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
;
715 BlockDriverState
*filtered
= bdrv_filter_bs(bs
);
717 if (drv
&& drv
->bdrv_probe_blocksizes
) {
718 return drv
->bdrv_probe_blocksizes(bs
, bsz
);
719 } else if (filtered
) {
720 return bdrv_probe_blocksizes(filtered
, bsz
);
727 * Try to get @bs's geometry (cyls, heads, sectors).
728 * On success, store them in @geo struct and return 0.
729 * On failure return -errno.
730 * @bs must not be empty.
732 int bdrv_probe_geometry(BlockDriverState
*bs
, HDGeometry
*geo
)
734 BlockDriver
*drv
= bs
->drv
;
735 BlockDriverState
*filtered
= bdrv_filter_bs(bs
);
737 if (drv
&& drv
->bdrv_probe_geometry
) {
738 return drv
->bdrv_probe_geometry(bs
, geo
);
739 } else if (filtered
) {
740 return bdrv_probe_geometry(filtered
, geo
);
747 * Create a uniquely-named empty temporary file.
748 * Return 0 upon success, otherwise a negative errno value.
750 int get_tmp_filename(char *filename
, int size
)
753 char temp_dir
[MAX_PATH
];
754 /* GetTempFileName requires that its output buffer (4th param)
755 have length MAX_PATH or greater. */
756 assert(size
>= MAX_PATH
);
757 return (GetTempPath(MAX_PATH
, temp_dir
)
758 && GetTempFileName(temp_dir
, "qem", 0, filename
)
759 ? 0 : -GetLastError());
763 tmpdir
= getenv("TMPDIR");
767 if (snprintf(filename
, size
, "%s/vl.XXXXXX", tmpdir
) >= size
) {
770 fd
= mkstemp(filename
);
774 if (close(fd
) != 0) {
783 * Detect host devices. By convention, /dev/cdrom[N] is always
784 * recognized as a host CDROM.
786 static BlockDriver
*find_hdev_driver(const char *filename
)
788 int score_max
= 0, score
;
789 BlockDriver
*drv
= NULL
, *d
;
791 QLIST_FOREACH(d
, &bdrv_drivers
, list
) {
792 if (d
->bdrv_probe_device
) {
793 score
= d
->bdrv_probe_device(filename
);
794 if (score
> score_max
) {
804 static BlockDriver
*bdrv_do_find_protocol(const char *protocol
)
808 QLIST_FOREACH(drv1
, &bdrv_drivers
, list
) {
809 if (drv1
->protocol_name
&& !strcmp(drv1
->protocol_name
, protocol
)) {
817 BlockDriver
*bdrv_find_protocol(const char *filename
,
818 bool allow_protocol_prefix
,
827 /* TODO Drivers without bdrv_file_open must be specified explicitly */
830 * XXX(hch): we really should not let host device detection
831 * override an explicit protocol specification, but moving this
832 * later breaks access to device names with colons in them.
833 * Thanks to the brain-dead persistent naming schemes on udev-
834 * based Linux systems those actually are quite common.
836 drv1
= find_hdev_driver(filename
);
841 if (!path_has_protocol(filename
) || !allow_protocol_prefix
) {
845 p
= strchr(filename
, ':');
848 if (len
> sizeof(protocol
) - 1)
849 len
= sizeof(protocol
) - 1;
850 memcpy(protocol
, filename
, len
);
851 protocol
[len
] = '\0';
853 drv1
= bdrv_do_find_protocol(protocol
);
858 for (i
= 0; i
< (int)ARRAY_SIZE(block_driver_modules
); ++i
) {
859 if (block_driver_modules
[i
].protocol_name
&&
860 !strcmp(block_driver_modules
[i
].protocol_name
, protocol
)) {
861 block_module_load_one(block_driver_modules
[i
].library_name
);
866 drv1
= bdrv_do_find_protocol(protocol
);
868 error_setg(errp
, "Unknown protocol '%s'", protocol
);
874 * Guess image format by probing its contents.
875 * This is not a good idea when your image is raw (CVE-2008-2004), but
876 * we do it anyway for backward compatibility.
878 * @buf contains the image's first @buf_size bytes.
879 * @buf_size is the buffer size in bytes (generally BLOCK_PROBE_BUF_SIZE,
880 * but can be smaller if the image file is smaller)
881 * @filename is its filename.
883 * For all block drivers, call the bdrv_probe() method to get its
885 * Return the first block driver with the highest probing score.
887 BlockDriver
*bdrv_probe_all(const uint8_t *buf
, int buf_size
,
888 const char *filename
)
890 int score_max
= 0, score
;
891 BlockDriver
*drv
= NULL
, *d
;
893 QLIST_FOREACH(d
, &bdrv_drivers
, list
) {
895 score
= d
->bdrv_probe(buf
, buf_size
, filename
);
896 if (score
> score_max
) {
906 static int find_image_format(BlockBackend
*file
, const char *filename
,
907 BlockDriver
**pdrv
, Error
**errp
)
910 uint8_t buf
[BLOCK_PROBE_BUF_SIZE
];
913 /* Return the raw BlockDriver * to scsi-generic devices or empty drives */
914 if (blk_is_sg(file
) || !blk_is_inserted(file
) || blk_getlength(file
) == 0) {
919 ret
= blk_pread(file
, 0, buf
, sizeof(buf
));
921 error_setg_errno(errp
, -ret
, "Could not read image for determining its "
927 drv
= bdrv_probe_all(buf
, ret
, filename
);
929 error_setg(errp
, "Could not determine image format: No compatible "
938 * Set the current 'total_sectors' value
939 * Return 0 on success, -errno on error.
941 int refresh_total_sectors(BlockDriverState
*bs
, int64_t hint
)
943 BlockDriver
*drv
= bs
->drv
;
949 /* Do not attempt drv->bdrv_getlength() on scsi-generic devices */
953 /* query actual device if possible, otherwise just trust the hint */
954 if (drv
->bdrv_getlength
) {
955 int64_t length
= drv
->bdrv_getlength(bs
);
959 hint
= DIV_ROUND_UP(length
, BDRV_SECTOR_SIZE
);
962 bs
->total_sectors
= hint
;
967 * Combines a QDict of new block driver @options with any missing options taken
968 * from @old_options, so that leaving out an option defaults to its old value.
970 static void bdrv_join_options(BlockDriverState
*bs
, QDict
*options
,
973 if (bs
->drv
&& bs
->drv
->bdrv_join_options
) {
974 bs
->drv
->bdrv_join_options(options
, old_options
);
976 qdict_join(options
, old_options
, false);
980 static BlockdevDetectZeroesOptions
bdrv_parse_detect_zeroes(QemuOpts
*opts
,
984 Error
*local_err
= NULL
;
985 char *value
= qemu_opt_get_del(opts
, "detect-zeroes");
986 BlockdevDetectZeroesOptions detect_zeroes
=
987 qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup
, value
,
988 BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF
, &local_err
);
991 error_propagate(errp
, local_err
);
992 return detect_zeroes
;
995 if (detect_zeroes
== BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP
&&
996 !(open_flags
& BDRV_O_UNMAP
))
998 error_setg(errp
, "setting detect-zeroes to unmap is not allowed "
999 "without setting discard operation to unmap");
1002 return detect_zeroes
;
1006 * Set open flags for aio engine
1008 * Return 0 on success, -1 if the engine specified is invalid
1010 int bdrv_parse_aio(const char *mode
, int *flags
)
1012 if (!strcmp(mode
, "threads")) {
1013 /* do nothing, default */
1014 } else if (!strcmp(mode
, "native")) {
1015 *flags
|= BDRV_O_NATIVE_AIO
;
1016 #ifdef CONFIG_LINUX_IO_URING
1017 } else if (!strcmp(mode
, "io_uring")) {
1018 *flags
|= BDRV_O_IO_URING
;
1028 * Set open flags for a given discard mode
1030 * Return 0 on success, -1 if the discard mode was invalid.
1032 int bdrv_parse_discard_flags(const char *mode
, int *flags
)
1034 *flags
&= ~BDRV_O_UNMAP
;
1036 if (!strcmp(mode
, "off") || !strcmp(mode
, "ignore")) {
1038 } else if (!strcmp(mode
, "on") || !strcmp(mode
, "unmap")) {
1039 *flags
|= BDRV_O_UNMAP
;
1048 * Set open flags for a given cache mode
1050 * Return 0 on success, -1 if the cache mode was invalid.
1052 int bdrv_parse_cache_mode(const char *mode
, int *flags
, bool *writethrough
)
1054 *flags
&= ~BDRV_O_CACHE_MASK
;
1056 if (!strcmp(mode
, "off") || !strcmp(mode
, "none")) {
1057 *writethrough
= false;
1058 *flags
|= BDRV_O_NOCACHE
;
1059 } else if (!strcmp(mode
, "directsync")) {
1060 *writethrough
= true;
1061 *flags
|= BDRV_O_NOCACHE
;
1062 } else if (!strcmp(mode
, "writeback")) {
1063 *writethrough
= false;
1064 } else if (!strcmp(mode
, "unsafe")) {
1065 *writethrough
= false;
1066 *flags
|= BDRV_O_NO_FLUSH
;
1067 } else if (!strcmp(mode
, "writethrough")) {
1068 *writethrough
= true;
1076 static char *bdrv_child_get_parent_desc(BdrvChild
*c
)
1078 BlockDriverState
*parent
= c
->opaque
;
1079 return g_strdup(bdrv_get_device_or_node_name(parent
));
1082 static void bdrv_child_cb_drained_begin(BdrvChild
*child
)
1084 BlockDriverState
*bs
= child
->opaque
;
1085 bdrv_do_drained_begin_quiesce(bs
, NULL
, false);
1088 static bool bdrv_child_cb_drained_poll(BdrvChild
*child
)
1090 BlockDriverState
*bs
= child
->opaque
;
1091 return bdrv_drain_poll(bs
, false, NULL
, false);
1094 static void bdrv_child_cb_drained_end(BdrvChild
*child
,
1095 int *drained_end_counter
)
1097 BlockDriverState
*bs
= child
->opaque
;
1098 bdrv_drained_end_no_poll(bs
, drained_end_counter
);
1101 static int bdrv_child_cb_inactivate(BdrvChild
*child
)
1103 BlockDriverState
*bs
= child
->opaque
;
1104 assert(bs
->open_flags
& BDRV_O_INACTIVE
);
1108 static bool bdrv_child_cb_can_set_aio_ctx(BdrvChild
*child
, AioContext
*ctx
,
1109 GSList
**ignore
, Error
**errp
)
1111 BlockDriverState
*bs
= child
->opaque
;
1112 return bdrv_can_set_aio_context(bs
, ctx
, ignore
, errp
);
1115 static void bdrv_child_cb_set_aio_ctx(BdrvChild
*child
, AioContext
*ctx
,
1118 BlockDriverState
*bs
= child
->opaque
;
1119 return bdrv_set_aio_context_ignore(bs
, ctx
, ignore
);
1123 * Returns the options and flags that a temporary snapshot should get, based on
1124 * the originally requested flags (the originally requested image will have
1125 * flags like a backing file)
1127 static void bdrv_temp_snapshot_options(int *child_flags
, QDict
*child_options
,
1128 int parent_flags
, QDict
*parent_options
)
1130 *child_flags
= (parent_flags
& ~BDRV_O_SNAPSHOT
) | BDRV_O_TEMPORARY
;
1132 /* For temporary files, unconditional cache=unsafe is fine */
1133 qdict_set_default_str(child_options
, BDRV_OPT_CACHE_DIRECT
, "off");
1134 qdict_set_default_str(child_options
, BDRV_OPT_CACHE_NO_FLUSH
, "on");
1136 /* Copy the read-only and discard options from the parent */
1137 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_READ_ONLY
);
1138 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_DISCARD
);
1140 /* aio=native doesn't work for cache.direct=off, so disable it for the
1141 * temporary snapshot */
1142 *child_flags
&= ~BDRV_O_NATIVE_AIO
;
1145 static void bdrv_backing_attach(BdrvChild
*c
)
1147 BlockDriverState
*parent
= c
->opaque
;
1148 BlockDriverState
*backing_hd
= c
->bs
;
1150 assert(!parent
->backing_blocker
);
1151 error_setg(&parent
->backing_blocker
,
1152 "node is used as backing hd of '%s'",
1153 bdrv_get_device_or_node_name(parent
));
1155 bdrv_refresh_filename(backing_hd
);
1157 parent
->open_flags
&= ~BDRV_O_NO_BACKING
;
1158 pstrcpy(parent
->backing_file
, sizeof(parent
->backing_file
),
1159 backing_hd
->filename
);
1160 pstrcpy(parent
->backing_format
, sizeof(parent
->backing_format
),
1161 backing_hd
->drv
? backing_hd
->drv
->format_name
: "");
1163 bdrv_op_block_all(backing_hd
, parent
->backing_blocker
);
1164 /* Otherwise we won't be able to commit or stream */
1165 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_COMMIT_TARGET
,
1166 parent
->backing_blocker
);
1167 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_STREAM
,
1168 parent
->backing_blocker
);
1170 * We do backup in 3 ways:
1172 * The target bs is new opened, and the source is top BDS
1173 * 2. blockdev backup
1174 * Both the source and the target are top BDSes.
1175 * 3. internal backup(used for block replication)
1176 * Both the source and the target are backing file
1178 * In case 1 and 2, neither the source nor the target is the backing file.
1179 * In case 3, we will block the top BDS, so there is only one block job
1180 * for the top BDS and its backing chain.
1182 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_BACKUP_SOURCE
,
1183 parent
->backing_blocker
);
1184 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_BACKUP_TARGET
,
1185 parent
->backing_blocker
);
1188 static void bdrv_backing_detach(BdrvChild
*c
)
1190 BlockDriverState
*parent
= c
->opaque
;
1192 assert(parent
->backing_blocker
);
1193 bdrv_op_unblock_all(c
->bs
, parent
->backing_blocker
);
1194 error_free(parent
->backing_blocker
);
1195 parent
->backing_blocker
= NULL
;
1198 static int bdrv_backing_update_filename(BdrvChild
*c
, BlockDriverState
*base
,
1199 const char *filename
, Error
**errp
)
1201 BlockDriverState
*parent
= c
->opaque
;
1202 bool read_only
= bdrv_is_read_only(parent
);
1206 ret
= bdrv_reopen_set_read_only(parent
, false, errp
);
1212 ret
= bdrv_change_backing_file(parent
, filename
,
1213 base
->drv
? base
->drv
->format_name
: "",
1216 error_setg_errno(errp
, -ret
, "Could not update backing file link");
1220 bdrv_reopen_set_read_only(parent
, true, NULL
);
1227 * Returns the options and flags that a generic child of a BDS should
1228 * get, based on the given options and flags for the parent BDS.
1230 static void bdrv_inherited_options(BdrvChildRole role
, bool parent_is_format
,
1231 int *child_flags
, QDict
*child_options
,
1232 int parent_flags
, QDict
*parent_options
)
1234 int flags
= parent_flags
;
1237 * First, decide whether to set, clear, or leave BDRV_O_PROTOCOL.
1238 * Generally, the question to answer is: Should this child be
1239 * format-probed by default?
1243 * Pure and non-filtered data children of non-format nodes should
1244 * be probed by default (even when the node itself has BDRV_O_PROTOCOL
1245 * set). This only affects a very limited set of drivers (namely
1246 * quorum and blkverify when this comment was written).
1247 * Force-clear BDRV_O_PROTOCOL then.
1249 if (!parent_is_format
&&
1250 (role
& BDRV_CHILD_DATA
) &&
1251 !(role
& (BDRV_CHILD_METADATA
| BDRV_CHILD_FILTERED
)))
1253 flags
&= ~BDRV_O_PROTOCOL
;
1257 * All children of format nodes (except for COW children) and all
1258 * metadata children in general should never be format-probed.
1259 * Force-set BDRV_O_PROTOCOL then.
1261 if ((parent_is_format
&& !(role
& BDRV_CHILD_COW
)) ||
1262 (role
& BDRV_CHILD_METADATA
))
1264 flags
|= BDRV_O_PROTOCOL
;
1268 * If the cache mode isn't explicitly set, inherit direct and no-flush from
1271 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_CACHE_DIRECT
);
1272 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_CACHE_NO_FLUSH
);
1273 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_FORCE_SHARE
);
1275 if (role
& BDRV_CHILD_COW
) {
1276 /* backing files are opened read-only by default */
1277 qdict_set_default_str(child_options
, BDRV_OPT_READ_ONLY
, "on");
1278 qdict_set_default_str(child_options
, BDRV_OPT_AUTO_READ_ONLY
, "off");
1280 /* Inherit the read-only option from the parent if it's not set */
1281 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_READ_ONLY
);
1282 qdict_copy_default(child_options
, parent_options
,
1283 BDRV_OPT_AUTO_READ_ONLY
);
1287 * bdrv_co_pdiscard() respects unmap policy for the parent, so we
1288 * can default to enable it on lower layers regardless of the
1291 qdict_set_default_str(child_options
, BDRV_OPT_DISCARD
, "unmap");
1293 /* Clear flags that only apply to the top layer */
1294 flags
&= ~(BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
| BDRV_O_COPY_ON_READ
);
1296 if (role
& BDRV_CHILD_METADATA
) {
1297 flags
&= ~BDRV_O_NO_IO
;
1299 if (role
& BDRV_CHILD_COW
) {
1300 flags
&= ~BDRV_O_TEMPORARY
;
1303 *child_flags
= flags
;
1306 static void bdrv_child_cb_attach(BdrvChild
*child
)
1308 BlockDriverState
*bs
= child
->opaque
;
1310 if (child
->role
& BDRV_CHILD_COW
) {
1311 bdrv_backing_attach(child
);
1314 bdrv_apply_subtree_drain(child
, bs
);
1317 static void bdrv_child_cb_detach(BdrvChild
*child
)
1319 BlockDriverState
*bs
= child
->opaque
;
1321 if (child
->role
& BDRV_CHILD_COW
) {
1322 bdrv_backing_detach(child
);
1325 bdrv_unapply_subtree_drain(child
, bs
);
1328 static int bdrv_child_cb_update_filename(BdrvChild
*c
, BlockDriverState
*base
,
1329 const char *filename
, Error
**errp
)
1331 if (c
->role
& BDRV_CHILD_COW
) {
1332 return bdrv_backing_update_filename(c
, base
, filename
, errp
);
1337 const BdrvChildClass child_of_bds
= {
1338 .parent_is_bds
= true,
1339 .get_parent_desc
= bdrv_child_get_parent_desc
,
1340 .inherit_options
= bdrv_inherited_options
,
1341 .drained_begin
= bdrv_child_cb_drained_begin
,
1342 .drained_poll
= bdrv_child_cb_drained_poll
,
1343 .drained_end
= bdrv_child_cb_drained_end
,
1344 .attach
= bdrv_child_cb_attach
,
1345 .detach
= bdrv_child_cb_detach
,
1346 .inactivate
= bdrv_child_cb_inactivate
,
1347 .can_set_aio_ctx
= bdrv_child_cb_can_set_aio_ctx
,
1348 .set_aio_ctx
= bdrv_child_cb_set_aio_ctx
,
1349 .update_filename
= bdrv_child_cb_update_filename
,
1352 static int bdrv_open_flags(BlockDriverState
*bs
, int flags
)
1354 int open_flags
= flags
;
1357 * Clear flags that are internal to the block layer before opening the
1360 open_flags
&= ~(BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
| BDRV_O_PROTOCOL
);
1365 static void update_flags_from_options(int *flags
, QemuOpts
*opts
)
1367 *flags
&= ~(BDRV_O_CACHE_MASK
| BDRV_O_RDWR
| BDRV_O_AUTO_RDONLY
);
1369 if (qemu_opt_get_bool_del(opts
, BDRV_OPT_CACHE_NO_FLUSH
, false)) {
1370 *flags
|= BDRV_O_NO_FLUSH
;
1373 if (qemu_opt_get_bool_del(opts
, BDRV_OPT_CACHE_DIRECT
, false)) {
1374 *flags
|= BDRV_O_NOCACHE
;
1377 if (!qemu_opt_get_bool_del(opts
, BDRV_OPT_READ_ONLY
, false)) {
1378 *flags
|= BDRV_O_RDWR
;
1381 if (qemu_opt_get_bool_del(opts
, BDRV_OPT_AUTO_READ_ONLY
, false)) {
1382 *flags
|= BDRV_O_AUTO_RDONLY
;
1386 static void update_options_from_flags(QDict
*options
, int flags
)
1388 if (!qdict_haskey(options
, BDRV_OPT_CACHE_DIRECT
)) {
1389 qdict_put_bool(options
, BDRV_OPT_CACHE_DIRECT
, flags
& BDRV_O_NOCACHE
);
1391 if (!qdict_haskey(options
, BDRV_OPT_CACHE_NO_FLUSH
)) {
1392 qdict_put_bool(options
, BDRV_OPT_CACHE_NO_FLUSH
,
1393 flags
& BDRV_O_NO_FLUSH
);
1395 if (!qdict_haskey(options
, BDRV_OPT_READ_ONLY
)) {
1396 qdict_put_bool(options
, BDRV_OPT_READ_ONLY
, !(flags
& BDRV_O_RDWR
));
1398 if (!qdict_haskey(options
, BDRV_OPT_AUTO_READ_ONLY
)) {
1399 qdict_put_bool(options
, BDRV_OPT_AUTO_READ_ONLY
,
1400 flags
& BDRV_O_AUTO_RDONLY
);
1404 static void bdrv_assign_node_name(BlockDriverState
*bs
,
1405 const char *node_name
,
1408 char *gen_node_name
= NULL
;
1411 node_name
= gen_node_name
= id_generate(ID_BLOCK
);
1412 } else if (!id_wellformed(node_name
)) {
1414 * Check for empty string or invalid characters, but not if it is
1415 * generated (generated names use characters not available to the user)
1417 error_setg(errp
, "Invalid node name");
1421 /* takes care of avoiding namespaces collisions */
1422 if (blk_by_name(node_name
)) {
1423 error_setg(errp
, "node-name=%s is conflicting with a device id",
1428 /* takes care of avoiding duplicates node names */
1429 if (bdrv_find_node(node_name
)) {
1430 error_setg(errp
, "Duplicate node name");
1434 /* Make sure that the node name isn't truncated */
1435 if (strlen(node_name
) >= sizeof(bs
->node_name
)) {
1436 error_setg(errp
, "Node name too long");
1440 /* copy node name into the bs and insert it into the graph list */
1441 pstrcpy(bs
->node_name
, sizeof(bs
->node_name
), node_name
);
1442 QTAILQ_INSERT_TAIL(&graph_bdrv_states
, bs
, node_list
);
1444 g_free(gen_node_name
);
1447 static int bdrv_open_driver(BlockDriverState
*bs
, BlockDriver
*drv
,
1448 const char *node_name
, QDict
*options
,
1449 int open_flags
, Error
**errp
)
1451 Error
*local_err
= NULL
;
1454 bdrv_assign_node_name(bs
, node_name
, &local_err
);
1456 error_propagate(errp
, local_err
);
1461 bs
->read_only
= !(bs
->open_flags
& BDRV_O_RDWR
);
1462 bs
->opaque
= g_malloc0(drv
->instance_size
);
1464 if (drv
->bdrv_file_open
) {
1465 assert(!drv
->bdrv_needs_filename
|| bs
->filename
[0]);
1466 ret
= drv
->bdrv_file_open(bs
, options
, open_flags
, &local_err
);
1467 } else if (drv
->bdrv_open
) {
1468 ret
= drv
->bdrv_open(bs
, options
, open_flags
, &local_err
);
1475 error_propagate(errp
, local_err
);
1476 } else if (bs
->filename
[0]) {
1477 error_setg_errno(errp
, -ret
, "Could not open '%s'", bs
->filename
);
1479 error_setg_errno(errp
, -ret
, "Could not open image");
1484 ret
= refresh_total_sectors(bs
, bs
->total_sectors
);
1486 error_setg_errno(errp
, -ret
, "Could not refresh total sector count");
1490 bdrv_refresh_limits(bs
, &local_err
);
1492 error_propagate(errp
, local_err
);
1496 assert(bdrv_opt_mem_align(bs
) != 0);
1497 assert(bdrv_min_mem_align(bs
) != 0);
1498 assert(is_power_of_2(bs
->bl
.request_alignment
));
1500 for (i
= 0; i
< bs
->quiesce_counter
; i
++) {
1501 if (drv
->bdrv_co_drain_begin
) {
1502 drv
->bdrv_co_drain_begin(bs
);
1509 if (bs
->file
!= NULL
) {
1510 bdrv_unref_child(bs
, bs
->file
);
1518 BlockDriverState
*bdrv_new_open_driver(BlockDriver
*drv
, const char *node_name
,
1519 int flags
, Error
**errp
)
1521 BlockDriverState
*bs
;
1525 bs
->open_flags
= flags
;
1526 bs
->explicit_options
= qdict_new();
1527 bs
->options
= qdict_new();
1530 update_options_from_flags(bs
->options
, flags
);
1532 ret
= bdrv_open_driver(bs
, drv
, node_name
, bs
->options
, flags
, errp
);
1534 qobject_unref(bs
->explicit_options
);
1535 bs
->explicit_options
= NULL
;
1536 qobject_unref(bs
->options
);
1545 QemuOptsList bdrv_runtime_opts
= {
1546 .name
= "bdrv_common",
1547 .head
= QTAILQ_HEAD_INITIALIZER(bdrv_runtime_opts
.head
),
1550 .name
= "node-name",
1551 .type
= QEMU_OPT_STRING
,
1552 .help
= "Node name of the block device node",
1556 .type
= QEMU_OPT_STRING
,
1557 .help
= "Block driver to use for the node",
1560 .name
= BDRV_OPT_CACHE_DIRECT
,
1561 .type
= QEMU_OPT_BOOL
,
1562 .help
= "Bypass software writeback cache on the host",
1565 .name
= BDRV_OPT_CACHE_NO_FLUSH
,
1566 .type
= QEMU_OPT_BOOL
,
1567 .help
= "Ignore flush requests",
1570 .name
= BDRV_OPT_READ_ONLY
,
1571 .type
= QEMU_OPT_BOOL
,
1572 .help
= "Node is opened in read-only mode",
1575 .name
= BDRV_OPT_AUTO_READ_ONLY
,
1576 .type
= QEMU_OPT_BOOL
,
1577 .help
= "Node can become read-only if opening read-write fails",
1580 .name
= "detect-zeroes",
1581 .type
= QEMU_OPT_STRING
,
1582 .help
= "try to optimize zero writes (off, on, unmap)",
1585 .name
= BDRV_OPT_DISCARD
,
1586 .type
= QEMU_OPT_STRING
,
1587 .help
= "discard operation (ignore/off, unmap/on)",
1590 .name
= BDRV_OPT_FORCE_SHARE
,
1591 .type
= QEMU_OPT_BOOL
,
1592 .help
= "always accept other writers (default: off)",
1594 { /* end of list */ }
1598 QemuOptsList bdrv_create_opts_simple
= {
1599 .name
= "simple-create-opts",
1600 .head
= QTAILQ_HEAD_INITIALIZER(bdrv_create_opts_simple
.head
),
1603 .name
= BLOCK_OPT_SIZE
,
1604 .type
= QEMU_OPT_SIZE
,
1605 .help
= "Virtual disk size"
1608 .name
= BLOCK_OPT_PREALLOC
,
1609 .type
= QEMU_OPT_STRING
,
1610 .help
= "Preallocation mode (allowed values: off)"
1612 { /* end of list */ }
1617 * Common part for opening disk images and files
1619 * Removes all processed options from *options.
1621 static int bdrv_open_common(BlockDriverState
*bs
, BlockBackend
*file
,
1622 QDict
*options
, Error
**errp
)
1624 int ret
, open_flags
;
1625 const char *filename
;
1626 const char *driver_name
= NULL
;
1627 const char *node_name
= NULL
;
1628 const char *discard
;
1631 Error
*local_err
= NULL
;
1633 assert(bs
->file
== NULL
);
1634 assert(options
!= NULL
&& bs
->options
!= options
);
1636 opts
= qemu_opts_create(&bdrv_runtime_opts
, NULL
, 0, &error_abort
);
1637 if (!qemu_opts_absorb_qdict(opts
, options
, errp
)) {
1642 update_flags_from_options(&bs
->open_flags
, opts
);
1644 driver_name
= qemu_opt_get(opts
, "driver");
1645 drv
= bdrv_find_format(driver_name
);
1646 assert(drv
!= NULL
);
1648 bs
->force_share
= qemu_opt_get_bool(opts
, BDRV_OPT_FORCE_SHARE
, false);
1650 if (bs
->force_share
&& (bs
->open_flags
& BDRV_O_RDWR
)) {
1652 BDRV_OPT_FORCE_SHARE
1653 "=on can only be used with read-only images");
1659 bdrv_refresh_filename(blk_bs(file
));
1660 filename
= blk_bs(file
)->filename
;
1663 * Caution: while qdict_get_try_str() is fine, getting
1664 * non-string types would require more care. When @options
1665 * come from -blockdev or blockdev_add, its members are typed
1666 * according to the QAPI schema, but when they come from
1667 * -drive, they're all QString.
1669 filename
= qdict_get_try_str(options
, "filename");
1672 if (drv
->bdrv_needs_filename
&& (!filename
|| !filename
[0])) {
1673 error_setg(errp
, "The '%s' block driver requires a file name",
1679 trace_bdrv_open_common(bs
, filename
?: "", bs
->open_flags
,
1682 bs
->read_only
= !(bs
->open_flags
& BDRV_O_RDWR
);
1684 if (use_bdrv_whitelist
&& !bdrv_is_whitelisted(drv
, bs
->read_only
)) {
1685 if (!bs
->read_only
&& bdrv_is_whitelisted(drv
, true)) {
1686 ret
= bdrv_apply_auto_read_only(bs
, NULL
, NULL
);
1692 !bs
->read_only
&& bdrv_is_whitelisted(drv
, true)
1693 ? "Driver '%s' can only be used for read-only devices"
1694 : "Driver '%s' is not whitelisted",
1700 /* bdrv_new() and bdrv_close() make it so */
1701 assert(atomic_read(&bs
->copy_on_read
) == 0);
1703 if (bs
->open_flags
& BDRV_O_COPY_ON_READ
) {
1704 if (!bs
->read_only
) {
1705 bdrv_enable_copy_on_read(bs
);
1707 error_setg(errp
, "Can't use copy-on-read on read-only device");
1713 discard
= qemu_opt_get(opts
, BDRV_OPT_DISCARD
);
1714 if (discard
!= NULL
) {
1715 if (bdrv_parse_discard_flags(discard
, &bs
->open_flags
) != 0) {
1716 error_setg(errp
, "Invalid discard option");
1723 bdrv_parse_detect_zeroes(opts
, bs
->open_flags
, &local_err
);
1725 error_propagate(errp
, local_err
);
1730 if (filename
!= NULL
) {
1731 pstrcpy(bs
->filename
, sizeof(bs
->filename
), filename
);
1733 bs
->filename
[0] = '\0';
1735 pstrcpy(bs
->exact_filename
, sizeof(bs
->exact_filename
), bs
->filename
);
1737 /* Open the image, either directly or using a protocol */
1738 open_flags
= bdrv_open_flags(bs
, bs
->open_flags
);
1739 node_name
= qemu_opt_get(opts
, "node-name");
1741 assert(!drv
->bdrv_file_open
|| file
== NULL
);
1742 ret
= bdrv_open_driver(bs
, drv
, node_name
, options
, open_flags
, errp
);
1747 qemu_opts_del(opts
);
1751 qemu_opts_del(opts
);
1755 static QDict
*parse_json_filename(const char *filename
, Error
**errp
)
1757 QObject
*options_obj
;
1761 ret
= strstart(filename
, "json:", &filename
);
1764 options_obj
= qobject_from_json(filename
, errp
);
1766 error_prepend(errp
, "Could not parse the JSON options: ");
1770 options
= qobject_to(QDict
, options_obj
);
1772 qobject_unref(options_obj
);
1773 error_setg(errp
, "Invalid JSON object given");
1777 qdict_flatten(options
);
1782 static void parse_json_protocol(QDict
*options
, const char **pfilename
,
1785 QDict
*json_options
;
1786 Error
*local_err
= NULL
;
1788 /* Parse json: pseudo-protocol */
1789 if (!*pfilename
|| !g_str_has_prefix(*pfilename
, "json:")) {
1793 json_options
= parse_json_filename(*pfilename
, &local_err
);
1795 error_propagate(errp
, local_err
);
1799 /* Options given in the filename have lower priority than options
1800 * specified directly */
1801 qdict_join(options
, json_options
, false);
1802 qobject_unref(json_options
);
1807 * Fills in default options for opening images and converts the legacy
1808 * filename/flags pair to option QDict entries.
1809 * The BDRV_O_PROTOCOL flag in *flags will be set or cleared accordingly if a
1810 * block driver has been specified explicitly.
1812 static int bdrv_fill_options(QDict
**options
, const char *filename
,
1813 int *flags
, Error
**errp
)
1815 const char *drvname
;
1816 bool protocol
= *flags
& BDRV_O_PROTOCOL
;
1817 bool parse_filename
= false;
1818 BlockDriver
*drv
= NULL
;
1819 Error
*local_err
= NULL
;
1822 * Caution: while qdict_get_try_str() is fine, getting non-string
1823 * types would require more care. When @options come from
1824 * -blockdev or blockdev_add, its members are typed according to
1825 * the QAPI schema, but when they come from -drive, they're all
1828 drvname
= qdict_get_try_str(*options
, "driver");
1830 drv
= bdrv_find_format(drvname
);
1832 error_setg(errp
, "Unknown driver '%s'", drvname
);
1835 /* If the user has explicitly specified the driver, this choice should
1836 * override the BDRV_O_PROTOCOL flag */
1837 protocol
= drv
->bdrv_file_open
;
1841 *flags
|= BDRV_O_PROTOCOL
;
1843 *flags
&= ~BDRV_O_PROTOCOL
;
1846 /* Translate cache options from flags into options */
1847 update_options_from_flags(*options
, *flags
);
1849 /* Fetch the file name from the options QDict if necessary */
1850 if (protocol
&& filename
) {
1851 if (!qdict_haskey(*options
, "filename")) {
1852 qdict_put_str(*options
, "filename", filename
);
1853 parse_filename
= true;
1855 error_setg(errp
, "Can't specify 'file' and 'filename' options at "
1861 /* Find the right block driver */
1862 /* See cautionary note on accessing @options above */
1863 filename
= qdict_get_try_str(*options
, "filename");
1865 if (!drvname
&& protocol
) {
1867 drv
= bdrv_find_protocol(filename
, parse_filename
, errp
);
1872 drvname
= drv
->format_name
;
1873 qdict_put_str(*options
, "driver", drvname
);
1875 error_setg(errp
, "Must specify either driver or file");
1880 assert(drv
|| !protocol
);
1882 /* Driver-specific filename parsing */
1883 if (drv
&& drv
->bdrv_parse_filename
&& parse_filename
) {
1884 drv
->bdrv_parse_filename(filename
, *options
, &local_err
);
1886 error_propagate(errp
, local_err
);
1890 if (!drv
->bdrv_needs_filename
) {
1891 qdict_del(*options
, "filename");
1898 static int bdrv_child_check_perm(BdrvChild
*c
, BlockReopenQueue
*q
,
1899 uint64_t perm
, uint64_t shared
,
1900 GSList
*ignore_children
,
1901 bool *tighten_restrictions
, Error
**errp
);
1902 static void bdrv_child_abort_perm_update(BdrvChild
*c
);
1903 static void bdrv_child_set_perm(BdrvChild
*c
, uint64_t perm
, uint64_t shared
);
1905 typedef struct BlockReopenQueueEntry
{
1908 BDRVReopenState state
;
1909 QTAILQ_ENTRY(BlockReopenQueueEntry
) entry
;
1910 } BlockReopenQueueEntry
;
1913 * Return the flags that @bs will have after the reopens in @q have
1914 * successfully completed. If @q is NULL (or @bs is not contained in @q),
1915 * return the current flags.
1917 static int bdrv_reopen_get_flags(BlockReopenQueue
*q
, BlockDriverState
*bs
)
1919 BlockReopenQueueEntry
*entry
;
1922 QTAILQ_FOREACH(entry
, q
, entry
) {
1923 if (entry
->state
.bs
== bs
) {
1924 return entry
->state
.flags
;
1929 return bs
->open_flags
;
1932 /* Returns whether the image file can be written to after the reopen queue @q
1933 * has been successfully applied, or right now if @q is NULL. */
1934 static bool bdrv_is_writable_after_reopen(BlockDriverState
*bs
,
1935 BlockReopenQueue
*q
)
1937 int flags
= bdrv_reopen_get_flags(q
, bs
);
1939 return (flags
& (BDRV_O_RDWR
| BDRV_O_INACTIVE
)) == BDRV_O_RDWR
;
1943 * Return whether the BDS can be written to. This is not necessarily
1944 * the same as !bdrv_is_read_only(bs), as inactivated images may not
1945 * be written to but do not count as read-only images.
1947 bool bdrv_is_writable(BlockDriverState
*bs
)
1949 return bdrv_is_writable_after_reopen(bs
, NULL
);
1952 static void bdrv_child_perm(BlockDriverState
*bs
, BlockDriverState
*child_bs
,
1953 BdrvChild
*c
, BdrvChildRole role
,
1954 BlockReopenQueue
*reopen_queue
,
1955 uint64_t parent_perm
, uint64_t parent_shared
,
1956 uint64_t *nperm
, uint64_t *nshared
)
1958 assert(bs
->drv
&& bs
->drv
->bdrv_child_perm
);
1959 bs
->drv
->bdrv_child_perm(bs
, c
, role
, reopen_queue
,
1960 parent_perm
, parent_shared
,
1962 /* TODO Take force_share from reopen_queue */
1963 if (child_bs
&& child_bs
->force_share
) {
1964 *nshared
= BLK_PERM_ALL
;
1969 * Check whether permissions on this node can be changed in a way that
1970 * @cumulative_perms and @cumulative_shared_perms are the new cumulative
1971 * permissions of all its parents. This involves checking whether all necessary
1972 * permission changes to child nodes can be performed.
1974 * Will set *tighten_restrictions to true if and only if new permissions have to
1975 * be taken or currently shared permissions are to be unshared. Otherwise,
1976 * errors are not fatal as long as the caller accepts that the restrictions
1977 * remain tighter than they need to be. The caller still has to abort the
1979 * @tighten_restrictions cannot be used together with @q: When reopening, we may
1980 * encounter fatal errors even though no restrictions are to be tightened. For
1981 * example, changing a node from RW to RO will fail if the WRITE permission is
1984 * A call to this function must always be followed by a call to bdrv_set_perm()
1985 * or bdrv_abort_perm_update().
1987 static int bdrv_check_perm(BlockDriverState
*bs
, BlockReopenQueue
*q
,
1988 uint64_t cumulative_perms
,
1989 uint64_t cumulative_shared_perms
,
1990 GSList
*ignore_children
,
1991 bool *tighten_restrictions
, Error
**errp
)
1993 BlockDriver
*drv
= bs
->drv
;
1997 assert(!q
|| !tighten_restrictions
);
1999 if (tighten_restrictions
) {
2000 uint64_t current_perms
, current_shared
;
2001 uint64_t added_perms
, removed_shared_perms
;
2003 bdrv_get_cumulative_perm(bs
, ¤t_perms
, ¤t_shared
);
2005 added_perms
= cumulative_perms
& ~current_perms
;
2006 removed_shared_perms
= current_shared
& ~cumulative_shared_perms
;
2008 *tighten_restrictions
= added_perms
|| removed_shared_perms
;
2011 /* Write permissions never work with read-only images */
2012 if ((cumulative_perms
& (BLK_PERM_WRITE
| BLK_PERM_WRITE_UNCHANGED
)) &&
2013 !bdrv_is_writable_after_reopen(bs
, q
))
2015 if (!bdrv_is_writable_after_reopen(bs
, NULL
)) {
2016 error_setg(errp
, "Block node is read-only");
2018 uint64_t current_perms
, current_shared
;
2019 bdrv_get_cumulative_perm(bs
, ¤t_perms
, ¤t_shared
);
2020 if (current_perms
& (BLK_PERM_WRITE
| BLK_PERM_WRITE_UNCHANGED
)) {
2021 error_setg(errp
, "Cannot make block node read-only, there is "
2024 error_setg(errp
, "Cannot make block node read-only and create "
2033 * Unaligned requests will automatically be aligned to bl.request_alignment
2034 * and without RESIZE we can't extend requests to write to space beyond the
2035 * end of the image, so it's required that the image size is aligned.
2037 if ((cumulative_perms
& (BLK_PERM_WRITE
| BLK_PERM_WRITE_UNCHANGED
)) &&
2038 !(cumulative_perms
& BLK_PERM_RESIZE
))
2040 if ((bs
->total_sectors
* BDRV_SECTOR_SIZE
) % bs
->bl
.request_alignment
) {
2041 error_setg(errp
, "Cannot get 'write' permission without 'resize': "
2042 "Image size is not a multiple of request "
2048 /* Check this node */
2053 if (drv
->bdrv_check_perm
) {
2054 return drv
->bdrv_check_perm(bs
, cumulative_perms
,
2055 cumulative_shared_perms
, errp
);
2058 /* Drivers that never have children can omit .bdrv_child_perm() */
2059 if (!drv
->bdrv_child_perm
) {
2060 assert(QLIST_EMPTY(&bs
->children
));
2064 /* Check all children */
2065 QLIST_FOREACH(c
, &bs
->children
, next
) {
2066 uint64_t cur_perm
, cur_shared
;
2067 bool child_tighten_restr
;
2069 bdrv_child_perm(bs
, c
->bs
, c
, c
->role
, q
,
2070 cumulative_perms
, cumulative_shared_perms
,
2071 &cur_perm
, &cur_shared
);
2072 ret
= bdrv_child_check_perm(c
, q
, cur_perm
, cur_shared
, ignore_children
,
2073 tighten_restrictions
? &child_tighten_restr
2076 if (tighten_restrictions
) {
2077 *tighten_restrictions
|= child_tighten_restr
;
2088 * Notifies drivers that after a previous bdrv_check_perm() call, the
2089 * permission update is not performed and any preparations made for it (e.g.
2090 * taken file locks) need to be undone.
2092 * This function recursively notifies all child nodes.
2094 static void bdrv_abort_perm_update(BlockDriverState
*bs
)
2096 BlockDriver
*drv
= bs
->drv
;
2103 if (drv
->bdrv_abort_perm_update
) {
2104 drv
->bdrv_abort_perm_update(bs
);
2107 QLIST_FOREACH(c
, &bs
->children
, next
) {
2108 bdrv_child_abort_perm_update(c
);
2112 static void bdrv_set_perm(BlockDriverState
*bs
, uint64_t cumulative_perms
,
2113 uint64_t cumulative_shared_perms
)
2115 BlockDriver
*drv
= bs
->drv
;
2122 /* Update this node */
2123 if (drv
->bdrv_set_perm
) {
2124 drv
->bdrv_set_perm(bs
, cumulative_perms
, cumulative_shared_perms
);
2127 /* Drivers that never have children can omit .bdrv_child_perm() */
2128 if (!drv
->bdrv_child_perm
) {
2129 assert(QLIST_EMPTY(&bs
->children
));
2133 /* Update all children */
2134 QLIST_FOREACH(c
, &bs
->children
, next
) {
2135 uint64_t cur_perm
, cur_shared
;
2136 bdrv_child_perm(bs
, c
->bs
, c
, c
->role
, NULL
,
2137 cumulative_perms
, cumulative_shared_perms
,
2138 &cur_perm
, &cur_shared
);
2139 bdrv_child_set_perm(c
, cur_perm
, cur_shared
);
2143 void bdrv_get_cumulative_perm(BlockDriverState
*bs
, uint64_t *perm
,
2144 uint64_t *shared_perm
)
2147 uint64_t cumulative_perms
= 0;
2148 uint64_t cumulative_shared_perms
= BLK_PERM_ALL
;
2150 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
2151 cumulative_perms
|= c
->perm
;
2152 cumulative_shared_perms
&= c
->shared_perm
;
2155 *perm
= cumulative_perms
;
2156 *shared_perm
= cumulative_shared_perms
;
2159 static char *bdrv_child_user_desc(BdrvChild
*c
)
2161 if (c
->klass
->get_parent_desc
) {
2162 return c
->klass
->get_parent_desc(c
);
2165 return g_strdup("another user");
2168 char *bdrv_perm_names(uint64_t perm
)
2174 { BLK_PERM_CONSISTENT_READ
, "consistent read" },
2175 { BLK_PERM_WRITE
, "write" },
2176 { BLK_PERM_WRITE_UNCHANGED
, "write unchanged" },
2177 { BLK_PERM_RESIZE
, "resize" },
2178 { BLK_PERM_GRAPH_MOD
, "change children" },
2182 GString
*result
= g_string_sized_new(30);
2183 struct perm_name
*p
;
2185 for (p
= permissions
; p
->name
; p
++) {
2186 if (perm
& p
->perm
) {
2187 if (result
->len
> 0) {
2188 g_string_append(result
, ", ");
2190 g_string_append(result
, p
->name
);
2194 return g_string_free(result
, FALSE
);
2198 * Checks whether a new reference to @bs can be added if the new user requires
2199 * @new_used_perm/@new_shared_perm as its permissions. If @ignore_children is
2200 * set, the BdrvChild objects in this list are ignored in the calculations;
2201 * this allows checking permission updates for an existing reference.
2203 * See bdrv_check_perm() for the semantics of @tighten_restrictions.
2205 * Needs to be followed by a call to either bdrv_set_perm() or
2206 * bdrv_abort_perm_update(). */
2207 static int bdrv_check_update_perm(BlockDriverState
*bs
, BlockReopenQueue
*q
,
2208 uint64_t new_used_perm
,
2209 uint64_t new_shared_perm
,
2210 GSList
*ignore_children
,
2211 bool *tighten_restrictions
,
2215 uint64_t cumulative_perms
= new_used_perm
;
2216 uint64_t cumulative_shared_perms
= new_shared_perm
;
2218 assert(!q
|| !tighten_restrictions
);
2220 /* There is no reason why anyone couldn't tolerate write_unchanged */
2221 assert(new_shared_perm
& BLK_PERM_WRITE_UNCHANGED
);
2223 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
2224 if (g_slist_find(ignore_children
, c
)) {
2228 if ((new_used_perm
& c
->shared_perm
) != new_used_perm
) {
2229 char *user
= bdrv_child_user_desc(c
);
2230 char *perm_names
= bdrv_perm_names(new_used_perm
& ~c
->shared_perm
);
2232 if (tighten_restrictions
) {
2233 *tighten_restrictions
= true;
2236 error_setg(errp
, "Conflicts with use by %s as '%s', which does not "
2238 user
, c
->name
, perm_names
, bdrv_get_node_name(c
->bs
));
2244 if ((c
->perm
& new_shared_perm
) != c
->perm
) {
2245 char *user
= bdrv_child_user_desc(c
);
2246 char *perm_names
= bdrv_perm_names(c
->perm
& ~new_shared_perm
);
2248 if (tighten_restrictions
) {
2249 *tighten_restrictions
= true;
2252 error_setg(errp
, "Conflicts with use by %s as '%s', which uses "
2254 user
, c
->name
, perm_names
, bdrv_get_node_name(c
->bs
));
2260 cumulative_perms
|= c
->perm
;
2261 cumulative_shared_perms
&= c
->shared_perm
;
2264 return bdrv_check_perm(bs
, q
, cumulative_perms
, cumulative_shared_perms
,
2265 ignore_children
, tighten_restrictions
, errp
);
2268 /* Needs to be followed by a call to either bdrv_child_set_perm() or
2269 * bdrv_child_abort_perm_update(). */
2270 static int bdrv_child_check_perm(BdrvChild
*c
, BlockReopenQueue
*q
,
2271 uint64_t perm
, uint64_t shared
,
2272 GSList
*ignore_children
,
2273 bool *tighten_restrictions
, Error
**errp
)
2277 ignore_children
= g_slist_prepend(g_slist_copy(ignore_children
), c
);
2278 ret
= bdrv_check_update_perm(c
->bs
, q
, perm
, shared
, ignore_children
,
2279 tighten_restrictions
, errp
);
2280 g_slist_free(ignore_children
);
2286 if (!c
->has_backup_perm
) {
2287 c
->has_backup_perm
= true;
2288 c
->backup_perm
= c
->perm
;
2289 c
->backup_shared_perm
= c
->shared_perm
;
2292 * Note: it's OK if c->has_backup_perm was already set, as we can find the
2293 * same child twice during check_perm procedure
2297 c
->shared_perm
= shared
;
2302 static void bdrv_child_set_perm(BdrvChild
*c
, uint64_t perm
, uint64_t shared
)
2304 uint64_t cumulative_perms
, cumulative_shared_perms
;
2306 c
->has_backup_perm
= false;
2309 c
->shared_perm
= shared
;
2311 bdrv_get_cumulative_perm(c
->bs
, &cumulative_perms
,
2312 &cumulative_shared_perms
);
2313 bdrv_set_perm(c
->bs
, cumulative_perms
, cumulative_shared_perms
);
2316 static void bdrv_child_abort_perm_update(BdrvChild
*c
)
2318 if (c
->has_backup_perm
) {
2319 c
->perm
= c
->backup_perm
;
2320 c
->shared_perm
= c
->backup_shared_perm
;
2321 c
->has_backup_perm
= false;
2324 bdrv_abort_perm_update(c
->bs
);
2327 int bdrv_child_try_set_perm(BdrvChild
*c
, uint64_t perm
, uint64_t shared
,
2330 Error
*local_err
= NULL
;
2332 bool tighten_restrictions
;
2334 ret
= bdrv_child_check_perm(c
, NULL
, perm
, shared
, NULL
,
2335 &tighten_restrictions
, &local_err
);
2337 bdrv_child_abort_perm_update(c
);
2338 if (tighten_restrictions
) {
2339 error_propagate(errp
, local_err
);
2342 * Our caller may intend to only loosen restrictions and
2343 * does not expect this function to fail. Errors are not
2344 * fatal in such a case, so we can just hide them from our
2347 error_free(local_err
);
2353 bdrv_child_set_perm(c
, perm
, shared
);
2358 int bdrv_child_refresh_perms(BlockDriverState
*bs
, BdrvChild
*c
, Error
**errp
)
2360 uint64_t parent_perms
, parent_shared
;
2361 uint64_t perms
, shared
;
2363 bdrv_get_cumulative_perm(bs
, &parent_perms
, &parent_shared
);
2364 bdrv_child_perm(bs
, c
->bs
, c
, c
->role
, NULL
,
2365 parent_perms
, parent_shared
, &perms
, &shared
);
2367 return bdrv_child_try_set_perm(c
, perms
, shared
, errp
);
2371 * Default implementation for .bdrv_child_perm() for block filters:
2372 * Forward CONSISTENT_READ, WRITE, WRITE_UNCHANGED, and RESIZE to the
2375 static void bdrv_filter_default_perms(BlockDriverState
*bs
, BdrvChild
*c
,
2377 BlockReopenQueue
*reopen_queue
,
2378 uint64_t perm
, uint64_t shared
,
2379 uint64_t *nperm
, uint64_t *nshared
)
2381 *nperm
= perm
& DEFAULT_PERM_PASSTHROUGH
;
2382 *nshared
= (shared
& DEFAULT_PERM_PASSTHROUGH
) | DEFAULT_PERM_UNCHANGED
;
2385 static void bdrv_default_perms_for_cow(BlockDriverState
*bs
, BdrvChild
*c
,
2387 BlockReopenQueue
*reopen_queue
,
2388 uint64_t perm
, uint64_t shared
,
2389 uint64_t *nperm
, uint64_t *nshared
)
2391 assert(role
& BDRV_CHILD_COW
);
2394 * We want consistent read from backing files if the parent needs it.
2395 * No other operations are performed on backing files.
2397 perm
&= BLK_PERM_CONSISTENT_READ
;
2400 * If the parent can deal with changing data, we're okay with a
2401 * writable and resizable backing file.
2402 * TODO Require !(perm & BLK_PERM_CONSISTENT_READ), too?
2404 if (shared
& BLK_PERM_WRITE
) {
2405 shared
= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2410 shared
|= BLK_PERM_CONSISTENT_READ
| BLK_PERM_GRAPH_MOD
|
2411 BLK_PERM_WRITE_UNCHANGED
;
2413 if (bs
->open_flags
& BDRV_O_INACTIVE
) {
2414 shared
|= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2421 static void bdrv_default_perms_for_storage(BlockDriverState
*bs
, BdrvChild
*c
,
2423 BlockReopenQueue
*reopen_queue
,
2424 uint64_t perm
, uint64_t shared
,
2425 uint64_t *nperm
, uint64_t *nshared
)
2429 assert(role
& (BDRV_CHILD_METADATA
| BDRV_CHILD_DATA
));
2431 flags
= bdrv_reopen_get_flags(reopen_queue
, bs
);
2434 * Apart from the modifications below, the same permissions are
2435 * forwarded and left alone as for filters
2437 bdrv_filter_default_perms(bs
, c
, role
, reopen_queue
,
2438 perm
, shared
, &perm
, &shared
);
2440 if (role
& BDRV_CHILD_METADATA
) {
2441 /* Format drivers may touch metadata even if the guest doesn't write */
2442 if (bdrv_is_writable_after_reopen(bs
, reopen_queue
)) {
2443 perm
|= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2447 * bs->file always needs to be consistent because of the
2448 * metadata. We can never allow other users to resize or write
2451 if (!(flags
& BDRV_O_NO_IO
)) {
2452 perm
|= BLK_PERM_CONSISTENT_READ
;
2454 shared
&= ~(BLK_PERM_WRITE
| BLK_PERM_RESIZE
);
2457 if (role
& BDRV_CHILD_DATA
) {
2459 * Technically, everything in this block is a subset of the
2460 * BDRV_CHILD_METADATA path taken above, and so this could
2461 * be an "else if" branch. However, that is not obvious, and
2462 * this function is not performance critical, therefore we let
2463 * this be an independent "if".
2467 * We cannot allow other users to resize the file because the
2468 * format driver might have some assumptions about the size
2469 * (e.g. because it is stored in metadata, or because the file
2470 * is split into fixed-size data files).
2472 shared
&= ~BLK_PERM_RESIZE
;
2475 * WRITE_UNCHANGED often cannot be performed as such on the
2476 * data file. For example, the qcow2 driver may still need to
2477 * write copied clusters on copy-on-read.
2479 if (perm
& BLK_PERM_WRITE_UNCHANGED
) {
2480 perm
|= BLK_PERM_WRITE
;
2484 * If the data file is written to, the format driver may
2485 * expect to be able to resize it by writing beyond the EOF.
2487 if (perm
& BLK_PERM_WRITE
) {
2488 perm
|= BLK_PERM_RESIZE
;
2492 if (bs
->open_flags
& BDRV_O_INACTIVE
) {
2493 shared
|= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2500 void bdrv_default_perms(BlockDriverState
*bs
, BdrvChild
*c
,
2501 BdrvChildRole role
, BlockReopenQueue
*reopen_queue
,
2502 uint64_t perm
, uint64_t shared
,
2503 uint64_t *nperm
, uint64_t *nshared
)
2505 if (role
& BDRV_CHILD_FILTERED
) {
2506 assert(!(role
& (BDRV_CHILD_DATA
| BDRV_CHILD_METADATA
|
2508 bdrv_filter_default_perms(bs
, c
, role
, reopen_queue
,
2509 perm
, shared
, nperm
, nshared
);
2510 } else if (role
& BDRV_CHILD_COW
) {
2511 assert(!(role
& (BDRV_CHILD_DATA
| BDRV_CHILD_METADATA
)));
2512 bdrv_default_perms_for_cow(bs
, c
, role
, reopen_queue
,
2513 perm
, shared
, nperm
, nshared
);
2514 } else if (role
& (BDRV_CHILD_METADATA
| BDRV_CHILD_DATA
)) {
2515 bdrv_default_perms_for_storage(bs
, c
, role
, reopen_queue
,
2516 perm
, shared
, nperm
, nshared
);
2518 g_assert_not_reached();
2522 uint64_t bdrv_qapi_perm_to_blk_perm(BlockPermission qapi_perm
)
2524 static const uint64_t permissions
[] = {
2525 [BLOCK_PERMISSION_CONSISTENT_READ
] = BLK_PERM_CONSISTENT_READ
,
2526 [BLOCK_PERMISSION_WRITE
] = BLK_PERM_WRITE
,
2527 [BLOCK_PERMISSION_WRITE_UNCHANGED
] = BLK_PERM_WRITE_UNCHANGED
,
2528 [BLOCK_PERMISSION_RESIZE
] = BLK_PERM_RESIZE
,
2529 [BLOCK_PERMISSION_GRAPH_MOD
] = BLK_PERM_GRAPH_MOD
,
2532 QEMU_BUILD_BUG_ON(ARRAY_SIZE(permissions
) != BLOCK_PERMISSION__MAX
);
2533 QEMU_BUILD_BUG_ON(1UL << ARRAY_SIZE(permissions
) != BLK_PERM_ALL
+ 1);
2535 assert(qapi_perm
< BLOCK_PERMISSION__MAX
);
2537 return permissions
[qapi_perm
];
2540 static void bdrv_replace_child_noperm(BdrvChild
*child
,
2541 BlockDriverState
*new_bs
)
2543 BlockDriverState
*old_bs
= child
->bs
;
2544 int new_bs_quiesce_counter
;
2547 assert(!child
->frozen
);
2549 if (old_bs
&& new_bs
) {
2550 assert(bdrv_get_aio_context(old_bs
) == bdrv_get_aio_context(new_bs
));
2553 new_bs_quiesce_counter
= (new_bs
? new_bs
->quiesce_counter
: 0);
2554 drain_saldo
= new_bs_quiesce_counter
- child
->parent_quiesce_counter
;
2557 * If the new child node is drained but the old one was not, flush
2558 * all outstanding requests to the old child node.
2560 while (drain_saldo
> 0 && child
->klass
->drained_begin
) {
2561 bdrv_parent_drained_begin_single(child
, true);
2566 /* Detach first so that the recursive drain sections coming from @child
2567 * are already gone and we only end the drain sections that came from
2569 if (child
->klass
->detach
) {
2570 child
->klass
->detach(child
);
2572 QLIST_REMOVE(child
, next_parent
);
2578 QLIST_INSERT_HEAD(&new_bs
->parents
, child
, next_parent
);
2581 * Detaching the old node may have led to the new node's
2582 * quiesce_counter having been decreased. Not a problem, we
2583 * just need to recognize this here and then invoke
2584 * drained_end appropriately more often.
2586 assert(new_bs
->quiesce_counter
<= new_bs_quiesce_counter
);
2587 drain_saldo
+= new_bs
->quiesce_counter
- new_bs_quiesce_counter
;
2589 /* Attach only after starting new drained sections, so that recursive
2590 * drain sections coming from @child don't get an extra .drained_begin
2592 if (child
->klass
->attach
) {
2593 child
->klass
->attach(child
);
2598 * If the old child node was drained but the new one is not, allow
2599 * requests to come in only after the new node has been attached.
2601 while (drain_saldo
< 0 && child
->klass
->drained_end
) {
2602 bdrv_parent_drained_end_single(child
);
2608 * Updates @child to change its reference to point to @new_bs, including
2609 * checking and applying the necessary permisson updates both to the old node
2612 * NULL is passed as @new_bs for removing the reference before freeing @child.
2614 * If @new_bs is not NULL, bdrv_check_perm() must be called beforehand, as this
2615 * function uses bdrv_set_perm() to update the permissions according to the new
2616 * reference that @new_bs gets.
2618 * Callers must ensure that child->frozen is false.
2620 static void bdrv_replace_child(BdrvChild
*child
, BlockDriverState
*new_bs
)
2622 BlockDriverState
*old_bs
= child
->bs
;
2623 uint64_t perm
, shared_perm
;
2625 /* Asserts that child->frozen == false */
2626 bdrv_replace_child_noperm(child
, new_bs
);
2629 * Start with the new node's permissions. If @new_bs is a (direct
2630 * or indirect) child of @old_bs, we must complete the permission
2631 * update on @new_bs before we loosen the restrictions on @old_bs.
2632 * Otherwise, bdrv_check_perm() on @old_bs would re-initiate
2633 * updating the permissions of @new_bs, and thus not purely loosen
2637 bdrv_get_cumulative_perm(new_bs
, &perm
, &shared_perm
);
2638 bdrv_set_perm(new_bs
, perm
, shared_perm
);
2642 /* Update permissions for old node. This is guaranteed to succeed
2643 * because we're just taking a parent away, so we're loosening
2645 bool tighten_restrictions
;
2648 bdrv_get_cumulative_perm(old_bs
, &perm
, &shared_perm
);
2649 ret
= bdrv_check_perm(old_bs
, NULL
, perm
, shared_perm
, NULL
,
2650 &tighten_restrictions
, NULL
);
2651 assert(tighten_restrictions
== false);
2653 /* We only tried to loosen restrictions, so errors are not fatal */
2654 bdrv_abort_perm_update(old_bs
);
2656 bdrv_set_perm(old_bs
, perm
, shared_perm
);
2659 /* When the parent requiring a non-default AioContext is removed, the
2660 * node moves back to the main AioContext */
2661 bdrv_try_set_aio_context(old_bs
, qemu_get_aio_context(), NULL
);
2666 * This function steals the reference to child_bs from the caller.
2667 * That reference is later dropped by bdrv_root_unref_child().
2669 * On failure NULL is returned, errp is set and the reference to
2670 * child_bs is also dropped.
2672 * The caller must hold the AioContext lock @child_bs, but not that of @ctx
2673 * (unless @child_bs is already in @ctx).
2675 BdrvChild
*bdrv_root_attach_child(BlockDriverState
*child_bs
,
2676 const char *child_name
,
2677 const BdrvChildClass
*child_class
,
2678 BdrvChildRole child_role
,
2680 uint64_t perm
, uint64_t shared_perm
,
2681 void *opaque
, Error
**errp
)
2684 Error
*local_err
= NULL
;
2687 ret
= bdrv_check_update_perm(child_bs
, NULL
, perm
, shared_perm
, NULL
, NULL
,
2690 bdrv_abort_perm_update(child_bs
);
2691 bdrv_unref(child_bs
);
2695 child
= g_new(BdrvChild
, 1);
2696 *child
= (BdrvChild
) {
2698 .name
= g_strdup(child_name
),
2699 .klass
= child_class
,
2702 .shared_perm
= shared_perm
,
2706 /* If the AioContexts don't match, first try to move the subtree of
2707 * child_bs into the AioContext of the new parent. If this doesn't work,
2708 * try moving the parent into the AioContext of child_bs instead. */
2709 if (bdrv_get_aio_context(child_bs
) != ctx
) {
2710 ret
= bdrv_try_set_aio_context(child_bs
, ctx
, &local_err
);
2711 if (ret
< 0 && child_class
->can_set_aio_ctx
) {
2712 GSList
*ignore
= g_slist_prepend(NULL
, child
);
2713 ctx
= bdrv_get_aio_context(child_bs
);
2714 if (child_class
->can_set_aio_ctx(child
, ctx
, &ignore
, NULL
)) {
2715 error_free(local_err
);
2717 g_slist_free(ignore
);
2718 ignore
= g_slist_prepend(NULL
, child
);
2719 child_class
->set_aio_ctx(child
, ctx
, &ignore
);
2721 g_slist_free(ignore
);
2724 error_propagate(errp
, local_err
);
2726 bdrv_abort_perm_update(child_bs
);
2727 bdrv_unref(child_bs
);
2732 /* This performs the matching bdrv_set_perm() for the above check. */
2733 bdrv_replace_child(child
, child_bs
);
2739 * This function transfers the reference to child_bs from the caller
2740 * to parent_bs. That reference is later dropped by parent_bs on
2741 * bdrv_close() or if someone calls bdrv_unref_child().
2743 * On failure NULL is returned, errp is set and the reference to
2744 * child_bs is also dropped.
2746 * If @parent_bs and @child_bs are in different AioContexts, the caller must
2747 * hold the AioContext lock for @child_bs, but not for @parent_bs.
2749 BdrvChild
*bdrv_attach_child(BlockDriverState
*parent_bs
,
2750 BlockDriverState
*child_bs
,
2751 const char *child_name
,
2752 const BdrvChildClass
*child_class
,
2753 BdrvChildRole child_role
,
2757 uint64_t perm
, shared_perm
;
2759 bdrv_get_cumulative_perm(parent_bs
, &perm
, &shared_perm
);
2761 assert(parent_bs
->drv
);
2762 bdrv_child_perm(parent_bs
, child_bs
, NULL
, child_role
, NULL
,
2763 perm
, shared_perm
, &perm
, &shared_perm
);
2765 child
= bdrv_root_attach_child(child_bs
, child_name
, child_class
,
2766 child_role
, bdrv_get_aio_context(parent_bs
),
2767 perm
, shared_perm
, parent_bs
, errp
);
2768 if (child
== NULL
) {
2772 QLIST_INSERT_HEAD(&parent_bs
->children
, child
, next
);
2776 static void bdrv_detach_child(BdrvChild
*child
)
2778 QLIST_SAFE_REMOVE(child
, next
);
2780 bdrv_replace_child(child
, NULL
);
2782 g_free(child
->name
);
2786 /* Callers must ensure that child->frozen is false. */
2787 void bdrv_root_unref_child(BdrvChild
*child
)
2789 BlockDriverState
*child_bs
;
2791 child_bs
= child
->bs
;
2792 bdrv_detach_child(child
);
2793 bdrv_unref(child_bs
);
2797 * Clear all inherits_from pointers from children and grandchildren of
2798 * @root that point to @root, where necessary.
2800 static void bdrv_unset_inherits_from(BlockDriverState
*root
, BdrvChild
*child
)
2804 if (child
->bs
->inherits_from
== root
) {
2806 * Remove inherits_from only when the last reference between root and
2807 * child->bs goes away.
2809 QLIST_FOREACH(c
, &root
->children
, next
) {
2810 if (c
!= child
&& c
->bs
== child
->bs
) {
2815 child
->bs
->inherits_from
= NULL
;
2819 QLIST_FOREACH(c
, &child
->bs
->children
, next
) {
2820 bdrv_unset_inherits_from(root
, c
);
2824 /* Callers must ensure that child->frozen is false. */
2825 void bdrv_unref_child(BlockDriverState
*parent
, BdrvChild
*child
)
2827 if (child
== NULL
) {
2831 bdrv_unset_inherits_from(parent
, child
);
2832 bdrv_root_unref_child(child
);
2836 static void bdrv_parent_cb_change_media(BlockDriverState
*bs
, bool load
)
2839 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
2840 if (c
->klass
->change_media
) {
2841 c
->klass
->change_media(c
, load
);
2846 /* Return true if you can reach parent going through child->inherits_from
2847 * recursively. If parent or child are NULL, return false */
2848 static bool bdrv_inherits_from_recursive(BlockDriverState
*child
,
2849 BlockDriverState
*parent
)
2851 while (child
&& child
!= parent
) {
2852 child
= child
->inherits_from
;
2855 return child
!= NULL
;
2859 * Return the BdrvChildRole for @bs's backing child. bs->backing is
2860 * mostly used for COW backing children (role = COW), but also for
2861 * filtered children (role = FILTERED | PRIMARY).
2863 static BdrvChildRole
bdrv_backing_role(BlockDriverState
*bs
)
2865 if (bs
->drv
&& bs
->drv
->is_filter
) {
2866 return BDRV_CHILD_FILTERED
| BDRV_CHILD_PRIMARY
;
2868 return BDRV_CHILD_COW
;
2873 * Sets the bs->backing link of a BDS. A new reference is created; callers
2874 * which don't need their own reference any more must call bdrv_unref().
2876 void bdrv_set_backing_hd(BlockDriverState
*bs
, BlockDriverState
*backing_hd
,
2879 bool update_inherits_from
= bdrv_chain_contains(bs
, backing_hd
) &&
2880 bdrv_inherits_from_recursive(backing_hd
, bs
);
2882 if (bdrv_is_backing_chain_frozen(bs
, child_bs(bs
->backing
), errp
)) {
2887 bdrv_ref(backing_hd
);
2891 /* Cannot be frozen, we checked that above */
2892 bdrv_unref_child(bs
, bs
->backing
);
2900 bs
->backing
= bdrv_attach_child(bs
, backing_hd
, "backing", &child_of_bds
,
2901 bdrv_backing_role(bs
), errp
);
2902 /* If backing_hd was already part of bs's backing chain, and
2903 * inherits_from pointed recursively to bs then let's update it to
2904 * point directly to bs (else it will become NULL). */
2905 if (bs
->backing
&& update_inherits_from
) {
2906 backing_hd
->inherits_from
= bs
;
2910 bdrv_refresh_limits(bs
, NULL
);
2914 * Opens the backing file for a BlockDriverState if not yet open
2916 * bdref_key specifies the key for the image's BlockdevRef in the options QDict.
2917 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
2918 * itself, all options starting with "${bdref_key}." are considered part of the
2921 * TODO Can this be unified with bdrv_open_image()?
2923 int bdrv_open_backing_file(BlockDriverState
*bs
, QDict
*parent_options
,
2924 const char *bdref_key
, Error
**errp
)
2926 char *backing_filename
= NULL
;
2927 char *bdref_key_dot
;
2928 const char *reference
= NULL
;
2930 bool implicit_backing
= false;
2931 BlockDriverState
*backing_hd
;
2933 QDict
*tmp_parent_options
= NULL
;
2934 Error
*local_err
= NULL
;
2936 if (bs
->backing
!= NULL
) {
2940 /* NULL means an empty set of options */
2941 if (parent_options
== NULL
) {
2942 tmp_parent_options
= qdict_new();
2943 parent_options
= tmp_parent_options
;
2946 bs
->open_flags
&= ~BDRV_O_NO_BACKING
;
2948 bdref_key_dot
= g_strdup_printf("%s.", bdref_key
);
2949 qdict_extract_subqdict(parent_options
, &options
, bdref_key_dot
);
2950 g_free(bdref_key_dot
);
2953 * Caution: while qdict_get_try_str() is fine, getting non-string
2954 * types would require more care. When @parent_options come from
2955 * -blockdev or blockdev_add, its members are typed according to
2956 * the QAPI schema, but when they come from -drive, they're all
2959 reference
= qdict_get_try_str(parent_options
, bdref_key
);
2960 if (reference
|| qdict_haskey(options
, "file.filename")) {
2961 /* keep backing_filename NULL */
2962 } else if (bs
->backing_file
[0] == '\0' && qdict_size(options
) == 0) {
2963 qobject_unref(options
);
2966 if (qdict_size(options
) == 0) {
2967 /* If the user specifies options that do not modify the
2968 * backing file's behavior, we might still consider it the
2969 * implicit backing file. But it's easier this way, and
2970 * just specifying some of the backing BDS's options is
2971 * only possible with -drive anyway (otherwise the QAPI
2972 * schema forces the user to specify everything). */
2973 implicit_backing
= !strcmp(bs
->auto_backing_file
, bs
->backing_file
);
2976 backing_filename
= bdrv_get_full_backing_filename(bs
, &local_err
);
2979 error_propagate(errp
, local_err
);
2980 qobject_unref(options
);
2985 if (!bs
->drv
|| !bs
->drv
->supports_backing
) {
2987 error_setg(errp
, "Driver doesn't support backing files");
2988 qobject_unref(options
);
2993 bs
->backing_format
[0] != '\0' && !qdict_haskey(options
, "driver")) {
2994 qdict_put_str(options
, "driver", bs
->backing_format
);
2997 backing_hd
= bdrv_open_inherit(backing_filename
, reference
, options
, 0, bs
,
2998 &child_of_bds
, bdrv_backing_role(bs
), errp
);
3000 bs
->open_flags
|= BDRV_O_NO_BACKING
;
3001 error_prepend(errp
, "Could not open backing file: ");
3006 if (implicit_backing
) {
3007 bdrv_refresh_filename(backing_hd
);
3008 pstrcpy(bs
->auto_backing_file
, sizeof(bs
->auto_backing_file
),
3009 backing_hd
->filename
);
3012 /* Hook up the backing file link; drop our reference, bs owns the
3013 * backing_hd reference now */
3014 bdrv_set_backing_hd(bs
, backing_hd
, &local_err
);
3015 bdrv_unref(backing_hd
);
3017 error_propagate(errp
, local_err
);
3022 qdict_del(parent_options
, bdref_key
);
3025 g_free(backing_filename
);
3026 qobject_unref(tmp_parent_options
);
3030 static BlockDriverState
*
3031 bdrv_open_child_bs(const char *filename
, QDict
*options
, const char *bdref_key
,
3032 BlockDriverState
*parent
, const BdrvChildClass
*child_class
,
3033 BdrvChildRole child_role
, bool allow_none
, Error
**errp
)
3035 BlockDriverState
*bs
= NULL
;
3036 QDict
*image_options
;
3037 char *bdref_key_dot
;
3038 const char *reference
;
3040 assert(child_class
!= NULL
);
3042 bdref_key_dot
= g_strdup_printf("%s.", bdref_key
);
3043 qdict_extract_subqdict(options
, &image_options
, bdref_key_dot
);
3044 g_free(bdref_key_dot
);
3047 * Caution: while qdict_get_try_str() is fine, getting non-string
3048 * types would require more care. When @options come from
3049 * -blockdev or blockdev_add, its members are typed according to
3050 * the QAPI schema, but when they come from -drive, they're all
3053 reference
= qdict_get_try_str(options
, bdref_key
);
3054 if (!filename
&& !reference
&& !qdict_size(image_options
)) {
3056 error_setg(errp
, "A block device must be specified for \"%s\"",
3059 qobject_unref(image_options
);
3063 bs
= bdrv_open_inherit(filename
, reference
, image_options
, 0,
3064 parent
, child_class
, child_role
, errp
);
3070 qdict_del(options
, bdref_key
);
3075 * Opens a disk image whose options are given as BlockdevRef in another block
3078 * If allow_none is true, no image will be opened if filename is false and no
3079 * BlockdevRef is given. NULL will be returned, but errp remains unset.
3081 * bdrev_key specifies the key for the image's BlockdevRef in the options QDict.
3082 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
3083 * itself, all options starting with "${bdref_key}." are considered part of the
3086 * The BlockdevRef will be removed from the options QDict.
3088 BdrvChild
*bdrv_open_child(const char *filename
,
3089 QDict
*options
, const char *bdref_key
,
3090 BlockDriverState
*parent
,
3091 const BdrvChildClass
*child_class
,
3092 BdrvChildRole child_role
,
3093 bool allow_none
, Error
**errp
)
3095 BlockDriverState
*bs
;
3097 bs
= bdrv_open_child_bs(filename
, options
, bdref_key
, parent
, child_class
,
3098 child_role
, allow_none
, errp
);
3103 return bdrv_attach_child(parent
, bs
, bdref_key
, child_class
, child_role
,
3108 * TODO Future callers may need to specify parent/child_class in order for
3109 * option inheritance to work. Existing callers use it for the root node.
3111 BlockDriverState
*bdrv_open_blockdev_ref(BlockdevRef
*ref
, Error
**errp
)
3113 BlockDriverState
*bs
= NULL
;
3114 QObject
*obj
= NULL
;
3115 QDict
*qdict
= NULL
;
3116 const char *reference
= NULL
;
3119 if (ref
->type
== QTYPE_QSTRING
) {
3120 reference
= ref
->u
.reference
;
3122 BlockdevOptions
*options
= &ref
->u
.definition
;
3123 assert(ref
->type
== QTYPE_QDICT
);
3125 v
= qobject_output_visitor_new(&obj
);
3126 visit_type_BlockdevOptions(v
, NULL
, &options
, &error_abort
);
3127 visit_complete(v
, &obj
);
3129 qdict
= qobject_to(QDict
, obj
);
3130 qdict_flatten(qdict
);
3132 /* bdrv_open_inherit() defaults to the values in bdrv_flags (for
3133 * compatibility with other callers) rather than what we want as the
3134 * real defaults. Apply the defaults here instead. */
3135 qdict_set_default_str(qdict
, BDRV_OPT_CACHE_DIRECT
, "off");
3136 qdict_set_default_str(qdict
, BDRV_OPT_CACHE_NO_FLUSH
, "off");
3137 qdict_set_default_str(qdict
, BDRV_OPT_READ_ONLY
, "off");
3138 qdict_set_default_str(qdict
, BDRV_OPT_AUTO_READ_ONLY
, "off");
3142 bs
= bdrv_open_inherit(NULL
, reference
, qdict
, 0, NULL
, NULL
, 0, errp
);
3149 static BlockDriverState
*bdrv_append_temp_snapshot(BlockDriverState
*bs
,
3151 QDict
*snapshot_options
,
3154 /* TODO: extra byte is a hack to ensure MAX_PATH space on Windows. */
3155 char *tmp_filename
= g_malloc0(PATH_MAX
+ 1);
3157 QemuOpts
*opts
= NULL
;
3158 BlockDriverState
*bs_snapshot
= NULL
;
3159 Error
*local_err
= NULL
;
3162 /* if snapshot, we create a temporary backing file and open it
3163 instead of opening 'filename' directly */
3165 /* Get the required size from the image */
3166 total_size
= bdrv_getlength(bs
);
3167 if (total_size
< 0) {
3168 error_setg_errno(errp
, -total_size
, "Could not get image size");
3172 /* Create the temporary image */
3173 ret
= get_tmp_filename(tmp_filename
, PATH_MAX
+ 1);
3175 error_setg_errno(errp
, -ret
, "Could not get temporary filename");
3179 opts
= qemu_opts_create(bdrv_qcow2
.create_opts
, NULL
, 0,
3181 qemu_opt_set_number(opts
, BLOCK_OPT_SIZE
, total_size
, &error_abort
);
3182 ret
= bdrv_create(&bdrv_qcow2
, tmp_filename
, opts
, errp
);
3183 qemu_opts_del(opts
);
3185 error_prepend(errp
, "Could not create temporary overlay '%s': ",
3190 /* Prepare options QDict for the temporary file */
3191 qdict_put_str(snapshot_options
, "file.driver", "file");
3192 qdict_put_str(snapshot_options
, "file.filename", tmp_filename
);
3193 qdict_put_str(snapshot_options
, "driver", "qcow2");
3195 bs_snapshot
= bdrv_open(NULL
, NULL
, snapshot_options
, flags
, errp
);
3196 snapshot_options
= NULL
;
3201 /* bdrv_append() consumes a strong reference to bs_snapshot
3202 * (i.e. it will call bdrv_unref() on it) even on error, so in
3203 * order to be able to return one, we have to increase
3204 * bs_snapshot's refcount here */
3205 bdrv_ref(bs_snapshot
);
3206 bdrv_append(bs_snapshot
, bs
, &local_err
);
3208 error_propagate(errp
, local_err
);
3214 qobject_unref(snapshot_options
);
3215 g_free(tmp_filename
);
3220 * Opens a disk image (raw, qcow2, vmdk, ...)
3222 * options is a QDict of options to pass to the block drivers, or NULL for an
3223 * empty set of options. The reference to the QDict belongs to the block layer
3224 * after the call (even on failure), so if the caller intends to reuse the
3225 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
3227 * If *pbs is NULL, a new BDS will be created with a pointer to it stored there.
3228 * If it is not NULL, the referenced BDS will be reused.
3230 * The reference parameter may be used to specify an existing block device which
3231 * should be opened. If specified, neither options nor a filename may be given,
3232 * nor can an existing BDS be reused (that is, *pbs has to be NULL).
3234 static BlockDriverState
*bdrv_open_inherit(const char *filename
,
3235 const char *reference
,
3236 QDict
*options
, int flags
,
3237 BlockDriverState
*parent
,
3238 const BdrvChildClass
*child_class
,
3239 BdrvChildRole child_role
,
3243 BlockBackend
*file
= NULL
;
3244 BlockDriverState
*bs
;
3245 BlockDriver
*drv
= NULL
;
3247 const char *drvname
;
3248 const char *backing
;
3249 Error
*local_err
= NULL
;
3250 QDict
*snapshot_options
= NULL
;
3251 int snapshot_flags
= 0;
3253 assert(!child_class
|| !flags
);
3254 assert(!child_class
== !parent
);
3257 bool options_non_empty
= options
? qdict_size(options
) : false;
3258 qobject_unref(options
);
3260 if (filename
|| options_non_empty
) {
3261 error_setg(errp
, "Cannot reference an existing block device with "
3262 "additional options or a new filename");
3266 bs
= bdrv_lookup_bs(reference
, reference
, errp
);
3277 /* NULL means an empty set of options */
3278 if (options
== NULL
) {
3279 options
= qdict_new();
3282 /* json: syntax counts as explicit options, as if in the QDict */
3283 parse_json_protocol(options
, &filename
, &local_err
);
3288 bs
->explicit_options
= qdict_clone_shallow(options
);
3291 bool parent_is_format
;
3294 parent_is_format
= parent
->drv
->is_format
;
3297 * parent->drv is not set yet because this node is opened for
3298 * (potential) format probing. That means that @parent is going
3299 * to be a format node.
3301 parent_is_format
= true;
3304 bs
->inherits_from
= parent
;
3305 child_class
->inherit_options(child_role
, parent_is_format
,
3307 parent
->open_flags
, parent
->options
);
3310 ret
= bdrv_fill_options(&options
, filename
, &flags
, &local_err
);
3316 * Set the BDRV_O_RDWR and BDRV_O_ALLOW_RDWR flags.
3317 * Caution: getting a boolean member of @options requires care.
3318 * When @options come from -blockdev or blockdev_add, members are
3319 * typed according to the QAPI schema, but when they come from
3320 * -drive, they're all QString.
3322 if (g_strcmp0(qdict_get_try_str(options
, BDRV_OPT_READ_ONLY
), "on") &&
3323 !qdict_get_try_bool(options
, BDRV_OPT_READ_ONLY
, false)) {
3324 flags
|= (BDRV_O_RDWR
| BDRV_O_ALLOW_RDWR
);
3326 flags
&= ~BDRV_O_RDWR
;
3329 if (flags
& BDRV_O_SNAPSHOT
) {
3330 snapshot_options
= qdict_new();
3331 bdrv_temp_snapshot_options(&snapshot_flags
, snapshot_options
,
3333 /* Let bdrv_backing_options() override "read-only" */
3334 qdict_del(options
, BDRV_OPT_READ_ONLY
);
3335 bdrv_inherited_options(BDRV_CHILD_COW
, true,
3336 &flags
, options
, flags
, options
);
3339 bs
->open_flags
= flags
;
3340 bs
->options
= options
;
3341 options
= qdict_clone_shallow(options
);
3343 /* Find the right image format driver */
3344 /* See cautionary note on accessing @options above */
3345 drvname
= qdict_get_try_str(options
, "driver");
3347 drv
= bdrv_find_format(drvname
);
3349 error_setg(errp
, "Unknown driver: '%s'", drvname
);
3354 assert(drvname
|| !(flags
& BDRV_O_PROTOCOL
));
3356 /* See cautionary note on accessing @options above */
3357 backing
= qdict_get_try_str(options
, "backing");
3358 if (qobject_to(QNull
, qdict_get(options
, "backing")) != NULL
||
3359 (backing
&& *backing
== '\0'))
3362 warn_report("Use of \"backing\": \"\" is deprecated; "
3363 "use \"backing\": null instead");
3365 flags
|= BDRV_O_NO_BACKING
;
3366 qdict_del(bs
->explicit_options
, "backing");
3367 qdict_del(bs
->options
, "backing");
3368 qdict_del(options
, "backing");
3371 /* Open image file without format layer. This BlockBackend is only used for
3372 * probing, the block drivers will do their own bdrv_open_child() for the
3373 * same BDS, which is why we put the node name back into options. */
3374 if ((flags
& BDRV_O_PROTOCOL
) == 0) {
3375 BlockDriverState
*file_bs
;
3377 file_bs
= bdrv_open_child_bs(filename
, options
, "file", bs
,
3378 &child_of_bds
, BDRV_CHILD_IMAGE
,
3383 if (file_bs
!= NULL
) {
3384 /* Not requesting BLK_PERM_CONSISTENT_READ because we're only
3385 * looking at the header to guess the image format. This works even
3386 * in cases where a guest would not see a consistent state. */
3387 file
= blk_new(bdrv_get_aio_context(file_bs
), 0, BLK_PERM_ALL
);
3388 blk_insert_bs(file
, file_bs
, &local_err
);
3389 bdrv_unref(file_bs
);
3394 qdict_put_str(options
, "file", bdrv_get_node_name(file_bs
));
3398 /* Image format probing */
3401 ret
= find_image_format(file
, filename
, &drv
, &local_err
);
3406 * This option update would logically belong in bdrv_fill_options(),
3407 * but we first need to open bs->file for the probing to work, while
3408 * opening bs->file already requires the (mostly) final set of options
3409 * so that cache mode etc. can be inherited.
3411 * Adding the driver later is somewhat ugly, but it's not an option
3412 * that would ever be inherited, so it's correct. We just need to make
3413 * sure to update both bs->options (which has the full effective
3414 * options for bs) and options (which has file.* already removed).
3416 qdict_put_str(bs
->options
, "driver", drv
->format_name
);
3417 qdict_put_str(options
, "driver", drv
->format_name
);
3419 error_setg(errp
, "Must specify either driver or file");
3423 /* BDRV_O_PROTOCOL must be set iff a protocol BDS is about to be created */
3424 assert(!!(flags
& BDRV_O_PROTOCOL
) == !!drv
->bdrv_file_open
);
3425 /* file must be NULL if a protocol BDS is about to be created
3426 * (the inverse results in an error message from bdrv_open_common()) */
3427 assert(!(flags
& BDRV_O_PROTOCOL
) || !file
);
3429 /* Open the image */
3430 ret
= bdrv_open_common(bs
, file
, options
, &local_err
);
3440 /* If there is a backing file, use it */
3441 if ((flags
& BDRV_O_NO_BACKING
) == 0) {
3442 ret
= bdrv_open_backing_file(bs
, options
, "backing", &local_err
);
3444 goto close_and_fail
;
3448 /* Remove all children options and references
3449 * from bs->options and bs->explicit_options */
3450 QLIST_FOREACH(child
, &bs
->children
, next
) {
3451 char *child_key_dot
;
3452 child_key_dot
= g_strdup_printf("%s.", child
->name
);
3453 qdict_extract_subqdict(bs
->explicit_options
, NULL
, child_key_dot
);
3454 qdict_extract_subqdict(bs
->options
, NULL
, child_key_dot
);
3455 qdict_del(bs
->explicit_options
, child
->name
);
3456 qdict_del(bs
->options
, child
->name
);
3457 g_free(child_key_dot
);
3460 /* Check if any unknown options were used */
3461 if (qdict_size(options
) != 0) {
3462 const QDictEntry
*entry
= qdict_first(options
);
3463 if (flags
& BDRV_O_PROTOCOL
) {
3464 error_setg(errp
, "Block protocol '%s' doesn't support the option "
3465 "'%s'", drv
->format_name
, entry
->key
);
3468 "Block format '%s' does not support the option '%s'",
3469 drv
->format_name
, entry
->key
);
3472 goto close_and_fail
;
3475 bdrv_parent_cb_change_media(bs
, true);
3477 qobject_unref(options
);
3480 /* For snapshot=on, create a temporary qcow2 overlay. bs points to the
3481 * temporary snapshot afterwards. */
3482 if (snapshot_flags
) {
3483 BlockDriverState
*snapshot_bs
;
3484 snapshot_bs
= bdrv_append_temp_snapshot(bs
, snapshot_flags
,
3485 snapshot_options
, &local_err
);
3486 snapshot_options
= NULL
;
3488 goto close_and_fail
;
3490 /* We are not going to return bs but the overlay on top of it
3491 * (snapshot_bs); thus, we have to drop the strong reference to bs
3492 * (which we obtained by calling bdrv_new()). bs will not be deleted,
3493 * though, because the overlay still has a reference to it. */
3502 qobject_unref(snapshot_options
);
3503 qobject_unref(bs
->explicit_options
);
3504 qobject_unref(bs
->options
);
3505 qobject_unref(options
);
3507 bs
->explicit_options
= NULL
;
3509 error_propagate(errp
, local_err
);
3514 qobject_unref(snapshot_options
);
3515 qobject_unref(options
);
3516 error_propagate(errp
, local_err
);
3520 BlockDriverState
*bdrv_open(const char *filename
, const char *reference
,
3521 QDict
*options
, int flags
, Error
**errp
)
3523 return bdrv_open_inherit(filename
, reference
, options
, flags
, NULL
,
3527 /* Return true if the NULL-terminated @list contains @str */
3528 static bool is_str_in_list(const char *str
, const char *const *list
)
3532 for (i
= 0; list
[i
] != NULL
; i
++) {
3533 if (!strcmp(str
, list
[i
])) {
3542 * Check that every option set in @bs->options is also set in
3545 * Options listed in the common_options list and in
3546 * @bs->drv->mutable_opts are skipped.
3548 * Return 0 on success, otherwise return -EINVAL and set @errp.
3550 static int bdrv_reset_options_allowed(BlockDriverState
*bs
,
3551 const QDict
*new_opts
, Error
**errp
)
3553 const QDictEntry
*e
;
3554 /* These options are common to all block drivers and are handled
3555 * in bdrv_reopen_prepare() so they can be left out of @new_opts */
3556 const char *const common_options
[] = {
3557 "node-name", "discard", "cache.direct", "cache.no-flush",
3558 "read-only", "auto-read-only", "detect-zeroes", NULL
3561 for (e
= qdict_first(bs
->options
); e
; e
= qdict_next(bs
->options
, e
)) {
3562 if (!qdict_haskey(new_opts
, e
->key
) &&
3563 !is_str_in_list(e
->key
, common_options
) &&
3564 !is_str_in_list(e
->key
, bs
->drv
->mutable_opts
)) {
3565 error_setg(errp
, "Option '%s' cannot be reset "
3566 "to its default value", e
->key
);
3575 * Returns true if @child can be reached recursively from @bs
3577 static bool bdrv_recurse_has_child(BlockDriverState
*bs
,
3578 BlockDriverState
*child
)
3586 QLIST_FOREACH(c
, &bs
->children
, next
) {
3587 if (bdrv_recurse_has_child(c
->bs
, child
)) {
3596 * Adds a BlockDriverState to a simple queue for an atomic, transactional
3597 * reopen of multiple devices.
3599 * bs_queue can either be an existing BlockReopenQueue that has had QTAILQ_INIT
3600 * already performed, or alternatively may be NULL a new BlockReopenQueue will
3601 * be created and initialized. This newly created BlockReopenQueue should be
3602 * passed back in for subsequent calls that are intended to be of the same
3605 * bs is the BlockDriverState to add to the reopen queue.
3607 * options contains the changed options for the associated bs
3608 * (the BlockReopenQueue takes ownership)
3610 * flags contains the open flags for the associated bs
3612 * returns a pointer to bs_queue, which is either the newly allocated
3613 * bs_queue, or the existing bs_queue being used.
3615 * bs must be drained between bdrv_reopen_queue() and bdrv_reopen_multiple().
3617 static BlockReopenQueue
*bdrv_reopen_queue_child(BlockReopenQueue
*bs_queue
,
3618 BlockDriverState
*bs
,
3620 const BdrvChildClass
*klass
,
3622 bool parent_is_format
,
3623 QDict
*parent_options
,
3629 BlockReopenQueueEntry
*bs_entry
;
3631 QDict
*old_options
, *explicit_options
, *options_copy
;
3635 /* Make sure that the caller remembered to use a drained section. This is
3636 * important to avoid graph changes between the recursive queuing here and
3637 * bdrv_reopen_multiple(). */
3638 assert(bs
->quiesce_counter
> 0);
3640 if (bs_queue
== NULL
) {
3641 bs_queue
= g_new0(BlockReopenQueue
, 1);
3642 QTAILQ_INIT(bs_queue
);
3646 options
= qdict_new();
3649 /* Check if this BlockDriverState is already in the queue */
3650 QTAILQ_FOREACH(bs_entry
, bs_queue
, entry
) {
3651 if (bs
== bs_entry
->state
.bs
) {
3657 * Precedence of options:
3658 * 1. Explicitly passed in options (highest)
3659 * 2. Retained from explicitly set options of bs
3660 * 3. Inherited from parent node
3661 * 4. Retained from effective options of bs
3664 /* Old explicitly set values (don't overwrite by inherited value) */
3665 if (bs_entry
|| keep_old_opts
) {
3666 old_options
= qdict_clone_shallow(bs_entry
?
3667 bs_entry
->state
.explicit_options
:
3668 bs
->explicit_options
);
3669 bdrv_join_options(bs
, options
, old_options
);
3670 qobject_unref(old_options
);
3673 explicit_options
= qdict_clone_shallow(options
);
3675 /* Inherit from parent node */
3676 if (parent_options
) {
3678 klass
->inherit_options(role
, parent_is_format
, &flags
, options
,
3679 parent_flags
, parent_options
);
3681 flags
= bdrv_get_flags(bs
);
3684 if (keep_old_opts
) {
3685 /* Old values are used for options that aren't set yet */
3686 old_options
= qdict_clone_shallow(bs
->options
);
3687 bdrv_join_options(bs
, options
, old_options
);
3688 qobject_unref(old_options
);
3691 /* We have the final set of options so let's update the flags */
3692 options_copy
= qdict_clone_shallow(options
);
3693 opts
= qemu_opts_create(&bdrv_runtime_opts
, NULL
, 0, &error_abort
);
3694 qemu_opts_absorb_qdict(opts
, options_copy
, NULL
);
3695 update_flags_from_options(&flags
, opts
);
3696 qemu_opts_del(opts
);
3697 qobject_unref(options_copy
);
3699 /* bdrv_open_inherit() sets and clears some additional flags internally */
3700 flags
&= ~BDRV_O_PROTOCOL
;
3701 if (flags
& BDRV_O_RDWR
) {
3702 flags
|= BDRV_O_ALLOW_RDWR
;
3706 bs_entry
= g_new0(BlockReopenQueueEntry
, 1);
3707 QTAILQ_INSERT_TAIL(bs_queue
, bs_entry
, entry
);
3709 qobject_unref(bs_entry
->state
.options
);
3710 qobject_unref(bs_entry
->state
.explicit_options
);
3713 bs_entry
->state
.bs
= bs
;
3714 bs_entry
->state
.options
= options
;
3715 bs_entry
->state
.explicit_options
= explicit_options
;
3716 bs_entry
->state
.flags
= flags
;
3718 /* This needs to be overwritten in bdrv_reopen_prepare() */
3719 bs_entry
->state
.perm
= UINT64_MAX
;
3720 bs_entry
->state
.shared_perm
= 0;
3723 * If keep_old_opts is false then it means that unspecified
3724 * options must be reset to their original value. We don't allow
3725 * resetting 'backing' but we need to know if the option is
3726 * missing in order to decide if we have to return an error.
3728 if (!keep_old_opts
) {
3729 bs_entry
->state
.backing_missing
=
3730 !qdict_haskey(options
, "backing") &&
3731 !qdict_haskey(options
, "backing.driver");
3734 QLIST_FOREACH(child
, &bs
->children
, next
) {
3735 QDict
*new_child_options
= NULL
;
3736 bool child_keep_old
= keep_old_opts
;
3738 /* reopen can only change the options of block devices that were
3739 * implicitly created and inherited options. For other (referenced)
3740 * block devices, a syntax like "backing.foo" results in an error. */
3741 if (child
->bs
->inherits_from
!= bs
) {
3745 /* Check if the options contain a child reference */
3746 if (qdict_haskey(options
, child
->name
)) {
3747 const char *childref
= qdict_get_try_str(options
, child
->name
);
3749 * The current child must not be reopened if the child
3750 * reference is null or points to a different node.
3752 if (g_strcmp0(childref
, child
->bs
->node_name
)) {
3756 * If the child reference points to the current child then
3757 * reopen it with its existing set of options (note that
3758 * it can still inherit new options from the parent).
3760 child_keep_old
= true;
3762 /* Extract child options ("child-name.*") */
3763 char *child_key_dot
= g_strdup_printf("%s.", child
->name
);
3764 qdict_extract_subqdict(explicit_options
, NULL
, child_key_dot
);
3765 qdict_extract_subqdict(options
, &new_child_options
, child_key_dot
);
3766 g_free(child_key_dot
);
3769 bdrv_reopen_queue_child(bs_queue
, child
->bs
, new_child_options
,
3770 child
->klass
, child
->role
, bs
->drv
->is_format
,
3771 options
, flags
, child_keep_old
);
3777 BlockReopenQueue
*bdrv_reopen_queue(BlockReopenQueue
*bs_queue
,
3778 BlockDriverState
*bs
,
3779 QDict
*options
, bool keep_old_opts
)
3781 return bdrv_reopen_queue_child(bs_queue
, bs
, options
, NULL
, 0, false,
3782 NULL
, 0, keep_old_opts
);
3786 * Reopen multiple BlockDriverStates atomically & transactionally.
3788 * The queue passed in (bs_queue) must have been built up previous
3789 * via bdrv_reopen_queue().
3791 * Reopens all BDS specified in the queue, with the appropriate
3792 * flags. All devices are prepared for reopen, and failure of any
3793 * device will cause all device changes to be abandoned, and intermediate
3796 * If all devices prepare successfully, then the changes are committed
3799 * All affected nodes must be drained between bdrv_reopen_queue() and
3800 * bdrv_reopen_multiple().
3802 int bdrv_reopen_multiple(BlockReopenQueue
*bs_queue
, Error
**errp
)
3805 BlockReopenQueueEntry
*bs_entry
, *next
;
3807 assert(bs_queue
!= NULL
);
3809 QTAILQ_FOREACH(bs_entry
, bs_queue
, entry
) {
3810 assert(bs_entry
->state
.bs
->quiesce_counter
> 0);
3811 if (bdrv_reopen_prepare(&bs_entry
->state
, bs_queue
, errp
)) {
3814 bs_entry
->prepared
= true;
3817 QTAILQ_FOREACH(bs_entry
, bs_queue
, entry
) {
3818 BDRVReopenState
*state
= &bs_entry
->state
;
3819 ret
= bdrv_check_perm(state
->bs
, bs_queue
, state
->perm
,
3820 state
->shared_perm
, NULL
, NULL
, errp
);
3824 /* Check if new_backing_bs would accept the new permissions */
3825 if (state
->replace_backing_bs
&& state
->new_backing_bs
) {
3826 uint64_t nperm
, nshared
;
3827 bdrv_child_perm(state
->bs
, state
->new_backing_bs
,
3828 NULL
, bdrv_backing_role(state
->bs
),
3829 bs_queue
, state
->perm
, state
->shared_perm
,
3831 ret
= bdrv_check_update_perm(state
->new_backing_bs
, NULL
,
3832 nperm
, nshared
, NULL
, NULL
, errp
);
3837 bs_entry
->perms_checked
= true;
3841 * If we reach this point, we have success and just need to apply the
3844 * Reverse order is used to comfort qcow2 driver: on commit it need to write
3845 * IN_USE flag to the image, to mark bitmaps in the image as invalid. But
3846 * children are usually goes after parents in reopen-queue, so go from last
3849 QTAILQ_FOREACH_REVERSE(bs_entry
, bs_queue
, entry
) {
3850 bdrv_reopen_commit(&bs_entry
->state
);
3855 QTAILQ_FOREACH_SAFE(bs_entry
, bs_queue
, entry
, next
) {
3856 BDRVReopenState
*state
= &bs_entry
->state
;
3858 if (!bs_entry
->perms_checked
) {
3863 bdrv_set_perm(state
->bs
, state
->perm
, state
->shared_perm
);
3865 bdrv_abort_perm_update(state
->bs
);
3866 if (state
->replace_backing_bs
&& state
->new_backing_bs
) {
3867 bdrv_abort_perm_update(state
->new_backing_bs
);
3873 QTAILQ_FOREACH_REVERSE(bs_entry
, bs_queue
, entry
) {
3874 BlockDriverState
*bs
= bs_entry
->state
.bs
;
3876 if (bs
->drv
->bdrv_reopen_commit_post
)
3877 bs
->drv
->bdrv_reopen_commit_post(&bs_entry
->state
);
3881 QTAILQ_FOREACH_SAFE(bs_entry
, bs_queue
, entry
, next
) {
3883 if (bs_entry
->prepared
) {
3884 bdrv_reopen_abort(&bs_entry
->state
);
3886 qobject_unref(bs_entry
->state
.explicit_options
);
3887 qobject_unref(bs_entry
->state
.options
);
3889 if (bs_entry
->state
.new_backing_bs
) {
3890 bdrv_unref(bs_entry
->state
.new_backing_bs
);
3899 int bdrv_reopen_set_read_only(BlockDriverState
*bs
, bool read_only
,
3903 BlockReopenQueue
*queue
;
3904 QDict
*opts
= qdict_new();
3906 qdict_put_bool(opts
, BDRV_OPT_READ_ONLY
, read_only
);
3908 bdrv_subtree_drained_begin(bs
);
3909 queue
= bdrv_reopen_queue(NULL
, bs
, opts
, true);
3910 ret
= bdrv_reopen_multiple(queue
, errp
);
3911 bdrv_subtree_drained_end(bs
);
3916 static BlockReopenQueueEntry
*find_parent_in_reopen_queue(BlockReopenQueue
*q
,
3919 BlockReopenQueueEntry
*entry
;
3921 QTAILQ_FOREACH(entry
, q
, entry
) {
3922 BlockDriverState
*bs
= entry
->state
.bs
;
3925 QLIST_FOREACH(child
, &bs
->children
, next
) {
3935 static void bdrv_reopen_perm(BlockReopenQueue
*q
, BlockDriverState
*bs
,
3936 uint64_t *perm
, uint64_t *shared
)
3939 BlockReopenQueueEntry
*parent
;
3940 uint64_t cumulative_perms
= 0;
3941 uint64_t cumulative_shared_perms
= BLK_PERM_ALL
;
3943 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
3944 parent
= find_parent_in_reopen_queue(q
, c
);
3946 cumulative_perms
|= c
->perm
;
3947 cumulative_shared_perms
&= c
->shared_perm
;
3949 uint64_t nperm
, nshared
;
3951 bdrv_child_perm(parent
->state
.bs
, bs
, c
, c
->role
, q
,
3952 parent
->state
.perm
, parent
->state
.shared_perm
,
3955 cumulative_perms
|= nperm
;
3956 cumulative_shared_perms
&= nshared
;
3959 *perm
= cumulative_perms
;
3960 *shared
= cumulative_shared_perms
;
3963 static bool bdrv_reopen_can_attach(BlockDriverState
*parent
,
3965 BlockDriverState
*new_child
,
3968 AioContext
*parent_ctx
= bdrv_get_aio_context(parent
);
3969 AioContext
*child_ctx
= bdrv_get_aio_context(new_child
);
3973 ignore
= g_slist_prepend(NULL
, child
);
3974 ret
= bdrv_can_set_aio_context(new_child
, parent_ctx
, &ignore
, NULL
);
3975 g_slist_free(ignore
);
3980 ignore
= g_slist_prepend(NULL
, child
);
3981 ret
= bdrv_can_set_aio_context(parent
, child_ctx
, &ignore
, errp
);
3982 g_slist_free(ignore
);
3987 * Take a BDRVReopenState and check if the value of 'backing' in the
3988 * reopen_state->options QDict is valid or not.
3990 * If 'backing' is missing from the QDict then return 0.
3992 * If 'backing' contains the node name of the backing file of
3993 * reopen_state->bs then return 0.
3995 * If 'backing' contains a different node name (or is null) then check
3996 * whether the current backing file can be replaced with the new one.
3997 * If that's the case then reopen_state->replace_backing_bs is set to
3998 * true and reopen_state->new_backing_bs contains a pointer to the new
3999 * backing BlockDriverState (or NULL).
4001 * Return 0 on success, otherwise return < 0 and set @errp.
4003 static int bdrv_reopen_parse_backing(BDRVReopenState
*reopen_state
,
4006 BlockDriverState
*bs
= reopen_state
->bs
;
4007 BlockDriverState
*overlay_bs
, *below_bs
, *new_backing_bs
;
4011 value
= qdict_get(reopen_state
->options
, "backing");
4012 if (value
== NULL
) {
4016 switch (qobject_type(value
)) {
4018 new_backing_bs
= NULL
;
4021 str
= qobject_get_try_str(value
);
4022 new_backing_bs
= bdrv_lookup_bs(NULL
, str
, errp
);
4023 if (new_backing_bs
== NULL
) {
4025 } else if (bdrv_recurse_has_child(new_backing_bs
, bs
)) {
4026 error_setg(errp
, "Making '%s' a backing file of '%s' "
4027 "would create a cycle", str
, bs
->node_name
);
4032 /* 'backing' does not allow any other data type */
4033 g_assert_not_reached();
4037 * Check AioContext compatibility so that the bdrv_set_backing_hd() call in
4038 * bdrv_reopen_commit() won't fail.
4040 if (new_backing_bs
) {
4041 if (!bdrv_reopen_can_attach(bs
, bs
->backing
, new_backing_bs
, errp
)) {
4047 * Ensure that @bs can really handle backing files, because we are
4048 * about to give it one (or swap the existing one)
4050 if (bs
->drv
->is_filter
) {
4051 /* Filters always have a file or a backing child */
4053 error_setg(errp
, "'%s' is a %s filter node that does not support a "
4054 "backing child", bs
->node_name
, bs
->drv
->format_name
);
4057 } else if (!bs
->drv
->supports_backing
) {
4058 error_setg(errp
, "Driver '%s' of node '%s' does not support backing "
4059 "files", bs
->drv
->format_name
, bs
->node_name
);
4064 * Find the "actual" backing file by skipping all links that point
4065 * to an implicit node, if any (e.g. a commit filter node).
4066 * We cannot use any of the bdrv_skip_*() functions here because
4067 * those return the first explicit node, while we are looking for
4071 for (below_bs
= bdrv_filter_or_cow_bs(overlay_bs
);
4072 below_bs
&& below_bs
->implicit
;
4073 below_bs
= bdrv_filter_or_cow_bs(overlay_bs
))
4075 overlay_bs
= below_bs
;
4078 /* If we want to replace the backing file we need some extra checks */
4079 if (new_backing_bs
!= bdrv_filter_or_cow_bs(overlay_bs
)) {
4080 /* Check for implicit nodes between bs and its backing file */
4081 if (bs
!= overlay_bs
) {
4082 error_setg(errp
, "Cannot change backing link if '%s' has "
4083 "an implicit backing file", bs
->node_name
);
4087 * Check if the backing link that we want to replace is frozen.
4089 * bdrv_filter_or_cow_child(overlay_bs) == overlay_bs->backing,
4090 * because we know that overlay_bs == bs, and that @bs
4091 * either is a filter that uses ->backing or a COW format BDS
4092 * with bs->drv->supports_backing == true.
4094 if (bdrv_is_backing_chain_frozen(overlay_bs
,
4095 child_bs(overlay_bs
->backing
), errp
))
4099 reopen_state
->replace_backing_bs
= true;
4100 if (new_backing_bs
) {
4101 bdrv_ref(new_backing_bs
);
4102 reopen_state
->new_backing_bs
= new_backing_bs
;
4110 * Prepares a BlockDriverState for reopen. All changes are staged in the
4111 * 'opaque' field of the BDRVReopenState, which is used and allocated by
4112 * the block driver layer .bdrv_reopen_prepare()
4114 * bs is the BlockDriverState to reopen
4115 * flags are the new open flags
4116 * queue is the reopen queue
4118 * Returns 0 on success, non-zero on error. On error errp will be set
4121 * On failure, bdrv_reopen_abort() will be called to clean up any data.
4122 * It is the responsibility of the caller to then call the abort() or
4123 * commit() for any other BDS that have been left in a prepare() state
4126 int bdrv_reopen_prepare(BDRVReopenState
*reopen_state
, BlockReopenQueue
*queue
,
4131 Error
*local_err
= NULL
;
4134 QDict
*orig_reopen_opts
;
4135 char *discard
= NULL
;
4137 bool drv_prepared
= false;
4139 assert(reopen_state
!= NULL
);
4140 assert(reopen_state
->bs
->drv
!= NULL
);
4141 drv
= reopen_state
->bs
->drv
;
4143 /* This function and each driver's bdrv_reopen_prepare() remove
4144 * entries from reopen_state->options as they are processed, so
4145 * we need to make a copy of the original QDict. */
4146 orig_reopen_opts
= qdict_clone_shallow(reopen_state
->options
);
4148 /* Process generic block layer options */
4149 opts
= qemu_opts_create(&bdrv_runtime_opts
, NULL
, 0, &error_abort
);
4150 if (!qemu_opts_absorb_qdict(opts
, reopen_state
->options
, errp
)) {
4155 /* This was already called in bdrv_reopen_queue_child() so the flags
4156 * are up-to-date. This time we simply want to remove the options from
4157 * QemuOpts in order to indicate that they have been processed. */
4158 old_flags
= reopen_state
->flags
;
4159 update_flags_from_options(&reopen_state
->flags
, opts
);
4160 assert(old_flags
== reopen_state
->flags
);
4162 discard
= qemu_opt_get_del(opts
, BDRV_OPT_DISCARD
);
4163 if (discard
!= NULL
) {
4164 if (bdrv_parse_discard_flags(discard
, &reopen_state
->flags
) != 0) {
4165 error_setg(errp
, "Invalid discard option");
4171 reopen_state
->detect_zeroes
=
4172 bdrv_parse_detect_zeroes(opts
, reopen_state
->flags
, &local_err
);
4174 error_propagate(errp
, local_err
);
4179 /* All other options (including node-name and driver) must be unchanged.
4180 * Put them back into the QDict, so that they are checked at the end
4181 * of this function. */
4182 qemu_opts_to_qdict(opts
, reopen_state
->options
);
4184 /* If we are to stay read-only, do not allow permission change
4185 * to r/w. Attempting to set to r/w may fail if either BDRV_O_ALLOW_RDWR is
4186 * not set, or if the BDS still has copy_on_read enabled */
4187 read_only
= !(reopen_state
->flags
& BDRV_O_RDWR
);
4188 ret
= bdrv_can_set_read_only(reopen_state
->bs
, read_only
, true, &local_err
);
4190 error_propagate(errp
, local_err
);
4194 /* Calculate required permissions after reopening */
4195 bdrv_reopen_perm(queue
, reopen_state
->bs
,
4196 &reopen_state
->perm
, &reopen_state
->shared_perm
);
4198 ret
= bdrv_flush(reopen_state
->bs
);
4200 error_setg_errno(errp
, -ret
, "Error flushing drive");
4204 if (drv
->bdrv_reopen_prepare
) {
4206 * If a driver-specific option is missing, it means that we
4207 * should reset it to its default value.
4208 * But not all options allow that, so we need to check it first.
4210 ret
= bdrv_reset_options_allowed(reopen_state
->bs
,
4211 reopen_state
->options
, errp
);
4216 ret
= drv
->bdrv_reopen_prepare(reopen_state
, queue
, &local_err
);
4218 if (local_err
!= NULL
) {
4219 error_propagate(errp
, local_err
);
4221 bdrv_refresh_filename(reopen_state
->bs
);
4222 error_setg(errp
, "failed while preparing to reopen image '%s'",
4223 reopen_state
->bs
->filename
);
4228 /* It is currently mandatory to have a bdrv_reopen_prepare()
4229 * handler for each supported drv. */
4230 error_setg(errp
, "Block format '%s' used by node '%s' "
4231 "does not support reopening files", drv
->format_name
,
4232 bdrv_get_device_or_node_name(reopen_state
->bs
));
4237 drv_prepared
= true;
4240 * We must provide the 'backing' option if the BDS has a backing
4241 * file or if the image file has a backing file name as part of
4242 * its metadata. Otherwise the 'backing' option can be omitted.
4244 if (drv
->supports_backing
&& reopen_state
->backing_missing
&&
4245 (reopen_state
->bs
->backing
|| reopen_state
->bs
->backing_file
[0])) {
4246 error_setg(errp
, "backing is missing for '%s'",
4247 reopen_state
->bs
->node_name
);
4253 * Allow changing the 'backing' option. The new value can be
4254 * either a reference to an existing node (using its node name)
4255 * or NULL to simply detach the current backing file.
4257 ret
= bdrv_reopen_parse_backing(reopen_state
, errp
);
4261 qdict_del(reopen_state
->options
, "backing");
4263 /* Options that are not handled are only okay if they are unchanged
4264 * compared to the old state. It is expected that some options are only
4265 * used for the initial open, but not reopen (e.g. filename) */
4266 if (qdict_size(reopen_state
->options
)) {
4267 const QDictEntry
*entry
= qdict_first(reopen_state
->options
);
4270 QObject
*new = entry
->value
;
4271 QObject
*old
= qdict_get(reopen_state
->bs
->options
, entry
->key
);
4273 /* Allow child references (child_name=node_name) as long as they
4274 * point to the current child (i.e. everything stays the same). */
4275 if (qobject_type(new) == QTYPE_QSTRING
) {
4277 QLIST_FOREACH(child
, &reopen_state
->bs
->children
, next
) {
4278 if (!strcmp(child
->name
, entry
->key
)) {
4284 const char *str
= qobject_get_try_str(new);
4285 if (!strcmp(child
->bs
->node_name
, str
)) {
4286 continue; /* Found child with this name, skip option */
4292 * TODO: When using -drive to specify blockdev options, all values
4293 * will be strings; however, when using -blockdev, blockdev-add or
4294 * filenames using the json:{} pseudo-protocol, they will be
4296 * In contrast, reopening options are (currently) always strings
4297 * (because you can only specify them through qemu-io; all other
4298 * callers do not specify any options).
4299 * Therefore, when using anything other than -drive to create a BDS,
4300 * this cannot detect non-string options as unchanged, because
4301 * qobject_is_equal() always returns false for objects of different
4302 * type. In the future, this should be remedied by correctly typing
4303 * all options. For now, this is not too big of an issue because
4304 * the user can simply omit options which cannot be changed anyway,
4305 * so they will stay unchanged.
4307 if (!qobject_is_equal(new, old
)) {
4308 error_setg(errp
, "Cannot change the option '%s'", entry
->key
);
4312 } while ((entry
= qdict_next(reopen_state
->options
, entry
)));
4317 /* Restore the original reopen_state->options QDict */
4318 qobject_unref(reopen_state
->options
);
4319 reopen_state
->options
= qobject_ref(orig_reopen_opts
);
4322 if (ret
< 0 && drv_prepared
) {
4323 /* drv->bdrv_reopen_prepare() has succeeded, so we need to
4324 * call drv->bdrv_reopen_abort() before signaling an error
4325 * (bdrv_reopen_multiple() will not call bdrv_reopen_abort()
4326 * when the respective bdrv_reopen_prepare() has failed) */
4327 if (drv
->bdrv_reopen_abort
) {
4328 drv
->bdrv_reopen_abort(reopen_state
);
4331 qemu_opts_del(opts
);
4332 qobject_unref(orig_reopen_opts
);
4338 * Takes the staged changes for the reopen from bdrv_reopen_prepare(), and
4339 * makes them final by swapping the staging BlockDriverState contents into
4340 * the active BlockDriverState contents.
4342 void bdrv_reopen_commit(BDRVReopenState
*reopen_state
)
4345 BlockDriverState
*bs
;
4348 assert(reopen_state
!= NULL
);
4349 bs
= reopen_state
->bs
;
4351 assert(drv
!= NULL
);
4353 /* If there are any driver level actions to take */
4354 if (drv
->bdrv_reopen_commit
) {
4355 drv
->bdrv_reopen_commit(reopen_state
);
4358 /* set BDS specific flags now */
4359 qobject_unref(bs
->explicit_options
);
4360 qobject_unref(bs
->options
);
4362 bs
->explicit_options
= reopen_state
->explicit_options
;
4363 bs
->options
= reopen_state
->options
;
4364 bs
->open_flags
= reopen_state
->flags
;
4365 bs
->read_only
= !(reopen_state
->flags
& BDRV_O_RDWR
);
4366 bs
->detect_zeroes
= reopen_state
->detect_zeroes
;
4368 if (reopen_state
->replace_backing_bs
) {
4369 qdict_del(bs
->explicit_options
, "backing");
4370 qdict_del(bs
->options
, "backing");
4373 /* Remove child references from bs->options and bs->explicit_options.
4374 * Child options were already removed in bdrv_reopen_queue_child() */
4375 QLIST_FOREACH(child
, &bs
->children
, next
) {
4376 qdict_del(bs
->explicit_options
, child
->name
);
4377 qdict_del(bs
->options
, child
->name
);
4381 * Change the backing file if a new one was specified. We do this
4382 * after updating bs->options, so bdrv_refresh_filename() (called
4383 * from bdrv_set_backing_hd()) has the new values.
4385 if (reopen_state
->replace_backing_bs
) {
4386 BlockDriverState
*old_backing_bs
= child_bs(bs
->backing
);
4387 assert(!old_backing_bs
|| !old_backing_bs
->implicit
);
4388 /* Abort the permission update on the backing bs we're detaching */
4389 if (old_backing_bs
) {
4390 bdrv_abort_perm_update(old_backing_bs
);
4392 bdrv_set_backing_hd(bs
, reopen_state
->new_backing_bs
, &error_abort
);
4395 bdrv_refresh_limits(bs
, NULL
);
4399 * Abort the reopen, and delete and free the staged changes in
4402 void bdrv_reopen_abort(BDRVReopenState
*reopen_state
)
4406 assert(reopen_state
!= NULL
);
4407 drv
= reopen_state
->bs
->drv
;
4408 assert(drv
!= NULL
);
4410 if (drv
->bdrv_reopen_abort
) {
4411 drv
->bdrv_reopen_abort(reopen_state
);
4416 static void bdrv_close(BlockDriverState
*bs
)
4418 BdrvAioNotifier
*ban
, *ban_next
;
4419 BdrvChild
*child
, *next
;
4421 assert(!bs
->refcnt
);
4423 bdrv_drained_begin(bs
); /* complete I/O */
4425 bdrv_drain(bs
); /* in case flush left pending I/O */
4428 if (bs
->drv
->bdrv_close
) {
4429 /* Must unfreeze all children, so bdrv_unref_child() works */
4430 bs
->drv
->bdrv_close(bs
);
4435 QLIST_FOREACH_SAFE(child
, &bs
->children
, next
, next
) {
4436 bdrv_unref_child(bs
, child
);
4443 atomic_set(&bs
->copy_on_read
, 0);
4444 bs
->backing_file
[0] = '\0';
4445 bs
->backing_format
[0] = '\0';
4446 bs
->total_sectors
= 0;
4447 bs
->encrypted
= false;
4449 qobject_unref(bs
->options
);
4450 qobject_unref(bs
->explicit_options
);
4452 bs
->explicit_options
= NULL
;
4453 qobject_unref(bs
->full_open_options
);
4454 bs
->full_open_options
= NULL
;
4456 bdrv_release_named_dirty_bitmaps(bs
);
4457 assert(QLIST_EMPTY(&bs
->dirty_bitmaps
));
4459 QLIST_FOREACH_SAFE(ban
, &bs
->aio_notifiers
, list
, ban_next
) {
4462 QLIST_INIT(&bs
->aio_notifiers
);
4463 bdrv_drained_end(bs
);
4466 void bdrv_close_all(void)
4468 assert(job_next(NULL
) == NULL
);
4469 nbd_export_close_all();
4471 /* Drop references from requests still in flight, such as canceled block
4472 * jobs whose AIO context has not been polled yet */
4475 blk_remove_all_bs();
4476 blockdev_close_all_bdrv_states();
4478 assert(QTAILQ_EMPTY(&all_bdrv_states
));
4481 static bool should_update_child(BdrvChild
*c
, BlockDriverState
*to
)
4487 if (c
->klass
->stay_at_node
) {
4491 /* If the child @c belongs to the BDS @to, replacing the current
4492 * c->bs by @to would mean to create a loop.
4494 * Such a case occurs when appending a BDS to a backing chain.
4495 * For instance, imagine the following chain:
4497 * guest device -> node A -> further backing chain...
4499 * Now we create a new BDS B which we want to put on top of this
4500 * chain, so we first attach A as its backing node:
4505 * guest device -> node A -> further backing chain...
4507 * Finally we want to replace A by B. When doing that, we want to
4508 * replace all pointers to A by pointers to B -- except for the
4509 * pointer from B because (1) that would create a loop, and (2)
4510 * that pointer should simply stay intact:
4512 * guest device -> node B
4515 * node A -> further backing chain...
4517 * In general, when replacing a node A (c->bs) by a node B (@to),
4518 * if A is a child of B, that means we cannot replace A by B there
4519 * because that would create a loop. Silently detaching A from B
4520 * is also not really an option. So overall just leaving A in
4521 * place there is the most sensible choice.
4523 * We would also create a loop in any cases where @c is only
4524 * indirectly referenced by @to. Prevent this by returning false
4525 * if @c is found (by breadth-first search) anywhere in the whole
4530 found
= g_hash_table_new(NULL
, NULL
);
4531 g_hash_table_add(found
, to
);
4532 queue
= g_queue_new();
4533 g_queue_push_tail(queue
, to
);
4535 while (!g_queue_is_empty(queue
)) {
4536 BlockDriverState
*v
= g_queue_pop_head(queue
);
4539 QLIST_FOREACH(c2
, &v
->children
, next
) {
4545 if (g_hash_table_contains(found
, c2
->bs
)) {
4549 g_queue_push_tail(queue
, c2
->bs
);
4550 g_hash_table_add(found
, c2
->bs
);
4554 g_queue_free(queue
);
4555 g_hash_table_destroy(found
);
4560 void bdrv_replace_node(BlockDriverState
*from
, BlockDriverState
*to
,
4563 BdrvChild
*c
, *next
;
4564 GSList
*list
= NULL
, *p
;
4565 uint64_t perm
= 0, shared
= BLK_PERM_ALL
;
4568 /* Make sure that @from doesn't go away until we have successfully attached
4569 * all of its parents to @to. */
4572 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
4573 assert(bdrv_get_aio_context(from
) == bdrv_get_aio_context(to
));
4574 bdrv_drained_begin(from
);
4576 /* Put all parents into @list and calculate their cumulative permissions */
4577 QLIST_FOREACH_SAFE(c
, &from
->parents
, next_parent
, next
) {
4578 assert(c
->bs
== from
);
4579 if (!should_update_child(c
, to
)) {
4583 error_setg(errp
, "Cannot change '%s' link to '%s'",
4584 c
->name
, from
->node_name
);
4587 list
= g_slist_prepend(list
, c
);
4589 shared
&= c
->shared_perm
;
4592 /* Check whether the required permissions can be granted on @to, ignoring
4593 * all BdrvChild in @list so that they can't block themselves. */
4594 ret
= bdrv_check_update_perm(to
, NULL
, perm
, shared
, list
, NULL
, errp
);
4596 bdrv_abort_perm_update(to
);
4600 /* Now actually perform the change. We performed the permission check for
4601 * all elements of @list at once, so set the permissions all at once at the
4603 for (p
= list
; p
!= NULL
; p
= p
->next
) {
4607 bdrv_replace_child_noperm(c
, to
);
4611 bdrv_get_cumulative_perm(to
, &perm
, &shared
);
4612 bdrv_set_perm(to
, perm
, shared
);
4616 bdrv_drained_end(from
);
4621 * Add new bs contents at the top of an image chain while the chain is
4622 * live, while keeping required fields on the top layer.
4624 * This will modify the BlockDriverState fields, and swap contents
4625 * between bs_new and bs_top. Both bs_new and bs_top are modified.
4627 * bs_new must not be attached to a BlockBackend.
4629 * This function does not create any image files.
4631 * bdrv_append() takes ownership of a bs_new reference and unrefs it because
4632 * that's what the callers commonly need. bs_new will be referenced by the old
4633 * parents of bs_top after bdrv_append() returns. If the caller needs to keep a
4634 * reference of its own, it must call bdrv_ref().
4636 void bdrv_append(BlockDriverState
*bs_new
, BlockDriverState
*bs_top
,
4639 Error
*local_err
= NULL
;
4641 bdrv_set_backing_hd(bs_new
, bs_top
, &local_err
);
4643 error_propagate(errp
, local_err
);
4647 bdrv_replace_node(bs_top
, bs_new
, &local_err
);
4649 error_propagate(errp
, local_err
);
4650 bdrv_set_backing_hd(bs_new
, NULL
, &error_abort
);
4654 /* bs_new is now referenced by its new parents, we don't need the
4655 * additional reference any more. */
4660 static void bdrv_delete(BlockDriverState
*bs
)
4662 assert(bdrv_op_blocker_is_empty(bs
));
4663 assert(!bs
->refcnt
);
4665 /* remove from list, if necessary */
4666 if (bs
->node_name
[0] != '\0') {
4667 QTAILQ_REMOVE(&graph_bdrv_states
, bs
, node_list
);
4669 QTAILQ_REMOVE(&all_bdrv_states
, bs
, bs_list
);
4677 * Run consistency checks on an image
4679 * Returns 0 if the check could be completed (it doesn't mean that the image is
4680 * free of errors) or -errno when an internal error occurred. The results of the
4681 * check are stored in res.
4683 static int coroutine_fn
bdrv_co_check(BlockDriverState
*bs
,
4684 BdrvCheckResult
*res
, BdrvCheckMode fix
)
4686 if (bs
->drv
== NULL
) {
4689 if (bs
->drv
->bdrv_co_check
== NULL
) {
4693 memset(res
, 0, sizeof(*res
));
4694 return bs
->drv
->bdrv_co_check(bs
, res
, fix
);
4697 typedef struct CheckCo
{
4698 BlockDriverState
*bs
;
4699 BdrvCheckResult
*res
;
4704 static void coroutine_fn
bdrv_check_co_entry(void *opaque
)
4706 CheckCo
*cco
= opaque
;
4707 cco
->ret
= bdrv_co_check(cco
->bs
, cco
->res
, cco
->fix
);
4711 int bdrv_check(BlockDriverState
*bs
,
4712 BdrvCheckResult
*res
, BdrvCheckMode fix
)
4718 .ret
= -EINPROGRESS
,
4722 if (qemu_in_coroutine()) {
4723 /* Fast-path if already in coroutine context */
4724 bdrv_check_co_entry(&cco
);
4726 co
= qemu_coroutine_create(bdrv_check_co_entry
, &cco
);
4727 bdrv_coroutine_enter(bs
, co
);
4728 BDRV_POLL_WHILE(bs
, cco
.ret
== -EINPROGRESS
);
4737 * -EINVAL - backing format specified, but no file
4738 * -ENOSPC - can't update the backing file because no space is left in the
4740 * -ENOTSUP - format driver doesn't support changing the backing file
4742 int bdrv_change_backing_file(BlockDriverState
*bs
, const char *backing_file
,
4743 const char *backing_fmt
, bool warn
)
4745 BlockDriver
*drv
= bs
->drv
;
4752 /* Backing file format doesn't make sense without a backing file */
4753 if (backing_fmt
&& !backing_file
) {
4757 if (warn
&& backing_file
&& !backing_fmt
) {
4758 warn_report("Deprecated use of backing file without explicit "
4759 "backing format, use of this image requires "
4760 "potentially unsafe format probing");
4763 if (drv
->bdrv_change_backing_file
!= NULL
) {
4764 ret
= drv
->bdrv_change_backing_file(bs
, backing_file
, backing_fmt
);
4770 pstrcpy(bs
->backing_file
, sizeof(bs
->backing_file
), backing_file
?: "");
4771 pstrcpy(bs
->backing_format
, sizeof(bs
->backing_format
), backing_fmt
?: "");
4772 pstrcpy(bs
->auto_backing_file
, sizeof(bs
->auto_backing_file
),
4773 backing_file
?: "");
4779 * Finds the first non-filter node above bs in the chain between
4780 * active and bs. The returned node is either an immediate parent of
4781 * bs, or there are only filter nodes between the two.
4783 * Returns NULL if bs is not found in active's image chain,
4784 * or if active == bs.
4786 * Returns the bottommost base image if bs == NULL.
4788 BlockDriverState
*bdrv_find_overlay(BlockDriverState
*active
,
4789 BlockDriverState
*bs
)
4791 bs
= bdrv_skip_filters(bs
);
4792 active
= bdrv_skip_filters(active
);
4795 BlockDriverState
*next
= bdrv_backing_chain_next(active
);
4805 /* Given a BDS, searches for the base layer. */
4806 BlockDriverState
*bdrv_find_base(BlockDriverState
*bs
)
4808 return bdrv_find_overlay(bs
, NULL
);
4812 * Return true if at least one of the COW (backing) and filter links
4813 * between @bs and @base is frozen. @errp is set if that's the case.
4814 * @base must be reachable from @bs, or NULL.
4816 bool bdrv_is_backing_chain_frozen(BlockDriverState
*bs
, BlockDriverState
*base
,
4819 BlockDriverState
*i
;
4822 for (i
= bs
; i
!= base
; i
= child_bs(child
)) {
4823 child
= bdrv_filter_or_cow_child(i
);
4825 if (child
&& child
->frozen
) {
4826 error_setg(errp
, "Cannot change '%s' link from '%s' to '%s'",
4827 child
->name
, i
->node_name
, child
->bs
->node_name
);
4836 * Freeze all COW (backing) and filter links between @bs and @base.
4837 * If any of the links is already frozen the operation is aborted and
4838 * none of the links are modified.
4839 * @base must be reachable from @bs, or NULL.
4840 * Returns 0 on success. On failure returns < 0 and sets @errp.
4842 int bdrv_freeze_backing_chain(BlockDriverState
*bs
, BlockDriverState
*base
,
4845 BlockDriverState
*i
;
4848 if (bdrv_is_backing_chain_frozen(bs
, base
, errp
)) {
4852 for (i
= bs
; i
!= base
; i
= child_bs(child
)) {
4853 child
= bdrv_filter_or_cow_child(i
);
4854 if (child
&& child
->bs
->never_freeze
) {
4855 error_setg(errp
, "Cannot freeze '%s' link to '%s'",
4856 child
->name
, child
->bs
->node_name
);
4861 for (i
= bs
; i
!= base
; i
= child_bs(child
)) {
4862 child
= bdrv_filter_or_cow_child(i
);
4864 child
->frozen
= true;
4872 * Unfreeze all COW (backing) and filter links between @bs and @base.
4873 * The caller must ensure that all links are frozen before using this
4875 * @base must be reachable from @bs, or NULL.
4877 void bdrv_unfreeze_backing_chain(BlockDriverState
*bs
, BlockDriverState
*base
)
4879 BlockDriverState
*i
;
4882 for (i
= bs
; i
!= base
; i
= child_bs(child
)) {
4883 child
= bdrv_filter_or_cow_child(i
);
4885 assert(child
->frozen
);
4886 child
->frozen
= false;
4892 * Drops images above 'base' up to and including 'top', and sets the image
4893 * above 'top' to have base as its backing file.
4895 * Requires that the overlay to 'top' is opened r/w, so that the backing file
4896 * information in 'bs' can be properly updated.
4898 * E.g., this will convert the following chain:
4899 * bottom <- base <- intermediate <- top <- active
4903 * bottom <- base <- active
4905 * It is allowed for bottom==base, in which case it converts:
4907 * base <- intermediate <- top <- active
4913 * If backing_file_str is non-NULL, it will be used when modifying top's
4914 * overlay image metadata.
4917 * if active == top, that is considered an error
4920 int bdrv_drop_intermediate(BlockDriverState
*top
, BlockDriverState
*base
,
4921 const char *backing_file_str
)
4923 BlockDriverState
*explicit_top
= top
;
4924 bool update_inherits_from
;
4925 BdrvChild
*c
, *next
;
4926 Error
*local_err
= NULL
;
4930 bdrv_subtree_drained_begin(top
);
4932 if (!top
->drv
|| !base
->drv
) {
4936 /* Make sure that base is in the backing chain of top */
4937 if (!bdrv_chain_contains(top
, base
)) {
4941 /* This function changes all links that point to top and makes
4942 * them point to base. Check that none of them is frozen. */
4943 QLIST_FOREACH(c
, &top
->parents
, next_parent
) {
4949 /* If 'base' recursively inherits from 'top' then we should set
4950 * base->inherits_from to top->inherits_from after 'top' and all
4951 * other intermediate nodes have been dropped.
4952 * If 'top' is an implicit node (e.g. "commit_top") we should skip
4953 * it because no one inherits from it. We use explicit_top for that. */
4954 explicit_top
= bdrv_skip_implicit_filters(explicit_top
);
4955 update_inherits_from
= bdrv_inherits_from_recursive(base
, explicit_top
);
4957 /* success - we can delete the intermediate states, and link top->base */
4958 /* TODO Check graph modification op blockers (BLK_PERM_GRAPH_MOD) once
4959 * we've figured out how they should work. */
4960 if (!backing_file_str
) {
4961 bdrv_refresh_filename(base
);
4962 backing_file_str
= base
->filename
;
4965 QLIST_FOREACH_SAFE(c
, &top
->parents
, next_parent
, next
) {
4966 /* Check whether we are allowed to switch c from top to base */
4967 GSList
*ignore_children
= g_slist_prepend(NULL
, c
);
4968 ret
= bdrv_check_update_perm(base
, NULL
, c
->perm
, c
->shared_perm
,
4969 ignore_children
, NULL
, &local_err
);
4970 g_slist_free(ignore_children
);
4972 error_report_err(local_err
);
4976 /* If so, update the backing file path in the image file */
4977 if (c
->klass
->update_filename
) {
4978 ret
= c
->klass
->update_filename(c
, base
, backing_file_str
,
4981 bdrv_abort_perm_update(base
);
4982 error_report_err(local_err
);
4988 * Do the actual switch in the in-memory graph.
4989 * Completes bdrv_check_update_perm() transaction internally.
4990 * c->frozen is false, we have checked that above.
4993 bdrv_replace_child(c
, base
);
4997 if (update_inherits_from
) {
4998 base
->inherits_from
= explicit_top
->inherits_from
;
5003 bdrv_subtree_drained_end(top
);
5009 * Implementation of BlockDriver.bdrv_get_allocated_file_size() that
5010 * sums the size of all data-bearing children. (This excludes backing
5013 static int64_t bdrv_sum_allocated_file_size(BlockDriverState
*bs
)
5016 int64_t child_size
, sum
= 0;
5018 QLIST_FOREACH(child
, &bs
->children
, next
) {
5019 if (child
->role
& (BDRV_CHILD_DATA
| BDRV_CHILD_METADATA
|
5020 BDRV_CHILD_FILTERED
))
5022 child_size
= bdrv_get_allocated_file_size(child
->bs
);
5023 if (child_size
< 0) {
5034 * Length of a allocated file in bytes. Sparse files are counted by actual
5035 * allocated space. Return < 0 if error or unknown.
5037 int64_t bdrv_get_allocated_file_size(BlockDriverState
*bs
)
5039 BlockDriver
*drv
= bs
->drv
;
5043 if (drv
->bdrv_get_allocated_file_size
) {
5044 return drv
->bdrv_get_allocated_file_size(bs
);
5047 if (drv
->bdrv_file_open
) {
5049 * Protocol drivers default to -ENOTSUP (most of their data is
5050 * not stored in any of their children (if they even have any),
5051 * so there is no generic way to figure it out).
5054 } else if (drv
->is_filter
) {
5055 /* Filter drivers default to the size of their filtered child */
5056 return bdrv_get_allocated_file_size(bdrv_filter_bs(bs
));
5058 /* Other drivers default to summing their children's sizes */
5059 return bdrv_sum_allocated_file_size(bs
);
5065 * @drv: Format driver
5066 * @opts: Creation options for new image
5067 * @in_bs: Existing image containing data for new image (may be NULL)
5068 * @errp: Error object
5069 * Returns: A #BlockMeasureInfo (free using qapi_free_BlockMeasureInfo())
5072 * Calculate file size required to create a new image.
5074 * If @in_bs is given then space for allocated clusters and zero clusters
5075 * from that image are included in the calculation. If @opts contains a
5076 * backing file that is shared by @in_bs then backing clusters may be omitted
5077 * from the calculation.
5079 * If @in_bs is NULL then the calculation includes no allocated clusters
5080 * unless a preallocation option is given in @opts.
5082 * Note that @in_bs may use a different BlockDriver from @drv.
5084 * If an error occurs the @errp pointer is set.
5086 BlockMeasureInfo
*bdrv_measure(BlockDriver
*drv
, QemuOpts
*opts
,
5087 BlockDriverState
*in_bs
, Error
**errp
)
5089 if (!drv
->bdrv_measure
) {
5090 error_setg(errp
, "Block driver '%s' does not support size measurement",
5095 return drv
->bdrv_measure(opts
, in_bs
, errp
);
5099 * Return number of sectors on success, -errno on error.
5101 int64_t bdrv_nb_sectors(BlockDriverState
*bs
)
5103 BlockDriver
*drv
= bs
->drv
;
5108 if (drv
->has_variable_length
) {
5109 int ret
= refresh_total_sectors(bs
, bs
->total_sectors
);
5114 return bs
->total_sectors
;
5118 * Return length in bytes on success, -errno on error.
5119 * The length is always a multiple of BDRV_SECTOR_SIZE.
5121 int64_t bdrv_getlength(BlockDriverState
*bs
)
5123 int64_t ret
= bdrv_nb_sectors(bs
);
5125 ret
= ret
> INT64_MAX
/ BDRV_SECTOR_SIZE
? -EFBIG
: ret
;
5126 return ret
< 0 ? ret
: ret
* BDRV_SECTOR_SIZE
;
5129 /* return 0 as number of sectors if no device present or error */
5130 void bdrv_get_geometry(BlockDriverState
*bs
, uint64_t *nb_sectors_ptr
)
5132 int64_t nb_sectors
= bdrv_nb_sectors(bs
);
5134 *nb_sectors_ptr
= nb_sectors
< 0 ? 0 : nb_sectors
;
5137 bool bdrv_is_sg(BlockDriverState
*bs
)
5143 * Return whether the given node supports compressed writes.
5145 bool bdrv_supports_compressed_writes(BlockDriverState
*bs
)
5147 BlockDriverState
*filtered
;
5149 if (!bs
->drv
|| !block_driver_can_compress(bs
->drv
)) {
5153 filtered
= bdrv_filter_bs(bs
);
5156 * Filters can only forward compressed writes, so we have to
5159 return bdrv_supports_compressed_writes(filtered
);
5165 const char *bdrv_get_format_name(BlockDriverState
*bs
)
5167 return bs
->drv
? bs
->drv
->format_name
: NULL
;
5170 static int qsort_strcmp(const void *a
, const void *b
)
5172 return strcmp(*(char *const *)a
, *(char *const *)b
);
5175 void bdrv_iterate_format(void (*it
)(void *opaque
, const char *name
),
5176 void *opaque
, bool read_only
)
5181 const char **formats
= NULL
;
5183 QLIST_FOREACH(drv
, &bdrv_drivers
, list
) {
5184 if (drv
->format_name
) {
5188 if (use_bdrv_whitelist
&& !bdrv_is_whitelisted(drv
, read_only
)) {
5192 while (formats
&& i
&& !found
) {
5193 found
= !strcmp(formats
[--i
], drv
->format_name
);
5197 formats
= g_renew(const char *, formats
, count
+ 1);
5198 formats
[count
++] = drv
->format_name
;
5203 for (i
= 0; i
< (int)ARRAY_SIZE(block_driver_modules
); i
++) {
5204 const char *format_name
= block_driver_modules
[i
].format_name
;
5210 if (use_bdrv_whitelist
&&
5211 !bdrv_format_is_whitelisted(format_name
, read_only
)) {
5215 while (formats
&& j
&& !found
) {
5216 found
= !strcmp(formats
[--j
], format_name
);
5220 formats
= g_renew(const char *, formats
, count
+ 1);
5221 formats
[count
++] = format_name
;
5226 qsort(formats
, count
, sizeof(formats
[0]), qsort_strcmp
);
5228 for (i
= 0; i
< count
; i
++) {
5229 it(opaque
, formats
[i
]);
5235 /* This function is to find a node in the bs graph */
5236 BlockDriverState
*bdrv_find_node(const char *node_name
)
5238 BlockDriverState
*bs
;
5242 QTAILQ_FOREACH(bs
, &graph_bdrv_states
, node_list
) {
5243 if (!strcmp(node_name
, bs
->node_name
)) {
5250 /* Put this QMP function here so it can access the static graph_bdrv_states. */
5251 BlockDeviceInfoList
*bdrv_named_nodes_list(bool flat
,
5254 BlockDeviceInfoList
*list
, *entry
;
5255 BlockDriverState
*bs
;
5258 QTAILQ_FOREACH(bs
, &graph_bdrv_states
, node_list
) {
5259 BlockDeviceInfo
*info
= bdrv_block_device_info(NULL
, bs
, flat
, errp
);
5261 qapi_free_BlockDeviceInfoList(list
);
5264 entry
= g_malloc0(sizeof(*entry
));
5265 entry
->value
= info
;
5273 #define QAPI_LIST_ADD(list, element) do { \
5274 typeof(list) _tmp = g_new(typeof(*(list)), 1); \
5275 _tmp->value = (element); \
5276 _tmp->next = (list); \
5280 typedef struct XDbgBlockGraphConstructor
{
5281 XDbgBlockGraph
*graph
;
5282 GHashTable
*graph_nodes
;
5283 } XDbgBlockGraphConstructor
;
5285 static XDbgBlockGraphConstructor
*xdbg_graph_new(void)
5287 XDbgBlockGraphConstructor
*gr
= g_new(XDbgBlockGraphConstructor
, 1);
5289 gr
->graph
= g_new0(XDbgBlockGraph
, 1);
5290 gr
->graph_nodes
= g_hash_table_new(NULL
, NULL
);
5295 static XDbgBlockGraph
*xdbg_graph_finalize(XDbgBlockGraphConstructor
*gr
)
5297 XDbgBlockGraph
*graph
= gr
->graph
;
5299 g_hash_table_destroy(gr
->graph_nodes
);
5305 static uintptr_t xdbg_graph_node_num(XDbgBlockGraphConstructor
*gr
, void *node
)
5307 uintptr_t ret
= (uintptr_t)g_hash_table_lookup(gr
->graph_nodes
, node
);
5314 * Start counting from 1, not 0, because 0 interferes with not-found (NULL)
5315 * answer of g_hash_table_lookup.
5317 ret
= g_hash_table_size(gr
->graph_nodes
) + 1;
5318 g_hash_table_insert(gr
->graph_nodes
, node
, (void *)ret
);
5323 static void xdbg_graph_add_node(XDbgBlockGraphConstructor
*gr
, void *node
,
5324 XDbgBlockGraphNodeType type
, const char *name
)
5326 XDbgBlockGraphNode
*n
;
5328 n
= g_new0(XDbgBlockGraphNode
, 1);
5330 n
->id
= xdbg_graph_node_num(gr
, node
);
5332 n
->name
= g_strdup(name
);
5334 QAPI_LIST_ADD(gr
->graph
->nodes
, n
);
5337 static void xdbg_graph_add_edge(XDbgBlockGraphConstructor
*gr
, void *parent
,
5338 const BdrvChild
*child
)
5340 BlockPermission qapi_perm
;
5341 XDbgBlockGraphEdge
*edge
;
5343 edge
= g_new0(XDbgBlockGraphEdge
, 1);
5345 edge
->parent
= xdbg_graph_node_num(gr
, parent
);
5346 edge
->child
= xdbg_graph_node_num(gr
, child
->bs
);
5347 edge
->name
= g_strdup(child
->name
);
5349 for (qapi_perm
= 0; qapi_perm
< BLOCK_PERMISSION__MAX
; qapi_perm
++) {
5350 uint64_t flag
= bdrv_qapi_perm_to_blk_perm(qapi_perm
);
5352 if (flag
& child
->perm
) {
5353 QAPI_LIST_ADD(edge
->perm
, qapi_perm
);
5355 if (flag
& child
->shared_perm
) {
5356 QAPI_LIST_ADD(edge
->shared_perm
, qapi_perm
);
5360 QAPI_LIST_ADD(gr
->graph
->edges
, edge
);
5364 XDbgBlockGraph
*bdrv_get_xdbg_block_graph(Error
**errp
)
5368 BlockDriverState
*bs
;
5370 XDbgBlockGraphConstructor
*gr
= xdbg_graph_new();
5372 for (blk
= blk_all_next(NULL
); blk
; blk
= blk_all_next(blk
)) {
5373 char *allocated_name
= NULL
;
5374 const char *name
= blk_name(blk
);
5377 name
= allocated_name
= blk_get_attached_dev_id(blk
);
5379 xdbg_graph_add_node(gr
, blk
, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_BACKEND
,
5381 g_free(allocated_name
);
5382 if (blk_root(blk
)) {
5383 xdbg_graph_add_edge(gr
, blk
, blk_root(blk
));
5387 for (job
= block_job_next(NULL
); job
; job
= block_job_next(job
)) {
5390 xdbg_graph_add_node(gr
, job
, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_JOB
,
5392 for (el
= job
->nodes
; el
; el
= el
->next
) {
5393 xdbg_graph_add_edge(gr
, job
, (BdrvChild
*)el
->data
);
5397 QTAILQ_FOREACH(bs
, &graph_bdrv_states
, node_list
) {
5398 xdbg_graph_add_node(gr
, bs
, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_DRIVER
,
5400 QLIST_FOREACH(child
, &bs
->children
, next
) {
5401 xdbg_graph_add_edge(gr
, bs
, child
);
5405 return xdbg_graph_finalize(gr
);
5408 BlockDriverState
*bdrv_lookup_bs(const char *device
,
5409 const char *node_name
,
5413 BlockDriverState
*bs
;
5416 blk
= blk_by_name(device
);
5421 error_setg(errp
, "Device '%s' has no medium", device
);
5429 bs
= bdrv_find_node(node_name
);
5436 error_setg(errp
, "Cannot find device=%s nor node_name=%s",
5437 device
? device
: "",
5438 node_name
? node_name
: "");
5442 /* If 'base' is in the same chain as 'top', return true. Otherwise,
5443 * return false. If either argument is NULL, return false. */
5444 bool bdrv_chain_contains(BlockDriverState
*top
, BlockDriverState
*base
)
5446 while (top
&& top
!= base
) {
5447 top
= bdrv_filter_or_cow_bs(top
);
5453 BlockDriverState
*bdrv_next_node(BlockDriverState
*bs
)
5456 return QTAILQ_FIRST(&graph_bdrv_states
);
5458 return QTAILQ_NEXT(bs
, node_list
);
5461 BlockDriverState
*bdrv_next_all_states(BlockDriverState
*bs
)
5464 return QTAILQ_FIRST(&all_bdrv_states
);
5466 return QTAILQ_NEXT(bs
, bs_list
);
5469 const char *bdrv_get_node_name(const BlockDriverState
*bs
)
5471 return bs
->node_name
;
5474 const char *bdrv_get_parent_name(const BlockDriverState
*bs
)
5479 /* If multiple parents have a name, just pick the first one. */
5480 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
5481 if (c
->klass
->get_name
) {
5482 name
= c
->klass
->get_name(c
);
5483 if (name
&& *name
) {
5492 /* TODO check what callers really want: bs->node_name or blk_name() */
5493 const char *bdrv_get_device_name(const BlockDriverState
*bs
)
5495 return bdrv_get_parent_name(bs
) ?: "";
5498 /* This can be used to identify nodes that might not have a device
5499 * name associated. Since node and device names live in the same
5500 * namespace, the result is unambiguous. The exception is if both are
5501 * absent, then this returns an empty (non-null) string. */
5502 const char *bdrv_get_device_or_node_name(const BlockDriverState
*bs
)
5504 return bdrv_get_parent_name(bs
) ?: bs
->node_name
;
5507 int bdrv_get_flags(BlockDriverState
*bs
)
5509 return bs
->open_flags
;
5512 int bdrv_has_zero_init_1(BlockDriverState
*bs
)
5517 int bdrv_has_zero_init(BlockDriverState
*bs
)
5519 BlockDriverState
*filtered
;
5525 /* If BS is a copy on write image, it is initialized to
5526 the contents of the base image, which may not be zeroes. */
5527 if (bdrv_cow_child(bs
)) {
5530 if (bs
->drv
->bdrv_has_zero_init
) {
5531 return bs
->drv
->bdrv_has_zero_init(bs
);
5534 filtered
= bdrv_filter_bs(bs
);
5536 return bdrv_has_zero_init(filtered
);
5543 bool bdrv_can_write_zeroes_with_unmap(BlockDriverState
*bs
)
5545 if (!(bs
->open_flags
& BDRV_O_UNMAP
)) {
5549 return bs
->supported_zero_flags
& BDRV_REQ_MAY_UNMAP
;
5552 void bdrv_get_backing_filename(BlockDriverState
*bs
,
5553 char *filename
, int filename_size
)
5555 pstrcpy(filename
, filename_size
, bs
->backing_file
);
5558 int bdrv_get_info(BlockDriverState
*bs
, BlockDriverInfo
*bdi
)
5560 BlockDriver
*drv
= bs
->drv
;
5561 /* if bs->drv == NULL, bs is closed, so there's nothing to do here */
5565 if (!drv
->bdrv_get_info
) {
5566 BlockDriverState
*filtered
= bdrv_filter_bs(bs
);
5568 return bdrv_get_info(filtered
, bdi
);
5572 memset(bdi
, 0, sizeof(*bdi
));
5573 return drv
->bdrv_get_info(bs
, bdi
);
5576 ImageInfoSpecific
*bdrv_get_specific_info(BlockDriverState
*bs
,
5579 BlockDriver
*drv
= bs
->drv
;
5580 if (drv
&& drv
->bdrv_get_specific_info
) {
5581 return drv
->bdrv_get_specific_info(bs
, errp
);
5586 BlockStatsSpecific
*bdrv_get_specific_stats(BlockDriverState
*bs
)
5588 BlockDriver
*drv
= bs
->drv
;
5589 if (!drv
|| !drv
->bdrv_get_specific_stats
) {
5592 return drv
->bdrv_get_specific_stats(bs
);
5595 void bdrv_debug_event(BlockDriverState
*bs
, BlkdebugEvent event
)
5597 if (!bs
|| !bs
->drv
|| !bs
->drv
->bdrv_debug_event
) {
5601 bs
->drv
->bdrv_debug_event(bs
, event
);
5604 static BlockDriverState
*bdrv_find_debug_node(BlockDriverState
*bs
)
5606 while (bs
&& bs
->drv
&& !bs
->drv
->bdrv_debug_breakpoint
) {
5607 bs
= bdrv_primary_bs(bs
);
5610 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_breakpoint
) {
5611 assert(bs
->drv
->bdrv_debug_remove_breakpoint
);
5618 int bdrv_debug_breakpoint(BlockDriverState
*bs
, const char *event
,
5621 bs
= bdrv_find_debug_node(bs
);
5623 return bs
->drv
->bdrv_debug_breakpoint(bs
, event
, tag
);
5629 int bdrv_debug_remove_breakpoint(BlockDriverState
*bs
, const char *tag
)
5631 bs
= bdrv_find_debug_node(bs
);
5633 return bs
->drv
->bdrv_debug_remove_breakpoint(bs
, tag
);
5639 int bdrv_debug_resume(BlockDriverState
*bs
, const char *tag
)
5641 while (bs
&& (!bs
->drv
|| !bs
->drv
->bdrv_debug_resume
)) {
5642 bs
= bdrv_primary_bs(bs
);
5645 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_resume
) {
5646 return bs
->drv
->bdrv_debug_resume(bs
, tag
);
5652 bool bdrv_debug_is_suspended(BlockDriverState
*bs
, const char *tag
)
5654 while (bs
&& bs
->drv
&& !bs
->drv
->bdrv_debug_is_suspended
) {
5655 bs
= bdrv_primary_bs(bs
);
5658 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_is_suspended
) {
5659 return bs
->drv
->bdrv_debug_is_suspended(bs
, tag
);
5665 /* backing_file can either be relative, or absolute, or a protocol. If it is
5666 * relative, it must be relative to the chain. So, passing in bs->filename
5667 * from a BDS as backing_file should not be done, as that may be relative to
5668 * the CWD rather than the chain. */
5669 BlockDriverState
*bdrv_find_backing_image(BlockDriverState
*bs
,
5670 const char *backing_file
)
5672 char *filename_full
= NULL
;
5673 char *backing_file_full
= NULL
;
5674 char *filename_tmp
= NULL
;
5675 int is_protocol
= 0;
5676 BlockDriverState
*curr_bs
= NULL
;
5677 BlockDriverState
*retval
= NULL
;
5678 BlockDriverState
*bs_below
;
5680 if (!bs
|| !bs
->drv
|| !backing_file
) {
5684 filename_full
= g_malloc(PATH_MAX
);
5685 backing_file_full
= g_malloc(PATH_MAX
);
5687 is_protocol
= path_has_protocol(backing_file
);
5690 * Being largely a legacy function, skip any filters here
5691 * (because filters do not have normal filenames, so they cannot
5692 * match anyway; and allowing json:{} filenames is a bit out of
5695 for (curr_bs
= bdrv_skip_filters(bs
);
5696 bdrv_cow_child(curr_bs
) != NULL
;
5699 bs_below
= bdrv_backing_chain_next(curr_bs
);
5701 /* If either of the filename paths is actually a protocol, then
5702 * compare unmodified paths; otherwise make paths relative */
5703 if (is_protocol
|| path_has_protocol(curr_bs
->backing_file
)) {
5704 char *backing_file_full_ret
;
5706 if (strcmp(backing_file
, curr_bs
->backing_file
) == 0) {
5710 /* Also check against the full backing filename for the image */
5711 backing_file_full_ret
= bdrv_get_full_backing_filename(curr_bs
,
5713 if (backing_file_full_ret
) {
5714 bool equal
= strcmp(backing_file
, backing_file_full_ret
) == 0;
5715 g_free(backing_file_full_ret
);
5722 /* If not an absolute filename path, make it relative to the current
5723 * image's filename path */
5724 filename_tmp
= bdrv_make_absolute_filename(curr_bs
, backing_file
,
5726 /* We are going to compare canonicalized absolute pathnames */
5727 if (!filename_tmp
|| !realpath(filename_tmp
, filename_full
)) {
5728 g_free(filename_tmp
);
5731 g_free(filename_tmp
);
5733 /* We need to make sure the backing filename we are comparing against
5734 * is relative to the current image filename (or absolute) */
5735 filename_tmp
= bdrv_get_full_backing_filename(curr_bs
, NULL
);
5736 if (!filename_tmp
|| !realpath(filename_tmp
, backing_file_full
)) {
5737 g_free(filename_tmp
);
5740 g_free(filename_tmp
);
5742 if (strcmp(backing_file_full
, filename_full
) == 0) {
5749 g_free(filename_full
);
5750 g_free(backing_file_full
);
5754 void bdrv_init(void)
5756 module_call_init(MODULE_INIT_BLOCK
);
5759 void bdrv_init_with_whitelist(void)
5761 use_bdrv_whitelist
= 1;
5765 static void coroutine_fn
bdrv_co_invalidate_cache(BlockDriverState
*bs
,
5768 BdrvChild
*child
, *parent
;
5769 uint64_t perm
, shared_perm
;
5770 Error
*local_err
= NULL
;
5772 BdrvDirtyBitmap
*bm
;
5778 QLIST_FOREACH(child
, &bs
->children
, next
) {
5779 bdrv_co_invalidate_cache(child
->bs
, &local_err
);
5781 error_propagate(errp
, local_err
);
5787 * Update permissions, they may differ for inactive nodes.
5789 * Note that the required permissions of inactive images are always a
5790 * subset of the permissions required after activating the image. This
5791 * allows us to just get the permissions upfront without restricting
5792 * drv->bdrv_invalidate_cache().
5794 * It also means that in error cases, we don't have to try and revert to
5795 * the old permissions (which is an operation that could fail, too). We can
5796 * just keep the extended permissions for the next time that an activation
5797 * of the image is tried.
5799 if (bs
->open_flags
& BDRV_O_INACTIVE
) {
5800 bs
->open_flags
&= ~BDRV_O_INACTIVE
;
5801 bdrv_get_cumulative_perm(bs
, &perm
, &shared_perm
);
5802 ret
= bdrv_check_perm(bs
, NULL
, perm
, shared_perm
, NULL
, NULL
, errp
);
5804 bs
->open_flags
|= BDRV_O_INACTIVE
;
5807 bdrv_set_perm(bs
, perm
, shared_perm
);
5809 if (bs
->drv
->bdrv_co_invalidate_cache
) {
5810 bs
->drv
->bdrv_co_invalidate_cache(bs
, &local_err
);
5812 bs
->open_flags
|= BDRV_O_INACTIVE
;
5813 error_propagate(errp
, local_err
);
5818 FOR_EACH_DIRTY_BITMAP(bs
, bm
) {
5819 bdrv_dirty_bitmap_skip_store(bm
, false);
5822 ret
= refresh_total_sectors(bs
, bs
->total_sectors
);
5824 bs
->open_flags
|= BDRV_O_INACTIVE
;
5825 error_setg_errno(errp
, -ret
, "Could not refresh total sector count");
5830 QLIST_FOREACH(parent
, &bs
->parents
, next_parent
) {
5831 if (parent
->klass
->activate
) {
5832 parent
->klass
->activate(parent
, &local_err
);
5834 bs
->open_flags
|= BDRV_O_INACTIVE
;
5835 error_propagate(errp
, local_err
);
5842 typedef struct InvalidateCacheCo
{
5843 BlockDriverState
*bs
;
5846 } InvalidateCacheCo
;
5848 static void coroutine_fn
bdrv_invalidate_cache_co_entry(void *opaque
)
5850 InvalidateCacheCo
*ico
= opaque
;
5851 bdrv_co_invalidate_cache(ico
->bs
, ico
->errp
);
5856 void bdrv_invalidate_cache(BlockDriverState
*bs
, Error
**errp
)
5859 InvalidateCacheCo ico
= {
5865 if (qemu_in_coroutine()) {
5866 /* Fast-path if already in coroutine context */
5867 bdrv_invalidate_cache_co_entry(&ico
);
5869 co
= qemu_coroutine_create(bdrv_invalidate_cache_co_entry
, &ico
);
5870 bdrv_coroutine_enter(bs
, co
);
5871 BDRV_POLL_WHILE(bs
, !ico
.done
);
5875 void bdrv_invalidate_cache_all(Error
**errp
)
5877 BlockDriverState
*bs
;
5878 Error
*local_err
= NULL
;
5879 BdrvNextIterator it
;
5881 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
5882 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
5884 aio_context_acquire(aio_context
);
5885 bdrv_invalidate_cache(bs
, &local_err
);
5886 aio_context_release(aio_context
);
5888 error_propagate(errp
, local_err
);
5889 bdrv_next_cleanup(&it
);
5895 static bool bdrv_has_bds_parent(BlockDriverState
*bs
, bool only_active
)
5899 QLIST_FOREACH(parent
, &bs
->parents
, next_parent
) {
5900 if (parent
->klass
->parent_is_bds
) {
5901 BlockDriverState
*parent_bs
= parent
->opaque
;
5902 if (!only_active
|| !(parent_bs
->open_flags
& BDRV_O_INACTIVE
)) {
5911 static int bdrv_inactivate_recurse(BlockDriverState
*bs
)
5913 BdrvChild
*child
, *parent
;
5914 bool tighten_restrictions
;
5915 uint64_t perm
, shared_perm
;
5922 /* Make sure that we don't inactivate a child before its parent.
5923 * It will be covered by recursion from the yet active parent. */
5924 if (bdrv_has_bds_parent(bs
, true)) {
5928 assert(!(bs
->open_flags
& BDRV_O_INACTIVE
));
5930 /* Inactivate this node */
5931 if (bs
->drv
->bdrv_inactivate
) {
5932 ret
= bs
->drv
->bdrv_inactivate(bs
);
5938 QLIST_FOREACH(parent
, &bs
->parents
, next_parent
) {
5939 if (parent
->klass
->inactivate
) {
5940 ret
= parent
->klass
->inactivate(parent
);
5947 bs
->open_flags
|= BDRV_O_INACTIVE
;
5949 /* Update permissions, they may differ for inactive nodes */
5950 bdrv_get_cumulative_perm(bs
, &perm
, &shared_perm
);
5951 ret
= bdrv_check_perm(bs
, NULL
, perm
, shared_perm
, NULL
,
5952 &tighten_restrictions
, NULL
);
5953 assert(tighten_restrictions
== false);
5955 /* We only tried to loosen restrictions, so errors are not fatal */
5956 bdrv_abort_perm_update(bs
);
5958 bdrv_set_perm(bs
, perm
, shared_perm
);
5962 /* Recursively inactivate children */
5963 QLIST_FOREACH(child
, &bs
->children
, next
) {
5964 ret
= bdrv_inactivate_recurse(child
->bs
);
5973 int bdrv_inactivate_all(void)
5975 BlockDriverState
*bs
= NULL
;
5976 BdrvNextIterator it
;
5978 GSList
*aio_ctxs
= NULL
, *ctx
;
5980 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
5981 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
5983 if (!g_slist_find(aio_ctxs
, aio_context
)) {
5984 aio_ctxs
= g_slist_prepend(aio_ctxs
, aio_context
);
5985 aio_context_acquire(aio_context
);
5989 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
5990 /* Nodes with BDS parents are covered by recursion from the last
5991 * parent that gets inactivated. Don't inactivate them a second
5992 * time if that has already happened. */
5993 if (bdrv_has_bds_parent(bs
, false)) {
5996 ret
= bdrv_inactivate_recurse(bs
);
5998 bdrv_next_cleanup(&it
);
6004 for (ctx
= aio_ctxs
; ctx
!= NULL
; ctx
= ctx
->next
) {
6005 AioContext
*aio_context
= ctx
->data
;
6006 aio_context_release(aio_context
);
6008 g_slist_free(aio_ctxs
);
6013 /**************************************************************/
6014 /* removable device support */
6017 * Return TRUE if the media is present
6019 bool bdrv_is_inserted(BlockDriverState
*bs
)
6021 BlockDriver
*drv
= bs
->drv
;
6027 if (drv
->bdrv_is_inserted
) {
6028 return drv
->bdrv_is_inserted(bs
);
6030 QLIST_FOREACH(child
, &bs
->children
, next
) {
6031 if (!bdrv_is_inserted(child
->bs
)) {
6039 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
6041 void bdrv_eject(BlockDriverState
*bs
, bool eject_flag
)
6043 BlockDriver
*drv
= bs
->drv
;
6045 if (drv
&& drv
->bdrv_eject
) {
6046 drv
->bdrv_eject(bs
, eject_flag
);
6051 * Lock or unlock the media (if it is locked, the user won't be able
6052 * to eject it manually).
6054 void bdrv_lock_medium(BlockDriverState
*bs
, bool locked
)
6056 BlockDriver
*drv
= bs
->drv
;
6058 trace_bdrv_lock_medium(bs
, locked
);
6060 if (drv
&& drv
->bdrv_lock_medium
) {
6061 drv
->bdrv_lock_medium(bs
, locked
);
6065 /* Get a reference to bs */
6066 void bdrv_ref(BlockDriverState
*bs
)
6071 /* Release a previously grabbed reference to bs.
6072 * If after releasing, reference count is zero, the BlockDriverState is
6074 void bdrv_unref(BlockDriverState
*bs
)
6079 assert(bs
->refcnt
> 0);
6080 if (--bs
->refcnt
== 0) {
6085 struct BdrvOpBlocker
{
6087 QLIST_ENTRY(BdrvOpBlocker
) list
;
6090 bool bdrv_op_is_blocked(BlockDriverState
*bs
, BlockOpType op
, Error
**errp
)
6092 BdrvOpBlocker
*blocker
;
6093 assert((int) op
>= 0 && op
< BLOCK_OP_TYPE_MAX
);
6094 if (!QLIST_EMPTY(&bs
->op_blockers
[op
])) {
6095 blocker
= QLIST_FIRST(&bs
->op_blockers
[op
]);
6096 error_propagate_prepend(errp
, error_copy(blocker
->reason
),
6097 "Node '%s' is busy: ",
6098 bdrv_get_device_or_node_name(bs
));
6104 void bdrv_op_block(BlockDriverState
*bs
, BlockOpType op
, Error
*reason
)
6106 BdrvOpBlocker
*blocker
;
6107 assert((int) op
>= 0 && op
< BLOCK_OP_TYPE_MAX
);
6109 blocker
= g_new0(BdrvOpBlocker
, 1);
6110 blocker
->reason
= reason
;
6111 QLIST_INSERT_HEAD(&bs
->op_blockers
[op
], blocker
, list
);
6114 void bdrv_op_unblock(BlockDriverState
*bs
, BlockOpType op
, Error
*reason
)
6116 BdrvOpBlocker
*blocker
, *next
;
6117 assert((int) op
>= 0 && op
< BLOCK_OP_TYPE_MAX
);
6118 QLIST_FOREACH_SAFE(blocker
, &bs
->op_blockers
[op
], list
, next
) {
6119 if (blocker
->reason
== reason
) {
6120 QLIST_REMOVE(blocker
, list
);
6126 void bdrv_op_block_all(BlockDriverState
*bs
, Error
*reason
)
6129 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
6130 bdrv_op_block(bs
, i
, reason
);
6134 void bdrv_op_unblock_all(BlockDriverState
*bs
, Error
*reason
)
6137 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
6138 bdrv_op_unblock(bs
, i
, reason
);
6142 bool bdrv_op_blocker_is_empty(BlockDriverState
*bs
)
6146 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
6147 if (!QLIST_EMPTY(&bs
->op_blockers
[i
])) {
6154 void bdrv_img_create(const char *filename
, const char *fmt
,
6155 const char *base_filename
, const char *base_fmt
,
6156 char *options
, uint64_t img_size
, int flags
, bool quiet
,
6159 QemuOptsList
*create_opts
= NULL
;
6160 QemuOpts
*opts
= NULL
;
6161 const char *backing_fmt
, *backing_file
;
6163 BlockDriver
*drv
, *proto_drv
;
6164 Error
*local_err
= NULL
;
6167 /* Find driver and parse its options */
6168 drv
= bdrv_find_format(fmt
);
6170 error_setg(errp
, "Unknown file format '%s'", fmt
);
6174 proto_drv
= bdrv_find_protocol(filename
, true, errp
);
6179 if (!drv
->create_opts
) {
6180 error_setg(errp
, "Format driver '%s' does not support image creation",
6185 if (!proto_drv
->create_opts
) {
6186 error_setg(errp
, "Protocol driver '%s' does not support image creation",
6187 proto_drv
->format_name
);
6191 /* Create parameter list */
6192 create_opts
= qemu_opts_append(create_opts
, drv
->create_opts
);
6193 create_opts
= qemu_opts_append(create_opts
, proto_drv
->create_opts
);
6195 opts
= qemu_opts_create(create_opts
, NULL
, 0, &error_abort
);
6197 /* Parse -o options */
6199 if (!qemu_opts_do_parse(opts
, options
, NULL
, errp
)) {
6204 if (!qemu_opt_get(opts
, BLOCK_OPT_SIZE
)) {
6205 qemu_opt_set_number(opts
, BLOCK_OPT_SIZE
, img_size
, &error_abort
);
6206 } else if (img_size
!= UINT64_C(-1)) {
6207 error_setg(errp
, "The image size must be specified only once");
6211 if (base_filename
) {
6212 if (!qemu_opt_set(opts
, BLOCK_OPT_BACKING_FILE
, base_filename
,
6214 error_setg(errp
, "Backing file not supported for file format '%s'",
6221 if (!qemu_opt_set(opts
, BLOCK_OPT_BACKING_FMT
, base_fmt
, NULL
)) {
6222 error_setg(errp
, "Backing file format not supported for file "
6223 "format '%s'", fmt
);
6228 backing_file
= qemu_opt_get(opts
, BLOCK_OPT_BACKING_FILE
);
6230 if (!strcmp(filename
, backing_file
)) {
6231 error_setg(errp
, "Error: Trying to create an image with the "
6232 "same filename as the backing file");
6235 if (backing_file
[0] == '\0') {
6236 error_setg(errp
, "Expected backing file name, got empty string");
6241 backing_fmt
= qemu_opt_get(opts
, BLOCK_OPT_BACKING_FMT
);
6243 /* The size for the image must always be specified, unless we have a backing
6244 * file and we have not been forbidden from opening it. */
6245 size
= qemu_opt_get_size(opts
, BLOCK_OPT_SIZE
, img_size
);
6246 if (backing_file
&& !(flags
& BDRV_O_NO_BACKING
)) {
6247 BlockDriverState
*bs
;
6250 QDict
*backing_options
= NULL
;
6253 bdrv_get_full_backing_filename_from_filename(filename
, backing_file
,
6258 assert(full_backing
);
6260 /* backing files always opened read-only */
6262 back_flags
&= ~(BDRV_O_RDWR
| BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
);
6264 backing_options
= qdict_new();
6266 qdict_put_str(backing_options
, "driver", backing_fmt
);
6268 qdict_put_bool(backing_options
, BDRV_OPT_FORCE_SHARE
, true);
6270 bs
= bdrv_open(full_backing
, NULL
, backing_options
, back_flags
,
6272 g_free(full_backing
);
6274 error_append_hint(&local_err
, "Could not open backing image.\n");
6278 warn_report("Deprecated use of backing file without explicit "
6279 "backing format (detected format of %s)",
6280 bs
->drv
->format_name
);
6281 if (bs
->drv
!= &bdrv_raw
) {
6283 * A probe of raw deserves the most attention:
6284 * leaving the backing format out of the image
6285 * will ensure bs->probed is set (ensuring we
6286 * don't accidentally commit into the backing
6287 * file), and allow more spots to warn the users
6288 * to fix their toolchain when opening this image
6289 * later. For other images, we can safely record
6290 * the format that we probed.
6292 backing_fmt
= bs
->drv
->format_name
;
6293 qemu_opt_set(opts
, BLOCK_OPT_BACKING_FMT
, backing_fmt
,
6298 /* Opened BS, have no size */
6299 size
= bdrv_getlength(bs
);
6301 error_setg_errno(errp
, -size
, "Could not get size of '%s'",
6306 qemu_opt_set_number(opts
, BLOCK_OPT_SIZE
, size
, &error_abort
);
6310 /* (backing_file && !(flags & BDRV_O_NO_BACKING)) */
6311 } else if (backing_file
&& !backing_fmt
) {
6312 warn_report("Deprecated use of unopened backing file without "
6313 "explicit backing format, use of this image requires "
6314 "potentially unsafe format probing");
6318 error_setg(errp
, "Image creation needs a size parameter");
6323 printf("Formatting '%s', fmt=%s ", filename
, fmt
);
6324 qemu_opts_print(opts
, " ");
6329 ret
= bdrv_create(drv
, filename
, opts
, &local_err
);
6331 if (ret
== -EFBIG
) {
6332 /* This is generally a better message than whatever the driver would
6333 * deliver (especially because of the cluster_size_hint), since that
6334 * is most probably not much different from "image too large". */
6335 const char *cluster_size_hint
= "";
6336 if (qemu_opt_get_size(opts
, BLOCK_OPT_CLUSTER_SIZE
, 0)) {
6337 cluster_size_hint
= " (try using a larger cluster size)";
6339 error_setg(errp
, "The image size is too large for file format '%s'"
6340 "%s", fmt
, cluster_size_hint
);
6341 error_free(local_err
);
6346 qemu_opts_del(opts
);
6347 qemu_opts_free(create_opts
);
6348 error_propagate(errp
, local_err
);
6351 AioContext
*bdrv_get_aio_context(BlockDriverState
*bs
)
6353 return bs
? bs
->aio_context
: qemu_get_aio_context();
6356 void bdrv_coroutine_enter(BlockDriverState
*bs
, Coroutine
*co
)
6358 aio_co_enter(bdrv_get_aio_context(bs
), co
);
6361 static void bdrv_do_remove_aio_context_notifier(BdrvAioNotifier
*ban
)
6363 QLIST_REMOVE(ban
, list
);
6367 static void bdrv_detach_aio_context(BlockDriverState
*bs
)
6369 BdrvAioNotifier
*baf
, *baf_tmp
;
6371 assert(!bs
->walking_aio_notifiers
);
6372 bs
->walking_aio_notifiers
= true;
6373 QLIST_FOREACH_SAFE(baf
, &bs
->aio_notifiers
, list
, baf_tmp
) {
6375 bdrv_do_remove_aio_context_notifier(baf
);
6377 baf
->detach_aio_context(baf
->opaque
);
6380 /* Never mind iterating again to check for ->deleted. bdrv_close() will
6381 * remove remaining aio notifiers if we aren't called again.
6383 bs
->walking_aio_notifiers
= false;
6385 if (bs
->drv
&& bs
->drv
->bdrv_detach_aio_context
) {
6386 bs
->drv
->bdrv_detach_aio_context(bs
);
6389 if (bs
->quiesce_counter
) {
6390 aio_enable_external(bs
->aio_context
);
6392 bs
->aio_context
= NULL
;
6395 static void bdrv_attach_aio_context(BlockDriverState
*bs
,
6396 AioContext
*new_context
)
6398 BdrvAioNotifier
*ban
, *ban_tmp
;
6400 if (bs
->quiesce_counter
) {
6401 aio_disable_external(new_context
);
6404 bs
->aio_context
= new_context
;
6406 if (bs
->drv
&& bs
->drv
->bdrv_attach_aio_context
) {
6407 bs
->drv
->bdrv_attach_aio_context(bs
, new_context
);
6410 assert(!bs
->walking_aio_notifiers
);
6411 bs
->walking_aio_notifiers
= true;
6412 QLIST_FOREACH_SAFE(ban
, &bs
->aio_notifiers
, list
, ban_tmp
) {
6414 bdrv_do_remove_aio_context_notifier(ban
);
6416 ban
->attached_aio_context(new_context
, ban
->opaque
);
6419 bs
->walking_aio_notifiers
= false;
6423 * Changes the AioContext used for fd handlers, timers, and BHs by this
6424 * BlockDriverState and all its children and parents.
6426 * Must be called from the main AioContext.
6428 * The caller must own the AioContext lock for the old AioContext of bs, but it
6429 * must not own the AioContext lock for new_context (unless new_context is the
6430 * same as the current context of bs).
6432 * @ignore will accumulate all visited BdrvChild object. The caller is
6433 * responsible for freeing the list afterwards.
6435 void bdrv_set_aio_context_ignore(BlockDriverState
*bs
,
6436 AioContext
*new_context
, GSList
**ignore
)
6438 AioContext
*old_context
= bdrv_get_aio_context(bs
);
6441 g_assert(qemu_get_current_aio_context() == qemu_get_aio_context());
6443 if (old_context
== new_context
) {
6447 bdrv_drained_begin(bs
);
6449 QLIST_FOREACH(child
, &bs
->children
, next
) {
6450 if (g_slist_find(*ignore
, child
)) {
6453 *ignore
= g_slist_prepend(*ignore
, child
);
6454 bdrv_set_aio_context_ignore(child
->bs
, new_context
, ignore
);
6456 QLIST_FOREACH(child
, &bs
->parents
, next_parent
) {
6457 if (g_slist_find(*ignore
, child
)) {
6460 assert(child
->klass
->set_aio_ctx
);
6461 *ignore
= g_slist_prepend(*ignore
, child
);
6462 child
->klass
->set_aio_ctx(child
, new_context
, ignore
);
6465 bdrv_detach_aio_context(bs
);
6467 /* Acquire the new context, if necessary */
6468 if (qemu_get_aio_context() != new_context
) {
6469 aio_context_acquire(new_context
);
6472 bdrv_attach_aio_context(bs
, new_context
);
6475 * If this function was recursively called from
6476 * bdrv_set_aio_context_ignore(), there may be nodes in the
6477 * subtree that have not yet been moved to the new AioContext.
6478 * Release the old one so bdrv_drained_end() can poll them.
6480 if (qemu_get_aio_context() != old_context
) {
6481 aio_context_release(old_context
);
6484 bdrv_drained_end(bs
);
6486 if (qemu_get_aio_context() != old_context
) {
6487 aio_context_acquire(old_context
);
6489 if (qemu_get_aio_context() != new_context
) {
6490 aio_context_release(new_context
);
6494 static bool bdrv_parent_can_set_aio_context(BdrvChild
*c
, AioContext
*ctx
,
6495 GSList
**ignore
, Error
**errp
)
6497 if (g_slist_find(*ignore
, c
)) {
6500 *ignore
= g_slist_prepend(*ignore
, c
);
6503 * A BdrvChildClass that doesn't handle AioContext changes cannot
6504 * tolerate any AioContext changes
6506 if (!c
->klass
->can_set_aio_ctx
) {
6507 char *user
= bdrv_child_user_desc(c
);
6508 error_setg(errp
, "Changing iothreads is not supported by %s", user
);
6512 if (!c
->klass
->can_set_aio_ctx(c
, ctx
, ignore
, errp
)) {
6513 assert(!errp
|| *errp
);
6519 bool bdrv_child_can_set_aio_context(BdrvChild
*c
, AioContext
*ctx
,
6520 GSList
**ignore
, Error
**errp
)
6522 if (g_slist_find(*ignore
, c
)) {
6525 *ignore
= g_slist_prepend(*ignore
, c
);
6526 return bdrv_can_set_aio_context(c
->bs
, ctx
, ignore
, errp
);
6529 /* @ignore will accumulate all visited BdrvChild object. The caller is
6530 * responsible for freeing the list afterwards. */
6531 bool bdrv_can_set_aio_context(BlockDriverState
*bs
, AioContext
*ctx
,
6532 GSList
**ignore
, Error
**errp
)
6536 if (bdrv_get_aio_context(bs
) == ctx
) {
6540 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
6541 if (!bdrv_parent_can_set_aio_context(c
, ctx
, ignore
, errp
)) {
6545 QLIST_FOREACH(c
, &bs
->children
, next
) {
6546 if (!bdrv_child_can_set_aio_context(c
, ctx
, ignore
, errp
)) {
6554 int bdrv_child_try_set_aio_context(BlockDriverState
*bs
, AioContext
*ctx
,
6555 BdrvChild
*ignore_child
, Error
**errp
)
6560 ignore
= ignore_child
? g_slist_prepend(NULL
, ignore_child
) : NULL
;
6561 ret
= bdrv_can_set_aio_context(bs
, ctx
, &ignore
, errp
);
6562 g_slist_free(ignore
);
6568 ignore
= ignore_child
? g_slist_prepend(NULL
, ignore_child
) : NULL
;
6569 bdrv_set_aio_context_ignore(bs
, ctx
, &ignore
);
6570 g_slist_free(ignore
);
6575 int bdrv_try_set_aio_context(BlockDriverState
*bs
, AioContext
*ctx
,
6578 return bdrv_child_try_set_aio_context(bs
, ctx
, NULL
, errp
);
6581 void bdrv_add_aio_context_notifier(BlockDriverState
*bs
,
6582 void (*attached_aio_context
)(AioContext
*new_context
, void *opaque
),
6583 void (*detach_aio_context
)(void *opaque
), void *opaque
)
6585 BdrvAioNotifier
*ban
= g_new(BdrvAioNotifier
, 1);
6586 *ban
= (BdrvAioNotifier
){
6587 .attached_aio_context
= attached_aio_context
,
6588 .detach_aio_context
= detach_aio_context
,
6592 QLIST_INSERT_HEAD(&bs
->aio_notifiers
, ban
, list
);
6595 void bdrv_remove_aio_context_notifier(BlockDriverState
*bs
,
6596 void (*attached_aio_context
)(AioContext
*,
6598 void (*detach_aio_context
)(void *),
6601 BdrvAioNotifier
*ban
, *ban_next
;
6603 QLIST_FOREACH_SAFE(ban
, &bs
->aio_notifiers
, list
, ban_next
) {
6604 if (ban
->attached_aio_context
== attached_aio_context
&&
6605 ban
->detach_aio_context
== detach_aio_context
&&
6606 ban
->opaque
== opaque
&&
6607 ban
->deleted
== false)
6609 if (bs
->walking_aio_notifiers
) {
6610 ban
->deleted
= true;
6612 bdrv_do_remove_aio_context_notifier(ban
);
6621 int bdrv_amend_options(BlockDriverState
*bs
, QemuOpts
*opts
,
6622 BlockDriverAmendStatusCB
*status_cb
, void *cb_opaque
,
6627 error_setg(errp
, "Node is ejected");
6630 if (!bs
->drv
->bdrv_amend_options
) {
6631 error_setg(errp
, "Block driver '%s' does not support option amendment",
6632 bs
->drv
->format_name
);
6635 return bs
->drv
->bdrv_amend_options(bs
, opts
, status_cb
,
6636 cb_opaque
, force
, errp
);
6640 * This function checks whether the given @to_replace is allowed to be
6641 * replaced by a node that always shows the same data as @bs. This is
6642 * used for example to verify whether the mirror job can replace
6643 * @to_replace by the target mirrored from @bs.
6644 * To be replaceable, @bs and @to_replace may either be guaranteed to
6645 * always show the same data (because they are only connected through
6646 * filters), or some driver may allow replacing one of its children
6647 * because it can guarantee that this child's data is not visible at
6648 * all (for example, for dissenting quorum children that have no other
6651 bool bdrv_recurse_can_replace(BlockDriverState
*bs
,
6652 BlockDriverState
*to_replace
)
6654 BlockDriverState
*filtered
;
6656 if (!bs
|| !bs
->drv
) {
6660 if (bs
== to_replace
) {
6664 /* See what the driver can do */
6665 if (bs
->drv
->bdrv_recurse_can_replace
) {
6666 return bs
->drv
->bdrv_recurse_can_replace(bs
, to_replace
);
6669 /* For filters without an own implementation, we can recurse on our own */
6670 filtered
= bdrv_filter_bs(bs
);
6672 return bdrv_recurse_can_replace(filtered
, to_replace
);
6680 * Check whether the given @node_name can be replaced by a node that
6681 * has the same data as @parent_bs. If so, return @node_name's BDS;
6684 * @node_name must be a (recursive) *child of @parent_bs (or this
6685 * function will return NULL).
6687 * The result (whether the node can be replaced or not) is only valid
6688 * for as long as no graph or permission changes occur.
6690 BlockDriverState
*check_to_replace_node(BlockDriverState
*parent_bs
,
6691 const char *node_name
, Error
**errp
)
6693 BlockDriverState
*to_replace_bs
= bdrv_find_node(node_name
);
6694 AioContext
*aio_context
;
6696 if (!to_replace_bs
) {
6697 error_setg(errp
, "Node name '%s' not found", node_name
);
6701 aio_context
= bdrv_get_aio_context(to_replace_bs
);
6702 aio_context_acquire(aio_context
);
6704 if (bdrv_op_is_blocked(to_replace_bs
, BLOCK_OP_TYPE_REPLACE
, errp
)) {
6705 to_replace_bs
= NULL
;
6709 /* We don't want arbitrary node of the BDS chain to be replaced only the top
6710 * most non filter in order to prevent data corruption.
6711 * Another benefit is that this tests exclude backing files which are
6712 * blocked by the backing blockers.
6714 if (!bdrv_recurse_can_replace(parent_bs
, to_replace_bs
)) {
6715 error_setg(errp
, "Cannot replace '%s' by a node mirrored from '%s', "
6716 "because it cannot be guaranteed that doing so would not "
6717 "lead to an abrupt change of visible data",
6718 node_name
, parent_bs
->node_name
);
6719 to_replace_bs
= NULL
;
6724 aio_context_release(aio_context
);
6725 return to_replace_bs
;
6729 * Iterates through the list of runtime option keys that are said to
6730 * be "strong" for a BDS. An option is called "strong" if it changes
6731 * a BDS's data. For example, the null block driver's "size" and
6732 * "read-zeroes" options are strong, but its "latency-ns" option is
6735 * If a key returned by this function ends with a dot, all options
6736 * starting with that prefix are strong.
6738 static const char *const *strong_options(BlockDriverState
*bs
,
6739 const char *const *curopt
)
6741 static const char *const global_options
[] = {
6742 "driver", "filename", NULL
6746 return &global_options
[0];
6750 if (curopt
== &global_options
[ARRAY_SIZE(global_options
) - 1] && bs
->drv
) {
6751 curopt
= bs
->drv
->strong_runtime_opts
;
6754 return (curopt
&& *curopt
) ? curopt
: NULL
;
6758 * Copies all strong runtime options from bs->options to the given
6759 * QDict. The set of strong option keys is determined by invoking
6762 * Returns true iff any strong option was present in bs->options (and
6763 * thus copied to the target QDict) with the exception of "filename"
6764 * and "driver". The caller is expected to use this value to decide
6765 * whether the existence of strong options prevents the generation of
6768 static bool append_strong_runtime_options(QDict
*d
, BlockDriverState
*bs
)
6770 bool found_any
= false;
6771 const char *const *option_name
= NULL
;
6777 while ((option_name
= strong_options(bs
, option_name
))) {
6778 bool option_given
= false;
6780 assert(strlen(*option_name
) > 0);
6781 if ((*option_name
)[strlen(*option_name
) - 1] != '.') {
6782 QObject
*entry
= qdict_get(bs
->options
, *option_name
);
6787 qdict_put_obj(d
, *option_name
, qobject_ref(entry
));
6788 option_given
= true;
6790 const QDictEntry
*entry
;
6791 for (entry
= qdict_first(bs
->options
); entry
;
6792 entry
= qdict_next(bs
->options
, entry
))
6794 if (strstart(qdict_entry_key(entry
), *option_name
, NULL
)) {
6795 qdict_put_obj(d
, qdict_entry_key(entry
),
6796 qobject_ref(qdict_entry_value(entry
)));
6797 option_given
= true;
6802 /* While "driver" and "filename" need to be included in a JSON filename,
6803 * their existence does not prohibit generation of a plain filename. */
6804 if (!found_any
&& option_given
&&
6805 strcmp(*option_name
, "driver") && strcmp(*option_name
, "filename"))
6811 if (!qdict_haskey(d
, "driver")) {
6812 /* Drivers created with bdrv_new_open_driver() may not have a
6813 * @driver option. Add it here. */
6814 qdict_put_str(d
, "driver", bs
->drv
->format_name
);
6820 /* Note: This function may return false positives; it may return true
6821 * even if opening the backing file specified by bs's image header
6822 * would result in exactly bs->backing. */
6823 static bool bdrv_backing_overridden(BlockDriverState
*bs
)
6826 return strcmp(bs
->auto_backing_file
,
6827 bs
->backing
->bs
->filename
);
6829 /* No backing BDS, so if the image header reports any backing
6830 * file, it must have been suppressed */
6831 return bs
->auto_backing_file
[0] != '\0';
6835 /* Updates the following BDS fields:
6836 * - exact_filename: A filename which may be used for opening a block device
6837 * which (mostly) equals the given BDS (even without any
6838 * other options; so reading and writing must return the same
6839 * results, but caching etc. may be different)
6840 * - full_open_options: Options which, when given when opening a block device
6841 * (without a filename), result in a BDS (mostly)
6842 * equalling the given one
6843 * - filename: If exact_filename is set, it is copied here. Otherwise,
6844 * full_open_options is converted to a JSON object, prefixed with
6845 * "json:" (for use through the JSON pseudo protocol) and put here.
6847 void bdrv_refresh_filename(BlockDriverState
*bs
)
6849 BlockDriver
*drv
= bs
->drv
;
6851 BlockDriverState
*primary_child_bs
;
6853 bool backing_overridden
;
6854 bool generate_json_filename
; /* Whether our default implementation should
6855 fill exact_filename (false) or not (true) */
6861 /* This BDS's file name may depend on any of its children's file names, so
6862 * refresh those first */
6863 QLIST_FOREACH(child
, &bs
->children
, next
) {
6864 bdrv_refresh_filename(child
->bs
);
6868 /* For implicit nodes, just copy everything from the single child */
6869 child
= QLIST_FIRST(&bs
->children
);
6870 assert(QLIST_NEXT(child
, next
) == NULL
);
6872 pstrcpy(bs
->exact_filename
, sizeof(bs
->exact_filename
),
6873 child
->bs
->exact_filename
);
6874 pstrcpy(bs
->filename
, sizeof(bs
->filename
), child
->bs
->filename
);
6876 qobject_unref(bs
->full_open_options
);
6877 bs
->full_open_options
= qobject_ref(child
->bs
->full_open_options
);
6882 backing_overridden
= bdrv_backing_overridden(bs
);
6884 if (bs
->open_flags
& BDRV_O_NO_IO
) {
6885 /* Without I/O, the backing file does not change anything.
6886 * Therefore, in such a case (primarily qemu-img), we can
6887 * pretend the backing file has not been overridden even if
6888 * it technically has been. */
6889 backing_overridden
= false;
6892 /* Gather the options QDict */
6894 generate_json_filename
= append_strong_runtime_options(opts
, bs
);
6895 generate_json_filename
|= backing_overridden
;
6897 if (drv
->bdrv_gather_child_options
) {
6898 /* Some block drivers may not want to present all of their children's
6899 * options, or name them differently from BdrvChild.name */
6900 drv
->bdrv_gather_child_options(bs
, opts
, backing_overridden
);
6902 QLIST_FOREACH(child
, &bs
->children
, next
) {
6903 if (child
== bs
->backing
&& !backing_overridden
) {
6904 /* We can skip the backing BDS if it has not been overridden */
6908 qdict_put(opts
, child
->name
,
6909 qobject_ref(child
->bs
->full_open_options
));
6912 if (backing_overridden
&& !bs
->backing
) {
6913 /* Force no backing file */
6914 qdict_put_null(opts
, "backing");
6918 qobject_unref(bs
->full_open_options
);
6919 bs
->full_open_options
= opts
;
6921 primary_child_bs
= bdrv_primary_bs(bs
);
6923 if (drv
->bdrv_refresh_filename
) {
6924 /* Obsolete information is of no use here, so drop the old file name
6925 * information before refreshing it */
6926 bs
->exact_filename
[0] = '\0';
6928 drv
->bdrv_refresh_filename(bs
);
6929 } else if (primary_child_bs
) {
6931 * Try to reconstruct valid information from the underlying
6932 * file -- this only works for format nodes (filter nodes
6933 * cannot be probed and as such must be selected by the user
6934 * either through an options dict, or through a special
6935 * filename which the filter driver must construct in its
6936 * .bdrv_refresh_filename() implementation).
6939 bs
->exact_filename
[0] = '\0';
6942 * We can use the underlying file's filename if:
6943 * - it has a filename,
6944 * - the current BDS is not a filter,
6945 * - the file is a protocol BDS, and
6946 * - opening that file (as this BDS's format) will automatically create
6947 * the BDS tree we have right now, that is:
6948 * - the user did not significantly change this BDS's behavior with
6949 * some explicit (strong) options
6950 * - no non-file child of this BDS has been overridden by the user
6951 * Both of these conditions are represented by generate_json_filename.
6953 if (primary_child_bs
->exact_filename
[0] &&
6954 primary_child_bs
->drv
->bdrv_file_open
&&
6955 !drv
->is_filter
&& !generate_json_filename
)
6957 strcpy(bs
->exact_filename
, primary_child_bs
->exact_filename
);
6961 if (bs
->exact_filename
[0]) {
6962 pstrcpy(bs
->filename
, sizeof(bs
->filename
), bs
->exact_filename
);
6964 QString
*json
= qobject_to_json(QOBJECT(bs
->full_open_options
));
6965 if (snprintf(bs
->filename
, sizeof(bs
->filename
), "json:%s",
6966 qstring_get_str(json
)) >= sizeof(bs
->filename
)) {
6967 /* Give user a hint if we truncated things. */
6968 strcpy(bs
->filename
+ sizeof(bs
->filename
) - 4, "...");
6970 qobject_unref(json
);
6974 char *bdrv_dirname(BlockDriverState
*bs
, Error
**errp
)
6976 BlockDriver
*drv
= bs
->drv
;
6977 BlockDriverState
*child_bs
;
6980 error_setg(errp
, "Node '%s' is ejected", bs
->node_name
);
6984 if (drv
->bdrv_dirname
) {
6985 return drv
->bdrv_dirname(bs
, errp
);
6988 child_bs
= bdrv_primary_bs(bs
);
6990 return bdrv_dirname(child_bs
, errp
);
6993 bdrv_refresh_filename(bs
);
6994 if (bs
->exact_filename
[0] != '\0') {
6995 return path_combine(bs
->exact_filename
, "");
6998 error_setg(errp
, "Cannot generate a base directory for %s nodes",
7004 * Hot add/remove a BDS's child. So the user can take a child offline when
7005 * it is broken and take a new child online
7007 void bdrv_add_child(BlockDriverState
*parent_bs
, BlockDriverState
*child_bs
,
7011 if (!parent_bs
->drv
|| !parent_bs
->drv
->bdrv_add_child
) {
7012 error_setg(errp
, "The node %s does not support adding a child",
7013 bdrv_get_device_or_node_name(parent_bs
));
7017 if (!QLIST_EMPTY(&child_bs
->parents
)) {
7018 error_setg(errp
, "The node %s already has a parent",
7019 child_bs
->node_name
);
7023 parent_bs
->drv
->bdrv_add_child(parent_bs
, child_bs
, errp
);
7026 void bdrv_del_child(BlockDriverState
*parent_bs
, BdrvChild
*child
, Error
**errp
)
7030 if (!parent_bs
->drv
|| !parent_bs
->drv
->bdrv_del_child
) {
7031 error_setg(errp
, "The node %s does not support removing a child",
7032 bdrv_get_device_or_node_name(parent_bs
));
7036 QLIST_FOREACH(tmp
, &parent_bs
->children
, next
) {
7043 error_setg(errp
, "The node %s does not have a child named %s",
7044 bdrv_get_device_or_node_name(parent_bs
),
7045 bdrv_get_device_or_node_name(child
->bs
));
7049 parent_bs
->drv
->bdrv_del_child(parent_bs
, child
, errp
);
7052 int bdrv_make_empty(BdrvChild
*c
, Error
**errp
)
7054 BlockDriver
*drv
= c
->bs
->drv
;
7057 assert(c
->perm
& (BLK_PERM_WRITE
| BLK_PERM_WRITE_UNCHANGED
));
7059 if (!drv
->bdrv_make_empty
) {
7060 error_setg(errp
, "%s does not support emptying nodes",
7065 ret
= drv
->bdrv_make_empty(c
->bs
);
7067 error_setg_errno(errp
, -ret
, "Failed to empty %s",
7076 * Return the child that @bs acts as an overlay for, and from which data may be
7077 * copied in COW or COR operations. Usually this is the backing file.
7079 BdrvChild
*bdrv_cow_child(BlockDriverState
*bs
)
7081 if (!bs
|| !bs
->drv
) {
7085 if (bs
->drv
->is_filter
) {
7093 assert(bs
->backing
->role
& BDRV_CHILD_COW
);
7098 * If @bs acts as a filter for exactly one of its children, return
7101 BdrvChild
*bdrv_filter_child(BlockDriverState
*bs
)
7105 if (!bs
|| !bs
->drv
) {
7109 if (!bs
->drv
->is_filter
) {
7113 /* Only one of @backing or @file may be used */
7114 assert(!(bs
->backing
&& bs
->file
));
7116 c
= bs
->backing
?: bs
->file
;
7121 assert(c
->role
& BDRV_CHILD_FILTERED
);
7126 * Return either the result of bdrv_cow_child() or bdrv_filter_child(),
7127 * whichever is non-NULL.
7129 * Return NULL if both are NULL.
7131 BdrvChild
*bdrv_filter_or_cow_child(BlockDriverState
*bs
)
7133 BdrvChild
*cow_child
= bdrv_cow_child(bs
);
7134 BdrvChild
*filter_child
= bdrv_filter_child(bs
);
7136 /* Filter nodes cannot have COW backing files */
7137 assert(!(cow_child
&& filter_child
));
7139 return cow_child
?: filter_child
;
7143 * Return the primary child of this node: For filters, that is the
7144 * filtered child. For other nodes, that is usually the child storing
7146 * (A generally more helpful description is that this is (usually) the
7147 * child that has the same filename as @bs.)
7149 * Drivers do not necessarily have a primary child; for example quorum
7152 BdrvChild
*bdrv_primary_child(BlockDriverState
*bs
)
7154 BdrvChild
*c
, *found
= NULL
;
7156 QLIST_FOREACH(c
, &bs
->children
, next
) {
7157 if (c
->role
& BDRV_CHILD_PRIMARY
) {
7166 static BlockDriverState
*bdrv_do_skip_filters(BlockDriverState
*bs
,
7167 bool stop_on_explicit_filter
)
7175 while (!(stop_on_explicit_filter
&& !bs
->implicit
)) {
7176 c
= bdrv_filter_child(bs
);
7179 * A filter that is embedded in a working block graph must
7180 * have a child. Assert this here so this function does
7181 * not return a filter node that is not expected by the
7184 assert(!bs
->drv
|| !bs
->drv
->is_filter
);
7190 * Note that this treats nodes with bs->drv == NULL as not being
7191 * filters (bs->drv == NULL should be replaced by something else
7193 * The advantage of this behavior is that this function will thus
7194 * always return a non-NULL value (given a non-NULL @bs).
7201 * Return the first BDS that has not been added implicitly or that
7202 * does not have a filtered child down the chain starting from @bs
7203 * (including @bs itself).
7205 BlockDriverState
*bdrv_skip_implicit_filters(BlockDriverState
*bs
)
7207 return bdrv_do_skip_filters(bs
, true);
7211 * Return the first BDS that does not have a filtered child down the
7212 * chain starting from @bs (including @bs itself).
7214 BlockDriverState
*bdrv_skip_filters(BlockDriverState
*bs
)
7216 return bdrv_do_skip_filters(bs
, false);
7220 * For a backing chain, return the first non-filter backing image of
7221 * the first non-filter image.
7223 BlockDriverState
*bdrv_backing_chain_next(BlockDriverState
*bs
)
7225 return bdrv_skip_filters(bdrv_cow_bs(bdrv_skip_filters(bs
)));