
, BdrvChild
*c
, Error
**errp
)
2658 uint64_t parent_perms
, parent_shared
;
2659 uint64_t perms
, shared
;
2661 GLOBAL_STATE_CODE();
2663 bdrv_get_cumulative_perm(bs
, &parent_perms
, &parent_shared
);
2664 bdrv_child_perm(bs
, c
->bs
, c
, c
->role
, NULL
,
2665 parent_perms
, parent_shared
, &perms
, &shared
);
2667 return bdrv_child_try_set_perm(c
, perms
, shared
, errp
);
2671 * Default implementation for .bdrv_child_perm() for block filters:
2672 * Forward CONSISTENT_READ, WRITE, WRITE_UNCHANGED, and RESIZE to the
2675 static void bdrv_filter_default_perms(BlockDriverState
*bs
, BdrvChild
*c
,
2677 BlockReopenQueue
*reopen_queue
,
2678 uint64_t perm
, uint64_t shared
,
2679 uint64_t *nperm
, uint64_t *nshared
)
2681 GLOBAL_STATE_CODE();
2682 *nperm
= perm
& DEFAULT_PERM_PASSTHROUGH
;
2683 *nshared
= (shared
& DEFAULT_PERM_PASSTHROUGH
) | DEFAULT_PERM_UNCHANGED
;
2686 static void bdrv_default_perms_for_cow(BlockDriverState
*bs
, BdrvChild
*c
,
2688 BlockReopenQueue
*reopen_queue
,
2689 uint64_t perm
, uint64_t shared
,
2690 uint64_t *nperm
, uint64_t *nshared
)
2692 assert(role
& BDRV_CHILD_COW
);
2693 GLOBAL_STATE_CODE();
2696 * We want consistent read from backing files if the parent needs it.
2697 * No other operations are performed on backing files.
2699 perm
&= BLK_PERM_CONSISTENT_READ
;
2702 * If the parent can deal with changing data, we're okay with a
2703 * writable and resizable backing file.
2704 * TODO Require !(perm & BLK_PERM_CONSISTENT_READ), too?
2706 if (shared
& BLK_PERM_WRITE
) {
2707 shared
= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2712 shared
|= BLK_PERM_CONSISTENT_READ
| BLK_PERM_WRITE_UNCHANGED
;
2714 if (bs
->open_flags
& BDRV_O_INACTIVE
) {
2715 shared
|= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2722 static void bdrv_default_perms_for_storage(BlockDriverState
*bs
, BdrvChild
*c
,
2724 BlockReopenQueue
*reopen_queue
,
2725 uint64_t perm
, uint64_t shared
,
2726 uint64_t *nperm
, uint64_t *nshared
)
2730 GLOBAL_STATE_CODE();
2731 assert(role
& (BDRV_CHILD_METADATA
| BDRV_CHILD_DATA
));
2733 flags
= bdrv_reopen_get_flags(reopen_queue
, bs
);
2736 * Apart from the modifications below, the same permissions are
2737 * forwarded and left alone as for filters
2739 bdrv_filter_default_perms(bs
, c
, role
, reopen_queue
,
2740 perm
, shared
, &perm
, &shared
);
2742 if (role
& BDRV_CHILD_METADATA
) {
2743 /* Format drivers may touch metadata even if the guest doesn't write */
2744 if (bdrv_is_writable_after_reopen(bs
, reopen_queue
)) {
2745 perm
|= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2749 * bs->file always needs to be consistent because of the
2750 * metadata. We can never allow other users to resize or write
2753 if (!(flags
& BDRV_O_NO_IO
)) {
2754 perm
|= BLK_PERM_CONSISTENT_READ
;
2756 shared
&= ~(BLK_PERM_WRITE
| BLK_PERM_RESIZE
);
2759 if (role
& BDRV_CHILD_DATA
) {
2761 * Technically, everything in this block is a subset of the
2762 * BDRV_CHILD_METADATA path taken above, and so this could
2763 * be an "else if" branch. However, that is not obvious, and
2764 * this function is not performance critical, therefore we let
2765 * this be an independent "if".
2769 * We cannot allow other users to resize the file because the
2770 * format driver might have some assumptions about the size
2771 * (e.g. because it is stored in metadata, or because the file
2772 * is split into fixed-size data files).
2774 shared
&= ~BLK_PERM_RESIZE
;
2777 * WRITE_UNCHANGED often cannot be performed as such on the
2778 * data file. For example, the qcow2 driver may still need to
2779 * write copied clusters on copy-on-read.
2781 if (perm
& BLK_PERM_WRITE_UNCHANGED
) {
2782 perm
|= BLK_PERM_WRITE
;
2786 * If the data file is written to, the format driver may
2787 * expect to be able to resize it by writing beyond the EOF.
2789 if (perm
& BLK_PERM_WRITE
) {
2790 perm
|= BLK_PERM_RESIZE
;
2794 if (bs
->open_flags
& BDRV_O_INACTIVE
) {
2795 shared
|= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2802 void bdrv_default_perms(BlockDriverState
*bs
, BdrvChild
*c
,
2803 BdrvChildRole role
, BlockReopenQueue
*reopen_queue
,
2804 uint64_t perm
, uint64_t shared
,
2805 uint64_t *nperm
, uint64_t *nshared
)
2807 GLOBAL_STATE_CODE();
2808 if (role
& BDRV_CHILD_FILTERED
) {
2809 assert(!(role
& (BDRV_CHILD_DATA
| BDRV_CHILD_METADATA
|
2811 bdrv_filter_default_perms(bs
, c
, role
, reopen_queue
,
2812 perm
, shared
, nperm
, nshared
);
2813 } else if (role
& BDRV_CHILD_COW
) {
2814 assert(!(role
& (BDRV_CHILD_DATA
| BDRV_CHILD_METADATA
)));
2815 bdrv_default_perms_for_cow(bs
, c
, role
, reopen_queue
,
2816 perm
, shared
, nperm
, nshared
);
2817 } else if (role
& (BDRV_CHILD_METADATA
| BDRV_CHILD_DATA
)) {
2818 bdrv_default_perms_for_storage(bs
, c
, role
, reopen_queue
,
2819 perm
, shared
, nperm
, nshared
);
2821 g_assert_not_reached();
2825 uint64_t bdrv_qapi_perm_to_blk_perm(BlockPermission qapi_perm
)
2827 static const uint64_t permissions
[] = {
2828 [BLOCK_PERMISSION_CONSISTENT_READ
] = BLK_PERM_CONSISTENT_READ
,
2829 [BLOCK_PERMISSION_WRITE
] = BLK_PERM_WRITE
,
2830 [BLOCK_PERMISSION_WRITE_UNCHANGED
] = BLK_PERM_WRITE_UNCHANGED
,
2831 [BLOCK_PERMISSION_RESIZE
] = BLK_PERM_RESIZE
,
2834 QEMU_BUILD_BUG_ON(ARRAY_SIZE(permissions
) != BLOCK_PERMISSION__MAX
);
2835 QEMU_BUILD_BUG_ON(1UL << ARRAY_SIZE(permissions
) != BLK_PERM_ALL
+ 1);
2837 assert(qapi_perm
< BLOCK_PERMISSION__MAX
);
2839 return permissions
[qapi_perm
];
2843 * Replaces the node that a BdrvChild points to without updating permissions.
2845 * If @new_bs is non-NULL, the parent of @child must already be drained through
2848 static void bdrv_replace_child_noperm(BdrvChild
*child
,
2849 BlockDriverState
*new_bs
)
2851 BlockDriverState
*old_bs
= child
->bs
;
2852 int new_bs_quiesce_counter
;
2854 assert(!child
->frozen
);
2857 * If we want to change the BdrvChild to point to a drained node as its new
2858 * child->bs, we need to make sure that its new parent is drained, too. In
2859 * other words, either child->quiesce_parent must already be true or we must
2860 * be able to set it and keep the parent's quiesce_counter consistent with
2861 * that, but without polling or starting new requests (this function
2862 * guarantees that it doesn't poll, and starting new requests would be
2863 * against the invariants of drain sections).
2865 * To keep things simple, we pick the first option (child->quiesce_parent
2866 * must already be true). We also generalise the rule a bit to make it
2867 * easier to verify in callers and more likely to be covered in test cases:
2868 * The parent must be quiesced through this child even if new_bs isn't
2869 * currently drained.
2871 * The only exception is for callers that always pass new_bs == NULL. In
2872 * this case, we obviously never need to consider the case of a drained
2873 * new_bs, so we can keep the callers simpler by allowing them not to drain
2876 assert(!new_bs
|| child
->quiesced_parent
);
2877 assert(old_bs
!= new_bs
);
2878 GLOBAL_STATE_CODE();
2880 if (old_bs
&& new_bs
) {
2881 assert(bdrv_get_aio_context(old_bs
) == bdrv_get_aio_context(new_bs
));
2884 /* TODO Pull this up into the callers to avoid polling here */
2885 bdrv_graph_wrlock();
2887 if (child
->klass
->detach
) {
2888 child
->klass
->detach(child
);
2890 QLIST_REMOVE(child
, next_parent
);
2896 QLIST_INSERT_HEAD(&new_bs
->parents
, child
, next_parent
);
2897 if (child
->klass
->attach
) {
2898 child
->klass
->attach(child
);
2901 bdrv_graph_wrunlock();
2904 * If the parent was drained through this BdrvChild previously, but new_bs
2905 * is not drained, allow requests to come in only after the new node has
2908 new_bs_quiesce_counter
= (new_bs
? new_bs
->quiesce_counter
: 0);
2909 if (!new_bs_quiesce_counter
&& child
->quiesced_parent
) {
2910 bdrv_parent_drained_end_single(child
);
2915 * Free the given @child.
2917 * The child must be empty (i.e. `child->bs == NULL`) and it must be
2918 * unused (i.e. not in a children list).
2920 static void bdrv_child_free(BdrvChild
*child
)
2923 GLOBAL_STATE_CODE();
2924 assert(!child
->next
.le_prev
); /* not in children list */
2926 g_free(child
->name
);
2930 typedef struct BdrvAttachChildCommonState
{
2932 AioContext
*old_parent_ctx
;
2933 AioContext
*old_child_ctx
;
2934 } BdrvAttachChildCommonState
;
2936 static void bdrv_attach_child_common_abort(void *opaque
)
2938 BdrvAttachChildCommonState
*s
= opaque
;
2939 BlockDriverState
*bs
= s
->child
->bs
;
2941 GLOBAL_STATE_CODE();
2942 bdrv_replace_child_noperm(s
->child
, NULL
);
2944 if (bdrv_get_aio_context(bs
) != s
->old_child_ctx
) {
2945 bdrv_try_change_aio_context(bs
, s
->old_child_ctx
, NULL
, &error_abort
);
2948 if (bdrv_child_get_parent_aio_context(s
->child
) != s
->old_parent_ctx
) {
2950 GHashTable
*visited
;
2955 /* No need to visit `child`, because it has been detached already */
2956 visited
= g_hash_table_new(NULL
, NULL
);
2957 ret
= s
->child
->klass
->change_aio_ctx(s
->child
, s
->old_parent_ctx
,
2958 visited
, tran
, &error_abort
);
2959 g_hash_table_destroy(visited
);
2961 /* transaction is supposed to always succeed */
2962 assert(ret
== true);
2967 bdrv_child_free(s
->child
);
2970 static TransactionActionDrv bdrv_attach_child_common_drv
= {
2971 .abort
= bdrv_attach_child_common_abort
,
2976 * Common part of attaching bdrv child to bs or to blk or to job
2978 * Function doesn't update permissions, caller is responsible for this.
2980 * Returns new created child.
2982 static BdrvChild
*bdrv_attach_child_common(BlockDriverState
*child_bs
,
2983 const char *child_name
,
2984 const BdrvChildClass
*child_class
,
2985 BdrvChildRole child_role
,
2986 uint64_t perm
, uint64_t shared_perm
,
2988 Transaction
*tran
, Error
**errp
)
2990 BdrvChild
*new_child
;
2991 AioContext
*parent_ctx
;
2992 AioContext
*child_ctx
= bdrv_get_aio_context(child_bs
);
2994 assert(child_class
->get_parent_desc
);
2995 GLOBAL_STATE_CODE();
2997 new_child
= g_new(BdrvChild
, 1);
2998 *new_child
= (BdrvChild
) {
3000 .name
= g_strdup(child_name
),
3001 .klass
= child_class
,
3004 .shared_perm
= shared_perm
,
3009 * If the AioContexts don't match, first try to move the subtree of
3010 * child_bs into the AioContext of the new parent. If this doesn't work,
3011 * try moving the parent into the AioContext of child_bs instead.
3013 parent_ctx
= bdrv_child_get_parent_aio_context(new_child
);
3014 if (child_ctx
!= parent_ctx
) {
3015 Error
*local_err
= NULL
;
3016 int ret
= bdrv_try_change_aio_context(child_bs
, parent_ctx
, NULL
,
3019 if (ret
< 0 && child_class
->change_aio_ctx
) {
3020 Transaction
*tran
= tran_new();
3021 GHashTable
*visited
= g_hash_table_new(NULL
, NULL
);
3024 g_hash_table_add(visited
, new_child
);
3025 ret_child
= child_class
->change_aio_ctx(new_child
, child_ctx
,
3026 visited
, tran
, NULL
);
3027 if (ret_child
== true) {
3028 error_free(local_err
);
3031 tran_finalize(tran
, ret_child
== true ? 0 : -1);
3032 g_hash_table_destroy(visited
);
3036 error_propagate(errp
, local_err
);
3037 bdrv_child_free(new_child
);
3044 * Let every new BdrvChild start with a drained parent. Inserting the child
3045 * in the graph with bdrv_replace_child_noperm() will undrain it if
3046 * @child_bs is not drained.
3048 * The child was only just created and is not yet visible in global state
3049 * until bdrv_replace_child_noperm() inserts it into the graph, so nobody
3050 * could have sent requests and polling is not necessary.
3052 * Note that this means that the parent isn't fully drained yet, we only
3053 * stop new requests from coming in. This is fine, we don't care about the
3054 * old requests here, they are not for this child. If another place enters a
3055 * drain section for the same parent, but wants it to be fully quiesced, it
3056 * will not run most of the the code in .drained_begin() again (which is not
3057 * a problem, we already did this), but it will still poll until the parent
3058 * is fully quiesced, so it will not be negatively affected either.
3060 bdrv_parent_drained_begin_single(new_child
);
3061 bdrv_replace_child_noperm(new_child
, child_bs
);
3063 BdrvAttachChildCommonState
*s
= g_new(BdrvAttachChildCommonState
, 1);
3064 *s
= (BdrvAttachChildCommonState
) {
3066 .old_parent_ctx
= parent_ctx
,
3067 .old_child_ctx
= child_ctx
,
3069 tran_add(tran
, &bdrv_attach_child_common_drv
, s
);
3075 * Function doesn't update permissions, caller is responsible for this.
3077 static BdrvChild
*bdrv_attach_child_noperm(BlockDriverState
*parent_bs
,
3078 BlockDriverState
*child_bs
,
3079 const char *child_name
,
3080 const BdrvChildClass
*child_class
,
3081 BdrvChildRole child_role
,
3085 uint64_t perm
, shared_perm
;
3087 assert(parent_bs
->drv
);
3088 GLOBAL_STATE_CODE();
3090 if (bdrv_recurse_has_child(child_bs
, parent_bs
)) {
3091 error_setg(errp
, "Making '%s' a %s child of '%s' would create a cycle",
3092 child_bs
->node_name
, child_name
, parent_bs
->node_name
);
3096 bdrv_get_cumulative_perm(parent_bs
, &perm
, &shared_perm
);
3097 bdrv_child_perm(parent_bs
, child_bs
, NULL
, child_role
, NULL
,
3098 perm
, shared_perm
, &perm
, &shared_perm
);
3100 return bdrv_attach_child_common(child_bs
, child_name
, child_class
,
3101 child_role
, perm
, shared_perm
, parent_bs
,
3106 * This function steals the reference to child_bs from the caller.
3107 * That reference is later dropped by bdrv_root_unref_child().
3109 * On failure NULL is returned, errp is set and the reference to
3110 * child_bs is also dropped.
3112 * The caller must hold the AioContext lock @child_bs, but not that of @ctx
3113 * (unless @child_bs is already in @ctx).
3115 BdrvChild
*bdrv_root_attach_child(BlockDriverState
*child_bs
,
3116 const char *child_name
,
3117 const BdrvChildClass
*child_class
,
3118 BdrvChildRole child_role
,
3119 uint64_t perm
, uint64_t shared_perm
,
3120 void *opaque
, Error
**errp
)
3124 Transaction
*tran
= tran_new();
3126 GLOBAL_STATE_CODE();
3128 child
= bdrv_attach_child_common(child_bs
, child_name
, child_class
,
3129 child_role
, perm
, shared_perm
, opaque
,
3136 ret
= bdrv_refresh_perms(child_bs
, tran
, errp
);
3139 tran_finalize(tran
, ret
);
3141 bdrv_unref(child_bs
);
3143 return ret
< 0 ? NULL
: child
;
3147 * This function transfers the reference to child_bs from the caller
3148 * to parent_bs. That reference is later dropped by parent_bs on
3149 * bdrv_close() or if someone calls bdrv_unref_child().
3151 * On failure NULL is returned, errp is set and the reference to
3152 * child_bs is also dropped.
3154 * If @parent_bs and @child_bs are in different AioContexts, the caller must
3155 * hold the AioContext lock for @child_bs, but not for @parent_bs.
3157 BdrvChild
*bdrv_attach_child(BlockDriverState
*parent_bs
,
3158 BlockDriverState
*child_bs
,
3159 const char *child_name
,
3160 const BdrvChildClass
*child_class
,
3161 BdrvChildRole child_role
,
3166 Transaction
*tran
= tran_new();
3168 GLOBAL_STATE_CODE();
3170 child
= bdrv_attach_child_noperm(parent_bs
, child_bs
, child_name
,
3171 child_class
, child_role
, tran
, errp
);
3177 ret
= bdrv_refresh_perms(parent_bs
, tran
, errp
);
3183 tran_finalize(tran
, ret
);
3185 bdrv_unref(child_bs
);
3187 return ret
< 0 ? NULL
: child
;
3190 /* Callers must ensure that child->frozen is false. */
3191 void bdrv_root_unref_child(BdrvChild
*child
)
3193 BlockDriverState
*child_bs
= child
->bs
;
3195 GLOBAL_STATE_CODE();
3196 bdrv_replace_child_noperm(child
, NULL
);
3197 bdrv_child_free(child
);
3201 * Update permissions for old node. We're just taking a parent away, so
3202 * we're loosening restrictions. Errors of permission update are not
3203 * fatal in this case, ignore them.
3205 bdrv_refresh_perms(child_bs
, NULL
, NULL
);
3208 * When the parent requiring a non-default AioContext is removed, the
3209 * node moves back to the main AioContext
3211 bdrv_try_change_aio_context(child_bs
, qemu_get_aio_context(), NULL
,
3215 bdrv_unref(child_bs
);
3218 typedef struct BdrvSetInheritsFrom
{
3219 BlockDriverState
*bs
;
3220 BlockDriverState
*old_inherits_from
;
3221 } BdrvSetInheritsFrom
;
3223 static void bdrv_set_inherits_from_abort(void *opaque
)
3225 BdrvSetInheritsFrom
*s
= opaque
;
3227 s
->bs
->inherits_from
= s
->old_inherits_from
;
3230 static TransactionActionDrv bdrv_set_inherits_from_drv
= {
3231 .abort
= bdrv_set_inherits_from_abort
,
3235 /* @tran is allowed to be NULL. In this case no rollback is possible */
3236 static void bdrv_set_inherits_from(BlockDriverState
*bs
,
3237 BlockDriverState
*new_inherits_from
,
3241 BdrvSetInheritsFrom
*s
= g_new(BdrvSetInheritsFrom
, 1);
3243 *s
= (BdrvSetInheritsFrom
) {
3245 .old_inherits_from
= bs
->inherits_from
,
3248 tran_add(tran
, &bdrv_set_inherits_from_drv
, s
);
3251 bs
->inherits_from
= new_inherits_from
;
3255 * Clear all inherits_from pointers from children and grandchildren of
3256 * @root that point to @root, where necessary.
3257 * @tran is allowed to be NULL. In this case no rollback is possible
3259 static void bdrv_unset_inherits_from(BlockDriverState
*root
, BdrvChild
*child
,
3264 if (child
->bs
->inherits_from
== root
) {
3266 * Remove inherits_from only when the last reference between root and
3267 * child->bs goes away.
3269 QLIST_FOREACH(c
, &root
->children
, next
) {
3270 if (c
!= child
&& c
->bs
== child
->bs
) {
3275 bdrv_set_inherits_from(child
->bs
, NULL
, tran
);
3279 QLIST_FOREACH(c
, &child
->bs
->children
, next
) {
3280 bdrv_unset_inherits_from(root
, c
, tran
);
3284 /* Callers must ensure that child->frozen is false. */
3285 void bdrv_unref_child(BlockDriverState
*parent
, BdrvChild
*child
)
3287 GLOBAL_STATE_CODE();
3288 if (child
== NULL
) {
3292 bdrv_unset_inherits_from(parent
, child
, NULL
);
3293 bdrv_root_unref_child(child
);
3297 static void bdrv_parent_cb_change_media(BlockDriverState
*bs
, bool load
)
3300 GLOBAL_STATE_CODE();
3301 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
3302 if (c
->klass
->change_media
) {
3303 c
->klass
->change_media(c
, load
);
3308 /* Return true if you can reach parent going through child->inherits_from
3309 * recursively. If parent or child are NULL, return false */
3310 static bool bdrv_inherits_from_recursive(BlockDriverState
*child
,
3311 BlockDriverState
*parent
)
3313 while (child
&& child
!= parent
) {
3314 child
= child
->inherits_from
;
3317 return child
!= NULL
;
3321 * Return the BdrvChildRole for @bs's backing child. bs->backing is
3322 * mostly used for COW backing children (role = COW), but also for
3323 * filtered children (role = FILTERED | PRIMARY).
3325 static BdrvChildRole
bdrv_backing_role(BlockDriverState
*bs
)
3327 if (bs
->drv
&& bs
->drv
->is_filter
) {
3328 return BDRV_CHILD_FILTERED
| BDRV_CHILD_PRIMARY
;
3330 return BDRV_CHILD_COW
;
3335 * Sets the bs->backing or bs->file link of a BDS. A new reference is created;
3336 * callers which don't need their own reference any more must call bdrv_unref().
3338 * Function doesn't update permissions, caller is responsible for this.
3340 static int bdrv_set_file_or_backing_noperm(BlockDriverState
*parent_bs
,
3341 BlockDriverState
*child_bs
,
3343 Transaction
*tran
, Error
**errp
)
3345 bool update_inherits_from
=
3346 bdrv_inherits_from_recursive(child_bs
, parent_bs
);
3347 BdrvChild
*child
= is_backing
? parent_bs
->backing
: parent_bs
->file
;
3350 GLOBAL_STATE_CODE();
3352 if (!parent_bs
->drv
) {
3354 * Node without drv is an object without a class :/. TODO: finally fix
3355 * qcow2 driver to never clear bs->drv and implement format corruption
3356 * handling in other way.
3358 error_setg(errp
, "Node corrupted");
3362 if (child
&& child
->frozen
) {
3363 error_setg(errp
, "Cannot change frozen '%s' link from '%s' to '%s'",
3364 child
->name
, parent_bs
->node_name
, child
->bs
->node_name
);
3368 if (is_backing
&& !parent_bs
->drv
->is_filter
&&
3369 !parent_bs
->drv
->supports_backing
)
3371 error_setg(errp
, "Driver '%s' of node '%s' does not support backing "
3372 "files", parent_bs
->drv
->format_name
, parent_bs
->node_name
);
3376 if (parent_bs
->drv
->is_filter
) {
3377 role
= BDRV_CHILD_FILTERED
| BDRV_CHILD_PRIMARY
;
3378 } else if (is_backing
) {
3379 role
= BDRV_CHILD_COW
;
3382 * We only can use same role as it is in existing child. We don't have
3383 * infrastructure to determine role of file child in generic way
3386 error_setg(errp
, "Cannot set file child to format node without "
3394 bdrv_unset_inherits_from(parent_bs
, child
, tran
);
3395 bdrv_remove_child(child
, tran
);
3402 child
= bdrv_attach_child_noperm(parent_bs
, child_bs
,
3403 is_backing
? "backing" : "file",
3404 &child_of_bds
, role
,
3412 * If inherits_from pointed recursively to bs then let's update it to
3413 * point directly to bs (else it will become NULL).
3415 if (update_inherits_from
) {
3416 bdrv_set_inherits_from(child_bs
, parent_bs
, tran
);
3420 bdrv_refresh_limits(parent_bs
, tran
, NULL
);
3425 static int bdrv_set_backing_noperm(BlockDriverState
*bs
,
3426 BlockDriverState
*backing_hd
,
3427 Transaction
*tran
, Error
**errp
)
3429 GLOBAL_STATE_CODE();
3430 return bdrv_set_file_or_backing_noperm(bs
, backing_hd
, true, tran
, errp
);
3433 int bdrv_set_backing_hd_drained(BlockDriverState
*bs
,
3434 BlockDriverState
*backing_hd
,
3438 Transaction
*tran
= tran_new();
3440 GLOBAL_STATE_CODE();
3441 assert(bs
->quiesce_counter
> 0);
3443 ret
= bdrv_set_backing_noperm(bs
, backing_hd
, tran
, errp
);
3448 ret
= bdrv_refresh_perms(bs
, tran
, errp
);
3450 tran_finalize(tran
, ret
);
3454 int bdrv_set_backing_hd(BlockDriverState
*bs
, BlockDriverState
*backing_hd
,
3458 GLOBAL_STATE_CODE();
3460 bdrv_drained_begin(bs
);
3461 ret
= bdrv_set_backing_hd_drained(bs
, backing_hd
, errp
);
3462 bdrv_drained_end(bs
);
3468 * Opens the backing file for a BlockDriverState if not yet open
3470 * bdref_key specifies the key for the image's BlockdevRef in the options QDict.
3471 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
3472 * itself, all options starting with "${bdref_key}." are considered part of the
3475 * TODO Can this be unified with bdrv_open_image()?
3477 int bdrv_open_backing_file(BlockDriverState
*bs
, QDict
*parent_options
,
3478 const char *bdref_key
, Error
**errp
)
3480 char *backing_filename
= NULL
;
3481 char *bdref_key_dot
;
3482 const char *reference
= NULL
;
3484 bool implicit_backing
= false;
3485 BlockDriverState
*backing_hd
;
3487 QDict
*tmp_parent_options
= NULL
;
3488 Error
*local_err
= NULL
;
3490 GLOBAL_STATE_CODE();
3492 if (bs
->backing
!= NULL
) {
3496 /* NULL means an empty set of options */
3497 if (parent_options
== NULL
) {
3498 tmp_parent_options
= qdict_new();
3499 parent_options
= tmp_parent_options
;
3502 bs
->open_flags
&= ~BDRV_O_NO_BACKING
;
3504 bdref_key_dot
= g_strdup_printf("%s.", bdref_key
);
3505 qdict_extract_subqdict(parent_options
, &options
, bdref_key_dot
);
3506 g_free(bdref_key_dot
);
3509 * Caution: while qdict_get_try_str() is fine, getting non-string
3510 * types would require more care. When @parent_options come from
3511 * -blockdev or blockdev_add, its members are typed according to
3512 * the QAPI schema, but when they come from -drive, they're all
3515 reference
= qdict_get_try_str(parent_options
, bdref_key
);
3516 if (reference
|| qdict_haskey(options
, "file.filename")) {
3517 /* keep backing_filename NULL */
3518 } else if (bs
->backing_file
[0] == '\0' && qdict_size(options
) == 0) {
3519 qobject_unref(options
);
3522 if (qdict_size(options
) == 0) {
3523 /* If the user specifies options that do not modify the
3524 * backing file's behavior, we might still consider it the
3525 * implicit backing file. But it's easier this way, and
3526 * just specifying some of the backing BDS's options is
3527 * only possible with -drive anyway (otherwise the QAPI
3528 * schema forces the user to specify everything). */
3529 implicit_backing
= !strcmp(bs
->auto_backing_file
, bs
->backing_file
);
3532 backing_filename
= bdrv_get_full_backing_filename(bs
, &local_err
);
3535 error_propagate(errp
, local_err
);
3536 qobject_unref(options
);
3541 if (!bs
->drv
|| !bs
->drv
->supports_backing
) {
3543 error_setg(errp
, "Driver doesn't support backing files");
3544 qobject_unref(options
);
3549 bs
->backing_format
[0] != '\0' && !qdict_haskey(options
, "driver")) {
3550 qdict_put_str(options
, "driver", bs
->backing_format
);
3553 backing_hd
= bdrv_open_inherit(backing_filename
, reference
, options
, 0, bs
,
3554 &child_of_bds
, bdrv_backing_role(bs
), errp
);
3556 bs
->open_flags
|= BDRV_O_NO_BACKING
;
3557 error_prepend(errp
, "Could not open backing file: ");
3562 if (implicit_backing
) {
3563 bdrv_refresh_filename(backing_hd
);
3564 pstrcpy(bs
->auto_backing_file
, sizeof(bs
->auto_backing_file
),
3565 backing_hd
->filename
);
3568 /* Hook up the backing file link; drop our reference, bs owns the
3569 * backing_hd reference now */
3570 ret
= bdrv_set_backing_hd(bs
, backing_hd
, errp
);
3571 bdrv_unref(backing_hd
);
3576 qdict_del(parent_options
, bdref_key
);
3579 g_free(backing_filename
);
3580 qobject_unref(tmp_parent_options
);
3584 static BlockDriverState
*
3585 bdrv_open_child_bs(const char *filename
, QDict
*options
, const char *bdref_key
,
3586 BlockDriverState
*parent
, const BdrvChildClass
*child_class
,
3587 BdrvChildRole child_role
, bool allow_none
, Error
**errp
)
3589 BlockDriverState
*bs
= NULL
;
3590 QDict
*image_options
;
3591 char *bdref_key_dot
;
3592 const char *reference
;
3594 assert(child_class
!= NULL
);
3596 bdref_key_dot
= g_strdup_printf("%s.", bdref_key
);
3597 qdict_extract_subqdict(options
, &image_options
, bdref_key_dot
);
3598 g_free(bdref_key_dot
);
3601 * Caution: while qdict_get_try_str() is fine, getting non-string
3602 * types would require more care. When @options come from
3603 * -blockdev or blockdev_add, its members are typed according to
3604 * the QAPI schema, but when they come from -drive, they're all
3607 reference
= qdict_get_try_str(options
, bdref_key
);
3608 if (!filename
&& !reference
&& !qdict_size(image_options
)) {
3610 error_setg(errp
, "A block device must be specified for \"%s\"",
3613 qobject_unref(image_options
);
3617 bs
= bdrv_open_inherit(filename
, reference
, image_options
, 0,
3618 parent
, child_class
, child_role
, errp
);
3624 qdict_del(options
, bdref_key
);
3629 * Opens a disk image whose options are given as BlockdevRef in another block
3632 * If allow_none is true, no image will be opened if filename is false and no
3633 * BlockdevRef is given. NULL will be returned, but errp remains unset.
3635 * bdrev_key specifies the key for the image's BlockdevRef in the options QDict.
3636 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
3637 * itself, all options starting with "${bdref_key}." are considered part of the
3640 * The BlockdevRef will be removed from the options QDict.
3642 BdrvChild
*bdrv_open_child(const char *filename
,
3643 QDict
*options
, const char *bdref_key
,
3644 BlockDriverState
*parent
,
3645 const BdrvChildClass
*child_class
,
3646 BdrvChildRole child_role
,
3647 bool allow_none
, Error
**errp
)
3649 BlockDriverState
*bs
;
3651 GLOBAL_STATE_CODE();
3653 bs
= bdrv_open_child_bs(filename
, options
, bdref_key
, parent
, child_class
,
3654 child_role
, allow_none
, errp
);
3659 return bdrv_attach_child(parent
, bs
, bdref_key
, child_class
, child_role
,
3664 * Wrapper on bdrv_open_child() for most popular case: open primary child of bs.
3666 int bdrv_open_file_child(const char *filename
,
3667 QDict
*options
, const char *bdref_key
,
3668 BlockDriverState
*parent
, Error
**errp
)
3672 /* commit_top and mirror_top don't use this function */
3673 assert(!parent
->drv
->filtered_child_is_backing
);
3674 role
= parent
->drv
->is_filter
?
3675 (BDRV_CHILD_FILTERED
| BDRV_CHILD_PRIMARY
) : BDRV_CHILD_IMAGE
;
3677 if (!bdrv_open_child(filename
, options
, bdref_key
, parent
,
3678 &child_of_bds
, role
, false, errp
))
3687 * TODO Future callers may need to specify parent/child_class in order for
3688 * option inheritance to work. Existing callers use it for the root node.
3690 BlockDriverState
*bdrv_open_blockdev_ref(BlockdevRef
*ref
, Error
**errp
)
3692 BlockDriverState
*bs
= NULL
;
3693 QObject
*obj
= NULL
;
3694 QDict
*qdict
= NULL
;
3695 const char *reference
= NULL
;
3698 GLOBAL_STATE_CODE();
3700 if (ref
->type
== QTYPE_QSTRING
) {
3701 reference
= ref
->u
.reference
;
3703 BlockdevOptions
*options
= &ref
->u
.definition
;
3704 assert(ref
->type
== QTYPE_QDICT
);
3706 v
= qobject_output_visitor_new(&obj
);
3707 visit_type_BlockdevOptions(v
, NULL
, &options
, &error_abort
);
3708 visit_complete(v
, &obj
);
3710 qdict
= qobject_to(QDict
, obj
);
3711 qdict_flatten(qdict
);
3713 /* bdrv_open_inherit() defaults to the values in bdrv_flags (for
3714 * compatibility with other callers) rather than what we want as the
3715 * real defaults. Apply the defaults here instead. */
3716 qdict_set_default_str(qdict
, BDRV_OPT_CACHE_DIRECT
, "off");
3717 qdict_set_default_str(qdict
, BDRV_OPT_CACHE_NO_FLUSH
, "off");
3718 qdict_set_default_str(qdict
, BDRV_OPT_READ_ONLY
, "off");
3719 qdict_set_default_str(qdict
, BDRV_OPT_AUTO_READ_ONLY
, "off");
3723 bs
= bdrv_open_inherit(NULL
, reference
, qdict
, 0, NULL
, NULL
, 0, errp
);
3730 static BlockDriverState
*bdrv_append_temp_snapshot(BlockDriverState
*bs
,
3732 QDict
*snapshot_options
,
3735 g_autofree
char *tmp_filename
= NULL
;
3737 QemuOpts
*opts
= NULL
;
3738 BlockDriverState
*bs_snapshot
= NULL
;
3741 GLOBAL_STATE_CODE();
3743 /* if snapshot, we create a temporary backing file and open it
3744 instead of opening 'filename' directly */
3746 /* Get the required size from the image */
3747 total_size
= bdrv_getlength(bs
);
3748 if (total_size
< 0) {
3749 error_setg_errno(errp
, -total_size
, "Could not get image size");
3753 /* Create the temporary image */
3754 tmp_filename
= create_tmp_file(errp
);
3755 if (!tmp_filename
) {
3759 opts
= qemu_opts_create(bdrv_qcow2
.create_opts
, NULL
, 0,
3761 qemu_opt_set_number(opts
, BLOCK_OPT_SIZE
, total_size
, &error_abort
);
3762 ret
= bdrv_create(&bdrv_qcow2
, tmp_filename
, opts
, errp
);
3763 qemu_opts_del(opts
);
3765 error_prepend(errp
, "Could not create temporary overlay '%s': ",
3770 /* Prepare options QDict for the temporary file */
3771 qdict_put_str(snapshot_options
, "file.driver", "file");
3772 qdict_put_str(snapshot_options
, "file.filename", tmp_filename
);
3773 qdict_put_str(snapshot_options
, "driver", "qcow2");
3775 bs_snapshot
= bdrv_open(NULL
, NULL
, snapshot_options
, flags
, errp
);
3776 snapshot_options
= NULL
;
3781 ret
= bdrv_append(bs_snapshot
, bs
, errp
);
3788 qobject_unref(snapshot_options
);
3793 * Opens a disk image (raw, qcow2, vmdk, ...)
3795 * options is a QDict of options to pass to the block drivers, or NULL for an
3796 * empty set of options. The reference to the QDict belongs to the block layer
3797 * after the call (even on failure), so if the caller intends to reuse the
3798 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
3800 * If *pbs is NULL, a new BDS will be created with a pointer to it stored there.
3801 * If it is not NULL, the referenced BDS will be reused.
3803 * The reference parameter may be used to specify an existing block device which
3804 * should be opened. If specified, neither options nor a filename may be given,
3805 * nor can an existing BDS be reused (that is, *pbs has to be NULL).
3807 * The caller must always hold @filename AioContext lock, because this
3808 * function eventually calls bdrv_refresh_total_sectors() which polls
3809 * when called from non-coroutine context.
3811 static BlockDriverState
* no_coroutine_fn
3812 bdrv_open_inherit(const char *filename
, const char *reference
, QDict
*options
,
3813 int flags
, BlockDriverState
*parent
,
3814 const BdrvChildClass
*child_class
, BdrvChildRole child_role
,
3818 BlockBackend
*file
= NULL
;
3819 BlockDriverState
*bs
;
3820 BlockDriver
*drv
= NULL
;
3822 const char *drvname
;
3823 const char *backing
;
3824 Error
*local_err
= NULL
;
3825 QDict
*snapshot_options
= NULL
;
3826 int snapshot_flags
= 0;
3828 assert(!child_class
|| !flags
);
3829 assert(!child_class
== !parent
);
3830 GLOBAL_STATE_CODE();
3831 assert(!qemu_in_coroutine());
3834 bool options_non_empty
= options
? qdict_size(options
) : false;
3835 qobject_unref(options
);
3837 if (filename
|| options_non_empty
) {
3838 error_setg(errp
, "Cannot reference an existing block device with "
3839 "additional options or a new filename");
3843 bs
= bdrv_lookup_bs(reference
, reference
, errp
);
3854 /* NULL means an empty set of options */
3855 if (options
== NULL
) {
3856 options
= qdict_new();
3859 /* json: syntax counts as explicit options, as if in the QDict */
3860 parse_json_protocol(options
, &filename
, &local_err
);
3865 bs
->explicit_options
= qdict_clone_shallow(options
);
3868 bool parent_is_format
;
3871 parent_is_format
= parent
->drv
->is_format
;
3874 * parent->drv is not set yet because this node is opened for
3875 * (potential) format probing. That means that @parent is going
3876 * to be a format node.
3878 parent_is_format
= true;
3881 bs
->inherits_from
= parent
;
3882 child_class
->inherit_options(child_role
, parent_is_format
,
3884 parent
->open_flags
, parent
->options
);
3887 ret
= bdrv_fill_options(&options
, filename
, &flags
, &local_err
);
3893 * Set the BDRV_O_RDWR and BDRV_O_ALLOW_RDWR flags.
3894 * Caution: getting a boolean member of @options requires care.
3895 * When @options come from -blockdev or blockdev_add, members are
3896 * typed according to the QAPI schema, but when they come from
3897 * -drive, they're all QString.
3899 if (g_strcmp0(qdict_get_try_str(options
, BDRV_OPT_READ_ONLY
), "on") &&
3900 !qdict_get_try_bool(options
, BDRV_OPT_READ_ONLY
, false)) {
3901 flags
|= (BDRV_O_RDWR
| BDRV_O_ALLOW_RDWR
);
3903 flags
&= ~BDRV_O_RDWR
;
3906 if (flags
& BDRV_O_SNAPSHOT
) {
3907 snapshot_options
= qdict_new();
3908 bdrv_temp_snapshot_options(&snapshot_flags
, snapshot_options
,
3910 /* Let bdrv_backing_options() override "read-only" */
3911 qdict_del(options
, BDRV_OPT_READ_ONLY
);
3912 bdrv_inherited_options(BDRV_CHILD_COW
, true,
3913 &flags
, options
, flags
, options
);
3916 bs
->open_flags
= flags
;
3917 bs
->options
= options
;
3918 options
= qdict_clone_shallow(options
);
3920 /* Find the right image format driver */
3921 /* See cautionary note on accessing @options above */
3922 drvname
= qdict_get_try_str(options
, "driver");
3924 drv
= bdrv_find_format(drvname
);
3926 error_setg(errp
, "Unknown driver: '%s'", drvname
);
3931 assert(drvname
|| !(flags
& BDRV_O_PROTOCOL
));
3933 /* See cautionary note on accessing @options above */
3934 backing
= qdict_get_try_str(options
, "backing");
3935 if (qobject_to(QNull
, qdict_get(options
, "backing")) != NULL
||
3936 (backing
&& *backing
== '\0'))
3939 warn_report("Use of \"backing\": \"\" is deprecated; "
3940 "use \"backing\": null instead");
3942 flags
|= BDRV_O_NO_BACKING
;
3943 qdict_del(bs
->explicit_options
, "backing");
3944 qdict_del(bs
->options
, "backing");
3945 qdict_del(options
, "backing");
3948 /* Open image file without format layer. This BlockBackend is only used for
3949 * probing, the block drivers will do their own bdrv_open_child() for the
3950 * same BDS, which is why we put the node name back into options. */
3951 if ((flags
& BDRV_O_PROTOCOL
) == 0) {
3952 BlockDriverState
*file_bs
;
3954 file_bs
= bdrv_open_child_bs(filename
, options
, "file", bs
,
3955 &child_of_bds
, BDRV_CHILD_IMAGE
,
3960 if (file_bs
!= NULL
) {
3961 /* Not requesting BLK_PERM_CONSISTENT_READ because we're only
3962 * looking at the header to guess the image format. This works even
3963 * in cases where a guest would not see a consistent state. */
3964 file
= blk_new(bdrv_get_aio_context(file_bs
), 0, BLK_PERM_ALL
);
3965 blk_insert_bs(file
, file_bs
, &local_err
);
3966 bdrv_unref(file_bs
);
3971 qdict_put_str(options
, "file", bdrv_get_node_name(file_bs
));
3975 /* Image format probing */
3978 ret
= find_image_format(file
, filename
, &drv
, &local_err
);
3983 * This option update would logically belong in bdrv_fill_options(),
3984 * but we first need to open bs->file for the probing to work, while
3985 * opening bs->file already requires the (mostly) final set of options
3986 * so that cache mode etc. can be inherited.
3988 * Adding the driver later is somewhat ugly, but it's not an option
3989 * that would ever be inherited, so it's correct. We just need to make
3990 * sure to update both bs->options (which has the full effective
3991 * options for bs) and options (which has file.* already removed).
3993 qdict_put_str(bs
->options
, "driver", drv
->format_name
);
3994 qdict_put_str(options
, "driver", drv
->format_name
);
3996 error_setg(errp
, "Must specify either driver or file");
4000 /* BDRV_O_PROTOCOL must be set iff a protocol BDS is about to be created */
4001 assert(!!(flags
& BDRV_O_PROTOCOL
) == !!drv
->bdrv_file_open
);
4002 /* file must be NULL if a protocol BDS is about to be created
4003 * (the inverse results in an error message from bdrv_open_common()) */
4004 assert(!(flags
& BDRV_O_PROTOCOL
) || !file
);
4006 /* Open the image */
4007 ret
= bdrv_open_common(bs
, file
, options
, &local_err
);
4017 /* If there is a backing file, use it */
4018 if ((flags
& BDRV_O_NO_BACKING
) == 0) {
4019 ret
= bdrv_open_backing_file(bs
, options
, "backing", &local_err
);
4021 goto close_and_fail
;
4025 /* Remove all children options and references
4026 * from bs->options and bs->explicit_options */
4027 QLIST_FOREACH(child
, &bs
->children
, next
) {
4028 char *child_key_dot
;
4029 child_key_dot
= g_strdup_printf("%s.", child
->name
);
4030 qdict_extract_subqdict(bs
->explicit_options
, NULL
, child_key_dot
);
4031 qdict_extract_subqdict(bs
->options
, NULL
, child_key_dot
);
4032 qdict_del(bs
->explicit_options
, child
->name
);
4033 qdict_del(bs
->options
, child
->name
);
4034 g_free(child_key_dot
);
4037 /* Check if any unknown options were used */
4038 if (qdict_size(options
) != 0) {
4039 const QDictEntry
*entry
= qdict_first(options
);
4040 if (flags
& BDRV_O_PROTOCOL
) {
4041 error_setg(errp
, "Block protocol '%s' doesn't support the option "
4042 "'%s'", drv
->format_name
, entry
->key
);
4045 "Block format '%s' does not support the option '%s'",
4046 drv
->format_name
, entry
->key
);
4049 goto close_and_fail
;
4052 bdrv_parent_cb_change_media(bs
, true);
4054 qobject_unref(options
);
4057 /* For snapshot=on, create a temporary qcow2 overlay. bs points to the
4058 * temporary snapshot afterwards. */
4059 if (snapshot_flags
) {
4060 BlockDriverState
*snapshot_bs
;
4061 snapshot_bs
= bdrv_append_temp_snapshot(bs
, snapshot_flags
,
4062 snapshot_options
, &local_err
);
4063 snapshot_options
= NULL
;
4065 goto close_and_fail
;
4067 /* We are not going to return bs but the overlay on top of it
4068 * (snapshot_bs); thus, we have to drop the strong reference to bs
4069 * (which we obtained by calling bdrv_new()). bs will not be deleted,
4070 * though, because the overlay still has a reference to it. */
4079 qobject_unref(snapshot_options
);
4080 qobject_unref(bs
->explicit_options
);
4081 qobject_unref(bs
->options
);
4082 qobject_unref(options
);
4084 bs
->explicit_options
= NULL
;
4086 error_propagate(errp
, local_err
);
4091 qobject_unref(snapshot_options
);
4092 qobject_unref(options
);
4093 error_propagate(errp
, local_err
);
4098 * The caller must always hold @filename AioContext lock, because this
4099 * function eventually calls bdrv_refresh_total_sectors() which polls
4100 * when called from non-coroutine context.
4102 BlockDriverState
*bdrv_open(const char *filename
, const char *reference
,
4103 QDict
*options
, int flags
, Error
**errp
)
4105 GLOBAL_STATE_CODE();
4107 return bdrv_open_inherit(filename
, reference
, options
, flags
, NULL
,
4111 /* Return true if the NULL-terminated @list contains @str */
4112 static bool is_str_in_list(const char *str
, const char *const *list
)
4116 for (i
= 0; list
[i
] != NULL
; i
++) {
4117 if (!strcmp(str
, list
[i
])) {
4126 * Check that every option set in @bs->options is also set in
4129 * Options listed in the common_options list and in
4130 * @bs->drv->mutable_opts are skipped.
4132 * Return 0 on success, otherwise return -EINVAL and set @errp.
4134 static int bdrv_reset_options_allowed(BlockDriverState
*bs
,
4135 const QDict
*new_opts
, Error
**errp
)
4137 const QDictEntry
*e
;
4138 /* These options are common to all block drivers and are handled
4139 * in bdrv_reopen_prepare() so they can be left out of @new_opts */
4140 const char *const common_options
[] = {
4141 "node-name", "discard", "cache.direct", "cache.no-flush",
4142 "read-only", "auto-read-only", "detect-zeroes", NULL
4145 for (e
= qdict_first(bs
->options
); e
; e
= qdict_next(bs
->options
, e
)) {
4146 if (!qdict_haskey(new_opts
, e
->key
) &&
4147 !is_str_in_list(e
->key
, common_options
) &&
4148 !is_str_in_list(e
->key
, bs
->drv
->mutable_opts
)) {
4149 error_setg(errp
, "Option '%s' cannot be reset "
4150 "to its default value", e
->key
);
4159 * Returns true if @child can be reached recursively from @bs
4161 static bool bdrv_recurse_has_child(BlockDriverState
*bs
,
4162 BlockDriverState
*child
)
4170 QLIST_FOREACH(c
, &bs
->children
, next
) {
4171 if (bdrv_recurse_has_child(c
->bs
, child
)) {
4180 * Adds a BlockDriverState to a simple queue for an atomic, transactional
4181 * reopen of multiple devices.
4183 * bs_queue can either be an existing BlockReopenQueue that has had QTAILQ_INIT
4184 * already performed, or alternatively may be NULL a new BlockReopenQueue will
4185 * be created and initialized. This newly created BlockReopenQueue should be
4186 * passed back in for subsequent calls that are intended to be of the same
4189 * bs is the BlockDriverState to add to the reopen queue.
4191 * options contains the changed options for the associated bs
4192 * (the BlockReopenQueue takes ownership)
4194 * flags contains the open flags for the associated bs
4196 * returns a pointer to bs_queue, which is either the newly allocated
4197 * bs_queue, or the existing bs_queue being used.
4199 * bs is drained here and undrained by bdrv_reopen_queue_free().
4201 * To be called with bs->aio_context locked.
4203 static BlockReopenQueue
*bdrv_reopen_queue_child(BlockReopenQueue
*bs_queue
,
4204 BlockDriverState
*bs
,
4206 const BdrvChildClass
*klass
,
4208 bool parent_is_format
,
4209 QDict
*parent_options
,
4215 BlockReopenQueueEntry
*bs_entry
;
4217 QDict
*old_options
, *explicit_options
, *options_copy
;
4221 GLOBAL_STATE_CODE();
4223 bdrv_drained_begin(bs
);
4225 if (bs_queue
== NULL
) {
4226 bs_queue
= g_new0(BlockReopenQueue
, 1);
4227 QTAILQ_INIT(bs_queue
);
4231 options
= qdict_new();
4234 /* Check if this BlockDriverState is already in the queue */
4235 QTAILQ_FOREACH(bs_entry
, bs_queue
, entry
) {
4236 if (bs
== bs_entry
->state
.bs
) {
4242 * Precedence of options:
4243 * 1. Explicitly passed in options (highest)
4244 * 2. Retained from explicitly set options of bs
4245 * 3. Inherited from parent node
4246 * 4. Retained from effective options of bs
4249 /* Old explicitly set values (don't overwrite by inherited value) */
4250 if (bs_entry
|| keep_old_opts
) {
4251 old_options
= qdict_clone_shallow(bs_entry
?
4252 bs_entry
->state
.explicit_options
:
4253 bs
->explicit_options
);
4254 bdrv_join_options(bs
, options
, old_options
);
4255 qobject_unref(old_options
);
4258 explicit_options
= qdict_clone_shallow(options
);
4260 /* Inherit from parent node */
4261 if (parent_options
) {
4263 klass
->inherit_options(role
, parent_is_format
, &flags
, options
,
4264 parent_flags
, parent_options
);
4266 flags
= bdrv_get_flags(bs
);
4269 if (keep_old_opts
) {
4270 /* Old values are used for options that aren't set yet */
4271 old_options
= qdict_clone_shallow(bs
->options
);
4272 bdrv_join_options(bs
, options
, old_options
);
4273 qobject_unref(old_options
);
4276 /* We have the final set of options so let's update the flags */
4277 options_copy
= qdict_clone_shallow(options
);
4278 opts
= qemu_opts_create(&bdrv_runtime_opts
, NULL
, 0, &error_abort
);
4279 qemu_opts_absorb_qdict(opts
, options_copy
, NULL
);
4280 update_flags_from_options(&flags
, opts
);
4281 qemu_opts_del(opts
);
4282 qobject_unref(options_copy
);
4284 /* bdrv_open_inherit() sets and clears some additional flags internally */
4285 flags
&= ~BDRV_O_PROTOCOL
;
4286 if (flags
& BDRV_O_RDWR
) {
4287 flags
|= BDRV_O_ALLOW_RDWR
;
4291 bs_entry
= g_new0(BlockReopenQueueEntry
, 1);
4292 QTAILQ_INSERT_TAIL(bs_queue
, bs_entry
, entry
);
4294 qobject_unref(bs_entry
->state
.options
);
4295 qobject_unref(bs_entry
->state
.explicit_options
);
4298 bs_entry
->state
.bs
= bs
;
4299 bs_entry
->state
.options
= options
;
4300 bs_entry
->state
.explicit_options
= explicit_options
;
4301 bs_entry
->state
.flags
= flags
;
4304 * If keep_old_opts is false then it means that unspecified
4305 * options must be reset to their original value. We don't allow
4306 * resetting 'backing' but we need to know if the option is
4307 * missing in order to decide if we have to return an error.
4309 if (!keep_old_opts
) {
4310 bs_entry
->state
.backing_missing
=
4311 !qdict_haskey(options
, "backing") &&
4312 !qdict_haskey(options
, "backing.driver");
4315 QLIST_FOREACH(child
, &bs
->children
, next
) {
4316 QDict
*new_child_options
= NULL
;
4317 bool child_keep_old
= keep_old_opts
;
4319 /* reopen can only change the options of block devices that were
4320 * implicitly created and inherited options. For other (referenced)
4321 * block devices, a syntax like "backing.foo" results in an error. */
4322 if (child
->bs
->inherits_from
!= bs
) {
4326 /* Check if the options contain a child reference */
4327 if (qdict_haskey(options
, child
->name
)) {
4328 const char *childref
= qdict_get_try_str(options
, child
->name
);
4330 * The current child must not be reopened if the child
4331 * reference is null or points to a different node.
4333 if (g_strcmp0(childref
, child
->bs
->node_name
)) {
4337 * If the child reference points to the current child then
4338 * reopen it with its existing set of options (note that
4339 * it can still inherit new options from the parent).
4341 child_keep_old
= true;
4343 /* Extract child options ("child-name.*") */
4344 char *child_key_dot
= g_strdup_printf("%s.", child
->name
);
4345 qdict_extract_subqdict(explicit_options
, NULL
, child_key_dot
);
4346 qdict_extract_subqdict(options
, &new_child_options
, child_key_dot
);
4347 g_free(child_key_dot
);
4350 bdrv_reopen_queue_child(bs_queue
, child
->bs
, new_child_options
,
4351 child
->klass
, child
->role
, bs
->drv
->is_format
,
4352 options
, flags
, child_keep_old
);
4358 /* To be called with bs->aio_context locked */
4359 BlockReopenQueue
*bdrv_reopen_queue(BlockReopenQueue
*bs_queue
,
4360 BlockDriverState
*bs
,
4361 QDict
*options
, bool keep_old_opts
)
4363 GLOBAL_STATE_CODE();
4365 return bdrv_reopen_queue_child(bs_queue
, bs
, options
, NULL
, 0, false,
4366 NULL
, 0, keep_old_opts
);
4369 void bdrv_reopen_queue_free(BlockReopenQueue
*bs_queue
)
4371 GLOBAL_STATE_CODE();
4373 BlockReopenQueueEntry
*bs_entry
, *next
;
4374 QTAILQ_FOREACH_SAFE(bs_entry
, bs_queue
, entry
, next
) {
4375 AioContext
*ctx
= bdrv_get_aio_context(bs_entry
->state
.bs
);
4377 aio_context_acquire(ctx
);
4378 bdrv_drained_end(bs_entry
->state
.bs
);
4379 aio_context_release(ctx
);
4381 qobject_unref(bs_entry
->state
.explicit_options
);
4382 qobject_unref(bs_entry
->state
.options
);
4390 * Reopen multiple BlockDriverStates atomically & transactionally.
4392 * The queue passed in (bs_queue) must have been built up previous
4393 * via bdrv_reopen_queue().
4395 * Reopens all BDS specified in the queue, with the appropriate
4396 * flags. All devices are prepared for reopen, and failure of any
4397 * device will cause all device changes to be abandoned, and intermediate
4400 * If all devices prepare successfully, then the changes are committed
4403 * All affected nodes must be drained between bdrv_reopen_queue() and
4404 * bdrv_reopen_multiple().
4406 * To be called from the main thread, with all other AioContexts unlocked.
4408 int bdrv_reopen_multiple(BlockReopenQueue
*bs_queue
, Error
**errp
)
4411 BlockReopenQueueEntry
*bs_entry
, *next
;
4413 Transaction
*tran
= tran_new();
4414 g_autoptr(GSList
) refresh_list
= NULL
;
4416 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
4417 assert(bs_queue
!= NULL
);
4418 GLOBAL_STATE_CODE();
4420 QTAILQ_FOREACH(bs_entry
, bs_queue
, entry
) {
4421 ctx
= bdrv_get_aio_context(bs_entry
->state
.bs
);
4422 aio_context_acquire(ctx
);
4423 ret
= bdrv_flush(bs_entry
->state
.bs
);
4424 aio_context_release(ctx
);
4426 error_setg_errno(errp
, -ret
, "Error flushing drive");
4431 QTAILQ_FOREACH(bs_entry
, bs_queue
, entry
) {
4432 assert(bs_entry
->state
.bs
->quiesce_counter
> 0);
4433 ctx
= bdrv_get_aio_context(bs_entry
->state
.bs
);
4434 aio_context_acquire(ctx
);
4435 ret
= bdrv_reopen_prepare(&bs_entry
->state
, bs_queue
, tran
, errp
);
4436 aio_context_release(ctx
);
4440 bs_entry
->prepared
= true;
4443 QTAILQ_FOREACH(bs_entry
, bs_queue
, entry
) {
4444 BDRVReopenState
*state
= &bs_entry
->state
;
4446 refresh_list
= g_slist_prepend(refresh_list
, state
->bs
);
4447 if (state
->old_backing_bs
) {
4448 refresh_list
= g_slist_prepend(refresh_list
, state
->old_backing_bs
);
4450 if (state
->old_file_bs
) {
4451 refresh_list
= g_slist_prepend(refresh_list
, state
->old_file_bs
);
4456 * Note that file-posix driver rely on permission update done during reopen
4457 * (even if no permission changed), because it wants "new" permissions for
4458 * reconfiguring the fd and that's why it does it in raw_check_perm(), not
4459 * in raw_reopen_prepare() which is called with "old" permissions.
4461 ret
= bdrv_list_refresh_perms(refresh_list
, bs_queue
, tran
, errp
);
4467 * If we reach this point, we have success and just need to apply the
4470 * Reverse order is used to comfort qcow2 driver: on commit it need to write
4471 * IN_USE flag to the image, to mark bitmaps in the image as invalid. But
4472 * children are usually goes after parents in reopen-queue, so go from last
4475 QTAILQ_FOREACH_REVERSE(bs_entry
, bs_queue
, entry
) {
4476 ctx
= bdrv_get_aio_context(bs_entry
->state
.bs
);
4477 aio_context_acquire(ctx
);
4478 bdrv_reopen_commit(&bs_entry
->state
);
4479 aio_context_release(ctx
);
4484 QTAILQ_FOREACH_REVERSE(bs_entry
, bs_queue
, entry
) {
4485 BlockDriverState
*bs
= bs_entry
->state
.bs
;
4487 if (bs
->drv
->bdrv_reopen_commit_post
) {
4488 ctx
= bdrv_get_aio_context(bs
);
4489 aio_context_acquire(ctx
);
4490 bs
->drv
->bdrv_reopen_commit_post(&bs_entry
->state
);
4491 aio_context_release(ctx
);
4500 QTAILQ_FOREACH_SAFE(bs_entry
, bs_queue
, entry
, next
) {
4501 if (bs_entry
->prepared
) {
4502 ctx
= bdrv_get_aio_context(bs_entry
->state
.bs
);
4503 aio_context_acquire(ctx
);
4504 bdrv_reopen_abort(&bs_entry
->state
);
4505 aio_context_release(ctx
);
4510 bdrv_reopen_queue_free(bs_queue
);
4515 int bdrv_reopen(BlockDriverState
*bs
, QDict
*opts
, bool keep_old_opts
,
4518 AioContext
*ctx
= bdrv_get_aio_context(bs
);
4519 BlockReopenQueue
*queue
;
4522 GLOBAL_STATE_CODE();
4524 queue
= bdrv_reopen_queue(NULL
, bs
, opts
, keep_old_opts
);
4526 if (ctx
!= qemu_get_aio_context()) {
4527 aio_context_release(ctx
);
4529 ret
= bdrv_reopen_multiple(queue
, errp
);
4531 if (ctx
!= qemu_get_aio_context()) {
4532 aio_context_acquire(ctx
);
4538 int bdrv_reopen_set_read_only(BlockDriverState
*bs
, bool read_only
,
4541 QDict
*opts
= qdict_new();
4543 GLOBAL_STATE_CODE();
4545 qdict_put_bool(opts
, BDRV_OPT_READ_ONLY
, read_only
);
4547 return bdrv_reopen(bs
, opts
, true, errp
);
4551 * Take a BDRVReopenState and check if the value of 'backing' in the
4552 * reopen_state->options QDict is valid or not.
4554 * If 'backing' is missing from the QDict then return 0.
4556 * If 'backing' contains the node name of the backing file of
4557 * reopen_state->bs then return 0.
4559 * If 'backing' contains a different node name (or is null) then check
4560 * whether the current backing file can be replaced with the new one.
4561 * If that's the case then reopen_state->replace_backing_bs is set to
4562 * true and reopen_state->new_backing_bs contains a pointer to the new
4563 * backing BlockDriverState (or NULL).
4565 * Return 0 on success, otherwise return < 0 and set @errp.
4567 static int bdrv_reopen_parse_file_or_backing(BDRVReopenState
*reopen_state
,
4568 bool is_backing
, Transaction
*tran
,
4571 BlockDriverState
*bs
= reopen_state
->bs
;
4572 BlockDriverState
*new_child_bs
;
4573 BlockDriverState
*old_child_bs
= is_backing
? child_bs(bs
->backing
) :
4575 const char *child_name
= is_backing
? "backing" : "file";
4579 GLOBAL_STATE_CODE();
4581 value
= qdict_get(reopen_state
->options
, child_name
);
4582 if (value
== NULL
) {
4586 switch (qobject_type(value
)) {
4588 assert(is_backing
); /* The 'file' option does not allow a null value */
4589 new_child_bs
= NULL
;
4592 str
= qstring_get_str(qobject_to(QString
, value
));
4593 new_child_bs
= bdrv_lookup_bs(NULL
, str
, errp
);
4594 if (new_child_bs
== NULL
) {
4596 } else if (bdrv_recurse_has_child(new_child_bs
, bs
)) {
4597 error_setg(errp
, "Making '%s' a %s child of '%s' would create a "
4598 "cycle", str
, child_name
, bs
->node_name
);
4604 * The options QDict has been flattened, so 'backing' and 'file'
4605 * do not allow any other data type here.
4607 g_assert_not_reached();
4610 if (old_child_bs
== new_child_bs
) {
4615 if (bdrv_skip_implicit_filters(old_child_bs
) == new_child_bs
) {
4619 if (old_child_bs
->implicit
) {
4620 error_setg(errp
, "Cannot replace implicit %s child of %s",
4621 child_name
, bs
->node_name
);
4626 if (bs
->drv
->is_filter
&& !old_child_bs
) {
4628 * Filters always have a file or a backing child, so we are trying to
4629 * change wrong child
4631 error_setg(errp
, "'%s' is a %s filter node that does not support a "
4632 "%s child", bs
->node_name
, bs
->drv
->format_name
, child_name
);
4637 reopen_state
->old_backing_bs
= old_child_bs
;
4639 reopen_state
->old_file_bs
= old_child_bs
;
4642 return bdrv_set_file_or_backing_noperm(bs
, new_child_bs
, is_backing
,
4647 * Prepares a BlockDriverState for reopen. All changes are staged in the
4648 * 'opaque' field of the BDRVReopenState, which is used and allocated by
4649 * the block driver layer .bdrv_reopen_prepare()
4651 * bs is the BlockDriverState to reopen
4652 * flags are the new open flags
4653 * queue is the reopen queue
4655 * Returns 0 on success, non-zero on error. On error errp will be set
4658 * On failure, bdrv_reopen_abort() will be called to clean up any data.
4659 * It is the responsibility of the caller to then call the abort() or
4660 * commit() for any other BDS that have been left in a prepare() state
4663 static int bdrv_reopen_prepare(BDRVReopenState
*reopen_state
,
4664 BlockReopenQueue
*queue
,
4665 Transaction
*change_child_tran
, Error
**errp
)
4669 Error
*local_err
= NULL
;
4672 QDict
*orig_reopen_opts
;
4673 char *discard
= NULL
;
4675 bool drv_prepared
= false;
4677 assert(reopen_state
!= NULL
);
4678 assert(reopen_state
->bs
->drv
!= NULL
);
4679 GLOBAL_STATE_CODE();
4680 drv
= reopen_state
->bs
->drv
;
4682 /* This function and each driver's bdrv_reopen_prepare() remove
4683 * entries from reopen_state->options as they are processed, so
4684 * we need to make a copy of the original QDict. */
4685 orig_reopen_opts
= qdict_clone_shallow(reopen_state
->options
);
4687 /* Process generic block layer options */
4688 opts
= qemu_opts_create(&bdrv_runtime_opts
, NULL
, 0, &error_abort
);
4689 if (!qemu_opts_absorb_qdict(opts
, reopen_state
->options
, errp
)) {
4694 /* This was already called in bdrv_reopen_queue_child() so the flags
4695 * are up-to-date. This time we simply want to remove the options from
4696 * QemuOpts in order to indicate that they have been processed. */
4697 old_flags
= reopen_state
->flags
;
4698 update_flags_from_options(&reopen_state
->flags
, opts
);
4699 assert(old_flags
== reopen_state
->flags
);
4701 discard
= qemu_opt_get_del(opts
, BDRV_OPT_DISCARD
);
4702 if (discard
!= NULL
) {
4703 if (bdrv_parse_discard_flags(discard
, &reopen_state
->flags
) != 0) {
4704 error_setg(errp
, "Invalid discard option");
4710 reopen_state
->detect_zeroes
=
4711 bdrv_parse_detect_zeroes(opts
, reopen_state
->flags
, &local_err
);
4713 error_propagate(errp
, local_err
);
4718 /* All other options (including node-name and driver) must be unchanged.
4719 * Put them back into the QDict, so that they are checked at the end
4720 * of this function. */
4721 qemu_opts_to_qdict(opts
, reopen_state
->options
);
4723 /* If we are to stay read-only, do not allow permission change
4724 * to r/w. Attempting to set to r/w may fail if either BDRV_O_ALLOW_RDWR is
4725 * not set, or if the BDS still has copy_on_read enabled */
4726 read_only
= !(reopen_state
->flags
& BDRV_O_RDWR
);
4727 ret
= bdrv_can_set_read_only(reopen_state
->bs
, read_only
, true, &local_err
);
4729 error_propagate(errp
, local_err
);
4733 if (drv
->bdrv_reopen_prepare
) {
4735 * If a driver-specific option is missing, it means that we
4736 * should reset it to its default value.
4737 * But not all options allow that, so we need to check it first.
4739 ret
= bdrv_reset_options_allowed(reopen_state
->bs
,
4740 reopen_state
->options
, errp
);
4745 ret
= drv
->bdrv_reopen_prepare(reopen_state
, queue
, &local_err
);
4747 if (local_err
!= NULL
) {
4748 error_propagate(errp
, local_err
);
4750 bdrv_refresh_filename(reopen_state
->bs
);
4751 error_setg(errp
, "failed while preparing to reopen image '%s'",
4752 reopen_state
->bs
->filename
);
4757 /* It is currently mandatory to have a bdrv_reopen_prepare()
4758 * handler for each supported drv. */
4759 error_setg(errp
, "Block format '%s' used by node '%s' "
4760 "does not support reopening files", drv
->format_name
,
4761 bdrv_get_device_or_node_name(reopen_state
->bs
));
4766 drv_prepared
= true;
4769 * We must provide the 'backing' option if the BDS has a backing
4770 * file or if the image file has a backing file name as part of
4771 * its metadata. Otherwise the 'backing' option can be omitted.
4773 if (drv
->supports_backing
&& reopen_state
->backing_missing
&&
4774 (reopen_state
->bs
->backing
|| reopen_state
->bs
->backing_file
[0])) {
4775 error_setg(errp
, "backing is missing for '%s'",
4776 reopen_state
->bs
->node_name
);
4782 * Allow changing the 'backing' option. The new value can be
4783 * either a reference to an existing node (using its node name)
4784 * or NULL to simply detach the current backing file.
4786 ret
= bdrv_reopen_parse_file_or_backing(reopen_state
, true,
4787 change_child_tran
, errp
);
4791 qdict_del(reopen_state
->options
, "backing");
4793 /* Allow changing the 'file' option. In this case NULL is not allowed */
4794 ret
= bdrv_reopen_parse_file_or_backing(reopen_state
, false,
4795 change_child_tran
, errp
);
4799 qdict_del(reopen_state
->options
, "file");
4801 /* Options that are not handled are only okay if they are unchanged
4802 * compared to the old state. It is expected that some options are only
4803 * used for the initial open, but not reopen (e.g. filename) */
4804 if (qdict_size(reopen_state
->options
)) {
4805 const QDictEntry
*entry
= qdict_first(reopen_state
->options
);
4808 QObject
*new = entry
->value
;
4809 QObject
*old
= qdict_get(reopen_state
->bs
->options
, entry
->key
);
4811 /* Allow child references (child_name=node_name) as long as they
4812 * point to the current child (i.e. everything stays the same). */
4813 if (qobject_type(new) == QTYPE_QSTRING
) {
4815 QLIST_FOREACH(child
, &reopen_state
->bs
->children
, next
) {
4816 if (!strcmp(child
->name
, entry
->key
)) {
4822 if (!strcmp(child
->bs
->node_name
,
4823 qstring_get_str(qobject_to(QString
, new)))) {
4824 continue; /* Found child with this name, skip option */
4830 * TODO: When using -drive to specify blockdev options, all values
4831 * will be strings; however, when using -blockdev, blockdev-add or
4832 * filenames using the json:{} pseudo-protocol, they will be
4834 * In contrast, reopening options are (currently) always strings
4835 * (because you can only specify them through qemu-io; all other
4836 * callers do not specify any options).
4837 * Therefore, when using anything other than -drive to create a BDS,
4838 * this cannot detect non-string options as unchanged, because
4839 * qobject_is_equal() always returns false for objects of different
4840 * type. In the future, this should be remedied by correctly typing
4841 * all options. For now, this is not too big of an issue because
4842 * the user can simply omit options which cannot be changed anyway,
4843 * so they will stay unchanged.
4845 if (!qobject_is_equal(new, old
)) {
4846 error_setg(errp
, "Cannot change the option '%s'", entry
->key
);
4850 } while ((entry
= qdict_next(reopen_state
->options
, entry
)));
4855 /* Restore the original reopen_state->options QDict */
4856 qobject_unref(reopen_state
->options
);
4857 reopen_state
->options
= qobject_ref(orig_reopen_opts
);
4860 if (ret
< 0 && drv_prepared
) {
4861 /* drv->bdrv_reopen_prepare() has succeeded, so we need to
4862 * call drv->bdrv_reopen_abort() before signaling an error
4863 * (bdrv_reopen_multiple() will not call bdrv_reopen_abort()
4864 * when the respective bdrv_reopen_prepare() has failed) */
4865 if (drv
->bdrv_reopen_abort
) {
4866 drv
->bdrv_reopen_abort(reopen_state
);
4869 qemu_opts_del(opts
);
4870 qobject_unref(orig_reopen_opts
);
4876 * Takes the staged changes for the reopen from bdrv_reopen_prepare(), and
4877 * makes them final by swapping the staging BlockDriverState contents into
4878 * the active BlockDriverState contents.
4880 static void bdrv_reopen_commit(BDRVReopenState
*reopen_state
)
4883 BlockDriverState
*bs
;
4886 assert(reopen_state
!= NULL
);
4887 bs
= reopen_state
->bs
;
4889 assert(drv
!= NULL
);
4890 GLOBAL_STATE_CODE();
4892 /* If there are any driver level actions to take */
4893 if (drv
->bdrv_reopen_commit
) {
4894 drv
->bdrv_reopen_commit(reopen_state
);
4897 /* set BDS specific flags now */
4898 qobject_unref(bs
->explicit_options
);
4899 qobject_unref(bs
->options
);
4900 qobject_ref(reopen_state
->explicit_options
);
4901 qobject_ref(reopen_state
->options
);
4903 bs
->explicit_options
= reopen_state
->explicit_options
;
4904 bs
->options
= reopen_state
->options
;
4905 bs
->open_flags
= reopen_state
->flags
;
4906 bs
->detect_zeroes
= reopen_state
->detect_zeroes
;
4908 /* Remove child references from bs->options and bs->explicit_options.
4909 * Child options were already removed in bdrv_reopen_queue_child() */
4910 QLIST_FOREACH(child
, &bs
->children
, next
) {
4911 qdict_del(bs
->explicit_options
, child
->name
);
4912 qdict_del(bs
->options
, child
->name
);
4914 /* backing is probably removed, so it's not handled by previous loop */
4915 qdict_del(bs
->explicit_options
, "backing");
4916 qdict_del(bs
->options
, "backing");
4918 bdrv_refresh_limits(bs
, NULL
, NULL
);
4922 * Abort the reopen, and delete and free the staged changes in
4925 static void bdrv_reopen_abort(BDRVReopenState
*reopen_state
)
4929 assert(reopen_state
!= NULL
);
4930 drv
= reopen_state
->bs
->drv
;
4931 assert(drv
!= NULL
);
4932 GLOBAL_STATE_CODE();
4934 if (drv
->bdrv_reopen_abort
) {
4935 drv
->bdrv_reopen_abort(reopen_state
);
4940 static void bdrv_close(BlockDriverState
*bs
)
4942 BdrvAioNotifier
*ban
, *ban_next
;
4943 BdrvChild
*child
, *next
;
4945 GLOBAL_STATE_CODE();
4946 assert(!bs
->refcnt
);
4948 bdrv_drained_begin(bs
); /* complete I/O */
4950 bdrv_drain(bs
); /* in case flush left pending I/O */
4953 if (bs
->drv
->bdrv_close
) {
4954 /* Must unfreeze all children, so bdrv_unref_child() works */
4955 bs
->drv
->bdrv_close(bs
);
4960 QLIST_FOREACH_SAFE(child
, &bs
->children
, next
, next
) {
4961 bdrv_unref_child(bs
, child
);
4964 assert(!bs
->backing
);
4968 qatomic_set(&bs
->copy_on_read
, 0);
4969 bs
->backing_file
[0] = '\0';
4970 bs
->backing_format
[0] = '\0';
4971 bs
->total_sectors
= 0;
4972 bs
->encrypted
= false;
4974 qobject_unref(bs
->options
);
4975 qobject_unref(bs
->explicit_options
);
4977 bs
->explicit_options
= NULL
;
4978 qobject_unref(bs
->full_open_options
);
4979 bs
->full_open_options
= NULL
;
4980 g_free(bs
->block_status_cache
);
4981 bs
->block_status_cache
= NULL
;
4983 bdrv_release_named_dirty_bitmaps(bs
);
4984 assert(QLIST_EMPTY(&bs
->dirty_bitmaps
));
4986 QLIST_FOREACH_SAFE(ban
, &bs
->aio_notifiers
, list
, ban_next
) {
4989 QLIST_INIT(&bs
->aio_notifiers
);
4990 bdrv_drained_end(bs
);
4993 * If we're still inside some bdrv_drain_all_begin()/end() sections, end
4994 * them now since this BDS won't exist anymore when bdrv_drain_all_end()
4997 if (bs
->quiesce_counter
) {
4998 bdrv_drain_all_end_quiesce(bs
);
5002 void bdrv_close_all(void)
5004 GLOBAL_STATE_CODE();
5005 assert(job_next(NULL
) == NULL
);
5007 /* Drop references from requests still in flight, such as canceled block
5008 * jobs whose AIO context has not been polled yet */
5011 blk_remove_all_bs();
5012 blockdev_close_all_bdrv_states();
5014 assert(QTAILQ_EMPTY(&all_bdrv_states
));
5017 static bool should_update_child(BdrvChild
*c
, BlockDriverState
*to
)
5023 if (c
->klass
->stay_at_node
) {
5027 /* If the child @c belongs to the BDS @to, replacing the current
5028 * c->bs by @to would mean to create a loop.
5030 * Such a case occurs when appending a BDS to a backing chain.
5031 * For instance, imagine the following chain:
5033 * guest device -> node A -> further backing chain...
5035 * Now we create a new BDS B which we want to put on top of this
5036 * chain, so we first attach A as its backing node:
5041 * guest device -> node A -> further backing chain...
5043 * Finally we want to replace A by B. When doing that, we want to
5044 * replace all pointers to A by pointers to B -- except for the
5045 * pointer from B because (1) that would create a loop, and (2)
5046 * that pointer should simply stay intact:
5048 * guest device -> node B
5051 * node A -> further backing chain...
5053 * In general, when replacing a node A (c->bs) by a node B (@to),
5054 * if A is a child of B, that means we cannot replace A by B there
5055 * because that would create a loop. Silently detaching A from B
5056 * is also not really an option. So overall just leaving A in
5057 * place there is the most sensible choice.
5059 * We would also create a loop in any cases where @c is only
5060 * indirectly referenced by @to. Prevent this by returning false
5061 * if @c is found (by breadth-first search) anywhere in the whole
5066 found
= g_hash_table_new(NULL
, NULL
);
5067 g_hash_table_add(found
, to
);
5068 queue
= g_queue_new();
5069 g_queue_push_tail(queue
, to
);
5071 while (!g_queue_is_empty(queue
)) {
5072 BlockDriverState
*v
= g_queue_pop_head(queue
);
5075 QLIST_FOREACH(c2
, &v
->children
, next
) {
5081 if (g_hash_table_contains(found
, c2
->bs
)) {
5085 g_queue_push_tail(queue
, c2
->bs
);
5086 g_hash_table_add(found
, c2
->bs
);
5090 g_queue_free(queue
);
5091 g_hash_table_destroy(found
);
5096 static void bdrv_remove_child_commit(void *opaque
)
5098 GLOBAL_STATE_CODE();
5099 bdrv_child_free(opaque
);
5102 static TransactionActionDrv bdrv_remove_child_drv
= {
5103 .commit
= bdrv_remove_child_commit
,
5106 /* Function doesn't update permissions, caller is responsible for this. */
5107 static void bdrv_remove_child(BdrvChild
*child
, Transaction
*tran
)
5114 BlockDriverState
*bs
= child
->bs
;
5115 bdrv_drained_begin(bs
);
5116 bdrv_replace_child_tran(child
, NULL
, tran
);
5117 bdrv_drained_end(bs
);
5120 tran_add(tran
, &bdrv_remove_child_drv
, child
);
5123 static void undrain_on_clean_cb(void *opaque
)
5125 bdrv_drained_end(opaque
);
5128 static TransactionActionDrv undrain_on_clean
= {
5129 .clean
= undrain_on_clean_cb
,
5132 static int bdrv_replace_node_noperm(BlockDriverState
*from
,
5133 BlockDriverState
*to
,
5134 bool auto_skip
, Transaction
*tran
,
5137 BdrvChild
*c
, *next
;
5139 GLOBAL_STATE_CODE();
5141 bdrv_drained_begin(from
);
5142 bdrv_drained_begin(to
);
5143 tran_add(tran
, &undrain_on_clean
, from
);
5144 tran_add(tran
, &undrain_on_clean
, to
);
5146 QLIST_FOREACH_SAFE(c
, &from
->parents
, next_parent
, next
) {
5147 assert(c
->bs
== from
);
5148 if (!should_update_child(c
, to
)) {
5152 error_setg(errp
, "Should not change '%s' link to '%s'",
5153 c
->name
, from
->node_name
);
5157 error_setg(errp
, "Cannot change '%s' link to '%s'",
5158 c
->name
, from
->node_name
);
5161 bdrv_replace_child_tran(c
, to
, tran
);
5168 * With auto_skip=true bdrv_replace_node_common skips updating from parents
5169 * if it creates a parent-child relation loop or if parent is block-job.
5171 * With auto_skip=false the error is returned if from has a parent which should
5174 * With @detach_subchain=true @to must be in a backing chain of @from. In this
5175 * case backing link of the cow-parent of @to is removed.
5177 static int bdrv_replace_node_common(BlockDriverState
*from
,
5178 BlockDriverState
*to
,
5179 bool auto_skip
, bool detach_subchain
,
5182 Transaction
*tran
= tran_new();
5183 g_autoptr(GSList
) refresh_list
= NULL
;
5184 BlockDriverState
*to_cow_parent
= NULL
;
5187 GLOBAL_STATE_CODE();
5189 if (detach_subchain
) {
5190 assert(bdrv_chain_contains(from
, to
));
5192 for (to_cow_parent
= from
;
5193 bdrv_filter_or_cow_bs(to_cow_parent
) != to
;
5194 to_cow_parent
= bdrv_filter_or_cow_bs(to_cow_parent
))
5200 /* Make sure that @from doesn't go away until we have successfully attached
5201 * all of its parents to @to. */
5204 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
5205 assert(bdrv_get_aio_context(from
) == bdrv_get_aio_context(to
));
5206 bdrv_drained_begin(from
);
5209 * Do the replacement without permission update.
5210 * Replacement may influence the permissions, we should calculate new
5211 * permissions based on new graph. If we fail, we'll roll-back the
5214 ret
= bdrv_replace_node_noperm(from
, to
, auto_skip
, tran
, errp
);
5219 if (detach_subchain
) {
5220 bdrv_remove_child(bdrv_filter_or_cow_child(to_cow_parent
), tran
);
5223 refresh_list
= g_slist_prepend(refresh_list
, to
);
5224 refresh_list
= g_slist_prepend(refresh_list
, from
);
5226 ret
= bdrv_list_refresh_perms(refresh_list
, NULL
, tran
, errp
);
5234 tran_finalize(tran
, ret
);
5236 bdrv_drained_end(from
);
5242 int bdrv_replace_node(BlockDriverState
*from
, BlockDriverState
*to
,
5245 GLOBAL_STATE_CODE();
5247 return bdrv_replace_node_common(from
, to
, true, false, errp
);
5250 int bdrv_drop_filter(BlockDriverState
*bs
, Error
**errp
)
5252 GLOBAL_STATE_CODE();
5254 return bdrv_replace_node_common(bs
, bdrv_filter_or_cow_bs(bs
), true, true,
5259 * Add new bs contents at the top of an image chain while the chain is
5260 * live, while keeping required fields on the top layer.
5262 * This will modify the BlockDriverState fields, and swap contents
5263 * between bs_new and bs_top. Both bs_new and bs_top are modified.
5265 * bs_new must not be attached to a BlockBackend and must not have backing
5268 * This function does not create any image files.
5270 * The caller must hold the AioContext lock for @bs_top.
5272 int bdrv_append(BlockDriverState
*bs_new
, BlockDriverState
*bs_top
,
5277 Transaction
*tran
= tran_new();
5278 AioContext
*old_context
, *new_context
= NULL
;
5280 GLOBAL_STATE_CODE();
5282 assert(!bs_new
->backing
);
5284 old_context
= bdrv_get_aio_context(bs_top
);
5286 child
= bdrv_attach_child_noperm(bs_new
, bs_top
, "backing",
5287 &child_of_bds
, bdrv_backing_role(bs_new
),
5295 * bdrv_attach_child_noperm could change the AioContext of bs_top.
5296 * bdrv_replace_node_noperm calls bdrv_drained_begin, so let's temporarily
5297 * hold the new AioContext, since bdrv_drained_begin calls BDRV_POLL_WHILE
5298 * that assumes the new lock is taken.
5300 new_context
= bdrv_get_aio_context(bs_top
);
5302 if (old_context
!= new_context
) {
5303 aio_context_release(old_context
);
5304 aio_context_acquire(new_context
);
5307 ret
= bdrv_replace_node_noperm(bs_top
, bs_new
, true, tran
, errp
);
5312 ret
= bdrv_refresh_perms(bs_new
, tran
, errp
);
5314 tran_finalize(tran
, ret
);
5316 bdrv_refresh_limits(bs_top
, NULL
, NULL
);
5318 if (new_context
&& old_context
!= new_context
) {
5319 aio_context_release(new_context
);
5320 aio_context_acquire(old_context
);
5326 /* Not for empty child */
5327 int bdrv_replace_child_bs(BdrvChild
*child
, BlockDriverState
*new_bs
,
5331 Transaction
*tran
= tran_new();
5332 g_autoptr(GSList
) refresh_list
= NULL
;
5333 BlockDriverState
*old_bs
= child
->bs
;
5335 GLOBAL_STATE_CODE();
5338 bdrv_drained_begin(old_bs
);
5339 bdrv_drained_begin(new_bs
);
5341 bdrv_replace_child_tran(child
, new_bs
, tran
);
5343 refresh_list
= g_slist_prepend(refresh_list
, old_bs
);
5344 refresh_list
= g_slist_prepend(refresh_list
, new_bs
);
5346 ret
= bdrv_list_refresh_perms(refresh_list
, NULL
, tran
, errp
);
5348 tran_finalize(tran
, ret
);
5350 bdrv_drained_end(old_bs
);
5351 bdrv_drained_end(new_bs
);
5357 static void bdrv_delete(BlockDriverState
*bs
)
5359 assert(bdrv_op_blocker_is_empty(bs
));
5360 assert(!bs
->refcnt
);
5361 GLOBAL_STATE_CODE();
5363 /* remove from list, if necessary */
5364 if (bs
->node_name
[0] != '\0') {
5365 QTAILQ_REMOVE(&graph_bdrv_states
, bs
, node_list
);
5367 QTAILQ_REMOVE(&all_bdrv_states
, bs
, bs_list
);
5376 * Replace @bs by newly created block node.
5378 * @options is a QDict of options to pass to the block drivers, or NULL for an
5379 * empty set of options. The reference to the QDict belongs to the block layer
5380 * after the call (even on failure), so if the caller intends to reuse the
5381 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
5383 BlockDriverState
*bdrv_insert_node(BlockDriverState
*bs
, QDict
*options
,
5384 int flags
, Error
**errp
)
5388 BlockDriverState
*new_node_bs
= NULL
;
5389 const char *drvname
, *node_name
;
5392 drvname
= qdict_get_try_str(options
, "driver");
5394 error_setg(errp
, "driver is not specified");
5398 drv
= bdrv_find_format(drvname
);
5400 error_setg(errp
, "Unknown driver: '%s'", drvname
);
5404 node_name
= qdict_get_try_str(options
, "node-name");
5406 GLOBAL_STATE_CODE();
5408 new_node_bs
= bdrv_new_open_driver_opts(drv
, node_name
, options
, flags
,
5410 options
= NULL
; /* bdrv_new_open_driver() eats options */
5412 error_prepend(errp
, "Could not create node: ");
5416 bdrv_drained_begin(bs
);
5417 ret
= bdrv_replace_node(bs
, new_node_bs
, errp
);
5418 bdrv_drained_end(bs
);
5421 error_prepend(errp
, "Could not replace node: ");
5428 qobject_unref(options
);
5429 bdrv_unref(new_node_bs
);
5434 * Run consistency checks on an image
5436 * Returns 0 if the check could be completed (it doesn't mean that the image is
5437 * free of errors) or -errno when an internal error occurred. The results of the
5438 * check are stored in res.
5440 int coroutine_fn
bdrv_co_check(BlockDriverState
*bs
,
5441 BdrvCheckResult
*res
, BdrvCheckMode fix
)
5444 assert_bdrv_graph_readable();
5445 if (bs
->drv
== NULL
) {
5448 if (bs
->drv
->bdrv_co_check
== NULL
) {
5452 memset(res
, 0, sizeof(*res
));
5453 return bs
->drv
->bdrv_co_check(bs
, res
, fix
);
5459 * -EINVAL - backing format specified, but no file
5460 * -ENOSPC - can't update the backing file because no space is left in the
5462 * -ENOTSUP - format driver doesn't support changing the backing file
5464 int bdrv_change_backing_file(BlockDriverState
*bs
, const char *backing_file
,
5465 const char *backing_fmt
, bool require
)
5467 BlockDriver
*drv
= bs
->drv
;
5470 GLOBAL_STATE_CODE();
5476 /* Backing file format doesn't make sense without a backing file */
5477 if (backing_fmt
&& !backing_file
) {
5481 if (require
&& backing_file
&& !backing_fmt
) {
5485 if (drv
->bdrv_change_backing_file
!= NULL
) {
5486 ret
= drv
->bdrv_change_backing_file(bs
, backing_file
, backing_fmt
);
5492 pstrcpy(bs
->backing_file
, sizeof(bs
->backing_file
), backing_file
?: "");
5493 pstrcpy(bs
->backing_format
, sizeof(bs
->backing_format
), backing_fmt
?: "");
5494 pstrcpy(bs
->auto_backing_file
, sizeof(bs
->auto_backing_file
),
5495 backing_file
?: "");
5501 * Finds the first non-filter node above bs in the chain between
5502 * active and bs. The returned node is either an immediate parent of
5503 * bs, or there are only filter nodes between the two.
5505 * Returns NULL if bs is not found in active's image chain,
5506 * or if active == bs.
5508 * Returns the bottommost base image if bs == NULL.
5510 BlockDriverState
*bdrv_find_overlay(BlockDriverState
*active
,
5511 BlockDriverState
*bs
)
5514 GLOBAL_STATE_CODE();
5516 bs
= bdrv_skip_filters(bs
);
5517 active
= bdrv_skip_filters(active
);
5520 BlockDriverState
*next
= bdrv_backing_chain_next(active
);
5530 /* Given a BDS, searches for the base layer. */
5531 BlockDriverState
*bdrv_find_base(BlockDriverState
*bs
)
5533 GLOBAL_STATE_CODE();
5535 return bdrv_find_overlay(bs
, NULL
);
5539 * Return true if at least one of the COW (backing) and filter links
5540 * between @bs and @base is frozen. @errp is set if that's the case.
5541 * @base must be reachable from @bs, or NULL.
5543 bool bdrv_is_backing_chain_frozen(BlockDriverState
*bs
, BlockDriverState
*base
,
5546 BlockDriverState
*i
;
5549 GLOBAL_STATE_CODE();
5551 for (i
= bs
; i
!= base
; i
= child_bs(child
)) {
5552 child
= bdrv_filter_or_cow_child(i
);
5554 if (child
&& child
->frozen
) {
5555 error_setg(errp
, "Cannot change '%s' link from '%s' to '%s'",
5556 child
->name
, i
->node_name
, child
->bs
->node_name
);
5565 * Freeze all COW (backing) and filter links between @bs and @base.
5566 * If any of the links is already frozen the operation is aborted and
5567 * none of the links are modified.
5568 * @base must be reachable from @bs, or NULL.
5569 * Returns 0 on success. On failure returns < 0 and sets @errp.
5571 int bdrv_freeze_backing_chain(BlockDriverState
*bs
, BlockDriverState
*base
,
5574 BlockDriverState
*i
;
5577 GLOBAL_STATE_CODE();
5579 if (bdrv_is_backing_chain_frozen(bs
, base
, errp
)) {
5583 for (i
= bs
; i
!= base
; i
= child_bs(child
)) {
5584 child
= bdrv_filter_or_cow_child(i
);
5585 if (child
&& child
->bs
->never_freeze
) {
5586 error_setg(errp
, "Cannot freeze '%s' link to '%s'",
5587 child
->name
, child
->bs
->node_name
);
5592 for (i
= bs
; i
!= base
; i
= child_bs(child
)) {
5593 child
= bdrv_filter_or_cow_child(i
);
5595 child
->frozen
= true;
5603 * Unfreeze all COW (backing) and filter links between @bs and @base.
5604 * The caller must ensure that all links are frozen before using this
5606 * @base must be reachable from @bs, or NULL.
5608 void bdrv_unfreeze_backing_chain(BlockDriverState
*bs
, BlockDriverState
*base
)
5610 BlockDriverState
*i
;
5613 GLOBAL_STATE_CODE();
5615 for (i
= bs
; i
!= base
; i
= child_bs(child
)) {
5616 child
= bdrv_filter_or_cow_child(i
);
5618 assert(child
->frozen
);
5619 child
->frozen
= false;
5625 * Drops images above 'base' up to and including 'top', and sets the image
5626 * above 'top' to have base as its backing file.
5628 * Requires that the overlay to 'top' is opened r/w, so that the backing file
5629 * information in 'bs' can be properly updated.
5631 * E.g., this will convert the following chain:
5632 * bottom <- base <- intermediate <- top <- active
5636 * bottom <- base <- active
5638 * It is allowed for bottom==base, in which case it converts:
5640 * base <- intermediate <- top <- active
5646 * If backing_file_str is non-NULL, it will be used when modifying top's
5647 * overlay image metadata.
5650 * if active == top, that is considered an error
5653 int bdrv_drop_intermediate(BlockDriverState
*top
, BlockDriverState
*base
,
5654 const char *backing_file_str
)
5656 BlockDriverState
*explicit_top
= top
;
5657 bool update_inherits_from
;
5659 Error
*local_err
= NULL
;
5661 g_autoptr(GSList
) updated_children
= NULL
;
5664 GLOBAL_STATE_CODE();
5667 bdrv_drained_begin(base
);
5669 if (!top
->drv
|| !base
->drv
) {
5673 /* Make sure that base is in the backing chain of top */
5674 if (!bdrv_chain_contains(top
, base
)) {
5678 /* If 'base' recursively inherits from 'top' then we should set
5679 * base->inherits_from to top->inherits_from after 'top' and all
5680 * other intermediate nodes have been dropped.
5681 * If 'top' is an implicit node (e.g. "commit_top") we should skip
5682 * it because no one inherits from it. We use explicit_top for that. */
5683 explicit_top
= bdrv_skip_implicit_filters(explicit_top
);
5684 update_inherits_from
= bdrv_inherits_from_recursive(base
, explicit_top
);
5686 /* success - we can delete the intermediate states, and link top->base */
5687 if (!backing_file_str
) {
5688 bdrv_refresh_filename(base
);
5689 backing_file_str
= base
->filename
;
5692 QLIST_FOREACH(c
, &top
->parents
, next_parent
) {
5693 updated_children
= g_slist_prepend(updated_children
, c
);
5697 * It seems correct to pass detach_subchain=true here, but it triggers
5698 * one more yet not fixed bug, when due to nested aio_poll loop we switch to
5699 * another drained section, which modify the graph (for example, removing
5700 * the child, which we keep in updated_children list). So, it's a TODO.
5702 * Note, bug triggered if pass detach_subchain=true here and run
5703 * test-bdrv-drain. test_drop_intermediate_poll() test-case will crash.
5706 bdrv_replace_node_common(top
, base
, false, false, &local_err
);
5708 error_report_err(local_err
);
5712 for (p
= updated_children
; p
; p
= p
->next
) {
5715 if (c
->klass
->update_filename
) {
5716 ret
= c
->klass
->update_filename(c
, base
, backing_file_str
,
5720 * TODO: Actually, we want to rollback all previous iterations
5721 * of this loop, and (which is almost impossible) previous
5722 * bdrv_replace_node()...
5724 * Note, that c->klass->update_filename may lead to permission
5725 * update, so it's a bad idea to call it inside permission
5726 * update transaction of bdrv_replace_node.
5728 error_report_err(local_err
);
5734 if (update_inherits_from
) {
5735 base
->inherits_from
= explicit_top
->inherits_from
;
5740 bdrv_drained_end(base
);
5746 * Implementation of BlockDriver.bdrv_co_get_allocated_file_size() that
5747 * sums the size of all data-bearing children. (This excludes backing
5750 static int64_t bdrv_sum_allocated_file_size(BlockDriverState
*bs
)
5753 int64_t child_size
, sum
= 0;
5755 QLIST_FOREACH(child
, &bs
->children
, next
) {
5756 if (child
->role
& (BDRV_CHILD_DATA
| BDRV_CHILD_METADATA
|
5757 BDRV_CHILD_FILTERED
))
5759 child_size
= bdrv_co_get_allocated_file_size(child
->bs
);
5760 if (child_size
< 0) {
5771 * Length of a allocated file in bytes. Sparse files are counted by actual
5772 * allocated space. Return < 0 if error or unknown.
5774 int64_t coroutine_fn
bdrv_co_get_allocated_file_size(BlockDriverState
*bs
)
5776 BlockDriver
*drv
= bs
->drv
;
5782 if (drv
->bdrv_co_get_allocated_file_size
) {
5783 return drv
->bdrv_co_get_allocated_file_size(bs
);
5786 if (drv
->bdrv_file_open
) {
5788 * Protocol drivers default to -ENOTSUP (most of their data is
5789 * not stored in any of their children (if they even have any),
5790 * so there is no generic way to figure it out).
5793 } else if (drv
->is_filter
) {
5794 /* Filter drivers default to the size of their filtered child */
5795 return bdrv_co_get_allocated_file_size(bdrv_filter_bs(bs
));
5797 /* Other drivers default to summing their children's sizes */
5798 return bdrv_sum_allocated_file_size(bs
);
5804 * @drv: Format driver
5805 * @opts: Creation options for new image
5806 * @in_bs: Existing image containing data for new image (may be NULL)
5807 * @errp: Error object
5808 * Returns: A #BlockMeasureInfo (free using qapi_free_BlockMeasureInfo())
5811 * Calculate file size required to create a new image.
5813 * If @in_bs is given then space for allocated clusters and zero clusters
5814 * from that image are included in the calculation. If @opts contains a
5815 * backing file that is shared by @in_bs then backing clusters may be omitted
5816 * from the calculation.
5818 * If @in_bs is NULL then the calculation includes no allocated clusters
5819 * unless a preallocation option is given in @opts.
5821 * Note that @in_bs may use a different BlockDriver from @drv.
5823 * If an error occurs the @errp pointer is set.
5825 BlockMeasureInfo
*bdrv_measure(BlockDriver
*drv
, QemuOpts
*opts
,
5826 BlockDriverState
*in_bs
, Error
**errp
)
5829 if (!drv
->bdrv_measure
) {
5830 error_setg(errp
, "Block driver '%s' does not support size measurement",
5835 return drv
->bdrv_measure(opts
, in_bs
, errp
);
5839 * Return number of sectors on success, -errno on error.
5841 int64_t coroutine_fn
bdrv_co_nb_sectors(BlockDriverState
*bs
)
5843 BlockDriver
*drv
= bs
->drv
;
5849 if (drv
->has_variable_length
) {
5850 int ret
= bdrv_co_refresh_total_sectors(bs
, bs
->total_sectors
);
5855 return bs
->total_sectors
;
5859 * Return length in bytes on success, -errno on error.
5860 * The length is always a multiple of BDRV_SECTOR_SIZE.
5862 int64_t coroutine_fn
bdrv_co_getlength(BlockDriverState
*bs
)
5867 ret
= bdrv_co_nb_sectors(bs
);
5871 if (ret
> INT64_MAX
/ BDRV_SECTOR_SIZE
) {
5874 return ret
* BDRV_SECTOR_SIZE
;
5877 /* return 0 as number of sectors if no device present or error */
5878 void bdrv_get_geometry(BlockDriverState
*bs
, uint64_t *nb_sectors_ptr
)
5880 int64_t nb_sectors
= bdrv_nb_sectors(bs
);
5883 *nb_sectors_ptr
= nb_sectors
< 0 ? 0 : nb_sectors
;
5886 bool bdrv_is_sg(BlockDriverState
*bs
)
5893 * Return whether the given node supports compressed writes.
5895 bool bdrv_supports_compressed_writes(BlockDriverState
*bs
)
5897 BlockDriverState
*filtered
;
5900 if (!bs
->drv
|| !block_driver_can_compress(bs
->drv
)) {
5904 filtered
= bdrv_filter_bs(bs
);
5907 * Filters can only forward compressed writes, so we have to
5910 return bdrv_supports_compressed_writes(filtered
);
5916 const char *bdrv_get_format_name(BlockDriverState
*bs
)
5919 return bs
->drv
? bs
->drv
->format_name
: NULL
;
5922 static int qsort_strcmp(const void *a
, const void *b
)
5924 return strcmp(*(char *const *)a
, *(char *const *)b
);
5927 void bdrv_iterate_format(void (*it
)(void *opaque
, const char *name
),
5928 void *opaque
, bool read_only
)
5933 const char **formats
= NULL
;
5935 GLOBAL_STATE_CODE();
5937 QLIST_FOREACH(drv
, &bdrv_drivers
, list
) {
5938 if (drv
->format_name
) {
5942 if (use_bdrv_whitelist
&& !bdrv_is_whitelisted(drv
, read_only
)) {
5946 while (formats
&& i
&& !found
) {
5947 found
= !strcmp(formats
[--i
], drv
->format_name
);
5951 formats
= g_renew(const char *, formats
, count
+ 1);
5952 formats
[count
++] = drv
->format_name
;
5957 for (i
= 0; i
< (int)ARRAY_SIZE(block_driver_modules
); i
++) {
5958 const char *format_name
= block_driver_modules
[i
].format_name
;
5964 if (use_bdrv_whitelist
&&
5965 !bdrv_format_is_whitelisted(format_name
, read_only
)) {
5969 while (formats
&& j
&& !found
) {
5970 found
= !strcmp(formats
[--j
], format_name
);
5974 formats
= g_renew(const char *, formats
, count
+ 1);
5975 formats
[count
++] = format_name
;
5980 qsort(formats
, count
, sizeof(formats
[0]), qsort_strcmp
);
5982 for (i
= 0; i
< count
; i
++) {
5983 it(opaque
, formats
[i
]);
5989 /* This function is to find a node in the bs graph */
5990 BlockDriverState
*bdrv_find_node(const char *node_name
)
5992 BlockDriverState
*bs
;
5995 GLOBAL_STATE_CODE();
5997 QTAILQ_FOREACH(bs
, &graph_bdrv_states
, node_list
) {
5998 if (!strcmp(node_name
, bs
->node_name
)) {
6005 /* Put this QMP function here so it can access the static graph_bdrv_states. */
6006 BlockDeviceInfoList
*bdrv_named_nodes_list(bool flat
,
6009 BlockDeviceInfoList
*list
;
6010 BlockDriverState
*bs
;
6012 GLOBAL_STATE_CODE();
6015 QTAILQ_FOREACH(bs
, &graph_bdrv_states
, node_list
) {
6016 BlockDeviceInfo
*info
= bdrv_block_device_info(NULL
, bs
, flat
, errp
);
6018 qapi_free_BlockDeviceInfoList(list
);
6021 QAPI_LIST_PREPEND(list
, info
);
6027 typedef struct XDbgBlockGraphConstructor
{
6028 XDbgBlockGraph
*graph
;
6029 GHashTable
*graph_nodes
;
6030 } XDbgBlockGraphConstructor
;
6032 static XDbgBlockGraphConstructor
*xdbg_graph_new(void)
6034 XDbgBlockGraphConstructor
*gr
= g_new(XDbgBlockGraphConstructor
, 1);
6036 gr
->graph
= g_new0(XDbgBlockGraph
, 1);
6037 gr
->graph_nodes
= g_hash_table_new(NULL
, NULL
);
6042 static XDbgBlockGraph
*xdbg_graph_finalize(XDbgBlockGraphConstructor
*gr
)
6044 XDbgBlockGraph
*graph
= gr
->graph
;
6046 g_hash_table_destroy(gr
->graph_nodes
);
6052 static uintptr_t xdbg_graph_node_num(XDbgBlockGraphConstructor
*gr
, void *node
)
6054 uintptr_t ret
= (uintptr_t)g_hash_table_lookup(gr
->graph_nodes
, node
);
6061 * Start counting from 1, not 0, because 0 interferes with not-found (NULL)
6062 * answer of g_hash_table_lookup.
6064 ret
= g_hash_table_size(gr
->graph_nodes
) + 1;
6065 g_hash_table_insert(gr
->graph_nodes
, node
, (void *)ret
);
6070 static void xdbg_graph_add_node(XDbgBlockGraphConstructor
*gr
, void *node
,
6071 XDbgBlockGraphNodeType type
, const char *name
)
6073 XDbgBlockGraphNode
*n
;
6075 n
= g_new0(XDbgBlockGraphNode
, 1);
6077 n
->id
= xdbg_graph_node_num(gr
, node
);
6079 n
->name
= g_strdup(name
);
6081 QAPI_LIST_PREPEND(gr
->graph
->nodes
, n
);
6084 static void xdbg_graph_add_edge(XDbgBlockGraphConstructor
*gr
, void *parent
,
6085 const BdrvChild
*child
)
6087 BlockPermission qapi_perm
;
6088 XDbgBlockGraphEdge
*edge
;
6089 GLOBAL_STATE_CODE();
6091 edge
= g_new0(XDbgBlockGraphEdge
, 1);
6093 edge
->parent
= xdbg_graph_node_num(gr
, parent
);
6094 edge
->child
= xdbg_graph_node_num(gr
, child
->bs
);
6095 edge
->name
= g_strdup(child
->name
);
6097 for (qapi_perm
= 0; qapi_perm
< BLOCK_PERMISSION__MAX
; qapi_perm
++) {
6098 uint64_t flag
= bdrv_qapi_perm_to_blk_perm(qapi_perm
);
6100 if (flag
& child
->perm
) {
6101 QAPI_LIST_PREPEND(edge
->perm
, qapi_perm
);
6103 if (flag
& child
->shared_perm
) {
6104 QAPI_LIST_PREPEND(edge
->shared_perm
, qapi_perm
);
6108 QAPI_LIST_PREPEND(gr
->graph
->edges
, edge
);
6112 XDbgBlockGraph
*bdrv_get_xdbg_block_graph(Error
**errp
)
6116 BlockDriverState
*bs
;
6118 XDbgBlockGraphConstructor
*gr
= xdbg_graph_new();
6120 GLOBAL_STATE_CODE();
6122 for (blk
= blk_all_next(NULL
); blk
; blk
= blk_all_next(blk
)) {
6123 char *allocated_name
= NULL
;
6124 const char *name
= blk_name(blk
);
6127 name
= allocated_name
= blk_get_attached_dev_id(blk
);
6129 xdbg_graph_add_node(gr
, blk
, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_BACKEND
,
6131 g_free(allocated_name
);
6132 if (blk_root(blk
)) {
6133 xdbg_graph_add_edge(gr
, blk
, blk_root(blk
));
6137 WITH_JOB_LOCK_GUARD() {
6138 for (job
= block_job_next_locked(NULL
); job
;
6139 job
= block_job_next_locked(job
)) {
6142 xdbg_graph_add_node(gr
, job
, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_JOB
,
6144 for (el
= job
->nodes
; el
; el
= el
->next
) {
6145 xdbg_graph_add_edge(gr
, job
, (BdrvChild
*)el
->data
);
6150 QTAILQ_FOREACH(bs
, &graph_bdrv_states
, node_list
) {
6151 xdbg_graph_add_node(gr
, bs
, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_DRIVER
,
6153 QLIST_FOREACH(child
, &bs
->children
, next
) {
6154 xdbg_graph_add_edge(gr
, bs
, child
);
6158 return xdbg_graph_finalize(gr
);
6161 BlockDriverState
*bdrv_lookup_bs(const char *device
,
6162 const char *node_name
,
6166 BlockDriverState
*bs
;
6168 GLOBAL_STATE_CODE();
6171 blk
= blk_by_name(device
);
6176 error_setg(errp
, "Device '%s' has no medium", device
);
6184 bs
= bdrv_find_node(node_name
);
6191 error_setg(errp
, "Cannot find device=\'%s\' nor node-name=\'%s\'",
6192 device
? device
: "",
6193 node_name
? node_name
: "");
6197 /* If 'base' is in the same chain as 'top', return true. Otherwise,
6198 * return false. If either argument is NULL, return false. */
6199 bool bdrv_chain_contains(BlockDriverState
*top
, BlockDriverState
*base
)
6202 GLOBAL_STATE_CODE();
6204 while (top
&& top
!= base
) {
6205 top
= bdrv_filter_or_cow_bs(top
);
6211 BlockDriverState
*bdrv_next_node(BlockDriverState
*bs
)
6213 GLOBAL_STATE_CODE();
6215 return QTAILQ_FIRST(&graph_bdrv_states
);
6217 return QTAILQ_NEXT(bs
, node_list
);
6220 BlockDriverState
*bdrv_next_all_states(BlockDriverState
*bs
)
6222 GLOBAL_STATE_CODE();
6224 return QTAILQ_FIRST(&all_bdrv_states
);
6226 return QTAILQ_NEXT(bs
, bs_list
);
6229 const char *bdrv_get_node_name(const BlockDriverState
*bs
)
6232 return bs
->node_name
;
6235 const char *bdrv_get_parent_name(const BlockDriverState
*bs
)
6241 /* If multiple parents have a name, just pick the first one. */
6242 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
6243 if (c
->klass
->get_name
) {
6244 name
= c
->klass
->get_name(c
);
6245 if (name
&& *name
) {
6254 /* TODO check what callers really want: bs->node_name or blk_name() */
6255 const char *bdrv_get_device_name(const BlockDriverState
*bs
)
6258 return bdrv_get_parent_name(bs
) ?: "";
6261 /* This can be used to identify nodes that might not have a device
6262 * name associated. Since node and device names live in the same
6263 * namespace, the result is unambiguous. The exception is if both are
6264 * absent, then this returns an empty (non-null) string. */
6265 const char *bdrv_get_device_or_node_name(const BlockDriverState
*bs
)
6268 return bdrv_get_parent_name(bs
) ?: bs
->node_name
;
6271 int bdrv_get_flags(BlockDriverState
*bs
)
6274 return bs
->open_flags
;
6277 int bdrv_has_zero_init_1(BlockDriverState
*bs
)
6279 GLOBAL_STATE_CODE();
6283 int bdrv_has_zero_init(BlockDriverState
*bs
)
6285 BlockDriverState
*filtered
;
6286 GLOBAL_STATE_CODE();
6292 /* If BS is a copy on write image, it is initialized to
6293 the contents of the base image, which may not be zeroes. */
6294 if (bdrv_cow_child(bs
)) {
6297 if (bs
->drv
->bdrv_has_zero_init
) {
6298 return bs
->drv
->bdrv_has_zero_init(bs
);
6301 filtered
= bdrv_filter_bs(bs
);
6303 return bdrv_has_zero_init(filtered
);
6310 bool bdrv_can_write_zeroes_with_unmap(BlockDriverState
*bs
)
6313 if (!(bs
->open_flags
& BDRV_O_UNMAP
)) {
6317 return bs
->supported_zero_flags
& BDRV_REQ_MAY_UNMAP
;
6320 void bdrv_get_backing_filename(BlockDriverState
*bs
,
6321 char *filename
, int filename_size
)
6324 pstrcpy(filename
, filename_size
, bs
->backing_file
);
6327 int coroutine_fn
bdrv_co_get_info(BlockDriverState
*bs
, BlockDriverInfo
*bdi
)
6330 BlockDriver
*drv
= bs
->drv
;
6332 /* if bs->drv == NULL, bs is closed, so there's nothing to do here */
6336 if (!drv
->bdrv_co_get_info
) {
6337 BlockDriverState
*filtered
= bdrv_filter_bs(bs
);
6339 return bdrv_co_get_info(filtered
, bdi
);
6343 memset(bdi
, 0, sizeof(*bdi
));
6344 ret
= drv
->bdrv_co_get_info(bs
, bdi
);
6349 if (bdi
->cluster_size
> BDRV_MAX_ALIGNMENT
) {
6356 ImageInfoSpecific
*bdrv_get_specific_info(BlockDriverState
*bs
,
6359 BlockDriver
*drv
= bs
->drv
;
6361 if (drv
&& drv
->bdrv_get_specific_info
) {
6362 return drv
->bdrv_get_specific_info(bs
, errp
);
6367 BlockStatsSpecific
*bdrv_get_specific_stats(BlockDriverState
*bs
)
6369 BlockDriver
*drv
= bs
->drv
;
6371 if (!drv
|| !drv
->bdrv_get_specific_stats
) {
6374 return drv
->bdrv_get_specific_stats(bs
);
6377 void coroutine_fn
bdrv_co_debug_event(BlockDriverState
*bs
, BlkdebugEvent event
)
6380 if (!bs
|| !bs
->drv
|| !bs
->drv
->bdrv_co_debug_event
) {
6384 bs
->drv
->bdrv_co_debug_event(bs
, event
);
6387 static BlockDriverState
*bdrv_find_debug_node(BlockDriverState
*bs
)
6389 GLOBAL_STATE_CODE();
6390 while (bs
&& bs
->drv
&& !bs
->drv
->bdrv_debug_breakpoint
) {
6391 bs
= bdrv_primary_bs(bs
);
6394 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_breakpoint
) {
6395 assert(bs
->drv
->bdrv_debug_remove_breakpoint
);
6402 int bdrv_debug_breakpoint(BlockDriverState
*bs
, const char *event
,
6405 GLOBAL_STATE_CODE();
6406 bs
= bdrv_find_debug_node(bs
);
6408 return bs
->drv
->bdrv_debug_breakpoint(bs
, event
, tag
);
6414 int bdrv_debug_remove_breakpoint(BlockDriverState
*bs
, const char *tag
)
6416 GLOBAL_STATE_CODE();
6417 bs
= bdrv_find_debug_node(bs
);
6419 return bs
->drv
->bdrv_debug_remove_breakpoint(bs
, tag
);
6425 int bdrv_debug_resume(BlockDriverState
*bs
, const char *tag
)
6427 GLOBAL_STATE_CODE();
6428 while (bs
&& (!bs
->drv
|| !bs
->drv
->bdrv_debug_resume
)) {
6429 bs
= bdrv_primary_bs(bs
);
6432 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_resume
) {
6433 return bs
->drv
->bdrv_debug_resume(bs
, tag
);
6439 bool bdrv_debug_is_suspended(BlockDriverState
*bs
, const char *tag
)
6441 GLOBAL_STATE_CODE();
6442 while (bs
&& bs
->drv
&& !bs
->drv
->bdrv_debug_is_suspended
) {
6443 bs
= bdrv_primary_bs(bs
);
6446 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_is_suspended
) {
6447 return bs
->drv
->bdrv_debug_is_suspended(bs
, tag
);
6453 /* backing_file can either be relative, or absolute, or a protocol. If it is
6454 * relative, it must be relative to the chain. So, passing in bs->filename
6455 * from a BDS as backing_file should not be done, as that may be relative to
6456 * the CWD rather than the chain. */
6457 BlockDriverState
*bdrv_find_backing_image(BlockDriverState
*bs
,
6458 const char *backing_file
)
6460 char *filename_full
= NULL
;
6461 char *backing_file_full
= NULL
;
6462 char *filename_tmp
= NULL
;
6463 int is_protocol
= 0;
6464 bool filenames_refreshed
= false;
6465 BlockDriverState
*curr_bs
= NULL
;
6466 BlockDriverState
*retval
= NULL
;
6467 BlockDriverState
*bs_below
;
6469 GLOBAL_STATE_CODE();
6471 if (!bs
|| !bs
->drv
|| !backing_file
) {
6475 filename_full
= g_malloc(PATH_MAX
);
6476 backing_file_full
= g_malloc(PATH_MAX
);
6478 is_protocol
= path_has_protocol(backing_file
);
6481 * Being largely a legacy function, skip any filters here
6482 * (because filters do not have normal filenames, so they cannot
6483 * match anyway; and allowing json:{} filenames is a bit out of
6486 for (curr_bs
= bdrv_skip_filters(bs
);
6487 bdrv_cow_child(curr_bs
) != NULL
;
6490 bs_below
= bdrv_backing_chain_next(curr_bs
);
6492 if (bdrv_backing_overridden(curr_bs
)) {
6494 * If the backing file was overridden, we can only compare
6495 * directly against the backing node's filename.
6498 if (!filenames_refreshed
) {
6500 * This will automatically refresh all of the
6501 * filenames in the rest of the backing chain, so we
6502 * only need to do this once.
6504 bdrv_refresh_filename(bs_below
);
6505 filenames_refreshed
= true;
6508 if (strcmp(backing_file
, bs_below
->filename
) == 0) {
6512 } else if (is_protocol
|| path_has_protocol(curr_bs
->backing_file
)) {
6514 * If either of the filename paths is actually a protocol, then
6515 * compare unmodified paths; otherwise make paths relative.
6517 char *backing_file_full_ret
;
6519 if (strcmp(backing_file
, curr_bs
->backing_file
) == 0) {
6523 /* Also check against the full backing filename for the image */
6524 backing_file_full_ret
= bdrv_get_full_backing_filename(curr_bs
,
6526 if (backing_file_full_ret
) {
6527 bool equal
= strcmp(backing_file
, backing_file_full_ret
) == 0;
6528 g_free(backing_file_full_ret
);
6535 /* If not an absolute filename path, make it relative to the current
6536 * image's filename path */
6537 filename_tmp
= bdrv_make_absolute_filename(curr_bs
, backing_file
,
6539 /* We are going to compare canonicalized absolute pathnames */
6540 if (!filename_tmp
|| !realpath(filename_tmp
, filename_full
)) {
6541 g_free(filename_tmp
);
6544 g_free(filename_tmp
);
6546 /* We need to make sure the backing filename we are comparing against
6547 * is relative to the current image filename (or absolute) */
6548 filename_tmp
= bdrv_get_full_backing_filename(curr_bs
, NULL
);
6549 if (!filename_tmp
|| !realpath(filename_tmp
, backing_file_full
)) {
6550 g_free(filename_tmp
);
6553 g_free(filename_tmp
);
6555 if (strcmp(backing_file_full
, filename_full
) == 0) {
6562 g_free(filename_full
);
6563 g_free(backing_file_full
);
6567 void bdrv_init(void)
6569 #ifdef CONFIG_BDRV_WHITELIST_TOOLS
6570 use_bdrv_whitelist
= 1;
6572 module_call_init(MODULE_INIT_BLOCK
);
6575 void bdrv_init_with_whitelist(void)
6577 use_bdrv_whitelist
= 1;
6581 int bdrv_activate(BlockDriverState
*bs
, Error
**errp
)
6583 BdrvChild
*child
, *parent
;
6584 Error
*local_err
= NULL
;
6586 BdrvDirtyBitmap
*bm
;
6588 GLOBAL_STATE_CODE();
6594 QLIST_FOREACH(child
, &bs
->children
, next
) {
6595 bdrv_activate(child
->bs
, &local_err
);
6597 error_propagate(errp
, local_err
);
6603 * Update permissions, they may differ for inactive nodes.
6605 * Note that the required permissions of inactive images are always a
6606 * subset of the permissions required after activating the image. This
6607 * allows us to just get the permissions upfront without restricting
6608 * bdrv_co_invalidate_cache().
6610 * It also means that in error cases, we don't have to try and revert to
6611 * the old permissions (which is an operation that could fail, too). We can
6612 * just keep the extended permissions for the next time that an activation
6613 * of the image is tried.
6615 if (bs
->open_flags
& BDRV_O_INACTIVE
) {
6616 bs
->open_flags
&= ~BDRV_O_INACTIVE
;
6617 ret
= bdrv_refresh_perms(bs
, NULL
, errp
);
6619 bs
->open_flags
|= BDRV_O_INACTIVE
;
6623 ret
= bdrv_invalidate_cache(bs
, errp
);
6625 bs
->open_flags
|= BDRV_O_INACTIVE
;
6629 FOR_EACH_DIRTY_BITMAP(bs
, bm
) {
6630 bdrv_dirty_bitmap_skip_store(bm
, false);
6633 ret
= bdrv_refresh_total_sectors(bs
, bs
->total_sectors
);
6635 bs
->open_flags
|= BDRV_O_INACTIVE
;
6636 error_setg_errno(errp
, -ret
, "Could not refresh total sector count");
6641 QLIST_FOREACH(parent
, &bs
->parents
, next_parent
) {
6642 if (parent
->klass
->activate
) {
6643 parent
->klass
->activate(parent
, &local_err
);
6645 bs
->open_flags
|= BDRV_O_INACTIVE
;
6646 error_propagate(errp
, local_err
);
6655 int coroutine_fn
bdrv_co_invalidate_cache(BlockDriverState
*bs
, Error
**errp
)
6657 Error
*local_err
= NULL
;
6660 assert(!(bs
->open_flags
& BDRV_O_INACTIVE
));
6661 assert_bdrv_graph_readable();
6663 if (bs
->drv
->bdrv_co_invalidate_cache
) {
6664 bs
->drv
->bdrv_co_invalidate_cache(bs
, &local_err
);
6666 error_propagate(errp
, local_err
);
6674 void bdrv_activate_all(Error
**errp
)
6676 BlockDriverState
*bs
;
6677 BdrvNextIterator it
;
6679 GLOBAL_STATE_CODE();
6681 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
6682 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
6685 aio_context_acquire(aio_context
);
6686 ret
= bdrv_activate(bs
, errp
);
6687 aio_context_release(aio_context
);
6689 bdrv_next_cleanup(&it
);
6695 static bool bdrv_has_bds_parent(BlockDriverState
*bs
, bool only_active
)
6698 GLOBAL_STATE_CODE();
6700 QLIST_FOREACH(parent
, &bs
->parents
, next_parent
) {
6701 if (parent
->klass
->parent_is_bds
) {
6702 BlockDriverState
*parent_bs
= parent
->opaque
;
6703 if (!only_active
|| !(parent_bs
->open_flags
& BDRV_O_INACTIVE
)) {
6712 static int bdrv_inactivate_recurse(BlockDriverState
*bs
)
6714 BdrvChild
*child
, *parent
;
6716 uint64_t cumulative_perms
, cumulative_shared_perms
;
6718 GLOBAL_STATE_CODE();
6724 /* Make sure that we don't inactivate a child before its parent.
6725 * It will be covered by recursion from the yet active parent. */
6726 if (bdrv_has_bds_parent(bs
, true)) {
6730 assert(!(bs
->open_flags
& BDRV_O_INACTIVE
));
6732 /* Inactivate this node */
6733 if (bs
->drv
->bdrv_inactivate
) {
6734 ret
= bs
->drv
->bdrv_inactivate(bs
);
6740 QLIST_FOREACH(parent
, &bs
->parents
, next_parent
) {
6741 if (parent
->klass
->inactivate
) {
6742 ret
= parent
->klass
->inactivate(parent
);
6749 bdrv_get_cumulative_perm(bs
, &cumulative_perms
,
6750 &cumulative_shared_perms
);
6751 if (cumulative_perms
& (BLK_PERM_WRITE
| BLK_PERM_WRITE_UNCHANGED
)) {
6752 /* Our inactive parents still need write access. Inactivation failed. */
6756 bs
->open_flags
|= BDRV_O_INACTIVE
;
6759 * Update permissions, they may differ for inactive nodes.
6760 * We only tried to loosen restrictions, so errors are not fatal, ignore
6763 bdrv_refresh_perms(bs
, NULL
, NULL
);
6765 /* Recursively inactivate children */
6766 QLIST_FOREACH(child
, &bs
->children
, next
) {
6767 ret
= bdrv_inactivate_recurse(child
->bs
);
6776 int bdrv_inactivate_all(void)
6778 BlockDriverState
*bs
= NULL
;
6779 BdrvNextIterator it
;
6781 GSList
*aio_ctxs
= NULL
, *ctx
;
6783 GLOBAL_STATE_CODE();
6785 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
6786 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
6788 if (!g_slist_find(aio_ctxs
, aio_context
)) {
6789 aio_ctxs
= g_slist_prepend(aio_ctxs
, aio_context
);
6790 aio_context_acquire(aio_context
);
6794 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
6795 /* Nodes with BDS parents are covered by recursion from the last
6796 * parent that gets inactivated. Don't inactivate them a second
6797 * time if that has already happened. */
6798 if (bdrv_has_bds_parent(bs
, false)) {
6801 ret
= bdrv_inactivate_recurse(bs
);
6803 bdrv_next_cleanup(&it
);
6809 for (ctx
= aio_ctxs
; ctx
!= NULL
; ctx
= ctx
->next
) {
6810 AioContext
*aio_context
= ctx
->data
;
6811 aio_context_release(aio_context
);
6813 g_slist_free(aio_ctxs
);
6818 /**************************************************************/
6819 /* removable device support */
6822 * Return TRUE if the media is present
6824 bool coroutine_fn
bdrv_co_is_inserted(BlockDriverState
*bs
)
6826 BlockDriver
*drv
= bs
->drv
;
6833 if (drv
->bdrv_co_is_inserted
) {
6834 return drv
->bdrv_co_is_inserted(bs
);
6836 QLIST_FOREACH(child
, &bs
->children
, next
) {
6837 if (!bdrv_co_is_inserted(child
->bs
)) {
6845 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
6847 void coroutine_fn
bdrv_co_eject(BlockDriverState
*bs
, bool eject_flag
)
6849 BlockDriver
*drv
= bs
->drv
;
6852 if (drv
&& drv
->bdrv_co_eject
) {
6853 drv
->bdrv_co_eject(bs
, eject_flag
);
6858 * Lock or unlock the media (if it is locked, the user won't be able
6859 * to eject it manually).
6861 void coroutine_fn
bdrv_co_lock_medium(BlockDriverState
*bs
, bool locked
)
6863 BlockDriver
*drv
= bs
->drv
;
6865 trace_bdrv_lock_medium(bs
, locked
);
6867 if (drv
&& drv
->bdrv_co_lock_medium
) {
6868 drv
->bdrv_co_lock_medium(bs
, locked
);
6872 /* Get a reference to bs */
6873 void bdrv_ref(BlockDriverState
*bs
)
6875 GLOBAL_STATE_CODE();
6879 /* Release a previously grabbed reference to bs.
6880 * If after releasing, reference count is zero, the BlockDriverState is
6882 void bdrv_unref(BlockDriverState
*bs
)
6884 GLOBAL_STATE_CODE();
6888 assert(bs
->refcnt
> 0);
6889 if (--bs
->refcnt
== 0) {
6894 struct BdrvOpBlocker
{
6896 QLIST_ENTRY(BdrvOpBlocker
) list
;
6899 bool bdrv_op_is_blocked(BlockDriverState
*bs
, BlockOpType op
, Error
**errp
)
6901 BdrvOpBlocker
*blocker
;
6902 GLOBAL_STATE_CODE();
6903 assert((int) op
>= 0 && op
< BLOCK_OP_TYPE_MAX
);
6904 if (!QLIST_EMPTY(&bs
->op_blockers
[op
])) {
6905 blocker
= QLIST_FIRST(&bs
->op_blockers
[op
]);
6906 error_propagate_prepend(errp
, error_copy(blocker
->reason
),
6907 "Node '%s' is busy: ",
6908 bdrv_get_device_or_node_name(bs
));
6914 void bdrv_op_block(BlockDriverState
*bs
, BlockOpType op
, Error
*reason
)
6916 BdrvOpBlocker
*blocker
;
6917 GLOBAL_STATE_CODE();
6918 assert((int) op
>= 0 && op
< BLOCK_OP_TYPE_MAX
);
6920 blocker
= g_new0(BdrvOpBlocker
, 1);
6921 blocker
->reason
= reason
;
6922 QLIST_INSERT_HEAD(&bs
->op_blockers
[op
], blocker
, list
);
6925 void bdrv_op_unblock(BlockDriverState
*bs
, BlockOpType op
, Error
*reason
)
6927 BdrvOpBlocker
*blocker
, *next
;
6928 GLOBAL_STATE_CODE();
6929 assert((int) op
>= 0 && op
< BLOCK_OP_TYPE_MAX
);
6930 QLIST_FOREACH_SAFE(blocker
, &bs
->op_blockers
[op
], list
, next
) {
6931 if (blocker
->reason
== reason
) {
6932 QLIST_REMOVE(blocker
, list
);
6938 void bdrv_op_block_all(BlockDriverState
*bs
, Error
*reason
)
6941 GLOBAL_STATE_CODE();
6942 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
6943 bdrv_op_block(bs
, i
, reason
);
6947 void bdrv_op_unblock_all(BlockDriverState
*bs
, Error
*reason
)
6950 GLOBAL_STATE_CODE();
6951 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
6952 bdrv_op_unblock(bs
, i
, reason
);
6956 bool bdrv_op_blocker_is_empty(BlockDriverState
*bs
)
6959 GLOBAL_STATE_CODE();
6960 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
6961 if (!QLIST_EMPTY(&bs
->op_blockers
[i
])) {
6969 * Must not be called while holding the lock of an AioContext other than the
6972 void bdrv_img_create(const char *filename
, const char *fmt
,
6973 const char *base_filename
, const char *base_fmt
,
6974 char *options
, uint64_t img_size
, int flags
, bool quiet
,
6977 QemuOptsList
*create_opts
= NULL
;
6978 QemuOpts
*opts
= NULL
;
6979 const char *backing_fmt
, *backing_file
;
6981 BlockDriver
*drv
, *proto_drv
;
6982 Error
*local_err
= NULL
;
6985 GLOBAL_STATE_CODE();
6987 /* Find driver and parse its options */
6988 drv
= bdrv_find_format(fmt
);
6990 error_setg(errp
, "Unknown file format '%s'", fmt
);
6994 proto_drv
= bdrv_find_protocol(filename
, true, errp
);
6999 if (!drv
->create_opts
) {
7000 error_setg(errp
, "Format driver '%s' does not support image creation",
7005 if (!proto_drv
->create_opts
) {
7006 error_setg(errp
, "Protocol driver '%s' does not support image creation",
7007 proto_drv
->format_name
);
7011 /* Create parameter list */
7012 create_opts
= qemu_opts_append(create_opts
, drv
->create_opts
);
7013 create_opts
= qemu_opts_append(create_opts
, proto_drv
->create_opts
);
7015 opts
= qemu_opts_create(create_opts
, NULL
, 0, &error_abort
);
7017 /* Parse -o options */
7019 if (!qemu_opts_do_parse(opts
, options
, NULL
, errp
)) {
7024 if (!qemu_opt_get(opts
, BLOCK_OPT_SIZE
)) {
7025 qemu_opt_set_number(opts
, BLOCK_OPT_SIZE
, img_size
, &error_abort
);
7026 } else if (img_size
!= UINT64_C(-1)) {
7027 error_setg(errp
, "The image size must be specified only once");
7031 if (base_filename
) {
7032 if (!qemu_opt_set(opts
, BLOCK_OPT_BACKING_FILE
, base_filename
,
7034 error_setg(errp
, "Backing file not supported for file format '%s'",
7041 if (!qemu_opt_set(opts
, BLOCK_OPT_BACKING_FMT
, base_fmt
, NULL
)) {
7042 error_setg(errp
, "Backing file format not supported for file "
7043 "format '%s'", fmt
);
7048 backing_file
= qemu_opt_get(opts
, BLOCK_OPT_BACKING_FILE
);
7050 if (!strcmp(filename
, backing_file
)) {
7051 error_setg(errp
, "Error: Trying to create an image with the "
7052 "same filename as the backing file");
7055 if (backing_file
[0] == '\0') {
7056 error_setg(errp
, "Expected backing file name, got empty string");
7061 backing_fmt
= qemu_opt_get(opts
, BLOCK_OPT_BACKING_FMT
);
7063 /* The size for the image must always be specified, unless we have a backing
7064 * file and we have not been forbidden from opening it. */
7065 size
= qemu_opt_get_size(opts
, BLOCK_OPT_SIZE
, img_size
);
7066 if (backing_file
&& !(flags
& BDRV_O_NO_BACKING
)) {
7067 BlockDriverState
*bs
;
7070 QDict
*backing_options
= NULL
;
7073 bdrv_get_full_backing_filename_from_filename(filename
, backing_file
,
7078 assert(full_backing
);
7081 * No need to do I/O here, which allows us to open encrypted
7082 * backing images without needing the secret
7085 back_flags
&= ~(BDRV_O_RDWR
| BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
);
7086 back_flags
|= BDRV_O_NO_IO
;
7088 backing_options
= qdict_new();
7090 qdict_put_str(backing_options
, "driver", backing_fmt
);
7092 qdict_put_bool(backing_options
, BDRV_OPT_FORCE_SHARE
, true);
7094 bs
= bdrv_open(full_backing
, NULL
, backing_options
, back_flags
,
7096 g_free(full_backing
);
7098 error_append_hint(&local_err
, "Could not open backing image.\n");
7102 error_setg(&local_err
,
7103 "Backing file specified without backing format");
7104 error_append_hint(&local_err
, "Detected format of %s.",
7105 bs
->drv
->format_name
);
7109 /* Opened BS, have no size */
7110 size
= bdrv_getlength(bs
);
7112 error_setg_errno(errp
, -size
, "Could not get size of '%s'",
7117 qemu_opt_set_number(opts
, BLOCK_OPT_SIZE
, size
, &error_abort
);
7121 /* (backing_file && !(flags & BDRV_O_NO_BACKING)) */
7122 } else if (backing_file
&& !backing_fmt
) {
7123 error_setg(&local_err
,
7124 "Backing file specified without backing format");
7129 error_setg(errp
, "Image creation needs a size parameter");
7134 printf("Formatting '%s', fmt=%s ", filename
, fmt
);
7135 qemu_opts_print(opts
, " ");
7140 ret
= bdrv_create(drv
, filename
, opts
, &local_err
);
7142 if (ret
== -EFBIG
) {
7143 /* This is generally a better message than whatever the driver would
7144 * deliver (especially because of the cluster_size_hint), since that
7145 * is most probably not much different from "image too large". */
7146 const char *cluster_size_hint
= "";
7147 if (qemu_opt_get_size(opts
, BLOCK_OPT_CLUSTER_SIZE
, 0)) {
7148 cluster_size_hint
= " (try using a larger cluster size)";
7150 error_setg(errp
, "The image size is too large for file format '%s'"
7151 "%s", fmt
, cluster_size_hint
);
7152 error_free(local_err
);
7157 qemu_opts_del(opts
);
7158 qemu_opts_free(create_opts
);
7159 error_propagate(errp
, local_err
);
7162 AioContext
*bdrv_get_aio_context(BlockDriverState
*bs
)
7165 return bs
? bs
->aio_context
: qemu_get_aio_context();
7168 AioContext
*coroutine_fn
bdrv_co_enter(BlockDriverState
*bs
)
7170 Coroutine
*self
= qemu_coroutine_self();
7171 AioContext
*old_ctx
= qemu_coroutine_get_aio_context(self
);
7172 AioContext
*new_ctx
;
7176 * Increase bs->in_flight to ensure that this operation is completed before
7177 * moving the node to a different AioContext. Read new_ctx only afterwards.
7179 bdrv_inc_in_flight(bs
);
7181 new_ctx
= bdrv_get_aio_context(bs
);
7182 aio_co_reschedule_self(new_ctx
);
7186 void coroutine_fn
bdrv_co_leave(BlockDriverState
*bs
, AioContext
*old_ctx
)
7189 aio_co_reschedule_self(old_ctx
);
7190 bdrv_dec_in_flight(bs
);
7193 void coroutine_fn
bdrv_co_lock(BlockDriverState
*bs
)
7195 AioContext
*ctx
= bdrv_get_aio_context(bs
);
7197 /* In the main thread, bs->aio_context won't change concurrently */
7198 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
7201 * We're in coroutine context, so we already hold the lock of the main
7202 * loop AioContext. Don't lock it twice to avoid deadlocks.
7204 assert(qemu_in_coroutine());
7205 if (ctx
!= qemu_get_aio_context()) {
7206 aio_context_acquire(ctx
);
7210 void coroutine_fn
bdrv_co_unlock(BlockDriverState
*bs
)
7212 AioContext
*ctx
= bdrv_get_aio_context(bs
);
7214 assert(qemu_in_coroutine());
7215 if (ctx
!= qemu_get_aio_context()) {
7216 aio_context_release(ctx
);
7220 static void bdrv_do_remove_aio_context_notifier(BdrvAioNotifier
*ban
)
7222 GLOBAL_STATE_CODE();
7223 QLIST_REMOVE(ban
, list
);
7227 static void bdrv_detach_aio_context(BlockDriverState
*bs
)
7229 BdrvAioNotifier
*baf
, *baf_tmp
;
7231 assert(!bs
->walking_aio_notifiers
);
7232 GLOBAL_STATE_CODE();
7233 bs
->walking_aio_notifiers
= true;
7234 QLIST_FOREACH_SAFE(baf
, &bs
->aio_notifiers
, list
, baf_tmp
) {
7236 bdrv_do_remove_aio_context_notifier(baf
);
7238 baf
->detach_aio_context(baf
->opaque
);
7241 /* Never mind iterating again to check for ->deleted. bdrv_close() will
7242 * remove remaining aio notifiers if we aren't called again.
7244 bs
->walking_aio_notifiers
= false;
7246 if (bs
->drv
&& bs
->drv
->bdrv_detach_aio_context
) {
7247 bs
->drv
->bdrv_detach_aio_context(bs
);
7250 if (bs
->quiesce_counter
) {
7251 aio_enable_external(bs
->aio_context
);
7253 bs
->aio_context
= NULL
;
7256 static void bdrv_attach_aio_context(BlockDriverState
*bs
,
7257 AioContext
*new_context
)
7259 BdrvAioNotifier
*ban
, *ban_tmp
;
7260 GLOBAL_STATE_CODE();
7262 if (bs
->quiesce_counter
) {
7263 aio_disable_external(new_context
);
7266 bs
->aio_context
= new_context
;
7268 if (bs
->drv
&& bs
->drv
->bdrv_attach_aio_context
) {
7269 bs
->drv
->bdrv_attach_aio_context(bs
, new_context
);
7272 assert(!bs
->walking_aio_notifiers
);
7273 bs
->walking_aio_notifiers
= true;
7274 QLIST_FOREACH_SAFE(ban
, &bs
->aio_notifiers
, list
, ban_tmp
) {
7276 bdrv_do_remove_aio_context_notifier(ban
);
7278 ban
->attached_aio_context(new_context
, ban
->opaque
);
7281 bs
->walking_aio_notifiers
= false;
7284 typedef struct BdrvStateSetAioContext
{
7285 AioContext
*new_ctx
;
7286 BlockDriverState
*bs
;
7287 } BdrvStateSetAioContext
;
7289 static bool bdrv_parent_change_aio_context(BdrvChild
*c
, AioContext
*ctx
,
7290 GHashTable
*visited
,
7294 GLOBAL_STATE_CODE();
7295 if (g_hash_table_contains(visited
, c
)) {
7298 g_hash_table_add(visited
, c
);
7301 * A BdrvChildClass that doesn't handle AioContext changes cannot
7302 * tolerate any AioContext changes
7304 if (!c
->klass
->change_aio_ctx
) {
7305 char *user
= bdrv_child_user_desc(c
);
7306 error_setg(errp
, "Changing iothreads is not supported by %s", user
);
7310 if (!c
->klass
->change_aio_ctx(c
, ctx
, visited
, tran
, errp
)) {
7311 assert(!errp
|| *errp
);
7317 bool bdrv_child_change_aio_context(BdrvChild
*c
, AioContext
*ctx
,
7318 GHashTable
*visited
, Transaction
*tran
,
7321 GLOBAL_STATE_CODE();
7322 if (g_hash_table_contains(visited
, c
)) {
7325 g_hash_table_add(visited
, c
);
7326 return bdrv_change_aio_context(c
->bs
, ctx
, visited
, tran
, errp
);
7329 static void bdrv_set_aio_context_clean(void *opaque
)
7331 BdrvStateSetAioContext
*state
= (BdrvStateSetAioContext
*) opaque
;
7332 BlockDriverState
*bs
= (BlockDriverState
*) state
->bs
;
7334 /* Paired with bdrv_drained_begin in bdrv_change_aio_context() */
7335 bdrv_drained_end(bs
);
7340 static void bdrv_set_aio_context_commit(void *opaque
)
7342 BdrvStateSetAioContext
*state
= (BdrvStateSetAioContext
*) opaque
;
7343 BlockDriverState
*bs
= (BlockDriverState
*) state
->bs
;
7344 AioContext
*new_context
= state
->new_ctx
;
7345 AioContext
*old_context
= bdrv_get_aio_context(bs
);
7348 * Take the old AioContex when detaching it from bs.
7349 * At this point, new_context lock is already acquired, and we are now
7350 * also taking old_context. This is safe as long as bdrv_detach_aio_context
7351 * does not call AIO_POLL_WHILE().
7353 if (old_context
!= qemu_get_aio_context()) {
7354 aio_context_acquire(old_context
);
7356 bdrv_detach_aio_context(bs
);
7357 if (old_context
!= qemu_get_aio_context()) {
7358 aio_context_release(old_context
);
7360 bdrv_attach_aio_context(bs
, new_context
);
7363 static TransactionActionDrv set_aio_context
= {
7364 .commit
= bdrv_set_aio_context_commit
,
7365 .clean
= bdrv_set_aio_context_clean
,
7369 * Changes the AioContext used for fd handlers, timers, and BHs by this
7370 * BlockDriverState and all its children and parents.
7372 * Must be called from the main AioContext.
7374 * The caller must own the AioContext lock for the old AioContext of bs, but it
7375 * must not own the AioContext lock for new_context (unless new_context is the
7376 * same as the current context of bs).
7378 * @visited will accumulate all visited BdrvChild objects. The caller is
7379 * responsible for freeing the list afterwards.
7381 static bool bdrv_change_aio_context(BlockDriverState
*bs
, AioContext
*ctx
,
7382 GHashTable
*visited
, Transaction
*tran
,
7386 BdrvStateSetAioContext
*state
;
7388 GLOBAL_STATE_CODE();
7390 if (bdrv_get_aio_context(bs
) == ctx
) {
7394 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
7395 if (!bdrv_parent_change_aio_context(c
, ctx
, visited
, tran
, errp
)) {
7400 QLIST_FOREACH(c
, &bs
->children
, next
) {
7401 if (!bdrv_child_change_aio_context(c
, ctx
, visited
, tran
, errp
)) {
7406 state
= g_new(BdrvStateSetAioContext
, 1);
7407 *state
= (BdrvStateSetAioContext
) {
7412 /* Paired with bdrv_drained_end in bdrv_set_aio_context_clean() */
7413 bdrv_drained_begin(bs
);
7415 tran_add(tran
, &set_aio_context
, state
);
7421 * Change bs's and recursively all of its parents' and children's AioContext
7422 * to the given new context, returning an error if that isn't possible.
7424 * If ignore_child is not NULL, that child (and its subgraph) will not
7427 * This function still requires the caller to take the bs current
7428 * AioContext lock, otherwise draining will fail since AIO_WAIT_WHILE
7429 * assumes the lock is always held if bs is in another AioContext.
7430 * For the same reason, it temporarily also holds the new AioContext, since
7431 * bdrv_drained_end calls BDRV_POLL_WHILE that assumes the lock is taken too.
7432 * Therefore the new AioContext lock must not be taken by the caller.
7434 int bdrv_try_change_aio_context(BlockDriverState
*bs
, AioContext
*ctx
,
7435 BdrvChild
*ignore_child
, Error
**errp
)
7438 GHashTable
*visited
;
7440 AioContext
*old_context
= bdrv_get_aio_context(bs
);
7441 GLOBAL_STATE_CODE();
7444 * Recursion phase: go through all nodes of the graph.
7445 * Take care of checking that all nodes support changing AioContext
7446 * and drain them, builing a linear list of callbacks to run if everything
7447 * is successful (the transaction itself).
7450 visited
= g_hash_table_new(NULL
, NULL
);
7452 g_hash_table_add(visited
, ignore_child
);
7454 ret
= bdrv_change_aio_context(bs
, ctx
, visited
, tran
, errp
);
7455 g_hash_table_destroy(visited
);
7458 * Linear phase: go through all callbacks collected in the transaction.
7459 * Run all callbacks collected in the recursion to switch all nodes
7460 * AioContext lock (transaction commit), or undo all changes done in the
7461 * recursion (transaction abort).
7465 /* Just run clean() callbacks. No AioContext changed. */
7471 * Release old AioContext, it won't be needed anymore, as all
7472 * bdrv_drained_begin() have been called already.
7474 if (qemu_get_aio_context() != old_context
) {
7475 aio_context_release(old_context
);
7479 * Acquire new AioContext since bdrv_drained_end() is going to be called
7480 * after we switched all nodes in the new AioContext, and the function
7481 * assumes that the lock of the bs is always taken.
7483 if (qemu_get_aio_context() != ctx
) {
7484 aio_context_acquire(ctx
);
7489 if (qemu_get_aio_context() != ctx
) {
7490 aio_context_release(ctx
);
7493 /* Re-acquire the old AioContext, since the caller takes and releases it. */
7494 if (qemu_get_aio_context() != old_context
) {
7495 aio_context_acquire(old_context
);
7501 void bdrv_add_aio_context_notifier(BlockDriverState
*bs
,
7502 void (*attached_aio_context
)(AioContext
*new_context
, void *opaque
),
7503 void (*detach_aio_context
)(void *opaque
), void *opaque
)
7505 BdrvAioNotifier
*ban
= g_new(BdrvAioNotifier
, 1);
7506 *ban
= (BdrvAioNotifier
){
7507 .attached_aio_context
= attached_aio_context
,
7508 .detach_aio_context
= detach_aio_context
,
7511 GLOBAL_STATE_CODE();
7513 QLIST_INSERT_HEAD(&bs
->aio_notifiers
, ban
, list
);
7516 void bdrv_remove_aio_context_notifier(BlockDriverState
*bs
,
7517 void (*attached_aio_context
)(AioContext
*,
7519 void (*detach_aio_context
)(void *),
7522 BdrvAioNotifier
*ban
, *ban_next
;
7523 GLOBAL_STATE_CODE();
7525 QLIST_FOREACH_SAFE(ban
, &bs
->aio_notifiers
, list
, ban_next
) {
7526 if (ban
->attached_aio_context
== attached_aio_context
&&
7527 ban
->detach_aio_context
== detach_aio_context
&&
7528 ban
->opaque
== opaque
&&
7529 ban
->deleted
== false)
7531 if (bs
->walking_aio_notifiers
) {
7532 ban
->deleted
= true;
7534 bdrv_do_remove_aio_context_notifier(ban
);
7543 int bdrv_amend_options(BlockDriverState
*bs
, QemuOpts
*opts
,
7544 BlockDriverAmendStatusCB
*status_cb
, void *cb_opaque
,
7548 GLOBAL_STATE_CODE();
7550 error_setg(errp
, "Node is ejected");
7553 if (!bs
->drv
->bdrv_amend_options
) {
7554 error_setg(errp
, "Block driver '%s' does not support option amendment",
7555 bs
->drv
->format_name
);
7558 return bs
->drv
->bdrv_amend_options(bs
, opts
, status_cb
,
7559 cb_opaque
, force
, errp
);
7563 * This function checks whether the given @to_replace is allowed to be
7564 * replaced by a node that always shows the same data as @bs. This is
7565 * used for example to verify whether the mirror job can replace
7566 * @to_replace by the target mirrored from @bs.
7567 * To be replaceable, @bs and @to_replace may either be guaranteed to
7568 * always show the same data (because they are only connected through
7569 * filters), or some driver may allow replacing one of its children
7570 * because it can guarantee that this child's data is not visible at
7571 * all (for example, for dissenting quorum children that have no other
7574 bool bdrv_recurse_can_replace(BlockDriverState
*bs
,
7575 BlockDriverState
*to_replace
)
7577 BlockDriverState
*filtered
;
7579 GLOBAL_STATE_CODE();
7581 if (!bs
|| !bs
->drv
) {
7585 if (bs
== to_replace
) {
7589 /* See what the driver can do */
7590 if (bs
->drv
->bdrv_recurse_can_replace
) {
7591 return bs
->drv
->bdrv_recurse_can_replace(bs
, to_replace
);
7594 /* For filters without an own implementation, we can recurse on our own */
7595 filtered
= bdrv_filter_bs(bs
);
7597 return bdrv_recurse_can_replace(filtered
, to_replace
);
7605 * Check whether the given @node_name can be replaced by a node that
7606 * has the same data as @parent_bs. If so, return @node_name's BDS;
7609 * @node_name must be a (recursive) *child of @parent_bs (or this
7610 * function will return NULL).
7612 * The result (whether the node can be replaced or not) is only valid
7613 * for as long as no graph or permission changes occur.
7615 BlockDriverState
*check_to_replace_node(BlockDriverState
*parent_bs
,
7616 const char *node_name
, Error
**errp
)
7618 BlockDriverState
*to_replace_bs
= bdrv_find_node(node_name
);
7619 AioContext
*aio_context
;
7621 GLOBAL_STATE_CODE();
7623 if (!to_replace_bs
) {
7624 error_setg(errp
, "Failed to find node with node-name='%s'", node_name
);
7628 aio_context
= bdrv_get_aio_context(to_replace_bs
);
7629 aio_context_acquire(aio_context
);
7631 if (bdrv_op_is_blocked(to_replace_bs
, BLOCK_OP_TYPE_REPLACE
, errp
)) {
7632 to_replace_bs
= NULL
;
7636 /* We don't want arbitrary node of the BDS chain to be replaced only the top
7637 * most non filter in order to prevent data corruption.
7638 * Another benefit is that this tests exclude backing files which are
7639 * blocked by the backing blockers.
7641 if (!bdrv_recurse_can_replace(parent_bs
, to_replace_bs
)) {
7642 error_setg(errp
, "Cannot replace '%s' by a node mirrored from '%s', "
7643 "because it cannot be guaranteed that doing so would not "
7644 "lead to an abrupt change of visible data",
7645 node_name
, parent_bs
->node_name
);
7646 to_replace_bs
= NULL
;
7651 aio_context_release(aio_context
);
7652 return to_replace_bs
;
7656 * Iterates through the list of runtime option keys that are said to
7657 * be "strong" for a BDS. An option is called "strong" if it changes
7658 * a BDS's data. For example, the null block driver's "size" and
7659 * "read-zeroes" options are strong, but its "latency-ns" option is
7662 * If a key returned by this function ends with a dot, all options
7663 * starting with that prefix are strong.
7665 static const char *const *strong_options(BlockDriverState
*bs
,
7666 const char *const *curopt
)
7668 static const char *const global_options
[] = {
7669 "driver", "filename", NULL
7673 return &global_options
[0];
7677 if (curopt
== &global_options
[ARRAY_SIZE(global_options
) - 1] && bs
->drv
) {
7678 curopt
= bs
->drv
->strong_runtime_opts
;
7681 return (curopt
&& *curopt
) ? curopt
: NULL
;
7685 * Copies all strong runtime options from bs->options to the given
7686 * QDict. The set of strong option keys is determined by invoking
7689 * Returns true iff any strong option was present in bs->options (and
7690 * thus copied to the target QDict) with the exception of "filename"
7691 * and "driver". The caller is expected to use this value to decide
7692 * whether the existence of strong options prevents the generation of
7695 static bool append_strong_runtime_options(QDict
*d
, BlockDriverState
*bs
)
7697 bool found_any
= false;
7698 const char *const *option_name
= NULL
;
7704 while ((option_name
= strong_options(bs
, option_name
))) {
7705 bool option_given
= false;
7707 assert(strlen(*option_name
) > 0);
7708 if ((*option_name
)[strlen(*option_name
) - 1] != '.') {
7709 QObject
*entry
= qdict_get(bs
->options
, *option_name
);
7714 qdict_put_obj(d
, *option_name
, qobject_ref(entry
));
7715 option_given
= true;
7717 const QDictEntry
*entry
;
7718 for (entry
= qdict_first(bs
->options
); entry
;
7719 entry
= qdict_next(bs
->options
, entry
))
7721 if (strstart(qdict_entry_key(entry
), *option_name
, NULL
)) {
7722 qdict_put_obj(d
, qdict_entry_key(entry
),
7723 qobject_ref(qdict_entry_value(entry
)));
7724 option_given
= true;
7729 /* While "driver" and "filename" need to be included in a JSON filename,
7730 * their existence does not prohibit generation of a plain filename. */
7731 if (!found_any
&& option_given
&&
7732 strcmp(*option_name
, "driver") && strcmp(*option_name
, "filename"))
7738 if (!qdict_haskey(d
, "driver")) {
7739 /* Drivers created with bdrv_new_open_driver() may not have a
7740 * @driver option. Add it here. */
7741 qdict_put_str(d
, "driver", bs
->drv
->format_name
);
7747 /* Note: This function may return false positives; it may return true
7748 * even if opening the backing file specified by bs's image header
7749 * would result in exactly bs->backing. */
7750 static bool bdrv_backing_overridden(BlockDriverState
*bs
)
7752 GLOBAL_STATE_CODE();
7754 return strcmp(bs
->auto_backing_file
,
7755 bs
->backing
->bs
->filename
);
7757 /* No backing BDS, so if the image header reports any backing
7758 * file, it must have been suppressed */
7759 return bs
->auto_backing_file
[0] != '\0';
7763 /* Updates the following BDS fields:
7764 * - exact_filename: A filename which may be used for opening a block device
7765 * which (mostly) equals the given BDS (even without any
7766 * other options; so reading and writing must return the same
7767 * results, but caching etc. may be different)
7768 * - full_open_options: Options which, when given when opening a block device
7769 * (without a filename), result in a BDS (mostly)
7770 * equalling the given one
7771 * - filename: If exact_filename is set, it is copied here. Otherwise,
7772 * full_open_options is converted to a JSON object, prefixed with
7773 * "json:" (for use through the JSON pseudo protocol) and put here.
7775 void bdrv_refresh_filename(BlockDriverState
*bs
)
7777 BlockDriver
*drv
= bs
->drv
;
7779 BlockDriverState
*primary_child_bs
;
7781 bool backing_overridden
;
7782 bool generate_json_filename
; /* Whether our default implementation should
7783 fill exact_filename (false) or not (true) */
7785 GLOBAL_STATE_CODE();
7791 /* This BDS's file name may depend on any of its children's file names, so
7792 * refresh those first */
7793 QLIST_FOREACH(child
, &bs
->children
, next
) {
7794 bdrv_refresh_filename(child
->bs
);
7798 /* For implicit nodes, just copy everything from the single child */
7799 child
= QLIST_FIRST(&bs
->children
);
7800 assert(QLIST_NEXT(child
, next
) == NULL
);
7802 pstrcpy(bs
->exact_filename
, sizeof(bs
->exact_filename
),
7803 child
->bs
->exact_filename
);
7804 pstrcpy(bs
->filename
, sizeof(bs
->filename
), child
->bs
->filename
);
7806 qobject_unref(bs
->full_open_options
);
7807 bs
->full_open_options
= qobject_ref(child
->bs
->full_open_options
);
7812 backing_overridden
= bdrv_backing_overridden(bs
);
7814 if (bs
->open_flags
& BDRV_O_NO_IO
) {
7815 /* Without I/O, the backing file does not change anything.
7816 * Therefore, in such a case (primarily qemu-img), we can
7817 * pretend the backing file has not been overridden even if
7818 * it technically has been. */
7819 backing_overridden
= false;
7822 /* Gather the options QDict */
7824 generate_json_filename
= append_strong_runtime_options(opts
, bs
);
7825 generate_json_filename
|= backing_overridden
;
7827 if (drv
->bdrv_gather_child_options
) {
7828 /* Some block drivers may not want to present all of their children's
7829 * options, or name them differently from BdrvChild.name */
7830 drv
->bdrv_gather_child_options(bs
, opts
, backing_overridden
);
7832 QLIST_FOREACH(child
, &bs
->children
, next
) {
7833 if (child
== bs
->backing
&& !backing_overridden
) {
7834 /* We can skip the backing BDS if it has not been overridden */
7838 qdict_put(opts
, child
->name
,
7839 qobject_ref(child
->bs
->full_open_options
));
7842 if (backing_overridden
&& !bs
->backing
) {
7843 /* Force no backing file */
7844 qdict_put_null(opts
, "backing");
7848 qobject_unref(bs
->full_open_options
);
7849 bs
->full_open_options
= opts
;
7851 primary_child_bs
= bdrv_primary_bs(bs
);
7853 if (drv
->bdrv_refresh_filename
) {
7854 /* Obsolete information is of no use here, so drop the old file name
7855 * information before refreshing it */
7856 bs
->exact_filename
[0] = '\0';
7858 drv
->bdrv_refresh_filename(bs
);
7859 } else if (primary_child_bs
) {
7861 * Try to reconstruct valid information from the underlying
7862 * file -- this only works for format nodes (filter nodes
7863 * cannot be probed and as such must be selected by the user
7864 * either through an options dict, or through a special
7865 * filename which the filter driver must construct in its
7866 * .bdrv_refresh_filename() implementation).
7869 bs
->exact_filename
[0] = '\0';
7872 * We can use the underlying file's filename if:
7873 * - it has a filename,
7874 * - the current BDS is not a filter,
7875 * - the file is a protocol BDS, and
7876 * - opening that file (as this BDS's format) will automatically create
7877 * the BDS tree we have right now, that is:
7878 * - the user did not significantly change this BDS's behavior with
7879 * some explicit (strong) options
7880 * - no non-file child of this BDS has been overridden by the user
7881 * Both of these conditions are represented by generate_json_filename.
7883 if (primary_child_bs
->exact_filename
[0] &&
7884 primary_child_bs
->drv
->bdrv_file_open
&&
7885 !drv
->is_filter
&& !generate_json_filename
)
7887 strcpy(bs
->exact_filename
, primary_child_bs
->exact_filename
);
7891 if (bs
->exact_filename
[0]) {
7892 pstrcpy(bs
->filename
, sizeof(bs
->filename
), bs
->exact_filename
);
7894 GString
*json
= qobject_to_json(QOBJECT(bs
->full_open_options
));
7895 if (snprintf(bs
->filename
, sizeof(bs
->filename
), "json:%s",
7896 json
->str
) >= sizeof(bs
->filename
)) {
7897 /* Give user a hint if we truncated things. */
7898 strcpy(bs
->filename
+ sizeof(bs
->filename
) - 4, "...");
7900 g_string_free(json
, true);
7904 char *bdrv_dirname(BlockDriverState
*bs
, Error
**errp
)
7906 BlockDriver
*drv
= bs
->drv
;
7907 BlockDriverState
*child_bs
;
7909 GLOBAL_STATE_CODE();
7912 error_setg(errp
, "Node '%s' is ejected", bs
->node_name
);
7916 if (drv
->bdrv_dirname
) {
7917 return drv
->bdrv_dirname(bs
, errp
);
7920 child_bs
= bdrv_primary_bs(bs
);
7922 return bdrv_dirname(child_bs
, errp
);
7925 bdrv_refresh_filename(bs
);
7926 if (bs
->exact_filename
[0] != '\0') {
7927 return path_combine(bs
->exact_filename
, "");
7930 error_setg(errp
, "Cannot generate a base directory for %s nodes",
7936 * Hot add/remove a BDS's child. So the user can take a child offline when
7937 * it is broken and take a new child online
7939 void bdrv_add_child(BlockDriverState
*parent_bs
, BlockDriverState
*child_bs
,
7942 GLOBAL_STATE_CODE();
7943 if (!parent_bs
->drv
|| !parent_bs
->drv
->bdrv_add_child
) {
7944 error_setg(errp
, "The node %s does not support adding a child",
7945 bdrv_get_device_or_node_name(parent_bs
));
7949 if (!QLIST_EMPTY(&child_bs
->parents
)) {
7950 error_setg(errp
, "The node %s already has a parent",
7951 child_bs
->node_name
);
7955 parent_bs
->drv
->bdrv_add_child(parent_bs
, child_bs
, errp
);
7958 void bdrv_del_child(BlockDriverState
*parent_bs
, BdrvChild
*child
, Error
**errp
)
7962 GLOBAL_STATE_CODE();
7963 if (!parent_bs
->drv
|| !parent_bs
->drv
->bdrv_del_child
) {
7964 error_setg(errp
, "The node %s does not support removing a child",
7965 bdrv_get_device_or_node_name(parent_bs
));
7969 QLIST_FOREACH(tmp
, &parent_bs
->children
, next
) {
7976 error_setg(errp
, "The node %s does not have a child named %s",
7977 bdrv_get_device_or_node_name(parent_bs
),
7978 bdrv_get_device_or_node_name(child
->bs
));
7982 parent_bs
->drv
->bdrv_del_child(parent_bs
, child
, errp
);
7985 int bdrv_make_empty(BdrvChild
*c
, Error
**errp
)
7987 BlockDriver
*drv
= c
->bs
->drv
;
7990 GLOBAL_STATE_CODE();
7991 assert(c
->perm
& (BLK_PERM_WRITE
| BLK_PERM_WRITE_UNCHANGED
));
7993 if (!drv
->bdrv_make_empty
) {
7994 error_setg(errp
, "%s does not support emptying nodes",
7999 ret
= drv
->bdrv_make_empty(c
->bs
);
8001 error_setg_errno(errp
, -ret
, "Failed to empty %s",
8010 * Return the child that @bs acts as an overlay for, and from which data may be
8011 * copied in COW or COR operations. Usually this is the backing file.
8013 BdrvChild
*bdrv_cow_child(BlockDriverState
*bs
)
8017 if (!bs
|| !bs
->drv
) {
8021 if (bs
->drv
->is_filter
) {
8029 assert(bs
->backing
->role
& BDRV_CHILD_COW
);
8034 * If @bs acts as a filter for exactly one of its children, return
8037 BdrvChild
*bdrv_filter_child(BlockDriverState
*bs
)
8042 if (!bs
|| !bs
->drv
) {
8046 if (!bs
->drv
->is_filter
) {
8050 /* Only one of @backing or @file may be used */
8051 assert(!(bs
->backing
&& bs
->file
));
8053 c
= bs
->backing
?: bs
->file
;
8058 assert(c
->role
& BDRV_CHILD_FILTERED
);
8063 * Return either the result of bdrv_cow_child() or bdrv_filter_child(),
8064 * whichever is non-NULL.
8066 * Return NULL if both are NULL.
8068 BdrvChild
*bdrv_filter_or_cow_child(BlockDriverState
*bs
)
8070 BdrvChild
*cow_child
= bdrv_cow_child(bs
);
8071 BdrvChild
*filter_child
= bdrv_filter_child(bs
);
8074 /* Filter nodes cannot have COW backing files */
8075 assert(!(cow_child
&& filter_child
));
8077 return cow_child
?: filter_child
;
8081 * Return the primary child of this node: For filters, that is the
8082 * filtered child. For other nodes, that is usually the child storing
8084 * (A generally more helpful description is that this is (usually) the
8085 * child that has the same filename as @bs.)
8087 * Drivers do not necessarily have a primary child; for example quorum
8090 BdrvChild
*bdrv_primary_child(BlockDriverState
*bs
)
8092 BdrvChild
*c
, *found
= NULL
;
8095 QLIST_FOREACH(c
, &bs
->children
, next
) {
8096 if (c
->role
& BDRV_CHILD_PRIMARY
) {
8105 static BlockDriverState
*bdrv_do_skip_filters(BlockDriverState
*bs
,
8106 bool stop_on_explicit_filter
)
8114 while (!(stop_on_explicit_filter
&& !bs
->implicit
)) {
8115 c
= bdrv_filter_child(bs
);
8118 * A filter that is embedded in a working block graph must
8119 * have a child. Assert this here so this function does
8120 * not return a filter node that is not expected by the
8123 assert(!bs
->drv
|| !bs
->drv
->is_filter
);
8129 * Note that this treats nodes with bs->drv == NULL as not being
8130 * filters (bs->drv == NULL should be replaced by something else
8132 * The advantage of this behavior is that this function will thus
8133 * always return a non-NULL value (given a non-NULL @bs).
8140 * Return the first BDS that has not been added implicitly or that
8141 * does not have a filtered child down the chain starting from @bs
8142 * (including @bs itself).
8144 BlockDriverState
*bdrv_skip_implicit_filters(BlockDriverState
*bs
)
8146 GLOBAL_STATE_CODE();
8147 return bdrv_do_skip_filters(bs
, true);
8151 * Return the first BDS that does not have a filtered child down the
8152 * chain starting from @bs (including @bs itself).
8154 BlockDriverState
*bdrv_skip_filters(BlockDriverState
*bs
)
8157 return bdrv_do_skip_filters(bs
, false);
8161 * For a backing chain, return the first non-filter backing image of
8162 * the first non-filter image.
8164 BlockDriverState
*bdrv_backing_chain_next(BlockDriverState
*bs
)
8167 return bdrv_skip_filters(bdrv_cow_bs(bdrv_skip_filters(bs
)));
8171 * Check whether [offset, offset + bytes) overlaps with the cached
8172 * block-status data region.
8174 * If so, and @pnum is not NULL, set *pnum to `bsc.data_end - offset`,
8175 * which is what bdrv_bsc_is_data()'s interface needs.
8176 * Otherwise, *pnum is not touched.
8178 static bool bdrv_bsc_range_overlaps_locked(BlockDriverState
*bs
,
8179 int64_t offset
, int64_t bytes
,
8182 BdrvBlockStatusCache
*bsc
= qatomic_rcu_read(&bs
->block_status_cache
);
8186 qatomic_read(&bsc
->valid
) &&
8187 ranges_overlap(offset
, bytes
, bsc
->data_start
,
8188 bsc
->data_end
- bsc
->data_start
);
8190 if (overlaps
&& pnum
) {
8191 *pnum
= bsc
->data_end
- offset
;
8198 * See block_int.h for this function's documentation.
8200 bool bdrv_bsc_is_data(BlockDriverState
*bs
, int64_t offset
, int64_t *pnum
)
8203 RCU_READ_LOCK_GUARD();
8204 return bdrv_bsc_range_overlaps_locked(bs
, offset
, 1, pnum
);
8208 * See block_int.h for this function's documentation.
8210 void bdrv_bsc_invalidate_range(BlockDriverState
*bs
,
8211 int64_t offset
, int64_t bytes
)
8214 RCU_READ_LOCK_GUARD();
8216 if (bdrv_bsc_range_overlaps_locked(bs
, offset
, bytes
, NULL
)) {
8217 qatomic_set(&bs
->block_status_cache
->valid
, false);
8222 * See block_int.h for this function's documentation.
8224 void bdrv_bsc_fill(BlockDriverState
*bs
, int64_t offset
, int64_t bytes
)
8226 BdrvBlockStatusCache
*new_bsc
= g_new(BdrvBlockStatusCache
, 1);
8227 BdrvBlockStatusCache
*old_bsc
;
8230 *new_bsc
= (BdrvBlockStatusCache
) {
8232 .data_start
= offset
,
8233 .data_end
= offset
+ bytes
,
8236 QEMU_LOCK_GUARD(&bs
->bsc_modify_lock
);
8238 old_bsc
= qatomic_rcu_read(&bs
->block_status_cache
);
8239 qatomic_rcu_set(&bs
->block_status_cache
, new_bsc
);
8241 g_free_rcu(old_bsc
, rcu
);