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 QDict
*parse_json_filename(const char *filename
, Error
**errp
)
1753 QObject
*options_obj
;
1757 ret
= strstart(filename
, "json:", &filename
);
1760 options_obj
= qobject_from_json(filename
, errp
);
1762 error_prepend(errp
, "Could not parse the JSON options: ");
1766 options
= qobject_to(QDict
, options_obj
);
1768 qobject_unref(options_obj
);
1769 error_setg(errp
, "Invalid JSON object given");
1773 qdict_flatten(options
);
1778 static void parse_json_protocol(QDict
*options
, const char **pfilename
,
1781 QDict
*json_options
;
1782 Error
*local_err
= NULL
;
1784 /* Parse json: pseudo-protocol */
1785 if (!*pfilename
|| !g_str_has_prefix(*pfilename
, "json:")) {
1789 json_options
= parse_json_filename(*pfilename
, &local_err
);
1791 error_propagate(errp
, local_err
);
1795 /* Options given in the filename have lower priority than options
1796 * specified directly */
1797 qdict_join(options
, json_options
, false);
1798 qobject_unref(json_options
);
1803 * Fills in default options for opening images and converts the legacy
1804 * filename/flags pair to option QDict entries.
1805 * The BDRV_O_PROTOCOL flag in *flags will be set or cleared accordingly if a
1806 * block driver has been specified explicitly.
1808 static int bdrv_fill_options(QDict
**options
, const char *filename
,
1809 int *flags
, Error
**errp
)
1811 const char *drvname
;
1812 bool protocol
= *flags
& BDRV_O_PROTOCOL
;
1813 bool parse_filename
= false;
1814 BlockDriver
*drv
= NULL
;
1815 Error
*local_err
= NULL
;
1818 * Caution: while qdict_get_try_str() is fine, getting non-string
1819 * types would require more care. When @options come from
1820 * -blockdev or blockdev_add, its members are typed according to
1821 * the QAPI schema, but when they come from -drive, they're all
1824 drvname
= qdict_get_try_str(*options
, "driver");
1826 drv
= bdrv_find_format(drvname
);
1828 error_setg(errp
, "Unknown driver '%s'", drvname
);
1831 /* If the user has explicitly specified the driver, this choice should
1832 * override the BDRV_O_PROTOCOL flag */
1833 protocol
= drv
->bdrv_file_open
;
1837 *flags
|= BDRV_O_PROTOCOL
;
1839 *flags
&= ~BDRV_O_PROTOCOL
;
1842 /* Translate cache options from flags into options */
1843 update_options_from_flags(*options
, *flags
);
1845 /* Fetch the file name from the options QDict if necessary */
1846 if (protocol
&& filename
) {
1847 if (!qdict_haskey(*options
, "filename")) {
1848 qdict_put_str(*options
, "filename", filename
);
1849 parse_filename
= true;
1851 error_setg(errp
, "Can't specify 'file' and 'filename' options at "
1857 /* Find the right block driver */
1858 /* See cautionary note on accessing @options above */
1859 filename
= qdict_get_try_str(*options
, "filename");
1861 if (!drvname
&& protocol
) {
1863 drv
= bdrv_find_protocol(filename
, parse_filename
, errp
);
1868 drvname
= drv
->format_name
;
1869 qdict_put_str(*options
, "driver", drvname
);
1871 error_setg(errp
, "Must specify either driver or file");
1876 assert(drv
|| !protocol
);
1878 /* Driver-specific filename parsing */
1879 if (drv
&& drv
->bdrv_parse_filename
&& parse_filename
) {
1880 drv
->bdrv_parse_filename(filename
, *options
, &local_err
);
1882 error_propagate(errp
, local_err
);
1886 if (!drv
->bdrv_needs_filename
) {
1887 qdict_del(*options
, "filename");
1894 static int bdrv_child_check_perm(BdrvChild
*c
, BlockReopenQueue
*q
,
1895 uint64_t perm
, uint64_t shared
,
1896 GSList
*ignore_children
,
1897 bool *tighten_restrictions
, Error
**errp
);
1898 static void bdrv_child_abort_perm_update(BdrvChild
*c
);
1899 static void bdrv_child_set_perm(BdrvChild
*c
, uint64_t perm
, uint64_t shared
);
1901 typedef struct BlockReopenQueueEntry
{
1904 BDRVReopenState state
;
1905 QTAILQ_ENTRY(BlockReopenQueueEntry
) entry
;
1906 } BlockReopenQueueEntry
;
1909 * Return the flags that @bs will have after the reopens in @q have
1910 * successfully completed. If @q is NULL (or @bs is not contained in @q),
1911 * return the current flags.
1913 static int bdrv_reopen_get_flags(BlockReopenQueue
*q
, BlockDriverState
*bs
)
1915 BlockReopenQueueEntry
*entry
;
1918 QTAILQ_FOREACH(entry
, q
, entry
) {
1919 if (entry
->state
.bs
== bs
) {
1920 return entry
->state
.flags
;
1925 return bs
->open_flags
;
1928 /* Returns whether the image file can be written to after the reopen queue @q
1929 * has been successfully applied, or right now if @q is NULL. */
1930 static bool bdrv_is_writable_after_reopen(BlockDriverState
*bs
,
1931 BlockReopenQueue
*q
)
1933 int flags
= bdrv_reopen_get_flags(q
, bs
);
1935 return (flags
& (BDRV_O_RDWR
| BDRV_O_INACTIVE
)) == BDRV_O_RDWR
;
1939 * Return whether the BDS can be written to. This is not necessarily
1940 * the same as !bdrv_is_read_only(bs), as inactivated images may not
1941 * be written to but do not count as read-only images.
1943 bool bdrv_is_writable(BlockDriverState
*bs
)
1945 return bdrv_is_writable_after_reopen(bs
, NULL
);
1948 static void bdrv_child_perm(BlockDriverState
*bs
, BlockDriverState
*child_bs
,
1949 BdrvChild
*c
, BdrvChildRole role
,
1950 BlockReopenQueue
*reopen_queue
,
1951 uint64_t parent_perm
, uint64_t parent_shared
,
1952 uint64_t *nperm
, uint64_t *nshared
)
1954 assert(bs
->drv
&& bs
->drv
->bdrv_child_perm
);
1955 bs
->drv
->bdrv_child_perm(bs
, c
, role
, reopen_queue
,
1956 parent_perm
, parent_shared
,
1958 /* TODO Take force_share from reopen_queue */
1959 if (child_bs
&& child_bs
->force_share
) {
1960 *nshared
= BLK_PERM_ALL
;
1965 * Check whether permissions on this node can be changed in a way that
1966 * @cumulative_perms and @cumulative_shared_perms are the new cumulative
1967 * permissions of all its parents. This involves checking whether all necessary
1968 * permission changes to child nodes can be performed.
1970 * Will set *tighten_restrictions to true if and only if new permissions have to
1971 * be taken or currently shared permissions are to be unshared. Otherwise,
1972 * errors are not fatal as long as the caller accepts that the restrictions
1973 * remain tighter than they need to be. The caller still has to abort the
1975 * @tighten_restrictions cannot be used together with @q: When reopening, we may
1976 * encounter fatal errors even though no restrictions are to be tightened. For
1977 * example, changing a node from RW to RO will fail if the WRITE permission is
1980 * A call to this function must always be followed by a call to bdrv_set_perm()
1981 * or bdrv_abort_perm_update().
1983 static int bdrv_check_perm(BlockDriverState
*bs
, BlockReopenQueue
*q
,
1984 uint64_t cumulative_perms
,
1985 uint64_t cumulative_shared_perms
,
1986 GSList
*ignore_children
,
1987 bool *tighten_restrictions
, Error
**errp
)
1989 BlockDriver
*drv
= bs
->drv
;
1993 assert(!q
|| !tighten_restrictions
);
1995 if (tighten_restrictions
) {
1996 uint64_t current_perms
, current_shared
;
1997 uint64_t added_perms
, removed_shared_perms
;
1999 bdrv_get_cumulative_perm(bs
, ¤t_perms
, ¤t_shared
);
2001 added_perms
= cumulative_perms
& ~current_perms
;
2002 removed_shared_perms
= current_shared
& ~cumulative_shared_perms
;
2004 *tighten_restrictions
= added_perms
|| removed_shared_perms
;
2007 /* Write permissions never work with read-only images */
2008 if ((cumulative_perms
& (BLK_PERM_WRITE
| BLK_PERM_WRITE_UNCHANGED
)) &&
2009 !bdrv_is_writable_after_reopen(bs
, q
))
2011 if (!bdrv_is_writable_after_reopen(bs
, NULL
)) {
2012 error_setg(errp
, "Block node is read-only");
2014 uint64_t current_perms
, current_shared
;
2015 bdrv_get_cumulative_perm(bs
, ¤t_perms
, ¤t_shared
);
2016 if (current_perms
& (BLK_PERM_WRITE
| BLK_PERM_WRITE_UNCHANGED
)) {
2017 error_setg(errp
, "Cannot make block node read-only, there is "
2020 error_setg(errp
, "Cannot make block node read-only and create "
2029 * Unaligned requests will automatically be aligned to bl.request_alignment
2030 * and without RESIZE we can't extend requests to write to space beyond the
2031 * end of the image, so it's required that the image size is aligned.
2033 if ((cumulative_perms
& (BLK_PERM_WRITE
| BLK_PERM_WRITE_UNCHANGED
)) &&
2034 !(cumulative_perms
& BLK_PERM_RESIZE
))
2036 if ((bs
->total_sectors
* BDRV_SECTOR_SIZE
) % bs
->bl
.request_alignment
) {
2037 error_setg(errp
, "Cannot get 'write' permission without 'resize': "
2038 "Image size is not a multiple of request "
2044 /* Check this node */
2049 if (drv
->bdrv_check_perm
) {
2050 return drv
->bdrv_check_perm(bs
, cumulative_perms
,
2051 cumulative_shared_perms
, errp
);
2054 /* Drivers that never have children can omit .bdrv_child_perm() */
2055 if (!drv
->bdrv_child_perm
) {
2056 assert(QLIST_EMPTY(&bs
->children
));
2060 /* Check all children */
2061 QLIST_FOREACH(c
, &bs
->children
, next
) {
2062 uint64_t cur_perm
, cur_shared
;
2063 bool child_tighten_restr
;
2065 bdrv_child_perm(bs
, c
->bs
, c
, c
->role
, q
,
2066 cumulative_perms
, cumulative_shared_perms
,
2067 &cur_perm
, &cur_shared
);
2068 ret
= bdrv_child_check_perm(c
, q
, cur_perm
, cur_shared
, ignore_children
,
2069 tighten_restrictions ?
&child_tighten_restr
2072 if (tighten_restrictions
) {
2073 *tighten_restrictions
|= child_tighten_restr
;
2084 * Notifies drivers that after a previous bdrv_check_perm() call, the
2085 * permission update is not performed and any preparations made for it (e.g.
2086 * taken file locks) need to be undone.
2088 * This function recursively notifies all child nodes.
2090 static void bdrv_abort_perm_update(BlockDriverState
*bs
)
2092 BlockDriver
*drv
= bs
->drv
;
2099 if (drv
->bdrv_abort_perm_update
) {
2100 drv
->bdrv_abort_perm_update(bs
);
2103 QLIST_FOREACH(c
, &bs
->children
, next
) {
2104 bdrv_child_abort_perm_update(c
);
2108 static void bdrv_set_perm(BlockDriverState
*bs
, uint64_t cumulative_perms
,
2109 uint64_t cumulative_shared_perms
)
2111 BlockDriver
*drv
= bs
->drv
;
2118 /* Update this node */
2119 if (drv
->bdrv_set_perm
) {
2120 drv
->bdrv_set_perm(bs
, cumulative_perms
, cumulative_shared_perms
);
2123 /* Drivers that never have children can omit .bdrv_child_perm() */
2124 if (!drv
->bdrv_child_perm
) {
2125 assert(QLIST_EMPTY(&bs
->children
));
2129 /* Update all children */
2130 QLIST_FOREACH(c
, &bs
->children
, next
) {
2131 uint64_t cur_perm
, cur_shared
;
2132 bdrv_child_perm(bs
, c
->bs
, c
, c
->role
, NULL
,
2133 cumulative_perms
, cumulative_shared_perms
,
2134 &cur_perm
, &cur_shared
);
2135 bdrv_child_set_perm(c
, cur_perm
, cur_shared
);
2139 void bdrv_get_cumulative_perm(BlockDriverState
*bs
, uint64_t *perm
,
2140 uint64_t *shared_perm
)
2143 uint64_t cumulative_perms
= 0;
2144 uint64_t cumulative_shared_perms
= BLK_PERM_ALL
;
2146 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
2147 cumulative_perms
|= c
->perm
;
2148 cumulative_shared_perms
&= c
->shared_perm
;
2151 *perm
= cumulative_perms
;
2152 *shared_perm
= cumulative_shared_perms
;
2155 static char *bdrv_child_user_desc(BdrvChild
*c
)
2157 if (c
->klass
->get_parent_desc
) {
2158 return c
->klass
->get_parent_desc(c
);
2161 return g_strdup("another user");
2164 char *bdrv_perm_names(uint64_t perm
)
2170 { BLK_PERM_CONSISTENT_READ
, "consistent read" },
2171 { BLK_PERM_WRITE
, "write" },
2172 { BLK_PERM_WRITE_UNCHANGED
, "write unchanged" },
2173 { BLK_PERM_RESIZE
, "resize" },
2174 { BLK_PERM_GRAPH_MOD
, "change children" },
2178 GString
*result
= g_string_sized_new(30);
2179 struct perm_name
*p
;
2181 for (p
= permissions
; p
->name
; p
++) {
2182 if (perm
& p
->perm
) {
2183 if (result
->len
> 0) {
2184 g_string_append(result
, ", ");
2186 g_string_append(result
, p
->name
);
2190 return g_string_free(result
, FALSE
);
2194 * Checks whether a new reference to @bs can be added if the new user requires
2195 * @new_used_perm/@new_shared_perm as its permissions. If @ignore_children is
2196 * set, the BdrvChild objects in this list are ignored in the calculations;
2197 * this allows checking permission updates for an existing reference.
2199 * See bdrv_check_perm() for the semantics of @tighten_restrictions.
2201 * Needs to be followed by a call to either bdrv_set_perm() or
2202 * bdrv_abort_perm_update(). */
2203 static int bdrv_check_update_perm(BlockDriverState
*bs
, BlockReopenQueue
*q
,
2204 uint64_t new_used_perm
,
2205 uint64_t new_shared_perm
,
2206 GSList
*ignore_children
,
2207 bool *tighten_restrictions
,
2211 uint64_t cumulative_perms
= new_used_perm
;
2212 uint64_t cumulative_shared_perms
= new_shared_perm
;
2214 assert(!q
|| !tighten_restrictions
);
2216 /* There is no reason why anyone couldn't tolerate write_unchanged */
2217 assert(new_shared_perm
& BLK_PERM_WRITE_UNCHANGED
);
2219 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
2220 if (g_slist_find(ignore_children
, c
)) {
2224 if ((new_used_perm
& c
->shared_perm
) != new_used_perm
) {
2225 char *user
= bdrv_child_user_desc(c
);
2226 char *perm_names
= bdrv_perm_names(new_used_perm
& ~c
->shared_perm
);
2228 if (tighten_restrictions
) {
2229 *tighten_restrictions
= true;
2232 error_setg(errp
, "Conflicts with use by %s as '%s', which does not "
2234 user
, c
->name
, perm_names
, bdrv_get_node_name(c
->bs
));
2240 if ((c
->perm
& new_shared_perm
) != c
->perm
) {
2241 char *user
= bdrv_child_user_desc(c
);
2242 char *perm_names
= bdrv_perm_names(c
->perm
& ~new_shared_perm
);
2244 if (tighten_restrictions
) {
2245 *tighten_restrictions
= true;
2248 error_setg(errp
, "Conflicts with use by %s as '%s', which uses "
2250 user
, c
->name
, perm_names
, bdrv_get_node_name(c
->bs
));
2256 cumulative_perms
|= c
->perm
;
2257 cumulative_shared_perms
&= c
->shared_perm
;
2260 return bdrv_check_perm(bs
, q
, cumulative_perms
, cumulative_shared_perms
,
2261 ignore_children
, tighten_restrictions
, errp
);
2264 /* Needs to be followed by a call to either bdrv_child_set_perm() or
2265 * bdrv_child_abort_perm_update(). */
2266 static int bdrv_child_check_perm(BdrvChild
*c
, BlockReopenQueue
*q
,
2267 uint64_t perm
, uint64_t shared
,
2268 GSList
*ignore_children
,
2269 bool *tighten_restrictions
, Error
**errp
)
2273 ignore_children
= g_slist_prepend(g_slist_copy(ignore_children
), c
);
2274 ret
= bdrv_check_update_perm(c
->bs
, q
, perm
, shared
, ignore_children
,
2275 tighten_restrictions
, errp
);
2276 g_slist_free(ignore_children
);
2282 if (!c
->has_backup_perm
) {
2283 c
->has_backup_perm
= true;
2284 c
->backup_perm
= c
->perm
;
2285 c
->backup_shared_perm
= c
->shared_perm
;
2288 * Note: it's OK if c->has_backup_perm was already set, as we can find the
2289 * same child twice during check_perm procedure
2293 c
->shared_perm
= shared
;
2298 static void bdrv_child_set_perm(BdrvChild
*c
, uint64_t perm
, uint64_t shared
)
2300 uint64_t cumulative_perms
, cumulative_shared_perms
;
2302 c
->has_backup_perm
= false;
2305 c
->shared_perm
= shared
;
2307 bdrv_get_cumulative_perm(c
->bs
, &cumulative_perms
,
2308 &cumulative_shared_perms
);
2309 bdrv_set_perm(c
->bs
, cumulative_perms
, cumulative_shared_perms
);
2312 static void bdrv_child_abort_perm_update(BdrvChild
*c
)
2314 if (c
->has_backup_perm
) {
2315 c
->perm
= c
->backup_perm
;
2316 c
->shared_perm
= c
->backup_shared_perm
;
2317 c
->has_backup_perm
= false;
2320 bdrv_abort_perm_update(c
->bs
);
2323 int bdrv_child_try_set_perm(BdrvChild
*c
, uint64_t perm
, uint64_t shared
,
2326 Error
*local_err
= NULL
;
2328 bool tighten_restrictions
;
2330 ret
= bdrv_child_check_perm(c
, NULL
, perm
, shared
, NULL
,
2331 &tighten_restrictions
, &local_err
);
2333 bdrv_child_abort_perm_update(c
);
2334 if (tighten_restrictions
) {
2335 error_propagate(errp
, local_err
);
2338 * Our caller may intend to only loosen restrictions and
2339 * does not expect this function to fail. Errors are not
2340 * fatal in such a case, so we can just hide them from our
2343 error_free(local_err
);
2349 bdrv_child_set_perm(c
, perm
, shared
);
2354 int bdrv_child_refresh_perms(BlockDriverState
*bs
, BdrvChild
*c
, Error
**errp
)
2356 uint64_t parent_perms
, parent_shared
;
2357 uint64_t perms
, shared
;
2359 bdrv_get_cumulative_perm(bs
, &parent_perms
, &parent_shared
);
2360 bdrv_child_perm(bs
, c
->bs
, c
, c
->role
, NULL
,
2361 parent_perms
, parent_shared
, &perms
, &shared
);
2363 return bdrv_child_try_set_perm(c
, perms
, shared
, errp
);
2367 * Default implementation for .bdrv_child_perm() for block filters:
2368 * Forward CONSISTENT_READ, WRITE, WRITE_UNCHANGED, and RESIZE to the
2371 static void bdrv_filter_default_perms(BlockDriverState
*bs
, BdrvChild
*c
,
2373 BlockReopenQueue
*reopen_queue
,
2374 uint64_t perm
, uint64_t shared
,
2375 uint64_t *nperm
, uint64_t *nshared
)
2377 *nperm
= perm
& DEFAULT_PERM_PASSTHROUGH
;
2378 *nshared
= (shared
& DEFAULT_PERM_PASSTHROUGH
) | DEFAULT_PERM_UNCHANGED
;
2381 static void bdrv_default_perms_for_cow(BlockDriverState
*bs
, BdrvChild
*c
,
2383 BlockReopenQueue
*reopen_queue
,
2384 uint64_t perm
, uint64_t shared
,
2385 uint64_t *nperm
, uint64_t *nshared
)
2387 assert(role
& BDRV_CHILD_COW
);
2390 * We want consistent read from backing files if the parent needs it.
2391 * No other operations are performed on backing files.
2393 perm
&= BLK_PERM_CONSISTENT_READ
;
2396 * If the parent can deal with changing data, we're okay with a
2397 * writable and resizable backing file.
2398 * TODO Require !(perm & BLK_PERM_CONSISTENT_READ), too?
2400 if (shared
& BLK_PERM_WRITE
) {
2401 shared
= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2406 shared
|= BLK_PERM_CONSISTENT_READ
| BLK_PERM_GRAPH_MOD
|
2407 BLK_PERM_WRITE_UNCHANGED
;
2409 if (bs
->open_flags
& BDRV_O_INACTIVE
) {
2410 shared
|= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2417 static void bdrv_default_perms_for_storage(BlockDriverState
*bs
, BdrvChild
*c
,
2419 BlockReopenQueue
*reopen_queue
,
2420 uint64_t perm
, uint64_t shared
,
2421 uint64_t *nperm
, uint64_t *nshared
)
2425 assert(role
& (BDRV_CHILD_METADATA
| BDRV_CHILD_DATA
));
2427 flags
= bdrv_reopen_get_flags(reopen_queue
, bs
);
2430 * Apart from the modifications below, the same permissions are
2431 * forwarded and left alone as for filters
2433 bdrv_filter_default_perms(bs
, c
, role
, reopen_queue
,
2434 perm
, shared
, &perm
, &shared
);
2436 if (role
& BDRV_CHILD_METADATA
) {
2437 /* Format drivers may touch metadata even if the guest doesn't write */
2438 if (bdrv_is_writable_after_reopen(bs
, reopen_queue
)) {
2439 perm
|= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2443 * bs->file always needs to be consistent because of the
2444 * metadata. We can never allow other users to resize or write
2447 if (!(flags
& BDRV_O_NO_IO
)) {
2448 perm
|= BLK_PERM_CONSISTENT_READ
;
2450 shared
&= ~(BLK_PERM_WRITE
| BLK_PERM_RESIZE
);
2453 if (role
& BDRV_CHILD_DATA
) {
2455 * Technically, everything in this block is a subset of the
2456 * BDRV_CHILD_METADATA path taken above, and so this could
2457 * be an "else if" branch. However, that is not obvious, and
2458 * this function is not performance critical, therefore we let
2459 * this be an independent "if".
2463 * We cannot allow other users to resize the file because the
2464 * format driver might have some assumptions about the size
2465 * (e.g. because it is stored in metadata, or because the file
2466 * is split into fixed-size data files).
2468 shared
&= ~BLK_PERM_RESIZE
;
2471 * WRITE_UNCHANGED often cannot be performed as such on the
2472 * data file. For example, the qcow2 driver may still need to
2473 * write copied clusters on copy-on-read.
2475 if (perm
& BLK_PERM_WRITE_UNCHANGED
) {
2476 perm
|= BLK_PERM_WRITE
;
2480 * If the data file is written to, the format driver may
2481 * expect to be able to resize it by writing beyond the EOF.
2483 if (perm
& BLK_PERM_WRITE
) {
2484 perm
|= BLK_PERM_RESIZE
;
2488 if (bs
->open_flags
& BDRV_O_INACTIVE
) {
2489 shared
|= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2496 void bdrv_default_perms(BlockDriverState
*bs
, BdrvChild
*c
,
2497 BdrvChildRole role
, BlockReopenQueue
*reopen_queue
,
2498 uint64_t perm
, uint64_t shared
,
2499 uint64_t *nperm
, uint64_t *nshared
)
2501 if (role
& BDRV_CHILD_FILTERED
) {
2502 assert(!(role
& (BDRV_CHILD_DATA
| BDRV_CHILD_METADATA
|
2504 bdrv_filter_default_perms(bs
, c
, role
, reopen_queue
,
2505 perm
, shared
, nperm
, nshared
);
2506 } else if (role
& BDRV_CHILD_COW
) {
2507 assert(!(role
& (BDRV_CHILD_DATA
| BDRV_CHILD_METADATA
)));
2508 bdrv_default_perms_for_cow(bs
, c
, role
, reopen_queue
,
2509 perm
, shared
, nperm
, nshared
);
2510 } else if (role
& (BDRV_CHILD_METADATA
| BDRV_CHILD_DATA
)) {
2511 bdrv_default_perms_for_storage(bs
, c
, role
, reopen_queue
,
2512 perm
, shared
, nperm
, nshared
);
2514 g_assert_not_reached();
2518 uint64_t bdrv_qapi_perm_to_blk_perm(BlockPermission qapi_perm
)
2520 static const uint64_t permissions
[] = {
2521 [BLOCK_PERMISSION_CONSISTENT_READ
] = BLK_PERM_CONSISTENT_READ
,
2522 [BLOCK_PERMISSION_WRITE
] = BLK_PERM_WRITE
,
2523 [BLOCK_PERMISSION_WRITE_UNCHANGED
] = BLK_PERM_WRITE_UNCHANGED
,
2524 [BLOCK_PERMISSION_RESIZE
] = BLK_PERM_RESIZE
,
2525 [BLOCK_PERMISSION_GRAPH_MOD
] = BLK_PERM_GRAPH_MOD
,
2528 QEMU_BUILD_BUG_ON(ARRAY_SIZE(permissions
) != BLOCK_PERMISSION__MAX
);
2529 QEMU_BUILD_BUG_ON(1UL << ARRAY_SIZE(permissions
) != BLK_PERM_ALL
+ 1);
2531 assert(qapi_perm
< BLOCK_PERMISSION__MAX
);
2533 return permissions
[qapi_perm
];
2536 static void bdrv_replace_child_noperm(BdrvChild
*child
,
2537 BlockDriverState
*new_bs
)
2539 BlockDriverState
*old_bs
= child
->bs
;
2540 int new_bs_quiesce_counter
;
2543 assert(!child
->frozen
);
2545 if (old_bs
&& new_bs
) {
2546 assert(bdrv_get_aio_context(old_bs
) == bdrv_get_aio_context(new_bs
));
2549 new_bs_quiesce_counter
= (new_bs ? new_bs
->quiesce_counter
: 0);
2550 drain_saldo
= new_bs_quiesce_counter
- child
->parent_quiesce_counter
;
2553 * If the new child node is drained but the old one was not, flush
2554 * all outstanding requests to the old child node.
2556 while (drain_saldo
> 0 && child
->klass
->drained_begin
) {
2557 bdrv_parent_drained_begin_single(child
, true);
2562 /* Detach first so that the recursive drain sections coming from @child
2563 * are already gone and we only end the drain sections that came from
2565 if (child
->klass
->detach
) {
2566 child
->klass
->detach(child
);
2568 QLIST_REMOVE(child
, next_parent
);
2574 QLIST_INSERT_HEAD(&new_bs
->parents
, child
, next_parent
);
2577 * Detaching the old node may have led to the new node's
2578 * quiesce_counter having been decreased. Not a problem, we
2579 * just need to recognize this here and then invoke
2580 * drained_end appropriately more often.
2582 assert(new_bs
->quiesce_counter
<= new_bs_quiesce_counter
);
2583 drain_saldo
+= new_bs
->quiesce_counter
- new_bs_quiesce_counter
;
2585 /* Attach only after starting new drained sections, so that recursive
2586 * drain sections coming from @child don't get an extra .drained_begin
2588 if (child
->klass
->attach
) {
2589 child
->klass
->attach(child
);
2594 * If the old child node was drained but the new one is not, allow
2595 * requests to come in only after the new node has been attached.
2597 while (drain_saldo
< 0 && child
->klass
->drained_end
) {
2598 bdrv_parent_drained_end_single(child
);
2604 * Updates @child to change its reference to point to @new_bs, including
2605 * checking and applying the necessary permisson updates both to the old node
2608 * NULL is passed as @new_bs for removing the reference before freeing @child.
2610 * If @new_bs is not NULL, bdrv_check_perm() must be called beforehand, as this
2611 * function uses bdrv_set_perm() to update the permissions according to the new
2612 * reference that @new_bs gets.
2614 static void bdrv_replace_child(BdrvChild
*child
, BlockDriverState
*new_bs
)
2616 BlockDriverState
*old_bs
= child
->bs
;
2617 uint64_t perm
, shared_perm
;
2619 bdrv_replace_child_noperm(child
, new_bs
);
2622 * Start with the new node's permissions. If @new_bs is a (direct
2623 * or indirect) child of @old_bs, we must complete the permission
2624 * update on @new_bs before we loosen the restrictions on @old_bs.
2625 * Otherwise, bdrv_check_perm() on @old_bs would re-initiate
2626 * updating the permissions of @new_bs, and thus not purely loosen
2630 bdrv_get_cumulative_perm(new_bs
, &perm
, &shared_perm
);
2631 bdrv_set_perm(new_bs
, perm
, shared_perm
);
2635 /* Update permissions for old node. This is guaranteed to succeed
2636 * because we're just taking a parent away, so we're loosening
2638 bool tighten_restrictions
;
2641 bdrv_get_cumulative_perm(old_bs
, &perm
, &shared_perm
);
2642 ret
= bdrv_check_perm(old_bs
, NULL
, perm
, shared_perm
, NULL
,
2643 &tighten_restrictions
, NULL
);
2644 assert(tighten_restrictions
== false);
2646 /* We only tried to loosen restrictions, so errors are not fatal */
2647 bdrv_abort_perm_update(old_bs
);
2649 bdrv_set_perm(old_bs
, perm
, shared_perm
);
2652 /* When the parent requiring a non-default AioContext is removed, the
2653 * node moves back to the main AioContext */
2654 bdrv_try_set_aio_context(old_bs
, qemu_get_aio_context(), NULL
);
2659 * This function steals the reference to child_bs from the caller.
2660 * That reference is later dropped by bdrv_root_unref_child().
2662 * On failure NULL is returned, errp is set and the reference to
2663 * child_bs is also dropped.
2665 * The caller must hold the AioContext lock @child_bs, but not that of @ctx
2666 * (unless @child_bs is already in @ctx).
2668 BdrvChild
*bdrv_root_attach_child(BlockDriverState
*child_bs
,
2669 const char *child_name
,
2670 const BdrvChildClass
*child_class
,
2671 BdrvChildRole child_role
,
2673 uint64_t perm
, uint64_t shared_perm
,
2674 void *opaque
, Error
**errp
)
2677 Error
*local_err
= NULL
;
2680 ret
= bdrv_check_update_perm(child_bs
, NULL
, perm
, shared_perm
, NULL
, NULL
,
2683 bdrv_abort_perm_update(child_bs
);
2684 bdrv_unref(child_bs
);
2688 child
= g_new(BdrvChild
, 1);
2689 *child
= (BdrvChild
) {
2691 .name
= g_strdup(child_name
),
2692 .klass
= child_class
,
2695 .shared_perm
= shared_perm
,
2699 /* If the AioContexts don't match, first try to move the subtree of
2700 * child_bs into the AioContext of the new parent. If this doesn't work,
2701 * try moving the parent into the AioContext of child_bs instead. */
2702 if (bdrv_get_aio_context(child_bs
) != ctx
) {
2703 ret
= bdrv_try_set_aio_context(child_bs
, ctx
, &local_err
);
2704 if (ret
< 0 && child_class
->can_set_aio_ctx
) {
2705 GSList
*ignore
= g_slist_prepend(NULL
, child
);
2706 ctx
= bdrv_get_aio_context(child_bs
);
2707 if (child_class
->can_set_aio_ctx(child
, ctx
, &ignore
, NULL
)) {
2708 error_free(local_err
);
2710 g_slist_free(ignore
);
2711 ignore
= g_slist_prepend(NULL
, child
);
2712 child_class
->set_aio_ctx(child
, ctx
, &ignore
);
2714 g_slist_free(ignore
);
2717 error_propagate(errp
, local_err
);
2719 bdrv_abort_perm_update(child_bs
);
2720 bdrv_unref(child_bs
);
2725 /* This performs the matching bdrv_set_perm() for the above check. */
2726 bdrv_replace_child(child
, child_bs
);
2732 * This function transfers the reference to child_bs from the caller
2733 * to parent_bs. That reference is later dropped by parent_bs on
2734 * bdrv_close() or if someone calls bdrv_unref_child().
2736 * On failure NULL is returned, errp is set and the reference to
2737 * child_bs is also dropped.
2739 * If @parent_bs and @child_bs are in different AioContexts, the caller must
2740 * hold the AioContext lock for @child_bs, but not for @parent_bs.
2742 BdrvChild
*bdrv_attach_child(BlockDriverState
*parent_bs
,
2743 BlockDriverState
*child_bs
,
2744 const char *child_name
,
2745 const BdrvChildClass
*child_class
,
2746 BdrvChildRole child_role
,
2750 uint64_t perm
, shared_perm
;
2752 bdrv_get_cumulative_perm(parent_bs
, &perm
, &shared_perm
);
2754 assert(parent_bs
->drv
);
2755 bdrv_child_perm(parent_bs
, child_bs
, NULL
, child_role
, NULL
,
2756 perm
, shared_perm
, &perm
, &shared_perm
);
2758 child
= bdrv_root_attach_child(child_bs
, child_name
, child_class
,
2759 child_role
, bdrv_get_aio_context(parent_bs
),
2760 perm
, shared_perm
, parent_bs
, errp
);
2761 if (child
== NULL
) {
2765 QLIST_INSERT_HEAD(&parent_bs
->children
, child
, next
);
2769 static void bdrv_detach_child(BdrvChild
*child
)
2771 QLIST_SAFE_REMOVE(child
, next
);
2773 bdrv_replace_child(child
, NULL
);
2775 g_free(child
->name
);
2779 void bdrv_root_unref_child(BdrvChild
*child
)
2781 BlockDriverState
*child_bs
;
2783 child_bs
= child
->bs
;
2784 bdrv_detach_child(child
);
2785 bdrv_unref(child_bs
);
2789 * Clear all inherits_from pointers from children and grandchildren of
2790 * @root that point to @root, where necessary.
2792 static void bdrv_unset_inherits_from(BlockDriverState
*root
, BdrvChild
*child
)
2796 if (child
->bs
->inherits_from
== root
) {
2798 * Remove inherits_from only when the last reference between root and
2799 * child->bs goes away.
2801 QLIST_FOREACH(c
, &root
->children
, next
) {
2802 if (c
!= child
&& c
->bs
== child
->bs
) {
2807 child
->bs
->inherits_from
= NULL
;
2811 QLIST_FOREACH(c
, &child
->bs
->children
, next
) {
2812 bdrv_unset_inherits_from(root
, c
);
2816 void bdrv_unref_child(BlockDriverState
*parent
, BdrvChild
*child
)
2818 if (child
== NULL
) {
2822 bdrv_unset_inherits_from(parent
, child
);
2823 bdrv_root_unref_child(child
);
2827 static void bdrv_parent_cb_change_media(BlockDriverState
*bs
, bool load
)
2830 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
2831 if (c
->klass
->change_media
) {
2832 c
->klass
->change_media(c
, load
);
2837 /* Return true if you can reach parent going through child->inherits_from
2838 * recursively. If parent or child are NULL, return false */
2839 static bool bdrv_inherits_from_recursive(BlockDriverState
*child
,
2840 BlockDriverState
*parent
)
2842 while (child
&& child
!= parent
) {
2843 child
= child
->inherits_from
;
2846 return child
!= NULL
;
2850 * Return the BdrvChildRole for @bs's backing child. bs->backing is
2851 * mostly used for COW backing children (role = COW), but also for
2852 * filtered children (role = FILTERED | PRIMARY).
2854 static BdrvChildRole
bdrv_backing_role(BlockDriverState
*bs
)
2856 if (bs
->drv
&& bs
->drv
->is_filter
) {
2857 return BDRV_CHILD_FILTERED
| BDRV_CHILD_PRIMARY
;
2859 return BDRV_CHILD_COW
;
2864 * Sets the backing file link of a BDS. A new reference is created; callers
2865 * which don't need their own reference any more must call bdrv_unref().
2867 void bdrv_set_backing_hd(BlockDriverState
*bs
, BlockDriverState
*backing_hd
,
2870 bool update_inherits_from
= bdrv_chain_contains(bs
, backing_hd
) &&
2871 bdrv_inherits_from_recursive(backing_hd
, bs
);
2873 if (bdrv_is_backing_chain_frozen(bs
, backing_bs(bs
), errp
)) {
2878 bdrv_ref(backing_hd
);
2882 bdrv_unref_child(bs
, bs
->backing
);
2890 bs
->backing
= bdrv_attach_child(bs
, backing_hd
, "backing", &child_of_bds
,
2891 bdrv_backing_role(bs
), errp
);
2892 /* If backing_hd was already part of bs's backing chain, and
2893 * inherits_from pointed recursively to bs then let's update it to
2894 * point directly to bs (else it will become NULL). */
2895 if (bs
->backing
&& update_inherits_from
) {
2896 backing_hd
->inherits_from
= bs
;
2900 bdrv_refresh_limits(bs
, NULL
);
2904 * Opens the backing file for a BlockDriverState if not yet open
2906 * bdref_key specifies the key for the image's BlockdevRef in the options QDict.
2907 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
2908 * itself, all options starting with "${bdref_key}." are considered part of the
2911 * TODO Can this be unified with bdrv_open_image()?
2913 int bdrv_open_backing_file(BlockDriverState
*bs
, QDict
*parent_options
,
2914 const char *bdref_key
, Error
**errp
)
2916 char *backing_filename
= NULL
;
2917 char *bdref_key_dot
;
2918 const char *reference
= NULL
;
2920 bool implicit_backing
= false;
2921 BlockDriverState
*backing_hd
;
2923 QDict
*tmp_parent_options
= NULL
;
2924 Error
*local_err
= NULL
;
2926 if (bs
->backing
!= NULL
) {
2930 /* NULL means an empty set of options */
2931 if (parent_options
== NULL
) {
2932 tmp_parent_options
= qdict_new();
2933 parent_options
= tmp_parent_options
;
2936 bs
->open_flags
&= ~BDRV_O_NO_BACKING
;
2938 bdref_key_dot
= g_strdup_printf("%s.", bdref_key
);
2939 qdict_extract_subqdict(parent_options
, &options
, bdref_key_dot
);
2940 g_free(bdref_key_dot
);
2943 * Caution: while qdict_get_try_str() is fine, getting non-string
2944 * types would require more care. When @parent_options come from
2945 * -blockdev or blockdev_add, its members are typed according to
2946 * the QAPI schema, but when they come from -drive, they're all
2949 reference
= qdict_get_try_str(parent_options
, bdref_key
);
2950 if (reference
|| qdict_haskey(options
, "file.filename")) {
2951 /* keep backing_filename NULL */
2952 } else if (bs
->backing_file
[0] == '\0' && qdict_size(options
) == 0) {
2953 qobject_unref(options
);
2956 if (qdict_size(options
) == 0) {
2957 /* If the user specifies options that do not modify the
2958 * backing file's behavior, we might still consider it the
2959 * implicit backing file. But it's easier this way, and
2960 * just specifying some of the backing BDS's options is
2961 * only possible with -drive anyway (otherwise the QAPI
2962 * schema forces the user to specify everything). */
2963 implicit_backing
= !strcmp(bs
->auto_backing_file
, bs
->backing_file
);
2966 backing_filename
= bdrv_get_full_backing_filename(bs
, &local_err
);
2969 error_propagate(errp
, local_err
);
2970 qobject_unref(options
);
2975 if (!bs
->drv
|| !bs
->drv
->supports_backing
) {
2977 error_setg(errp
, "Driver doesn't support backing files");
2978 qobject_unref(options
);
2983 bs
->backing_format
[0] != '\0' && !qdict_haskey(options
, "driver")) {
2984 qdict_put_str(options
, "driver", bs
->backing_format
);
2987 backing_hd
= bdrv_open_inherit(backing_filename
, reference
, options
, 0, bs
,
2988 &child_of_bds
, bdrv_backing_role(bs
), errp
);
2990 bs
->open_flags
|= BDRV_O_NO_BACKING
;
2991 error_prepend(errp
, "Could not open backing file: ");
2996 if (implicit_backing
) {
2997 bdrv_refresh_filename(backing_hd
);
2998 pstrcpy(bs
->auto_backing_file
, sizeof(bs
->auto_backing_file
),
2999 backing_hd
->filename
);
3002 /* Hook up the backing file link; drop our reference, bs owns the
3003 * backing_hd reference now */
3004 bdrv_set_backing_hd(bs
, backing_hd
, &local_err
);
3005 bdrv_unref(backing_hd
);
3007 error_propagate(errp
, local_err
);
3012 qdict_del(parent_options
, bdref_key
);
3015 g_free(backing_filename
);
3016 qobject_unref(tmp_parent_options
);
3020 static BlockDriverState
*
3021 bdrv_open_child_bs(const char *filename
, QDict
*options
, const char *bdref_key
,
3022 BlockDriverState
*parent
, const BdrvChildClass
*child_class
,
3023 BdrvChildRole child_role
, bool allow_none
, Error
**errp
)
3025 BlockDriverState
*bs
= NULL
;
3026 QDict
*image_options
;
3027 char *bdref_key_dot
;
3028 const char *reference
;
3030 assert(child_class
!= NULL
);
3032 bdref_key_dot
= g_strdup_printf("%s.", bdref_key
);
3033 qdict_extract_subqdict(options
, &image_options
, bdref_key_dot
);
3034 g_free(bdref_key_dot
);
3037 * Caution: while qdict_get_try_str() is fine, getting non-string
3038 * types would require more care. When @options come from
3039 * -blockdev or blockdev_add, its members are typed according to
3040 * the QAPI schema, but when they come from -drive, they're all
3043 reference
= qdict_get_try_str(options
, bdref_key
);
3044 if (!filename
&& !reference
&& !qdict_size(image_options
)) {
3046 error_setg(errp
, "A block device must be specified for \"%s\"",
3049 qobject_unref(image_options
);
3053 bs
= bdrv_open_inherit(filename
, reference
, image_options
, 0,
3054 parent
, child_class
, child_role
, errp
);
3060 qdict_del(options
, bdref_key
);
3065 * Opens a disk image whose options are given as BlockdevRef in another block
3068 * If allow_none is true, no image will be opened if filename is false and no
3069 * BlockdevRef is given. NULL will be returned, but errp remains unset.
3071 * bdrev_key specifies the key for the image's BlockdevRef in the options QDict.
3072 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
3073 * itself, all options starting with "${bdref_key}." are considered part of the
3076 * The BlockdevRef will be removed from the options QDict.
3078 BdrvChild
*bdrv_open_child(const char *filename
,
3079 QDict
*options
, const char *bdref_key
,
3080 BlockDriverState
*parent
,
3081 const BdrvChildClass
*child_class
,
3082 BdrvChildRole child_role
,
3083 bool allow_none
, Error
**errp
)
3085 BlockDriverState
*bs
;
3087 bs
= bdrv_open_child_bs(filename
, options
, bdref_key
, parent
, child_class
,
3088 child_role
, allow_none
, errp
);
3093 return bdrv_attach_child(parent
, bs
, bdref_key
, child_class
, child_role
,
3098 * TODO Future callers may need to specify parent/child_class in order for
3099 * option inheritance to work. Existing callers use it for the root node.
3101 BlockDriverState
*bdrv_open_blockdev_ref(BlockdevRef
*ref
, Error
**errp
)
3103 BlockDriverState
*bs
= NULL
;
3104 QObject
*obj
= NULL
;
3105 QDict
*qdict
= NULL
;
3106 const char *reference
= NULL
;
3109 if (ref
->type
== QTYPE_QSTRING
) {
3110 reference
= ref
->u
.reference
;
3112 BlockdevOptions
*options
= &ref
->u
.definition
;
3113 assert(ref
->type
== QTYPE_QDICT
);
3115 v
= qobject_output_visitor_new(&obj
);
3116 visit_type_BlockdevOptions(v
, NULL
, &options
, &error_abort
);
3117 visit_complete(v
, &obj
);
3119 qdict
= qobject_to(QDict
, obj
);
3120 qdict_flatten(qdict
);
3122 /* bdrv_open_inherit() defaults to the values in bdrv_flags (for
3123 * compatibility with other callers) rather than what we want as the
3124 * real defaults. Apply the defaults here instead. */
3125 qdict_set_default_str(qdict
, BDRV_OPT_CACHE_DIRECT
, "off");
3126 qdict_set_default_str(qdict
, BDRV_OPT_CACHE_NO_FLUSH
, "off");
3127 qdict_set_default_str(qdict
, BDRV_OPT_READ_ONLY
, "off");
3128 qdict_set_default_str(qdict
, BDRV_OPT_AUTO_READ_ONLY
, "off");
3132 bs
= bdrv_open_inherit(NULL
, reference
, qdict
, 0, NULL
, NULL
, 0, errp
);
3139 static BlockDriverState
*bdrv_append_temp_snapshot(BlockDriverState
*bs
,
3141 QDict
*snapshot_options
,
3144 /* TODO: extra byte is a hack to ensure MAX_PATH space on Windows. */
3145 char *tmp_filename
= g_malloc0(PATH_MAX
+ 1);
3147 QemuOpts
*opts
= NULL
;
3148 BlockDriverState
*bs_snapshot
= NULL
;
3149 Error
*local_err
= NULL
;
3152 /* if snapshot, we create a temporary backing file and open it
3153 instead of opening 'filename' directly */
3155 /* Get the required size from the image */
3156 total_size
= bdrv_getlength(bs
);
3157 if (total_size
< 0) {
3158 error_setg_errno(errp
, -total_size
, "Could not get image size");
3162 /* Create the temporary image */
3163 ret
= get_tmp_filename(tmp_filename
, PATH_MAX
+ 1);
3165 error_setg_errno(errp
, -ret
, "Could not get temporary filename");
3169 opts
= qemu_opts_create(bdrv_qcow2
.create_opts
, NULL
, 0,
3171 qemu_opt_set_number(opts
, BLOCK_OPT_SIZE
, total_size
, &error_abort
);
3172 ret
= bdrv_create(&bdrv_qcow2
, tmp_filename
, opts
, errp
);
3173 qemu_opts_del(opts
);
3175 error_prepend(errp
, "Could not create temporary overlay '%s': ",
3180 /* Prepare options QDict for the temporary file */
3181 qdict_put_str(snapshot_options
, "file.driver", "file");
3182 qdict_put_str(snapshot_options
, "file.filename", tmp_filename
);
3183 qdict_put_str(snapshot_options
, "driver", "qcow2");
3185 bs_snapshot
= bdrv_open(NULL
, NULL
, snapshot_options
, flags
, errp
);
3186 snapshot_options
= NULL
;
3191 /* bdrv_append() consumes a strong reference to bs_snapshot
3192 * (i.e. it will call bdrv_unref() on it) even on error, so in
3193 * order to be able to return one, we have to increase
3194 * bs_snapshot's refcount here */
3195 bdrv_ref(bs_snapshot
);
3196 bdrv_append(bs_snapshot
, bs
, &local_err
);
3198 error_propagate(errp
, local_err
);
3204 qobject_unref(snapshot_options
);
3205 g_free(tmp_filename
);
3210 * Opens a disk image (raw, qcow2, vmdk, ...)
3212 * options is a QDict of options to pass to the block drivers, or NULL for an
3213 * empty set of options. The reference to the QDict belongs to the block layer
3214 * after the call (even on failure), so if the caller intends to reuse the
3215 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
3217 * If *pbs is NULL, a new BDS will be created with a pointer to it stored there.
3218 * If it is not NULL, the referenced BDS will be reused.
3220 * The reference parameter may be used to specify an existing block device which
3221 * should be opened. If specified, neither options nor a filename may be given,
3222 * nor can an existing BDS be reused (that is, *pbs has to be NULL).
3224 static BlockDriverState
*bdrv_open_inherit(const char *filename
,
3225 const char *reference
,
3226 QDict
*options
, int flags
,
3227 BlockDriverState
*parent
,
3228 const BdrvChildClass
*child_class
,
3229 BdrvChildRole child_role
,
3233 BlockBackend
*file
= NULL
;
3234 BlockDriverState
*bs
;
3235 BlockDriver
*drv
= NULL
;
3237 const char *drvname
;
3238 const char *backing
;
3239 Error
*local_err
= NULL
;
3240 QDict
*snapshot_options
= NULL
;
3241 int snapshot_flags
= 0;
3243 assert(!child_class
|| !flags
);
3244 assert(!child_class
== !parent
);
3247 bool options_non_empty
= options ?
qdict_size(options
) : false;
3248 qobject_unref(options
);
3250 if (filename
|| options_non_empty
) {
3251 error_setg(errp
, "Cannot reference an existing block device with "
3252 "additional options or a new filename");
3256 bs
= bdrv_lookup_bs(reference
, reference
, errp
);
3267 /* NULL means an empty set of options */
3268 if (options
== NULL
) {
3269 options
= qdict_new();
3272 /* json: syntax counts as explicit options, as if in the QDict */
3273 parse_json_protocol(options
, &filename
, &local_err
);
3278 bs
->explicit_options
= qdict_clone_shallow(options
);
3281 bool parent_is_format
;
3284 parent_is_format
= parent
->drv
->is_format
;
3287 * parent->drv is not set yet because this node is opened for
3288 * (potential) format probing. That means that @parent is going
3289 * to be a format node.
3291 parent_is_format
= true;
3294 bs
->inherits_from
= parent
;
3295 child_class
->inherit_options(child_role
, parent_is_format
,
3297 parent
->open_flags
, parent
->options
);
3300 ret
= bdrv_fill_options(&options
, filename
, &flags
, &local_err
);
3306 * Set the BDRV_O_RDWR and BDRV_O_ALLOW_RDWR flags.
3307 * Caution: getting a boolean member of @options requires care.
3308 * When @options come from -blockdev or blockdev_add, members are
3309 * typed according to the QAPI schema, but when they come from
3310 * -drive, they're all QString.
3312 if (g_strcmp0(qdict_get_try_str(options
, BDRV_OPT_READ_ONLY
), "on") &&
3313 !qdict_get_try_bool(options
, BDRV_OPT_READ_ONLY
, false)) {
3314 flags
|= (BDRV_O_RDWR
| BDRV_O_ALLOW_RDWR
);
3316 flags
&= ~BDRV_O_RDWR
;
3319 if (flags
& BDRV_O_SNAPSHOT
) {
3320 snapshot_options
= qdict_new();
3321 bdrv_temp_snapshot_options(&snapshot_flags
, snapshot_options
,
3323 /* Let bdrv_backing_options() override "read-only" */
3324 qdict_del(options
, BDRV_OPT_READ_ONLY
);
3325 bdrv_inherited_options(BDRV_CHILD_COW
, true,
3326 &flags
, options
, flags
, options
);
3329 bs
->open_flags
= flags
;
3330 bs
->options
= options
;
3331 options
= qdict_clone_shallow(options
);
3333 /* Find the right image format driver */
3334 /* See cautionary note on accessing @options above */
3335 drvname
= qdict_get_try_str(options
, "driver");
3337 drv
= bdrv_find_format(drvname
);
3339 error_setg(errp
, "Unknown driver: '%s'", drvname
);
3344 assert(drvname
|| !(flags
& BDRV_O_PROTOCOL
));
3346 /* See cautionary note on accessing @options above */
3347 backing
= qdict_get_try_str(options
, "backing");
3348 if (qobject_to(QNull
, qdict_get(options
, "backing")) != NULL
||
3349 (backing
&& *backing
== '\0'))
3352 warn_report("Use of \"backing\": \"\" is deprecated; "
3353 "use \"backing\": null instead");
3355 flags
|= BDRV_O_NO_BACKING
;
3356 qdict_del(bs
->explicit_options
, "backing");
3357 qdict_del(bs
->options
, "backing");
3358 qdict_del(options
, "backing");
3361 /* Open image file without format layer. This BlockBackend is only used for
3362 * probing, the block drivers will do their own bdrv_open_child() for the
3363 * same BDS, which is why we put the node name back into options. */
3364 if ((flags
& BDRV_O_PROTOCOL
) == 0) {
3365 BlockDriverState
*file_bs
;
3367 file_bs
= bdrv_open_child_bs(filename
, options
, "file", bs
,
3368 &child_of_bds
, BDRV_CHILD_IMAGE
,
3373 if (file_bs
!= NULL
) {
3374 /* Not requesting BLK_PERM_CONSISTENT_READ because we're only
3375 * looking at the header to guess the image format. This works even
3376 * in cases where a guest would not see a consistent state. */
3377 file
= blk_new(bdrv_get_aio_context(file_bs
), 0, BLK_PERM_ALL
);
3378 blk_insert_bs(file
, file_bs
, &local_err
);
3379 bdrv_unref(file_bs
);
3384 qdict_put_str(options
, "file", bdrv_get_node_name(file_bs
));
3388 /* Image format probing */
3391 ret
= find_image_format(file
, filename
, &drv
, &local_err
);
3396 * This option update would logically belong in bdrv_fill_options(),
3397 * but we first need to open bs->file for the probing to work, while
3398 * opening bs->file already requires the (mostly) final set of options
3399 * so that cache mode etc. can be inherited.
3401 * Adding the driver later is somewhat ugly, but it's not an option
3402 * that would ever be inherited, so it's correct. We just need to make
3403 * sure to update both bs->options (which has the full effective
3404 * options for bs) and options (which has file.* already removed).
3406 qdict_put_str(bs
->options
, "driver", drv
->format_name
);
3407 qdict_put_str(options
, "driver", drv
->format_name
);
3409 error_setg(errp
, "Must specify either driver or file");
3413 /* BDRV_O_PROTOCOL must be set iff a protocol BDS is about to be created */
3414 assert(!!(flags
& BDRV_O_PROTOCOL
) == !!drv
->bdrv_file_open
);
3415 /* file must be NULL if a protocol BDS is about to be created
3416 * (the inverse results in an error message from bdrv_open_common()) */
3417 assert(!(flags
& BDRV_O_PROTOCOL
) || !file
);
3419 /* Open the image */
3420 ret
= bdrv_open_common(bs
, file
, options
, &local_err
);
3430 /* If there is a backing file, use it */
3431 if ((flags
& BDRV_O_NO_BACKING
) == 0) {
3432 ret
= bdrv_open_backing_file(bs
, options
, "backing", &local_err
);
3434 goto close_and_fail
;
3438 /* Remove all children options and references
3439 * from bs->options and bs->explicit_options */
3440 QLIST_FOREACH(child
, &bs
->children
, next
) {
3441 char *child_key_dot
;
3442 child_key_dot
= g_strdup_printf("%s.", child
->name
);
3443 qdict_extract_subqdict(bs
->explicit_options
, NULL
, child_key_dot
);
3444 qdict_extract_subqdict(bs
->options
, NULL
, child_key_dot
);
3445 qdict_del(bs
->explicit_options
, child
->name
);
3446 qdict_del(bs
->options
, child
->name
);
3447 g_free(child_key_dot
);
3450 /* Check if any unknown options were used */
3451 if (qdict_size(options
) != 0) {
3452 const QDictEntry
*entry
= qdict_first(options
);
3453 if (flags
& BDRV_O_PROTOCOL
) {
3454 error_setg(errp
, "Block protocol '%s' doesn't support the option "
3455 "'%s'", drv
->format_name
, entry
->key
);
3458 "Block format '%s' does not support the option '%s'",
3459 drv
->format_name
, entry
->key
);
3462 goto close_and_fail
;
3465 bdrv_parent_cb_change_media(bs
, true);
3467 qobject_unref(options
);
3470 /* For snapshot=on, create a temporary qcow2 overlay. bs points to the
3471 * temporary snapshot afterwards. */
3472 if (snapshot_flags
) {
3473 BlockDriverState
*snapshot_bs
;
3474 snapshot_bs
= bdrv_append_temp_snapshot(bs
, snapshot_flags
,
3475 snapshot_options
, &local_err
);
3476 snapshot_options
= NULL
;
3478 goto close_and_fail
;
3480 /* We are not going to return bs but the overlay on top of it
3481 * (snapshot_bs); thus, we have to drop the strong reference to bs
3482 * (which we obtained by calling bdrv_new()). bs will not be deleted,
3483 * though, because the overlay still has a reference to it. */
3492 qobject_unref(snapshot_options
);
3493 qobject_unref(bs
->explicit_options
);
3494 qobject_unref(bs
->options
);
3495 qobject_unref(options
);
3497 bs
->explicit_options
= NULL
;
3499 error_propagate(errp
, local_err
);
3504 qobject_unref(snapshot_options
);
3505 qobject_unref(options
);
3506 error_propagate(errp
, local_err
);
3510 BlockDriverState
*bdrv_open(const char *filename
, const char *reference
,
3511 QDict
*options
, int flags
, Error
**errp
)
3513 return bdrv_open_inherit(filename
, reference
, options
, flags
, NULL
,
3517 /* Return true if the NULL-terminated @list contains @str */
3518 static bool is_str_in_list(const char *str
, const char *const *list
)
3522 for (i
= 0; list
[i
] != NULL
; i
++) {
3523 if (!strcmp(str
, list
[i
])) {
3532 * Check that every option set in @bs->options is also set in
3535 * Options listed in the common_options list and in
3536 * @bs->drv->mutable_opts are skipped.
3538 * Return 0 on success, otherwise return -EINVAL and set @errp.
3540 static int bdrv_reset_options_allowed(BlockDriverState
*bs
,
3541 const QDict
*new_opts
, Error
**errp
)
3543 const QDictEntry
*e
;
3544 /* These options are common to all block drivers and are handled
3545 * in bdrv_reopen_prepare() so they can be left out of @new_opts */
3546 const char *const common_options
[] = {
3547 "node-name", "discard", "cache.direct", "cache.no-flush",
3548 "read-only", "auto-read-only", "detect-zeroes", NULL
3551 for (e
= qdict_first(bs
->options
); e
; e
= qdict_next(bs
->options
, e
)) {
3552 if (!qdict_haskey(new_opts
, e
->key
) &&
3553 !is_str_in_list(e
->key
, common_options
) &&
3554 !is_str_in_list(e
->key
, bs
->drv
->mutable_opts
)) {
3555 error_setg(errp
, "Option '%s' cannot be reset "
3556 "to its default value", e
->key
);