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 "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 BdrvChildRole
*child_role
,
82 /* If non-zero, use only whitelisted block drivers */
83 static int use_bdrv_whitelist
;
86 static int is_windows_drive_prefix(const char *filename
)
88 return (((filename
[0] >= 'a' && filename
[0] <= 'z') ||
89 (filename
[0] >= 'A' && filename
[0] <= 'Z')) &&
93 int is_windows_drive(const char *filename
)
95 if (is_windows_drive_prefix(filename
) &&
98 if (strstart(filename
, "\\\\.\\", NULL
) ||
99 strstart(filename
, "//./", NULL
))
105 size_t bdrv_opt_mem_align(BlockDriverState
*bs
)
107 if (!bs
|| !bs
->drv
) {
108 /* page size or 4k (hdd sector size) should be on the safe side */
109 return MAX(4096, qemu_real_host_page_size
);
112 return bs
->bl
.opt_mem_alignment
;
115 size_t bdrv_min_mem_align(BlockDriverState
*bs
)
117 if (!bs
|| !bs
->drv
) {
118 /* page size or 4k (hdd sector size) should be on the safe side */
119 return MAX(4096, qemu_real_host_page_size
);
122 return bs
->bl
.min_mem_alignment
;
125 /* check if the path starts with "<protocol>:" */
126 int path_has_protocol(const char *path
)
131 if (is_windows_drive(path
) ||
132 is_windows_drive_prefix(path
)) {
135 p
= path
+ strcspn(path
, ":/\\");
137 p
= path
+ strcspn(path
, ":/");
143 int path_is_absolute(const char *path
)
146 /* specific case for names like: "\\.\d:" */
147 if (is_windows_drive(path
) || is_windows_drive_prefix(path
)) {
150 return (*path
== '/' || *path
== '\\');
152 return (*path
== '/');
156 /* if filename is absolute, just return its duplicate. Otherwise, build a
157 path to it by considering it is relative to base_path. URL are
159 char *path_combine(const char *base_path
, const char *filename
)
161 const char *protocol_stripped
= NULL
;
166 if (path_is_absolute(filename
)) {
167 return g_strdup(filename
);
170 if (path_has_protocol(base_path
)) {
171 protocol_stripped
= strchr(base_path
, ':');
172 if (protocol_stripped
) {
176 p
= protocol_stripped
?: base_path
;
178 p1
= strrchr(base_path
, '/');
182 p2
= strrchr(base_path
, '\\');
183 if (!p1
|| p2
> p1
) {
198 result
= g_malloc(len
+ strlen(filename
) + 1);
199 memcpy(result
, base_path
, len
);
200 strcpy(result
+ len
, filename
);
206 * Helper function for bdrv_parse_filename() implementations to remove optional
207 * protocol prefixes (especially "file:") from a filename and for putting the
208 * stripped filename into the options QDict if there is such a prefix.
210 void bdrv_parse_filename_strip_prefix(const char *filename
, const char *prefix
,
213 if (strstart(filename
, prefix
, &filename
)) {
214 /* Stripping the explicit protocol prefix may result in a protocol
215 * prefix being (wrongly) detected (if the filename contains a colon) */
216 if (path_has_protocol(filename
)) {
217 QString
*fat_filename
;
219 /* This means there is some colon before the first slash; therefore,
220 * this cannot be an absolute path */
221 assert(!path_is_absolute(filename
));
223 /* And we can thus fix the protocol detection issue by prefixing it
225 fat_filename
= qstring_from_str("./");
226 qstring_append(fat_filename
, filename
);
228 assert(!path_has_protocol(qstring_get_str(fat_filename
)));
230 qdict_put(options
, "filename", fat_filename
);
232 /* If no protocol prefix was detected, we can use the shortened
234 qdict_put_str(options
, "filename", filename
);
240 /* Returns whether the image file is opened as read-only. Note that this can
241 * return false and writing to the image file is still not possible because the
242 * image is inactivated. */
243 bool bdrv_is_read_only(BlockDriverState
*bs
)
245 return bs
->read_only
;
248 int bdrv_can_set_read_only(BlockDriverState
*bs
, bool read_only
,
249 bool ignore_allow_rdw
, Error
**errp
)
251 /* Do not set read_only if copy_on_read is enabled */
252 if (bs
->copy_on_read
&& read_only
) {
253 error_setg(errp
, "Can't set node '%s' to r/o with copy-on-read enabled",
254 bdrv_get_device_or_node_name(bs
));
258 /* Do not clear read_only if it is prohibited */
259 if (!read_only
&& !(bs
->open_flags
& BDRV_O_ALLOW_RDWR
) &&
262 error_setg(errp
, "Node '%s' is read only",
263 bdrv_get_device_or_node_name(bs
));
271 * Called by a driver that can only provide a read-only image.
273 * Returns 0 if the node is already read-only or it could switch the node to
274 * read-only because BDRV_O_AUTO_RDONLY is set.
276 * Returns -EACCES if the node is read-write and BDRV_O_AUTO_RDONLY is not set
277 * or bdrv_can_set_read_only() forbids making the node read-only. If @errmsg
278 * is not NULL, it is used as the error message for the Error object.
280 int bdrv_apply_auto_read_only(BlockDriverState
*bs
, const char *errmsg
,
285 if (!(bs
->open_flags
& BDRV_O_RDWR
)) {
288 if (!(bs
->open_flags
& BDRV_O_AUTO_RDONLY
)) {
292 ret
= bdrv_can_set_read_only(bs
, true, false, NULL
);
297 bs
->read_only
= true;
298 bs
->open_flags
&= ~BDRV_O_RDWR
;
303 error_setg(errp
, "%s", errmsg
?: "Image is read-only");
308 * If @backing is empty, this function returns NULL without setting
309 * @errp. In all other cases, NULL will only be returned with @errp
312 * Therefore, a return value of NULL without @errp set means that
313 * there is no backing file; if @errp is set, there is one but its
314 * absolute filename cannot be generated.
316 char *bdrv_get_full_backing_filename_from_filename(const char *backed
,
320 if (backing
[0] == '\0') {
322 } else if (path_has_protocol(backing
) || path_is_absolute(backing
)) {
323 return g_strdup(backing
);
324 } else if (backed
[0] == '\0' || strstart(backed
, "json:", NULL
)) {
325 error_setg(errp
, "Cannot use relative backing file names for '%s'",
329 return path_combine(backed
, backing
);
334 * If @filename is empty or NULL, this function returns NULL without
335 * setting @errp. In all other cases, NULL will only be returned with
338 static char *bdrv_make_absolute_filename(BlockDriverState
*relative_to
,
339 const char *filename
, Error
**errp
)
341 char *dir
, *full_name
;
343 if (!filename
|| filename
[0] == '\0') {
345 } else if (path_has_protocol(filename
) || path_is_absolute(filename
)) {
346 return g_strdup(filename
);
349 dir
= bdrv_dirname(relative_to
, errp
);
354 full_name
= g_strconcat(dir
, filename
, NULL
);
359 char *bdrv_get_full_backing_filename(BlockDriverState
*bs
, Error
**errp
)
361 return bdrv_make_absolute_filename(bs
, bs
->backing_file
, errp
);
364 void bdrv_register(BlockDriver
*bdrv
)
366 assert(bdrv
->format_name
);
367 QLIST_INSERT_HEAD(&bdrv_drivers
, bdrv
, list
);
370 BlockDriverState
*bdrv_new(void)
372 BlockDriverState
*bs
;
375 bs
= g_new0(BlockDriverState
, 1);
376 QLIST_INIT(&bs
->dirty_bitmaps
);
377 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
378 QLIST_INIT(&bs
->op_blockers
[i
]);
380 notifier_with_return_list_init(&bs
->before_write_notifiers
);
381 qemu_co_mutex_init(&bs
->reqs_lock
);
382 qemu_mutex_init(&bs
->dirty_bitmap_mutex
);
384 bs
->aio_context
= qemu_get_aio_context();
386 qemu_co_queue_init(&bs
->flush_queue
);
388 for (i
= 0; i
< bdrv_drain_all_count
; i
++) {
389 bdrv_drained_begin(bs
);
392 QTAILQ_INSERT_TAIL(&all_bdrv_states
, bs
, bs_list
);
397 static BlockDriver
*bdrv_do_find_format(const char *format_name
)
401 QLIST_FOREACH(drv1
, &bdrv_drivers
, list
) {
402 if (!strcmp(drv1
->format_name
, format_name
)) {
410 BlockDriver
*bdrv_find_format(const char *format_name
)
415 drv1
= bdrv_do_find_format(format_name
);
420 /* The driver isn't registered, maybe we need to load a module */
421 for (i
= 0; i
< (int)ARRAY_SIZE(block_driver_modules
); ++i
) {
422 if (!strcmp(block_driver_modules
[i
].format_name
, format_name
)) {
423 block_module_load_one(block_driver_modules
[i
].library_name
);
428 return bdrv_do_find_format(format_name
);
431 static int bdrv_format_is_whitelisted(const char *format_name
, bool read_only
)
433 static const char *whitelist_rw
[] = {
434 CONFIG_BDRV_RW_WHITELIST
436 static const char *whitelist_ro
[] = {
437 CONFIG_BDRV_RO_WHITELIST
441 if (!whitelist_rw
[0] && !whitelist_ro
[0]) {
442 return 1; /* no whitelist, anything goes */
445 for (p
= whitelist_rw
; *p
; p
++) {
446 if (!strcmp(format_name
, *p
)) {
451 for (p
= whitelist_ro
; *p
; p
++) {
452 if (!strcmp(format_name
, *p
)) {
460 int bdrv_is_whitelisted(BlockDriver
*drv
, bool read_only
)
462 return bdrv_format_is_whitelisted(drv
->format_name
, read_only
);
465 bool bdrv_uses_whitelist(void)
467 return use_bdrv_whitelist
;
470 typedef struct CreateCo
{
478 static void coroutine_fn
bdrv_create_co_entry(void *opaque
)
480 Error
*local_err
= NULL
;
483 CreateCo
*cco
= opaque
;
486 ret
= cco
->drv
->bdrv_co_create_opts(cco
->drv
,
487 cco
->filename
, cco
->opts
, &local_err
);
488 error_propagate(&cco
->err
, local_err
);
492 int bdrv_create(BlockDriver
*drv
, const char* filename
,
493 QemuOpts
*opts
, Error
**errp
)
500 .filename
= g_strdup(filename
),
506 if (!drv
->bdrv_co_create_opts
) {
507 error_setg(errp
, "Driver '%s' does not support image creation", drv
->format_name
);
512 if (qemu_in_coroutine()) {
513 /* Fast-path if already in coroutine context */
514 bdrv_create_co_entry(&cco
);
516 co
= qemu_coroutine_create(bdrv_create_co_entry
, &cco
);
517 qemu_coroutine_enter(co
);
518 while (cco
.ret
== NOT_DONE
) {
519 aio_poll(qemu_get_aio_context(), true);
526 error_propagate(errp
, cco
.err
);
528 error_setg_errno(errp
, -ret
, "Could not create image");
533 g_free(cco
.filename
);
538 * Helper function for bdrv_create_file_fallback(): Resize @blk to at
539 * least the given @minimum_size.
541 * On success, return @blk's actual length.
542 * Otherwise, return -errno.
544 static int64_t create_file_fallback_truncate(BlockBackend
*blk
,
545 int64_t minimum_size
, Error
**errp
)
547 Error
*local_err
= NULL
;
551 ret
= blk_truncate(blk
, minimum_size
, false, PREALLOC_MODE_OFF
, &local_err
);
552 if (ret
< 0 && ret
!= -ENOTSUP
) {
553 error_propagate(errp
, local_err
);
557 size
= blk_getlength(blk
);
559 error_free(local_err
);
560 error_setg_errno(errp
, -size
,
561 "Failed to inquire the new image file's length");
565 if (size
< minimum_size
) {
566 /* Need to grow the image, but we failed to do that */
567 error_propagate(errp
, local_err
);
571 error_free(local_err
);
578 * Helper function for bdrv_create_file_fallback(): Zero the first
579 * sector to remove any potentially pre-existing image header.
581 static int create_file_fallback_zero_first_sector(BlockBackend
*blk
,
582 int64_t current_size
,
585 int64_t bytes_to_clear
;
588 bytes_to_clear
= MIN(current_size
, BDRV_SECTOR_SIZE
);
589 if (bytes_to_clear
) {
590 ret
= blk_pwrite_zeroes(blk
, 0, bytes_to_clear
, BDRV_REQ_MAY_UNMAP
);
592 error_setg_errno(errp
, -ret
,
593 "Failed to clear the new image's first sector");
601 static int bdrv_create_file_fallback(const char *filename
, BlockDriver
*drv
,
602 QemuOpts
*opts
, Error
**errp
)
608 PreallocMode prealloc
;
609 Error
*local_err
= NULL
;
612 size
= qemu_opt_get_size_del(opts
, BLOCK_OPT_SIZE
, 0);
613 buf
= qemu_opt_get_del(opts
, BLOCK_OPT_PREALLOC
);
614 prealloc
= qapi_enum_parse(&PreallocMode_lookup
, buf
,
615 PREALLOC_MODE_OFF
, &local_err
);
618 error_propagate(errp
, local_err
);
622 if (prealloc
!= PREALLOC_MODE_OFF
) {
623 error_setg(errp
, "Unsupported preallocation mode '%s'",
624 PreallocMode_str(prealloc
));
628 options
= qdict_new();
629 qdict_put_str(options
, "driver", drv
->format_name
);
631 blk
= blk_new_open(filename
, NULL
, options
,
632 BDRV_O_RDWR
| BDRV_O_RESIZE
, errp
);
634 error_prepend(errp
, "Protocol driver '%s' does not support image "
635 "creation, and opening the image failed: ",
640 size
= create_file_fallback_truncate(blk
, size
, errp
);
646 ret
= create_file_fallback_zero_first_sector(blk
, size
, errp
);
657 int bdrv_create_file(const char *filename
, QemuOpts
*opts
, Error
**errp
)
661 drv
= bdrv_find_protocol(filename
, true, errp
);
666 if (drv
->bdrv_co_create_opts
) {
667 return bdrv_create(drv
, filename
, opts
, errp
);
669 return bdrv_create_file_fallback(filename
, drv
, opts
, errp
);
673 int coroutine_fn
bdrv_co_delete_file(BlockDriverState
*bs
, Error
**errp
)
675 Error
*local_err
= NULL
;
681 error_setg(errp
, "Block node '%s' is not opened", bs
->filename
);
685 if (!bs
->drv
->bdrv_co_delete_file
) {
686 error_setg(errp
, "Driver '%s' does not support image deletion",
687 bs
->drv
->format_name
);
691 ret
= bs
->drv
->bdrv_co_delete_file(bs
, &local_err
);
693 error_propagate(errp
, local_err
);
700 * Try to get @bs's logical and physical block size.
701 * On success, store them in @bsz struct and return 0.
702 * On failure return -errno.
703 * @bs must not be empty.
705 int bdrv_probe_blocksizes(BlockDriverState
*bs
, BlockSizes
*bsz
)
707 BlockDriver
*drv
= bs
->drv
;
709 if (drv
&& drv
->bdrv_probe_blocksizes
) {
710 return drv
->bdrv_probe_blocksizes(bs
, bsz
);
711 } else if (drv
&& drv
->is_filter
&& bs
->file
) {
712 return bdrv_probe_blocksizes(bs
->file
->bs
, bsz
);
719 * Try to get @bs's geometry (cyls, heads, sectors).
720 * On success, store them in @geo struct and return 0.
721 * On failure return -errno.
722 * @bs must not be empty.
724 int bdrv_probe_geometry(BlockDriverState
*bs
, HDGeometry
*geo
)
726 BlockDriver
*drv
= bs
->drv
;
728 if (drv
&& drv
->bdrv_probe_geometry
) {
729 return drv
->bdrv_probe_geometry(bs
, geo
);
730 } else if (drv
&& drv
->is_filter
&& bs
->file
) {
731 return bdrv_probe_geometry(bs
->file
->bs
, geo
);
738 * Create a uniquely-named empty temporary file.
739 * Return 0 upon success, otherwise a negative errno value.
741 int get_tmp_filename(char *filename
, int size
)
744 char temp_dir
[MAX_PATH
];
745 /* GetTempFileName requires that its output buffer (4th param)
746 have length MAX_PATH or greater. */
747 assert(size
>= MAX_PATH
);
748 return (GetTempPath(MAX_PATH
, temp_dir
)
749 && GetTempFileName(temp_dir
, "qem", 0, filename
)
750 ? 0 : -GetLastError());
754 tmpdir
= getenv("TMPDIR");
758 if (snprintf(filename
, size
, "%s/vl.XXXXXX", tmpdir
) >= size
) {
761 fd
= mkstemp(filename
);
765 if (close(fd
) != 0) {
774 * Detect host devices. By convention, /dev/cdrom[N] is always
775 * recognized as a host CDROM.
777 static BlockDriver
*find_hdev_driver(const char *filename
)
779 int score_max
= 0, score
;
780 BlockDriver
*drv
= NULL
, *d
;
782 QLIST_FOREACH(d
, &bdrv_drivers
, list
) {
783 if (d
->bdrv_probe_device
) {
784 score
= d
->bdrv_probe_device(filename
);
785 if (score
> score_max
) {
795 static BlockDriver
*bdrv_do_find_protocol(const char *protocol
)
799 QLIST_FOREACH(drv1
, &bdrv_drivers
, list
) {
800 if (drv1
->protocol_name
&& !strcmp(drv1
->protocol_name
, protocol
)) {
808 BlockDriver
*bdrv_find_protocol(const char *filename
,
809 bool allow_protocol_prefix
,
818 /* TODO Drivers without bdrv_file_open must be specified explicitly */
821 * XXX(hch): we really should not let host device detection
822 * override an explicit protocol specification, but moving this
823 * later breaks access to device names with colons in them.
824 * Thanks to the brain-dead persistent naming schemes on udev-
825 * based Linux systems those actually are quite common.
827 drv1
= find_hdev_driver(filename
);
832 if (!path_has_protocol(filename
) || !allow_protocol_prefix
) {
836 p
= strchr(filename
, ':');
839 if (len
> sizeof(protocol
) - 1)
840 len
= sizeof(protocol
) - 1;
841 memcpy(protocol
, filename
, len
);
842 protocol
[len
] = '\0';
844 drv1
= bdrv_do_find_protocol(protocol
);
849 for (i
= 0; i
< (int)ARRAY_SIZE(block_driver_modules
); ++i
) {
850 if (block_driver_modules
[i
].protocol_name
&&
851 !strcmp(block_driver_modules
[i
].protocol_name
, protocol
)) {
852 block_module_load_one(block_driver_modules
[i
].library_name
);
857 drv1
= bdrv_do_find_protocol(protocol
);
859 error_setg(errp
, "Unknown protocol '%s'", protocol
);
865 * Guess image format by probing its contents.
866 * This is not a good idea when your image is raw (CVE-2008-2004), but
867 * we do it anyway for backward compatibility.
869 * @buf contains the image's first @buf_size bytes.
870 * @buf_size is the buffer size in bytes (generally BLOCK_PROBE_BUF_SIZE,
871 * but can be smaller if the image file is smaller)
872 * @filename is its filename.
874 * For all block drivers, call the bdrv_probe() method to get its
876 * Return the first block driver with the highest probing score.
878 BlockDriver
*bdrv_probe_all(const uint8_t *buf
, int buf_size
,
879 const char *filename
)
881 int score_max
= 0, score
;
882 BlockDriver
*drv
= NULL
, *d
;
884 QLIST_FOREACH(d
, &bdrv_drivers
, list
) {
886 score
= d
->bdrv_probe(buf
, buf_size
, filename
);
887 if (score
> score_max
) {
897 static int find_image_format(BlockBackend
*file
, const char *filename
,
898 BlockDriver
**pdrv
, Error
**errp
)
901 uint8_t buf
[BLOCK_PROBE_BUF_SIZE
];
904 /* Return the raw BlockDriver * to scsi-generic devices or empty drives */
905 if (blk_is_sg(file
) || !blk_is_inserted(file
) || blk_getlength(file
) == 0) {
910 ret
= blk_pread(file
, 0, buf
, sizeof(buf
));
912 error_setg_errno(errp
, -ret
, "Could not read image for determining its "
918 drv
= bdrv_probe_all(buf
, ret
, filename
);
920 error_setg(errp
, "Could not determine image format: No compatible "
929 * Set the current 'total_sectors' value
930 * Return 0 on success, -errno on error.
932 int refresh_total_sectors(BlockDriverState
*bs
, int64_t hint
)
934 BlockDriver
*drv
= bs
->drv
;
940 /* Do not attempt drv->bdrv_getlength() on scsi-generic devices */
944 /* query actual device if possible, otherwise just trust the hint */
945 if (drv
->bdrv_getlength
) {
946 int64_t length
= drv
->bdrv_getlength(bs
);
950 hint
= DIV_ROUND_UP(length
, BDRV_SECTOR_SIZE
);
953 bs
->total_sectors
= hint
;
958 * Combines a QDict of new block driver @options with any missing options taken
959 * from @old_options, so that leaving out an option defaults to its old value.
961 static void bdrv_join_options(BlockDriverState
*bs
, QDict
*options
,
964 if (bs
->drv
&& bs
->drv
->bdrv_join_options
) {
965 bs
->drv
->bdrv_join_options(options
, old_options
);
967 qdict_join(options
, old_options
, false);
971 static BlockdevDetectZeroesOptions
bdrv_parse_detect_zeroes(QemuOpts
*opts
,
975 Error
*local_err
= NULL
;
976 char *value
= qemu_opt_get_del(opts
, "detect-zeroes");
977 BlockdevDetectZeroesOptions detect_zeroes
=
978 qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup
, value
,
979 BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF
, &local_err
);
982 error_propagate(errp
, local_err
);
983 return detect_zeroes
;
986 if (detect_zeroes
== BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP
&&
987 !(open_flags
& BDRV_O_UNMAP
))
989 error_setg(errp
, "setting detect-zeroes to unmap is not allowed "
990 "without setting discard operation to unmap");
993 return detect_zeroes
;
997 * Set open flags for aio engine
999 * Return 0 on success, -1 if the engine specified is invalid
1001 int bdrv_parse_aio(const char *mode
, int *flags
)
1003 if (!strcmp(mode
, "threads")) {
1004 /* do nothing, default */
1005 } else if (!strcmp(mode
, "native")) {
1006 *flags
|= BDRV_O_NATIVE_AIO
;
1007 #ifdef CONFIG_LINUX_IO_URING
1008 } else if (!strcmp(mode
, "io_uring")) {
1009 *flags
|= BDRV_O_IO_URING
;
1019 * Set open flags for a given discard mode
1021 * Return 0 on success, -1 if the discard mode was invalid.
1023 int bdrv_parse_discard_flags(const char *mode
, int *flags
)
1025 *flags
&= ~BDRV_O_UNMAP
;
1027 if (!strcmp(mode
, "off") || !strcmp(mode
, "ignore")) {
1029 } else if (!strcmp(mode
, "on") || !strcmp(mode
, "unmap")) {
1030 *flags
|= BDRV_O_UNMAP
;
1039 * Set open flags for a given cache mode
1041 * Return 0 on success, -1 if the cache mode was invalid.
1043 int bdrv_parse_cache_mode(const char *mode
, int *flags
, bool *writethrough
)
1045 *flags
&= ~BDRV_O_CACHE_MASK
;
1047 if (!strcmp(mode
, "off") || !strcmp(mode
, "none")) {
1048 *writethrough
= false;
1049 *flags
|= BDRV_O_NOCACHE
;
1050 } else if (!strcmp(mode
, "directsync")) {
1051 *writethrough
= true;
1052 *flags
|= BDRV_O_NOCACHE
;
1053 } else if (!strcmp(mode
, "writeback")) {
1054 *writethrough
= false;
1055 } else if (!strcmp(mode
, "unsafe")) {
1056 *writethrough
= false;
1057 *flags
|= BDRV_O_NO_FLUSH
;
1058 } else if (!strcmp(mode
, "writethrough")) {
1059 *writethrough
= true;
1067 static char *bdrv_child_get_parent_desc(BdrvChild
*c
)
1069 BlockDriverState
*parent
= c
->opaque
;
1070 return g_strdup(bdrv_get_device_or_node_name(parent
));
1073 static void bdrv_child_cb_drained_begin(BdrvChild
*child
)
1075 BlockDriverState
*bs
= child
->opaque
;
1076 bdrv_do_drained_begin_quiesce(bs
, NULL
, false);
1079 static bool bdrv_child_cb_drained_poll(BdrvChild
*child
)
1081 BlockDriverState
*bs
= child
->opaque
;
1082 return bdrv_drain_poll(bs
, false, NULL
, false);
1085 static void bdrv_child_cb_drained_end(BdrvChild
*child
,
1086 int *drained_end_counter
)
1088 BlockDriverState
*bs
= child
->opaque
;
1089 bdrv_drained_end_no_poll(bs
, drained_end_counter
);
1092 static void bdrv_child_cb_attach(BdrvChild
*child
)
1094 BlockDriverState
*bs
= child
->opaque
;
1095 bdrv_apply_subtree_drain(child
, bs
);
1098 static void bdrv_child_cb_detach(BdrvChild
*child
)
1100 BlockDriverState
*bs
= child
->opaque
;
1101 bdrv_unapply_subtree_drain(child
, bs
);
1104 static int bdrv_child_cb_inactivate(BdrvChild
*child
)
1106 BlockDriverState
*bs
= child
->opaque
;
1107 assert(bs
->open_flags
& BDRV_O_INACTIVE
);
1111 static bool bdrv_child_cb_can_set_aio_ctx(BdrvChild
*child
, AioContext
*ctx
,
1112 GSList
**ignore
, Error
**errp
)
1114 BlockDriverState
*bs
= child
->opaque
;
1115 return bdrv_can_set_aio_context(bs
, ctx
, ignore
, errp
);
1118 static void bdrv_child_cb_set_aio_ctx(BdrvChild
*child
, AioContext
*ctx
,
1121 BlockDriverState
*bs
= child
->opaque
;
1122 return bdrv_set_aio_context_ignore(bs
, ctx
, ignore
);
1126 * Returns the options and flags that a temporary snapshot should get, based on
1127 * the originally requested flags (the originally requested image will have
1128 * flags like a backing file)
1130 static void bdrv_temp_snapshot_options(int *child_flags
, QDict
*child_options
,
1131 int parent_flags
, QDict
*parent_options
)
1133 *child_flags
= (parent_flags
& ~BDRV_O_SNAPSHOT
) | BDRV_O_TEMPORARY
;
1135 /* For temporary files, unconditional cache=unsafe is fine */
1136 qdict_set_default_str(child_options
, BDRV_OPT_CACHE_DIRECT
, "off");
1137 qdict_set_default_str(child_options
, BDRV_OPT_CACHE_NO_FLUSH
, "on");
1139 /* Copy the read-only and discard options from the parent */
1140 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_READ_ONLY
);
1141 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_DISCARD
);
1143 /* aio=native doesn't work for cache.direct=off, so disable it for the
1144 * temporary snapshot */
1145 *child_flags
&= ~BDRV_O_NATIVE_AIO
;
1149 * Returns the options and flags that bs->file should get if a protocol driver
1150 * is expected, based on the given options and flags for the parent BDS
1152 static void bdrv_inherited_options(int *child_flags
, QDict
*child_options
,
1153 int parent_flags
, QDict
*parent_options
)
1155 int flags
= parent_flags
;
1157 /* Enable protocol handling, disable format probing for bs->file */
1158 flags
|= BDRV_O_PROTOCOL
;
1160 /* If the cache mode isn't explicitly set, inherit direct and no-flush from
1162 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_CACHE_DIRECT
);
1163 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_CACHE_NO_FLUSH
);
1164 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_FORCE_SHARE
);
1166 /* Inherit the read-only option from the parent if it's not set */
1167 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_READ_ONLY
);
1168 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_AUTO_READ_ONLY
);
1170 /* Our block drivers take care to send flushes and respect unmap policy,
1171 * so we can default to enable both on lower layers regardless of the
1172 * corresponding parent options. */
1173 qdict_set_default_str(child_options
, BDRV_OPT_DISCARD
, "unmap");
1175 /* Clear flags that only apply to the top layer */
1176 flags
&= ~(BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
| BDRV_O_COPY_ON_READ
|
1179 *child_flags
= flags
;
1182 const BdrvChildRole child_file
= {
1183 .parent_is_bds
= true,
1184 .get_parent_desc
= bdrv_child_get_parent_desc
,
1185 .inherit_options
= bdrv_inherited_options
,
1186 .drained_begin
= bdrv_child_cb_drained_begin
,
1187 .drained_poll
= bdrv_child_cb_drained_poll
,
1188 .drained_end
= bdrv_child_cb_drained_end
,
1189 .attach
= bdrv_child_cb_attach
,
1190 .detach
= bdrv_child_cb_detach
,
1191 .inactivate
= bdrv_child_cb_inactivate
,
1192 .can_set_aio_ctx
= bdrv_child_cb_can_set_aio_ctx
,
1193 .set_aio_ctx
= bdrv_child_cb_set_aio_ctx
,
1197 * Returns the options and flags that bs->file should get if the use of formats
1198 * (and not only protocols) is permitted for it, based on the given options and
1199 * flags for the parent BDS
1201 static void bdrv_inherited_fmt_options(int *child_flags
, QDict
*child_options
,
1202 int parent_flags
, QDict
*parent_options
)
1204 child_file
.inherit_options(child_flags
, child_options
,
1205 parent_flags
, parent_options
);
1207 *child_flags
&= ~(BDRV_O_PROTOCOL
| BDRV_O_NO_IO
);
1210 const BdrvChildRole child_format
= {
1211 .parent_is_bds
= true,
1212 .get_parent_desc
= bdrv_child_get_parent_desc
,
1213 .inherit_options
= bdrv_inherited_fmt_options
,
1214 .drained_begin
= bdrv_child_cb_drained_begin
,
1215 .drained_poll
= bdrv_child_cb_drained_poll
,
1216 .drained_end
= bdrv_child_cb_drained_end
,
1217 .attach
= bdrv_child_cb_attach
,
1218 .detach
= bdrv_child_cb_detach
,
1219 .inactivate
= bdrv_child_cb_inactivate
,
1220 .can_set_aio_ctx
= bdrv_child_cb_can_set_aio_ctx
,
1221 .set_aio_ctx
= bdrv_child_cb_set_aio_ctx
,
1224 static void bdrv_backing_attach(BdrvChild
*c
)
1226 BlockDriverState
*parent
= c
->opaque
;
1227 BlockDriverState
*backing_hd
= c
->bs
;
1229 assert(!parent
->backing_blocker
);
1230 error_setg(&parent
->backing_blocker
,
1231 "node is used as backing hd of '%s'",
1232 bdrv_get_device_or_node_name(parent
));
1234 bdrv_refresh_filename(backing_hd
);
1236 parent
->open_flags
&= ~BDRV_O_NO_BACKING
;
1237 pstrcpy(parent
->backing_file
, sizeof(parent
->backing_file
),
1238 backing_hd
->filename
);
1239 pstrcpy(parent
->backing_format
, sizeof(parent
->backing_format
),
1240 backing_hd
->drv
? backing_hd
->drv
->format_name
: "");
1242 bdrv_op_block_all(backing_hd
, parent
->backing_blocker
);
1243 /* Otherwise we won't be able to commit or stream */
1244 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_COMMIT_TARGET
,
1245 parent
->backing_blocker
);
1246 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_STREAM
,
1247 parent
->backing_blocker
);
1249 * We do backup in 3 ways:
1251 * The target bs is new opened, and the source is top BDS
1252 * 2. blockdev backup
1253 * Both the source and the target are top BDSes.
1254 * 3. internal backup(used for block replication)
1255 * Both the source and the target are backing file
1257 * In case 1 and 2, neither the source nor the target is the backing file.
1258 * In case 3, we will block the top BDS, so there is only one block job
1259 * for the top BDS and its backing chain.
1261 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_BACKUP_SOURCE
,
1262 parent
->backing_blocker
);
1263 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_BACKUP_TARGET
,
1264 parent
->backing_blocker
);
1266 bdrv_child_cb_attach(c
);
1269 static void bdrv_backing_detach(BdrvChild
*c
)
1271 BlockDriverState
*parent
= c
->opaque
;
1273 assert(parent
->backing_blocker
);
1274 bdrv_op_unblock_all(c
->bs
, parent
->backing_blocker
);
1275 error_free(parent
->backing_blocker
);
1276 parent
->backing_blocker
= NULL
;
1278 bdrv_child_cb_detach(c
);
1282 * Returns the options and flags that bs->backing should get, based on the
1283 * given options and flags for the parent BDS
1285 static void bdrv_backing_options(int *child_flags
, QDict
*child_options
,
1286 int parent_flags
, QDict
*parent_options
)
1288 int flags
= parent_flags
;
1290 /* The cache mode is inherited unmodified for backing files; except WCE,
1291 * which is only applied on the top level (BlockBackend) */
1292 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_CACHE_DIRECT
);
1293 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_CACHE_NO_FLUSH
);
1294 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_FORCE_SHARE
);
1296 /* backing files always opened read-only */
1297 qdict_set_default_str(child_options
, BDRV_OPT_READ_ONLY
, "on");
1298 qdict_set_default_str(child_options
, BDRV_OPT_AUTO_READ_ONLY
, "off");
1299 flags
&= ~BDRV_O_COPY_ON_READ
;
1301 /* snapshot=on is handled on the top layer */
1302 flags
&= ~(BDRV_O_SNAPSHOT
| BDRV_O_TEMPORARY
);
1304 *child_flags
= flags
;
1307 static int bdrv_backing_update_filename(BdrvChild
*c
, BlockDriverState
*base
,
1308 const char *filename
, Error
**errp
)
1310 BlockDriverState
*parent
= c
->opaque
;
1311 bool read_only
= bdrv_is_read_only(parent
);
1315 ret
= bdrv_reopen_set_read_only(parent
, false, errp
);
1321 ret
= bdrv_change_backing_file(parent
, filename
,
1322 base
->drv
? base
->drv
->format_name
: "");
1324 error_setg_errno(errp
, -ret
, "Could not update backing file link");
1328 bdrv_reopen_set_read_only(parent
, true, NULL
);
1334 const BdrvChildRole child_backing
= {
1335 .parent_is_bds
= true,
1336 .get_parent_desc
= bdrv_child_get_parent_desc
,
1337 .attach
= bdrv_backing_attach
,
1338 .detach
= bdrv_backing_detach
,
1339 .inherit_options
= bdrv_backing_options
,
1340 .drained_begin
= bdrv_child_cb_drained_begin
,
1341 .drained_poll
= bdrv_child_cb_drained_poll
,
1342 .drained_end
= bdrv_child_cb_drained_end
,
1343 .inactivate
= bdrv_child_cb_inactivate
,
1344 .update_filename
= bdrv_backing_update_filename
,
1345 .can_set_aio_ctx
= bdrv_child_cb_can_set_aio_ctx
,
1346 .set_aio_ctx
= bdrv_child_cb_set_aio_ctx
,
1349 static int bdrv_open_flags(BlockDriverState
*bs
, int flags
)
1351 int open_flags
= flags
;
1354 * Clear flags that are internal to the block layer before opening the
1357 open_flags
&= ~(BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
| BDRV_O_PROTOCOL
);
1362 static void update_flags_from_options(int *flags
, QemuOpts
*opts
)
1364 *flags
&= ~(BDRV_O_CACHE_MASK
| BDRV_O_RDWR
| BDRV_O_AUTO_RDONLY
);
1366 if (qemu_opt_get_bool_del(opts
, BDRV_OPT_CACHE_NO_FLUSH
, false)) {
1367 *flags
|= BDRV_O_NO_FLUSH
;
1370 if (qemu_opt_get_bool_del(opts
, BDRV_OPT_CACHE_DIRECT
, false)) {
1371 *flags
|= BDRV_O_NOCACHE
;
1374 if (!qemu_opt_get_bool_del(opts
, BDRV_OPT_READ_ONLY
, false)) {
1375 *flags
|= BDRV_O_RDWR
;
1378 if (qemu_opt_get_bool_del(opts
, BDRV_OPT_AUTO_READ_ONLY
, false)) {
1379 *flags
|= BDRV_O_AUTO_RDONLY
;
1383 static void update_options_from_flags(QDict
*options
, int flags
)
1385 if (!qdict_haskey(options
, BDRV_OPT_CACHE_DIRECT
)) {
1386 qdict_put_bool(options
, BDRV_OPT_CACHE_DIRECT
, flags
& BDRV_O_NOCACHE
);
1388 if (!qdict_haskey(options
, BDRV_OPT_CACHE_NO_FLUSH
)) {
1389 qdict_put_bool(options
, BDRV_OPT_CACHE_NO_FLUSH
,
1390 flags
& BDRV_O_NO_FLUSH
);
1392 if (!qdict_haskey(options
, BDRV_OPT_READ_ONLY
)) {
1393 qdict_put_bool(options
, BDRV_OPT_READ_ONLY
, !(flags
& BDRV_O_RDWR
));
1395 if (!qdict_haskey(options
, BDRV_OPT_AUTO_READ_ONLY
)) {
1396 qdict_put_bool(options
, BDRV_OPT_AUTO_READ_ONLY
,
1397 flags
& BDRV_O_AUTO_RDONLY
);
1401 static void bdrv_assign_node_name(BlockDriverState
*bs
,
1402 const char *node_name
,
1405 char *gen_node_name
= NULL
;
1408 node_name
= gen_node_name
= id_generate(ID_BLOCK
);
1409 } else if (!id_wellformed(node_name
)) {
1411 * Check for empty string or invalid characters, but not if it is
1412 * generated (generated names use characters not available to the user)
1414 error_setg(errp
, "Invalid node name");
1418 /* takes care of avoiding namespaces collisions */
1419 if (blk_by_name(node_name
)) {
1420 error_setg(errp
, "node-name=%s is conflicting with a device id",
1425 /* takes care of avoiding duplicates node names */
1426 if (bdrv_find_node(node_name
)) {
1427 error_setg(errp
, "Duplicate node name");
1431 /* Make sure that the node name isn't truncated */
1432 if (strlen(node_name
) >= sizeof(bs
->node_name
)) {
1433 error_setg(errp
, "Node name too long");
1437 /* copy node name into the bs and insert it into the graph list */
1438 pstrcpy(bs
->node_name
, sizeof(bs
->node_name
), node_name
);
1439 QTAILQ_INSERT_TAIL(&graph_bdrv_states
, bs
, node_list
);
1441 g_free(gen_node_name
);
1444 static int bdrv_open_driver(BlockDriverState
*bs
, BlockDriver
*drv
,
1445 const char *node_name
, QDict
*options
,
1446 int open_flags
, Error
**errp
)
1448 Error
*local_err
= NULL
;
1451 bdrv_assign_node_name(bs
, node_name
, &local_err
);
1453 error_propagate(errp
, local_err
);
1458 bs
->read_only
= !(bs
->open_flags
& BDRV_O_RDWR
);
1459 bs
->opaque
= g_malloc0(drv
->instance_size
);
1461 if (drv
->bdrv_file_open
) {
1462 assert(!drv
->bdrv_needs_filename
|| bs
->filename
[0]);
1463 ret
= drv
->bdrv_file_open(bs
, options
, open_flags
, &local_err
);
1464 } else if (drv
->bdrv_open
) {
1465 ret
= drv
->bdrv_open(bs
, options
, open_flags
, &local_err
);
1472 error_propagate(errp
, local_err
);
1473 } else if (bs
->filename
[0]) {
1474 error_setg_errno(errp
, -ret
, "Could not open '%s'", bs
->filename
);
1476 error_setg_errno(errp
, -ret
, "Could not open image");
1481 ret
= refresh_total_sectors(bs
, bs
->total_sectors
);
1483 error_setg_errno(errp
, -ret
, "Could not refresh total sector count");
1487 bdrv_refresh_limits(bs
, &local_err
);
1489 error_propagate(errp
, local_err
);
1493 assert(bdrv_opt_mem_align(bs
) != 0);
1494 assert(bdrv_min_mem_align(bs
) != 0);
1495 assert(is_power_of_2(bs
->bl
.request_alignment
));
1497 for (i
= 0; i
< bs
->quiesce_counter
; i
++) {
1498 if (drv
->bdrv_co_drain_begin
) {
1499 drv
->bdrv_co_drain_begin(bs
);
1506 if (bs
->file
!= NULL
) {
1507 bdrv_unref_child(bs
, bs
->file
);
1515 BlockDriverState
*bdrv_new_open_driver(BlockDriver
*drv
, const char *node_name
,
1516 int flags
, Error
**errp
)
1518 BlockDriverState
*bs
;
1522 bs
->open_flags
= flags
;
1523 bs
->explicit_options
= qdict_new();
1524 bs
->options
= qdict_new();
1527 update_options_from_flags(bs
->options
, flags
);
1529 ret
= bdrv_open_driver(bs
, drv
, node_name
, bs
->options
, flags
, errp
);
1531 qobject_unref(bs
->explicit_options
);
1532 bs
->explicit_options
= NULL
;
1533 qobject_unref(bs
->options
);
1542 QemuOptsList bdrv_runtime_opts
= {
1543 .name
= "bdrv_common",
1544 .head
= QTAILQ_HEAD_INITIALIZER(bdrv_runtime_opts
.head
),
1547 .name
= "node-name",
1548 .type
= QEMU_OPT_STRING
,
1549 .help
= "Node name of the block device node",
1553 .type
= QEMU_OPT_STRING
,
1554 .help
= "Block driver to use for the node",
1557 .name
= BDRV_OPT_CACHE_DIRECT
,
1558 .type
= QEMU_OPT_BOOL
,
1559 .help
= "Bypass software writeback cache on the host",
1562 .name
= BDRV_OPT_CACHE_NO_FLUSH
,
1563 .type
= QEMU_OPT_BOOL
,
1564 .help
= "Ignore flush requests",
1567 .name
= BDRV_OPT_READ_ONLY
,
1568 .type
= QEMU_OPT_BOOL
,
1569 .help
= "Node is opened in read-only mode",
1572 .name
= BDRV_OPT_AUTO_READ_ONLY
,
1573 .type
= QEMU_OPT_BOOL
,
1574 .help
= "Node can become read-only if opening read-write fails",
1577 .name
= "detect-zeroes",
1578 .type
= QEMU_OPT_STRING
,
1579 .help
= "try to optimize zero writes (off, on, unmap)",
1582 .name
= BDRV_OPT_DISCARD
,
1583 .type
= QEMU_OPT_STRING
,
1584 .help
= "discard operation (ignore/off, unmap/on)",
1587 .name
= BDRV_OPT_FORCE_SHARE
,
1588 .type
= QEMU_OPT_BOOL
,
1589 .help
= "always accept other writers (default: off)",
1591 { /* end of list */ }
1595 static QemuOptsList fallback_create_opts
= {
1596 .name
= "fallback-create-opts",
1597 .head
= QTAILQ_HEAD_INITIALIZER(fallback_create_opts
.head
),
1600 .name
= BLOCK_OPT_SIZE
,
1601 .type
= QEMU_OPT_SIZE
,
1602 .help
= "Virtual disk size"
1605 .name
= BLOCK_OPT_PREALLOC
,
1606 .type
= QEMU_OPT_STRING
,
1607 .help
= "Preallocation mode (allowed values: off)"
1609 { /* end of list */ }
1614 * Common part for opening disk images and files
1616 * Removes all processed options from *options.
1618 static int bdrv_open_common(BlockDriverState
*bs
, BlockBackend
*file
,
1619 QDict
*options
, Error
**errp
)
1621 int ret
, open_flags
;
1622 const char *filename
;
1623 const char *driver_name
= NULL
;
1624 const char *node_name
= NULL
;
1625 const char *discard
;
1628 Error
*local_err
= NULL
;
1630 assert(bs
->file
== NULL
);
1631 assert(options
!= NULL
&& bs
->options
!= options
);
1633 opts
= qemu_opts_create(&bdrv_runtime_opts
, NULL
, 0, &error_abort
);
1634 qemu_opts_absorb_qdict(opts
, options
, &local_err
);
1636 error_propagate(errp
, local_err
);
1641 update_flags_from_options(&bs
->open_flags
, opts
);
1643 driver_name
= qemu_opt_get(opts
, "driver");
1644 drv
= bdrv_find_format(driver_name
);
1645 assert(drv
!= NULL
);
1647 bs
->force_share
= qemu_opt_get_bool(opts
, BDRV_OPT_FORCE_SHARE
, false);
1649 if (bs
->force_share
&& (bs
->open_flags
& BDRV_O_RDWR
)) {
1651 BDRV_OPT_FORCE_SHARE
1652 "=on can only be used with read-only images");
1658 bdrv_refresh_filename(blk_bs(file
));
1659 filename
= blk_bs(file
)->filename
;
1662 * Caution: while qdict_get_try_str() is fine, getting
1663 * non-string types would require more care. When @options
1664 * come from -blockdev or blockdev_add, its members are typed
1665 * according to the QAPI schema, but when they come from
1666 * -drive, they're all QString.
1668 filename
= qdict_get_try_str(options
, "filename");
1671 if (drv
->bdrv_needs_filename
&& (!filename
|| !filename
[0])) {
1672 error_setg(errp
, "The '%s' block driver requires a file name",
1678 trace_bdrv_open_common(bs
, filename
?: "", bs
->open_flags
,
1681 bs
->read_only
= !(bs
->open_flags
& BDRV_O_RDWR
);
1683 if (use_bdrv_whitelist
&& !bdrv_is_whitelisted(drv
, bs
->read_only
)) {
1684 if (!bs
->read_only
&& bdrv_is_whitelisted(drv
, true)) {
1685 ret
= bdrv_apply_auto_read_only(bs
, NULL
, NULL
);
1691 !bs
->read_only
&& bdrv_is_whitelisted(drv
, true)
1692 ? "Driver '%s' can only be used for read-only devices"
1693 : "Driver '%s' is not whitelisted",
1699 /* bdrv_new() and bdrv_close() make it so */
1700 assert(atomic_read(&bs
->copy_on_read
) == 0);
1702 if (bs
->open_flags
& BDRV_O_COPY_ON_READ
) {
1703 if (!bs
->read_only
) {
1704 bdrv_enable_copy_on_read(bs
);
1706 error_setg(errp
, "Can't use copy-on-read on read-only device");
1712 discard
= qemu_opt_get(opts
, BDRV_OPT_DISCARD
);
1713 if (discard
!= NULL
) {
1714 if (bdrv_parse_discard_flags(discard
, &bs
->open_flags
) != 0) {
1715 error_setg(errp
, "Invalid discard option");
1722 bdrv_parse_detect_zeroes(opts
, bs
->open_flags
, &local_err
);
1724 error_propagate(errp
, local_err
);
1729 if (filename
!= NULL
) {
1730 pstrcpy(bs
->filename
, sizeof(bs
->filename
), filename
);
1732 bs
->filename
[0] = '\0';
1734 pstrcpy(bs
->exact_filename
, sizeof(bs
->exact_filename
), bs
->filename
);
1736 /* Open the image, either directly or using a protocol */
1737 open_flags
= bdrv_open_flags(bs
, bs
->open_flags
);
1738 node_name
= qemu_opt_get(opts
, "node-name");
1740 assert(!drv
->bdrv_file_open
|| file
== NULL
);
1741 ret
= bdrv_open_driver(bs
, drv
, node_name
, options
, open_flags
, errp
);
1746 qemu_opts_del(opts
);
1750 qemu_opts_del(opts
);
1754 static QDict
*parse_json_filename(const char *filename
, Error
**errp
)
1756 QObject
*options_obj
;
1760 ret
= strstart(filename
, "json:", &filename
);
1763 options_obj
= qobject_from_json(filename
, errp
);
1765 error_prepend(errp
, "Could not parse the JSON options: ");
1769 options
= qobject_to(QDict
, options_obj
);
1771 qobject_unref(options_obj
);
1772 error_setg(errp
, "Invalid JSON object given");
1776 qdict_flatten(options
);
1781 static void parse_json_protocol(QDict
*options
, const char **pfilename
,
1784 QDict
*json_options
;
1785 Error
*local_err
= NULL
;
1787 /* Parse json: pseudo-protocol */
1788 if (!*pfilename
|| !g_str_has_prefix(*pfilename
, "json:")) {
1792 json_options
= parse_json_filename(*pfilename
, &local_err
);
1794 error_propagate(errp
, local_err
);
1798 /* Options given in the filename have lower priority than options
1799 * specified directly */
1800 qdict_join(options
, json_options
, false);
1801 qobject_unref(json_options
);
1806 * Fills in default options for opening images and converts the legacy
1807 * filename/flags pair to option QDict entries.
1808 * The BDRV_O_PROTOCOL flag in *flags will be set or cleared accordingly if a
1809 * block driver has been specified explicitly.
1811 static int bdrv_fill_options(QDict
**options
, const char *filename
,
1812 int *flags
, Error
**errp
)
1814 const char *drvname
;
1815 bool protocol
= *flags
& BDRV_O_PROTOCOL
;
1816 bool parse_filename
= false;
1817 BlockDriver
*drv
= NULL
;
1818 Error
*local_err
= NULL
;
1821 * Caution: while qdict_get_try_str() is fine, getting non-string
1822 * types would require more care. When @options come from
1823 * -blockdev or blockdev_add, its members are typed according to
1824 * the QAPI schema, but when they come from -drive, they're all
1827 drvname
= qdict_get_try_str(*options
, "driver");
1829 drv
= bdrv_find_format(drvname
);
1831 error_setg(errp
, "Unknown driver '%s'", drvname
);
1834 /* If the user has explicitly specified the driver, this choice should
1835 * override the BDRV_O_PROTOCOL flag */
1836 protocol
= drv
->bdrv_file_open
;
1840 *flags
|= BDRV_O_PROTOCOL
;
1842 *flags
&= ~BDRV_O_PROTOCOL
;
1845 /* Translate cache options from flags into options */
1846 update_options_from_flags(*options
, *flags
);
1848 /* Fetch the file name from the options QDict if necessary */
1849 if (protocol
&& filename
) {
1850 if (!qdict_haskey(*options
, "filename")) {
1851 qdict_put_str(*options
, "filename", filename
);
1852 parse_filename
= true;
1854 error_setg(errp
, "Can't specify 'file' and 'filename' options at "
1860 /* Find the right block driver */
1861 /* See cautionary note on accessing @options above */
1862 filename
= qdict_get_try_str(*options
, "filename");
1864 if (!drvname
&& protocol
) {
1866 drv
= bdrv_find_protocol(filename
, parse_filename
, errp
);
1871 drvname
= drv
->format_name
;
1872 qdict_put_str(*options
, "driver", drvname
);
1874 error_setg(errp
, "Must specify either driver or file");
1879 assert(drv
|| !protocol
);
1881 /* Driver-specific filename parsing */
1882 if (drv
&& drv
->bdrv_parse_filename
&& parse_filename
) {
1883 drv
->bdrv_parse_filename(filename
, *options
, &local_err
);
1885 error_propagate(errp
, local_err
);
1889 if (!drv
->bdrv_needs_filename
) {
1890 qdict_del(*options
, "filename");
1897 static int bdrv_child_check_perm(BdrvChild
*c
, BlockReopenQueue
*q
,
1898 uint64_t perm
, uint64_t shared
,
1899 GSList
*ignore_children
,
1900 bool *tighten_restrictions
, Error
**errp
);
1901 static void bdrv_child_abort_perm_update(BdrvChild
*c
);
1902 static void bdrv_child_set_perm(BdrvChild
*c
, uint64_t perm
, uint64_t shared
);
1904 typedef struct BlockReopenQueueEntry
{
1907 BDRVReopenState state
;
1908 QTAILQ_ENTRY(BlockReopenQueueEntry
) entry
;
1909 } BlockReopenQueueEntry
;
1912 * Return the flags that @bs will have after the reopens in @q have
1913 * successfully completed. If @q is NULL (or @bs is not contained in @q),
1914 * return the current flags.
1916 static int bdrv_reopen_get_flags(BlockReopenQueue
*q
, BlockDriverState
*bs
)
1918 BlockReopenQueueEntry
*entry
;
1921 QTAILQ_FOREACH(entry
, q
, entry
) {
1922 if (entry
->state
.bs
== bs
) {
1923 return entry
->state
.flags
;
1928 return bs
->open_flags
;
1931 /* Returns whether the image file can be written to after the reopen queue @q
1932 * has been successfully applied, or right now if @q is NULL. */
1933 static bool bdrv_is_writable_after_reopen(BlockDriverState
*bs
,
1934 BlockReopenQueue
*q
)
1936 int flags
= bdrv_reopen_get_flags(q
, bs
);
1938 return (flags
& (BDRV_O_RDWR
| BDRV_O_INACTIVE
)) == BDRV_O_RDWR
;
1942 * Return whether the BDS can be written to. This is not necessarily
1943 * the same as !bdrv_is_read_only(bs), as inactivated images may not
1944 * be written to but do not count as read-only images.
1946 bool bdrv_is_writable(BlockDriverState
*bs
)
1948 return bdrv_is_writable_after_reopen(bs
, NULL
);
1951 static void bdrv_child_perm(BlockDriverState
*bs
, BlockDriverState
*child_bs
,
1952 BdrvChild
*c
, const BdrvChildRole
*role
,
1953 BlockReopenQueue
*reopen_queue
,
1954 uint64_t parent_perm
, uint64_t parent_shared
,
1955 uint64_t *nperm
, uint64_t *nshared
)
1957 assert(bs
->drv
&& bs
->drv
->bdrv_child_perm
);
1958 bs
->drv
->bdrv_child_perm(bs
, c
, role
, reopen_queue
,
1959 parent_perm
, parent_shared
,
1961 /* TODO Take force_share from reopen_queue */
1962 if (child_bs
&& child_bs
->force_share
) {
1963 *nshared
= BLK_PERM_ALL
;
1968 * Check whether permissions on this node can be changed in a way that
1969 * @cumulative_perms and @cumulative_shared_perms are the new cumulative
1970 * permissions of all its parents. This involves checking whether all necessary
1971 * permission changes to child nodes can be performed.
1973 * Will set *tighten_restrictions to true if and only if new permissions have to
1974 * be taken or currently shared permissions are to be unshared. Otherwise,
1975 * errors are not fatal as long as the caller accepts that the restrictions
1976 * remain tighter than they need to be. The caller still has to abort the
1978 * @tighten_restrictions cannot be used together with @q: When reopening, we may
1979 * encounter fatal errors even though no restrictions are to be tightened. For
1980 * example, changing a node from RW to RO will fail if the WRITE permission is
1983 * A call to this function must always be followed by a call to bdrv_set_perm()
1984 * or bdrv_abort_perm_update().
1986 static int bdrv_check_perm(BlockDriverState
*bs
, BlockReopenQueue
*q
,
1987 uint64_t cumulative_perms
,
1988 uint64_t cumulative_shared_perms
,
1989 GSList
*ignore_children
,
1990 bool *tighten_restrictions
, Error
**errp
)
1992 BlockDriver
*drv
= bs
->drv
;
1996 assert(!q
|| !tighten_restrictions
);
1998 if (tighten_restrictions
) {
1999 uint64_t current_perms
, current_shared
;
2000 uint64_t added_perms
, removed_shared_perms
;
2002 bdrv_get_cumulative_perm(bs
, ¤t_perms
, ¤t_shared
);
2004 added_perms
= cumulative_perms
& ~current_perms
;
2005 removed_shared_perms
= current_shared
& ~cumulative_shared_perms
;
2007 *tighten_restrictions
= added_perms
|| removed_shared_perms
;
2010 /* Write permissions never work with read-only images */
2011 if ((cumulative_perms
& (BLK_PERM_WRITE
| BLK_PERM_WRITE_UNCHANGED
)) &&
2012 !bdrv_is_writable_after_reopen(bs
, q
))
2014 if (!bdrv_is_writable_after_reopen(bs
, NULL
)) {
2015 error_setg(errp
, "Block node is read-only");
2017 uint64_t current_perms
, current_shared
;
2018 bdrv_get_cumulative_perm(bs
, ¤t_perms
, ¤t_shared
);
2019 if (current_perms
& (BLK_PERM_WRITE
| BLK_PERM_WRITE_UNCHANGED
)) {
2020 error_setg(errp
, "Cannot make block node read-only, there is "
2023 error_setg(errp
, "Cannot make block node read-only and create "
2031 /* Check this node */
2036 if (drv
->bdrv_check_perm
) {
2037 return drv
->bdrv_check_perm(bs
, cumulative_perms
,
2038 cumulative_shared_perms
, errp
);
2041 /* Drivers that never have children can omit .bdrv_child_perm() */
2042 if (!drv
->bdrv_child_perm
) {
2043 assert(QLIST_EMPTY(&bs
->children
));
2047 /* Check all children */
2048 QLIST_FOREACH(c
, &bs
->children
, next
) {
2049 uint64_t cur_perm
, cur_shared
;
2050 bool child_tighten_restr
;
2052 bdrv_child_perm(bs
, c
->bs
, c
, c
->role
, q
,
2053 cumulative_perms
, cumulative_shared_perms
,
2054 &cur_perm
, &cur_shared
);
2055 ret
= bdrv_child_check_perm(c
, q
, cur_perm
, cur_shared
, ignore_children
,
2056 tighten_restrictions
? &child_tighten_restr
2059 if (tighten_restrictions
) {
2060 *tighten_restrictions
|= child_tighten_restr
;
2071 * Notifies drivers that after a previous bdrv_check_perm() call, the
2072 * permission update is not performed and any preparations made for it (e.g.
2073 * taken file locks) need to be undone.
2075 * This function recursively notifies all child nodes.
2077 static void bdrv_abort_perm_update(BlockDriverState
*bs
)
2079 BlockDriver
*drv
= bs
->drv
;
2086 if (drv
->bdrv_abort_perm_update
) {
2087 drv
->bdrv_abort_perm_update(bs
);
2090 QLIST_FOREACH(c
, &bs
->children
, next
) {
2091 bdrv_child_abort_perm_update(c
);
2095 static void bdrv_set_perm(BlockDriverState
*bs
, uint64_t cumulative_perms
,
2096 uint64_t cumulative_shared_perms
)
2098 BlockDriver
*drv
= bs
->drv
;
2105 /* Update this node */
2106 if (drv
->bdrv_set_perm
) {
2107 drv
->bdrv_set_perm(bs
, cumulative_perms
, cumulative_shared_perms
);
2110 /* Drivers that never have children can omit .bdrv_child_perm() */
2111 if (!drv
->bdrv_child_perm
) {
2112 assert(QLIST_EMPTY(&bs
->children
));
2116 /* Update all children */
2117 QLIST_FOREACH(c
, &bs
->children
, next
) {
2118 uint64_t cur_perm
, cur_shared
;
2119 bdrv_child_perm(bs
, c
->bs
, c
, c
->role
, NULL
,
2120 cumulative_perms
, cumulative_shared_perms
,
2121 &cur_perm
, &cur_shared
);
2122 bdrv_child_set_perm(c
, cur_perm
, cur_shared
);
2126 void bdrv_get_cumulative_perm(BlockDriverState
*bs
, uint64_t *perm
,
2127 uint64_t *shared_perm
)
2130 uint64_t cumulative_perms
= 0;
2131 uint64_t cumulative_shared_perms
= BLK_PERM_ALL
;
2133 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
2134 cumulative_perms
|= c
->perm
;
2135 cumulative_shared_perms
&= c
->shared_perm
;
2138 *perm
= cumulative_perms
;
2139 *shared_perm
= cumulative_shared_perms
;
2142 static char *bdrv_child_user_desc(BdrvChild
*c
)
2144 if (c
->role
->get_parent_desc
) {
2145 return c
->role
->get_parent_desc(c
);
2148 return g_strdup("another user");
2151 char *bdrv_perm_names(uint64_t perm
)
2157 { BLK_PERM_CONSISTENT_READ
, "consistent read" },
2158 { BLK_PERM_WRITE
, "write" },
2159 { BLK_PERM_WRITE_UNCHANGED
, "write unchanged" },
2160 { BLK_PERM_RESIZE
, "resize" },
2161 { BLK_PERM_GRAPH_MOD
, "change children" },
2165 GString
*result
= g_string_sized_new(30);
2166 struct perm_name
*p
;
2168 for (p
= permissions
; p
->name
; p
++) {
2169 if (perm
& p
->perm
) {
2170 if (result
->len
> 0) {
2171 g_string_append(result
, ", ");
2173 g_string_append(result
, p
->name
);
2177 return g_string_free(result
, FALSE
);
2181 * Checks whether a new reference to @bs can be added if the new user requires
2182 * @new_used_perm/@new_shared_perm as its permissions. If @ignore_children is
2183 * set, the BdrvChild objects in this list are ignored in the calculations;
2184 * this allows checking permission updates for an existing reference.
2186 * See bdrv_check_perm() for the semantics of @tighten_restrictions.
2188 * Needs to be followed by a call to either bdrv_set_perm() or
2189 * bdrv_abort_perm_update(). */
2190 static int bdrv_check_update_perm(BlockDriverState
*bs
, BlockReopenQueue
*q
,
2191 uint64_t new_used_perm
,
2192 uint64_t new_shared_perm
,
2193 GSList
*ignore_children
,
2194 bool *tighten_restrictions
,
2198 uint64_t cumulative_perms
= new_used_perm
;
2199 uint64_t cumulative_shared_perms
= new_shared_perm
;
2201 assert(!q
|| !tighten_restrictions
);
2203 /* There is no reason why anyone couldn't tolerate write_unchanged */
2204 assert(new_shared_perm
& BLK_PERM_WRITE_UNCHANGED
);
2206 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
2207 if (g_slist_find(ignore_children
, c
)) {
2211 if ((new_used_perm
& c
->shared_perm
) != new_used_perm
) {
2212 char *user
= bdrv_child_user_desc(c
);
2213 char *perm_names
= bdrv_perm_names(new_used_perm
& ~c
->shared_perm
);
2215 if (tighten_restrictions
) {
2216 *tighten_restrictions
= true;
2219 error_setg(errp
, "Conflicts with use by %s as '%s', which does not "
2221 user
, c
->name
, perm_names
, bdrv_get_node_name(c
->bs
));
2227 if ((c
->perm
& new_shared_perm
) != c
->perm
) {
2228 char *user
= bdrv_child_user_desc(c
);
2229 char *perm_names
= bdrv_perm_names(c
->perm
& ~new_shared_perm
);
2231 if (tighten_restrictions
) {
2232 *tighten_restrictions
= true;
2235 error_setg(errp
, "Conflicts with use by %s as '%s', which uses "
2237 user
, c
->name
, perm_names
, bdrv_get_node_name(c
->bs
));
2243 cumulative_perms
|= c
->perm
;
2244 cumulative_shared_perms
&= c
->shared_perm
;
2247 return bdrv_check_perm(bs
, q
, cumulative_perms
, cumulative_shared_perms
,
2248 ignore_children
, tighten_restrictions
, errp
);
2251 /* Needs to be followed by a call to either bdrv_child_set_perm() or
2252 * bdrv_child_abort_perm_update(). */
2253 static int bdrv_child_check_perm(BdrvChild
*c
, BlockReopenQueue
*q
,
2254 uint64_t perm
, uint64_t shared
,
2255 GSList
*ignore_children
,
2256 bool *tighten_restrictions
, Error
**errp
)
2260 ignore_children
= g_slist_prepend(g_slist_copy(ignore_children
), c
);
2261 ret
= bdrv_check_update_perm(c
->bs
, q
, perm
, shared
, ignore_children
,
2262 tighten_restrictions
, errp
);
2263 g_slist_free(ignore_children
);
2269 if (!c
->has_backup_perm
) {
2270 c
->has_backup_perm
= true;
2271 c
->backup_perm
= c
->perm
;
2272 c
->backup_shared_perm
= c
->shared_perm
;
2275 * Note: it's OK if c->has_backup_perm was already set, as we can find the
2276 * same child twice during check_perm procedure
2280 c
->shared_perm
= shared
;
2285 static void bdrv_child_set_perm(BdrvChild
*c
, uint64_t perm
, uint64_t shared
)
2287 uint64_t cumulative_perms
, cumulative_shared_perms
;
2289 c
->has_backup_perm
= false;
2292 c
->shared_perm
= shared
;
2294 bdrv_get_cumulative_perm(c
->bs
, &cumulative_perms
,
2295 &cumulative_shared_perms
);
2296 bdrv_set_perm(c
->bs
, cumulative_perms
, cumulative_shared_perms
);
2299 static void bdrv_child_abort_perm_update(BdrvChild
*c
)
2301 if (c
->has_backup_perm
) {
2302 c
->perm
= c
->backup_perm
;
2303 c
->shared_perm
= c
->backup_shared_perm
;
2304 c
->has_backup_perm
= false;
2307 bdrv_abort_perm_update(c
->bs
);
2310 int bdrv_child_try_set_perm(BdrvChild
*c
, uint64_t perm
, uint64_t shared
,
2313 Error
*local_err
= NULL
;
2315 bool tighten_restrictions
;
2317 ret
= bdrv_child_check_perm(c
, NULL
, perm
, shared
, NULL
,
2318 &tighten_restrictions
, &local_err
);
2320 bdrv_child_abort_perm_update(c
);
2321 if (tighten_restrictions
) {
2322 error_propagate(errp
, local_err
);
2325 * Our caller may intend to only loosen restrictions and
2326 * does not expect this function to fail. Errors are not
2327 * fatal in such a case, so we can just hide them from our
2330 error_free(local_err
);
2336 bdrv_child_set_perm(c
, perm
, shared
);
2341 int bdrv_child_refresh_perms(BlockDriverState
*bs
, BdrvChild
*c
, Error
**errp
)
2343 uint64_t parent_perms
, parent_shared
;
2344 uint64_t perms
, shared
;
2346 bdrv_get_cumulative_perm(bs
, &parent_perms
, &parent_shared
);
2347 bdrv_child_perm(bs
, c
->bs
, c
, c
->role
, NULL
, parent_perms
, parent_shared
,
2350 return bdrv_child_try_set_perm(c
, perms
, shared
, errp
);
2353 void bdrv_filter_default_perms(BlockDriverState
*bs
, BdrvChild
*c
,
2354 const BdrvChildRole
*role
,
2355 BlockReopenQueue
*reopen_queue
,
2356 uint64_t perm
, uint64_t shared
,
2357 uint64_t *nperm
, uint64_t *nshared
)
2359 *nperm
= perm
& DEFAULT_PERM_PASSTHROUGH
;
2360 *nshared
= (shared
& DEFAULT_PERM_PASSTHROUGH
) | DEFAULT_PERM_UNCHANGED
;
2363 void bdrv_format_default_perms(BlockDriverState
*bs
, BdrvChild
*c
,
2364 const BdrvChildRole
*role
,
2365 BlockReopenQueue
*reopen_queue
,
2366 uint64_t perm
, uint64_t shared
,
2367 uint64_t *nperm
, uint64_t *nshared
)
2369 bool backing
= (role
== &child_backing
);
2370 assert(role
== &child_backing
|| role
== &child_file
);
2373 int flags
= bdrv_reopen_get_flags(reopen_queue
, bs
);
2375 /* Apart from the modifications below, the same permissions are
2376 * forwarded and left alone as for filters */
2377 bdrv_filter_default_perms(bs
, c
, role
, reopen_queue
, perm
, shared
,
2380 /* Format drivers may touch metadata even if the guest doesn't write */
2381 if (bdrv_is_writable_after_reopen(bs
, reopen_queue
)) {
2382 perm
|= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2385 /* bs->file always needs to be consistent because of the metadata. We
2386 * can never allow other users to resize or write to it. */
2387 if (!(flags
& BDRV_O_NO_IO
)) {
2388 perm
|= BLK_PERM_CONSISTENT_READ
;
2390 shared
&= ~(BLK_PERM_WRITE
| BLK_PERM_RESIZE
);
2392 /* We want consistent read from backing files if the parent needs it.
2393 * No other operations are performed on backing files. */
2394 perm
&= BLK_PERM_CONSISTENT_READ
;
2396 /* If the parent can deal with changing data, we're okay with a
2397 * writable and resizable backing file. */
2398 /* TODO Require !(perm & BLK_PERM_CONSISTENT_READ), too? */
2399 if (shared
& BLK_PERM_WRITE
) {
2400 shared
= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2405 shared
|= BLK_PERM_CONSISTENT_READ
| BLK_PERM_GRAPH_MOD
|
2406 BLK_PERM_WRITE_UNCHANGED
;
2409 if (bs
->open_flags
& BDRV_O_INACTIVE
) {
2410 shared
|= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2417 uint64_t bdrv_qapi_perm_to_blk_perm(BlockPermission qapi_perm
)
2419 static const uint64_t permissions
[] = {
2420 [BLOCK_PERMISSION_CONSISTENT_READ
] = BLK_PERM_CONSISTENT_READ
,
2421 [BLOCK_PERMISSION_WRITE
] = BLK_PERM_WRITE
,
2422 [BLOCK_PERMISSION_WRITE_UNCHANGED
] = BLK_PERM_WRITE_UNCHANGED
,
2423 [BLOCK_PERMISSION_RESIZE
] = BLK_PERM_RESIZE
,
2424 [BLOCK_PERMISSION_GRAPH_MOD
] = BLK_PERM_GRAPH_MOD
,
2427 QEMU_BUILD_BUG_ON(ARRAY_SIZE(permissions
) != BLOCK_PERMISSION__MAX
);
2428 QEMU_BUILD_BUG_ON(1UL << ARRAY_SIZE(permissions
) != BLK_PERM_ALL
+ 1);
2430 assert(qapi_perm
< BLOCK_PERMISSION__MAX
);
2432 return permissions
[qapi_perm
];
2435 static void bdrv_replace_child_noperm(BdrvChild
*child
,
2436 BlockDriverState
*new_bs
)
2438 BlockDriverState
*old_bs
= child
->bs
;
2439 int new_bs_quiesce_counter
;
2442 assert(!child
->frozen
);
2444 if (old_bs
&& new_bs
) {
2445 assert(bdrv_get_aio_context(old_bs
) == bdrv_get_aio_context(new_bs
));
2448 new_bs_quiesce_counter
= (new_bs
? new_bs
->quiesce_counter
: 0);
2449 drain_saldo
= new_bs_quiesce_counter
- child
->parent_quiesce_counter
;
2452 * If the new child node is drained but the old one was not, flush
2453 * all outstanding requests to the old child node.
2455 while (drain_saldo
> 0 && child
->role
->drained_begin
) {
2456 bdrv_parent_drained_begin_single(child
, true);
2461 /* Detach first so that the recursive drain sections coming from @child
2462 * are already gone and we only end the drain sections that came from
2464 if (child
->role
->detach
) {
2465 child
->role
->detach(child
);
2467 QLIST_REMOVE(child
, next_parent
);
2473 QLIST_INSERT_HEAD(&new_bs
->parents
, child
, next_parent
);
2476 * Detaching the old node may have led to the new node's
2477 * quiesce_counter having been decreased. Not a problem, we
2478 * just need to recognize this here and then invoke
2479 * drained_end appropriately more often.
2481 assert(new_bs
->quiesce_counter
<= new_bs_quiesce_counter
);
2482 drain_saldo
+= new_bs
->quiesce_counter
- new_bs_quiesce_counter
;
2484 /* Attach only after starting new drained sections, so that recursive
2485 * drain sections coming from @child don't get an extra .drained_begin
2487 if (child
->role
->attach
) {
2488 child
->role
->attach(child
);
2493 * If the old child node was drained but the new one is not, allow
2494 * requests to come in only after the new node has been attached.
2496 while (drain_saldo
< 0 && child
->role
->drained_end
) {
2497 bdrv_parent_drained_end_single(child
);
2503 * Updates @child to change its reference to point to @new_bs, including
2504 * checking and applying the necessary permisson updates both to the old node
2507 * NULL is passed as @new_bs for removing the reference before freeing @child.
2509 * If @new_bs is not NULL, bdrv_check_perm() must be called beforehand, as this
2510 * function uses bdrv_set_perm() to update the permissions according to the new
2511 * reference that @new_bs gets.
2513 static void bdrv_replace_child(BdrvChild
*child
, BlockDriverState
*new_bs
)
2515 BlockDriverState
*old_bs
= child
->bs
;
2516 uint64_t perm
, shared_perm
;
2518 bdrv_replace_child_noperm(child
, new_bs
);
2521 * Start with the new node's permissions. If @new_bs is a (direct
2522 * or indirect) child of @old_bs, we must complete the permission
2523 * update on @new_bs before we loosen the restrictions on @old_bs.
2524 * Otherwise, bdrv_check_perm() on @old_bs would re-initiate
2525 * updating the permissions of @new_bs, and thus not purely loosen
2529 bdrv_get_cumulative_perm(new_bs
, &perm
, &shared_perm
);
2530 bdrv_set_perm(new_bs
, perm
, shared_perm
);
2534 /* Update permissions for old node. This is guaranteed to succeed
2535 * because we're just taking a parent away, so we're loosening
2537 bool tighten_restrictions
;
2540 bdrv_get_cumulative_perm(old_bs
, &perm
, &shared_perm
);
2541 ret
= bdrv_check_perm(old_bs
, NULL
, perm
, shared_perm
, NULL
,
2542 &tighten_restrictions
, NULL
);
2543 assert(tighten_restrictions
== false);
2545 /* We only tried to loosen restrictions, so errors are not fatal */
2546 bdrv_abort_perm_update(old_bs
);
2548 bdrv_set_perm(old_bs
, perm
, shared_perm
);
2551 /* When the parent requiring a non-default AioContext is removed, the
2552 * node moves back to the main AioContext */
2553 bdrv_try_set_aio_context(old_bs
, qemu_get_aio_context(), NULL
);
2558 * This function steals the reference to child_bs from the caller.
2559 * That reference is later dropped by bdrv_root_unref_child().
2561 * On failure NULL is returned, errp is set and the reference to
2562 * child_bs is also dropped.
2564 * The caller must hold the AioContext lock @child_bs, but not that of @ctx
2565 * (unless @child_bs is already in @ctx).
2567 BdrvChild
*bdrv_root_attach_child(BlockDriverState
*child_bs
,
2568 const char *child_name
,
2569 const BdrvChildRole
*child_role
,
2571 uint64_t perm
, uint64_t shared_perm
,
2572 void *opaque
, Error
**errp
)
2575 Error
*local_err
= NULL
;
2578 ret
= bdrv_check_update_perm(child_bs
, NULL
, perm
, shared_perm
, NULL
, NULL
,
2581 bdrv_abort_perm_update(child_bs
);
2582 bdrv_unref(child_bs
);
2586 child
= g_new(BdrvChild
, 1);
2587 *child
= (BdrvChild
) {
2589 .name
= g_strdup(child_name
),
2592 .shared_perm
= shared_perm
,
2596 /* If the AioContexts don't match, first try to move the subtree of
2597 * child_bs into the AioContext of the new parent. If this doesn't work,
2598 * try moving the parent into the AioContext of child_bs instead. */
2599 if (bdrv_get_aio_context(child_bs
) != ctx
) {
2600 ret
= bdrv_try_set_aio_context(child_bs
, ctx
, &local_err
);
2601 if (ret
< 0 && child_role
->can_set_aio_ctx
) {
2602 GSList
*ignore
= g_slist_prepend(NULL
, child
);
2603 ctx
= bdrv_get_aio_context(child_bs
);
2604 if (child_role
->can_set_aio_ctx(child
, ctx
, &ignore
, NULL
)) {
2605 error_free(local_err
);
2607 g_slist_free(ignore
);
2608 ignore
= g_slist_prepend(NULL
, child
);
2609 child_role
->set_aio_ctx(child
, ctx
, &ignore
);
2611 g_slist_free(ignore
);
2614 error_propagate(errp
, local_err
);
2616 bdrv_abort_perm_update(child_bs
);
2621 /* This performs the matching bdrv_set_perm() for the above check. */
2622 bdrv_replace_child(child
, child_bs
);
2628 * This function transfers the reference to child_bs from the caller
2629 * to parent_bs. That reference is later dropped by parent_bs on
2630 * bdrv_close() or if someone calls bdrv_unref_child().
2632 * On failure NULL is returned, errp is set and the reference to
2633 * child_bs is also dropped.
2635 * If @parent_bs and @child_bs are in different AioContexts, the caller must
2636 * hold the AioContext lock for @child_bs, but not for @parent_bs.
2638 BdrvChild
*bdrv_attach_child(BlockDriverState
*parent_bs
,
2639 BlockDriverState
*child_bs
,
2640 const char *child_name
,
2641 const BdrvChildRole
*child_role
,
2645 uint64_t perm
, shared_perm
;
2647 bdrv_get_cumulative_perm(parent_bs
, &perm
, &shared_perm
);
2649 assert(parent_bs
->drv
);
2650 bdrv_child_perm(parent_bs
, child_bs
, NULL
, child_role
, NULL
,
2651 perm
, shared_perm
, &perm
, &shared_perm
);
2653 child
= bdrv_root_attach_child(child_bs
, child_name
, child_role
,
2654 bdrv_get_aio_context(parent_bs
),
2655 perm
, shared_perm
, parent_bs
, errp
);
2656 if (child
== NULL
) {
2660 QLIST_INSERT_HEAD(&parent_bs
->children
, child
, next
);
2664 static void bdrv_detach_child(BdrvChild
*child
)
2666 QLIST_SAFE_REMOVE(child
, next
);
2668 bdrv_replace_child(child
, NULL
);
2670 g_free(child
->name
);
2674 void bdrv_root_unref_child(BdrvChild
*child
)
2676 BlockDriverState
*child_bs
;
2678 child_bs
= child
->bs
;
2679 bdrv_detach_child(child
);
2680 bdrv_unref(child_bs
);
2684 * Clear all inherits_from pointers from children and grandchildren of
2685 * @root that point to @root, where necessary.
2687 static void bdrv_unset_inherits_from(BlockDriverState
*root
, BdrvChild
*child
)
2691 if (child
->bs
->inherits_from
== root
) {
2693 * Remove inherits_from only when the last reference between root and
2694 * child->bs goes away.
2696 QLIST_FOREACH(c
, &root
->children
, next
) {
2697 if (c
!= child
&& c
->bs
== child
->bs
) {
2702 child
->bs
->inherits_from
= NULL
;
2706 QLIST_FOREACH(c
, &child
->bs
->children
, next
) {
2707 bdrv_unset_inherits_from(root
, c
);
2711 void bdrv_unref_child(BlockDriverState
*parent
, BdrvChild
*child
)
2713 if (child
== NULL
) {
2717 bdrv_unset_inherits_from(parent
, child
);
2718 bdrv_root_unref_child(child
);
2722 static void bdrv_parent_cb_change_media(BlockDriverState
*bs
, bool load
)
2725 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
2726 if (c
->role
->change_media
) {
2727 c
->role
->change_media(c
, load
);
2732 /* Return true if you can reach parent going through child->inherits_from
2733 * recursively. If parent or child are NULL, return false */
2734 static bool bdrv_inherits_from_recursive(BlockDriverState
*child
,
2735 BlockDriverState
*parent
)
2737 while (child
&& child
!= parent
) {
2738 child
= child
->inherits_from
;
2741 return child
!= NULL
;
2745 * Sets the backing file link of a BDS. A new reference is created; callers
2746 * which don't need their own reference any more must call bdrv_unref().
2748 void bdrv_set_backing_hd(BlockDriverState
*bs
, BlockDriverState
*backing_hd
,
2751 bool update_inherits_from
= bdrv_chain_contains(bs
, backing_hd
) &&
2752 bdrv_inherits_from_recursive(backing_hd
, bs
);
2754 if (bdrv_is_backing_chain_frozen(bs
, backing_bs(bs
), errp
)) {
2759 bdrv_ref(backing_hd
);
2763 bdrv_unref_child(bs
, bs
->backing
);
2771 bs
->backing
= bdrv_attach_child(bs
, backing_hd
, "backing", &child_backing
,
2773 /* If backing_hd was already part of bs's backing chain, and
2774 * inherits_from pointed recursively to bs then let's update it to
2775 * point directly to bs (else it will become NULL). */
2776 if (bs
->backing
&& update_inherits_from
) {
2777 backing_hd
->inherits_from
= bs
;
2781 bdrv_refresh_limits(bs
, NULL
);
2785 * Opens the backing file for a BlockDriverState if not yet open
2787 * bdref_key specifies the key for the image's BlockdevRef in the options QDict.
2788 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
2789 * itself, all options starting with "${bdref_key}." are considered part of the
2792 * TODO Can this be unified with bdrv_open_image()?
2794 int bdrv_open_backing_file(BlockDriverState
*bs
, QDict
*parent_options
,
2795 const char *bdref_key
, Error
**errp
)
2797 char *backing_filename
= NULL
;
2798 char *bdref_key_dot
;
2799 const char *reference
= NULL
;
2801 bool implicit_backing
= false;
2802 BlockDriverState
*backing_hd
;
2804 QDict
*tmp_parent_options
= NULL
;
2805 Error
*local_err
= NULL
;
2807 if (bs
->backing
!= NULL
) {
2811 /* NULL means an empty set of options */
2812 if (parent_options
== NULL
) {
2813 tmp_parent_options
= qdict_new();
2814 parent_options
= tmp_parent_options
;
2817 bs
->open_flags
&= ~BDRV_O_NO_BACKING
;
2819 bdref_key_dot
= g_strdup_printf("%s.", bdref_key
);
2820 qdict_extract_subqdict(parent_options
, &options
, bdref_key_dot
);
2821 g_free(bdref_key_dot
);
2824 * Caution: while qdict_get_try_str() is fine, getting non-string
2825 * types would require more care. When @parent_options come from
2826 * -blockdev or blockdev_add, its members are typed according to
2827 * the QAPI schema, but when they come from -drive, they're all
2830 reference
= qdict_get_try_str(parent_options
, bdref_key
);
2831 if (reference
|| qdict_haskey(options
, "file.filename")) {
2832 /* keep backing_filename NULL */
2833 } else if (bs
->backing_file
[0] == '\0' && qdict_size(options
) == 0) {
2834 qobject_unref(options
);
2837 if (qdict_size(options
) == 0) {
2838 /* If the user specifies options that do not modify the
2839 * backing file's behavior, we might still consider it the
2840 * implicit backing file. But it's easier this way, and
2841 * just specifying some of the backing BDS's options is
2842 * only possible with -drive anyway (otherwise the QAPI
2843 * schema forces the user to specify everything). */
2844 implicit_backing
= !strcmp(bs
->auto_backing_file
, bs
->backing_file
);
2847 backing_filename
= bdrv_get_full_backing_filename(bs
, &local_err
);
2850 error_propagate(errp
, local_err
);
2851 qobject_unref(options
);
2856 if (!bs
->drv
|| !bs
->drv
->supports_backing
) {
2858 error_setg(errp
, "Driver doesn't support backing files");
2859 qobject_unref(options
);
2864 bs
->backing_format
[0] != '\0' && !qdict_haskey(options
, "driver")) {
2865 qdict_put_str(options
, "driver", bs
->backing_format
);
2868 backing_hd
= bdrv_open_inherit(backing_filename
, reference
, options
, 0, bs
,
2869 &child_backing
, errp
);
2871 bs
->open_flags
|= BDRV_O_NO_BACKING
;
2872 error_prepend(errp
, "Could not open backing file: ");
2877 if (implicit_backing
) {
2878 bdrv_refresh_filename(backing_hd
);
2879 pstrcpy(bs
->auto_backing_file
, sizeof(bs
->auto_backing_file
),
2880 backing_hd
->filename
);
2883 /* Hook up the backing file link; drop our reference, bs owns the
2884 * backing_hd reference now */
2885 bdrv_set_backing_hd(bs
, backing_hd
, &local_err
);
2886 bdrv_unref(backing_hd
);
2888 error_propagate(errp
, local_err
);
2893 qdict_del(parent_options
, bdref_key
);
2896 g_free(backing_filename
);
2897 qobject_unref(tmp_parent_options
);
2901 static BlockDriverState
*
2902 bdrv_open_child_bs(const char *filename
, QDict
*options
, const char *bdref_key
,
2903 BlockDriverState
*parent
, const BdrvChildRole
*child_role
,
2904 bool allow_none
, Error
**errp
)
2906 BlockDriverState
*bs
= NULL
;
2907 QDict
*image_options
;
2908 char *bdref_key_dot
;
2909 const char *reference
;
2911 assert(child_role
!= NULL
);
2913 bdref_key_dot
= g_strdup_printf("%s.", bdref_key
);
2914 qdict_extract_subqdict(options
, &image_options
, bdref_key_dot
);
2915 g_free(bdref_key_dot
);
2918 * Caution: while qdict_get_try_str() is fine, getting non-string
2919 * types would require more care. When @options come from
2920 * -blockdev or blockdev_add, its members are typed according to
2921 * the QAPI schema, but when they come from -drive, they're all
2924 reference
= qdict_get_try_str(options
, bdref_key
);
2925 if (!filename
&& !reference
&& !qdict_size(image_options
)) {
2927 error_setg(errp
, "A block device must be specified for \"%s\"",
2930 qobject_unref(image_options
);
2934 bs
= bdrv_open_inherit(filename
, reference
, image_options
, 0,
2935 parent
, child_role
, errp
);
2941 qdict_del(options
, bdref_key
);
2946 * Opens a disk image whose options are given as BlockdevRef in another block
2949 * If allow_none is true, no image will be opened if filename is false and no
2950 * BlockdevRef is given. NULL will be returned, but errp remains unset.
2952 * bdrev_key specifies the key for the image's BlockdevRef in the options QDict.
2953 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
2954 * itself, all options starting with "${bdref_key}." are considered part of the
2957 * The BlockdevRef will be removed from the options QDict.
2959 BdrvChild
*bdrv_open_child(const char *filename
,
2960 QDict
*options
, const char *bdref_key
,
2961 BlockDriverState
*parent
,
2962 const BdrvChildRole
*child_role
,
2963 bool allow_none
, Error
**errp
)
2965 BlockDriverState
*bs
;
2967 bs
= bdrv_open_child_bs(filename
, options
, bdref_key
, parent
, child_role
,
2973 return bdrv_attach_child(parent
, bs
, bdref_key
, child_role
, errp
);
2976 /* TODO Future callers may need to specify parent/child_role in order for
2977 * option inheritance to work. Existing callers use it for the root node. */
2978 BlockDriverState
*bdrv_open_blockdev_ref(BlockdevRef
*ref
, Error
**errp
)
2980 BlockDriverState
*bs
= NULL
;
2981 Error
*local_err
= NULL
;
2982 QObject
*obj
= NULL
;
2983 QDict
*qdict
= NULL
;
2984 const char *reference
= NULL
;
2987 if (ref
->type
== QTYPE_QSTRING
) {
2988 reference
= ref
->u
.reference
;
2990 BlockdevOptions
*options
= &ref
->u
.definition
;
2991 assert(ref
->type
== QTYPE_QDICT
);
2993 v
= qobject_output_visitor_new(&obj
);
2994 visit_type_BlockdevOptions(v
, NULL
, &options
, &local_err
);
2996 error_propagate(errp
, local_err
);
2999 visit_complete(v
, &obj
);
3001 qdict
= qobject_to(QDict
, obj
);
3002 qdict_flatten(qdict
);
3004 /* bdrv_open_inherit() defaults to the values in bdrv_flags (for
3005 * compatibility with other callers) rather than what we want as the
3006 * real defaults. Apply the defaults here instead. */
3007 qdict_set_default_str(qdict
, BDRV_OPT_CACHE_DIRECT
, "off");
3008 qdict_set_default_str(qdict
, BDRV_OPT_CACHE_NO_FLUSH
, "off");
3009 qdict_set_default_str(qdict
, BDRV_OPT_READ_ONLY
, "off");
3010 qdict_set_default_str(qdict
, BDRV_OPT_AUTO_READ_ONLY
, "off");
3014 bs
= bdrv_open_inherit(NULL
, reference
, qdict
, 0, NULL
, NULL
, errp
);
3023 static BlockDriverState
*bdrv_append_temp_snapshot(BlockDriverState
*bs
,
3025 QDict
*snapshot_options
,
3028 /* TODO: extra byte is a hack to ensure MAX_PATH space on Windows. */
3029 char *tmp_filename
= g_malloc0(PATH_MAX
+ 1);
3031 QemuOpts
*opts
= NULL
;
3032 BlockDriverState
*bs_snapshot
= NULL
;
3033 Error
*local_err
= NULL
;
3036 /* if snapshot, we create a temporary backing file and open it
3037 instead of opening 'filename' directly */
3039 /* Get the required size from the image */
3040 total_size
= bdrv_getlength(bs
);
3041 if (total_size
< 0) {
3042 error_setg_errno(errp
, -total_size
, "Could not get image size");
3046 /* Create the temporary image */
3047 ret
= get_tmp_filename(tmp_filename
, PATH_MAX
+ 1);
3049 error_setg_errno(errp
, -ret
, "Could not get temporary filename");
3053 opts
= qemu_opts_create(bdrv_qcow2
.create_opts
, NULL
, 0,
3055 qemu_opt_set_number(opts
, BLOCK_OPT_SIZE
, total_size
, &error_abort
);
3056 ret
= bdrv_create(&bdrv_qcow2
, tmp_filename
, opts
, errp
);
3057 qemu_opts_del(opts
);
3059 error_prepend(errp
, "Could not create temporary overlay '%s': ",
3064 /* Prepare options QDict for the temporary file */
3065 qdict_put_str(snapshot_options
, "file.driver", "file");
3066 qdict_put_str(snapshot_options
, "file.filename", tmp_filename
);
3067 qdict_put_str(snapshot_options
, "driver", "qcow2");
3069 bs_snapshot
= bdrv_open(NULL
, NULL
, snapshot_options
, flags
, errp
);
3070 snapshot_options
= NULL
;
3075 /* bdrv_append() consumes a strong reference to bs_snapshot
3076 * (i.e. it will call bdrv_unref() on it) even on error, so in
3077 * order to be able to return one, we have to increase
3078 * bs_snapshot's refcount here */
3079 bdrv_ref(bs_snapshot
);
3080 bdrv_append(bs_snapshot
, bs
, &local_err
);
3082 error_propagate(errp
, local_err
);
3088 qobject_unref(snapshot_options
);
3089 g_free(tmp_filename
);
3094 * Opens a disk image (raw, qcow2, vmdk, ...)
3096 * options is a QDict of options to pass to the block drivers, or NULL for an
3097 * empty set of options. The reference to the QDict belongs to the block layer
3098 * after the call (even on failure), so if the caller intends to reuse the
3099 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
3101 * If *pbs is NULL, a new BDS will be created with a pointer to it stored there.
3102 * If it is not NULL, the referenced BDS will be reused.
3104 * The reference parameter may be used to specify an existing block device which
3105 * should be opened. If specified, neither options nor a filename may be given,
3106 * nor can an existing BDS be reused (that is, *pbs has to be NULL).
3108 static BlockDriverState
*bdrv_open_inherit(const char *filename
,
3109 const char *reference
,
3110 QDict
*options
, int flags
,
3111 BlockDriverState
*parent
,
3112 const BdrvChildRole
*child_role
,
3116 BlockBackend
*file
= NULL
;
3117 BlockDriverState
*bs
;
3118 BlockDriver
*drv
= NULL
;
3120 const char *drvname
;
3121 const char *backing
;
3122 Error
*local_err
= NULL
;
3123 QDict
*snapshot_options
= NULL
;
3124 int snapshot_flags
= 0;
3126 assert(!child_role
|| !flags
);
3127 assert(!child_role
== !parent
);
3130 bool options_non_empty
= options
? qdict_size(options
) : false;
3131 qobject_unref(options
);
3133 if (filename
|| options_non_empty
) {
3134 error_setg(errp
, "Cannot reference an existing block device with "
3135 "additional options or a new filename");
3139 bs
= bdrv_lookup_bs(reference
, reference
, errp
);
3150 /* NULL means an empty set of options */
3151 if (options
== NULL
) {
3152 options
= qdict_new();
3155 /* json: syntax counts as explicit options, as if in the QDict */
3156 parse_json_protocol(options
, &filename
, &local_err
);
3161 bs
->explicit_options
= qdict_clone_shallow(options
);
3164 bs
->inherits_from
= parent
;
3165 child_role
->inherit_options(&flags
, options
,
3166 parent
->open_flags
, parent
->options
);
3169 ret
= bdrv_fill_options(&options
, filename
, &flags
, &local_err
);
3175 * Set the BDRV_O_RDWR and BDRV_O_ALLOW_RDWR flags.
3176 * Caution: getting a boolean member of @options requires care.
3177 * When @options come from -blockdev or blockdev_add, members are
3178 * typed according to the QAPI schema, but when they come from
3179 * -drive, they're all QString.
3181 if (g_strcmp0(qdict_get_try_str(options
, BDRV_OPT_READ_ONLY
), "on") &&
3182 !qdict_get_try_bool(options
, BDRV_OPT_READ_ONLY
, false)) {
3183 flags
|= (BDRV_O_RDWR
| BDRV_O_ALLOW_RDWR
);
3185 flags
&= ~BDRV_O_RDWR
;
3188 if (flags
& BDRV_O_SNAPSHOT
) {
3189 snapshot_options
= qdict_new();
3190 bdrv_temp_snapshot_options(&snapshot_flags
, snapshot_options
,
3192 /* Let bdrv_backing_options() override "read-only" */
3193 qdict_del(options
, BDRV_OPT_READ_ONLY
);
3194 bdrv_backing_options(&flags
, options
, flags
, options
);
3197 bs
->open_flags
= flags
;
3198 bs
->options
= options
;
3199 options
= qdict_clone_shallow(options
);
3201 /* Find the right image format driver */
3202 /* See cautionary note on accessing @options above */
3203 drvname
= qdict_get_try_str(options
, "driver");
3205 drv
= bdrv_find_format(drvname
);
3207 error_setg(errp
, "Unknown driver: '%s'", drvname
);
3212 assert(drvname
|| !(flags
& BDRV_O_PROTOCOL
));
3214 /* See cautionary note on accessing @options above */
3215 backing
= qdict_get_try_str(options
, "backing");
3216 if (qobject_to(QNull
, qdict_get(options
, "backing")) != NULL
||
3217 (backing
&& *backing
== '\0'))
3220 warn_report("Use of \"backing\": \"\" is deprecated; "
3221 "use \"backing\": null instead");
3223 flags
|= BDRV_O_NO_BACKING
;
3224 qdict_del(bs
->explicit_options
, "backing");
3225 qdict_del(bs
->options
, "backing");
3226 qdict_del(options
, "backing");
3229 /* Open image file without format layer. This BlockBackend is only used for
3230 * probing, the block drivers will do their own bdrv_open_child() for the
3231 * same BDS, which is why we put the node name back into options. */
3232 if ((flags
& BDRV_O_PROTOCOL
) == 0) {
3233 BlockDriverState
*file_bs
;
3235 file_bs
= bdrv_open_child_bs(filename
, options
, "file", bs
,
3236 &child_file
, true, &local_err
);
3240 if (file_bs
!= NULL
) {
3241 /* Not requesting BLK_PERM_CONSISTENT_READ because we're only
3242 * looking at the header to guess the image format. This works even
3243 * in cases where a guest would not see a consistent state. */
3244 file
= blk_new(bdrv_get_aio_context(file_bs
), 0, BLK_PERM_ALL
);
3245 blk_insert_bs(file
, file_bs
, &local_err
);
3246 bdrv_unref(file_bs
);
3251 qdict_put_str(options
, "file", bdrv_get_node_name(file_bs
));
3255 /* Image format probing */
3258 ret
= find_image_format(file
, filename
, &drv
, &local_err
);
3263 * This option update would logically belong in bdrv_fill_options(),
3264 * but we first need to open bs->file for the probing to work, while
3265 * opening bs->file already requires the (mostly) final set of options
3266 * so that cache mode etc. can be inherited.
3268 * Adding the driver later is somewhat ugly, but it's not an option
3269 * that would ever be inherited, so it's correct. We just need to make
3270 * sure to update both bs->options (which has the full effective
3271 * options for bs) and options (which has file.* already removed).
3273 qdict_put_str(bs
->options
, "driver", drv
->format_name
);
3274 qdict_put_str(options
, "driver", drv
->format_name
);
3276 error_setg(errp
, "Must specify either driver or file");
3280 /* BDRV_O_PROTOCOL must be set iff a protocol BDS is about to be created */
3281 assert(!!(flags
& BDRV_O_PROTOCOL
) == !!drv
->bdrv_file_open
);
3282 /* file must be NULL if a protocol BDS is about to be created
3283 * (the inverse results in an error message from bdrv_open_common()) */
3284 assert(!(flags
& BDRV_O_PROTOCOL
) || !file
);
3286 /* Open the image */
3287 ret
= bdrv_open_common(bs
, file
, options
, &local_err
);
3297 /* If there is a backing file, use it */
3298 if ((flags
& BDRV_O_NO_BACKING
) == 0) {
3299 ret
= bdrv_open_backing_file(bs
, options
, "backing", &local_err
);
3301 goto close_and_fail
;
3305 /* Remove all children options and references
3306 * from bs->options and bs->explicit_options */
3307 QLIST_FOREACH(child
, &bs
->children
, next
) {
3308 char *child_key_dot
;
3309 child_key_dot
= g_strdup_printf("%s.", child
->name
);
3310 qdict_extract_subqdict(bs
->explicit_options
, NULL
, child_key_dot
);
3311 qdict_extract_subqdict(bs
->options
, NULL
, child_key_dot
);
3312 qdict_del(bs
->explicit_options
, child
->name
);
3313 qdict_del(bs
->options
, child
->name
);
3314 g_free(child_key_dot
);
3317 /* Check if any unknown options were used */
3318 if (qdict_size(options
) != 0) {
3319 const QDictEntry
*entry
= qdict_first(options
);
3320 if (flags
& BDRV_O_PROTOCOL
) {
3321 error_setg(errp
, "Block protocol '%s' doesn't support the option "
3322 "'%s'", drv
->format_name
, entry
->key
);
3325 "Block format '%s' does not support the option '%s'",
3326 drv
->format_name
, entry
->key
);
3329 goto close_and_fail
;
3332 bdrv_parent_cb_change_media(bs
, true);
3334 qobject_unref(options
);
3337 /* For snapshot=on, create a temporary qcow2 overlay. bs points to the
3338 * temporary snapshot afterwards. */
3339 if (snapshot_flags
) {
3340 BlockDriverState
*snapshot_bs
;
3341 snapshot_bs
= bdrv_append_temp_snapshot(bs
, snapshot_flags
,
3342 snapshot_options
, &local_err
);
3343 snapshot_options
= NULL
;
3345 goto close_and_fail
;
3347 /* We are not going to return bs but the overlay on top of it
3348 * (snapshot_bs); thus, we have to drop the strong reference to bs
3349 * (which we obtained by calling bdrv_new()). bs will not be deleted,
3350 * though, because the overlay still has a reference to it. */
3359 qobject_unref(snapshot_options
);
3360 qobject_unref(bs
->explicit_options
);
3361 qobject_unref(bs
->options
);
3362 qobject_unref(options
);
3364 bs
->explicit_options
= NULL
;
3366 error_propagate(errp
, local_err
);
3371 qobject_unref(snapshot_options
);
3372 qobject_unref(options
);
3373 error_propagate(errp
, local_err
);
3377 BlockDriverState
*bdrv_open(const char *filename
, const char *reference
,
3378 QDict
*options
, int flags
, Error
**errp
)
3380 return bdrv_open_inherit(filename
, reference
, options
, flags
, NULL
,
3384 /* Return true if the NULL-terminated @list contains @str */
3385 static bool is_str_in_list(const char *str
, const char *const *list
)
3389 for (i
= 0; list
[i
] != NULL
; i
++) {
3390 if (!strcmp(str
, list
[i
])) {
3399 * Check that every option set in @bs->options is also set in
3402 * Options listed in the common_options list and in
3403 * @bs->drv->mutable_opts are skipped.
3405 * Return 0 on success, otherwise return -EINVAL and set @errp.
3407 static int bdrv_reset_options_allowed(BlockDriverState
*bs
,
3408 const QDict
*new_opts
, Error
**errp
)
3410 const QDictEntry
*e
;
3411 /* These options are common to all block drivers and are handled
3412 * in bdrv_reopen_prepare() so they can be left out of @new_opts */
3413 const char *const common_options
[] = {
3414 "node-name", "discard", "cache.direct", "cache.no-flush",
3415 "read-only", "auto-read-only", "detect-zeroes", NULL
3418 for (e
= qdict_first(bs
->options
); e
; e
= qdict_next(bs
->options
, e
)) {
3419 if (!qdict_haskey(new_opts
, e
->key
) &&
3420 !is_str_in_list(e
->key
, common_options
) &&
3421 !is_str_in_list(e
->key
, bs
->drv
->mutable_opts
)) {
3422 error_setg(errp
, "Option '%s' cannot be reset "
3423 "to its default value", e
->key
);
3432 * Returns true if @child can be reached recursively from @bs
3434 static bool bdrv_recurse_has_child(BlockDriverState
*bs
,
3435 BlockDriverState
*child
)
3443 QLIST_FOREACH(c
, &bs
->children
, next
) {
3444 if (bdrv_recurse_has_child(c
->bs
, child
)) {
3453 * Adds a BlockDriverState to a simple queue for an atomic, transactional
3454 * reopen of multiple devices.
3456 * bs_queue can either be an existing BlockReopenQueue that has had QTAILQ_INIT
3457 * already performed, or alternatively may be NULL a new BlockReopenQueue will
3458 * be created and initialized. This newly created BlockReopenQueue should be
3459 * passed back in for subsequent calls that are intended to be of the same
3462 * bs is the BlockDriverState to add to the reopen queue.
3464 * options contains the changed options for the associated bs
3465 * (the BlockReopenQueue takes ownership)
3467 * flags contains the open flags for the associated bs
3469 * returns a pointer to bs_queue, which is either the newly allocated
3470 * bs_queue, or the existing bs_queue being used.
3472 * bs must be drained between bdrv_reopen_queue() and bdrv_reopen_multiple().
3474 static BlockReopenQueue
*bdrv_reopen_queue_child(BlockReopenQueue
*bs_queue
,
3475 BlockDriverState
*bs
,
3477 const BdrvChildRole
*role
,
3478 QDict
*parent_options
,
3484 BlockReopenQueueEntry
*bs_entry
;
3486 QDict
*old_options
, *explicit_options
, *options_copy
;
3490 /* Make sure that the caller remembered to use a drained section. This is
3491 * important to avoid graph changes between the recursive queuing here and
3492 * bdrv_reopen_multiple(). */
3493 assert(bs
->quiesce_counter
> 0);
3495 if (bs_queue
== NULL
) {
3496 bs_queue
= g_new0(BlockReopenQueue
, 1);
3497 QTAILQ_INIT(bs_queue
);
3501 options
= qdict_new();
3504 /* Check if this BlockDriverState is already in the queue */
3505 QTAILQ_FOREACH(bs_entry
, bs_queue
, entry
) {
3506 if (bs
== bs_entry
->state
.bs
) {
3512 * Precedence of options:
3513 * 1. Explicitly passed in options (highest)
3514 * 2. Retained from explicitly set options of bs
3515 * 3. Inherited from parent node
3516 * 4. Retained from effective options of bs
3519 /* Old explicitly set values (don't overwrite by inherited value) */
3520 if (bs_entry
|| keep_old_opts
) {
3521 old_options
= qdict_clone_shallow(bs_entry
?
3522 bs_entry
->state
.explicit_options
:
3523 bs
->explicit_options
);
3524 bdrv_join_options(bs
, options
, old_options
);
3525 qobject_unref(old_options
);
3528 explicit_options
= qdict_clone_shallow(options
);
3530 /* Inherit from parent node */
3531 if (parent_options
) {
3533 role
->inherit_options(&flags
, options
, parent_flags
, parent_options
);
3535 flags
= bdrv_get_flags(bs
);
3538 if (keep_old_opts
) {
3539 /* Old values are used for options that aren't set yet */
3540 old_options
= qdict_clone_shallow(bs
->options
);
3541 bdrv_join_options(bs
, options
, old_options
);
3542 qobject_unref(old_options
);
3545 /* We have the final set of options so let's update the flags */
3546 options_copy
= qdict_clone_shallow(options
);
3547 opts
= qemu_opts_create(&bdrv_runtime_opts
, NULL
, 0, &error_abort
);
3548 qemu_opts_absorb_qdict(opts
, options_copy
, NULL
);
3549 update_flags_from_options(&flags
, opts
);
3550 qemu_opts_del(opts
);
3551 qobject_unref(options_copy
);
3553 /* bdrv_open_inherit() sets and clears some additional flags internally */
3554 flags
&= ~BDRV_O_PROTOCOL
;
3555 if (flags
& BDRV_O_RDWR
) {
3556 flags
|= BDRV_O_ALLOW_RDWR
;
3560 bs_entry
= g_new0(BlockReopenQueueEntry
, 1);
3561 QTAILQ_INSERT_TAIL(bs_queue
, bs_entry
, entry
);
3563 qobject_unref(bs_entry
->state
.options
);
3564 qobject_unref(bs_entry
->state
.explicit_options
);
3567 bs_entry
->state
.bs
= bs
;
3568 bs_entry
->state
.options
= options
;
3569 bs_entry
->state
.explicit_options
= explicit_options
;
3570 bs_entry
->state
.flags
= flags
;
3572 /* This needs to be overwritten in bdrv_reopen_prepare() */
3573 bs_entry
->state
.perm
= UINT64_MAX
;
3574 bs_entry
->state
.shared_perm
= 0;
3577 * If keep_old_opts is false then it means that unspecified
3578 * options must be reset to their original value. We don't allow
3579 * resetting 'backing' but we need to know if the option is
3580 * missing in order to decide if we have to return an error.
3582 if (!keep_old_opts
) {
3583 bs_entry
->state
.backing_missing
=
3584 !qdict_haskey(options
, "backing") &&
3585 !qdict_haskey(options
, "backing.driver");
3588 QLIST_FOREACH(child
, &bs
->children
, next
) {
3589 QDict
*new_child_options
= NULL
;
3590 bool child_keep_old
= keep_old_opts
;
3592 /* reopen can only change the options of block devices that were
3593 * implicitly created and inherited options. For other (referenced)
3594 * block devices, a syntax like "backing.foo" results in an error. */
3595 if (child
->bs
->inherits_from
!= bs
) {
3599 /* Check if the options contain a child reference */
3600 if (qdict_haskey(options
, child
->name
)) {
3601 const char *childref
= qdict_get_try_str(options
, child
->name
);
3603 * The current child must not be reopened if the child
3604 * reference is null or points to a different node.
3606 if (g_strcmp0(childref
, child
->bs
->node_name
)) {
3610 * If the child reference points to the current child then
3611 * reopen it with its existing set of options (note that
3612 * it can still inherit new options from the parent).
3614 child_keep_old
= true;
3616 /* Extract child options ("child-name.*") */
3617 char *child_key_dot
= g_strdup_printf("%s.", child
->name
);
3618 qdict_extract_subqdict(explicit_options
, NULL
, child_key_dot
);
3619 qdict_extract_subqdict(options
, &new_child_options
, child_key_dot
);
3620 g_free(child_key_dot
);
3623 bdrv_reopen_queue_child(bs_queue
, child
->bs
, new_child_options
,
3624 child
->role
, options
, flags
, child_keep_old
);
3630 BlockReopenQueue
*bdrv_reopen_queue(BlockReopenQueue
*bs_queue
,
3631 BlockDriverState
*bs
,
3632 QDict
*options
, bool keep_old_opts
)
3634 return bdrv_reopen_queue_child(bs_queue
, bs
, options
, NULL
, NULL
, 0,
3639 * Reopen multiple BlockDriverStates atomically & transactionally.
3641 * The queue passed in (bs_queue) must have been built up previous
3642 * via bdrv_reopen_queue().
3644 * Reopens all BDS specified in the queue, with the appropriate
3645 * flags. All devices are prepared for reopen, and failure of any
3646 * device will cause all device changes to be abandoned, and intermediate
3649 * If all devices prepare successfully, then the changes are committed
3652 * All affected nodes must be drained between bdrv_reopen_queue() and
3653 * bdrv_reopen_multiple().
3655 int bdrv_reopen_multiple(BlockReopenQueue
*bs_queue
, Error
**errp
)
3658 BlockReopenQueueEntry
*bs_entry
, *next
;
3660 assert(bs_queue
!= NULL
);
3662 QTAILQ_FOREACH(bs_entry
, bs_queue
, entry
) {
3663 assert(bs_entry
->state
.bs
->quiesce_counter
> 0);
3664 if (bdrv_reopen_prepare(&bs_entry
->state
, bs_queue
, errp
)) {
3667 bs_entry
->prepared
= true;
3670 QTAILQ_FOREACH(bs_entry
, bs_queue
, entry
) {
3671 BDRVReopenState
*state
= &bs_entry
->state
;
3672 ret
= bdrv_check_perm(state
->bs
, bs_queue
, state
->perm
,
3673 state
->shared_perm
, NULL
, NULL
, errp
);
3677 /* Check if new_backing_bs would accept the new permissions */
3678 if (state
->replace_backing_bs
&& state
->new_backing_bs
) {
3679 uint64_t nperm
, nshared
;
3680 bdrv_child_perm(state
->bs
, state
->new_backing_bs
,
3681 NULL
, &child_backing
, bs_queue
,
3682 state
->perm
, state
->shared_perm
,
3684 ret
= bdrv_check_update_perm(state
->new_backing_bs
, NULL
,
3685 nperm
, nshared
, NULL
, NULL
, errp
);
3690 bs_entry
->perms_checked
= true;
3694 * If we reach this point, we have success and just need to apply the
3697 * Reverse order is used to comfort qcow2 driver: on commit it need to write
3698 * IN_USE flag to the image, to mark bitmaps in the image as invalid. But
3699 * children are usually goes after parents in reopen-queue, so go from last
3702 QTAILQ_FOREACH_REVERSE(bs_entry
, bs_queue
, entry
) {
3703 bdrv_reopen_commit(&bs_entry
->state
);
3708 QTAILQ_FOREACH_SAFE(bs_entry
, bs_queue
, entry
, next
) {
3709 BDRVReopenState
*state
= &bs_entry
->state
;
3711 if (!bs_entry
->perms_checked
) {
3716 bdrv_set_perm(state
->bs
, state
->perm
, state
->shared_perm
);
3718 bdrv_abort_perm_update(state
->bs
);
3719 if (state
->replace_backing_bs
&& state
->new_backing_bs
) {
3720 bdrv_abort_perm_update(state
->new_backing_bs
);
3726 QTAILQ_FOREACH_REVERSE(bs_entry
, bs_queue
, entry
) {
3727 BlockDriverState
*bs
= bs_entry
->state
.bs
;
3729 if (bs
->drv
->bdrv_reopen_commit_post
)
3730 bs
->drv
->bdrv_reopen_commit_post(&bs_entry
->state
);
3734 QTAILQ_FOREACH_SAFE(bs_entry
, bs_queue
, entry
, next
) {
3736 if (bs_entry
->prepared
) {
3737 bdrv_reopen_abort(&bs_entry
->state
);
3739 qobject_unref(bs_entry
->state
.explicit_options
);
3740 qobject_unref(bs_entry
->state
.options
);
3742 if (bs_entry
->state
.new_backing_bs
) {
3743 bdrv_unref(bs_entry
->state
.new_backing_bs
);
3752 int bdrv_reopen_set_read_only(BlockDriverState
*bs
, bool read_only
,
3756 BlockReopenQueue
*queue
;
3757 QDict
*opts
= qdict_new();
3759 qdict_put_bool(opts
, BDRV_OPT_READ_ONLY
, read_only
);
3761 bdrv_subtree_drained_begin(bs
);
3762 queue
= bdrv_reopen_queue(NULL
, bs
, opts
, true);
3763 ret
= bdrv_reopen_multiple(queue
, errp
);
3764 bdrv_subtree_drained_end(bs
);
3769 static BlockReopenQueueEntry
*find_parent_in_reopen_queue(BlockReopenQueue
*q
,
3772 BlockReopenQueueEntry
*entry
;
3774 QTAILQ_FOREACH(entry
, q
, entry
) {
3775 BlockDriverState
*bs
= entry
->state
.bs
;
3778 QLIST_FOREACH(child
, &bs
->children
, next
) {
3788 static void bdrv_reopen_perm(BlockReopenQueue
*q
, BlockDriverState
*bs
,
3789 uint64_t *perm
, uint64_t *shared
)
3792 BlockReopenQueueEntry
*parent
;
3793 uint64_t cumulative_perms
= 0;
3794 uint64_t cumulative_shared_perms
= BLK_PERM_ALL
;
3796 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
3797 parent
= find_parent_in_reopen_queue(q
, c
);
3799 cumulative_perms
|= c
->perm
;
3800 cumulative_shared_perms
&= c
->shared_perm
;
3802 uint64_t nperm
, nshared
;
3804 bdrv_child_perm(parent
->state
.bs
, bs
, c
, c
->role
, q
,
3805 parent
->state
.perm
, parent
->state
.shared_perm
,
3808 cumulative_perms
|= nperm
;
3809 cumulative_shared_perms
&= nshared
;
3812 *perm
= cumulative_perms
;
3813 *shared
= cumulative_shared_perms
;
3816 static bool bdrv_reopen_can_attach(BlockDriverState
*parent
,
3818 BlockDriverState
*new_child
,
3821 AioContext
*parent_ctx
= bdrv_get_aio_context(parent
);
3822 AioContext
*child_ctx
= bdrv_get_aio_context(new_child
);
3826 ignore
= g_slist_prepend(NULL
, child
);
3827 ret
= bdrv_can_set_aio_context(new_child
, parent_ctx
, &ignore
, NULL
);
3828 g_slist_free(ignore
);
3833 ignore
= g_slist_prepend(NULL
, child
);
3834 ret
= bdrv_can_set_aio_context(parent
, child_ctx
, &ignore
, errp
);
3835 g_slist_free(ignore
);
3840 * Take a BDRVReopenState and check if the value of 'backing' in the
3841 * reopen_state->options QDict is valid or not.
3843 * If 'backing' is missing from the QDict then return 0.
3845 * If 'backing' contains the node name of the backing file of
3846 * reopen_state->bs then return 0.
3848 * If 'backing' contains a different node name (or is null) then check
3849 * whether the current backing file can be replaced with the new one.
3850 * If that's the case then reopen_state->replace_backing_bs is set to
3851 * true and reopen_state->new_backing_bs contains a pointer to the new
3852 * backing BlockDriverState (or NULL).
3854 * Return 0 on success, otherwise return < 0 and set @errp.
3856 static int bdrv_reopen_parse_backing(BDRVReopenState
*reopen_state
,
3859 BlockDriverState
*bs
= reopen_state
->bs
;
3860 BlockDriverState
*overlay_bs
, *new_backing_bs
;
3864 value
= qdict_get(reopen_state
->options
, "backing");
3865 if (value
== NULL
) {
3869 switch (qobject_type(value
)) {
3871 new_backing_bs
= NULL
;
3874 str
= qobject_get_try_str(value
);
3875 new_backing_bs
= bdrv_lookup_bs(NULL
, str
, errp
);
3876 if (new_backing_bs
== NULL
) {
3878 } else if (bdrv_recurse_has_child(new_backing_bs
, bs
)) {
3879 error_setg(errp
, "Making '%s' a backing file of '%s' "
3880 "would create a cycle", str
, bs
->node_name
);
3885 /* 'backing' does not allow any other data type */
3886 g_assert_not_reached();
3890 * Check AioContext compatibility so that the bdrv_set_backing_hd() call in
3891 * bdrv_reopen_commit() won't fail.
3893 if (new_backing_bs
) {
3894 if (!bdrv_reopen_can_attach(bs
, bs
->backing
, new_backing_bs
, errp
)) {
3900 * Find the "actual" backing file by skipping all links that point
3901 * to an implicit node, if any (e.g. a commit filter node).
3904 while (backing_bs(overlay_bs
) && backing_bs(overlay_bs
)->implicit
) {
3905 overlay_bs
= backing_bs(overlay_bs
);
3908 /* If we want to replace the backing file we need some extra checks */
3909 if (new_backing_bs
!= backing_bs(overlay_bs
)) {
3910 /* Check for implicit nodes between bs and its backing file */
3911 if (bs
!= overlay_bs
) {
3912 error_setg(errp
, "Cannot change backing link if '%s' has "
3913 "an implicit backing file", bs
->node_name
);
3916 /* Check if the backing link that we want to replace is frozen */
3917 if (bdrv_is_backing_chain_frozen(overlay_bs
, backing_bs(overlay_bs
),
3921 reopen_state
->replace_backing_bs
= true;
3922 if (new_backing_bs
) {
3923 bdrv_ref(new_backing_bs
);
3924 reopen_state
->new_backing_bs
= new_backing_bs
;
3932 * Prepares a BlockDriverState for reopen. All changes are staged in the
3933 * 'opaque' field of the BDRVReopenState, which is used and allocated by
3934 * the block driver layer .bdrv_reopen_prepare()
3936 * bs is the BlockDriverState to reopen
3937 * flags are the new open flags
3938 * queue is the reopen queue
3940 * Returns 0 on success, non-zero on error. On error errp will be set
3943 * On failure, bdrv_reopen_abort() will be called to clean up any data.
3944 * It is the responsibility of the caller to then call the abort() or
3945 * commit() for any other BDS that have been left in a prepare() state
3948 int bdrv_reopen_prepare(BDRVReopenState
*reopen_state
, BlockReopenQueue
*queue
,
3953 Error
*local_err
= NULL
;
3956 QDict
*orig_reopen_opts
;
3957 char *discard
= NULL
;
3959 bool drv_prepared
= false;
3961 assert(reopen_state
!= NULL
);
3962 assert(reopen_state
->bs
->drv
!= NULL
);
3963 drv
= reopen_state
->bs
->drv
;
3965 /* This function and each driver's bdrv_reopen_prepare() remove
3966 * entries from reopen_state->options as they are processed, so
3967 * we need to make a copy of the original QDict. */
3968 orig_reopen_opts
= qdict_clone_shallow(reopen_state
->options
);
3970 /* Process generic block layer options */
3971 opts
= qemu_opts_create(&bdrv_runtime_opts
, NULL
, 0, &error_abort
);
3972 qemu_opts_absorb_qdict(opts
, reopen_state
->options
, &local_err
);
3974 error_propagate(errp
, local_err
);
3979 /* This was already called in bdrv_reopen_queue_child() so the flags
3980 * are up-to-date. This time we simply want to remove the options from
3981 * QemuOpts in order to indicate that they have been processed. */
3982 old_flags
= reopen_state
->flags
;
3983 update_flags_from_options(&reopen_state
->flags
, opts
);
3984 assert(old_flags
== reopen_state
->flags
);
3986 discard
= qemu_opt_get_del(opts
, BDRV_OPT_DISCARD
);
3987 if (discard
!= NULL
) {
3988 if (bdrv_parse_discard_flags(discard
, &reopen_state
->flags
) != 0) {
3989 error_setg(errp
, "Invalid discard option");
3995 reopen_state
->detect_zeroes
=
3996 bdrv_parse_detect_zeroes(opts
, reopen_state
->flags
, &local_err
);
3998 error_propagate(errp
, local_err
);
4003 /* All other options (including node-name and driver) must be unchanged.
4004 * Put them back into the QDict, so that they are checked at the end
4005 * of this function. */
4006 qemu_opts_to_qdict(opts
, reopen_state
->options
);
4008 /* If we are to stay read-only, do not allow permission change
4009 * to r/w. Attempting to set to r/w may fail if either BDRV_O_ALLOW_RDWR is
4010 * not set, or if the BDS still has copy_on_read enabled */
4011 read_only
= !(reopen_state
->flags
& BDRV_O_RDWR
);
4012 ret
= bdrv_can_set_read_only(reopen_state
->bs
, read_only
, true, &local_err
);
4014 error_propagate(errp
, local_err
);
4018 /* Calculate required permissions after reopening */
4019 bdrv_reopen_perm(queue
, reopen_state
->bs
,
4020 &reopen_state
->perm
, &reopen_state
->shared_perm
);
4022 ret
= bdrv_flush(reopen_state
->bs
);
4024 error_setg_errno(errp
, -ret
, "Error flushing drive");
4028 if (drv
->bdrv_reopen_prepare
) {
4030 * If a driver-specific option is missing, it means that we
4031 * should reset it to its default value.
4032 * But not all options allow that, so we need to check it first.
4034 ret
= bdrv_reset_options_allowed(reopen_state
->bs
,
4035 reopen_state
->options
, errp
);
4040 ret
= drv
->bdrv_reopen_prepare(reopen_state
, queue
, &local_err
);
4042 if (local_err
!= NULL
) {
4043 error_propagate(errp
, local_err
);
4045 bdrv_refresh_filename(reopen_state
->bs
);
4046 error_setg(errp
, "failed while preparing to reopen image '%s'",
4047 reopen_state
->bs
->filename
);
4052 /* It is currently mandatory to have a bdrv_reopen_prepare()
4053 * handler for each supported drv. */
4054 error_setg(errp
, "Block format '%s' used by node '%s' "
4055 "does not support reopening files", drv
->format_name
,
4056 bdrv_get_device_or_node_name(reopen_state
->bs
));
4061 drv_prepared
= true;
4064 * We must provide the 'backing' option if the BDS has a backing
4065 * file or if the image file has a backing file name as part of
4066 * its metadata. Otherwise the 'backing' option can be omitted.
4068 if (drv
->supports_backing
&& reopen_state
->backing_missing
&&
4069 (backing_bs(reopen_state
->bs
) || reopen_state
->bs
->backing_file
[0])) {
4070 error_setg(errp
, "backing is missing for '%s'",
4071 reopen_state
->bs
->node_name
);
4077 * Allow changing the 'backing' option. The new value can be
4078 * either a reference to an existing node (using its node name)
4079 * or NULL to simply detach the current backing file.
4081 ret
= bdrv_reopen_parse_backing(reopen_state
, errp
);
4085 qdict_del(reopen_state
->options
, "backing");
4087 /* Options that are not handled are only okay if they are unchanged
4088 * compared to the old state. It is expected that some options are only
4089 * used for the initial open, but not reopen (e.g. filename) */
4090 if (qdict_size(reopen_state
->options
)) {
4091 const QDictEntry
*entry
= qdict_first(reopen_state
->options
);
4094 QObject
*new = entry
->value
;
4095 QObject
*old
= qdict_get(reopen_state
->bs
->options
, entry
->key
);
4097 /* Allow child references (child_name=node_name) as long as they
4098 * point to the current child (i.e. everything stays the same). */
4099 if (qobject_type(new) == QTYPE_QSTRING
) {
4101 QLIST_FOREACH(child
, &reopen_state
->bs
->children
, next
) {
4102 if (!strcmp(child
->name
, entry
->key
)) {
4108 const char *str
= qobject_get_try_str(new);
4109 if (!strcmp(child
->bs
->node_name
, str
)) {
4110 continue; /* Found child with this name, skip option */
4116 * TODO: When using -drive to specify blockdev options, all values
4117 * will be strings; however, when using -blockdev, blockdev-add or
4118 * filenames using the json:{} pseudo-protocol, they will be
4120 * In contrast, reopening options are (currently) always strings
4121 * (because you can only specify them through qemu-io; all other
4122 * callers do not specify any options).
4123 * Therefore, when using anything other than -drive to create a BDS,
4124 * this cannot detect non-string options as unchanged, because
4125 * qobject_is_equal() always returns false for objects of different
4126 * type. In the future, this should be remedied by correctly typing
4127 * all options. For now, this is not too big of an issue because
4128 * the user can simply omit options which cannot be changed anyway,
4129 * so they will stay unchanged.
4131 if (!qobject_is_equal(new, old
)) {
4132 error_setg(errp
, "Cannot change the option '%s'", entry
->key
);
4136 } while ((entry
= qdict_next(reopen_state
->options
, entry
)));
4141 /* Restore the original reopen_state->options QDict */
4142 qobject_unref(reopen_state
->options
);
4143 reopen_state
->options
= qobject_ref(orig_reopen_opts
);
4146 if (ret
< 0 && drv_prepared
) {
4147 /* drv->bdrv_reopen_prepare() has succeeded, so we need to
4148 * call drv->bdrv_reopen_abort() before signaling an error
4149 * (bdrv_reopen_multiple() will not call bdrv_reopen_abort()
4150 * when the respective bdrv_reopen_prepare() has failed) */
4151 if (drv
->bdrv_reopen_abort
) {
4152 drv
->bdrv_reopen_abort(reopen_state
);
4155 qemu_opts_del(opts
);
4156 qobject_unref(orig_reopen_opts
);
4162 * Takes the staged changes for the reopen from bdrv_reopen_prepare(), and
4163 * makes them final by swapping the staging BlockDriverState contents into
4164 * the active BlockDriverState contents.
4166 void bdrv_reopen_commit(BDRVReopenState
*reopen_state
)
4169 BlockDriverState
*bs
;
4172 assert(reopen_state
!= NULL
);
4173 bs
= reopen_state
->bs
;
4175 assert(drv
!= NULL
);
4177 /* If there are any driver level actions to take */
4178 if (drv
->bdrv_reopen_commit
) {
4179 drv
->bdrv_reopen_commit(reopen_state
);
4182 /* set BDS specific flags now */
4183 qobject_unref(bs
->explicit_options
);
4184 qobject_unref(bs
->options
);
4186 bs
->explicit_options
= reopen_state
->explicit_options
;
4187 bs
->options
= reopen_state
->options
;
4188 bs
->open_flags
= reopen_state
->flags
;
4189 bs
->read_only
= !(reopen_state
->flags
& BDRV_O_RDWR
);
4190 bs
->detect_zeroes
= reopen_state
->detect_zeroes
;
4192 if (reopen_state
->replace_backing_bs
) {
4193 qdict_del(bs
->explicit_options
, "backing");
4194 qdict_del(bs
->options
, "backing");
4197 /* Remove child references from bs->options and bs->explicit_options.
4198 * Child options were already removed in bdrv_reopen_queue_child() */
4199 QLIST_FOREACH(child
, &bs
->children
, next
) {
4200 qdict_del(bs
->explicit_options
, child
->name
);
4201 qdict_del(bs
->options
, child
->name
);
4205 * Change the backing file if a new one was specified. We do this
4206 * after updating bs->options, so bdrv_refresh_filename() (called
4207 * from bdrv_set_backing_hd()) has the new values.
4209 if (reopen_state
->replace_backing_bs
) {
4210 BlockDriverState
*old_backing_bs
= backing_bs(bs
);
4211 assert(!old_backing_bs
|| !old_backing_bs
->implicit
);
4212 /* Abort the permission update on the backing bs we're detaching */
4213 if (old_backing_bs
) {
4214 bdrv_abort_perm_update(old_backing_bs
);
4216 bdrv_set_backing_hd(bs
, reopen_state
->new_backing_bs
, &error_abort
);
4219 bdrv_refresh_limits(bs
, NULL
);
4223 * Abort the reopen, and delete and free the staged changes in
4226 void bdrv_reopen_abort(BDRVReopenState
*reopen_state
)
4230 assert(reopen_state
!= NULL
);
4231 drv
= reopen_state
->bs
->drv
;
4232 assert(drv
!= NULL
);
4234 if (drv
->bdrv_reopen_abort
) {
4235 drv
->bdrv_reopen_abort(reopen_state
);
4240 static void bdrv_close(BlockDriverState
*bs
)
4242 BdrvAioNotifier
*ban
, *ban_next
;
4243 BdrvChild
*child
, *next
;
4245 assert(!bs
->refcnt
);
4247 bdrv_drained_begin(bs
); /* complete I/O */
4249 bdrv_drain(bs
); /* in case flush left pending I/O */
4252 if (bs
->drv
->bdrv_close
) {
4253 bs
->drv
->bdrv_close(bs
);
4258 QLIST_FOREACH_SAFE(child
, &bs
->children
, next
, next
) {
4259 bdrv_unref_child(bs
, child
);
4266 atomic_set(&bs
->copy_on_read
, 0);
4267 bs
->backing_file
[0] = '\0';
4268 bs
->backing_format
[0] = '\0';
4269 bs
->total_sectors
= 0;
4270 bs
->encrypted
= false;
4272 qobject_unref(bs
->options
);
4273 qobject_unref(bs
->explicit_options
);
4275 bs
->explicit_options
= NULL
;
4276 qobject_unref(bs
->full_open_options
);
4277 bs
->full_open_options
= NULL
;
4279 bdrv_release_named_dirty_bitmaps(bs
);
4280 assert(QLIST_EMPTY(&bs
->dirty_bitmaps
));
4282 QLIST_FOREACH_SAFE(ban
, &bs
->aio_notifiers
, list
, ban_next
) {
4285 QLIST_INIT(&bs
->aio_notifiers
);
4286 bdrv_drained_end(bs
);
4289 void bdrv_close_all(void)
4291 assert(job_next(NULL
) == NULL
);
4292 nbd_export_close_all();
4294 /* Drop references from requests still in flight, such as canceled block
4295 * jobs whose AIO context has not been polled yet */
4298 blk_remove_all_bs();
4299 blockdev_close_all_bdrv_states();
4301 assert(QTAILQ_EMPTY(&all_bdrv_states
));
4304 static bool should_update_child(BdrvChild
*c
, BlockDriverState
*to
)
4310 if (c
->role
->stay_at_node
) {
4314 /* If the child @c belongs to the BDS @to, replacing the current
4315 * c->bs by @to would mean to create a loop.
4317 * Such a case occurs when appending a BDS to a backing chain.
4318 * For instance, imagine the following chain:
4320 * guest device -> node A -> further backing chain...
4322 * Now we create a new BDS B which we want to put on top of this
4323 * chain, so we first attach A as its backing node:
4328 * guest device -> node A -> further backing chain...
4330 * Finally we want to replace A by B. When doing that, we want to
4331 * replace all pointers to A by pointers to B -- except for the
4332 * pointer from B because (1) that would create a loop, and (2)
4333 * that pointer should simply stay intact:
4335 * guest device -> node B
4338 * node A -> further backing chain...
4340 * In general, when replacing a node A (c->bs) by a node B (@to),
4341 * if A is a child of B, that means we cannot replace A by B there
4342 * because that would create a loop. Silently detaching A from B
4343 * is also not really an option. So overall just leaving A in
4344 * place there is the most sensible choice.
4346 * We would also create a loop in any cases where @c is only
4347 * indirectly referenced by @to. Prevent this by returning false
4348 * if @c is found (by breadth-first search) anywhere in the whole
4353 found
= g_hash_table_new(NULL
, NULL
);
4354 g_hash_table_add(found
, to
);
4355 queue
= g_queue_new();
4356 g_queue_push_tail(queue
, to
);
4358 while (!g_queue_is_empty(queue
)) {
4359 BlockDriverState
*v
= g_queue_pop_head(queue
);
4362 QLIST_FOREACH(c2
, &v
->children
, next
) {
4368 if (g_hash_table_contains(found
, c2
->bs
)) {
4372 g_queue_push_tail(queue
, c2
->bs
);
4373 g_hash_table_add(found
, c2
->bs
);
4377 g_queue_free(queue
);
4378 g_hash_table_destroy(found
);
4383 void bdrv_replace_node(BlockDriverState
*from
, BlockDriverState
*to
,
4386 BdrvChild
*c
, *next
;
4387 GSList
*list
= NULL
, *p
;
4388 uint64_t perm
= 0, shared
= BLK_PERM_ALL
;
4391 /* Make sure that @from doesn't go away until we have successfully attached
4392 * all of its parents to @to. */
4395 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
4396 assert(bdrv_get_aio_context(from
) == bdrv_get_aio_context(to
));
4397 bdrv_drained_begin(from
);
4399 /* Put all parents into @list and calculate their cumulative permissions */
4400 QLIST_FOREACH_SAFE(c
, &from
->parents
, next_parent
, next
) {
4401 assert(c
->bs
== from
);
4402 if (!should_update_child(c
, to
)) {
4406 error_setg(errp
, "Cannot change '%s' link to '%s'",
4407 c
->name
, from
->node_name
);
4410 list
= g_slist_prepend(list
, c
);
4412 shared
&= c
->shared_perm
;
4415 /* Check whether the required permissions can be granted on @to, ignoring
4416 * all BdrvChild in @list so that they can't block themselves. */
4417 ret
= bdrv_check_update_perm(to
, NULL
, perm
, shared
, list
, NULL
, errp
);
4419 bdrv_abort_perm_update(to
);
4423 /* Now actually perform the change. We performed the permission check for
4424 * all elements of @list at once, so set the permissions all at once at the
4426 for (p
= list
; p
!= NULL
; p
= p
->next
) {
4430 bdrv_replace_child_noperm(c
, to
);
4434 bdrv_get_cumulative_perm(to
, &perm
, &shared
);
4435 bdrv_set_perm(to
, perm
, shared
);
4439 bdrv_drained_end(from
);
4444 * Add new bs contents at the top of an image chain while the chain is
4445 * live, while keeping required fields on the top layer.
4447 * This will modify the BlockDriverState fields, and swap contents
4448 * between bs_new and bs_top. Both bs_new and bs_top are modified.
4450 * bs_new must not be attached to a BlockBackend.
4452 * This function does not create any image files.
4454 * bdrv_append() takes ownership of a bs_new reference and unrefs it because
4455 * that's what the callers commonly need. bs_new will be referenced by the old
4456 * parents of bs_top after bdrv_append() returns. If the caller needs to keep a
4457 * reference of its own, it must call bdrv_ref().
4459 void bdrv_append(BlockDriverState
*bs_new
, BlockDriverState
*bs_top
,
4462 Error
*local_err
= NULL
;
4464 bdrv_set_backing_hd(bs_new
, bs_top
, &local_err
);
4466 error_propagate(errp
, local_err
);
4470 bdrv_replace_node(bs_top
, bs_new
, &local_err
);
4472 error_propagate(errp
, local_err
);
4473 bdrv_set_backing_hd(bs_new
, NULL
, &error_abort
);
4477 /* bs_new is now referenced by its new parents, we don't need the
4478 * additional reference any more. */
4483 static void bdrv_delete(BlockDriverState
*bs
)
4485 assert(bdrv_op_blocker_is_empty(bs
));
4486 assert(!bs
->refcnt
);
4488 /* remove from list, if necessary */
4489 if (bs
->node_name
[0] != '\0') {
4490 QTAILQ_REMOVE(&graph_bdrv_states
, bs
, node_list
);
4492 QTAILQ_REMOVE(&all_bdrv_states
, bs
, bs_list
);
4500 * Run consistency checks on an image
4502 * Returns 0 if the check could be completed (it doesn't mean that the image is
4503 * free of errors) or -errno when an internal error occurred. The results of the
4504 * check are stored in res.
4506 static int coroutine_fn
bdrv_co_check(BlockDriverState
*bs
,
4507 BdrvCheckResult
*res
, BdrvCheckMode fix
)
4509 if (bs
->drv
== NULL
) {
4512 if (bs
->drv
->bdrv_co_check
== NULL
) {
4516 memset(res
, 0, sizeof(*res
));
4517 return bs
->drv
->bdrv_co_check(bs
, res
, fix
);
4520 typedef struct CheckCo
{
4521 BlockDriverState
*bs
;
4522 BdrvCheckResult
*res
;
4527 static void coroutine_fn
bdrv_check_co_entry(void *opaque
)
4529 CheckCo
*cco
= opaque
;
4530 cco
->ret
= bdrv_co_check(cco
->bs
, cco
->res
, cco
->fix
);
4534 int bdrv_check(BlockDriverState
*bs
,
4535 BdrvCheckResult
*res
, BdrvCheckMode fix
)
4541 .ret
= -EINPROGRESS
,
4545 if (qemu_in_coroutine()) {
4546 /* Fast-path if already in coroutine context */
4547 bdrv_check_co_entry(&cco
);
4549 co
= qemu_coroutine_create(bdrv_check_co_entry
, &cco
);
4550 bdrv_coroutine_enter(bs
, co
);
4551 BDRV_POLL_WHILE(bs
, cco
.ret
== -EINPROGRESS
);
4560 * -EINVAL - backing format specified, but no file
4561 * -ENOSPC - can't update the backing file because no space is left in the
4563 * -ENOTSUP - format driver doesn't support changing the backing file
4565 int bdrv_change_backing_file(BlockDriverState
*bs
,
4566 const char *backing_file
, const char *backing_fmt
)
4568 BlockDriver
*drv
= bs
->drv
;
4575 /* Backing file format doesn't make sense without a backing file */
4576 if (backing_fmt
&& !backing_file
) {
4580 if (drv
->bdrv_change_backing_file
!= NULL
) {
4581 ret
= drv
->bdrv_change_backing_file(bs
, backing_file
, backing_fmt
);
4587 pstrcpy(bs
->backing_file
, sizeof(bs
->backing_file
), backing_file
?: "");
4588 pstrcpy(bs
->backing_format
, sizeof(bs
->backing_format
), backing_fmt
?: "");
4589 pstrcpy(bs
->auto_backing_file
, sizeof(bs
->auto_backing_file
),
4590 backing_file
?: "");
4596 * Finds the image layer in the chain that has 'bs' as its backing file.
4598 * active is the current topmost image.
4600 * Returns NULL if bs is not found in active's image chain,
4601 * or if active == bs.
4603 * Returns the bottommost base image if bs == NULL.
4605 BlockDriverState
*bdrv_find_overlay(BlockDriverState
*active
,
4606 BlockDriverState
*bs
)
4608 while (active
&& bs
!= backing_bs(active
)) {
4609 active
= backing_bs(active
);
4615 /* Given a BDS, searches for the base layer. */
4616 BlockDriverState
*bdrv_find_base(BlockDriverState
*bs
)
4618 return bdrv_find_overlay(bs
, NULL
);
4622 * Return true if at least one of the backing links between @bs and
4623 * @base is frozen. @errp is set if that's the case.
4624 * @base must be reachable from @bs, or NULL.
4626 bool bdrv_is_backing_chain_frozen(BlockDriverState
*bs
, BlockDriverState
*base
,
4629 BlockDriverState
*i
;
4631 for (i
= bs
; i
!= base
; i
= backing_bs(i
)) {
4632 if (i
->backing
&& i
->backing
->frozen
) {
4633 error_setg(errp
, "Cannot change '%s' link from '%s' to '%s'",
4634 i
->backing
->name
, i
->node_name
,
4635 backing_bs(i
)->node_name
);
4644 * Freeze all backing links between @bs and @base.
4645 * If any of the links is already frozen the operation is aborted and
4646 * none of the links are modified.
4647 * @base must be reachable from @bs, or NULL.
4648 * Returns 0 on success. On failure returns < 0 and sets @errp.
4650 int bdrv_freeze_backing_chain(BlockDriverState
*bs
, BlockDriverState
*base
,
4653 BlockDriverState
*i
;
4655 if (bdrv_is_backing_chain_frozen(bs
, base
, errp
)) {
4659 for (i
= bs
; i
!= base
; i
= backing_bs(i
)) {
4660 if (i
->backing
&& backing_bs(i
)->never_freeze
) {
4661 error_setg(errp
, "Cannot freeze '%s' link to '%s'",
4662 i
->backing
->name
, backing_bs(i
)->node_name
);
4667 for (i
= bs
; i
!= base
; i
= backing_bs(i
)) {
4669 i
->backing
->frozen
= true;
4677 * Unfreeze all backing links between @bs and @base. The caller must
4678 * ensure that all links are frozen before using this function.
4679 * @base must be reachable from @bs, or NULL.
4681 void bdrv_unfreeze_backing_chain(BlockDriverState
*bs
, BlockDriverState
*base
)
4683 BlockDriverState
*i
;
4685 for (i
= bs
; i
!= base
; i
= backing_bs(i
)) {
4687 assert(i
->backing
->frozen
);
4688 i
->backing
->frozen
= false;
4694 * Drops images above 'base' up to and including 'top', and sets the image
4695 * above 'top' to have base as its backing file.
4697 * Requires that the overlay to 'top' is opened r/w, so that the backing file
4698 * information in 'bs' can be properly updated.
4700 * E.g., this will convert the following chain:
4701 * bottom <- base <- intermediate <- top <- active
4705 * bottom <- base <- active
4707 * It is allowed for bottom==base, in which case it converts:
4709 * base <- intermediate <- top <- active
4715 * If backing_file_str is non-NULL, it will be used when modifying top's
4716 * overlay image metadata.
4719 * if active == top, that is considered an error
4722 int bdrv_drop_intermediate(BlockDriverState
*top
, BlockDriverState
*base
,
4723 const char *backing_file_str
)
4725 BlockDriverState
*explicit_top
= top
;
4726 bool update_inherits_from
;
4727 BdrvChild
*c
, *next
;
4728 Error
*local_err
= NULL
;
4732 bdrv_subtree_drained_begin(top
);
4734 if (!top
->drv
|| !base
->drv
) {
4738 /* Make sure that base is in the backing chain of top */
4739 if (!bdrv_chain_contains(top
, base
)) {
4743 /* This function changes all links that point to top and makes
4744 * them point to base. Check that none of them is frozen. */
4745 QLIST_FOREACH(c
, &top
->parents
, next_parent
) {
4751 /* If 'base' recursively inherits from 'top' then we should set
4752 * base->inherits_from to top->inherits_from after 'top' and all
4753 * other intermediate nodes have been dropped.
4754 * If 'top' is an implicit node (e.g. "commit_top") we should skip
4755 * it because no one inherits from it. We use explicit_top for that. */
4756 while (explicit_top
&& explicit_top
->implicit
) {
4757 explicit_top
= backing_bs(explicit_top
);
4759 update_inherits_from
= bdrv_inherits_from_recursive(base
, explicit_top
);
4761 /* success - we can delete the intermediate states, and link top->base */
4762 /* TODO Check graph modification op blockers (BLK_PERM_GRAPH_MOD) once
4763 * we've figured out how they should work. */
4764 if (!backing_file_str
) {
4765 bdrv_refresh_filename(base
);
4766 backing_file_str
= base
->filename
;
4769 QLIST_FOREACH_SAFE(c
, &top
->parents
, next_parent
, next
) {
4770 /* Check whether we are allowed to switch c from top to base */
4771 GSList
*ignore_children
= g_slist_prepend(NULL
, c
);
4772 ret
= bdrv_check_update_perm(base
, NULL
, c
->perm
, c
->shared_perm
,
4773 ignore_children
, NULL
, &local_err
);
4774 g_slist_free(ignore_children
);
4776 error_report_err(local_err
);
4780 /* If so, update the backing file path in the image file */
4781 if (c
->role
->update_filename
) {
4782 ret
= c
->role
->update_filename(c
, base
, backing_file_str
,
4785 bdrv_abort_perm_update(base
);
4786 error_report_err(local_err
);
4791 /* Do the actual switch in the in-memory graph.
4792 * Completes bdrv_check_update_perm() transaction internally. */
4794 bdrv_replace_child(c
, base
);
4798 if (update_inherits_from
) {
4799 base
->inherits_from
= explicit_top
->inherits_from
;
4804 bdrv_subtree_drained_end(top
);
4810 * Length of a allocated file in bytes. Sparse files are counted by actual
4811 * allocated space. Return < 0 if error or unknown.
4813 int64_t bdrv_get_allocated_file_size(BlockDriverState
*bs
)
4815 BlockDriver
*drv
= bs
->drv
;
4819 if (drv
->bdrv_get_allocated_file_size
) {
4820 return drv
->bdrv_get_allocated_file_size(bs
);
4823 return bdrv_get_allocated_file_size(bs
->file
->bs
);
4830 * @drv: Format driver
4831 * @opts: Creation options for new image
4832 * @in_bs: Existing image containing data for new image (may be NULL)
4833 * @errp: Error object
4834 * Returns: A #BlockMeasureInfo (free using qapi_free_BlockMeasureInfo())
4837 * Calculate file size required to create a new image.
4839 * If @in_bs is given then space for allocated clusters and zero clusters
4840 * from that image are included in the calculation. If @opts contains a
4841 * backing file that is shared by @in_bs then backing clusters may be omitted
4842 * from the calculation.
4844 * If @in_bs is NULL then the calculation includes no allocated clusters
4845 * unless a preallocation option is given in @opts.
4847 * Note that @in_bs may use a different BlockDriver from @drv.
4849 * If an error occurs the @errp pointer is set.
4851 BlockMeasureInfo
*bdrv_measure(BlockDriver
*drv
, QemuOpts
*opts
,
4852 BlockDriverState
*in_bs
, Error
**errp
)
4854 if (!drv
->bdrv_measure
) {
4855 error_setg(errp
, "Block driver '%s' does not support size measurement",
4860 return drv
->bdrv_measure(opts
, in_bs
, errp
);
4864 * Return number of sectors on success, -errno on error.
4866 int64_t bdrv_nb_sectors(BlockDriverState
*bs
)
4868 BlockDriver
*drv
= bs
->drv
;
4873 if (drv
->has_variable_length
) {
4874 int ret
= refresh_total_sectors(bs
, bs
->total_sectors
);
4879 return bs
->total_sectors
;
4883 * Return length in bytes on success, -errno on error.
4884 * The length is always a multiple of BDRV_SECTOR_SIZE.
4886 int64_t bdrv_getlength(BlockDriverState
*bs
)
4888 int64_t ret
= bdrv_nb_sectors(bs
);
4890 ret
= ret
> INT64_MAX
/ BDRV_SECTOR_SIZE
? -EFBIG
: ret
;
4891 return ret
< 0 ? ret
: ret
* BDRV_SECTOR_SIZE
;
4894 /* return 0 as number of sectors if no device present or error */
4895 void bdrv_get_geometry(BlockDriverState
*bs
, uint64_t *nb_sectors_ptr
)
4897 int64_t nb_sectors
= bdrv_nb_sectors(bs
);
4899 *nb_sectors_ptr
= nb_sectors
< 0 ? 0 : nb_sectors
;
4902 bool bdrv_is_sg(BlockDriverState
*bs
)
4907 bool bdrv_is_encrypted(BlockDriverState
*bs
)
4909 if (bs
->backing
&& bs
->backing
->bs
->encrypted
) {
4912 return bs
->encrypted
;
4915 const char *bdrv_get_format_name(BlockDriverState
*bs
)
4917 return bs
->drv
? bs
->drv
->format_name
: NULL
;
4920 static int qsort_strcmp(const void *a
, const void *b
)
4922 return strcmp(*(char *const *)a
, *(char *const *)b
);
4925 void bdrv_iterate_format(void (*it
)(void *opaque
, const char *name
),
4926 void *opaque
, bool read_only
)
4931 const char **formats
= NULL
;
4933 QLIST_FOREACH(drv
, &bdrv_drivers
, list
) {
4934 if (drv
->format_name
) {
4938 if (use_bdrv_whitelist
&& !bdrv_is_whitelisted(drv
, read_only
)) {
4942 while (formats
&& i
&& !found
) {
4943 found
= !strcmp(formats
[--i
], drv
->format_name
);
4947 formats
= g_renew(const char *, formats
, count
+ 1);
4948 formats
[count
++] = drv
->format_name
;
4953 for (i
= 0; i
< (int)ARRAY_SIZE(block_driver_modules
); i
++) {
4954 const char *format_name
= block_driver_modules
[i
].format_name
;
4960 if (use_bdrv_whitelist
&&
4961 !bdrv_format_is_whitelisted(format_name
, read_only
)) {
4965 while (formats
&& j
&& !found
) {
4966 found
= !strcmp(formats
[--j
], format_name
);
4970 formats
= g_renew(const char *, formats
, count
+ 1);
4971 formats
[count
++] = format_name
;
4976 qsort(formats
, count
, sizeof(formats
[0]), qsort_strcmp
);
4978 for (i
= 0; i
< count
; i
++) {
4979 it(opaque
, formats
[i
]);
4985 /* This function is to find a node in the bs graph */
4986 BlockDriverState
*bdrv_find_node(const char *node_name
)
4988 BlockDriverState
*bs
;
4992 QTAILQ_FOREACH(bs
, &graph_bdrv_states
, node_list
) {
4993 if (!strcmp(node_name
, bs
->node_name
)) {
5000 /* Put this QMP function here so it can access the static graph_bdrv_states. */
5001 BlockDeviceInfoList
*bdrv_named_nodes_list(bool flat
,
5004 BlockDeviceInfoList
*list
, *entry
;
5005 BlockDriverState
*bs
;
5008 QTAILQ_FOREACH(bs
, &graph_bdrv_states
, node_list
) {
5009 BlockDeviceInfo
*info
= bdrv_block_device_info(NULL
, bs
, flat
, errp
);
5011 qapi_free_BlockDeviceInfoList(list
);
5014 entry
= g_malloc0(sizeof(*entry
));
5015 entry
->value
= info
;
5023 #define QAPI_LIST_ADD(list, element) do { \
5024 typeof(list) _tmp = g_new(typeof(*(list)), 1); \
5025 _tmp->value = (element); \
5026 _tmp->next = (list); \
5030 typedef struct XDbgBlockGraphConstructor
{
5031 XDbgBlockGraph
*graph
;
5032 GHashTable
*graph_nodes
;
5033 } XDbgBlockGraphConstructor
;
5035 static XDbgBlockGraphConstructor
*xdbg_graph_new(void)
5037 XDbgBlockGraphConstructor
*gr
= g_new(XDbgBlockGraphConstructor
, 1);
5039 gr
->graph
= g_new0(XDbgBlockGraph
, 1);
5040 gr
->graph_nodes
= g_hash_table_new(NULL
, NULL
);
5045 static XDbgBlockGraph
*xdbg_graph_finalize(XDbgBlockGraphConstructor
*gr
)
5047 XDbgBlockGraph
*graph
= gr
->graph
;
5049 g_hash_table_destroy(gr
->graph_nodes
);
5055 static uintptr_t xdbg_graph_node_num(XDbgBlockGraphConstructor
*gr
, void *node
)
5057 uintptr_t ret
= (uintptr_t)g_hash_table_lookup(gr
->graph_nodes
, node
);
5064 * Start counting from 1, not 0, because 0 interferes with not-found (NULL)
5065 * answer of g_hash_table_lookup.
5067 ret
= g_hash_table_size(gr
->graph_nodes
) + 1;
5068 g_hash_table_insert(gr
->graph_nodes
, node
, (void *)ret
);
5073 static void xdbg_graph_add_node(XDbgBlockGraphConstructor
*gr
, void *node
,
5074 XDbgBlockGraphNodeType type
, const char *name
)
5076 XDbgBlockGraphNode
*n
;
5078 n
= g_new0(XDbgBlockGraphNode
, 1);
5080 n
->id
= xdbg_graph_node_num(gr
, node
);
5082 n
->name
= g_strdup(name
);
5084 QAPI_LIST_ADD(gr
->graph
->nodes
, n
);
5087 static void xdbg_graph_add_edge(XDbgBlockGraphConstructor
*gr
, void *parent
,
5088 const BdrvChild
*child
)
5090 BlockPermission qapi_perm
;
5091 XDbgBlockGraphEdge
*edge
;
5093 edge
= g_new0(XDbgBlockGraphEdge
, 1);
5095 edge
->parent
= xdbg_graph_node_num(gr
, parent
);
5096 edge
->child
= xdbg_graph_node_num(gr
, child
->bs
);
5097 edge
->name
= g_strdup(child
->name
);
5099 for (qapi_perm
= 0; qapi_perm
< BLOCK_PERMISSION__MAX
; qapi_perm
++) {
5100 uint64_t flag
= bdrv_qapi_perm_to_blk_perm(qapi_perm
);
5102 if (flag
& child
->perm
) {
5103 QAPI_LIST_ADD(edge
->perm
, qapi_perm
);
5105 if (flag
& child
->shared_perm
) {
5106 QAPI_LIST_ADD(edge
->shared_perm
, qapi_perm
);
5110 QAPI_LIST_ADD(gr
->graph
->edges
, edge
);
5114 XDbgBlockGraph
*bdrv_get_xdbg_block_graph(Error
**errp
)
5118 BlockDriverState
*bs
;
5120 XDbgBlockGraphConstructor
*gr
= xdbg_graph_new();
5122 for (blk
= blk_all_next(NULL
); blk
; blk
= blk_all_next(blk
)) {
5123 char *allocated_name
= NULL
;
5124 const char *name
= blk_name(blk
);
5127 name
= allocated_name
= blk_get_attached_dev_id(blk
);
5129 xdbg_graph_add_node(gr
, blk
, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_BACKEND
,
5131 g_free(allocated_name
);
5132 if (blk_root(blk
)) {
5133 xdbg_graph_add_edge(gr
, blk
, blk_root(blk
));
5137 for (job
= block_job_next(NULL
); job
; job
= block_job_next(job
)) {
5140 xdbg_graph_add_node(gr
, job
, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_JOB
,
5142 for (el
= job
->nodes
; el
; el
= el
->next
) {
5143 xdbg_graph_add_edge(gr
, job
, (BdrvChild
*)el
->data
);
5147 QTAILQ_FOREACH(bs
, &graph_bdrv_states
, node_list
) {
5148 xdbg_graph_add_node(gr
, bs
, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_DRIVER
,
5150 QLIST_FOREACH(child
, &bs
->children
, next
) {
5151 xdbg_graph_add_edge(gr
, bs
, child
);
5155 return xdbg_graph_finalize(gr
);
5158 BlockDriverState
*bdrv_lookup_bs(const char *device
,
5159 const char *node_name
,
5163 BlockDriverState
*bs
;
5166 blk
= blk_by_name(device
);
5171 error_setg(errp
, "Device '%s' has no medium", device
);
5179 bs
= bdrv_find_node(node_name
);
5186 error_setg(errp
, "Cannot find device=%s nor node_name=%s",
5187 device
? device
: "",
5188 node_name
? node_name
: "");
5192 /* If 'base' is in the same chain as 'top', return true. Otherwise,
5193 * return false. If either argument is NULL, return false. */
5194 bool bdrv_chain_contains(BlockDriverState
*top
, BlockDriverState
*base
)
5196 while (top
&& top
!= base
) {
5197 top
= backing_bs(top
);
5203 BlockDriverState
*bdrv_next_node(BlockDriverState
*bs
)
5206 return QTAILQ_FIRST(&graph_bdrv_states
);
5208 return QTAILQ_NEXT(bs
, node_list
);
5211 BlockDriverState
*bdrv_next_all_states(BlockDriverState
*bs
)
5214 return QTAILQ_FIRST(&all_bdrv_states
);
5216 return QTAILQ_NEXT(bs
, bs_list
);
5219 const char *bdrv_get_node_name(const BlockDriverState
*bs
)
5221 return bs
->node_name
;
5224 const char *bdrv_get_parent_name(const BlockDriverState
*bs
)
5229 /* If multiple parents have a name, just pick the first one. */
5230 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
5231 if (c
->role
->get_name
) {
5232 name
= c
->role
->get_name(c
);
5233 if (name
&& *name
) {
5242 /* TODO check what callers really want: bs->node_name or blk_name() */
5243 const char *bdrv_get_device_name(const BlockDriverState
*bs
)
5245 return bdrv_get_parent_name(bs
) ?: "";
5248 /* This can be used to identify nodes that might not have a device
5249 * name associated. Since node and device names live in the same
5250 * namespace, the result is unambiguous. The exception is if both are
5251 * absent, then this returns an empty (non-null) string. */
5252 const char *bdrv_get_device_or_node_name(const BlockDriverState
*bs
)
5254 return bdrv_get_parent_name(bs
) ?: bs
->node_name
;
5257 int bdrv_get_flags(BlockDriverState
*bs
)
5259 return bs
->open_flags
;
5262 int bdrv_has_zero_init_1(BlockDriverState
*bs
)
5267 int bdrv_has_zero_init(BlockDriverState
*bs
)
5273 /* If BS is a copy on write image, it is initialized to
5274 the contents of the base image, which may not be zeroes. */
5278 if (bs
->drv
->bdrv_has_zero_init
) {
5279 return bs
->drv
->bdrv_has_zero_init(bs
);
5281 if (bs
->file
&& bs
->drv
->is_filter
) {
5282 return bdrv_has_zero_init(bs
->file
->bs
);
5289 int bdrv_has_zero_init_truncate(BlockDriverState
*bs
)
5296 /* Depends on the backing image length, but better safe than sorry */
5299 if (bs
->drv
->bdrv_has_zero_init_truncate
) {
5300 return bs
->drv
->bdrv_has_zero_init_truncate(bs
);
5302 if (bs
->file
&& bs
->drv
->is_filter
) {
5303 return bdrv_has_zero_init_truncate(bs
->file
->bs
);
5310 bool bdrv_unallocated_blocks_are_zero(BlockDriverState
*bs
)
5312 BlockDriverInfo bdi
;
5318 if (bdrv_get_info(bs
, &bdi
) == 0) {
5319 return bdi
.unallocated_blocks_are_zero
;
5325 bool bdrv_can_write_zeroes_with_unmap(BlockDriverState
*bs
)
5327 if (!(bs
->open_flags
& BDRV_O_UNMAP
)) {
5331 return bs
->supported_zero_flags
& BDRV_REQ_MAY_UNMAP
;
5334 void bdrv_get_backing_filename(BlockDriverState
*bs
,
5335 char *filename
, int filename_size
)
5337 pstrcpy(filename
, filename_size
, bs
->backing_file
);
5340 int bdrv_get_info(BlockDriverState
*bs
, BlockDriverInfo
*bdi
)
5342 BlockDriver
*drv
= bs
->drv
;
5343 /* if bs->drv == NULL, bs is closed, so there's nothing to do here */
5347 if (!drv
->bdrv_get_info
) {
5348 if (bs
->file
&& drv
->is_filter
) {
5349 return bdrv_get_info(bs
->file
->bs
, bdi
);
5353 memset(bdi
, 0, sizeof(*bdi
));
5354 return drv
->bdrv_get_info(bs
, bdi
);
5357 ImageInfoSpecific
*bdrv_get_specific_info(BlockDriverState
*bs
,
5360 BlockDriver
*drv
= bs
->drv
;
5361 if (drv
&& drv
->bdrv_get_specific_info
) {
5362 return drv
->bdrv_get_specific_info(bs
, errp
);
5367 BlockStatsSpecific
*bdrv_get_specific_stats(BlockDriverState
*bs
)
5369 BlockDriver
*drv
= bs
->drv
;
5370 if (!drv
|| !drv
->bdrv_get_specific_stats
) {
5373 return drv
->bdrv_get_specific_stats(bs
);
5376 void bdrv_debug_event(BlockDriverState
*bs
, BlkdebugEvent event
)
5378 if (!bs
|| !bs
->drv
|| !bs
->drv
->bdrv_debug_event
) {
5382 bs
->drv
->bdrv_debug_event(bs
, event
);
5385 static BlockDriverState
*bdrv_find_debug_node(BlockDriverState
*bs
)
5387 while (bs
&& bs
->drv
&& !bs
->drv
->bdrv_debug_breakpoint
) {
5393 if (bs
->drv
->is_filter
&& bs
->backing
) {
5394 bs
= bs
->backing
->bs
;
5401 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_breakpoint
) {
5402 assert(bs
->drv
->bdrv_debug_remove_breakpoint
);
5409 int bdrv_debug_breakpoint(BlockDriverState
*bs
, const char *event
,
5412 bs
= bdrv_find_debug_node(bs
);
5414 return bs
->drv
->bdrv_debug_breakpoint(bs
, event
, tag
);
5420 int bdrv_debug_remove_breakpoint(BlockDriverState
*bs
, const char *tag
)
5422 bs
= bdrv_find_debug_node(bs
);
5424 return bs
->drv
->bdrv_debug_remove_breakpoint(bs
, tag
);
5430 int bdrv_debug_resume(BlockDriverState
*bs
, const char *tag
)
5432 while (bs
&& (!bs
->drv
|| !bs
->drv
->bdrv_debug_resume
)) {
5433 bs
= bs
->file
? bs
->file
->bs
: NULL
;
5436 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_resume
) {
5437 return bs
->drv
->bdrv_debug_resume(bs
, tag
);
5443 bool bdrv_debug_is_suspended(BlockDriverState
*bs
, const char *tag
)
5445 while (bs
&& bs
->drv
&& !bs
->drv
->bdrv_debug_is_suspended
) {
5446 bs
= bs
->file
? bs
->file
->bs
: NULL
;
5449 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_is_suspended
) {
5450 return bs
->drv
->bdrv_debug_is_suspended(bs
, tag
);
5456 /* backing_file can either be relative, or absolute, or a protocol. If it is
5457 * relative, it must be relative to the chain. So, passing in bs->filename
5458 * from a BDS as backing_file should not be done, as that may be relative to
5459 * the CWD rather than the chain. */
5460 BlockDriverState
*bdrv_find_backing_image(BlockDriverState
*bs
,
5461 const char *backing_file
)
5463 char *filename_full
= NULL
;
5464 char *backing_file_full
= NULL
;
5465 char *filename_tmp
= NULL
;
5466 int is_protocol
= 0;
5467 BlockDriverState
*curr_bs
= NULL
;
5468 BlockDriverState
*retval
= NULL
;
5470 if (!bs
|| !bs
->drv
|| !backing_file
) {
5474 filename_full
= g_malloc(PATH_MAX
);
5475 backing_file_full
= g_malloc(PATH_MAX
);
5477 is_protocol
= path_has_protocol(backing_file
);
5479 for (curr_bs
= bs
; curr_bs
->backing
; curr_bs
= curr_bs
->backing
->bs
) {
5481 /* If either of the filename paths is actually a protocol, then
5482 * compare unmodified paths; otherwise make paths relative */
5483 if (is_protocol
|| path_has_protocol(curr_bs
->backing_file
)) {
5484 char *backing_file_full_ret
;
5486 if (strcmp(backing_file
, curr_bs
->backing_file
) == 0) {
5487 retval
= curr_bs
->backing
->bs
;
5490 /* Also check against the full backing filename for the image */
5491 backing_file_full_ret
= bdrv_get_full_backing_filename(curr_bs
,
5493 if (backing_file_full_ret
) {
5494 bool equal
= strcmp(backing_file
, backing_file_full_ret
) == 0;
5495 g_free(backing_file_full_ret
);
5497 retval
= curr_bs
->backing
->bs
;
5502 /* If not an absolute filename path, make it relative to the current
5503 * image's filename path */
5504 filename_tmp
= bdrv_make_absolute_filename(curr_bs
, backing_file
,
5506 /* We are going to compare canonicalized absolute pathnames */
5507 if (!filename_tmp
|| !realpath(filename_tmp
, filename_full
)) {
5508 g_free(filename_tmp
);
5511 g_free(filename_tmp
);
5513 /* We need to make sure the backing filename we are comparing against
5514 * is relative to the current image filename (or absolute) */
5515 filename_tmp
= bdrv_get_full_backing_filename(curr_bs
, NULL
);
5516 if (!filename_tmp
|| !realpath(filename_tmp
, backing_file_full
)) {
5517 g_free(filename_tmp
);
5520 g_free(filename_tmp
);
5522 if (strcmp(backing_file_full
, filename_full
) == 0) {
5523 retval
= curr_bs
->backing
->bs
;
5529 g_free(filename_full
);
5530 g_free(backing_file_full
);
5534 void bdrv_init(void)
5536 module_call_init(MODULE_INIT_BLOCK
);
5539 void bdrv_init_with_whitelist(void)
5541 use_bdrv_whitelist
= 1;
5545 static void coroutine_fn
bdrv_co_invalidate_cache(BlockDriverState
*bs
,
5548 BdrvChild
*child
, *parent
;
5549 uint64_t perm
, shared_perm
;
5550 Error
*local_err
= NULL
;
5552 BdrvDirtyBitmap
*bm
;
5558 QLIST_FOREACH(child
, &bs
->children
, next
) {
5559 bdrv_co_invalidate_cache(child
->bs
, &local_err
);
5561 error_propagate(errp
, local_err
);
5567 * Update permissions, they may differ for inactive nodes.
5569 * Note that the required permissions of inactive images are always a
5570 * subset of the permissions required after activating the image. This
5571 * allows us to just get the permissions upfront without restricting
5572 * drv->bdrv_invalidate_cache().
5574 * It also means that in error cases, we don't have to try and revert to
5575 * the old permissions (which is an operation that could fail, too). We can
5576 * just keep the extended permissions for the next time that an activation
5577 * of the image is tried.
5579 if (bs
->open_flags
& BDRV_O_INACTIVE
) {
5580 bs
->open_flags
&= ~BDRV_O_INACTIVE
;
5581 bdrv_get_cumulative_perm(bs
, &perm
, &shared_perm
);
5582 ret
= bdrv_check_perm(bs
, NULL
, perm
, shared_perm
, NULL
, NULL
, &local_err
);
5584 bs
->open_flags
|= BDRV_O_INACTIVE
;
5585 error_propagate(errp
, local_err
);
5588 bdrv_set_perm(bs
, perm
, shared_perm
);
5590 if (bs
->drv
->bdrv_co_invalidate_cache
) {
5591 bs
->drv
->bdrv_co_invalidate_cache(bs
, &local_err
);
5593 bs
->open_flags
|= BDRV_O_INACTIVE
;
5594 error_propagate(errp
, local_err
);
5599 FOR_EACH_DIRTY_BITMAP(bs
, bm
) {
5600 bdrv_dirty_bitmap_skip_store(bm
, false);
5603 ret
= refresh_total_sectors(bs
, bs
->total_sectors
);
5605 bs
->open_flags
|= BDRV_O_INACTIVE
;
5606 error_setg_errno(errp
, -ret
, "Could not refresh total sector count");
5611 QLIST_FOREACH(parent
, &bs
->parents
, next_parent
) {
5612 if (parent
->role
->activate
) {
5613 parent
->role
->activate(parent
, &local_err
);
5615 bs
->open_flags
|= BDRV_O_INACTIVE
;
5616 error_propagate(errp
, local_err
);
5623 typedef struct InvalidateCacheCo
{
5624 BlockDriverState
*bs
;
5627 } InvalidateCacheCo
;
5629 static void coroutine_fn
bdrv_invalidate_cache_co_entry(void *opaque
)
5631 InvalidateCacheCo
*ico
= opaque
;
5632 bdrv_co_invalidate_cache(ico
->bs
, ico
->errp
);
5637 void bdrv_invalidate_cache(BlockDriverState
*bs
, Error
**errp
)
5640 InvalidateCacheCo ico
= {
5646 if (qemu_in_coroutine()) {
5647 /* Fast-path if already in coroutine context */
5648 bdrv_invalidate_cache_co_entry(&ico
);
5650 co
= qemu_coroutine_create(bdrv_invalidate_cache_co_entry
, &ico
);
5651 bdrv_coroutine_enter(bs
, co
);
5652 BDRV_POLL_WHILE(bs
, !ico
.done
);
5656 void bdrv_invalidate_cache_all(Error
**errp
)
5658 BlockDriverState
*bs
;
5659 Error
*local_err
= NULL
;
5660 BdrvNextIterator it
;
5662 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
5663 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
5665 aio_context_acquire(aio_context
);
5666 bdrv_invalidate_cache(bs
, &local_err
);
5667 aio_context_release(aio_context
);
5669 error_propagate(errp
, local_err
);
5670 bdrv_next_cleanup(&it
);
5676 static bool bdrv_has_bds_parent(BlockDriverState
*bs
, bool only_active
)
5680 QLIST_FOREACH(parent
, &bs
->parents
, next_parent
) {
5681 if (parent
->role
->parent_is_bds
) {
5682 BlockDriverState
*parent_bs
= parent
->opaque
;
5683 if (!only_active
|| !(parent_bs
->open_flags
& BDRV_O_INACTIVE
)) {
5692 static int bdrv_inactivate_recurse(BlockDriverState
*bs
)
5694 BdrvChild
*child
, *parent
;
5695 bool tighten_restrictions
;
5696 uint64_t perm
, shared_perm
;
5703 /* Make sure that we don't inactivate a child before its parent.
5704 * It will be covered by recursion from the yet active parent. */
5705 if (bdrv_has_bds_parent(bs
, true)) {
5709 assert(!(bs
->open_flags
& BDRV_O_INACTIVE
));
5711 /* Inactivate this node */
5712 if (bs
->drv
->bdrv_inactivate
) {
5713 ret
= bs
->drv
->bdrv_inactivate(bs
);
5719 QLIST_FOREACH(parent
, &bs
->parents
, next_parent
) {
5720 if (parent
->role
->inactivate
) {
5721 ret
= parent
->role
->inactivate(parent
);
5728 bs
->open_flags
|= BDRV_O_INACTIVE
;
5730 /* Update permissions, they may differ for inactive nodes */
5731 bdrv_get_cumulative_perm(bs
, &perm
, &shared_perm
);
5732 ret
= bdrv_check_perm(bs
, NULL
, perm
, shared_perm
, NULL
,
5733 &tighten_restrictions
, NULL
);
5734 assert(tighten_restrictions
== false);
5736 /* We only tried to loosen restrictions, so errors are not fatal */
5737 bdrv_abort_perm_update(bs
);
5739 bdrv_set_perm(bs
, perm
, shared_perm
);
5743 /* Recursively inactivate children */
5744 QLIST_FOREACH(child
, &bs
->children
, next
) {
5745 ret
= bdrv_inactivate_recurse(child
->bs
);
5754 int bdrv_inactivate_all(void)
5756 BlockDriverState
*bs
= NULL
;
5757 BdrvNextIterator it
;
5759 GSList
*aio_ctxs
= NULL
, *ctx
;
5761 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
5762 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
5764 if (!g_slist_find(aio_ctxs
, aio_context
)) {
5765 aio_ctxs
= g_slist_prepend(aio_ctxs
, aio_context
);
5766 aio_context_acquire(aio_context
);
5770 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
5771 /* Nodes with BDS parents are covered by recursion from the last
5772 * parent that gets inactivated. Don't inactivate them a second
5773 * time if that has already happened. */
5774 if (bdrv_has_bds_parent(bs
, false)) {
5777 ret
= bdrv_inactivate_recurse(bs
);
5779 bdrv_next_cleanup(&it
);
5785 for (ctx
= aio_ctxs
; ctx
!= NULL
; ctx
= ctx
->next
) {
5786 AioContext
*aio_context
= ctx
->data
;
5787 aio_context_release(aio_context
);
5789 g_slist_free(aio_ctxs
);
5794 /**************************************************************/
5795 /* removable device support */
5798 * Return TRUE if the media is present
5800 bool bdrv_is_inserted(BlockDriverState
*bs
)
5802 BlockDriver
*drv
= bs
->drv
;
5808 if (drv
->bdrv_is_inserted
) {
5809 return drv
->bdrv_is_inserted(bs
);
5811 QLIST_FOREACH(child
, &bs
->children
, next
) {
5812 if (!bdrv_is_inserted(child
->bs
)) {
5820 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
5822 void bdrv_eject(BlockDriverState
*bs
, bool eject_flag
)
5824 BlockDriver
*drv
= bs
->drv
;
5826 if (drv
&& drv
->bdrv_eject
) {
5827 drv
->bdrv_eject(bs
, eject_flag
);
5832 * Lock or unlock the media (if it is locked, the user won't be able
5833 * to eject it manually).
5835 void bdrv_lock_medium(BlockDriverState
*bs
, bool locked
)
5837 BlockDriver
*drv
= bs
->drv
;
5839 trace_bdrv_lock_medium(bs
, locked
);
5841 if (drv
&& drv
->bdrv_lock_medium
) {
5842 drv
->bdrv_lock_medium(bs
, locked
);
5846 /* Get a reference to bs */
5847 void bdrv_ref(BlockDriverState
*bs
)
5852 /* Release a previously grabbed reference to bs.
5853 * If after releasing, reference count is zero, the BlockDriverState is
5855 void bdrv_unref(BlockDriverState
*bs
)
5860 assert(bs
->refcnt
> 0);
5861 if (--bs
->refcnt
== 0) {
5866 struct BdrvOpBlocker
{
5868 QLIST_ENTRY(BdrvOpBlocker
) list
;
5871 bool bdrv_op_is_blocked(BlockDriverState
*bs
, BlockOpType op
, Error
**errp
)
5873 BdrvOpBlocker
*blocker
;
5874 assert((int) op
>= 0 && op
< BLOCK_OP_TYPE_MAX
);
5875 if (!QLIST_EMPTY(&bs
->op_blockers
[op
])) {
5876 blocker
= QLIST_FIRST(&bs
->op_blockers
[op
]);
5877 error_propagate_prepend(errp
, error_copy(blocker
->reason
),
5878 "Node '%s' is busy: ",
5879 bdrv_get_device_or_node_name(bs
));
5885 void bdrv_op_block(BlockDriverState
*bs
, BlockOpType op
, Error
*reason
)
5887 BdrvOpBlocker
*blocker
;
5888 assert((int) op
>= 0 && op
< BLOCK_OP_TYPE_MAX
);
5890 blocker
= g_new0(BdrvOpBlocker
, 1);
5891 blocker
->reason
= reason
;
5892 QLIST_INSERT_HEAD(&bs
->op_blockers
[op
], blocker
, list
);
5895 void bdrv_op_unblock(BlockDriverState
*bs
, BlockOpType op
, Error
*reason
)
5897 BdrvOpBlocker
*blocker
, *next
;
5898 assert((int) op
>= 0 && op
< BLOCK_OP_TYPE_MAX
);
5899 QLIST_FOREACH_SAFE(blocker
, &bs
->op_blockers
[op
], list
, next
) {
5900 if (blocker
->reason
== reason
) {
5901 QLIST_REMOVE(blocker
, list
);
5907 void bdrv_op_block_all(BlockDriverState
*bs
, Error
*reason
)
5910 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
5911 bdrv_op_block(bs
, i
, reason
);
5915 void bdrv_op_unblock_all(BlockDriverState
*bs
, Error
*reason
)
5918 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
5919 bdrv_op_unblock(bs
, i
, reason
);
5923 bool bdrv_op_blocker_is_empty(BlockDriverState
*bs
)
5927 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
5928 if (!QLIST_EMPTY(&bs
->op_blockers
[i
])) {
5935 void bdrv_img_create(const char *filename
, const char *fmt
,
5936 const char *base_filename
, const char *base_fmt
,
5937 char *options
, uint64_t img_size
, int flags
, bool quiet
,
5940 QemuOptsList
*create_opts
= NULL
;
5941 QemuOpts
*opts
= NULL
;
5942 const char *backing_fmt
, *backing_file
;
5944 BlockDriver
*drv
, *proto_drv
;
5945 Error
*local_err
= NULL
;
5948 /* Find driver and parse its options */
5949 drv
= bdrv_find_format(fmt
);
5951 error_setg(errp
, "Unknown file format '%s'", fmt
);
5955 proto_drv
= bdrv_find_protocol(filename
, true, errp
);
5960 if (!drv
->create_opts
) {
5961 error_setg(errp
, "Format driver '%s' does not support image creation",
5966 /* Create parameter list */
5967 create_opts
= qemu_opts_append(create_opts
, drv
->create_opts
);
5968 if (proto_drv
->create_opts
) {
5969 create_opts
= qemu_opts_append(create_opts
, proto_drv
->create_opts
);
5971 create_opts
= qemu_opts_append(create_opts
, &fallback_create_opts
);
5974 opts
= qemu_opts_create(create_opts
, NULL
, 0, &error_abort
);
5976 /* Parse -o options */
5978 qemu_opts_do_parse(opts
, options
, NULL
, &local_err
);
5984 if (!qemu_opt_get(opts
, BLOCK_OPT_SIZE
)) {
5985 qemu_opt_set_number(opts
, BLOCK_OPT_SIZE
, img_size
, &error_abort
);
5986 } else if (img_size
!= UINT64_C(-1)) {
5987 error_setg(errp
, "The image size must be specified only once");
5991 if (base_filename
) {
5992 qemu_opt_set(opts
, BLOCK_OPT_BACKING_FILE
, base_filename
, &local_err
);
5994 error_setg(errp
, "Backing file not supported for file format '%s'",
6001 qemu_opt_set(opts
, BLOCK_OPT_BACKING_FMT
, base_fmt
, &local_err
);
6003 error_setg(errp
, "Backing file format not supported for file "
6004 "format '%s'", fmt
);
6009 backing_file
= qemu_opt_get(opts
, BLOCK_OPT_BACKING_FILE
);
6011 if (!strcmp(filename
, backing_file
)) {
6012 error_setg(errp
, "Error: Trying to create an image with the "
6013 "same filename as the backing file");
6018 backing_fmt
= qemu_opt_get(opts
, BLOCK_OPT_BACKING_FMT
);
6020 /* The size for the image must always be specified, unless we have a backing
6021 * file and we have not been forbidden from opening it. */
6022 size
= qemu_opt_get_size(opts
, BLOCK_OPT_SIZE
, img_size
);
6023 if (backing_file
&& !(flags
& BDRV_O_NO_BACKING
)) {
6024 BlockDriverState
*bs
;
6027 QDict
*backing_options
= NULL
;
6030 bdrv_get_full_backing_filename_from_filename(filename
, backing_file
,
6035 assert(full_backing
);
6037 /* backing files always opened read-only */
6039 back_flags
&= ~(BDRV_O_RDWR
| BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
);
6041 backing_options
= qdict_new();
6043 qdict_put_str(backing_options
, "driver", backing_fmt
);
6045 qdict_put_bool(backing_options
, BDRV_OPT_FORCE_SHARE
, true);
6047 bs
= bdrv_open(full_backing
, NULL
, backing_options
, back_flags
,
6049 g_free(full_backing
);
6050 if (!bs
&& size
!= -1) {
6051 /* Couldn't open BS, but we have a size, so it's nonfatal */
6052 warn_reportf_err(local_err
,
6053 "Could not verify backing image. "
6054 "This may become an error in future versions.\n");
6057 /* Couldn't open bs, do not have size */
6058 error_append_hint(&local_err
,
6059 "Could not open backing image to determine size.\n");
6063 /* Opened BS, have no size */
6064 size
= bdrv_getlength(bs
);
6066 error_setg_errno(errp
, -size
, "Could not get size of '%s'",
6071 qemu_opt_set_number(opts
, BLOCK_OPT_SIZE
, size
, &error_abort
);
6075 } /* (backing_file && !(flags & BDRV_O_NO_BACKING)) */
6078 error_setg(errp
, "Image creation needs a size parameter");
6083 printf("Formatting '%s', fmt=%s ", filename
, fmt
);
6084 qemu_opts_print(opts
, " ");
6088 ret
= bdrv_create(drv
, filename
, opts
, &local_err
);
6090 if (ret
== -EFBIG
) {
6091 /* This is generally a better message than whatever the driver would
6092 * deliver (especially because of the cluster_size_hint), since that
6093 * is most probably not much different from "image too large". */
6094 const char *cluster_size_hint
= "";
6095 if (qemu_opt_get_size(opts
, BLOCK_OPT_CLUSTER_SIZE
, 0)) {
6096 cluster_size_hint
= " (try using a larger cluster size)";
6098 error_setg(errp
, "The image size is too large for file format '%s'"
6099 "%s", fmt
, cluster_size_hint
);
6100 error_free(local_err
);
6105 qemu_opts_del(opts
);
6106 qemu_opts_free(create_opts
);
6107 error_propagate(errp
, local_err
);
6110 AioContext
*bdrv_get_aio_context(BlockDriverState
*bs
)
6112 return bs
? bs
->aio_context
: qemu_get_aio_context();
6115 void bdrv_coroutine_enter(BlockDriverState
*bs
, Coroutine
*co
)
6117 aio_co_enter(bdrv_get_aio_context(bs
), co
);
6120 static void bdrv_do_remove_aio_context_notifier(BdrvAioNotifier
*ban
)
6122 QLIST_REMOVE(ban
, list
);
6126 static void bdrv_detach_aio_context(BlockDriverState
*bs
)
6128 BdrvAioNotifier
*baf
, *baf_tmp
;
6130 assert(!bs
->walking_aio_notifiers
);
6131 bs
->walking_aio_notifiers
= true;
6132 QLIST_FOREACH_SAFE(baf
, &bs
->aio_notifiers
, list
, baf_tmp
) {
6134 bdrv_do_remove_aio_context_notifier(baf
);
6136 baf
->detach_aio_context(baf
->opaque
);
6139 /* Never mind iterating again to check for ->deleted. bdrv_close() will
6140 * remove remaining aio notifiers if we aren't called again.
6142 bs
->walking_aio_notifiers
= false;
6144 if (bs
->drv
&& bs
->drv
->bdrv_detach_aio_context
) {
6145 bs
->drv
->bdrv_detach_aio_context(bs
);
6148 if (bs
->quiesce_counter
) {
6149 aio_enable_external(bs
->aio_context
);
6151 bs
->aio_context
= NULL
;
6154 static void bdrv_attach_aio_context(BlockDriverState
*bs
,
6155 AioContext
*new_context
)
6157 BdrvAioNotifier
*ban
, *ban_tmp
;
6159 if (bs
->quiesce_counter
) {
6160 aio_disable_external(new_context
);
6163 bs
->aio_context
= new_context
;
6165 if (bs
->drv
&& bs
->drv
->bdrv_attach_aio_context
) {
6166 bs
->drv
->bdrv_attach_aio_context(bs
, new_context
);
6169 assert(!bs
->walking_aio_notifiers
);
6170 bs
->walking_aio_notifiers
= true;
6171 QLIST_FOREACH_SAFE(ban
, &bs
->aio_notifiers
, list
, ban_tmp
) {
6173 bdrv_do_remove_aio_context_notifier(ban
);
6175 ban
->attached_aio_context(new_context
, ban
->opaque
);
6178 bs
->walking_aio_notifiers
= false;
6182 * Changes the AioContext used for fd handlers, timers, and BHs by this
6183 * BlockDriverState and all its children and parents.
6185 * Must be called from the main AioContext.
6187 * The caller must own the AioContext lock for the old AioContext of bs, but it
6188 * must not own the AioContext lock for new_context (unless new_context is the
6189 * same as the current context of bs).
6191 * @ignore will accumulate all visited BdrvChild object. The caller is
6192 * responsible for freeing the list afterwards.
6194 void bdrv_set_aio_context_ignore(BlockDriverState
*bs
,
6195 AioContext
*new_context
, GSList
**ignore
)
6197 AioContext
*old_context
= bdrv_get_aio_context(bs
);
6200 g_assert(qemu_get_current_aio_context() == qemu_get_aio_context());
6202 if (old_context
== new_context
) {
6206 bdrv_drained_begin(bs
);
6208 QLIST_FOREACH(child
, &bs
->children
, next
) {
6209 if (g_slist_find(*ignore
, child
)) {
6212 *ignore
= g_slist_prepend(*ignore
, child
);
6213 bdrv_set_aio_context_ignore(child
->bs
, new_context
, ignore
);
6215 QLIST_FOREACH(child
, &bs
->parents
, next_parent
) {
6216 if (g_slist_find(*ignore
, child
)) {
6219 assert(child
->role
->set_aio_ctx
);
6220 *ignore
= g_slist_prepend(*ignore
, child
);
6221 child
->role
->set_aio_ctx(child
, new_context
, ignore
);
6224 bdrv_detach_aio_context(bs
);
6226 /* Acquire the new context, if necessary */
6227 if (qemu_get_aio_context() != new_context
) {
6228 aio_context_acquire(new_context
);
6231 bdrv_attach_aio_context(bs
, new_context
);
6234 * If this function was recursively called from
6235 * bdrv_set_aio_context_ignore(), there may be nodes in the
6236 * subtree that have not yet been moved to the new AioContext.
6237 * Release the old one so bdrv_drained_end() can poll them.
6239 if (qemu_get_aio_context() != old_context
) {
6240 aio_context_release(old_context
);
6243 bdrv_drained_end(bs
);
6245 if (qemu_get_aio_context() != old_context
) {
6246 aio_context_acquire(old_context
);
6248 if (qemu_get_aio_context() != new_context
) {
6249 aio_context_release(new_context
);
6253 static bool bdrv_parent_can_set_aio_context(BdrvChild
*c
, AioContext
*ctx
,
6254 GSList
**ignore
, Error
**errp
)
6256 if (g_slist_find(*ignore
, c
)) {
6259 *ignore
= g_slist_prepend(*ignore
, c
);
6261 /* A BdrvChildRole that doesn't handle AioContext changes cannot
6262 * tolerate any AioContext changes */
6263 if (!c
->role
->can_set_aio_ctx
) {
6264 char *user
= bdrv_child_user_desc(c
);
6265 error_setg(errp
, "Changing iothreads is not supported by %s", user
);
6269 if (!c
->role
->can_set_aio_ctx(c
, ctx
, ignore
, errp
)) {
6270 assert(!errp
|| *errp
);
6276 bool bdrv_child_can_set_aio_context(BdrvChild
*c
, AioContext
*ctx
,
6277 GSList
**ignore
, Error
**errp
)
6279 if (g_slist_find(*ignore
, c
)) {
6282 *ignore
= g_slist_prepend(*ignore
, c
);
6283 return bdrv_can_set_aio_context(c
->bs
, ctx
, ignore
, errp
);
6286 /* @ignore will accumulate all visited BdrvChild object. The caller is
6287 * responsible for freeing the list afterwards. */
6288 bool bdrv_can_set_aio_context(BlockDriverState
*bs
, AioContext
*ctx
,
6289 GSList
**ignore
, Error
**errp
)
6293 if (bdrv_get_aio_context(bs
) == ctx
) {
6297 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
6298 if (!bdrv_parent_can_set_aio_context(c
, ctx
, ignore
, errp
)) {
6302 QLIST_FOREACH(c
, &bs
->children
, next
) {
6303 if (!bdrv_child_can_set_aio_context(c
, ctx
, ignore
, errp
)) {
6311 int bdrv_child_try_set_aio_context(BlockDriverState
*bs
, AioContext
*ctx
,
6312 BdrvChild
*ignore_child
, Error
**errp
)
6317 ignore
= ignore_child
? g_slist_prepend(NULL
, ignore_child
) : NULL
;
6318 ret
= bdrv_can_set_aio_context(bs
, ctx
, &ignore
, errp
);
6319 g_slist_free(ignore
);
6325 ignore
= ignore_child
? g_slist_prepend(NULL
, ignore_child
) : NULL
;
6326 bdrv_set_aio_context_ignore(bs
, ctx
, &ignore
);
6327 g_slist_free(ignore
);
6332 int bdrv_try_set_aio_context(BlockDriverState
*bs
, AioContext
*ctx
,
6335 return bdrv_child_try_set_aio_context(bs
, ctx
, NULL
, errp
);
6338 void bdrv_add_aio_context_notifier(BlockDriverState
*bs
,
6339 void (*attached_aio_context
)(AioContext
*new_context
, void *opaque
),
6340 void (*detach_aio_context
)(void *opaque
), void *opaque
)
6342 BdrvAioNotifier
*ban
= g_new(BdrvAioNotifier
, 1);
6343 *ban
= (BdrvAioNotifier
){
6344 .attached_aio_context
= attached_aio_context
,
6345 .detach_aio_context
= detach_aio_context
,
6349 QLIST_INSERT_HEAD(&bs
->aio_notifiers
, ban
, list
);
6352 void bdrv_remove_aio_context_notifier(BlockDriverState
*bs
,
6353 void (*attached_aio_context
)(AioContext
*,
6355 void (*detach_aio_context
)(void *),
6358 BdrvAioNotifier
*ban
, *ban_next
;
6360 QLIST_FOREACH_SAFE(ban
, &bs
->aio_notifiers
, list
, ban_next
) {
6361 if (ban
->attached_aio_context
== attached_aio_context
&&
6362 ban
->detach_aio_context
== detach_aio_context
&&
6363 ban
->opaque
== opaque
&&
6364 ban
->deleted
== false)
6366 if (bs
->walking_aio_notifiers
) {
6367 ban
->deleted
= true;
6369 bdrv_do_remove_aio_context_notifier(ban
);
6378 int bdrv_amend_options(BlockDriverState
*bs
, QemuOpts
*opts
,
6379 BlockDriverAmendStatusCB
*status_cb
, void *cb_opaque
,
6383 error_setg(errp
, "Node is ejected");
6386 if (!bs
->drv
->bdrv_amend_options
) {
6387 error_setg(errp
, "Block driver '%s' does not support option amendment",
6388 bs
->drv
->format_name
);
6391 return bs
->drv
->bdrv_amend_options(bs
, opts
, status_cb
, cb_opaque
, errp
);
6395 * This function checks whether the given @to_replace is allowed to be
6396 * replaced by a node that always shows the same data as @bs. This is
6397 * used for example to verify whether the mirror job can replace
6398 * @to_replace by the target mirrored from @bs.
6399 * To be replaceable, @bs and @to_replace may either be guaranteed to
6400 * always show the same data (because they are only connected through
6401 * filters), or some driver may allow replacing one of its children
6402 * because it can guarantee that this child's data is not visible at
6403 * all (for example, for dissenting quorum children that have no other
6406 bool bdrv_recurse_can_replace(BlockDriverState
*bs
,
6407 BlockDriverState
*to_replace
)
6409 if (!bs
|| !bs
->drv
) {
6413 if (bs
== to_replace
) {
6417 /* See what the driver can do */
6418 if (bs
->drv
->bdrv_recurse_can_replace
) {
6419 return bs
->drv
->bdrv_recurse_can_replace(bs
, to_replace
);
6422 /* For filters without an own implementation, we can recurse on our own */
6423 if (bs
->drv
->is_filter
) {
6424 BdrvChild
*child
= bs
->file
?: bs
->backing
;
6425 return bdrv_recurse_can_replace(child
->bs
, to_replace
);
6433 * Check whether the given @node_name can be replaced by a node that
6434 * has the same data as @parent_bs. If so, return @node_name's BDS;
6437 * @node_name must be a (recursive) *child of @parent_bs (or this
6438 * function will return NULL).
6440 * The result (whether the node can be replaced or not) is only valid
6441 * for as long as no graph or permission changes occur.
6443 BlockDriverState
*check_to_replace_node(BlockDriverState
*parent_bs
,
6444 const char *node_name
, Error
**errp
)
6446 BlockDriverState
*to_replace_bs
= bdrv_find_node(node_name
);
6447 AioContext
*aio_context
;
6449 if (!to_replace_bs
) {
6450 error_setg(errp
, "Node name '%s' not found", node_name
);
6454 aio_context
= bdrv_get_aio_context(to_replace_bs
);
6455 aio_context_acquire(aio_context
);
6457 if (bdrv_op_is_blocked(to_replace_bs
, BLOCK_OP_TYPE_REPLACE
, errp
)) {
6458 to_replace_bs
= NULL
;
6462 /* We don't want arbitrary node of the BDS chain to be replaced only the top
6463 * most non filter in order to prevent data corruption.
6464 * Another benefit is that this tests exclude backing files which are
6465 * blocked by the backing blockers.
6467 if (!bdrv_recurse_can_replace(parent_bs
, to_replace_bs
)) {
6468 error_setg(errp
, "Cannot replace '%s' by a node mirrored from '%s', "
6469 "because it cannot be guaranteed that doing so would not "
6470 "lead to an abrupt change of visible data",
6471 node_name
, parent_bs
->node_name
);
6472 to_replace_bs
= NULL
;
6477 aio_context_release(aio_context
);
6478 return to_replace_bs
;
6482 * Iterates through the list of runtime option keys that are said to
6483 * be "strong" for a BDS. An option is called "strong" if it changes
6484 * a BDS's data. For example, the null block driver's "size" and
6485 * "read-zeroes" options are strong, but its "latency-ns" option is
6488 * If a key returned by this function ends with a dot, all options
6489 * starting with that prefix are strong.
6491 static const char *const *strong_options(BlockDriverState
*bs
,
6492 const char *const *curopt
)
6494 static const char *const global_options
[] = {
6495 "driver", "filename", NULL
6499 return &global_options
[0];
6503 if (curopt
== &global_options
[ARRAY_SIZE(global_options
) - 1] && bs
->drv
) {
6504 curopt
= bs
->drv
->strong_runtime_opts
;
6507 return (curopt
&& *curopt
) ? curopt
: NULL
;
6511 * Copies all strong runtime options from bs->options to the given
6512 * QDict. The set of strong option keys is determined by invoking
6515 * Returns true iff any strong option was present in bs->options (and
6516 * thus copied to the target QDict) with the exception of "filename"
6517 * and "driver". The caller is expected to use this value to decide
6518 * whether the existence of strong options prevents the generation of
6521 static bool append_strong_runtime_options(QDict
*d
, BlockDriverState
*bs
)
6523 bool found_any
= false;
6524 const char *const *option_name
= NULL
;
6530 while ((option_name
= strong_options(bs
, option_name
))) {
6531 bool option_given
= false;
6533 assert(strlen(*option_name
) > 0);
6534 if ((*option_name
)[strlen(*option_name
) - 1] != '.') {
6535 QObject
*entry
= qdict_get(bs
->options
, *option_name
);
6540 qdict_put_obj(d
, *option_name
, qobject_ref(entry
));
6541 option_given
= true;
6543 const QDictEntry
*entry
;
6544 for (entry
= qdict_first(bs
->options
); entry
;
6545 entry
= qdict_next(bs
->options
, entry
))
6547 if (strstart(qdict_entry_key(entry
), *option_name
, NULL
)) {
6548 qdict_put_obj(d
, qdict_entry_key(entry
),
6549 qobject_ref(qdict_entry_value(entry
)));
6550 option_given
= true;
6555 /* While "driver" and "filename" need to be included in a JSON filename,
6556 * their existence does not prohibit generation of a plain filename. */
6557 if (!found_any
&& option_given
&&
6558 strcmp(*option_name
, "driver") && strcmp(*option_name
, "filename"))
6564 if (!qdict_haskey(d
, "driver")) {
6565 /* Drivers created with bdrv_new_open_driver() may not have a
6566 * @driver option. Add it here. */
6567 qdict_put_str(d
, "driver", bs
->drv
->format_name
);
6573 /* Note: This function may return false positives; it may return true
6574 * even if opening the backing file specified by bs's image header
6575 * would result in exactly bs->backing. */
6576 static bool bdrv_backing_overridden(BlockDriverState
*bs
)
6579 return strcmp(bs
->auto_backing_file
,
6580 bs
->backing
->bs
->filename
);
6582 /* No backing BDS, so if the image header reports any backing
6583 * file, it must have been suppressed */
6584 return bs
->auto_backing_file
[0] != '\0';
6588 /* Updates the following BDS fields:
6589 * - exact_filename: A filename which may be used for opening a block device
6590 * which (mostly) equals the given BDS (even without any
6591 * other options; so reading and writing must return the same
6592 * results, but caching etc. may be different)
6593 * - full_open_options: Options which, when given when opening a block device
6594 * (without a filename), result in a BDS (mostly)
6595 * equalling the given one
6596 * - filename: If exact_filename is set, it is copied here. Otherwise,
6597 * full_open_options is converted to a JSON object, prefixed with
6598 * "json:" (for use through the JSON pseudo protocol) and put here.
6600 void bdrv_refresh_filename(BlockDriverState
*bs
)
6602 BlockDriver
*drv
= bs
->drv
;
6605 bool backing_overridden
;
6606 bool generate_json_filename
; /* Whether our default implementation should
6607 fill exact_filename (false) or not (true) */
6613 /* This BDS's file name may depend on any of its children's file names, so
6614 * refresh those first */
6615 QLIST_FOREACH(child
, &bs
->children
, next
) {
6616 bdrv_refresh_filename(child
->bs
);
6620 /* For implicit nodes, just copy everything from the single child */
6621 child
= QLIST_FIRST(&bs
->children
);
6622 assert(QLIST_NEXT(child
, next
) == NULL
);
6624 pstrcpy(bs
->exact_filename
, sizeof(bs
->exact_filename
),
6625 child
->bs
->exact_filename
);
6626 pstrcpy(bs
->filename
, sizeof(bs
->filename
), child
->bs
->filename
);
6628 qobject_unref(bs
->full_open_options
);
6629 bs
->full_open_options
= qobject_ref(child
->bs
->full_open_options
);
6634 backing_overridden
= bdrv_backing_overridden(bs
);
6636 if (bs
->open_flags
& BDRV_O_NO_IO
) {
6637 /* Without I/O, the backing file does not change anything.
6638 * Therefore, in such a case (primarily qemu-img), we can
6639 * pretend the backing file has not been overridden even if
6640 * it technically has been. */
6641 backing_overridden
= false;
6644 /* Gather the options QDict */
6646 generate_json_filename
= append_strong_runtime_options(opts
, bs
);
6647 generate_json_filename
|= backing_overridden
;
6649 if (drv
->bdrv_gather_child_options
) {
6650 /* Some block drivers may not want to present all of their children's
6651 * options, or name them differently from BdrvChild.name */
6652 drv
->bdrv_gather_child_options(bs
, opts
, backing_overridden
);
6654 QLIST_FOREACH(child
, &bs
->children
, next
) {
6655 if (child
->role
== &child_backing
&& !backing_overridden
) {
6656 /* We can skip the backing BDS if it has not been overridden */
6660 qdict_put(opts
, child
->name
,
6661 qobject_ref(child
->bs
->full_open_options
));
6664 if (backing_overridden
&& !bs
->backing
) {
6665 /* Force no backing file */
6666 qdict_put_null(opts
, "backing");
6670 qobject_unref(bs
->full_open_options
);
6671 bs
->full_open_options
= opts
;
6673 if (drv
->bdrv_refresh_filename
) {
6674 /* Obsolete information is of no use here, so drop the old file name
6675 * information before refreshing it */
6676 bs
->exact_filename
[0] = '\0';
6678 drv
->bdrv_refresh_filename(bs
);
6679 } else if (bs
->file
) {
6680 /* Try to reconstruct valid information from the underlying file */
6682 bs
->exact_filename
[0] = '\0';
6685 * We can use the underlying file's filename if:
6686 * - it has a filename,
6687 * - the file is a protocol BDS, and
6688 * - opening that file (as this BDS's format) will automatically create
6689 * the BDS tree we have right now, that is:
6690 * - the user did not significantly change this BDS's behavior with
6691 * some explicit (strong) options
6692 * - no non-file child of this BDS has been overridden by the user
6693 * Both of these conditions are represented by generate_json_filename.
6695 if (bs
->file
->bs
->exact_filename
[0] &&
6696 bs
->file
->bs
->drv
->bdrv_file_open
&&
6697 !generate_json_filename
)
6699 strcpy(bs
->exact_filename
, bs
->file
->bs
->exact_filename
);
6703 if (bs
->exact_filename
[0]) {
6704 pstrcpy(bs
->filename
, sizeof(bs
->filename
), bs
->exact_filename
);
6706 QString
*json
= qobject_to_json(QOBJECT(bs
->full_open_options
));
6707 snprintf(bs
->filename
, sizeof(bs
->filename
), "json:%s",
6708 qstring_get_str(json
));
6709 qobject_unref(json
);
6713 char *bdrv_dirname(BlockDriverState
*bs
, Error
**errp
)
6715 BlockDriver
*drv
= bs
->drv
;
6718 error_setg(errp
, "Node '%s' is ejected", bs
->node_name
);
6722 if (drv
->bdrv_dirname
) {
6723 return drv
->bdrv_dirname(bs
, errp
);
6727 return bdrv_dirname(bs
->file
->bs
, errp
);
6730 bdrv_refresh_filename(bs
);
6731 if (bs
->exact_filename
[0] != '\0') {
6732 return path_combine(bs
->exact_filename
, "");
6735 error_setg(errp
, "Cannot generate a base directory for %s nodes",
6741 * Hot add/remove a BDS's child. So the user can take a child offline when
6742 * it is broken and take a new child online
6744 void bdrv_add_child(BlockDriverState
*parent_bs
, BlockDriverState
*child_bs
,
6748 if (!parent_bs
->drv
|| !parent_bs
->drv
->bdrv_add_child
) {
6749 error_setg(errp
, "The node %s does not support adding a child",
6750 bdrv_get_device_or_node_name(parent_bs
));
6754 if (!QLIST_EMPTY(&child_bs
->parents
)) {
6755 error_setg(errp
, "The node %s already has a parent",
6756 child_bs
->node_name
);
6760 parent_bs
->drv
->bdrv_add_child(parent_bs
, child_bs
, errp
);
6763 void bdrv_del_child(BlockDriverState
*parent_bs
, BdrvChild
*child
, Error
**errp
)
6767 if (!parent_bs
->drv
|| !parent_bs
->drv
->bdrv_del_child
) {
6768 error_setg(errp
, "The node %s does not support removing a child",
6769 bdrv_get_device_or_node_name(parent_bs
));
6773 QLIST_FOREACH(tmp
, &parent_bs
->children
, next
) {
6780 error_setg(errp
, "The node %s does not have a child named %s",
6781 bdrv_get_device_or_node_name(parent_bs
),
6782 bdrv_get_device_or_node_name(child
->bs
));
6786 parent_bs
->drv
->bdrv_del_child(parent_bs
, child
, errp
);