2 * QEMU System Emulator block driver
4 * Copyright (c) 2003 Fabrice Bellard
5 * Copyright (c) 2020 Virtuozzo International GmbH.
7 * Permission is hereby granted, free of charge, to any person obtaining a copy
8 * of this software and associated documentation files (the "Software"), to deal
9 * in the Software without restriction, including without limitation the rights
10 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
11 * copies of the Software, and to permit persons to whom the Software is
12 * furnished to do so, subject to the following conditions:
14 * The above copyright notice and this permission notice shall be included in
15 * all copies or substantial portions of the Software.
17 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
18 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
19 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
20 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
21 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
22 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
26 #include "qemu/osdep.h"
27 #include "block/trace.h"
28 #include "block/block_int.h"
29 #include "block/blockjob.h"
30 #include "block/dirty-bitmap.h"
31 #include "block/fuse.h"
32 #include "block/nbd.h"
33 #include "block/qdict.h"
34 #include "qemu/error-report.h"
35 #include "block/module_block.h"
36 #include "qemu/main-loop.h"
37 #include "qemu/module.h"
38 #include "qapi/error.h"
39 #include "qapi/qmp/qdict.h"
40 #include "qapi/qmp/qjson.h"
41 #include "qapi/qmp/qnull.h"
42 #include "qapi/qmp/qstring.h"
43 #include "qapi/qobject-output-visitor.h"
44 #include "qapi/qapi-visit-block-core.h"
45 #include "sysemu/block-backend.h"
46 #include "qemu/notify.h"
47 #include "qemu/option.h"
48 #include "qemu/coroutine.h"
49 #include "block/qapi.h"
50 #include "qemu/timer.h"
51 #include "qemu/cutils.h"
53 #include "qemu/range.h"
55 #include "block/coroutines.h"
58 #include <sys/ioctl.h>
59 #include <sys/queue.h>
60 #if defined(HAVE_SYS_DISK_H)
69 #define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */
71 /* Protected by BQL */
72 static QTAILQ_HEAD(, BlockDriverState
) graph_bdrv_states
=
73 QTAILQ_HEAD_INITIALIZER(graph_bdrv_states
);
75 /* Protected by BQL */
76 static QTAILQ_HEAD(, BlockDriverState
) all_bdrv_states
=
77 QTAILQ_HEAD_INITIALIZER(all_bdrv_states
);
79 /* Protected by BQL */
80 static QLIST_HEAD(, BlockDriver
) bdrv_drivers
=
81 QLIST_HEAD_INITIALIZER(bdrv_drivers
);
83 static BlockDriverState
*bdrv_open_inherit(const char *filename
,
84 const char *reference
,
85 QDict
*options
, int flags
,
86 BlockDriverState
*parent
,
87 const BdrvChildClass
*child_class
,
88 BdrvChildRole child_role
,
91 static bool bdrv_recurse_has_child(BlockDriverState
*bs
,
92 BlockDriverState
*child
);
94 static void GRAPH_WRLOCK
95 bdrv_replace_child_noperm(BdrvChild
*child
, BlockDriverState
*new_bs
);
97 static void GRAPH_WRLOCK
98 bdrv_remove_child(BdrvChild
*child
, Transaction
*tran
);
100 static int bdrv_reopen_prepare(BDRVReopenState
*reopen_state
,
101 BlockReopenQueue
*queue
,
102 Transaction
*change_child_tran
, Error
**errp
);
103 static void bdrv_reopen_commit(BDRVReopenState
*reopen_state
);
104 static void bdrv_reopen_abort(BDRVReopenState
*reopen_state
);
106 static bool bdrv_backing_overridden(BlockDriverState
*bs
);
108 static bool bdrv_change_aio_context(BlockDriverState
*bs
, AioContext
*ctx
,
109 GHashTable
*visited
, Transaction
*tran
,
112 /* If non-zero, use only whitelisted block drivers */
113 static int use_bdrv_whitelist
;
116 static int is_windows_drive_prefix(const char *filename
)
118 return (((filename
[0] >= 'a' && filename
[0] <= 'z') ||
119 (filename
[0] >= 'A' && filename
[0] <= 'Z')) &&
123 int is_windows_drive(const char *filename
)
125 if (is_windows_drive_prefix(filename
) &&
128 if (strstart(filename
, "\\\\.\\", NULL
) ||
129 strstart(filename
, "//./", NULL
))
135 size_t bdrv_opt_mem_align(BlockDriverState
*bs
)
137 if (!bs
|| !bs
->drv
) {
138 /* page size or 4k (hdd sector size) should be on the safe side */
139 return MAX(4096, qemu_real_host_page_size());
143 return bs
->bl
.opt_mem_alignment
;
146 size_t bdrv_min_mem_align(BlockDriverState
*bs
)
148 if (!bs
|| !bs
->drv
) {
149 /* page size or 4k (hdd sector size) should be on the safe side */
150 return MAX(4096, qemu_real_host_page_size());
154 return bs
->bl
.min_mem_alignment
;
157 /* check if the path starts with "<protocol>:" */
158 int path_has_protocol(const char *path
)
163 if (is_windows_drive(path
) ||
164 is_windows_drive_prefix(path
)) {
167 p
= path
+ strcspn(path
, ":/\\");
169 p
= path
+ strcspn(path
, ":/");
175 int path_is_absolute(const char *path
)
178 /* specific case for names like: "\\.\d:" */
179 if (is_windows_drive(path
) || is_windows_drive_prefix(path
)) {
182 return (*path
== '/' || *path
== '\\');
184 return (*path
== '/');
188 /* if filename is absolute, just return its duplicate. Otherwise, build a
189 path to it by considering it is relative to base_path. URL are
191 char *path_combine(const char *base_path
, const char *filename
)
193 const char *protocol_stripped
= NULL
;
198 if (path_is_absolute(filename
)) {
199 return g_strdup(filename
);
202 if (path_has_protocol(base_path
)) {
203 protocol_stripped
= strchr(base_path
, ':');
204 if (protocol_stripped
) {
208 p
= protocol_stripped
?: base_path
;
210 p1
= strrchr(base_path
, '/');
214 p2
= strrchr(base_path
, '\\');
215 if (!p1
|| p2
> p1
) {
230 result
= g_malloc(len
+ strlen(filename
) + 1);
231 memcpy(result
, base_path
, len
);
232 strcpy(result
+ len
, filename
);
238 * Helper function for bdrv_parse_filename() implementations to remove optional
239 * protocol prefixes (especially "file:") from a filename and for putting the
240 * stripped filename into the options QDict if there is such a prefix.
242 void bdrv_parse_filename_strip_prefix(const char *filename
, const char *prefix
,
245 if (strstart(filename
, prefix
, &filename
)) {
246 /* Stripping the explicit protocol prefix may result in a protocol
247 * prefix being (wrongly) detected (if the filename contains a colon) */
248 if (path_has_protocol(filename
)) {
249 GString
*fat_filename
;
251 /* This means there is some colon before the first slash; therefore,
252 * this cannot be an absolute path */
253 assert(!path_is_absolute(filename
));
255 /* And we can thus fix the protocol detection issue by prefixing it
257 fat_filename
= g_string_new("./");
258 g_string_append(fat_filename
, filename
);
260 assert(!path_has_protocol(fat_filename
->str
));
262 qdict_put(options
, "filename",
263 qstring_from_gstring(fat_filename
));
265 /* If no protocol prefix was detected, we can use the shortened
267 qdict_put_str(options
, "filename", filename
);
273 /* Returns whether the image file is opened as read-only. Note that this can
274 * return false and writing to the image file is still not possible because the
275 * image is inactivated. */
276 bool bdrv_is_read_only(BlockDriverState
*bs
)
279 return !(bs
->open_flags
& BDRV_O_RDWR
);
282 static int GRAPH_RDLOCK
283 bdrv_can_set_read_only(BlockDriverState
*bs
, bool read_only
,
284 bool ignore_allow_rdw
, Error
**errp
)
288 /* Do not set read_only if copy_on_read is enabled */
289 if (bs
->copy_on_read
&& read_only
) {
290 error_setg(errp
, "Can't set node '%s' to r/o with copy-on-read enabled",
291 bdrv_get_device_or_node_name(bs
));
295 /* Do not clear read_only if it is prohibited */
296 if (!read_only
&& !(bs
->open_flags
& BDRV_O_ALLOW_RDWR
) &&
299 error_setg(errp
, "Node '%s' is read only",
300 bdrv_get_device_or_node_name(bs
));
308 * Called by a driver that can only provide a read-only image.
310 * Returns 0 if the node is already read-only or it could switch the node to
311 * read-only because BDRV_O_AUTO_RDONLY is set.
313 * Returns -EACCES if the node is read-write and BDRV_O_AUTO_RDONLY is not set
314 * or bdrv_can_set_read_only() forbids making the node read-only. If @errmsg
315 * is not NULL, it is used as the error message for the Error object.
317 int bdrv_apply_auto_read_only(BlockDriverState
*bs
, const char *errmsg
,
323 if (!(bs
->open_flags
& BDRV_O_RDWR
)) {
326 if (!(bs
->open_flags
& BDRV_O_AUTO_RDONLY
)) {
330 ret
= bdrv_can_set_read_only(bs
, true, false, NULL
);
335 bs
->open_flags
&= ~BDRV_O_RDWR
;
340 error_setg(errp
, "%s", errmsg
?: "Image is read-only");
345 * If @backing is empty, this function returns NULL without setting
346 * @errp. In all other cases, NULL will only be returned with @errp
349 * Therefore, a return value of NULL without @errp set means that
350 * there is no backing file; if @errp is set, there is one but its
351 * absolute filename cannot be generated.
353 char *bdrv_get_full_backing_filename_from_filename(const char *backed
,
357 if (backing
[0] == '\0') {
359 } else if (path_has_protocol(backing
) || path_is_absolute(backing
)) {
360 return g_strdup(backing
);
361 } else if (backed
[0] == '\0' || strstart(backed
, "json:", NULL
)) {
362 error_setg(errp
, "Cannot use relative backing file names for '%s'",
366 return path_combine(backed
, backing
);
371 * If @filename is empty or NULL, this function returns NULL without
372 * setting @errp. In all other cases, NULL will only be returned with
375 static char * GRAPH_RDLOCK
376 bdrv_make_absolute_filename(BlockDriverState
*relative_to
,
377 const char *filename
, Error
**errp
)
379 char *dir
, *full_name
;
381 if (!filename
|| filename
[0] == '\0') {
383 } else if (path_has_protocol(filename
) || path_is_absolute(filename
)) {
384 return g_strdup(filename
);
387 dir
= bdrv_dirname(relative_to
, errp
);
392 full_name
= g_strconcat(dir
, filename
, NULL
);
397 char *bdrv_get_full_backing_filename(BlockDriverState
*bs
, Error
**errp
)
400 return bdrv_make_absolute_filename(bs
, bs
->backing_file
, errp
);
403 void bdrv_register(BlockDriver
*bdrv
)
405 assert(bdrv
->format_name
);
407 QLIST_INSERT_HEAD(&bdrv_drivers
, bdrv
, list
);
410 BlockDriverState
*bdrv_new(void)
412 BlockDriverState
*bs
;
417 bs
= g_new0(BlockDriverState
, 1);
418 QLIST_INIT(&bs
->dirty_bitmaps
);
419 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
420 QLIST_INIT(&bs
->op_blockers
[i
]);
422 qemu_mutex_init(&bs
->reqs_lock
);
423 qemu_mutex_init(&bs
->dirty_bitmap_mutex
);
425 bs
->aio_context
= qemu_get_aio_context();
427 qemu_co_queue_init(&bs
->flush_queue
);
429 qemu_co_mutex_init(&bs
->bsc_modify_lock
);
430 bs
->block_status_cache
= g_new0(BdrvBlockStatusCache
, 1);
432 for (i
= 0; i
< bdrv_drain_all_count
; i
++) {
433 bdrv_drained_begin(bs
);
436 QTAILQ_INSERT_TAIL(&all_bdrv_states
, bs
, bs_list
);
441 static BlockDriver
*bdrv_do_find_format(const char *format_name
)
446 QLIST_FOREACH(drv1
, &bdrv_drivers
, list
) {
447 if (!strcmp(drv1
->format_name
, format_name
)) {
455 BlockDriver
*bdrv_find_format(const char *format_name
)
462 drv1
= bdrv_do_find_format(format_name
);
467 /* The driver isn't registered, maybe we need to load a module */
468 for (i
= 0; i
< (int)ARRAY_SIZE(block_driver_modules
); ++i
) {
469 if (!strcmp(block_driver_modules
[i
].format_name
, format_name
)) {
470 Error
*local_err
= NULL
;
471 int rv
= block_module_load(block_driver_modules
[i
].library_name
,
474 return bdrv_do_find_format(format_name
);
476 error_report_err(local_err
);
484 static int bdrv_format_is_whitelisted(const char *format_name
, bool read_only
)
486 static const char *whitelist_rw
[] = {
487 CONFIG_BDRV_RW_WHITELIST
490 static const char *whitelist_ro
[] = {
491 CONFIG_BDRV_RO_WHITELIST
496 if (!whitelist_rw
[0] && !whitelist_ro
[0]) {
497 return 1; /* no whitelist, anything goes */
500 for (p
= whitelist_rw
; *p
; p
++) {
501 if (!strcmp(format_name
, *p
)) {
506 for (p
= whitelist_ro
; *p
; p
++) {
507 if (!strcmp(format_name
, *p
)) {
515 int bdrv_is_whitelisted(BlockDriver
*drv
, bool read_only
)
518 return bdrv_format_is_whitelisted(drv
->format_name
, read_only
);
521 bool bdrv_uses_whitelist(void)
523 return use_bdrv_whitelist
;
526 typedef struct CreateCo
{
534 int coroutine_fn
bdrv_co_create(BlockDriver
*drv
, const char *filename
,
535 QemuOpts
*opts
, Error
**errp
)
541 if (!drv
->bdrv_co_create_opts
) {
542 error_setg(errp
, "Driver '%s' does not support image creation",
547 ret
= drv
->bdrv_co_create_opts(drv
, filename
, opts
, errp
);
548 if (ret
< 0 && !*errp
) {
549 error_setg_errno(errp
, -ret
, "Could not create image");
556 * Helper function for bdrv_create_file_fallback(): Resize @blk to at
557 * least the given @minimum_size.
559 * On success, return @blk's actual length.
560 * Otherwise, return -errno.
562 static int64_t coroutine_fn GRAPH_UNLOCKED
563 create_file_fallback_truncate(BlockBackend
*blk
, int64_t minimum_size
,
566 Error
*local_err
= NULL
;
572 ret
= blk_co_truncate(blk
, minimum_size
, false, PREALLOC_MODE_OFF
, 0,
574 if (ret
< 0 && ret
!= -ENOTSUP
) {
575 error_propagate(errp
, local_err
);
579 size
= blk_co_getlength(blk
);
581 error_free(local_err
);
582 error_setg_errno(errp
, -size
,
583 "Failed to inquire the new image file's length");
587 if (size
< minimum_size
) {
588 /* Need to grow the image, but we failed to do that */
589 error_propagate(errp
, local_err
);
593 error_free(local_err
);
600 * Helper function for bdrv_create_file_fallback(): Zero the first
601 * sector to remove any potentially pre-existing image header.
603 static int coroutine_fn
604 create_file_fallback_zero_first_sector(BlockBackend
*blk
,
605 int64_t current_size
,
608 int64_t bytes_to_clear
;
613 bytes_to_clear
= MIN(current_size
, BDRV_SECTOR_SIZE
);
614 if (bytes_to_clear
) {
615 ret
= blk_co_pwrite_zeroes(blk
, 0, bytes_to_clear
, BDRV_REQ_MAY_UNMAP
);
617 error_setg_errno(errp
, -ret
,
618 "Failed to clear the new image's first sector");
627 * Simple implementation of bdrv_co_create_opts for protocol drivers
628 * which only support creation via opening a file
629 * (usually existing raw storage device)
631 int coroutine_fn
bdrv_co_create_opts_simple(BlockDriver
*drv
,
632 const char *filename
,
640 PreallocMode prealloc
;
641 Error
*local_err
= NULL
;
646 size
= qemu_opt_get_size_del(opts
, BLOCK_OPT_SIZE
, 0);
647 buf
= qemu_opt_get_del(opts
, BLOCK_OPT_PREALLOC
);
648 prealloc
= qapi_enum_parse(&PreallocMode_lookup
, buf
,
649 PREALLOC_MODE_OFF
, &local_err
);
652 error_propagate(errp
, local_err
);
656 if (prealloc
!= PREALLOC_MODE_OFF
) {
657 error_setg(errp
, "Unsupported preallocation mode '%s'",
658 PreallocMode_str(prealloc
));
662 options
= qdict_new();
663 qdict_put_str(options
, "driver", drv
->format_name
);
665 blk
= blk_co_new_open(filename
, NULL
, options
,
666 BDRV_O_RDWR
| BDRV_O_RESIZE
, errp
);
668 error_prepend(errp
, "Protocol driver '%s' does not support creating "
669 "new images, so an existing image must be selected as "
670 "the target; however, opening the given target as an "
671 "existing image failed: ",
676 size
= create_file_fallback_truncate(blk
, size
, errp
);
682 ret
= create_file_fallback_zero_first_sector(blk
, size
, errp
);
693 int coroutine_fn
bdrv_co_create_file(const char *filename
, QemuOpts
*opts
,
696 QemuOpts
*protocol_opts
;
703 drv
= bdrv_find_protocol(filename
, true, errp
);
708 if (!drv
->create_opts
) {
709 error_setg(errp
, "Driver '%s' does not support image creation",
715 * 'opts' contains a QemuOptsList with a combination of format and protocol
718 * The format properly removes its options, but the default values remain
719 * in 'opts->list'. So if the protocol has options with the same name
720 * (e.g. rbd has 'cluster_size' as qcow2), it will see the default values
721 * of the format, since for overlapping options, the format wins.
723 * To avoid this issue, lets convert QemuOpts to QDict, in this way we take
724 * only the set options, and then convert it back to QemuOpts, using the
725 * create_opts of the protocol. So the new QemuOpts, will contain only the
728 qdict
= qemu_opts_to_qdict(opts
, NULL
);
729 protocol_opts
= qemu_opts_from_qdict(drv
->create_opts
, qdict
, errp
);
730 if (protocol_opts
== NULL
) {
735 ret
= bdrv_co_create(drv
, filename
, protocol_opts
, errp
);
737 qemu_opts_del(protocol_opts
);
738 qobject_unref(qdict
);
742 int coroutine_fn
bdrv_co_delete_file(BlockDriverState
*bs
, Error
**errp
)
744 Error
*local_err
= NULL
;
749 assert_bdrv_graph_readable();
752 error_setg(errp
, "Block node '%s' is not opened", bs
->filename
);
756 if (!bs
->drv
->bdrv_co_delete_file
) {
757 error_setg(errp
, "Driver '%s' does not support image deletion",
758 bs
->drv
->format_name
);
762 ret
= bs
->drv
->bdrv_co_delete_file(bs
, &local_err
);
764 error_propagate(errp
, local_err
);
770 void coroutine_fn
bdrv_co_delete_file_noerr(BlockDriverState
*bs
)
772 Error
*local_err
= NULL
;
780 ret
= bdrv_co_delete_file(bs
, &local_err
);
782 * ENOTSUP will happen if the block driver doesn't support
783 * the 'bdrv_co_delete_file' interface. This is a predictable
784 * scenario and shouldn't be reported back to the user.
786 if (ret
== -ENOTSUP
) {
787 error_free(local_err
);
788 } else if (ret
< 0) {
789 error_report_err(local_err
);
794 * Try to get @bs's logical and physical block size.
795 * On success, store them in @bsz struct and return 0.
796 * On failure return -errno.
797 * @bs must not be empty.
799 int bdrv_probe_blocksizes(BlockDriverState
*bs
, BlockSizes
*bsz
)
801 BlockDriver
*drv
= bs
->drv
;
802 BlockDriverState
*filtered
= bdrv_filter_bs(bs
);
805 if (drv
&& drv
->bdrv_probe_blocksizes
) {
806 return drv
->bdrv_probe_blocksizes(bs
, bsz
);
807 } else if (filtered
) {
808 return bdrv_probe_blocksizes(filtered
, bsz
);
815 * Try to get @bs's geometry (cyls, heads, sectors).
816 * On success, store them in @geo struct and return 0.
817 * On failure return -errno.
818 * @bs must not be empty.
820 int bdrv_probe_geometry(BlockDriverState
*bs
, HDGeometry
*geo
)
822 BlockDriver
*drv
= bs
->drv
;
823 BlockDriverState
*filtered
;
826 GRAPH_RDLOCK_GUARD_MAINLOOP();
828 if (drv
&& drv
->bdrv_probe_geometry
) {
829 return drv
->bdrv_probe_geometry(bs
, geo
);
832 filtered
= bdrv_filter_bs(bs
);
834 return bdrv_probe_geometry(filtered
, geo
);
841 * Create a uniquely-named empty temporary file.
842 * Return the actual file name used upon success, otherwise NULL.
843 * This string should be freed with g_free() when not needed any longer.
845 * Note: creating a temporary file for the caller to (re)open is
846 * inherently racy. Use g_file_open_tmp() instead whenever practical.
848 char *create_tmp_file(Error
**errp
)
852 g_autofree
char *filename
= NULL
;
854 tmpdir
= g_get_tmp_dir();
857 * See commit 69bef79 ("block: use /var/tmp instead of /tmp for -snapshot")
859 * This function is used to create temporary disk images (like -snapshot),
860 * so the files can become very large. /tmp is often a tmpfs where as
861 * /var/tmp is usually on a disk, so more appropriate for disk images.
863 if (!g_strcmp0(tmpdir
, "/tmp")) {
868 filename
= g_strdup_printf("%s/vl.XXXXXX", tmpdir
);
869 fd
= g_mkstemp(filename
);
871 error_setg_errno(errp
, errno
, "Could not open temporary file '%s'",
877 return g_steal_pointer(&filename
);
881 * Detect host devices. By convention, /dev/cdrom[N] is always
882 * recognized as a host CDROM.
884 static BlockDriver
*find_hdev_driver(const char *filename
)
886 int score_max
= 0, score
;
887 BlockDriver
*drv
= NULL
, *d
;
890 QLIST_FOREACH(d
, &bdrv_drivers
, list
) {
891 if (d
->bdrv_probe_device
) {
892 score
= d
->bdrv_probe_device(filename
);
893 if (score
> score_max
) {
903 static BlockDriver
*bdrv_do_find_protocol(const char *protocol
)
908 QLIST_FOREACH(drv1
, &bdrv_drivers
, list
) {
909 if (drv1
->protocol_name
&& !strcmp(drv1
->protocol_name
, protocol
)) {
917 BlockDriver
*bdrv_find_protocol(const char *filename
,
918 bool allow_protocol_prefix
,
928 /* TODO Drivers without bdrv_file_open must be specified explicitly */
931 * XXX(hch): we really should not let host device detection
932 * override an explicit protocol specification, but moving this
933 * later breaks access to device names with colons in them.
934 * Thanks to the brain-dead persistent naming schemes on udev-
935 * based Linux systems those actually are quite common.
937 drv1
= find_hdev_driver(filename
);
942 if (!path_has_protocol(filename
) || !allow_protocol_prefix
) {
946 p
= strchr(filename
, ':');
949 if (len
> sizeof(protocol
) - 1)
950 len
= sizeof(protocol
) - 1;
951 memcpy(protocol
, filename
, len
);
952 protocol
[len
] = '\0';
954 drv1
= bdrv_do_find_protocol(protocol
);
959 for (i
= 0; i
< (int)ARRAY_SIZE(block_driver_modules
); ++i
) {
960 if (block_driver_modules
[i
].protocol_name
&&
961 !strcmp(block_driver_modules
[i
].protocol_name
, protocol
)) {
962 int rv
= block_module_load(block_driver_modules
[i
].library_name
, errp
);
964 drv1
= bdrv_do_find_protocol(protocol
);
973 error_setg(errp
, "Unknown protocol '%s'", protocol
);
979 * Guess image format by probing its contents.
980 * This is not a good idea when your image is raw (CVE-2008-2004), but
981 * we do it anyway for backward compatibility.
983 * @buf contains the image's first @buf_size bytes.
984 * @buf_size is the buffer size in bytes (generally BLOCK_PROBE_BUF_SIZE,
985 * but can be smaller if the image file is smaller)
986 * @filename is its filename.
988 * For all block drivers, call the bdrv_probe() method to get its
990 * Return the first block driver with the highest probing score.
992 BlockDriver
*bdrv_probe_all(const uint8_t *buf
, int buf_size
,
993 const char *filename
)
995 int score_max
= 0, score
;
996 BlockDriver
*drv
= NULL
, *d
;
999 QLIST_FOREACH(d
, &bdrv_drivers
, list
) {
1000 if (d
->bdrv_probe
) {
1001 score
= d
->bdrv_probe(buf
, buf_size
, filename
);
1002 if (score
> score_max
) {
1012 static int find_image_format(BlockBackend
*file
, const char *filename
,
1013 BlockDriver
**pdrv
, Error
**errp
)
1016 uint8_t buf
[BLOCK_PROBE_BUF_SIZE
];
1019 GLOBAL_STATE_CODE();
1021 /* Return the raw BlockDriver * to scsi-generic devices or empty drives */
1022 if (blk_is_sg(file
) || !blk_is_inserted(file
) || blk_getlength(file
) == 0) {
1027 ret
= blk_pread(file
, 0, sizeof(buf
), buf
, 0);
1029 error_setg_errno(errp
, -ret
, "Could not read image for determining its "
1035 drv
= bdrv_probe_all(buf
, sizeof(buf
), filename
);
1037 error_setg(errp
, "Could not determine image format: No compatible "
1048 * Set the current 'total_sectors' value
1049 * Return 0 on success, -errno on error.
1051 int coroutine_fn
bdrv_co_refresh_total_sectors(BlockDriverState
*bs
,
1054 BlockDriver
*drv
= bs
->drv
;
1056 assert_bdrv_graph_readable();
1062 /* Do not attempt drv->bdrv_co_getlength() on scsi-generic devices */
1066 /* query actual device if possible, otherwise just trust the hint */
1067 if (drv
->bdrv_co_getlength
) {
1068 int64_t length
= drv
->bdrv_co_getlength(bs
);
1072 hint
= DIV_ROUND_UP(length
, BDRV_SECTOR_SIZE
);
1075 bs
->total_sectors
= hint
;
1077 if (bs
->total_sectors
* BDRV_SECTOR_SIZE
> BDRV_MAX_LENGTH
) {
1085 * Combines a QDict of new block driver @options with any missing options taken
1086 * from @old_options, so that leaving out an option defaults to its old value.
1088 static void bdrv_join_options(BlockDriverState
*bs
, QDict
*options
,
1091 GLOBAL_STATE_CODE();
1092 if (bs
->drv
&& bs
->drv
->bdrv_join_options
) {
1093 bs
->drv
->bdrv_join_options(options
, old_options
);
1095 qdict_join(options
, old_options
, false);
1099 static BlockdevDetectZeroesOptions
bdrv_parse_detect_zeroes(QemuOpts
*opts
,
1103 Error
*local_err
= NULL
;
1104 char *value
= qemu_opt_get_del(opts
, "detect-zeroes");
1105 BlockdevDetectZeroesOptions detect_zeroes
=
1106 qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup
, value
,
1107 BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF
, &local_err
);
1108 GLOBAL_STATE_CODE();
1111 error_propagate(errp
, local_err
);
1112 return detect_zeroes
;
1115 if (detect_zeroes
== BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP
&&
1116 !(open_flags
& BDRV_O_UNMAP
))
1118 error_setg(errp
, "setting detect-zeroes to unmap is not allowed "
1119 "without setting discard operation to unmap");
1122 return detect_zeroes
;
1126 * Set open flags for aio engine
1128 * Return 0 on success, -1 if the engine specified is invalid
1130 int bdrv_parse_aio(const char *mode
, int *flags
)
1132 if (!strcmp(mode
, "threads")) {
1133 /* do nothing, default */
1134 } else if (!strcmp(mode
, "native")) {
1135 *flags
|= BDRV_O_NATIVE_AIO
;
1136 #ifdef CONFIG_LINUX_IO_URING
1137 } else if (!strcmp(mode
, "io_uring")) {
1138 *flags
|= BDRV_O_IO_URING
;
1148 * Set open flags for a given discard mode
1150 * Return 0 on success, -1 if the discard mode was invalid.
1152 int bdrv_parse_discard_flags(const char *mode
, int *flags
)
1154 *flags
&= ~BDRV_O_UNMAP
;
1156 if (!strcmp(mode
, "off") || !strcmp(mode
, "ignore")) {
1158 } else if (!strcmp(mode
, "on") || !strcmp(mode
, "unmap")) {
1159 *flags
|= BDRV_O_UNMAP
;
1168 * Set open flags for a given cache mode
1170 * Return 0 on success, -1 if the cache mode was invalid.
1172 int bdrv_parse_cache_mode(const char *mode
, int *flags
, bool *writethrough
)
1174 *flags
&= ~BDRV_O_CACHE_MASK
;
1176 if (!strcmp(mode
, "off") || !strcmp(mode
, "none")) {
1177 *writethrough
= false;
1178 *flags
|= BDRV_O_NOCACHE
;
1179 } else if (!strcmp(mode
, "directsync")) {
1180 *writethrough
= true;
1181 *flags
|= BDRV_O_NOCACHE
;
1182 } else if (!strcmp(mode
, "writeback")) {
1183 *writethrough
= false;
1184 } else if (!strcmp(mode
, "unsafe")) {
1185 *writethrough
= false;
1186 *flags
|= BDRV_O_NO_FLUSH
;
1187 } else if (!strcmp(mode
, "writethrough")) {
1188 *writethrough
= true;
1196 static char *bdrv_child_get_parent_desc(BdrvChild
*c
)
1198 BlockDriverState
*parent
= c
->opaque
;
1199 return g_strdup_printf("node '%s'", bdrv_get_node_name(parent
));
1202 static void GRAPH_RDLOCK
bdrv_child_cb_drained_begin(BdrvChild
*child
)
1204 BlockDriverState
*bs
= child
->opaque
;
1205 bdrv_do_drained_begin_quiesce(bs
, NULL
);
1208 static bool GRAPH_RDLOCK
bdrv_child_cb_drained_poll(BdrvChild
*child
)
1210 BlockDriverState
*bs
= child
->opaque
;
1211 return bdrv_drain_poll(bs
, NULL
, false);
1214 static void GRAPH_RDLOCK
bdrv_child_cb_drained_end(BdrvChild
*child
)
1216 BlockDriverState
*bs
= child
->opaque
;
1217 bdrv_drained_end(bs
);
1220 static int bdrv_child_cb_inactivate(BdrvChild
*child
)
1222 BlockDriverState
*bs
= child
->opaque
;
1223 GLOBAL_STATE_CODE();
1224 assert(bs
->open_flags
& BDRV_O_INACTIVE
);
1228 static bool bdrv_child_cb_change_aio_ctx(BdrvChild
*child
, AioContext
*ctx
,
1229 GHashTable
*visited
, Transaction
*tran
,
1232 BlockDriverState
*bs
= child
->opaque
;
1233 return bdrv_change_aio_context(bs
, ctx
, visited
, tran
, errp
);
1237 * Returns the options and flags that a temporary snapshot should get, based on
1238 * the originally requested flags (the originally requested image will have
1239 * flags like a backing file)
1241 static void bdrv_temp_snapshot_options(int *child_flags
, QDict
*child_options
,
1242 int parent_flags
, QDict
*parent_options
)
1244 GLOBAL_STATE_CODE();
1245 *child_flags
= (parent_flags
& ~BDRV_O_SNAPSHOT
) | BDRV_O_TEMPORARY
;
1247 /* For temporary files, unconditional cache=unsafe is fine */
1248 qdict_set_default_str(child_options
, BDRV_OPT_CACHE_DIRECT
, "off");
1249 qdict_set_default_str(child_options
, BDRV_OPT_CACHE_NO_FLUSH
, "on");
1251 /* Copy the read-only and discard options from the parent */
1252 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_READ_ONLY
);
1253 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_DISCARD
);
1255 /* aio=native doesn't work for cache.direct=off, so disable it for the
1256 * temporary snapshot */
1257 *child_flags
&= ~BDRV_O_NATIVE_AIO
;
1260 static void GRAPH_WRLOCK
bdrv_backing_attach(BdrvChild
*c
)
1262 BlockDriverState
*parent
= c
->opaque
;
1263 BlockDriverState
*backing_hd
= c
->bs
;
1265 GLOBAL_STATE_CODE();
1266 assert(!parent
->backing_blocker
);
1267 error_setg(&parent
->backing_blocker
,
1268 "node is used as backing hd of '%s'",
1269 bdrv_get_device_or_node_name(parent
));
1271 bdrv_refresh_filename(backing_hd
);
1273 parent
->open_flags
&= ~BDRV_O_NO_BACKING
;
1275 bdrv_op_block_all(backing_hd
, parent
->backing_blocker
);
1276 /* Otherwise we won't be able to commit or stream */
1277 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_COMMIT_TARGET
,
1278 parent
->backing_blocker
);
1279 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_STREAM
,
1280 parent
->backing_blocker
);
1282 * We do backup in 3 ways:
1284 * The target bs is new opened, and the source is top BDS
1285 * 2. blockdev backup
1286 * Both the source and the target are top BDSes.
1287 * 3. internal backup(used for block replication)
1288 * Both the source and the target are backing file
1290 * In case 1 and 2, neither the source nor the target is the backing file.
1291 * In case 3, we will block the top BDS, so there is only one block job
1292 * for the top BDS and its backing chain.
1294 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_BACKUP_SOURCE
,
1295 parent
->backing_blocker
);
1296 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_BACKUP_TARGET
,
1297 parent
->backing_blocker
);
1300 static void bdrv_backing_detach(BdrvChild
*c
)
1302 BlockDriverState
*parent
= c
->opaque
;
1304 GLOBAL_STATE_CODE();
1305 assert(parent
->backing_blocker
);
1306 bdrv_op_unblock_all(c
->bs
, parent
->backing_blocker
);
1307 error_free(parent
->backing_blocker
);
1308 parent
->backing_blocker
= NULL
;
1311 static int bdrv_backing_update_filename(BdrvChild
*c
, BlockDriverState
*base
,
1312 const char *filename
, Error
**errp
)
1314 BlockDriverState
*parent
= c
->opaque
;
1315 bool read_only
= bdrv_is_read_only(parent
);
1317 GLOBAL_STATE_CODE();
1320 ret
= bdrv_reopen_set_read_only(parent
, false, errp
);
1326 ret
= bdrv_change_backing_file(parent
, filename
,
1327 base
->drv
? base
->drv
->format_name
: "",
1330 error_setg_errno(errp
, -ret
, "Could not update backing file link");
1334 bdrv_reopen_set_read_only(parent
, true, NULL
);
1341 * Returns the options and flags that a generic child of a BDS should
1342 * get, based on the given options and flags for the parent BDS.
1344 static void bdrv_inherited_options(BdrvChildRole role
, bool parent_is_format
,
1345 int *child_flags
, QDict
*child_options
,
1346 int parent_flags
, QDict
*parent_options
)
1348 int flags
= parent_flags
;
1349 GLOBAL_STATE_CODE();
1352 * First, decide whether to set, clear, or leave BDRV_O_PROTOCOL.
1353 * Generally, the question to answer is: Should this child be
1354 * format-probed by default?
1358 * Pure and non-filtered data children of non-format nodes should
1359 * be probed by default (even when the node itself has BDRV_O_PROTOCOL
1360 * set). This only affects a very limited set of drivers (namely
1361 * quorum and blkverify when this comment was written).
1362 * Force-clear BDRV_O_PROTOCOL then.
1364 if (!parent_is_format
&&
1365 (role
& BDRV_CHILD_DATA
) &&
1366 !(role
& (BDRV_CHILD_METADATA
| BDRV_CHILD_FILTERED
)))
1368 flags
&= ~BDRV_O_PROTOCOL
;
1372 * All children of format nodes (except for COW children) and all
1373 * metadata children in general should never be format-probed.
1374 * Force-set BDRV_O_PROTOCOL then.
1376 if ((parent_is_format
&& !(role
& BDRV_CHILD_COW
)) ||
1377 (role
& BDRV_CHILD_METADATA
))
1379 flags
|= BDRV_O_PROTOCOL
;
1383 * If the cache mode isn't explicitly set, inherit direct and no-flush from
1386 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_CACHE_DIRECT
);
1387 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_CACHE_NO_FLUSH
);
1388 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_FORCE_SHARE
);
1390 if (role
& BDRV_CHILD_COW
) {
1391 /* backing files are opened read-only by default */
1392 qdict_set_default_str(child_options
, BDRV_OPT_READ_ONLY
, "on");
1393 qdict_set_default_str(child_options
, BDRV_OPT_AUTO_READ_ONLY
, "off");
1395 /* Inherit the read-only option from the parent if it's not set */
1396 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_READ_ONLY
);
1397 qdict_copy_default(child_options
, parent_options
,
1398 BDRV_OPT_AUTO_READ_ONLY
);
1402 * bdrv_co_pdiscard() respects unmap policy for the parent, so we
1403 * can default to enable it on lower layers regardless of the
1406 qdict_set_default_str(child_options
, BDRV_OPT_DISCARD
, "unmap");
1408 /* Clear flags that only apply to the top layer */
1409 flags
&= ~(BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
| BDRV_O_COPY_ON_READ
);
1411 if (role
& BDRV_CHILD_METADATA
) {
1412 flags
&= ~BDRV_O_NO_IO
;
1414 if (role
& BDRV_CHILD_COW
) {
1415 flags
&= ~BDRV_O_TEMPORARY
;
1418 *child_flags
= flags
;
1421 static void GRAPH_WRLOCK
bdrv_child_cb_attach(BdrvChild
*child
)
1423 BlockDriverState
*bs
= child
->opaque
;
1425 assert_bdrv_graph_writable();
1426 QLIST_INSERT_HEAD(&bs
->children
, child
, next
);
1427 if (bs
->drv
->is_filter
|| (child
->role
& BDRV_CHILD_FILTERED
)) {
1429 * Here we handle filters and block/raw-format.c when it behave like
1430 * filter. They generally have a single PRIMARY child, which is also the
1431 * FILTERED child, and that they may have multiple more children, which
1432 * are neither PRIMARY nor FILTERED. And never we have a COW child here.
1433 * So bs->file will be the PRIMARY child, unless the PRIMARY child goes
1434 * into bs->backing on exceptional cases; and bs->backing will be
1437 assert(!(child
->role
& BDRV_CHILD_COW
));
1438 if (child
->role
& BDRV_CHILD_PRIMARY
) {
1439 assert(child
->role
& BDRV_CHILD_FILTERED
);
1440 assert(!bs
->backing
);
1443 if (bs
->drv
->filtered_child_is_backing
) {
1444 bs
->backing
= child
;
1449 assert(!(child
->role
& BDRV_CHILD_FILTERED
));
1451 } else if (child
->role
& BDRV_CHILD_COW
) {
1452 assert(bs
->drv
->supports_backing
);
1453 assert(!(child
->role
& BDRV_CHILD_PRIMARY
));
1454 assert(!bs
->backing
);
1455 bs
->backing
= child
;
1456 bdrv_backing_attach(child
);
1457 } else if (child
->role
& BDRV_CHILD_PRIMARY
) {
1463 static void GRAPH_WRLOCK
bdrv_child_cb_detach(BdrvChild
*child
)
1465 BlockDriverState
*bs
= child
->opaque
;
1467 if (child
->role
& BDRV_CHILD_COW
) {
1468 bdrv_backing_detach(child
);
1471 assert_bdrv_graph_writable();
1472 QLIST_REMOVE(child
, next
);
1473 if (child
== bs
->backing
) {
1474 assert(child
!= bs
->file
);
1476 } else if (child
== bs
->file
) {
1481 static int bdrv_child_cb_update_filename(BdrvChild
*c
, BlockDriverState
*base
,
1482 const char *filename
, Error
**errp
)
1484 if (c
->role
& BDRV_CHILD_COW
) {
1485 return bdrv_backing_update_filename(c
, base
, filename
, errp
);
1490 AioContext
*child_of_bds_get_parent_aio_context(BdrvChild
*c
)
1492 BlockDriverState
*bs
= c
->opaque
;
1495 return bdrv_get_aio_context(bs
);
1498 const BdrvChildClass child_of_bds
= {
1499 .parent_is_bds
= true,
1500 .get_parent_desc
= bdrv_child_get_parent_desc
,
1501 .inherit_options
= bdrv_inherited_options
,
1502 .drained_begin
= bdrv_child_cb_drained_begin
,
1503 .drained_poll
= bdrv_child_cb_drained_poll
,
1504 .drained_end
= bdrv_child_cb_drained_end
,
1505 .attach
= bdrv_child_cb_attach
,
1506 .detach
= bdrv_child_cb_detach
,
1507 .inactivate
= bdrv_child_cb_inactivate
,
1508 .change_aio_ctx
= bdrv_child_cb_change_aio_ctx
,
1509 .update_filename
= bdrv_child_cb_update_filename
,
1510 .get_parent_aio_context
= child_of_bds_get_parent_aio_context
,
1513 AioContext
*bdrv_child_get_parent_aio_context(BdrvChild
*c
)
1516 return c
->klass
->get_parent_aio_context(c
);
1519 static int bdrv_open_flags(BlockDriverState
*bs
, int flags
)
1521 int open_flags
= flags
;
1522 GLOBAL_STATE_CODE();
1525 * Clear flags that are internal to the block layer before opening the
1528 open_flags
&= ~(BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
| BDRV_O_PROTOCOL
);
1533 static void update_flags_from_options(int *flags
, QemuOpts
*opts
)
1535 GLOBAL_STATE_CODE();
1537 *flags
&= ~(BDRV_O_CACHE_MASK
| BDRV_O_RDWR
| BDRV_O_AUTO_RDONLY
);
1539 if (qemu_opt_get_bool_del(opts
, BDRV_OPT_CACHE_NO_FLUSH
, false)) {
1540 *flags
|= BDRV_O_NO_FLUSH
;
1543 if (qemu_opt_get_bool_del(opts
, BDRV_OPT_CACHE_DIRECT
, false)) {
1544 *flags
|= BDRV_O_NOCACHE
;
1547 if (!qemu_opt_get_bool_del(opts
, BDRV_OPT_READ_ONLY
, false)) {
1548 *flags
|= BDRV_O_RDWR
;
1551 if (qemu_opt_get_bool_del(opts
, BDRV_OPT_AUTO_READ_ONLY
, false)) {
1552 *flags
|= BDRV_O_AUTO_RDONLY
;
1556 static void update_options_from_flags(QDict
*options
, int flags
)
1558 GLOBAL_STATE_CODE();
1559 if (!qdict_haskey(options
, BDRV_OPT_CACHE_DIRECT
)) {
1560 qdict_put_bool(options
, BDRV_OPT_CACHE_DIRECT
, flags
& BDRV_O_NOCACHE
);
1562 if (!qdict_haskey(options
, BDRV_OPT_CACHE_NO_FLUSH
)) {
1563 qdict_put_bool(options
, BDRV_OPT_CACHE_NO_FLUSH
,
1564 flags
& BDRV_O_NO_FLUSH
);
1566 if (!qdict_haskey(options
, BDRV_OPT_READ_ONLY
)) {
1567 qdict_put_bool(options
, BDRV_OPT_READ_ONLY
, !(flags
& BDRV_O_RDWR
));
1569 if (!qdict_haskey(options
, BDRV_OPT_AUTO_READ_ONLY
)) {
1570 qdict_put_bool(options
, BDRV_OPT_AUTO_READ_ONLY
,
1571 flags
& BDRV_O_AUTO_RDONLY
);
1575 static void bdrv_assign_node_name(BlockDriverState
*bs
,
1576 const char *node_name
,
1579 char *gen_node_name
= NULL
;
1580 GLOBAL_STATE_CODE();
1583 node_name
= gen_node_name
= id_generate(ID_BLOCK
);
1584 } else if (!id_wellformed(node_name
)) {
1586 * Check for empty string or invalid characters, but not if it is
1587 * generated (generated names use characters not available to the user)
1589 error_setg(errp
, "Invalid node-name: '%s'", node_name
);
1593 /* takes care of avoiding namespaces collisions */
1594 if (blk_by_name(node_name
)) {
1595 error_setg(errp
, "node-name=%s is conflicting with a device id",
1600 /* takes care of avoiding duplicates node names */
1601 if (bdrv_find_node(node_name
)) {
1602 error_setg(errp
, "Duplicate nodes with node-name='%s'", node_name
);
1606 /* Make sure that the node name isn't truncated */
1607 if (strlen(node_name
) >= sizeof(bs
->node_name
)) {
1608 error_setg(errp
, "Node name too long");
1612 /* copy node name into the bs and insert it into the graph list */
1613 pstrcpy(bs
->node_name
, sizeof(bs
->node_name
), node_name
);
1614 QTAILQ_INSERT_TAIL(&graph_bdrv_states
, bs
, node_list
);
1616 g_free(gen_node_name
);
1620 * The caller must always hold @bs AioContext lock, because this function calls
1621 * bdrv_refresh_total_sectors() which polls when called from non-coroutine
1624 static int no_coroutine_fn GRAPH_UNLOCKED
1625 bdrv_open_driver(BlockDriverState
*bs
, BlockDriver
*drv
, const char *node_name
,
1626 QDict
*options
, int open_flags
, Error
**errp
)
1629 Error
*local_err
= NULL
;
1631 GLOBAL_STATE_CODE();
1633 bdrv_assign_node_name(bs
, node_name
, &local_err
);
1635 error_propagate(errp
, local_err
);
1640 bs
->opaque
= g_malloc0(drv
->instance_size
);
1642 if (drv
->bdrv_file_open
) {
1643 assert(!drv
->bdrv_needs_filename
|| bs
->filename
[0]);
1644 ret
= drv
->bdrv_file_open(bs
, options
, open_flags
, &local_err
);
1645 } else if (drv
->bdrv_open
) {
1646 ret
= drv
->bdrv_open(bs
, options
, open_flags
, &local_err
);
1653 error_propagate(errp
, local_err
);
1654 } else if (bs
->filename
[0]) {
1655 error_setg_errno(errp
, -ret
, "Could not open '%s'", bs
->filename
);
1657 error_setg_errno(errp
, -ret
, "Could not open image");
1662 assert(!(bs
->supported_read_flags
& ~BDRV_REQ_MASK
));
1663 assert(!(bs
->supported_write_flags
& ~BDRV_REQ_MASK
));
1666 * Always allow the BDRV_REQ_REGISTERED_BUF optimization hint. This saves
1667 * drivers that pass read/write requests through to a child the trouble of
1668 * declaring support explicitly.
1670 * Drivers must not propagate this flag accidentally when they initiate I/O
1671 * to a bounce buffer. That case should be rare though.
1673 bs
->supported_read_flags
|= BDRV_REQ_REGISTERED_BUF
;
1674 bs
->supported_write_flags
|= BDRV_REQ_REGISTERED_BUF
;
1676 /* Get the context after .bdrv_open, it can change the context */
1677 ctx
= bdrv_get_aio_context(bs
);
1678 aio_context_acquire(ctx
);
1680 ret
= bdrv_refresh_total_sectors(bs
, bs
->total_sectors
);
1682 error_setg_errno(errp
, -ret
, "Could not refresh total sector count");
1683 aio_context_release(ctx
);
1687 bdrv_graph_rdlock_main_loop();
1688 bdrv_refresh_limits(bs
, NULL
, &local_err
);
1689 bdrv_graph_rdunlock_main_loop();
1690 aio_context_release(ctx
);
1693 error_propagate(errp
, local_err
);
1697 assert(bdrv_opt_mem_align(bs
) != 0);
1698 assert(bdrv_min_mem_align(bs
) != 0);
1699 assert(is_power_of_2(bs
->bl
.request_alignment
));
1701 for (i
= 0; i
< bs
->quiesce_counter
; i
++) {
1702 if (drv
->bdrv_drain_begin
) {
1703 drv
->bdrv_drain_begin(bs
);
1710 if (bs
->file
!= NULL
) {
1711 bdrv_graph_wrlock(NULL
);
1712 bdrv_unref_child(bs
, bs
->file
);
1713 bdrv_graph_wrunlock();
1722 * Create and open a block node.
1724 * @options is a QDict of options to pass to the block drivers, or NULL for an
1725 * empty set of options. The reference to the QDict belongs to the block layer
1726 * after the call (even on failure), so if the caller intends to reuse the
1727 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
1729 BlockDriverState
*bdrv_new_open_driver_opts(BlockDriver
*drv
,
1730 const char *node_name
,
1731 QDict
*options
, int flags
,
1734 BlockDriverState
*bs
;
1737 GLOBAL_STATE_CODE();
1740 bs
->open_flags
= flags
;
1741 bs
->options
= options
?: qdict_new();
1742 bs
->explicit_options
= qdict_clone_shallow(bs
->options
);
1745 update_options_from_flags(bs
->options
, flags
);
1747 ret
= bdrv_open_driver(bs
, drv
, node_name
, bs
->options
, flags
, errp
);
1749 qobject_unref(bs
->explicit_options
);
1750 bs
->explicit_options
= NULL
;
1751 qobject_unref(bs
->options
);
1760 /* Create and open a block node. */
1761 BlockDriverState
*bdrv_new_open_driver(BlockDriver
*drv
, const char *node_name
,
1762 int flags
, Error
**errp
)
1764 GLOBAL_STATE_CODE();
1765 return bdrv_new_open_driver_opts(drv
, node_name
, NULL
, flags
, errp
);
1768 QemuOptsList bdrv_runtime_opts
= {
1769 .name
= "bdrv_common",
1770 .head
= QTAILQ_HEAD_INITIALIZER(bdrv_runtime_opts
.head
),
1773 .name
= "node-name",
1774 .type
= QEMU_OPT_STRING
,
1775 .help
= "Node name of the block device node",
1779 .type
= QEMU_OPT_STRING
,
1780 .help
= "Block driver to use for the node",
1783 .name
= BDRV_OPT_CACHE_DIRECT
,
1784 .type
= QEMU_OPT_BOOL
,
1785 .help
= "Bypass software writeback cache on the host",
1788 .name
= BDRV_OPT_CACHE_NO_FLUSH
,
1789 .type
= QEMU_OPT_BOOL
,
1790 .help
= "Ignore flush requests",
1793 .name
= BDRV_OPT_READ_ONLY
,
1794 .type
= QEMU_OPT_BOOL
,
1795 .help
= "Node is opened in read-only mode",
1798 .name
= BDRV_OPT_AUTO_READ_ONLY
,
1799 .type
= QEMU_OPT_BOOL
,
1800 .help
= "Node can become read-only if opening read-write fails",
1803 .name
= "detect-zeroes",
1804 .type
= QEMU_OPT_STRING
,
1805 .help
= "try to optimize zero writes (off, on, unmap)",
1808 .name
= BDRV_OPT_DISCARD
,
1809 .type
= QEMU_OPT_STRING
,
1810 .help
= "discard operation (ignore/off, unmap/on)",
1813 .name
= BDRV_OPT_FORCE_SHARE
,
1814 .type
= QEMU_OPT_BOOL
,
1815 .help
= "always accept other writers (default: off)",
1817 { /* end of list */ }
1821 QemuOptsList bdrv_create_opts_simple
= {
1822 .name
= "simple-create-opts",
1823 .head
= QTAILQ_HEAD_INITIALIZER(bdrv_create_opts_simple
.head
),
1826 .name
= BLOCK_OPT_SIZE
,
1827 .type
= QEMU_OPT_SIZE
,
1828 .help
= "Virtual disk size"
1831 .name
= BLOCK_OPT_PREALLOC
,
1832 .type
= QEMU_OPT_STRING
,
1833 .help
= "Preallocation mode (allowed values: off)"
1835 { /* end of list */ }
1840 * Common part for opening disk images and files
1842 * Removes all processed options from *options.
1844 static int bdrv_open_common(BlockDriverState
*bs
, BlockBackend
*file
,
1845 QDict
*options
, Error
**errp
)
1847 int ret
, open_flags
;
1848 const char *filename
;
1849 const char *driver_name
= NULL
;
1850 const char *node_name
= NULL
;
1851 const char *discard
;
1854 Error
*local_err
= NULL
;
1857 assert(bs
->file
== NULL
);
1858 assert(options
!= NULL
&& bs
->options
!= options
);
1859 GLOBAL_STATE_CODE();
1861 opts
= qemu_opts_create(&bdrv_runtime_opts
, NULL
, 0, &error_abort
);
1862 if (!qemu_opts_absorb_qdict(opts
, options
, errp
)) {
1867 update_flags_from_options(&bs
->open_flags
, opts
);
1869 driver_name
= qemu_opt_get(opts
, "driver");
1870 drv
= bdrv_find_format(driver_name
);
1871 assert(drv
!= NULL
);
1873 bs
->force_share
= qemu_opt_get_bool(opts
, BDRV_OPT_FORCE_SHARE
, false);
1875 if (bs
->force_share
&& (bs
->open_flags
& BDRV_O_RDWR
)) {
1877 BDRV_OPT_FORCE_SHARE
1878 "=on can only be used with read-only images");
1884 bdrv_graph_rdlock_main_loop();
1885 bdrv_refresh_filename(blk_bs(file
));
1886 bdrv_graph_rdunlock_main_loop();
1888 filename
= blk_bs(file
)->filename
;
1891 * Caution: while qdict_get_try_str() is fine, getting
1892 * non-string types would require more care. When @options
1893 * come from -blockdev or blockdev_add, its members are typed
1894 * according to the QAPI schema, but when they come from
1895 * -drive, they're all QString.
1897 filename
= qdict_get_try_str(options
, "filename");
1900 if (drv
->bdrv_needs_filename
&& (!filename
|| !filename
[0])) {
1901 error_setg(errp
, "The '%s' block driver requires a file name",
1907 trace_bdrv_open_common(bs
, filename
?: "", bs
->open_flags
,
1910 ro
= bdrv_is_read_only(bs
);
1912 if (use_bdrv_whitelist
&& !bdrv_is_whitelisted(drv
, ro
)) {
1913 if (!ro
&& bdrv_is_whitelisted(drv
, true)) {
1914 bdrv_graph_rdlock_main_loop();
1915 ret
= bdrv_apply_auto_read_only(bs
, NULL
, NULL
);
1916 bdrv_graph_rdunlock_main_loop();
1922 !ro
&& bdrv_is_whitelisted(drv
, true)
1923 ? "Driver '%s' can only be used for read-only devices"
1924 : "Driver '%s' is not whitelisted",
1930 /* bdrv_new() and bdrv_close() make it so */
1931 assert(qatomic_read(&bs
->copy_on_read
) == 0);
1933 if (bs
->open_flags
& BDRV_O_COPY_ON_READ
) {
1935 bdrv_enable_copy_on_read(bs
);
1937 error_setg(errp
, "Can't use copy-on-read on read-only device");
1943 discard
= qemu_opt_get(opts
, BDRV_OPT_DISCARD
);
1944 if (discard
!= NULL
) {
1945 if (bdrv_parse_discard_flags(discard
, &bs
->open_flags
) != 0) {
1946 error_setg(errp
, "Invalid discard option");
1953 bdrv_parse_detect_zeroes(opts
, bs
->open_flags
, &local_err
);
1955 error_propagate(errp
, local_err
);
1960 if (filename
!= NULL
) {
1961 pstrcpy(bs
->filename
, sizeof(bs
->filename
), filename
);
1963 bs
->filename
[0] = '\0';
1965 pstrcpy(bs
->exact_filename
, sizeof(bs
->exact_filename
), bs
->filename
);
1967 /* Open the image, either directly or using a protocol */
1968 open_flags
= bdrv_open_flags(bs
, bs
->open_flags
);
1969 node_name
= qemu_opt_get(opts
, "node-name");
1971 assert(!drv
->bdrv_file_open
|| file
== NULL
);
1972 ret
= bdrv_open_driver(bs
, drv
, node_name
, options
, open_flags
, errp
);
1977 qemu_opts_del(opts
);
1981 qemu_opts_del(opts
);
1985 static QDict
*parse_json_filename(const char *filename
, Error
**errp
)
1987 QObject
*options_obj
;
1990 GLOBAL_STATE_CODE();
1992 ret
= strstart(filename
, "json:", &filename
);
1995 options_obj
= qobject_from_json(filename
, errp
);
1997 error_prepend(errp
, "Could not parse the JSON options: ");
2001 options
= qobject_to(QDict
, options_obj
);
2003 qobject_unref(options_obj
);
2004 error_setg(errp
, "Invalid JSON object given");
2008 qdict_flatten(options
);
2013 static void parse_json_protocol(QDict
*options
, const char **pfilename
,
2016 QDict
*json_options
;
2017 Error
*local_err
= NULL
;
2018 GLOBAL_STATE_CODE();
2020 /* Parse json: pseudo-protocol */
2021 if (!*pfilename
|| !g_str_has_prefix(*pfilename
, "json:")) {
2025 json_options
= parse_json_filename(*pfilename
, &local_err
);
2027 error_propagate(errp
, local_err
);
2031 /* Options given in the filename have lower priority than options
2032 * specified directly */
2033 qdict_join(options
, json_options
, false);
2034 qobject_unref(json_options
);
2039 * Fills in default options for opening images and converts the legacy
2040 * filename/flags pair to option QDict entries.
2041 * The BDRV_O_PROTOCOL flag in *flags will be set or cleared accordingly if a
2042 * block driver has been specified explicitly.
2044 static int bdrv_fill_options(QDict
**options
, const char *filename
,
2045 int *flags
, Error
**errp
)
2047 const char *drvname
;
2048 bool protocol
= *flags
& BDRV_O_PROTOCOL
;
2049 bool parse_filename
= false;
2050 BlockDriver
*drv
= NULL
;
2051 Error
*local_err
= NULL
;
2053 GLOBAL_STATE_CODE();
2056 * Caution: while qdict_get_try_str() is fine, getting non-string
2057 * types would require more care. When @options come from
2058 * -blockdev or blockdev_add, its members are typed according to
2059 * the QAPI schema, but when they come from -drive, they're all
2062 drvname
= qdict_get_try_str(*options
, "driver");
2064 drv
= bdrv_find_format(drvname
);
2066 error_setg(errp
, "Unknown driver '%s'", drvname
);
2069 /* If the user has explicitly specified the driver, this choice should
2070 * override the BDRV_O_PROTOCOL flag */
2071 protocol
= drv
->bdrv_file_open
;
2075 *flags
|= BDRV_O_PROTOCOL
;
2077 *flags
&= ~BDRV_O_PROTOCOL
;
2080 /* Translate cache options from flags into options */
2081 update_options_from_flags(*options
, *flags
);
2083 /* Fetch the file name from the options QDict if necessary */
2084 if (protocol
&& filename
) {
2085 if (!qdict_haskey(*options
, "filename")) {
2086 qdict_put_str(*options
, "filename", filename
);
2087 parse_filename
= true;
2089 error_setg(errp
, "Can't specify 'file' and 'filename' options at "
2095 /* Find the right block driver */
2096 /* See cautionary note on accessing @options above */
2097 filename
= qdict_get_try_str(*options
, "filename");
2099 if (!drvname
&& protocol
) {
2101 drv
= bdrv_find_protocol(filename
, parse_filename
, errp
);
2106 drvname
= drv
->format_name
;
2107 qdict_put_str(*options
, "driver", drvname
);
2109 error_setg(errp
, "Must specify either driver or file");
2114 assert(drv
|| !protocol
);
2116 /* Driver-specific filename parsing */
2117 if (drv
&& drv
->bdrv_parse_filename
&& parse_filename
) {
2118 drv
->bdrv_parse_filename(filename
, *options
, &local_err
);
2120 error_propagate(errp
, local_err
);
2124 if (!drv
->bdrv_needs_filename
) {
2125 qdict_del(*options
, "filename");
2132 typedef struct BlockReopenQueueEntry
{
2134 BDRVReopenState state
;
2135 QTAILQ_ENTRY(BlockReopenQueueEntry
) entry
;
2136 } BlockReopenQueueEntry
;
2139 * Return the flags that @bs will have after the reopens in @q have
2140 * successfully completed. If @q is NULL (or @bs is not contained in @q),
2141 * return the current flags.
2143 static int bdrv_reopen_get_flags(BlockReopenQueue
*q
, BlockDriverState
*bs
)
2145 BlockReopenQueueEntry
*entry
;
2148 QTAILQ_FOREACH(entry
, q
, entry
) {
2149 if (entry
->state
.bs
== bs
) {
2150 return entry
->state
.flags
;
2155 return bs
->open_flags
;
2158 /* Returns whether the image file can be written to after the reopen queue @q
2159 * has been successfully applied, or right now if @q is NULL. */
2160 static bool bdrv_is_writable_after_reopen(BlockDriverState
*bs
,
2161 BlockReopenQueue
*q
)
2163 int flags
= bdrv_reopen_get_flags(q
, bs
);
2165 return (flags
& (BDRV_O_RDWR
| BDRV_O_INACTIVE
)) == BDRV_O_RDWR
;
2169 * Return whether the BDS can be written to. This is not necessarily
2170 * the same as !bdrv_is_read_only(bs), as inactivated images may not
2171 * be written to but do not count as read-only images.
2173 bool bdrv_is_writable(BlockDriverState
*bs
)
2176 return bdrv_is_writable_after_reopen(bs
, NULL
);
2179 static char *bdrv_child_user_desc(BdrvChild
*c
)
2181 GLOBAL_STATE_CODE();
2182 return c
->klass
->get_parent_desc(c
);
2186 * Check that @a allows everything that @b needs. @a and @b must reference same
2189 static bool bdrv_a_allow_b(BdrvChild
*a
, BdrvChild
*b
, Error
**errp
)
2191 const char *child_bs_name
;
2192 g_autofree
char *a_user
= NULL
;
2193 g_autofree
char *b_user
= NULL
;
2194 g_autofree
char *perms
= NULL
;
2197 assert(a
->bs
== b
->bs
);
2198 GLOBAL_STATE_CODE();
2200 if ((b
->perm
& a
->shared_perm
) == b
->perm
) {
2204 child_bs_name
= bdrv_get_node_name(b
->bs
);
2205 a_user
= bdrv_child_user_desc(a
);
2206 b_user
= bdrv_child_user_desc(b
);
2207 perms
= bdrv_perm_names(b
->perm
& ~a
->shared_perm
);
2209 error_setg(errp
, "Permission conflict on node '%s': permissions '%s' are "
2210 "both required by %s (uses node '%s' as '%s' child) and "
2211 "unshared by %s (uses node '%s' as '%s' child).",
2212 child_bs_name
, perms
,
2213 b_user
, child_bs_name
, b
->name
,
2214 a_user
, child_bs_name
, a
->name
);
2219 static bool GRAPH_RDLOCK
2220 bdrv_parent_perms_conflict(BlockDriverState
*bs
, Error
**errp
)
2223 GLOBAL_STATE_CODE();
2226 * During the loop we'll look at each pair twice. That's correct because
2227 * bdrv_a_allow_b() is asymmetric and we should check each pair in both
2230 QLIST_FOREACH(a
, &bs
->parents
, next_parent
) {
2231 QLIST_FOREACH(b
, &bs
->parents
, next_parent
) {
2236 if (!bdrv_a_allow_b(a
, b
, errp
)) {
2245 static void GRAPH_RDLOCK
2246 bdrv_child_perm(BlockDriverState
*bs
, BlockDriverState
*child_bs
,
2247 BdrvChild
*c
, BdrvChildRole role
,
2248 BlockReopenQueue
*reopen_queue
,
2249 uint64_t parent_perm
, uint64_t parent_shared
,
2250 uint64_t *nperm
, uint64_t *nshared
)
2252 assert(bs
->drv
&& bs
->drv
->bdrv_child_perm
);
2253 GLOBAL_STATE_CODE();
2254 bs
->drv
->bdrv_child_perm(bs
, c
, role
, reopen_queue
,
2255 parent_perm
, parent_shared
,
2257 /* TODO Take force_share from reopen_queue */
2258 if (child_bs
&& child_bs
->force_share
) {
2259 *nshared
= BLK_PERM_ALL
;
2264 * Adds the whole subtree of @bs (including @bs itself) to the @list (except for
2265 * nodes that are already in the @list, of course) so that final list is
2266 * topologically sorted. Return the result (GSList @list object is updated, so
2267 * don't use old reference after function call).
2269 * On function start @list must be already topologically sorted and for any node
2270 * in the @list the whole subtree of the node must be in the @list as well. The
2271 * simplest way to satisfy this criteria: use only result of
2272 * bdrv_topological_dfs() or NULL as @list parameter.
2274 static GSList
* GRAPH_RDLOCK
2275 bdrv_topological_dfs(GSList
*list
, GHashTable
*found
, BlockDriverState
*bs
)
2278 g_autoptr(GHashTable
) local_found
= NULL
;
2280 GLOBAL_STATE_CODE();
2284 found
= local_found
= g_hash_table_new(NULL
, NULL
);
2287 if (g_hash_table_contains(found
, bs
)) {
2290 g_hash_table_add(found
, bs
);
2292 QLIST_FOREACH(child
, &bs
->children
, next
) {
2293 list
= bdrv_topological_dfs(list
, found
, child
->bs
);
2296 return g_slist_prepend(list
, bs
);
2299 typedef struct BdrvChildSetPermState
{
2302 uint64_t old_shared_perm
;
2303 } BdrvChildSetPermState
;
2305 static void bdrv_child_set_perm_abort(void *opaque
)
2307 BdrvChildSetPermState
*s
= opaque
;
2309 GLOBAL_STATE_CODE();
2311 s
->child
->perm
= s
->old_perm
;
2312 s
->child
->shared_perm
= s
->old_shared_perm
;
2315 static TransactionActionDrv bdrv_child_set_pem_drv
= {
2316 .abort
= bdrv_child_set_perm_abort
,
2320 static void bdrv_child_set_perm(BdrvChild
*c
, uint64_t perm
,
2321 uint64_t shared
, Transaction
*tran
)
2323 BdrvChildSetPermState
*s
= g_new(BdrvChildSetPermState
, 1);
2324 GLOBAL_STATE_CODE();
2326 *s
= (BdrvChildSetPermState
) {
2328 .old_perm
= c
->perm
,
2329 .old_shared_perm
= c
->shared_perm
,
2333 c
->shared_perm
= shared
;
2335 tran_add(tran
, &bdrv_child_set_pem_drv
, s
);
2338 static void GRAPH_RDLOCK
bdrv_drv_set_perm_commit(void *opaque
)
2340 BlockDriverState
*bs
= opaque
;
2341 uint64_t cumulative_perms
, cumulative_shared_perms
;
2342 GLOBAL_STATE_CODE();
2344 if (bs
->drv
->bdrv_set_perm
) {
2345 bdrv_get_cumulative_perm(bs
, &cumulative_perms
,
2346 &cumulative_shared_perms
);
2347 bs
->drv
->bdrv_set_perm(bs
, cumulative_perms
, cumulative_shared_perms
);
2351 static void GRAPH_RDLOCK
bdrv_drv_set_perm_abort(void *opaque
)
2353 BlockDriverState
*bs
= opaque
;
2354 GLOBAL_STATE_CODE();
2356 if (bs
->drv
->bdrv_abort_perm_update
) {
2357 bs
->drv
->bdrv_abort_perm_update(bs
);
2361 TransactionActionDrv bdrv_drv_set_perm_drv
= {
2362 .abort
= bdrv_drv_set_perm_abort
,
2363 .commit
= bdrv_drv_set_perm_commit
,
2367 * After calling this function, the transaction @tran may only be completed
2368 * while holding a reader lock for the graph.
2370 static int GRAPH_RDLOCK
2371 bdrv_drv_set_perm(BlockDriverState
*bs
, uint64_t perm
, uint64_t shared_perm
,
2372 Transaction
*tran
, Error
**errp
)
2374 GLOBAL_STATE_CODE();
2379 if (bs
->drv
->bdrv_check_perm
) {
2380 int ret
= bs
->drv
->bdrv_check_perm(bs
, perm
, shared_perm
, errp
);
2387 tran_add(tran
, &bdrv_drv_set_perm_drv
, bs
);
2393 typedef struct BdrvReplaceChildState
{
2395 BlockDriverState
*old_bs
;
2396 } BdrvReplaceChildState
;
2398 static void GRAPH_WRLOCK
bdrv_replace_child_commit(void *opaque
)
2400 BdrvReplaceChildState
*s
= opaque
;
2401 GLOBAL_STATE_CODE();
2403 bdrv_schedule_unref(s
->old_bs
);
2406 static void GRAPH_WRLOCK
bdrv_replace_child_abort(void *opaque
)
2408 BdrvReplaceChildState
*s
= opaque
;
2409 BlockDriverState
*new_bs
= s
->child
->bs
;
2411 GLOBAL_STATE_CODE();
2412 assert_bdrv_graph_writable();
2414 /* old_bs reference is transparently moved from @s to @s->child */
2415 if (!s
->child
->bs
) {
2417 * The parents were undrained when removing old_bs from the child. New
2418 * requests can't have been made, though, because the child was empty.
2420 * TODO Make bdrv_replace_child_noperm() transactionable to avoid
2421 * undraining the parent in the first place. Once this is done, having
2422 * new_bs drained when calling bdrv_replace_child_tran() is not a
2423 * requirement any more.
2425 bdrv_parent_drained_begin_single(s
->child
);
2426 assert(!bdrv_parent_drained_poll_single(s
->child
));
2428 assert(s
->child
->quiesced_parent
);
2429 bdrv_replace_child_noperm(s
->child
, s
->old_bs
);
2434 static TransactionActionDrv bdrv_replace_child_drv
= {
2435 .commit
= bdrv_replace_child_commit
,
2436 .abort
= bdrv_replace_child_abort
,
2441 * bdrv_replace_child_tran
2443 * Note: real unref of old_bs is done only on commit.
2445 * Both @child->bs and @new_bs (if non-NULL) must be drained. @new_bs must be
2446 * kept drained until the transaction is completed.
2448 * After calling this function, the transaction @tran may only be completed
2449 * while holding a writer lock for the graph.
2451 * The function doesn't update permissions, caller is responsible for this.
2453 static void GRAPH_WRLOCK
2454 bdrv_replace_child_tran(BdrvChild
*child
, BlockDriverState
*new_bs
,
2457 BdrvReplaceChildState
*s
= g_new(BdrvReplaceChildState
, 1);
2459 assert(child
->quiesced_parent
);
2460 assert(!new_bs
|| new_bs
->quiesce_counter
);
2462 *s
= (BdrvReplaceChildState
) {
2464 .old_bs
= child
->bs
,
2466 tran_add(tran
, &bdrv_replace_child_drv
, s
);
2472 bdrv_replace_child_noperm(child
, new_bs
);
2473 /* old_bs reference is transparently moved from @child to @s */
2477 * Refresh permissions in @bs subtree. The function is intended to be called
2478 * after some graph modification that was done without permission update.
2480 * After calling this function, the transaction @tran may only be completed
2481 * while holding a reader lock for the graph.
2483 static int GRAPH_RDLOCK
2484 bdrv_node_refresh_perm(BlockDriverState
*bs
, BlockReopenQueue
*q
,
2485 Transaction
*tran
, Error
**errp
)
2487 BlockDriver
*drv
= bs
->drv
;
2490 uint64_t cumulative_perms
, cumulative_shared_perms
;
2491 GLOBAL_STATE_CODE();
2493 bdrv_get_cumulative_perm(bs
, &cumulative_perms
, &cumulative_shared_perms
);
2495 /* Write permissions never work with read-only images */
2496 if ((cumulative_perms
& (BLK_PERM_WRITE
| BLK_PERM_WRITE_UNCHANGED
)) &&
2497 !bdrv_is_writable_after_reopen(bs
, q
))
2499 if (!bdrv_is_writable_after_reopen(bs
, NULL
)) {
2500 error_setg(errp
, "Block node is read-only");
2502 error_setg(errp
, "Read-only block node '%s' cannot support "
2503 "read-write users", bdrv_get_node_name(bs
));
2510 * Unaligned requests will automatically be aligned to bl.request_alignment
2511 * and without RESIZE we can't extend requests to write to space beyond the
2512 * end of the image, so it's required that the image size is aligned.
2514 if ((cumulative_perms
& (BLK_PERM_WRITE
| BLK_PERM_WRITE_UNCHANGED
)) &&
2515 !(cumulative_perms
& BLK_PERM_RESIZE
))
2517 if ((bs
->total_sectors
* BDRV_SECTOR_SIZE
) % bs
->bl
.request_alignment
) {
2518 error_setg(errp
, "Cannot get 'write' permission without 'resize': "
2519 "Image size is not a multiple of request "
2525 /* Check this node */
2530 ret
= bdrv_drv_set_perm(bs
, cumulative_perms
, cumulative_shared_perms
, tran
,
2536 /* Drivers that never have children can omit .bdrv_child_perm() */
2537 if (!drv
->bdrv_child_perm
) {
2538 assert(QLIST_EMPTY(&bs
->children
));
2542 /* Check all children */
2543 QLIST_FOREACH(c
, &bs
->children
, next
) {
2544 uint64_t cur_perm
, cur_shared
;
2546 bdrv_child_perm(bs
, c
->bs
, c
, c
->role
, q
,
2547 cumulative_perms
, cumulative_shared_perms
,
2548 &cur_perm
, &cur_shared
);
2549 bdrv_child_set_perm(c
, cur_perm
, cur_shared
, tran
);
2556 * @list is a product of bdrv_topological_dfs() (may be called several times) -
2557 * a topologically sorted subgraph.
2559 * After calling this function, the transaction @tran may only be completed
2560 * while holding a reader lock for the graph.
2562 static int GRAPH_RDLOCK
2563 bdrv_do_refresh_perms(GSList
*list
, BlockReopenQueue
*q
, Transaction
*tran
,
2567 BlockDriverState
*bs
;
2568 GLOBAL_STATE_CODE();
2570 for ( ; list
; list
= list
->next
) {
2573 if (bdrv_parent_perms_conflict(bs
, errp
)) {
2577 ret
= bdrv_node_refresh_perm(bs
, q
, tran
, errp
);
2587 * @list is any list of nodes. List is completed by all subtrees and
2588 * topologically sorted. It's not a problem if some node occurs in the @list
2591 * After calling this function, the transaction @tran may only be completed
2592 * while holding a reader lock for the graph.
2594 static int GRAPH_RDLOCK
2595 bdrv_list_refresh_perms(GSList
*list
, BlockReopenQueue
*q
, Transaction
*tran
,
2598 g_autoptr(GHashTable
) found
= g_hash_table_new(NULL
, NULL
);
2599 g_autoptr(GSList
) refresh_list
= NULL
;
2601 for ( ; list
; list
= list
->next
) {
2602 refresh_list
= bdrv_topological_dfs(refresh_list
, found
, list
->data
);
2605 return bdrv_do_refresh_perms(refresh_list
, q
, tran
, errp
);
2608 void bdrv_get_cumulative_perm(BlockDriverState
*bs
, uint64_t *perm
,
2609 uint64_t *shared_perm
)
2612 uint64_t cumulative_perms
= 0;
2613 uint64_t cumulative_shared_perms
= BLK_PERM_ALL
;
2615 GLOBAL_STATE_CODE();
2617 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
2618 cumulative_perms
|= c
->perm
;
2619 cumulative_shared_perms
&= c
->shared_perm
;
2622 *perm
= cumulative_perms
;
2623 *shared_perm
= cumulative_shared_perms
;
2626 char *bdrv_perm_names(uint64_t perm
)
2632 { BLK_PERM_CONSISTENT_READ
, "consistent read" },
2633 { BLK_PERM_WRITE
, "write" },
2634 { BLK_PERM_WRITE_UNCHANGED
, "write unchanged" },
2635 { BLK_PERM_RESIZE
, "resize" },
2639 GString
*result
= g_string_sized_new(30);
2640 struct perm_name
*p
;
2642 for (p
= permissions
; p
->name
; p
++) {
2643 if (perm
& p
->perm
) {
2644 if (result
->len
> 0) {
2645 g_string_append(result
, ", ");
2647 g_string_append(result
, p
->name
);
2651 return g_string_free(result
, FALSE
);
2656 * @tran is allowed to be NULL. In this case no rollback is possible.
2658 * After calling this function, the transaction @tran may only be completed
2659 * while holding a reader lock for the graph.
2661 static int GRAPH_RDLOCK
2662 bdrv_refresh_perms(BlockDriverState
*bs
, Transaction
*tran
, Error
**errp
)
2665 Transaction
*local_tran
= NULL
;
2666 g_autoptr(GSList
) list
= bdrv_topological_dfs(NULL
, NULL
, bs
);
2667 GLOBAL_STATE_CODE();
2670 tran
= local_tran
= tran_new();
2673 ret
= bdrv_do_refresh_perms(list
, NULL
, tran
, errp
);
2676 tran_finalize(local_tran
, ret
);
2682 int bdrv_child_try_set_perm(BdrvChild
*c
, uint64_t perm
, uint64_t shared
,
2685 Error
*local_err
= NULL
;
2686 Transaction
*tran
= tran_new();
2689 GLOBAL_STATE_CODE();
2691 bdrv_child_set_perm(c
, perm
, shared
, tran
);
2693 ret
= bdrv_refresh_perms(c
->bs
, tran
, &local_err
);
2695 tran_finalize(tran
, ret
);
2698 if ((perm
& ~c
->perm
) || (c
->shared_perm
& ~shared
)) {
2699 /* tighten permissions */
2700 error_propagate(errp
, local_err
);
2703 * Our caller may intend to only loosen restrictions and
2704 * does not expect this function to fail. Errors are not
2705 * fatal in such a case, so we can just hide them from our
2708 error_free(local_err
);
2716 int bdrv_child_refresh_perms(BlockDriverState
*bs
, BdrvChild
*c
, Error
**errp
)
2718 uint64_t parent_perms
, parent_shared
;
2719 uint64_t perms
, shared
;
2721 GLOBAL_STATE_CODE();
2723 bdrv_get_cumulative_perm(bs
, &parent_perms
, &parent_shared
);
2724 bdrv_child_perm(bs
, c
->bs
, c
, c
->role
, NULL
,
2725 parent_perms
, parent_shared
, &perms
, &shared
);
2727 return bdrv_child_try_set_perm(c
, perms
, shared
, errp
);
2731 * Default implementation for .bdrv_child_perm() for block filters:
2732 * Forward CONSISTENT_READ, WRITE, WRITE_UNCHANGED, and RESIZE to the
2735 static void bdrv_filter_default_perms(BlockDriverState
*bs
, BdrvChild
*c
,
2737 BlockReopenQueue
*reopen_queue
,
2738 uint64_t perm
, uint64_t shared
,
2739 uint64_t *nperm
, uint64_t *nshared
)
2741 GLOBAL_STATE_CODE();
2742 *nperm
= perm
& DEFAULT_PERM_PASSTHROUGH
;
2743 *nshared
= (shared
& DEFAULT_PERM_PASSTHROUGH
) | DEFAULT_PERM_UNCHANGED
;
2746 static void bdrv_default_perms_for_cow(BlockDriverState
*bs
, BdrvChild
*c
,
2748 BlockReopenQueue
*reopen_queue
,
2749 uint64_t perm
, uint64_t shared
,
2750 uint64_t *nperm
, uint64_t *nshared
)
2752 assert(role
& BDRV_CHILD_COW
);
2753 GLOBAL_STATE_CODE();
2756 * We want consistent read from backing files if the parent needs it.
2757 * No other operations are performed on backing files.
2759 perm
&= BLK_PERM_CONSISTENT_READ
;
2762 * If the parent can deal with changing data, we're okay with a
2763 * writable and resizable backing file.
2764 * TODO Require !(perm & BLK_PERM_CONSISTENT_READ), too?
2766 if (shared
& BLK_PERM_WRITE
) {
2767 shared
= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2772 shared
|= BLK_PERM_CONSISTENT_READ
| BLK_PERM_WRITE_UNCHANGED
;
2774 if (bs
->open_flags
& BDRV_O_INACTIVE
) {
2775 shared
|= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2782 static void bdrv_default_perms_for_storage(BlockDriverState
*bs
, BdrvChild
*c
,
2784 BlockReopenQueue
*reopen_queue
,
2785 uint64_t perm
, uint64_t shared
,
2786 uint64_t *nperm
, uint64_t *nshared
)
2790 GLOBAL_STATE_CODE();
2791 assert(role
& (BDRV_CHILD_METADATA
| BDRV_CHILD_DATA
));
2793 flags
= bdrv_reopen_get_flags(reopen_queue
, bs
);
2796 * Apart from the modifications below, the same permissions are
2797 * forwarded and left alone as for filters
2799 bdrv_filter_default_perms(bs
, c
, role
, reopen_queue
,
2800 perm
, shared
, &perm
, &shared
);
2802 if (role
& BDRV_CHILD_METADATA
) {
2803 /* Format drivers may touch metadata even if the guest doesn't write */
2804 if (bdrv_is_writable_after_reopen(bs
, reopen_queue
)) {
2805 perm
|= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2809 * bs->file always needs to be consistent because of the
2810 * metadata. We can never allow other users to resize or write
2813 if (!(flags
& BDRV_O_NO_IO
)) {
2814 perm
|= BLK_PERM_CONSISTENT_READ
;
2816 shared
&= ~(BLK_PERM_WRITE
| BLK_PERM_RESIZE
);
2819 if (role
& BDRV_CHILD_DATA
) {
2821 * Technically, everything in this block is a subset of the
2822 * BDRV_CHILD_METADATA path taken above, and so this could
2823 * be an "else if" branch. However, that is not obvious, and
2824 * this function is not performance critical, therefore we let
2825 * this be an independent "if".
2829 * We cannot allow other users to resize the file because the
2830 * format driver might have some assumptions about the size
2831 * (e.g. because it is stored in metadata, or because the file
2832 * is split into fixed-size data files).
2834 shared
&= ~BLK_PERM_RESIZE
;
2837 * WRITE_UNCHANGED often cannot be performed as such on the
2838 * data file. For example, the qcow2 driver may still need to
2839 * write copied clusters on copy-on-read.
2841 if (perm
& BLK_PERM_WRITE_UNCHANGED
) {
2842 perm
|= BLK_PERM_WRITE
;
2846 * If the data file is written to, the format driver may
2847 * expect to be able to resize it by writing beyond the EOF.
2849 if (perm
& BLK_PERM_WRITE
) {
2850 perm
|= BLK_PERM_RESIZE
;
2854 if (bs
->open_flags
& BDRV_O_INACTIVE
) {
2855 shared
|= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2862 void bdrv_default_perms(BlockDriverState
*bs
, BdrvChild
*c
,
2863 BdrvChildRole role
, BlockReopenQueue
*reopen_queue
,
2864 uint64_t perm
, uint64_t shared
,
2865 uint64_t *nperm
, uint64_t *nshared
)
2867 GLOBAL_STATE_CODE();
2868 if (role
& BDRV_CHILD_FILTERED
) {
2869 assert(!(role
& (BDRV_CHILD_DATA
| BDRV_CHILD_METADATA
|
2871 bdrv_filter_default_perms(bs
, c
, role
, reopen_queue
,
2872 perm
, shared
, nperm
, nshared
);
2873 } else if (role
& BDRV_CHILD_COW
) {
2874 assert(!(role
& (BDRV_CHILD_DATA
| BDRV_CHILD_METADATA
)));
2875 bdrv_default_perms_for_cow(bs
, c
, role
, reopen_queue
,
2876 perm
, shared
, nperm
, nshared
);
2877 } else if (role
& (BDRV_CHILD_METADATA
| BDRV_CHILD_DATA
)) {
2878 bdrv_default_perms_for_storage(bs
, c
, role
, reopen_queue
,
2879 perm
, shared
, nperm
, nshared
);
2881 g_assert_not_reached();
2885 uint64_t bdrv_qapi_perm_to_blk_perm(BlockPermission qapi_perm
)
2887 static const uint64_t permissions
[] = {
2888 [BLOCK_PERMISSION_CONSISTENT_READ
] = BLK_PERM_CONSISTENT_READ
,
2889 [BLOCK_PERMISSION_WRITE
] = BLK_PERM_WRITE
,
2890 [BLOCK_PERMISSION_WRITE_UNCHANGED
] = BLK_PERM_WRITE_UNCHANGED
,
2891 [BLOCK_PERMISSION_RESIZE
] = BLK_PERM_RESIZE
,
2894 QEMU_BUILD_BUG_ON(ARRAY_SIZE(permissions
) != BLOCK_PERMISSION__MAX
);
2895 QEMU_BUILD_BUG_ON(1UL << ARRAY_SIZE(permissions
) != BLK_PERM_ALL
+ 1);
2897 assert(qapi_perm
< BLOCK_PERMISSION__MAX
);
2899 return permissions
[qapi_perm
];
2903 * Replaces the node that a BdrvChild points to without updating permissions.
2905 * If @new_bs is non-NULL, the parent of @child must already be drained through
2906 * @child and the caller must hold the AioContext lock for @new_bs.
2908 static void GRAPH_WRLOCK
2909 bdrv_replace_child_noperm(BdrvChild
*child
, BlockDriverState
*new_bs
)
2911 BlockDriverState
*old_bs
= child
->bs
;
2912 int new_bs_quiesce_counter
;
2914 assert(!child
->frozen
);
2917 * If we want to change the BdrvChild to point to a drained node as its new
2918 * child->bs, we need to make sure that its new parent is drained, too. In
2919 * other words, either child->quiesce_parent must already be true or we must
2920 * be able to set it and keep the parent's quiesce_counter consistent with
2921 * that, but without polling or starting new requests (this function
2922 * guarantees that it doesn't poll, and starting new requests would be
2923 * against the invariants of drain sections).
2925 * To keep things simple, we pick the first option (child->quiesce_parent
2926 * must already be true). We also generalise the rule a bit to make it
2927 * easier to verify in callers and more likely to be covered in test cases:
2928 * The parent must be quiesced through this child even if new_bs isn't
2929 * currently drained.
2931 * The only exception is for callers that always pass new_bs == NULL. In
2932 * this case, we obviously never need to consider the case of a drained
2933 * new_bs, so we can keep the callers simpler by allowing them not to drain
2936 assert(!new_bs
|| child
->quiesced_parent
);
2937 assert(old_bs
!= new_bs
);
2938 GLOBAL_STATE_CODE();
2940 if (old_bs
&& new_bs
) {
2941 assert(bdrv_get_aio_context(old_bs
) == bdrv_get_aio_context(new_bs
));
2945 if (child
->klass
->detach
) {
2946 child
->klass
->detach(child
);
2948 QLIST_REMOVE(child
, next_parent
);
2954 QLIST_INSERT_HEAD(&new_bs
->parents
, child
, next_parent
);
2955 if (child
->klass
->attach
) {
2956 child
->klass
->attach(child
);
2961 * If the parent was drained through this BdrvChild previously, but new_bs
2962 * is not drained, allow requests to come in only after the new node has
2965 new_bs_quiesce_counter
= (new_bs
? new_bs
->quiesce_counter
: 0);
2966 if (!new_bs_quiesce_counter
&& child
->quiesced_parent
) {
2967 bdrv_parent_drained_end_single(child
);
2972 * Free the given @child.
2974 * The child must be empty (i.e. `child->bs == NULL`) and it must be
2975 * unused (i.e. not in a children list).
2977 static void bdrv_child_free(BdrvChild
*child
)
2980 GLOBAL_STATE_CODE();
2981 GRAPH_RDLOCK_GUARD_MAINLOOP();
2983 assert(!child
->next
.le_prev
); /* not in children list */
2985 g_free(child
->name
);
2989 typedef struct BdrvAttachChildCommonState
{
2991 AioContext
*old_parent_ctx
;
2992 AioContext
*old_child_ctx
;
2993 } BdrvAttachChildCommonState
;
2995 static void GRAPH_WRLOCK
bdrv_attach_child_common_abort(void *opaque
)
2997 BdrvAttachChildCommonState
*s
= opaque
;
2998 BlockDriverState
*bs
= s
->child
->bs
;
3000 GLOBAL_STATE_CODE();
3001 assert_bdrv_graph_writable();
3003 bdrv_replace_child_noperm(s
->child
, NULL
);
3005 if (bdrv_get_aio_context(bs
) != s
->old_child_ctx
) {
3006 bdrv_try_change_aio_context(bs
, s
->old_child_ctx
, NULL
, &error_abort
);
3009 if (bdrv_child_get_parent_aio_context(s
->child
) != s
->old_parent_ctx
) {
3011 GHashTable
*visited
;
3016 /* No need to visit `child`, because it has been detached already */
3017 visited
= g_hash_table_new(NULL
, NULL
);
3018 ret
= s
->child
->klass
->change_aio_ctx(s
->child
, s
->old_parent_ctx
,
3019 visited
, tran
, &error_abort
);
3020 g_hash_table_destroy(visited
);
3022 /* transaction is supposed to always succeed */
3023 assert(ret
== true);
3027 bdrv_schedule_unref(bs
);
3028 bdrv_child_free(s
->child
);
3031 static TransactionActionDrv bdrv_attach_child_common_drv
= {
3032 .abort
= bdrv_attach_child_common_abort
,
3037 * Common part of attaching bdrv child to bs or to blk or to job
3039 * Function doesn't update permissions, caller is responsible for this.
3041 * After calling this function, the transaction @tran may only be completed
3042 * while holding a writer lock for the graph.
3044 * Returns new created child.
3046 * The caller must hold the AioContext lock for @child_bs. Both @parent_bs and
3047 * @child_bs can move to a different AioContext in this function. Callers must
3048 * make sure that their AioContext locking is still correct after this.
3050 static BdrvChild
* GRAPH_WRLOCK
3051 bdrv_attach_child_common(BlockDriverState
*child_bs
,
3052 const char *child_name
,
3053 const BdrvChildClass
*child_class
,
3054 BdrvChildRole child_role
,
3055 uint64_t perm
, uint64_t shared_perm
,
3057 Transaction
*tran
, Error
**errp
)
3059 BdrvChild
*new_child
;
3060 AioContext
*parent_ctx
, *new_child_ctx
;
3061 AioContext
*child_ctx
= bdrv_get_aio_context(child_bs
);
3063 assert(child_class
->get_parent_desc
);
3064 GLOBAL_STATE_CODE();
3066 new_child
= g_new(BdrvChild
, 1);
3067 *new_child
= (BdrvChild
) {
3069 .name
= g_strdup(child_name
),
3070 .klass
= child_class
,
3073 .shared_perm
= shared_perm
,
3078 * If the AioContexts don't match, first try to move the subtree of
3079 * child_bs into the AioContext of the new parent. If this doesn't work,
3080 * try moving the parent into the AioContext of child_bs instead.
3082 parent_ctx
= bdrv_child_get_parent_aio_context(new_child
);
3083 if (child_ctx
!= parent_ctx
) {
3084 Error
*local_err
= NULL
;
3085 int ret
= bdrv_try_change_aio_context(child_bs
, parent_ctx
, NULL
,
3088 if (ret
< 0 && child_class
->change_aio_ctx
) {
3089 Transaction
*aio_ctx_tran
= tran_new();
3090 GHashTable
*visited
= g_hash_table_new(NULL
, NULL
);
3093 g_hash_table_add(visited
, new_child
);
3094 ret_child
= child_class
->change_aio_ctx(new_child
, child_ctx
,
3095 visited
, aio_ctx_tran
,
3097 if (ret_child
== true) {
3098 error_free(local_err
);
3101 tran_finalize(aio_ctx_tran
, ret_child
== true ? 0 : -1);
3102 g_hash_table_destroy(visited
);
3106 error_propagate(errp
, local_err
);
3107 bdrv_child_free(new_child
);
3112 new_child_ctx
= bdrv_get_aio_context(child_bs
);
3113 if (new_child_ctx
!= child_ctx
) {
3114 aio_context_release(child_ctx
);
3115 aio_context_acquire(new_child_ctx
);
3120 * Let every new BdrvChild start with a drained parent. Inserting the child
3121 * in the graph with bdrv_replace_child_noperm() will undrain it if
3122 * @child_bs is not drained.
3124 * The child was only just created and is not yet visible in global state
3125 * until bdrv_replace_child_noperm() inserts it into the graph, so nobody
3126 * could have sent requests and polling is not necessary.
3128 * Note that this means that the parent isn't fully drained yet, we only
3129 * stop new requests from coming in. This is fine, we don't care about the
3130 * old requests here, they are not for this child. If another place enters a
3131 * drain section for the same parent, but wants it to be fully quiesced, it
3132 * will not run most of the the code in .drained_begin() again (which is not
3133 * a problem, we already did this), but it will still poll until the parent
3134 * is fully quiesced, so it will not be negatively affected either.
3136 bdrv_parent_drained_begin_single(new_child
);
3137 bdrv_replace_child_noperm(new_child
, child_bs
);
3139 BdrvAttachChildCommonState
*s
= g_new(BdrvAttachChildCommonState
, 1);
3140 *s
= (BdrvAttachChildCommonState
) {
3142 .old_parent_ctx
= parent_ctx
,
3143 .old_child_ctx
= child_ctx
,
3145 tran_add(tran
, &bdrv_attach_child_common_drv
, s
);
3147 if (new_child_ctx
!= child_ctx
) {
3148 aio_context_release(new_child_ctx
);
3149 aio_context_acquire(child_ctx
);
3156 * Function doesn't update permissions, caller is responsible for this.
3158 * The caller must hold the AioContext lock for @child_bs. Both @parent_bs and
3159 * @child_bs can move to a different AioContext in this function. Callers must
3160 * make sure that their AioContext locking is still correct after this.
3162 * After calling this function, the transaction @tran may only be completed
3163 * while holding a writer lock for the graph.
3165 static BdrvChild
* GRAPH_WRLOCK
3166 bdrv_attach_child_noperm(BlockDriverState
*parent_bs
,
3167 BlockDriverState
*child_bs
,
3168 const char *child_name
,
3169 const BdrvChildClass
*child_class
,
3170 BdrvChildRole child_role
,
3174 uint64_t perm
, shared_perm
;
3176 assert(parent_bs
->drv
);
3177 GLOBAL_STATE_CODE();
3179 if (bdrv_recurse_has_child(child_bs
, parent_bs
)) {
3180 error_setg(errp
, "Making '%s' a %s child of '%s' would create a cycle",
3181 child_bs
->node_name
, child_name
, parent_bs
->node_name
);
3185 bdrv_get_cumulative_perm(parent_bs
, &perm
, &shared_perm
);
3186 bdrv_child_perm(parent_bs
, child_bs
, NULL
, child_role
, NULL
,
3187 perm
, shared_perm
, &perm
, &shared_perm
);
3189 return bdrv_attach_child_common(child_bs
, child_name
, child_class
,
3190 child_role
, perm
, shared_perm
, parent_bs
,
3195 * This function steals the reference to child_bs from the caller.
3196 * That reference is later dropped by bdrv_root_unref_child().
3198 * On failure NULL is returned, errp is set and the reference to
3199 * child_bs is also dropped.
3201 * The caller must hold the AioContext lock @child_bs, but not that of @ctx
3202 * (unless @child_bs is already in @ctx).
3204 BdrvChild
*bdrv_root_attach_child(BlockDriverState
*child_bs
,
3205 const char *child_name
,
3206 const BdrvChildClass
*child_class
,
3207 BdrvChildRole child_role
,
3208 uint64_t perm
, uint64_t shared_perm
,
3209 void *opaque
, Error
**errp
)
3213 Transaction
*tran
= tran_new();
3215 GLOBAL_STATE_CODE();
3217 child
= bdrv_attach_child_common(child_bs
, child_name
, child_class
,
3218 child_role
, perm
, shared_perm
, opaque
,
3225 ret
= bdrv_refresh_perms(child_bs
, tran
, errp
);
3228 tran_finalize(tran
, ret
);
3230 bdrv_schedule_unref(child_bs
);
3232 return ret
< 0 ? NULL
: child
;
3236 * This function transfers the reference to child_bs from the caller
3237 * to parent_bs. That reference is later dropped by parent_bs on
3238 * bdrv_close() or if someone calls bdrv_unref_child().
3240 * On failure NULL is returned, errp is set and the reference to
3241 * child_bs is also dropped.
3243 * If @parent_bs and @child_bs are in different AioContexts, the caller must
3244 * hold the AioContext lock for @child_bs, but not for @parent_bs.
3246 BdrvChild
*bdrv_attach_child(BlockDriverState
*parent_bs
,
3247 BlockDriverState
*child_bs
,
3248 const char *child_name
,
3249 const BdrvChildClass
*child_class
,
3250 BdrvChildRole child_role
,
3255 Transaction
*tran
= tran_new();
3257 GLOBAL_STATE_CODE();
3259 child
= bdrv_attach_child_noperm(parent_bs
, child_bs
, child_name
,
3260 child_class
, child_role
, tran
, errp
);
3266 ret
= bdrv_refresh_perms(parent_bs
, tran
, errp
);
3272 tran_finalize(tran
, ret
);
3274 bdrv_schedule_unref(child_bs
);
3276 return ret
< 0 ? NULL
: child
;
3279 /* Callers must ensure that child->frozen is false. */
3280 void bdrv_root_unref_child(BdrvChild
*child
)
3282 BlockDriverState
*child_bs
= child
->bs
;
3284 GLOBAL_STATE_CODE();
3285 bdrv_replace_child_noperm(child
, NULL
);
3286 bdrv_child_free(child
);
3290 * Update permissions for old node. We're just taking a parent away, so
3291 * we're loosening restrictions. Errors of permission update are not
3292 * fatal in this case, ignore them.
3294 bdrv_refresh_perms(child_bs
, NULL
, NULL
);
3297 * When the parent requiring a non-default AioContext is removed, the
3298 * node moves back to the main AioContext
3300 bdrv_try_change_aio_context(child_bs
, qemu_get_aio_context(), NULL
,
3304 bdrv_schedule_unref(child_bs
);
3307 typedef struct BdrvSetInheritsFrom
{
3308 BlockDriverState
*bs
;
3309 BlockDriverState
*old_inherits_from
;
3310 } BdrvSetInheritsFrom
;
3312 static void bdrv_set_inherits_from_abort(void *opaque
)
3314 BdrvSetInheritsFrom
*s
= opaque
;
3316 s
->bs
->inherits_from
= s
->old_inherits_from
;
3319 static TransactionActionDrv bdrv_set_inherits_from_drv
= {
3320 .abort
= bdrv_set_inherits_from_abort
,
3324 /* @tran is allowed to be NULL. In this case no rollback is possible */
3325 static void bdrv_set_inherits_from(BlockDriverState
*bs
,
3326 BlockDriverState
*new_inherits_from
,
3330 BdrvSetInheritsFrom
*s
= g_new(BdrvSetInheritsFrom
, 1);
3332 *s
= (BdrvSetInheritsFrom
) {
3334 .old_inherits_from
= bs
->inherits_from
,
3337 tran_add(tran
, &bdrv_set_inherits_from_drv
, s
);
3340 bs
->inherits_from
= new_inherits_from
;
3344 * Clear all inherits_from pointers from children and grandchildren of
3345 * @root that point to @root, where necessary.
3346 * @tran is allowed to be NULL. In this case no rollback is possible
3348 static void GRAPH_WRLOCK
3349 bdrv_unset_inherits_from(BlockDriverState
*root
, BdrvChild
*child
,
3354 if (child
->bs
->inherits_from
== root
) {
3356 * Remove inherits_from only when the last reference between root and
3357 * child->bs goes away.
3359 QLIST_FOREACH(c
, &root
->children
, next
) {
3360 if (c
!= child
&& c
->bs
== child
->bs
) {
3365 bdrv_set_inherits_from(child
->bs
, NULL
, tran
);
3369 QLIST_FOREACH(c
, &child
->bs
->children
, next
) {
3370 bdrv_unset_inherits_from(root
, c
, tran
);
3374 /* Callers must ensure that child->frozen is false. */
3375 void bdrv_unref_child(BlockDriverState
*parent
, BdrvChild
*child
)
3377 GLOBAL_STATE_CODE();
3378 if (child
== NULL
) {
3382 bdrv_unset_inherits_from(parent
, child
, NULL
);
3383 bdrv_root_unref_child(child
);
3387 static void GRAPH_RDLOCK
3388 bdrv_parent_cb_change_media(BlockDriverState
*bs
, bool load
)
3391 GLOBAL_STATE_CODE();
3392 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
3393 if (c
->klass
->change_media
) {
3394 c
->klass
->change_media(c
, load
);
3399 /* Return true if you can reach parent going through child->inherits_from
3400 * recursively. If parent or child are NULL, return false */
3401 static bool bdrv_inherits_from_recursive(BlockDriverState
*child
,
3402 BlockDriverState
*parent
)
3404 while (child
&& child
!= parent
) {
3405 child
= child
->inherits_from
;
3408 return child
!= NULL
;
3412 * Return the BdrvChildRole for @bs's backing child. bs->backing is
3413 * mostly used for COW backing children (role = COW), but also for
3414 * filtered children (role = FILTERED | PRIMARY).
3416 static BdrvChildRole
bdrv_backing_role(BlockDriverState
*bs
)
3418 if (bs
->drv
&& bs
->drv
->is_filter
) {
3419 return BDRV_CHILD_FILTERED
| BDRV_CHILD_PRIMARY
;
3421 return BDRV_CHILD_COW
;
3426 * Sets the bs->backing or bs->file link of a BDS. A new reference is created;
3427 * callers which don't need their own reference any more must call bdrv_unref().
3429 * If the respective child is already present (i.e. we're detaching a node),
3430 * that child node must be drained.
3432 * Function doesn't update permissions, caller is responsible for this.
3434 * The caller must hold the AioContext lock for @child_bs. Both @parent_bs and
3435 * @child_bs can move to a different AioContext in this function. Callers must
3436 * make sure that their AioContext locking is still correct after this.
3438 * After calling this function, the transaction @tran may only be completed
3439 * while holding a writer lock for the graph.
3441 static int GRAPH_WRLOCK
3442 bdrv_set_file_or_backing_noperm(BlockDriverState
*parent_bs
,
3443 BlockDriverState
*child_bs
,
3445 Transaction
*tran
, Error
**errp
)
3447 bool update_inherits_from
=
3448 bdrv_inherits_from_recursive(child_bs
, parent_bs
);
3449 BdrvChild
*child
= is_backing
? parent_bs
->backing
: parent_bs
->file
;
3452 GLOBAL_STATE_CODE();
3454 if (!parent_bs
->drv
) {
3456 * Node without drv is an object without a class :/. TODO: finally fix
3457 * qcow2 driver to never clear bs->drv and implement format corruption
3458 * handling in other way.
3460 error_setg(errp
, "Node corrupted");
3464 if (child
&& child
->frozen
) {
3465 error_setg(errp
, "Cannot change frozen '%s' link from '%s' to '%s'",
3466 child
->name
, parent_bs
->node_name
, child
->bs
->node_name
);
3470 if (is_backing
&& !parent_bs
->drv
->is_filter
&&
3471 !parent_bs
->drv
->supports_backing
)
3473 error_setg(errp
, "Driver '%s' of node '%s' does not support backing "
3474 "files", parent_bs
->drv
->format_name
, parent_bs
->node_name
);
3478 if (parent_bs
->drv
->is_filter
) {
3479 role
= BDRV_CHILD_FILTERED
| BDRV_CHILD_PRIMARY
;
3480 } else if (is_backing
) {
3481 role
= BDRV_CHILD_COW
;
3484 * We only can use same role as it is in existing child. We don't have
3485 * infrastructure to determine role of file child in generic way
3488 error_setg(errp
, "Cannot set file child to format node without "
3496 assert(child
->bs
->quiesce_counter
);
3497 bdrv_unset_inherits_from(parent_bs
, child
, tran
);
3498 bdrv_remove_child(child
, tran
);
3505 child
= bdrv_attach_child_noperm(parent_bs
, child_bs
,
3506 is_backing
? "backing" : "file",
3507 &child_of_bds
, role
,
3515 * If inherits_from pointed recursively to bs then let's update it to
3516 * point directly to bs (else it will become NULL).
3518 if (update_inherits_from
) {
3519 bdrv_set_inherits_from(child_bs
, parent_bs
, tran
);
3523 bdrv_refresh_limits(parent_bs
, tran
, NULL
);
3529 * The caller must hold the AioContext lock for @backing_hd. Both @bs and
3530 * @backing_hd can move to a different AioContext in this function. Callers must
3531 * make sure that their AioContext locking is still correct after this.
3533 * If a backing child is already present (i.e. we're detaching a node), that
3534 * child node must be drained.
3536 * After calling this function, the transaction @tran may only be completed
3537 * while holding a writer lock for the graph.
3539 static int GRAPH_WRLOCK
3540 bdrv_set_backing_noperm(BlockDriverState
*bs
,
3541 BlockDriverState
*backing_hd
,
3542 Transaction
*tran
, Error
**errp
)
3544 GLOBAL_STATE_CODE();
3545 return bdrv_set_file_or_backing_noperm(bs
, backing_hd
, true, tran
, errp
);
3548 int bdrv_set_backing_hd_drained(BlockDriverState
*bs
,
3549 BlockDriverState
*backing_hd
,
3553 Transaction
*tran
= tran_new();
3555 GLOBAL_STATE_CODE();
3556 assert(bs
->quiesce_counter
> 0);
3558 assert(bs
->backing
->bs
->quiesce_counter
> 0);
3560 bdrv_graph_wrlock(backing_hd
);
3562 ret
= bdrv_set_backing_noperm(bs
, backing_hd
, tran
, errp
);
3567 ret
= bdrv_refresh_perms(bs
, tran
, errp
);
3569 tran_finalize(tran
, ret
);
3570 bdrv_graph_wrunlock();
3574 int bdrv_set_backing_hd(BlockDriverState
*bs
, BlockDriverState
*backing_hd
,
3577 BlockDriverState
*drain_bs
= bs
->backing
? bs
->backing
->bs
: bs
;
3579 GLOBAL_STATE_CODE();
3582 bdrv_drained_begin(drain_bs
);
3583 ret
= bdrv_set_backing_hd_drained(bs
, backing_hd
, errp
);
3584 bdrv_drained_end(drain_bs
);
3585 bdrv_unref(drain_bs
);
3591 * Opens the backing file for a BlockDriverState if not yet open
3593 * bdref_key specifies the key for the image's BlockdevRef in the options QDict.
3594 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
3595 * itself, all options starting with "${bdref_key}." are considered part of the
3598 * The caller must hold the main AioContext lock.
3600 * TODO Can this be unified with bdrv_open_image()?
3602 int bdrv_open_backing_file(BlockDriverState
*bs
, QDict
*parent_options
,
3603 const char *bdref_key
, Error
**errp
)
3605 char *backing_filename
= NULL
;
3606 char *bdref_key_dot
;
3607 const char *reference
= NULL
;
3609 bool implicit_backing
= false;
3610 BlockDriverState
*backing_hd
;
3611 AioContext
*backing_hd_ctx
;
3613 QDict
*tmp_parent_options
= NULL
;
3614 Error
*local_err
= NULL
;
3616 GLOBAL_STATE_CODE();
3618 if (bs
->backing
!= NULL
) {
3622 /* NULL means an empty set of options */
3623 if (parent_options
== NULL
) {
3624 tmp_parent_options
= qdict_new();
3625 parent_options
= tmp_parent_options
;
3628 bs
->open_flags
&= ~BDRV_O_NO_BACKING
;
3630 bdref_key_dot
= g_strdup_printf("%s.", bdref_key
);
3631 qdict_extract_subqdict(parent_options
, &options
, bdref_key_dot
);
3632 g_free(bdref_key_dot
);
3635 * Caution: while qdict_get_try_str() is fine, getting non-string
3636 * types would require more care. When @parent_options come from
3637 * -blockdev or blockdev_add, its members are typed according to
3638 * the QAPI schema, but when they come from -drive, they're all
3641 reference
= qdict_get_try_str(parent_options
, bdref_key
);
3642 if (reference
|| qdict_haskey(options
, "file.filename")) {
3643 /* keep backing_filename NULL */
3644 } else if (bs
->backing_file
[0] == '\0' && qdict_size(options
) == 0) {
3645 qobject_unref(options
);
3648 if (qdict_size(options
) == 0) {
3649 /* If the user specifies options that do not modify the
3650 * backing file's behavior, we might still consider it the
3651 * implicit backing file. But it's easier this way, and
3652 * just specifying some of the backing BDS's options is
3653 * only possible with -drive anyway (otherwise the QAPI
3654 * schema forces the user to specify everything). */
3655 implicit_backing
= !strcmp(bs
->auto_backing_file
, bs
->backing_file
);
3658 bdrv_graph_rdlock_main_loop();
3659 backing_filename
= bdrv_get_full_backing_filename(bs
, &local_err
);
3660 bdrv_graph_rdunlock_main_loop();
3664 error_propagate(errp
, local_err
);
3665 qobject_unref(options
);
3670 if (!bs
->drv
|| !bs
->drv
->supports_backing
) {
3672 error_setg(errp
, "Driver doesn't support backing files");
3673 qobject_unref(options
);
3678 bs
->backing_format
[0] != '\0' && !qdict_haskey(options
, "driver")) {
3679 qdict_put_str(options
, "driver", bs
->backing_format
);
3682 backing_hd
= bdrv_open_inherit(backing_filename
, reference
, options
, 0, bs
,
3683 &child_of_bds
, bdrv_backing_role(bs
), errp
);
3685 bs
->open_flags
|= BDRV_O_NO_BACKING
;
3686 error_prepend(errp
, "Could not open backing file: ");
3691 if (implicit_backing
) {
3692 bdrv_graph_rdlock_main_loop();
3693 bdrv_refresh_filename(backing_hd
);
3694 bdrv_graph_rdunlock_main_loop();
3695 pstrcpy(bs
->auto_backing_file
, sizeof(bs
->auto_backing_file
),
3696 backing_hd
->filename
);
3699 /* Hook up the backing file link; drop our reference, bs owns the
3700 * backing_hd reference now */
3701 backing_hd_ctx
= bdrv_get_aio_context(backing_hd
);
3702 aio_context_acquire(backing_hd_ctx
);
3703 ret
= bdrv_set_backing_hd(bs
, backing_hd
, errp
);
3704 bdrv_unref(backing_hd
);
3705 aio_context_release(backing_hd_ctx
);
3711 qdict_del(parent_options
, bdref_key
);
3714 g_free(backing_filename
);
3715 qobject_unref(tmp_parent_options
);
3719 static BlockDriverState
*
3720 bdrv_open_child_bs(const char *filename
, QDict
*options
, const char *bdref_key
,
3721 BlockDriverState
*parent
, const BdrvChildClass
*child_class
,
3722 BdrvChildRole child_role
, bool allow_none
, Error
**errp
)
3724 BlockDriverState
*bs
= NULL
;
3725 QDict
*image_options
;
3726 char *bdref_key_dot
;
3727 const char *reference
;
3729 assert(child_class
!= NULL
);
3731 bdref_key_dot
= g_strdup_printf("%s.", bdref_key
);
3732 qdict_extract_subqdict(options
, &image_options
, bdref_key_dot
);
3733 g_free(bdref_key_dot
);
3736 * Caution: while qdict_get_try_str() is fine, getting non-string
3737 * types would require more care. When @options come from
3738 * -blockdev or blockdev_add, its members are typed according to
3739 * the QAPI schema, but when they come from -drive, they're all
3742 reference
= qdict_get_try_str(options
, bdref_key
);
3743 if (!filename
&& !reference
&& !qdict_size(image_options
)) {
3745 error_setg(errp
, "A block device must be specified for \"%s\"",
3748 qobject_unref(image_options
);
3752 bs
= bdrv_open_inherit(filename
, reference
, image_options
, 0,
3753 parent
, child_class
, child_role
, errp
);
3759 qdict_del(options
, bdref_key
);
3764 * Opens a disk image whose options are given as BlockdevRef in another block
3767 * If allow_none is true, no image will be opened if filename is false and no
3768 * BlockdevRef is given. NULL will be returned, but errp remains unset.
3770 * bdrev_key specifies the key for the image's BlockdevRef in the options QDict.
3771 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
3772 * itself, all options starting with "${bdref_key}." are considered part of the
3775 * The BlockdevRef will be removed from the options QDict.
3777 * The caller must hold the lock of the main AioContext and no other AioContext.
3778 * @parent can move to a different AioContext in this function. Callers must
3779 * make sure that their AioContext locking is still correct after this.
3781 BdrvChild
*bdrv_open_child(const char *filename
,
3782 QDict
*options
, const char *bdref_key
,
3783 BlockDriverState
*parent
,
3784 const BdrvChildClass
*child_class
,
3785 BdrvChildRole child_role
,
3786 bool allow_none
, Error
**errp
)
3788 BlockDriverState
*bs
;
3792 GLOBAL_STATE_CODE();
3794 bs
= bdrv_open_child_bs(filename
, options
, bdref_key
, parent
, child_class
,
3795 child_role
, allow_none
, errp
);
3800 bdrv_graph_wrlock(NULL
);
3801 ctx
= bdrv_get_aio_context(bs
);
3802 aio_context_acquire(ctx
);
3803 child
= bdrv_attach_child(parent
, bs
, bdref_key
, child_class
, child_role
,
3805 aio_context_release(ctx
);
3806 bdrv_graph_wrunlock();
3812 * Wrapper on bdrv_open_child() for most popular case: open primary child of bs.
3814 * The caller must hold the lock of the main AioContext and no other AioContext.
3815 * @parent can move to a different AioContext in this function. Callers must
3816 * make sure that their AioContext locking is still correct after this.
3818 int bdrv_open_file_child(const char *filename
,
3819 QDict
*options
, const char *bdref_key
,
3820 BlockDriverState
*parent
, Error
**errp
)
3824 /* commit_top and mirror_top don't use this function */
3825 assert(!parent
->drv
->filtered_child_is_backing
);
3826 role
= parent
->drv
->is_filter
?
3827 (BDRV_CHILD_FILTERED
| BDRV_CHILD_PRIMARY
) : BDRV_CHILD_IMAGE
;
3829 if (!bdrv_open_child(filename
, options
, bdref_key
, parent
,
3830 &child_of_bds
, role
, false, errp
))
3839 * TODO Future callers may need to specify parent/child_class in order for
3840 * option inheritance to work. Existing callers use it for the root node.
3842 BlockDriverState
*bdrv_open_blockdev_ref(BlockdevRef
*ref
, Error
**errp
)
3844 BlockDriverState
*bs
= NULL
;
3845 QObject
*obj
= NULL
;
3846 QDict
*qdict
= NULL
;
3847 const char *reference
= NULL
;
3850 GLOBAL_STATE_CODE();
3852 if (ref
->type
== QTYPE_QSTRING
) {
3853 reference
= ref
->u
.reference
;
3855 BlockdevOptions
*options
= &ref
->u
.definition
;
3856 assert(ref
->type
== QTYPE_QDICT
);
3858 v
= qobject_output_visitor_new(&obj
);
3859 visit_type_BlockdevOptions(v
, NULL
, &options
, &error_abort
);
3860 visit_complete(v
, &obj
);
3862 qdict
= qobject_to(QDict
, obj
);
3863 qdict_flatten(qdict
);
3865 /* bdrv_open_inherit() defaults to the values in bdrv_flags (for
3866 * compatibility with other callers) rather than what we want as the
3867 * real defaults. Apply the defaults here instead. */
3868 qdict_set_default_str(qdict
, BDRV_OPT_CACHE_DIRECT
, "off");
3869 qdict_set_default_str(qdict
, BDRV_OPT_CACHE_NO_FLUSH
, "off");
3870 qdict_set_default_str(qdict
, BDRV_OPT_READ_ONLY
, "off");
3871 qdict_set_default_str(qdict
, BDRV_OPT_AUTO_READ_ONLY
, "off");
3875 bs
= bdrv_open_inherit(NULL
, reference
, qdict
, 0, NULL
, NULL
, 0, errp
);
3882 static BlockDriverState
*bdrv_append_temp_snapshot(BlockDriverState
*bs
,
3884 QDict
*snapshot_options
,
3887 g_autofree
char *tmp_filename
= NULL
;
3889 QemuOpts
*opts
= NULL
;
3890 BlockDriverState
*bs_snapshot
= NULL
;
3891 AioContext
*ctx
= bdrv_get_aio_context(bs
);
3894 GLOBAL_STATE_CODE();
3896 /* if snapshot, we create a temporary backing file and open it
3897 instead of opening 'filename' directly */
3899 /* Get the required size from the image */
3900 aio_context_acquire(ctx
);
3901 total_size
= bdrv_getlength(bs
);
3902 aio_context_release(ctx
);
3904 if (total_size
< 0) {
3905 error_setg_errno(errp
, -total_size
, "Could not get image size");
3909 /* Create the temporary image */
3910 tmp_filename
= create_tmp_file(errp
);
3911 if (!tmp_filename
) {
3915 opts
= qemu_opts_create(bdrv_qcow2
.create_opts
, NULL
, 0,
3917 qemu_opt_set_number(opts
, BLOCK_OPT_SIZE
, total_size
, &error_abort
);
3918 ret
= bdrv_create(&bdrv_qcow2
, tmp_filename
, opts
, errp
);
3919 qemu_opts_del(opts
);
3921 error_prepend(errp
, "Could not create temporary overlay '%s': ",
3926 /* Prepare options QDict for the temporary file */
3927 qdict_put_str(snapshot_options
, "file.driver", "file");
3928 qdict_put_str(snapshot_options
, "file.filename", tmp_filename
);
3929 qdict_put_str(snapshot_options
, "driver", "qcow2");
3931 bs_snapshot
= bdrv_open(NULL
, NULL
, snapshot_options
, flags
, errp
);
3932 snapshot_options
= NULL
;
3937 aio_context_acquire(ctx
);
3938 ret
= bdrv_append(bs_snapshot
, bs
, errp
);
3939 aio_context_release(ctx
);
3947 qobject_unref(snapshot_options
);
3952 * Opens a disk image (raw, qcow2, vmdk, ...)
3954 * options is a QDict of options to pass to the block drivers, or NULL for an
3955 * empty set of options. The reference to the QDict belongs to the block layer
3956 * after the call (even on failure), so if the caller intends to reuse the
3957 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
3959 * If *pbs is NULL, a new BDS will be created with a pointer to it stored there.
3960 * If it is not NULL, the referenced BDS will be reused.
3962 * The reference parameter may be used to specify an existing block device which
3963 * should be opened. If specified, neither options nor a filename may be given,
3964 * nor can an existing BDS be reused (that is, *pbs has to be NULL).
3966 * The caller must always hold the main AioContext lock.
3968 static BlockDriverState
* no_coroutine_fn
3969 bdrv_open_inherit(const char *filename
, const char *reference
, QDict
*options
,
3970 int flags
, BlockDriverState
*parent
,
3971 const BdrvChildClass
*child_class
, BdrvChildRole child_role
,
3975 BlockBackend
*file
= NULL
;
3976 BlockDriverState
*bs
;
3977 BlockDriver
*drv
= NULL
;
3979 const char *drvname
;
3980 const char *backing
;
3981 Error
*local_err
= NULL
;
3982 QDict
*snapshot_options
= NULL
;
3983 int snapshot_flags
= 0;
3984 AioContext
*ctx
= qemu_get_aio_context();
3986 assert(!child_class
|| !flags
);
3987 assert(!child_class
== !parent
);
3988 GLOBAL_STATE_CODE();
3989 assert(!qemu_in_coroutine());
3991 /* TODO We'll eventually have to take a writer lock in this function */
3992 GRAPH_RDLOCK_GUARD_MAINLOOP();
3995 bool options_non_empty
= options
? qdict_size(options
) : false;
3996 qobject_unref(options
);
3998 if (filename
|| options_non_empty
) {
3999 error_setg(errp
, "Cannot reference an existing block device with "
4000 "additional options or a new filename");
4004 bs
= bdrv_lookup_bs(reference
, reference
, errp
);
4015 /* NULL means an empty set of options */
4016 if (options
== NULL
) {
4017 options
= qdict_new();
4020 /* json: syntax counts as explicit options, as if in the QDict */
4021 parse_json_protocol(options
, &filename
, &local_err
);
4026 bs
->explicit_options
= qdict_clone_shallow(options
);
4029 bool parent_is_format
;
4032 parent_is_format
= parent
->drv
->is_format
;
4035 * parent->drv is not set yet because this node is opened for
4036 * (potential) format probing. That means that @parent is going
4037 * to be a format node.
4039 parent_is_format
= true;
4042 bs
->inherits_from
= parent
;
4043 child_class
->inherit_options(child_role
, parent_is_format
,
4045 parent
->open_flags
, parent
->options
);
4048 ret
= bdrv_fill_options(&options
, filename
, &flags
, &local_err
);
4054 * Set the BDRV_O_RDWR and BDRV_O_ALLOW_RDWR flags.
4055 * Caution: getting a boolean member of @options requires care.
4056 * When @options come from -blockdev or blockdev_add, members are
4057 * typed according to the QAPI schema, but when they come from
4058 * -drive, they're all QString.
4060 if (g_strcmp0(qdict_get_try_str(options
, BDRV_OPT_READ_ONLY
), "on") &&
4061 !qdict_get_try_bool(options
, BDRV_OPT_READ_ONLY
, false)) {
4062 flags
|= (BDRV_O_RDWR
| BDRV_O_ALLOW_RDWR
);
4064 flags
&= ~BDRV_O_RDWR
;
4067 if (flags
& BDRV_O_SNAPSHOT
) {
4068 snapshot_options
= qdict_new();
4069 bdrv_temp_snapshot_options(&snapshot_flags
, snapshot_options
,
4071 /* Let bdrv_backing_options() override "read-only" */
4072 qdict_del(options
, BDRV_OPT_READ_ONLY
);
4073 bdrv_inherited_options(BDRV_CHILD_COW
, true,
4074 &flags
, options
, flags
, options
);
4077 bs
->open_flags
= flags
;
4078 bs
->options
= options
;
4079 options
= qdict_clone_shallow(options
);
4081 /* Find the right image format driver */
4082 /* See cautionary note on accessing @options above */
4083 drvname
= qdict_get_try_str(options
, "driver");
4085 drv
= bdrv_find_format(drvname
);
4087 error_setg(errp
, "Unknown driver: '%s'", drvname
);
4092 assert(drvname
|| !(flags
& BDRV_O_PROTOCOL
));
4094 /* See cautionary note on accessing @options above */
4095 backing
= qdict_get_try_str(options
, "backing");
4096 if (qobject_to(QNull
, qdict_get(options
, "backing")) != NULL
||
4097 (backing
&& *backing
== '\0'))
4100 warn_report("Use of \"backing\": \"\" is deprecated; "
4101 "use \"backing\": null instead");
4103 flags
|= BDRV_O_NO_BACKING
;
4104 qdict_del(bs
->explicit_options
, "backing");
4105 qdict_del(bs
->options
, "backing");
4106 qdict_del(options
, "backing");
4109 /* Open image file without format layer. This BlockBackend is only used for
4110 * probing, the block drivers will do their own bdrv_open_child() for the
4111 * same BDS, which is why we put the node name back into options. */
4112 if ((flags
& BDRV_O_PROTOCOL
) == 0) {
4113 BlockDriverState
*file_bs
;
4115 file_bs
= bdrv_open_child_bs(filename
, options
, "file", bs
,
4116 &child_of_bds
, BDRV_CHILD_IMAGE
,
4121 if (file_bs
!= NULL
) {
4122 /* Not requesting BLK_PERM_CONSISTENT_READ because we're only
4123 * looking at the header to guess the image format. This works even
4124 * in cases where a guest would not see a consistent state. */
4125 ctx
= bdrv_get_aio_context(file_bs
);
4126 aio_context_acquire(ctx
);
4127 file
= blk_new(ctx
, 0, BLK_PERM_ALL
);
4128 blk_insert_bs(file
, file_bs
, &local_err
);
4129 bdrv_unref(file_bs
);
4130 aio_context_release(ctx
);
4136 qdict_put_str(options
, "file", bdrv_get_node_name(file_bs
));
4140 /* Image format probing */
4143 ret
= find_image_format(file
, filename
, &drv
, &local_err
);
4148 * This option update would logically belong in bdrv_fill_options(),
4149 * but we first need to open bs->file for the probing to work, while
4150 * opening bs->file already requires the (mostly) final set of options
4151 * so that cache mode etc. can be inherited.
4153 * Adding the driver later is somewhat ugly, but it's not an option
4154 * that would ever be inherited, so it's correct. We just need to make
4155 * sure to update both bs->options (which has the full effective
4156 * options for bs) and options (which has file.* already removed).
4158 qdict_put_str(bs
->options
, "driver", drv
->format_name
);
4159 qdict_put_str(options
, "driver", drv
->format_name
);
4161 error_setg(errp
, "Must specify either driver or file");
4165 /* BDRV_O_PROTOCOL must be set iff a protocol BDS is about to be created */
4166 assert(!!(flags
& BDRV_O_PROTOCOL
) == !!drv
->bdrv_file_open
);
4167 /* file must be NULL if a protocol BDS is about to be created
4168 * (the inverse results in an error message from bdrv_open_common()) */
4169 assert(!(flags
& BDRV_O_PROTOCOL
) || !file
);
4171 /* Open the image */
4172 ret
= bdrv_open_common(bs
, file
, options
, &local_err
);
4177 /* The AioContext could have changed during bdrv_open_common() */
4178 ctx
= bdrv_get_aio_context(bs
);
4181 aio_context_acquire(ctx
);
4183 aio_context_release(ctx
);
4187 /* If there is a backing file, use it */
4188 if ((flags
& BDRV_O_NO_BACKING
) == 0) {
4189 ret
= bdrv_open_backing_file(bs
, options
, "backing", &local_err
);
4191 goto close_and_fail
;
4195 /* Remove all children options and references
4196 * from bs->options and bs->explicit_options */
4197 QLIST_FOREACH(child
, &bs
->children
, next
) {
4198 char *child_key_dot
;
4199 child_key_dot
= g_strdup_printf("%s.", child
->name
);
4200 qdict_extract_subqdict(bs
->explicit_options
, NULL
, child_key_dot
);
4201 qdict_extract_subqdict(bs
->options
, NULL
, child_key_dot
);
4202 qdict_del(bs
->explicit_options
, child
->name
);
4203 qdict_del(bs
->options
, child
->name
);
4204 g_free(child_key_dot
);
4207 /* Check if any unknown options were used */
4208 if (qdict_size(options
) != 0) {
4209 const QDictEntry
*entry
= qdict_first(options
);
4210 if (flags
& BDRV_O_PROTOCOL
) {
4211 error_setg(errp
, "Block protocol '%s' doesn't support the option "
4212 "'%s'", drv
->format_name
, entry
->key
);
4215 "Block format '%s' does not support the option '%s'",
4216 drv
->format_name
, entry
->key
);
4219 goto close_and_fail
;
4222 bdrv_parent_cb_change_media(bs
, true);
4224 qobject_unref(options
);
4227 /* For snapshot=on, create a temporary qcow2 overlay. bs points to the
4228 * temporary snapshot afterwards. */
4229 if (snapshot_flags
) {
4230 BlockDriverState
*snapshot_bs
;
4231 snapshot_bs
= bdrv_append_temp_snapshot(bs
, snapshot_flags
,
4232 snapshot_options
, &local_err
);
4233 snapshot_options
= NULL
;
4235 goto close_and_fail
;
4237 /* We are not going to return bs but the overlay on top of it
4238 * (snapshot_bs); thus, we have to drop the strong reference to bs
4239 * (which we obtained by calling bdrv_new()). bs will not be deleted,
4240 * though, because the overlay still has a reference to it. */
4241 aio_context_acquire(ctx
);
4243 aio_context_release(ctx
);
4250 aio_context_acquire(ctx
);
4252 qobject_unref(snapshot_options
);
4253 qobject_unref(bs
->explicit_options
);
4254 qobject_unref(bs
->options
);
4255 qobject_unref(options
);
4257 bs
->explicit_options
= NULL
;
4259 aio_context_release(ctx
);
4260 error_propagate(errp
, local_err
);
4264 aio_context_acquire(ctx
);
4266 aio_context_release(ctx
);
4267 qobject_unref(snapshot_options
);
4268 qobject_unref(options
);
4269 error_propagate(errp
, local_err
);
4273 /* The caller must always hold the main AioContext lock. */
4274 BlockDriverState
*bdrv_open(const char *filename
, const char *reference
,
4275 QDict
*options
, int flags
, Error
**errp
)
4277 GLOBAL_STATE_CODE();
4279 return bdrv_open_inherit(filename
, reference
, options
, flags
, NULL
,
4283 /* Return true if the NULL-terminated @list contains @str */
4284 static bool is_str_in_list(const char *str
, const char *const *list
)
4288 for (i
= 0; list
[i
] != NULL
; i
++) {
4289 if (!strcmp(str
, list
[i
])) {
4298 * Check that every option set in @bs->options is also set in
4301 * Options listed in the common_options list and in
4302 * @bs->drv->mutable_opts are skipped.
4304 * Return 0 on success, otherwise return -EINVAL and set @errp.
4306 static int bdrv_reset_options_allowed(BlockDriverState
*bs
,
4307 const QDict
*new_opts
, Error
**errp
)
4309 const QDictEntry
*e
;
4310 /* These options are common to all block drivers and are handled
4311 * in bdrv_reopen_prepare() so they can be left out of @new_opts */
4312 const char *const common_options
[] = {
4313 "node-name", "discard", "cache.direct", "cache.no-flush",
4314 "read-only", "auto-read-only", "detect-zeroes", NULL
4317 for (e
= qdict_first(bs
->options
); e
; e
= qdict_next(bs
->options
, e
)) {
4318 if (!qdict_haskey(new_opts
, e
->key
) &&
4319 !is_str_in_list(e
->key
, common_options
) &&
4320 !is_str_in_list(e
->key
, bs
->drv
->mutable_opts
)) {
4321 error_setg(errp
, "Option '%s' cannot be reset "
4322 "to its default value", e
->key
);
4331 * Returns true if @child can be reached recursively from @bs
4333 static bool GRAPH_RDLOCK
4334 bdrv_recurse_has_child(BlockDriverState
*bs
, BlockDriverState
*child
)
4342 QLIST_FOREACH(c
, &bs
->children
, next
) {
4343 if (bdrv_recurse_has_child(c
->bs
, child
)) {
4352 * Adds a BlockDriverState to a simple queue for an atomic, transactional
4353 * reopen of multiple devices.
4355 * bs_queue can either be an existing BlockReopenQueue that has had QTAILQ_INIT
4356 * already performed, or alternatively may be NULL a new BlockReopenQueue will
4357 * be created and initialized. This newly created BlockReopenQueue should be
4358 * passed back in for subsequent calls that are intended to be of the same
4361 * bs is the BlockDriverState to add to the reopen queue.
4363 * options contains the changed options for the associated bs
4364 * (the BlockReopenQueue takes ownership)
4366 * flags contains the open flags for the associated bs
4368 * returns a pointer to bs_queue, which is either the newly allocated
4369 * bs_queue, or the existing bs_queue being used.
4371 * bs is drained here and undrained by bdrv_reopen_queue_free().
4373 * To be called with bs->aio_context locked.
4375 static BlockReopenQueue
* GRAPH_RDLOCK
4376 bdrv_reopen_queue_child(BlockReopenQueue
*bs_queue
, BlockDriverState
*bs
,
4377 QDict
*options
, const BdrvChildClass
*klass
,
4378 BdrvChildRole role
, bool parent_is_format
,
4379 QDict
*parent_options
, int parent_flags
,
4384 BlockReopenQueueEntry
*bs_entry
;
4386 QDict
*old_options
, *explicit_options
, *options_copy
;
4390 GLOBAL_STATE_CODE();
4393 * Strictly speaking, draining is illegal under GRAPH_RDLOCK. We know that
4394 * we've been called with bdrv_graph_rdlock_main_loop(), though, so it's ok
4397 bdrv_drained_begin(bs
);
4399 if (bs_queue
== NULL
) {
4400 bs_queue
= g_new0(BlockReopenQueue
, 1);
4401 QTAILQ_INIT(bs_queue
);
4405 options
= qdict_new();
4408 /* Check if this BlockDriverState is already in the queue */
4409 QTAILQ_FOREACH(bs_entry
, bs_queue
, entry
) {
4410 if (bs
== bs_entry
->state
.bs
) {
4416 * Precedence of options:
4417 * 1. Explicitly passed in options (highest)
4418 * 2. Retained from explicitly set options of bs
4419 * 3. Inherited from parent node
4420 * 4. Retained from effective options of bs
4423 /* Old explicitly set values (don't overwrite by inherited value) */
4424 if (bs_entry
|| keep_old_opts
) {
4425 old_options
= qdict_clone_shallow(bs_entry
?
4426 bs_entry
->state
.explicit_options
:
4427 bs
->explicit_options
);
4428 bdrv_join_options(bs
, options
, old_options
);
4429 qobject_unref(old_options
);
4432 explicit_options
= qdict_clone_shallow(options
);
4434 /* Inherit from parent node */
4435 if (parent_options
) {
4437 klass
->inherit_options(role
, parent_is_format
, &flags
, options
,
4438 parent_flags
, parent_options
);
4440 flags
= bdrv_get_flags(bs
);
4443 if (keep_old_opts
) {
4444 /* Old values are used for options that aren't set yet */
4445 old_options
= qdict_clone_shallow(bs
->options
);
4446 bdrv_join_options(bs
, options
, old_options
);
4447 qobject_unref(old_options
);
4450 /* We have the final set of options so let's update the flags */
4451 options_copy
= qdict_clone_shallow(options
);
4452 opts
= qemu_opts_create(&bdrv_runtime_opts
, NULL
, 0, &error_abort
);
4453 qemu_opts_absorb_qdict(opts
, options_copy
, NULL
);
4454 update_flags_from_options(&flags
, opts
);
4455 qemu_opts_del(opts
);
4456 qobject_unref(options_copy
);
4458 /* bdrv_open_inherit() sets and clears some additional flags internally */
4459 flags
&= ~BDRV_O_PROTOCOL
;
4460 if (flags
& BDRV_O_RDWR
) {
4461 flags
|= BDRV_O_ALLOW_RDWR
;
4465 bs_entry
= g_new0(BlockReopenQueueEntry
, 1);
4466 QTAILQ_INSERT_TAIL(bs_queue
, bs_entry
, entry
);
4468 qobject_unref(bs_entry
->state
.options
);
4469 qobject_unref(bs_entry
->state
.explicit_options
);
4472 bs_entry
->state
.bs
= bs
;
4473 bs_entry
->state
.options
= options
;
4474 bs_entry
->state
.explicit_options
= explicit_options
;
4475 bs_entry
->state
.flags
= flags
;
4478 * If keep_old_opts is false then it means that unspecified
4479 * options must be reset to their original value. We don't allow
4480 * resetting 'backing' but we need to know if the option is
4481 * missing in order to decide if we have to return an error.
4483 if (!keep_old_opts
) {
4484 bs_entry
->state
.backing_missing
=
4485 !qdict_haskey(options
, "backing") &&
4486 !qdict_haskey(options
, "backing.driver");
4489 QLIST_FOREACH(child
, &bs
->children
, next
) {
4490 QDict
*new_child_options
= NULL
;
4491 bool child_keep_old
= keep_old_opts
;
4493 /* reopen can only change the options of block devices that were
4494 * implicitly created and inherited options. For other (referenced)
4495 * block devices, a syntax like "backing.foo" results in an error. */
4496 if (child
->bs
->inherits_from
!= bs
) {
4500 /* Check if the options contain a child reference */
4501 if (qdict_haskey(options
, child
->name
)) {
4502 const char *childref
= qdict_get_try_str(options
, child
->name
);
4504 * The current child must not be reopened if the child
4505 * reference is null or points to a different node.
4507 if (g_strcmp0(childref
, child
->bs
->node_name
)) {
4511 * If the child reference points to the current child then
4512 * reopen it with its existing set of options (note that
4513 * it can still inherit new options from the parent).
4515 child_keep_old
= true;
4517 /* Extract child options ("child-name.*") */
4518 char *child_key_dot
= g_strdup_printf("%s.", child
->name
);
4519 qdict_extract_subqdict(explicit_options
, NULL
, child_key_dot
);
4520 qdict_extract_subqdict(options
, &new_child_options
, child_key_dot
);
4521 g_free(child_key_dot
);
4524 bdrv_reopen_queue_child(bs_queue
, child
->bs
, new_child_options
,
4525 child
->klass
, child
->role
, bs
->drv
->is_format
,
4526 options
, flags
, child_keep_old
);
4532 /* To be called with bs->aio_context locked */
4533 BlockReopenQueue
*bdrv_reopen_queue(BlockReopenQueue
*bs_queue
,
4534 BlockDriverState
*bs
,
4535 QDict
*options
, bool keep_old_opts
)
4537 GLOBAL_STATE_CODE();
4538 GRAPH_RDLOCK_GUARD_MAINLOOP();
4540 return bdrv_reopen_queue_child(bs_queue
, bs
, options
, NULL
, 0, false,
4541 NULL
, 0, keep_old_opts
);
4544 void bdrv_reopen_queue_free(BlockReopenQueue
*bs_queue
)
4546 GLOBAL_STATE_CODE();
4548 BlockReopenQueueEntry
*bs_entry
, *next
;
4549 QTAILQ_FOREACH_SAFE(bs_entry
, bs_queue
, entry
, next
) {
4550 AioContext
*ctx
= bdrv_get_aio_context(bs_entry
->state
.bs
);
4552 aio_context_acquire(ctx
);
4553 bdrv_drained_end(bs_entry
->state
.bs
);
4554 aio_context_release(ctx
);
4556 qobject_unref(bs_entry
->state
.explicit_options
);
4557 qobject_unref(bs_entry
->state
.options
);
4565 * Reopen multiple BlockDriverStates atomically & transactionally.
4567 * The queue passed in (bs_queue) must have been built up previous
4568 * via bdrv_reopen_queue().
4570 * Reopens all BDS specified in the queue, with the appropriate
4571 * flags. All devices are prepared for reopen, and failure of any
4572 * device will cause all device changes to be abandoned, and intermediate
4575 * If all devices prepare successfully, then the changes are committed
4578 * All affected nodes must be drained between bdrv_reopen_queue() and
4579 * bdrv_reopen_multiple().
4581 * To be called from the main thread, with all other AioContexts unlocked.
4583 int bdrv_reopen_multiple(BlockReopenQueue
*bs_queue
, Error
**errp
)
4586 BlockReopenQueueEntry
*bs_entry
, *next
;
4588 Transaction
*tran
= tran_new();
4589 g_autoptr(GSList
) refresh_list
= NULL
;
4591 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
4592 assert(bs_queue
!= NULL
);
4593 GLOBAL_STATE_CODE();
4595 QTAILQ_FOREACH(bs_entry
, bs_queue
, entry
) {
4596 ctx
= bdrv_get_aio_context(bs_entry
->state
.bs
);
4597 aio_context_acquire(ctx
);
4598 ret
= bdrv_flush(bs_entry
->state
.bs
);
4599 aio_context_release(ctx
);
4601 error_setg_errno(errp
, -ret
, "Error flushing drive");
4606 QTAILQ_FOREACH(bs_entry
, bs_queue
, entry
) {
4607 assert(bs_entry
->state
.bs
->quiesce_counter
> 0);
4608 ctx
= bdrv_get_aio_context(bs_entry
->state
.bs
);
4609 aio_context_acquire(ctx
);
4610 ret
= bdrv_reopen_prepare(&bs_entry
->state
, bs_queue
, tran
, errp
);
4611 aio_context_release(ctx
);
4615 bs_entry
->prepared
= true;
4618 QTAILQ_FOREACH(bs_entry
, bs_queue
, entry
) {
4619 BDRVReopenState
*state
= &bs_entry
->state
;
4621 refresh_list
= g_slist_prepend(refresh_list
, state
->bs
);
4622 if (state
->old_backing_bs
) {
4623 refresh_list
= g_slist_prepend(refresh_list
, state
->old_backing_bs
);
4625 if (state
->old_file_bs
) {
4626 refresh_list
= g_slist_prepend(refresh_list
, state
->old_file_bs
);
4631 * Note that file-posix driver rely on permission update done during reopen
4632 * (even if no permission changed), because it wants "new" permissions for
4633 * reconfiguring the fd and that's why it does it in raw_check_perm(), not
4634 * in raw_reopen_prepare() which is called with "old" permissions.
4636 bdrv_graph_rdlock_main_loop();
4637 ret
= bdrv_list_refresh_perms(refresh_list
, bs_queue
, tran
, errp
);
4638 bdrv_graph_rdunlock_main_loop();
4645 * If we reach this point, we have success and just need to apply the
4648 * Reverse order is used to comfort qcow2 driver: on commit it need to write
4649 * IN_USE flag to the image, to mark bitmaps in the image as invalid. But
4650 * children are usually goes after parents in reopen-queue, so go from last
4653 QTAILQ_FOREACH_REVERSE(bs_entry
, bs_queue
, entry
) {
4654 ctx
= bdrv_get_aio_context(bs_entry
->state
.bs
);
4655 aio_context_acquire(ctx
);
4656 bdrv_reopen_commit(&bs_entry
->state
);
4657 aio_context_release(ctx
);
4660 bdrv_graph_wrlock(NULL
);
4662 bdrv_graph_wrunlock();
4664 QTAILQ_FOREACH_REVERSE(bs_entry
, bs_queue
, entry
) {
4665 BlockDriverState
*bs
= bs_entry
->state
.bs
;
4667 if (bs
->drv
->bdrv_reopen_commit_post
) {
4668 ctx
= bdrv_get_aio_context(bs
);
4669 aio_context_acquire(ctx
);
4670 bs
->drv
->bdrv_reopen_commit_post(&bs_entry
->state
);
4671 aio_context_release(ctx
);
4679 bdrv_graph_wrlock(NULL
);
4681 bdrv_graph_wrunlock();
4683 QTAILQ_FOREACH_SAFE(bs_entry
, bs_queue
, entry
, next
) {
4684 if (bs_entry
->prepared
) {
4685 ctx
= bdrv_get_aio_context(bs_entry
->state
.bs
);
4686 aio_context_acquire(ctx
);
4687 bdrv_reopen_abort(&bs_entry
->state
);
4688 aio_context_release(ctx
);
4693 bdrv_reopen_queue_free(bs_queue
);
4698 int bdrv_reopen(BlockDriverState
*bs
, QDict
*opts
, bool keep_old_opts
,
4701 AioContext
*ctx
= bdrv_get_aio_context(bs
);
4702 BlockReopenQueue
*queue
;
4705 GLOBAL_STATE_CODE();
4707 queue
= bdrv_reopen_queue(NULL
, bs
, opts
, keep_old_opts
);
4709 if (ctx
!= qemu_get_aio_context()) {
4710 aio_context_release(ctx
);
4712 ret
= bdrv_reopen_multiple(queue
, errp
);
4714 if (ctx
!= qemu_get_aio_context()) {
4715 aio_context_acquire(ctx
);
4721 int bdrv_reopen_set_read_only(BlockDriverState
*bs
, bool read_only
,
4724 QDict
*opts
= qdict_new();
4726 GLOBAL_STATE_CODE();
4728 qdict_put_bool(opts
, BDRV_OPT_READ_ONLY
, read_only
);
4730 return bdrv_reopen(bs
, opts
, true, errp
);
4734 * Take a BDRVReopenState and check if the value of 'backing' in the
4735 * reopen_state->options QDict is valid or not.
4737 * If 'backing' is missing from the QDict then return 0.
4739 * If 'backing' contains the node name of the backing file of
4740 * reopen_state->bs then return 0.
4742 * If 'backing' contains a different node name (or is null) then check
4743 * whether the current backing file can be replaced with the new one.
4744 * If that's the case then reopen_state->replace_backing_bs is set to
4745 * true and reopen_state->new_backing_bs contains a pointer to the new
4746 * backing BlockDriverState (or NULL).
4748 * After calling this function, the transaction @tran may only be completed
4749 * while holding a writer lock for the graph.
4751 * Return 0 on success, otherwise return < 0 and set @errp.
4753 * The caller must hold the AioContext lock of @reopen_state->bs.
4754 * @reopen_state->bs can move to a different AioContext in this function.
4755 * Callers must make sure that their AioContext locking is still correct after
4758 static int GRAPH_UNLOCKED
4759 bdrv_reopen_parse_file_or_backing(BDRVReopenState
*reopen_state
,
4760 bool is_backing
, Transaction
*tran
,
4763 BlockDriverState
*bs
= reopen_state
->bs
;
4764 BlockDriverState
*new_child_bs
;
4765 BlockDriverState
*old_child_bs
= is_backing
? child_bs(bs
->backing
) :
4767 const char *child_name
= is_backing
? "backing" : "file";
4770 AioContext
*ctx
, *old_ctx
;
4774 GLOBAL_STATE_CODE();
4776 value
= qdict_get(reopen_state
->options
, child_name
);
4777 if (value
== NULL
) {
4781 switch (qobject_type(value
)) {
4783 assert(is_backing
); /* The 'file' option does not allow a null value */
4784 new_child_bs
= NULL
;
4787 str
= qstring_get_str(qobject_to(QString
, value
));
4788 new_child_bs
= bdrv_lookup_bs(NULL
, str
, errp
);
4789 if (new_child_bs
== NULL
) {
4793 bdrv_graph_rdlock_main_loop();
4794 has_child
= bdrv_recurse_has_child(new_child_bs
, bs
);
4795 bdrv_graph_rdunlock_main_loop();
4798 error_setg(errp
, "Making '%s' a %s child of '%s' would create a "
4799 "cycle", str
, child_name
, bs
->node_name
);
4805 * The options QDict has been flattened, so 'backing' and 'file'
4806 * do not allow any other data type here.
4808 g_assert_not_reached();
4811 if (old_child_bs
== new_child_bs
) {
4816 if (bdrv_skip_implicit_filters(old_child_bs
) == new_child_bs
) {
4820 if (old_child_bs
->implicit
) {
4821 error_setg(errp
, "Cannot replace implicit %s child of %s",
4822 child_name
, bs
->node_name
);
4827 if (bs
->drv
->is_filter
&& !old_child_bs
) {
4829 * Filters always have a file or a backing child, so we are trying to
4830 * change wrong child
4832 error_setg(errp
, "'%s' is a %s filter node that does not support a "
4833 "%s child", bs
->node_name
, bs
->drv
->format_name
, child_name
);
4838 reopen_state
->old_backing_bs
= old_child_bs
;
4840 reopen_state
->old_file_bs
= old_child_bs
;
4844 bdrv_ref(old_child_bs
);
4845 bdrv_drained_begin(old_child_bs
);
4848 old_ctx
= bdrv_get_aio_context(bs
);
4849 ctx
= bdrv_get_aio_context(new_child_bs
);
4850 if (old_ctx
!= ctx
) {
4851 aio_context_release(old_ctx
);
4852 aio_context_acquire(ctx
);
4855 bdrv_graph_wrlock(new_child_bs
);
4857 ret
= bdrv_set_file_or_backing_noperm(bs
, new_child_bs
, is_backing
,
4860 bdrv_graph_wrunlock();
4862 if (old_ctx
!= ctx
) {
4863 aio_context_release(ctx
);
4864 aio_context_acquire(old_ctx
);
4868 bdrv_drained_end(old_child_bs
);
4869 bdrv_unref(old_child_bs
);
4876 * Prepares a BlockDriverState for reopen. All changes are staged in the
4877 * 'opaque' field of the BDRVReopenState, which is used and allocated by
4878 * the block driver layer .bdrv_reopen_prepare()
4880 * bs is the BlockDriverState to reopen
4881 * flags are the new open flags
4882 * queue is the reopen queue
4884 * Returns 0 on success, non-zero on error. On error errp will be set
4887 * On failure, bdrv_reopen_abort() will be called to clean up any data.
4888 * It is the responsibility of the caller to then call the abort() or
4889 * commit() for any other BDS that have been left in a prepare() state
4891 * The caller must hold the AioContext lock of @reopen_state->bs.
4893 * After calling this function, the transaction @change_child_tran may only be
4894 * completed while holding a writer lock for the graph.
4896 static int GRAPH_UNLOCKED
4897 bdrv_reopen_prepare(BDRVReopenState
*reopen_state
, BlockReopenQueue
*queue
,
4898 Transaction
*change_child_tran
, Error
**errp
)
4902 Error
*local_err
= NULL
;
4905 QDict
*orig_reopen_opts
;
4906 char *discard
= NULL
;
4908 bool drv_prepared
= false;
4910 assert(reopen_state
!= NULL
);
4911 assert(reopen_state
->bs
->drv
!= NULL
);
4912 GLOBAL_STATE_CODE();
4913 drv
= reopen_state
->bs
->drv
;
4915 /* This function and each driver's bdrv_reopen_prepare() remove
4916 * entries from reopen_state->options as they are processed, so
4917 * we need to make a copy of the original QDict. */
4918 orig_reopen_opts
= qdict_clone_shallow(reopen_state
->options
);
4920 /* Process generic block layer options */
4921 opts
= qemu_opts_create(&bdrv_runtime_opts
, NULL
, 0, &error_abort
);
4922 if (!qemu_opts_absorb_qdict(opts
, reopen_state
->options
, errp
)) {
4927 /* This was already called in bdrv_reopen_queue_child() so the flags
4928 * are up-to-date. This time we simply want to remove the options from
4929 * QemuOpts in order to indicate that they have been processed. */
4930 old_flags
= reopen_state
->flags
;
4931 update_flags_from_options(&reopen_state
->flags
, opts
);
4932 assert(old_flags
== reopen_state
->flags
);
4934 discard
= qemu_opt_get_del(opts
, BDRV_OPT_DISCARD
);
4935 if (discard
!= NULL
) {
4936 if (bdrv_parse_discard_flags(discard
, &reopen_state
->flags
) != 0) {
4937 error_setg(errp
, "Invalid discard option");
4943 reopen_state
->detect_zeroes
=
4944 bdrv_parse_detect_zeroes(opts
, reopen_state
->flags
, &local_err
);
4946 error_propagate(errp
, local_err
);
4951 /* All other options (including node-name and driver) must be unchanged.
4952 * Put them back into the QDict, so that they are checked at the end
4953 * of this function. */
4954 qemu_opts_to_qdict(opts
, reopen_state
->options
);
4956 /* If we are to stay read-only, do not allow permission change
4957 * to r/w. Attempting to set to r/w may fail if either BDRV_O_ALLOW_RDWR is
4958 * not set, or if the BDS still has copy_on_read enabled */
4959 read_only
= !(reopen_state
->flags
& BDRV_O_RDWR
);
4961 bdrv_graph_rdlock_main_loop();
4962 ret
= bdrv_can_set_read_only(reopen_state
->bs
, read_only
, true, &local_err
);
4963 bdrv_graph_rdunlock_main_loop();
4965 error_propagate(errp
, local_err
);
4969 if (drv
->bdrv_reopen_prepare
) {
4971 * If a driver-specific option is missing, it means that we
4972 * should reset it to its default value.
4973 * But not all options allow that, so we need to check it first.
4975 ret
= bdrv_reset_options_allowed(reopen_state
->bs
,
4976 reopen_state
->options
, errp
);
4981 ret
= drv
->bdrv_reopen_prepare(reopen_state
, queue
, &local_err
);
4983 if (local_err
!= NULL
) {
4984 error_propagate(errp
, local_err
);
4986 bdrv_graph_rdlock_main_loop();
4987 bdrv_refresh_filename(reopen_state
->bs
);
4988 bdrv_graph_rdunlock_main_loop();
4989 error_setg(errp
, "failed while preparing to reopen image '%s'",
4990 reopen_state
->bs
->filename
);
4995 /* It is currently mandatory to have a bdrv_reopen_prepare()
4996 * handler for each supported drv. */
4997 bdrv_graph_rdlock_main_loop();
4998 error_setg(errp
, "Block format '%s' used by node '%s' "
4999 "does not support reopening files", drv
->format_name
,
5000 bdrv_get_device_or_node_name(reopen_state
->bs
));
5001 bdrv_graph_rdunlock_main_loop();
5006 drv_prepared
= true;
5009 * We must provide the 'backing' option if the BDS has a backing
5010 * file or if the image file has a backing file name as part of
5011 * its metadata. Otherwise the 'backing' option can be omitted.
5013 if (drv
->supports_backing
&& reopen_state
->backing_missing
&&
5014 (reopen_state
->bs
->backing
|| reopen_state
->bs
->backing_file
[0])) {
5015 error_setg(errp
, "backing is missing for '%s'",
5016 reopen_state
->bs
->node_name
);
5022 * Allow changing the 'backing' option. The new value can be
5023 * either a reference to an existing node (using its node name)
5024 * or NULL to simply detach the current backing file.
5026 ret
= bdrv_reopen_parse_file_or_backing(reopen_state
, true,
5027 change_child_tran
, errp
);
5031 qdict_del(reopen_state
->options
, "backing");
5033 /* Allow changing the 'file' option. In this case NULL is not allowed */
5034 ret
= bdrv_reopen_parse_file_or_backing(reopen_state
, false,
5035 change_child_tran
, errp
);
5039 qdict_del(reopen_state
->options
, "file");
5041 /* Options that are not handled are only okay if they are unchanged
5042 * compared to the old state. It is expected that some options are only
5043 * used for the initial open, but not reopen (e.g. filename) */
5044 if (qdict_size(reopen_state
->options
)) {
5045 const QDictEntry
*entry
= qdict_first(reopen_state
->options
);
5047 GRAPH_RDLOCK_GUARD_MAINLOOP();
5050 QObject
*new = entry
->value
;
5051 QObject
*old
= qdict_get(reopen_state
->bs
->options
, entry
->key
);
5053 /* Allow child references (child_name=node_name) as long as they
5054 * point to the current child (i.e. everything stays the same). */
5055 if (qobject_type(new) == QTYPE_QSTRING
) {
5057 QLIST_FOREACH(child
, &reopen_state
->bs
->children
, next
) {
5058 if (!strcmp(child
->name
, entry
->key
)) {
5064 if (!strcmp(child
->bs
->node_name
,
5065 qstring_get_str(qobject_to(QString
, new)))) {
5066 continue; /* Found child with this name, skip option */
5072 * TODO: When using -drive to specify blockdev options, all values
5073 * will be strings; however, when using -blockdev, blockdev-add or
5074 * filenames using the json:{} pseudo-protocol, they will be
5076 * In contrast, reopening options are (currently) always strings
5077 * (because you can only specify them through qemu-io; all other
5078 * callers do not specify any options).
5079 * Therefore, when using anything other than -drive to create a BDS,
5080 * this cannot detect non-string options as unchanged, because
5081 * qobject_is_equal() always returns false for objects of different
5082 * type. In the future, this should be remedied by correctly typing
5083 * all options. For now, this is not too big of an issue because
5084 * the user can simply omit options which cannot be changed anyway,
5085 * so they will stay unchanged.
5087 if (!qobject_is_equal(new, old
)) {
5088 error_setg(errp
, "Cannot change the option '%s'", entry
->key
);
5092 } while ((entry
= qdict_next(reopen_state
->options
, entry
)));
5097 /* Restore the original reopen_state->options QDict */
5098 qobject_unref(reopen_state
->options
);
5099 reopen_state
->options
= qobject_ref(orig_reopen_opts
);
5102 if (ret
< 0 && drv_prepared
) {
5103 /* drv->bdrv_reopen_prepare() has succeeded, so we need to
5104 * call drv->bdrv_reopen_abort() before signaling an error
5105 * (bdrv_reopen_multiple() will not call bdrv_reopen_abort()
5106 * when the respective bdrv_reopen_prepare() has failed) */
5107 if (drv
->bdrv_reopen_abort
) {
5108 drv
->bdrv_reopen_abort(reopen_state
);
5111 qemu_opts_del(opts
);
5112 qobject_unref(orig_reopen_opts
);
5118 * Takes the staged changes for the reopen from bdrv_reopen_prepare(), and
5119 * makes them final by swapping the staging BlockDriverState contents into
5120 * the active BlockDriverState contents.
5122 static void GRAPH_UNLOCKED
bdrv_reopen_commit(BDRVReopenState
*reopen_state
)
5125 BlockDriverState
*bs
;
5128 assert(reopen_state
!= NULL
);
5129 bs
= reopen_state
->bs
;
5131 assert(drv
!= NULL
);
5132 GLOBAL_STATE_CODE();
5134 /* If there are any driver level actions to take */
5135 if (drv
->bdrv_reopen_commit
) {
5136 drv
->bdrv_reopen_commit(reopen_state
);
5139 GRAPH_RDLOCK_GUARD_MAINLOOP();
5141 /* set BDS specific flags now */
5142 qobject_unref(bs
->explicit_options
);
5143 qobject_unref(bs
->options
);
5144 qobject_ref(reopen_state
->explicit_options
);
5145 qobject_ref(reopen_state
->options
);
5147 bs
->explicit_options
= reopen_state
->explicit_options
;
5148 bs
->options
= reopen_state
->options
;
5149 bs
->open_flags
= reopen_state
->flags
;
5150 bs
->detect_zeroes
= reopen_state
->detect_zeroes
;
5152 /* Remove child references from bs->options and bs->explicit_options.
5153 * Child options were already removed in bdrv_reopen_queue_child() */
5154 QLIST_FOREACH(child
, &bs
->children
, next
) {
5155 qdict_del(bs
->explicit_options
, child
->name
);
5156 qdict_del(bs
->options
, child
->name
);
5158 /* backing is probably removed, so it's not handled by previous loop */
5159 qdict_del(bs
->explicit_options
, "backing");
5160 qdict_del(bs
->options
, "backing");
5162 bdrv_refresh_limits(bs
, NULL
, NULL
);
5163 bdrv_refresh_total_sectors(bs
, bs
->total_sectors
);
5167 * Abort the reopen, and delete and free the staged changes in
5170 static void GRAPH_UNLOCKED
bdrv_reopen_abort(BDRVReopenState
*reopen_state
)
5174 assert(reopen_state
!= NULL
);
5175 drv
= reopen_state
->bs
->drv
;
5176 assert(drv
!= NULL
);
5177 GLOBAL_STATE_CODE();
5179 if (drv
->bdrv_reopen_abort
) {
5180 drv
->bdrv_reopen_abort(reopen_state
);
5185 static void bdrv_close(BlockDriverState
*bs
)
5187 BdrvAioNotifier
*ban
, *ban_next
;
5188 BdrvChild
*child
, *next
;
5190 GLOBAL_STATE_CODE();
5191 assert(!bs
->refcnt
);
5193 bdrv_drained_begin(bs
); /* complete I/O */
5195 bdrv_drain(bs
); /* in case flush left pending I/O */
5198 if (bs
->drv
->bdrv_close
) {
5199 /* Must unfreeze all children, so bdrv_unref_child() works */
5200 bs
->drv
->bdrv_close(bs
);
5205 bdrv_graph_wrlock(bs
);
5206 QLIST_FOREACH_SAFE(child
, &bs
->children
, next
, next
) {
5207 bdrv_unref_child(bs
, child
);
5209 bdrv_graph_wrunlock();
5211 assert(!bs
->backing
);
5215 qatomic_set(&bs
->copy_on_read
, 0);
5216 bs
->backing_file
[0] = '\0';
5217 bs
->backing_format
[0] = '\0';
5218 bs
->total_sectors
= 0;
5219 bs
->encrypted
= false;
5221 qobject_unref(bs
->options
);
5222 qobject_unref(bs
->explicit_options
);
5224 bs
->explicit_options
= NULL
;
5225 qobject_unref(bs
->full_open_options
);
5226 bs
->full_open_options
= NULL
;
5227 g_free(bs
->block_status_cache
);
5228 bs
->block_status_cache
= NULL
;
5230 bdrv_release_named_dirty_bitmaps(bs
);
5231 assert(QLIST_EMPTY(&bs
->dirty_bitmaps
));
5233 QLIST_FOREACH_SAFE(ban
, &bs
->aio_notifiers
, list
, ban_next
) {
5236 QLIST_INIT(&bs
->aio_notifiers
);
5237 bdrv_drained_end(bs
);
5240 * If we're still inside some bdrv_drain_all_begin()/end() sections, end
5241 * them now since this BDS won't exist anymore when bdrv_drain_all_end()
5244 if (bs
->quiesce_counter
) {
5245 bdrv_drain_all_end_quiesce(bs
);
5249 void bdrv_close_all(void)
5251 GLOBAL_STATE_CODE();
5252 assert(job_next(NULL
) == NULL
);
5254 /* Drop references from requests still in flight, such as canceled block
5255 * jobs whose AIO context has not been polled yet */
5258 blk_remove_all_bs();
5259 blockdev_close_all_bdrv_states();
5261 assert(QTAILQ_EMPTY(&all_bdrv_states
));
5264 static bool GRAPH_RDLOCK
should_update_child(BdrvChild
*c
, BlockDriverState
*to
)
5270 if (c
->klass
->stay_at_node
) {
5274 /* If the child @c belongs to the BDS @to, replacing the current
5275 * c->bs by @to would mean to create a loop.
5277 * Such a case occurs when appending a BDS to a backing chain.
5278 * For instance, imagine the following chain:
5280 * guest device -> node A -> further backing chain...
5282 * Now we create a new BDS B which we want to put on top of this
5283 * chain, so we first attach A as its backing node:
5288 * guest device -> node A -> further backing chain...
5290 * Finally we want to replace A by B. When doing that, we want to
5291 * replace all pointers to A by pointers to B -- except for the
5292 * pointer from B because (1) that would create a loop, and (2)
5293 * that pointer should simply stay intact:
5295 * guest device -> node B
5298 * node A -> further backing chain...
5300 * In general, when replacing a node A (c->bs) by a node B (@to),
5301 * if A is a child of B, that means we cannot replace A by B there
5302 * because that would create a loop. Silently detaching A from B
5303 * is also not really an option. So overall just leaving A in
5304 * place there is the most sensible choice.
5306 * We would also create a loop in any cases where @c is only
5307 * indirectly referenced by @to. Prevent this by returning false
5308 * if @c is found (by breadth-first search) anywhere in the whole
5313 found
= g_hash_table_new(NULL
, NULL
);
5314 g_hash_table_add(found
, to
);
5315 queue
= g_queue_new();
5316 g_queue_push_tail(queue
, to
);
5318 while (!g_queue_is_empty(queue
)) {
5319 BlockDriverState
*v
= g_queue_pop_head(queue
);
5322 QLIST_FOREACH(c2
, &v
->children
, next
) {
5328 if (g_hash_table_contains(found
, c2
->bs
)) {
5332 g_queue_push_tail(queue
, c2
->bs
);
5333 g_hash_table_add(found
, c2
->bs
);
5337 g_queue_free(queue
);
5338 g_hash_table_destroy(found
);
5343 static void bdrv_remove_child_commit(void *opaque
)
5345 GLOBAL_STATE_CODE();
5346 bdrv_child_free(opaque
);
5349 static TransactionActionDrv bdrv_remove_child_drv
= {
5350 .commit
= bdrv_remove_child_commit
,
5354 * Function doesn't update permissions, caller is responsible for this.
5356 * @child->bs (if non-NULL) must be drained.
5358 * After calling this function, the transaction @tran may only be completed
5359 * while holding a writer lock for the graph.
5361 static void GRAPH_WRLOCK
bdrv_remove_child(BdrvChild
*child
, Transaction
*tran
)
5368 assert(child
->quiesced_parent
);
5369 bdrv_replace_child_tran(child
, NULL
, tran
);
5372 tran_add(tran
, &bdrv_remove_child_drv
, child
);
5376 * Both @from and @to (if non-NULL) must be drained. @to must be kept drained
5377 * until the transaction is completed.
5379 * After calling this function, the transaction @tran may only be completed
5380 * while holding a writer lock for the graph.
5382 static int GRAPH_WRLOCK
5383 bdrv_replace_node_noperm(BlockDriverState
*from
,
5384 BlockDriverState
*to
,
5385 bool auto_skip
, Transaction
*tran
,
5388 BdrvChild
*c
, *next
;
5390 GLOBAL_STATE_CODE();
5392 assert(from
->quiesce_counter
);
5393 assert(to
->quiesce_counter
);
5395 QLIST_FOREACH_SAFE(c
, &from
->parents
, next_parent
, next
) {
5396 assert(c
->bs
== from
);
5397 if (!should_update_child(c
, to
)) {
5401 error_setg(errp
, "Should not change '%s' link to '%s'",
5402 c
->name
, from
->node_name
);
5406 error_setg(errp
, "Cannot change '%s' link to '%s'",
5407 c
->name
, from
->node_name
);
5410 bdrv_replace_child_tran(c
, to
, tran
);
5417 * With auto_skip=true bdrv_replace_node_common skips updating from parents
5418 * if it creates a parent-child relation loop or if parent is block-job.
5420 * With auto_skip=false the error is returned if from has a parent which should
5423 * With @detach_subchain=true @to must be in a backing chain of @from. In this
5424 * case backing link of the cow-parent of @to is removed.
5426 static int bdrv_replace_node_common(BlockDriverState
*from
,
5427 BlockDriverState
*to
,
5428 bool auto_skip
, bool detach_subchain
,
5431 Transaction
*tran
= tran_new();
5432 g_autoptr(GSList
) refresh_list
= NULL
;
5433 BlockDriverState
*to_cow_parent
= NULL
;
5436 GLOBAL_STATE_CODE();
5438 /* Make sure that @from doesn't go away until we have successfully attached
5439 * all of its parents to @to. */
5442 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
5443 assert(bdrv_get_aio_context(from
) == bdrv_get_aio_context(to
));
5444 bdrv_drained_begin(from
);
5445 bdrv_drained_begin(to
);
5447 bdrv_graph_wrlock(to
);
5449 if (detach_subchain
) {
5450 assert(bdrv_chain_contains(from
, to
));
5452 for (to_cow_parent
= from
;
5453 bdrv_filter_or_cow_bs(to_cow_parent
) != to
;
5454 to_cow_parent
= bdrv_filter_or_cow_bs(to_cow_parent
))
5461 * Do the replacement without permission update.
5462 * Replacement may influence the permissions, we should calculate new
5463 * permissions based on new graph. If we fail, we'll roll-back the
5466 ret
= bdrv_replace_node_noperm(from
, to
, auto_skip
, tran
, errp
);
5471 if (detach_subchain
) {
5472 /* to_cow_parent is already drained because from is drained */
5473 bdrv_remove_child(bdrv_filter_or_cow_child(to_cow_parent
), tran
);
5476 refresh_list
= g_slist_prepend(refresh_list
, to
);
5477 refresh_list
= g_slist_prepend(refresh_list
, from
);
5479 ret
= bdrv_list_refresh_perms(refresh_list
, NULL
, tran
, errp
);
5487 tran_finalize(tran
, ret
);
5488 bdrv_graph_wrunlock();
5490 bdrv_drained_end(to
);
5491 bdrv_drained_end(from
);
5497 int bdrv_replace_node(BlockDriverState
*from
, BlockDriverState
*to
,
5500 GLOBAL_STATE_CODE();
5502 return bdrv_replace_node_common(from
, to
, true, false, errp
);
5505 int bdrv_drop_filter(BlockDriverState
*bs
, Error
**errp
)
5507 BlockDriverState
*child_bs
;
5509 GLOBAL_STATE_CODE();
5510 bdrv_graph_rdlock_main_loop();
5511 child_bs
= bdrv_filter_or_cow_bs(bs
);
5512 bdrv_graph_rdunlock_main_loop();
5514 return bdrv_replace_node_common(bs
, child_bs
, true, true, errp
);
5518 * Add new bs contents at the top of an image chain while the chain is
5519 * live, while keeping required fields on the top layer.
5521 * This will modify the BlockDriverState fields, and swap contents
5522 * between bs_new and bs_top. Both bs_new and bs_top are modified.
5524 * bs_new must not be attached to a BlockBackend and must not have backing
5527 * This function does not create any image files.
5529 * The caller must hold the AioContext lock for @bs_top.
5531 int bdrv_append(BlockDriverState
*bs_new
, BlockDriverState
*bs_top
,
5536 Transaction
*tran
= tran_new();
5537 AioContext
*old_context
, *new_context
= NULL
;
5539 GLOBAL_STATE_CODE();
5541 assert(!bs_new
->backing
);
5543 old_context
= bdrv_get_aio_context(bs_top
);
5544 bdrv_drained_begin(bs_top
);
5547 * bdrv_drained_begin() requires that only the AioContext of the drained
5548 * node is locked, and at this point it can still differ from the AioContext
5551 new_context
= bdrv_get_aio_context(bs_new
);
5552 aio_context_release(old_context
);
5553 aio_context_acquire(new_context
);
5554 bdrv_drained_begin(bs_new
);
5555 aio_context_release(new_context
);
5556 aio_context_acquire(old_context
);
5559 bdrv_graph_wrlock(bs_top
);
5561 child
= bdrv_attach_child_noperm(bs_new
, bs_top
, "backing",
5562 &child_of_bds
, bdrv_backing_role(bs_new
),
5570 * bdrv_attach_child_noperm could change the AioContext of bs_top and
5571 * bs_new, but at least they are in the same AioContext now. This is the
5572 * AioContext that we need to lock for the rest of the function.
5574 new_context
= bdrv_get_aio_context(bs_top
);
5576 if (old_context
!= new_context
) {
5577 aio_context_release(old_context
);
5578 aio_context_acquire(new_context
);
5581 ret
= bdrv_replace_node_noperm(bs_top
, bs_new
, true, tran
, errp
);
5586 ret
= bdrv_refresh_perms(bs_new
, tran
, errp
);
5588 tran_finalize(tran
, ret
);
5590 bdrv_refresh_limits(bs_top
, NULL
, NULL
);
5591 bdrv_graph_wrunlock();
5593 bdrv_drained_end(bs_top
);
5594 bdrv_drained_end(bs_new
);
5596 if (new_context
&& old_context
!= new_context
) {
5597 aio_context_release(new_context
);
5598 aio_context_acquire(old_context
);
5604 /* Not for empty child */
5605 int bdrv_replace_child_bs(BdrvChild
*child
, BlockDriverState
*new_bs
,
5609 Transaction
*tran
= tran_new();
5610 g_autoptr(GSList
) refresh_list
= NULL
;
5611 BlockDriverState
*old_bs
= child
->bs
;
5613 GLOBAL_STATE_CODE();
5616 bdrv_drained_begin(old_bs
);
5617 bdrv_drained_begin(new_bs
);
5618 bdrv_graph_wrlock(new_bs
);
5620 bdrv_replace_child_tran(child
, new_bs
, tran
);
5622 refresh_list
= g_slist_prepend(refresh_list
, old_bs
);
5623 refresh_list
= g_slist_prepend(refresh_list
, new_bs
);
5625 ret
= bdrv_list_refresh_perms(refresh_list
, NULL
, tran
, errp
);
5627 tran_finalize(tran
, ret
);
5629 bdrv_graph_wrunlock();
5630 bdrv_drained_end(old_bs
);
5631 bdrv_drained_end(new_bs
);
5637 static void bdrv_delete(BlockDriverState
*bs
)
5639 assert(bdrv_op_blocker_is_empty(bs
));
5640 assert(!bs
->refcnt
);
5641 GLOBAL_STATE_CODE();
5643 /* remove from list, if necessary */
5644 if (bs
->node_name
[0] != '\0') {
5645 QTAILQ_REMOVE(&graph_bdrv_states
, bs
, node_list
);
5647 QTAILQ_REMOVE(&all_bdrv_states
, bs
, bs_list
);
5651 qemu_mutex_destroy(&bs
->reqs_lock
);
5658 * Replace @bs by newly created block node.
5660 * @options is a QDict of options to pass to the block drivers, or NULL for an
5661 * empty set of options. The reference to the QDict belongs to the block layer
5662 * after the call (even on failure), so if the caller intends to reuse the
5663 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
5665 * The caller holds the AioContext lock for @bs. It must make sure that @bs
5666 * stays in the same AioContext, i.e. @options must not refer to nodes in a
5667 * different AioContext.
5669 BlockDriverState
*bdrv_insert_node(BlockDriverState
*bs
, QDict
*options
,
5670 int flags
, Error
**errp
)
5674 AioContext
*ctx
= bdrv_get_aio_context(bs
);
5675 BlockDriverState
*new_node_bs
= NULL
;
5676 const char *drvname
, *node_name
;
5679 drvname
= qdict_get_try_str(options
, "driver");
5681 error_setg(errp
, "driver is not specified");
5685 drv
= bdrv_find_format(drvname
);
5687 error_setg(errp
, "Unknown driver: '%s'", drvname
);
5691 node_name
= qdict_get_try_str(options
, "node-name");
5693 GLOBAL_STATE_CODE();
5695 aio_context_release(ctx
);
5696 aio_context_acquire(qemu_get_aio_context());
5697 new_node_bs
= bdrv_new_open_driver_opts(drv
, node_name
, options
, flags
,
5699 aio_context_release(qemu_get_aio_context());
5700 aio_context_acquire(ctx
);
5701 assert(bdrv_get_aio_context(bs
) == ctx
);
5703 options
= NULL
; /* bdrv_new_open_driver() eats options */
5705 error_prepend(errp
, "Could not create node: ");
5709 bdrv_drained_begin(bs
);
5710 ret
= bdrv_replace_node(bs
, new_node_bs
, errp
);
5711 bdrv_drained_end(bs
);
5714 error_prepend(errp
, "Could not replace node: ");
5721 qobject_unref(options
);
5722 bdrv_unref(new_node_bs
);
5727 * Run consistency checks on an image
5729 * Returns 0 if the check could be completed (it doesn't mean that the image is
5730 * free of errors) or -errno when an internal error occurred. The results of the
5731 * check are stored in res.
5733 int coroutine_fn
bdrv_co_check(BlockDriverState
*bs
,
5734 BdrvCheckResult
*res
, BdrvCheckMode fix
)
5737 assert_bdrv_graph_readable();
5738 if (bs
->drv
== NULL
) {
5741 if (bs
->drv
->bdrv_co_check
== NULL
) {
5745 memset(res
, 0, sizeof(*res
));
5746 return bs
->drv
->bdrv_co_check(bs
, res
, fix
);
5752 * -EINVAL - backing format specified, but no file
5753 * -ENOSPC - can't update the backing file because no space is left in the
5755 * -ENOTSUP - format driver doesn't support changing the backing file
5757 int bdrv_change_backing_file(BlockDriverState
*bs
, const char *backing_file
,
5758 const char *backing_fmt
, bool require
)
5760 BlockDriver
*drv
= bs
->drv
;
5763 GLOBAL_STATE_CODE();
5769 /* Backing file format doesn't make sense without a backing file */
5770 if (backing_fmt
&& !backing_file
) {
5774 if (require
&& backing_file
&& !backing_fmt
) {
5778 if (drv
->bdrv_change_backing_file
!= NULL
) {
5779 ret
= drv
->bdrv_change_backing_file(bs
, backing_file
, backing_fmt
);
5785 pstrcpy(bs
->backing_file
, sizeof(bs
->backing_file
), backing_file
?: "");
5786 pstrcpy(bs
->backing_format
, sizeof(bs
->backing_format
), backing_fmt
?: "");
5787 pstrcpy(bs
->auto_backing_file
, sizeof(bs
->auto_backing_file
),
5788 backing_file
?: "");
5794 * Finds the first non-filter node above bs in the chain between
5795 * active and bs. The returned node is either an immediate parent of
5796 * bs, or there are only filter nodes between the two.
5798 * Returns NULL if bs is not found in active's image chain,
5799 * or if active == bs.
5801 * Returns the bottommost base image if bs == NULL.
5803 BlockDriverState
*bdrv_find_overlay(BlockDriverState
*active
,
5804 BlockDriverState
*bs
)
5807 GLOBAL_STATE_CODE();
5809 bs
= bdrv_skip_filters(bs
);
5810 active
= bdrv_skip_filters(active
);
5813 BlockDriverState
*next
= bdrv_backing_chain_next(active
);
5823 /* Given a BDS, searches for the base layer. */
5824 BlockDriverState
*bdrv_find_base(BlockDriverState
*bs
)
5826 GLOBAL_STATE_CODE();
5828 return bdrv_find_overlay(bs
, NULL
);
5832 * Return true if at least one of the COW (backing) and filter links
5833 * between @bs and @base is frozen. @errp is set if that's the case.
5834 * @base must be reachable from @bs, or NULL.
5836 bool bdrv_is_backing_chain_frozen(BlockDriverState
*bs
, BlockDriverState
*base
,
5839 BlockDriverState
*i
;
5842 GLOBAL_STATE_CODE();
5844 for (i
= bs
; i
!= base
; i
= child_bs(child
)) {
5845 child
= bdrv_filter_or_cow_child(i
);
5847 if (child
&& child
->frozen
) {
5848 error_setg(errp
, "Cannot change '%s' link from '%s' to '%s'",
5849 child
->name
, i
->node_name
, child
->bs
->node_name
);
5858 * Freeze all COW (backing) and filter links between @bs and @base.
5859 * If any of the links is already frozen the operation is aborted and
5860 * none of the links are modified.
5861 * @base must be reachable from @bs, or NULL.
5862 * Returns 0 on success. On failure returns < 0 and sets @errp.
5864 int bdrv_freeze_backing_chain(BlockDriverState
*bs
, BlockDriverState
*base
,
5867 BlockDriverState
*i
;
5870 GLOBAL_STATE_CODE();
5872 if (bdrv_is_backing_chain_frozen(bs
, base
, errp
)) {
5876 for (i
= bs
; i
!= base
; i
= child_bs(child
)) {
5877 child
= bdrv_filter_or_cow_child(i
);
5878 if (child
&& child
->bs
->never_freeze
) {
5879 error_setg(errp
, "Cannot freeze '%s' link to '%s'",
5880 child
->name
, child
->bs
->node_name
);
5885 for (i
= bs
; i
!= base
; i
= child_bs(child
)) {
5886 child
= bdrv_filter_or_cow_child(i
);
5888 child
->frozen
= true;
5896 * Unfreeze all COW (backing) and filter links between @bs and @base.
5897 * The caller must ensure that all links are frozen before using this
5899 * @base must be reachable from @bs, or NULL.
5901 void bdrv_unfreeze_backing_chain(BlockDriverState
*bs
, BlockDriverState
*base
)
5903 BlockDriverState
*i
;
5906 GLOBAL_STATE_CODE();
5908 for (i
= bs
; i
!= base
; i
= child_bs(child
)) {
5909 child
= bdrv_filter_or_cow_child(i
);
5911 assert(child
->frozen
);
5912 child
->frozen
= false;
5918 * Drops images above 'base' up to and including 'top', and sets the image
5919 * above 'top' to have base as its backing file.
5921 * Requires that the overlay to 'top' is opened r/w, so that the backing file
5922 * information in 'bs' can be properly updated.
5924 * E.g., this will convert the following chain:
5925 * bottom <- base <- intermediate <- top <- active
5929 * bottom <- base <- active
5931 * It is allowed for bottom==base, in which case it converts:
5933 * base <- intermediate <- top <- active
5939 * If backing_file_str is non-NULL, it will be used when modifying top's
5940 * overlay image metadata.
5943 * if active == top, that is considered an error
5946 int bdrv_drop_intermediate(BlockDriverState
*top
, BlockDriverState
*base
,
5947 const char *backing_file_str
)
5949 BlockDriverState
*explicit_top
= top
;
5950 bool update_inherits_from
;
5952 Error
*local_err
= NULL
;
5954 g_autoptr(GSList
) updated_children
= NULL
;
5957 GLOBAL_STATE_CODE();
5960 bdrv_drained_begin(base
);
5961 bdrv_graph_rdlock_main_loop();
5963 if (!top
->drv
|| !base
->drv
) {
5967 /* Make sure that base is in the backing chain of top */
5968 if (!bdrv_chain_contains(top
, base
)) {
5972 /* If 'base' recursively inherits from 'top' then we should set
5973 * base->inherits_from to top->inherits_from after 'top' and all
5974 * other intermediate nodes have been dropped.
5975 * If 'top' is an implicit node (e.g. "commit_top") we should skip
5976 * it because no one inherits from it. We use explicit_top for that. */
5977 explicit_top
= bdrv_skip_implicit_filters(explicit_top
);
5978 update_inherits_from
= bdrv_inherits_from_recursive(base
, explicit_top
);
5980 /* success - we can delete the intermediate states, and link top->base */
5981 if (!backing_file_str
) {
5982 bdrv_refresh_filename(base
);
5983 backing_file_str
= base
->filename
;
5986 QLIST_FOREACH(c
, &top
->parents
, next_parent
) {
5987 updated_children
= g_slist_prepend(updated_children
, c
);
5991 * It seems correct to pass detach_subchain=true here, but it triggers
5992 * one more yet not fixed bug, when due to nested aio_poll loop we switch to
5993 * another drained section, which modify the graph (for example, removing
5994 * the child, which we keep in updated_children list). So, it's a TODO.
5996 * Note, bug triggered if pass detach_subchain=true here and run
5997 * test-bdrv-drain. test_drop_intermediate_poll() test-case will crash.
6000 bdrv_replace_node_common(top
, base
, false, false, &local_err
);
6002 error_report_err(local_err
);
6006 for (p
= updated_children
; p
; p
= p
->next
) {
6009 if (c
->klass
->update_filename
) {
6010 ret
= c
->klass
->update_filename(c
, base
, backing_file_str
,
6014 * TODO: Actually, we want to rollback all previous iterations
6015 * of this loop, and (which is almost impossible) previous
6016 * bdrv_replace_node()...
6018 * Note, that c->klass->update_filename may lead to permission
6019 * update, so it's a bad idea to call it inside permission
6020 * update transaction of bdrv_replace_node.
6022 error_report_err(local_err
);
6028 if (update_inherits_from
) {
6029 base
->inherits_from
= explicit_top
->inherits_from
;
6034 bdrv_graph_rdunlock_main_loop();
6035 bdrv_drained_end(base
);
6041 * Implementation of BlockDriver.bdrv_co_get_allocated_file_size() that
6042 * sums the size of all data-bearing children. (This excludes backing
6045 static int64_t coroutine_fn GRAPH_RDLOCK
6046 bdrv_sum_allocated_file_size(BlockDriverState
*bs
)
6049 int64_t child_size
, sum
= 0;
6051 QLIST_FOREACH(child
, &bs
->children
, next
) {
6052 if (child
->role
& (BDRV_CHILD_DATA
| BDRV_CHILD_METADATA
|
6053 BDRV_CHILD_FILTERED
))
6055 child_size
= bdrv_co_get_allocated_file_size(child
->bs
);
6056 if (child_size
< 0) {
6067 * Length of a allocated file in bytes. Sparse files are counted by actual
6068 * allocated space. Return < 0 if error or unknown.
6070 int64_t coroutine_fn
bdrv_co_get_allocated_file_size(BlockDriverState
*bs
)
6072 BlockDriver
*drv
= bs
->drv
;
6074 assert_bdrv_graph_readable();
6079 if (drv
->bdrv_co_get_allocated_file_size
) {
6080 return drv
->bdrv_co_get_allocated_file_size(bs
);
6083 if (drv
->bdrv_file_open
) {
6085 * Protocol drivers default to -ENOTSUP (most of their data is
6086 * not stored in any of their children (if they even have any),
6087 * so there is no generic way to figure it out).
6090 } else if (drv
->is_filter
) {
6091 /* Filter drivers default to the size of their filtered child */
6092 return bdrv_co_get_allocated_file_size(bdrv_filter_bs(bs
));
6094 /* Other drivers default to summing their children's sizes */
6095 return bdrv_sum_allocated_file_size(bs
);
6101 * @drv: Format driver
6102 * @opts: Creation options for new image
6103 * @in_bs: Existing image containing data for new image (may be NULL)
6104 * @errp: Error object
6105 * Returns: A #BlockMeasureInfo (free using qapi_free_BlockMeasureInfo())
6108 * Calculate file size required to create a new image.
6110 * If @in_bs is given then space for allocated clusters and zero clusters
6111 * from that image are included in the calculation. If @opts contains a
6112 * backing file that is shared by @in_bs then backing clusters may be omitted
6113 * from the calculation.
6115 * If @in_bs is NULL then the calculation includes no allocated clusters
6116 * unless a preallocation option is given in @opts.
6118 * Note that @in_bs may use a different BlockDriver from @drv.
6120 * If an error occurs the @errp pointer is set.
6122 BlockMeasureInfo
*bdrv_measure(BlockDriver
*drv
, QemuOpts
*opts
,
6123 BlockDriverState
*in_bs
, Error
**errp
)
6126 if (!drv
->bdrv_measure
) {
6127 error_setg(errp
, "Block driver '%s' does not support size measurement",
6132 return drv
->bdrv_measure(opts
, in_bs
, errp
);
6136 * Return number of sectors on success, -errno on error.
6138 int64_t coroutine_fn
bdrv_co_nb_sectors(BlockDriverState
*bs
)
6140 BlockDriver
*drv
= bs
->drv
;
6142 assert_bdrv_graph_readable();
6147 if (bs
->bl
.has_variable_length
) {
6148 int ret
= bdrv_co_refresh_total_sectors(bs
, bs
->total_sectors
);
6153 return bs
->total_sectors
;
6157 * This wrapper is written by hand because this function is in the hot I/O path,
6158 * via blk_get_geometry.
6160 int64_t coroutine_mixed_fn
bdrv_nb_sectors(BlockDriverState
*bs
)
6162 BlockDriver
*drv
= bs
->drv
;
6168 if (bs
->bl
.has_variable_length
) {
6169 int ret
= bdrv_refresh_total_sectors(bs
, bs
->total_sectors
);
6175 return bs
->total_sectors
;
6179 * Return length in bytes on success, -errno on error.
6180 * The length is always a multiple of BDRV_SECTOR_SIZE.
6182 int64_t coroutine_fn
bdrv_co_getlength(BlockDriverState
*bs
)
6186 assert_bdrv_graph_readable();
6188 ret
= bdrv_co_nb_sectors(bs
);
6192 if (ret
> INT64_MAX
/ BDRV_SECTOR_SIZE
) {
6195 return ret
* BDRV_SECTOR_SIZE
;
6198 bool bdrv_is_sg(BlockDriverState
*bs
)
6205 * Return whether the given node supports compressed writes.
6207 bool bdrv_supports_compressed_writes(BlockDriverState
*bs
)
6209 BlockDriverState
*filtered
;
6212 if (!bs
->drv
|| !block_driver_can_compress(bs
->drv
)) {
6216 filtered
= bdrv_filter_bs(bs
);
6219 * Filters can only forward compressed writes, so we have to
6222 return bdrv_supports_compressed_writes(filtered
);
6228 const char *bdrv_get_format_name(BlockDriverState
*bs
)
6231 return bs
->drv
? bs
->drv
->format_name
: NULL
;
6234 static int qsort_strcmp(const void *a
, const void *b
)
6236 return strcmp(*(char *const *)a
, *(char *const *)b
);
6239 void bdrv_iterate_format(void (*it
)(void *opaque
, const char *name
),
6240 void *opaque
, bool read_only
)
6245 const char **formats
= NULL
;
6247 GLOBAL_STATE_CODE();
6249 QLIST_FOREACH(drv
, &bdrv_drivers
, list
) {
6250 if (drv
->format_name
) {
6253 if (use_bdrv_whitelist
&& !bdrv_is_whitelisted(drv
, read_only
)) {
6258 while (formats
&& i
&& !found
) {
6259 found
= !strcmp(formats
[--i
], drv
->format_name
);
6263 formats
= g_renew(const char *, formats
, count
+ 1);
6264 formats
[count
++] = drv
->format_name
;
6269 for (i
= 0; i
< (int)ARRAY_SIZE(block_driver_modules
); i
++) {
6270 const char *format_name
= block_driver_modules
[i
].format_name
;
6276 if (use_bdrv_whitelist
&&
6277 !bdrv_format_is_whitelisted(format_name
, read_only
)) {
6281 while (formats
&& j
&& !found
) {
6282 found
= !strcmp(formats
[--j
], format_name
);
6286 formats
= g_renew(const char *, formats
, count
+ 1);
6287 formats
[count
++] = format_name
;
6292 qsort(formats
, count
, sizeof(formats
[0]), qsort_strcmp
);
6294 for (i
= 0; i
< count
; i
++) {
6295 it(opaque
, formats
[i
]);
6301 /* This function is to find a node in the bs graph */
6302 BlockDriverState
*bdrv_find_node(const char *node_name
)
6304 BlockDriverState
*bs
;
6307 GLOBAL_STATE_CODE();
6309 QTAILQ_FOREACH(bs
, &graph_bdrv_states
, node_list
) {
6310 if (!strcmp(node_name
, bs
->node_name
)) {
6317 /* Put this QMP function here so it can access the static graph_bdrv_states. */
6318 BlockDeviceInfoList
*bdrv_named_nodes_list(bool flat
,
6321 BlockDeviceInfoList
*list
;
6322 BlockDriverState
*bs
;
6324 GLOBAL_STATE_CODE();
6325 GRAPH_RDLOCK_GUARD_MAINLOOP();
6328 QTAILQ_FOREACH(bs
, &graph_bdrv_states
, node_list
) {
6329 BlockDeviceInfo
*info
= bdrv_block_device_info(NULL
, bs
, flat
, errp
);
6331 qapi_free_BlockDeviceInfoList(list
);
6334 QAPI_LIST_PREPEND(list
, info
);
6340 typedef struct XDbgBlockGraphConstructor
{
6341 XDbgBlockGraph
*graph
;
6342 GHashTable
*graph_nodes
;
6343 } XDbgBlockGraphConstructor
;
6345 static XDbgBlockGraphConstructor
*xdbg_graph_new(void)
6347 XDbgBlockGraphConstructor
*gr
= g_new(XDbgBlockGraphConstructor
, 1);
6349 gr
->graph
= g_new0(XDbgBlockGraph
, 1);
6350 gr
->graph_nodes
= g_hash_table_new(NULL
, NULL
);
6355 static XDbgBlockGraph
*xdbg_graph_finalize(XDbgBlockGraphConstructor
*gr
)
6357 XDbgBlockGraph
*graph
= gr
->graph
;
6359 g_hash_table_destroy(gr
->graph_nodes
);
6365 static uintptr_t xdbg_graph_node_num(XDbgBlockGraphConstructor
*gr
, void *node
)
6367 uintptr_t ret
= (uintptr_t)g_hash_table_lookup(gr
->graph_nodes
, node
);
6374 * Start counting from 1, not 0, because 0 interferes with not-found (NULL)
6375 * answer of g_hash_table_lookup.
6377 ret
= g_hash_table_size(gr
->graph_nodes
) + 1;
6378 g_hash_table_insert(gr
->graph_nodes
, node
, (void *)ret
);
6383 static void xdbg_graph_add_node(XDbgBlockGraphConstructor
*gr
, void *node
,
6384 XDbgBlockGraphNodeType type
, const char *name
)
6386 XDbgBlockGraphNode
*n
;
6388 n
= g_new0(XDbgBlockGraphNode
, 1);
6390 n
->id
= xdbg_graph_node_num(gr
, node
);
6392 n
->name
= g_strdup(name
);
6394 QAPI_LIST_PREPEND(gr
->graph
->nodes
, n
);
6397 static void xdbg_graph_add_edge(XDbgBlockGraphConstructor
*gr
, void *parent
,
6398 const BdrvChild
*child
)
6400 BlockPermission qapi_perm
;
6401 XDbgBlockGraphEdge
*edge
;
6402 GLOBAL_STATE_CODE();
6404 edge
= g_new0(XDbgBlockGraphEdge
, 1);
6406 edge
->parent
= xdbg_graph_node_num(gr
, parent
);
6407 edge
->child
= xdbg_graph_node_num(gr
, child
->bs
);
6408 edge
->name
= g_strdup(child
->name
);
6410 for (qapi_perm
= 0; qapi_perm
< BLOCK_PERMISSION__MAX
; qapi_perm
++) {
6411 uint64_t flag
= bdrv_qapi_perm_to_blk_perm(qapi_perm
);
6413 if (flag
& child
->perm
) {
6414 QAPI_LIST_PREPEND(edge
->perm
, qapi_perm
);
6416 if (flag
& child
->shared_perm
) {
6417 QAPI_LIST_PREPEND(edge
->shared_perm
, qapi_perm
);
6421 QAPI_LIST_PREPEND(gr
->graph
->edges
, edge
);
6425 XDbgBlockGraph
*bdrv_get_xdbg_block_graph(Error
**errp
)
6429 BlockDriverState
*bs
;
6431 XDbgBlockGraphConstructor
*gr
= xdbg_graph_new();
6433 GLOBAL_STATE_CODE();
6435 for (blk
= blk_all_next(NULL
); blk
; blk
= blk_all_next(blk
)) {
6436 char *allocated_name
= NULL
;
6437 const char *name
= blk_name(blk
);
6440 name
= allocated_name
= blk_get_attached_dev_id(blk
);
6442 xdbg_graph_add_node(gr
, blk
, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_BACKEND
,
6444 g_free(allocated_name
);
6445 if (blk_root(blk
)) {
6446 xdbg_graph_add_edge(gr
, blk
, blk_root(blk
));
6450 WITH_JOB_LOCK_GUARD() {
6451 for (job
= block_job_next_locked(NULL
); job
;
6452 job
= block_job_next_locked(job
)) {
6455 xdbg_graph_add_node(gr
, job
, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_JOB
,
6457 for (el
= job
->nodes
; el
; el
= el
->next
) {
6458 xdbg_graph_add_edge(gr
, job
, (BdrvChild
*)el
->data
);
6463 QTAILQ_FOREACH(bs
, &graph_bdrv_states
, node_list
) {
6464 xdbg_graph_add_node(gr
, bs
, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_DRIVER
,
6466 QLIST_FOREACH(child
, &bs
->children
, next
) {
6467 xdbg_graph_add_edge(gr
, bs
, child
);
6471 return xdbg_graph_finalize(gr
);
6474 BlockDriverState
*bdrv_lookup_bs(const char *device
,
6475 const char *node_name
,
6479 BlockDriverState
*bs
;
6481 GLOBAL_STATE_CODE();
6484 blk
= blk_by_name(device
);
6489 error_setg(errp
, "Device '%s' has no medium", device
);
6497 bs
= bdrv_find_node(node_name
);
6504 error_setg(errp
, "Cannot find device=\'%s\' nor node-name=\'%s\'",
6505 device
? device
: "",
6506 node_name
? node_name
: "");
6510 /* If 'base' is in the same chain as 'top', return true. Otherwise,
6511 * return false. If either argument is NULL, return false. */
6512 bool bdrv_chain_contains(BlockDriverState
*top
, BlockDriverState
*base
)
6515 GLOBAL_STATE_CODE();
6516 GRAPH_RDLOCK_GUARD_MAINLOOP();
6518 while (top
&& top
!= base
) {
6519 top
= bdrv_filter_or_cow_bs(top
);
6525 BlockDriverState
*bdrv_next_node(BlockDriverState
*bs
)
6527 GLOBAL_STATE_CODE();
6529 return QTAILQ_FIRST(&graph_bdrv_states
);
6531 return QTAILQ_NEXT(bs
, node_list
);
6534 BlockDriverState
*bdrv_next_all_states(BlockDriverState
*bs
)
6536 GLOBAL_STATE_CODE();
6538 return QTAILQ_FIRST(&all_bdrv_states
);
6540 return QTAILQ_NEXT(bs
, bs_list
);
6543 const char *bdrv_get_node_name(const BlockDriverState
*bs
)
6546 return bs
->node_name
;
6549 const char *bdrv_get_parent_name(const BlockDriverState
*bs
)
6555 /* If multiple parents have a name, just pick the first one. */
6556 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
6557 if (c
->klass
->get_name
) {
6558 name
= c
->klass
->get_name(c
);
6559 if (name
&& *name
) {
6568 /* TODO check what callers really want: bs->node_name or blk_name() */
6569 const char *bdrv_get_device_name(const BlockDriverState
*bs
)
6572 return bdrv_get_parent_name(bs
) ?: "";
6575 /* This can be used to identify nodes that might not have a device
6576 * name associated. Since node and device names live in the same
6577 * namespace, the result is unambiguous. The exception is if both are
6578 * absent, then this returns an empty (non-null) string. */
6579 const char *bdrv_get_device_or_node_name(const BlockDriverState
*bs
)
6582 return bdrv_get_parent_name(bs
) ?: bs
->node_name
;
6585 int bdrv_get_flags(BlockDriverState
*bs
)
6588 return bs
->open_flags
;
6591 int bdrv_has_zero_init_1(BlockDriverState
*bs
)
6593 GLOBAL_STATE_CODE();
6597 int coroutine_mixed_fn
bdrv_has_zero_init(BlockDriverState
*bs
)
6599 BlockDriverState
*filtered
;
6600 GLOBAL_STATE_CODE();
6606 /* If BS is a copy on write image, it is initialized to
6607 the contents of the base image, which may not be zeroes. */
6608 if (bdrv_cow_child(bs
)) {
6611 if (bs
->drv
->bdrv_has_zero_init
) {
6612 return bs
->drv
->bdrv_has_zero_init(bs
);
6615 filtered
= bdrv_filter_bs(bs
);
6617 return bdrv_has_zero_init(filtered
);
6624 bool bdrv_can_write_zeroes_with_unmap(BlockDriverState
*bs
)
6627 if (!(bs
->open_flags
& BDRV_O_UNMAP
)) {
6631 return bs
->supported_zero_flags
& BDRV_REQ_MAY_UNMAP
;
6634 void bdrv_get_backing_filename(BlockDriverState
*bs
,
6635 char *filename
, int filename_size
)
6638 pstrcpy(filename
, filename_size
, bs
->backing_file
);
6641 int coroutine_fn
bdrv_co_get_info(BlockDriverState
*bs
, BlockDriverInfo
*bdi
)
6644 BlockDriver
*drv
= bs
->drv
;
6646 assert_bdrv_graph_readable();
6648 /* if bs->drv == NULL, bs is closed, so there's nothing to do here */
6652 if (!drv
->bdrv_co_get_info
) {
6653 BlockDriverState
*filtered
= bdrv_filter_bs(bs
);
6655 return bdrv_co_get_info(filtered
, bdi
);
6659 memset(bdi
, 0, sizeof(*bdi
));
6660 ret
= drv
->bdrv_co_get_info(bs
, bdi
);
6661 if (bdi
->subcluster_size
== 0) {
6663 * If the driver left this unset, subclusters are not supported.
6664 * Then it is safe to treat each cluster as having only one subcluster.
6666 bdi
->subcluster_size
= bdi
->cluster_size
;
6672 if (bdi
->cluster_size
> BDRV_MAX_ALIGNMENT
) {
6679 ImageInfoSpecific
*bdrv_get_specific_info(BlockDriverState
*bs
,
6682 BlockDriver
*drv
= bs
->drv
;
6684 if (drv
&& drv
->bdrv_get_specific_info
) {
6685 return drv
->bdrv_get_specific_info(bs
, errp
);
6690 BlockStatsSpecific
*bdrv_get_specific_stats(BlockDriverState
*bs
)
6692 BlockDriver
*drv
= bs
->drv
;
6694 if (!drv
|| !drv
->bdrv_get_specific_stats
) {
6697 return drv
->bdrv_get_specific_stats(bs
);
6700 void coroutine_fn
bdrv_co_debug_event(BlockDriverState
*bs
, BlkdebugEvent event
)
6703 assert_bdrv_graph_readable();
6705 if (!bs
|| !bs
->drv
|| !bs
->drv
->bdrv_co_debug_event
) {
6709 bs
->drv
->bdrv_co_debug_event(bs
, event
);
6712 static BlockDriverState
* GRAPH_RDLOCK
6713 bdrv_find_debug_node(BlockDriverState
*bs
)
6715 GLOBAL_STATE_CODE();
6716 while (bs
&& bs
->drv
&& !bs
->drv
->bdrv_debug_breakpoint
) {
6717 bs
= bdrv_primary_bs(bs
);
6720 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_breakpoint
) {
6721 assert(bs
->drv
->bdrv_debug_remove_breakpoint
);
6728 int bdrv_debug_breakpoint(BlockDriverState
*bs
, const char *event
,
6731 GLOBAL_STATE_CODE();
6732 GRAPH_RDLOCK_GUARD_MAINLOOP();
6734 bs
= bdrv_find_debug_node(bs
);
6736 return bs
->drv
->bdrv_debug_breakpoint(bs
, event
, tag
);
6742 int bdrv_debug_remove_breakpoint(BlockDriverState
*bs
, const char *tag
)
6744 GLOBAL_STATE_CODE();
6745 GRAPH_RDLOCK_GUARD_MAINLOOP();
6747 bs
= bdrv_find_debug_node(bs
);
6749 return bs
->drv
->bdrv_debug_remove_breakpoint(bs
, tag
);
6755 int bdrv_debug_resume(BlockDriverState
*bs
, const char *tag
)
6757 GLOBAL_STATE_CODE();
6758 GRAPH_RDLOCK_GUARD_MAINLOOP();
6760 while (bs
&& (!bs
->drv
|| !bs
->drv
->bdrv_debug_resume
)) {
6761 bs
= bdrv_primary_bs(bs
);
6764 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_resume
) {
6765 return bs
->drv
->bdrv_debug_resume(bs
, tag
);
6771 bool bdrv_debug_is_suspended(BlockDriverState
*bs
, const char *tag
)
6773 GLOBAL_STATE_CODE();
6774 GRAPH_RDLOCK_GUARD_MAINLOOP();
6776 while (bs
&& bs
->drv
&& !bs
->drv
->bdrv_debug_is_suspended
) {
6777 bs
= bdrv_primary_bs(bs
);
6780 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_is_suspended
) {
6781 return bs
->drv
->bdrv_debug_is_suspended(bs
, tag
);
6787 /* backing_file can either be relative, or absolute, or a protocol. If it is
6788 * relative, it must be relative to the chain. So, passing in bs->filename
6789 * from a BDS as backing_file should not be done, as that may be relative to
6790 * the CWD rather than the chain. */
6791 BlockDriverState
*bdrv_find_backing_image(BlockDriverState
*bs
,
6792 const char *backing_file
)
6794 char *filename_full
= NULL
;
6795 char *backing_file_full
= NULL
;
6796 char *filename_tmp
= NULL
;
6797 int is_protocol
= 0;
6798 bool filenames_refreshed
= false;
6799 BlockDriverState
*curr_bs
= NULL
;
6800 BlockDriverState
*retval
= NULL
;
6801 BlockDriverState
*bs_below
;
6803 GLOBAL_STATE_CODE();
6804 GRAPH_RDLOCK_GUARD_MAINLOOP();
6806 if (!bs
|| !bs
->drv
|| !backing_file
) {
6810 filename_full
= g_malloc(PATH_MAX
);
6811 backing_file_full
= g_malloc(PATH_MAX
);
6813 is_protocol
= path_has_protocol(backing_file
);
6816 * Being largely a legacy function, skip any filters here
6817 * (because filters do not have normal filenames, so they cannot
6818 * match anyway; and allowing json:{} filenames is a bit out of
6821 for (curr_bs
= bdrv_skip_filters(bs
);
6822 bdrv_cow_child(curr_bs
) != NULL
;
6825 bs_below
= bdrv_backing_chain_next(curr_bs
);
6827 if (bdrv_backing_overridden(curr_bs
)) {
6829 * If the backing file was overridden, we can only compare
6830 * directly against the backing node's filename.
6833 if (!filenames_refreshed
) {
6835 * This will automatically refresh all of the
6836 * filenames in the rest of the backing chain, so we
6837 * only need to do this once.
6839 bdrv_refresh_filename(bs_below
);
6840 filenames_refreshed
= true;
6843 if (strcmp(backing_file
, bs_below
->filename
) == 0) {
6847 } else if (is_protocol
|| path_has_protocol(curr_bs
->backing_file
)) {
6849 * If either of the filename paths is actually a protocol, then
6850 * compare unmodified paths; otherwise make paths relative.
6852 char *backing_file_full_ret
;
6854 if (strcmp(backing_file
, curr_bs
->backing_file
) == 0) {
6858 /* Also check against the full backing filename for the image */
6859 backing_file_full_ret
= bdrv_get_full_backing_filename(curr_bs
,
6861 if (backing_file_full_ret
) {
6862 bool equal
= strcmp(backing_file
, backing_file_full_ret
) == 0;
6863 g_free(backing_file_full_ret
);
6870 /* If not an absolute filename path, make it relative to the current
6871 * image's filename path */
6872 filename_tmp
= bdrv_make_absolute_filename(curr_bs
, backing_file
,
6874 /* We are going to compare canonicalized absolute pathnames */
6875 if (!filename_tmp
|| !realpath(filename_tmp
, filename_full
)) {
6876 g_free(filename_tmp
);
6879 g_free(filename_tmp
);
6881 /* We need to make sure the backing filename we are comparing against
6882 * is relative to the current image filename (or absolute) */
6883 filename_tmp
= bdrv_get_full_backing_filename(curr_bs
, NULL
);
6884 if (!filename_tmp
|| !realpath(filename_tmp
, backing_file_full
)) {
6885 g_free(filename_tmp
);
6888 g_free(filename_tmp
);
6890 if (strcmp(backing_file_full
, filename_full
) == 0) {
6897 g_free(filename_full
);
6898 g_free(backing_file_full
);
6902 void bdrv_init(void)
6904 #ifdef CONFIG_BDRV_WHITELIST_TOOLS
6905 use_bdrv_whitelist
= 1;
6907 module_call_init(MODULE_INIT_BLOCK
);
6910 void bdrv_init_with_whitelist(void)
6912 use_bdrv_whitelist
= 1;
6916 int bdrv_activate(BlockDriverState
*bs
, Error
**errp
)
6918 BdrvChild
*child
, *parent
;
6919 Error
*local_err
= NULL
;
6921 BdrvDirtyBitmap
*bm
;
6923 GLOBAL_STATE_CODE();
6924 GRAPH_RDLOCK_GUARD_MAINLOOP();
6930 QLIST_FOREACH(child
, &bs
->children
, next
) {
6931 bdrv_activate(child
->bs
, &local_err
);
6933 error_propagate(errp
, local_err
);
6939 * Update permissions, they may differ for inactive nodes.
6941 * Note that the required permissions of inactive images are always a
6942 * subset of the permissions required after activating the image. This
6943 * allows us to just get the permissions upfront without restricting
6944 * bdrv_co_invalidate_cache().
6946 * It also means that in error cases, we don't have to try and revert to
6947 * the old permissions (which is an operation that could fail, too). We can
6948 * just keep the extended permissions for the next time that an activation
6949 * of the image is tried.
6951 if (bs
->open_flags
& BDRV_O_INACTIVE
) {
6952 bs
->open_flags
&= ~BDRV_O_INACTIVE
;
6953 ret
= bdrv_refresh_perms(bs
, NULL
, errp
);
6955 bs
->open_flags
|= BDRV_O_INACTIVE
;
6959 ret
= bdrv_invalidate_cache(bs
, errp
);
6961 bs
->open_flags
|= BDRV_O_INACTIVE
;
6965 FOR_EACH_DIRTY_BITMAP(bs
, bm
) {
6966 bdrv_dirty_bitmap_skip_store(bm
, false);
6969 ret
= bdrv_refresh_total_sectors(bs
, bs
->total_sectors
);
6971 bs
->open_flags
|= BDRV_O_INACTIVE
;
6972 error_setg_errno(errp
, -ret
, "Could not refresh total sector count");
6977 QLIST_FOREACH(parent
, &bs
->parents
, next_parent
) {
6978 if (parent
->klass
->activate
) {
6979 parent
->klass
->activate(parent
, &local_err
);
6981 bs
->open_flags
|= BDRV_O_INACTIVE
;
6982 error_propagate(errp
, local_err
);
6991 int coroutine_fn
bdrv_co_invalidate_cache(BlockDriverState
*bs
, Error
**errp
)
6993 Error
*local_err
= NULL
;
6996 assert(!(bs
->open_flags
& BDRV_O_INACTIVE
));
6997 assert_bdrv_graph_readable();
6999 if (bs
->drv
->bdrv_co_invalidate_cache
) {
7000 bs
->drv
->bdrv_co_invalidate_cache(bs
, &local_err
);
7002 error_propagate(errp
, local_err
);
7010 void bdrv_activate_all(Error
**errp
)
7012 BlockDriverState
*bs
;
7013 BdrvNextIterator it
;
7015 GLOBAL_STATE_CODE();
7016 GRAPH_RDLOCK_GUARD_MAINLOOP();
7018 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
7019 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
7022 aio_context_acquire(aio_context
);
7023 ret
= bdrv_activate(bs
, errp
);
7024 aio_context_release(aio_context
);
7026 bdrv_next_cleanup(&it
);
7032 static bool GRAPH_RDLOCK
7033 bdrv_has_bds_parent(BlockDriverState
*bs
, bool only_active
)
7036 GLOBAL_STATE_CODE();
7038 QLIST_FOREACH(parent
, &bs
->parents
, next_parent
) {
7039 if (parent
->klass
->parent_is_bds
) {
7040 BlockDriverState
*parent_bs
= parent
->opaque
;
7041 if (!only_active
|| !(parent_bs
->open_flags
& BDRV_O_INACTIVE
)) {
7050 static int GRAPH_RDLOCK
bdrv_inactivate_recurse(BlockDriverState
*bs
)
7052 BdrvChild
*child
, *parent
;
7054 uint64_t cumulative_perms
, cumulative_shared_perms
;
7056 GLOBAL_STATE_CODE();
7062 /* Make sure that we don't inactivate a child before its parent.
7063 * It will be covered by recursion from the yet active parent. */
7064 if (bdrv_has_bds_parent(bs
, true)) {
7068 assert(!(bs
->open_flags
& BDRV_O_INACTIVE
));
7070 /* Inactivate this node */
7071 if (bs
->drv
->bdrv_inactivate
) {
7072 ret
= bs
->drv
->bdrv_inactivate(bs
);
7078 QLIST_FOREACH(parent
, &bs
->parents
, next_parent
) {
7079 if (parent
->klass
->inactivate
) {
7080 ret
= parent
->klass
->inactivate(parent
);
7087 bdrv_get_cumulative_perm(bs
, &cumulative_perms
,
7088 &cumulative_shared_perms
);
7089 if (cumulative_perms
& (BLK_PERM_WRITE
| BLK_PERM_WRITE_UNCHANGED
)) {
7090 /* Our inactive parents still need write access. Inactivation failed. */
7094 bs
->open_flags
|= BDRV_O_INACTIVE
;
7097 * Update permissions, they may differ for inactive nodes.
7098 * We only tried to loosen restrictions, so errors are not fatal, ignore
7101 bdrv_refresh_perms(bs
, NULL
, NULL
);
7103 /* Recursively inactivate children */
7104 QLIST_FOREACH(child
, &bs
->children
, next
) {
7105 ret
= bdrv_inactivate_recurse(child
->bs
);
7114 int bdrv_inactivate_all(void)
7116 BlockDriverState
*bs
= NULL
;
7117 BdrvNextIterator it
;
7119 GSList
*aio_ctxs
= NULL
, *ctx
;
7121 GLOBAL_STATE_CODE();
7122 GRAPH_RDLOCK_GUARD_MAINLOOP();
7124 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
7125 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
7127 if (!g_slist_find(aio_ctxs
, aio_context
)) {
7128 aio_ctxs
= g_slist_prepend(aio_ctxs
, aio_context
);
7129 aio_context_acquire(aio_context
);
7133 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
7134 /* Nodes with BDS parents are covered by recursion from the last
7135 * parent that gets inactivated. Don't inactivate them a second
7136 * time if that has already happened. */
7137 if (bdrv_has_bds_parent(bs
, false)) {
7140 ret
= bdrv_inactivate_recurse(bs
);
7142 bdrv_next_cleanup(&it
);
7148 for (ctx
= aio_ctxs
; ctx
!= NULL
; ctx
= ctx
->next
) {
7149 AioContext
*aio_context
= ctx
->data
;
7150 aio_context_release(aio_context
);
7152 g_slist_free(aio_ctxs
);
7157 /**************************************************************/
7158 /* removable device support */
7161 * Return TRUE if the media is present
7163 bool coroutine_fn
bdrv_co_is_inserted(BlockDriverState
*bs
)
7165 BlockDriver
*drv
= bs
->drv
;
7168 assert_bdrv_graph_readable();
7173 if (drv
->bdrv_co_is_inserted
) {
7174 return drv
->bdrv_co_is_inserted(bs
);
7176 QLIST_FOREACH(child
, &bs
->children
, next
) {
7177 if (!bdrv_co_is_inserted(child
->bs
)) {
7185 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
7187 void coroutine_fn
bdrv_co_eject(BlockDriverState
*bs
, bool eject_flag
)
7189 BlockDriver
*drv
= bs
->drv
;
7191 assert_bdrv_graph_readable();
7193 if (drv
&& drv
->bdrv_co_eject
) {
7194 drv
->bdrv_co_eject(bs
, eject_flag
);
7199 * Lock or unlock the media (if it is locked, the user won't be able
7200 * to eject it manually).
7202 void coroutine_fn
bdrv_co_lock_medium(BlockDriverState
*bs
, bool locked
)
7204 BlockDriver
*drv
= bs
->drv
;
7206 assert_bdrv_graph_readable();
7207 trace_bdrv_lock_medium(bs
, locked
);
7209 if (drv
&& drv
->bdrv_co_lock_medium
) {
7210 drv
->bdrv_co_lock_medium(bs
, locked
);
7214 /* Get a reference to bs */
7215 void bdrv_ref(BlockDriverState
*bs
)
7217 GLOBAL_STATE_CODE();
7221 /* Release a previously grabbed reference to bs.
7222 * If after releasing, reference count is zero, the BlockDriverState is
7224 void bdrv_unref(BlockDriverState
*bs
)
7226 GLOBAL_STATE_CODE();
7230 assert(bs
->refcnt
> 0);
7231 if (--bs
->refcnt
== 0) {
7237 * Release a BlockDriverState reference while holding the graph write lock.
7239 * Calling bdrv_unref() directly is forbidden while holding the graph lock
7240 * because bdrv_close() both involves polling and taking the graph lock
7241 * internally. bdrv_schedule_unref() instead delays decreasing the refcount and
7242 * possibly closing @bs until the graph lock is released.
7244 void bdrv_schedule_unref(BlockDriverState
*bs
)
7249 aio_bh_schedule_oneshot(qemu_get_aio_context(),
7250 (QEMUBHFunc
*) bdrv_unref
, bs
);
7253 struct BdrvOpBlocker
{
7255 QLIST_ENTRY(BdrvOpBlocker
) list
;
7258 bool bdrv_op_is_blocked(BlockDriverState
*bs
, BlockOpType op
, Error
**errp
)
7260 BdrvOpBlocker
*blocker
;
7261 GLOBAL_STATE_CODE();
7263 assert((int) op
>= 0 && op
< BLOCK_OP_TYPE_MAX
);
7264 if (!QLIST_EMPTY(&bs
->op_blockers
[op
])) {
7265 blocker
= QLIST_FIRST(&bs
->op_blockers
[op
]);
7266 error_propagate_prepend(errp
, error_copy(blocker
->reason
),
7267 "Node '%s' is busy: ",
7268 bdrv_get_device_or_node_name(bs
));
7274 void bdrv_op_block(BlockDriverState
*bs
, BlockOpType op
, Error
*reason
)
7276 BdrvOpBlocker
*blocker
;
7277 GLOBAL_STATE_CODE();
7278 assert((int) op
>= 0 && op
< BLOCK_OP_TYPE_MAX
);
7280 blocker
= g_new0(BdrvOpBlocker
, 1);
7281 blocker
->reason
= reason
;
7282 QLIST_INSERT_HEAD(&bs
->op_blockers
[op
], blocker
, list
);
7285 void bdrv_op_unblock(BlockDriverState
*bs
, BlockOpType op
, Error
*reason
)
7287 BdrvOpBlocker
*blocker
, *next
;
7288 GLOBAL_STATE_CODE();
7289 assert((int) op
>= 0 && op
< BLOCK_OP_TYPE_MAX
);
7290 QLIST_FOREACH_SAFE(blocker
, &bs
->op_blockers
[op
], list
, next
) {
7291 if (blocker
->reason
== reason
) {
7292 QLIST_REMOVE(blocker
, list
);
7298 void bdrv_op_block_all(BlockDriverState
*bs
, Error
*reason
)
7301 GLOBAL_STATE_CODE();
7302 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
7303 bdrv_op_block(bs
, i
, reason
);
7307 void bdrv_op_unblock_all(BlockDriverState
*bs
, Error
*reason
)
7310 GLOBAL_STATE_CODE();
7311 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
7312 bdrv_op_unblock(bs
, i
, reason
);
7316 bool bdrv_op_blocker_is_empty(BlockDriverState
*bs
)
7319 GLOBAL_STATE_CODE();
7320 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
7321 if (!QLIST_EMPTY(&bs
->op_blockers
[i
])) {
7329 * Must not be called while holding the lock of an AioContext other than the
7332 void bdrv_img_create(const char *filename
, const char *fmt
,
7333 const char *base_filename
, const char *base_fmt
,
7334 char *options
, uint64_t img_size
, int flags
, bool quiet
,
7337 QemuOptsList
*create_opts
= NULL
;
7338 QemuOpts
*opts
= NULL
;
7339 const char *backing_fmt
, *backing_file
;
7341 BlockDriver
*drv
, *proto_drv
;
7342 Error
*local_err
= NULL
;
7345 GLOBAL_STATE_CODE();
7347 /* Find driver and parse its options */
7348 drv
= bdrv_find_format(fmt
);
7350 error_setg(errp
, "Unknown file format '%s'", fmt
);
7354 proto_drv
= bdrv_find_protocol(filename
, true, errp
);
7359 if (!drv
->create_opts
) {
7360 error_setg(errp
, "Format driver '%s' does not support image creation",
7365 if (!proto_drv
->create_opts
) {
7366 error_setg(errp
, "Protocol driver '%s' does not support image creation",
7367 proto_drv
->format_name
);
7371 aio_context_acquire(qemu_get_aio_context());
7373 /* Create parameter list */
7374 create_opts
= qemu_opts_append(create_opts
, drv
->create_opts
);
7375 create_opts
= qemu_opts_append(create_opts
, proto_drv
->create_opts
);
7377 opts
= qemu_opts_create(create_opts
, NULL
, 0, &error_abort
);
7379 /* Parse -o options */
7381 if (!qemu_opts_do_parse(opts
, options
, NULL
, errp
)) {
7386 if (!qemu_opt_get(opts
, BLOCK_OPT_SIZE
)) {
7387 qemu_opt_set_number(opts
, BLOCK_OPT_SIZE
, img_size
, &error_abort
);
7388 } else if (img_size
!= UINT64_C(-1)) {
7389 error_setg(errp
, "The image size must be specified only once");
7393 if (base_filename
) {
7394 if (!qemu_opt_set(opts
, BLOCK_OPT_BACKING_FILE
, base_filename
,
7396 error_setg(errp
, "Backing file not supported for file format '%s'",
7403 if (!qemu_opt_set(opts
, BLOCK_OPT_BACKING_FMT
, base_fmt
, NULL
)) {
7404 error_setg(errp
, "Backing file format not supported for file "
7405 "format '%s'", fmt
);
7410 backing_file
= qemu_opt_get(opts
, BLOCK_OPT_BACKING_FILE
);
7412 if (!strcmp(filename
, backing_file
)) {
7413 error_setg(errp
, "Error: Trying to create an image with the "
7414 "same filename as the backing file");
7417 if (backing_file
[0] == '\0') {
7418 error_setg(errp
, "Expected backing file name, got empty string");
7423 backing_fmt
= qemu_opt_get(opts
, BLOCK_OPT_BACKING_FMT
);
7425 /* The size for the image must always be specified, unless we have a backing
7426 * file and we have not been forbidden from opening it. */
7427 size
= qemu_opt_get_size(opts
, BLOCK_OPT_SIZE
, img_size
);
7428 if (backing_file
&& !(flags
& BDRV_O_NO_BACKING
)) {
7429 BlockDriverState
*bs
;
7432 QDict
*backing_options
= NULL
;
7435 bdrv_get_full_backing_filename_from_filename(filename
, backing_file
,
7440 assert(full_backing
);
7443 * No need to do I/O here, which allows us to open encrypted
7444 * backing images without needing the secret
7447 back_flags
&= ~(BDRV_O_RDWR
| BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
);
7448 back_flags
|= BDRV_O_NO_IO
;
7450 backing_options
= qdict_new();
7452 qdict_put_str(backing_options
, "driver", backing_fmt
);
7454 qdict_put_bool(backing_options
, BDRV_OPT_FORCE_SHARE
, true);
7456 bs
= bdrv_open(full_backing
, NULL
, backing_options
, back_flags
,
7458 g_free(full_backing
);
7460 error_append_hint(&local_err
, "Could not open backing image.\n");
7464 error_setg(&local_err
,
7465 "Backing file specified without backing format");
7466 error_append_hint(&local_err
, "Detected format of %s.\n",
7467 bs
->drv
->format_name
);
7471 /* Opened BS, have no size */
7472 size
= bdrv_getlength(bs
);
7474 error_setg_errno(errp
, -size
, "Could not get size of '%s'",
7479 qemu_opt_set_number(opts
, BLOCK_OPT_SIZE
, size
, &error_abort
);
7483 /* (backing_file && !(flags & BDRV_O_NO_BACKING)) */
7484 } else if (backing_file
&& !backing_fmt
) {
7485 error_setg(&local_err
,
7486 "Backing file specified without backing format");
7491 error_setg(errp
, "Image creation needs a size parameter");
7496 printf("Formatting '%s', fmt=%s ", filename
, fmt
);
7497 qemu_opts_print(opts
, " ");
7502 ret
= bdrv_create(drv
, filename
, opts
, &local_err
);
7504 if (ret
== -EFBIG
) {
7505 /* This is generally a better message than whatever the driver would
7506 * deliver (especially because of the cluster_size_hint), since that
7507 * is most probably not much different from "image too large". */
7508 const char *cluster_size_hint
= "";
7509 if (qemu_opt_get_size(opts
, BLOCK_OPT_CLUSTER_SIZE
, 0)) {
7510 cluster_size_hint
= " (try using a larger cluster size)";
7512 error_setg(errp
, "The image size is too large for file format '%s'"
7513 "%s", fmt
, cluster_size_hint
);
7514 error_free(local_err
);
7519 qemu_opts_del(opts
);
7520 qemu_opts_free(create_opts
);
7521 error_propagate(errp
, local_err
);
7522 aio_context_release(qemu_get_aio_context());
7525 AioContext
*bdrv_get_aio_context(BlockDriverState
*bs
)
7528 return bs
? bs
->aio_context
: qemu_get_aio_context();
7531 AioContext
*coroutine_fn
bdrv_co_enter(BlockDriverState
*bs
)
7533 Coroutine
*self
= qemu_coroutine_self();
7534 AioContext
*old_ctx
= qemu_coroutine_get_aio_context(self
);
7535 AioContext
*new_ctx
;
7539 * Increase bs->in_flight to ensure that this operation is completed before
7540 * moving the node to a different AioContext. Read new_ctx only afterwards.
7542 bdrv_inc_in_flight(bs
);
7544 new_ctx
= bdrv_get_aio_context(bs
);
7545 aio_co_reschedule_self(new_ctx
);
7549 void coroutine_fn
bdrv_co_leave(BlockDriverState
*bs
, AioContext
*old_ctx
)
7552 aio_co_reschedule_self(old_ctx
);
7553 bdrv_dec_in_flight(bs
);
7556 void coroutine_fn
bdrv_co_lock(BlockDriverState
*bs
)
7558 AioContext
*ctx
= bdrv_get_aio_context(bs
);
7560 /* In the main thread, bs->aio_context won't change concurrently */
7561 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
7564 * We're in coroutine context, so we already hold the lock of the main
7565 * loop AioContext. Don't lock it twice to avoid deadlocks.
7567 assert(qemu_in_coroutine());
7568 if (ctx
!= qemu_get_aio_context()) {
7569 aio_context_acquire(ctx
);
7573 void coroutine_fn
bdrv_co_unlock(BlockDriverState
*bs
)
7575 AioContext
*ctx
= bdrv_get_aio_context(bs
);
7577 assert(qemu_in_coroutine());
7578 if (ctx
!= qemu_get_aio_context()) {
7579 aio_context_release(ctx
);
7583 static void bdrv_do_remove_aio_context_notifier(BdrvAioNotifier
*ban
)
7585 GLOBAL_STATE_CODE();
7586 QLIST_REMOVE(ban
, list
);
7590 static void bdrv_detach_aio_context(BlockDriverState
*bs
)
7592 BdrvAioNotifier
*baf
, *baf_tmp
;
7594 assert(!bs
->walking_aio_notifiers
);
7595 GLOBAL_STATE_CODE();
7596 bs
->walking_aio_notifiers
= true;
7597 QLIST_FOREACH_SAFE(baf
, &bs
->aio_notifiers
, list
, baf_tmp
) {
7599 bdrv_do_remove_aio_context_notifier(baf
);
7601 baf
->detach_aio_context(baf
->opaque
);
7604 /* Never mind iterating again to check for ->deleted. bdrv_close() will
7605 * remove remaining aio notifiers if we aren't called again.
7607 bs
->walking_aio_notifiers
= false;
7609 if (bs
->drv
&& bs
->drv
->bdrv_detach_aio_context
) {
7610 bs
->drv
->bdrv_detach_aio_context(bs
);
7613 bs
->aio_context
= NULL
;
7616 static void bdrv_attach_aio_context(BlockDriverState
*bs
,
7617 AioContext
*new_context
)
7619 BdrvAioNotifier
*ban
, *ban_tmp
;
7620 GLOBAL_STATE_CODE();
7622 bs
->aio_context
= new_context
;
7624 if (bs
->drv
&& bs
->drv
->bdrv_attach_aio_context
) {
7625 bs
->drv
->bdrv_attach_aio_context(bs
, new_context
);
7628 assert(!bs
->walking_aio_notifiers
);
7629 bs
->walking_aio_notifiers
= true;
7630 QLIST_FOREACH_SAFE(ban
, &bs
->aio_notifiers
, list
, ban_tmp
) {
7632 bdrv_do_remove_aio_context_notifier(ban
);
7634 ban
->attached_aio_context(new_context
, ban
->opaque
);
7637 bs
->walking_aio_notifiers
= false;
7640 typedef struct BdrvStateSetAioContext
{
7641 AioContext
*new_ctx
;
7642 BlockDriverState
*bs
;
7643 } BdrvStateSetAioContext
;
7645 static bool bdrv_parent_change_aio_context(BdrvChild
*c
, AioContext
*ctx
,
7646 GHashTable
*visited
,
7650 GLOBAL_STATE_CODE();
7651 if (g_hash_table_contains(visited
, c
)) {
7654 g_hash_table_add(visited
, c
);
7657 * A BdrvChildClass that doesn't handle AioContext changes cannot
7658 * tolerate any AioContext changes
7660 if (!c
->klass
->change_aio_ctx
) {
7661 char *user
= bdrv_child_user_desc(c
);
7662 error_setg(errp
, "Changing iothreads is not supported by %s", user
);
7666 if (!c
->klass
->change_aio_ctx(c
, ctx
, visited
, tran
, errp
)) {
7667 assert(!errp
|| *errp
);
7673 bool bdrv_child_change_aio_context(BdrvChild
*c
, AioContext
*ctx
,
7674 GHashTable
*visited
, Transaction
*tran
,
7677 GLOBAL_STATE_CODE();
7678 if (g_hash_table_contains(visited
, c
)) {
7681 g_hash_table_add(visited
, c
);
7682 return bdrv_change_aio_context(c
->bs
, ctx
, visited
, tran
, errp
);
7685 static void bdrv_set_aio_context_clean(void *opaque
)
7687 BdrvStateSetAioContext
*state
= (BdrvStateSetAioContext
*) opaque
;
7688 BlockDriverState
*bs
= (BlockDriverState
*) state
->bs
;
7690 /* Paired with bdrv_drained_begin in bdrv_change_aio_context() */
7691 bdrv_drained_end(bs
);
7696 static void bdrv_set_aio_context_commit(void *opaque
)
7698 BdrvStateSetAioContext
*state
= (BdrvStateSetAioContext
*) opaque
;
7699 BlockDriverState
*bs
= (BlockDriverState
*) state
->bs
;
7700 AioContext
*new_context
= state
->new_ctx
;
7701 AioContext
*old_context
= bdrv_get_aio_context(bs
);
7704 * Take the old AioContex when detaching it from bs.
7705 * At this point, new_context lock is already acquired, and we are now
7706 * also taking old_context. This is safe as long as bdrv_detach_aio_context
7707 * does not call AIO_POLL_WHILE().
7709 if (old_context
!= qemu_get_aio_context()) {
7710 aio_context_acquire(old_context
);
7712 bdrv_detach_aio_context(bs
);
7713 if (old_context
!= qemu_get_aio_context()) {
7714 aio_context_release(old_context
);
7716 bdrv_attach_aio_context(bs
, new_context
);
7719 static TransactionActionDrv set_aio_context
= {
7720 .commit
= bdrv_set_aio_context_commit
,
7721 .clean
= bdrv_set_aio_context_clean
,
7725 * Changes the AioContext used for fd handlers, timers, and BHs by this
7726 * BlockDriverState and all its children and parents.
7728 * Must be called from the main AioContext.
7730 * The caller must own the AioContext lock for the old AioContext of bs, but it
7731 * must not own the AioContext lock for new_context (unless new_context is the
7732 * same as the current context of bs).
7734 * @visited will accumulate all visited BdrvChild objects. The caller is
7735 * responsible for freeing the list afterwards.
7737 static bool bdrv_change_aio_context(BlockDriverState
*bs
, AioContext
*ctx
,
7738 GHashTable
*visited
, Transaction
*tran
,
7742 BdrvStateSetAioContext
*state
;
7744 GLOBAL_STATE_CODE();
7746 if (bdrv_get_aio_context(bs
) == ctx
) {
7750 bdrv_graph_rdlock_main_loop();
7751 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
7752 if (!bdrv_parent_change_aio_context(c
, ctx
, visited
, tran
, errp
)) {
7753 bdrv_graph_rdunlock_main_loop();
7758 QLIST_FOREACH(c
, &bs
->children
, next
) {
7759 if (!bdrv_child_change_aio_context(c
, ctx
, visited
, tran
, errp
)) {
7760 bdrv_graph_rdunlock_main_loop();
7764 bdrv_graph_rdunlock_main_loop();
7766 state
= g_new(BdrvStateSetAioContext
, 1);
7767 *state
= (BdrvStateSetAioContext
) {
7772 /* Paired with bdrv_drained_end in bdrv_set_aio_context_clean() */
7773 bdrv_drained_begin(bs
);
7775 tran_add(tran
, &set_aio_context
, state
);
7781 * Change bs's and recursively all of its parents' and children's AioContext
7782 * to the given new context, returning an error if that isn't possible.
7784 * If ignore_child is not NULL, that child (and its subgraph) will not
7787 * This function still requires the caller to take the bs current
7788 * AioContext lock, otherwise draining will fail since AIO_WAIT_WHILE
7789 * assumes the lock is always held if bs is in another AioContext.
7790 * For the same reason, it temporarily also holds the new AioContext, since
7791 * bdrv_drained_end calls BDRV_POLL_WHILE that assumes the lock is taken too.
7792 * Therefore the new AioContext lock must not be taken by the caller.
7794 int bdrv_try_change_aio_context(BlockDriverState
*bs
, AioContext
*ctx
,
7795 BdrvChild
*ignore_child
, Error
**errp
)
7798 GHashTable
*visited
;
7800 AioContext
*old_context
= bdrv_get_aio_context(bs
);
7801 GLOBAL_STATE_CODE();
7804 * Recursion phase: go through all nodes of the graph.
7805 * Take care of checking that all nodes support changing AioContext
7806 * and drain them, building a linear list of callbacks to run if everything
7807 * is successful (the transaction itself).
7810 visited
= g_hash_table_new(NULL
, NULL
);
7812 g_hash_table_add(visited
, ignore_child
);
7814 ret
= bdrv_change_aio_context(bs
, ctx
, visited
, tran
, errp
);
7815 g_hash_table_destroy(visited
);
7818 * Linear phase: go through all callbacks collected in the transaction.
7819 * Run all callbacks collected in the recursion to switch all nodes
7820 * AioContext lock (transaction commit), or undo all changes done in the
7821 * recursion (transaction abort).
7825 /* Just run clean() callbacks. No AioContext changed. */
7831 * Release old AioContext, it won't be needed anymore, as all
7832 * bdrv_drained_begin() have been called already.
7834 if (qemu_get_aio_context() != old_context
) {
7835 aio_context_release(old_context
);
7839 * Acquire new AioContext since bdrv_drained_end() is going to be called
7840 * after we switched all nodes in the new AioContext, and the function
7841 * assumes that the lock of the bs is always taken.
7843 if (qemu_get_aio_context() != ctx
) {
7844 aio_context_acquire(ctx
);
7849 if (qemu_get_aio_context() != ctx
) {
7850 aio_context_release(ctx
);
7853 /* Re-acquire the old AioContext, since the caller takes and releases it. */
7854 if (qemu_get_aio_context() != old_context
) {
7855 aio_context_acquire(old_context
);
7861 void bdrv_add_aio_context_notifier(BlockDriverState
*bs
,
7862 void (*attached_aio_context
)(AioContext
*new_context
, void *opaque
),
7863 void (*detach_aio_context
)(void *opaque
), void *opaque
)
7865 BdrvAioNotifier
*ban
= g_new(BdrvAioNotifier
, 1);
7866 *ban
= (BdrvAioNotifier
){
7867 .attached_aio_context
= attached_aio_context
,
7868 .detach_aio_context
= detach_aio_context
,
7871 GLOBAL_STATE_CODE();
7873 QLIST_INSERT_HEAD(&bs
->aio_notifiers
, ban
, list
);
7876 void bdrv_remove_aio_context_notifier(BlockDriverState
*bs
,
7877 void (*attached_aio_context
)(AioContext
*,
7879 void (*detach_aio_context
)(void *),
7882 BdrvAioNotifier
*ban
, *ban_next
;
7883 GLOBAL_STATE_CODE();
7885 QLIST_FOREACH_SAFE(ban
, &bs
->aio_notifiers
, list
, ban_next
) {
7886 if (ban
->attached_aio_context
== attached_aio_context
&&
7887 ban
->detach_aio_context
== detach_aio_context
&&
7888 ban
->opaque
== opaque
&&
7889 ban
->deleted
== false)
7891 if (bs
->walking_aio_notifiers
) {
7892 ban
->deleted
= true;
7894 bdrv_do_remove_aio_context_notifier(ban
);
7903 int bdrv_amend_options(BlockDriverState
*bs
, QemuOpts
*opts
,
7904 BlockDriverAmendStatusCB
*status_cb
, void *cb_opaque
,
7908 GLOBAL_STATE_CODE();
7910 error_setg(errp
, "Node is ejected");
7913 if (!bs
->drv
->bdrv_amend_options
) {
7914 error_setg(errp
, "Block driver '%s' does not support option amendment",
7915 bs
->drv
->format_name
);
7918 return bs
->drv
->bdrv_amend_options(bs
, opts
, status_cb
,
7919 cb_opaque
, force
, errp
);
7923 * This function checks whether the given @to_replace is allowed to be
7924 * replaced by a node that always shows the same data as @bs. This is
7925 * used for example to verify whether the mirror job can replace
7926 * @to_replace by the target mirrored from @bs.
7927 * To be replaceable, @bs and @to_replace may either be guaranteed to
7928 * always show the same data (because they are only connected through
7929 * filters), or some driver may allow replacing one of its children
7930 * because it can guarantee that this child's data is not visible at
7931 * all (for example, for dissenting quorum children that have no other
7934 bool bdrv_recurse_can_replace(BlockDriverState
*bs
,
7935 BlockDriverState
*to_replace
)
7937 BlockDriverState
*filtered
;
7939 GLOBAL_STATE_CODE();
7941 if (!bs
|| !bs
->drv
) {
7945 if (bs
== to_replace
) {
7949 /* See what the driver can do */
7950 if (bs
->drv
->bdrv_recurse_can_replace
) {
7951 return bs
->drv
->bdrv_recurse_can_replace(bs
, to_replace
);
7954 /* For filters without an own implementation, we can recurse on our own */
7955 filtered
= bdrv_filter_bs(bs
);
7957 return bdrv_recurse_can_replace(filtered
, to_replace
);
7965 * Check whether the given @node_name can be replaced by a node that
7966 * has the same data as @parent_bs. If so, return @node_name's BDS;
7969 * @node_name must be a (recursive) *child of @parent_bs (or this
7970 * function will return NULL).
7972 * The result (whether the node can be replaced or not) is only valid
7973 * for as long as no graph or permission changes occur.
7975 BlockDriverState
*check_to_replace_node(BlockDriverState
*parent_bs
,
7976 const char *node_name
, Error
**errp
)
7978 BlockDriverState
*to_replace_bs
= bdrv_find_node(node_name
);
7979 AioContext
*aio_context
;
7981 GLOBAL_STATE_CODE();
7983 if (!to_replace_bs
) {
7984 error_setg(errp
, "Failed to find node with node-name='%s'", node_name
);
7988 aio_context
= bdrv_get_aio_context(to_replace_bs
);
7989 aio_context_acquire(aio_context
);
7991 if (bdrv_op_is_blocked(to_replace_bs
, BLOCK_OP_TYPE_REPLACE
, errp
)) {
7992 to_replace_bs
= NULL
;
7996 /* We don't want arbitrary node of the BDS chain to be replaced only the top
7997 * most non filter in order to prevent data corruption.
7998 * Another benefit is that this tests exclude backing files which are
7999 * blocked by the backing blockers.
8001 if (!bdrv_recurse_can_replace(parent_bs
, to_replace_bs
)) {
8002 error_setg(errp
, "Cannot replace '%s' by a node mirrored from '%s', "
8003 "because it cannot be guaranteed that doing so would not "
8004 "lead to an abrupt change of visible data",
8005 node_name
, parent_bs
->node_name
);
8006 to_replace_bs
= NULL
;
8011 aio_context_release(aio_context
);
8012 return to_replace_bs
;
8016 * Iterates through the list of runtime option keys that are said to
8017 * be "strong" for a BDS. An option is called "strong" if it changes
8018 * a BDS's data. For example, the null block driver's "size" and
8019 * "read-zeroes" options are strong, but its "latency-ns" option is
8022 * If a key returned by this function ends with a dot, all options
8023 * starting with that prefix are strong.
8025 static const char *const *strong_options(BlockDriverState
*bs
,
8026 const char *const *curopt
)
8028 static const char *const global_options
[] = {
8029 "driver", "filename", NULL
8033 return &global_options
[0];
8037 if (curopt
== &global_options
[ARRAY_SIZE(global_options
) - 1] && bs
->drv
) {
8038 curopt
= bs
->drv
->strong_runtime_opts
;
8041 return (curopt
&& *curopt
) ? curopt
: NULL
;
8045 * Copies all strong runtime options from bs->options to the given
8046 * QDict. The set of strong option keys is determined by invoking
8049 * Returns true iff any strong option was present in bs->options (and
8050 * thus copied to the target QDict) with the exception of "filename"
8051 * and "driver". The caller is expected to use this value to decide
8052 * whether the existence of strong options prevents the generation of
8055 static bool append_strong_runtime_options(QDict
*d
, BlockDriverState
*bs
)
8057 bool found_any
= false;
8058 const char *const *option_name
= NULL
;
8064 while ((option_name
= strong_options(bs
, option_name
))) {
8065 bool option_given
= false;
8067 assert(strlen(*option_name
) > 0);
8068 if ((*option_name
)[strlen(*option_name
) - 1] != '.') {
8069 QObject
*entry
= qdict_get(bs
->options
, *option_name
);
8074 qdict_put_obj(d
, *option_name
, qobject_ref(entry
));
8075 option_given
= true;
8077 const QDictEntry
*entry
;
8078 for (entry
= qdict_first(bs
->options
); entry
;
8079 entry
= qdict_next(bs
->options
, entry
))
8081 if (strstart(qdict_entry_key(entry
), *option_name
, NULL
)) {
8082 qdict_put_obj(d
, qdict_entry_key(entry
),
8083 qobject_ref(qdict_entry_value(entry
)));
8084 option_given
= true;
8089 /* While "driver" and "filename" need to be included in a JSON filename,
8090 * their existence does not prohibit generation of a plain filename. */
8091 if (!found_any
&& option_given
&&
8092 strcmp(*option_name
, "driver") && strcmp(*option_name
, "filename"))
8098 if (!qdict_haskey(d
, "driver")) {
8099 /* Drivers created with bdrv_new_open_driver() may not have a
8100 * @driver option. Add it here. */
8101 qdict_put_str(d
, "driver", bs
->drv
->format_name
);
8107 /* Note: This function may return false positives; it may return true
8108 * even if opening the backing file specified by bs's image header
8109 * would result in exactly bs->backing. */
8110 static bool bdrv_backing_overridden(BlockDriverState
*bs
)
8112 GLOBAL_STATE_CODE();
8114 return strcmp(bs
->auto_backing_file
,
8115 bs
->backing
->bs
->filename
);
8117 /* No backing BDS, so if the image header reports any backing
8118 * file, it must have been suppressed */
8119 return bs
->auto_backing_file
[0] != '\0';
8123 /* Updates the following BDS fields:
8124 * - exact_filename: A filename which may be used for opening a block device
8125 * which (mostly) equals the given BDS (even without any
8126 * other options; so reading and writing must return the same
8127 * results, but caching etc. may be different)
8128 * - full_open_options: Options which, when given when opening a block device
8129 * (without a filename), result in a BDS (mostly)
8130 * equalling the given one
8131 * - filename: If exact_filename is set, it is copied here. Otherwise,
8132 * full_open_options is converted to a JSON object, prefixed with
8133 * "json:" (for use through the JSON pseudo protocol) and put here.
8135 void bdrv_refresh_filename(BlockDriverState
*bs
)
8137 BlockDriver
*drv
= bs
->drv
;
8139 BlockDriverState
*primary_child_bs
;
8141 bool backing_overridden
;
8142 bool generate_json_filename
; /* Whether our default implementation should
8143 fill exact_filename (false) or not (true) */
8145 GLOBAL_STATE_CODE();
8151 /* This BDS's file name may depend on any of its children's file names, so
8152 * refresh those first */
8153 QLIST_FOREACH(child
, &bs
->children
, next
) {
8154 bdrv_refresh_filename(child
->bs
);
8158 /* For implicit nodes, just copy everything from the single child */
8159 child
= QLIST_FIRST(&bs
->children
);
8160 assert(QLIST_NEXT(child
, next
) == NULL
);
8162 pstrcpy(bs
->exact_filename
, sizeof(bs
->exact_filename
),
8163 child
->bs
->exact_filename
);
8164 pstrcpy(bs
->filename
, sizeof(bs
->filename
), child
->bs
->filename
);
8166 qobject_unref(bs
->full_open_options
);
8167 bs
->full_open_options
= qobject_ref(child
->bs
->full_open_options
);
8172 backing_overridden
= bdrv_backing_overridden(bs
);
8174 if (bs
->open_flags
& BDRV_O_NO_IO
) {
8175 /* Without I/O, the backing file does not change anything.
8176 * Therefore, in such a case (primarily qemu-img), we can
8177 * pretend the backing file has not been overridden even if
8178 * it technically has been. */
8179 backing_overridden
= false;
8182 /* Gather the options QDict */
8184 generate_json_filename
= append_strong_runtime_options(opts
, bs
);
8185 generate_json_filename
|= backing_overridden
;
8187 if (drv
->bdrv_gather_child_options
) {
8188 /* Some block drivers may not want to present all of their children's
8189 * options, or name them differently from BdrvChild.name */
8190 drv
->bdrv_gather_child_options(bs
, opts
, backing_overridden
);
8192 QLIST_FOREACH(child
, &bs
->children
, next
) {
8193 if (child
== bs
->backing
&& !backing_overridden
) {
8194 /* We can skip the backing BDS if it has not been overridden */
8198 qdict_put(opts
, child
->name
,
8199 qobject_ref(child
->bs
->full_open_options
));
8202 if (backing_overridden
&& !bs
->backing
) {
8203 /* Force no backing file */
8204 qdict_put_null(opts
, "backing");
8208 qobject_unref(bs
->full_open_options
);
8209 bs
->full_open_options
= opts
;
8211 primary_child_bs
= bdrv_primary_bs(bs
);
8213 if (drv
->bdrv_refresh_filename
) {
8214 /* Obsolete information is of no use here, so drop the old file name
8215 * information before refreshing it */
8216 bs
->exact_filename
[0] = '\0';
8218 drv
->bdrv_refresh_filename(bs
);
8219 } else if (primary_child_bs
) {
8221 * Try to reconstruct valid information from the underlying
8222 * file -- this only works for format nodes (filter nodes
8223 * cannot be probed and as such must be selected by the user
8224 * either through an options dict, or through a special
8225 * filename which the filter driver must construct in its
8226 * .bdrv_refresh_filename() implementation).
8229 bs
->exact_filename
[0] = '\0';
8232 * We can use the underlying file's filename if:
8233 * - it has a filename,
8234 * - the current BDS is not a filter,
8235 * - the file is a protocol BDS, and
8236 * - opening that file (as this BDS's format) will automatically create
8237 * the BDS tree we have right now, that is:
8238 * - the user did not significantly change this BDS's behavior with
8239 * some explicit (strong) options
8240 * - no non-file child of this BDS has been overridden by the user
8241 * Both of these conditions are represented by generate_json_filename.
8243 if (primary_child_bs
->exact_filename
[0] &&
8244 primary_child_bs
->drv
->bdrv_file_open
&&
8245 !drv
->is_filter
&& !generate_json_filename
)
8247 strcpy(bs
->exact_filename
, primary_child_bs
->exact_filename
);
8251 if (bs
->exact_filename
[0]) {
8252 pstrcpy(bs
->filename
, sizeof(bs
->filename
), bs
->exact_filename
);
8254 GString
*json
= qobject_to_json(QOBJECT(bs
->full_open_options
));
8255 if (snprintf(bs
->filename
, sizeof(bs
->filename
), "json:%s",
8256 json
->str
) >= sizeof(bs
->filename
)) {
8257 /* Give user a hint if we truncated things. */
8258 strcpy(bs
->filename
+ sizeof(bs
->filename
) - 4, "...");
8260 g_string_free(json
, true);
8264 char *bdrv_dirname(BlockDriverState
*bs
, Error
**errp
)
8266 BlockDriver
*drv
= bs
->drv
;
8267 BlockDriverState
*child_bs
;
8269 GLOBAL_STATE_CODE();
8272 error_setg(errp
, "Node '%s' is ejected", bs
->node_name
);
8276 if (drv
->bdrv_dirname
) {
8277 return drv
->bdrv_dirname(bs
, errp
);
8280 child_bs
= bdrv_primary_bs(bs
);
8282 return bdrv_dirname(child_bs
, errp
);
8285 bdrv_refresh_filename(bs
);
8286 if (bs
->exact_filename
[0] != '\0') {
8287 return path_combine(bs
->exact_filename
, "");
8290 error_setg(errp
, "Cannot generate a base directory for %s nodes",
8296 * Hot add/remove a BDS's child. So the user can take a child offline when
8297 * it is broken and take a new child online
8299 void bdrv_add_child(BlockDriverState
*parent_bs
, BlockDriverState
*child_bs
,
8302 GLOBAL_STATE_CODE();
8303 if (!parent_bs
->drv
|| !parent_bs
->drv
->bdrv_add_child
) {
8304 error_setg(errp
, "The node %s does not support adding a child",
8305 bdrv_get_device_or_node_name(parent_bs
));
8310 * Non-zoned block drivers do not follow zoned storage constraints
8311 * (i.e. sequential writes to zones). Refuse mixing zoned and non-zoned
8312 * drivers in a graph.
8314 if (!parent_bs
->drv
->supports_zoned_children
&&
8315 child_bs
->bl
.zoned
== BLK_Z_HM
) {
8317 * The host-aware model allows zoned storage constraints and random
8318 * write. Allow mixing host-aware and non-zoned drivers. Using
8319 * host-aware device as a regular device.
8321 error_setg(errp
, "Cannot add a %s child to a %s parent",
8322 child_bs
->bl
.zoned
== BLK_Z_HM
? "zoned" : "non-zoned",
8323 parent_bs
->drv
->supports_zoned_children
?
8324 "support zoned children" : "not support zoned children");
8328 if (!QLIST_EMPTY(&child_bs
->parents
)) {
8329 error_setg(errp
, "The node %s already has a parent",
8330 child_bs
->node_name
);
8334 parent_bs
->drv
->bdrv_add_child(parent_bs
, child_bs
, errp
);
8337 void bdrv_del_child(BlockDriverState
*parent_bs
, BdrvChild
*child
, Error
**errp
)
8341 GLOBAL_STATE_CODE();
8342 if (!parent_bs
->drv
|| !parent_bs
->drv
->bdrv_del_child
) {
8343 error_setg(errp
, "The node %s does not support removing a child",
8344 bdrv_get_device_or_node_name(parent_bs
));
8348 QLIST_FOREACH(tmp
, &parent_bs
->children
, next
) {
8355 error_setg(errp
, "The node %s does not have a child named %s",
8356 bdrv_get_device_or_node_name(parent_bs
),
8357 bdrv_get_device_or_node_name(child
->bs
));
8361 parent_bs
->drv
->bdrv_del_child(parent_bs
, child
, errp
);
8364 int bdrv_make_empty(BdrvChild
*c
, Error
**errp
)
8366 BlockDriver
*drv
= c
->bs
->drv
;
8369 GLOBAL_STATE_CODE();
8370 assert(c
->perm
& (BLK_PERM_WRITE
| BLK_PERM_WRITE_UNCHANGED
));
8372 if (!drv
->bdrv_make_empty
) {
8373 error_setg(errp
, "%s does not support emptying nodes",
8378 ret
= drv
->bdrv_make_empty(c
->bs
);
8380 error_setg_errno(errp
, -ret
, "Failed to empty %s",
8389 * Return the child that @bs acts as an overlay for, and from which data may be
8390 * copied in COW or COR operations. Usually this is the backing file.
8392 BdrvChild
*bdrv_cow_child(BlockDriverState
*bs
)
8396 if (!bs
|| !bs
->drv
) {
8400 if (bs
->drv
->is_filter
) {
8408 assert(bs
->backing
->role
& BDRV_CHILD_COW
);
8413 * If @bs acts as a filter for exactly one of its children, return
8416 BdrvChild
*bdrv_filter_child(BlockDriverState
*bs
)
8421 if (!bs
|| !bs
->drv
) {
8425 if (!bs
->drv
->is_filter
) {
8429 /* Only one of @backing or @file may be used */
8430 assert(!(bs
->backing
&& bs
->file
));
8432 c
= bs
->backing
?: bs
->file
;
8437 assert(c
->role
& BDRV_CHILD_FILTERED
);
8442 * Return either the result of bdrv_cow_child() or bdrv_filter_child(),
8443 * whichever is non-NULL.
8445 * Return NULL if both are NULL.
8447 BdrvChild
*bdrv_filter_or_cow_child(BlockDriverState
*bs
)
8449 BdrvChild
*cow_child
= bdrv_cow_child(bs
);
8450 BdrvChild
*filter_child
= bdrv_filter_child(bs
);
8453 /* Filter nodes cannot have COW backing files */
8454 assert(!(cow_child
&& filter_child
));
8456 return cow_child
?: filter_child
;
8460 * Return the primary child of this node: For filters, that is the
8461 * filtered child. For other nodes, that is usually the child storing
8463 * (A generally more helpful description is that this is (usually) the
8464 * child that has the same filename as @bs.)
8466 * Drivers do not necessarily have a primary child; for example quorum
8469 BdrvChild
*bdrv_primary_child(BlockDriverState
*bs
)
8471 BdrvChild
*c
, *found
= NULL
;
8474 QLIST_FOREACH(c
, &bs
->children
, next
) {
8475 if (c
->role
& BDRV_CHILD_PRIMARY
) {
8484 static BlockDriverState
*bdrv_do_skip_filters(BlockDriverState
*bs
,
8485 bool stop_on_explicit_filter
)
8493 while (!(stop_on_explicit_filter
&& !bs
->implicit
)) {
8494 c
= bdrv_filter_child(bs
);
8497 * A filter that is embedded in a working block graph must
8498 * have a child. Assert this here so this function does
8499 * not return a filter node that is not expected by the
8502 assert(!bs
->drv
|| !bs
->drv
->is_filter
);
8508 * Note that this treats nodes with bs->drv == NULL as not being
8509 * filters (bs->drv == NULL should be replaced by something else
8511 * The advantage of this behavior is that this function will thus
8512 * always return a non-NULL value (given a non-NULL @bs).
8519 * Return the first BDS that has not been added implicitly or that
8520 * does not have a filtered child down the chain starting from @bs
8521 * (including @bs itself).
8523 BlockDriverState
*bdrv_skip_implicit_filters(BlockDriverState
*bs
)
8525 GLOBAL_STATE_CODE();
8526 return bdrv_do_skip_filters(bs
, true);
8530 * Return the first BDS that does not have a filtered child down the
8531 * chain starting from @bs (including @bs itself).
8533 BlockDriverState
*bdrv_skip_filters(BlockDriverState
*bs
)
8536 return bdrv_do_skip_filters(bs
, false);
8540 * For a backing chain, return the first non-filter backing image of
8541 * the first non-filter image.
8543 BlockDriverState
*bdrv_backing_chain_next(BlockDriverState
*bs
)
8546 return bdrv_skip_filters(bdrv_cow_bs(bdrv_skip_filters(bs
)));
8550 * Check whether [offset, offset + bytes) overlaps with the cached
8551 * block-status data region.
8553 * If so, and @pnum is not NULL, set *pnum to `bsc.data_end - offset`,
8554 * which is what bdrv_bsc_is_data()'s interface needs.
8555 * Otherwise, *pnum is not touched.
8557 static bool bdrv_bsc_range_overlaps_locked(BlockDriverState
*bs
,
8558 int64_t offset
, int64_t bytes
,
8561 BdrvBlockStatusCache
*bsc
= qatomic_rcu_read(&bs
->block_status_cache
);
8565 qatomic_read(&bsc
->valid
) &&
8566 ranges_overlap(offset
, bytes
, bsc
->data_start
,
8567 bsc
->data_end
- bsc
->data_start
);
8569 if (overlaps
&& pnum
) {
8570 *pnum
= bsc
->data_end
- offset
;
8577 * See block_int.h for this function's documentation.
8579 bool bdrv_bsc_is_data(BlockDriverState
*bs
, int64_t offset
, int64_t *pnum
)
8582 RCU_READ_LOCK_GUARD();
8583 return bdrv_bsc_range_overlaps_locked(bs
, offset
, 1, pnum
);
8587 * See block_int.h for this function's documentation.
8589 void bdrv_bsc_invalidate_range(BlockDriverState
*bs
,
8590 int64_t offset
, int64_t bytes
)
8593 RCU_READ_LOCK_GUARD();
8595 if (bdrv_bsc_range_overlaps_locked(bs
, offset
, bytes
, NULL
)) {
8596 qatomic_set(&bs
->block_status_cache
->valid
, false);
8601 * See block_int.h for this function's documentation.
8603 void bdrv_bsc_fill(BlockDriverState
*bs
, int64_t offset
, int64_t bytes
)
8605 BdrvBlockStatusCache
*new_bsc
= g_new(BdrvBlockStatusCache
, 1);
8606 BdrvBlockStatusCache
*old_bsc
;
8609 *new_bsc
= (BdrvBlockStatusCache
) {
8611 .data_start
= offset
,
8612 .data_end
= offset
+ bytes
,
8615 QEMU_LOCK_GUARD(&bs
->bsc_modify_lock
);
8617 old_bsc
= qatomic_rcu_read(&bs
->block_status_cache
);
8618 qatomic_rcu_set(&bs
->block_status_cache
, new_bsc
);
8620 g_free_rcu(old_bsc
, rcu
);