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
->filename
, cco
->opts
, &local_err
);
487 error_propagate(&cco
->err
, local_err
);
491 int bdrv_create(BlockDriver
*drv
, const char* filename
,
492 QemuOpts
*opts
, Error
**errp
)
499 .filename
= g_strdup(filename
),
505 if (!drv
->bdrv_co_create_opts
) {
506 error_setg(errp
, "Driver '%s' does not support image creation", drv
->format_name
);
511 if (qemu_in_coroutine()) {
512 /* Fast-path if already in coroutine context */
513 bdrv_create_co_entry(&cco
);
515 co
= qemu_coroutine_create(bdrv_create_co_entry
, &cco
);
516 qemu_coroutine_enter(co
);
517 while (cco
.ret
== NOT_DONE
) {
518 aio_poll(qemu_get_aio_context(), true);
525 error_propagate(errp
, cco
.err
);
527 error_setg_errno(errp
, -ret
, "Could not create image");
532 g_free(cco
.filename
);
537 * Helper function for bdrv_create_file_fallback(): Resize @blk to at
538 * least the given @minimum_size.
540 * On success, return @blk's actual length.
541 * Otherwise, return -errno.
543 static int64_t create_file_fallback_truncate(BlockBackend
*blk
,
544 int64_t minimum_size
, Error
**errp
)
546 Error
*local_err
= NULL
;
550 ret
= blk_truncate(blk
, minimum_size
, false, PREALLOC_MODE_OFF
, &local_err
);
551 if (ret
< 0 && ret
!= -ENOTSUP
) {
552 error_propagate(errp
, local_err
);
556 size
= blk_getlength(blk
);
558 error_free(local_err
);
559 error_setg_errno(errp
, -size
,
560 "Failed to inquire the new image file's length");
564 if (size
< minimum_size
) {
565 /* Need to grow the image, but we failed to do that */
566 error_propagate(errp
, local_err
);
570 error_free(local_err
);
577 * Helper function for bdrv_create_file_fallback(): Zero the first
578 * sector to remove any potentially pre-existing image header.
580 static int create_file_fallback_zero_first_sector(BlockBackend
*blk
,
581 int64_t current_size
,
584 int64_t bytes_to_clear
;
587 bytes_to_clear
= MIN(current_size
, BDRV_SECTOR_SIZE
);
588 if (bytes_to_clear
) {
589 ret
= blk_pwrite_zeroes(blk
, 0, bytes_to_clear
, BDRV_REQ_MAY_UNMAP
);
591 error_setg_errno(errp
, -ret
,
592 "Failed to clear the new image's first sector");
600 static int bdrv_create_file_fallback(const char *filename
, BlockDriver
*drv
,
601 QemuOpts
*opts
, Error
**errp
)
607 PreallocMode prealloc
;
608 Error
*local_err
= NULL
;
611 size
= qemu_opt_get_size_del(opts
, BLOCK_OPT_SIZE
, 0);
612 buf
= qemu_opt_get_del(opts
, BLOCK_OPT_PREALLOC
);
613 prealloc
= qapi_enum_parse(&PreallocMode_lookup
, buf
,
614 PREALLOC_MODE_OFF
, &local_err
);
617 error_propagate(errp
, local_err
);
621 if (prealloc
!= PREALLOC_MODE_OFF
) {
622 error_setg(errp
, "Unsupported preallocation mode '%s'",
623 PreallocMode_str(prealloc
));
627 options
= qdict_new();
628 qdict_put_str(options
, "driver", drv
->format_name
);
630 blk
= blk_new_open(filename
, NULL
, options
,
631 BDRV_O_RDWR
| BDRV_O_RESIZE
, errp
);
633 error_prepend(errp
, "Protocol driver '%s' does not support image "
634 "creation, and opening the image failed: ",
639 size
= create_file_fallback_truncate(blk
, size
, errp
);
645 ret
= create_file_fallback_zero_first_sector(blk
, size
, errp
);
656 int bdrv_create_file(const char *filename
, QemuOpts
*opts
, Error
**errp
)
660 drv
= bdrv_find_protocol(filename
, true, errp
);
665 if (drv
->bdrv_co_create_opts
) {
666 return bdrv_create(drv
, filename
, opts
, errp
);
668 return bdrv_create_file_fallback(filename
, drv
, opts
, errp
);
672 int coroutine_fn
bdrv_co_delete_file(BlockDriverState
*bs
, Error
**errp
)
674 Error
*local_err
= NULL
;
680 error_setg(errp
, "Block node '%s' is not opened", bs
->filename
);
684 if (!bs
->drv
->bdrv_co_delete_file
) {
685 error_setg(errp
, "Driver '%s' does not support image deletion",
686 bs
->drv
->format_name
);
690 ret
= bs
->drv
->bdrv_co_delete_file(bs
, &local_err
);
692 error_propagate(errp
, local_err
);
699 * Try to get @bs's logical and physical block size.
700 * On success, store them in @bsz struct and return 0.
701 * On failure return -errno.
702 * @bs must not be empty.
704 int bdrv_probe_blocksizes(BlockDriverState
*bs
, BlockSizes
*bsz
)
706 BlockDriver
*drv
= bs
->drv
;
708 if (drv
&& drv
->bdrv_probe_blocksizes
) {
709 return drv
->bdrv_probe_blocksizes(bs
, bsz
);
710 } else if (drv
&& drv
->is_filter
&& bs
->file
) {
711 return bdrv_probe_blocksizes(bs
->file
->bs
, bsz
);
718 * Try to get @bs's geometry (cyls, heads, sectors).
719 * On success, store them in @geo struct and return 0.
720 * On failure return -errno.
721 * @bs must not be empty.
723 int bdrv_probe_geometry(BlockDriverState
*bs
, HDGeometry
*geo
)
725 BlockDriver
*drv
= bs
->drv
;
727 if (drv
&& drv
->bdrv_probe_geometry
) {
728 return drv
->bdrv_probe_geometry(bs
, geo
);
729 } else if (drv
&& drv
->is_filter
&& bs
->file
) {
730 return bdrv_probe_geometry(bs
->file
->bs
, geo
);
737 * Create a uniquely-named empty temporary file.
738 * Return 0 upon success, otherwise a negative errno value.
740 int get_tmp_filename(char *filename
, int size
)
743 char temp_dir
[MAX_PATH
];
744 /* GetTempFileName requires that its output buffer (4th param)
745 have length MAX_PATH or greater. */
746 assert(size
>= MAX_PATH
);
747 return (GetTempPath(MAX_PATH
, temp_dir
)
748 && GetTempFileName(temp_dir
, "qem", 0, filename
)
749 ? 0 : -GetLastError());
753 tmpdir
= getenv("TMPDIR");
757 if (snprintf(filename
, size
, "%s/vl.XXXXXX", tmpdir
) >= size
) {
760 fd
= mkstemp(filename
);
764 if (close(fd
) != 0) {
773 * Detect host devices. By convention, /dev/cdrom[N] is always
774 * recognized as a host CDROM.
776 static BlockDriver
*find_hdev_driver(const char *filename
)
778 int score_max
= 0, score
;
779 BlockDriver
*drv
= NULL
, *d
;
781 QLIST_FOREACH(d
, &bdrv_drivers
, list
) {
782 if (d
->bdrv_probe_device
) {
783 score
= d
->bdrv_probe_device(filename
);
784 if (score
> score_max
) {
794 static BlockDriver
*bdrv_do_find_protocol(const char *protocol
)
798 QLIST_FOREACH(drv1
, &bdrv_drivers
, list
) {
799 if (drv1
->protocol_name
&& !strcmp(drv1
->protocol_name
, protocol
)) {
807 BlockDriver
*bdrv_find_protocol(const char *filename
,
808 bool allow_protocol_prefix
,
817 /* TODO Drivers without bdrv_file_open must be specified explicitly */
820 * XXX(hch): we really should not let host device detection
821 * override an explicit protocol specification, but moving this
822 * later breaks access to device names with colons in them.
823 * Thanks to the brain-dead persistent naming schemes on udev-
824 * based Linux systems those actually are quite common.
826 drv1
= find_hdev_driver(filename
);
831 if (!path_has_protocol(filename
) || !allow_protocol_prefix
) {
835 p
= strchr(filename
, ':');
838 if (len
> sizeof(protocol
) - 1)
839 len
= sizeof(protocol
) - 1;
840 memcpy(protocol
, filename
, len
);
841 protocol
[len
] = '\0';
843 drv1
= bdrv_do_find_protocol(protocol
);
848 for (i
= 0; i
< (int)ARRAY_SIZE(block_driver_modules
); ++i
) {
849 if (block_driver_modules
[i
].protocol_name
&&
850 !strcmp(block_driver_modules
[i
].protocol_name
, protocol
)) {
851 block_module_load_one(block_driver_modules
[i
].library_name
);
856 drv1
= bdrv_do_find_protocol(protocol
);
858 error_setg(errp
, "Unknown protocol '%s'", protocol
);
864 * Guess image format by probing its contents.
865 * This is not a good idea when your image is raw (CVE-2008-2004), but
866 * we do it anyway for backward compatibility.
868 * @buf contains the image's first @buf_size bytes.
869 * @buf_size is the buffer size in bytes (generally BLOCK_PROBE_BUF_SIZE,
870 * but can be smaller if the image file is smaller)
871 * @filename is its filename.
873 * For all block drivers, call the bdrv_probe() method to get its
875 * Return the first block driver with the highest probing score.
877 BlockDriver
*bdrv_probe_all(const uint8_t *buf
, int buf_size
,
878 const char *filename
)
880 int score_max
= 0, score
;
881 BlockDriver
*drv
= NULL
, *d
;
883 QLIST_FOREACH(d
, &bdrv_drivers
, list
) {
885 score
= d
->bdrv_probe(buf
, buf_size
, filename
);
886 if (score
> score_max
) {
896 static int find_image_format(BlockBackend
*file
, const char *filename
,
897 BlockDriver
**pdrv
, Error
**errp
)
900 uint8_t buf
[BLOCK_PROBE_BUF_SIZE
];
903 /* Return the raw BlockDriver * to scsi-generic devices or empty drives */
904 if (blk_is_sg(file
) || !blk_is_inserted(file
) || blk_getlength(file
) == 0) {
909 ret
= blk_pread(file
, 0, buf
, sizeof(buf
));
911 error_setg_errno(errp
, -ret
, "Could not read image for determining its "
917 drv
= bdrv_probe_all(buf
, ret
, filename
);
919 error_setg(errp
, "Could not determine image format: No compatible "
928 * Set the current 'total_sectors' value
929 * Return 0 on success, -errno on error.
931 int refresh_total_sectors(BlockDriverState
*bs
, int64_t hint
)
933 BlockDriver
*drv
= bs
->drv
;
939 /* Do not attempt drv->bdrv_getlength() on scsi-generic devices */
943 /* query actual device if possible, otherwise just trust the hint */
944 if (drv
->bdrv_getlength
) {
945 int64_t length
= drv
->bdrv_getlength(bs
);
949 hint
= DIV_ROUND_UP(length
, BDRV_SECTOR_SIZE
);
952 bs
->total_sectors
= hint
;
957 * Combines a QDict of new block driver @options with any missing options taken
958 * from @old_options, so that leaving out an option defaults to its old value.
960 static void bdrv_join_options(BlockDriverState
*bs
, QDict
*options
,
963 if (bs
->drv
&& bs
->drv
->bdrv_join_options
) {
964 bs
->drv
->bdrv_join_options(options
, old_options
);
966 qdict_join(options
, old_options
, false);
970 static BlockdevDetectZeroesOptions
bdrv_parse_detect_zeroes(QemuOpts
*opts
,
974 Error
*local_err
= NULL
;
975 char *value
= qemu_opt_get_del(opts
, "detect-zeroes");
976 BlockdevDetectZeroesOptions detect_zeroes
=
977 qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup
, value
,
978 BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF
, &local_err
);
981 error_propagate(errp
, local_err
);
982 return detect_zeroes
;
985 if (detect_zeroes
== BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP
&&
986 !(open_flags
& BDRV_O_UNMAP
))
988 error_setg(errp
, "setting detect-zeroes to unmap is not allowed "
989 "without setting discard operation to unmap");
992 return detect_zeroes
;
996 * Set open flags for aio engine
998 * Return 0 on success, -1 if the engine specified is invalid
1000 int bdrv_parse_aio(const char *mode
, int *flags
)
1002 if (!strcmp(mode
, "threads")) {
1003 /* do nothing, default */
1004 } else if (!strcmp(mode
, "native")) {
1005 *flags
|= BDRV_O_NATIVE_AIO
;
1006 #ifdef CONFIG_LINUX_IO_URING
1007 } else if (!strcmp(mode
, "io_uring")) {
1008 *flags
|= BDRV_O_IO_URING
;
1018 * Set open flags for a given discard mode
1020 * Return 0 on success, -1 if the discard mode was invalid.
1022 int bdrv_parse_discard_flags(const char *mode
, int *flags
)
1024 *flags
&= ~BDRV_O_UNMAP
;
1026 if (!strcmp(mode
, "off") || !strcmp(mode
, "ignore")) {
1028 } else if (!strcmp(mode
, "on") || !strcmp(mode
, "unmap")) {
1029 *flags
|= BDRV_O_UNMAP
;
1038 * Set open flags for a given cache mode
1040 * Return 0 on success, -1 if the cache mode was invalid.
1042 int bdrv_parse_cache_mode(const char *mode
, int *flags
, bool *writethrough
)
1044 *flags
&= ~BDRV_O_CACHE_MASK
;
1046 if (!strcmp(mode
, "off") || !strcmp(mode
, "none")) {
1047 *writethrough
= false;
1048 *flags
|= BDRV_O_NOCACHE
;
1049 } else if (!strcmp(mode
, "directsync")) {
1050 *writethrough
= true;
1051 *flags
|= BDRV_O_NOCACHE
;
1052 } else if (!strcmp(mode
, "writeback")) {
1053 *writethrough
= false;
1054 } else if (!strcmp(mode
, "unsafe")) {
1055 *writethrough
= false;
1056 *flags
|= BDRV_O_NO_FLUSH
;
1057 } else if (!strcmp(mode
, "writethrough")) {
1058 *writethrough
= true;
1066 static char *bdrv_child_get_parent_desc(BdrvChild
*c
)
1068 BlockDriverState
*parent
= c
->opaque
;
1069 return g_strdup(bdrv_get_device_or_node_name(parent
));
1072 static void bdrv_child_cb_drained_begin(BdrvChild
*child
)
1074 BlockDriverState
*bs
= child
->opaque
;
1075 bdrv_do_drained_begin_quiesce(bs
, NULL
, false);
1078 static bool bdrv_child_cb_drained_poll(BdrvChild
*child
)
1080 BlockDriverState
*bs
= child
->opaque
;
1081 return bdrv_drain_poll(bs
, false, NULL
, false);
1084 static void bdrv_child_cb_drained_end(BdrvChild
*child
,
1085 int *drained_end_counter
)
1087 BlockDriverState
*bs
= child
->opaque
;
1088 bdrv_drained_end_no_poll(bs
, drained_end_counter
);
1091 static void bdrv_child_cb_attach(BdrvChild
*child
)
1093 BlockDriverState
*bs
= child
->opaque
;
1094 bdrv_apply_subtree_drain(child
, bs
);
1097 static void bdrv_child_cb_detach(BdrvChild
*child
)
1099 BlockDriverState
*bs
= child
->opaque
;
1100 bdrv_unapply_subtree_drain(child
, bs
);
1103 static int bdrv_child_cb_inactivate(BdrvChild
*child
)
1105 BlockDriverState
*bs
= child
->opaque
;
1106 assert(bs
->open_flags
& BDRV_O_INACTIVE
);
1110 static bool bdrv_child_cb_can_set_aio_ctx(BdrvChild
*child
, AioContext
*ctx
,
1111 GSList
**ignore
, Error
**errp
)
1113 BlockDriverState
*bs
= child
->opaque
;
1114 return bdrv_can_set_aio_context(bs
, ctx
, ignore
, errp
);
1117 static void bdrv_child_cb_set_aio_ctx(BdrvChild
*child
, AioContext
*ctx
,
1120 BlockDriverState
*bs
= child
->opaque
;
1121 return bdrv_set_aio_context_ignore(bs
, ctx
, ignore
);
1125 * Returns the options and flags that a temporary snapshot should get, based on
1126 * the originally requested flags (the originally requested image will have
1127 * flags like a backing file)
1129 static void bdrv_temp_snapshot_options(int *child_flags
, QDict
*child_options
,
1130 int parent_flags
, QDict
*parent_options
)
1132 *child_flags
= (parent_flags
& ~BDRV_O_SNAPSHOT
) | BDRV_O_TEMPORARY
;
1134 /* For temporary files, unconditional cache=unsafe is fine */
1135 qdict_set_default_str(child_options
, BDRV_OPT_CACHE_DIRECT
, "off");
1136 qdict_set_default_str(child_options
, BDRV_OPT_CACHE_NO_FLUSH
, "on");
1138 /* Copy the read-only and discard options from the parent */
1139 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_READ_ONLY
);
1140 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_DISCARD
);
1142 /* aio=native doesn't work for cache.direct=off, so disable it for the
1143 * temporary snapshot */
1144 *child_flags
&= ~BDRV_O_NATIVE_AIO
;
1148 * Returns the options and flags that bs->file should get if a protocol driver
1149 * is expected, based on the given options and flags for the parent BDS
1151 static void bdrv_inherited_options(int *child_flags
, QDict
*child_options
,
1152 int parent_flags
, QDict
*parent_options
)
1154 int flags
= parent_flags
;
1156 /* Enable protocol handling, disable format probing for bs->file */
1157 flags
|= BDRV_O_PROTOCOL
;
1159 /* If the cache mode isn't explicitly set, inherit direct and no-flush from
1161 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_CACHE_DIRECT
);
1162 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_CACHE_NO_FLUSH
);
1163 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_FORCE_SHARE
);
1165 /* Inherit the read-only option from the parent if it's not set */
1166 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_READ_ONLY
);
1167 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_AUTO_READ_ONLY
);
1169 /* Our block drivers take care to send flushes and respect unmap policy,
1170 * so we can default to enable both on lower layers regardless of the
1171 * corresponding parent options. */
1172 qdict_set_default_str(child_options
, BDRV_OPT_DISCARD
, "unmap");
1174 /* Clear flags that only apply to the top layer */
1175 flags
&= ~(BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
| BDRV_O_COPY_ON_READ
|
1178 *child_flags
= flags
;
1181 const BdrvChildRole child_file
= {
1182 .parent_is_bds
= true,
1183 .get_parent_desc
= bdrv_child_get_parent_desc
,
1184 .inherit_options
= bdrv_inherited_options
,
1185 .drained_begin
= bdrv_child_cb_drained_begin
,
1186 .drained_poll
= bdrv_child_cb_drained_poll
,
1187 .drained_end
= bdrv_child_cb_drained_end
,
1188 .attach
= bdrv_child_cb_attach
,
1189 .detach
= bdrv_child_cb_detach
,
1190 .inactivate
= bdrv_child_cb_inactivate
,
1191 .can_set_aio_ctx
= bdrv_child_cb_can_set_aio_ctx
,
1192 .set_aio_ctx
= bdrv_child_cb_set_aio_ctx
,
1196 * Returns the options and flags that bs->file should get if the use of formats
1197 * (and not only protocols) is permitted for it, based on the given options and
1198 * flags for the parent BDS
1200 static void bdrv_inherited_fmt_options(int *child_flags
, QDict
*child_options
,
1201 int parent_flags
, QDict
*parent_options
)
1203 child_file
.inherit_options(child_flags
, child_options
,
1204 parent_flags
, parent_options
);
1206 *child_flags
&= ~(BDRV_O_PROTOCOL
| BDRV_O_NO_IO
);
1209 const BdrvChildRole child_format
= {
1210 .parent_is_bds
= true,
1211 .get_parent_desc
= bdrv_child_get_parent_desc
,
1212 .inherit_options
= bdrv_inherited_fmt_options
,
1213 .drained_begin
= bdrv_child_cb_drained_begin
,
1214 .drained_poll
= bdrv_child_cb_drained_poll
,
1215 .drained_end
= bdrv_child_cb_drained_end
,
1216 .attach
= bdrv_child_cb_attach
,
1217 .detach
= bdrv_child_cb_detach
,
1218 .inactivate
= bdrv_child_cb_inactivate
,
1219 .can_set_aio_ctx
= bdrv_child_cb_can_set_aio_ctx
,
1220 .set_aio_ctx
= bdrv_child_cb_set_aio_ctx
,
1223 static void bdrv_backing_attach(BdrvChild
*c
)
1225 BlockDriverState
*parent
= c
->opaque
;
1226 BlockDriverState
*backing_hd
= c
->bs
;
1228 assert(!parent
->backing_blocker
);
1229 error_setg(&parent
->backing_blocker
,
1230 "node is used as backing hd of '%s'",
1231 bdrv_get_device_or_node_name(parent
));
1233 bdrv_refresh_filename(backing_hd
);
1235 parent
->open_flags
&= ~BDRV_O_NO_BACKING
;
1236 pstrcpy(parent
->backing_file
, sizeof(parent
->backing_file
),
1237 backing_hd
->filename
);
1238 pstrcpy(parent
->backing_format
, sizeof(parent
->backing_format
),
1239 backing_hd
->drv
? backing_hd
->drv
->format_name
: "");
1241 bdrv_op_block_all(backing_hd
, parent
->backing_blocker
);
1242 /* Otherwise we won't be able to commit or stream */
1243 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_COMMIT_TARGET
,
1244 parent
->backing_blocker
);
1245 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_STREAM
,
1246 parent
->backing_blocker
);
1248 * We do backup in 3 ways:
1250 * The target bs is new opened, and the source is top BDS
1251 * 2. blockdev backup
1252 * Both the source and the target are top BDSes.
1253 * 3. internal backup(used for block replication)
1254 * Both the source and the target are backing file
1256 * In case 1 and 2, neither the source nor the target is the backing file.
1257 * In case 3, we will block the top BDS, so there is only one block job
1258 * for the top BDS and its backing chain.
1260 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_BACKUP_SOURCE
,
1261 parent
->backing_blocker
);
1262 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_BACKUP_TARGET
,
1263 parent
->backing_blocker
);
1265 bdrv_child_cb_attach(c
);
1268 static void bdrv_backing_detach(BdrvChild
*c
)
1270 BlockDriverState
*parent
= c
->opaque
;
1272 assert(parent
->backing_blocker
);
1273 bdrv_op_unblock_all(c
->bs
, parent
->backing_blocker
);
1274 error_free(parent
->backing_blocker
);
1275 parent
->backing_blocker
= NULL
;
1277 bdrv_child_cb_detach(c
);
1281 * Returns the options and flags that bs->backing should get, based on the
1282 * given options and flags for the parent BDS
1284 static void bdrv_backing_options(int *child_flags
, QDict
*child_options
,
1285 int parent_flags
, QDict
*parent_options
)
1287 int flags
= parent_flags
;
1289 /* The cache mode is inherited unmodified for backing files; except WCE,
1290 * which is only applied on the top level (BlockBackend) */
1291 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_CACHE_DIRECT
);
1292 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_CACHE_NO_FLUSH
);
1293 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_FORCE_SHARE
);
1295 /* backing files always opened read-only */
1296 qdict_set_default_str(child_options
, BDRV_OPT_READ_ONLY
, "on");
1297 qdict_set_default_str(child_options
, BDRV_OPT_AUTO_READ_ONLY
, "off");
1298 flags
&= ~BDRV_O_COPY_ON_READ
;
1300 /* snapshot=on is handled on the top layer */
1301 flags
&= ~(BDRV_O_SNAPSHOT
| BDRV_O_TEMPORARY
);
1303 *child_flags
= flags
;
1306 static int bdrv_backing_update_filename(BdrvChild
*c
, BlockDriverState
*base
,
1307 const char *filename
, Error
**errp
)
1309 BlockDriverState
*parent
= c
->opaque
;
1310 bool read_only
= bdrv_is_read_only(parent
);
1314 ret
= bdrv_reopen_set_read_only(parent
, false, errp
);
1320 ret
= bdrv_change_backing_file(parent
, filename
,
1321 base
->drv
? base
->drv
->format_name
: "");
1323 error_setg_errno(errp
, -ret
, "Could not update backing file link");
1327 bdrv_reopen_set_read_only(parent
, true, NULL
);
1333 const BdrvChildRole child_backing
= {
1334 .parent_is_bds
= true,
1335 .get_parent_desc
= bdrv_child_get_parent_desc
,
1336 .attach
= bdrv_backing_attach
,
1337 .detach
= bdrv_backing_detach
,
1338 .inherit_options
= bdrv_backing_options
,
1339 .drained_begin
= bdrv_child_cb_drained_begin
,
1340 .drained_poll
= bdrv_child_cb_drained_poll
,
1341 .drained_end
= bdrv_child_cb_drained_end
,
1342 .inactivate
= bdrv_child_cb_inactivate
,
1343 .update_filename
= bdrv_backing_update_filename
,
1344 .can_set_aio_ctx
= bdrv_child_cb_can_set_aio_ctx
,
1345 .set_aio_ctx
= bdrv_child_cb_set_aio_ctx
,
1348 static int bdrv_open_flags(BlockDriverState
*bs
, int flags
)
1350 int open_flags
= flags
;
1353 * Clear flags that are internal to the block layer before opening the
1356 open_flags
&= ~(BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
| BDRV_O_PROTOCOL
);
1361 static void update_flags_from_options(int *flags
, QemuOpts
*opts
)
1363 *flags
&= ~(BDRV_O_CACHE_MASK
| BDRV_O_RDWR
| BDRV_O_AUTO_RDONLY
);
1365 if (qemu_opt_get_bool_del(opts
, BDRV_OPT_CACHE_NO_FLUSH
, false)) {
1366 *flags
|= BDRV_O_NO_FLUSH
;
1369 if (qemu_opt_get_bool_del(opts
, BDRV_OPT_CACHE_DIRECT
, false)) {
1370 *flags
|= BDRV_O_NOCACHE
;
1373 if (!qemu_opt_get_bool_del(opts
, BDRV_OPT_READ_ONLY
, false)) {
1374 *flags
|= BDRV_O_RDWR
;
1377 if (qemu_opt_get_bool_del(opts
, BDRV_OPT_AUTO_READ_ONLY
, false)) {
1378 *flags
|= BDRV_O_AUTO_RDONLY
;
1382 static void update_options_from_flags(QDict
*options
, int flags
)
1384 if (!qdict_haskey(options
, BDRV_OPT_CACHE_DIRECT
)) {
1385 qdict_put_bool(options
, BDRV_OPT_CACHE_DIRECT
, flags
& BDRV_O_NOCACHE
);
1387 if (!qdict_haskey(options
, BDRV_OPT_CACHE_NO_FLUSH
)) {
1388 qdict_put_bool(options
, BDRV_OPT_CACHE_NO_FLUSH
,
1389 flags
& BDRV_O_NO_FLUSH
);
1391 if (!qdict_haskey(options
, BDRV_OPT_READ_ONLY
)) {
1392 qdict_put_bool(options
, BDRV_OPT_READ_ONLY
, !(flags
& BDRV_O_RDWR
));
1394 if (!qdict_haskey(options
, BDRV_OPT_AUTO_READ_ONLY
)) {
1395 qdict_put_bool(options
, BDRV_OPT_AUTO_READ_ONLY
,
1396 flags
& BDRV_O_AUTO_RDONLY
);
1400 static void bdrv_assign_node_name(BlockDriverState
*bs
,
1401 const char *node_name
,
1404 char *gen_node_name
= NULL
;
1407 node_name
= gen_node_name
= id_generate(ID_BLOCK
);
1408 } else if (!id_wellformed(node_name
)) {
1410 * Check for empty string or invalid characters, but not if it is
1411 * generated (generated names use characters not available to the user)
1413 error_setg(errp
, "Invalid node name");
1417 /* takes care of avoiding namespaces collisions */
1418 if (blk_by_name(node_name
)) {
1419 error_setg(errp
, "node-name=%s is conflicting with a device id",
1424 /* takes care of avoiding duplicates node names */
1425 if (bdrv_find_node(node_name
)) {
1426 error_setg(errp
, "Duplicate node name");
1430 /* Make sure that the node name isn't truncated */
1431 if (strlen(node_name
) >= sizeof(bs
->node_name
)) {
1432 error_setg(errp
, "Node name too long");
1436 /* copy node name into the bs and insert it into the graph list */
1437 pstrcpy(bs
->node_name
, sizeof(bs
->node_name
), node_name
);
1438 QTAILQ_INSERT_TAIL(&graph_bdrv_states
, bs
, node_list
);
1440 g_free(gen_node_name
);
1443 static int bdrv_open_driver(BlockDriverState
*bs
, BlockDriver
*drv
,
1444 const char *node_name
, QDict
*options
,
1445 int open_flags
, Error
**errp
)
1447 Error
*local_err
= NULL
;
1450 bdrv_assign_node_name(bs
, node_name
, &local_err
);
1452 error_propagate(errp
, local_err
);
1457 bs
->read_only
= !(bs
->open_flags
& BDRV_O_RDWR
);
1458 bs
->opaque
= g_malloc0(drv
->instance_size
);
1460 if (drv
->bdrv_file_open
) {
1461 assert(!drv
->bdrv_needs_filename
|| bs
->filename
[0]);
1462 ret
= drv
->bdrv_file_open(bs
, options
, open_flags
, &local_err
);
1463 } else if (drv
->bdrv_open
) {
1464 ret
= drv
->bdrv_open(bs
, options
, open_flags
, &local_err
);
1471 error_propagate(errp
, local_err
);
1472 } else if (bs
->filename
[0]) {
1473 error_setg_errno(errp
, -ret
, "Could not open '%s'", bs
->filename
);
1475 error_setg_errno(errp
, -ret
, "Could not open image");
1480 ret
= refresh_total_sectors(bs
, bs
->total_sectors
);
1482 error_setg_errno(errp
, -ret
, "Could not refresh total sector count");
1486 bdrv_refresh_limits(bs
, &local_err
);
1488 error_propagate(errp
, local_err
);
1492 assert(bdrv_opt_mem_align(bs
) != 0);
1493 assert(bdrv_min_mem_align(bs
) != 0);
1494 assert(is_power_of_2(bs
->bl
.request_alignment
));
1496 for (i
= 0; i
< bs
->quiesce_counter
; i
++) {
1497 if (drv
->bdrv_co_drain_begin
) {
1498 drv
->bdrv_co_drain_begin(bs
);
1505 if (bs
->file
!= NULL
) {
1506 bdrv_unref_child(bs
, bs
->file
);
1514 BlockDriverState
*bdrv_new_open_driver(BlockDriver
*drv
, const char *node_name
,
1515 int flags
, Error
**errp
)
1517 BlockDriverState
*bs
;
1521 bs
->open_flags
= flags
;
1522 bs
->explicit_options
= qdict_new();
1523 bs
->options
= qdict_new();
1526 update_options_from_flags(bs
->options
, flags
);
1528 ret
= bdrv_open_driver(bs
, drv
, node_name
, bs
->options
, flags
, errp
);
1530 qobject_unref(bs
->explicit_options
);
1531 bs
->explicit_options
= NULL
;
1532 qobject_unref(bs
->options
);
1541 QemuOptsList bdrv_runtime_opts
= {
1542 .name
= "bdrv_common",
1543 .head
= QTAILQ_HEAD_INITIALIZER(bdrv_runtime_opts
.head
),
1546 .name
= "node-name",
1547 .type
= QEMU_OPT_STRING
,
1548 .help
= "Node name of the block device node",
1552 .type
= QEMU_OPT_STRING
,
1553 .help
= "Block driver to use for the node",
1556 .name
= BDRV_OPT_CACHE_DIRECT
,
1557 .type
= QEMU_OPT_BOOL
,
1558 .help
= "Bypass software writeback cache on the host",
1561 .name
= BDRV_OPT_CACHE_NO_FLUSH
,
1562 .type
= QEMU_OPT_BOOL
,
1563 .help
= "Ignore flush requests",
1566 .name
= BDRV_OPT_READ_ONLY
,
1567 .type
= QEMU_OPT_BOOL
,
1568 .help
= "Node is opened in read-only mode",
1571 .name
= BDRV_OPT_AUTO_READ_ONLY
,
1572 .type
= QEMU_OPT_BOOL
,
1573 .help
= "Node can become read-only if opening read-write fails",
1576 .name
= "detect-zeroes",
1577 .type
= QEMU_OPT_STRING
,
1578 .help
= "try to optimize zero writes (off, on, unmap)",
1581 .name
= BDRV_OPT_DISCARD
,
1582 .type
= QEMU_OPT_STRING
,
1583 .help
= "discard operation (ignore/off, unmap/on)",
1586 .name
= BDRV_OPT_FORCE_SHARE
,
1587 .type
= QEMU_OPT_BOOL
,
1588 .help
= "always accept other writers (default: off)",
1590 { /* end of list */ }
1594 static QemuOptsList fallback_create_opts
= {
1595 .name
= "fallback-create-opts",
1596 .head
= QTAILQ_HEAD_INITIALIZER(fallback_create_opts
.head
),
1599 .name
= BLOCK_OPT_SIZE
,
1600 .type
= QEMU_OPT_SIZE
,
1601 .help
= "Virtual disk size"
1604 .name
= BLOCK_OPT_PREALLOC
,
1605 .type
= QEMU_OPT_STRING
,
1606 .help
= "Preallocation mode (allowed values: off)"
1608 { /* end of list */ }
1613 * Common part for opening disk images and files
1615 * Removes all processed options from *options.
1617 static int bdrv_open_common(BlockDriverState
*bs
, BlockBackend
*file
,
1618 QDict
*options
, Error
**errp
)
1620 int ret
, open_flags
;
1621 const char *filename
;
1622 const char *driver_name
= NULL
;
1623 const char *node_name
= NULL
;
1624 const char *discard
;
1627 Error
*local_err
= NULL
;
1629 assert(bs
->file
== NULL
);
1630 assert(options
!= NULL
&& bs
->options
!= options
);
1632 opts
= qemu_opts_create(&bdrv_runtime_opts
, NULL
, 0, &error_abort
);
1633 qemu_opts_absorb_qdict(opts
, options
, &local_err
);
1635 error_propagate(errp
, local_err
);
1640 update_flags_from_options(&bs
->open_flags
, opts
);
1642 driver_name
= qemu_opt_get(opts
, "driver");
1643 drv
= bdrv_find_format(driver_name
);
1644 assert(drv
!= NULL
);
1646 bs
->force_share
= qemu_opt_get_bool(opts
, BDRV_OPT_FORCE_SHARE
, false);
1648 if (bs
->force_share
&& (bs
->open_flags
& BDRV_O_RDWR
)) {
1650 BDRV_OPT_FORCE_SHARE
1651 "=on can only be used with read-only images");
1657 bdrv_refresh_filename(blk_bs(file
));
1658 filename
= blk_bs(file
)->filename
;
1661 * Caution: while qdict_get_try_str() is fine, getting
1662 * non-string types would require more care. When @options
1663 * come from -blockdev or blockdev_add, its members are typed
1664 * according to the QAPI schema, but when they come from
1665 * -drive, they're all QString.
1667 filename
= qdict_get_try_str(options
, "filename");
1670 if (drv
->bdrv_needs_filename
&& (!filename
|| !filename
[0])) {
1671 error_setg(errp
, "The '%s' block driver requires a file name",
1677 trace_bdrv_open_common(bs
, filename
?: "", bs
->open_flags
,
1680 bs
->read_only
= !(bs
->open_flags
& BDRV_O_RDWR
);
1682 if (use_bdrv_whitelist
&& !bdrv_is_whitelisted(drv
, bs
->read_only
)) {
1683 if (!bs
->read_only
&& bdrv_is_whitelisted(drv
, true)) {
1684 ret
= bdrv_apply_auto_read_only(bs
, NULL
, NULL
);
1690 !bs
->read_only
&& bdrv_is_whitelisted(drv
, true)
1691 ? "Driver '%s' can only be used for read-only devices"
1692 : "Driver '%s' is not whitelisted",
1698 /* bdrv_new() and bdrv_close() make it so */
1699 assert(atomic_read(&bs
->copy_on_read
) == 0);
1701 if (bs
->open_flags
& BDRV_O_COPY_ON_READ
) {
1702 if (!bs
->read_only
) {
1703 bdrv_enable_copy_on_read(bs
);
1705 error_setg(errp
, "Can't use copy-on-read on read-only device");
1711 discard
= qemu_opt_get(opts
, BDRV_OPT_DISCARD
);
1712 if (discard
!= NULL
) {
1713 if (bdrv_parse_discard_flags(discard
, &bs
->open_flags
) != 0) {
1714 error_setg(errp
, "Invalid discard option");
1721 bdrv_parse_detect_zeroes(opts
, bs
->open_flags
, &local_err
);
1723 error_propagate(errp
, local_err
);
1728 if (filename
!= NULL
) {
1729 pstrcpy(bs
->filename
, sizeof(bs
->filename
), filename
);
1731 bs
->filename
[0] = '\0';
1733 pstrcpy(bs
->exact_filename
, sizeof(bs
->exact_filename
), bs
->filename
);
1735 /* Open the image, either directly or using a protocol */
1736 open_flags
= bdrv_open_flags(bs
, bs
->open_flags
);
1737 node_name
= qemu_opt_get(opts
, "node-name");
1739 assert(!drv
->bdrv_file_open
|| file
== NULL
);
1740 ret
= bdrv_open_driver(bs
, drv
, node_name
, options
, open_flags
, errp
);
1745 qemu_opts_del(opts
);
1749 qemu_opts_del(opts
);
1753 static QDict
*parse_json_filename(const char *filename
, Error
**errp
)
1755 QObject
*options_obj
;
1759 ret
= strstart(filename
, "json:", &filename
);
1762 options_obj
= qobject_from_json(filename
, errp
);
1764 error_prepend(errp
, "Could not parse the JSON options: ");
1768 options
= qobject_to(QDict
, options_obj
);
1770 qobject_unref(options_obj
);
1771 error_setg(errp
, "Invalid JSON object given");
1775 qdict_flatten(options
);
1780 static void parse_json_protocol(QDict
*options
, const char **pfilename
,
1783 QDict
*json_options
;
1784 Error
*local_err
= NULL
;
1786 /* Parse json: pseudo-protocol */
1787 if (!*pfilename
|| !g_str_has_prefix(*pfilename
, "json:")) {
1791 json_options
= parse_json_filename(*pfilename
, &local_err
);
1793 error_propagate(errp
, local_err
);
1797 /* Options given in the filename have lower priority than options
1798 * specified directly */
1799 qdict_join(options
, json_options
, false);
1800 qobject_unref(json_options
);
1805 * Fills in default options for opening images and converts the legacy
1806 * filename/flags pair to option QDict entries.
1807 * The BDRV_O_PROTOCOL flag in *flags will be set or cleared accordingly if a
1808 * block driver has been specified explicitly.
1810 static int bdrv_fill_options(QDict
**options
, const char *filename
,
1811 int *flags
, Error
**errp
)
1813 const char *drvname
;
1814 bool protocol
= *flags
& BDRV_O_PROTOCOL
;
1815 bool parse_filename
= false;
1816 BlockDriver
*drv
= NULL
;
1817 Error
*local_err
= NULL
;
1820 * Caution: while qdict_get_try_str() is fine, getting non-string
1821 * types would require more care. When @options come from
1822 * -blockdev or blockdev_add, its members are typed according to
1823 * the QAPI schema, but when they come from -drive, they're all
1826 drvname
= qdict_get_try_str(*options
, "driver");
1828 drv
= bdrv_find_format(drvname
);
1830 error_setg(errp
, "Unknown driver '%s'", drvname
);
1833 /* If the user has explicitly specified the driver, this choice should
1834 * override the BDRV_O_PROTOCOL flag */
1835 protocol
= drv
->bdrv_file_open
;
1839 *flags
|= BDRV_O_PROTOCOL
;
1841 *flags
&= ~BDRV_O_PROTOCOL
;
1844 /* Translate cache options from flags into options */
1845 update_options_from_flags(*options
, *flags
);
1847 /* Fetch the file name from the options QDict if necessary */
1848 if (protocol
&& filename
) {
1849 if (!qdict_haskey(*options
, "filename")) {
1850 qdict_put_str(*options
, "filename", filename
);
1851 parse_filename
= true;
1853 error_setg(errp
, "Can't specify 'file' and 'filename' options at "
1859 /* Find the right block driver */
1860 /* See cautionary note on accessing @options above */
1861 filename
= qdict_get_try_str(*options
, "filename");
1863 if (!drvname
&& protocol
) {
1865 drv
= bdrv_find_protocol(filename
, parse_filename
, errp
);
1870 drvname
= drv
->format_name
;
1871 qdict_put_str(*options
, "driver", drvname
);
1873 error_setg(errp
, "Must specify either driver or file");
1878 assert(drv
|| !protocol
);
1880 /* Driver-specific filename parsing */
1881 if (drv
&& drv
->bdrv_parse_filename
&& parse_filename
) {
1882 drv
->bdrv_parse_filename(filename
, *options
, &local_err
);
1884 error_propagate(errp
, local_err
);
1888 if (!drv
->bdrv_needs_filename
) {
1889 qdict_del(*options
, "filename");
1896 static int bdrv_child_check_perm(BdrvChild
*c
, BlockReopenQueue
*q
,
1897 uint64_t perm
, uint64_t shared
,
1898 GSList
*ignore_children
,
1899 bool *tighten_restrictions
, Error
**errp
);
1900 static void bdrv_child_abort_perm_update(BdrvChild
*c
);
1901 static void bdrv_child_set_perm(BdrvChild
*c
, uint64_t perm
, uint64_t shared
);
1903 typedef struct BlockReopenQueueEntry
{
1906 BDRVReopenState state
;
1907 QTAILQ_ENTRY(BlockReopenQueueEntry
) entry
;
1908 } BlockReopenQueueEntry
;
1911 * Return the flags that @bs will have after the reopens in @q have
1912 * successfully completed. If @q is NULL (or @bs is not contained in @q),
1913 * return the current flags.
1915 static int bdrv_reopen_get_flags(BlockReopenQueue
*q
, BlockDriverState
*bs
)
1917 BlockReopenQueueEntry
*entry
;
1920 QTAILQ_FOREACH(entry
, q
, entry
) {
1921 if (entry
->state
.bs
== bs
) {
1922 return entry
->state
.flags
;
1927 return bs
->open_flags
;
1930 /* Returns whether the image file can be written to after the reopen queue @q
1931 * has been successfully applied, or right now if @q is NULL. */
1932 static bool bdrv_is_writable_after_reopen(BlockDriverState
*bs
,
1933 BlockReopenQueue
*q
)
1935 int flags
= bdrv_reopen_get_flags(q
, bs
);
1937 return (flags
& (BDRV_O_RDWR
| BDRV_O_INACTIVE
)) == BDRV_O_RDWR
;
1941 * Return whether the BDS can be written to. This is not necessarily
1942 * the same as !bdrv_is_read_only(bs), as inactivated images may not
1943 * be written to but do not count as read-only images.
1945 bool bdrv_is_writable(BlockDriverState
*bs
)
1947 return bdrv_is_writable_after_reopen(bs
, NULL
);
1950 static void bdrv_child_perm(BlockDriverState
*bs
, BlockDriverState
*child_bs
,
1951 BdrvChild
*c
, const BdrvChildRole
*role
,
1952 BlockReopenQueue
*reopen_queue
,
1953 uint64_t parent_perm
, uint64_t parent_shared
,
1954 uint64_t *nperm
, uint64_t *nshared
)
1956 assert(bs
->drv
&& bs
->drv
->bdrv_child_perm
);
1957 bs
->drv
->bdrv_child_perm(bs
, c
, role
, reopen_queue
,
1958 parent_perm
, parent_shared
,
1960 /* TODO Take force_share from reopen_queue */
1961 if (child_bs
&& child_bs
->force_share
) {
1962 *nshared
= BLK_PERM_ALL
;
1967 * Check whether permissions on this node can be changed in a way that
1968 * @cumulative_perms and @cumulative_shared_perms are the new cumulative
1969 * permissions of all its parents. This involves checking whether all necessary
1970 * permission changes to child nodes can be performed.
1972 * Will set *tighten_restrictions to true if and only if new permissions have to
1973 * be taken or currently shared permissions are to be unshared. Otherwise,
1974 * errors are not fatal as long as the caller accepts that the restrictions
1975 * remain tighter than they need to be. The caller still has to abort the
1977 * @tighten_restrictions cannot be used together with @q: When reopening, we may
1978 * encounter fatal errors even though no restrictions are to be tightened. For
1979 * example, changing a node from RW to RO will fail if the WRITE permission is
1982 * A call to this function must always be followed by a call to bdrv_set_perm()
1983 * or bdrv_abort_perm_update().
1985 static int bdrv_check_perm(BlockDriverState
*bs
, BlockReopenQueue
*q
,
1986 uint64_t cumulative_perms
,
1987 uint64_t cumulative_shared_perms
,
1988 GSList
*ignore_children
,
1989 bool *tighten_restrictions
, Error
**errp
)
1991 BlockDriver
*drv
= bs
->drv
;
1995 assert(!q
|| !tighten_restrictions
);
1997 if (tighten_restrictions
) {
1998 uint64_t current_perms
, current_shared
;
1999 uint64_t added_perms
, removed_shared_perms
;
2001 bdrv_get_cumulative_perm(bs
, ¤t_perms
, ¤t_shared
);
2003 added_perms
= cumulative_perms
& ~current_perms
;
2004 removed_shared_perms
= current_shared
& ~cumulative_shared_perms
;
2006 *tighten_restrictions
= added_perms
|| removed_shared_perms
;
2009 /* Write permissions never work with read-only images */
2010 if ((cumulative_perms
& (BLK_PERM_WRITE
| BLK_PERM_WRITE_UNCHANGED
)) &&
2011 !bdrv_is_writable_after_reopen(bs
, q
))
2013 if (!bdrv_is_writable_after_reopen(bs
, NULL
)) {
2014 error_setg(errp
, "Block node is read-only");
2016 uint64_t current_perms
, current_shared
;
2017 bdrv_get_cumulative_perm(bs
, ¤t_perms
, ¤t_shared
);
2018 if (current_perms
& (BLK_PERM_WRITE
| BLK_PERM_WRITE_UNCHANGED
)) {
2019 error_setg(errp
, "Cannot make block node read-only, there is "
2022 error_setg(errp
, "Cannot make block node read-only and create "
2030 /* Check this node */
2035 if (drv
->bdrv_check_perm
) {
2036 return drv
->bdrv_check_perm(bs
, cumulative_perms
,
2037 cumulative_shared_perms
, errp
);
2040 /* Drivers that never have children can omit .bdrv_child_perm() */
2041 if (!drv
->bdrv_child_perm
) {
2042 assert(QLIST_EMPTY(&bs
->children
));
2046 /* Check all children */
2047 QLIST_FOREACH(c
, &bs
->children
, next
) {
2048 uint64_t cur_perm
, cur_shared
;
2049 bool child_tighten_restr
;
2051 bdrv_child_perm(bs
, c
->bs
, c
, c
->role
, q
,
2052 cumulative_perms
, cumulative_shared_perms
,
2053 &cur_perm
, &cur_shared
);
2054 ret
= bdrv_child_check_perm(c
, q
, cur_perm
, cur_shared
, ignore_children
,
2055 tighten_restrictions
? &child_tighten_restr
2058 if (tighten_restrictions
) {
2059 *tighten_restrictions
|= child_tighten_restr
;
2070 * Notifies drivers that after a previous bdrv_check_perm() call, the
2071 * permission update is not performed and any preparations made for it (e.g.
2072 * taken file locks) need to be undone.
2074 * This function recursively notifies all child nodes.
2076 static void bdrv_abort_perm_update(BlockDriverState
*bs
)
2078 BlockDriver
*drv
= bs
->drv
;
2085 if (drv
->bdrv_abort_perm_update
) {
2086 drv
->bdrv_abort_perm_update(bs
);
2089 QLIST_FOREACH(c
, &bs
->children
, next
) {
2090 bdrv_child_abort_perm_update(c
);
2094 static void bdrv_set_perm(BlockDriverState
*bs
, uint64_t cumulative_perms
,
2095 uint64_t cumulative_shared_perms
)
2097 BlockDriver
*drv
= bs
->drv
;
2104 /* Update this node */
2105 if (drv
->bdrv_set_perm
) {
2106 drv
->bdrv_set_perm(bs
, cumulative_perms
, cumulative_shared_perms
);
2109 /* Drivers that never have children can omit .bdrv_child_perm() */
2110 if (!drv
->bdrv_child_perm
) {
2111 assert(QLIST_EMPTY(&bs
->children
));
2115 /* Update all children */
2116 QLIST_FOREACH(c
, &bs
->children
, next
) {
2117 uint64_t cur_perm
, cur_shared
;
2118 bdrv_child_perm(bs
, c
->bs
, c
, c
->role
, NULL
,
2119 cumulative_perms
, cumulative_shared_perms
,
2120 &cur_perm
, &cur_shared
);
2121 bdrv_child_set_perm(c
, cur_perm
, cur_shared
);
2125 void bdrv_get_cumulative_perm(BlockDriverState
*bs
, uint64_t *perm
,
2126 uint64_t *shared_perm
)
2129 uint64_t cumulative_perms
= 0;
2130 uint64_t cumulative_shared_perms
= BLK_PERM_ALL
;
2132 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
2133 cumulative_perms
|= c
->perm
;
2134 cumulative_shared_perms
&= c
->shared_perm
;
2137 *perm
= cumulative_perms
;
2138 *shared_perm
= cumulative_shared_perms
;
2141 static char *bdrv_child_user_desc(BdrvChild
*c
)
2143 if (c
->role
->get_parent_desc
) {
2144 return c
->role
->get_parent_desc(c
);
2147 return g_strdup("another user");
2150 char *bdrv_perm_names(uint64_t perm
)
2156 { BLK_PERM_CONSISTENT_READ
, "consistent read" },
2157 { BLK_PERM_WRITE
, "write" },
2158 { BLK_PERM_WRITE_UNCHANGED
, "write unchanged" },
2159 { BLK_PERM_RESIZE
, "resize" },
2160 { BLK_PERM_GRAPH_MOD
, "change children" },
2164 GString
*result
= g_string_sized_new(30);
2165 struct perm_name
*p
;
2167 for (p
= permissions
; p
->name
; p
++) {
2168 if (perm
& p
->perm
) {
2169 if (result
->len
> 0) {
2170 g_string_append(result
, ", ");
2172 g_string_append(result
, p
->name
);
2176 return g_string_free(result
, FALSE
);
2180 * Checks whether a new reference to @bs can be added if the new user requires
2181 * @new_used_perm/@new_shared_perm as its permissions. If @ignore_children is
2182 * set, the BdrvChild objects in this list are ignored in the calculations;
2183 * this allows checking permission updates for an existing reference.
2185 * See bdrv_check_perm() for the semantics of @tighten_restrictions.
2187 * Needs to be followed by a call to either bdrv_set_perm() or
2188 * bdrv_abort_perm_update(). */
2189 static int bdrv_check_update_perm(BlockDriverState
*bs
, BlockReopenQueue
*q
,
2190 uint64_t new_used_perm
,
2191 uint64_t new_shared_perm
,
2192 GSList
*ignore_children
,
2193 bool *tighten_restrictions
,
2197 uint64_t cumulative_perms
= new_used_perm
;
2198 uint64_t cumulative_shared_perms
= new_shared_perm
;
2200 assert(!q
|| !tighten_restrictions
);
2202 /* There is no reason why anyone couldn't tolerate write_unchanged */
2203 assert(new_shared_perm
& BLK_PERM_WRITE_UNCHANGED
);
2205 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
2206 if (g_slist_find(ignore_children
, c
)) {
2210 if ((new_used_perm
& c
->shared_perm
) != new_used_perm
) {
2211 char *user
= bdrv_child_user_desc(c
);
2212 char *perm_names
= bdrv_perm_names(new_used_perm
& ~c
->shared_perm
);
2214 if (tighten_restrictions
) {
2215 *tighten_restrictions
= true;
2218 error_setg(errp
, "Conflicts with use by %s as '%s', which does not "
2220 user
, c
->name
, perm_names
, bdrv_get_node_name(c
->bs
));
2226 if ((c
->perm
& new_shared_perm
) != c
->perm
) {
2227 char *user
= bdrv_child_user_desc(c
);
2228 char *perm_names
= bdrv_perm_names(c
->perm
& ~new_shared_perm
);
2230 if (tighten_restrictions
) {
2231 *tighten_restrictions
= true;
2234 error_setg(errp
, "Conflicts with use by %s as '%s', which uses "
2236 user
, c
->name
, perm_names
, bdrv_get_node_name(c
->bs
));
2242 cumulative_perms
|= c
->perm
;
2243 cumulative_shared_perms
&= c
->shared_perm
;
2246 return bdrv_check_perm(bs
, q
, cumulative_perms
, cumulative_shared_perms
,
2247 ignore_children
, tighten_restrictions
, errp
);
2250 /* Needs to be followed by a call to either bdrv_child_set_perm() or
2251 * bdrv_child_abort_perm_update(). */
2252 static int bdrv_child_check_perm(BdrvChild
*c
, BlockReopenQueue
*q
,
2253 uint64_t perm
, uint64_t shared
,
2254 GSList
*ignore_children
,
2255 bool *tighten_restrictions
, Error
**errp
)
2259 ignore_children
= g_slist_prepend(g_slist_copy(ignore_children
), c
);
2260 ret
= bdrv_check_update_perm(c
->bs
, q
, perm
, shared
, ignore_children
,
2261 tighten_restrictions
, errp
);
2262 g_slist_free(ignore_children
);
2268 if (!c
->has_backup_perm
) {
2269 c
->has_backup_perm
= true;
2270 c
->backup_perm
= c
->perm
;
2271 c
->backup_shared_perm
= c
->shared_perm
;
2274 * Note: it's OK if c->has_backup_perm was already set, as we can find the
2275 * same child twice during check_perm procedure
2279 c
->shared_perm
= shared
;
2284 static void bdrv_child_set_perm(BdrvChild
*c
, uint64_t perm
, uint64_t shared
)
2286 uint64_t cumulative_perms
, cumulative_shared_perms
;
2288 c
->has_backup_perm
= false;
2291 c
->shared_perm
= shared
;
2293 bdrv_get_cumulative_perm(c
->bs
, &cumulative_perms
,
2294 &cumulative_shared_perms
);
2295 bdrv_set_perm(c
->bs
, cumulative_perms
, cumulative_shared_perms
);
2298 static void bdrv_child_abort_perm_update(BdrvChild
*c
)
2300 if (c
->has_backup_perm
) {
2301 c
->perm
= c
->backup_perm
;
2302 c
->shared_perm
= c
->backup_shared_perm
;
2303 c
->has_backup_perm
= false;
2306 bdrv_abort_perm_update(c
->bs
);
2309 int bdrv_child_try_set_perm(BdrvChild
*c
, uint64_t perm
, uint64_t shared
,
2312 Error
*local_err
= NULL
;
2314 bool tighten_restrictions
;
2316 ret
= bdrv_child_check_perm(c
, NULL
, perm
, shared
, NULL
,
2317 &tighten_restrictions
, &local_err
);
2319 bdrv_child_abort_perm_update(c
);
2320 if (tighten_restrictions
) {
2321 error_propagate(errp
, local_err
);
2324 * Our caller may intend to only loosen restrictions and
2325 * does not expect this function to fail. Errors are not
2326 * fatal in such a case, so we can just hide them from our
2329 error_free(local_err
);
2335 bdrv_child_set_perm(c
, perm
, shared
);
2340 int bdrv_child_refresh_perms(BlockDriverState
*bs
, BdrvChild
*c
, Error
**errp
)
2342 uint64_t parent_perms
, parent_shared
;
2343 uint64_t perms
, shared
;
2345 bdrv_get_cumulative_perm(bs
, &parent_perms
, &parent_shared
);
2346 bdrv_child_perm(bs
, c
->bs
, c
, c
->role
, NULL
, parent_perms
, parent_shared
,
2349 return bdrv_child_try_set_perm(c
, perms
, shared
, errp
);
2352 void bdrv_filter_default_perms(BlockDriverState
*bs
, BdrvChild
*c
,
2353 const BdrvChildRole
*role
,
2354 BlockReopenQueue
*reopen_queue
,
2355 uint64_t perm
, uint64_t shared
,
2356 uint64_t *nperm
, uint64_t *nshared
)
2358 *nperm
= perm
& DEFAULT_PERM_PASSTHROUGH
;
2359 *nshared
= (shared
& DEFAULT_PERM_PASSTHROUGH
) | DEFAULT_PERM_UNCHANGED
;
2362 void bdrv_format_default_perms(BlockDriverState
*bs
, BdrvChild
*c
,
2363 const BdrvChildRole
*role
,
2364 BlockReopenQueue
*reopen_queue
,
2365 uint64_t perm
, uint64_t shared
,
2366 uint64_t *nperm
, uint64_t *nshared
)
2368 bool backing
= (role
== &child_backing
);
2369 assert(role
== &child_backing
|| role
== &child_file
);
2372 int flags
= bdrv_reopen_get_flags(reopen_queue
, bs
);
2374 /* Apart from the modifications below, the same permissions are
2375 * forwarded and left alone as for filters */
2376 bdrv_filter_default_perms(bs
, c
, role
, reopen_queue
, perm
, shared
,
2379 /* Format drivers may touch metadata even if the guest doesn't write */
2380 if (bdrv_is_writable_after_reopen(bs
, reopen_queue
)) {
2381 perm
|= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2384 /* bs->file always needs to be consistent because of the metadata. We
2385 * can never allow other users to resize or write to it. */
2386 if (!(flags
& BDRV_O_NO_IO
)) {
2387 perm
|= BLK_PERM_CONSISTENT_READ
;
2389 shared
&= ~(BLK_PERM_WRITE
| BLK_PERM_RESIZE
);
2391 /* We want consistent read from backing files if the parent needs it.
2392 * No other operations are performed on backing files. */
2393 perm
&= BLK_PERM_CONSISTENT_READ
;
2395 /* If the parent can deal with changing data, we're okay with a
2396 * writable and resizable backing file. */
2397 /* TODO Require !(perm & BLK_PERM_CONSISTENT_READ), too? */
2398 if (shared
& BLK_PERM_WRITE
) {
2399 shared
= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2404 shared
|= BLK_PERM_CONSISTENT_READ
| BLK_PERM_GRAPH_MOD
|
2405 BLK_PERM_WRITE_UNCHANGED
;
2408 if (bs
->open_flags
& BDRV_O_INACTIVE
) {
2409 shared
|= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2416 uint64_t bdrv_qapi_perm_to_blk_perm(BlockPermission qapi_perm
)
2418 static const uint64_t permissions
[] = {
2419 [BLOCK_PERMISSION_CONSISTENT_READ
] = BLK_PERM_CONSISTENT_READ
,
2420 [BLOCK_PERMISSION_WRITE
] = BLK_PERM_WRITE
,
2421 [BLOCK_PERMISSION_WRITE_UNCHANGED
] = BLK_PERM_WRITE_UNCHANGED
,
2422 [BLOCK_PERMISSION_RESIZE
] = BLK_PERM_RESIZE
,
2423 [BLOCK_PERMISSION_GRAPH_MOD
] = BLK_PERM_GRAPH_MOD
,
2426 QEMU_BUILD_BUG_ON(ARRAY_SIZE(permissions
) != BLOCK_PERMISSION__MAX
);
2427 QEMU_BUILD_BUG_ON(1UL << ARRAY_SIZE(permissions
) != BLK_PERM_ALL
+ 1);
2429 assert(qapi_perm
< BLOCK_PERMISSION__MAX
);
2431 return permissions
[qapi_perm
];
2434 static void bdrv_replace_child_noperm(BdrvChild
*child
,
2435 BlockDriverState
*new_bs
)
2437 BlockDriverState
*old_bs
= child
->bs
;
2438 int new_bs_quiesce_counter
;
2441 assert(!child
->frozen
);
2443 if (old_bs
&& new_bs
) {
2444 assert(bdrv_get_aio_context(old_bs
) == bdrv_get_aio_context(new_bs
));
2447 new_bs_quiesce_counter
= (new_bs
? new_bs
->quiesce_counter
: 0);
2448 drain_saldo
= new_bs_quiesce_counter
- child
->parent_quiesce_counter
;
2451 * If the new child node is drained but the old one was not, flush
2452 * all outstanding requests to the old child node.
2454 while (drain_saldo
> 0 && child
->role
->drained_begin
) {
2455 bdrv_parent_drained_begin_single(child
, true);
2460 /* Detach first so that the recursive drain sections coming from @child
2461 * are already gone and we only end the drain sections that came from
2463 if (child
->role
->detach
) {
2464 child
->role
->detach(child
);
2466 QLIST_REMOVE(child
, next_parent
);
2472 QLIST_INSERT_HEAD(&new_bs
->parents
, child
, next_parent
);
2475 * Detaching the old node may have led to the new node's
2476 * quiesce_counter having been decreased. Not a problem, we
2477 * just need to recognize this here and then invoke
2478 * drained_end appropriately more often.
2480 assert(new_bs
->quiesce_counter
<= new_bs_quiesce_counter
);
2481 drain_saldo
+= new_bs
->quiesce_counter
- new_bs_quiesce_counter
;
2483 /* Attach only after starting new drained sections, so that recursive
2484 * drain sections coming from @child don't get an extra .drained_begin
2486 if (child
->role
->attach
) {
2487 child
->role
->attach(child
);
2492 * If the old child node was drained but the new one is not, allow
2493 * requests to come in only after the new node has been attached.
2495 while (drain_saldo
< 0 && child
->role
->drained_end
) {
2496 bdrv_parent_drained_end_single(child
);
2502 * Updates @child to change its reference to point to @new_bs, including
2503 * checking and applying the necessary permisson updates both to the old node
2506 * NULL is passed as @new_bs for removing the reference before freeing @child.
2508 * If @new_bs is not NULL, bdrv_check_perm() must be called beforehand, as this
2509 * function uses bdrv_set_perm() to update the permissions according to the new
2510 * reference that @new_bs gets.
2512 static void bdrv_replace_child(BdrvChild
*child
, BlockDriverState
*new_bs
)
2514 BlockDriverState
*old_bs
= child
->bs
;
2515 uint64_t perm
, shared_perm
;
2517 bdrv_replace_child_noperm(child
, new_bs
);
2520 * Start with the new node's permissions. If @new_bs is a (direct
2521 * or indirect) child of @old_bs, we must complete the permission
2522 * update on @new_bs before we loosen the restrictions on @old_bs.
2523 * Otherwise, bdrv_check_perm() on @old_bs would re-initiate
2524 * updating the permissions of @new_bs, and thus not purely loosen
2528 bdrv_get_cumulative_perm(new_bs
, &perm
, &shared_perm
);
2529 bdrv_set_perm(new_bs
, perm
, shared_perm
);
2533 /* Update permissions for old node. This is guaranteed to succeed
2534 * because we're just taking a parent away, so we're loosening
2536 bool tighten_restrictions
;
2539 bdrv_get_cumulative_perm(old_bs
, &perm
, &shared_perm
);
2540 ret
= bdrv_check_perm(old_bs
, NULL
, perm
, shared_perm
, NULL
,
2541 &tighten_restrictions
, NULL
);
2542 assert(tighten_restrictions
== false);
2544 /* We only tried to loosen restrictions, so errors are not fatal */
2545 bdrv_abort_perm_update(old_bs
);
2547 bdrv_set_perm(old_bs
, perm
, shared_perm
);
2550 /* When the parent requiring a non-default AioContext is removed, the
2551 * node moves back to the main AioContext */
2552 bdrv_try_set_aio_context(old_bs
, qemu_get_aio_context(), NULL
);
2557 * This function steals the reference to child_bs from the caller.
2558 * That reference is later dropped by bdrv_root_unref_child().
2560 * On failure NULL is returned, errp is set and the reference to
2561 * child_bs is also dropped.
2563 * The caller must hold the AioContext lock @child_bs, but not that of @ctx
2564 * (unless @child_bs is already in @ctx).
2566 BdrvChild
*bdrv_root_attach_child(BlockDriverState
*child_bs
,
2567 const char *child_name
,
2568 const BdrvChildRole
*child_role
,
2570 uint64_t perm
, uint64_t shared_perm
,
2571 void *opaque
, Error
**errp
)
2574 Error
*local_err
= NULL
;
2577 ret
= bdrv_check_update_perm(child_bs
, NULL
, perm
, shared_perm
, NULL
, NULL
,
2580 bdrv_abort_perm_update(child_bs
);
2581 bdrv_unref(child_bs
);
2585 child
= g_new(BdrvChild
, 1);
2586 *child
= (BdrvChild
) {
2588 .name
= g_strdup(child_name
),
2591 .shared_perm
= shared_perm
,
2595 /* If the AioContexts don't match, first try to move the subtree of
2596 * child_bs into the AioContext of the new parent. If this doesn't work,
2597 * try moving the parent into the AioContext of child_bs instead. */
2598 if (bdrv_get_aio_context(child_bs
) != ctx
) {
2599 ret
= bdrv_try_set_aio_context(child_bs
, ctx
, &local_err
);
2600 if (ret
< 0 && child_role
->can_set_aio_ctx
) {
2601 GSList
*ignore
= g_slist_prepend(NULL
, child
);
2602 ctx
= bdrv_get_aio_context(child_bs
);
2603 if (child_role
->can_set_aio_ctx(child
, ctx
, &ignore
, NULL
)) {
2604 error_free(local_err
);
2606 g_slist_free(ignore
);
2607 ignore
= g_slist_prepend(NULL
, child
);
2608 child_role
->set_aio_ctx(child
, ctx
, &ignore
);
2610 g_slist_free(ignore
);
2613 error_propagate(errp
, local_err
);
2615 bdrv_abort_perm_update(child_bs
);
2620 /* This performs the matching bdrv_set_perm() for the above check. */
2621 bdrv_replace_child(child
, child_bs
);
2627 * This function transfers the reference to child_bs from the caller
2628 * to parent_bs. That reference is later dropped by parent_bs on
2629 * bdrv_close() or if someone calls bdrv_unref_child().
2631 * On failure NULL is returned, errp is set and the reference to
2632 * child_bs is also dropped.
2634 * If @parent_bs and @child_bs are in different AioContexts, the caller must
2635 * hold the AioContext lock for @child_bs, but not for @parent_bs.
2637 BdrvChild
*bdrv_attach_child(BlockDriverState
*parent_bs
,
2638 BlockDriverState
*child_bs
,
2639 const char *child_name
,
2640 const BdrvChildRole
*child_role
,
2644 uint64_t perm
, shared_perm
;
2646 bdrv_get_cumulative_perm(parent_bs
, &perm
, &shared_perm
);
2648 assert(parent_bs
->drv
);
2649 bdrv_child_perm(parent_bs
, child_bs
, NULL
, child_role
, NULL
,
2650 perm
, shared_perm
, &perm
, &shared_perm
);
2652 child
= bdrv_root_attach_child(child_bs
, child_name
, child_role
,
2653 bdrv_get_aio_context(parent_bs
),
2654 perm
, shared_perm
, parent_bs
, errp
);
2655 if (child
== NULL
) {
2659 QLIST_INSERT_HEAD(&parent_bs
->children
, child
, next
);
2663 static void bdrv_detach_child(BdrvChild
*child
)
2665 QLIST_SAFE_REMOVE(child
, next
);
2667 bdrv_replace_child(child
, NULL
);
2669 g_free(child
->name
);
2673 void bdrv_root_unref_child(BdrvChild
*child
)
2675 BlockDriverState
*child_bs
;
2677 child_bs
= child
->bs
;
2678 bdrv_detach_child(child
);
2679 bdrv_unref(child_bs
);
2683 * Clear all inherits_from pointers from children and grandchildren of
2684 * @root that point to @root, where necessary.
2686 static void bdrv_unset_inherits_from(BlockDriverState
*root
, BdrvChild
*child
)
2690 if (child
->bs
->inherits_from
== root
) {
2692 * Remove inherits_from only when the last reference between root and
2693 * child->bs goes away.
2695 QLIST_FOREACH(c
, &root
->children
, next
) {
2696 if (c
!= child
&& c
->bs
== child
->bs
) {
2701 child
->bs
->inherits_from
= NULL
;
2705 QLIST_FOREACH(c
, &child
->bs
->children
, next
) {
2706 bdrv_unset_inherits_from(root
, c
);
2710 void bdrv_unref_child(BlockDriverState
*parent
, BdrvChild
*child
)
2712 if (child
== NULL
) {
2716 bdrv_unset_inherits_from(parent
, child
);
2717 bdrv_root_unref_child(child
);
2721 static void bdrv_parent_cb_change_media(BlockDriverState
*bs
, bool load
)
2724 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
2725 if (c
->role
->change_media
) {
2726 c
->role
->change_media(c
, load
);
2731 /* Return true if you can reach parent going through child->inherits_from
2732 * recursively. If parent or child are NULL, return false */
2733 static bool bdrv_inherits_from_recursive(BlockDriverState
*child
,
2734 BlockDriverState
*parent
)
2736 while (child
&& child
!= parent
) {
2737 child
= child
->inherits_from
;
2740 return child
!= NULL
;
2744 * Sets the backing file link of a BDS. A new reference is created; callers
2745 * which don't need their own reference any more must call bdrv_unref().
2747 void bdrv_set_backing_hd(BlockDriverState
*bs
, BlockDriverState
*backing_hd
,
2750 bool update_inherits_from
= bdrv_chain_contains(bs
, backing_hd
) &&
2751 bdrv_inherits_from_recursive(backing_hd
, bs
);
2753 if (bdrv_is_backing_chain_frozen(bs
, backing_bs(bs
), errp
)) {
2758 bdrv_ref(backing_hd
);
2762 bdrv_unref_child(bs
, bs
->backing
);
2770 bs
->backing
= bdrv_attach_child(bs
, backing_hd
, "backing", &child_backing
,
2772 /* If backing_hd was already part of bs's backing chain, and
2773 * inherits_from pointed recursively to bs then let's update it to
2774 * point directly to bs (else it will become NULL). */
2775 if (bs
->backing
&& update_inherits_from
) {
2776 backing_hd
->inherits_from
= bs
;
2780 bdrv_refresh_limits(bs
, NULL
);
2784 * Opens the backing file for a BlockDriverState if not yet open
2786 * bdref_key specifies the key for the image's BlockdevRef in the options QDict.
2787 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
2788 * itself, all options starting with "${bdref_key}." are considered part of the
2791 * TODO Can this be unified with bdrv_open_image()?
2793 int bdrv_open_backing_file(BlockDriverState
*bs
, QDict
*parent_options
,
2794 const char *bdref_key
, Error
**errp
)
2796 char *backing_filename
= NULL
;
2797 char *bdref_key_dot
;
2798 const char *reference
= NULL
;
2800 bool implicit_backing
= false;
2801 BlockDriverState
*backing_hd
;
2803 QDict
*tmp_parent_options
= NULL
;
2804 Error
*local_err
= NULL
;
2806 if (bs
->backing
!= NULL
) {
2810 /* NULL means an empty set of options */
2811 if (parent_options
== NULL
) {
2812 tmp_parent_options
= qdict_new();
2813 parent_options
= tmp_parent_options
;
2816 bs
->open_flags
&= ~BDRV_O_NO_BACKING
;
2818 bdref_key_dot
= g_strdup_printf("%s.", bdref_key
);
2819 qdict_extract_subqdict(parent_options
, &options
, bdref_key_dot
);
2820 g_free(bdref_key_dot
);
2823 * Caution: while qdict_get_try_str() is fine, getting non-string
2824 * types would require more care. When @parent_options come from
2825 * -blockdev or blockdev_add, its members are typed according to
2826 * the QAPI schema, but when they come from -drive, they're all
2829 reference
= qdict_get_try_str(parent_options
, bdref_key
);
2830 if (reference
|| qdict_haskey(options
, "file.filename")) {
2831 /* keep backing_filename NULL */
2832 } else if (bs
->backing_file
[0] == '\0' && qdict_size(options
) == 0) {
2833 qobject_unref(options
);
2836 if (qdict_size(options
) == 0) {
2837 /* If the user specifies options that do not modify the
2838 * backing file's behavior, we might still consider it the
2839 * implicit backing file. But it's easier this way, and
2840 * just specifying some of the backing BDS's options is
2841 * only possible with -drive anyway (otherwise the QAPI
2842 * schema forces the user to specify everything). */
2843 implicit_backing
= !strcmp(bs
->auto_backing_file
, bs
->backing_file
);
2846 backing_filename
= bdrv_get_full_backing_filename(bs
, &local_err
);
2849 error_propagate(errp
, local_err
);
2850 qobject_unref(options
);
2855 if (!bs
->drv
|| !bs
->drv
->supports_backing
) {
2857 error_setg(errp
, "Driver doesn't support backing files");
2858 qobject_unref(options
);
2863 bs
->backing_format
[0] != '\0' && !qdict_haskey(options
, "driver")) {
2864 qdict_put_str(options
, "driver", bs
->backing_format
);
2867 backing_hd
= bdrv_open_inherit(backing_filename
, reference
, options
, 0, bs
,
2868 &child_backing
, errp
);
2870 bs
->open_flags
|= BDRV_O_NO_BACKING
;
2871 error_prepend(errp
, "Could not open backing file: ");
2876 if (implicit_backing
) {
2877 bdrv_refresh_filename(backing_hd
);
2878 pstrcpy(bs
->auto_backing_file
, sizeof(bs
->auto_backing_file
),
2879 backing_hd
->filename
);
2882 /* Hook up the backing file link; drop our reference, bs owns the
2883 * backing_hd reference now */
2884 bdrv_set_backing_hd(bs
, backing_hd
, &local_err
);
2885 bdrv_unref(backing_hd
);
2887 error_propagate(errp
, local_err
);
2892 qdict_del(parent_options
, bdref_key
);
2895 g_free(backing_filename
);
2896 qobject_unref(tmp_parent_options
);
2900 static BlockDriverState
*
2901 bdrv_open_child_bs(const char *filename
, QDict
*options
, const char *bdref_key
,
2902 BlockDriverState
*parent
, const BdrvChildRole
*child_role
,
2903 bool allow_none
, Error
**errp
)
2905 BlockDriverState
*bs
= NULL
;
2906 QDict
*image_options
;
2907 char *bdref_key_dot
;
2908 const char *reference
;
2910 assert(child_role
!= NULL
);
2912 bdref_key_dot
= g_strdup_printf("%s.", bdref_key
);
2913 qdict_extract_subqdict(options
, &image_options
, bdref_key_dot
);
2914 g_free(bdref_key_dot
);
2917 * Caution: while qdict_get_try_str() is fine, getting non-string
2918 * types would require more care. When @options come from
2919 * -blockdev or blockdev_add, its members are typed according to
2920 * the QAPI schema, but when they come from -drive, they're all
2923 reference
= qdict_get_try_str(options
, bdref_key
);
2924 if (!filename
&& !reference
&& !qdict_size(image_options
)) {
2926 error_setg(errp
, "A block device must be specified for \"%s\"",
2929 qobject_unref(image_options
);
2933 bs
= bdrv_open_inherit(filename
, reference
, image_options
, 0,
2934 parent
, child_role
, errp
);
2940 qdict_del(options
, bdref_key
);
2945 * Opens a disk image whose options are given as BlockdevRef in another block
2948 * If allow_none is true, no image will be opened if filename is false and no
2949 * BlockdevRef is given. NULL will be returned, but errp remains unset.
2951 * bdrev_key specifies the key for the image's BlockdevRef in the options QDict.
2952 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
2953 * itself, all options starting with "${bdref_key}." are considered part of the
2956 * The BlockdevRef will be removed from the options QDict.
2958 BdrvChild
*bdrv_open_child(const char *filename
,
2959 QDict
*options
, const char *bdref_key
,
2960 BlockDriverState
*parent
,
2961 const BdrvChildRole
*child_role
,
2962 bool allow_none
, Error
**errp
)
2964 BlockDriverState
*bs
;
2966 bs
= bdrv_open_child_bs(filename
, options
, bdref_key
, parent
, child_role
,
2972 return bdrv_attach_child(parent
, bs
, bdref_key
, child_role
, errp
);
2975 /* TODO Future callers may need to specify parent/child_role in order for
2976 * option inheritance to work. Existing callers use it for the root node. */
2977 BlockDriverState
*bdrv_open_blockdev_ref(BlockdevRef
*ref
, Error
**errp
)
2979 BlockDriverState
*bs
= NULL
;
2980 Error
*local_err
= NULL
;
2981 QObject
*obj
= NULL
;
2982 QDict
*qdict
= NULL
;
2983 const char *reference
= NULL
;
2986 if (ref
->type
== QTYPE_QSTRING
) {
2987 reference
= ref
->u
.reference
;
2989 BlockdevOptions
*options
= &ref
->u
.definition
;
2990 assert(ref
->type
== QTYPE_QDICT
);
2992 v
= qobject_output_visitor_new(&obj
);
2993 visit_type_BlockdevOptions(v
, NULL
, &options
, &local_err
);
2995 error_propagate(errp
, local_err
);
2998 visit_complete(v
, &obj
);
3000 qdict
= qobject_to(QDict
, obj
);
3001 qdict_flatten(qdict
);
3003 /* bdrv_open_inherit() defaults to the values in bdrv_flags (for
3004 * compatibility with other callers) rather than what we want as the
3005 * real defaults. Apply the defaults here instead. */
3006 qdict_set_default_str(qdict
, BDRV_OPT_CACHE_DIRECT
, "off");
3007 qdict_set_default_str(qdict
, BDRV_OPT_CACHE_NO_FLUSH
, "off");
3008 qdict_set_default_str(qdict
, BDRV_OPT_READ_ONLY
, "off");
3009 qdict_set_default_str(qdict
, BDRV_OPT_AUTO_READ_ONLY
, "off");
3013 bs
= bdrv_open_inherit(NULL
, reference
, qdict
, 0, NULL
, NULL
, errp
);
3022 static BlockDriverState
*bdrv_append_temp_snapshot(BlockDriverState
*bs
,
3024 QDict
*snapshot_options
,
3027 /* TODO: extra byte is a hack to ensure MAX_PATH space on Windows. */
3028 char *tmp_filename
= g_malloc0(PATH_MAX
+ 1);
3030 QemuOpts
*opts
= NULL
;
3031 BlockDriverState
*bs_snapshot
= NULL
;
3032 Error
*local_err
= NULL
;
3035 /* if snapshot, we create a temporary backing file and open it
3036 instead of opening 'filename' directly */
3038 /* Get the required size from the image */
3039 total_size
= bdrv_getlength(bs
);
3040 if (total_size
< 0) {
3041 error_setg_errno(errp
, -total_size
, "Could not get image size");
3045 /* Create the temporary image */
3046 ret
= get_tmp_filename(tmp_filename
, PATH_MAX
+ 1);
3048 error_setg_errno(errp
, -ret
, "Could not get temporary filename");
3052 opts
= qemu_opts_create(bdrv_qcow2
.create_opts
, NULL
, 0,
3054 qemu_opt_set_number(opts
, BLOCK_OPT_SIZE
, total_size
, &error_abort
);
3055 ret
= bdrv_create(&bdrv_qcow2
, tmp_filename
, opts
, errp
);
3056 qemu_opts_del(opts
);
3058 error_prepend(errp
, "Could not create temporary overlay '%s': ",
3063 /* Prepare options QDict for the temporary file */
3064 qdict_put_str(snapshot_options
, "file.driver", "file");
3065 qdict_put_str(snapshot_options
, "file.filename", tmp_filename
);
3066 qdict_put_str(snapshot_options
, "driver", "qcow2");
3068 bs_snapshot
= bdrv_open(NULL
, NULL
, snapshot_options
, flags
, errp
);
3069 snapshot_options
= NULL
;
3074 /* bdrv_append() consumes a strong reference to bs_snapshot
3075 * (i.e. it will call bdrv_unref() on it) even on error, so in
3076 * order to be able to return one, we have to increase
3077 * bs_snapshot's refcount here */
3078 bdrv_ref(bs_snapshot
);
3079 bdrv_append(bs_snapshot
, bs
, &local_err
);
3081 error_propagate(errp
, local_err
);
3087 qobject_unref(snapshot_options
);
3088 g_free(tmp_filename
);
3093 * Opens a disk image (raw, qcow2, vmdk, ...)
3095 * options is a QDict of options to pass to the block drivers, or NULL for an
3096 * empty set of options. The reference to the QDict belongs to the block layer
3097 * after the call (even on failure), so if the caller intends to reuse the
3098 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
3100 * If *pbs is NULL, a new BDS will be created with a pointer to it stored there.
3101 * If it is not NULL, the referenced BDS will be reused.
3103 * The reference parameter may be used to specify an existing block device which
3104 * should be opened. If specified, neither options nor a filename may be given,
3105 * nor can an existing BDS be reused (that is, *pbs has to be NULL).
3107 static BlockDriverState
*bdrv_open_inherit(const char *filename
,
3108 const char *reference
,
3109 QDict
*options
, int flags
,
3110 BlockDriverState
*parent
,
3111 const BdrvChildRole
*child_role
,
3115 BlockBackend
*file
= NULL
;
3116 BlockDriverState
*bs
;
3117 BlockDriver
*drv
= NULL
;
3119 const char *drvname
;
3120 const char *backing
;
3121 Error
*local_err
= NULL
;
3122 QDict
*snapshot_options
= NULL
;
3123 int snapshot_flags
= 0;
3125 assert(!child_role
|| !flags
);
3126 assert(!child_role
== !parent
);
3129 bool options_non_empty
= options
? qdict_size(options
) : false;
3130 qobject_unref(options
);
3132 if (filename
|| options_non_empty
) {
3133 error_setg(errp
, "Cannot reference an existing block device with "
3134 "additional options or a new filename");
3138 bs
= bdrv_lookup_bs(reference
, reference
, errp
);
3149 /* NULL means an empty set of options */
3150 if (options
== NULL
) {
3151 options
= qdict_new();
3154 /* json: syntax counts as explicit options, as if in the QDict */
3155 parse_json_protocol(options
, &filename
, &local_err
);
3160 bs
->explicit_options
= qdict_clone_shallow(options
);
3163 bs
->inherits_from
= parent
;
3164 child_role
->inherit_options(&flags
, options
,
3165 parent
->open_flags
, parent
->options
);
3168 ret
= bdrv_fill_options(&options
, filename
, &flags
, &local_err
);
3174 * Set the BDRV_O_RDWR and BDRV_O_ALLOW_RDWR flags.
3175 * Caution: getting a boolean member of @options requires care.
3176 * When @options come from -blockdev or blockdev_add, members are
3177 * typed according to the QAPI schema, but when they come from
3178 * -drive, they're all QString.
3180 if (g_strcmp0(qdict_get_try_str(options
, BDRV_OPT_READ_ONLY
), "on") &&
3181 !qdict_get_try_bool(options
, BDRV_OPT_READ_ONLY
, false)) {
3182 flags
|= (BDRV_O_RDWR
| BDRV_O_ALLOW_RDWR
);
3184 flags
&= ~BDRV_O_RDWR
;
3187 if (flags
& BDRV_O_SNAPSHOT
) {
3188 snapshot_options
= qdict_new();
3189 bdrv_temp_snapshot_options(&snapshot_flags
, snapshot_options
,
3191 /* Let bdrv_backing_options() override "read-only" */
3192 qdict_del(options
, BDRV_OPT_READ_ONLY
);
3193 bdrv_backing_options(&flags
, options
, flags
, options
);
3196 bs
->open_flags
= flags
;
3197 bs
->options
= options
;
3198 options
= qdict_clone_shallow(options
);
3200 /* Find the right image format driver */
3201 /* See cautionary note on accessing @options above */
3202 drvname
= qdict_get_try_str(options
, "driver");
3204 drv
= bdrv_find_format(drvname
);
3206 error_setg(errp
, "Unknown driver: '%s'", drvname
);
3211 assert(drvname
|| !(flags
& BDRV_O_PROTOCOL
));
3213 /* See cautionary note on accessing @options above */
3214 backing
= qdict_get_try_str(options
, "backing");
3215 if (qobject_to(QNull
, qdict_get(options
, "backing")) != NULL
||
3216 (backing
&& *backing
== '\0'))
3219 warn_report("Use of \"backing\": \"\" is deprecated; "
3220 "use \"backing\": null instead");
3222 flags
|= BDRV_O_NO_BACKING
;
3223 qdict_del(bs
->explicit_options
, "backing");
3224 qdict_del(bs
->options
, "backing");
3225 qdict_del(options
, "backing");
3228 /* Open image file without format layer. This BlockBackend is only used for
3229 * probing, the block drivers will do their own bdrv_open_child() for the
3230 * same BDS, which is why we put the node name back into options. */
3231 if ((flags
& BDRV_O_PROTOCOL
) == 0) {
3232 BlockDriverState
*file_bs
;
3234 file_bs
= bdrv_open_child_bs(filename
, options
, "file", bs
,
3235 &child_file
, true, &local_err
);
3239 if (file_bs
!= NULL
) {
3240 /* Not requesting BLK_PERM_CONSISTENT_READ because we're only
3241 * looking at the header to guess the image format. This works even
3242 * in cases where a guest would not see a consistent state. */
3243 file
= blk_new(bdrv_get_aio_context(file_bs
), 0, BLK_PERM_ALL
);
3244 blk_insert_bs(file
, file_bs
, &local_err
);
3245 bdrv_unref(file_bs
);
3250 qdict_put_str(options
, "file", bdrv_get_node_name(file_bs
));
3254 /* Image format probing */
3257 ret
= find_image_format(file
, filename
, &drv
, &local_err
);
3262 * This option update would logically belong in bdrv_fill_options(),
3263 * but we first need to open bs->file for the probing to work, while
3264 * opening bs->file already requires the (mostly) final set of options
3265 * so that cache mode etc. can be inherited.
3267 * Adding the driver later is somewhat ugly, but it's not an option
3268 * that would ever be inherited, so it's correct. We just need to make
3269 * sure to update both bs->options (which has the full effective
3270 * options for bs) and options (which has file.* already removed).
3272 qdict_put_str(bs
->options
, "driver", drv
->format_name
);
3273 qdict_put_str(options
, "driver", drv
->format_name
);
3275 error_setg(errp
, "Must specify either driver or file");
3279 /* BDRV_O_PROTOCOL must be set iff a protocol BDS is about to be created */
3280 assert(!!(flags
& BDRV_O_PROTOCOL
) == !!drv
->bdrv_file_open
);
3281 /* file must be NULL if a protocol BDS is about to be created
3282 * (the inverse results in an error message from bdrv_open_common()) */
3283 assert(!(flags
& BDRV_O_PROTOCOL
) || !file
);
3285 /* Open the image */
3286 ret
= bdrv_open_common(bs
, file
, options
, &local_err
);
3296 /* If there is a backing file, use it */
3297 if ((flags
& BDRV_O_NO_BACKING
) == 0) {
3298 ret
= bdrv_open_backing_file(bs
, options
, "backing", &local_err
);
3300 goto close_and_fail
;
3304 /* Remove all children options and references
3305 * from bs->options and bs->explicit_options */
3306 QLIST_FOREACH(child
, &bs
->children
, next
) {
3307 char *child_key_dot
;
3308 child_key_dot
= g_strdup_printf("%s.", child
->name
);
3309 qdict_extract_subqdict(bs
->explicit_options
, NULL
, child_key_dot
);
3310 qdict_extract_subqdict(bs
->options
, NULL
, child_key_dot
);
3311 qdict_del(bs
->explicit_options
, child
->name
);
3312 qdict_del(bs
->options
, child
->name
);
3313 g_free(child_key_dot
);
3316 /* Check if any unknown options were used */
3317 if (qdict_size(options
) != 0) {
3318 const QDictEntry
*entry
= qdict_first(options
);
3319 if (flags
& BDRV_O_PROTOCOL
) {
3320 error_setg(errp
, "Block protocol '%s' doesn't support the option "
3321 "'%s'", drv
->format_name
, entry
->key
);
3324 "Block format '%s' does not support the option '%s'",
3325 drv
->format_name
, entry
->key
);
3328 goto close_and_fail
;
3331 bdrv_parent_cb_change_media(bs
, true);
3333 qobject_unref(options
);
3336 /* For snapshot=on, create a temporary qcow2 overlay. bs points to the
3337 * temporary snapshot afterwards. */
3338 if (snapshot_flags
) {
3339 BlockDriverState
*snapshot_bs
;
3340 snapshot_bs
= bdrv_append_temp_snapshot(bs
, snapshot_flags
,
3341 snapshot_options
, &local_err
);
3342 snapshot_options
= NULL
;
3344 goto close_and_fail
;
3346 /* We are not going to return bs but the overlay on top of it
3347 * (snapshot_bs); thus, we have to drop the strong reference to bs
3348 * (which we obtained by calling bdrv_new()). bs will not be deleted,
3349 * though, because the overlay still has a reference to it. */
3358 qobject_unref(snapshot_options
);
3359 qobject_unref(bs
->explicit_options
);
3360 qobject_unref(bs
->options
);
3361 qobject_unref(options
);
3363 bs
->explicit_options
= NULL
;
3365 error_propagate(errp
, local_err
);
3370 qobject_unref(snapshot_options
);
3371 qobject_unref(options
);
3372 error_propagate(errp
, local_err
);
3376 BlockDriverState
*bdrv_open(const char *filename
, const char *reference
,
3377 QDict
*options
, int flags
, Error
**errp
)
3379 return bdrv_open_inherit(filename
, reference
, options
, flags
, NULL
,
3383 /* Return true if the NULL-terminated @list contains @str */
3384 static bool is_str_in_list(const char *str
, const char *const *list
)
3388 for (i
= 0; list
[i
] != NULL
; i
++) {
3389 if (!strcmp(str
, list
[i
])) {
3398 * Check that every option set in @bs->options is also set in
3401 * Options listed in the common_options list and in
3402 * @bs->drv->mutable_opts are skipped.
3404 * Return 0 on success, otherwise return -EINVAL and set @errp.
3406 static int bdrv_reset_options_allowed(BlockDriverState
*bs
,
3407 const QDict
*new_opts
, Error
**errp
)
3409 const QDictEntry
*e
;
3410 /* These options are common to all block drivers and are handled
3411 * in bdrv_reopen_prepare() so they can be left out of @new_opts */
3412 const char *const common_options
[] = {
3413 "node-name", "discard", "cache.direct", "cache.no-flush",
3414 "read-only", "auto-read-only", "detect-zeroes", NULL
3417 for (e
= qdict_first(bs
->options
); e
; e
= qdict_next(bs
->options
, e
)) {
3418 if (!qdict_haskey(new_opts
, e
->key
) &&
3419 !is_str_in_list(e
->key
, common_options
) &&
3420 !is_str_in_list(e
->key
, bs
->drv
->mutable_opts
)) {
3421 error_setg(errp
, "Option '%s' cannot be reset "
3422 "to its default value", e
->key
);
3431 * Returns true if @child can be reached recursively from @bs
3433 static bool bdrv_recurse_has_child(BlockDriverState
*bs
,
3434 BlockDriverState
*child
)
3442 QLIST_FOREACH(c
, &bs
->children
, next
) {
3443 if (bdrv_recurse_has_child(c
->bs
, child
)) {
3452 * Adds a BlockDriverState to a simple queue for an atomic, transactional
3453 * reopen of multiple devices.
3455 * bs_queue can either be an existing BlockReopenQueue that has had QTAILQ_INIT
3456 * already performed, or alternatively may be NULL a new BlockReopenQueue will
3457 * be created and initialized. This newly created BlockReopenQueue should be
3458 * passed back in for subsequent calls that are intended to be of the same
3461 * bs is the BlockDriverState to add to the reopen queue.
3463 * options contains the changed options for the associated bs
3464 * (the BlockReopenQueue takes ownership)
3466 * flags contains the open flags for the associated bs
3468 * returns a pointer to bs_queue, which is either the newly allocated
3469 * bs_queue, or the existing bs_queue being used.
3471 * bs must be drained between bdrv_reopen_queue() and bdrv_reopen_multiple().
3473 static BlockReopenQueue
*bdrv_reopen_queue_child(BlockReopenQueue
*bs_queue
,
3474 BlockDriverState
*bs
,
3476 const BdrvChildRole
*role
,
3477 QDict
*parent_options
,
3483 BlockReopenQueueEntry
*bs_entry
;
3485 QDict
*old_options
, *explicit_options
, *options_copy
;
3489 /* Make sure that the caller remembered to use a drained section. This is
3490 * important to avoid graph changes between the recursive queuing here and
3491 * bdrv_reopen_multiple(). */
3492 assert(bs
->quiesce_counter
> 0);
3494 if (bs_queue
== NULL
) {
3495 bs_queue
= g_new0(BlockReopenQueue
, 1);
3496 QTAILQ_INIT(bs_queue
);
3500 options
= qdict_new();
3503 /* Check if this BlockDriverState is already in the queue */
3504 QTAILQ_FOREACH(bs_entry
, bs_queue
, entry
) {
3505 if (bs
== bs_entry
->state
.bs
) {
3511 * Precedence of options:
3512 * 1. Explicitly passed in options (highest)
3513 * 2. Retained from explicitly set options of bs
3514 * 3. Inherited from parent node
3515 * 4. Retained from effective options of bs
3518 /* Old explicitly set values (don't overwrite by inherited value) */
3519 if (bs_entry
|| keep_old_opts
) {
3520 old_options
= qdict_clone_shallow(bs_entry
?
3521 bs_entry
->state
.explicit_options
:
3522 bs
->explicit_options
);
3523 bdrv_join_options(bs
, options
, old_options
);
3524 qobject_unref(old_options
);
3527 explicit_options
= qdict_clone_shallow(options
);
3529 /* Inherit from parent node */
3530 if (parent_options
) {
3532 role
->inherit_options(&flags
, options
, parent_flags
, parent_options
);
3534 flags
= bdrv_get_flags(bs
);
3537 if (keep_old_opts
) {
3538 /* Old values are used for options that aren't set yet */
3539 old_options
= qdict_clone_shallow(bs
->options
);
3540 bdrv_join_options(bs
, options
, old_options
);
3541 qobject_unref(old_options
);
3544 /* We have the final set of options so let's update the flags */
3545 options_copy
= qdict_clone_shallow(options
);
3546 opts
= qemu_opts_create(&bdrv_runtime_opts
, NULL
, 0, &error_abort
);
3547 qemu_opts_absorb_qdict(opts
, options_copy
, NULL
);
3548 update_flags_from_options(&flags
, opts
);
3549 qemu_opts_del(opts
);
3550 qobject_unref(options_copy
);
3552 /* bdrv_open_inherit() sets and clears some additional flags internally */
3553 flags
&= ~BDRV_O_PROTOCOL
;
3554 if (flags
& BDRV_O_RDWR
) {
3555 flags
|= BDRV_O_ALLOW_RDWR
;
3559 bs_entry
= g_new0(BlockReopenQueueEntry
, 1);
3560 QTAILQ_INSERT_TAIL(bs_queue
, bs_entry
, entry
);
3562 qobject_unref(bs_entry
->state
.options
);
3563 qobject_unref(bs_entry
->state
.explicit_options
);
3566 bs_entry
->state
.bs
= bs
;
3567 bs_entry
->state
.options
= options
;
3568 bs_entry
->state
.explicit_options
= explicit_options
;
3569 bs_entry
->state
.flags
= flags
;
3571 /* This needs to be overwritten in bdrv_reopen_prepare() */
3572 bs_entry
->state
.perm
= UINT64_MAX
;
3573 bs_entry
->state
.shared_perm
= 0;
3576 * If keep_old_opts is false then it means that unspecified
3577 * options must be reset to their original value. We don't allow
3578 * resetting 'backing' but we need to know if the option is
3579 * missing in order to decide if we have to return an error.
3581 if (!keep_old_opts
) {
3582 bs_entry
->state
.backing_missing
=
3583 !qdict_haskey(options
, "backing") &&
3584 !qdict_haskey(options
, "backing.driver");
3587 QLIST_FOREACH(child
, &bs
->children
, next
) {
3588 QDict
*new_child_options
= NULL
;
3589 bool child_keep_old
= keep_old_opts
;
3591 /* reopen can only change the options of block devices that were
3592 * implicitly created and inherited options. For other (referenced)
3593 * block devices, a syntax like "backing.foo" results in an error. */
3594 if (child
->bs
->inherits_from
!= bs
) {
3598 /* Check if the options contain a child reference */
3599 if (qdict_haskey(options
, child
->name
)) {
3600 const char *childref
= qdict_get_try_str(options
, child
->name
);
3602 * The current child must not be reopened if the child
3603 * reference is null or points to a different node.
3605 if (g_strcmp0(childref
, child
->bs
->node_name
)) {
3609 * If the child reference points to the current child then
3610 * reopen it with its existing set of options (note that
3611 * it can still inherit new options from the parent).
3613 child_keep_old
= true;
3615 /* Extract child options ("child-name.*") */
3616 char *child_key_dot
= g_strdup_printf("%s.", child
->name
);
3617 qdict_extract_subqdict(explicit_options
, NULL
, child_key_dot
);
3618 qdict_extract_subqdict(options
, &new_child_options
, child_key_dot
);
3619 g_free(child_key_dot
);
3622 bdrv_reopen_queue_child(bs_queue
, child
->bs
, new_child_options
,
3623 child
->role
, options
, flags
, child_keep_old
);
3629 BlockReopenQueue
*bdrv_reopen_queue(BlockReopenQueue
*bs_queue
,
3630 BlockDriverState
*bs
,
3631 QDict
*options
, bool keep_old_opts
)
3633 return bdrv_reopen_queue_child(bs_queue
, bs
, options
, NULL
, NULL
, 0,
3638 * Reopen multiple BlockDriverStates atomically & transactionally.
3640 * The queue passed in (bs_queue) must have been built up previous
3641 * via bdrv_reopen_queue().
3643 * Reopens all BDS specified in the queue, with the appropriate
3644 * flags. All devices are prepared for reopen, and failure of any
3645 * device will cause all device changes to be abandoned, and intermediate
3648 * If all devices prepare successfully, then the changes are committed
3651 * All affected nodes must be drained between bdrv_reopen_queue() and
3652 * bdrv_reopen_multiple().
3654 int bdrv_reopen_multiple(BlockReopenQueue
*bs_queue
, Error
**errp
)
3657 BlockReopenQueueEntry
*bs_entry
, *next
;
3659 assert(bs_queue
!= NULL
);
3661 QTAILQ_FOREACH(bs_entry
, bs_queue
, entry
) {
3662 assert(bs_entry
->state
.bs
->quiesce_counter
> 0);
3663 if (bdrv_reopen_prepare(&bs_entry
->state
, bs_queue
, errp
)) {
3666 bs_entry
->prepared
= true;
3669 QTAILQ_FOREACH(bs_entry
, bs_queue
, entry
) {
3670 BDRVReopenState
*state
= &bs_entry
->state
;
3671 ret
= bdrv_check_perm(state
->bs
, bs_queue
, state
->perm
,
3672 state
->shared_perm
, NULL
, NULL
, errp
);
3676 /* Check if new_backing_bs would accept the new permissions */
3677 if (state
->replace_backing_bs
&& state
->new_backing_bs
) {
3678 uint64_t nperm
, nshared
;
3679 bdrv_child_perm(state
->bs
, state
->new_backing_bs
,
3680 NULL
, &child_backing
, bs_queue
,
3681 state
->perm
, state
->shared_perm
,
3683 ret
= bdrv_check_update_perm(state
->new_backing_bs
, NULL
,
3684 nperm
, nshared
, NULL
, NULL
, errp
);
3689 bs_entry
->perms_checked
= true;
3693 * If we reach this point, we have success and just need to apply the
3696 * Reverse order is used to comfort qcow2 driver: on commit it need to write
3697 * IN_USE flag to the image, to mark bitmaps in the image as invalid. But
3698 * children are usually goes after parents in reopen-queue, so go from last
3701 QTAILQ_FOREACH_REVERSE(bs_entry
, bs_queue
, entry
) {
3702 bdrv_reopen_commit(&bs_entry
->state
);
3707 QTAILQ_FOREACH_SAFE(bs_entry
, bs_queue
, entry
, next
) {
3708 BDRVReopenState
*state
= &bs_entry
->state
;
3710 if (!bs_entry
->perms_checked
) {
3715 bdrv_set_perm(state
->bs
, state
->perm
, state
->shared_perm
);
3717 bdrv_abort_perm_update(state
->bs
);
3718 if (state
->replace_backing_bs
&& state
->new_backing_bs
) {
3719 bdrv_abort_perm_update(state
->new_backing_bs
);
3725 QTAILQ_FOREACH_REVERSE(bs_entry
, bs_queue
, entry
) {
3726 BlockDriverState
*bs
= bs_entry
->state
.bs
;
3728 if (bs
->drv
->bdrv_reopen_commit_post
)
3729 bs
->drv
->bdrv_reopen_commit_post(&bs_entry
->state
);
3733 QTAILQ_FOREACH_SAFE(bs_entry
, bs_queue
, entry
, next
) {
3735 if (bs_entry
->prepared
) {
3736 bdrv_reopen_abort(&bs_entry
->state
);
3738 qobject_unref(bs_entry
->state
.explicit_options
);
3739 qobject_unref(bs_entry
->state
.options
);
3741 if (bs_entry
->state
.new_backing_bs
) {
3742 bdrv_unref(bs_entry
->state
.new_backing_bs
);
3751 int bdrv_reopen_set_read_only(BlockDriverState
*bs
, bool read_only
,
3755 BlockReopenQueue
*queue
;
3756 QDict
*opts
= qdict_new();
3758 qdict_put_bool(opts
, BDRV_OPT_READ_ONLY
, read_only
);
3760 bdrv_subtree_drained_begin(bs
);
3761 queue
= bdrv_reopen_queue(NULL
, bs
, opts
, true);
3762 ret
= bdrv_reopen_multiple(queue
, errp
);
3763 bdrv_subtree_drained_end(bs
);
3768 static BlockReopenQueueEntry
*find_parent_in_reopen_queue(BlockReopenQueue
*q
,
3771 BlockReopenQueueEntry
*entry
;
3773 QTAILQ_FOREACH(entry
, q
, entry
) {
3774 BlockDriverState
*bs
= entry
->state
.bs
;
3777 QLIST_FOREACH(child
, &bs
->children
, next
) {
3787 static void bdrv_reopen_perm(BlockReopenQueue
*q
, BlockDriverState
*bs
,
3788 uint64_t *perm
, uint64_t *shared
)
3791 BlockReopenQueueEntry
*parent
;
3792 uint64_t cumulative_perms
= 0;
3793 uint64_t cumulative_shared_perms
= BLK_PERM_ALL
;
3795 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
3796 parent
= find_parent_in_reopen_queue(q
, c
);
3798 cumulative_perms
|= c
->perm
;
3799 cumulative_shared_perms
&= c
->shared_perm
;
3801 uint64_t nperm
, nshared
;
3803 bdrv_child_perm(parent
->state
.bs
, bs
, c
, c
->role
, q
,
3804 parent
->state
.perm
, parent
->state
.shared_perm
,
3807 cumulative_perms
|= nperm
;
3808 cumulative_shared_perms
&= nshared
;
3811 *perm
= cumulative_perms
;
3812 *shared
= cumulative_shared_perms
;
3815 static bool bdrv_reopen_can_attach(BlockDriverState
*parent
,
3817 BlockDriverState
*new_child
,
3820 AioContext
*parent_ctx
= bdrv_get_aio_context(parent
);
3821 AioContext
*child_ctx
= bdrv_get_aio_context(new_child
);
3825 ignore
= g_slist_prepend(NULL
, child
);
3826 ret
= bdrv_can_set_aio_context(new_child
, parent_ctx
, &ignore
, NULL
);
3827 g_slist_free(ignore
);
3832 ignore
= g_slist_prepend(NULL
, child
);
3833 ret
= bdrv_can_set_aio_context(parent
, child_ctx
, &ignore
, errp
);
3834 g_slist_free(ignore
);
3839 * Take a BDRVReopenState and check if the value of 'backing' in the
3840 * reopen_state->options QDict is valid or not.
3842 * If 'backing' is missing from the QDict then return 0.
3844 * If 'backing' contains the node name of the backing file of
3845 * reopen_state->bs then return 0.
3847 * If 'backing' contains a different node name (or is null) then check
3848 * whether the current backing file can be replaced with the new one.
3849 * If that's the case then reopen_state->replace_backing_bs is set to
3850 * true and reopen_state->new_backing_bs contains a pointer to the new
3851 * backing BlockDriverState (or NULL).
3853 * Return 0 on success, otherwise return < 0 and set @errp.
3855 static int bdrv_reopen_parse_backing(BDRVReopenState
*reopen_state
,
3858 BlockDriverState
*bs
= reopen_state
->bs
;
3859 BlockDriverState
*overlay_bs
, *new_backing_bs
;
3863 value
= qdict_get(reopen_state
->options
, "backing");
3864 if (value
== NULL
) {
3868 switch (qobject_type(value
)) {
3870 new_backing_bs
= NULL
;
3873 str
= qobject_get_try_str(value
);
3874 new_backing_bs
= bdrv_lookup_bs(NULL
, str
, errp
);
3875 if (new_backing_bs
== NULL
) {
3877 } else if (bdrv_recurse_has_child(new_backing_bs
, bs
)) {
3878 error_setg(errp
, "Making '%s' a backing file of '%s' "
3879 "would create a cycle", str
, bs
->node_name
);
3884 /* 'backing' does not allow any other data type */
3885 g_assert_not_reached();
3889 * Check AioContext compatibility so that the bdrv_set_backing_hd() call in
3890 * bdrv_reopen_commit() won't fail.
3892 if (new_backing_bs
) {
3893 if (!bdrv_reopen_can_attach(bs
, bs
->backing
, new_backing_bs
, errp
)) {
3899 * Find the "actual" backing file by skipping all links that point
3900 * to an implicit node, if any (e.g. a commit filter node).
3903 while (backing_bs(overlay_bs
) && backing_bs(overlay_bs
)->implicit
) {
3904 overlay_bs
= backing_bs(overlay_bs
);
3907 /* If we want to replace the backing file we need some extra checks */
3908 if (new_backing_bs
!= backing_bs(overlay_bs
)) {
3909 /* Check for implicit nodes between bs and its backing file */
3910 if (bs
!= overlay_bs
) {
3911 error_setg(errp
, "Cannot change backing link if '%s' has "
3912 "an implicit backing file", bs
->node_name
);
3915 /* Check if the backing link that we want to replace is frozen */
3916 if (bdrv_is_backing_chain_frozen(overlay_bs
, backing_bs(overlay_bs
),
3920 reopen_state
->replace_backing_bs
= true;
3921 if (new_backing_bs
) {
3922 bdrv_ref(new_backing_bs
);
3923 reopen_state
->new_backing_bs
= new_backing_bs
;
3931 * Prepares a BlockDriverState for reopen. All changes are staged in the
3932 * 'opaque' field of the BDRVReopenState, which is used and allocated by
3933 * the block driver layer .bdrv_reopen_prepare()
3935 * bs is the BlockDriverState to reopen
3936 * flags are the new open flags
3937 * queue is the reopen queue
3939 * Returns 0 on success, non-zero on error. On error errp will be set
3942 * On failure, bdrv_reopen_abort() will be called to clean up any data.
3943 * It is the responsibility of the caller to then call the abort() or
3944 * commit() for any other BDS that have been left in a prepare() state
3947 int bdrv_reopen_prepare(BDRVReopenState
*reopen_state
, BlockReopenQueue
*queue
,
3952 Error
*local_err
= NULL
;
3955 QDict
*orig_reopen_opts
;
3956 char *discard
= NULL
;
3958 bool drv_prepared
= false;
3960 assert(reopen_state
!= NULL
);
3961 assert(reopen_state
->bs
->drv
!= NULL
);
3962 drv
= reopen_state
->bs
->drv
;
3964 /* This function and each driver's bdrv_reopen_prepare() remove
3965 * entries from reopen_state->options as they are processed, so
3966 * we need to make a copy of the original QDict. */
3967 orig_reopen_opts
= qdict_clone_shallow(reopen_state
->options
);
3969 /* Process generic block layer options */
3970 opts
= qemu_opts_create(&bdrv_runtime_opts
, NULL
, 0, &error_abort
);
3971 qemu_opts_absorb_qdict(opts
, reopen_state
->options
, &local_err
);
3973 error_propagate(errp
, local_err
);
3978 /* This was already called in bdrv_reopen_queue_child() so the flags
3979 * are up-to-date. This time we simply want to remove the options from
3980 * QemuOpts in order to indicate that they have been processed. */
3981 old_flags
= reopen_state
->flags
;
3982 update_flags_from_options(&reopen_state
->flags
, opts
);
3983 assert(old_flags
== reopen_state
->flags
);
3985 discard
= qemu_opt_get_del(opts
, BDRV_OPT_DISCARD
);
3986 if (discard
!= NULL
) {
3987 if (bdrv_parse_discard_flags(discard
, &reopen_state
->flags
) != 0) {
3988 error_setg(errp
, "Invalid discard option");
3994 reopen_state
->detect_zeroes
=
3995 bdrv_parse_detect_zeroes(opts
, reopen_state
->flags
, &local_err
);
3997 error_propagate(errp
, local_err
);
4002 /* All other options (including node-name and driver) must be unchanged.
4003 * Put them back into the QDict, so that they are checked at the end
4004 * of this function. */
4005 qemu_opts_to_qdict(opts
, reopen_state
->options
);
4007 /* If we are to stay read-only, do not allow permission change
4008 * to r/w. Attempting to set to r/w may fail if either BDRV_O_ALLOW_RDWR is
4009 * not set, or if the BDS still has copy_on_read enabled */
4010 read_only
= !(reopen_state
->flags
& BDRV_O_RDWR
);
4011 ret
= bdrv_can_set_read_only(reopen_state
->bs
, read_only
, true, &local_err
);
4013 error_propagate(errp
, local_err
);
4017 /* Calculate required permissions after reopening */
4018 bdrv_reopen_perm(queue
, reopen_state
->bs
,
4019 &reopen_state
->perm
, &reopen_state
->shared_perm
);
4021 ret
= bdrv_flush(reopen_state
->bs
);
4023 error_setg_errno(errp
, -ret
, "Error flushing drive");
4027 if (drv
->bdrv_reopen_prepare
) {
4029 * If a driver-specific option is missing, it means that we
4030 * should reset it to its default value.
4031 * But not all options allow that, so we need to check it first.
4033 ret
= bdrv_reset_options_allowed(reopen_state
->bs
,
4034 reopen_state
->options
, errp
);
4039 ret
= drv
->bdrv_reopen_prepare(reopen_state
, queue
, &local_err
);
4041 if (local_err
!= NULL
) {
4042 error_propagate(errp
, local_err
);
4044 bdrv_refresh_filename(reopen_state
->bs
);
4045 error_setg(errp
, "failed while preparing to reopen image '%s'",
4046 reopen_state
->bs
->filename
);
4051 /* It is currently mandatory to have a bdrv_reopen_prepare()
4052 * handler for each supported drv. */
4053 error_setg(errp
, "Block format '%s' used by node '%s' "
4054 "does not support reopening files", drv
->format_name
,
4055 bdrv_get_device_or_node_name(reopen_state
->bs
));
4060 drv_prepared
= true;
4063 * We must provide the 'backing' option if the BDS has a backing
4064 * file or if the image file has a backing file name as part of
4065 * its metadata. Otherwise the 'backing' option can be omitted.
4067 if (drv
->supports_backing
&& reopen_state
->backing_missing
&&
4068 (backing_bs(reopen_state
->bs
) || reopen_state
->bs
->backing_file
[0])) {
4069 error_setg(errp
, "backing is missing for '%s'",
4070 reopen_state
->bs
->node_name
);
4076 * Allow changing the 'backing' option. The new value can be
4077 * either a reference to an existing node (using its node name)
4078 * or NULL to simply detach the current backing file.
4080 ret
= bdrv_reopen_parse_backing(reopen_state
, errp
);
4084 qdict_del(reopen_state
->options
, "backing");
4086 /* Options that are not handled are only okay if they are unchanged
4087 * compared to the old state. It is expected that some options are only
4088 * used for the initial open, but not reopen (e.g. filename) */
4089 if (qdict_size(reopen_state
->options
)) {
4090 const QDictEntry
*entry
= qdict_first(reopen_state
->options
);
4093 QObject
*new = entry
->value
;
4094 QObject
*old
= qdict_get(reopen_state
->bs
->options
, entry
->key
);
4096 /* Allow child references (child_name=node_name) as long as they
4097 * point to the current child (i.e. everything stays the same). */
4098 if (qobject_type(new) == QTYPE_QSTRING
) {
4100 QLIST_FOREACH(child
, &reopen_state
->bs
->children
, next
) {
4101 if (!strcmp(child
->name
, entry
->key
)) {
4107 const char *str
= qobject_get_try_str(new);
4108 if (!strcmp(child
->bs
->node_name
, str
)) {
4109 continue; /* Found child with this name, skip option */
4115 * TODO: When using -drive to specify blockdev options, all values
4116 * will be strings; however, when using -blockdev, blockdev-add or
4117 * filenames using the json:{} pseudo-protocol, they will be
4119 * In contrast, reopening options are (currently) always strings
4120 * (because you can only specify them through qemu-io; all other
4121 * callers do not specify any options).
4122 * Therefore, when using anything other than -drive to create a BDS,
4123 * this cannot detect non-string options as unchanged, because
4124 * qobject_is_equal() always returns false for objects of different
4125 * type. In the future, this should be remedied by correctly typing
4126 * all options. For now, this is not too big of an issue because
4127 * the user can simply omit options which cannot be changed anyway,
4128 * so they will stay unchanged.
4130 if (!qobject_is_equal(new, old
)) {
4131 error_setg(errp
, "Cannot change the option '%s'", entry
->key
);
4135 } while ((entry
= qdict_next(reopen_state
->options
, entry
)));
4140 /* Restore the original reopen_state->options QDict */
4141 qobject_unref(reopen_state
->options
);
4142 reopen_state
->options
= qobject_ref(orig_reopen_opts
);
4145 if (ret
< 0 && drv_prepared
) {
4146 /* drv->bdrv_reopen_prepare() has succeeded, so we need to
4147 * call drv->bdrv_reopen_abort() before signaling an error
4148 * (bdrv_reopen_multiple() will not call bdrv_reopen_abort()
4149 * when the respective bdrv_reopen_prepare() has failed) */
4150 if (drv
->bdrv_reopen_abort
) {
4151 drv
->bdrv_reopen_abort(reopen_state
);
4154 qemu_opts_del(opts
);
4155 qobject_unref(orig_reopen_opts
);
4161 * Takes the staged changes for the reopen from bdrv_reopen_prepare(), and
4162 * makes them final by swapping the staging BlockDriverState contents into
4163 * the active BlockDriverState contents.
4165 void bdrv_reopen_commit(BDRVReopenState
*reopen_state
)
4168 BlockDriverState
*bs
;
4171 assert(reopen_state
!= NULL
);
4172 bs
= reopen_state
->bs
;
4174 assert(drv
!= NULL
);
4176 /* If there are any driver level actions to take */
4177 if (drv
->bdrv_reopen_commit
) {
4178 drv
->bdrv_reopen_commit(reopen_state
);
4181 /* set BDS specific flags now */
4182 qobject_unref(bs
->explicit_options
);
4183 qobject_unref(bs
->options
);
4185 bs
->explicit_options
= reopen_state
->explicit_options
;
4186 bs
->options
= reopen_state
->options
;
4187 bs
->open_flags
= reopen_state
->flags
;
4188 bs
->read_only
= !(reopen_state
->flags
& BDRV_O_RDWR
);
4189 bs
->detect_zeroes
= reopen_state
->detect_zeroes
;
4191 if (reopen_state
->replace_backing_bs
) {
4192 qdict_del(bs
->explicit_options
, "backing");
4193 qdict_del(bs
->options
, "backing");
4196 /* Remove child references from bs->options and bs->explicit_options.
4197 * Child options were already removed in bdrv_reopen_queue_child() */
4198 QLIST_FOREACH(child
, &bs
->children
, next
) {
4199 qdict_del(bs
->explicit_options
, child
->name
);
4200 qdict_del(bs
->options
, child
->name
);
4204 * Change the backing file if a new one was specified. We do this
4205 * after updating bs->options, so bdrv_refresh_filename() (called
4206 * from bdrv_set_backing_hd()) has the new values.
4208 if (reopen_state
->replace_backing_bs
) {
4209 BlockDriverState
*old_backing_bs
= backing_bs(bs
);
4210 assert(!old_backing_bs
|| !old_backing_bs
->implicit
);
4211 /* Abort the permission update on the backing bs we're detaching */
4212 if (old_backing_bs
) {
4213 bdrv_abort_perm_update(old_backing_bs
);
4215 bdrv_set_backing_hd(bs
, reopen_state
->new_backing_bs
, &error_abort
);
4218 bdrv_refresh_limits(bs
, NULL
);
4222 * Abort the reopen, and delete and free the staged changes in
4225 void bdrv_reopen_abort(BDRVReopenState
*reopen_state
)
4229 assert(reopen_state
!= NULL
);
4230 drv
= reopen_state
->bs
->drv
;
4231 assert(drv
!= NULL
);
4233 if (drv
->bdrv_reopen_abort
) {
4234 drv
->bdrv_reopen_abort(reopen_state
);
4239 static void bdrv_close(BlockDriverState
*bs
)
4241 BdrvAioNotifier
*ban
, *ban_next
;
4242 BdrvChild
*child
, *next
;
4244 assert(!bs
->refcnt
);
4246 bdrv_drained_begin(bs
); /* complete I/O */
4248 bdrv_drain(bs
); /* in case flush left pending I/O */
4251 if (bs
->drv
->bdrv_close
) {
4252 bs
->drv
->bdrv_close(bs
);
4257 QLIST_FOREACH_SAFE(child
, &bs
->children
, next
, next
) {
4258 bdrv_unref_child(bs
, child
);
4265 atomic_set(&bs
->copy_on_read
, 0);
4266 bs
->backing_file
[0] = '\0';
4267 bs
->backing_format
[0] = '\0';
4268 bs
->total_sectors
= 0;
4269 bs
->encrypted
= false;
4271 qobject_unref(bs
->options
);
4272 qobject_unref(bs
->explicit_options
);
4274 bs
->explicit_options
= NULL
;
4275 qobject_unref(bs
->full_open_options
);
4276 bs
->full_open_options
= NULL
;
4278 bdrv_release_named_dirty_bitmaps(bs
);
4279 assert(QLIST_EMPTY(&bs
->dirty_bitmaps
));
4281 QLIST_FOREACH_SAFE(ban
, &bs
->aio_notifiers
, list
, ban_next
) {
4284 QLIST_INIT(&bs
->aio_notifiers
);
4285 bdrv_drained_end(bs
);
4288 void bdrv_close_all(void)
4290 assert(job_next(NULL
) == NULL
);
4291 nbd_export_close_all();
4293 /* Drop references from requests still in flight, such as canceled block
4294 * jobs whose AIO context has not been polled yet */
4297 blk_remove_all_bs();
4298 blockdev_close_all_bdrv_states();
4300 assert(QTAILQ_EMPTY(&all_bdrv_states
));
4303 static bool should_update_child(BdrvChild
*c
, BlockDriverState
*to
)
4309 if (c
->role
->stay_at_node
) {
4313 /* If the child @c belongs to the BDS @to, replacing the current
4314 * c->bs by @to would mean to create a loop.
4316 * Such a case occurs when appending a BDS to a backing chain.
4317 * For instance, imagine the following chain:
4319 * guest device -> node A -> further backing chain...
4321 * Now we create a new BDS B which we want to put on top of this
4322 * chain, so we first attach A as its backing node:
4327 * guest device -> node A -> further backing chain...
4329 * Finally we want to replace A by B. When doing that, we want to
4330 * replace all pointers to A by pointers to B -- except for the
4331 * pointer from B because (1) that would create a loop, and (2)
4332 * that pointer should simply stay intact:
4334 * guest device -> node B
4337 * node A -> further backing chain...
4339 * In general, when replacing a node A (c->bs) by a node B (@to),
4340 * if A is a child of B, that means we cannot replace A by B there
4341 * because that would create a loop. Silently detaching A from B
4342 * is also not really an option. So overall just leaving A in
4343 * place there is the most sensible choice.
4345 * We would also create a loop in any cases where @c is only
4346 * indirectly referenced by @to. Prevent this by returning false
4347 * if @c is found (by breadth-first search) anywhere in the whole
4352 found
= g_hash_table_new(NULL
, NULL
);
4353 g_hash_table_add(found
, to
);
4354 queue
= g_queue_new();
4355 g_queue_push_tail(queue
, to
);
4357 while (!g_queue_is_empty(queue
)) {
4358 BlockDriverState
*v
= g_queue_pop_head(queue
);
4361 QLIST_FOREACH(c2
, &v
->children
, next
) {
4367 if (g_hash_table_contains(found
, c2
->bs
)) {
4371 g_queue_push_tail(queue
, c2
->bs
);
4372 g_hash_table_add(found
, c2
->bs
);
4376 g_queue_free(queue
);
4377 g_hash_table_destroy(found
);
4382 void bdrv_replace_node(BlockDriverState
*from
, BlockDriverState
*to
,
4385 BdrvChild
*c
, *next
;
4386 GSList
*list
= NULL
, *p
;
4387 uint64_t perm
= 0, shared
= BLK_PERM_ALL
;
4390 /* Make sure that @from doesn't go away until we have successfully attached
4391 * all of its parents to @to. */
4394 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
4395 assert(bdrv_get_aio_context(from
) == bdrv_get_aio_context(to
));
4396 bdrv_drained_begin(from
);
4398 /* Put all parents into @list and calculate their cumulative permissions */
4399 QLIST_FOREACH_SAFE(c
, &from
->parents
, next_parent
, next
) {
4400 assert(c
->bs
== from
);
4401 if (!should_update_child(c
, to
)) {
4405 error_setg(errp
, "Cannot change '%s' link to '%s'",
4406 c
->name
, from
->node_name
);
4409 list
= g_slist_prepend(list
, c
);
4411 shared
&= c
->shared_perm
;
4414 /* Check whether the required permissions can be granted on @to, ignoring
4415 * all BdrvChild in @list so that they can't block themselves. */
4416 ret
= bdrv_check_update_perm(to
, NULL
, perm
, shared
, list
, NULL
, errp
);
4418 bdrv_abort_perm_update(to
);
4422 /* Now actually perform the change. We performed the permission check for
4423 * all elements of @list at once, so set the permissions all at once at the
4425 for (p
= list
; p
!= NULL
; p
= p
->next
) {
4429 bdrv_replace_child_noperm(c
, to
);
4433 bdrv_get_cumulative_perm(to
, &perm
, &shared
);
4434 bdrv_set_perm(to
, perm
, shared
);
4438 bdrv_drained_end(from
);
4443 * Add new bs contents at the top of an image chain while the chain is
4444 * live, while keeping required fields on the top layer.
4446 * This will modify the BlockDriverState fields, and swap contents
4447 * between bs_new and bs_top. Both bs_new and bs_top are modified.
4449 * bs_new must not be attached to a BlockBackend.
4451 * This function does not create any image files.
4453 * bdrv_append() takes ownership of a bs_new reference and unrefs it because
4454 * that's what the callers commonly need. bs_new will be referenced by the old
4455 * parents of bs_top after bdrv_append() returns. If the caller needs to keep a
4456 * reference of its own, it must call bdrv_ref().
4458 void bdrv_append(BlockDriverState
*bs_new
, BlockDriverState
*bs_top
,
4461 Error
*local_err
= NULL
;
4463 bdrv_set_backing_hd(bs_new
, bs_top
, &local_err
);
4465 error_propagate(errp
, local_err
);
4469 bdrv_replace_node(bs_top
, bs_new
, &local_err
);
4471 error_propagate(errp
, local_err
);
4472 bdrv_set_backing_hd(bs_new
, NULL
, &error_abort
);
4476 /* bs_new is now referenced by its new parents, we don't need the
4477 * additional reference any more. */
4482 static void bdrv_delete(BlockDriverState
*bs
)
4484 assert(bdrv_op_blocker_is_empty(bs
));
4485 assert(!bs
->refcnt
);
4487 /* remove from list, if necessary */
4488 if (bs
->node_name
[0] != '\0') {
4489 QTAILQ_REMOVE(&graph_bdrv_states
, bs
, node_list
);
4491 QTAILQ_REMOVE(&all_bdrv_states
, bs
, bs_list
);
4499 * Run consistency checks on an image
4501 * Returns 0 if the check could be completed (it doesn't mean that the image is
4502 * free of errors) or -errno when an internal error occurred. The results of the
4503 * check are stored in res.
4505 static int coroutine_fn
bdrv_co_check(BlockDriverState
*bs
,
4506 BdrvCheckResult
*res
, BdrvCheckMode fix
)
4508 if (bs
->drv
== NULL
) {
4511 if (bs
->drv
->bdrv_co_check
== NULL
) {
4515 memset(res
, 0, sizeof(*res
));
4516 return bs
->drv
->bdrv_co_check(bs
, res
, fix
);
4519 typedef struct CheckCo
{
4520 BlockDriverState
*bs
;
4521 BdrvCheckResult
*res
;
4526 static void coroutine_fn
bdrv_check_co_entry(void *opaque
)
4528 CheckCo
*cco
= opaque
;
4529 cco
->ret
= bdrv_co_check(cco
->bs
, cco
->res
, cco
->fix
);
4533 int bdrv_check(BlockDriverState
*bs
,
4534 BdrvCheckResult
*res
, BdrvCheckMode fix
)
4540 .ret
= -EINPROGRESS
,
4544 if (qemu_in_coroutine()) {
4545 /* Fast-path if already in coroutine context */
4546 bdrv_check_co_entry(&cco
);
4548 co
= qemu_coroutine_create(bdrv_check_co_entry
, &cco
);
4549 bdrv_coroutine_enter(bs
, co
);
4550 BDRV_POLL_WHILE(bs
, cco
.ret
== -EINPROGRESS
);
4559 * -EINVAL - backing format specified, but no file
4560 * -ENOSPC - can't update the backing file because no space is left in the
4562 * -ENOTSUP - format driver doesn't support changing the backing file
4564 int bdrv_change_backing_file(BlockDriverState
*bs
,
4565 const char *backing_file
, const char *backing_fmt
)
4567 BlockDriver
*drv
= bs
->drv
;
4574 /* Backing file format doesn't make sense without a backing file */
4575 if (backing_fmt
&& !backing_file
) {
4579 if (drv
->bdrv_change_backing_file
!= NULL
) {
4580 ret
= drv
->bdrv_change_backing_file(bs
, backing_file
, backing_fmt
);
4586 pstrcpy(bs
->backing_file
, sizeof(bs
->backing_file
), backing_file
?: "");
4587 pstrcpy(bs
->backing_format
, sizeof(bs
->backing_format
), backing_fmt
?: "");
4588 pstrcpy(bs
->auto_backing_file
, sizeof(bs
->auto_backing_file
),
4589 backing_file
?: "");
4595 * Finds the image layer in the chain that has 'bs' as its backing file.
4597 * active is the current topmost image.
4599 * Returns NULL if bs is not found in active's image chain,
4600 * or if active == bs.
4602 * Returns the bottommost base image if bs == NULL.
4604 BlockDriverState
*bdrv_find_overlay(BlockDriverState
*active
,
4605 BlockDriverState
*bs
)
4607 while (active
&& bs
!= backing_bs(active
)) {
4608 active
= backing_bs(active
);
4614 /* Given a BDS, searches for the base layer. */
4615 BlockDriverState
*bdrv_find_base(BlockDriverState
*bs
)
4617 return bdrv_find_overlay(bs
, NULL
);
4621 * Return true if at least one of the backing links between @bs and
4622 * @base is frozen. @errp is set if that's the case.
4623 * @base must be reachable from @bs, or NULL.
4625 bool bdrv_is_backing_chain_frozen(BlockDriverState
*bs
, BlockDriverState
*base
,
4628 BlockDriverState
*i
;
4630 for (i
= bs
; i
!= base
; i
= backing_bs(i
)) {
4631 if (i
->backing
&& i
->backing
->frozen
) {
4632 error_setg(errp
, "Cannot change '%s' link from '%s' to '%s'",
4633 i
->backing
->name
, i
->node_name
,
4634 backing_bs(i
)->node_name
);
4643 * Freeze all backing links between @bs and @base.
4644 * If any of the links is already frozen the operation is aborted and
4645 * none of the links are modified.
4646 * @base must be reachable from @bs, or NULL.
4647 * Returns 0 on success. On failure returns < 0 and sets @errp.
4649 int bdrv_freeze_backing_chain(BlockDriverState
*bs
, BlockDriverState
*base
,
4652 BlockDriverState
*i
;
4654 if (bdrv_is_backing_chain_frozen(bs
, base
, errp
)) {
4658 for (i
= bs
; i
!= base
; i
= backing_bs(i
)) {
4659 if (i
->backing
&& backing_bs(i
)->never_freeze
) {
4660 error_setg(errp
, "Cannot freeze '%s' link to '%s'",
4661 i
->backing
->name
, backing_bs(i
)->node_name
);
4666 for (i
= bs
; i
!= base
; i
= backing_bs(i
)) {
4668 i
->backing
->frozen
= true;
4676 * Unfreeze all backing links between @bs and @base. The caller must
4677 * ensure that all links are frozen before using this function.
4678 * @base must be reachable from @bs, or NULL.
4680 void bdrv_unfreeze_backing_chain(BlockDriverState
*bs
, BlockDriverState
*base
)
4682 BlockDriverState
*i
;
4684 for (i
= bs
; i
!= base
; i
= backing_bs(i
)) {
4686 assert(i
->backing
->frozen
);
4687 i
->backing
->frozen
= false;
4693 * Drops images above 'base' up to and including 'top', and sets the image
4694 * above 'top' to have base as its backing file.
4696 * Requires that the overlay to 'top' is opened r/w, so that the backing file
4697 * information in 'bs' can be properly updated.
4699 * E.g., this will convert the following chain:
4700 * bottom <- base <- intermediate <- top <- active
4704 * bottom <- base <- active
4706 * It is allowed for bottom==base, in which case it converts:
4708 * base <- intermediate <- top <- active
4714 * If backing_file_str is non-NULL, it will be used when modifying top's
4715 * overlay image metadata.
4718 * if active == top, that is considered an error
4721 int bdrv_drop_intermediate(BlockDriverState
*top
, BlockDriverState
*base
,
4722 const char *backing_file_str
)
4724 BlockDriverState
*explicit_top
= top
;
4725 bool update_inherits_from
;
4726 BdrvChild
*c
, *next
;
4727 Error
*local_err
= NULL
;
4731 bdrv_subtree_drained_begin(top
);
4733 if (!top
->drv
|| !base
->drv
) {
4737 /* Make sure that base is in the backing chain of top */
4738 if (!bdrv_chain_contains(top
, base
)) {
4742 /* This function changes all links that point to top and makes
4743 * them point to base. Check that none of them is frozen. */
4744 QLIST_FOREACH(c
, &top
->parents
, next_parent
) {
4750 /* If 'base' recursively inherits from 'top' then we should set
4751 * base->inherits_from to top->inherits_from after 'top' and all
4752 * other intermediate nodes have been dropped.
4753 * If 'top' is an implicit node (e.g. "commit_top") we should skip
4754 * it because no one inherits from it. We use explicit_top for that. */
4755 while (explicit_top
&& explicit_top
->implicit
) {
4756 explicit_top
= backing_bs(explicit_top
);
4758 update_inherits_from
= bdrv_inherits_from_recursive(base
, explicit_top
);
4760 /* success - we can delete the intermediate states, and link top->base */
4761 /* TODO Check graph modification op blockers (BLK_PERM_GRAPH_MOD) once
4762 * we've figured out how they should work. */
4763 if (!backing_file_str
) {
4764 bdrv_refresh_filename(base
);
4765 backing_file_str
= base
->filename
;
4768 QLIST_FOREACH_SAFE(c
, &top
->parents
, next_parent
, next
) {
4769 /* Check whether we are allowed to switch c from top to base */
4770 GSList
*ignore_children
= g_slist_prepend(NULL
, c
);
4771 ret
= bdrv_check_update_perm(base
, NULL
, c
->perm
, c
->shared_perm
,
4772 ignore_children
, NULL
, &local_err
);
4773 g_slist_free(ignore_children
);
4775 error_report_err(local_err
);
4779 /* If so, update the backing file path in the image file */
4780 if (c
->role
->update_filename
) {
4781 ret
= c
->role
->update_filename(c
, base
, backing_file_str
,
4784 bdrv_abort_perm_update(base
);
4785 error_report_err(local_err
);
4790 /* Do the actual switch in the in-memory graph.
4791 * Completes bdrv_check_update_perm() transaction internally. */
4793 bdrv_replace_child(c
, base
);
4797 if (update_inherits_from
) {
4798 base
->inherits_from
= explicit_top
->inherits_from
;
4803 bdrv_subtree_drained_end(top
);
4809 * Length of a allocated file in bytes. Sparse files are counted by actual
4810 * allocated space. Return < 0 if error or unknown.
4812 int64_t bdrv_get_allocated_file_size(BlockDriverState
*bs
)
4814 BlockDriver
*drv
= bs
->drv
;
4818 if (drv
->bdrv_get_allocated_file_size
) {
4819 return drv
->bdrv_get_allocated_file_size(bs
);
4822 return bdrv_get_allocated_file_size(bs
->file
->bs
);
4829 * @drv: Format driver
4830 * @opts: Creation options for new image
4831 * @in_bs: Existing image containing data for new image (may be NULL)
4832 * @errp: Error object
4833 * Returns: A #BlockMeasureInfo (free using qapi_free_BlockMeasureInfo())
4836 * Calculate file size required to create a new image.
4838 * If @in_bs is given then space for allocated clusters and zero clusters
4839 * from that image are included in the calculation. If @opts contains a
4840 * backing file that is shared by @in_bs then backing clusters may be omitted
4841 * from the calculation.
4843 * If @in_bs is NULL then the calculation includes no allocated clusters
4844 * unless a preallocation option is given in @opts.
4846 * Note that @in_bs may use a different BlockDriver from @drv.
4848 * If an error occurs the @errp pointer is set.
4850 BlockMeasureInfo
*bdrv_measure(BlockDriver
*drv
, QemuOpts
*opts
,
4851 BlockDriverState
*in_bs
, Error
**errp
)
4853 if (!drv
->bdrv_measure
) {
4854 error_setg(errp
, "Block driver '%s' does not support size measurement",
4859 return drv
->bdrv_measure(opts
, in_bs
, errp
);
4863 * Return number of sectors on success, -errno on error.
4865 int64_t bdrv_nb_sectors(BlockDriverState
*bs
)
4867 BlockDriver
*drv
= bs
->drv
;
4872 if (drv
->has_variable_length
) {
4873 int ret
= refresh_total_sectors(bs
, bs
->total_sectors
);
4878 return bs
->total_sectors
;
4882 * Return length in bytes on success, -errno on error.
4883 * The length is always a multiple of BDRV_SECTOR_SIZE.
4885 int64_t bdrv_getlength(BlockDriverState
*bs
)
4887 int64_t ret
= bdrv_nb_sectors(bs
);
4889 ret
= ret
> INT64_MAX
/ BDRV_SECTOR_SIZE
? -EFBIG
: ret
;
4890 return ret
< 0 ? ret
: ret
* BDRV_SECTOR_SIZE
;
4893 /* return 0 as number of sectors if no device present or error */
4894 void bdrv_get_geometry(BlockDriverState
*bs
, uint64_t *nb_sectors_ptr
)
4896 int64_t nb_sectors
= bdrv_nb_sectors(bs
);
4898 *nb_sectors_ptr
= nb_sectors
< 0 ? 0 : nb_sectors
;
4901 bool bdrv_is_sg(BlockDriverState
*bs
)
4906 bool bdrv_is_encrypted(BlockDriverState
*bs
)
4908 if (bs
->backing
&& bs
->backing
->bs
->encrypted
) {
4911 return bs
->encrypted
;
4914 const char *bdrv_get_format_name(BlockDriverState
*bs
)
4916 return bs
->drv
? bs
->drv
->format_name
: NULL
;
4919 static int qsort_strcmp(const void *a
, const void *b
)
4921 return strcmp(*(char *const *)a
, *(char *const *)b
);
4924 void bdrv_iterate_format(void (*it
)(void *opaque
, const char *name
),
4925 void *opaque
, bool read_only
)
4930 const char **formats
= NULL
;
4932 QLIST_FOREACH(drv
, &bdrv_drivers
, list
) {
4933 if (drv
->format_name
) {
4937 if (use_bdrv_whitelist
&& !bdrv_is_whitelisted(drv
, read_only
)) {
4941 while (formats
&& i
&& !found
) {
4942 found
= !strcmp(formats
[--i
], drv
->format_name
);
4946 formats
= g_renew(const char *, formats
, count
+ 1);
4947 formats
[count
++] = drv
->format_name
;
4952 for (i
= 0; i
< (int)ARRAY_SIZE(block_driver_modules
); i
++) {
4953 const char *format_name
= block_driver_modules
[i
].format_name
;
4959 if (use_bdrv_whitelist
&&
4960 !bdrv_format_is_whitelisted(format_name
, read_only
)) {
4964 while (formats
&& j
&& !found
) {
4965 found
= !strcmp(formats
[--j
], format_name
);
4969 formats
= g_renew(const char *, formats
, count
+ 1);
4970 formats
[count
++] = format_name
;
4975 qsort(formats
, count
, sizeof(formats
[0]), qsort_strcmp
);
4977 for (i
= 0; i
< count
; i
++) {
4978 it(opaque
, formats
[i
]);
4984 /* This function is to find a node in the bs graph */
4985 BlockDriverState
*bdrv_find_node(const char *node_name
)
4987 BlockDriverState
*bs
;
4991 QTAILQ_FOREACH(bs
, &graph_bdrv_states
, node_list
) {
4992 if (!strcmp(node_name
, bs
->node_name
)) {
4999 /* Put this QMP function here so it can access the static graph_bdrv_states. */
5000 BlockDeviceInfoList
*bdrv_named_nodes_list(bool flat
,
5003 BlockDeviceInfoList
*list
, *entry
;
5004 BlockDriverState
*bs
;
5007 QTAILQ_FOREACH(bs
, &graph_bdrv_states
, node_list
) {
5008 BlockDeviceInfo
*info
= bdrv_block_device_info(NULL
, bs
, flat
, errp
);
5010 qapi_free_BlockDeviceInfoList(list
);
5013 entry
= g_malloc0(sizeof(*entry
));
5014 entry
->value
= info
;
5022 #define QAPI_LIST_ADD(list, element) do { \
5023 typeof(list) _tmp = g_new(typeof(*(list)), 1); \
5024 _tmp->value = (element); \
5025 _tmp->next = (list); \
5029 typedef struct XDbgBlockGraphConstructor
{
5030 XDbgBlockGraph
*graph
;
5031 GHashTable
*graph_nodes
;
5032 } XDbgBlockGraphConstructor
;
5034 static XDbgBlockGraphConstructor
*xdbg_graph_new(void)
5036 XDbgBlockGraphConstructor
*gr
= g_new(XDbgBlockGraphConstructor
, 1);
5038 gr
->graph
= g_new0(XDbgBlockGraph
, 1);
5039 gr
->graph_nodes
= g_hash_table_new(NULL
, NULL
);
5044 static XDbgBlockGraph
*xdbg_graph_finalize(XDbgBlockGraphConstructor
*gr
)
5046 XDbgBlockGraph
*graph
= gr
->graph
;
5048 g_hash_table_destroy(gr
->graph_nodes
);
5054 static uintptr_t xdbg_graph_node_num(XDbgBlockGraphConstructor
*gr
, void *node
)
5056 uintptr_t ret
= (uintptr_t)g_hash_table_lookup(gr
->graph_nodes
, node
);
5063 * Start counting from 1, not 0, because 0 interferes with not-found (NULL)
5064 * answer of g_hash_table_lookup.
5066 ret
= g_hash_table_size(gr
->graph_nodes
) + 1;
5067 g_hash_table_insert(gr
->graph_nodes
, node
, (void *)ret
);
5072 static void xdbg_graph_add_node(XDbgBlockGraphConstructor
*gr
, void *node
,
5073 XDbgBlockGraphNodeType type
, const char *name
)
5075 XDbgBlockGraphNode
*n
;
5077 n
= g_new0(XDbgBlockGraphNode
, 1);
5079 n
->id
= xdbg_graph_node_num(gr
, node
);
5081 n
->name
= g_strdup(name
);
5083 QAPI_LIST_ADD(gr
->graph
->nodes
, n
);
5086 static void xdbg_graph_add_edge(XDbgBlockGraphConstructor
*gr
, void *parent
,
5087 const BdrvChild
*child
)
5089 BlockPermission qapi_perm
;
5090 XDbgBlockGraphEdge
*edge
;
5092 edge
= g_new0(XDbgBlockGraphEdge
, 1);
5094 edge
->parent
= xdbg_graph_node_num(gr
, parent
);
5095 edge
->child
= xdbg_graph_node_num(gr
, child
->bs
);
5096 edge
->name
= g_strdup(child
->name
);
5098 for (qapi_perm
= 0; qapi_perm
< BLOCK_PERMISSION__MAX
; qapi_perm
++) {
5099 uint64_t flag
= bdrv_qapi_perm_to_blk_perm(qapi_perm
);
5101 if (flag
& child
->perm
) {
5102 QAPI_LIST_ADD(edge
->perm
, qapi_perm
);
5104 if (flag
& child
->shared_perm
) {
5105 QAPI_LIST_ADD(edge
->shared_perm
, qapi_perm
);
5109 QAPI_LIST_ADD(gr
->graph
->edges
, edge
);
5113 XDbgBlockGraph
*bdrv_get_xdbg_block_graph(Error
**errp
)
5117 BlockDriverState
*bs
;
5119 XDbgBlockGraphConstructor
*gr
= xdbg_graph_new();
5121 for (blk
= blk_all_next(NULL
); blk
; blk
= blk_all_next(blk
)) {
5122 char *allocated_name
= NULL
;
5123 const char *name
= blk_name(blk
);
5126 name
= allocated_name
= blk_get_attached_dev_id(blk
);
5128 xdbg_graph_add_node(gr
, blk
, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_BACKEND
,
5130 g_free(allocated_name
);
5131 if (blk_root(blk
)) {
5132 xdbg_graph_add_edge(gr
, blk
, blk_root(blk
));
5136 for (job
= block_job_next(NULL
); job
; job
= block_job_next(job
)) {
5139 xdbg_graph_add_node(gr
, job
, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_JOB
,
5141 for (el
= job
->nodes
; el
; el
= el
->next
) {
5142 xdbg_graph_add_edge(gr
, job
, (BdrvChild
*)el
->data
);
5146 QTAILQ_FOREACH(bs
, &graph_bdrv_states
, node_list
) {
5147 xdbg_graph_add_node(gr
, bs
, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_DRIVER
,
5149 QLIST_FOREACH(child
, &bs
->children
, next
) {
5150 xdbg_graph_add_edge(gr
, bs
, child
);
5154 return xdbg_graph_finalize(gr
);
5157 BlockDriverState
*bdrv_lookup_bs(const char *device
,
5158 const char *node_name
,
5162 BlockDriverState
*bs
;
5165 blk
= blk_by_name(device
);
5170 error_setg(errp
, "Device '%s' has no medium", device
);
5178 bs
= bdrv_find_node(node_name
);
5185 error_setg(errp
, "Cannot find device=%s nor node_name=%s",
5186 device
? device
: "",
5187 node_name
? node_name
: "");
5191 /* If 'base' is in the same chain as 'top', return true. Otherwise,
5192 * return false. If either argument is NULL, return false. */
5193 bool bdrv_chain_contains(BlockDriverState
*top
, BlockDriverState
*base
)
5195 while (top
&& top
!= base
) {
5196 top
= backing_bs(top
);
5202 BlockDriverState
*bdrv_next_node(BlockDriverState
*bs
)
5205 return QTAILQ_FIRST(&graph_bdrv_states
);
5207 return QTAILQ_NEXT(bs
, node_list
);
5210 BlockDriverState
*bdrv_next_all_states(BlockDriverState
*bs
)
5213 return QTAILQ_FIRST(&all_bdrv_states
);
5215 return QTAILQ_NEXT(bs
, bs_list
);
5218 const char *bdrv_get_node_name(const BlockDriverState
*bs
)
5220 return bs
->node_name
;
5223 const char *bdrv_get_parent_name(const BlockDriverState
*bs
)
5228 /* If multiple parents have a name, just pick the first one. */
5229 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
5230 if (c
->role
->get_name
) {
5231 name
= c
->role
->get_name(c
);
5232 if (name
&& *name
) {
5241 /* TODO check what callers really want: bs->node_name or blk_name() */
5242 const char *bdrv_get_device_name(const BlockDriverState
*bs
)
5244 return bdrv_get_parent_name(bs
) ?: "";
5247 /* This can be used to identify nodes that might not have a device
5248 * name associated. Since node and device names live in the same
5249 * namespace, the result is unambiguous. The exception is if both are
5250 * absent, then this returns an empty (non-null) string. */
5251 const char *bdrv_get_device_or_node_name(const BlockDriverState
*bs
)
5253 return bdrv_get_parent_name(bs
) ?: bs
->node_name
;
5256 int bdrv_get_flags(BlockDriverState
*bs
)
5258 return bs
->open_flags
;
5261 int bdrv_has_zero_init_1(BlockDriverState
*bs
)
5266 int bdrv_has_zero_init(BlockDriverState
*bs
)
5272 /* If BS is a copy on write image, it is initialized to
5273 the contents of the base image, which may not be zeroes. */
5277 if (bs
->drv
->bdrv_has_zero_init
) {
5278 return bs
->drv
->bdrv_has_zero_init(bs
);
5280 if (bs
->file
&& bs
->drv
->is_filter
) {
5281 return bdrv_has_zero_init(bs
->file
->bs
);
5288 int bdrv_has_zero_init_truncate(BlockDriverState
*bs
)
5295 /* Depends on the backing image length, but better safe than sorry */
5298 if (bs
->drv
->bdrv_has_zero_init_truncate
) {
5299 return bs
->drv
->bdrv_has_zero_init_truncate(bs
);
5301 if (bs
->file
&& bs
->drv
->is_filter
) {
5302 return bdrv_has_zero_init_truncate(bs
->file
->bs
);
5309 bool bdrv_unallocated_blocks_are_zero(BlockDriverState
*bs
)
5311 BlockDriverInfo bdi
;
5317 if (bdrv_get_info(bs
, &bdi
) == 0) {
5318 return bdi
.unallocated_blocks_are_zero
;
5324 bool bdrv_can_write_zeroes_with_unmap(BlockDriverState
*bs
)
5326 if (!(bs
->open_flags
& BDRV_O_UNMAP
)) {
5330 return bs
->supported_zero_flags
& BDRV_REQ_MAY_UNMAP
;
5333 void bdrv_get_backing_filename(BlockDriverState
*bs
,
5334 char *filename
, int filename_size
)
5336 pstrcpy(filename
, filename_size
, bs
->backing_file
);
5339 int bdrv_get_info(BlockDriverState
*bs
, BlockDriverInfo
*bdi
)
5341 BlockDriver
*drv
= bs
->drv
;
5342 /* if bs->drv == NULL, bs is closed, so there's nothing to do here */
5346 if (!drv
->bdrv_get_info
) {
5347 if (bs
->file
&& drv
->is_filter
) {
5348 return bdrv_get_info(bs
->file
->bs
, bdi
);
5352 memset(bdi
, 0, sizeof(*bdi
));
5353 return drv
->bdrv_get_info(bs
, bdi
);
5356 ImageInfoSpecific
*bdrv_get_specific_info(BlockDriverState
*bs
,
5359 BlockDriver
*drv
= bs
->drv
;
5360 if (drv
&& drv
->bdrv_get_specific_info
) {
5361 return drv
->bdrv_get_specific_info(bs
, errp
);
5366 BlockStatsSpecific
*bdrv_get_specific_stats(BlockDriverState
*bs
)
5368 BlockDriver
*drv
= bs
->drv
;
5369 if (!drv
|| !drv
->bdrv_get_specific_stats
) {
5372 return drv
->bdrv_get_specific_stats(bs
);
5375 void bdrv_debug_event(BlockDriverState
*bs
, BlkdebugEvent event
)
5377 if (!bs
|| !bs
->drv
|| !bs
->drv
->bdrv_debug_event
) {
5381 bs
->drv
->bdrv_debug_event(bs
, event
);
5384 static BlockDriverState
*bdrv_find_debug_node(BlockDriverState
*bs
)
5386 while (bs
&& bs
->drv
&& !bs
->drv
->bdrv_debug_breakpoint
) {
5392 if (bs
->drv
->is_filter
&& bs
->backing
) {
5393 bs
= bs
->backing
->bs
;
5400 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_breakpoint
) {
5401 assert(bs
->drv
->bdrv_debug_remove_breakpoint
);
5408 int bdrv_debug_breakpoint(BlockDriverState
*bs
, const char *event
,
5411 bs
= bdrv_find_debug_node(bs
);
5413 return bs
->drv
->bdrv_debug_breakpoint(bs
, event
, tag
);
5419 int bdrv_debug_remove_breakpoint(BlockDriverState
*bs
, const char *tag
)
5421 bs
= bdrv_find_debug_node(bs
);
5423 return bs
->drv
->bdrv_debug_remove_breakpoint(bs
, tag
);
5429 int bdrv_debug_resume(BlockDriverState
*bs
, const char *tag
)
5431 while (bs
&& (!bs
->drv
|| !bs
->drv
->bdrv_debug_resume
)) {
5432 bs
= bs
->file
? bs
->file
->bs
: NULL
;
5435 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_resume
) {
5436 return bs
->drv
->bdrv_debug_resume(bs
, tag
);
5442 bool bdrv_debug_is_suspended(BlockDriverState
*bs
, const char *tag
)
5444 while (bs
&& bs
->drv
&& !bs
->drv
->bdrv_debug_is_suspended
) {
5445 bs
= bs
->file
? bs
->file
->bs
: NULL
;
5448 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_is_suspended
) {
5449 return bs
->drv
->bdrv_debug_is_suspended(bs
, tag
);
5455 /* backing_file can either be relative, or absolute, or a protocol. If it is
5456 * relative, it must be relative to the chain. So, passing in bs->filename
5457 * from a BDS as backing_file should not be done, as that may be relative to
5458 * the CWD rather than the chain. */
5459 BlockDriverState
*bdrv_find_backing_image(BlockDriverState
*bs
,
5460 const char *backing_file
)
5462 char *filename_full
= NULL
;
5463 char *backing_file_full
= NULL
;
5464 char *filename_tmp
= NULL
;
5465 int is_protocol
= 0;
5466 BlockDriverState
*curr_bs
= NULL
;
5467 BlockDriverState
*retval
= NULL
;
5469 if (!bs
|| !bs
->drv
|| !backing_file
) {
5473 filename_full
= g_malloc(PATH_MAX
);
5474 backing_file_full
= g_malloc(PATH_MAX
);
5476 is_protocol
= path_has_protocol(backing_file
);
5478 for (curr_bs
= bs
; curr_bs
->backing
; curr_bs
= curr_bs
->backing
->bs
) {
5480 /* If either of the filename paths is actually a protocol, then
5481 * compare unmodified paths; otherwise make paths relative */
5482 if (is_protocol
|| path_has_protocol(curr_bs
->backing_file
)) {
5483 char *backing_file_full_ret
;
5485 if (strcmp(backing_file
, curr_bs
->backing_file
) == 0) {
5486 retval
= curr_bs
->backing
->bs
;
5489 /* Also check against the full backing filename for the image */
5490 backing_file_full_ret
= bdrv_get_full_backing_filename(curr_bs
,
5492 if (backing_file_full_ret
) {
5493 bool equal
= strcmp(backing_file
, backing_file_full_ret
) == 0;
5494 g_free(backing_file_full_ret
);
5496 retval
= curr_bs
->backing
->bs
;
5501 /* If not an absolute filename path, make it relative to the current
5502 * image's filename path */
5503 filename_tmp
= bdrv_make_absolute_filename(curr_bs
, backing_file
,
5505 /* We are going to compare canonicalized absolute pathnames */
5506 if (!filename_tmp
|| !realpath(filename_tmp
, filename_full
)) {
5507 g_free(filename_tmp
);
5510 g_free(filename_tmp
);
5512 /* We need to make sure the backing filename we are comparing against
5513 * is relative to the current image filename (or absolute) */
5514 filename_tmp
= bdrv_get_full_backing_filename(curr_bs
, NULL
);
5515 if (!filename_tmp
|| !realpath(filename_tmp
, backing_file_full
)) {
5516 g_free(filename_tmp
);
5519 g_free(filename_tmp
);
5521 if (strcmp(backing_file_full
, filename_full
) == 0) {
5522 retval
= curr_bs
->backing
->bs
;
5528 g_free(filename_full
);
5529 g_free(backing_file_full
);
5533 void bdrv_init(void)
5535 module_call_init(MODULE_INIT_BLOCK
);
5538 void bdrv_init_with_whitelist(void)
5540 use_bdrv_whitelist
= 1;
5544 static void coroutine_fn
bdrv_co_invalidate_cache(BlockDriverState
*bs
,
5547 BdrvChild
*child
, *parent
;
5548 uint64_t perm
, shared_perm
;
5549 Error
*local_err
= NULL
;
5551 BdrvDirtyBitmap
*bm
;
5557 QLIST_FOREACH(child
, &bs
->children
, next
) {
5558 bdrv_co_invalidate_cache(child
->bs
, &local_err
);
5560 error_propagate(errp
, local_err
);
5566 * Update permissions, they may differ for inactive nodes.
5568 * Note that the required permissions of inactive images are always a
5569 * subset of the permissions required after activating the image. This
5570 * allows us to just get the permissions upfront without restricting
5571 * drv->bdrv_invalidate_cache().
5573 * It also means that in error cases, we don't have to try and revert to
5574 * the old permissions (which is an operation that could fail, too). We can
5575 * just keep the extended permissions for the next time that an activation
5576 * of the image is tried.
5578 if (bs
->open_flags
& BDRV_O_INACTIVE
) {
5579 bs
->open_flags
&= ~BDRV_O_INACTIVE
;
5580 bdrv_get_cumulative_perm(bs
, &perm
, &shared_perm
);
5581 ret
= bdrv_check_perm(bs
, NULL
, perm
, shared_perm
, NULL
, NULL
, &local_err
);
5583 bs
->open_flags
|= BDRV_O_INACTIVE
;
5584 error_propagate(errp
, local_err
);
5587 bdrv_set_perm(bs
, perm
, shared_perm
);
5589 if (bs
->drv
->bdrv_co_invalidate_cache
) {
5590 bs
->drv
->bdrv_co_invalidate_cache(bs
, &local_err
);
5592 bs
->open_flags
|= BDRV_O_INACTIVE
;
5593 error_propagate(errp
, local_err
);
5598 FOR_EACH_DIRTY_BITMAP(bs
, bm
) {
5599 bdrv_dirty_bitmap_skip_store(bm
, false);
5602 ret
= refresh_total_sectors(bs
, bs
->total_sectors
);
5604 bs
->open_flags
|= BDRV_O_INACTIVE
;
5605 error_setg_errno(errp
, -ret
, "Could not refresh total sector count");
5610 QLIST_FOREACH(parent
, &bs
->parents
, next_parent
) {
5611 if (parent
->role
->activate
) {
5612 parent
->role
->activate(parent
, &local_err
);
5614 bs
->open_flags
|= BDRV_O_INACTIVE
;
5615 error_propagate(errp
, local_err
);
5622 typedef struct InvalidateCacheCo
{
5623 BlockDriverState
*bs
;
5626 } InvalidateCacheCo
;
5628 static void coroutine_fn
bdrv_invalidate_cache_co_entry(void *opaque
)
5630 InvalidateCacheCo
*ico
= opaque
;
5631 bdrv_co_invalidate_cache(ico
->bs
, ico
->errp
);
5636 void bdrv_invalidate_cache(BlockDriverState
*bs
, Error
**errp
)
5639 InvalidateCacheCo ico
= {
5645 if (qemu_in_coroutine()) {
5646 /* Fast-path if already in coroutine context */
5647 bdrv_invalidate_cache_co_entry(&ico
);
5649 co
= qemu_coroutine_create(bdrv_invalidate_cache_co_entry
, &ico
);
5650 bdrv_coroutine_enter(bs
, co
);
5651 BDRV_POLL_WHILE(bs
, !ico
.done
);
5655 void bdrv_invalidate_cache_all(Error
**errp
)
5657 BlockDriverState
*bs
;
5658 Error
*local_err
= NULL
;
5659 BdrvNextIterator it
;
5661 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
5662 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
5664 aio_context_acquire(aio_context
);
5665 bdrv_invalidate_cache(bs
, &local_err
);
5666 aio_context_release(aio_context
);
5668 error_propagate(errp
, local_err
);
5669 bdrv_next_cleanup(&it
);
5675 static bool bdrv_has_bds_parent(BlockDriverState
*bs
, bool only_active
)
5679 QLIST_FOREACH(parent
, &bs
->parents
, next_parent
) {
5680 if (parent
->role
->parent_is_bds
) {
5681 BlockDriverState
*parent_bs
= parent
->opaque
;
5682 if (!only_active
|| !(parent_bs
->open_flags
& BDRV_O_INACTIVE
)) {
5691 static int bdrv_inactivate_recurse(BlockDriverState
*bs
)
5693 BdrvChild
*child
, *parent
;
5694 bool tighten_restrictions
;
5695 uint64_t perm
, shared_perm
;
5702 /* Make sure that we don't inactivate a child before its parent.
5703 * It will be covered by recursion from the yet active parent. */
5704 if (bdrv_has_bds_parent(bs
, true)) {
5708 assert(!(bs
->open_flags
& BDRV_O_INACTIVE
));
5710 /* Inactivate this node */
5711 if (bs
->drv
->bdrv_inactivate
) {
5712 ret
= bs
->drv
->bdrv_inactivate(bs
);
5718 QLIST_FOREACH(parent
, &bs
->parents
, next_parent
) {
5719 if (parent
->role
->inactivate
) {
5720 ret
= parent
->role
->inactivate(parent
);
5727 bs
->open_flags
|= BDRV_O_INACTIVE
;
5729 /* Update permissions, they may differ for inactive nodes */
5730 bdrv_get_cumulative_perm(bs
, &perm
, &shared_perm
);
5731 ret
= bdrv_check_perm(bs
, NULL
, perm
, shared_perm
, NULL
,
5732 &tighten_restrictions
, NULL
);
5733 assert(tighten_restrictions
== false);
5735 /* We only tried to loosen restrictions, so errors are not fatal */
5736 bdrv_abort_perm_update(bs
);
5738 bdrv_set_perm(bs
, perm
, shared_perm
);
5742 /* Recursively inactivate children */
5743 QLIST_FOREACH(child
, &bs
->children
, next
) {
5744 ret
= bdrv_inactivate_recurse(child
->bs
);
5753 int bdrv_inactivate_all(void)
5755 BlockDriverState
*bs
= NULL
;
5756 BdrvNextIterator it
;
5758 GSList
*aio_ctxs
= NULL
, *ctx
;
5760 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
5761 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
5763 if (!g_slist_find(aio_ctxs
, aio_context
)) {
5764 aio_ctxs
= g_slist_prepend(aio_ctxs
, aio_context
);
5765 aio_context_acquire(aio_context
);
5769 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
5770 /* Nodes with BDS parents are covered by recursion from the last
5771 * parent that gets inactivated. Don't inactivate them a second
5772 * time if that has already happened. */
5773 if (bdrv_has_bds_parent(bs
, false)) {
5776 ret
= bdrv_inactivate_recurse(bs
);
5778 bdrv_next_cleanup(&it
);
5784 for (ctx
= aio_ctxs
; ctx
!= NULL
; ctx
= ctx
->next
) {
5785 AioContext
*aio_context
= ctx
->data
;
5786 aio_context_release(aio_context
);
5788 g_slist_free(aio_ctxs
);
5793 /**************************************************************/
5794 /* removable device support */
5797 * Return TRUE if the media is present
5799 bool bdrv_is_inserted(BlockDriverState
*bs
)
5801 BlockDriver
*drv
= bs
->drv
;
5807 if (drv
->bdrv_is_inserted
) {
5808 return drv
->bdrv_is_inserted(bs
);
5810 QLIST_FOREACH(child
, &bs
->children
, next
) {
5811 if (!bdrv_is_inserted(child
->bs
)) {
5819 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
5821 void bdrv_eject(BlockDriverState
*bs
, bool eject_flag
)
5823 BlockDriver
*drv
= bs
->drv
;
5825 if (drv
&& drv
->bdrv_eject
) {
5826 drv
->bdrv_eject(bs
, eject_flag
);
5831 * Lock or unlock the media (if it is locked, the user won't be able
5832 * to eject it manually).
5834 void bdrv_lock_medium(BlockDriverState
*bs
, bool locked
)
5836 BlockDriver
*drv
= bs
->drv
;
5838 trace_bdrv_lock_medium(bs
, locked
);
5840 if (drv
&& drv
->bdrv_lock_medium
) {
5841 drv
->bdrv_lock_medium(bs
, locked
);
5845 /* Get a reference to bs */
5846 void bdrv_ref(BlockDriverState
*bs
)
5851 /* Release a previously grabbed reference to bs.
5852 * If after releasing, reference count is zero, the BlockDriverState is
5854 void bdrv_unref(BlockDriverState
*bs
)
5859 assert(bs
->refcnt
> 0);
5860 if (--bs
->refcnt
== 0) {
5865 struct BdrvOpBlocker
{
5867 QLIST_ENTRY(BdrvOpBlocker
) list
;
5870 bool bdrv_op_is_blocked(BlockDriverState
*bs
, BlockOpType op
, Error
**errp
)
5872 BdrvOpBlocker
*blocker
;
5873 assert((int) op
>= 0 && op
< BLOCK_OP_TYPE_MAX
);
5874 if (!QLIST_EMPTY(&bs
->op_blockers
[op
])) {
5875 blocker
= QLIST_FIRST(&bs
->op_blockers
[op
]);
5876 error_propagate_prepend(errp
, error_copy(blocker
->reason
),
5877 "Node '%s' is busy: ",
5878 bdrv_get_device_or_node_name(bs
));
5884 void bdrv_op_block(BlockDriverState
*bs
, BlockOpType op
, Error
*reason
)
5886 BdrvOpBlocker
*blocker
;
5887 assert((int) op
>= 0 && op
< BLOCK_OP_TYPE_MAX
);
5889 blocker
= g_new0(BdrvOpBlocker
, 1);
5890 blocker
->reason
= reason
;
5891 QLIST_INSERT_HEAD(&bs
->op_blockers
[op
], blocker
, list
);
5894 void bdrv_op_unblock(BlockDriverState
*bs
, BlockOpType op
, Error
*reason
)
5896 BdrvOpBlocker
*blocker
, *next
;
5897 assert((int) op
>= 0 && op
< BLOCK_OP_TYPE_MAX
);
5898 QLIST_FOREACH_SAFE(blocker
, &bs
->op_blockers
[op
], list
, next
) {
5899 if (blocker
->reason
== reason
) {
5900 QLIST_REMOVE(blocker
, list
);
5906 void bdrv_op_block_all(BlockDriverState
*bs
, Error
*reason
)
5909 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
5910 bdrv_op_block(bs
, i
, reason
);
5914 void bdrv_op_unblock_all(BlockDriverState
*bs
, Error
*reason
)
5917 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
5918 bdrv_op_unblock(bs
, i
, reason
);
5922 bool bdrv_op_blocker_is_empty(BlockDriverState
*bs
)
5926 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
5927 if (!QLIST_EMPTY(&bs
->op_blockers
[i
])) {
5934 void bdrv_img_create(const char *filename
, const char *fmt
,
5935 const char *base_filename
, const char *base_fmt
,
5936 char *options
, uint64_t img_size
, int flags
, bool quiet
,
5939 QemuOptsList
*create_opts
= NULL
;
5940 QemuOpts
*opts
= NULL
;
5941 const char *backing_fmt
, *backing_file
;
5943 BlockDriver
*drv
, *proto_drv
;
5944 Error
*local_err
= NULL
;
5947 /* Find driver and parse its options */
5948 drv
= bdrv_find_format(fmt
);
5950 error_setg(errp
, "Unknown file format '%s'", fmt
);
5954 proto_drv
= bdrv_find_protocol(filename
, true, errp
);
5959 if (!drv
->create_opts
) {
5960 error_setg(errp
, "Format driver '%s' does not support image creation",
5965 /* Create parameter list */
5966 create_opts
= qemu_opts_append(create_opts
, drv
->create_opts
);
5967 if (proto_drv
->create_opts
) {
5968 create_opts
= qemu_opts_append(create_opts
, proto_drv
->create_opts
);
5970 create_opts
= qemu_opts_append(create_opts
, &fallback_create_opts
);
5973 opts
= qemu_opts_create(create_opts
, NULL
, 0, &error_abort
);
5975 /* Parse -o options */
5977 qemu_opts_do_parse(opts
, options
, NULL
, &local_err
);
5983 if (!qemu_opt_get(opts
, BLOCK_OPT_SIZE
)) {
5984 qemu_opt_set_number(opts
, BLOCK_OPT_SIZE
, img_size
, &error_abort
);
5985 } else if (img_size
!= UINT64_C(-1)) {
5986 error_setg(errp
, "The image size must be specified only once");
5990 if (base_filename
) {
5991 qemu_opt_set(opts
, BLOCK_OPT_BACKING_FILE
, base_filename
, &local_err
);
5993 error_setg(errp
, "Backing file not supported for file format '%s'",
6000 qemu_opt_set(opts
, BLOCK_OPT_BACKING_FMT
, base_fmt
, &local_err
);
6002 error_setg(errp
, "Backing file format not supported for file "
6003 "format '%s'", fmt
);
6008 backing_file
= qemu_opt_get(opts
, BLOCK_OPT_BACKING_FILE
);
6010 if (!strcmp(filename
, backing_file
)) {
6011 error_setg(errp
, "Error: Trying to create an image with the "
6012 "same filename as the backing file");
6017 backing_fmt
= qemu_opt_get(opts
, BLOCK_OPT_BACKING_FMT
);
6019 /* The size for the image must always be specified, unless we have a backing
6020 * file and we have not been forbidden from opening it. */
6021 size
= qemu_opt_get_size(opts
, BLOCK_OPT_SIZE
, img_size
);
6022 if (backing_file
&& !(flags
& BDRV_O_NO_BACKING
)) {
6023 BlockDriverState
*bs
;
6026 QDict
*backing_options
= NULL
;
6029 bdrv_get_full_backing_filename_from_filename(filename
, backing_file
,
6034 assert(full_backing
);
6036 /* backing files always opened read-only */
6038 back_flags
&= ~(BDRV_O_RDWR
| BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
);
6040 backing_options
= qdict_new();
6042 qdict_put_str(backing_options
, "driver", backing_fmt
);
6044 qdict_put_bool(backing_options
, BDRV_OPT_FORCE_SHARE
, true);
6046 bs
= bdrv_open(full_backing
, NULL
, backing_options
, back_flags
,
6048 g_free(full_backing
);
6049 if (!bs
&& size
!= -1) {
6050 /* Couldn't open BS, but we have a size, so it's nonfatal */
6051 warn_reportf_err(local_err
,
6052 "Could not verify backing image. "
6053 "This may become an error in future versions.\n");
6056 /* Couldn't open bs, do not have size */
6057 error_append_hint(&local_err
,
6058 "Could not open backing image to determine size.\n");
6062 /* Opened BS, have no size */
6063 size
= bdrv_getlength(bs
);
6065 error_setg_errno(errp
, -size
, "Could not get size of '%s'",
6070 qemu_opt_set_number(opts
, BLOCK_OPT_SIZE
, size
, &error_abort
);
6074 } /* (backing_file && !(flags & BDRV_O_NO_BACKING)) */
6077 error_setg(errp
, "Image creation needs a size parameter");
6082 printf("Formatting '%s', fmt=%s ", filename
, fmt
);
6083 qemu_opts_print(opts
, " ");
6087 ret
= bdrv_create(drv
, filename
, opts
, &local_err
);
6089 if (ret
== -EFBIG
) {
6090 /* This is generally a better message than whatever the driver would
6091 * deliver (especially because of the cluster_size_hint), since that
6092 * is most probably not much different from "image too large". */
6093 const char *cluster_size_hint
= "";
6094 if (qemu_opt_get_size(opts
, BLOCK_OPT_CLUSTER_SIZE
, 0)) {
6095 cluster_size_hint
= " (try using a larger cluster size)";
6097 error_setg(errp
, "The image size is too large for file format '%s'"
6098 "%s", fmt
, cluster_size_hint
);
6099 error_free(local_err
);
6104 qemu_opts_del(opts
);
6105 qemu_opts_free(create_opts
);
6106 error_propagate(errp
, local_err
);
6109 AioContext
*bdrv_get_aio_context(BlockDriverState
*bs
)
6111 return bs
? bs
->aio_context
: qemu_get_aio_context();
6114 void bdrv_coroutine_enter(BlockDriverState
*bs
, Coroutine
*co
)
6116 aio_co_enter(bdrv_get_aio_context(bs
), co
);
6119 static void bdrv_do_remove_aio_context_notifier(BdrvAioNotifier
*ban
)
6121 QLIST_REMOVE(ban
, list
);
6125 static void bdrv_detach_aio_context(BlockDriverState
*bs
)
6127 BdrvAioNotifier
*baf
, *baf_tmp
;
6129 assert(!bs
->walking_aio_notifiers
);
6130 bs
->walking_aio_notifiers
= true;
6131 QLIST_FOREACH_SAFE(baf
, &bs
->aio_notifiers
, list
, baf_tmp
) {
6133 bdrv_do_remove_aio_context_notifier(baf
);
6135 baf
->detach_aio_context(baf
->opaque
);
6138 /* Never mind iterating again to check for ->deleted. bdrv_close() will
6139 * remove remaining aio notifiers if we aren't called again.
6141 bs
->walking_aio_notifiers
= false;
6143 if (bs
->drv
&& bs
->drv
->bdrv_detach_aio_context
) {
6144 bs
->drv
->bdrv_detach_aio_context(bs
);
6147 if (bs
->quiesce_counter
) {
6148 aio_enable_external(bs
->aio_context
);
6150 bs
->aio_context
= NULL
;
6153 static void bdrv_attach_aio_context(BlockDriverState
*bs
,
6154 AioContext
*new_context
)
6156 BdrvAioNotifier
*ban
, *ban_tmp
;
6158 if (bs
->quiesce_counter
) {
6159 aio_disable_external(new_context
);
6162 bs
->aio_context
= new_context
;
6164 if (bs
->drv
&& bs
->drv
->bdrv_attach_aio_context
) {
6165 bs
->drv
->bdrv_attach_aio_context(bs
, new_context
);
6168 assert(!bs
->walking_aio_notifiers
);
6169 bs
->walking_aio_notifiers
= true;
6170 QLIST_FOREACH_SAFE(ban
, &bs
->aio_notifiers
, list
, ban_tmp
) {
6172 bdrv_do_remove_aio_context_notifier(ban
);
6174 ban
->attached_aio_context(new_context
, ban
->opaque
);
6177 bs
->walking_aio_notifiers
= false;
6181 * Changes the AioContext used for fd handlers, timers, and BHs by this
6182 * BlockDriverState and all its children and parents.
6184 * Must be called from the main AioContext.
6186 * The caller must own the AioContext lock for the old AioContext of bs, but it
6187 * must not own the AioContext lock for new_context (unless new_context is the
6188 * same as the current context of bs).
6190 * @ignore will accumulate all visited BdrvChild object. The caller is
6191 * responsible for freeing the list afterwards.
6193 void bdrv_set_aio_context_ignore(BlockDriverState
*bs
,
6194 AioContext
*new_context
, GSList
**ignore
)
6196 AioContext
*old_context
= bdrv_get_aio_context(bs
);
6199 g_assert(qemu_get_current_aio_context() == qemu_get_aio_context());
6201 if (old_context
== new_context
) {
6205 bdrv_drained_begin(bs
);
6207 QLIST_FOREACH(child
, &bs
->children
, next
) {
6208 if (g_slist_find(*ignore
, child
)) {
6211 *ignore
= g_slist_prepend(*ignore
, child
);
6212 bdrv_set_aio_context_ignore(child
->bs
, new_context
, ignore
);
6214 QLIST_FOREACH(child
, &bs
->parents
, next_parent
) {
6215 if (g_slist_find(*ignore
, child
)) {
6218 assert(child
->role
->set_aio_ctx
);
6219 *ignore
= g_slist_prepend(*ignore
, child
);
6220 child
->role
->set_aio_ctx(child
, new_context
, ignore
);
6223 bdrv_detach_aio_context(bs
);
6225 /* Acquire the new context, if necessary */
6226 if (qemu_get_aio_context() != new_context
) {
6227 aio_context_acquire(new_context
);
6230 bdrv_attach_aio_context(bs
, new_context
);
6233 * If this function was recursively called from
6234 * bdrv_set_aio_context_ignore(), there may be nodes in the
6235 * subtree that have not yet been moved to the new AioContext.
6236 * Release the old one so bdrv_drained_end() can poll them.
6238 if (qemu_get_aio_context() != old_context
) {
6239 aio_context_release(old_context
);
6242 bdrv_drained_end(bs
);
6244 if (qemu_get_aio_context() != old_context
) {
6245 aio_context_acquire(old_context
);
6247 if (qemu_get_aio_context() != new_context
) {
6248 aio_context_release(new_context
);
6252 static bool bdrv_parent_can_set_aio_context(BdrvChild
*c
, AioContext
*ctx
,
6253 GSList
**ignore
, Error
**errp
)
6255 if (g_slist_find(*ignore
, c
)) {
6258 *ignore
= g_slist_prepend(*ignore
, c
);
6260 /* A BdrvChildRole that doesn't handle AioContext changes cannot
6261 * tolerate any AioContext changes */
6262 if (!c
->role
->can_set_aio_ctx
) {
6263 char *user
= bdrv_child_user_desc(c
);
6264 error_setg(errp
, "Changing iothreads is not supported by %s", user
);
6268 if (!c
->role
->can_set_aio_ctx(c
, ctx
, ignore
, errp
)) {
6269 assert(!errp
|| *errp
);
6275 bool bdrv_child_can_set_aio_context(BdrvChild
*c
, AioContext
*ctx
,
6276 GSList
**ignore
, Error
**errp
)
6278 if (g_slist_find(*ignore
, c
)) {
6281 *ignore
= g_slist_prepend(*ignore
, c
);
6282 return bdrv_can_set_aio_context(c
->bs
, ctx
, ignore
, errp
);
6285 /* @ignore will accumulate all visited BdrvChild object. The caller is
6286 * responsible for freeing the list afterwards. */
6287 bool bdrv_can_set_aio_context(BlockDriverState
*bs
, AioContext
*ctx
,
6288 GSList
**ignore
, Error
**errp
)
6292 if (bdrv_get_aio_context(bs
) == ctx
) {
6296 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
6297 if (!bdrv_parent_can_set_aio_context(c
, ctx
, ignore
, errp
)) {
6301 QLIST_FOREACH(c
, &bs
->children
, next
) {
6302 if (!bdrv_child_can_set_aio_context(c
, ctx
, ignore
, errp
)) {
6310 int bdrv_child_try_set_aio_context(BlockDriverState
*bs
, AioContext
*ctx
,
6311 BdrvChild
*ignore_child
, Error
**errp
)
6316 ignore
= ignore_child
? g_slist_prepend(NULL
, ignore_child
) : NULL
;
6317 ret
= bdrv_can_set_aio_context(bs
, ctx
, &ignore
, errp
);
6318 g_slist_free(ignore
);
6324 ignore
= ignore_child
? g_slist_prepend(NULL
, ignore_child
) : NULL
;
6325 bdrv_set_aio_context_ignore(bs
, ctx
, &ignore
);
6326 g_slist_free(ignore
);
6331 int bdrv_try_set_aio_context(BlockDriverState
*bs
, AioContext
*ctx
,
6334 return bdrv_child_try_set_aio_context(bs
, ctx
, NULL
, errp
);
6337 void bdrv_add_aio_context_notifier(BlockDriverState
*bs
,
6338 void (*attached_aio_context
)(AioContext
*new_context
, void *opaque
),
6339 void (*detach_aio_context
)(void *opaque
), void *opaque
)
6341 BdrvAioNotifier
*ban
= g_new(BdrvAioNotifier
, 1);
6342 *ban
= (BdrvAioNotifier
){
6343 .attached_aio_context
= attached_aio_context
,
6344 .detach_aio_context
= detach_aio_context
,
6348 QLIST_INSERT_HEAD(&bs
->aio_notifiers
, ban
, list
);
6351 void bdrv_remove_aio_context_notifier(BlockDriverState
*bs
,
6352 void (*attached_aio_context
)(AioContext
*,
6354 void (*detach_aio_context
)(void *),
6357 BdrvAioNotifier
*ban
, *ban_next
;
6359 QLIST_FOREACH_SAFE(ban
, &bs
->aio_notifiers
, list
, ban_next
) {
6360 if (ban
->attached_aio_context
== attached_aio_context
&&
6361 ban
->detach_aio_context
== detach_aio_context
&&
6362 ban
->opaque
== opaque
&&
6363 ban
->deleted
== false)
6365 if (bs
->walking_aio_notifiers
) {
6366 ban
->deleted
= true;
6368 bdrv_do_remove_aio_context_notifier(ban
);
6377 int bdrv_amend_options(BlockDriverState
*bs
, QemuOpts
*opts
,
6378 BlockDriverAmendStatusCB
*status_cb
, void *cb_opaque
,
6382 error_setg(errp
, "Node is ejected");
6385 if (!bs
->drv
->bdrv_amend_options
) {
6386 error_setg(errp
, "Block driver '%s' does not support option amendment",
6387 bs
->drv
->format_name
);
6390 return bs
->drv
->bdrv_amend_options(bs
, opts
, status_cb
, cb_opaque
, errp
);
6394 * This function checks whether the given @to_replace is allowed to be
6395 * replaced by a node that always shows the same data as @bs. This is
6396 * used for example to verify whether the mirror job can replace
6397 * @to_replace by the target mirrored from @bs.
6398 * To be replaceable, @bs and @to_replace may either be guaranteed to
6399 * always show the same data (because they are only connected through
6400 * filters), or some driver may allow replacing one of its children
6401 * because it can guarantee that this child's data is not visible at
6402 * all (for example, for dissenting quorum children that have no other
6405 bool bdrv_recurse_can_replace(BlockDriverState
*bs
,
6406 BlockDriverState
*to_replace
)
6408 if (!bs
|| !bs
->drv
) {
6412 if (bs
== to_replace
) {
6416 /* See what the driver can do */
6417 if (bs
->drv
->bdrv_recurse_can_replace
) {
6418 return bs
->drv
->bdrv_recurse_can_replace(bs
, to_replace
);
6421 /* For filters without an own implementation, we can recurse on our own */
6422 if (bs
->drv
->is_filter
) {
6423 BdrvChild
*child
= bs
->file
?: bs
->backing
;
6424 return bdrv_recurse_can_replace(child
->bs
, to_replace
);
6432 * Check whether the given @node_name can be replaced by a node that
6433 * has the same data as @parent_bs. If so, return @node_name's BDS;
6436 * @node_name must be a (recursive) *child of @parent_bs (or this
6437 * function will return NULL).
6439 * The result (whether the node can be replaced or not) is only valid
6440 * for as long as no graph or permission changes occur.
6442 BlockDriverState
*check_to_replace_node(BlockDriverState
*parent_bs
,
6443 const char *node_name
, Error
**errp
)
6445 BlockDriverState
*to_replace_bs
= bdrv_find_node(node_name
);
6446 AioContext
*aio_context
;
6448 if (!to_replace_bs
) {
6449 error_setg(errp
, "Node name '%s' not found", node_name
);
6453 aio_context
= bdrv_get_aio_context(to_replace_bs
);
6454 aio_context_acquire(aio_context
);
6456 if (bdrv_op_is_blocked(to_replace_bs
, BLOCK_OP_TYPE_REPLACE
, errp
)) {
6457 to_replace_bs
= NULL
;
6461 /* We don't want arbitrary node of the BDS chain to be replaced only the top
6462 * most non filter in order to prevent data corruption.
6463 * Another benefit is that this tests exclude backing files which are
6464 * blocked by the backing blockers.
6466 if (!bdrv_recurse_can_replace(parent_bs
, to_replace_bs
)) {
6467 error_setg(errp
, "Cannot replace '%s' by a node mirrored from '%s', "
6468 "because it cannot be guaranteed that doing so would not "
6469 "lead to an abrupt change of visible data",
6470 node_name
, parent_bs
->node_name
);
6471 to_replace_bs
= NULL
;
6476 aio_context_release(aio_context
);
6477 return to_replace_bs
;
6481 * Iterates through the list of runtime option keys that are said to
6482 * be "strong" for a BDS. An option is called "strong" if it changes
6483 * a BDS's data. For example, the null block driver's "size" and
6484 * "read-zeroes" options are strong, but its "latency-ns" option is
6487 * If a key returned by this function ends with a dot, all options
6488 * starting with that prefix are strong.
6490 static const char *const *strong_options(BlockDriverState
*bs
,
6491 const char *const *curopt
)
6493 static const char *const global_options
[] = {
6494 "driver", "filename", NULL
6498 return &global_options
[0];
6502 if (curopt
== &global_options
[ARRAY_SIZE(global_options
) - 1] && bs
->drv
) {
6503 curopt
= bs
->drv
->strong_runtime_opts
;
6506 return (curopt
&& *curopt
) ? curopt
: NULL
;
6510 * Copies all strong runtime options from bs->options to the given
6511 * QDict. The set of strong option keys is determined by invoking
6514 * Returns true iff any strong option was present in bs->options (and
6515 * thus copied to the target QDict) with the exception of "filename"
6516 * and "driver". The caller is expected to use this value to decide
6517 * whether the existence of strong options prevents the generation of
6520 static bool append_strong_runtime_options(QDict
*d
, BlockDriverState
*bs
)
6522 bool found_any
= false;
6523 const char *const *option_name
= NULL
;
6529 while ((option_name
= strong_options(bs
, option_name
))) {
6530 bool option_given
= false;
6532 assert(strlen(*option_name
) > 0);
6533 if ((*option_name
)[strlen(*option_name
) - 1] != '.') {
6534 QObject
*entry
= qdict_get(bs
->options
, *option_name
);
6539 qdict_put_obj(d
, *option_name
, qobject_ref(entry
));
6540 option_given
= true;
6542 const QDictEntry
*entry
;
6543 for (entry
= qdict_first(bs
->options
); entry
;
6544 entry
= qdict_next(bs
->options
, entry
))
6546 if (strstart(qdict_entry_key(entry
), *option_name
, NULL
)) {
6547 qdict_put_obj(d
, qdict_entry_key(entry
),
6548 qobject_ref(qdict_entry_value(entry
)));
6549 option_given
= true;
6554 /* While "driver" and "filename" need to be included in a JSON filename,
6555 * their existence does not prohibit generation of a plain filename. */
6556 if (!found_any
&& option_given
&&
6557 strcmp(*option_name
, "driver") && strcmp(*option_name
, "filename"))
6563 if (!qdict_haskey(d
, "driver")) {
6564 /* Drivers created with bdrv_new_open_driver() may not have a
6565 * @driver option. Add it here. */
6566 qdict_put_str(d
, "driver", bs
->drv
->format_name
);
6572 /* Note: This function may return false positives; it may return true
6573 * even if opening the backing file specified by bs's image header
6574 * would result in exactly bs->backing. */
6575 static bool bdrv_backing_overridden(BlockDriverState
*bs
)
6578 return strcmp(bs
->auto_backing_file
,
6579 bs
->backing
->bs
->filename
);
6581 /* No backing BDS, so if the image header reports any backing
6582 * file, it must have been suppressed */
6583 return bs
->auto_backing_file
[0] != '\0';
6587 /* Updates the following BDS fields:
6588 * - exact_filename: A filename which may be used for opening a block device
6589 * which (mostly) equals the given BDS (even without any
6590 * other options; so reading and writing must return the same
6591 * results, but caching etc. may be different)
6592 * - full_open_options: Options which, when given when opening a block device
6593 * (without a filename), result in a BDS (mostly)
6594 * equalling the given one
6595 * - filename: If exact_filename is set, it is copied here. Otherwise,
6596 * full_open_options is converted to a JSON object, prefixed with
6597 * "json:" (for use through the JSON pseudo protocol) and put here.
6599 void bdrv_refresh_filename(BlockDriverState
*bs
)
6601 BlockDriver
*drv
= bs
->drv
;
6604 bool backing_overridden
;
6605 bool generate_json_filename
; /* Whether our default implementation should
6606 fill exact_filename (false) or not (true) */
6612 /* This BDS's file name may depend on any of its children's file names, so
6613 * refresh those first */
6614 QLIST_FOREACH(child
, &bs
->children
, next
) {
6615 bdrv_refresh_filename(child
->bs
);
6619 /* For implicit nodes, just copy everything from the single child */
6620 child
= QLIST_FIRST(&bs
->children
);
6621 assert(QLIST_NEXT(child
, next
) == NULL
);
6623 pstrcpy(bs
->exact_filename
, sizeof(bs
->exact_filename
),
6624 child
->bs
->exact_filename
);
6625 pstrcpy(bs
->filename
, sizeof(bs
->filename
), child
->bs
->filename
);
6627 qobject_unref(bs
->full_open_options
);
6628 bs
->full_open_options
= qobject_ref(child
->bs
->full_open_options
);
6633 backing_overridden
= bdrv_backing_overridden(bs
);
6635 if (bs
->open_flags
& BDRV_O_NO_IO
) {
6636 /* Without I/O, the backing file does not change anything.
6637 * Therefore, in such a case (primarily qemu-img), we can
6638 * pretend the backing file has not been overridden even if
6639 * it technically has been. */
6640 backing_overridden
= false;
6643 /* Gather the options QDict */
6645 generate_json_filename
= append_strong_runtime_options(opts
, bs
);
6646 generate_json_filename
|= backing_overridden
;
6648 if (drv
->bdrv_gather_child_options
) {
6649 /* Some block drivers may not want to present all of their children's
6650 * options, or name them differently from BdrvChild.name */
6651 drv
->bdrv_gather_child_options(bs
, opts
, backing_overridden
);
6653 QLIST_FOREACH(child
, &bs
->children
, next
) {
6654 if (child
->role
== &child_backing
&& !backing_overridden
) {
6655 /* We can skip the backing BDS if it has not been overridden */
6659 qdict_put(opts
, child
->name
,
6660 qobject_ref(child
->bs
->full_open_options
));
6663 if (backing_overridden
&& !bs
->backing
) {
6664 /* Force no backing file */
6665 qdict_put_null(opts
, "backing");
6669 qobject_unref(bs
->full_open_options
);
6670 bs
->full_open_options
= opts
;
6672 if (drv
->bdrv_refresh_filename
) {
6673 /* Obsolete information is of no use here, so drop the old file name
6674 * information before refreshing it */
6675 bs
->exact_filename
[0] = '\0';
6677 drv
->bdrv_refresh_filename(bs
);
6678 } else if (bs
->file
) {
6679 /* Try to reconstruct valid information from the underlying file */
6681 bs
->exact_filename
[0] = '\0';
6684 * We can use the underlying file's filename if:
6685 * - it has a filename,
6686 * - the file is a protocol BDS, and
6687 * - opening that file (as this BDS's format) will automatically create
6688 * the BDS tree we have right now, that is:
6689 * - the user did not significantly change this BDS's behavior with
6690 * some explicit (strong) options
6691 * - no non-file child of this BDS has been overridden by the user
6692 * Both of these conditions are represented by generate_json_filename.
6694 if (bs
->file
->bs
->exact_filename
[0] &&
6695 bs
->file
->bs
->drv
->bdrv_file_open
&&
6696 !generate_json_filename
)
6698 strcpy(bs
->exact_filename
, bs
->file
->bs
->exact_filename
);
6702 if (bs
->exact_filename
[0]) {
6703 pstrcpy(bs
->filename
, sizeof(bs
->filename
), bs
->exact_filename
);
6705 QString
*json
= qobject_to_json(QOBJECT(bs
->full_open_options
));
6706 snprintf(bs
->filename
, sizeof(bs
->filename
), "json:%s",
6707 qstring_get_str(json
));
6708 qobject_unref(json
);
6712 char *bdrv_dirname(BlockDriverState
*bs
, Error
**errp
)
6714 BlockDriver
*drv
= bs
->drv
;
6717 error_setg(errp
, "Node '%s' is ejected", bs
->node_name
);
6721 if (drv
->bdrv_dirname
) {
6722 return drv
->bdrv_dirname(bs
, errp
);
6726 return bdrv_dirname(bs
->file
->bs
, errp
);
6729 bdrv_refresh_filename(bs
);
6730 if (bs
->exact_filename
[0] != '\0') {
6731 return path_combine(bs
->exact_filename
, "");
6734 error_setg(errp
, "Cannot generate a base directory for %s nodes",
6740 * Hot add/remove a BDS's child. So the user can take a child offline when
6741 * it is broken and take a new child online
6743 void bdrv_add_child(BlockDriverState
*parent_bs
, BlockDriverState
*child_bs
,
6747 if (!parent_bs
->drv
|| !parent_bs
->drv
->bdrv_add_child
) {
6748 error_setg(errp
, "The node %s does not support adding a child",
6749 bdrv_get_device_or_node_name(parent_bs
));
6753 if (!QLIST_EMPTY(&child_bs
->parents
)) {
6754 error_setg(errp
, "The node %s already has a parent",
6755 child_bs
->node_name
);
6759 parent_bs
->drv
->bdrv_add_child(parent_bs
, child_bs
, errp
);
6762 void bdrv_del_child(BlockDriverState
*parent_bs
, BdrvChild
*child
, Error
**errp
)
6766 if (!parent_bs
->drv
|| !parent_bs
->drv
->bdrv_del_child
) {
6767 error_setg(errp
, "The node %s does not support removing a child",
6768 bdrv_get_device_or_node_name(parent_bs
));
6772 QLIST_FOREACH(tmp
, &parent_bs
->children
, next
) {
6779 error_setg(errp
, "The node %s does not have a child named %s",
6780 bdrv_get_device_or_node_name(parent_bs
),
6781 bdrv_get_device_or_node_name(child
->bs
));
6785 parent_bs
->drv
->bdrv_del_child(parent_bs
, child
, errp
);