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
,
82 /* If non-zero, use only whitelisted block drivers */
83 static int use_bdrv_whitelist
;
86 static int is_windows_drive_prefix(const char *filename
)
88 return (((filename
[0] >= 'a' && filename
[0] <= 'z') ||
89 (filename
[0] >= 'A' && filename
[0] <= 'Z')) &&
93 int is_windows_drive(const char *filename
)
95 if (is_windows_drive_prefix(filename
) &&
98 if (strstart(filename
, "\\\\.\\", NULL
) ||
99 strstart(filename
, "//./", NULL
))
105 size_t bdrv_opt_mem_align(BlockDriverState
*bs
)
107 if (!bs
|| !bs
->drv
) {
108 /* page size or 4k (hdd sector size) should be on the safe side */
109 return MAX(4096, qemu_real_host_page_size
);
112 return bs
->bl
.opt_mem_alignment
;
115 size_t bdrv_min_mem_align(BlockDriverState
*bs
)
117 if (!bs
|| !bs
->drv
) {
118 /* page size or 4k (hdd sector size) should be on the safe side */
119 return MAX(4096, qemu_real_host_page_size
);
122 return bs
->bl
.min_mem_alignment
;
125 /* check if the path starts with "<protocol>:" */
126 int path_has_protocol(const char *path
)
131 if (is_windows_drive(path
) ||
132 is_windows_drive_prefix(path
)) {
135 p
= path
+ strcspn(path
, ":/\\");
137 p
= path
+ strcspn(path
, ":/");
143 int path_is_absolute(const char *path
)
146 /* specific case for names like: "\\.\d:" */
147 if (is_windows_drive(path
) || is_windows_drive_prefix(path
)) {
150 return (*path
== '/' || *path
== '\\');
152 return (*path
== '/');
156 /* if filename is absolute, just return its duplicate. Otherwise, build a
157 path to it by considering it is relative to base_path. URL are
159 char *path_combine(const char *base_path
, const char *filename
)
161 const char *protocol_stripped
= NULL
;
166 if (path_is_absolute(filename
)) {
167 return g_strdup(filename
);
170 if (path_has_protocol(base_path
)) {
171 protocol_stripped
= strchr(base_path
, ':');
172 if (protocol_stripped
) {
176 p
= protocol_stripped
?: base_path
;
178 p1
= strrchr(base_path
, '/');
182 p2
= strrchr(base_path
, '\\');
183 if (!p1
|| p2
> p1
) {
198 result
= g_malloc(len
+ strlen(filename
) + 1);
199 memcpy(result
, base_path
, len
);
200 strcpy(result
+ len
, filename
);
206 * Helper function for bdrv_parse_filename() implementations to remove optional
207 * protocol prefixes (especially "file:") from a filename and for putting the
208 * stripped filename into the options QDict if there is such a prefix.
210 void bdrv_parse_filename_strip_prefix(const char *filename
, const char *prefix
,
213 if (strstart(filename
, prefix
, &filename
)) {
214 /* Stripping the explicit protocol prefix may result in a protocol
215 * prefix being (wrongly) detected (if the filename contains a colon) */
216 if (path_has_protocol(filename
)) {
217 QString
*fat_filename
;
219 /* This means there is some colon before the first slash; therefore,
220 * this cannot be an absolute path */
221 assert(!path_is_absolute(filename
));
223 /* And we can thus fix the protocol detection issue by prefixing it
225 fat_filename
= qstring_from_str("./");
226 qstring_append(fat_filename
, filename
);
228 assert(!path_has_protocol(qstring_get_str(fat_filename
)));
230 qdict_put(options
, "filename", fat_filename
);
232 /* If no protocol prefix was detected, we can use the shortened
234 qdict_put_str(options
, "filename", filename
);
240 /* Returns whether the image file is opened as read-only. Note that this can
241 * return false and writing to the image file is still not possible because the
242 * image is inactivated. */
243 bool bdrv_is_read_only(BlockDriverState
*bs
)
245 return bs
->read_only
;
248 int bdrv_can_set_read_only(BlockDriverState
*bs
, bool read_only
,
249 bool ignore_allow_rdw
, Error
**errp
)
251 /* Do not set read_only if copy_on_read is enabled */
252 if (bs
->copy_on_read
&& read_only
) {
253 error_setg(errp
, "Can't set node '%s' to r/o with copy-on-read enabled",
254 bdrv_get_device_or_node_name(bs
));
258 /* Do not clear read_only if it is prohibited */
259 if (!read_only
&& !(bs
->open_flags
& BDRV_O_ALLOW_RDWR
) &&
262 error_setg(errp
, "Node '%s' is read only",
263 bdrv_get_device_or_node_name(bs
));
271 * Called by a driver that can only provide a read-only image.
273 * Returns 0 if the node is already read-only or it could switch the node to
274 * read-only because BDRV_O_AUTO_RDONLY is set.
276 * Returns -EACCES if the node is read-write and BDRV_O_AUTO_RDONLY is not set
277 * or bdrv_can_set_read_only() forbids making the node read-only. If @errmsg
278 * is not NULL, it is used as the error message for the Error object.
280 int bdrv_apply_auto_read_only(BlockDriverState
*bs
, const char *errmsg
,
285 if (!(bs
->open_flags
& BDRV_O_RDWR
)) {
288 if (!(bs
->open_flags
& BDRV_O_AUTO_RDONLY
)) {
292 ret
= bdrv_can_set_read_only(bs
, true, false, NULL
);
297 bs
->read_only
= true;
298 bs
->open_flags
&= ~BDRV_O_RDWR
;
303 error_setg(errp
, "%s", errmsg
?: "Image is read-only");
308 * If @backing is empty, this function returns NULL without setting
309 * @errp. In all other cases, NULL will only be returned with @errp
312 * Therefore, a return value of NULL without @errp set means that
313 * there is no backing file; if @errp is set, there is one but its
314 * absolute filename cannot be generated.
316 char *bdrv_get_full_backing_filename_from_filename(const char *backed
,
320 if (backing
[0] == '\0') {
322 } else if (path_has_protocol(backing
) || path_is_absolute(backing
)) {
323 return g_strdup(backing
);
324 } else if (backed
[0] == '\0' || strstart(backed
, "json:", NULL
)) {
325 error_setg(errp
, "Cannot use relative backing file names for '%s'",
329 return path_combine(backed
, backing
);
334 * If @filename is empty or NULL, this function returns NULL without
335 * setting @errp. In all other cases, NULL will only be returned with
338 static char *bdrv_make_absolute_filename(BlockDriverState
*relative_to
,
339 const char *filename
, Error
**errp
)
341 char *dir
, *full_name
;
343 if (!filename
|| filename
[0] == '\0') {
345 } else if (path_has_protocol(filename
) || path_is_absolute(filename
)) {
346 return g_strdup(filename
);
349 dir
= bdrv_dirname(relative_to
, errp
);
354 full_name
= g_strconcat(dir
, filename
, NULL
);
359 char *bdrv_get_full_backing_filename(BlockDriverState
*bs
, Error
**errp
)
361 return bdrv_make_absolute_filename(bs
, bs
->backing_file
, errp
);
364 void bdrv_register(BlockDriver
*bdrv
)
366 assert(bdrv
->format_name
);
367 QLIST_INSERT_HEAD(&bdrv_drivers
, bdrv
, list
);
370 BlockDriverState
*bdrv_new(void)
372 BlockDriverState
*bs
;
375 bs
= g_new0(BlockDriverState
, 1);
376 QLIST_INIT(&bs
->dirty_bitmaps
);
377 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
378 QLIST_INIT(&bs
->op_blockers
[i
]);
380 notifier_with_return_list_init(&bs
->before_write_notifiers
);
381 qemu_co_mutex_init(&bs
->reqs_lock
);
382 qemu_mutex_init(&bs
->dirty_bitmap_mutex
);
384 bs
->aio_context
= qemu_get_aio_context();
386 qemu_co_queue_init(&bs
->flush_queue
);
388 for (i
= 0; i
< bdrv_drain_all_count
; i
++) {
389 bdrv_drained_begin(bs
);
392 QTAILQ_INSERT_TAIL(&all_bdrv_states
, bs
, bs_list
);
397 static BlockDriver
*bdrv_do_find_format(const char *format_name
)
401 QLIST_FOREACH(drv1
, &bdrv_drivers
, list
) {
402 if (!strcmp(drv1
->format_name
, format_name
)) {
410 BlockDriver
*bdrv_find_format(const char *format_name
)
415 drv1
= bdrv_do_find_format(format_name
);
420 /* The driver isn't registered, maybe we need to load a module */
421 for (i
= 0; i
< (int)ARRAY_SIZE(block_driver_modules
); ++i
) {
422 if (!strcmp(block_driver_modules
[i
].format_name
, format_name
)) {
423 block_module_load_one(block_driver_modules
[i
].library_name
);
428 return bdrv_do_find_format(format_name
);
431 static int bdrv_format_is_whitelisted(const char *format_name
, bool read_only
)
433 static const char *whitelist_rw
[] = {
434 CONFIG_BDRV_RW_WHITELIST
436 static const char *whitelist_ro
[] = {
437 CONFIG_BDRV_RO_WHITELIST
441 if (!whitelist_rw
[0] && !whitelist_ro
[0]) {
442 return 1; /* no whitelist, anything goes */
445 for (p
= whitelist_rw
; *p
; p
++) {
446 if (!strcmp(format_name
, *p
)) {
451 for (p
= whitelist_ro
; *p
; p
++) {
452 if (!strcmp(format_name
, *p
)) {
460 int bdrv_is_whitelisted(BlockDriver
*drv
, bool read_only
)
462 return bdrv_format_is_whitelisted(drv
->format_name
, read_only
);
465 bool bdrv_uses_whitelist(void)
467 return use_bdrv_whitelist
;
470 typedef struct CreateCo
{
478 static void coroutine_fn
bdrv_create_co_entry(void *opaque
)
480 Error
*local_err
= NULL
;
483 CreateCo
*cco
= opaque
;
486 ret
= cco
->drv
->bdrv_co_create_opts(cco
->drv
,
487 cco
->filename
, cco
->opts
, &local_err
);
488 error_propagate(&cco
->err
, local_err
);
492 int bdrv_create(BlockDriver
*drv
, const char* filename
,
493 QemuOpts
*opts
, Error
**errp
)
500 .filename
= g_strdup(filename
),
506 if (!drv
->bdrv_co_create_opts
) {
507 error_setg(errp
, "Driver '%s' does not support image creation", drv
->format_name
);
512 if (qemu_in_coroutine()) {
513 /* Fast-path if already in coroutine context */
514 bdrv_create_co_entry(&cco
);
516 co
= qemu_coroutine_create(bdrv_create_co_entry
, &cco
);
517 qemu_coroutine_enter(co
);
518 while (cco
.ret
== NOT_DONE
) {
519 aio_poll(qemu_get_aio_context(), true);
526 error_propagate(errp
, cco
.err
);
528 error_setg_errno(errp
, -ret
, "Could not create image");
533 g_free(cco
.filename
);
538 * Helper function for bdrv_create_file_fallback(): Resize @blk to at
539 * least the given @minimum_size.
541 * On success, return @blk's actual length.
542 * Otherwise, return -errno.
544 static int64_t create_file_fallback_truncate(BlockBackend
*blk
,
545 int64_t minimum_size
, Error
**errp
)
547 Error
*local_err
= NULL
;
551 ret
= blk_truncate(blk
, minimum_size
, false, PREALLOC_MODE_OFF
, 0,
553 if (ret
< 0 && ret
!= -ENOTSUP
) {
554 error_propagate(errp
, local_err
);
558 size
= blk_getlength(blk
);
560 error_free(local_err
);
561 error_setg_errno(errp
, -size
,
562 "Failed to inquire the new image file's length");
566 if (size
< minimum_size
) {
567 /* Need to grow the image, but we failed to do that */
568 error_propagate(errp
, local_err
);
572 error_free(local_err
);
579 * Helper function for bdrv_create_file_fallback(): Zero the first
580 * sector to remove any potentially pre-existing image header.
582 static int create_file_fallback_zero_first_sector(BlockBackend
*blk
,
583 int64_t current_size
,
586 int64_t bytes_to_clear
;
589 bytes_to_clear
= MIN(current_size
, BDRV_SECTOR_SIZE
);
590 if (bytes_to_clear
) {
591 ret
= blk_pwrite_zeroes(blk
, 0, bytes_to_clear
, BDRV_REQ_MAY_UNMAP
);
593 error_setg_errno(errp
, -ret
,
594 "Failed to clear the new image's first sector");
603 * Simple implementation of bdrv_co_create_opts for protocol drivers
604 * which only support creation via opening a file
605 * (usually existing raw storage device)
607 int coroutine_fn
bdrv_co_create_opts_simple(BlockDriver
*drv
,
608 const char *filename
,
616 PreallocMode prealloc
;
617 Error
*local_err
= NULL
;
620 size
= qemu_opt_get_size_del(opts
, BLOCK_OPT_SIZE
, 0);
621 buf
= qemu_opt_get_del(opts
, BLOCK_OPT_PREALLOC
);
622 prealloc
= qapi_enum_parse(&PreallocMode_lookup
, buf
,
623 PREALLOC_MODE_OFF
, &local_err
);
626 error_propagate(errp
, local_err
);
630 if (prealloc
!= PREALLOC_MODE_OFF
) {
631 error_setg(errp
, "Unsupported preallocation mode '%s'",
632 PreallocMode_str(prealloc
));
636 options
= qdict_new();
637 qdict_put_str(options
, "driver", drv
->format_name
);
639 blk
= blk_new_open(filename
, NULL
, options
,
640 BDRV_O_RDWR
| BDRV_O_RESIZE
, errp
);
642 error_prepend(errp
, "Protocol driver '%s' does not support image "
643 "creation, and opening the image failed: ",
648 size
= create_file_fallback_truncate(blk
, size
, errp
);
654 ret
= create_file_fallback_zero_first_sector(blk
, size
, errp
);
665 int bdrv_create_file(const char *filename
, QemuOpts
*opts
, Error
**errp
)
669 drv
= bdrv_find_protocol(filename
, true, errp
);
674 return bdrv_create(drv
, filename
, opts
, errp
);
677 int coroutine_fn
bdrv_co_delete_file(BlockDriverState
*bs
, Error
**errp
)
679 Error
*local_err
= NULL
;
685 error_setg(errp
, "Block node '%s' is not opened", bs
->filename
);
689 if (!bs
->drv
->bdrv_co_delete_file
) {
690 error_setg(errp
, "Driver '%s' does not support image deletion",
691 bs
->drv
->format_name
);
695 ret
= bs
->drv
->bdrv_co_delete_file(bs
, &local_err
);
697 error_propagate(errp
, local_err
);
704 * Try to get @bs's logical and physical block size.
705 * On success, store them in @bsz struct and return 0.
706 * On failure return -errno.
707 * @bs must not be empty.
709 int bdrv_probe_blocksizes(BlockDriverState
*bs
, BlockSizes
*bsz
)
711 BlockDriver
*drv
= bs
->drv
;
713 if (drv
&& drv
->bdrv_probe_blocksizes
) {
714 return drv
->bdrv_probe_blocksizes(bs
, bsz
);
715 } else if (drv
&& drv
->is_filter
&& bs
->file
) {
716 return bdrv_probe_blocksizes(bs
->file
->bs
, bsz
);
723 * Try to get @bs's geometry (cyls, heads, sectors).
724 * On success, store them in @geo struct and return 0.
725 * On failure return -errno.
726 * @bs must not be empty.
728 int bdrv_probe_geometry(BlockDriverState
*bs
, HDGeometry
*geo
)
730 BlockDriver
*drv
= bs
->drv
;
732 if (drv
&& drv
->bdrv_probe_geometry
) {
733 return drv
->bdrv_probe_geometry(bs
, geo
);
734 } else if (drv
&& drv
->is_filter
&& bs
->file
) {
735 return bdrv_probe_geometry(bs
->file
->bs
, geo
);
742 * Create a uniquely-named empty temporary file.
743 * Return 0 upon success, otherwise a negative errno value.
745 int get_tmp_filename(char *filename
, int size
)
748 char temp_dir
[MAX_PATH
];
749 /* GetTempFileName requires that its output buffer (4th param)
750 have length MAX_PATH or greater. */
751 assert(size
>= MAX_PATH
);
752 return (GetTempPath(MAX_PATH
, temp_dir
)
753 && GetTempFileName(temp_dir
, "qem", 0, filename
)
754 ? 0 : -GetLastError());
758 tmpdir
= getenv("TMPDIR");
762 if (snprintf(filename
, size
, "%s/vl.XXXXXX", tmpdir
) >= size
) {
765 fd
= mkstemp(filename
);
769 if (close(fd
) != 0) {
778 * Detect host devices. By convention, /dev/cdrom[N] is always
779 * recognized as a host CDROM.
781 static BlockDriver
*find_hdev_driver(const char *filename
)
783 int score_max
= 0, score
;
784 BlockDriver
*drv
= NULL
, *d
;
786 QLIST_FOREACH(d
, &bdrv_drivers
, list
) {
787 if (d
->bdrv_probe_device
) {
788 score
= d
->bdrv_probe_device(filename
);
789 if (score
> score_max
) {
799 static BlockDriver
*bdrv_do_find_protocol(const char *protocol
)
803 QLIST_FOREACH(drv1
, &bdrv_drivers
, list
) {
804 if (drv1
->protocol_name
&& !strcmp(drv1
->protocol_name
, protocol
)) {
812 BlockDriver
*bdrv_find_protocol(const char *filename
,
813 bool allow_protocol_prefix
,
822 /* TODO Drivers without bdrv_file_open must be specified explicitly */
825 * XXX(hch): we really should not let host device detection
826 * override an explicit protocol specification, but moving this
827 * later breaks access to device names with colons in them.
828 * Thanks to the brain-dead persistent naming schemes on udev-
829 * based Linux systems those actually are quite common.
831 drv1
= find_hdev_driver(filename
);
836 if (!path_has_protocol(filename
) || !allow_protocol_prefix
) {
840 p
= strchr(filename
, ':');
843 if (len
> sizeof(protocol
) - 1)
844 len
= sizeof(protocol
) - 1;
845 memcpy(protocol
, filename
, len
);
846 protocol
[len
] = '\0';
848 drv1
= bdrv_do_find_protocol(protocol
);
853 for (i
= 0; i
< (int)ARRAY_SIZE(block_driver_modules
); ++i
) {
854 if (block_driver_modules
[i
].protocol_name
&&
855 !strcmp(block_driver_modules
[i
].protocol_name
, protocol
)) {
856 block_module_load_one(block_driver_modules
[i
].library_name
);
861 drv1
= bdrv_do_find_protocol(protocol
);
863 error_setg(errp
, "Unknown protocol '%s'", protocol
);
869 * Guess image format by probing its contents.
870 * This is not a good idea when your image is raw (CVE-2008-2004), but
871 * we do it anyway for backward compatibility.
873 * @buf contains the image's first @buf_size bytes.
874 * @buf_size is the buffer size in bytes (generally BLOCK_PROBE_BUF_SIZE,
875 * but can be smaller if the image file is smaller)
876 * @filename is its filename.
878 * For all block drivers, call the bdrv_probe() method to get its
880 * Return the first block driver with the highest probing score.
882 BlockDriver
*bdrv_probe_all(const uint8_t *buf
, int buf_size
,
883 const char *filename
)
885 int score_max
= 0, score
;
886 BlockDriver
*drv
= NULL
, *d
;
888 QLIST_FOREACH(d
, &bdrv_drivers
, list
) {
890 score
= d
->bdrv_probe(buf
, buf_size
, filename
);
891 if (score
> score_max
) {
901 static int find_image_format(BlockBackend
*file
, const char *filename
,
902 BlockDriver
**pdrv
, Error
**errp
)
905 uint8_t buf
[BLOCK_PROBE_BUF_SIZE
];
908 /* Return the raw BlockDriver * to scsi-generic devices or empty drives */
909 if (blk_is_sg(file
) || !blk_is_inserted(file
) || blk_getlength(file
) == 0) {
914 ret
= blk_pread(file
, 0, buf
, sizeof(buf
));
916 error_setg_errno(errp
, -ret
, "Could not read image for determining its "
922 drv
= bdrv_probe_all(buf
, ret
, filename
);
924 error_setg(errp
, "Could not determine image format: No compatible "
933 * Set the current 'total_sectors' value
934 * Return 0 on success, -errno on error.
936 int refresh_total_sectors(BlockDriverState
*bs
, int64_t hint
)
938 BlockDriver
*drv
= bs
->drv
;
944 /* Do not attempt drv->bdrv_getlength() on scsi-generic devices */
948 /* query actual device if possible, otherwise just trust the hint */
949 if (drv
->bdrv_getlength
) {
950 int64_t length
= drv
->bdrv_getlength(bs
);
954 hint
= DIV_ROUND_UP(length
, BDRV_SECTOR_SIZE
);
957 bs
->total_sectors
= hint
;
962 * Combines a QDict of new block driver @options with any missing options taken
963 * from @old_options, so that leaving out an option defaults to its old value.
965 static void bdrv_join_options(BlockDriverState
*bs
, QDict
*options
,
968 if (bs
->drv
&& bs
->drv
->bdrv_join_options
) {
969 bs
->drv
->bdrv_join_options(options
, old_options
);
971 qdict_join(options
, old_options
, false);
975 static BlockdevDetectZeroesOptions
bdrv_parse_detect_zeroes(QemuOpts
*opts
,
979 Error
*local_err
= NULL
;
980 char *value
= qemu_opt_get_del(opts
, "detect-zeroes");
981 BlockdevDetectZeroesOptions detect_zeroes
=
982 qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup
, value
,
983 BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF
, &local_err
);
986 error_propagate(errp
, local_err
);
987 return detect_zeroes
;
990 if (detect_zeroes
== BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP
&&
991 !(open_flags
& BDRV_O_UNMAP
))
993 error_setg(errp
, "setting detect-zeroes to unmap is not allowed "
994 "without setting discard operation to unmap");
997 return detect_zeroes
;
1001 * Set open flags for aio engine
1003 * Return 0 on success, -1 if the engine specified is invalid
1005 int bdrv_parse_aio(const char *mode
, int *flags
)
1007 if (!strcmp(mode
, "threads")) {
1008 /* do nothing, default */
1009 } else if (!strcmp(mode
, "native")) {
1010 *flags
|= BDRV_O_NATIVE_AIO
;
1011 #ifdef CONFIG_LINUX_IO_URING
1012 } else if (!strcmp(mode
, "io_uring")) {
1013 *flags
|= BDRV_O_IO_URING
;
1023 * Set open flags for a given discard mode
1025 * Return 0 on success, -1 if the discard mode was invalid.
1027 int bdrv_parse_discard_flags(const char *mode
, int *flags
)
1029 *flags
&= ~BDRV_O_UNMAP
;
1031 if (!strcmp(mode
, "off") || !strcmp(mode
, "ignore")) {
1033 } else if (!strcmp(mode
, "on") || !strcmp(mode
, "unmap")) {
1034 *flags
|= BDRV_O_UNMAP
;
1043 * Set open flags for a given cache mode
1045 * Return 0 on success, -1 if the cache mode was invalid.
1047 int bdrv_parse_cache_mode(const char *mode
, int *flags
, bool *writethrough
)
1049 *flags
&= ~BDRV_O_CACHE_MASK
;
1051 if (!strcmp(mode
, "off") || !strcmp(mode
, "none")) {
1052 *writethrough
= false;
1053 *flags
|= BDRV_O_NOCACHE
;
1054 } else if (!strcmp(mode
, "directsync")) {
1055 *writethrough
= true;
1056 *flags
|= BDRV_O_NOCACHE
;
1057 } else if (!strcmp(mode
, "writeback")) {
1058 *writethrough
= false;
1059 } else if (!strcmp(mode
, "unsafe")) {
1060 *writethrough
= false;
1061 *flags
|= BDRV_O_NO_FLUSH
;
1062 } else if (!strcmp(mode
, "writethrough")) {
1063 *writethrough
= true;
1071 static char *bdrv_child_get_parent_desc(BdrvChild
*c
)
1073 BlockDriverState
*parent
= c
->opaque
;
1074 return g_strdup(bdrv_get_device_or_node_name(parent
));
1077 static void bdrv_child_cb_drained_begin(BdrvChild
*child
)
1079 BlockDriverState
*bs
= child
->opaque
;
1080 bdrv_do_drained_begin_quiesce(bs
, NULL
, false);
1083 static bool bdrv_child_cb_drained_poll(BdrvChild
*child
)
1085 BlockDriverState
*bs
= child
->opaque
;
1086 return bdrv_drain_poll(bs
, false, NULL
, false);
1089 static void bdrv_child_cb_drained_end(BdrvChild
*child
,
1090 int *drained_end_counter
)
1092 BlockDriverState
*bs
= child
->opaque
;
1093 bdrv_drained_end_no_poll(bs
, drained_end_counter
);
1096 static void bdrv_child_cb_attach(BdrvChild
*child
)
1098 BlockDriverState
*bs
= child
->opaque
;
1099 bdrv_apply_subtree_drain(child
, bs
);
1102 static void bdrv_child_cb_detach(BdrvChild
*child
)
1104 BlockDriverState
*bs
= child
->opaque
;
1105 bdrv_unapply_subtree_drain(child
, bs
);
1108 static int bdrv_child_cb_inactivate(BdrvChild
*child
)
1110 BlockDriverState
*bs
= child
->opaque
;
1111 assert(bs
->open_flags
& BDRV_O_INACTIVE
);
1115 static bool bdrv_child_cb_can_set_aio_ctx(BdrvChild
*child
, AioContext
*ctx
,
1116 GSList
**ignore
, Error
**errp
)
1118 BlockDriverState
*bs
= child
->opaque
;
1119 return bdrv_can_set_aio_context(bs
, ctx
, ignore
, errp
);
1122 static void bdrv_child_cb_set_aio_ctx(BdrvChild
*child
, AioContext
*ctx
,
1125 BlockDriverState
*bs
= child
->opaque
;
1126 return bdrv_set_aio_context_ignore(bs
, ctx
, ignore
);
1130 * Returns the options and flags that a temporary snapshot should get, based on
1131 * the originally requested flags (the originally requested image will have
1132 * flags like a backing file)
1134 static void bdrv_temp_snapshot_options(int *child_flags
, QDict
*child_options
,
1135 int parent_flags
, QDict
*parent_options
)
1137 *child_flags
= (parent_flags
& ~BDRV_O_SNAPSHOT
) | BDRV_O_TEMPORARY
;
1139 /* For temporary files, unconditional cache=unsafe is fine */
1140 qdict_set_default_str(child_options
, BDRV_OPT_CACHE_DIRECT
, "off");
1141 qdict_set_default_str(child_options
, BDRV_OPT_CACHE_NO_FLUSH
, "on");
1143 /* Copy the read-only and discard options from the parent */
1144 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_READ_ONLY
);
1145 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_DISCARD
);
1147 /* aio=native doesn't work for cache.direct=off, so disable it for the
1148 * temporary snapshot */
1149 *child_flags
&= ~BDRV_O_NATIVE_AIO
;
1153 * Returns the options and flags that bs->file should get if a protocol driver
1154 * is expected, based on the given options and flags for the parent BDS
1156 static void bdrv_inherited_options(int *child_flags
, QDict
*child_options
,
1157 int parent_flags
, QDict
*parent_options
)
1159 int flags
= parent_flags
;
1161 /* Enable protocol handling, disable format probing for bs->file */
1162 flags
|= BDRV_O_PROTOCOL
;
1164 /* If the cache mode isn't explicitly set, inherit direct and no-flush from
1166 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_CACHE_DIRECT
);
1167 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_CACHE_NO_FLUSH
);
1168 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_FORCE_SHARE
);
1170 /* Inherit the read-only option from the parent if it's not set */
1171 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_READ_ONLY
);
1172 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_AUTO_READ_ONLY
);
1174 /* Our block drivers take care to send flushes and respect unmap policy,
1175 * so we can default to enable both on lower layers regardless of the
1176 * corresponding parent options. */
1177 qdict_set_default_str(child_options
, BDRV_OPT_DISCARD
, "unmap");
1179 /* Clear flags that only apply to the top layer */
1180 flags
&= ~(BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
| BDRV_O_COPY_ON_READ
|
1183 *child_flags
= flags
;
1186 const BdrvChildClass child_file
= {
1187 .parent_is_bds
= true,
1188 .get_parent_desc
= bdrv_child_get_parent_desc
,
1189 .inherit_options
= bdrv_inherited_options
,
1190 .drained_begin
= bdrv_child_cb_drained_begin
,
1191 .drained_poll
= bdrv_child_cb_drained_poll
,
1192 .drained_end
= bdrv_child_cb_drained_end
,
1193 .attach
= bdrv_child_cb_attach
,
1194 .detach
= bdrv_child_cb_detach
,
1195 .inactivate
= bdrv_child_cb_inactivate
,
1196 .can_set_aio_ctx
= bdrv_child_cb_can_set_aio_ctx
,
1197 .set_aio_ctx
= bdrv_child_cb_set_aio_ctx
,
1201 * Returns the options and flags that bs->file should get if the use of formats
1202 * (and not only protocols) is permitted for it, based on the given options and
1203 * flags for the parent BDS
1205 static void bdrv_inherited_fmt_options(int *child_flags
, QDict
*child_options
,
1206 int parent_flags
, QDict
*parent_options
)
1208 child_file
.inherit_options(child_flags
, child_options
,
1209 parent_flags
, parent_options
);
1211 *child_flags
&= ~(BDRV_O_PROTOCOL
| BDRV_O_NO_IO
);
1214 const BdrvChildClass child_format
= {
1215 .parent_is_bds
= true,
1216 .get_parent_desc
= bdrv_child_get_parent_desc
,
1217 .inherit_options
= bdrv_inherited_fmt_options
,
1218 .drained_begin
= bdrv_child_cb_drained_begin
,
1219 .drained_poll
= bdrv_child_cb_drained_poll
,
1220 .drained_end
= bdrv_child_cb_drained_end
,
1221 .attach
= bdrv_child_cb_attach
,
1222 .detach
= bdrv_child_cb_detach
,
1223 .inactivate
= bdrv_child_cb_inactivate
,
1224 .can_set_aio_ctx
= bdrv_child_cb_can_set_aio_ctx
,
1225 .set_aio_ctx
= bdrv_child_cb_set_aio_ctx
,
1228 static void bdrv_backing_attach(BdrvChild
*c
)
1230 BlockDriverState
*parent
= c
->opaque
;
1231 BlockDriverState
*backing_hd
= c
->bs
;
1233 assert(!parent
->backing_blocker
);
1234 error_setg(&parent
->backing_blocker
,
1235 "node is used as backing hd of '%s'",
1236 bdrv_get_device_or_node_name(parent
));
1238 bdrv_refresh_filename(backing_hd
);
1240 parent
->open_flags
&= ~BDRV_O_NO_BACKING
;
1241 pstrcpy(parent
->backing_file
, sizeof(parent
->backing_file
),
1242 backing_hd
->filename
);
1243 pstrcpy(parent
->backing_format
, sizeof(parent
->backing_format
),
1244 backing_hd
->drv
? backing_hd
->drv
->format_name
: "");
1246 bdrv_op_block_all(backing_hd
, parent
->backing_blocker
);
1247 /* Otherwise we won't be able to commit or stream */
1248 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_COMMIT_TARGET
,
1249 parent
->backing_blocker
);
1250 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_STREAM
,
1251 parent
->backing_blocker
);
1253 * We do backup in 3 ways:
1255 * The target bs is new opened, and the source is top BDS
1256 * 2. blockdev backup
1257 * Both the source and the target are top BDSes.
1258 * 3. internal backup(used for block replication)
1259 * Both the source and the target are backing file
1261 * In case 1 and 2, neither the source nor the target is the backing file.
1262 * In case 3, we will block the top BDS, so there is only one block job
1263 * for the top BDS and its backing chain.
1265 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_BACKUP_SOURCE
,
1266 parent
->backing_blocker
);
1267 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_BACKUP_TARGET
,
1268 parent
->backing_blocker
);
1270 bdrv_child_cb_attach(c
);
1273 static void bdrv_backing_detach(BdrvChild
*c
)
1275 BlockDriverState
*parent
= c
->opaque
;
1277 assert(parent
->backing_blocker
);
1278 bdrv_op_unblock_all(c
->bs
, parent
->backing_blocker
);
1279 error_free(parent
->backing_blocker
);
1280 parent
->backing_blocker
= NULL
;
1282 bdrv_child_cb_detach(c
);
1286 * Returns the options and flags that bs->backing should get, based on the
1287 * given options and flags for the parent BDS
1289 static void bdrv_backing_options(int *child_flags
, QDict
*child_options
,
1290 int parent_flags
, QDict
*parent_options
)
1292 int flags
= parent_flags
;
1294 /* The cache mode is inherited unmodified for backing files; except WCE,
1295 * which is only applied on the top level (BlockBackend) */
1296 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_CACHE_DIRECT
);
1297 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_CACHE_NO_FLUSH
);
1298 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_FORCE_SHARE
);
1300 /* backing files always opened read-only */
1301 qdict_set_default_str(child_options
, BDRV_OPT_READ_ONLY
, "on");
1302 qdict_set_default_str(child_options
, BDRV_OPT_AUTO_READ_ONLY
, "off");
1303 flags
&= ~BDRV_O_COPY_ON_READ
;
1305 /* snapshot=on is handled on the top layer */
1306 flags
&= ~(BDRV_O_SNAPSHOT
| BDRV_O_TEMPORARY
);
1308 *child_flags
= flags
;
1311 static int bdrv_backing_update_filename(BdrvChild
*c
, BlockDriverState
*base
,
1312 const char *filename
, Error
**errp
)
1314 BlockDriverState
*parent
= c
->opaque
;
1315 bool read_only
= bdrv_is_read_only(parent
);
1319 ret
= bdrv_reopen_set_read_only(parent
, false, errp
);
1325 ret
= bdrv_change_backing_file(parent
, filename
,
1326 base
->drv
? base
->drv
->format_name
: "");
1328 error_setg_errno(errp
, -ret
, "Could not update backing file link");
1332 bdrv_reopen_set_read_only(parent
, true, NULL
);
1338 const BdrvChildClass child_backing
= {
1339 .parent_is_bds
= true,
1340 .get_parent_desc
= bdrv_child_get_parent_desc
,
1341 .attach
= bdrv_backing_attach
,
1342 .detach
= bdrv_backing_detach
,
1343 .inherit_options
= bdrv_backing_options
,
1344 .drained_begin
= bdrv_child_cb_drained_begin
,
1345 .drained_poll
= bdrv_child_cb_drained_poll
,
1346 .drained_end
= bdrv_child_cb_drained_end
,
1347 .inactivate
= bdrv_child_cb_inactivate
,
1348 .update_filename
= bdrv_backing_update_filename
,
1349 .can_set_aio_ctx
= bdrv_child_cb_can_set_aio_ctx
,
1350 .set_aio_ctx
= bdrv_child_cb_set_aio_ctx
,
1353 static int bdrv_open_flags(BlockDriverState
*bs
, int flags
)
1355 int open_flags
= flags
;
1358 * Clear flags that are internal to the block layer before opening the
1361 open_flags
&= ~(BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
| BDRV_O_PROTOCOL
);
1366 static void update_flags_from_options(int *flags
, QemuOpts
*opts
)
1368 *flags
&= ~(BDRV_O_CACHE_MASK
| BDRV_O_RDWR
| BDRV_O_AUTO_RDONLY
);
1370 if (qemu_opt_get_bool_del(opts
, BDRV_OPT_CACHE_NO_FLUSH
, false)) {
1371 *flags
|= BDRV_O_NO_FLUSH
;
1374 if (qemu_opt_get_bool_del(opts
, BDRV_OPT_CACHE_DIRECT
, false)) {
1375 *flags
|= BDRV_O_NOCACHE
;
1378 if (!qemu_opt_get_bool_del(opts
, BDRV_OPT_READ_ONLY
, false)) {
1379 *flags
|= BDRV_O_RDWR
;
1382 if (qemu_opt_get_bool_del(opts
, BDRV_OPT_AUTO_READ_ONLY
, false)) {
1383 *flags
|= BDRV_O_AUTO_RDONLY
;
1387 static void update_options_from_flags(QDict
*options
, int flags
)
1389 if (!qdict_haskey(options
, BDRV_OPT_CACHE_DIRECT
)) {
1390 qdict_put_bool(options
, BDRV_OPT_CACHE_DIRECT
, flags
& BDRV_O_NOCACHE
);
1392 if (!qdict_haskey(options
, BDRV_OPT_CACHE_NO_FLUSH
)) {
1393 qdict_put_bool(options
, BDRV_OPT_CACHE_NO_FLUSH
,
1394 flags
& BDRV_O_NO_FLUSH
);
1396 if (!qdict_haskey(options
, BDRV_OPT_READ_ONLY
)) {
1397 qdict_put_bool(options
, BDRV_OPT_READ_ONLY
, !(flags
& BDRV_O_RDWR
));
1399 if (!qdict_haskey(options
, BDRV_OPT_AUTO_READ_ONLY
)) {
1400 qdict_put_bool(options
, BDRV_OPT_AUTO_READ_ONLY
,
1401 flags
& BDRV_O_AUTO_RDONLY
);
1405 static void bdrv_assign_node_name(BlockDriverState
*bs
,
1406 const char *node_name
,
1409 char *gen_node_name
= NULL
;
1412 node_name
= gen_node_name
= id_generate(ID_BLOCK
);
1413 } else if (!id_wellformed(node_name
)) {
1415 * Check for empty string or invalid characters, but not if it is
1416 * generated (generated names use characters not available to the user)
1418 error_setg(errp
, "Invalid node name");
1422 /* takes care of avoiding namespaces collisions */
1423 if (blk_by_name(node_name
)) {
1424 error_setg(errp
, "node-name=%s is conflicting with a device id",
1429 /* takes care of avoiding duplicates node names */
1430 if (bdrv_find_node(node_name
)) {
1431 error_setg(errp
, "Duplicate node name");
1435 /* Make sure that the node name isn't truncated */
1436 if (strlen(node_name
) >= sizeof(bs
->node_name
)) {
1437 error_setg(errp
, "Node name too long");
1441 /* copy node name into the bs and insert it into the graph list */
1442 pstrcpy(bs
->node_name
, sizeof(bs
->node_name
), node_name
);
1443 QTAILQ_INSERT_TAIL(&graph_bdrv_states
, bs
, node_list
);
1445 g_free(gen_node_name
);
1448 static int bdrv_open_driver(BlockDriverState
*bs
, BlockDriver
*drv
,
1449 const char *node_name
, QDict
*options
,
1450 int open_flags
, Error
**errp
)
1452 Error
*local_err
= NULL
;
1455 bdrv_assign_node_name(bs
, node_name
, &local_err
);
1457 error_propagate(errp
, local_err
);
1462 bs
->read_only
= !(bs
->open_flags
& BDRV_O_RDWR
);
1463 bs
->opaque
= g_malloc0(drv
->instance_size
);
1465 if (drv
->bdrv_file_open
) {
1466 assert(!drv
->bdrv_needs_filename
|| bs
->filename
[0]);
1467 ret
= drv
->bdrv_file_open(bs
, options
, open_flags
, &local_err
);
1468 } else if (drv
->bdrv_open
) {
1469 ret
= drv
->bdrv_open(bs
, options
, open_flags
, &local_err
);
1476 error_propagate(errp
, local_err
);
1477 } else if (bs
->filename
[0]) {
1478 error_setg_errno(errp
, -ret
, "Could not open '%s'", bs
->filename
);
1480 error_setg_errno(errp
, -ret
, "Could not open image");
1485 ret
= refresh_total_sectors(bs
, bs
->total_sectors
);
1487 error_setg_errno(errp
, -ret
, "Could not refresh total sector count");
1491 bdrv_refresh_limits(bs
, &local_err
);
1493 error_propagate(errp
, local_err
);
1497 assert(bdrv_opt_mem_align(bs
) != 0);
1498 assert(bdrv_min_mem_align(bs
) != 0);
1499 assert(is_power_of_2(bs
->bl
.request_alignment
));
1501 for (i
= 0; i
< bs
->quiesce_counter
; i
++) {
1502 if (drv
->bdrv_co_drain_begin
) {
1503 drv
->bdrv_co_drain_begin(bs
);
1510 if (bs
->file
!= NULL
) {
1511 bdrv_unref_child(bs
, bs
->file
);
1519 BlockDriverState
*bdrv_new_open_driver(BlockDriver
*drv
, const char *node_name
,
1520 int flags
, Error
**errp
)
1522 BlockDriverState
*bs
;
1526 bs
->open_flags
= flags
;
1527 bs
->explicit_options
= qdict_new();
1528 bs
->options
= qdict_new();
1531 update_options_from_flags(bs
->options
, flags
);
1533 ret
= bdrv_open_driver(bs
, drv
, node_name
, bs
->options
, flags
, errp
);
1535 qobject_unref(bs
->explicit_options
);
1536 bs
->explicit_options
= NULL
;
1537 qobject_unref(bs
->options
);
1546 QemuOptsList bdrv_runtime_opts
= {
1547 .name
= "bdrv_common",
1548 .head
= QTAILQ_HEAD_INITIALIZER(bdrv_runtime_opts
.head
),
1551 .name
= "node-name",
1552 .type
= QEMU_OPT_STRING
,
1553 .help
= "Node name of the block device node",
1557 .type
= QEMU_OPT_STRING
,
1558 .help
= "Block driver to use for the node",
1561 .name
= BDRV_OPT_CACHE_DIRECT
,
1562 .type
= QEMU_OPT_BOOL
,
1563 .help
= "Bypass software writeback cache on the host",
1566 .name
= BDRV_OPT_CACHE_NO_FLUSH
,
1567 .type
= QEMU_OPT_BOOL
,
1568 .help
= "Ignore flush requests",
1571 .name
= BDRV_OPT_READ_ONLY
,
1572 .type
= QEMU_OPT_BOOL
,
1573 .help
= "Node is opened in read-only mode",
1576 .name
= BDRV_OPT_AUTO_READ_ONLY
,
1577 .type
= QEMU_OPT_BOOL
,
1578 .help
= "Node can become read-only if opening read-write fails",
1581 .name
= "detect-zeroes",
1582 .type
= QEMU_OPT_STRING
,
1583 .help
= "try to optimize zero writes (off, on, unmap)",
1586 .name
= BDRV_OPT_DISCARD
,
1587 .type
= QEMU_OPT_STRING
,
1588 .help
= "discard operation (ignore/off, unmap/on)",
1591 .name
= BDRV_OPT_FORCE_SHARE
,
1592 .type
= QEMU_OPT_BOOL
,
1593 .help
= "always accept other writers (default: off)",
1595 { /* end of list */ }
1599 QemuOptsList bdrv_create_opts_simple
= {
1600 .name
= "simple-create-opts",
1601 .head
= QTAILQ_HEAD_INITIALIZER(bdrv_create_opts_simple
.head
),
1604 .name
= BLOCK_OPT_SIZE
,
1605 .type
= QEMU_OPT_SIZE
,
1606 .help
= "Virtual disk size"
1609 .name
= BLOCK_OPT_PREALLOC
,
1610 .type
= QEMU_OPT_STRING
,
1611 .help
= "Preallocation mode (allowed values: off)"
1613 { /* end of list */ }
1618 * Common part for opening disk images and files
1620 * Removes all processed options from *options.
1622 static int bdrv_open_common(BlockDriverState
*bs
, BlockBackend
*file
,
1623 QDict
*options
, Error
**errp
)
1625 int ret
, open_flags
;
1626 const char *filename
;
1627 const char *driver_name
= NULL
;
1628 const char *node_name
= NULL
;
1629 const char *discard
;
1632 Error
*local_err
= NULL
;
1634 assert(bs
->file
== NULL
);
1635 assert(options
!= NULL
&& bs
->options
!= options
);
1637 opts
= qemu_opts_create(&bdrv_runtime_opts
, NULL
, 0, &error_abort
);
1638 qemu_opts_absorb_qdict(opts
, options
, &local_err
);
1640 error_propagate(errp
, local_err
);
1645 update_flags_from_options(&bs
->open_flags
, opts
);
1647 driver_name
= qemu_opt_get(opts
, "driver");
1648 drv
= bdrv_find_format(driver_name
);
1649 assert(drv
!= NULL
);
1651 bs
->force_share
= qemu_opt_get_bool(opts
, BDRV_OPT_FORCE_SHARE
, false);
1653 if (bs
->force_share
&& (bs
->open_flags
& BDRV_O_RDWR
)) {
1655 BDRV_OPT_FORCE_SHARE
1656 "=on can only be used with read-only images");
1662 bdrv_refresh_filename(blk_bs(file
));
1663 filename
= blk_bs(file
)->filename
;
1666 * Caution: while qdict_get_try_str() is fine, getting
1667 * non-string types would require more care. When @options
1668 * come from -blockdev or blockdev_add, its members are typed
1669 * according to the QAPI schema, but when they come from
1670 * -drive, they're all QString.
1672 filename
= qdict_get_try_str(options
, "filename");
1675 if (drv
->bdrv_needs_filename
&& (!filename
|| !filename
[0])) {
1676 error_setg(errp
, "The '%s' block driver requires a file name",
1682 trace_bdrv_open_common(bs
, filename
?: "", bs
->open_flags
,
1685 bs
->read_only
= !(bs
->open_flags
& BDRV_O_RDWR
);
1687 if (use_bdrv_whitelist
&& !bdrv_is_whitelisted(drv
, bs
->read_only
)) {
1688 if (!bs
->read_only
&& bdrv_is_whitelisted(drv
, true)) {
1689 ret
= bdrv_apply_auto_read_only(bs
, NULL
, NULL
);
1695 !bs
->read_only
&& bdrv_is_whitelisted(drv
, true)
1696 ? "Driver '%s' can only be used for read-only devices"
1697 : "Driver '%s' is not whitelisted",
1703 /* bdrv_new() and bdrv_close() make it so */
1704 assert(atomic_read(&bs
->copy_on_read
) == 0);
1706 if (bs
->open_flags
& BDRV_O_COPY_ON_READ
) {
1707 if (!bs
->read_only
) {
1708 bdrv_enable_copy_on_read(bs
);
1710 error_setg(errp
, "Can't use copy-on-read on read-only device");
1716 discard
= qemu_opt_get(opts
, BDRV_OPT_DISCARD
);
1717 if (discard
!= NULL
) {
1718 if (bdrv_parse_discard_flags(discard
, &bs
->open_flags
) != 0) {
1719 error_setg(errp
, "Invalid discard option");
1726 bdrv_parse_detect_zeroes(opts
, bs
->open_flags
, &local_err
);
1728 error_propagate(errp
, local_err
);
1733 if (filename
!= NULL
) {
1734 pstrcpy(bs
->filename
, sizeof(bs
->filename
), filename
);
1736 bs
->filename
[0] = '\0';
1738 pstrcpy(bs
->exact_filename
, sizeof(bs
->exact_filename
), bs
->filename
);
1740 /* Open the image, either directly or using a protocol */
1741 open_flags
= bdrv_open_flags(bs
, bs
->open_flags
);
1742 node_name
= qemu_opt_get(opts
, "node-name");
1744 assert(!drv
->bdrv_file_open
|| file
== NULL
);
1745 ret
= bdrv_open_driver(bs
, drv
, node_name
, options
, open_flags
, errp
);
1750 qemu_opts_del(opts
);
1754 qemu_opts_del(opts
);
1758 static QDict
*parse_json_filename(const char *filename
, Error
**errp
)
1760 QObject
*options_obj
;
1764 ret
= strstart(filename
, "json:", &filename
);
1767 options_obj
= qobject_from_json(filename
, errp
);
1769 error_prepend(errp
, "Could not parse the JSON options: ");
1773 options
= qobject_to(QDict
, options_obj
);
1775 qobject_unref(options_obj
);
1776 error_setg(errp
, "Invalid JSON object given");
1780 qdict_flatten(options
);
1785 static void parse_json_protocol(QDict
*options
, const char **pfilename
,
1788 QDict
*json_options
;
1789 Error
*local_err
= NULL
;
1791 /* Parse json: pseudo-protocol */
1792 if (!*pfilename
|| !g_str_has_prefix(*pfilename
, "json:")) {
1796 json_options
= parse_json_filename(*pfilename
, &local_err
);
1798 error_propagate(errp
, local_err
);
1802 /* Options given in the filename have lower priority than options
1803 * specified directly */
1804 qdict_join(options
, json_options
, false);
1805 qobject_unref(json_options
);
1810 * Fills in default options for opening images and converts the legacy
1811 * filename/flags pair to option QDict entries.
1812 * The BDRV_O_PROTOCOL flag in *flags will be set or cleared accordingly if a
1813 * block driver has been specified explicitly.
1815 static int bdrv_fill_options(QDict
**options
, const char *filename
,
1816 int *flags
, Error
**errp
)
1818 const char *drvname
;
1819 bool protocol
= *flags
& BDRV_O_PROTOCOL
;
1820 bool parse_filename
= false;
1821 BlockDriver
*drv
= NULL
;
1822 Error
*local_err
= NULL
;
1825 * Caution: while qdict_get_try_str() is fine, getting non-string
1826 * types would require more care. When @options come from
1827 * -blockdev or blockdev_add, its members are typed according to
1828 * the QAPI schema, but when they come from -drive, they're all
1831 drvname
= qdict_get_try_str(*options
, "driver");
1833 drv
= bdrv_find_format(drvname
);
1835 error_setg(errp
, "Unknown driver '%s'", drvname
);
1838 /* If the user has explicitly specified the driver, this choice should
1839 * override the BDRV_O_PROTOCOL flag */
1840 protocol
= drv
->bdrv_file_open
;
1844 *flags
|= BDRV_O_PROTOCOL
;
1846 *flags
&= ~BDRV_O_PROTOCOL
;
1849 /* Translate cache options from flags into options */
1850 update_options_from_flags(*options
, *flags
);
1852 /* Fetch the file name from the options QDict if necessary */
1853 if (protocol
&& filename
) {
1854 if (!qdict_haskey(*options
, "filename")) {
1855 qdict_put_str(*options
, "filename", filename
);
1856 parse_filename
= true;
1858 error_setg(errp
, "Can't specify 'file' and 'filename' options at "
1864 /* Find the right block driver */
1865 /* See cautionary note on accessing @options above */
1866 filename
= qdict_get_try_str(*options
, "filename");
1868 if (!drvname
&& protocol
) {
1870 drv
= bdrv_find_protocol(filename
, parse_filename
, errp
);
1875 drvname
= drv
->format_name
;
1876 qdict_put_str(*options
, "driver", drvname
);
1878 error_setg(errp
, "Must specify either driver or file");
1883 assert(drv
|| !protocol
);
1885 /* Driver-specific filename parsing */
1886 if (drv
&& drv
->bdrv_parse_filename
&& parse_filename
) {
1887 drv
->bdrv_parse_filename(filename
, *options
, &local_err
);
1889 error_propagate(errp
, local_err
);
1893 if (!drv
->bdrv_needs_filename
) {
1894 qdict_del(*options
, "filename");
1901 static int bdrv_child_check_perm(BdrvChild
*c
, BlockReopenQueue
*q
,
1902 uint64_t perm
, uint64_t shared
,
1903 GSList
*ignore_children
,
1904 bool *tighten_restrictions
, Error
**errp
);
1905 static void bdrv_child_abort_perm_update(BdrvChild
*c
);
1906 static void bdrv_child_set_perm(BdrvChild
*c
, uint64_t perm
, uint64_t shared
);
1908 typedef struct BlockReopenQueueEntry
{
1911 BDRVReopenState state
;
1912 QTAILQ_ENTRY(BlockReopenQueueEntry
) entry
;
1913 } BlockReopenQueueEntry
;
1916 * Return the flags that @bs will have after the reopens in @q have
1917 * successfully completed. If @q is NULL (or @bs is not contained in @q),
1918 * return the current flags.
1920 static int bdrv_reopen_get_flags(BlockReopenQueue
*q
, BlockDriverState
*bs
)
1922 BlockReopenQueueEntry
*entry
;
1925 QTAILQ_FOREACH(entry
, q
, entry
) {
1926 if (entry
->state
.bs
== bs
) {
1927 return entry
->state
.flags
;
1932 return bs
->open_flags
;
1935 /* Returns whether the image file can be written to after the reopen queue @q
1936 * has been successfully applied, or right now if @q is NULL. */
1937 static bool bdrv_is_writable_after_reopen(BlockDriverState
*bs
,
1938 BlockReopenQueue
*q
)
1940 int flags
= bdrv_reopen_get_flags(q
, bs
);
1942 return (flags
& (BDRV_O_RDWR
| BDRV_O_INACTIVE
)) == BDRV_O_RDWR
;
1946 * Return whether the BDS can be written to. This is not necessarily
1947 * the same as !bdrv_is_read_only(bs), as inactivated images may not
1948 * be written to but do not count as read-only images.
1950 bool bdrv_is_writable(BlockDriverState
*bs
)
1952 return bdrv_is_writable_after_reopen(bs
, NULL
);
1955 static void bdrv_child_perm(BlockDriverState
*bs
, BlockDriverState
*child_bs
,
1956 BdrvChild
*c
, const BdrvChildClass
*child_class
,
1957 BlockReopenQueue
*reopen_queue
,
1958 uint64_t parent_perm
, uint64_t parent_shared
,
1959 uint64_t *nperm
, uint64_t *nshared
)
1961 assert(bs
->drv
&& bs
->drv
->bdrv_child_perm
);
1962 bs
->drv
->bdrv_child_perm(bs
, c
, child_class
, reopen_queue
,
1963 parent_perm
, parent_shared
,
1965 /* TODO Take force_share from reopen_queue */
1966 if (child_bs
&& child_bs
->force_share
) {
1967 *nshared
= BLK_PERM_ALL
;
1972 * Check whether permissions on this node can be changed in a way that
1973 * @cumulative_perms and @cumulative_shared_perms are the new cumulative
1974 * permissions of all its parents. This involves checking whether all necessary
1975 * permission changes to child nodes can be performed.
1977 * Will set *tighten_restrictions to true if and only if new permissions have to
1978 * be taken or currently shared permissions are to be unshared. Otherwise,
1979 * errors are not fatal as long as the caller accepts that the restrictions
1980 * remain tighter than they need to be. The caller still has to abort the
1982 * @tighten_restrictions cannot be used together with @q: When reopening, we may
1983 * encounter fatal errors even though no restrictions are to be tightened. For
1984 * example, changing a node from RW to RO will fail if the WRITE permission is
1987 * A call to this function must always be followed by a call to bdrv_set_perm()
1988 * or bdrv_abort_perm_update().
1990 static int bdrv_check_perm(BlockDriverState
*bs
, BlockReopenQueue
*q
,
1991 uint64_t cumulative_perms
,
1992 uint64_t cumulative_shared_perms
,
1993 GSList
*ignore_children
,
1994 bool *tighten_restrictions
, Error
**errp
)
1996 BlockDriver
*drv
= bs
->drv
;
2000 assert(!q
|| !tighten_restrictions
);
2002 if (tighten_restrictions
) {
2003 uint64_t current_perms
, current_shared
;
2004 uint64_t added_perms
, removed_shared_perms
;
2006 bdrv_get_cumulative_perm(bs
, ¤t_perms
, ¤t_shared
);
2008 added_perms
= cumulative_perms
& ~current_perms
;
2009 removed_shared_perms
= current_shared
& ~cumulative_shared_perms
;
2011 *tighten_restrictions
= added_perms
|| removed_shared_perms
;
2014 /* Write permissions never work with read-only images */
2015 if ((cumulative_perms
& (BLK_PERM_WRITE
| BLK_PERM_WRITE_UNCHANGED
)) &&
2016 !bdrv_is_writable_after_reopen(bs
, q
))
2018 if (!bdrv_is_writable_after_reopen(bs
, NULL
)) {
2019 error_setg(errp
, "Block node is read-only");
2021 uint64_t current_perms
, current_shared
;
2022 bdrv_get_cumulative_perm(bs
, ¤t_perms
, ¤t_shared
);
2023 if (current_perms
& (BLK_PERM_WRITE
| BLK_PERM_WRITE_UNCHANGED
)) {
2024 error_setg(errp
, "Cannot make block node read-only, there is "
2027 error_setg(errp
, "Cannot make block node read-only and create "
2035 /* Check this node */
2040 if (drv
->bdrv_check_perm
) {
2041 return drv
->bdrv_check_perm(bs
, cumulative_perms
,
2042 cumulative_shared_perms
, errp
);
2045 /* Drivers that never have children can omit .bdrv_child_perm() */
2046 if (!drv
->bdrv_child_perm
) {
2047 assert(QLIST_EMPTY(&bs
->children
));
2051 /* Check all children */
2052 QLIST_FOREACH(c
, &bs
->children
, next
) {
2053 uint64_t cur_perm
, cur_shared
;
2054 bool child_tighten_restr
;
2056 bdrv_child_perm(bs
, c
->bs
, c
, c
->klass
, q
,
2057 cumulative_perms
, cumulative_shared_perms
,
2058 &cur_perm
, &cur_shared
);
2059 ret
= bdrv_child_check_perm(c
, q
, cur_perm
, cur_shared
, ignore_children
,
2060 tighten_restrictions
? &child_tighten_restr
2063 if (tighten_restrictions
) {
2064 *tighten_restrictions
|= child_tighten_restr
;
2075 * Notifies drivers that after a previous bdrv_check_perm() call, the
2076 * permission update is not performed and any preparations made for it (e.g.
2077 * taken file locks) need to be undone.
2079 * This function recursively notifies all child nodes.
2081 static void bdrv_abort_perm_update(BlockDriverState
*bs
)
2083 BlockDriver
*drv
= bs
->drv
;
2090 if (drv
->bdrv_abort_perm_update
) {
2091 drv
->bdrv_abort_perm_update(bs
);
2094 QLIST_FOREACH(c
, &bs
->children
, next
) {
2095 bdrv_child_abort_perm_update(c
);
2099 static void bdrv_set_perm(BlockDriverState
*bs
, uint64_t cumulative_perms
,
2100 uint64_t cumulative_shared_perms
)
2102 BlockDriver
*drv
= bs
->drv
;
2109 /* Update this node */
2110 if (drv
->bdrv_set_perm
) {
2111 drv
->bdrv_set_perm(bs
, cumulative_perms
, cumulative_shared_perms
);
2114 /* Drivers that never have children can omit .bdrv_child_perm() */
2115 if (!drv
->bdrv_child_perm
) {
2116 assert(QLIST_EMPTY(&bs
->children
));
2120 /* Update all children */
2121 QLIST_FOREACH(c
, &bs
->children
, next
) {
2122 uint64_t cur_perm
, cur_shared
;
2123 bdrv_child_perm(bs
, c
->bs
, c
, c
->klass
, NULL
,
2124 cumulative_perms
, cumulative_shared_perms
,
2125 &cur_perm
, &cur_shared
);
2126 bdrv_child_set_perm(c
, cur_perm
, cur_shared
);
2130 void bdrv_get_cumulative_perm(BlockDriverState
*bs
, uint64_t *perm
,
2131 uint64_t *shared_perm
)
2134 uint64_t cumulative_perms
= 0;
2135 uint64_t cumulative_shared_perms
= BLK_PERM_ALL
;
2137 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
2138 cumulative_perms
|= c
->perm
;
2139 cumulative_shared_perms
&= c
->shared_perm
;
2142 *perm
= cumulative_perms
;
2143 *shared_perm
= cumulative_shared_perms
;
2146 static char *bdrv_child_user_desc(BdrvChild
*c
)
2148 if (c
->klass
->get_parent_desc
) {
2149 return c
->klass
->get_parent_desc(c
);
2152 return g_strdup("another user");
2155 char *bdrv_perm_names(uint64_t perm
)
2161 { BLK_PERM_CONSISTENT_READ
, "consistent read" },
2162 { BLK_PERM_WRITE
, "write" },
2163 { BLK_PERM_WRITE_UNCHANGED
, "write unchanged" },
2164 { BLK_PERM_RESIZE
, "resize" },
2165 { BLK_PERM_GRAPH_MOD
, "change children" },
2169 GString
*result
= g_string_sized_new(30);
2170 struct perm_name
*p
;
2172 for (p
= permissions
; p
->name
; p
++) {
2173 if (perm
& p
->perm
) {
2174 if (result
->len
> 0) {
2175 g_string_append(result
, ", ");
2177 g_string_append(result
, p
->name
);
2181 return g_string_free(result
, FALSE
);
2185 * Checks whether a new reference to @bs can be added if the new user requires
2186 * @new_used_perm/@new_shared_perm as its permissions. If @ignore_children is
2187 * set, the BdrvChild objects in this list are ignored in the calculations;
2188 * this allows checking permission updates for an existing reference.
2190 * See bdrv_check_perm() for the semantics of @tighten_restrictions.
2192 * Needs to be followed by a call to either bdrv_set_perm() or
2193 * bdrv_abort_perm_update(). */
2194 static int bdrv_check_update_perm(BlockDriverState
*bs
, BlockReopenQueue
*q
,
2195 uint64_t new_used_perm
,
2196 uint64_t new_shared_perm
,
2197 GSList
*ignore_children
,
2198 bool *tighten_restrictions
,
2202 uint64_t cumulative_perms
= new_used_perm
;
2203 uint64_t cumulative_shared_perms
= new_shared_perm
;
2205 assert(!q
|| !tighten_restrictions
);
2207 /* There is no reason why anyone couldn't tolerate write_unchanged */
2208 assert(new_shared_perm
& BLK_PERM_WRITE_UNCHANGED
);
2210 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
2211 if (g_slist_find(ignore_children
, c
)) {
2215 if ((new_used_perm
& c
->shared_perm
) != new_used_perm
) {
2216 char *user
= bdrv_child_user_desc(c
);
2217 char *perm_names
= bdrv_perm_names(new_used_perm
& ~c
->shared_perm
);
2219 if (tighten_restrictions
) {
2220 *tighten_restrictions
= true;
2223 error_setg(errp
, "Conflicts with use by %s as '%s', which does not "
2225 user
, c
->name
, perm_names
, bdrv_get_node_name(c
->bs
));
2231 if ((c
->perm
& new_shared_perm
) != c
->perm
) {
2232 char *user
= bdrv_child_user_desc(c
);
2233 char *perm_names
= bdrv_perm_names(c
->perm
& ~new_shared_perm
);
2235 if (tighten_restrictions
) {
2236 *tighten_restrictions
= true;
2239 error_setg(errp
, "Conflicts with use by %s as '%s', which uses "
2241 user
, c
->name
, perm_names
, bdrv_get_node_name(c
->bs
));
2247 cumulative_perms
|= c
->perm
;
2248 cumulative_shared_perms
&= c
->shared_perm
;
2251 return bdrv_check_perm(bs
, q
, cumulative_perms
, cumulative_shared_perms
,
2252 ignore_children
, tighten_restrictions
, errp
);
2255 /* Needs to be followed by a call to either bdrv_child_set_perm() or
2256 * bdrv_child_abort_perm_update(). */
2257 static int bdrv_child_check_perm(BdrvChild
*c
, BlockReopenQueue
*q
,
2258 uint64_t perm
, uint64_t shared
,
2259 GSList
*ignore_children
,
2260 bool *tighten_restrictions
, Error
**errp
)
2264 ignore_children
= g_slist_prepend(g_slist_copy(ignore_children
), c
);
2265 ret
= bdrv_check_update_perm(c
->bs
, q
, perm
, shared
, ignore_children
,
2266 tighten_restrictions
, errp
);
2267 g_slist_free(ignore_children
);
2273 if (!c
->has_backup_perm
) {
2274 c
->has_backup_perm
= true;
2275 c
->backup_perm
= c
->perm
;
2276 c
->backup_shared_perm
= c
->shared_perm
;
2279 * Note: it's OK if c->has_backup_perm was already set, as we can find the
2280 * same child twice during check_perm procedure
2284 c
->shared_perm
= shared
;
2289 static void bdrv_child_set_perm(BdrvChild
*c
, uint64_t perm
, uint64_t shared
)
2291 uint64_t cumulative_perms
, cumulative_shared_perms
;
2293 c
->has_backup_perm
= false;
2296 c
->shared_perm
= shared
;
2298 bdrv_get_cumulative_perm(c
->bs
, &cumulative_perms
,
2299 &cumulative_shared_perms
);
2300 bdrv_set_perm(c
->bs
, cumulative_perms
, cumulative_shared_perms
);
2303 static void bdrv_child_abort_perm_update(BdrvChild
*c
)
2305 if (c
->has_backup_perm
) {
2306 c
->perm
= c
->backup_perm
;
2307 c
->shared_perm
= c
->backup_shared_perm
;
2308 c
->has_backup_perm
= false;
2311 bdrv_abort_perm_update(c
->bs
);
2314 int bdrv_child_try_set_perm(BdrvChild
*c
, uint64_t perm
, uint64_t shared
,
2317 Error
*local_err
= NULL
;
2319 bool tighten_restrictions
;
2321 ret
= bdrv_child_check_perm(c
, NULL
, perm
, shared
, NULL
,
2322 &tighten_restrictions
, &local_err
);
2324 bdrv_child_abort_perm_update(c
);
2325 if (tighten_restrictions
) {
2326 error_propagate(errp
, local_err
);
2329 * Our caller may intend to only loosen restrictions and
2330 * does not expect this function to fail. Errors are not
2331 * fatal in such a case, so we can just hide them from our
2334 error_free(local_err
);
2340 bdrv_child_set_perm(c
, perm
, shared
);
2345 int bdrv_child_refresh_perms(BlockDriverState
*bs
, BdrvChild
*c
, Error
**errp
)
2347 uint64_t parent_perms
, parent_shared
;
2348 uint64_t perms
, shared
;
2350 bdrv_get_cumulative_perm(bs
, &parent_perms
, &parent_shared
);
2351 bdrv_child_perm(bs
, c
->bs
, c
, c
->klass
, NULL
, parent_perms
, parent_shared
,
2354 return bdrv_child_try_set_perm(c
, perms
, shared
, errp
);
2357 void bdrv_filter_default_perms(BlockDriverState
*bs
, BdrvChild
*c
,
2358 const BdrvChildClass
*child_class
,
2359 BlockReopenQueue
*reopen_queue
,
2360 uint64_t perm
, uint64_t shared
,
2361 uint64_t *nperm
, uint64_t *nshared
)
2363 *nperm
= perm
& DEFAULT_PERM_PASSTHROUGH
;
2364 *nshared
= (shared
& DEFAULT_PERM_PASSTHROUGH
) | DEFAULT_PERM_UNCHANGED
;
2367 void bdrv_format_default_perms(BlockDriverState
*bs
, BdrvChild
*c
,
2368 const BdrvChildClass
*child_class
,
2369 BlockReopenQueue
*reopen_queue
,
2370 uint64_t perm
, uint64_t shared
,
2371 uint64_t *nperm
, uint64_t *nshared
)
2373 bool backing
= (child_class
== &child_backing
);
2374 assert(child_class
== &child_backing
|| child_class
== &child_file
);
2377 int flags
= bdrv_reopen_get_flags(reopen_queue
, bs
);
2379 /* Apart from the modifications below, the same permissions are
2380 * forwarded and left alone as for filters */
2381 bdrv_filter_default_perms(bs
, c
, child_class
, reopen_queue
,
2382 perm
, shared
, &perm
, &shared
);
2384 /* Format drivers may touch metadata even if the guest doesn't write */
2385 if (bdrv_is_writable_after_reopen(bs
, reopen_queue
)) {
2386 perm
|= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2389 /* bs->file always needs to be consistent because of the metadata. We
2390 * can never allow other users to resize or write to it. */
2391 if (!(flags
& BDRV_O_NO_IO
)) {
2392 perm
|= BLK_PERM_CONSISTENT_READ
;
2394 shared
&= ~(BLK_PERM_WRITE
| BLK_PERM_RESIZE
);
2396 /* We want consistent read from backing files if the parent needs it.
2397 * No other operations are performed on backing files. */
2398 perm
&= BLK_PERM_CONSISTENT_READ
;
2400 /* If the parent can deal with changing data, we're okay with a
2401 * writable and resizable backing file. */
2402 /* TODO Require !(perm & BLK_PERM_CONSISTENT_READ), too? */
2403 if (shared
& BLK_PERM_WRITE
) {
2404 shared
= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2409 shared
|= BLK_PERM_CONSISTENT_READ
| BLK_PERM_GRAPH_MOD
|
2410 BLK_PERM_WRITE_UNCHANGED
;
2413 if (bs
->open_flags
& BDRV_O_INACTIVE
) {
2414 shared
|= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2421 uint64_t bdrv_qapi_perm_to_blk_perm(BlockPermission qapi_perm
)
2423 static const uint64_t permissions
[] = {
2424 [BLOCK_PERMISSION_CONSISTENT_READ
] = BLK_PERM_CONSISTENT_READ
,
2425 [BLOCK_PERMISSION_WRITE
] = BLK_PERM_WRITE
,
2426 [BLOCK_PERMISSION_WRITE_UNCHANGED
] = BLK_PERM_WRITE_UNCHANGED
,
2427 [BLOCK_PERMISSION_RESIZE
] = BLK_PERM_RESIZE
,
2428 [BLOCK_PERMISSION_GRAPH_MOD
] = BLK_PERM_GRAPH_MOD
,
2431 QEMU_BUILD_BUG_ON(ARRAY_SIZE(permissions
) != BLOCK_PERMISSION__MAX
);
2432 QEMU_BUILD_BUG_ON(1UL << ARRAY_SIZE(permissions
) != BLK_PERM_ALL
+ 1);
2434 assert(qapi_perm
< BLOCK_PERMISSION__MAX
);
2436 return permissions
[qapi_perm
];
2439 static void bdrv_replace_child_noperm(BdrvChild
*child
,
2440 BlockDriverState
*new_bs
)
2442 BlockDriverState
*old_bs
= child
->bs
;
2443 int new_bs_quiesce_counter
;
2446 assert(!child
->frozen
);
2448 if (old_bs
&& new_bs
) {
2449 assert(bdrv_get_aio_context(old_bs
) == bdrv_get_aio_context(new_bs
));
2452 new_bs_quiesce_counter
= (new_bs
? new_bs
->quiesce_counter
: 0);
2453 drain_saldo
= new_bs_quiesce_counter
- child
->parent_quiesce_counter
;
2456 * If the new child node is drained but the old one was not, flush
2457 * all outstanding requests to the old child node.
2459 while (drain_saldo
> 0 && child
->klass
->drained_begin
) {
2460 bdrv_parent_drained_begin_single(child
, true);
2465 /* Detach first so that the recursive drain sections coming from @child
2466 * are already gone and we only end the drain sections that came from
2468 if (child
->klass
->detach
) {
2469 child
->klass
->detach(child
);
2471 QLIST_REMOVE(child
, next_parent
);
2477 QLIST_INSERT_HEAD(&new_bs
->parents
, child
, next_parent
);
2480 * Detaching the old node may have led to the new node's
2481 * quiesce_counter having been decreased. Not a problem, we
2482 * just need to recognize this here and then invoke
2483 * drained_end appropriately more often.
2485 assert(new_bs
->quiesce_counter
<= new_bs_quiesce_counter
);
2486 drain_saldo
+= new_bs
->quiesce_counter
- new_bs_quiesce_counter
;
2488 /* Attach only after starting new drained sections, so that recursive
2489 * drain sections coming from @child don't get an extra .drained_begin
2491 if (child
->klass
->attach
) {
2492 child
->klass
->attach(child
);
2497 * If the old child node was drained but the new one is not, allow
2498 * requests to come in only after the new node has been attached.
2500 while (drain_saldo
< 0 && child
->klass
->drained_end
) {
2501 bdrv_parent_drained_end_single(child
);
2507 * Updates @child to change its reference to point to @new_bs, including
2508 * checking and applying the necessary permisson updates both to the old node
2511 * NULL is passed as @new_bs for removing the reference before freeing @child.
2513 * If @new_bs is not NULL, bdrv_check_perm() must be called beforehand, as this
2514 * function uses bdrv_set_perm() to update the permissions according to the new
2515 * reference that @new_bs gets.
2517 static void bdrv_replace_child(BdrvChild
*child
, BlockDriverState
*new_bs
)
2519 BlockDriverState
*old_bs
= child
->bs
;
2520 uint64_t perm
, shared_perm
;
2522 bdrv_replace_child_noperm(child
, new_bs
);
2525 * Start with the new node's permissions. If @new_bs is a (direct
2526 * or indirect) child of @old_bs, we must complete the permission
2527 * update on @new_bs before we loosen the restrictions on @old_bs.
2528 * Otherwise, bdrv_check_perm() on @old_bs would re-initiate
2529 * updating the permissions of @new_bs, and thus not purely loosen
2533 bdrv_get_cumulative_perm(new_bs
, &perm
, &shared_perm
);
2534 bdrv_set_perm(new_bs
, perm
, shared_perm
);
2538 /* Update permissions for old node. This is guaranteed to succeed
2539 * because we're just taking a parent away, so we're loosening
2541 bool tighten_restrictions
;
2544 bdrv_get_cumulative_perm(old_bs
, &perm
, &shared_perm
);
2545 ret
= bdrv_check_perm(old_bs
, NULL
, perm
, shared_perm
, NULL
,
2546 &tighten_restrictions
, NULL
);
2547 assert(tighten_restrictions
== false);
2549 /* We only tried to loosen restrictions, so errors are not fatal */
2550 bdrv_abort_perm_update(old_bs
);
2552 bdrv_set_perm(old_bs
, perm
, shared_perm
);
2555 /* When the parent requiring a non-default AioContext is removed, the
2556 * node moves back to the main AioContext */
2557 bdrv_try_set_aio_context(old_bs
, qemu_get_aio_context(), NULL
);
2562 * This function steals the reference to child_bs from the caller.
2563 * That reference is later dropped by bdrv_root_unref_child().
2565 * On failure NULL is returned, errp is set and the reference to
2566 * child_bs is also dropped.
2568 * The caller must hold the AioContext lock @child_bs, but not that of @ctx
2569 * (unless @child_bs is already in @ctx).
2571 BdrvChild
*bdrv_root_attach_child(BlockDriverState
*child_bs
,
2572 const char *child_name
,
2573 const BdrvChildClass
*child_class
,
2574 BdrvChildRole child_role
,
2576 uint64_t perm
, uint64_t shared_perm
,
2577 void *opaque
, Error
**errp
)
2580 Error
*local_err
= NULL
;
2583 ret
= bdrv_check_update_perm(child_bs
, NULL
, perm
, shared_perm
, NULL
, NULL
,
2586 bdrv_abort_perm_update(child_bs
);
2587 bdrv_unref(child_bs
);
2591 child
= g_new(BdrvChild
, 1);
2592 *child
= (BdrvChild
) {
2594 .name
= g_strdup(child_name
),
2595 .klass
= child_class
,
2598 .shared_perm
= shared_perm
,
2602 /* If the AioContexts don't match, first try to move the subtree of
2603 * child_bs into the AioContext of the new parent. If this doesn't work,
2604 * try moving the parent into the AioContext of child_bs instead. */
2605 if (bdrv_get_aio_context(child_bs
) != ctx
) {
2606 ret
= bdrv_try_set_aio_context(child_bs
, ctx
, &local_err
);
2607 if (ret
< 0 && child_class
->can_set_aio_ctx
) {
2608 GSList
*ignore
= g_slist_prepend(NULL
, child
);
2609 ctx
= bdrv_get_aio_context(child_bs
);
2610 if (child_class
->can_set_aio_ctx(child
, ctx
, &ignore
, NULL
)) {
2611 error_free(local_err
);
2613 g_slist_free(ignore
);
2614 ignore
= g_slist_prepend(NULL
, child
);
2615 child_class
->set_aio_ctx(child
, ctx
, &ignore
);
2617 g_slist_free(ignore
);
2620 error_propagate(errp
, local_err
);
2622 bdrv_abort_perm_update(child_bs
);
2623 bdrv_unref(child_bs
);
2628 /* This performs the matching bdrv_set_perm() for the above check. */
2629 bdrv_replace_child(child
, child_bs
);
2635 * This function transfers the reference to child_bs from the caller
2636 * to parent_bs. That reference is later dropped by parent_bs on
2637 * bdrv_close() or if someone calls bdrv_unref_child().
2639 * On failure NULL is returned, errp is set and the reference to
2640 * child_bs is also dropped.
2642 * If @parent_bs and @child_bs are in different AioContexts, the caller must
2643 * hold the AioContext lock for @child_bs, but not for @parent_bs.
2645 BdrvChild
*bdrv_attach_child(BlockDriverState
*parent_bs
,
2646 BlockDriverState
*child_bs
,
2647 const char *child_name
,
2648 const BdrvChildClass
*child_class
,
2649 BdrvChildRole child_role
,
2653 uint64_t perm
, shared_perm
;
2655 bdrv_get_cumulative_perm(parent_bs
, &perm
, &shared_perm
);
2657 assert(parent_bs
->drv
);
2658 bdrv_child_perm(parent_bs
, child_bs
, NULL
, child_class
, NULL
,
2659 perm
, shared_perm
, &perm
, &shared_perm
);
2661 child
= bdrv_root_attach_child(child_bs
, child_name
, child_class
,
2662 child_role
, bdrv_get_aio_context(parent_bs
),
2663 perm
, shared_perm
, parent_bs
, errp
);
2664 if (child
== NULL
) {
2668 QLIST_INSERT_HEAD(&parent_bs
->children
, child
, next
);
2672 static void bdrv_detach_child(BdrvChild
*child
)
2674 QLIST_SAFE_REMOVE(child
, next
);
2676 bdrv_replace_child(child
, NULL
);
2678 g_free(child
->name
);
2682 void bdrv_root_unref_child(BdrvChild
*child
)
2684 BlockDriverState
*child_bs
;
2686 child_bs
= child
->bs
;
2687 bdrv_detach_child(child
);
2688 bdrv_unref(child_bs
);
2692 * Clear all inherits_from pointers from children and grandchildren of
2693 * @root that point to @root, where necessary.
2695 static void bdrv_unset_inherits_from(BlockDriverState
*root
, BdrvChild
*child
)
2699 if (child
->bs
->inherits_from
== root
) {
2701 * Remove inherits_from only when the last reference between root and
2702 * child->bs goes away.
2704 QLIST_FOREACH(c
, &root
->children
, next
) {
2705 if (c
!= child
&& c
->bs
== child
->bs
) {
2710 child
->bs
->inherits_from
= NULL
;
2714 QLIST_FOREACH(c
, &child
->bs
->children
, next
) {
2715 bdrv_unset_inherits_from(root
, c
);
2719 void bdrv_unref_child(BlockDriverState
*parent
, BdrvChild
*child
)
2721 if (child
== NULL
) {
2725 bdrv_unset_inherits_from(parent
, child
);
2726 bdrv_root_unref_child(child
);
2730 static void bdrv_parent_cb_change_media(BlockDriverState
*bs
, bool load
)
2733 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
2734 if (c
->klass
->change_media
) {
2735 c
->klass
->change_media(c
, load
);
2740 /* Return true if you can reach parent going through child->inherits_from
2741 * recursively. If parent or child are NULL, return false */
2742 static bool bdrv_inherits_from_recursive(BlockDriverState
*child
,
2743 BlockDriverState
*parent
)
2745 while (child
&& child
!= parent
) {
2746 child
= child
->inherits_from
;
2749 return child
!= NULL
;
2753 * Sets the backing file link of a BDS. A new reference is created; callers
2754 * which don't need their own reference any more must call bdrv_unref().
2756 void bdrv_set_backing_hd(BlockDriverState
*bs
, BlockDriverState
*backing_hd
,
2759 bool update_inherits_from
= bdrv_chain_contains(bs
, backing_hd
) &&
2760 bdrv_inherits_from_recursive(backing_hd
, bs
);
2762 if (bdrv_is_backing_chain_frozen(bs
, backing_bs(bs
), errp
)) {
2767 bdrv_ref(backing_hd
);
2771 bdrv_unref_child(bs
, bs
->backing
);
2779 bs
->backing
= bdrv_attach_child(bs
, backing_hd
, "backing", &child_backing
,
2781 /* If backing_hd was already part of bs's backing chain, and
2782 * inherits_from pointed recursively to bs then let's update it to
2783 * point directly to bs (else it will become NULL). */
2784 if (bs
->backing
&& update_inherits_from
) {
2785 backing_hd
->inherits_from
= bs
;
2789 bdrv_refresh_limits(bs
, NULL
);
2793 * Opens the backing file for a BlockDriverState if not yet open
2795 * bdref_key specifies the key for the image's BlockdevRef in the options QDict.
2796 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
2797 * itself, all options starting with "${bdref_key}." are considered part of the
2800 * TODO Can this be unified with bdrv_open_image()?
2802 int bdrv_open_backing_file(BlockDriverState
*bs
, QDict
*parent_options
,
2803 const char *bdref_key
, Error
**errp
)
2805 char *backing_filename
= NULL
;
2806 char *bdref_key_dot
;
2807 const char *reference
= NULL
;
2809 bool implicit_backing
= false;
2810 BlockDriverState
*backing_hd
;
2812 QDict
*tmp_parent_options
= NULL
;
2813 Error
*local_err
= NULL
;
2815 if (bs
->backing
!= NULL
) {
2819 /* NULL means an empty set of options */
2820 if (parent_options
== NULL
) {
2821 tmp_parent_options
= qdict_new();
2822 parent_options
= tmp_parent_options
;
2825 bs
->open_flags
&= ~BDRV_O_NO_BACKING
;
2827 bdref_key_dot
= g_strdup_printf("%s.", bdref_key
);
2828 qdict_extract_subqdict(parent_options
, &options
, bdref_key_dot
);
2829 g_free(bdref_key_dot
);
2832 * Caution: while qdict_get_try_str() is fine, getting non-string
2833 * types would require more care. When @parent_options come from
2834 * -blockdev or blockdev_add, its members are typed according to
2835 * the QAPI schema, but when they come from -drive, they're all
2838 reference
= qdict_get_try_str(parent_options
, bdref_key
);
2839 if (reference
|| qdict_haskey(options
, "file.filename")) {
2840 /* keep backing_filename NULL */
2841 } else if (bs
->backing_file
[0] == '\0' && qdict_size(options
) == 0) {
2842 qobject_unref(options
);
2845 if (qdict_size(options
) == 0) {
2846 /* If the user specifies options that do not modify the
2847 * backing file's behavior, we might still consider it the
2848 * implicit backing file. But it's easier this way, and
2849 * just specifying some of the backing BDS's options is
2850 * only possible with -drive anyway (otherwise the QAPI
2851 * schema forces the user to specify everything). */
2852 implicit_backing
= !strcmp(bs
->auto_backing_file
, bs
->backing_file
);
2855 backing_filename
= bdrv_get_full_backing_filename(bs
, &local_err
);
2858 error_propagate(errp
, local_err
);
2859 qobject_unref(options
);
2864 if (!bs
->drv
|| !bs
->drv
->supports_backing
) {
2866 error_setg(errp
, "Driver doesn't support backing files");
2867 qobject_unref(options
);
2872 bs
->backing_format
[0] != '\0' && !qdict_haskey(options
, "driver")) {
2873 qdict_put_str(options
, "driver", bs
->backing_format
);
2876 backing_hd
= bdrv_open_inherit(backing_filename
, reference
, options
, 0, bs
,
2877 &child_backing
, errp
);
2879 bs
->open_flags
|= BDRV_O_NO_BACKING
;
2880 error_prepend(errp
, "Could not open backing file: ");
2885 if (implicit_backing
) {
2886 bdrv_refresh_filename(backing_hd
);
2887 pstrcpy(bs
->auto_backing_file
, sizeof(bs
->auto_backing_file
),
2888 backing_hd
->filename
);
2891 /* Hook up the backing file link; drop our reference, bs owns the
2892 * backing_hd reference now */
2893 bdrv_set_backing_hd(bs
, backing_hd
, &local_err
);
2894 bdrv_unref(backing_hd
);
2896 error_propagate(errp
, local_err
);
2901 qdict_del(parent_options
, bdref_key
);
2904 g_free(backing_filename
);
2905 qobject_unref(tmp_parent_options
);
2909 static BlockDriverState
*
2910 bdrv_open_child_bs(const char *filename
, QDict
*options
, const char *bdref_key
,
2911 BlockDriverState
*parent
, const BdrvChildClass
*child_class
,
2912 bool allow_none
, Error
**errp
)
2914 BlockDriverState
*bs
= NULL
;
2915 QDict
*image_options
;
2916 char *bdref_key_dot
;
2917 const char *reference
;
2919 assert(child_class
!= NULL
);
2921 bdref_key_dot
= g_strdup_printf("%s.", bdref_key
);
2922 qdict_extract_subqdict(options
, &image_options
, bdref_key_dot
);
2923 g_free(bdref_key_dot
);
2926 * Caution: while qdict_get_try_str() is fine, getting non-string
2927 * types would require more care. When @options come from
2928 * -blockdev or blockdev_add, its members are typed according to
2929 * the QAPI schema, but when they come from -drive, they're all
2932 reference
= qdict_get_try_str(options
, bdref_key
);
2933 if (!filename
&& !reference
&& !qdict_size(image_options
)) {
2935 error_setg(errp
, "A block device must be specified for \"%s\"",
2938 qobject_unref(image_options
);
2942 bs
= bdrv_open_inherit(filename
, reference
, image_options
, 0,
2943 parent
, child_class
, errp
);
2949 qdict_del(options
, bdref_key
);
2954 * Opens a disk image whose options are given as BlockdevRef in another block
2957 * If allow_none is true, no image will be opened if filename is false and no
2958 * BlockdevRef is given. NULL will be returned, but errp remains unset.
2960 * bdrev_key specifies the key for the image's BlockdevRef in the options QDict.
2961 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
2962 * itself, all options starting with "${bdref_key}." are considered part of the
2965 * The BlockdevRef will be removed from the options QDict.
2967 BdrvChild
*bdrv_open_child(const char *filename
,
2968 QDict
*options
, const char *bdref_key
,
2969 BlockDriverState
*parent
,
2970 const BdrvChildClass
*child_class
,
2971 BdrvChildRole child_role
,
2972 bool allow_none
, Error
**errp
)
2974 BlockDriverState
*bs
;
2976 bs
= bdrv_open_child_bs(filename
, options
, bdref_key
, parent
, child_class
,
2982 return bdrv_attach_child(parent
, bs
, bdref_key
, child_class
, child_role
,
2987 * TODO Future callers may need to specify parent/child_class in order for
2988 * option inheritance to work. Existing callers use it for the root node.
2990 BlockDriverState
*bdrv_open_blockdev_ref(BlockdevRef
*ref
, Error
**errp
)
2992 BlockDriverState
*bs
= NULL
;
2993 QObject
*obj
= NULL
;
2994 QDict
*qdict
= NULL
;
2995 const char *reference
= NULL
;
2998 if (ref
->type
== QTYPE_QSTRING
) {
2999 reference
= ref
->u
.reference
;
3001 BlockdevOptions
*options
= &ref
->u
.definition
;
3002 assert(ref
->type
== QTYPE_QDICT
);
3004 v
= qobject_output_visitor_new(&obj
);
3005 visit_type_BlockdevOptions(v
, NULL
, &options
, &error_abort
);
3006 visit_complete(v
, &obj
);
3008 qdict
= qobject_to(QDict
, obj
);
3009 qdict_flatten(qdict
);
3011 /* bdrv_open_inherit() defaults to the values in bdrv_flags (for
3012 * compatibility with other callers) rather than what we want as the
3013 * real defaults. Apply the defaults here instead. */
3014 qdict_set_default_str(qdict
, BDRV_OPT_CACHE_DIRECT
, "off");
3015 qdict_set_default_str(qdict
, BDRV_OPT_CACHE_NO_FLUSH
, "off");
3016 qdict_set_default_str(qdict
, BDRV_OPT_READ_ONLY
, "off");
3017 qdict_set_default_str(qdict
, BDRV_OPT_AUTO_READ_ONLY
, "off");
3021 bs
= bdrv_open_inherit(NULL
, reference
, qdict
, 0, NULL
, NULL
, errp
);
3028 static BlockDriverState
*bdrv_append_temp_snapshot(BlockDriverState
*bs
,
3030 QDict
*snapshot_options
,
3033 /* TODO: extra byte is a hack to ensure MAX_PATH space on Windows. */
3034 char *tmp_filename
= g_malloc0(PATH_MAX
+ 1);
3036 QemuOpts
*opts
= NULL
;
3037 BlockDriverState
*bs_snapshot
= NULL
;
3038 Error
*local_err
= NULL
;
3041 /* if snapshot, we create a temporary backing file and open it
3042 instead of opening 'filename' directly */
3044 /* Get the required size from the image */
3045 total_size
= bdrv_getlength(bs
);
3046 if (total_size
< 0) {
3047 error_setg_errno(errp
, -total_size
, "Could not get image size");
3051 /* Create the temporary image */
3052 ret
= get_tmp_filename(tmp_filename
, PATH_MAX
+ 1);
3054 error_setg_errno(errp
, -ret
, "Could not get temporary filename");
3058 opts
= qemu_opts_create(bdrv_qcow2
.create_opts
, NULL
, 0,
3060 qemu_opt_set_number(opts
, BLOCK_OPT_SIZE
, total_size
, &error_abort
);
3061 ret
= bdrv_create(&bdrv_qcow2
, tmp_filename
, opts
, errp
);
3062 qemu_opts_del(opts
);
3064 error_prepend(errp
, "Could not create temporary overlay '%s': ",
3069 /* Prepare options QDict for the temporary file */
3070 qdict_put_str(snapshot_options
, "file.driver", "file");
3071 qdict_put_str(snapshot_options
, "file.filename", tmp_filename
);
3072 qdict_put_str(snapshot_options
, "driver", "qcow2");
3074 bs_snapshot
= bdrv_open(NULL
, NULL
, snapshot_options
, flags
, errp
);
3075 snapshot_options
= NULL
;
3080 /* bdrv_append() consumes a strong reference to bs_snapshot
3081 * (i.e. it will call bdrv_unref() on it) even on error, so in
3082 * order to be able to return one, we have to increase
3083 * bs_snapshot's refcount here */
3084 bdrv_ref(bs_snapshot
);
3085 bdrv_append(bs_snapshot
, bs
, &local_err
);
3087 error_propagate(errp
, local_err
);
3093 qobject_unref(snapshot_options
);
3094 g_free(tmp_filename
);
3099 * Opens a disk image (raw, qcow2, vmdk, ...)
3101 * options is a QDict of options to pass to the block drivers, or NULL for an
3102 * empty set of options. The reference to the QDict belongs to the block layer
3103 * after the call (even on failure), so if the caller intends to reuse the
3104 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
3106 * If *pbs is NULL, a new BDS will be created with a pointer to it stored there.
3107 * If it is not NULL, the referenced BDS will be reused.
3109 * The reference parameter may be used to specify an existing block device which
3110 * should be opened. If specified, neither options nor a filename may be given,
3111 * nor can an existing BDS be reused (that is, *pbs has to be NULL).
3113 static BlockDriverState
*bdrv_open_inherit(const char *filename
,
3114 const char *reference
,
3115 QDict
*options
, int flags
,
3116 BlockDriverState
*parent
,
3117 const BdrvChildClass
*child_class
,
3121 BlockBackend
*file
= NULL
;
3122 BlockDriverState
*bs
;
3123 BlockDriver
*drv
= NULL
;
3125 const char *drvname
;
3126 const char *backing
;
3127 Error
*local_err
= NULL
;
3128 QDict
*snapshot_options
= NULL
;
3129 int snapshot_flags
= 0;
3131 assert(!child_class
|| !flags
);
3132 assert(!child_class
== !parent
);
3135 bool options_non_empty
= options
? qdict_size(options
) : false;
3136 qobject_unref(options
);
3138 if (filename
|| options_non_empty
) {
3139 error_setg(errp
, "Cannot reference an existing block device with "
3140 "additional options or a new filename");
3144 bs
= bdrv_lookup_bs(reference
, reference
, errp
);
3155 /* NULL means an empty set of options */
3156 if (options
== NULL
) {
3157 options
= qdict_new();
3160 /* json: syntax counts as explicit options, as if in the QDict */
3161 parse_json_protocol(options
, &filename
, &local_err
);
3166 bs
->explicit_options
= qdict_clone_shallow(options
);
3169 bs
->inherits_from
= parent
;
3170 child_class
->inherit_options(&flags
, options
,
3171 parent
->open_flags
, parent
->options
);
3174 ret
= bdrv_fill_options(&options
, filename
, &flags
, &local_err
);
3180 * Set the BDRV_O_RDWR and BDRV_O_ALLOW_RDWR flags.
3181 * Caution: getting a boolean member of @options requires care.
3182 * When @options come from -blockdev or blockdev_add, members are
3183 * typed according to the QAPI schema, but when they come from
3184 * -drive, they're all QString.
3186 if (g_strcmp0(qdict_get_try_str(options
, BDRV_OPT_READ_ONLY
), "on") &&
3187 !qdict_get_try_bool(options
, BDRV_OPT_READ_ONLY
, false)) {
3188 flags
|= (BDRV_O_RDWR
| BDRV_O_ALLOW_RDWR
);
3190 flags
&= ~BDRV_O_RDWR
;
3193 if (flags
& BDRV_O_SNAPSHOT
) {
3194 snapshot_options
= qdict_new();
3195 bdrv_temp_snapshot_options(&snapshot_flags
, snapshot_options
,
3197 /* Let bdrv_backing_options() override "read-only" */
3198 qdict_del(options
, BDRV_OPT_READ_ONLY
);
3199 bdrv_backing_options(&flags
, options
, flags
, options
);
3202 bs
->open_flags
= flags
;
3203 bs
->options
= options
;
3204 options
= qdict_clone_shallow(options
);
3206 /* Find the right image format driver */
3207 /* See cautionary note on accessing @options above */
3208 drvname
= qdict_get_try_str(options
, "driver");
3210 drv
= bdrv_find_format(drvname
);
3212 error_setg(errp
, "Unknown driver: '%s'", drvname
);
3217 assert(drvname
|| !(flags
& BDRV_O_PROTOCOL
));
3219 /* See cautionary note on accessing @options above */
3220 backing
= qdict_get_try_str(options
, "backing");
3221 if (qobject_to(QNull
, qdict_get(options
, "backing")) != NULL
||
3222 (backing
&& *backing
== '\0'))
3225 warn_report("Use of \"backing\": \"\" is deprecated; "
3226 "use \"backing\": null instead");
3228 flags
|= BDRV_O_NO_BACKING
;
3229 qdict_del(bs
->explicit_options
, "backing");
3230 qdict_del(bs
->options
, "backing");
3231 qdict_del(options
, "backing");
3234 /* Open image file without format layer. This BlockBackend is only used for
3235 * probing, the block drivers will do their own bdrv_open_child() for the
3236 * same BDS, which is why we put the node name back into options. */
3237 if ((flags
& BDRV_O_PROTOCOL
) == 0) {
3238 BlockDriverState
*file_bs
;
3240 file_bs
= bdrv_open_child_bs(filename
, options
, "file", bs
,
3241 &child_file
, true, &local_err
);
3245 if (file_bs
!= NULL
) {
3246 /* Not requesting BLK_PERM_CONSISTENT_READ because we're only
3247 * looking at the header to guess the image format. This works even
3248 * in cases where a guest would not see a consistent state. */
3249 file
= blk_new(bdrv_get_aio_context(file_bs
), 0, BLK_PERM_ALL
);
3250 blk_insert_bs(file
, file_bs
, &local_err
);
3251 bdrv_unref(file_bs
);
3256 qdict_put_str(options
, "file", bdrv_get_node_name(file_bs
));
3260 /* Image format probing */
3263 ret
= find_image_format(file
, filename
, &drv
, &local_err
);
3268 * This option update would logically belong in bdrv_fill_options(),
3269 * but we first need to open bs->file for the probing to work, while
3270 * opening bs->file already requires the (mostly) final set of options
3271 * so that cache mode etc. can be inherited.
3273 * Adding the driver later is somewhat ugly, but it's not an option
3274 * that would ever be inherited, so it's correct. We just need to make
3275 * sure to update both bs->options (which has the full effective
3276 * options for bs) and options (which has file.* already removed).
3278 qdict_put_str(bs
->options
, "driver", drv
->format_name
);
3279 qdict_put_str(options
, "driver", drv
->format_name
);
3281 error_setg(errp
, "Must specify either driver or file");
3285 /* BDRV_O_PROTOCOL must be set iff a protocol BDS is about to be created */
3286 assert(!!(flags
& BDRV_O_PROTOCOL
) == !!drv
->bdrv_file_open
);
3287 /* file must be NULL if a protocol BDS is about to be created
3288 * (the inverse results in an error message from bdrv_open_common()) */
3289 assert(!(flags
& BDRV_O_PROTOCOL
) || !file
);
3291 /* Open the image */
3292 ret
= bdrv_open_common(bs
, file
, options
, &local_err
);
3302 /* If there is a backing file, use it */
3303 if ((flags
& BDRV_O_NO_BACKING
) == 0) {
3304 ret
= bdrv_open_backing_file(bs
, options
, "backing", &local_err
);
3306 goto close_and_fail
;
3310 /* Remove all children options and references
3311 * from bs->options and bs->explicit_options */
3312 QLIST_FOREACH(child
, &bs
->children
, next
) {
3313 char *child_key_dot
;
3314 child_key_dot
= g_strdup_printf("%s.", child
->name
);
3315 qdict_extract_subqdict(bs
->explicit_options
, NULL
, child_key_dot
);
3316 qdict_extract_subqdict(bs
->options
, NULL
, child_key_dot
);
3317 qdict_del(bs
->explicit_options
, child
->name
);
3318 qdict_del(bs
->options
, child
->name
);
3319 g_free(child_key_dot
);
3322 /* Check if any unknown options were used */
3323 if (qdict_size(options
) != 0) {
3324 const QDictEntry
*entry
= qdict_first(options
);
3325 if (flags
& BDRV_O_PROTOCOL
) {
3326 error_setg(errp
, "Block protocol '%s' doesn't support the option "
3327 "'%s'", drv
->format_name
, entry
->key
);
3330 "Block format '%s' does not support the option '%s'",
3331 drv
->format_name
, entry
->key
);
3334 goto close_and_fail
;
3337 bdrv_parent_cb_change_media(bs
, true);
3339 qobject_unref(options
);
3342 /* For snapshot=on, create a temporary qcow2 overlay. bs points to the
3343 * temporary snapshot afterwards. */
3344 if (snapshot_flags
) {
3345 BlockDriverState
*snapshot_bs
;
3346 snapshot_bs
= bdrv_append_temp_snapshot(bs
, snapshot_flags
,
3347 snapshot_options
, &local_err
);
3348 snapshot_options
= NULL
;
3350 goto close_and_fail
;
3352 /* We are not going to return bs but the overlay on top of it
3353 * (snapshot_bs); thus, we have to drop the strong reference to bs
3354 * (which we obtained by calling bdrv_new()). bs will not be deleted,
3355 * though, because the overlay still has a reference to it. */
3364 qobject_unref(snapshot_options
);
3365 qobject_unref(bs
->explicit_options
);
3366 qobject_unref(bs
->options
);
3367 qobject_unref(options
);
3369 bs
->explicit_options
= NULL
;
3371 error_propagate(errp
, local_err
);
3376 qobject_unref(snapshot_options
);
3377 qobject_unref(options
);
3378 error_propagate(errp
, local_err
);
3382 BlockDriverState
*bdrv_open(const char *filename
, const char *reference
,
3383 QDict
*options
, int flags
, Error
**errp
)
3385 return bdrv_open_inherit(filename
, reference
, options
, flags
, NULL
,
3389 /* Return true if the NULL-terminated @list contains @str */
3390 static bool is_str_in_list(const char *str
, const char *const *list
)
3394 for (i
= 0; list
[i
] != NULL
; i
++) {
3395 if (!strcmp(str
, list
[i
])) {
3404 * Check that every option set in @bs->options is also set in
3407 * Options listed in the common_options list and in
3408 * @bs->drv->mutable_opts are skipped.
3410 * Return 0 on success, otherwise return -EINVAL and set @errp.
3412 static int bdrv_reset_options_allowed(BlockDriverState
*bs
,
3413 const QDict
*new_opts
, Error
**errp
)
3415 const QDictEntry
*e
;
3416 /* These options are common to all block drivers and are handled
3417 * in bdrv_reopen_prepare() so they can be left out of @new_opts */
3418 const char *const common_options
[] = {
3419 "node-name", "discard", "cache.direct", "cache.no-flush",
3420 "read-only", "auto-read-only", "detect-zeroes", NULL
3423 for (e
= qdict_first(bs
->options
); e
; e
= qdict_next(bs
->options
, e
)) {
3424 if (!qdict_haskey(new_opts
, e
->key
) &&
3425 !is_str_in_list(e
->key
, common_options
) &&
3426 !is_str_in_list(e
->key
, bs
->drv
->mutable_opts
)) {
3427 error_setg(errp
, "Option '%s' cannot be reset "
3428 "to its default value", e
->key
);
3437 * Returns true if @child can be reached recursively from @bs
3439 static bool bdrv_recurse_has_child(BlockDriverState
*bs
,
3440 BlockDriverState
*child
)
3448 QLIST_FOREACH(c
, &bs
->children
, next
) {
3449 if (bdrv_recurse_has_child(c
->bs
, child
)) {
3458 * Adds a BlockDriverState to a simple queue for an atomic, transactional
3459 * reopen of multiple devices.
3461 * bs_queue can either be an existing BlockReopenQueue that has had QTAILQ_INIT
3462 * already performed, or alternatively may be NULL a new BlockReopenQueue will
3463 * be created and initialized. This newly created BlockReopenQueue should be
3464 * passed back in for subsequent calls that are intended to be of the same
3467 * bs is the BlockDriverState to add to the reopen queue.
3469 * options contains the changed options for the associated bs
3470 * (the BlockReopenQueue takes ownership)
3472 * flags contains the open flags for the associated bs
3474 * returns a pointer to bs_queue, which is either the newly allocated
3475 * bs_queue, or the existing bs_queue being used.
3477 * bs must be drained between bdrv_reopen_queue() and bdrv_reopen_multiple().
3479 static BlockReopenQueue
*bdrv_reopen_queue_child(BlockReopenQueue
*bs_queue
,
3480 BlockDriverState
*bs
,
3482 const BdrvChildClass
*klass
,
3483 QDict
*parent_options
,
3489 BlockReopenQueueEntry
*bs_entry
;
3491 QDict
*old_options
, *explicit_options
, *options_copy
;
3495 /* Make sure that the caller remembered to use a drained section. This is
3496 * important to avoid graph changes between the recursive queuing here and
3497 * bdrv_reopen_multiple(). */
3498 assert(bs
->quiesce_counter
> 0);
3500 if (bs_queue
== NULL
) {
3501 bs_queue
= g_new0(BlockReopenQueue
, 1);
3502 QTAILQ_INIT(bs_queue
);
3506 options
= qdict_new();
3509 /* Check if this BlockDriverState is already in the queue */
3510 QTAILQ_FOREACH(bs_entry
, bs_queue
, entry
) {
3511 if (bs
== bs_entry
->state
.bs
) {
3517 * Precedence of options:
3518 * 1. Explicitly passed in options (highest)
3519 * 2. Retained from explicitly set options of bs
3520 * 3. Inherited from parent node
3521 * 4. Retained from effective options of bs
3524 /* Old explicitly set values (don't overwrite by inherited value) */
3525 if (bs_entry
|| keep_old_opts
) {
3526 old_options
= qdict_clone_shallow(bs_entry
?
3527 bs_entry
->state
.explicit_options
:
3528 bs
->explicit_options
);
3529 bdrv_join_options(bs
, options
, old_options
);
3530 qobject_unref(old_options
);
3533 explicit_options
= qdict_clone_shallow(options
);
3535 /* Inherit from parent node */
3536 if (parent_options
) {
3538 klass
->inherit_options(&flags
, options
, parent_flags
, parent_options
);
3540 flags
= bdrv_get_flags(bs
);
3543 if (keep_old_opts
) {
3544 /* Old values are used for options that aren't set yet */
3545 old_options
= qdict_clone_shallow(bs
->options
);
3546 bdrv_join_options(bs
, options
, old_options
);
3547 qobject_unref(old_options
);
3550 /* We have the final set of options so let's update the flags */
3551 options_copy
= qdict_clone_shallow(options
);
3552 opts
= qemu_opts_create(&bdrv_runtime_opts
, NULL
, 0, &error_abort
);
3553 qemu_opts_absorb_qdict(opts
, options_copy
, NULL
);
3554 update_flags_from_options(&flags
, opts
);
3555 qemu_opts_del(opts
);
3556 qobject_unref(options_copy
);
3558 /* bdrv_open_inherit() sets and clears some additional flags internally */
3559 flags
&= ~BDRV_O_PROTOCOL
;
3560 if (flags
& BDRV_O_RDWR
) {
3561 flags
|= BDRV_O_ALLOW_RDWR
;
3565 bs_entry
= g_new0(BlockReopenQueueEntry
, 1);
3566 QTAILQ_INSERT_TAIL(bs_queue
, bs_entry
, entry
);
3568 qobject_unref(bs_entry
->state
.options
);
3569 qobject_unref(bs_entry
->state
.explicit_options
);
3572 bs_entry
->state
.bs
= bs
;
3573 bs_entry
->state
.options
= options
;
3574 bs_entry
->state
.explicit_options
= explicit_options
;
3575 bs_entry
->state
.flags
= flags
;
3577 /* This needs to be overwritten in bdrv_reopen_prepare() */
3578 bs_entry
->state
.perm
= UINT64_MAX
;
3579 bs_entry
->state
.shared_perm
= 0;
3582 * If keep_old_opts is false then it means that unspecified
3583 * options must be reset to their original value. We don't allow
3584 * resetting 'backing' but we need to know if the option is
3585 * missing in order to decide if we have to return an error.
3587 if (!keep_old_opts
) {
3588 bs_entry
->state
.backing_missing
=
3589 !qdict_haskey(options
, "backing") &&
3590 !qdict_haskey(options
, "backing.driver");
3593 QLIST_FOREACH(child
, &bs
->children
, next
) {
3594 QDict
*new_child_options
= NULL
;
3595 bool child_keep_old
= keep_old_opts
;
3597 /* reopen can only change the options of block devices that were
3598 * implicitly created and inherited options. For other (referenced)
3599 * block devices, a syntax like "backing.foo" results in an error. */
3600 if (child
->bs
->inherits_from
!= bs
) {
3604 /* Check if the options contain a child reference */
3605 if (qdict_haskey(options
, child
->name
)) {
3606 const char *childref
= qdict_get_try_str(options
, child
->name
);
3608 * The current child must not be reopened if the child
3609 * reference is null or points to a different node.
3611 if (g_strcmp0(childref
, child
->bs
->node_name
)) {
3615 * If the child reference points to the current child then
3616 * reopen it with its existing set of options (note that
3617 * it can still inherit new options from the parent).
3619 child_keep_old
= true;
3621 /* Extract child options ("child-name.*") */
3622 char *child_key_dot
= g_strdup_printf("%s.", child
->name
);
3623 qdict_extract_subqdict(explicit_options
, NULL
, child_key_dot
);
3624 qdict_extract_subqdict(options
, &new_child_options
, child_key_dot
);
3625 g_free(child_key_dot
);
3628 bdrv_reopen_queue_child(bs_queue
, child
->bs
, new_child_options
,
3629 child
->klass
, options
, flags
, child_keep_old
);
3635 BlockReopenQueue
*bdrv_reopen_queue(BlockReopenQueue
*bs_queue
,
3636 BlockDriverState
*bs
,
3637 QDict
*options
, bool keep_old_opts
)
3639 return bdrv_reopen_queue_child(bs_queue
, bs
, options
, NULL
, NULL
, 0,
3644 * Reopen multiple BlockDriverStates atomically & transactionally.
3646 * The queue passed in (bs_queue) must have been built up previous
3647 * via bdrv_reopen_queue().
3649 * Reopens all BDS specified in the queue, with the appropriate
3650 * flags. All devices are prepared for reopen, and failure of any
3651 * device will cause all device changes to be abandoned, and intermediate
3654 * If all devices prepare successfully, then the changes are committed
3657 * All affected nodes must be drained between bdrv_reopen_queue() and
3658 * bdrv_reopen_multiple().
3660 int bdrv_reopen_multiple(BlockReopenQueue
*bs_queue
, Error
**errp
)
3663 BlockReopenQueueEntry
*bs_entry
, *next
;
3665 assert(bs_queue
!= NULL
);
3667 QTAILQ_FOREACH(bs_entry
, bs_queue
, entry
) {
3668 assert(bs_entry
->state
.bs
->quiesce_counter
> 0);
3669 if (bdrv_reopen_prepare(&bs_entry
->state
, bs_queue
, errp
)) {
3672 bs_entry
->prepared
= true;
3675 QTAILQ_FOREACH(bs_entry
, bs_queue
, entry
) {
3676 BDRVReopenState
*state
= &bs_entry
->state
;
3677 ret
= bdrv_check_perm(state
->bs
, bs_queue
, state
->perm
,
3678 state
->shared_perm
, NULL
, NULL
, errp
);
3682 /* Check if new_backing_bs would accept the new permissions */
3683 if (state
->replace_backing_bs
&& state
->new_backing_bs
) {
3684 uint64_t nperm
, nshared
;
3685 bdrv_child_perm(state
->bs
, state
->new_backing_bs
,
3686 NULL
, &child_backing
, bs_queue
,
3687 state
->perm
, state
->shared_perm
,
3689 ret
= bdrv_check_update_perm(state
->new_backing_bs
, NULL
,
3690 nperm
, nshared
, NULL
, NULL
, errp
);
3695 bs_entry
->perms_checked
= true;
3699 * If we reach this point, we have success and just need to apply the
3702 * Reverse order is used to comfort qcow2 driver: on commit it need to write
3703 * IN_USE flag to the image, to mark bitmaps in the image as invalid. But
3704 * children are usually goes after parents in reopen-queue, so go from last
3707 QTAILQ_FOREACH_REVERSE(bs_entry
, bs_queue
, entry
) {
3708 bdrv_reopen_commit(&bs_entry
->state
);
3713 QTAILQ_FOREACH_SAFE(bs_entry
, bs_queue
, entry
, next
) {
3714 BDRVReopenState
*state
= &bs_entry
->state
;
3716 if (!bs_entry
->perms_checked
) {
3721 bdrv_set_perm(state
->bs
, state
->perm
, state
->shared_perm
);
3723 bdrv_abort_perm_update(state
->bs
);
3724 if (state
->replace_backing_bs
&& state
->new_backing_bs
) {
3725 bdrv_abort_perm_update(state
->new_backing_bs
);
3731 QTAILQ_FOREACH_REVERSE(bs_entry
, bs_queue
, entry
) {
3732 BlockDriverState
*bs
= bs_entry
->state
.bs
;
3734 if (bs
->drv
->bdrv_reopen_commit_post
)
3735 bs
->drv
->bdrv_reopen_commit_post(&bs_entry
->state
);
3739 QTAILQ_FOREACH_SAFE(bs_entry
, bs_queue
, entry
, next
) {
3741 if (bs_entry
->prepared
) {
3742 bdrv_reopen_abort(&bs_entry
->state
);
3744 qobject_unref(bs_entry
->state
.explicit_options
);
3745 qobject_unref(bs_entry
->state
.options
);
3747 if (bs_entry
->state
.new_backing_bs
) {
3748 bdrv_unref(bs_entry
->state
.new_backing_bs
);
3757 int bdrv_reopen_set_read_only(BlockDriverState
*bs
, bool read_only
,
3761 BlockReopenQueue
*queue
;
3762 QDict
*opts
= qdict_new();
3764 qdict_put_bool(opts
, BDRV_OPT_READ_ONLY
, read_only
);
3766 bdrv_subtree_drained_begin(bs
);
3767 queue
= bdrv_reopen_queue(NULL
, bs
, opts
, true);
3768 ret
= bdrv_reopen_multiple(queue
, errp
);
3769 bdrv_subtree_drained_end(bs
);
3774 static BlockReopenQueueEntry
*find_parent_in_reopen_queue(BlockReopenQueue
*q
,
3777 BlockReopenQueueEntry
*entry
;
3779 QTAILQ_FOREACH(entry
, q
, entry
) {
3780 BlockDriverState
*bs
= entry
->state
.bs
;
3783 QLIST_FOREACH(child
, &bs
->children
, next
) {
3793 static void bdrv_reopen_perm(BlockReopenQueue
*q
, BlockDriverState
*bs
,
3794 uint64_t *perm
, uint64_t *shared
)
3797 BlockReopenQueueEntry
*parent
;
3798 uint64_t cumulative_perms
= 0;
3799 uint64_t cumulative_shared_perms
= BLK_PERM_ALL
;
3801 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
3802 parent
= find_parent_in_reopen_queue(q
, c
);
3804 cumulative_perms
|= c
->perm
;
3805 cumulative_shared_perms
&= c
->shared_perm
;
3807 uint64_t nperm
, nshared
;
3809 bdrv_child_perm(parent
->state
.bs
, bs
, c
, c
->klass
, q
,
3810 parent
->state
.perm
, parent
->state
.shared_perm
,
3813 cumulative_perms
|= nperm
;
3814 cumulative_shared_perms
&= nshared
;
3817 *perm
= cumulative_perms
;
3818 *shared
= cumulative_shared_perms
;
3821 static bool bdrv_reopen_can_attach(BlockDriverState
*parent
,
3823 BlockDriverState
*new_child
,
3826 AioContext
*parent_ctx
= bdrv_get_aio_context(parent
);
3827 AioContext
*child_ctx
= bdrv_get_aio_context(new_child
);
3831 ignore
= g_slist_prepend(NULL
, child
);
3832 ret
= bdrv_can_set_aio_context(new_child
, parent_ctx
, &ignore
, NULL
);
3833 g_slist_free(ignore
);
3838 ignore
= g_slist_prepend(NULL
, child
);
3839 ret
= bdrv_can_set_aio_context(parent
, child_ctx
, &ignore
, errp
);
3840 g_slist_free(ignore
);
3845 * Take a BDRVReopenState and check if the value of 'backing' in the
3846 * reopen_state->options QDict is valid or not.
3848 * If 'backing' is missing from the QDict then return 0.
3850 * If 'backing' contains the node name of the backing file of
3851 * reopen_state->bs then return 0.
3853 * If 'backing' contains a different node name (or is null) then check
3854 * whether the current backing file can be replaced with the new one.
3855 * If that's the case then reopen_state->replace_backing_bs is set to
3856 * true and reopen_state->new_backing_bs contains a pointer to the new
3857 * backing BlockDriverState (or NULL).
3859 * Return 0 on success, otherwise return < 0 and set @errp.
3861 static int bdrv_reopen_parse_backing(BDRVReopenState
*reopen_state
,
3864 BlockDriverState
*bs
= reopen_state
->bs
;
3865 BlockDriverState
*overlay_bs
, *new_backing_bs
;
3869 value
= qdict_get(reopen_state
->options
, "backing");
3870 if (value
== NULL
) {
3874 switch (qobject_type(value
)) {
3876 new_backing_bs
= NULL
;
3879 str
= qobject_get_try_str(value
);
3880 new_backing_bs
= bdrv_lookup_bs(NULL
, str
, errp
);
3881 if (new_backing_bs
== NULL
) {
3883 } else if (bdrv_recurse_has_child(new_backing_bs
, bs
)) {
3884 error_setg(errp
, "Making '%s' a backing file of '%s' "
3885 "would create a cycle", str
, bs
->node_name
);
3890 /* 'backing' does not allow any other data type */
3891 g_assert_not_reached();
3895 * Check AioContext compatibility so that the bdrv_set_backing_hd() call in
3896 * bdrv_reopen_commit() won't fail.
3898 if (new_backing_bs
) {
3899 if (!bdrv_reopen_can_attach(bs
, bs
->backing
, new_backing_bs
, errp
)) {
3905 * Find the "actual" backing file by skipping all links that point
3906 * to an implicit node, if any (e.g. a commit filter node).
3909 while (backing_bs(overlay_bs
) && backing_bs(overlay_bs
)->implicit
) {
3910 overlay_bs
= backing_bs(overlay_bs
);
3913 /* If we want to replace the backing file we need some extra checks */
3914 if (new_backing_bs
!= backing_bs(overlay_bs
)) {
3915 /* Check for implicit nodes between bs and its backing file */
3916 if (bs
!= overlay_bs
) {
3917 error_setg(errp
, "Cannot change backing link if '%s' has "
3918 "an implicit backing file", bs
->node_name
);
3921 /* Check if the backing link that we want to replace is frozen */
3922 if (bdrv_is_backing_chain_frozen(overlay_bs
, backing_bs(overlay_bs
),
3926 reopen_state
->replace_backing_bs
= true;
3927 if (new_backing_bs
) {
3928 bdrv_ref(new_backing_bs
);
3929 reopen_state
->new_backing_bs
= new_backing_bs
;
3937 * Prepares a BlockDriverState for reopen. All changes are staged in the
3938 * 'opaque' field of the BDRVReopenState, which is used and allocated by
3939 * the block driver layer .bdrv_reopen_prepare()
3941 * bs is the BlockDriverState to reopen
3942 * flags are the new open flags
3943 * queue is the reopen queue
3945 * Returns 0 on success, non-zero on error. On error errp will be set
3948 * On failure, bdrv_reopen_abort() will be called to clean up any data.
3949 * It is the responsibility of the caller to then call the abort() or
3950 * commit() for any other BDS that have been left in a prepare() state
3953 int bdrv_reopen_prepare(BDRVReopenState
*reopen_state
, BlockReopenQueue
*queue
,
3958 Error
*local_err
= NULL
;
3961 QDict
*orig_reopen_opts
;
3962 char *discard
= NULL
;
3964 bool drv_prepared
= false;
3966 assert(reopen_state
!= NULL
);
3967 assert(reopen_state
->bs
->drv
!= NULL
);
3968 drv
= reopen_state
->bs
->drv
;
3970 /* This function and each driver's bdrv_reopen_prepare() remove
3971 * entries from reopen_state->options as they are processed, so
3972 * we need to make a copy of the original QDict. */
3973 orig_reopen_opts
= qdict_clone_shallow(reopen_state
->options
);
3975 /* Process generic block layer options */
3976 opts
= qemu_opts_create(&bdrv_runtime_opts
, NULL
, 0, &error_abort
);
3977 qemu_opts_absorb_qdict(opts
, reopen_state
->options
, &local_err
);
3979 error_propagate(errp
, local_err
);
3984 /* This was already called in bdrv_reopen_queue_child() so the flags
3985 * are up-to-date. This time we simply want to remove the options from
3986 * QemuOpts in order to indicate that they have been processed. */
3987 old_flags
= reopen_state
->flags
;
3988 update_flags_from_options(&reopen_state
->flags
, opts
);
3989 assert(old_flags
== reopen_state
->flags
);
3991 discard
= qemu_opt_get_del(opts
, BDRV_OPT_DISCARD
);
3992 if (discard
!= NULL
) {
3993 if (bdrv_parse_discard_flags(discard
, &reopen_state
->flags
) != 0) {
3994 error_setg(errp
, "Invalid discard option");
4000 reopen_state
->detect_zeroes
=
4001 bdrv_parse_detect_zeroes(opts
, reopen_state
->flags
, &local_err
);
4003 error_propagate(errp
, local_err
);
4008 /* All other options (including node-name and driver) must be unchanged.
4009 * Put them back into the QDict, so that they are checked at the end
4010 * of this function. */
4011 qemu_opts_to_qdict(opts
, reopen_state
->options
);
4013 /* If we are to stay read-only, do not allow permission change
4014 * to r/w. Attempting to set to r/w may fail if either BDRV_O_ALLOW_RDWR is
4015 * not set, or if the BDS still has copy_on_read enabled */
4016 read_only
= !(reopen_state
->flags
& BDRV_O_RDWR
);
4017 ret
= bdrv_can_set_read_only(reopen_state
->bs
, read_only
, true, &local_err
);
4019 error_propagate(errp
, local_err
);
4023 /* Calculate required permissions after reopening */
4024 bdrv_reopen_perm(queue
, reopen_state
->bs
,
4025 &reopen_state
->perm
, &reopen_state
->shared_perm
);
4027 ret
= bdrv_flush(reopen_state
->bs
);
4029 error_setg_errno(errp
, -ret
, "Error flushing drive");
4033 if (drv
->bdrv_reopen_prepare
) {
4035 * If a driver-specific option is missing, it means that we
4036 * should reset it to its default value.
4037 * But not all options allow that, so we need to check it first.
4039 ret
= bdrv_reset_options_allowed(reopen_state
->bs
,
4040 reopen_state
->options
, errp
);
4045 ret
= drv
->bdrv_reopen_prepare(reopen_state
, queue
, &local_err
);
4047 if (local_err
!= NULL
) {
4048 error_propagate(errp
, local_err
);
4050 bdrv_refresh_filename(reopen_state
->bs
);
4051 error_setg(errp
, "failed while preparing to reopen image '%s'",
4052 reopen_state
->bs
->filename
);
4057 /* It is currently mandatory to have a bdrv_reopen_prepare()
4058 * handler for each supported drv. */
4059 error_setg(errp
, "Block format '%s' used by node '%s' "
4060 "does not support reopening files", drv
->format_name
,
4061 bdrv_get_device_or_node_name(reopen_state
->bs
));
4066 drv_prepared
= true;
4069 * We must provide the 'backing' option if the BDS has a backing
4070 * file or if the image file has a backing file name as part of
4071 * its metadata. Otherwise the 'backing' option can be omitted.
4073 if (drv
->supports_backing
&& reopen_state
->backing_missing
&&
4074 (backing_bs(reopen_state
->bs
) || reopen_state
->bs
->backing_file
[0])) {
4075 error_setg(errp
, "backing is missing for '%s'",
4076 reopen_state
->bs
->node_name
);
4082 * Allow changing the 'backing' option. The new value can be
4083 * either a reference to an existing node (using its node name)
4084 * or NULL to simply detach the current backing file.
4086 ret
= bdrv_reopen_parse_backing(reopen_state
, errp
);
4090 qdict_del(reopen_state
->options
, "backing");
4092 /* Options that are not handled are only okay if they are unchanged
4093 * compared to the old state. It is expected that some options are only
4094 * used for the initial open, but not reopen (e.g. filename) */
4095 if (qdict_size(reopen_state
->options
)) {
4096 const QDictEntry
*entry
= qdict_first(reopen_state
->options
);
4099 QObject
*new = entry
->value
;
4100 QObject
*old
= qdict_get(reopen_state
->bs
->options
, entry
->key
);
4102 /* Allow child references (child_name=node_name) as long as they
4103 * point to the current child (i.e. everything stays the same). */
4104 if (qobject_type(new) == QTYPE_QSTRING
) {
4106 QLIST_FOREACH(child
, &reopen_state
->bs
->children
, next
) {
4107 if (!strcmp(child
->name
, entry
->key
)) {
4113 const char *str
= qobject_get_try_str(new);
4114 if (!strcmp(child
->bs
->node_name
, str
)) {
4115 continue; /* Found child with this name, skip option */
4121 * TODO: When using -drive to specify blockdev options, all values
4122 * will be strings; however, when using -blockdev, blockdev-add or
4123 * filenames using the json:{} pseudo-protocol, they will be
4125 * In contrast, reopening options are (currently) always strings
4126 * (because you can only specify them through qemu-io; all other
4127 * callers do not specify any options).
4128 * Therefore, when using anything other than -drive to create a BDS,
4129 * this cannot detect non-string options as unchanged, because
4130 * qobject_is_equal() always returns false for objects of different
4131 * type. In the future, this should be remedied by correctly typing
4132 * all options. For now, this is not too big of an issue because
4133 * the user can simply omit options which cannot be changed anyway,
4134 * so they will stay unchanged.
4136 if (!qobject_is_equal(new, old
)) {
4137 error_setg(errp
, "Cannot change the option '%s'", entry
->key
);
4141 } while ((entry
= qdict_next(reopen_state
->options
, entry
)));
4146 /* Restore the original reopen_state->options QDict */
4147 qobject_unref(reopen_state
->options
);
4148 reopen_state
->options
= qobject_ref(orig_reopen_opts
);
4151 if (ret
< 0 && drv_prepared
) {
4152 /* drv->bdrv_reopen_prepare() has succeeded, so we need to
4153 * call drv->bdrv_reopen_abort() before signaling an error
4154 * (bdrv_reopen_multiple() will not call bdrv_reopen_abort()
4155 * when the respective bdrv_reopen_prepare() has failed) */
4156 if (drv
->bdrv_reopen_abort
) {
4157 drv
->bdrv_reopen_abort(reopen_state
);
4160 qemu_opts_del(opts
);
4161 qobject_unref(orig_reopen_opts
);
4167 * Takes the staged changes for the reopen from bdrv_reopen_prepare(), and
4168 * makes them final by swapping the staging BlockDriverState contents into
4169 * the active BlockDriverState contents.
4171 void bdrv_reopen_commit(BDRVReopenState
*reopen_state
)
4174 BlockDriverState
*bs
;
4177 assert(reopen_state
!= NULL
);
4178 bs
= reopen_state
->bs
;
4180 assert(drv
!= NULL
);
4182 /* If there are any driver level actions to take */
4183 if (drv
->bdrv_reopen_commit
) {
4184 drv
->bdrv_reopen_commit(reopen_state
);
4187 /* set BDS specific flags now */
4188 qobject_unref(bs
->explicit_options
);
4189 qobject_unref(bs
->options
);
4191 bs
->explicit_options
= reopen_state
->explicit_options
;
4192 bs
->options
= reopen_state
->options
;
4193 bs
->open_flags
= reopen_state
->flags
;
4194 bs
->read_only
= !(reopen_state
->flags
& BDRV_O_RDWR
);
4195 bs
->detect_zeroes
= reopen_state
->detect_zeroes
;
4197 if (reopen_state
->replace_backing_bs
) {
4198 qdict_del(bs
->explicit_options
, "backing");
4199 qdict_del(bs
->options
, "backing");
4202 /* Remove child references from bs->options and bs->explicit_options.
4203 * Child options were already removed in bdrv_reopen_queue_child() */
4204 QLIST_FOREACH(child
, &bs
->children
, next
) {
4205 qdict_del(bs
->explicit_options
, child
->name
);
4206 qdict_del(bs
->options
, child
->name
);
4210 * Change the backing file if a new one was specified. We do this
4211 * after updating bs->options, so bdrv_refresh_filename() (called
4212 * from bdrv_set_backing_hd()) has the new values.
4214 if (reopen_state
->replace_backing_bs
) {
4215 BlockDriverState
*old_backing_bs
= backing_bs(bs
);
4216 assert(!old_backing_bs
|| !old_backing_bs
->implicit
);
4217 /* Abort the permission update on the backing bs we're detaching */
4218 if (old_backing_bs
) {
4219 bdrv_abort_perm_update(old_backing_bs
);
4221 bdrv_set_backing_hd(bs
, reopen_state
->new_backing_bs
, &error_abort
);
4224 bdrv_refresh_limits(bs
, NULL
);
4228 * Abort the reopen, and delete and free the staged changes in
4231 void bdrv_reopen_abort(BDRVReopenState
*reopen_state
)
4235 assert(reopen_state
!= NULL
);
4236 drv
= reopen_state
->bs
->drv
;
4237 assert(drv
!= NULL
);
4239 if (drv
->bdrv_reopen_abort
) {
4240 drv
->bdrv_reopen_abort(reopen_state
);
4245 static void bdrv_close(BlockDriverState
*bs
)
4247 BdrvAioNotifier
*ban
, *ban_next
;
4248 BdrvChild
*child
, *next
;
4250 assert(!bs
->refcnt
);
4252 bdrv_drained_begin(bs
); /* complete I/O */
4254 bdrv_drain(bs
); /* in case flush left pending I/O */
4257 if (bs
->drv
->bdrv_close
) {
4258 bs
->drv
->bdrv_close(bs
);
4263 QLIST_FOREACH_SAFE(child
, &bs
->children
, next
, next
) {
4264 bdrv_unref_child(bs
, child
);
4271 atomic_set(&bs
->copy_on_read
, 0);
4272 bs
->backing_file
[0] = '\0';
4273 bs
->backing_format
[0] = '\0';
4274 bs
->total_sectors
= 0;
4275 bs
->encrypted
= false;
4277 qobject_unref(bs
->options
);
4278 qobject_unref(bs
->explicit_options
);
4280 bs
->explicit_options
= NULL
;
4281 qobject_unref(bs
->full_open_options
);
4282 bs
->full_open_options
= NULL
;
4284 bdrv_release_named_dirty_bitmaps(bs
);
4285 assert(QLIST_EMPTY(&bs
->dirty_bitmaps
));
4287 QLIST_FOREACH_SAFE(ban
, &bs
->aio_notifiers
, list
, ban_next
) {
4290 QLIST_INIT(&bs
->aio_notifiers
);
4291 bdrv_drained_end(bs
);
4294 void bdrv_close_all(void)
4296 assert(job_next(NULL
) == NULL
);
4297 nbd_export_close_all();
4299 /* Drop references from requests still in flight, such as canceled block
4300 * jobs whose AIO context has not been polled yet */
4303 blk_remove_all_bs();
4304 blockdev_close_all_bdrv_states();
4306 assert(QTAILQ_EMPTY(&all_bdrv_states
));
4309 static bool should_update_child(BdrvChild
*c
, BlockDriverState
*to
)
4315 if (c
->klass
->stay_at_node
) {
4319 /* If the child @c belongs to the BDS @to, replacing the current
4320 * c->bs by @to would mean to create a loop.
4322 * Such a case occurs when appending a BDS to a backing chain.
4323 * For instance, imagine the following chain:
4325 * guest device -> node A -> further backing chain...
4327 * Now we create a new BDS B which we want to put on top of this
4328 * chain, so we first attach A as its backing node:
4333 * guest device -> node A -> further backing chain...
4335 * Finally we want to replace A by B. When doing that, we want to
4336 * replace all pointers to A by pointers to B -- except for the
4337 * pointer from B because (1) that would create a loop, and (2)
4338 * that pointer should simply stay intact:
4340 * guest device -> node B
4343 * node A -> further backing chain...
4345 * In general, when replacing a node A (c->bs) by a node B (@to),
4346 * if A is a child of B, that means we cannot replace A by B there
4347 * because that would create a loop. Silently detaching A from B
4348 * is also not really an option. So overall just leaving A in
4349 * place there is the most sensible choice.
4351 * We would also create a loop in any cases where @c is only
4352 * indirectly referenced by @to. Prevent this by returning false
4353 * if @c is found (by breadth-first search) anywhere in the whole
4358 found
= g_hash_table_new(NULL
, NULL
);
4359 g_hash_table_add(found
, to
);
4360 queue
= g_queue_new();
4361 g_queue_push_tail(queue
, to
);
4363 while (!g_queue_is_empty(queue
)) {
4364 BlockDriverState
*v
= g_queue_pop_head(queue
);
4367 QLIST_FOREACH(c2
, &v
->children
, next
) {
4373 if (g_hash_table_contains(found
, c2
->bs
)) {
4377 g_queue_push_tail(queue
, c2
->bs
);
4378 g_hash_table_add(found
, c2
->bs
);
4382 g_queue_free(queue
);
4383 g_hash_table_destroy(found
);
4388 void bdrv_replace_node(BlockDriverState
*from
, BlockDriverState
*to
,
4391 BdrvChild
*c
, *next
;
4392 GSList
*list
= NULL
, *p
;
4393 uint64_t perm
= 0, shared
= BLK_PERM_ALL
;
4396 /* Make sure that @from doesn't go away until we have successfully attached
4397 * all of its parents to @to. */
4400 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
4401 assert(bdrv_get_aio_context(from
) == bdrv_get_aio_context(to
));
4402 bdrv_drained_begin(from
);
4404 /* Put all parents into @list and calculate their cumulative permissions */
4405 QLIST_FOREACH_SAFE(c
, &from
->parents
, next_parent
, next
) {
4406 assert(c
->bs
== from
);
4407 if (!should_update_child(c
, to
)) {
4411 error_setg(errp
, "Cannot change '%s' link to '%s'",
4412 c
->name
, from
->node_name
);
4415 list
= g_slist_prepend(list
, c
);
4417 shared
&= c
->shared_perm
;
4420 /* Check whether the required permissions can be granted on @to, ignoring
4421 * all BdrvChild in @list so that they can't block themselves. */
4422 ret
= bdrv_check_update_perm(to
, NULL
, perm
, shared
, list
, NULL
, errp
);
4424 bdrv_abort_perm_update(to
);
4428 /* Now actually perform the change. We performed the permission check for
4429 * all elements of @list at once, so set the permissions all at once at the
4431 for (p
= list
; p
!= NULL
; p
= p
->next
) {
4435 bdrv_replace_child_noperm(c
, to
);
4439 bdrv_get_cumulative_perm(to
, &perm
, &shared
);
4440 bdrv_set_perm(to
, perm
, shared
);
4444 bdrv_drained_end(from
);
4449 * Add new bs contents at the top of an image chain while the chain is
4450 * live, while keeping required fields on the top layer.
4452 * This will modify the BlockDriverState fields, and swap contents
4453 * between bs_new and bs_top. Both bs_new and bs_top are modified.
4455 * bs_new must not be attached to a BlockBackend.
4457 * This function does not create any image files.
4459 * bdrv_append() takes ownership of a bs_new reference and unrefs it because
4460 * that's what the callers commonly need. bs_new will be referenced by the old
4461 * parents of bs_top after bdrv_append() returns. If the caller needs to keep a
4462 * reference of its own, it must call bdrv_ref().
4464 void bdrv_append(BlockDriverState
*bs_new
, BlockDriverState
*bs_top
,
4467 Error
*local_err
= NULL
;
4469 bdrv_set_backing_hd(bs_new
, bs_top
, &local_err
);
4471 error_propagate(errp
, local_err
);
4475 bdrv_replace_node(bs_top
, bs_new
, &local_err
);
4477 error_propagate(errp
, local_err
);
4478 bdrv_set_backing_hd(bs_new
, NULL
, &error_abort
);
4482 /* bs_new is now referenced by its new parents, we don't need the
4483 * additional reference any more. */
4488 static void bdrv_delete(BlockDriverState
*bs
)
4490 assert(bdrv_op_blocker_is_empty(bs
));
4491 assert(!bs
->refcnt
);
4493 /* remove from list, if necessary */
4494 if (bs
->node_name
[0] != '\0') {
4495 QTAILQ_REMOVE(&graph_bdrv_states
, bs
, node_list
);
4497 QTAILQ_REMOVE(&all_bdrv_states
, bs
, bs_list
);
4505 * Run consistency checks on an image
4507 * Returns 0 if the check could be completed (it doesn't mean that the image is
4508 * free of errors) or -errno when an internal error occurred. The results of the
4509 * check are stored in res.
4511 static int coroutine_fn
bdrv_co_check(BlockDriverState
*bs
,
4512 BdrvCheckResult
*res
, BdrvCheckMode fix
)
4514 if (bs
->drv
== NULL
) {
4517 if (bs
->drv
->bdrv_co_check
== NULL
) {
4521 memset(res
, 0, sizeof(*res
));
4522 return bs
->drv
->bdrv_co_check(bs
, res
, fix
);
4525 typedef struct CheckCo
{
4526 BlockDriverState
*bs
;
4527 BdrvCheckResult
*res
;
4532 static void coroutine_fn
bdrv_check_co_entry(void *opaque
)
4534 CheckCo
*cco
= opaque
;
4535 cco
->ret
= bdrv_co_check(cco
->bs
, cco
->res
, cco
->fix
);
4539 int bdrv_check(BlockDriverState
*bs
,
4540 BdrvCheckResult
*res
, BdrvCheckMode fix
)
4546 .ret
= -EINPROGRESS
,
4550 if (qemu_in_coroutine()) {
4551 /* Fast-path if already in coroutine context */
4552 bdrv_check_co_entry(&cco
);
4554 co
= qemu_coroutine_create(bdrv_check_co_entry
, &cco
);
4555 bdrv_coroutine_enter(bs
, co
);
4556 BDRV_POLL_WHILE(bs
, cco
.ret
== -EINPROGRESS
);
4565 * -EINVAL - backing format specified, but no file
4566 * -ENOSPC - can't update the backing file because no space is left in the
4568 * -ENOTSUP - format driver doesn't support changing the backing file
4570 int bdrv_change_backing_file(BlockDriverState
*bs
,
4571 const char *backing_file
, const char *backing_fmt
)
4573 BlockDriver
*drv
= bs
->drv
;
4580 /* Backing file format doesn't make sense without a backing file */
4581 if (backing_fmt
&& !backing_file
) {
4585 if (drv
->bdrv_change_backing_file
!= NULL
) {
4586 ret
= drv
->bdrv_change_backing_file(bs
, backing_file
, backing_fmt
);
4592 pstrcpy(bs
->backing_file
, sizeof(bs
->backing_file
), backing_file
?: "");
4593 pstrcpy(bs
->backing_format
, sizeof(bs
->backing_format
), backing_fmt
?: "");
4594 pstrcpy(bs
->auto_backing_file
, sizeof(bs
->auto_backing_file
),
4595 backing_file
?: "");
4601 * Finds the image layer in the chain that has 'bs' as its backing file.
4603 * active is the current topmost image.
4605 * Returns NULL if bs is not found in active's image chain,
4606 * or if active == bs.
4608 * Returns the bottommost base image if bs == NULL.
4610 BlockDriverState
*bdrv_find_overlay(BlockDriverState
*active
,
4611 BlockDriverState
*bs
)
4613 while (active
&& bs
!= backing_bs(active
)) {
4614 active
= backing_bs(active
);
4620 /* Given a BDS, searches for the base layer. */
4621 BlockDriverState
*bdrv_find_base(BlockDriverState
*bs
)
4623 return bdrv_find_overlay(bs
, NULL
);
4627 * Return true if at least one of the backing links between @bs and
4628 * @base is frozen. @errp is set if that's the case.
4629 * @base must be reachable from @bs, or NULL.
4631 bool bdrv_is_backing_chain_frozen(BlockDriverState
*bs
, BlockDriverState
*base
,
4634 BlockDriverState
*i
;
4636 for (i
= bs
; i
!= base
; i
= backing_bs(i
)) {
4637 if (i
->backing
&& i
->backing
->frozen
) {
4638 error_setg(errp
, "Cannot change '%s' link from '%s' to '%s'",
4639 i
->backing
->name
, i
->node_name
,
4640 backing_bs(i
)->node_name
);
4649 * Freeze all backing links between @bs and @base.
4650 * If any of the links is already frozen the operation is aborted and
4651 * none of the links are modified.
4652 * @base must be reachable from @bs, or NULL.
4653 * Returns 0 on success. On failure returns < 0 and sets @errp.
4655 int bdrv_freeze_backing_chain(BlockDriverState
*bs
, BlockDriverState
*base
,
4658 BlockDriverState
*i
;
4660 if (bdrv_is_backing_chain_frozen(bs
, base
, errp
)) {
4664 for (i
= bs
; i
!= base
; i
= backing_bs(i
)) {
4665 if (i
->backing
&& backing_bs(i
)->never_freeze
) {
4666 error_setg(errp
, "Cannot freeze '%s' link to '%s'",
4667 i
->backing
->name
, backing_bs(i
)->node_name
);
4672 for (i
= bs
; i
!= base
; i
= backing_bs(i
)) {
4674 i
->backing
->frozen
= true;
4682 * Unfreeze all backing links between @bs and @base. The caller must
4683 * ensure that all links are frozen before using this function.
4684 * @base must be reachable from @bs, or NULL.
4686 void bdrv_unfreeze_backing_chain(BlockDriverState
*bs
, BlockDriverState
*base
)
4688 BlockDriverState
*i
;
4690 for (i
= bs
; i
!= base
; i
= backing_bs(i
)) {
4692 assert(i
->backing
->frozen
);
4693 i
->backing
->frozen
= false;
4699 * Drops images above 'base' up to and including 'top', and sets the image
4700 * above 'top' to have base as its backing file.
4702 * Requires that the overlay to 'top' is opened r/w, so that the backing file
4703 * information in 'bs' can be properly updated.
4705 * E.g., this will convert the following chain:
4706 * bottom <- base <- intermediate <- top <- active
4710 * bottom <- base <- active
4712 * It is allowed for bottom==base, in which case it converts:
4714 * base <- intermediate <- top <- active
4720 * If backing_file_str is non-NULL, it will be used when modifying top's
4721 * overlay image metadata.
4724 * if active == top, that is considered an error
4727 int bdrv_drop_intermediate(BlockDriverState
*top
, BlockDriverState
*base
,
4728 const char *backing_file_str
)
4730 BlockDriverState
*explicit_top
= top
;
4731 bool update_inherits_from
;
4732 BdrvChild
*c
, *next
;
4733 Error
*local_err
= NULL
;
4737 bdrv_subtree_drained_begin(top
);
4739 if (!top
->drv
|| !base
->drv
) {
4743 /* Make sure that base is in the backing chain of top */
4744 if (!bdrv_chain_contains(top
, base
)) {
4748 /* This function changes all links that point to top and makes
4749 * them point to base. Check that none of them is frozen. */
4750 QLIST_FOREACH(c
, &top
->parents
, next_parent
) {
4756 /* If 'base' recursively inherits from 'top' then we should set
4757 * base->inherits_from to top->inherits_from after 'top' and all
4758 * other intermediate nodes have been dropped.
4759 * If 'top' is an implicit node (e.g. "commit_top") we should skip
4760 * it because no one inherits from it. We use explicit_top for that. */
4761 while (explicit_top
&& explicit_top
->implicit
) {
4762 explicit_top
= backing_bs(explicit_top
);
4764 update_inherits_from
= bdrv_inherits_from_recursive(base
, explicit_top
);
4766 /* success - we can delete the intermediate states, and link top->base */
4767 /* TODO Check graph modification op blockers (BLK_PERM_GRAPH_MOD) once
4768 * we've figured out how they should work. */
4769 if (!backing_file_str
) {
4770 bdrv_refresh_filename(base
);
4771 backing_file_str
= base
->filename
;
4774 QLIST_FOREACH_SAFE(c
, &top
->parents
, next_parent
, next
) {
4775 /* Check whether we are allowed to switch c from top to base */
4776 GSList
*ignore_children
= g_slist_prepend(NULL
, c
);
4777 ret
= bdrv_check_update_perm(base
, NULL
, c
->perm
, c
->shared_perm
,
4778 ignore_children
, NULL
, &local_err
);
4779 g_slist_free(ignore_children
);
4781 error_report_err(local_err
);
4785 /* If so, update the backing file path in the image file */
4786 if (c
->klass
->update_filename
) {
4787 ret
= c
->klass
->update_filename(c
, base
, backing_file_str
,
4790 bdrv_abort_perm_update(base
);
4791 error_report_err(local_err
);
4796 /* Do the actual switch in the in-memory graph.
4797 * Completes bdrv_check_update_perm() transaction internally. */
4799 bdrv_replace_child(c
, base
);
4803 if (update_inherits_from
) {
4804 base
->inherits_from
= explicit_top
->inherits_from
;
4809 bdrv_subtree_drained_end(top
);
4815 * Length of a allocated file in bytes. Sparse files are counted by actual
4816 * allocated space. Return < 0 if error or unknown.
4818 int64_t bdrv_get_allocated_file_size(BlockDriverState
*bs
)
4820 BlockDriver
*drv
= bs
->drv
;
4824 if (drv
->bdrv_get_allocated_file_size
) {
4825 return drv
->bdrv_get_allocated_file_size(bs
);
4828 return bdrv_get_allocated_file_size(bs
->file
->bs
);
4835 * @drv: Format driver
4836 * @opts: Creation options for new image
4837 * @in_bs: Existing image containing data for new image (may be NULL)
4838 * @errp: Error object
4839 * Returns: A #BlockMeasureInfo (free using qapi_free_BlockMeasureInfo())
4842 * Calculate file size required to create a new image.
4844 * If @in_bs is given then space for allocated clusters and zero clusters
4845 * from that image are included in the calculation. If @opts contains a
4846 * backing file that is shared by @in_bs then backing clusters may be omitted
4847 * from the calculation.
4849 * If @in_bs is NULL then the calculation includes no allocated clusters
4850 * unless a preallocation option is given in @opts.
4852 * Note that @in_bs may use a different BlockDriver from @drv.
4854 * If an error occurs the @errp pointer is set.
4856 BlockMeasureInfo
*bdrv_measure(BlockDriver
*drv
, QemuOpts
*opts
,
4857 BlockDriverState
*in_bs
, Error
**errp
)
4859 if (!drv
->bdrv_measure
) {
4860 error_setg(errp
, "Block driver '%s' does not support size measurement",
4865 return drv
->bdrv_measure(opts
, in_bs
, errp
);
4869 * Return number of sectors on success, -errno on error.
4871 int64_t bdrv_nb_sectors(BlockDriverState
*bs
)
4873 BlockDriver
*drv
= bs
->drv
;
4878 if (drv
->has_variable_length
) {
4879 int ret
= refresh_total_sectors(bs
, bs
->total_sectors
);
4884 return bs
->total_sectors
;
4888 * Return length in bytes on success, -errno on error.
4889 * The length is always a multiple of BDRV_SECTOR_SIZE.
4891 int64_t bdrv_getlength(BlockDriverState
*bs
)
4893 int64_t ret
= bdrv_nb_sectors(bs
);
4895 ret
= ret
> INT64_MAX
/ BDRV_SECTOR_SIZE
? -EFBIG
: ret
;
4896 return ret
< 0 ? ret
: ret
* BDRV_SECTOR_SIZE
;
4899 /* return 0 as number of sectors if no device present or error */
4900 void bdrv_get_geometry(BlockDriverState
*bs
, uint64_t *nb_sectors_ptr
)
4902 int64_t nb_sectors
= bdrv_nb_sectors(bs
);
4904 *nb_sectors_ptr
= nb_sectors
< 0 ? 0 : nb_sectors
;
4907 bool bdrv_is_sg(BlockDriverState
*bs
)
4912 bool bdrv_is_encrypted(BlockDriverState
*bs
)
4914 if (bs
->backing
&& bs
->backing
->bs
->encrypted
) {
4917 return bs
->encrypted
;
4920 const char *bdrv_get_format_name(BlockDriverState
*bs
)
4922 return bs
->drv
? bs
->drv
->format_name
: NULL
;
4925 static int qsort_strcmp(const void *a
, const void *b
)
4927 return strcmp(*(char *const *)a
, *(char *const *)b
);
4930 void bdrv_iterate_format(void (*it
)(void *opaque
, const char *name
),
4931 void *opaque
, bool read_only
)
4936 const char **formats
= NULL
;
4938 QLIST_FOREACH(drv
, &bdrv_drivers
, list
) {
4939 if (drv
->format_name
) {
4943 if (use_bdrv_whitelist
&& !bdrv_is_whitelisted(drv
, read_only
)) {
4947 while (formats
&& i
&& !found
) {
4948 found
= !strcmp(formats
[--i
], drv
->format_name
);
4952 formats
= g_renew(const char *, formats
, count
+ 1);
4953 formats
[count
++] = drv
->format_name
;
4958 for (i
= 0; i
< (int)ARRAY_SIZE(block_driver_modules
); i
++) {
4959 const char *format_name
= block_driver_modules
[i
].format_name
;
4965 if (use_bdrv_whitelist
&&
4966 !bdrv_format_is_whitelisted(format_name
, read_only
)) {
4970 while (formats
&& j
&& !found
) {
4971 found
= !strcmp(formats
[--j
], format_name
);
4975 formats
= g_renew(const char *, formats
, count
+ 1);
4976 formats
[count
++] = format_name
;
4981 qsort(formats
, count
, sizeof(formats
[0]), qsort_strcmp
);
4983 for (i
= 0; i
< count
; i
++) {
4984 it(opaque
, formats
[i
]);
4990 /* This function is to find a node in the bs graph */
4991 BlockDriverState
*bdrv_find_node(const char *node_name
)
4993 BlockDriverState
*bs
;
4997 QTAILQ_FOREACH(bs
, &graph_bdrv_states
, node_list
) {
4998 if (!strcmp(node_name
, bs
->node_name
)) {
5005 /* Put this QMP function here so it can access the static graph_bdrv_states. */
5006 BlockDeviceInfoList
*bdrv_named_nodes_list(bool flat
,
5009 BlockDeviceInfoList
*list
, *entry
;
5010 BlockDriverState
*bs
;
5013 QTAILQ_FOREACH(bs
, &graph_bdrv_states
, node_list
) {
5014 BlockDeviceInfo
*info
= bdrv_block_device_info(NULL
, bs
, flat
, errp
);
5016 qapi_free_BlockDeviceInfoList(list
);
5019 entry
= g_malloc0(sizeof(*entry
));
5020 entry
->value
= info
;
5028 #define QAPI_LIST_ADD(list, element) do { \
5029 typeof(list) _tmp = g_new(typeof(*(list)), 1); \
5030 _tmp->value = (element); \
5031 _tmp->next = (list); \
5035 typedef struct XDbgBlockGraphConstructor
{
5036 XDbgBlockGraph
*graph
;
5037 GHashTable
*graph_nodes
;
5038 } XDbgBlockGraphConstructor
;
5040 static XDbgBlockGraphConstructor
*xdbg_graph_new(void)
5042 XDbgBlockGraphConstructor
*gr
= g_new(XDbgBlockGraphConstructor
, 1);
5044 gr
->graph
= g_new0(XDbgBlockGraph
, 1);
5045 gr
->graph_nodes
= g_hash_table_new(NULL
, NULL
);
5050 static XDbgBlockGraph
*xdbg_graph_finalize(XDbgBlockGraphConstructor
*gr
)
5052 XDbgBlockGraph
*graph
= gr
->graph
;
5054 g_hash_table_destroy(gr
->graph_nodes
);
5060 static uintptr_t xdbg_graph_node_num(XDbgBlockGraphConstructor
*gr
, void *node
)
5062 uintptr_t ret
= (uintptr_t)g_hash_table_lookup(gr
->graph_nodes
, node
);
5069 * Start counting from 1, not 0, because 0 interferes with not-found (NULL)
5070 * answer of g_hash_table_lookup.
5072 ret
= g_hash_table_size(gr
->graph_nodes
) + 1;
5073 g_hash_table_insert(gr
->graph_nodes
, node
, (void *)ret
);
5078 static void xdbg_graph_add_node(XDbgBlockGraphConstructor
*gr
, void *node
,
5079 XDbgBlockGraphNodeType type
, const char *name
)
5081 XDbgBlockGraphNode
*n
;
5083 n
= g_new0(XDbgBlockGraphNode
, 1);
5085 n
->id
= xdbg_graph_node_num(gr
, node
);
5087 n
->name
= g_strdup(name
);
5089 QAPI_LIST_ADD(gr
->graph
->nodes
, n
);
5092 static void xdbg_graph_add_edge(XDbgBlockGraphConstructor
*gr
, void *parent
,
5093 const BdrvChild
*child
)
5095 BlockPermission qapi_perm
;
5096 XDbgBlockGraphEdge
*edge
;
5098 edge
= g_new0(XDbgBlockGraphEdge
, 1);
5100 edge
->parent
= xdbg_graph_node_num(gr
, parent
);
5101 edge
->child
= xdbg_graph_node_num(gr
, child
->bs
);
5102 edge
->name
= g_strdup(child
->name
);
5104 for (qapi_perm
= 0; qapi_perm
< BLOCK_PERMISSION__MAX
; qapi_perm
++) {
5105 uint64_t flag
= bdrv_qapi_perm_to_blk_perm(qapi_perm
);
5107 if (flag
& child
->perm
) {
5108 QAPI_LIST_ADD(edge
->perm
, qapi_perm
);
5110 if (flag
& child
->shared_perm
) {
5111 QAPI_LIST_ADD(edge
->shared_perm
, qapi_perm
);
5115 QAPI_LIST_ADD(gr
->graph
->edges
, edge
);
5119 XDbgBlockGraph
*bdrv_get_xdbg_block_graph(Error
**errp
)
5123 BlockDriverState
*bs
;
5125 XDbgBlockGraphConstructor
*gr
= xdbg_graph_new();
5127 for (blk
= blk_all_next(NULL
); blk
; blk
= blk_all_next(blk
)) {
5128 char *allocated_name
= NULL
;
5129 const char *name
= blk_name(blk
);
5132 name
= allocated_name
= blk_get_attached_dev_id(blk
);
5134 xdbg_graph_add_node(gr
, blk
, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_BACKEND
,
5136 g_free(allocated_name
);
5137 if (blk_root(blk
)) {
5138 xdbg_graph_add_edge(gr
, blk
, blk_root(blk
));
5142 for (job
= block_job_next(NULL
); job
; job
= block_job_next(job
)) {
5145 xdbg_graph_add_node(gr
, job
, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_JOB
,
5147 for (el
= job
->nodes
; el
; el
= el
->next
) {
5148 xdbg_graph_add_edge(gr
, job
, (BdrvChild
*)el
->data
);
5152 QTAILQ_FOREACH(bs
, &graph_bdrv_states
, node_list
) {
5153 xdbg_graph_add_node(gr
, bs
, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_DRIVER
,
5155 QLIST_FOREACH(child
, &bs
->children
, next
) {
5156 xdbg_graph_add_edge(gr
, bs
, child
);
5160 return xdbg_graph_finalize(gr
);
5163 BlockDriverState
*bdrv_lookup_bs(const char *device
,
5164 const char *node_name
,
5168 BlockDriverState
*bs
;
5171 blk
= blk_by_name(device
);
5176 error_setg(errp
, "Device '%s' has no medium", device
);
5184 bs
= bdrv_find_node(node_name
);
5191 error_setg(errp
, "Cannot find device=%s nor node_name=%s",
5192 device
? device
: "",
5193 node_name
? node_name
: "");
5197 /* If 'base' is in the same chain as 'top', return true. Otherwise,
5198 * return false. If either argument is NULL, return false. */
5199 bool bdrv_chain_contains(BlockDriverState
*top
, BlockDriverState
*base
)
5201 while (top
&& top
!= base
) {
5202 top
= backing_bs(top
);
5208 BlockDriverState
*bdrv_next_node(BlockDriverState
*bs
)
5211 return QTAILQ_FIRST(&graph_bdrv_states
);
5213 return QTAILQ_NEXT(bs
, node_list
);
5216 BlockDriverState
*bdrv_next_all_states(BlockDriverState
*bs
)
5219 return QTAILQ_FIRST(&all_bdrv_states
);
5221 return QTAILQ_NEXT(bs
, bs_list
);
5224 const char *bdrv_get_node_name(const BlockDriverState
*bs
)
5226 return bs
->node_name
;
5229 const char *bdrv_get_parent_name(const BlockDriverState
*bs
)
5234 /* If multiple parents have a name, just pick the first one. */
5235 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
5236 if (c
->klass
->get_name
) {
5237 name
= c
->klass
->get_name(c
);
5238 if (name
&& *name
) {
5247 /* TODO check what callers really want: bs->node_name or blk_name() */
5248 const char *bdrv_get_device_name(const BlockDriverState
*bs
)
5250 return bdrv_get_parent_name(bs
) ?: "";
5253 /* This can be used to identify nodes that might not have a device
5254 * name associated. Since node and device names live in the same
5255 * namespace, the result is unambiguous. The exception is if both are
5256 * absent, then this returns an empty (non-null) string. */
5257 const char *bdrv_get_device_or_node_name(const BlockDriverState
*bs
)
5259 return bdrv_get_parent_name(bs
) ?: bs
->node_name
;
5262 int bdrv_get_flags(BlockDriverState
*bs
)
5264 return bs
->open_flags
;
5267 int bdrv_has_zero_init_1(BlockDriverState
*bs
)
5272 int bdrv_has_zero_init(BlockDriverState
*bs
)
5278 /* If BS is a copy on write image, it is initialized to
5279 the contents of the base image, which may not be zeroes. */
5283 if (bs
->drv
->bdrv_has_zero_init
) {
5284 return bs
->drv
->bdrv_has_zero_init(bs
);
5286 if (bs
->file
&& bs
->drv
->is_filter
) {
5287 return bdrv_has_zero_init(bs
->file
->bs
);
5294 bool bdrv_unallocated_blocks_are_zero(BlockDriverState
*bs
)
5296 BlockDriverInfo bdi
;
5302 if (bdrv_get_info(bs
, &bdi
) == 0) {
5303 return bdi
.unallocated_blocks_are_zero
;
5309 bool bdrv_can_write_zeroes_with_unmap(BlockDriverState
*bs
)
5311 if (!(bs
->open_flags
& BDRV_O_UNMAP
)) {
5315 return bs
->supported_zero_flags
& BDRV_REQ_MAY_UNMAP
;
5318 void bdrv_get_backing_filename(BlockDriverState
*bs
,
5319 char *filename
, int filename_size
)
5321 pstrcpy(filename
, filename_size
, bs
->backing_file
);
5324 int bdrv_get_info(BlockDriverState
*bs
, BlockDriverInfo
*bdi
)
5326 BlockDriver
*drv
= bs
->drv
;
5327 /* if bs->drv == NULL, bs is closed, so there's nothing to do here */
5331 if (!drv
->bdrv_get_info
) {
5332 if (bs
->file
&& drv
->is_filter
) {
5333 return bdrv_get_info(bs
->file
->bs
, bdi
);
5337 memset(bdi
, 0, sizeof(*bdi
));
5338 return drv
->bdrv_get_info(bs
, bdi
);
5341 ImageInfoSpecific
*bdrv_get_specific_info(BlockDriverState
*bs
,
5344 BlockDriver
*drv
= bs
->drv
;
5345 if (drv
&& drv
->bdrv_get_specific_info
) {
5346 return drv
->bdrv_get_specific_info(bs
, errp
);
5351 BlockStatsSpecific
*bdrv_get_specific_stats(BlockDriverState
*bs
)
5353 BlockDriver
*drv
= bs
->drv
;
5354 if (!drv
|| !drv
->bdrv_get_specific_stats
) {
5357 return drv
->bdrv_get_specific_stats(bs
);
5360 void bdrv_debug_event(BlockDriverState
*bs
, BlkdebugEvent event
)
5362 if (!bs
|| !bs
->drv
|| !bs
->drv
->bdrv_debug_event
) {
5366 bs
->drv
->bdrv_debug_event(bs
, event
);
5369 static BlockDriverState
*bdrv_find_debug_node(BlockDriverState
*bs
)
5371 while (bs
&& bs
->drv
&& !bs
->drv
->bdrv_debug_breakpoint
) {
5377 if (bs
->drv
->is_filter
&& bs
->backing
) {
5378 bs
= bs
->backing
->bs
;
5385 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_breakpoint
) {
5386 assert(bs
->drv
->bdrv_debug_remove_breakpoint
);
5393 int bdrv_debug_breakpoint(BlockDriverState
*bs
, const char *event
,
5396 bs
= bdrv_find_debug_node(bs
);
5398 return bs
->drv
->bdrv_debug_breakpoint(bs
, event
, tag
);
5404 int bdrv_debug_remove_breakpoint(BlockDriverState
*bs
, const char *tag
)
5406 bs
= bdrv_find_debug_node(bs
);
5408 return bs
->drv
->bdrv_debug_remove_breakpoint(bs
, tag
);
5414 int bdrv_debug_resume(BlockDriverState
*bs
, const char *tag
)
5416 while (bs
&& (!bs
->drv
|| !bs
->drv
->bdrv_debug_resume
)) {
5417 bs
= bs
->file
? bs
->file
->bs
: NULL
;
5420 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_resume
) {
5421 return bs
->drv
->bdrv_debug_resume(bs
, tag
);
5427 bool bdrv_debug_is_suspended(BlockDriverState
*bs
, const char *tag
)
5429 while (bs
&& bs
->drv
&& !bs
->drv
->bdrv_debug_is_suspended
) {
5430 bs
= bs
->file
? bs
->file
->bs
: NULL
;
5433 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_is_suspended
) {
5434 return bs
->drv
->bdrv_debug_is_suspended(bs
, tag
);
5440 /* backing_file can either be relative, or absolute, or a protocol. If it is
5441 * relative, it must be relative to the chain. So, passing in bs->filename
5442 * from a BDS as backing_file should not be done, as that may be relative to
5443 * the CWD rather than the chain. */
5444 BlockDriverState
*bdrv_find_backing_image(BlockDriverState
*bs
,
5445 const char *backing_file
)
5447 char *filename_full
= NULL
;
5448 char *backing_file_full
= NULL
;
5449 char *filename_tmp
= NULL
;
5450 int is_protocol
= 0;
5451 BlockDriverState
*curr_bs
= NULL
;
5452 BlockDriverState
*retval
= NULL
;
5454 if (!bs
|| !bs
->drv
|| !backing_file
) {
5458 filename_full
= g_malloc(PATH_MAX
);
5459 backing_file_full
= g_malloc(PATH_MAX
);
5461 is_protocol
= path_has_protocol(backing_file
);
5463 for (curr_bs
= bs
; curr_bs
->backing
; curr_bs
= curr_bs
->backing
->bs
) {
5465 /* If either of the filename paths is actually a protocol, then
5466 * compare unmodified paths; otherwise make paths relative */
5467 if (is_protocol
|| path_has_protocol(curr_bs
->backing_file
)) {
5468 char *backing_file_full_ret
;
5470 if (strcmp(backing_file
, curr_bs
->backing_file
) == 0) {
5471 retval
= curr_bs
->backing
->bs
;
5474 /* Also check against the full backing filename for the image */
5475 backing_file_full_ret
= bdrv_get_full_backing_filename(curr_bs
,
5477 if (backing_file_full_ret
) {
5478 bool equal
= strcmp(backing_file
, backing_file_full_ret
) == 0;
5479 g_free(backing_file_full_ret
);
5481 retval
= curr_bs
->backing
->bs
;
5486 /* If not an absolute filename path, make it relative to the current
5487 * image's filename path */
5488 filename_tmp
= bdrv_make_absolute_filename(curr_bs
, backing_file
,
5490 /* We are going to compare canonicalized absolute pathnames */
5491 if (!filename_tmp
|| !realpath(filename_tmp
, filename_full
)) {
5492 g_free(filename_tmp
);
5495 g_free(filename_tmp
);
5497 /* We need to make sure the backing filename we are comparing against
5498 * is relative to the current image filename (or absolute) */
5499 filename_tmp
= bdrv_get_full_backing_filename(curr_bs
, NULL
);
5500 if (!filename_tmp
|| !realpath(filename_tmp
, backing_file_full
)) {
5501 g_free(filename_tmp
);
5504 g_free(filename_tmp
);
5506 if (strcmp(backing_file_full
, filename_full
) == 0) {
5507 retval
= curr_bs
->backing
->bs
;
5513 g_free(filename_full
);
5514 g_free(backing_file_full
);
5518 void bdrv_init(void)
5520 module_call_init(MODULE_INIT_BLOCK
);
5523 void bdrv_init_with_whitelist(void)
5525 use_bdrv_whitelist
= 1;
5529 static void coroutine_fn
bdrv_co_invalidate_cache(BlockDriverState
*bs
,
5532 BdrvChild
*child
, *parent
;
5533 uint64_t perm
, shared_perm
;
5534 Error
*local_err
= NULL
;
5536 BdrvDirtyBitmap
*bm
;
5542 QLIST_FOREACH(child
, &bs
->children
, next
) {
5543 bdrv_co_invalidate_cache(child
->bs
, &local_err
);
5545 error_propagate(errp
, local_err
);
5551 * Update permissions, they may differ for inactive nodes.
5553 * Note that the required permissions of inactive images are always a
5554 * subset of the permissions required after activating the image. This
5555 * allows us to just get the permissions upfront without restricting
5556 * drv->bdrv_invalidate_cache().
5558 * It also means that in error cases, we don't have to try and revert to
5559 * the old permissions (which is an operation that could fail, too). We can
5560 * just keep the extended permissions for the next time that an activation
5561 * of the image is tried.
5563 if (bs
->open_flags
& BDRV_O_INACTIVE
) {
5564 bs
->open_flags
&= ~BDRV_O_INACTIVE
;
5565 bdrv_get_cumulative_perm(bs
, &perm
, &shared_perm
);
5566 ret
= bdrv_check_perm(bs
, NULL
, perm
, shared_perm
, NULL
, NULL
, &local_err
);
5568 bs
->open_flags
|= BDRV_O_INACTIVE
;
5569 error_propagate(errp
, local_err
);
5572 bdrv_set_perm(bs
, perm
, shared_perm
);
5574 if (bs
->drv
->bdrv_co_invalidate_cache
) {
5575 bs
->drv
->bdrv_co_invalidate_cache(bs
, &local_err
);
5577 bs
->open_flags
|= BDRV_O_INACTIVE
;
5578 error_propagate(errp
, local_err
);
5583 FOR_EACH_DIRTY_BITMAP(bs
, bm
) {
5584 bdrv_dirty_bitmap_skip_store(bm
, false);
5587 ret
= refresh_total_sectors(bs
, bs
->total_sectors
);
5589 bs
->open_flags
|= BDRV_O_INACTIVE
;
5590 error_setg_errno(errp
, -ret
, "Could not refresh total sector count");
5595 QLIST_FOREACH(parent
, &bs
->parents
, next_parent
) {
5596 if (parent
->klass
->activate
) {
5597 parent
->klass
->activate(parent
, &local_err
);
5599 bs
->open_flags
|= BDRV_O_INACTIVE
;
5600 error_propagate(errp
, local_err
);
5607 typedef struct InvalidateCacheCo
{
5608 BlockDriverState
*bs
;
5611 } InvalidateCacheCo
;
5613 static void coroutine_fn
bdrv_invalidate_cache_co_entry(void *opaque
)
5615 InvalidateCacheCo
*ico
= opaque
;
5616 bdrv_co_invalidate_cache(ico
->bs
, ico
->errp
);
5621 void bdrv_invalidate_cache(BlockDriverState
*bs
, Error
**errp
)
5624 InvalidateCacheCo ico
= {
5630 if (qemu_in_coroutine()) {
5631 /* Fast-path if already in coroutine context */
5632 bdrv_invalidate_cache_co_entry(&ico
);
5634 co
= qemu_coroutine_create(bdrv_invalidate_cache_co_entry
, &ico
);
5635 bdrv_coroutine_enter(bs
, co
);
5636 BDRV_POLL_WHILE(bs
, !ico
.done
);
5640 void bdrv_invalidate_cache_all(Error
**errp
)
5642 BlockDriverState
*bs
;
5643 Error
*local_err
= NULL
;
5644 BdrvNextIterator it
;
5646 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
5647 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
5649 aio_context_acquire(aio_context
);
5650 bdrv_invalidate_cache(bs
, &local_err
);
5651 aio_context_release(aio_context
);
5653 error_propagate(errp
, local_err
);
5654 bdrv_next_cleanup(&it
);
5660 static bool bdrv_has_bds_parent(BlockDriverState
*bs
, bool only_active
)
5664 QLIST_FOREACH(parent
, &bs
->parents
, next_parent
) {
5665 if (parent
->klass
->parent_is_bds
) {
5666 BlockDriverState
*parent_bs
= parent
->opaque
;
5667 if (!only_active
|| !(parent_bs
->open_flags
& BDRV_O_INACTIVE
)) {
5676 static int bdrv_inactivate_recurse(BlockDriverState
*bs
)
5678 BdrvChild
*child
, *parent
;
5679 bool tighten_restrictions
;
5680 uint64_t perm
, shared_perm
;
5687 /* Make sure that we don't inactivate a child before its parent.
5688 * It will be covered by recursion from the yet active parent. */
5689 if (bdrv_has_bds_parent(bs
, true)) {
5693 assert(!(bs
->open_flags
& BDRV_O_INACTIVE
));
5695 /* Inactivate this node */
5696 if (bs
->drv
->bdrv_inactivate
) {
5697 ret
= bs
->drv
->bdrv_inactivate(bs
);
5703 QLIST_FOREACH(parent
, &bs
->parents
, next_parent
) {
5704 if (parent
->klass
->inactivate
) {
5705 ret
= parent
->klass
->inactivate(parent
);
5712 bs
->open_flags
|= BDRV_O_INACTIVE
;
5714 /* Update permissions, they may differ for inactive nodes */
5715 bdrv_get_cumulative_perm(bs
, &perm
, &shared_perm
);
5716 ret
= bdrv_check_perm(bs
, NULL
, perm
, shared_perm
, NULL
,
5717 &tighten_restrictions
, NULL
);
5718 assert(tighten_restrictions
== false);
5720 /* We only tried to loosen restrictions, so errors are not fatal */
5721 bdrv_abort_perm_update(bs
);
5723 bdrv_set_perm(bs
, perm
, shared_perm
);
5727 /* Recursively inactivate children */
5728 QLIST_FOREACH(child
, &bs
->children
, next
) {
5729 ret
= bdrv_inactivate_recurse(child
->bs
);
5738 int bdrv_inactivate_all(void)
5740 BlockDriverState
*bs
= NULL
;
5741 BdrvNextIterator it
;
5743 GSList
*aio_ctxs
= NULL
, *ctx
;
5745 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
5746 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
5748 if (!g_slist_find(aio_ctxs
, aio_context
)) {
5749 aio_ctxs
= g_slist_prepend(aio_ctxs
, aio_context
);
5750 aio_context_acquire(aio_context
);
5754 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
5755 /* Nodes with BDS parents are covered by recursion from the last
5756 * parent that gets inactivated. Don't inactivate them a second
5757 * time if that has already happened. */
5758 if (bdrv_has_bds_parent(bs
, false)) {
5761 ret
= bdrv_inactivate_recurse(bs
);
5763 bdrv_next_cleanup(&it
);
5769 for (ctx
= aio_ctxs
; ctx
!= NULL
; ctx
= ctx
->next
) {
5770 AioContext
*aio_context
= ctx
->data
;
5771 aio_context_release(aio_context
);
5773 g_slist_free(aio_ctxs
);
5778 /**************************************************************/
5779 /* removable device support */
5782 * Return TRUE if the media is present
5784 bool bdrv_is_inserted(BlockDriverState
*bs
)
5786 BlockDriver
*drv
= bs
->drv
;
5792 if (drv
->bdrv_is_inserted
) {
5793 return drv
->bdrv_is_inserted(bs
);
5795 QLIST_FOREACH(child
, &bs
->children
, next
) {
5796 if (!bdrv_is_inserted(child
->bs
)) {
5804 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
5806 void bdrv_eject(BlockDriverState
*bs
, bool eject_flag
)
5808 BlockDriver
*drv
= bs
->drv
;
5810 if (drv
&& drv
->bdrv_eject
) {
5811 drv
->bdrv_eject(bs
, eject_flag
);
5816 * Lock or unlock the media (if it is locked, the user won't be able
5817 * to eject it manually).
5819 void bdrv_lock_medium(BlockDriverState
*bs
, bool locked
)
5821 BlockDriver
*drv
= bs
->drv
;
5823 trace_bdrv_lock_medium(bs
, locked
);
5825 if (drv
&& drv
->bdrv_lock_medium
) {
5826 drv
->bdrv_lock_medium(bs
, locked
);
5830 /* Get a reference to bs */
5831 void bdrv_ref(BlockDriverState
*bs
)
5836 /* Release a previously grabbed reference to bs.
5837 * If after releasing, reference count is zero, the BlockDriverState is
5839 void bdrv_unref(BlockDriverState
*bs
)
5844 assert(bs
->refcnt
> 0);
5845 if (--bs
->refcnt
== 0) {
5850 struct BdrvOpBlocker
{
5852 QLIST_ENTRY(BdrvOpBlocker
) list
;
5855 bool bdrv_op_is_blocked(BlockDriverState
*bs
, BlockOpType op
, Error
**errp
)
5857 BdrvOpBlocker
*blocker
;
5858 assert((int) op
>= 0 && op
< BLOCK_OP_TYPE_MAX
);
5859 if (!QLIST_EMPTY(&bs
->op_blockers
[op
])) {
5860 blocker
= QLIST_FIRST(&bs
->op_blockers
[op
]);
5861 error_propagate_prepend(errp
, error_copy(blocker
->reason
),
5862 "Node '%s' is busy: ",
5863 bdrv_get_device_or_node_name(bs
));
5869 void bdrv_op_block(BlockDriverState
*bs
, BlockOpType op
, Error
*reason
)
5871 BdrvOpBlocker
*blocker
;
5872 assert((int) op
>= 0 && op
< BLOCK_OP_TYPE_MAX
);
5874 blocker
= g_new0(BdrvOpBlocker
, 1);
5875 blocker
->reason
= reason
;
5876 QLIST_INSERT_HEAD(&bs
->op_blockers
[op
], blocker
, list
);
5879 void bdrv_op_unblock(BlockDriverState
*bs
, BlockOpType op
, Error
*reason
)
5881 BdrvOpBlocker
*blocker
, *next
;
5882 assert((int) op
>= 0 && op
< BLOCK_OP_TYPE_MAX
);
5883 QLIST_FOREACH_SAFE(blocker
, &bs
->op_blockers
[op
], list
, next
) {
5884 if (blocker
->reason
== reason
) {
5885 QLIST_REMOVE(blocker
, list
);
5891 void bdrv_op_block_all(BlockDriverState
*bs
, Error
*reason
)
5894 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
5895 bdrv_op_block(bs
, i
, reason
);
5899 void bdrv_op_unblock_all(BlockDriverState
*bs
, Error
*reason
)
5902 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
5903 bdrv_op_unblock(bs
, i
, reason
);
5907 bool bdrv_op_blocker_is_empty(BlockDriverState
*bs
)
5911 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
5912 if (!QLIST_EMPTY(&bs
->op_blockers
[i
])) {
5919 void bdrv_img_create(const char *filename
, const char *fmt
,
5920 const char *base_filename
, const char *base_fmt
,
5921 char *options
, uint64_t img_size
, int flags
, bool quiet
,
5924 QemuOptsList
*create_opts
= NULL
;
5925 QemuOpts
*opts
= NULL
;
5926 const char *backing_fmt
, *backing_file
;
5928 BlockDriver
*drv
, *proto_drv
;
5929 Error
*local_err
= NULL
;
5932 /* Find driver and parse its options */
5933 drv
= bdrv_find_format(fmt
);
5935 error_setg(errp
, "Unknown file format '%s'", fmt
);
5939 proto_drv
= bdrv_find_protocol(filename
, true, errp
);
5944 if (!drv
->create_opts
) {
5945 error_setg(errp
, "Format driver '%s' does not support image creation",
5950 if (!proto_drv
->create_opts
) {
5951 error_setg(errp
, "Protocol driver '%s' does not support image creation",
5952 proto_drv
->format_name
);
5956 /* Create parameter list */
5957 create_opts
= qemu_opts_append(create_opts
, drv
->create_opts
);
5958 create_opts
= qemu_opts_append(create_opts
, proto_drv
->create_opts
);
5960 opts
= qemu_opts_create(create_opts
, NULL
, 0, &error_abort
);
5962 /* Parse -o options */
5964 qemu_opts_do_parse(opts
, options
, NULL
, &local_err
);
5970 if (!qemu_opt_get(opts
, BLOCK_OPT_SIZE
)) {
5971 qemu_opt_set_number(opts
, BLOCK_OPT_SIZE
, img_size
, &error_abort
);
5972 } else if (img_size
!= UINT64_C(-1)) {
5973 error_setg(errp
, "The image size must be specified only once");
5977 if (base_filename
) {
5978 qemu_opt_set(opts
, BLOCK_OPT_BACKING_FILE
, base_filename
, &local_err
);
5980 error_setg(errp
, "Backing file not supported for file format '%s'",
5987 qemu_opt_set(opts
, BLOCK_OPT_BACKING_FMT
, base_fmt
, &local_err
);
5989 error_setg(errp
, "Backing file format not supported for file "
5990 "format '%s'", fmt
);
5995 backing_file
= qemu_opt_get(opts
, BLOCK_OPT_BACKING_FILE
);
5997 if (!strcmp(filename
, backing_file
)) {
5998 error_setg(errp
, "Error: Trying to create an image with the "
5999 "same filename as the backing file");
6004 backing_fmt
= qemu_opt_get(opts
, BLOCK_OPT_BACKING_FMT
);
6006 /* The size for the image must always be specified, unless we have a backing
6007 * file and we have not been forbidden from opening it. */
6008 size
= qemu_opt_get_size(opts
, BLOCK_OPT_SIZE
, img_size
);
6009 if (backing_file
&& !(flags
& BDRV_O_NO_BACKING
)) {
6010 BlockDriverState
*bs
;
6013 QDict
*backing_options
= NULL
;
6016 bdrv_get_full_backing_filename_from_filename(filename
, backing_file
,
6021 assert(full_backing
);
6023 /* backing files always opened read-only */
6025 back_flags
&= ~(BDRV_O_RDWR
| BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
);
6027 backing_options
= qdict_new();
6029 qdict_put_str(backing_options
, "driver", backing_fmt
);
6031 qdict_put_bool(backing_options
, BDRV_OPT_FORCE_SHARE
, true);
6033 bs
= bdrv_open(full_backing
, NULL
, backing_options
, back_flags
,
6035 g_free(full_backing
);
6036 if (!bs
&& size
!= -1) {
6037 /* Couldn't open BS, but we have a size, so it's nonfatal */
6038 warn_reportf_err(local_err
,
6039 "Could not verify backing image. "
6040 "This may become an error in future versions.\n");
6043 /* Couldn't open bs, do not have size */
6044 error_append_hint(&local_err
,
6045 "Could not open backing image to determine size.\n");
6049 /* Opened BS, have no size */
6050 size
= bdrv_getlength(bs
);
6052 error_setg_errno(errp
, -size
, "Could not get size of '%s'",
6057 qemu_opt_set_number(opts
, BLOCK_OPT_SIZE
, size
, &error_abort
);
6061 } /* (backing_file && !(flags & BDRV_O_NO_BACKING)) */
6064 error_setg(errp
, "Image creation needs a size parameter");
6069 printf("Formatting '%s', fmt=%s ", filename
, fmt
);
6070 qemu_opts_print(opts
, " ");
6074 ret
= bdrv_create(drv
, filename
, opts
, &local_err
);
6076 if (ret
== -EFBIG
) {
6077 /* This is generally a better message than whatever the driver would
6078 * deliver (especially because of the cluster_size_hint), since that
6079 * is most probably not much different from "image too large". */
6080 const char *cluster_size_hint
= "";
6081 if (qemu_opt_get_size(opts
, BLOCK_OPT_CLUSTER_SIZE
, 0)) {
6082 cluster_size_hint
= " (try using a larger cluster size)";
6084 error_setg(errp
, "The image size is too large for file format '%s'"
6085 "%s", fmt
, cluster_size_hint
);
6086 error_free(local_err
);
6091 qemu_opts_del(opts
);
6092 qemu_opts_free(create_opts
);
6093 error_propagate(errp
, local_err
);
6096 AioContext
*bdrv_get_aio_context(BlockDriverState
*bs
)
6098 return bs
? bs
->aio_context
: qemu_get_aio_context();
6101 void bdrv_coroutine_enter(BlockDriverState
*bs
, Coroutine
*co
)
6103 aio_co_enter(bdrv_get_aio_context(bs
), co
);
6106 static void bdrv_do_remove_aio_context_notifier(BdrvAioNotifier
*ban
)
6108 QLIST_REMOVE(ban
, list
);
6112 static void bdrv_detach_aio_context(BlockDriverState
*bs
)
6114 BdrvAioNotifier
*baf
, *baf_tmp
;
6116 assert(!bs
->walking_aio_notifiers
);
6117 bs
->walking_aio_notifiers
= true;
6118 QLIST_FOREACH_SAFE(baf
, &bs
->aio_notifiers
, list
, baf_tmp
) {
6120 bdrv_do_remove_aio_context_notifier(baf
);
6122 baf
->detach_aio_context(baf
->opaque
);
6125 /* Never mind iterating again to check for ->deleted. bdrv_close() will
6126 * remove remaining aio notifiers if we aren't called again.
6128 bs
->walking_aio_notifiers
= false;
6130 if (bs
->drv
&& bs
->drv
->bdrv_detach_aio_context
) {
6131 bs
->drv
->bdrv_detach_aio_context(bs
);
6134 if (bs
->quiesce_counter
) {
6135 aio_enable_external(bs
->aio_context
);
6137 bs
->aio_context
= NULL
;
6140 static void bdrv_attach_aio_context(BlockDriverState
*bs
,
6141 AioContext
*new_context
)
6143 BdrvAioNotifier
*ban
, *ban_tmp
;
6145 if (bs
->quiesce_counter
) {
6146 aio_disable_external(new_context
);
6149 bs
->aio_context
= new_context
;
6151 if (bs
->drv
&& bs
->drv
->bdrv_attach_aio_context
) {
6152 bs
->drv
->bdrv_attach_aio_context(bs
, new_context
);
6155 assert(!bs
->walking_aio_notifiers
);
6156 bs
->walking_aio_notifiers
= true;
6157 QLIST_FOREACH_SAFE(ban
, &bs
->aio_notifiers
, list
, ban_tmp
) {
6159 bdrv_do_remove_aio_context_notifier(ban
);
6161 ban
->attached_aio_context(new_context
, ban
->opaque
);
6164 bs
->walking_aio_notifiers
= false;
6168 * Changes the AioContext used for fd handlers, timers, and BHs by this
6169 * BlockDriverState and all its children and parents.
6171 * Must be called from the main AioContext.
6173 * The caller must own the AioContext lock for the old AioContext of bs, but it
6174 * must not own the AioContext lock for new_context (unless new_context is the
6175 * same as the current context of bs).
6177 * @ignore will accumulate all visited BdrvChild object. The caller is
6178 * responsible for freeing the list afterwards.
6180 void bdrv_set_aio_context_ignore(BlockDriverState
*bs
,
6181 AioContext
*new_context
, GSList
**ignore
)
6183 AioContext
*old_context
= bdrv_get_aio_context(bs
);
6186 g_assert(qemu_get_current_aio_context() == qemu_get_aio_context());
6188 if (old_context
== new_context
) {
6192 bdrv_drained_begin(bs
);
6194 QLIST_FOREACH(child
, &bs
->children
, next
) {
6195 if (g_slist_find(*ignore
, child
)) {
6198 *ignore
= g_slist_prepend(*ignore
, child
);
6199 bdrv_set_aio_context_ignore(child
->bs
, new_context
, ignore
);
6201 QLIST_FOREACH(child
, &bs
->parents
, next_parent
) {
6202 if (g_slist_find(*ignore
, child
)) {
6205 assert(child
->klass
->set_aio_ctx
);
6206 *ignore
= g_slist_prepend(*ignore
, child
);
6207 child
->klass
->set_aio_ctx(child
, new_context
, ignore
);
6210 bdrv_detach_aio_context(bs
);
6212 /* Acquire the new context, if necessary */
6213 if (qemu_get_aio_context() != new_context
) {
6214 aio_context_acquire(new_context
);
6217 bdrv_attach_aio_context(bs
, new_context
);
6220 * If this function was recursively called from
6221 * bdrv_set_aio_context_ignore(), there may be nodes in the
6222 * subtree that have not yet been moved to the new AioContext.
6223 * Release the old one so bdrv_drained_end() can poll them.
6225 if (qemu_get_aio_context() != old_context
) {
6226 aio_context_release(old_context
);
6229 bdrv_drained_end(bs
);
6231 if (qemu_get_aio_context() != old_context
) {
6232 aio_context_acquire(old_context
);
6234 if (qemu_get_aio_context() != new_context
) {
6235 aio_context_release(new_context
);
6239 static bool bdrv_parent_can_set_aio_context(BdrvChild
*c
, AioContext
*ctx
,
6240 GSList
**ignore
, Error
**errp
)
6242 if (g_slist_find(*ignore
, c
)) {
6245 *ignore
= g_slist_prepend(*ignore
, c
);
6248 * A BdrvChildClass that doesn't handle AioContext changes cannot
6249 * tolerate any AioContext changes
6251 if (!c
->klass
->can_set_aio_ctx
) {
6252 char *user
= bdrv_child_user_desc(c
);
6253 error_setg(errp
, "Changing iothreads is not supported by %s", user
);
6257 if (!c
->klass
->can_set_aio_ctx(c
, ctx
, ignore
, errp
)) {
6258 assert(!errp
|| *errp
);
6264 bool bdrv_child_can_set_aio_context(BdrvChild
*c
, AioContext
*ctx
,
6265 GSList
**ignore
, Error
**errp
)
6267 if (g_slist_find(*ignore
, c
)) {
6270 *ignore
= g_slist_prepend(*ignore
, c
);
6271 return bdrv_can_set_aio_context(c
->bs
, ctx
, ignore
, errp
);
6274 /* @ignore will accumulate all visited BdrvChild object. The caller is
6275 * responsible for freeing the list afterwards. */
6276 bool bdrv_can_set_aio_context(BlockDriverState
*bs
, AioContext
*ctx
,
6277 GSList
**ignore
, Error
**errp
)
6281 if (bdrv_get_aio_context(bs
) == ctx
) {
6285 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
6286 if (!bdrv_parent_can_set_aio_context(c
, ctx
, ignore
, errp
)) {
6290 QLIST_FOREACH(c
, &bs
->children
, next
) {
6291 if (!bdrv_child_can_set_aio_context(c
, ctx
, ignore
, errp
)) {
6299 int bdrv_child_try_set_aio_context(BlockDriverState
*bs
, AioContext
*ctx
,
6300 BdrvChild
*ignore_child
, Error
**errp
)
6305 ignore
= ignore_child
? g_slist_prepend(NULL
, ignore_child
) : NULL
;
6306 ret
= bdrv_can_set_aio_context(bs
, ctx
, &ignore
, errp
);
6307 g_slist_free(ignore
);
6313 ignore
= ignore_child
? g_slist_prepend(NULL
, ignore_child
) : NULL
;
6314 bdrv_set_aio_context_ignore(bs
, ctx
, &ignore
);
6315 g_slist_free(ignore
);
6320 int bdrv_try_set_aio_context(BlockDriverState
*bs
, AioContext
*ctx
,
6323 return bdrv_child_try_set_aio_context(bs
, ctx
, NULL
, errp
);
6326 void bdrv_add_aio_context_notifier(BlockDriverState
*bs
,
6327 void (*attached_aio_context
)(AioContext
*new_context
, void *opaque
),
6328 void (*detach_aio_context
)(void *opaque
), void *opaque
)
6330 BdrvAioNotifier
*ban
= g_new(BdrvAioNotifier
, 1);
6331 *ban
= (BdrvAioNotifier
){
6332 .attached_aio_context
= attached_aio_context
,
6333 .detach_aio_context
= detach_aio_context
,
6337 QLIST_INSERT_HEAD(&bs
->aio_notifiers
, ban
, list
);
6340 void bdrv_remove_aio_context_notifier(BlockDriverState
*bs
,
6341 void (*attached_aio_context
)(AioContext
*,
6343 void (*detach_aio_context
)(void *),
6346 BdrvAioNotifier
*ban
, *ban_next
;
6348 QLIST_FOREACH_SAFE(ban
, &bs
->aio_notifiers
, list
, ban_next
) {
6349 if (ban
->attached_aio_context
== attached_aio_context
&&
6350 ban
->detach_aio_context
== detach_aio_context
&&
6351 ban
->opaque
== opaque
&&
6352 ban
->deleted
== false)
6354 if (bs
->walking_aio_notifiers
) {
6355 ban
->deleted
= true;
6357 bdrv_do_remove_aio_context_notifier(ban
);
6366 int bdrv_amend_options(BlockDriverState
*bs
, QemuOpts
*opts
,
6367 BlockDriverAmendStatusCB
*status_cb
, void *cb_opaque
,
6371 error_setg(errp
, "Node is ejected");
6374 if (!bs
->drv
->bdrv_amend_options
) {
6375 error_setg(errp
, "Block driver '%s' does not support option amendment",
6376 bs
->drv
->format_name
);
6379 return bs
->drv
->bdrv_amend_options(bs
, opts
, status_cb
, cb_opaque
, errp
);
6383 * This function checks whether the given @to_replace is allowed to be
6384 * replaced by a node that always shows the same data as @bs. This is
6385 * used for example to verify whether the mirror job can replace
6386 * @to_replace by the target mirrored from @bs.
6387 * To be replaceable, @bs and @to_replace may either be guaranteed to
6388 * always show the same data (because they are only connected through
6389 * filters), or some driver may allow replacing one of its children
6390 * because it can guarantee that this child's data is not visible at
6391 * all (for example, for dissenting quorum children that have no other
6394 bool bdrv_recurse_can_replace(BlockDriverState
*bs
,
6395 BlockDriverState
*to_replace
)
6397 if (!bs
|| !bs
->drv
) {
6401 if (bs
== to_replace
) {
6405 /* See what the driver can do */
6406 if (bs
->drv
->bdrv_recurse_can_replace
) {
6407 return bs
->drv
->bdrv_recurse_can_replace(bs
, to_replace
);
6410 /* For filters without an own implementation, we can recurse on our own */
6411 if (bs
->drv
->is_filter
) {
6412 BdrvChild
*child
= bs
->file
?: bs
->backing
;
6413 return bdrv_recurse_can_replace(child
->bs
, to_replace
);
6421 * Check whether the given @node_name can be replaced by a node that
6422 * has the same data as @parent_bs. If so, return @node_name's BDS;
6425 * @node_name must be a (recursive) *child of @parent_bs (or this
6426 * function will return NULL).
6428 * The result (whether the node can be replaced or not) is only valid
6429 * for as long as no graph or permission changes occur.
6431 BlockDriverState
*check_to_replace_node(BlockDriverState
*parent_bs
,
6432 const char *node_name
, Error
**errp
)
6434 BlockDriverState
*to_replace_bs
= bdrv_find_node(node_name
);
6435 AioContext
*aio_context
;
6437 if (!to_replace_bs
) {
6438 error_setg(errp
, "Node name '%s' not found", node_name
);
6442 aio_context
= bdrv_get_aio_context(to_replace_bs
);
6443 aio_context_acquire(aio_context
);
6445 if (bdrv_op_is_blocked(to_replace_bs
, BLOCK_OP_TYPE_REPLACE
, errp
)) {
6446 to_replace_bs
= NULL
;
6450 /* We don't want arbitrary node of the BDS chain to be replaced only the top
6451 * most non filter in order to prevent data corruption.
6452 * Another benefit is that this tests exclude backing files which are
6453 * blocked by the backing blockers.
6455 if (!bdrv_recurse_can_replace(parent_bs
, to_replace_bs
)) {
6456 error_setg(errp
, "Cannot replace '%s' by a node mirrored from '%s', "
6457 "because it cannot be guaranteed that doing so would not "
6458 "lead to an abrupt change of visible data",
6459 node_name
, parent_bs
->node_name
);
6460 to_replace_bs
= NULL
;
6465 aio_context_release(aio_context
);
6466 return to_replace_bs
;
6470 * Iterates through the list of runtime option keys that are said to
6471 * be "strong" for a BDS. An option is called "strong" if it changes
6472 * a BDS's data. For example, the null block driver's "size" and
6473 * "read-zeroes" options are strong, but its "latency-ns" option is
6476 * If a key returned by this function ends with a dot, all options
6477 * starting with that prefix are strong.
6479 static const char *const *strong_options(BlockDriverState
*bs
,
6480 const char *const *curopt
)
6482 static const char *const global_options
[] = {
6483 "driver", "filename", NULL
6487 return &global_options
[0];
6491 if (curopt
== &global_options
[ARRAY_SIZE(global_options
) - 1] && bs
->drv
) {
6492 curopt
= bs
->drv
->strong_runtime_opts
;
6495 return (curopt
&& *curopt
) ? curopt
: NULL
;
6499 * Copies all strong runtime options from bs->options to the given
6500 * QDict. The set of strong option keys is determined by invoking
6503 * Returns true iff any strong option was present in bs->options (and
6504 * thus copied to the target QDict) with the exception of "filename"
6505 * and "driver". The caller is expected to use this value to decide
6506 * whether the existence of strong options prevents the generation of
6509 static bool append_strong_runtime_options(QDict
*d
, BlockDriverState
*bs
)
6511 bool found_any
= false;
6512 const char *const *option_name
= NULL
;
6518 while ((option_name
= strong_options(bs
, option_name
))) {
6519 bool option_given
= false;
6521 assert(strlen(*option_name
) > 0);
6522 if ((*option_name
)[strlen(*option_name
) - 1] != '.') {
6523 QObject
*entry
= qdict_get(bs
->options
, *option_name
);
6528 qdict_put_obj(d
, *option_name
, qobject_ref(entry
));
6529 option_given
= true;
6531 const QDictEntry
*entry
;
6532 for (entry
= qdict_first(bs
->options
); entry
;
6533 entry
= qdict_next(bs
->options
, entry
))
6535 if (strstart(qdict_entry_key(entry
), *option_name
, NULL
)) {
6536 qdict_put_obj(d
, qdict_entry_key(entry
),
6537 qobject_ref(qdict_entry_value(entry
)));
6538 option_given
= true;
6543 /* While "driver" and "filename" need to be included in a JSON filename,
6544 * their existence does not prohibit generation of a plain filename. */
6545 if (!found_any
&& option_given
&&
6546 strcmp(*option_name
, "driver") && strcmp(*option_name
, "filename"))
6552 if (!qdict_haskey(d
, "driver")) {
6553 /* Drivers created with bdrv_new_open_driver() may not have a
6554 * @driver option. Add it here. */
6555 qdict_put_str(d
, "driver", bs
->drv
->format_name
);
6561 /* Note: This function may return false positives; it may return true
6562 * even if opening the backing file specified by bs's image header
6563 * would result in exactly bs->backing. */
6564 static bool bdrv_backing_overridden(BlockDriverState
*bs
)
6567 return strcmp(bs
->auto_backing_file
,
6568 bs
->backing
->bs
->filename
);
6570 /* No backing BDS, so if the image header reports any backing
6571 * file, it must have been suppressed */
6572 return bs
->auto_backing_file
[0] != '\0';
6576 /* Updates the following BDS fields:
6577 * - exact_filename: A filename which may be used for opening a block device
6578 * which (mostly) equals the given BDS (even without any
6579 * other options; so reading and writing must return the same
6580 * results, but caching etc. may be different)
6581 * - full_open_options: Options which, when given when opening a block device
6582 * (without a filename), result in a BDS (mostly)
6583 * equalling the given one
6584 * - filename: If exact_filename is set, it is copied here. Otherwise,
6585 * full_open_options is converted to a JSON object, prefixed with
6586 * "json:" (for use through the JSON pseudo protocol) and put here.
6588 void bdrv_refresh_filename(BlockDriverState
*bs
)
6590 BlockDriver
*drv
= bs
->drv
;
6593 bool backing_overridden
;
6594 bool generate_json_filename
; /* Whether our default implementation should
6595 fill exact_filename (false) or not (true) */
6601 /* This BDS's file name may depend on any of its children's file names, so
6602 * refresh those first */
6603 QLIST_FOREACH(child
, &bs
->children
, next
) {
6604 bdrv_refresh_filename(child
->bs
);
6608 /* For implicit nodes, just copy everything from the single child */
6609 child
= QLIST_FIRST(&bs
->children
);
6610 assert(QLIST_NEXT(child
, next
) == NULL
);
6612 pstrcpy(bs
->exact_filename
, sizeof(bs
->exact_filename
),
6613 child
->bs
->exact_filename
);
6614 pstrcpy(bs
->filename
, sizeof(bs
->filename
), child
->bs
->filename
);
6616 qobject_unref(bs
->full_open_options
);
6617 bs
->full_open_options
= qobject_ref(child
->bs
->full_open_options
);
6622 backing_overridden
= bdrv_backing_overridden(bs
);
6624 if (bs
->open_flags
& BDRV_O_NO_IO
) {
6625 /* Without I/O, the backing file does not change anything.
6626 * Therefore, in such a case (primarily qemu-img), we can
6627 * pretend the backing file has not been overridden even if
6628 * it technically has been. */
6629 backing_overridden
= false;
6632 /* Gather the options QDict */
6634 generate_json_filename
= append_strong_runtime_options(opts
, bs
);
6635 generate_json_filename
|= backing_overridden
;
6637 if (drv
->bdrv_gather_child_options
) {
6638 /* Some block drivers may not want to present all of their children's
6639 * options, or name them differently from BdrvChild.name */
6640 drv
->bdrv_gather_child_options(bs
, opts
, backing_overridden
);
6642 QLIST_FOREACH(child
, &bs
->children
, next
) {
6643 if (child
->klass
== &child_backing
&& !backing_overridden
) {
6644 /* We can skip the backing BDS if it has not been overridden */
6648 qdict_put(opts
, child
->name
,
6649 qobject_ref(child
->bs
->full_open_options
));
6652 if (backing_overridden
&& !bs
->backing
) {
6653 /* Force no backing file */
6654 qdict_put_null(opts
, "backing");
6658 qobject_unref(bs
->full_open_options
);
6659 bs
->full_open_options
= opts
;
6661 if (drv
->bdrv_refresh_filename
) {
6662 /* Obsolete information is of no use here, so drop the old file name
6663 * information before refreshing it */
6664 bs
->exact_filename
[0] = '\0';
6666 drv
->bdrv_refresh_filename(bs
);
6667 } else if (bs
->file
) {
6668 /* Try to reconstruct valid information from the underlying file */
6670 bs
->exact_filename
[0] = '\0';
6673 * We can use the underlying file's filename if:
6674 * - it has a filename,
6675 * - the file is a protocol BDS, and
6676 * - opening that file (as this BDS's format) will automatically create
6677 * the BDS tree we have right now, that is:
6678 * - the user did not significantly change this BDS's behavior with
6679 * some explicit (strong) options
6680 * - no non-file child of this BDS has been overridden by the user
6681 * Both of these conditions are represented by generate_json_filename.
6683 if (bs
->file
->bs
->exact_filename
[0] &&
6684 bs
->file
->bs
->drv
->bdrv_file_open
&&
6685 !generate_json_filename
)
6687 strcpy(bs
->exact_filename
, bs
->file
->bs
->exact_filename
);
6691 if (bs
->exact_filename
[0]) {
6692 pstrcpy(bs
->filename
, sizeof(bs
->filename
), bs
->exact_filename
);
6694 QString
*json
= qobject_to_json(QOBJECT(bs
->full_open_options
));
6695 snprintf(bs
->filename
, sizeof(bs
->filename
), "json:%s",
6696 qstring_get_str(json
));
6697 qobject_unref(json
);
6701 char *bdrv_dirname(BlockDriverState
*bs
, Error
**errp
)
6703 BlockDriver
*drv
= bs
->drv
;
6706 error_setg(errp
, "Node '%s' is ejected", bs
->node_name
);
6710 if (drv
->bdrv_dirname
) {
6711 return drv
->bdrv_dirname(bs
, errp
);
6715 return bdrv_dirname(bs
->file
->bs
, errp
);
6718 bdrv_refresh_filename(bs
);
6719 if (bs
->exact_filename
[0] != '\0') {
6720 return path_combine(bs
->exact_filename
, "");
6723 error_setg(errp
, "Cannot generate a base directory for %s nodes",
6729 * Hot add/remove a BDS's child. So the user can take a child offline when
6730 * it is broken and take a new child online
6732 void bdrv_add_child(BlockDriverState
*parent_bs
, BlockDriverState
*child_bs
,
6736 if (!parent_bs
->drv
|| !parent_bs
->drv
->bdrv_add_child
) {
6737 error_setg(errp
, "The node %s does not support adding a child",
6738 bdrv_get_device_or_node_name(parent_bs
));
6742 if (!QLIST_EMPTY(&child_bs
->parents
)) {
6743 error_setg(errp
, "The node %s already has a parent",
6744 child_bs
->node_name
);
6748 parent_bs
->drv
->bdrv_add_child(parent_bs
, child_bs
, errp
);
6751 void bdrv_del_child(BlockDriverState
*parent_bs
, BdrvChild
*child
, Error
**errp
)
6755 if (!parent_bs
->drv
|| !parent_bs
->drv
->bdrv_del_child
) {
6756 error_setg(errp
, "The node %s does not support removing a child",
6757 bdrv_get_device_or_node_name(parent_bs
));
6761 QLIST_FOREACH(tmp
, &parent_bs
->children
, next
) {
6768 error_setg(errp
, "The node %s does not have a child named %s",
6769 bdrv_get_device_or_node_name(parent_bs
),
6770 bdrv_get_device_or_node_name(child
->bs
));
6774 parent_bs
->drv
->bdrv_del_child(parent_bs
, child
, errp
);
6777 int bdrv_make_empty(BdrvChild
*c
, Error
**errp
)
6779 BlockDriver
*drv
= c
->bs
->drv
;
6782 assert(c
->perm
& (BLK_PERM_WRITE
| BLK_PERM_WRITE_UNCHANGED
));
6784 if (!drv
->bdrv_make_empty
) {
6785 error_setg(errp
, "%s does not support emptying nodes",
6790 ret
= drv
->bdrv_make_empty(c
->bs
);
6792 error_setg_errno(errp
, -ret
, "Failed to empty %s",