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
;
743 assert_bdrv_graph_readable();
746 error_setg(errp
, "Block node '%s' is not opened", bs
->filename
);
750 if (!bs
->drv
->bdrv_co_delete_file
) {
751 error_setg(errp
, "Driver '%s' does not support image deletion",
752 bs
->drv
->format_name
);
756 ret
= bs
->drv
->bdrv_co_delete_file(bs
, &local_err
);
758 error_propagate(errp
, local_err
);
764 void coroutine_fn
bdrv_co_delete_file_noerr(BlockDriverState
*bs
)
766 Error
*local_err
= NULL
;
774 ret
= bdrv_co_delete_file(bs
, &local_err
);
776 * ENOTSUP will happen if the block driver doesn't support
777 * the 'bdrv_co_delete_file' interface. This is a predictable
778 * scenario and shouldn't be reported back to the user.
780 if (ret
== -ENOTSUP
) {
781 error_free(local_err
);
782 } else if (ret
< 0) {
783 error_report_err(local_err
);
788 * Try to get @bs's logical and physical block size.
789 * On success, store them in @bsz struct and return 0.
790 * On failure return -errno.
791 * @bs must not be empty.
793 int bdrv_probe_blocksizes(BlockDriverState
*bs
, BlockSizes
*bsz
)
795 BlockDriver
*drv
= bs
->drv
;
796 BlockDriverState
*filtered
= bdrv_filter_bs(bs
);
799 if (drv
&& drv
->bdrv_probe_blocksizes
) {
800 return drv
->bdrv_probe_blocksizes(bs
, bsz
);
801 } else if (filtered
) {
802 return bdrv_probe_blocksizes(filtered
, bsz
);
809 * Try to get @bs's geometry (cyls, heads, sectors).
810 * On success, store them in @geo struct and return 0.
811 * On failure return -errno.
812 * @bs must not be empty.
814 int bdrv_probe_geometry(BlockDriverState
*bs
, HDGeometry
*geo
)
816 BlockDriver
*drv
= bs
->drv
;
817 BlockDriverState
*filtered
= bdrv_filter_bs(bs
);
820 if (drv
&& drv
->bdrv_probe_geometry
) {
821 return drv
->bdrv_probe_geometry(bs
, geo
);
822 } else if (filtered
) {
823 return bdrv_probe_geometry(filtered
, geo
);
830 * Create a uniquely-named empty temporary file.
831 * Return the actual file name used upon success, otherwise NULL.
832 * This string should be freed with g_free() when not needed any longer.
834 * Note: creating a temporary file for the caller to (re)open is
835 * inherently racy. Use g_file_open_tmp() instead whenever practical.
837 char *create_tmp_file(Error
**errp
)
841 g_autofree
char *filename
= NULL
;
843 tmpdir
= g_get_tmp_dir();
846 * See commit 69bef79 ("block: use /var/tmp instead of /tmp for -snapshot")
848 * This function is used to create temporary disk images (like -snapshot),
849 * so the files can become very large. /tmp is often a tmpfs where as
850 * /var/tmp is usually on a disk, so more appropriate for disk images.
852 if (!g_strcmp0(tmpdir
, "/tmp")) {
857 filename
= g_strdup_printf("%s/vl.XXXXXX", tmpdir
);
858 fd
= g_mkstemp(filename
);
860 error_setg_errno(errp
, errno
, "Could not open temporary file '%s'",
866 return g_steal_pointer(&filename
);
870 * Detect host devices. By convention, /dev/cdrom[N] is always
871 * recognized as a host CDROM.
873 static BlockDriver
*find_hdev_driver(const char *filename
)
875 int score_max
= 0, score
;
876 BlockDriver
*drv
= NULL
, *d
;
879 QLIST_FOREACH(d
, &bdrv_drivers
, list
) {
880 if (d
->bdrv_probe_device
) {
881 score
= d
->bdrv_probe_device(filename
);
882 if (score
> score_max
) {
892 static BlockDriver
*bdrv_do_find_protocol(const char *protocol
)
897 QLIST_FOREACH(drv1
, &bdrv_drivers
, list
) {
898 if (drv1
->protocol_name
&& !strcmp(drv1
->protocol_name
, protocol
)) {
906 BlockDriver
*bdrv_find_protocol(const char *filename
,
907 bool allow_protocol_prefix
,
917 /* TODO Drivers without bdrv_file_open must be specified explicitly */
920 * XXX(hch): we really should not let host device detection
921 * override an explicit protocol specification, but moving this
922 * later breaks access to device names with colons in them.
923 * Thanks to the brain-dead persistent naming schemes on udev-
924 * based Linux systems those actually are quite common.
926 drv1
= find_hdev_driver(filename
);
931 if (!path_has_protocol(filename
) || !allow_protocol_prefix
) {
935 p
= strchr(filename
, ':');
938 if (len
> sizeof(protocol
) - 1)
939 len
= sizeof(protocol
) - 1;
940 memcpy(protocol
, filename
, len
);
941 protocol
[len
] = '\0';
943 drv1
= bdrv_do_find_protocol(protocol
);
948 for (i
= 0; i
< (int)ARRAY_SIZE(block_driver_modules
); ++i
) {
949 if (block_driver_modules
[i
].protocol_name
&&
950 !strcmp(block_driver_modules
[i
].protocol_name
, protocol
)) {
951 int rv
= block_module_load(block_driver_modules
[i
].library_name
, errp
);
953 drv1
= bdrv_do_find_protocol(protocol
);
962 error_setg(errp
, "Unknown protocol '%s'", protocol
);
968 * Guess image format by probing its contents.
969 * This is not a good idea when your image is raw (CVE-2008-2004), but
970 * we do it anyway for backward compatibility.
972 * @buf contains the image's first @buf_size bytes.
973 * @buf_size is the buffer size in bytes (generally BLOCK_PROBE_BUF_SIZE,
974 * but can be smaller if the image file is smaller)
975 * @filename is its filename.
977 * For all block drivers, call the bdrv_probe() method to get its
979 * Return the first block driver with the highest probing score.
981 BlockDriver
*bdrv_probe_all(const uint8_t *buf
, int buf_size
,
982 const char *filename
)
984 int score_max
= 0, score
;
985 BlockDriver
*drv
= NULL
, *d
;
988 QLIST_FOREACH(d
, &bdrv_drivers
, list
) {
990 score
= d
->bdrv_probe(buf
, buf_size
, filename
);
991 if (score
> score_max
) {
1001 static int find_image_format(BlockBackend
*file
, const char *filename
,
1002 BlockDriver
**pdrv
, Error
**errp
)
1005 uint8_t buf
[BLOCK_PROBE_BUF_SIZE
];
1008 GLOBAL_STATE_CODE();
1010 /* Return the raw BlockDriver * to scsi-generic devices or empty drives */
1011 if (blk_is_sg(file
) || !blk_is_inserted(file
) || blk_getlength(file
) == 0) {
1016 ret
= blk_pread(file
, 0, sizeof(buf
), buf
, 0);
1018 error_setg_errno(errp
, -ret
, "Could not read image for determining its "
1024 drv
= bdrv_probe_all(buf
, sizeof(buf
), filename
);
1026 error_setg(errp
, "Could not determine image format: No compatible "
1037 * Set the current 'total_sectors' value
1038 * Return 0 on success, -errno on error.
1040 int coroutine_fn
bdrv_co_refresh_total_sectors(BlockDriverState
*bs
,
1043 BlockDriver
*drv
= bs
->drv
;
1045 assert_bdrv_graph_readable();
1051 /* Do not attempt drv->bdrv_co_getlength() on scsi-generic devices */
1055 /* query actual device if possible, otherwise just trust the hint */
1056 if (drv
->bdrv_co_getlength
) {
1057 int64_t length
= drv
->bdrv_co_getlength(bs
);
1061 hint
= DIV_ROUND_UP(length
, BDRV_SECTOR_SIZE
);
1064 bs
->total_sectors
= hint
;
1066 if (bs
->total_sectors
* BDRV_SECTOR_SIZE
> BDRV_MAX_LENGTH
) {
1074 * Combines a QDict of new block driver @options with any missing options taken
1075 * from @old_options, so that leaving out an option defaults to its old value.
1077 static void bdrv_join_options(BlockDriverState
*bs
, QDict
*options
,
1080 GLOBAL_STATE_CODE();
1081 if (bs
->drv
&& bs
->drv
->bdrv_join_options
) {
1082 bs
->drv
->bdrv_join_options(options
, old_options
);
1084 qdict_join(options
, old_options
, false);
1088 static BlockdevDetectZeroesOptions
bdrv_parse_detect_zeroes(QemuOpts
*opts
,
1092 Error
*local_err
= NULL
;
1093 char *value
= qemu_opt_get_del(opts
, "detect-zeroes");
1094 BlockdevDetectZeroesOptions detect_zeroes
=
1095 qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup
, value
,
1096 BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF
, &local_err
);
1097 GLOBAL_STATE_CODE();
1100 error_propagate(errp
, local_err
);
1101 return detect_zeroes
;
1104 if (detect_zeroes
== BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP
&&
1105 !(open_flags
& BDRV_O_UNMAP
))
1107 error_setg(errp
, "setting detect-zeroes to unmap is not allowed "
1108 "without setting discard operation to unmap");
1111 return detect_zeroes
;
1115 * Set open flags for aio engine
1117 * Return 0 on success, -1 if the engine specified is invalid
1119 int bdrv_parse_aio(const char *mode
, int *flags
)
1121 if (!strcmp(mode
, "threads")) {
1122 /* do nothing, default */
1123 } else if (!strcmp(mode
, "native")) {
1124 *flags
|= BDRV_O_NATIVE_AIO
;
1125 #ifdef CONFIG_LINUX_IO_URING
1126 } else if (!strcmp(mode
, "io_uring")) {
1127 *flags
|= BDRV_O_IO_URING
;
1137 * Set open flags for a given discard mode
1139 * Return 0 on success, -1 if the discard mode was invalid.
1141 int bdrv_parse_discard_flags(const char *mode
, int *flags
)
1143 *flags
&= ~BDRV_O_UNMAP
;
1145 if (!strcmp(mode
, "off") || !strcmp(mode
, "ignore")) {
1147 } else if (!strcmp(mode
, "on") || !strcmp(mode
, "unmap")) {
1148 *flags
|= BDRV_O_UNMAP
;
1157 * Set open flags for a given cache mode
1159 * Return 0 on success, -1 if the cache mode was invalid.
1161 int bdrv_parse_cache_mode(const char *mode
, int *flags
, bool *writethrough
)
1163 *flags
&= ~BDRV_O_CACHE_MASK
;
1165 if (!strcmp(mode
, "off") || !strcmp(mode
, "none")) {
1166 *writethrough
= false;
1167 *flags
|= BDRV_O_NOCACHE
;
1168 } else if (!strcmp(mode
, "directsync")) {
1169 *writethrough
= true;
1170 *flags
|= BDRV_O_NOCACHE
;
1171 } else if (!strcmp(mode
, "writeback")) {
1172 *writethrough
= false;
1173 } else if (!strcmp(mode
, "unsafe")) {
1174 *writethrough
= false;
1175 *flags
|= BDRV_O_NO_FLUSH
;
1176 } else if (!strcmp(mode
, "writethrough")) {
1177 *writethrough
= true;
1185 static char *bdrv_child_get_parent_desc(BdrvChild
*c
)
1187 BlockDriverState
*parent
= c
->opaque
;
1188 return g_strdup_printf("node '%s'", bdrv_get_node_name(parent
));
1191 static void bdrv_child_cb_drained_begin(BdrvChild
*child
)
1193 BlockDriverState
*bs
= child
->opaque
;
1194 bdrv_do_drained_begin_quiesce(bs
, NULL
);
1197 static bool bdrv_child_cb_drained_poll(BdrvChild
*child
)
1199 BlockDriverState
*bs
= child
->opaque
;
1200 return bdrv_drain_poll(bs
, NULL
, false);
1203 static void bdrv_child_cb_drained_end(BdrvChild
*child
)
1205 BlockDriverState
*bs
= child
->opaque
;
1206 bdrv_drained_end(bs
);
1209 static int bdrv_child_cb_inactivate(BdrvChild
*child
)
1211 BlockDriverState
*bs
= child
->opaque
;
1212 GLOBAL_STATE_CODE();
1213 assert(bs
->open_flags
& BDRV_O_INACTIVE
);
1217 static bool bdrv_child_cb_change_aio_ctx(BdrvChild
*child
, AioContext
*ctx
,
1218 GHashTable
*visited
, Transaction
*tran
,
1221 BlockDriverState
*bs
= child
->opaque
;
1222 return bdrv_change_aio_context(bs
, ctx
, visited
, tran
, errp
);
1226 * Returns the options and flags that a temporary snapshot should get, based on
1227 * the originally requested flags (the originally requested image will have
1228 * flags like a backing file)
1230 static void bdrv_temp_snapshot_options(int *child_flags
, QDict
*child_options
,
1231 int parent_flags
, QDict
*parent_options
)
1233 GLOBAL_STATE_CODE();
1234 *child_flags
= (parent_flags
& ~BDRV_O_SNAPSHOT
) | BDRV_O_TEMPORARY
;
1236 /* For temporary files, unconditional cache=unsafe is fine */
1237 qdict_set_default_str(child_options
, BDRV_OPT_CACHE_DIRECT
, "off");
1238 qdict_set_default_str(child_options
, BDRV_OPT_CACHE_NO_FLUSH
, "on");
1240 /* Copy the read-only and discard options from the parent */
1241 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_READ_ONLY
);
1242 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_DISCARD
);
1244 /* aio=native doesn't work for cache.direct=off, so disable it for the
1245 * temporary snapshot */
1246 *child_flags
&= ~BDRV_O_NATIVE_AIO
;
1249 static void bdrv_backing_attach(BdrvChild
*c
)
1251 BlockDriverState
*parent
= c
->opaque
;
1252 BlockDriverState
*backing_hd
= c
->bs
;
1254 GLOBAL_STATE_CODE();
1255 assert(!parent
->backing_blocker
);
1256 error_setg(&parent
->backing_blocker
,
1257 "node is used as backing hd of '%s'",
1258 bdrv_get_device_or_node_name(parent
));
1260 bdrv_refresh_filename(backing_hd
);
1262 parent
->open_flags
&= ~BDRV_O_NO_BACKING
;
1264 bdrv_op_block_all(backing_hd
, parent
->backing_blocker
);
1265 /* Otherwise we won't be able to commit or stream */
1266 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_COMMIT_TARGET
,
1267 parent
->backing_blocker
);
1268 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_STREAM
,
1269 parent
->backing_blocker
);
1271 * We do backup in 3 ways:
1273 * The target bs is new opened, and the source is top BDS
1274 * 2. blockdev backup
1275 * Both the source and the target are top BDSes.
1276 * 3. internal backup(used for block replication)
1277 * Both the source and the target are backing file
1279 * In case 1 and 2, neither the source nor the target is the backing file.
1280 * In case 3, we will block the top BDS, so there is only one block job
1281 * for the top BDS and its backing chain.
1283 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_BACKUP_SOURCE
,
1284 parent
->backing_blocker
);
1285 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_BACKUP_TARGET
,
1286 parent
->backing_blocker
);
1289 static void bdrv_backing_detach(BdrvChild
*c
)
1291 BlockDriverState
*parent
= c
->opaque
;
1293 GLOBAL_STATE_CODE();
1294 assert(parent
->backing_blocker
);
1295 bdrv_op_unblock_all(c
->bs
, parent
->backing_blocker
);
1296 error_free(parent
->backing_blocker
);
1297 parent
->backing_blocker
= NULL
;
1300 static int bdrv_backing_update_filename(BdrvChild
*c
, BlockDriverState
*base
,
1301 const char *filename
, Error
**errp
)
1303 BlockDriverState
*parent
= c
->opaque
;
1304 bool read_only
= bdrv_is_read_only(parent
);
1306 GLOBAL_STATE_CODE();
1309 ret
= bdrv_reopen_set_read_only(parent
, false, errp
);
1315 ret
= bdrv_change_backing_file(parent
, filename
,
1316 base
->drv
? base
->drv
->format_name
: "",
1319 error_setg_errno(errp
, -ret
, "Could not update backing file link");
1323 bdrv_reopen_set_read_only(parent
, true, NULL
);
1330 * Returns the options and flags that a generic child of a BDS should
1331 * get, based on the given options and flags for the parent BDS.
1333 static void bdrv_inherited_options(BdrvChildRole role
, bool parent_is_format
,
1334 int *child_flags
, QDict
*child_options
,
1335 int parent_flags
, QDict
*parent_options
)
1337 int flags
= parent_flags
;
1338 GLOBAL_STATE_CODE();
1341 * First, decide whether to set, clear, or leave BDRV_O_PROTOCOL.
1342 * Generally, the question to answer is: Should this child be
1343 * format-probed by default?
1347 * Pure and non-filtered data children of non-format nodes should
1348 * be probed by default (even when the node itself has BDRV_O_PROTOCOL
1349 * set). This only affects a very limited set of drivers (namely
1350 * quorum and blkverify when this comment was written).
1351 * Force-clear BDRV_O_PROTOCOL then.
1353 if (!parent_is_format
&&
1354 (role
& BDRV_CHILD_DATA
) &&
1355 !(role
& (BDRV_CHILD_METADATA
| BDRV_CHILD_FILTERED
)))
1357 flags
&= ~BDRV_O_PROTOCOL
;
1361 * All children of format nodes (except for COW children) and all
1362 * metadata children in general should never be format-probed.
1363 * Force-set BDRV_O_PROTOCOL then.
1365 if ((parent_is_format
&& !(role
& BDRV_CHILD_COW
)) ||
1366 (role
& BDRV_CHILD_METADATA
))
1368 flags
|= BDRV_O_PROTOCOL
;
1372 * If the cache mode isn't explicitly set, inherit direct and no-flush from
1375 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_CACHE_DIRECT
);
1376 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_CACHE_NO_FLUSH
);
1377 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_FORCE_SHARE
);
1379 if (role
& BDRV_CHILD_COW
) {
1380 /* backing files are opened read-only by default */
1381 qdict_set_default_str(child_options
, BDRV_OPT_READ_ONLY
, "on");
1382 qdict_set_default_str(child_options
, BDRV_OPT_AUTO_READ_ONLY
, "off");
1384 /* Inherit the read-only option from the parent if it's not set */
1385 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_READ_ONLY
);
1386 qdict_copy_default(child_options
, parent_options
,
1387 BDRV_OPT_AUTO_READ_ONLY
);
1391 * bdrv_co_pdiscard() respects unmap policy for the parent, so we
1392 * can default to enable it on lower layers regardless of the
1395 qdict_set_default_str(child_options
, BDRV_OPT_DISCARD
, "unmap");
1397 /* Clear flags that only apply to the top layer */
1398 flags
&= ~(BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
| BDRV_O_COPY_ON_READ
);
1400 if (role
& BDRV_CHILD_METADATA
) {
1401 flags
&= ~BDRV_O_NO_IO
;
1403 if (role
& BDRV_CHILD_COW
) {
1404 flags
&= ~BDRV_O_TEMPORARY
;
1407 *child_flags
= flags
;
1410 static void GRAPH_WRLOCK
bdrv_child_cb_attach(BdrvChild
*child
)
1412 BlockDriverState
*bs
= child
->opaque
;
1414 assert_bdrv_graph_writable();
1415 QLIST_INSERT_HEAD(&bs
->children
, child
, next
);
1416 if (bs
->drv
->is_filter
|| (child
->role
& BDRV_CHILD_FILTERED
)) {
1418 * Here we handle filters and block/raw-format.c when it behave like
1419 * filter. They generally have a single PRIMARY child, which is also the
1420 * FILTERED child, and that they may have multiple more children, which
1421 * are neither PRIMARY nor FILTERED. And never we have a COW child here.
1422 * So bs->file will be the PRIMARY child, unless the PRIMARY child goes
1423 * into bs->backing on exceptional cases; and bs->backing will be
1426 assert(!(child
->role
& BDRV_CHILD_COW
));
1427 if (child
->role
& BDRV_CHILD_PRIMARY
) {
1428 assert(child
->role
& BDRV_CHILD_FILTERED
);
1429 assert(!bs
->backing
);
1432 if (bs
->drv
->filtered_child_is_backing
) {
1433 bs
->backing
= child
;
1438 assert(!(child
->role
& BDRV_CHILD_FILTERED
));
1440 } else if (child
->role
& BDRV_CHILD_COW
) {
1441 assert(bs
->drv
->supports_backing
);
1442 assert(!(child
->role
& BDRV_CHILD_PRIMARY
));
1443 assert(!bs
->backing
);
1444 bs
->backing
= child
;
1445 bdrv_backing_attach(child
);
1446 } else if (child
->role
& BDRV_CHILD_PRIMARY
) {
1452 static void GRAPH_WRLOCK
bdrv_child_cb_detach(BdrvChild
*child
)
1454 BlockDriverState
*bs
= child
->opaque
;
1456 if (child
->role
& BDRV_CHILD_COW
) {
1457 bdrv_backing_detach(child
);
1460 assert_bdrv_graph_writable();
1461 QLIST_REMOVE(child
, next
);
1462 if (child
== bs
->backing
) {
1463 assert(child
!= bs
->file
);
1465 } else if (child
== bs
->file
) {
1470 static int bdrv_child_cb_update_filename(BdrvChild
*c
, BlockDriverState
*base
,
1471 const char *filename
, Error
**errp
)
1473 if (c
->role
& BDRV_CHILD_COW
) {
1474 return bdrv_backing_update_filename(c
, base
, filename
, errp
);
1479 AioContext
*child_of_bds_get_parent_aio_context(BdrvChild
*c
)
1481 BlockDriverState
*bs
= c
->opaque
;
1484 return bdrv_get_aio_context(bs
);
1487 const BdrvChildClass child_of_bds
= {
1488 .parent_is_bds
= true,
1489 .get_parent_desc
= bdrv_child_get_parent_desc
,
1490 .inherit_options
= bdrv_inherited_options
,
1491 .drained_begin
= bdrv_child_cb_drained_begin
,
1492 .drained_poll
= bdrv_child_cb_drained_poll
,
1493 .drained_end
= bdrv_child_cb_drained_end
,
1494 .attach
= bdrv_child_cb_attach
,
1495 .detach
= bdrv_child_cb_detach
,
1496 .inactivate
= bdrv_child_cb_inactivate
,
1497 .change_aio_ctx
= bdrv_child_cb_change_aio_ctx
,
1498 .update_filename
= bdrv_child_cb_update_filename
,
1499 .get_parent_aio_context
= child_of_bds_get_parent_aio_context
,
1502 AioContext
*bdrv_child_get_parent_aio_context(BdrvChild
*c
)
1505 return c
->klass
->get_parent_aio_context(c
);
1508 static int bdrv_open_flags(BlockDriverState
*bs
, int flags
)
1510 int open_flags
= flags
;
1511 GLOBAL_STATE_CODE();
1514 * Clear flags that are internal to the block layer before opening the
1517 open_flags
&= ~(BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
| BDRV_O_PROTOCOL
);
1522 static void update_flags_from_options(int *flags
, QemuOpts
*opts
)
1524 GLOBAL_STATE_CODE();
1526 *flags
&= ~(BDRV_O_CACHE_MASK
| BDRV_O_RDWR
| BDRV_O_AUTO_RDONLY
);
1528 if (qemu_opt_get_bool_del(opts
, BDRV_OPT_CACHE_NO_FLUSH
, false)) {
1529 *flags
|= BDRV_O_NO_FLUSH
;
1532 if (qemu_opt_get_bool_del(opts
, BDRV_OPT_CACHE_DIRECT
, false)) {
1533 *flags
|= BDRV_O_NOCACHE
;
1536 if (!qemu_opt_get_bool_del(opts
, BDRV_OPT_READ_ONLY
, false)) {
1537 *flags
|= BDRV_O_RDWR
;
1540 if (qemu_opt_get_bool_del(opts
, BDRV_OPT_AUTO_READ_ONLY
, false)) {
1541 *flags
|= BDRV_O_AUTO_RDONLY
;
1545 static void update_options_from_flags(QDict
*options
, int flags
)
1547 GLOBAL_STATE_CODE();
1548 if (!qdict_haskey(options
, BDRV_OPT_CACHE_DIRECT
)) {
1549 qdict_put_bool(options
, BDRV_OPT_CACHE_DIRECT
, flags
& BDRV_O_NOCACHE
);
1551 if (!qdict_haskey(options
, BDRV_OPT_CACHE_NO_FLUSH
)) {
1552 qdict_put_bool(options
, BDRV_OPT_CACHE_NO_FLUSH
,
1553 flags
& BDRV_O_NO_FLUSH
);
1555 if (!qdict_haskey(options
, BDRV_OPT_READ_ONLY
)) {
1556 qdict_put_bool(options
, BDRV_OPT_READ_ONLY
, !(flags
& BDRV_O_RDWR
));
1558 if (!qdict_haskey(options
, BDRV_OPT_AUTO_READ_ONLY
)) {
1559 qdict_put_bool(options
, BDRV_OPT_AUTO_READ_ONLY
,
1560 flags
& BDRV_O_AUTO_RDONLY
);
1564 static void bdrv_assign_node_name(BlockDriverState
*bs
,
1565 const char *node_name
,
1568 char *gen_node_name
= NULL
;
1569 GLOBAL_STATE_CODE();
1572 node_name
= gen_node_name
= id_generate(ID_BLOCK
);
1573 } else if (!id_wellformed(node_name
)) {
1575 * Check for empty string or invalid characters, but not if it is
1576 * generated (generated names use characters not available to the user)
1578 error_setg(errp
, "Invalid node-name: '%s'", node_name
);
1582 /* takes care of avoiding namespaces collisions */
1583 if (blk_by_name(node_name
)) {
1584 error_setg(errp
, "node-name=%s is conflicting with a device id",
1589 /* takes care of avoiding duplicates node names */
1590 if (bdrv_find_node(node_name
)) {
1591 error_setg(errp
, "Duplicate nodes with node-name='%s'", node_name
);
1595 /* Make sure that the node name isn't truncated */
1596 if (strlen(node_name
) >= sizeof(bs
->node_name
)) {
1597 error_setg(errp
, "Node name too long");
1601 /* copy node name into the bs and insert it into the graph list */
1602 pstrcpy(bs
->node_name
, sizeof(bs
->node_name
), node_name
);
1603 QTAILQ_INSERT_TAIL(&graph_bdrv_states
, bs
, node_list
);
1605 g_free(gen_node_name
);
1609 * The caller must always hold @bs AioContext lock, because this function calls
1610 * bdrv_refresh_total_sectors() which polls when called from non-coroutine
1613 static int bdrv_open_driver(BlockDriverState
*bs
, BlockDriver
*drv
,
1614 const char *node_name
, QDict
*options
,
1615 int open_flags
, Error
**errp
)
1617 Error
*local_err
= NULL
;
1619 GLOBAL_STATE_CODE();
1621 bdrv_assign_node_name(bs
, node_name
, &local_err
);
1623 error_propagate(errp
, local_err
);
1628 bs
->opaque
= g_malloc0(drv
->instance_size
);
1630 if (drv
->bdrv_file_open
) {
1631 assert(!drv
->bdrv_needs_filename
|| bs
->filename
[0]);
1632 ret
= drv
->bdrv_file_open(bs
, options
, open_flags
, &local_err
);
1633 } else if (drv
->bdrv_open
) {
1634 ret
= drv
->bdrv_open(bs
, options
, open_flags
, &local_err
);
1641 error_propagate(errp
, local_err
);
1642 } else if (bs
->filename
[0]) {
1643 error_setg_errno(errp
, -ret
, "Could not open '%s'", bs
->filename
);
1645 error_setg_errno(errp
, -ret
, "Could not open image");
1650 assert(!(bs
->supported_read_flags
& ~BDRV_REQ_MASK
));
1651 assert(!(bs
->supported_write_flags
& ~BDRV_REQ_MASK
));
1654 * Always allow the BDRV_REQ_REGISTERED_BUF optimization hint. This saves
1655 * drivers that pass read/write requests through to a child the trouble of
1656 * declaring support explicitly.
1658 * Drivers must not propagate this flag accidentally when they initiate I/O
1659 * to a bounce buffer. That case should be rare though.
1661 bs
->supported_read_flags
|= BDRV_REQ_REGISTERED_BUF
;
1662 bs
->supported_write_flags
|= BDRV_REQ_REGISTERED_BUF
;
1664 ret
= bdrv_refresh_total_sectors(bs
, bs
->total_sectors
);
1666 error_setg_errno(errp
, -ret
, "Could not refresh total sector count");
1670 bdrv_refresh_limits(bs
, NULL
, &local_err
);
1672 error_propagate(errp
, local_err
);
1676 assert(bdrv_opt_mem_align(bs
) != 0);
1677 assert(bdrv_min_mem_align(bs
) != 0);
1678 assert(is_power_of_2(bs
->bl
.request_alignment
));
1680 for (i
= 0; i
< bs
->quiesce_counter
; i
++) {
1681 if (drv
->bdrv_drain_begin
) {
1682 drv
->bdrv_drain_begin(bs
);
1689 if (bs
->file
!= NULL
) {
1690 bdrv_unref_child(bs
, bs
->file
);
1699 * Create and open a block node.
1701 * @options is a QDict of options to pass to the block drivers, or NULL for an
1702 * empty set of options. The reference to the QDict belongs to the block layer
1703 * after the call (even on failure), so if the caller intends to reuse the
1704 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
1706 BlockDriverState
*bdrv_new_open_driver_opts(BlockDriver
*drv
,
1707 const char *node_name
,
1708 QDict
*options
, int flags
,
1711 BlockDriverState
*bs
;
1714 GLOBAL_STATE_CODE();
1717 bs
->open_flags
= flags
;
1718 bs
->options
= options
?: qdict_new();
1719 bs
->explicit_options
= qdict_clone_shallow(bs
->options
);
1722 update_options_from_flags(bs
->options
, flags
);
1724 ret
= bdrv_open_driver(bs
, drv
, node_name
, bs
->options
, flags
, errp
);
1726 qobject_unref(bs
->explicit_options
);
1727 bs
->explicit_options
= NULL
;
1728 qobject_unref(bs
->options
);
1737 /* Create and open a block node. */
1738 BlockDriverState
*bdrv_new_open_driver(BlockDriver
*drv
, const char *node_name
,
1739 int flags
, Error
**errp
)
1741 GLOBAL_STATE_CODE();
1742 return bdrv_new_open_driver_opts(drv
, node_name
, NULL
, flags
, errp
);
1745 QemuOptsList bdrv_runtime_opts
= {
1746 .name
= "bdrv_common",
1747 .head
= QTAILQ_HEAD_INITIALIZER(bdrv_runtime_opts
.head
),
1750 .name
= "node-name",
1751 .type
= QEMU_OPT_STRING
,
1752 .help
= "Node name of the block device node",
1756 .type
= QEMU_OPT_STRING
,
1757 .help
= "Block driver to use for the node",
1760 .name
= BDRV_OPT_CACHE_DIRECT
,
1761 .type
= QEMU_OPT_BOOL
,
1762 .help
= "Bypass software writeback cache on the host",
1765 .name
= BDRV_OPT_CACHE_NO_FLUSH
,
1766 .type
= QEMU_OPT_BOOL
,
1767 .help
= "Ignore flush requests",
1770 .name
= BDRV_OPT_READ_ONLY
,
1771 .type
= QEMU_OPT_BOOL
,
1772 .help
= "Node is opened in read-only mode",
1775 .name
= BDRV_OPT_AUTO_READ_ONLY
,
1776 .type
= QEMU_OPT_BOOL
,
1777 .help
= "Node can become read-only if opening read-write fails",
1780 .name
= "detect-zeroes",
1781 .type
= QEMU_OPT_STRING
,
1782 .help
= "try to optimize zero writes (off, on, unmap)",
1785 .name
= BDRV_OPT_DISCARD
,
1786 .type
= QEMU_OPT_STRING
,
1787 .help
= "discard operation (ignore/off, unmap/on)",
1790 .name
= BDRV_OPT_FORCE_SHARE
,
1791 .type
= QEMU_OPT_BOOL
,
1792 .help
= "always accept other writers (default: off)",
1794 { /* end of list */ }
1798 QemuOptsList bdrv_create_opts_simple
= {
1799 .name
= "simple-create-opts",
1800 .head
= QTAILQ_HEAD_INITIALIZER(bdrv_create_opts_simple
.head
),
1803 .name
= BLOCK_OPT_SIZE
,
1804 .type
= QEMU_OPT_SIZE
,
1805 .help
= "Virtual disk size"
1808 .name
= BLOCK_OPT_PREALLOC
,
1809 .type
= QEMU_OPT_STRING
,
1810 .help
= "Preallocation mode (allowed values: off)"
1812 { /* end of list */ }
1817 * Common part for opening disk images and files
1819 * Removes all processed options from *options.
1821 static int bdrv_open_common(BlockDriverState
*bs
, BlockBackend
*file
,
1822 QDict
*options
, Error
**errp
)
1824 int ret
, open_flags
;
1825 const char *filename
;
1826 const char *driver_name
= NULL
;
1827 const char *node_name
= NULL
;
1828 const char *discard
;
1831 Error
*local_err
= NULL
;
1834 assert(bs
->file
== NULL
);
1835 assert(options
!= NULL
&& bs
->options
!= options
);
1836 GLOBAL_STATE_CODE();
1838 opts
= qemu_opts_create(&bdrv_runtime_opts
, NULL
, 0, &error_abort
);
1839 if (!qemu_opts_absorb_qdict(opts
, options
, errp
)) {
1844 update_flags_from_options(&bs
->open_flags
, opts
);
1846 driver_name
= qemu_opt_get(opts
, "driver");
1847 drv
= bdrv_find_format(driver_name
);
1848 assert(drv
!= NULL
);
1850 bs
->force_share
= qemu_opt_get_bool(opts
, BDRV_OPT_FORCE_SHARE
, false);
1852 if (bs
->force_share
&& (bs
->open_flags
& BDRV_O_RDWR
)) {
1854 BDRV_OPT_FORCE_SHARE
1855 "=on can only be used with read-only images");
1861 bdrv_refresh_filename(blk_bs(file
));
1862 filename
= blk_bs(file
)->filename
;
1865 * Caution: while qdict_get_try_str() is fine, getting
1866 * non-string types would require more care. When @options
1867 * come from -blockdev or blockdev_add, its members are typed
1868 * according to the QAPI schema, but when they come from
1869 * -drive, they're all QString.
1871 filename
= qdict_get_try_str(options
, "filename");
1874 if (drv
->bdrv_needs_filename
&& (!filename
|| !filename
[0])) {
1875 error_setg(errp
, "The '%s' block driver requires a file name",
1881 trace_bdrv_open_common(bs
, filename
?: "", bs
->open_flags
,
1884 ro
= bdrv_is_read_only(bs
);
1886 if (use_bdrv_whitelist
&& !bdrv_is_whitelisted(drv
, ro
)) {
1887 if (!ro
&& bdrv_is_whitelisted(drv
, true)) {
1888 ret
= bdrv_apply_auto_read_only(bs
, NULL
, NULL
);
1894 !ro
&& bdrv_is_whitelisted(drv
, true)
1895 ? "Driver '%s' can only be used for read-only devices"
1896 : "Driver '%s' is not whitelisted",
1902 /* bdrv_new() and bdrv_close() make it so */
1903 assert(qatomic_read(&bs
->copy_on_read
) == 0);
1905 if (bs
->open_flags
& BDRV_O_COPY_ON_READ
) {
1907 bdrv_enable_copy_on_read(bs
);
1909 error_setg(errp
, "Can't use copy-on-read on read-only device");
1915 discard
= qemu_opt_get(opts
, BDRV_OPT_DISCARD
);
1916 if (discard
!= NULL
) {
1917 if (bdrv_parse_discard_flags(discard
, &bs
->open_flags
) != 0) {
1918 error_setg(errp
, "Invalid discard option");
1925 bdrv_parse_detect_zeroes(opts
, bs
->open_flags
, &local_err
);
1927 error_propagate(errp
, local_err
);
1932 if (filename
!= NULL
) {
1933 pstrcpy(bs
->filename
, sizeof(bs
->filename
), filename
);
1935 bs
->filename
[0] = '\0';
1937 pstrcpy(bs
->exact_filename
, sizeof(bs
->exact_filename
), bs
->filename
);
1939 /* Open the image, either directly or using a protocol */
1940 open_flags
= bdrv_open_flags(bs
, bs
->open_flags
);
1941 node_name
= qemu_opt_get(opts
, "node-name");
1943 assert(!drv
->bdrv_file_open
|| file
== NULL
);
1944 ret
= bdrv_open_driver(bs
, drv
, node_name
, options
, open_flags
, errp
);
1949 qemu_opts_del(opts
);
1953 qemu_opts_del(opts
);
1957 static G_GNUC_PRINTF(1, 0)
1958 QDict
*parse_json_filename(const char *filename
, Error
**errp
)
1960 QObject
*options_obj
;
1963 GLOBAL_STATE_CODE();
1965 ret
= strstart(filename
, "json:", &filename
);
1968 options_obj
= qobject_from_json(filename
, errp
);
1970 error_prepend(errp
, "Could not parse the JSON options: ");
1974 options
= qobject_to(QDict
, options_obj
);
1976 qobject_unref(options_obj
);
1977 error_setg(errp
, "Invalid JSON object given");
1981 qdict_flatten(options
);
1986 static void parse_json_protocol(QDict
*options
, const char **pfilename
,
1989 QDict
*json_options
;
1990 Error
*local_err
= NULL
;
1991 GLOBAL_STATE_CODE();
1993 /* Parse json: pseudo-protocol */
1994 if (!*pfilename
|| !g_str_has_prefix(*pfilename
, "json:")) {
1998 json_options
= parse_json_filename(*pfilename
, &local_err
);
2000 error_propagate(errp
, local_err
);
2004 /* Options given in the filename have lower priority than options
2005 * specified directly */
2006 qdict_join(options
, json_options
, false);
2007 qobject_unref(json_options
);
2012 * Fills in default options for opening images and converts the legacy
2013 * filename/flags pair to option QDict entries.
2014 * The BDRV_O_PROTOCOL flag in *flags will be set or cleared accordingly if a
2015 * block driver has been specified explicitly.
2017 static int bdrv_fill_options(QDict
**options
, const char *filename
,
2018 int *flags
, Error
**errp
)
2020 const char *drvname
;
2021 bool protocol
= *flags
& BDRV_O_PROTOCOL
;
2022 bool parse_filename
= false;
2023 BlockDriver
*drv
= NULL
;
2024 Error
*local_err
= NULL
;
2026 GLOBAL_STATE_CODE();
2029 * Caution: while qdict_get_try_str() is fine, getting non-string
2030 * types would require more care. When @options come from
2031 * -blockdev or blockdev_add, its members are typed according to
2032 * the QAPI schema, but when they come from -drive, they're all
2035 drvname
= qdict_get_try_str(*options
, "driver");
2037 drv
= bdrv_find_format(drvname
);
2039 error_setg(errp
, "Unknown driver '%s'", drvname
);
2042 /* If the user has explicitly specified the driver, this choice should
2043 * override the BDRV_O_PROTOCOL flag */
2044 protocol
= drv
->bdrv_file_open
;
2048 *flags
|= BDRV_O_PROTOCOL
;
2050 *flags
&= ~BDRV_O_PROTOCOL
;
2053 /* Translate cache options from flags into options */
2054 update_options_from_flags(*options
, *flags
);
2056 /* Fetch the file name from the options QDict if necessary */
2057 if (protocol
&& filename
) {
2058 if (!qdict_haskey(*options
, "filename")) {
2059 qdict_put_str(*options
, "filename", filename
);
2060 parse_filename
= true;
2062 error_setg(errp
, "Can't specify 'file' and 'filename' options at "
2068 /* Find the right block driver */
2069 /* See cautionary note on accessing @options above */
2070 filename
= qdict_get_try_str(*options
, "filename");
2072 if (!drvname
&& protocol
) {
2074 drv
= bdrv_find_protocol(filename
, parse_filename
, errp
);
2079 drvname
= drv
->format_name
;
2080 qdict_put_str(*options
, "driver", drvname
);
2082 error_setg(errp
, "Must specify either driver or file");
2087 assert(drv
|| !protocol
);
2089 /* Driver-specific filename parsing */
2090 if (drv
&& drv
->bdrv_parse_filename
&& parse_filename
) {
2091 drv
->bdrv_parse_filename(filename
, *options
, &local_err
);
2093 error_propagate(errp
, local_err
);
2097 if (!drv
->bdrv_needs_filename
) {
2098 qdict_del(*options
, "filename");
2105 typedef struct BlockReopenQueueEntry
{
2108 BDRVReopenState state
;
2109 QTAILQ_ENTRY(BlockReopenQueueEntry
) entry
;
2110 } BlockReopenQueueEntry
;
2113 * Return the flags that @bs will have after the reopens in @q have
2114 * successfully completed. If @q is NULL (or @bs is not contained in @q),
2115 * return the current flags.
2117 static int bdrv_reopen_get_flags(BlockReopenQueue
*q
, BlockDriverState
*bs
)
2119 BlockReopenQueueEntry
*entry
;
2122 QTAILQ_FOREACH(entry
, q
, entry
) {
2123 if (entry
->state
.bs
== bs
) {
2124 return entry
->state
.flags
;
2129 return bs
->open_flags
;
2132 /* Returns whether the image file can be written to after the reopen queue @q
2133 * has been successfully applied, or right now if @q is NULL. */
2134 static bool bdrv_is_writable_after_reopen(BlockDriverState
*bs
,
2135 BlockReopenQueue
*q
)
2137 int flags
= bdrv_reopen_get_flags(q
, bs
);
2139 return (flags
& (BDRV_O_RDWR
| BDRV_O_INACTIVE
)) == BDRV_O_RDWR
;
2143 * Return whether the BDS can be written to. This is not necessarily
2144 * the same as !bdrv_is_read_only(bs), as inactivated images may not
2145 * be written to but do not count as read-only images.
2147 bool bdrv_is_writable(BlockDriverState
*bs
)
2150 return bdrv_is_writable_after_reopen(bs
, NULL
);
2153 static char *bdrv_child_user_desc(BdrvChild
*c
)
2155 GLOBAL_STATE_CODE();
2156 return c
->klass
->get_parent_desc(c
);
2160 * Check that @a allows everything that @b needs. @a and @b must reference same
2163 static bool bdrv_a_allow_b(BdrvChild
*a
, BdrvChild
*b
, Error
**errp
)
2165 const char *child_bs_name
;
2166 g_autofree
char *a_user
= NULL
;
2167 g_autofree
char *b_user
= NULL
;
2168 g_autofree
char *perms
= NULL
;
2171 assert(a
->bs
== b
->bs
);
2172 GLOBAL_STATE_CODE();
2174 if ((b
->perm
& a
->shared_perm
) == b
->perm
) {
2178 child_bs_name
= bdrv_get_node_name(b
->bs
);
2179 a_user
= bdrv_child_user_desc(a
);
2180 b_user
= bdrv_child_user_desc(b
);
2181 perms
= bdrv_perm_names(b
->perm
& ~a
->shared_perm
);
2183 error_setg(errp
, "Permission conflict on node '%s': permissions '%s' are "
2184 "both required by %s (uses node '%s' as '%s' child) and "
2185 "unshared by %s (uses node '%s' as '%s' child).",
2186 child_bs_name
, perms
,
2187 b_user
, child_bs_name
, b
->name
,
2188 a_user
, child_bs_name
, a
->name
);
2193 static bool bdrv_parent_perms_conflict(BlockDriverState
*bs
, Error
**errp
)
2196 GLOBAL_STATE_CODE();
2199 * During the loop we'll look at each pair twice. That's correct because
2200 * bdrv_a_allow_b() is asymmetric and we should check each pair in both
2203 QLIST_FOREACH(a
, &bs
->parents
, next_parent
) {
2204 QLIST_FOREACH(b
, &bs
->parents
, next_parent
) {
2209 if (!bdrv_a_allow_b(a
, b
, errp
)) {
2218 static void bdrv_child_perm(BlockDriverState
*bs
, BlockDriverState
*child_bs
,
2219 BdrvChild
*c
, BdrvChildRole role
,
2220 BlockReopenQueue
*reopen_queue
,
2221 uint64_t parent_perm
, uint64_t parent_shared
,
2222 uint64_t *nperm
, uint64_t *nshared
)
2224 assert(bs
->drv
&& bs
->drv
->bdrv_child_perm
);
2225 GLOBAL_STATE_CODE();
2226 bs
->drv
->bdrv_child_perm(bs
, c
, role
, reopen_queue
,
2227 parent_perm
, parent_shared
,
2229 /* TODO Take force_share from reopen_queue */
2230 if (child_bs
&& child_bs
->force_share
) {
2231 *nshared
= BLK_PERM_ALL
;
2236 * Adds the whole subtree of @bs (including @bs itself) to the @list (except for
2237 * nodes that are already in the @list, of course) so that final list is
2238 * topologically sorted. Return the result (GSList @list object is updated, so
2239 * don't use old reference after function call).
2241 * On function start @list must be already topologically sorted and for any node
2242 * in the @list the whole subtree of the node must be in the @list as well. The
2243 * simplest way to satisfy this criteria: use only result of
2244 * bdrv_topological_dfs() or NULL as @list parameter.
2246 static GSList
*bdrv_topological_dfs(GSList
*list
, GHashTable
*found
,
2247 BlockDriverState
*bs
)
2250 g_autoptr(GHashTable
) local_found
= NULL
;
2252 GLOBAL_STATE_CODE();
2256 found
= local_found
= g_hash_table_new(NULL
, NULL
);
2259 if (g_hash_table_contains(found
, bs
)) {
2262 g_hash_table_add(found
, bs
);
2264 QLIST_FOREACH(child
, &bs
->children
, next
) {
2265 list
= bdrv_topological_dfs(list
, found
, child
->bs
);
2268 return g_slist_prepend(list
, bs
);
2271 typedef struct BdrvChildSetPermState
{
2274 uint64_t old_shared_perm
;
2275 } BdrvChildSetPermState
;
2277 static void bdrv_child_set_perm_abort(void *opaque
)
2279 BdrvChildSetPermState
*s
= opaque
;
2281 GLOBAL_STATE_CODE();
2283 s
->child
->perm
= s
->old_perm
;
2284 s
->child
->shared_perm
= s
->old_shared_perm
;
2287 static TransactionActionDrv bdrv_child_set_pem_drv
= {
2288 .abort
= bdrv_child_set_perm_abort
,
2292 static void bdrv_child_set_perm(BdrvChild
*c
, uint64_t perm
,
2293 uint64_t shared
, Transaction
*tran
)
2295 BdrvChildSetPermState
*s
= g_new(BdrvChildSetPermState
, 1);
2296 GLOBAL_STATE_CODE();
2298 *s
= (BdrvChildSetPermState
) {
2300 .old_perm
= c
->perm
,
2301 .old_shared_perm
= c
->shared_perm
,
2305 c
->shared_perm
= shared
;
2307 tran_add(tran
, &bdrv_child_set_pem_drv
, s
);
2310 static void bdrv_drv_set_perm_commit(void *opaque
)
2312 BlockDriverState
*bs
= opaque
;
2313 uint64_t cumulative_perms
, cumulative_shared_perms
;
2314 GLOBAL_STATE_CODE();
2316 if (bs
->drv
->bdrv_set_perm
) {
2317 bdrv_get_cumulative_perm(bs
, &cumulative_perms
,
2318 &cumulative_shared_perms
);
2319 bs
->drv
->bdrv_set_perm(bs
, cumulative_perms
, cumulative_shared_perms
);
2323 static void bdrv_drv_set_perm_abort(void *opaque
)
2325 BlockDriverState
*bs
= opaque
;
2326 GLOBAL_STATE_CODE();
2328 if (bs
->drv
->bdrv_abort_perm_update
) {
2329 bs
->drv
->bdrv_abort_perm_update(bs
);
2333 TransactionActionDrv bdrv_drv_set_perm_drv
= {
2334 .abort
= bdrv_drv_set_perm_abort
,
2335 .commit
= bdrv_drv_set_perm_commit
,
2338 static int bdrv_drv_set_perm(BlockDriverState
*bs
, uint64_t perm
,
2339 uint64_t shared_perm
, Transaction
*tran
,
2342 GLOBAL_STATE_CODE();
2347 if (bs
->drv
->bdrv_check_perm
) {
2348 int ret
= bs
->drv
->bdrv_check_perm(bs
, perm
, shared_perm
, errp
);
2355 tran_add(tran
, &bdrv_drv_set_perm_drv
, bs
);
2361 typedef struct BdrvReplaceChildState
{
2363 BlockDriverState
*old_bs
;
2364 } BdrvReplaceChildState
;
2366 static void bdrv_replace_child_commit(void *opaque
)
2368 BdrvReplaceChildState
*s
= opaque
;
2369 GLOBAL_STATE_CODE();
2371 bdrv_unref(s
->old_bs
);
2374 static void bdrv_replace_child_abort(void *opaque
)
2376 BdrvReplaceChildState
*s
= opaque
;
2377 BlockDriverState
*new_bs
= s
->child
->bs
;
2379 GLOBAL_STATE_CODE();
2380 /* old_bs reference is transparently moved from @s to @s->child */
2381 if (!s
->child
->bs
) {
2383 * The parents were undrained when removing old_bs from the child. New
2384 * requests can't have been made, though, because the child was empty.
2386 * TODO Make bdrv_replace_child_noperm() transactionable to avoid
2387 * undraining the parent in the first place. Once this is done, having
2388 * new_bs drained when calling bdrv_replace_child_tran() is not a
2389 * requirement any more.
2391 bdrv_parent_drained_begin_single(s
->child
);
2392 assert(!bdrv_parent_drained_poll_single(s
->child
));
2394 assert(s
->child
->quiesced_parent
);
2395 bdrv_replace_child_noperm(s
->child
, s
->old_bs
);
2399 static TransactionActionDrv bdrv_replace_child_drv
= {
2400 .commit
= bdrv_replace_child_commit
,
2401 .abort
= bdrv_replace_child_abort
,
2406 * bdrv_replace_child_tran
2408 * Note: real unref of old_bs is done only on commit.
2410 * Both @child->bs and @new_bs (if non-NULL) must be drained. @new_bs must be
2411 * kept drained until the transaction is completed.
2413 * The function doesn't update permissions, caller is responsible for this.
2415 static void bdrv_replace_child_tran(BdrvChild
*child
, BlockDriverState
*new_bs
,
2418 BdrvReplaceChildState
*s
= g_new(BdrvReplaceChildState
, 1);
2420 assert(child
->quiesced_parent
);
2421 assert(!new_bs
|| new_bs
->quiesce_counter
);
2423 *s
= (BdrvReplaceChildState
) {
2425 .old_bs
= child
->bs
,
2427 tran_add(tran
, &bdrv_replace_child_drv
, s
);
2432 bdrv_replace_child_noperm(child
, new_bs
);
2433 /* old_bs reference is transparently moved from @child to @s */
2437 * Refresh permissions in @bs subtree. The function is intended to be called
2438 * after some graph modification that was done without permission update.
2440 static int bdrv_node_refresh_perm(BlockDriverState
*bs
, BlockReopenQueue
*q
,
2441 Transaction
*tran
, Error
**errp
)
2443 BlockDriver
*drv
= bs
->drv
;
2446 uint64_t cumulative_perms
, cumulative_shared_perms
;
2447 GLOBAL_STATE_CODE();
2449 bdrv_get_cumulative_perm(bs
, &cumulative_perms
, &cumulative_shared_perms
);
2451 /* Write permissions never work with read-only images */
2452 if ((cumulative_perms
& (BLK_PERM_WRITE
| BLK_PERM_WRITE_UNCHANGED
)) &&
2453 !bdrv_is_writable_after_reopen(bs
, q
))
2455 if (!bdrv_is_writable_after_reopen(bs
, NULL
)) {
2456 error_setg(errp
, "Block node is read-only");
2458 error_setg(errp
, "Read-only block node '%s' cannot support "
2459 "read-write users", bdrv_get_node_name(bs
));
2466 * Unaligned requests will automatically be aligned to bl.request_alignment
2467 * and without RESIZE we can't extend requests to write to space beyond the
2468 * end of the image, so it's required that the image size is aligned.
2470 if ((cumulative_perms
& (BLK_PERM_WRITE
| BLK_PERM_WRITE_UNCHANGED
)) &&
2471 !(cumulative_perms
& BLK_PERM_RESIZE
))
2473 if ((bs
->total_sectors
* BDRV_SECTOR_SIZE
) % bs
->bl
.request_alignment
) {
2474 error_setg(errp
, "Cannot get 'write' permission without 'resize': "
2475 "Image size is not a multiple of request "
2481 /* Check this node */
2486 ret
= bdrv_drv_set_perm(bs
, cumulative_perms
, cumulative_shared_perms
, tran
,
2492 /* Drivers that never have children can omit .bdrv_child_perm() */
2493 if (!drv
->bdrv_child_perm
) {
2494 assert(QLIST_EMPTY(&bs
->children
));
2498 /* Check all children */
2499 QLIST_FOREACH(c
, &bs
->children
, next
) {
2500 uint64_t cur_perm
, cur_shared
;
2502 bdrv_child_perm(bs
, c
->bs
, c
, c
->role
, q
,
2503 cumulative_perms
, cumulative_shared_perms
,
2504 &cur_perm
, &cur_shared
);
2505 bdrv_child_set_perm(c
, cur_perm
, cur_shared
, tran
);
2512 * @list is a product of bdrv_topological_dfs() (may be called several times) -
2513 * a topologically sorted subgraph.
2515 static int bdrv_do_refresh_perms(GSList
*list
, BlockReopenQueue
*q
,
2516 Transaction
*tran
, Error
**errp
)
2519 BlockDriverState
*bs
;
2520 GLOBAL_STATE_CODE();
2522 for ( ; list
; list
= list
->next
) {
2525 if (bdrv_parent_perms_conflict(bs
, errp
)) {
2529 ret
= bdrv_node_refresh_perm(bs
, q
, tran
, errp
);
2539 * @list is any list of nodes. List is completed by all subtrees and
2540 * topologically sorted. It's not a problem if some node occurs in the @list
2543 static int bdrv_list_refresh_perms(GSList
*list
, BlockReopenQueue
*q
,
2544 Transaction
*tran
, Error
**errp
)
2546 g_autoptr(GHashTable
) found
= g_hash_table_new(NULL
, NULL
);
2547 g_autoptr(GSList
) refresh_list
= NULL
;
2549 for ( ; list
; list
= list
->next
) {
2550 refresh_list
= bdrv_topological_dfs(refresh_list
, found
, list
->data
);
2553 return bdrv_do_refresh_perms(refresh_list
, q
, tran
, errp
);
2556 void bdrv_get_cumulative_perm(BlockDriverState
*bs
, uint64_t *perm
,
2557 uint64_t *shared_perm
)
2560 uint64_t cumulative_perms
= 0;
2561 uint64_t cumulative_shared_perms
= BLK_PERM_ALL
;
2563 GLOBAL_STATE_CODE();
2565 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
2566 cumulative_perms
|= c
->perm
;
2567 cumulative_shared_perms
&= c
->shared_perm
;
2570 *perm
= cumulative_perms
;
2571 *shared_perm
= cumulative_shared_perms
;
2574 char *bdrv_perm_names(uint64_t perm
)
2580 { BLK_PERM_CONSISTENT_READ
, "consistent read" },
2581 { BLK_PERM_WRITE
, "write" },
2582 { BLK_PERM_WRITE_UNCHANGED
, "write unchanged" },
2583 { BLK_PERM_RESIZE
, "resize" },
2587 GString
*result
= g_string_sized_new(30);
2588 struct perm_name
*p
;
2590 for (p
= permissions
; p
->name
; p
++) {
2591 if (perm
& p
->perm
) {
2592 if (result
->len
> 0) {
2593 g_string_append(result
, ", ");
2595 g_string_append(result
, p
->name
);
2599 return g_string_free(result
, FALSE
);
2603 /* @tran is allowed to be NULL. In this case no rollback is possible */
2604 static int bdrv_refresh_perms(BlockDriverState
*bs
, Transaction
*tran
,
2608 Transaction
*local_tran
= NULL
;
2609 g_autoptr(GSList
) list
= bdrv_topological_dfs(NULL
, NULL
, bs
);
2610 GLOBAL_STATE_CODE();
2613 tran
= local_tran
= tran_new();
2616 ret
= bdrv_do_refresh_perms(list
, NULL
, tran
, errp
);
2619 tran_finalize(local_tran
, ret
);
2625 int bdrv_child_try_set_perm(BdrvChild
*c
, uint64_t perm
, uint64_t shared
,
2628 Error
*local_err
= NULL
;
2629 Transaction
*tran
= tran_new();
2632 GLOBAL_STATE_CODE();
2634 bdrv_child_set_perm(c
, perm
, shared
, tran
);
2636 ret
= bdrv_refresh_perms(c
->bs
, tran
, &local_err
);
2638 tran_finalize(tran
, ret
);
2641 if ((perm
& ~c
->perm
) || (c
->shared_perm
& ~shared
)) {
2642 /* tighten permissions */
2643 error_propagate(errp
, local_err
);
2646 * Our caller may intend to only loosen restrictions and
2647 * does not expect this function to fail. Errors are not
2648 * fatal in such a case, so we can just hide them from our
2651 error_free(local_err
);
2659 int bdrv_child_refresh_perms(BlockDriverState
*bs
, BdrvChild
*c
, Error
**errp
)
2661 uint64_t parent_perms
, parent_shared
;
2662 uint64_t perms
, shared
;
2664 GLOBAL_STATE_CODE();
2666 bdrv_get_cumulative_perm(bs
, &parent_perms
, &parent_shared
);
2667 bdrv_child_perm(bs
, c
->bs
, c
, c
->role
, NULL
,
2668 parent_perms
, parent_shared
, &perms
, &shared
);
2670 return bdrv_child_try_set_perm(c
, perms
, shared
, errp
);
2674 * Default implementation for .bdrv_child_perm() for block filters:
2675 * Forward CONSISTENT_READ, WRITE, WRITE_UNCHANGED, and RESIZE to the
2678 static void bdrv_filter_default_perms(BlockDriverState
*bs
, BdrvChild
*c
,
2680 BlockReopenQueue
*reopen_queue
,
2681 uint64_t perm
, uint64_t shared
,
2682 uint64_t *nperm
, uint64_t *nshared
)
2684 GLOBAL_STATE_CODE();
2685 *nperm
= perm
& DEFAULT_PERM_PASSTHROUGH
;
2686 *nshared
= (shared
& DEFAULT_PERM_PASSTHROUGH
) | DEFAULT_PERM_UNCHANGED
;
2689 static void bdrv_default_perms_for_cow(BlockDriverState
*bs
, BdrvChild
*c
,
2691 BlockReopenQueue
*reopen_queue
,
2692 uint64_t perm
, uint64_t shared
,
2693 uint64_t *nperm
, uint64_t *nshared
)
2695 assert(role
& BDRV_CHILD_COW
);
2696 GLOBAL_STATE_CODE();
2699 * We want consistent read from backing files if the parent needs it.
2700 * No other operations are performed on backing files.
2702 perm
&= BLK_PERM_CONSISTENT_READ
;
2705 * If the parent can deal with changing data, we're okay with a
2706 * writable and resizable backing file.
2707 * TODO Require !(perm & BLK_PERM_CONSISTENT_READ), too?
2709 if (shared
& BLK_PERM_WRITE
) {
2710 shared
= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2715 shared
|= BLK_PERM_CONSISTENT_READ
| BLK_PERM_WRITE_UNCHANGED
;
2717 if (bs
->open_flags
& BDRV_O_INACTIVE
) {
2718 shared
|= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2725 static void bdrv_default_perms_for_storage(BlockDriverState
*bs
, BdrvChild
*c
,
2727 BlockReopenQueue
*reopen_queue
,
2728 uint64_t perm
, uint64_t shared
,
2729 uint64_t *nperm
, uint64_t *nshared
)
2733 GLOBAL_STATE_CODE();
2734 assert(role
& (BDRV_CHILD_METADATA
| BDRV_CHILD_DATA
));
2736 flags
= bdrv_reopen_get_flags(reopen_queue
, bs
);
2739 * Apart from the modifications below, the same permissions are
2740 * forwarded and left alone as for filters
2742 bdrv_filter_default_perms(bs
, c
, role
, reopen_queue
,
2743 perm
, shared
, &perm
, &shared
);
2745 if (role
& BDRV_CHILD_METADATA
) {
2746 /* Format drivers may touch metadata even if the guest doesn't write */
2747 if (bdrv_is_writable_after_reopen(bs
, reopen_queue
)) {
2748 perm
|= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2752 * bs->file always needs to be consistent because of the
2753 * metadata. We can never allow other users to resize or write
2756 if (!(flags
& BDRV_O_NO_IO
)) {
2757 perm
|= BLK_PERM_CONSISTENT_READ
;
2759 shared
&= ~(BLK_PERM_WRITE
| BLK_PERM_RESIZE
);
2762 if (role
& BDRV_CHILD_DATA
) {
2764 * Technically, everything in this block is a subset of the
2765 * BDRV_CHILD_METADATA path taken above, and so this could
2766 * be an "else if" branch. However, that is not obvious, and
2767 * this function is not performance critical, therefore we let
2768 * this be an independent "if".
2772 * We cannot allow other users to resize the file because the
2773 * format driver might have some assumptions about the size
2774 * (e.g. because it is stored in metadata, or because the file
2775 * is split into fixed-size data files).
2777 shared
&= ~BLK_PERM_RESIZE
;
2780 * WRITE_UNCHANGED often cannot be performed as such on the
2781 * data file. For example, the qcow2 driver may still need to
2782 * write copied clusters on copy-on-read.
2784 if (perm
& BLK_PERM_WRITE_UNCHANGED
) {
2785 perm
|= BLK_PERM_WRITE
;
2789 * If the data file is written to, the format driver may
2790 * expect to be able to resize it by writing beyond the EOF.
2792 if (perm
& BLK_PERM_WRITE
) {
2793 perm
|= BLK_PERM_RESIZE
;
2797 if (bs
->open_flags
& BDRV_O_INACTIVE
) {
2798 shared
|= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2805 void bdrv_default_perms(BlockDriverState
*bs
, BdrvChild
*c
,
2806 BdrvChildRole role
, BlockReopenQueue
*reopen_queue
,
2807 uint64_t perm
, uint64_t shared
,
2808 uint64_t *nperm
, uint64_t *nshared
)
2810 GLOBAL_STATE_CODE();
2811 if (role
& BDRV_CHILD_FILTERED
) {
2812 assert(!(role
& (BDRV_CHILD_DATA
| BDRV_CHILD_METADATA
|
2814 bdrv_filter_default_perms(bs
, c
, role
, reopen_queue
,
2815 perm
, shared
, nperm
, nshared
);
2816 } else if (role
& BDRV_CHILD_COW
) {
2817 assert(!(role
& (BDRV_CHILD_DATA
| BDRV_CHILD_METADATA
)));
2818 bdrv_default_perms_for_cow(bs
, c
, role
, reopen_queue
,
2819 perm
, shared
, nperm
, nshared
);
2820 } else if (role
& (BDRV_CHILD_METADATA
| BDRV_CHILD_DATA
)) {
2821 bdrv_default_perms_for_storage(bs
, c
, role
, reopen_queue
,
2822 perm
, shared
, nperm
, nshared
);
2824 g_assert_not_reached();
2828 uint64_t bdrv_qapi_perm_to_blk_perm(BlockPermission qapi_perm
)
2830 static const uint64_t permissions
[] = {
2831 [BLOCK_PERMISSION_CONSISTENT_READ
] = BLK_PERM_CONSISTENT_READ
,
2832 [BLOCK_PERMISSION_WRITE
] = BLK_PERM_WRITE
,
2833 [BLOCK_PERMISSION_WRITE_UNCHANGED
] = BLK_PERM_WRITE_UNCHANGED
,
2834 [BLOCK_PERMISSION_RESIZE
] = BLK_PERM_RESIZE
,
2837 QEMU_BUILD_BUG_ON(ARRAY_SIZE(permissions
) != BLOCK_PERMISSION__MAX
);
2838 QEMU_BUILD_BUG_ON(1UL << ARRAY_SIZE(permissions
) != BLK_PERM_ALL
+ 1);
2840 assert(qapi_perm
< BLOCK_PERMISSION__MAX
);
2842 return permissions
[qapi_perm
];
2846 * Replaces the node that a BdrvChild points to without updating permissions.
2848 * If @new_bs is non-NULL, the parent of @child must already be drained through
2851 static void bdrv_replace_child_noperm(BdrvChild
*child
,
2852 BlockDriverState
*new_bs
)
2854 BlockDriverState
*old_bs
= child
->bs
;
2855 int new_bs_quiesce_counter
;
2857 assert(!child
->frozen
);
2860 * If we want to change the BdrvChild to point to a drained node as its new
2861 * child->bs, we need to make sure that its new parent is drained, too. In
2862 * other words, either child->quiesce_parent must already be true or we must
2863 * be able to set it and keep the parent's quiesce_counter consistent with
2864 * that, but without polling or starting new requests (this function
2865 * guarantees that it doesn't poll, and starting new requests would be
2866 * against the invariants of drain sections).
2868 * To keep things simple, we pick the first option (child->quiesce_parent
2869 * must already be true). We also generalise the rule a bit to make it
2870 * easier to verify in callers and more likely to be covered in test cases:
2871 * The parent must be quiesced through this child even if new_bs isn't
2872 * currently drained.
2874 * The only exception is for callers that always pass new_bs == NULL. In
2875 * this case, we obviously never need to consider the case of a drained
2876 * new_bs, so we can keep the callers simpler by allowing them not to drain
2879 assert(!new_bs
|| child
->quiesced_parent
);
2880 assert(old_bs
!= new_bs
);
2881 GLOBAL_STATE_CODE();
2883 if (old_bs
&& new_bs
) {
2884 assert(bdrv_get_aio_context(old_bs
) == bdrv_get_aio_context(new_bs
));
2887 /* TODO Pull this up into the callers to avoid polling here */
2888 bdrv_graph_wrlock();
2890 if (child
->klass
->detach
) {
2891 child
->klass
->detach(child
);
2893 QLIST_REMOVE(child
, next_parent
);
2899 QLIST_INSERT_HEAD(&new_bs
->parents
, child
, next_parent
);
2900 if (child
->klass
->attach
) {
2901 child
->klass
->attach(child
);
2904 bdrv_graph_wrunlock();
2907 * If the parent was drained through this BdrvChild previously, but new_bs
2908 * is not drained, allow requests to come in only after the new node has
2911 new_bs_quiesce_counter
= (new_bs
? new_bs
->quiesce_counter
: 0);
2912 if (!new_bs_quiesce_counter
&& child
->quiesced_parent
) {
2913 bdrv_parent_drained_end_single(child
);
2918 * Free the given @child.
2920 * The child must be empty (i.e. `child->bs == NULL`) and it must be
2921 * unused (i.e. not in a children list).
2923 static void bdrv_child_free(BdrvChild
*child
)
2926 GLOBAL_STATE_CODE();
2927 assert(!child
->next
.le_prev
); /* not in children list */
2929 g_free(child
->name
);
2933 typedef struct BdrvAttachChildCommonState
{
2935 AioContext
*old_parent_ctx
;
2936 AioContext
*old_child_ctx
;
2937 } BdrvAttachChildCommonState
;
2939 static void bdrv_attach_child_common_abort(void *opaque
)
2941 BdrvAttachChildCommonState
*s
= opaque
;
2942 BlockDriverState
*bs
= s
->child
->bs
;
2944 GLOBAL_STATE_CODE();
2945 bdrv_replace_child_noperm(s
->child
, NULL
);
2947 if (bdrv_get_aio_context(bs
) != s
->old_child_ctx
) {
2948 bdrv_try_change_aio_context(bs
, s
->old_child_ctx
, NULL
, &error_abort
);
2951 if (bdrv_child_get_parent_aio_context(s
->child
) != s
->old_parent_ctx
) {
2953 GHashTable
*visited
;
2958 /* No need to visit `child`, because it has been detached already */
2959 visited
= g_hash_table_new(NULL
, NULL
);
2960 ret
= s
->child
->klass
->change_aio_ctx(s
->child
, s
->old_parent_ctx
,
2961 visited
, tran
, &error_abort
);
2962 g_hash_table_destroy(visited
);
2964 /* transaction is supposed to always succeed */
2965 assert(ret
== true);
2970 bdrv_child_free(s
->child
);
2973 static TransactionActionDrv bdrv_attach_child_common_drv
= {
2974 .abort
= bdrv_attach_child_common_abort
,
2979 * Common part of attaching bdrv child to bs or to blk or to job
2981 * Function doesn't update permissions, caller is responsible for this.
2983 * Returns new created child.
2985 static BdrvChild
*bdrv_attach_child_common(BlockDriverState
*child_bs
,
2986 const char *child_name
,
2987 const BdrvChildClass
*child_class
,
2988 BdrvChildRole child_role
,
2989 uint64_t perm
, uint64_t shared_perm
,
2991 Transaction
*tran
, Error
**errp
)
2993 BdrvChild
*new_child
;
2994 AioContext
*parent_ctx
;
2995 AioContext
*child_ctx
= bdrv_get_aio_context(child_bs
);
2997 assert(child_class
->get_parent_desc
);
2998 GLOBAL_STATE_CODE();
3000 new_child
= g_new(BdrvChild
, 1);
3001 *new_child
= (BdrvChild
) {
3003 .name
= g_strdup(child_name
),
3004 .klass
= child_class
,
3007 .shared_perm
= shared_perm
,
3012 * If the AioContexts don't match, first try to move the subtree of
3013 * child_bs into the AioContext of the new parent. If this doesn't work,
3014 * try moving the parent into the AioContext of child_bs instead.
3016 parent_ctx
= bdrv_child_get_parent_aio_context(new_child
);
3017 if (child_ctx
!= parent_ctx
) {
3018 Error
*local_err
= NULL
;
3019 int ret
= bdrv_try_change_aio_context(child_bs
, parent_ctx
, NULL
,
3022 if (ret
< 0 && child_class
->change_aio_ctx
) {
3023 Transaction
*tran
= tran_new();
3024 GHashTable
*visited
= g_hash_table_new(NULL
, NULL
);
3027 g_hash_table_add(visited
, new_child
);
3028 ret_child
= child_class
->change_aio_ctx(new_child
, child_ctx
,
3029 visited
, tran
, NULL
);
3030 if (ret_child
== true) {
3031 error_free(local_err
);
3034 tran_finalize(tran
, ret_child
== true ? 0 : -1);
3035 g_hash_table_destroy(visited
);
3039 error_propagate(errp
, local_err
);
3040 bdrv_child_free(new_child
);
3047 * Let every new BdrvChild start with a drained parent. Inserting the child
3048 * in the graph with bdrv_replace_child_noperm() will undrain it if
3049 * @child_bs is not drained.
3051 * The child was only just created and is not yet visible in global state
3052 * until bdrv_replace_child_noperm() inserts it into the graph, so nobody
3053 * could have sent requests and polling is not necessary.
3055 * Note that this means that the parent isn't fully drained yet, we only
3056 * stop new requests from coming in. This is fine, we don't care about the
3057 * old requests here, they are not for this child. If another place enters a
3058 * drain section for the same parent, but wants it to be fully quiesced, it
3059 * will not run most of the the code in .drained_begin() again (which is not
3060 * a problem, we already did this), but it will still poll until the parent
3061 * is fully quiesced, so it will not be negatively affected either.
3063 bdrv_parent_drained_begin_single(new_child
);
3064 bdrv_replace_child_noperm(new_child
, child_bs
);
3066 BdrvAttachChildCommonState
*s
= g_new(BdrvAttachChildCommonState
, 1);
3067 *s
= (BdrvAttachChildCommonState
) {
3069 .old_parent_ctx
= parent_ctx
,
3070 .old_child_ctx
= child_ctx
,
3072 tran_add(tran
, &bdrv_attach_child_common_drv
, s
);
3078 * Function doesn't update permissions, caller is responsible for this.
3080 static BdrvChild
*bdrv_attach_child_noperm(BlockDriverState
*parent_bs
,
3081 BlockDriverState
*child_bs
,
3082 const char *child_name
,
3083 const BdrvChildClass
*child_class
,
3084 BdrvChildRole child_role
,
3088 uint64_t perm
, shared_perm
;
3090 assert(parent_bs
->drv
);
3091 GLOBAL_STATE_CODE();
3093 if (bdrv_recurse_has_child(child_bs
, parent_bs
)) {
3094 error_setg(errp
, "Making '%s' a %s child of '%s' would create a cycle",
3095 child_bs
->node_name
, child_name
, parent_bs
->node_name
);
3099 bdrv_get_cumulative_perm(parent_bs
, &perm
, &shared_perm
);
3100 bdrv_child_perm(parent_bs
, child_bs
, NULL
, child_role
, NULL
,
3101 perm
, shared_perm
, &perm
, &shared_perm
);
3103 return bdrv_attach_child_common(child_bs
, child_name
, child_class
,
3104 child_role
, perm
, shared_perm
, parent_bs
,
3109 * This function steals the reference to child_bs from the caller.
3110 * That reference is later dropped by bdrv_root_unref_child().
3112 * On failure NULL is returned, errp is set and the reference to
3113 * child_bs is also dropped.
3115 * The caller must hold the AioContext lock @child_bs, but not that of @ctx
3116 * (unless @child_bs is already in @ctx).
3118 BdrvChild
*bdrv_root_attach_child(BlockDriverState
*child_bs
,
3119 const char *child_name
,
3120 const BdrvChildClass
*child_class
,
3121 BdrvChildRole child_role
,
3122 uint64_t perm
, uint64_t shared_perm
,
3123 void *opaque
, Error
**errp
)
3127 Transaction
*tran
= tran_new();
3129 GLOBAL_STATE_CODE();
3131 child
= bdrv_attach_child_common(child_bs
, child_name
, child_class
,
3132 child_role
, perm
, shared_perm
, opaque
,
3139 ret
= bdrv_refresh_perms(child_bs
, tran
, errp
);
3142 tran_finalize(tran
, ret
);
3144 bdrv_unref(child_bs
);
3146 return ret
< 0 ? NULL
: child
;
3150 * This function transfers the reference to child_bs from the caller
3151 * to parent_bs. That reference is later dropped by parent_bs on
3152 * bdrv_close() or if someone calls bdrv_unref_child().
3154 * On failure NULL is returned, errp is set and the reference to
3155 * child_bs is also dropped.
3157 * If @parent_bs and @child_bs are in different AioContexts, the caller must
3158 * hold the AioContext lock for @child_bs, but not for @parent_bs.
3160 BdrvChild
*bdrv_attach_child(BlockDriverState
*parent_bs
,
3161 BlockDriverState
*child_bs
,
3162 const char *child_name
,
3163 const BdrvChildClass
*child_class
,
3164 BdrvChildRole child_role
,
3169 Transaction
*tran
= tran_new();
3171 GLOBAL_STATE_CODE();
3173 child
= bdrv_attach_child_noperm(parent_bs
, child_bs
, child_name
,
3174 child_class
, child_role
, tran
, errp
);
3180 ret
= bdrv_refresh_perms(parent_bs
, tran
, errp
);
3186 tran_finalize(tran
, ret
);
3188 bdrv_unref(child_bs
);
3190 return ret
< 0 ? NULL
: child
;
3193 /* Callers must ensure that child->frozen is false. */
3194 void bdrv_root_unref_child(BdrvChild
*child
)
3196 BlockDriverState
*child_bs
= child
->bs
;
3198 GLOBAL_STATE_CODE();
3199 bdrv_replace_child_noperm(child
, NULL
);
3200 bdrv_child_free(child
);
3204 * Update permissions for old node. We're just taking a parent away, so
3205 * we're loosening restrictions. Errors of permission update are not
3206 * fatal in this case, ignore them.
3208 bdrv_refresh_perms(child_bs
, NULL
, NULL
);
3211 * When the parent requiring a non-default AioContext is removed, the
3212 * node moves back to the main AioContext
3214 bdrv_try_change_aio_context(child_bs
, qemu_get_aio_context(), NULL
,
3218 bdrv_unref(child_bs
);
3221 typedef struct BdrvSetInheritsFrom
{
3222 BlockDriverState
*bs
;
3223 BlockDriverState
*old_inherits_from
;
3224 } BdrvSetInheritsFrom
;
3226 static void bdrv_set_inherits_from_abort(void *opaque
)
3228 BdrvSetInheritsFrom
*s
= opaque
;
3230 s
->bs
->inherits_from
= s
->old_inherits_from
;
3233 static TransactionActionDrv bdrv_set_inherits_from_drv
= {
3234 .abort
= bdrv_set_inherits_from_abort
,
3238 /* @tran is allowed to be NULL. In this case no rollback is possible */
3239 static void bdrv_set_inherits_from(BlockDriverState
*bs
,
3240 BlockDriverState
*new_inherits_from
,
3244 BdrvSetInheritsFrom
*s
= g_new(BdrvSetInheritsFrom
, 1);
3246 *s
= (BdrvSetInheritsFrom
) {
3248 .old_inherits_from
= bs
->inherits_from
,
3251 tran_add(tran
, &bdrv_set_inherits_from_drv
, s
);
3254 bs
->inherits_from
= new_inherits_from
;
3258 * Clear all inherits_from pointers from children and grandchildren of
3259 * @root that point to @root, where necessary.
3260 * @tran is allowed to be NULL. In this case no rollback is possible
3262 static void bdrv_unset_inherits_from(BlockDriverState
*root
, BdrvChild
*child
,
3267 if (child
->bs
->inherits_from
== root
) {
3269 * Remove inherits_from only when the last reference between root and
3270 * child->bs goes away.
3272 QLIST_FOREACH(c
, &root
->children
, next
) {
3273 if (c
!= child
&& c
->bs
== child
->bs
) {
3278 bdrv_set_inherits_from(child
->bs
, NULL
, tran
);
3282 QLIST_FOREACH(c
, &child
->bs
->children
, next
) {
3283 bdrv_unset_inherits_from(root
, c
, tran
);
3287 /* Callers must ensure that child->frozen is false. */
3288 void bdrv_unref_child(BlockDriverState
*parent
, BdrvChild
*child
)
3290 GLOBAL_STATE_CODE();
3291 if (child
== NULL
) {
3295 bdrv_unset_inherits_from(parent
, child
, NULL
);
3296 bdrv_root_unref_child(child
);
3300 static void bdrv_parent_cb_change_media(BlockDriverState
*bs
, bool load
)
3303 GLOBAL_STATE_CODE();
3304 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
3305 if (c
->klass
->change_media
) {
3306 c
->klass
->change_media(c
, load
);
3311 /* Return true if you can reach parent going through child->inherits_from
3312 * recursively. If parent or child are NULL, return false */
3313 static bool bdrv_inherits_from_recursive(BlockDriverState
*child
,
3314 BlockDriverState
*parent
)
3316 while (child
&& child
!= parent
) {
3317 child
= child
->inherits_from
;
3320 return child
!= NULL
;
3324 * Return the BdrvChildRole for @bs's backing child. bs->backing is
3325 * mostly used for COW backing children (role = COW), but also for
3326 * filtered children (role = FILTERED | PRIMARY).
3328 static BdrvChildRole
bdrv_backing_role(BlockDriverState
*bs
)
3330 if (bs
->drv
&& bs
->drv
->is_filter
) {
3331 return BDRV_CHILD_FILTERED
| BDRV_CHILD_PRIMARY
;
3333 return BDRV_CHILD_COW
;
3338 * Sets the bs->backing or bs->file link of a BDS. A new reference is created;
3339 * callers which don't need their own reference any more must call bdrv_unref().
3341 * Function doesn't update permissions, caller is responsible for this.
3343 static int bdrv_set_file_or_backing_noperm(BlockDriverState
*parent_bs
,
3344 BlockDriverState
*child_bs
,
3346 Transaction
*tran
, Error
**errp
)
3348 bool update_inherits_from
=
3349 bdrv_inherits_from_recursive(child_bs
, parent_bs
);
3350 BdrvChild
*child
= is_backing
? parent_bs
->backing
: parent_bs
->file
;
3353 GLOBAL_STATE_CODE();
3355 if (!parent_bs
->drv
) {
3357 * Node without drv is an object without a class :/. TODO: finally fix
3358 * qcow2 driver to never clear bs->drv and implement format corruption
3359 * handling in other way.
3361 error_setg(errp
, "Node corrupted");
3365 if (child
&& child
->frozen
) {
3366 error_setg(errp
, "Cannot change frozen '%s' link from '%s' to '%s'",
3367 child
->name
, parent_bs
->node_name
, child
->bs
->node_name
);
3371 if (is_backing
&& !parent_bs
->drv
->is_filter
&&
3372 !parent_bs
->drv
->supports_backing
)
3374 error_setg(errp
, "Driver '%s' of node '%s' does not support backing "
3375 "files", parent_bs
->drv
->format_name
, parent_bs
->node_name
);
3379 if (parent_bs
->drv
->is_filter
) {
3380 role
= BDRV_CHILD_FILTERED
| BDRV_CHILD_PRIMARY
;
3381 } else if (is_backing
) {
3382 role
= BDRV_CHILD_COW
;
3385 * We only can use same role as it is in existing child. We don't have
3386 * infrastructure to determine role of file child in generic way
3389 error_setg(errp
, "Cannot set file child to format node without "
3397 bdrv_unset_inherits_from(parent_bs
, child
, tran
);
3398 bdrv_remove_child(child
, tran
);
3405 child
= bdrv_attach_child_noperm(parent_bs
, child_bs
,
3406 is_backing
? "backing" : "file",
3407 &child_of_bds
, role
,
3415 * If inherits_from pointed recursively to bs then let's update it to
3416 * point directly to bs (else it will become NULL).
3418 if (update_inherits_from
) {
3419 bdrv_set_inherits_from(child_bs
, parent_bs
, tran
);
3423 bdrv_refresh_limits(parent_bs
, tran
, NULL
);
3428 static int bdrv_set_backing_noperm(BlockDriverState
*bs
,
3429 BlockDriverState
*backing_hd
,
3430 Transaction
*tran
, Error
**errp
)
3432 GLOBAL_STATE_CODE();
3433 return bdrv_set_file_or_backing_noperm(bs
, backing_hd
, true, tran
, errp
);
3436 int bdrv_set_backing_hd_drained(BlockDriverState
*bs
,
3437 BlockDriverState
*backing_hd
,
3441 Transaction
*tran
= tran_new();
3443 GLOBAL_STATE_CODE();
3444 assert(bs
->quiesce_counter
> 0);
3446 ret
= bdrv_set_backing_noperm(bs
, backing_hd
, tran
, errp
);
3451 ret
= bdrv_refresh_perms(bs
, tran
, errp
);
3453 tran_finalize(tran
, ret
);
3457 int bdrv_set_backing_hd(BlockDriverState
*bs
, BlockDriverState
*backing_hd
,
3461 GLOBAL_STATE_CODE();
3463 bdrv_drained_begin(bs
);
3464 ret
= bdrv_set_backing_hd_drained(bs
, backing_hd
, errp
);
3465 bdrv_drained_end(bs
);
3471 * Opens the backing file for a BlockDriverState if not yet open
3473 * bdref_key specifies the key for the image's BlockdevRef in the options QDict.
3474 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
3475 * itself, all options starting with "${bdref_key}." are considered part of the
3478 * TODO Can this be unified with bdrv_open_image()?
3480 int bdrv_open_backing_file(BlockDriverState
*bs
, QDict
*parent_options
,
3481 const char *bdref_key
, Error
**errp
)
3483 char *backing_filename
= NULL
;
3484 char *bdref_key_dot
;
3485 const char *reference
= NULL
;
3487 bool implicit_backing
= false;
3488 BlockDriverState
*backing_hd
;
3490 QDict
*tmp_parent_options
= NULL
;
3491 Error
*local_err
= NULL
;
3493 GLOBAL_STATE_CODE();
3495 if (bs
->backing
!= NULL
) {
3499 /* NULL means an empty set of options */
3500 if (parent_options
== NULL
) {
3501 tmp_parent_options
= qdict_new();
3502 parent_options
= tmp_parent_options
;
3505 bs
->open_flags
&= ~BDRV_O_NO_BACKING
;
3507 bdref_key_dot
= g_strdup_printf("%s.", bdref_key
);
3508 qdict_extract_subqdict(parent_options
, &options
, bdref_key_dot
);
3509 g_free(bdref_key_dot
);
3512 * Caution: while qdict_get_try_str() is fine, getting non-string
3513 * types would require more care. When @parent_options come from
3514 * -blockdev or blockdev_add, its members are typed according to
3515 * the QAPI schema, but when they come from -drive, they're all
3518 reference
= qdict_get_try_str(parent_options
, bdref_key
);
3519 if (reference
|| qdict_haskey(options
, "file.filename")) {
3520 /* keep backing_filename NULL */
3521 } else if (bs
->backing_file
[0] == '\0' && qdict_size(options
) == 0) {
3522 qobject_unref(options
);
3525 if (qdict_size(options
) == 0) {
3526 /* If the user specifies options that do not modify the
3527 * backing file's behavior, we might still consider it the
3528 * implicit backing file. But it's easier this way, and
3529 * just specifying some of the backing BDS's options is
3530 * only possible with -drive anyway (otherwise the QAPI
3531 * schema forces the user to specify everything). */
3532 implicit_backing
= !strcmp(bs
->auto_backing_file
, bs
->backing_file
);
3535 backing_filename
= bdrv_get_full_backing_filename(bs
, &local_err
);
3538 error_propagate(errp
, local_err
);
3539 qobject_unref(options
);
3544 if (!bs
->drv
|| !bs
->drv
->supports_backing
) {
3546 error_setg(errp
, "Driver doesn't support backing files");
3547 qobject_unref(options
);
3552 bs
->backing_format
[0] != '\0' && !qdict_haskey(options
, "driver")) {
3553 qdict_put_str(options
, "driver", bs
->backing_format
);
3556 backing_hd
= bdrv_open_inherit(backing_filename
, reference
, options
, 0, bs
,
3557 &child_of_bds
, bdrv_backing_role(bs
), errp
);
3559 bs
->open_flags
|= BDRV_O_NO_BACKING
;
3560 error_prepend(errp
, "Could not open backing file: ");
3565 if (implicit_backing
) {
3566 bdrv_refresh_filename(backing_hd
);
3567 pstrcpy(bs
->auto_backing_file
, sizeof(bs
->auto_backing_file
),
3568 backing_hd
->filename
);
3571 /* Hook up the backing file link; drop our reference, bs owns the
3572 * backing_hd reference now */
3573 ret
= bdrv_set_backing_hd(bs
, backing_hd
, errp
);
3574 bdrv_unref(backing_hd
);
3579 qdict_del(parent_options
, bdref_key
);
3582 g_free(backing_filename
);
3583 qobject_unref(tmp_parent_options
);
3587 static BlockDriverState
*
3588 bdrv_open_child_bs(const char *filename
, QDict
*options
, const char *bdref_key
,
3589 BlockDriverState
*parent
, const BdrvChildClass
*child_class
,
3590 BdrvChildRole child_role
, bool allow_none
, Error
**errp
)
3592 BlockDriverState
*bs
= NULL
;
3593 QDict
*image_options
;
3594 char *bdref_key_dot
;
3595 const char *reference
;
3597 assert(child_class
!= NULL
);
3599 bdref_key_dot
= g_strdup_printf("%s.", bdref_key
);
3600 qdict_extract_subqdict(options
, &image_options
, bdref_key_dot
);
3601 g_free(bdref_key_dot
);
3604 * Caution: while qdict_get_try_str() is fine, getting non-string
3605 * types would require more care. When @options come from
3606 * -blockdev or blockdev_add, its members are typed according to
3607 * the QAPI schema, but when they come from -drive, they're all
3610 reference
= qdict_get_try_str(options
, bdref_key
);
3611 if (!filename
&& !reference
&& !qdict_size(image_options
)) {
3613 error_setg(errp
, "A block device must be specified for \"%s\"",
3616 qobject_unref(image_options
);
3620 bs
= bdrv_open_inherit(filename
, reference
, image_options
, 0,
3621 parent
, child_class
, child_role
, errp
);
3627 qdict_del(options
, bdref_key
);
3632 * Opens a disk image whose options are given as BlockdevRef in another block
3635 * If allow_none is true, no image will be opened if filename is false and no
3636 * BlockdevRef is given. NULL will be returned, but errp remains unset.
3638 * bdrev_key specifies the key for the image's BlockdevRef in the options QDict.
3639 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
3640 * itself, all options starting with "${bdref_key}." are considered part of the
3643 * The BlockdevRef will be removed from the options QDict.
3645 BdrvChild
*bdrv_open_child(const char *filename
,
3646 QDict
*options
, const char *bdref_key
,
3647 BlockDriverState
*parent
,
3648 const BdrvChildClass
*child_class
,
3649 BdrvChildRole child_role
,
3650 bool allow_none
, Error
**errp
)
3652 BlockDriverState
*bs
;
3654 GLOBAL_STATE_CODE();
3656 bs
= bdrv_open_child_bs(filename
, options
, bdref_key
, parent
, child_class
,
3657 child_role
, allow_none
, errp
);
3662 return bdrv_attach_child(parent
, bs
, bdref_key
, child_class
, child_role
,
3667 * Wrapper on bdrv_open_child() for most popular case: open primary child of bs.
3669 int bdrv_open_file_child(const char *filename
,
3670 QDict
*options
, const char *bdref_key
,
3671 BlockDriverState
*parent
, Error
**errp
)
3675 /* commit_top and mirror_top don't use this function */
3676 assert(!parent
->drv
->filtered_child_is_backing
);
3677 role
= parent
->drv
->is_filter
?
3678 (BDRV_CHILD_FILTERED
| BDRV_CHILD_PRIMARY
) : BDRV_CHILD_IMAGE
;
3680 if (!bdrv_open_child(filename
, options
, bdref_key
, parent
,
3681 &child_of_bds
, role
, false, errp
))
3690 * TODO Future callers may need to specify parent/child_class in order for
3691 * option inheritance to work. Existing callers use it for the root node.
3693 BlockDriverState
*bdrv_open_blockdev_ref(BlockdevRef
*ref
, Error
**errp
)
3695 BlockDriverState
*bs
= NULL
;
3696 QObject
*obj
= NULL
;
3697 QDict
*qdict
= NULL
;
3698 const char *reference
= NULL
;
3701 GLOBAL_STATE_CODE();
3703 if (ref
->type
== QTYPE_QSTRING
) {
3704 reference
= ref
->u
.reference
;
3706 BlockdevOptions
*options
= &ref
->u
.definition
;
3707 assert(ref
->type
== QTYPE_QDICT
);
3709 v
= qobject_output_visitor_new(&obj
);
3710 visit_type_BlockdevOptions(v
, NULL
, &options
, &error_abort
);
3711 visit_complete(v
, &obj
);
3713 qdict
= qobject_to(QDict
, obj
);
3714 qdict_flatten(qdict
);
3716 /* bdrv_open_inherit() defaults to the values in bdrv_flags (for
3717 * compatibility with other callers) rather than what we want as the
3718 * real defaults. Apply the defaults here instead. */
3719 qdict_set_default_str(qdict
, BDRV_OPT_CACHE_DIRECT
, "off");
3720 qdict_set_default_str(qdict
, BDRV_OPT_CACHE_NO_FLUSH
, "off");
3721 qdict_set_default_str(qdict
, BDRV_OPT_READ_ONLY
, "off");
3722 qdict_set_default_str(qdict
, BDRV_OPT_AUTO_READ_ONLY
, "off");
3726 bs
= bdrv_open_inherit(NULL
, reference
, qdict
, 0, NULL
, NULL
, 0, errp
);
3733 static BlockDriverState
*bdrv_append_temp_snapshot(BlockDriverState
*bs
,
3735 QDict
*snapshot_options
,
3738 g_autofree
char *tmp_filename
= NULL
;
3740 QemuOpts
*opts
= NULL
;
3741 BlockDriverState
*bs_snapshot
= NULL
;
3744 GLOBAL_STATE_CODE();
3746 /* if snapshot, we create a temporary backing file and open it
3747 instead of opening 'filename' directly */
3749 /* Get the required size from the image */
3750 total_size
= bdrv_getlength(bs
);
3751 if (total_size
< 0) {
3752 error_setg_errno(errp
, -total_size
, "Could not get image size");
3756 /* Create the temporary image */
3757 tmp_filename
= create_tmp_file(errp
);
3758 if (!tmp_filename
) {
3762 opts
= qemu_opts_create(bdrv_qcow2
.create_opts
, NULL
, 0,
3764 qemu_opt_set_number(opts
, BLOCK_OPT_SIZE
, total_size
, &error_abort
);
3765 ret
= bdrv_create(&bdrv_qcow2
, tmp_filename
, opts
, errp
);
3766 qemu_opts_del(opts
);
3768 error_prepend(errp
, "Could not create temporary overlay '%s': ",
3773 /* Prepare options QDict for the temporary file */
3774 qdict_put_str(snapshot_options
, "file.driver", "file");
3775 qdict_put_str(snapshot_options
, "file.filename", tmp_filename
);
3776 qdict_put_str(snapshot_options
, "driver", "qcow2");
3778 bs_snapshot
= bdrv_open(NULL
, NULL
, snapshot_options
, flags
, errp
);
3779 snapshot_options
= NULL
;
3784 ret
= bdrv_append(bs_snapshot
, bs
, errp
);
3791 qobject_unref(snapshot_options
);
3796 * Opens a disk image (raw, qcow2, vmdk, ...)
3798 * options is a QDict of options to pass to the block drivers, or NULL for an
3799 * empty set of options. The reference to the QDict belongs to the block layer
3800 * after the call (even on failure), so if the caller intends to reuse the
3801 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
3803 * If *pbs is NULL, a new BDS will be created with a pointer to it stored there.
3804 * If it is not NULL, the referenced BDS will be reused.
3806 * The reference parameter may be used to specify an existing block device which
3807 * should be opened. If specified, neither options nor a filename may be given,
3808 * nor can an existing BDS be reused (that is, *pbs has to be NULL).
3810 * The caller must always hold @filename AioContext lock, because this
3811 * function eventually calls bdrv_refresh_total_sectors() which polls
3812 * when called from non-coroutine context.
3814 static BlockDriverState
* no_coroutine_fn
3815 bdrv_open_inherit(const char *filename
, const char *reference
, QDict
*options
,
3816 int flags
, BlockDriverState
*parent
,
3817 const BdrvChildClass
*child_class
, BdrvChildRole child_role
,
3821 BlockBackend
*file
= NULL
;
3822 BlockDriverState
*bs
;
3823 BlockDriver
*drv
= NULL
;
3825 const char *drvname
;
3826 const char *backing
;
3827 Error
*local_err
= NULL
;
3828 QDict
*snapshot_options
= NULL
;
3829 int snapshot_flags
= 0;
3831 assert(!child_class
|| !flags
);
3832 assert(!child_class
== !parent
);
3833 GLOBAL_STATE_CODE();
3834 assert(!qemu_in_coroutine());
3837 bool options_non_empty
= options
? qdict_size(options
) : false;
3838 qobject_unref(options
);
3840 if (filename
|| options_non_empty
) {
3841 error_setg(errp
, "Cannot reference an existing block device with "
3842 "additional options or a new filename");
3846 bs
= bdrv_lookup_bs(reference
, reference
, errp
);
3857 /* NULL means an empty set of options */
3858 if (options
== NULL
) {
3859 options
= qdict_new();
3862 /* json: syntax counts as explicit options, as if in the QDict */
3863 parse_json_protocol(options
, &filename
, &local_err
);
3868 bs
->explicit_options
= qdict_clone_shallow(options
);
3871 bool parent_is_format
;
3874 parent_is_format
= parent
->drv
->is_format
;
3877 * parent->drv is not set yet because this node is opened for
3878 * (potential) format probing. That means that @parent is going
3879 * to be a format node.
3881 parent_is_format
= true;
3884 bs
->inherits_from
= parent
;
3885 child_class
->inherit_options(child_role
, parent_is_format
,
3887 parent
->open_flags
, parent
->options
);
3890 ret
= bdrv_fill_options(&options
, filename
, &flags
, &local_err
);
3896 * Set the BDRV_O_RDWR and BDRV_O_ALLOW_RDWR flags.
3897 * Caution: getting a boolean member of @options requires care.
3898 * When @options come from -blockdev or blockdev_add, members are
3899 * typed according to the QAPI schema, but when they come from
3900 * -drive, they're all QString.
3902 if (g_strcmp0(qdict_get_try_str(options
, BDRV_OPT_READ_ONLY
), "on") &&
3903 !qdict_get_try_bool(options
, BDRV_OPT_READ_ONLY
, false)) {
3904 flags
|= (BDRV_O_RDWR
| BDRV_O_ALLOW_RDWR
);
3906 flags
&= ~BDRV_O_RDWR
;
3909 if (flags
& BDRV_O_SNAPSHOT
) {
3910 snapshot_options
= qdict_new();
3911 bdrv_temp_snapshot_options(&snapshot_flags
, snapshot_options
,
3913 /* Let bdrv_backing_options() override "read-only" */
3914 qdict_del(options
, BDRV_OPT_READ_ONLY
);
3915 bdrv_inherited_options(BDRV_CHILD_COW
, true,
3916 &flags
, options
, flags
, options
);
3919 bs
->open_flags
= flags
;
3920 bs
->options
= options
;
3921 options
= qdict_clone_shallow(options
);
3923 /* Find the right image format driver */
3924 /* See cautionary note on accessing @options above */
3925 drvname
= qdict_get_try_str(options
, "driver");
3927 drv
= bdrv_find_format(drvname
);
3929 error_setg(errp
, "Unknown driver: '%s'", drvname
);
3934 assert(drvname
|| !(flags
& BDRV_O_PROTOCOL
));
3936 /* See cautionary note on accessing @options above */
3937 backing
= qdict_get_try_str(options
, "backing");
3938 if (qobject_to(QNull
, qdict_get(options
, "backing")) != NULL
||
3939 (backing
&& *backing
== '\0'))
3942 warn_report("Use of \"backing\": \"\" is deprecated; "
3943 "use \"backing\": null instead");
3945 flags
|= BDRV_O_NO_BACKING
;
3946 qdict_del(bs
->explicit_options
, "backing");
3947 qdict_del(bs
->options
, "backing");
3948 qdict_del(options
, "backing");
3951 /* Open image file without format layer. This BlockBackend is only used for
3952 * probing, the block drivers will do their own bdrv_open_child() for the
3953 * same BDS, which is why we put the node name back into options. */
3954 if ((flags
& BDRV_O_PROTOCOL
) == 0) {
3955 BlockDriverState
*file_bs
;
3957 file_bs
= bdrv_open_child_bs(filename
, options
, "file", bs
,
3958 &child_of_bds
, BDRV_CHILD_IMAGE
,
3963 if (file_bs
!= NULL
) {
3964 /* Not requesting BLK_PERM_CONSISTENT_READ because we're only
3965 * looking at the header to guess the image format. This works even
3966 * in cases where a guest would not see a consistent state. */
3967 file
= blk_new(bdrv_get_aio_context(file_bs
), 0, BLK_PERM_ALL
);
3968 blk_insert_bs(file
, file_bs
, &local_err
);
3969 bdrv_unref(file_bs
);
3974 qdict_put_str(options
, "file", bdrv_get_node_name(file_bs
));
3978 /* Image format probing */
3981 ret
= find_image_format(file
, filename
, &drv
, &local_err
);
3986 * This option update would logically belong in bdrv_fill_options(),
3987 * but we first need to open bs->file for the probing to work, while
3988 * opening bs->file already requires the (mostly) final set of options
3989 * so that cache mode etc. can be inherited.
3991 * Adding the driver later is somewhat ugly, but it's not an option
3992 * that would ever be inherited, so it's correct. We just need to make
3993 * sure to update both bs->options (which has the full effective
3994 * options for bs) and options (which has file.* already removed).
3996 qdict_put_str(bs
->options
, "driver", drv
->format_name
);
3997 qdict_put_str(options
, "driver", drv
->format_name
);
3999 error_setg(errp
, "Must specify either driver or file");
4003 /* BDRV_O_PROTOCOL must be set iff a protocol BDS is about to be created */
4004 assert(!!(flags
& BDRV_O_PROTOCOL
) == !!drv
->bdrv_file_open
);
4005 /* file must be NULL if a protocol BDS is about to be created
4006 * (the inverse results in an error message from bdrv_open_common()) */
4007 assert(!(flags
& BDRV_O_PROTOCOL
) || !file
);
4009 /* Open the image */
4010 ret
= bdrv_open_common(bs
, file
, options
, &local_err
);
4020 /* If there is a backing file, use it */
4021 if ((flags
& BDRV_O_NO_BACKING
) == 0) {
4022 ret
= bdrv_open_backing_file(bs
, options
, "backing", &local_err
);
4024 goto close_and_fail
;
4028 /* Remove all children options and references
4029 * from bs->options and bs->explicit_options */
4030 QLIST_FOREACH(child
, &bs
->children
, next
) {
4031 char *child_key_dot
;
4032 child_key_dot
= g_strdup_printf("%s.", child
->name
);
4033 qdict_extract_subqdict(bs
->explicit_options
, NULL
, child_key_dot
);
4034 qdict_extract_subqdict(bs
->options
, NULL
, child_key_dot
);
4035 qdict_del(bs
->explicit_options
, child
->name
);
4036 qdict_del(bs
->options
, child
->name
);
4037 g_free(child_key_dot
);
4040 /* Check if any unknown options were used */
4041 if (qdict_size(options
) != 0) {
4042 const QDictEntry
*entry
= qdict_first(options
);
4043 if (flags
& BDRV_O_PROTOCOL
) {
4044 error_setg(errp
, "Block protocol '%s' doesn't support the option "
4045 "'%s'", drv
->format_name
, entry
->key
);
4048 "Block format '%s' does not support the option '%s'",
4049 drv
->format_name
, entry
->key
);
4052 goto close_and_fail
;
4055 bdrv_parent_cb_change_media(bs
, true);
4057 qobject_unref(options
);
4060 /* For snapshot=on, create a temporary qcow2 overlay. bs points to the
4061 * temporary snapshot afterwards. */
4062 if (snapshot_flags
) {
4063 BlockDriverState
*snapshot_bs
;
4064 snapshot_bs
= bdrv_append_temp_snapshot(bs
, snapshot_flags
,
4065 snapshot_options
, &local_err
);
4066 snapshot_options
= NULL
;
4068 goto close_and_fail
;
4070 /* We are not going to return bs but the overlay on top of it
4071 * (snapshot_bs); thus, we have to drop the strong reference to bs
4072 * (which we obtained by calling bdrv_new()). bs will not be deleted,
4073 * though, because the overlay still has a reference to it. */
4082 qobject_unref(snapshot_options
);
4083 qobject_unref(bs
->explicit_options
);
4084 qobject_unref(bs
->options
);
4085 qobject_unref(options
);
4087 bs
->explicit_options
= NULL
;
4089 error_propagate(errp
, local_err
);
4094 qobject_unref(snapshot_options
);
4095 qobject_unref(options
);
4096 error_propagate(errp
, local_err
);
4101 * The caller must always hold @filename AioContext lock, because this
4102 * function eventually calls bdrv_refresh_total_sectors() which polls
4103 * when called from non-coroutine context.
4105 BlockDriverState
*bdrv_open(const char *filename
, const char *reference
,
4106 QDict
*options
, int flags
, Error
**errp
)
4108 GLOBAL_STATE_CODE();
4110 return bdrv_open_inherit(filename
, reference
, options
, flags
, NULL
,
4114 /* Return true if the NULL-terminated @list contains @str */
4115 static bool is_str_in_list(const char *str
, const char *const *list
)
4119 for (i
= 0; list
[i
] != NULL
; i
++) {
4120 if (!strcmp(str
, list
[i
])) {
4129 * Check that every option set in @bs->options is also set in
4132 * Options listed in the common_options list and in
4133 * @bs->drv->mutable_opts are skipped.
4135 * Return 0 on success, otherwise return -EINVAL and set @errp.
4137 static int bdrv_reset_options_allowed(BlockDriverState
*bs
,
4138 const QDict
*new_opts
, Error
**errp
)
4140 const QDictEntry
*e
;
4141 /* These options are common to all block drivers and are handled
4142 * in bdrv_reopen_prepare() so they can be left out of @new_opts */
4143 const char *const common_options
[] = {
4144 "node-name", "discard", "cache.direct", "cache.no-flush",
4145 "read-only", "auto-read-only", "detect-zeroes", NULL
4148 for (e
= qdict_first(bs
->options
); e
; e
= qdict_next(bs
->options
, e
)) {
4149 if (!qdict_haskey(new_opts
, e
->key
) &&
4150 !is_str_in_list(e
->key
, common_options
) &&
4151 !is_str_in_list(e
->key
, bs
->drv
->mutable_opts
)) {
4152 error_setg(errp
, "Option '%s' cannot be reset "
4153 "to its default value", e
->key
);
4162 * Returns true if @child can be reached recursively from @bs
4164 static bool bdrv_recurse_has_child(BlockDriverState
*bs
,
4165 BlockDriverState
*child
)
4173 QLIST_FOREACH(c
, &bs
->children
, next
) {
4174 if (bdrv_recurse_has_child(c
->bs
, child
)) {
4183 * Adds a BlockDriverState to a simple queue for an atomic, transactional
4184 * reopen of multiple devices.
4186 * bs_queue can either be an existing BlockReopenQueue that has had QTAILQ_INIT
4187 * already performed, or alternatively may be NULL a new BlockReopenQueue will
4188 * be created and initialized. This newly created BlockReopenQueue should be
4189 * passed back in for subsequent calls that are intended to be of the same
4192 * bs is the BlockDriverState to add to the reopen queue.
4194 * options contains the changed options for the associated bs
4195 * (the BlockReopenQueue takes ownership)
4197 * flags contains the open flags for the associated bs
4199 * returns a pointer to bs_queue, which is either the newly allocated
4200 * bs_queue, or the existing bs_queue being used.
4202 * bs is drained here and undrained by bdrv_reopen_queue_free().
4204 * To be called with bs->aio_context locked.
4206 static BlockReopenQueue
*bdrv_reopen_queue_child(BlockReopenQueue
*bs_queue
,
4207 BlockDriverState
*bs
,
4209 const BdrvChildClass
*klass
,
4211 bool parent_is_format
,
4212 QDict
*parent_options
,
4218 BlockReopenQueueEntry
*bs_entry
;
4220 QDict
*old_options
, *explicit_options
, *options_copy
;
4224 GLOBAL_STATE_CODE();
4226 bdrv_drained_begin(bs
);
4228 if (bs_queue
== NULL
) {
4229 bs_queue
= g_new0(BlockReopenQueue
, 1);
4230 QTAILQ_INIT(bs_queue
);
4234 options
= qdict_new();
4237 /* Check if this BlockDriverState is already in the queue */
4238 QTAILQ_FOREACH(bs_entry
, bs_queue
, entry
) {
4239 if (bs
== bs_entry
->state
.bs
) {
4245 * Precedence of options:
4246 * 1. Explicitly passed in options (highest)
4247 * 2. Retained from explicitly set options of bs
4248 * 3. Inherited from parent node
4249 * 4. Retained from effective options of bs
4252 /* Old explicitly set values (don't overwrite by inherited value) */
4253 if (bs_entry
|| keep_old_opts
) {
4254 old_options
= qdict_clone_shallow(bs_entry
?
4255 bs_entry
->state
.explicit_options
:
4256 bs
->explicit_options
);
4257 bdrv_join_options(bs
, options
, old_options
);
4258 qobject_unref(old_options
);
4261 explicit_options
= qdict_clone_shallow(options
);
4263 /* Inherit from parent node */
4264 if (parent_options
) {
4266 klass
->inherit_options(role
, parent_is_format
, &flags
, options
,
4267 parent_flags
, parent_options
);
4269 flags
= bdrv_get_flags(bs
);
4272 if (keep_old_opts
) {
4273 /* Old values are used for options that aren't set yet */
4274 old_options
= qdict_clone_shallow(bs
->options
);
4275 bdrv_join_options(bs
, options
, old_options
);
4276 qobject_unref(old_options
);
4279 /* We have the final set of options so let's update the flags */
4280 options_copy
= qdict_clone_shallow(options
);
4281 opts
= qemu_opts_create(&bdrv_runtime_opts
, NULL
, 0, &error_abort
);
4282 qemu_opts_absorb_qdict(opts
, options_copy
, NULL
);
4283 update_flags_from_options(&flags
, opts
);
4284 qemu_opts_del(opts
);
4285 qobject_unref(options_copy
);
4287 /* bdrv_open_inherit() sets and clears some additional flags internally */
4288 flags
&= ~BDRV_O_PROTOCOL
;
4289 if (flags
& BDRV_O_RDWR
) {
4290 flags
|= BDRV_O_ALLOW_RDWR
;
4294 bs_entry
= g_new0(BlockReopenQueueEntry
, 1);
4295 QTAILQ_INSERT_TAIL(bs_queue
, bs_entry
, entry
);
4297 qobject_unref(bs_entry
->state
.options
);
4298 qobject_unref(bs_entry
->state
.explicit_options
);
4301 bs_entry
->state
.bs
= bs
;
4302 bs_entry
->state
.options
= options
;
4303 bs_entry
->state
.explicit_options
= explicit_options
;
4304 bs_entry
->state
.flags
= flags
;
4307 * If keep_old_opts is false then it means that unspecified
4308 * options must be reset to their original value. We don't allow
4309 * resetting 'backing' but we need to know if the option is
4310 * missing in order to decide if we have to return an error.
4312 if (!keep_old_opts
) {
4313 bs_entry
->state
.backing_missing
=
4314 !qdict_haskey(options
, "backing") &&
4315 !qdict_haskey(options
, "backing.driver");
4318 QLIST_FOREACH(child
, &bs
->children
, next
) {
4319 QDict
*new_child_options
= NULL
;
4320 bool child_keep_old
= keep_old_opts
;
4322 /* reopen can only change the options of block devices that were
4323 * implicitly created and inherited options. For other (referenced)
4324 * block devices, a syntax like "backing.foo" results in an error. */
4325 if (child
->bs
->inherits_from
!= bs
) {
4329 /* Check if the options contain a child reference */
4330 if (qdict_haskey(options
, child
->name
)) {
4331 const char *childref
= qdict_get_try_str(options
, child
->name
);
4333 * The current child must not be reopened if the child
4334 * reference is null or points to a different node.
4336 if (g_strcmp0(childref
, child
->bs
->node_name
)) {
4340 * If the child reference points to the current child then
4341 * reopen it with its existing set of options (note that
4342 * it can still inherit new options from the parent).
4344 child_keep_old
= true;
4346 /* Extract child options ("child-name.*") */
4347 char *child_key_dot
= g_strdup_printf("%s.", child
->name
);
4348 qdict_extract_subqdict(explicit_options
, NULL
, child_key_dot
);
4349 qdict_extract_subqdict(options
, &new_child_options
, child_key_dot
);
4350 g_free(child_key_dot
);
4353 bdrv_reopen_queue_child(bs_queue
, child
->bs
, new_child_options
,
4354 child
->klass
, child
->role
, bs
->drv
->is_format
,
4355 options
, flags
, child_keep_old
);
4361 /* To be called with bs->aio_context locked */
4362 BlockReopenQueue
*bdrv_reopen_queue(BlockReopenQueue
*bs_queue
,
4363 BlockDriverState
*bs
,
4364 QDict
*options
, bool keep_old_opts
)
4366 GLOBAL_STATE_CODE();
4368 return bdrv_reopen_queue_child(bs_queue
, bs
, options
, NULL
, 0, false,
4369 NULL
, 0, keep_old_opts
);
4372 void bdrv_reopen_queue_free(BlockReopenQueue
*bs_queue
)
4374 GLOBAL_STATE_CODE();
4376 BlockReopenQueueEntry
*bs_entry
, *next
;
4377 QTAILQ_FOREACH_SAFE(bs_entry
, bs_queue
, entry
, next
) {
4378 AioContext
*ctx
= bdrv_get_aio_context(bs_entry
->state
.bs
);
4380 aio_context_acquire(ctx
);
4381 bdrv_drained_end(bs_entry
->state
.bs
);
4382 aio_context_release(ctx
);
4384 qobject_unref(bs_entry
->state
.explicit_options
);
4385 qobject_unref(bs_entry
->state
.options
);
4393 * Reopen multiple BlockDriverStates atomically & transactionally.
4395 * The queue passed in (bs_queue) must have been built up previous
4396 * via bdrv_reopen_queue().
4398 * Reopens all BDS specified in the queue, with the appropriate
4399 * flags. All devices are prepared for reopen, and failure of any
4400 * device will cause all device changes to be abandoned, and intermediate
4403 * If all devices prepare successfully, then the changes are committed
4406 * All affected nodes must be drained between bdrv_reopen_queue() and
4407 * bdrv_reopen_multiple().
4409 * To be called from the main thread, with all other AioContexts unlocked.
4411 int bdrv_reopen_multiple(BlockReopenQueue
*bs_queue
, Error
**errp
)
4414 BlockReopenQueueEntry
*bs_entry
, *next
;
4416 Transaction
*tran
= tran_new();
4417 g_autoptr(GSList
) refresh_list
= NULL
;
4419 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
4420 assert(bs_queue
!= NULL
);
4421 GLOBAL_STATE_CODE();
4423 QTAILQ_FOREACH(bs_entry
, bs_queue
, entry
) {
4424 ctx
= bdrv_get_aio_context(bs_entry
->state
.bs
);
4425 aio_context_acquire(ctx
);
4426 ret
= bdrv_flush(bs_entry
->state
.bs
);
4427 aio_context_release(ctx
);
4429 error_setg_errno(errp
, -ret
, "Error flushing drive");
4434 QTAILQ_FOREACH(bs_entry
, bs_queue
, entry
) {
4435 assert(bs_entry
->state
.bs
->quiesce_counter
> 0);
4436 ctx
= bdrv_get_aio_context(bs_entry
->state
.bs
);
4437 aio_context_acquire(ctx
);
4438 ret
= bdrv_reopen_prepare(&bs_entry
->state
, bs_queue
, tran
, errp
);
4439 aio_context_release(ctx
);
4443 bs_entry
->prepared
= true;
4446 QTAILQ_FOREACH(bs_entry
, bs_queue
, entry
) {
4447 BDRVReopenState
*state
= &bs_entry
->state
;
4449 refresh_list
= g_slist_prepend(refresh_list
, state
->bs
);
4450 if (state
->old_backing_bs
) {
4451 refresh_list
= g_slist_prepend(refresh_list
, state
->old_backing_bs
);
4453 if (state
->old_file_bs
) {
4454 refresh_list
= g_slist_prepend(refresh_list
, state
->old_file_bs
);
4459 * Note that file-posix driver rely on permission update done during reopen
4460 * (even if no permission changed), because it wants "new" permissions for
4461 * reconfiguring the fd and that's why it does it in raw_check_perm(), not
4462 * in raw_reopen_prepare() which is called with "old" permissions.
4464 ret
= bdrv_list_refresh_perms(refresh_list
, bs_queue
, tran
, errp
);
4470 * If we reach this point, we have success and just need to apply the
4473 * Reverse order is used to comfort qcow2 driver: on commit it need to write
4474 * IN_USE flag to the image, to mark bitmaps in the image as invalid. But
4475 * children are usually goes after parents in reopen-queue, so go from last
4478 QTAILQ_FOREACH_REVERSE(bs_entry
, bs_queue
, entry
) {
4479 ctx
= bdrv_get_aio_context(bs_entry
->state
.bs
);
4480 aio_context_acquire(ctx
);
4481 bdrv_reopen_commit(&bs_entry
->state
);
4482 aio_context_release(ctx
);
4487 QTAILQ_FOREACH_REVERSE(bs_entry
, bs_queue
, entry
) {
4488 BlockDriverState
*bs
= bs_entry
->state
.bs
;
4490 if (bs
->drv
->bdrv_reopen_commit_post
) {
4491 ctx
= bdrv_get_aio_context(bs
);
4492 aio_context_acquire(ctx
);
4493 bs
->drv
->bdrv_reopen_commit_post(&bs_entry
->state
);
4494 aio_context_release(ctx
);
4503 QTAILQ_FOREACH_SAFE(bs_entry
, bs_queue
, entry
, next
) {
4504 if (bs_entry
->prepared
) {
4505 ctx
= bdrv_get_aio_context(bs_entry
->state
.bs
);
4506 aio_context_acquire(ctx
);
4507 bdrv_reopen_abort(&bs_entry
->state
);
4508 aio_context_release(ctx
);
4513 bdrv_reopen_queue_free(bs_queue
);
4518 int bdrv_reopen(BlockDriverState
*bs
, QDict
*opts
, bool keep_old_opts
,
4521 AioContext
*ctx
= bdrv_get_aio_context(bs
);
4522 BlockReopenQueue
*queue
;
4525 GLOBAL_STATE_CODE();
4527 queue
= bdrv_reopen_queue(NULL
, bs
, opts
, keep_old_opts
);
4529 if (ctx
!= qemu_get_aio_context()) {
4530 aio_context_release(ctx
);
4532 ret
= bdrv_reopen_multiple(queue
, errp
);
4534 if (ctx
!= qemu_get_aio_context()) {
4535 aio_context_acquire(ctx
);
4541 int bdrv_reopen_set_read_only(BlockDriverState
*bs
, bool read_only
,
4544 QDict
*opts
= qdict_new();
4546 GLOBAL_STATE_CODE();
4548 qdict_put_bool(opts
, BDRV_OPT_READ_ONLY
, read_only
);
4550 return bdrv_reopen(bs
, opts
, true, errp
);
4554 * Take a BDRVReopenState and check if the value of 'backing' in the
4555 * reopen_state->options QDict is valid or not.
4557 * If 'backing' is missing from the QDict then return 0.
4559 * If 'backing' contains the node name of the backing file of
4560 * reopen_state->bs then return 0.
4562 * If 'backing' contains a different node name (or is null) then check
4563 * whether the current backing file can be replaced with the new one.
4564 * If that's the case then reopen_state->replace_backing_bs is set to
4565 * true and reopen_state->new_backing_bs contains a pointer to the new
4566 * backing BlockDriverState (or NULL).
4568 * Return 0 on success, otherwise return < 0 and set @errp.
4570 static int bdrv_reopen_parse_file_or_backing(BDRVReopenState
*reopen_state
,
4571 bool is_backing
, Transaction
*tran
,
4574 BlockDriverState
*bs
= reopen_state
->bs
;
4575 BlockDriverState
*new_child_bs
;
4576 BlockDriverState
*old_child_bs
= is_backing
? child_bs(bs
->backing
) :
4578 const char *child_name
= is_backing
? "backing" : "file";
4582 GLOBAL_STATE_CODE();
4584 value
= qdict_get(reopen_state
->options
, child_name
);
4585 if (value
== NULL
) {
4589 switch (qobject_type(value
)) {
4591 assert(is_backing
); /* The 'file' option does not allow a null value */
4592 new_child_bs
= NULL
;
4595 str
= qstring_get_str(qobject_to(QString
, value
));
4596 new_child_bs
= bdrv_lookup_bs(NULL
, str
, errp
);
4597 if (new_child_bs
== NULL
) {
4599 } else if (bdrv_recurse_has_child(new_child_bs
, bs
)) {
4600 error_setg(errp
, "Making '%s' a %s child of '%s' would create a "
4601 "cycle", str
, child_name
, bs
->node_name
);
4607 * The options QDict has been flattened, so 'backing' and 'file'
4608 * do not allow any other data type here.
4610 g_assert_not_reached();
4613 if (old_child_bs
== new_child_bs
) {
4618 if (bdrv_skip_implicit_filters(old_child_bs
) == new_child_bs
) {
4622 if (old_child_bs
->implicit
) {
4623 error_setg(errp
, "Cannot replace implicit %s child of %s",
4624 child_name
, bs
->node_name
);
4629 if (bs
->drv
->is_filter
&& !old_child_bs
) {
4631 * Filters always have a file or a backing child, so we are trying to
4632 * change wrong child
4634 error_setg(errp
, "'%s' is a %s filter node that does not support a "
4635 "%s child", bs
->node_name
, bs
->drv
->format_name
, child_name
);
4640 reopen_state
->old_backing_bs
= old_child_bs
;
4642 reopen_state
->old_file_bs
= old_child_bs
;
4645 return bdrv_set_file_or_backing_noperm(bs
, new_child_bs
, is_backing
,
4650 * Prepares a BlockDriverState for reopen. All changes are staged in the
4651 * 'opaque' field of the BDRVReopenState, which is used and allocated by
4652 * the block driver layer .bdrv_reopen_prepare()
4654 * bs is the BlockDriverState to reopen
4655 * flags are the new open flags
4656 * queue is the reopen queue
4658 * Returns 0 on success, non-zero on error. On error errp will be set
4661 * On failure, bdrv_reopen_abort() will be called to clean up any data.
4662 * It is the responsibility of the caller to then call the abort() or
4663 * commit() for any other BDS that have been left in a prepare() state
4666 static int bdrv_reopen_prepare(BDRVReopenState
*reopen_state
,
4667 BlockReopenQueue
*queue
,
4668 Transaction
*change_child_tran
, Error
**errp
)
4672 Error
*local_err
= NULL
;
4675 QDict
*orig_reopen_opts
;
4676 char *discard
= NULL
;
4678 bool drv_prepared
= false;
4680 assert(reopen_state
!= NULL
);
4681 assert(reopen_state
->bs
->drv
!= NULL
);
4682 GLOBAL_STATE_CODE();
4683 drv
= reopen_state
->bs
->drv
;
4685 /* This function and each driver's bdrv_reopen_prepare() remove
4686 * entries from reopen_state->options as they are processed, so
4687 * we need to make a copy of the original QDict. */
4688 orig_reopen_opts
= qdict_clone_shallow(reopen_state
->options
);
4690 /* Process generic block layer options */
4691 opts
= qemu_opts_create(&bdrv_runtime_opts
, NULL
, 0, &error_abort
);
4692 if (!qemu_opts_absorb_qdict(opts
, reopen_state
->options
, errp
)) {
4697 /* This was already called in bdrv_reopen_queue_child() so the flags
4698 * are up-to-date. This time we simply want to remove the options from
4699 * QemuOpts in order to indicate that they have been processed. */
4700 old_flags
= reopen_state
->flags
;
4701 update_flags_from_options(&reopen_state
->flags
, opts
);
4702 assert(old_flags
== reopen_state
->flags
);
4704 discard
= qemu_opt_get_del(opts
, BDRV_OPT_DISCARD
);
4705 if (discard
!= NULL
) {
4706 if (bdrv_parse_discard_flags(discard
, &reopen_state
->flags
) != 0) {
4707 error_setg(errp
, "Invalid discard option");
4713 reopen_state
->detect_zeroes
=
4714 bdrv_parse_detect_zeroes(opts
, reopen_state
->flags
, &local_err
);
4716 error_propagate(errp
, local_err
);
4721 /* All other options (including node-name and driver) must be unchanged.
4722 * Put them back into the QDict, so that they are checked at the end
4723 * of this function. */
4724 qemu_opts_to_qdict(opts
, reopen_state
->options
);
4726 /* If we are to stay read-only, do not allow permission change
4727 * to r/w. Attempting to set to r/w may fail if either BDRV_O_ALLOW_RDWR is
4728 * not set, or if the BDS still has copy_on_read enabled */
4729 read_only
= !(reopen_state
->flags
& BDRV_O_RDWR
);
4730 ret
= bdrv_can_set_read_only(reopen_state
->bs
, read_only
, true, &local_err
);
4732 error_propagate(errp
, local_err
);
4736 if (drv
->bdrv_reopen_prepare
) {
4738 * If a driver-specific option is missing, it means that we
4739 * should reset it to its default value.
4740 * But not all options allow that, so we need to check it first.
4742 ret
= bdrv_reset_options_allowed(reopen_state
->bs
,
4743 reopen_state
->options
, errp
);
4748 ret
= drv
->bdrv_reopen_prepare(reopen_state
, queue
, &local_err
);
4750 if (local_err
!= NULL
) {
4751 error_propagate(errp
, local_err
);
4753 bdrv_refresh_filename(reopen_state
->bs
);
4754 error_setg(errp
, "failed while preparing to reopen image '%s'",
4755 reopen_state
->bs
->filename
);
4760 /* It is currently mandatory to have a bdrv_reopen_prepare()
4761 * handler for each supported drv. */
4762 error_setg(errp
, "Block format '%s' used by node '%s' "
4763 "does not support reopening files", drv
->format_name
,
4764 bdrv_get_device_or_node_name(reopen_state
->bs
));
4769 drv_prepared
= true;
4772 * We must provide the 'backing' option if the BDS has a backing
4773 * file or if the image file has a backing file name as part of
4774 * its metadata. Otherwise the 'backing' option can be omitted.
4776 if (drv
->supports_backing
&& reopen_state
->backing_missing
&&
4777 (reopen_state
->bs
->backing
|| reopen_state
->bs
->backing_file
[0])) {
4778 error_setg(errp
, "backing is missing for '%s'",
4779 reopen_state
->bs
->node_name
);
4785 * Allow changing the 'backing' option. The new value can be
4786 * either a reference to an existing node (using its node name)
4787 * or NULL to simply detach the current backing file.
4789 ret
= bdrv_reopen_parse_file_or_backing(reopen_state
, true,
4790 change_child_tran
, errp
);
4794 qdict_del(reopen_state
->options
, "backing");
4796 /* Allow changing the 'file' option. In this case NULL is not allowed */
4797 ret
= bdrv_reopen_parse_file_or_backing(reopen_state
, false,
4798 change_child_tran
, errp
);
4802 qdict_del(reopen_state
->options
, "file");
4804 /* Options that are not handled are only okay if they are unchanged
4805 * compared to the old state. It is expected that some options are only
4806 * used for the initial open, but not reopen (e.g. filename) */
4807 if (qdict_size(reopen_state
->options
)) {
4808 const QDictEntry
*entry
= qdict_first(reopen_state
->options
);
4811 QObject
*new = entry
->value
;
4812 QObject
*old
= qdict_get(reopen_state
->bs
->options
, entry
->key
);
4814 /* Allow child references (child_name=node_name) as long as they
4815 * point to the current child (i.e. everything stays the same). */
4816 if (qobject_type(new) == QTYPE_QSTRING
) {
4818 QLIST_FOREACH(child
, &reopen_state
->bs
->children
, next
) {
4819 if (!strcmp(child
->name
, entry
->key
)) {
4825 if (!strcmp(child
->bs
->node_name
,
4826 qstring_get_str(qobject_to(QString
, new)))) {
4827 continue; /* Found child with this name, skip option */
4833 * TODO: When using -drive to specify blockdev options, all values
4834 * will be strings; however, when using -blockdev, blockdev-add or
4835 * filenames using the json:{} pseudo-protocol, they will be
4837 * In contrast, reopening options are (currently) always strings
4838 * (because you can only specify them through qemu-io; all other
4839 * callers do not specify any options).
4840 * Therefore, when using anything other than -drive to create a BDS,
4841 * this cannot detect non-string options as unchanged, because
4842 * qobject_is_equal() always returns false for objects of different
4843 * type. In the future, this should be remedied by correctly typing
4844 * all options. For now, this is not too big of an issue because
4845 * the user can simply omit options which cannot be changed anyway,
4846 * so they will stay unchanged.
4848 if (!qobject_is_equal(new, old
)) {
4849 error_setg(errp
, "Cannot change the option '%s'", entry
->key
);
4853 } while ((entry
= qdict_next(reopen_state
->options
, entry
)));
4858 /* Restore the original reopen_state->options QDict */
4859 qobject_unref(reopen_state
->options
);
4860 reopen_state
->options
= qobject_ref(orig_reopen_opts
);
4863 if (ret
< 0 && drv_prepared
) {
4864 /* drv->bdrv_reopen_prepare() has succeeded, so we need to
4865 * call drv->bdrv_reopen_abort() before signaling an error
4866 * (bdrv_reopen_multiple() will not call bdrv_reopen_abort()
4867 * when the respective bdrv_reopen_prepare() has failed) */
4868 if (drv
->bdrv_reopen_abort
) {
4869 drv
->bdrv_reopen_abort(reopen_state
);
4872 qemu_opts_del(opts
);
4873 qobject_unref(orig_reopen_opts
);
4879 * Takes the staged changes for the reopen from bdrv_reopen_prepare(), and
4880 * makes them final by swapping the staging BlockDriverState contents into
4881 * the active BlockDriverState contents.
4883 static void bdrv_reopen_commit(BDRVReopenState
*reopen_state
)
4886 BlockDriverState
*bs
;
4889 assert(reopen_state
!= NULL
);
4890 bs
= reopen_state
->bs
;
4892 assert(drv
!= NULL
);
4893 GLOBAL_STATE_CODE();
4895 /* If there are any driver level actions to take */
4896 if (drv
->bdrv_reopen_commit
) {
4897 drv
->bdrv_reopen_commit(reopen_state
);
4900 /* set BDS specific flags now */
4901 qobject_unref(bs
->explicit_options
);
4902 qobject_unref(bs
->options
);
4903 qobject_ref(reopen_state
->explicit_options
);
4904 qobject_ref(reopen_state
->options
);
4906 bs
->explicit_options
= reopen_state
->explicit_options
;
4907 bs
->options
= reopen_state
->options
;
4908 bs
->open_flags
= reopen_state
->flags
;
4909 bs
->detect_zeroes
= reopen_state
->detect_zeroes
;
4911 /* Remove child references from bs->options and bs->explicit_options.
4912 * Child options were already removed in bdrv_reopen_queue_child() */
4913 QLIST_FOREACH(child
, &bs
->children
, next
) {
4914 qdict_del(bs
->explicit_options
, child
->name
);
4915 qdict_del(bs
->options
, child
->name
);
4917 /* backing is probably removed, so it's not handled by previous loop */
4918 qdict_del(bs
->explicit_options
, "backing");
4919 qdict_del(bs
->options
, "backing");
4921 bdrv_refresh_limits(bs
, NULL
, NULL
);
4925 * Abort the reopen, and delete and free the staged changes in
4928 static void bdrv_reopen_abort(BDRVReopenState
*reopen_state
)
4932 assert(reopen_state
!= NULL
);
4933 drv
= reopen_state
->bs
->drv
;
4934 assert(drv
!= NULL
);
4935 GLOBAL_STATE_CODE();
4937 if (drv
->bdrv_reopen_abort
) {
4938 drv
->bdrv_reopen_abort(reopen_state
);
4943 static void bdrv_close(BlockDriverState
*bs
)
4945 BdrvAioNotifier
*ban
, *ban_next
;
4946 BdrvChild
*child
, *next
;
4948 GLOBAL_STATE_CODE();
4949 assert(!bs
->refcnt
);
4951 bdrv_drained_begin(bs
); /* complete I/O */
4953 bdrv_drain(bs
); /* in case flush left pending I/O */
4956 if (bs
->drv
->bdrv_close
) {
4957 /* Must unfreeze all children, so bdrv_unref_child() works */
4958 bs
->drv
->bdrv_close(bs
);
4963 QLIST_FOREACH_SAFE(child
, &bs
->children
, next
, next
) {
4964 bdrv_unref_child(bs
, child
);
4967 assert(!bs
->backing
);
4971 qatomic_set(&bs
->copy_on_read
, 0);
4972 bs
->backing_file
[0] = '\0';
4973 bs
->backing_format
[0] = '\0';
4974 bs
->total_sectors
= 0;
4975 bs
->encrypted
= false;
4977 qobject_unref(bs
->options
);
4978 qobject_unref(bs
->explicit_options
);
4980 bs
->explicit_options
= NULL
;
4981 qobject_unref(bs
->full_open_options
);
4982 bs
->full_open_options
= NULL
;
4983 g_free(bs
->block_status_cache
);
4984 bs
->block_status_cache
= NULL
;
4986 bdrv_release_named_dirty_bitmaps(bs
);
4987 assert(QLIST_EMPTY(&bs
->dirty_bitmaps
));
4989 QLIST_FOREACH_SAFE(ban
, &bs
->aio_notifiers
, list
, ban_next
) {
4992 QLIST_INIT(&bs
->aio_notifiers
);
4993 bdrv_drained_end(bs
);
4996 * If we're still inside some bdrv_drain_all_begin()/end() sections, end
4997 * them now since this BDS won't exist anymore when bdrv_drain_all_end()
5000 if (bs
->quiesce_counter
) {
5001 bdrv_drain_all_end_quiesce(bs
);
5005 void bdrv_close_all(void)
5007 GLOBAL_STATE_CODE();
5008 assert(job_next(NULL
) == NULL
);
5010 /* Drop references from requests still in flight, such as canceled block
5011 * jobs whose AIO context has not been polled yet */
5014 blk_remove_all_bs();
5015 blockdev_close_all_bdrv_states();
5017 assert(QTAILQ_EMPTY(&all_bdrv_states
));
5020 static bool should_update_child(BdrvChild
*c
, BlockDriverState
*to
)
5026 if (c
->klass
->stay_at_node
) {
5030 /* If the child @c belongs to the BDS @to, replacing the current
5031 * c->bs by @to would mean to create a loop.
5033 * Such a case occurs when appending a BDS to a backing chain.
5034 * For instance, imagine the following chain:
5036 * guest device -> node A -> further backing chain...
5038 * Now we create a new BDS B which we want to put on top of this
5039 * chain, so we first attach A as its backing node:
5044 * guest device -> node A -> further backing chain...
5046 * Finally we want to replace A by B. When doing that, we want to
5047 * replace all pointers to A by pointers to B -- except for the
5048 * pointer from B because (1) that would create a loop, and (2)
5049 * that pointer should simply stay intact:
5051 * guest device -> node B
5054 * node A -> further backing chain...
5056 * In general, when replacing a node A (c->bs) by a node B (@to),
5057 * if A is a child of B, that means we cannot replace A by B there
5058 * because that would create a loop. Silently detaching A from B
5059 * is also not really an option. So overall just leaving A in
5060 * place there is the most sensible choice.
5062 * We would also create a loop in any cases where @c is only
5063 * indirectly referenced by @to. Prevent this by returning false
5064 * if @c is found (by breadth-first search) anywhere in the whole
5069 found
= g_hash_table_new(NULL
, NULL
);
5070 g_hash_table_add(found
, to
);
5071 queue
= g_queue_new();
5072 g_queue_push_tail(queue
, to
);
5074 while (!g_queue_is_empty(queue
)) {
5075 BlockDriverState
*v
= g_queue_pop_head(queue
);
5078 QLIST_FOREACH(c2
, &v
->children
, next
) {
5084 if (g_hash_table_contains(found
, c2
->bs
)) {
5088 g_queue_push_tail(queue
, c2
->bs
);
5089 g_hash_table_add(found
, c2
->bs
);
5093 g_queue_free(queue
);
5094 g_hash_table_destroy(found
);
5099 static void bdrv_remove_child_commit(void *opaque
)
5101 GLOBAL_STATE_CODE();
5102 bdrv_child_free(opaque
);
5105 static TransactionActionDrv bdrv_remove_child_drv
= {
5106 .commit
= bdrv_remove_child_commit
,
5109 /* Function doesn't update permissions, caller is responsible for this. */
5110 static void bdrv_remove_child(BdrvChild
*child
, Transaction
*tran
)
5117 BlockDriverState
*bs
= child
->bs
;
5118 bdrv_drained_begin(bs
);
5119 bdrv_replace_child_tran(child
, NULL
, tran
);
5120 bdrv_drained_end(bs
);
5123 tran_add(tran
, &bdrv_remove_child_drv
, child
);
5126 static void undrain_on_clean_cb(void *opaque
)
5128 bdrv_drained_end(opaque
);
5131 static TransactionActionDrv undrain_on_clean
= {
5132 .clean
= undrain_on_clean_cb
,
5135 static int bdrv_replace_node_noperm(BlockDriverState
*from
,
5136 BlockDriverState
*to
,
5137 bool auto_skip
, Transaction
*tran
,
5140 BdrvChild
*c
, *next
;
5142 GLOBAL_STATE_CODE();
5144 bdrv_drained_begin(from
);
5145 bdrv_drained_begin(to
);
5146 tran_add(tran
, &undrain_on_clean
, from
);
5147 tran_add(tran
, &undrain_on_clean
, to
);
5149 QLIST_FOREACH_SAFE(c
, &from
->parents
, next_parent
, next
) {
5150 assert(c
->bs
== from
);
5151 if (!should_update_child(c
, to
)) {
5155 error_setg(errp
, "Should not change '%s' link to '%s'",
5156 c
->name
, from
->node_name
);
5160 error_setg(errp
, "Cannot change '%s' link to '%s'",
5161 c
->name
, from
->node_name
);
5164 bdrv_replace_child_tran(c
, to
, tran
);
5171 * With auto_skip=true bdrv_replace_node_common skips updating from parents
5172 * if it creates a parent-child relation loop or if parent is block-job.
5174 * With auto_skip=false the error is returned if from has a parent which should
5177 * With @detach_subchain=true @to must be in a backing chain of @from. In this
5178 * case backing link of the cow-parent of @to is removed.
5180 static int bdrv_replace_node_common(BlockDriverState
*from
,
5181 BlockDriverState
*to
,
5182 bool auto_skip
, bool detach_subchain
,
5185 Transaction
*tran
= tran_new();
5186 g_autoptr(GSList
) refresh_list
= NULL
;
5187 BlockDriverState
*to_cow_parent
= NULL
;
5190 GLOBAL_STATE_CODE();
5192 if (detach_subchain
) {
5193 assert(bdrv_chain_contains(from
, to
));
5195 for (to_cow_parent
= from
;
5196 bdrv_filter_or_cow_bs(to_cow_parent
) != to
;
5197 to_cow_parent
= bdrv_filter_or_cow_bs(to_cow_parent
))
5203 /* Make sure that @from doesn't go away until we have successfully attached
5204 * all of its parents to @to. */
5207 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
5208 assert(bdrv_get_aio_context(from
) == bdrv_get_aio_context(to
));
5209 bdrv_drained_begin(from
);
5212 * Do the replacement without permission update.
5213 * Replacement may influence the permissions, we should calculate new
5214 * permissions based on new graph. If we fail, we'll roll-back the
5217 ret
= bdrv_replace_node_noperm(from
, to
, auto_skip
, tran
, errp
);
5222 if (detach_subchain
) {
5223 bdrv_remove_child(bdrv_filter_or_cow_child(to_cow_parent
), tran
);
5226 refresh_list
= g_slist_prepend(refresh_list
, to
);
5227 refresh_list
= g_slist_prepend(refresh_list
, from
);
5229 ret
= bdrv_list_refresh_perms(refresh_list
, NULL
, tran
, errp
);
5237 tran_finalize(tran
, ret
);
5239 bdrv_drained_end(from
);
5245 int bdrv_replace_node(BlockDriverState
*from
, BlockDriverState
*to
,
5248 GLOBAL_STATE_CODE();
5250 return bdrv_replace_node_common(from
, to
, true, false, errp
);
5253 int bdrv_drop_filter(BlockDriverState
*bs
, Error
**errp
)
5255 GLOBAL_STATE_CODE();
5257 return bdrv_replace_node_common(bs
, bdrv_filter_or_cow_bs(bs
), true, true,
5262 * Add new bs contents at the top of an image chain while the chain is
5263 * live, while keeping required fields on the top layer.
5265 * This will modify the BlockDriverState fields, and swap contents
5266 * between bs_new and bs_top. Both bs_new and bs_top are modified.
5268 * bs_new must not be attached to a BlockBackend and must not have backing
5271 * This function does not create any image files.
5273 * The caller must hold the AioContext lock for @bs_top.
5275 int bdrv_append(BlockDriverState
*bs_new
, BlockDriverState
*bs_top
,
5280 Transaction
*tran
= tran_new();
5281 AioContext
*old_context
, *new_context
= NULL
;
5283 GLOBAL_STATE_CODE();
5285 assert(!bs_new
->backing
);
5287 old_context
= bdrv_get_aio_context(bs_top
);
5289 child
= bdrv_attach_child_noperm(bs_new
, bs_top
, "backing",
5290 &child_of_bds
, bdrv_backing_role(bs_new
),
5298 * bdrv_attach_child_noperm could change the AioContext of bs_top.
5299 * bdrv_replace_node_noperm calls bdrv_drained_begin, so let's temporarily
5300 * hold the new AioContext, since bdrv_drained_begin calls BDRV_POLL_WHILE
5301 * that assumes the new lock is taken.
5303 new_context
= bdrv_get_aio_context(bs_top
);
5305 if (old_context
!= new_context
) {
5306 aio_context_release(old_context
);
5307 aio_context_acquire(new_context
);
5310 ret
= bdrv_replace_node_noperm(bs_top
, bs_new
, true, tran
, errp
);
5315 ret
= bdrv_refresh_perms(bs_new
, tran
, errp
);
5317 tran_finalize(tran
, ret
);
5319 bdrv_refresh_limits(bs_top
, NULL
, NULL
);
5321 if (new_context
&& old_context
!= new_context
) {
5322 aio_context_release(new_context
);
5323 aio_context_acquire(old_context
);
5329 /* Not for empty child */
5330 int bdrv_replace_child_bs(BdrvChild
*child
, BlockDriverState
*new_bs
,
5334 Transaction
*tran
= tran_new();
5335 g_autoptr(GSList
) refresh_list
= NULL
;
5336 BlockDriverState
*old_bs
= child
->bs
;
5338 GLOBAL_STATE_CODE();
5341 bdrv_drained_begin(old_bs
);
5342 bdrv_drained_begin(new_bs
);
5344 bdrv_replace_child_tran(child
, new_bs
, tran
);
5346 refresh_list
= g_slist_prepend(refresh_list
, old_bs
);
5347 refresh_list
= g_slist_prepend(refresh_list
, new_bs
);
5349 ret
= bdrv_list_refresh_perms(refresh_list
, NULL
, tran
, errp
);
5351 tran_finalize(tran
, ret
);
5353 bdrv_drained_end(old_bs
);
5354 bdrv_drained_end(new_bs
);
5360 static void bdrv_delete(BlockDriverState
*bs
)
5362 assert(bdrv_op_blocker_is_empty(bs
));
5363 assert(!bs
->refcnt
);
5364 GLOBAL_STATE_CODE();
5366 /* remove from list, if necessary */
5367 if (bs
->node_name
[0] != '\0') {
5368 QTAILQ_REMOVE(&graph_bdrv_states
, bs
, node_list
);
5370 QTAILQ_REMOVE(&all_bdrv_states
, bs
, bs_list
);
5379 * Replace @bs by newly created block node.
5381 * @options is a QDict of options to pass to the block drivers, or NULL for an
5382 * empty set of options. The reference to the QDict belongs to the block layer
5383 * after the call (even on failure), so if the caller intends to reuse the
5384 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
5386 BlockDriverState
*bdrv_insert_node(BlockDriverState
*bs
, QDict
*options
,
5387 int flags
, Error
**errp
)
5391 BlockDriverState
*new_node_bs
= NULL
;
5392 const char *drvname
, *node_name
;
5395 drvname
= qdict_get_try_str(options
, "driver");
5397 error_setg(errp
, "driver is not specified");
5401 drv
= bdrv_find_format(drvname
);
5403 error_setg(errp
, "Unknown driver: '%s'", drvname
);
5407 node_name
= qdict_get_try_str(options
, "node-name");
5409 GLOBAL_STATE_CODE();
5411 new_node_bs
= bdrv_new_open_driver_opts(drv
, node_name
, options
, flags
,
5413 options
= NULL
; /* bdrv_new_open_driver() eats options */
5415 error_prepend(errp
, "Could not create node: ");
5419 bdrv_drained_begin(bs
);
5420 ret
= bdrv_replace_node(bs
, new_node_bs
, errp
);
5421 bdrv_drained_end(bs
);
5424 error_prepend(errp
, "Could not replace node: ");
5431 qobject_unref(options
);
5432 bdrv_unref(new_node_bs
);
5437 * Run consistency checks on an image
5439 * Returns 0 if the check could be completed (it doesn't mean that the image is
5440 * free of errors) or -errno when an internal error occurred. The results of the
5441 * check are stored in res.
5443 int coroutine_fn
bdrv_co_check(BlockDriverState
*bs
,
5444 BdrvCheckResult
*res
, BdrvCheckMode fix
)
5447 assert_bdrv_graph_readable();
5448 if (bs
->drv
== NULL
) {
5451 if (bs
->drv
->bdrv_co_check
== NULL
) {
5455 memset(res
, 0, sizeof(*res
));
5456 return bs
->drv
->bdrv_co_check(bs
, res
, fix
);
5462 * -EINVAL - backing format specified, but no file
5463 * -ENOSPC - can't update the backing file because no space is left in the
5465 * -ENOTSUP - format driver doesn't support changing the backing file
5467 int bdrv_change_backing_file(BlockDriverState
*bs
, const char *backing_file
,
5468 const char *backing_fmt
, bool require
)
5470 BlockDriver
*drv
= bs
->drv
;
5473 GLOBAL_STATE_CODE();
5479 /* Backing file format doesn't make sense without a backing file */
5480 if (backing_fmt
&& !backing_file
) {
5484 if (require
&& backing_file
&& !backing_fmt
) {
5488 if (drv
->bdrv_change_backing_file
!= NULL
) {
5489 ret
= drv
->bdrv_change_backing_file(bs
, backing_file
, backing_fmt
);
5495 pstrcpy(bs
->backing_file
, sizeof(bs
->backing_file
), backing_file
?: "");
5496 pstrcpy(bs
->backing_format
, sizeof(bs
->backing_format
), backing_fmt
?: "");
5497 pstrcpy(bs
->auto_backing_file
, sizeof(bs
->auto_backing_file
),
5498 backing_file
?: "");
5504 * Finds the first non-filter node above bs in the chain between
5505 * active and bs. The returned node is either an immediate parent of
5506 * bs, or there are only filter nodes between the two.
5508 * Returns NULL if bs is not found in active's image chain,
5509 * or if active == bs.
5511 * Returns the bottommost base image if bs == NULL.
5513 BlockDriverState
*bdrv_find_overlay(BlockDriverState
*active
,
5514 BlockDriverState
*bs
)
5517 GLOBAL_STATE_CODE();
5519 bs
= bdrv_skip_filters(bs
);
5520 active
= bdrv_skip_filters(active
);
5523 BlockDriverState
*next
= bdrv_backing_chain_next(active
);
5533 /* Given a BDS, searches for the base layer. */
5534 BlockDriverState
*bdrv_find_base(BlockDriverState
*bs
)
5536 GLOBAL_STATE_CODE();
5538 return bdrv_find_overlay(bs
, NULL
);
5542 * Return true if at least one of the COW (backing) and filter links
5543 * between @bs and @base is frozen. @errp is set if that's the case.
5544 * @base must be reachable from @bs, or NULL.
5546 bool bdrv_is_backing_chain_frozen(BlockDriverState
*bs
, BlockDriverState
*base
,
5549 BlockDriverState
*i
;
5552 GLOBAL_STATE_CODE();
5554 for (i
= bs
; i
!= base
; i
= child_bs(child
)) {
5555 child
= bdrv_filter_or_cow_child(i
);
5557 if (child
&& child
->frozen
) {
5558 error_setg(errp
, "Cannot change '%s' link from '%s' to '%s'",
5559 child
->name
, i
->node_name
, child
->bs
->node_name
);
5568 * Freeze all COW (backing) and filter links between @bs and @base.
5569 * If any of the links is already frozen the operation is aborted and
5570 * none of the links are modified.
5571 * @base must be reachable from @bs, or NULL.
5572 * Returns 0 on success. On failure returns < 0 and sets @errp.
5574 int bdrv_freeze_backing_chain(BlockDriverState
*bs
, BlockDriverState
*base
,
5577 BlockDriverState
*i
;
5580 GLOBAL_STATE_CODE();
5582 if (bdrv_is_backing_chain_frozen(bs
, base
, errp
)) {
5586 for (i
= bs
; i
!= base
; i
= child_bs(child
)) {
5587 child
= bdrv_filter_or_cow_child(i
);
5588 if (child
&& child
->bs
->never_freeze
) {
5589 error_setg(errp
, "Cannot freeze '%s' link to '%s'",
5590 child
->name
, child
->bs
->node_name
);
5595 for (i
= bs
; i
!= base
; i
= child_bs(child
)) {
5596 child
= bdrv_filter_or_cow_child(i
);
5598 child
->frozen
= true;
5606 * Unfreeze all COW (backing) and filter links between @bs and @base.
5607 * The caller must ensure that all links are frozen before using this
5609 * @base must be reachable from @bs, or NULL.
5611 void bdrv_unfreeze_backing_chain(BlockDriverState
*bs
, BlockDriverState
*base
)
5613 BlockDriverState
*i
;
5616 GLOBAL_STATE_CODE();
5618 for (i
= bs
; i
!= base
; i
= child_bs(child
)) {
5619 child
= bdrv_filter_or_cow_child(i
);
5621 assert(child
->frozen
);
5622 child
->frozen
= false;
5628 * Drops images above 'base' up to and including 'top', and sets the image
5629 * above 'top' to have base as its backing file.
5631 * Requires that the overlay to 'top' is opened r/w, so that the backing file
5632 * information in 'bs' can be properly updated.
5634 * E.g., this will convert the following chain:
5635 * bottom <- base <- intermediate <- top <- active
5639 * bottom <- base <- active
5641 * It is allowed for bottom==base, in which case it converts:
5643 * base <- intermediate <- top <- active
5649 * If backing_file_str is non-NULL, it will be used when modifying top's
5650 * overlay image metadata.
5653 * if active == top, that is considered an error
5656 int bdrv_drop_intermediate(BlockDriverState
*top
, BlockDriverState
*base
,
5657 const char *backing_file_str
)
5659 BlockDriverState
*explicit_top
= top
;
5660 bool update_inherits_from
;
5662 Error
*local_err
= NULL
;
5664 g_autoptr(GSList
) updated_children
= NULL
;
5667 GLOBAL_STATE_CODE();
5670 bdrv_drained_begin(base
);
5672 if (!top
->drv
|| !base
->drv
) {
5676 /* Make sure that base is in the backing chain of top */
5677 if (!bdrv_chain_contains(top
, base
)) {
5681 /* If 'base' recursively inherits from 'top' then we should set
5682 * base->inherits_from to top->inherits_from after 'top' and all
5683 * other intermediate nodes have been dropped.
5684 * If 'top' is an implicit node (e.g. "commit_top") we should skip
5685 * it because no one inherits from it. We use explicit_top for that. */
5686 explicit_top
= bdrv_skip_implicit_filters(explicit_top
);
5687 update_inherits_from
= bdrv_inherits_from_recursive(base
, explicit_top
);
5689 /* success - we can delete the intermediate states, and link top->base */
5690 if (!backing_file_str
) {
5691 bdrv_refresh_filename(base
);
5692 backing_file_str
= base
->filename
;
5695 QLIST_FOREACH(c
, &top
->parents
, next_parent
) {
5696 updated_children
= g_slist_prepend(updated_children
, c
);
5700 * It seems correct to pass detach_subchain=true here, but it triggers
5701 * one more yet not fixed bug, when due to nested aio_poll loop we switch to
5702 * another drained section, which modify the graph (for example, removing
5703 * the child, which we keep in updated_children list). So, it's a TODO.
5705 * Note, bug triggered if pass detach_subchain=true here and run
5706 * test-bdrv-drain. test_drop_intermediate_poll() test-case will crash.
5709 bdrv_replace_node_common(top
, base
, false, false, &local_err
);
5711 error_report_err(local_err
);
5715 for (p
= updated_children
; p
; p
= p
->next
) {
5718 if (c
->klass
->update_filename
) {
5719 ret
= c
->klass
->update_filename(c
, base
, backing_file_str
,
5723 * TODO: Actually, we want to rollback all previous iterations
5724 * of this loop, and (which is almost impossible) previous
5725 * bdrv_replace_node()...
5727 * Note, that c->klass->update_filename may lead to permission
5728 * update, so it's a bad idea to call it inside permission
5729 * update transaction of bdrv_replace_node.
5731 error_report_err(local_err
);
5737 if (update_inherits_from
) {
5738 base
->inherits_from
= explicit_top
->inherits_from
;
5743 bdrv_drained_end(base
);
5749 * Implementation of BlockDriver.bdrv_co_get_allocated_file_size() that
5750 * sums the size of all data-bearing children. (This excludes backing
5753 static int64_t bdrv_sum_allocated_file_size(BlockDriverState
*bs
)
5756 int64_t child_size
, sum
= 0;
5758 QLIST_FOREACH(child
, &bs
->children
, next
) {
5759 if (child
->role
& (BDRV_CHILD_DATA
| BDRV_CHILD_METADATA
|
5760 BDRV_CHILD_FILTERED
))
5762 child_size
= bdrv_co_get_allocated_file_size(child
->bs
);
5763 if (child_size
< 0) {
5774 * Length of a allocated file in bytes. Sparse files are counted by actual
5775 * allocated space. Return < 0 if error or unknown.
5777 int64_t coroutine_fn
bdrv_co_get_allocated_file_size(BlockDriverState
*bs
)
5779 BlockDriver
*drv
= bs
->drv
;
5785 if (drv
->bdrv_co_get_allocated_file_size
) {
5786 return drv
->bdrv_co_get_allocated_file_size(bs
);
5789 if (drv
->bdrv_file_open
) {
5791 * Protocol drivers default to -ENOTSUP (most of their data is
5792 * not stored in any of their children (if they even have any),
5793 * so there is no generic way to figure it out).
5796 } else if (drv
->is_filter
) {
5797 /* Filter drivers default to the size of their filtered child */
5798 return bdrv_co_get_allocated_file_size(bdrv_filter_bs(bs
));
5800 /* Other drivers default to summing their children's sizes */
5801 return bdrv_sum_allocated_file_size(bs
);
5807 * @drv: Format driver
5808 * @opts: Creation options for new image
5809 * @in_bs: Existing image containing data for new image (may be NULL)
5810 * @errp: Error object
5811 * Returns: A #BlockMeasureInfo (free using qapi_free_BlockMeasureInfo())
5814 * Calculate file size required to create a new image.
5816 * If @in_bs is given then space for allocated clusters and zero clusters
5817 * from that image are included in the calculation. If @opts contains a
5818 * backing file that is shared by @in_bs then backing clusters may be omitted
5819 * from the calculation.
5821 * If @in_bs is NULL then the calculation includes no allocated clusters
5822 * unless a preallocation option is given in @opts.
5824 * Note that @in_bs may use a different BlockDriver from @drv.
5826 * If an error occurs the @errp pointer is set.
5828 BlockMeasureInfo
*bdrv_measure(BlockDriver
*drv
, QemuOpts
*opts
,
5829 BlockDriverState
*in_bs
, Error
**errp
)
5832 if (!drv
->bdrv_measure
) {
5833 error_setg(errp
, "Block driver '%s' does not support size measurement",
5838 return drv
->bdrv_measure(opts
, in_bs
, errp
);
5842 * Return number of sectors on success, -errno on error.
5844 int64_t coroutine_fn
bdrv_co_nb_sectors(BlockDriverState
*bs
)
5846 BlockDriver
*drv
= bs
->drv
;
5848 assert_bdrv_graph_readable();
5853 if (drv
->has_variable_length
) {
5854 int ret
= bdrv_co_refresh_total_sectors(bs
, bs
->total_sectors
);
5859 return bs
->total_sectors
;
5863 * Return length in bytes on success, -errno on error.
5864 * The length is always a multiple of BDRV_SECTOR_SIZE.
5866 int64_t coroutine_fn
bdrv_co_getlength(BlockDriverState
*bs
)
5870 assert_bdrv_graph_readable();
5872 ret
= bdrv_co_nb_sectors(bs
);
5876 if (ret
> INT64_MAX
/ BDRV_SECTOR_SIZE
) {
5879 return ret
* BDRV_SECTOR_SIZE
;
5882 /* return 0 as number of sectors if no device present or error */
5883 void coroutine_fn
bdrv_co_get_geometry(BlockDriverState
*bs
,
5884 uint64_t *nb_sectors_ptr
)
5886 int64_t nb_sectors
= bdrv_co_nb_sectors(bs
);
5889 *nb_sectors_ptr
= nb_sectors
< 0 ? 0 : nb_sectors
;
5892 bool bdrv_is_sg(BlockDriverState
*bs
)
5899 * Return whether the given node supports compressed writes.
5901 bool bdrv_supports_compressed_writes(BlockDriverState
*bs
)
5903 BlockDriverState
*filtered
;
5906 if (!bs
->drv
|| !block_driver_can_compress(bs
->drv
)) {
5910 filtered
= bdrv_filter_bs(bs
);
5913 * Filters can only forward compressed writes, so we have to
5916 return bdrv_supports_compressed_writes(filtered
);
5922 const char *bdrv_get_format_name(BlockDriverState
*bs
)
5925 return bs
->drv
? bs
->drv
->format_name
: NULL
;
5928 static int qsort_strcmp(const void *a
, const void *b
)
5930 return strcmp(*(char *const *)a
, *(char *const *)b
);
5933 void bdrv_iterate_format(void (*it
)(void *opaque
, const char *name
),
5934 void *opaque
, bool read_only
)
5939 const char **formats
= NULL
;
5941 GLOBAL_STATE_CODE();
5943 QLIST_FOREACH(drv
, &bdrv_drivers
, list
) {
5944 if (drv
->format_name
) {
5948 if (use_bdrv_whitelist
&& !bdrv_is_whitelisted(drv
, read_only
)) {
5952 while (formats
&& i
&& !found
) {
5953 found
= !strcmp(formats
[--i
], drv
->format_name
);
5957 formats
= g_renew(const char *, formats
, count
+ 1);
5958 formats
[count
++] = drv
->format_name
;
5963 for (i
= 0; i
< (int)ARRAY_SIZE(block_driver_modules
); i
++) {
5964 const char *format_name
= block_driver_modules
[i
].format_name
;
5970 if (use_bdrv_whitelist
&&
5971 !bdrv_format_is_whitelisted(format_name
, read_only
)) {
5975 while (formats
&& j
&& !found
) {
5976 found
= !strcmp(formats
[--j
], format_name
);
5980 formats
= g_renew(const char *, formats
, count
+ 1);
5981 formats
[count
++] = format_name
;
5986 qsort(formats
, count
, sizeof(formats
[0]), qsort_strcmp
);
5988 for (i
= 0; i
< count
; i
++) {
5989 it(opaque
, formats
[i
]);
5995 /* This function is to find a node in the bs graph */
5996 BlockDriverState
*bdrv_find_node(const char *node_name
)
5998 BlockDriverState
*bs
;
6001 GLOBAL_STATE_CODE();
6003 QTAILQ_FOREACH(bs
, &graph_bdrv_states
, node_list
) {
6004 if (!strcmp(node_name
, bs
->node_name
)) {
6011 /* Put this QMP function here so it can access the static graph_bdrv_states. */
6012 BlockDeviceInfoList
*bdrv_named_nodes_list(bool flat
,
6015 BlockDeviceInfoList
*list
;
6016 BlockDriverState
*bs
;
6018 GLOBAL_STATE_CODE();
6021 QTAILQ_FOREACH(bs
, &graph_bdrv_states
, node_list
) {
6022 BlockDeviceInfo
*info
= bdrv_block_device_info(NULL
, bs
, flat
, errp
);
6024 qapi_free_BlockDeviceInfoList(list
);
6027 QAPI_LIST_PREPEND(list
, info
);
6033 typedef struct XDbgBlockGraphConstructor
{
6034 XDbgBlockGraph
*graph
;
6035 GHashTable
*graph_nodes
;
6036 } XDbgBlockGraphConstructor
;
6038 static XDbgBlockGraphConstructor
*xdbg_graph_new(void)
6040 XDbgBlockGraphConstructor
*gr
= g_new(XDbgBlockGraphConstructor
, 1);
6042 gr
->graph
= g_new0(XDbgBlockGraph
, 1);
6043 gr
->graph_nodes
= g_hash_table_new(NULL
, NULL
);
6048 static XDbgBlockGraph
*xdbg_graph_finalize(XDbgBlockGraphConstructor
*gr
)
6050 XDbgBlockGraph
*graph
= gr
->graph
;
6052 g_hash_table_destroy(gr
->graph_nodes
);
6058 static uintptr_t xdbg_graph_node_num(XDbgBlockGraphConstructor
*gr
, void *node
)
6060 uintptr_t ret
= (uintptr_t)g_hash_table_lookup(gr
->graph_nodes
, node
);
6067 * Start counting from 1, not 0, because 0 interferes with not-found (NULL)
6068 * answer of g_hash_table_lookup.
6070 ret
= g_hash_table_size(gr
->graph_nodes
) + 1;
6071 g_hash_table_insert(gr
->graph_nodes
, node
, (void *)ret
);
6076 static void xdbg_graph_add_node(XDbgBlockGraphConstructor
*gr
, void *node
,
6077 XDbgBlockGraphNodeType type
, const char *name
)
6079 XDbgBlockGraphNode
*n
;
6081 n
= g_new0(XDbgBlockGraphNode
, 1);
6083 n
->id
= xdbg_graph_node_num(gr
, node
);
6085 n
->name
= g_strdup(name
);
6087 QAPI_LIST_PREPEND(gr
->graph
->nodes
, n
);
6090 static void xdbg_graph_add_edge(XDbgBlockGraphConstructor
*gr
, void *parent
,
6091 const BdrvChild
*child
)
6093 BlockPermission qapi_perm
;
6094 XDbgBlockGraphEdge
*edge
;
6095 GLOBAL_STATE_CODE();
6097 edge
= g_new0(XDbgBlockGraphEdge
, 1);
6099 edge
->parent
= xdbg_graph_node_num(gr
, parent
);
6100 edge
->child
= xdbg_graph_node_num(gr
, child
->bs
);
6101 edge
->name
= g_strdup(child
->name
);
6103 for (qapi_perm
= 0; qapi_perm
< BLOCK_PERMISSION__MAX
; qapi_perm
++) {
6104 uint64_t flag
= bdrv_qapi_perm_to_blk_perm(qapi_perm
);
6106 if (flag
& child
->perm
) {
6107 QAPI_LIST_PREPEND(edge
->perm
, qapi_perm
);
6109 if (flag
& child
->shared_perm
) {
6110 QAPI_LIST_PREPEND(edge
->shared_perm
, qapi_perm
);
6114 QAPI_LIST_PREPEND(gr
->graph
->edges
, edge
);
6118 XDbgBlockGraph
*bdrv_get_xdbg_block_graph(Error
**errp
)
6122 BlockDriverState
*bs
;
6124 XDbgBlockGraphConstructor
*gr
= xdbg_graph_new();
6126 GLOBAL_STATE_CODE();
6128 for (blk
= blk_all_next(NULL
); blk
; blk
= blk_all_next(blk
)) {
6129 char *allocated_name
= NULL
;
6130 const char *name
= blk_name(blk
);
6133 name
= allocated_name
= blk_get_attached_dev_id(blk
);
6135 xdbg_graph_add_node(gr
, blk
, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_BACKEND
,
6137 g_free(allocated_name
);
6138 if (blk_root(blk
)) {
6139 xdbg_graph_add_edge(gr
, blk
, blk_root(blk
));
6143 WITH_JOB_LOCK_GUARD() {
6144 for (job
= block_job_next_locked(NULL
); job
;
6145 job
= block_job_next_locked(job
)) {
6148 xdbg_graph_add_node(gr
, job
, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_JOB
,
6150 for (el
= job
->nodes
; el
; el
= el
->next
) {
6151 xdbg_graph_add_edge(gr
, job
, (BdrvChild
*)el
->data
);
6156 QTAILQ_FOREACH(bs
, &graph_bdrv_states
, node_list
) {
6157 xdbg_graph_add_node(gr
, bs
, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_DRIVER
,
6159 QLIST_FOREACH(child
, &bs
->children
, next
) {
6160 xdbg_graph_add_edge(gr
, bs
, child
);
6164 return xdbg_graph_finalize(gr
);
6167 BlockDriverState
*bdrv_lookup_bs(const char *device
,
6168 const char *node_name
,
6172 BlockDriverState
*bs
;
6174 GLOBAL_STATE_CODE();
6177 blk
= blk_by_name(device
);
6182 error_setg(errp
, "Device '%s' has no medium", device
);
6190 bs
= bdrv_find_node(node_name
);
6197 error_setg(errp
, "Cannot find device=\'%s\' nor node-name=\'%s\'",
6198 device
? device
: "",
6199 node_name
? node_name
: "");
6203 /* If 'base' is in the same chain as 'top', return true. Otherwise,
6204 * return false. If either argument is NULL, return false. */
6205 bool bdrv_chain_contains(BlockDriverState
*top
, BlockDriverState
*base
)
6208 GLOBAL_STATE_CODE();
6210 while (top
&& top
!= base
) {
6211 top
= bdrv_filter_or_cow_bs(top
);
6217 BlockDriverState
*bdrv_next_node(BlockDriverState
*bs
)
6219 GLOBAL_STATE_CODE();
6221 return QTAILQ_FIRST(&graph_bdrv_states
);
6223 return QTAILQ_NEXT(bs
, node_list
);
6226 BlockDriverState
*bdrv_next_all_states(BlockDriverState
*bs
)
6228 GLOBAL_STATE_CODE();
6230 return QTAILQ_FIRST(&all_bdrv_states
);
6232 return QTAILQ_NEXT(bs
, bs_list
);
6235 const char *bdrv_get_node_name(const BlockDriverState
*bs
)
6238 return bs
->node_name
;
6241 const char *bdrv_get_parent_name(const BlockDriverState
*bs
)
6247 /* If multiple parents have a name, just pick the first one. */
6248 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
6249 if (c
->klass
->get_name
) {
6250 name
= c
->klass
->get_name(c
);
6251 if (name
&& *name
) {
6260 /* TODO check what callers really want: bs->node_name or blk_name() */
6261 const char *bdrv_get_device_name(const BlockDriverState
*bs
)
6264 return bdrv_get_parent_name(bs
) ?: "";
6267 /* This can be used to identify nodes that might not have a device
6268 * name associated. Since node and device names live in the same
6269 * namespace, the result is unambiguous. The exception is if both are
6270 * absent, then this returns an empty (non-null) string. */
6271 const char *bdrv_get_device_or_node_name(const BlockDriverState
*bs
)
6274 return bdrv_get_parent_name(bs
) ?: bs
->node_name
;
6277 int bdrv_get_flags(BlockDriverState
*bs
)
6280 return bs
->open_flags
;
6283 int bdrv_has_zero_init_1(BlockDriverState
*bs
)
6285 GLOBAL_STATE_CODE();
6289 int bdrv_has_zero_init(BlockDriverState
*bs
)
6291 BlockDriverState
*filtered
;
6292 GLOBAL_STATE_CODE();
6298 /* If BS is a copy on write image, it is initialized to
6299 the contents of the base image, which may not be zeroes. */
6300 if (bdrv_cow_child(bs
)) {
6303 if (bs
->drv
->bdrv_has_zero_init
) {
6304 return bs
->drv
->bdrv_has_zero_init(bs
);
6307 filtered
= bdrv_filter_bs(bs
);
6309 return bdrv_has_zero_init(filtered
);
6316 bool bdrv_can_write_zeroes_with_unmap(BlockDriverState
*bs
)
6319 if (!(bs
->open_flags
& BDRV_O_UNMAP
)) {
6323 return bs
->supported_zero_flags
& BDRV_REQ_MAY_UNMAP
;
6326 void bdrv_get_backing_filename(BlockDriverState
*bs
,
6327 char *filename
, int filename_size
)
6330 pstrcpy(filename
, filename_size
, bs
->backing_file
);
6333 int coroutine_fn
bdrv_co_get_info(BlockDriverState
*bs
, BlockDriverInfo
*bdi
)
6336 BlockDriver
*drv
= bs
->drv
;
6338 /* if bs->drv == NULL, bs is closed, so there's nothing to do here */
6342 if (!drv
->bdrv_co_get_info
) {
6343 BlockDriverState
*filtered
= bdrv_filter_bs(bs
);
6345 return bdrv_co_get_info(filtered
, bdi
);
6349 memset(bdi
, 0, sizeof(*bdi
));
6350 ret
= drv
->bdrv_co_get_info(bs
, bdi
);
6355 if (bdi
->cluster_size
> BDRV_MAX_ALIGNMENT
) {
6362 ImageInfoSpecific
*bdrv_get_specific_info(BlockDriverState
*bs
,
6365 BlockDriver
*drv
= bs
->drv
;
6367 if (drv
&& drv
->bdrv_get_specific_info
) {
6368 return drv
->bdrv_get_specific_info(bs
, errp
);
6373 BlockStatsSpecific
*bdrv_get_specific_stats(BlockDriverState
*bs
)
6375 BlockDriver
*drv
= bs
->drv
;
6377 if (!drv
|| !drv
->bdrv_get_specific_stats
) {
6380 return drv
->bdrv_get_specific_stats(bs
);
6383 void coroutine_fn
bdrv_co_debug_event(BlockDriverState
*bs
, BlkdebugEvent event
)
6386 if (!bs
|| !bs
->drv
|| !bs
->drv
->bdrv_co_debug_event
) {
6390 bs
->drv
->bdrv_co_debug_event(bs
, event
);
6393 static BlockDriverState
*bdrv_find_debug_node(BlockDriverState
*bs
)
6395 GLOBAL_STATE_CODE();
6396 while (bs
&& bs
->drv
&& !bs
->drv
->bdrv_debug_breakpoint
) {
6397 bs
= bdrv_primary_bs(bs
);
6400 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_breakpoint
) {
6401 assert(bs
->drv
->bdrv_debug_remove_breakpoint
);
6408 int bdrv_debug_breakpoint(BlockDriverState
*bs
, const char *event
,
6411 GLOBAL_STATE_CODE();
6412 bs
= bdrv_find_debug_node(bs
);
6414 return bs
->drv
->bdrv_debug_breakpoint(bs
, event
, tag
);
6420 int bdrv_debug_remove_breakpoint(BlockDriverState
*bs
, const char *tag
)
6422 GLOBAL_STATE_CODE();
6423 bs
= bdrv_find_debug_node(bs
);
6425 return bs
->drv
->bdrv_debug_remove_breakpoint(bs
, tag
);
6431 int bdrv_debug_resume(BlockDriverState
*bs
, const char *tag
)
6433 GLOBAL_STATE_CODE();
6434 while (bs
&& (!bs
->drv
|| !bs
->drv
->bdrv_debug_resume
)) {
6435 bs
= bdrv_primary_bs(bs
);
6438 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_resume
) {
6439 return bs
->drv
->bdrv_debug_resume(bs
, tag
);
6445 bool bdrv_debug_is_suspended(BlockDriverState
*bs
, const char *tag
)
6447 GLOBAL_STATE_CODE();
6448 while (bs
&& bs
->drv
&& !bs
->drv
->bdrv_debug_is_suspended
) {
6449 bs
= bdrv_primary_bs(bs
);
6452 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_is_suspended
) {
6453 return bs
->drv
->bdrv_debug_is_suspended(bs
, tag
);
6459 /* backing_file can either be relative, or absolute, or a protocol. If it is
6460 * relative, it must be relative to the chain. So, passing in bs->filename
6461 * from a BDS as backing_file should not be done, as that may be relative to
6462 * the CWD rather than the chain. */
6463 BlockDriverState
*bdrv_find_backing_image(BlockDriverState
*bs
,
6464 const char *backing_file
)
6466 char *filename_full
= NULL
;
6467 char *backing_file_full
= NULL
;
6468 char *filename_tmp
= NULL
;
6469 int is_protocol
= 0;
6470 bool filenames_refreshed
= false;
6471 BlockDriverState
*curr_bs
= NULL
;
6472 BlockDriverState
*retval
= NULL
;
6473 BlockDriverState
*bs_below
;
6475 GLOBAL_STATE_CODE();
6477 if (!bs
|| !bs
->drv
|| !backing_file
) {
6481 filename_full
= g_malloc(PATH_MAX
);
6482 backing_file_full
= g_malloc(PATH_MAX
);
6484 is_protocol
= path_has_protocol(backing_file
);
6487 * Being largely a legacy function, skip any filters here
6488 * (because filters do not have normal filenames, so they cannot
6489 * match anyway; and allowing json:{} filenames is a bit out of
6492 for (curr_bs
= bdrv_skip_filters(bs
);
6493 bdrv_cow_child(curr_bs
) != NULL
;
6496 bs_below
= bdrv_backing_chain_next(curr_bs
);
6498 if (bdrv_backing_overridden(curr_bs
)) {
6500 * If the backing file was overridden, we can only compare
6501 * directly against the backing node's filename.
6504 if (!filenames_refreshed
) {
6506 * This will automatically refresh all of the
6507 * filenames in the rest of the backing chain, so we
6508 * only need to do this once.
6510 bdrv_refresh_filename(bs_below
);
6511 filenames_refreshed
= true;
6514 if (strcmp(backing_file
, bs_below
->filename
) == 0) {
6518 } else if (is_protocol
|| path_has_protocol(curr_bs
->backing_file
)) {
6520 * If either of the filename paths is actually a protocol, then
6521 * compare unmodified paths; otherwise make paths relative.
6523 char *backing_file_full_ret
;
6525 if (strcmp(backing_file
, curr_bs
->backing_file
) == 0) {
6529 /* Also check against the full backing filename for the image */
6530 backing_file_full_ret
= bdrv_get_full_backing_filename(curr_bs
,
6532 if (backing_file_full_ret
) {
6533 bool equal
= strcmp(backing_file
, backing_file_full_ret
) == 0;
6534 g_free(backing_file_full_ret
);
6541 /* If not an absolute filename path, make it relative to the current
6542 * image's filename path */
6543 filename_tmp
= bdrv_make_absolute_filename(curr_bs
, backing_file
,
6545 /* We are going to compare canonicalized absolute pathnames */
6546 if (!filename_tmp
|| !realpath(filename_tmp
, filename_full
)) {
6547 g_free(filename_tmp
);
6550 g_free(filename_tmp
);
6552 /* We need to make sure the backing filename we are comparing against
6553 * is relative to the current image filename (or absolute) */
6554 filename_tmp
= bdrv_get_full_backing_filename(curr_bs
, NULL
);
6555 if (!filename_tmp
|| !realpath(filename_tmp
, backing_file_full
)) {
6556 g_free(filename_tmp
);
6559 g_free(filename_tmp
);
6561 if (strcmp(backing_file_full
, filename_full
) == 0) {
6568 g_free(filename_full
);
6569 g_free(backing_file_full
);
6573 void bdrv_init(void)
6575 #ifdef CONFIG_BDRV_WHITELIST_TOOLS
6576 use_bdrv_whitelist
= 1;
6578 module_call_init(MODULE_INIT_BLOCK
);
6581 void bdrv_init_with_whitelist(void)
6583 use_bdrv_whitelist
= 1;
6587 int bdrv_activate(BlockDriverState
*bs
, Error
**errp
)
6589 BdrvChild
*child
, *parent
;
6590 Error
*local_err
= NULL
;
6592 BdrvDirtyBitmap
*bm
;
6594 GLOBAL_STATE_CODE();
6600 QLIST_FOREACH(child
, &bs
->children
, next
) {
6601 bdrv_activate(child
->bs
, &local_err
);
6603 error_propagate(errp
, local_err
);
6609 * Update permissions, they may differ for inactive nodes.
6611 * Note that the required permissions of inactive images are always a
6612 * subset of the permissions required after activating the image. This
6613 * allows us to just get the permissions upfront without restricting
6614 * bdrv_co_invalidate_cache().
6616 * It also means that in error cases, we don't have to try and revert to
6617 * the old permissions (which is an operation that could fail, too). We can
6618 * just keep the extended permissions for the next time that an activation
6619 * of the image is tried.
6621 if (bs
->open_flags
& BDRV_O_INACTIVE
) {
6622 bs
->open_flags
&= ~BDRV_O_INACTIVE
;
6623 ret
= bdrv_refresh_perms(bs
, NULL
, errp
);
6625 bs
->open_flags
|= BDRV_O_INACTIVE
;
6629 ret
= bdrv_invalidate_cache(bs
, errp
);
6631 bs
->open_flags
|= BDRV_O_INACTIVE
;
6635 FOR_EACH_DIRTY_BITMAP(bs
, bm
) {
6636 bdrv_dirty_bitmap_skip_store(bm
, false);
6639 ret
= bdrv_refresh_total_sectors(bs
, bs
->total_sectors
);
6641 bs
->open_flags
|= BDRV_O_INACTIVE
;
6642 error_setg_errno(errp
, -ret
, "Could not refresh total sector count");
6647 QLIST_FOREACH(parent
, &bs
->parents
, next_parent
) {
6648 if (parent
->klass
->activate
) {
6649 parent
->klass
->activate(parent
, &local_err
);
6651 bs
->open_flags
|= BDRV_O_INACTIVE
;
6652 error_propagate(errp
, local_err
);
6661 int coroutine_fn
bdrv_co_invalidate_cache(BlockDriverState
*bs
, Error
**errp
)
6663 Error
*local_err
= NULL
;
6666 assert(!(bs
->open_flags
& BDRV_O_INACTIVE
));
6667 assert_bdrv_graph_readable();
6669 if (bs
->drv
->bdrv_co_invalidate_cache
) {
6670 bs
->drv
->bdrv_co_invalidate_cache(bs
, &local_err
);
6672 error_propagate(errp
, local_err
);
6680 void bdrv_activate_all(Error
**errp
)
6682 BlockDriverState
*bs
;
6683 BdrvNextIterator it
;
6685 GLOBAL_STATE_CODE();
6687 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
6688 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
6691 aio_context_acquire(aio_context
);
6692 ret
= bdrv_activate(bs
, errp
);
6693 aio_context_release(aio_context
);
6695 bdrv_next_cleanup(&it
);
6701 static bool bdrv_has_bds_parent(BlockDriverState
*bs
, bool only_active
)
6704 GLOBAL_STATE_CODE();
6706 QLIST_FOREACH(parent
, &bs
->parents
, next_parent
) {
6707 if (parent
->klass
->parent_is_bds
) {
6708 BlockDriverState
*parent_bs
= parent
->opaque
;
6709 if (!only_active
|| !(parent_bs
->open_flags
& BDRV_O_INACTIVE
)) {
6718 static int bdrv_inactivate_recurse(BlockDriverState
*bs
)
6720 BdrvChild
*child
, *parent
;
6722 uint64_t cumulative_perms
, cumulative_shared_perms
;
6724 GLOBAL_STATE_CODE();
6730 /* Make sure that we don't inactivate a child before its parent.
6731 * It will be covered by recursion from the yet active parent. */
6732 if (bdrv_has_bds_parent(bs
, true)) {
6736 assert(!(bs
->open_flags
& BDRV_O_INACTIVE
));
6738 /* Inactivate this node */
6739 if (bs
->drv
->bdrv_inactivate
) {
6740 ret
= bs
->drv
->bdrv_inactivate(bs
);
6746 QLIST_FOREACH(parent
, &bs
->parents
, next_parent
) {
6747 if (parent
->klass
->inactivate
) {
6748 ret
= parent
->klass
->inactivate(parent
);
6755 bdrv_get_cumulative_perm(bs
, &cumulative_perms
,
6756 &cumulative_shared_perms
);
6757 if (cumulative_perms
& (BLK_PERM_WRITE
| BLK_PERM_WRITE_UNCHANGED
)) {
6758 /* Our inactive parents still need write access. Inactivation failed. */
6762 bs
->open_flags
|= BDRV_O_INACTIVE
;
6765 * Update permissions, they may differ for inactive nodes.
6766 * We only tried to loosen restrictions, so errors are not fatal, ignore
6769 bdrv_refresh_perms(bs
, NULL
, NULL
);
6771 /* Recursively inactivate children */
6772 QLIST_FOREACH(child
, &bs
->children
, next
) {
6773 ret
= bdrv_inactivate_recurse(child
->bs
);
6782 int bdrv_inactivate_all(void)
6784 BlockDriverState
*bs
= NULL
;
6785 BdrvNextIterator it
;
6787 GSList
*aio_ctxs
= NULL
, *ctx
;
6789 GLOBAL_STATE_CODE();
6791 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
6792 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
6794 if (!g_slist_find(aio_ctxs
, aio_context
)) {
6795 aio_ctxs
= g_slist_prepend(aio_ctxs
, aio_context
);
6796 aio_context_acquire(aio_context
);
6800 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
6801 /* Nodes with BDS parents are covered by recursion from the last
6802 * parent that gets inactivated. Don't inactivate them a second
6803 * time if that has already happened. */
6804 if (bdrv_has_bds_parent(bs
, false)) {
6807 ret
= bdrv_inactivate_recurse(bs
);
6809 bdrv_next_cleanup(&it
);
6815 for (ctx
= aio_ctxs
; ctx
!= NULL
; ctx
= ctx
->next
) {
6816 AioContext
*aio_context
= ctx
->data
;
6817 aio_context_release(aio_context
);
6819 g_slist_free(aio_ctxs
);
6824 /**************************************************************/
6825 /* removable device support */
6828 * Return TRUE if the media is present
6830 bool coroutine_fn
bdrv_co_is_inserted(BlockDriverState
*bs
)
6832 BlockDriver
*drv
= bs
->drv
;
6835 assert_bdrv_graph_readable();
6840 if (drv
->bdrv_co_is_inserted
) {
6841 return drv
->bdrv_co_is_inserted(bs
);
6843 QLIST_FOREACH(child
, &bs
->children
, next
) {
6844 if (!bdrv_co_is_inserted(child
->bs
)) {
6852 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
6854 void coroutine_fn
bdrv_co_eject(BlockDriverState
*bs
, bool eject_flag
)
6856 BlockDriver
*drv
= bs
->drv
;
6858 assert_bdrv_graph_readable();
6860 if (drv
&& drv
->bdrv_co_eject
) {
6861 drv
->bdrv_co_eject(bs
, eject_flag
);
6866 * Lock or unlock the media (if it is locked, the user won't be able
6867 * to eject it manually).
6869 void coroutine_fn
bdrv_co_lock_medium(BlockDriverState
*bs
, bool locked
)
6871 BlockDriver
*drv
= bs
->drv
;
6873 assert_bdrv_graph_readable();
6874 trace_bdrv_lock_medium(bs
, locked
);
6876 if (drv
&& drv
->bdrv_co_lock_medium
) {
6877 drv
->bdrv_co_lock_medium(bs
, locked
);
6881 /* Get a reference to bs */
6882 void bdrv_ref(BlockDriverState
*bs
)
6884 GLOBAL_STATE_CODE();
6888 /* Release a previously grabbed reference to bs.
6889 * If after releasing, reference count is zero, the BlockDriverState is
6891 void bdrv_unref(BlockDriverState
*bs
)
6893 GLOBAL_STATE_CODE();
6897 assert(bs
->refcnt
> 0);
6898 if (--bs
->refcnt
== 0) {
6903 struct BdrvOpBlocker
{
6905 QLIST_ENTRY(BdrvOpBlocker
) list
;
6908 bool bdrv_op_is_blocked(BlockDriverState
*bs
, BlockOpType op
, Error
**errp
)
6910 BdrvOpBlocker
*blocker
;
6911 GLOBAL_STATE_CODE();
6912 assert((int) op
>= 0 && op
< BLOCK_OP_TYPE_MAX
);
6913 if (!QLIST_EMPTY(&bs
->op_blockers
[op
])) {
6914 blocker
= QLIST_FIRST(&bs
->op_blockers
[op
]);
6915 error_propagate_prepend(errp
, error_copy(blocker
->reason
),
6916 "Node '%s' is busy: ",
6917 bdrv_get_device_or_node_name(bs
));
6923 void bdrv_op_block(BlockDriverState
*bs
, BlockOpType op
, Error
*reason
)
6925 BdrvOpBlocker
*blocker
;
6926 GLOBAL_STATE_CODE();
6927 assert((int) op
>= 0 && op
< BLOCK_OP_TYPE_MAX
);
6929 blocker
= g_new0(BdrvOpBlocker
, 1);
6930 blocker
->reason
= reason
;
6931 QLIST_INSERT_HEAD(&bs
->op_blockers
[op
], blocker
, list
);
6934 void bdrv_op_unblock(BlockDriverState
*bs
, BlockOpType op
, Error
*reason
)
6936 BdrvOpBlocker
*blocker
, *next
;
6937 GLOBAL_STATE_CODE();
6938 assert((int) op
>= 0 && op
< BLOCK_OP_TYPE_MAX
);
6939 QLIST_FOREACH_SAFE(blocker
, &bs
->op_blockers
[op
], list
, next
) {
6940 if (blocker
->reason
== reason
) {
6941 QLIST_REMOVE(blocker
, list
);
6947 void bdrv_op_block_all(BlockDriverState
*bs
, Error
*reason
)
6950 GLOBAL_STATE_CODE();
6951 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
6952 bdrv_op_block(bs
, i
, reason
);
6956 void bdrv_op_unblock_all(BlockDriverState
*bs
, Error
*reason
)
6959 GLOBAL_STATE_CODE();
6960 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
6961 bdrv_op_unblock(bs
, i
, reason
);
6965 bool bdrv_op_blocker_is_empty(BlockDriverState
*bs
)
6968 GLOBAL_STATE_CODE();
6969 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
6970 if (!QLIST_EMPTY(&bs
->op_blockers
[i
])) {
6978 * Must not be called while holding the lock of an AioContext other than the
6981 void bdrv_img_create(const char *filename
, const char *fmt
,
6982 const char *base_filename
, const char *base_fmt
,
6983 char *options
, uint64_t img_size
, int flags
, bool quiet
,
6986 QemuOptsList
*create_opts
= NULL
;
6987 QemuOpts
*opts
= NULL
;
6988 const char *backing_fmt
, *backing_file
;
6990 BlockDriver
*drv
, *proto_drv
;
6991 Error
*local_err
= NULL
;
6994 GLOBAL_STATE_CODE();
6996 /* Find driver and parse its options */
6997 drv
= bdrv_find_format(fmt
);
6999 error_setg(errp
, "Unknown file format '%s'", fmt
);
7003 proto_drv
= bdrv_find_protocol(filename
, true, errp
);
7008 if (!drv
->create_opts
) {
7009 error_setg(errp
, "Format driver '%s' does not support image creation",
7014 if (!proto_drv
->create_opts
) {
7015 error_setg(errp
, "Protocol driver '%s' does not support image creation",
7016 proto_drv
->format_name
);
7020 /* Create parameter list */
7021 create_opts
= qemu_opts_append(create_opts
, drv
->create_opts
);
7022 create_opts
= qemu_opts_append(create_opts
, proto_drv
->create_opts
);
7024 opts
= qemu_opts_create(create_opts
, NULL
, 0, &error_abort
);
7026 /* Parse -o options */
7028 if (!qemu_opts_do_parse(opts
, options
, NULL
, errp
)) {
7033 if (!qemu_opt_get(opts
, BLOCK_OPT_SIZE
)) {
7034 qemu_opt_set_number(opts
, BLOCK_OPT_SIZE
, img_size
, &error_abort
);
7035 } else if (img_size
!= UINT64_C(-1)) {
7036 error_setg(errp
, "The image size must be specified only once");
7040 if (base_filename
) {
7041 if (!qemu_opt_set(opts
, BLOCK_OPT_BACKING_FILE
, base_filename
,
7043 error_setg(errp
, "Backing file not supported for file format '%s'",
7050 if (!qemu_opt_set(opts
, BLOCK_OPT_BACKING_FMT
, base_fmt
, NULL
)) {
7051 error_setg(errp
, "Backing file format not supported for file "
7052 "format '%s'", fmt
);
7057 backing_file
= qemu_opt_get(opts
, BLOCK_OPT_BACKING_FILE
);
7059 if (!strcmp(filename
, backing_file
)) {
7060 error_setg(errp
, "Error: Trying to create an image with the "
7061 "same filename as the backing file");
7064 if (backing_file
[0] == '\0') {
7065 error_setg(errp
, "Expected backing file name, got empty string");
7070 backing_fmt
= qemu_opt_get(opts
, BLOCK_OPT_BACKING_FMT
);
7072 /* The size for the image must always be specified, unless we have a backing
7073 * file and we have not been forbidden from opening it. */
7074 size
= qemu_opt_get_size(opts
, BLOCK_OPT_SIZE
, img_size
);
7075 if (backing_file
&& !(flags
& BDRV_O_NO_BACKING
)) {
7076 BlockDriverState
*bs
;
7079 QDict
*backing_options
= NULL
;
7082 bdrv_get_full_backing_filename_from_filename(filename
, backing_file
,
7087 assert(full_backing
);
7090 * No need to do I/O here, which allows us to open encrypted
7091 * backing images without needing the secret
7094 back_flags
&= ~(BDRV_O_RDWR
| BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
);
7095 back_flags
|= BDRV_O_NO_IO
;
7097 backing_options
= qdict_new();
7099 qdict_put_str(backing_options
, "driver", backing_fmt
);
7101 qdict_put_bool(backing_options
, BDRV_OPT_FORCE_SHARE
, true);
7103 bs
= bdrv_open(full_backing
, NULL
, backing_options
, back_flags
,
7105 g_free(full_backing
);
7107 error_append_hint(&local_err
, "Could not open backing image.\n");
7111 error_setg(&local_err
,
7112 "Backing file specified without backing format");
7113 error_append_hint(&local_err
, "Detected format of %s.",
7114 bs
->drv
->format_name
);
7118 /* Opened BS, have no size */
7119 size
= bdrv_getlength(bs
);
7121 error_setg_errno(errp
, -size
, "Could not get size of '%s'",
7126 qemu_opt_set_number(opts
, BLOCK_OPT_SIZE
, size
, &error_abort
);
7130 /* (backing_file && !(flags & BDRV_O_NO_BACKING)) */
7131 } else if (backing_file
&& !backing_fmt
) {
7132 error_setg(&local_err
,
7133 "Backing file specified without backing format");
7138 error_setg(errp
, "Image creation needs a size parameter");
7143 printf("Formatting '%s', fmt=%s ", filename
, fmt
);
7144 qemu_opts_print(opts
, " ");
7149 ret
= bdrv_create(drv
, filename
, opts
, &local_err
);
7151 if (ret
== -EFBIG
) {
7152 /* This is generally a better message than whatever the driver would
7153 * deliver (especially because of the cluster_size_hint), since that
7154 * is most probably not much different from "image too large". */
7155 const char *cluster_size_hint
= "";
7156 if (qemu_opt_get_size(opts
, BLOCK_OPT_CLUSTER_SIZE
, 0)) {
7157 cluster_size_hint
= " (try using a larger cluster size)";
7159 error_setg(errp
, "The image size is too large for file format '%s'"
7160 "%s", fmt
, cluster_size_hint
);
7161 error_free(local_err
);
7166 qemu_opts_del(opts
);
7167 qemu_opts_free(create_opts
);
7168 error_propagate(errp
, local_err
);
7171 AioContext
*bdrv_get_aio_context(BlockDriverState
*bs
)
7174 return bs
? bs
->aio_context
: qemu_get_aio_context();
7177 AioContext
*coroutine_fn
bdrv_co_enter(BlockDriverState
*bs
)
7179 Coroutine
*self
= qemu_coroutine_self();
7180 AioContext
*old_ctx
= qemu_coroutine_get_aio_context(self
);
7181 AioContext
*new_ctx
;
7185 * Increase bs->in_flight to ensure that this operation is completed before
7186 * moving the node to a different AioContext. Read new_ctx only afterwards.
7188 bdrv_inc_in_flight(bs
);
7190 new_ctx
= bdrv_get_aio_context(bs
);
7191 aio_co_reschedule_self(new_ctx
);
7195 void coroutine_fn
bdrv_co_leave(BlockDriverState
*bs
, AioContext
*old_ctx
)
7198 aio_co_reschedule_self(old_ctx
);
7199 bdrv_dec_in_flight(bs
);
7202 void coroutine_fn
bdrv_co_lock(BlockDriverState
*bs
)
7204 AioContext
*ctx
= bdrv_get_aio_context(bs
);
7206 /* In the main thread, bs->aio_context won't change concurrently */
7207 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
7210 * We're in coroutine context, so we already hold the lock of the main
7211 * loop AioContext. Don't lock it twice to avoid deadlocks.
7213 assert(qemu_in_coroutine());
7214 if (ctx
!= qemu_get_aio_context()) {
7215 aio_context_acquire(ctx
);
7219 void coroutine_fn
bdrv_co_unlock(BlockDriverState
*bs
)
7221 AioContext
*ctx
= bdrv_get_aio_context(bs
);
7223 assert(qemu_in_coroutine());
7224 if (ctx
!= qemu_get_aio_context()) {
7225 aio_context_release(ctx
);
7229 static void bdrv_do_remove_aio_context_notifier(BdrvAioNotifier
*ban
)
7231 GLOBAL_STATE_CODE();
7232 QLIST_REMOVE(ban
, list
);
7236 static void bdrv_detach_aio_context(BlockDriverState
*bs
)
7238 BdrvAioNotifier
*baf
, *baf_tmp
;
7240 assert(!bs
->walking_aio_notifiers
);
7241 GLOBAL_STATE_CODE();
7242 bs
->walking_aio_notifiers
= true;
7243 QLIST_FOREACH_SAFE(baf
, &bs
->aio_notifiers
, list
, baf_tmp
) {
7245 bdrv_do_remove_aio_context_notifier(baf
);
7247 baf
->detach_aio_context(baf
->opaque
);
7250 /* Never mind iterating again to check for ->deleted. bdrv_close() will
7251 * remove remaining aio notifiers if we aren't called again.
7253 bs
->walking_aio_notifiers
= false;
7255 if (bs
->drv
&& bs
->drv
->bdrv_detach_aio_context
) {
7256 bs
->drv
->bdrv_detach_aio_context(bs
);
7259 if (bs
->quiesce_counter
) {
7260 aio_enable_external(bs
->aio_context
);
7262 bs
->aio_context
= NULL
;
7265 static void bdrv_attach_aio_context(BlockDriverState
*bs
,
7266 AioContext
*new_context
)
7268 BdrvAioNotifier
*ban
, *ban_tmp
;
7269 GLOBAL_STATE_CODE();
7271 if (bs
->quiesce_counter
) {
7272 aio_disable_external(new_context
);
7275 bs
->aio_context
= new_context
;
7277 if (bs
->drv
&& bs
->drv
->bdrv_attach_aio_context
) {
7278 bs
->drv
->bdrv_attach_aio_context(bs
, new_context
);
7281 assert(!bs
->walking_aio_notifiers
);
7282 bs
->walking_aio_notifiers
= true;
7283 QLIST_FOREACH_SAFE(ban
, &bs
->aio_notifiers
, list
, ban_tmp
) {
7285 bdrv_do_remove_aio_context_notifier(ban
);
7287 ban
->attached_aio_context(new_context
, ban
->opaque
);
7290 bs
->walking_aio_notifiers
= false;
7293 typedef struct BdrvStateSetAioContext
{
7294 AioContext
*new_ctx
;
7295 BlockDriverState
*bs
;
7296 } BdrvStateSetAioContext
;
7298 static bool bdrv_parent_change_aio_context(BdrvChild
*c
, AioContext
*ctx
,
7299 GHashTable
*visited
,
7303 GLOBAL_STATE_CODE();
7304 if (g_hash_table_contains(visited
, c
)) {
7307 g_hash_table_add(visited
, c
);
7310 * A BdrvChildClass that doesn't handle AioContext changes cannot
7311 * tolerate any AioContext changes
7313 if (!c
->klass
->change_aio_ctx
) {
7314 char *user
= bdrv_child_user_desc(c
);
7315 error_setg(errp
, "Changing iothreads is not supported by %s", user
);
7319 if (!c
->klass
->change_aio_ctx(c
, ctx
, visited
, tran
, errp
)) {
7320 assert(!errp
|| *errp
);
7326 bool bdrv_child_change_aio_context(BdrvChild
*c
, AioContext
*ctx
,
7327 GHashTable
*visited
, Transaction
*tran
,
7330 GLOBAL_STATE_CODE();
7331 if (g_hash_table_contains(visited
, c
)) {
7334 g_hash_table_add(visited
, c
);
7335 return bdrv_change_aio_context(c
->bs
, ctx
, visited
, tran
, errp
);
7338 static void bdrv_set_aio_context_clean(void *opaque
)
7340 BdrvStateSetAioContext
*state
= (BdrvStateSetAioContext
*) opaque
;
7341 BlockDriverState
*bs
= (BlockDriverState
*) state
->bs
;
7343 /* Paired with bdrv_drained_begin in bdrv_change_aio_context() */
7344 bdrv_drained_end(bs
);
7349 static void bdrv_set_aio_context_commit(void *opaque
)
7351 BdrvStateSetAioContext
*state
= (BdrvStateSetAioContext
*) opaque
;
7352 BlockDriverState
*bs
= (BlockDriverState
*) state
->bs
;
7353 AioContext
*new_context
= state
->new_ctx
;
7354 AioContext
*old_context
= bdrv_get_aio_context(bs
);
7357 * Take the old AioContex when detaching it from bs.
7358 * At this point, new_context lock is already acquired, and we are now
7359 * also taking old_context. This is safe as long as bdrv_detach_aio_context
7360 * does not call AIO_POLL_WHILE().
7362 if (old_context
!= qemu_get_aio_context()) {
7363 aio_context_acquire(old_context
);
7365 bdrv_detach_aio_context(bs
);
7366 if (old_context
!= qemu_get_aio_context()) {
7367 aio_context_release(old_context
);
7369 bdrv_attach_aio_context(bs
, new_context
);
7372 static TransactionActionDrv set_aio_context
= {
7373 .commit
= bdrv_set_aio_context_commit
,
7374 .clean
= bdrv_set_aio_context_clean
,
7378 * Changes the AioContext used for fd handlers, timers, and BHs by this
7379 * BlockDriverState and all its children and parents.
7381 * Must be called from the main AioContext.
7383 * The caller must own the AioContext lock for the old AioContext of bs, but it
7384 * must not own the AioContext lock for new_context (unless new_context is the
7385 * same as the current context of bs).
7387 * @visited will accumulate all visited BdrvChild objects. The caller is
7388 * responsible for freeing the list afterwards.
7390 static bool bdrv_change_aio_context(BlockDriverState
*bs
, AioContext
*ctx
,
7391 GHashTable
*visited
, Transaction
*tran
,
7395 BdrvStateSetAioContext
*state
;
7397 GLOBAL_STATE_CODE();
7399 if (bdrv_get_aio_context(bs
) == ctx
) {
7403 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
7404 if (!bdrv_parent_change_aio_context(c
, ctx
, visited
, tran
, errp
)) {
7409 QLIST_FOREACH(c
, &bs
->children
, next
) {
7410 if (!bdrv_child_change_aio_context(c
, ctx
, visited
, tran
, errp
)) {
7415 state
= g_new(BdrvStateSetAioContext
, 1);
7416 *state
= (BdrvStateSetAioContext
) {
7421 /* Paired with bdrv_drained_end in bdrv_set_aio_context_clean() */
7422 bdrv_drained_begin(bs
);
7424 tran_add(tran
, &set_aio_context
, state
);
7430 * Change bs's and recursively all of its parents' and children's AioContext
7431 * to the given new context, returning an error if that isn't possible.
7433 * If ignore_child is not NULL, that child (and its subgraph) will not
7436 * This function still requires the caller to take the bs current
7437 * AioContext lock, otherwise draining will fail since AIO_WAIT_WHILE
7438 * assumes the lock is always held if bs is in another AioContext.
7439 * For the same reason, it temporarily also holds the new AioContext, since
7440 * bdrv_drained_end calls BDRV_POLL_WHILE that assumes the lock is taken too.
7441 * Therefore the new AioContext lock must not be taken by the caller.
7443 int bdrv_try_change_aio_context(BlockDriverState
*bs
, AioContext
*ctx
,
7444 BdrvChild
*ignore_child
, Error
**errp
)
7447 GHashTable
*visited
;
7449 AioContext
*old_context
= bdrv_get_aio_context(bs
);
7450 GLOBAL_STATE_CODE();
7453 * Recursion phase: go through all nodes of the graph.
7454 * Take care of checking that all nodes support changing AioContext
7455 * and drain them, builing a linear list of callbacks to run if everything
7456 * is successful (the transaction itself).
7459 visited
= g_hash_table_new(NULL
, NULL
);
7461 g_hash_table_add(visited
, ignore_child
);
7463 ret
= bdrv_change_aio_context(bs
, ctx
, visited
, tran
, errp
);
7464 g_hash_table_destroy(visited
);
7467 * Linear phase: go through all callbacks collected in the transaction.
7468 * Run all callbacks collected in the recursion to switch all nodes
7469 * AioContext lock (transaction commit), or undo all changes done in the
7470 * recursion (transaction abort).
7474 /* Just run clean() callbacks. No AioContext changed. */
7480 * Release old AioContext, it won't be needed anymore, as all
7481 * bdrv_drained_begin() have been called already.
7483 if (qemu_get_aio_context() != old_context
) {
7484 aio_context_release(old_context
);
7488 * Acquire new AioContext since bdrv_drained_end() is going to be called
7489 * after we switched all nodes in the new AioContext, and the function
7490 * assumes that the lock of the bs is always taken.
7492 if (qemu_get_aio_context() != ctx
) {
7493 aio_context_acquire(ctx
);
7498 if (qemu_get_aio_context() != ctx
) {
7499 aio_context_release(ctx
);
7502 /* Re-acquire the old AioContext, since the caller takes and releases it. */
7503 if (qemu_get_aio_context() != old_context
) {
7504 aio_context_acquire(old_context
);
7510 void bdrv_add_aio_context_notifier(BlockDriverState
*bs
,
7511 void (*attached_aio_context
)(AioContext
*new_context
, void *opaque
),
7512 void (*detach_aio_context
)(void *opaque
), void *opaque
)
7514 BdrvAioNotifier
*ban
= g_new(BdrvAioNotifier
, 1);
7515 *ban
= (BdrvAioNotifier
){
7516 .attached_aio_context
= attached_aio_context
,
7517 .detach_aio_context
= detach_aio_context
,
7520 GLOBAL_STATE_CODE();
7522 QLIST_INSERT_HEAD(&bs
->aio_notifiers
, ban
, list
);
7525 void bdrv_remove_aio_context_notifier(BlockDriverState
*bs
,
7526 void (*attached_aio_context
)(AioContext
*,
7528 void (*detach_aio_context
)(void *),
7531 BdrvAioNotifier
*ban
, *ban_next
;
7532 GLOBAL_STATE_CODE();
7534 QLIST_FOREACH_SAFE(ban
, &bs
->aio_notifiers
, list
, ban_next
) {
7535 if (ban
->attached_aio_context
== attached_aio_context
&&
7536 ban
->detach_aio_context
== detach_aio_context
&&
7537 ban
->opaque
== opaque
&&
7538 ban
->deleted
== false)
7540 if (bs
->walking_aio_notifiers
) {
7541 ban
->deleted
= true;
7543 bdrv_do_remove_aio_context_notifier(ban
);
7552 int bdrv_amend_options(BlockDriverState
*bs
, QemuOpts
*opts
,
7553 BlockDriverAmendStatusCB
*status_cb
, void *cb_opaque
,
7557 GLOBAL_STATE_CODE();
7559 error_setg(errp
, "Node is ejected");
7562 if (!bs
->drv
->bdrv_amend_options
) {
7563 error_setg(errp
, "Block driver '%s' does not support option amendment",
7564 bs
->drv
->format_name
);
7567 return bs
->drv
->bdrv_amend_options(bs
, opts
, status_cb
,
7568 cb_opaque
, force
, errp
);
7572 * This function checks whether the given @to_replace is allowed to be
7573 * replaced by a node that always shows the same data as @bs. This is
7574 * used for example to verify whether the mirror job can replace
7575 * @to_replace by the target mirrored from @bs.
7576 * To be replaceable, @bs and @to_replace may either be guaranteed to
7577 * always show the same data (because they are only connected through
7578 * filters), or some driver may allow replacing one of its children
7579 * because it can guarantee that this child's data is not visible at
7580 * all (for example, for dissenting quorum children that have no other
7583 bool bdrv_recurse_can_replace(BlockDriverState
*bs
,
7584 BlockDriverState
*to_replace
)
7586 BlockDriverState
*filtered
;
7588 GLOBAL_STATE_CODE();
7590 if (!bs
|| !bs
->drv
) {
7594 if (bs
== to_replace
) {
7598 /* See what the driver can do */
7599 if (bs
->drv
->bdrv_recurse_can_replace
) {
7600 return bs
->drv
->bdrv_recurse_can_replace(bs
, to_replace
);
7603 /* For filters without an own implementation, we can recurse on our own */
7604 filtered
= bdrv_filter_bs(bs
);
7606 return bdrv_recurse_can_replace(filtered
, to_replace
);
7614 * Check whether the given @node_name can be replaced by a node that
7615 * has the same data as @parent_bs. If so, return @node_name's BDS;
7618 * @node_name must be a (recursive) *child of @parent_bs (or this
7619 * function will return NULL).
7621 * The result (whether the node can be replaced or not) is only valid
7622 * for as long as no graph or permission changes occur.
7624 BlockDriverState
*check_to_replace_node(BlockDriverState
*parent_bs
,
7625 const char *node_name
, Error
**errp
)
7627 BlockDriverState
*to_replace_bs
= bdrv_find_node(node_name
);
7628 AioContext
*aio_context
;
7630 GLOBAL_STATE_CODE();
7632 if (!to_replace_bs
) {
7633 error_setg(errp
, "Failed to find node with node-name='%s'", node_name
);
7637 aio_context
= bdrv_get_aio_context(to_replace_bs
);
7638 aio_context_acquire(aio_context
);
7640 if (bdrv_op_is_blocked(to_replace_bs
, BLOCK_OP_TYPE_REPLACE
, errp
)) {
7641 to_replace_bs
= NULL
;
7645 /* We don't want arbitrary node of the BDS chain to be replaced only the top
7646 * most non filter in order to prevent data corruption.
7647 * Another benefit is that this tests exclude backing files which are
7648 * blocked by the backing blockers.
7650 if (!bdrv_recurse_can_replace(parent_bs
, to_replace_bs
)) {
7651 error_setg(errp
, "Cannot replace '%s' by a node mirrored from '%s', "
7652 "because it cannot be guaranteed that doing so would not "
7653 "lead to an abrupt change of visible data",
7654 node_name
, parent_bs
->node_name
);
7655 to_replace_bs
= NULL
;
7660 aio_context_release(aio_context
);
7661 return to_replace_bs
;
7665 * Iterates through the list of runtime option keys that are said to
7666 * be "strong" for a BDS. An option is called "strong" if it changes
7667 * a BDS's data. For example, the null block driver's "size" and
7668 * "read-zeroes" options are strong, but its "latency-ns" option is
7671 * If a key returned by this function ends with a dot, all options
7672 * starting with that prefix are strong.
7674 static const char *const *strong_options(BlockDriverState
*bs
,
7675 const char *const *curopt
)
7677 static const char *const global_options
[] = {
7678 "driver", "filename", NULL
7682 return &global_options
[0];
7686 if (curopt
== &global_options
[ARRAY_SIZE(global_options
) - 1] && bs
->drv
) {
7687 curopt
= bs
->drv
->strong_runtime_opts
;
7690 return (curopt
&& *curopt
) ? curopt
: NULL
;
7694 * Copies all strong runtime options from bs->options to the given
7695 * QDict. The set of strong option keys is determined by invoking
7698 * Returns true iff any strong option was present in bs->options (and
7699 * thus copied to the target QDict) with the exception of "filename"
7700 * and "driver". The caller is expected to use this value to decide
7701 * whether the existence of strong options prevents the generation of
7704 static bool append_strong_runtime_options(QDict
*d
, BlockDriverState
*bs
)
7706 bool found_any
= false;
7707 const char *const *option_name
= NULL
;
7713 while ((option_name
= strong_options(bs
, option_name
))) {
7714 bool option_given
= false;
7716 assert(strlen(*option_name
) > 0);
7717 if ((*option_name
)[strlen(*option_name
) - 1] != '.') {
7718 QObject
*entry
= qdict_get(bs
->options
, *option_name
);
7723 qdict_put_obj(d
, *option_name
, qobject_ref(entry
));
7724 option_given
= true;
7726 const QDictEntry
*entry
;
7727 for (entry
= qdict_first(bs
->options
); entry
;
7728 entry
= qdict_next(bs
->options
, entry
))
7730 if (strstart(qdict_entry_key(entry
), *option_name
, NULL
)) {
7731 qdict_put_obj(d
, qdict_entry_key(entry
),
7732 qobject_ref(qdict_entry_value(entry
)));
7733 option_given
= true;
7738 /* While "driver" and "filename" need to be included in a JSON filename,
7739 * their existence does not prohibit generation of a plain filename. */
7740 if (!found_any
&& option_given
&&
7741 strcmp(*option_name
, "driver") && strcmp(*option_name
, "filename"))
7747 if (!qdict_haskey(d
, "driver")) {
7748 /* Drivers created with bdrv_new_open_driver() may not have a
7749 * @driver option. Add it here. */
7750 qdict_put_str(d
, "driver", bs
->drv
->format_name
);
7756 /* Note: This function may return false positives; it may return true
7757 * even if opening the backing file specified by bs's image header
7758 * would result in exactly bs->backing. */
7759 static bool bdrv_backing_overridden(BlockDriverState
*bs
)
7761 GLOBAL_STATE_CODE();
7763 return strcmp(bs
->auto_backing_file
,
7764 bs
->backing
->bs
->filename
);
7766 /* No backing BDS, so if the image header reports any backing
7767 * file, it must have been suppressed */
7768 return bs
->auto_backing_file
[0] != '\0';
7772 /* Updates the following BDS fields:
7773 * - exact_filename: A filename which may be used for opening a block device
7774 * which (mostly) equals the given BDS (even without any
7775 * other options; so reading and writing must return the same
7776 * results, but caching etc. may be different)
7777 * - full_open_options: Options which, when given when opening a block device
7778 * (without a filename), result in a BDS (mostly)
7779 * equalling the given one
7780 * - filename: If exact_filename is set, it is copied here. Otherwise,
7781 * full_open_options is converted to a JSON object, prefixed with
7782 * "json:" (for use through the JSON pseudo protocol) and put here.
7784 void bdrv_refresh_filename(BlockDriverState
*bs
)
7786 BlockDriver
*drv
= bs
->drv
;
7788 BlockDriverState
*primary_child_bs
;
7790 bool backing_overridden
;
7791 bool generate_json_filename
; /* Whether our default implementation should
7792 fill exact_filename (false) or not (true) */
7794 GLOBAL_STATE_CODE();
7800 /* This BDS's file name may depend on any of its children's file names, so
7801 * refresh those first */
7802 QLIST_FOREACH(child
, &bs
->children
, next
) {
7803 bdrv_refresh_filename(child
->bs
);
7807 /* For implicit nodes, just copy everything from the single child */
7808 child
= QLIST_FIRST(&bs
->children
);
7809 assert(QLIST_NEXT(child
, next
) == NULL
);
7811 pstrcpy(bs
->exact_filename
, sizeof(bs
->exact_filename
),
7812 child
->bs
->exact_filename
);
7813 pstrcpy(bs
->filename
, sizeof(bs
->filename
), child
->bs
->filename
);
7815 qobject_unref(bs
->full_open_options
);
7816 bs
->full_open_options
= qobject_ref(child
->bs
->full_open_options
);
7821 backing_overridden
= bdrv_backing_overridden(bs
);
7823 if (bs
->open_flags
& BDRV_O_NO_IO
) {
7824 /* Without I/O, the backing file does not change anything.
7825 * Therefore, in such a case (primarily qemu-img), we can
7826 * pretend the backing file has not been overridden even if
7827 * it technically has been. */
7828 backing_overridden
= false;
7831 /* Gather the options QDict */
7833 generate_json_filename
= append_strong_runtime_options(opts
, bs
);
7834 generate_json_filename
|= backing_overridden
;
7836 if (drv
->bdrv_gather_child_options
) {
7837 /* Some block drivers may not want to present all of their children's
7838 * options, or name them differently from BdrvChild.name */
7839 drv
->bdrv_gather_child_options(bs
, opts
, backing_overridden
);
7841 QLIST_FOREACH(child
, &bs
->children
, next
) {
7842 if (child
== bs
->backing
&& !backing_overridden
) {
7843 /* We can skip the backing BDS if it has not been overridden */
7847 qdict_put(opts
, child
->name
,
7848 qobject_ref(child
->bs
->full_open_options
));
7851 if (backing_overridden
&& !bs
->backing
) {
7852 /* Force no backing file */
7853 qdict_put_null(opts
, "backing");
7857 qobject_unref(bs
->full_open_options
);
7858 bs
->full_open_options
= opts
;
7860 primary_child_bs
= bdrv_primary_bs(bs
);
7862 if (drv
->bdrv_refresh_filename
) {
7863 /* Obsolete information is of no use here, so drop the old file name
7864 * information before refreshing it */
7865 bs
->exact_filename
[0] = '\0';
7867 drv
->bdrv_refresh_filename(bs
);
7868 } else if (primary_child_bs
) {
7870 * Try to reconstruct valid information from the underlying
7871 * file -- this only works for format nodes (filter nodes
7872 * cannot be probed and as such must be selected by the user
7873 * either through an options dict, or through a special
7874 * filename which the filter driver must construct in its
7875 * .bdrv_refresh_filename() implementation).
7878 bs
->exact_filename
[0] = '\0';
7881 * We can use the underlying file's filename if:
7882 * - it has a filename,
7883 * - the current BDS is not a filter,
7884 * - the file is a protocol BDS, and
7885 * - opening that file (as this BDS's format) will automatically create
7886 * the BDS tree we have right now, that is:
7887 * - the user did not significantly change this BDS's behavior with
7888 * some explicit (strong) options
7889 * - no non-file child of this BDS has been overridden by the user
7890 * Both of these conditions are represented by generate_json_filename.
7892 if (primary_child_bs
->exact_filename
[0] &&
7893 primary_child_bs
->drv
->bdrv_file_open
&&
7894 !drv
->is_filter
&& !generate_json_filename
)
7896 strcpy(bs
->exact_filename
, primary_child_bs
->exact_filename
);
7900 if (bs
->exact_filename
[0]) {
7901 pstrcpy(bs
->filename
, sizeof(bs
->filename
), bs
->exact_filename
);
7903 GString
*json
= qobject_to_json(QOBJECT(bs
->full_open_options
));
7904 if (snprintf(bs
->filename
, sizeof(bs
->filename
), "json:%s",
7905 json
->str
) >= sizeof(bs
->filename
)) {
7906 /* Give user a hint if we truncated things. */
7907 strcpy(bs
->filename
+ sizeof(bs
->filename
) - 4, "...");
7909 g_string_free(json
, true);
7913 char *bdrv_dirname(BlockDriverState
*bs
, Error
**errp
)
7915 BlockDriver
*drv
= bs
->drv
;
7916 BlockDriverState
*child_bs
;
7918 GLOBAL_STATE_CODE();
7921 error_setg(errp
, "Node '%s' is ejected", bs
->node_name
);
7925 if (drv
->bdrv_dirname
) {
7926 return drv
->bdrv_dirname(bs
, errp
);
7929 child_bs
= bdrv_primary_bs(bs
);
7931 return bdrv_dirname(child_bs
, errp
);
7934 bdrv_refresh_filename(bs
);
7935 if (bs
->exact_filename
[0] != '\0') {
7936 return path_combine(bs
->exact_filename
, "");
7939 error_setg(errp
, "Cannot generate a base directory for %s nodes",
7945 * Hot add/remove a BDS's child. So the user can take a child offline when
7946 * it is broken and take a new child online
7948 void bdrv_add_child(BlockDriverState
*parent_bs
, BlockDriverState
*child_bs
,
7951 GLOBAL_STATE_CODE();
7952 if (!parent_bs
->drv
|| !parent_bs
->drv
->bdrv_add_child
) {
7953 error_setg(errp
, "The node %s does not support adding a child",
7954 bdrv_get_device_or_node_name(parent_bs
));
7958 if (!QLIST_EMPTY(&child_bs
->parents
)) {
7959 error_setg(errp
, "The node %s already has a parent",
7960 child_bs
->node_name
);
7964 parent_bs
->drv
->bdrv_add_child(parent_bs
, child_bs
, errp
);
7967 void bdrv_del_child(BlockDriverState
*parent_bs
, BdrvChild
*child
, Error
**errp
)
7971 GLOBAL_STATE_CODE();
7972 if (!parent_bs
->drv
|| !parent_bs
->drv
->bdrv_del_child
) {
7973 error_setg(errp
, "The node %s does not support removing a child",
7974 bdrv_get_device_or_node_name(parent_bs
));
7978 QLIST_FOREACH(tmp
, &parent_bs
->children
, next
) {
7985 error_setg(errp
, "The node %s does not have a child named %s",
7986 bdrv_get_device_or_node_name(parent_bs
),
7987 bdrv_get_device_or_node_name(child
->bs
));
7991 parent_bs
->drv
->bdrv_del_child(parent_bs
, child
, errp
);
7994 int bdrv_make_empty(BdrvChild
*c
, Error
**errp
)
7996 BlockDriver
*drv
= c
->bs
->drv
;
7999 GLOBAL_STATE_CODE();
8000 assert(c
->perm
& (BLK_PERM_WRITE
| BLK_PERM_WRITE_UNCHANGED
));
8002 if (!drv
->bdrv_make_empty
) {
8003 error_setg(errp
, "%s does not support emptying nodes",
8008 ret
= drv
->bdrv_make_empty(c
->bs
);
8010 error_setg_errno(errp
, -ret
, "Failed to empty %s",
8019 * Return the child that @bs acts as an overlay for, and from which data may be
8020 * copied in COW or COR operations. Usually this is the backing file.
8022 BdrvChild
*bdrv_cow_child(BlockDriverState
*bs
)
8026 if (!bs
|| !bs
->drv
) {
8030 if (bs
->drv
->is_filter
) {
8038 assert(bs
->backing
->role
& BDRV_CHILD_COW
);
8043 * If @bs acts as a filter for exactly one of its children, return
8046 BdrvChild
*bdrv_filter_child(BlockDriverState
*bs
)
8051 if (!bs
|| !bs
->drv
) {
8055 if (!bs
->drv
->is_filter
) {
8059 /* Only one of @backing or @file may be used */
8060 assert(!(bs
->backing
&& bs
->file
));
8062 c
= bs
->backing
?: bs
->file
;
8067 assert(c
->role
& BDRV_CHILD_FILTERED
);
8072 * Return either the result of bdrv_cow_child() or bdrv_filter_child(),
8073 * whichever is non-NULL.
8075 * Return NULL if both are NULL.
8077 BdrvChild
*bdrv_filter_or_cow_child(BlockDriverState
*bs
)
8079 BdrvChild
*cow_child
= bdrv_cow_child(bs
);
8080 BdrvChild
*filter_child
= bdrv_filter_child(bs
);
8083 /* Filter nodes cannot have COW backing files */
8084 assert(!(cow_child
&& filter_child
));
8086 return cow_child
?: filter_child
;
8090 * Return the primary child of this node: For filters, that is the
8091 * filtered child. For other nodes, that is usually the child storing
8093 * (A generally more helpful description is that this is (usually) the
8094 * child that has the same filename as @bs.)
8096 * Drivers do not necessarily have a primary child; for example quorum
8099 BdrvChild
*bdrv_primary_child(BlockDriverState
*bs
)
8101 BdrvChild
*c
, *found
= NULL
;
8104 QLIST_FOREACH(c
, &bs
->children
, next
) {
8105 if (c
->role
& BDRV_CHILD_PRIMARY
) {
8114 static BlockDriverState
*bdrv_do_skip_filters(BlockDriverState
*bs
,
8115 bool stop_on_explicit_filter
)
8123 while (!(stop_on_explicit_filter
&& !bs
->implicit
)) {
8124 c
= bdrv_filter_child(bs
);
8127 * A filter that is embedded in a working block graph must
8128 * have a child. Assert this here so this function does
8129 * not return a filter node that is not expected by the
8132 assert(!bs
->drv
|| !bs
->drv
->is_filter
);
8138 * Note that this treats nodes with bs->drv == NULL as not being
8139 * filters (bs->drv == NULL should be replaced by something else
8141 * The advantage of this behavior is that this function will thus
8142 * always return a non-NULL value (given a non-NULL @bs).
8149 * Return the first BDS that has not been added implicitly or that
8150 * does not have a filtered child down the chain starting from @bs
8151 * (including @bs itself).
8153 BlockDriverState
*bdrv_skip_implicit_filters(BlockDriverState
*bs
)
8155 GLOBAL_STATE_CODE();
8156 return bdrv_do_skip_filters(bs
, true);
8160 * Return the first BDS that does not have a filtered child down the
8161 * chain starting from @bs (including @bs itself).
8163 BlockDriverState
*bdrv_skip_filters(BlockDriverState
*bs
)
8166 return bdrv_do_skip_filters(bs
, false);
8170 * For a backing chain, return the first non-filter backing image of
8171 * the first non-filter image.
8173 BlockDriverState
*bdrv_backing_chain_next(BlockDriverState
*bs
)
8176 return bdrv_skip_filters(bdrv_cow_bs(bdrv_skip_filters(bs
)));
8180 * Check whether [offset, offset + bytes) overlaps with the cached
8181 * block-status data region.
8183 * If so, and @pnum is not NULL, set *pnum to `bsc.data_end - offset`,
8184 * which is what bdrv_bsc_is_data()'s interface needs.
8185 * Otherwise, *pnum is not touched.
8187 static bool bdrv_bsc_range_overlaps_locked(BlockDriverState
*bs
,
8188 int64_t offset
, int64_t bytes
,
8191 BdrvBlockStatusCache
*bsc
= qatomic_rcu_read(&bs
->block_status_cache
);
8195 qatomic_read(&bsc
->valid
) &&
8196 ranges_overlap(offset
, bytes
, bsc
->data_start
,
8197 bsc
->data_end
- bsc
->data_start
);
8199 if (overlaps
&& pnum
) {
8200 *pnum
= bsc
->data_end
- offset
;
8207 * See block_int.h for this function's documentation.
8209 bool bdrv_bsc_is_data(BlockDriverState
*bs
, int64_t offset
, int64_t *pnum
)
8212 RCU_READ_LOCK_GUARD();
8213 return bdrv_bsc_range_overlaps_locked(bs
, offset
, 1, pnum
);
8217 * See block_int.h for this function's documentation.
8219 void bdrv_bsc_invalidate_range(BlockDriverState
*bs
,
8220 int64_t offset
, int64_t bytes
)
8223 RCU_READ_LOCK_GUARD();
8225 if (bdrv_bsc_range_overlaps_locked(bs
, offset
, bytes
, NULL
)) {
8226 qatomic_set(&bs
->block_status_cache
->valid
, false);
8231 * See block_int.h for this function's documentation.
8233 void bdrv_bsc_fill(BlockDriverState
*bs
, int64_t offset
, int64_t bytes
)
8235 BdrvBlockStatusCache
*new_bsc
= g_new(BdrvBlockStatusCache
, 1);
8236 BdrvBlockStatusCache
*old_bsc
;
8239 *new_bsc
= (BdrvBlockStatusCache
) {
8241 .data_start
= offset
,
8242 .data_end
= offset
+ bytes
,
8245 QEMU_LOCK_GUARD(&bs
->bsc_modify_lock
);
8247 old_bsc
= qatomic_rcu_read(&bs
->block_status_cache
);
8248 qatomic_rcu_set(&bs
->block_status_cache
, new_bsc
);
8250 g_free_rcu(old_bsc
, rcu
);