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
: "",
1212 error_setg_errno(errp
, -ret
, "Could not update backing file link");
1216 bdrv_reopen_set_read_only(parent
, true, NULL
);
1223 * Returns the options and flags that a generic child of a BDS should
1224 * get, based on the given options and flags for the parent BDS.
1226 static void bdrv_inherited_options(BdrvChildRole role
, bool parent_is_format
,
1227 int *child_flags
, QDict
*child_options
,
1228 int parent_flags
, QDict
*parent_options
)
1230 int flags
= parent_flags
;
1233 * First, decide whether to set, clear, or leave BDRV_O_PROTOCOL.
1234 * Generally, the question to answer is: Should this child be
1235 * format-probed by default?
1239 * Pure and non-filtered data children of non-format nodes should
1240 * be probed by default (even when the node itself has BDRV_O_PROTOCOL
1241 * set). This only affects a very limited set of drivers (namely
1242 * quorum and blkverify when this comment was written).
1243 * Force-clear BDRV_O_PROTOCOL then.
1245 if (!parent_is_format
&&
1246 (role
& BDRV_CHILD_DATA
) &&
1247 !(role
& (BDRV_CHILD_METADATA
| BDRV_CHILD_FILTERED
)))
1249 flags
&= ~BDRV_O_PROTOCOL
;
1253 * All children of format nodes (except for COW children) and all
1254 * metadata children in general should never be format-probed.
1255 * Force-set BDRV_O_PROTOCOL then.
1257 if ((parent_is_format
&& !(role
& BDRV_CHILD_COW
)) ||
1258 (role
& BDRV_CHILD_METADATA
))
1260 flags
|= BDRV_O_PROTOCOL
;
1264 * If the cache mode isn't explicitly set, inherit direct and no-flush from
1267 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_CACHE_DIRECT
);
1268 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_CACHE_NO_FLUSH
);
1269 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_FORCE_SHARE
);
1271 if (role
& BDRV_CHILD_COW
) {
1272 /* backing files are opened read-only by default */
1273 qdict_set_default_str(child_options
, BDRV_OPT_READ_ONLY
, "on");
1274 qdict_set_default_str(child_options
, BDRV_OPT_AUTO_READ_ONLY
, "off");
1276 /* Inherit the read-only option from the parent if it's not set */
1277 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_READ_ONLY
);
1278 qdict_copy_default(child_options
, parent_options
,
1279 BDRV_OPT_AUTO_READ_ONLY
);
1283 * bdrv_co_pdiscard() respects unmap policy for the parent, so we
1284 * can default to enable it on lower layers regardless of the
1287 qdict_set_default_str(child_options
, BDRV_OPT_DISCARD
, "unmap");
1289 /* Clear flags that only apply to the top layer */
1290 flags
&= ~(BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
| BDRV_O_COPY_ON_READ
);
1292 if (role
& BDRV_CHILD_METADATA
) {
1293 flags
&= ~BDRV_O_NO_IO
;
1295 if (role
& BDRV_CHILD_COW
) {
1296 flags
&= ~BDRV_O_TEMPORARY
;
1299 *child_flags
= flags
;
1302 static void bdrv_child_cb_attach(BdrvChild
*child
)
1304 BlockDriverState
*bs
= child
->opaque
;
1306 if (child
->role
& BDRV_CHILD_COW
) {
1307 bdrv_backing_attach(child
);
1310 bdrv_apply_subtree_drain(child
, bs
);
1313 static void bdrv_child_cb_detach(BdrvChild
*child
)
1315 BlockDriverState
*bs
= child
->opaque
;
1317 if (child
->role
& BDRV_CHILD_COW
) {
1318 bdrv_backing_detach(child
);
1321 bdrv_unapply_subtree_drain(child
, bs
);
1324 static int bdrv_child_cb_update_filename(BdrvChild
*c
, BlockDriverState
*base
,
1325 const char *filename
, Error
**errp
)
1327 if (c
->role
& BDRV_CHILD_COW
) {
1328 return bdrv_backing_update_filename(c
, base
, filename
, errp
);
1333 const BdrvChildClass child_of_bds
= {
1334 .parent_is_bds
= true,
1335 .get_parent_desc
= bdrv_child_get_parent_desc
,
1336 .inherit_options
= bdrv_inherited_options
,
1337 .drained_begin
= bdrv_child_cb_drained_begin
,
1338 .drained_poll
= bdrv_child_cb_drained_poll
,
1339 .drained_end
= bdrv_child_cb_drained_end
,
1340 .attach
= bdrv_child_cb_attach
,
1341 .detach
= bdrv_child_cb_detach
,
1342 .inactivate
= bdrv_child_cb_inactivate
,
1343 .can_set_aio_ctx
= bdrv_child_cb_can_set_aio_ctx
,
1344 .set_aio_ctx
= bdrv_child_cb_set_aio_ctx
,
1345 .update_filename
= bdrv_child_cb_update_filename
,
1348 static int bdrv_open_flags(BlockDriverState
*bs
, int flags
)
1350 int open_flags
= flags
;
1353 * Clear flags that are internal to the block layer before opening the
1356 open_flags
&= ~(BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
| BDRV_O_PROTOCOL
);
1361 static void update_flags_from_options(int *flags
, QemuOpts
*opts
)
1363 *flags
&= ~(BDRV_O_CACHE_MASK
| BDRV_O_RDWR
| BDRV_O_AUTO_RDONLY
);
1365 if (qemu_opt_get_bool_del(opts
, BDRV_OPT_CACHE_NO_FLUSH
, false)) {
1366 *flags
|= BDRV_O_NO_FLUSH
;
1369 if (qemu_opt_get_bool_del(opts
, BDRV_OPT_CACHE_DIRECT
, false)) {
1370 *flags
|= BDRV_O_NOCACHE
;
1373 if (!qemu_opt_get_bool_del(opts
, BDRV_OPT_READ_ONLY
, false)) {
1374 *flags
|= BDRV_O_RDWR
;
1377 if (qemu_opt_get_bool_del(opts
, BDRV_OPT_AUTO_READ_ONLY
, false)) {
1378 *flags
|= BDRV_O_AUTO_RDONLY
;
1382 static void update_options_from_flags(QDict
*options
, int flags
)
1384 if (!qdict_haskey(options
, BDRV_OPT_CACHE_DIRECT
)) {
1385 qdict_put_bool(options
, BDRV_OPT_CACHE_DIRECT
, flags
& BDRV_O_NOCACHE
);
1387 if (!qdict_haskey(options
, BDRV_OPT_CACHE_NO_FLUSH
)) {
1388 qdict_put_bool(options
, BDRV_OPT_CACHE_NO_FLUSH
,
1389 flags
& BDRV_O_NO_FLUSH
);
1391 if (!qdict_haskey(options
, BDRV_OPT_READ_ONLY
)) {
1392 qdict_put_bool(options
, BDRV_OPT_READ_ONLY
, !(flags
& BDRV_O_RDWR
));
1394 if (!qdict_haskey(options
, BDRV_OPT_AUTO_READ_ONLY
)) {
1395 qdict_put_bool(options
, BDRV_OPT_AUTO_READ_ONLY
,
1396 flags
& BDRV_O_AUTO_RDONLY
);
1400 static void bdrv_assign_node_name(BlockDriverState
*bs
,
1401 const char *node_name
,
1404 char *gen_node_name
= NULL
;
1407 node_name
= gen_node_name
= id_generate(ID_BLOCK
);
1408 } else if (!id_wellformed(node_name
)) {
1410 * Check for empty string or invalid characters, but not if it is
1411 * generated (generated names use characters not available to the user)
1413 error_setg(errp
, "Invalid node name");
1417 /* takes care of avoiding namespaces collisions */
1418 if (blk_by_name(node_name
)) {
1419 error_setg(errp
, "node-name=%s is conflicting with a device id",
1424 /* takes care of avoiding duplicates node names */
1425 if (bdrv_find_node(node_name
)) {
1426 error_setg(errp
, "Duplicate node name");
1430 /* Make sure that the node name isn't truncated */
1431 if (strlen(node_name
) >= sizeof(bs
->node_name
)) {
1432 error_setg(errp
, "Node name too long");
1436 /* copy node name into the bs and insert it into the graph list */
1437 pstrcpy(bs
->node_name
, sizeof(bs
->node_name
), node_name
);
1438 QTAILQ_INSERT_TAIL(&graph_bdrv_states
, bs
, node_list
);
1440 g_free(gen_node_name
);
1443 static int bdrv_open_driver(BlockDriverState
*bs
, BlockDriver
*drv
,
1444 const char *node_name
, QDict
*options
,
1445 int open_flags
, Error
**errp
)
1447 Error
*local_err
= NULL
;
1450 bdrv_assign_node_name(bs
, node_name
, &local_err
);
1452 error_propagate(errp
, local_err
);
1457 bs
->read_only
= !(bs
->open_flags
& BDRV_O_RDWR
);
1458 bs
->opaque
= g_malloc0(drv
->instance_size
);
1460 if (drv
->bdrv_file_open
) {
1461 assert(!drv
->bdrv_needs_filename
|| bs
->filename
[0]);
1462 ret
= drv
->bdrv_file_open(bs
, options
, open_flags
, &local_err
);
1463 } else if (drv
->bdrv_open
) {
1464 ret
= drv
->bdrv_open(bs
, options
, open_flags
, &local_err
);
1471 error_propagate(errp
, local_err
);
1472 } else if (bs
->filename
[0]) {
1473 error_setg_errno(errp
, -ret
, "Could not open '%s'", bs
->filename
);
1475 error_setg_errno(errp
, -ret
, "Could not open image");
1480 ret
= refresh_total_sectors(bs
, bs
->total_sectors
);
1482 error_setg_errno(errp
, -ret
, "Could not refresh total sector count");
1486 bdrv_refresh_limits(bs
, &local_err
);
1488 error_propagate(errp
, local_err
);
1492 assert(bdrv_opt_mem_align(bs
) != 0);
1493 assert(bdrv_min_mem_align(bs
) != 0);
1494 assert(is_power_of_2(bs
->bl
.request_alignment
));
1496 for (i
= 0; i
< bs
->quiesce_counter
; i
++) {
1497 if (drv
->bdrv_co_drain_begin
) {
1498 drv
->bdrv_co_drain_begin(bs
);
1505 if (bs
->file
!= NULL
) {
1506 bdrv_unref_child(bs
, bs
->file
);
1514 BlockDriverState
*bdrv_new_open_driver(BlockDriver
*drv
, const char *node_name
,
1515 int flags
, Error
**errp
)
1517 BlockDriverState
*bs
;
1521 bs
->open_flags
= flags
;
1522 bs
->explicit_options
= qdict_new();
1523 bs
->options
= qdict_new();
1526 update_options_from_flags(bs
->options
, flags
);
1528 ret
= bdrv_open_driver(bs
, drv
, node_name
, bs
->options
, flags
, errp
);
1530 qobject_unref(bs
->explicit_options
);
1531 bs
->explicit_options
= NULL
;
1532 qobject_unref(bs
->options
);
1541 QemuOptsList bdrv_runtime_opts
= {
1542 .name
= "bdrv_common",
1543 .head
= QTAILQ_HEAD_INITIALIZER(bdrv_runtime_opts
.head
),
1546 .name
= "node-name",
1547 .type
= QEMU_OPT_STRING
,
1548 .help
= "Node name of the block device node",
1552 .type
= QEMU_OPT_STRING
,
1553 .help
= "Block driver to use for the node",
1556 .name
= BDRV_OPT_CACHE_DIRECT
,
1557 .type
= QEMU_OPT_BOOL
,
1558 .help
= "Bypass software writeback cache on the host",
1561 .name
= BDRV_OPT_CACHE_NO_FLUSH
,
1562 .type
= QEMU_OPT_BOOL
,
1563 .help
= "Ignore flush requests",
1566 .name
= BDRV_OPT_READ_ONLY
,
1567 .type
= QEMU_OPT_BOOL
,
1568 .help
= "Node is opened in read-only mode",
1571 .name
= BDRV_OPT_AUTO_READ_ONLY
,
1572 .type
= QEMU_OPT_BOOL
,
1573 .help
= "Node can become read-only if opening read-write fails",
1576 .name
= "detect-zeroes",
1577 .type
= QEMU_OPT_STRING
,
1578 .help
= "try to optimize zero writes (off, on, unmap)",
1581 .name
= BDRV_OPT_DISCARD
,
1582 .type
= QEMU_OPT_STRING
,
1583 .help
= "discard operation (ignore/off, unmap/on)",
1586 .name
= BDRV_OPT_FORCE_SHARE
,
1587 .type
= QEMU_OPT_BOOL
,
1588 .help
= "always accept other writers (default: off)",
1590 { /* end of list */ }
1594 QemuOptsList bdrv_create_opts_simple
= {
1595 .name
= "simple-create-opts",
1596 .head
= QTAILQ_HEAD_INITIALIZER(bdrv_create_opts_simple
.head
),
1599 .name
= BLOCK_OPT_SIZE
,
1600 .type
= QEMU_OPT_SIZE
,
1601 .help
= "Virtual disk size"
1604 .name
= BLOCK_OPT_PREALLOC
,
1605 .type
= QEMU_OPT_STRING
,
1606 .help
= "Preallocation mode (allowed values: off)"
1608 { /* end of list */ }
1613 * Common part for opening disk images and files
1615 * Removes all processed options from *options.
1617 static int bdrv_open_common(BlockDriverState
*bs
, BlockBackend
*file
,
1618 QDict
*options
, Error
**errp
)
1620 int ret
, open_flags
;
1621 const char *filename
;
1622 const char *driver_name
= NULL
;
1623 const char *node_name
= NULL
;
1624 const char *discard
;
1627 Error
*local_err
= NULL
;
1629 assert(bs
->file
== NULL
);
1630 assert(options
!= NULL
&& bs
->options
!= options
);
1632 opts
= qemu_opts_create(&bdrv_runtime_opts
, NULL
, 0, &error_abort
);
1633 if (!qemu_opts_absorb_qdict(opts
, options
, errp
)) {
1638 update_flags_from_options(&bs
->open_flags
, opts
);
1640 driver_name
= qemu_opt_get(opts
, "driver");
1641 drv
= bdrv_find_format(driver_name
);
1642 assert(drv
!= NULL
);
1644 bs
->force_share
= qemu_opt_get_bool(opts
, BDRV_OPT_FORCE_SHARE
, false);
1646 if (bs
->force_share
&& (bs
->open_flags
& BDRV_O_RDWR
)) {
1648 BDRV_OPT_FORCE_SHARE
1649 "=on can only be used with read-only images");
1655 bdrv_refresh_filename(blk_bs(file
));
1656 filename
= blk_bs(file
)->filename
;
1659 * Caution: while qdict_get_try_str() is fine, getting
1660 * non-string types would require more care. When @options
1661 * come from -blockdev or blockdev_add, its members are typed
1662 * according to the QAPI schema, but when they come from
1663 * -drive, they're all QString.
1665 filename
= qdict_get_try_str(options
, "filename");
1668 if (drv
->bdrv_needs_filename
&& (!filename
|| !filename
[0])) {
1669 error_setg(errp
, "The '%s' block driver requires a file name",
1675 trace_bdrv_open_common(bs
, filename
?: "", bs
->open_flags
,
1678 bs
->read_only
= !(bs
->open_flags
& BDRV_O_RDWR
);
1680 if (use_bdrv_whitelist
&& !bdrv_is_whitelisted(drv
, bs
->read_only
)) {
1681 if (!bs
->read_only
&& bdrv_is_whitelisted(drv
, true)) {
1682 ret
= bdrv_apply_auto_read_only(bs
, NULL
, NULL
);
1688 !bs
->read_only
&& bdrv_is_whitelisted(drv
, true)
1689 ? "Driver '%s' can only be used for read-only devices"
1690 : "Driver '%s' is not whitelisted",
1696 /* bdrv_new() and bdrv_close() make it so */
1697 assert(atomic_read(&bs
->copy_on_read
) == 0);
1699 if (bs
->open_flags
& BDRV_O_COPY_ON_READ
) {
1700 if (!bs
->read_only
) {
1701 bdrv_enable_copy_on_read(bs
);
1703 error_setg(errp
, "Can't use copy-on-read on read-only device");
1709 discard
= qemu_opt_get(opts
, BDRV_OPT_DISCARD
);
1710 if (discard
!= NULL
) {
1711 if (bdrv_parse_discard_flags(discard
, &bs
->open_flags
) != 0) {
1712 error_setg(errp
, "Invalid discard option");
1719 bdrv_parse_detect_zeroes(opts
, bs
->open_flags
, &local_err
);
1721 error_propagate(errp
, local_err
);
1726 if (filename
!= NULL
) {
1727 pstrcpy(bs
->filename
, sizeof(bs
->filename
), filename
);
1729 bs
->filename
[0] = '\0';
1731 pstrcpy(bs
->exact_filename
, sizeof(bs
->exact_filename
), bs
->filename
);
1733 /* Open the image, either directly or using a protocol */
1734 open_flags
= bdrv_open_flags(bs
, bs
->open_flags
);
1735 node_name
= qemu_opt_get(opts
, "node-name");
1737 assert(!drv
->bdrv_file_open
|| file
== NULL
);
1738 ret
= bdrv_open_driver(bs
, drv
, node_name
, options
, open_flags
, errp
);
1743 qemu_opts_del(opts
);
1747 qemu_opts_del(opts
);
1751 static GCC_FMT_ATTR(1, 0)
1752 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
, BdrvChildRole role
,
1951 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
, 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 "
2030 * Unaligned requests will automatically be aligned to bl.request_alignment
2031 * and without RESIZE we can't extend requests to write to space beyond the
2032 * end of the image, so it's required that the image size is aligned.
2034 if ((cumulative_perms
& (BLK_PERM_WRITE
| BLK_PERM_WRITE_UNCHANGED
)) &&
2035 !(cumulative_perms
& BLK_PERM_RESIZE
))
2037 if ((bs
->total_sectors
* BDRV_SECTOR_SIZE
) % bs
->bl
.request_alignment
) {
2038 error_setg(errp
, "Cannot get 'write' permission without 'resize': "
2039 "Image size is not a multiple of request "
2045 /* Check this node */
2050 if (drv
->bdrv_check_perm
) {
2051 return drv
->bdrv_check_perm(bs
, cumulative_perms
,
2052 cumulative_shared_perms
, errp
);
2055 /* Drivers that never have children can omit .bdrv_child_perm() */
2056 if (!drv
->bdrv_child_perm
) {
2057 assert(QLIST_EMPTY(&bs
->children
));
2061 /* Check all children */
2062 QLIST_FOREACH(c
, &bs
->children
, next
) {
2063 uint64_t cur_perm
, cur_shared
;
2064 bool child_tighten_restr
;
2066 bdrv_child_perm(bs
, c
->bs
, c
, c
->role
, q
,
2067 cumulative_perms
, cumulative_shared_perms
,
2068 &cur_perm
, &cur_shared
);
2069 ret
= bdrv_child_check_perm(c
, q
, cur_perm
, cur_shared
, ignore_children
,
2070 tighten_restrictions
? &child_tighten_restr
2073 if (tighten_restrictions
) {
2074 *tighten_restrictions
|= child_tighten_restr
;
2085 * Notifies drivers that after a previous bdrv_check_perm() call, the
2086 * permission update is not performed and any preparations made for it (e.g.
2087 * taken file locks) need to be undone.
2089 * This function recursively notifies all child nodes.
2091 static void bdrv_abort_perm_update(BlockDriverState
*bs
)
2093 BlockDriver
*drv
= bs
->drv
;
2100 if (drv
->bdrv_abort_perm_update
) {
2101 drv
->bdrv_abort_perm_update(bs
);
2104 QLIST_FOREACH(c
, &bs
->children
, next
) {
2105 bdrv_child_abort_perm_update(c
);
2109 static void bdrv_set_perm(BlockDriverState
*bs
, uint64_t cumulative_perms
,
2110 uint64_t cumulative_shared_perms
)
2112 BlockDriver
*drv
= bs
->drv
;
2119 /* Update this node */
2120 if (drv
->bdrv_set_perm
) {
2121 drv
->bdrv_set_perm(bs
, cumulative_perms
, cumulative_shared_perms
);
2124 /* Drivers that never have children can omit .bdrv_child_perm() */
2125 if (!drv
->bdrv_child_perm
) {
2126 assert(QLIST_EMPTY(&bs
->children
));
2130 /* Update all children */
2131 QLIST_FOREACH(c
, &bs
->children
, next
) {
2132 uint64_t cur_perm
, cur_shared
;
2133 bdrv_child_perm(bs
, c
->bs
, c
, c
->role
, NULL
,
2134 cumulative_perms
, cumulative_shared_perms
,
2135 &cur_perm
, &cur_shared
);
2136 bdrv_child_set_perm(c
, cur_perm
, cur_shared
);
2140 void bdrv_get_cumulative_perm(BlockDriverState
*bs
, uint64_t *perm
,
2141 uint64_t *shared_perm
)
2144 uint64_t cumulative_perms
= 0;
2145 uint64_t cumulative_shared_perms
= BLK_PERM_ALL
;
2147 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
2148 cumulative_perms
|= c
->perm
;
2149 cumulative_shared_perms
&= c
->shared_perm
;
2152 *perm
= cumulative_perms
;
2153 *shared_perm
= cumulative_shared_perms
;
2156 static char *bdrv_child_user_desc(BdrvChild
*c
)
2158 if (c
->klass
->get_parent_desc
) {
2159 return c
->klass
->get_parent_desc(c
);
2162 return g_strdup("another user");
2165 char *bdrv_perm_names(uint64_t perm
)
2171 { BLK_PERM_CONSISTENT_READ
, "consistent read" },
2172 { BLK_PERM_WRITE
, "write" },
2173 { BLK_PERM_WRITE_UNCHANGED
, "write unchanged" },
2174 { BLK_PERM_RESIZE
, "resize" },
2175 { BLK_PERM_GRAPH_MOD
, "change children" },
2179 GString
*result
= g_string_sized_new(30);
2180 struct perm_name
*p
;
2182 for (p
= permissions
; p
->name
; p
++) {
2183 if (perm
& p
->perm
) {
2184 if (result
->len
> 0) {
2185 g_string_append(result
, ", ");
2187 g_string_append(result
, p
->name
);
2191 return g_string_free(result
, FALSE
);
2195 * Checks whether a new reference to @bs can be added if the new user requires
2196 * @new_used_perm/@new_shared_perm as its permissions. If @ignore_children is
2197 * set, the BdrvChild objects in this list are ignored in the calculations;
2198 * this allows checking permission updates for an existing reference.
2200 * See bdrv_check_perm() for the semantics of @tighten_restrictions.
2202 * Needs to be followed by a call to either bdrv_set_perm() or
2203 * bdrv_abort_perm_update(). */
2204 static int bdrv_check_update_perm(BlockDriverState
*bs
, BlockReopenQueue
*q
,
2205 uint64_t new_used_perm
,
2206 uint64_t new_shared_perm
,
2207 GSList
*ignore_children
,
2208 bool *tighten_restrictions
,
2212 uint64_t cumulative_perms
= new_used_perm
;
2213 uint64_t cumulative_shared_perms
= new_shared_perm
;
2215 assert(!q
|| !tighten_restrictions
);
2217 /* There is no reason why anyone couldn't tolerate write_unchanged */
2218 assert(new_shared_perm
& BLK_PERM_WRITE_UNCHANGED
);
2220 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
2221 if (g_slist_find(ignore_children
, c
)) {
2225 if ((new_used_perm
& c
->shared_perm
) != new_used_perm
) {
2226 char *user
= bdrv_child_user_desc(c
);
2227 char *perm_names
= bdrv_perm_names(new_used_perm
& ~c
->shared_perm
);
2229 if (tighten_restrictions
) {
2230 *tighten_restrictions
= true;
2233 error_setg(errp
, "Conflicts with use by %s as '%s', which does not "
2235 user
, c
->name
, perm_names
, bdrv_get_node_name(c
->bs
));
2241 if ((c
->perm
& new_shared_perm
) != c
->perm
) {
2242 char *user
= bdrv_child_user_desc(c
);
2243 char *perm_names
= bdrv_perm_names(c
->perm
& ~new_shared_perm
);
2245 if (tighten_restrictions
) {
2246 *tighten_restrictions
= true;
2249 error_setg(errp
, "Conflicts with use by %s as '%s', which uses "
2251 user
, c
->name
, perm_names
, bdrv_get_node_name(c
->bs
));
2257 cumulative_perms
|= c
->perm
;
2258 cumulative_shared_perms
&= c
->shared_perm
;
2261 return bdrv_check_perm(bs
, q
, cumulative_perms
, cumulative_shared_perms
,
2262 ignore_children
, tighten_restrictions
, errp
);
2265 /* Needs to be followed by a call to either bdrv_child_set_perm() or
2266 * bdrv_child_abort_perm_update(). */
2267 static int bdrv_child_check_perm(BdrvChild
*c
, BlockReopenQueue
*q
,
2268 uint64_t perm
, uint64_t shared
,
2269 GSList
*ignore_children
,
2270 bool *tighten_restrictions
, Error
**errp
)
2274 ignore_children
= g_slist_prepend(g_slist_copy(ignore_children
), c
);
2275 ret
= bdrv_check_update_perm(c
->bs
, q
, perm
, shared
, ignore_children
,
2276 tighten_restrictions
, errp
);
2277 g_slist_free(ignore_children
);
2283 if (!c
->has_backup_perm
) {
2284 c
->has_backup_perm
= true;
2285 c
->backup_perm
= c
->perm
;
2286 c
->backup_shared_perm
= c
->shared_perm
;
2289 * Note: it's OK if c->has_backup_perm was already set, as we can find the
2290 * same child twice during check_perm procedure
2294 c
->shared_perm
= shared
;
2299 static void bdrv_child_set_perm(BdrvChild
*c
, uint64_t perm
, uint64_t shared
)
2301 uint64_t cumulative_perms
, cumulative_shared_perms
;
2303 c
->has_backup_perm
= false;
2306 c
->shared_perm
= shared
;
2308 bdrv_get_cumulative_perm(c
->bs
, &cumulative_perms
,
2309 &cumulative_shared_perms
);
2310 bdrv_set_perm(c
->bs
, cumulative_perms
, cumulative_shared_perms
);
2313 static void bdrv_child_abort_perm_update(BdrvChild
*c
)
2315 if (c
->has_backup_perm
) {
2316 c
->perm
= c
->backup_perm
;
2317 c
->shared_perm
= c
->backup_shared_perm
;
2318 c
->has_backup_perm
= false;
2321 bdrv_abort_perm_update(c
->bs
);
2324 int bdrv_child_try_set_perm(BdrvChild
*c
, uint64_t perm
, uint64_t shared
,
2327 Error
*local_err
= NULL
;
2329 bool tighten_restrictions
;
2331 ret
= bdrv_child_check_perm(c
, NULL
, perm
, shared
, NULL
,
2332 &tighten_restrictions
, &local_err
);
2334 bdrv_child_abort_perm_update(c
);
2335 if (tighten_restrictions
) {
2336 error_propagate(errp
, local_err
);
2339 * Our caller may intend to only loosen restrictions and
2340 * does not expect this function to fail. Errors are not
2341 * fatal in such a case, so we can just hide them from our
2344 error_free(local_err
);
2350 bdrv_child_set_perm(c
, perm
, shared
);
2355 int bdrv_child_refresh_perms(BlockDriverState
*bs
, BdrvChild
*c
, Error
**errp
)
2357 uint64_t parent_perms
, parent_shared
;
2358 uint64_t perms
, shared
;
2360 bdrv_get_cumulative_perm(bs
, &parent_perms
, &parent_shared
);
2361 bdrv_child_perm(bs
, c
->bs
, c
, c
->role
, NULL
,
2362 parent_perms
, parent_shared
, &perms
, &shared
);
2364 return bdrv_child_try_set_perm(c
, perms
, shared
, errp
);
2368 * Default implementation for .bdrv_child_perm() for block filters:
2369 * Forward CONSISTENT_READ, WRITE, WRITE_UNCHANGED, and RESIZE to the
2372 static void bdrv_filter_default_perms(BlockDriverState
*bs
, BdrvChild
*c
,
2374 BlockReopenQueue
*reopen_queue
,
2375 uint64_t perm
, uint64_t shared
,
2376 uint64_t *nperm
, uint64_t *nshared
)
2378 *nperm
= perm
& DEFAULT_PERM_PASSTHROUGH
;
2379 *nshared
= (shared
& DEFAULT_PERM_PASSTHROUGH
) | DEFAULT_PERM_UNCHANGED
;
2382 static void bdrv_default_perms_for_cow(BlockDriverState
*bs
, BdrvChild
*c
,
2384 BlockReopenQueue
*reopen_queue
,
2385 uint64_t perm
, uint64_t shared
,
2386 uint64_t *nperm
, uint64_t *nshared
)
2388 assert(role
& BDRV_CHILD_COW
);
2391 * We want consistent read from backing files if the parent needs it.
2392 * No other operations are performed on backing files.
2394 perm
&= BLK_PERM_CONSISTENT_READ
;
2397 * If the parent can deal with changing data, we're okay with a
2398 * writable and resizable backing file.
2399 * TODO Require !(perm & BLK_PERM_CONSISTENT_READ), too?
2401 if (shared
& BLK_PERM_WRITE
) {
2402 shared
= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2407 shared
|= BLK_PERM_CONSISTENT_READ
| BLK_PERM_GRAPH_MOD
|
2408 BLK_PERM_WRITE_UNCHANGED
;
2410 if (bs
->open_flags
& BDRV_O_INACTIVE
) {
2411 shared
|= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2418 static void bdrv_default_perms_for_storage(BlockDriverState
*bs
, BdrvChild
*c
,
2420 BlockReopenQueue
*reopen_queue
,
2421 uint64_t perm
, uint64_t shared
,
2422 uint64_t *nperm
, uint64_t *nshared
)
2426 assert(role
& (BDRV_CHILD_METADATA
| BDRV_CHILD_DATA
));
2428 flags
= bdrv_reopen_get_flags(reopen_queue
, bs
);
2431 * Apart from the modifications below, the same permissions are
2432 * forwarded and left alone as for filters
2434 bdrv_filter_default_perms(bs
, c
, role
, reopen_queue
,
2435 perm
, shared
, &perm
, &shared
);
2437 if (role
& BDRV_CHILD_METADATA
) {
2438 /* Format drivers may touch metadata even if the guest doesn't write */
2439 if (bdrv_is_writable_after_reopen(bs
, reopen_queue
)) {
2440 perm
|= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2444 * bs->file always needs to be consistent because of the
2445 * metadata. We can never allow other users to resize or write
2448 if (!(flags
& BDRV_O_NO_IO
)) {
2449 perm
|= BLK_PERM_CONSISTENT_READ
;
2451 shared
&= ~(BLK_PERM_WRITE
| BLK_PERM_RESIZE
);
2454 if (role
& BDRV_CHILD_DATA
) {
2456 * Technically, everything in this block is a subset of the
2457 * BDRV_CHILD_METADATA path taken above, and so this could
2458 * be an "else if" branch. However, that is not obvious, and
2459 * this function is not performance critical, therefore we let
2460 * this be an independent "if".
2464 * We cannot allow other users to resize the file because the
2465 * format driver might have some assumptions about the size
2466 * (e.g. because it is stored in metadata, or because the file
2467 * is split into fixed-size data files).
2469 shared
&= ~BLK_PERM_RESIZE
;
2472 * WRITE_UNCHANGED often cannot be performed as such on the
2473 * data file. For example, the qcow2 driver may still need to
2474 * write copied clusters on copy-on-read.
2476 if (perm
& BLK_PERM_WRITE_UNCHANGED
) {
2477 perm
|= BLK_PERM_WRITE
;
2481 * If the data file is written to, the format driver may
2482 * expect to be able to resize it by writing beyond the EOF.
2484 if (perm
& BLK_PERM_WRITE
) {
2485 perm
|= BLK_PERM_RESIZE
;
2489 if (bs
->open_flags
& BDRV_O_INACTIVE
) {
2490 shared
|= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2497 void bdrv_default_perms(BlockDriverState
*bs
, BdrvChild
*c
,
2498 BdrvChildRole role
, BlockReopenQueue
*reopen_queue
,
2499 uint64_t perm
, uint64_t shared
,
2500 uint64_t *nperm
, uint64_t *nshared
)
2502 if (role
& BDRV_CHILD_FILTERED
) {
2503 assert(!(role
& (BDRV_CHILD_DATA
| BDRV_CHILD_METADATA
|
2505 bdrv_filter_default_perms(bs
, c
, role
, reopen_queue
,
2506 perm
, shared
, nperm
, nshared
);
2507 } else if (role
& BDRV_CHILD_COW
) {
2508 assert(!(role
& (BDRV_CHILD_DATA
| BDRV_CHILD_METADATA
)));
2509 bdrv_default_perms_for_cow(bs
, c
, role
, reopen_queue
,
2510 perm
, shared
, nperm
, nshared
);
2511 } else if (role
& (BDRV_CHILD_METADATA
| BDRV_CHILD_DATA
)) {
2512 bdrv_default_perms_for_storage(bs
, c
, role
, reopen_queue
,
2513 perm
, shared
, nperm
, nshared
);
2515 g_assert_not_reached();
2519 uint64_t bdrv_qapi_perm_to_blk_perm(BlockPermission qapi_perm
)
2521 static const uint64_t permissions
[] = {
2522 [BLOCK_PERMISSION_CONSISTENT_READ
] = BLK_PERM_CONSISTENT_READ
,
2523 [BLOCK_PERMISSION_WRITE
] = BLK_PERM_WRITE
,
2524 [BLOCK_PERMISSION_WRITE_UNCHANGED
] = BLK_PERM_WRITE_UNCHANGED
,
2525 [BLOCK_PERMISSION_RESIZE
] = BLK_PERM_RESIZE
,
2526 [BLOCK_PERMISSION_GRAPH_MOD
] = BLK_PERM_GRAPH_MOD
,
2529 QEMU_BUILD_BUG_ON(ARRAY_SIZE(permissions
) != BLOCK_PERMISSION__MAX
);
2530 QEMU_BUILD_BUG_ON(1UL << ARRAY_SIZE(permissions
) != BLK_PERM_ALL
+ 1);
2532 assert(qapi_perm
< BLOCK_PERMISSION__MAX
);
2534 return permissions
[qapi_perm
];
2537 static void bdrv_replace_child_noperm(BdrvChild
*child
,
2538 BlockDriverState
*new_bs
)
2540 BlockDriverState
*old_bs
= child
->bs
;
2541 int new_bs_quiesce_counter
;
2544 assert(!child
->frozen
);
2546 if (old_bs
&& new_bs
) {
2547 assert(bdrv_get_aio_context(old_bs
) == bdrv_get_aio_context(new_bs
));
2550 new_bs_quiesce_counter
= (new_bs
? new_bs
->quiesce_counter
: 0);
2551 drain_saldo
= new_bs_quiesce_counter
- child
->parent_quiesce_counter
;
2554 * If the new child node is drained but the old one was not, flush
2555 * all outstanding requests to the old child node.
2557 while (drain_saldo
> 0 && child
->klass
->drained_begin
) {
2558 bdrv_parent_drained_begin_single(child
, true);
2563 /* Detach first so that the recursive drain sections coming from @child
2564 * are already gone and we only end the drain sections that came from
2566 if (child
->klass
->detach
) {
2567 child
->klass
->detach(child
);
2569 QLIST_REMOVE(child
, next_parent
);
2575 QLIST_INSERT_HEAD(&new_bs
->parents
, child
, next_parent
);
2578 * Detaching the old node may have led to the new node's
2579 * quiesce_counter having been decreased. Not a problem, we
2580 * just need to recognize this here and then invoke
2581 * drained_end appropriately more often.
2583 assert(new_bs
->quiesce_counter
<= new_bs_quiesce_counter
);
2584 drain_saldo
+= new_bs
->quiesce_counter
- new_bs_quiesce_counter
;
2586 /* Attach only after starting new drained sections, so that recursive
2587 * drain sections coming from @child don't get an extra .drained_begin
2589 if (child
->klass
->attach
) {
2590 child
->klass
->attach(child
);
2595 * If the old child node was drained but the new one is not, allow
2596 * requests to come in only after the new node has been attached.
2598 while (drain_saldo
< 0 && child
->klass
->drained_end
) {
2599 bdrv_parent_drained_end_single(child
);
2605 * Updates @child to change its reference to point to @new_bs, including
2606 * checking and applying the necessary permisson updates both to the old node
2609 * NULL is passed as @new_bs for removing the reference before freeing @child.
2611 * If @new_bs is not NULL, bdrv_check_perm() must be called beforehand, as this
2612 * function uses bdrv_set_perm() to update the permissions according to the new
2613 * reference that @new_bs gets.
2615 static void bdrv_replace_child(BdrvChild
*child
, BlockDriverState
*new_bs
)
2617 BlockDriverState
*old_bs
= child
->bs
;
2618 uint64_t perm
, shared_perm
;
2620 bdrv_replace_child_noperm(child
, new_bs
);
2623 * Start with the new node's permissions. If @new_bs is a (direct
2624 * or indirect) child of @old_bs, we must complete the permission
2625 * update on @new_bs before we loosen the restrictions on @old_bs.
2626 * Otherwise, bdrv_check_perm() on @old_bs would re-initiate
2627 * updating the permissions of @new_bs, and thus not purely loosen
2631 bdrv_get_cumulative_perm(new_bs
, &perm
, &shared_perm
);
2632 bdrv_set_perm(new_bs
, perm
, shared_perm
);
2636 /* Update permissions for old node. This is guaranteed to succeed
2637 * because we're just taking a parent away, so we're loosening
2639 bool tighten_restrictions
;
2642 bdrv_get_cumulative_perm(old_bs
, &perm
, &shared_perm
);
2643 ret
= bdrv_check_perm(old_bs
, NULL
, perm
, shared_perm
, NULL
,
2644 &tighten_restrictions
, NULL
);
2645 assert(tighten_restrictions
== false);
2647 /* We only tried to loosen restrictions, so errors are not fatal */
2648 bdrv_abort_perm_update(old_bs
);
2650 bdrv_set_perm(old_bs
, perm
, shared_perm
);
2653 /* When the parent requiring a non-default AioContext is removed, the
2654 * node moves back to the main AioContext */
2655 bdrv_try_set_aio_context(old_bs
, qemu_get_aio_context(), NULL
);
2660 * This function steals the reference to child_bs from the caller.
2661 * That reference is later dropped by bdrv_root_unref_child().
2663 * On failure NULL is returned, errp is set and the reference to
2664 * child_bs is also dropped.
2666 * The caller must hold the AioContext lock @child_bs, but not that of @ctx
2667 * (unless @child_bs is already in @ctx).
2669 BdrvChild
*bdrv_root_attach_child(BlockDriverState
*child_bs
,
2670 const char *child_name
,
2671 const BdrvChildClass
*child_class
,
2672 BdrvChildRole child_role
,
2674 uint64_t perm
, uint64_t shared_perm
,
2675 void *opaque
, Error
**errp
)
2678 Error
*local_err
= NULL
;
2681 ret
= bdrv_check_update_perm(child_bs
, NULL
, perm
, shared_perm
, NULL
, NULL
,
2684 bdrv_abort_perm_update(child_bs
);
2685 bdrv_unref(child_bs
);
2689 child
= g_new(BdrvChild
, 1);
2690 *child
= (BdrvChild
) {
2692 .name
= g_strdup(child_name
),
2693 .klass
= child_class
,
2696 .shared_perm
= shared_perm
,
2700 /* If the AioContexts don't match, first try to move the subtree of
2701 * child_bs into the AioContext of the new parent. If this doesn't work,
2702 * try moving the parent into the AioContext of child_bs instead. */
2703 if (bdrv_get_aio_context(child_bs
) != ctx
) {
2704 ret
= bdrv_try_set_aio_context(child_bs
, ctx
, &local_err
);
2705 if (ret
< 0 && child_class
->can_set_aio_ctx
) {
2706 GSList
*ignore
= g_slist_prepend(NULL
, child
);
2707 ctx
= bdrv_get_aio_context(child_bs
);
2708 if (child_class
->can_set_aio_ctx(child
, ctx
, &ignore
, NULL
)) {
2709 error_free(local_err
);
2711 g_slist_free(ignore
);
2712 ignore
= g_slist_prepend(NULL
, child
);
2713 child_class
->set_aio_ctx(child
, ctx
, &ignore
);
2715 g_slist_free(ignore
);
2718 error_propagate(errp
, local_err
);
2720 bdrv_abort_perm_update(child_bs
);
2721 bdrv_unref(child_bs
);
2726 /* This performs the matching bdrv_set_perm() for the above check. */
2727 bdrv_replace_child(child
, child_bs
);
2733 * This function transfers the reference to child_bs from the caller
2734 * to parent_bs. That reference is later dropped by parent_bs on
2735 * bdrv_close() or if someone calls bdrv_unref_child().
2737 * On failure NULL is returned, errp is set and the reference to
2738 * child_bs is also dropped.
2740 * If @parent_bs and @child_bs are in different AioContexts, the caller must
2741 * hold the AioContext lock for @child_bs, but not for @parent_bs.
2743 BdrvChild
*bdrv_attach_child(BlockDriverState
*parent_bs
,
2744 BlockDriverState
*child_bs
,
2745 const char *child_name
,
2746 const BdrvChildClass
*child_class
,
2747 BdrvChildRole child_role
,
2751 uint64_t perm
, shared_perm
;
2753 bdrv_get_cumulative_perm(parent_bs
, &perm
, &shared_perm
);
2755 assert(parent_bs
->drv
);
2756 bdrv_child_perm(parent_bs
, child_bs
, NULL
, child_role
, NULL
,
2757 perm
, shared_perm
, &perm
, &shared_perm
);
2759 child
= bdrv_root_attach_child(child_bs
, child_name
, child_class
,
2760 child_role
, bdrv_get_aio_context(parent_bs
),
2761 perm
, shared_perm
, parent_bs
, errp
);
2762 if (child
== NULL
) {
2766 QLIST_INSERT_HEAD(&parent_bs
->children
, child
, next
);
2770 static void bdrv_detach_child(BdrvChild
*child
)
2772 QLIST_SAFE_REMOVE(child
, next
);
2774 bdrv_replace_child(child
, NULL
);
2776 g_free(child
->name
);
2780 void bdrv_root_unref_child(BdrvChild
*child
)
2782 BlockDriverState
*child_bs
;
2784 child_bs
= child
->bs
;
2785 bdrv_detach_child(child
);
2786 bdrv_unref(child_bs
);
2790 * Clear all inherits_from pointers from children and grandchildren of
2791 * @root that point to @root, where necessary.
2793 static void bdrv_unset_inherits_from(BlockDriverState
*root
, BdrvChild
*child
)
2797 if (child
->bs
->inherits_from
== root
) {
2799 * Remove inherits_from only when the last reference between root and
2800 * child->bs goes away.
2802 QLIST_FOREACH(c
, &root
->children
, next
) {
2803 if (c
!= child
&& c
->bs
== child
->bs
) {
2808 child
->bs
->inherits_from
= NULL
;
2812 QLIST_FOREACH(c
, &child
->bs
->children
, next
) {
2813 bdrv_unset_inherits_from(root
, c
);
2817 void bdrv_unref_child(BlockDriverState
*parent
, BdrvChild
*child
)
2819 if (child
== NULL
) {
2823 bdrv_unset_inherits_from(parent
, child
);
2824 bdrv_root_unref_child(child
);
2828 static void bdrv_parent_cb_change_media(BlockDriverState
*bs
, bool load
)
2831 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
2832 if (c
->klass
->change_media
) {
2833 c
->klass
->change_media(c
, load
);
2838 /* Return true if you can reach parent going through child->inherits_from
2839 * recursively. If parent or child are NULL, return false */
2840 static bool bdrv_inherits_from_recursive(BlockDriverState
*child
,
2841 BlockDriverState
*parent
)
2843 while (child
&& child
!= parent
) {
2844 child
= child
->inherits_from
;
2847 return child
!= NULL
;
2851 * Return the BdrvChildRole for @bs's backing child. bs->backing is
2852 * mostly used for COW backing children (role = COW), but also for
2853 * filtered children (role = FILTERED | PRIMARY).
2855 static BdrvChildRole
bdrv_backing_role(BlockDriverState
*bs
)
2857 if (bs
->drv
&& bs
->drv
->is_filter
) {
2858 return BDRV_CHILD_FILTERED
| BDRV_CHILD_PRIMARY
;
2860 return BDRV_CHILD_COW
;
2865 * Sets the backing file link of a BDS. A new reference is created; callers
2866 * which don't need their own reference any more must call bdrv_unref().
2868 void bdrv_set_backing_hd(BlockDriverState
*bs
, BlockDriverState
*backing_hd
,
2871 bool update_inherits_from
= bdrv_chain_contains(bs
, backing_hd
) &&
2872 bdrv_inherits_from_recursive(backing_hd
, bs
);
2874 if (bdrv_is_backing_chain_frozen(bs
, backing_bs(bs
), errp
)) {
2879 bdrv_ref(backing_hd
);
2883 bdrv_unref_child(bs
, bs
->backing
);
2891 bs
->backing
= bdrv_attach_child(bs
, backing_hd
, "backing", &child_of_bds
,
2892 bdrv_backing_role(bs
), errp
);
2893 /* If backing_hd was already part of bs's backing chain, and
2894 * inherits_from pointed recursively to bs then let's update it to
2895 * point directly to bs (else it will become NULL). */
2896 if (bs
->backing
&& update_inherits_from
) {
2897 backing_hd
->inherits_from
= bs
;
2901 bdrv_refresh_limits(bs
, NULL
);
2905 * Opens the backing file for a BlockDriverState if not yet open
2907 * bdref_key specifies the key for the image's BlockdevRef in the options QDict.
2908 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
2909 * itself, all options starting with "${bdref_key}." are considered part of the
2912 * TODO Can this be unified with bdrv_open_image()?
2914 int bdrv_open_backing_file(BlockDriverState
*bs
, QDict
*parent_options
,
2915 const char *bdref_key
, Error
**errp
)
2917 char *backing_filename
= NULL
;
2918 char *bdref_key_dot
;
2919 const char *reference
= NULL
;
2921 bool implicit_backing
= false;
2922 BlockDriverState
*backing_hd
;
2924 QDict
*tmp_parent_options
= NULL
;
2925 Error
*local_err
= NULL
;
2927 if (bs
->backing
!= NULL
) {
2931 /* NULL means an empty set of options */
2932 if (parent_options
== NULL
) {
2933 tmp_parent_options
= qdict_new();
2934 parent_options
= tmp_parent_options
;
2937 bs
->open_flags
&= ~BDRV_O_NO_BACKING
;
2939 bdref_key_dot
= g_strdup_printf("%s.", bdref_key
);
2940 qdict_extract_subqdict(parent_options
, &options
, bdref_key_dot
);
2941 g_free(bdref_key_dot
);
2944 * Caution: while qdict_get_try_str() is fine, getting non-string
2945 * types would require more care. When @parent_options come from
2946 * -blockdev or blockdev_add, its members are typed according to
2947 * the QAPI schema, but when they come from -drive, they're all
2950 reference
= qdict_get_try_str(parent_options
, bdref_key
);
2951 if (reference
|| qdict_haskey(options
, "file.filename")) {
2952 /* keep backing_filename NULL */
2953 } else if (bs
->backing_file
[0] == '\0' && qdict_size(options
) == 0) {
2954 qobject_unref(options
);
2957 if (qdict_size(options
) == 0) {
2958 /* If the user specifies options that do not modify the
2959 * backing file's behavior, we might still consider it the
2960 * implicit backing file. But it's easier this way, and
2961 * just specifying some of the backing BDS's options is
2962 * only possible with -drive anyway (otherwise the QAPI
2963 * schema forces the user to specify everything). */
2964 implicit_backing
= !strcmp(bs
->auto_backing_file
, bs
->backing_file
);
2967 backing_filename
= bdrv_get_full_backing_filename(bs
, &local_err
);
2970 error_propagate(errp
, local_err
);
2971 qobject_unref(options
);
2976 if (!bs
->drv
|| !bs
->drv
->supports_backing
) {
2978 error_setg(errp
, "Driver doesn't support backing files");
2979 qobject_unref(options
);
2984 bs
->backing_format
[0] != '\0' && !qdict_haskey(options
, "driver")) {
2985 qdict_put_str(options
, "driver", bs
->backing_format
);
2988 backing_hd
= bdrv_open_inherit(backing_filename
, reference
, options
, 0, bs
,
2989 &child_of_bds
, bdrv_backing_role(bs
), errp
);
2991 bs
->open_flags
|= BDRV_O_NO_BACKING
;
2992 error_prepend(errp
, "Could not open backing file: ");
2997 if (implicit_backing
) {
2998 bdrv_refresh_filename(backing_hd
);
2999 pstrcpy(bs
->auto_backing_file
, sizeof(bs
->auto_backing_file
),
3000 backing_hd
->filename
);
3003 /* Hook up the backing file link; drop our reference, bs owns the
3004 * backing_hd reference now */
3005 bdrv_set_backing_hd(bs
, backing_hd
, &local_err
);
3006 bdrv_unref(backing_hd
);
3008 error_propagate(errp
, local_err
);
3013 qdict_del(parent_options
, bdref_key
);
3016 g_free(backing_filename
);
3017 qobject_unref(tmp_parent_options
);
3021 static BlockDriverState
*
3022 bdrv_open_child_bs(const char *filename
, QDict
*options
, const char *bdref_key
,
3023 BlockDriverState
*parent
, const BdrvChildClass
*child_class
,
3024 BdrvChildRole child_role
, bool allow_none
, Error
**errp
)
3026 BlockDriverState
*bs
= NULL
;
3027 QDict
*image_options
;
3028 char *bdref_key_dot
;
3029 const char *reference
;
3031 assert(child_class
!= NULL
);
3033 bdref_key_dot
= g_strdup_printf("%s.", bdref_key
);
3034 qdict_extract_subqdict(options
, &image_options
, bdref_key_dot
);
3035 g_free(bdref_key_dot
);
3038 * Caution: while qdict_get_try_str() is fine, getting non-string
3039 * types would require more care. When @options come from
3040 * -blockdev or blockdev_add, its members are typed according to
3041 * the QAPI schema, but when they come from -drive, they're all
3044 reference
= qdict_get_try_str(options
, bdref_key
);
3045 if (!filename
&& !reference
&& !qdict_size(image_options
)) {
3047 error_setg(errp
, "A block device must be specified for \"%s\"",
3050 qobject_unref(image_options
);
3054 bs
= bdrv_open_inherit(filename
, reference
, image_options
, 0,
3055 parent
, child_class
, child_role
, errp
);
3061 qdict_del(options
, bdref_key
);
3066 * Opens a disk image whose options are given as BlockdevRef in another block
3069 * If allow_none is true, no image will be opened if filename is false and no
3070 * BlockdevRef is given. NULL will be returned, but errp remains unset.
3072 * bdrev_key specifies the key for the image's BlockdevRef in the options QDict.
3073 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
3074 * itself, all options starting with "${bdref_key}." are considered part of the
3077 * The BlockdevRef will be removed from the options QDict.
3079 BdrvChild
*bdrv_open_child(const char *filename
,
3080 QDict
*options
, const char *bdref_key
,
3081 BlockDriverState
*parent
,
3082 const BdrvChildClass
*child_class
,
3083 BdrvChildRole child_role
,
3084 bool allow_none
, Error
**errp
)
3086 BlockDriverState
*bs
;
3088 bs
= bdrv_open_child_bs(filename
, options
, bdref_key
, parent
, child_class
,
3089 child_role
, allow_none
, errp
);
3094 return bdrv_attach_child(parent
, bs
, bdref_key
, child_class
, child_role
,
3099 * TODO Future callers may need to specify parent/child_class in order for
3100 * option inheritance to work. Existing callers use it for the root node.
3102 BlockDriverState
*bdrv_open_blockdev_ref(BlockdevRef
*ref
, Error
**errp
)
3104 BlockDriverState
*bs
= NULL
;
3105 QObject
*obj
= NULL
;
3106 QDict
*qdict
= NULL
;
3107 const char *reference
= NULL
;
3110 if (ref
->type
== QTYPE_QSTRING
) {
3111 reference
= ref
->u
.reference
;
3113 BlockdevOptions
*options
= &ref
->u
.definition
;
3114 assert(ref
->type
== QTYPE_QDICT
);
3116 v
= qobject_output_visitor_new(&obj
);
3117 visit_type_BlockdevOptions(v
, NULL
, &options
, &error_abort
);
3118 visit_complete(v
, &obj
);
3120 qdict
= qobject_to(QDict
, obj
);
3121 qdict_flatten(qdict
);
3123 /* bdrv_open_inherit() defaults to the values in bdrv_flags (for
3124 * compatibility with other callers) rather than what we want as the
3125 * real defaults. Apply the defaults here instead. */
3126 qdict_set_default_str(qdict
, BDRV_OPT_CACHE_DIRECT
, "off");
3127 qdict_set_default_str(qdict
, BDRV_OPT_CACHE_NO_FLUSH
, "off");
3128 qdict_set_default_str(qdict
, BDRV_OPT_READ_ONLY
, "off");
3129 qdict_set_default_str(qdict
, BDRV_OPT_AUTO_READ_ONLY
, "off");
3133 bs
= bdrv_open_inherit(NULL
, reference
, qdict
, 0, NULL
, NULL
, 0, errp
);
3140 static BlockDriverState
*bdrv_append_temp_snapshot(BlockDriverState
*bs
,
3142 QDict
*snapshot_options
,
3145 /* TODO: extra byte is a hack to ensure MAX_PATH space on Windows. */
3146 char *tmp_filename
= g_malloc0(PATH_MAX
+ 1);
3148 QemuOpts
*opts
= NULL
;
3149 BlockDriverState
*bs_snapshot
= NULL
;
3150 Error
*local_err
= NULL
;
3153 /* if snapshot, we create a temporary backing file and open it
3154 instead of opening 'filename' directly */
3156 /* Get the required size from the image */
3157 total_size
= bdrv_getlength(bs
);
3158 if (total_size
< 0) {
3159 error_setg_errno(errp
, -total_size
, "Could not get image size");
3163 /* Create the temporary image */
3164 ret
= get_tmp_filename(tmp_filename
, PATH_MAX
+ 1);
3166 error_setg_errno(errp
, -ret
, "Could not get temporary filename");
3170 opts
= qemu_opts_create(bdrv_qcow2
.create_opts
, NULL
, 0,
3172 qemu_opt_set_number(opts
, BLOCK_OPT_SIZE
, total_size
, &error_abort
);
3173 ret
= bdrv_create(&bdrv_qcow2
, tmp_filename
, opts
, errp
);
3174 qemu_opts_del(opts
);
3176 error_prepend(errp
, "Could not create temporary overlay '%s': ",
3181 /* Prepare options QDict for the temporary file */
3182 qdict_put_str(snapshot_options
, "file.driver", "file");
3183 qdict_put_str(snapshot_options
, "file.filename", tmp_filename
);
3184 qdict_put_str(snapshot_options
, "driver", "qcow2");
3186 bs_snapshot
= bdrv_open(NULL
, NULL
, snapshot_options
, flags
, errp
);
3187 snapshot_options
= NULL
;
3192 /* bdrv_append() consumes a strong reference to bs_snapshot
3193 * (i.e. it will call bdrv_unref() on it) even on error, so in
3194 * order to be able to return one, we have to increase
3195 * bs_snapshot's refcount here */
3196 bdrv_ref(bs_snapshot
);
3197 bdrv_append(bs_snapshot
, bs
, &local_err
);
3199 error_propagate(errp
, local_err
);
3205 qobject_unref(snapshot_options
);
3206 g_free(tmp_filename
);
3211 * Opens a disk image (raw, qcow2, vmdk, ...)
3213 * options is a QDict of options to pass to the block drivers, or NULL for an
3214 * empty set of options. The reference to the QDict belongs to the block layer
3215 * after the call (even on failure), so if the caller intends to reuse the
3216 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
3218 * If *pbs is NULL, a new BDS will be created with a pointer to it stored there.
3219 * If it is not NULL, the referenced BDS will be reused.
3221 * The reference parameter may be used to specify an existing block device which
3222 * should be opened. If specified, neither options nor a filename may be given,
3223 * nor can an existing BDS be reused (that is, *pbs has to be NULL).
3225 static BlockDriverState
*bdrv_open_inherit(const char *filename
,
3226 const char *reference
,
3227 QDict
*options
, int flags
,
3228 BlockDriverState
*parent
,
3229 const BdrvChildClass
*child_class
,
3230 BdrvChildRole child_role
,
3234 BlockBackend
*file
= NULL
;
3235 BlockDriverState
*bs
;
3236 BlockDriver
*drv
= NULL
;
3238 const char *drvname
;
3239 const char *backing
;
3240 Error
*local_err
= NULL
;
3241 QDict
*snapshot_options
= NULL
;
3242 int snapshot_flags
= 0;
3244 assert(!child_class
|| !flags
);
3245 assert(!child_class
== !parent
);
3248 bool options_non_empty
= options
? qdict_size(options
) : false;
3249 qobject_unref(options
);
3251 if (filename
|| options_non_empty
) {
3252 error_setg(errp
, "Cannot reference an existing block device with "
3253 "additional options or a new filename");
3257 bs
= bdrv_lookup_bs(reference
, reference
, errp
);
3268 /* NULL means an empty set of options */
3269 if (options
== NULL
) {
3270 options
= qdict_new();
3273 /* json: syntax counts as explicit options, as if in the QDict */
3274 parse_json_protocol(options
, &filename
, &local_err
);
3279 bs
->explicit_options
= qdict_clone_shallow(options
);
3282 bool parent_is_format
;
3285 parent_is_format
= parent
->drv
->is_format
;
3288 * parent->drv is not set yet because this node is opened for
3289 * (potential) format probing. That means that @parent is going
3290 * to be a format node.
3292 parent_is_format
= true;
3295 bs
->inherits_from
= parent
;
3296 child_class
->inherit_options(child_role
, parent_is_format
,
3298 parent
->open_flags
, parent
->options
);
3301 ret
= bdrv_fill_options(&options
, filename
, &flags
, &local_err
);
3307 * Set the BDRV_O_RDWR and BDRV_O_ALLOW_RDWR flags.
3308 * Caution: getting a boolean member of @options requires care.
3309 * When @options come from -blockdev or blockdev_add, members are
3310 * typed according to the QAPI schema, but when they come from
3311 * -drive, they're all QString.
3313 if (g_strcmp0(qdict_get_try_str(options
, BDRV_OPT_READ_ONLY
), "on") &&
3314 !qdict_get_try_bool(options
, BDRV_OPT_READ_ONLY
, false)) {
3315 flags
|= (BDRV_O_RDWR
| BDRV_O_ALLOW_RDWR
);
3317 flags
&= ~BDRV_O_RDWR
;
3320 if (flags
& BDRV_O_SNAPSHOT
) {
3321 snapshot_options
= qdict_new();
3322 bdrv_temp_snapshot_options(&snapshot_flags
, snapshot_options
,
3324 /* Let bdrv_backing_options() override "read-only" */
3325 qdict_del(options
, BDRV_OPT_READ_ONLY
);
3326 bdrv_inherited_options(BDRV_CHILD_COW
, true,
3327 &flags
, options
, flags
, options
);
3330 bs
->open_flags
= flags
;
3331 bs
->options
= options
;
3332 options
= qdict_clone_shallow(options
);
3334 /* Find the right image format driver */
3335 /* See cautionary note on accessing @options above */
3336 drvname
= qdict_get_try_str(options
, "driver");
3338 drv
= bdrv_find_format(drvname
);
3340 error_setg(errp
, "Unknown driver: '%s'", drvname
);
3345 assert(drvname
|| !(flags
& BDRV_O_PROTOCOL
));
3347 /* See cautionary note on accessing @options above */
3348 backing
= qdict_get_try_str(options
, "backing");
3349 if (qobject_to(QNull
, qdict_get(options
, "backing")) != NULL
||
3350 (backing
&& *backing
== '\0'))
3353 warn_report("Use of \"backing\": \"\" is deprecated; "
3354 "use \"backing\": null instead");
3356 flags
|= BDRV_O_NO_BACKING
;
3357 qdict_del(bs
->explicit_options
, "backing");
3358 qdict_del(bs
->options
, "backing");
3359 qdict_del(options
, "backing");
3362 /* Open image file without format layer. This BlockBackend is only used for
3363 * probing, the block drivers will do their own bdrv_open_child() for the
3364 * same BDS, which is why we put the node name back into options. */
3365 if ((flags
& BDRV_O_PROTOCOL
) == 0) {
3366 BlockDriverState
*file_bs
;
3368 file_bs
= bdrv_open_child_bs(filename
, options
, "file", bs
,
3369 &child_of_bds
, BDRV_CHILD_IMAGE
,
3374 if (file_bs
!= NULL
) {
3375 /* Not requesting BLK_PERM_CONSISTENT_READ because we're only
3376 * looking at the header to guess the image format. This works even
3377 * in cases where a guest would not see a consistent state. */
3378 file
= blk_new(bdrv_get_aio_context(file_bs
), 0, BLK_PERM_ALL
);
3379 blk_insert_bs(file
, file_bs
, &local_err
);
3380 bdrv_unref(file_bs
);
3385 qdict_put_str(options
, "file", bdrv_get_node_name(file_bs
));
3389 /* Image format probing */
3392 ret
= find_image_format(file
, filename
, &drv
, &local_err
);
3397 * This option update would logically belong in bdrv_fill_options(),
3398 * but we first need to open bs->file for the probing to work, while
3399 * opening bs->file already requires the (mostly) final set of options
3400 * so that cache mode etc. can be inherited.
3402 * Adding the driver later is somewhat ugly, but it's not an option
3403 * that would ever be inherited, so it's correct. We just need to make
3404 * sure to update both bs->options (which has the full effective
3405 * options for bs) and options (which has file.* already removed).
3407 qdict_put_str(bs
->options
, "driver", drv
->format_name
);
3408 qdict_put_str(options
, "driver", drv
->format_name
);
3410 error_setg(errp
, "Must specify either driver or file");
3414 /* BDRV_O_PROTOCOL must be set iff a protocol BDS is about to be created */
3415 assert(!!(flags
& BDRV_O_PROTOCOL
) == !!drv
->bdrv_file_open
);
3416 /* file must be NULL if a protocol BDS is about to be created
3417 * (the inverse results in an error message from bdrv_open_common()) */
3418 assert(!(flags
& BDRV_O_PROTOCOL
) || !file
);
3420 /* Open the image */
3421 ret
= bdrv_open_common(bs
, file
, options
, &local_err
);
3431 /* If there is a backing file, use it */
3432 if ((flags
& BDRV_O_NO_BACKING
) == 0) {
3433 ret
= bdrv_open_backing_file(bs
, options
, "backing", &local_err
);
3435 goto close_and_fail
;
3439 /* Remove all children options and references
3440 * from bs->options and bs->explicit_options */
3441 QLIST_FOREACH(child
, &bs
->children
, next
) {
3442 char *child_key_dot
;
3443 child_key_dot
= g_strdup_printf("%s.", child
->name
);
3444 qdict_extract_subqdict(bs
->explicit_options
, NULL
, child_key_dot
);
3445 qdict_extract_subqdict(bs
->options
, NULL
, child_key_dot
);
3446 qdict_del(bs
->explicit_options
, child
->name
);
3447 qdict_del(bs
->options
, child
->name
);
3448 g_free(child_key_dot
);
3451 /* Check if any unknown options were used */
3452 if (qdict_size(options
) != 0) {
3453 const QDictEntry
*entry
= qdict_first(options
);
3454 if (flags
& BDRV_O_PROTOCOL
) {
3455 error_setg(errp
, "Block protocol '%s' doesn't support the option "
3456 "'%s'", drv
->format_name
, entry
->key
);
3459 "Block format '%s' does not support the option '%s'",
3460 drv
->format_name
, entry
->key
);
3463 goto close_and_fail
;
3466 bdrv_parent_cb_change_media(bs
, true);
3468 qobject_unref(options
);
3471 /* For snapshot=on, create a temporary qcow2 overlay. bs points to the
3472 * temporary snapshot afterwards. */
3473 if (snapshot_flags
) {
3474 BlockDriverState
*snapshot_bs
;
3475 snapshot_bs
= bdrv_append_temp_snapshot(bs
, snapshot_flags
,
3476 snapshot_options
, &local_err
);
3477 snapshot_options
= NULL
;
3479 goto close_and_fail
;
3481 /* We are not going to return bs but the overlay on top of it
3482 * (snapshot_bs); thus, we have to drop the strong reference to bs
3483 * (which we obtained by calling bdrv_new()). bs will not be deleted,
3484 * though, because the overlay still has a reference to it. */
3493 qobject_unref(snapshot_options
);
3494 qobject_unref(bs
->explicit_options
);
3495 qobject_unref(bs
->options
);
3496 qobject_unref(options
);
3498 bs
->explicit_options
= NULL
;
3500 error_propagate(errp
, local_err
);
3505 qobject_unref(snapshot_options
);
3506 qobject_unref(options
);
3507 error_propagate(errp
, local_err
);
3511 BlockDriverState
*bdrv_open(const char *filename
, const char *reference
,
3512 QDict
*options
, int flags
, Error
**errp
)
3514 return bdrv_open_inherit(filename
, reference
, options
, flags
, NULL
,
3518 /* Return true if the NULL-terminated @list contains @str */
3519 static bool is_str_in_list(const char *str
, const char *const *list
)
3523 for (i
= 0; list
[i
] != NULL
; i
++) {
3524 if (!strcmp(str
, list
[i
])) {
3533 * Check that every option set in @bs->options is also set in
3536 * Options listed in the common_options list and in
3537 * @bs->drv->mutable_opts are skipped.
3539 * Return 0 on success, otherwise return -EINVAL and set @errp.
3541 static int bdrv_reset_options_allowed(BlockDriverState
*bs
,
3542 const QDict
*new_opts
, Error
**errp
)
3544 const QDictEntry
*e
;
3545 /* These options are common to all block drivers and are handled
3546 * in bdrv_reopen_prepare() so they can be left out of @new_opts */
3547 const char *const common_options
[] = {
3548 "node-name", "discard", "cache.direct", "cache.no-flush",
3549 "read-only", "auto-read-only", "detect-zeroes", NULL
3552 for (e
= qdict_first(bs
->options
); e
; e
= qdict_next(bs
->options
, e
)) {
3553 if (!qdict_haskey(new_opts
, e
->key
) &&
3554 !is_str_in_list(e
->key
, common_options
) &&
3555 !is_str_in_list(e
->key
, bs
->drv
->mutable_opts
)) {
3556 error_setg(errp
, "Option '%s' cannot be reset "
3557 "to its default value", e
->key
);
3566 * Returns true if @child can be reached recursively from @bs
3568 static bool bdrv_recurse_has_child(BlockDriverState
*bs
,
3569 BlockDriverState
*child
)
3577 QLIST_FOREACH(c
, &bs
->children
, next
) {
3578 if (bdrv_recurse_has_child(c
->bs
, child
)) {
3587 * Adds a BlockDriverState to a simple queue for an atomic, transactional
3588 * reopen of multiple devices.
3590 * bs_queue can either be an existing BlockReopenQueue that has had QTAILQ_INIT
3591 * already performed, or alternatively may be NULL a new BlockReopenQueue will
3592 * be created and initialized. This newly created BlockReopenQueue should be
3593 * passed back in for subsequent calls that are intended to be of the same
3596 * bs is the BlockDriverState to add to the reopen queue.
3598 * options contains the changed options for the associated bs
3599 * (the BlockReopenQueue takes ownership)
3601 * flags contains the open flags for the associated bs
3603 * returns a pointer to bs_queue, which is either the newly allocated
3604 * bs_queue, or the existing bs_queue being used.
3606 * bs must be drained between bdrv_reopen_queue() and bdrv_reopen_multiple().
3608 static BlockReopenQueue
*bdrv_reopen_queue_child(BlockReopenQueue
*bs_queue
,
3609 BlockDriverState
*bs
,
3611 const BdrvChildClass
*klass
,
3613 bool parent_is_format
,
3614 QDict
*parent_options
,
3620 BlockReopenQueueEntry
*bs_entry
;
3622 QDict
*old_options
, *explicit_options
, *options_copy
;
3626 /* Make sure that the caller remembered to use a drained section. This is
3627 * important to avoid graph changes between the recursive queuing here and
3628 * bdrv_reopen_multiple(). */
3629 assert(bs
->quiesce_counter
> 0);
3631 if (bs_queue
== NULL
) {
3632 bs_queue
= g_new0(BlockReopenQueue
, 1);
3633 QTAILQ_INIT(bs_queue
);
3637 options
= qdict_new();
3640 /* Check if this BlockDriverState is already in the queue */
3641 QTAILQ_FOREACH(bs_entry
, bs_queue
, entry
) {
3642 if (bs
== bs_entry
->state
.bs
) {
3648 * Precedence of options:
3649 * 1. Explicitly passed in options (highest)
3650 * 2. Retained from explicitly set options of bs
3651 * 3. Inherited from parent node
3652 * 4. Retained from effective options of bs
3655 /* Old explicitly set values (don't overwrite by inherited value) */
3656 if (bs_entry
|| keep_old_opts
) {
3657 old_options
= qdict_clone_shallow(bs_entry
?
3658 bs_entry
->state
.explicit_options
:
3659 bs
->explicit_options
);
3660 bdrv_join_options(bs
, options
, old_options
);
3661 qobject_unref(old_options
);
3664 explicit_options
= qdict_clone_shallow(options
);
3666 /* Inherit from parent node */
3667 if (parent_options
) {
3669 klass
->inherit_options(role
, parent_is_format
, &flags
, options
,
3670 parent_flags
, parent_options
);
3672 flags
= bdrv_get_flags(bs
);
3675 if (keep_old_opts
) {
3676 /* Old values are used for options that aren't set yet */
3677 old_options
= qdict_clone_shallow(bs
->options
);
3678 bdrv_join_options(bs
, options
, old_options
);
3679 qobject_unref(old_options
);
3682 /* We have the final set of options so let's update the flags */
3683 options_copy
= qdict_clone_shallow(options
);
3684 opts
= qemu_opts_create(&bdrv_runtime_opts
, NULL
, 0, &error_abort
);
3685 qemu_opts_absorb_qdict(opts
, options_copy
, NULL
);
3686 update_flags_from_options(&flags
, opts
);
3687 qemu_opts_del(opts
);
3688 qobject_unref(options_copy
);
3690 /* bdrv_open_inherit() sets and clears some additional flags internally */
3691 flags
&= ~BDRV_O_PROTOCOL
;
3692 if (flags
& BDRV_O_RDWR
) {
3693 flags
|= BDRV_O_ALLOW_RDWR
;
3697 bs_entry
= g_new0(BlockReopenQueueEntry
, 1);
3698 QTAILQ_INSERT_TAIL(bs_queue
, bs_entry
, entry
);
3700 qobject_unref(bs_entry
->state
.options
);
3701 qobject_unref(bs_entry
->state
.explicit_options
);
3704 bs_entry
->state
.bs
= bs
;
3705 bs_entry
->state
.options
= options
;
3706 bs_entry
->state
.explicit_options
= explicit_options
;
3707 bs_entry
->state
.flags
= flags
;
3709 /* This needs to be overwritten in bdrv_reopen_prepare() */
3710 bs_entry
->state
.perm
= UINT64_MAX
;
3711 bs_entry
->state
.shared_perm
= 0;
3714 * If keep_old_opts is false then it means that unspecified
3715 * options must be reset to their original value. We don't allow
3716 * resetting 'backing' but we need to know if the option is
3717 * missing in order to decide if we have to return an error.
3719 if (!keep_old_opts
) {
3720 bs_entry
->state
.backing_missing
=
3721 !qdict_haskey(options
, "backing") &&
3722 !qdict_haskey(options
, "backing.driver");
3725 QLIST_FOREACH(child
, &bs
->children
, next
) {
3726 QDict
*new_child_options
= NULL
;
3727 bool child_keep_old
= keep_old_opts
;
3729 /* reopen can only change the options of block devices that were
3730 * implicitly created and inherited options. For other (referenced)
3731 * block devices, a syntax like "backing.foo" results in an error. */
3732 if (child
->bs
->inherits_from
!= bs
) {
3736 /* Check if the options contain a child reference */
3737 if (qdict_haskey(options
, child
->name
)) {
3738 const char *childref
= qdict_get_try_str(options
, child
->name
);
3740 * The current child must not be reopened if the child
3741 * reference is null or points to a different node.
3743 if (g_strcmp0(childref
, child
->bs
->node_name
)) {
3747 * If the child reference points to the current child then
3748 * reopen it with its existing set of options (note that
3749 * it can still inherit new options from the parent).
3751 child_keep_old
= true;
3753 /* Extract child options ("child-name.*") */
3754 char *child_key_dot
= g_strdup_printf("%s.", child
->name
);
3755 qdict_extract_subqdict(explicit_options
, NULL
, child_key_dot
);
3756 qdict_extract_subqdict(options
, &new_child_options
, child_key_dot
);
3757 g_free(child_key_dot
);
3760 bdrv_reopen_queue_child(bs_queue
, child
->bs
, new_child_options
,
3761 child
->klass
, child
->role
, bs
->drv
->is_format
,
3762 options
, flags
, child_keep_old
);
3768 BlockReopenQueue
*bdrv_reopen_queue(BlockReopenQueue
*bs_queue
,
3769 BlockDriverState
*bs
,
3770 QDict
*options
, bool keep_old_opts
)
3772 return bdrv_reopen_queue_child(bs_queue
, bs
, options
, NULL
, 0, false,
3773 NULL
, 0, keep_old_opts
);
3777 * Reopen multiple BlockDriverStates atomically & transactionally.
3779 * The queue passed in (bs_queue) must have been built up previous
3780 * via bdrv_reopen_queue().
3782 * Reopens all BDS specified in the queue, with the appropriate
3783 * flags. All devices are prepared for reopen, and failure of any
3784 * device will cause all device changes to be abandoned, and intermediate
3787 * If all devices prepare successfully, then the changes are committed
3790 * All affected nodes must be drained between bdrv_reopen_queue() and
3791 * bdrv_reopen_multiple().
3793 int bdrv_reopen_multiple(BlockReopenQueue
*bs_queue
, Error
**errp
)
3796 BlockReopenQueueEntry
*bs_entry
, *next
;
3798 assert(bs_queue
!= NULL
);
3800 QTAILQ_FOREACH(bs_entry
, bs_queue
, entry
) {
3801 assert(bs_entry
->state
.bs
->quiesce_counter
> 0);
3802 if (bdrv_reopen_prepare(&bs_entry
->state
, bs_queue
, errp
)) {
3805 bs_entry
->prepared
= true;
3808 QTAILQ_FOREACH(bs_entry
, bs_queue
, entry
) {
3809 BDRVReopenState
*state
= &bs_entry
->state
;
3810 ret
= bdrv_check_perm(state
->bs
, bs_queue
, state
->perm
,
3811 state
->shared_perm
, NULL
, NULL
, errp
);
3815 /* Check if new_backing_bs would accept the new permissions */
3816 if (state
->replace_backing_bs
&& state
->new_backing_bs
) {
3817 uint64_t nperm
, nshared
;
3818 bdrv_child_perm(state
->bs
, state
->new_backing_bs
,
3819 NULL
, bdrv_backing_role(state
->bs
),
3820 bs_queue
, state
->perm
, state
->shared_perm
,
3822 ret
= bdrv_check_update_perm(state
->new_backing_bs
, NULL
,
3823 nperm
, nshared
, NULL
, NULL
, errp
);
3828 bs_entry
->perms_checked
= true;
3832 * If we reach this point, we have success and just need to apply the
3835 * Reverse order is used to comfort qcow2 driver: on commit it need to write
3836 * IN_USE flag to the image, to mark bitmaps in the image as invalid. But
3837 * children are usually goes after parents in reopen-queue, so go from last
3840 QTAILQ_FOREACH_REVERSE(bs_entry
, bs_queue
, entry
) {
3841 bdrv_reopen_commit(&bs_entry
->state
);
3846 QTAILQ_FOREACH_SAFE(bs_entry
, bs_queue
, entry
, next
) {
3847 BDRVReopenState
*state
= &bs_entry
->state
;
3849 if (!bs_entry
->perms_checked
) {
3854 bdrv_set_perm(state
->bs
, state
->perm
, state
->shared_perm
);
3856 bdrv_abort_perm_update(state
->bs
);
3857 if (state
->replace_backing_bs
&& state
->new_backing_bs
) {
3858 bdrv_abort_perm_update(state
->new_backing_bs
);
3864 QTAILQ_FOREACH_REVERSE(bs_entry
, bs_queue
, entry
) {
3865 BlockDriverState
*bs
= bs_entry
->state
.bs
;
3867 if (bs
->drv
->bdrv_reopen_commit_post
)
3868 bs
->drv
->bdrv_reopen_commit_post(&bs_entry
->state
);
3872 QTAILQ_FOREACH_SAFE(bs_entry
, bs_queue
, entry
, next
) {
3874 if (bs_entry
->prepared
) {
3875 bdrv_reopen_abort(&bs_entry
->state
);
3877 qobject_unref(bs_entry
->state
.explicit_options
);
3878 qobject_unref(bs_entry
->state
.options
);
3880 if (bs_entry
->state
.new_backing_bs
) {
3881 bdrv_unref(bs_entry
->state
.new_backing_bs
);
3890 int bdrv_reopen_set_read_only(BlockDriverState
*bs
, bool read_only
,
3894 BlockReopenQueue
*queue
;
3895 QDict
*opts
= qdict_new();
3897 qdict_put_bool(opts
, BDRV_OPT_READ_ONLY
, read_only
);
3899 bdrv_subtree_drained_begin(bs
);
3900 queue
= bdrv_reopen_queue(NULL
, bs
, opts
, true);
3901 ret
= bdrv_reopen_multiple(queue
, errp
);
3902 bdrv_subtree_drained_end(bs
);
3907 static BlockReopenQueueEntry
*find_parent_in_reopen_queue(BlockReopenQueue
*q
,
3910 BlockReopenQueueEntry
*entry
;
3912 QTAILQ_FOREACH(entry
, q
, entry
) {
3913 BlockDriverState
*bs
= entry
->state
.bs
;
3916 QLIST_FOREACH(child
, &bs
->children
, next
) {
3926 static void bdrv_reopen_perm(BlockReopenQueue
*q
, BlockDriverState
*bs
,
3927 uint64_t *perm
, uint64_t *shared
)
3930 BlockReopenQueueEntry
*parent
;
3931 uint64_t cumulative_perms
= 0;
3932 uint64_t cumulative_shared_perms
= BLK_PERM_ALL
;
3934 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
3935 parent
= find_parent_in_reopen_queue(q
, c
);
3937 cumulative_perms
|= c
->perm
;
3938 cumulative_shared_perms
&= c
->shared_perm
;
3940 uint64_t nperm
, nshared
;
3942 bdrv_child_perm(parent
->state
.bs
, bs
, c
, c
->role
, q
,
3943 parent
->state
.perm
, parent
->state
.shared_perm
,
3946 cumulative_perms
|= nperm
;
3947 cumulative_shared_perms
&= nshared
;
3950 *perm
= cumulative_perms
;
3951 *shared
= cumulative_shared_perms
;
3954 static bool bdrv_reopen_can_attach(BlockDriverState
*parent
,
3956 BlockDriverState
*new_child
,
3959 AioContext
*parent_ctx
= bdrv_get_aio_context(parent
);
3960 AioContext
*child_ctx
= bdrv_get_aio_context(new_child
);
3964 ignore
= g_slist_prepend(NULL
, child
);
3965 ret
= bdrv_can_set_aio_context(new_child
, parent_ctx
, &ignore
, NULL
);
3966 g_slist_free(ignore
);
3971 ignore
= g_slist_prepend(NULL
, child
);
3972 ret
= bdrv_can_set_aio_context(parent
, child_ctx
, &ignore
, errp
);
3973 g_slist_free(ignore
);
3978 * Take a BDRVReopenState and check if the value of 'backing' in the
3979 * reopen_state->options QDict is valid or not.
3981 * If 'backing' is missing from the QDict then return 0.
3983 * If 'backing' contains the node name of the backing file of
3984 * reopen_state->bs then return 0.
3986 * If 'backing' contains a different node name (or is null) then check
3987 * whether the current backing file can be replaced with the new one.
3988 * If that's the case then reopen_state->replace_backing_bs is set to
3989 * true and reopen_state->new_backing_bs contains a pointer to the new
3990 * backing BlockDriverState (or NULL).
3992 * Return 0 on success, otherwise return < 0 and set @errp.
3994 static int bdrv_reopen_parse_backing(BDRVReopenState
*reopen_state
,
3997 BlockDriverState
*bs
= reopen_state
->bs
;
3998 BlockDriverState
*overlay_bs
, *new_backing_bs
;
4002 value
= qdict_get(reopen_state
->options
, "backing");
4003 if (value
== NULL
) {
4007 switch (qobject_type(value
)) {
4009 new_backing_bs
= NULL
;
4012 str
= qobject_get_try_str(value
);
4013 new_backing_bs
= bdrv_lookup_bs(NULL
, str
, errp
);
4014 if (new_backing_bs
== NULL
) {
4016 } else if (bdrv_recurse_has_child(new_backing_bs
, bs
)) {
4017 error_setg(errp
, "Making '%s' a backing file of '%s' "
4018 "would create a cycle", str
, bs
->node_name
);
4023 /* 'backing' does not allow any other data type */
4024 g_assert_not_reached();
4028 * Check AioContext compatibility so that the bdrv_set_backing_hd() call in
4029 * bdrv_reopen_commit() won't fail.
4031 if (new_backing_bs
) {
4032 if (!bdrv_reopen_can_attach(bs
, bs
->backing
, new_backing_bs
, errp
)) {
4038 * Find the "actual" backing file by skipping all links that point
4039 * to an implicit node, if any (e.g. a commit filter node).
4042 while (backing_bs(overlay_bs
) && backing_bs(overlay_bs
)->implicit
) {
4043 overlay_bs
= backing_bs(overlay_bs
);
4046 /* If we want to replace the backing file we need some extra checks */
4047 if (new_backing_bs
!= backing_bs(overlay_bs
)) {
4048 /* Check for implicit nodes between bs and its backing file */
4049 if (bs
!= overlay_bs
) {
4050 error_setg(errp
, "Cannot change backing link if '%s' has "
4051 "an implicit backing file", bs
->node_name
);
4054 /* Check if the backing link that we want to replace is frozen */
4055 if (bdrv_is_backing_chain_frozen(overlay_bs
, backing_bs(overlay_bs
),
4059 reopen_state
->replace_backing_bs
= true;
4060 if (new_backing_bs
) {
4061 bdrv_ref(new_backing_bs
);
4062 reopen_state
->new_backing_bs
= new_backing_bs
;
4070 * Prepares a BlockDriverState for reopen. All changes are staged in the
4071 * 'opaque' field of the BDRVReopenState, which is used and allocated by
4072 * the block driver layer .bdrv_reopen_prepare()
4074 * bs is the BlockDriverState to reopen
4075 * flags are the new open flags
4076 * queue is the reopen queue
4078 * Returns 0 on success, non-zero on error. On error errp will be set
4081 * On failure, bdrv_reopen_abort() will be called to clean up any data.
4082 * It is the responsibility of the caller to then call the abort() or
4083 * commit() for any other BDS that have been left in a prepare() state
4086 int bdrv_reopen_prepare(BDRVReopenState
*reopen_state
, BlockReopenQueue
*queue
,
4091 Error
*local_err
= NULL
;
4094 QDict
*orig_reopen_opts
;
4095 char *discard
= NULL
;
4097 bool drv_prepared
= false;
4099 assert(reopen_state
!= NULL
);
4100 assert(reopen_state
->bs
->drv
!= NULL
);
4101 drv
= reopen_state
->bs
->drv
;
4103 /* This function and each driver's bdrv_reopen_prepare() remove
4104 * entries from reopen_state->options as they are processed, so
4105 * we need to make a copy of the original QDict. */
4106 orig_reopen_opts
= qdict_clone_shallow(reopen_state
->options
);
4108 /* Process generic block layer options */
4109 opts
= qemu_opts_create(&bdrv_runtime_opts
, NULL
, 0, &error_abort
);
4110 if (!qemu_opts_absorb_qdict(opts
, reopen_state
->options
, errp
)) {
4115 /* This was already called in bdrv_reopen_queue_child() so the flags
4116 * are up-to-date. This time we simply want to remove the options from
4117 * QemuOpts in order to indicate that they have been processed. */
4118 old_flags
= reopen_state
->flags
;
4119 update_flags_from_options(&reopen_state
->flags
, opts
);
4120 assert(old_flags
== reopen_state
->flags
);
4122 discard
= qemu_opt_get_del(opts
, BDRV_OPT_DISCARD
);
4123 if (discard
!= NULL
) {
4124 if (bdrv_parse_discard_flags(discard
, &reopen_state
->flags
) != 0) {
4125 error_setg(errp
, "Invalid discard option");
4131 reopen_state
->detect_zeroes
=
4132 bdrv_parse_detect_zeroes(opts
, reopen_state
->flags
, &local_err
);
4134 error_propagate(errp
, local_err
);
4139 /* All other options (including node-name and driver) must be unchanged.
4140 * Put them back into the QDict, so that they are checked at the end
4141 * of this function. */
4142 qemu_opts_to_qdict(opts
, reopen_state
->options
);
4144 /* If we are to stay read-only, do not allow permission change
4145 * to r/w. Attempting to set to r/w may fail if either BDRV_O_ALLOW_RDWR is
4146 * not set, or if the BDS still has copy_on_read enabled */
4147 read_only
= !(reopen_state
->flags
& BDRV_O_RDWR
);
4148 ret
= bdrv_can_set_read_only(reopen_state
->bs
, read_only
, true, &local_err
);
4150 error_propagate(errp
, local_err
);
4154 /* Calculate required permissions after reopening */
4155 bdrv_reopen_perm(queue
, reopen_state
->bs
,
4156 &reopen_state
->perm
, &reopen_state
->shared_perm
);
4158 ret
= bdrv_flush(reopen_state
->bs
);
4160 error_setg_errno(errp
, -ret
, "Error flushing drive");
4164 if (drv
->bdrv_reopen_prepare
) {
4166 * If a driver-specific option is missing, it means that we
4167 * should reset it to its default value.
4168 * But not all options allow that, so we need to check it first.
4170 ret
= bdrv_reset_options_allowed(reopen_state
->bs
,
4171 reopen_state
->options
, errp
);
4176 ret
= drv
->bdrv_reopen_prepare(reopen_state
, queue
, &local_err
);
4178 if (local_err
!= NULL
) {
4179 error_propagate(errp
, local_err
);
4181 bdrv_refresh_filename(reopen_state
->bs
);
4182 error_setg(errp
, "failed while preparing to reopen image '%s'",
4183 reopen_state
->bs
->filename
);
4188 /* It is currently mandatory to have a bdrv_reopen_prepare()
4189 * handler for each supported drv. */
4190 error_setg(errp
, "Block format '%s' used by node '%s' "
4191 "does not support reopening files", drv
->format_name
,
4192 bdrv_get_device_or_node_name(reopen_state
->bs
));
4197 drv_prepared
= true;
4200 * We must provide the 'backing' option if the BDS has a backing
4201 * file or if the image file has a backing file name as part of
4202 * its metadata. Otherwise the 'backing' option can be omitted.
4204 if (drv
->supports_backing
&& reopen_state
->backing_missing
&&
4205 (backing_bs(reopen_state
->bs
) || reopen_state
->bs
->backing_file
[0])) {
4206 error_setg(errp
, "backing is missing for '%s'",
4207 reopen_state
->bs
->node_name
);
4213 * Allow changing the 'backing' option. The new value can be
4214 * either a reference to an existing node (using its node name)
4215 * or NULL to simply detach the current backing file.
4217 ret
= bdrv_reopen_parse_backing(reopen_state
, errp
);
4221 qdict_del(reopen_state
->options
, "backing");
4223 /* Options that are not handled are only okay if they are unchanged
4224 * compared to the old state. It is expected that some options are only
4225 * used for the initial open, but not reopen (e.g. filename) */
4226 if (qdict_size(reopen_state
->options
)) {
4227 const QDictEntry
*entry
= qdict_first(reopen_state
->options
);
4230 QObject
*new = entry
->value
;
4231 QObject
*old
= qdict_get(reopen_state
->bs
->options
, entry
->key
);
4233 /* Allow child references (child_name=node_name) as long as they
4234 * point to the current child (i.e. everything stays the same). */
4235 if (qobject_type(new) == QTYPE_QSTRING
) {
4237 QLIST_FOREACH(child
, &reopen_state
->bs
->children
, next
) {
4238 if (!strcmp(child
->name
, entry
->key
)) {
4244 const char *str
= qobject_get_try_str(new);
4245 if (!strcmp(child
->bs
->node_name
, str
)) {
4246 continue; /* Found child with this name, skip option */
4252 * TODO: When using -drive to specify blockdev options, all values
4253 * will be strings; however, when using -blockdev, blockdev-add or
4254 * filenames using the json:{} pseudo-protocol, they will be
4256 * In contrast, reopening options are (currently) always strings
4257 * (because you can only specify them through qemu-io; all other
4258 * callers do not specify any options).
4259 * Therefore, when using anything other than -drive to create a BDS,
4260 * this cannot detect non-string options as unchanged, because
4261 * qobject_is_equal() always returns false for objects of different
4262 * type. In the future, this should be remedied by correctly typing
4263 * all options. For now, this is not too big of an issue because
4264 * the user can simply omit options which cannot be changed anyway,
4265 * so they will stay unchanged.
4267 if (!qobject_is_equal(new, old
)) {
4268 error_setg(errp
, "Cannot change the option '%s'", entry
->key
);
4272 } while ((entry
= qdict_next(reopen_state
->options
, entry
)));
4277 /* Restore the original reopen_state->options QDict */
4278 qobject_unref(reopen_state
->options
);
4279 reopen_state
->options
= qobject_ref(orig_reopen_opts
);
4282 if (ret
< 0 && drv_prepared
) {
4283 /* drv->bdrv_reopen_prepare() has succeeded, so we need to
4284 * call drv->bdrv_reopen_abort() before signaling an error
4285 * (bdrv_reopen_multiple() will not call bdrv_reopen_abort()
4286 * when the respective bdrv_reopen_prepare() has failed) */
4287 if (drv
->bdrv_reopen_abort
) {
4288 drv
->bdrv_reopen_abort(reopen_state
);
4291 qemu_opts_del(opts
);
4292 qobject_unref(orig_reopen_opts
);
4298 * Takes the staged changes for the reopen from bdrv_reopen_prepare(), and
4299 * makes them final by swapping the staging BlockDriverState contents into
4300 * the active BlockDriverState contents.
4302 void bdrv_reopen_commit(BDRVReopenState
*reopen_state
)
4305 BlockDriverState
*bs
;
4308 assert(reopen_state
!= NULL
);
4309 bs
= reopen_state
->bs
;
4311 assert(drv
!= NULL
);
4313 /* If there are any driver level actions to take */
4314 if (drv
->bdrv_reopen_commit
) {
4315 drv
->bdrv_reopen_commit(reopen_state
);
4318 /* set BDS specific flags now */
4319 qobject_unref(bs
->explicit_options
);
4320 qobject_unref(bs
->options
);
4322 bs
->explicit_options
= reopen_state
->explicit_options
;
4323 bs
->options
= reopen_state
->options
;
4324 bs
->open_flags
= reopen_state
->flags
;
4325 bs
->read_only
= !(reopen_state
->flags
& BDRV_O_RDWR
);
4326 bs
->detect_zeroes
= reopen_state
->detect_zeroes
;
4328 if (reopen_state
->replace_backing_bs
) {
4329 qdict_del(bs
->explicit_options
, "backing");
4330 qdict_del(bs
->options
, "backing");
4333 /* Remove child references from bs->options and bs->explicit_options.
4334 * Child options were already removed in bdrv_reopen_queue_child() */
4335 QLIST_FOREACH(child
, &bs
->children
, next
) {
4336 qdict_del(bs
->explicit_options
, child
->name
);
4337 qdict_del(bs
->options
, child
->name
);
4341 * Change the backing file if a new one was specified. We do this
4342 * after updating bs->options, so bdrv_refresh_filename() (called
4343 * from bdrv_set_backing_hd()) has the new values.
4345 if (reopen_state
->replace_backing_bs
) {
4346 BlockDriverState
*old_backing_bs
= backing_bs(bs
);
4347 assert(!old_backing_bs
|| !old_backing_bs
->implicit
);
4348 /* Abort the permission update on the backing bs we're detaching */
4349 if (old_backing_bs
) {
4350 bdrv_abort_perm_update(old_backing_bs
);
4352 bdrv_set_backing_hd(bs
, reopen_state
->new_backing_bs
, &error_abort
);
4355 bdrv_refresh_limits(bs
, NULL
);
4359 * Abort the reopen, and delete and free the staged changes in
4362 void bdrv_reopen_abort(BDRVReopenState
*reopen_state
)
4366 assert(reopen_state
!= NULL
);
4367 drv
= reopen_state
->bs
->drv
;
4368 assert(drv
!= NULL
);
4370 if (drv
->bdrv_reopen_abort
) {
4371 drv
->bdrv_reopen_abort(reopen_state
);
4376 static void bdrv_close(BlockDriverState
*bs
)
4378 BdrvAioNotifier
*ban
, *ban_next
;
4379 BdrvChild
*child
, *next
;
4381 assert(!bs
->refcnt
);
4383 bdrv_drained_begin(bs
); /* complete I/O */
4385 bdrv_drain(bs
); /* in case flush left pending I/O */
4388 if (bs
->drv
->bdrv_close
) {
4389 bs
->drv
->bdrv_close(bs
);
4394 QLIST_FOREACH_SAFE(child
, &bs
->children
, next
, next
) {
4395 bdrv_unref_child(bs
, child
);
4402 atomic_set(&bs
->copy_on_read
, 0);
4403 bs
->backing_file
[0] = '\0';
4404 bs
->backing_format
[0] = '\0';
4405 bs
->total_sectors
= 0;
4406 bs
->encrypted
= false;
4408 qobject_unref(bs
->options
);
4409 qobject_unref(bs
->explicit_options
);
4411 bs
->explicit_options
= NULL
;
4412 qobject_unref(bs
->full_open_options
);
4413 bs
->full_open_options
= NULL
;
4415 bdrv_release_named_dirty_bitmaps(bs
);
4416 assert(QLIST_EMPTY(&bs
->dirty_bitmaps
));
4418 QLIST_FOREACH_SAFE(ban
, &bs
->aio_notifiers
, list
, ban_next
) {
4421 QLIST_INIT(&bs
->aio_notifiers
);
4422 bdrv_drained_end(bs
);
4425 void bdrv_close_all(void)
4427 assert(job_next(NULL
) == NULL
);
4428 nbd_export_close_all();
4430 /* Drop references from requests still in flight, such as canceled block
4431 * jobs whose AIO context has not been polled yet */
4434 blk_remove_all_bs();
4435 blockdev_close_all_bdrv_states();
4437 assert(QTAILQ_EMPTY(&all_bdrv_states
));
4440 static bool should_update_child(BdrvChild
*c
, BlockDriverState
*to
)
4446 if (c
->klass
->stay_at_node
) {
4450 /* If the child @c belongs to the BDS @to, replacing the current
4451 * c->bs by @to would mean to create a loop.
4453 * Such a case occurs when appending a BDS to a backing chain.
4454 * For instance, imagine the following chain:
4456 * guest device -> node A -> further backing chain...
4458 * Now we create a new BDS B which we want to put on top of this
4459 * chain, so we first attach A as its backing node:
4464 * guest device -> node A -> further backing chain...
4466 * Finally we want to replace A by B. When doing that, we want to
4467 * replace all pointers to A by pointers to B -- except for the
4468 * pointer from B because (1) that would create a loop, and (2)
4469 * that pointer should simply stay intact:
4471 * guest device -> node B
4474 * node A -> further backing chain...
4476 * In general, when replacing a node A (c->bs) by a node B (@to),
4477 * if A is a child of B, that means we cannot replace A by B there
4478 * because that would create a loop. Silently detaching A from B
4479 * is also not really an option. So overall just leaving A in
4480 * place there is the most sensible choice.
4482 * We would also create a loop in any cases where @c is only
4483 * indirectly referenced by @to. Prevent this by returning false
4484 * if @c is found (by breadth-first search) anywhere in the whole
4489 found
= g_hash_table_new(NULL
, NULL
);
4490 g_hash_table_add(found
, to
);
4491 queue
= g_queue_new();
4492 g_queue_push_tail(queue
, to
);
4494 while (!g_queue_is_empty(queue
)) {
4495 BlockDriverState
*v
= g_queue_pop_head(queue
);
4498 QLIST_FOREACH(c2
, &v
->children
, next
) {
4504 if (g_hash_table_contains(found
, c2
->bs
)) {
4508 g_queue_push_tail(queue
, c2
->bs
);
4509 g_hash_table_add(found
, c2
->bs
);
4513 g_queue_free(queue
);
4514 g_hash_table_destroy(found
);
4519 void bdrv_replace_node(BlockDriverState
*from
, BlockDriverState
*to
,
4522 BdrvChild
*c
, *next
;
4523 GSList
*list
= NULL
, *p
;
4524 uint64_t perm
= 0, shared
= BLK_PERM_ALL
;
4527 /* Make sure that @from doesn't go away until we have successfully attached
4528 * all of its parents to @to. */
4531 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
4532 assert(bdrv_get_aio_context(from
) == bdrv_get_aio_context(to
));
4533 bdrv_drained_begin(from
);
4535 /* Put all parents into @list and calculate their cumulative permissions */
4536 QLIST_FOREACH_SAFE(c
, &from
->parents
, next_parent
, next
) {
4537 assert(c
->bs
== from
);
4538 if (!should_update_child(c
, to
)) {
4542 error_setg(errp
, "Cannot change '%s' link to '%s'",
4543 c
->name
, from
->node_name
);
4546 list
= g_slist_prepend(list
, c
);
4548 shared
&= c
->shared_perm
;
4551 /* Check whether the required permissions can be granted on @to, ignoring
4552 * all BdrvChild in @list so that they can't block themselves. */
4553 ret
= bdrv_check_update_perm(to
, NULL
, perm
, shared
, list
, NULL
, errp
);
4555 bdrv_abort_perm_update(to
);
4559 /* Now actually perform the change. We performed the permission check for
4560 * all elements of @list at once, so set the permissions all at once at the
4562 for (p
= list
; p
!= NULL
; p
= p
->next
) {
4566 bdrv_replace_child_noperm(c
, to
);
4570 bdrv_get_cumulative_perm(to
, &perm
, &shared
);
4571 bdrv_set_perm(to
, perm
, shared
);
4575 bdrv_drained_end(from
);
4580 * Add new bs contents at the top of an image chain while the chain is
4581 * live, while keeping required fields on the top layer.
4583 * This will modify the BlockDriverState fields, and swap contents
4584 * between bs_new and bs_top. Both bs_new and bs_top are modified.
4586 * bs_new must not be attached to a BlockBackend.
4588 * This function does not create any image files.
4590 * bdrv_append() takes ownership of a bs_new reference and unrefs it because
4591 * that's what the callers commonly need. bs_new will be referenced by the old
4592 * parents of bs_top after bdrv_append() returns. If the caller needs to keep a
4593 * reference of its own, it must call bdrv_ref().
4595 void bdrv_append(BlockDriverState
*bs_new
, BlockDriverState
*bs_top
,
4598 Error
*local_err
= NULL
;
4600 bdrv_set_backing_hd(bs_new
, bs_top
, &local_err
);
4602 error_propagate(errp
, local_err
);
4606 bdrv_replace_node(bs_top
, bs_new
, &local_err
);
4608 error_propagate(errp
, local_err
);
4609 bdrv_set_backing_hd(bs_new
, NULL
, &error_abort
);
4613 /* bs_new is now referenced by its new parents, we don't need the
4614 * additional reference any more. */
4619 static void bdrv_delete(BlockDriverState
*bs
)
4621 assert(bdrv_op_blocker_is_empty(bs
));
4622 assert(!bs
->refcnt
);
4624 /* remove from list, if necessary */
4625 if (bs
->node_name
[0] != '\0') {
4626 QTAILQ_REMOVE(&graph_bdrv_states
, bs
, node_list
);
4628 QTAILQ_REMOVE(&all_bdrv_states
, bs
, bs_list
);
4636 * Run consistency checks on an image
4638 * Returns 0 if the check could be completed (it doesn't mean that the image is
4639 * free of errors) or -errno when an internal error occurred. The results of the
4640 * check are stored in res.
4642 static int coroutine_fn
bdrv_co_check(BlockDriverState
*bs
,
4643 BdrvCheckResult
*res
, BdrvCheckMode fix
)
4645 if (bs
->drv
== NULL
) {
4648 if (bs
->drv
->bdrv_co_check
== NULL
) {
4652 memset(res
, 0, sizeof(*res
));
4653 return bs
->drv
->bdrv_co_check(bs
, res
, fix
);
4656 typedef struct CheckCo
{
4657 BlockDriverState
*bs
;
4658 BdrvCheckResult
*res
;
4663 static void coroutine_fn
bdrv_check_co_entry(void *opaque
)
4665 CheckCo
*cco
= opaque
;
4666 cco
->ret
= bdrv_co_check(cco
->bs
, cco
->res
, cco
->fix
);
4670 int bdrv_check(BlockDriverState
*bs
,
4671 BdrvCheckResult
*res
, BdrvCheckMode fix
)
4677 .ret
= -EINPROGRESS
,
4681 if (qemu_in_coroutine()) {
4682 /* Fast-path if already in coroutine context */
4683 bdrv_check_co_entry(&cco
);
4685 co
= qemu_coroutine_create(bdrv_check_co_entry
, &cco
);
4686 bdrv_coroutine_enter(bs
, co
);
4687 BDRV_POLL_WHILE(bs
, cco
.ret
== -EINPROGRESS
);
4696 * -EINVAL - backing format specified, but no file
4697 * -ENOSPC - can't update the backing file because no space is left in the
4699 * -ENOTSUP - format driver doesn't support changing the backing file
4701 int bdrv_change_backing_file(BlockDriverState
*bs
, const char *backing_file
,
4702 const char *backing_fmt
, bool warn
)
4704 BlockDriver
*drv
= bs
->drv
;
4711 /* Backing file format doesn't make sense without a backing file */
4712 if (backing_fmt
&& !backing_file
) {
4716 if (warn
&& backing_file
&& !backing_fmt
) {
4717 warn_report("Deprecated use of backing file without explicit "
4718 "backing format, use of this image requires "
4719 "potentially unsafe format probing");
4722 if (drv
->bdrv_change_backing_file
!= NULL
) {
4723 ret
= drv
->bdrv_change_backing_file(bs
, backing_file
, backing_fmt
);
4729 pstrcpy(bs
->backing_file
, sizeof(bs
->backing_file
), backing_file
?: "");
4730 pstrcpy(bs
->backing_format
, sizeof(bs
->backing_format
), backing_fmt
?: "");
4731 pstrcpy(bs
->auto_backing_file
, sizeof(bs
->auto_backing_file
),
4732 backing_file
?: "");
4738 * Finds the image layer in the chain that has 'bs' as its backing file.
4740 * active is the current topmost image.
4742 * Returns NULL if bs is not found in active's image chain,
4743 * or if active == bs.
4745 * Returns the bottommost base image if bs == NULL.
4747 BlockDriverState
*bdrv_find_overlay(BlockDriverState
*active
,
4748 BlockDriverState
*bs
)
4750 while (active
&& bs
!= backing_bs(active
)) {
4751 active
= backing_bs(active
);
4757 /* Given a BDS, searches for the base layer. */
4758 BlockDriverState
*bdrv_find_base(BlockDriverState
*bs
)
4760 return bdrv_find_overlay(bs
, NULL
);
4764 * Return true if at least one of the backing links between @bs and
4765 * @base is frozen. @errp is set if that's the case.
4766 * @base must be reachable from @bs, or NULL.
4768 bool bdrv_is_backing_chain_frozen(BlockDriverState
*bs
, BlockDriverState
*base
,
4771 BlockDriverState
*i
;
4773 for (i
= bs
; i
!= base
; i
= backing_bs(i
)) {
4774 if (i
->backing
&& i
->backing
->frozen
) {
4775 error_setg(errp
, "Cannot change '%s' link from '%s' to '%s'",
4776 i
->backing
->name
, i
->node_name
,
4777 backing_bs(i
)->node_name
);
4786 * Freeze all backing links between @bs and @base.
4787 * If any of the links is already frozen the operation is aborted and
4788 * none of the links are modified.
4789 * @base must be reachable from @bs, or NULL.
4790 * Returns 0 on success. On failure returns < 0 and sets @errp.
4792 int bdrv_freeze_backing_chain(BlockDriverState
*bs
, BlockDriverState
*base
,
4795 BlockDriverState
*i
;
4797 if (bdrv_is_backing_chain_frozen(bs
, base
, errp
)) {
4801 for (i
= bs
; i
!= base
; i
= backing_bs(i
)) {
4802 if (i
->backing
&& backing_bs(i
)->never_freeze
) {
4803 error_setg(errp
, "Cannot freeze '%s' link to '%s'",
4804 i
->backing
->name
, backing_bs(i
)->node_name
);
4809 for (i
= bs
; i
!= base
; i
= backing_bs(i
)) {
4811 i
->backing
->frozen
= true;
4819 * Unfreeze all backing links between @bs and @base. The caller must
4820 * ensure that all links are frozen before using this function.
4821 * @base must be reachable from @bs, or NULL.
4823 void bdrv_unfreeze_backing_chain(BlockDriverState
*bs
, BlockDriverState
*base
)
4825 BlockDriverState
*i
;
4827 for (i
= bs
; i
!= base
; i
= backing_bs(i
)) {
4829 assert(i
->backing
->frozen
);
4830 i
->backing
->frozen
= false;
4836 * Drops images above 'base' up to and including 'top', and sets the image
4837 * above 'top' to have base as its backing file.
4839 * Requires that the overlay to 'top' is opened r/w, so that the backing file
4840 * information in 'bs' can be properly updated.
4842 * E.g., this will convert the following chain:
4843 * bottom <- base <- intermediate <- top <- active
4847 * bottom <- base <- active
4849 * It is allowed for bottom==base, in which case it converts:
4851 * base <- intermediate <- top <- active
4857 * If backing_file_str is non-NULL, it will be used when modifying top's
4858 * overlay image metadata.
4861 * if active == top, that is considered an error
4864 int bdrv_drop_intermediate(BlockDriverState
*top
, BlockDriverState
*base
,
4865 const char *backing_file_str
)
4867 BlockDriverState
*explicit_top
= top
;
4868 bool update_inherits_from
;
4869 BdrvChild
*c
, *next
;
4870 Error
*local_err
= NULL
;
4874 bdrv_subtree_drained_begin(top
);
4876 if (!top
->drv
|| !base
->drv
) {
4880 /* Make sure that base is in the backing chain of top */
4881 if (!bdrv_chain_contains(top
, base
)) {
4885 /* This function changes all links that point to top and makes
4886 * them point to base. Check that none of them is frozen. */
4887 QLIST_FOREACH(c
, &top
->parents
, next_parent
) {
4893 /* If 'base' recursively inherits from 'top' then we should set
4894 * base->inherits_from to top->inherits_from after 'top' and all
4895 * other intermediate nodes have been dropped.
4896 * If 'top' is an implicit node (e.g. "commit_top") we should skip
4897 * it because no one inherits from it. We use explicit_top for that. */
4898 while (explicit_top
&& explicit_top
->implicit
) {
4899 explicit_top
= backing_bs(explicit_top
);
4901 update_inherits_from
= bdrv_inherits_from_recursive(base
, explicit_top
);
4903 /* success - we can delete the intermediate states, and link top->base */
4904 /* TODO Check graph modification op blockers (BLK_PERM_GRAPH_MOD) once
4905 * we've figured out how they should work. */
4906 if (!backing_file_str
) {
4907 bdrv_refresh_filename(base
);
4908 backing_file_str
= base
->filename
;
4911 QLIST_FOREACH_SAFE(c
, &top
->parents
, next_parent
, next
) {
4912 /* Check whether we are allowed to switch c from top to base */
4913 GSList
*ignore_children
= g_slist_prepend(NULL
, c
);
4914 ret
= bdrv_check_update_perm(base
, NULL
, c
->perm
, c
->shared_perm
,
4915 ignore_children
, NULL
, &local_err
);
4916 g_slist_free(ignore_children
);
4918 error_report_err(local_err
);
4922 /* If so, update the backing file path in the image file */
4923 if (c
->klass
->update_filename
) {
4924 ret
= c
->klass
->update_filename(c
, base
, backing_file_str
,
4927 bdrv_abort_perm_update(base
);
4928 error_report_err(local_err
);
4933 /* Do the actual switch in the in-memory graph.
4934 * Completes bdrv_check_update_perm() transaction internally. */
4936 bdrv_replace_child(c
, base
);
4940 if (update_inherits_from
) {
4941 base
->inherits_from
= explicit_top
->inherits_from
;
4946 bdrv_subtree_drained_end(top
);
4952 * Length of a allocated file in bytes. Sparse files are counted by actual
4953 * allocated space. Return < 0 if error or unknown.
4955 int64_t bdrv_get_allocated_file_size(BlockDriverState
*bs
)
4957 BlockDriver
*drv
= bs
->drv
;
4961 if (drv
->bdrv_get_allocated_file_size
) {
4962 return drv
->bdrv_get_allocated_file_size(bs
);
4965 return bdrv_get_allocated_file_size(bs
->file
->bs
);
4972 * @drv: Format driver
4973 * @opts: Creation options for new image
4974 * @in_bs: Existing image containing data for new image (may be NULL)
4975 * @errp: Error object
4976 * Returns: A #BlockMeasureInfo (free using qapi_free_BlockMeasureInfo())
4979 * Calculate file size required to create a new image.
4981 * If @in_bs is given then space for allocated clusters and zero clusters
4982 * from that image are included in the calculation. If @opts contains a
4983 * backing file that is shared by @in_bs then backing clusters may be omitted
4984 * from the calculation.
4986 * If @in_bs is NULL then the calculation includes no allocated clusters
4987 * unless a preallocation option is given in @opts.
4989 * Note that @in_bs may use a different BlockDriver from @drv.
4991 * If an error occurs the @errp pointer is set.
4993 BlockMeasureInfo
*bdrv_measure(BlockDriver
*drv
, QemuOpts
*opts
,
4994 BlockDriverState
*in_bs
, Error
**errp
)
4996 if (!drv
->bdrv_measure
) {
4997 error_setg(errp
, "Block driver '%s' does not support size measurement",
5002 return drv
->bdrv_measure(opts
, in_bs
, errp
);
5006 * Return number of sectors on success, -errno on error.
5008 int64_t bdrv_nb_sectors(BlockDriverState
*bs
)
5010 BlockDriver
*drv
= bs
->drv
;
5015 if (drv
->has_variable_length
) {
5016 int ret
= refresh_total_sectors(bs
, bs
->total_sectors
);
5021 return bs
->total_sectors
;
5025 * Return length in bytes on success, -errno on error.
5026 * The length is always a multiple of BDRV_SECTOR_SIZE.
5028 int64_t bdrv_getlength(BlockDriverState
*bs
)
5030 int64_t ret
= bdrv_nb_sectors(bs
);
5032 ret
= ret
> INT64_MAX
/ BDRV_SECTOR_SIZE
? -EFBIG
: ret
;
5033 return ret
< 0 ? ret
: ret
* BDRV_SECTOR_SIZE
;
5036 /* return 0 as number of sectors if no device present or error */
5037 void bdrv_get_geometry(BlockDriverState
*bs
, uint64_t *nb_sectors_ptr
)
5039 int64_t nb_sectors
= bdrv_nb_sectors(bs
);
5041 *nb_sectors_ptr
= nb_sectors
< 0 ? 0 : nb_sectors
;
5044 bool bdrv_is_sg(BlockDriverState
*bs
)
5049 bool bdrv_is_encrypted(BlockDriverState
*bs
)
5051 if (bs
->backing
&& bs
->backing
->bs
->encrypted
) {
5054 return bs
->encrypted
;
5057 const char *bdrv_get_format_name(BlockDriverState
*bs
)
5059 return bs
->drv
? bs
->drv
->format_name
: NULL
;
5062 static int qsort_strcmp(const void *a
, const void *b
)
5064 return strcmp(*(char *const *)a
, *(char *const *)b
);
5067 void bdrv_iterate_format(void (*it
)(void *opaque
, const char *name
),
5068 void *opaque
, bool read_only
)
5073 const char **formats
= NULL
;
5075 QLIST_FOREACH(drv
, &bdrv_drivers
, list
) {
5076 if (drv
->format_name
) {
5080 if (use_bdrv_whitelist
&& !bdrv_is_whitelisted(drv
, read_only
)) {
5084 while (formats
&& i
&& !found
) {
5085 found
= !strcmp(formats
[--i
], drv
->format_name
);
5089 formats
= g_renew(const char *, formats
, count
+ 1);
5090 formats
[count
++] = drv
->format_name
;
5095 for (i
= 0; i
< (int)ARRAY_SIZE(block_driver_modules
); i
++) {
5096 const char *format_name
= block_driver_modules
[i
].format_name
;
5102 if (use_bdrv_whitelist
&&
5103 !bdrv_format_is_whitelisted(format_name
, read_only
)) {
5107 while (formats
&& j
&& !found
) {
5108 found
= !strcmp(formats
[--j
], format_name
);
5112 formats
= g_renew(const char *, formats
, count
+ 1);
5113 formats
[count
++] = format_name
;
5118 qsort(formats
, count
, sizeof(formats
[0]), qsort_strcmp
);
5120 for (i
= 0; i
< count
; i
++) {
5121 it(opaque
, formats
[i
]);
5127 /* This function is to find a node in the bs graph */
5128 BlockDriverState
*bdrv_find_node(const char *node_name
)
5130 BlockDriverState
*bs
;
5134 QTAILQ_FOREACH(bs
, &graph_bdrv_states
, node_list
) {
5135 if (!strcmp(node_name
, bs
->node_name
)) {
5142 /* Put this QMP function here so it can access the static graph_bdrv_states. */
5143 BlockDeviceInfoList
*bdrv_named_nodes_list(bool flat
,
5146 BlockDeviceInfoList
*list
, *entry
;
5147 BlockDriverState
*bs
;
5150 QTAILQ_FOREACH(bs
, &graph_bdrv_states
, node_list
) {
5151 BlockDeviceInfo
*info
= bdrv_block_device_info(NULL
, bs
, flat
, errp
);
5153 qapi_free_BlockDeviceInfoList(list
);
5156 entry
= g_malloc0(sizeof(*entry
));
5157 entry
->value
= info
;
5165 #define QAPI_LIST_ADD(list, element) do { \
5166 typeof(list) _tmp = g_new(typeof(*(list)), 1); \
5167 _tmp->value = (element); \
5168 _tmp->next = (list); \
5172 typedef struct XDbgBlockGraphConstructor
{
5173 XDbgBlockGraph
*graph
;
5174 GHashTable
*graph_nodes
;
5175 } XDbgBlockGraphConstructor
;
5177 static XDbgBlockGraphConstructor
*xdbg_graph_new(void)
5179 XDbgBlockGraphConstructor
*gr
= g_new(XDbgBlockGraphConstructor
, 1);
5181 gr
->graph
= g_new0(XDbgBlockGraph
, 1);
5182 gr
->graph_nodes
= g_hash_table_new(NULL
, NULL
);
5187 static XDbgBlockGraph
*xdbg_graph_finalize(XDbgBlockGraphConstructor
*gr
)
5189 XDbgBlockGraph
*graph
= gr
->graph
;
5191 g_hash_table_destroy(gr
->graph_nodes
);
5197 static uintptr_t xdbg_graph_node_num(XDbgBlockGraphConstructor
*gr
, void *node
)
5199 uintptr_t ret
= (uintptr_t)g_hash_table_lookup(gr
->graph_nodes
, node
);
5206 * Start counting from 1, not 0, because 0 interferes with not-found (NULL)
5207 * answer of g_hash_table_lookup.
5209 ret
= g_hash_table_size(gr
->graph_nodes
) + 1;
5210 g_hash_table_insert(gr
->graph_nodes
, node
, (void *)ret
);
5215 static void xdbg_graph_add_node(XDbgBlockGraphConstructor
*gr
, void *node
,
5216 XDbgBlockGraphNodeType type
, const char *name
)
5218 XDbgBlockGraphNode
*n
;
5220 n
= g_new0(XDbgBlockGraphNode
, 1);
5222 n
->id
= xdbg_graph_node_num(gr
, node
);
5224 n
->name
= g_strdup(name
);
5226 QAPI_LIST_ADD(gr
->graph
->nodes
, n
);
5229 static void xdbg_graph_add_edge(XDbgBlockGraphConstructor
*gr
, void *parent
,
5230 const BdrvChild
*child
)
5232 BlockPermission qapi_perm
;
5233 XDbgBlockGraphEdge
*edge
;
5235 edge
= g_new0(XDbgBlockGraphEdge
, 1);
5237 edge
->parent
= xdbg_graph_node_num(gr
, parent
);
5238 edge
->child
= xdbg_graph_node_num(gr
, child
->bs
);
5239 edge
->name
= g_strdup(child
->name
);
5241 for (qapi_perm
= 0; qapi_perm
< BLOCK_PERMISSION__MAX
; qapi_perm
++) {
5242 uint64_t flag
= bdrv_qapi_perm_to_blk_perm(qapi_perm
);
5244 if (flag
& child
->perm
) {
5245 QAPI_LIST_ADD(edge
->perm
, qapi_perm
);
5247 if (flag
& child
->shared_perm
) {
5248 QAPI_LIST_ADD(edge
->shared_perm
, qapi_perm
);
5252 QAPI_LIST_ADD(gr
->graph
->edges
, edge
);
5256 XDbgBlockGraph
*bdrv_get_xdbg_block_graph(Error
**errp
)
5260 BlockDriverState
*bs
;
5262 XDbgBlockGraphConstructor
*gr
= xdbg_graph_new();
5264 for (blk
= blk_all_next(NULL
); blk
; blk
= blk_all_next(blk
)) {
5265 char *allocated_name
= NULL
;
5266 const char *name
= blk_name(blk
);
5269 name
= allocated_name
= blk_get_attached_dev_id(blk
);
5271 xdbg_graph_add_node(gr
, blk
, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_BACKEND
,
5273 g_free(allocated_name
);
5274 if (blk_root(blk
)) {
5275 xdbg_graph_add_edge(gr
, blk
, blk_root(blk
));
5279 for (job
= block_job_next(NULL
); job
; job
= block_job_next(job
)) {
5282 xdbg_graph_add_node(gr
, job
, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_JOB
,
5284 for (el
= job
->nodes
; el
; el
= el
->next
) {
5285 xdbg_graph_add_edge(gr
, job
, (BdrvChild
*)el
->data
);
5289 QTAILQ_FOREACH(bs
, &graph_bdrv_states
, node_list
) {
5290 xdbg_graph_add_node(gr
, bs
, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_DRIVER
,
5292 QLIST_FOREACH(child
, &bs
->children
, next
) {
5293 xdbg_graph_add_edge(gr
, bs
, child
);
5297 return xdbg_graph_finalize(gr
);
5300 BlockDriverState
*bdrv_lookup_bs(const char *device
,
5301 const char *node_name
,
5305 BlockDriverState
*bs
;
5308 blk
= blk_by_name(device
);
5313 error_setg(errp
, "Device '%s' has no medium", device
);
5321 bs
= bdrv_find_node(node_name
);
5328 error_setg(errp
, "Cannot find device=%s nor node_name=%s",
5329 device
? device
: "",
5330 node_name
? node_name
: "");
5334 /* If 'base' is in the same chain as 'top', return true. Otherwise,
5335 * return false. If either argument is NULL, return false. */
5336 bool bdrv_chain_contains(BlockDriverState
*top
, BlockDriverState
*base
)
5338 while (top
&& top
!= base
) {
5339 top
= backing_bs(top
);
5345 BlockDriverState
*bdrv_next_node(BlockDriverState
*bs
)
5348 return QTAILQ_FIRST(&graph_bdrv_states
);
5350 return QTAILQ_NEXT(bs
, node_list
);
5353 BlockDriverState
*bdrv_next_all_states(BlockDriverState
*bs
)
5356 return QTAILQ_FIRST(&all_bdrv_states
);
5358 return QTAILQ_NEXT(bs
, bs_list
);
5361 const char *bdrv_get_node_name(const BlockDriverState
*bs
)
5363 return bs
->node_name
;
5366 const char *bdrv_get_parent_name(const BlockDriverState
*bs
)
5371 /* If multiple parents have a name, just pick the first one. */
5372 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
5373 if (c
->klass
->get_name
) {
5374 name
= c
->klass
->get_name(c
);
5375 if (name
&& *name
) {
5384 /* TODO check what callers really want: bs->node_name or blk_name() */
5385 const char *bdrv_get_device_name(const BlockDriverState
*bs
)
5387 return bdrv_get_parent_name(bs
) ?: "";
5390 /* This can be used to identify nodes that might not have a device
5391 * name associated. Since node and device names live in the same
5392 * namespace, the result is unambiguous. The exception is if both are
5393 * absent, then this returns an empty (non-null) string. */
5394 const char *bdrv_get_device_or_node_name(const BlockDriverState
*bs
)
5396 return bdrv_get_parent_name(bs
) ?: bs
->node_name
;
5399 int bdrv_get_flags(BlockDriverState
*bs
)
5401 return bs
->open_flags
;
5404 int bdrv_has_zero_init_1(BlockDriverState
*bs
)
5409 int bdrv_has_zero_init(BlockDriverState
*bs
)
5415 /* If BS is a copy on write image, it is initialized to
5416 the contents of the base image, which may not be zeroes. */
5420 if (bs
->drv
->bdrv_has_zero_init
) {
5421 return bs
->drv
->bdrv_has_zero_init(bs
);
5423 if (bs
->file
&& bs
->drv
->is_filter
) {
5424 return bdrv_has_zero_init(bs
->file
->bs
);
5431 bool bdrv_can_write_zeroes_with_unmap(BlockDriverState
*bs
)
5433 if (!(bs
->open_flags
& BDRV_O_UNMAP
)) {
5437 return bs
->supported_zero_flags
& BDRV_REQ_MAY_UNMAP
;
5440 void bdrv_get_backing_filename(BlockDriverState
*bs
,
5441 char *filename
, int filename_size
)
5443 pstrcpy(filename
, filename_size
, bs
->backing_file
);
5446 int bdrv_get_info(BlockDriverState
*bs
, BlockDriverInfo
*bdi
)
5448 BlockDriver
*drv
= bs
->drv
;
5449 /* if bs->drv == NULL, bs is closed, so there's nothing to do here */
5453 if (!drv
->bdrv_get_info
) {
5454 if (bs
->file
&& drv
->is_filter
) {
5455 return bdrv_get_info(bs
->file
->bs
, bdi
);
5459 memset(bdi
, 0, sizeof(*bdi
));
5460 return drv
->bdrv_get_info(bs
, bdi
);
5463 ImageInfoSpecific
*bdrv_get_specific_info(BlockDriverState
*bs
,
5466 BlockDriver
*drv
= bs
->drv
;
5467 if (drv
&& drv
->bdrv_get_specific_info
) {
5468 return drv
->bdrv_get_specific_info(bs
, errp
);
5473 BlockStatsSpecific
*bdrv_get_specific_stats(BlockDriverState
*bs
)
5475 BlockDriver
*drv
= bs
->drv
;
5476 if (!drv
|| !drv
->bdrv_get_specific_stats
) {
5479 return drv
->bdrv_get_specific_stats(bs
);
5482 void bdrv_debug_event(BlockDriverState
*bs
, BlkdebugEvent event
)
5484 if (!bs
|| !bs
->drv
|| !bs
->drv
->bdrv_debug_event
) {
5488 bs
->drv
->bdrv_debug_event(bs
, event
);
5491 static BlockDriverState
*bdrv_find_debug_node(BlockDriverState
*bs
)
5493 while (bs
&& bs
->drv
&& !bs
->drv
->bdrv_debug_breakpoint
) {
5499 if (bs
->drv
->is_filter
&& bs
->backing
) {
5500 bs
= bs
->backing
->bs
;
5507 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_breakpoint
) {
5508 assert(bs
->drv
->bdrv_debug_remove_breakpoint
);
5515 int bdrv_debug_breakpoint(BlockDriverState
*bs
, const char *event
,
5518 bs
= bdrv_find_debug_node(bs
);
5520 return bs
->drv
->bdrv_debug_breakpoint(bs
, event
, tag
);
5526 int bdrv_debug_remove_breakpoint(BlockDriverState
*bs
, const char *tag
)
5528 bs
= bdrv_find_debug_node(bs
);
5530 return bs
->drv
->bdrv_debug_remove_breakpoint(bs
, tag
);
5536 int bdrv_debug_resume(BlockDriverState
*bs
, const char *tag
)
5538 while (bs
&& (!bs
->drv
|| !bs
->drv
->bdrv_debug_resume
)) {
5539 bs
= bs
->file
? bs
->file
->bs
: NULL
;
5542 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_resume
) {
5543 return bs
->drv
->bdrv_debug_resume(bs
, tag
);
5549 bool bdrv_debug_is_suspended(BlockDriverState
*bs
, const char *tag
)
5551 while (bs
&& bs
->drv
&& !bs
->drv
->bdrv_debug_is_suspended
) {
5552 bs
= bs
->file
? bs
->file
->bs
: NULL
;
5555 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_is_suspended
) {
5556 return bs
->drv
->bdrv_debug_is_suspended(bs
, tag
);
5562 /* backing_file can either be relative, or absolute, or a protocol. If it is
5563 * relative, it must be relative to the chain. So, passing in bs->filename
5564 * from a BDS as backing_file should not be done, as that may be relative to
5565 * the CWD rather than the chain. */
5566 BlockDriverState
*bdrv_find_backing_image(BlockDriverState
*bs
,
5567 const char *backing_file
)
5569 char *filename_full
= NULL
;
5570 char *backing_file_full
= NULL
;
5571 char *filename_tmp
= NULL
;
5572 int is_protocol
= 0;
5573 BlockDriverState
*curr_bs
= NULL
;
5574 BlockDriverState
*retval
= NULL
;
5576 if (!bs
|| !bs
->drv
|| !backing_file
) {
5580 filename_full
= g_malloc(PATH_MAX
);
5581 backing_file_full
= g_malloc(PATH_MAX
);
5583 is_protocol
= path_has_protocol(backing_file
);
5585 for (curr_bs
= bs
; curr_bs
->backing
; curr_bs
= curr_bs
->backing
->bs
) {
5587 /* If either of the filename paths is actually a protocol, then
5588 * compare unmodified paths; otherwise make paths relative */
5589 if (is_protocol
|| path_has_protocol(curr_bs
->backing_file
)) {
5590 char *backing_file_full_ret
;
5592 if (strcmp(backing_file
, curr_bs
->backing_file
) == 0) {
5593 retval
= curr_bs
->backing
->bs
;
5596 /* Also check against the full backing filename for the image */
5597 backing_file_full_ret
= bdrv_get_full_backing_filename(curr_bs
,
5599 if (backing_file_full_ret
) {
5600 bool equal
= strcmp(backing_file
, backing_file_full_ret
) == 0;
5601 g_free(backing_file_full_ret
);
5603 retval
= curr_bs
->backing
->bs
;
5608 /* If not an absolute filename path, make it relative to the current
5609 * image's filename path */
5610 filename_tmp
= bdrv_make_absolute_filename(curr_bs
, backing_file
,
5612 /* We are going to compare canonicalized absolute pathnames */
5613 if (!filename_tmp
|| !realpath(filename_tmp
, filename_full
)) {
5614 g_free(filename_tmp
);
5617 g_free(filename_tmp
);
5619 /* We need to make sure the backing filename we are comparing against
5620 * is relative to the current image filename (or absolute) */
5621 filename_tmp
= bdrv_get_full_backing_filename(curr_bs
, NULL
);
5622 if (!filename_tmp
|| !realpath(filename_tmp
, backing_file_full
)) {
5623 g_free(filename_tmp
);
5626 g_free(filename_tmp
);
5628 if (strcmp(backing_file_full
, filename_full
) == 0) {
5629 retval
= curr_bs
->backing
->bs
;
5635 g_free(filename_full
);
5636 g_free(backing_file_full
);
5640 void bdrv_init(void)
5642 module_call_init(MODULE_INIT_BLOCK
);
5645 void bdrv_init_with_whitelist(void)
5647 use_bdrv_whitelist
= 1;
5651 static void coroutine_fn
bdrv_co_invalidate_cache(BlockDriverState
*bs
,
5654 BdrvChild
*child
, *parent
;
5655 uint64_t perm
, shared_perm
;
5656 Error
*local_err
= NULL
;
5658 BdrvDirtyBitmap
*bm
;
5664 QLIST_FOREACH(child
, &bs
->children
, next
) {
5665 bdrv_co_invalidate_cache(child
->bs
, &local_err
);
5667 error_propagate(errp
, local_err
);
5673 * Update permissions, they may differ for inactive nodes.
5675 * Note that the required permissions of inactive images are always a
5676 * subset of the permissions required after activating the image. This
5677 * allows us to just get the permissions upfront without restricting
5678 * drv->bdrv_invalidate_cache().
5680 * It also means that in error cases, we don't have to try and revert to
5681 * the old permissions (which is an operation that could fail, too). We can
5682 * just keep the extended permissions for the next time that an activation
5683 * of the image is tried.
5685 if (bs
->open_flags
& BDRV_O_INACTIVE
) {
5686 bs
->open_flags
&= ~BDRV_O_INACTIVE
;
5687 bdrv_get_cumulative_perm(bs
, &perm
, &shared_perm
);
5688 ret
= bdrv_check_perm(bs
, NULL
, perm
, shared_perm
, NULL
, NULL
, errp
);
5690 bs
->open_flags
|= BDRV_O_INACTIVE
;
5693 bdrv_set_perm(bs
, perm
, shared_perm
);
5695 if (bs
->drv
->bdrv_co_invalidate_cache
) {
5696 bs
->drv
->bdrv_co_invalidate_cache(bs
, &local_err
);
5698 bs
->open_flags
|= BDRV_O_INACTIVE
;
5699 error_propagate(errp
, local_err
);
5704 FOR_EACH_DIRTY_BITMAP(bs
, bm
) {
5705 bdrv_dirty_bitmap_skip_store(bm
, false);
5708 ret
= refresh_total_sectors(bs
, bs
->total_sectors
);
5710 bs
->open_flags
|= BDRV_O_INACTIVE
;
5711 error_setg_errno(errp
, -ret
, "Could not refresh total sector count");
5716 QLIST_FOREACH(parent
, &bs
->parents
, next_parent
) {
5717 if (parent
->klass
->activate
) {
5718 parent
->klass
->activate(parent
, &local_err
);
5720 bs
->open_flags
|= BDRV_O_INACTIVE
;
5721 error_propagate(errp
, local_err
);
5728 typedef struct InvalidateCacheCo
{
5729 BlockDriverState
*bs
;
5732 } InvalidateCacheCo
;
5734 static void coroutine_fn
bdrv_invalidate_cache_co_entry(void *opaque
)
5736 InvalidateCacheCo
*ico
= opaque
;
5737 bdrv_co_invalidate_cache(ico
->bs
, ico
->errp
);
5742 void bdrv_invalidate_cache(BlockDriverState
*bs
, Error
**errp
)
5745 InvalidateCacheCo ico
= {
5751 if (qemu_in_coroutine()) {
5752 /* Fast-path if already in coroutine context */
5753 bdrv_invalidate_cache_co_entry(&ico
);
5755 co
= qemu_coroutine_create(bdrv_invalidate_cache_co_entry
, &ico
);
5756 bdrv_coroutine_enter(bs
, co
);
5757 BDRV_POLL_WHILE(bs
, !ico
.done
);
5761 void bdrv_invalidate_cache_all(Error
**errp
)
5763 BlockDriverState
*bs
;
5764 Error
*local_err
= NULL
;
5765 BdrvNextIterator it
;
5767 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
5768 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
5770 aio_context_acquire(aio_context
);
5771 bdrv_invalidate_cache(bs
, &local_err
);
5772 aio_context_release(aio_context
);
5774 error_propagate(errp
, local_err
);
5775 bdrv_next_cleanup(&it
);
5781 static bool bdrv_has_bds_parent(BlockDriverState
*bs
, bool only_active
)
5785 QLIST_FOREACH(parent
, &bs
->parents
, next_parent
) {
5786 if (parent
->klass
->parent_is_bds
) {
5787 BlockDriverState
*parent_bs
= parent
->opaque
;
5788 if (!only_active
|| !(parent_bs
->open_flags
& BDRV_O_INACTIVE
)) {
5797 static int bdrv_inactivate_recurse(BlockDriverState
*bs
)
5799 BdrvChild
*child
, *parent
;
5800 bool tighten_restrictions
;
5801 uint64_t perm
, shared_perm
;
5808 /* Make sure that we don't inactivate a child before its parent.
5809 * It will be covered by recursion from the yet active parent. */
5810 if (bdrv_has_bds_parent(bs
, true)) {
5814 assert(!(bs
->open_flags
& BDRV_O_INACTIVE
));
5816 /* Inactivate this node */
5817 if (bs
->drv
->bdrv_inactivate
) {
5818 ret
= bs
->drv
->bdrv_inactivate(bs
);
5824 QLIST_FOREACH(parent
, &bs
->parents
, next_parent
) {
5825 if (parent
->klass
->inactivate
) {
5826 ret
= parent
->klass
->inactivate(parent
);
5833 bs
->open_flags
|= BDRV_O_INACTIVE
;
5835 /* Update permissions, they may differ for inactive nodes */
5836 bdrv_get_cumulative_perm(bs
, &perm
, &shared_perm
);
5837 ret
= bdrv_check_perm(bs
, NULL
, perm
, shared_perm
, NULL
,
5838 &tighten_restrictions
, NULL
);
5839 assert(tighten_restrictions
== false);
5841 /* We only tried to loosen restrictions, so errors are not fatal */
5842 bdrv_abort_perm_update(bs
);
5844 bdrv_set_perm(bs
, perm
, shared_perm
);
5848 /* Recursively inactivate children */
5849 QLIST_FOREACH(child
, &bs
->children
, next
) {
5850 ret
= bdrv_inactivate_recurse(child
->bs
);
5859 int bdrv_inactivate_all(void)
5861 BlockDriverState
*bs
= NULL
;
5862 BdrvNextIterator it
;
5864 GSList
*aio_ctxs
= NULL
, *ctx
;
5866 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
5867 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
5869 if (!g_slist_find(aio_ctxs
, aio_context
)) {
5870 aio_ctxs
= g_slist_prepend(aio_ctxs
, aio_context
);
5871 aio_context_acquire(aio_context
);
5875 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
5876 /* Nodes with BDS parents are covered by recursion from the last
5877 * parent that gets inactivated. Don't inactivate them a second
5878 * time if that has already happened. */
5879 if (bdrv_has_bds_parent(bs
, false)) {
5882 ret
= bdrv_inactivate_recurse(bs
);
5884 bdrv_next_cleanup(&it
);
5890 for (ctx
= aio_ctxs
; ctx
!= NULL
; ctx
= ctx
->next
) {
5891 AioContext
*aio_context
= ctx
->data
;
5892 aio_context_release(aio_context
);
5894 g_slist_free(aio_ctxs
);
5899 /**************************************************************/
5900 /* removable device support */
5903 * Return TRUE if the media is present
5905 bool bdrv_is_inserted(BlockDriverState
*bs
)
5907 BlockDriver
*drv
= bs
->drv
;
5913 if (drv
->bdrv_is_inserted
) {
5914 return drv
->bdrv_is_inserted(bs
);
5916 QLIST_FOREACH(child
, &bs
->children
, next
) {
5917 if (!bdrv_is_inserted(child
->bs
)) {
5925 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
5927 void bdrv_eject(BlockDriverState
*bs
, bool eject_flag
)
5929 BlockDriver
*drv
= bs
->drv
;
5931 if (drv
&& drv
->bdrv_eject
) {
5932 drv
->bdrv_eject(bs
, eject_flag
);
5937 * Lock or unlock the media (if it is locked, the user won't be able
5938 * to eject it manually).
5940 void bdrv_lock_medium(BlockDriverState
*bs
, bool locked
)
5942 BlockDriver
*drv
= bs
->drv
;
5944 trace_bdrv_lock_medium(bs
, locked
);
5946 if (drv
&& drv
->bdrv_lock_medium
) {
5947 drv
->bdrv_lock_medium(bs
, locked
);
5951 /* Get a reference to bs */
5952 void bdrv_ref(BlockDriverState
*bs
)
5957 /* Release a previously grabbed reference to bs.
5958 * If after releasing, reference count is zero, the BlockDriverState is
5960 void bdrv_unref(BlockDriverState
*bs
)
5965 assert(bs
->refcnt
> 0);
5966 if (--bs
->refcnt
== 0) {
5971 struct BdrvOpBlocker
{
5973 QLIST_ENTRY(BdrvOpBlocker
) list
;
5976 bool bdrv_op_is_blocked(BlockDriverState
*bs
, BlockOpType op
, Error
**errp
)
5978 BdrvOpBlocker
*blocker
;
5979 assert((int) op
>= 0 && op
< BLOCK_OP_TYPE_MAX
);
5980 if (!QLIST_EMPTY(&bs
->op_blockers
[op
])) {
5981 blocker
= QLIST_FIRST(&bs
->op_blockers
[op
]);
5982 error_propagate_prepend(errp
, error_copy(blocker
->reason
),
5983 "Node '%s' is busy: ",
5984 bdrv_get_device_or_node_name(bs
));
5990 void bdrv_op_block(BlockDriverState
*bs
, BlockOpType op
, Error
*reason
)
5992 BdrvOpBlocker
*blocker
;
5993 assert((int) op
>= 0 && op
< BLOCK_OP_TYPE_MAX
);
5995 blocker
= g_new0(BdrvOpBlocker
, 1);
5996 blocker
->reason
= reason
;
5997 QLIST_INSERT_HEAD(&bs
->op_blockers
[op
], blocker
, list
);
6000 void bdrv_op_unblock(BlockDriverState
*bs
, BlockOpType op
, Error
*reason
)
6002 BdrvOpBlocker
*blocker
, *next
;
6003 assert((int) op
>= 0 && op
< BLOCK_OP_TYPE_MAX
);
6004 QLIST_FOREACH_SAFE(blocker
, &bs
->op_blockers
[op
], list
, next
) {
6005 if (blocker
->reason
== reason
) {
6006 QLIST_REMOVE(blocker
, list
);
6012 void bdrv_op_block_all(BlockDriverState
*bs
, Error
*reason
)
6015 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
6016 bdrv_op_block(bs
, i
, reason
);
6020 void bdrv_op_unblock_all(BlockDriverState
*bs
, Error
*reason
)
6023 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
6024 bdrv_op_unblock(bs
, i
, reason
);
6028 bool bdrv_op_blocker_is_empty(BlockDriverState
*bs
)
6032 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
6033 if (!QLIST_EMPTY(&bs
->op_blockers
[i
])) {
6040 void bdrv_img_create(const char *filename
, const char *fmt
,
6041 const char *base_filename
, const char *base_fmt
,
6042 char *options
, uint64_t img_size
, int flags
, bool quiet
,
6045 QemuOptsList
*create_opts
= NULL
;
6046 QemuOpts
*opts
= NULL
;
6047 const char *backing_fmt
, *backing_file
;
6049 BlockDriver
*drv
, *proto_drv
;
6050 Error
*local_err
= NULL
;
6053 /* Find driver and parse its options */
6054 drv
= bdrv_find_format(fmt
);
6056 error_setg(errp
, "Unknown file format '%s'", fmt
);
6060 proto_drv
= bdrv_find_protocol(filename
, true, errp
);
6065 if (!drv
->create_opts
) {
6066 error_setg(errp
, "Format driver '%s' does not support image creation",
6071 if (!proto_drv
->create_opts
) {
6072 error_setg(errp
, "Protocol driver '%s' does not support image creation",
6073 proto_drv
->format_name
);
6077 /* Create parameter list */
6078 create_opts
= qemu_opts_append(create_opts
, drv
->create_opts
);
6079 create_opts
= qemu_opts_append(create_opts
, proto_drv
->create_opts
);
6081 opts
= qemu_opts_create(create_opts
, NULL
, 0, &error_abort
);
6083 /* Parse -o options */
6085 if (!qemu_opts_do_parse(opts
, options
, NULL
, errp
)) {
6090 if (!qemu_opt_get(opts
, BLOCK_OPT_SIZE
)) {
6091 qemu_opt_set_number(opts
, BLOCK_OPT_SIZE
, img_size
, &error_abort
);
6092 } else if (img_size
!= UINT64_C(-1)) {
6093 error_setg(errp
, "The image size must be specified only once");
6097 if (base_filename
) {
6098 if (!qemu_opt_set(opts
, BLOCK_OPT_BACKING_FILE
, base_filename
,
6100 error_setg(errp
, "Backing file not supported for file format '%s'",
6107 if (!qemu_opt_set(opts
, BLOCK_OPT_BACKING_FMT
, base_fmt
, NULL
)) {
6108 error_setg(errp
, "Backing file format not supported for file "
6109 "format '%s'", fmt
);
6114 backing_file
= qemu_opt_get(opts
, BLOCK_OPT_BACKING_FILE
);
6116 if (!strcmp(filename
, backing_file
)) {
6117 error_setg(errp
, "Error: Trying to create an image with the "
6118 "same filename as the backing file");
6123 backing_fmt
= qemu_opt_get(opts
, BLOCK_OPT_BACKING_FMT
);
6125 /* The size for the image must always be specified, unless we have a backing
6126 * file and we have not been forbidden from opening it. */
6127 size
= qemu_opt_get_size(opts
, BLOCK_OPT_SIZE
, img_size
);
6128 if (backing_file
&& !(flags
& BDRV_O_NO_BACKING
)) {
6129 BlockDriverState
*bs
;
6132 QDict
*backing_options
= NULL
;
6135 bdrv_get_full_backing_filename_from_filename(filename
, backing_file
,
6140 assert(full_backing
);
6142 /* backing files always opened read-only */
6144 back_flags
&= ~(BDRV_O_RDWR
| BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
);
6146 backing_options
= qdict_new();
6148 qdict_put_str(backing_options
, "driver", backing_fmt
);
6150 qdict_put_bool(backing_options
, BDRV_OPT_FORCE_SHARE
, true);
6152 bs
= bdrv_open(full_backing
, NULL
, backing_options
, back_flags
,
6154 g_free(full_backing
);
6156 error_append_hint(&local_err
, "Could not open backing image.\n");
6160 warn_report("Deprecated use of backing file without explicit "
6161 "backing format (detected format of %s)",
6162 bs
->drv
->format_name
);
6163 if (bs
->drv
!= &bdrv_raw
) {
6165 * A probe of raw deserves the most attention:
6166 * leaving the backing format out of the image
6167 * will ensure bs->probed is set (ensuring we
6168 * don't accidentally commit into the backing
6169 * file), and allow more spots to warn the users
6170 * to fix their toolchain when opening this image
6171 * later. For other images, we can safely record
6172 * the format that we probed.
6174 backing_fmt
= bs
->drv
->format_name
;
6175 qemu_opt_set(opts
, BLOCK_OPT_BACKING_FMT
, backing_fmt
,
6180 /* Opened BS, have no size */
6181 size
= bdrv_getlength(bs
);
6183 error_setg_errno(errp
, -size
, "Could not get size of '%s'",
6188 qemu_opt_set_number(opts
, BLOCK_OPT_SIZE
, size
, &error_abort
);
6192 /* (backing_file && !(flags & BDRV_O_NO_BACKING)) */
6193 } else if (backing_file
&& !backing_fmt
) {
6194 warn_report("Deprecated use of unopened backing file without "
6195 "explicit backing format, use of this image requires "
6196 "potentially unsafe format probing");
6200 error_setg(errp
, "Image creation needs a size parameter");
6205 printf("Formatting '%s', fmt=%s ", filename
, fmt
);
6206 qemu_opts_print(opts
, " ");
6211 ret
= bdrv_create(drv
, filename
, opts
, &local_err
);
6213 if (ret
== -EFBIG
) {
6214 /* This is generally a better message than whatever the driver would
6215 * deliver (especially because of the cluster_size_hint), since that
6216 * is most probably not much different from "image too large". */
6217 const char *cluster_size_hint
= "";
6218 if (qemu_opt_get_size(opts
, BLOCK_OPT_CLUSTER_SIZE
, 0)) {
6219 cluster_size_hint
= " (try using a larger cluster size)";
6221 error_setg(errp
, "The image size is too large for file format '%s'"
6222 "%s", fmt
, cluster_size_hint
);
6223 error_free(local_err
);
6228 qemu_opts_del(opts
);
6229 qemu_opts_free(create_opts
);
6230 error_propagate(errp
, local_err
);
6233 AioContext
*bdrv_get_aio_context(BlockDriverState
*bs
)
6235 return bs
? bs
->aio_context
: qemu_get_aio_context();
6238 void bdrv_coroutine_enter(BlockDriverState
*bs
, Coroutine
*co
)
6240 aio_co_enter(bdrv_get_aio_context(bs
), co
);
6243 static void bdrv_do_remove_aio_context_notifier(BdrvAioNotifier
*ban
)
6245 QLIST_REMOVE(ban
, list
);
6249 static void bdrv_detach_aio_context(BlockDriverState
*bs
)
6251 BdrvAioNotifier
*baf
, *baf_tmp
;
6253 assert(!bs
->walking_aio_notifiers
);
6254 bs
->walking_aio_notifiers
= true;
6255 QLIST_FOREACH_SAFE(baf
, &bs
->aio_notifiers
, list
, baf_tmp
) {
6257 bdrv_do_remove_aio_context_notifier(baf
);
6259 baf
->detach_aio_context(baf
->opaque
);
6262 /* Never mind iterating again to check for ->deleted. bdrv_close() will
6263 * remove remaining aio notifiers if we aren't called again.
6265 bs
->walking_aio_notifiers
= false;
6267 if (bs
->drv
&& bs
->drv
->bdrv_detach_aio_context
) {
6268 bs
->drv
->bdrv_detach_aio_context(bs
);
6271 if (bs
->quiesce_counter
) {
6272 aio_enable_external(bs
->aio_context
);
6274 bs
->aio_context
= NULL
;
6277 static void bdrv_attach_aio_context(BlockDriverState
*bs
,
6278 AioContext
*new_context
)
6280 BdrvAioNotifier
*ban
, *ban_tmp
;
6282 if (bs
->quiesce_counter
) {
6283 aio_disable_external(new_context
);
6286 bs
->aio_context
= new_context
;
6288 if (bs
->drv
&& bs
->drv
->bdrv_attach_aio_context
) {
6289 bs
->drv
->bdrv_attach_aio_context(bs
, new_context
);
6292 assert(!bs
->walking_aio_notifiers
);
6293 bs
->walking_aio_notifiers
= true;
6294 QLIST_FOREACH_SAFE(ban
, &bs
->aio_notifiers
, list
, ban_tmp
) {
6296 bdrv_do_remove_aio_context_notifier(ban
);
6298 ban
->attached_aio_context(new_context
, ban
->opaque
);
6301 bs
->walking_aio_notifiers
= false;
6305 * Changes the AioContext used for fd handlers, timers, and BHs by this
6306 * BlockDriverState and all its children and parents.
6308 * Must be called from the main AioContext.
6310 * The caller must own the AioContext lock for the old AioContext of bs, but it
6311 * must not own the AioContext lock for new_context (unless new_context is the
6312 * same as the current context of bs).
6314 * @ignore will accumulate all visited BdrvChild object. The caller is
6315 * responsible for freeing the list afterwards.
6317 void bdrv_set_aio_context_ignore(BlockDriverState
*bs
,
6318 AioContext
*new_context
, GSList
**ignore
)
6320 AioContext
*old_context
= bdrv_get_aio_context(bs
);
6323 g_assert(qemu_get_current_aio_context() == qemu_get_aio_context());
6325 if (old_context
== new_context
) {
6329 bdrv_drained_begin(bs
);
6331 QLIST_FOREACH(child
, &bs
->children
, next
) {
6332 if (g_slist_find(*ignore
, child
)) {
6335 *ignore
= g_slist_prepend(*ignore
, child
);
6336 bdrv_set_aio_context_ignore(child
->bs
, new_context
, ignore
);
6338 QLIST_FOREACH(child
, &bs
->parents
, next_parent
) {
6339 if (g_slist_find(*ignore
, child
)) {
6342 assert(child
->klass
->set_aio_ctx
);
6343 *ignore
= g_slist_prepend(*ignore
, child
);
6344 child
->klass
->set_aio_ctx(child
, new_context
, ignore
);
6347 bdrv_detach_aio_context(bs
);
6349 /* Acquire the new context, if necessary */
6350 if (qemu_get_aio_context() != new_context
) {
6351 aio_context_acquire(new_context
);
6354 bdrv_attach_aio_context(bs
, new_context
);
6357 * If this function was recursively called from
6358 * bdrv_set_aio_context_ignore(), there may be nodes in the
6359 * subtree that have not yet been moved to the new AioContext.
6360 * Release the old one so bdrv_drained_end() can poll them.
6362 if (qemu_get_aio_context() != old_context
) {
6363 aio_context_release(old_context
);
6366 bdrv_drained_end(bs
);
6368 if (qemu_get_aio_context() != old_context
) {
6369 aio_context_acquire(old_context
);
6371 if (qemu_get_aio_context() != new_context
) {
6372 aio_context_release(new_context
);
6376 static bool bdrv_parent_can_set_aio_context(BdrvChild
*c
, AioContext
*ctx
,
6377 GSList
**ignore
, Error
**errp
)
6379 if (g_slist_find(*ignore
, c
)) {
6382 *ignore
= g_slist_prepend(*ignore
, c
);
6385 * A BdrvChildClass that doesn't handle AioContext changes cannot
6386 * tolerate any AioContext changes
6388 if (!c
->klass
->can_set_aio_ctx
) {
6389 char *user
= bdrv_child_user_desc(c
);
6390 error_setg(errp
, "Changing iothreads is not supported by %s", user
);
6394 if (!c
->klass
->can_set_aio_ctx(c
, ctx
, ignore
, errp
)) {
6395 assert(!errp
|| *errp
);
6401 bool bdrv_child_can_set_aio_context(BdrvChild
*c
, AioContext
*ctx
,
6402 GSList
**ignore
, Error
**errp
)
6404 if (g_slist_find(*ignore
, c
)) {
6407 *ignore
= g_slist_prepend(*ignore
, c
);
6408 return bdrv_can_set_aio_context(c
->bs
, ctx
, ignore
, errp
);
6411 /* @ignore will accumulate all visited BdrvChild object. The caller is
6412 * responsible for freeing the list afterwards. */
6413 bool bdrv_can_set_aio_context(BlockDriverState
*bs
, AioContext
*ctx
,
6414 GSList
**ignore
, Error
**errp
)
6418 if (bdrv_get_aio_context(bs
) == ctx
) {
6422 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
6423 if (!bdrv_parent_can_set_aio_context(c
, ctx
, ignore
, errp
)) {
6427 QLIST_FOREACH(c
, &bs
->children
, next
) {
6428 if (!bdrv_child_can_set_aio_context(c
, ctx
, ignore
, errp
)) {
6436 int bdrv_child_try_set_aio_context(BlockDriverState
*bs
, AioContext
*ctx
,
6437 BdrvChild
*ignore_child
, Error
**errp
)
6442 ignore
= ignore_child
? g_slist_prepend(NULL
, ignore_child
) : NULL
;
6443 ret
= bdrv_can_set_aio_context(bs
, ctx
, &ignore
, errp
);
6444 g_slist_free(ignore
);
6450 ignore
= ignore_child
? g_slist_prepend(NULL
, ignore_child
) : NULL
;
6451 bdrv_set_aio_context_ignore(bs
, ctx
, &ignore
);
6452 g_slist_free(ignore
);
6457 int bdrv_try_set_aio_context(BlockDriverState
*bs
, AioContext
*ctx
,
6460 return bdrv_child_try_set_aio_context(bs
, ctx
, NULL
, errp
);
6463 void bdrv_add_aio_context_notifier(BlockDriverState
*bs
,
6464 void (*attached_aio_context
)(AioContext
*new_context
, void *opaque
),
6465 void (*detach_aio_context
)(void *opaque
), void *opaque
)
6467 BdrvAioNotifier
*ban
= g_new(BdrvAioNotifier
, 1);
6468 *ban
= (BdrvAioNotifier
){
6469 .attached_aio_context
= attached_aio_context
,
6470 .detach_aio_context
= detach_aio_context
,
6474 QLIST_INSERT_HEAD(&bs
->aio_notifiers
, ban
, list
);
6477 void bdrv_remove_aio_context_notifier(BlockDriverState
*bs
,
6478 void (*attached_aio_context
)(AioContext
*,
6480 void (*detach_aio_context
)(void *),
6483 BdrvAioNotifier
*ban
, *ban_next
;
6485 QLIST_FOREACH_SAFE(ban
, &bs
->aio_notifiers
, list
, ban_next
) {
6486 if (ban
->attached_aio_context
== attached_aio_context
&&
6487 ban
->detach_aio_context
== detach_aio_context
&&
6488 ban
->opaque
== opaque
&&
6489 ban
->deleted
== false)
6491 if (bs
->walking_aio_notifiers
) {
6492 ban
->deleted
= true;
6494 bdrv_do_remove_aio_context_notifier(ban
);
6503 int bdrv_amend_options(BlockDriverState
*bs
, QemuOpts
*opts
,
6504 BlockDriverAmendStatusCB
*status_cb
, void *cb_opaque
,
6509 error_setg(errp
, "Node is ejected");
6512 if (!bs
->drv
->bdrv_amend_options
) {
6513 error_setg(errp
, "Block driver '%s' does not support option amendment",
6514 bs
->drv
->format_name
);
6517 return bs
->drv
->bdrv_amend_options(bs
, opts
, status_cb
,
6518 cb_opaque
, force
, errp
);
6522 * This function checks whether the given @to_replace is allowed to be
6523 * replaced by a node that always shows the same data as @bs. This is
6524 * used for example to verify whether the mirror job can replace
6525 * @to_replace by the target mirrored from @bs.
6526 * To be replaceable, @bs and @to_replace may either be guaranteed to
6527 * always show the same data (because they are only connected through
6528 * filters), or some driver may allow replacing one of its children
6529 * because it can guarantee that this child's data is not visible at
6530 * all (for example, for dissenting quorum children that have no other
6533 bool bdrv_recurse_can_replace(BlockDriverState
*bs
,
6534 BlockDriverState
*to_replace
)
6536 if (!bs
|| !bs
->drv
) {
6540 if (bs
== to_replace
) {
6544 /* See what the driver can do */
6545 if (bs
->drv
->bdrv_recurse_can_replace
) {
6546 return bs
->drv
->bdrv_recurse_can_replace(bs
, to_replace
);
6549 /* For filters without an own implementation, we can recurse on our own */
6550 if (bs
->drv
->is_filter
) {
6551 BdrvChild
*child
= bs
->file
?: bs
->backing
;
6552 return bdrv_recurse_can_replace(child
->bs
, to_replace
);
6560 * Check whether the given @node_name can be replaced by a node that
6561 * has the same data as @parent_bs. If so, return @node_name's BDS;
6564 * @node_name must be a (recursive) *child of @parent_bs (or this
6565 * function will return NULL).
6567 * The result (whether the node can be replaced or not) is only valid
6568 * for as long as no graph or permission changes occur.
6570 BlockDriverState
*check_to_replace_node(BlockDriverState
*parent_bs
,
6571 const char *node_name
, Error
**errp
)
6573 BlockDriverState
*to_replace_bs
= bdrv_find_node(node_name
);
6574 AioContext
*aio_context
;
6576 if (!to_replace_bs
) {
6577 error_setg(errp
, "Node name '%s' not found", node_name
);
6581 aio_context
= bdrv_get_aio_context(to_replace_bs
);
6582 aio_context_acquire(aio_context
);
6584 if (bdrv_op_is_blocked(to_replace_bs
, BLOCK_OP_TYPE_REPLACE
, errp
)) {
6585 to_replace_bs
= NULL
;
6589 /* We don't want arbitrary node of the BDS chain to be replaced only the top
6590 * most non filter in order to prevent data corruption.
6591 * Another benefit is that this tests exclude backing files which are
6592 * blocked by the backing blockers.
6594 if (!bdrv_recurse_can_replace(parent_bs
, to_replace_bs
)) {
6595 error_setg(errp
, "Cannot replace '%s' by a node mirrored from '%s', "
6596 "because it cannot be guaranteed that doing so would not "
6597 "lead to an abrupt change of visible data",
6598 node_name
, parent_bs
->node_name
);
6599 to_replace_bs
= NULL
;
6604 aio_context_release(aio_context
);
6605 return to_replace_bs
;
6609 * Iterates through the list of runtime option keys that are said to
6610 * be "strong" for a BDS. An option is called "strong" if it changes
6611 * a BDS's data. For example, the null block driver's "size" and
6612 * "read-zeroes" options are strong, but its "latency-ns" option is
6615 * If a key returned by this function ends with a dot, all options
6616 * starting with that prefix are strong.
6618 static const char *const *strong_options(BlockDriverState
*bs
,
6619 const char *const *curopt
)
6621 static const char *const global_options
[] = {
6622 "driver", "filename", NULL
6626 return &global_options
[0];
6630 if (curopt
== &global_options
[ARRAY_SIZE(global_options
) - 1] && bs
->drv
) {
6631 curopt
= bs
->drv
->strong_runtime_opts
;
6634 return (curopt
&& *curopt
) ? curopt
: NULL
;
6638 * Copies all strong runtime options from bs->options to the given
6639 * QDict. The set of strong option keys is determined by invoking
6642 * Returns true iff any strong option was present in bs->options (and
6643 * thus copied to the target QDict) with the exception of "filename"
6644 * and "driver". The caller is expected to use this value to decide
6645 * whether the existence of strong options prevents the generation of
6648 static bool append_strong_runtime_options(QDict
*d
, BlockDriverState
*bs
)
6650 bool found_any
= false;
6651 const char *const *option_name
= NULL
;
6657 while ((option_name
= strong_options(bs
, option_name
))) {
6658 bool option_given
= false;
6660 assert(strlen(*option_name
) > 0);
6661 if ((*option_name
)[strlen(*option_name
) - 1] != '.') {
6662 QObject
*entry
= qdict_get(bs
->options
, *option_name
);
6667 qdict_put_obj(d
, *option_name
, qobject_ref(entry
));
6668 option_given
= true;
6670 const QDictEntry
*entry
;
6671 for (entry
= qdict_first(bs
->options
); entry
;
6672 entry
= qdict_next(bs
->options
, entry
))
6674 if (strstart(qdict_entry_key(entry
), *option_name
, NULL
)) {
6675 qdict_put_obj(d
, qdict_entry_key(entry
),
6676 qobject_ref(qdict_entry_value(entry
)));
6677 option_given
= true;
6682 /* While "driver" and "filename" need to be included in a JSON filename,
6683 * their existence does not prohibit generation of a plain filename. */
6684 if (!found_any
&& option_given
&&
6685 strcmp(*option_name
, "driver") && strcmp(*option_name
, "filename"))
6691 if (!qdict_haskey(d
, "driver")) {
6692 /* Drivers created with bdrv_new_open_driver() may not have a
6693 * @driver option. Add it here. */
6694 qdict_put_str(d
, "driver", bs
->drv
->format_name
);
6700 /* Note: This function may return false positives; it may return true
6701 * even if opening the backing file specified by bs's image header
6702 * would result in exactly bs->backing. */
6703 static bool bdrv_backing_overridden(BlockDriverState
*bs
)
6706 return strcmp(bs
->auto_backing_file
,
6707 bs
->backing
->bs
->filename
);
6709 /* No backing BDS, so if the image header reports any backing
6710 * file, it must have been suppressed */
6711 return bs
->auto_backing_file
[0] != '\0';
6715 /* Updates the following BDS fields:
6716 * - exact_filename: A filename which may be used for opening a block device
6717 * which (mostly) equals the given BDS (even without any
6718 * other options; so reading and writing must return the same
6719 * results, but caching etc. may be different)
6720 * - full_open_options: Options which, when given when opening a block device
6721 * (without a filename), result in a BDS (mostly)
6722 * equalling the given one
6723 * - filename: If exact_filename is set, it is copied here. Otherwise,
6724 * full_open_options is converted to a JSON object, prefixed with
6725 * "json:" (for use through the JSON pseudo protocol) and put here.
6727 void bdrv_refresh_filename(BlockDriverState
*bs
)
6729 BlockDriver
*drv
= bs
->drv
;
6732 bool backing_overridden
;
6733 bool generate_json_filename
; /* Whether our default implementation should
6734 fill exact_filename (false) or not (true) */
6740 /* This BDS's file name may depend on any of its children's file names, so
6741 * refresh those first */
6742 QLIST_FOREACH(child
, &bs
->children
, next
) {
6743 bdrv_refresh_filename(child
->bs
);
6747 /* For implicit nodes, just copy everything from the single child */
6748 child
= QLIST_FIRST(&bs
->children
);
6749 assert(QLIST_NEXT(child
, next
) == NULL
);
6751 pstrcpy(bs
->exact_filename
, sizeof(bs
->exact_filename
),
6752 child
->bs
->exact_filename
);
6753 pstrcpy(bs
->filename
, sizeof(bs
->filename
), child
->bs
->filename
);
6755 qobject_unref(bs
->full_open_options
);
6756 bs
->full_open_options
= qobject_ref(child
->bs
->full_open_options
);
6761 backing_overridden
= bdrv_backing_overridden(bs
);
6763 if (bs
->open_flags
& BDRV_O_NO_IO
) {
6764 /* Without I/O, the backing file does not change anything.
6765 * Therefore, in such a case (primarily qemu-img), we can
6766 * pretend the backing file has not been overridden even if
6767 * it technically has been. */
6768 backing_overridden
= false;
6771 /* Gather the options QDict */
6773 generate_json_filename
= append_strong_runtime_options(opts
, bs
);
6774 generate_json_filename
|= backing_overridden
;
6776 if (drv
->bdrv_gather_child_options
) {
6777 /* Some block drivers may not want to present all of their children's
6778 * options, or name them differently from BdrvChild.name */
6779 drv
->bdrv_gather_child_options(bs
, opts
, backing_overridden
);
6781 QLIST_FOREACH(child
, &bs
->children
, next
) {
6782 if (child
== bs
->backing
&& !backing_overridden
) {
6783 /* We can skip the backing BDS if it has not been overridden */
6787 qdict_put(opts
, child
->name
,
6788 qobject_ref(child
->bs
->full_open_options
));
6791 if (backing_overridden
&& !bs
->backing
) {
6792 /* Force no backing file */
6793 qdict_put_null(opts
, "backing");
6797 qobject_unref(bs
->full_open_options
);
6798 bs
->full_open_options
= opts
;
6800 if (drv
->bdrv_refresh_filename
) {
6801 /* Obsolete information is of no use here, so drop the old file name
6802 * information before refreshing it */
6803 bs
->exact_filename
[0] = '\0';
6805 drv
->bdrv_refresh_filename(bs
);
6806 } else if (bs
->file
) {
6807 /* Try to reconstruct valid information from the underlying file */
6809 bs
->exact_filename
[0] = '\0';
6812 * We can use the underlying file's filename if:
6813 * - it has a filename,
6814 * - the file is a protocol BDS, and
6815 * - opening that file (as this BDS's format) will automatically create
6816 * the BDS tree we have right now, that is:
6817 * - the user did not significantly change this BDS's behavior with
6818 * some explicit (strong) options
6819 * - no non-file child of this BDS has been overridden by the user
6820 * Both of these conditions are represented by generate_json_filename.
6822 if (bs
->file
->bs
->exact_filename
[0] &&
6823 bs
->file
->bs
->drv
->bdrv_file_open
&&
6824 !generate_json_filename
)
6826 strcpy(bs
->exact_filename
, bs
->file
->bs
->exact_filename
);
6830 if (bs
->exact_filename
[0]) {
6831 pstrcpy(bs
->filename
, sizeof(bs
->filename
), bs
->exact_filename
);
6833 QString
*json
= qobject_to_json(QOBJECT(bs
->full_open_options
));
6834 if (snprintf(bs
->filename
, sizeof(bs
->filename
), "json:%s",
6835 qstring_get_str(json
)) >= sizeof(bs
->filename
)) {
6836 /* Give user a hint if we truncated things. */
6837 strcpy(bs
->filename
+ sizeof(bs
->filename
) - 4, "...");
6839 qobject_unref(json
);
6843 char *bdrv_dirname(BlockDriverState
*bs
, Error
**errp
)
6845 BlockDriver
*drv
= bs
->drv
;
6848 error_setg(errp
, "Node '%s' is ejected", bs
->node_name
);
6852 if (drv
->bdrv_dirname
) {
6853 return drv
->bdrv_dirname(bs
, errp
);
6857 return bdrv_dirname(bs
->file
->bs
, errp
);
6860 bdrv_refresh_filename(bs
);
6861 if (bs
->exact_filename
[0] != '\0') {
6862 return path_combine(bs
->exact_filename
, "");
6865 error_setg(errp
, "Cannot generate a base directory for %s nodes",
6871 * Hot add/remove a BDS's child. So the user can take a child offline when
6872 * it is broken and take a new child online
6874 void bdrv_add_child(BlockDriverState
*parent_bs
, BlockDriverState
*child_bs
,
6878 if (!parent_bs
->drv
|| !parent_bs
->drv
->bdrv_add_child
) {
6879 error_setg(errp
, "The node %s does not support adding a child",
6880 bdrv_get_device_or_node_name(parent_bs
));
6884 if (!QLIST_EMPTY(&child_bs
->parents
)) {
6885 error_setg(errp
, "The node %s already has a parent",
6886 child_bs
->node_name
);
6890 parent_bs
->drv
->bdrv_add_child(parent_bs
, child_bs
, errp
);
6893 void bdrv_del_child(BlockDriverState
*parent_bs
, BdrvChild
*child
, Error
**errp
)
6897 if (!parent_bs
->drv
|| !parent_bs
->drv
->bdrv_del_child
) {
6898 error_setg(errp
, "The node %s does not support removing a child",
6899 bdrv_get_device_or_node_name(parent_bs
));
6903 QLIST_FOREACH(tmp
, &parent_bs
->children
, next
) {
6910 error_setg(errp
, "The node %s does not have a child named %s",
6911 bdrv_get_device_or_node_name(parent_bs
),
6912 bdrv_get_device_or_node_name(child
->bs
));
6916 parent_bs
->drv
->bdrv_del_child(parent_bs
, child
, errp
);
6919 int bdrv_make_empty(BdrvChild
*c
, Error
**errp
)
6921 BlockDriver
*drv
= c
->bs
->drv
;
6924 assert(c
->perm
& (BLK_PERM_WRITE
| BLK_PERM_WRITE_UNCHANGED
));
6926 if (!drv
->bdrv_make_empty
) {
6927 error_setg(errp
, "%s does not support emptying nodes",
6932 ret
= drv
->bdrv_make_empty(c
->bs
);
6934 error_setg_errno(errp
, -ret
, "Failed to empty %s",