2 * QEMU System Emulator block driver
4 * Copyright (c) 2003 Fabrice Bellard
6 * Permission is hereby granted, free of charge, to any person obtaining a copy
7 * of this software and associated documentation files (the "Software"), to deal
8 * in the Software without restriction, including without limitation the rights
9 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10 * copies of the Software, and to permit persons to whom the Software is
11 * furnished to do so, subject to the following conditions:
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
25 #include "qemu/osdep.h"
26 #include "block/trace.h"
27 #include "block/block_int.h"
28 #include "block/blockjob.h"
29 #include "block/nbd.h"
30 #include "block/qdict.h"
31 #include "qemu/error-report.h"
32 #include "block/module_block.h"
33 #include "qemu/main-loop.h"
34 #include "qemu/module.h"
35 #include "qapi/error.h"
36 #include "qapi/qmp/qdict.h"
37 #include "qapi/qmp/qjson.h"
38 #include "qapi/qmp/qnull.h"
39 #include "qapi/qmp/qstring.h"
40 #include "qapi/qobject-output-visitor.h"
41 #include "qapi/qapi-visit-block-core.h"
42 #include "sysemu/block-backend.h"
43 #include "sysemu/sysemu.h"
44 #include "qemu/notify.h"
45 #include "qemu/option.h"
46 #include "qemu/coroutine.h"
47 #include "block/qapi.h"
48 #include "qemu/timer.h"
49 #include "qemu/cutils.h"
53 #include <sys/ioctl.h>
54 #include <sys/queue.h>
64 #define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */
66 static QTAILQ_HEAD(, BlockDriverState
) graph_bdrv_states
=
67 QTAILQ_HEAD_INITIALIZER(graph_bdrv_states
);
69 static QTAILQ_HEAD(, BlockDriverState
) all_bdrv_states
=
70 QTAILQ_HEAD_INITIALIZER(all_bdrv_states
);
72 static QLIST_HEAD(, BlockDriver
) bdrv_drivers
=
73 QLIST_HEAD_INITIALIZER(bdrv_drivers
);
75 static BlockDriverState
*bdrv_open_inherit(const char *filename
,
76 const char *reference
,
77 QDict
*options
, int flags
,
78 BlockDriverState
*parent
,
79 const BdrvChildClass
*child_class
,
80 BdrvChildRole child_role
,
83 /* If non-zero, use only whitelisted block drivers */
84 static int use_bdrv_whitelist
;
87 static int is_windows_drive_prefix(const char *filename
)
89 return (((filename
[0] >= 'a' && filename
[0] <= 'z') ||
90 (filename
[0] >= 'A' && filename
[0] <= 'Z')) &&
94 int is_windows_drive(const char *filename
)
96 if (is_windows_drive_prefix(filename
) &&
99 if (strstart(filename
, "\\\\.\\", NULL
) ||
100 strstart(filename
, "//./", NULL
))
106 size_t bdrv_opt_mem_align(BlockDriverState
*bs
)
108 if (!bs
|| !bs
->drv
) {
109 /* page size or 4k (hdd sector size) should be on the safe side */
110 return MAX(4096, qemu_real_host_page_size
);
113 return bs
->bl
.opt_mem_alignment
;
116 size_t bdrv_min_mem_align(BlockDriverState
*bs
)
118 if (!bs
|| !bs
->drv
) {
119 /* page size or 4k (hdd sector size) should be on the safe side */
120 return MAX(4096, qemu_real_host_page_size
);
123 return bs
->bl
.min_mem_alignment
;
126 /* check if the path starts with "<protocol>:" */
127 int path_has_protocol(const char *path
)
132 if (is_windows_drive(path
) ||
133 is_windows_drive_prefix(path
)) {
136 p
= path
+ strcspn(path
, ":/\\");
138 p
= path
+ strcspn(path
, ":/");
144 int path_is_absolute(const char *path
)
147 /* specific case for names like: "\\.\d:" */
148 if (is_windows_drive(path
) || is_windows_drive_prefix(path
)) {
151 return (*path
== '/' || *path
== '\\');
153 return (*path
== '/');
157 /* if filename is absolute, just return its duplicate. Otherwise, build a
158 path to it by considering it is relative to base_path. URL are
160 char *path_combine(const char *base_path
, const char *filename
)
162 const char *protocol_stripped
= NULL
;
167 if (path_is_absolute(filename
)) {
168 return g_strdup(filename
);
171 if (path_has_protocol(base_path
)) {
172 protocol_stripped
= strchr(base_path
, ':');
173 if (protocol_stripped
) {
177 p
= protocol_stripped
?: base_path
;
179 p1
= strrchr(base_path
, '/');
183 p2
= strrchr(base_path
, '\\');
184 if (!p1
|| p2
> p1
) {
199 result
= g_malloc(len
+ strlen(filename
) + 1);
200 memcpy(result
, base_path
, len
);
201 strcpy(result
+ len
, filename
);
207 * Helper function for bdrv_parse_filename() implementations to remove optional
208 * protocol prefixes (especially "file:") from a filename and for putting the
209 * stripped filename into the options QDict if there is such a prefix.
211 void bdrv_parse_filename_strip_prefix(const char *filename
, const char *prefix
,
214 if (strstart(filename
, prefix
, &filename
)) {
215 /* Stripping the explicit protocol prefix may result in a protocol
216 * prefix being (wrongly) detected (if the filename contains a colon) */
217 if (path_has_protocol(filename
)) {
218 QString
*fat_filename
;
220 /* This means there is some colon before the first slash; therefore,
221 * this cannot be an absolute path */
222 assert(!path_is_absolute(filename
));
224 /* And we can thus fix the protocol detection issue by prefixing it
226 fat_filename
= qstring_from_str("./");
227 qstring_append(fat_filename
, filename
);
229 assert(!path_has_protocol(qstring_get_str(fat_filename
)));
231 qdict_put(options
, "filename", fat_filename
);
233 /* If no protocol prefix was detected, we can use the shortened
235 qdict_put_str(options
, "filename", filename
);
241 /* Returns whether the image file is opened as read-only. Note that this can
242 * return false and writing to the image file is still not possible because the
243 * image is inactivated. */
244 bool bdrv_is_read_only(BlockDriverState
*bs
)
246 return bs
->read_only
;
249 int bdrv_can_set_read_only(BlockDriverState
*bs
, bool read_only
,
250 bool ignore_allow_rdw
, Error
**errp
)
252 /* Do not set read_only if copy_on_read is enabled */
253 if (bs
->copy_on_read
&& read_only
) {
254 error_setg(errp
, "Can't set node '%s' to r/o with copy-on-read enabled",
255 bdrv_get_device_or_node_name(bs
));
259 /* Do not clear read_only if it is prohibited */
260 if (!read_only
&& !(bs
->open_flags
& BDRV_O_ALLOW_RDWR
) &&
263 error_setg(errp
, "Node '%s' is read only",
264 bdrv_get_device_or_node_name(bs
));
272 * Called by a driver that can only provide a read-only image.
274 * Returns 0 if the node is already read-only or it could switch the node to
275 * read-only because BDRV_O_AUTO_RDONLY is set.
277 * Returns -EACCES if the node is read-write and BDRV_O_AUTO_RDONLY is not set
278 * or bdrv_can_set_read_only() forbids making the node read-only. If @errmsg
279 * is not NULL, it is used as the error message for the Error object.
281 int bdrv_apply_auto_read_only(BlockDriverState
*bs
, const char *errmsg
,
286 if (!(bs
->open_flags
& BDRV_O_RDWR
)) {
289 if (!(bs
->open_flags
& BDRV_O_AUTO_RDONLY
)) {
293 ret
= bdrv_can_set_read_only(bs
, true, false, NULL
);
298 bs
->read_only
= true;
299 bs
->open_flags
&= ~BDRV_O_RDWR
;
304 error_setg(errp
, "%s", errmsg
?: "Image is read-only");
309 * If @backing is empty, this function returns NULL without setting
310 * @errp. In all other cases, NULL will only be returned with @errp
313 * Therefore, a return value of NULL without @errp set means that
314 * there is no backing file; if @errp is set, there is one but its
315 * absolute filename cannot be generated.
317 char *bdrv_get_full_backing_filename_from_filename(const char *backed
,
321 if (backing
[0] == '\0') {
323 } else if (path_has_protocol(backing
) || path_is_absolute(backing
)) {
324 return g_strdup(backing
);
325 } else if (backed
[0] == '\0' || strstart(backed
, "json:", NULL
)) {
326 error_setg(errp
, "Cannot use relative backing file names for '%s'",
330 return path_combine(backed
, backing
);
335 * If @filename is empty or NULL, this function returns NULL without
336 * setting @errp. In all other cases, NULL will only be returned with
339 static char *bdrv_make_absolute_filename(BlockDriverState
*relative_to
,
340 const char *filename
, Error
**errp
)
342 char *dir
, *full_name
;
344 if (!filename
|| filename
[0] == '\0') {
346 } else if (path_has_protocol(filename
) || path_is_absolute(filename
)) {
347 return g_strdup(filename
);
350 dir
= bdrv_dirname(relative_to
, errp
);
355 full_name
= g_strconcat(dir
, filename
, NULL
);
360 char *bdrv_get_full_backing_filename(BlockDriverState
*bs
, Error
**errp
)
362 return bdrv_make_absolute_filename(bs
, bs
->backing_file
, errp
);
365 void bdrv_register(BlockDriver
*bdrv
)
367 assert(bdrv
->format_name
);
368 QLIST_INSERT_HEAD(&bdrv_drivers
, bdrv
, list
);
371 BlockDriverState
*bdrv_new(void)
373 BlockDriverState
*bs
;
376 bs
= g_new0(BlockDriverState
, 1);
377 QLIST_INIT(&bs
->dirty_bitmaps
);
378 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
379 QLIST_INIT(&bs
->op_blockers
[i
]);
381 notifier_with_return_list_init(&bs
->before_write_notifiers
);
382 qemu_co_mutex_init(&bs
->reqs_lock
);
383 qemu_mutex_init(&bs
->dirty_bitmap_mutex
);
385 bs
->aio_context
= qemu_get_aio_context();
387 qemu_co_queue_init(&bs
->flush_queue
);
389 for (i
= 0; i
< bdrv_drain_all_count
; i
++) {
390 bdrv_drained_begin(bs
);
393 QTAILQ_INSERT_TAIL(&all_bdrv_states
, bs
, bs_list
);
398 static BlockDriver
*bdrv_do_find_format(const char *format_name
)
402 QLIST_FOREACH(drv1
, &bdrv_drivers
, list
) {
403 if (!strcmp(drv1
->format_name
, format_name
)) {
411 BlockDriver
*bdrv_find_format(const char *format_name
)
416 drv1
= bdrv_do_find_format(format_name
);
421 /* The driver isn't registered, maybe we need to load a module */
422 for (i
= 0; i
< (int)ARRAY_SIZE(block_driver_modules
); ++i
) {
423 if (!strcmp(block_driver_modules
[i
].format_name
, format_name
)) {
424 block_module_load_one(block_driver_modules
[i
].library_name
);
429 return bdrv_do_find_format(format_name
);
432 static int bdrv_format_is_whitelisted(const char *format_name
, bool read_only
)
434 static const char *whitelist_rw
[] = {
435 CONFIG_BDRV_RW_WHITELIST
438 static const char *whitelist_ro
[] = {
439 CONFIG_BDRV_RO_WHITELIST
444 if (!whitelist_rw
[0] && !whitelist_ro
[0]) {
445 return 1; /* no whitelist, anything goes */
448 for (p
= whitelist_rw
; *p
; p
++) {
449 if (!strcmp(format_name
, *p
)) {
454 for (p
= whitelist_ro
; *p
; p
++) {
455 if (!strcmp(format_name
, *p
)) {
463 int bdrv_is_whitelisted(BlockDriver
*drv
, bool read_only
)
465 return bdrv_format_is_whitelisted(drv
->format_name
, read_only
);
468 bool bdrv_uses_whitelist(void)
470 return use_bdrv_whitelist
;
473 typedef struct CreateCo
{
481 static void coroutine_fn
bdrv_create_co_entry(void *opaque
)
483 Error
*local_err
= NULL
;
486 CreateCo
*cco
= opaque
;
489 ret
= cco
->drv
->bdrv_co_create_opts(cco
->drv
,
490 cco
->filename
, cco
->opts
, &local_err
);
491 error_propagate(&cco
->err
, local_err
);
495 int bdrv_create(BlockDriver
*drv
, const char* filename
,
496 QemuOpts
*opts
, Error
**errp
)
503 .filename
= g_strdup(filename
),
509 if (!drv
->bdrv_co_create_opts
) {
510 error_setg(errp
, "Driver '%s' does not support image creation", drv
->format_name
);
515 if (qemu_in_coroutine()) {
516 /* Fast-path if already in coroutine context */
517 bdrv_create_co_entry(&cco
);
519 co
= qemu_coroutine_create(bdrv_create_co_entry
, &cco
);
520 qemu_coroutine_enter(co
);
521 while (cco
.ret
== NOT_DONE
) {
522 aio_poll(qemu_get_aio_context(), true);
529 error_propagate(errp
, cco
.err
);
531 error_setg_errno(errp
, -ret
, "Could not create image");
536 g_free(cco
.filename
);
541 * Helper function for bdrv_create_file_fallback(): Resize @blk to at
542 * least the given @minimum_size.
544 * On success, return @blk's actual length.
545 * Otherwise, return -errno.
547 static int64_t create_file_fallback_truncate(BlockBackend
*blk
,
548 int64_t minimum_size
, Error
**errp
)
550 Error
*local_err
= NULL
;
554 ret
= blk_truncate(blk
, minimum_size
, false, PREALLOC_MODE_OFF
, 0,
556 if (ret
< 0 && ret
!= -ENOTSUP
) {
557 error_propagate(errp
, local_err
);
561 size
= blk_getlength(blk
);
563 error_free(local_err
);
564 error_setg_errno(errp
, -size
,
565 "Failed to inquire the new image file's length");
569 if (size
< minimum_size
) {
570 /* Need to grow the image, but we failed to do that */
571 error_propagate(errp
, local_err
);
575 error_free(local_err
);
582 * Helper function for bdrv_create_file_fallback(): Zero the first
583 * sector to remove any potentially pre-existing image header.
585 static int create_file_fallback_zero_first_sector(BlockBackend
*blk
,
586 int64_t current_size
,
589 int64_t bytes_to_clear
;
592 bytes_to_clear
= MIN(current_size
, BDRV_SECTOR_SIZE
);
593 if (bytes_to_clear
) {
594 ret
= blk_pwrite_zeroes(blk
, 0, bytes_to_clear
, BDRV_REQ_MAY_UNMAP
);
596 error_setg_errno(errp
, -ret
,
597 "Failed to clear the new image's first sector");
606 * Simple implementation of bdrv_co_create_opts for protocol drivers
607 * which only support creation via opening a file
608 * (usually existing raw storage device)
610 int coroutine_fn
bdrv_co_create_opts_simple(BlockDriver
*drv
,
611 const char *filename
,
619 PreallocMode prealloc
;
620 Error
*local_err
= NULL
;
623 size
= qemu_opt_get_size_del(opts
, BLOCK_OPT_SIZE
, 0);
624 buf
= qemu_opt_get_del(opts
, BLOCK_OPT_PREALLOC
);
625 prealloc
= qapi_enum_parse(&PreallocMode_lookup
, buf
,
626 PREALLOC_MODE_OFF
, &local_err
);
629 error_propagate(errp
, local_err
);
633 if (prealloc
!= PREALLOC_MODE_OFF
) {
634 error_setg(errp
, "Unsupported preallocation mode '%s'",
635 PreallocMode_str(prealloc
));
639 options
= qdict_new();
640 qdict_put_str(options
, "driver", drv
->format_name
);
642 blk
= blk_new_open(filename
, NULL
, options
,
643 BDRV_O_RDWR
| BDRV_O_RESIZE
, errp
);
645 error_prepend(errp
, "Protocol driver '%s' does not support image "
646 "creation, and opening the image failed: ",
651 size
= create_file_fallback_truncate(blk
, size
, errp
);
657 ret
= create_file_fallback_zero_first_sector(blk
, size
, errp
);
668 int bdrv_create_file(const char *filename
, QemuOpts
*opts
, Error
**errp
)
672 drv
= bdrv_find_protocol(filename
, true, errp
);
677 return bdrv_create(drv
, filename
, opts
, errp
);
680 int coroutine_fn
bdrv_co_delete_file(BlockDriverState
*bs
, Error
**errp
)
682 Error
*local_err
= NULL
;
688 error_setg(errp
, "Block node '%s' is not opened", bs
->filename
);
692 if (!bs
->drv
->bdrv_co_delete_file
) {
693 error_setg(errp
, "Driver '%s' does not support image deletion",
694 bs
->drv
->format_name
);
698 ret
= bs
->drv
->bdrv_co_delete_file(bs
, &local_err
);
700 error_propagate(errp
, local_err
);
707 * Try to get @bs's logical and physical block size.
708 * On success, store them in @bsz struct and return 0.
709 * On failure return -errno.
710 * @bs must not be empty.
712 int bdrv_probe_blocksizes(BlockDriverState
*bs
, BlockSizes
*bsz
)
714 BlockDriver
*drv
= bs
->drv
;
716 if (drv
&& drv
->bdrv_probe_blocksizes
) {
717 return drv
->bdrv_probe_blocksizes(bs
, bsz
);
718 } else if (drv
&& drv
->is_filter
&& bs
->file
) {
719 return bdrv_probe_blocksizes(bs
->file
->bs
, bsz
);
726 * Try to get @bs's geometry (cyls, heads, sectors).
727 * On success, store them in @geo struct and return 0.
728 * On failure return -errno.
729 * @bs must not be empty.
731 int bdrv_probe_geometry(BlockDriverState
*bs
, HDGeometry
*geo
)
733 BlockDriver
*drv
= bs
->drv
;
735 if (drv
&& drv
->bdrv_probe_geometry
) {
736 return drv
->bdrv_probe_geometry(bs
, geo
);
737 } else if (drv
&& drv
->is_filter
&& bs
->file
) {
738 return bdrv_probe_geometry(bs
->file
->bs
, geo
);
745 * Create a uniquely-named empty temporary file.
746 * Return 0 upon success, otherwise a negative errno value.
748 int get_tmp_filename(char *filename
, int size
)
751 char temp_dir
[MAX_PATH
];
752 /* GetTempFileName requires that its output buffer (4th param)
753 have length MAX_PATH or greater. */
754 assert(size
>= MAX_PATH
);
755 return (GetTempPath(MAX_PATH
, temp_dir
)
756 && GetTempFileName(temp_dir
, "qem", 0, filename
)
757 ? 0 : -GetLastError());
761 tmpdir
= getenv("TMPDIR");
765 if (snprintf(filename
, size
, "%s/vl.XXXXXX", tmpdir
) >= size
) {
768 fd
= mkstemp(filename
);
772 if (close(fd
) != 0) {
781 * Detect host devices. By convention, /dev/cdrom[N] is always
782 * recognized as a host CDROM.
784 static BlockDriver
*find_hdev_driver(const char *filename
)
786 int score_max
= 0, score
;
787 BlockDriver
*drv
= NULL
, *d
;
789 QLIST_FOREACH(d
, &bdrv_drivers
, list
) {
790 if (d
->bdrv_probe_device
) {
791 score
= d
->bdrv_probe_device(filename
);
792 if (score
> score_max
) {
802 static BlockDriver
*bdrv_do_find_protocol(const char *protocol
)
806 QLIST_FOREACH(drv1
, &bdrv_drivers
, list
) {
807 if (drv1
->protocol_name
&& !strcmp(drv1
->protocol_name
, protocol
)) {
815 BlockDriver
*bdrv_find_protocol(const char *filename
,
816 bool allow_protocol_prefix
,
825 /* TODO Drivers without bdrv_file_open must be specified explicitly */
828 * XXX(hch): we really should not let host device detection
829 * override an explicit protocol specification, but moving this
830 * later breaks access to device names with colons in them.
831 * Thanks to the brain-dead persistent naming schemes on udev-
832 * based Linux systems those actually are quite common.
834 drv1
= find_hdev_driver(filename
);
839 if (!path_has_protocol(filename
) || !allow_protocol_prefix
) {
843 p
= strchr(filename
, ':');
846 if (len
> sizeof(protocol
) - 1)
847 len
= sizeof(protocol
) - 1;
848 memcpy(protocol
, filename
, len
);
849 protocol
[len
] = '\0';
851 drv1
= bdrv_do_find_protocol(protocol
);
856 for (i
= 0; i
< (int)ARRAY_SIZE(block_driver_modules
); ++i
) {
857 if (block_driver_modules
[i
].protocol_name
&&
858 !strcmp(block_driver_modules
[i
].protocol_name
, protocol
)) {
859 block_module_load_one(block_driver_modules
[i
].library_name
);
864 drv1
= bdrv_do_find_protocol(protocol
);
866 error_setg(errp
, "Unknown protocol '%s'", protocol
);
872 * Guess image format by probing its contents.
873 * This is not a good idea when your image is raw (CVE-2008-2004), but
874 * we do it anyway for backward compatibility.
876 * @buf contains the image's first @buf_size bytes.
877 * @buf_size is the buffer size in bytes (generally BLOCK_PROBE_BUF_SIZE,
878 * but can be smaller if the image file is smaller)
879 * @filename is its filename.
881 * For all block drivers, call the bdrv_probe() method to get its
883 * Return the first block driver with the highest probing score.
885 BlockDriver
*bdrv_probe_all(const uint8_t *buf
, int buf_size
,
886 const char *filename
)
888 int score_max
= 0, score
;
889 BlockDriver
*drv
= NULL
, *d
;
891 QLIST_FOREACH(d
, &bdrv_drivers
, list
) {
893 score
= d
->bdrv_probe(buf
, buf_size
, filename
);
894 if (score
> score_max
) {
904 static int find_image_format(BlockBackend
*file
, const char *filename
,
905 BlockDriver
**pdrv
, Error
**errp
)
908 uint8_t buf
[BLOCK_PROBE_BUF_SIZE
];
911 /* Return the raw BlockDriver * to scsi-generic devices or empty drives */
912 if (blk_is_sg(file
) || !blk_is_inserted(file
) || blk_getlength(file
) == 0) {
917 ret
= blk_pread(file
, 0, buf
, sizeof(buf
));
919 error_setg_errno(errp
, -ret
, "Could not read image for determining its "
925 drv
= bdrv_probe_all(buf
, ret
, filename
);
927 error_setg(errp
, "Could not determine image format: No compatible "
936 * Set the current 'total_sectors' value
937 * Return 0 on success, -errno on error.
939 int refresh_total_sectors(BlockDriverState
*bs
, int64_t hint
)
941 BlockDriver
*drv
= bs
->drv
;
947 /* Do not attempt drv->bdrv_getlength() on scsi-generic devices */
951 /* query actual device if possible, otherwise just trust the hint */
952 if (drv
->bdrv_getlength
) {
953 int64_t length
= drv
->bdrv_getlength(bs
);
957 hint
= DIV_ROUND_UP(length
, BDRV_SECTOR_SIZE
);
960 bs
->total_sectors
= hint
;
965 * Combines a QDict of new block driver @options with any missing options taken
966 * from @old_options, so that leaving out an option defaults to its old value.
968 static void bdrv_join_options(BlockDriverState
*bs
, QDict
*options
,
971 if (bs
->drv
&& bs
->drv
->bdrv_join_options
) {
972 bs
->drv
->bdrv_join_options(options
, old_options
);
974 qdict_join(options
, old_options
, false);
978 static BlockdevDetectZeroesOptions
bdrv_parse_detect_zeroes(QemuOpts
*opts
,
982 Error
*local_err
= NULL
;
983 char *value
= qemu_opt_get_del(opts
, "detect-zeroes");
984 BlockdevDetectZeroesOptions detect_zeroes
=
985 qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup
, value
,
986 BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF
, &local_err
);
989 error_propagate(errp
, local_err
);
990 return detect_zeroes
;
993 if (detect_zeroes
== BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP
&&
994 !(open_flags
& BDRV_O_UNMAP
))
996 error_setg(errp
, "setting detect-zeroes to unmap is not allowed "
997 "without setting discard operation to unmap");
1000 return detect_zeroes
;
1004 * Set open flags for aio engine
1006 * Return 0 on success, -1 if the engine specified is invalid
1008 int bdrv_parse_aio(const char *mode
, int *flags
)
1010 if (!strcmp(mode
, "threads")) {
1011 /* do nothing, default */
1012 } else if (!strcmp(mode
, "native")) {
1013 *flags
|= BDRV_O_NATIVE_AIO
;
1014 #ifdef CONFIG_LINUX_IO_URING
1015 } else if (!strcmp(mode
, "io_uring")) {
1016 *flags
|= BDRV_O_IO_URING
;
1026 * Set open flags for a given discard mode
1028 * Return 0 on success, -1 if the discard mode was invalid.
1030 int bdrv_parse_discard_flags(const char *mode
, int *flags
)
1032 *flags
&= ~BDRV_O_UNMAP
;
1034 if (!strcmp(mode
, "off") || !strcmp(mode
, "ignore")) {
1036 } else if (!strcmp(mode
, "on") || !strcmp(mode
, "unmap")) {
1037 *flags
|= BDRV_O_UNMAP
;
1046 * Set open flags for a given cache mode
1048 * Return 0 on success, -1 if the cache mode was invalid.
1050 int bdrv_parse_cache_mode(const char *mode
, int *flags
, bool *writethrough
)
1052 *flags
&= ~BDRV_O_CACHE_MASK
;
1054 if (!strcmp(mode
, "off") || !strcmp(mode
, "none")) {
1055 *writethrough
= false;
1056 *flags
|= BDRV_O_NOCACHE
;
1057 } else if (!strcmp(mode
, "directsync")) {
1058 *writethrough
= true;
1059 *flags
|= BDRV_O_NOCACHE
;
1060 } else if (!strcmp(mode
, "writeback")) {
1061 *writethrough
= false;
1062 } else if (!strcmp(mode
, "unsafe")) {
1063 *writethrough
= false;
1064 *flags
|= BDRV_O_NO_FLUSH
;
1065 } else if (!strcmp(mode
, "writethrough")) {
1066 *writethrough
= true;
1074 static char *bdrv_child_get_parent_desc(BdrvChild
*c
)
1076 BlockDriverState
*parent
= c
->opaque
;
1077 return g_strdup(bdrv_get_device_or_node_name(parent
));
1080 static void bdrv_child_cb_drained_begin(BdrvChild
*child
)
1082 BlockDriverState
*bs
= child
->opaque
;
1083 bdrv_do_drained_begin_quiesce(bs
, NULL
, false);
1086 static bool bdrv_child_cb_drained_poll(BdrvChild
*child
)
1088 BlockDriverState
*bs
= child
->opaque
;
1089 return bdrv_drain_poll(bs
, false, NULL
, false);
1092 static void bdrv_child_cb_drained_end(BdrvChild
*child
,
1093 int *drained_end_counter
)
1095 BlockDriverState
*bs
= child
->opaque
;
1096 bdrv_drained_end_no_poll(bs
, drained_end_counter
);
1099 static int bdrv_child_cb_inactivate(BdrvChild
*child
)
1101 BlockDriverState
*bs
= child
->opaque
;
1102 assert(bs
->open_flags
& BDRV_O_INACTIVE
);
1106 static bool bdrv_child_cb_can_set_aio_ctx(BdrvChild
*child
, AioContext
*ctx
,
1107 GSList
**ignore
, Error
**errp
)
1109 BlockDriverState
*bs
= child
->opaque
;
1110 return bdrv_can_set_aio_context(bs
, ctx
, ignore
, errp
);
1113 static void bdrv_child_cb_set_aio_ctx(BdrvChild
*child
, AioContext
*ctx
,
1116 BlockDriverState
*bs
= child
->opaque
;
1117 return bdrv_set_aio_context_ignore(bs
, ctx
, ignore
);
1121 * Returns the options and flags that a temporary snapshot should get, based on
1122 * the originally requested flags (the originally requested image will have
1123 * flags like a backing file)
1125 static void bdrv_temp_snapshot_options(int *child_flags
, QDict
*child_options
,
1126 int parent_flags
, QDict
*parent_options
)
1128 *child_flags
= (parent_flags
& ~BDRV_O_SNAPSHOT
) | BDRV_O_TEMPORARY
;
1130 /* For temporary files, unconditional cache=unsafe is fine */
1131 qdict_set_default_str(child_options
, BDRV_OPT_CACHE_DIRECT
, "off");
1132 qdict_set_default_str(child_options
, BDRV_OPT_CACHE_NO_FLUSH
, "on");
1134 /* Copy the read-only and discard options from the parent */
1135 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_READ_ONLY
);
1136 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_DISCARD
);
1138 /* aio=native doesn't work for cache.direct=off, so disable it for the
1139 * temporary snapshot */
1140 *child_flags
&= ~BDRV_O_NATIVE_AIO
;
1143 static void bdrv_backing_attach(BdrvChild
*c
)
1145 BlockDriverState
*parent
= c
->opaque
;
1146 BlockDriverState
*backing_hd
= c
->bs
;
1148 assert(!parent
->backing_blocker
);
1149 error_setg(&parent
->backing_blocker
,
1150 "node is used as backing hd of '%s'",
1151 bdrv_get_device_or_node_name(parent
));
1153 bdrv_refresh_filename(backing_hd
);
1155 parent
->open_flags
&= ~BDRV_O_NO_BACKING
;
1156 pstrcpy(parent
->backing_file
, sizeof(parent
->backing_file
),
1157 backing_hd
->filename
);
1158 pstrcpy(parent
->backing_format
, sizeof(parent
->backing_format
),
1159 backing_hd
->drv
? backing_hd
->drv
->format_name
: "");
1161 bdrv_op_block_all(backing_hd
, parent
->backing_blocker
);
1162 /* Otherwise we won't be able to commit or stream */
1163 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_COMMIT_TARGET
,
1164 parent
->backing_blocker
);
1165 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_STREAM
,
1166 parent
->backing_blocker
);
1168 * We do backup in 3 ways:
1170 * The target bs is new opened, and the source is top BDS
1171 * 2. blockdev backup
1172 * Both the source and the target are top BDSes.
1173 * 3. internal backup(used for block replication)
1174 * Both the source and the target are backing file
1176 * In case 1 and 2, neither the source nor the target is the backing file.
1177 * In case 3, we will block the top BDS, so there is only one block job
1178 * for the top BDS and its backing chain.
1180 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_BACKUP_SOURCE
,
1181 parent
->backing_blocker
);
1182 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_BACKUP_TARGET
,
1183 parent
->backing_blocker
);
1186 static void bdrv_backing_detach(BdrvChild
*c
)
1188 BlockDriverState
*parent
= c
->opaque
;
1190 assert(parent
->backing_blocker
);
1191 bdrv_op_unblock_all(c
->bs
, parent
->backing_blocker
);
1192 error_free(parent
->backing_blocker
);
1193 parent
->backing_blocker
= NULL
;
1196 static int bdrv_backing_update_filename(BdrvChild
*c
, BlockDriverState
*base
,
1197 const char *filename
, Error
**errp
)
1199 BlockDriverState
*parent
= c
->opaque
;
1200 bool read_only
= bdrv_is_read_only(parent
);
1204 ret
= bdrv_reopen_set_read_only(parent
, false, errp
);
1210 ret
= bdrv_change_backing_file(parent
, filename
,
1211 base
->drv
? base
->drv
->format_name
: "",
1214 error_setg_errno(errp
, -ret
, "Could not update backing file link");
1218 bdrv_reopen_set_read_only(parent
, true, NULL
);
1225 * Returns the options and flags that a generic child of a BDS should
1226 * get, based on the given options and flags for the parent BDS.
1228 static void bdrv_inherited_options(BdrvChildRole role
, bool parent_is_format
,
1229 int *child_flags
, QDict
*child_options
,
1230 int parent_flags
, QDict
*parent_options
)
1232 int flags
= parent_flags
;
1235 * First, decide whether to set, clear, or leave BDRV_O_PROTOCOL.
1236 * Generally, the question to answer is: Should this child be
1237 * format-probed by default?
1241 * Pure and non-filtered data children of non-format nodes should
1242 * be probed by default (even when the node itself has BDRV_O_PROTOCOL
1243 * set). This only affects a very limited set of drivers (namely
1244 * quorum and blkverify when this comment was written).
1245 * Force-clear BDRV_O_PROTOCOL then.
1247 if (!parent_is_format
&&
1248 (role
& BDRV_CHILD_DATA
) &&
1249 !(role
& (BDRV_CHILD_METADATA
| BDRV_CHILD_FILTERED
)))
1251 flags
&= ~BDRV_O_PROTOCOL
;
1255 * All children of format nodes (except for COW children) and all
1256 * metadata children in general should never be format-probed.
1257 * Force-set BDRV_O_PROTOCOL then.
1259 if ((parent_is_format
&& !(role
& BDRV_CHILD_COW
)) ||
1260 (role
& BDRV_CHILD_METADATA
))
1262 flags
|= BDRV_O_PROTOCOL
;
1266 * If the cache mode isn't explicitly set, inherit direct and no-flush from
1269 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_CACHE_DIRECT
);
1270 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_CACHE_NO_FLUSH
);
1271 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_FORCE_SHARE
);
1273 if (role
& BDRV_CHILD_COW
) {
1274 /* backing files are opened read-only by default */
1275 qdict_set_default_str(child_options
, BDRV_OPT_READ_ONLY
, "on");
1276 qdict_set_default_str(child_options
, BDRV_OPT_AUTO_READ_ONLY
, "off");
1278 /* Inherit the read-only option from the parent if it's not set */
1279 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_READ_ONLY
);
1280 qdict_copy_default(child_options
, parent_options
,
1281 BDRV_OPT_AUTO_READ_ONLY
);
1285 * bdrv_co_pdiscard() respects unmap policy for the parent, so we
1286 * can default to enable it on lower layers regardless of the
1289 qdict_set_default_str(child_options
, BDRV_OPT_DISCARD
, "unmap");
1291 /* Clear flags that only apply to the top layer */
1292 flags
&= ~(BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
| BDRV_O_COPY_ON_READ
);
1294 if (role
& BDRV_CHILD_METADATA
) {
1295 flags
&= ~BDRV_O_NO_IO
;
1297 if (role
& BDRV_CHILD_COW
) {
1298 flags
&= ~BDRV_O_TEMPORARY
;
1301 *child_flags
= flags
;
1304 static void bdrv_child_cb_attach(BdrvChild
*child
)
1306 BlockDriverState
*bs
= child
->opaque
;
1308 if (child
->role
& BDRV_CHILD_COW
) {
1309 bdrv_backing_attach(child
);
1312 bdrv_apply_subtree_drain(child
, bs
);
1315 static void bdrv_child_cb_detach(BdrvChild
*child
)
1317 BlockDriverState
*bs
= child
->opaque
;
1319 if (child
->role
& BDRV_CHILD_COW
) {
1320 bdrv_backing_detach(child
);
1323 bdrv_unapply_subtree_drain(child
, bs
);
1326 static int bdrv_child_cb_update_filename(BdrvChild
*c
, BlockDriverState
*base
,
1327 const char *filename
, Error
**errp
)
1329 if (c
->role
& BDRV_CHILD_COW
) {
1330 return bdrv_backing_update_filename(c
, base
, filename
, errp
);
1335 const BdrvChildClass child_of_bds
= {
1336 .parent_is_bds
= true,
1337 .get_parent_desc
= bdrv_child_get_parent_desc
,
1338 .inherit_options
= bdrv_inherited_options
,
1339 .drained_begin
= bdrv_child_cb_drained_begin
,
1340 .drained_poll
= bdrv_child_cb_drained_poll
,
1341 .drained_end
= bdrv_child_cb_drained_end
,
1342 .attach
= bdrv_child_cb_attach
,
1343 .detach
= bdrv_child_cb_detach
,
1344 .inactivate
= bdrv_child_cb_inactivate
,
1345 .can_set_aio_ctx
= bdrv_child_cb_can_set_aio_ctx
,
1346 .set_aio_ctx
= bdrv_child_cb_set_aio_ctx
,
1347 .update_filename
= bdrv_child_cb_update_filename
,
1350 static int bdrv_open_flags(BlockDriverState
*bs
, int flags
)
1352 int open_flags
= flags
;
1355 * Clear flags that are internal to the block layer before opening the
1358 open_flags
&= ~(BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
| BDRV_O_PROTOCOL
);
1363 static void update_flags_from_options(int *flags
, QemuOpts
*opts
)
1365 *flags
&= ~(BDRV_O_CACHE_MASK
| BDRV_O_RDWR
| BDRV_O_AUTO_RDONLY
);
1367 if (qemu_opt_get_bool_del(opts
, BDRV_OPT_CACHE_NO_FLUSH
, false)) {
1368 *flags
|= BDRV_O_NO_FLUSH
;
1371 if (qemu_opt_get_bool_del(opts
, BDRV_OPT_CACHE_DIRECT
, false)) {
1372 *flags
|= BDRV_O_NOCACHE
;
1375 if (!qemu_opt_get_bool_del(opts
, BDRV_OPT_READ_ONLY
, false)) {
1376 *flags
|= BDRV_O_RDWR
;
1379 if (qemu_opt_get_bool_del(opts
, BDRV_OPT_AUTO_READ_ONLY
, false)) {
1380 *flags
|= BDRV_O_AUTO_RDONLY
;
1384 static void update_options_from_flags(QDict
*options
, int flags
)
1386 if (!qdict_haskey(options
, BDRV_OPT_CACHE_DIRECT
)) {
1387 qdict_put_bool(options
, BDRV_OPT_CACHE_DIRECT
, flags
& BDRV_O_NOCACHE
);
1389 if (!qdict_haskey(options
, BDRV_OPT_CACHE_NO_FLUSH
)) {
1390 qdict_put_bool(options
, BDRV_OPT_CACHE_NO_FLUSH
,
1391 flags
& BDRV_O_NO_FLUSH
);
1393 if (!qdict_haskey(options
, BDRV_OPT_READ_ONLY
)) {
1394 qdict_put_bool(options
, BDRV_OPT_READ_ONLY
, !(flags
& BDRV_O_RDWR
));
1396 if (!qdict_haskey(options
, BDRV_OPT_AUTO_READ_ONLY
)) {
1397 qdict_put_bool(options
, BDRV_OPT_AUTO_READ_ONLY
,
1398 flags
& BDRV_O_AUTO_RDONLY
);
1402 static void bdrv_assign_node_name(BlockDriverState
*bs
,
1403 const char *node_name
,
1406 char *gen_node_name
= NULL
;
1409 node_name
= gen_node_name
= id_generate(ID_BLOCK
);
1410 } else if (!id_wellformed(node_name
)) {
1412 * Check for empty string or invalid characters, but not if it is
1413 * generated (generated names use characters not available to the user)
1415 error_setg(errp
, "Invalid node name");
1419 /* takes care of avoiding namespaces collisions */
1420 if (blk_by_name(node_name
)) {
1421 error_setg(errp
, "node-name=%s is conflicting with a device id",
1426 /* takes care of avoiding duplicates node names */
1427 if (bdrv_find_node(node_name
)) {
1428 error_setg(errp
, "Duplicate node name");
1432 /* Make sure that the node name isn't truncated */
1433 if (strlen(node_name
) >= sizeof(bs
->node_name
)) {
1434 error_setg(errp
, "Node name too long");
1438 /* copy node name into the bs and insert it into the graph list */
1439 pstrcpy(bs
->node_name
, sizeof(bs
->node_name
), node_name
);
1440 QTAILQ_INSERT_TAIL(&graph_bdrv_states
, bs
, node_list
);
1442 g_free(gen_node_name
);
1445 static int bdrv_open_driver(BlockDriverState
*bs
, BlockDriver
*drv
,
1446 const char *node_name
, QDict
*options
,
1447 int open_flags
, Error
**errp
)
1449 Error
*local_err
= NULL
;
1452 bdrv_assign_node_name(bs
, node_name
, &local_err
);
1454 error_propagate(errp
, local_err
);
1459 bs
->read_only
= !(bs
->open_flags
& BDRV_O_RDWR
);
1460 bs
->opaque
= g_malloc0(drv
->instance_size
);
1462 if (drv
->bdrv_file_open
) {
1463 assert(!drv
->bdrv_needs_filename
|| bs
->filename
[0]);
1464 ret
= drv
->bdrv_file_open(bs
, options
, open_flags
, &local_err
);
1465 } else if (drv
->bdrv_open
) {
1466 ret
= drv
->bdrv_open(bs
, options
, open_flags
, &local_err
);
1473 error_propagate(errp
, local_err
);
1474 } else if (bs
->filename
[0]) {
1475 error_setg_errno(errp
, -ret
, "Could not open '%s'", bs
->filename
);
1477 error_setg_errno(errp
, -ret
, "Could not open image");
1482 ret
= refresh_total_sectors(bs
, bs
->total_sectors
);
1484 error_setg_errno(errp
, -ret
, "Could not refresh total sector count");
1488 bdrv_refresh_limits(bs
, &local_err
);
1490 error_propagate(errp
, local_err
);
1494 assert(bdrv_opt_mem_align(bs
) != 0);
1495 assert(bdrv_min_mem_align(bs
) != 0);
1496 assert(is_power_of_2(bs
->bl
.request_alignment
));
1498 for (i
= 0; i
< bs
->quiesce_counter
; i
++) {
1499 if (drv
->bdrv_co_drain_begin
) {
1500 drv
->bdrv_co_drain_begin(bs
);
1507 if (bs
->file
!= NULL
) {
1508 bdrv_unref_child(bs
, bs
->file
);
1516 BlockDriverState
*bdrv_new_open_driver(BlockDriver
*drv
, const char *node_name
,
1517 int flags
, Error
**errp
)
1519 BlockDriverState
*bs
;
1523 bs
->open_flags
= flags
;
1524 bs
->explicit_options
= qdict_new();
1525 bs
->options
= qdict_new();
1528 update_options_from_flags(bs
->options
, flags
);
1530 ret
= bdrv_open_driver(bs
, drv
, node_name
, bs
->options
, flags
, errp
);
1532 qobject_unref(bs
->explicit_options
);
1533 bs
->explicit_options
= NULL
;
1534 qobject_unref(bs
->options
);
1543 QemuOptsList bdrv_runtime_opts
= {
1544 .name
= "bdrv_common",
1545 .head
= QTAILQ_HEAD_INITIALIZER(bdrv_runtime_opts
.head
),
1548 .name
= "node-name",
1549 .type
= QEMU_OPT_STRING
,
1550 .help
= "Node name of the block device node",
1554 .type
= QEMU_OPT_STRING
,
1555 .help
= "Block driver to use for the node",
1558 .name
= BDRV_OPT_CACHE_DIRECT
,
1559 .type
= QEMU_OPT_BOOL
,
1560 .help
= "Bypass software writeback cache on the host",
1563 .name
= BDRV_OPT_CACHE_NO_FLUSH
,
1564 .type
= QEMU_OPT_BOOL
,
1565 .help
= "Ignore flush requests",
1568 .name
= BDRV_OPT_READ_ONLY
,
1569 .type
= QEMU_OPT_BOOL
,
1570 .help
= "Node is opened in read-only mode",
1573 .name
= BDRV_OPT_AUTO_READ_ONLY
,
1574 .type
= QEMU_OPT_BOOL
,
1575 .help
= "Node can become read-only if opening read-write fails",
1578 .name
= "detect-zeroes",
1579 .type
= QEMU_OPT_STRING
,
1580 .help
= "try to optimize zero writes (off, on, unmap)",
1583 .name
= BDRV_OPT_DISCARD
,
1584 .type
= QEMU_OPT_STRING
,
1585 .help
= "discard operation (ignore/off, unmap/on)",
1588 .name
= BDRV_OPT_FORCE_SHARE
,
1589 .type
= QEMU_OPT_BOOL
,
1590 .help
= "always accept other writers (default: off)",
1592 { /* end of list */ }
1596 QemuOptsList bdrv_create_opts_simple
= {
1597 .name
= "simple-create-opts",
1598 .head
= QTAILQ_HEAD_INITIALIZER(bdrv_create_opts_simple
.head
),
1601 .name
= BLOCK_OPT_SIZE
,
1602 .type
= QEMU_OPT_SIZE
,
1603 .help
= "Virtual disk size"
1606 .name
= BLOCK_OPT_PREALLOC
,
1607 .type
= QEMU_OPT_STRING
,
1608 .help
= "Preallocation mode (allowed values: off)"
1610 { /* end of list */ }
1615 * Common part for opening disk images and files
1617 * Removes all processed options from *options.
1619 static int bdrv_open_common(BlockDriverState
*bs
, BlockBackend
*file
,
1620 QDict
*options
, Error
**errp
)
1622 int ret
, open_flags
;
1623 const char *filename
;
1624 const char *driver_name
= NULL
;
1625 const char *node_name
= NULL
;
1626 const char *discard
;
1629 Error
*local_err
= NULL
;
1631 assert(bs
->file
== NULL
);
1632 assert(options
!= NULL
&& bs
->options
!= options
);
1634 opts
= qemu_opts_create(&bdrv_runtime_opts
, NULL
, 0, &error_abort
);
1635 if (!qemu_opts_absorb_qdict(opts
, options
, errp
)) {
1640 update_flags_from_options(&bs
->open_flags
, opts
);
1642 driver_name
= qemu_opt_get(opts
, "driver");
1643 drv
= bdrv_find_format(driver_name
);
1644 assert(drv
!= NULL
);
1646 bs
->force_share
= qemu_opt_get_bool(opts
, BDRV_OPT_FORCE_SHARE
, false);
1648 if (bs
->force_share
&& (bs
->open_flags
& BDRV_O_RDWR
)) {
1650 BDRV_OPT_FORCE_SHARE
1651 "=on can only be used with read-only images");
1657 bdrv_refresh_filename(blk_bs(file
));
1658 filename
= blk_bs(file
)->filename
;
1661 * Caution: while qdict_get_try_str() is fine, getting
1662 * non-string types would require more care. When @options
1663 * come from -blockdev or blockdev_add, its members are typed
1664 * according to the QAPI schema, but when they come from
1665 * -drive, they're all QString.
1667 filename
= qdict_get_try_str(options
, "filename");
1670 if (drv
->bdrv_needs_filename
&& (!filename
|| !filename
[0])) {
1671 error_setg(errp
, "The '%s' block driver requires a file name",
1677 trace_bdrv_open_common(bs
, filename
?: "", bs
->open_flags
,
1680 bs
->read_only
= !(bs
->open_flags
& BDRV_O_RDWR
);
1682 if (use_bdrv_whitelist
&& !bdrv_is_whitelisted(drv
, bs
->read_only
)) {
1683 if (!bs
->read_only
&& bdrv_is_whitelisted(drv
, true)) {
1684 ret
= bdrv_apply_auto_read_only(bs
, NULL
, NULL
);
1690 !bs
->read_only
&& bdrv_is_whitelisted(drv
, true)
1691 ? "Driver '%s' can only be used for read-only devices"
1692 : "Driver '%s' is not whitelisted",
1698 /* bdrv_new() and bdrv_close() make it so */
1699 assert(atomic_read(&bs
->copy_on_read
) == 0);
1701 if (bs
->open_flags
& BDRV_O_COPY_ON_READ
) {
1702 if (!bs
->read_only
) {
1703 bdrv_enable_copy_on_read(bs
);
1705 error_setg(errp
, "Can't use copy-on-read on read-only device");
1711 discard
= qemu_opt_get(opts
, BDRV_OPT_DISCARD
);
1712 if (discard
!= NULL
) {
1713 if (bdrv_parse_discard_flags(discard
, &bs
->open_flags
) != 0) {
1714 error_setg(errp
, "Invalid discard option");
1721 bdrv_parse_detect_zeroes(opts
, bs
->open_flags
, &local_err
);
1723 error_propagate(errp
, local_err
);
1728 if (filename
!= NULL
) {
1729 pstrcpy(bs
->filename
, sizeof(bs
->filename
), filename
);
1731 bs
->filename
[0] = '\0';
1733 pstrcpy(bs
->exact_filename
, sizeof(bs
->exact_filename
), bs
->filename
);
1735 /* Open the image, either directly or using a protocol */
1736 open_flags
= bdrv_open_flags(bs
, bs
->open_flags
);
1737 node_name
= qemu_opt_get(opts
, "node-name");
1739 assert(!drv
->bdrv_file_open
|| file
== NULL
);
1740 ret
= bdrv_open_driver(bs
, drv
, node_name
, options
, open_flags
, errp
);
1745 qemu_opts_del(opts
);
1749 qemu_opts_del(opts
);
1753 static QDict
*parse_json_filename(const char *filename
, Error
**errp
)
1755 QObject
*options_obj
;
1759 ret
= strstart(filename
, "json:", &filename
);
1762 options_obj
= qobject_from_json(filename
, errp
);
1764 error_prepend(errp
, "Could not parse the JSON options: ");
1768 options
= qobject_to(QDict
, options_obj
);
1770 qobject_unref(options_obj
);
1771 error_setg(errp
, "Invalid JSON object given");
1775 qdict_flatten(options
);
1780 static void parse_json_protocol(QDict
*options
, const char **pfilename
,
1783 QDict
*json_options
;
1784 Error
*local_err
= NULL
;
1786 /* Parse json: pseudo-protocol */
1787 if (!*pfilename
|| !g_str_has_prefix(*pfilename
, "json:")) {
1791 json_options
= parse_json_filename(*pfilename
, &local_err
);
1793 error_propagate(errp
, local_err
);
1797 /* Options given in the filename have lower priority than options
1798 * specified directly */
1799 qdict_join(options
, json_options
, false);
1800 qobject_unref(json_options
);
1805 * Fills in default options for opening images and converts the legacy
1806 * filename/flags pair to option QDict entries.
1807 * The BDRV_O_PROTOCOL flag in *flags will be set or cleared accordingly if a
1808 * block driver has been specified explicitly.
1810 static int bdrv_fill_options(QDict
**options
, const char *filename
,
1811 int *flags
, Error
**errp
)
1813 const char *drvname
;
1814 bool protocol
= *flags
& BDRV_O_PROTOCOL
;
1815 bool parse_filename
= false;
1816 BlockDriver
*drv
= NULL
;
1817 Error
*local_err
= NULL
;
1820 * Caution: while qdict_get_try_str() is fine, getting non-string
1821 * types would require more care. When @options come from
1822 * -blockdev or blockdev_add, its members are typed according to
1823 * the QAPI schema, but when they come from -drive, they're all
1826 drvname
= qdict_get_try_str(*options
, "driver");
1828 drv
= bdrv_find_format(drvname
);
1830 error_setg(errp
, "Unknown driver '%s'", drvname
);
1833 /* If the user has explicitly specified the driver, this choice should
1834 * override the BDRV_O_PROTOCOL flag */
1835 protocol
= drv
->bdrv_file_open
;
1839 *flags
|= BDRV_O_PROTOCOL
;
1841 *flags
&= ~BDRV_O_PROTOCOL
;
1844 /* Translate cache options from flags into options */
1845 update_options_from_flags(*options
, *flags
);
1847 /* Fetch the file name from the options QDict if necessary */
1848 if (protocol
&& filename
) {
1849 if (!qdict_haskey(*options
, "filename")) {
1850 qdict_put_str(*options
, "filename", filename
);
1851 parse_filename
= true;
1853 error_setg(errp
, "Can't specify 'file' and 'filename' options at "
1859 /* Find the right block driver */
1860 /* See cautionary note on accessing @options above */
1861 filename
= qdict_get_try_str(*options
, "filename");
1863 if (!drvname
&& protocol
) {
1865 drv
= bdrv_find_protocol(filename
, parse_filename
, errp
);
1870 drvname
= drv
->format_name
;
1871 qdict_put_str(*options
, "driver", drvname
);
1873 error_setg(errp
, "Must specify either driver or file");
1878 assert(drv
|| !protocol
);
1880 /* Driver-specific filename parsing */
1881 if (drv
&& drv
->bdrv_parse_filename
&& parse_filename
) {
1882 drv
->bdrv_parse_filename(filename
, *options
, &local_err
);
1884 error_propagate(errp
, local_err
);
1888 if (!drv
->bdrv_needs_filename
) {
1889 qdict_del(*options
, "filename");
1896 static int bdrv_child_check_perm(BdrvChild
*c
, BlockReopenQueue
*q
,
1897 uint64_t perm
, uint64_t shared
,
1898 GSList
*ignore_children
,
1899 bool *tighten_restrictions
, Error
**errp
);
1900 static void bdrv_child_abort_perm_update(BdrvChild
*c
);
1901 static void bdrv_child_set_perm(BdrvChild
*c
, uint64_t perm
, uint64_t shared
);
1903 typedef struct BlockReopenQueueEntry
{
1906 BDRVReopenState state
;
1907 QTAILQ_ENTRY(BlockReopenQueueEntry
) entry
;
1908 } BlockReopenQueueEntry
;
1911 * Return the flags that @bs will have after the reopens in @q have
1912 * successfully completed. If @q is NULL (or @bs is not contained in @q),
1913 * return the current flags.
1915 static int bdrv_reopen_get_flags(BlockReopenQueue
*q
, BlockDriverState
*bs
)
1917 BlockReopenQueueEntry
*entry
;
1920 QTAILQ_FOREACH(entry
, q
, entry
) {
1921 if (entry
->state
.bs
== bs
) {
1922 return entry
->state
.flags
;
1927 return bs
->open_flags
;
1930 /* Returns whether the image file can be written to after the reopen queue @q
1931 * has been successfully applied, or right now if @q is NULL. */
1932 static bool bdrv_is_writable_after_reopen(BlockDriverState
*bs
,
1933 BlockReopenQueue
*q
)
1935 int flags
= bdrv_reopen_get_flags(q
, bs
);
1937 return (flags
& (BDRV_O_RDWR
| BDRV_O_INACTIVE
)) == BDRV_O_RDWR
;
1941 * Return whether the BDS can be written to. This is not necessarily
1942 * the same as !bdrv_is_read_only(bs), as inactivated images may not
1943 * be written to but do not count as read-only images.
1945 bool bdrv_is_writable(BlockDriverState
*bs
)
1947 return bdrv_is_writable_after_reopen(bs
, NULL
);
1950 static void bdrv_child_perm(BlockDriverState
*bs
, BlockDriverState
*child_bs
,
1951 BdrvChild
*c
, BdrvChildRole role
,
1952 BlockReopenQueue
*reopen_queue
,
1953 uint64_t parent_perm
, uint64_t parent_shared
,
1954 uint64_t *nperm
, uint64_t *nshared
)
1956 assert(bs
->drv
&& bs
->drv
->bdrv_child_perm
);
1957 bs
->drv
->bdrv_child_perm(bs
, c
, role
, reopen_queue
,
1958 parent_perm
, parent_shared
,
1960 /* TODO Take force_share from reopen_queue */
1961 if (child_bs
&& child_bs
->force_share
) {
1962 *nshared
= BLK_PERM_ALL
;
1967 * Check whether permissions on this node can be changed in a way that
1968 * @cumulative_perms and @cumulative_shared_perms are the new cumulative
1969 * permissions of all its parents. This involves checking whether all necessary
1970 * permission changes to child nodes can be performed.
1972 * Will set *tighten_restrictions to true if and only if new permissions have to
1973 * be taken or currently shared permissions are to be unshared. Otherwise,
1974 * errors are not fatal as long as the caller accepts that the restrictions
1975 * remain tighter than they need to be. The caller still has to abort the
1977 * @tighten_restrictions cannot be used together with @q: When reopening, we may
1978 * encounter fatal errors even though no restrictions are to be tightened. For
1979 * example, changing a node from RW to RO will fail if the WRITE permission is
1982 * A call to this function must always be followed by a call to bdrv_set_perm()
1983 * or bdrv_abort_perm_update().
1985 static int bdrv_check_perm(BlockDriverState
*bs
, BlockReopenQueue
*q
,
1986 uint64_t cumulative_perms
,
1987 uint64_t cumulative_shared_perms
,
1988 GSList
*ignore_children
,
1989 bool *tighten_restrictions
, Error
**errp
)
1991 BlockDriver
*drv
= bs
->drv
;
1995 assert(!q
|| !tighten_restrictions
);
1997 if (tighten_restrictions
) {
1998 uint64_t current_perms
, current_shared
;
1999 uint64_t added_perms
, removed_shared_perms
;
2001 bdrv_get_cumulative_perm(bs
, ¤t_perms
, ¤t_shared
);
2003 added_perms
= cumulative_perms
& ~current_perms
;
2004 removed_shared_perms
= current_shared
& ~cumulative_shared_perms
;
2006 *tighten_restrictions
= added_perms
|| removed_shared_perms
;
2009 /* Write permissions never work with read-only images */
2010 if ((cumulative_perms
& (BLK_PERM_WRITE
| BLK_PERM_WRITE_UNCHANGED
)) &&
2011 !bdrv_is_writable_after_reopen(bs
, q
))
2013 if (!bdrv_is_writable_after_reopen(bs
, NULL
)) {
2014 error_setg(errp
, "Block node is read-only");
2016 uint64_t current_perms
, current_shared
;
2017 bdrv_get_cumulative_perm(bs
, ¤t_perms
, ¤t_shared
);
2018 if (current_perms
& (BLK_PERM_WRITE
| BLK_PERM_WRITE_UNCHANGED
)) {
2019 error_setg(errp
, "Cannot make block node read-only, there is "
2022 error_setg(errp
, "Cannot make block node read-only and create "
2031 * Unaligned requests will automatically be aligned to bl.request_alignment
2032 * and without RESIZE we can't extend requests to write to space beyond the
2033 * end of the image, so it's required that the image size is aligned.
2035 if ((cumulative_perms
& (BLK_PERM_WRITE
| BLK_PERM_WRITE_UNCHANGED
)) &&
2036 !(cumulative_perms
& BLK_PERM_RESIZE
))
2038 if ((bs
->total_sectors
* BDRV_SECTOR_SIZE
) % bs
->bl
.request_alignment
) {
2039 error_setg(errp
, "Cannot get 'write' permission without 'resize': "
2040 "Image size is not a multiple of request "
2046 /* Check this node */
2051 if (drv
->bdrv_check_perm
) {
2052 return drv
->bdrv_check_perm(bs
, cumulative_perms
,
2053 cumulative_shared_perms
, errp
);
2056 /* Drivers that never have children can omit .bdrv_child_perm() */
2057 if (!drv
->bdrv_child_perm
) {
2058 assert(QLIST_EMPTY(&bs
->children
));
2062 /* Check all children */
2063 QLIST_FOREACH(c
, &bs
->children
, next
) {
2064 uint64_t cur_perm
, cur_shared
;
2065 bool child_tighten_restr
;
2067 bdrv_child_perm(bs
, c
->bs
, c
, c
->role
, q
,
2068 cumulative_perms
, cumulative_shared_perms
,
2069 &cur_perm
, &cur_shared
);
2070 ret
= bdrv_child_check_perm(c
, q
, cur_perm
, cur_shared
, ignore_children
,
2071 tighten_restrictions
? &child_tighten_restr
2074 if (tighten_restrictions
) {
2075 *tighten_restrictions
|= child_tighten_restr
;
2086 * Notifies drivers that after a previous bdrv_check_perm() call, the
2087 * permission update is not performed and any preparations made for it (e.g.
2088 * taken file locks) need to be undone.
2090 * This function recursively notifies all child nodes.
2092 static void bdrv_abort_perm_update(BlockDriverState
*bs
)
2094 BlockDriver
*drv
= bs
->drv
;
2101 if (drv
->bdrv_abort_perm_update
) {
2102 drv
->bdrv_abort_perm_update(bs
);
2105 QLIST_FOREACH(c
, &bs
->children
, next
) {
2106 bdrv_child_abort_perm_update(c
);
2110 static void bdrv_set_perm(BlockDriverState
*bs
, uint64_t cumulative_perms
,
2111 uint64_t cumulative_shared_perms
)
2113 BlockDriver
*drv
= bs
->drv
;
2120 /* Update this node */
2121 if (drv
->bdrv_set_perm
) {
2122 drv
->bdrv_set_perm(bs
, cumulative_perms
, cumulative_shared_perms
);
2125 /* Drivers that never have children can omit .bdrv_child_perm() */
2126 if (!drv
->bdrv_child_perm
) {
2127 assert(QLIST_EMPTY(&bs
->children
));
2131 /* Update all children */
2132 QLIST_FOREACH(c
, &bs
->children
, next
) {
2133 uint64_t cur_perm
, cur_shared
;
2134 bdrv_child_perm(bs
, c
->bs
, c
, c
->role
, NULL
,
2135 cumulative_perms
, cumulative_shared_perms
,
2136 &cur_perm
, &cur_shared
);
2137 bdrv_child_set_perm(c
, cur_perm
, cur_shared
);
2141 void bdrv_get_cumulative_perm(BlockDriverState
*bs
, uint64_t *perm
,
2142 uint64_t *shared_perm
)
2145 uint64_t cumulative_perms
= 0;
2146 uint64_t cumulative_shared_perms
= BLK_PERM_ALL
;
2148 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
2149 cumulative_perms
|= c
->perm
;
2150 cumulative_shared_perms
&= c
->shared_perm
;
2153 *perm
= cumulative_perms
;
2154 *shared_perm
= cumulative_shared_perms
;
2157 static char *bdrv_child_user_desc(BdrvChild
*c
)
2159 if (c
->klass
->get_parent_desc
) {
2160 return c
->klass
->get_parent_desc(c
);
2163 return g_strdup("another user");
2166 char *bdrv_perm_names(uint64_t perm
)
2172 { BLK_PERM_CONSISTENT_READ
, "consistent read" },
2173 { BLK_PERM_WRITE
, "write" },
2174 { BLK_PERM_WRITE_UNCHANGED
, "write unchanged" },
2175 { BLK_PERM_RESIZE
, "resize" },
2176 { BLK_PERM_GRAPH_MOD
, "change children" },
2180 GString
*result
= g_string_sized_new(30);
2181 struct perm_name
*p
;
2183 for (p
= permissions
; p
->name
; p
++) {
2184 if (perm
& p
->perm
) {
2185 if (result
->len
> 0) {
2186 g_string_append(result
, ", ");
2188 g_string_append(result
, p
->name
);
2192 return g_string_free(result
, FALSE
);
2196 * Checks whether a new reference to @bs can be added if the new user requires
2197 * @new_used_perm/@new_shared_perm as its permissions. If @ignore_children is
2198 * set, the BdrvChild objects in this list are ignored in the calculations;
2199 * this allows checking permission updates for an existing reference.
2201 * See bdrv_check_perm() for the semantics of @tighten_restrictions.
2203 * Needs to be followed by a call to either bdrv_set_perm() or
2204 * bdrv_abort_perm_update(). */
2205 static int bdrv_check_update_perm(BlockDriverState
*bs
, BlockReopenQueue
*q
,
2206 uint64_t new_used_perm
,
2207 uint64_t new_shared_perm
,
2208 GSList
*ignore_children
,
2209 bool *tighten_restrictions
,
2213 uint64_t cumulative_perms
= new_used_perm
;
2214 uint64_t cumulative_shared_perms
= new_shared_perm
;
2216 assert(!q
|| !tighten_restrictions
);
2218 /* There is no reason why anyone couldn't tolerate write_unchanged */
2219 assert(new_shared_perm
& BLK_PERM_WRITE_UNCHANGED
);
2221 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
2222 if (g_slist_find(ignore_children
, c
)) {
2226 if ((new_used_perm
& c
->shared_perm
) != new_used_perm
) {
2227 char *user
= bdrv_child_user_desc(c
);
2228 char *perm_names
= bdrv_perm_names(new_used_perm
& ~c
->shared_perm
);
2230 if (tighten_restrictions
) {
2231 *tighten_restrictions
= true;
2234 error_setg(errp
, "Conflicts with use by %s as '%s', which does not "
2236 user
, c
->name
, perm_names
, bdrv_get_node_name(c
->bs
));
2242 if ((c
->perm
& new_shared_perm
) != c
->perm
) {
2243 char *user
= bdrv_child_user_desc(c
);
2244 char *perm_names
= bdrv_perm_names(c
->perm
& ~new_shared_perm
);
2246 if (tighten_restrictions
) {
2247 *tighten_restrictions
= true;
2250 error_setg(errp
, "Conflicts with use by %s as '%s', which uses "
2252 user
, c
->name
, perm_names
, bdrv_get_node_name(c
->bs
));
2258 cumulative_perms
|= c
->perm
;
2259 cumulative_shared_perms
&= c
->shared_perm
;
2262 return bdrv_check_perm(bs
, q
, cumulative_perms
, cumulative_shared_perms
,
2263 ignore_children
, tighten_restrictions
, errp
);
2266 /* Needs to be followed by a call to either bdrv_child_set_perm() or
2267 * bdrv_child_abort_perm_update(). */
2268 static int bdrv_child_check_perm(BdrvChild
*c
, BlockReopenQueue
*q
,
2269 uint64_t perm
, uint64_t shared
,
2270 GSList
*ignore_children
,
2271 bool *tighten_restrictions
, Error
**errp
)
2275 ignore_children
= g_slist_prepend(g_slist_copy(ignore_children
), c
);
2276 ret
= bdrv_check_update_perm(c
->bs
, q
, perm
, shared
, ignore_children
,
2277 tighten_restrictions
, errp
);
2278 g_slist_free(ignore_children
);
2284 if (!c
->has_backup_perm
) {
2285 c
->has_backup_perm
= true;
2286 c
->backup_perm
= c
->perm
;
2287 c
->backup_shared_perm
= c
->shared_perm
;
2290 * Note: it's OK if c->has_backup_perm was already set, as we can find the
2291 * same child twice during check_perm procedure
2295 c
->shared_perm
= shared
;
2300 static void bdrv_child_set_perm(BdrvChild
*c
, uint64_t perm
, uint64_t shared
)
2302 uint64_t cumulative_perms
, cumulative_shared_perms
;
2304 c
->has_backup_perm
= false;
2307 c
->shared_perm
= shared
;
2309 bdrv_get_cumulative_perm(c
->bs
, &cumulative_perms
,
2310 &cumulative_shared_perms
);
2311 bdrv_set_perm(c
->bs
, cumulative_perms
, cumulative_shared_perms
);
2314 static void bdrv_child_abort_perm_update(BdrvChild
*c
)
2316 if (c
->has_backup_perm
) {
2317 c
->perm
= c
->backup_perm
;
2318 c
->shared_perm
= c
->backup_shared_perm
;
2319 c
->has_backup_perm
= false;
2322 bdrv_abort_perm_update(c
->bs
);
2325 int bdrv_child_try_set_perm(BdrvChild
*c
, uint64_t perm
, uint64_t shared
,
2328 Error
*local_err
= NULL
;
2330 bool tighten_restrictions
;
2332 ret
= bdrv_child_check_perm(c
, NULL
, perm
, shared
, NULL
,
2333 &tighten_restrictions
, &local_err
);
2335 bdrv_child_abort_perm_update(c
);
2336 if (tighten_restrictions
) {
2337 error_propagate(errp
, local_err
);
2340 * Our caller may intend to only loosen restrictions and
2341 * does not expect this function to fail. Errors are not
2342 * fatal in such a case, so we can just hide them from our
2345 error_free(local_err
);
2351 bdrv_child_set_perm(c
, perm
, shared
);
2356 int bdrv_child_refresh_perms(BlockDriverState
*bs
, BdrvChild
*c
, Error
**errp
)
2358 uint64_t parent_perms
, parent_shared
;
2359 uint64_t perms
, shared
;
2361 bdrv_get_cumulative_perm(bs
, &parent_perms
, &parent_shared
);
2362 bdrv_child_perm(bs
, c
->bs
, c
, c
->role
, NULL
,
2363 parent_perms
, parent_shared
, &perms
, &shared
);
2365 return bdrv_child_try_set_perm(c
, perms
, shared
, errp
);
2369 * Default implementation for .bdrv_child_perm() for block filters:
2370 * Forward CONSISTENT_READ, WRITE, WRITE_UNCHANGED, and RESIZE to the
2373 static void bdrv_filter_default_perms(BlockDriverState
*bs
, BdrvChild
*c
,
2375 BlockReopenQueue
*reopen_queue
,
2376 uint64_t perm
, uint64_t shared
,
2377 uint64_t *nperm
, uint64_t *nshared
)
2379 *nperm
= perm
& DEFAULT_PERM_PASSTHROUGH
;
2380 *nshared
= (shared
& DEFAULT_PERM_PASSTHROUGH
) | DEFAULT_PERM_UNCHANGED
;
2383 static void bdrv_default_perms_for_cow(BlockDriverState
*bs
, BdrvChild
*c
,
2385 BlockReopenQueue
*reopen_queue
,
2386 uint64_t perm
, uint64_t shared
,
2387 uint64_t *nperm
, uint64_t *nshared
)
2389 assert(role
& BDRV_CHILD_COW
);
2392 * We want consistent read from backing files if the parent needs it.
2393 * No other operations are performed on backing files.
2395 perm
&= BLK_PERM_CONSISTENT_READ
;
2398 * If the parent can deal with changing data, we're okay with a
2399 * writable and resizable backing file.
2400 * TODO Require !(perm & BLK_PERM_CONSISTENT_READ), too?
2402 if (shared
& BLK_PERM_WRITE
) {
2403 shared
= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2408 shared
|= BLK_PERM_CONSISTENT_READ
| BLK_PERM_GRAPH_MOD
|
2409 BLK_PERM_WRITE_UNCHANGED
;
2411 if (bs
->open_flags
& BDRV_O_INACTIVE
) {
2412 shared
|= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2419 static void bdrv_default_perms_for_storage(BlockDriverState
*bs
, BdrvChild
*c
,
2421 BlockReopenQueue
*reopen_queue
,
2422 uint64_t perm
, uint64_t shared
,
2423 uint64_t *nperm
, uint64_t *nshared
)
2427 assert(role
& (BDRV_CHILD_METADATA
| BDRV_CHILD_DATA
));
2429 flags
= bdrv_reopen_get_flags(reopen_queue
, bs
);
2432 * Apart from the modifications below, the same permissions are
2433 * forwarded and left alone as for filters
2435 bdrv_filter_default_perms(bs
, c
, role
, reopen_queue
,
2436 perm
, shared
, &perm
, &shared
);
2438 if (role
& BDRV_CHILD_METADATA
) {
2439 /* Format drivers may touch metadata even if the guest doesn't write */
2440 if (bdrv_is_writable_after_reopen(bs
, reopen_queue
)) {
2441 perm
|= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2445 * bs->file always needs to be consistent because of the
2446 * metadata. We can never allow other users to resize or write
2449 if (!(flags
& BDRV_O_NO_IO
)) {
2450 perm
|= BLK_PERM_CONSISTENT_READ
;
2452 shared
&= ~(BLK_PERM_WRITE
| BLK_PERM_RESIZE
);
2455 if (role
& BDRV_CHILD_DATA
) {
2457 * Technically, everything in this block is a subset of the
2458 * BDRV_CHILD_METADATA path taken above, and so this could
2459 * be an "else if" branch. However, that is not obvious, and
2460 * this function is not performance critical, therefore we let
2461 * this be an independent "if".
2465 * We cannot allow other users to resize the file because the
2466 * format driver might have some assumptions about the size
2467 * (e.g. because it is stored in metadata, or because the file
2468 * is split into fixed-size data files).
2470 shared
&= ~BLK_PERM_RESIZE
;
2473 * WRITE_UNCHANGED often cannot be performed as such on the
2474 * data file. For example, the qcow2 driver may still need to
2475 * write copied clusters on copy-on-read.
2477 if (perm
& BLK_PERM_WRITE_UNCHANGED
) {
2478 perm
|= BLK_PERM_WRITE
;
2482 * If the data file is written to, the format driver may
2483 * expect to be able to resize it by writing beyond the EOF.
2485 if (perm
& BLK_PERM_WRITE
) {
2486 perm
|= BLK_PERM_RESIZE
;
2490 if (bs
->open_flags
& BDRV_O_INACTIVE
) {
2491 shared
|= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2498 void bdrv_default_perms(BlockDriverState
*bs
, BdrvChild
*c
,
2499 BdrvChildRole role
, BlockReopenQueue
*reopen_queue
,
2500 uint64_t perm
, uint64_t shared
,
2501 uint64_t *nperm
, uint64_t *nshared
)
2503 if (role
& BDRV_CHILD_FILTERED
) {
2504 assert(!(role
& (BDRV_CHILD_DATA
| BDRV_CHILD_METADATA
|
2506 bdrv_filter_default_perms(bs
, c
, role
, reopen_queue
,
2507 perm
, shared
, nperm
, nshared
);
2508 } else if (role
& BDRV_CHILD_COW
) {
2509 assert(!(role
& (BDRV_CHILD_DATA
| BDRV_CHILD_METADATA
)));
2510 bdrv_default_perms_for_cow(bs
, c
, role
, reopen_queue
,
2511 perm
, shared
, nperm
, nshared
);
2512 } else if (role
& (BDRV_CHILD_METADATA
| BDRV_CHILD_DATA
)) {
2513 bdrv_default_perms_for_storage(bs
, c
, role
, reopen_queue
,
2514 perm
, shared
, nperm
, nshared
);
2516 g_assert_not_reached();
2520 uint64_t bdrv_qapi_perm_to_blk_perm(BlockPermission qapi_perm
)
2522 static const uint64_t permissions
[] = {
2523 [BLOCK_PERMISSION_CONSISTENT_READ
] = BLK_PERM_CONSISTENT_READ
,
2524 [BLOCK_PERMISSION_WRITE
] = BLK_PERM_WRITE
,
2525 [BLOCK_PERMISSION_WRITE_UNCHANGED
] = BLK_PERM_WRITE_UNCHANGED
,
2526 [BLOCK_PERMISSION_RESIZE
] = BLK_PERM_RESIZE
,
2527 [BLOCK_PERMISSION_GRAPH_MOD
] = BLK_PERM_GRAPH_MOD
,
2530 QEMU_BUILD_BUG_ON(ARRAY_SIZE(permissions
) != BLOCK_PERMISSION__MAX
);
2531 QEMU_BUILD_BUG_ON(1UL << ARRAY_SIZE(permissions
) != BLK_PERM_ALL
+ 1);
2533 assert(qapi_perm
< BLOCK_PERMISSION__MAX
);
2535 return permissions
[qapi_perm
];
2538 static void bdrv_replace_child_noperm(BdrvChild
*child
,
2539 BlockDriverState
*new_bs
)
2541 BlockDriverState
*old_bs
= child
->bs
;
2542 int new_bs_quiesce_counter
;
2545 assert(!child
->frozen
);
2547 if (old_bs
&& new_bs
) {
2548 assert(bdrv_get_aio_context(old_bs
) == bdrv_get_aio_context(new_bs
));
2551 new_bs_quiesce_counter
= (new_bs
? new_bs
->quiesce_counter
: 0);
2552 drain_saldo
= new_bs_quiesce_counter
- child
->parent_quiesce_counter
;
2555 * If the new child node is drained but the old one was not, flush
2556 * all outstanding requests to the old child node.
2558 while (drain_saldo
> 0 && child
->klass
->drained_begin
) {
2559 bdrv_parent_drained_begin_single(child
, true);
2564 /* Detach first so that the recursive drain sections coming from @child
2565 * are already gone and we only end the drain sections that came from
2567 if (child
->klass
->detach
) {
2568 child
->klass
->detach(child
);
2570 QLIST_REMOVE(child
, next_parent
);
2576 QLIST_INSERT_HEAD(&new_bs
->parents
, child
, next_parent
);
2579 * Detaching the old node may have led to the new node's
2580 * quiesce_counter having been decreased. Not a problem, we
2581 * just need to recognize this here and then invoke
2582 * drained_end appropriately more often.
2584 assert(new_bs
->quiesce_counter
<= new_bs_quiesce_counter
);
2585 drain_saldo
+= new_bs
->quiesce_counter
- new_bs_quiesce_counter
;
2587 /* Attach only after starting new drained sections, so that recursive
2588 * drain sections coming from @child don't get an extra .drained_begin
2590 if (child
->klass
->attach
) {
2591 child
->klass
->attach(child
);
2596 * If the old child node was drained but the new one is not, allow
2597 * requests to come in only after the new node has been attached.
2599 while (drain_saldo
< 0 && child
->klass
->drained_end
) {
2600 bdrv_parent_drained_end_single(child
);
2606 * Updates @child to change its reference to point to @new_bs, including
2607 * checking and applying the necessary permisson updates both to the old node
2610 * NULL is passed as @new_bs for removing the reference before freeing @child.
2612 * If @new_bs is not NULL, bdrv_check_perm() must be called beforehand, as this
2613 * function uses bdrv_set_perm() to update the permissions according to the new
2614 * reference that @new_bs gets.
2616 * Callers must ensure that child->frozen is false.
2618 static void bdrv_replace_child(BdrvChild
*child
, BlockDriverState
*new_bs
)
2620 BlockDriverState
*old_bs
= child
->bs
;
2621 uint64_t perm
, shared_perm
;
2623 /* Asserts that child->frozen == false */
2624 bdrv_replace_child_noperm(child
, new_bs
);
2627 * Start with the new node's permissions. If @new_bs is a (direct
2628 * or indirect) child of @old_bs, we must complete the permission
2629 * update on @new_bs before we loosen the restrictions on @old_bs.
2630 * Otherwise, bdrv_check_perm() on @old_bs would re-initiate
2631 * updating the permissions of @new_bs, and thus not purely loosen
2635 bdrv_get_cumulative_perm(new_bs
, &perm
, &shared_perm
);
2636 bdrv_set_perm(new_bs
, perm
, shared_perm
);
2640 /* Update permissions for old node. This is guaranteed to succeed
2641 * because we're just taking a parent away, so we're loosening
2643 bool tighten_restrictions
;
2646 bdrv_get_cumulative_perm(old_bs
, &perm
, &shared_perm
);
2647 ret
= bdrv_check_perm(old_bs
, NULL
, perm
, shared_perm
, NULL
,
2648 &tighten_restrictions
, NULL
);
2649 assert(tighten_restrictions
== false);
2651 /* We only tried to loosen restrictions, so errors are not fatal */
2652 bdrv_abort_perm_update(old_bs
);
2654 bdrv_set_perm(old_bs
, perm
, shared_perm
);
2657 /* When the parent requiring a non-default AioContext is removed, the
2658 * node moves back to the main AioContext */
2659 bdrv_try_set_aio_context(old_bs
, qemu_get_aio_context(), NULL
);
2664 * This function steals the reference to child_bs from the caller.
2665 * That reference is later dropped by bdrv_root_unref_child().
2667 * On failure NULL is returned, errp is set and the reference to
2668 * child_bs is also dropped.
2670 * The caller must hold the AioContext lock @child_bs, but not that of @ctx
2671 * (unless @child_bs is already in @ctx).
2673 BdrvChild
*bdrv_root_attach_child(BlockDriverState
*child_bs
,
2674 const char *child_name
,
2675 const BdrvChildClass
*child_class
,
2676 BdrvChildRole child_role
,
2678 uint64_t perm
, uint64_t shared_perm
,
2679 void *opaque
, Error
**errp
)
2682 Error
*local_err
= NULL
;
2685 ret
= bdrv_check_update_perm(child_bs
, NULL
, perm
, shared_perm
, NULL
, NULL
,
2688 bdrv_abort_perm_update(child_bs
);
2689 bdrv_unref(child_bs
);
2693 child
= g_new(BdrvChild
, 1);
2694 *child
= (BdrvChild
) {
2696 .name
= g_strdup(child_name
),
2697 .klass
= child_class
,
2700 .shared_perm
= shared_perm
,
2704 /* If the AioContexts don't match, first try to move the subtree of
2705 * child_bs into the AioContext of the new parent. If this doesn't work,
2706 * try moving the parent into the AioContext of child_bs instead. */
2707 if (bdrv_get_aio_context(child_bs
) != ctx
) {
2708 ret
= bdrv_try_set_aio_context(child_bs
, ctx
, &local_err
);
2709 if (ret
< 0 && child_class
->can_set_aio_ctx
) {
2710 GSList
*ignore
= g_slist_prepend(NULL
, child
);
2711 ctx
= bdrv_get_aio_context(child_bs
);
2712 if (child_class
->can_set_aio_ctx(child
, ctx
, &ignore
, NULL
)) {
2713 error_free(local_err
);
2715 g_slist_free(ignore
);
2716 ignore
= g_slist_prepend(NULL
, child
);
2717 child_class
->set_aio_ctx(child
, ctx
, &ignore
);
2719 g_slist_free(ignore
);
2722 error_propagate(errp
, local_err
);
2724 bdrv_abort_perm_update(child_bs
);
2725 bdrv_unref(child_bs
);
2730 /* This performs the matching bdrv_set_perm() for the above check. */
2731 bdrv_replace_child(child
, child_bs
);
2737 * This function transfers the reference to child_bs from the caller
2738 * to parent_bs. That reference is later dropped by parent_bs on
2739 * bdrv_close() or if someone calls bdrv_unref_child().
2741 * On failure NULL is returned, errp is set and the reference to
2742 * child_bs is also dropped.
2744 * If @parent_bs and @child_bs are in different AioContexts, the caller must
2745 * hold the AioContext lock for @child_bs, but not for @parent_bs.
2747 BdrvChild
*bdrv_attach_child(BlockDriverState
*parent_bs
,
2748 BlockDriverState
*child_bs
,
2749 const char *child_name
,
2750 const BdrvChildClass
*child_class
,
2751 BdrvChildRole child_role
,
2755 uint64_t perm
, shared_perm
;
2757 bdrv_get_cumulative_perm(parent_bs
, &perm
, &shared_perm
);
2759 assert(parent_bs
->drv
);
2760 bdrv_child_perm(parent_bs
, child_bs
, NULL
, child_role
, NULL
,
2761 perm
, shared_perm
, &perm
, &shared_perm
);
2763 child
= bdrv_root_attach_child(child_bs
, child_name
, child_class
,
2764 child_role
, bdrv_get_aio_context(parent_bs
),
2765 perm
, shared_perm
, parent_bs
, errp
);
2766 if (child
== NULL
) {
2770 QLIST_INSERT_HEAD(&parent_bs
->children
, child
, next
);
2774 static void bdrv_detach_child(BdrvChild
*child
)
2776 QLIST_SAFE_REMOVE(child
, next
);
2778 bdrv_replace_child(child
, NULL
);
2780 g_free(child
->name
);
2784 /* Callers must ensure that child->frozen is false. */
2785 void bdrv_root_unref_child(BdrvChild
*child
)
2787 BlockDriverState
*child_bs
;
2789 child_bs
= child
->bs
;
2790 bdrv_detach_child(child
);
2791 bdrv_unref(child_bs
);
2795 * Clear all inherits_from pointers from children and grandchildren of
2796 * @root that point to @root, where necessary.
2798 static void bdrv_unset_inherits_from(BlockDriverState
*root
, BdrvChild
*child
)
2802 if (child
->bs
->inherits_from
== root
) {
2804 * Remove inherits_from only when the last reference between root and
2805 * child->bs goes away.
2807 QLIST_FOREACH(c
, &root
->children
, next
) {
2808 if (c
!= child
&& c
->bs
== child
->bs
) {
2813 child
->bs
->inherits_from
= NULL
;
2817 QLIST_FOREACH(c
, &child
->bs
->children
, next
) {
2818 bdrv_unset_inherits_from(root
, c
);
2822 /* Callers must ensure that child->frozen is false. */
2823 void bdrv_unref_child(BlockDriverState
*parent
, BdrvChild
*child
)
2825 if (child
== NULL
) {
2829 bdrv_unset_inherits_from(parent
, child
);
2830 bdrv_root_unref_child(child
);
2834 static void bdrv_parent_cb_change_media(BlockDriverState
*bs
, bool load
)
2837 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
2838 if (c
->klass
->change_media
) {
2839 c
->klass
->change_media(c
, load
);
2844 /* Return true if you can reach parent going through child->inherits_from
2845 * recursively. If parent or child are NULL, return false */
2846 static bool bdrv_inherits_from_recursive(BlockDriverState
*child
,
2847 BlockDriverState
*parent
)
2849 while (child
&& child
!= parent
) {
2850 child
= child
->inherits_from
;
2853 return child
!= NULL
;
2857 * Return the BdrvChildRole for @bs's backing child. bs->backing is
2858 * mostly used for COW backing children (role = COW), but also for
2859 * filtered children (role = FILTERED | PRIMARY).
2861 static BdrvChildRole
bdrv_backing_role(BlockDriverState
*bs
)
2863 if (bs
->drv
&& bs
->drv
->is_filter
) {
2864 return BDRV_CHILD_FILTERED
| BDRV_CHILD_PRIMARY
;
2866 return BDRV_CHILD_COW
;
2871 * Sets the bs->backing link of a BDS. A new reference is created; callers
2872 * which don't need their own reference any more must call bdrv_unref().
2874 void bdrv_set_backing_hd(BlockDriverState
*bs
, BlockDriverState
*backing_hd
,
2877 bool update_inherits_from
= bdrv_chain_contains(bs
, backing_hd
) &&
2878 bdrv_inherits_from_recursive(backing_hd
, bs
);
2880 if (bdrv_is_backing_chain_frozen(bs
, child_bs(bs
->backing
), errp
)) {
2885 bdrv_ref(backing_hd
);
2889 /* Cannot be frozen, we checked that above */
2890 bdrv_unref_child(bs
, bs
->backing
);
2898 bs
->backing
= bdrv_attach_child(bs
, backing_hd
, "backing", &child_of_bds
,
2899 bdrv_backing_role(bs
), errp
);
2900 /* If backing_hd was already part of bs's backing chain, and
2901 * inherits_from pointed recursively to bs then let's update it to
2902 * point directly to bs (else it will become NULL). */
2903 if (bs
->backing
&& update_inherits_from
) {
2904 backing_hd
->inherits_from
= bs
;
2908 bdrv_refresh_limits(bs
, NULL
);
2912 * Opens the backing file for a BlockDriverState if not yet open
2914 * bdref_key specifies the key for the image's BlockdevRef in the options QDict.
2915 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
2916 * itself, all options starting with "${bdref_key}." are considered part of the
2919 * TODO Can this be unified with bdrv_open_image()?
2921 int bdrv_open_backing_file(BlockDriverState
*bs
, QDict
*parent_options
,
2922 const char *bdref_key
, Error
**errp
)
2924 char *backing_filename
= NULL
;
2925 char *bdref_key_dot
;
2926 const char *reference
= NULL
;
2928 bool implicit_backing
= false;
2929 BlockDriverState
*backing_hd
;
2931 QDict
*tmp_parent_options
= NULL
;
2932 Error
*local_err
= NULL
;
2934 if (bs
->backing
!= NULL
) {
2938 /* NULL means an empty set of options */
2939 if (parent_options
== NULL
) {
2940 tmp_parent_options
= qdict_new();
2941 parent_options
= tmp_parent_options
;
2944 bs
->open_flags
&= ~BDRV_O_NO_BACKING
;
2946 bdref_key_dot
= g_strdup_printf("%s.", bdref_key
);
2947 qdict_extract_subqdict(parent_options
, &options
, bdref_key_dot
);
2948 g_free(bdref_key_dot
);
2951 * Caution: while qdict_get_try_str() is fine, getting non-string
2952 * types would require more care. When @parent_options come from
2953 * -blockdev or blockdev_add, its members are typed according to
2954 * the QAPI schema, but when they come from -drive, they're all
2957 reference
= qdict_get_try_str(parent_options
, bdref_key
);
2958 if (reference
|| qdict_haskey(options
, "file.filename")) {
2959 /* keep backing_filename NULL */
2960 } else if (bs
->backing_file
[0] == '\0' && qdict_size(options
) == 0) {
2961 qobject_unref(options
);
2964 if (qdict_size(options
) == 0) {
2965 /* If the user specifies options that do not modify the
2966 * backing file's behavior, we might still consider it the
2967 * implicit backing file. But it's easier this way, and
2968 * just specifying some of the backing BDS's options is
2969 * only possible with -drive anyway (otherwise the QAPI
2970 * schema forces the user to specify everything). */
2971 implicit_backing
= !strcmp(bs
->auto_backing_file
, bs
->backing_file
);
2974 backing_filename
= bdrv_get_full_backing_filename(bs
, &local_err
);
2977 error_propagate(errp
, local_err
);
2978 qobject_unref(options
);
2983 if (!bs
->drv
|| !bs
->drv
->supports_backing
) {
2985 error_setg(errp
, "Driver doesn't support backing files");
2986 qobject_unref(options
);
2991 bs
->backing_format
[0] != '\0' && !qdict_haskey(options
, "driver")) {
2992 qdict_put_str(options
, "driver", bs
->backing_format
);
2995 backing_hd
= bdrv_open_inherit(backing_filename
, reference
, options
, 0, bs
,
2996 &child_of_bds
, bdrv_backing_role(bs
), errp
);
2998 bs
->open_flags
|= BDRV_O_NO_BACKING
;
2999 error_prepend(errp
, "Could not open backing file: ");
3004 if (implicit_backing
) {
3005 bdrv_refresh_filename(backing_hd
);
3006 pstrcpy(bs
->auto_backing_file
, sizeof(bs
->auto_backing_file
),
3007 backing_hd
->filename
);
3010 /* Hook up the backing file link; drop our reference, bs owns the
3011 * backing_hd reference now */
3012 bdrv_set_backing_hd(bs
, backing_hd
, &local_err
);
3013 bdrv_unref(backing_hd
);
3015 error_propagate(errp
, local_err
);
3020 qdict_del(parent_options
, bdref_key
);
3023 g_free(backing_filename
);
3024 qobject_unref(tmp_parent_options
);
3028 static BlockDriverState
*
3029 bdrv_open_child_bs(const char *filename
, QDict
*options
, const char *bdref_key
,
3030 BlockDriverState
*parent
, const BdrvChildClass
*child_class
,
3031 BdrvChildRole child_role
, bool allow_none
, Error
**errp
)
3033 BlockDriverState
*bs
= NULL
;
3034 QDict
*image_options
;
3035 char *bdref_key_dot
;
3036 const char *reference
;
3038 assert(child_class
!= NULL
);
3040 bdref_key_dot
= g_strdup_printf("%s.", bdref_key
);
3041 qdict_extract_subqdict(options
, &image_options
, bdref_key_dot
);
3042 g_free(bdref_key_dot
);
3045 * Caution: while qdict_get_try_str() is fine, getting non-string
3046 * types would require more care. When @options come from
3047 * -blockdev or blockdev_add, its members are typed according to
3048 * the QAPI schema, but when they come from -drive, they're all
3051 reference
= qdict_get_try_str(options
, bdref_key
);
3052 if (!filename
&& !reference
&& !qdict_size(image_options
)) {
3054 error_setg(errp
, "A block device must be specified for \"%s\"",
3057 qobject_unref(image_options
);
3061 bs
= bdrv_open_inherit(filename
, reference
, image_options
, 0,
3062 parent
, child_class
, child_role
, errp
);
3068 qdict_del(options
, bdref_key
);
3073 * Opens a disk image whose options are given as BlockdevRef in another block
3076 * If allow_none is true, no image will be opened if filename is false and no
3077 * BlockdevRef is given. NULL will be returned, but errp remains unset.
3079 * bdrev_key specifies the key for the image's BlockdevRef in the options QDict.
3080 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
3081 * itself, all options starting with "${bdref_key}." are considered part of the
3084 * The BlockdevRef will be removed from the options QDict.
3086 BdrvChild
*bdrv_open_child(const char *filename
,
3087 QDict
*options
, const char *bdref_key
,
3088 BlockDriverState
*parent
,
3089 const BdrvChildClass
*child_class
,
3090 BdrvChildRole child_role
,
3091 bool allow_none
, Error
**errp
)
3093 BlockDriverState
*bs
;
3095 bs
= bdrv_open_child_bs(filename
, options
, bdref_key
, parent
, child_class
,
3096 child_role
, allow_none
, errp
);
3101 return bdrv_attach_child(parent
, bs
, bdref_key
, child_class
, child_role
,
3106 * TODO Future callers may need to specify parent/child_class in order for
3107 * option inheritance to work. Existing callers use it for the root node.
3109 BlockDriverState
*bdrv_open_blockdev_ref(BlockdevRef
*ref
, Error
**errp
)
3111 BlockDriverState
*bs
= NULL
;
3112 QObject
*obj
= NULL
;
3113 QDict
*qdict
= NULL
;
3114 const char *reference
= NULL
;
3117 if (ref
->type
== QTYPE_QSTRING
) {
3118 reference
= ref
->u
.reference
;
3120 BlockdevOptions
*options
= &ref
->u
.definition
;
3121 assert(ref
->type
== QTYPE_QDICT
);
3123 v
= qobject_output_visitor_new(&obj
);
3124 visit_type_BlockdevOptions(v
, NULL
, &options
, &error_abort
);
3125 visit_complete(v
, &obj
);
3127 qdict
= qobject_to(QDict
, obj
);
3128 qdict_flatten(qdict
);
3130 /* bdrv_open_inherit() defaults to the values in bdrv_flags (for
3131 * compatibility with other callers) rather than what we want as the
3132 * real defaults. Apply the defaults here instead. */
3133 qdict_set_default_str(qdict
, BDRV_OPT_CACHE_DIRECT
, "off");
3134 qdict_set_default_str(qdict
, BDRV_OPT_CACHE_NO_FLUSH
, "off");
3135 qdict_set_default_str(qdict
, BDRV_OPT_READ_ONLY
, "off");
3136 qdict_set_default_str(qdict
, BDRV_OPT_AUTO_READ_ONLY
, "off");
3140 bs
= bdrv_open_inherit(NULL
, reference
, qdict
, 0, NULL
, NULL
, 0, errp
);
3147 static BlockDriverState
*bdrv_append_temp_snapshot(BlockDriverState
*bs
,
3149 QDict
*snapshot_options
,
3152 /* TODO: extra byte is a hack to ensure MAX_PATH space on Windows. */
3153 char *tmp_filename
= g_malloc0(PATH_MAX
+ 1);
3155 QemuOpts
*opts
= NULL
;
3156 BlockDriverState
*bs_snapshot
= NULL
;
3157 Error
*local_err
= NULL
;
3160 /* if snapshot, we create a temporary backing file and open it
3161 instead of opening 'filename' directly */
3163 /* Get the required size from the image */
3164 total_size
= bdrv_getlength(bs
);
3165 if (total_size
< 0) {
3166 error_setg_errno(errp
, -total_size
, "Could not get image size");
3170 /* Create the temporary image */
3171 ret
= get_tmp_filename(tmp_filename
, PATH_MAX
+ 1);
3173 error_setg_errno(errp
, -ret
, "Could not get temporary filename");
3177 opts
= qemu_opts_create(bdrv_qcow2
.create_opts
, NULL
, 0,
3179 qemu_opt_set_number(opts
, BLOCK_OPT_SIZE
, total_size
, &error_abort
);
3180 ret
= bdrv_create(&bdrv_qcow2
, tmp_filename
, opts
, errp
);
3181 qemu_opts_del(opts
);
3183 error_prepend(errp
, "Could not create temporary overlay '%s': ",
3188 /* Prepare options QDict for the temporary file */
3189 qdict_put_str(snapshot_options
, "file.driver", "file");
3190 qdict_put_str(snapshot_options
, "file.filename", tmp_filename
);
3191 qdict_put_str(snapshot_options
, "driver", "qcow2");
3193 bs_snapshot
= bdrv_open(NULL
, NULL
, snapshot_options
, flags
, errp
);
3194 snapshot_options
= NULL
;
3199 /* bdrv_append() consumes a strong reference to bs_snapshot
3200 * (i.e. it will call bdrv_unref() on it) even on error, so in
3201 * order to be able to return one, we have to increase
3202 * bs_snapshot's refcount here */
3203 bdrv_ref(bs_snapshot
);
3204 bdrv_append(bs_snapshot
, bs
, &local_err
);
3206 error_propagate(errp
, local_err
);
3212 qobject_unref(snapshot_options
);
3213 g_free(tmp_filename
);
3218 * Opens a disk image (raw, qcow2, vmdk, ...)
3220 * options is a QDict of options to pass to the block drivers, or NULL for an
3221 * empty set of options. The reference to the QDict belongs to the block layer
3222 * after the call (even on failure), so if the caller intends to reuse the
3223 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
3225 * If *pbs is NULL, a new BDS will be created with a pointer to it stored there.
3226 * If it is not NULL, the referenced BDS will be reused.
3228 * The reference parameter may be used to specify an existing block device which
3229 * should be opened. If specified, neither options nor a filename may be given,
3230 * nor can an existing BDS be reused (that is, *pbs has to be NULL).
3232 static BlockDriverState
*bdrv_open_inherit(const char *filename
,
3233 const char *reference
,
3234 QDict
*options
, int flags
,
3235 BlockDriverState
*parent
,
3236 const BdrvChildClass
*child_class
,
3237 BdrvChildRole child_role
,
3241 BlockBackend
*file
= NULL
;
3242 BlockDriverState
*bs
;
3243 BlockDriver
*drv
= NULL
;
3245 const char *drvname
;
3246 const char *backing
;
3247 Error
*local_err
= NULL
;
3248 QDict
*snapshot_options
= NULL
;
3249 int snapshot_flags
= 0;
3251 assert(!child_class
|| !flags
);
3252 assert(!child_class
== !parent
);
3255 bool options_non_empty
= options
? qdict_size(options
) : false;
3256 qobject_unref(options
);
3258 if (filename
|| options_non_empty
) {
3259 error_setg(errp
, "Cannot reference an existing block device with "
3260 "additional options or a new filename");
3264 bs
= bdrv_lookup_bs(reference
, reference
, errp
);
3275 /* NULL means an empty set of options */
3276 if (options
== NULL
) {
3277 options
= qdict_new();
3280 /* json: syntax counts as explicit options, as if in the QDict */
3281 parse_json_protocol(options
, &filename
, &local_err
);
3286 bs
->explicit_options
= qdict_clone_shallow(options
);
3289 bool parent_is_format
;
3292 parent_is_format
= parent
->drv
->is_format
;
3295 * parent->drv is not set yet because this node is opened for
3296 * (potential) format probing. That means that @parent is going
3297 * to be a format node.
3299 parent_is_format
= true;
3302 bs
->inherits_from
= parent
;
3303 child_class
->inherit_options(child_role
, parent_is_format
,
3305 parent
->open_flags
, parent
->options
);
3308 ret
= bdrv_fill_options(&options
, filename
, &flags
, &local_err
);
3314 * Set the BDRV_O_RDWR and BDRV_O_ALLOW_RDWR flags.
3315 * Caution: getting a boolean member of @options requires care.
3316 * When @options come from -blockdev or blockdev_add, members are
3317 * typed according to the QAPI schema, but when they come from
3318 * -drive, they're all QString.
3320 if (g_strcmp0(qdict_get_try_str(options
, BDRV_OPT_READ_ONLY
), "on") &&
3321 !qdict_get_try_bool(options
, BDRV_OPT_READ_ONLY
, false)) {
3322 flags
|= (BDRV_O_RDWR
| BDRV_O_ALLOW_RDWR
);
3324 flags
&= ~BDRV_O_RDWR
;
3327 if (flags
& BDRV_O_SNAPSHOT
) {
3328 snapshot_options
= qdict_new();
3329 bdrv_temp_snapshot_options(&snapshot_flags
, snapshot_options
,
3331 /* Let bdrv_backing_options() override "read-only" */
3332 qdict_del(options
, BDRV_OPT_READ_ONLY
);
3333 bdrv_inherited_options(BDRV_CHILD_COW
, true,
3334 &flags
, options
, flags
, options
);
3337 bs
->open_flags
= flags
;
3338 bs
->options
= options
;
3339 options
= qdict_clone_shallow(options
);
3341 /* Find the right image format driver */
3342 /* See cautionary note on accessing @options above */
3343 drvname
= qdict_get_try_str(options
, "driver");
3345 drv
= bdrv_find_format(drvname
);
3347 error_setg(errp
, "Unknown driver: '%s'", drvname
);
3352 assert(drvname
|| !(flags
& BDRV_O_PROTOCOL
));
3354 /* See cautionary note on accessing @options above */
3355 backing
= qdict_get_try_str(options
, "backing");
3356 if (qobject_to(QNull
, qdict_get(options
, "backing")) != NULL
||
3357 (backing
&& *backing
== '\0'))
3360 warn_report("Use of \"backing\": \"\" is deprecated; "
3361 "use \"backing\": null instead");
3363 flags
|= BDRV_O_NO_BACKING
;
3364 qdict_del(bs
->explicit_options
, "backing");
3365 qdict_del(bs
->options
, "backing");
3366 qdict_del(options
, "backing");
3369 /* Open image file without format layer. This BlockBackend is only used for
3370 * probing, the block drivers will do their own bdrv_open_child() for the
3371 * same BDS, which is why we put the node name back into options. */
3372 if ((flags
& BDRV_O_PROTOCOL
) == 0) {
3373 BlockDriverState
*file_bs
;
3375 file_bs
= bdrv_open_child_bs(filename
, options
, "file", bs
,
3376 &child_of_bds
, BDRV_CHILD_IMAGE
,
3381 if (file_bs
!= NULL
) {
3382 /* Not requesting BLK_PERM_CONSISTENT_READ because we're only
3383 * looking at the header to guess the image format. This works even
3384 * in cases where a guest would not see a consistent state. */
3385 file
= blk_new(bdrv_get_aio_context(file_bs
), 0, BLK_PERM_ALL
);
3386 blk_insert_bs(file
, file_bs
, &local_err
);
3387 bdrv_unref(file_bs
);
3392 qdict_put_str(options
, "file", bdrv_get_node_name(file_bs
));
3396 /* Image format probing */
3399 ret
= find_image_format(file
, filename
, &drv
, &local_err
);
3404 * This option update would logically belong in bdrv_fill_options(),
3405 * but we first need to open bs->file for the probing to work, while
3406 * opening bs->file already requires the (mostly) final set of options
3407 * so that cache mode etc. can be inherited.
3409 * Adding the driver later is somewhat ugly, but it's not an option
3410 * that would ever be inherited, so it's correct. We just need to make
3411 * sure to update both bs->options (which has the full effective
3412 * options for bs) and options (which has file.* already removed).
3414 qdict_put_str(bs
->options
, "driver", drv
->format_name
);
3415 qdict_put_str(options
, "driver", drv
->format_name
);
3417 error_setg(errp
, "Must specify either driver or file");
3421 /* BDRV_O_PROTOCOL must be set iff a protocol BDS is about to be created */
3422 assert(!!(flags
& BDRV_O_PROTOCOL
) == !!drv
->bdrv_file_open
);
3423 /* file must be NULL if a protocol BDS is about to be created
3424 * (the inverse results in an error message from bdrv_open_common()) */
3425 assert(!(flags
& BDRV_O_PROTOCOL
) || !file
);
3427 /* Open the image */
3428 ret
= bdrv_open_common(bs
, file
, options
, &local_err
);
3438 /* If there is a backing file, use it */
3439 if ((flags
& BDRV_O_NO_BACKING
) == 0) {
3440 ret
= bdrv_open_backing_file(bs
, options
, "backing", &local_err
);
3442 goto close_and_fail
;
3446 /* Remove all children options and references
3447 * from bs->options and bs->explicit_options */
3448 QLIST_FOREACH(child
, &bs
->children
, next
) {
3449 char *child_key_dot
;
3450 child_key_dot
= g_strdup_printf("%s.", child
->name
);
3451 qdict_extract_subqdict(bs
->explicit_options
, NULL
, child_key_dot
);
3452 qdict_extract_subqdict(bs
->options
, NULL
, child_key_dot
);
3453 qdict_del(bs
->explicit_options
, child
->name
);
3454 qdict_del(bs
->options
, child
->name
);
3455 g_free(child_key_dot
);
3458 /* Check if any unknown options were used */
3459 if (qdict_size(options
) != 0) {
3460 const QDictEntry
*entry
= qdict_first(options
);
3461 if (flags
& BDRV_O_PROTOCOL
) {
3462 error_setg(errp
, "Block protocol '%s' doesn't support the option "
3463 "'%s'", drv
->format_name
, entry
->key
);
3466 "Block format '%s' does not support the option '%s'",
3467 drv
->format_name
, entry
->key
);
3470 goto close_and_fail
;
3473 bdrv_parent_cb_change_media(bs
, true);
3475 qobject_unref(options
);
3478 /* For snapshot=on, create a temporary qcow2 overlay. bs points to the
3479 * temporary snapshot afterwards. */
3480 if (snapshot_flags
) {
3481 BlockDriverState
*snapshot_bs
;
3482 snapshot_bs
= bdrv_append_temp_snapshot(bs
, snapshot_flags
,
3483 snapshot_options
, &local_err
);
3484 snapshot_options
= NULL
;
3486 goto close_and_fail
;
3488 /* We are not going to return bs but the overlay on top of it
3489 * (snapshot_bs); thus, we have to drop the strong reference to bs
3490 * (which we obtained by calling bdrv_new()). bs will not be deleted,
3491 * though, because the overlay still has a reference to it. */
3500 qobject_unref(snapshot_options
);
3501 qobject_unref(bs
->explicit_options
);
3502 qobject_unref(bs
->options
);
3503 qobject_unref(options
);
3505 bs
->explicit_options
= NULL
;
3507 error_propagate(errp
, local_err
);
3512 qobject_unref(snapshot_options
);
3513 qobject_unref(options
);
3514 error_propagate(errp
, local_err
);
3518 BlockDriverState
*bdrv_open(const char *filename
, const char *reference
,
3519 QDict
*options
, int flags
, Error
**errp
)
3521 return bdrv_open_inherit(filename
, reference
, options
, flags
, NULL
,
3525 /* Return true if the NULL-terminated @list contains @str */
3526 static bool is_str_in_list(const char *str
, const char *const *list
)
3530 for (i
= 0; list
[i
] != NULL
; i
++) {
3531 if (!strcmp(str
, list
[i
])) {
3540 * Check that every option set in @bs->options is also set in
3543 * Options listed in the common_options list and in
3544 * @bs->drv->mutable_opts are skipped.
3546 * Return 0 on success, otherwise return -EINVAL and set @errp.
3548 static int bdrv_reset_options_allowed(BlockDriverState
*bs
,
3549 const QDict
*new_opts
, Error
**errp
)
3551 const QDictEntry
*e
;
3552 /* These options are common to all block drivers and are handled
3553 * in bdrv_reopen_prepare() so they can be left out of @new_opts */
3554 const char *const common_options
[] = {
3555 "node-name", "discard", "cache.direct", "cache.no-flush",
3556 "read-only", "auto-read-only", "detect-zeroes", NULL
3559 for (e
= qdict_first(bs
->options
); e
; e
= qdict_next(bs
->options
, e
)) {
3560 if (!qdict_haskey(new_opts
, e
->key
) &&
3561 !is_str_in_list(e
->key
, common_options
) &&
3562 !is_str_in_list(e
->key
, bs
->drv
->mutable_opts
)) {
3563 error_setg(errp
, "Option '%s' cannot be reset "
3564 "to its default value", e
->key
);
3573 * Returns true if @child can be reached recursively from @bs
3575 static bool bdrv_recurse_has_child(BlockDriverState
*bs
,
3576 BlockDriverState
*child
)
3584 QLIST_FOREACH(c
, &bs
->children
, next
) {
3585 if (bdrv_recurse_has_child(c
->bs
, child
)) {
3594 * Adds a BlockDriverState to a simple queue for an atomic, transactional
3595 * reopen of multiple devices.
3597 * bs_queue can either be an existing BlockReopenQueue that has had QTAILQ_INIT
3598 * already performed, or alternatively may be NULL a new BlockReopenQueue will
3599 * be created and initialized. This newly created BlockReopenQueue should be
3600 * passed back in for subsequent calls that are intended to be of the same
3603 * bs is the BlockDriverState to add to the reopen queue.
3605 * options contains the changed options for the associated bs
3606 * (the BlockReopenQueue takes ownership)
3608 * flags contains the open flags for the associated bs
3610 * returns a pointer to bs_queue, which is either the newly allocated
3611 * bs_queue, or the existing bs_queue being used.
3613 * bs must be drained between bdrv_reopen_queue() and bdrv_reopen_multiple().
3615 static BlockReopenQueue
*bdrv_reopen_queue_child(BlockReopenQueue
*bs_queue
,
3616 BlockDriverState
*bs
,
3618 const BdrvChildClass
*klass
,
3620 bool parent_is_format
,
3621 QDict
*parent_options
,
3627 BlockReopenQueueEntry
*bs_entry
;
3629 QDict
*old_options
, *explicit_options
, *options_copy
;
3633 /* Make sure that the caller remembered to use a drained section. This is
3634 * important to avoid graph changes between the recursive queuing here and
3635 * bdrv_reopen_multiple(). */
3636 assert(bs
->quiesce_counter
> 0);
3638 if (bs_queue
== NULL
) {
3639 bs_queue
= g_new0(BlockReopenQueue
, 1);
3640 QTAILQ_INIT(bs_queue
);
3644 options
= qdict_new();
3647 /* Check if this BlockDriverState is already in the queue */
3648 QTAILQ_FOREACH(bs_entry
, bs_queue
, entry
) {
3649 if (bs
== bs_entry
->state
.bs
) {
3655 * Precedence of options:
3656 * 1. Explicitly passed in options (highest)
3657 * 2. Retained from explicitly set options of bs
3658 * 3. Inherited from parent node
3659 * 4. Retained from effective options of bs
3662 /* Old explicitly set values (don't overwrite by inherited value) */
3663 if (bs_entry
|| keep_old_opts
) {
3664 old_options
= qdict_clone_shallow(bs_entry
?
3665 bs_entry
->state
.explicit_options
:
3666 bs
->explicit_options
);
3667 bdrv_join_options(bs
, options
, old_options
);
3668 qobject_unref(old_options
);
3671 explicit_options
= qdict_clone_shallow(options
);
3673 /* Inherit from parent node */
3674 if (parent_options
) {
3676 klass
->inherit_options(role
, parent_is_format
, &flags
, options
,
3677 parent_flags
, parent_options
);
3679 flags
= bdrv_get_flags(bs
);
3682 if (keep_old_opts
) {
3683 /* Old values are used for options that aren't set yet */
3684 old_options
= qdict_clone_shallow(bs
->options
);
3685 bdrv_join_options(bs
, options
, old_options
);
3686 qobject_unref(old_options
);
3689 /* We have the final set of options so let's update the flags */
3690 options_copy
= qdict_clone_shallow(options
);
3691 opts
= qemu_opts_create(&bdrv_runtime_opts
, NULL
, 0, &error_abort
);
3692 qemu_opts_absorb_qdict(opts
, options_copy
, NULL
);
3693 update_flags_from_options(&flags
, opts
);
3694 qemu_opts_del(opts
);
3695 qobject_unref(options_copy
);
3697 /* bdrv_open_inherit() sets and clears some additional flags internally */
3698 flags
&= ~BDRV_O_PROTOCOL
;
3699 if (flags
& BDRV_O_RDWR
) {
3700 flags
|= BDRV_O_ALLOW_RDWR
;
3704 bs_entry
= g_new0(BlockReopenQueueEntry
, 1);
3705 QTAILQ_INSERT_TAIL(bs_queue
, bs_entry
, entry
);
3707 qobject_unref(bs_entry
->state
.options
);
3708 qobject_unref(bs_entry
->state
.explicit_options
);
3711 bs_entry
->state
.bs
= bs
;
3712 bs_entry
->state
.options
= options
;
3713 bs_entry
->state
.explicit_options
= explicit_options
;
3714 bs_entry
->state
.flags
= flags
;
3716 /* This needs to be overwritten in bdrv_reopen_prepare() */
3717 bs_entry
->state
.perm
= UINT64_MAX
;
3718 bs_entry
->state
.shared_perm
= 0;
3721 * If keep_old_opts is false then it means that unspecified
3722 * options must be reset to their original value. We don't allow
3723 * resetting 'backing' but we need to know if the option is
3724 * missing in order to decide if we have to return an error.
3726 if (!keep_old_opts
) {
3727 bs_entry
->state
.backing_missing
=
3728 !qdict_haskey(options
, "backing") &&
3729 !qdict_haskey(options
, "backing.driver");
3732 QLIST_FOREACH(child
, &bs
->children
, next
) {
3733 QDict
*new_child_options
= NULL
;
3734 bool child_keep_old
= keep_old_opts
;
3736 /* reopen can only change the options of block devices that were
3737 * implicitly created and inherited options. For other (referenced)
3738 * block devices, a syntax like "backing.foo" results in an error. */
3739 if (child
->bs
->inherits_from
!= bs
) {
3743 /* Check if the options contain a child reference */
3744 if (qdict_haskey(options
, child
->name
)) {
3745 const char *childref
= qdict_get_try_str(options
, child
->name
);
3747 * The current child must not be reopened if the child
3748 * reference is null or points to a different node.
3750 if (g_strcmp0(childref
, child
->bs
->node_name
)) {
3754 * If the child reference points to the current child then
3755 * reopen it with its existing set of options (note that
3756 * it can still inherit new options from the parent).
3758 child_keep_old
= true;
3760 /* Extract child options ("child-name.*") */
3761 char *child_key_dot
= g_strdup_printf("%s.", child
->name
);
3762 qdict_extract_subqdict(explicit_options
, NULL
, child_key_dot
);
3763 qdict_extract_subqdict(options
, &new_child_options
, child_key_dot
);
3764 g_free(child_key_dot
);
3767 bdrv_reopen_queue_child(bs_queue
, child
->bs
, new_child_options
,
3768 child
->klass
, child
->role
, bs
->drv
->is_format
,
3769 options
, flags
, child_keep_old
);
3775 BlockReopenQueue
*bdrv_reopen_queue(BlockReopenQueue
*bs_queue
,
3776 BlockDriverState
*bs
,
3777 QDict
*options
, bool keep_old_opts
)
3779 return bdrv_reopen_queue_child(bs_queue
, bs
, options
, NULL
, 0, false,
3780 NULL
, 0, keep_old_opts
);
3784 * Reopen multiple BlockDriverStates atomically & transactionally.
3786 * The queue passed in (bs_queue) must have been built up previous
3787 * via bdrv_reopen_queue().
3789 * Reopens all BDS specified in the queue, with the appropriate
3790 * flags. All devices are prepared for reopen, and failure of any
3791 * device will cause all device changes to be abandoned, and intermediate
3794 * If all devices prepare successfully, then the changes are committed
3797 * All affected nodes must be drained between bdrv_reopen_queue() and
3798 * bdrv_reopen_multiple().
3800 int bdrv_reopen_multiple(BlockReopenQueue
*bs_queue
, Error
**errp
)
3803 BlockReopenQueueEntry
*bs_entry
, *next
;
3805 assert(bs_queue
!= NULL
);
3807 QTAILQ_FOREACH(bs_entry
, bs_queue
, entry
) {
3808 assert(bs_entry
->state
.bs
->quiesce_counter
> 0);
3809 if (bdrv_reopen_prepare(&bs_entry
->state
, bs_queue
, errp
)) {
3812 bs_entry
->prepared
= true;
3815 QTAILQ_FOREACH(bs_entry
, bs_queue
, entry
) {
3816 BDRVReopenState
*state
= &bs_entry
->state
;
3817 ret
= bdrv_check_perm(state
->bs
, bs_queue
, state
->perm
,
3818 state
->shared_perm
, NULL
, NULL
, errp
);
3822 /* Check if new_backing_bs would accept the new permissions */
3823 if (state
->replace_backing_bs
&& state
->new_backing_bs
) {
3824 uint64_t nperm
, nshared
;
3825 bdrv_child_perm(state
->bs
, state
->new_backing_bs
,
3826 NULL
, bdrv_backing_role(state
->bs
),
3827 bs_queue
, state
->perm
, state
->shared_perm
,
3829 ret
= bdrv_check_update_perm(state
->new_backing_bs
, NULL
,
3830 nperm
, nshared
, NULL
, NULL
, errp
);
3835 bs_entry
->perms_checked
= true;
3839 * If we reach this point, we have success and just need to apply the
3842 * Reverse order is used to comfort qcow2 driver: on commit it need to write
3843 * IN_USE flag to the image, to mark bitmaps in the image as invalid. But
3844 * children are usually goes after parents in reopen-queue, so go from last
3847 QTAILQ_FOREACH_REVERSE(bs_entry
, bs_queue
, entry
) {
3848 bdrv_reopen_commit(&bs_entry
->state
);
3853 QTAILQ_FOREACH_SAFE(bs_entry
, bs_queue
, entry
, next
) {
3854 BDRVReopenState
*state
= &bs_entry
->state
;
3856 if (!bs_entry
->perms_checked
) {
3861 bdrv_set_perm(state
->bs
, state
->perm
, state
->shared_perm
);
3863 bdrv_abort_perm_update(state
->bs
);
3864 if (state
->replace_backing_bs
&& state
->new_backing_bs
) {
3865 bdrv_abort_perm_update(state
->new_backing_bs
);
3871 QTAILQ_FOREACH_REVERSE(bs_entry
, bs_queue
, entry
) {
3872 BlockDriverState
*bs
= bs_entry
->state
.bs
;
3874 if (bs
->drv
->bdrv_reopen_commit_post
)
3875 bs
->drv
->bdrv_reopen_commit_post(&bs_entry
->state
);
3879 QTAILQ_FOREACH_SAFE(bs_entry
, bs_queue
, entry
, next
) {
3881 if (bs_entry
->prepared
) {
3882 bdrv_reopen_abort(&bs_entry
->state
);
3884 qobject_unref(bs_entry
->state
.explicit_options
);
3885 qobject_unref(bs_entry
->state
.options
);
3887 if (bs_entry
->state
.new_backing_bs
) {
3888 bdrv_unref(bs_entry
->state
.new_backing_bs
);
3897 int bdrv_reopen_set_read_only(BlockDriverState
*bs
, bool read_only
,
3901 BlockReopenQueue
*queue
;
3902 QDict
*opts
= qdict_new();
3904 qdict_put_bool(opts
, BDRV_OPT_READ_ONLY
, read_only
);
3906 bdrv_subtree_drained_begin(bs
);
3907 queue
= bdrv_reopen_queue(NULL
, bs
, opts
, true);
3908 ret
= bdrv_reopen_multiple(queue
, errp
);
3909 bdrv_subtree_drained_end(bs
);
3914 static BlockReopenQueueEntry
*find_parent_in_reopen_queue(BlockReopenQueue
*q
,
3917 BlockReopenQueueEntry
*entry
;
3919 QTAILQ_FOREACH(entry
, q
, entry
) {
3920 BlockDriverState
*bs
= entry
->state
.bs
;
3923 QLIST_FOREACH(child
, &bs
->children
, next
) {
3933 static void bdrv_reopen_perm(BlockReopenQueue
*q
, BlockDriverState
*bs
,
3934 uint64_t *perm
, uint64_t *shared
)
3937 BlockReopenQueueEntry
*parent
;
3938 uint64_t cumulative_perms
= 0;
3939 uint64_t cumulative_shared_perms
= BLK_PERM_ALL
;
3941 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
3942 parent
= find_parent_in_reopen_queue(q
, c
);
3944 cumulative_perms
|= c
->perm
;
3945 cumulative_shared_perms
&= c
->shared_perm
;
3947 uint64_t nperm
, nshared
;
3949 bdrv_child_perm(parent
->state
.bs
, bs
, c
, c
->role
, q
,
3950 parent
->state
.perm
, parent
->state
.shared_perm
,
3953 cumulative_perms
|= nperm
;
3954 cumulative_shared_perms
&= nshared
;
3957 *perm
= cumulative_perms
;
3958 *shared
= cumulative_shared_perms
;
3961 static bool bdrv_reopen_can_attach(BlockDriverState
*parent
,
3963 BlockDriverState
*new_child
,
3966 AioContext
*parent_ctx
= bdrv_get_aio_context(parent
);
3967 AioContext
*child_ctx
= bdrv_get_aio_context(new_child
);
3971 ignore
= g_slist_prepend(NULL
, child
);
3972 ret
= bdrv_can_set_aio_context(new_child
, parent_ctx
, &ignore
, NULL
);
3973 g_slist_free(ignore
);
3978 ignore
= g_slist_prepend(NULL
, child
);
3979 ret
= bdrv_can_set_aio_context(parent
, child_ctx
, &ignore
, errp
);
3980 g_slist_free(ignore
);
3985 * Take a BDRVReopenState and check if the value of 'backing' in the
3986 * reopen_state->options QDict is valid or not.
3988 * If 'backing' is missing from the QDict then return 0.
3990 * If 'backing' contains the node name of the backing file of
3991 * reopen_state->bs then return 0.
3993 * If 'backing' contains a different node name (or is null) then check
3994 * whether the current backing file can be replaced with the new one.
3995 * If that's the case then reopen_state->replace_backing_bs is set to
3996 * true and reopen_state->new_backing_bs contains a pointer to the new
3997 * backing BlockDriverState (or NULL).
3999 * Return 0 on success, otherwise return < 0 and set @errp.
4001 static int bdrv_reopen_parse_backing(BDRVReopenState
*reopen_state
,
4004 BlockDriverState
*bs
= reopen_state
->bs
;
4005 BlockDriverState
*overlay_bs
, *new_backing_bs
;
4009 value
= qdict_get(reopen_state
->options
, "backing");
4010 if (value
== NULL
) {
4014 switch (qobject_type(value
)) {
4016 new_backing_bs
= NULL
;
4019 str
= qobject_get_try_str(value
);
4020 new_backing_bs
= bdrv_lookup_bs(NULL
, str
, errp
);
4021 if (new_backing_bs
== NULL
) {
4023 } else if (bdrv_recurse_has_child(new_backing_bs
, bs
)) {
4024 error_setg(errp
, "Making '%s' a backing file of '%s' "
4025 "would create a cycle", str
, bs
->node_name
);
4030 /* 'backing' does not allow any other data type */
4031 g_assert_not_reached();
4035 * Check AioContext compatibility so that the bdrv_set_backing_hd() call in
4036 * bdrv_reopen_commit() won't fail.
4038 if (new_backing_bs
) {
4039 if (!bdrv_reopen_can_attach(bs
, bs
->backing
, new_backing_bs
, errp
)) {
4045 * Find the "actual" backing file by skipping all links that point
4046 * to an implicit node, if any (e.g. a commit filter node).
4049 while (backing_bs(overlay_bs
) && backing_bs(overlay_bs
)->implicit
) {
4050 overlay_bs
= backing_bs(overlay_bs
);
4053 /* If we want to replace the backing file we need some extra checks */
4054 if (new_backing_bs
!= backing_bs(overlay_bs
)) {
4055 /* Check for implicit nodes between bs and its backing file */
4056 if (bs
!= overlay_bs
) {
4057 error_setg(errp
, "Cannot change backing link if '%s' has "
4058 "an implicit backing file", bs
->node_name
);
4061 /* Check if the backing link that we want to replace is frozen */
4062 if (bdrv_is_backing_chain_frozen(overlay_bs
, backing_bs(overlay_bs
),
4066 reopen_state
->replace_backing_bs
= true;
4067 if (new_backing_bs
) {
4068 bdrv_ref(new_backing_bs
);
4069 reopen_state
->new_backing_bs
= new_backing_bs
;
4077 * Prepares a BlockDriverState for reopen. All changes are staged in the
4078 * 'opaque' field of the BDRVReopenState, which is used and allocated by
4079 * the block driver layer .bdrv_reopen_prepare()
4081 * bs is the BlockDriverState to reopen
4082 * flags are the new open flags
4083 * queue is the reopen queue
4085 * Returns 0 on success, non-zero on error. On error errp will be set
4088 * On failure, bdrv_reopen_abort() will be called to clean up any data.
4089 * It is the responsibility of the caller to then call the abort() or
4090 * commit() for any other BDS that have been left in a prepare() state
4093 int bdrv_reopen_prepare(BDRVReopenState
*reopen_state
, BlockReopenQueue
*queue
,
4098 Error
*local_err
= NULL
;
4101 QDict
*orig_reopen_opts
;
4102 char *discard
= NULL
;
4104 bool drv_prepared
= false;
4106 assert(reopen_state
!= NULL
);
4107 assert(reopen_state
->bs
->drv
!= NULL
);
4108 drv
= reopen_state
->bs
->drv
;
4110 /* This function and each driver's bdrv_reopen_prepare() remove
4111 * entries from reopen_state->options as they are processed, so
4112 * we need to make a copy of the original QDict. */
4113 orig_reopen_opts
= qdict_clone_shallow(reopen_state
->options
);
4115 /* Process generic block layer options */
4116 opts
= qemu_opts_create(&bdrv_runtime_opts
, NULL
, 0, &error_abort
);
4117 if (!qemu_opts_absorb_qdict(opts
, reopen_state
->options
, errp
)) {
4122 /* This was already called in bdrv_reopen_queue_child() so the flags
4123 * are up-to-date. This time we simply want to remove the options from
4124 * QemuOpts in order to indicate that they have been processed. */
4125 old_flags
= reopen_state
->flags
;
4126 update_flags_from_options(&reopen_state
->flags
, opts
);
4127 assert(old_flags
== reopen_state
->flags
);
4129 discard
= qemu_opt_get_del(opts
, BDRV_OPT_DISCARD
);
4130 if (discard
!= NULL
) {
4131 if (bdrv_parse_discard_flags(discard
, &reopen_state
->flags
) != 0) {
4132 error_setg(errp
, "Invalid discard option");
4138 reopen_state
->detect_zeroes
=
4139 bdrv_parse_detect_zeroes(opts
, reopen_state
->flags
, &local_err
);
4141 error_propagate(errp
, local_err
);
4146 /* All other options (including node-name and driver) must be unchanged.
4147 * Put them back into the QDict, so that they are checked at the end
4148 * of this function. */
4149 qemu_opts_to_qdict(opts
, reopen_state
->options
);
4151 /* If we are to stay read-only, do not allow permission change
4152 * to r/w. Attempting to set to r/w may fail if either BDRV_O_ALLOW_RDWR is
4153 * not set, or if the BDS still has copy_on_read enabled */
4154 read_only
= !(reopen_state
->flags
& BDRV_O_RDWR
);
4155 ret
= bdrv_can_set_read_only(reopen_state
->bs
, read_only
, true, &local_err
);
4157 error_propagate(errp
, local_err
);
4161 /* Calculate required permissions after reopening */
4162 bdrv_reopen_perm(queue
, reopen_state
->bs
,
4163 &reopen_state
->perm
, &reopen_state
->shared_perm
);
4165 ret
= bdrv_flush(reopen_state
->bs
);
4167 error_setg_errno(errp
, -ret
, "Error flushing drive");
4171 if (drv
->bdrv_reopen_prepare
) {
4173 * If a driver-specific option is missing, it means that we
4174 * should reset it to its default value.
4175 * But not all options allow that, so we need to check it first.
4177 ret
= bdrv_reset_options_allowed(reopen_state
->bs
,
4178 reopen_state
->options
, errp
);
4183 ret
= drv
->bdrv_reopen_prepare(reopen_state
, queue
, &local_err
);
4185 if (local_err
!= NULL
) {
4186 error_propagate(errp
, local_err
);
4188 bdrv_refresh_filename(reopen_state
->bs
);
4189 error_setg(errp
, "failed while preparing to reopen image '%s'",
4190 reopen_state
->bs
->filename
);
4195 /* It is currently mandatory to have a bdrv_reopen_prepare()
4196 * handler for each supported drv. */
4197 error_setg(errp
, "Block format '%s' used by node '%s' "
4198 "does not support reopening files", drv
->format_name
,
4199 bdrv_get_device_or_node_name(reopen_state
->bs
));
4204 drv_prepared
= true;
4207 * We must provide the 'backing' option if the BDS has a backing
4208 * file or if the image file has a backing file name as part of
4209 * its metadata. Otherwise the 'backing' option can be omitted.
4211 if (drv
->supports_backing
&& reopen_state
->backing_missing
&&
4212 (backing_bs(reopen_state
->bs
) || reopen_state
->bs
->backing_file
[0])) {
4213 error_setg(errp
, "backing is missing for '%s'",
4214 reopen_state
->bs
->node_name
);
4220 * Allow changing the 'backing' option. The new value can be
4221 * either a reference to an existing node (using its node name)
4222 * or NULL to simply detach the current backing file.
4224 ret
= bdrv_reopen_parse_backing(reopen_state
, errp
);
4228 qdict_del(reopen_state
->options
, "backing");
4230 /* Options that are not handled are only okay if they are unchanged
4231 * compared to the old state. It is expected that some options are only
4232 * used for the initial open, but not reopen (e.g. filename) */
4233 if (qdict_size(reopen_state
->options
)) {
4234 const QDictEntry
*entry
= qdict_first(reopen_state
->options
);
4237 QObject
*new = entry
->value
;
4238 QObject
*old
= qdict_get(reopen_state
->bs
->options
, entry
->key
);
4240 /* Allow child references (child_name=node_name) as long as they
4241 * point to the current child (i.e. everything stays the same). */
4242 if (qobject_type(new) == QTYPE_QSTRING
) {
4244 QLIST_FOREACH(child
, &reopen_state
->bs
->children
, next
) {
4245 if (!strcmp(child
->name
, entry
->key
)) {
4251 const char *str
= qobject_get_try_str(new);
4252 if (!strcmp(child
->bs
->node_name
, str
)) {
4253 continue; /* Found child with this name, skip option */
4259 * TODO: When using -drive to specify blockdev options, all values
4260 * will be strings; however, when using -blockdev, blockdev-add or
4261 * filenames using the json:{} pseudo-protocol, they will be
4263 * In contrast, reopening options are (currently) always strings
4264 * (because you can only specify them through qemu-io; all other
4265 * callers do not specify any options).
4266 * Therefore, when using anything other than -drive to create a BDS,
4267 * this cannot detect non-string options as unchanged, because
4268 * qobject_is_equal() always returns false for objects of different
4269 * type. In the future, this should be remedied by correctly typing
4270 * all options. For now, this is not too big of an issue because
4271 * the user can simply omit options which cannot be changed anyway,
4272 * so they will stay unchanged.
4274 if (!qobject_is_equal(new, old
)) {
4275 error_setg(errp
, "Cannot change the option '%s'", entry
->key
);
4279 } while ((entry
= qdict_next(reopen_state
->options
, entry
)));
4284 /* Restore the original reopen_state->options QDict */
4285 qobject_unref(reopen_state
->options
);
4286 reopen_state
->options
= qobject_ref(orig_reopen_opts
);
4289 if (ret
< 0 && drv_prepared
) {
4290 /* drv->bdrv_reopen_prepare() has succeeded, so we need to
4291 * call drv->bdrv_reopen_abort() before signaling an error
4292 * (bdrv_reopen_multiple() will not call bdrv_reopen_abort()
4293 * when the respective bdrv_reopen_prepare() has failed) */
4294 if (drv
->bdrv_reopen_abort
) {
4295 drv
->bdrv_reopen_abort(reopen_state
);
4298 qemu_opts_del(opts
);
4299 qobject_unref(orig_reopen_opts
);
4305 * Takes the staged changes for the reopen from bdrv_reopen_prepare(), and
4306 * makes them final by swapping the staging BlockDriverState contents into
4307 * the active BlockDriverState contents.
4309 void bdrv_reopen_commit(BDRVReopenState
*reopen_state
)
4312 BlockDriverState
*bs
;
4315 assert(reopen_state
!= NULL
);
4316 bs
= reopen_state
->bs
;
4318 assert(drv
!= NULL
);
4320 /* If there are any driver level actions to take */
4321 if (drv
->bdrv_reopen_commit
) {
4322 drv
->bdrv_reopen_commit(reopen_state
);
4325 /* set BDS specific flags now */
4326 qobject_unref(bs
->explicit_options
);
4327 qobject_unref(bs
->options
);
4329 bs
->explicit_options
= reopen_state
->explicit_options
;
4330 bs
->options
= reopen_state
->options
;
4331 bs
->open_flags
= reopen_state
->flags
;
4332 bs
->read_only
= !(reopen_state
->flags
& BDRV_O_RDWR
);
4333 bs
->detect_zeroes
= reopen_state
->detect_zeroes
;
4335 if (reopen_state
->replace_backing_bs
) {
4336 qdict_del(bs
->explicit_options
, "backing");
4337 qdict_del(bs
->options
, "backing");
4340 /* Remove child references from bs->options and bs->explicit_options.
4341 * Child options were already removed in bdrv_reopen_queue_child() */
4342 QLIST_FOREACH(child
, &bs
->children
, next
) {
4343 qdict_del(bs
->explicit_options
, child
->name
);
4344 qdict_del(bs
->options
, child
->name
);
4348 * Change the backing file if a new one was specified. We do this
4349 * after updating bs->options, so bdrv_refresh_filename() (called
4350 * from bdrv_set_backing_hd()) has the new values.
4352 if (reopen_state
->replace_backing_bs
) {
4353 BlockDriverState
*old_backing_bs
= backing_bs(bs
);
4354 assert(!old_backing_bs
|| !old_backing_bs
->implicit
);
4355 /* Abort the permission update on the backing bs we're detaching */
4356 if (old_backing_bs
) {
4357 bdrv_abort_perm_update(old_backing_bs
);
4359 bdrv_set_backing_hd(bs
, reopen_state
->new_backing_bs
, &error_abort
);
4362 bdrv_refresh_limits(bs
, NULL
);
4366 * Abort the reopen, and delete and free the staged changes in
4369 void bdrv_reopen_abort(BDRVReopenState
*reopen_state
)
4373 assert(reopen_state
!= NULL
);
4374 drv
= reopen_state
->bs
->drv
;
4375 assert(drv
!= NULL
);
4377 if (drv
->bdrv_reopen_abort
) {
4378 drv
->bdrv_reopen_abort(reopen_state
);
4383 static void bdrv_close(BlockDriverState
*bs
)
4385 BdrvAioNotifier
*ban
, *ban_next
;
4386 BdrvChild
*child
, *next
;
4388 assert(!bs
->refcnt
);
4390 bdrv_drained_begin(bs
); /* complete I/O */
4392 bdrv_drain(bs
); /* in case flush left pending I/O */
4395 if (bs
->drv
->bdrv_close
) {
4396 /* Must unfreeze all children, so bdrv_unref_child() works */
4397 bs
->drv
->bdrv_close(bs
);
4402 QLIST_FOREACH_SAFE(child
, &bs
->children
, next
, next
) {
4403 bdrv_unref_child(bs
, child
);
4410 atomic_set(&bs
->copy_on_read
, 0);
4411 bs
->backing_file
[0] = '\0';
4412 bs
->backing_format
[0] = '\0';
4413 bs
->total_sectors
= 0;
4414 bs
->encrypted
= false;
4416 qobject_unref(bs
->options
);
4417 qobject_unref(bs
->explicit_options
);
4419 bs
->explicit_options
= NULL
;
4420 qobject_unref(bs
->full_open_options
);
4421 bs
->full_open_options
= NULL
;
4423 bdrv_release_named_dirty_bitmaps(bs
);
4424 assert(QLIST_EMPTY(&bs
->dirty_bitmaps
));
4426 QLIST_FOREACH_SAFE(ban
, &bs
->aio_notifiers
, list
, ban_next
) {
4429 QLIST_INIT(&bs
->aio_notifiers
);
4430 bdrv_drained_end(bs
);
4433 void bdrv_close_all(void)
4435 assert(job_next(NULL
) == NULL
);
4436 nbd_export_close_all();
4438 /* Drop references from requests still in flight, such as canceled block
4439 * jobs whose AIO context has not been polled yet */
4442 blk_remove_all_bs();
4443 blockdev_close_all_bdrv_states();
4445 assert(QTAILQ_EMPTY(&all_bdrv_states
));
4448 static bool should_update_child(BdrvChild
*c
, BlockDriverState
*to
)
4454 if (c
->klass
->stay_at_node
) {
4458 /* If the child @c belongs to the BDS @to, replacing the current
4459 * c->bs by @to would mean to create a loop.
4461 * Such a case occurs when appending a BDS to a backing chain.
4462 * For instance, imagine the following chain:
4464 * guest device -> node A -> further backing chain...
4466 * Now we create a new BDS B which we want to put on top of this
4467 * chain, so we first attach A as its backing node:
4472 * guest device -> node A -> further backing chain...
4474 * Finally we want to replace A by B. When doing that, we want to
4475 * replace all pointers to A by pointers to B -- except for the
4476 * pointer from B because (1) that would create a loop, and (2)
4477 * that pointer should simply stay intact:
4479 * guest device -> node B
4482 * node A -> further backing chain...
4484 * In general, when replacing a node A (c->bs) by a node B (@to),
4485 * if A is a child of B, that means we cannot replace A by B there
4486 * because that would create a loop. Silently detaching A from B
4487 * is also not really an option. So overall just leaving A in
4488 * place there is the most sensible choice.
4490 * We would also create a loop in any cases where @c is only
4491 * indirectly referenced by @to. Prevent this by returning false
4492 * if @c is found (by breadth-first search) anywhere in the whole
4497 found
= g_hash_table_new(NULL
, NULL
);
4498 g_hash_table_add(found
, to
);
4499 queue
= g_queue_new();
4500 g_queue_push_tail(queue
, to
);
4502 while (!g_queue_is_empty(queue
)) {
4503 BlockDriverState
*v
= g_queue_pop_head(queue
);
4506 QLIST_FOREACH(c2
, &v
->children
, next
) {
4512 if (g_hash_table_contains(found
, c2
->bs
)) {
4516 g_queue_push_tail(queue
, c2
->bs
);
4517 g_hash_table_add(found
, c2
->bs
);
4521 g_queue_free(queue
);
4522 g_hash_table_destroy(found
);
4527 void bdrv_replace_node(BlockDriverState
*from
, BlockDriverState
*to
,
4530 BdrvChild
*c
, *next
;
4531 GSList
*list
= NULL
, *p
;
4532 uint64_t perm
= 0, shared
= BLK_PERM_ALL
;
4535 /* Make sure that @from doesn't go away until we have successfully attached
4536 * all of its parents to @to. */
4539 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
4540 assert(bdrv_get_aio_context(from
) == bdrv_get_aio_context(to
));
4541 bdrv_drained_begin(from
);
4543 /* Put all parents into @list and calculate their cumulative permissions */
4544 QLIST_FOREACH_SAFE(c
, &from
->parents
, next_parent
, next
) {
4545 assert(c
->bs
== from
);
4546 if (!should_update_child(c
, to
)) {
4550 error_setg(errp
, "Cannot change '%s' link to '%s'",
4551 c
->name
, from
->node_name
);
4554 list
= g_slist_prepend(list
, c
);
4556 shared
&= c
->shared_perm
;
4559 /* Check whether the required permissions can be granted on @to, ignoring
4560 * all BdrvChild in @list so that they can't block themselves. */
4561 ret
= bdrv_check_update_perm(to
, NULL
, perm
, shared
, list
, NULL
, errp
);
4563 bdrv_abort_perm_update(to
);
4567 /* Now actually perform the change. We performed the permission check for
4568 * all elements of @list at once, so set the permissions all at once at the
4570 for (p
= list
; p
!= NULL
; p
= p
->next
) {
4574 bdrv_replace_child_noperm(c
, to
);
4578 bdrv_get_cumulative_perm(to
, &perm
, &shared
);
4579 bdrv_set_perm(to
, perm
, shared
);
4583 bdrv_drained_end(from
);
4588 * Add new bs contents at the top of an image chain while the chain is
4589 * live, while keeping required fields on the top layer.
4591 * This will modify the BlockDriverState fields, and swap contents
4592 * between bs_new and bs_top. Both bs_new and bs_top are modified.
4594 * bs_new must not be attached to a BlockBackend.
4596 * This function does not create any image files.
4598 * bdrv_append() takes ownership of a bs_new reference and unrefs it because
4599 * that's what the callers commonly need. bs_new will be referenced by the old
4600 * parents of bs_top after bdrv_append() returns. If the caller needs to keep a
4601 * reference of its own, it must call bdrv_ref().
4603 void bdrv_append(BlockDriverState
*bs_new
, BlockDriverState
*bs_top
,
4606 Error
*local_err
= NULL
;
4608 bdrv_set_backing_hd(bs_new
, bs_top
, &local_err
);
4610 error_propagate(errp
, local_err
);
4614 bdrv_replace_node(bs_top
, bs_new
, &local_err
);
4616 error_propagate(errp
, local_err
);
4617 bdrv_set_backing_hd(bs_new
, NULL
, &error_abort
);
4621 /* bs_new is now referenced by its new parents, we don't need the
4622 * additional reference any more. */
4627 static void bdrv_delete(BlockDriverState
*bs
)
4629 assert(bdrv_op_blocker_is_empty(bs
));
4630 assert(!bs
->refcnt
);
4632 /* remove from list, if necessary */
4633 if (bs
->node_name
[0] != '\0') {
4634 QTAILQ_REMOVE(&graph_bdrv_states
, bs
, node_list
);
4636 QTAILQ_REMOVE(&all_bdrv_states
, bs
, bs_list
);
4644 * Run consistency checks on an image
4646 * Returns 0 if the check could be completed (it doesn't mean that the image is
4647 * free of errors) or -errno when an internal error occurred. The results of the
4648 * check are stored in res.
4650 static int coroutine_fn
bdrv_co_check(BlockDriverState
*bs
,
4651 BdrvCheckResult
*res
, BdrvCheckMode fix
)
4653 if (bs
->drv
== NULL
) {
4656 if (bs
->drv
->bdrv_co_check
== NULL
) {
4660 memset(res
, 0, sizeof(*res
));
4661 return bs
->drv
->bdrv_co_check(bs
, res
, fix
);
4664 typedef struct CheckCo
{
4665 BlockDriverState
*bs
;
4666 BdrvCheckResult
*res
;
4671 static void coroutine_fn
bdrv_check_co_entry(void *opaque
)
4673 CheckCo
*cco
= opaque
;
4674 cco
->ret
= bdrv_co_check(cco
->bs
, cco
->res
, cco
->fix
);
4678 int bdrv_check(BlockDriverState
*bs
,
4679 BdrvCheckResult
*res
, BdrvCheckMode fix
)
4685 .ret
= -EINPROGRESS
,
4689 if (qemu_in_coroutine()) {
4690 /* Fast-path if already in coroutine context */
4691 bdrv_check_co_entry(&cco
);
4693 co
= qemu_coroutine_create(bdrv_check_co_entry
, &cco
);
4694 bdrv_coroutine_enter(bs
, co
);
4695 BDRV_POLL_WHILE(bs
, cco
.ret
== -EINPROGRESS
);
4704 * -EINVAL - backing format specified, but no file
4705 * -ENOSPC - can't update the backing file because no space is left in the
4707 * -ENOTSUP - format driver doesn't support changing the backing file
4709 int bdrv_change_backing_file(BlockDriverState
*bs
, const char *backing_file
,
4710 const char *backing_fmt
, bool warn
)
4712 BlockDriver
*drv
= bs
->drv
;
4719 /* Backing file format doesn't make sense without a backing file */
4720 if (backing_fmt
&& !backing_file
) {
4724 if (warn
&& backing_file
&& !backing_fmt
) {
4725 warn_report("Deprecated use of backing file without explicit "
4726 "backing format, use of this image requires "
4727 "potentially unsafe format probing");
4730 if (drv
->bdrv_change_backing_file
!= NULL
) {
4731 ret
= drv
->bdrv_change_backing_file(bs
, backing_file
, backing_fmt
);
4737 pstrcpy(bs
->backing_file
, sizeof(bs
->backing_file
), backing_file
?: "");
4738 pstrcpy(bs
->backing_format
, sizeof(bs
->backing_format
), backing_fmt
?: "");
4739 pstrcpy(bs
->auto_backing_file
, sizeof(bs
->auto_backing_file
),
4740 backing_file
?: "");
4746 * Finds the image layer in the chain that has 'bs' as its backing file.
4748 * active is the current topmost image.
4750 * Returns NULL if bs is not found in active's image chain,
4751 * or if active == bs.
4753 * Returns the bottommost base image if bs == NULL.
4755 BlockDriverState
*bdrv_find_overlay(BlockDriverState
*active
,
4756 BlockDriverState
*bs
)
4758 while (active
&& bs
!= backing_bs(active
)) {
4759 active
= backing_bs(active
);
4765 /* Given a BDS, searches for the base layer. */
4766 BlockDriverState
*bdrv_find_base(BlockDriverState
*bs
)
4768 return bdrv_find_overlay(bs
, NULL
);
4772 * Return true if at least one of the COW (backing) and filter links
4773 * between @bs and @base is frozen. @errp is set if that's the case.
4774 * @base must be reachable from @bs, or NULL.
4776 bool bdrv_is_backing_chain_frozen(BlockDriverState
*bs
, BlockDriverState
*base
,
4779 BlockDriverState
*i
;
4782 for (i
= bs
; i
!= base
; i
= child_bs(child
)) {
4783 child
= bdrv_filter_or_cow_child(i
);
4785 if (child
&& child
->frozen
) {
4786 error_setg(errp
, "Cannot change '%s' link from '%s' to '%s'",
4787 child
->name
, i
->node_name
, child
->bs
->node_name
);
4796 * Freeze all COW (backing) and filter links between @bs and @base.
4797 * If any of the links is already frozen the operation is aborted and
4798 * none of the links are modified.
4799 * @base must be reachable from @bs, or NULL.
4800 * Returns 0 on success. On failure returns < 0 and sets @errp.
4802 int bdrv_freeze_backing_chain(BlockDriverState
*bs
, BlockDriverState
*base
,
4805 BlockDriverState
*i
;
4808 if (bdrv_is_backing_chain_frozen(bs
, base
, errp
)) {
4812 for (i
= bs
; i
!= base
; i
= child_bs(child
)) {
4813 child
= bdrv_filter_or_cow_child(i
);
4814 if (child
&& child
->bs
->never_freeze
) {
4815 error_setg(errp
, "Cannot freeze '%s' link to '%s'",
4816 child
->name
, child
->bs
->node_name
);
4821 for (i
= bs
; i
!= base
; i
= child_bs(child
)) {
4822 child
= bdrv_filter_or_cow_child(i
);
4824 child
->frozen
= true;
4832 * Unfreeze all COW (backing) and filter links between @bs and @base.
4833 * The caller must ensure that all links are frozen before using this
4835 * @base must be reachable from @bs, or NULL.
4837 void bdrv_unfreeze_backing_chain(BlockDriverState
*bs
, BlockDriverState
*base
)
4839 BlockDriverState
*i
;
4842 for (i
= bs
; i
!= base
; i
= child_bs(child
)) {
4843 child
= bdrv_filter_or_cow_child(i
);
4845 assert(child
->frozen
);
4846 child
->frozen
= false;
4852 * Drops images above 'base' up to and including 'top', and sets the image
4853 * above 'top' to have base as its backing file.
4855 * Requires that the overlay to 'top' is opened r/w, so that the backing file
4856 * information in 'bs' can be properly updated.
4858 * E.g., this will convert the following chain:
4859 * bottom <- base <- intermediate <- top <- active
4863 * bottom <- base <- active
4865 * It is allowed for bottom==base, in which case it converts:
4867 * base <- intermediate <- top <- active
4873 * If backing_file_str is non-NULL, it will be used when modifying top's
4874 * overlay image metadata.
4877 * if active == top, that is considered an error
4880 int bdrv_drop_intermediate(BlockDriverState
*top
, BlockDriverState
*base
,
4881 const char *backing_file_str
)
4883 BlockDriverState
*explicit_top
= top
;
4884 bool update_inherits_from
;
4885 BdrvChild
*c
, *next
;
4886 Error
*local_err
= NULL
;
4890 bdrv_subtree_drained_begin(top
);
4892 if (!top
->drv
|| !base
->drv
) {
4896 /* Make sure that base is in the backing chain of top */
4897 if (!bdrv_chain_contains(top
, base
)) {
4901 /* This function changes all links that point to top and makes
4902 * them point to base. Check that none of them is frozen. */
4903 QLIST_FOREACH(c
, &top
->parents
, next_parent
) {
4909 /* If 'base' recursively inherits from 'top' then we should set
4910 * base->inherits_from to top->inherits_from after 'top' and all
4911 * other intermediate nodes have been dropped.
4912 * If 'top' is an implicit node (e.g. "commit_top") we should skip
4913 * it because no one inherits from it. We use explicit_top for that. */
4914 while (explicit_top
&& explicit_top
->implicit
) {
4915 explicit_top
= backing_bs(explicit_top
);
4917 update_inherits_from
= bdrv_inherits_from_recursive(base
, explicit_top
);
4919 /* success - we can delete the intermediate states, and link top->base */
4920 /* TODO Check graph modification op blockers (BLK_PERM_GRAPH_MOD) once
4921 * we've figured out how they should work. */
4922 if (!backing_file_str
) {
4923 bdrv_refresh_filename(base
);
4924 backing_file_str
= base
->filename
;
4927 QLIST_FOREACH_SAFE(c
, &top
->parents
, next_parent
, next
) {
4928 /* Check whether we are allowed to switch c from top to base */
4929 GSList
*ignore_children
= g_slist_prepend(NULL
, c
);
4930 ret
= bdrv_check_update_perm(base
, NULL
, c
->perm
, c
->shared_perm
,
4931 ignore_children
, NULL
, &local_err
);
4932 g_slist_free(ignore_children
);
4934 error_report_err(local_err
);
4938 /* If so, update the backing file path in the image file */
4939 if (c
->klass
->update_filename
) {
4940 ret
= c
->klass
->update_filename(c
, base
, backing_file_str
,
4943 bdrv_abort_perm_update(base
);
4944 error_report_err(local_err
);
4950 * Do the actual switch in the in-memory graph.
4951 * Completes bdrv_check_update_perm() transaction internally.
4952 * c->frozen is false, we have checked that above.
4955 bdrv_replace_child(c
, base
);
4959 if (update_inherits_from
) {
4960 base
->inherits_from
= explicit_top
->inherits_from
;
4965 bdrv_subtree_drained_end(top
);
4971 * Length of a allocated file in bytes. Sparse files are counted by actual
4972 * allocated space. Return < 0 if error or unknown.
4974 int64_t bdrv_get_allocated_file_size(BlockDriverState
*bs
)
4976 BlockDriver
*drv
= bs
->drv
;
4980 if (drv
->bdrv_get_allocated_file_size
) {
4981 return drv
->bdrv_get_allocated_file_size(bs
);
4984 return bdrv_get_allocated_file_size(bs
->file
->bs
);
4991 * @drv: Format driver
4992 * @opts: Creation options for new image
4993 * @in_bs: Existing image containing data for new image (may be NULL)
4994 * @errp: Error object
4995 * Returns: A #BlockMeasureInfo (free using qapi_free_BlockMeasureInfo())
4998 * Calculate file size required to create a new image.
5000 * If @in_bs is given then space for allocated clusters and zero clusters
5001 * from that image are included in the calculation. If @opts contains a
5002 * backing file that is shared by @in_bs then backing clusters may be omitted
5003 * from the calculation.
5005 * If @in_bs is NULL then the calculation includes no allocated clusters
5006 * unless a preallocation option is given in @opts.
5008 * Note that @in_bs may use a different BlockDriver from @drv.
5010 * If an error occurs the @errp pointer is set.
5012 BlockMeasureInfo
*bdrv_measure(BlockDriver
*drv
, QemuOpts
*opts
,
5013 BlockDriverState
*in_bs
, Error
**errp
)
5015 if (!drv
->bdrv_measure
) {
5016 error_setg(errp
, "Block driver '%s' does not support size measurement",
5021 return drv
->bdrv_measure(opts
, in_bs
, errp
);
5025 * Return number of sectors on success, -errno on error.
5027 int64_t bdrv_nb_sectors(BlockDriverState
*bs
)
5029 BlockDriver
*drv
= bs
->drv
;
5034 if (drv
->has_variable_length
) {
5035 int ret
= refresh_total_sectors(bs
, bs
->total_sectors
);
5040 return bs
->total_sectors
;
5044 * Return length in bytes on success, -errno on error.
5045 * The length is always a multiple of BDRV_SECTOR_SIZE.
5047 int64_t bdrv_getlength(BlockDriverState
*bs
)
5049 int64_t ret
= bdrv_nb_sectors(bs
);
5051 ret
= ret
> INT64_MAX
/ BDRV_SECTOR_SIZE
? -EFBIG
: ret
;
5052 return ret
< 0 ? ret
: ret
* BDRV_SECTOR_SIZE
;
5055 /* return 0 as number of sectors if no device present or error */
5056 void bdrv_get_geometry(BlockDriverState
*bs
, uint64_t *nb_sectors_ptr
)
5058 int64_t nb_sectors
= bdrv_nb_sectors(bs
);
5060 *nb_sectors_ptr
= nb_sectors
< 0 ? 0 : nb_sectors
;
5063 bool bdrv_is_sg(BlockDriverState
*bs
)
5068 bool bdrv_is_encrypted(BlockDriverState
*bs
)
5070 if (bs
->backing
&& bs
->backing
->bs
->encrypted
) {
5073 return bs
->encrypted
;
5076 const char *bdrv_get_format_name(BlockDriverState
*bs
)
5078 return bs
->drv
? bs
->drv
->format_name
: NULL
;
5081 static int qsort_strcmp(const void *a
, const void *b
)
5083 return strcmp(*(char *const *)a
, *(char *const *)b
);
5086 void bdrv_iterate_format(void (*it
)(void *opaque
, const char *name
),
5087 void *opaque
, bool read_only
)
5092 const char **formats
= NULL
;
5094 QLIST_FOREACH(drv
, &bdrv_drivers
, list
) {
5095 if (drv
->format_name
) {
5099 if (use_bdrv_whitelist
&& !bdrv_is_whitelisted(drv
, read_only
)) {
5103 while (formats
&& i
&& !found
) {
5104 found
= !strcmp(formats
[--i
], drv
->format_name
);
5108 formats
= g_renew(const char *, formats
, count
+ 1);
5109 formats
[count
++] = drv
->format_name
;
5114 for (i
= 0; i
< (int)ARRAY_SIZE(block_driver_modules
); i
++) {
5115 const char *format_name
= block_driver_modules
[i
].format_name
;
5121 if (use_bdrv_whitelist
&&
5122 !bdrv_format_is_whitelisted(format_name
, read_only
)) {
5126 while (formats
&& j
&& !found
) {
5127 found
= !strcmp(formats
[--j
], format_name
);
5131 formats
= g_renew(const char *, formats
, count
+ 1);
5132 formats
[count
++] = format_name
;
5137 qsort(formats
, count
, sizeof(formats
[0]), qsort_strcmp
);
5139 for (i
= 0; i
< count
; i
++) {
5140 it(opaque
, formats
[i
]);
5146 /* This function is to find a node in the bs graph */
5147 BlockDriverState
*bdrv_find_node(const char *node_name
)
5149 BlockDriverState
*bs
;
5153 QTAILQ_FOREACH(bs
, &graph_bdrv_states
, node_list
) {
5154 if (!strcmp(node_name
, bs
->node_name
)) {
5161 /* Put this QMP function here so it can access the static graph_bdrv_states. */
5162 BlockDeviceInfoList
*bdrv_named_nodes_list(bool flat
,
5165 BlockDeviceInfoList
*list
, *entry
;
5166 BlockDriverState
*bs
;
5169 QTAILQ_FOREACH(bs
, &graph_bdrv_states
, node_list
) {
5170 BlockDeviceInfo
*info
= bdrv_block_device_info(NULL
, bs
, flat
, errp
);
5172 qapi_free_BlockDeviceInfoList(list
);
5175 entry
= g_malloc0(sizeof(*entry
));
5176 entry
->value
= info
;
5184 #define QAPI_LIST_ADD(list, element) do { \
5185 typeof(list) _tmp = g_new(typeof(*(list)), 1); \
5186 _tmp->value = (element); \
5187 _tmp->next = (list); \
5191 typedef struct XDbgBlockGraphConstructor
{
5192 XDbgBlockGraph
*graph
;
5193 GHashTable
*graph_nodes
;
5194 } XDbgBlockGraphConstructor
;
5196 static XDbgBlockGraphConstructor
*xdbg_graph_new(void)
5198 XDbgBlockGraphConstructor
*gr
= g_new(XDbgBlockGraphConstructor
, 1);
5200 gr
->graph
= g_new0(XDbgBlockGraph
, 1);
5201 gr
->graph_nodes
= g_hash_table_new(NULL
, NULL
);
5206 static XDbgBlockGraph
*xdbg_graph_finalize(XDbgBlockGraphConstructor
*gr
)
5208 XDbgBlockGraph
*graph
= gr
->graph
;
5210 g_hash_table_destroy(gr
->graph_nodes
);
5216 static uintptr_t xdbg_graph_node_num(XDbgBlockGraphConstructor
*gr
, void *node
)
5218 uintptr_t ret
= (uintptr_t)g_hash_table_lookup(gr
->graph_nodes
, node
);
5225 * Start counting from 1, not 0, because 0 interferes with not-found (NULL)
5226 * answer of g_hash_table_lookup.
5228 ret
= g_hash_table_size(gr
->graph_nodes
) + 1;
5229 g_hash_table_insert(gr
->graph_nodes
, node
, (void *)ret
);
5234 static void xdbg_graph_add_node(XDbgBlockGraphConstructor
*gr
, void *node
,
5235 XDbgBlockGraphNodeType type
, const char *name
)
5237 XDbgBlockGraphNode
*n
;
5239 n
= g_new0(XDbgBlockGraphNode
, 1);
5241 n
->id
= xdbg_graph_node_num(gr
, node
);
5243 n
->name
= g_strdup(name
);
5245 QAPI_LIST_ADD(gr
->graph
->nodes
, n
);
5248 static void xdbg_graph_add_edge(XDbgBlockGraphConstructor
*gr
, void *parent
,
5249 const BdrvChild
*child
)
5251 BlockPermission qapi_perm
;
5252 XDbgBlockGraphEdge
*edge
;
5254 edge
= g_new0(XDbgBlockGraphEdge
, 1);
5256 edge
->parent
= xdbg_graph_node_num(gr
, parent
);
5257 edge
->child
= xdbg_graph_node_num(gr
, child
->bs
);
5258 edge
->name
= g_strdup(child
->name
);
5260 for (qapi_perm
= 0; qapi_perm
< BLOCK_PERMISSION__MAX
; qapi_perm
++) {
5261 uint64_t flag
= bdrv_qapi_perm_to_blk_perm(qapi_perm
);
5263 if (flag
& child
->perm
) {
5264 QAPI_LIST_ADD(edge
->perm
, qapi_perm
);
5266 if (flag
& child
->shared_perm
) {
5267 QAPI_LIST_ADD(edge
->shared_perm
, qapi_perm
);
5271 QAPI_LIST_ADD(gr
->graph
->edges
, edge
);
5275 XDbgBlockGraph
*bdrv_get_xdbg_block_graph(Error
**errp
)
5279 BlockDriverState
*bs
;
5281 XDbgBlockGraphConstructor
*gr
= xdbg_graph_new();
5283 for (blk
= blk_all_next(NULL
); blk
; blk
= blk_all_next(blk
)) {
5284 char *allocated_name
= NULL
;
5285 const char *name
= blk_name(blk
);
5288 name
= allocated_name
= blk_get_attached_dev_id(blk
);
5290 xdbg_graph_add_node(gr
, blk
, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_BACKEND
,
5292 g_free(allocated_name
);
5293 if (blk_root(blk
)) {
5294 xdbg_graph_add_edge(gr
, blk
, blk_root(blk
));
5298 for (job
= block_job_next(NULL
); job
; job
= block_job_next(job
)) {
5301 xdbg_graph_add_node(gr
, job
, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_JOB
,
5303 for (el
= job
->nodes
; el
; el
= el
->next
) {
5304 xdbg_graph_add_edge(gr
, job
, (BdrvChild
*)el
->data
);
5308 QTAILQ_FOREACH(bs
, &graph_bdrv_states
, node_list
) {
5309 xdbg_graph_add_node(gr
, bs
, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_DRIVER
,
5311 QLIST_FOREACH(child
, &bs
->children
, next
) {
5312 xdbg_graph_add_edge(gr
, bs
, child
);
5316 return xdbg_graph_finalize(gr
);
5319 BlockDriverState
*bdrv_lookup_bs(const char *device
,
5320 const char *node_name
,
5324 BlockDriverState
*bs
;
5327 blk
= blk_by_name(device
);
5332 error_setg(errp
, "Device '%s' has no medium", device
);
5340 bs
= bdrv_find_node(node_name
);
5347 error_setg(errp
, "Cannot find device=%s nor node_name=%s",
5348 device
? device
: "",
5349 node_name
? node_name
: "");
5353 /* If 'base' is in the same chain as 'top', return true. Otherwise,
5354 * return false. If either argument is NULL, return false. */
5355 bool bdrv_chain_contains(BlockDriverState
*top
, BlockDriverState
*base
)
5357 while (top
&& top
!= base
) {
5358 top
= backing_bs(top
);
5364 BlockDriverState
*bdrv_next_node(BlockDriverState
*bs
)
5367 return QTAILQ_FIRST(&graph_bdrv_states
);
5369 return QTAILQ_NEXT(bs
, node_list
);
5372 BlockDriverState
*bdrv_next_all_states(BlockDriverState
*bs
)
5375 return QTAILQ_FIRST(&all_bdrv_states
);
5377 return QTAILQ_NEXT(bs
, bs_list
);
5380 const char *bdrv_get_node_name(const BlockDriverState
*bs
)
5382 return bs
->node_name
;
5385 const char *bdrv_get_parent_name(const BlockDriverState
*bs
)
5390 /* If multiple parents have a name, just pick the first one. */
5391 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
5392 if (c
->klass
->get_name
) {
5393 name
= c
->klass
->get_name(c
);
5394 if (name
&& *name
) {
5403 /* TODO check what callers really want: bs->node_name or blk_name() */
5404 const char *bdrv_get_device_name(const BlockDriverState
*bs
)
5406 return bdrv_get_parent_name(bs
) ?: "";
5409 /* This can be used to identify nodes that might not have a device
5410 * name associated. Since node and device names live in the same
5411 * namespace, the result is unambiguous. The exception is if both are
5412 * absent, then this returns an empty (non-null) string. */
5413 const char *bdrv_get_device_or_node_name(const BlockDriverState
*bs
)
5415 return bdrv_get_parent_name(bs
) ?: bs
->node_name
;
5418 int bdrv_get_flags(BlockDriverState
*bs
)
5420 return bs
->open_flags
;
5423 int bdrv_has_zero_init_1(BlockDriverState
*bs
)
5428 int bdrv_has_zero_init(BlockDriverState
*bs
)
5434 /* If BS is a copy on write image, it is initialized to
5435 the contents of the base image, which may not be zeroes. */
5436 if (bdrv_cow_child(bs
)) {
5439 if (bs
->drv
->bdrv_has_zero_init
) {
5440 return bs
->drv
->bdrv_has_zero_init(bs
);
5442 if (bs
->file
&& bs
->drv
->is_filter
) {
5443 return bdrv_has_zero_init(bs
->file
->bs
);
5450 bool bdrv_can_write_zeroes_with_unmap(BlockDriverState
*bs
)
5452 if (!(bs
->open_flags
& BDRV_O_UNMAP
)) {
5456 return bs
->supported_zero_flags
& BDRV_REQ_MAY_UNMAP
;
5459 void bdrv_get_backing_filename(BlockDriverState
*bs
,
5460 char *filename
, int filename_size
)
5462 pstrcpy(filename
, filename_size
, bs
->backing_file
);
5465 int bdrv_get_info(BlockDriverState
*bs
, BlockDriverInfo
*bdi
)
5467 BlockDriver
*drv
= bs
->drv
;
5468 /* if bs->drv == NULL, bs is closed, so there's nothing to do here */
5472 if (!drv
->bdrv_get_info
) {
5473 if (bs
->file
&& drv
->is_filter
) {
5474 return bdrv_get_info(bs
->file
->bs
, bdi
);
5478 memset(bdi
, 0, sizeof(*bdi
));
5479 return drv
->bdrv_get_info(bs
, bdi
);
5482 ImageInfoSpecific
*bdrv_get_specific_info(BlockDriverState
*bs
,
5485 BlockDriver
*drv
= bs
->drv
;
5486 if (drv
&& drv
->bdrv_get_specific_info
) {
5487 return drv
->bdrv_get_specific_info(bs
, errp
);
5492 BlockStatsSpecific
*bdrv_get_specific_stats(BlockDriverState
*bs
)
5494 BlockDriver
*drv
= bs
->drv
;
5495 if (!drv
|| !drv
->bdrv_get_specific_stats
) {
5498 return drv
->bdrv_get_specific_stats(bs
);
5501 void bdrv_debug_event(BlockDriverState
*bs
, BlkdebugEvent event
)
5503 if (!bs
|| !bs
->drv
|| !bs
->drv
->bdrv_debug_event
) {
5507 bs
->drv
->bdrv_debug_event(bs
, event
);
5510 static BlockDriverState
*bdrv_find_debug_node(BlockDriverState
*bs
)
5512 while (bs
&& bs
->drv
&& !bs
->drv
->bdrv_debug_breakpoint
) {
5518 if (bs
->drv
->is_filter
&& bs
->backing
) {
5519 bs
= bs
->backing
->bs
;
5526 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_breakpoint
) {
5527 assert(bs
->drv
->bdrv_debug_remove_breakpoint
);
5534 int bdrv_debug_breakpoint(BlockDriverState
*bs
, const char *event
,
5537 bs
= bdrv_find_debug_node(bs
);
5539 return bs
->drv
->bdrv_debug_breakpoint(bs
, event
, tag
);
5545 int bdrv_debug_remove_breakpoint(BlockDriverState
*bs
, const char *tag
)
5547 bs
= bdrv_find_debug_node(bs
);
5549 return bs
->drv
->bdrv_debug_remove_breakpoint(bs
, tag
);
5555 int bdrv_debug_resume(BlockDriverState
*bs
, const char *tag
)
5557 while (bs
&& (!bs
->drv
|| !bs
->drv
->bdrv_debug_resume
)) {
5558 bs
= bs
->file
? bs
->file
->bs
: NULL
;
5561 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_resume
) {
5562 return bs
->drv
->bdrv_debug_resume(bs
, tag
);
5568 bool bdrv_debug_is_suspended(BlockDriverState
*bs
, const char *tag
)
5570 while (bs
&& bs
->drv
&& !bs
->drv
->bdrv_debug_is_suspended
) {
5571 bs
= bs
->file
? bs
->file
->bs
: NULL
;
5574 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_is_suspended
) {
5575 return bs
->drv
->bdrv_debug_is_suspended(bs
, tag
);
5581 /* backing_file can either be relative, or absolute, or a protocol. If it is
5582 * relative, it must be relative to the chain. So, passing in bs->filename
5583 * from a BDS as backing_file should not be done, as that may be relative to
5584 * the CWD rather than the chain. */
5585 BlockDriverState
*bdrv_find_backing_image(BlockDriverState
*bs
,
5586 const char *backing_file
)
5588 char *filename_full
= NULL
;
5589 char *backing_file_full
= NULL
;
5590 char *filename_tmp
= NULL
;
5591 int is_protocol
= 0;
5592 BlockDriverState
*curr_bs
= NULL
;
5593 BlockDriverState
*retval
= NULL
;
5595 if (!bs
|| !bs
->drv
|| !backing_file
) {
5599 filename_full
= g_malloc(PATH_MAX
);
5600 backing_file_full
= g_malloc(PATH_MAX
);
5602 is_protocol
= path_has_protocol(backing_file
);
5604 for (curr_bs
= bs
; curr_bs
->backing
; curr_bs
= curr_bs
->backing
->bs
) {
5606 /* If either of the filename paths is actually a protocol, then
5607 * compare unmodified paths; otherwise make paths relative */
5608 if (is_protocol
|| path_has_protocol(curr_bs
->backing_file
)) {
5609 char *backing_file_full_ret
;
5611 if (strcmp(backing_file
, curr_bs
->backing_file
) == 0) {
5612 retval
= curr_bs
->backing
->bs
;
5615 /* Also check against the full backing filename for the image */
5616 backing_file_full_ret
= bdrv_get_full_backing_filename(curr_bs
,
5618 if (backing_file_full_ret
) {
5619 bool equal
= strcmp(backing_file
, backing_file_full_ret
) == 0;
5620 g_free(backing_file_full_ret
);
5622 retval
= curr_bs
->backing
->bs
;
5627 /* If not an absolute filename path, make it relative to the current
5628 * image's filename path */
5629 filename_tmp
= bdrv_make_absolute_filename(curr_bs
, backing_file
,
5631 /* We are going to compare canonicalized absolute pathnames */
5632 if (!filename_tmp
|| !realpath(filename_tmp
, filename_full
)) {
5633 g_free(filename_tmp
);
5636 g_free(filename_tmp
);
5638 /* We need to make sure the backing filename we are comparing against
5639 * is relative to the current image filename (or absolute) */
5640 filename_tmp
= bdrv_get_full_backing_filename(curr_bs
, NULL
);
5641 if (!filename_tmp
|| !realpath(filename_tmp
, backing_file_full
)) {
5642 g_free(filename_tmp
);
5645 g_free(filename_tmp
);
5647 if (strcmp(backing_file_full
, filename_full
) == 0) {
5648 retval
= curr_bs
->backing
->bs
;
5654 g_free(filename_full
);
5655 g_free(backing_file_full
);
5659 void bdrv_init(void)
5661 module_call_init(MODULE_INIT_BLOCK
);
5664 void bdrv_init_with_whitelist(void)
5666 use_bdrv_whitelist
= 1;
5670 static void coroutine_fn
bdrv_co_invalidate_cache(BlockDriverState
*bs
,
5673 BdrvChild
*child
, *parent
;
5674 uint64_t perm
, shared_perm
;
5675 Error
*local_err
= NULL
;
5677 BdrvDirtyBitmap
*bm
;
5683 QLIST_FOREACH(child
, &bs
->children
, next
) {
5684 bdrv_co_invalidate_cache(child
->bs
, &local_err
);
5686 error_propagate(errp
, local_err
);
5692 * Update permissions, they may differ for inactive nodes.
5694 * Note that the required permissions of inactive images are always a
5695 * subset of the permissions required after activating the image. This
5696 * allows us to just get the permissions upfront without restricting
5697 * drv->bdrv_invalidate_cache().
5699 * It also means that in error cases, we don't have to try and revert to
5700 * the old permissions (which is an operation that could fail, too). We can
5701 * just keep the extended permissions for the next time that an activation
5702 * of the image is tried.
5704 if (bs
->open_flags
& BDRV_O_INACTIVE
) {
5705 bs
->open_flags
&= ~BDRV_O_INACTIVE
;
5706 bdrv_get_cumulative_perm(bs
, &perm
, &shared_perm
);
5707 ret
= bdrv_check_perm(bs
, NULL
, perm
, shared_perm
, NULL
, NULL
, errp
);
5709 bs
->open_flags
|= BDRV_O_INACTIVE
;
5712 bdrv_set_perm(bs
, perm
, shared_perm
);
5714 if (bs
->drv
->bdrv_co_invalidate_cache
) {
5715 bs
->drv
->bdrv_co_invalidate_cache(bs
, &local_err
);
5717 bs
->open_flags
|= BDRV_O_INACTIVE
;
5718 error_propagate(errp
, local_err
);
5723 FOR_EACH_DIRTY_BITMAP(bs
, bm
) {
5724 bdrv_dirty_bitmap_skip_store(bm
, false);
5727 ret
= refresh_total_sectors(bs
, bs
->total_sectors
);
5729 bs
->open_flags
|= BDRV_O_INACTIVE
;
5730 error_setg_errno(errp
, -ret
, "Could not refresh total sector count");
5735 QLIST_FOREACH(parent
, &bs
->parents
, next_parent
) {
5736 if (parent
->klass
->activate
) {
5737 parent
->klass
->activate(parent
, &local_err
);
5739 bs
->open_flags
|= BDRV_O_INACTIVE
;
5740 error_propagate(errp
, local_err
);
5747 typedef struct InvalidateCacheCo
{
5748 BlockDriverState
*bs
;
5751 } InvalidateCacheCo
;
5753 static void coroutine_fn
bdrv_invalidate_cache_co_entry(void *opaque
)
5755 InvalidateCacheCo
*ico
= opaque
;
5756 bdrv_co_invalidate_cache(ico
->bs
, ico
->errp
);
5761 void bdrv_invalidate_cache(BlockDriverState
*bs
, Error
**errp
)
5764 InvalidateCacheCo ico
= {
5770 if (qemu_in_coroutine()) {
5771 /* Fast-path if already in coroutine context */
5772 bdrv_invalidate_cache_co_entry(&ico
);
5774 co
= qemu_coroutine_create(bdrv_invalidate_cache_co_entry
, &ico
);
5775 bdrv_coroutine_enter(bs
, co
);
5776 BDRV_POLL_WHILE(bs
, !ico
.done
);
5780 void bdrv_invalidate_cache_all(Error
**errp
)
5782 BlockDriverState
*bs
;
5783 Error
*local_err
= NULL
;
5784 BdrvNextIterator it
;
5786 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
5787 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
5789 aio_context_acquire(aio_context
);
5790 bdrv_invalidate_cache(bs
, &local_err
);
5791 aio_context_release(aio_context
);
5793 error_propagate(errp
, local_err
);
5794 bdrv_next_cleanup(&it
);
5800 static bool bdrv_has_bds_parent(BlockDriverState
*bs
, bool only_active
)
5804 QLIST_FOREACH(parent
, &bs
->parents
, next_parent
) {
5805 if (parent
->klass
->parent_is_bds
) {
5806 BlockDriverState
*parent_bs
= parent
->opaque
;
5807 if (!only_active
|| !(parent_bs
->open_flags
& BDRV_O_INACTIVE
)) {
5816 static int bdrv_inactivate_recurse(BlockDriverState
*bs
)
5818 BdrvChild
*child
, *parent
;
5819 bool tighten_restrictions
;
5820 uint64_t perm
, shared_perm
;
5827 /* Make sure that we don't inactivate a child before its parent.
5828 * It will be covered by recursion from the yet active parent. */
5829 if (bdrv_has_bds_parent(bs
, true)) {
5833 assert(!(bs
->open_flags
& BDRV_O_INACTIVE
));
5835 /* Inactivate this node */
5836 if (bs
->drv
->bdrv_inactivate
) {
5837 ret
= bs
->drv
->bdrv_inactivate(bs
);
5843 QLIST_FOREACH(parent
, &bs
->parents
, next_parent
) {
5844 if (parent
->klass
->inactivate
) {
5845 ret
= parent
->klass
->inactivate(parent
);
5852 bs
->open_flags
|= BDRV_O_INACTIVE
;
5854 /* Update permissions, they may differ for inactive nodes */
5855 bdrv_get_cumulative_perm(bs
, &perm
, &shared_perm
);
5856 ret
= bdrv_check_perm(bs
, NULL
, perm
, shared_perm
, NULL
,
5857 &tighten_restrictions
, NULL
);
5858 assert(tighten_restrictions
== false);
5860 /* We only tried to loosen restrictions, so errors are not fatal */
5861 bdrv_abort_perm_update(bs
);
5863 bdrv_set_perm(bs
, perm
, shared_perm
);
5867 /* Recursively inactivate children */
5868 QLIST_FOREACH(child
, &bs
->children
, next
) {
5869 ret
= bdrv_inactivate_recurse(child
->bs
);
5878 int bdrv_inactivate_all(void)
5880 BlockDriverState
*bs
= NULL
;
5881 BdrvNextIterator it
;
5883 GSList
*aio_ctxs
= NULL
, *ctx
;
5885 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
5886 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
5888 if (!g_slist_find(aio_ctxs
, aio_context
)) {
5889 aio_ctxs
= g_slist_prepend(aio_ctxs
, aio_context
);
5890 aio_context_acquire(aio_context
);
5894 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
5895 /* Nodes with BDS parents are covered by recursion from the last
5896 * parent that gets inactivated. Don't inactivate them a second
5897 * time if that has already happened. */
5898 if (bdrv_has_bds_parent(bs
, false)) {
5901 ret
= bdrv_inactivate_recurse(bs
);
5903 bdrv_next_cleanup(&it
);
5909 for (ctx
= aio_ctxs
; ctx
!= NULL
; ctx
= ctx
->next
) {
5910 AioContext
*aio_context
= ctx
->data
;
5911 aio_context_release(aio_context
);
5913 g_slist_free(aio_ctxs
);
5918 /**************************************************************/
5919 /* removable device support */
5922 * Return TRUE if the media is present
5924 bool bdrv_is_inserted(BlockDriverState
*bs
)
5926 BlockDriver
*drv
= bs
->drv
;
5932 if (drv
->bdrv_is_inserted
) {
5933 return drv
->bdrv_is_inserted(bs
);
5935 QLIST_FOREACH(child
, &bs
->children
, next
) {
5936 if (!bdrv_is_inserted(child
->bs
)) {
5944 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
5946 void bdrv_eject(BlockDriverState
*bs
, bool eject_flag
)
5948 BlockDriver
*drv
= bs
->drv
;
5950 if (drv
&& drv
->bdrv_eject
) {
5951 drv
->bdrv_eject(bs
, eject_flag
);
5956 * Lock or unlock the media (if it is locked, the user won't be able
5957 * to eject it manually).
5959 void bdrv_lock_medium(BlockDriverState
*bs
, bool locked
)
5961 BlockDriver
*drv
= bs
->drv
;
5963 trace_bdrv_lock_medium(bs
, locked
);
5965 if (drv
&& drv
->bdrv_lock_medium
) {
5966 drv
->bdrv_lock_medium(bs
, locked
);
5970 /* Get a reference to bs */
5971 void bdrv_ref(BlockDriverState
*bs
)
5976 /* Release a previously grabbed reference to bs.
5977 * If after releasing, reference count is zero, the BlockDriverState is
5979 void bdrv_unref(BlockDriverState
*bs
)
5984 assert(bs
->refcnt
> 0);
5985 if (--bs
->refcnt
== 0) {
5990 struct BdrvOpBlocker
{
5992 QLIST_ENTRY(BdrvOpBlocker
) list
;
5995 bool bdrv_op_is_blocked(BlockDriverState
*bs
, BlockOpType op
, Error
**errp
)
5997 BdrvOpBlocker
*blocker
;
5998 assert((int) op
>= 0 && op
< BLOCK_OP_TYPE_MAX
);
5999 if (!QLIST_EMPTY(&bs
->op_blockers
[op
])) {
6000 blocker
= QLIST_FIRST(&bs
->op_blockers
[op
]);
6001 error_propagate_prepend(errp
, error_copy(blocker
->reason
),
6002 "Node '%s' is busy: ",
6003 bdrv_get_device_or_node_name(bs
));
6009 void bdrv_op_block(BlockDriverState
*bs
, BlockOpType op
, Error
*reason
)
6011 BdrvOpBlocker
*blocker
;
6012 assert((int) op
>= 0 && op
< BLOCK_OP_TYPE_MAX
);
6014 blocker
= g_new0(BdrvOpBlocker
, 1);
6015 blocker
->reason
= reason
;
6016 QLIST_INSERT_HEAD(&bs
->op_blockers
[op
], blocker
, list
);
6019 void bdrv_op_unblock(BlockDriverState
*bs
, BlockOpType op
, Error
*reason
)
6021 BdrvOpBlocker
*blocker
, *next
;
6022 assert((int) op
>= 0 && op
< BLOCK_OP_TYPE_MAX
);
6023 QLIST_FOREACH_SAFE(blocker
, &bs
->op_blockers
[op
], list
, next
) {
6024 if (blocker
->reason
== reason
) {
6025 QLIST_REMOVE(blocker
, list
);
6031 void bdrv_op_block_all(BlockDriverState
*bs
, Error
*reason
)
6034 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
6035 bdrv_op_block(bs
, i
, reason
);
6039 void bdrv_op_unblock_all(BlockDriverState
*bs
, Error
*reason
)
6042 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
6043 bdrv_op_unblock(bs
, i
, reason
);
6047 bool bdrv_op_blocker_is_empty(BlockDriverState
*bs
)
6051 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
6052 if (!QLIST_EMPTY(&bs
->op_blockers
[i
])) {
6059 void bdrv_img_create(const char *filename
, const char *fmt
,
6060 const char *base_filename
, const char *base_fmt
,
6061 char *options
, uint64_t img_size
, int flags
, bool quiet
,
6064 QemuOptsList
*create_opts
= NULL
;
6065 QemuOpts
*opts
= NULL
;
6066 const char *backing_fmt
, *backing_file
;
6068 BlockDriver
*drv
, *proto_drv
;
6069 Error
*local_err
= NULL
;
6072 /* Find driver and parse its options */
6073 drv
= bdrv_find_format(fmt
);
6075 error_setg(errp
, "Unknown file format '%s'", fmt
);
6079 proto_drv
= bdrv_find_protocol(filename
, true, errp
);
6084 if (!drv
->create_opts
) {
6085 error_setg(errp
, "Format driver '%s' does not support image creation",
6090 if (!proto_drv
->create_opts
) {
6091 error_setg(errp
, "Protocol driver '%s' does not support image creation",
6092 proto_drv
->format_name
);
6096 /* Create parameter list */
6097 create_opts
= qemu_opts_append(create_opts
, drv
->create_opts
);
6098 create_opts
= qemu_opts_append(create_opts
, proto_drv
->create_opts
);
6100 opts
= qemu_opts_create(create_opts
, NULL
, 0, &error_abort
);
6102 /* Parse -o options */
6104 if (!qemu_opts_do_parse(opts
, options
, NULL
, errp
)) {
6109 if (!qemu_opt_get(opts
, BLOCK_OPT_SIZE
)) {
6110 qemu_opt_set_number(opts
, BLOCK_OPT_SIZE
, img_size
, &error_abort
);
6111 } else if (img_size
!= UINT64_C(-1)) {
6112 error_setg(errp
, "The image size must be specified only once");
6116 if (base_filename
) {
6117 if (!qemu_opt_set(opts
, BLOCK_OPT_BACKING_FILE
, base_filename
,
6119 error_setg(errp
, "Backing file not supported for file format '%s'",
6126 if (!qemu_opt_set(opts
, BLOCK_OPT_BACKING_FMT
, base_fmt
, NULL
)) {
6127 error_setg(errp
, "Backing file format not supported for file "
6128 "format '%s'", fmt
);
6133 backing_file
= qemu_opt_get(opts
, BLOCK_OPT_BACKING_FILE
);
6135 if (!strcmp(filename
, backing_file
)) {
6136 error_setg(errp
, "Error: Trying to create an image with the "
6137 "same filename as the backing file");
6140 if (backing_file
[0] == '\0') {
6141 error_setg(errp
, "Expected backing file name, got empty string");
6146 backing_fmt
= qemu_opt_get(opts
, BLOCK_OPT_BACKING_FMT
);
6148 /* The size for the image must always be specified, unless we have a backing
6149 * file and we have not been forbidden from opening it. */
6150 size
= qemu_opt_get_size(opts
, BLOCK_OPT_SIZE
, img_size
);
6151 if (backing_file
&& !(flags
& BDRV_O_NO_BACKING
)) {
6152 BlockDriverState
*bs
;
6155 QDict
*backing_options
= NULL
;
6158 bdrv_get_full_backing_filename_from_filename(filename
, backing_file
,
6163 assert(full_backing
);
6165 /* backing files always opened read-only */
6167 back_flags
&= ~(BDRV_O_RDWR
| BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
);
6169 backing_options
= qdict_new();
6171 qdict_put_str(backing_options
, "driver", backing_fmt
);
6173 qdict_put_bool(backing_options
, BDRV_OPT_FORCE_SHARE
, true);
6175 bs
= bdrv_open(full_backing
, NULL
, backing_options
, back_flags
,
6177 g_free(full_backing
);
6179 error_append_hint(&local_err
, "Could not open backing image.\n");
6183 warn_report("Deprecated use of backing file without explicit "
6184 "backing format (detected format of %s)",
6185 bs
->drv
->format_name
);
6186 if (bs
->drv
!= &bdrv_raw
) {
6188 * A probe of raw deserves the most attention:
6189 * leaving the backing format out of the image
6190 * will ensure bs->probed is set (ensuring we
6191 * don't accidentally commit into the backing
6192 * file), and allow more spots to warn the users
6193 * to fix their toolchain when opening this image
6194 * later. For other images, we can safely record
6195 * the format that we probed.
6197 backing_fmt
= bs
->drv
->format_name
;
6198 qemu_opt_set(opts
, BLOCK_OPT_BACKING_FMT
, backing_fmt
,
6203 /* Opened BS, have no size */
6204 size
= bdrv_getlength(bs
);
6206 error_setg_errno(errp
, -size
, "Could not get size of '%s'",
6211 qemu_opt_set_number(opts
, BLOCK_OPT_SIZE
, size
, &error_abort
);
6215 /* (backing_file && !(flags & BDRV_O_NO_BACKING)) */
6216 } else if (backing_file
&& !backing_fmt
) {
6217 warn_report("Deprecated use of unopened backing file without "
6218 "explicit backing format, use of this image requires "
6219 "potentially unsafe format probing");
6223 error_setg(errp
, "Image creation needs a size parameter");
6228 printf("Formatting '%s', fmt=%s ", filename
, fmt
);
6229 qemu_opts_print(opts
, " ");
6234 ret
= bdrv_create(drv
, filename
, opts
, &local_err
);
6236 if (ret
== -EFBIG
) {
6237 /* This is generally a better message than whatever the driver would
6238 * deliver (especially because of the cluster_size_hint), since that
6239 * is most probably not much different from "image too large". */
6240 const char *cluster_size_hint
= "";
6241 if (qemu_opt_get_size(opts
, BLOCK_OPT_CLUSTER_SIZE
, 0)) {
6242 cluster_size_hint
= " (try using a larger cluster size)";
6244 error_setg(errp
, "The image size is too large for file format '%s'"
6245 "%s", fmt
, cluster_size_hint
);
6246 error_free(local_err
);
6251 qemu_opts_del(opts
);
6252 qemu_opts_free(create_opts
);
6253 error_propagate(errp
, local_err
);
6256 AioContext
*bdrv_get_aio_context(BlockDriverState
*bs
)
6258 return bs
? bs
->aio_context
: qemu_get_aio_context();
6261 void bdrv_coroutine_enter(BlockDriverState
*bs
, Coroutine
*co
)
6263 aio_co_enter(bdrv_get_aio_context(bs
), co
);
6266 static void bdrv_do_remove_aio_context_notifier(BdrvAioNotifier
*ban
)
6268 QLIST_REMOVE(ban
, list
);
6272 static void bdrv_detach_aio_context(BlockDriverState
*bs
)
6274 BdrvAioNotifier
*baf
, *baf_tmp
;
6276 assert(!bs
->walking_aio_notifiers
);
6277 bs
->walking_aio_notifiers
= true;
6278 QLIST_FOREACH_SAFE(baf
, &bs
->aio_notifiers
, list
, baf_tmp
) {
6280 bdrv_do_remove_aio_context_notifier(baf
);
6282 baf
->detach_aio_context(baf
->opaque
);
6285 /* Never mind iterating again to check for ->deleted. bdrv_close() will
6286 * remove remaining aio notifiers if we aren't called again.
6288 bs
->walking_aio_notifiers
= false;
6290 if (bs
->drv
&& bs
->drv
->bdrv_detach_aio_context
) {
6291 bs
->drv
->bdrv_detach_aio_context(bs
);
6294 if (bs
->quiesce_counter
) {
6295 aio_enable_external(bs
->aio_context
);
6297 bs
->aio_context
= NULL
;
6300 static void bdrv_attach_aio_context(BlockDriverState
*bs
,
6301 AioContext
*new_context
)
6303 BdrvAioNotifier
*ban
, *ban_tmp
;
6305 if (bs
->quiesce_counter
) {
6306 aio_disable_external(new_context
);
6309 bs
->aio_context
= new_context
;
6311 if (bs
->drv
&& bs
->drv
->bdrv_attach_aio_context
) {
6312 bs
->drv
->bdrv_attach_aio_context(bs
, new_context
);
6315 assert(!bs
->walking_aio_notifiers
);
6316 bs
->walking_aio_notifiers
= true;
6317 QLIST_FOREACH_SAFE(ban
, &bs
->aio_notifiers
, list
, ban_tmp
) {
6319 bdrv_do_remove_aio_context_notifier(ban
);
6321 ban
->attached_aio_context(new_context
, ban
->opaque
);
6324 bs
->walking_aio_notifiers
= false;
6328 * Changes the AioContext used for fd handlers, timers, and BHs by this
6329 * BlockDriverState and all its children and parents.
6331 * Must be called from the main AioContext.
6333 * The caller must own the AioContext lock for the old AioContext of bs, but it
6334 * must not own the AioContext lock for new_context (unless new_context is the
6335 * same as the current context of bs).
6337 * @ignore will accumulate all visited BdrvChild object. The caller is
6338 * responsible for freeing the list afterwards.
6340 void bdrv_set_aio_context_ignore(BlockDriverState
*bs
,
6341 AioContext
*new_context
, GSList
**ignore
)
6343 AioContext
*old_context
= bdrv_get_aio_context(bs
);
6346 g_assert(qemu_get_current_aio_context() == qemu_get_aio_context());
6348 if (old_context
== new_context
) {
6352 bdrv_drained_begin(bs
);
6354 QLIST_FOREACH(child
, &bs
->children
, next
) {
6355 if (g_slist_find(*ignore
, child
)) {
6358 *ignore
= g_slist_prepend(*ignore
, child
);
6359 bdrv_set_aio_context_ignore(child
->bs
, new_context
, ignore
);
6361 QLIST_FOREACH(child
, &bs
->parents
, next_parent
) {
6362 if (g_slist_find(*ignore
, child
)) {
6365 assert(child
->klass
->set_aio_ctx
);
6366 *ignore
= g_slist_prepend(*ignore
, child
);
6367 child
->klass
->set_aio_ctx(child
, new_context
, ignore
);
6370 bdrv_detach_aio_context(bs
);
6372 /* Acquire the new context, if necessary */
6373 if (qemu_get_aio_context() != new_context
) {
6374 aio_context_acquire(new_context
);
6377 bdrv_attach_aio_context(bs
, new_context
);
6380 * If this function was recursively called from
6381 * bdrv_set_aio_context_ignore(), there may be nodes in the
6382 * subtree that have not yet been moved to the new AioContext.
6383 * Release the old one so bdrv_drained_end() can poll them.
6385 if (qemu_get_aio_context() != old_context
) {
6386 aio_context_release(old_context
);
6389 bdrv_drained_end(bs
);
6391 if (qemu_get_aio_context() != old_context
) {
6392 aio_context_acquire(old_context
);
6394 if (qemu_get_aio_context() != new_context
) {
6395 aio_context_release(new_context
);
6399 static bool bdrv_parent_can_set_aio_context(BdrvChild
*c
, AioContext
*ctx
,
6400 GSList
**ignore
, Error
**errp
)
6402 if (g_slist_find(*ignore
, c
)) {
6405 *ignore
= g_slist_prepend(*ignore
, c
);
6408 * A BdrvChildClass that doesn't handle AioContext changes cannot
6409 * tolerate any AioContext changes
6411 if (!c
->klass
->can_set_aio_ctx
) {
6412 char *user
= bdrv_child_user_desc(c
);
6413 error_setg(errp
, "Changing iothreads is not supported by %s", user
);
6417 if (!c
->klass
->can_set_aio_ctx(c
, ctx
, ignore
, errp
)) {
6418 assert(!errp
|| *errp
);
6424 bool bdrv_child_can_set_aio_context(BdrvChild
*c
, AioContext
*ctx
,
6425 GSList
**ignore
, Error
**errp
)
6427 if (g_slist_find(*ignore
, c
)) {
6430 *ignore
= g_slist_prepend(*ignore
, c
);
6431 return bdrv_can_set_aio_context(c
->bs
, ctx
, ignore
, errp
);
6434 /* @ignore will accumulate all visited BdrvChild object. The caller is
6435 * responsible for freeing the list afterwards. */
6436 bool bdrv_can_set_aio_context(BlockDriverState
*bs
, AioContext
*ctx
,
6437 GSList
**ignore
, Error
**errp
)
6441 if (bdrv_get_aio_context(bs
) == ctx
) {
6445 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
6446 if (!bdrv_parent_can_set_aio_context(c
, ctx
, ignore
, errp
)) {
6450 QLIST_FOREACH(c
, &bs
->children
, next
) {
6451 if (!bdrv_child_can_set_aio_context(c
, ctx
, ignore
, errp
)) {
6459 int bdrv_child_try_set_aio_context(BlockDriverState
*bs
, AioContext
*ctx
,
6460 BdrvChild
*ignore_child
, Error
**errp
)
6465 ignore
= ignore_child
? g_slist_prepend(NULL
, ignore_child
) : NULL
;
6466 ret
= bdrv_can_set_aio_context(bs
, ctx
, &ignore
, errp
);
6467 g_slist_free(ignore
);
6473 ignore
= ignore_child
? g_slist_prepend(NULL
, ignore_child
) : NULL
;
6474 bdrv_set_aio_context_ignore(bs
, ctx
, &ignore
);
6475 g_slist_free(ignore
);
6480 int bdrv_try_set_aio_context(BlockDriverState
*bs
, AioContext
*ctx
,
6483 return bdrv_child_try_set_aio_context(bs
, ctx
, NULL
, errp
);
6486 void bdrv_add_aio_context_notifier(BlockDriverState
*bs
,
6487 void (*attached_aio_context
)(AioContext
*new_context
, void *opaque
),
6488 void (*detach_aio_context
)(void *opaque
), void *opaque
)
6490 BdrvAioNotifier
*ban
= g_new(BdrvAioNotifier
, 1);
6491 *ban
= (BdrvAioNotifier
){
6492 .attached_aio_context
= attached_aio_context
,
6493 .detach_aio_context
= detach_aio_context
,
6497 QLIST_INSERT_HEAD(&bs
->aio_notifiers
, ban
, list
);
6500 void bdrv_remove_aio_context_notifier(BlockDriverState
*bs
,
6501 void (*attached_aio_context
)(AioContext
*,
6503 void (*detach_aio_context
)(void *),
6506 BdrvAioNotifier
*ban
, *ban_next
;
6508 QLIST_FOREACH_SAFE(ban
, &bs
->aio_notifiers
, list
, ban_next
) {
6509 if (ban
->attached_aio_context
== attached_aio_context
&&
6510 ban
->detach_aio_context
== detach_aio_context
&&
6511 ban
->opaque
== opaque
&&
6512 ban
->deleted
== false)
6514 if (bs
->walking_aio_notifiers
) {
6515 ban
->deleted
= true;
6517 bdrv_do_remove_aio_context_notifier(ban
);
6526 int bdrv_amend_options(BlockDriverState
*bs
, QemuOpts
*opts
,
6527 BlockDriverAmendStatusCB
*status_cb
, void *cb_opaque
,
6532 error_setg(errp
, "Node is ejected");
6535 if (!bs
->drv
->bdrv_amend_options
) {
6536 error_setg(errp
, "Block driver '%s' does not support option amendment",
6537 bs
->drv
->format_name
);
6540 return bs
->drv
->bdrv_amend_options(bs
, opts
, status_cb
,
6541 cb_opaque
, force
, errp
);
6545 * This function checks whether the given @to_replace is allowed to be
6546 * replaced by a node that always shows the same data as @bs. This is
6547 * used for example to verify whether the mirror job can replace
6548 * @to_replace by the target mirrored from @bs.
6549 * To be replaceable, @bs and @to_replace may either be guaranteed to
6550 * always show the same data (because they are only connected through
6551 * filters), or some driver may allow replacing one of its children
6552 * because it can guarantee that this child's data is not visible at
6553 * all (for example, for dissenting quorum children that have no other
6556 bool bdrv_recurse_can_replace(BlockDriverState
*bs
,
6557 BlockDriverState
*to_replace
)
6559 if (!bs
|| !bs
->drv
) {
6563 if (bs
== to_replace
) {
6567 /* See what the driver can do */
6568 if (bs
->drv
->bdrv_recurse_can_replace
) {
6569 return bs
->drv
->bdrv_recurse_can_replace(bs
, to_replace
);
6572 /* For filters without an own implementation, we can recurse on our own */
6573 if (bs
->drv
->is_filter
) {
6574 BdrvChild
*child
= bs
->file
?: bs
->backing
;
6575 return bdrv_recurse_can_replace(child
->bs
, to_replace
);
6583 * Check whether the given @node_name can be replaced by a node that
6584 * has the same data as @parent_bs. If so, return @node_name's BDS;
6587 * @node_name must be a (recursive) *child of @parent_bs (or this
6588 * function will return NULL).
6590 * The result (whether the node can be replaced or not) is only valid
6591 * for as long as no graph or permission changes occur.
6593 BlockDriverState
*check_to_replace_node(BlockDriverState
*parent_bs
,
6594 const char *node_name
, Error
**errp
)
6596 BlockDriverState
*to_replace_bs
= bdrv_find_node(node_name
);
6597 AioContext
*aio_context
;
6599 if (!to_replace_bs
) {
6600 error_setg(errp
, "Node name '%s' not found", node_name
);
6604 aio_context
= bdrv_get_aio_context(to_replace_bs
);
6605 aio_context_acquire(aio_context
);
6607 if (bdrv_op_is_blocked(to_replace_bs
, BLOCK_OP_TYPE_REPLACE
, errp
)) {
6608 to_replace_bs
= NULL
;
6612 /* We don't want arbitrary node of the BDS chain to be replaced only the top
6613 * most non filter in order to prevent data corruption.
6614 * Another benefit is that this tests exclude backing files which are
6615 * blocked by the backing blockers.
6617 if (!bdrv_recurse_can_replace(parent_bs
, to_replace_bs
)) {
6618 error_setg(errp
, "Cannot replace '%s' by a node mirrored from '%s', "
6619 "because it cannot be guaranteed that doing so would not "
6620 "lead to an abrupt change of visible data",
6621 node_name
, parent_bs
->node_name
);
6622 to_replace_bs
= NULL
;
6627 aio_context_release(aio_context
);
6628 return to_replace_bs
;
6632 * Iterates through the list of runtime option keys that are said to
6633 * be "strong" for a BDS. An option is called "strong" if it changes
6634 * a BDS's data. For example, the null block driver's "size" and
6635 * "read-zeroes" options are strong, but its "latency-ns" option is
6638 * If a key returned by this function ends with a dot, all options
6639 * starting with that prefix are strong.
6641 static const char *const *strong_options(BlockDriverState
*bs
,
6642 const char *const *curopt
)
6644 static const char *const global_options
[] = {
6645 "driver", "filename", NULL
6649 return &global_options
[0];
6653 if (curopt
== &global_options
[ARRAY_SIZE(global_options
) - 1] && bs
->drv
) {
6654 curopt
= bs
->drv
->strong_runtime_opts
;
6657 return (curopt
&& *curopt
) ? curopt
: NULL
;
6661 * Copies all strong runtime options from bs->options to the given
6662 * QDict. The set of strong option keys is determined by invoking
6665 * Returns true iff any strong option was present in bs->options (and
6666 * thus copied to the target QDict) with the exception of "filename"
6667 * and "driver". The caller is expected to use this value to decide
6668 * whether the existence of strong options prevents the generation of
6671 static bool append_strong_runtime_options(QDict
*d
, BlockDriverState
*bs
)
6673 bool found_any
= false;
6674 const char *const *option_name
= NULL
;
6680 while ((option_name
= strong_options(bs
, option_name
))) {
6681 bool option_given
= false;
6683 assert(strlen(*option_name
) > 0);
6684 if ((*option_name
)[strlen(*option_name
) - 1] != '.') {
6685 QObject
*entry
= qdict_get(bs
->options
, *option_name
);
6690 qdict_put_obj(d
, *option_name
, qobject_ref(entry
));
6691 option_given
= true;
6693 const QDictEntry
*entry
;
6694 for (entry
= qdict_first(bs
->options
); entry
;
6695 entry
= qdict_next(bs
->options
, entry
))
6697 if (strstart(qdict_entry_key(entry
), *option_name
, NULL
)) {
6698 qdict_put_obj(d
, qdict_entry_key(entry
),
6699 qobject_ref(qdict_entry_value(entry
)));
6700 option_given
= true;
6705 /* While "driver" and "filename" need to be included in a JSON filename,
6706 * their existence does not prohibit generation of a plain filename. */
6707 if (!found_any
&& option_given
&&
6708 strcmp(*option_name
, "driver") && strcmp(*option_name
, "filename"))
6714 if (!qdict_haskey(d
, "driver")) {
6715 /* Drivers created with bdrv_new_open_driver() may not have a
6716 * @driver option. Add it here. */
6717 qdict_put_str(d
, "driver", bs
->drv
->format_name
);
6723 /* Note: This function may return false positives; it may return true
6724 * even if opening the backing file specified by bs's image header
6725 * would result in exactly bs->backing. */
6726 static bool bdrv_backing_overridden(BlockDriverState
*bs
)
6729 return strcmp(bs
->auto_backing_file
,
6730 bs
->backing
->bs
->filename
);
6732 /* No backing BDS, so if the image header reports any backing
6733 * file, it must have been suppressed */
6734 return bs
->auto_backing_file
[0] != '\0';
6738 /* Updates the following BDS fields:
6739 * - exact_filename: A filename which may be used for opening a block device
6740 * which (mostly) equals the given BDS (even without any
6741 * other options; so reading and writing must return the same
6742 * results, but caching etc. may be different)
6743 * - full_open_options: Options which, when given when opening a block device
6744 * (without a filename), result in a BDS (mostly)
6745 * equalling the given one
6746 * - filename: If exact_filename is set, it is copied here. Otherwise,
6747 * full_open_options is converted to a JSON object, prefixed with
6748 * "json:" (for use through the JSON pseudo protocol) and put here.
6750 void bdrv_refresh_filename(BlockDriverState
*bs
)
6752 BlockDriver
*drv
= bs
->drv
;
6755 bool backing_overridden
;
6756 bool generate_json_filename
; /* Whether our default implementation should
6757 fill exact_filename (false) or not (true) */
6763 /* This BDS's file name may depend on any of its children's file names, so
6764 * refresh those first */
6765 QLIST_FOREACH(child
, &bs
->children
, next
) {
6766 bdrv_refresh_filename(child
->bs
);
6770 /* For implicit nodes, just copy everything from the single child */
6771 child
= QLIST_FIRST(&bs
->children
);
6772 assert(QLIST_NEXT(child
, next
) == NULL
);
6774 pstrcpy(bs
->exact_filename
, sizeof(bs
->exact_filename
),
6775 child
->bs
->exact_filename
);
6776 pstrcpy(bs
->filename
, sizeof(bs
->filename
), child
->bs
->filename
);
6778 qobject_unref(bs
->full_open_options
);
6779 bs
->full_open_options
= qobject_ref(child
->bs
->full_open_options
);
6784 backing_overridden
= bdrv_backing_overridden(bs
);
6786 if (bs
->open_flags
& BDRV_O_NO_IO
) {
6787 /* Without I/O, the backing file does not change anything.
6788 * Therefore, in such a case (primarily qemu-img), we can
6789 * pretend the backing file has not been overridden even if
6790 * it technically has been. */
6791 backing_overridden
= false;
6794 /* Gather the options QDict */
6796 generate_json_filename
= append_strong_runtime_options(opts
, bs
);
6797 generate_json_filename
|= backing_overridden
;
6799 if (drv
->bdrv_gather_child_options
) {
6800 /* Some block drivers may not want to present all of their children's
6801 * options, or name them differently from BdrvChild.name */
6802 drv
->bdrv_gather_child_options(bs
, opts
, backing_overridden
);
6804 QLIST_FOREACH(child
, &bs
->children
, next
) {
6805 if (child
== bs
->backing
&& !backing_overridden
) {
6806 /* We can skip the backing BDS if it has not been overridden */
6810 qdict_put(opts
, child
->name
,
6811 qobject_ref(child
->bs
->full_open_options
));
6814 if (backing_overridden
&& !bs
->backing
) {
6815 /* Force no backing file */
6816 qdict_put_null(opts
, "backing");
6820 qobject_unref(bs
->full_open_options
);
6821 bs
->full_open_options
= opts
;
6823 if (drv
->bdrv_refresh_filename
) {
6824 /* Obsolete information is of no use here, so drop the old file name
6825 * information before refreshing it */
6826 bs
->exact_filename
[0] = '\0';
6828 drv
->bdrv_refresh_filename(bs
);
6829 } else if (bs
->file
) {
6830 /* Try to reconstruct valid information from the underlying file */
6832 bs
->exact_filename
[0] = '\0';
6835 * We can use the underlying file's filename if:
6836 * - it has a filename,
6837 * - the file is a protocol BDS, and
6838 * - opening that file (as this BDS's format) will automatically create
6839 * the BDS tree we have right now, that is:
6840 * - the user did not significantly change this BDS's behavior with
6841 * some explicit (strong) options
6842 * - no non-file child of this BDS has been overridden by the user
6843 * Both of these conditions are represented by generate_json_filename.
6845 if (bs
->file
->bs
->exact_filename
[0] &&
6846 bs
->file
->bs
->drv
->bdrv_file_open
&&
6847 !generate_json_filename
)
6849 strcpy(bs
->exact_filename
, bs
->file
->bs
->exact_filename
);
6853 if (bs
->exact_filename
[0]) {
6854 pstrcpy(bs
->filename
, sizeof(bs
->filename
), bs
->exact_filename
);
6856 QString
*json
= qobject_to_json(QOBJECT(bs
->full_open_options
));
6857 if (snprintf(bs
->filename
, sizeof(bs
->filename
), "json:%s",
6858 qstring_get_str(json
)) >= sizeof(bs
->filename
)) {
6859 /* Give user a hint if we truncated things. */
6860 strcpy(bs
->filename
+ sizeof(bs
->filename
) - 4, "...");
6862 qobject_unref(json
);
6866 char *bdrv_dirname(BlockDriverState
*bs
, Error
**errp
)
6868 BlockDriver
*drv
= bs
->drv
;
6871 error_setg(errp
, "Node '%s' is ejected", bs
->node_name
);
6875 if (drv
->bdrv_dirname
) {
6876 return drv
->bdrv_dirname(bs
, errp
);
6880 return bdrv_dirname(bs
->file
->bs
, errp
);
6883 bdrv_refresh_filename(bs
);
6884 if (bs
->exact_filename
[0] != '\0') {
6885 return path_combine(bs
->exact_filename
, "");
6888 error_setg(errp
, "Cannot generate a base directory for %s nodes",
6894 * Hot add/remove a BDS's child. So the user can take a child offline when
6895 * it is broken and take a new child online
6897 void bdrv_add_child(BlockDriverState
*parent_bs
, BlockDriverState
*child_bs
,
6901 if (!parent_bs
->drv
|| !parent_bs
->drv
->bdrv_add_child
) {
6902 error_setg(errp
, "The node %s does not support adding a child",
6903 bdrv_get_device_or_node_name(parent_bs
));
6907 if (!QLIST_EMPTY(&child_bs
->parents
)) {
6908 error_setg(errp
, "The node %s already has a parent",
6909 child_bs
->node_name
);
6913 parent_bs
->drv
->bdrv_add_child(parent_bs
, child_bs
, errp
);
6916 void bdrv_del_child(BlockDriverState
*parent_bs
, BdrvChild
*child
, Error
**errp
)
6920 if (!parent_bs
->drv
|| !parent_bs
->drv
->bdrv_del_child
) {
6921 error_setg(errp
, "The node %s does not support removing a child",
6922 bdrv_get_device_or_node_name(parent_bs
));
6926 QLIST_FOREACH(tmp
, &parent_bs
->children
, next
) {
6933 error_setg(errp
, "The node %s does not have a child named %s",
6934 bdrv_get_device_or_node_name(parent_bs
),
6935 bdrv_get_device_or_node_name(child
->bs
));
6939 parent_bs
->drv
->bdrv_del_child(parent_bs
, child
, errp
);
6942 int bdrv_make_empty(BdrvChild
*c
, Error
**errp
)
6944 BlockDriver
*drv
= c
->bs
->drv
;
6947 assert(c
->perm
& (BLK_PERM_WRITE
| BLK_PERM_WRITE_UNCHANGED
));
6949 if (!drv
->bdrv_make_empty
) {
6950 error_setg(errp
, "%s does not support emptying nodes",
6955 ret
= drv
->bdrv_make_empty(c
->bs
);
6957 error_setg_errno(errp
, -ret
, "Failed to empty %s",
6966 * Return the child that @bs acts as an overlay for, and from which data may be
6967 * copied in COW or COR operations. Usually this is the backing file.
6969 BdrvChild
*bdrv_cow_child(BlockDriverState
*bs
)
6971 if (!bs
|| !bs
->drv
) {
6975 if (bs
->drv
->is_filter
) {
6983 assert(bs
->backing
->role
& BDRV_CHILD_COW
);
6988 * If @bs acts as a filter for exactly one of its children, return
6991 BdrvChild
*bdrv_filter_child(BlockDriverState
*bs
)
6995 if (!bs
|| !bs
->drv
) {
6999 if (!bs
->drv
->is_filter
) {
7003 /* Only one of @backing or @file may be used */
7004 assert(!(bs
->backing
&& bs
->file
));
7006 c
= bs
->backing
?: bs
->file
;
7011 assert(c
->role
& BDRV_CHILD_FILTERED
);
7016 * Return either the result of bdrv_cow_child() or bdrv_filter_child(),
7017 * whichever is non-NULL.
7019 * Return NULL if both are NULL.
7021 BdrvChild
*bdrv_filter_or_cow_child(BlockDriverState
*bs
)
7023 BdrvChild
*cow_child
= bdrv_cow_child(bs
);
7024 BdrvChild
*filter_child
= bdrv_filter_child(bs
);
7026 /* Filter nodes cannot have COW backing files */
7027 assert(!(cow_child
&& filter_child
));
7029 return cow_child
?: filter_child
;
7033 * Return the primary child of this node: For filters, that is the
7034 * filtered child. For other nodes, that is usually the child storing
7036 * (A generally more helpful description is that this is (usually) the
7037 * child that has the same filename as @bs.)
7039 * Drivers do not necessarily have a primary child; for example quorum
7042 BdrvChild
*bdrv_primary_child(BlockDriverState
*bs
)
7044 BdrvChild
*c
, *found
= NULL
;
7046 QLIST_FOREACH(c
, &bs
->children
, next
) {
7047 if (c
->role
& BDRV_CHILD_PRIMARY
) {
7056 static BlockDriverState
*bdrv_do_skip_filters(BlockDriverState
*bs
,
7057 bool stop_on_explicit_filter
)
7065 while (!(stop_on_explicit_filter
&& !bs
->implicit
)) {
7066 c
= bdrv_filter_child(bs
);
7069 * A filter that is embedded in a working block graph must
7070 * have a child. Assert this here so this function does
7071 * not return a filter node that is not expected by the
7074 assert(!bs
->drv
|| !bs
->drv
->is_filter
);
7080 * Note that this treats nodes with bs->drv == NULL as not being
7081 * filters (bs->drv == NULL should be replaced by something else
7083 * The advantage of this behavior is that this function will thus
7084 * always return a non-NULL value (given a non-NULL @bs).
7091 * Return the first BDS that has not been added implicitly or that
7092 * does not have a filtered child down the chain starting from @bs
7093 * (including @bs itself).
7095 BlockDriverState
*bdrv_skip_implicit_filters(BlockDriverState
*bs
)
7097 return bdrv_do_skip_filters(bs
, true);
7101 * Return the first BDS that does not have a filtered child down the
7102 * chain starting from @bs (including @bs itself).
7104 BlockDriverState
*bdrv_skip_filters(BlockDriverState
*bs
)
7106 return bdrv_do_skip_filters(bs
, false);
7110 * For a backing chain, return the first non-filter backing image of
7111 * the first non-filter image.
7113 BlockDriverState
*bdrv_backing_chain_next(BlockDriverState
*bs
)
7115 return bdrv_skip_filters(bdrv_cow_bs(bdrv_skip_filters(bs
)));