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 BdrvChildClass
*child_class
,
80 BdrvChildRole child_role
,
83 /* If non-zero, use only whitelisted block drivers */
84 static int use_bdrv_whitelist
;
87 static int is_windows_drive_prefix(const char *filename
)
89 return (((filename
[0] >= 'a' && filename
[0] <= 'z') ||
90 (filename
[0] >= 'A' && filename
[0] <= 'Z')) &&
94 int is_windows_drive(const char *filename
)
96 if (is_windows_drive_prefix(filename
) &&
99 if (strstart(filename
, "\\\\.\\", NULL
) ||
100 strstart(filename
, "//./", NULL
))
106 size_t bdrv_opt_mem_align(BlockDriverState
*bs
)
108 if (!bs
|| !bs
->drv
) {
109 /* page size or 4k (hdd sector size) should be on the safe side */
110 return MAX(4096, qemu_real_host_page_size
);
113 return bs
->bl
.opt_mem_alignment
;
116 size_t bdrv_min_mem_align(BlockDriverState
*bs
)
118 if (!bs
|| !bs
->drv
) {
119 /* page size or 4k (hdd sector size) should be on the safe side */
120 return MAX(4096, qemu_real_host_page_size
);
123 return bs
->bl
.min_mem_alignment
;
126 /* check if the path starts with "<protocol>:" */
127 int path_has_protocol(const char *path
)
132 if (is_windows_drive(path
) ||
133 is_windows_drive_prefix(path
)) {
136 p
= path
+ strcspn(path
, ":/\\");
138 p
= path
+ strcspn(path
, ":/");
144 int path_is_absolute(const char *path
)
147 /* specific case for names like: "\\.\d:" */
148 if (is_windows_drive(path
) || is_windows_drive_prefix(path
)) {
151 return (*path
== '/' || *path
== '\\');
153 return (*path
== '/');
157 /* if filename is absolute, just return its duplicate. Otherwise, build a
158 path to it by considering it is relative to base_path. URL are
160 char *path_combine(const char *base_path
, const char *filename
)
162 const char *protocol_stripped
= NULL
;
167 if (path_is_absolute(filename
)) {
168 return g_strdup(filename
);
171 if (path_has_protocol(base_path
)) {
172 protocol_stripped
= strchr(base_path
, ':');
173 if (protocol_stripped
) {
177 p
= protocol_stripped
?: base_path
;
179 p1
= strrchr(base_path
, '/');
183 p2
= strrchr(base_path
, '\\');
184 if (!p1
|| p2
> p1
) {
199 result
= g_malloc(len
+ strlen(filename
) + 1);
200 memcpy(result
, base_path
, len
);
201 strcpy(result
+ len
, filename
);
207 * Helper function for bdrv_parse_filename() implementations to remove optional
208 * protocol prefixes (especially "file:") from a filename and for putting the
209 * stripped filename into the options QDict if there is such a prefix.
211 void bdrv_parse_filename_strip_prefix(const char *filename
, const char *prefix
,
214 if (strstart(filename
, prefix
, &filename
)) {
215 /* Stripping the explicit protocol prefix may result in a protocol
216 * prefix being (wrongly) detected (if the filename contains a colon) */
217 if (path_has_protocol(filename
)) {
218 QString
*fat_filename
;
220 /* This means there is some colon before the first slash; therefore,
221 * this cannot be an absolute path */
222 assert(!path_is_absolute(filename
));
224 /* And we can thus fix the protocol detection issue by prefixing it
226 fat_filename
= qstring_from_str("./");
227 qstring_append(fat_filename
, filename
);
229 assert(!path_has_protocol(qstring_get_str(fat_filename
)));
231 qdict_put(options
, "filename", fat_filename
);
233 /* If no protocol prefix was detected, we can use the shortened
235 qdict_put_str(options
, "filename", filename
);
241 /* Returns whether the image file is opened as read-only. Note that this can
242 * return false and writing to the image file is still not possible because the
243 * image is inactivated. */
244 bool bdrv_is_read_only(BlockDriverState
*bs
)
246 return bs
->read_only
;
249 int bdrv_can_set_read_only(BlockDriverState
*bs
, bool read_only
,
250 bool ignore_allow_rdw
, Error
**errp
)
252 /* Do not set read_only if copy_on_read is enabled */
253 if (bs
->copy_on_read
&& read_only
) {
254 error_setg(errp
, "Can't set node '%s' to r/o with copy-on-read enabled",
255 bdrv_get_device_or_node_name(bs
));
259 /* Do not clear read_only if it is prohibited */
260 if (!read_only
&& !(bs
->open_flags
& BDRV_O_ALLOW_RDWR
) &&
263 error_setg(errp
, "Node '%s' is read only",
264 bdrv_get_device_or_node_name(bs
));
272 * Called by a driver that can only provide a read-only image.
274 * Returns 0 if the node is already read-only or it could switch the node to
275 * read-only because BDRV_O_AUTO_RDONLY is set.
277 * Returns -EACCES if the node is read-write and BDRV_O_AUTO_RDONLY is not set
278 * or bdrv_can_set_read_only() forbids making the node read-only. If @errmsg
279 * is not NULL, it is used as the error message for the Error object.
281 int bdrv_apply_auto_read_only(BlockDriverState
*bs
, const char *errmsg
,
286 if (!(bs
->open_flags
& BDRV_O_RDWR
)) {
289 if (!(bs
->open_flags
& BDRV_O_AUTO_RDONLY
)) {
293 ret
= bdrv_can_set_read_only(bs
, true, false, NULL
);
298 bs
->read_only
= true;
299 bs
->open_flags
&= ~BDRV_O_RDWR
;
304 error_setg(errp
, "%s", errmsg
?: "Image is read-only");
309 * If @backing is empty, this function returns NULL without setting
310 * @errp. In all other cases, NULL will only be returned with @errp
313 * Therefore, a return value of NULL without @errp set means that
314 * there is no backing file; if @errp is set, there is one but its
315 * absolute filename cannot be generated.
317 char *bdrv_get_full_backing_filename_from_filename(const char *backed
,
321 if (backing
[0] == '\0') {
323 } else if (path_has_protocol(backing
) || path_is_absolute(backing
)) {
324 return g_strdup(backing
);
325 } else if (backed
[0] == '\0' || strstart(backed
, "json:", NULL
)) {
326 error_setg(errp
, "Cannot use relative backing file names for '%s'",
330 return path_combine(backed
, backing
);
335 * If @filename is empty or NULL, this function returns NULL without
336 * setting @errp. In all other cases, NULL will only be returned with
339 static char *bdrv_make_absolute_filename(BlockDriverState
*relative_to
,
340 const char *filename
, Error
**errp
)
342 char *dir
, *full_name
;
344 if (!filename
|| filename
[0] == '\0') {
346 } else if (path_has_protocol(filename
) || path_is_absolute(filename
)) {
347 return g_strdup(filename
);
350 dir
= bdrv_dirname(relative_to
, errp
);
355 full_name
= g_strconcat(dir
, filename
, NULL
);
360 char *bdrv_get_full_backing_filename(BlockDriverState
*bs
, Error
**errp
)
362 return bdrv_make_absolute_filename(bs
, bs
->backing_file
, errp
);
365 void bdrv_register(BlockDriver
*bdrv
)
367 assert(bdrv
->format_name
);
368 QLIST_INSERT_HEAD(&bdrv_drivers
, bdrv
, list
);
371 BlockDriverState
*bdrv_new(void)
373 BlockDriverState
*bs
;
376 bs
= g_new0(BlockDriverState
, 1);
377 QLIST_INIT(&bs
->dirty_bitmaps
);
378 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
379 QLIST_INIT(&bs
->op_blockers
[i
]);
381 notifier_with_return_list_init(&bs
->before_write_notifiers
);
382 qemu_co_mutex_init(&bs
->reqs_lock
);
383 qemu_mutex_init(&bs
->dirty_bitmap_mutex
);
385 bs
->aio_context
= qemu_get_aio_context();
387 qemu_co_queue_init(&bs
->flush_queue
);
389 for (i
= 0; i
< bdrv_drain_all_count
; i
++) {
390 bdrv_drained_begin(bs
);
393 QTAILQ_INSERT_TAIL(&all_bdrv_states
, bs
, bs_list
);
398 static BlockDriver
*bdrv_do_find_format(const char *format_name
)
402 QLIST_FOREACH(drv1
, &bdrv_drivers
, list
) {
403 if (!strcmp(drv1
->format_name
, format_name
)) {
411 BlockDriver
*bdrv_find_format(const char *format_name
)
416 drv1
= bdrv_do_find_format(format_name
);
421 /* The driver isn't registered, maybe we need to load a module */
422 for (i
= 0; i
< (int)ARRAY_SIZE(block_driver_modules
); ++i
) {
423 if (!strcmp(block_driver_modules
[i
].format_name
, format_name
)) {
424 block_module_load_one(block_driver_modules
[i
].library_name
);
429 return bdrv_do_find_format(format_name
);
432 static int bdrv_format_is_whitelisted(const char *format_name
, bool read_only
)
434 static const char *whitelist_rw
[] = {
435 CONFIG_BDRV_RW_WHITELIST
437 static const char *whitelist_ro
[] = {
438 CONFIG_BDRV_RO_WHITELIST
442 if (!whitelist_rw
[0] && !whitelist_ro
[0]) {
443 return 1; /* no whitelist, anything goes */
446 for (p
= whitelist_rw
; *p
; p
++) {
447 if (!strcmp(format_name
, *p
)) {
452 for (p
= whitelist_ro
; *p
; p
++) {
453 if (!strcmp(format_name
, *p
)) {
461 int bdrv_is_whitelisted(BlockDriver
*drv
, bool read_only
)
463 return bdrv_format_is_whitelisted(drv
->format_name
, read_only
);
466 bool bdrv_uses_whitelist(void)
468 return use_bdrv_whitelist
;
471 typedef struct CreateCo
{
479 static void coroutine_fn
bdrv_create_co_entry(void *opaque
)
481 Error
*local_err
= NULL
;
484 CreateCo
*cco
= opaque
;
487 ret
= cco
->drv
->bdrv_co_create_opts(cco
->drv
,
488 cco
->filename
, cco
->opts
, &local_err
);
489 error_propagate(&cco
->err
, local_err
);
493 int bdrv_create(BlockDriver
*drv
, const char* filename
,
494 QemuOpts
*opts
, Error
**errp
)
501 .filename
= g_strdup(filename
),
507 if (!drv
->bdrv_co_create_opts
) {
508 error_setg(errp
, "Driver '%s' does not support image creation", drv
->format_name
);
513 if (qemu_in_coroutine()) {
514 /* Fast-path if already in coroutine context */
515 bdrv_create_co_entry(&cco
);
517 co
= qemu_coroutine_create(bdrv_create_co_entry
, &cco
);
518 qemu_coroutine_enter(co
);
519 while (cco
.ret
== NOT_DONE
) {
520 aio_poll(qemu_get_aio_context(), true);
527 error_propagate(errp
, cco
.err
);
529 error_setg_errno(errp
, -ret
, "Could not create image");
534 g_free(cco
.filename
);
539 * Helper function for bdrv_create_file_fallback(): Resize @blk to at
540 * least the given @minimum_size.
542 * On success, return @blk's actual length.
543 * Otherwise, return -errno.
545 static int64_t create_file_fallback_truncate(BlockBackend
*blk
,
546 int64_t minimum_size
, Error
**errp
)
548 Error
*local_err
= NULL
;
552 ret
= blk_truncate(blk
, minimum_size
, false, PREALLOC_MODE_OFF
, 0,
554 if (ret
< 0 && ret
!= -ENOTSUP
) {
555 error_propagate(errp
, local_err
);
559 size
= blk_getlength(blk
);
561 error_free(local_err
);
562 error_setg_errno(errp
, -size
,
563 "Failed to inquire the new image file's length");
567 if (size
< minimum_size
) {
568 /* Need to grow the image, but we failed to do that */
569 error_propagate(errp
, local_err
);
573 error_free(local_err
);
580 * Helper function for bdrv_create_file_fallback(): Zero the first
581 * sector to remove any potentially pre-existing image header.
583 static int create_file_fallback_zero_first_sector(BlockBackend
*blk
,
584 int64_t current_size
,
587 int64_t bytes_to_clear
;
590 bytes_to_clear
= MIN(current_size
, BDRV_SECTOR_SIZE
);
591 if (bytes_to_clear
) {
592 ret
= blk_pwrite_zeroes(blk
, 0, bytes_to_clear
, BDRV_REQ_MAY_UNMAP
);
594 error_setg_errno(errp
, -ret
,
595 "Failed to clear the new image's first sector");
604 * Simple implementation of bdrv_co_create_opts for protocol drivers
605 * which only support creation via opening a file
606 * (usually existing raw storage device)
608 int coroutine_fn
bdrv_co_create_opts_simple(BlockDriver
*drv
,
609 const char *filename
,
617 PreallocMode prealloc
;
618 Error
*local_err
= NULL
;
621 size
= qemu_opt_get_size_del(opts
, BLOCK_OPT_SIZE
, 0);
622 buf
= qemu_opt_get_del(opts
, BLOCK_OPT_PREALLOC
);
623 prealloc
= qapi_enum_parse(&PreallocMode_lookup
, buf
,
624 PREALLOC_MODE_OFF
, &local_err
);
627 error_propagate(errp
, local_err
);
631 if (prealloc
!= PREALLOC_MODE_OFF
) {
632 error_setg(errp
, "Unsupported preallocation mode '%s'",
633 PreallocMode_str(prealloc
));
637 options
= qdict_new();
638 qdict_put_str(options
, "driver", drv
->format_name
);
640 blk
= blk_new_open(filename
, NULL
, options
,
641 BDRV_O_RDWR
| BDRV_O_RESIZE
, errp
);
643 error_prepend(errp
, "Protocol driver '%s' does not support image "
644 "creation, and opening the image failed: ",
649 size
= create_file_fallback_truncate(blk
, size
, errp
);
655 ret
= create_file_fallback_zero_first_sector(blk
, size
, errp
);
666 int bdrv_create_file(const char *filename
, QemuOpts
*opts
, Error
**errp
)
670 drv
= bdrv_find_protocol(filename
, true, errp
);
675 return bdrv_create(drv
, filename
, opts
, errp
);
678 int coroutine_fn
bdrv_co_delete_file(BlockDriverState
*bs
, Error
**errp
)
680 Error
*local_err
= NULL
;
686 error_setg(errp
, "Block node '%s' is not opened", bs
->filename
);
690 if (!bs
->drv
->bdrv_co_delete_file
) {
691 error_setg(errp
, "Driver '%s' does not support image deletion",
692 bs
->drv
->format_name
);
696 ret
= bs
->drv
->bdrv_co_delete_file(bs
, &local_err
);
698 error_propagate(errp
, local_err
);
705 * Try to get @bs's logical and physical block size.
706 * On success, store them in @bsz struct and return 0.
707 * On failure return -errno.
708 * @bs must not be empty.
710 int bdrv_probe_blocksizes(BlockDriverState
*bs
, BlockSizes
*bsz
)
712 BlockDriver
*drv
= bs
->drv
;
714 if (drv
&& drv
->bdrv_probe_blocksizes
) {
715 return drv
->bdrv_probe_blocksizes(bs
, bsz
);
716 } else if (drv
&& drv
->is_filter
&& bs
->file
) {
717 return bdrv_probe_blocksizes(bs
->file
->bs
, bsz
);
724 * Try to get @bs's geometry (cyls, heads, sectors).
725 * On success, store them in @geo struct and return 0.
726 * On failure return -errno.
727 * @bs must not be empty.
729 int bdrv_probe_geometry(BlockDriverState
*bs
, HDGeometry
*geo
)
731 BlockDriver
*drv
= bs
->drv
;
733 if (drv
&& drv
->bdrv_probe_geometry
) {
734 return drv
->bdrv_probe_geometry(bs
, geo
);
735 } else if (drv
&& drv
->is_filter
&& bs
->file
) {
736 return bdrv_probe_geometry(bs
->file
->bs
, geo
);
743 * Create a uniquely-named empty temporary file.
744 * Return 0 upon success, otherwise a negative errno value.
746 int get_tmp_filename(char *filename
, int size
)
749 char temp_dir
[MAX_PATH
];
750 /* GetTempFileName requires that its output buffer (4th param)
751 have length MAX_PATH or greater. */
752 assert(size
>= MAX_PATH
);
753 return (GetTempPath(MAX_PATH
, temp_dir
)
754 && GetTempFileName(temp_dir
, "qem", 0, filename
)
755 ? 0 : -GetLastError());
759 tmpdir
= getenv("TMPDIR");
763 if (snprintf(filename
, size
, "%s/vl.XXXXXX", tmpdir
) >= size
) {
766 fd
= mkstemp(filename
);
770 if (close(fd
) != 0) {
779 * Detect host devices. By convention, /dev/cdrom[N] is always
780 * recognized as a host CDROM.
782 static BlockDriver
*find_hdev_driver(const char *filename
)
784 int score_max
= 0, score
;
785 BlockDriver
*drv
= NULL
, *d
;
787 QLIST_FOREACH(d
, &bdrv_drivers
, list
) {
788 if (d
->bdrv_probe_device
) {
789 score
= d
->bdrv_probe_device(filename
);
790 if (score
> score_max
) {
800 static BlockDriver
*bdrv_do_find_protocol(const char *protocol
)
804 QLIST_FOREACH(drv1
, &bdrv_drivers
, list
) {
805 if (drv1
->protocol_name
&& !strcmp(drv1
->protocol_name
, protocol
)) {
813 BlockDriver
*bdrv_find_protocol(const char *filename
,
814 bool allow_protocol_prefix
,
823 /* TODO Drivers without bdrv_file_open must be specified explicitly */
826 * XXX(hch): we really should not let host device detection
827 * override an explicit protocol specification, but moving this
828 * later breaks access to device names with colons in them.
829 * Thanks to the brain-dead persistent naming schemes on udev-
830 * based Linux systems those actually are quite common.
832 drv1
= find_hdev_driver(filename
);
837 if (!path_has_protocol(filename
) || !allow_protocol_prefix
) {
841 p
= strchr(filename
, ':');
844 if (len
> sizeof(protocol
) - 1)
845 len
= sizeof(protocol
) - 1;
846 memcpy(protocol
, filename
, len
);
847 protocol
[len
] = '\0';
849 drv1
= bdrv_do_find_protocol(protocol
);
854 for (i
= 0; i
< (int)ARRAY_SIZE(block_driver_modules
); ++i
) {
855 if (block_driver_modules
[i
].protocol_name
&&
856 !strcmp(block_driver_modules
[i
].protocol_name
, protocol
)) {
857 block_module_load_one(block_driver_modules
[i
].library_name
);
862 drv1
= bdrv_do_find_protocol(protocol
);
864 error_setg(errp
, "Unknown protocol '%s'", protocol
);
870 * Guess image format by probing its contents.
871 * This is not a good idea when your image is raw (CVE-2008-2004), but
872 * we do it anyway for backward compatibility.
874 * @buf contains the image's first @buf_size bytes.
875 * @buf_size is the buffer size in bytes (generally BLOCK_PROBE_BUF_SIZE,
876 * but can be smaller if the image file is smaller)
877 * @filename is its filename.
879 * For all block drivers, call the bdrv_probe() method to get its
881 * Return the first block driver with the highest probing score.
883 BlockDriver
*bdrv_probe_all(const uint8_t *buf
, int buf_size
,
884 const char *filename
)
886 int score_max
= 0, score
;
887 BlockDriver
*drv
= NULL
, *d
;
889 QLIST_FOREACH(d
, &bdrv_drivers
, list
) {
891 score
= d
->bdrv_probe(buf
, buf_size
, filename
);
892 if (score
> score_max
) {
902 static int find_image_format(BlockBackend
*file
, const char *filename
,
903 BlockDriver
**pdrv
, Error
**errp
)
906 uint8_t buf
[BLOCK_PROBE_BUF_SIZE
];
909 /* Return the raw BlockDriver * to scsi-generic devices or empty drives */
910 if (blk_is_sg(file
) || !blk_is_inserted(file
) || blk_getlength(file
) == 0) {
915 ret
= blk_pread(file
, 0, buf
, sizeof(buf
));
917 error_setg_errno(errp
, -ret
, "Could not read image for determining its "
923 drv
= bdrv_probe_all(buf
, ret
, filename
);
925 error_setg(errp
, "Could not determine image format: No compatible "
934 * Set the current 'total_sectors' value
935 * Return 0 on success, -errno on error.
937 int refresh_total_sectors(BlockDriverState
*bs
, int64_t hint
)
939 BlockDriver
*drv
= bs
->drv
;
945 /* Do not attempt drv->bdrv_getlength() on scsi-generic devices */
949 /* query actual device if possible, otherwise just trust the hint */
950 if (drv
->bdrv_getlength
) {
951 int64_t length
= drv
->bdrv_getlength(bs
);
955 hint
= DIV_ROUND_UP(length
, BDRV_SECTOR_SIZE
);
958 bs
->total_sectors
= hint
;
963 * Combines a QDict of new block driver @options with any missing options taken
964 * from @old_options, so that leaving out an option defaults to its old value.
966 static void bdrv_join_options(BlockDriverState
*bs
, QDict
*options
,
969 if (bs
->drv
&& bs
->drv
->bdrv_join_options
) {
970 bs
->drv
->bdrv_join_options(options
, old_options
);
972 qdict_join(options
, old_options
, false);
976 static BlockdevDetectZeroesOptions
bdrv_parse_detect_zeroes(QemuOpts
*opts
,
980 Error
*local_err
= NULL
;
981 char *value
= qemu_opt_get_del(opts
, "detect-zeroes");
982 BlockdevDetectZeroesOptions detect_zeroes
=
983 qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup
, value
,
984 BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF
, &local_err
);
987 error_propagate(errp
, local_err
);
988 return detect_zeroes
;
991 if (detect_zeroes
== BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP
&&
992 !(open_flags
& BDRV_O_UNMAP
))
994 error_setg(errp
, "setting detect-zeroes to unmap is not allowed "
995 "without setting discard operation to unmap");
998 return detect_zeroes
;
1002 * Set open flags for aio engine
1004 * Return 0 on success, -1 if the engine specified is invalid
1006 int bdrv_parse_aio(const char *mode
, int *flags
)
1008 if (!strcmp(mode
, "threads")) {
1009 /* do nothing, default */
1010 } else if (!strcmp(mode
, "native")) {
1011 *flags
|= BDRV_O_NATIVE_AIO
;
1012 #ifdef CONFIG_LINUX_IO_URING
1013 } else if (!strcmp(mode
, "io_uring")) {
1014 *flags
|= BDRV_O_IO_URING
;
1024 * Set open flags for a given discard mode
1026 * Return 0 on success, -1 if the discard mode was invalid.
1028 int bdrv_parse_discard_flags(const char *mode
, int *flags
)
1030 *flags
&= ~BDRV_O_UNMAP
;
1032 if (!strcmp(mode
, "off") || !strcmp(mode
, "ignore")) {
1034 } else if (!strcmp(mode
, "on") || !strcmp(mode
, "unmap")) {
1035 *flags
|= BDRV_O_UNMAP
;
1044 * Set open flags for a given cache mode
1046 * Return 0 on success, -1 if the cache mode was invalid.
1048 int bdrv_parse_cache_mode(const char *mode
, int *flags
, bool *writethrough
)
1050 *flags
&= ~BDRV_O_CACHE_MASK
;
1052 if (!strcmp(mode
, "off") || !strcmp(mode
, "none")) {
1053 *writethrough
= false;
1054 *flags
|= BDRV_O_NOCACHE
;
1055 } else if (!strcmp(mode
, "directsync")) {
1056 *writethrough
= true;
1057 *flags
|= BDRV_O_NOCACHE
;
1058 } else if (!strcmp(mode
, "writeback")) {
1059 *writethrough
= false;
1060 } else if (!strcmp(mode
, "unsafe")) {
1061 *writethrough
= false;
1062 *flags
|= BDRV_O_NO_FLUSH
;
1063 } else if (!strcmp(mode
, "writethrough")) {
1064 *writethrough
= true;
1072 static char *bdrv_child_get_parent_desc(BdrvChild
*c
)
1074 BlockDriverState
*parent
= c
->opaque
;
1075 return g_strdup(bdrv_get_device_or_node_name(parent
));
1078 static void bdrv_child_cb_drained_begin(BdrvChild
*child
)
1080 BlockDriverState
*bs
= child
->opaque
;
1081 bdrv_do_drained_begin_quiesce(bs
, NULL
, false);
1084 static bool bdrv_child_cb_drained_poll(BdrvChild
*child
)
1086 BlockDriverState
*bs
= child
->opaque
;
1087 return bdrv_drain_poll(bs
, false, NULL
, false);
1090 static void bdrv_child_cb_drained_end(BdrvChild
*child
,
1091 int *drained_end_counter
)
1093 BlockDriverState
*bs
= child
->opaque
;
1094 bdrv_drained_end_no_poll(bs
, drained_end_counter
);
1097 static int bdrv_child_cb_inactivate(BdrvChild
*child
)
1099 BlockDriverState
*bs
= child
->opaque
;
1100 assert(bs
->open_flags
& BDRV_O_INACTIVE
);
1104 static bool bdrv_child_cb_can_set_aio_ctx(BdrvChild
*child
, AioContext
*ctx
,
1105 GSList
**ignore
, Error
**errp
)
1107 BlockDriverState
*bs
= child
->opaque
;
1108 return bdrv_can_set_aio_context(bs
, ctx
, ignore
, errp
);
1111 static void bdrv_child_cb_set_aio_ctx(BdrvChild
*child
, AioContext
*ctx
,
1114 BlockDriverState
*bs
= child
->opaque
;
1115 return bdrv_set_aio_context_ignore(bs
, ctx
, ignore
);
1119 * Returns the options and flags that a temporary snapshot should get, based on
1120 * the originally requested flags (the originally requested image will have
1121 * flags like a backing file)
1123 static void bdrv_temp_snapshot_options(int *child_flags
, QDict
*child_options
,
1124 int parent_flags
, QDict
*parent_options
)
1126 *child_flags
= (parent_flags
& ~BDRV_O_SNAPSHOT
) | BDRV_O_TEMPORARY
;
1128 /* For temporary files, unconditional cache=unsafe is fine */
1129 qdict_set_default_str(child_options
, BDRV_OPT_CACHE_DIRECT
, "off");
1130 qdict_set_default_str(child_options
, BDRV_OPT_CACHE_NO_FLUSH
, "on");
1132 /* Copy the read-only and discard options from the parent */
1133 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_READ_ONLY
);
1134 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_DISCARD
);
1136 /* aio=native doesn't work for cache.direct=off, so disable it for the
1137 * temporary snapshot */
1138 *child_flags
&= ~BDRV_O_NATIVE_AIO
;
1141 static void bdrv_backing_attach(BdrvChild
*c
)
1143 BlockDriverState
*parent
= c
->opaque
;
1144 BlockDriverState
*backing_hd
= c
->bs
;
1146 assert(!parent
->backing_blocker
);
1147 error_setg(&parent
->backing_blocker
,
1148 "node is used as backing hd of '%s'",
1149 bdrv_get_device_or_node_name(parent
));
1151 bdrv_refresh_filename(backing_hd
);
1153 parent
->open_flags
&= ~BDRV_O_NO_BACKING
;
1154 pstrcpy(parent
->backing_file
, sizeof(parent
->backing_file
),
1155 backing_hd
->filename
);
1156 pstrcpy(parent
->backing_format
, sizeof(parent
->backing_format
),
1157 backing_hd
->drv
? backing_hd
->drv
->format_name
: "");
1159 bdrv_op_block_all(backing_hd
, parent
->backing_blocker
);
1160 /* Otherwise we won't be able to commit or stream */
1161 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_COMMIT_TARGET
,
1162 parent
->backing_blocker
);
1163 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_STREAM
,
1164 parent
->backing_blocker
);
1166 * We do backup in 3 ways:
1168 * The target bs is new opened, and the source is top BDS
1169 * 2. blockdev backup
1170 * Both the source and the target are top BDSes.
1171 * 3. internal backup(used for block replication)
1172 * Both the source and the target are backing file
1174 * In case 1 and 2, neither the source nor the target is the backing file.
1175 * In case 3, we will block the top BDS, so there is only one block job
1176 * for the top BDS and its backing chain.
1178 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_BACKUP_SOURCE
,
1179 parent
->backing_blocker
);
1180 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_BACKUP_TARGET
,
1181 parent
->backing_blocker
);
1184 static void bdrv_backing_detach(BdrvChild
*c
)
1186 BlockDriverState
*parent
= c
->opaque
;
1188 assert(parent
->backing_blocker
);
1189 bdrv_op_unblock_all(c
->bs
, parent
->backing_blocker
);
1190 error_free(parent
->backing_blocker
);
1191 parent
->backing_blocker
= NULL
;
1194 static int bdrv_backing_update_filename(BdrvChild
*c
, BlockDriverState
*base
,
1195 const char *filename
, Error
**errp
)
1197 BlockDriverState
*parent
= c
->opaque
;
1198 bool read_only
= bdrv_is_read_only(parent
);
1202 ret
= bdrv_reopen_set_read_only(parent
, false, errp
);
1208 ret
= bdrv_change_backing_file(parent
, filename
,
1209 base
->drv
? base
->drv
->format_name
: "");
1211 error_setg_errno(errp
, -ret
, "Could not update backing file link");
1215 bdrv_reopen_set_read_only(parent
, true, NULL
);
1222 * Returns the options and flags that a generic child of a BDS should
1223 * get, based on the given options and flags for the parent BDS.
1225 static void bdrv_inherited_options(BdrvChildRole role
, bool parent_is_format
,
1226 int *child_flags
, QDict
*child_options
,
1227 int parent_flags
, QDict
*parent_options
)
1229 int flags
= parent_flags
;
1232 * First, decide whether to set, clear, or leave BDRV_O_PROTOCOL.
1233 * Generally, the question to answer is: Should this child be
1234 * format-probed by default?
1238 * Pure and non-filtered data children of non-format nodes should
1239 * be probed by default (even when the node itself has BDRV_O_PROTOCOL
1240 * set). This only affects a very limited set of drivers (namely
1241 * quorum and blkverify when this comment was written).
1242 * Force-clear BDRV_O_PROTOCOL then.
1244 if (!parent_is_format
&&
1245 (role
& BDRV_CHILD_DATA
) &&
1246 !(role
& (BDRV_CHILD_METADATA
| BDRV_CHILD_FILTERED
)))
1248 flags
&= ~BDRV_O_PROTOCOL
;
1252 * All children of format nodes (except for COW children) and all
1253 * metadata children in general should never be format-probed.
1254 * Force-set BDRV_O_PROTOCOL then.
1256 if ((parent_is_format
&& !(role
& BDRV_CHILD_COW
)) ||
1257 (role
& BDRV_CHILD_METADATA
))
1259 flags
|= BDRV_O_PROTOCOL
;
1263 * If the cache mode isn't explicitly set, inherit direct and no-flush from
1266 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_CACHE_DIRECT
);
1267 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_CACHE_NO_FLUSH
);
1268 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_FORCE_SHARE
);
1270 if (role
& BDRV_CHILD_COW
) {
1271 /* backing files are opened read-only by default */
1272 qdict_set_default_str(child_options
, BDRV_OPT_READ_ONLY
, "on");
1273 qdict_set_default_str(child_options
, BDRV_OPT_AUTO_READ_ONLY
, "off");
1275 /* Inherit the read-only option from the parent if it's not set */
1276 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_READ_ONLY
);
1277 qdict_copy_default(child_options
, parent_options
,
1278 BDRV_OPT_AUTO_READ_ONLY
);
1282 * bdrv_co_pdiscard() respects unmap policy for the parent, so we
1283 * can default to enable it on lower layers regardless of the
1286 qdict_set_default_str(child_options
, BDRV_OPT_DISCARD
, "unmap");
1288 /* Clear flags that only apply to the top layer */
1289 flags
&= ~(BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
| BDRV_O_COPY_ON_READ
);
1291 if (role
& BDRV_CHILD_METADATA
) {
1292 flags
&= ~BDRV_O_NO_IO
;
1294 if (role
& BDRV_CHILD_COW
) {
1295 flags
&= ~BDRV_O_TEMPORARY
;
1298 *child_flags
= flags
;
1301 static void bdrv_child_cb_attach(BdrvChild
*child
)
1303 BlockDriverState
*bs
= child
->opaque
;
1305 if (child
->role
& BDRV_CHILD_COW
) {
1306 bdrv_backing_attach(child
);
1309 bdrv_apply_subtree_drain(child
, bs
);
1312 static void bdrv_child_cb_detach(BdrvChild
*child
)
1314 BlockDriverState
*bs
= child
->opaque
;
1316 if (child
->role
& BDRV_CHILD_COW
) {
1317 bdrv_backing_detach(child
);
1320 bdrv_unapply_subtree_drain(child
, bs
);
1323 static int bdrv_child_cb_update_filename(BdrvChild
*c
, BlockDriverState
*base
,
1324 const char *filename
, Error
**errp
)
1326 if (c
->role
& BDRV_CHILD_COW
) {
1327 return bdrv_backing_update_filename(c
, base
, filename
, errp
);
1332 const BdrvChildClass child_of_bds
= {
1333 .parent_is_bds
= true,
1334 .get_parent_desc
= bdrv_child_get_parent_desc
,
1335 .inherit_options
= bdrv_inherited_options
,
1336 .drained_begin
= bdrv_child_cb_drained_begin
,
1337 .drained_poll
= bdrv_child_cb_drained_poll
,
1338 .drained_end
= bdrv_child_cb_drained_end
,
1339 .attach
= bdrv_child_cb_attach
,
1340 .detach
= bdrv_child_cb_detach
,
1341 .inactivate
= bdrv_child_cb_inactivate
,
1342 .can_set_aio_ctx
= bdrv_child_cb_can_set_aio_ctx
,
1343 .set_aio_ctx
= bdrv_child_cb_set_aio_ctx
,
1344 .update_filename
= bdrv_child_cb_update_filename
,
1347 static int bdrv_open_flags(BlockDriverState
*bs
, int flags
)
1349 int open_flags
= flags
;
1352 * Clear flags that are internal to the block layer before opening the
1355 open_flags
&= ~(BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
| BDRV_O_PROTOCOL
);
1360 static void update_flags_from_options(int *flags
, QemuOpts
*opts
)
1362 *flags
&= ~(BDRV_O_CACHE_MASK
| BDRV_O_RDWR
| BDRV_O_AUTO_RDONLY
);
1364 if (qemu_opt_get_bool_del(opts
, BDRV_OPT_CACHE_NO_FLUSH
, false)) {
1365 *flags
|= BDRV_O_NO_FLUSH
;
1368 if (qemu_opt_get_bool_del(opts
, BDRV_OPT_CACHE_DIRECT
, false)) {
1369 *flags
|= BDRV_O_NOCACHE
;
1372 if (!qemu_opt_get_bool_del(opts
, BDRV_OPT_READ_ONLY
, false)) {
1373 *flags
|= BDRV_O_RDWR
;
1376 if (qemu_opt_get_bool_del(opts
, BDRV_OPT_AUTO_READ_ONLY
, false)) {
1377 *flags
|= BDRV_O_AUTO_RDONLY
;
1381 static void update_options_from_flags(QDict
*options
, int flags
)
1383 if (!qdict_haskey(options
, BDRV_OPT_CACHE_DIRECT
)) {
1384 qdict_put_bool(options
, BDRV_OPT_CACHE_DIRECT
, flags
& BDRV_O_NOCACHE
);
1386 if (!qdict_haskey(options
, BDRV_OPT_CACHE_NO_FLUSH
)) {
1387 qdict_put_bool(options
, BDRV_OPT_CACHE_NO_FLUSH
,
1388 flags
& BDRV_O_NO_FLUSH
);
1390 if (!qdict_haskey(options
, BDRV_OPT_READ_ONLY
)) {
1391 qdict_put_bool(options
, BDRV_OPT_READ_ONLY
, !(flags
& BDRV_O_RDWR
));
1393 if (!qdict_haskey(options
, BDRV_OPT_AUTO_READ_ONLY
)) {
1394 qdict_put_bool(options
, BDRV_OPT_AUTO_READ_ONLY
,
1395 flags
& BDRV_O_AUTO_RDONLY
);
1399 static void bdrv_assign_node_name(BlockDriverState
*bs
,
1400 const char *node_name
,
1403 char *gen_node_name
= NULL
;
1406 node_name
= gen_node_name
= id_generate(ID_BLOCK
);
1407 } else if (!id_wellformed(node_name
)) {
1409 * Check for empty string or invalid characters, but not if it is
1410 * generated (generated names use characters not available to the user)
1412 error_setg(errp
, "Invalid node name");
1416 /* takes care of avoiding namespaces collisions */
1417 if (blk_by_name(node_name
)) {
1418 error_setg(errp
, "node-name=%s is conflicting with a device id",
1423 /* takes care of avoiding duplicates node names */
1424 if (bdrv_find_node(node_name
)) {
1425 error_setg(errp
, "Duplicate node name");
1429 /* Make sure that the node name isn't truncated */
1430 if (strlen(node_name
) >= sizeof(bs
->node_name
)) {
1431 error_setg(errp
, "Node name too long");
1435 /* copy node name into the bs and insert it into the graph list */
1436 pstrcpy(bs
->node_name
, sizeof(bs
->node_name
), node_name
);
1437 QTAILQ_INSERT_TAIL(&graph_bdrv_states
, bs
, node_list
);
1439 g_free(gen_node_name
);
1442 static int bdrv_open_driver(BlockDriverState
*bs
, BlockDriver
*drv
,
1443 const char *node_name
, QDict
*options
,
1444 int open_flags
, Error
**errp
)
1446 Error
*local_err
= NULL
;
1449 bdrv_assign_node_name(bs
, node_name
, &local_err
);
1451 error_propagate(errp
, local_err
);
1456 bs
->read_only
= !(bs
->open_flags
& BDRV_O_RDWR
);
1457 bs
->opaque
= g_malloc0(drv
->instance_size
);
1459 if (drv
->bdrv_file_open
) {
1460 assert(!drv
->bdrv_needs_filename
|| bs
->filename
[0]);
1461 ret
= drv
->bdrv_file_open(bs
, options
, open_flags
, &local_err
);
1462 } else if (drv
->bdrv_open
) {
1463 ret
= drv
->bdrv_open(bs
, options
, open_flags
, &local_err
);
1470 error_propagate(errp
, local_err
);
1471 } else if (bs
->filename
[0]) {
1472 error_setg_errno(errp
, -ret
, "Could not open '%s'", bs
->filename
);
1474 error_setg_errno(errp
, -ret
, "Could not open image");
1479 ret
= refresh_total_sectors(bs
, bs
->total_sectors
);
1481 error_setg_errno(errp
, -ret
, "Could not refresh total sector count");
1485 bdrv_refresh_limits(bs
, &local_err
);
1487 error_propagate(errp
, local_err
);
1491 assert(bdrv_opt_mem_align(bs
) != 0);
1492 assert(bdrv_min_mem_align(bs
) != 0);
1493 assert(is_power_of_2(bs
->bl
.request_alignment
));
1495 for (i
= 0; i
< bs
->quiesce_counter
; i
++) {
1496 if (drv
->bdrv_co_drain_begin
) {
1497 drv
->bdrv_co_drain_begin(bs
);
1504 if (bs
->file
!= NULL
) {
1505 bdrv_unref_child(bs
, bs
->file
);
1513 BlockDriverState
*bdrv_new_open_driver(BlockDriver
*drv
, const char *node_name
,
1514 int flags
, Error
**errp
)
1516 BlockDriverState
*bs
;
1520 bs
->open_flags
= flags
;
1521 bs
->explicit_options
= qdict_new();
1522 bs
->options
= qdict_new();
1525 update_options_from_flags(bs
->options
, flags
);
1527 ret
= bdrv_open_driver(bs
, drv
, node_name
, bs
->options
, flags
, errp
);
1529 qobject_unref(bs
->explicit_options
);
1530 bs
->explicit_options
= NULL
;
1531 qobject_unref(bs
->options
);
1540 QemuOptsList bdrv_runtime_opts
= {
1541 .name
= "bdrv_common",
1542 .head
= QTAILQ_HEAD_INITIALIZER(bdrv_runtime_opts
.head
),
1545 .name
= "node-name",
1546 .type
= QEMU_OPT_STRING
,
1547 .help
= "Node name of the block device node",
1551 .type
= QEMU_OPT_STRING
,
1552 .help
= "Block driver to use for the node",
1555 .name
= BDRV_OPT_CACHE_DIRECT
,
1556 .type
= QEMU_OPT_BOOL
,
1557 .help
= "Bypass software writeback cache on the host",
1560 .name
= BDRV_OPT_CACHE_NO_FLUSH
,
1561 .type
= QEMU_OPT_BOOL
,
1562 .help
= "Ignore flush requests",
1565 .name
= BDRV_OPT_READ_ONLY
,
1566 .type
= QEMU_OPT_BOOL
,
1567 .help
= "Node is opened in read-only mode",
1570 .name
= BDRV_OPT_AUTO_READ_ONLY
,
1571 .type
= QEMU_OPT_BOOL
,
1572 .help
= "Node can become read-only if opening read-write fails",
1575 .name
= "detect-zeroes",
1576 .type
= QEMU_OPT_STRING
,
1577 .help
= "try to optimize zero writes (off, on, unmap)",
1580 .name
= BDRV_OPT_DISCARD
,
1581 .type
= QEMU_OPT_STRING
,
1582 .help
= "discard operation (ignore/off, unmap/on)",
1585 .name
= BDRV_OPT_FORCE_SHARE
,
1586 .type
= QEMU_OPT_BOOL
,
1587 .help
= "always accept other writers (default: off)",
1589 { /* end of list */ }
1593 QemuOptsList bdrv_create_opts_simple
= {
1594 .name
= "simple-create-opts",
1595 .head
= QTAILQ_HEAD_INITIALIZER(bdrv_create_opts_simple
.head
),
1598 .name
= BLOCK_OPT_SIZE
,
1599 .type
= QEMU_OPT_SIZE
,
1600 .help
= "Virtual disk size"
1603 .name
= BLOCK_OPT_PREALLOC
,
1604 .type
= QEMU_OPT_STRING
,
1605 .help
= "Preallocation mode (allowed values: off)"
1607 { /* end of list */ }
1612 * Common part for opening disk images and files
1614 * Removes all processed options from *options.
1616 static int bdrv_open_common(BlockDriverState
*bs
, BlockBackend
*file
,
1617 QDict
*options
, Error
**errp
)
1619 int ret
, open_flags
;
1620 const char *filename
;
1621 const char *driver_name
= NULL
;
1622 const char *node_name
= NULL
;
1623 const char *discard
;
1626 Error
*local_err
= NULL
;
1628 assert(bs
->file
== NULL
);
1629 assert(options
!= NULL
&& bs
->options
!= options
);
1631 opts
= qemu_opts_create(&bdrv_runtime_opts
, NULL
, 0, &error_abort
);
1632 qemu_opts_absorb_qdict(opts
, options
, &local_err
);
1634 error_propagate(errp
, local_err
);
1639 update_flags_from_options(&bs
->open_flags
, opts
);
1641 driver_name
= qemu_opt_get(opts
, "driver");
1642 drv
= bdrv_find_format(driver_name
);
1643 assert(drv
!= NULL
);
1645 bs
->force_share
= qemu_opt_get_bool(opts
, BDRV_OPT_FORCE_SHARE
, false);
1647 if (bs
->force_share
&& (bs
->open_flags
& BDRV_O_RDWR
)) {
1649 BDRV_OPT_FORCE_SHARE
1650 "=on can only be used with read-only images");
1656 bdrv_refresh_filename(blk_bs(file
));
1657 filename
= blk_bs(file
)->filename
;
1660 * Caution: while qdict_get_try_str() is fine, getting
1661 * non-string types would require more care. When @options
1662 * come from -blockdev or blockdev_add, its members are typed
1663 * according to the QAPI schema, but when they come from
1664 * -drive, they're all QString.
1666 filename
= qdict_get_try_str(options
, "filename");
1669 if (drv
->bdrv_needs_filename
&& (!filename
|| !filename
[0])) {
1670 error_setg(errp
, "The '%s' block driver requires a file name",
1676 trace_bdrv_open_common(bs
, filename
?: "", bs
->open_flags
,
1679 bs
->read_only
= !(bs
->open_flags
& BDRV_O_RDWR
);
1681 if (use_bdrv_whitelist
&& !bdrv_is_whitelisted(drv
, bs
->read_only
)) {
1682 if (!bs
->read_only
&& bdrv_is_whitelisted(drv
, true)) {
1683 ret
= bdrv_apply_auto_read_only(bs
, NULL
, NULL
);
1689 !bs
->read_only
&& bdrv_is_whitelisted(drv
, true)
1690 ? "Driver '%s' can only be used for read-only devices"
1691 : "Driver '%s' is not whitelisted",
1697 /* bdrv_new() and bdrv_close() make it so */
1698 assert(atomic_read(&bs
->copy_on_read
) == 0);
1700 if (bs
->open_flags
& BDRV_O_COPY_ON_READ
) {
1701 if (!bs
->read_only
) {
1702 bdrv_enable_copy_on_read(bs
);
1704 error_setg(errp
, "Can't use copy-on-read on read-only device");
1710 discard
= qemu_opt_get(opts
, BDRV_OPT_DISCARD
);
1711 if (discard
!= NULL
) {
1712 if (bdrv_parse_discard_flags(discard
, &bs
->open_flags
) != 0) {
1713 error_setg(errp
, "Invalid discard option");
1720 bdrv_parse_detect_zeroes(opts
, bs
->open_flags
, &local_err
);
1722 error_propagate(errp
, local_err
);
1727 if (filename
!= NULL
) {
1728 pstrcpy(bs
->filename
, sizeof(bs
->filename
), filename
);
1730 bs
->filename
[0] = '\0';
1732 pstrcpy(bs
->exact_filename
, sizeof(bs
->exact_filename
), bs
->filename
);
1734 /* Open the image, either directly or using a protocol */
1735 open_flags
= bdrv_open_flags(bs
, bs
->open_flags
);
1736 node_name
= qemu_opt_get(opts
, "node-name");
1738 assert(!drv
->bdrv_file_open
|| file
== NULL
);
1739 ret
= bdrv_open_driver(bs
, drv
, node_name
, options
, open_flags
, errp
);
1744 qemu_opts_del(opts
);
1748 qemu_opts_del(opts
);
1752 static QDict
*parse_json_filename(const char *filename
, Error
**errp
)
1754 QObject
*options_obj
;
1758 ret
= strstart(filename
, "json:", &filename
);
1761 options_obj
= qobject_from_json(filename
, errp
);
1763 error_prepend(errp
, "Could not parse the JSON options: ");
1767 options
= qobject_to(QDict
, options_obj
);
1769 qobject_unref(options_obj
);
1770 error_setg(errp
, "Invalid JSON object given");
1774 qdict_flatten(options
);
1779 static void parse_json_protocol(QDict
*options
, const char **pfilename
,
1782 QDict
*json_options
;
1783 Error
*local_err
= NULL
;
1785 /* Parse json: pseudo-protocol */
1786 if (!*pfilename
|| !g_str_has_prefix(*pfilename
, "json:")) {
1790 json_options
= parse_json_filename(*pfilename
, &local_err
);
1792 error_propagate(errp
, local_err
);
1796 /* Options given in the filename have lower priority than options
1797 * specified directly */
1798 qdict_join(options
, json_options
, false);
1799 qobject_unref(json_options
);
1804 * Fills in default options for opening images and converts the legacy
1805 * filename/flags pair to option QDict entries.
1806 * The BDRV_O_PROTOCOL flag in *flags will be set or cleared accordingly if a
1807 * block driver has been specified explicitly.
1809 static int bdrv_fill_options(QDict
**options
, const char *filename
,
1810 int *flags
, Error
**errp
)
1812 const char *drvname
;
1813 bool protocol
= *flags
& BDRV_O_PROTOCOL
;
1814 bool parse_filename
= false;
1815 BlockDriver
*drv
= NULL
;
1816 Error
*local_err
= NULL
;
1819 * Caution: while qdict_get_try_str() is fine, getting non-string
1820 * types would require more care. When @options come from
1821 * -blockdev or blockdev_add, its members are typed according to
1822 * the QAPI schema, but when they come from -drive, they're all
1825 drvname
= qdict_get_try_str(*options
, "driver");
1827 drv
= bdrv_find_format(drvname
);
1829 error_setg(errp
, "Unknown driver '%s'", drvname
);
1832 /* If the user has explicitly specified the driver, this choice should
1833 * override the BDRV_O_PROTOCOL flag */
1834 protocol
= drv
->bdrv_file_open
;
1838 *flags
|= BDRV_O_PROTOCOL
;
1840 *flags
&= ~BDRV_O_PROTOCOL
;
1843 /* Translate cache options from flags into options */
1844 update_options_from_flags(*options
, *flags
);
1846 /* Fetch the file name from the options QDict if necessary */
1847 if (protocol
&& filename
) {
1848 if (!qdict_haskey(*options
, "filename")) {
1849 qdict_put_str(*options
, "filename", filename
);
1850 parse_filename
= true;
1852 error_setg(errp
, "Can't specify 'file' and 'filename' options at "
1858 /* Find the right block driver */
1859 /* See cautionary note on accessing @options above */
1860 filename
= qdict_get_try_str(*options
, "filename");
1862 if (!drvname
&& protocol
) {
1864 drv
= bdrv_find_protocol(filename
, parse_filename
, errp
);
1869 drvname
= drv
->format_name
;
1870 qdict_put_str(*options
, "driver", drvname
);
1872 error_setg(errp
, "Must specify either driver or file");
1877 assert(drv
|| !protocol
);
1879 /* Driver-specific filename parsing */
1880 if (drv
&& drv
->bdrv_parse_filename
&& parse_filename
) {
1881 drv
->bdrv_parse_filename(filename
, *options
, &local_err
);
1883 error_propagate(errp
, local_err
);
1887 if (!drv
->bdrv_needs_filename
) {
1888 qdict_del(*options
, "filename");
1895 static int bdrv_child_check_perm(BdrvChild
*c
, BlockReopenQueue
*q
,
1896 uint64_t perm
, uint64_t shared
,
1897 GSList
*ignore_children
,
1898 bool *tighten_restrictions
, Error
**errp
);
1899 static void bdrv_child_abort_perm_update(BdrvChild
*c
);
1900 static void bdrv_child_set_perm(BdrvChild
*c
, uint64_t perm
, uint64_t shared
);
1902 typedef struct BlockReopenQueueEntry
{
1905 BDRVReopenState state
;
1906 QTAILQ_ENTRY(BlockReopenQueueEntry
) entry
;
1907 } BlockReopenQueueEntry
;
1910 * Return the flags that @bs will have after the reopens in @q have
1911 * successfully completed. If @q is NULL (or @bs is not contained in @q),
1912 * return the current flags.
1914 static int bdrv_reopen_get_flags(BlockReopenQueue
*q
, BlockDriverState
*bs
)
1916 BlockReopenQueueEntry
*entry
;
1919 QTAILQ_FOREACH(entry
, q
, entry
) {
1920 if (entry
->state
.bs
== bs
) {
1921 return entry
->state
.flags
;
1926 return bs
->open_flags
;
1929 /* Returns whether the image file can be written to after the reopen queue @q
1930 * has been successfully applied, or right now if @q is NULL. */
1931 static bool bdrv_is_writable_after_reopen(BlockDriverState
*bs
,
1932 BlockReopenQueue
*q
)
1934 int flags
= bdrv_reopen_get_flags(q
, bs
);
1936 return (flags
& (BDRV_O_RDWR
| BDRV_O_INACTIVE
)) == BDRV_O_RDWR
;
1940 * Return whether the BDS can be written to. This is not necessarily
1941 * the same as !bdrv_is_read_only(bs), as inactivated images may not
1942 * be written to but do not count as read-only images.
1944 bool bdrv_is_writable(BlockDriverState
*bs
)
1946 return bdrv_is_writable_after_reopen(bs
, NULL
);
1949 static void bdrv_child_perm(BlockDriverState
*bs
, BlockDriverState
*child_bs
,
1950 BdrvChild
*c
, const BdrvChildClass
*child_class
,
1951 BdrvChildRole role
, BlockReopenQueue
*reopen_queue
,
1952 uint64_t parent_perm
, uint64_t parent_shared
,
1953 uint64_t *nperm
, uint64_t *nshared
)
1955 assert(bs
->drv
&& bs
->drv
->bdrv_child_perm
);
1956 bs
->drv
->bdrv_child_perm(bs
, c
, child_class
, role
, reopen_queue
,
1957 parent_perm
, parent_shared
,
1959 /* TODO Take force_share from reopen_queue */
1960 if (child_bs
&& child_bs
->force_share
) {
1961 *nshared
= BLK_PERM_ALL
;
1966 * Check whether permissions on this node can be changed in a way that
1967 * @cumulative_perms and @cumulative_shared_perms are the new cumulative
1968 * permissions of all its parents. This involves checking whether all necessary
1969 * permission changes to child nodes can be performed.
1971 * Will set *tighten_restrictions to true if and only if new permissions have to
1972 * be taken or currently shared permissions are to be unshared. Otherwise,
1973 * errors are not fatal as long as the caller accepts that the restrictions
1974 * remain tighter than they need to be. The caller still has to abort the
1976 * @tighten_restrictions cannot be used together with @q: When reopening, we may
1977 * encounter fatal errors even though no restrictions are to be tightened. For
1978 * example, changing a node from RW to RO will fail if the WRITE permission is
1981 * A call to this function must always be followed by a call to bdrv_set_perm()
1982 * or bdrv_abort_perm_update().
1984 static int bdrv_check_perm(BlockDriverState
*bs
, BlockReopenQueue
*q
,
1985 uint64_t cumulative_perms
,
1986 uint64_t cumulative_shared_perms
,
1987 GSList
*ignore_children
,
1988 bool *tighten_restrictions
, Error
**errp
)
1990 BlockDriver
*drv
= bs
->drv
;
1994 assert(!q
|| !tighten_restrictions
);
1996 if (tighten_restrictions
) {
1997 uint64_t current_perms
, current_shared
;
1998 uint64_t added_perms
, removed_shared_perms
;
2000 bdrv_get_cumulative_perm(bs
, ¤t_perms
, ¤t_shared
);
2002 added_perms
= cumulative_perms
& ~current_perms
;
2003 removed_shared_perms
= current_shared
& ~cumulative_shared_perms
;
2005 *tighten_restrictions
= added_perms
|| removed_shared_perms
;
2008 /* Write permissions never work with read-only images */
2009 if ((cumulative_perms
& (BLK_PERM_WRITE
| BLK_PERM_WRITE_UNCHANGED
)) &&
2010 !bdrv_is_writable_after_reopen(bs
, q
))
2012 if (!bdrv_is_writable_after_reopen(bs
, NULL
)) {
2013 error_setg(errp
, "Block node is read-only");
2015 uint64_t current_perms
, current_shared
;
2016 bdrv_get_cumulative_perm(bs
, ¤t_perms
, ¤t_shared
);
2017 if (current_perms
& (BLK_PERM_WRITE
| BLK_PERM_WRITE_UNCHANGED
)) {
2018 error_setg(errp
, "Cannot make block node read-only, there is "
2021 error_setg(errp
, "Cannot make block node read-only and create "
2029 /* Check this node */
2034 if (drv
->bdrv_check_perm
) {
2035 return drv
->bdrv_check_perm(bs
, cumulative_perms
,
2036 cumulative_shared_perms
, errp
);
2039 /* Drivers that never have children can omit .bdrv_child_perm() */
2040 if (!drv
->bdrv_child_perm
) {
2041 assert(QLIST_EMPTY(&bs
->children
));
2045 /* Check all children */
2046 QLIST_FOREACH(c
, &bs
->children
, next
) {
2047 uint64_t cur_perm
, cur_shared
;
2048 bool child_tighten_restr
;
2050 bdrv_child_perm(bs
, c
->bs
, c
, c
->klass
, c
->role
, q
,
2051 cumulative_perms
, cumulative_shared_perms
,
2052 &cur_perm
, &cur_shared
);
2053 ret
= bdrv_child_check_perm(c
, q
, cur_perm
, cur_shared
, ignore_children
,
2054 tighten_restrictions
? &child_tighten_restr
2057 if (tighten_restrictions
) {
2058 *tighten_restrictions
|= child_tighten_restr
;
2069 * Notifies drivers that after a previous bdrv_check_perm() call, the
2070 * permission update is not performed and any preparations made for it (e.g.
2071 * taken file locks) need to be undone.
2073 * This function recursively notifies all child nodes.
2075 static void bdrv_abort_perm_update(BlockDriverState
*bs
)
2077 BlockDriver
*drv
= bs
->drv
;
2084 if (drv
->bdrv_abort_perm_update
) {
2085 drv
->bdrv_abort_perm_update(bs
);
2088 QLIST_FOREACH(c
, &bs
->children
, next
) {
2089 bdrv_child_abort_perm_update(c
);
2093 static void bdrv_set_perm(BlockDriverState
*bs
, uint64_t cumulative_perms
,
2094 uint64_t cumulative_shared_perms
)
2096 BlockDriver
*drv
= bs
->drv
;
2103 /* Update this node */
2104 if (drv
->bdrv_set_perm
) {
2105 drv
->bdrv_set_perm(bs
, cumulative_perms
, cumulative_shared_perms
);
2108 /* Drivers that never have children can omit .bdrv_child_perm() */
2109 if (!drv
->bdrv_child_perm
) {
2110 assert(QLIST_EMPTY(&bs
->children
));
2114 /* Update all children */
2115 QLIST_FOREACH(c
, &bs
->children
, next
) {
2116 uint64_t cur_perm
, cur_shared
;
2117 bdrv_child_perm(bs
, c
->bs
, c
, c
->klass
, c
->role
, NULL
,
2118 cumulative_perms
, cumulative_shared_perms
,
2119 &cur_perm
, &cur_shared
);
2120 bdrv_child_set_perm(c
, cur_perm
, cur_shared
);
2124 void bdrv_get_cumulative_perm(BlockDriverState
*bs
, uint64_t *perm
,
2125 uint64_t *shared_perm
)
2128 uint64_t cumulative_perms
= 0;
2129 uint64_t cumulative_shared_perms
= BLK_PERM_ALL
;
2131 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
2132 cumulative_perms
|= c
->perm
;
2133 cumulative_shared_perms
&= c
->shared_perm
;
2136 *perm
= cumulative_perms
;
2137 *shared_perm
= cumulative_shared_perms
;
2140 static char *bdrv_child_user_desc(BdrvChild
*c
)
2142 if (c
->klass
->get_parent_desc
) {
2143 return c
->klass
->get_parent_desc(c
);
2146 return g_strdup("another user");
2149 char *bdrv_perm_names(uint64_t perm
)
2155 { BLK_PERM_CONSISTENT_READ
, "consistent read" },
2156 { BLK_PERM_WRITE
, "write" },
2157 { BLK_PERM_WRITE_UNCHANGED
, "write unchanged" },
2158 { BLK_PERM_RESIZE
, "resize" },
2159 { BLK_PERM_GRAPH_MOD
, "change children" },
2163 GString
*result
= g_string_sized_new(30);
2164 struct perm_name
*p
;
2166 for (p
= permissions
; p
->name
; p
++) {
2167 if (perm
& p
->perm
) {
2168 if (result
->len
> 0) {
2169 g_string_append(result
, ", ");
2171 g_string_append(result
, p
->name
);
2175 return g_string_free(result
, FALSE
);
2179 * Checks whether a new reference to @bs can be added if the new user requires
2180 * @new_used_perm/@new_shared_perm as its permissions. If @ignore_children is
2181 * set, the BdrvChild objects in this list are ignored in the calculations;
2182 * this allows checking permission updates for an existing reference.
2184 * See bdrv_check_perm() for the semantics of @tighten_restrictions.
2186 * Needs to be followed by a call to either bdrv_set_perm() or
2187 * bdrv_abort_perm_update(). */
2188 static int bdrv_check_update_perm(BlockDriverState
*bs
, BlockReopenQueue
*q
,
2189 uint64_t new_used_perm
,
2190 uint64_t new_shared_perm
,
2191 GSList
*ignore_children
,
2192 bool *tighten_restrictions
,
2196 uint64_t cumulative_perms
= new_used_perm
;
2197 uint64_t cumulative_shared_perms
= new_shared_perm
;
2199 assert(!q
|| !tighten_restrictions
);
2201 /* There is no reason why anyone couldn't tolerate write_unchanged */
2202 assert(new_shared_perm
& BLK_PERM_WRITE_UNCHANGED
);
2204 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
2205 if (g_slist_find(ignore_children
, c
)) {
2209 if ((new_used_perm
& c
->shared_perm
) != new_used_perm
) {
2210 char *user
= bdrv_child_user_desc(c
);
2211 char *perm_names
= bdrv_perm_names(new_used_perm
& ~c
->shared_perm
);
2213 if (tighten_restrictions
) {
2214 *tighten_restrictions
= true;
2217 error_setg(errp
, "Conflicts with use by %s as '%s', which does not "
2219 user
, c
->name
, perm_names
, bdrv_get_node_name(c
->bs
));
2225 if ((c
->perm
& new_shared_perm
) != c
->perm
) {
2226 char *user
= bdrv_child_user_desc(c
);
2227 char *perm_names
= bdrv_perm_names(c
->perm
& ~new_shared_perm
);
2229 if (tighten_restrictions
) {
2230 *tighten_restrictions
= true;
2233 error_setg(errp
, "Conflicts with use by %s as '%s', which uses "
2235 user
, c
->name
, perm_names
, bdrv_get_node_name(c
->bs
));
2241 cumulative_perms
|= c
->perm
;
2242 cumulative_shared_perms
&= c
->shared_perm
;
2245 return bdrv_check_perm(bs
, q
, cumulative_perms
, cumulative_shared_perms
,
2246 ignore_children
, tighten_restrictions
, errp
);
2249 /* Needs to be followed by a call to either bdrv_child_set_perm() or
2250 * bdrv_child_abort_perm_update(). */
2251 static int bdrv_child_check_perm(BdrvChild
*c
, BlockReopenQueue
*q
,
2252 uint64_t perm
, uint64_t shared
,
2253 GSList
*ignore_children
,
2254 bool *tighten_restrictions
, Error
**errp
)
2258 ignore_children
= g_slist_prepend(g_slist_copy(ignore_children
), c
);
2259 ret
= bdrv_check_update_perm(c
->bs
, q
, perm
, shared
, ignore_children
,
2260 tighten_restrictions
, errp
);
2261 g_slist_free(ignore_children
);
2267 if (!c
->has_backup_perm
) {
2268 c
->has_backup_perm
= true;
2269 c
->backup_perm
= c
->perm
;
2270 c
->backup_shared_perm
= c
->shared_perm
;
2273 * Note: it's OK if c->has_backup_perm was already set, as we can find the
2274 * same child twice during check_perm procedure
2278 c
->shared_perm
= shared
;
2283 static void bdrv_child_set_perm(BdrvChild
*c
, uint64_t perm
, uint64_t shared
)
2285 uint64_t cumulative_perms
, cumulative_shared_perms
;
2287 c
->has_backup_perm
= false;
2290 c
->shared_perm
= shared
;
2292 bdrv_get_cumulative_perm(c
->bs
, &cumulative_perms
,
2293 &cumulative_shared_perms
);
2294 bdrv_set_perm(c
->bs
, cumulative_perms
, cumulative_shared_perms
);
2297 static void bdrv_child_abort_perm_update(BdrvChild
*c
)
2299 if (c
->has_backup_perm
) {
2300 c
->perm
= c
->backup_perm
;
2301 c
->shared_perm
= c
->backup_shared_perm
;
2302 c
->has_backup_perm
= false;
2305 bdrv_abort_perm_update(c
->bs
);
2308 int bdrv_child_try_set_perm(BdrvChild
*c
, uint64_t perm
, uint64_t shared
,
2311 Error
*local_err
= NULL
;
2313 bool tighten_restrictions
;
2315 ret
= bdrv_child_check_perm(c
, NULL
, perm
, shared
, NULL
,
2316 &tighten_restrictions
, &local_err
);
2318 bdrv_child_abort_perm_update(c
);
2319 if (tighten_restrictions
) {
2320 error_propagate(errp
, local_err
);
2323 * Our caller may intend to only loosen restrictions and
2324 * does not expect this function to fail. Errors are not
2325 * fatal in such a case, so we can just hide them from our
2328 error_free(local_err
);
2334 bdrv_child_set_perm(c
, perm
, shared
);
2339 int bdrv_child_refresh_perms(BlockDriverState
*bs
, BdrvChild
*c
, Error
**errp
)
2341 uint64_t parent_perms
, parent_shared
;
2342 uint64_t perms
, shared
;
2344 bdrv_get_cumulative_perm(bs
, &parent_perms
, &parent_shared
);
2345 bdrv_child_perm(bs
, c
->bs
, c
, c
->klass
, c
->role
, NULL
,
2346 parent_perms
, parent_shared
, &perms
, &shared
);
2348 return bdrv_child_try_set_perm(c
, perms
, shared
, errp
);
2352 * Default implementation for .bdrv_child_perm() for block filters:
2353 * Forward CONSISTENT_READ, WRITE, WRITE_UNCHANGED, and RESIZE to the
2356 static void bdrv_filter_default_perms(BlockDriverState
*bs
, BdrvChild
*c
,
2357 const BdrvChildClass
*child_class
,
2359 BlockReopenQueue
*reopen_queue
,
2360 uint64_t perm
, uint64_t shared
,
2361 uint64_t *nperm
, uint64_t *nshared
)
2363 *nperm
= perm
& DEFAULT_PERM_PASSTHROUGH
;
2364 *nshared
= (shared
& DEFAULT_PERM_PASSTHROUGH
) | DEFAULT_PERM_UNCHANGED
;
2367 static void bdrv_default_perms_for_cow(BlockDriverState
*bs
, BdrvChild
*c
,
2368 const BdrvChildClass
*child_class
,
2370 BlockReopenQueue
*reopen_queue
,
2371 uint64_t perm
, uint64_t shared
,
2372 uint64_t *nperm
, uint64_t *nshared
)
2374 assert(child_class
== &child_of_bds
&& (role
& BDRV_CHILD_COW
));
2377 * We want consistent read from backing files if the parent needs it.
2378 * No other operations are performed on backing files.
2380 perm
&= BLK_PERM_CONSISTENT_READ
;
2383 * If the parent can deal with changing data, we're okay with a
2384 * writable and resizable backing file.
2385 * TODO Require !(perm & BLK_PERM_CONSISTENT_READ), too?
2387 if (shared
& BLK_PERM_WRITE
) {
2388 shared
= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2393 shared
|= BLK_PERM_CONSISTENT_READ
| BLK_PERM_GRAPH_MOD
|
2394 BLK_PERM_WRITE_UNCHANGED
;
2396 if (bs
->open_flags
& BDRV_O_INACTIVE
) {
2397 shared
|= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2404 static void bdrv_default_perms_for_storage(BlockDriverState
*bs
, BdrvChild
*c
,
2405 const BdrvChildClass
*child_class
,
2407 BlockReopenQueue
*reopen_queue
,
2408 uint64_t perm
, uint64_t shared
,
2409 uint64_t *nperm
, uint64_t *nshared
)
2413 assert(child_class
== &child_of_bds
&&
2414 (role
& (BDRV_CHILD_METADATA
| BDRV_CHILD_DATA
)));
2416 flags
= bdrv_reopen_get_flags(reopen_queue
, bs
);
2419 * Apart from the modifications below, the same permissions are
2420 * forwarded and left alone as for filters
2422 bdrv_filter_default_perms(bs
, c
, child_class
, role
, reopen_queue
,
2423 perm
, shared
, &perm
, &shared
);
2425 if (role
& BDRV_CHILD_METADATA
) {
2426 /* Format drivers may touch metadata even if the guest doesn't write */
2427 if (bdrv_is_writable_after_reopen(bs
, reopen_queue
)) {
2428 perm
|= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2432 * bs->file always needs to be consistent because of the
2433 * metadata. We can never allow other users to resize or write
2436 if (!(flags
& BDRV_O_NO_IO
)) {
2437 perm
|= BLK_PERM_CONSISTENT_READ
;
2439 shared
&= ~(BLK_PERM_WRITE
| BLK_PERM_RESIZE
);
2442 if (role
& BDRV_CHILD_DATA
) {
2444 * Technically, everything in this block is a subset of the
2445 * BDRV_CHILD_METADATA path taken above, and so this could
2446 * be an "else if" branch. However, that is not obvious, and
2447 * this function is not performance critical, therefore we let
2448 * this be an independent "if".
2452 * We cannot allow other users to resize the file because the
2453 * format driver might have some assumptions about the size
2454 * (e.g. because it is stored in metadata, or because the file
2455 * is split into fixed-size data files).
2457 shared
&= ~BLK_PERM_RESIZE
;
2460 * WRITE_UNCHANGED often cannot be performed as such on the
2461 * data file. For example, the qcow2 driver may still need to
2462 * write copied clusters on copy-on-read.
2464 if (perm
& BLK_PERM_WRITE_UNCHANGED
) {
2465 perm
|= BLK_PERM_WRITE
;
2469 * If the data file is written to, the format driver may
2470 * expect to be able to resize it by writing beyond the EOF.
2472 if (perm
& BLK_PERM_WRITE
) {
2473 perm
|= BLK_PERM_RESIZE
;
2477 if (bs
->open_flags
& BDRV_O_INACTIVE
) {
2478 shared
|= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2485 void bdrv_default_perms(BlockDriverState
*bs
, BdrvChild
*c
,
2486 const BdrvChildClass
*child_class
, BdrvChildRole role
,
2487 BlockReopenQueue
*reopen_queue
,
2488 uint64_t perm
, uint64_t shared
,
2489 uint64_t *nperm
, uint64_t *nshared
)
2491 assert(child_class
== &child_of_bds
);
2493 if (role
& BDRV_CHILD_FILTERED
) {
2494 assert(!(role
& (BDRV_CHILD_DATA
| BDRV_CHILD_METADATA
|
2496 bdrv_filter_default_perms(bs
, c
, child_class
, role
, reopen_queue
,
2497 perm
, shared
, nperm
, nshared
);
2498 } else if (role
& BDRV_CHILD_COW
) {
2499 assert(!(role
& (BDRV_CHILD_DATA
| BDRV_CHILD_METADATA
)));
2500 bdrv_default_perms_for_cow(bs
, c
, child_class
, role
, reopen_queue
,
2501 perm
, shared
, nperm
, nshared
);
2502 } else if (role
& (BDRV_CHILD_METADATA
| BDRV_CHILD_DATA
)) {
2503 bdrv_default_perms_for_storage(bs
, c
, child_class
, role
, reopen_queue
,
2504 perm
, shared
, nperm
, nshared
);
2506 g_assert_not_reached();
2510 uint64_t bdrv_qapi_perm_to_blk_perm(BlockPermission qapi_perm
)
2512 static const uint64_t permissions
[] = {
2513 [BLOCK_PERMISSION_CONSISTENT_READ
] = BLK_PERM_CONSISTENT_READ
,
2514 [BLOCK_PERMISSION_WRITE
] = BLK_PERM_WRITE
,
2515 [BLOCK_PERMISSION_WRITE_UNCHANGED
] = BLK_PERM_WRITE_UNCHANGED
,
2516 [BLOCK_PERMISSION_RESIZE
] = BLK_PERM_RESIZE
,
2517 [BLOCK_PERMISSION_GRAPH_MOD
] = BLK_PERM_GRAPH_MOD
,
2520 QEMU_BUILD_BUG_ON(ARRAY_SIZE(permissions
) != BLOCK_PERMISSION__MAX
);
2521 QEMU_BUILD_BUG_ON(1UL << ARRAY_SIZE(permissions
) != BLK_PERM_ALL
+ 1);
2523 assert(qapi_perm
< BLOCK_PERMISSION__MAX
);
2525 return permissions
[qapi_perm
];
2528 static void bdrv_replace_child_noperm(BdrvChild
*child
,
2529 BlockDriverState
*new_bs
)
2531 BlockDriverState
*old_bs
= child
->bs
;
2532 int new_bs_quiesce_counter
;
2535 assert(!child
->frozen
);
2537 if (old_bs
&& new_bs
) {
2538 assert(bdrv_get_aio_context(old_bs
) == bdrv_get_aio_context(new_bs
));
2541 new_bs_quiesce_counter
= (new_bs
? new_bs
->quiesce_counter
: 0);
2542 drain_saldo
= new_bs_quiesce_counter
- child
->parent_quiesce_counter
;
2545 * If the new child node is drained but the old one was not, flush
2546 * all outstanding requests to the old child node.
2548 while (drain_saldo
> 0 && child
->klass
->drained_begin
) {
2549 bdrv_parent_drained_begin_single(child
, true);
2554 /* Detach first so that the recursive drain sections coming from @child
2555 * are already gone and we only end the drain sections that came from
2557 if (child
->klass
->detach
) {
2558 child
->klass
->detach(child
);
2560 QLIST_REMOVE(child
, next_parent
);
2566 QLIST_INSERT_HEAD(&new_bs
->parents
, child
, next_parent
);
2569 * Detaching the old node may have led to the new node's
2570 * quiesce_counter having been decreased. Not a problem, we
2571 * just need to recognize this here and then invoke
2572 * drained_end appropriately more often.
2574 assert(new_bs
->quiesce_counter
<= new_bs_quiesce_counter
);
2575 drain_saldo
+= new_bs
->quiesce_counter
- new_bs_quiesce_counter
;
2577 /* Attach only after starting new drained sections, so that recursive
2578 * drain sections coming from @child don't get an extra .drained_begin
2580 if (child
->klass
->attach
) {
2581 child
->klass
->attach(child
);
2586 * If the old child node was drained but the new one is not, allow
2587 * requests to come in only after the new node has been attached.
2589 while (drain_saldo
< 0 && child
->klass
->drained_end
) {
2590 bdrv_parent_drained_end_single(child
);
2596 * Updates @child to change its reference to point to @new_bs, including
2597 * checking and applying the necessary permisson updates both to the old node
2600 * NULL is passed as @new_bs for removing the reference before freeing @child.
2602 * If @new_bs is not NULL, bdrv_check_perm() must be called beforehand, as this
2603 * function uses bdrv_set_perm() to update the permissions according to the new
2604 * reference that @new_bs gets.
2606 static void bdrv_replace_child(BdrvChild
*child
, BlockDriverState
*new_bs
)
2608 BlockDriverState
*old_bs
= child
->bs
;
2609 uint64_t perm
, shared_perm
;
2611 bdrv_replace_child_noperm(child
, new_bs
);
2614 * Start with the new node's permissions. If @new_bs is a (direct
2615 * or indirect) child of @old_bs, we must complete the permission
2616 * update on @new_bs before we loosen the restrictions on @old_bs.
2617 * Otherwise, bdrv_check_perm() on @old_bs would re-initiate
2618 * updating the permissions of @new_bs, and thus not purely loosen
2622 bdrv_get_cumulative_perm(new_bs
, &perm
, &shared_perm
);
2623 bdrv_set_perm(new_bs
, perm
, shared_perm
);
2627 /* Update permissions for old node. This is guaranteed to succeed
2628 * because we're just taking a parent away, so we're loosening
2630 bool tighten_restrictions
;
2633 bdrv_get_cumulative_perm(old_bs
, &perm
, &shared_perm
);
2634 ret
= bdrv_check_perm(old_bs
, NULL
, perm
, shared_perm
, NULL
,
2635 &tighten_restrictions
, NULL
);
2636 assert(tighten_restrictions
== false);
2638 /* We only tried to loosen restrictions, so errors are not fatal */
2639 bdrv_abort_perm_update(old_bs
);
2641 bdrv_set_perm(old_bs
, perm
, shared_perm
);
2644 /* When the parent requiring a non-default AioContext is removed, the
2645 * node moves back to the main AioContext */
2646 bdrv_try_set_aio_context(old_bs
, qemu_get_aio_context(), NULL
);
2651 * This function steals the reference to child_bs from the caller.
2652 * That reference is later dropped by bdrv_root_unref_child().
2654 * On failure NULL is returned, errp is set and the reference to
2655 * child_bs is also dropped.
2657 * The caller must hold the AioContext lock @child_bs, but not that of @ctx
2658 * (unless @child_bs is already in @ctx).
2660 BdrvChild
*bdrv_root_attach_child(BlockDriverState
*child_bs
,
2661 const char *child_name
,
2662 const BdrvChildClass
*child_class
,
2663 BdrvChildRole child_role
,
2665 uint64_t perm
, uint64_t shared_perm
,
2666 void *opaque
, Error
**errp
)
2669 Error
*local_err
= NULL
;
2672 ret
= bdrv_check_update_perm(child_bs
, NULL
, perm
, shared_perm
, NULL
, NULL
,
2675 bdrv_abort_perm_update(child_bs
);
2676 bdrv_unref(child_bs
);
2680 child
= g_new(BdrvChild
, 1);
2681 *child
= (BdrvChild
) {
2683 .name
= g_strdup(child_name
),
2684 .klass
= child_class
,
2687 .shared_perm
= shared_perm
,
2691 /* If the AioContexts don't match, first try to move the subtree of
2692 * child_bs into the AioContext of the new parent. If this doesn't work,
2693 * try moving the parent into the AioContext of child_bs instead. */
2694 if (bdrv_get_aio_context(child_bs
) != ctx
) {
2695 ret
= bdrv_try_set_aio_context(child_bs
, ctx
, &local_err
);
2696 if (ret
< 0 && child_class
->can_set_aio_ctx
) {
2697 GSList
*ignore
= g_slist_prepend(NULL
, child
);
2698 ctx
= bdrv_get_aio_context(child_bs
);
2699 if (child_class
->can_set_aio_ctx(child
, ctx
, &ignore
, NULL
)) {
2700 error_free(local_err
);
2702 g_slist_free(ignore
);
2703 ignore
= g_slist_prepend(NULL
, child
);
2704 child_class
->set_aio_ctx(child
, ctx
, &ignore
);
2706 g_slist_free(ignore
);
2709 error_propagate(errp
, local_err
);
2711 bdrv_abort_perm_update(child_bs
);
2712 bdrv_unref(child_bs
);
2717 /* This performs the matching bdrv_set_perm() for the above check. */
2718 bdrv_replace_child(child
, child_bs
);
2724 * This function transfers the reference to child_bs from the caller
2725 * to parent_bs. That reference is later dropped by parent_bs on
2726 * bdrv_close() or if someone calls bdrv_unref_child().
2728 * On failure NULL is returned, errp is set and the reference to
2729 * child_bs is also dropped.
2731 * If @parent_bs and @child_bs are in different AioContexts, the caller must
2732 * hold the AioContext lock for @child_bs, but not for @parent_bs.
2734 BdrvChild
*bdrv_attach_child(BlockDriverState
*parent_bs
,
2735 BlockDriverState
*child_bs
,
2736 const char *child_name
,
2737 const BdrvChildClass
*child_class
,
2738 BdrvChildRole child_role
,
2742 uint64_t perm
, shared_perm
;
2744 bdrv_get_cumulative_perm(parent_bs
, &perm
, &shared_perm
);
2746 assert(parent_bs
->drv
);
2747 bdrv_child_perm(parent_bs
, child_bs
, NULL
, child_class
, child_role
, NULL
,
2748 perm
, shared_perm
, &perm
, &shared_perm
);
2750 child
= bdrv_root_attach_child(child_bs
, child_name
, child_class
,
2751 child_role
, bdrv_get_aio_context(parent_bs
),
2752 perm
, shared_perm
, parent_bs
, errp
);
2753 if (child
== NULL
) {
2757 QLIST_INSERT_HEAD(&parent_bs
->children
, child
, next
);
2761 static void bdrv_detach_child(BdrvChild
*child
)
2763 QLIST_SAFE_REMOVE(child
, next
);
2765 bdrv_replace_child(child
, NULL
);
2767 g_free(child
->name
);
2771 void bdrv_root_unref_child(BdrvChild
*child
)
2773 BlockDriverState
*child_bs
;
2775 child_bs
= child
->bs
;
2776 bdrv_detach_child(child
);
2777 bdrv_unref(child_bs
);
2781 * Clear all inherits_from pointers from children and grandchildren of
2782 * @root that point to @root, where necessary.
2784 static void bdrv_unset_inherits_from(BlockDriverState
*root
, BdrvChild
*child
)
2788 if (child
->bs
->inherits_from
== root
) {
2790 * Remove inherits_from only when the last reference between root and
2791 * child->bs goes away.
2793 QLIST_FOREACH(c
, &root
->children
, next
) {
2794 if (c
!= child
&& c
->bs
== child
->bs
) {
2799 child
->bs
->inherits_from
= NULL
;
2803 QLIST_FOREACH(c
, &child
->bs
->children
, next
) {
2804 bdrv_unset_inherits_from(root
, c
);
2808 void bdrv_unref_child(BlockDriverState
*parent
, BdrvChild
*child
)
2810 if (child
== NULL
) {
2814 bdrv_unset_inherits_from(parent
, child
);
2815 bdrv_root_unref_child(child
);
2819 static void bdrv_parent_cb_change_media(BlockDriverState
*bs
, bool load
)
2822 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
2823 if (c
->klass
->change_media
) {
2824 c
->klass
->change_media(c
, load
);
2829 /* Return true if you can reach parent going through child->inherits_from
2830 * recursively. If parent or child are NULL, return false */
2831 static bool bdrv_inherits_from_recursive(BlockDriverState
*child
,
2832 BlockDriverState
*parent
)
2834 while (child
&& child
!= parent
) {
2835 child
= child
->inherits_from
;
2838 return child
!= NULL
;
2842 * Return the BdrvChildRole for @bs's backing child. bs->backing is
2843 * mostly used for COW backing children (role = COW), but also for
2844 * filtered children (role = FILTERED | PRIMARY).
2846 static BdrvChildRole
bdrv_backing_role(BlockDriverState
*bs
)
2848 if (bs
->drv
&& bs
->drv
->is_filter
) {
2849 return BDRV_CHILD_FILTERED
| BDRV_CHILD_PRIMARY
;
2851 return BDRV_CHILD_COW
;
2856 * Sets the backing file link of a BDS. A new reference is created; callers
2857 * which don't need their own reference any more must call bdrv_unref().
2859 void bdrv_set_backing_hd(BlockDriverState
*bs
, BlockDriverState
*backing_hd
,
2862 bool update_inherits_from
= bdrv_chain_contains(bs
, backing_hd
) &&
2863 bdrv_inherits_from_recursive(backing_hd
, bs
);
2865 if (bdrv_is_backing_chain_frozen(bs
, backing_bs(bs
), errp
)) {
2870 bdrv_ref(backing_hd
);
2874 bdrv_unref_child(bs
, bs
->backing
);
2882 bs
->backing
= bdrv_attach_child(bs
, backing_hd
, "backing", &child_of_bds
,
2883 bdrv_backing_role(bs
), errp
);
2884 /* If backing_hd was already part of bs's backing chain, and
2885 * inherits_from pointed recursively to bs then let's update it to
2886 * point directly to bs (else it will become NULL). */
2887 if (bs
->backing
&& update_inherits_from
) {
2888 backing_hd
->inherits_from
= bs
;
2892 bdrv_refresh_limits(bs
, NULL
);
2896 * Opens the backing file for a BlockDriverState if not yet open
2898 * bdref_key specifies the key for the image's BlockdevRef in the options QDict.
2899 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
2900 * itself, all options starting with "${bdref_key}." are considered part of the
2903 * TODO Can this be unified with bdrv_open_image()?
2905 int bdrv_open_backing_file(BlockDriverState
*bs
, QDict
*parent_options
,
2906 const char *bdref_key
, Error
**errp
)
2908 char *backing_filename
= NULL
;
2909 char *bdref_key_dot
;
2910 const char *reference
= NULL
;
2912 bool implicit_backing
= false;
2913 BlockDriverState
*backing_hd
;
2915 QDict
*tmp_parent_options
= NULL
;
2916 Error
*local_err
= NULL
;
2918 if (bs
->backing
!= NULL
) {
2922 /* NULL means an empty set of options */
2923 if (parent_options
== NULL
) {
2924 tmp_parent_options
= qdict_new();
2925 parent_options
= tmp_parent_options
;
2928 bs
->open_flags
&= ~BDRV_O_NO_BACKING
;
2930 bdref_key_dot
= g_strdup_printf("%s.", bdref_key
);
2931 qdict_extract_subqdict(parent_options
, &options
, bdref_key_dot
);
2932 g_free(bdref_key_dot
);
2935 * Caution: while qdict_get_try_str() is fine, getting non-string
2936 * types would require more care. When @parent_options come from
2937 * -blockdev or blockdev_add, its members are typed according to
2938 * the QAPI schema, but when they come from -drive, they're all
2941 reference
= qdict_get_try_str(parent_options
, bdref_key
);
2942 if (reference
|| qdict_haskey(options
, "file.filename")) {
2943 /* keep backing_filename NULL */
2944 } else if (bs
->backing_file
[0] == '\0' && qdict_size(options
) == 0) {
2945 qobject_unref(options
);
2948 if (qdict_size(options
) == 0) {
2949 /* If the user specifies options that do not modify the
2950 * backing file's behavior, we might still consider it the
2951 * implicit backing file. But it's easier this way, and
2952 * just specifying some of the backing BDS's options is
2953 * only possible with -drive anyway (otherwise the QAPI
2954 * schema forces the user to specify everything). */
2955 implicit_backing
= !strcmp(bs
->auto_backing_file
, bs
->backing_file
);
2958 backing_filename
= bdrv_get_full_backing_filename(bs
, &local_err
);
2961 error_propagate(errp
, local_err
);
2962 qobject_unref(options
);
2967 if (!bs
->drv
|| !bs
->drv
->supports_backing
) {
2969 error_setg(errp
, "Driver doesn't support backing files");
2970 qobject_unref(options
);
2975 bs
->backing_format
[0] != '\0' && !qdict_haskey(options
, "driver")) {
2976 qdict_put_str(options
, "driver", bs
->backing_format
);
2979 backing_hd
= bdrv_open_inherit(backing_filename
, reference
, options
, 0, bs
,
2980 &child_of_bds
, bdrv_backing_role(bs
), errp
);
2982 bs
->open_flags
|= BDRV_O_NO_BACKING
;
2983 error_prepend(errp
, "Could not open backing file: ");
2988 if (implicit_backing
) {
2989 bdrv_refresh_filename(backing_hd
);
2990 pstrcpy(bs
->auto_backing_file
, sizeof(bs
->auto_backing_file
),
2991 backing_hd
->filename
);
2994 /* Hook up the backing file link; drop our reference, bs owns the
2995 * backing_hd reference now */
2996 bdrv_set_backing_hd(bs
, backing_hd
, &local_err
);
2997 bdrv_unref(backing_hd
);
2999 error_propagate(errp
, local_err
);
3004 qdict_del(parent_options
, bdref_key
);
3007 g_free(backing_filename
);
3008 qobject_unref(tmp_parent_options
);
3012 static BlockDriverState
*
3013 bdrv_open_child_bs(const char *filename
, QDict
*options
, const char *bdref_key
,
3014 BlockDriverState
*parent
, const BdrvChildClass
*child_class
,
3015 BdrvChildRole child_role
, bool allow_none
, Error
**errp
)
3017 BlockDriverState
*bs
= NULL
;
3018 QDict
*image_options
;
3019 char *bdref_key_dot
;
3020 const char *reference
;
3022 assert(child_class
!= NULL
);
3024 bdref_key_dot
= g_strdup_printf("%s.", bdref_key
);
3025 qdict_extract_subqdict(options
, &image_options
, bdref_key_dot
);
3026 g_free(bdref_key_dot
);
3029 * Caution: while qdict_get_try_str() is fine, getting non-string
3030 * types would require more care. When @options come from
3031 * -blockdev or blockdev_add, its members are typed according to
3032 * the QAPI schema, but when they come from -drive, they're all
3035 reference
= qdict_get_try_str(options
, bdref_key
);
3036 if (!filename
&& !reference
&& !qdict_size(image_options
)) {
3038 error_setg(errp
, "A block device must be specified for \"%s\"",
3041 qobject_unref(image_options
);
3045 bs
= bdrv_open_inherit(filename
, reference
, image_options
, 0,
3046 parent
, child_class
, child_role
, errp
);
3052 qdict_del(options
, bdref_key
);
3057 * Opens a disk image whose options are given as BlockdevRef in another block
3060 * If allow_none is true, no image will be opened if filename is false and no
3061 * BlockdevRef is given. NULL will be returned, but errp remains unset.
3063 * bdrev_key specifies the key for the image's BlockdevRef in the options QDict.
3064 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
3065 * itself, all options starting with "${bdref_key}." are considered part of the
3068 * The BlockdevRef will be removed from the options QDict.
3070 BdrvChild
*bdrv_open_child(const char *filename
,
3071 QDict
*options
, const char *bdref_key
,
3072 BlockDriverState
*parent
,
3073 const BdrvChildClass
*child_class
,
3074 BdrvChildRole child_role
,
3075 bool allow_none
, Error
**errp
)
3077 BlockDriverState
*bs
;
3079 bs
= bdrv_open_child_bs(filename
, options
, bdref_key
, parent
, child_class
,
3080 child_role
, allow_none
, errp
);
3085 return bdrv_attach_child(parent
, bs
, bdref_key
, child_class
, child_role
,
3090 * TODO Future callers may need to specify parent/child_class in order for
3091 * option inheritance to work. Existing callers use it for the root node.
3093 BlockDriverState
*bdrv_open_blockdev_ref(BlockdevRef
*ref
, Error
**errp
)
3095 BlockDriverState
*bs
= NULL
;
3096 QObject
*obj
= NULL
;
3097 QDict
*qdict
= NULL
;
3098 const char *reference
= NULL
;
3101 if (ref
->type
== QTYPE_QSTRING
) {
3102 reference
= ref
->u
.reference
;
3104 BlockdevOptions
*options
= &ref
->u
.definition
;
3105 assert(ref
->type
== QTYPE_QDICT
);
3107 v
= qobject_output_visitor_new(&obj
);
3108 visit_type_BlockdevOptions(v
, NULL
, &options
, &error_abort
);
3109 visit_complete(v
, &obj
);
3111 qdict
= qobject_to(QDict
, obj
);
3112 qdict_flatten(qdict
);
3114 /* bdrv_open_inherit() defaults to the values in bdrv_flags (for
3115 * compatibility with other callers) rather than what we want as the
3116 * real defaults. Apply the defaults here instead. */
3117 qdict_set_default_str(qdict
, BDRV_OPT_CACHE_DIRECT
, "off");
3118 qdict_set_default_str(qdict
, BDRV_OPT_CACHE_NO_FLUSH
, "off");
3119 qdict_set_default_str(qdict
, BDRV_OPT_READ_ONLY
, "off");
3120 qdict_set_default_str(qdict
, BDRV_OPT_AUTO_READ_ONLY
, "off");
3124 bs
= bdrv_open_inherit(NULL
, reference
, qdict
, 0, NULL
, NULL
, 0, errp
);
3131 static BlockDriverState
*bdrv_append_temp_snapshot(BlockDriverState
*bs
,
3133 QDict
*snapshot_options
,
3136 /* TODO: extra byte is a hack to ensure MAX_PATH space on Windows. */
3137 char *tmp_filename
= g_malloc0(PATH_MAX
+ 1);
3139 QemuOpts
*opts
= NULL
;
3140 BlockDriverState
*bs_snapshot
= NULL
;
3141 Error
*local_err
= NULL
;
3144 /* if snapshot, we create a temporary backing file and open it
3145 instead of opening 'filename' directly */
3147 /* Get the required size from the image */
3148 total_size
= bdrv_getlength(bs
);
3149 if (total_size
< 0) {
3150 error_setg_errno(errp
, -total_size
, "Could not get image size");
3154 /* Create the temporary image */
3155 ret
= get_tmp_filename(tmp_filename
, PATH_MAX
+ 1);
3157 error_setg_errno(errp
, -ret
, "Could not get temporary filename");
3161 opts
= qemu_opts_create(bdrv_qcow2
.create_opts
, NULL
, 0,
3163 qemu_opt_set_number(opts
, BLOCK_OPT_SIZE
, total_size
, &error_abort
);
3164 ret
= bdrv_create(&bdrv_qcow2
, tmp_filename
, opts
, errp
);
3165 qemu_opts_del(opts
);
3167 error_prepend(errp
, "Could not create temporary overlay '%s': ",
3172 /* Prepare options QDict for the temporary file */
3173 qdict_put_str(snapshot_options
, "file.driver", "file");
3174 qdict_put_str(snapshot_options
, "file.filename", tmp_filename
);
3175 qdict_put_str(snapshot_options
, "driver", "qcow2");
3177 bs_snapshot
= bdrv_open(NULL
, NULL
, snapshot_options
, flags
, errp
);
3178 snapshot_options
= NULL
;
3183 /* bdrv_append() consumes a strong reference to bs_snapshot
3184 * (i.e. it will call bdrv_unref() on it) even on error, so in
3185 * order to be able to return one, we have to increase
3186 * bs_snapshot's refcount here */
3187 bdrv_ref(bs_snapshot
);
3188 bdrv_append(bs_snapshot
, bs
, &local_err
);
3190 error_propagate(errp
, local_err
);
3196 qobject_unref(snapshot_options
);
3197 g_free(tmp_filename
);
3202 * Opens a disk image (raw, qcow2, vmdk, ...)
3204 * options is a QDict of options to pass to the block drivers, or NULL for an
3205 * empty set of options. The reference to the QDict belongs to the block layer
3206 * after the call (even on failure), so if the caller intends to reuse the
3207 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
3209 * If *pbs is NULL, a new BDS will be created with a pointer to it stored there.
3210 * If it is not NULL, the referenced BDS will be reused.
3212 * The reference parameter may be used to specify an existing block device which
3213 * should be opened. If specified, neither options nor a filename may be given,
3214 * nor can an existing BDS be reused (that is, *pbs has to be NULL).
3216 static BlockDriverState
*bdrv_open_inherit(const char *filename
,
3217 const char *reference
,
3218 QDict
*options
, int flags
,
3219 BlockDriverState
*parent
,
3220 const BdrvChildClass
*child_class
,
3221 BdrvChildRole child_role
,
3225 BlockBackend
*file
= NULL
;
3226 BlockDriverState
*bs
;
3227 BlockDriver
*drv
= NULL
;
3229 const char *drvname
;
3230 const char *backing
;
3231 Error
*local_err
= NULL
;
3232 QDict
*snapshot_options
= NULL
;
3233 int snapshot_flags
= 0;
3235 assert(!child_class
|| !flags
);
3236 assert(!child_class
== !parent
);
3239 bool options_non_empty
= options
? qdict_size(options
) : false;
3240 qobject_unref(options
);
3242 if (filename
|| options_non_empty
) {
3243 error_setg(errp
, "Cannot reference an existing block device with "
3244 "additional options or a new filename");
3248 bs
= bdrv_lookup_bs(reference
, reference
, errp
);
3259 /* NULL means an empty set of options */
3260 if (options
== NULL
) {
3261 options
= qdict_new();
3264 /* json: syntax counts as explicit options, as if in the QDict */
3265 parse_json_protocol(options
, &filename
, &local_err
);
3270 bs
->explicit_options
= qdict_clone_shallow(options
);
3273 bool parent_is_format
;
3276 parent_is_format
= parent
->drv
->is_format
;
3279 * parent->drv is not set yet because this node is opened for
3280 * (potential) format probing. That means that @parent is going
3281 * to be a format node.
3283 parent_is_format
= true;
3286 bs
->inherits_from
= parent
;
3287 child_class
->inherit_options(child_role
, parent_is_format
,
3289 parent
->open_flags
, parent
->options
);
3292 ret
= bdrv_fill_options(&options
, filename
, &flags
, &local_err
);
3298 * Set the BDRV_O_RDWR and BDRV_O_ALLOW_RDWR flags.
3299 * Caution: getting a boolean member of @options requires care.
3300 * When @options come from -blockdev or blockdev_add, members are
3301 * typed according to the QAPI schema, but when they come from
3302 * -drive, they're all QString.
3304 if (g_strcmp0(qdict_get_try_str(options
, BDRV_OPT_READ_ONLY
), "on") &&
3305 !qdict_get_try_bool(options
, BDRV_OPT_READ_ONLY
, false)) {
3306 flags
|= (BDRV_O_RDWR
| BDRV_O_ALLOW_RDWR
);
3308 flags
&= ~BDRV_O_RDWR
;
3311 if (flags
& BDRV_O_SNAPSHOT
) {
3312 snapshot_options
= qdict_new();
3313 bdrv_temp_snapshot_options(&snapshot_flags
, snapshot_options
,
3315 /* Let bdrv_backing_options() override "read-only" */
3316 qdict_del(options
, BDRV_OPT_READ_ONLY
);
3317 bdrv_inherited_options(BDRV_CHILD_COW
, true,
3318 &flags
, options
, flags
, options
);
3321 bs
->open_flags
= flags
;
3322 bs
->options
= options
;
3323 options
= qdict_clone_shallow(options
);
3325 /* Find the right image format driver */
3326 /* See cautionary note on accessing @options above */
3327 drvname
= qdict_get_try_str(options
, "driver");
3329 drv
= bdrv_find_format(drvname
);
3331 error_setg(errp
, "Unknown driver: '%s'", drvname
);
3336 assert(drvname
|| !(flags
& BDRV_O_PROTOCOL
));
3338 /* See cautionary note on accessing @options above */
3339 backing
= qdict_get_try_str(options
, "backing");
3340 if (qobject_to(QNull
, qdict_get(options
, "backing")) != NULL
||
3341 (backing
&& *backing
== '\0'))
3344 warn_report("Use of \"backing\": \"\" is deprecated; "
3345 "use \"backing\": null instead");
3347 flags
|= BDRV_O_NO_BACKING
;
3348 qdict_del(bs
->explicit_options
, "backing");
3349 qdict_del(bs
->options
, "backing");
3350 qdict_del(options
, "backing");
3353 /* Open image file without format layer. This BlockBackend is only used for
3354 * probing, the block drivers will do their own bdrv_open_child() for the
3355 * same BDS, which is why we put the node name back into options. */
3356 if ((flags
& BDRV_O_PROTOCOL
) == 0) {
3357 BlockDriverState
*file_bs
;
3359 file_bs
= bdrv_open_child_bs(filename
, options
, "file", bs
,
3360 &child_of_bds
, BDRV_CHILD_IMAGE
,
3365 if (file_bs
!= NULL
) {
3366 /* Not requesting BLK_PERM_CONSISTENT_READ because we're only
3367 * looking at the header to guess the image format. This works even
3368 * in cases where a guest would not see a consistent state. */
3369 file
= blk_new(bdrv_get_aio_context(file_bs
), 0, BLK_PERM_ALL
);
3370 blk_insert_bs(file
, file_bs
, &local_err
);
3371 bdrv_unref(file_bs
);
3376 qdict_put_str(options
, "file", bdrv_get_node_name(file_bs
));
3380 /* Image format probing */
3383 ret
= find_image_format(file
, filename
, &drv
, &local_err
);
3388 * This option update would logically belong in bdrv_fill_options(),
3389 * but we first need to open bs->file for the probing to work, while
3390 * opening bs->file already requires the (mostly) final set of options
3391 * so that cache mode etc. can be inherited.
3393 * Adding the driver later is somewhat ugly, but it's not an option
3394 * that would ever be inherited, so it's correct. We just need to make
3395 * sure to update both bs->options (which has the full effective
3396 * options for bs) and options (which has file.* already removed).
3398 qdict_put_str(bs
->options
, "driver", drv
->format_name
);
3399 qdict_put_str(options
, "driver", drv
->format_name
);
3401 error_setg(errp
, "Must specify either driver or file");
3405 /* BDRV_O_PROTOCOL must be set iff a protocol BDS is about to be created */
3406 assert(!!(flags
& BDRV_O_PROTOCOL
) == !!drv
->bdrv_file_open
);
3407 /* file must be NULL if a protocol BDS is about to be created
3408 * (the inverse results in an error message from bdrv_open_common()) */
3409 assert(!(flags
& BDRV_O_PROTOCOL
) || !file
);
3411 /* Open the image */
3412 ret
= bdrv_open_common(bs
, file
, options
, &local_err
);
3422 /* If there is a backing file, use it */
3423 if ((flags
& BDRV_O_NO_BACKING
) == 0) {
3424 ret
= bdrv_open_backing_file(bs
, options
, "backing", &local_err
);
3426 goto close_and_fail
;
3430 /* Remove all children options and references
3431 * from bs->options and bs->explicit_options */
3432 QLIST_FOREACH(child
, &bs
->children
, next
) {
3433 char *child_key_dot
;
3434 child_key_dot
= g_strdup_printf("%s.", child
->name
);
3435 qdict_extract_subqdict(bs
->explicit_options
, NULL
, child_key_dot
);
3436 qdict_extract_subqdict(bs
->options
, NULL
, child_key_dot
);
3437 qdict_del(bs
->explicit_options
, child
->name
);
3438 qdict_del(bs
->options
, child
->name
);
3439 g_free(child_key_dot
);
3442 /* Check if any unknown options were used */
3443 if (qdict_size(options
) != 0) {
3444 const QDictEntry
*entry
= qdict_first(options
);
3445 if (flags
& BDRV_O_PROTOCOL
) {
3446 error_setg(errp
, "Block protocol '%s' doesn't support the option "
3447 "'%s'", drv
->format_name
, entry
->key
);
3450 "Block format '%s' does not support the option '%s'",
3451 drv
->format_name
, entry
->key
);
3454 goto close_and_fail
;
3457 bdrv_parent_cb_change_media(bs
, true);
3459 qobject_unref(options
);
3462 /* For snapshot=on, create a temporary qcow2 overlay. bs points to the
3463 * temporary snapshot afterwards. */
3464 if (snapshot_flags
) {
3465 BlockDriverState
*snapshot_bs
;
3466 snapshot_bs
= bdrv_append_temp_snapshot(bs
, snapshot_flags
,
3467 snapshot_options
, &local_err
);
3468 snapshot_options
= NULL
;
3470 goto close_and_fail
;
3472 /* We are not going to return bs but the overlay on top of it
3473 * (snapshot_bs); thus, we have to drop the strong reference to bs
3474 * (which we obtained by calling bdrv_new()). bs will not be deleted,
3475 * though, because the overlay still has a reference to it. */
3484 qobject_unref(snapshot_options
);
3485 qobject_unref(bs
->explicit_options
);
3486 qobject_unref(bs
->options
);
3487 qobject_unref(options
);
3489 bs
->explicit_options
= NULL
;
3491 error_propagate(errp
, local_err
);
3496 qobject_unref(snapshot_options
);
3497 qobject_unref(options
);
3498 error_propagate(errp
, local_err
);
3502 BlockDriverState
*bdrv_open(const char *filename
, const char *reference
,
3503 QDict
*options
, int flags
, Error
**errp
)
3505 return bdrv_open_inherit(filename
, reference
, options
, flags
, NULL
,
3509 /* Return true if the NULL-terminated @list contains @str */
3510 static bool is_str_in_list(const char *str
, const char *const *list
)
3514 for (i
= 0; list
[i
] != NULL
; i
++) {
3515 if (!strcmp(str
, list
[i
])) {
3524 * Check that every option set in @bs->options is also set in
3527 * Options listed in the common_options list and in
3528 * @bs->drv->mutable_opts are skipped.
3530 * Return 0 on success, otherwise return -EINVAL and set @errp.
3532 static int bdrv_reset_options_allowed(BlockDriverState
*bs
,
3533 const QDict
*new_opts
, Error
**errp
)
3535 const QDictEntry
*e
;
3536 /* These options are common to all block drivers and are handled
3537 * in bdrv_reopen_prepare() so they can be left out of @new_opts */
3538 const char *const common_options
[] = {
3539 "node-name", "discard", "cache.direct", "cache.no-flush",
3540 "read-only", "auto-read-only", "detect-zeroes", NULL
3543 for (e
= qdict_first(bs
->options
); e
; e
= qdict_next(bs
->options
, e
)) {
3544 if (!qdict_haskey(new_opts
, e
->key
) &&
3545 !is_str_in_list(e
->key
, common_options
) &&
3546 !is_str_in_list(e
->key
, bs
->drv
->mutable_opts
)) {
3547 error_setg(errp
, "Option '%s' cannot be reset "
3548 "to its default value", e
->key
);
3557 * Returns true if @child can be reached recursively from @bs
3559 static bool bdrv_recurse_has_child(BlockDriverState
*bs
,
3560 BlockDriverState
*child
)
3568 QLIST_FOREACH(c
, &bs
->children
, next
) {
3569 if (bdrv_recurse_has_child(c
->bs
, child
)) {
3578 * Adds a BlockDriverState to a simple queue for an atomic, transactional
3579 * reopen of multiple devices.
3581 * bs_queue can either be an existing BlockReopenQueue that has had QTAILQ_INIT
3582 * already performed, or alternatively may be NULL a new BlockReopenQueue will
3583 * be created and initialized. This newly created BlockReopenQueue should be
3584 * passed back in for subsequent calls that are intended to be of the same
3587 * bs is the BlockDriverState to add to the reopen queue.
3589 * options contains the changed options for the associated bs
3590 * (the BlockReopenQueue takes ownership)
3592 * flags contains the open flags for the associated bs
3594 * returns a pointer to bs_queue, which is either the newly allocated
3595 * bs_queue, or the existing bs_queue being used.
3597 * bs must be drained between bdrv_reopen_queue() and bdrv_reopen_multiple().
3599 static BlockReopenQueue
*bdrv_reopen_queue_child(BlockReopenQueue
*bs_queue
,
3600 BlockDriverState
*bs
,
3602 const BdrvChildClass
*klass
,
3604 bool parent_is_format
,
3605 QDict
*parent_options
,
3611 BlockReopenQueueEntry
*bs_entry
;
3613 QDict
*old_options
, *explicit_options
, *options_copy
;
3617 /* Make sure that the caller remembered to use a drained section. This is
3618 * important to avoid graph changes between the recursive queuing here and
3619 * bdrv_reopen_multiple(). */
3620 assert(bs
->quiesce_counter
> 0);
3622 if (bs_queue
== NULL
) {
3623 bs_queue
= g_new0(BlockReopenQueue
, 1);
3624 QTAILQ_INIT(bs_queue
);
3628 options
= qdict_new();
3631 /* Check if this BlockDriverState is already in the queue */
3632 QTAILQ_FOREACH(bs_entry
, bs_queue
, entry
) {
3633 if (bs
== bs_entry
->state
.bs
) {
3639 * Precedence of options:
3640 * 1. Explicitly passed in options (highest)
3641 * 2. Retained from explicitly set options of bs
3642 * 3. Inherited from parent node
3643 * 4. Retained from effective options of bs
3646 /* Old explicitly set values (don't overwrite by inherited value) */
3647 if (bs_entry
|| keep_old_opts
) {
3648 old_options
= qdict_clone_shallow(bs_entry
?
3649 bs_entry
->state
.explicit_options
:
3650 bs
->explicit_options
);
3651 bdrv_join_options(bs
, options
, old_options
);
3652 qobject_unref(old_options
);
3655 explicit_options
= qdict_clone_shallow(options
);
3657 /* Inherit from parent node */
3658 if (parent_options
) {
3660 klass
->inherit_options(role
, parent_is_format
, &flags
, options
,
3661 parent_flags
, parent_options
);
3663 flags
= bdrv_get_flags(bs
);
3666 if (keep_old_opts
) {
3667 /* Old values are used for options that aren't set yet */
3668 old_options
= qdict_clone_shallow(bs
->options
);
3669 bdrv_join_options(bs
, options
, old_options
);
3670 qobject_unref(old_options
);
3673 /* We have the final set of options so let's update the flags */
3674 options_copy
= qdict_clone_shallow(options
);
3675 opts
= qemu_opts_create(&bdrv_runtime_opts
, NULL
, 0, &error_abort
);
3676 qemu_opts_absorb_qdict(opts
, options_copy
, NULL
);
3677 update_flags_from_options(&flags
, opts
);
3678 qemu_opts_del(opts
);
3679 qobject_unref(options_copy
);
3681 /* bdrv_open_inherit() sets and clears some additional flags internally */
3682 flags
&= ~BDRV_O_PROTOCOL
;
3683 if (flags
& BDRV_O_RDWR
) {
3684 flags
|= BDRV_O_ALLOW_RDWR
;
3688 bs_entry
= g_new0(BlockReopenQueueEntry
, 1);
3689 QTAILQ_INSERT_TAIL(bs_queue
, bs_entry
, entry
);
3691 qobject_unref(bs_entry
->state
.options
);
3692 qobject_unref(bs_entry
->state
.explicit_options
);
3695 bs_entry
->state
.bs
= bs
;
3696 bs_entry
->state
.options
= options
;
3697 bs_entry
->state
.explicit_options
= explicit_options
;
3698 bs_entry
->state
.flags
= flags
;
3700 /* This needs to be overwritten in bdrv_reopen_prepare() */
3701 bs_entry
->state
.perm
= UINT64_MAX
;
3702 bs_entry
->state
.shared_perm
= 0;
3705 * If keep_old_opts is false then it means that unspecified
3706 * options must be reset to their original value. We don't allow
3707 * resetting 'backing' but we need to know if the option is
3708 * missing in order to decide if we have to return an error.
3710 if (!keep_old_opts
) {
3711 bs_entry
->state
.backing_missing
=
3712 !qdict_haskey(options
, "backing") &&
3713 !qdict_haskey(options
, "backing.driver");
3716 QLIST_FOREACH(child
, &bs
->children
, next
) {
3717 QDict
*new_child_options
= NULL
;
3718 bool child_keep_old
= keep_old_opts
;
3720 /* reopen can only change the options of block devices that were
3721 * implicitly created and inherited options. For other (referenced)
3722 * block devices, a syntax like "backing.foo" results in an error. */
3723 if (child
->bs
->inherits_from
!= bs
) {
3727 /* Check if the options contain a child reference */
3728 if (qdict_haskey(options
, child
->name
)) {
3729 const char *childref
= qdict_get_try_str(options
, child
->name
);
3731 * The current child must not be reopened if the child
3732 * reference is null or points to a different node.
3734 if (g_strcmp0(childref
, child
->bs
->node_name
)) {
3738 * If the child reference points to the current child then
3739 * reopen it with its existing set of options (note that
3740 * it can still inherit new options from the parent).
3742 child_keep_old
= true;
3744 /* Extract child options ("child-name.*") */
3745 char *child_key_dot
= g_strdup_printf("%s.", child
->name
);
3746 qdict_extract_subqdict(explicit_options
, NULL
, child_key_dot
);
3747 qdict_extract_subqdict(options
, &new_child_options
, child_key_dot
);
3748 g_free(child_key_dot
);
3751 bdrv_reopen_queue_child(bs_queue
, child
->bs
, new_child_options
,
3752 child
->klass
, child
->role
, bs
->drv
->is_format
,
3753 options
, flags
, child_keep_old
);
3759 BlockReopenQueue
*bdrv_reopen_queue(BlockReopenQueue
*bs_queue
,
3760 BlockDriverState
*bs
,
3761 QDict
*options
, bool keep_old_opts
)
3763 return bdrv_reopen_queue_child(bs_queue
, bs
, options
, NULL
, 0, false,
3764 NULL
, 0, keep_old_opts
);
3768 * Reopen multiple BlockDriverStates atomically & transactionally.
3770 * The queue passed in (bs_queue) must have been built up previous
3771 * via bdrv_reopen_queue().
3773 * Reopens all BDS specified in the queue, with the appropriate
3774 * flags. All devices are prepared for reopen, and failure of any
3775 * device will cause all device changes to be abandoned, and intermediate
3778 * If all devices prepare successfully, then the changes are committed
3781 * All affected nodes must be drained between bdrv_reopen_queue() and
3782 * bdrv_reopen_multiple().
3784 int bdrv_reopen_multiple(BlockReopenQueue
*bs_queue
, Error
**errp
)
3787 BlockReopenQueueEntry
*bs_entry
, *next
;
3789 assert(bs_queue
!= NULL
);
3791 QTAILQ_FOREACH(bs_entry
, bs_queue
, entry
) {
3792 assert(bs_entry
->state
.bs
->quiesce_counter
> 0);
3793 if (bdrv_reopen_prepare(&bs_entry
->state
, bs_queue
, errp
)) {
3796 bs_entry
->prepared
= true;
3799 QTAILQ_FOREACH(bs_entry
, bs_queue
, entry
) {
3800 BDRVReopenState
*state
= &bs_entry
->state
;
3801 ret
= bdrv_check_perm(state
->bs
, bs_queue
, state
->perm
,
3802 state
->shared_perm
, NULL
, NULL
, errp
);
3806 /* Check if new_backing_bs would accept the new permissions */
3807 if (state
->replace_backing_bs
&& state
->new_backing_bs
) {
3808 uint64_t nperm
, nshared
;
3809 bdrv_child_perm(state
->bs
, state
->new_backing_bs
,
3810 NULL
, &child_of_bds
, bdrv_backing_role(state
->bs
),
3811 bs_queue
, state
->perm
, state
->shared_perm
,
3813 ret
= bdrv_check_update_perm(state
->new_backing_bs
, NULL
,
3814 nperm
, nshared
, NULL
, NULL
, errp
);
3819 bs_entry
->perms_checked
= true;
3823 * If we reach this point, we have success and just need to apply the
3826 * Reverse order is used to comfort qcow2 driver: on commit it need to write
3827 * IN_USE flag to the image, to mark bitmaps in the image as invalid. But
3828 * children are usually goes after parents in reopen-queue, so go from last
3831 QTAILQ_FOREACH_REVERSE(bs_entry
, bs_queue
, entry
) {
3832 bdrv_reopen_commit(&bs_entry
->state
);
3837 QTAILQ_FOREACH_SAFE(bs_entry
, bs_queue
, entry
, next
) {
3838 BDRVReopenState
*state
= &bs_entry
->state
;
3840 if (!bs_entry
->perms_checked
) {
3845 bdrv_set_perm(state
->bs
, state
->perm
, state
->shared_perm
);
3847 bdrv_abort_perm_update(state
->bs
);
3848 if (state
->replace_backing_bs
&& state
->new_backing_bs
) {
3849 bdrv_abort_perm_update(state
->new_backing_bs
);
3855 QTAILQ_FOREACH_REVERSE(bs_entry
, bs_queue
, entry
) {
3856 BlockDriverState
*bs
= bs_entry
->state
.bs
;
3858 if (bs
->drv
->bdrv_reopen_commit_post
)
3859 bs
->drv
->bdrv_reopen_commit_post(&bs_entry
->state
);
3863 QTAILQ_FOREACH_SAFE(bs_entry
, bs_queue
, entry
, next
) {
3865 if (bs_entry
->prepared
) {
3866 bdrv_reopen_abort(&bs_entry
->state
);
3868 qobject_unref(bs_entry
->state
.explicit_options
);
3869 qobject_unref(bs_entry
->state
.options
);
3871 if (bs_entry
->state
.new_backing_bs
) {
3872 bdrv_unref(bs_entry
->state
.new_backing_bs
);
3881 int bdrv_reopen_set_read_only(BlockDriverState
*bs
, bool read_only
,
3885 BlockReopenQueue
*queue
;
3886 QDict
*opts
= qdict_new();
3888 qdict_put_bool(opts
, BDRV_OPT_READ_ONLY
, read_only
);
3890 bdrv_subtree_drained_begin(bs
);
3891 queue
= bdrv_reopen_queue(NULL
, bs
, opts
, true);
3892 ret
= bdrv_reopen_multiple(queue
, errp
);
3893 bdrv_subtree_drained_end(bs
);
3898 static BlockReopenQueueEntry
*find_parent_in_reopen_queue(BlockReopenQueue
*q
,
3901 BlockReopenQueueEntry
*entry
;
3903 QTAILQ_FOREACH(entry
, q
, entry
) {
3904 BlockDriverState
*bs
= entry
->state
.bs
;
3907 QLIST_FOREACH(child
, &bs
->children
, next
) {
3917 static void bdrv_reopen_perm(BlockReopenQueue
*q
, BlockDriverState
*bs
,
3918 uint64_t *perm
, uint64_t *shared
)
3921 BlockReopenQueueEntry
*parent
;
3922 uint64_t cumulative_perms
= 0;
3923 uint64_t cumulative_shared_perms
= BLK_PERM_ALL
;
3925 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
3926 parent
= find_parent_in_reopen_queue(q
, c
);
3928 cumulative_perms
|= c
->perm
;
3929 cumulative_shared_perms
&= c
->shared_perm
;
3931 uint64_t nperm
, nshared
;
3933 bdrv_child_perm(parent
->state
.bs
, bs
, c
, c
->klass
, c
->role
, q
,
3934 parent
->state
.perm
, parent
->state
.shared_perm
,
3937 cumulative_perms
|= nperm
;
3938 cumulative_shared_perms
&= nshared
;
3941 *perm
= cumulative_perms
;
3942 *shared
= cumulative_shared_perms
;
3945 static bool bdrv_reopen_can_attach(BlockDriverState
*parent
,
3947 BlockDriverState
*new_child
,
3950 AioContext
*parent_ctx
= bdrv_get_aio_context(parent
);
3951 AioContext
*child_ctx
= bdrv_get_aio_context(new_child
);
3955 ignore
= g_slist_prepend(NULL
, child
);
3956 ret
= bdrv_can_set_aio_context(new_child
, parent_ctx
, &ignore
, NULL
);
3957 g_slist_free(ignore
);
3962 ignore
= g_slist_prepend(NULL
, child
);
3963 ret
= bdrv_can_set_aio_context(parent
, child_ctx
, &ignore
, errp
);
3964 g_slist_free(ignore
);
3969 * Take a BDRVReopenState and check if the value of 'backing' in the
3970 * reopen_state->options QDict is valid or not.
3972 * If 'backing' is missing from the QDict then return 0.
3974 * If 'backing' contains the node name of the backing file of
3975 * reopen_state->bs then return 0.
3977 * If 'backing' contains a different node name (or is null) then check
3978 * whether the current backing file can be replaced with the new one.
3979 * If that's the case then reopen_state->replace_backing_bs is set to
3980 * true and reopen_state->new_backing_bs contains a pointer to the new
3981 * backing BlockDriverState (or NULL).
3983 * Return 0 on success, otherwise return < 0 and set @errp.
3985 static int bdrv_reopen_parse_backing(BDRVReopenState
*reopen_state
,
3988 BlockDriverState
*bs
= reopen_state
->bs
;
3989 BlockDriverState
*overlay_bs
, *new_backing_bs
;
3993 value
= qdict_get(reopen_state
->options
, "backing");
3994 if (value
== NULL
) {
3998 switch (qobject_type(value
)) {
4000 new_backing_bs
= NULL
;
4003 str
= qobject_get_try_str(value
);
4004 new_backing_bs
= bdrv_lookup_bs(NULL
, str
, errp
);
4005 if (new_backing_bs
== NULL
) {
4007 } else if (bdrv_recurse_has_child(new_backing_bs
, bs
)) {
4008 error_setg(errp
, "Making '%s' a backing file of '%s' "
4009 "would create a cycle", str
, bs
->node_name
);
4014 /* 'backing' does not allow any other data type */
4015 g_assert_not_reached();
4019 * Check AioContext compatibility so that the bdrv_set_backing_hd() call in
4020 * bdrv_reopen_commit() won't fail.
4022 if (new_backing_bs
) {
4023 if (!bdrv_reopen_can_attach(bs
, bs
->backing
, new_backing_bs
, errp
)) {
4029 * Find the "actual" backing file by skipping all links that point
4030 * to an implicit node, if any (e.g. a commit filter node).
4033 while (backing_bs(overlay_bs
) && backing_bs(overlay_bs
)->implicit
) {
4034 overlay_bs
= backing_bs(overlay_bs
);
4037 /* If we want to replace the backing file we need some extra checks */
4038 if (new_backing_bs
!= backing_bs(overlay_bs
)) {
4039 /* Check for implicit nodes between bs and its backing file */
4040 if (bs
!= overlay_bs
) {
4041 error_setg(errp
, "Cannot change backing link if '%s' has "
4042 "an implicit backing file", bs
->node_name
);
4045 /* Check if the backing link that we want to replace is frozen */
4046 if (bdrv_is_backing_chain_frozen(overlay_bs
, backing_bs(overlay_bs
),
4050 reopen_state
->replace_backing_bs
= true;
4051 if (new_backing_bs
) {
4052 bdrv_ref(new_backing_bs
);
4053 reopen_state
->new_backing_bs
= new_backing_bs
;
4061 * Prepares a BlockDriverState for reopen. All changes are staged in the
4062 * 'opaque' field of the BDRVReopenState, which is used and allocated by
4063 * the block driver layer .bdrv_reopen_prepare()
4065 * bs is the BlockDriverState to reopen
4066 * flags are the new open flags
4067 * queue is the reopen queue
4069 * Returns 0 on success, non-zero on error. On error errp will be set
4072 * On failure, bdrv_reopen_abort() will be called to clean up any data.
4073 * It is the responsibility of the caller to then call the abort() or
4074 * commit() for any other BDS that have been left in a prepare() state
4077 int bdrv_reopen_prepare(BDRVReopenState
*reopen_state
, BlockReopenQueue
*queue
,
4082 Error
*local_err
= NULL
;
4085 QDict
*orig_reopen_opts
;
4086 char *discard
= NULL
;
4088 bool drv_prepared
= false;
4090 assert(reopen_state
!= NULL
);
4091 assert(reopen_state
->bs
->drv
!= NULL
);
4092 drv
= reopen_state
->bs
->drv
;
4094 /* This function and each driver's bdrv_reopen_prepare() remove
4095 * entries from reopen_state->options as they are processed, so
4096 * we need to make a copy of the original QDict. */
4097 orig_reopen_opts
= qdict_clone_shallow(reopen_state
->options
);
4099 /* Process generic block layer options */
4100 opts
= qemu_opts_create(&bdrv_runtime_opts
, NULL
, 0, &error_abort
);
4101 qemu_opts_absorb_qdict(opts
, reopen_state
->options
, &local_err
);
4103 error_propagate(errp
, local_err
);
4108 /* This was already called in bdrv_reopen_queue_child() so the flags
4109 * are up-to-date. This time we simply want to remove the options from
4110 * QemuOpts in order to indicate that they have been processed. */
4111 old_flags
= reopen_state
->flags
;
4112 update_flags_from_options(&reopen_state
->flags
, opts
);
4113 assert(old_flags
== reopen_state
->flags
);
4115 discard
= qemu_opt_get_del(opts
, BDRV_OPT_DISCARD
);
4116 if (discard
!= NULL
) {
4117 if (bdrv_parse_discard_flags(discard
, &reopen_state
->flags
) != 0) {
4118 error_setg(errp
, "Invalid discard option");
4124 reopen_state
->detect_zeroes
=
4125 bdrv_parse_detect_zeroes(opts
, reopen_state
->flags
, &local_err
);
4127 error_propagate(errp
, local_err
);
4132 /* All other options (including node-name and driver) must be unchanged.
4133 * Put them back into the QDict, so that they are checked at the end
4134 * of this function. */
4135 qemu_opts_to_qdict(opts
, reopen_state
->options
);
4137 /* If we are to stay read-only, do not allow permission change
4138 * to r/w. Attempting to set to r/w may fail if either BDRV_O_ALLOW_RDWR is
4139 * not set, or if the BDS still has copy_on_read enabled */
4140 read_only
= !(reopen_state
->flags
& BDRV_O_RDWR
);
4141 ret
= bdrv_can_set_read_only(reopen_state
->bs
, read_only
, true, &local_err
);
4143 error_propagate(errp
, local_err
);
4147 /* Calculate required permissions after reopening */
4148 bdrv_reopen_perm(queue
, reopen_state
->bs
,
4149 &reopen_state
->perm
, &reopen_state
->shared_perm
);
4151 ret
= bdrv_flush(reopen_state
->bs
);
4153 error_setg_errno(errp
, -ret
, "Error flushing drive");
4157 if (drv
->bdrv_reopen_prepare
) {
4159 * If a driver-specific option is missing, it means that we
4160 * should reset it to its default value.
4161 * But not all options allow that, so we need to check it first.
4163 ret
= bdrv_reset_options_allowed(reopen_state
->bs
,
4164 reopen_state
->options
, errp
);
4169 ret
= drv
->bdrv_reopen_prepare(reopen_state
, queue
, &local_err
);
4171 if (local_err
!= NULL
) {
4172 error_propagate(errp
, local_err
);
4174 bdrv_refresh_filename(reopen_state
->bs
);
4175 error_setg(errp
, "failed while preparing to reopen image '%s'",
4176 reopen_state
->bs
->filename
);
4181 /* It is currently mandatory to have a bdrv_reopen_prepare()
4182 * handler for each supported drv. */
4183 error_setg(errp
, "Block format '%s' used by node '%s' "
4184 "does not support reopening files", drv
->format_name
,
4185 bdrv_get_device_or_node_name(reopen_state
->bs
));
4190 drv_prepared
= true;
4193 * We must provide the 'backing' option if the BDS has a backing
4194 * file or if the image file has a backing file name as part of
4195 * its metadata. Otherwise the 'backing' option can be omitted.
4197 if (drv
->supports_backing
&& reopen_state
->backing_missing
&&
4198 (backing_bs(reopen_state
->bs
) || reopen_state
->bs
->backing_file
[0])) {
4199 error_setg(errp
, "backing is missing for '%s'",
4200 reopen_state
->bs
->node_name
);
4206 * Allow changing the 'backing' option. The new value can be
4207 * either a reference to an existing node (using its node name)
4208 * or NULL to simply detach the current backing file.
4210 ret
= bdrv_reopen_parse_backing(reopen_state
, errp
);
4214 qdict_del(reopen_state
->options
, "backing");
4216 /* Options that are not handled are only okay if they are unchanged
4217 * compared to the old state. It is expected that some options are only
4218 * used for the initial open, but not reopen (e.g. filename) */
4219 if (qdict_size(reopen_state
->options
)) {
4220 const QDictEntry
*entry
= qdict_first(reopen_state
->options
);
4223 QObject
*new = entry
->value
;
4224 QObject
*old
= qdict_get(reopen_state
->bs
->options
, entry
->key
);
4226 /* Allow child references (child_name=node_name) as long as they
4227 * point to the current child (i.e. everything stays the same). */
4228 if (qobject_type(new) == QTYPE_QSTRING
) {
4230 QLIST_FOREACH(child
, &reopen_state
->bs
->children
, next
) {
4231 if (!strcmp(child
->name
, entry
->key
)) {
4237 const char *str
= qobject_get_try_str(new);
4238 if (!strcmp(child
->bs
->node_name
, str
)) {
4239 continue; /* Found child with this name, skip option */
4245 * TODO: When using -drive to specify blockdev options, all values
4246 * will be strings; however, when using -blockdev, blockdev-add or
4247 * filenames using the json:{} pseudo-protocol, they will be
4249 * In contrast, reopening options are (currently) always strings
4250 * (because you can only specify them through qemu-io; all other
4251 * callers do not specify any options).
4252 * Therefore, when using anything other than -drive to create a BDS,
4253 * this cannot detect non-string options as unchanged, because
4254 * qobject_is_equal() always returns false for objects of different
4255 * type. In the future, this should be remedied by correctly typing
4256 * all options. For now, this is not too big of an issue because
4257 * the user can simply omit options which cannot be changed anyway,
4258 * so they will stay unchanged.
4260 if (!qobject_is_equal(new, old
)) {
4261 error_setg(errp
, "Cannot change the option '%s'", entry
->key
);
4265 } while ((entry
= qdict_next(reopen_state
->options
, entry
)));
4270 /* Restore the original reopen_state->options QDict */
4271 qobject_unref(reopen_state
->options
);
4272 reopen_state
->options
= qobject_ref(orig_reopen_opts
);
4275 if (ret
< 0 && drv_prepared
) {
4276 /* drv->bdrv_reopen_prepare() has succeeded, so we need to
4277 * call drv->bdrv_reopen_abort() before signaling an error
4278 * (bdrv_reopen_multiple() will not call bdrv_reopen_abort()
4279 * when the respective bdrv_reopen_prepare() has failed) */
4280 if (drv
->bdrv_reopen_abort
) {
4281 drv
->bdrv_reopen_abort(reopen_state
);
4284 qemu_opts_del(opts
);
4285 qobject_unref(orig_reopen_opts
);
4291 * Takes the staged changes for the reopen from bdrv_reopen_prepare(), and
4292 * makes them final by swapping the staging BlockDriverState contents into
4293 * the active BlockDriverState contents.
4295 void bdrv_reopen_commit(BDRVReopenState
*reopen_state
)
4298 BlockDriverState
*bs
;
4301 assert(reopen_state
!= NULL
);
4302 bs
= reopen_state
->bs
;
4304 assert(drv
!= NULL
);
4306 /* If there are any driver level actions to take */
4307 if (drv
->bdrv_reopen_commit
) {
4308 drv
->bdrv_reopen_commit(reopen_state
);
4311 /* set BDS specific flags now */
4312 qobject_unref(bs
->explicit_options
);
4313 qobject_unref(bs
->options
);
4315 bs
->explicit_options
= reopen_state
->explicit_options
;
4316 bs
->options
= reopen_state
->options
;
4317 bs
->open_flags
= reopen_state
->flags
;
4318 bs
->read_only
= !(reopen_state
->flags
& BDRV_O_RDWR
);
4319 bs
->detect_zeroes
= reopen_state
->detect_zeroes
;
4321 if (reopen_state
->replace_backing_bs
) {
4322 qdict_del(bs
->explicit_options
, "backing");
4323 qdict_del(bs
->options
, "backing");
4326 /* Remove child references from bs->options and bs->explicit_options.
4327 * Child options were already removed in bdrv_reopen_queue_child() */
4328 QLIST_FOREACH(child
, &bs
->children
, next
) {
4329 qdict_del(bs
->explicit_options
, child
->name
);
4330 qdict_del(bs
->options
, child
->name
);
4334 * Change the backing file if a new one was specified. We do this
4335 * after updating bs->options, so bdrv_refresh_filename() (called
4336 * from bdrv_set_backing_hd()) has the new values.
4338 if (reopen_state
->replace_backing_bs
) {
4339 BlockDriverState
*old_backing_bs
= backing_bs(bs
);
4340 assert(!old_backing_bs
|| !old_backing_bs
->implicit
);
4341 /* Abort the permission update on the backing bs we're detaching */
4342 if (old_backing_bs
) {
4343 bdrv_abort_perm_update(old_backing_bs
);
4345 bdrv_set_backing_hd(bs
, reopen_state
->new_backing_bs
, &error_abort
);
4348 bdrv_refresh_limits(bs
, NULL
);
4352 * Abort the reopen, and delete and free the staged changes in
4355 void bdrv_reopen_abort(BDRVReopenState
*reopen_state
)
4359 assert(reopen_state
!= NULL
);
4360 drv
= reopen_state
->bs
->drv
;
4361 assert(drv
!= NULL
);
4363 if (drv
->bdrv_reopen_abort
) {
4364 drv
->bdrv_reopen_abort(reopen_state
);
4369 static void bdrv_close(BlockDriverState
*bs
)
4371 BdrvAioNotifier
*ban
, *ban_next
;
4372 BdrvChild
*child
, *next
;
4374 assert(!bs
->refcnt
);
4376 bdrv_drained_begin(bs
); /* complete I/O */
4378 bdrv_drain(bs
); /* in case flush left pending I/O */
4381 if (bs
->drv
->bdrv_close
) {
4382 bs
->drv
->bdrv_close(bs
);
4387 QLIST_FOREACH_SAFE(child
, &bs
->children
, next
, next
) {
4388 bdrv_unref_child(bs
, child
);
4395 atomic_set(&bs
->copy_on_read
, 0);
4396 bs
->backing_file
[0] = '\0';
4397 bs
->backing_format
[0] = '\0';
4398 bs
->total_sectors
= 0;
4399 bs
->encrypted
= false;
4401 qobject_unref(bs
->options
);
4402 qobject_unref(bs
->explicit_options
);
4404 bs
->explicit_options
= NULL
;
4405 qobject_unref(bs
->full_open_options
);
4406 bs
->full_open_options
= NULL
;
4408 bdrv_release_named_dirty_bitmaps(bs
);
4409 assert(QLIST_EMPTY(&bs
->dirty_bitmaps
));
4411 QLIST_FOREACH_SAFE(ban
, &bs
->aio_notifiers
, list
, ban_next
) {
4414 QLIST_INIT(&bs
->aio_notifiers
);
4415 bdrv_drained_end(bs
);
4418 void bdrv_close_all(void)
4420 assert(job_next(NULL
) == NULL
);
4421 nbd_export_close_all();
4423 /* Drop references from requests still in flight, such as canceled block
4424 * jobs whose AIO context has not been polled yet */
4427 blk_remove_all_bs();
4428 blockdev_close_all_bdrv_states();
4430 assert(QTAILQ_EMPTY(&all_bdrv_states
));
4433 static bool should_update_child(BdrvChild
*c
, BlockDriverState
*to
)
4439 if (c
->klass
->stay_at_node
) {
4443 /* If the child @c belongs to the BDS @to, replacing the current
4444 * c->bs by @to would mean to create a loop.
4446 * Such a case occurs when appending a BDS to a backing chain.
4447 * For instance, imagine the following chain:
4449 * guest device -> node A -> further backing chain...
4451 * Now we create a new BDS B which we want to put on top of this
4452 * chain, so we first attach A as its backing node:
4457 * guest device -> node A -> further backing chain...
4459 * Finally we want to replace A by B. When doing that, we want to
4460 * replace all pointers to A by pointers to B -- except for the
4461 * pointer from B because (1) that would create a loop, and (2)
4462 * that pointer should simply stay intact:
4464 * guest device -> node B
4467 * node A -> further backing chain...
4469 * In general, when replacing a node A (c->bs) by a node B (@to),
4470 * if A is a child of B, that means we cannot replace A by B there
4471 * because that would create a loop. Silently detaching A from B
4472 * is also not really an option. So overall just leaving A in
4473 * place there is the most sensible choice.
4475 * We would also create a loop in any cases where @c is only
4476 * indirectly referenced by @to. Prevent this by returning false
4477 * if @c is found (by breadth-first search) anywhere in the whole
4482 found
= g_hash_table_new(NULL
, NULL
);
4483 g_hash_table_add(found
, to
);
4484 queue
= g_queue_new();
4485 g_queue_push_tail(queue
, to
);
4487 while (!g_queue_is_empty(queue
)) {
4488 BlockDriverState
*v
= g_queue_pop_head(queue
);
4491 QLIST_FOREACH(c2
, &v
->children
, next
) {
4497 if (g_hash_table_contains(found
, c2
->bs
)) {
4501 g_queue_push_tail(queue
, c2
->bs
);
4502 g_hash_table_add(found
, c2
->bs
);
4506 g_queue_free(queue
);
4507 g_hash_table_destroy(found
);
4512 void bdrv_replace_node(BlockDriverState
*from
, BlockDriverState
*to
,
4515 BdrvChild
*c
, *next
;
4516 GSList
*list
= NULL
, *p
;
4517 uint64_t perm
= 0, shared
= BLK_PERM_ALL
;
4520 /* Make sure that @from doesn't go away until we have successfully attached
4521 * all of its parents to @to. */
4524 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
4525 assert(bdrv_get_aio_context(from
) == bdrv_get_aio_context(to
));
4526 bdrv_drained_begin(from
);
4528 /* Put all parents into @list and calculate their cumulative permissions */
4529 QLIST_FOREACH_SAFE(c
, &from
->parents
, next_parent
, next
) {
4530 assert(c
->bs
== from
);
4531 if (!should_update_child(c
, to
)) {
4535 error_setg(errp
, "Cannot change '%s' link to '%s'",
4536 c
->name
, from
->node_name
);
4539 list
= g_slist_prepend(list
, c
);
4541 shared
&= c
->shared_perm
;
4544 /* Check whether the required permissions can be granted on @to, ignoring
4545 * all BdrvChild in @list so that they can't block themselves. */
4546 ret
= bdrv_check_update_perm(to
, NULL
, perm
, shared
, list
, NULL
, errp
);
4548 bdrv_abort_perm_update(to
);
4552 /* Now actually perform the change. We performed the permission check for
4553 * all elements of @list at once, so set the permissions all at once at the
4555 for (p
= list
; p
!= NULL
; p
= p
->next
) {
4559 bdrv_replace_child_noperm(c
, to
);
4563 bdrv_get_cumulative_perm(to
, &perm
, &shared
);
4564 bdrv_set_perm(to
, perm
, shared
);
4568 bdrv_drained_end(from
);
4573 * Add new bs contents at the top of an image chain while the chain is
4574 * live, while keeping required fields on the top layer.
4576 * This will modify the BlockDriverState fields, and swap contents
4577 * between bs_new and bs_top. Both bs_new and bs_top are modified.
4579 * bs_new must not be attached to a BlockBackend.
4581 * This function does not create any image files.
4583 * bdrv_append() takes ownership of a bs_new reference and unrefs it because
4584 * that's what the callers commonly need. bs_new will be referenced by the old
4585 * parents of bs_top after bdrv_append() returns. If the caller needs to keep a
4586 * reference of its own, it must call bdrv_ref().
4588 void bdrv_append(BlockDriverState
*bs_new
, BlockDriverState
*bs_top
,
4591 Error
*local_err
= NULL
;
4593 bdrv_set_backing_hd(bs_new
, bs_top
, &local_err
);
4595 error_propagate(errp
, local_err
);
4599 bdrv_replace_node(bs_top
, bs_new
, &local_err
);
4601 error_propagate(errp
, local_err
);
4602 bdrv_set_backing_hd(bs_new
, NULL
, &error_abort
);
4606 /* bs_new is now referenced by its new parents, we don't need the
4607 * additional reference any more. */
4612 static void bdrv_delete(BlockDriverState
*bs
)
4614 assert(bdrv_op_blocker_is_empty(bs
));
4615 assert(!bs
->refcnt
);
4617 /* remove from list, if necessary */
4618 if (bs
->node_name
[0] != '\0') {
4619 QTAILQ_REMOVE(&graph_bdrv_states
, bs
, node_list
);
4621 QTAILQ_REMOVE(&all_bdrv_states
, bs
, bs_list
);
4629 * Run consistency checks on an image
4631 * Returns 0 if the check could be completed (it doesn't mean that the image is
4632 * free of errors) or -errno when an internal error occurred. The results of the
4633 * check are stored in res.
4635 static int coroutine_fn
bdrv_co_check(BlockDriverState
*bs
,
4636 BdrvCheckResult
*res
, BdrvCheckMode fix
)
4638 if (bs
->drv
== NULL
) {
4641 if (bs
->drv
->bdrv_co_check
== NULL
) {
4645 memset(res
, 0, sizeof(*res
));
4646 return bs
->drv
->bdrv_co_check(bs
, res
, fix
);
4649 typedef struct CheckCo
{
4650 BlockDriverState
*bs
;
4651 BdrvCheckResult
*res
;
4656 static void coroutine_fn
bdrv_check_co_entry(void *opaque
)
4658 CheckCo
*cco
= opaque
;
4659 cco
->ret
= bdrv_co_check(cco
->bs
, cco
->res
, cco
->fix
);
4663 int bdrv_check(BlockDriverState
*bs
,
4664 BdrvCheckResult
*res
, BdrvCheckMode fix
)
4670 .ret
= -EINPROGRESS
,
4674 if (qemu_in_coroutine()) {
4675 /* Fast-path if already in coroutine context */
4676 bdrv_check_co_entry(&cco
);
4678 co
= qemu_coroutine_create(bdrv_check_co_entry
, &cco
);
4679 bdrv_coroutine_enter(bs
, co
);
4680 BDRV_POLL_WHILE(bs
, cco
.ret
== -EINPROGRESS
);
4689 * -EINVAL - backing format specified, but no file
4690 * -ENOSPC - can't update the backing file because no space is left in the
4692 * -ENOTSUP - format driver doesn't support changing the backing file
4694 int bdrv_change_backing_file(BlockDriverState
*bs
,
4695 const char *backing_file
, const char *backing_fmt
)
4697 BlockDriver
*drv
= bs
->drv
;
4704 /* Backing file format doesn't make sense without a backing file */
4705 if (backing_fmt
&& !backing_file
) {
4709 if (drv
->bdrv_change_backing_file
!= NULL
) {
4710 ret
= drv
->bdrv_change_backing_file(bs
, backing_file
, backing_fmt
);
4716 pstrcpy(bs
->backing_file
, sizeof(bs
->backing_file
), backing_file
?: "");
4717 pstrcpy(bs
->backing_format
, sizeof(bs
->backing_format
), backing_fmt
?: "");
4718 pstrcpy(bs
->auto_backing_file
, sizeof(bs
->auto_backing_file
),
4719 backing_file
?: "");
4725 * Finds the image layer in the chain that has 'bs' as its backing file.
4727 * active is the current topmost image.
4729 * Returns NULL if bs is not found in active's image chain,
4730 * or if active == bs.
4732 * Returns the bottommost base image if bs == NULL.
4734 BlockDriverState
*bdrv_find_overlay(BlockDriverState
*active
,
4735 BlockDriverState
*bs
)
4737 while (active
&& bs
!= backing_bs(active
)) {
4738 active
= backing_bs(active
);
4744 /* Given a BDS, searches for the base layer. */
4745 BlockDriverState
*bdrv_find_base(BlockDriverState
*bs
)
4747 return bdrv_find_overlay(bs
, NULL
);
4751 * Return true if at least one of the backing links between @bs and
4752 * @base is frozen. @errp is set if that's the case.
4753 * @base must be reachable from @bs, or NULL.
4755 bool bdrv_is_backing_chain_frozen(BlockDriverState
*bs
, BlockDriverState
*base
,
4758 BlockDriverState
*i
;
4760 for (i
= bs
; i
!= base
; i
= backing_bs(i
)) {
4761 if (i
->backing
&& i
->backing
->frozen
) {
4762 error_setg(errp
, "Cannot change '%s' link from '%s' to '%s'",
4763 i
->backing
->name
, i
->node_name
,
4764 backing_bs(i
)->node_name
);
4773 * Freeze all backing links between @bs and @base.
4774 * If any of the links is already frozen the operation is aborted and
4775 * none of the links are modified.
4776 * @base must be reachable from @bs, or NULL.
4777 * Returns 0 on success. On failure returns < 0 and sets @errp.
4779 int bdrv_freeze_backing_chain(BlockDriverState
*bs
, BlockDriverState
*base
,
4782 BlockDriverState
*i
;
4784 if (bdrv_is_backing_chain_frozen(bs
, base
, errp
)) {
4788 for (i
= bs
; i
!= base
; i
= backing_bs(i
)) {
4789 if (i
->backing
&& backing_bs(i
)->never_freeze
) {
4790 error_setg(errp
, "Cannot freeze '%s' link to '%s'",
4791 i
->backing
->name
, backing_bs(i
)->node_name
);
4796 for (i
= bs
; i
!= base
; i
= backing_bs(i
)) {
4798 i
->backing
->frozen
= true;
4806 * Unfreeze all backing links between @bs and @base. The caller must
4807 * ensure that all links are frozen before using this function.
4808 * @base must be reachable from @bs, or NULL.
4810 void bdrv_unfreeze_backing_chain(BlockDriverState
*bs
, BlockDriverState
*base
)
4812 BlockDriverState
*i
;
4814 for (i
= bs
; i
!= base
; i
= backing_bs(i
)) {
4816 assert(i
->backing
->frozen
);
4817 i
->backing
->frozen
= false;
4823 * Drops images above 'base' up to and including 'top', and sets the image
4824 * above 'top' to have base as its backing file.
4826 * Requires that the overlay to 'top' is opened r/w, so that the backing file
4827 * information in 'bs' can be properly updated.
4829 * E.g., this will convert the following chain:
4830 * bottom <- base <- intermediate <- top <- active
4834 * bottom <- base <- active
4836 * It is allowed for bottom==base, in which case it converts:
4838 * base <- intermediate <- top <- active
4844 * If backing_file_str is non-NULL, it will be used when modifying top's
4845 * overlay image metadata.
4848 * if active == top, that is considered an error
4851 int bdrv_drop_intermediate(BlockDriverState
*top
, BlockDriverState
*base
,
4852 const char *backing_file_str
)
4854 BlockDriverState
*explicit_top
= top
;
4855 bool update_inherits_from
;
4856 BdrvChild
*c
, *next
;
4857 Error
*local_err
= NULL
;
4861 bdrv_subtree_drained_begin(top
);
4863 if (!top
->drv
|| !base
->drv
) {
4867 /* Make sure that base is in the backing chain of top */
4868 if (!bdrv_chain_contains(top
, base
)) {
4872 /* This function changes all links that point to top and makes
4873 * them point to base. Check that none of them is frozen. */
4874 QLIST_FOREACH(c
, &top
->parents
, next_parent
) {
4880 /* If 'base' recursively inherits from 'top' then we should set
4881 * base->inherits_from to top->inherits_from after 'top' and all
4882 * other intermediate nodes have been dropped.
4883 * If 'top' is an implicit node (e.g. "commit_top") we should skip
4884 * it because no one inherits from it. We use explicit_top for that. */
4885 while (explicit_top
&& explicit_top
->implicit
) {
4886 explicit_top
= backing_bs(explicit_top
);
4888 update_inherits_from
= bdrv_inherits_from_recursive(base
, explicit_top
);
4890 /* success - we can delete the intermediate states, and link top->base */
4891 /* TODO Check graph modification op blockers (BLK_PERM_GRAPH_MOD) once
4892 * we've figured out how they should work. */
4893 if (!backing_file_str
) {
4894 bdrv_refresh_filename(base
);
4895 backing_file_str
= base
->filename
;
4898 QLIST_FOREACH_SAFE(c
, &top
->parents
, next_parent
, next
) {
4899 /* Check whether we are allowed to switch c from top to base */
4900 GSList
*ignore_children
= g_slist_prepend(NULL
, c
);
4901 ret
= bdrv_check_update_perm(base
, NULL
, c
->perm
, c
->shared_perm
,
4902 ignore_children
, NULL
, &local_err
);
4903 g_slist_free(ignore_children
);
4905 error_report_err(local_err
);
4909 /* If so, update the backing file path in the image file */
4910 if (c
->klass
->update_filename
) {
4911 ret
= c
->klass
->update_filename(c
, base
, backing_file_str
,
4914 bdrv_abort_perm_update(base
);
4915 error_report_err(local_err
);
4920 /* Do the actual switch in the in-memory graph.
4921 * Completes bdrv_check_update_perm() transaction internally. */
4923 bdrv_replace_child(c
, base
);
4927 if (update_inherits_from
) {
4928 base
->inherits_from
= explicit_top
->inherits_from
;
4933 bdrv_subtree_drained_end(top
);
4939 * Length of a allocated file in bytes. Sparse files are counted by actual
4940 * allocated space. Return < 0 if error or unknown.
4942 int64_t bdrv_get_allocated_file_size(BlockDriverState
*bs
)
4944 BlockDriver
*drv
= bs
->drv
;
4948 if (drv
->bdrv_get_allocated_file_size
) {
4949 return drv
->bdrv_get_allocated_file_size(bs
);
4952 return bdrv_get_allocated_file_size(bs
->file
->bs
);
4959 * @drv: Format driver
4960 * @opts: Creation options for new image
4961 * @in_bs: Existing image containing data for new image (may be NULL)
4962 * @errp: Error object
4963 * Returns: A #BlockMeasureInfo (free using qapi_free_BlockMeasureInfo())
4966 * Calculate file size required to create a new image.
4968 * If @in_bs is given then space for allocated clusters and zero clusters
4969 * from that image are included in the calculation. If @opts contains a
4970 * backing file that is shared by @in_bs then backing clusters may be omitted
4971 * from the calculation.
4973 * If @in_bs is NULL then the calculation includes no allocated clusters
4974 * unless a preallocation option is given in @opts.
4976 * Note that @in_bs may use a different BlockDriver from @drv.
4978 * If an error occurs the @errp pointer is set.
4980 BlockMeasureInfo
*bdrv_measure(BlockDriver
*drv
, QemuOpts
*opts
,
4981 BlockDriverState
*in_bs
, Error
**errp
)
4983 if (!drv
->bdrv_measure
) {
4984 error_setg(errp
, "Block driver '%s' does not support size measurement",
4989 return drv
->bdrv_measure(opts
, in_bs
, errp
);
4993 * Return number of sectors on success, -errno on error.
4995 int64_t bdrv_nb_sectors(BlockDriverState
*bs
)
4997 BlockDriver
*drv
= bs
->drv
;
5002 if (drv
->has_variable_length
) {
5003 int ret
= refresh_total_sectors(bs
, bs
->total_sectors
);
5008 return bs
->total_sectors
;
5012 * Return length in bytes on success, -errno on error.
5013 * The length is always a multiple of BDRV_SECTOR_SIZE.
5015 int64_t bdrv_getlength(BlockDriverState
*bs
)
5017 int64_t ret
= bdrv_nb_sectors(bs
);
5019 ret
= ret
> INT64_MAX
/ BDRV_SECTOR_SIZE
? -EFBIG
: ret
;
5020 return ret
< 0 ? ret
: ret
* BDRV_SECTOR_SIZE
;
5023 /* return 0 as number of sectors if no device present or error */
5024 void bdrv_get_geometry(BlockDriverState
*bs
, uint64_t *nb_sectors_ptr
)
5026 int64_t nb_sectors
= bdrv_nb_sectors(bs
);
5028 *nb_sectors_ptr
= nb_sectors
< 0 ? 0 : nb_sectors
;
5031 bool bdrv_is_sg(BlockDriverState
*bs
)
5036 bool bdrv_is_encrypted(BlockDriverState
*bs
)
5038 if (bs
->backing
&& bs
->backing
->bs
->encrypted
) {
5041 return bs
->encrypted
;
5044 const char *bdrv_get_format_name(BlockDriverState
*bs
)
5046 return bs
->drv
? bs
->drv
->format_name
: NULL
;
5049 static int qsort_strcmp(const void *a
, const void *b
)
5051 return strcmp(*(char *const *)a
, *(char *const *)b
);
5054 void bdrv_iterate_format(void (*it
)(void *opaque
, const char *name
),
5055 void *opaque
, bool read_only
)
5060 const char **formats
= NULL
;
5062 QLIST_FOREACH(drv
, &bdrv_drivers
, list
) {
5063 if (drv
->format_name
) {
5067 if (use_bdrv_whitelist
&& !bdrv_is_whitelisted(drv
, read_only
)) {
5071 while (formats
&& i
&& !found
) {
5072 found
= !strcmp(formats
[--i
], drv
->format_name
);
5076 formats
= g_renew(const char *, formats
, count
+ 1);
5077 formats
[count
++] = drv
->format_name
;
5082 for (i
= 0; i
< (int)ARRAY_SIZE(block_driver_modules
); i
++) {
5083 const char *format_name
= block_driver_modules
[i
].format_name
;
5089 if (use_bdrv_whitelist
&&
5090 !bdrv_format_is_whitelisted(format_name
, read_only
)) {
5094 while (formats
&& j
&& !found
) {
5095 found
= !strcmp(formats
[--j
], format_name
);
5099 formats
= g_renew(const char *, formats
, count
+ 1);
5100 formats
[count
++] = format_name
;
5105 qsort(formats
, count
, sizeof(formats
[0]), qsort_strcmp
);
5107 for (i
= 0; i
< count
; i
++) {
5108 it(opaque
, formats
[i
]);
5114 /* This function is to find a node in the bs graph */
5115 BlockDriverState
*bdrv_find_node(const char *node_name
)
5117 BlockDriverState
*bs
;
5121 QTAILQ_FOREACH(bs
, &graph_bdrv_states
, node_list
) {
5122 if (!strcmp(node_name
, bs
->node_name
)) {
5129 /* Put this QMP function here so it can access the static graph_bdrv_states. */
5130 BlockDeviceInfoList
*bdrv_named_nodes_list(bool flat
,
5133 BlockDeviceInfoList
*list
, *entry
;
5134 BlockDriverState
*bs
;
5137 QTAILQ_FOREACH(bs
, &graph_bdrv_states
, node_list
) {
5138 BlockDeviceInfo
*info
= bdrv_block_device_info(NULL
, bs
, flat
, errp
);
5140 qapi_free_BlockDeviceInfoList(list
);
5143 entry
= g_malloc0(sizeof(*entry
));
5144 entry
->value
= info
;
5152 #define QAPI_LIST_ADD(list, element) do { \
5153 typeof(list) _tmp = g_new(typeof(*(list)), 1); \
5154 _tmp->value = (element); \
5155 _tmp->next = (list); \
5159 typedef struct XDbgBlockGraphConstructor
{
5160 XDbgBlockGraph
*graph
;
5161 GHashTable
*graph_nodes
;
5162 } XDbgBlockGraphConstructor
;
5164 static XDbgBlockGraphConstructor
*xdbg_graph_new(void)
5166 XDbgBlockGraphConstructor
*gr
= g_new(XDbgBlockGraphConstructor
, 1);
5168 gr
->graph
= g_new0(XDbgBlockGraph
, 1);
5169 gr
->graph_nodes
= g_hash_table_new(NULL
, NULL
);
5174 static XDbgBlockGraph
*xdbg_graph_finalize(XDbgBlockGraphConstructor
*gr
)
5176 XDbgBlockGraph
*graph
= gr
->graph
;
5178 g_hash_table_destroy(gr
->graph_nodes
);
5184 static uintptr_t xdbg_graph_node_num(XDbgBlockGraphConstructor
*gr
, void *node
)
5186 uintptr_t ret
= (uintptr_t)g_hash_table_lookup(gr
->graph_nodes
, node
);
5193 * Start counting from 1, not 0, because 0 interferes with not-found (NULL)
5194 * answer of g_hash_table_lookup.
5196 ret
= g_hash_table_size(gr
->graph_nodes
) + 1;
5197 g_hash_table_insert(gr
->graph_nodes
, node
, (void *)ret
);
5202 static void xdbg_graph_add_node(XDbgBlockGraphConstructor
*gr
, void *node
,
5203 XDbgBlockGraphNodeType type
, const char *name
)
5205 XDbgBlockGraphNode
*n
;
5207 n
= g_new0(XDbgBlockGraphNode
, 1);
5209 n
->id
= xdbg_graph_node_num(gr
, node
);
5211 n
->name
= g_strdup(name
);
5213 QAPI_LIST_ADD(gr
->graph
->nodes
, n
);
5216 static void xdbg_graph_add_edge(XDbgBlockGraphConstructor
*gr
, void *parent
,
5217 const BdrvChild
*child
)
5219 BlockPermission qapi_perm
;
5220 XDbgBlockGraphEdge
*edge
;
5222 edge
= g_new0(XDbgBlockGraphEdge
, 1);
5224 edge
->parent
= xdbg_graph_node_num(gr
, parent
);
5225 edge
->child
= xdbg_graph_node_num(gr
, child
->bs
);
5226 edge
->name
= g_strdup(child
->name
);
5228 for (qapi_perm
= 0; qapi_perm
< BLOCK_PERMISSION__MAX
; qapi_perm
++) {
5229 uint64_t flag
= bdrv_qapi_perm_to_blk_perm(qapi_perm
);
5231 if (flag
& child
->perm
) {
5232 QAPI_LIST_ADD(edge
->perm
, qapi_perm
);
5234 if (flag
& child
->shared_perm
) {
5235 QAPI_LIST_ADD(edge
->shared_perm
, qapi_perm
);
5239 QAPI_LIST_ADD(gr
->graph
->edges
, edge
);
5243 XDbgBlockGraph
*bdrv_get_xdbg_block_graph(Error
**errp
)
5247 BlockDriverState
*bs
;
5249 XDbgBlockGraphConstructor
*gr
= xdbg_graph_new();
5251 for (blk
= blk_all_next(NULL
); blk
; blk
= blk_all_next(blk
)) {
5252 char *allocated_name
= NULL
;
5253 const char *name
= blk_name(blk
);
5256 name
= allocated_name
= blk_get_attached_dev_id(blk
);
5258 xdbg_graph_add_node(gr
, blk
, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_BACKEND
,
5260 g_free(allocated_name
);
5261 if (blk_root(blk
)) {
5262 xdbg_graph_add_edge(gr
, blk
, blk_root(blk
));
5266 for (job
= block_job_next(NULL
); job
; job
= block_job_next(job
)) {
5269 xdbg_graph_add_node(gr
, job
, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_JOB
,
5271 for (el
= job
->nodes
; el
; el
= el
->next
) {
5272 xdbg_graph_add_edge(gr
, job
, (BdrvChild
*)el
->data
);
5276 QTAILQ_FOREACH(bs
, &graph_bdrv_states
, node_list
) {
5277 xdbg_graph_add_node(gr
, bs
, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_DRIVER
,
5279 QLIST_FOREACH(child
, &bs
->children
, next
) {
5280 xdbg_graph_add_edge(gr
, bs
, child
);
5284 return xdbg_graph_finalize(gr
);
5287 BlockDriverState
*bdrv_lookup_bs(const char *device
,
5288 const char *node_name
,
5292 BlockDriverState
*bs
;
5295 blk
= blk_by_name(device
);
5300 error_setg(errp
, "Device '%s' has no medium", device
);
5308 bs
= bdrv_find_node(node_name
);
5315 error_setg(errp
, "Cannot find device=%s nor node_name=%s",
5316 device
? device
: "",
5317 node_name
? node_name
: "");
5321 /* If 'base' is in the same chain as 'top', return true. Otherwise,
5322 * return false. If either argument is NULL, return false. */
5323 bool bdrv_chain_contains(BlockDriverState
*top
, BlockDriverState
*base
)
5325 while (top
&& top
!= base
) {
5326 top
= backing_bs(top
);
5332 BlockDriverState
*bdrv_next_node(BlockDriverState
*bs
)
5335 return QTAILQ_FIRST(&graph_bdrv_states
);
5337 return QTAILQ_NEXT(bs
, node_list
);
5340 BlockDriverState
*bdrv_next_all_states(BlockDriverState
*bs
)
5343 return QTAILQ_FIRST(&all_bdrv_states
);
5345 return QTAILQ_NEXT(bs
, bs_list
);
5348 const char *bdrv_get_node_name(const BlockDriverState
*bs
)
5350 return bs
->node_name
;
5353 const char *bdrv_get_parent_name(const BlockDriverState
*bs
)
5358 /* If multiple parents have a name, just pick the first one. */
5359 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
5360 if (c
->klass
->get_name
) {
5361 name
= c
->klass
->get_name(c
);
5362 if (name
&& *name
) {
5371 /* TODO check what callers really want: bs->node_name or blk_name() */
5372 const char *bdrv_get_device_name(const BlockDriverState
*bs
)
5374 return bdrv_get_parent_name(bs
) ?: "";
5377 /* This can be used to identify nodes that might not have a device
5378 * name associated. Since node and device names live in the same
5379 * namespace, the result is unambiguous. The exception is if both are
5380 * absent, then this returns an empty (non-null) string. */
5381 const char *bdrv_get_device_or_node_name(const BlockDriverState
*bs
)
5383 return bdrv_get_parent_name(bs
) ?: bs
->node_name
;
5386 int bdrv_get_flags(BlockDriverState
*bs
)
5388 return bs
->open_flags
;
5391 int bdrv_has_zero_init_1(BlockDriverState
*bs
)
5396 int bdrv_has_zero_init(BlockDriverState
*bs
)
5402 /* If BS is a copy on write image, it is initialized to
5403 the contents of the base image, which may not be zeroes. */
5407 if (bs
->drv
->bdrv_has_zero_init
) {
5408 return bs
->drv
->bdrv_has_zero_init(bs
);
5410 if (bs
->file
&& bs
->drv
->is_filter
) {
5411 return bdrv_has_zero_init(bs
->file
->bs
);
5418 bool bdrv_unallocated_blocks_are_zero(BlockDriverState
*bs
)
5420 BlockDriverInfo bdi
;
5426 if (bdrv_get_info(bs
, &bdi
) == 0) {
5427 return bdi
.unallocated_blocks_are_zero
;
5433 bool bdrv_can_write_zeroes_with_unmap(BlockDriverState
*bs
)
5435 if (!(bs
->open_flags
& BDRV_O_UNMAP
)) {
5439 return bs
->supported_zero_flags
& BDRV_REQ_MAY_UNMAP
;
5442 void bdrv_get_backing_filename(BlockDriverState
*bs
,
5443 char *filename
, int filename_size
)
5445 pstrcpy(filename
, filename_size
, bs
->backing_file
);
5448 int bdrv_get_info(BlockDriverState
*bs
, BlockDriverInfo
*bdi
)
5450 BlockDriver
*drv
= bs
->drv
;
5451 /* if bs->drv == NULL, bs is closed, so there's nothing to do here */
5455 if (!drv
->bdrv_get_info
) {
5456 if (bs
->file
&& drv
->is_filter
) {
5457 return bdrv_get_info(bs
->file
->bs
, bdi
);
5461 memset(bdi
, 0, sizeof(*bdi
));
5462 return drv
->bdrv_get_info(bs
, bdi
);
5465 ImageInfoSpecific
*bdrv_get_specific_info(BlockDriverState
*bs
,
5468 BlockDriver
*drv
= bs
->drv
;
5469 if (drv
&& drv
->bdrv_get_specific_info
) {
5470 return drv
->bdrv_get_specific_info(bs
, errp
);
5475 BlockStatsSpecific
*bdrv_get_specific_stats(BlockDriverState
*bs
)
5477 BlockDriver
*drv
= bs
->drv
;
5478 if (!drv
|| !drv
->bdrv_get_specific_stats
) {
5481 return drv
->bdrv_get_specific_stats(bs
);
5484 void bdrv_debug_event(BlockDriverState
*bs
, BlkdebugEvent event
)
5486 if (!bs
|| !bs
->drv
|| !bs
->drv
->bdrv_debug_event
) {
5490 bs
->drv
->bdrv_debug_event(bs
, event
);
5493 static BlockDriverState
*bdrv_find_debug_node(BlockDriverState
*bs
)
5495 while (bs
&& bs
->drv
&& !bs
->drv
->bdrv_debug_breakpoint
) {
5501 if (bs
->drv
->is_filter
&& bs
->backing
) {
5502 bs
= bs
->backing
->bs
;
5509 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_breakpoint
) {
5510 assert(bs
->drv
->bdrv_debug_remove_breakpoint
);
5517 int bdrv_debug_breakpoint(BlockDriverState
*bs
, const char *event
,
5520 bs
= bdrv_find_debug_node(bs
);
5522 return bs
->drv
->bdrv_debug_breakpoint(bs
, event
, tag
);
5528 int bdrv_debug_remove_breakpoint(BlockDriverState
*bs
, const char *tag
)
5530 bs
= bdrv_find_debug_node(bs
);
5532 return bs
->drv
->bdrv_debug_remove_breakpoint(bs
, tag
);
5538 int bdrv_debug_resume(BlockDriverState
*bs
, const char *tag
)
5540 while (bs
&& (!bs
->drv
|| !bs
->drv
->bdrv_debug_resume
)) {
5541 bs
= bs
->file
? bs
->file
->bs
: NULL
;
5544 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_resume
) {
5545 return bs
->drv
->bdrv_debug_resume(bs
, tag
);
5551 bool bdrv_debug_is_suspended(BlockDriverState
*bs
, const char *tag
)
5553 while (bs
&& bs
->drv
&& !bs
->drv
->bdrv_debug_is_suspended
) {
5554 bs
= bs
->file
? bs
->file
->bs
: NULL
;
5557 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_is_suspended
) {
5558 return bs
->drv
->bdrv_debug_is_suspended(bs
, tag
);
5564 /* backing_file can either be relative, or absolute, or a protocol. If it is
5565 * relative, it must be relative to the chain. So, passing in bs->filename
5566 * from a BDS as backing_file should not be done, as that may be relative to
5567 * the CWD rather than the chain. */
5568 BlockDriverState
*bdrv_find_backing_image(BlockDriverState
*bs
,
5569 const char *backing_file
)
5571 char *filename_full
= NULL
;
5572 char *backing_file_full
= NULL
;
5573 char *filename_tmp
= NULL
;
5574 int is_protocol
= 0;
5575 BlockDriverState
*curr_bs
= NULL
;
5576 BlockDriverState
*retval
= NULL
;
5578 if (!bs
|| !bs
->drv
|| !backing_file
) {
5582 filename_full
= g_malloc(PATH_MAX
);
5583 backing_file_full
= g_malloc(PATH_MAX
);
5585 is_protocol
= path_has_protocol(backing_file
);
5587 for (curr_bs
= bs
; curr_bs
->backing
; curr_bs
= curr_bs
->backing
->bs
) {
5589 /* If either of the filename paths is actually a protocol, then
5590 * compare unmodified paths; otherwise make paths relative */
5591 if (is_protocol
|| path_has_protocol(curr_bs
->backing_file
)) {
5592 char *backing_file_full_ret
;
5594 if (strcmp(backing_file
, curr_bs
->backing_file
) == 0) {
5595 retval
= curr_bs
->backing
->bs
;
5598 /* Also check against the full backing filename for the image */
5599 backing_file_full_ret
= bdrv_get_full_backing_filename(curr_bs
,
5601 if (backing_file_full_ret
) {
5602 bool equal
= strcmp(backing_file
, backing_file_full_ret
) == 0;
5603 g_free(backing_file_full_ret
);
5605 retval
= curr_bs
->backing
->bs
;
5610 /* If not an absolute filename path, make it relative to the current
5611 * image's filename path */
5612 filename_tmp
= bdrv_make_absolute_filename(curr_bs
, backing_file
,
5614 /* We are going to compare canonicalized absolute pathnames */
5615 if (!filename_tmp
|| !realpath(filename_tmp
, filename_full
)) {
5616 g_free(filename_tmp
);
5619 g_free(filename_tmp
);
5621 /* We need to make sure the backing filename we are comparing against
5622 * is relative to the current image filename (or absolute) */
5623 filename_tmp
= bdrv_get_full_backing_filename(curr_bs
, NULL
);
5624 if (!filename_tmp
|| !realpath(filename_tmp
, backing_file_full
)) {
5625 g_free(filename_tmp
);
5628 g_free(filename_tmp
);
5630 if (strcmp(backing_file_full
, filename_full
) == 0) {
5631 retval
= curr_bs
->backing
->bs
;
5637 g_free(filename_full
);
5638 g_free(backing_file_full
);
5642 void bdrv_init(void)
5644 module_call_init(MODULE_INIT_BLOCK
);
5647 void bdrv_init_with_whitelist(void)
5649 use_bdrv_whitelist
= 1;
5653 static void coroutine_fn
bdrv_co_invalidate_cache(BlockDriverState
*bs
,
5656 BdrvChild
*child
, *parent
;
5657 uint64_t perm
, shared_perm
;
5658 Error
*local_err
= NULL
;
5660 BdrvDirtyBitmap
*bm
;
5666 QLIST_FOREACH(child
, &bs
->children
, next
) {
5667 bdrv_co_invalidate_cache(child
->bs
, &local_err
);
5669 error_propagate(errp
, local_err
);
5675 * Update permissions, they may differ for inactive nodes.
5677 * Note that the required permissions of inactive images are always a
5678 * subset of the permissions required after activating the image. This
5679 * allows us to just get the permissions upfront without restricting
5680 * drv->bdrv_invalidate_cache().
5682 * It also means that in error cases, we don't have to try and revert to
5683 * the old permissions (which is an operation that could fail, too). We can
5684 * just keep the extended permissions for the next time that an activation
5685 * of the image is tried.
5687 if (bs
->open_flags
& BDRV_O_INACTIVE
) {
5688 bs
->open_flags
&= ~BDRV_O_INACTIVE
;
5689 bdrv_get_cumulative_perm(bs
, &perm
, &shared_perm
);
5690 ret
= bdrv_check_perm(bs
, NULL
, perm
, shared_perm
, NULL
, NULL
, &local_err
);
5692 bs
->open_flags
|= BDRV_O_INACTIVE
;
5693 error_propagate(errp
, local_err
);
5696 bdrv_set_perm(bs
, perm
, shared_perm
);
5698 if (bs
->drv
->bdrv_co_invalidate_cache
) {
5699 bs
->drv
->bdrv_co_invalidate_cache(bs
, &local_err
);
5701 bs
->open_flags
|= BDRV_O_INACTIVE
;
5702 error_propagate(errp
, local_err
);
5707 FOR_EACH_DIRTY_BITMAP(bs
, bm
) {
5708 bdrv_dirty_bitmap_skip_store(bm
, false);
5711 ret
= refresh_total_sectors(bs
, bs
->total_sectors
);
5713 bs
->open_flags
|= BDRV_O_INACTIVE
;
5714 error_setg_errno(errp
, -ret
, "Could not refresh total sector count");
5719 QLIST_FOREACH(parent
, &bs
->parents
, next_parent
) {
5720 if (parent
->klass
->activate
) {
5721 parent
->klass
->activate(parent
, &local_err
);
5723 bs
->open_flags
|= BDRV_O_INACTIVE
;
5724 error_propagate(errp
, local_err
);
5731 typedef struct InvalidateCacheCo
{
5732 BlockDriverState
*bs
;
5735 } InvalidateCacheCo
;
5737 static void coroutine_fn
bdrv_invalidate_cache_co_entry(void *opaque
)
5739 InvalidateCacheCo
*ico
= opaque
;
5740 bdrv_co_invalidate_cache(ico
->bs
, ico
->errp
);
5745 void bdrv_invalidate_cache(BlockDriverState
*bs
, Error
**errp
)
5748 InvalidateCacheCo ico
= {
5754 if (qemu_in_coroutine()) {
5755 /* Fast-path if already in coroutine context */
5756 bdrv_invalidate_cache_co_entry(&ico
);
5758 co
= qemu_coroutine_create(bdrv_invalidate_cache_co_entry
, &ico
);
5759 bdrv_coroutine_enter(bs
, co
);
5760 BDRV_POLL_WHILE(bs
, !ico
.done
);
5764 void bdrv_invalidate_cache_all(Error
**errp
)
5766 BlockDriverState
*bs
;
5767 Error
*local_err
= NULL
;
5768 BdrvNextIterator it
;
5770 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
5771 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
5773 aio_context_acquire(aio_context
);
5774 bdrv_invalidate_cache(bs
, &local_err
);
5775 aio_context_release(aio_context
);
5777 error_propagate(errp
, local_err
);
5778 bdrv_next_cleanup(&it
);
5784 static bool bdrv_has_bds_parent(BlockDriverState
*bs
, bool only_active
)
5788 QLIST_FOREACH(parent
, &bs
->parents
, next_parent
) {
5789 if (parent
->klass
->parent_is_bds
) {
5790 BlockDriverState
*parent_bs
= parent
->opaque
;
5791 if (!only_active
|| !(parent_bs
->open_flags
& BDRV_O_INACTIVE
)) {
5800 static int bdrv_inactivate_recurse(BlockDriverState
*bs
)
5802 BdrvChild
*child
, *parent
;
5803 bool tighten_restrictions
;
5804 uint64_t perm
, shared_perm
;
5811 /* Make sure that we don't inactivate a child before its parent.
5812 * It will be covered by recursion from the yet active parent. */
5813 if (bdrv_has_bds_parent(bs
, true)) {
5817 assert(!(bs
->open_flags
& BDRV_O_INACTIVE
));
5819 /* Inactivate this node */
5820 if (bs
->drv
->bdrv_inactivate
) {
5821 ret
= bs
->drv
->bdrv_inactivate(bs
);
5827 QLIST_FOREACH(parent
, &bs
->parents
, next_parent
) {
5828 if (parent
->klass
->inactivate
) {
5829 ret
= parent
->klass
->inactivate(parent
);
5836 bs
->open_flags
|= BDRV_O_INACTIVE
;
5838 /* Update permissions, they may differ for inactive nodes */
5839 bdrv_get_cumulative_perm(bs
, &perm
, &shared_perm
);
5840 ret
= bdrv_check_perm(bs
, NULL
, perm
, shared_perm
, NULL
,
5841 &tighten_restrictions
, NULL
);
5842 assert(tighten_restrictions
== false);
5844 /* We only tried to loosen restrictions, so errors are not fatal */
5845 bdrv_abort_perm_update(bs
);
5847 bdrv_set_perm(bs
, perm
, shared_perm
);
5851 /* Recursively inactivate children */
5852 QLIST_FOREACH(child
, &bs
->children
, next
) {
5853 ret
= bdrv_inactivate_recurse(child
->bs
);
5862 int bdrv_inactivate_all(void)
5864 BlockDriverState
*bs
= NULL
;
5865 BdrvNextIterator it
;
5867 GSList
*aio_ctxs
= NULL
, *ctx
;
5869 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
5870 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
5872 if (!g_slist_find(aio_ctxs
, aio_context
)) {
5873 aio_ctxs
= g_slist_prepend(aio_ctxs
, aio_context
);
5874 aio_context_acquire(aio_context
);
5878 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
5879 /* Nodes with BDS parents are covered by recursion from the last
5880 * parent that gets inactivated. Don't inactivate them a second
5881 * time if that has already happened. */
5882 if (bdrv_has_bds_parent(bs
, false)) {
5885 ret
= bdrv_inactivate_recurse(bs
);
5887 bdrv_next_cleanup(&it
);
5893 for (ctx
= aio_ctxs
; ctx
!= NULL
; ctx
= ctx
->next
) {
5894 AioContext
*aio_context
= ctx
->data
;
5895 aio_context_release(aio_context
);
5897 g_slist_free(aio_ctxs
);
5902 /**************************************************************/
5903 /* removable device support */
5906 * Return TRUE if the media is present
5908 bool bdrv_is_inserted(BlockDriverState
*bs
)
5910 BlockDriver
*drv
= bs
->drv
;
5916 if (drv
->bdrv_is_inserted
) {
5917 return drv
->bdrv_is_inserted(bs
);
5919 QLIST_FOREACH(child
, &bs
->children
, next
) {
5920 if (!bdrv_is_inserted(child
->bs
)) {
5928 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
5930 void bdrv_eject(BlockDriverState
*bs
, bool eject_flag
)
5932 BlockDriver
*drv
= bs
->drv
;
5934 if (drv
&& drv
->bdrv_eject
) {
5935 drv
->bdrv_eject(bs
, eject_flag
);
5940 * Lock or unlock the media (if it is locked, the user won't be able
5941 * to eject it manually).
5943 void bdrv_lock_medium(BlockDriverState
*bs
, bool locked
)
5945 BlockDriver
*drv
= bs
->drv
;
5947 trace_bdrv_lock_medium(bs
, locked
);
5949 if (drv
&& drv
->bdrv_lock_medium
) {
5950 drv
->bdrv_lock_medium(bs
, locked
);
5954 /* Get a reference to bs */
5955 void bdrv_ref(BlockDriverState
*bs
)
5960 /* Release a previously grabbed reference to bs.
5961 * If after releasing, reference count is zero, the BlockDriverState is
5963 void bdrv_unref(BlockDriverState
*bs
)
5968 assert(bs
->refcnt
> 0);
5969 if (--bs
->refcnt
== 0) {
5974 struct BdrvOpBlocker
{
5976 QLIST_ENTRY(BdrvOpBlocker
) list
;
5979 bool bdrv_op_is_blocked(BlockDriverState
*bs
, BlockOpType op
, Error
**errp
)
5981 BdrvOpBlocker
*blocker
;
5982 assert((int) op
>= 0 && op
< BLOCK_OP_TYPE_MAX
);
5983 if (!QLIST_EMPTY(&bs
->op_blockers
[op
])) {
5984 blocker
= QLIST_FIRST(&bs
->op_blockers
[op
]);
5985 error_propagate_prepend(errp
, error_copy(blocker
->reason
),
5986 "Node '%s' is busy: ",
5987 bdrv_get_device_or_node_name(bs
));
5993 void bdrv_op_block(BlockDriverState
*bs
, BlockOpType op
, Error
*reason
)
5995 BdrvOpBlocker
*blocker
;
5996 assert((int) op
>= 0 && op
< BLOCK_OP_TYPE_MAX
);
5998 blocker
= g_new0(BdrvOpBlocker
, 1);
5999 blocker
->reason
= reason
;
6000 QLIST_INSERT_HEAD(&bs
->op_blockers
[op
], blocker
, list
);
6003 void bdrv_op_unblock(BlockDriverState
*bs
, BlockOpType op
, Error
*reason
)
6005 BdrvOpBlocker
*blocker
, *next
;
6006 assert((int) op
>= 0 && op
< BLOCK_OP_TYPE_MAX
);
6007 QLIST_FOREACH_SAFE(blocker
, &bs
->op_blockers
[op
], list
, next
) {
6008 if (blocker
->reason
== reason
) {
6009 QLIST_REMOVE(blocker
, list
);
6015 void bdrv_op_block_all(BlockDriverState
*bs
, Error
*reason
)
6018 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
6019 bdrv_op_block(bs
, i
, reason
);
6023 void bdrv_op_unblock_all(BlockDriverState
*bs
, Error
*reason
)
6026 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
6027 bdrv_op_unblock(bs
, i
, reason
);
6031 bool bdrv_op_blocker_is_empty(BlockDriverState
*bs
)
6035 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
6036 if (!QLIST_EMPTY(&bs
->op_blockers
[i
])) {
6043 void bdrv_img_create(const char *filename
, const char *fmt
,
6044 const char *base_filename
, const char *base_fmt
,
6045 char *options
, uint64_t img_size
, int flags
, bool quiet
,
6048 QemuOptsList
*create_opts
= NULL
;
6049 QemuOpts
*opts
= NULL
;
6050 const char *backing_fmt
, *backing_file
;
6052 BlockDriver
*drv
, *proto_drv
;
6053 Error
*local_err
= NULL
;
6056 /* Find driver and parse its options */
6057 drv
= bdrv_find_format(fmt
);
6059 error_setg(errp
, "Unknown file format '%s'", fmt
);
6063 proto_drv
= bdrv_find_protocol(filename
, true, errp
);
6068 if (!drv
->create_opts
) {
6069 error_setg(errp
, "Format driver '%s' does not support image creation",
6074 if (!proto_drv
->create_opts
) {
6075 error_setg(errp
, "Protocol driver '%s' does not support image creation",
6076 proto_drv
->format_name
);
6080 /* Create parameter list */
6081 create_opts
= qemu_opts_append(create_opts
, drv
->create_opts
);
6082 create_opts
= qemu_opts_append(create_opts
, proto_drv
->create_opts
);
6084 opts
= qemu_opts_create(create_opts
, NULL
, 0, &error_abort
);
6086 /* Parse -o options */
6088 qemu_opts_do_parse(opts
, options
, NULL
, &local_err
);
6094 if (!qemu_opt_get(opts
, BLOCK_OPT_SIZE
)) {
6095 qemu_opt_set_number(opts
, BLOCK_OPT_SIZE
, img_size
, &error_abort
);
6096 } else if (img_size
!= UINT64_C(-1)) {
6097 error_setg(errp
, "The image size must be specified only once");
6101 if (base_filename
) {
6102 qemu_opt_set(opts
, BLOCK_OPT_BACKING_FILE
, base_filename
, &local_err
);
6104 error_setg(errp
, "Backing file not supported for file format '%s'",
6111 qemu_opt_set(opts
, BLOCK_OPT_BACKING_FMT
, base_fmt
, &local_err
);
6113 error_setg(errp
, "Backing file format not supported for file "
6114 "format '%s'", fmt
);
6119 backing_file
= qemu_opt_get(opts
, BLOCK_OPT_BACKING_FILE
);
6121 if (!strcmp(filename
, backing_file
)) {
6122 error_setg(errp
, "Error: Trying to create an image with the "
6123 "same filename as the backing file");
6128 backing_fmt
= qemu_opt_get(opts
, BLOCK_OPT_BACKING_FMT
);
6130 /* The size for the image must always be specified, unless we have a backing
6131 * file and we have not been forbidden from opening it. */
6132 size
= qemu_opt_get_size(opts
, BLOCK_OPT_SIZE
, img_size
);
6133 if (backing_file
&& !(flags
& BDRV_O_NO_BACKING
)) {
6134 BlockDriverState
*bs
;
6137 QDict
*backing_options
= NULL
;
6140 bdrv_get_full_backing_filename_from_filename(filename
, backing_file
,
6145 assert(full_backing
);
6147 /* backing files always opened read-only */
6149 back_flags
&= ~(BDRV_O_RDWR
| BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
);
6151 backing_options
= qdict_new();
6153 qdict_put_str(backing_options
, "driver", backing_fmt
);
6155 qdict_put_bool(backing_options
, BDRV_OPT_FORCE_SHARE
, true);
6157 bs
= bdrv_open(full_backing
, NULL
, backing_options
, back_flags
,
6159 g_free(full_backing
);
6160 if (!bs
&& size
!= -1) {
6161 /* Couldn't open BS, but we have a size, so it's nonfatal */
6162 warn_reportf_err(local_err
,
6163 "Could not verify backing image. "
6164 "This may become an error in future versions.\n");
6167 /* Couldn't open bs, do not have size */
6168 error_append_hint(&local_err
,
6169 "Could not open backing image to determine size.\n");
6173 /* Opened BS, have no size */
6174 size
= bdrv_getlength(bs
);
6176 error_setg_errno(errp
, -size
, "Could not get size of '%s'",
6181 qemu_opt_set_number(opts
, BLOCK_OPT_SIZE
, size
, &error_abort
);
6185 } /* (backing_file && !(flags & BDRV_O_NO_BACKING)) */
6188 error_setg(errp
, "Image creation needs a size parameter");
6193 printf("Formatting '%s', fmt=%s ", filename
, fmt
);
6194 qemu_opts_print(opts
, " ");
6198 ret
= bdrv_create(drv
, filename
, opts
, &local_err
);
6200 if (ret
== -EFBIG
) {
6201 /* This is generally a better message than whatever the driver would
6202 * deliver (especially because of the cluster_size_hint), since that
6203 * is most probably not much different from "image too large". */
6204 const char *cluster_size_hint
= "";
6205 if (qemu_opt_get_size(opts
, BLOCK_OPT_CLUSTER_SIZE
, 0)) {
6206 cluster_size_hint
= " (try using a larger cluster size)";
6208 error_setg(errp
, "The image size is too large for file format '%s'"
6209 "%s", fmt
, cluster_size_hint
);
6210 error_free(local_err
);
6215 qemu_opts_del(opts
);
6216 qemu_opts_free(create_opts
);
6217 error_propagate(errp
, local_err
);
6220 AioContext
*bdrv_get_aio_context(BlockDriverState
*bs
)
6222 return bs
? bs
->aio_context
: qemu_get_aio_context();
6225 void bdrv_coroutine_enter(BlockDriverState
*bs
, Coroutine
*co
)
6227 aio_co_enter(bdrv_get_aio_context(bs
), co
);
6230 static void bdrv_do_remove_aio_context_notifier(BdrvAioNotifier
*ban
)
6232 QLIST_REMOVE(ban
, list
);
6236 static void bdrv_detach_aio_context(BlockDriverState
*bs
)
6238 BdrvAioNotifier
*baf
, *baf_tmp
;
6240 assert(!bs
->walking_aio_notifiers
);
6241 bs
->walking_aio_notifiers
= true;
6242 QLIST_FOREACH_SAFE(baf
, &bs
->aio_notifiers
, list
, baf_tmp
) {
6244 bdrv_do_remove_aio_context_notifier(baf
);
6246 baf
->detach_aio_context(baf
->opaque
);
6249 /* Never mind iterating again to check for ->deleted. bdrv_close() will
6250 * remove remaining aio notifiers if we aren't called again.
6252 bs
->walking_aio_notifiers
= false;
6254 if (bs
->drv
&& bs
->drv
->bdrv_detach_aio_context
) {
6255 bs
->drv
->bdrv_detach_aio_context(bs
);
6258 if (bs
->quiesce_counter
) {
6259 aio_enable_external(bs
->aio_context
);
6261 bs
->aio_context
= NULL
;
6264 static void bdrv_attach_aio_context(BlockDriverState
*bs
,
6265 AioContext
*new_context
)
6267 BdrvAioNotifier
*ban
, *ban_tmp
;
6269 if (bs
->quiesce_counter
) {
6270 aio_disable_external(new_context
);
6273 bs
->aio_context
= new_context
;
6275 if (bs
->drv
&& bs
->drv
->bdrv_attach_aio_context
) {
6276 bs
->drv
->bdrv_attach_aio_context(bs
, new_context
);
6279 assert(!bs
->walking_aio_notifiers
);
6280 bs
->walking_aio_notifiers
= true;
6281 QLIST_FOREACH_SAFE(ban
, &bs
->aio_notifiers
, list
, ban_tmp
) {
6283 bdrv_do_remove_aio_context_notifier(ban
);
6285 ban
->attached_aio_context(new_context
, ban
->opaque
);
6288 bs
->walking_aio_notifiers
= false;
6292 * Changes the AioContext used for fd handlers, timers, and BHs by this
6293 * BlockDriverState and all its children and parents.
6295 * Must be called from the main AioContext.
6297 * The caller must own the AioContext lock for the old AioContext of bs, but it
6298 * must not own the AioContext lock for new_context (unless new_context is the
6299 * same as the current context of bs).
6301 * @ignore will accumulate all visited BdrvChild object. The caller is
6302 * responsible for freeing the list afterwards.
6304 void bdrv_set_aio_context_ignore(BlockDriverState
*bs
,
6305 AioContext
*new_context
, GSList
**ignore
)
6307 AioContext
*old_context
= bdrv_get_aio_context(bs
);
6310 g_assert(qemu_get_current_aio_context() == qemu_get_aio_context());
6312 if (old_context
== new_context
) {
6316 bdrv_drained_begin(bs
);
6318 QLIST_FOREACH(child
, &bs
->children
, next
) {
6319 if (g_slist_find(*ignore
, child
)) {
6322 *ignore
= g_slist_prepend(*ignore
, child
);
6323 bdrv_set_aio_context_ignore(child
->bs
, new_context
, ignore
);
6325 QLIST_FOREACH(child
, &bs
->parents
, next_parent
) {
6326 if (g_slist_find(*ignore
, child
)) {
6329 assert(child
->klass
->set_aio_ctx
);
6330 *ignore
= g_slist_prepend(*ignore
, child
);
6331 child
->klass
->set_aio_ctx(child
, new_context
, ignore
);
6334 bdrv_detach_aio_context(bs
);
6336 /* Acquire the new context, if necessary */
6337 if (qemu_get_aio_context() != new_context
) {
6338 aio_context_acquire(new_context
);
6341 bdrv_attach_aio_context(bs
, new_context
);
6344 * If this function was recursively called from
6345 * bdrv_set_aio_context_ignore(), there may be nodes in the
6346 * subtree that have not yet been moved to the new AioContext.
6347 * Release the old one so bdrv_drained_end() can poll them.
6349 if (qemu_get_aio_context() != old_context
) {
6350 aio_context_release(old_context
);
6353 bdrv_drained_end(bs
);
6355 if (qemu_get_aio_context() != old_context
) {
6356 aio_context_acquire(old_context
);
6358 if (qemu_get_aio_context() != new_context
) {
6359 aio_context_release(new_context
);
6363 static bool bdrv_parent_can_set_aio_context(BdrvChild
*c
, AioContext
*ctx
,
6364 GSList
**ignore
, Error
**errp
)
6366 if (g_slist_find(*ignore
, c
)) {
6369 *ignore
= g_slist_prepend(*ignore
, c
);
6372 * A BdrvChildClass that doesn't handle AioContext changes cannot
6373 * tolerate any AioContext changes
6375 if (!c
->klass
->can_set_aio_ctx
) {
6376 char *user
= bdrv_child_user_desc(c
);
6377 error_setg(errp
, "Changing iothreads is not supported by %s", user
);
6381 if (!c
->klass
->can_set_aio_ctx(c
, ctx
, ignore
, errp
)) {
6382 assert(!errp
|| *errp
);
6388 bool bdrv_child_can_set_aio_context(BdrvChild
*c
, AioContext
*ctx
,
6389 GSList
**ignore
, Error
**errp
)
6391 if (g_slist_find(*ignore
, c
)) {
6394 *ignore
= g_slist_prepend(*ignore
, c
);
6395 return bdrv_can_set_aio_context(c
->bs
, ctx
, ignore
, errp
);
6398 /* @ignore will accumulate all visited BdrvChild object. The caller is
6399 * responsible for freeing the list afterwards. */
6400 bool bdrv_can_set_aio_context(BlockDriverState
*bs
, AioContext
*ctx
,
6401 GSList
**ignore
, Error
**errp
)
6405 if (bdrv_get_aio_context(bs
) == ctx
) {
6409 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
6410 if (!bdrv_parent_can_set_aio_context(c
, ctx
, ignore
, errp
)) {
6414 QLIST_FOREACH(c
, &bs
->children
, next
) {
6415 if (!bdrv_child_can_set_aio_context(c
, ctx
, ignore
, errp
)) {
6423 int bdrv_child_try_set_aio_context(BlockDriverState
*bs
, AioContext
*ctx
,
6424 BdrvChild
*ignore_child
, Error
**errp
)
6429 ignore
= ignore_child
? g_slist_prepend(NULL
, ignore_child
) : NULL
;
6430 ret
= bdrv_can_set_aio_context(bs
, ctx
, &ignore
, errp
);
6431 g_slist_free(ignore
);
6437 ignore
= ignore_child
? g_slist_prepend(NULL
, ignore_child
) : NULL
;
6438 bdrv_set_aio_context_ignore(bs
, ctx
, &ignore
);
6439 g_slist_free(ignore
);
6444 int bdrv_try_set_aio_context(BlockDriverState
*bs
, AioContext
*ctx
,
6447 return bdrv_child_try_set_aio_context(bs
, ctx
, NULL
, errp
);
6450 void bdrv_add_aio_context_notifier(BlockDriverState
*bs
,
6451 void (*attached_aio_context
)(AioContext
*new_context
, void *opaque
),
6452 void (*detach_aio_context
)(void *opaque
), void *opaque
)
6454 BdrvAioNotifier
*ban
= g_new(BdrvAioNotifier
, 1);
6455 *ban
= (BdrvAioNotifier
){
6456 .attached_aio_context
= attached_aio_context
,
6457 .detach_aio_context
= detach_aio_context
,
6461 QLIST_INSERT_HEAD(&bs
->aio_notifiers
, ban
, list
);
6464 void bdrv_remove_aio_context_notifier(BlockDriverState
*bs
,
6465 void (*attached_aio_context
)(AioContext
*,
6467 void (*detach_aio_context
)(void *),
6470 BdrvAioNotifier
*ban
, *ban_next
;
6472 QLIST_FOREACH_SAFE(ban
, &bs
->aio_notifiers
, list
, ban_next
) {
6473 if (ban
->attached_aio_context
== attached_aio_context
&&
6474 ban
->detach_aio_context
== detach_aio_context
&&
6475 ban
->opaque
== opaque
&&
6476 ban
->deleted
== false)
6478 if (bs
->walking_aio_notifiers
) {
6479 ban
->deleted
= true;
6481 bdrv_do_remove_aio_context_notifier(ban
);
6490 int bdrv_amend_options(BlockDriverState
*bs
, QemuOpts
*opts
,
6491 BlockDriverAmendStatusCB
*status_cb
, void *cb_opaque
,
6495 error_setg(errp
, "Node is ejected");
6498 if (!bs
->drv
->bdrv_amend_options
) {
6499 error_setg(errp
, "Block driver '%s' does not support option amendment",
6500 bs
->drv
->format_name
);
6503 return bs
->drv
->bdrv_amend_options(bs
, opts
, status_cb
, cb_opaque
, errp
);
6507 * This function checks whether the given @to_replace is allowed to be
6508 * replaced by a node that always shows the same data as @bs. This is
6509 * used for example to verify whether the mirror job can replace
6510 * @to_replace by the target mirrored from @bs.
6511 * To be replaceable, @bs and @to_replace may either be guaranteed to
6512 * always show the same data (because they are only connected through
6513 * filters), or some driver may allow replacing one of its children
6514 * because it can guarantee that this child's data is not visible at
6515 * all (for example, for dissenting quorum children that have no other
6518 bool bdrv_recurse_can_replace(BlockDriverState
*bs
,
6519 BlockDriverState
*to_replace
)
6521 if (!bs
|| !bs
->drv
) {
6525 if (bs
== to_replace
) {
6529 /* See what the driver can do */
6530 if (bs
->drv
->bdrv_recurse_can_replace
) {
6531 return bs
->drv
->bdrv_recurse_can_replace(bs
, to_replace
);
6534 /* For filters without an own implementation, we can recurse on our own */
6535 if (bs
->drv
->is_filter
) {
6536 BdrvChild
*child
= bs
->file
?: bs
->backing
;
6537 return bdrv_recurse_can_replace(child
->bs
, to_replace
);
6545 * Check whether the given @node_name can be replaced by a node that
6546 * has the same data as @parent_bs. If so, return @node_name's BDS;
6549 * @node_name must be a (recursive) *child of @parent_bs (or this
6550 * function will return NULL).
6552 * The result (whether the node can be replaced or not) is only valid
6553 * for as long as no graph or permission changes occur.
6555 BlockDriverState
*check_to_replace_node(BlockDriverState
*parent_bs
,
6556 const char *node_name
, Error
**errp
)
6558 BlockDriverState
*to_replace_bs
= bdrv_find_node(node_name
);
6559 AioContext
*aio_context
;
6561 if (!to_replace_bs
) {
6562 error_setg(errp
, "Node name '%s' not found", node_name
);
6566 aio_context
= bdrv_get_aio_context(to_replace_bs
);
6567 aio_context_acquire(aio_context
);
6569 if (bdrv_op_is_blocked(to_replace_bs
, BLOCK_OP_TYPE_REPLACE
, errp
)) {
6570 to_replace_bs
= NULL
;
6574 /* We don't want arbitrary node of the BDS chain to be replaced only the top
6575 * most non filter in order to prevent data corruption.
6576 * Another benefit is that this tests exclude backing files which are
6577 * blocked by the backing blockers.
6579 if (!bdrv_recurse_can_replace(parent_bs
, to_replace_bs
)) {
6580 error_setg(errp
, "Cannot replace '%s' by a node mirrored from '%s', "
6581 "because it cannot be guaranteed that doing so would not "
6582 "lead to an abrupt change of visible data",
6583 node_name
, parent_bs
->node_name
);
6584 to_replace_bs
= NULL
;
6589 aio_context_release(aio_context
);
6590 return to_replace_bs
;
6594 * Iterates through the list of runtime option keys that are said to
6595 * be "strong" for a BDS. An option is called "strong" if it changes
6596 * a BDS's data. For example, the null block driver's "size" and
6597 * "read-zeroes" options are strong, but its "latency-ns" option is
6600 * If a key returned by this function ends with a dot, all options
6601 * starting with that prefix are strong.
6603 static const char *const *strong_options(BlockDriverState
*bs
,
6604 const char *const *curopt
)
6606 static const char *const global_options
[] = {
6607 "driver", "filename", NULL
6611 return &global_options
[0];
6615 if (curopt
== &global_options
[ARRAY_SIZE(global_options
) - 1] && bs
->drv
) {
6616 curopt
= bs
->drv
->strong_runtime_opts
;
6619 return (curopt
&& *curopt
) ? curopt
: NULL
;
6623 * Copies all strong runtime options from bs->options to the given
6624 * QDict. The set of strong option keys is determined by invoking
6627 * Returns true iff any strong option was present in bs->options (and
6628 * thus copied to the target QDict) with the exception of "filename"
6629 * and "driver". The caller is expected to use this value to decide
6630 * whether the existence of strong options prevents the generation of
6633 static bool append_strong_runtime_options(QDict
*d
, BlockDriverState
*bs
)
6635 bool found_any
= false;
6636 const char *const *option_name
= NULL
;
6642 while ((option_name
= strong_options(bs
, option_name
))) {
6643 bool option_given
= false;
6645 assert(strlen(*option_name
) > 0);
6646 if ((*option_name
)[strlen(*option_name
) - 1] != '.') {
6647 QObject
*entry
= qdict_get(bs
->options
, *option_name
);
6652 qdict_put_obj(d
, *option_name
, qobject_ref(entry
));
6653 option_given
= true;
6655 const QDictEntry
*entry
;
6656 for (entry
= qdict_first(bs
->options
); entry
;
6657 entry
= qdict_next(bs
->options
, entry
))
6659 if (strstart(qdict_entry_key(entry
), *option_name
, NULL
)) {
6660 qdict_put_obj(d
, qdict_entry_key(entry
),
6661 qobject_ref(qdict_entry_value(entry
)));
6662 option_given
= true;
6667 /* While "driver" and "filename" need to be included in a JSON filename,
6668 * their existence does not prohibit generation of a plain filename. */
6669 if (!found_any
&& option_given
&&
6670 strcmp(*option_name
, "driver") && strcmp(*option_name
, "filename"))
6676 if (!qdict_haskey(d
, "driver")) {
6677 /* Drivers created with bdrv_new_open_driver() may not have a
6678 * @driver option. Add it here. */
6679 qdict_put_str(d
, "driver", bs
->drv
->format_name
);
6685 /* Note: This function may return false positives; it may return true
6686 * even if opening the backing file specified by bs's image header
6687 * would result in exactly bs->backing. */
6688 static bool bdrv_backing_overridden(BlockDriverState
*bs
)
6691 return strcmp(bs
->auto_backing_file
,
6692 bs
->backing
->bs
->filename
);
6694 /* No backing BDS, so if the image header reports any backing
6695 * file, it must have been suppressed */
6696 return bs
->auto_backing_file
[0] != '\0';
6700 /* Updates the following BDS fields:
6701 * - exact_filename: A filename which may be used for opening a block device
6702 * which (mostly) equals the given BDS (even without any
6703 * other options; so reading and writing must return the same
6704 * results, but caching etc. may be different)
6705 * - full_open_options: Options which, when given when opening a block device
6706 * (without a filename), result in a BDS (mostly)
6707 * equalling the given one
6708 * - filename: If exact_filename is set, it is copied here. Otherwise,
6709 * full_open_options is converted to a JSON object, prefixed with
6710 * "json:" (for use through the JSON pseudo protocol) and put here.
6712 void bdrv_refresh_filename(BlockDriverState
*bs
)
6714 BlockDriver
*drv
= bs
->drv
;
6717 bool backing_overridden
;
6718 bool generate_json_filename
; /* Whether our default implementation should
6719 fill exact_filename (false) or not (true) */
6725 /* This BDS's file name may depend on any of its children's file names, so
6726 * refresh those first */
6727 QLIST_FOREACH(child
, &bs
->children
, next
) {
6728 bdrv_refresh_filename(child
->bs
);
6732 /* For implicit nodes, just copy everything from the single child */
6733 child
= QLIST_FIRST(&bs
->children
);
6734 assert(QLIST_NEXT(child
, next
) == NULL
);
6736 pstrcpy(bs
->exact_filename
, sizeof(bs
->exact_filename
),
6737 child
->bs
->exact_filename
);
6738 pstrcpy(bs
->filename
, sizeof(bs
->filename
), child
->bs
->filename
);
6740 qobject_unref(bs
->full_open_options
);
6741 bs
->full_open_options
= qobject_ref(child
->bs
->full_open_options
);
6746 backing_overridden
= bdrv_backing_overridden(bs
);
6748 if (bs
->open_flags
& BDRV_O_NO_IO
) {
6749 /* Without I/O, the backing file does not change anything.
6750 * Therefore, in such a case (primarily qemu-img), we can
6751 * pretend the backing file has not been overridden even if
6752 * it technically has been. */
6753 backing_overridden
= false;
6756 /* Gather the options QDict */
6758 generate_json_filename
= append_strong_runtime_options(opts
, bs
);
6759 generate_json_filename
|= backing_overridden
;
6761 if (drv
->bdrv_gather_child_options
) {
6762 /* Some block drivers may not want to present all of their children's
6763 * options, or name them differently from BdrvChild.name */
6764 drv
->bdrv_gather_child_options(bs
, opts
, backing_overridden
);
6766 QLIST_FOREACH(child
, &bs
->children
, next
) {
6767 if (child
== bs
->backing
&& !backing_overridden
) {
6768 /* We can skip the backing BDS if it has not been overridden */
6772 qdict_put(opts
, child
->name
,
6773 qobject_ref(child
->bs
->full_open_options
));
6776 if (backing_overridden
&& !bs
->backing
) {
6777 /* Force no backing file */
6778 qdict_put_null(opts
, "backing");
6782 qobject_unref(bs
->full_open_options
);
6783 bs
->full_open_options
= opts
;
6785 if (drv
->bdrv_refresh_filename
) {
6786 /* Obsolete information is of no use here, so drop the old file name
6787 * information before refreshing it */
6788 bs
->exact_filename
[0] = '\0';
6790 drv
->bdrv_refresh_filename(bs
);
6791 } else if (bs
->file
) {
6792 /* Try to reconstruct valid information from the underlying file */
6794 bs
->exact_filename
[0] = '\0';
6797 * We can use the underlying file's filename if:
6798 * - it has a filename,
6799 * - the file is a protocol BDS, and
6800 * - opening that file (as this BDS's format) will automatically create
6801 * the BDS tree we have right now, that is:
6802 * - the user did not significantly change this BDS's behavior with
6803 * some explicit (strong) options
6804 * - no non-file child of this BDS has been overridden by the user
6805 * Both of these conditions are represented by generate_json_filename.
6807 if (bs
->file
->bs
->exact_filename
[0] &&
6808 bs
->file
->bs
->drv
->bdrv_file_open
&&
6809 !generate_json_filename
)
6811 strcpy(bs
->exact_filename
, bs
->file
->bs
->exact_filename
);
6815 if (bs
->exact_filename
[0]) {
6816 pstrcpy(bs
->filename
, sizeof(bs
->filename
), bs
->exact_filename
);
6818 QString
*json
= qobject_to_json(QOBJECT(bs
->full_open_options
));
6819 snprintf(bs
->filename
, sizeof(bs
->filename
), "json:%s",
6820 qstring_get_str(json
));
6821 qobject_unref(json
);
6825 char *bdrv_dirname(BlockDriverState
*bs
, Error
**errp
)
6827 BlockDriver
*drv
= bs
->drv
;
6830 error_setg(errp
, "Node '%s' is ejected", bs
->node_name
);
6834 if (drv
->bdrv_dirname
) {
6835 return drv
->bdrv_dirname(bs
, errp
);
6839 return bdrv_dirname(bs
->file
->bs
, errp
);
6842 bdrv_refresh_filename(bs
);
6843 if (bs
->exact_filename
[0] != '\0') {
6844 return path_combine(bs
->exact_filename
, "");
6847 error_setg(errp
, "Cannot generate a base directory for %s nodes",
6853 * Hot add/remove a BDS's child. So the user can take a child offline when
6854 * it is broken and take a new child online
6856 void bdrv_add_child(BlockDriverState
*parent_bs
, BlockDriverState
*child_bs
,
6860 if (!parent_bs
->drv
|| !parent_bs
->drv
->bdrv_add_child
) {
6861 error_setg(errp
, "The node %s does not support adding a child",
6862 bdrv_get_device_or_node_name(parent_bs
));
6866 if (!QLIST_EMPTY(&child_bs
->parents
)) {
6867 error_setg(errp
, "The node %s already has a parent",
6868 child_bs
->node_name
);
6872 parent_bs
->drv
->bdrv_add_child(parent_bs
, child_bs
, errp
);
6875 void bdrv_del_child(BlockDriverState
*parent_bs
, BdrvChild
*child
, Error
**errp
)
6879 if (!parent_bs
->drv
|| !parent_bs
->drv
->bdrv_del_child
) {
6880 error_setg(errp
, "The node %s does not support removing a child",
6881 bdrv_get_device_or_node_name(parent_bs
));
6885 QLIST_FOREACH(tmp
, &parent_bs
->children
, next
) {
6892 error_setg(errp
, "The node %s does not have a child named %s",
6893 bdrv_get_device_or_node_name(parent_bs
),
6894 bdrv_get_device_or_node_name(child
->bs
));
6898 parent_bs
->drv
->bdrv_del_child(parent_bs
, child
, errp
);
6901 int bdrv_make_empty(BdrvChild
*c
, Error
**errp
)
6903 BlockDriver
*drv
= c
->bs
->drv
;
6906 assert(c
->perm
& (BLK_PERM_WRITE
| BLK_PERM_WRITE_UNCHANGED
));
6908 if (!drv
->bdrv_make_empty
) {
6909 error_setg(errp
, "%s does not support emptying nodes",
6914 ret
= drv
->bdrv_make_empty(c
->bs
);
6916 error_setg_errno(errp
, -ret
, "Failed to empty %s",