2 * QEMU System Emulator block driver
4 * Copyright (c) 2003 Fabrice Bellard
6 * Permission is hereby granted, free of charge, to any person obtaining a copy
7 * of this software and associated documentation files (the "Software"), to deal
8 * in the Software without restriction, including without limitation the rights
9 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10 * copies of the Software, and to permit persons to whom the Software is
11 * furnished to do so, subject to the following conditions:
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
25 #include "qemu/osdep.h"
26 #include "block/trace.h"
27 #include "block/block_int.h"
28 #include "block/blockjob.h"
29 #include "block/nbd.h"
30 #include "block/qdict.h"
31 #include "qemu/error-report.h"
32 #include "module_block.h"
33 #include "qemu/main-loop.h"
34 #include "qemu/module.h"
35 #include "qapi/error.h"
36 #include "qapi/qmp/qdict.h"
37 #include "qapi/qmp/qjson.h"
38 #include "qapi/qmp/qnull.h"
39 #include "qapi/qmp/qstring.h"
40 #include "qapi/qobject-output-visitor.h"
41 #include "qapi/qapi-visit-block-core.h"
42 #include "sysemu/block-backend.h"
43 #include "sysemu/sysemu.h"
44 #include "qemu/notify.h"
45 #include "qemu/option.h"
46 #include "qemu/coroutine.h"
47 #include "block/qapi.h"
48 #include "qemu/timer.h"
49 #include "qemu/cutils.h"
53 #include <sys/ioctl.h>
54 #include <sys/queue.h>
64 #define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */
66 static QTAILQ_HEAD(, BlockDriverState
) graph_bdrv_states
=
67 QTAILQ_HEAD_INITIALIZER(graph_bdrv_states
);
69 static QTAILQ_HEAD(, BlockDriverState
) all_bdrv_states
=
70 QTAILQ_HEAD_INITIALIZER(all_bdrv_states
);
72 static QLIST_HEAD(, BlockDriver
) bdrv_drivers
=
73 QLIST_HEAD_INITIALIZER(bdrv_drivers
);
75 static BlockDriverState
*bdrv_open_inherit(const char *filename
,
76 const char *reference
,
77 QDict
*options
, int flags
,
78 BlockDriverState
*parent
,
79 const BdrvChildRole
*child_role
,
82 /* If non-zero, use only whitelisted block drivers */
83 static int use_bdrv_whitelist
;
86 static int is_windows_drive_prefix(const char *filename
)
88 return (((filename
[0] >= 'a' && filename
[0] <= 'z') ||
89 (filename
[0] >= 'A' && filename
[0] <= 'Z')) &&
93 int is_windows_drive(const char *filename
)
95 if (is_windows_drive_prefix(filename
) &&
98 if (strstart(filename
, "\\\\.\\", NULL
) ||
99 strstart(filename
, "//./", NULL
))
105 size_t bdrv_opt_mem_align(BlockDriverState
*bs
)
107 if (!bs
|| !bs
->drv
) {
108 /* page size or 4k (hdd sector size) should be on the safe side */
109 return MAX(4096, getpagesize());
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, getpagesize());
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 QLIST_INSERT_HEAD(&bdrv_drivers
, bdrv
, list
);
369 BlockDriverState
*bdrv_new(void)
371 BlockDriverState
*bs
;
374 bs
= g_new0(BlockDriverState
, 1);
375 QLIST_INIT(&bs
->dirty_bitmaps
);
376 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
377 QLIST_INIT(&bs
->op_blockers
[i
]);
379 notifier_with_return_list_init(&bs
->before_write_notifiers
);
380 qemu_co_mutex_init(&bs
->reqs_lock
);
381 qemu_mutex_init(&bs
->dirty_bitmap_mutex
);
383 bs
->aio_context
= qemu_get_aio_context();
385 qemu_co_queue_init(&bs
->flush_queue
);
387 for (i
= 0; i
< bdrv_drain_all_count
; i
++) {
388 bdrv_drained_begin(bs
);
391 QTAILQ_INSERT_TAIL(&all_bdrv_states
, bs
, bs_list
);
396 static BlockDriver
*bdrv_do_find_format(const char *format_name
)
400 QLIST_FOREACH(drv1
, &bdrv_drivers
, list
) {
401 if (!strcmp(drv1
->format_name
, format_name
)) {
409 BlockDriver
*bdrv_find_format(const char *format_name
)
414 drv1
= bdrv_do_find_format(format_name
);
419 /* The driver isn't registered, maybe we need to load a module */
420 for (i
= 0; i
< (int)ARRAY_SIZE(block_driver_modules
); ++i
) {
421 if (!strcmp(block_driver_modules
[i
].format_name
, format_name
)) {
422 block_module_load_one(block_driver_modules
[i
].library_name
);
427 return bdrv_do_find_format(format_name
);
430 static int bdrv_format_is_whitelisted(const char *format_name
, bool read_only
)
432 static const char *whitelist_rw
[] = {
433 CONFIG_BDRV_RW_WHITELIST
435 static const char *whitelist_ro
[] = {
436 CONFIG_BDRV_RO_WHITELIST
440 if (!whitelist_rw
[0] && !whitelist_ro
[0]) {
441 return 1; /* no whitelist, anything goes */
444 for (p
= whitelist_rw
; *p
; p
++) {
445 if (!strcmp(format_name
, *p
)) {
450 for (p
= whitelist_ro
; *p
; p
++) {
451 if (!strcmp(format_name
, *p
)) {
459 int bdrv_is_whitelisted(BlockDriver
*drv
, bool read_only
)
461 return bdrv_format_is_whitelisted(drv
->format_name
, read_only
);
464 bool bdrv_uses_whitelist(void)
466 return use_bdrv_whitelist
;
469 typedef struct CreateCo
{
477 static void coroutine_fn
bdrv_create_co_entry(void *opaque
)
479 Error
*local_err
= NULL
;
482 CreateCo
*cco
= opaque
;
485 ret
= cco
->drv
->bdrv_co_create_opts(cco
->filename
, cco
->opts
, &local_err
);
486 error_propagate(&cco
->err
, local_err
);
490 int bdrv_create(BlockDriver
*drv
, const char* filename
,
491 QemuOpts
*opts
, Error
**errp
)
498 .filename
= g_strdup(filename
),
504 if (!drv
->bdrv_co_create_opts
) {
505 error_setg(errp
, "Driver '%s' does not support image creation", drv
->format_name
);
510 if (qemu_in_coroutine()) {
511 /* Fast-path if already in coroutine context */
512 bdrv_create_co_entry(&cco
);
514 co
= qemu_coroutine_create(bdrv_create_co_entry
, &cco
);
515 qemu_coroutine_enter(co
);
516 while (cco
.ret
== NOT_DONE
) {
517 aio_poll(qemu_get_aio_context(), true);
524 error_propagate(errp
, cco
.err
);
526 error_setg_errno(errp
, -ret
, "Could not create image");
531 g_free(cco
.filename
);
535 int bdrv_create_file(const char *filename
, QemuOpts
*opts
, Error
**errp
)
538 Error
*local_err
= NULL
;
541 drv
= bdrv_find_protocol(filename
, true, errp
);
546 ret
= bdrv_create(drv
, filename
, opts
, &local_err
);
547 error_propagate(errp
, local_err
);
552 * Try to get @bs's logical and physical block size.
553 * On success, store them in @bsz struct and return 0.
554 * On failure return -errno.
555 * @bs must not be empty.
557 int bdrv_probe_blocksizes(BlockDriverState
*bs
, BlockSizes
*bsz
)
559 BlockDriver
*drv
= bs
->drv
;
561 if (drv
&& drv
->bdrv_probe_blocksizes
) {
562 return drv
->bdrv_probe_blocksizes(bs
, bsz
);
563 } else if (drv
&& drv
->is_filter
&& bs
->file
) {
564 return bdrv_probe_blocksizes(bs
->file
->bs
, bsz
);
571 * Try to get @bs's geometry (cyls, heads, sectors).
572 * On success, store them in @geo struct and return 0.
573 * On failure return -errno.
574 * @bs must not be empty.
576 int bdrv_probe_geometry(BlockDriverState
*bs
, HDGeometry
*geo
)
578 BlockDriver
*drv
= bs
->drv
;
580 if (drv
&& drv
->bdrv_probe_geometry
) {
581 return drv
->bdrv_probe_geometry(bs
, geo
);
582 } else if (drv
&& drv
->is_filter
&& bs
->file
) {
583 return bdrv_probe_geometry(bs
->file
->bs
, geo
);
590 * Create a uniquely-named empty temporary file.
591 * Return 0 upon success, otherwise a negative errno value.
593 int get_tmp_filename(char *filename
, int size
)
596 char temp_dir
[MAX_PATH
];
597 /* GetTempFileName requires that its output buffer (4th param)
598 have length MAX_PATH or greater. */
599 assert(size
>= MAX_PATH
);
600 return (GetTempPath(MAX_PATH
, temp_dir
)
601 && GetTempFileName(temp_dir
, "qem", 0, filename
)
602 ? 0 : -GetLastError());
606 tmpdir
= getenv("TMPDIR");
610 if (snprintf(filename
, size
, "%s/vl.XXXXXX", tmpdir
) >= size
) {
613 fd
= mkstemp(filename
);
617 if (close(fd
) != 0) {
626 * Detect host devices. By convention, /dev/cdrom[N] is always
627 * recognized as a host CDROM.
629 static BlockDriver
*find_hdev_driver(const char *filename
)
631 int score_max
= 0, score
;
632 BlockDriver
*drv
= NULL
, *d
;
634 QLIST_FOREACH(d
, &bdrv_drivers
, list
) {
635 if (d
->bdrv_probe_device
) {
636 score
= d
->bdrv_probe_device(filename
);
637 if (score
> score_max
) {
647 static BlockDriver
*bdrv_do_find_protocol(const char *protocol
)
651 QLIST_FOREACH(drv1
, &bdrv_drivers
, list
) {
652 if (drv1
->protocol_name
&& !strcmp(drv1
->protocol_name
, protocol
)) {
660 BlockDriver
*bdrv_find_protocol(const char *filename
,
661 bool allow_protocol_prefix
,
670 /* TODO Drivers without bdrv_file_open must be specified explicitly */
673 * XXX(hch): we really should not let host device detection
674 * override an explicit protocol specification, but moving this
675 * later breaks access to device names with colons in them.
676 * Thanks to the brain-dead persistent naming schemes on udev-
677 * based Linux systems those actually are quite common.
679 drv1
= find_hdev_driver(filename
);
684 if (!path_has_protocol(filename
) || !allow_protocol_prefix
) {
688 p
= strchr(filename
, ':');
691 if (len
> sizeof(protocol
) - 1)
692 len
= sizeof(protocol
) - 1;
693 memcpy(protocol
, filename
, len
);
694 protocol
[len
] = '\0';
696 drv1
= bdrv_do_find_protocol(protocol
);
701 for (i
= 0; i
< (int)ARRAY_SIZE(block_driver_modules
); ++i
) {
702 if (block_driver_modules
[i
].protocol_name
&&
703 !strcmp(block_driver_modules
[i
].protocol_name
, protocol
)) {
704 block_module_load_one(block_driver_modules
[i
].library_name
);
709 drv1
= bdrv_do_find_protocol(protocol
);
711 error_setg(errp
, "Unknown protocol '%s'", protocol
);
717 * Guess image format by probing its contents.
718 * This is not a good idea when your image is raw (CVE-2008-2004), but
719 * we do it anyway for backward compatibility.
721 * @buf contains the image's first @buf_size bytes.
722 * @buf_size is the buffer size in bytes (generally BLOCK_PROBE_BUF_SIZE,
723 * but can be smaller if the image file is smaller)
724 * @filename is its filename.
726 * For all block drivers, call the bdrv_probe() method to get its
728 * Return the first block driver with the highest probing score.
730 BlockDriver
*bdrv_probe_all(const uint8_t *buf
, int buf_size
,
731 const char *filename
)
733 int score_max
= 0, score
;
734 BlockDriver
*drv
= NULL
, *d
;
736 QLIST_FOREACH(d
, &bdrv_drivers
, list
) {
738 score
= d
->bdrv_probe(buf
, buf_size
, filename
);
739 if (score
> score_max
) {
749 static int find_image_format(BlockBackend
*file
, const char *filename
,
750 BlockDriver
**pdrv
, Error
**errp
)
753 uint8_t buf
[BLOCK_PROBE_BUF_SIZE
];
756 /* Return the raw BlockDriver * to scsi-generic devices or empty drives */
757 if (blk_is_sg(file
) || !blk_is_inserted(file
) || blk_getlength(file
) == 0) {
762 ret
= blk_pread(file
, 0, buf
, sizeof(buf
));
764 error_setg_errno(errp
, -ret
, "Could not read image for determining its "
770 drv
= bdrv_probe_all(buf
, ret
, filename
);
772 error_setg(errp
, "Could not determine image format: No compatible "
781 * Set the current 'total_sectors' value
782 * Return 0 on success, -errno on error.
784 int refresh_total_sectors(BlockDriverState
*bs
, int64_t hint
)
786 BlockDriver
*drv
= bs
->drv
;
792 /* Do not attempt drv->bdrv_getlength() on scsi-generic devices */
796 /* query actual device if possible, otherwise just trust the hint */
797 if (drv
->bdrv_getlength
) {
798 int64_t length
= drv
->bdrv_getlength(bs
);
802 hint
= DIV_ROUND_UP(length
, BDRV_SECTOR_SIZE
);
805 bs
->total_sectors
= hint
;
810 * Combines a QDict of new block driver @options with any missing options taken
811 * from @old_options, so that leaving out an option defaults to its old value.
813 static void bdrv_join_options(BlockDriverState
*bs
, QDict
*options
,
816 if (bs
->drv
&& bs
->drv
->bdrv_join_options
) {
817 bs
->drv
->bdrv_join_options(options
, old_options
);
819 qdict_join(options
, old_options
, false);
823 static BlockdevDetectZeroesOptions
bdrv_parse_detect_zeroes(QemuOpts
*opts
,
827 Error
*local_err
= NULL
;
828 char *value
= qemu_opt_get_del(opts
, "detect-zeroes");
829 BlockdevDetectZeroesOptions detect_zeroes
=
830 qapi_enum_parse(&BlockdevDetectZeroesOptions_lookup
, value
,
831 BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF
, &local_err
);
834 error_propagate(errp
, local_err
);
835 return detect_zeroes
;
838 if (detect_zeroes
== BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP
&&
839 !(open_flags
& BDRV_O_UNMAP
))
841 error_setg(errp
, "setting detect-zeroes to unmap is not allowed "
842 "without setting discard operation to unmap");
845 return detect_zeroes
;
849 * Set open flags for a given discard mode
851 * Return 0 on success, -1 if the discard mode was invalid.
853 int bdrv_parse_discard_flags(const char *mode
, int *flags
)
855 *flags
&= ~BDRV_O_UNMAP
;
857 if (!strcmp(mode
, "off") || !strcmp(mode
, "ignore")) {
859 } else if (!strcmp(mode
, "on") || !strcmp(mode
, "unmap")) {
860 *flags
|= BDRV_O_UNMAP
;
869 * Set open flags for a given cache mode
871 * Return 0 on success, -1 if the cache mode was invalid.
873 int bdrv_parse_cache_mode(const char *mode
, int *flags
, bool *writethrough
)
875 *flags
&= ~BDRV_O_CACHE_MASK
;
877 if (!strcmp(mode
, "off") || !strcmp(mode
, "none")) {
878 *writethrough
= false;
879 *flags
|= BDRV_O_NOCACHE
;
880 } else if (!strcmp(mode
, "directsync")) {
881 *writethrough
= true;
882 *flags
|= BDRV_O_NOCACHE
;
883 } else if (!strcmp(mode
, "writeback")) {
884 *writethrough
= false;
885 } else if (!strcmp(mode
, "unsafe")) {
886 *writethrough
= false;
887 *flags
|= BDRV_O_NO_FLUSH
;
888 } else if (!strcmp(mode
, "writethrough")) {
889 *writethrough
= true;
897 static char *bdrv_child_get_parent_desc(BdrvChild
*c
)
899 BlockDriverState
*parent
= c
->opaque
;
900 return g_strdup(bdrv_get_device_or_node_name(parent
));
903 static void bdrv_child_cb_drained_begin(BdrvChild
*child
)
905 BlockDriverState
*bs
= child
->opaque
;
906 bdrv_do_drained_begin_quiesce(bs
, NULL
, false);
909 static bool bdrv_child_cb_drained_poll(BdrvChild
*child
)
911 BlockDriverState
*bs
= child
->opaque
;
912 return bdrv_drain_poll(bs
, false, NULL
, false);
915 static void bdrv_child_cb_drained_end(BdrvChild
*child
,
916 int *drained_end_counter
)
918 BlockDriverState
*bs
= child
->opaque
;
919 bdrv_drained_end_no_poll(bs
, drained_end_counter
);
922 static void bdrv_child_cb_attach(BdrvChild
*child
)
924 BlockDriverState
*bs
= child
->opaque
;
925 bdrv_apply_subtree_drain(child
, bs
);
928 static void bdrv_child_cb_detach(BdrvChild
*child
)
930 BlockDriverState
*bs
= child
->opaque
;
931 bdrv_unapply_subtree_drain(child
, bs
);
934 static int bdrv_child_cb_inactivate(BdrvChild
*child
)
936 BlockDriverState
*bs
= child
->opaque
;
937 assert(bs
->open_flags
& BDRV_O_INACTIVE
);
941 static bool bdrv_child_cb_can_set_aio_ctx(BdrvChild
*child
, AioContext
*ctx
,
942 GSList
**ignore
, Error
**errp
)
944 BlockDriverState
*bs
= child
->opaque
;
945 return bdrv_can_set_aio_context(bs
, ctx
, ignore
, errp
);
948 static void bdrv_child_cb_set_aio_ctx(BdrvChild
*child
, AioContext
*ctx
,
951 BlockDriverState
*bs
= child
->opaque
;
952 return bdrv_set_aio_context_ignore(bs
, ctx
, ignore
);
956 * Returns the options and flags that a temporary snapshot should get, based on
957 * the originally requested flags (the originally requested image will have
958 * flags like a backing file)
960 static void bdrv_temp_snapshot_options(int *child_flags
, QDict
*child_options
,
961 int parent_flags
, QDict
*parent_options
)
963 *child_flags
= (parent_flags
& ~BDRV_O_SNAPSHOT
) | BDRV_O_TEMPORARY
;
965 /* For temporary files, unconditional cache=unsafe is fine */
966 qdict_set_default_str(child_options
, BDRV_OPT_CACHE_DIRECT
, "off");
967 qdict_set_default_str(child_options
, BDRV_OPT_CACHE_NO_FLUSH
, "on");
969 /* Copy the read-only and discard options from the parent */
970 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_READ_ONLY
);
971 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_DISCARD
);
973 /* aio=native doesn't work for cache.direct=off, so disable it for the
974 * temporary snapshot */
975 *child_flags
&= ~BDRV_O_NATIVE_AIO
;
979 * Returns the options and flags that bs->file should get if a protocol driver
980 * is expected, based on the given options and flags for the parent BDS
982 static void bdrv_inherited_options(int *child_flags
, QDict
*child_options
,
983 int parent_flags
, QDict
*parent_options
)
985 int flags
= parent_flags
;
987 /* Enable protocol handling, disable format probing for bs->file */
988 flags
|= BDRV_O_PROTOCOL
;
990 /* If the cache mode isn't explicitly set, inherit direct and no-flush from
992 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_CACHE_DIRECT
);
993 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_CACHE_NO_FLUSH
);
994 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_FORCE_SHARE
);
996 /* Inherit the read-only option from the parent if it's not set */
997 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_READ_ONLY
);
998 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_AUTO_READ_ONLY
);
1000 /* Our block drivers take care to send flushes and respect unmap policy,
1001 * so we can default to enable both on lower layers regardless of the
1002 * corresponding parent options. */
1003 qdict_set_default_str(child_options
, BDRV_OPT_DISCARD
, "unmap");
1005 /* Clear flags that only apply to the top layer */
1006 flags
&= ~(BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
| BDRV_O_COPY_ON_READ
|
1009 *child_flags
= flags
;
1012 const BdrvChildRole child_file
= {
1013 .parent_is_bds
= true,
1014 .get_parent_desc
= bdrv_child_get_parent_desc
,
1015 .inherit_options
= bdrv_inherited_options
,
1016 .drained_begin
= bdrv_child_cb_drained_begin
,
1017 .drained_poll
= bdrv_child_cb_drained_poll
,
1018 .drained_end
= bdrv_child_cb_drained_end
,
1019 .attach
= bdrv_child_cb_attach
,
1020 .detach
= bdrv_child_cb_detach
,
1021 .inactivate
= bdrv_child_cb_inactivate
,
1022 .can_set_aio_ctx
= bdrv_child_cb_can_set_aio_ctx
,
1023 .set_aio_ctx
= bdrv_child_cb_set_aio_ctx
,
1027 * Returns the options and flags that bs->file should get if the use of formats
1028 * (and not only protocols) is permitted for it, based on the given options and
1029 * flags for the parent BDS
1031 static void bdrv_inherited_fmt_options(int *child_flags
, QDict
*child_options
,
1032 int parent_flags
, QDict
*parent_options
)
1034 child_file
.inherit_options(child_flags
, child_options
,
1035 parent_flags
, parent_options
);
1037 *child_flags
&= ~(BDRV_O_PROTOCOL
| BDRV_O_NO_IO
);
1040 const BdrvChildRole child_format
= {
1041 .parent_is_bds
= true,
1042 .get_parent_desc
= bdrv_child_get_parent_desc
,
1043 .inherit_options
= bdrv_inherited_fmt_options
,
1044 .drained_begin
= bdrv_child_cb_drained_begin
,
1045 .drained_poll
= bdrv_child_cb_drained_poll
,
1046 .drained_end
= bdrv_child_cb_drained_end
,
1047 .attach
= bdrv_child_cb_attach
,
1048 .detach
= bdrv_child_cb_detach
,
1049 .inactivate
= bdrv_child_cb_inactivate
,
1050 .can_set_aio_ctx
= bdrv_child_cb_can_set_aio_ctx
,
1051 .set_aio_ctx
= bdrv_child_cb_set_aio_ctx
,
1054 static void bdrv_backing_attach(BdrvChild
*c
)
1056 BlockDriverState
*parent
= c
->opaque
;
1057 BlockDriverState
*backing_hd
= c
->bs
;
1059 assert(!parent
->backing_blocker
);
1060 error_setg(&parent
->backing_blocker
,
1061 "node is used as backing hd of '%s'",
1062 bdrv_get_device_or_node_name(parent
));
1064 bdrv_refresh_filename(backing_hd
);
1066 parent
->open_flags
&= ~BDRV_O_NO_BACKING
;
1067 pstrcpy(parent
->backing_file
, sizeof(parent
->backing_file
),
1068 backing_hd
->filename
);
1069 pstrcpy(parent
->backing_format
, sizeof(parent
->backing_format
),
1070 backing_hd
->drv
? backing_hd
->drv
->format_name
: "");
1072 bdrv_op_block_all(backing_hd
, parent
->backing_blocker
);
1073 /* Otherwise we won't be able to commit or stream */
1074 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_COMMIT_TARGET
,
1075 parent
->backing_blocker
);
1076 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_STREAM
,
1077 parent
->backing_blocker
);
1079 * We do backup in 3 ways:
1081 * The target bs is new opened, and the source is top BDS
1082 * 2. blockdev backup
1083 * Both the source and the target are top BDSes.
1084 * 3. internal backup(used for block replication)
1085 * Both the source and the target are backing file
1087 * In case 1 and 2, neither the source nor the target is the backing file.
1088 * In case 3, we will block the top BDS, so there is only one block job
1089 * for the top BDS and its backing chain.
1091 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_BACKUP_SOURCE
,
1092 parent
->backing_blocker
);
1093 bdrv_op_unblock(backing_hd
, BLOCK_OP_TYPE_BACKUP_TARGET
,
1094 parent
->backing_blocker
);
1096 bdrv_child_cb_attach(c
);
1099 static void bdrv_backing_detach(BdrvChild
*c
)
1101 BlockDriverState
*parent
= c
->opaque
;
1103 assert(parent
->backing_blocker
);
1104 bdrv_op_unblock_all(c
->bs
, parent
->backing_blocker
);
1105 error_free(parent
->backing_blocker
);
1106 parent
->backing_blocker
= NULL
;
1108 bdrv_child_cb_detach(c
);
1112 * Returns the options and flags that bs->backing should get, based on the
1113 * given options and flags for the parent BDS
1115 static void bdrv_backing_options(int *child_flags
, QDict
*child_options
,
1116 int parent_flags
, QDict
*parent_options
)
1118 int flags
= parent_flags
;
1120 /* The cache mode is inherited unmodified for backing files; except WCE,
1121 * which is only applied on the top level (BlockBackend) */
1122 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_CACHE_DIRECT
);
1123 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_CACHE_NO_FLUSH
);
1124 qdict_copy_default(child_options
, parent_options
, BDRV_OPT_FORCE_SHARE
);
1126 /* backing files always opened read-only */
1127 qdict_set_default_str(child_options
, BDRV_OPT_READ_ONLY
, "on");
1128 qdict_set_default_str(child_options
, BDRV_OPT_AUTO_READ_ONLY
, "off");
1129 flags
&= ~BDRV_O_COPY_ON_READ
;
1131 /* snapshot=on is handled on the top layer */
1132 flags
&= ~(BDRV_O_SNAPSHOT
| BDRV_O_TEMPORARY
);
1134 *child_flags
= flags
;
1137 static int bdrv_backing_update_filename(BdrvChild
*c
, BlockDriverState
*base
,
1138 const char *filename
, Error
**errp
)
1140 BlockDriverState
*parent
= c
->opaque
;
1141 bool read_only
= bdrv_is_read_only(parent
);
1145 ret
= bdrv_reopen_set_read_only(parent
, false, errp
);
1151 ret
= bdrv_change_backing_file(parent
, filename
,
1152 base
->drv
? base
->drv
->format_name
: "");
1154 error_setg_errno(errp
, -ret
, "Could not update backing file link");
1158 bdrv_reopen_set_read_only(parent
, true, NULL
);
1164 const BdrvChildRole child_backing
= {
1165 .parent_is_bds
= true,
1166 .get_parent_desc
= bdrv_child_get_parent_desc
,
1167 .attach
= bdrv_backing_attach
,
1168 .detach
= bdrv_backing_detach
,
1169 .inherit_options
= bdrv_backing_options
,
1170 .drained_begin
= bdrv_child_cb_drained_begin
,
1171 .drained_poll
= bdrv_child_cb_drained_poll
,
1172 .drained_end
= bdrv_child_cb_drained_end
,
1173 .inactivate
= bdrv_child_cb_inactivate
,
1174 .update_filename
= bdrv_backing_update_filename
,
1175 .can_set_aio_ctx
= bdrv_child_cb_can_set_aio_ctx
,
1176 .set_aio_ctx
= bdrv_child_cb_set_aio_ctx
,
1179 static int bdrv_open_flags(BlockDriverState
*bs
, int flags
)
1181 int open_flags
= flags
;
1184 * Clear flags that are internal to the block layer before opening the
1187 open_flags
&= ~(BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
| BDRV_O_PROTOCOL
);
1192 static void update_flags_from_options(int *flags
, QemuOpts
*opts
)
1194 *flags
&= ~(BDRV_O_CACHE_MASK
| BDRV_O_RDWR
| BDRV_O_AUTO_RDONLY
);
1196 if (qemu_opt_get_bool_del(opts
, BDRV_OPT_CACHE_NO_FLUSH
, false)) {
1197 *flags
|= BDRV_O_NO_FLUSH
;
1200 if (qemu_opt_get_bool_del(opts
, BDRV_OPT_CACHE_DIRECT
, false)) {
1201 *flags
|= BDRV_O_NOCACHE
;
1204 if (!qemu_opt_get_bool_del(opts
, BDRV_OPT_READ_ONLY
, false)) {
1205 *flags
|= BDRV_O_RDWR
;
1208 if (qemu_opt_get_bool_del(opts
, BDRV_OPT_AUTO_READ_ONLY
, false)) {
1209 *flags
|= BDRV_O_AUTO_RDONLY
;
1213 static void update_options_from_flags(QDict
*options
, int flags
)
1215 if (!qdict_haskey(options
, BDRV_OPT_CACHE_DIRECT
)) {
1216 qdict_put_bool(options
, BDRV_OPT_CACHE_DIRECT
, flags
& BDRV_O_NOCACHE
);
1218 if (!qdict_haskey(options
, BDRV_OPT_CACHE_NO_FLUSH
)) {
1219 qdict_put_bool(options
, BDRV_OPT_CACHE_NO_FLUSH
,
1220 flags
& BDRV_O_NO_FLUSH
);
1222 if (!qdict_haskey(options
, BDRV_OPT_READ_ONLY
)) {
1223 qdict_put_bool(options
, BDRV_OPT_READ_ONLY
, !(flags
& BDRV_O_RDWR
));
1225 if (!qdict_haskey(options
, BDRV_OPT_AUTO_READ_ONLY
)) {
1226 qdict_put_bool(options
, BDRV_OPT_AUTO_READ_ONLY
,
1227 flags
& BDRV_O_AUTO_RDONLY
);
1231 static void bdrv_assign_node_name(BlockDriverState
*bs
,
1232 const char *node_name
,
1235 char *gen_node_name
= NULL
;
1238 node_name
= gen_node_name
= id_generate(ID_BLOCK
);
1239 } else if (!id_wellformed(node_name
)) {
1241 * Check for empty string or invalid characters, but not if it is
1242 * generated (generated names use characters not available to the user)
1244 error_setg(errp
, "Invalid node name");
1248 /* takes care of avoiding namespaces collisions */
1249 if (blk_by_name(node_name
)) {
1250 error_setg(errp
, "node-name=%s is conflicting with a device id",
1255 /* takes care of avoiding duplicates node names */
1256 if (bdrv_find_node(node_name
)) {
1257 error_setg(errp
, "Duplicate node name");
1261 /* Make sure that the node name isn't truncated */
1262 if (strlen(node_name
) >= sizeof(bs
->node_name
)) {
1263 error_setg(errp
, "Node name too long");
1267 /* copy node name into the bs and insert it into the graph list */
1268 pstrcpy(bs
->node_name
, sizeof(bs
->node_name
), node_name
);
1269 QTAILQ_INSERT_TAIL(&graph_bdrv_states
, bs
, node_list
);
1271 g_free(gen_node_name
);
1274 static int bdrv_open_driver(BlockDriverState
*bs
, BlockDriver
*drv
,
1275 const char *node_name
, QDict
*options
,
1276 int open_flags
, Error
**errp
)
1278 Error
*local_err
= NULL
;
1281 bdrv_assign_node_name(bs
, node_name
, &local_err
);
1283 error_propagate(errp
, local_err
);
1288 bs
->read_only
= !(bs
->open_flags
& BDRV_O_RDWR
);
1289 bs
->opaque
= g_malloc0(drv
->instance_size
);
1291 if (drv
->bdrv_file_open
) {
1292 assert(!drv
->bdrv_needs_filename
|| bs
->filename
[0]);
1293 ret
= drv
->bdrv_file_open(bs
, options
, open_flags
, &local_err
);
1294 } else if (drv
->bdrv_open
) {
1295 ret
= drv
->bdrv_open(bs
, options
, open_flags
, &local_err
);
1302 error_propagate(errp
, local_err
);
1303 } else if (bs
->filename
[0]) {
1304 error_setg_errno(errp
, -ret
, "Could not open '%s'", bs
->filename
);
1306 error_setg_errno(errp
, -ret
, "Could not open image");
1311 ret
= refresh_total_sectors(bs
, bs
->total_sectors
);
1313 error_setg_errno(errp
, -ret
, "Could not refresh total sector count");
1317 bdrv_refresh_limits(bs
, &local_err
);
1319 error_propagate(errp
, local_err
);
1323 assert(bdrv_opt_mem_align(bs
) != 0);
1324 assert(bdrv_min_mem_align(bs
) != 0);
1325 assert(is_power_of_2(bs
->bl
.request_alignment
));
1327 for (i
= 0; i
< bs
->quiesce_counter
; i
++) {
1328 if (drv
->bdrv_co_drain_begin
) {
1329 drv
->bdrv_co_drain_begin(bs
);
1336 if (bs
->file
!= NULL
) {
1337 bdrv_unref_child(bs
, bs
->file
);
1345 BlockDriverState
*bdrv_new_open_driver(BlockDriver
*drv
, const char *node_name
,
1346 int flags
, Error
**errp
)
1348 BlockDriverState
*bs
;
1352 bs
->open_flags
= flags
;
1353 bs
->explicit_options
= qdict_new();
1354 bs
->options
= qdict_new();
1357 update_options_from_flags(bs
->options
, flags
);
1359 ret
= bdrv_open_driver(bs
, drv
, node_name
, bs
->options
, flags
, errp
);
1361 qobject_unref(bs
->explicit_options
);
1362 bs
->explicit_options
= NULL
;
1363 qobject_unref(bs
->options
);
1372 QemuOptsList bdrv_runtime_opts
= {
1373 .name
= "bdrv_common",
1374 .head
= QTAILQ_HEAD_INITIALIZER(bdrv_runtime_opts
.head
),
1377 .name
= "node-name",
1378 .type
= QEMU_OPT_STRING
,
1379 .help
= "Node name of the block device node",
1383 .type
= QEMU_OPT_STRING
,
1384 .help
= "Block driver to use for the node",
1387 .name
= BDRV_OPT_CACHE_DIRECT
,
1388 .type
= QEMU_OPT_BOOL
,
1389 .help
= "Bypass software writeback cache on the host",
1392 .name
= BDRV_OPT_CACHE_NO_FLUSH
,
1393 .type
= QEMU_OPT_BOOL
,
1394 .help
= "Ignore flush requests",
1397 .name
= BDRV_OPT_READ_ONLY
,
1398 .type
= QEMU_OPT_BOOL
,
1399 .help
= "Node is opened in read-only mode",
1402 .name
= BDRV_OPT_AUTO_READ_ONLY
,
1403 .type
= QEMU_OPT_BOOL
,
1404 .help
= "Node can become read-only if opening read-write fails",
1407 .name
= "detect-zeroes",
1408 .type
= QEMU_OPT_STRING
,
1409 .help
= "try to optimize zero writes (off, on, unmap)",
1412 .name
= BDRV_OPT_DISCARD
,
1413 .type
= QEMU_OPT_STRING
,
1414 .help
= "discard operation (ignore/off, unmap/on)",
1417 .name
= BDRV_OPT_FORCE_SHARE
,
1418 .type
= QEMU_OPT_BOOL
,
1419 .help
= "always accept other writers (default: off)",
1421 { /* end of list */ }
1426 * Common part for opening disk images and files
1428 * Removes all processed options from *options.
1430 static int bdrv_open_common(BlockDriverState
*bs
, BlockBackend
*file
,
1431 QDict
*options
, Error
**errp
)
1433 int ret
, open_flags
;
1434 const char *filename
;
1435 const char *driver_name
= NULL
;
1436 const char *node_name
= NULL
;
1437 const char *discard
;
1440 Error
*local_err
= NULL
;
1442 assert(bs
->file
== NULL
);
1443 assert(options
!= NULL
&& bs
->options
!= options
);
1445 opts
= qemu_opts_create(&bdrv_runtime_opts
, NULL
, 0, &error_abort
);
1446 qemu_opts_absorb_qdict(opts
, options
, &local_err
);
1448 error_propagate(errp
, local_err
);
1453 update_flags_from_options(&bs
->open_flags
, opts
);
1455 driver_name
= qemu_opt_get(opts
, "driver");
1456 drv
= bdrv_find_format(driver_name
);
1457 assert(drv
!= NULL
);
1459 bs
->force_share
= qemu_opt_get_bool(opts
, BDRV_OPT_FORCE_SHARE
, false);
1461 if (bs
->force_share
&& (bs
->open_flags
& BDRV_O_RDWR
)) {
1463 BDRV_OPT_FORCE_SHARE
1464 "=on can only be used with read-only images");
1470 bdrv_refresh_filename(blk_bs(file
));
1471 filename
= blk_bs(file
)->filename
;
1474 * Caution: while qdict_get_try_str() is fine, getting
1475 * non-string types would require more care. When @options
1476 * come from -blockdev or blockdev_add, its members are typed
1477 * according to the QAPI schema, but when they come from
1478 * -drive, they're all QString.
1480 filename
= qdict_get_try_str(options
, "filename");
1483 if (drv
->bdrv_needs_filename
&& (!filename
|| !filename
[0])) {
1484 error_setg(errp
, "The '%s' block driver requires a file name",
1490 trace_bdrv_open_common(bs
, filename
?: "", bs
->open_flags
,
1493 bs
->read_only
= !(bs
->open_flags
& BDRV_O_RDWR
);
1495 if (use_bdrv_whitelist
&& !bdrv_is_whitelisted(drv
, bs
->read_only
)) {
1496 if (!bs
->read_only
&& bdrv_is_whitelisted(drv
, true)) {
1497 ret
= bdrv_apply_auto_read_only(bs
, NULL
, NULL
);
1503 !bs
->read_only
&& bdrv_is_whitelisted(drv
, true)
1504 ? "Driver '%s' can only be used for read-only devices"
1505 : "Driver '%s' is not whitelisted",
1511 /* bdrv_new() and bdrv_close() make it so */
1512 assert(atomic_read(&bs
->copy_on_read
) == 0);
1514 if (bs
->open_flags
& BDRV_O_COPY_ON_READ
) {
1515 if (!bs
->read_only
) {
1516 bdrv_enable_copy_on_read(bs
);
1518 error_setg(errp
, "Can't use copy-on-read on read-only device");
1524 discard
= qemu_opt_get(opts
, BDRV_OPT_DISCARD
);
1525 if (discard
!= NULL
) {
1526 if (bdrv_parse_discard_flags(discard
, &bs
->open_flags
) != 0) {
1527 error_setg(errp
, "Invalid discard option");
1534 bdrv_parse_detect_zeroes(opts
, bs
->open_flags
, &local_err
);
1536 error_propagate(errp
, local_err
);
1541 if (filename
!= NULL
) {
1542 pstrcpy(bs
->filename
, sizeof(bs
->filename
), filename
);
1544 bs
->filename
[0] = '\0';
1546 pstrcpy(bs
->exact_filename
, sizeof(bs
->exact_filename
), bs
->filename
);
1548 /* Open the image, either directly or using a protocol */
1549 open_flags
= bdrv_open_flags(bs
, bs
->open_flags
);
1550 node_name
= qemu_opt_get(opts
, "node-name");
1552 assert(!drv
->bdrv_file_open
|| file
== NULL
);
1553 ret
= bdrv_open_driver(bs
, drv
, node_name
, options
, open_flags
, errp
);
1558 qemu_opts_del(opts
);
1562 qemu_opts_del(opts
);
1566 static QDict
*parse_json_filename(const char *filename
, Error
**errp
)
1568 QObject
*options_obj
;
1572 ret
= strstart(filename
, "json:", &filename
);
1575 options_obj
= qobject_from_json(filename
, errp
);
1577 error_prepend(errp
, "Could not parse the JSON options: ");
1581 options
= qobject_to(QDict
, options_obj
);
1583 qobject_unref(options_obj
);
1584 error_setg(errp
, "Invalid JSON object given");
1588 qdict_flatten(options
);
1593 static void parse_json_protocol(QDict
*options
, const char **pfilename
,
1596 QDict
*json_options
;
1597 Error
*local_err
= NULL
;
1599 /* Parse json: pseudo-protocol */
1600 if (!*pfilename
|| !g_str_has_prefix(*pfilename
, "json:")) {
1604 json_options
= parse_json_filename(*pfilename
, &local_err
);
1606 error_propagate(errp
, local_err
);
1610 /* Options given in the filename have lower priority than options
1611 * specified directly */
1612 qdict_join(options
, json_options
, false);
1613 qobject_unref(json_options
);
1618 * Fills in default options for opening images and converts the legacy
1619 * filename/flags pair to option QDict entries.
1620 * The BDRV_O_PROTOCOL flag in *flags will be set or cleared accordingly if a
1621 * block driver has been specified explicitly.
1623 static int bdrv_fill_options(QDict
**options
, const char *filename
,
1624 int *flags
, Error
**errp
)
1626 const char *drvname
;
1627 bool protocol
= *flags
& BDRV_O_PROTOCOL
;
1628 bool parse_filename
= false;
1629 BlockDriver
*drv
= NULL
;
1630 Error
*local_err
= NULL
;
1633 * Caution: while qdict_get_try_str() is fine, getting non-string
1634 * types would require more care. When @options come from
1635 * -blockdev or blockdev_add, its members are typed according to
1636 * the QAPI schema, but when they come from -drive, they're all
1639 drvname
= qdict_get_try_str(*options
, "driver");
1641 drv
= bdrv_find_format(drvname
);
1643 error_setg(errp
, "Unknown driver '%s'", drvname
);
1646 /* If the user has explicitly specified the driver, this choice should
1647 * override the BDRV_O_PROTOCOL flag */
1648 protocol
= drv
->bdrv_file_open
;
1652 *flags
|= BDRV_O_PROTOCOL
;
1654 *flags
&= ~BDRV_O_PROTOCOL
;
1657 /* Translate cache options from flags into options */
1658 update_options_from_flags(*options
, *flags
);
1660 /* Fetch the file name from the options QDict if necessary */
1661 if (protocol
&& filename
) {
1662 if (!qdict_haskey(*options
, "filename")) {
1663 qdict_put_str(*options
, "filename", filename
);
1664 parse_filename
= true;
1666 error_setg(errp
, "Can't specify 'file' and 'filename' options at "
1672 /* Find the right block driver */
1673 /* See cautionary note on accessing @options above */
1674 filename
= qdict_get_try_str(*options
, "filename");
1676 if (!drvname
&& protocol
) {
1678 drv
= bdrv_find_protocol(filename
, parse_filename
, errp
);
1683 drvname
= drv
->format_name
;
1684 qdict_put_str(*options
, "driver", drvname
);
1686 error_setg(errp
, "Must specify either driver or file");
1691 assert(drv
|| !protocol
);
1693 /* Driver-specific filename parsing */
1694 if (drv
&& drv
->bdrv_parse_filename
&& parse_filename
) {
1695 drv
->bdrv_parse_filename(filename
, *options
, &local_err
);
1697 error_propagate(errp
, local_err
);
1701 if (!drv
->bdrv_needs_filename
) {
1702 qdict_del(*options
, "filename");
1709 static int bdrv_child_check_perm(BdrvChild
*c
, BlockReopenQueue
*q
,
1710 uint64_t perm
, uint64_t shared
,
1711 GSList
*ignore_children
,
1712 bool *tighten_restrictions
, Error
**errp
);
1713 static void bdrv_child_abort_perm_update(BdrvChild
*c
);
1714 static void bdrv_child_set_perm(BdrvChild
*c
, uint64_t perm
, uint64_t shared
);
1715 static void bdrv_get_cumulative_perm(BlockDriverState
*bs
, uint64_t *perm
,
1716 uint64_t *shared_perm
);
1718 typedef struct BlockReopenQueueEntry
{
1721 BDRVReopenState state
;
1722 QSIMPLEQ_ENTRY(BlockReopenQueueEntry
) entry
;
1723 } BlockReopenQueueEntry
;
1726 * Return the flags that @bs will have after the reopens in @q have
1727 * successfully completed. If @q is NULL (or @bs is not contained in @q),
1728 * return the current flags.
1730 static int bdrv_reopen_get_flags(BlockReopenQueue
*q
, BlockDriverState
*bs
)
1732 BlockReopenQueueEntry
*entry
;
1735 QSIMPLEQ_FOREACH(entry
, q
, entry
) {
1736 if (entry
->state
.bs
== bs
) {
1737 return entry
->state
.flags
;
1742 return bs
->open_flags
;
1745 /* Returns whether the image file can be written to after the reopen queue @q
1746 * has been successfully applied, or right now if @q is NULL. */
1747 static bool bdrv_is_writable_after_reopen(BlockDriverState
*bs
,
1748 BlockReopenQueue
*q
)
1750 int flags
= bdrv_reopen_get_flags(q
, bs
);
1752 return (flags
& (BDRV_O_RDWR
| BDRV_O_INACTIVE
)) == BDRV_O_RDWR
;
1756 * Return whether the BDS can be written to. This is not necessarily
1757 * the same as !bdrv_is_read_only(bs), as inactivated images may not
1758 * be written to but do not count as read-only images.
1760 bool bdrv_is_writable(BlockDriverState
*bs
)
1762 return bdrv_is_writable_after_reopen(bs
, NULL
);
1765 static void bdrv_child_perm(BlockDriverState
*bs
, BlockDriverState
*child_bs
,
1766 BdrvChild
*c
, const BdrvChildRole
*role
,
1767 BlockReopenQueue
*reopen_queue
,
1768 uint64_t parent_perm
, uint64_t parent_shared
,
1769 uint64_t *nperm
, uint64_t *nshared
)
1771 assert(bs
->drv
&& bs
->drv
->bdrv_child_perm
);
1772 bs
->drv
->bdrv_child_perm(bs
, c
, role
, reopen_queue
,
1773 parent_perm
, parent_shared
,
1775 /* TODO Take force_share from reopen_queue */
1776 if (child_bs
&& child_bs
->force_share
) {
1777 *nshared
= BLK_PERM_ALL
;
1782 * Check whether permissions on this node can be changed in a way that
1783 * @cumulative_perms and @cumulative_shared_perms are the new cumulative
1784 * permissions of all its parents. This involves checking whether all necessary
1785 * permission changes to child nodes can be performed.
1787 * Will set *tighten_restrictions to true if and only if new permissions have to
1788 * be taken or currently shared permissions are to be unshared. Otherwise,
1789 * errors are not fatal as long as the caller accepts that the restrictions
1790 * remain tighter than they need to be. The caller still has to abort the
1792 * @tighten_restrictions cannot be used together with @q: When reopening, we may
1793 * encounter fatal errors even though no restrictions are to be tightened. For
1794 * example, changing a node from RW to RO will fail if the WRITE permission is
1797 * A call to this function must always be followed by a call to bdrv_set_perm()
1798 * or bdrv_abort_perm_update().
1800 static int bdrv_check_perm(BlockDriverState
*bs
, BlockReopenQueue
*q
,
1801 uint64_t cumulative_perms
,
1802 uint64_t cumulative_shared_perms
,
1803 GSList
*ignore_children
,
1804 bool *tighten_restrictions
, Error
**errp
)
1806 BlockDriver
*drv
= bs
->drv
;
1810 assert(!q
|| !tighten_restrictions
);
1812 if (tighten_restrictions
) {
1813 uint64_t current_perms
, current_shared
;
1814 uint64_t added_perms
, removed_shared_perms
;
1816 bdrv_get_cumulative_perm(bs
, ¤t_perms
, ¤t_shared
);
1818 added_perms
= cumulative_perms
& ~current_perms
;
1819 removed_shared_perms
= current_shared
& ~cumulative_shared_perms
;
1821 *tighten_restrictions
= added_perms
|| removed_shared_perms
;
1824 /* Write permissions never work with read-only images */
1825 if ((cumulative_perms
& (BLK_PERM_WRITE
| BLK_PERM_WRITE_UNCHANGED
)) &&
1826 !bdrv_is_writable_after_reopen(bs
, q
))
1828 if (!bdrv_is_writable_after_reopen(bs
, NULL
)) {
1829 error_setg(errp
, "Block node is read-only");
1831 uint64_t current_perms
, current_shared
;
1832 bdrv_get_cumulative_perm(bs
, ¤t_perms
, ¤t_shared
);
1833 if (current_perms
& (BLK_PERM_WRITE
| BLK_PERM_WRITE_UNCHANGED
)) {
1834 error_setg(errp
, "Cannot make block node read-only, there is "
1837 error_setg(errp
, "Cannot make block node read-only and create "
1845 /* Check this node */
1850 if (drv
->bdrv_check_perm
) {
1851 return drv
->bdrv_check_perm(bs
, cumulative_perms
,
1852 cumulative_shared_perms
, errp
);
1855 /* Drivers that never have children can omit .bdrv_child_perm() */
1856 if (!drv
->bdrv_child_perm
) {
1857 assert(QLIST_EMPTY(&bs
->children
));
1861 /* Check all children */
1862 QLIST_FOREACH(c
, &bs
->children
, next
) {
1863 uint64_t cur_perm
, cur_shared
;
1864 bool child_tighten_restr
;
1866 bdrv_child_perm(bs
, c
->bs
, c
, c
->role
, q
,
1867 cumulative_perms
, cumulative_shared_perms
,
1868 &cur_perm
, &cur_shared
);
1869 ret
= bdrv_child_check_perm(c
, q
, cur_perm
, cur_shared
, ignore_children
,
1870 tighten_restrictions
? &child_tighten_restr
1873 if (tighten_restrictions
) {
1874 *tighten_restrictions
|= child_tighten_restr
;
1885 * Notifies drivers that after a previous bdrv_check_perm() call, the
1886 * permission update is not performed and any preparations made for it (e.g.
1887 * taken file locks) need to be undone.
1889 * This function recursively notifies all child nodes.
1891 static void bdrv_abort_perm_update(BlockDriverState
*bs
)
1893 BlockDriver
*drv
= bs
->drv
;
1900 if (drv
->bdrv_abort_perm_update
) {
1901 drv
->bdrv_abort_perm_update(bs
);
1904 QLIST_FOREACH(c
, &bs
->children
, next
) {
1905 bdrv_child_abort_perm_update(c
);
1909 static void bdrv_set_perm(BlockDriverState
*bs
, uint64_t cumulative_perms
,
1910 uint64_t cumulative_shared_perms
)
1912 BlockDriver
*drv
= bs
->drv
;
1919 /* Update this node */
1920 if (drv
->bdrv_set_perm
) {
1921 drv
->bdrv_set_perm(bs
, cumulative_perms
, cumulative_shared_perms
);
1924 /* Drivers that never have children can omit .bdrv_child_perm() */
1925 if (!drv
->bdrv_child_perm
) {
1926 assert(QLIST_EMPTY(&bs
->children
));
1930 /* Update all children */
1931 QLIST_FOREACH(c
, &bs
->children
, next
) {
1932 uint64_t cur_perm
, cur_shared
;
1933 bdrv_child_perm(bs
, c
->bs
, c
, c
->role
, NULL
,
1934 cumulative_perms
, cumulative_shared_perms
,
1935 &cur_perm
, &cur_shared
);
1936 bdrv_child_set_perm(c
, cur_perm
, cur_shared
);
1940 static void bdrv_get_cumulative_perm(BlockDriverState
*bs
, uint64_t *perm
,
1941 uint64_t *shared_perm
)
1944 uint64_t cumulative_perms
= 0;
1945 uint64_t cumulative_shared_perms
= BLK_PERM_ALL
;
1947 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
1948 cumulative_perms
|= c
->perm
;
1949 cumulative_shared_perms
&= c
->shared_perm
;
1952 *perm
= cumulative_perms
;
1953 *shared_perm
= cumulative_shared_perms
;
1956 static char *bdrv_child_user_desc(BdrvChild
*c
)
1958 if (c
->role
->get_parent_desc
) {
1959 return c
->role
->get_parent_desc(c
);
1962 return g_strdup("another user");
1965 char *bdrv_perm_names(uint64_t perm
)
1971 { BLK_PERM_CONSISTENT_READ
, "consistent read" },
1972 { BLK_PERM_WRITE
, "write" },
1973 { BLK_PERM_WRITE_UNCHANGED
, "write unchanged" },
1974 { BLK_PERM_RESIZE
, "resize" },
1975 { BLK_PERM_GRAPH_MOD
, "change children" },
1979 char *result
= g_strdup("");
1980 struct perm_name
*p
;
1982 for (p
= permissions
; p
->name
; p
++) {
1983 if (perm
& p
->perm
) {
1985 result
= g_strdup_printf("%s%s%s", old
, *old
? ", " : "", p
->name
);
1994 * Checks whether a new reference to @bs can be added if the new user requires
1995 * @new_used_perm/@new_shared_perm as its permissions. If @ignore_children is
1996 * set, the BdrvChild objects in this list are ignored in the calculations;
1997 * this allows checking permission updates for an existing reference.
1999 * See bdrv_check_perm() for the semantics of @tighten_restrictions.
2001 * Needs to be followed by a call to either bdrv_set_perm() or
2002 * bdrv_abort_perm_update(). */
2003 static int bdrv_check_update_perm(BlockDriverState
*bs
, BlockReopenQueue
*q
,
2004 uint64_t new_used_perm
,
2005 uint64_t new_shared_perm
,
2006 GSList
*ignore_children
,
2007 bool *tighten_restrictions
,
2011 uint64_t cumulative_perms
= new_used_perm
;
2012 uint64_t cumulative_shared_perms
= new_shared_perm
;
2014 assert(!q
|| !tighten_restrictions
);
2016 /* There is no reason why anyone couldn't tolerate write_unchanged */
2017 assert(new_shared_perm
& BLK_PERM_WRITE_UNCHANGED
);
2019 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
2020 if (g_slist_find(ignore_children
, c
)) {
2024 if ((new_used_perm
& c
->shared_perm
) != new_used_perm
) {
2025 char *user
= bdrv_child_user_desc(c
);
2026 char *perm_names
= bdrv_perm_names(new_used_perm
& ~c
->shared_perm
);
2028 if (tighten_restrictions
) {
2029 *tighten_restrictions
= true;
2032 error_setg(errp
, "Conflicts with use by %s as '%s', which does not "
2034 user
, c
->name
, perm_names
, bdrv_get_node_name(c
->bs
));
2040 if ((c
->perm
& new_shared_perm
) != c
->perm
) {
2041 char *user
= bdrv_child_user_desc(c
);
2042 char *perm_names
= bdrv_perm_names(c
->perm
& ~new_shared_perm
);
2044 if (tighten_restrictions
) {
2045 *tighten_restrictions
= true;
2048 error_setg(errp
, "Conflicts with use by %s as '%s', which uses "
2050 user
, c
->name
, perm_names
, bdrv_get_node_name(c
->bs
));
2056 cumulative_perms
|= c
->perm
;
2057 cumulative_shared_perms
&= c
->shared_perm
;
2060 return bdrv_check_perm(bs
, q
, cumulative_perms
, cumulative_shared_perms
,
2061 ignore_children
, tighten_restrictions
, errp
);
2064 /* Needs to be followed by a call to either bdrv_child_set_perm() or
2065 * bdrv_child_abort_perm_update(). */
2066 static int bdrv_child_check_perm(BdrvChild
*c
, BlockReopenQueue
*q
,
2067 uint64_t perm
, uint64_t shared
,
2068 GSList
*ignore_children
,
2069 bool *tighten_restrictions
, Error
**errp
)
2073 ignore_children
= g_slist_prepend(g_slist_copy(ignore_children
), c
);
2074 ret
= bdrv_check_update_perm(c
->bs
, q
, perm
, shared
, ignore_children
,
2075 tighten_restrictions
, errp
);
2076 g_slist_free(ignore_children
);
2082 if (!c
->has_backup_perm
) {
2083 c
->has_backup_perm
= true;
2084 c
->backup_perm
= c
->perm
;
2085 c
->backup_shared_perm
= c
->shared_perm
;
2088 * Note: it's OK if c->has_backup_perm was already set, as we can find the
2089 * same child twice during check_perm procedure
2093 c
->shared_perm
= shared
;
2098 static void bdrv_child_set_perm(BdrvChild
*c
, uint64_t perm
, uint64_t shared
)
2100 uint64_t cumulative_perms
, cumulative_shared_perms
;
2102 c
->has_backup_perm
= false;
2105 c
->shared_perm
= shared
;
2107 bdrv_get_cumulative_perm(c
->bs
, &cumulative_perms
,
2108 &cumulative_shared_perms
);
2109 bdrv_set_perm(c
->bs
, cumulative_perms
, cumulative_shared_perms
);
2112 static void bdrv_child_abort_perm_update(BdrvChild
*c
)
2114 if (c
->has_backup_perm
) {
2115 c
->perm
= c
->backup_perm
;
2116 c
->shared_perm
= c
->backup_shared_perm
;
2117 c
->has_backup_perm
= false;
2120 bdrv_abort_perm_update(c
->bs
);
2123 int bdrv_child_try_set_perm(BdrvChild
*c
, uint64_t perm
, uint64_t shared
,
2126 Error
*local_err
= NULL
;
2128 bool tighten_restrictions
;
2130 ret
= bdrv_child_check_perm(c
, NULL
, perm
, shared
, NULL
,
2131 &tighten_restrictions
, &local_err
);
2133 bdrv_child_abort_perm_update(c
);
2134 if (tighten_restrictions
) {
2135 error_propagate(errp
, local_err
);
2138 * Our caller may intend to only loosen restrictions and
2139 * does not expect this function to fail. Errors are not
2140 * fatal in such a case, so we can just hide them from our
2143 error_free(local_err
);
2149 bdrv_child_set_perm(c
, perm
, shared
);
2154 int bdrv_child_refresh_perms(BlockDriverState
*bs
, BdrvChild
*c
, Error
**errp
)
2156 uint64_t parent_perms
, parent_shared
;
2157 uint64_t perms
, shared
;
2159 bdrv_get_cumulative_perm(bs
, &parent_perms
, &parent_shared
);
2160 bdrv_child_perm(bs
, c
->bs
, c
, c
->role
, NULL
, parent_perms
, parent_shared
,
2163 return bdrv_child_try_set_perm(c
, perms
, shared
, errp
);
2166 void bdrv_filter_default_perms(BlockDriverState
*bs
, BdrvChild
*c
,
2167 const BdrvChildRole
*role
,
2168 BlockReopenQueue
*reopen_queue
,
2169 uint64_t perm
, uint64_t shared
,
2170 uint64_t *nperm
, uint64_t *nshared
)
2173 *nperm
= perm
& DEFAULT_PERM_PASSTHROUGH
;
2174 *nshared
= (shared
& DEFAULT_PERM_PASSTHROUGH
) | DEFAULT_PERM_UNCHANGED
;
2178 *nperm
= (perm
& DEFAULT_PERM_PASSTHROUGH
) |
2179 (c
->perm
& DEFAULT_PERM_UNCHANGED
);
2180 *nshared
= (shared
& DEFAULT_PERM_PASSTHROUGH
) |
2181 (c
->shared_perm
& DEFAULT_PERM_UNCHANGED
);
2184 void bdrv_format_default_perms(BlockDriverState
*bs
, BdrvChild
*c
,
2185 const BdrvChildRole
*role
,
2186 BlockReopenQueue
*reopen_queue
,
2187 uint64_t perm
, uint64_t shared
,
2188 uint64_t *nperm
, uint64_t *nshared
)
2190 bool backing
= (role
== &child_backing
);
2191 assert(role
== &child_backing
|| role
== &child_file
);
2194 int flags
= bdrv_reopen_get_flags(reopen_queue
, bs
);
2196 /* Apart from the modifications below, the same permissions are
2197 * forwarded and left alone as for filters */
2198 bdrv_filter_default_perms(bs
, c
, role
, reopen_queue
, perm
, shared
,
2201 /* Format drivers may touch metadata even if the guest doesn't write */
2202 if (bdrv_is_writable_after_reopen(bs
, reopen_queue
)) {
2203 perm
|= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2206 /* bs->file always needs to be consistent because of the metadata. We
2207 * can never allow other users to resize or write to it. */
2208 if (!(flags
& BDRV_O_NO_IO
)) {
2209 perm
|= BLK_PERM_CONSISTENT_READ
;
2211 shared
&= ~(BLK_PERM_WRITE
| BLK_PERM_RESIZE
);
2213 /* We want consistent read from backing files if the parent needs it.
2214 * No other operations are performed on backing files. */
2215 perm
&= BLK_PERM_CONSISTENT_READ
;
2217 /* If the parent can deal with changing data, we're okay with a
2218 * writable and resizable backing file. */
2219 /* TODO Require !(perm & BLK_PERM_CONSISTENT_READ), too? */
2220 if (shared
& BLK_PERM_WRITE
) {
2221 shared
= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2226 shared
|= BLK_PERM_CONSISTENT_READ
| BLK_PERM_GRAPH_MOD
|
2227 BLK_PERM_WRITE_UNCHANGED
;
2230 if (bs
->open_flags
& BDRV_O_INACTIVE
) {
2231 shared
|= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2238 static void bdrv_replace_child_noperm(BdrvChild
*child
,
2239 BlockDriverState
*new_bs
)
2241 BlockDriverState
*old_bs
= child
->bs
;
2244 assert(!child
->frozen
);
2246 if (old_bs
&& new_bs
) {
2247 assert(bdrv_get_aio_context(old_bs
) == bdrv_get_aio_context(new_bs
));
2250 /* Detach first so that the recursive drain sections coming from @child
2251 * are already gone and we only end the drain sections that came from
2253 if (child
->role
->detach
) {
2254 child
->role
->detach(child
);
2256 while (child
->parent_quiesce_counter
) {
2257 bdrv_parent_drained_end_single(child
);
2259 QLIST_REMOVE(child
, next_parent
);
2261 assert(child
->parent_quiesce_counter
== 0);
2267 QLIST_INSERT_HEAD(&new_bs
->parents
, child
, next_parent
);
2268 if (new_bs
->quiesce_counter
) {
2269 int num
= new_bs
->quiesce_counter
;
2270 if (child
->role
->parent_is_bds
) {
2271 num
-= bdrv_drain_all_count
;
2274 for (i
= 0; i
< num
; i
++) {
2275 bdrv_parent_drained_begin_single(child
, true);
2279 /* Attach only after starting new drained sections, so that recursive
2280 * drain sections coming from @child don't get an extra .drained_begin
2282 if (child
->role
->attach
) {
2283 child
->role
->attach(child
);
2289 * Updates @child to change its reference to point to @new_bs, including
2290 * checking and applying the necessary permisson updates both to the old node
2293 * NULL is passed as @new_bs for removing the reference before freeing @child.
2295 * If @new_bs is not NULL, bdrv_check_perm() must be called beforehand, as this
2296 * function uses bdrv_set_perm() to update the permissions according to the new
2297 * reference that @new_bs gets.
2299 static void bdrv_replace_child(BdrvChild
*child
, BlockDriverState
*new_bs
)
2301 BlockDriverState
*old_bs
= child
->bs
;
2302 uint64_t perm
, shared_perm
;
2304 bdrv_replace_child_noperm(child
, new_bs
);
2307 * Start with the new node's permissions. If @new_bs is a (direct
2308 * or indirect) child of @old_bs, we must complete the permission
2309 * update on @new_bs before we loosen the restrictions on @old_bs.
2310 * Otherwise, bdrv_check_perm() on @old_bs would re-initiate
2311 * updating the permissions of @new_bs, and thus not purely loosen
2315 bdrv_get_cumulative_perm(new_bs
, &perm
, &shared_perm
);
2316 bdrv_set_perm(new_bs
, perm
, shared_perm
);
2320 /* Update permissions for old node. This is guaranteed to succeed
2321 * because we're just taking a parent away, so we're loosening
2323 bool tighten_restrictions
;
2326 bdrv_get_cumulative_perm(old_bs
, &perm
, &shared_perm
);
2327 ret
= bdrv_check_perm(old_bs
, NULL
, perm
, shared_perm
, NULL
,
2328 &tighten_restrictions
, NULL
);
2329 assert(tighten_restrictions
== false);
2331 /* We only tried to loosen restrictions, so errors are not fatal */
2332 bdrv_abort_perm_update(old_bs
);
2334 bdrv_set_perm(old_bs
, perm
, shared_perm
);
2337 /* When the parent requiring a non-default AioContext is removed, the
2338 * node moves back to the main AioContext */
2339 bdrv_try_set_aio_context(old_bs
, qemu_get_aio_context(), NULL
);
2344 * This function steals the reference to child_bs from the caller.
2345 * That reference is later dropped by bdrv_root_unref_child().
2347 * On failure NULL is returned, errp is set and the reference to
2348 * child_bs is also dropped.
2350 * The caller must hold the AioContext lock @child_bs, but not that of @ctx
2351 * (unless @child_bs is already in @ctx).
2353 BdrvChild
*bdrv_root_attach_child(BlockDriverState
*child_bs
,
2354 const char *child_name
,
2355 const BdrvChildRole
*child_role
,
2357 uint64_t perm
, uint64_t shared_perm
,
2358 void *opaque
, Error
**errp
)
2361 Error
*local_err
= NULL
;
2364 ret
= bdrv_check_update_perm(child_bs
, NULL
, perm
, shared_perm
, NULL
, NULL
,
2367 bdrv_abort_perm_update(child_bs
);
2368 bdrv_unref(child_bs
);
2372 child
= g_new(BdrvChild
, 1);
2373 *child
= (BdrvChild
) {
2375 .name
= g_strdup(child_name
),
2378 .shared_perm
= shared_perm
,
2382 /* If the AioContexts don't match, first try to move the subtree of
2383 * child_bs into the AioContext of the new parent. If this doesn't work,
2384 * try moving the parent into the AioContext of child_bs instead. */
2385 if (bdrv_get_aio_context(child_bs
) != ctx
) {
2386 ret
= bdrv_try_set_aio_context(child_bs
, ctx
, &local_err
);
2387 if (ret
< 0 && child_role
->can_set_aio_ctx
) {
2388 GSList
*ignore
= g_slist_prepend(NULL
, child
);;
2389 ctx
= bdrv_get_aio_context(child_bs
);
2390 if (child_role
->can_set_aio_ctx(child
, ctx
, &ignore
, NULL
)) {
2391 error_free(local_err
);
2393 g_slist_free(ignore
);
2394 ignore
= g_slist_prepend(NULL
, child
);;
2395 child_role
->set_aio_ctx(child
, ctx
, &ignore
);
2397 g_slist_free(ignore
);
2400 error_propagate(errp
, local_err
);
2402 bdrv_abort_perm_update(child_bs
);
2407 /* This performs the matching bdrv_set_perm() for the above check. */
2408 bdrv_replace_child(child
, child_bs
);
2414 * This function transfers the reference to child_bs from the caller
2415 * to parent_bs. That reference is later dropped by parent_bs on
2416 * bdrv_close() or if someone calls bdrv_unref_child().
2418 * On failure NULL is returned, errp is set and the reference to
2419 * child_bs is also dropped.
2421 * If @parent_bs and @child_bs are in different AioContexts, the caller must
2422 * hold the AioContext lock for @child_bs, but not for @parent_bs.
2424 BdrvChild
*bdrv_attach_child(BlockDriverState
*parent_bs
,
2425 BlockDriverState
*child_bs
,
2426 const char *child_name
,
2427 const BdrvChildRole
*child_role
,
2431 uint64_t perm
, shared_perm
;
2433 bdrv_get_cumulative_perm(parent_bs
, &perm
, &shared_perm
);
2435 assert(parent_bs
->drv
);
2436 bdrv_child_perm(parent_bs
, child_bs
, NULL
, child_role
, NULL
,
2437 perm
, shared_perm
, &perm
, &shared_perm
);
2439 child
= bdrv_root_attach_child(child_bs
, child_name
, child_role
,
2440 bdrv_get_aio_context(parent_bs
),
2441 perm
, shared_perm
, parent_bs
, errp
);
2442 if (child
== NULL
) {
2446 QLIST_INSERT_HEAD(&parent_bs
->children
, child
, next
);
2450 static void bdrv_detach_child(BdrvChild
*child
)
2452 if (child
->next
.le_prev
) {
2453 QLIST_REMOVE(child
, next
);
2454 child
->next
.le_prev
= NULL
;
2457 bdrv_replace_child(child
, NULL
);
2459 g_free(child
->name
);
2463 void bdrv_root_unref_child(BdrvChild
*child
)
2465 BlockDriverState
*child_bs
;
2467 child_bs
= child
->bs
;
2468 bdrv_detach_child(child
);
2469 bdrv_unref(child_bs
);
2473 * Clear all inherits_from pointers from children and grandchildren of
2474 * @root that point to @root, where necessary.
2476 static void bdrv_unset_inherits_from(BlockDriverState
*root
, BdrvChild
*child
)
2480 if (child
->bs
->inherits_from
== root
) {
2482 * Remove inherits_from only when the last reference between root and
2483 * child->bs goes away.
2485 QLIST_FOREACH(c
, &root
->children
, next
) {
2486 if (c
!= child
&& c
->bs
== child
->bs
) {
2491 child
->bs
->inherits_from
= NULL
;
2495 QLIST_FOREACH(c
, &child
->bs
->children
, next
) {
2496 bdrv_unset_inherits_from(root
, c
);
2500 void bdrv_unref_child(BlockDriverState
*parent
, BdrvChild
*child
)
2502 if (child
== NULL
) {
2506 bdrv_unset_inherits_from(parent
, child
);
2507 bdrv_root_unref_child(child
);
2511 static void bdrv_parent_cb_change_media(BlockDriverState
*bs
, bool load
)
2514 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
2515 if (c
->role
->change_media
) {
2516 c
->role
->change_media(c
, load
);
2521 /* Return true if you can reach parent going through child->inherits_from
2522 * recursively. If parent or child are NULL, return false */
2523 static bool bdrv_inherits_from_recursive(BlockDriverState
*child
,
2524 BlockDriverState
*parent
)
2526 while (child
&& child
!= parent
) {
2527 child
= child
->inherits_from
;
2530 return child
!= NULL
;
2534 * Sets the backing file link of a BDS. A new reference is created; callers
2535 * which don't need their own reference any more must call bdrv_unref().
2537 void bdrv_set_backing_hd(BlockDriverState
*bs
, BlockDriverState
*backing_hd
,
2540 bool update_inherits_from
= bdrv_chain_contains(bs
, backing_hd
) &&
2541 bdrv_inherits_from_recursive(backing_hd
, bs
);
2543 if (bdrv_is_backing_chain_frozen(bs
, backing_bs(bs
), errp
)) {
2548 bdrv_ref(backing_hd
);
2552 bdrv_unref_child(bs
, bs
->backing
);
2560 bs
->backing
= bdrv_attach_child(bs
, backing_hd
, "backing", &child_backing
,
2562 /* If backing_hd was already part of bs's backing chain, and
2563 * inherits_from pointed recursively to bs then let's update it to
2564 * point directly to bs (else it will become NULL). */
2565 if (bs
->backing
&& update_inherits_from
) {
2566 backing_hd
->inherits_from
= bs
;
2570 bdrv_refresh_limits(bs
, NULL
);
2574 * Opens the backing file for a BlockDriverState if not yet open
2576 * bdref_key specifies the key for the image's BlockdevRef in the options QDict.
2577 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
2578 * itself, all options starting with "${bdref_key}." are considered part of the
2581 * TODO Can this be unified with bdrv_open_image()?
2583 int bdrv_open_backing_file(BlockDriverState
*bs
, QDict
*parent_options
,
2584 const char *bdref_key
, Error
**errp
)
2586 char *backing_filename
= NULL
;
2587 char *bdref_key_dot
;
2588 const char *reference
= NULL
;
2590 bool implicit_backing
= false;
2591 BlockDriverState
*backing_hd
;
2593 QDict
*tmp_parent_options
= NULL
;
2594 Error
*local_err
= NULL
;
2596 if (bs
->backing
!= NULL
) {
2600 /* NULL means an empty set of options */
2601 if (parent_options
== NULL
) {
2602 tmp_parent_options
= qdict_new();
2603 parent_options
= tmp_parent_options
;
2606 bs
->open_flags
&= ~BDRV_O_NO_BACKING
;
2608 bdref_key_dot
= g_strdup_printf("%s.", bdref_key
);
2609 qdict_extract_subqdict(parent_options
, &options
, bdref_key_dot
);
2610 g_free(bdref_key_dot
);
2613 * Caution: while qdict_get_try_str() is fine, getting non-string
2614 * types would require more care. When @parent_options come from
2615 * -blockdev or blockdev_add, its members are typed according to
2616 * the QAPI schema, but when they come from -drive, they're all
2619 reference
= qdict_get_try_str(parent_options
, bdref_key
);
2620 if (reference
|| qdict_haskey(options
, "file.filename")) {
2621 /* keep backing_filename NULL */
2622 } else if (bs
->backing_file
[0] == '\0' && qdict_size(options
) == 0) {
2623 qobject_unref(options
);
2626 if (qdict_size(options
) == 0) {
2627 /* If the user specifies options that do not modify the
2628 * backing file's behavior, we might still consider it the
2629 * implicit backing file. But it's easier this way, and
2630 * just specifying some of the backing BDS's options is
2631 * only possible with -drive anyway (otherwise the QAPI
2632 * schema forces the user to specify everything). */
2633 implicit_backing
= !strcmp(bs
->auto_backing_file
, bs
->backing_file
);
2636 backing_filename
= bdrv_get_full_backing_filename(bs
, &local_err
);
2639 error_propagate(errp
, local_err
);
2640 qobject_unref(options
);
2645 if (!bs
->drv
|| !bs
->drv
->supports_backing
) {
2647 error_setg(errp
, "Driver doesn't support backing files");
2648 qobject_unref(options
);
2653 bs
->backing_format
[0] != '\0' && !qdict_haskey(options
, "driver")) {
2654 qdict_put_str(options
, "driver", bs
->backing_format
);
2657 backing_hd
= bdrv_open_inherit(backing_filename
, reference
, options
, 0, bs
,
2658 &child_backing
, errp
);
2660 bs
->open_flags
|= BDRV_O_NO_BACKING
;
2661 error_prepend(errp
, "Could not open backing file: ");
2666 if (implicit_backing
) {
2667 bdrv_refresh_filename(backing_hd
);
2668 pstrcpy(bs
->auto_backing_file
, sizeof(bs
->auto_backing_file
),
2669 backing_hd
->filename
);
2672 /* Hook up the backing file link; drop our reference, bs owns the
2673 * backing_hd reference now */
2674 bdrv_set_backing_hd(bs
, backing_hd
, &local_err
);
2675 bdrv_unref(backing_hd
);
2677 error_propagate(errp
, local_err
);
2682 qdict_del(parent_options
, bdref_key
);
2685 g_free(backing_filename
);
2686 qobject_unref(tmp_parent_options
);
2690 static BlockDriverState
*
2691 bdrv_open_child_bs(const char *filename
, QDict
*options
, const char *bdref_key
,
2692 BlockDriverState
*parent
, const BdrvChildRole
*child_role
,
2693 bool allow_none
, Error
**errp
)
2695 BlockDriverState
*bs
= NULL
;
2696 QDict
*image_options
;
2697 char *bdref_key_dot
;
2698 const char *reference
;
2700 assert(child_role
!= NULL
);
2702 bdref_key_dot
= g_strdup_printf("%s.", bdref_key
);
2703 qdict_extract_subqdict(options
, &image_options
, bdref_key_dot
);
2704 g_free(bdref_key_dot
);
2707 * Caution: while qdict_get_try_str() is fine, getting non-string
2708 * types would require more care. When @options come from
2709 * -blockdev or blockdev_add, its members are typed according to
2710 * the QAPI schema, but when they come from -drive, they're all
2713 reference
= qdict_get_try_str(options
, bdref_key
);
2714 if (!filename
&& !reference
&& !qdict_size(image_options
)) {
2716 error_setg(errp
, "A block device must be specified for \"%s\"",
2719 qobject_unref(image_options
);
2723 bs
= bdrv_open_inherit(filename
, reference
, image_options
, 0,
2724 parent
, child_role
, errp
);
2730 qdict_del(options
, bdref_key
);
2735 * Opens a disk image whose options are given as BlockdevRef in another block
2738 * If allow_none is true, no image will be opened if filename is false and no
2739 * BlockdevRef is given. NULL will be returned, but errp remains unset.
2741 * bdrev_key specifies the key for the image's BlockdevRef in the options QDict.
2742 * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
2743 * itself, all options starting with "${bdref_key}." are considered part of the
2746 * The BlockdevRef will be removed from the options QDict.
2748 BdrvChild
*bdrv_open_child(const char *filename
,
2749 QDict
*options
, const char *bdref_key
,
2750 BlockDriverState
*parent
,
2751 const BdrvChildRole
*child_role
,
2752 bool allow_none
, Error
**errp
)
2754 BlockDriverState
*bs
;
2756 bs
= bdrv_open_child_bs(filename
, options
, bdref_key
, parent
, child_role
,
2762 return bdrv_attach_child(parent
, bs
, bdref_key
, child_role
, errp
);
2765 /* TODO Future callers may need to specify parent/child_role in order for
2766 * option inheritance to work. Existing callers use it for the root node. */
2767 BlockDriverState
*bdrv_open_blockdev_ref(BlockdevRef
*ref
, Error
**errp
)
2769 BlockDriverState
*bs
= NULL
;
2770 Error
*local_err
= NULL
;
2771 QObject
*obj
= NULL
;
2772 QDict
*qdict
= NULL
;
2773 const char *reference
= NULL
;
2776 if (ref
->type
== QTYPE_QSTRING
) {
2777 reference
= ref
->u
.reference
;
2779 BlockdevOptions
*options
= &ref
->u
.definition
;
2780 assert(ref
->type
== QTYPE_QDICT
);
2782 v
= qobject_output_visitor_new(&obj
);
2783 visit_type_BlockdevOptions(v
, NULL
, &options
, &local_err
);
2785 error_propagate(errp
, local_err
);
2788 visit_complete(v
, &obj
);
2790 qdict
= qobject_to(QDict
, obj
);
2791 qdict_flatten(qdict
);
2793 /* bdrv_open_inherit() defaults to the values in bdrv_flags (for
2794 * compatibility with other callers) rather than what we want as the
2795 * real defaults. Apply the defaults here instead. */
2796 qdict_set_default_str(qdict
, BDRV_OPT_CACHE_DIRECT
, "off");
2797 qdict_set_default_str(qdict
, BDRV_OPT_CACHE_NO_FLUSH
, "off");
2798 qdict_set_default_str(qdict
, BDRV_OPT_READ_ONLY
, "off");
2799 qdict_set_default_str(qdict
, BDRV_OPT_AUTO_READ_ONLY
, "off");
2803 bs
= bdrv_open_inherit(NULL
, reference
, qdict
, 0, NULL
, NULL
, errp
);
2812 static BlockDriverState
*bdrv_append_temp_snapshot(BlockDriverState
*bs
,
2814 QDict
*snapshot_options
,
2817 /* TODO: extra byte is a hack to ensure MAX_PATH space on Windows. */
2818 char *tmp_filename
= g_malloc0(PATH_MAX
+ 1);
2820 QemuOpts
*opts
= NULL
;
2821 BlockDriverState
*bs_snapshot
= NULL
;
2822 Error
*local_err
= NULL
;
2825 /* if snapshot, we create a temporary backing file and open it
2826 instead of opening 'filename' directly */
2828 /* Get the required size from the image */
2829 total_size
= bdrv_getlength(bs
);
2830 if (total_size
< 0) {
2831 error_setg_errno(errp
, -total_size
, "Could not get image size");
2835 /* Create the temporary image */
2836 ret
= get_tmp_filename(tmp_filename
, PATH_MAX
+ 1);
2838 error_setg_errno(errp
, -ret
, "Could not get temporary filename");
2842 opts
= qemu_opts_create(bdrv_qcow2
.create_opts
, NULL
, 0,
2844 qemu_opt_set_number(opts
, BLOCK_OPT_SIZE
, total_size
, &error_abort
);
2845 ret
= bdrv_create(&bdrv_qcow2
, tmp_filename
, opts
, errp
);
2846 qemu_opts_del(opts
);
2848 error_prepend(errp
, "Could not create temporary overlay '%s': ",
2853 /* Prepare options QDict for the temporary file */
2854 qdict_put_str(snapshot_options
, "file.driver", "file");
2855 qdict_put_str(snapshot_options
, "file.filename", tmp_filename
);
2856 qdict_put_str(snapshot_options
, "driver", "qcow2");
2858 bs_snapshot
= bdrv_open(NULL
, NULL
, snapshot_options
, flags
, errp
);
2859 snapshot_options
= NULL
;
2864 /* bdrv_append() consumes a strong reference to bs_snapshot
2865 * (i.e. it will call bdrv_unref() on it) even on error, so in
2866 * order to be able to return one, we have to increase
2867 * bs_snapshot's refcount here */
2868 bdrv_ref(bs_snapshot
);
2869 bdrv_append(bs_snapshot
, bs
, &local_err
);
2871 error_propagate(errp
, local_err
);
2877 qobject_unref(snapshot_options
);
2878 g_free(tmp_filename
);
2883 * Opens a disk image (raw, qcow2, vmdk, ...)
2885 * options is a QDict of options to pass to the block drivers, or NULL for an
2886 * empty set of options. The reference to the QDict belongs to the block layer
2887 * after the call (even on failure), so if the caller intends to reuse the
2888 * dictionary, it needs to use qobject_ref() before calling bdrv_open.
2890 * If *pbs is NULL, a new BDS will be created with a pointer to it stored there.
2891 * If it is not NULL, the referenced BDS will be reused.
2893 * The reference parameter may be used to specify an existing block device which
2894 * should be opened. If specified, neither options nor a filename may be given,
2895 * nor can an existing BDS be reused (that is, *pbs has to be NULL).
2897 static BlockDriverState
*bdrv_open_inherit(const char *filename
,
2898 const char *reference
,
2899 QDict
*options
, int flags
,
2900 BlockDriverState
*parent
,
2901 const BdrvChildRole
*child_role
,
2905 BlockBackend
*file
= NULL
;
2906 BlockDriverState
*bs
;
2907 BlockDriver
*drv
= NULL
;
2909 const char *drvname
;
2910 const char *backing
;
2911 Error
*local_err
= NULL
;
2912 QDict
*snapshot_options
= NULL
;
2913 int snapshot_flags
= 0;
2915 assert(!child_role
|| !flags
);
2916 assert(!child_role
== !parent
);
2919 bool options_non_empty
= options
? qdict_size(options
) : false;
2920 qobject_unref(options
);
2922 if (filename
|| options_non_empty
) {
2923 error_setg(errp
, "Cannot reference an existing block device with "
2924 "additional options or a new filename");
2928 bs
= bdrv_lookup_bs(reference
, reference
, errp
);
2939 /* NULL means an empty set of options */
2940 if (options
== NULL
) {
2941 options
= qdict_new();
2944 /* json: syntax counts as explicit options, as if in the QDict */
2945 parse_json_protocol(options
, &filename
, &local_err
);
2950 bs
->explicit_options
= qdict_clone_shallow(options
);
2953 bs
->inherits_from
= parent
;
2954 child_role
->inherit_options(&flags
, options
,
2955 parent
->open_flags
, parent
->options
);
2958 ret
= bdrv_fill_options(&options
, filename
, &flags
, &local_err
);
2964 * Set the BDRV_O_RDWR and BDRV_O_ALLOW_RDWR flags.
2965 * Caution: getting a boolean member of @options requires care.
2966 * When @options come from -blockdev or blockdev_add, members are
2967 * typed according to the QAPI schema, but when they come from
2968 * -drive, they're all QString.
2970 if (g_strcmp0(qdict_get_try_str(options
, BDRV_OPT_READ_ONLY
), "on") &&
2971 !qdict_get_try_bool(options
, BDRV_OPT_READ_ONLY
, false)) {
2972 flags
|= (BDRV_O_RDWR
| BDRV_O_ALLOW_RDWR
);
2974 flags
&= ~BDRV_O_RDWR
;
2977 if (flags
& BDRV_O_SNAPSHOT
) {
2978 snapshot_options
= qdict_new();
2979 bdrv_temp_snapshot_options(&snapshot_flags
, snapshot_options
,
2981 /* Let bdrv_backing_options() override "read-only" */
2982 qdict_del(options
, BDRV_OPT_READ_ONLY
);
2983 bdrv_backing_options(&flags
, options
, flags
, options
);
2986 bs
->open_flags
= flags
;
2987 bs
->options
= options
;
2988 options
= qdict_clone_shallow(options
);
2990 /* Find the right image format driver */
2991 /* See cautionary note on accessing @options above */
2992 drvname
= qdict_get_try_str(options
, "driver");
2994 drv
= bdrv_find_format(drvname
);
2996 error_setg(errp
, "Unknown driver: '%s'", drvname
);
3001 assert(drvname
|| !(flags
& BDRV_O_PROTOCOL
));
3003 /* See cautionary note on accessing @options above */
3004 backing
= qdict_get_try_str(options
, "backing");
3005 if (qobject_to(QNull
, qdict_get(options
, "backing")) != NULL
||
3006 (backing
&& *backing
== '\0'))
3009 warn_report("Use of \"backing\": \"\" is deprecated; "
3010 "use \"backing\": null instead");
3012 flags
|= BDRV_O_NO_BACKING
;
3013 qdict_del(options
, "backing");
3016 /* Open image file without format layer. This BlockBackend is only used for
3017 * probing, the block drivers will do their own bdrv_open_child() for the
3018 * same BDS, which is why we put the node name back into options. */
3019 if ((flags
& BDRV_O_PROTOCOL
) == 0) {
3020 BlockDriverState
*file_bs
;
3022 file_bs
= bdrv_open_child_bs(filename
, options
, "file", bs
,
3023 &child_file
, true, &local_err
);
3027 if (file_bs
!= NULL
) {
3028 /* Not requesting BLK_PERM_CONSISTENT_READ because we're only
3029 * looking at the header to guess the image format. This works even
3030 * in cases where a guest would not see a consistent state. */
3031 file
= blk_new(bdrv_get_aio_context(file_bs
), 0, BLK_PERM_ALL
);
3032 blk_insert_bs(file
, file_bs
, &local_err
);
3033 bdrv_unref(file_bs
);
3038 qdict_put_str(options
, "file", bdrv_get_node_name(file_bs
));
3042 /* Image format probing */
3045 ret
= find_image_format(file
, filename
, &drv
, &local_err
);
3050 * This option update would logically belong in bdrv_fill_options(),
3051 * but we first need to open bs->file for the probing to work, while
3052 * opening bs->file already requires the (mostly) final set of options
3053 * so that cache mode etc. can be inherited.
3055 * Adding the driver later is somewhat ugly, but it's not an option
3056 * that would ever be inherited, so it's correct. We just need to make
3057 * sure to update both bs->options (which has the full effective
3058 * options for bs) and options (which has file.* already removed).
3060 qdict_put_str(bs
->options
, "driver", drv
->format_name
);
3061 qdict_put_str(options
, "driver", drv
->format_name
);
3063 error_setg(errp
, "Must specify either driver or file");
3067 /* BDRV_O_PROTOCOL must be set iff a protocol BDS is about to be created */
3068 assert(!!(flags
& BDRV_O_PROTOCOL
) == !!drv
->bdrv_file_open
);
3069 /* file must be NULL if a protocol BDS is about to be created
3070 * (the inverse results in an error message from bdrv_open_common()) */
3071 assert(!(flags
& BDRV_O_PROTOCOL
) || !file
);
3073 /* Open the image */
3074 ret
= bdrv_open_common(bs
, file
, options
, &local_err
);
3084 /* If there is a backing file, use it */
3085 if ((flags
& BDRV_O_NO_BACKING
) == 0) {
3086 ret
= bdrv_open_backing_file(bs
, options
, "backing", &local_err
);
3088 goto close_and_fail
;
3092 /* Remove all children options and references
3093 * from bs->options and bs->explicit_options */
3094 QLIST_FOREACH(child
, &bs
->children
, next
) {
3095 char *child_key_dot
;
3096 child_key_dot
= g_strdup_printf("%s.", child
->name
);
3097 qdict_extract_subqdict(bs
->explicit_options
, NULL
, child_key_dot
);
3098 qdict_extract_subqdict(bs
->options
, NULL
, child_key_dot
);
3099 qdict_del(bs
->explicit_options
, child
->name
);
3100 qdict_del(bs
->options
, child
->name
);
3101 g_free(child_key_dot
);
3104 /* Check if any unknown options were used */
3105 if (qdict_size(options
) != 0) {
3106 const QDictEntry
*entry
= qdict_first(options
);
3107 if (flags
& BDRV_O_PROTOCOL
) {
3108 error_setg(errp
, "Block protocol '%s' doesn't support the option "
3109 "'%s'", drv
->format_name
, entry
->key
);
3112 "Block format '%s' does not support the option '%s'",
3113 drv
->format_name
, entry
->key
);
3116 goto close_and_fail
;
3119 bdrv_parent_cb_change_media(bs
, true);
3121 qobject_unref(options
);
3124 /* For snapshot=on, create a temporary qcow2 overlay. bs points to the
3125 * temporary snapshot afterwards. */
3126 if (snapshot_flags
) {
3127 BlockDriverState
*snapshot_bs
;
3128 snapshot_bs
= bdrv_append_temp_snapshot(bs
, snapshot_flags
,
3129 snapshot_options
, &local_err
);
3130 snapshot_options
= NULL
;
3132 goto close_and_fail
;
3134 /* We are not going to return bs but the overlay on top of it
3135 * (snapshot_bs); thus, we have to drop the strong reference to bs
3136 * (which we obtained by calling bdrv_new()). bs will not be deleted,
3137 * though, because the overlay still has a reference to it. */
3146 qobject_unref(snapshot_options
);
3147 qobject_unref(bs
->explicit_options
);
3148 qobject_unref(bs
->options
);
3149 qobject_unref(options
);
3151 bs
->explicit_options
= NULL
;
3153 error_propagate(errp
, local_err
);
3158 qobject_unref(snapshot_options
);
3159 qobject_unref(options
);
3160 error_propagate(errp
, local_err
);
3164 BlockDriverState
*bdrv_open(const char *filename
, const char *reference
,
3165 QDict
*options
, int flags
, Error
**errp
)
3167 return bdrv_open_inherit(filename
, reference
, options
, flags
, NULL
,
3171 /* Return true if the NULL-terminated @list contains @str */
3172 static bool is_str_in_list(const char *str
, const char *const *list
)
3176 for (i
= 0; list
[i
] != NULL
; i
++) {
3177 if (!strcmp(str
, list
[i
])) {
3186 * Check that every option set in @bs->options is also set in
3189 * Options listed in the common_options list and in
3190 * @bs->drv->mutable_opts are skipped.
3192 * Return 0 on success, otherwise return -EINVAL and set @errp.
3194 static int bdrv_reset_options_allowed(BlockDriverState
*bs
,
3195 const QDict
*new_opts
, Error
**errp
)
3197 const QDictEntry
*e
;
3198 /* These options are common to all block drivers and are handled
3199 * in bdrv_reopen_prepare() so they can be left out of @new_opts */
3200 const char *const common_options
[] = {
3201 "node-name", "discard", "cache.direct", "cache.no-flush",
3202 "read-only", "auto-read-only", "detect-zeroes", NULL
3205 for (e
= qdict_first(bs
->options
); e
; e
= qdict_next(bs
->options
, e
)) {
3206 if (!qdict_haskey(new_opts
, e
->key
) &&
3207 !is_str_in_list(e
->key
, common_options
) &&
3208 !is_str_in_list(e
->key
, bs
->drv
->mutable_opts
)) {
3209 error_setg(errp
, "Option '%s' cannot be reset "
3210 "to its default value", e
->key
);
3219 * Returns true if @child can be reached recursively from @bs
3221 static bool bdrv_recurse_has_child(BlockDriverState
*bs
,
3222 BlockDriverState
*child
)
3230 QLIST_FOREACH(c
, &bs
->children
, next
) {
3231 if (bdrv_recurse_has_child(c
->bs
, child
)) {
3240 * Adds a BlockDriverState to a simple queue for an atomic, transactional
3241 * reopen of multiple devices.
3243 * bs_queue can either be an existing BlockReopenQueue that has had QSIMPLE_INIT
3244 * already performed, or alternatively may be NULL a new BlockReopenQueue will
3245 * be created and initialized. This newly created BlockReopenQueue should be
3246 * passed back in for subsequent calls that are intended to be of the same
3249 * bs is the BlockDriverState to add to the reopen queue.
3251 * options contains the changed options for the associated bs
3252 * (the BlockReopenQueue takes ownership)
3254 * flags contains the open flags for the associated bs
3256 * returns a pointer to bs_queue, which is either the newly allocated
3257 * bs_queue, or the existing bs_queue being used.
3259 * bs must be drained between bdrv_reopen_queue() and bdrv_reopen_multiple().
3261 static BlockReopenQueue
*bdrv_reopen_queue_child(BlockReopenQueue
*bs_queue
,
3262 BlockDriverState
*bs
,
3264 const BdrvChildRole
*role
,
3265 QDict
*parent_options
,
3271 BlockReopenQueueEntry
*bs_entry
;
3273 QDict
*old_options
, *explicit_options
, *options_copy
;
3277 /* Make sure that the caller remembered to use a drained section. This is
3278 * important to avoid graph changes between the recursive queuing here and
3279 * bdrv_reopen_multiple(). */
3280 assert(bs
->quiesce_counter
> 0);
3282 if (bs_queue
== NULL
) {
3283 bs_queue
= g_new0(BlockReopenQueue
, 1);
3284 QSIMPLEQ_INIT(bs_queue
);
3288 options
= qdict_new();
3291 /* Check if this BlockDriverState is already in the queue */
3292 QSIMPLEQ_FOREACH(bs_entry
, bs_queue
, entry
) {
3293 if (bs
== bs_entry
->state
.bs
) {
3299 * Precedence of options:
3300 * 1. Explicitly passed in options (highest)
3301 * 2. Retained from explicitly set options of bs
3302 * 3. Inherited from parent node
3303 * 4. Retained from effective options of bs
3306 /* Old explicitly set values (don't overwrite by inherited value) */
3307 if (bs_entry
|| keep_old_opts
) {
3308 old_options
= qdict_clone_shallow(bs_entry
?
3309 bs_entry
->state
.explicit_options
:
3310 bs
->explicit_options
);
3311 bdrv_join_options(bs
, options
, old_options
);
3312 qobject_unref(old_options
);
3315 explicit_options
= qdict_clone_shallow(options
);
3317 /* Inherit from parent node */
3318 if (parent_options
) {
3320 role
->inherit_options(&flags
, options
, parent_flags
, parent_options
);
3322 flags
= bdrv_get_flags(bs
);
3325 if (keep_old_opts
) {
3326 /* Old values are used for options that aren't set yet */
3327 old_options
= qdict_clone_shallow(bs
->options
);
3328 bdrv_join_options(bs
, options
, old_options
);
3329 qobject_unref(old_options
);
3332 /* We have the final set of options so let's update the flags */
3333 options_copy
= qdict_clone_shallow(options
);
3334 opts
= qemu_opts_create(&bdrv_runtime_opts
, NULL
, 0, &error_abort
);
3335 qemu_opts_absorb_qdict(opts
, options_copy
, NULL
);
3336 update_flags_from_options(&flags
, opts
);
3337 qemu_opts_del(opts
);
3338 qobject_unref(options_copy
);
3340 /* bdrv_open_inherit() sets and clears some additional flags internally */
3341 flags
&= ~BDRV_O_PROTOCOL
;
3342 if (flags
& BDRV_O_RDWR
) {
3343 flags
|= BDRV_O_ALLOW_RDWR
;
3347 bs_entry
= g_new0(BlockReopenQueueEntry
, 1);
3348 QSIMPLEQ_INSERT_TAIL(bs_queue
, bs_entry
, entry
);
3350 qobject_unref(bs_entry
->state
.options
);
3351 qobject_unref(bs_entry
->state
.explicit_options
);
3354 bs_entry
->state
.bs
= bs
;
3355 bs_entry
->state
.options
= options
;
3356 bs_entry
->state
.explicit_options
= explicit_options
;
3357 bs_entry
->state
.flags
= flags
;
3359 /* This needs to be overwritten in bdrv_reopen_prepare() */
3360 bs_entry
->state
.perm
= UINT64_MAX
;
3361 bs_entry
->state
.shared_perm
= 0;
3364 * If keep_old_opts is false then it means that unspecified
3365 * options must be reset to their original value. We don't allow
3366 * resetting 'backing' but we need to know if the option is
3367 * missing in order to decide if we have to return an error.
3369 if (!keep_old_opts
) {
3370 bs_entry
->state
.backing_missing
=
3371 !qdict_haskey(options
, "backing") &&
3372 !qdict_haskey(options
, "backing.driver");
3375 QLIST_FOREACH(child
, &bs
->children
, next
) {
3376 QDict
*new_child_options
= NULL
;
3377 bool child_keep_old
= keep_old_opts
;
3379 /* reopen can only change the options of block devices that were
3380 * implicitly created and inherited options. For other (referenced)
3381 * block devices, a syntax like "backing.foo" results in an error. */
3382 if (child
->bs
->inherits_from
!= bs
) {
3386 /* Check if the options contain a child reference */
3387 if (qdict_haskey(options
, child
->name
)) {
3388 const char *childref
= qdict_get_try_str(options
, child
->name
);
3390 * The current child must not be reopened if the child
3391 * reference is null or points to a different node.
3393 if (g_strcmp0(childref
, child
->bs
->node_name
)) {
3397 * If the child reference points to the current child then
3398 * reopen it with its existing set of options (note that
3399 * it can still inherit new options from the parent).
3401 child_keep_old
= true;
3403 /* Extract child options ("child-name.*") */
3404 char *child_key_dot
= g_strdup_printf("%s.", child
->name
);
3405 qdict_extract_subqdict(explicit_options
, NULL
, child_key_dot
);
3406 qdict_extract_subqdict(options
, &new_child_options
, child_key_dot
);
3407 g_free(child_key_dot
);
3410 bdrv_reopen_queue_child(bs_queue
, child
->bs
, new_child_options
,
3411 child
->role
, options
, flags
, child_keep_old
);
3417 BlockReopenQueue
*bdrv_reopen_queue(BlockReopenQueue
*bs_queue
,
3418 BlockDriverState
*bs
,
3419 QDict
*options
, bool keep_old_opts
)
3421 return bdrv_reopen_queue_child(bs_queue
, bs
, options
, NULL
, NULL
, 0,
3426 * Reopen multiple BlockDriverStates atomically & transactionally.
3428 * The queue passed in (bs_queue) must have been built up previous
3429 * via bdrv_reopen_queue().
3431 * Reopens all BDS specified in the queue, with the appropriate
3432 * flags. All devices are prepared for reopen, and failure of any
3433 * device will cause all device changes to be abandoned, and intermediate
3436 * If all devices prepare successfully, then the changes are committed
3439 * All affected nodes must be drained between bdrv_reopen_queue() and
3440 * bdrv_reopen_multiple().
3442 int bdrv_reopen_multiple(BlockReopenQueue
*bs_queue
, Error
**errp
)
3445 BlockReopenQueueEntry
*bs_entry
, *next
;
3447 assert(bs_queue
!= NULL
);
3449 QSIMPLEQ_FOREACH(bs_entry
, bs_queue
, entry
) {
3450 assert(bs_entry
->state
.bs
->quiesce_counter
> 0);
3451 if (bdrv_reopen_prepare(&bs_entry
->state
, bs_queue
, errp
)) {
3454 bs_entry
->prepared
= true;
3457 QSIMPLEQ_FOREACH(bs_entry
, bs_queue
, entry
) {
3458 BDRVReopenState
*state
= &bs_entry
->state
;
3459 ret
= bdrv_check_perm(state
->bs
, bs_queue
, state
->perm
,
3460 state
->shared_perm
, NULL
, NULL
, errp
);
3464 /* Check if new_backing_bs would accept the new permissions */
3465 if (state
->replace_backing_bs
&& state
->new_backing_bs
) {
3466 uint64_t nperm
, nshared
;
3467 bdrv_child_perm(state
->bs
, state
->new_backing_bs
,
3468 NULL
, &child_backing
, bs_queue
,
3469 state
->perm
, state
->shared_perm
,
3471 ret
= bdrv_check_update_perm(state
->new_backing_bs
, NULL
,
3472 nperm
, nshared
, NULL
, NULL
, errp
);
3477 bs_entry
->perms_checked
= true;
3480 /* If we reach this point, we have success and just need to apply the
3483 QSIMPLEQ_FOREACH(bs_entry
, bs_queue
, entry
) {
3484 bdrv_reopen_commit(&bs_entry
->state
);
3489 QSIMPLEQ_FOREACH_SAFE(bs_entry
, bs_queue
, entry
, next
) {
3490 BDRVReopenState
*state
= &bs_entry
->state
;
3492 if (!bs_entry
->perms_checked
) {
3497 bdrv_set_perm(state
->bs
, state
->perm
, state
->shared_perm
);
3499 bdrv_abort_perm_update(state
->bs
);
3500 if (state
->replace_backing_bs
&& state
->new_backing_bs
) {
3501 bdrv_abort_perm_update(state
->new_backing_bs
);
3506 QSIMPLEQ_FOREACH_SAFE(bs_entry
, bs_queue
, entry
, next
) {
3508 if (bs_entry
->prepared
) {
3509 bdrv_reopen_abort(&bs_entry
->state
);
3511 qobject_unref(bs_entry
->state
.explicit_options
);
3512 qobject_unref(bs_entry
->state
.options
);
3514 if (bs_entry
->state
.new_backing_bs
) {
3515 bdrv_unref(bs_entry
->state
.new_backing_bs
);
3524 int bdrv_reopen_set_read_only(BlockDriverState
*bs
, bool read_only
,
3528 BlockReopenQueue
*queue
;
3529 QDict
*opts
= qdict_new();
3531 qdict_put_bool(opts
, BDRV_OPT_READ_ONLY
, read_only
);
3533 bdrv_subtree_drained_begin(bs
);
3534 queue
= bdrv_reopen_queue(NULL
, bs
, opts
, true);
3535 ret
= bdrv_reopen_multiple(queue
, errp
);
3536 bdrv_subtree_drained_end(bs
);
3541 static BlockReopenQueueEntry
*find_parent_in_reopen_queue(BlockReopenQueue
*q
,
3544 BlockReopenQueueEntry
*entry
;
3546 QSIMPLEQ_FOREACH(entry
, q
, entry
) {
3547 BlockDriverState
*bs
= entry
->state
.bs
;
3550 QLIST_FOREACH(child
, &bs
->children
, next
) {
3560 static void bdrv_reopen_perm(BlockReopenQueue
*q
, BlockDriverState
*bs
,
3561 uint64_t *perm
, uint64_t *shared
)
3564 BlockReopenQueueEntry
*parent
;
3565 uint64_t cumulative_perms
= 0;
3566 uint64_t cumulative_shared_perms
= BLK_PERM_ALL
;
3568 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
3569 parent
= find_parent_in_reopen_queue(q
, c
);
3571 cumulative_perms
|= c
->perm
;
3572 cumulative_shared_perms
&= c
->shared_perm
;
3574 uint64_t nperm
, nshared
;
3576 bdrv_child_perm(parent
->state
.bs
, bs
, c
, c
->role
, q
,
3577 parent
->state
.perm
, parent
->state
.shared_perm
,
3580 cumulative_perms
|= nperm
;
3581 cumulative_shared_perms
&= nshared
;
3584 *perm
= cumulative_perms
;
3585 *shared
= cumulative_shared_perms
;
3589 * Take a BDRVReopenState and check if the value of 'backing' in the
3590 * reopen_state->options QDict is valid or not.
3592 * If 'backing' is missing from the QDict then return 0.
3594 * If 'backing' contains the node name of the backing file of
3595 * reopen_state->bs then return 0.
3597 * If 'backing' contains a different node name (or is null) then check
3598 * whether the current backing file can be replaced with the new one.
3599 * If that's the case then reopen_state->replace_backing_bs is set to
3600 * true and reopen_state->new_backing_bs contains a pointer to the new
3601 * backing BlockDriverState (or NULL).
3603 * Return 0 on success, otherwise return < 0 and set @errp.
3605 static int bdrv_reopen_parse_backing(BDRVReopenState
*reopen_state
,
3608 BlockDriverState
*bs
= reopen_state
->bs
;
3609 BlockDriverState
*overlay_bs
, *new_backing_bs
;
3613 value
= qdict_get(reopen_state
->options
, "backing");
3614 if (value
== NULL
) {
3618 switch (qobject_type(value
)) {
3620 new_backing_bs
= NULL
;
3623 str
= qobject_get_try_str(value
);
3624 new_backing_bs
= bdrv_lookup_bs(NULL
, str
, errp
);
3625 if (new_backing_bs
== NULL
) {
3627 } else if (bdrv_recurse_has_child(new_backing_bs
, bs
)) {
3628 error_setg(errp
, "Making '%s' a backing file of '%s' "
3629 "would create a cycle", str
, bs
->node_name
);
3634 /* 'backing' does not allow any other data type */
3635 g_assert_not_reached();
3639 * TODO: before removing the x- prefix from x-blockdev-reopen we
3640 * should move the new backing file into the right AioContext
3641 * instead of returning an error.
3643 if (new_backing_bs
) {
3644 if (bdrv_get_aio_context(new_backing_bs
) != bdrv_get_aio_context(bs
)) {
3645 error_setg(errp
, "Cannot use a new backing file "
3646 "with a different AioContext");
3652 * Find the "actual" backing file by skipping all links that point
3653 * to an implicit node, if any (e.g. a commit filter node).
3656 while (backing_bs(overlay_bs
) && backing_bs(overlay_bs
)->implicit
) {
3657 overlay_bs
= backing_bs(overlay_bs
);
3660 /* If we want to replace the backing file we need some extra checks */
3661 if (new_backing_bs
!= backing_bs(overlay_bs
)) {
3662 /* Check for implicit nodes between bs and its backing file */
3663 if (bs
!= overlay_bs
) {
3664 error_setg(errp
, "Cannot change backing link if '%s' has "
3665 "an implicit backing file", bs
->node_name
);
3668 /* Check if the backing link that we want to replace is frozen */
3669 if (bdrv_is_backing_chain_frozen(overlay_bs
, backing_bs(overlay_bs
),
3673 reopen_state
->replace_backing_bs
= true;
3674 if (new_backing_bs
) {
3675 bdrv_ref(new_backing_bs
);
3676 reopen_state
->new_backing_bs
= new_backing_bs
;
3684 * Prepares a BlockDriverState for reopen. All changes are staged in the
3685 * 'opaque' field of the BDRVReopenState, which is used and allocated by
3686 * the block driver layer .bdrv_reopen_prepare()
3688 * bs is the BlockDriverState to reopen
3689 * flags are the new open flags
3690 * queue is the reopen queue
3692 * Returns 0 on success, non-zero on error. On error errp will be set
3695 * On failure, bdrv_reopen_abort() will be called to clean up any data.
3696 * It is the responsibility of the caller to then call the abort() or
3697 * commit() for any other BDS that have been left in a prepare() state
3700 int bdrv_reopen_prepare(BDRVReopenState
*reopen_state
, BlockReopenQueue
*queue
,
3705 Error
*local_err
= NULL
;
3708 QDict
*orig_reopen_opts
;
3709 char *discard
= NULL
;
3711 bool drv_prepared
= false;
3713 assert(reopen_state
!= NULL
);
3714 assert(reopen_state
->bs
->drv
!= NULL
);
3715 drv
= reopen_state
->bs
->drv
;
3717 /* This function and each driver's bdrv_reopen_prepare() remove
3718 * entries from reopen_state->options as they are processed, so
3719 * we need to make a copy of the original QDict. */
3720 orig_reopen_opts
= qdict_clone_shallow(reopen_state
->options
);
3722 /* Process generic block layer options */
3723 opts
= qemu_opts_create(&bdrv_runtime_opts
, NULL
, 0, &error_abort
);
3724 qemu_opts_absorb_qdict(opts
, reopen_state
->options
, &local_err
);
3726 error_propagate(errp
, local_err
);
3731 /* This was already called in bdrv_reopen_queue_child() so the flags
3732 * are up-to-date. This time we simply want to remove the options from
3733 * QemuOpts in order to indicate that they have been processed. */
3734 old_flags
= reopen_state
->flags
;
3735 update_flags_from_options(&reopen_state
->flags
, opts
);
3736 assert(old_flags
== reopen_state
->flags
);
3738 discard
= qemu_opt_get_del(opts
, BDRV_OPT_DISCARD
);
3739 if (discard
!= NULL
) {
3740 if (bdrv_parse_discard_flags(discard
, &reopen_state
->flags
) != 0) {
3741 error_setg(errp
, "Invalid discard option");
3747 reopen_state
->detect_zeroes
=
3748 bdrv_parse_detect_zeroes(opts
, reopen_state
->flags
, &local_err
);
3750 error_propagate(errp
, local_err
);
3755 /* All other options (including node-name and driver) must be unchanged.
3756 * Put them back into the QDict, so that they are checked at the end
3757 * of this function. */
3758 qemu_opts_to_qdict(opts
, reopen_state
->options
);
3760 /* If we are to stay read-only, do not allow permission change
3761 * to r/w. Attempting to set to r/w may fail if either BDRV_O_ALLOW_RDWR is
3762 * not set, or if the BDS still has copy_on_read enabled */
3763 read_only
= !(reopen_state
->flags
& BDRV_O_RDWR
);
3764 ret
= bdrv_can_set_read_only(reopen_state
->bs
, read_only
, true, &local_err
);
3766 error_propagate(errp
, local_err
);
3770 /* Calculate required permissions after reopening */
3771 bdrv_reopen_perm(queue
, reopen_state
->bs
,
3772 &reopen_state
->perm
, &reopen_state
->shared_perm
);
3774 ret
= bdrv_flush(reopen_state
->bs
);
3776 error_setg_errno(errp
, -ret
, "Error flushing drive");
3780 if (drv
->bdrv_reopen_prepare
) {
3782 * If a driver-specific option is missing, it means that we
3783 * should reset it to its default value.
3784 * But not all options allow that, so we need to check it first.
3786 ret
= bdrv_reset_options_allowed(reopen_state
->bs
,
3787 reopen_state
->options
, errp
);
3792 ret
= drv
->bdrv_reopen_prepare(reopen_state
, queue
, &local_err
);
3794 if (local_err
!= NULL
) {
3795 error_propagate(errp
, local_err
);
3797 bdrv_refresh_filename(reopen_state
->bs
);
3798 error_setg(errp
, "failed while preparing to reopen image '%s'",
3799 reopen_state
->bs
->filename
);
3804 /* It is currently mandatory to have a bdrv_reopen_prepare()
3805 * handler for each supported drv. */
3806 error_setg(errp
, "Block format '%s' used by node '%s' "
3807 "does not support reopening files", drv
->format_name
,
3808 bdrv_get_device_or_node_name(reopen_state
->bs
));
3813 drv_prepared
= true;
3816 * We must provide the 'backing' option if the BDS has a backing
3817 * file or if the image file has a backing file name as part of
3818 * its metadata. Otherwise the 'backing' option can be omitted.
3820 if (drv
->supports_backing
&& reopen_state
->backing_missing
&&
3821 (backing_bs(reopen_state
->bs
) || reopen_state
->bs
->backing_file
[0])) {
3822 error_setg(errp
, "backing is missing for '%s'",
3823 reopen_state
->bs
->node_name
);
3829 * Allow changing the 'backing' option. The new value can be
3830 * either a reference to an existing node (using its node name)
3831 * or NULL to simply detach the current backing file.
3833 ret
= bdrv_reopen_parse_backing(reopen_state
, errp
);
3837 qdict_del(reopen_state
->options
, "backing");
3839 /* Options that are not handled are only okay if they are unchanged
3840 * compared to the old state. It is expected that some options are only
3841 * used for the initial open, but not reopen (e.g. filename) */
3842 if (qdict_size(reopen_state
->options
)) {
3843 const QDictEntry
*entry
= qdict_first(reopen_state
->options
);
3846 QObject
*new = entry
->value
;
3847 QObject
*old
= qdict_get(reopen_state
->bs
->options
, entry
->key
);
3849 /* Allow child references (child_name=node_name) as long as they
3850 * point to the current child (i.e. everything stays the same). */
3851 if (qobject_type(new) == QTYPE_QSTRING
) {
3853 QLIST_FOREACH(child
, &reopen_state
->bs
->children
, next
) {
3854 if (!strcmp(child
->name
, entry
->key
)) {
3860 const char *str
= qobject_get_try_str(new);
3861 if (!strcmp(child
->bs
->node_name
, str
)) {
3862 continue; /* Found child with this name, skip option */
3868 * TODO: When using -drive to specify blockdev options, all values
3869 * will be strings; however, when using -blockdev, blockdev-add or
3870 * filenames using the json:{} pseudo-protocol, they will be
3872 * In contrast, reopening options are (currently) always strings
3873 * (because you can only specify them through qemu-io; all other
3874 * callers do not specify any options).
3875 * Therefore, when using anything other than -drive to create a BDS,
3876 * this cannot detect non-string options as unchanged, because
3877 * qobject_is_equal() always returns false for objects of different
3878 * type. In the future, this should be remedied by correctly typing
3879 * all options. For now, this is not too big of an issue because
3880 * the user can simply omit options which cannot be changed anyway,
3881 * so they will stay unchanged.
3883 if (!qobject_is_equal(new, old
)) {
3884 error_setg(errp
, "Cannot change the option '%s'", entry
->key
);
3888 } while ((entry
= qdict_next(reopen_state
->options
, entry
)));
3893 /* Restore the original reopen_state->options QDict */
3894 qobject_unref(reopen_state
->options
);
3895 reopen_state
->options
= qobject_ref(orig_reopen_opts
);
3898 if (ret
< 0 && drv_prepared
) {
3899 /* drv->bdrv_reopen_prepare() has succeeded, so we need to
3900 * call drv->bdrv_reopen_abort() before signaling an error
3901 * (bdrv_reopen_multiple() will not call bdrv_reopen_abort()
3902 * when the respective bdrv_reopen_prepare() has failed) */
3903 if (drv
->bdrv_reopen_abort
) {
3904 drv
->bdrv_reopen_abort(reopen_state
);
3907 qemu_opts_del(opts
);
3908 qobject_unref(orig_reopen_opts
);
3914 * Takes the staged changes for the reopen from bdrv_reopen_prepare(), and
3915 * makes them final by swapping the staging BlockDriverState contents into
3916 * the active BlockDriverState contents.
3918 void bdrv_reopen_commit(BDRVReopenState
*reopen_state
)
3921 BlockDriverState
*bs
;
3923 bool old_can_write
, new_can_write
;
3925 assert(reopen_state
!= NULL
);
3926 bs
= reopen_state
->bs
;
3928 assert(drv
!= NULL
);
3931 !bdrv_is_read_only(bs
) && !(bdrv_get_flags(bs
) & BDRV_O_INACTIVE
);
3933 /* If there are any driver level actions to take */
3934 if (drv
->bdrv_reopen_commit
) {
3935 drv
->bdrv_reopen_commit(reopen_state
);
3938 /* set BDS specific flags now */
3939 qobject_unref(bs
->explicit_options
);
3940 qobject_unref(bs
->options
);
3942 bs
->explicit_options
= reopen_state
->explicit_options
;
3943 bs
->options
= reopen_state
->options
;
3944 bs
->open_flags
= reopen_state
->flags
;
3945 bs
->read_only
= !(reopen_state
->flags
& BDRV_O_RDWR
);
3946 bs
->detect_zeroes
= reopen_state
->detect_zeroes
;
3948 if (reopen_state
->replace_backing_bs
) {
3949 qdict_del(bs
->explicit_options
, "backing");
3950 qdict_del(bs
->options
, "backing");
3953 /* Remove child references from bs->options and bs->explicit_options.
3954 * Child options were already removed in bdrv_reopen_queue_child() */
3955 QLIST_FOREACH(child
, &bs
->children
, next
) {
3956 qdict_del(bs
->explicit_options
, child
->name
);
3957 qdict_del(bs
->options
, child
->name
);
3961 * Change the backing file if a new one was specified. We do this
3962 * after updating bs->options, so bdrv_refresh_filename() (called
3963 * from bdrv_set_backing_hd()) has the new values.
3965 if (reopen_state
->replace_backing_bs
) {
3966 BlockDriverState
*old_backing_bs
= backing_bs(bs
);
3967 assert(!old_backing_bs
|| !old_backing_bs
->implicit
);
3968 /* Abort the permission update on the backing bs we're detaching */
3969 if (old_backing_bs
) {
3970 bdrv_abort_perm_update(old_backing_bs
);
3972 bdrv_set_backing_hd(bs
, reopen_state
->new_backing_bs
, &error_abort
);
3975 bdrv_refresh_limits(bs
, NULL
);
3978 !bdrv_is_read_only(bs
) && !(bdrv_get_flags(bs
) & BDRV_O_INACTIVE
);
3979 if (!old_can_write
&& new_can_write
&& drv
->bdrv_reopen_bitmaps_rw
) {
3980 Error
*local_err
= NULL
;
3981 if (drv
->bdrv_reopen_bitmaps_rw(bs
, &local_err
) < 0) {
3982 /* This is not fatal, bitmaps just left read-only, so all following
3983 * writes will fail. User can remove read-only bitmaps to unblock
3986 error_reportf_err(local_err
,
3987 "%s: Failed to make dirty bitmaps writable: ",
3988 bdrv_get_node_name(bs
));
3994 * Abort the reopen, and delete and free the staged changes in
3997 void bdrv_reopen_abort(BDRVReopenState
*reopen_state
)
4001 assert(reopen_state
!= NULL
);
4002 drv
= reopen_state
->bs
->drv
;
4003 assert(drv
!= NULL
);
4005 if (drv
->bdrv_reopen_abort
) {
4006 drv
->bdrv_reopen_abort(reopen_state
);
4011 static void bdrv_close(BlockDriverState
*bs
)
4013 BdrvAioNotifier
*ban
, *ban_next
;
4014 BdrvChild
*child
, *next
;
4016 assert(!bs
->refcnt
);
4018 bdrv_drained_begin(bs
); /* complete I/O */
4020 bdrv_drain(bs
); /* in case flush left pending I/O */
4023 if (bs
->drv
->bdrv_close
) {
4024 bs
->drv
->bdrv_close(bs
);
4029 QLIST_FOREACH_SAFE(child
, &bs
->children
, next
, next
) {
4030 bdrv_unref_child(bs
, child
);
4037 atomic_set(&bs
->copy_on_read
, 0);
4038 bs
->backing_file
[0] = '\0';
4039 bs
->backing_format
[0] = '\0';
4040 bs
->total_sectors
= 0;
4041 bs
->encrypted
= false;
4043 qobject_unref(bs
->options
);
4044 qobject_unref(bs
->explicit_options
);
4046 bs
->explicit_options
= NULL
;
4047 qobject_unref(bs
->full_open_options
);
4048 bs
->full_open_options
= NULL
;
4050 bdrv_release_named_dirty_bitmaps(bs
);
4051 assert(QLIST_EMPTY(&bs
->dirty_bitmaps
));
4053 QLIST_FOREACH_SAFE(ban
, &bs
->aio_notifiers
, list
, ban_next
) {
4056 QLIST_INIT(&bs
->aio_notifiers
);
4057 bdrv_drained_end(bs
);
4060 void bdrv_close_all(void)
4062 assert(job_next(NULL
) == NULL
);
4063 nbd_export_close_all();
4065 /* Drop references from requests still in flight, such as canceled block
4066 * jobs whose AIO context has not been polled yet */
4069 blk_remove_all_bs();
4070 blockdev_close_all_bdrv_states();
4072 assert(QTAILQ_EMPTY(&all_bdrv_states
));
4075 static bool should_update_child(BdrvChild
*c
, BlockDriverState
*to
)
4081 if (c
->role
->stay_at_node
) {
4085 /* If the child @c belongs to the BDS @to, replacing the current
4086 * c->bs by @to would mean to create a loop.
4088 * Such a case occurs when appending a BDS to a backing chain.
4089 * For instance, imagine the following chain:
4091 * guest device -> node A -> further backing chain...
4093 * Now we create a new BDS B which we want to put on top of this
4094 * chain, so we first attach A as its backing node:
4099 * guest device -> node A -> further backing chain...
4101 * Finally we want to replace A by B. When doing that, we want to
4102 * replace all pointers to A by pointers to B -- except for the
4103 * pointer from B because (1) that would create a loop, and (2)
4104 * that pointer should simply stay intact:
4106 * guest device -> node B
4109 * node A -> further backing chain...
4111 * In general, when replacing a node A (c->bs) by a node B (@to),
4112 * if A is a child of B, that means we cannot replace A by B there
4113 * because that would create a loop. Silently detaching A from B
4114 * is also not really an option. So overall just leaving A in
4115 * place there is the most sensible choice.
4117 * We would also create a loop in any cases where @c is only
4118 * indirectly referenced by @to. Prevent this by returning false
4119 * if @c is found (by breadth-first search) anywhere in the whole
4124 found
= g_hash_table_new(NULL
, NULL
);
4125 g_hash_table_add(found
, to
);
4126 queue
= g_queue_new();
4127 g_queue_push_tail(queue
, to
);
4129 while (!g_queue_is_empty(queue
)) {
4130 BlockDriverState
*v
= g_queue_pop_head(queue
);
4133 QLIST_FOREACH(c2
, &v
->children
, next
) {
4139 if (g_hash_table_contains(found
, c2
->bs
)) {
4143 g_queue_push_tail(queue
, c2
->bs
);
4144 g_hash_table_add(found
, c2
->bs
);
4148 g_queue_free(queue
);
4149 g_hash_table_destroy(found
);
4154 void bdrv_replace_node(BlockDriverState
*from
, BlockDriverState
*to
,
4157 BdrvChild
*c
, *next
;
4158 GSList
*list
= NULL
, *p
;
4159 uint64_t old_perm
, old_shared
;
4160 uint64_t perm
= 0, shared
= BLK_PERM_ALL
;
4163 /* Make sure that @from doesn't go away until we have successfully attached
4164 * all of its parents to @to. */
4167 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
4168 bdrv_drained_begin(from
);
4170 /* Put all parents into @list and calculate their cumulative permissions */
4171 QLIST_FOREACH_SAFE(c
, &from
->parents
, next_parent
, next
) {
4172 assert(c
->bs
== from
);
4173 if (!should_update_child(c
, to
)) {
4177 error_setg(errp
, "Cannot change '%s' link to '%s'",
4178 c
->name
, from
->node_name
);
4181 list
= g_slist_prepend(list
, c
);
4183 shared
&= c
->shared_perm
;
4186 /* Check whether the required permissions can be granted on @to, ignoring
4187 * all BdrvChild in @list so that they can't block themselves. */
4188 ret
= bdrv_check_update_perm(to
, NULL
, perm
, shared
, list
, NULL
, errp
);
4190 bdrv_abort_perm_update(to
);
4194 /* Now actually perform the change. We performed the permission check for
4195 * all elements of @list at once, so set the permissions all at once at the
4197 for (p
= list
; p
!= NULL
; p
= p
->next
) {
4201 bdrv_replace_child_noperm(c
, to
);
4205 bdrv_get_cumulative_perm(to
, &old_perm
, &old_shared
);
4206 bdrv_set_perm(to
, old_perm
| perm
, old_shared
| shared
);
4210 bdrv_drained_end(from
);
4215 * Add new bs contents at the top of an image chain while the chain is
4216 * live, while keeping required fields on the top layer.
4218 * This will modify the BlockDriverState fields, and swap contents
4219 * between bs_new and bs_top. Both bs_new and bs_top are modified.
4221 * bs_new must not be attached to a BlockBackend.
4223 * This function does not create any image files.
4225 * bdrv_append() takes ownership of a bs_new reference and unrefs it because
4226 * that's what the callers commonly need. bs_new will be referenced by the old
4227 * parents of bs_top after bdrv_append() returns. If the caller needs to keep a
4228 * reference of its own, it must call bdrv_ref().
4230 void bdrv_append(BlockDriverState
*bs_new
, BlockDriverState
*bs_top
,
4233 Error
*local_err
= NULL
;
4235 bdrv_set_backing_hd(bs_new
, bs_top
, &local_err
);
4237 error_propagate(errp
, local_err
);
4241 bdrv_replace_node(bs_top
, bs_new
, &local_err
);
4243 error_propagate(errp
, local_err
);
4244 bdrv_set_backing_hd(bs_new
, NULL
, &error_abort
);
4248 /* bs_new is now referenced by its new parents, we don't need the
4249 * additional reference any more. */
4254 static void bdrv_delete(BlockDriverState
*bs
)
4256 assert(bdrv_op_blocker_is_empty(bs
));
4257 assert(!bs
->refcnt
);
4259 /* remove from list, if necessary */
4260 if (bs
->node_name
[0] != '\0') {
4261 QTAILQ_REMOVE(&graph_bdrv_states
, bs
, node_list
);
4263 QTAILQ_REMOVE(&all_bdrv_states
, bs
, bs_list
);
4271 * Run consistency checks on an image
4273 * Returns 0 if the check could be completed (it doesn't mean that the image is
4274 * free of errors) or -errno when an internal error occurred. The results of the
4275 * check are stored in res.
4277 static int coroutine_fn
bdrv_co_check(BlockDriverState
*bs
,
4278 BdrvCheckResult
*res
, BdrvCheckMode fix
)
4280 if (bs
->drv
== NULL
) {
4283 if (bs
->drv
->bdrv_co_check
== NULL
) {
4287 memset(res
, 0, sizeof(*res
));
4288 return bs
->drv
->bdrv_co_check(bs
, res
, fix
);
4291 typedef struct CheckCo
{
4292 BlockDriverState
*bs
;
4293 BdrvCheckResult
*res
;
4298 static void coroutine_fn
bdrv_check_co_entry(void *opaque
)
4300 CheckCo
*cco
= opaque
;
4301 cco
->ret
= bdrv_co_check(cco
->bs
, cco
->res
, cco
->fix
);
4305 int bdrv_check(BlockDriverState
*bs
,
4306 BdrvCheckResult
*res
, BdrvCheckMode fix
)
4312 .ret
= -EINPROGRESS
,
4316 if (qemu_in_coroutine()) {
4317 /* Fast-path if already in coroutine context */
4318 bdrv_check_co_entry(&cco
);
4320 co
= qemu_coroutine_create(bdrv_check_co_entry
, &cco
);
4321 bdrv_coroutine_enter(bs
, co
);
4322 BDRV_POLL_WHILE(bs
, cco
.ret
== -EINPROGRESS
);
4331 * -EINVAL - backing format specified, but no file
4332 * -ENOSPC - can't update the backing file because no space is left in the
4334 * -ENOTSUP - format driver doesn't support changing the backing file
4336 int bdrv_change_backing_file(BlockDriverState
*bs
,
4337 const char *backing_file
, const char *backing_fmt
)
4339 BlockDriver
*drv
= bs
->drv
;
4346 /* Backing file format doesn't make sense without a backing file */
4347 if (backing_fmt
&& !backing_file
) {
4351 if (drv
->bdrv_change_backing_file
!= NULL
) {
4352 ret
= drv
->bdrv_change_backing_file(bs
, backing_file
, backing_fmt
);
4358 pstrcpy(bs
->backing_file
, sizeof(bs
->backing_file
), backing_file
?: "");
4359 pstrcpy(bs
->backing_format
, sizeof(bs
->backing_format
), backing_fmt
?: "");
4360 pstrcpy(bs
->auto_backing_file
, sizeof(bs
->auto_backing_file
),
4361 backing_file
?: "");
4367 * Finds the image layer in the chain that has 'bs' as its backing file.
4369 * active is the current topmost image.
4371 * Returns NULL if bs is not found in active's image chain,
4372 * or if active == bs.
4374 * Returns the bottommost base image if bs == NULL.
4376 BlockDriverState
*bdrv_find_overlay(BlockDriverState
*active
,
4377 BlockDriverState
*bs
)
4379 while (active
&& bs
!= backing_bs(active
)) {
4380 active
= backing_bs(active
);
4386 /* Given a BDS, searches for the base layer. */
4387 BlockDriverState
*bdrv_find_base(BlockDriverState
*bs
)
4389 return bdrv_find_overlay(bs
, NULL
);
4393 * Return true if at least one of the backing links between @bs and
4394 * @base is frozen. @errp is set if that's the case.
4395 * @base must be reachable from @bs, or NULL.
4397 bool bdrv_is_backing_chain_frozen(BlockDriverState
*bs
, BlockDriverState
*base
,
4400 BlockDriverState
*i
;
4402 for (i
= bs
; i
!= base
; i
= backing_bs(i
)) {
4403 if (i
->backing
&& i
->backing
->frozen
) {
4404 error_setg(errp
, "Cannot change '%s' link from '%s' to '%s'",
4405 i
->backing
->name
, i
->node_name
,
4406 backing_bs(i
)->node_name
);
4415 * Freeze all backing links between @bs and @base.
4416 * If any of the links is already frozen the operation is aborted and
4417 * none of the links are modified.
4418 * @base must be reachable from @bs, or NULL.
4419 * Returns 0 on success. On failure returns < 0 and sets @errp.
4421 int bdrv_freeze_backing_chain(BlockDriverState
*bs
, BlockDriverState
*base
,
4424 BlockDriverState
*i
;
4426 if (bdrv_is_backing_chain_frozen(bs
, base
, errp
)) {
4430 for (i
= bs
; i
!= base
; i
= backing_bs(i
)) {
4431 if (i
->backing
&& backing_bs(i
)->never_freeze
) {
4432 error_setg(errp
, "Cannot freeze '%s' link to '%s'",
4433 i
->backing
->name
, backing_bs(i
)->node_name
);
4438 for (i
= bs
; i
!= base
; i
= backing_bs(i
)) {
4440 i
->backing
->frozen
= true;
4448 * Unfreeze all backing links between @bs and @base. The caller must
4449 * ensure that all links are frozen before using this function.
4450 * @base must be reachable from @bs, or NULL.
4452 void bdrv_unfreeze_backing_chain(BlockDriverState
*bs
, BlockDriverState
*base
)
4454 BlockDriverState
*i
;
4456 for (i
= bs
; i
!= base
; i
= backing_bs(i
)) {
4458 assert(i
->backing
->frozen
);
4459 i
->backing
->frozen
= false;
4465 * Drops images above 'base' up to and including 'top', and sets the image
4466 * above 'top' to have base as its backing file.
4468 * Requires that the overlay to 'top' is opened r/w, so that the backing file
4469 * information in 'bs' can be properly updated.
4471 * E.g., this will convert the following chain:
4472 * bottom <- base <- intermediate <- top <- active
4476 * bottom <- base <- active
4478 * It is allowed for bottom==base, in which case it converts:
4480 * base <- intermediate <- top <- active
4486 * If backing_file_str is non-NULL, it will be used when modifying top's
4487 * overlay image metadata.
4490 * if active == top, that is considered an error
4493 int bdrv_drop_intermediate(BlockDriverState
*top
, BlockDriverState
*base
,
4494 const char *backing_file_str
)
4496 BlockDriverState
*explicit_top
= top
;
4497 bool update_inherits_from
;
4498 BdrvChild
*c
, *next
;
4499 Error
*local_err
= NULL
;
4504 if (!top
->drv
|| !base
->drv
) {
4508 /* Make sure that base is in the backing chain of top */
4509 if (!bdrv_chain_contains(top
, base
)) {
4513 /* This function changes all links that point to top and makes
4514 * them point to base. Check that none of them is frozen. */
4515 QLIST_FOREACH(c
, &top
->parents
, next_parent
) {
4521 /* If 'base' recursively inherits from 'top' then we should set
4522 * base->inherits_from to top->inherits_from after 'top' and all
4523 * other intermediate nodes have been dropped.
4524 * If 'top' is an implicit node (e.g. "commit_top") we should skip
4525 * it because no one inherits from it. We use explicit_top for that. */
4526 while (explicit_top
&& explicit_top
->implicit
) {
4527 explicit_top
= backing_bs(explicit_top
);
4529 update_inherits_from
= bdrv_inherits_from_recursive(base
, explicit_top
);
4531 /* success - we can delete the intermediate states, and link top->base */
4532 /* TODO Check graph modification op blockers (BLK_PERM_GRAPH_MOD) once
4533 * we've figured out how they should work. */
4534 if (!backing_file_str
) {
4535 bdrv_refresh_filename(base
);
4536 backing_file_str
= base
->filename
;
4539 QLIST_FOREACH_SAFE(c
, &top
->parents
, next_parent
, next
) {
4540 /* Check whether we are allowed to switch c from top to base */
4541 GSList
*ignore_children
= g_slist_prepend(NULL
, c
);
4542 ret
= bdrv_check_update_perm(base
, NULL
, c
->perm
, c
->shared_perm
,
4543 ignore_children
, NULL
, &local_err
);
4544 g_slist_free(ignore_children
);
4546 error_report_err(local_err
);
4550 /* If so, update the backing file path in the image file */
4551 if (c
->role
->update_filename
) {
4552 ret
= c
->role
->update_filename(c
, base
, backing_file_str
,
4555 bdrv_abort_perm_update(base
);
4556 error_report_err(local_err
);
4561 /* Do the actual switch in the in-memory graph.
4562 * Completes bdrv_check_update_perm() transaction internally. */
4564 bdrv_replace_child(c
, base
);
4568 if (update_inherits_from
) {
4569 base
->inherits_from
= explicit_top
->inherits_from
;
4579 * Length of a allocated file in bytes. Sparse files are counted by actual
4580 * allocated space. Return < 0 if error or unknown.
4582 int64_t bdrv_get_allocated_file_size(BlockDriverState
*bs
)
4584 BlockDriver
*drv
= bs
->drv
;
4588 if (drv
->bdrv_get_allocated_file_size
) {
4589 return drv
->bdrv_get_allocated_file_size(bs
);
4592 return bdrv_get_allocated_file_size(bs
->file
->bs
);
4599 * @drv: Format driver
4600 * @opts: Creation options for new image
4601 * @in_bs: Existing image containing data for new image (may be NULL)
4602 * @errp: Error object
4603 * Returns: A #BlockMeasureInfo (free using qapi_free_BlockMeasureInfo())
4606 * Calculate file size required to create a new image.
4608 * If @in_bs is given then space for allocated clusters and zero clusters
4609 * from that image are included in the calculation. If @opts contains a
4610 * backing file that is shared by @in_bs then backing clusters may be omitted
4611 * from the calculation.
4613 * If @in_bs is NULL then the calculation includes no allocated clusters
4614 * unless a preallocation option is given in @opts.
4616 * Note that @in_bs may use a different BlockDriver from @drv.
4618 * If an error occurs the @errp pointer is set.
4620 BlockMeasureInfo
*bdrv_measure(BlockDriver
*drv
, QemuOpts
*opts
,
4621 BlockDriverState
*in_bs
, Error
**errp
)
4623 if (!drv
->bdrv_measure
) {
4624 error_setg(errp
, "Block driver '%s' does not support size measurement",
4629 return drv
->bdrv_measure(opts
, in_bs
, errp
);
4633 * Return number of sectors on success, -errno on error.
4635 int64_t bdrv_nb_sectors(BlockDriverState
*bs
)
4637 BlockDriver
*drv
= bs
->drv
;
4642 if (drv
->has_variable_length
) {
4643 int ret
= refresh_total_sectors(bs
, bs
->total_sectors
);
4648 return bs
->total_sectors
;
4652 * Return length in bytes on success, -errno on error.
4653 * The length is always a multiple of BDRV_SECTOR_SIZE.
4655 int64_t bdrv_getlength(BlockDriverState
*bs
)
4657 int64_t ret
= bdrv_nb_sectors(bs
);
4659 ret
= ret
> INT64_MAX
/ BDRV_SECTOR_SIZE
? -EFBIG
: ret
;
4660 return ret
< 0 ? ret
: ret
* BDRV_SECTOR_SIZE
;
4663 /* return 0 as number of sectors if no device present or error */
4664 void bdrv_get_geometry(BlockDriverState
*bs
, uint64_t *nb_sectors_ptr
)
4666 int64_t nb_sectors
= bdrv_nb_sectors(bs
);
4668 *nb_sectors_ptr
= nb_sectors
< 0 ? 0 : nb_sectors
;
4671 bool bdrv_is_sg(BlockDriverState
*bs
)
4676 bool bdrv_is_encrypted(BlockDriverState
*bs
)
4678 if (bs
->backing
&& bs
->backing
->bs
->encrypted
) {
4681 return bs
->encrypted
;
4684 const char *bdrv_get_format_name(BlockDriverState
*bs
)
4686 return bs
->drv
? bs
->drv
->format_name
: NULL
;
4689 static int qsort_strcmp(const void *a
, const void *b
)
4691 return strcmp(*(char *const *)a
, *(char *const *)b
);
4694 void bdrv_iterate_format(void (*it
)(void *opaque
, const char *name
),
4695 void *opaque
, bool read_only
)
4700 const char **formats
= NULL
;
4702 QLIST_FOREACH(drv
, &bdrv_drivers
, list
) {
4703 if (drv
->format_name
) {
4707 if (use_bdrv_whitelist
&& !bdrv_is_whitelisted(drv
, read_only
)) {
4711 while (formats
&& i
&& !found
) {
4712 found
= !strcmp(formats
[--i
], drv
->format_name
);
4716 formats
= g_renew(const char *, formats
, count
+ 1);
4717 formats
[count
++] = drv
->format_name
;
4722 for (i
= 0; i
< (int)ARRAY_SIZE(block_driver_modules
); i
++) {
4723 const char *format_name
= block_driver_modules
[i
].format_name
;
4729 if (use_bdrv_whitelist
&&
4730 !bdrv_format_is_whitelisted(format_name
, read_only
)) {
4734 while (formats
&& j
&& !found
) {
4735 found
= !strcmp(formats
[--j
], format_name
);
4739 formats
= g_renew(const char *, formats
, count
+ 1);
4740 formats
[count
++] = format_name
;
4745 qsort(formats
, count
, sizeof(formats
[0]), qsort_strcmp
);
4747 for (i
= 0; i
< count
; i
++) {
4748 it(opaque
, formats
[i
]);
4754 /* This function is to find a node in the bs graph */
4755 BlockDriverState
*bdrv_find_node(const char *node_name
)
4757 BlockDriverState
*bs
;
4761 QTAILQ_FOREACH(bs
, &graph_bdrv_states
, node_list
) {
4762 if (!strcmp(node_name
, bs
->node_name
)) {
4769 /* Put this QMP function here so it can access the static graph_bdrv_states. */
4770 BlockDeviceInfoList
*bdrv_named_nodes_list(Error
**errp
)
4772 BlockDeviceInfoList
*list
, *entry
;
4773 BlockDriverState
*bs
;
4776 QTAILQ_FOREACH(bs
, &graph_bdrv_states
, node_list
) {
4777 BlockDeviceInfo
*info
= bdrv_block_device_info(NULL
, bs
, errp
);
4779 qapi_free_BlockDeviceInfoList(list
);
4782 entry
= g_malloc0(sizeof(*entry
));
4783 entry
->value
= info
;
4791 #define QAPI_LIST_ADD(list, element) do { \
4792 typeof(list) _tmp = g_new(typeof(*(list)), 1); \
4793 _tmp->value = (element); \
4794 _tmp->next = (list); \
4798 typedef struct XDbgBlockGraphConstructor
{
4799 XDbgBlockGraph
*graph
;
4800 GHashTable
*graph_nodes
;
4801 } XDbgBlockGraphConstructor
;
4803 static XDbgBlockGraphConstructor
*xdbg_graph_new(void)
4805 XDbgBlockGraphConstructor
*gr
= g_new(XDbgBlockGraphConstructor
, 1);
4807 gr
->graph
= g_new0(XDbgBlockGraph
, 1);
4808 gr
->graph_nodes
= g_hash_table_new(NULL
, NULL
);
4813 static XDbgBlockGraph
*xdbg_graph_finalize(XDbgBlockGraphConstructor
*gr
)
4815 XDbgBlockGraph
*graph
= gr
->graph
;
4817 g_hash_table_destroy(gr
->graph_nodes
);
4823 static uintptr_t xdbg_graph_node_num(XDbgBlockGraphConstructor
*gr
, void *node
)
4825 uintptr_t ret
= (uintptr_t)g_hash_table_lookup(gr
->graph_nodes
, node
);
4832 * Start counting from 1, not 0, because 0 interferes with not-found (NULL)
4833 * answer of g_hash_table_lookup.
4835 ret
= g_hash_table_size(gr
->graph_nodes
) + 1;
4836 g_hash_table_insert(gr
->graph_nodes
, node
, (void *)ret
);
4841 static void xdbg_graph_add_node(XDbgBlockGraphConstructor
*gr
, void *node
,
4842 XDbgBlockGraphNodeType type
, const char *name
)
4844 XDbgBlockGraphNode
*n
;
4846 n
= g_new0(XDbgBlockGraphNode
, 1);
4848 n
->id
= xdbg_graph_node_num(gr
, node
);
4850 n
->name
= g_strdup(name
);
4852 QAPI_LIST_ADD(gr
->graph
->nodes
, n
);
4855 static void xdbg_graph_add_edge(XDbgBlockGraphConstructor
*gr
, void *parent
,
4856 const BdrvChild
*child
)
4860 BlockPermission num
;
4863 static const PermissionMap permissions
[] = {
4864 { BLK_PERM_CONSISTENT_READ
, BLOCK_PERMISSION_CONSISTENT_READ
},
4865 { BLK_PERM_WRITE
, BLOCK_PERMISSION_WRITE
},
4866 { BLK_PERM_WRITE_UNCHANGED
, BLOCK_PERMISSION_WRITE_UNCHANGED
},
4867 { BLK_PERM_RESIZE
, BLOCK_PERMISSION_RESIZE
},
4868 { BLK_PERM_GRAPH_MOD
, BLOCK_PERMISSION_GRAPH_MOD
},
4871 const PermissionMap
*p
;
4872 XDbgBlockGraphEdge
*edge
;
4874 QEMU_BUILD_BUG_ON(1UL << (ARRAY_SIZE(permissions
) - 1) != BLK_PERM_ALL
+ 1);
4876 edge
= g_new0(XDbgBlockGraphEdge
, 1);
4878 edge
->parent
= xdbg_graph_node_num(gr
, parent
);
4879 edge
->child
= xdbg_graph_node_num(gr
, child
->bs
);
4880 edge
->name
= g_strdup(child
->name
);
4882 for (p
= permissions
; p
->flag
; p
++) {
4883 if (p
->flag
& child
->perm
) {
4884 QAPI_LIST_ADD(edge
->perm
, p
->num
);
4886 if (p
->flag
& child
->shared_perm
) {
4887 QAPI_LIST_ADD(edge
->shared_perm
, p
->num
);
4891 QAPI_LIST_ADD(gr
->graph
->edges
, edge
);
4895 XDbgBlockGraph
*bdrv_get_xdbg_block_graph(Error
**errp
)
4899 BlockDriverState
*bs
;
4901 XDbgBlockGraphConstructor
*gr
= xdbg_graph_new();
4903 for (blk
= blk_all_next(NULL
); blk
; blk
= blk_all_next(blk
)) {
4904 char *allocated_name
= NULL
;
4905 const char *name
= blk_name(blk
);
4908 name
= allocated_name
= blk_get_attached_dev_id(blk
);
4910 xdbg_graph_add_node(gr
, blk
, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_BACKEND
,
4912 g_free(allocated_name
);
4913 if (blk_root(blk
)) {
4914 xdbg_graph_add_edge(gr
, blk
, blk_root(blk
));
4918 for (job
= block_job_next(NULL
); job
; job
= block_job_next(job
)) {
4921 xdbg_graph_add_node(gr
, job
, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_JOB
,
4923 for (el
= job
->nodes
; el
; el
= el
->next
) {
4924 xdbg_graph_add_edge(gr
, job
, (BdrvChild
*)el
->data
);
4928 QTAILQ_FOREACH(bs
, &graph_bdrv_states
, node_list
) {
4929 xdbg_graph_add_node(gr
, bs
, X_DBG_BLOCK_GRAPH_NODE_TYPE_BLOCK_DRIVER
,
4931 QLIST_FOREACH(child
, &bs
->children
, next
) {
4932 xdbg_graph_add_edge(gr
, bs
, child
);
4936 return xdbg_graph_finalize(gr
);
4939 BlockDriverState
*bdrv_lookup_bs(const char *device
,
4940 const char *node_name
,
4944 BlockDriverState
*bs
;
4947 blk
= blk_by_name(device
);
4952 error_setg(errp
, "Device '%s' has no medium", device
);
4960 bs
= bdrv_find_node(node_name
);
4967 error_setg(errp
, "Cannot find device=%s nor node_name=%s",
4968 device
? device
: "",
4969 node_name
? node_name
: "");
4973 /* If 'base' is in the same chain as 'top', return true. Otherwise,
4974 * return false. If either argument is NULL, return false. */
4975 bool bdrv_chain_contains(BlockDriverState
*top
, BlockDriverState
*base
)
4977 while (top
&& top
!= base
) {
4978 top
= backing_bs(top
);
4984 BlockDriverState
*bdrv_next_node(BlockDriverState
*bs
)
4987 return QTAILQ_FIRST(&graph_bdrv_states
);
4989 return QTAILQ_NEXT(bs
, node_list
);
4992 BlockDriverState
*bdrv_next_all_states(BlockDriverState
*bs
)
4995 return QTAILQ_FIRST(&all_bdrv_states
);
4997 return QTAILQ_NEXT(bs
, bs_list
);
5000 const char *bdrv_get_node_name(const BlockDriverState
*bs
)
5002 return bs
->node_name
;
5005 const char *bdrv_get_parent_name(const BlockDriverState
*bs
)
5010 /* If multiple parents have a name, just pick the first one. */
5011 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
5012 if (c
->role
->get_name
) {
5013 name
= c
->role
->get_name(c
);
5014 if (name
&& *name
) {
5023 /* TODO check what callers really want: bs->node_name or blk_name() */
5024 const char *bdrv_get_device_name(const BlockDriverState
*bs
)
5026 return bdrv_get_parent_name(bs
) ?: "";
5029 /* This can be used to identify nodes that might not have a device
5030 * name associated. Since node and device names live in the same
5031 * namespace, the result is unambiguous. The exception is if both are
5032 * absent, then this returns an empty (non-null) string. */
5033 const char *bdrv_get_device_or_node_name(const BlockDriverState
*bs
)
5035 return bdrv_get_parent_name(bs
) ?: bs
->node_name
;
5038 int bdrv_get_flags(BlockDriverState
*bs
)
5040 return bs
->open_flags
;
5043 int bdrv_has_zero_init_1(BlockDriverState
*bs
)
5048 int bdrv_has_zero_init(BlockDriverState
*bs
)
5054 /* If BS is a copy on write image, it is initialized to
5055 the contents of the base image, which may not be zeroes. */
5059 if (bs
->drv
->bdrv_has_zero_init
) {
5060 return bs
->drv
->bdrv_has_zero_init(bs
);
5062 if (bs
->file
&& bs
->drv
->is_filter
) {
5063 return bdrv_has_zero_init(bs
->file
->bs
);
5070 bool bdrv_unallocated_blocks_are_zero(BlockDriverState
*bs
)
5072 BlockDriverInfo bdi
;
5078 if (bdrv_get_info(bs
, &bdi
) == 0) {
5079 return bdi
.unallocated_blocks_are_zero
;
5085 bool bdrv_can_write_zeroes_with_unmap(BlockDriverState
*bs
)
5087 if (!(bs
->open_flags
& BDRV_O_UNMAP
)) {
5091 return bs
->supported_zero_flags
& BDRV_REQ_MAY_UNMAP
;
5094 void bdrv_get_backing_filename(BlockDriverState
*bs
,
5095 char *filename
, int filename_size
)
5097 pstrcpy(filename
, filename_size
, bs
->backing_file
);
5100 int bdrv_get_info(BlockDriverState
*bs
, BlockDriverInfo
*bdi
)
5102 BlockDriver
*drv
= bs
->drv
;
5103 /* if bs->drv == NULL, bs is closed, so there's nothing to do here */
5107 if (!drv
->bdrv_get_info
) {
5108 if (bs
->file
&& drv
->is_filter
) {
5109 return bdrv_get_info(bs
->file
->bs
, bdi
);
5113 memset(bdi
, 0, sizeof(*bdi
));
5114 return drv
->bdrv_get_info(bs
, bdi
);
5117 ImageInfoSpecific
*bdrv_get_specific_info(BlockDriverState
*bs
,
5120 BlockDriver
*drv
= bs
->drv
;
5121 if (drv
&& drv
->bdrv_get_specific_info
) {
5122 return drv
->bdrv_get_specific_info(bs
, errp
);
5127 void bdrv_debug_event(BlockDriverState
*bs
, BlkdebugEvent event
)
5129 if (!bs
|| !bs
->drv
|| !bs
->drv
->bdrv_debug_event
) {
5133 bs
->drv
->bdrv_debug_event(bs
, event
);
5136 int bdrv_debug_breakpoint(BlockDriverState
*bs
, const char *event
,
5139 while (bs
&& bs
->drv
&& !bs
->drv
->bdrv_debug_breakpoint
) {
5140 bs
= bs
->file
? bs
->file
->bs
: NULL
;
5143 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_breakpoint
) {
5144 return bs
->drv
->bdrv_debug_breakpoint(bs
, event
, tag
);
5150 int bdrv_debug_remove_breakpoint(BlockDriverState
*bs
, const char *tag
)
5152 while (bs
&& bs
->drv
&& !bs
->drv
->bdrv_debug_remove_breakpoint
) {
5153 bs
= bs
->file
? bs
->file
->bs
: NULL
;
5156 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_remove_breakpoint
) {
5157 return bs
->drv
->bdrv_debug_remove_breakpoint(bs
, tag
);
5163 int bdrv_debug_resume(BlockDriverState
*bs
, const char *tag
)
5165 while (bs
&& (!bs
->drv
|| !bs
->drv
->bdrv_debug_resume
)) {
5166 bs
= bs
->file
? bs
->file
->bs
: NULL
;
5169 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_resume
) {
5170 return bs
->drv
->bdrv_debug_resume(bs
, tag
);
5176 bool bdrv_debug_is_suspended(BlockDriverState
*bs
, const char *tag
)
5178 while (bs
&& bs
->drv
&& !bs
->drv
->bdrv_debug_is_suspended
) {
5179 bs
= bs
->file
? bs
->file
->bs
: NULL
;
5182 if (bs
&& bs
->drv
&& bs
->drv
->bdrv_debug_is_suspended
) {
5183 return bs
->drv
->bdrv_debug_is_suspended(bs
, tag
);
5189 /* backing_file can either be relative, or absolute, or a protocol. If it is
5190 * relative, it must be relative to the chain. So, passing in bs->filename
5191 * from a BDS as backing_file should not be done, as that may be relative to
5192 * the CWD rather than the chain. */
5193 BlockDriverState
*bdrv_find_backing_image(BlockDriverState
*bs
,
5194 const char *backing_file
)
5196 char *filename_full
= NULL
;
5197 char *backing_file_full
= NULL
;
5198 char *filename_tmp
= NULL
;
5199 int is_protocol
= 0;
5200 BlockDriverState
*curr_bs
= NULL
;
5201 BlockDriverState
*retval
= NULL
;
5203 if (!bs
|| !bs
->drv
|| !backing_file
) {
5207 filename_full
= g_malloc(PATH_MAX
);
5208 backing_file_full
= g_malloc(PATH_MAX
);
5210 is_protocol
= path_has_protocol(backing_file
);
5212 for (curr_bs
= bs
; curr_bs
->backing
; curr_bs
= curr_bs
->backing
->bs
) {
5214 /* If either of the filename paths is actually a protocol, then
5215 * compare unmodified paths; otherwise make paths relative */
5216 if (is_protocol
|| path_has_protocol(curr_bs
->backing_file
)) {
5217 char *backing_file_full_ret
;
5219 if (strcmp(backing_file
, curr_bs
->backing_file
) == 0) {
5220 retval
= curr_bs
->backing
->bs
;
5223 /* Also check against the full backing filename for the image */
5224 backing_file_full_ret
= bdrv_get_full_backing_filename(curr_bs
,
5226 if (backing_file_full_ret
) {
5227 bool equal
= strcmp(backing_file
, backing_file_full_ret
) == 0;
5228 g_free(backing_file_full_ret
);
5230 retval
= curr_bs
->backing
->bs
;
5235 /* If not an absolute filename path, make it relative to the current
5236 * image's filename path */
5237 filename_tmp
= bdrv_make_absolute_filename(curr_bs
, backing_file
,
5239 /* We are going to compare canonicalized absolute pathnames */
5240 if (!filename_tmp
|| !realpath(filename_tmp
, filename_full
)) {
5241 g_free(filename_tmp
);
5244 g_free(filename_tmp
);
5246 /* We need to make sure the backing filename we are comparing against
5247 * is relative to the current image filename (or absolute) */
5248 filename_tmp
= bdrv_get_full_backing_filename(curr_bs
, NULL
);
5249 if (!filename_tmp
|| !realpath(filename_tmp
, backing_file_full
)) {
5250 g_free(filename_tmp
);
5253 g_free(filename_tmp
);
5255 if (strcmp(backing_file_full
, filename_full
) == 0) {
5256 retval
= curr_bs
->backing
->bs
;
5262 g_free(filename_full
);
5263 g_free(backing_file_full
);
5267 void bdrv_init(void)
5269 module_call_init(MODULE_INIT_BLOCK
);
5272 void bdrv_init_with_whitelist(void)
5274 use_bdrv_whitelist
= 1;
5278 static void coroutine_fn
bdrv_co_invalidate_cache(BlockDriverState
*bs
,
5281 BdrvChild
*child
, *parent
;
5282 uint64_t perm
, shared_perm
;
5283 Error
*local_err
= NULL
;
5285 BdrvDirtyBitmap
*bm
;
5291 if (!(bs
->open_flags
& BDRV_O_INACTIVE
)) {
5295 QLIST_FOREACH(child
, &bs
->children
, next
) {
5296 bdrv_co_invalidate_cache(child
->bs
, &local_err
);
5298 error_propagate(errp
, local_err
);
5304 * Update permissions, they may differ for inactive nodes.
5306 * Note that the required permissions of inactive images are always a
5307 * subset of the permissions required after activating the image. This
5308 * allows us to just get the permissions upfront without restricting
5309 * drv->bdrv_invalidate_cache().
5311 * It also means that in error cases, we don't have to try and revert to
5312 * the old permissions (which is an operation that could fail, too). We can
5313 * just keep the extended permissions for the next time that an activation
5314 * of the image is tried.
5316 bs
->open_flags
&= ~BDRV_O_INACTIVE
;
5317 bdrv_get_cumulative_perm(bs
, &perm
, &shared_perm
);
5318 ret
= bdrv_check_perm(bs
, NULL
, perm
, shared_perm
, NULL
, NULL
, &local_err
);
5320 bs
->open_flags
|= BDRV_O_INACTIVE
;
5321 error_propagate(errp
, local_err
);
5324 bdrv_set_perm(bs
, perm
, shared_perm
);
5326 if (bs
->drv
->bdrv_co_invalidate_cache
) {
5327 bs
->drv
->bdrv_co_invalidate_cache(bs
, &local_err
);
5329 bs
->open_flags
|= BDRV_O_INACTIVE
;
5330 error_propagate(errp
, local_err
);
5335 for (bm
= bdrv_dirty_bitmap_next(bs
, NULL
); bm
;
5336 bm
= bdrv_dirty_bitmap_next(bs
, bm
))
5338 bdrv_dirty_bitmap_set_migration(bm
, false);
5341 ret
= refresh_total_sectors(bs
, bs
->total_sectors
);
5343 bs
->open_flags
|= BDRV_O_INACTIVE
;
5344 error_setg_errno(errp
, -ret
, "Could not refresh total sector count");
5348 QLIST_FOREACH(parent
, &bs
->parents
, next_parent
) {
5349 if (parent
->role
->activate
) {
5350 parent
->role
->activate(parent
, &local_err
);
5352 bs
->open_flags
|= BDRV_O_INACTIVE
;
5353 error_propagate(errp
, local_err
);
5360 typedef struct InvalidateCacheCo
{
5361 BlockDriverState
*bs
;
5364 } InvalidateCacheCo
;
5366 static void coroutine_fn
bdrv_invalidate_cache_co_entry(void *opaque
)
5368 InvalidateCacheCo
*ico
= opaque
;
5369 bdrv_co_invalidate_cache(ico
->bs
, ico
->errp
);
5374 void bdrv_invalidate_cache(BlockDriverState
*bs
, Error
**errp
)
5377 InvalidateCacheCo ico
= {
5383 if (qemu_in_coroutine()) {
5384 /* Fast-path if already in coroutine context */
5385 bdrv_invalidate_cache_co_entry(&ico
);
5387 co
= qemu_coroutine_create(bdrv_invalidate_cache_co_entry
, &ico
);
5388 bdrv_coroutine_enter(bs
, co
);
5389 BDRV_POLL_WHILE(bs
, !ico
.done
);
5393 void bdrv_invalidate_cache_all(Error
**errp
)
5395 BlockDriverState
*bs
;
5396 Error
*local_err
= NULL
;
5397 BdrvNextIterator it
;
5399 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
5400 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
5402 aio_context_acquire(aio_context
);
5403 bdrv_invalidate_cache(bs
, &local_err
);
5404 aio_context_release(aio_context
);
5406 error_propagate(errp
, local_err
);
5407 bdrv_next_cleanup(&it
);
5413 static bool bdrv_has_bds_parent(BlockDriverState
*bs
, bool only_active
)
5417 QLIST_FOREACH(parent
, &bs
->parents
, next_parent
) {
5418 if (parent
->role
->parent_is_bds
) {
5419 BlockDriverState
*parent_bs
= parent
->opaque
;
5420 if (!only_active
|| !(parent_bs
->open_flags
& BDRV_O_INACTIVE
)) {
5429 static int bdrv_inactivate_recurse(BlockDriverState
*bs
)
5431 BdrvChild
*child
, *parent
;
5432 bool tighten_restrictions
;
5433 uint64_t perm
, shared_perm
;
5440 /* Make sure that we don't inactivate a child before its parent.
5441 * It will be covered by recursion from the yet active parent. */
5442 if (bdrv_has_bds_parent(bs
, true)) {
5446 assert(!(bs
->open_flags
& BDRV_O_INACTIVE
));
5448 /* Inactivate this node */
5449 if (bs
->drv
->bdrv_inactivate
) {
5450 ret
= bs
->drv
->bdrv_inactivate(bs
);
5456 QLIST_FOREACH(parent
, &bs
->parents
, next_parent
) {
5457 if (parent
->role
->inactivate
) {
5458 ret
= parent
->role
->inactivate(parent
);
5465 bs
->open_flags
|= BDRV_O_INACTIVE
;
5467 /* Update permissions, they may differ for inactive nodes */
5468 bdrv_get_cumulative_perm(bs
, &perm
, &shared_perm
);
5469 ret
= bdrv_check_perm(bs
, NULL
, perm
, shared_perm
, NULL
,
5470 &tighten_restrictions
, NULL
);
5471 assert(tighten_restrictions
== false);
5473 /* We only tried to loosen restrictions, so errors are not fatal */
5474 bdrv_abort_perm_update(bs
);
5476 bdrv_set_perm(bs
, perm
, shared_perm
);
5480 /* Recursively inactivate children */
5481 QLIST_FOREACH(child
, &bs
->children
, next
) {
5482 ret
= bdrv_inactivate_recurse(child
->bs
);
5491 int bdrv_inactivate_all(void)
5493 BlockDriverState
*bs
= NULL
;
5494 BdrvNextIterator it
;
5496 GSList
*aio_ctxs
= NULL
, *ctx
;
5498 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
5499 AioContext
*aio_context
= bdrv_get_aio_context(bs
);
5501 if (!g_slist_find(aio_ctxs
, aio_context
)) {
5502 aio_ctxs
= g_slist_prepend(aio_ctxs
, aio_context
);
5503 aio_context_acquire(aio_context
);
5507 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
5508 /* Nodes with BDS parents are covered by recursion from the last
5509 * parent that gets inactivated. Don't inactivate them a second
5510 * time if that has already happened. */
5511 if (bdrv_has_bds_parent(bs
, false)) {
5514 ret
= bdrv_inactivate_recurse(bs
);
5516 bdrv_next_cleanup(&it
);
5522 for (ctx
= aio_ctxs
; ctx
!= NULL
; ctx
= ctx
->next
) {
5523 AioContext
*aio_context
= ctx
->data
;
5524 aio_context_release(aio_context
);
5526 g_slist_free(aio_ctxs
);
5531 /**************************************************************/
5532 /* removable device support */
5535 * Return TRUE if the media is present
5537 bool bdrv_is_inserted(BlockDriverState
*bs
)
5539 BlockDriver
*drv
= bs
->drv
;
5545 if (drv
->bdrv_is_inserted
) {
5546 return drv
->bdrv_is_inserted(bs
);
5548 QLIST_FOREACH(child
, &bs
->children
, next
) {
5549 if (!bdrv_is_inserted(child
->bs
)) {
5557 * If eject_flag is TRUE, eject the media. Otherwise, close the tray
5559 void bdrv_eject(BlockDriverState
*bs
, bool eject_flag
)
5561 BlockDriver
*drv
= bs
->drv
;
5563 if (drv
&& drv
->bdrv_eject
) {
5564 drv
->bdrv_eject(bs
, eject_flag
);
5569 * Lock or unlock the media (if it is locked, the user won't be able
5570 * to eject it manually).
5572 void bdrv_lock_medium(BlockDriverState
*bs
, bool locked
)
5574 BlockDriver
*drv
= bs
->drv
;
5576 trace_bdrv_lock_medium(bs
, locked
);
5578 if (drv
&& drv
->bdrv_lock_medium
) {
5579 drv
->bdrv_lock_medium(bs
, locked
);
5583 /* Get a reference to bs */
5584 void bdrv_ref(BlockDriverState
*bs
)
5589 /* Release a previously grabbed reference to bs.
5590 * If after releasing, reference count is zero, the BlockDriverState is
5592 void bdrv_unref(BlockDriverState
*bs
)
5597 assert(bs
->refcnt
> 0);
5598 if (--bs
->refcnt
== 0) {
5603 struct BdrvOpBlocker
{
5605 QLIST_ENTRY(BdrvOpBlocker
) list
;
5608 bool bdrv_op_is_blocked(BlockDriverState
*bs
, BlockOpType op
, Error
**errp
)
5610 BdrvOpBlocker
*blocker
;
5611 assert((int) op
>= 0 && op
< BLOCK_OP_TYPE_MAX
);
5612 if (!QLIST_EMPTY(&bs
->op_blockers
[op
])) {
5613 blocker
= QLIST_FIRST(&bs
->op_blockers
[op
]);
5614 error_propagate_prepend(errp
, error_copy(blocker
->reason
),
5615 "Node '%s' is busy: ",
5616 bdrv_get_device_or_node_name(bs
));
5622 void bdrv_op_block(BlockDriverState
*bs
, BlockOpType op
, Error
*reason
)
5624 BdrvOpBlocker
*blocker
;
5625 assert((int) op
>= 0 && op
< BLOCK_OP_TYPE_MAX
);
5627 blocker
= g_new0(BdrvOpBlocker
, 1);
5628 blocker
->reason
= reason
;
5629 QLIST_INSERT_HEAD(&bs
->op_blockers
[op
], blocker
, list
);
5632 void bdrv_op_unblock(BlockDriverState
*bs
, BlockOpType op
, Error
*reason
)
5634 BdrvOpBlocker
*blocker
, *next
;
5635 assert((int) op
>= 0 && op
< BLOCK_OP_TYPE_MAX
);
5636 QLIST_FOREACH_SAFE(blocker
, &bs
->op_blockers
[op
], list
, next
) {
5637 if (blocker
->reason
== reason
) {
5638 QLIST_REMOVE(blocker
, list
);
5644 void bdrv_op_block_all(BlockDriverState
*bs
, Error
*reason
)
5647 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
5648 bdrv_op_block(bs
, i
, reason
);
5652 void bdrv_op_unblock_all(BlockDriverState
*bs
, Error
*reason
)
5655 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
5656 bdrv_op_unblock(bs
, i
, reason
);
5660 bool bdrv_op_blocker_is_empty(BlockDriverState
*bs
)
5664 for (i
= 0; i
< BLOCK_OP_TYPE_MAX
; i
++) {
5665 if (!QLIST_EMPTY(&bs
->op_blockers
[i
])) {
5672 void bdrv_img_create(const char *filename
, const char *fmt
,
5673 const char *base_filename
, const char *base_fmt
,
5674 char *options
, uint64_t img_size
, int flags
, bool quiet
,
5677 QemuOptsList
*create_opts
= NULL
;
5678 QemuOpts
*opts
= NULL
;
5679 const char *backing_fmt
, *backing_file
;
5681 BlockDriver
*drv
, *proto_drv
;
5682 Error
*local_err
= NULL
;
5685 /* Find driver and parse its options */
5686 drv
= bdrv_find_format(fmt
);
5688 error_setg(errp
, "Unknown file format '%s'", fmt
);
5692 proto_drv
= bdrv_find_protocol(filename
, true, errp
);
5697 if (!drv
->create_opts
) {
5698 error_setg(errp
, "Format driver '%s' does not support image creation",
5703 if (!proto_drv
->create_opts
) {
5704 error_setg(errp
, "Protocol driver '%s' does not support image creation",
5705 proto_drv
->format_name
);
5709 create_opts
= qemu_opts_append(create_opts
, drv
->create_opts
);
5710 create_opts
= qemu_opts_append(create_opts
, proto_drv
->create_opts
);
5712 /* Create parameter list with default values */
5713 opts
= qemu_opts_create(create_opts
, NULL
, 0, &error_abort
);
5714 qemu_opt_set_number(opts
, BLOCK_OPT_SIZE
, img_size
, &error_abort
);
5716 /* Parse -o options */
5718 qemu_opts_do_parse(opts
, options
, NULL
, &local_err
);
5724 if (base_filename
) {
5725 qemu_opt_set(opts
, BLOCK_OPT_BACKING_FILE
, base_filename
, &local_err
);
5727 error_setg(errp
, "Backing file not supported for file format '%s'",
5734 qemu_opt_set(opts
, BLOCK_OPT_BACKING_FMT
, base_fmt
, &local_err
);
5736 error_setg(errp
, "Backing file format not supported for file "
5737 "format '%s'", fmt
);
5742 backing_file
= qemu_opt_get(opts
, BLOCK_OPT_BACKING_FILE
);
5744 if (!strcmp(filename
, backing_file
)) {
5745 error_setg(errp
, "Error: Trying to create an image with the "
5746 "same filename as the backing file");
5751 backing_fmt
= qemu_opt_get(opts
, BLOCK_OPT_BACKING_FMT
);
5753 /* The size for the image must always be specified, unless we have a backing
5754 * file and we have not been forbidden from opening it. */
5755 size
= qemu_opt_get_size(opts
, BLOCK_OPT_SIZE
, img_size
);
5756 if (backing_file
&& !(flags
& BDRV_O_NO_BACKING
)) {
5757 BlockDriverState
*bs
;
5760 QDict
*backing_options
= NULL
;
5763 bdrv_get_full_backing_filename_from_filename(filename
, backing_file
,
5768 assert(full_backing
);
5770 /* backing files always opened read-only */
5772 back_flags
&= ~(BDRV_O_RDWR
| BDRV_O_SNAPSHOT
| BDRV_O_NO_BACKING
);
5774 backing_options
= qdict_new();
5776 qdict_put_str(backing_options
, "driver", backing_fmt
);
5778 qdict_put_bool(backing_options
, BDRV_OPT_FORCE_SHARE
, true);
5780 bs
= bdrv_open(full_backing
, NULL
, backing_options
, back_flags
,
5782 g_free(full_backing
);
5783 if (!bs
&& size
!= -1) {
5784 /* Couldn't open BS, but we have a size, so it's nonfatal */
5785 warn_reportf_err(local_err
,
5786 "Could not verify backing image. "
5787 "This may become an error in future versions.\n");
5790 /* Couldn't open bs, do not have size */
5791 error_append_hint(&local_err
,
5792 "Could not open backing image to determine size.\n");
5796 /* Opened BS, have no size */
5797 size
= bdrv_getlength(bs
);
5799 error_setg_errno(errp
, -size
, "Could not get size of '%s'",
5804 qemu_opt_set_number(opts
, BLOCK_OPT_SIZE
, size
, &error_abort
);
5808 } /* (backing_file && !(flags & BDRV_O_NO_BACKING)) */
5811 error_setg(errp
, "Image creation needs a size parameter");
5816 printf("Formatting '%s', fmt=%s ", filename
, fmt
);
5817 qemu_opts_print(opts
, " ");
5821 ret
= bdrv_create(drv
, filename
, opts
, &local_err
);
5823 if (ret
== -EFBIG
) {
5824 /* This is generally a better message than whatever the driver would
5825 * deliver (especially because of the cluster_size_hint), since that
5826 * is most probably not much different from "image too large". */
5827 const char *cluster_size_hint
= "";
5828 if (qemu_opt_get_size(opts
, BLOCK_OPT_CLUSTER_SIZE
, 0)) {
5829 cluster_size_hint
= " (try using a larger cluster size)";
5831 error_setg(errp
, "The image size is too large for file format '%s'"
5832 "%s", fmt
, cluster_size_hint
);
5833 error_free(local_err
);
5838 qemu_opts_del(opts
);
5839 qemu_opts_free(create_opts
);
5840 error_propagate(errp
, local_err
);
5843 AioContext
*bdrv_get_aio_context(BlockDriverState
*bs
)
5845 return bs
? bs
->aio_context
: qemu_get_aio_context();
5848 void bdrv_coroutine_enter(BlockDriverState
*bs
, Coroutine
*co
)
5850 aio_co_enter(bdrv_get_aio_context(bs
), co
);
5853 static void bdrv_do_remove_aio_context_notifier(BdrvAioNotifier
*ban
)
5855 QLIST_REMOVE(ban
, list
);
5859 static void bdrv_detach_aio_context(BlockDriverState
*bs
)
5861 BdrvAioNotifier
*baf
, *baf_tmp
;
5863 assert(!bs
->walking_aio_notifiers
);
5864 bs
->walking_aio_notifiers
= true;
5865 QLIST_FOREACH_SAFE(baf
, &bs
->aio_notifiers
, list
, baf_tmp
) {
5867 bdrv_do_remove_aio_context_notifier(baf
);
5869 baf
->detach_aio_context(baf
->opaque
);
5872 /* Never mind iterating again to check for ->deleted. bdrv_close() will
5873 * remove remaining aio notifiers if we aren't called again.
5875 bs
->walking_aio_notifiers
= false;
5877 if (bs
->drv
&& bs
->drv
->bdrv_detach_aio_context
) {
5878 bs
->drv
->bdrv_detach_aio_context(bs
);
5881 if (bs
->quiesce_counter
) {
5882 aio_enable_external(bs
->aio_context
);
5884 bs
->aio_context
= NULL
;
5887 static void bdrv_attach_aio_context(BlockDriverState
*bs
,
5888 AioContext
*new_context
)
5890 BdrvAioNotifier
*ban
, *ban_tmp
;
5892 if (bs
->quiesce_counter
) {
5893 aio_disable_external(new_context
);
5896 bs
->aio_context
= new_context
;
5898 if (bs
->drv
&& bs
->drv
->bdrv_attach_aio_context
) {
5899 bs
->drv
->bdrv_attach_aio_context(bs
, new_context
);
5902 assert(!bs
->walking_aio_notifiers
);
5903 bs
->walking_aio_notifiers
= true;
5904 QLIST_FOREACH_SAFE(ban
, &bs
->aio_notifiers
, list
, ban_tmp
) {
5906 bdrv_do_remove_aio_context_notifier(ban
);
5908 ban
->attached_aio_context(new_context
, ban
->opaque
);
5911 bs
->walking_aio_notifiers
= false;
5915 * Changes the AioContext used for fd handlers, timers, and BHs by this
5916 * BlockDriverState and all its children and parents.
5918 * Must be called from the main AioContext.
5920 * The caller must own the AioContext lock for the old AioContext of bs, but it
5921 * must not own the AioContext lock for new_context (unless new_context is the
5922 * same as the current context of bs).
5924 * @ignore will accumulate all visited BdrvChild object. The caller is
5925 * responsible for freeing the list afterwards.
5927 void bdrv_set_aio_context_ignore(BlockDriverState
*bs
,
5928 AioContext
*new_context
, GSList
**ignore
)
5930 AioContext
*old_context
= bdrv_get_aio_context(bs
);
5933 g_assert(qemu_get_current_aio_context() == qemu_get_aio_context());
5935 if (old_context
== new_context
) {
5939 bdrv_drained_begin(bs
);
5941 QLIST_FOREACH(child
, &bs
->children
, next
) {
5942 if (g_slist_find(*ignore
, child
)) {
5945 *ignore
= g_slist_prepend(*ignore
, child
);
5946 bdrv_set_aio_context_ignore(child
->bs
, new_context
, ignore
);
5948 QLIST_FOREACH(child
, &bs
->parents
, next_parent
) {
5949 if (g_slist_find(*ignore
, child
)) {
5952 assert(child
->role
->set_aio_ctx
);
5953 *ignore
= g_slist_prepend(*ignore
, child
);
5954 child
->role
->set_aio_ctx(child
, new_context
, ignore
);
5957 bdrv_detach_aio_context(bs
);
5959 /* Acquire the new context, if necessary */
5960 if (qemu_get_aio_context() != new_context
) {
5961 aio_context_acquire(new_context
);
5964 bdrv_attach_aio_context(bs
, new_context
);
5967 * If this function was recursively called from
5968 * bdrv_set_aio_context_ignore(), there may be nodes in the
5969 * subtree that have not yet been moved to the new AioContext.
5970 * Release the old one so bdrv_drained_end() can poll them.
5972 if (qemu_get_aio_context() != old_context
) {
5973 aio_context_release(old_context
);
5976 bdrv_drained_end(bs
);
5978 if (qemu_get_aio_context() != old_context
) {
5979 aio_context_acquire(old_context
);
5981 if (qemu_get_aio_context() != new_context
) {
5982 aio_context_release(new_context
);
5986 static bool bdrv_parent_can_set_aio_context(BdrvChild
*c
, AioContext
*ctx
,
5987 GSList
**ignore
, Error
**errp
)
5989 if (g_slist_find(*ignore
, c
)) {
5992 *ignore
= g_slist_prepend(*ignore
, c
);
5994 /* A BdrvChildRole that doesn't handle AioContext changes cannot
5995 * tolerate any AioContext changes */
5996 if (!c
->role
->can_set_aio_ctx
) {
5997 char *user
= bdrv_child_user_desc(c
);
5998 error_setg(errp
, "Changing iothreads is not supported by %s", user
);
6002 if (!c
->role
->can_set_aio_ctx(c
, ctx
, ignore
, errp
)) {
6003 assert(!errp
|| *errp
);
6009 bool bdrv_child_can_set_aio_context(BdrvChild
*c
, AioContext
*ctx
,
6010 GSList
**ignore
, Error
**errp
)
6012 if (g_slist_find(*ignore
, c
)) {
6015 *ignore
= g_slist_prepend(*ignore
, c
);
6016 return bdrv_can_set_aio_context(c
->bs
, ctx
, ignore
, errp
);
6019 /* @ignore will accumulate all visited BdrvChild object. The caller is
6020 * responsible for freeing the list afterwards. */
6021 bool bdrv_can_set_aio_context(BlockDriverState
*bs
, AioContext
*ctx
,
6022 GSList
**ignore
, Error
**errp
)
6026 if (bdrv_get_aio_context(bs
) == ctx
) {
6030 QLIST_FOREACH(c
, &bs
->parents
, next_parent
) {
6031 if (!bdrv_parent_can_set_aio_context(c
, ctx
, ignore
, errp
)) {
6035 QLIST_FOREACH(c
, &bs
->children
, next
) {
6036 if (!bdrv_child_can_set_aio_context(c
, ctx
, ignore
, errp
)) {
6044 int bdrv_child_try_set_aio_context(BlockDriverState
*bs
, AioContext
*ctx
,
6045 BdrvChild
*ignore_child
, Error
**errp
)
6050 ignore
= ignore_child
? g_slist_prepend(NULL
, ignore_child
) : NULL
;
6051 ret
= bdrv_can_set_aio_context(bs
, ctx
, &ignore
, errp
);
6052 g_slist_free(ignore
);
6058 ignore
= ignore_child
? g_slist_prepend(NULL
, ignore_child
) : NULL
;
6059 bdrv_set_aio_context_ignore(bs
, ctx
, &ignore
);
6060 g_slist_free(ignore
);
6065 int bdrv_try_set_aio_context(BlockDriverState
*bs
, AioContext
*ctx
,
6068 return bdrv_child_try_set_aio_context(bs
, ctx
, NULL
, errp
);
6071 void bdrv_add_aio_context_notifier(BlockDriverState
*bs
,
6072 void (*attached_aio_context
)(AioContext
*new_context
, void *opaque
),
6073 void (*detach_aio_context
)(void *opaque
), void *opaque
)
6075 BdrvAioNotifier
*ban
= g_new(BdrvAioNotifier
, 1);
6076 *ban
= (BdrvAioNotifier
){
6077 .attached_aio_context
= attached_aio_context
,
6078 .detach_aio_context
= detach_aio_context
,
6082 QLIST_INSERT_HEAD(&bs
->aio_notifiers
, ban
, list
);
6085 void bdrv_remove_aio_context_notifier(BlockDriverState
*bs
,
6086 void (*attached_aio_context
)(AioContext
*,
6088 void (*detach_aio_context
)(void *),
6091 BdrvAioNotifier
*ban
, *ban_next
;
6093 QLIST_FOREACH_SAFE(ban
, &bs
->aio_notifiers
, list
, ban_next
) {
6094 if (ban
->attached_aio_context
== attached_aio_context
&&
6095 ban
->detach_aio_context
== detach_aio_context
&&
6096 ban
->opaque
== opaque
&&
6097 ban
->deleted
== false)
6099 if (bs
->walking_aio_notifiers
) {
6100 ban
->deleted
= true;
6102 bdrv_do_remove_aio_context_notifier(ban
);
6111 int bdrv_amend_options(BlockDriverState
*bs
, QemuOpts
*opts
,
6112 BlockDriverAmendStatusCB
*status_cb
, void *cb_opaque
,
6116 error_setg(errp
, "Node is ejected");
6119 if (!bs
->drv
->bdrv_amend_options
) {
6120 error_setg(errp
, "Block driver '%s' does not support option amendment",
6121 bs
->drv
->format_name
);
6124 return bs
->drv
->bdrv_amend_options(bs
, opts
, status_cb
, cb_opaque
, errp
);
6127 /* This function will be called by the bdrv_recurse_is_first_non_filter method
6128 * of block filter and by bdrv_is_first_non_filter.
6129 * It is used to test if the given bs is the candidate or recurse more in the
6132 bool bdrv_recurse_is_first_non_filter(BlockDriverState
*bs
,
6133 BlockDriverState
*candidate
)
6135 /* return false if basic checks fails */
6136 if (!bs
|| !bs
->drv
) {
6140 /* the code reached a non block filter driver -> check if the bs is
6141 * the same as the candidate. It's the recursion termination condition.
6143 if (!bs
->drv
->is_filter
) {
6144 return bs
== candidate
;
6146 /* Down this path the driver is a block filter driver */
6148 /* If the block filter recursion method is defined use it to recurse down
6151 if (bs
->drv
->bdrv_recurse_is_first_non_filter
) {
6152 return bs
->drv
->bdrv_recurse_is_first_non_filter(bs
, candidate
);
6155 /* the driver is a block filter but don't allow to recurse -> return false
6160 /* This function checks if the candidate is the first non filter bs down it's
6161 * bs chain. Since we don't have pointers to parents it explore all bs chains
6162 * from the top. Some filters can choose not to pass down the recursion.
6164 bool bdrv_is_first_non_filter(BlockDriverState
*candidate
)
6166 BlockDriverState
*bs
;
6167 BdrvNextIterator it
;
6169 /* walk down the bs forest recursively */
6170 for (bs
= bdrv_first(&it
); bs
; bs
= bdrv_next(&it
)) {
6173 /* try to recurse in this top level bs */
6174 perm
= bdrv_recurse_is_first_non_filter(bs
, candidate
);
6176 /* candidate is the first non filter */
6178 bdrv_next_cleanup(&it
);
6186 BlockDriverState
*check_to_replace_node(BlockDriverState
*parent_bs
,
6187 const char *node_name
, Error
**errp
)
6189 BlockDriverState
*to_replace_bs
= bdrv_find_node(node_name
);
6190 AioContext
*aio_context
;
6192 if (!to_replace_bs
) {
6193 error_setg(errp
, "Node name '%s' not found", node_name
);
6197 aio_context
= bdrv_get_aio_context(to_replace_bs
);
6198 aio_context_acquire(aio_context
);
6200 if (bdrv_op_is_blocked(to_replace_bs
, BLOCK_OP_TYPE_REPLACE
, errp
)) {
6201 to_replace_bs
= NULL
;
6205 /* We don't want arbitrary node of the BDS chain to be replaced only the top
6206 * most non filter in order to prevent data corruption.
6207 * Another benefit is that this tests exclude backing files which are
6208 * blocked by the backing blockers.
6210 if (!bdrv_recurse_is_first_non_filter(parent_bs
, to_replace_bs
)) {
6211 error_setg(errp
, "Only top most non filter can be replaced");
6212 to_replace_bs
= NULL
;
6217 aio_context_release(aio_context
);
6218 return to_replace_bs
;
6222 * Iterates through the list of runtime option keys that are said to
6223 * be "strong" for a BDS. An option is called "strong" if it changes
6224 * a BDS's data. For example, the null block driver's "size" and
6225 * "read-zeroes" options are strong, but its "latency-ns" option is
6228 * If a key returned by this function ends with a dot, all options
6229 * starting with that prefix are strong.
6231 static const char *const *strong_options(BlockDriverState
*bs
,
6232 const char *const *curopt
)
6234 static const char *const global_options
[] = {
6235 "driver", "filename", NULL
6239 return &global_options
[0];
6243 if (curopt
== &global_options
[ARRAY_SIZE(global_options
) - 1] && bs
->drv
) {
6244 curopt
= bs
->drv
->strong_runtime_opts
;
6247 return (curopt
&& *curopt
) ? curopt
: NULL
;
6251 * Copies all strong runtime options from bs->options to the given
6252 * QDict. The set of strong option keys is determined by invoking
6255 * Returns true iff any strong option was present in bs->options (and
6256 * thus copied to the target QDict) with the exception of "filename"
6257 * and "driver". The caller is expected to use this value to decide
6258 * whether the existence of strong options prevents the generation of
6261 static bool append_strong_runtime_options(QDict
*d
, BlockDriverState
*bs
)
6263 bool found_any
= false;
6264 const char *const *option_name
= NULL
;
6270 while ((option_name
= strong_options(bs
, option_name
))) {
6271 bool option_given
= false;
6273 assert(strlen(*option_name
) > 0);
6274 if ((*option_name
)[strlen(*option_name
) - 1] != '.') {
6275 QObject
*entry
= qdict_get(bs
->options
, *option_name
);
6280 qdict_put_obj(d
, *option_name
, qobject_ref(entry
));
6281 option_given
= true;
6283 const QDictEntry
*entry
;
6284 for (entry
= qdict_first(bs
->options
); entry
;
6285 entry
= qdict_next(bs
->options
, entry
))
6287 if (strstart(qdict_entry_key(entry
), *option_name
, NULL
)) {
6288 qdict_put_obj(d
, qdict_entry_key(entry
),
6289 qobject_ref(qdict_entry_value(entry
)));
6290 option_given
= true;
6295 /* While "driver" and "filename" need to be included in a JSON filename,
6296 * their existence does not prohibit generation of a plain filename. */
6297 if (!found_any
&& option_given
&&
6298 strcmp(*option_name
, "driver") && strcmp(*option_name
, "filename"))
6304 if (!qdict_haskey(d
, "driver")) {
6305 /* Drivers created with bdrv_new_open_driver() may not have a
6306 * @driver option. Add it here. */
6307 qdict_put_str(d
, "driver", bs
->drv
->format_name
);
6313 /* Note: This function may return false positives; it may return true
6314 * even if opening the backing file specified by bs's image header
6315 * would result in exactly bs->backing. */
6316 static bool bdrv_backing_overridden(BlockDriverState
*bs
)
6319 return strcmp(bs
->auto_backing_file
,
6320 bs
->backing
->bs
->filename
);
6322 /* No backing BDS, so if the image header reports any backing
6323 * file, it must have been suppressed */
6324 return bs
->auto_backing_file
[0] != '\0';
6328 /* Updates the following BDS fields:
6329 * - exact_filename: A filename which may be used for opening a block device
6330 * which (mostly) equals the given BDS (even without any
6331 * other options; so reading and writing must return the same
6332 * results, but caching etc. may be different)
6333 * - full_open_options: Options which, when given when opening a block device
6334 * (without a filename), result in a BDS (mostly)
6335 * equalling the given one
6336 * - filename: If exact_filename is set, it is copied here. Otherwise,
6337 * full_open_options is converted to a JSON object, prefixed with
6338 * "json:" (for use through the JSON pseudo protocol) and put here.
6340 void bdrv_refresh_filename(BlockDriverState
*bs
)
6342 BlockDriver
*drv
= bs
->drv
;
6345 bool backing_overridden
;
6346 bool generate_json_filename
; /* Whether our default implementation should
6347 fill exact_filename (false) or not (true) */
6353 /* This BDS's file name may depend on any of its children's file names, so
6354 * refresh those first */
6355 QLIST_FOREACH(child
, &bs
->children
, next
) {
6356 bdrv_refresh_filename(child
->bs
);
6360 /* For implicit nodes, just copy everything from the single child */
6361 child
= QLIST_FIRST(&bs
->children
);
6362 assert(QLIST_NEXT(child
, next
) == NULL
);
6364 pstrcpy(bs
->exact_filename
, sizeof(bs
->exact_filename
),
6365 child
->bs
->exact_filename
);
6366 pstrcpy(bs
->filename
, sizeof(bs
->filename
), child
->bs
->filename
);
6368 bs
->full_open_options
= qobject_ref(child
->bs
->full_open_options
);
6373 backing_overridden
= bdrv_backing_overridden(bs
);
6375 if (bs
->open_flags
& BDRV_O_NO_IO
) {
6376 /* Without I/O, the backing file does not change anything.
6377 * Therefore, in such a case (primarily qemu-img), we can
6378 * pretend the backing file has not been overridden even if
6379 * it technically has been. */
6380 backing_overridden
= false;
6383 /* Gather the options QDict */
6385 generate_json_filename
= append_strong_runtime_options(opts
, bs
);
6386 generate_json_filename
|= backing_overridden
;
6388 if (drv
->bdrv_gather_child_options
) {
6389 /* Some block drivers may not want to present all of their children's
6390 * options, or name them differently from BdrvChild.name */
6391 drv
->bdrv_gather_child_options(bs
, opts
, backing_overridden
);
6393 QLIST_FOREACH(child
, &bs
->children
, next
) {
6394 if (child
->role
== &child_backing
&& !backing_overridden
) {
6395 /* We can skip the backing BDS if it has not been overridden */
6399 qdict_put(opts
, child
->name
,
6400 qobject_ref(child
->bs
->full_open_options
));
6403 if (backing_overridden
&& !bs
->backing
) {
6404 /* Force no backing file */
6405 qdict_put_null(opts
, "backing");
6409 qobject_unref(bs
->full_open_options
);
6410 bs
->full_open_options
= opts
;
6412 if (drv
->bdrv_refresh_filename
) {
6413 /* Obsolete information is of no use here, so drop the old file name
6414 * information before refreshing it */
6415 bs
->exact_filename
[0] = '\0';
6417 drv
->bdrv_refresh_filename(bs
);
6418 } else if (bs
->file
) {
6419 /* Try to reconstruct valid information from the underlying file */
6421 bs
->exact_filename
[0] = '\0';
6424 * We can use the underlying file's filename if:
6425 * - it has a filename,
6426 * - the file is a protocol BDS, and
6427 * - opening that file (as this BDS's format) will automatically create
6428 * the BDS tree we have right now, that is:
6429 * - the user did not significantly change this BDS's behavior with
6430 * some explicit (strong) options
6431 * - no non-file child of this BDS has been overridden by the user
6432 * Both of these conditions are represented by generate_json_filename.
6434 if (bs
->file
->bs
->exact_filename
[0] &&
6435 bs
->file
->bs
->drv
->bdrv_file_open
&&
6436 !generate_json_filename
)
6438 strcpy(bs
->exact_filename
, bs
->file
->bs
->exact_filename
);
6442 if (bs
->exact_filename
[0]) {
6443 pstrcpy(bs
->filename
, sizeof(bs
->filename
), bs
->exact_filename
);
6445 QString
*json
= qobject_to_json(QOBJECT(bs
->full_open_options
));
6446 snprintf(bs
->filename
, sizeof(bs
->filename
), "json:%s",
6447 qstring_get_str(json
));
6448 qobject_unref(json
);
6452 char *bdrv_dirname(BlockDriverState
*bs
, Error
**errp
)
6454 BlockDriver
*drv
= bs
->drv
;
6457 error_setg(errp
, "Node '%s' is ejected", bs
->node_name
);
6461 if (drv
->bdrv_dirname
) {
6462 return drv
->bdrv_dirname(bs
, errp
);
6466 return bdrv_dirname(bs
->file
->bs
, errp
);
6469 bdrv_refresh_filename(bs
);
6470 if (bs
->exact_filename
[0] != '\0') {
6471 return path_combine(bs
->exact_filename
, "");
6474 error_setg(errp
, "Cannot generate a base directory for %s nodes",
6480 * Hot add/remove a BDS's child. So the user can take a child offline when
6481 * it is broken and take a new child online
6483 void bdrv_add_child(BlockDriverState
*parent_bs
, BlockDriverState
*child_bs
,
6487 if (!parent_bs
->drv
|| !parent_bs
->drv
->bdrv_add_child
) {
6488 error_setg(errp
, "The node %s does not support adding a child",
6489 bdrv_get_device_or_node_name(parent_bs
));
6493 if (!QLIST_EMPTY(&child_bs
->parents
)) {
6494 error_setg(errp
, "The node %s already has a parent",
6495 child_bs
->node_name
);
6499 parent_bs
->drv
->bdrv_add_child(parent_bs
, child_bs
, errp
);
6502 void bdrv_del_child(BlockDriverState
*parent_bs
, BdrvChild
*child
, Error
**errp
)
6506 if (!parent_bs
->drv
|| !parent_bs
->drv
->bdrv_del_child
) {
6507 error_setg(errp
, "The node %s does not support removing a child",
6508 bdrv_get_device_or_node_name(parent_bs
));
6512 QLIST_FOREACH(tmp
, &parent_bs
->children
, next
) {
6519 error_setg(errp
, "The node %s does not have a child named %s",
6520 bdrv_get_device_or_node_name(parent_bs
),
6521 bdrv_get_device_or_node_name(child
->bs
));
6525 parent_bs
->drv
->bdrv_del_child(parent_bs
, child
, errp
);
6528 bool bdrv_can_store_new_dirty_bitmap(BlockDriverState
*bs
, const char *name
,
6529 uint32_t granularity
, Error
**errp
)
6531 BlockDriver
*drv
= bs
->drv
;
6534 error_setg_errno(errp
, ENOMEDIUM
,
6535 "Can't store persistent bitmaps to %s",
6536 bdrv_get_device_or_node_name(bs
));
6540 if (!drv
->bdrv_can_store_new_dirty_bitmap
) {
6541 error_setg_errno(errp
, ENOTSUP
,
6542 "Can't store persistent bitmaps to %s",
6543 bdrv_get_device_or_node_name(bs
));
6547 return drv
->bdrv_can_store_new_dirty_bitmap(bs
, name
, granularity
, errp
);